[Mlir-commits] [mlir] [mlir][math] Rsqrt math expand pass expects static shaped operand (PR #129006)

[llvmlistbot at llvm.org]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-mlir-math

Author: Kai Sasaki (Lewuathe)

<details>
<summary>Changes</summary>

Similar to the issue reported in
https://github.com/llvm/llvm-project/pull/128299#pullrequestreview-2636142506, ExpandMath pattern for rsqrt expects the static shaped operands. Otherwise, it crashes due to the assertion violation.

See: https://github.com/llvm/llvm-project/pull/128299

---
Full diff: https://github.com/llvm/llvm-project/pull/129006.diff


2 Files Affected:

- (modified) mlir/lib/Dialect/Math/Transforms/ExpandPatterns.cpp (+5) 
- (modified) mlir/test/Dialect/Math/expand-math.mlir (+26) 


``````````diff
diff --git a/mlir/lib/Dialect/Math/Transforms/ExpandPatterns.cpp b/mlir/lib/Dialect/Math/Transforms/ExpandPatterns.cpp
index bb592c667549c..7b5350ca26b60 100644
--- a/mlir/lib/Dialect/Math/Transforms/ExpandPatterns.cpp
+++ b/mlir/lib/Dialect/Math/Transforms/ExpandPatterns.cpp
@@ -646,6 +646,11 @@ static LogicalResult convertRsqrtOp(math::RsqrtOp op,
 
   auto operand = op.getOperand();
   auto operandTy = operand.getType();
+  // Operand type must be shatic shaped type to create const float.
+  auto shapedOperandType = dyn_cast<ShapedType>(operandTy);
+  if (shapedOperandType && !shapedOperandType.hasStaticShape())
+    return failure();
+
   auto eTy = getElementTypeOrSelf(operandTy);
   if (!isa<FloatType>(eTy))
     return failure();
diff --git a/mlir/test/Dialect/Math/expand-math.mlir b/mlir/test/Dialect/Math/expand-math.mlir
index 946a411e4cc4b..8743efec5ecb4 100644
--- a/mlir/test/Dialect/Math/expand-math.mlir
+++ b/mlir/test/Dialect/Math/expand-math.mlir
@@ -787,3 +787,29 @@ func.func @unranked_ceil_op(%arg: tensor<*xf32>) -> tensor<*xf32>{
   %a = math.ceil %arg : tensor<*xf32>
   return %a: tensor<*xf32>
 }
+
+// -----
+
+// CHECK-LABEL:    func.func @non_static_shape_rsqrt_op
+// CHECK-SAME:     (%[[ARG:.*]]: tensor<?xf32>)
+// CHECK-SAME:     -> tensor<?xf32>
+// CHECK:          %[[CEIL:.*]] = math.rsqrt %[[ARG]] : tensor<?xf32>
+// CHECK:          return %[[CEIL]] : tensor<?xf32>
+
+func.func @non_static_shape_rsqrt_op(%arg: tensor<?xf32>) -> tensor<?xf32>{
+  %a = math.rsqrt %arg : tensor<?xf32>
+  return %a: tensor<?xf32>
+}
+
+// -----
+
+// CHECK-LABEL:    func.func @unranked_rsqrt_op
+// CHECK-SAME:     (%[[ARG:.*]]: tensor<*xf32>)
+// CHECK-SAME:     -> tensor<*xf32>
+// CHECK:          %[[CEIL:.*]] = math.rsqrt %[[ARG]] : tensor<*xf32>
+// CHECK:          return %[[CEIL]] : tensor<*xf32>
+
+func.func @unranked_rsqrt_op(%arg: tensor<*xf32>) -> tensor<*xf32>{
+  %a = math.rsqrt %arg : tensor<*xf32>
+  return %a: tensor<*xf32>
+}

``````````

</details>


https://github.com/llvm/llvm-project/pull/129006