projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: aa2cee9) | patch
Port mlir-cuda-runner to use dialect conversion framework.
authorStephan Herhut <herhut@google.com>
Wed, 28 Aug 2019 08:50:28 +0000 (01:50 -0700)
committerA. Unique TensorFlower <gardener@tensorflow.org>
Wed, 28 Aug 2019 08:50:57 +0000 (01:50 -0700)
commit545c3e489f82cd14bbafbf45b99ef8be81ce6b62
tree0661224dc752b95e5d9ea8d2fd71b7ed7a88e796tree | snapshot
parentaa2cee9cf53678dde2950087548f502009fec814commit | diff
Port mlir-cuda-runner to use dialect conversion framework.

Instead of lowering the program in two steps (Standard->LLVM followed
by GPU->NVVM), leading to invalid IR inbetween, the runner now uses
one pattern based rewrite step to go directly from Standard+GPU to
LLVM+NVVM.

PiperOrigin-RevId: 265861934
mlir/include/mlir/Conversion/GPUToNVVM/GPUToNVVMPass.h diff | blob | history
mlir/include/mlir/Dialect/LLVMIR/NVVMDialect.h diff | blob | history
mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp diff | blob | history
mlir/tools/mlir-cuda-runner/mlir-cuda-runner.cpp diff | blob | history
Domain: System / Toolchain;