[TEST] Fix test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma (#4402)

There are 6 failed, 18 passed for
`test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma`.
The failures are cause by triton.runtime.errors.OutOfResources: out of
resource: shared memory, Required: 196608, Hardware limit: 131072.

This change checks and marks out of memory cases as XFAIL.

Author: LiyangLingIntel

python/test/unit/language/test_tensor_descriptor.py

@@ -1270,6 +1270,12 @@ def test_mxfp8_mxfp4_matmul_tma(M, N, K, BLOCK_M, BLOCK_N, BLOCK_K, NUM_STAGES,
     if BLOCK_K < K and is_cuda() and torch.cuda.get_device_capability(0)[0] != 10:
         pytest.skip("Currently broken on hopper")
 
+    required_sm = BLOCK_M * BLOCK_K * 2 + BLOCK_K * BLOCK_N * 2
+    max_sm = triton.runtime.driver.active.utils.get_device_properties(
+        triton.runtime.driver.active.get_current_device())["max_shared_mem"]
+    if is_xpu() and required_sm > max_sm:
+        pytest.xfail(f"Not enough shared memory for the given block size ({BLOCK_M}, {BLOCK_N}, {BLOCK_K})")
+
     a = torch.randint(20, 40, (M, K), dtype=torch.uint8).view(torch.float8_e5m2).to(device)
 
     dtype_src_str = "float8e5"

scripts/skiplist/default/language.txt

@@ -1,13 +1,2 @@
 # https://github.com/intel/intel-xpu-backend-for-triton/issues/3870
 python/test/unit/language/test_core.py::test_dot3d[8-2-64-64-64-32-32-float32-float32]
-# https://github.com/intel/intel-xpu-backend-for-triton/issues/4222
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[1-128-128-256-8192-8192-8192]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[1-128-256-128-8192-8192-8192]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[1-128-256-256-1024-512-256]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[1-128-256-256-128-256-256]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[1-128-256-256-8192-8192-8192]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[3-128-128-256-8192-8192-8192]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[3-128-256-128-8192-8192-8192]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[3-128-256-256-1024-512-256]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[3-128-256-256-128-256-256]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[3-128-256-256-8192-8192-8192]

scripts/skiplist/lts/language.txt

@@ -369,15 +369,5 @@ python/test/unit/language/test_pipeliner.py::test_indirect_matmul[5-128-128-64]
 python/test/unit/language/test_pipeliner.py::test_indirect_matmul[5-128-64-128]
 python/test/unit/language/test_core.py::test_convert_mma2mma[mma_pair0-float16-256-256]
 python/test/unit/language/test_matmul.py::test_lhs_in_tmem
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[1-128-128-256-8192-8192-8192]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[1-128-256-128-8192-8192-8192]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[1-128-256-256-1024-512-256]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[1-128-256-256-128-256-256]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[1-128-256-256-8192-8192-8192]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[3-128-128-256-8192-8192-8192]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[3-128-256-128-8192-8192-8192]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[3-128-256-256-1024-512-256]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[3-128-256-256-128-256-256]
-python/test/unit/language/test_tensor_descriptor.py::test_mxfp8_mxfp4_matmul_tma[3-128-256-256-8192-8192-8192]
 python/test/unit/language/test_tensor_descriptor.py::test_tensor_descriptor_batched_gemm_2d_tma
 python/test/unit/language/test_tensor_descriptor.py::test_tensor_descriptor_batched_gemm_3d_tma