PR #20340: Fix missing template value

Imported from GitHub PR #20340 Fixes a bug introduced in this change: google/tsl#2944 The change makes use of a template variable `%{compiler}`, that is not defined for this file. This causes the `-fno-canonical-system-headers` option to be set for Clang builds, and Clang will fail with an error about that command line flag not being defined. Copybara import of the project: -- 75a3d3f by Charles Hofer <[email protected]>: Fix missing template value -- e08537b by Charles Hofer <[email protected]>: Change flag to compiler_is_clang -- 373f359 by Charles Hofer <[email protected]>: Fix typo -- 2be3c30 by Harsha HS <[email protected]>: [ROCm] Add cuda-only tags for nvidia profiler test Merging this change closes #20340 FUTURE_COPYBARA_INTEGRATE_REVIEW=#20340 from ROCm:fix-missing-template-value 2be3c30 PiperOrigin-RevId: 713547337
openxla · Jan 9, 2025 · a2c93e8 · a2c93e8
1 parent b9ac712
commit a2c93e8
Showing 1 changed file with 5 additions and 2 deletions.
diff --git a/xla/backends/profiler/gpu/BUILD b/xla/backends/profiler/gpu/BUILD
@@ -435,7 +435,7 @@ cuda_library(
         "ptxas-options=-v",
     ]),
     local_defines = if_oss(["NVTX_VERSION_3_1=1"]),
-    tags = ["requires-gpu-nvidia"],
+    tags = ["cuda-only"],
     visibility = ["//visibility:public"],
 )
 
@@ -445,7 +445,10 @@ xla_test(
     srcs = ["nvtx_with_cuda_kernels_test.cc"],
     backends = ["gpu"],
     copts = tf_profiler_copts() + tsl_copts(),
-    tags = ["no_mac"],
+    tags = [
+        "cuda-only",
+        "no_mac",
+    ],
     deps = [
         ":nvtx_with_cuda_kernels",
         "@com_google_googletest//:gtest_main",