[Mlir-commits] [mlir] [mlir] Add `vector.store` of `memref.subview` memref alias folding (PR #72184)

[llvmlistbot at llvm.org]

https://github.com/Max191 updated https://github.com/llvm/llvm-project/pull/72184

>From 5a82f18745f272f43ee6caabe18c920aa5dea358 Mon Sep 17 00:00:00 2001
From: Max Dawkins <max.dawkins at gmail.com>
Date: Fri, 10 Nov 2023 11:27:13 -0500
Subject: [PATCH] [mlir] Add `vector.store` and `vector.maskedstore` of
 `memref.subview` memref alias folding

---
 .../MemRef/Transforms/FoldMemRefAliasOps.cpp  | 15 ++++++++
 .../Dialect/MemRef/fold-memref-alias-ops.mlir | 35 +++++++++++++++++++
 2 files changed, 50 insertions(+)

diff --git a/mlir/lib/Dialect/MemRef/Transforms/FoldMemRefAliasOps.cpp b/mlir/lib/Dialect/MemRef/Transforms/FoldMemRefAliasOps.cpp
index b78c4510ff88585..aa44455ada7f9aa 100644
--- a/mlir/lib/Dialect/MemRef/Transforms/FoldMemRefAliasOps.cpp
+++ b/mlir/lib/Dialect/MemRef/Transforms/FoldMemRefAliasOps.cpp
@@ -187,8 +187,12 @@ static Value getMemRefOperand(nvgpu::LdMatrixOp op) {
 
 static Value getMemRefOperand(vector::LoadOp op) { return op.getBase(); }
 
+static Value getMemRefOperand(vector::StoreOp op) { return op.getBase(); }
+
 static Value getMemRefOperand(vector::MaskedLoadOp op) { return op.getBase(); }
 
+static Value getMemRefOperand(vector::MaskedStoreOp op) { return op.getBase(); }
+
 static Value getMemRefOperand(vector::TransferWriteOp op) {
   return op.getSource();
 }
@@ -557,6 +561,15 @@ LogicalResult StoreOpOfSubViewOpFolder<OpTy>::matchAndRewrite(
                 subViewOp.getDroppedDims())),
             op.getMask(), op.getInBoundsAttr());
       })
+      .Case([&](vector::StoreOp op) {
+        rewriter.replaceOpWithNewOp<vector::StoreOp>(
+            op, op.getValueToStore(), subViewOp.getSource(), sourceIndices);
+      })
+      .Case([&](vector::MaskedStoreOp op) {
+        rewriter.replaceOpWithNewOp<vector::MaskedStoreOp>(
+            op, subViewOp.getSource(), sourceIndices, op.getMask(),
+            op.getValueToStore());
+      })
       .Case([&](gpu::SubgroupMmaStoreMatrixOp op) {
         rewriter.replaceOpWithNewOp<gpu::SubgroupMmaStoreMatrixOp>(
             op, op.getSrc(), subViewOp.getSource(), sourceIndices,
@@ -698,6 +711,8 @@ void memref::populateFoldMemRefAliasOpPatterns(RewritePatternSet &patterns) {
                StoreOpOfSubViewOpFolder<affine::AffineStoreOp>,
                StoreOpOfSubViewOpFolder<memref::StoreOp>,
                StoreOpOfSubViewOpFolder<vector::TransferWriteOp>,
+               StoreOpOfSubViewOpFolder<vector::StoreOp>,
+               StoreOpOfSubViewOpFolder<vector::MaskedStoreOp>,
                StoreOpOfSubViewOpFolder<gpu::SubgroupMmaStoreMatrixOp>,
                LoadOpOfExpandShapeOpFolder<affine::AffineLoadOp>,
                LoadOpOfExpandShapeOpFolder<memref::LoadOp>,
diff --git a/mlir/test/Dialect/MemRef/fold-memref-alias-ops.mlir b/mlir/test/Dialect/MemRef/fold-memref-alias-ops.mlir
index 8fe87bc8c57c300..96b72e042b9e0d6 100644
--- a/mlir/test/Dialect/MemRef/fold-memref-alias-ops.mlir
+++ b/mlir/test/Dialect/MemRef/fold-memref-alias-ops.mlir
@@ -803,3 +803,38 @@ func.func @fold_vector_maskedload(
 // CHECK-SAME:   %[[ARG3:[a-zA-Z0-9_]+]]: vector<32xi1>
 // CHECK-SAME:   %[[ARG4:[a-zA-Z0-9_]+]]: vector<32xf32>
 //      CHECK:   vector.maskedload %[[ARG0]][%[[ARG1]], %[[ARG2]]], %[[ARG3]], %[[ARG4]] : memref<12x32xf32>, vector<32xi1>, vector<32xf32> into vector<32xf32>
+
+// -----
+
+func.func @fold_vector_store(
+  %arg0 : memref<12x32xf32>, %arg1 : index, %arg2 : index, %arg3: vector<2x32xf32>) -> () {
+  %0 = memref.subview %arg0[%arg1, %arg2][1, 1][1, 1] : memref<12x32xf32> to memref<f32, strided<[], offset: ?>>
+  vector.store %arg3, %0[] : memref<f32, strided<[], offset: ?>>, vector<2x32xf32>
+  return
+}
+
+//      CHECK: func @fold_vector_store
+// CHECK-SAME:   %[[ARG0:[a-zA-Z0-9_]+]]: memref<12x32xf32>
+// CHECK-SAME:   %[[ARG1:[a-zA-Z0-9_]+]]: index
+// CHECK-SAME:   %[[ARG2:[a-zA-Z0-9_]+]]: index
+// CHECK-SAME:   %[[ARG3:[a-zA-Z0-9_]+]]: vector<2x32xf32>
+//      CHECK:   vector.store %[[ARG3]], %[[ARG0]][%[[ARG1]], %[[ARG2]]] :  memref<12x32xf32>, vector<2x32xf32>
+//      CHECK:   return
+
+// -----
+
+func.func @fold_vector_maskedstore(
+  %arg0 : memref<12x32xf32>, %arg1 : index, %arg2 : index, %arg3: vector<32xi1>, %arg4: vector<32xf32>) -> () {
+  %0 = memref.subview %arg0[%arg1, %arg2][1, 1][1, 1] : memref<12x32xf32> to memref<f32, strided<[], offset: ?>>
+  vector.maskedstore %0[], %arg3, %arg4 : memref<f32, strided<[], offset: ?>>, vector<32xi1>, vector<32xf32>
+  return
+}
+
+//      CHECK: func @fold_vector_maskedstore
+// CHECK-SAME:   %[[ARG0:[a-zA-Z0-9_]+]]: memref<12x32xf32>
+// CHECK-SAME:   %[[ARG1:[a-zA-Z0-9_]+]]: index
+// CHECK-SAME:   %[[ARG2:[a-zA-Z0-9_]+]]: index
+// CHECK-SAME:   %[[ARG3:[a-zA-Z0-9_]+]]: vector<32xi1>
+// CHECK-SAME:   %[[ARG4:[a-zA-Z0-9_]+]]: vector<32xf32>
+//      CHECK:   vector.maskedstore %[[ARG0]][%[[ARG1]], %[[ARG2]]], %[[ARG3]], %[[ARG4]] : memref<12x32xf32>, vector<32xi1>, vector<32xf32>
+//      CHECK:   return