use kwarg arch from ipex (#105) (#107)

* use cc from ipex * refine code (cherry picked from commit c3ecd6d)
intel · Sep 26, 2023 · 69998dd · 69998dd
1 parent 7fc9f9b
commit 69998dd
Showing 1 changed file with 6 additions and 4 deletions.
diff --git a/python/__init__.py b/python/__init__.py
@@ -420,10 +420,12 @@ def get_kernel_bin(self):
         return "spvbin"
 
     def get_architecture_descriptor(self, **kwargs):
-        dev_props = self.driver.utils.get_device_properties(torch.xpu.device(torch.xpu.current_device()).sycl_device)  # noqa: E501
-        max_work_group_size = dev_props['max_work_group_size']
-        max_num_sub_groups = dev_props['max_num_sub_groups']
-        sub_group_sizes = dev_props['sub_group_sizes']
+        arch = kwargs.get("arch", None)
+        if arch is None:
+            arch = self.get_device_properties(self.get_current_device())
+        max_work_group_size = arch['max_work_group_size']
+        max_num_sub_groups = arch['max_num_sub_groups']
+        sub_group_sizes = arch['sub_group_sizes']
         # TODO: chose a reasonable subgroup size
         threads_per_warp = 32
         assert threads_per_warp in sub_group_sizes, "Current platform does not support threads_per_warp to be 32"  # noqa: E501