[Mlir-commits] [mlir] [mlir][linalg] Implement Conv2D using Winograd Conv2D algorithm (PR #96181)

[Hsiangkai Wang]

================
@@ -0,0 +1,321 @@
+//===- WinogradConv2D.cpp - Winograd Conv2D implementation ----------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// Implement Winograd Conv2D algorithm. The implementation is based on the
+// paper: Fast Algorithms for Convolutional Neural Networks
+// (https://arxiv.org/abs/1509.09308)
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Dialect/Linalg/IR/Linalg.h"
+#include "mlir/Dialect/Tensor/IR/Tensor.h"
+#include "mlir/Dialect/Tosa/Utils/ConversionUtils.h"
+#include "mlir/Transforms/GreedyPatternRewriteDriver.h"
+#include "llvm/Support/MathExtras.h"
+
+namespace mlir {
+namespace linalg {
+
+namespace {
+
+using TransformMapKeyTy = std::pair<int, int>;
+
+// We use F(m, r) to define the size of minimal filtering algorithms.
+// m is the output dimension and r is the filter dimension. We can get
+// the input dimension, alpha, from the formula, alpha = m + r - 1.
+//
+// For example, when m = 2 and r = 3, we know its input size is 4.
+// The Conv2D will operate on 4x4 input data with 3x3 filter and get
+// 2x2 output result.
+constexpr TransformMapKeyTy F_2_3{2, 3};
+constexpr TransformMapKeyTy F_4_3{4, 3};
+constexpr TransformMapKeyTy F_2_5{2, 5};
+
+Value collapse2DData(RewriterBase &rewriter, Location loc, Value data) {
+  auto type = cast<ShapedType>(data.getType());
+  auto elementType = type.getElementType();
+  auto shape = type.getShape();
+  auto collapseType = RankedTensorType::get(
+      {shape[0] * shape[1] * shape[2] * shape[3], shape[4], shape[5]},
+      elementType);
+  SmallVector<ReassociationIndices> reassociation = {{0, 1, 2, 3}, {4}, {5}};
+  return rewriter.create<tensor::CollapseShapeOp>(loc, collapseType, data,
+                                                  reassociation);
+}
+
+// This function generates linalg.batch_matmul to multiply input with filter.
+// linalg.batch_matmul only supports 3-dimension data sets. We can treat
+// tileH x tileW x H x W data as the 1-dimension data array. That is to convert
+// [tileH, tileW, H, W, N, C] to [tileH x tileW x H x W, N, C]. In this way, we
+// can convert 6-dimension input data to 3-dimension representation that is
+// suitable for linalg.batch_matmul.
+//
+// Batched matmul will do the matrix multiply with the reduction on channel.
+//
+// We get
+//
+// %collapsed_input = tensor.collapse_shape %input
+// %collapsed_filter = tensor.collapse_shape %filter
+// %ret = linalg.batch_matmul %collapsed_input, %collapsed_filter
+// %expanded_ret = tensor.expand_shape %ret
+//
+// After this function, we get return value with data layout
+// (tileH, tileW, H, W, N, F).
+Value matrixMultiply(RewriterBase &rewriter, Location loc,
+                     Value transformedFilter, Value transformedInput) {
+  auto collapseFilter = collapse2DData(rewriter, loc, transformedFilter);
+  auto collapseInput = collapse2DData(rewriter, loc, transformedInput);
+
+  // Batched matrix multiply
+  auto filterType = cast<ShapedType>(transformedFilter.getType());
+  auto filterShape = filterType.getShape();
+  auto inputType = cast<ShapedType>(transformedInput.getType());
+  auto inputElemType = inputType.getElementType();
+  auto inputShape = inputType.getShape();
+
+  auto matmulType = RankedTensorType::get(
+      {inputShape[0] * inputShape[1] * inputShape[2] * inputShape[3],
+       inputShape[4], filterShape[5]},
+      inputElemType);
+  Value init = rewriter.create<tensor::EmptyOp>(loc, matmulType.getShape(),
+                                                inputElemType);
+
+  auto matmulOp = rewriter.create<linalg::BatchMatmulOp>(
+      loc, matmulType, ValueRange({collapseInput, collapseFilter}),
+      ValueRange{init});
+
+  // Expand matmul result
+  SmallVector<ReassociationIndices> reassociation = {{0, 1, 2, 3}, {4}, {5}};
+  auto expandType =
+      RankedTensorType::get({inputShape[0], inputShape[1], inputShape[2],
+                             inputShape[3], inputShape[4], filterShape[5]},
+                            inputElemType);
+  auto expandOutput = rewriter.create<tensor::ExpandShapeOp>(
+      loc, expandType, matmulOp.getResult(0), reassociation);
+  return expandOutput;
+}
+
+Value insertToAlignedTensor(RewriterBase &rewriter, Location loc, Value value,
+                            RankedTensorType alignedType) {
+  Value alignedInput = rewriter.create<tensor::EmptyOp>(
+      loc, alignedType.getShape(), alignedType.getElementType());
+
+  auto zeroIndex = rewriter.getIndexAttr(0);
+  auto oneIndex = rewriter.getIndexAttr(1);
+  SmallVector<OpFoldResult, 4> offsets(4, zeroIndex);
+  SmallVector<OpFoldResult, 4> strides(4, oneIndex);
+
+  auto valueType = cast<ShapedType>(value.getType());
+  auto valueShape = valueType.getShape();
+  SmallVector<OpFoldResult, 4> sizes;
+  sizes.emplace_back(rewriter.getIndexAttr(valueShape[0]));
+  sizes.emplace_back(rewriter.getIndexAttr(valueShape[1]));
+  sizes.emplace_back(rewriter.getIndexAttr(valueShape[2]));
+  sizes.emplace_back(rewriter.getIndexAttr(valueShape[3]));
----------------
Hsiangkai wrote:

Thanks for your tips. It is more elegant to use `getAsOpFoldResult`. 

https://github.com/llvm/llvm-project/pull/96181