Merge branch 'main' into feature/da-monotonic

Author: kasuga-fj

.ci/all_requirements.txt

@@ -0,0 +1,85 @@
+# Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+# See https://llvm.org/LICENSE.txt for license information.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+"""Caches .lit_test_times.txt files between premerge invocations.
+
+.lit_test_times.txt files are used by lit to order tests to best take advantage
+of parallelism. Having them around and up to date can result in a ~15%
+improvement in test times. This script downloading cached test time files and
+uploading new versions to the GCS buckets used for caching.
+"""
+
+import sys
+import os
+import logging
+import multiprocessing.pool
+import pathlib
+import platform
+import glob
+
+from google.cloud import storage
+
+GCS_PARALLELISM = 100
+
+
+def _get_blob_prefix():
+    return f"lit_timing_{platform.system().lower()}"
+
+
+def _maybe_upload_timing_file(bucket, timing_file_path):
+    if os.path.exists(timing_file_path):
+        timing_file_blob = bucket.blob(_get_blob_prefix() + "/" + timing_file_path)
+        timing_file_blob.upload_from_filename(timing_file_path)
+
+
+def upload_timing_files(storage_client, bucket_name: str):
+    bucket = storage_client.bucket(bucket_name)
+    with multiprocessing.pool.ThreadPool(GCS_PARALLELISM) as thread_pool:
+        futures = []
+        for timing_file_path in glob.glob("**/.lit_test_times.txt", recursive=True):
+            futures.append(
+                thread_pool.apply_async(
+                    _maybe_upload_timing_file, (bucket, timing_file_path)
+                )
+            )
+        for future in futures:
+            future.get()
+    print("Done uploading")
+
+
+def _maybe_download_timing_file(blob):
+    file_name = blob.name.removeprefix(_get_blob_prefix() + "/")
+    pathlib.Path(os.path.dirname(file_name)).mkdir(parents=True, exist_ok=True)
+    blob.download_to_filename(file_name)
+
+
+def download_timing_files(storage_client, bucket_name: str):
+    bucket = storage_client.bucket(bucket_name)
+    blobs = bucket.list_blobs(prefix=_get_blob_prefix())
+    with multiprocessing.pool.ThreadPool(GCS_PARALLELISM) as thread_pool:
+        futures = []
+        for timing_file_blob in blobs:
+            futures.append(
+                thread_pool.apply_async(
+                    _maybe_download_timing_file, (timing_file_blob,)
+                )
+            )
+        for future in futures:
+            future.get()
+    print("Done downloading")
+
+
+if __name__ == "__main__":
+    if len(sys.argv) != 2:
+        logging.fatal("Expected usage is cache_lit_timing_files.py <upload/download>")
+        sys.exit(1)
+    action = sys.argv[1]
+    storage_client = storage.Client()
+    bucket_name = os.environ["CACHE_GCS_BUCKET"]
+    if action == "download":
+        download_timing_files(storage_client, bucket_name)
+    elif action == "upload":
+        upload_timing_files(storage_client, bucket_name)
+    else:
+        logging.fatal("Expected usage is cache_lit_timing_files.py <upload/download>")
+        sys.exit(1)

.ci/cache_lit_timing_files.py

@@ -333,6 +333,7 @@ def get_env_variables(modified_files: list[str], platform: str) -> Set[str]:
     current_platform = platform.system()
     if len(sys.argv) == 2:
         current_platform = sys.argv[1]
-    env_variables = get_env_variables(sys.stdin.readlines(), current_platform)
+    changed_files = [line.strip() for line in sys.stdin.readlines()]
+    env_variables = get_env_variables(changed_files, current_platform)
     for env_variable in env_variables:
         print(f"{env_variable}='{env_variables[env_variable]}'")

.ci/compute_projects.py

@@ -27,6 +27,13 @@ def _parse_ninja_log(ninja_log: list[str]) -> list[tuple[str, str]]:
             # We hit the end of the log without finding a build failure, go to
             # the next log.
             return failures
+        # If we are doing a build with LLVM_ENABLE_RUNTIMES, we can have nested
+        # ninja invocations. The sub-ninja will print that a subcommand failed,
+        # and then the outer ninja will list the command that failed. We should
+        # ignore the outer failure.
+        if ninja_log[index - 1].startswith("ninja: build stopped:"):
+            index += 1
+            continue
         # We are trying to parse cases like the following:
         #
         # [4/5] test/4.stamp

.ci/generate_test_report_lib.py

@@ -126,6 +126,39 @@ def test_ninja_log_multiple_failures(self):
             ),
         )
 
+    # Test that we can correctly handle the runtimes build. the LLVM runtimes
+    # build will involve ninja invoking more ninja processes within the
+    # runtimes directory. This means that we see two failures for a failure in
+    # the runtimes build: one from the inner ninja containing the actual action
+    # that failed, and one for the sub ninja invocation that failed.
+    def test_ninja_log_runtimes_failure(self):
+        failures = generate_test_report_lib.find_failure_in_ninja_logs(
+            [
+                [
+                    "[1/5] test/1.stamp",
+                    "[2/5] test/2.stamp",
+                    "FAILED: touch test/2.stamp",
+                    "Wow! This system is really broken!",
+                    "ninja: build stopped: subcommand failed.",
+                    "FAILED: running check-runtime failed.",
+                    "<some random command>",
+                    "ninja: build stopped: subcommand failed.",
+                ]
+            ]
+        )
+        self.assertEqual(len(failures), 1)
+        self.assertEqual(
+            failures[0],
+            (
+                "test/2.stamp",
+                dedent(
+                    """\
+                    FAILED: touch test/2.stamp
+                    Wow! This system is really broken!"""
+                ),
+            ),
+        )
+
     def test_title_only(self):
         self.assertEqual(
             generate_test_report_lib.generate_report("Foo", 0, [], []),

.ci/generate_test_report_lib_test.py

@@ -70,6 +70,7 @@
 # by trial and error).
 GRAFANA_METRIC_MAX_AGE_MN = 120
 
+
 @dataclass
 class JobMetrics:
     job_name: str
@@ -243,6 +244,7 @@ def clean_up_libcxx_job_name(old_name: str) -> str:
     new_name = stage + "_" + remainder
     return new_name
 
+
 def github_get_metrics(
     github_repo: github.Repository, last_workflows_seen_as_completed: set[int]
 ) -> tuple[list[JobMetrics], int]:
@@ -336,11 +338,29 @@ def github_get_metrics(
                 name_suffix = GITHUB_JOB_TO_TRACK[name_prefix][job.name]
             metric_name = name_prefix + "_" + name_suffix
 
+            ag_metric_name = None
+            if libcxx_testing:
+                job_key = None
+                if job.name.find("stage1") != -1:
+                    job_key = "stage1"
+                elif job.name.find("stage2") != -1:
+                    job_key = "stage2"
+                elif job.name.find("stage3") != -1:
+                    job_key = "stage3"
+                if job_key:
+                    ag_name = (
+                        name_prefix + "_" + GITHUB_JOB_TO_TRACK[name_prefix][job_key]
+                    )
+
             if task.status != "completed":
                 if job.status == "queued":
                     queued_count[metric_name] += 1
+                    if libcxx_testing:
+                        queued_count[ag_name] += 1
                 elif job.status == "in_progress":
                     running_count[metric_name] += 1
+                    if libcxx_testing:
+                        running_count[ag_name] += 1
                 continue
 
             job_result = int(job.conclusion == "success" or job.conclusion == "skipped")

.ci/metrics/metrics.py

@@ -409,5 +409,6 @@ def test_clean_up_libcxx_job_name(self):
         out_name4 = metrics.clean_up_libcxx_job_name(bad_name)
         self.assertEqual(out_name4, bad_name)
 
+
 if __name__ == "__main__":
     unittest.main()

.ci/metrics/metrics_test.py

@@ -1 +1,2 @@
 junitparser==3.2.0
+google-cloud-storage==3.3.0

.ci/requirements.txt

@@ -112,6 +112,9 @@
 /mlir/**/NVGPU*/ @grypp
 /mlir/test/**/CUDA/ @grypp
 
+# MLIR GPU Dialect
+/mlir/**/GPU*/ @fabianmcg
+
 # MLIR NVVM Dialect in MLIR
 /mlir/**/LLVMIR/**/BasicPtxBuilderInterface* @grypp
 /mlir/**/NVVM* @grypp

.github/CODEOWNERS

@@ -2,7 +2,7 @@ FROM docker.io/library/ubuntu:24.04 as base
 ENV LLVM_SYSROOT=/opt/llvm
 
 FROM base as stage1-toolchain
-ENV LLVM_VERSION=20.1.8
+ENV LLVM_VERSION=21.1.0
 
 RUN apt-get update && \
     apt-get install -y \
@@ -73,8 +73,8 @@ RUN apt-get update && \
 # caching), so we manually install it here.
 # TODO(boomanaiden154): We should return to installing this from the apt
 # repository once a version containing the necessary bug fixes is available.
-RUN curl -L 'https://github.com/mozilla/sccache/releases/download/v0.10.0/sccache-v0.10.0-x86_64-unknown-linux-musl.tar.gz' > /tmp/sccache.tar.gz && \
-    echo "1fbb35e135660d04a2d5e42b59c7874d39b3deb17de56330b25b713ec59f849b /tmp/sccache.tar.gz" | sha256sum -c && \
+RUN curl -L "https://github.com/mozilla/sccache/releases/download/v0.10.0/sccache-v0.10.0-$(arch)-unknown-linux-musl.tar.gz" > /tmp/sccache.tar.gz && \
+    echo $( [ $(arch) = 'x86_64' ] && echo "1fbb35e135660d04a2d5e42b59c7874d39b3deb17de56330b25b713ec59f849b" || echo "d6a1ce4acd02b937cd61bc675a8be029a60f7bc167594c33d75732bbc0a07400")  /tmp/sccache.tar.gz | sha256sum -c && \
     tar xzf /tmp/sccache.tar.gz -O --wildcards '*/sccache' > '/usr/local/bin/sccache' && \
     rm /tmp/sccache.tar.gz && \
     chmod +x /usr/local/bin/sccache