[all-commits] [llvm/llvm-project] f68c40: [mlir][linalg] Implement Conv2D using Winograd Con...

[Hsiangkai Wang via All-commits]

  Branch: refs/heads/users/hsiangkai/winograd-decompose
  Home:   https://github.com/llvm/llvm-project
  Commit: f68c400a7e9f6591738e28e671aa69a8076a8cc3
      https://github.com/llvm/llvm-project/commit/f68c400a7e9f6591738e28e671aa69a8076a8cc3
  Author: Hsiangkai Wang <hsiangkai.wang at arm.com>
  Date:   2024-06-20 (Thu, 20 Jun 2024)

  Changed paths:
    M mlir/include/mlir/Dialect/Linalg/IR/LinalgOps.td
    M mlir/include/mlir/Dialect/Linalg/Transforms/Transforms.h
    M mlir/lib/Dialect/Linalg/IR/LinalgOps.cpp
    M mlir/lib/Dialect/Linalg/Transforms/CMakeLists.txt
    A mlir/lib/Dialect/Linalg/Transforms/WinogradConv2D.cpp
    A mlir/test/Dialect/Linalg/winograd-conv2d.mlir
    M mlir/test/lib/Dialect/Linalg/TestLinalgTransforms.cpp

  Log Message:
  -----------
  [mlir][linalg] Implement Conv2D using Winograd Conv2D algorithm

Define high level winograd operators and convert conv_2d_nhwc_fhwc into
winograd operators. According to Winograd Conv2D algorithm, we need
three transform operators for input, filter, and output transformation.

The formula of Winograd Conv2D algorithm is

Y = A^T x [(G x g x G^T) @ (B^T x d x B)] x A

filter transform: G x g x G^T
input transform: B^T x d x B
output transform: A^T x y x A

The implementation is based on the paper, Fast Algorithm for
Convolutional Neural Networks. (https://arxiv.org/abs/1509.09308)


  Commit: a173f8b42e16b4d4cdef81f07cd393120cc3ce44
      https://github.com/llvm/llvm-project/commit/a173f8b42e16b4d4cdef81f07cd393120cc3ce44
  Author: Hsiangkai Wang <hsiangkai.wang at arm.com>
  Date:   2024-06-20 (Thu, 20 Jun 2024)

  Changed paths:
    M mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td
    M mlir/include/mlir/Dialect/Linalg/Transforms/Transforms.h
    M mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp
    M mlir/lib/Dialect/Linalg/Transforms/WinogradConv2D.cpp
    A mlir/test/Dialect/Linalg/transform-winograd-conv2d.mlir

  Log Message:
  -----------
  [mlir][linalg] Add transform operator for Winograd Conv2D algorithm

Add a transform operator structured.winograd_conv2d to convert
linalg.conv_2d_nhwc_fhwc to Linalg winograd operators.


  Commit: 8edd982f9fccc89af77cdd9236e69a76d4e983c5
      https://github.com/llvm/llvm-project/commit/8edd982f9fccc89af77cdd9236e69a76d4e983c5
  Author: Hsiangkai Wang <hsiangkai.wang at arm.com>
  Date:   2024-06-20 (Thu, 20 Jun 2024)

  Changed paths:
    M mlir/include/mlir/Dialect/Linalg/Transforms/Transforms.h
    M mlir/lib/Dialect/Linalg/Transforms/WinogradConv2D.cpp
    A mlir/test/Dialect/Linalg/winograd-conv2d-rewrite.mlir
    M mlir/test/lib/Dialect/Linalg/TestLinalgTransforms.cpp

  Log Message:
  -----------
  [mlir][linalg] Decompose winograd operators

Convert Linalg winograd_filter_transform, winograd_input_transform, and
winograd_output_transform into nested loops with matrix multiplication
with constant transform matrices.

Support several configurations of Winograd Conv2D, including F(2, 3),
F(4, 3) and F(2, 5). These configurations show that the implementation
can support different kernel size (3 and 5) and different output size
(2 and 4). Besides symetric kernel size 3x3 and 5x5, this patch also
supports 1x3, 3x1, 1x5, and 5x1 kernels.

The implementation is based on the paper, Fast Algorithm for
Convolutional Neural Networks. (https://arxiv.org/abs/1509.09308)


Compare: https://github.com/llvm/llvm-project/compare/f68c400a7e9f%5E...8edd982f9fcc

To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications