This is an archive of the discontinued LLVM Phabricator instance.

Differential D74472

Fix MLIR build when the NVPTX target isn't configured
ClosedPublic

Authored by mehdi_amini on Feb 12 2020, 2:12 AM.

Details

Reviewers
mravishankar
herhut
ftynse
Commits
rG7b635880ab57: Fix MLIR build when the NVPTX target isn't configured

Diff Detail

Event Timeline

mehdi_amini created this revision.Feb 12 2020, 2:12 AM
Herald added a reviewer: mravishankar. · View Herald TranscriptFeb 12 2020, 2:12 AM
Herald added a project: Restricted Project. · View Herald Transcript
Herald added subscribers: llvm-commits, Joonsoo, liufengdb and 13 others. · View Herald Transcript
Harbormaster failed remote builds in B46314: Diff 244108!Feb 12 2020, 2:33 AM
herhut added a reviewer: herhut.Feb 12 2020, 2:45 AM
herhut added inline comments.
mlir/tools/mlir-opt/CMakeLists.txt
78–79

Does the mlir-cuda-runner CMakefile need this, as well? Or can this be hidden as a transitive dependency of MLIRGPUtoCUDATransforms?

mehdi_amini updated this revision to Diff 244117.Feb 12 2020, 3:14 AM

Fix build with bfd

mehdi_amini updated this revision to Diff 244118.Feb 12 2020, 3:15 AM

Update

Harbormaster failed remote builds in B46317: Diff 244117!Feb 12 2020, 3:37 AM
Harbormaster failed remote builds in B46318: Diff 244118!
ftynse accepted this revision.Feb 12 2020, 3:45 AM
This revision was not accepted when it landed; it landed in state Needs Review.Feb 12 2020, 4:41 AM
Closed by commit rG7b635880ab57: Fix MLIR build when the NVPTX target isn't configured (authored by • joker-eph-DISABLED). · Explain Why
This revision was automatically updated to reflect the committed changes.
ftynse added a comment.Feb 12 2020, 5:30 AM

It looks like similar changes may be needed to mlir-cuda-runner and, potentially, mlir-cpu-runner.

herhut added a comment.Feb 12 2020, 6:14 AM

https://reviews.llvm.org/D74480 for the mlir-cuda-runner changes. I had to include many dependencies that I assume should come with the JitRunner. Not sure what the overall goal is but this makes things build.

Revision Contents

PathSize
mlir/
lib/
Conversion/
GPUToCUDA/
27 lines
7 lines
tools/
mlir-opt/
23 lines

Diff 244137

mlir/lib/Conversion/GPUToCUDA/CMakeLists.txt

if(MLIR_CUDA_CONVERSIONS_ENABLED)
add_llvm_library(MLIRGPUtoCUDATransforms add_llvm_library(MLIRGPUtoCUDATransforms
ConvertKernelFuncToCubin.cpp ConvertKernelFuncToCubin.cpp
ConvertLaunchFuncToCudaCalls.cpp ConvertLaunchFuncToCudaCalls.cpp
) )
target_link_libraries(MLIRGPUtoCUDATransforms target_link_libraries(MLIRGPUtoCUDATransforms
LLVMNVPTXCodeGen
LLVMNVPTXDesc
LLVMNVPTXInfo
MLIRGPU MLIRGPU
MLIRLLVMIR MLIRLLVMIR
MLIRNVVMIR MLIRNVVMIR
MLIRPass MLIRPass
MLIRTargetNVVMIR MLIRTargetNVVMIR
) )
endif()

mlir/lib/Conversion/GPUToCUDA/ConvertKernelFuncToCubin.cpp

Show First 20 Lines • Show All 51 Lines • ▼ Show 20 Linespublic:
GpuKernelToCubinPass( GpuKernelToCubinPass(
CubinGenerator cubinGenerator = compilePtxToCubinForTesting) CubinGenerator cubinGenerator = compilePtxToCubinForTesting)
: cubinGenerator(cubinGenerator) {} : cubinGenerator(cubinGenerator) {}
void runOnOperation() override { void runOnOperation() override {
gpu::GPUModuleOp module = getOperation(); gpu::GPUModuleOp module = getOperation();
// Make sure the NVPTX target is initialized. // Make sure the NVPTX target is initialized.
LLVMInitializeNVPTXTarget(); llvm::InitializeAllTargets();
LLVMInitializeNVPTXTargetInfo(); llvm::InitializeAllTargetMCs();
LLVMInitializeNVPTXTargetMC(); llvm::InitializeAllAsmPrinters();
LLVMInitializeNVPTXAsmPrinter();
auto llvmModule = translateModuleToNVVMIR(module); auto llvmModule = translateModuleToNVVMIR(module);
if (!llvmModule) if (!llvmModule)
return signalPassFailure(); return signalPassFailure();
// Translate the module to CUBIN and attach the result as attribute to the // Translate the module to CUBIN and attach the result as attribute to the
// module. // module.
if (auto cubinAttr = translateGPUModuleToCubinAnnotation( if (auto cubinAttr = translateGPUModuleToCubinAnnotation(
▲ Show 20 Lines • Show All 93 Lines • Show Last 20 Lines

mlir/tools/mlir-opt/CMakeLists.txt

Show All 25 Linesset(LIBS
MLIRDialect MLIRDialect
MLIRLoopsToGPU MLIRLoopsToGPU
MLIRLinalgToLLVM MLIRLinalgToLLVM
MLIRLoopToStandard MLIRLoopToStandard
MLIREDSC MLIREDSC
MLIRFxpMathOps MLIRFxpMathOps
MLIRGPU MLIRGPU
MLIRGPUtoCUDATransforms
MLIRGPUtoNVVMTransforms MLIRGPUtoNVVMTransforms
MLIRGPUtoROCDLTransforms MLIRGPUtoROCDLTransforms
MLIRGPUtoSPIRVTransforms MLIRGPUtoSPIRVTransforms
MLIRLinalgOps MLIRLinalgOps
MLIRLinalgAnalysis MLIRLinalgAnalysis
MLIRLinalgEDSC MLIRLinalgEDSC
MLIRLinalgTransforms MLIRLinalgTransforms
MLIRLinalgUtils MLIRLinalgUtils
Show All 27 Linesset(LIBS
MLIRVectorToLLVM MLIRVectorToLLVM
MLIRVectorToLoops MLIRVectorToLoops
MLIRIR MLIRIR
MLIROptLib MLIROptLib
LLVMSupport LLVMSupport
LLVMCore LLVMCore
LLVMAsmParser LLVMAsmParser
) )
if(MLIR_CUDA_CONVERSIONS_ENABLED)
list(APPEND LIBS
LLVMNVPTXCodeGen
LLVMNVPTXDesc
LLVMNVPTXInfo
MLIRGPUtoCUDATransforms
MLIRTargetNVVMIR
)
endif()
add_llvm_tool(mlir-opt add_llvm_tool(mlir-opt
herhutUnsubmitted
Not Done
ReplyInline Actions

Does the mlir-cuda-runner CMakefile need this, as well? Or can this be hidden as a transitive dependency of MLIRGPUtoCUDATransforms?

herhut: Does the `mlir-cuda-runner` CMakefile need this, as well? Or can this be hidden as a transitive…
mlir-opt.cpp mlir-opt.cpp
) )
# Manually expand the target library, since our MLIR libraries
# aren't plugged into the LLVM dependency tracking. If we don't
# do this then we can't insert the CodeGen library after ourselves
llvm_expand_pseudo_components(TARGET_LIBS AllTargetsCodeGens)
# Prepend LLVM in front of every target, this is how the library
# are named with CMake
SET(targets_to_link)
FOREACH(t ${TARGET_LIBS})
LIST(APPEND targets_to_link "LLVM${t}")
ENDFOREACH(t)
llvm_update_compile_flags(mlir-opt) llvm_update_compile_flags(mlir-opt)
target_link_libraries(mlir-opt PRIVATE ${LIBS}) target_link_libraries(mlir-opt PRIVATE ${LIBS} ${targets_to_link})