luau/CodeGen/src/IrLoweringX64.cpp

2152 lines
73 KiB
C++
Raw Permalink Normal View History

// This file is part of the Luau programming language and is licensed under MIT License; see LICENSE.txt for details
#include "IrLoweringX64.h"
#include "Luau/DenseHash.h"
#include "Luau/IrData.h"
Sync to upstream/release/562 (#828) * Fixed rare use-after-free in analysis during table unification A lot of work these past months went into two new Luau components: * A near full rewrite of the typechecker using a new deferred constraint resolution system * Native code generation for AoT/JiT compilation of VM bytecode into x64 (avx)/arm64 instructions Both of these components are far from finished and we don't provide documentation on building and using them at this point. However, curious community members expressed interest in learning about changes that go into these components each week, so we are now listing them here in the 'sync' pull request descriptions. --- New typechecker can be enabled by setting DebugLuauDeferredConstraintResolution flag to 'true'. It is considered unstable right now, so try it at your own risk. Even though it already provides better type inference than the current one in some cases, our main goal right now is to reach feature parity with current typechecker. Features which improve over the capabilities of the current typechecker are marked as '(NEW)'. Changes to new typechecker: * Regular for loop index and parameters are now typechecked * Invalid type annotations on local variables are ignored to improve autocomplete * Fixed missing autocomplete type suggestions for function arguments * Type reduction is now performed to produce simpler types to be presented to the user (error messages, custom LSPs) * Internally, complex types like '((number | string) & ~(false?)) | string' can be produced, which is just 'string | number' when simplified * Fixed spots where support for unknown and never types was missing * (NEW) Length operator '#' is now valid to use on top table type, this type comes up when doing typeof(x) == "table" guards and isn't available in current typechecker --- Changes to native code generation: * Additional math library fast calls are now lowered to x64: math.ldexp, math.round, math.frexp, math.modf, math.sign and math.clamp
2023-02-03 19:26:13 +00:00
#include "Luau/IrUtils.h"
#include "Luau/IrCallWrapperX64.h"
#include "EmitBuiltinsX64.h"
#include "EmitCommonX64.h"
#include "EmitInstructionX64.h"
#include "NativeState.h"
#include "lstate.h"
#include "lgc.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
2023-03-03 20:21:14 +00:00
namespace X64
{
IrLoweringX64::IrLoweringX64(AssemblyBuilderX64& build, ModuleHelpers& helpers, IrFunction& function, LoweringStats* stats)
: build(build)
, helpers(helpers)
, function(function)
, stats(stats)
, regs(build, function, stats)
, valueTracker(function)
, exitHandlerMap(~0u)
{
valueTracker.setRestoreCallack(&regs, [](void* context, IrInst& inst) {
((IrRegAllocX64*)context)->restore(inst, false);
});
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
build.align(kFunctionAlignment, X64::AlignmentDataX64::Ud2);
}
void IrLoweringX64::storeDoubleAsFloat(OperandX64 dst, IrOp src)
{
ScopedRegX64 tmp{regs, SizeX64::xmmword};
if (src.kind == IrOpKind::Constant)
{
build.vmovss(tmp.reg, build.f32(float(doubleOp(src))));
}
else if (src.kind == IrOpKind::Inst)
{
build.vcvtsd2ss(tmp.reg, regOp(src), regOp(src));
}
else
{
LUAU_ASSERT(!"Unsupported instruction form");
}
build.vmovss(dst, tmp.reg);
}
void IrLoweringX64::lowerInst(IrInst& inst, uint32_t index, const IrBlock& next)
{
regs.currInstIdx = index;
valueTracker.beforeInstLowering(inst);
switch (inst.cmd)
{
case IrCmd::LOAD_TAG:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocReg(SizeX64::dword, index);
if (inst.a.kind == IrOpKind::VmReg)
build.mov(inst.regX64, luauRegTag(vmRegOp(inst.a)));
else if (inst.a.kind == IrOpKind::VmConst)
build.mov(inst.regX64, luauConstantTag(vmConstOp(inst.a)));
// If we have a register, we assume it's a pointer to TValue
// We might introduce explicit operand types in the future to make this more robust
else if (inst.a.kind == IrOpKind::Inst)
build.mov(inst.regX64, dword[regOp(inst.a) + offsetof(TValue, tt)]);
else
LUAU_ASSERT(!"Unsupported instruction form");
break;
case IrCmd::LOAD_POINTER:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocReg(SizeX64::qword, index);
if (inst.a.kind == IrOpKind::VmReg)
build.mov(inst.regX64, luauRegValue(vmRegOp(inst.a)));
else if (inst.a.kind == IrOpKind::VmConst)
build.mov(inst.regX64, luauConstantValue(vmConstOp(inst.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
2023-03-17 19:20:37 +00:00
// If we have a register, we assume it's a pointer to TValue
// We might introduce explicit operand types in the future to make this more robust
else if (inst.a.kind == IrOpKind::Inst)
build.mov(inst.regX64, qword[regOp(inst.a) + offsetof(TValue, value)]);
else
LUAU_ASSERT(!"Unsupported instruction form");
break;
case IrCmd::LOAD_DOUBLE:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocReg(SizeX64::xmmword, index);
if (inst.a.kind == IrOpKind::VmReg)
build.vmovsd(inst.regX64, luauRegValue(vmRegOp(inst.a)));
else if (inst.a.kind == IrOpKind::VmConst)
build.vmovsd(inst.regX64, luauConstantValue(vmConstOp(inst.a)));
else
LUAU_ASSERT(!"Unsupported instruction form");
break;
case IrCmd::LOAD_INT:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocReg(SizeX64::dword, index);
build.mov(inst.regX64, luauRegValueInt(vmRegOp(inst.a)));
break;
case IrCmd::LOAD_TVALUE:
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>
2023-08-18 19:15:41 +01:00
{
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocReg(SizeX64::xmmword, index);
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>
2023-08-18 19:15:41 +01:00
int addrOffset = inst.b.kind != IrOpKind::None ? intOp(inst.b) : 0;
if (inst.a.kind == IrOpKind::VmReg)
build.vmovups(inst.regX64, luauReg(vmRegOp(inst.a)));
else if (inst.a.kind == IrOpKind::VmConst)
build.vmovups(inst.regX64, luauConstant(vmConstOp(inst.a)));
else if (inst.a.kind == IrOpKind::Inst)
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>
2023-08-18 19:15:41 +01:00
build.vmovups(inst.regX64, xmmword[regOp(inst.a) + addrOffset]);
else
LUAU_ASSERT(!"Unsupported instruction form");
break;
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>
2023-08-18 19:15:41 +01:00
}
case IrCmd::LOAD_ENV:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocReg(SizeX64::qword, index);
build.mov(inst.regX64, sClosure);
build.mov(inst.regX64, qword[inst.regX64 + offsetof(Closure, env)]);
break;
case IrCmd::GET_ARR_ADDR:
if (inst.b.kind == IrOpKind::Inst)
{
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::qword, index, {inst.b});
if (dwordReg(inst.regX64) != regOp(inst.b))
build.mov(dwordReg(inst.regX64), regOp(inst.b));
build.shl(dwordReg(inst.regX64), kTValueSizeLog2);
build.add(inst.regX64, qword[regOp(inst.a) + offsetof(Table, array)]);
}
else if (inst.b.kind == IrOpKind::Constant)
{
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::qword, index, {inst.a});
build.mov(inst.regX64, qword[regOp(inst.a) + offsetof(Table, array)]);
if (intOp(inst.b) != 0)
build.lea(inst.regX64, addr[inst.regX64 + intOp(inst.b) * sizeof(TValue)]);
}
else
{
LUAU_ASSERT(!"Unsupported instruction form");
}
break;
case IrCmd::GET_SLOT_NODE_ADDR:
{
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocReg(SizeX64::qword, index);
ScopedRegX64 tmp{regs, SizeX64::qword};
getTableNodeAtCachedSlot(build, tmp.reg, inst.regX64, regOp(inst.a), uintOp(inst.b));
break;
}
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
2023-03-17 19:20:37 +00:00
case IrCmd::GET_HASH_NODE_ADDR:
{
// Custom bit shift value can only be placed in cl
ScopedRegX64 shiftTmp{regs, regs.takeReg(rcx, kInvalidInstIdx)};
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
2023-03-17 19:20:37 +00:00
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocReg(SizeX64::qword, index);
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
2023-03-17 19:20:37 +00:00
ScopedRegX64 tmp{regs, SizeX64::qword};
build.mov(inst.regX64, qword[regOp(inst.a) + offsetof(Table, node)]);
build.mov(dwordReg(tmp.reg), 1);
build.mov(byteReg(shiftTmp.reg), byte[regOp(inst.a) + offsetof(Table, lsizenode)]);
build.shl(dwordReg(tmp.reg), byteReg(shiftTmp.reg));
build.dec(dwordReg(tmp.reg));
build.and_(dwordReg(tmp.reg), uintOp(inst.b));
build.shl(tmp.reg, kLuaNodeSizeLog2);
build.add(inst.regX64, tmp.reg);
break;
};
Sync to upstream/release/588 (#992) Type checker/autocomplete: * `Luau::autocomplete` no longer performs typechecking internally, make sure to run `Frontend::check` before performing autocomplete requests * Autocomplete string suggestions without "" are now only suggested inside the "" * Autocomplete suggestions now include `function (anonymous autofilled)` key with a full suggestion for the function expression (with arguments included) stored in `AutocompleteEntry::insertText` * `AutocompleteEntry::indexedWithSelf` is provided for function call suggestions made with `:` * Cyclic modules now see each other type exports as `any` to prevent memory use-after-free (similar to module return type) Runtime: * Updated inline/loop unroll cost model to better handle assignments (Fixes https://github.com/Roblox/luau/issues/978) * `math.noise` speed was improved by ~30% * `table.concat` speed was improved by ~5-7% * `tonumber` and `tostring` now have fastcall paths that execute ~1.5x and ~2.5x faster respectively (fixes #777) * Fixed crash in `luaL_typename` when index refers to a non-existing value * Fixed potential out of memory scenario when using `string.sub` or `string.char` in a loop * Fixed behavior of some fastcall builtins when called without arguments under -O2 to match original functions * Support for native code execution in VM is now enabled by default (note: native code still has to be generated explicitly) * `Codegen::compile` now accepts `CodeGen_OnlyNativeModules` flag. When set, only modules that have a `--!native` hot-comment at the top will be compiled to native code In our new typechecker: * Generic type packs are no longer considered to be variadic during unification * Timeout and cancellation now works in new solver * Fixed false positive errors around 'table' and 'function' type refinements * Table literals now use covariant unification rules. This is sound since literal has no type specified and has no aliases * Fixed issues with blocked types escaping the constraint solver * Fixed more places where error messages that should've been suppressed were still reported * Fixed errors when iterating over a top table type In our native code generation (jit): * 'DebugLuauAbortingChecks' flag is now supported on A64 * LOP_NEWCLOSURE has been translated to IR
2023-07-28 16:13:53 +01:00
case IrCmd::GET_CLOSURE_UPVAL_ADDR:
{
inst.regX64 = regs.allocRegOrReuse(SizeX64::qword, index, {inst.a});
if (inst.a.kind == IrOpKind::Undef)
{
build.mov(inst.regX64, sClosure);
}
else
{
RegisterX64 cl = regOp(inst.a);
if (inst.regX64 != cl)
build.mov(inst.regX64, cl);
}
build.add(inst.regX64, offsetof(Closure, l.uprefs) + sizeof(TValue) * vmUpvalueOp(inst.b));
break;
}
case IrCmd::STORE_TAG:
if (inst.b.kind == IrOpKind::Constant)
Sync to upstream/release/588 (#992) Type checker/autocomplete: * `Luau::autocomplete` no longer performs typechecking internally, make sure to run `Frontend::check` before performing autocomplete requests * Autocomplete string suggestions without "" are now only suggested inside the "" * Autocomplete suggestions now include `function (anonymous autofilled)` key with a full suggestion for the function expression (with arguments included) stored in `AutocompleteEntry::insertText` * `AutocompleteEntry::indexedWithSelf` is provided for function call suggestions made with `:` * Cyclic modules now see each other type exports as `any` to prevent memory use-after-free (similar to module return type) Runtime: * Updated inline/loop unroll cost model to better handle assignments (Fixes https://github.com/Roblox/luau/issues/978) * `math.noise` speed was improved by ~30% * `table.concat` speed was improved by ~5-7% * `tonumber` and `tostring` now have fastcall paths that execute ~1.5x and ~2.5x faster respectively (fixes #777) * Fixed crash in `luaL_typename` when index refers to a non-existing value * Fixed potential out of memory scenario when using `string.sub` or `string.char` in a loop * Fixed behavior of some fastcall builtins when called without arguments under -O2 to match original functions * Support for native code execution in VM is now enabled by default (note: native code still has to be generated explicitly) * `Codegen::compile` now accepts `CodeGen_OnlyNativeModules` flag. When set, only modules that have a `--!native` hot-comment at the top will be compiled to native code In our new typechecker: * Generic type packs are no longer considered to be variadic during unification * Timeout and cancellation now works in new solver * Fixed false positive errors around 'table' and 'function' type refinements * Table literals now use covariant unification rules. This is sound since literal has no type specified and has no aliases * Fixed issues with blocked types escaping the constraint solver * Fixed more places where error messages that should've been suppressed were still reported * Fixed errors when iterating over a top table type In our native code generation (jit): * 'DebugLuauAbortingChecks' flag is now supported on A64 * LOP_NEWCLOSURE has been translated to IR
2023-07-28 16:13:53 +01:00
{
if (inst.a.kind == IrOpKind::Inst)
build.mov(dword[regOp(inst.a) + offsetof(TValue, tt)], tagOp(inst.b));
else
build.mov(luauRegTag(vmRegOp(inst.a)), tagOp(inst.b));
}
else
{
LUAU_ASSERT(!"Unsupported instruction form");
}
break;
case IrCmd::STORE_POINTER:
{
OperandX64 valueLhs = inst.a.kind == IrOpKind::Inst ? qword[regOp(inst.a) + offsetof(TValue, value)] : luauRegValue(vmRegOp(inst.a));
if (inst.b.kind == IrOpKind::Constant)
{
LUAU_ASSERT(intOp(inst.b) == 0);
build.mov(valueLhs, 0);
}
else if (inst.b.kind == IrOpKind::Inst)
{
build.mov(valueLhs, regOp(inst.b));
}
Sync to upstream/release/588 (#992) Type checker/autocomplete: * `Luau::autocomplete` no longer performs typechecking internally, make sure to run `Frontend::check` before performing autocomplete requests * Autocomplete string suggestions without "" are now only suggested inside the "" * Autocomplete suggestions now include `function (anonymous autofilled)` key with a full suggestion for the function expression (with arguments included) stored in `AutocompleteEntry::insertText` * `AutocompleteEntry::indexedWithSelf` is provided for function call suggestions made with `:` * Cyclic modules now see each other type exports as `any` to prevent memory use-after-free (similar to module return type) Runtime: * Updated inline/loop unroll cost model to better handle assignments (Fixes https://github.com/Roblox/luau/issues/978) * `math.noise` speed was improved by ~30% * `table.concat` speed was improved by ~5-7% * `tonumber` and `tostring` now have fastcall paths that execute ~1.5x and ~2.5x faster respectively (fixes #777) * Fixed crash in `luaL_typename` when index refers to a non-existing value * Fixed potential out of memory scenario when using `string.sub` or `string.char` in a loop * Fixed behavior of some fastcall builtins when called without arguments under -O2 to match original functions * Support for native code execution in VM is now enabled by default (note: native code still has to be generated explicitly) * `Codegen::compile` now accepts `CodeGen_OnlyNativeModules` flag. When set, only modules that have a `--!native` hot-comment at the top will be compiled to native code In our new typechecker: * Generic type packs are no longer considered to be variadic during unification * Timeout and cancellation now works in new solver * Fixed false positive errors around 'table' and 'function' type refinements * Table literals now use covariant unification rules. This is sound since literal has no type specified and has no aliases * Fixed issues with blocked types escaping the constraint solver * Fixed more places where error messages that should've been suppressed were still reported * Fixed errors when iterating over a top table type In our native code generation (jit): * 'DebugLuauAbortingChecks' flag is now supported on A64 * LOP_NEWCLOSURE has been translated to IR
2023-07-28 16:13:53 +01:00
else
{
LUAU_ASSERT(!"Unsupported instruction form");
}
break;
}
case IrCmd::STORE_EXTRA:
if (inst.b.kind == IrOpKind::Constant)
{
if (inst.a.kind == IrOpKind::Inst)
build.mov(dword[regOp(inst.a) + offsetof(TValue, extra)], intOp(inst.b));
else
build.mov(luauRegExtra(vmRegOp(inst.a)), intOp(inst.b));
}
else
{
LUAU_ASSERT(!"Unsupported instruction form");
}
break;
case IrCmd::STORE_DOUBLE:
Sync to upstream/release/603 (#1097) # What's changed? - Record the location of properties for table types (closes #802) - Implement stricter UTF-8 validations as per the RFC (https://github.com/luau-lang/rfcs/pull/1) - Implement `buffer` as a new type in both the old and new solvers. - Changed errors produced by some `buffer` builtins to be a bit more generic to avoid platform-dependent error messages. - Fixed a bug where `Unifier` would copy some persistent types, tripping some internal assertions. - Type checking rules on relational operators is now a little bit more lax. - Improve dead code elimination for some `if` statements with complex always-false conditions ## New type solver - Dataflow analysis now generates phi nodes on exit of branches. - Dataflow analysis avoids producing a new definition for locals or properties that are not owned by that loop. - If a function parameter has been constrained to `never`, report errors at all uses of that parameter within that function. - Switch to using the new `Luau::Set` to replace `std::unordered_set` to alleviate some poor allocation characteristics which was negatively affecting overall performance. - Subtyping can now report many failing reasons instead of just the first one that we happened to find during the test. - Subtyping now also report reasons for type pack mismatches. - When visiting `if` statements or expressions, the resulting context are the common terms in both branches. ## Native codegen - Implement support for `buffer` builtins to its IR for x64 and A64. - Optimized `table.insert` by not inserting a table barrier if it is fastcalled with a constant. ## Internal Contributors Co-authored-by: Aaron Weiss <aaronweiss@roblox.com> Co-authored-by: Alexander McCord <amccord@roblox.com> Co-authored-by: Andy Friesen <afriesen@roblox.com> Co-authored-by: Arseny Kapoulkine <arseny@roblox.com> Co-authored-by: Aviral Goel <agoel@roblox.com> Co-authored-by: Lily Brown <lbrown@roblox.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-11-10 21:10:07 +00:00
{
OperandX64 valueLhs = inst.a.kind == IrOpKind::Inst ? qword[regOp(inst.a) + offsetof(TValue, value)] : luauRegValue(vmRegOp(inst.a));
if (inst.b.kind == IrOpKind::Constant)
{
ScopedRegX64 tmp{regs, SizeX64::xmmword};
build.vmovsd(tmp.reg, build.f64(doubleOp(inst.b)));
Sync to upstream/release/603 (#1097) # What's changed? - Record the location of properties for table types (closes #802) - Implement stricter UTF-8 validations as per the RFC (https://github.com/luau-lang/rfcs/pull/1) - Implement `buffer` as a new type in both the old and new solvers. - Changed errors produced by some `buffer` builtins to be a bit more generic to avoid platform-dependent error messages. - Fixed a bug where `Unifier` would copy some persistent types, tripping some internal assertions. - Type checking rules on relational operators is now a little bit more lax. - Improve dead code elimination for some `if` statements with complex always-false conditions ## New type solver - Dataflow analysis now generates phi nodes on exit of branches. - Dataflow analysis avoids producing a new definition for locals or properties that are not owned by that loop. - If a function parameter has been constrained to `never`, report errors at all uses of that parameter within that function. - Switch to using the new `Luau::Set` to replace `std::unordered_set` to alleviate some poor allocation characteristics which was negatively affecting overall performance. - Subtyping can now report many failing reasons instead of just the first one that we happened to find during the test. - Subtyping now also report reasons for type pack mismatches. - When visiting `if` statements or expressions, the resulting context are the common terms in both branches. ## Native codegen - Implement support for `buffer` builtins to its IR for x64 and A64. - Optimized `table.insert` by not inserting a table barrier if it is fastcalled with a constant. ## Internal Contributors Co-authored-by: Aaron Weiss <aaronweiss@roblox.com> Co-authored-by: Alexander McCord <amccord@roblox.com> Co-authored-by: Andy Friesen <afriesen@roblox.com> Co-authored-by: Arseny Kapoulkine <arseny@roblox.com> Co-authored-by: Aviral Goel <agoel@roblox.com> Co-authored-by: Lily Brown <lbrown@roblox.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-11-10 21:10:07 +00:00
build.vmovsd(valueLhs, tmp.reg);
}
else if (inst.b.kind == IrOpKind::Inst)
{
Sync to upstream/release/603 (#1097) # What's changed? - Record the location of properties for table types (closes #802) - Implement stricter UTF-8 validations as per the RFC (https://github.com/luau-lang/rfcs/pull/1) - Implement `buffer` as a new type in both the old and new solvers. - Changed errors produced by some `buffer` builtins to be a bit more generic to avoid platform-dependent error messages. - Fixed a bug where `Unifier` would copy some persistent types, tripping some internal assertions. - Type checking rules on relational operators is now a little bit more lax. - Improve dead code elimination for some `if` statements with complex always-false conditions ## New type solver - Dataflow analysis now generates phi nodes on exit of branches. - Dataflow analysis avoids producing a new definition for locals or properties that are not owned by that loop. - If a function parameter has been constrained to `never`, report errors at all uses of that parameter within that function. - Switch to using the new `Luau::Set` to replace `std::unordered_set` to alleviate some poor allocation characteristics which was negatively affecting overall performance. - Subtyping can now report many failing reasons instead of just the first one that we happened to find during the test. - Subtyping now also report reasons for type pack mismatches. - When visiting `if` statements or expressions, the resulting context are the common terms in both branches. ## Native codegen - Implement support for `buffer` builtins to its IR for x64 and A64. - Optimized `table.insert` by not inserting a table barrier if it is fastcalled with a constant. ## Internal Contributors Co-authored-by: Aaron Weiss <aaronweiss@roblox.com> Co-authored-by: Alexander McCord <amccord@roblox.com> Co-authored-by: Andy Friesen <afriesen@roblox.com> Co-authored-by: Arseny Kapoulkine <arseny@roblox.com> Co-authored-by: Aviral Goel <agoel@roblox.com> Co-authored-by: Lily Brown <lbrown@roblox.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-11-10 21:10:07 +00:00
build.vmovsd(valueLhs, regOp(inst.b));
}
else
{
LUAU_ASSERT(!"Unsupported instruction form");
}
break;
Sync to upstream/release/603 (#1097) # What's changed? - Record the location of properties for table types (closes #802) - Implement stricter UTF-8 validations as per the RFC (https://github.com/luau-lang/rfcs/pull/1) - Implement `buffer` as a new type in both the old and new solvers. - Changed errors produced by some `buffer` builtins to be a bit more generic to avoid platform-dependent error messages. - Fixed a bug where `Unifier` would copy some persistent types, tripping some internal assertions. - Type checking rules on relational operators is now a little bit more lax. - Improve dead code elimination for some `if` statements with complex always-false conditions ## New type solver - Dataflow analysis now generates phi nodes on exit of branches. - Dataflow analysis avoids producing a new definition for locals or properties that are not owned by that loop. - If a function parameter has been constrained to `never`, report errors at all uses of that parameter within that function. - Switch to using the new `Luau::Set` to replace `std::unordered_set` to alleviate some poor allocation characteristics which was negatively affecting overall performance. - Subtyping can now report many failing reasons instead of just the first one that we happened to find during the test. - Subtyping now also report reasons for type pack mismatches. - When visiting `if` statements or expressions, the resulting context are the common terms in both branches. ## Native codegen - Implement support for `buffer` builtins to its IR for x64 and A64. - Optimized `table.insert` by not inserting a table barrier if it is fastcalled with a constant. ## Internal Contributors Co-authored-by: Aaron Weiss <aaronweiss@roblox.com> Co-authored-by: Alexander McCord <amccord@roblox.com> Co-authored-by: Andy Friesen <afriesen@roblox.com> Co-authored-by: Arseny Kapoulkine <arseny@roblox.com> Co-authored-by: Aviral Goel <agoel@roblox.com> Co-authored-by: Lily Brown <lbrown@roblox.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-11-10 21:10:07 +00:00
}
case IrCmd::STORE_INT:
if (inst.b.kind == IrOpKind::Constant)
build.mov(luauRegValueInt(vmRegOp(inst.a)), intOp(inst.b));
else if (inst.b.kind == IrOpKind::Inst)
build.mov(luauRegValueInt(vmRegOp(inst.a)), regOp(inst.b));
else
LUAU_ASSERT(!"Unsupported instruction form");
break;
case IrCmd::STORE_VECTOR:
storeDoubleAsFloat(luauRegValueVector(vmRegOp(inst.a), 0), inst.b);
storeDoubleAsFloat(luauRegValueVector(vmRegOp(inst.a), 1), inst.c);
storeDoubleAsFloat(luauRegValueVector(vmRegOp(inst.a), 2), inst.d);
break;
case IrCmd::STORE_TVALUE:
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>
2023-08-18 19:15:41 +01:00
{
int addrOffset = inst.c.kind != IrOpKind::None ? intOp(inst.c) : 0;
if (inst.a.kind == IrOpKind::VmReg)
build.vmovups(luauReg(vmRegOp(inst.a)), regOp(inst.b));
else if (inst.a.kind == IrOpKind::Inst)
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>
2023-08-18 19:15:41 +01:00
build.vmovups(xmmword[regOp(inst.a) + addrOffset], regOp(inst.b));
else
LUAU_ASSERT(!"Unsupported instruction form");
break;
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>
2023-08-18 19:15:41 +01:00
}
case IrCmd::STORE_SPLIT_TVALUE:
{
int addrOffset = inst.d.kind != IrOpKind::None ? intOp(inst.d) : 0;
OperandX64 tagLhs = inst.a.kind == IrOpKind::Inst ? dword[regOp(inst.a) + offsetof(TValue, tt) + addrOffset] : luauRegTag(vmRegOp(inst.a));
build.mov(tagLhs, tagOp(inst.b));
if (tagOp(inst.b) == LUA_TBOOLEAN)
{
OperandX64 valueLhs =
inst.a.kind == IrOpKind::Inst ? dword[regOp(inst.a) + offsetof(TValue, value) + addrOffset] : luauRegValueInt(vmRegOp(inst.a));
build.mov(valueLhs, inst.c.kind == IrOpKind::Constant ? OperandX64(intOp(inst.c)) : regOp(inst.c));
}
else if (tagOp(inst.b) == LUA_TNUMBER)
{
OperandX64 valueLhs =
inst.a.kind == IrOpKind::Inst ? qword[regOp(inst.a) + offsetof(TValue, value) + addrOffset] : luauRegValue(vmRegOp(inst.a));
if (inst.c.kind == IrOpKind::Constant)
{
ScopedRegX64 tmp{regs, SizeX64::xmmword};
build.vmovsd(tmp.reg, build.f64(doubleOp(inst.c)));
build.vmovsd(valueLhs, tmp.reg);
}
else
{
build.vmovsd(valueLhs, regOp(inst.c));
}
}
else if (isGCO(tagOp(inst.b)))
{
OperandX64 valueLhs =
inst.a.kind == IrOpKind::Inst ? qword[regOp(inst.a) + offsetof(TValue, value) + addrOffset] : luauRegValue(vmRegOp(inst.a));
build.mov(valueLhs, regOp(inst.c));
}
else
{
LUAU_ASSERT(!"Unsupported instruction form");
}
break;
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>
2023-08-18 19:15:41 +01:00
}
case IrCmd::ADD_INT:
{
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a});
if (inst.a.kind == IrOpKind::Constant)
{
build.lea(inst.regX64, addr[regOp(inst.b) + intOp(inst.a)]);
}
else if (inst.a.kind == IrOpKind::Inst)
{
if (inst.regX64 == regOp(inst.a))
{
if (inst.b.kind == IrOpKind::Inst)
build.add(inst.regX64, regOp(inst.b));
else if (intOp(inst.b) == 1)
build.inc(inst.regX64);
else
build.add(inst.regX64, intOp(inst.b));
}
else
{
if (inst.b.kind == IrOpKind::Inst)
build.lea(inst.regX64, addr[regOp(inst.a) + regOp(inst.b)]);
else
build.lea(inst.regX64, addr[regOp(inst.a) + intOp(inst.b)]);
}
}
else
{
LUAU_ASSERT(!"Unsupported instruction form");
}
break;
}
case IrCmd::SUB_INT:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a});
if (inst.regX64 == regOp(inst.a) && intOp(inst.b) == 1)
build.dec(inst.regX64);
else if (inst.regX64 == regOp(inst.a))
build.sub(inst.regX64, intOp(inst.b));
else
build.lea(inst.regX64, addr[regOp(inst.a) - intOp(inst.b)]);
break;
case IrCmd::ADD_NUM:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::xmmword, index, {inst.a, inst.b});
if (inst.a.kind == IrOpKind::Constant)
{
ScopedRegX64 tmp{regs, SizeX64::xmmword};
build.vmovsd(tmp.reg, memRegDoubleOp(inst.a));
build.vaddsd(inst.regX64, tmp.reg, memRegDoubleOp(inst.b));
}
else
{
build.vaddsd(inst.regX64, regOp(inst.a), memRegDoubleOp(inst.b));
}
break;
case IrCmd::SUB_NUM:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::xmmword, index, {inst.a, inst.b});
if (inst.a.kind == IrOpKind::Constant)
{
ScopedRegX64 tmp{regs, SizeX64::xmmword};
build.vmovsd(tmp.reg, memRegDoubleOp(inst.a));
build.vsubsd(inst.regX64, tmp.reg, memRegDoubleOp(inst.b));
}
else
{
build.vsubsd(inst.regX64, regOp(inst.a), memRegDoubleOp(inst.b));
}
break;
case IrCmd::MUL_NUM:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::xmmword, index, {inst.a, inst.b});
if (inst.a.kind == IrOpKind::Constant)
{
ScopedRegX64 tmp{regs, SizeX64::xmmword};
build.vmovsd(tmp.reg, memRegDoubleOp(inst.a));
build.vmulsd(inst.regX64, tmp.reg, memRegDoubleOp(inst.b));
}
else
{
build.vmulsd(inst.regX64, regOp(inst.a), memRegDoubleOp(inst.b));
}
break;
case IrCmd::DIV_NUM:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::xmmword, index, {inst.a, inst.b});
if (inst.a.kind == IrOpKind::Constant)
{
ScopedRegX64 tmp{regs, SizeX64::xmmword};
build.vmovsd(tmp.reg, memRegDoubleOp(inst.a));
build.vdivsd(inst.regX64, tmp.reg, memRegDoubleOp(inst.b));
}
else
{
build.vdivsd(inst.regX64, regOp(inst.a), memRegDoubleOp(inst.b));
}
break;
case IrCmd::IDIV_NUM:
inst.regX64 = regs.allocRegOrReuse(SizeX64::xmmword, index, {inst.a, inst.b});
if (inst.a.kind == IrOpKind::Constant)
{
ScopedRegX64 tmp{regs, SizeX64::xmmword};
build.vmovsd(tmp.reg, memRegDoubleOp(inst.a));
build.vdivsd(inst.regX64, tmp.reg, memRegDoubleOp(inst.b));
}
else
{
build.vdivsd(inst.regX64, regOp(inst.a), memRegDoubleOp(inst.b));
}
build.vroundsd(inst.regX64, inst.regX64, inst.regX64, RoundingModeX64::RoundToNegativeInfinity);
break;
case IrCmd::MOD_NUM:
{
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::xmmword, index, {inst.a, inst.b});
ScopedRegX64 optLhsTmp{regs};
RegisterX64 lhs;
if (inst.a.kind == IrOpKind::Constant)
{
optLhsTmp.alloc(SizeX64::xmmword);
build.vmovsd(optLhsTmp.reg, memRegDoubleOp(inst.a));
lhs = optLhsTmp.reg;
}
else
{
lhs = regOp(inst.a);
}
if (inst.b.kind == IrOpKind::Inst)
{
ScopedRegX64 tmp{regs, SizeX64::xmmword};
build.vdivsd(tmp.reg, lhs, memRegDoubleOp(inst.b));
build.vroundsd(tmp.reg, tmp.reg, tmp.reg, RoundingModeX64::RoundToNegativeInfinity);
build.vmulsd(tmp.reg, tmp.reg, memRegDoubleOp(inst.b));
build.vsubsd(inst.regX64, lhs, tmp.reg);
}
else
{
ScopedRegX64 tmp1{regs, SizeX64::xmmword};
ScopedRegX64 tmp2{regs, SizeX64::xmmword};
build.vmovsd(tmp1.reg, memRegDoubleOp(inst.b));
build.vdivsd(tmp2.reg, lhs, tmp1.reg);
build.vroundsd(tmp2.reg, tmp2.reg, tmp2.reg, RoundingModeX64::RoundToNegativeInfinity);
build.vmulsd(tmp1.reg, tmp2.reg, tmp1.reg);
build.vsubsd(inst.regX64, lhs, tmp1.reg);
}
break;
}
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
2023-03-03 20:21:14 +00:00
case IrCmd::MIN_NUM:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::xmmword, index, {inst.a, inst.b});
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
2023-03-03 20:21:14 +00:00
if (inst.a.kind == IrOpKind::Constant)
{
ScopedRegX64 tmp{regs, SizeX64::xmmword};
build.vmovsd(tmp.reg, memRegDoubleOp(inst.a));
build.vminsd(inst.regX64, tmp.reg, memRegDoubleOp(inst.b));
}
else
{
build.vminsd(inst.regX64, regOp(inst.a), memRegDoubleOp(inst.b));
}
break;
case IrCmd::MAX_NUM:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::xmmword, index, {inst.a, inst.b});
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
2023-03-03 20:21:14 +00:00
if (inst.a.kind == IrOpKind::Constant)
{
ScopedRegX64 tmp{regs, SizeX64::xmmword};
build.vmovsd(tmp.reg, memRegDoubleOp(inst.a));
build.vmaxsd(inst.regX64, tmp.reg, memRegDoubleOp(inst.b));
}
else
{
build.vmaxsd(inst.regX64, regOp(inst.a), memRegDoubleOp(inst.b));
}
break;
case IrCmd::UNM_NUM:
{
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::xmmword, index, {inst.a});
RegisterX64 src = regOp(inst.a);
if (inst.regX64 == src)
{
build.vxorpd(inst.regX64, inst.regX64, build.f64(-0.0));
}
else
{
build.vmovsd(inst.regX64, src, src);
build.vxorpd(inst.regX64, inst.regX64, build.f64(-0.0));
}
break;
}
case IrCmd::FLOOR_NUM:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::xmmword, index, {inst.a});
build.vroundsd(inst.regX64, inst.regX64, memRegDoubleOp(inst.a), RoundingModeX64::RoundToNegativeInfinity);
break;
case IrCmd::CEIL_NUM:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::xmmword, index, {inst.a});
build.vroundsd(inst.regX64, inst.regX64, memRegDoubleOp(inst.a), RoundingModeX64::RoundToPositiveInfinity);
break;
case IrCmd::ROUND_NUM:
{
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::xmmword, index, {inst.a});
ScopedRegX64 tmp1{regs, SizeX64::xmmword};
ScopedRegX64 tmp2{regs, SizeX64::xmmword};
if (inst.a.kind != IrOpKind::Inst)
build.vmovsd(inst.regX64, memRegDoubleOp(inst.a));
else if (regOp(inst.a) != inst.regX64)
build.vmovsd(inst.regX64, inst.regX64, regOp(inst.a));
build.vandpd(tmp1.reg, inst.regX64, build.f64x2(-0.0, -0.0));
build.vmovsd(tmp2.reg, build.i64(0x3fdfffffffffffff)); // 0.49999999999999994
build.vorpd(tmp1.reg, tmp1.reg, tmp2.reg);
build.vaddsd(inst.regX64, inst.regX64, tmp1.reg);
build.vroundsd(inst.regX64, inst.regX64, inst.regX64, RoundingModeX64::RoundToZero);
break;
}
case IrCmd::SQRT_NUM:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::xmmword, index, {inst.a});
build.vsqrtsd(inst.regX64, inst.regX64, memRegDoubleOp(inst.a));
break;
case IrCmd::ABS_NUM:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::xmmword, index, {inst.a});
if (inst.a.kind != IrOpKind::Inst)
build.vmovsd(inst.regX64, memRegDoubleOp(inst.a));
else if (regOp(inst.a) != inst.regX64)
build.vmovsd(inst.regX64, inst.regX64, regOp(inst.a));
build.vandpd(inst.regX64, inst.regX64, build.i64(~(1LL << 63)));
break;
case IrCmd::NOT_ANY:
{
// TODO: if we have a single user which is a STORE_INT, we are missing the opportunity to write directly to target
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a, inst.b});
Label saveone, savezero, exit;
if (inst.a.kind == IrOpKind::Constant)
{
// Other cases should've been constant folded
LUAU_ASSERT(tagOp(inst.a) == LUA_TBOOLEAN);
}
else
{
build.cmp(regOp(inst.a), LUA_TNIL);
build.jcc(ConditionX64::Equal, saveone);
build.cmp(regOp(inst.a), LUA_TBOOLEAN);
build.jcc(ConditionX64::NotEqual, savezero);
}
if (inst.b.kind == IrOpKind::Constant)
{
// If value is 1, we fallthrough to storing 0
if (intOp(inst.b) == 0)
build.jmp(saveone);
}
else
{
build.cmp(regOp(inst.b), 0);
build.jcc(ConditionX64::Equal, saveone);
}
build.setLabel(savezero);
build.mov(inst.regX64, 0);
build.jmp(exit);
build.setLabel(saveone);
build.mov(inst.regX64, 1);
build.setLabel(exit);
break;
}
case IrCmd::CMP_ANY:
{
IrCondition cond = conditionOp(inst.c);
IrCallWrapperX64 callWrap(regs, build);
callWrap.addArgument(SizeX64::qword, rState);
callWrap.addArgument(SizeX64::qword, luauRegAddress(vmRegOp(inst.a)));
callWrap.addArgument(SizeX64::qword, luauRegAddress(vmRegOp(inst.b)));
if (cond == IrCondition::LessEqual)
callWrap.call(qword[rNativeContext + offsetof(NativeContext, luaV_lessequal)]);
else if (cond == IrCondition::Less)
callWrap.call(qword[rNativeContext + offsetof(NativeContext, luaV_lessthan)]);
else if (cond == IrCondition::Equal)
callWrap.call(qword[rNativeContext + offsetof(NativeContext, luaV_equalval)]);
else
LUAU_ASSERT(!"Unsupported condition");
emitUpdateBase(build);
inst.regX64 = regs.takeReg(eax, index);
break;
}
case IrCmd::JUMP:
jumpOrAbortOnUndef(inst.a, next);
break;
case IrCmd::JUMP_IF_TRUTHY:
jumpIfTruthy(build, vmRegOp(inst.a), labelOp(inst.b), labelOp(inst.c));
jumpOrFallthrough(blockOp(inst.c), next);
break;
case IrCmd::JUMP_IF_FALSY:
jumpIfFalsy(build, vmRegOp(inst.a), labelOp(inst.b), labelOp(inst.c));
jumpOrFallthrough(blockOp(inst.c), next);
break;
case IrCmd::JUMP_EQ_TAG:
{
LUAU_ASSERT(inst.b.kind == IrOpKind::Inst || inst.b.kind == IrOpKind::Constant);
OperandX64 opb = inst.b.kind == IrOpKind::Inst ? regOp(inst.b) : OperandX64(tagOp(inst.b));
if (inst.a.kind == IrOpKind::Constant)
build.cmp(opb, tagOp(inst.a));
else
build.cmp(memRegTagOp(inst.a), opb);
if (isFallthroughBlock(blockOp(inst.d), next))
{
build.jcc(ConditionX64::Equal, labelOp(inst.c));
jumpOrFallthrough(blockOp(inst.d), next);
}
else
{
build.jcc(ConditionX64::NotEqual, labelOp(inst.d));
jumpOrFallthrough(blockOp(inst.c), next);
}
break;
}
case IrCmd::JUMP_CMP_INT:
{
IrCondition cond = conditionOp(inst.c);
if ((cond == IrCondition::Equal || cond == IrCondition::NotEqual) && intOp(inst.b) == 0)
{
bool invert = cond == IrCondition::NotEqual;
build.test(regOp(inst.a), regOp(inst.a));
if (isFallthroughBlock(blockOp(inst.d), next))
{
build.jcc(invert ? ConditionX64::Zero : ConditionX64::NotZero, labelOp(inst.e));
jumpOrFallthrough(blockOp(inst.d), next);
}
else
{
build.jcc(invert ? ConditionX64::NotZero : ConditionX64::Zero, labelOp(inst.d));
jumpOrFallthrough(blockOp(inst.e), next);
}
}
else
{
build.cmp(regOp(inst.a), intOp(inst.b));
build.jcc(getConditionInt(cond), labelOp(inst.d));
jumpOrFallthrough(blockOp(inst.e), next);
}
break;
}
case IrCmd::JUMP_EQ_POINTER:
build.cmp(regOp(inst.a), regOp(inst.b));
build.jcc(ConditionX64::Equal, labelOp(inst.c));
jumpOrFallthrough(blockOp(inst.d), next);
break;
case IrCmd::JUMP_CMP_NUM:
{
IrCondition cond = conditionOp(inst.c);
ScopedRegX64 tmp{regs, SizeX64::xmmword};
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
2023-03-03 20:21:14 +00:00
jumpOnNumberCmp(build, tmp.reg, memRegDoubleOp(inst.a), memRegDoubleOp(inst.b), cond, labelOp(inst.d));
jumpOrFallthrough(blockOp(inst.e), next);
break;
}
Sync to upstream/release/600 (#1076) ### What's Changed - Improve readability of unions and intersections by limiting the number of elements of those types that can be presented on a single line (gated under `FFlag::LuauToStringSimpleCompositeTypesSingleLine`) - Adds a new option to the compiler `--record-stats` to record and output compilation statistics - `if...then...else` expressions are now optimized into `AND/OR` form when possible. ### VM - Add a new `buffer` type to Luau based on the [buffer RFC](https://github.com/Roblox/luau/pull/739) and additional C API functions to work with it; this release does not include the library. - Internal C API to work with string buffers has been updated to align with Lua version more closely ### Native Codegen - Added support for new X64 instruction (rev) and new A64 instruction (bswap) in the assembler - Simplified the way numerical loop condition is translated to IR ### New Type Solver - Operator inference now handled by type families - Created a new system called `Type Paths` to explain why subtyping tests fail in order to improve the quality of error messages. - Systematic changes to implement Data Flow analysis in the new solver (`Breadcrumb` removed and replaced with `RefinementKey`) --- Co-authored-by: Aaron Weiss <aaronweiss@roblox.com> Co-authored-by: Alexander McCord <amccord@roblox.com> Co-authored-by: Andy Friesen <afriesen@roblox.com> Co-authored-by: Aviral Goel <agoel@roblox.com> Co-authored-by: Lily Brown <lbrown@roblox.com> Co-authored-by: Vighnesh Vijay <vvijay@roblox.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com> --------- Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com> Co-authored-by: Andy Friesen <afriesen@roblox.com> Co-authored-by: Lily Brown <lbrown@roblox.com> Co-authored-by: Aaron Weiss <aaronweiss@roblox.com> Co-authored-by: Alexander McCord <amccord@roblox.com> Co-authored-by: Aviral Goel <agoel@roblox.com>
2023-10-21 02:10:30 +01:00
case IrCmd::JUMP_FORN_LOOP_COND:
{
ScopedRegX64 tmp1{regs, SizeX64::xmmword};
ScopedRegX64 tmp2{regs, SizeX64::xmmword};
ScopedRegX64 tmp3{regs, SizeX64::xmmword};
RegisterX64 index = inst.a.kind == IrOpKind::Inst ? regOp(inst.a) : tmp1.reg;
RegisterX64 limit = inst.b.kind == IrOpKind::Inst ? regOp(inst.b) : tmp2.reg;
if (inst.a.kind != IrOpKind::Inst)
build.vmovsd(tmp1.reg, memRegDoubleOp(inst.a));
if (inst.b.kind != IrOpKind::Inst)
build.vmovsd(tmp2.reg, memRegDoubleOp(inst.b));
Label direct;
// step > 0
jumpOnNumberCmp(build, tmp3.reg, memRegDoubleOp(inst.c), build.f64(0.0), IrCondition::Greater, direct);
// !(limit <= index)
jumpOnNumberCmp(build, noreg, limit, index, IrCondition::NotLessEqual, labelOp(inst.e));
build.jmp(labelOp(inst.d));
// !(index <= limit)
build.setLabel(direct);
jumpOnNumberCmp(build, noreg, index, limit, IrCondition::NotLessEqual, labelOp(inst.e));
jumpOrFallthrough(blockOp(inst.d), next);
break;
}
case IrCmd::TABLE_LEN:
{
IrCallWrapperX64 callWrap(regs, build, index);
callWrap.addArgument(SizeX64::qword, regOp(inst.a), inst.a);
callWrap.call(qword[rNativeContext + offsetof(NativeContext, luaH_getn)]);
inst.regX64 = regs.takeReg(eax, index);
break;
}
case IrCmd::TABLE_SETNUM:
{
IrCallWrapperX64 callWrap(regs, build, index);
callWrap.addArgument(SizeX64::qword, rState);
callWrap.addArgument(SizeX64::qword, regOp(inst.a), inst.a);
callWrap.addArgument(SizeX64::dword, regOp(inst.b), inst.b);
callWrap.call(qword[rNativeContext + offsetof(NativeContext, luaH_setnum)]);
inst.regX64 = regs.takeReg(rax, index);
break;
}
case IrCmd::STRING_LEN:
{
RegisterX64 ptr = regOp(inst.a);
inst.regX64 = regs.allocReg(SizeX64::dword, index);
build.mov(inst.regX64, dword[ptr + offsetof(TString, len)]);
break;
}
case IrCmd::NEW_TABLE:
{
IrCallWrapperX64 callWrap(regs, build, index);
callWrap.addArgument(SizeX64::qword, rState);
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
callWrap.addArgument(SizeX64::dword, int32_t(uintOp(inst.a)));
callWrap.addArgument(SizeX64::dword, int32_t(uintOp(inst.b)));
callWrap.call(qword[rNativeContext + offsetof(NativeContext, luaH_new)]);
inst.regX64 = regs.takeReg(rax, index);
break;
}
case IrCmd::DUP_TABLE:
{
IrCallWrapperX64 callWrap(regs, build, index);
callWrap.addArgument(SizeX64::qword, rState);
callWrap.addArgument(SizeX64::qword, regOp(inst.a), inst.a);
callWrap.call(qword[rNativeContext + offsetof(NativeContext, luaH_clone)]);
inst.regX64 = regs.takeReg(rax, index);
break;
}
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
2023-03-17 19:20:37 +00:00
case IrCmd::TRY_NUM_TO_INDEX:
{
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocReg(SizeX64::dword, index);
ScopedRegX64 tmp{regs, SizeX64::xmmword};
convertNumberToIndexOrJump(build, tmp.reg, regOp(inst.a), inst.regX64, labelOp(inst.b));
break;
}
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
2023-03-17 19:20:37 +00:00
case IrCmd::TRY_CALL_FASTGETTM:
{
ScopedRegX64 tmp{regs, SizeX64::qword};
build.mov(tmp.reg, qword[regOp(inst.a) + offsetof(Table, metatable)]);
regs.freeLastUseReg(function.instOp(inst.a), index); // Release before the call if it's the last use
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
2023-03-17 19:20:37 +00:00
build.test(tmp.reg, tmp.reg);
build.jcc(ConditionX64::Zero, labelOp(inst.c)); // No metatable
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
2023-03-17 19:20:37 +00:00
build.test(byte[tmp.reg + offsetof(Table, tmcache)], 1 << intOp(inst.b));
build.jcc(ConditionX64::NotZero, labelOp(inst.c)); // No tag method
ScopedRegX64 tmp2{regs, SizeX64::qword};
build.mov(tmp2.reg, qword[rState + offsetof(lua_State, global)]);
{
ScopedSpills spillGuard(regs);
IrCallWrapperX64 callWrap(regs, build, index);
callWrap.addArgument(SizeX64::qword, tmp);
callWrap.addArgument(SizeX64::qword, intOp(inst.b));
callWrap.addArgument(SizeX64::qword, qword[tmp2.release() + offsetof(global_State, tmname) + intOp(inst.b) * sizeof(TString*)]);
callWrap.call(qword[rNativeContext + offsetof(NativeContext, luaT_gettm)]);
}
build.test(rax, rax);
build.jcc(ConditionX64::Zero, labelOp(inst.c)); // No tag method
inst.regX64 = regs.takeReg(rax, index);
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
2023-03-17 19:20:37 +00:00
break;
}
case IrCmd::INT_TO_NUM:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
inst.regX64 = regs.allocReg(SizeX64::xmmword, index);
build.vcvtsi2sd(inst.regX64, inst.regX64, regOp(inst.a));
break;
case IrCmd::UINT_TO_NUM:
inst.regX64 = regs.allocReg(SizeX64::xmmword, index);
// AVX has no uint->double conversion; the source must come from UINT op and they all should clear top 32 bits so we can usually
// use 64-bit reg; the one exception is NUM_TO_UINT which doesn't clear top bits
if (IrCmd source = function.instOp(inst.a).cmd; source == IrCmd::NUM_TO_UINT)
{
ScopedRegX64 tmp{regs, SizeX64::dword};
build.mov(tmp.reg, regOp(inst.a));
build.vcvtsi2sd(inst.regX64, inst.regX64, qwordReg(tmp.reg));
}
else
{
LUAU_ASSERT(source != IrCmd::SUBSTITUTE); // we don't process substitutions
build.vcvtsi2sd(inst.regX64, inst.regX64, qwordReg(regOp(inst.a)));
}
break;
case IrCmd::NUM_TO_INT:
inst.regX64 = regs.allocReg(SizeX64::dword, index);
build.vcvttsd2si(inst.regX64, memRegDoubleOp(inst.a));
break;
case IrCmd::NUM_TO_UINT:
inst.regX64 = regs.allocReg(SizeX64::dword, index);
build.vcvttsd2si(qwordReg(inst.regX64), memRegDoubleOp(inst.a));
break;
case IrCmd::ADJUST_STACK_TO_REG:
{
ScopedRegX64 tmp{regs, SizeX64::qword};
if (inst.b.kind == IrOpKind::Constant)
{
build.lea(tmp.reg, addr[rBase + (vmRegOp(inst.a) + intOp(inst.b)) * sizeof(TValue)]);
build.mov(qword[rState + offsetof(lua_State, top)], tmp.reg);
}
else if (inst.b.kind == IrOpKind::Inst)
{
build.mov(dwordReg(tmp.reg), regOp(inst.b));
build.shl(tmp.reg, kTValueSizeLog2);
build.lea(tmp.reg, addr[rBase + tmp.reg + vmRegOp(inst.a) * sizeof(TValue)]);
build.mov(qword[rState + offsetof(lua_State, top)], tmp.reg);
}
else
{
LUAU_ASSERT(!"Unsupported instruction form");
}
break;
}
case IrCmd::ADJUST_STACK_TO_TOP:
{
ScopedRegX64 tmp{regs, SizeX64::qword};
build.mov(tmp.reg, qword[rState + offsetof(lua_State, ci)]);
build.mov(tmp.reg, qword[tmp.reg + offsetof(CallInfo, top)]);
build.mov(qword[rState + offsetof(lua_State, top)], tmp.reg);
break;
}
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
2023-03-03 20:21:14 +00:00
case IrCmd::FASTCALL:
{
OperandX64 arg2 = inst.d.kind != IrOpKind::Undef ? memRegDoubleOp(inst.d) : OperandX64{0};
emitBuiltin(regs, build, uintOp(inst.a), vmRegOp(inst.b), vmRegOp(inst.c), arg2, intOp(inst.e), intOp(inst.f));
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
2023-03-03 20:21:14 +00:00
break;
}
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
2023-03-03 20:21:14 +00:00
case IrCmd::INVOKE_FASTCALL:
{
unsigned bfid = uintOp(inst.a);
OperandX64 args = 0;
if (inst.d.kind == IrOpKind::VmReg)
args = luauRegAddress(vmRegOp(inst.d));
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
2023-03-03 20:21:14 +00:00
else if (inst.d.kind == IrOpKind::VmConst)
args = luauConstantAddress(vmConstOp(inst.d));
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
2023-03-03 20:21:14 +00:00
else
LUAU_ASSERT(inst.d.kind == IrOpKind::Undef);
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
2023-03-03 20:21:14 +00:00
int ra = vmRegOp(inst.b);
int arg = vmRegOp(inst.c);
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
2023-03-03 20:21:14 +00:00
int nparams = intOp(inst.e);
int nresults = intOp(inst.f);
IrCallWrapperX64 callWrap(regs, build, index);
callWrap.addArgument(SizeX64::qword, rState);
callWrap.addArgument(SizeX64::qword, luauRegAddress(ra));
callWrap.addArgument(SizeX64::qword, luauRegAddress(arg));
callWrap.addArgument(SizeX64::dword, nresults);
callWrap.addArgument(SizeX64::qword, args);
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
2023-03-03 20:21:14 +00:00
if (nparams == LUA_MULTRET)
{
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
RegisterX64 reg = callWrap.suggestNextArgumentRegister(SizeX64::qword);
ScopedRegX64 tmp{regs, SizeX64::qword};
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
// L->top - (ra + 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
2023-03-03 20:21:14 +00:00
build.mov(reg, qword[rState + offsetof(lua_State, top)]);
build.lea(tmp.reg, addr[rBase + (ra + 1) * sizeof(TValue)]);
build.sub(reg, tmp.reg);
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
2023-03-03 20:21:14 +00:00
build.shr(reg, kTValueSizeLog2);
callWrap.addArgument(SizeX64::dword, dwordReg(reg));
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
2023-03-03 20:21:14 +00:00
}
else
{
callWrap.addArgument(SizeX64::dword, nparams);
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
2023-03-03 20:21:14 +00:00
}
ScopedRegX64 func{regs, SizeX64::qword};
build.mov(func.reg, qword[rNativeContext + offsetof(NativeContext, luauF_table) + bfid * sizeof(luau_FastFunction)]);
callWrap.call(func.release());
inst.regX64 = regs.takeReg(eax, index); // Result of a builtin call is returned in eax
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
2023-03-03 20:21:14 +00:00
break;
}
case IrCmd::CHECK_FASTCALL_RES:
{
RegisterX64 res = regOp(inst.a);
build.test(res, res); // test here will set SF=1 for a negative number and it always sets OF to 0
build.jcc(ConditionX64::Less, labelOp(inst.b)); // jl jumps if SF != OF
break;
}
case IrCmd::DO_ARITH:
Add SUBRK and DIVRK bytecode instructions to bytecode v5 (#1115) Right now, we can compile R\*K for all arithmetic instructions, but K\*R gets compiled into two instructions (LOADN/LOADK + arithmetic opcode). This is problematic since it leads to reduced performance for some code. However, we'd like to avoid adding reverse variants of ADDK et al for all opcodes to avoid the increase in I$ footprint for interpreter. Looking at the arithmetic instructions, % and // don't have interesting use cases for K\*V; ^ is sometimes used with constant on the left hand side but this would need to call pow() by necessity in all cases so it would be slow regardless of the dispatch overhead. This leaves the four basic arithmetic operations. For + and \*, we can implement a compiler-side optimization in the future that transforms K\*R to R\*K automatically. This could either be done unconditionally at -O2, or conditionally based on the type of the value (driven by type annotations / inference) -- this technically changes behavior in presence of metamethods, although it might be sensible to just always do this because non-commutative +/* are evil. However, for - and / it is impossible for the compiler to optimize this in the future, so we need dedicated opcodes. This only increases the interpreter size by ~300 bytes (~1.5%) on X64. This makes spectral-norm and math-partial-sums 6% faster; maybe more importantly, voxelgen gets 1.5% faster (so this change does have real-world impact). To avoid the proliferation of bytecode versions this change piggybacks on the bytecode version bump that was just made in 604 for vector constants; we would still be able to enable these independently but we'll consider v5 complete when both are enabled. Related: #626 --------- Co-authored-by: vegorov-rbx <75688451+vegorov-rbx@users.noreply.github.com>
2023-11-28 15:35:01 +00:00
{
OperandX64 opb = inst.b.kind == IrOpKind::VmReg ? luauRegAddress(vmRegOp(inst.b)) : luauConstantAddress(vmConstOp(inst.b));
OperandX64 opc = inst.c.kind == IrOpKind::VmReg ? luauRegAddress(vmRegOp(inst.c)) : luauConstantAddress(vmConstOp(inst.c));
callArithHelper(regs, build, vmRegOp(inst.a), opb, opc, TMS(intOp(inst.d)));
break;
Add SUBRK and DIVRK bytecode instructions to bytecode v5 (#1115) Right now, we can compile R\*K for all arithmetic instructions, but K\*R gets compiled into two instructions (LOADN/LOADK + arithmetic opcode). This is problematic since it leads to reduced performance for some code. However, we'd like to avoid adding reverse variants of ADDK et al for all opcodes to avoid the increase in I$ footprint for interpreter. Looking at the arithmetic instructions, % and // don't have interesting use cases for K\*V; ^ is sometimes used with constant on the left hand side but this would need to call pow() by necessity in all cases so it would be slow regardless of the dispatch overhead. This leaves the four basic arithmetic operations. For + and \*, we can implement a compiler-side optimization in the future that transforms K\*R to R\*K automatically. This could either be done unconditionally at -O2, or conditionally based on the type of the value (driven by type annotations / inference) -- this technically changes behavior in presence of metamethods, although it might be sensible to just always do this because non-commutative +/* are evil. However, for - and / it is impossible for the compiler to optimize this in the future, so we need dedicated opcodes. This only increases the interpreter size by ~300 bytes (~1.5%) on X64. This makes spectral-norm and math-partial-sums 6% faster; maybe more importantly, voxelgen gets 1.5% faster (so this change does have real-world impact). To avoid the proliferation of bytecode versions this change piggybacks on the bytecode version bump that was just made in 604 for vector constants; we would still be able to enable these independently but we'll consider v5 complete when both are enabled. Related: #626 --------- Co-authored-by: vegorov-rbx <75688451+vegorov-rbx@users.noreply.github.com>
2023-11-28 15:35:01 +00:00
}
case IrCmd::DO_LEN:
callLengthHelper(regs, build, vmRegOp(inst.a), vmRegOp(inst.b));
break;
case IrCmd::GET_TABLE:
if (inst.c.kind == IrOpKind::VmReg)
{
callGetTable(regs, build, vmRegOp(inst.b), luauRegAddress(vmRegOp(inst.c)), vmRegOp(inst.a));
}
else if (inst.c.kind == IrOpKind::Constant)
{
TValue n = {};
setnvalue(&n, uintOp(inst.c));
callGetTable(regs, build, vmRegOp(inst.b), build.bytes(&n, sizeof(n)), vmRegOp(inst.a));
}
else
{
LUAU_ASSERT(!"Unsupported instruction form");
}
break;
case IrCmd::SET_TABLE:
if (inst.c.kind == IrOpKind::VmReg)
{
callSetTable(regs, build, vmRegOp(inst.b), luauRegAddress(vmRegOp(inst.c)), vmRegOp(inst.a));
}
else if (inst.c.kind == IrOpKind::Constant)
{
TValue n = {};
setnvalue(&n, uintOp(inst.c));
callSetTable(regs, build, vmRegOp(inst.b), build.bytes(&n, sizeof(n)), vmRegOp(inst.a));
}
else
{
LUAU_ASSERT(!"Unsupported instruction form");
}
break;
case IrCmd::GET_IMPORT:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
{
ScopedRegX64 tmp1{regs, SizeX64::qword};
build.mov(tmp1.reg, sClosure);
IrCallWrapperX64 callWrap(regs, build, index);
callWrap.addArgument(SizeX64::qword, rState);
callWrap.addArgument(SizeX64::qword, qword[tmp1.release() + offsetof(Closure, env)]);
callWrap.addArgument(SizeX64::qword, rConstants);
callWrap.addArgument(SizeX64::qword, luauRegAddress(vmRegOp(inst.a)));
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
callWrap.addArgument(SizeX64::dword, uintOp(inst.b));
callWrap.addArgument(SizeX64::dword, 0);
callWrap.call(qword[rNativeContext + offsetof(NativeContext, luaV_getimport)]);
emitUpdateBase(build);
break;
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
}
case IrCmd::CONCAT:
{
IrCallWrapperX64 callWrap(regs, build, index);
callWrap.addArgument(SizeX64::qword, rState);
callWrap.addArgument(SizeX64::dword, int32_t(uintOp(inst.b)));
callWrap.addArgument(SizeX64::dword, int32_t(vmRegOp(inst.a) + uintOp(inst.b) - 1));
callWrap.call(qword[rNativeContext + offsetof(NativeContext, luaV_concat)]);
emitUpdateBase(build);
break;
}
case IrCmd::GET_UPVALUE:
{
ScopedRegX64 tmp1{regs, SizeX64::qword};
ScopedRegX64 tmp2{regs, SizeX64::xmmword};
build.mov(tmp1.reg, sClosure);
build.add(tmp1.reg, offsetof(Closure, l.uprefs) + sizeof(TValue) * vmUpvalueOp(inst.b));
// uprefs[] is either an actual value, or it points to UpVal object which has a pointer to value
Label skip;
build.cmp(dword[tmp1.reg + offsetof(TValue, tt)], LUA_TUPVAL);
build.jcc(ConditionX64::NotEqual, skip);
// UpVal.v points to the value (either on stack, or on heap inside each UpVal, but we can deref it unconditionally)
build.mov(tmp1.reg, qword[tmp1.reg + offsetof(TValue, value.gc)]);
build.mov(tmp1.reg, qword[tmp1.reg + offsetof(UpVal, v)]);
build.setLabel(skip);
build.vmovups(tmp2.reg, xmmword[tmp1.reg]);
build.vmovups(luauReg(vmRegOp(inst.a)), tmp2.reg);
break;
}
case IrCmd::SET_UPVALUE:
{
ScopedRegX64 tmp1{regs, SizeX64::qword};
ScopedRegX64 tmp2{regs, SizeX64::qword};
build.mov(tmp1.reg, sClosure);
build.mov(tmp2.reg, qword[tmp1.reg + offsetof(Closure, l.uprefs) + sizeof(TValue) * vmUpvalueOp(inst.a) + offsetof(TValue, value.gc)]);
build.mov(tmp1.reg, qword[tmp2.reg + offsetof(UpVal, v)]);
{
ScopedRegX64 tmp3{regs, SizeX64::xmmword};
build.vmovups(tmp3.reg, luauReg(vmRegOp(inst.b)));
build.vmovups(xmmword[tmp1.reg], tmp3.reg);
}
tmp1.free();
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>
2023-08-18 19:15:41 +01:00
if (inst.c.kind == IrOpKind::Undef || isGCO(tagOp(inst.c)))
callBarrierObject(regs, build, tmp2.release(), {}, inst.b, inst.c.kind == IrOpKind::Undef ? -1 : tagOp(inst.c));
break;
}
case IrCmd::CHECK_TAG:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
build.cmp(memRegTagOp(inst.a), tagOp(inst.b));
jumpOrAbortOnUndef(ConditionX64::NotEqual, inst.c, next);
break;
case IrCmd::CHECK_TRUTHY:
{
// Constant tags which don't require boolean value check should've been removed in constant folding
LUAU_ASSERT(inst.a.kind != IrOpKind::Constant || tagOp(inst.a) == LUA_TBOOLEAN);
Label skip;
if (inst.a.kind != IrOpKind::Constant)
{
// Fail to fallback on 'nil' (falsy)
build.cmp(memRegTagOp(inst.a), LUA_TNIL);
jumpOrAbortOnUndef(ConditionX64::Equal, inst.c, next);
// Skip value test if it's not a boolean (truthy)
build.cmp(memRegTagOp(inst.a), LUA_TBOOLEAN);
build.jcc(ConditionX64::NotEqual, skip);
}
// fail to fallback on 'false' boolean value (falsy)
build.cmp(memRegUintOp(inst.b), 0);
jumpOrAbortOnUndef(ConditionX64::Equal, inst.c, next);
if (inst.a.kind != IrOpKind::Constant)
build.setLabel(skip);
break;
}
case IrCmd::CHECK_READONLY:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
build.cmp(byte[regOp(inst.a) + offsetof(Table, readonly)], 0);
jumpOrAbortOnUndef(ConditionX64::NotEqual, inst.b, next);
break;
case IrCmd::CHECK_NO_METATABLE:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
build.cmp(qword[regOp(inst.a) + offsetof(Table, metatable)], 0);
jumpOrAbortOnUndef(ConditionX64::NotEqual, inst.b, next);
break;
case IrCmd::CHECK_SAFE_ENV:
{
ScopedRegX64 tmp{regs, SizeX64::qword};
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
build.mov(tmp.reg, sClosure);
build.mov(tmp.reg, qword[tmp.reg + offsetof(Closure, env)]);
build.cmp(byte[tmp.reg + offsetof(Table, safeenv)], 0);
jumpOrAbortOnUndef(ConditionX64::Equal, inst.a, next);
break;
}
case IrCmd::CHECK_ARRAY_SIZE:
if (inst.b.kind == IrOpKind::Inst)
build.cmp(dword[regOp(inst.a) + offsetof(Table, sizearray)], regOp(inst.b));
else if (inst.b.kind == IrOpKind::Constant)
build.cmp(dword[regOp(inst.a) + offsetof(Table, sizearray)], intOp(inst.b));
else
LUAU_ASSERT(!"Unsupported instruction form");
jumpOrAbortOnUndef(ConditionX64::BelowEqual, inst.c, next);
break;
case IrCmd::JUMP_SLOT_MATCH:
case IrCmd::CHECK_SLOT_MATCH:
{
Label abort; // Used when guard aborts execution
const IrOp& mismatchOp = inst.cmd == IrCmd::JUMP_SLOT_MATCH ? inst.d : inst.c;
Label& mismatch = mismatchOp.kind == IrOpKind::Undef ? abort : labelOp(mismatchOp);
ScopedRegX64 tmp{regs, SizeX64::qword};
// Check if node key tag is a string
build.mov(dwordReg(tmp.reg), luauNodeKeyTag(regOp(inst.a)));
build.and_(dwordReg(tmp.reg), kTKeyTagMask);
build.cmp(dwordReg(tmp.reg), LUA_TSTRING);
build.jcc(ConditionX64::NotEqual, mismatch);
// Check that node key value matches the expected one
build.mov(tmp.reg, luauConstantValue(vmConstOp(inst.b)));
build.cmp(tmp.reg, luauNodeKeyValue(regOp(inst.a)));
build.jcc(ConditionX64::NotEqual, mismatch);
// Check that node value is not nil
build.cmp(dword[regOp(inst.a) + offsetof(LuaNode, val) + offsetof(TValue, tt)], LUA_TNIL);
build.jcc(ConditionX64::Equal, mismatch);
if (inst.cmd == IrCmd::JUMP_SLOT_MATCH)
{
jumpOrFallthrough(blockOp(inst.c), next);
}
else if (mismatchOp.kind == IrOpKind::Undef)
{
Label skip;
build.jmp(skip);
build.setLabel(abort);
build.ud2();
build.setLabel(skip);
}
break;
}
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
2023-03-17 19:20:37 +00:00
case IrCmd::CHECK_NODE_NO_NEXT:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
{
ScopedRegX64 tmp{regs, SizeX64::dword};
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>
2023-05-19 20:37:30 +01:00
build.mov(tmp.reg, dword[regOp(inst.a) + offsetof(LuaNode, key) + kOffsetOfTKeyTagNext]);
build.shr(tmp.reg, kTKeyTagBits);
jumpOrAbortOnUndef(ConditionX64::NotZero, inst.b, next);
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
2023-03-17 19:20:37 +00:00
break;
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
}
case IrCmd::CHECK_NODE_VALUE:
{
build.cmp(dword[regOp(inst.a) + offsetof(LuaNode, val) + offsetof(TValue, tt)], LUA_TNIL);
jumpOrAbortOnUndef(ConditionX64::Equal, inst.b, next);
break;
}
Sync to upstream/release/603 (#1097) # What's changed? - Record the location of properties for table types (closes #802) - Implement stricter UTF-8 validations as per the RFC (https://github.com/luau-lang/rfcs/pull/1) - Implement `buffer` as a new type in both the old and new solvers. - Changed errors produced by some `buffer` builtins to be a bit more generic to avoid platform-dependent error messages. - Fixed a bug where `Unifier` would copy some persistent types, tripping some internal assertions. - Type checking rules on relational operators is now a little bit more lax. - Improve dead code elimination for some `if` statements with complex always-false conditions ## New type solver - Dataflow analysis now generates phi nodes on exit of branches. - Dataflow analysis avoids producing a new definition for locals or properties that are not owned by that loop. - If a function parameter has been constrained to `never`, report errors at all uses of that parameter within that function. - Switch to using the new `Luau::Set` to replace `std::unordered_set` to alleviate some poor allocation characteristics which was negatively affecting overall performance. - Subtyping can now report many failing reasons instead of just the first one that we happened to find during the test. - Subtyping now also report reasons for type pack mismatches. - When visiting `if` statements or expressions, the resulting context are the common terms in both branches. ## Native codegen - Implement support for `buffer` builtins to its IR for x64 and A64. - Optimized `table.insert` by not inserting a table barrier if it is fastcalled with a constant. ## Internal Contributors Co-authored-by: Aaron Weiss <aaronweiss@roblox.com> Co-authored-by: Alexander McCord <amccord@roblox.com> Co-authored-by: Andy Friesen <afriesen@roblox.com> Co-authored-by: Arseny Kapoulkine <arseny@roblox.com> Co-authored-by: Aviral Goel <agoel@roblox.com> Co-authored-by: Lily Brown <lbrown@roblox.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-11-10 21:10:07 +00:00
case IrCmd::CHECK_BUFFER_LEN:
{
int accessSize = intOp(inst.c);
LUAU_ASSERT(accessSize > 0);
if (inst.b.kind == IrOpKind::Inst)
{
if (accessSize == 1)
{
// Simpler check for a single byte access
build.cmp(dword[regOp(inst.a) + offsetof(Buffer, len)], regOp(inst.b));
jumpOrAbortOnUndef(ConditionX64::BelowEqual, inst.d, next);
}
else
{
ScopedRegX64 tmp1{regs, SizeX64::qword};
ScopedRegX64 tmp2{regs, SizeX64::dword};
// To perform the bounds check using a single branch, we take index that is limited to 32 bit int
// Access size is then added using a 64 bit addition
// This will make sure that addition will not wrap around for values like 0xffffffff
if (IrCmd source = function.instOp(inst.b).cmd; source == IrCmd::NUM_TO_INT)
{
// When previous operation is a conversion to an integer (common case), it is guaranteed to have high register bits cleared
build.lea(tmp1.reg, addr[qwordReg(regOp(inst.b)) + accessSize]);
}
else
{
// When the source of the index is unknown, it could contain garbage in the high bits, so we zero-extend it explicitly
build.mov(dwordReg(tmp1.reg), regOp(inst.b));
build.add(tmp1.reg, accessSize);
}
build.mov(tmp2.reg, dword[regOp(inst.a) + offsetof(Buffer, len)]);
build.cmp(qwordReg(tmp2.reg), tmp1.reg);
jumpOrAbortOnUndef(ConditionX64::Below, inst.d, next);
}
}
else if (inst.b.kind == IrOpKind::Constant)
{
int offset = intOp(inst.b);
// Constant folding can take care of it, but for safety we avoid overflow/underflow cases here
if (offset < 0 || unsigned(offset) + unsigned(accessSize) >= unsigned(INT_MAX))
jumpOrAbortOnUndef(inst.d, next);
else
build.cmp(dword[regOp(inst.a) + offsetof(Buffer, len)], offset + accessSize);
jumpOrAbortOnUndef(ConditionX64::Below, inst.d, next);
}
else
{
LUAU_ASSERT(!"Unsupported instruction form");
}
break;
}
case IrCmd::INTERRUPT:
{
unsigned pcpos = uintOp(inst.a);
// We unconditionally spill values here because that allows us to ignore register state when we synthesize interrupt handler
// This can be changed in the future if we can somehow record interrupt handler code separately
// Since interrupts are loop edges or call/ret, we don't have a significant opportunity for register reuse here anyway
regs.preserveAndFreeInstValues();
ScopedRegX64 tmp{regs, SizeX64::qword};
Label self;
build.mov(tmp.reg, qword[rState + offsetof(lua_State, global)]);
build.cmp(qword[tmp.reg + offsetof(global_State, cb.interrupt)], 0);
build.jcc(ConditionX64::NotEqual, self);
Label next = build.setLabel();
interruptHandlers.push_back({self, pcpos, next});
break;
}
case IrCmd::CHECK_GC:
callStepGc(regs, build);
break;
case IrCmd::BARRIER_OBJ:
callBarrierObject(regs, build, regOp(inst.a), inst.a, inst.b, inst.c.kind == IrOpKind::Undef ? -1 : tagOp(inst.c));
break;
case IrCmd::BARRIER_TABLE_BACK:
callBarrierTableFast(regs, build, regOp(inst.a), inst.a);
break;
case IrCmd::BARRIER_TABLE_FORWARD:
{
Label skip;
ScopedRegX64 tmp{regs, SizeX64::qword};
checkObjectBarrierConditions(build, tmp.reg, regOp(inst.a), inst.b, inst.c.kind == IrOpKind::Undef ? -1 : tagOp(inst.c), skip);
{
ScopedSpills spillGuard(regs);
IrCallWrapperX64 callWrap(regs, build, index);
callWrap.addArgument(SizeX64::qword, rState);
callWrap.addArgument(SizeX64::qword, regOp(inst.a), inst.a);
callWrap.addArgument(SizeX64::qword, tmp);
callWrap.call(qword[rNativeContext + offsetof(NativeContext, luaC_barriertable)]);
}
build.setLabel(skip);
break;
}
case IrCmd::SET_SAVEDPC:
{
ScopedRegX64 tmp1{regs, SizeX64::qword};
ScopedRegX64 tmp2{regs, SizeX64::qword};
build.mov(tmp2.reg, sCode);
build.add(tmp2.reg, uintOp(inst.a) * sizeof(Instruction));
build.mov(tmp1.reg, qword[rState + offsetof(lua_State, ci)]);
build.mov(qword[tmp1.reg + offsetof(CallInfo, savedpc)], tmp2.reg);
break;
}
case IrCmd::CLOSE_UPVALS:
{
Label next;
ScopedRegX64 tmp1{regs, SizeX64::qword};
ScopedRegX64 tmp2{regs, SizeX64::qword};
// L->openupval != 0
build.mov(tmp1.reg, qword[rState + offsetof(lua_State, openupval)]);
build.test(tmp1.reg, tmp1.reg);
build.jcc(ConditionX64::Zero, next);
// ra <= L->openuval->v
build.lea(tmp2.reg, addr[rBase + vmRegOp(inst.a) * sizeof(TValue)]);
build.cmp(tmp2.reg, qword[tmp1.reg + offsetof(UpVal, v)]);
build.jcc(ConditionX64::Above, next);
tmp1.free();
{
ScopedSpills spillGuard(regs);
IrCallWrapperX64 callWrap(regs, build, index);
callWrap.addArgument(SizeX64::qword, rState);
callWrap.addArgument(SizeX64::qword, tmp2);
callWrap.call(qword[rNativeContext + offsetof(NativeContext, luaF_close)]);
}
build.setLabel(next);
break;
}
case IrCmd::CAPTURE:
// No-op right now
break;
// Fallbacks to non-IR instruction implementations
case IrCmd::SETLIST:
regs.assertAllFree();
emitInstSetList(
regs, build, vmRegOp(inst.b), vmRegOp(inst.c), intOp(inst.d), uintOp(inst.e), inst.f.kind == IrOpKind::Undef ? -1 : int(uintOp(inst.f)));
break;
case IrCmd::CALL:
regs.assertAllFree();
regs.assertNoSpills();
emitInstCall(build, helpers, vmRegOp(inst.a), intOp(inst.b), intOp(inst.c));
break;
case IrCmd::RETURN:
regs.assertAllFree();
regs.assertNoSpills();
emitInstReturn(build, helpers, vmRegOp(inst.a), intOp(inst.b), function.variadic);
break;
case IrCmd::FORGLOOP:
regs.assertAllFree();
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
emitInstForGLoop(build, vmRegOp(inst.a), intOp(inst.b), labelOp(inst.c));
jumpOrFallthrough(blockOp(inst.d), next);
break;
case IrCmd::FORGLOOP_FALLBACK:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
{
IrCallWrapperX64 callWrap(regs, build, index);
callWrap.addArgument(SizeX64::qword, rState);
callWrap.addArgument(SizeX64::dword, vmRegOp(inst.a));
callWrap.addArgument(SizeX64::dword, intOp(inst.b));
callWrap.call(qword[rNativeContext + offsetof(NativeContext, forgLoopNonTableFallback)]);
emitUpdateBase(build);
build.test(al, al);
build.jcc(ConditionX64::NotZero, labelOp(inst.c));
jumpOrFallthrough(blockOp(inst.d), next);
break;
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
}
case IrCmd::FORGPREP_XNEXT_FALLBACK:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
{
IrCallWrapperX64 callWrap(regs, build, index);
callWrap.addArgument(SizeX64::qword, rState);
callWrap.addArgument(SizeX64::qword, luauRegAddress(vmRegOp(inst.b)));
callWrap.addArgument(SizeX64::dword, uintOp(inst.a) + 1);
callWrap.call(qword[rNativeContext + offsetof(NativeContext, forgPrepXnextFallback)]);
jumpOrFallthrough(blockOp(inst.c), next);
break;
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
}
case IrCmd::COVERAGE:
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
{
ScopedRegX64 tmp1{regs, SizeX64::qword};
ScopedRegX64 tmp2{regs, SizeX64::dword};
ScopedRegX64 tmp3{regs, SizeX64::dword};
build.mov(tmp1.reg, sCode);
build.add(tmp1.reg, uintOp(inst.a) * sizeof(Instruction));
// hits = LUAU_INSN_E(*pc)
build.mov(tmp2.reg, dword[tmp1.reg]);
build.sar(tmp2.reg, 8);
// hits = (hits < (1 << 23) - 1) ? hits + 1 : hits;
build.xor_(tmp3.reg, tmp3.reg);
build.cmp(tmp2.reg, (1 << 23) - 1);
build.setcc(ConditionX64::NotEqual, byteReg(tmp3.reg));
build.add(tmp2.reg, tmp3.reg);
// VM_PATCH_E(pc, hits);
build.sal(tmp2.reg, 8);
build.movzx(tmp3.reg, byte[tmp1.reg]);
build.or_(tmp3.reg, tmp2.reg);
build.mov(dword[tmp1.reg], tmp3.reg);
break;
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
}
// Full instruction fallbacks
case IrCmd::FALLBACK_GETGLOBAL:
LUAU_ASSERT(inst.b.kind == IrOpKind::VmReg);
LUAU_ASSERT(inst.c.kind == IrOpKind::VmConst);
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>
2023-05-19 20:37:30 +01:00
emitFallback(regs, build, offsetof(NativeContext, executeGETGLOBAL), uintOp(inst.a));
break;
case IrCmd::FALLBACK_SETGLOBAL:
LUAU_ASSERT(inst.b.kind == IrOpKind::VmReg);
LUAU_ASSERT(inst.c.kind == IrOpKind::VmConst);
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>
2023-05-19 20:37:30 +01:00
emitFallback(regs, build, offsetof(NativeContext, executeSETGLOBAL), uintOp(inst.a));
break;
case IrCmd::FALLBACK_GETTABLEKS:
LUAU_ASSERT(inst.b.kind == IrOpKind::VmReg);
LUAU_ASSERT(inst.c.kind == IrOpKind::VmReg);
LUAU_ASSERT(inst.d.kind == IrOpKind::VmConst);
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>
2023-05-19 20:37:30 +01:00
emitFallback(regs, build, offsetof(NativeContext, executeGETTABLEKS), uintOp(inst.a));
break;
case IrCmd::FALLBACK_SETTABLEKS:
LUAU_ASSERT(inst.b.kind == IrOpKind::VmReg);
LUAU_ASSERT(inst.c.kind == IrOpKind::VmReg);
LUAU_ASSERT(inst.d.kind == IrOpKind::VmConst);
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>
2023-05-19 20:37:30 +01:00
emitFallback(regs, build, offsetof(NativeContext, executeSETTABLEKS), uintOp(inst.a));
break;
case IrCmd::FALLBACK_NAMECALL:
LUAU_ASSERT(inst.b.kind == IrOpKind::VmReg);
LUAU_ASSERT(inst.c.kind == IrOpKind::VmReg);
LUAU_ASSERT(inst.d.kind == IrOpKind::VmConst);
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>
2023-05-19 20:37:30 +01:00
emitFallback(regs, build, offsetof(NativeContext, executeNAMECALL), uintOp(inst.a));
break;
case IrCmd::FALLBACK_PREPVARARGS:
LUAU_ASSERT(inst.b.kind == IrOpKind::Constant);
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>
2023-05-19 20:37:30 +01:00
emitFallback(regs, build, offsetof(NativeContext, executePREPVARARGS), uintOp(inst.a));
break;
case IrCmd::FALLBACK_GETVARARGS:
LUAU_ASSERT(inst.b.kind == IrOpKind::VmReg);
LUAU_ASSERT(inst.c.kind == IrOpKind::Constant);
if (intOp(inst.c) == LUA_MULTRET)
{
IrCallWrapperX64 callWrap(regs, build);
callWrap.addArgument(SizeX64::qword, rState);
RegisterX64 reg = callWrap.suggestNextArgumentRegister(SizeX64::qword);
build.mov(reg, sCode);
callWrap.addArgument(SizeX64::qword, addr[reg + uintOp(inst.a) * sizeof(Instruction)]);
callWrap.addArgument(SizeX64::qword, rBase);
callWrap.addArgument(SizeX64::dword, vmRegOp(inst.b));
callWrap.call(qword[rNativeContext + offsetof(NativeContext, executeGETVARARGSMultRet)]);
emitUpdateBase(build);
}
else
{
IrCallWrapperX64 callWrap(regs, build);
callWrap.addArgument(SizeX64::qword, rState);
callWrap.addArgument(SizeX64::qword, rBase);
callWrap.addArgument(SizeX64::dword, vmRegOp(inst.b));
callWrap.addArgument(SizeX64::dword, intOp(inst.c));
callWrap.call(qword[rNativeContext + offsetof(NativeContext, executeGETVARARGSConst)]);
}
break;
Sync to upstream/release/588 (#992) Type checker/autocomplete: * `Luau::autocomplete` no longer performs typechecking internally, make sure to run `Frontend::check` before performing autocomplete requests * Autocomplete string suggestions without "" are now only suggested inside the "" * Autocomplete suggestions now include `function (anonymous autofilled)` key with a full suggestion for the function expression (with arguments included) stored in `AutocompleteEntry::insertText` * `AutocompleteEntry::indexedWithSelf` is provided for function call suggestions made with `:` * Cyclic modules now see each other type exports as `any` to prevent memory use-after-free (similar to module return type) Runtime: * Updated inline/loop unroll cost model to better handle assignments (Fixes https://github.com/Roblox/luau/issues/978) * `math.noise` speed was improved by ~30% * `table.concat` speed was improved by ~5-7% * `tonumber` and `tostring` now have fastcall paths that execute ~1.5x and ~2.5x faster respectively (fixes #777) * Fixed crash in `luaL_typename` when index refers to a non-existing value * Fixed potential out of memory scenario when using `string.sub` or `string.char` in a loop * Fixed behavior of some fastcall builtins when called without arguments under -O2 to match original functions * Support for native code execution in VM is now enabled by default (note: native code still has to be generated explicitly) * `Codegen::compile` now accepts `CodeGen_OnlyNativeModules` flag. When set, only modules that have a `--!native` hot-comment at the top will be compiled to native code In our new typechecker: * Generic type packs are no longer considered to be variadic during unification * Timeout and cancellation now works in new solver * Fixed false positive errors around 'table' and 'function' type refinements * Table literals now use covariant unification rules. This is sound since literal has no type specified and has no aliases * Fixed issues with blocked types escaping the constraint solver * Fixed more places where error messages that should've been suppressed were still reported * Fixed errors when iterating over a top table type In our native code generation (jit): * 'DebugLuauAbortingChecks' flag is now supported on A64 * LOP_NEWCLOSURE has been translated to IR
2023-07-28 16:13:53 +01:00
case IrCmd::NEWCLOSURE:
{
ScopedRegX64 tmp2{regs, SizeX64::qword};
build.mov(tmp2.reg, sClosure);
build.mov(tmp2.reg, qword[tmp2.reg + offsetof(Closure, l.p)]);
build.mov(tmp2.reg, qword[tmp2.reg + offsetof(Proto, p)]);
build.mov(tmp2.reg, qword[tmp2.reg + sizeof(Proto*) * uintOp(inst.c)]);
IrCallWrapperX64 callWrap(regs, build, index);
callWrap.addArgument(SizeX64::qword, rState);
callWrap.addArgument(SizeX64::dword, uintOp(inst.a), inst.a);
callWrap.addArgument(SizeX64::qword, regOp(inst.b), inst.b);
callWrap.addArgument(SizeX64::qword, tmp2);
Sync to upstream/release/588 (#992) Type checker/autocomplete: * `Luau::autocomplete` no longer performs typechecking internally, make sure to run `Frontend::check` before performing autocomplete requests * Autocomplete string suggestions without "" are now only suggested inside the "" * Autocomplete suggestions now include `function (anonymous autofilled)` key with a full suggestion for the function expression (with arguments included) stored in `AutocompleteEntry::insertText` * `AutocompleteEntry::indexedWithSelf` is provided for function call suggestions made with `:` * Cyclic modules now see each other type exports as `any` to prevent memory use-after-free (similar to module return type) Runtime: * Updated inline/loop unroll cost model to better handle assignments (Fixes https://github.com/Roblox/luau/issues/978) * `math.noise` speed was improved by ~30% * `table.concat` speed was improved by ~5-7% * `tonumber` and `tostring` now have fastcall paths that execute ~1.5x and ~2.5x faster respectively (fixes #777) * Fixed crash in `luaL_typename` when index refers to a non-existing value * Fixed potential out of memory scenario when using `string.sub` or `string.char` in a loop * Fixed behavior of some fastcall builtins when called without arguments under -O2 to match original functions * Support for native code execution in VM is now enabled by default (note: native code still has to be generated explicitly) * `Codegen::compile` now accepts `CodeGen_OnlyNativeModules` flag. When set, only modules that have a `--!native` hot-comment at the top will be compiled to native code In our new typechecker: * Generic type packs are no longer considered to be variadic during unification * Timeout and cancellation now works in new solver * Fixed false positive errors around 'table' and 'function' type refinements * Table literals now use covariant unification rules. This is sound since literal has no type specified and has no aliases * Fixed issues with blocked types escaping the constraint solver * Fixed more places where error messages that should've been suppressed were still reported * Fixed errors when iterating over a top table type In our native code generation (jit): * 'DebugLuauAbortingChecks' flag is now supported on A64 * LOP_NEWCLOSURE has been translated to IR
2023-07-28 16:13:53 +01:00
callWrap.call(qword[rNativeContext + offsetof(NativeContext, luaF_newLclosure)]);
inst.regX64 = regs.takeReg(rax, index);
break;
Sync to upstream/release/588 (#992) Type checker/autocomplete: * `Luau::autocomplete` no longer performs typechecking internally, make sure to run `Frontend::check` before performing autocomplete requests * Autocomplete string suggestions without "" are now only suggested inside the "" * Autocomplete suggestions now include `function (anonymous autofilled)` key with a full suggestion for the function expression (with arguments included) stored in `AutocompleteEntry::insertText` * `AutocompleteEntry::indexedWithSelf` is provided for function call suggestions made with `:` * Cyclic modules now see each other type exports as `any` to prevent memory use-after-free (similar to module return type) Runtime: * Updated inline/loop unroll cost model to better handle assignments (Fixes https://github.com/Roblox/luau/issues/978) * `math.noise` speed was improved by ~30% * `table.concat` speed was improved by ~5-7% * `tonumber` and `tostring` now have fastcall paths that execute ~1.5x and ~2.5x faster respectively (fixes #777) * Fixed crash in `luaL_typename` when index refers to a non-existing value * Fixed potential out of memory scenario when using `string.sub` or `string.char` in a loop * Fixed behavior of some fastcall builtins when called without arguments under -O2 to match original functions * Support for native code execution in VM is now enabled by default (note: native code still has to be generated explicitly) * `Codegen::compile` now accepts `CodeGen_OnlyNativeModules` flag. When set, only modules that have a `--!native` hot-comment at the top will be compiled to native code In our new typechecker: * Generic type packs are no longer considered to be variadic during unification * Timeout and cancellation now works in new solver * Fixed false positive errors around 'table' and 'function' type refinements * Table literals now use covariant unification rules. This is sound since literal has no type specified and has no aliases * Fixed issues with blocked types escaping the constraint solver * Fixed more places where error messages that should've been suppressed were still reported * Fixed errors when iterating over a top table type In our native code generation (jit): * 'DebugLuauAbortingChecks' flag is now supported on A64 * LOP_NEWCLOSURE has been translated to IR
2023-07-28 16:13:53 +01:00
}
case IrCmd::FALLBACK_DUPCLOSURE:
LUAU_ASSERT(inst.b.kind == IrOpKind::VmReg);
LUAU_ASSERT(inst.c.kind == IrOpKind::VmConst);
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>
2023-05-19 20:37:30 +01:00
emitFallback(regs, build, offsetof(NativeContext, executeDUPCLOSURE), uintOp(inst.a));
break;
case IrCmd::FALLBACK_FORGPREP:
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>
2023-05-19 20:37:30 +01:00
emitFallback(regs, build, offsetof(NativeContext, executeFORGPREP), uintOp(inst.a));
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
jumpOrFallthrough(blockOp(inst.c), next);
break;
case IrCmd::BITAND_UINT:
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a});
if (inst.a.kind != IrOpKind::Inst || inst.regX64 != regOp(inst.a))
build.mov(inst.regX64, memRegUintOp(inst.a));
build.and_(inst.regX64, memRegUintOp(inst.b));
break;
case IrCmd::BITXOR_UINT:
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a});
if (inst.a.kind != IrOpKind::Inst || inst.regX64 != regOp(inst.a))
build.mov(inst.regX64, memRegUintOp(inst.a));
build.xor_(inst.regX64, memRegUintOp(inst.b));
break;
case IrCmd::BITOR_UINT:
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a});
if (inst.a.kind != IrOpKind::Inst || inst.regX64 != regOp(inst.a))
build.mov(inst.regX64, memRegUintOp(inst.a));
build.or_(inst.regX64, memRegUintOp(inst.b));
break;
case IrCmd::BITNOT_UINT:
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a});
if (inst.a.kind != IrOpKind::Inst || inst.regX64 != regOp(inst.a))
build.mov(inst.regX64, memRegUintOp(inst.a));
build.not_(inst.regX64);
break;
case IrCmd::BITLSHIFT_UINT:
{
ScopedRegX64 shiftTmp{regs};
// Custom bit shift value can only be placed in cl
// but we use it if the shift value is not a constant stored in b
if (inst.b.kind != IrOpKind::Constant)
shiftTmp.take(ecx);
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a});
if (inst.a.kind != IrOpKind::Inst || inst.regX64 != regOp(inst.a))
build.mov(inst.regX64, memRegUintOp(inst.a));
if (inst.b.kind == IrOpKind::Constant)
{
// if shift value is a constant, we extract the byte-sized shift amount
int8_t shift = int8_t(unsigned(intOp(inst.b)));
build.shl(inst.regX64, shift);
}
else
{
build.mov(shiftTmp.reg, memRegUintOp(inst.b));
build.shl(inst.regX64, byteReg(shiftTmp.reg));
}
break;
}
case IrCmd::BITRSHIFT_UINT:
{
ScopedRegX64 shiftTmp{regs};
// Custom bit shift value can only be placed in cl
// but we use it if the shift value is not a constant stored in b
if (inst.b.kind != IrOpKind::Constant)
shiftTmp.take(ecx);
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a});
if (inst.a.kind != IrOpKind::Inst || inst.regX64 != regOp(inst.a))
build.mov(inst.regX64, memRegUintOp(inst.a));
if (inst.b.kind == IrOpKind::Constant)
{
// if shift value is a constant, we extract the byte-sized shift amount
int8_t shift = int8_t(unsigned(intOp(inst.b)));
build.shr(inst.regX64, shift);
}
else
{
build.mov(shiftTmp.reg, memRegUintOp(inst.b));
build.shr(inst.regX64, byteReg(shiftTmp.reg));
}
break;
}
case IrCmd::BITARSHIFT_UINT:
{
ScopedRegX64 shiftTmp{regs};
// Custom bit shift value can only be placed in cl
// but we use it if the shift value is not a constant stored in b
if (inst.b.kind != IrOpKind::Constant)
shiftTmp.take(ecx);
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a});
if (inst.a.kind != IrOpKind::Inst || inst.regX64 != regOp(inst.a))
build.mov(inst.regX64, memRegUintOp(inst.a));
if (inst.b.kind == IrOpKind::Constant)
{
// if shift value is a constant, we extract the byte-sized shift amount
int8_t shift = int8_t(unsigned(intOp(inst.b)));
build.sar(inst.regX64, shift);
}
else
{
build.mov(shiftTmp.reg, memRegUintOp(inst.b));
build.sar(inst.regX64, byteReg(shiftTmp.reg));
}
break;
}
case IrCmd::BITLROTATE_UINT:
{
ScopedRegX64 shiftTmp{regs};
// Custom bit shift value can only be placed in cl
// but we use it if the shift value is not a constant stored in b
if (inst.b.kind != IrOpKind::Constant)
shiftTmp.take(ecx);
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a});
if (inst.a.kind != IrOpKind::Inst || inst.regX64 != regOp(inst.a))
build.mov(inst.regX64, memRegUintOp(inst.a));
if (inst.b.kind == IrOpKind::Constant)
{
// if shift value is a constant, we extract the byte-sized shift amount
int8_t shift = int8_t(unsigned(intOp(inst.b)));
build.rol(inst.regX64, shift);
}
else
{
build.mov(shiftTmp.reg, memRegUintOp(inst.b));
build.rol(inst.regX64, byteReg(shiftTmp.reg));
}
break;
}
case IrCmd::BITRROTATE_UINT:
{
ScopedRegX64 shiftTmp{regs};
// Custom bit shift value can only be placed in cl
// but we use it if the shift value is not a constant stored in b
if (inst.b.kind != IrOpKind::Constant)
shiftTmp.take(ecx);
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a});
if (inst.a.kind != IrOpKind::Inst || inst.regX64 != regOp(inst.a))
build.mov(inst.regX64, memRegUintOp(inst.a));
if (inst.b.kind == IrOpKind::Constant)
{
// if shift value is a constant, we extract the byte-sized shift amount
int8_t shift = int8_t(unsigned(intOp(inst.b)));
build.ror(inst.regX64, shift);
}
else
{
build.mov(shiftTmp.reg, memRegUintOp(inst.b));
build.ror(inst.regX64, byteReg(shiftTmp.reg));
}
break;
}
case IrCmd::BITCOUNTLZ_UINT:
{
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a});
Label zero, exit;
build.test(regOp(inst.a), regOp(inst.a));
build.jcc(ConditionX64::Equal, zero);
build.bsr(inst.regX64, regOp(inst.a));
build.xor_(inst.regX64, 0x1f);
build.jmp(exit);
build.setLabel(zero);
build.mov(inst.regX64, 32);
build.setLabel(exit);
break;
}
case IrCmd::BITCOUNTRZ_UINT:
{
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a});
Label zero, exit;
build.test(regOp(inst.a), regOp(inst.a));
build.jcc(ConditionX64::Equal, zero);
build.bsf(inst.regX64, regOp(inst.a));
build.jmp(exit);
build.setLabel(zero);
build.mov(inst.regX64, 32);
build.setLabel(exit);
break;
}
case IrCmd::BYTESWAP_UINT:
{
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a});
if (inst.a.kind != IrOpKind::Inst || inst.regX64 != regOp(inst.a))
build.mov(inst.regX64, memRegUintOp(inst.a));
build.bswap(inst.regX64);
break;
}
case IrCmd::INVOKE_LIBM:
{
IrCallWrapperX64 callWrap(regs, build, index);
callWrap.addArgument(SizeX64::xmmword, memRegDoubleOp(inst.b), inst.b);
if (inst.c.kind != IrOpKind::None)
{
bool isInt = (inst.c.kind == IrOpKind::Constant) ? constOp(inst.c).kind == IrConstKind::Int
: getCmdValueKind(function.instOp(inst.c).cmd) == IrValueKind::Int;
if (isInt)
callWrap.addArgument(SizeX64::dword, memRegUintOp(inst.c), inst.c);
else
callWrap.addArgument(SizeX64::xmmword, memRegDoubleOp(inst.c), inst.c);
}
callWrap.call(qword[rNativeContext + getNativeContextOffset(uintOp(inst.a))]);
inst.regX64 = regs.takeReg(xmm0, index);
break;
}
case IrCmd::GET_TYPE:
{
inst.regX64 = regs.allocReg(SizeX64::qword, index);
build.mov(inst.regX64, qword[rState + offsetof(lua_State, global)]);
if (inst.a.kind == IrOpKind::Inst)
build.mov(inst.regX64, qword[inst.regX64 + qwordReg(regOp(inst.a)) * sizeof(TString*) + offsetof(global_State, ttname)]);
else if (inst.a.kind == IrOpKind::Constant)
build.mov(inst.regX64, qword[inst.regX64 + tagOp(inst.a) * sizeof(TString*) + offsetof(global_State, ttname)]);
else
LUAU_ASSERT(!"Unsupported instruction form");
break;
}
case IrCmd::GET_TYPEOF:
{
IrCallWrapperX64 callWrap(regs, build);
callWrap.addArgument(SizeX64::qword, rState);
callWrap.addArgument(SizeX64::qword, luauRegAddress(vmRegOp(inst.a)));
callWrap.call(qword[rNativeContext + offsetof(NativeContext, luaT_objtypenamestr)]);
inst.regX64 = regs.takeReg(rax, index);
break;
}
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
Sync to upstream/release/588 (#992) Type checker/autocomplete: * `Luau::autocomplete` no longer performs typechecking internally, make sure to run `Frontend::check` before performing autocomplete requests * Autocomplete string suggestions without "" are now only suggested inside the "" * Autocomplete suggestions now include `function (anonymous autofilled)` key with a full suggestion for the function expression (with arguments included) stored in `AutocompleteEntry::insertText` * `AutocompleteEntry::indexedWithSelf` is provided for function call suggestions made with `:` * Cyclic modules now see each other type exports as `any` to prevent memory use-after-free (similar to module return type) Runtime: * Updated inline/loop unroll cost model to better handle assignments (Fixes https://github.com/Roblox/luau/issues/978) * `math.noise` speed was improved by ~30% * `table.concat` speed was improved by ~5-7% * `tonumber` and `tostring` now have fastcall paths that execute ~1.5x and ~2.5x faster respectively (fixes #777) * Fixed crash in `luaL_typename` when index refers to a non-existing value * Fixed potential out of memory scenario when using `string.sub` or `string.char` in a loop * Fixed behavior of some fastcall builtins when called without arguments under -O2 to match original functions * Support for native code execution in VM is now enabled by default (note: native code still has to be generated explicitly) * `Codegen::compile` now accepts `CodeGen_OnlyNativeModules` flag. When set, only modules that have a `--!native` hot-comment at the top will be compiled to native code In our new typechecker: * Generic type packs are no longer considered to be variadic during unification * Timeout and cancellation now works in new solver * Fixed false positive errors around 'table' and 'function' type refinements * Table literals now use covariant unification rules. This is sound since literal has no type specified and has no aliases * Fixed issues with blocked types escaping the constraint solver * Fixed more places where error messages that should've been suppressed were still reported * Fixed errors when iterating over a top table type In our native code generation (jit): * 'DebugLuauAbortingChecks' flag is now supported on A64 * LOP_NEWCLOSURE has been translated to IR
2023-07-28 16:13:53 +01:00
case IrCmd::FINDUPVAL:
{
IrCallWrapperX64 callWrap(regs, build);
callWrap.addArgument(SizeX64::qword, rState);
callWrap.addArgument(SizeX64::qword, luauRegAddress(vmRegOp(inst.a)));
callWrap.call(qword[rNativeContext + offsetof(NativeContext, luaF_findupval)]);
inst.regX64 = regs.takeReg(rax, index);
break;
}
Sync to upstream/release/603 (#1097) # What's changed? - Record the location of properties for table types (closes #802) - Implement stricter UTF-8 validations as per the RFC (https://github.com/luau-lang/rfcs/pull/1) - Implement `buffer` as a new type in both the old and new solvers. - Changed errors produced by some `buffer` builtins to be a bit more generic to avoid platform-dependent error messages. - Fixed a bug where `Unifier` would copy some persistent types, tripping some internal assertions. - Type checking rules on relational operators is now a little bit more lax. - Improve dead code elimination for some `if` statements with complex always-false conditions ## New type solver - Dataflow analysis now generates phi nodes on exit of branches. - Dataflow analysis avoids producing a new definition for locals or properties that are not owned by that loop. - If a function parameter has been constrained to `never`, report errors at all uses of that parameter within that function. - Switch to using the new `Luau::Set` to replace `std::unordered_set` to alleviate some poor allocation characteristics which was negatively affecting overall performance. - Subtyping can now report many failing reasons instead of just the first one that we happened to find during the test. - Subtyping now also report reasons for type pack mismatches. - When visiting `if` statements or expressions, the resulting context are the common terms in both branches. ## Native codegen - Implement support for `buffer` builtins to its IR for x64 and A64. - Optimized `table.insert` by not inserting a table barrier if it is fastcalled with a constant. ## Internal Contributors Co-authored-by: Aaron Weiss <aaronweiss@roblox.com> Co-authored-by: Alexander McCord <amccord@roblox.com> Co-authored-by: Andy Friesen <afriesen@roblox.com> Co-authored-by: Arseny Kapoulkine <arseny@roblox.com> Co-authored-by: Aviral Goel <agoel@roblox.com> Co-authored-by: Lily Brown <lbrown@roblox.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-11-10 21:10:07 +00:00
case IrCmd::BUFFER_READI8:
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a, inst.b});
build.movsx(inst.regX64, byte[bufferAddrOp(inst.a, inst.b)]);
break;
case IrCmd::BUFFER_READU8:
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a, inst.b});
build.movzx(inst.regX64, byte[bufferAddrOp(inst.a, inst.b)]);
break;
case IrCmd::BUFFER_WRITEI8:
{
OperandX64 value = inst.c.kind == IrOpKind::Inst ? byteReg(regOp(inst.c)) : OperandX64(int8_t(intOp(inst.c)));
build.mov(byte[bufferAddrOp(inst.a, inst.b)], value);
Sync to upstream/release/603 (#1097) # What's changed? - Record the location of properties for table types (closes #802) - Implement stricter UTF-8 validations as per the RFC (https://github.com/luau-lang/rfcs/pull/1) - Implement `buffer` as a new type in both the old and new solvers. - Changed errors produced by some `buffer` builtins to be a bit more generic to avoid platform-dependent error messages. - Fixed a bug where `Unifier` would copy some persistent types, tripping some internal assertions. - Type checking rules on relational operators is now a little bit more lax. - Improve dead code elimination for some `if` statements with complex always-false conditions ## New type solver - Dataflow analysis now generates phi nodes on exit of branches. - Dataflow analysis avoids producing a new definition for locals or properties that are not owned by that loop. - If a function parameter has been constrained to `never`, report errors at all uses of that parameter within that function. - Switch to using the new `Luau::Set` to replace `std::unordered_set` to alleviate some poor allocation characteristics which was negatively affecting overall performance. - Subtyping can now report many failing reasons instead of just the first one that we happened to find during the test. - Subtyping now also report reasons for type pack mismatches. - When visiting `if` statements or expressions, the resulting context are the common terms in both branches. ## Native codegen - Implement support for `buffer` builtins to its IR for x64 and A64. - Optimized `table.insert` by not inserting a table barrier if it is fastcalled with a constant. ## Internal Contributors Co-authored-by: Aaron Weiss <aaronweiss@roblox.com> Co-authored-by: Alexander McCord <amccord@roblox.com> Co-authored-by: Andy Friesen <afriesen@roblox.com> Co-authored-by: Arseny Kapoulkine <arseny@roblox.com> Co-authored-by: Aviral Goel <agoel@roblox.com> Co-authored-by: Lily Brown <lbrown@roblox.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-11-10 21:10:07 +00:00
break;
}
case IrCmd::BUFFER_READI16:
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a, inst.b});
build.movsx(inst.regX64, word[bufferAddrOp(inst.a, inst.b)]);
break;
case IrCmd::BUFFER_READU16:
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a, inst.b});
build.movzx(inst.regX64, word[bufferAddrOp(inst.a, inst.b)]);
break;
case IrCmd::BUFFER_WRITEI16:
{
OperandX64 value = inst.c.kind == IrOpKind::Inst ? wordReg(regOp(inst.c)) : OperandX64(int16_t(intOp(inst.c)));
Sync to upstream/release/603 (#1097) # What's changed? - Record the location of properties for table types (closes #802) - Implement stricter UTF-8 validations as per the RFC (https://github.com/luau-lang/rfcs/pull/1) - Implement `buffer` as a new type in both the old and new solvers. - Changed errors produced by some `buffer` builtins to be a bit more generic to avoid platform-dependent error messages. - Fixed a bug where `Unifier` would copy some persistent types, tripping some internal assertions. - Type checking rules on relational operators is now a little bit more lax. - Improve dead code elimination for some `if` statements with complex always-false conditions ## New type solver - Dataflow analysis now generates phi nodes on exit of branches. - Dataflow analysis avoids producing a new definition for locals or properties that are not owned by that loop. - If a function parameter has been constrained to `never`, report errors at all uses of that parameter within that function. - Switch to using the new `Luau::Set` to replace `std::unordered_set` to alleviate some poor allocation characteristics which was negatively affecting overall performance. - Subtyping can now report many failing reasons instead of just the first one that we happened to find during the test. - Subtyping now also report reasons for type pack mismatches. - When visiting `if` statements or expressions, the resulting context are the common terms in both branches. ## Native codegen - Implement support for `buffer` builtins to its IR for x64 and A64. - Optimized `table.insert` by not inserting a table barrier if it is fastcalled with a constant. ## Internal Contributors Co-authored-by: Aaron Weiss <aaronweiss@roblox.com> Co-authored-by: Alexander McCord <amccord@roblox.com> Co-authored-by: Andy Friesen <afriesen@roblox.com> Co-authored-by: Arseny Kapoulkine <arseny@roblox.com> Co-authored-by: Aviral Goel <agoel@roblox.com> Co-authored-by: Lily Brown <lbrown@roblox.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-11-10 21:10:07 +00:00
build.mov(word[bufferAddrOp(inst.a, inst.b)], value);
Sync to upstream/release/603 (#1097) # What's changed? - Record the location of properties for table types (closes #802) - Implement stricter UTF-8 validations as per the RFC (https://github.com/luau-lang/rfcs/pull/1) - Implement `buffer` as a new type in both the old and new solvers. - Changed errors produced by some `buffer` builtins to be a bit more generic to avoid platform-dependent error messages. - Fixed a bug where `Unifier` would copy some persistent types, tripping some internal assertions. - Type checking rules on relational operators is now a little bit more lax. - Improve dead code elimination for some `if` statements with complex always-false conditions ## New type solver - Dataflow analysis now generates phi nodes on exit of branches. - Dataflow analysis avoids producing a new definition for locals or properties that are not owned by that loop. - If a function parameter has been constrained to `never`, report errors at all uses of that parameter within that function. - Switch to using the new `Luau::Set` to replace `std::unordered_set` to alleviate some poor allocation characteristics which was negatively affecting overall performance. - Subtyping can now report many failing reasons instead of just the first one that we happened to find during the test. - Subtyping now also report reasons for type pack mismatches. - When visiting `if` statements or expressions, the resulting context are the common terms in both branches. ## Native codegen - Implement support for `buffer` builtins to its IR for x64 and A64. - Optimized `table.insert` by not inserting a table barrier if it is fastcalled with a constant. ## Internal Contributors Co-authored-by: Aaron Weiss <aaronweiss@roblox.com> Co-authored-by: Alexander McCord <amccord@roblox.com> Co-authored-by: Andy Friesen <afriesen@roblox.com> Co-authored-by: Arseny Kapoulkine <arseny@roblox.com> Co-authored-by: Aviral Goel <agoel@roblox.com> Co-authored-by: Lily Brown <lbrown@roblox.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-11-10 21:10:07 +00:00
break;
}
case IrCmd::BUFFER_READI32:
inst.regX64 = regs.allocRegOrReuse(SizeX64::dword, index, {inst.a, inst.b});
build.mov(inst.regX64, dword[bufferAddrOp(inst.a, inst.b)]);
break;
case IrCmd::BUFFER_WRITEI32:
{
OperandX64 value = inst.c.kind == IrOpKind::Inst ? regOp(inst.c) : OperandX64(intOp(inst.c));
build.mov(dword[bufferAddrOp(inst.a, inst.b)], value);
break;
}
case IrCmd::BUFFER_READF32:
inst.regX64 = regs.allocReg(SizeX64::xmmword, index);
build.vcvtss2sd(inst.regX64, inst.regX64, dword[bufferAddrOp(inst.a, inst.b)]);
break;
case IrCmd::BUFFER_WRITEF32:
storeDoubleAsFloat(dword[bufferAddrOp(inst.a, inst.b)], inst.c);
break;
case IrCmd::BUFFER_READF64:
inst.regX64 = regs.allocReg(SizeX64::xmmword, index);
build.vmovsd(inst.regX64, qword[bufferAddrOp(inst.a, inst.b)]);
break;
case IrCmd::BUFFER_WRITEF64:
if (inst.c.kind == IrOpKind::Constant)
{
ScopedRegX64 tmp{regs, SizeX64::xmmword};
build.vmovsd(tmp.reg, build.f64(doubleOp(inst.c)));
build.vmovsd(qword[bufferAddrOp(inst.a, inst.b)], tmp.reg);
}
else if (inst.c.kind == IrOpKind::Inst)
{
build.vmovsd(qword[bufferAddrOp(inst.a, inst.b)], regOp(inst.c));
}
else
{
LUAU_ASSERT(!"Unsupported instruction form");
}
break;
Sync to upstream/release/572 (#899) * Fixed exported types not being suggested in autocomplete * `T...` is now convertible to `...any` (Fixes https://github.com/Roblox/luau/issues/767) * Fixed issue with `T?` not being convertible to `T | T` or `T?` (sometimes when internal pointer identity is different) * Fixed potential crash in missing table key error suggestion to use a similar existing key * `lua_topointer` now returns a pointer for strings C++ API Changes: * `prepareModuleScope` callback has moved from TypeChecker to Frontend * For LSPs, AstQuery functions (and `isWithinComment`) can be used without full Frontend data A lot of changes in our two experimental components as well. In our work on the new type-solver, the following issues were fixed: * Fixed table union and intersection indexing * Correct custom type environments are now used * Fixed issue with values of `free & number` type not accepted in numeric operations And these are the changes in native code generation (JIT): * arm64 lowering is almost complete with support for 99% of IR commands and all fastcalls * Fixed x64 assembly encoding for extended byte registers * More external x64 calls are aware of register allocator * `math.min`/`math.max` with more than 2 arguments are now lowered to IR as well * Fixed correctness issues with `math` library calls with multiple results in variadic context and with x64 register conflicts * x64 register allocator learnt to restore values from VM memory instead of always using stack spills * x64 exception unwind information now supports multiple functions and fixes function start offset in Dwarf2 info
2023-04-14 19:06:22 +01:00
// Pseudo instructions
case IrCmd::NOP:
case IrCmd::SUBSTITUTE:
LUAU_ASSERT(!"Pseudo instructions should not be lowered");
break;
}
valueTracker.afterInstLowering(inst, index);
regs.freeLastUseRegs(inst, index);
}
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>
2023-08-18 19:15:41 +01:00
void IrLoweringX64::finishBlock(const IrBlock& curr, const IrBlock& next)
{
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>
2023-08-18 19:15:41 +01:00
if (!regs.spills.empty())
{
// If we have spills remaining, we have to immediately lower the successor block
for (uint32_t predIdx : predecessors(function.cfg, function.getBlockIndex(next)))
Sync to upstream/release/605 (#1118) - Implemented [Require by String with Relative Paths](https://github.com/luau-lang/rfcs/blob/master/docs/new-require-by-string-semantics.md) RFC - Implemented [Require by String with Aliases](https://github.com/luau-lang/rfcs/blob/master/docs/require-by-string-aliases.md) RFC with support for `paths` and `alias` arrays in .luarc - Added SUBRK and DIVRK bytecode instructions to speed up constant-number and constant/number operations - Added `--vector-lib`, `--vector-ctor` and `--vector-type` options to luau-compile to support code with vectors New Solver - Correctness fixes to subtyping - Improvements to dataflow analysis Native Code Generation - Added bytecode analysis pass to predict type tags used in operations - Fixed rare cases of numerical loops being generated without an interrupt instruction - Restored optimization data propagation into the linear block - Duplicate buffer length checks are optimized away Miscellaneous - Small performance improvements to new non-strict mode - Introduced more scripts for fuzzing Luau and processing the results, including fuzzer build support for CMake Co-authored-by: Alexander McCord <amccord@roblox.com> Co-authored-by: Andy Friesen <afriesen@roblox.com> Co-authored-by: Aviral Goel <agoel@roblox.com> Co-authored-by: David Cope <dcope@roblox.com> Co-authored-by: Lily Brown <lbrown@roblox.com> Co-authored-by: Vighnesh Vijay <vvijay@roblox.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com> --------- Co-authored-by: Aaron Weiss <aaronweiss@roblox.com> Co-authored-by: Alexander McCord <amccord@roblox.com> Co-authored-by: Andy Friesen <afriesen@roblox.com> Co-authored-by: Aviral Goel <agoel@roblox.com> Co-authored-by: David Cope <dcope@roblox.com> Co-authored-by: Lily Brown <lbrown@roblox.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-12-02 07:46:57 +00:00
LUAU_ASSERT(predIdx == function.getBlockIndex(curr) || function.blocks[predIdx].kind == IrBlockKind::Dead);
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>
2023-08-18 19:15:41 +01:00
// And the next block cannot be a join block in cfg
LUAU_ASSERT(next.useCount == 1);
}
}
void IrLoweringX64::finishFunction()
{
if (build.logText)
build.logAppend("; interrupt handlers\n");
for (InterruptHandler& handler : interruptHandlers)
{
build.setLabel(handler.self);
build.mov(eax, handler.pcpos + 1);
build.lea(rbx, handler.next);
build.jmp(helpers.interrupt);
}
if (build.logText)
build.logAppend("; exit handlers\n");
for (ExitHandler& handler : exitHandlers)
{
LUAU_ASSERT(handler.pcpos != kVmExitEntryGuardPc);
build.setLabel(handler.self);
build.mov(edx, handler.pcpos * sizeof(Instruction));
build.jmp(helpers.updatePcAndContinueInVm);
}
if (stats)
{
if (regs.maxUsedSlot > kSpillSlots)
stats->regAllocErrors++;
if (regs.maxUsedSlot > stats->maxSpillSlotsUsed)
stats->maxSpillSlotsUsed = regs.maxUsedSlot;
}
}
bool IrLoweringX64::hasError() const
{
// If register allocator had to use more stack slots than we have available, this function can't run natively
if (regs.maxUsedSlot > kSpillSlots)
return true;
return false;
}
bool IrLoweringX64::isFallthroughBlock(const IrBlock& target, const IrBlock& next)
{
return target.start == next.start;
}
Label& IrLoweringX64::getTargetLabel(IrOp op, Label& fresh)
{
if (op.kind == IrOpKind::Undef)
return fresh;
if (op.kind == IrOpKind::VmExit)
{
// Special exit case that doesn't have to update pcpos
if (vmExitOp(op) == kVmExitEntryGuardPc)
return helpers.exitContinueVmClearNativeFlag;
if (uint32_t* index = exitHandlerMap.find(vmExitOp(op)))
return exitHandlers[*index].self;
return fresh;
}
return labelOp(op);
}
void IrLoweringX64::finalizeTargetLabel(IrOp op, Label& fresh)
{
if (op.kind == IrOpKind::VmExit && fresh.id != 0 && fresh.id != helpers.exitContinueVmClearNativeFlag.id)
{
exitHandlerMap[vmExitOp(op)] = uint32_t(exitHandlers.size());
exitHandlers.push_back({fresh, vmExitOp(op)});
}
}
void IrLoweringX64::jumpOrFallthrough(IrBlock& target, const IrBlock& next)
{
if (!isFallthroughBlock(target, next))
build.jmp(target.label);
}
void IrLoweringX64::jumpOrAbortOnUndef(ConditionX64 cond, IrOp target, const IrBlock& next)
{
Label fresh;
Label& label = getTargetLabel(target, fresh);
if (target.kind == IrOpKind::Undef)
{
if (cond == ConditionX64::Count)
{
build.ud2(); // Unconditional jump to abort is just an abort
}
else
{
build.jcc(getReverseCondition(cond), label);
build.ud2();
build.setLabel(label);
}
}
else if (cond == ConditionX64::Count)
{
// Unconditional jump can be skipped if it's a fallthrough
if (target.kind == IrOpKind::VmExit || !isFallthroughBlock(blockOp(target), next))
build.jmp(label);
}
else
{
build.jcc(cond, label);
}
finalizeTargetLabel(target, fresh);
}
void IrLoweringX64::jumpOrAbortOnUndef(IrOp target, const IrBlock& next)
{
jumpOrAbortOnUndef(ConditionX64::Count, target, next);
}
OperandX64 IrLoweringX64::memRegDoubleOp(IrOp op)
{
switch (op.kind)
{
case IrOpKind::Inst:
return regOp(op);
case IrOpKind::Constant:
return build.f64(doubleOp(op));
case IrOpKind::VmReg:
return luauRegValue(vmRegOp(op));
case IrOpKind::VmConst:
return luauConstantValue(vmConstOp(op));
default:
LUAU_ASSERT(!"Unsupported operand kind");
}
return noreg;
}
OperandX64 IrLoweringX64::memRegUintOp(IrOp op)
{
switch (op.kind)
{
case IrOpKind::Inst:
return regOp(op);
case IrOpKind::Constant:
return OperandX64(unsigned(intOp(op)));
case IrOpKind::VmReg:
return luauRegValueInt(vmRegOp(op));
default:
LUAU_ASSERT(!"Unsupported operand kind");
}
return noreg;
}
OperandX64 IrLoweringX64::memRegTagOp(IrOp op)
{
switch (op.kind)
{
case IrOpKind::Inst:
return regOp(op);
case IrOpKind::VmReg:
return luauRegTag(vmRegOp(op));
case IrOpKind::VmConst:
return luauConstantTag(vmConstOp(op));
default:
LUAU_ASSERT(!"Unsupported operand kind");
}
return noreg;
}
RegisterX64 IrLoweringX64::regOp(IrOp op)
{
IrInst& inst = function.instOp(op);
if (inst.spilled || inst.needsReload)
regs.restore(inst, false);
LUAU_ASSERT(inst.regX64 != noreg);
return inst.regX64;
}
Sync to upstream/release/603 (#1097) # What's changed? - Record the location of properties for table types (closes #802) - Implement stricter UTF-8 validations as per the RFC (https://github.com/luau-lang/rfcs/pull/1) - Implement `buffer` as a new type in both the old and new solvers. - Changed errors produced by some `buffer` builtins to be a bit more generic to avoid platform-dependent error messages. - Fixed a bug where `Unifier` would copy some persistent types, tripping some internal assertions. - Type checking rules on relational operators is now a little bit more lax. - Improve dead code elimination for some `if` statements with complex always-false conditions ## New type solver - Dataflow analysis now generates phi nodes on exit of branches. - Dataflow analysis avoids producing a new definition for locals or properties that are not owned by that loop. - If a function parameter has been constrained to `never`, report errors at all uses of that parameter within that function. - Switch to using the new `Luau::Set` to replace `std::unordered_set` to alleviate some poor allocation characteristics which was negatively affecting overall performance. - Subtyping can now report many failing reasons instead of just the first one that we happened to find during the test. - Subtyping now also report reasons for type pack mismatches. - When visiting `if` statements or expressions, the resulting context are the common terms in both branches. ## Native codegen - Implement support for `buffer` builtins to its IR for x64 and A64. - Optimized `table.insert` by not inserting a table barrier if it is fastcalled with a constant. ## Internal Contributors Co-authored-by: Aaron Weiss <aaronweiss@roblox.com> Co-authored-by: Alexander McCord <amccord@roblox.com> Co-authored-by: Andy Friesen <afriesen@roblox.com> Co-authored-by: Arseny Kapoulkine <arseny@roblox.com> Co-authored-by: Aviral Goel <agoel@roblox.com> Co-authored-by: Lily Brown <lbrown@roblox.com> Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-11-10 21:10:07 +00:00
OperandX64 IrLoweringX64::bufferAddrOp(IrOp bufferOp, IrOp indexOp)
{
if (indexOp.kind == IrOpKind::Inst)
return regOp(bufferOp) + qwordReg(regOp(indexOp)) + offsetof(Buffer, data);
else if (indexOp.kind == IrOpKind::Constant)
return regOp(bufferOp) + intOp(indexOp) + offsetof(Buffer, data);
LUAU_ASSERT(!"Unsupported instruction form");
return noreg;
}
IrConst IrLoweringX64::constOp(IrOp op) const
{
return function.constOp(op);
}
uint8_t IrLoweringX64::tagOp(IrOp op) const
{
return function.tagOp(op);
}
int IrLoweringX64::intOp(IrOp op) const
{
return function.intOp(op);
}
unsigned IrLoweringX64::uintOp(IrOp op) const
{
return function.uintOp(op);
}
double IrLoweringX64::doubleOp(IrOp op) const
{
return function.doubleOp(op);
}
IrBlock& IrLoweringX64::blockOp(IrOp op) const
{
return function.blockOp(op);
}
Label& IrLoweringX64::labelOp(IrOp op) const
{
return blockOp(op).label;
}
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
2023-03-03 20:21:14 +00:00
} // namespace X64
} // namespace CodeGen
} // namespace Luau