mirrors/luau
1
0
Fork 0
mirror of https://github.com/luau-lang/luau.git synced 2024-11-15 14:25:44 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity
16fbfe912c
luau/tests/IrBuilder.test.cpp

2822 lines
88 KiB
C++
Raw Normal View History

Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
// This file is part of the Luau programming language and is licensed under MIT License; see LICENSE.txt for details
#include "Luau/IrBuilder.h"
#include "Luau/IrAnalysis.h"
#include "Luau/IrDump.h"
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
#include "Luau/IrUtils.h"
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
#include "Luau/OptimizeConstProp.h"
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
#include "Luau/OptimizeFinalX64.h"
Sync to upstream/release/593 (#1024) - Updated Roblox copyright to 2023 - Floor division operator `//` (implements #832) - Autocomplete now shows `end` within `do` blocks - Restore BraceType when using `Lexer::lookahead` (fixes #1019) # New typechecker - Subtyping tests between metatables and tables - Subtyping tests between string singletons and tables - Subtyping tests for class types # Native codegen - Fixed macOS test failure (wrong spill restore offset) - Fixed clobbering of non-volatile xmm registers on Windows - Fixed wrong storage location of SSA reg spills - Implemented A64 support for add/sub extended - Eliminated zextReg from A64 lowering - Remove identical table slot lookups - Propagate values from predecessor into the linear block - Disabled reuse slot optimization - Keep `LuaNode::val` check for nil when optimizing `CHECK_SLOT_MATCH` - Implemented IR translation of `table.insert` builtin - Fixed mmap error handling on macOS/Linux # Tooling - Used `|` as a column separator instead of `+` in `bench.py` - Added a `table.sort` micro-benchmark - Switched `libprotobuf-mutator` to a less problematic version
2023-09-02 01:58:27 +08:00
#include "ScopedFlags.h"
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
#include "doctest.h"
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
#include <limits.h>
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
using namespace Luau::CodeGen;
class IrBuilderFixture
{
public:
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
void constantFold()
{
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
for (IrBlock& block : build.function.blocks)
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
{
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
if (block.kind == IrBlockKind::Dead)
continue;
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
for (size_t i = block.start; i <= block.finish; i++)
{
IrInst& inst = build.function.instructions[i];
applySubstitutions(build.function, inst);
foldConstants(build, build.function, block, uint32_t(i));
}
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
}
}
template<typename F>
void withOneBlock(F&& f)
{
IrOp main = build.block(IrBlockKind::Internal);
IrOp a = build.block(IrBlockKind::Internal);
build.beginBlock(main);
f(a);
build.beginBlock(a);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(1));
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
};
template<typename F>
void withTwoBlocks(F&& f)
{
IrOp main = build.block(IrBlockKind::Internal);
IrOp a = build.block(IrBlockKind::Internal);
IrOp b = build.block(IrBlockKind::Internal);
build.beginBlock(main);
f(a, b);
build.beginBlock(a);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(1));
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
build.beginBlock(b);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(2));
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
};
void checkEq(IrOp instOp, const IrInst& inst)
{
const IrInst& target = build.function.instOp(instOp);
CHECK(target.cmd == inst.cmd);
Sync to upstream/release/568 (#865) * A small subset of control-flow refinements have been added to recognize type options that are unreachable after a conditional/unconditional code block. (Fixes https://github.com/Roblox/luau/issues/356). Some examples: ```lua local function f(x: string?) if not x then return end -- x is 'string' here end ``` Throwing calls like `error` or `assert(false)` instead of 'return' are also recognized. Existing complex refinements like type/typeof and tagged union checks are expected to work, among others. To enable this feature, `LuauTinyControlFlowAnalysis` exclusion has to be removed from `ExperimentalFlags.h`. If will become enabled unconditionally in the near future. * Linter has been integrated into the typechecker analysis so that type-aware lint warnings can work in any mode `Frontend::lint` methods were deprecated, `Frontend::check` has to be used instead with `runLintChecks` option set. Resulting lint warning are located inside `CheckResult`. * Fixed large performance drop and increased memory consumption when array is filled at an offset (Fixes https://github.com/Roblox/luau/issues/590) * Part of [Type error suppression RFC](https://github.com/Roblox/luau/blob/master/rfcs/type-error-suppression.md) was implemented making subtyping checks with `any` type transitive. --- In our work on the new type-solver: * `--!nocheck` mode no longer reports type errors * New solver will not be used for `--!nonstrict` modules until all issues with strict mode typechecking are fixed * Added control-flow aware type refinements mentioned earlier In native code generation: * `LOP_NAMECALL` has been translated to IR * `type` and `typeof` builtin fastcalls have been translated to IR/assembly * Additional steps were taken towards arm64 support
2023-03-18 03:20:37 +08:00
CHECK(target.a == inst.a);
CHECK(target.b == inst.b);
CHECK(target.c == inst.c);
CHECK(target.d == inst.d);
CHECK(target.e == inst.e);
CHECK(target.f == inst.f);
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
}
Sync to upstream/release/580 (#951) * Added luau-compile executable target to build/test compilation without having full REPL included. In our new typechecker: * Fixed the order in which constraints are checked to get more deterministic errors in different environments * Fixed `isNumber`/`isString` checks to fix false positive errors in binary comparisons * CannotCallNonFunction error is reported when calling an intersection type of non-functions In our native code generation (jit): * Outlined X64 return instruction code to improve code size * Improved performance of return instruction on A64 * Added construction of a dominator tree for future optimizations
2023-06-10 01:08:00 +08:00
void defineCfgTree(const std::vector<std::vector<uint32_t>>& successorSets)
{
for (const std::vector<uint32_t>& successorSet : successorSets)
{
build.beginBlock(build.block(IrBlockKind::Internal));
build.function.cfg.successorsOffsets.push_back(uint32_t(build.function.cfg.successors.size()));
build.function.cfg.successors.insert(build.function.cfg.successors.end(), successorSet.begin(), successorSet.end());
}
// Brute-force the predecessor list
for (int i = 0; i < int(build.function.blocks.size()); i++)
{
build.function.cfg.predecessorsOffsets.push_back(uint32_t(build.function.cfg.predecessors.size()));
for (int k = 0; k < int(build.function.blocks.size()); k++)
{
for (uint32_t succIdx : successors(build.function.cfg, k))
{
if (succIdx == uint32_t(i))
build.function.cfg.predecessors.push_back(k);
}
}
}
computeCfgImmediateDominators(build.function);
computeCfgDominanceTreeChildren(build.function);
}
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
IrBuilder build;
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
// Luau.VM headers are not accessible
static const int tnil = 0;
static const int tboolean = 1;
static const int tnumber = 3;
Sync to upstream/release/578 (#939) * Fixed gcc warning about uninitialized `std::optional` * Fixed inlining of functions when they are used to compute their own arguments In the new type solver: * Type families that are not part of a function signature cannot be resolved at instantiation time and will now produce an error. This will be relaxed in the future when we get constraint clauses on function signatures (internally) * `never` type is now comparable * Improved typechecking of `for..in` statements * Fixed checks for number type in `Add` type family * Performance was improved, with particularly large gains on large projects And in native code generation (jit): * We eliminated the call instruction overhead when native code support is enabled in the VM * Small optimizations to arm64 lowering * Reworked LOP_GETIMPORT handling to reduce assembly code size * Fixed non-deterministic binary output * Fixed bad code generation caused by incorrect SSA to VM register links invalidation --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Andy Friesen <afriesen@roblox.com>
2023-05-26 05:36:34 +08:00
static const int tstring = 5;
static const int ttable = 6;
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
};
TEST_SUITE_BEGIN("Optimization");
TEST_CASE_FIXTURE(IrBuilderFixture, "FinalX64OptCheckTag")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp fallback = build.block(IrBlockKind::Fallback);
build.beginBlock(block);
IrOp tag1 = build.inst(IrCmd::LOAD_TAG, build.vmReg(2));
build.inst(IrCmd::CHECK_TAG, tag1, build.constTag(0), fallback);
IrOp tag2 = build.inst(IrCmd::LOAD_TAG, build.vmConst(5));
build.inst(IrCmd::CHECK_TAG, tag2, build.constTag(0), fallback);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
build.beginBlock(fallback);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(1));
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
updateUseCounts(build.function);
optimizeMemoryOperandsX64(build.function);
// Load from memory is 'inlined' into CHECK_TAG
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
bb_0:
CHECK_TAG R2, tnil, bb_fallback_1
CHECK_TAG K5, tnil, bb_fallback_1
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
bb_fallback_1:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 1u
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "FinalX64OptBinaryArith")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
IrOp opA = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(1));
IrOp opB = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(2));
build.inst(IrCmd::ADD_NUM, opA, opB);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
updateUseCounts(build.function);
optimizeMemoryOperandsX64(build.function);
// Load from memory is 'inlined' into second argument
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
bb_0:
%0 = LOAD_DOUBLE R1
%2 = ADD_NUM %0, R2
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "FinalX64OptEqTag1")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp trueBlock = build.block(IrBlockKind::Internal);
IrOp falseBlock = build.block(IrBlockKind::Internal);
build.beginBlock(block);
IrOp opA = build.inst(IrCmd::LOAD_TAG, build.vmReg(1));
IrOp opB = build.inst(IrCmd::LOAD_TAG, build.vmReg(2));
build.inst(IrCmd::JUMP_EQ_TAG, opA, opB, trueBlock, falseBlock);
build.beginBlock(trueBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
build.beginBlock(falseBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
updateUseCounts(build.function);
optimizeMemoryOperandsX64(build.function);
// Load from memory is 'inlined' into first argument
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
bb_0:
%1 = LOAD_TAG R2
JUMP_EQ_TAG R1, %1, bb_1, bb_2
bb_1:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
bb_2:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "FinalX64OptEqTag2")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp trueBlock = build.block(IrBlockKind::Internal);
IrOp falseBlock = build.block(IrBlockKind::Internal);
build.beginBlock(block);
IrOp opA = build.inst(IrCmd::LOAD_TAG, build.vmReg(1));
IrOp opB = build.inst(IrCmd::LOAD_TAG, build.vmReg(2));
build.inst(IrCmd::STORE_TAG, build.vmReg(6), opA);
build.inst(IrCmd::JUMP_EQ_TAG, opA, opB, trueBlock, falseBlock);
build.beginBlock(trueBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
build.beginBlock(falseBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
updateUseCounts(build.function);
optimizeMemoryOperandsX64(build.function);
// Load from memory is 'inlined' into second argument is it can't be done for the first one
// We also swap first and second argument to generate memory access on the LHS
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
bb_0:
%0 = LOAD_TAG R1
STORE_TAG R6, %0
JUMP_EQ_TAG R2, %0, bb_1, bb_2
bb_1:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
bb_2:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "FinalX64OptEqTag3")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp trueBlock = build.block(IrBlockKind::Internal);
IrOp falseBlock = build.block(IrBlockKind::Internal);
build.beginBlock(block);
IrOp table = build.inst(IrCmd::LOAD_POINTER, build.vmReg(1));
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
IrOp arrElem = build.inst(IrCmd::GET_ARR_ADDR, table, build.constInt(0));
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
IrOp opA = build.inst(IrCmd::LOAD_TAG, arrElem);
build.inst(IrCmd::JUMP_EQ_TAG, opA, build.constTag(0), trueBlock, falseBlock);
build.beginBlock(trueBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
build.beginBlock(falseBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
updateUseCounts(build.function);
optimizeMemoryOperandsX64(build.function);
// Load from memory is 'inlined' into first argument
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
bb_0:
%0 = LOAD_POINTER R1
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
%1 = GET_ARR_ADDR %0, 0i
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
%2 = LOAD_TAG %1
JUMP_EQ_TAG %2, tnil, bb_1, bb_2
bb_1:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
bb_2:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "FinalX64OptJumpCmpNum")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp trueBlock = build.block(IrBlockKind::Internal);
IrOp falseBlock = build.block(IrBlockKind::Internal);
build.beginBlock(block);
IrOp opA = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(1));
IrOp opB = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(2));
build.inst(IrCmd::JUMP_CMP_NUM, opA, opB, trueBlock, falseBlock);
build.beginBlock(trueBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
build.beginBlock(falseBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
updateUseCounts(build.function);
optimizeMemoryOperandsX64(build.function);
// Load from memory is 'inlined' into first argument
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
bb_0:
%1 = LOAD_DOUBLE R2
JUMP_CMP_NUM R1, %1, bb_1, bb_2
bb_1:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
bb_2:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/563 (#833) * Fix a bug where reading a property from an unsealed table caused inference to improperly infer the existence of that property. * Fix #827 We have also made a lot of progress on the new solver and the JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. We are mostly working to tighten up how the new solver handles refinements and updates to unsealed tables to bring it up to the same level as the old solver. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-11 03:40:38 +08:00
)");
}
TEST_SUITE_END();
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
TEST_SUITE_BEGIN("ConstantFolding");
TEST_CASE_FIXTURE(IrBuilderFixture, "Numeric")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
build.inst(IrCmd::STORE_INT, build.vmReg(0), build.inst(IrCmd::ADD_INT, build.constInt(10), build.constInt(20)));
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
build.inst(IrCmd::STORE_INT, build.vmReg(1), build.inst(IrCmd::ADD_INT, build.constInt(INT_MAX), build.constInt(1)));
build.inst(IrCmd::STORE_INT, build.vmReg(2), build.inst(IrCmd::SUB_INT, build.constInt(10), build.constInt(20)));
build.inst(IrCmd::STORE_INT, build.vmReg(3), build.inst(IrCmd::SUB_INT, build.constInt(INT_MIN), build.constInt(1)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(4), build.inst(IrCmd::ADD_NUM, build.constDouble(2), build.constDouble(5)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(5), build.inst(IrCmd::SUB_NUM, build.constDouble(2), build.constDouble(5)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(6), build.inst(IrCmd::MUL_NUM, build.constDouble(2), build.constDouble(5)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(7), build.inst(IrCmd::DIV_NUM, build.constDouble(2), build.constDouble(5)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(8), build.inst(IrCmd::MOD_NUM, build.constDouble(5), build.constDouble(2)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(10), build.inst(IrCmd::MIN_NUM, build.constDouble(5), build.constDouble(2)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(11), build.inst(IrCmd::MAX_NUM, build.constDouble(5), build.constDouble(2)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(12), build.inst(IrCmd::UNM_NUM, build.constDouble(5)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(13), build.inst(IrCmd::FLOOR_NUM, build.constDouble(2.5)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(14), build.inst(IrCmd::CEIL_NUM, build.constDouble(2.5)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(15), build.inst(IrCmd::ROUND_NUM, build.constDouble(2.5)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(16), build.inst(IrCmd::SQRT_NUM, build.constDouble(16)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(17), build.inst(IrCmd::ABS_NUM, build.constDouble(-4)));
build.inst(IrCmd::STORE_INT, build.vmReg(18), build.inst(IrCmd::NOT_ANY, build.constTag(tnil), build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(1))));
build.inst(
IrCmd::STORE_INT, build.vmReg(19), build.inst(IrCmd::NOT_ANY, build.constTag(tnumber), build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(1))));
build.inst(IrCmd::STORE_INT, build.vmReg(20), build.inst(IrCmd::NOT_ANY, build.constTag(tboolean), build.constInt(0)));
build.inst(IrCmd::STORE_INT, build.vmReg(21), build.inst(IrCmd::NOT_ANY, build.constTag(tboolean), build.constInt(1)));
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
updateUseCounts(build.function);
constantFold();
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
STORE_INT R0, 30i
STORE_INT R1, -2147483648i
STORE_INT R2, -10i
STORE_INT R3, 2147483647i
STORE_DOUBLE R4, 7
STORE_DOUBLE R5, -3
STORE_DOUBLE R6, 10
STORE_DOUBLE R7, 0.40000000000000002
STORE_DOUBLE R8, 1
STORE_DOUBLE R10, 2
STORE_DOUBLE R11, 5
STORE_DOUBLE R12, -5
STORE_DOUBLE R13, 2
STORE_DOUBLE R14, 3
STORE_DOUBLE R15, 3
STORE_DOUBLE R16, 4
STORE_DOUBLE R17, 4
STORE_INT R18, 1i
STORE_INT R19, 0i
STORE_INT R20, 1i
STORE_INT R21, 0i
RETURN 0u
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
)");
}
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "NumericConversions")
{
IrOp block = build.block(IrBlockKind::Internal);
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
build.beginBlock(block);
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(0), build.inst(IrCmd::INT_TO_NUM, build.constInt(8)));
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(1), build.inst(IrCmd::UINT_TO_NUM, build.constInt(0xdeee0000u)));
build.inst(IrCmd::STORE_INT, build.vmReg(2), build.inst(IrCmd::NUM_TO_INT, build.constDouble(200.0)));
build.inst(IrCmd::STORE_INT, build.vmReg(3), build.inst(IrCmd::NUM_TO_UINT, build.constDouble(3740139520.0)));
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
updateUseCounts(build.function);
constantFold();
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
bb_0:
STORE_DOUBLE R0, 8
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
STORE_DOUBLE R1, 3740139520
STORE_INT R2, 200i
STORE_INT R3, -554827776i
RETURN 0u
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "NumericConversionsBlocked")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
IrOp nan = build.inst(IrCmd::DIV_NUM, build.constDouble(0.0), build.constDouble(0.0));
build.inst(IrCmd::STORE_INT, build.vmReg(0), build.inst(IrCmd::NUM_TO_INT, build.constDouble(1e20)));
build.inst(IrCmd::STORE_INT, build.vmReg(1), build.inst(IrCmd::NUM_TO_UINT, build.constDouble(-10)));
build.inst(IrCmd::STORE_INT, build.vmReg(2), build.inst(IrCmd::NUM_TO_INT, nan));
build.inst(IrCmd::STORE_INT, build.vmReg(3), build.inst(IrCmd::NUM_TO_UINT, nan));
build.inst(IrCmd::RETURN, build.constUint(0));
updateUseCounts(build.function);
constantFold();
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%1 = NUM_TO_INT 1e+20
STORE_INT R0, %1
%3 = NUM_TO_UINT -10
STORE_INT R1, %3
%5 = NUM_TO_INT nan
STORE_INT R2, %5
%7 = NUM_TO_UINT nan
STORE_INT R3, %7
RETURN 0u
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "Bit32")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
IrOp unk = build.inst(IrCmd::LOAD_INT, build.vmReg(0));
build.inst(IrCmd::STORE_INT, build.vmReg(0), build.inst(IrCmd::BITAND_UINT, build.constInt(0xfe), build.constInt(0xe)));
build.inst(IrCmd::STORE_INT, build.vmReg(1), build.inst(IrCmd::BITAND_UINT, unk, build.constInt(0)));
build.inst(IrCmd::STORE_INT, build.vmReg(2), build.inst(IrCmd::BITAND_UINT, build.constInt(0), unk));
build.inst(IrCmd::STORE_INT, build.vmReg(3), build.inst(IrCmd::BITAND_UINT, unk, build.constInt(~0u)));
build.inst(IrCmd::STORE_INT, build.vmReg(4), build.inst(IrCmd::BITAND_UINT, build.constInt(~0u), unk));
build.inst(IrCmd::STORE_INT, build.vmReg(5), build.inst(IrCmd::BITXOR_UINT, build.constInt(0xfe), build.constInt(0xe)));
build.inst(IrCmd::STORE_INT, build.vmReg(6), build.inst(IrCmd::BITXOR_UINT, unk, build.constInt(0)));
build.inst(IrCmd::STORE_INT, build.vmReg(7), build.inst(IrCmd::BITXOR_UINT, build.constInt(0), unk));
build.inst(IrCmd::STORE_INT, build.vmReg(8), build.inst(IrCmd::BITXOR_UINT, unk, build.constInt(~0u)));
build.inst(IrCmd::STORE_INT, build.vmReg(9), build.inst(IrCmd::BITXOR_UINT, build.constInt(~0u), unk));
build.inst(IrCmd::STORE_INT, build.vmReg(10), build.inst(IrCmd::BITOR_UINT, build.constInt(0xf0), build.constInt(0xe)));
build.inst(IrCmd::STORE_INT, build.vmReg(11), build.inst(IrCmd::BITOR_UINT, unk, build.constInt(0)));
build.inst(IrCmd::STORE_INT, build.vmReg(12), build.inst(IrCmd::BITOR_UINT, build.constInt(0), unk));
build.inst(IrCmd::STORE_INT, build.vmReg(13), build.inst(IrCmd::BITOR_UINT, unk, build.constInt(~0u)));
build.inst(IrCmd::STORE_INT, build.vmReg(14), build.inst(IrCmd::BITOR_UINT, build.constInt(~0u), unk));
build.inst(IrCmd::STORE_INT, build.vmReg(15), build.inst(IrCmd::BITNOT_UINT, build.constInt(0xe)));
build.inst(IrCmd::STORE_INT, build.vmReg(16), build.inst(IrCmd::BITLSHIFT_UINT, build.constInt(0xf0), build.constInt(4)));
build.inst(IrCmd::STORE_INT, build.vmReg(17), build.inst(IrCmd::BITLSHIFT_UINT, unk, build.constInt(0)));
build.inst(IrCmd::STORE_INT, build.vmReg(18), build.inst(IrCmd::BITRSHIFT_UINT, build.constInt(0xdeee0000u), build.constInt(8)));
build.inst(IrCmd::STORE_INT, build.vmReg(19), build.inst(IrCmd::BITRSHIFT_UINT, unk, build.constInt(0)));
build.inst(IrCmd::STORE_INT, build.vmReg(20), build.inst(IrCmd::BITARSHIFT_UINT, build.constInt(0xdeee0000u), build.constInt(8)));
build.inst(IrCmd::STORE_INT, build.vmReg(21), build.inst(IrCmd::BITARSHIFT_UINT, unk, build.constInt(0)));
build.inst(IrCmd::STORE_INT, build.vmReg(22), build.inst(IrCmd::BITLROTATE_UINT, build.constInt(0xdeee0000u), build.constInt(8)));
build.inst(IrCmd::STORE_INT, build.vmReg(23), build.inst(IrCmd::BITLROTATE_UINT, unk, build.constInt(0)));
build.inst(IrCmd::STORE_INT, build.vmReg(24), build.inst(IrCmd::BITRROTATE_UINT, build.constInt(0xdeee0000u), build.constInt(8)));
build.inst(IrCmd::STORE_INT, build.vmReg(25), build.inst(IrCmd::BITRROTATE_UINT, unk, build.constInt(0)));
build.inst(IrCmd::STORE_INT, build.vmReg(26), build.inst(IrCmd::BITCOUNTLZ_UINT, build.constInt(0xff00)));
build.inst(IrCmd::STORE_INT, build.vmReg(27), build.inst(IrCmd::BITCOUNTLZ_UINT, build.constInt(0)));
build.inst(IrCmd::STORE_INT, build.vmReg(28), build.inst(IrCmd::BITCOUNTRZ_UINT, build.constInt(0xff00)));
build.inst(IrCmd::STORE_INT, build.vmReg(29), build.inst(IrCmd::BITCOUNTRZ_UINT, build.constInt(0)));
build.inst(IrCmd::RETURN, build.constUint(0));
updateUseCounts(build.function);
constantFold();
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%0 = LOAD_INT R0
STORE_INT R0, 14i
STORE_INT R1, 0i
STORE_INT R2, 0i
STORE_INT R3, %0
STORE_INT R4, %0
STORE_INT R5, 240i
STORE_INT R6, %0
STORE_INT R7, %0
%17 = BITNOT_UINT %0
STORE_INT R8, %17
%19 = BITNOT_UINT %0
STORE_INT R9, %19
STORE_INT R10, 254i
STORE_INT R11, %0
STORE_INT R12, %0
STORE_INT R13, -1i
STORE_INT R14, -1i
STORE_INT R15, -15i
STORE_INT R16, 3840i
STORE_INT R17, %0
STORE_INT R18, 14609920i
STORE_INT R19, %0
STORE_INT R20, -2167296i
STORE_INT R21, %0
STORE_INT R22, -301989666i
STORE_INT R23, %0
STORE_INT R24, 14609920i
STORE_INT R25, %0
STORE_INT R26, 16i
STORE_INT R27, 32i
STORE_INT R28, 8i
STORE_INT R29, 32i
RETURN 0u
)");
}
Sync to upstream/release/583 (#974) * Fixed indexing table intersections using `x["prop"]` syntax: https://github.com/Roblox/luau/pull/971 * Add console output codepage for Windows: https://github.com/Roblox/luau/pull/967 * Added `Frontend::parse` for a fast source graph preparation * luau_load should check GC * Work toward a type-diff system for nicer error messages New Solver * Correctly suppress errors in more cases * Further improvements to typechecking of function calls and return statements * Crash fixes * Propagate refinements drawn from the condition of a while loop into the loop body JIT * Fix accidental bailout for math.frexp/modf/sign in A64 * Work toward bringing type annotation info in * Do not propagate Luau IR constants of wrong type into load instructions * CHECK_SAFEENV exits to VM on failure * Implement error handling in A64 reg allocator * Inline the string.len builtin * Do not enter native code of a function if arguments don’t match --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-07-08 04:10:48 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "Bit32RangeReduction")
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
build.inst(IrCmd::STORE_INT, build.vmReg(10), build.inst(IrCmd::BITLSHIFT_UINT, build.constInt(0xf), build.constInt(-10)));
build.inst(IrCmd::STORE_INT, build.vmReg(10), build.inst(IrCmd::BITLSHIFT_UINT, build.constInt(0xf), build.constInt(140)));
Sync to upstream/release/583 (#974) * Fixed indexing table intersections using `x["prop"]` syntax: https://github.com/Roblox/luau/pull/971 * Add console output codepage for Windows: https://github.com/Roblox/luau/pull/967 * Added `Frontend::parse` for a fast source graph preparation * luau_load should check GC * Work toward a type-diff system for nicer error messages New Solver * Correctly suppress errors in more cases * Further improvements to typechecking of function calls and return statements * Crash fixes * Propagate refinements drawn from the condition of a while loop into the loop body JIT * Fix accidental bailout for math.frexp/modf/sign in A64 * Work toward bringing type annotation info in * Do not propagate Luau IR constants of wrong type into load instructions * CHECK_SAFEENV exits to VM on failure * Implement error handling in A64 reg allocator * Inline the string.len builtin * Do not enter native code of a function if arguments don’t match --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-07-08 04:10:48 +08:00
build.inst(IrCmd::STORE_INT, build.vmReg(10), build.inst(IrCmd::BITRSHIFT_UINT, build.constInt(0xffffff), build.constInt(-10)));
build.inst(IrCmd::STORE_INT, build.vmReg(10), build.inst(IrCmd::BITRSHIFT_UINT, build.constInt(0xffffff), build.constInt(140)));
build.inst(IrCmd::STORE_INT, build.vmReg(10), build.inst(IrCmd::BITARSHIFT_UINT, build.constInt(0xffffff), build.constInt(-10)));
build.inst(IrCmd::STORE_INT, build.vmReg(10), build.inst(IrCmd::BITARSHIFT_UINT, build.constInt(0xffffff), build.constInt(140)));
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
updateUseCounts(build.function);
constantFold();
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
Sync to upstream/release/583 (#974) * Fixed indexing table intersections using `x["prop"]` syntax: https://github.com/Roblox/luau/pull/971 * Add console output codepage for Windows: https://github.com/Roblox/luau/pull/967 * Added `Frontend::parse` for a fast source graph preparation * luau_load should check GC * Work toward a type-diff system for nicer error messages New Solver * Correctly suppress errors in more cases * Further improvements to typechecking of function calls and return statements * Crash fixes * Propagate refinements drawn from the condition of a while loop into the loop body JIT * Fix accidental bailout for math.frexp/modf/sign in A64 * Work toward bringing type annotation info in * Do not propagate Luau IR constants of wrong type into load instructions * CHECK_SAFEENV exits to VM on failure * Implement error handling in A64 reg allocator * Inline the string.len builtin * Do not enter native code of a function if arguments don’t match --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-07-08 04:10:48 +08:00
STORE_INT R10, 62914560i
STORE_INT R10, 61440i
STORE_INT R10, 3i
STORE_INT R10, 4095i
STORE_INT R10, 3i
STORE_INT R10, 4095i
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
RETURN 0u
)");
}
Sync to upstream/release/575 (#919) * `Luau.Analyze.CLI` now has experimental support for concurrent type checking. Use the option `-jN` where `N` is the number of threads to spawn. * Improve typechecking performance by ~17% by making the function `Luau::follow` much more efficient. * Tighten up the type of `os.date` * Removed `ParseOptions::allowTypeAnnotations` and `ParseOptions::supportContinueStatement` New solver * Improve the reliability of function overload resolution * More work toward supporting parallel type checking * Fix a bug in inference of `==` and `~=` which would erroneously infer that the operands were `boolean` * Better error reporting when `for...in` loops are used incorrectly. CodeGen * Fix unwind registration when libunwind is used on Linux * Fixed replaced IR instruction use count * Convert X64 unwind info generation to standard prologue * Implement A64 unwind info support for Dwarf2 * Live in/out data for linear blocks is now created * Add side-exit VM register requirements to the IR dump * Reuse ConstPropState between block chains * Remove redundant base update --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-05-06 05:52:49 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "ReplacementPreservesUses")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
IrOp unk = build.inst(IrCmd::LOAD_INT, build.vmReg(0));
build.inst(IrCmd::STORE_INT, build.vmReg(8), build.inst(IrCmd::BITXOR_UINT, unk, build.constInt(~0u)));
build.inst(IrCmd::RETURN, build.constUint(0));
updateUseCounts(build.function);
constantFold();
CHECK("\n" + toString(build.function, /* includeUseInfo */ true) == R"(
bb_0: ; useCount: 0
%0 = LOAD_INT R0 ; useCount: 1, lastUse: %0
%1 = BITNOT_UINT %0 ; useCount: 1, lastUse: %0
STORE_INT R8, %1 ; %2
RETURN 0u ; %3
)");
}
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "NumericNan")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
IrOp nan = build.inst(IrCmd::DIV_NUM, build.constDouble(0.0), build.constDouble(0.0));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(0), build.inst(IrCmd::MIN_NUM, nan, build.constDouble(2)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(0), build.inst(IrCmd::MIN_NUM, build.constDouble(1), nan));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(0), build.inst(IrCmd::MAX_NUM, nan, build.constDouble(2)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(0), build.inst(IrCmd::MAX_NUM, build.constDouble(1), nan));
build.inst(IrCmd::RETURN, build.constUint(0));
updateUseCounts(build.function);
constantFold();
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
STORE_DOUBLE R0, 2
STORE_DOUBLE R0, nan
STORE_DOUBLE R0, 2
STORE_DOUBLE R0, nan
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "ControlFlowEq")
{
withTwoBlocks([this](IrOp a, IrOp b) {
build.inst(IrCmd::JUMP_EQ_TAG, build.constTag(tnil), build.constTag(tnil), a, b);
});
withTwoBlocks([this](IrOp a, IrOp b) {
build.inst(IrCmd::JUMP_EQ_TAG, build.constTag(tnil), build.constTag(tnumber), a, b);
});
withTwoBlocks([this](IrOp a, IrOp b) {
Sync to upstream/release/594 (#1036) * Fixed `Frontend::markDirty` not working on modules that were not typechecked yet * Fixed generic variadic function unification succeeding when it should have reported an error New Type Solver: * Implemented semantic subtyping check for function types Native Code Generation: * Improved performance of numerical loops with a constant step * Simplified IR for `bit32.extract` calls extracting first/last bits * Improved performance of NaN checks
2023-09-08 08:13:49 +08:00
build.inst(IrCmd::JUMP_CMP_INT, build.constInt(0), build.constInt(0), build.cond(IrCondition::Equal), a, b);
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
});
withTwoBlocks([this](IrOp a, IrOp b) {
Sync to upstream/release/594 (#1036) * Fixed `Frontend::markDirty` not working on modules that were not typechecked yet * Fixed generic variadic function unification succeeding when it should have reported an error New Type Solver: * Implemented semantic subtyping check for function types Native Code Generation: * Improved performance of numerical loops with a constant step * Simplified IR for `bit32.extract` calls extracting first/last bits * Improved performance of NaN checks
2023-09-08 08:13:49 +08:00
build.inst(IrCmd::JUMP_CMP_INT, build.constInt(0), build.constInt(1), build.cond(IrCondition::Equal), a, b);
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
});
updateUseCounts(build.function);
constantFold();
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
bb_0:
JUMP bb_1
bb_1:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 1u
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
bb_3:
JUMP bb_5
bb_5:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 2u
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
bb_6:
JUMP bb_7
bb_7:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 1u
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
bb_9:
JUMP bb_11
bb_11:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 2u
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "NumToIndex")
{
withOneBlock([this](IrOp a) {
Sync to upstream/release/568 (#865) * A small subset of control-flow refinements have been added to recognize type options that are unreachable after a conditional/unconditional code block. (Fixes https://github.com/Roblox/luau/issues/356). Some examples: ```lua local function f(x: string?) if not x then return end -- x is 'string' here end ``` Throwing calls like `error` or `assert(false)` instead of 'return' are also recognized. Existing complex refinements like type/typeof and tagged union checks are expected to work, among others. To enable this feature, `LuauTinyControlFlowAnalysis` exclusion has to be removed from `ExperimentalFlags.h`. If will become enabled unconditionally in the near future. * Linter has been integrated into the typechecker analysis so that type-aware lint warnings can work in any mode `Frontend::lint` methods were deprecated, `Frontend::check` has to be used instead with `runLintChecks` option set. Resulting lint warning are located inside `CheckResult`. * Fixed large performance drop and increased memory consumption when array is filled at an offset (Fixes https://github.com/Roblox/luau/issues/590) * Part of [Type error suppression RFC](https://github.com/Roblox/luau/blob/master/rfcs/type-error-suppression.md) was implemented making subtyping checks with `any` type transitive. --- In our work on the new type-solver: * `--!nocheck` mode no longer reports type errors * New solver will not be used for `--!nonstrict` modules until all issues with strict mode typechecking are fixed * Added control-flow aware type refinements mentioned earlier In native code generation: * `LOP_NAMECALL` has been translated to IR * `type` and `typeof` builtin fastcalls have been translated to IR/assembly * Additional steps were taken towards arm64 support
2023-03-18 03:20:37 +08:00
build.inst(IrCmd::STORE_INT, build.vmReg(0), build.inst(IrCmd::TRY_NUM_TO_INDEX, build.constDouble(4), a));
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
});
withOneBlock([this](IrOp a) {
Sync to upstream/release/568 (#865) * A small subset of control-flow refinements have been added to recognize type options that are unreachable after a conditional/unconditional code block. (Fixes https://github.com/Roblox/luau/issues/356). Some examples: ```lua local function f(x: string?) if not x then return end -- x is 'string' here end ``` Throwing calls like `error` or `assert(false)` instead of 'return' are also recognized. Existing complex refinements like type/typeof and tagged union checks are expected to work, among others. To enable this feature, `LuauTinyControlFlowAnalysis` exclusion has to be removed from `ExperimentalFlags.h`. If will become enabled unconditionally in the near future. * Linter has been integrated into the typechecker analysis so that type-aware lint warnings can work in any mode `Frontend::lint` methods were deprecated, `Frontend::check` has to be used instead with `runLintChecks` option set. Resulting lint warning are located inside `CheckResult`. * Fixed large performance drop and increased memory consumption when array is filled at an offset (Fixes https://github.com/Roblox/luau/issues/590) * Part of [Type error suppression RFC](https://github.com/Roblox/luau/blob/master/rfcs/type-error-suppression.md) was implemented making subtyping checks with `any` type transitive. --- In our work on the new type-solver: * `--!nocheck` mode no longer reports type errors * New solver will not be used for `--!nonstrict` modules until all issues with strict mode typechecking are fixed * Added control-flow aware type refinements mentioned earlier In native code generation: * `LOP_NAMECALL` has been translated to IR * `type` and `typeof` builtin fastcalls have been translated to IR/assembly * Additional steps were taken towards arm64 support
2023-03-18 03:20:37 +08:00
build.inst(IrCmd::STORE_INT, build.vmReg(0), build.inst(IrCmd::TRY_NUM_TO_INDEX, build.constDouble(1.2), a));
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
});
withOneBlock([this](IrOp a) {
IrOp nan = build.inst(IrCmd::DIV_NUM, build.constDouble(0.0), build.constDouble(0.0));
Sync to upstream/release/568 (#865) * A small subset of control-flow refinements have been added to recognize type options that are unreachable after a conditional/unconditional code block. (Fixes https://github.com/Roblox/luau/issues/356). Some examples: ```lua local function f(x: string?) if not x then return end -- x is 'string' here end ``` Throwing calls like `error` or `assert(false)` instead of 'return' are also recognized. Existing complex refinements like type/typeof and tagged union checks are expected to work, among others. To enable this feature, `LuauTinyControlFlowAnalysis` exclusion has to be removed from `ExperimentalFlags.h`. If will become enabled unconditionally in the near future. * Linter has been integrated into the typechecker analysis so that type-aware lint warnings can work in any mode `Frontend::lint` methods were deprecated, `Frontend::check` has to be used instead with `runLintChecks` option set. Resulting lint warning are located inside `CheckResult`. * Fixed large performance drop and increased memory consumption when array is filled at an offset (Fixes https://github.com/Roblox/luau/issues/590) * Part of [Type error suppression RFC](https://github.com/Roblox/luau/blob/master/rfcs/type-error-suppression.md) was implemented making subtyping checks with `any` type transitive. --- In our work on the new type-solver: * `--!nocheck` mode no longer reports type errors * New solver will not be used for `--!nonstrict` modules until all issues with strict mode typechecking are fixed * Added control-flow aware type refinements mentioned earlier In native code generation: * `LOP_NAMECALL` has been translated to IR * `type` and `typeof` builtin fastcalls have been translated to IR/assembly * Additional steps were taken towards arm64 support
2023-03-18 03:20:37 +08:00
build.inst(IrCmd::STORE_INT, build.vmReg(0), build.inst(IrCmd::TRY_NUM_TO_INDEX, nan, a));
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
});
updateUseCounts(build.function);
constantFold();
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
bb_0:
STORE_INT R0, 4i
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
bb_2:
JUMP bb_3
bb_3:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 1u
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
bb_4:
JUMP bb_5
bb_5:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 1u
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "Guards")
{
withOneBlock([this](IrOp a) {
build.inst(IrCmd::CHECK_TAG, build.constTag(tnumber), build.constTag(tnumber), a);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
});
withOneBlock([this](IrOp a) {
build.inst(IrCmd::CHECK_TAG, build.constTag(tnil), build.constTag(tnumber), a);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
});
updateUseCounts(build.function);
constantFold();
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
bb_0:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
bb_2:
JUMP bb_3
bb_3:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 1u
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "ControlFlowCmpNum")
{
auto compareFold = [this](IrOp lhs, IrOp rhs, IrCondition cond, bool result) {
IrOp instOp;
IrInst instExpected;
withTwoBlocks([&](IrOp a, IrOp b) {
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
IrOp nan = build.inst(IrCmd::DIV_NUM, build.constDouble(0.0), build.constDouble(0.0));
instOp = build.inst(
IrCmd::JUMP_CMP_NUM, lhs.kind == IrOpKind::None ? nan : lhs, rhs.kind == IrOpKind::None ? nan : rhs, build.cond(cond), a, b);
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
instExpected = IrInst{IrCmd::JUMP, result ? a : b};
});
updateUseCounts(build.function);
constantFold();
checkEq(instOp, instExpected);
};
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
IrOp nan; // Empty operand is used to signal a placement of a 'nan'
Sync to upstream/release/564 (#841) This week we only have updates to new type solver and JIT. Both projects are still in the process of being built out. Neither are ready for general use yet. In the new solver, we fixed issues with recursive type aliases. Duplicated type parameters are once again reported, exported types are being recorder and function argument names are placed inside function types. We also made improvements to restore parts of bidirectional type tracking. On native code generation side, namecall instruction lowering was fixed, we fixed inconsistencies in IR command definitions and added utility function to help with constant folding.
2023-02-18 07:41:51 +08:00
compareFold(build.constDouble(1), build.constDouble(1), IrCondition::Equal, true);
compareFold(build.constDouble(1), build.constDouble(2), IrCondition::Equal, false);
compareFold(nan, nan, IrCondition::Equal, false);
compareFold(build.constDouble(1), build.constDouble(1), IrCondition::NotEqual, false);
compareFold(build.constDouble(1), build.constDouble(2), IrCondition::NotEqual, true);
compareFold(nan, nan, IrCondition::NotEqual, true);
compareFold(build.constDouble(1), build.constDouble(1), IrCondition::Less, false);
compareFold(build.constDouble(1), build.constDouble(2), IrCondition::Less, true);
compareFold(build.constDouble(2), build.constDouble(1), IrCondition::Less, false);
compareFold(build.constDouble(1), nan, IrCondition::Less, false);
compareFold(build.constDouble(1), build.constDouble(1), IrCondition::NotLess, true);
compareFold(build.constDouble(1), build.constDouble(2), IrCondition::NotLess, false);
compareFold(build.constDouble(2), build.constDouble(1), IrCondition::NotLess, true);
compareFold(build.constDouble(1), nan, IrCondition::NotLess, true);
compareFold(build.constDouble(1), build.constDouble(1), IrCondition::LessEqual, true);
compareFold(build.constDouble(1), build.constDouble(2), IrCondition::LessEqual, true);
compareFold(build.constDouble(2), build.constDouble(1), IrCondition::LessEqual, false);
compareFold(build.constDouble(1), nan, IrCondition::LessEqual, false);
compareFold(build.constDouble(1), build.constDouble(1), IrCondition::NotLessEqual, false);
compareFold(build.constDouble(1), build.constDouble(2), IrCondition::NotLessEqual, false);
compareFold(build.constDouble(2), build.constDouble(1), IrCondition::NotLessEqual, true);
compareFold(build.constDouble(1), nan, IrCondition::NotLessEqual, true);
compareFold(build.constDouble(1), build.constDouble(1), IrCondition::Greater, false);
compareFold(build.constDouble(1), build.constDouble(2), IrCondition::Greater, false);
compareFold(build.constDouble(2), build.constDouble(1), IrCondition::Greater, true);
compareFold(build.constDouble(1), nan, IrCondition::Greater, false);
compareFold(build.constDouble(1), build.constDouble(1), IrCondition::NotGreater, true);
compareFold(build.constDouble(1), build.constDouble(2), IrCondition::NotGreater, true);
compareFold(build.constDouble(2), build.constDouble(1), IrCondition::NotGreater, false);
compareFold(build.constDouble(1), nan, IrCondition::NotGreater, true);
compareFold(build.constDouble(1), build.constDouble(1), IrCondition::GreaterEqual, true);
compareFold(build.constDouble(1), build.constDouble(2), IrCondition::GreaterEqual, false);
compareFold(build.constDouble(2), build.constDouble(1), IrCondition::GreaterEqual, true);
compareFold(build.constDouble(1), nan, IrCondition::GreaterEqual, false);
compareFold(build.constDouble(1), build.constDouble(1), IrCondition::NotGreaterEqual, false);
compareFold(build.constDouble(1), build.constDouble(2), IrCondition::NotGreaterEqual, true);
compareFold(build.constDouble(2), build.constDouble(1), IrCondition::NotGreaterEqual, false);
compareFold(build.constDouble(1), nan, IrCondition::NotGreaterEqual, true);
}
TEST_SUITE_END();
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
TEST_SUITE_BEGIN("ConstantPropagation");
TEST_CASE_FIXTURE(IrBuilderFixture, "RememberTagsAndValues")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.constTag(tnumber));
build.inst(IrCmd::STORE_INT, build.vmReg(1), build.constInt(10));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(2), build.constDouble(0.5));
// We know constants from those loads
build.inst(IrCmd::STORE_TAG, build.vmReg(3), build.inst(IrCmd::LOAD_TAG, build.vmReg(0)));
build.inst(IrCmd::STORE_INT, build.vmReg(4), build.inst(IrCmd::LOAD_INT, build.vmReg(1)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(5), build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(2)));
// We know that these overrides have no effect
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.constTag(tnumber));
build.inst(IrCmd::STORE_INT, build.vmReg(1), build.constInt(10));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(2), build.constDouble(0.5));
// But we can invalidate them with unknown values
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.inst(IrCmd::LOAD_TAG, build.vmReg(6)));
build.inst(IrCmd::STORE_INT, build.vmReg(1), build.inst(IrCmd::LOAD_INT, build.vmReg(7)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(2), build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(8)));
// So now the constant stores have to be made
build.inst(IrCmd::STORE_TAG, build.vmReg(9), build.inst(IrCmd::LOAD_TAG, build.vmReg(0)));
build.inst(IrCmd::STORE_INT, build.vmReg(10), build.inst(IrCmd::LOAD_INT, build.vmReg(1)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(11), build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(2)));
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
STORE_TAG R0, tnumber
STORE_INT R1, 10i
STORE_DOUBLE R2, 0.5
STORE_TAG R3, tnumber
STORE_INT R4, 10i
STORE_DOUBLE R5, 0.5
%12 = LOAD_TAG R6
STORE_TAG R0, %12
%14 = LOAD_INT R7
STORE_INT R1, %14
%16 = LOAD_DOUBLE R8
STORE_DOUBLE R2, %16
%18 = LOAD_TAG R0
STORE_TAG R9, %18
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
STORE_INT R10, %14
STORE_DOUBLE R11, %16
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "PropagateThroughTvalue")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.constTag(tnumber));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(0), build.constDouble(0.5));
IrOp tv = build.inst(IrCmd::LOAD_TVALUE, build.vmReg(0));
build.inst(IrCmd::STORE_TVALUE, build.vmReg(1), tv);
// We know constants from those loads
build.inst(IrCmd::STORE_TAG, build.vmReg(3), build.inst(IrCmd::LOAD_TAG, build.vmReg(1)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(3), build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(1)));
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
STORE_TAG R0, tnumber
STORE_DOUBLE R0, 0.5
Sync to upstream/release/591 (#1012) * Fix a use-after-free bug in the new type cloning algorithm * Tighten up the type of `coroutine.wrap`. It is now `<A..., R...>(f: (A...) -> R...) -> ((A...) -> R...)` * Break `.luaurc` out into a separate library target `Luau.Config`. This makes it easier for applications to reason about config files without also depending on the type inference engine. * Move typechecking limits into `FrontendOptions`. This allows embedders more finely-grained control over autocomplete's internal time limits. * Fix stability issue with debugger onprotectederror callback allowing break in non-yieldable contexts New solver: * Initial work toward [Local Type Inference](https://github.com/Roblox/luau/blob/0e1082108fd6fb3a32dfdf5f1766ea3fc1391328/rfcs/local-type-inference.md) * Introduce a new subtyping test. This will be much nicer than the old test because it is completely separate both from actual type inference and from error reporting. Native code generation: * Added function to compute iterated dominance frontier * Optimize barriers in SET_UPVALUE when tag is known * Cache lua_State::global in a register on A64 * Optimize constant stores in A64 lowering * Track table array size state to optimize array size checks * Add split tag/value store into a VM register * Check that spills can outlive the block only in specific conditions --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-08-19 02:15:41 +08:00
STORE_SPLIT_TVALUE R1, tnumber, 0.5
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
STORE_TAG R3, tnumber
STORE_DOUBLE R3, 0.5
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "SkipCheckTag")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp fallback = build.block(IrBlockKind::Fallback);
build.beginBlock(block);
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.constTag(tnumber));
build.inst(IrCmd::CHECK_TAG, build.inst(IrCmd::LOAD_TAG, build.vmReg(0)), build.constTag(tnumber), fallback);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.beginBlock(fallback);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(1));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
STORE_TAG R0, tnumber
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "SkipOncePerBlockChecks")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
build.inst(IrCmd::CHECK_SAFE_ENV);
build.inst(IrCmd::CHECK_SAFE_ENV);
build.inst(IrCmd::CHECK_GC);
build.inst(IrCmd::CHECK_GC);
build.inst(IrCmd::DO_LEN, build.vmReg(1), build.vmReg(2)); // Can make env unsafe
build.inst(IrCmd::CHECK_SAFE_ENV);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
CHECK_SAFE_ENV
CHECK_GC
DO_LEN R1, R2
CHECK_SAFE_ENV
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "RememberTableState")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp fallback = build.block(IrBlockKind::Fallback);
build.beginBlock(block);
IrOp table = build.inst(IrCmd::LOAD_POINTER, build.vmReg(0));
build.inst(IrCmd::CHECK_NO_METATABLE, table, fallback);
build.inst(IrCmd::CHECK_READONLY, table, fallback);
build.inst(IrCmd::CHECK_NO_METATABLE, table, fallback);
build.inst(IrCmd::CHECK_READONLY, table, fallback);
build.inst(IrCmd::DO_LEN, build.vmReg(1), build.vmReg(2)); // Can access all heap memory
build.inst(IrCmd::CHECK_NO_METATABLE, table, fallback);
build.inst(IrCmd::CHECK_READONLY, table, fallback);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.beginBlock(fallback);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(1));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
%0 = LOAD_POINTER R0
CHECK_NO_METATABLE %0, bb_fallback_1
CHECK_READONLY %0, bb_fallback_1
DO_LEN R1, R2
CHECK_NO_METATABLE %0, bb_fallback_1
CHECK_READONLY %0, bb_fallback_1
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_fallback_1:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 1u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
Sync to upstream/release/591 (#1012) * Fix a use-after-free bug in the new type cloning algorithm * Tighten up the type of `coroutine.wrap`. It is now `<A..., R...>(f: (A...) -> R...) -> ((A...) -> R...)` * Break `.luaurc` out into a separate library target `Luau.Config`. This makes it easier for applications to reason about config files without also depending on the type inference engine. * Move typechecking limits into `FrontendOptions`. This allows embedders more finely-grained control over autocomplete's internal time limits. * Fix stability issue with debugger onprotectederror callback allowing break in non-yieldable contexts New solver: * Initial work toward [Local Type Inference](https://github.com/Roblox/luau/blob/0e1082108fd6fb3a32dfdf5f1766ea3fc1391328/rfcs/local-type-inference.md) * Introduce a new subtyping test. This will be much nicer than the old test because it is completely separate both from actual type inference and from error reporting. Native code generation: * Added function to compute iterated dominance frontier * Optimize barriers in SET_UPVALUE when tag is known * Cache lua_State::global in a register on A64 * Optimize constant stores in A64 lowering * Track table array size state to optimize array size checks * Add split tag/value store into a VM register * Check that spills can outlive the block only in specific conditions --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-08-19 02:15:41 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "RememberNewTableState")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp fallback = build.block(IrBlockKind::Fallback);
build.beginBlock(block);
IrOp newtable = build.inst(IrCmd::NEW_TABLE, build.constUint(16), build.constUint(32));
build.inst(IrCmd::STORE_POINTER, build.vmReg(0), newtable);
IrOp table = build.inst(IrCmd::LOAD_POINTER, build.vmReg(0));
build.inst(IrCmd::CHECK_NO_METATABLE, table, fallback);
build.inst(IrCmd::CHECK_READONLY, table, fallback);
build.inst(IrCmd::CHECK_ARRAY_SIZE, table, build.constInt(14), fallback);
build.inst(IrCmd::SET_TABLE, build.vmReg(1), build.vmReg(0), build.constUint(13)); // Invalidate table knowledge
build.inst(IrCmd::CHECK_NO_METATABLE, table, fallback);
build.inst(IrCmd::CHECK_READONLY, table, fallback);
build.inst(IrCmd::CHECK_ARRAY_SIZE, table, build.constInt(14), fallback);
build.inst(IrCmd::RETURN, build.constUint(0));
build.beginBlock(fallback);
build.inst(IrCmd::RETURN, build.constUint(1));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%0 = NEW_TABLE 16u, 32u
STORE_POINTER R0, %0
SET_TABLE R1, R0, 13u
CHECK_NO_METATABLE %0, bb_fallback_1
CHECK_READONLY %0, bb_fallback_1
CHECK_ARRAY_SIZE %0, 14i, bb_fallback_1
RETURN 0u
bb_fallback_1:
RETURN 1u
)");
}
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "SkipUselessBarriers")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.constTag(tnumber));
IrOp table = build.inst(IrCmd::LOAD_POINTER, build.vmReg(1));
Sync to upstream/release/582 (#960) * Optimized operations like instantiation and module export for very large types In our new typechecker: * Typechecking of function calls was rewritten to handle more cases correctly * Fixed a crash that can happen after self-referential type is exported from a module * Fixed a false positive error in string comparison * Added handling of `for...in` variable type annotations and fixed issues with the iterator call inside * Self-referential 'hasProp' and 'setProp' constraints are now handled correctly In our native code generation (jit): * Added '--target' argument to luau-compile to test multiple architectures different from host architecture * GC barrier tag check is skipped if type is already known to be GC-collectable * Added GET_TYPE/GET_TYPEOF instructions for type/typeof fast-calls * Improved code size of interrupt handlers on X64
2023-06-24 14:19:39 +08:00
build.inst(IrCmd::BARRIER_TABLE_FORWARD, table, build.vmReg(0), build.undef());
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
IrOp something = build.inst(IrCmd::LOAD_POINTER, build.vmReg(2));
Sync to upstream/release/582 (#960) * Optimized operations like instantiation and module export for very large types In our new typechecker: * Typechecking of function calls was rewritten to handle more cases correctly * Fixed a crash that can happen after self-referential type is exported from a module * Fixed a false positive error in string comparison * Added handling of `for...in` variable type annotations and fixed issues with the iterator call inside * Self-referential 'hasProp' and 'setProp' constraints are now handled correctly In our native code generation (jit): * Added '--target' argument to luau-compile to test multiple architectures different from host architecture * GC barrier tag check is skipped if type is already known to be GC-collectable * Added GET_TYPE/GET_TYPEOF instructions for type/typeof fast-calls * Improved code size of interrupt handlers on X64
2023-06-24 14:19:39 +08:00
build.inst(IrCmd::BARRIER_OBJ, something, build.vmReg(0), build.undef());
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
STORE_TAG R0, tnumber
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "ConcatInvalidation")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.constTag(tnumber));
build.inst(IrCmd::STORE_INT, build.vmReg(1), build.constInt(10));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(2), build.constDouble(0.5));
Sync to upstream/release/569 (#878) All of our changes this week have been focused on the new type solver and the JIT. As we march toward feature parity with the old solver, we've tightened up a bunch of lingering issues with overload resolution, unsealed tables, and type normalization. We've also fixed a bunch of crashes and assertion failures in the new solver. On the JIT front, we've started work on an A64 backend, improved the IR analysis in a bunch of cases, and implemented assembly generation for the builtin functions `type()` and `typeof()`. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-25 02:03:04 +08:00
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(3), build.constDouble(2.0));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/569 (#878) All of our changes this week have been focused on the new type solver and the JIT. As we march toward feature parity with the old solver, we've tightened up a bunch of lingering issues with overload resolution, unsealed tables, and type normalization. We've also fixed a bunch of crashes and assertion failures in the new solver. On the JIT front, we've started work on an A64 backend, improved the IR analysis in a bunch of cases, and implemented assembly generation for the builtin functions `type()` and `typeof()`. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-25 02:03:04 +08:00
build.inst(IrCmd::CONCAT, build.vmReg(0), build.constUint(3));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/569 (#878) All of our changes this week have been focused on the new type solver and the JIT. As we march toward feature parity with the old solver, we've tightened up a bunch of lingering issues with overload resolution, unsealed tables, and type normalization. We've also fixed a bunch of crashes and assertion failures in the new solver. On the JIT front, we've started work on an A64 backend, improved the IR analysis in a bunch of cases, and implemented assembly generation for the builtin functions `type()` and `typeof()`. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-25 02:03:04 +08:00
build.inst(IrCmd::STORE_TAG, build.vmReg(4), build.inst(IrCmd::LOAD_TAG, build.vmReg(0)));
build.inst(IrCmd::STORE_INT, build.vmReg(5), build.inst(IrCmd::LOAD_INT, build.vmReg(1)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(6), build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(2)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(7), build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(3)));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
STORE_TAG R0, tnumber
STORE_INT R1, 10i
STORE_DOUBLE R2, 0.5
Sync to upstream/release/569 (#878) All of our changes this week have been focused on the new type solver and the JIT. As we march toward feature parity with the old solver, we've tightened up a bunch of lingering issues with overload resolution, unsealed tables, and type normalization. We've also fixed a bunch of crashes and assertion failures in the new solver. On the JIT front, we've started work on an A64 backend, improved the IR analysis in a bunch of cases, and implemented assembly generation for the builtin functions `type()` and `typeof()`. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-25 02:03:04 +08:00
STORE_DOUBLE R3, 2
CONCAT R0, 3u
%5 = LOAD_TAG R0
STORE_TAG R4, %5
%7 = LOAD_INT R1
STORE_INT R5, %7
%9 = LOAD_DOUBLE R2
STORE_DOUBLE R6, %9
STORE_DOUBLE R7, 2
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "BuiltinFastcallsMayInvalidateMemory")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp fallback = build.block(IrBlockKind::Fallback);
build.beginBlock(block);
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(0), build.constDouble(0.5));
IrOp table = build.inst(IrCmd::LOAD_POINTER, build.vmReg(0));
build.inst(IrCmd::CHECK_NO_METATABLE, table, fallback);
build.inst(IrCmd::CHECK_READONLY, table, fallback);
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
build.inst(IrCmd::INVOKE_FASTCALL, build.constUint(LBF_SETMETATABLE), build.vmReg(1), build.vmReg(2), build.vmReg(3), build.constInt(3),
build.constInt(1));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.inst(IrCmd::CHECK_NO_METATABLE, table, fallback);
build.inst(IrCmd::CHECK_READONLY, table, fallback);
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(1), build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0))); // At least R0 wasn't touched
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.beginBlock(fallback);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(1));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
STORE_DOUBLE R0, 0.5
%1 = LOAD_POINTER R0
CHECK_NO_METATABLE %1, bb_fallback_1
CHECK_READONLY %1, bb_fallback_1
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
%4 = INVOKE_FASTCALL 61u, R1, R2, R3, 3i, 1i
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
CHECK_NO_METATABLE %1, bb_fallback_1
CHECK_READONLY %1, bb_fallback_1
STORE_DOUBLE R1, 0.5
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_fallback_1:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 1u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "RedundantStoreCheckConstantType")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
build.inst(IrCmd::STORE_INT, build.vmReg(0), build.constInt(10));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(0), build.constDouble(0.5));
build.inst(IrCmd::STORE_INT, build.vmReg(0), build.constInt(10));
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
STORE_INT R0, 10i
STORE_DOUBLE R0, 0.5
STORE_INT R0, 10i
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "TagCheckPropagation")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp fallback = build.block(IrBlockKind::Fallback);
build.beginBlock(block);
IrOp unknown = build.inst(IrCmd::LOAD_TAG, build.vmReg(0));
build.inst(IrCmd::CHECK_TAG, unknown, build.constTag(tnumber), fallback);
build.inst(IrCmd::CHECK_TAG, unknown, build.constTag(tnumber), fallback);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.beginBlock(fallback);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(1));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
%0 = LOAD_TAG R0
CHECK_TAG %0, tnumber, bb_fallback_1
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 0u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_fallback_1:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 1u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "TagCheckPropagationConflicting")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp fallback = build.block(IrBlockKind::Fallback);
build.beginBlock(block);
IrOp unknown = build.inst(IrCmd::LOAD_TAG, build.vmReg(0));
build.inst(IrCmd::CHECK_TAG, unknown, build.constTag(tnumber), fallback);
build.inst(IrCmd::CHECK_TAG, unknown, build.constTag(tnil), fallback);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.beginBlock(fallback);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(1));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
%0 = LOAD_TAG R0
CHECK_TAG %0, tnumber, bb_fallback_1
JUMP bb_fallback_1
bb_fallback_1:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 1u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "TruthyTestRemoval")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp trueBlock = build.block(IrBlockKind::Internal);
IrOp falseBlock = build.block(IrBlockKind::Internal);
IrOp fallback = build.block(IrBlockKind::Fallback);
build.beginBlock(block);
IrOp unknown = build.inst(IrCmd::LOAD_TAG, build.vmReg(1));
build.inst(IrCmd::CHECK_TAG, unknown, build.constTag(tnumber), fallback);
build.inst(IrCmd::JUMP_IF_TRUTHY, build.vmReg(1), trueBlock, falseBlock);
build.beginBlock(trueBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(1));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.beginBlock(falseBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(2));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.beginBlock(fallback);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(3));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
%0 = LOAD_TAG R1
CHECK_TAG %0, tnumber, bb_fallback_3
JUMP bb_1
bb_1:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 1u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_fallback_3:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 3u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "FalsyTestRemoval")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp trueBlock = build.block(IrBlockKind::Internal);
IrOp falseBlock = build.block(IrBlockKind::Internal);
IrOp fallback = build.block(IrBlockKind::Fallback);
build.beginBlock(block);
IrOp unknown = build.inst(IrCmd::LOAD_TAG, build.vmReg(1));
build.inst(IrCmd::CHECK_TAG, unknown, build.constTag(tnumber), fallback);
build.inst(IrCmd::JUMP_IF_FALSY, build.vmReg(1), trueBlock, falseBlock);
build.beginBlock(trueBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(1));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.beginBlock(falseBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(2));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.beginBlock(fallback);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(3));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
%0 = LOAD_TAG R1
CHECK_TAG %0, tnumber, bb_fallback_3
JUMP bb_2
bb_2:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 2u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_fallback_3:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 3u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "TagEqRemoval")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp trueBlock = build.block(IrBlockKind::Internal);
IrOp falseBlock = build.block(IrBlockKind::Internal);
build.beginBlock(block);
IrOp tag = build.inst(IrCmd::LOAD_TAG, build.vmReg(1));
build.inst(IrCmd::CHECK_TAG, tag, build.constTag(tboolean));
build.inst(IrCmd::JUMP_EQ_TAG, tag, build.constTag(tnumber), trueBlock, falseBlock);
build.beginBlock(trueBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(1));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.beginBlock(falseBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(2));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
%0 = LOAD_TAG R1
CHECK_TAG %0, tboolean
JUMP bb_2
bb_2:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 2u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "IntEqRemoval")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp trueBlock = build.block(IrBlockKind::Internal);
IrOp falseBlock = build.block(IrBlockKind::Internal);
build.beginBlock(block);
build.inst(IrCmd::STORE_INT, build.vmReg(1), build.constInt(5));
Sync to upstream/release/578 (#939) * Fixed gcc warning about uninitialized `std::optional` * Fixed inlining of functions when they are used to compute their own arguments In the new type solver: * Type families that are not part of a function signature cannot be resolved at instantiation time and will now produce an error. This will be relaxed in the future when we get constraint clauses on function signatures (internally) * `never` type is now comparable * Improved typechecking of `for..in` statements * Fixed checks for number type in `Add` type family * Performance was improved, with particularly large gains on large projects And in native code generation (jit): * We eliminated the call instruction overhead when native code support is enabled in the VM * Small optimizations to arm64 lowering * Reworked LOP_GETIMPORT handling to reduce assembly code size * Fixed non-deterministic binary output * Fixed bad code generation caused by incorrect SSA to VM register links invalidation --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Andy Friesen <afriesen@roblox.com>
2023-05-26 05:36:34 +08:00
IrOp value = build.inst(IrCmd::LOAD_INT, build.vmReg(1));
Sync to upstream/release/594 (#1036) * Fixed `Frontend::markDirty` not working on modules that were not typechecked yet * Fixed generic variadic function unification succeeding when it should have reported an error New Type Solver: * Implemented semantic subtyping check for function types Native Code Generation: * Improved performance of numerical loops with a constant step * Simplified IR for `bit32.extract` calls extracting first/last bits * Improved performance of NaN checks
2023-09-08 08:13:49 +08:00
build.inst(IrCmd::JUMP_CMP_INT, value, build.constInt(5), build.cond(IrCondition::Equal), trueBlock, falseBlock);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.beginBlock(trueBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(1));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.beginBlock(falseBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(2));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
STORE_INT R1, 5i
JUMP bb_1
bb_1:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 1u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "NumCmpRemoval")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp trueBlock = build.block(IrBlockKind::Internal);
IrOp falseBlock = build.block(IrBlockKind::Internal);
build.beginBlock(block);
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(1), build.constDouble(4.0));
Sync to upstream/release/578 (#939) * Fixed gcc warning about uninitialized `std::optional` * Fixed inlining of functions when they are used to compute their own arguments In the new type solver: * Type families that are not part of a function signature cannot be resolved at instantiation time and will now produce an error. This will be relaxed in the future when we get constraint clauses on function signatures (internally) * `never` type is now comparable * Improved typechecking of `for..in` statements * Fixed checks for number type in `Add` type family * Performance was improved, with particularly large gains on large projects And in native code generation (jit): * We eliminated the call instruction overhead when native code support is enabled in the VM * Small optimizations to arm64 lowering * Reworked LOP_GETIMPORT handling to reduce assembly code size * Fixed non-deterministic binary output * Fixed bad code generation caused by incorrect SSA to VM register links invalidation --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Andy Friesen <afriesen@roblox.com>
2023-05-26 05:36:34 +08:00
IrOp value = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(1));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.inst(IrCmd::JUMP_CMP_NUM, value, build.constDouble(8.0), build.cond(IrCondition::Greater), trueBlock, falseBlock);
build.beginBlock(trueBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(1));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.beginBlock(falseBlock);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(2));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
STORE_DOUBLE R1, 4
JUMP bb_2
bb_2:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 2u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "DataFlowsThroughDirectJumpToUniqueSuccessor")
{
IrOp block1 = build.block(IrBlockKind::Internal);
IrOp block2 = build.block(IrBlockKind::Internal);
build.beginBlock(block1);
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.constTag(tnumber));
build.inst(IrCmd::JUMP, block2);
build.beginBlock(block2);
build.inst(IrCmd::STORE_TAG, build.vmReg(1), build.inst(IrCmd::LOAD_TAG, build.vmReg(0)));
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(1));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
STORE_TAG R0, tnumber
JUMP bb_1
bb_1:
STORE_TAG R1, tnumber
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 1u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "DataDoesNotFlowThroughDirectJumpToNonUniqueSuccessor")
{
IrOp block1 = build.block(IrBlockKind::Internal);
IrOp block2 = build.block(IrBlockKind::Internal);
IrOp block3 = build.block(IrBlockKind::Internal);
build.beginBlock(block1);
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.constTag(tnumber));
build.inst(IrCmd::JUMP, block2);
build.beginBlock(block2);
build.inst(IrCmd::STORE_TAG, build.vmReg(1), build.inst(IrCmd::LOAD_TAG, build.vmReg(0)));
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.constUint(1));
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
build.beginBlock(block3);
build.inst(IrCmd::JUMP, block2);
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_0:
STORE_TAG R0, tnumber
JUMP bb_1
bb_1:
%2 = LOAD_TAG R0
STORE_TAG R1, %2
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN 1u
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
bb_2:
JUMP bb_1
)");
}
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "EntryBlockUseRemoval")
{
IrOp entry = build.block(IrBlockKind::Internal);
IrOp exit = build.block(IrBlockKind::Internal);
IrOp repeat = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.constTag(tnumber));
build.inst(IrCmd::JUMP_IF_TRUTHY, build.vmReg(0), exit, repeat);
build.beginBlock(exit);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(0));
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
build.beginBlock(repeat);
build.inst(IrCmd::INTERRUPT, build.constUint(0));
build.inst(IrCmd::JUMP, entry);
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
STORE_TAG R0, tnumber
JUMP bb_1
bb_1:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R0, 0i
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "RecursiveSccUseRemoval1")
{
IrOp entry = build.block(IrBlockKind::Internal);
IrOp block = build.block(IrBlockKind::Internal);
IrOp exit = build.block(IrBlockKind::Internal);
IrOp repeat = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(0));
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
build.beginBlock(block);
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.constTag(tnumber));
build.inst(IrCmd::JUMP_IF_TRUTHY, build.vmReg(0), exit, repeat);
build.beginBlock(exit);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(0));
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
build.beginBlock(repeat);
build.inst(IrCmd::INTERRUPT, build.constUint(0));
build.inst(IrCmd::JUMP, block);
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R0, 0i
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
bb_1:
STORE_TAG R0, tnumber
JUMP bb_2
bb_2:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R0, 0i
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "RecursiveSccUseRemoval2")
{
IrOp entry = build.block(IrBlockKind::Internal);
IrOp exit1 = build.block(IrBlockKind::Internal);
IrOp block = build.block(IrBlockKind::Internal);
IrOp exit2 = build.block(IrBlockKind::Internal);
IrOp repeat = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
Sync to upstream/release/594 (#1036) * Fixed `Frontend::markDirty` not working on modules that were not typechecked yet * Fixed generic variadic function unification succeeding when it should have reported an error New Type Solver: * Implemented semantic subtyping check for function types Native Code Generation: * Improved performance of numerical loops with a constant step * Simplified IR for `bit32.extract` calls extracting first/last bits * Improved performance of NaN checks
2023-09-08 08:13:49 +08:00
build.inst(IrCmd::JUMP_CMP_INT, build.constInt(0), build.constInt(1), build.cond(IrCondition::Equal), block, exit1);
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
build.beginBlock(exit1);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(0));
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
build.beginBlock(block);
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.constTag(tnumber));
build.inst(IrCmd::JUMP_IF_TRUTHY, build.vmReg(0), exit2, repeat);
build.beginBlock(exit2);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(0));
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
build.beginBlock(repeat);
build.inst(IrCmd::INTERRUPT, build.constUint(0));
build.inst(IrCmd::JUMP, block);
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
JUMP bb_1
bb_1:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R0, 0i
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
bb_2:
STORE_TAG R0, tnumber
JUMP bb_3
bb_3:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R0, 0i
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
)");
}
Sync to upstream/release/577 (#934) Lots of things going on this week: * Fix a crash that could occur in the presence of a cyclic union. We shouldn't be creating cyclic unions, but we shouldn't be crashing when they arise either. * Minor cleanup of `luau_precall` * Internal change to make L->top handling slightly more uniform * Optimize SETGLOBAL & GETGLOBAL fallback C functions. * https://github.com/Roblox/luau/pull/929 * The syntax to the `luau-reduce` commandline tool has changed. It now accepts a script, a command to execute, and an error to search for. It no longer automatically passes the script to the command which makes it a lot more flexible. Also be warned that it edits the script it is passed **in place**. Do not point it at something that is not in source control! New solver * Switch to a greedier but more fallible algorithm for simplifying union and intersection types that are created as part of refinement calculation. This has much better and more predictable performance. * Fix a constraint cycle in recursive function calls. * Much improved inference of binary addition. Functions like `function add(x, y) return x + y end` can now be inferred without annotations. We also accurately typecheck calls to functions like this. * Many small bugfixes surrounding things like table indexers * Add support for indexers on class types. This was previously added to the old solver; we now add it to the new one for feature parity. JIT * https://github.com/Roblox/luau/pull/931 * Fuse key.value and key.tt loads for CEHCK_SLOT_MATCH in A64 * Implement remaining aliases of BFM for A64 * Implement new callinfo flag for A64 * Add instruction simplification for int->num->int conversion chains * Don't even load execdata for X64 calls * Treat opcode fallbacks the same as manually written fallbacks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-05-20 03:37:30 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "IntNumIntPeepholes")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
IrOp i1 = build.inst(IrCmd::LOAD_INT, build.vmReg(0));
IrOp u1 = build.inst(IrCmd::LOAD_INT, build.vmReg(1));
IrOp ni1 = build.inst(IrCmd::INT_TO_NUM, i1);
IrOp nu1 = build.inst(IrCmd::UINT_TO_NUM, u1);
IrOp i2 = build.inst(IrCmd::NUM_TO_INT, ni1);
IrOp u2 = build.inst(IrCmd::NUM_TO_UINT, nu1);
build.inst(IrCmd::STORE_INT, build.vmReg(0), i2);
build.inst(IrCmd::STORE_INT, build.vmReg(1), u2);
build.inst(IrCmd::RETURN, build.constUint(2));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%0 = LOAD_INT R0
%1 = LOAD_INT R1
STORE_INT R0, %0
STORE_INT R1, %1
RETURN 2u
)");
}
Sync to upstream/release/578 (#939) * Fixed gcc warning about uninitialized `std::optional` * Fixed inlining of functions when they are used to compute their own arguments In the new type solver: * Type families that are not part of a function signature cannot be resolved at instantiation time and will now produce an error. This will be relaxed in the future when we get constraint clauses on function signatures (internally) * `never` type is now comparable * Improved typechecking of `for..in` statements * Fixed checks for number type in `Add` type family * Performance was improved, with particularly large gains on large projects And in native code generation (jit): * We eliminated the call instruction overhead when native code support is enabled in the VM * Small optimizations to arm64 lowering * Reworked LOP_GETIMPORT handling to reduce assembly code size * Fixed non-deterministic binary output * Fixed bad code generation caused by incorrect SSA to VM register links invalidation --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Andy Friesen <afriesen@roblox.com>
2023-05-26 05:36:34 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "InvalidateReglinkVersion")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp fallback = build.block(IrBlockKind::Fallback);
build.beginBlock(block);
build.inst(IrCmd::STORE_TAG, build.vmReg(2), build.constTag(tstring));
IrOp tv2 = build.inst(IrCmd::LOAD_TVALUE, build.vmReg(2));
build.inst(IrCmd::STORE_TVALUE, build.vmReg(1), tv2);
Sync to upstream/release/591 (#1012) * Fix a use-after-free bug in the new type cloning algorithm * Tighten up the type of `coroutine.wrap`. It is now `<A..., R...>(f: (A...) -> R...) -> ((A...) -> R...)` * Break `.luaurc` out into a separate library target `Luau.Config`. This makes it easier for applications to reason about config files without also depending on the type inference engine. * Move typechecking limits into `FrontendOptions`. This allows embedders more finely-grained control over autocomplete's internal time limits. * Fix stability issue with debugger onprotectederror callback allowing break in non-yieldable contexts New solver: * Initial work toward [Local Type Inference](https://github.com/Roblox/luau/blob/0e1082108fd6fb3a32dfdf5f1766ea3fc1391328/rfcs/local-type-inference.md) * Introduce a new subtyping test. This will be much nicer than the old test because it is completely separate both from actual type inference and from error reporting. Native code generation: * Added function to compute iterated dominance frontier * Optimize barriers in SET_UPVALUE when tag is known * Cache lua_State::global in a register on A64 * Optimize constant stores in A64 lowering * Track table array size state to optimize array size checks * Add split tag/value store into a VM register * Check that spills can outlive the block only in specific conditions --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-08-19 02:15:41 +08:00
IrOp ft = build.inst(IrCmd::NEW_TABLE, build.constUint(0), build.constUint(0));
Sync to upstream/release/578 (#939) * Fixed gcc warning about uninitialized `std::optional` * Fixed inlining of functions when they are used to compute their own arguments In the new type solver: * Type families that are not part of a function signature cannot be resolved at instantiation time and will now produce an error. This will be relaxed in the future when we get constraint clauses on function signatures (internally) * `never` type is now comparable * Improved typechecking of `for..in` statements * Fixed checks for number type in `Add` type family * Performance was improved, with particularly large gains on large projects And in native code generation (jit): * We eliminated the call instruction overhead when native code support is enabled in the VM * Small optimizations to arm64 lowering * Reworked LOP_GETIMPORT handling to reduce assembly code size * Fixed non-deterministic binary output * Fixed bad code generation caused by incorrect SSA to VM register links invalidation --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Andy Friesen <afriesen@roblox.com>
2023-05-26 05:36:34 +08:00
build.inst(IrCmd::STORE_POINTER, build.vmReg(2), ft);
build.inst(IrCmd::STORE_TAG, build.vmReg(2), build.constTag(ttable));
IrOp tv1 = build.inst(IrCmd::LOAD_TVALUE, build.vmReg(1));
build.inst(IrCmd::STORE_TVALUE, build.vmReg(0), tv1);
IrOp tag = build.inst(IrCmd::LOAD_TAG, build.vmReg(0));
build.inst(IrCmd::CHECK_TAG, tag, build.constTag(ttable), fallback);
build.inst(IrCmd::RETURN, build.constUint(0));
build.beginBlock(fallback);
build.inst(IrCmd::RETURN, build.constUint(1));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
STORE_TAG R2, tstring
%1 = LOAD_TVALUE R2
STORE_TVALUE R1, %1
Sync to upstream/release/591 (#1012) * Fix a use-after-free bug in the new type cloning algorithm * Tighten up the type of `coroutine.wrap`. It is now `<A..., R...>(f: (A...) -> R...) -> ((A...) -> R...)` * Break `.luaurc` out into a separate library target `Luau.Config`. This makes it easier for applications to reason about config files without also depending on the type inference engine. * Move typechecking limits into `FrontendOptions`. This allows embedders more finely-grained control over autocomplete's internal time limits. * Fix stability issue with debugger onprotectederror callback allowing break in non-yieldable contexts New solver: * Initial work toward [Local Type Inference](https://github.com/Roblox/luau/blob/0e1082108fd6fb3a32dfdf5f1766ea3fc1391328/rfcs/local-type-inference.md) * Introduce a new subtyping test. This will be much nicer than the old test because it is completely separate both from actual type inference and from error reporting. Native code generation: * Added function to compute iterated dominance frontier * Optimize barriers in SET_UPVALUE when tag is known * Cache lua_State::global in a register on A64 * Optimize constant stores in A64 lowering * Track table array size state to optimize array size checks * Add split tag/value store into a VM register * Check that spills can outlive the block only in specific conditions --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-08-19 02:15:41 +08:00
%3 = NEW_TABLE 0u, 0u
Sync to upstream/release/578 (#939) * Fixed gcc warning about uninitialized `std::optional` * Fixed inlining of functions when they are used to compute their own arguments In the new type solver: * Type families that are not part of a function signature cannot be resolved at instantiation time and will now produce an error. This will be relaxed in the future when we get constraint clauses on function signatures (internally) * `never` type is now comparable * Improved typechecking of `for..in` statements * Fixed checks for number type in `Add` type family * Performance was improved, with particularly large gains on large projects And in native code generation (jit): * We eliminated the call instruction overhead when native code support is enabled in the VM * Small optimizations to arm64 lowering * Reworked LOP_GETIMPORT handling to reduce assembly code size * Fixed non-deterministic binary output * Fixed bad code generation caused by incorrect SSA to VM register links invalidation --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Andy Friesen <afriesen@roblox.com>
2023-05-26 05:36:34 +08:00
STORE_POINTER R2, %3
STORE_TAG R2, ttable
STORE_TVALUE R0, %1
%8 = LOAD_TAG R0
CHECK_TAG %8, ttable, bb_fallback_1
RETURN 0u
bb_fallback_1:
RETURN 1u
)");
}
Sync to upstream/release/565 (#845) We've made a few small changes to reduce the amount of stack we use when typechecking nested method calls (eg `foo:bar():baz():quux()`). We've also fixed a small bytecode compiler issue that caused us to emit redundant jump instructions in code that conditionally uses `break` or `continue`. On the new solver, we've switched to a new, better way to handle augmentations to unsealed tables. We've also made some substantial improvements to type inference and error reporting on function calls. These things should both be on par with the old solver now. The main improvements to the native code generator have been elimination of some redundant type tag checks. Also, we are starting to inline particular fastcalls directly to IR. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-02-25 05:49:38 +08:00
TEST_SUITE_END();
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
TEST_SUITE_BEGIN("LinearExecutionFlowExtraction");
TEST_CASE_FIXTURE(IrBuilderFixture, "SimplePathExtraction")
{
IrOp block1 = build.block(IrBlockKind::Internal);
IrOp fallback1 = build.block(IrBlockKind::Fallback);
IrOp block2 = build.block(IrBlockKind::Internal);
IrOp fallback2 = build.block(IrBlockKind::Fallback);
IrOp block3 = build.block(IrBlockKind::Internal);
IrOp block4 = build.block(IrBlockKind::Internal);
build.beginBlock(block1);
IrOp tag1 = build.inst(IrCmd::LOAD_TAG, build.vmReg(2));
build.inst(IrCmd::CHECK_TAG, tag1, build.constTag(tnumber), fallback1);
build.inst(IrCmd::JUMP, block2);
build.beginBlock(fallback1);
build.inst(IrCmd::DO_LEN, build.vmReg(1), build.vmReg(2));
build.inst(IrCmd::JUMP, block2);
build.beginBlock(block2);
IrOp tag2 = build.inst(IrCmd::LOAD_TAG, build.vmReg(2));
build.inst(IrCmd::CHECK_TAG, tag2, build.constTag(tnumber), fallback2);
build.inst(IrCmd::JUMP, block3);
build.beginBlock(fallback2);
build.inst(IrCmd::DO_LEN, build.vmReg(0), build.vmReg(2));
build.inst(IrCmd::JUMP, block3);
build.beginBlock(block3);
build.inst(IrCmd::JUMP, block4);
build.beginBlock(block4);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(0));
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
createLinearBlocks(build, true);
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
bb_0:
%0 = LOAD_TAG R2
CHECK_TAG %0, tnumber, bb_fallback_1
JUMP bb_linear_6
bb_fallback_1:
DO_LEN R1, R2
JUMP bb_2
bb_2:
%5 = LOAD_TAG R2
CHECK_TAG %5, tnumber, bb_fallback_3
JUMP bb_4
bb_fallback_3:
DO_LEN R0, R2
JUMP bb_4
bb_4:
JUMP bb_5
bb_5:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R0, 0i
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
bb_linear_6:
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R0, 0i
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "NoPathExtractionForBlocksWithLiveOutValues")
{
IrOp block1 = build.block(IrBlockKind::Internal);
IrOp fallback1 = build.block(IrBlockKind::Fallback);
IrOp block2 = build.block(IrBlockKind::Internal);
IrOp fallback2 = build.block(IrBlockKind::Fallback);
IrOp block3 = build.block(IrBlockKind::Internal);
IrOp block4a = build.block(IrBlockKind::Internal);
IrOp block4b = build.block(IrBlockKind::Internal);
build.beginBlock(block1);
IrOp tag1 = build.inst(IrCmd::LOAD_TAG, build.vmReg(2));
build.inst(IrCmd::CHECK_TAG, tag1, build.constTag(tnumber), fallback1);
build.inst(IrCmd::JUMP, block2);
build.beginBlock(fallback1);
build.inst(IrCmd::DO_LEN, build.vmReg(1), build.vmReg(2));
build.inst(IrCmd::JUMP, block2);
build.beginBlock(block2);
IrOp tag2 = build.inst(IrCmd::LOAD_TAG, build.vmReg(2));
build.inst(IrCmd::CHECK_TAG, tag2, build.constTag(tnumber), fallback2);
build.inst(IrCmd::JUMP, block3);
build.beginBlock(fallback2);
build.inst(IrCmd::DO_LEN, build.vmReg(0), build.vmReg(2));
build.inst(IrCmd::JUMP, block3);
build.beginBlock(block3);
IrOp tag3a = build.inst(IrCmd::LOAD_TAG, build.vmReg(3));
build.inst(IrCmd::JUMP_EQ_TAG, tag3a, build.constTag(tnil), block4a, block4b);
build.beginBlock(block4a);
build.inst(IrCmd::STORE_TAG, build.vmReg(0), tag3a);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(0));
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
build.beginBlock(block4b);
build.inst(IrCmd::STORE_TAG, build.vmReg(0), tag3a);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(0));
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
createLinearBlocks(build, true);
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
bb_0:
%0 = LOAD_TAG R2
CHECK_TAG %0, tnumber, bb_fallback_1
JUMP bb_2
bb_fallback_1:
DO_LEN R1, R2
JUMP bb_2
bb_2:
%5 = LOAD_TAG R2
CHECK_TAG %5, tnumber, bb_fallback_3
JUMP bb_4
bb_fallback_3:
DO_LEN R0, R2
JUMP bb_4
bb_4:
%10 = LOAD_TAG R3
JUMP_EQ_TAG %10, tnil, bb_5, bb_6
bb_5:
STORE_TAG R0, %10
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R0, 0i
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
bb_6:
STORE_TAG R0, %10
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R0, 0i
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "InfiniteLoopInPathAnalysis")
{
IrOp block1 = build.block(IrBlockKind::Internal);
IrOp block2 = build.block(IrBlockKind::Internal);
build.beginBlock(block1);
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.constTag(tnumber));
build.inst(IrCmd::JUMP, block2);
build.beginBlock(block2);
build.inst(IrCmd::STORE_TAG, build.vmReg(1), build.constTag(tboolean));
build.inst(IrCmd::JUMP, block2);
updateUseCounts(build.function);
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
constPropInBlockChains(build, true);
createLinearBlocks(build, true);
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
bb_0:
STORE_TAG R0, tnumber
JUMP bb_1
bb_1:
STORE_TAG R1, tboolean
JUMP bb_1
)");
}
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "PartialStoreInvalidation")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
build.inst(IrCmd::STORE_TVALUE, build.vmReg(1), build.inst(IrCmd::LOAD_TVALUE, build.vmReg(0)));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(0), build.constDouble(0.5));
build.inst(IrCmd::STORE_TVALUE, build.vmReg(1), build.inst(IrCmd::LOAD_TVALUE, build.vmReg(0))); // Should be reloaded
Sync to upstream/release/591 (#1012) * Fix a use-after-free bug in the new type cloning algorithm * Tighten up the type of `coroutine.wrap`. It is now `<A..., R...>(f: (A...) -> R...) -> ((A...) -> R...)` * Break `.luaurc` out into a separate library target `Luau.Config`. This makes it easier for applications to reason about config files without also depending on the type inference engine. * Move typechecking limits into `FrontendOptions`. This allows embedders more finely-grained control over autocomplete's internal time limits. * Fix stability issue with debugger onprotectederror callback allowing break in non-yieldable contexts New solver: * Initial work toward [Local Type Inference](https://github.com/Roblox/luau/blob/0e1082108fd6fb3a32dfdf5f1766ea3fc1391328/rfcs/local-type-inference.md) * Introduce a new subtyping test. This will be much nicer than the old test because it is completely separate both from actual type inference and from error reporting. Native code generation: * Added function to compute iterated dominance frontier * Optimize barriers in SET_UPVALUE when tag is known * Cache lua_State::global in a register on A64 * Optimize constant stores in A64 lowering * Track table array size state to optimize array size checks * Add split tag/value store into a VM register * Check that spills can outlive the block only in specific conditions --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-08-19 02:15:41 +08:00
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.constTag(tnumber));
build.inst(IrCmd::STORE_TVALUE, build.vmReg(1), build.inst(IrCmd::LOAD_TVALUE, build.vmReg(0)));
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
build.inst(IrCmd::RETURN, build.constUint(0));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%0 = LOAD_TVALUE R0
STORE_TVALUE R1, %0
STORE_DOUBLE R0, 0.5
%3 = LOAD_TVALUE R0
STORE_TVALUE R1, %3
Sync to upstream/release/591 (#1012) * Fix a use-after-free bug in the new type cloning algorithm * Tighten up the type of `coroutine.wrap`. It is now `<A..., R...>(f: (A...) -> R...) -> ((A...) -> R...)` * Break `.luaurc` out into a separate library target `Luau.Config`. This makes it easier for applications to reason about config files without also depending on the type inference engine. * Move typechecking limits into `FrontendOptions`. This allows embedders more finely-grained control over autocomplete's internal time limits. * Fix stability issue with debugger onprotectederror callback allowing break in non-yieldable contexts New solver: * Initial work toward [Local Type Inference](https://github.com/Roblox/luau/blob/0e1082108fd6fb3a32dfdf5f1766ea3fc1391328/rfcs/local-type-inference.md) * Introduce a new subtyping test. This will be much nicer than the old test because it is completely separate both from actual type inference and from error reporting. Native code generation: * Added function to compute iterated dominance frontier * Optimize barriers in SET_UPVALUE when tag is known * Cache lua_State::global in a register on A64 * Optimize constant stores in A64 lowering * Track table array size state to optimize array size checks * Add split tag/value store into a VM register * Check that spills can outlive the block only in specific conditions --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-08-19 02:15:41 +08:00
STORE_TAG R0, tnumber
STORE_SPLIT_TVALUE R1, tnumber, 0.5
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
RETURN 0u
)");
}
Sync to upstream/release/579 (#943) A pretty small changelist this week: * When type inference fails to find any matching overload for a function, we were declining to commit any changes to the type graph at all. This was resulting in confusing type errors in certain cases. Now, when a matching overload cannot be found, we always commit to the first overload we tried. JIT * Fix missing variadic register invalidation in FALLBACK_GETVARARGS * Add a missing null pointer check for the result of luaT_gettm --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-06-03 03:52:15 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "VaridicRegisterRangeInvalidation")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
build.inst(IrCmd::STORE_TAG, build.vmReg(2), build.constTag(tnumber));
build.inst(IrCmd::FALLBACK_GETVARARGS, build.constUint(0), build.vmReg(1), build.constInt(-1));
build.inst(IrCmd::STORE_TAG, build.vmReg(2), build.constTag(tnumber));
build.inst(IrCmd::RETURN, build.constUint(0));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
STORE_TAG R2, tnumber
FALLBACK_GETVARARGS 0u, R1, -1i
STORE_TAG R2, tnumber
RETURN 0u
)");
}
Sync to upstream/release/583 (#974) * Fixed indexing table intersections using `x["prop"]` syntax: https://github.com/Roblox/luau/pull/971 * Add console output codepage for Windows: https://github.com/Roblox/luau/pull/967 * Added `Frontend::parse` for a fast source graph preparation * luau_load should check GC * Work toward a type-diff system for nicer error messages New Solver * Correctly suppress errors in more cases * Further improvements to typechecking of function calls and return statements * Crash fixes * Propagate refinements drawn from the condition of a while loop into the loop body JIT * Fix accidental bailout for math.frexp/modf/sign in A64 * Work toward bringing type annotation info in * Do not propagate Luau IR constants of wrong type into load instructions * CHECK_SAFEENV exits to VM on failure * Implement error handling in A64 reg allocator * Inline the string.len builtin * Do not enter native code of a function if arguments don’t match --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-07-08 04:10:48 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "LoadPropagatesOnlyRightType")
{
IrOp block = build.block(IrBlockKind::Internal);
build.beginBlock(block);
build.inst(IrCmd::STORE_INT, build.vmReg(0), build.constInt(2));
IrOp value1 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(1), value1);
IrOp value2 = build.inst(IrCmd::LOAD_INT, build.vmReg(1));
build.inst(IrCmd::STORE_INT, build.vmReg(2), value2);
build.inst(IrCmd::RETURN, build.constUint(0));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
STORE_INT R0, 2i
%1 = LOAD_DOUBLE R0
STORE_DOUBLE R1, %1
%3 = LOAD_INT R1
STORE_INT R2, %3
RETURN 0u
)");
}
Sync to upstream/release/593 (#1024) - Updated Roblox copyright to 2023 - Floor division operator `//` (implements #832) - Autocomplete now shows `end` within `do` blocks - Restore BraceType when using `Lexer::lookahead` (fixes #1019) # New typechecker - Subtyping tests between metatables and tables - Subtyping tests between string singletons and tables - Subtyping tests for class types # Native codegen - Fixed macOS test failure (wrong spill restore offset) - Fixed clobbering of non-volatile xmm registers on Windows - Fixed wrong storage location of SSA reg spills - Implemented A64 support for add/sub extended - Eliminated zextReg from A64 lowering - Remove identical table slot lookups - Propagate values from predecessor into the linear block - Disabled reuse slot optimization - Keep `LuaNode::val` check for nil when optimizing `CHECK_SLOT_MATCH` - Implemented IR translation of `table.insert` builtin - Fixed mmap error handling on macOS/Linux # Tooling - Used `|` as a column separator instead of `+` in `bench.py` - Added a `table.sort` micro-benchmark - Switched `libprotobuf-mutator` to a less problematic version
2023-09-02 01:58:27 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "DuplicateHashSlotChecks")
{
ScopedFastFlag luauReuseHashSlots{"LuauReuseHashSlots2", true};
IrOp block = build.block(IrBlockKind::Internal);
IrOp fallback = build.block(IrBlockKind::Fallback);
build.beginBlock(block);
// This roughly corresponds to 'return t.a + t.a'
IrOp table1 = build.inst(IrCmd::LOAD_POINTER, build.vmReg(1));
IrOp slot1 = build.inst(IrCmd::GET_SLOT_NODE_ADDR, table1, build.constUint(3), build.vmConst(1));
build.inst(IrCmd::CHECK_SLOT_MATCH, slot1, build.vmConst(1), fallback);
IrOp value1 = build.inst(IrCmd::LOAD_TVALUE, slot1, build.constInt(0));
build.inst(IrCmd::STORE_TVALUE, build.vmReg(3), value1);
IrOp slot1b = build.inst(IrCmd::GET_SLOT_NODE_ADDR, table1, build.constUint(8), build.vmConst(1)); // This will be removed
build.inst(IrCmd::CHECK_SLOT_MATCH, slot1b, build.vmConst(1), fallback); // Key will be replaced with undef here
IrOp value1b = build.inst(IrCmd::LOAD_TVALUE, slot1b, build.constInt(0));
build.inst(IrCmd::STORE_TVALUE, build.vmReg(4), value1b);
IrOp a = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(3));
IrOp b = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(4));
IrOp sum = build.inst(IrCmd::ADD_NUM, a, b);
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(2), sum);
build.inst(IrCmd::RETURN, build.vmReg(2), build.constUint(1));
build.beginBlock(fallback);
build.inst(IrCmd::RETURN, build.vmReg(0), build.constUint(1));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
// In the future, we might even see duplicate identical TValue loads go away
// In the future, we might even see loads of different VM regs with the same value go away
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%0 = LOAD_POINTER R1
%1 = GET_SLOT_NODE_ADDR %0, 3u, K1
CHECK_SLOT_MATCH %1, K1, bb_fallback_1
%3 = LOAD_TVALUE %1, 0i
STORE_TVALUE R3, %3
CHECK_NODE_VALUE %1, bb_fallback_1
%7 = LOAD_TVALUE %1, 0i
STORE_TVALUE R4, %7
%9 = LOAD_DOUBLE R3
%10 = LOAD_DOUBLE R4
%11 = ADD_NUM %9, %10
STORE_DOUBLE R2, %11
RETURN R2, 1u
bb_fallback_1:
RETURN R0, 1u
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "DuplicateHashSlotChecksAvoidNil")
{
ScopedFastFlag luauReuseHashSlots{"LuauReuseHashSlots2", true};
IrOp block = build.block(IrBlockKind::Internal);
IrOp fallback = build.block(IrBlockKind::Fallback);
build.beginBlock(block);
IrOp table1 = build.inst(IrCmd::LOAD_POINTER, build.vmReg(1));
IrOp slot1 = build.inst(IrCmd::GET_SLOT_NODE_ADDR, table1, build.constUint(3), build.vmConst(1));
build.inst(IrCmd::CHECK_SLOT_MATCH, slot1, build.vmConst(1), fallback);
IrOp value1 = build.inst(IrCmd::LOAD_TVALUE, slot1, build.constInt(0));
build.inst(IrCmd::STORE_TVALUE, build.vmReg(3), value1);
IrOp table2 = build.inst(IrCmd::LOAD_POINTER, build.vmReg(2));
IrOp slot2 = build.inst(IrCmd::GET_SLOT_NODE_ADDR, table2, build.constUint(6), build.vmConst(1));
build.inst(IrCmd::CHECK_SLOT_MATCH, slot2, build.vmConst(1), fallback);
build.inst(IrCmd::CHECK_READONLY, table2, fallback);
build.inst(IrCmd::STORE_TAG, build.vmReg(4), build.constTag(tnil));
IrOp valueNil = build.inst(IrCmd::LOAD_TVALUE, build.vmReg(4));
build.inst(IrCmd::STORE_TVALUE, slot2, valueNil, build.constInt(0));
// In the future, we might get to track that value became 'nil' and that fallback will be taken
IrOp slot1b = build.inst(IrCmd::GET_SLOT_NODE_ADDR, table1, build.constUint(8), build.vmConst(1)); // This will be removed
build.inst(IrCmd::CHECK_SLOT_MATCH, slot1b, build.vmConst(1), fallback); // Key will be replaced with undef here
IrOp value1b = build.inst(IrCmd::LOAD_TVALUE, slot1b, build.constInt(0));
build.inst(IrCmd::STORE_TVALUE, build.vmReg(3), value1b);
IrOp slot2b = build.inst(IrCmd::GET_SLOT_NODE_ADDR, table2, build.constUint(11), build.vmConst(1)); // This will be removed
build.inst(IrCmd::CHECK_SLOT_MATCH, slot2b, build.vmConst(1), fallback); // Key will be replaced with undef here
build.inst(IrCmd::CHECK_READONLY, table2, fallback);
build.inst(IrCmd::STORE_SPLIT_TVALUE, slot2b, build.constTag(tnumber), build.constDouble(1), build.constInt(0));
build.inst(IrCmd::RETURN, build.vmReg(3), build.constUint(2));
build.beginBlock(fallback);
build.inst(IrCmd::RETURN, build.vmReg(1), build.constUint(2));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%0 = LOAD_POINTER R1
%1 = GET_SLOT_NODE_ADDR %0, 3u, K1
CHECK_SLOT_MATCH %1, K1, bb_fallback_1
%3 = LOAD_TVALUE %1, 0i
STORE_TVALUE R3, %3
%5 = LOAD_POINTER R2
%6 = GET_SLOT_NODE_ADDR %5, 6u, K1
CHECK_SLOT_MATCH %6, K1, bb_fallback_1
CHECK_READONLY %5, bb_fallback_1
STORE_TAG R4, tnil
%10 = LOAD_TVALUE R4
STORE_TVALUE %6, %10, 0i
CHECK_NODE_VALUE %1, bb_fallback_1
%14 = LOAD_TVALUE %1, 0i
STORE_TVALUE R3, %14
CHECK_NODE_VALUE %6, bb_fallback_1
STORE_SPLIT_TVALUE %6, tnumber, 1, 0i
RETURN R3, 2u
bb_fallback_1:
RETURN R1, 2u
)");
}
Sync to upstream/release/566 (#853) * Fixed incorrect lexeme generated for string parts in the middle of an interpolated string (Fixes https://github.com/Roblox/luau/issues/744) * DeprecatedApi lint can report some issues without type inference information * Fixed performance of autocomplete requests when suggestions have large intersection types (Solves https://github.com/Roblox/luau/discussions/847) * Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC: Deprecate table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md)) * With -O2 optimization level, we now optimize builtin calls based on known argument/return count. Note that this change can be observable if `getfenv/setfenv` is used to substitute a builtin, especially if arity is different. Fastcall heavy tests show a 1-2% improvement. * Luau can now be built with clang-cl (Fixes https://github.com/Roblox/luau/issues/736) We also made many improvements to our experimental components. For our new type solver: * Overhauled data flow analysis system, fixed issues with 'repeat' loops, global variables and type annotations * Type refinements now work on generic table indexing with a string literal * Type refinements will properly track potentially 'nil' values (like t[x] for a missing key) and their further refinements * Internal top table type is now isomorphic to `{}` which fixes issues when `typeof(v) == 'table'` type refinement is handled * References to non-existent types in type annotations no longer resolve to 'error' type like in old solver * Improved handling of class unions in property access expressions * Fixed default type packs * Unsealed tables can now have metatables * Restored expected types for function arguments And for native code generation: * Added min and max IR instructions mapping to vminsd/vmaxsd on x64 * We now speculatively extract direct execution fast-paths based on expected types of expressions which provides better optimization opportunities inside a single basic block * Translated existing math fastcalls to IR form to improve tag guard removal and constant propagation
2023-03-04 04:21:14 +08:00
TEST_SUITE_END();
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
TEST_SUITE_BEGIN("Analysis");
TEST_CASE_FIXTURE(IrBuilderFixture, "SimpleDiamond")
{
IrOp entry = build.block(IrBlockKind::Internal);
IrOp a = build.block(IrBlockKind::Internal);
IrOp b = build.block(IrBlockKind::Internal);
IrOp exit = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
build.inst(IrCmd::JUMP_EQ_TAG, build.inst(IrCmd::LOAD_TAG, build.vmReg(0)), build.constTag(tnumber), a, b);
build.beginBlock(a);
build.inst(IrCmd::STORE_TVALUE, build.vmReg(2), build.inst(IrCmd::LOAD_TVALUE, build.vmReg(1)));
build.inst(IrCmd::JUMP, exit);
build.beginBlock(b);
build.inst(IrCmd::STORE_TVALUE, build.vmReg(3), build.inst(IrCmd::LOAD_TVALUE, build.vmReg(1)));
build.inst(IrCmd::JUMP, exit);
build.beginBlock(exit);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(2), build.constInt(2));
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
updateUseCounts(build.function);
computeCfgInfo(build.function);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
; successors: bb_1, bb_2
; in regs: R0, R1, R2, R3
; out regs: R1, R2, R3
%0 = LOAD_TAG R0
JUMP_EQ_TAG %0, tnumber, bb_1, bb_2
bb_1:
; predecessors: bb_0
; successors: bb_3
; in regs: R1, R3
; out regs: R2, R3
%2 = LOAD_TVALUE R1
STORE_TVALUE R2, %2
JUMP bb_3
bb_2:
; predecessors: bb_0
; successors: bb_3
; in regs: R1, R2
; out regs: R2, R3
%5 = LOAD_TVALUE R1
STORE_TVALUE R3, %5
JUMP bb_3
bb_3:
; predecessors: bb_1, bb_2
; in regs: R2, R3
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R2, 2i
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "ImplicitFixedRegistersInVarargCall")
{
IrOp entry = build.block(IrBlockKind::Internal);
IrOp exit = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
build.inst(IrCmd::FALLBACK_GETVARARGS, build.constUint(0), build.vmReg(3), build.constInt(-1));
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::CALL, build.vmReg(0), build.constInt(-1), build.constInt(5));
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
build.inst(IrCmd::JUMP, exit);
build.beginBlock(exit);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(5));
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
updateUseCounts(build.function);
computeCfgInfo(build.function);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
; successors: bb_1
; in regs: R0, R1, R2
; out regs: R0, R1, R2, R3, R4
FALLBACK_GETVARARGS 0u, R3, -1i
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
CALL R0, -1i, 5i
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
JUMP bb_1
bb_1:
; predecessors: bb_0
; in regs: R0, R1, R2, R3, R4
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R0, 5i
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "ExplicitUseOfRegisterInVarargSequence")
{
IrOp entry = build.block(IrBlockKind::Internal);
IrOp exit = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
build.inst(IrCmd::FALLBACK_GETVARARGS, build.constUint(0), build.vmReg(1), build.constInt(-1));
Sync to upstream/release/569 (#878) All of our changes this week have been focused on the new type solver and the JIT. As we march toward feature parity with the old solver, we've tightened up a bunch of lingering issues with overload resolution, unsealed tables, and type normalization. We've also fixed a bunch of crashes and assertion failures in the new solver. On the JIT front, we've started work on an A64 backend, improved the IR analysis in a bunch of cases, and implemented assembly generation for the builtin functions `type()` and `typeof()`. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-25 02:03:04 +08:00
IrOp results = build.inst(
IrCmd::INVOKE_FASTCALL, build.constUint(0), build.vmReg(0), build.vmReg(1), build.vmReg(2), build.constInt(-1), build.constInt(-1));
build.inst(IrCmd::ADJUST_STACK_TO_REG, build.vmReg(0), results);
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
build.inst(IrCmd::JUMP, exit);
build.beginBlock(exit);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(-1));
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
updateUseCounts(build.function);
computeCfgInfo(build.function);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
; successors: bb_1
; out regs: R0...
FALLBACK_GETVARARGS 0u, R1, -1i
%1 = INVOKE_FASTCALL 0u, R0, R1, R2, -1i, -1i
Sync to upstream/release/569 (#878) All of our changes this week have been focused on the new type solver and the JIT. As we march toward feature parity with the old solver, we've tightened up a bunch of lingering issues with overload resolution, unsealed tables, and type normalization. We've also fixed a bunch of crashes and assertion failures in the new solver. On the JIT front, we've started work on an A64 backend, improved the IR analysis in a bunch of cases, and implemented assembly generation for the builtin functions `type()` and `typeof()`. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-25 02:03:04 +08:00
ADJUST_STACK_TO_REG R0, %1
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
JUMP bb_1
bb_1:
; predecessors: bb_0
; in regs: R0...
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R0, -1i
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "VariadicSequenceRestart")
{
IrOp entry = build.block(IrBlockKind::Internal);
IrOp exit = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::CALL, build.vmReg(1), build.constInt(0), build.constInt(-1));
build.inst(IrCmd::CALL, build.vmReg(0), build.constInt(-1), build.constInt(-1));
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
build.inst(IrCmd::JUMP, exit);
build.beginBlock(exit);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(-1));
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
updateUseCounts(build.function);
computeCfgInfo(build.function);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
; successors: bb_1
; in regs: R0, R1
; out regs: R0...
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
CALL R1, 0i, -1i
CALL R0, -1i, -1i
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
JUMP bb_1
bb_1:
; predecessors: bb_0
; in regs: R0...
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R0, -1i
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "FallbackDoesNotFlowUp")
{
IrOp entry = build.block(IrBlockKind::Internal);
IrOp fallback = build.block(IrBlockKind::Fallback);
IrOp exit = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
build.inst(IrCmd::FALLBACK_GETVARARGS, build.constUint(0), build.vmReg(1), build.constInt(-1));
build.inst(IrCmd::CHECK_TAG, build.inst(IrCmd::LOAD_TAG, build.vmReg(0)), build.constTag(tnumber), fallback);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::CALL, build.vmReg(0), build.constInt(-1), build.constInt(-1));
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
build.inst(IrCmd::JUMP, exit);
build.beginBlock(fallback);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::CALL, build.vmReg(0), build.constInt(-1), build.constInt(-1));
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
build.inst(IrCmd::JUMP, exit);
build.beginBlock(exit);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(-1));
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
updateUseCounts(build.function);
computeCfgInfo(build.function);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
; successors: bb_fallback_1, bb_2
; in regs: R0
; out regs: R0...
FALLBACK_GETVARARGS 0u, R1, -1i
%1 = LOAD_TAG R0
CHECK_TAG %1, tnumber, bb_fallback_1
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
CALL R0, -1i, -1i
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
JUMP bb_2
bb_fallback_1:
; predecessors: bb_0
; successors: bb_2
; in regs: R0, R1...
; out regs: R0...
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
CALL R0, -1i, -1i
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
JUMP bb_2
bb_2:
; predecessors: bb_0, bb_fallback_1
; in regs: R0...
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R0, -1i
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
)");
}
Sync to upstream/release/568 (#865) * A small subset of control-flow refinements have been added to recognize type options that are unreachable after a conditional/unconditional code block. (Fixes https://github.com/Roblox/luau/issues/356). Some examples: ```lua local function f(x: string?) if not x then return end -- x is 'string' here end ``` Throwing calls like `error` or `assert(false)` instead of 'return' are also recognized. Existing complex refinements like type/typeof and tagged union checks are expected to work, among others. To enable this feature, `LuauTinyControlFlowAnalysis` exclusion has to be removed from `ExperimentalFlags.h`. If will become enabled unconditionally in the near future. * Linter has been integrated into the typechecker analysis so that type-aware lint warnings can work in any mode `Frontend::lint` methods were deprecated, `Frontend::check` has to be used instead with `runLintChecks` option set. Resulting lint warning are located inside `CheckResult`. * Fixed large performance drop and increased memory consumption when array is filled at an offset (Fixes https://github.com/Roblox/luau/issues/590) * Part of [Type error suppression RFC](https://github.com/Roblox/luau/blob/master/rfcs/type-error-suppression.md) was implemented making subtyping checks with `any` type transitive. --- In our work on the new type-solver: * `--!nocheck` mode no longer reports type errors * New solver will not be used for `--!nonstrict` modules until all issues with strict mode typechecking are fixed * Added control-flow aware type refinements mentioned earlier In native code generation: * `LOP_NAMECALL` has been translated to IR * `type` and `typeof` builtin fastcalls have been translated to IR/assembly * Additional steps were taken towards arm64 support
2023-03-18 03:20:37 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "VariadicSequencePeeling")
{
IrOp entry = build.block(IrBlockKind::Internal);
IrOp a = build.block(IrBlockKind::Internal);
IrOp b = build.block(IrBlockKind::Internal);
IrOp exit = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
build.inst(IrCmd::FALLBACK_GETVARARGS, build.constUint(0), build.vmReg(3), build.constInt(-1));
build.inst(IrCmd::JUMP_EQ_TAG, build.inst(IrCmd::LOAD_TAG, build.vmReg(0)), build.constTag(tnumber), a, b);
build.beginBlock(a);
build.inst(IrCmd::STORE_TVALUE, build.vmReg(2), build.inst(IrCmd::LOAD_TVALUE, build.vmReg(0)));
build.inst(IrCmd::JUMP, exit);
build.beginBlock(b);
build.inst(IrCmd::STORE_TVALUE, build.vmReg(2), build.inst(IrCmd::LOAD_TVALUE, build.vmReg(1)));
build.inst(IrCmd::JUMP, exit);
build.beginBlock(exit);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(2), build.constInt(-1));
Sync to upstream/release/568 (#865) * A small subset of control-flow refinements have been added to recognize type options that are unreachable after a conditional/unconditional code block. (Fixes https://github.com/Roblox/luau/issues/356). Some examples: ```lua local function f(x: string?) if not x then return end -- x is 'string' here end ``` Throwing calls like `error` or `assert(false)` instead of 'return' are also recognized. Existing complex refinements like type/typeof and tagged union checks are expected to work, among others. To enable this feature, `LuauTinyControlFlowAnalysis` exclusion has to be removed from `ExperimentalFlags.h`. If will become enabled unconditionally in the near future. * Linter has been integrated into the typechecker analysis so that type-aware lint warnings can work in any mode `Frontend::lint` methods were deprecated, `Frontend::check` has to be used instead with `runLintChecks` option set. Resulting lint warning are located inside `CheckResult`. * Fixed large performance drop and increased memory consumption when array is filled at an offset (Fixes https://github.com/Roblox/luau/issues/590) * Part of [Type error suppression RFC](https://github.com/Roblox/luau/blob/master/rfcs/type-error-suppression.md) was implemented making subtyping checks with `any` type transitive. --- In our work on the new type-solver: * `--!nocheck` mode no longer reports type errors * New solver will not be used for `--!nonstrict` modules until all issues with strict mode typechecking are fixed * Added control-flow aware type refinements mentioned earlier In native code generation: * `LOP_NAMECALL` has been translated to IR * `type` and `typeof` builtin fastcalls have been translated to IR/assembly * Additional steps were taken towards arm64 support
2023-03-18 03:20:37 +08:00
updateUseCounts(build.function);
computeCfgInfo(build.function);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
; successors: bb_1, bb_2
; in regs: R0, R1
; out regs: R0, R1, R3...
FALLBACK_GETVARARGS 0u, R3, -1i
%1 = LOAD_TAG R0
JUMP_EQ_TAG %1, tnumber, bb_1, bb_2
bb_1:
; predecessors: bb_0
; successors: bb_3
; in regs: R0, R3...
; out regs: R2...
%3 = LOAD_TVALUE R0
STORE_TVALUE R2, %3
JUMP bb_3
bb_2:
; predecessors: bb_0
; successors: bb_3
; in regs: R1, R3...
; out regs: R2...
%6 = LOAD_TVALUE R1
STORE_TVALUE R2, %6
JUMP bb_3
bb_3:
; predecessors: bb_1, bb_2
; in regs: R2...
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R2, -1i
Sync to upstream/release/569 (#878) All of our changes this week have been focused on the new type solver and the JIT. As we march toward feature parity with the old solver, we've tightened up a bunch of lingering issues with overload resolution, unsealed tables, and type normalization. We've also fixed a bunch of crashes and assertion failures in the new solver. On the JIT front, we've started work on an A64 backend, improved the IR analysis in a bunch of cases, and implemented assembly generation for the builtin functions `type()` and `typeof()`. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-25 02:03:04 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "BuiltinVariadicStart")
{
IrOp entry = build.block(IrBlockKind::Internal);
IrOp exit = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(1), build.constDouble(1.0));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(2), build.constDouble(2.0));
build.inst(IrCmd::ADJUST_STACK_TO_REG, build.vmReg(2), build.constInt(1));
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::CALL, build.vmReg(1), build.constInt(-1), build.constInt(1));
Sync to upstream/release/569 (#878) All of our changes this week have been focused on the new type solver and the JIT. As we march toward feature parity with the old solver, we've tightened up a bunch of lingering issues with overload resolution, unsealed tables, and type normalization. We've also fixed a bunch of crashes and assertion failures in the new solver. On the JIT front, we've started work on an A64 backend, improved the IR analysis in a bunch of cases, and implemented assembly generation for the builtin functions `type()` and `typeof()`. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-25 02:03:04 +08:00
build.inst(IrCmd::JUMP, exit);
build.beginBlock(exit);
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(2));
Sync to upstream/release/569 (#878) All of our changes this week have been focused on the new type solver and the JIT. As we march toward feature parity with the old solver, we've tightened up a bunch of lingering issues with overload resolution, unsealed tables, and type normalization. We've also fixed a bunch of crashes and assertion failures in the new solver. On the JIT front, we've started work on an A64 backend, improved the IR analysis in a bunch of cases, and implemented assembly generation for the builtin functions `type()` and `typeof()`. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-25 02:03:04 +08:00
updateUseCounts(build.function);
computeCfgInfo(build.function);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
; successors: bb_1
; in regs: R0
; out regs: R0, R1
STORE_DOUBLE R1, 1
STORE_DOUBLE R2, 2
ADJUST_STACK_TO_REG R2, 1i
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
CALL R1, -1i, 1i
Sync to upstream/release/569 (#878) All of our changes this week have been focused on the new type solver and the JIT. As we march toward feature parity with the old solver, we've tightened up a bunch of lingering issues with overload resolution, unsealed tables, and type normalization. We've also fixed a bunch of crashes and assertion failures in the new solver. On the JIT front, we've started work on an A64 backend, improved the IR analysis in a bunch of cases, and implemented assembly generation for the builtin functions `type()` and `typeof()`. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-25 02:03:04 +08:00
JUMP bb_1
bb_1:
; predecessors: bb_0
; in regs: R0, R1
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R0, 2i
Sync to upstream/release/569 (#878) All of our changes this week have been focused on the new type solver and the JIT. As we march toward feature parity with the old solver, we've tightened up a bunch of lingering issues with overload resolution, unsealed tables, and type normalization. We've also fixed a bunch of crashes and assertion failures in the new solver. On the JIT front, we've started work on an A64 backend, improved the IR analysis in a bunch of cases, and implemented assembly generation for the builtin functions `type()` and `typeof()`. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-25 02:03:04 +08:00
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "SetTable")
{
IrOp entry = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
build.inst(IrCmd::SET_TABLE, build.vmReg(0), build.vmReg(1), build.constUint(1));
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(1));
Sync to upstream/release/569 (#878) All of our changes this week have been focused on the new type solver and the JIT. As we march toward feature parity with the old solver, we've tightened up a bunch of lingering issues with overload resolution, unsealed tables, and type normalization. We've also fixed a bunch of crashes and assertion failures in the new solver. On the JIT front, we've started work on an A64 backend, improved the IR analysis in a bunch of cases, and implemented assembly generation for the builtin functions `type()` and `typeof()`. --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-25 02:03:04 +08:00
updateUseCounts(build.function);
computeCfgInfo(build.function);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
; in regs: R0, R1
SET_TABLE R0, R1, 1u
Sync to upstream/release/570 (#885) Once again, all of our changes this week are for new type solver and the JIT. In the new type solver, we fixed cyclic type alias handling and multiple stability issues. In the JIT, our main progress was for arm64, where, after lowering 36% of instructions, we start seeing first Luau functions executing natively. For x64, we performed code cleanup and refactoring to allow for future optimizations.
2023-04-01 02:42:49 +08:00
RETURN R0, 1i
Sync to upstream/release/568 (#865) * A small subset of control-flow refinements have been added to recognize type options that are unreachable after a conditional/unconditional code block. (Fixes https://github.com/Roblox/luau/issues/356). Some examples: ```lua local function f(x: string?) if not x then return end -- x is 'string' here end ``` Throwing calls like `error` or `assert(false)` instead of 'return' are also recognized. Existing complex refinements like type/typeof and tagged union checks are expected to work, among others. To enable this feature, `LuauTinyControlFlowAnalysis` exclusion has to be removed from `ExperimentalFlags.h`. If will become enabled unconditionally in the near future. * Linter has been integrated into the typechecker analysis so that type-aware lint warnings can work in any mode `Frontend::lint` methods were deprecated, `Frontend::check` has to be used instead with `runLintChecks` option set. Resulting lint warning are located inside `CheckResult`. * Fixed large performance drop and increased memory consumption when array is filled at an offset (Fixes https://github.com/Roblox/luau/issues/590) * Part of [Type error suppression RFC](https://github.com/Roblox/luau/blob/master/rfcs/type-error-suppression.md) was implemented making subtyping checks with `any` type transitive. --- In our work on the new type-solver: * `--!nocheck` mode no longer reports type errors * New solver will not be used for `--!nonstrict` modules until all issues with strict mode typechecking are fixed * Added control-flow aware type refinements mentioned earlier In native code generation: * `LOP_NAMECALL` has been translated to IR * `type` and `typeof` builtin fastcalls have been translated to IR/assembly * Additional steps were taken towards arm64 support
2023-03-18 03:20:37 +08:00
)");
}
Sync to upstream/release/580 (#951) * Added luau-compile executable target to build/test compilation without having full REPL included. In our new typechecker: * Fixed the order in which constraints are checked to get more deterministic errors in different environments * Fixed `isNumber`/`isString` checks to fix false positive errors in binary comparisons * CannotCallNonFunction error is reported when calling an intersection type of non-functions In our native code generation (jit): * Outlined X64 return instruction code to improve code size * Improved performance of return instruction on A64 * Added construction of a dominator tree for future optimizations
2023-06-10 01:08:00 +08:00
// 'A Simple, Fast Dominance Algorithm' [Keith D. Cooper, et al]. Figure 2.
TEST_CASE_FIXTURE(IrBuilderFixture, "DominanceVerification1")
{
defineCfgTree({{1, 2}, {3}, {4}, {4}, {3}});
CHECK(build.function.cfg.idoms == std::vector<uint32_t>{{~0u, 0, 0, 0, 0}});
}
// 'A Linear Time Algorithm for Placing Phi-Nodes' [Vugranam C.Sreedhar]. Figure 1.
TEST_CASE_FIXTURE(IrBuilderFixture, "DominanceVerification2")
{
defineCfgTree({{1, 16}, {2, 3, 4}, {4, 7}, {9}, {5}, {6}, {2, 8}, {8}, {7, 15}, {10, 11}, {12}, {12}, {13}, {3, 14, 15}, {12}, {16}, {}});
CHECK(build.function.cfg.idoms == std::vector<uint32_t>{~0u, 0, 1, 1, 1, 4, 5, 1, 1, 3, 9, 9, 9, 12, 13, 1, 0});
}
// 'A Linear Time Algorithm for Placing Phi-Nodes' [Vugranam C.Sreedhar]. Figure 4.
TEST_CASE_FIXTURE(IrBuilderFixture, "DominanceVerification3")
{
defineCfgTree({{1, 2}, {3}, {3, 4}, {5}, {5, 6}, {7}, {7}, {}});
CHECK(build.function.cfg.idoms == std::vector<uint32_t>{~0u, 0, 0, 0, 2, 0, 4, 0});
}
Sync to upstream/release/591 (#1012) * Fix a use-after-free bug in the new type cloning algorithm * Tighten up the type of `coroutine.wrap`. It is now `<A..., R...>(f: (A...) -> R...) -> ((A...) -> R...)` * Break `.luaurc` out into a separate library target `Luau.Config`. This makes it easier for applications to reason about config files without also depending on the type inference engine. * Move typechecking limits into `FrontendOptions`. This allows embedders more finely-grained control over autocomplete's internal time limits. * Fix stability issue with debugger onprotectederror callback allowing break in non-yieldable contexts New solver: * Initial work toward [Local Type Inference](https://github.com/Roblox/luau/blob/0e1082108fd6fb3a32dfdf5f1766ea3fc1391328/rfcs/local-type-inference.md) * Introduce a new subtyping test. This will be much nicer than the old test because it is completely separate both from actual type inference and from error reporting. Native code generation: * Added function to compute iterated dominance frontier * Optimize barriers in SET_UPVALUE when tag is known * Cache lua_State::global in a register on A64 * Optimize constant stores in A64 lowering * Track table array size state to optimize array size checks * Add split tag/value store into a VM register * Check that spills can outlive the block only in specific conditions --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-08-19 02:15:41 +08:00
// 'Static Single Assignment Book' Figure 4.1
TEST_CASE_FIXTURE(IrBuilderFixture, "DominanceVerification4")
{
defineCfgTree({{1}, {2, 10}, {3, 7}, {4}, {5}, {4, 6}, {1}, {8}, {5, 9}, {7}, {}});
IdfContext ctx;
computeIteratedDominanceFrontierForDefs(ctx, build.function, {0, 2, 3, 6}, {1, 2, 3, 4, 5, 6, 7, 8, 9, 10});
CHECK(ctx.idf == std::vector<uint32_t>{1, 4, 5});
}
// 'Static Single Assignment Book' Figure 4.5
TEST_CASE_FIXTURE(IrBuilderFixture, "DominanceVerification4")
{
defineCfgTree({{1}, {2}, {3, 7}, {4, 5}, {6}, {6}, {8}, {8}, {9}, {10, 11}, {11}, {9, 12}, {2}});
IdfContext ctx;
computeIteratedDominanceFrontierForDefs(ctx, build.function, {4, 5, 7, 12}, {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12});
CHECK(ctx.idf == std::vector<uint32_t>{2, 6, 8});
// Pruned form, when variable is only live-in in limited set of blocks
computeIteratedDominanceFrontierForDefs(ctx, build.function, {4, 5, 7, 12}, {6, 8, 9});
CHECK(ctx.idf == std::vector<uint32_t>{6, 8});
}
Sync to upstream/release/567 (#860) * Fix #817 * Fix #850 * Optimize math.floor/ceil/round with SSE4.1 * Results in a ~7-9% speedup on the math-cordic benchmark. * Optimized table.sort. * table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x faster when using a simple predicate. Performance may improve further in the future. * Reorganize the memory ownership of builtin type definitions. * This is a small initial step toward affording parallel typechecking. The new type solver is coming along nicely. We are working on fixing crashes and bugs. A few major changes to native codegen landed this week: * Fixed lowering of Luau IR mod instruction when first argument is a constant * Added VM register data-flow/capture analysis * Fixed issues with optimizations in unreachable blocks --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-03-11 04:21:07 +08:00
TEST_SUITE_END();
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
TEST_SUITE_BEGIN("ValueNumbering");
TEST_CASE_FIXTURE(IrBuilderFixture, "RemoveDuplicateCalculation")
{
IrOp entry = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
IrOp op1 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0));
IrOp op2 = build.inst(IrCmd::UNM_NUM, op1);
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(1), op2);
IrOp op3 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0)); // Load propagation is tested here
IrOp op4 = build.inst(IrCmd::UNM_NUM, op3); // And allows value numbering to trigger here
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(2), op4);
build.inst(IrCmd::RETURN, build.vmReg(1), build.constInt(2));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%0 = LOAD_DOUBLE R0
%1 = UNM_NUM %0
STORE_DOUBLE R1, %1
STORE_DOUBLE R2, %1
RETURN R1, 2i
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "LateTableStateLink")
{
IrOp block = build.block(IrBlockKind::Internal);
IrOp fallback = build.block(IrBlockKind::Fallback);
build.beginBlock(block);
IrOp tmp = build.inst(IrCmd::DUP_TABLE, build.vmReg(0));
build.inst(IrCmd::STORE_POINTER, build.vmReg(0), tmp); // Late tmp -> R0 link is tested here
IrOp table = build.inst(IrCmd::LOAD_POINTER, build.vmReg(0)); // Store to load propagation test
build.inst(IrCmd::CHECK_NO_METATABLE, table, fallback);
build.inst(IrCmd::CHECK_READONLY, table, fallback);
build.inst(IrCmd::CHECK_NO_METATABLE, table, fallback);
build.inst(IrCmd::CHECK_READONLY, table, fallback);
build.inst(IrCmd::RETURN, build.constUint(0));
build.beginBlock(fallback);
build.inst(IrCmd::RETURN, build.constUint(1));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%0 = DUP_TABLE R0
STORE_POINTER R0, %0
CHECK_NO_METATABLE %0, bb_fallback_1
CHECK_READONLY %0, bb_fallback_1
RETURN 0u
bb_fallback_1:
RETURN 1u
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "RegisterVersioning")
{
IrOp entry = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
IrOp op1 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0));
IrOp op2 = build.inst(IrCmd::UNM_NUM, op1);
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(0), op2);
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.constTag(tnumber)); // Doesn't prevent previous store propagation
IrOp op3 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0)); // No longer 'op1'
IrOp op4 = build.inst(IrCmd::UNM_NUM, op3);
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(1), op4);
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(2));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%0 = LOAD_DOUBLE R0
%1 = UNM_NUM %0
STORE_DOUBLE R0, %1
STORE_TAG R0, tnumber
%5 = UNM_NUM %1
STORE_DOUBLE R1, %5
RETURN R0, 2i
)");
}
// This can be relaxed in the future when SETLIST becomes aware of register allocator
TEST_CASE_FIXTURE(IrBuilderFixture, "SetListIsABlocker")
{
IrOp entry = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
IrOp op1 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0));
build.inst(IrCmd::SETLIST);
IrOp op2 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0));
IrOp sum = build.inst(IrCmd::ADD_NUM, op1, op2);
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(0), sum);
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(1));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%0 = LOAD_DOUBLE R0
SETLIST
%2 = LOAD_DOUBLE R0
%3 = ADD_NUM %0, %2
STORE_DOUBLE R0, %3
RETURN R0, 1i
)");
}
// Luau call will reuse the same stack and spills will be lost
// However, in the future we might propagate values that can be rematerialized
TEST_CASE_FIXTURE(IrBuilderFixture, "CallIsABlocker")
{
IrOp entry = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
IrOp op1 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0));
build.inst(IrCmd::CALL, build.vmReg(1), build.constInt(1), build.vmReg(2), build.constInt(1));
IrOp op2 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0));
IrOp sum = build.inst(IrCmd::ADD_NUM, op1, op2);
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(1), sum);
build.inst(IrCmd::RETURN, build.vmReg(1), build.constInt(2));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%0 = LOAD_DOUBLE R0
CALL R1, 1i, R2, 1i
%2 = LOAD_DOUBLE R0
%3 = ADD_NUM %0, %2
STORE_DOUBLE R1, %3
RETURN R1, 2i
)");
}
// While constant propagation correctly versions captured registers, IrValueLocationTracking doesn't (yet)
TEST_CASE_FIXTURE(IrBuilderFixture, "NoPropagationOfCapturedRegs")
{
IrOp entry = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
Sync to upstream/release/578 (#939) * Fixed gcc warning about uninitialized `std::optional` * Fixed inlining of functions when they are used to compute their own arguments In the new type solver: * Type families that are not part of a function signature cannot be resolved at instantiation time and will now produce an error. This will be relaxed in the future when we get constraint clauses on function signatures (internally) * `never` type is now comparable * Improved typechecking of `for..in` statements * Fixed checks for number type in `Add` type family * Performance was improved, with particularly large gains on large projects And in native code generation (jit): * We eliminated the call instruction overhead when native code support is enabled in the VM * Small optimizations to arm64 lowering * Reworked LOP_GETIMPORT handling to reduce assembly code size * Fixed non-deterministic binary output * Fixed bad code generation caused by incorrect SSA to VM register links invalidation --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Andy Friesen <afriesen@roblox.com>
2023-05-26 05:36:34 +08:00
build.inst(IrCmd::CAPTURE, build.vmReg(0), build.constUint(1));
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
IrOp op1 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0));
IrOp op2 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0));
IrOp sum = build.inst(IrCmd::ADD_NUM, op1, op2);
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(1), sum);
build.inst(IrCmd::RETURN, build.vmReg(1), build.constInt(1));
updateUseCounts(build.function);
computeCfgInfo(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
; captured regs: R0
bb_0:
; in regs: R0
Sync to upstream/release/578 (#939) * Fixed gcc warning about uninitialized `std::optional` * Fixed inlining of functions when they are used to compute their own arguments In the new type solver: * Type families that are not part of a function signature cannot be resolved at instantiation time and will now produce an error. This will be relaxed in the future when we get constraint clauses on function signatures (internally) * `never` type is now comparable * Improved typechecking of `for..in` statements * Fixed checks for number type in `Add` type family * Performance was improved, with particularly large gains on large projects And in native code generation (jit): * We eliminated the call instruction overhead when native code support is enabled in the VM * Small optimizations to arm64 lowering * Reworked LOP_GETIMPORT handling to reduce assembly code size * Fixed non-deterministic binary output * Fixed bad code generation caused by incorrect SSA to VM register links invalidation --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Andy Friesen <afriesen@roblox.com>
2023-05-26 05:36:34 +08:00
CAPTURE R0, 1u
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
%1 = LOAD_DOUBLE R0
%2 = LOAD_DOUBLE R0
%3 = ADD_NUM %1, %2
STORE_DOUBLE R1, %3
RETURN R1, 1i
)");
}
Sync to upstream/release/579 (#943) A pretty small changelist this week: * When type inference fails to find any matching overload for a function, we were declining to commit any changes to the type graph at all. This was resulting in confusing type errors in certain cases. Now, when a matching overload cannot be found, we always commit to the first overload we tried. JIT * Fix missing variadic register invalidation in FALLBACK_GETVARARGS * Add a missing null pointer check for the result of luaT_gettm --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-06-03 03:52:15 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "NoDeadLoadReuse")
{
IrOp entry = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
IrOp op1 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0));
IrOp op1i = build.inst(IrCmd::NUM_TO_INT, op1);
IrOp res = build.inst(IrCmd::BITAND_UINT, op1i, build.constInt(0));
IrOp resd = build.inst(IrCmd::INT_TO_NUM, res);
IrOp op2 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0));
IrOp sum = build.inst(IrCmd::ADD_NUM, resd, op2);
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(1), sum);
build.inst(IrCmd::RETURN, build.vmReg(1), build.constInt(1));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%4 = LOAD_DOUBLE R0
%5 = ADD_NUM 0, %4
STORE_DOUBLE R1, %5
RETURN R1, 1i
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "NoDeadValueReuse")
{
IrOp entry = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
IrOp op1 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0));
IrOp op1i = build.inst(IrCmd::NUM_TO_INT, op1);
IrOp res = build.inst(IrCmd::BITAND_UINT, op1i, build.constInt(0));
IrOp op2i = build.inst(IrCmd::NUM_TO_INT, op1);
IrOp sum = build.inst(IrCmd::ADD_INT, res, op2i);
IrOp resd = build.inst(IrCmd::INT_TO_NUM, sum);
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(1), resd);
build.inst(IrCmd::RETURN, build.vmReg(1), build.constInt(1));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%0 = LOAD_DOUBLE R0
%3 = NUM_TO_INT %0
%4 = ADD_INT 0i, %3
%5 = INT_TO_NUM %4
STORE_DOUBLE R1, %5
RETURN R1, 1i
)");
}
Sync to upstream/release/591 (#1012) * Fix a use-after-free bug in the new type cloning algorithm * Tighten up the type of `coroutine.wrap`. It is now `<A..., R...>(f: (A...) -> R...) -> ((A...) -> R...)` * Break `.luaurc` out into a separate library target `Luau.Config`. This makes it easier for applications to reason about config files without also depending on the type inference engine. * Move typechecking limits into `FrontendOptions`. This allows embedders more finely-grained control over autocomplete's internal time limits. * Fix stability issue with debugger onprotectederror callback allowing break in non-yieldable contexts New solver: * Initial work toward [Local Type Inference](https://github.com/Roblox/luau/blob/0e1082108fd6fb3a32dfdf5f1766ea3fc1391328/rfcs/local-type-inference.md) * Introduce a new subtyping test. This will be much nicer than the old test because it is completely separate both from actual type inference and from error reporting. Native code generation: * Added function to compute iterated dominance frontier * Optimize barriers in SET_UPVALUE when tag is known * Cache lua_State::global in a register on A64 * Optimize constant stores in A64 lowering * Track table array size state to optimize array size checks * Add split tag/value store into a VM register * Check that spills can outlive the block only in specific conditions --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-08-19 02:15:41 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "TValueLoadToSplitStore")
{
IrOp entry = build.block(IrBlockKind::Internal);
IrOp fallback = build.block(IrBlockKind::Fallback);
build.beginBlock(entry);
IrOp op1 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(0));
IrOp op1v2 = build.inst(IrCmd::ADD_NUM, op1, build.constDouble(4.0));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(1), op1v2);
build.inst(IrCmd::STORE_TAG, build.vmReg(1), build.constTag(tnumber));
// Check that this TValue store will be replaced by a split store
IrOp tv = build.inst(IrCmd::LOAD_TVALUE, build.vmReg(1));
build.inst(IrCmd::STORE_TVALUE, build.vmReg(2), tv);
// Check that tag and value can be extracted from R2 now (removing the fallback)
IrOp tag2 = build.inst(IrCmd::LOAD_TAG, build.vmReg(2));
build.inst(IrCmd::CHECK_TAG, tag2, build.constTag(tnumber), fallback);
IrOp op2 = build.inst(IrCmd::LOAD_DOUBLE, build.vmReg(2));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(3), op2);
build.inst(IrCmd::STORE_TAG, build.vmReg(3), build.constTag(tnumber));
build.inst(IrCmd::RETURN, build.vmReg(1), build.constInt(1));
build.beginBlock(fallback);
build.inst(IrCmd::RETURN, build.vmReg(2), build.constInt(1));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%0 = LOAD_DOUBLE R0
%1 = ADD_NUM %0, 4
STORE_DOUBLE R1, %1
STORE_TAG R1, tnumber
STORE_SPLIT_TVALUE R2, tnumber, %1
STORE_DOUBLE R3, %1
STORE_TAG R3, tnumber
RETURN R1, 1i
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "TagStoreUpdatesValueVersion")
{
IrOp entry = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
IrOp op1 = build.inst(IrCmd::LOAD_POINTER, build.vmReg(0));
build.inst(IrCmd::STORE_POINTER, build.vmReg(1), op1);
build.inst(IrCmd::STORE_TAG, build.vmReg(1), build.constTag(tstring));
IrOp str = build.inst(IrCmd::LOAD_POINTER, build.vmReg(1));
build.inst(IrCmd::STORE_POINTER, build.vmReg(2), str);
build.inst(IrCmd::STORE_TAG, build.vmReg(2), build.constTag(tstring));
build.inst(IrCmd::RETURN, build.vmReg(1), build.constInt(1));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
%0 = LOAD_POINTER R0
STORE_POINTER R1, %0
STORE_TAG R1, tstring
STORE_POINTER R2, %0
STORE_TAG R2, tstring
RETURN R1, 1i
)");
}
TEST_CASE_FIXTURE(IrBuilderFixture, "TagStoreUpdatesSetUpval")
{
IrOp entry = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
build.inst(IrCmd::STORE_TAG, build.vmReg(0), build.constTag(tnumber));
build.inst(IrCmd::STORE_DOUBLE, build.vmReg(0), build.constDouble(0.5));
build.inst(IrCmd::SET_UPVALUE, build.vmUpvalue(0), build.vmReg(0), build.undef());
build.inst(IrCmd::RETURN, build.vmReg(0), build.constInt(0));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
STORE_TAG R0, tnumber
STORE_DOUBLE R0, 0.5
SET_UPVALUE U0, R0, tnumber
RETURN R0, 0i
)");
}
Sync to upstream/release/594 (#1036) * Fixed `Frontend::markDirty` not working on modules that were not typechecked yet * Fixed generic variadic function unification succeeding when it should have reported an error New Type Solver: * Implemented semantic subtyping check for function types Native Code Generation: * Improved performance of numerical loops with a constant step * Simplified IR for `bit32.extract` calls extracting first/last bits * Improved performance of NaN checks
2023-09-08 08:13:49 +08:00
TEST_CASE_FIXTURE(IrBuilderFixture, "TagSelfEqualityCheckRemoval")
{
ScopedFastFlag luauMergeTagLoads{"LuauMergeTagLoads", true};
IrOp entry = build.block(IrBlockKind::Internal);
IrOp trueBlock = build.block(IrBlockKind::Internal);
IrOp falseBlock = build.block(IrBlockKind::Internal);
build.beginBlock(entry);
IrOp tag1 = build.inst(IrCmd::LOAD_TAG, build.vmReg(0));
IrOp tag2 = build.inst(IrCmd::LOAD_TAG, build.vmReg(0));
build.inst(IrCmd::JUMP_EQ_TAG, tag1, tag2, trueBlock, falseBlock);
build.beginBlock(trueBlock);
build.inst(IrCmd::RETURN, build.constUint(1));
build.beginBlock(falseBlock);
build.inst(IrCmd::RETURN, build.constUint(2));
updateUseCounts(build.function);
constPropInBlockChains(build, true);
CHECK("\n" + toString(build.function, /* includeUseInfo */ false) == R"(
bb_0:
JUMP bb_1
bb_1:
RETURN 1u
)");
}
Sync to upstream/release/574 (#910) * Added a limit on how many instructions the Compiler can safely produce (reported by @TheGreatSageEqualToHeaven) C++ API Changes: * With work started on read-only and write-only properties, `Property::type` member variable has been replaced with `TypeId type()` and `setType(TypeId)` functions. * New `LazyType` unwrap callback now has a `void` return type, all that's required from the callback is to write into `unwrapped` field. In our work on the new type solver, the following issues were fixed: * Work has started to support https://github.com/Roblox/luau/pull/77 and https://github.com/Roblox/luau/pull/79 * Refinements are no longer applied on l-values, removing some false-positive errors * Improved overload resolution against expected result type * `Frontend::prepareModuleScope` now works in the new solver * Cofinite strings are now comparable And these are the changes in native code generation (JIT): * Fixed MIN_NUM and MAX_NUM constant fold when one of the arguments is NaN * Added constant folding for number conversions and bit operations * Value spilling and rematerialization is now supported on arm64 * Improved FASTCALL2K IR generation to support second argument constant * Added value numbering and load/store propagation optimizations * Added STORE_VECTOR on arm64, completing the IR lowering on this target
2023-04-29 03:55:13 +08:00
TEST_SUITE_END();
Reference in New Issue Copy Permalink