Allow specification of the workgroup size for GPUToSPIRV lowering.
authorMahesh Ravishankar <ravishankarm@google.com>
Thu, 5 Dec 2019 19:31:28 +0000 (11:31 -0800)
committerA. Unique TensorFlower <gardener@tensorflow.org>
Thu, 5 Dec 2019 19:31:57 +0000 (11:31 -0800)
commit4d61a79db46e1cab990444cbcb84467faff7f042
tree8e15f13bb127d3b2b8c8430e42fdc75f568d52a3
parent037044b0ae4bc72f7c53db4f8435f7d163d2b68a
Allow specification of the workgroup size for GPUToSPIRV lowering.

SPIR-V/Vulkan spec requires the workgroups size to be specified with
the spv.ExecutionMode operation. This was hard-wired to be set to a
particular value. It is now changed to be configurable by clients of
the pass or of the patterns that implement the lowering from GPU to
SPIRV.

PiperOrigin-RevId: 284017482
mlir/include/mlir/Conversion/GPUToSPIRV/ConvertGPUToSPIRV.h
mlir/include/mlir/Conversion/GPUToSPIRV/ConvertGPUToSPIRVPass.h
mlir/lib/Conversion/GPUToSPIRV/CMakeLists.txt
mlir/lib/Conversion/GPUToSPIRV/ConvertGPUToSPIRV.cpp
mlir/lib/Conversion/GPUToSPIRV/ConvertGPUToSPIRVPass.cpp
mlir/test/Conversion/GPUToSPIRV/simple.mlir