tc

.ci/pytorch/build.sh

@@ -29,6 +29,9 @@ env
 # https://github.com/pytorch/pytorch/blob/0b6c0898e6c352c8ea93daec854e704b41485375/.ci/docker/common/install_cache.sh#L97
 export PATH="/opt/cache/lib:$PATH"
 
+# Turn off -MD / -MMD compiler flags to increase sccache hit rate
+export COMPILE_NO_MD=1
+
 if [[ "$BUILD_ENVIRONMENT" == *cuda* ]]; then
   # Use jemalloc during compilation to mitigate https://github.com/pytorch/pytorch/issues/116289
   export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so.2
@@ -289,8 +292,15 @@ else
       python -mpip install numpy==2.0.2
 
       WERROR=1 python setup.py clean
+      sccache --stop-server
+      export SCCACHE_LOG_LEVEL=debug
+      export SCCACHE_ERROR_LOG=/tmp/sccache_errors.log
+      export SCCACHE_LOG=debug
+      export RUST_LOG=sccache::server=debug
+      sccache --start-server
 
       WERROR=1 python -m build --wheel --no-isolation
+      mv /tmp/sccache_errors.log dist/
     else
       python setup.py clean
       if [[ "$BUILD_ENVIRONMENT" == *xla* ]]; then

.ci/pytorch/common-build.sh

@@ -2,6 +2,8 @@
 # Required environment variables:
 #   $BUILD_ENVIRONMENT (should be set by your Docker image)
 
+set -e -x -o pipefail
+
 if [[ "$BUILD_ENVIRONMENT" != *win-* ]]; then
     # Save the absolute path in case later we chdir (as occurs in the gpu perf test)
     script_dir="$( cd "$(dirname "${BASH_SOURCE[0]}")" || exit ; pwd -P )"
@@ -45,14 +47,14 @@ if [[ "$BUILD_ENVIRONMENT" != *win-* ]]; then
             # explicitly
             echo "Skipping sccache server initialization, setting environment variables"
             export SCCACHE_IDLE_TIMEOUT=0
-            export SCCACHE_ERROR_LOG=~/sccache_error.log
-            export RUST_LOG=sccache::server=error
+            export SCCACHE_ERROR_LOG=/tmp/sccache_error.log
+            export RUST_LOG=sccache::server=debug
         elif [[ "${BUILD_ENVIRONMENT}" == *rocm* ]]; then
-            SCCACHE_ERROR_LOG=~/sccache_error.log SCCACHE_IDLE_TIMEOUT=0 sccache --start-server
+            SCCACHE_ERROR_LOG=/tmp/sccache_error.log SCCACHE_IDLE_TIMEOUT=0 sccache --start-server
         else
             # increasing SCCACHE_IDLE_TIMEOUT so that extension_backend_test.cpp can build after this PR:
             # https://github.com/pytorch/pytorch/pull/16645
-            SCCACHE_ERROR_LOG=~/sccache_error.log SCCACHE_IDLE_TIMEOUT=0 RUST_LOG=sccache::server=error sccache --start-server
+            SCCACHE_ERROR_LOG=/tmp/sccache_error.log SCCACHE_IDLE_TIMEOUT=0 RUST_LOG=sccache::server=error sccache --start-server
         fi
 
         # Report sccache stats for easier debugging. It's ok if this commands

.github/workflows/pull.yml

@@ -1,16 +1,16 @@
 name: pull
 
 on:
-  pull_request:
-    branches-ignore:
-      - nightly
-  push:
-    branches:
-      - main
-      - release/*
-      - landchecks/*
-    tags:
-      - ciflow/pull/*
+  # pull_request:
+  #   branches-ignore:
+  #     - nightly
+  # push:
+  #   branches:
+  #     - main
+  #     - release/*
+  #     - landchecks/*
+  #   tags:
+  #     - ciflow/pull/*
   workflow_dispatch:
   schedule:
     - cron: 29 8 * * *  # about 1:29am PDT

.github/workflows/trunk.yml

@@ -47,22 +47,6 @@ jobs:
       curr_branch: ${{ github.head_ref || github.ref_name }}
       curr_ref_type: ${{ github.ref_type }}
 
-  libtorch-linux-jammy-cuda12_8-py3_10-gcc11-debug-build:
-    name: libtorch-linux-jammy-cuda12.8-py3.10-gcc11-debug
-    uses: ./.github/workflows/_linux-build.yml
-    needs: get-label-type
-    with:
-      build-environment: libtorch-linux-jammy-cuda12.8-py3.10-gcc11
-      docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11
-      build-generates-artifacts: false
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runner: "linux.4xlarge"
-      test-matrix: |
-        { include: [
-          { config: "default", shard: 1, num_shards: 1 },
-        ]}
-    secrets: inherit
-
   linux-jammy-cuda12_8-py3_10-gcc11-build:
     name: linux-jammy-cuda12.8-py3.10-gcc11
     uses: ./.github/workflows/_linux-build.yml
@@ -85,167 +69,3 @@ jobs:
           { config: "pr_time_benchmarks", shard: 1, num_shards: 1, runner: "linux.g4dn.metal.nvidia.gpu" },
         ]}
     secrets: inherit
-
-  linux-jammy-cuda12_8-py3_10-gcc11-test:
-    name: linux-jammy-cuda12.8-py3.10-gcc11
-    uses: ./.github/workflows/_linux-test.yml
-    needs:
-      - linux-jammy-cuda12_8-py3_10-gcc11-build
-      - target-determination
-    with:
-      timeout-minutes: 360
-      build-environment: linux-jammy-cuda12.8-py3.10-gcc11
-      docker-image: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-build.outputs.docker-image }}
-      test-matrix: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-build.outputs.test-matrix }}
-    secrets: inherit
-
-
-  # no-ops builds test USE_PER_OPERATOR_HEADERS=0 where ATen/ops is not generated
-  linux-jammy-cuda12_8-py3_10-gcc11-no-ops-build:
-    name: linux-jammy-cuda12.8-py3.10-gcc11-no-ops
-    uses: ./.github/workflows/_linux-build.yml
-    needs: get-label-type
-    with:
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build-environment: linux-jammy-cuda12.8-py3.10-gcc11-no-ops
-      docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11
-      test-matrix: |
-        { include: [
-          { config: "default", shard: 1, num_shards: 1 },
-        ]}
-    secrets: inherit
-
-  macos-py3-arm64-build:
-    if: github.repository_owner == 'pytorch'
-    name: macos-py3-arm64
-    uses: ./.github/workflows/_mac-build.yml
-    with:
-      sync-tag: macos-py3-arm64-build
-      build-environment: macos-py3-arm64
-      runner-type: macos-m1-stable
-      build-generates-artifacts: true
-      # To match the one pre-installed in the m1 runners
-      python-version: 3.12.7
-      test-matrix: |
-        { include: [
-          { config: "default", shard: 1, num_shards: 3, runner: "macos-m1-stable" },
-          { config: "default", shard: 2, num_shards: 3, runner: "macos-m1-stable" },
-          { config: "default", shard: 3, num_shards: 3, runner: "macos-m1-stable" },
-          { config: "mps", shard: 1, num_shards: 1, runner: "macos-m1-14" },
-          { config: "mps", shard: 1, num_shards: 1, runner: "macos-m2-15" },
-        ]}
-    secrets: inherit
-
-  macos-py3-arm64-test:
-    name: macos-py3-arm64
-    uses: ./.github/workflows/_mac-test.yml
-    needs:
-      - macos-py3-arm64-build
-      - target-determination
-    with:
-      build-environment: macos-py3-arm64
-      # Same as the build job
-      python-version: 3.12.7
-      test-matrix: ${{ needs.macos-py3-arm64-build.outputs.test-matrix }}
-      disable-monitor: false
-    secrets: inherit
-
-  win-vs2022-cpu-py3-build:
-    name: win-vs2022-cpu-py3
-    uses: ./.github/workflows/_win-build.yml
-    needs: get-label-type
-    with:
-      build-environment: win-vs2022-cpu-py3
-      cuda-version: cpu
-      runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
-      test-matrix: |
-        { include: [
-          { config: "default", shard: 1, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral" },
-          { config: "default", shard: 2, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral" },
-          { config: "default", shard: 3, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral" },
-          { config: "default", shard: 4, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral" },
-        ]}
-    secrets: inherit
-
-  win-vs2022-cpu-py3-test:
-    name: win-vs2022-cpu-py3
-    uses: ./.github/workflows/_win-test.yml
-    needs:
-      - win-vs2022-cpu-py3-build
-      - target-determination
-    with:
-      build-environment: win-vs2022-cpu-py3
-      cuda-version: cpu
-      test-matrix: ${{ needs.win-vs2022-cpu-py3-build.outputs.test-matrix }}
-      disable-monitor: false
-    secrets: inherit
-
-  win-vs2022-cuda12_6-py3-build:
-    name: win-vs2022-cuda12.6-py3
-    uses: ./.github/workflows/_win-build.yml
-    needs: get-label-type
-    with:
-      build-environment: win-vs2022-cuda12.6-py3
-      cuda-version: "12.6"
-      runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
-    secrets: inherit
-
-  inductor-build:
-    name: inductor-build
-    uses: ./.github/workflows/_linux-build.yml
-    needs: get-label-type
-    with:
-      build-environment: linux-jammy-cuda12.8-py3.12-gcc9-sm80
-      docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc9-inductor-benchmarks
-      cuda-arch-list: '8.0'
-    secrets: inherit
-
-  verify-cachebench-cpu-build:
-    name: verify-cachebench-cpu-build
-    uses: ./.github/workflows/_linux-build.yml
-    needs: get-label-type
-    with:
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build-environment: linux-jammy-py3.10-gcc11
-      docker-image-name: ci-image:pytorch-linux-jammy-py3-gcc11-inductor-benchmarks
-      test-matrix: |
-        { include: [
-          { config: "verify_cachebench", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
-        ]}
-    secrets: inherit
-
-  verify-cachebench-cpu-test:
-    name: verify-cachebench-cpu-test
-    uses: ./.github/workflows/_linux-test.yml
-    needs:
-      - verify-cachebench-cpu-build
-      - target-determination
-    with:
-      build-environment: linux-jammy-py3.10-gcc11
-      docker-image: ${{ needs.verify-cachebench-cpu-build.outputs.docker-image }}
-      test-matrix: ${{ needs.verify-cachebench-cpu-build.outputs.test-matrix }}
-    secrets: inherit
-
-  linux-jammy-py3-clang12-executorch-build:
-    name: linux-jammy-py3-clang12-executorch
-    uses: ./.github/workflows/_linux-build.yml
-    needs: get-label-type
-    with:
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build-environment: linux-jammy-py3-clang12-executorch
-      docker-image-name: ci-image:pytorch-linux-jammy-py3-clang12-executorch
-      test-matrix: |
-        { include: [
-          { config: "executorch", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
-        ]}
-    secrets: inherit
-
-  linux-jammy-py3-clang12-executorch-test:
-    name: linux-jammy-py3-clang12-executorch
-    uses: ./.github/workflows/_linux-test.yml
-    needs: linux-jammy-py3-clang12-executorch-build
-    with:
-      build-environment: linux-jammy-py3-clang12-executorch
-      docker-image: ${{ needs.linux-jammy-py3-clang12-executorch-build.outputs.docker-image }}
-      test-matrix: ${{ needs.linux-jammy-py3-clang12-executorch-build.outputs.test-matrix }}
-    secrets: inherit

CMakeLists.txt

@@ -420,6 +420,14 @@ if(USE_CCACHE)
   endif()
 endif()
 
+# Optionally disable -MD / -MMD if COMPILE_NO_MD is set in the environment
+if(DEFINED ENV{COMPILE_NO_MD})
+  message(STATUS "COMPILE_NO_MD is set — disabling compiler dependency file flags (-MD/-MMD)")
+  foreach(lang C CXX CUDA HIP ASM)
+    set(CMAKE_DEPFILE_FLAGS_${lang} "")
+  endforeach()
+endif()
+
 # Since TensorPipe does not support Windows, set it to OFF when WIN32 detected
 # On Windows platform, if user does not install libuv in build conda env and
 # does not set libuv_ROOT environment variable. Set USE_DISTRIBUTED to OFF.
@@ -1487,3 +1495,10 @@ else()
     ]])
   endif()
 endif()
+
+foreach(lang C CXX CUDA)
+  foreach(flg "" "_DEBUG" "_RELEASE" "_RELWITHDEBINFO")
+    string(REPLACE "-MD" "" CMAKE_${lang}_FLAGS${flg} "${CMAKE_${lang}_FLAGS${flg}}")
+    string(REPLACE "-MMD" "" CMAKE_${lang}_FLAGS${flg} "${CMAKE_${lang}_FLAGS${flg}}")
+  endforeach()
+endforeach()