From 7ad3108ef2cbc04bc03d7ca6331c49e535d3aa2f Mon Sep 17 00:00:00 2001 From: eqy Date: Thu, 15 Aug 2024 23:43:28 +0000 Subject: [PATCH] [CUTLASS][FP8] Skip scaled_mm rowwise test on sm89 (#133612) Rowwise implementation currently uses sm90-specific features incl. TMA CC @drisspg Pull Request resolved: https://github.com/pytorch/pytorch/pull/133612 Approved by: https://github.com/Skylion007 --- test/test_matmul_cuda.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/test/test_matmul_cuda.py b/test/test_matmul_cuda.py index 0dea610bf82..70235665cb6 100644 --- a/test/test_matmul_cuda.py +++ b/test/test_matmul_cuda.py @@ -16,6 +16,7 @@ from torch.quantization._quantized_conversions import ( from torch.testing import make_tensor from torch.testing._internal.common_cuda import ( SM53OrLater, + SM90OrLater, _get_torch_cuda_version, PLATFORM_SUPPORTS_FP8 ) @@ -664,6 +665,7 @@ class TestFP8MatmulCuda(TestCase): ) @unittest.skipIf(not PLATFORM_SUPPORTS_FP8 or IS_WINDOWS, f8_msg) + @unittest.skipIf(not SM90OrLater, "rowwise implementation is currently sm90 specific") @skipIfRocm() @parametrize("base_dtype", [torch.bfloat16]) def test_scaled_mm_vs_emulated_row_wise(self, base_dtype):