skip the sparse rowwise fp8 cutlass kernel test on ROCm (#1906)

vkuzo · web-flow · commit b1ecf659233c · 2025-03-17T14:58:24.000-07:00
Update

[ghstack-poisoned]
diff --git a/test/test_ops_rowwise_scaled_linear_sparse_cutlass.py b/test/test_ops_rowwise_scaled_linear_sparse_cutlass.py
@@ -18,6 +18,7 @@
     _float8_cutlass_quant_sparse,
 )
 from torchao.sparsity.utils import create_semi_structured_tensor
+from torchao.testing.utils import skip_if_rocm
 
 DTYPES = [torch.float16, torch.bfloat16]
 XQ_WQ_DTYPES = [
@@ -104,6 +105,7 @@ def run_test_for_op(
     )
 
 
+@skip_if_rocm("does not yet work on ROCm")
 @pytest.mark.skipif(not torch.cuda.is_available(), reason="CUDA not available")
 @pytest.mark.skipif(not SM90OrLater, reason="FP8 is only supported on H100+ devices")
 @pytest.mark.parametrize(

Original file line number	Diff line number	Diff line change
`@@ -18,6 +18,7 @@`
`18`	`18`	`_float8_cutlass_quant_sparse,`
`19`	`19`	`)`
`20`	`20`	`from torchao.sparsity.utils import create_semi_structured_tensor`
	`21`	`+from torchao.testing.utils import skip_if_rocm`
`21`	`22`
`22`	`23`	`DTYPES = [torch.float16, torch.bfloat16]`
`23`	`24`	`XQ_WQ_DTYPES = [`
`@@ -104,6 +105,7 @@ def run_test_for_op(`
`104`	`105`	`)`
`105`	`106`
`106`	`107`
	`108`	`+@skip_if_rocm("does not yet work on ROCm")`
`107`	`109`	`@pytest.mark.skipif(not torch.cuda.is_available(), reason="CUDA not available")`
`108`	`110`	`@pytest.mark.skipif(not SM90OrLater, reason="FP8 is only supported on H100+ devices")`
`109`	`111`	`@pytest.mark.parametrize(`