Merge branch 'fix-blockfreq-unroll-no-runtime' into fix-blockfreq-unroll-epilogue

Author: jdenny-ornl

.ci/all_requirements.txt

@@ -194,6 +194,10 @@ ml-dtypes==0.5.1 ; python_version < "3.13" \
     --hash=sha256:d13755f8e8445b3870114e5b6240facaa7cb0c3361e54beba3e07fa912a6e12b \
     --hash=sha256:fd918d4e6a4e0c110e2e05be7a7814d10dc1b95872accbf6512b80a109b71ae1
     # via -r mlir/python/requirements.txt
+nanobind==2.9.2 \
+    --hash=sha256:c37957ffd5eac7eda349cff3622ecd32e5ee1244ecc912c99b5bc8188bafd16e \
+    --hash=sha256:e7608472de99d375759814cab3e2c94aba3f9ec80e62cfef8ced495ca5c27d6e
+    # via -r mlir/python/requirements.txt
 numpy==2.0.2 \
     --hash=sha256:0123ffdaa88fa4ab64835dcbde75dcdf89c453c922f18dced6e27c90d1d0ec5a \
     --hash=sha256:11a76c372d1d37437857280aa142086476136a8c0f373b2e648ab2c8f18fb195 \
@@ -295,6 +299,10 @@ pyasn1-modules==0.4.2 \
     --hash=sha256:29253a9207ce32b64c3ac6600edc75368f98473906e8fd1043bd6b5b1de2c14a \
     --hash=sha256:677091de870a80aae844b1ca6134f54652fa2c8c5a52aa396440ac3106e941e6
     # via google-auth
+pybind11==2.13.6 \
+    --hash=sha256:237c41e29157b962835d356b370ededd57594a26d5894a795960f0047cb5caf5 \
+    --hash=sha256:ba6af10348c12b24e92fa086b39cfba0eff619b61ac77c406167d813b096d39a
+    # via -r mlir/python/requirements.txt
 pyyaml==6.0.1 \
     --hash=sha256:04ac92ad1925b2cff1db0cfebffb6ffc43457495c9b3c39d3fcae417d7125dc5 \
     --hash=sha256:062582fca9fabdd2c8b54a3ef1c978d786e0f6b3a1510e0ac93ef59e0ddae2bc \

.ci/cache_lit_timing_files.py

@@ -17,6 +17,7 @@
 import glob
 
 from google.cloud import storage
+from google.api_core import exceptions
 
 GCS_PARALLELISM = 100
 
@@ -50,7 +51,14 @@ def _maybe_download_timing_file(blob):
 
 def download_timing_files(storage_client, bucket_name: str):
     bucket = storage_client.bucket(bucket_name)
-    blobs = bucket.list_blobs(prefix="lit_timing")
+    try:
+        blobs = bucket.list_blobs(prefix="lit_timing")
+    except exceptions.ClientError as client_error:
+        print(
+            "::warning file=cache_lit_timing_files.py::Failed to list blobs "
+            "in bucket."
+        )
+        sys.exit(0)
     with multiprocessing.pool.ThreadPool(GCS_PARALLELISM) as thread_pool:
         futures = []
         for timing_file_blob in blobs:
@@ -60,7 +68,13 @@ def download_timing_files(storage_client, bucket_name: str):
                 )
             )
         for future in futures:
-            future.get()
+            future.wait()
+            if not future.successful():
+                print(
+                    "::warning file=cache_lit_timing_files.py::Failed to "
+                    "download lit timing file."
+                )
+                continue
     print("Done downloading")
 
 

.ci/generate_test_report_github.py

@@ -8,10 +8,15 @@
 
 import generate_test_report_lib
 
-PLATFORM_TITLES = {
-    "Windows": ":window: Windows x64 Test Results",
-    "Linux": ":penguin: Linux x64 Test Results",
-}
+def compute_platform_title() -> str:
+    logo = ":window:" if platform.system() == "Windows" else ":penguin:"
+    # On Linux the machine value is x86_64 on Windows it is AMD64.
+    if platform.machine() == "x86_64" or platform.machine() == "AMD64":
+        arch = "x64"
+    else:
+        arch = platform.machine()
+    return f"{logo} {platform.system()} {arch} Test Results"
+
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
@@ -22,7 +27,7 @@
     args = parser.parse_args()
 
     report = generate_test_report_lib.generate_report_from_files(
-        PLATFORM_TITLES[platform.system()], args.return_code, args.build_test_logs
+        compute_platform_title(), args.return_code, args.build_test_logs
     )
 
     print(report)

.ci/utils.sh

@@ -56,6 +56,7 @@ function start-group {
 export PIP_BREAK_SYSTEM_PACKAGES=1
 pip install -q -r "${MONOREPO_ROOT}"/.ci/all_requirements.txt
 
-if [[ "$GITHUB_ACTIONS" != "" ]]; then
+# The ARM64 builders run on AWS and don't have access to the GCS cache.
+if [[ "$GITHUB_ACTIONS" != "" ]] && [[ "$RUNNER_ARCH" != "ARM64" ]]; then
   python .ci/cache_lit_timing_files.py download
 fi

.github/new-prs-labeler.yml

@@ -722,7 +722,12 @@ mlgo:
   - llvm/include/llvm/Analysis/IR2Vec.h
   - llvm/lib/Analysis/IR2Vec.cpp
   - llvm/lib/Analysis/models/**
+  - llvm/include/llvm/CodeGen/MIR2Vec.h
+  - llvm/lib/CodeGen/MIR2Vec.cpp
   - llvm/test/Analysis/IR2Vec/**
+  - llvm/test/CodeGen/MIR2Vec/**
+  - llvm/unittests/Analysis/IR2VecTest.cpp
+  - llvm/unittests/CodeGen/MIR2VecTest.cpp
   - llvm/tools/llvm-ir2vec/**
   - llvm/docs/CommandGuide/llvm-ir2vec.rst
 
@@ -1091,8 +1096,8 @@ clang:openmp:
   - llvm/test/Transforms/OpenMP/**
 
 clang:temporal-safety:
-  - clang/include/clang/Analysis/Analyses/LifetimeSafety*
-  - clang/lib/Analysis/LifetimeSafety*
+  - clang/include/clang/Analysis/Analyses/LifetimeSafety/**
+  - clang/lib/Analysis/LifetimeSafety/**
   - clang/unittests/Analysis/LifetimeSafety*
   - clang/test/Sema/*lifetime-safety*
   - clang/test/Sema/*lifetime-analysis*

.github/workflows/build-ci-container-tooling.yml

@@ -1,4 +1,4 @@
-name: Build CI Container
+name: Build CI Tooling Containers
 
 permissions:
   contents: read
@@ -72,7 +72,7 @@ jobs:
       - name: Test Container
         run: |
           # Use --pull=never to ensure we are testing the just built image.
-          podman run --pull=never --rm -it ${{ steps.vars.outputs.container-name-format-tag }} /usr/bin/bash -x -c 'cd $HOME && clang-format --version | grep version && black --version | grep black'
+          podman run --pull=never --rm -it ${{ steps.vars.outputs.container-name-format-tag }} /usr/bin/bash -x -c 'cd $HOME && clang-format --version | grep version && git-clang-format -h | grep usage && black --version | grep black'
           podman run --pull=never --rm -it ${{ steps.vars.outputs.container-name-lint-tag }} /usr/bin/bash -x -c 'cd $HOME && clang-tidy --version | grep version && clang-tidy-diff.py -h | grep usage'
 
   push-ci-container:
@@ -101,7 +101,7 @@ jobs:
           }
 
           podman login -u ${{ github.actor }} -p $GITHUB_TOKEN ghcr.io
-          for f in $(find . -iname *.tar); do
+          for f in $(find . -iname '*.tar'); do
             image_name=$(podman load -q -i $f | sed 's/Loaded image: //g')
             push_container $image_name
 

.github/workflows/build-ci-container.yml

@@ -103,7 +103,7 @@ jobs:
           }
 
           podman login -u ${{ github.actor }} -p $GITHUB_TOKEN ghcr.io
-          for f in $(find . -iname *.tar); do
+          for f in $(find . -iname '*.tar'); do
             image_name=$(podman load -q -i $f | sed 's/Loaded image: //g')
             push_container $image_name
 

.github/workflows/containers/github-action-ci-tooling/Dockerfile

@@ -10,7 +10,8 @@ RUN apt-get update && \
     tar -xvJf llvm.tar.xz -C /llvm-extract \
         # Only unpack these tools to save space on Github runner.
         LLVM-${LLVM_VERSION}-Linux-X64/bin/clang-tidy \
-        LLVM-${LLVM_VERSION}-Linux-X64/bin/clang-format && \
+        LLVM-${LLVM_VERSION}-Linux-X64/bin/clang-format \
+        LLVM-${LLVM_VERSION}-Linux-X64/bin/git-clang-format && \
     rm llvm.tar.xz
 
 
@@ -35,7 +36,9 @@ RUN apt-get update && \
 FROM base AS ci-container-code-format
 ARG LLVM_VERSION
 
-COPY --from=llvm-downloader /llvm-extract/LLVM-${LLVM_VERSION}-Linux-X64/bin/clang-format ${LLVM_SYSROOT}/bin/clang-format
+COPY --from=llvm-downloader /llvm-extract/LLVM-${LLVM_VERSION}-Linux-X64/bin/clang-format \
+                            /llvm-extract/LLVM-${LLVM_VERSION}-Linux-X64/bin/git-clang-format \
+                            ${LLVM_SYSROOT}/bin/
 
 ENV PATH=${LLVM_SYSROOT}/bin:${PATH}
 

.github/workflows/containers/github-action-ci/Dockerfile

@@ -62,7 +62,6 @@ RUN apt-get update && \
     # Having a symlink from python to python3 enables code sharing between
     # the Linux and Windows pipelines.
     python3-pip \
-    python3-venv \
     file \
     tzdata \
     python-is-python3 && \

.github/workflows/premerge.yaml

@@ -24,17 +24,45 @@ concurrency:
 
 jobs:
   premerge-checks-linux:
-    name: Build and Test Linux
+    name: Build and Test Linux${{ (startsWith(matrix.runs-on, 'depot-ubuntu-24.04-arm') && ' AArch64') || '' }}
     if: >-
         github.repository_owner == 'llvm' &&
         (github.event_name != 'pull_request' || github.event.action != 'closed')
-    runs-on: llvm-premerge-linux-runners
+    strategy:
+      fail-fast: false
+      matrix:
+        runs-on:
+          - depot-ubuntu-24.04-arm-16
+          - llvm-premerge-linux-runners
+    runs-on: ${{ matrix.runs-on }}
+    container:
+      # The llvm-premerge agents are already containers and running the
+      # this same image, so we can't use a container for the github action
+      # job.  The depot containers are running on VMs, so we can use a
+      # container.  This helps ensure the build environment is as close
+      # as possible on both the depot runners and the llvm-premerge runners.
+      image: ${{ (startsWith(matrix.runs-on, 'depot-ubuntu-24.04-arm') && format('ghcr.io/{0}/arm64v8/ci-ubuntu-24.04',github.repository_owner) ) || null }}
+      # --privileged is needed to run the lldb tests that disable aslr.
+      # The SCCACHE environment variables are need to be copied from the host
+      # to the container to make sure it is configured correctly to use the
+      # depot cache.
+      options: >-
+         --privileged
+         --env SCCACHE_WEBDAV_ENDPOINT
+         --env SCCACHE_WEBDAV_TOKEN
+    defaults:
+      run:
+        # The run step defaults to using sh as the shell when running in a
+        # container, so make bash the default to ensure consistency between
+        # container and non-container jobs.
+        shell: bash
     steps:
       - name: Checkout LLVM
         uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
         with:
           fetch-depth: 2
       - name: Build and Test
+        continue-on-error: ${{ runner.arch == 'ARM64' }}
         run: |
           git config --global --add safe.directory '*'
 
@@ -54,11 +82,16 @@ jobs:
           export CC=/opt/llvm/bin/clang
           export CXX=/opt/llvm/bin/clang++
 
-          # This environment variable is passes into the container through the
-          # runner pod definition. This differs between our two clusters which
-          # why we do not hardcode it.
-          export SCCACHE_GCS_BUCKET=$CACHE_GCS_BUCKET
-          export SCCACHE_GCS_RW_MODE=READ_WRITE
+          # The linux-premerge runners are hosted on GCP and have a different
+          # cache setup than the depot runners.
+          if [[ "${{ matrix.runs-on }}" = "llvm-premerge-linux-runners" ]]; then
+            # This environment variable is passes into the container through the
+            # runner pod definition. This differs between our two clusters which
+            # why we do not hardcode it.
+            export SCCACHE_GCS_BUCKET=$CACHE_GCS_BUCKET
+            export SCCACHE_GCS_RW_MODE=READ_WRITE
+          fi
+          env
 
           # Set the idle timeout to zero to ensure sccache runs for the
           # entire duration of the job. Otherwise it might stop if we run
@@ -78,7 +111,7 @@ jobs:
         if: '!cancelled()'
         uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
         with:
-          name: Premerge Artifacts (Linux)
+          name: Premerge Artifacts (Linux ${{ runner.arch }})
           path: artifacts/
           retention-days: 5
           include-hidden-files: 'true'