Merge pull request triton-lang#19 from ROCmSoftwarePlatform/unskip_te…

…st_reduce

reduce the skips for test_reduce functions

python/test/unit/language/test_core.py

@@ -892,7 +892,8 @@ def copy_kernel(input_ptr, output_ptr, n_elements, BLOCK_SIZE: tl.constexpr):
 def test_reduce1d(op, dtype_str, shape, device='cuda'):
     check_type_supported(dtype_str)  # bfloat16 on cc < 80 will not be tested
     if torch.version.hip is not None:
-        pytest.skip(f"test_reduce1d currently has segfaults on ROCM")
+        if dtype_str in ["int8", "int16", "uint8", "uint16"]:
+            pytest.skip(f"test_reduce1d[{dtype_str}] skipped on ROCM")
 
     # triton kernel
     @triton.jit
@@ -953,7 +954,8 @@ def kernel(X, Z, BLOCK: tl.constexpr):
 def test_reduce2d(op, dtype_str, shape, axis, device='cuda'):
     check_type_supported(dtype_str)  # bfloat16 on cc < 80 will not be tested
     if torch.version.hip is not None:
-        pytest.skip(f"test_reduce2d currently has segfaults on ROCM")
+        if dtype_str in ["int8", "int16", "uint8", "uint16"]:
+            pytest.skip(f"test_reduce2d[{dtype_str}] skipped on ROCM")
     # triton kernel
     @triton.jit
     def kernel(X, Z, BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr, AXIS: tl.constexpr):