Address review comments

Author: oplavsic

python/test/unit/language/test_matmul.py

@@ -575,12 +575,11 @@ def _gemm_afp4_wfp4_kernel_preshuffled_scales_cdna4(a_ptr, b_ptr, c_ptr, a_scale
     tl.store(c_ptrs, c, mask=c_mask, cache_modifier=".wt")
 
 
-@pytest.mark.parametrize("M, N, K", [(1024, 1024, 1024), [512, 1024, 2048], [2048, 2048, 2048]])
-@pytest.mark.parametrize("BLOCK_M, BLOCK_N, BLOCK_K", [(256, 256, 256), (128, 128, 256), (128, 128, 512), [32, 32, 64]])
+@pytest.mark.parametrize("M, N, K", [(1024, 1024, 1024)])
+@pytest.mark.parametrize("BLOCK_M, BLOCK_N, BLOCK_K", [(128, 128, 256), (64, 64, 512), [32, 32, 64]])
 @pytest.mark.parametrize("matrix_instr_nonkdim", [16, 32])
 @pytest.mark.parametrize("preshuffle", [True, False])
-@pytest.mark.skipif(is_cuda(), reason="AMD specific scale shuffling")
-@pytest.mark.skipif(not is_hip_cdna4(), reason="Requires hardware support for scaled mfma instructions")
+@pytest.mark.skipif(is_hip() and not is_hip_cdna4(), reason="Requires hardware support for scaled mfma instructions")
 def test_preshuffle_scale_mxfp_cdna4(M, N, K, BLOCK_M, BLOCK_N, BLOCK_K, matrix_instr_nonkdim, preshuffle, device):
     # This test primarily evaluates correctness for efficient scale packing for MFMA-scaled instructions.
     #