Introduce rocm-only tag and remove if_rocm_is_configured

This replaces all `if_rocm_is_configured` guards in `stream_executor/rocm/...` with a filtering tag `rocm-only`. The CUDA build on the CI gets adjusted to skip those targets.

This uncovered some additional problems that get fixed as well:

- A wrong library name for the hipfft library in the Bazel CUDA configuration
- A wrong test case in the RocmVersionParser test that so far has not been running anywhere.
- Missing tags for the platform alias targets in `stream_executor/BUILD`

PiperOrigin-RevId: 678548961

.github/workflows/bazel_dependency_violations.yml

@@ -29,7 +29,7 @@ jobs:
   dependency-violations:
     strategy:
       matrix:
-        tag: [gpu, cuda-only]
+        tag: [gpu, cuda-only, rocm-only]
     name: no-${{ matrix.tag }}-targets-in-cpu-build
     runs-on: ubuntu-22.04
     defaults:

build_tools/ci/build.py

@@ -213,9 +213,9 @@ def nvidia_gpu_build_with_compute_capability(
       image_url=_DEFAULT_IMAGE,
       target_patterns=_XLA_DEFAULT_TARGET_PATTERNS,
       configs=configs,
-      test_tag_filters=("-no_oss", "requires-gpu-nvidia", "gpu")
+      test_tag_filters=("-no_oss", "requires-gpu-nvidia", "gpu", "-rocm-only")
       + extra_gpu_tags,
-      build_tag_filters=("-no_oss", "requires-gpu-nvidia", "gpu"),
+      build_tag_filters=("-no_oss", "requires-gpu-nvidia", "gpu", "-rocm-only"),
       options={
           "run_under": "//tools/ci_build/gpu_build:parallel_gpu_execute",
           "repo_env": f"TF_CUDA_COMPUTE_CAPABILITIES={compute_capability/10}",

build_tools/ci/golden_commands.txt

@@ -19,7 +19,7 @@ $KOKORO_ARTIFACTS_DIR/github/xla/.kokoro/generate_index_html.sh index.html
 nvidia-smi
 parallel --ungroup --retries 3 --delay 15 docker pull ::: gcr.io/tensorflow-sigs/build:latest-python3.11
 docker run --detach --name=xla_ci --rm --interactive --tty --volume=./github:/github --workdir=/github/xla gcr.io/tensorflow-sigs/build:latest-python3.11 bash
-docker exec xla_ci bazel test --build_tag_filters=-no_oss,requires-gpu-nvidia,gpu --test_tag_filters=-no_oss,requires-gpu-nvidia,gpu,requires-gpu-sm75-only,requires-gpu-sm60,requires-gpu-sm70,-requires-gpu-sm80,-requires-gpu-sm80-only,-requires-gpu-sm90,-requires-gpu-sm90-only,-requires-gpu-amd --config=warnings --config=rbe_linux_cuda_nvcc --run_under=//tools/ci_build/gpu_build:parallel_gpu_execute --repo_env=TF_CUDA_COMPUTE_CAPABILITIES=7.5 --@cuda_driver//:enable_forward_compatibility=true --test_output=errors --verbose_failures --keep_going --nobuild_tests_only --profile=profile.json.gz --flaky_test_attempts=3 --jobs=150 --bes_upload_mode=fully_async -- //xla/... //build_tools/... @tsl//tsl/...
+docker exec xla_ci bazel test --build_tag_filters=-no_oss,requires-gpu-nvidia,gpu,-rocm-only --test_tag_filters=-no_oss,requires-gpu-nvidia,gpu,-rocm-only,requires-gpu-sm75-only,requires-gpu-sm60,requires-gpu-sm70,-requires-gpu-sm80,-requires-gpu-sm80-only,-requires-gpu-sm90,-requires-gpu-sm90-only,-requires-gpu-amd --config=warnings --config=rbe_linux_cuda_nvcc --run_under=//tools/ci_build/gpu_build:parallel_gpu_execute --repo_env=TF_CUDA_COMPUTE_CAPABILITIES=7.5 --@cuda_driver//:enable_forward_compatibility=true --test_output=errors --verbose_failures --keep_going --nobuild_tests_only --profile=profile.json.gz --flaky_test_attempts=3 --jobs=150 --bes_upload_mode=fully_async -- //xla/... //build_tools/... @tsl//tsl/...
 docker exec xla_ci bazel analyze-profile profile.json.gz
 docker stop xla_ci
 # END BuildType.GPU

build_tools/dependencies/aspects.bzl

@@ -80,3 +80,11 @@ validate_cuda_only_tag = aspect(
     implementation = _cuda_only_tag_violation_aspect_impl,
     attr_aspects = ["deps"],
 )
+
+def _rocm_only_tag_violation_aspect_impl(target, ctx):
+    return _dependency_violation_aspect_impl(target, ctx, "rocm-only")
+
+validate_rocm_only_tag = aspect(
+    implementation = _rocm_only_tag_violation_aspect_impl,
+    attr_aspects = ["deps"],
+)

build_tools/lint/tags.py

@@ -65,6 +65,7 @@
     "gpu": "Catch-all tag for targets that should be built/tested on GPU CI",
     "cpu": "Catch-all tag for targets that should be built/tested on CPU CI.",
     "cuda-only": "Targets that require the CUDA backend to be enabled.",
+    "rocm-only": "Targets that require the ROCm backend to be enabled.",
     # Below tags are generated by `xla_test`.
     "broken": "Test will be marked with other tags to disable in `xla_test`.",
     "xla_interpreter": "Uses interpreter backend.",

third_party/tsl/third_party/gpus/rocm_configure.bzl

@@ -476,7 +476,7 @@ def _create_dummy_repository(repository_ctx):
             "%{hipblas_lib}": _lib_name("hipblas"),
             "%{miopen_lib}": _lib_name("miopen"),
             "%{rccl_lib}": _lib_name("rccl"),
-            "%{hipfft_or_rocfft}": _lib_name("hipfft"),
+            "%{hipfft_or_rocfft}": "hipfft",
             "%{hipfft_or_rocfft_lib}": _lib_name("hipfft"),
             "%{hiprand_lib}": _lib_name("hiprand"),
             "%{hipsparse_lib}": _lib_name("hipsparse"),

xla/backends/profiler/gpu/BUILD

@@ -26,14 +26,6 @@ tsl_gpu_library(
     name = "device_tracer",
     srcs = tf_additional_device_tracer_srcs(),
     copts = tf_profiler_copts() + tsl_copts(),
-    cuda_deps = [
-        ":cupti_buffer_events",
-        ":cupti_collector",
-        ":cupti_tracer",
-        ":cupti_wrapper",
-        ":rocm_collector",
-        ":rocm_tracer",
-    ],
     deps = [
         ":cupti_utils",
         "//xla/tsl/util:env_var",
@@ -47,7 +39,17 @@ tsl_gpu_library(
         "@tsl//tsl/profiler/lib:profiler_interface",
         "@tsl//tsl/profiler/protobuf:xplane_proto_cc",
         "@tsl//tsl/profiler/utils:time_utils",
-    ],
+    ] + if_cuda([
+        # keep sorted
+        ":cupti_buffer_events",
+        ":cupti_collector",
+        ":cupti_tracer",
+        ":cupti_wrapper",
+    ]) + if_rocm([
+        # keep sorted
+        ":rocm_collector",
+        ":rocm_tracer",
+    ]),
     alwayslink = 1,
 )
 
@@ -218,6 +220,10 @@ tsl_gpu_library(
     srcs = if_rocm(["rocm_collector.cc"]),
     hdrs = if_rocm(["rocm_collector.h"]),
     copts = tf_profiler_copts() + tsl_copts(),
+    tags = [
+        "gpu",
+        "rocm-only",
+    ],
     visibility = ["//visibility:public"],
     deps = [
         "//xla/stream_executor/rocm:roctracer_wrapper",
@@ -253,6 +259,10 @@ tsl_gpu_library(
     srcs = if_rocm(["rocm_tracer.cc"]),
     hdrs = if_rocm(["rocm_tracer.h"]),
     copts = tf_profiler_copts() + tsl_copts(),
+    tags = [
+        "gpu",
+        "rocm-only",
+    ],
     visibility = ["//visibility:public"],
     deps = [
         ":rocm_collector",

xla/stream_executor/BUILD

@@ -925,14 +925,25 @@ xla_cc_test(
 alias(
     name = "cuda_platform",
     actual = "//xla/stream_executor/cuda:all_runtime",
+    tags = [
+        "cuda-only",
+        "gpu",
+    ],
 )
 
 alias(
     name = "rocm_platform",
     actual = "//xla/stream_executor/rocm:all_runtime",
+    tags = [
+        "gpu",
+        "rocm-only",
+    ],
 )
 
 alias(
     name = "sycl_platform",
     actual = "//xla/stream_executor/sycl:all_runtime",
+    tags = [
+        "gpu",
+    ],
 )