Merge branch 'main' into chao/avg

Author: Chao1Han

.github/workflows/_linux_build.yml

@@ -120,7 +120,8 @@ jobs:
                 echo "Wheel build successful, update last commit in the issue https://github.com/intel/torch-xpu-ops/issues/1280"
                 gh --repo $repo issue view $commit_issue --json body -q .body | sed "s;${last_commit};${current_commit};g" | sed '/^$/d' > new_body.txt
                 gh --repo $repo issue edit $commit_issue --body-file new_body.txt
-              else
+              fi
+              if [ ! -f dist/torch*.whl ]; then
                 echo "Wheel build failed, use last commit in the issue https://github.com/intel/torch-xpu-ops/issues/1280"
                 gh --repo $repo issue comment $commit_issue -b "Wheel build failed with commit [${current_commit}](https://github.com/pytorch/pytorch/tree/${current_commit}), refer ${build_url}. CC @intel/torch-xpu-ops-maintain @EikanWang @riverliuintel @fengyuan14 @xytintel @etaf @chuanqi129 @mengfei25"
                 git clean -df .

test/regressions/test_torchvision_roi_ops.py

@@ -471,7 +471,6 @@ def expected_fn(
     )  # , ids=str)
     @pytest.mark.parametrize("contiguous", (True, False))
     @pytest.mark.parametrize("deterministic", (True, False))
-    @pytest.mark.opcheck_only_one
     def test_forward(
         self, device, contiguous, deterministic, aligned, x_dtype, rois_dtype=None
     ):
@@ -486,22 +485,6 @@ def test_forward(
             aligned=aligned,
         )
 
-    @pytest.mark.parametrize("aligned", (True, False))
-    @pytest.mark.parametrize("deterministic", (True, False))
-    @pytest.mark.parametrize("x_dtype", (torch.float, torch.half))
-    @pytest.mark.parametrize("rois_dtype", (torch.float, torch.half))
-    @pytest.mark.opcheck_only_one
-    def test_autocast(self, aligned, deterministic, x_dtype, rois_dtype):
-        with torch.amp.autocast("xpu"):
-            self.test_forward(
-                torch.device("xpu"),
-                contiguous=False,
-                deterministic=deterministic,
-                aligned=aligned,
-                x_dtype=x_dtype,
-                rois_dtype=rois_dtype,
-            )
-
 
 class TestPSRoIAlign(RoIOpTester):
     mps_backward_atol = 5e-2

test/xpu/skip_list_common.py

@@ -1421,6 +1421,9 @@
         "test_numeric_check_leak_tunableop_rocm_xpu_float32",
         "test_dump_results_on_exit_tunableop_xpu_float32",
         "test_rotating_buffer_tunableop_xpu_float32",
+        "test_gemm_bias_tunableop_xpu_bfloat16",
+        "test_scaled_gemm_tunableop_xpu_float8_e4m3fnuz",
+        "test_scaled_gemm_tunableop_xpu_float8_e5m2fnuz",
         # CUDA bias cases added in latest PyTorch
         # AttributeError: module 'torch._C' has no attribute '_cuda_tunableop_enable'
         "test_matmul_check_entries_tunableop_xpu_float16",