[Mlir-commits] [mlir] [mlir][linalg] unfold projected permutation. (PR #114704)

[Andrzej Warzyński]

================
@@ -0,0 +1,243 @@
+//===- UnfoldProjectedPermutation.cpp - extract projected projections   ---===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+#include "mlir/Dialect/Linalg/Transforms/Transforms.h"
+#include <utility>
+
+#include "mlir/Dialect/Affine/IR/AffineOps.h"
+#include "mlir/Dialect/Linalg/IR/Linalg.h"
+#include <map>
+#include <optional>
+#include <vector>
+
+using namespace mlir;
+using namespace mlir::linalg;
+
+namespace {
+
+/// This file implements pattern to decompose the input operand(s) of a
+/// linalg.generic that has a `transpose`, `broadcast` or a mixture of two,
+/// into explicit transpose and broadcast. Having them folded into the
+/// linalg.generic is a good optimization but sometimes we may want to unwrap
+/// i.e. `unfold` them as explicit transpose and broadcast. This rewrite
+/// pattern helps do it for each input operand. This is useful for instance
+/// when trying to recognize named ops.
+///
+/// The transpose, broadcast, or mixture of both, are expressed in the affine
+/// map of the operand. Technically it is essentially `projected permutation`.
+///
+///  Example
+///
+/// ```mlir
+///
+/// #projection = affine_map<(d0, d1, d2, d3, d4) -> (d2, d3, d1)>
+/// #identity   = affine_map<(d0, d1, d2, d3, d4) -> (d0, d1, d2, d3, d4)>
+/// ...
+///    %res = linalg.generic
+///       { indexing_maps = [#projection, #identity, #identity],
+///       iterator_types = ["parallel", "parallel", "parallel",
+///                         "parallel", "parallel"]}
+///       ins(%x, %y : tensor<7x8x9xf32>, tensor<5x9x7x8x10xf32>)
+///       outs(%z : tensor<5x9x7x8x10xf32>) {
+///         ^bb0(%in: f32, %in_1: f32, %out: f32):
+///              %div = arith.divf %in, %in_1 : f32
+///              linalg.yield %div : f32
+///    } -> tensor<5x9x7x8x10xf32>
+/// ```
+///
+/// In the above IR operand `%x` map is a projected-permutation. This can be
+/// unfolded as:
+///
+/// ```mlir
+///   ...
+///   %x_trans = linalg.transpose
+///                   ins(%x : tensor<7x8x9xf32>)
+///                   outs(%e1 : tensor<9x7x8xf32>) permutation = [2, 0, 1]
+///   ...
+///   %x_trans_bc = linalg.broadcast
+///                   ins(%x_trans : tensor<9x7x8xf32>)
+///                   outs(%e2 : tensor<5x9x7x8x10xf32>) dimensions = [0, 4]
+///   %2 = linalg.div
+///           ins(%x_trans_bc, %y :
+///                  tensor<5x9x7x8x10xf32>, tensor<5x9x7x8x10xf32>)
+///           outs(%arg2 : tensor<5x9x7x8x10xf32>) -> tensor<5x9x7x8x10xf32>
+///
+/// Note that linalg.generic has been 'specialized' to linalg.div.
+///
+/// To unfold it is more effective to transpose first and then do the broadcast.
+/// However, if transpose is done first, the permutation map needs to be
+/// expressed in terms of reduced dimension (as broadcast hasn't happened yet).
+/// Also, the broadcast dimensions in a linalg.generic come from other operands
+/// (those not broadcasted along that particular dimension). We work this out
+/// by computing the convex-polyhedron shape of the linalg.gneric iteration
+/// space from shapes of all the operands (inputs and outputs).
+///
+struct UnfoldProjectedPermutation : public OpRewritePattern<GenericOp> {
+  using OpRewritePattern<GenericOp>::OpRewritePattern;
+
+  LogicalResult matchAndRewrite(GenericOp genericOp,
+                                PatternRewriter &rewriter) const override;
+};
+
+/// For the given `map` determine what dimensions are transposed
+/// and what dimensions are broadcasted.
+/// Returns :
+///  `isTransposed, isBroadcast,
+///   transpose-permutation, broadcast-dimensions`
+///
+std::tuple<bool, bool, SmallVector<int64_t>, SmallVector<int64_t>>
+computeTransposeBroadcast(AffineMap &map) {
+  assert(map.isProjectedPermutation(false) && "not a projection");
+  int64_t minorSize = map.getNumResults();
+
+  SmallVector<int64_t> minorResult;
+  for (int64_t i = 0; i < minorSize; ++i) {
+    auto expr = cast<AffineDimExpr>(map.getResults()[i]);
+    minorResult.push_back(expr.getPosition());
+  }
+
+  // If dims are not monotonically increasing then transpose is present.
+  SmallVector<int64_t> sortedResMap(minorResult);
+  std::sort(sortedResMap.begin(), sortedResMap.end());
+  bool hasTranspose = !std::equal(minorResult.begin(), minorResult.end(),
+                                  sortedResMap.begin(), sortedResMap.end());
+
+  // Walk the sorted map result to determine which dimensions are broadcasted.
+  SmallVector<int64_t> broadcast;
+  for (int64_t i = 0, j = 0; i < map.getNumInputs(); ++i) {
+    if (j < minorSize && sortedResMap[j] == i) {
+      j++;
+      continue;
+    }
+    broadcast.push_back(i);
+  }
+  bool hasBroadcast = !broadcast.empty();
+
+  /// Consider an operand `x : tensor<7x8x9>` of a genericOp that has
+  /// affine map `affine_map<(d0, d1, d2, d3, d4) -> (d2, d3, d1)>`
+  /// `x`s access is both transposed and brodcast. But when specifying
+  /// the `linalg.transpose(x : tensor<7x8x9>)` the dimensions need to be
+  /// specified as `affine_map<(d0,d1,d2) -> (d1, d2, d0)` instead of
+  /// refering to d3, d4. Therefore, re-base the transpose dimensions so
+  /// that they start from d0.
+  std::map<int64_t, int64_t> minorMap;
+  for (int64_t i = 0; i < minorSize; ++i)
+    minorMap.insert({sortedResMap[i], i});
+
+  // Re-map the dimensions.
+  SmallVector<int64_t> remappedResult(minorSize);
+  for (int64_t i = 0; i < minorSize; ++i)
+    remappedResult[i] = minorMap[minorResult[i]];
+
+  /// Calculate the permutation for the transpose.
+  SmallVector<int64_t> permutation(minorSize);
+  for (unsigned i = 0; i < minorSize; ++i) {
+    permutation[remappedResult[i]] = i;
+  }
+
+  return {hasTranspose, hasBroadcast, permutation, broadcast};
+}
+
+LogicalResult
+UnfoldProjectedPermutation::matchAndRewrite(GenericOp op,
+                                            PatternRewriter &rewriter) const {
+  if (!op.hasPureTensorSemantics() || op.isSingleInputOutput() ||
+      op.isSingleYieldOp() || !op.isAllParallelLoops())
+    return failure();
+
+  // All maps need to be projected permutations.
+  for (auto &opOperand : op->getOpOperands()) {
+    auto map = op.getMatchingIndexingMap(&opOperand);
+    if (!map.isProjectedPermutation(false))
+      return failure();
+  }
+
+  // Currently we handle only static shapes.
----------------
banach-space wrote:

OK, rather than documenting what the code does, could you add a comment saying "why"? Or what's missing? From what you are saying, we'd need to add logic to compute dynamic sizes of the input tensors for ops like `EmptyOp`? And probably sth else as well?

https://github.com/llvm/llvm-project/pull/114704