Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix GPUXToSpirv Pass to iterate for all gpuModules #426

Merged
merged 1 commit into from
Oct 31, 2022
Merged

Fix GPUXToSpirv Pass to iterate for all gpuModules #426

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
64 changes: 34 additions & 30 deletions lib/Conversion/GPUToSPIRV/GPUToSPIRVPass.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,43 +64,47 @@ void GPUXToSPIRVPass::runOnOperation() {
gpuModules.push_back(builder.clone(*moduleOp.getOperation()));
});

// Map MemRef memory space to SPIR-V storage class first if requested.
if (mapMemorySpace) {
std::unique_ptr<mlir::ConversionTarget> target =
mlir::spirv::getMemorySpaceToStorageClassTarget(*context);
mlir::spirv::MemorySpaceToStorageClassMap memorySpaceMap =
mlir::spirv::mapMemorySpaceToVulkanStorageClass;
mlir::spirv::MemorySpaceToStorageClassConverter converter(memorySpaceMap);
for (auto gpuModule : gpuModules) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

so this means upstream is not supporting iterating over all gpu modules?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

upstream does it as well ..here https://mlir.llvm.org/doxygen/GPUToSPIRVPass_8cpp_source.html ..line 68

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I noticed that (Upstream iterating over gpu module) as well. Did not add the change to #415 since that would make the PR even bigger.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, I see it now, when I added this pass, upstream did not have this covered. Looks good.

Copy link
Contributor

@Hardcode84 Hardcode84 Oct 31, 2022 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I did this change both in main and in upstream some time ago, to support attaching different TargetEnv's to different gpu modules https://reviews.llvm.org/D135907


mlir::RewritePatternSet patterns(context);
mlir::spirv::populateMemorySpaceToStorageClassPatterns(converter, patterns);
// Map MemRef memory space to SPIR-V storage class first if requested.
if (mapMemorySpace) {
std::unique_ptr<mlir::ConversionTarget> target =
mlir::spirv::getMemorySpaceToStorageClassTarget(*context);
mlir::spirv::MemorySpaceToStorageClassMap memorySpaceMap =
mlir::spirv::mapMemorySpaceToVulkanStorageClass;
mlir::spirv::MemorySpaceToStorageClassConverter converter(memorySpaceMap);

if (failed(applyFullConversion(gpuModules, *target, std::move(patterns))))
return signalPassFailure();
}
mlir::RewritePatternSet patterns(context);
mlir::spirv::populateMemorySpaceToStorageClassPatterns(converter,
patterns);

auto targetAttr = mlir::spirv::lookupTargetEnvOrDefault(module);
std::unique_ptr<mlir::ConversionTarget> target =
mlir::SPIRVConversionTarget::get(targetAttr);
if (failed(applyFullConversion(gpuModule, *target, std::move(patterns))))
return signalPassFailure();
}

mlir::SPIRVTypeConverter typeConverter(targetAttr);
mlir::RewritePatternSet patterns(context);
auto targetAttr = mlir::spirv::lookupTargetEnvOrDefault(gpuModule);
std::unique_ptr<mlir::ConversionTarget> target =
mlir::SPIRVConversionTarget::get(targetAttr);

//------- Upstream Conversion------------
mlir::populateGPUToSPIRVPatterns(typeConverter, patterns);
mlir::arith::populateArithToSPIRVPatterns(typeConverter, patterns);
mlir::populateMemRefToSPIRVPatterns(typeConverter, patterns);
mlir::populateFuncToSPIRVPatterns(typeConverter, patterns);
// ---------------------------------------
mlir::SPIRVTypeConverter typeConverter(targetAttr);
mlir::RewritePatternSet patterns(context);

//------- Upstream Conversion------------
mlir::populateGPUToSPIRVPatterns(typeConverter, patterns);
mlir::arith::populateArithToSPIRVPatterns(typeConverter, patterns);
mlir::populateMemRefToSPIRVPatterns(typeConverter, patterns);
mlir::populateFuncToSPIRVPatterns(typeConverter, patterns);
// ---------------------------------------

// IMEX GPUToSPIRV extension
mlir::ScfToSPIRVContext scfToSpirvCtx;
mlir::populateSCFToSPIRVPatterns(typeConverter, scfToSpirvCtx, patterns);
mlir::cf::populateControlFlowToSPIRVPatterns(typeConverter, patterns);
mlir::populateMathToSPIRVPatterns(typeConverter, patterns);
// IMEX GPUToSPIRV extension
mlir::ScfToSPIRVContext scfToSpirvCtx;
mlir::populateSCFToSPIRVPatterns(typeConverter, scfToSpirvCtx, patterns);
mlir::cf::populateControlFlowToSPIRVPatterns(typeConverter, patterns);
mlir::populateMathToSPIRVPatterns(typeConverter, patterns);

if (failed(applyFullConversion(gpuModules, *target, std::move(patterns))))
return signalPassFailure();
if (failed(applyFullConversion(gpuModule, *target, std::move(patterns))))
return signalPassFailure();
}
}

std::unique_ptr<::mlir::OperationPass<::mlir::ModuleOp>>
Expand Down