luau/CodeGen/include/Luau/IrAnalysis.h

181 lines
5.8 KiB
C
Raw Normal View History

// This file is part of the Luau programming language and is licensed under MIT License; see LICENSE.txt for details
#pragma once
#include "Luau/CodeGenCommon.h"
#include <bitset>
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
#include <queue>
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
#include <utility>
#include <vector>
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
#include <stdint.h>
namespace Luau
{
namespace CodeGen
{
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
struct IrBlock;
struct IrFunction;
void updateUseCounts(IrFunction& function);
void updateLastUseLocations(IrFunction& function, const std::vector<uint32_t>& sortedBlocks);
uint32_t getNextInstUse(IrFunction& function, uint32_t targetInstIdx, uint32_t startInstIdx);
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
// Returns how many values are coming into the block (live in) and how many are coming out of the block (live out)
std::pair<uint32_t, uint32_t> getLiveInOutValueCount(IrFunction& function, IrBlock& block);
uint32_t getLiveInValueCount(IrFunction& function, IrBlock& block);
uint32_t getLiveOutValueCount(IrFunction& function, IrBlock& block);
struct RegisterSet
{
std::bitset<256> regs;
// If variadic sequence is active, we track register from which it starts
bool varargSeq = false;
uint8_t varargStart = 0;
};
void requireVariadicSequence(RegisterSet& sourceRs, const RegisterSet& defRs, uint8_t varargStart);
struct BlockOrdering
{
uint32_t depth = 0;
uint32_t preOrder = ~0u;
uint32_t postOrder = ~0u;
bool visited = false;
};
struct CfgInfo
{
std::vector<uint32_t> predecessors;
std::vector<uint32_t> predecessorsOffsets;
std::vector<uint32_t> successors;
std::vector<uint32_t> successorsOffsets;
// Immediate dominators (unique parent in the dominator tree)
std::vector<uint32_t> idoms;
// Children in the dominator tree
std::vector<uint32_t> domChildren;
std::vector<uint32_t> domChildrenOffsets;
std::vector<BlockOrdering> domOrdering;
// VM registers that are live when the block is entered
// Additionally, an active variadic sequence can exist at the entry of the block
std::vector<RegisterSet> in;
// VM registers that are defined inside the block
// It can also contain a variadic sequence definition if that hasn't been consumed inside the block
// Note that this means that checking 'def' set might not be enough to say that register has not been written to
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
std::vector<RegisterSet> def;
// VM registers that are coming out from the block
// These might be registers that are defined inside the block or have been defined at the entry of the block
// Additionally, an active variadic sequence can exist at the exit of the block
std::vector<RegisterSet> out;
// VM registers captured by nested closures
// This set can never have an active variadic sequence
RegisterSet captured;
};
// A quick refresher on dominance and dominator trees:
// * If A is a dominator of B (A dom B), you can never execute B without executing A first
// * A is a strict dominator of B (A sdom B) is similar to previous one but A != B
// * Immediate dominator node N (idom N) is a unique node T so that T sdom N,
// but T does not strictly dominate any other node that dominates N.
// * Dominance frontier is a set of nodes where dominance of a node X ends.
// In practice this is where values established by node X might no longer hold because of join edges from other nodes coming in.
// This is also where PHI instructions in SSA are placed.
void computeCfgImmediateDominators(IrFunction& function);
void computeCfgDominanceTreeChildren(IrFunction& function);
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
struct IdfContext
{
struct BlockAndOrdering
{
uint32_t blockIdx;
BlockOrdering ordering;
bool operator<(const BlockAndOrdering& rhs) const
{
if (ordering.depth != rhs.ordering.depth)
return ordering.depth < rhs.ordering.depth;
return ordering.preOrder < rhs.ordering.preOrder;
}
};
// Using priority queue to work on nodes in the order from the bottom of the dominator tree to the top
// If the depth of keys is equal, DFS order is used to provide strong ordering
std::priority_queue<BlockAndOrdering> queue;
std::vector<uint32_t> worklist;
struct IdfVisitMarks
{
bool seenInQueue = false;
bool seenInWorklist = false;
};
std::vector<IdfVisitMarks> visits;
std::vector<uint32_t> idf;
};
// Compute iterated dominance frontier (IDF or DF+) for a variable, given the set of blocks where that variable is defined
// Providing a set of blocks where the variable is a live-in at the entry helps produce a pruned SSA form (inserted phi nodes will not be dead)
//
// 'Iterated' comes from the definition where we recompute the IDFn+1 = DF(S) while adding IDFn to S until a fixed point is reached
// Iterated dominance frontier has been shown to be equal to the set of nodes where phi instructions have to be inserted
void computeIteratedDominanceFrontierForDefs(
IdfContext& ctx, const IrFunction& function, const std::vector<uint32_t>& defBlocks, const std::vector<uint32_t>& liveInBlocks);
// Function used to update all CFG data
void computeCfgInfo(IrFunction& function);
struct BlockIteratorWrapper
{
const uint32_t* itBegin = nullptr;
const uint32_t* itEnd = nullptr;
bool empty() const
{
return itBegin == itEnd;
}
size_t size() const
{
return size_t(itEnd - itBegin);
}
const uint32_t* begin() const
{
return itBegin;
}
const uint32_t* end() const
{
return itEnd;
}
uint32_t operator[](size_t pos) const
{
CODEGEN_ASSERT(pos < size_t(itEnd - itBegin));
return itBegin[pos];
}
};
BlockIteratorWrapper predecessors(const CfgInfo& cfg, uint32_t blockIdx);
BlockIteratorWrapper successors(const CfgInfo& cfg, uint32_t blockIdx);
BlockIteratorWrapper domChildren(const CfgInfo& cfg, uint32_t blockIdx);
} // namespace CodeGen
} // namespace Luau