[all-commits] [llvm/llvm-project] eaaf7a: [MLIR][GPU][NVVM] Add conversion of warp synchrono...

[Navdeep Kumar via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: eaaf7a6a09da905cc314201f93e2be11773726a0
      https://github.com/llvm/llvm-project/commit/eaaf7a6a09da905cc314201f93e2be11773726a0
  Author: Navdeep Kumar <navdeep.navdeep37 at gmail.com>
  Date:   2021-05-21 (Fri, 21 May 2021)

  Changed paths:
    M mlir/include/mlir/Conversion/GPUToNVVM/GPUToNVVMPass.h
    M mlir/lib/Conversion/GPUToNVVM/CMakeLists.txt
    M mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
    A mlir/lib/Conversion/GPUToNVVM/WmmaOpsToNvvm.cpp
    A mlir/test/Conversion/GPUToNVVM/wmma-ops-to-nvvm.mlir

  Log Message:
  -----------
  [MLIR][GPU][NVVM] Add conversion of warp synchronous matrix-multiply accumulate GPU ops

Add conversion of warp synchronous matrix-multiply
accumulate GPU ops
Add conversion of warp synchronous matrix-multiply accumulate GPU ops to
NVVM ops. The following conversions are added :-
  1.) subgroup_mma_load_matrix -> wmma.m16n16k16.load.[a,b,c]..row.stride
  2.) subgroup_mma_store_matrix -> wmma.m16n16k16.store.d.[f16,f32].row.stride
  3.) subgroup_mma_compute -> wmma.m16n16k16.mma.row.row.[f16,f32].[f16,f32]

Reviewed By: bondhugula, ftynse

Differential Revision: https://reviews.llvm.org/D95331