Misc fixes on test configs

- add missing test mark
- disable xla preallocate memory

Signed-off-by: Jay Gu <jagu@nvidia.com>

Author: haijieg

pytest.ini

@@ -10,3 +10,4 @@ markers =
     use_mlir: mark tests that depend on mlir which requires the "internal" extension
 env =
     CUDA_TILE_COMPILER_TIMEOUT_SEC=60
+    XLA_PYTHON_CLIENT_PREALLOCATE=false

test/conftest.py

@@ -12,7 +12,10 @@
 from functools import cache
 
 from cuda.tile._bytecode.version import BytecodeVersion
-from cuda.tile._compile import _get_max_supported_bytecode_version, _SUPPORTED_VERSIONS
+from cuda.tile._compile import (
+        _get_max_supported_bytecode_version,
+        _SUPPORTED_VERSIONS,
+        _find_compiler_bin)
 from cuda.tile._cext import dev_features_enabled
 from util import require_blackwell_or_newer, require_hopper_or_newer
 
@@ -39,6 +42,16 @@ def strict_importorskip(modname, *args, **kwargs):
         pytest.importorskip = strict_importorskip
 
 
+def pytest_sessionstart(session):
+    """
+    Called after the Session object has been created and
+    before performing collection and entering the run test loop.
+    """
+    print("Tile compiler path:", _find_compiler_bin().path)
+    print("Dev features enabled:", dev_features_enabled())
+    print("Bytecode version:", get_tileiras_version().as_string())
+
+
 @cache
 def get_tileiras_version():
     return _get_max_supported_bytecode_version(tempfile.gettempdir(),

test/test_token_order.py

@@ -1402,6 +1402,7 @@ def compile_kernel(self, kernel):
         Y = torch.arange(tile_size, device="cuda", dtype=torch.int32)
         return get_bytecode(kernel, (X, Y, tile_size))
 
+    @requires_tileiras(BytecodeVersion.V_13_3)
     @pytest.mark.parametrize("kernel, check_directive", make_cases(
         (tv_atomic, TVAtomicCheckDirective),
     ))