[Mlir-commits] [mlir] [mlir][test][gpu] Migrate CUDA tests to the TargetAttr compilation wo… (PR #65768)

[Fabian Mora]

================
@@ -212,8 +212,18 @@ def have_host_jit_feature_support(feature_name):
 if have_host_jit_feature_support("jit"):
     config.available_features.add("host-supports-jit")
 
+# Find if the host has the required CUDA toolkit tools to run the integration tests.
+# This tools are required for compiling GPU modules into fatbins.
+def have_host_required_cuda_tools():
+    return lit.util.which("ptxas") != None and lit.util.which("fatbinary") != None
+
 if config.run_cuda_tests:
     config.available_features.add("host-supports-nvptx")
+    if have_host_required_cuda_tools():
+        config.available_features.add("host-supports-cuda-runner")
+        config.enable_cuda_runner_tests = True
+    else:
+        config.enable_cuda_runner_tests = False
----------------
fabianmcg wrote:

Changed.

https://github.com/llvm/llvm-project/pull/65768