[all-commits] [llvm/llvm-project] 86ad0a: [mlir:Async] Implement recursive async work splitt...

[Eugene Zhulenev via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 86ad0af87054c3cccd68d32e103a6f1f6c6194c7
      https://github.com/llvm/llvm-project/commit/86ad0af87054c3cccd68d32e103a6f1f6c6194c7
  Author: Eugene Zhulenev <ezhulenev at google.com>
  Date:   2021-06-25 (Fri, 25 Jun 2021)

  Changed paths:
    M mlir/include/mlir/Dialect/Async/Passes.h
    M mlir/include/mlir/Dialect/Async/Passes.td
    M mlir/lib/Dialect/Async/Transforms/AsyncParallelFor.cpp
    A mlir/test/Dialect/Async/async-parallel-for-async-dispatch.mlir
    A mlir/test/Dialect/Async/async-parallel-for-seq-dispatch.mlir
    R mlir/test/Dialect/Async/async-parallel-for.mlir
    M mlir/test/Integration/Dialect/Async/CPU/microbench-linalg-async-parallel-for.mlir
    A mlir/test/Integration/Dialect/Async/CPU/microbench-scf-async-parallel-for.mlir
    M mlir/test/Integration/Dialect/Async/CPU/test-async-parallel-for-1d.mlir
    M mlir/test/Integration/Dialect/Async/CPU/test-async-parallel-for-2d.mlir

  Log Message:
  -----------
  [mlir:Async] Implement recursive async work splitting for scf.parallel operation (async-parallel-for pass)

Depends On D104780

Recursive work splitting instead of sequential async tasks submission gives ~20%-30% speedup in microbenchmarks.

Algorithm outline:
1. Collapse scf.parallel dimensions into a single dimension
2. Compute the block size for the parallel operations from the 1d problem size
3. Launch parallel tasks
4. Each parallel task reconstructs its own bounds in the original multi-dimensional iteration space
5. Each parallel task computes the original parallel operation body using scf.for loop nest

Reviewed By: herhut

Differential Revision: https://reviews.llvm.org/D104850