Merge branch 'main' into mengfeil/enhance-workflow

Author: mengfei25

CMakeLists.txt

@@ -40,7 +40,7 @@ include(${TORCH_XPU_OPS_ROOT}/cmake/ONEMKL.cmake)
 include(${TORCH_XPU_OPS_ROOT}/cmake/BuildFlags.cmake)
 
 option(USE_XCCL "Build with XCCL support" OFF)
-option(USE_C10D_XCCL "Build with XCCL support for C10D" OFF)
+option(USE_C10D_XCCL "Build with XCCL support for C10D" ON)
 
 # -- [ Re-generate the macros file for https://github.com/pytorch/pytorch/pull/147161
 macro(update_caffe2_macros_file)

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