[Mlir-commits] [mlir] [mlir][LLVM] Fix incorrect verification of atomicrmw f{min, max}imumnum (PR #190474)

[llvmlistbot at llvm.org]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-mlir-llvm

Author: Ivan R. Ivanov (ivanradanov)

<details>
<summary>Changes</summary>

Fix llvm.atomicrmw fminimumnum and fmaximumnum to correctly take the float operation verification path.

---
Full diff: https://github.com/llvm/llvm-project/pull/190474.diff


2 Files Affected:

- (modified) mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp (+3-1) 
- (modified) mlir/test/Dialect/LLVMIR/roundtrip.mlir (+2) 


``````````diff
diff --git a/mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp b/mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp
index 88b4857b52957..2b603efadb476 100644
--- a/mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp
+++ b/mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp
@@ -3663,7 +3663,9 @@ LogicalResult AtomicRMWOp::verify() {
   if (getBinOp() == AtomicBinOp::fadd || getBinOp() == AtomicBinOp::fsub ||
       getBinOp() == AtomicBinOp::fmin || getBinOp() == AtomicBinOp::fmax ||
       getBinOp() == AtomicBinOp::fminimum ||
-      getBinOp() == AtomicBinOp::fmaximum) {
+      getBinOp() == AtomicBinOp::fmaximum ||
+      getBinOp() == AtomicBinOp::fminimumnum ||
+      getBinOp() == AtomicBinOp::fmaximumnum) {
     if (isCompatibleVectorType(valType)) {
       if (isScalableVectorType(valType))
         return emitOpError("expected LLVM IR fixed vector type");
diff --git a/mlir/test/Dialect/LLVMIR/roundtrip.mlir b/mlir/test/Dialect/LLVMIR/roundtrip.mlir
index ee8c990f34d29..83d1f1b8e2884 100644
--- a/mlir/test/Dialect/LLVMIR/roundtrip.mlir
+++ b/mlir/test/Dialect/LLVMIR/roundtrip.mlir
@@ -547,6 +547,8 @@ func.func @atomicrmw(%ptr : !llvm.ptr, %f32 : f32, %f16_vec : vector<2xf16>) {
   %1 = llvm.atomicrmw volatile fsub %ptr, %f32 syncscope("singlethread") monotonic {alignment = 16 : i64} : !llvm.ptr, f32
   // CHECK: llvm.atomicrmw fmin %{{.*}}, %{{.*}} monotonic : !llvm.ptr, vector<2xf16>
   %2 = llvm.atomicrmw fmin %ptr, %f16_vec monotonic : !llvm.ptr, vector<2xf16>
+  // CHECK: llvm.atomicrmw fminimumnum %{{.*}}, %{{.*}} monotonic : !llvm.ptr, f32
+  %3 = llvm.atomicrmw fminimumnum %ptr, %f32 monotonic : !llvm.ptr, f32
   llvm.return
 }
 

``````````

</details>


https://github.com/llvm/llvm-project/pull/190474