[PATCH] D73769: Make gpu.launch implicitly capture uses of values defined above.

[Stephan Herhut via Phabricator via llvm-commits]

herhut created this revision.
Herald added subscribers: llvm-commits, Joonsoo, liufengdb, aartbik, lucyrfox, mgester, arpith-jacob, csigg, antiagainst, shauheen, burmako, jpienaar, rriddle, mehdi_amini.
Herald added a reviewer: nicolasvasilache.
Herald added a project: LLVM.
herhut added a reviewer: ftynse.

In the original design, gpu.launch required explicit capture of uses
and passing them as operands to the gpu.launch operation. This was
motivated by infrastructure reestrictions rather than design. This
change lifts the requirement and removes the concept of kernel
arguments from gpu.launch. Instead, the kernel outlining
transformation now does the explicit capturing.

This is a breaking change for users of gpu.launch.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D73769

Files:
  mlir/include/mlir/Dialect/GPU/GPUOps.td
  mlir/lib/Conversion/LoopsToGPU/LoopsToGPU.cpp
  mlir/lib/Dialect/GPU/IR/GPUDialect.cpp
  mlir/lib/Dialect/GPU/Transforms/KernelOutlining.cpp
  mlir/test/Conversion/LoopsToGPU/imperfect_2D.mlir
  mlir/test/Conversion/LoopsToGPU/linalg_to_gpu.mlir
  mlir/test/Conversion/LoopsToGPU/step_one.mlir
  mlir/test/Dialect/GPU/canonicalize.mlir
  mlir/test/Dialect/GPU/invalid.mlir
  mlir/test/Dialect/GPU/ops.mlir
  mlir/test/Dialect/GPU/outlining.mlir
  mlir/test/mlir-cuda-runner/all-reduce-op.mlir
  mlir/test/mlir-cuda-runner/all-reduce-region.mlir
  mlir/test/mlir-cuda-runner/gpu-to-cubin.mlir
  mlir/test/mlir-cuda-runner/shuffle.mlir

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D73769.241651.patch
Type: text/x-patch
Size: 41403 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200131/78bb376e/attachment.bin>