[Mlir-commits] [mlir] 1cade86 - [mlir][arith] Fold `(a * b) / b -> a` (#121534)
llvmlistbot at llvm.org
llvmlistbot at llvm.org
Fri Jan 3 09:03:03 PST 2025
Author: Ivan Butygin
Date: 2025-01-03T20:02:59+03:00
New Revision: 1cade8699719c934a8debb7bef9fdc3ff11e9602
URL: https://github.com/llvm/llvm-project/commit/1cade8699719c934a8debb7bef9fdc3ff11e9602
DIFF: https://github.com/llvm/llvm-project/commit/1cade8699719c934a8debb7bef9fdc3ff11e9602.diff
LOG: [mlir][arith] Fold `(a * b) / b -> a` (#121534)
If overflow flags allow it.
Alive2 check: https://alive2.llvm.org/ce/z/5XWjWE
Added:
Modified:
mlir/lib/Dialect/Arith/IR/ArithOps.cpp
mlir/test/Dialect/Arith/canonicalize.mlir
Removed:
################################################################################
diff --git a/mlir/lib/Dialect/Arith/IR/ArithOps.cpp b/mlir/lib/Dialect/Arith/IR/ArithOps.cpp
index d8b314a3fa43c0..e016a6e16e59ff 100644
--- a/mlir/lib/Dialect/Arith/IR/ArithOps.cpp
+++ b/mlir/lib/Dialect/Arith/IR/ArithOps.cpp
@@ -580,11 +580,31 @@ void arith::MulUIExtendedOp::getCanonicalizationPatterns(
// DivUIOp
//===----------------------------------------------------------------------===//
+/// Fold `(a * b) / b -> a`
+static Value foldDivMul(Value lhs, Value rhs,
+ arith::IntegerOverflowFlags ovfFlags) {
+ auto mul = lhs.getDefiningOp<mlir::arith::MulIOp>();
+ if (!mul || !bitEnumContainsAll(mul.getOverflowFlags(), ovfFlags))
+ return {};
+
+ if (mul.getLhs() == rhs)
+ return mul.getRhs();
+
+ if (mul.getRhs() == rhs)
+ return mul.getLhs();
+
+ return {};
+}
+
OpFoldResult arith::DivUIOp::fold(FoldAdaptor adaptor) {
// divui (x, 1) -> x.
if (matchPattern(adaptor.getRhs(), m_One()))
return getLhs();
+ // (a * b) / b -> a
+ if (Value val = foldDivMul(getLhs(), getRhs(), IntegerOverflowFlags::nuw))
+ return val;
+
// Don't fold if it would require a division by zero.
bool div0 = false;
auto result = constFoldBinaryOp<IntegerAttr>(adaptor.getOperands(),
@@ -621,6 +641,10 @@ OpFoldResult arith::DivSIOp::fold(FoldAdaptor adaptor) {
if (matchPattern(adaptor.getRhs(), m_One()))
return getLhs();
+ // (a * b) / b -> a
+ if (Value val = foldDivMul(getLhs(), getRhs(), IntegerOverflowFlags::nsw))
+ return val;
+
// Don't fold if it would overflow or if it requires a division by zero.
bool overflowOrDiv0 = false;
auto result = constFoldBinaryOp<IntegerAttr>(
diff --git a/mlir/test/Dialect/Arith/canonicalize.mlir b/mlir/test/Dialect/Arith/canonicalize.mlir
index 6a186a0c6ceca0..522711b08f289d 100644
--- a/mlir/test/Dialect/Arith/canonicalize.mlir
+++ b/mlir/test/Dialect/Arith/canonicalize.mlir
@@ -2060,6 +2060,70 @@ func.func @test_divf1(%arg0 : f32, %arg1 : f32) -> (f32) {
// -----
+func.func @fold_divui_of_muli_0(%arg0 : index, %arg1 : index) -> index {
+ %0 = arith.muli %arg0, %arg1 overflow<nuw> : index
+ %1 = arith.divui %0, %arg0 : index
+ return %1 : index
+}
+// CHECK-LABEL: func @fold_divui_of_muli_0(
+// CHECK-SAME: %[[ARG0:.+]]: index,
+// CHECK-SAME: %[[ARG1:.+]]: index)
+// CHECK: return %[[ARG1]]
+
+func.func @fold_divui_of_muli_1(%arg0 : index, %arg1 : index) -> index {
+ %0 = arith.muli %arg0, %arg1 overflow<nuw> : index
+ %1 = arith.divui %0, %arg1 : index
+ return %1 : index
+}
+// CHECK-LABEL: func @fold_divui_of_muli_1(
+// CHECK-SAME: %[[ARG0:.+]]: index,
+// CHECK-SAME: %[[ARG1:.+]]: index)
+// CHECK: return %[[ARG0]]
+
+func.func @fold_divsi_of_muli_0(%arg0 : index, %arg1 : index) -> index {
+ %0 = arith.muli %arg0, %arg1 overflow<nsw> : index
+ %1 = arith.divsi %0, %arg0 : index
+ return %1 : index
+}
+// CHECK-LABEL: func @fold_divsi_of_muli_0(
+// CHECK-SAME: %[[ARG0:.+]]: index,
+// CHECK-SAME: %[[ARG1:.+]]: index)
+// CHECK: return %[[ARG1]]
+
+func.func @fold_divsi_of_muli_1(%arg0 : index, %arg1 : index) -> index {
+ %0 = arith.muli %arg0, %arg1 overflow<nsw> : index
+ %1 = arith.divsi %0, %arg1 : index
+ return %1 : index
+}
+// CHECK-LABEL: func @fold_divsi_of_muli_1(
+// CHECK-SAME: %[[ARG0:.+]]: index,
+// CHECK-SAME: %[[ARG1:.+]]: index)
+// CHECK: return %[[ARG0]]
+
+// Do not fold divui(mul(a, v), v) -> a with nuw attribute.
+func.func @no_fold_divui_of_muli(%arg0 : index, %arg1 : index) -> index {
+ %0 = arith.muli %arg0, %arg1 : index
+ %1 = arith.divui %0, %arg0 : index
+ return %1 : index
+}
+// CHECK-LABEL: func @no_fold_divui_of_muli
+// CHECK: %[[T0:.+]] = arith.muli
+// CHECK: %[[T1:.+]] = arith.divui %[[T0]],
+// CHECK: return %[[T1]]
+
+// Do not fold divsi(mul(a, v), v) -> a with nuw attribute.
+func.func @no_fold_divsi_of_muli(%arg0 : index, %arg1 : index) -> index {
+ %0 = arith.muli %arg0, %arg1 : index
+ %1 = arith.divsi %0, %arg0 : index
+ return %1 : index
+}
+// CHECK-LABEL: func @no_fold_divsi_of_muli
+// CHECK: %[[T0:.+]] = arith.muli
+// CHECK: %[[T1:.+]] = arith.divsi %[[T0]],
+// CHECK: return %[[T1]]
+
+// -----
+
// CHECK-LABEL: @test_cmpf(
func.func @test_cmpf(%arg0 : f32) -> (i1, i1, i1, i1) {
// CHECK-DAG: %[[T:.*]] = arith.constant true
More information about the Mlir-commits
mailing list