clang format

triton-lang · Feb 8, 2025 · 2101aad · 2101aad
1 parent ed7c8b1
commit 2101aad
Show file tree

Hide file tree

Showing 3 changed files with 13 additions and 4 deletions.
diff --git a/python/test/unit/language/test_core.py b/python/test/unit/language/test_core.py
@@ -3421,6 +3421,8 @@ def get_test_dot_small_mn_fma_cases():
 
 
 def get_test_dot_double_rate_cases():
+    if not is_hip_cdna():
+        return []
     return [(32, 32, 16, 4, False, False, 'None', 'ieee', 'float16', 'float32', 1, None),
             (32, 32, 16, 4, False, False, 'None', 'ieee', 'bfloat16', 'float32', 1, None),
             (16, 16, 32, 4, False, False, 'None', 'ieee', 'float16', 'float32', 1, None),

diff --git a/python/triton/_internal_testing.py b/python/triton/_internal_testing.py
@@ -59,9 +59,15 @@ def is_hip_mi300():
         return False
     return target.arch in ('gfx940', 'gfx941', 'gfx942')
 
+def is_hip_mi350():
+    target = get_current_target()
+    if target is None or target.backend != 'hip':
+        return False
+    return target.arch in ('gfx950')
+
 
 def is_hip_cdna():
-    return is_hip_mi200() or is_hip_mi300()
+    return is_hip_mi200() or is_hip_mi300() or is_hip_mi350()
 
 
 def is_xpu():

diff --git a/third_party/amd/lib/TritonAMDGPUTransforms/AccelerateAMDMatmul.cpp b/third_party/amd/lib/TritonAMDGPUTransforms/AccelerateAMDMatmul.cpp
@@ -159,8 +159,8 @@ FailureOr<MfmaInsn> chooseMfmaInstruction(RankedTensorType cType,
 FailureOr<MfmaInsn> chooseMfmaInstruction(tt::DotOp dot, int mfmaVersion,
                                           int nonKDim) {
   RankedTensorType aType = dot.getA().getType();
-  bool allowXF32 = dot.getInputPrecision() == InputPrecision::TF32 &&
-                   mfmaVersion == 3;
+  bool allowXF32 =
+      dot.getInputPrecision() == InputPrecision::TF32 && mfmaVersion == 3;
   return chooseMfmaInstruction(dot.getC().getType(), aType.getElementType(),
                                dot.getB().getType().getElementType(),
                                aType.getShape().back(), mfmaVersion, allowXF32,
@@ -1028,7 +1028,8 @@ class TritonAMDGPUAccelerateMatmulPass
     case ISAFamily::CDNA2:
     case ISAFamily::CDNA3:
       patterns.add<::BlockedToMFMA, ::ScaledBlockedToMFMA>(
-          context, getMfmaVersion(archGenerationName), matrixInstructionSize, kPack,
+          context, getMfmaVersion(archGenerationName), matrixInstructionSize,
+          kPack,
           /*benefit=*/2);
       break;
     case ISAFamily::RDNA3: