[TEST] Only check_threads_supported on XPU (#3736)

7d24ef4
accidentally reduced the pass rate on interpreter mode, as
`get_current_device` returns xpu.
Pass rate: 89.76% -> 89.99%

Signed-off-by: Whitney Tsang <[email protected]>

Author: whitneywhtsang

python/test/unit/language/test_core.py

@@ -137,9 +137,10 @@ def check_type_supported(dtype, device):
             pytest.xfail("float64 not supported on current xpu hardware")
 
 
-def check_threads_supported(num_warps, threads_per_warp):
-    device = triton.runtime.driver.active.get_current_device()
-    props = triton.runtime.driver.active.utils.get_device_properties(device)
+def check_threads_supported(num_warps, threads_per_warp, device):
+    if device != "xpu":
+        return
+    props = triton.runtime.driver.active.utils.get_device_properties(triton.runtime.driver.active.get_current_device())
     if threads_per_warp not in props['sub_group_sizes']:
         pytest.xfail('unsupported warp size')
     if threads_per_warp * num_warps > props['max_work_group_size']:
@@ -2366,7 +2367,7 @@ def get_reduced_dtype(dtype_str, op):
                          [(64, 16), (4, THREADS_PER_WARP)] if is_xpu() else [(4, THREADS_PER_WARP)])
 def test_reduce1d(op, dtype_str, shape, num_ctas, num_warps, threads_per_warp, device):
     check_type_supported(dtype_str, device)  # bfloat16 on cc < 80 will not be tested
-    check_threads_supported(num_warps, threads_per_warp)
+    check_threads_supported(num_warps, threads_per_warp, device)
 
     # triton kernel
     @triton.jit
@@ -2475,7 +2476,7 @@ def kernel(X, Z, BLOCK: tl.constexpr):
                          [(64, 16), (4, THREADS_PER_WARP)] if is_xpu() else [(4, THREADS_PER_WARP)])
 def test_reduce(op, dtype_str, shape, axis, keep_dims, num_ctas, num_warps, threads_per_warp, device):
     check_type_supported(dtype_str, device)  # bfloat16 on cc < 80 will not be tested
-    check_threads_supported(num_warps, threads_per_warp)
+    check_threads_supported(num_warps, threads_per_warp, device)
 
     @triton.jit
     def kernel(X, Z, BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr, BLOCK_K: tl.constexpr, IS_3D: tl.constexpr,