[rocm7.0_internal_testing] unskip rowwise tests (#2567)

Signed-off-by: Jagadish Krishnamoorthy <[email protected]>

Author: jagadish-amd

test/test_matmul_cuda.py

@@ -46,7 +46,6 @@
     parametrize,
     run_tests,
     skipIfRocm,
-    skipIfRocmVersionAndArch,
     skipIfRocmVersionLessThan,
     TEST_CUDA,
     TEST_WITH_ROCM,
@@ -908,7 +907,6 @@ def test_float8_scale_fast_accum(self, device) -> None:
         out_fp8_s = torch._scaled_mm(x, y, scale_a=scale_a, scale_b=scale_b, use_fast_accum=True)
         self.assertEqual(out_fp8, out_fp8_s)
 
-    @skipIfRocmVersionAndArch((7, 1), "gfx950")
     @unittest.skipIf(not PLATFORM_SUPPORTS_FP8 or IS_WINDOWS, f8_msg)
     @unittest.skipIf(not SM89OrLater, "rowwise implementation is currently sm89+ specific")
     @parametrize("use_fast_accum", [True, False])
@@ -1014,7 +1012,6 @@ def test_float8_error_messages(self, device) -> None:
                 out_dtype=torch.bfloat16,
             )
 
-    @skipIfRocmVersionAndArch((7, 1), "gfx950")
     @unittest.skipIf(not PLATFORM_SUPPORTS_FP8 or IS_WINDOWS, f8_msg)
     @unittest.skipIf(not SM89OrLater, "rowwise implementation is currently sm89+ specific")
     @parametrize("base_dtype", [torch.bfloat16])