[Mlir-commits] [mlir] [mlir][vector] Add mask elimination transform (PR #99314)

[Andrzej Warzyński]

================
@@ -0,0 +1,129 @@
+//===- VectorMaskElimination.cpp - Eliminate Vector Masks -----------------===//
+//
+// 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/Arith/IR/Arith.h"
+#include "mlir/Dialect/Utils/StaticValueUtils.h"
+#include "mlir/Dialect/Vector/IR/ScalableValueBoundsConstraintSet.h"
+#include "mlir/Dialect/Vector/Transforms/VectorRewritePatterns.h"
+#include "mlir/Dialect/Vector/Transforms/VectorTransforms.h"
+#include "mlir/Interfaces/FunctionInterfaces.h"
+
+using namespace mlir;
+using namespace mlir::vector;
+namespace {
+
+/// If `value` is a constant multiple of `vector.vscale` return the multiplier.
+std::optional<int64_t> getConstantVscaleMultiplier(Value value) {
+  if (value.getDefiningOp<vector::VectorScaleOp>())
+    return 1;
+  auto mul = value.getDefiningOp<arith::MulIOp>();
+  if (!mul)
+    return {};
+  auto lhs = mul.getLhs();
+  auto rhs = mul.getRhs();
+  if (lhs.getDefiningOp<vector::VectorScaleOp>())
+    return getConstantIntValue(rhs);
+  if (rhs.getDefiningOp<vector::VectorScaleOp>())
+    return getConstantIntValue(lhs);
+  return {};
+}
+
+/// Attempts to resolve a (scalable) CreateMaskOp to an all-true constant mask.
+/// All-true masks can then be eliminated by simple folds.
+LogicalResult resolveAllTrueCreateMaskOp(IRRewriter &rewriter,
+                                         vector::CreateMaskOp createMaskOp,
+                                         VscaleRange vscaleRange) {
+  auto maskType = createMaskOp.getVectorType();
+  auto maskTypeDimScalableFlags = maskType.getScalableDims();
+  auto maskTypeDimSizes = maskType.getShape();
+
+  struct UnknownMaskDim {
+    size_t position;
+    Value dimSize;
+  };
+
+  // Check for any dims that could be (partially) false before doing the more
+  // expensive value bounds computations.
+  SmallVector<UnknownMaskDim> unknownDims;
+  for (auto [i, dimSize] : llvm::enumerate(createMaskOp.getOperands())) {
+    if (auto intSize = getConstantIntValue(dimSize)) {
+      // Mask not all-true for this dim.
+      if (maskTypeDimScalableFlags[i] || intSize < maskTypeDimSizes[i])
+        return failure();
----------------
banach-space wrote:

IIUC, `getConstantIntValue` would return `nullopt` for `vscale`? Shouldn't it always be failure if the mask is scalable and `intSize` is _not_ a multiple of `vscale` (i.e. a compile-time known int).

https://github.com/llvm/llvm-project/pull/99314