This is an archive of the discontinued LLVM Phabricator instance.

Differential D92850

Add sqrt lowering from standard to NVVM
ClosedPublic

Authored by frgossen on Dec 8 2020, 7:58 AM.

Details

Reviewers
herhut
bkramer
pifon2a
Commits
rGb4750f58d89f: Add sqrt lowering from standard to NVVM

Diff Detail

Unit TestsFailed

TimeTest
90 msx64 windows > LLVM.CodeGen/XCore::threads.ll

Event Timeline

frgossen created this revision.Dec 8 2020, 7:58 AM
Herald added subscribers: teijeong, rdzhabarov, tatianashp and 16 others. · View Herald TranscriptDec 8 2020, 7:58 AM
frgossen requested review of this revision.Dec 8 2020, 7:58 AM
Herald added a reviewer: herhut. · View Herald TranscriptDec 8 2020, 7:58 AM
Herald added a project: Restricted Project. · View Herald Transcript
Herald added subscribers: stephenneuendorffer, nicolasvasilache. · View Herald Transcript
frgossen added reviewers: bkramer, pifon2a.Dec 8 2020, 7:59 AM
frgossen updated this revision to Diff 310220.Dec 8 2020, 8:02 AM

Add test case

pifon2a accepted this revision.Dec 8 2020, 8:06 AM
This revision is now accepted and ready to land.Dec 8 2020, 8:06 AM
This revision was landed with ongoing or failed builds.Dec 8 2020, 8:08 AM
Closed by commit rGb4750f58d89f: Add sqrt lowering from standard to NVVM (authored by frgossen). · Explain Why
This revision was automatically updated to reflect the committed changes.
frgossen added a commit: rGb4750f58d89f: Add sqrt lowering from standard to NVVM.
Harbormaster completed remote builds in B81466: Diff 310218.Dec 8 2020, 8:18 AM
Harbormaster completed remote builds in B81467: Diff 310220.Dec 8 2020, 8:24 AM

Revision Contents

PathSize
mlir/
lib/
Conversion/
GPUToNVVM/
4 lines

Diff 310218

mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp

Show First 20 Lines • Show All 132 Lines • ▼ Show 20 Lines void runOnOperation() override {
applyPatternsAndFoldGreedily(m, std::move(patterns)); applyPatternsAndFoldGreedily(m, std::move(patterns));
populateStdToLLVMConversionPatterns(converter, llvmPatterns); populateStdToLLVMConversionPatterns(converter, llvmPatterns);
populateGpuToNVVMConversionPatterns(converter, llvmPatterns); populateGpuToNVVMConversionPatterns(converter, llvmPatterns);
LLVMConversionTarget target(getContext()); LLVMConversionTarget target(getContext());
target.addIllegalDialect<gpu::GPUDialect>(); target.addIllegalDialect<gpu::GPUDialect>();
target.addIllegalOp<LLVM::CosOp, LLVM::ExpOp, LLVM::FAbsOp, LLVM::FCeilOp, target.addIllegalOp<LLVM::CosOp, LLVM::ExpOp, LLVM::FAbsOp, LLVM::FCeilOp,
LLVM::FFloorOp, LLVM::LogOp, LLVM::Log10Op, LLVM::FFloorOp, LLVM::LogOp, LLVM::Log10Op,
LLVM::Log2Op, LLVM::SinOp>(); LLVM::Log2Op, LLVM::SinOp, LLVM::SqrtOp>();
target.addIllegalOp<FuncOp>(); target.addIllegalOp<FuncOp>();
target.addLegalDialect<NVVM::NVVMDialect>(); target.addLegalDialect<NVVM::NVVMDialect>();
// TODO: Remove once we support replacing non-root ops. // TODO: Remove once we support replacing non-root ops.
target.addLegalOp<gpu::YieldOp, gpu::GPUModuleOp, gpu::ModuleEndOp>(); target.addLegalOp<gpu::YieldOp, gpu::GPUModuleOp, gpu::ModuleEndOp>();
if (failed(applyPartialConversion(m, target, std::move(llvmPatterns)))) if (failed(applyPartialConversion(m, target, std::move(llvmPatterns))))
signalPassFailure(); signalPassFailure();
} }
}; };
Show All 32 Linesvoid mlir::populateGpuToNVVMConversionPatterns(
patterns.insert<OpToFuncCallLowering<Log10Op>>(converter, "__nv_log10f", patterns.insert<OpToFuncCallLowering<Log10Op>>(converter, "__nv_log10f",
"__nv_log10"); "__nv_log10");
patterns.insert<OpToFuncCallLowering<Log2Op>>(converter, "__nv_log2f", patterns.insert<OpToFuncCallLowering<Log2Op>>(converter, "__nv_log2f",
"__nv_log2"); "__nv_log2");
patterns.insert<OpToFuncCallLowering<RsqrtOp>>(converter, "__nv_rsqrtf", patterns.insert<OpToFuncCallLowering<RsqrtOp>>(converter, "__nv_rsqrtf",
"__nv_rsqrt"); "__nv_rsqrt");
patterns.insert<OpToFuncCallLowering<SinOp>>(converter, "__nv_sinf", patterns.insert<OpToFuncCallLowering<SinOp>>(converter, "__nv_sinf",
"__nv_sin"); "__nv_sin");
patterns.insert<OpToFuncCallLowering<SqrtOp>>(converter, "__nv_sqrtf",
"__nv_sqrt");
patterns.insert<OpToFuncCallLowering<TanhOp>>(converter, "__nv_tanhf", patterns.insert<OpToFuncCallLowering<TanhOp>>(converter, "__nv_tanhf",
"__nv_tanh"); "__nv_tanh");
} }
std::unique_ptr<OperationPass<gpu::GPUModuleOp>> std::unique_ptr<OperationPass<gpu::GPUModuleOp>>
mlir::createLowerGpuOpsToNVVMOpsPass(unsigned indexBitwidth) { mlir::createLowerGpuOpsToNVVMOpsPass(unsigned indexBitwidth) {
return std::make_unique<LowerGpuOpsToNVVMOpsPass>(indexBitwidth); return std::make_unique<LowerGpuOpsToNVVMOpsPass>(indexBitwidth);
} }