Merge branch 'findhao/operatorbench2' into findhao/operatorbench3

.ci/docker/build.sh

@@ -355,12 +355,6 @@ case "$image" in
     CONDA_CMAKE=yes
     VISION=yes
     ;;
-  pytorch-linux-jammy-py3-clang18-asan)
-    ANACONDA_PYTHON_VERSION=3.10
-    CLANG_VERSION=18
-    CONDA_CMAKE=yes
-    VISION=yes
-    ;;
   pytorch-linux-jammy-py3.9-gcc11)
     ANACONDA_PYTHON_VERSION=3.9
     GCC_VERSION=11
@@ -387,13 +381,6 @@ case "$image" in
     HALIDE=yes
     TRITON=yes
     ;;
-  pytorch-linux-jammy-py3.12-triton-cpu)
-    CUDA_VERSION=12.4
-    ANACONDA_PYTHON_VERSION=3.12
-    GCC_VERSION=11
-    CONDA_CMAKE=yes
-    TRITON_CPU=yes
-    ;;
   pytorch-linux-focal-linter)
     # TODO: Use 3.9 here because of this issue https://github.com/python/mypy/issues/13627.
     # We will need to update mypy version eventually, but that's for another day. The task
@@ -523,7 +510,6 @@ docker build \
        --build-arg "UCC_COMMIT=${UCC_COMMIT}" \
        --build-arg "CONDA_CMAKE=${CONDA_CMAKE}" \
        --build-arg "TRITON=${TRITON}" \
-       --build-arg "TRITON_CPU=${TRITON_CPU}" \
        --build-arg "ONNX=${ONNX}" \
        --build-arg "DOCS=${DOCS}" \
        --build-arg "INDUCTOR_BENCHMARKS=${INDUCTOR_BENCHMARKS}" \

.ci/docker/ci_commit_pins/triton-cpu.txt

@@ -1 +0,0 @@
-6a333f1b05671f6fada4ba7bbfae4a02a9d96f4f

.ci/docker/common/install_clang.sh

@@ -13,17 +13,11 @@ if [ -n "$CLANG_VERSION" ]; then
   elif [[ $UBUNTU_VERSION == 22.04 ]]; then
     # work around ubuntu apt-get conflicts
     sudo apt-get -y -f install
-    wget --no-check-certificate -O - https://apt.llvm.org/llvm-snapshot.gpg.key | sudo apt-key add  -
-    if [[ $CLANG_VERSION == 18 ]]; then
-      apt-add-repository "deb http://apt.llvm.org/jammy/ llvm-toolchain-jammy-18 main"
-    fi
   fi
 
   sudo apt-get update
-  apt-get install -y --no-install-recommends clang-"$CLANG_VERSION" llvm-"$CLANG_VERSION"
-  if [[ $CLANG_VERSION == 18 ]]; then
-    apt-get install -y --no-install-recommends libomp-18-dev
-  fi
+  apt-get install -y --no-install-recommends clang-"$CLANG_VERSION"
+  apt-get install -y --no-install-recommends llvm-"$CLANG_VERSION"
 
   # Install dev version of LLVM.
   if [ -n "$LLVMDEV" ]; then

.ci/docker/common/install_onnx.sh

@@ -32,7 +32,7 @@ pip_install coloredlogs packaging
 
 pip_install onnxruntime==1.18.1
 pip_install onnx==1.16.2
-pip_install onnxscript==0.1.0.dev20241008 --no-deps
+pip_install onnxscript==0.1.0.dev20240831 --no-deps
 # required by onnxscript
 pip_install ml_dtypes
 

.ci/docker/common/install_triton.sh

@@ -15,11 +15,8 @@ conda_reinstall() {
 if [ -n "${XPU_VERSION}" ]; then
   TRITON_REPO="https://github.com/intel/intel-xpu-backend-for-triton"
   TRITON_TEXT_FILE="triton-xpu"
-elif [ -n "${TRITON_CPU}" ]; then
-  TRITON_REPO="https://github.com/triton-lang/triton-cpu"
-  TRITON_TEXT_FILE="triton-cpu"
 else
-  TRITON_REPO="https://github.com/triton-lang/triton"
+  TRITON_REPO="https://github.com/openai/triton"
   TRITON_TEXT_FILE="triton"
 fi
 
@@ -47,10 +44,9 @@ chown -R jenkins /var/lib/jenkins/triton
 chgrp -R jenkins /var/lib/jenkins/triton
 pushd /var/lib/jenkins/
 
-as_jenkins git clone --recursive ${TRITON_REPO} triton
+as_jenkins git clone ${TRITON_REPO} triton
 cd triton
 as_jenkins git checkout ${TRITON_PINNED_COMMIT}
-as_jenkins git submodule update --init --recursive
 cd python
 
 # TODO: remove patch setup.py once we have a proper fix for https://github.com/triton-lang/triton/issues/4527

.ci/docker/requirements-ci.txt

@@ -139,9 +139,9 @@ opt-einsum==3.3
 #Pinned versions: 3.3
 #test that import: test_linalg.py
 
-optree==0.13.0
+optree==0.12.1
 #Description: A library for tree manipulation
-#Pinned versions: 0.13.0
+#Pinned versions: 0.12.1
 #test that import: test_vmap.py, test_aotdispatch.py, test_dynamic_shapes.py,
 #test_pytree.py, test_ops.py, test_control_flow.py, test_modules.py,
 #common_utils.py, test_eager_transforms.py, test_python_dispatch.py,

.ci/docker/ubuntu/Dockerfile

@@ -147,13 +147,6 @@ COPY ci_commit_pins/triton.txt triton.txt
 RUN if [ -n "${TRITON}" ]; then bash ./install_triton.sh; fi
 RUN rm install_triton.sh common_utils.sh triton.txt
 
-ARG TRITON_CPU
-COPY ./common/install_triton.sh install_triton.sh
-COPY ./common/common_utils.sh common_utils.sh
-COPY ci_commit_pins/triton-cpu.txt triton-cpu.txt
-RUN if [ -n "${TRITON_CPU}" ]; then bash ./install_triton.sh; fi
-RUN rm install_triton.sh common_utils.sh triton-cpu.txt
-
 ARG EXECUTORCH
 # Build and install executorch
 COPY ./common/install_executorch.sh install_executorch.sh

.ci/pytorch/build.sh

@@ -178,7 +178,7 @@ fi
 # sccache will fail for CUDA builds if all cores are used for compiling
 # gcc 7 with sccache seems to have intermittent OOM issue if all cores are used
 if [ -z "$MAX_JOBS" ]; then
-  if { [[ "$BUILD_ENVIRONMENT" == *cuda* ]]; } && which sccache > /dev/null; then
+  if { [[ "$BUILD_ENVIRONMENT" == *cuda* ]] || [[ "$BUILD_ENVIRONMENT" == *gcc7* ]]; } && which sccache > /dev/null; then
     export MAX_JOBS=$(($(nproc) - 1))
   fi
 fi
@@ -218,6 +218,10 @@ if [[ "${BUILD_ENVIRONMENT}" == *-pch* ]]; then
     export USE_PRECOMPILED_HEADERS=1
 fi
 
+if [[ "${BUILD_ENVIRONMENT}" == *linux-focal-py3.7-gcc7-build*  ]]; then
+  export USE_GLOO_WITH_OPENSSL=ON
+fi
+
 if [[ "${BUILD_ENVIRONMENT}" != *android* && "${BUILD_ENVIRONMENT}" != *cuda* ]]; then
   export BUILD_STATIC_RUNTIME_BENCHMARK=ON
 fi

.ci/pytorch/common_utils.sh

@@ -191,22 +191,9 @@ function install_torchrec_and_fbgemm() {
   pip_uninstall torchrec-nightly
   pip_uninstall fbgemm-gpu-nightly
   pip_install setuptools-git-versioning scikit-build pyre-extensions
-
-  # TODO (huydhn): I still have no clue on why sccache doesn't work with only fbgemm_gpu here, but it
-  # seems to be an sccache-related issue
-  if [[ "$IS_A100_RUNNER" == "1" ]]; then
-    unset CMAKE_CUDA_COMPILER_LAUNCHER
-    sudo mv /opt/cache/bin /opt/cache/bin-backup
-  fi
-
   # See https://github.com/pytorch/pytorch/issues/106971
   CUDA_PATH=/usr/local/cuda-12.1 pip_install --no-use-pep517 --user "git+https://github.com/pytorch/FBGEMM.git@${fbgemm_commit}#egg=fbgemm-gpu&subdirectory=fbgemm_gpu"
   pip_install --no-use-pep517 --user "git+https://github.com/pytorch/torchrec.git@${torchrec_commit}"
-
-  if [[ "$IS_A100_RUNNER" == "1" ]]; then
-    export CMAKE_CUDA_COMPILER_LAUNCHER=/opt/cache/bin/sccache
-    sudo mv /opt/cache/bin-backup /opt/cache/bin
-  fi
 }
 
 function clone_pytorch_xla() {

.ci/pytorch/test.sh

@@ -403,7 +403,7 @@ pr_time_benchmarks() {
   PYTHONPATH=$(pwd)/benchmarks/dynamo/pr_time_benchmarks source benchmarks/dynamo/pr_time_benchmarks/benchmark_runner.sh "$TEST_REPORTS_DIR/pr_time_benchmarks_results.csv" "benchmarks/dynamo/pr_time_benchmarks/benchmarks"
   echo "benchmark results on current PR: "
   cat  "$TEST_REPORTS_DIR/pr_time_benchmarks_results.csv"
-  PYTHONPATH=$(pwd)/benchmarks/dynamo/pr_time_benchmarks python benchmarks/dynamo/pr_time_benchmarks/check_results.py "benchmarks/dynamo/pr_time_benchmarks/expected_results.csv" "$TEST_REPORTS_DIR/pr_time_benchmarks_results.csv" "$TEST_REPORTS_DIR/new_expected_results.csv"
+  PYTHONPATH=$(pwd)/benchmarks/dynamo/pr_time_benchmarks python benchmarks/dynamo/pr_time_benchmarks/check_results.py "benchmarks/dynamo/pr_time_benchmarks/expected_results.csv" "$TEST_REPORTS_DIR/pr_time_benchmarks_results.csv"
 }
 
 if [[ "${TEST_CONFIG}" == *pr_time_benchmarks* ]]; then
@@ -606,11 +606,6 @@ test_inductor_halide() {
   assert_git_not_dirty
 }
 
-test_inductor_triton_cpu() {
-  python test/run_test.py --include inductor/test_triton_cpu_backend.py --verbose
-  assert_git_not_dirty
-}
-
 test_dynamo_benchmark() {
   # Usage: test_dynamo_benchmark huggingface 0
   TEST_REPORTS_DIR=$(pwd)/test/test-reports
@@ -665,6 +660,15 @@ test_inductor_torchbench_smoketest_perf() {
   # The threshold value needs to be actively maintained to make this check useful
   python benchmarks/dynamo/check_perf_csv.py -f "$TEST_REPORTS_DIR/inductor_training_smoketest.csv" -t 1.4
 
+  TORCHINDUCTOR_ABI_COMPATIBLE=1 python benchmarks/dynamo/torchbench.py --device cuda --performance --bfloat16 --inference \
+    --export-aot-inductor --only nanogpt --output "$TEST_REPORTS_DIR/inductor_inference_smoketest.csv"
+  # The threshold value needs to be actively maintained to make this check useful
+  # The perf number of nanogpt seems not very stable, e.g.
+  # https://github.com/pytorch/pytorch/actions/runs/7158691360/job/19491437314,
+  # and thus we lower its threshold to reduce flakiness. If this continues to be a problem,
+  # we switch to use some other model.
+  python benchmarks/dynamo/check_perf_csv.py -f "$TEST_REPORTS_DIR/inductor_inference_smoketest.csv" -t 4.9
+
   # Check memory compression ratio for a few models
   for test in hf_Albert timm_vision_transformer; do
     python benchmarks/dynamo/torchbench.py --device cuda --performance --backend inductor --amp --training \
@@ -1435,8 +1439,6 @@ elif [[ "${TEST_CONFIG}" == *inductor_distributed* ]]; then
   test_inductor_distributed
 elif [[ "${TEST_CONFIG}" == *inductor-halide* ]]; then
   test_inductor_halide
-elif [[ "${TEST_CONFIG}" == *inductor-triton-cpu* ]]; then
-  test_inductor_triton_cpu
 elif [[ "${TEST_CONFIG}" == *inductor-micro-benchmark* ]]; then
   test_inductor_micro_benchmark
 elif [[ "${TEST_CONFIG}" == *huggingface* ]]; then
@@ -1460,7 +1462,7 @@ elif [[ "${TEST_CONFIG}" == *torchbench* ]]; then
   # https://github.com/opencv/opencv-python/issues/885
   pip_install opencv-python==4.8.0.74
   if [[ "${TEST_CONFIG}" == *inductor_torchbench_smoketest_perf* ]]; then
-    checkout_install_torchbench hf_Bert hf_Albert timm_vision_transformer
+    checkout_install_torchbench hf_Bert hf_Albert nanogpt timm_vision_transformer
     PYTHONPATH=$(pwd)/torchbench test_inductor_torchbench_smoketest_perf
   elif [[ "${TEST_CONFIG}" == *inductor_torchbench_cpu_smoketest_perf* ]]; then
     checkout_install_torchbench timm_vision_transformer phlippe_densenet basic_gnn_edgecnn \

.ci/pytorch/win-build.sh

@@ -26,7 +26,7 @@ fi
 export SCRIPT_HELPERS_DIR=$SCRIPT_PARENT_DIR/win-test-helpers
 
 set +ex
-grep -E -R 'PyLong_(From|As)(Unsigned|)Long\(' --exclude=python_numbers.h  --exclude=pythoncapi_compat.h --exclude=eval_frame.c torch/
+grep -E -R 'PyLong_(From|As)(Unsigned|)Long\(' --exclude=python_numbers.h --exclude=eval_frame.c torch/
 PYLONG_API_CHECK=$?
 if [[ $PYLONG_API_CHECK == 0 ]]; then
   echo "Usage of PyLong_{From,As}{Unsigned}Long API may lead to overflow errors on Windows"

.circleci/scripts/binary_linux_test.sh

@@ -27,11 +27,12 @@ if [[ "$PACKAGE_TYPE" == conda ]]; then
   source activate testenv >/dev/null
 elif [[ "$PACKAGE_TYPE" != libtorch ]]; then
   python_path="/opt/python/cp\$python_nodot-cp\${python_nodot}"
-  if [[ "\$python_nodot" = *t ]]; then
-    python_digits="\$(echo $DESIRED_PYTHON | tr -cd [:digit:])"
-    python_path="/opt/python/cp\$python_digits-cp\${python_digits}t"
+  # Prior to Python 3.8 paths were suffixed with an 'm'
+  if [[ -d  "\${python_path}/bin" ]]; then
+    export PATH="\${python_path}/bin:\$PATH"
+  elif [[ -d "\${python_path}m/bin" ]]; then
+    export PATH="\${python_path}m/bin:\$PATH"
   fi
-  export PATH="\${python_path}/bin:\$PATH"
 fi
 
 EXTRA_CONDA_FLAGS=""

.clang-format

@@ -44,9 +44,7 @@ ContinuationIndentWidth: 4
 Cpp11BracedListStyle: true
 DerivePointerAlignment: false
 DisableFormat:   false
-ForEachMacros:
-  - FOR_EACH_RANGE
-  - FOR_EACH
+ForEachMacros:   [ FOR_EACH_RANGE, FOR_EACH, ]
 IncludeCategories:
   - Regex:           '^<.*\.h(pp)?>'
     Priority:        1
@@ -60,24 +58,6 @@ IndentWrappedFunctionNames: false
 KeepEmptyLinesAtTheStartOfBlocks: false
 MacroBlockBegin: ''
 MacroBlockEnd:   ''
-Macros:
-  - >-
-    PyObject_HEAD_INIT(type)={
-        /* this is not exactly match with PyObject_HEAD_INIT in Python source code
-         * but it is enough for clang-format */
-        { 0xFFFFFFFF },
-        (type)
-    },
-  - >-
-    PyVarObject_HEAD_INIT(type, size)={
-        {
-            /* manually expand PyObject_HEAD_INIT(type) above
-             * because clang-format do not support recursive expansion */
-            { 0xFFFFFFFF },
-            (type)
-        },
-        (size)
-    },
 MaxEmptyLinesToKeep: 1
 NamespaceIndentation: None
 PenaltyBreakBeforeFirstCallParameter: 1
@@ -99,11 +79,7 @@ SpacesInContainerLiterals: true
 SpacesInCStyleCastParentheses: false
 SpacesInParentheses: false
 SpacesInSquareBrackets: false
-Standard:        c++17
-StatementMacros:
-  - PyObject_HEAD
-  - PyObject_VAR_HEAD
-  - PyException_HEAD
+Standard:        Cpp11
 TabWidth:        8
 UseTab:          Never
 ---

.github/ISSUE_TEMPLATE.md

@@ -0,0 +1,38 @@
+If you have a question or would like help and support, please ask at our
+[forums](https://discuss.pytorch.org/).
+
+If you are submitting a feature request, please preface the title with [feature request].
+If you are submitting a bug report, please fill in the following details.
+
+## Issue description
+
+Provide a short description.
+
+## Code example
+
+Please try to provide a minimal example to repro the bug.
+Error messages and stack traces are also helpful.
+
+## System Info
+Please copy and paste the output from our
+[environment collection script](https://raw.githubusercontent.com/pytorch/pytorch/main/torch/utils/collect_env.py)
+(or fill out the checklist below manually).
+
+You can get the script and run it with:
+```
+wget https://raw.githubusercontent.com/pytorch/pytorch/main/torch/utils/collect_env.py
+# For security purposes, please check the contents of collect_env.py before running it.
+python collect_env.py
+```
+
+- PyTorch or Caffe2:
+- How you installed PyTorch (conda, pip, source):
+- Build command you used (if compiling from source):
+- OS:
+- PyTorch version:
+- Python version:
+- CUDA/cuDNN version:
+- GPU models and configuration:
+- GCC version (if compiling from source):
+- CMake version:
+- Versions of any other relevant libraries:

.github/actions/checkout-pytorch/action.yml

@@ -18,14 +18,8 @@ inputs:
 runs:
   using: composite
   steps:
-    - name: Check if in a container runner
-      shell: bash
-      id: check_container_runner
-      run: echo "IN_CONTAINER_RUNNER=$(if [ -f /.inarc ] || [ -f /.incontainer ]; then echo true ; else echo false; fi)" >> "$GITHUB_OUTPUT"
-
     - name: Clean workspace
       shell: bash
-      if: ${{ steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' }}
       env:
         NO_SUDO: ${{ inputs.no-sudo }}
       run: |

.github/actions/linux-test/action.yml

@@ -85,25 +85,15 @@ runs:
       with:
         docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
-    - name: Check if in a container runner
+    - name: Check if in a ARC runner
       shell: bash
-      id: check_container_runner
-      run: echo "IN_CONTAINER_RUNNER=$(if [ -f /.inarc ] || [ -f /.incontainer ]; then echo true ; else echo false; fi)" >> "$GITHUB_OUTPUT"
+      id: check_arc_runner
+      run: echo "IN_ARC_RUNNER=$([ -f /.inarc ] && echo true || echo false)" >> "$GITHUB_OUTPUT"
 
     - name: Install nvidia driver, nvidia-docker runtime, set GPU_FLAG
       id: install-nvidia-driver
       uses: pytorch/test-infra/.github/actions/setup-nvidia@main
-      if: ${{ contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu') && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' }}
-
-    - name: Setup GPU_FLAG for docker run
-      id: setup-gpu-flag
-      run: echo "GPU_FLAG=--gpus all -e NVIDIA_DRIVER_CAPABILITIES=all" >> "${GITHUB_ENV}"
-      if: ${{ contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu') && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'true' }}
-
-    - name: Setup SCCACHE_SERVER_PORT environment for docker run when on container
-      id: setup-sscache-port-flag
-      run: echo "SCCACHE_SERVER_PORT_DOCKER_FLAG=-e SCCACHE_SERVER_PORT=$((RUNNER_UID + 4226))" >> "${GITHUB_ENV}"
-      if: ${{ steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'true' }}
+      if: ${{ contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu') && steps.check_arc_runner.outputs.IN_ARC_RUNNER == 'false' }}
 
     - name: Lock NVIDIA A100 40GB Frequency
       shell: bash
@@ -111,7 +101,7 @@ runs:
         sudo nvidia-smi -pm 1
         sudo nvidia-smi -ac 1215,1410
         nvidia-smi
-      if: ${{ contains(matrix.runner, 'a100') && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' }}
+      if: contains(matrix.runner, 'a100')
 
     - name: Start monitoring script
       id: monitor-script
@@ -182,7 +172,6 @@ runs:
         NO_TD: ${{ steps.keep-going.outputs.ci-no-td }}
         TD_DISTRIBUTED: ${{ steps.keep-going.outputs.ci-td-distributed }}
         SCCACHE_BUCKET: ossci-compiler-cache-circleci-v2
-        SCCACHE_REGION: us-east-1
         SCCACHE_S3_KEY_PREFIX: ${{ github.workflow }}
         SHM_SIZE: ${{ contains(inputs.build-environment, 'cuda') && '2g' || '1g' }}
         DOCKER_IMAGE: ${{ inputs.docker-image }}
@@ -192,9 +181,6 @@ runs:
         PYTORCH_TEST_RERUN_DISABLED_TESTS: ${{ matrix.rerun_disabled_tests && '1' || '0' }}
         DASHBOARD_TAG: ${{ inputs.dashboard-tag }}
         HUGGING_FACE_HUB_TOKEN: ${{ inputs.HUGGING_FACE_HUB_TOKEN }}
-        SCRIBE_GRAPHQL_ACCESS_TOKEN: ${{ secrets.SCRIBE_GRAPHQL_ACCESS_TOKEN }}
-        IS_A100_RUNNER: ${{ contains(matrix.runner, 'a100') && '1' || '0' }}
-
       shell: bash
       run: |
         set -x
@@ -213,7 +199,6 @@ runs:
         # shellcheck disable=SC2086,SC2090
         container_name=$(docker run \
           ${GPU_FLAG:-} \
-          ${SCCACHE_SERVER_PORT_DOCKER_FLAG:-} \
           -e BUILD_ENVIRONMENT \
           -e PR_NUMBER \
           -e GITHUB_ACTIONS \
@@ -242,7 +227,6 @@ runs:
           -e PR_LABELS \
           -e MAX_JOBS="$(nproc --ignore=2)" \
           -e SCCACHE_BUCKET \
-          -e SCCACHE_REGION \
           -e SCCACHE_S3_KEY_PREFIX \
           -e XLA_CUDA \
           -e XLA_CLANG_CACHE_S3_BUCKET_NAME \
@@ -250,9 +234,7 @@ runs:
           -e PYTORCH_TEST_RERUN_DISABLED_TESTS \
           -e SKIP_SCCACHE_INITIALIZATION=1 \
           -e HUGGING_FACE_HUB_TOKEN \
-          -e SCRIBE_GRAPHQL_ACCESS_TOKEN \
           -e DASHBOARD_TAG \
-          -e IS_A100_RUNNER \
           --env-file="/tmp/github_env_${GITHUB_RUN_ID}" \
           --security-opt seccomp=unconfined \
           --cap-add=SYS_PTRACE \
@@ -323,7 +305,7 @@ runs:
 
     - name: Teardown Linux
       uses: pytorch/test-infra/.github/actions/teardown-linux@main
-      if: always() && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false'
+      if: always()
 
     # NB: We are currently having an intermittent GPU-related issue on G5 runners with
     # A10G GPU. Once this happens, trying to reset the GPU as done in setup-nvidia does

.github/actions/setup-linux/action.yml

@@ -28,14 +28,14 @@ runs:
         echo "instance-type: $(get_ec2_metadata instance-type)"
         echo "system info $(uname -a)"
 
-    - name: Check if in a container runner
+    - name: Check if in a ARC runner
       shell: bash
-      id: check_container_runner
-      run: echo "IN_CONTAINER_RUNNER=$(if [ -f /.inarc ] || [ -f /.incontainer ]; then echo true ; else echo false; fi)" >> "$GITHUB_OUTPUT"
+      id: check_arc_runner
+      run: echo "IN_ARC_RUNNER=$([ -f /.inarc ] && echo true || echo false)"  >> $GITHUB_OUTPUT
 
     - name: Start docker if docker deamon is not running
       shell: bash
-      if: ${{ steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' }}
+      if: ${{ steps.check_arc_runner.outputs.IN_ARC_RUNNER == 'false' }}
       run: |
         if systemctl is-active --quiet docker; then
             echo "Docker daemon is running...";
@@ -73,7 +73,7 @@ runs:
         env | grep '^CI' >> "/tmp/github_env_${GITHUB_RUN_ID}"
 
     - name: Kill any existing containers, clean up images
-      if: ${{ steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' }}
+      if: ${{ steps.check_arc_runner.outputs.IN_ARC_RUNNER == 'false' }}
       shell: bash
       run: |
         # ignore expansion of "docker ps -q" since it could be empty
@@ -116,7 +116,7 @@ runs:
     - name: Check that the docker daemon is running
       shell: bash
       continue-on-error: true
-      if: ${{ steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'true' }}
+      if: ${{ steps.check_arc_runner.outputs.IN_ARC_RUNNER == 'true' }}
       run: |
         set +x
 

.github/ci_commit_pins/audio.txt

@@ -1 +1 @@
-3f0569939c4369bec943fc27d1c9d8dfbc828c26
+ba696ea3dfec4cbe693bf06a84c75dc196077f5b

.github/pytorch-probot.yml

@@ -16,7 +16,6 @@ ciflow_push_tags:
 - ciflow/nightly
 - ciflow/periodic
 - ciflow/rocm
-- ciflow/s390
 - ciflow/slow
 - ciflow/trunk
 - ciflow/unstable

.github/requirements/pip-requirements-iOS.txt

@@ -1,4 +1,4 @@
 # iOS simulator requirements
 coremltools==5.0b5
 protobuf==3.20.2
-optree==0.13.0
+optree==0.12.1

.github/requirements/pip-requirements-macOS.txt

@@ -27,7 +27,7 @@ pytest-cpp==2.3.0
 rockset==1.0.3
 z3-solver==4.12.2.0
 tensorboard==2.13.0
-optree==0.13.0
+optree==0.12.1
 # NB: test_hparams_* from test_tensorboard is failing with protobuf 5.26.0 in
 # which the stringify metadata is wrong when escaping double quote
 protobuf==3.20.2

.github/scripts/generate_binary_build_matrix.py

@@ -333,7 +333,7 @@ def generate_wheels_matrix(
         package_type = "manywheel"
 
     if python_versions is None:
-        python_versions = FULL_PYTHON_VERSIONS + ["3.13", "3.13t"]
+        python_versions = FULL_PYTHON_VERSIONS + ["3.13"]
 
     if arches is None:
         # Define default compute archivectures
@@ -369,13 +369,7 @@ def generate_wheels_matrix(
             # TODO: Enable python 3.13 on rocm, aarch64, windows
             if (
                 gpu_arch_type == "rocm" or (os != "linux" and os != "linux-s390x")
-            ) and (python_version == "3.13" or python_version == "3.13t"):
-                continue
-
-            # TODO: Enable python 3.13t on xpu and cpu-s390x
-            if (
-                gpu_arch_type == "xpu" or gpu_arch_type == "cpu-s390x"
-            ) and python_version == "3.13t":
+            ) and python_version == "3.13":
                 continue
 
             if use_split_build and (

.github/scripts/runner_determinator.py

@@ -1,9 +1,5 @@
 # flake8: noqa: G004
 
-# Note: Copies of this script in runner_determinator.py and _runner-determinator.yml
-#       must be kept in sync. You can do it easily by running the following command:
-#           python .github/scripts/update_runner_determinator.py
-
 """
 This runner determinator is used to determine which set of runners to run a
 GitHub job on. It uses the first comment of a GitHub issue (by default
@@ -83,9 +79,6 @@ class Experiment(NamedTuple):
     rollout_perc: float = (
         0  # Percentage of workflows to experiment on when user is not opted-in.
     )
-    all_branches: bool = (
-        False  # If True, the experiment is also enabled on the exception branches
-    )
 
     # Add more fields as needed
 
@@ -219,7 +212,7 @@ def get_potential_pr_author(
 
 def is_exception_branch(branch: str) -> bool:
     """
-    Branches that get opted out of experiments by default, until they're explicitly enabled.
+    Branches that get opted out of all experiments and should always use Meta runners
     """
     return branch.split("/")[0] in {"main", "nightly", "release", "landchecks"}
 
@@ -345,10 +338,7 @@ def is_user_opted_in(user: str, user_optins: UserOptins, experiment_name: str) -
 
 
 def get_runner_prefix(
-    rollout_state: str,
-    workflow_requestors: Iterable[str],
-    branch: str,
-    is_canary: bool = False,
+    rollout_state: str, workflow_requestors: Iterable[str], is_canary: bool = False
 ) -> str:
     settings = parse_settings(rollout_state)
     user_optins = parse_users(rollout_state)
@@ -358,12 +348,6 @@ def get_runner_prefix(
     for experiment_name, experiment_settings in settings.experiments.items():
         enabled = False
 
-        if not experiment_settings.all_branches and is_exception_branch(branch):
-            log.info(
-                f"Branch {branch} is an exception branch. Not enabling experiment {experiment_name}."
-            )
-            continue
-
         # Is any workflow_requestor opted in to this experiment?
         opted_in_users = [
             requestor
@@ -423,34 +407,35 @@ def get_rollout_state_from_issue(github_token: str, repo: str, issue_num: int) -
 def main() -> None:
     args = parse_args()
 
-    runner_label_prefix = DEFAULT_LABEL_PREFIX
-
-    try:
-        rollout_state = get_rollout_state_from_issue(
-            args.github_token, args.github_issue_repo, args.github_issue
+    if args.github_ref_type == "branch" and is_exception_branch(args.github_branch):
+        log.info(
+            f"Exception branch: '{args.github_branch}', using Meta runners and no experiments."
         )
+        runner_label_prefix = DEFAULT_LABEL_PREFIX
+    else:
+        try:
+            rollout_state = get_rollout_state_from_issue(
+                args.github_token, args.github_issue_repo, args.github_issue
+            )
 
-        username = get_potential_pr_author(
-            args.github_token,
-            args.github_repo,
-            args.github_actor,
-            args.github_ref_type,
-            args.github_branch,
-        )
+            username = get_potential_pr_author(
+                args.github_token,
+                args.github_repo,
+                args.github_actor,
+                args.github_ref_type,
+                args.github_branch,
+            )
 
-        is_canary = args.github_repo == "pytorch/pytorch-canary"
+            is_canary = args.github_repo == "pytorch/pytorch-canary"
 
-        runner_label_prefix = get_runner_prefix(
-            rollout_state,
-            (args.github_issue_owner, username),
-            args.github_branch,
-            is_canary,
-        )
+            runner_label_prefix = get_runner_prefix(
+                rollout_state, (args.github_issue_owner, username), is_canary
+            )
 
-    except Exception as e:
-        log.error(
-            f"Failed to get issue. Defaulting to Meta runners and no experiments. Exception: {e}"
-        )
+        except Exception as e:
+            log.error(
+                f"Failed to get issue. Defaulting to Meta runners and no experiments. Exception: {e}"
+            )
 
     set_github_output(GH_OUTPUT_KEY_LABEL_TYPE, runner_label_prefix)
 

.github/scripts/test_runner_determinator.py

@@ -4,10 +4,6 @@ from unittest.mock import Mock, patch
 import runner_determinator as rd
 
 
-USER_BRANCH = "somebranch"
-EXCEPTION_BRANCH = "main"
-
-
 class TestRunnerDeterminatorIssueParser(TestCase):
     def test_parse_settings(self) -> None:
         settings_text = """
@@ -70,40 +66,6 @@ class TestRunnerDeterminatorIssueParser(TestCase):
             "otherExp settings not parsed correctly",
         )
 
-    def test_parse_all_branches_setting(self) -> None:
-        settings_text = """
-        ```
-        experiments:
-            lf:
-                rollout_perc: 25
-                all_branches: true
-            otherExp:
-                all_branches: True
-                rollout_perc: 0
-        ```
-
-        ---
-
-        Users:
-        @User1,lf
-        @User2,lf,otherExp
-
-        """
-
-        settings = rd.parse_settings(settings_text)
-
-        self.assertTupleEqual(
-            rd.Experiment(rollout_perc=25, all_branches=True),
-            settings.experiments["lf"],
-            "lf settings not parsed correctly",
-        )
-        self.assertTrue(settings.experiments["otherExp"].all_branches)
-        self.assertTupleEqual(
-            rd.Experiment(rollout_perc=0, all_branches=True),
-            settings.experiments["otherExp"],
-            "otherExp settings not parsed correctly",
-        )
-
     def test_parse_users(self) -> None:
         settings_text = """
         experiments:
@@ -157,7 +119,7 @@ class TestRunnerDeterminatorGetRunnerPrefix(TestCase):
         @User2,lf,otherExp
 
         """
-        prefix = rd.get_runner_prefix(settings_text, ["User1"], USER_BRANCH)
+        prefix = rd.get_runner_prefix(settings_text, ["User1"])
         self.assertEqual("lf.", prefix, "Runner prefix not correct for User1")
 
     def test_opted_in_user_two_experiments(self) -> None:
@@ -174,7 +136,7 @@ class TestRunnerDeterminatorGetRunnerPrefix(TestCase):
         @User2,lf,otherExp
 
         """
-        prefix = rd.get_runner_prefix(settings_text, ["User2"], USER_BRANCH)
+        prefix = rd.get_runner_prefix(settings_text, ["User2"])
         self.assertEqual("lf.otherExp.", prefix, "Runner prefix not correct for User2")
 
     @patch("random.uniform", return_value=50)
@@ -192,7 +154,7 @@ class TestRunnerDeterminatorGetRunnerPrefix(TestCase):
         @User2,lf,otherExp
 
         """
-        prefix = rd.get_runner_prefix(settings_text, ["User3"], USER_BRANCH)
+        prefix = rd.get_runner_prefix(settings_text, ["User3"])
         self.assertEqual("", prefix, "Runner prefix not correct for user")
 
     @patch("random.uniform", return_value=10)
@@ -212,7 +174,7 @@ class TestRunnerDeterminatorGetRunnerPrefix(TestCase):
         """
 
         # User3 is opted out, but is pulled into both experiments by the 10% rollout
-        prefix = rd.get_runner_prefix(settings_text, ["User3"], USER_BRANCH)
+        prefix = rd.get_runner_prefix(settings_text, ["User3"])
         self.assertEqual("lf.otherExp.", prefix, "Runner prefix not correct for user")
 
     def test_lf_prefix_always_comes_first(self) -> None:
@@ -230,7 +192,7 @@ class TestRunnerDeterminatorGetRunnerPrefix(TestCase):
 
         """
 
-        prefix = rd.get_runner_prefix(settings_text, ["User2"], USER_BRANCH)
+        prefix = rd.get_runner_prefix(settings_text, ["User2"])
         self.assertEqual("lf.otherExp.", prefix, "Runner prefix not correct for user")
 
     def test_ignores_commented_users(self) -> None:
@@ -248,7 +210,7 @@ class TestRunnerDeterminatorGetRunnerPrefix(TestCase):
 
         """
 
-        prefix = rd.get_runner_prefix(settings_text, ["User1"], USER_BRANCH)
+        prefix = rd.get_runner_prefix(settings_text, ["User1"])
         self.assertEqual("", prefix, "Runner prefix not correct for user")
 
     def test_ignores_extra_experiments(self) -> None:
@@ -267,44 +229,9 @@ class TestRunnerDeterminatorGetRunnerPrefix(TestCase):
 
         """
 
-        prefix = rd.get_runner_prefix(settings_text, ["User1"], USER_BRANCH)
+        prefix = rd.get_runner_prefix(settings_text, ["User1"])
         self.assertEqual("lf.otherExp.", prefix, "Runner prefix not correct for user")
 
-    def test_disables_experiment_on_exception_branches_when_not_explicitly_opted_in(
-        self,
-    ) -> None:
-        settings_text = """
-        experiments:
-            lf:
-                rollout_perc: 100
-        ---
-
-        Users:
-        @User,lf,otherExp
-
-        """
-
-        prefix = rd.get_runner_prefix(settings_text, ["User1"], EXCEPTION_BRANCH)
-        self.assertEqual("", prefix, "Runner prefix not correct for user")
-
-    def test_allows_experiment_on_exception_branches_when_explicitly_opted_in(
-        self,
-    ) -> None:
-        settings_text = """
-        experiments:
-            lf:
-                rollout_perc: 100
-                all_branches: true
-        ---
-
-        Users:
-        @User,lf,otherExp
-
-        """
-
-        prefix = rd.get_runner_prefix(settings_text, ["User1"], EXCEPTION_BRANCH)
-        self.assertEqual("lf.", prefix, "Runner prefix not correct for user")
-
 
 if __name__ == "__main__":
     main()

.github/scripts/test_trymerge.py

@@ -12,7 +12,7 @@ import json
 import os
 import warnings
 from hashlib import sha256
-from typing import Any, List, Optional
+from typing import Any, Dict, List, Optional
 from unittest import main, mock, skip, TestCase
 from urllib.error import HTTPError
 
@@ -24,6 +24,7 @@ from trymerge import (
     find_matching_merge_rule,
     get_classifications,
     get_drci_classifications,
+    get_rockset_results,
     gh_get_team_members,
     GitHubPR,
     JobCheckState,
@@ -41,6 +42,7 @@ if "GIT_REMOTE_URL" not in os.environ:
     os.environ["GIT_REMOTE_URL"] = "https://github.com/pytorch/pytorch"
 
 GQL_MOCKS = "gql_mocks.json.gz"
+ROCKSET_MOCKS = "rockset_mocks.json.gz"
 DRCI_MOCKS = "drci_mocks.json.gz"
 
 
@@ -75,11 +77,16 @@ def mock_query(
         if err.code == 401 or err.code == 403:
             err_msg = f"If you are seeing this message during workflow run, please make sure to update {file_name}"
             err_msg += f" locally, by deleting it and running {os.path.basename(__file__)} with"
-            err_msg += " GitHub Personal Access Token passed via GITHUB_TOKEN"
+            err_msg += " GitHub Personal Access Token passed via GITHUB_TOKEN,"
+            err_msg += " the rockset api key passed via ROCKSET_API_KEY,"
             err_msg += " and drci api key passed via DRCI_BOT_KEY environment variables"
-            if os.getenv("GITHUB_TOKEN") is None or os.getenv("DRCI_BOT_KEY") is None:
+            if (
+                os.getenv("GITHUB_TOKEN") is None
+                or os.getenv("ROCKSET_API_KEY") is None
+                or os.getenv("DRCI_BOT_KEY") is None
+            ):
                 err_msg = (
-                    "Failed to update cached queries as GITHUB_TOKEN or DRCI_BOT_KEY "
+                    "Failed to update cached queries as GITHUB_TOKEN or ROCKSET_API_KEY or DRCI_BOT_KEY "
                     + "is not defined. "
                     + err_msg
                 )
@@ -103,6 +110,16 @@ def mocked_gh_graphql(query: str, **kwargs: Any) -> Any:
     return mock_query(gh_graphql_wrapper, GQL_MOCKS, key_function, query, kwargs)
 
 
+def mocked_rockset_results(head_sha: str, merge_base: str, num_retries: int = 3) -> Any:
+    return mock_query(
+        get_rockset_results,
+        ROCKSET_MOCKS,
+        lambda x, y: f"{x} {y}",
+        head_sha,
+        merge_base,
+    )
+
+
 def mocked_drci_classifications(pr_num: int, project: str, num_retries: int = 3) -> Any:
     return mock_query(
         get_drci_classifications,
@@ -256,6 +273,10 @@ def xla_merge_rules(repo: Any, org: str, project: str) -> List[MergeRule]:
     ]
 
 
+def empty_rockset_results(head_sha: str, merge_base: str) -> List[Dict[str, Any]]:
+    return []
+
+
 class DummyGitRepo(GitRepo):
     def __init__(self) -> None:
         super().__init__(get_git_repo_dir(), get_git_remote_name())
@@ -267,6 +288,7 @@ class DummyGitRepo(GitRepo):
         return "super awsome commit message"
 
 
+@mock.patch("trymerge.get_rockset_results", side_effect=empty_rockset_results)
 @mock.patch("trymerge.gh_graphql", side_effect=mocked_gh_graphql)
 @mock.patch(
     "trymerge.get_drci_classifications", side_effect=mocked_drci_classifications
@@ -582,6 +604,7 @@ class TestTryMerge(TestCase):
             mocked_gh_fetch_merge_base.assert_called_once()
 
 
+@mock.patch("trymerge.get_rockset_results", side_effect=mocked_rockset_results)
 @mock.patch("trymerge.gh_graphql", side_effect=mocked_gh_graphql)
 @mock.patch("trymerge.gh_fetch_merge_base", return_value="")
 @mock.patch(
@@ -820,7 +843,7 @@ class TestBypassFailures(TestCase):
         checks = pr.get_checkrun_conclusions()
 
         # Known flaky failure takes precedence over ignore current (need to set the
-        # merge base here to get the results from Dr. CI, and that categorize the
+        # merge base here to get the results from Rockset, and that categorize the
         # broken trunk failure too
         checks = get_classifications(
             pr.pr_num,
@@ -906,6 +929,7 @@ class TestBypassFailures(TestCase):
         )
 
 
+@mock.patch("trymerge.get_rockset_results", side_effect=mocked_rockset_results)
 @mock.patch("trymerge.gh_graphql", side_effect=mocked_gh_graphql)
 @mock.patch("trymerge.gh_fetch_merge_base", return_value="")
 @mock.patch("trymerge.get_drci_classifications", return_value={})
@@ -984,6 +1008,7 @@ class TestBypassFailuresOnSandCastle(TestCase):
         self.assertTrue(len(failed) == 2)
 
 
+@mock.patch("trymerge.get_rockset_results", side_effect=mocked_rockset_results)
 @mock.patch("trymerge.gh_graphql", side_effect=mocked_gh_graphql)
 @mock.patch("trymerge.gh_fetch_merge_base", return_value="")
 @mock.patch(

.github/scripts/trymerge.py

@@ -452,6 +452,8 @@ RE_DIFF_REV = re.compile(r"^Differential Revision:.+?(D[0-9]+)", re.MULTILINE)
 CIFLOW_LABEL = re.compile(r"^ciflow/.+")
 CIFLOW_TRUNK_LABEL = re.compile(r"^ciflow/trunk")
 MERGE_RULE_PATH = Path(".github") / "merge_rules.yaml"
+ROCKSET_MERGES_COLLECTION = "merges"
+ROCKSET_MERGES_WORKSPACE = "commons"
 REMOTE_MAIN_BRANCH = "origin/main"
 DRCI_CHECKRUN_NAME = "Dr.CI"
 INTERNAL_CHANGES_CHECKRUN_NAME = "Meta Internal-Only Changes Check"
@@ -1178,7 +1180,7 @@ class GitHubPR:
         merge_commit_sha = repo.rev_parse(name=self.default_branch())
 
         if comment_id and self.pr_num:
-            # Finally, upload the record to s3. The list of pending and failed
+            # Finally, upload the record to Rockset. The list of pending and failed
             # checks are at the time of the merge
             save_merge_record(
                 comment_id=comment_id,
@@ -1200,7 +1202,7 @@ class GitHubPR:
                 ignore_current=bool(ignore_current_checks),
             )
         else:
-            print("Missing comment ID or PR number, couldn't upload to s3")
+            print("Missing comment ID or PR number, couldn't upload to Rockset")
 
         # Usually Github will see that the commit has "resolves <pr_num>" in the
         # commit message and close the PR, but sometimes it doesn't, leading to
@@ -1479,7 +1481,7 @@ def find_matching_merge_rule(
 
         # Categorize all checks when skip_mandatory_checks (force merge) is set. Do it here
         # where the list of checks is readily available. These records will be saved into
-        # s3 merge records
+        # Rockset merge records
         (
             pending_mandatory_checks,
             failed_mandatory_checks,
@@ -1566,7 +1568,7 @@ def save_merge_record(
     This saves the merge records as a json, which can later be uploaded to s3
     """
 
-    # Prepare the record to be written into s3
+    # Prepare the record to be written into Rockset
     data = [
         {
             "comment_id": comment_id,
@@ -1588,8 +1590,7 @@ def save_merge_record(
             "ignore_current": ignore_current,
             "error": error,
             # This is a unique identifier for the record for deduping purposes
-            # in Rockset.  Any unique string would work.  This will not be used
-            # after we migrate off Rockset
+            # in rockset.  Any unique string would work
             "_id": f"{project}-{pr_num}-{comment_id}-{os.environ.get('GITHUB_RUN_ID')}",
         }
     ]
@@ -1599,6 +1600,36 @@ def save_merge_record(
         json.dump(data, f)
 
 
+@retries_decorator(rc=[])
+def get_rockset_results(head_sha: str, merge_base: str) -> List[Dict[str, Any]]:
+    query = f"""
+SELECT
+    w.name as workflow_name,
+    j.id,
+    j.name,
+    j.conclusion,
+    j.completed_at,
+    j.html_url,
+    j.head_sha,
+    j.torchci_classification.captures as failure_captures,
+    LENGTH(j.steps) as steps,
+FROM
+    commons.workflow_job j join commons.workflow_run w on w.id = j.run_id
+where
+    j.head_sha in ('{head_sha}','{merge_base}')
+"""
+    try:
+        import rockset  # type: ignore[import]
+
+        res = rockset.RocksetClient(
+            host="api.usw2a1.rockset.com", api_key=os.environ["ROCKSET_API_KEY"]
+        ).sql(query)
+        return cast(List[Dict[str, Any]], res.results)
+    except ModuleNotFoundError:
+        print("Could not use RockSet as rocket dependency is missing")
+        return []
+
+
 @retries_decorator()
 def get_drci_classifications(pr_num: int, project: str = "pytorch") -> Any:
     """
@@ -2036,7 +2067,7 @@ def categorize_checks(
     pending_checks: List[Tuple[str, Optional[str], Optional[int]]] = []
     failed_checks: List[Tuple[str, Optional[str], Optional[int]]] = []
 
-    # failed_checks_categorization is used to keep track of all ignorable failures when saving the merge record on s3
+    # failed_checks_categorization is used to keep track of all ignorable failures when saving the merge record on Rockset
     failed_checks_categorization: Dict[str, List[Any]] = defaultdict(list)
 
     # If required_checks is not set or empty, consider all names are relevant
@@ -2095,7 +2126,7 @@ def categorize_checks(
     ):
         failed_checks = failed_checks + flaky_or_broken_trunk
 
-    # The list of failed_checks_categorization is returned so that it can be saved into the s3 merge record
+    # The list of failed_checks_categorization is returned so that it can be saved into the Rockset merge record
     return (pending_checks, failed_checks, failed_checks_categorization)
 
 
@@ -2379,7 +2410,7 @@ def main() -> None:
         handle_exception(e)
 
         if args.comment_id and args.pr_num:
-            # Finally, upload the record to s3, we don't have access to the
+            # Finally, upload the record to Rockset, we don't have access to the
             # list of pending and failed checks here, but they are not really
             # needed at the moment
             save_merge_record(
@@ -2402,7 +2433,7 @@ def main() -> None:
                 error=str(e),
             )
         else:
-            print("Missing comment ID or PR number, couldn't upload to s3")
+            print("Missing comment ID or PR number, couldn't upload to Rockset")
     finally:
         if not args.check_mergeability:
             gh_remove_label(

.github/scripts/update_runner_determinator.py

@@ -1,31 +0,0 @@
-#!/usr/bin/env python3
-
-import re
-
-
-# Read the contents of runner_determinator.py
-with open(".github/scripts/runner_determinator.py") as script_file:
-    script_content = script_file.read()
-
-# Indent the script content by 10 spaces to match destination indentation
-indented_script_content = "\n".join(
-    [" " * 10 + line if line else line for line in script_content.splitlines()]
-)
-
-# Read the contents of _runner-determinator.yml
-with open(".github/workflows/_runner-determinator.yml") as yml_file:
-    yml_content = yml_file.read()
-
-# Replace the content between the markers
-new_yml_content = re.sub(
-    r"(cat <<EOF > runner_determinator.py\n)(.*?)(\n\s+EOF)",
-    lambda match: match.group(1) + indented_script_content + match.group(3),
-    yml_content,
-    flags=re.DOTALL,
-)
-
-# Save the modified content back to _runner-determinator.yml
-with open(".github/workflows/_runner-determinator.yml", "w") as yml_file:
-    yml_file.write(new_yml_content)
-
-print("Updated _runner-determinator.yml with the contents of runner_determinator.py")

.github/templates/linux_binary_build_workflow.yml.j2

@@ -68,7 +68,6 @@ jobs:
     needs: get-label-type
     with:!{{ upload.binary_env_as_input(config) }}
       {%- if "aarch64" in build_environment %}
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
       {%- elif "s390x" in build_environment %}
@@ -103,7 +102,6 @@ jobs:
       build_name: !{{ config["build_name"] }}
       build_environment: !{{ build_environment }}
       {%- if "aarch64" in build_environment %}
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
       {%- elif "s390x" in build_environment %}

.github/workflows/_bazel-build-test.yml

@@ -91,14 +91,14 @@ jobs:
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
-      - name: Check if in a container runner
+      - name: Check if in a ARC runner
         shell: bash
-        id: check_container_runner
-        run: echo "IN_CONTAINER_RUNNER=$(if [ -f /.inarc ] || [ -f /.incontainer ]; then echo true ; else echo false; fi)" >> "$GITHUB_OUTPUT"
+        id: check_arc_runner
+        run: echo "IN_ARC_RUNNER=$([ -f /.inarc ] && echo true || echo false)" >> "$GITHUB_OUTPUT"
 
       - name: Install nvidia driver, nvidia-docker runtime, set GPU_FLAG
         uses: pytorch/test-infra/.github/actions/setup-nvidia@main
-        if: ${{ inputs.cuda-version != 'cpu' && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' }}
+        if: ${{ inputs.cuda-version != 'cpu' && steps.check_arc_runner.outputs.IN_ARC_RUNNER == 'false' }}
 
       - name: Output disk space left
         run: |

.github/workflows/_linux-test.yml

@@ -114,32 +114,22 @@ jobs:
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
-      - name: Check if in a container runner
+      - name: Check if in a ARC runner
         shell: bash
-        id: check_container_runner
-        run: echo "IN_CONTAINER_RUNNER=$(if [ -f /.inarc ] || [ -f /.incontainer ]; then echo true ; else echo false; fi)" >> "$GITHUB_OUTPUT"
+        id: check_arc_runner
+        run: echo "IN_ARC_RUNNER=$([ -f /.inarc ] && echo true || echo false)" >> "$GITHUB_OUTPUT"
 
       - name: Install nvidia driver, nvidia-docker runtime, set GPU_FLAG
         id: install-nvidia-driver
         uses: pytorch/test-infra/.github/actions/setup-nvidia@main
-        if: ${{ contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu') && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' }}
-
-      - name: Setup GPU_FLAG for docker run
-        id: setup-gpu-flag
-        run: echo "GPU_FLAG=--gpus all -e NVIDIA_DRIVER_CAPABILITIES=all" >> "${GITHUB_ENV}"
-        if: ${{ contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu') && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'true' }}
-
-      - name: Setup SCCACHE_SERVER_PORT environment for docker run when on container
-        id: setup-sscache-port-flag
-        run: echo "SCCACHE_SERVER_PORT_DOCKER_FLAG=-e SCCACHE_SERVER_PORT=$((RUNNER_UID + 4226))" >> "${GITHUB_ENV}"
-        if: ${{ steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'true' }}
+        if: ${{ contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu') && steps.check_arc_runner.outputs.IN_ARC_RUNNER == 'false' }}
 
       - name: Lock NVIDIA A100 40GB Frequency
         run: |
           sudo nvidia-smi -pm 1
           sudo nvidia-smi -ac 1215,1410
           nvidia-smi
-        if: ${{ contains(matrix.runner, 'a100') && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' }}
+        if: contains(matrix.runner, 'a100')
 
       - name: Start monitoring script
         id: monitor-script
@@ -218,7 +208,6 @@ jobs:
           NO_TD: ${{ steps.keep-going.outputs.ci-no-td }}
           TD_DISTRIBUTED: ${{ steps.keep-going.outputs.ci-td-distributed }}
           SCCACHE_BUCKET: ossci-compiler-cache-circleci-v2
-          SCCACHE_REGION: us-east-1
           SCCACHE_S3_KEY_PREFIX: ${{ github.workflow }}
           SHM_SIZE: ${{ contains(inputs.build-environment, 'cuda') && '2g' || '1g' }}
           DOCKER_IMAGE: ${{ inputs.docker-image }}
@@ -229,7 +218,6 @@ jobs:
           DASHBOARD_TAG: ${{ inputs.dashboard-tag }}
           HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
           SCRIBE_GRAPHQL_ACCESS_TOKEN: ${{ secrets.SCRIBE_GRAPHQL_ACCESS_TOKEN }}
-          IS_A100_RUNNER: ${{ contains(matrix.runner, 'a100') && '1' || '0' }}
 
         run: |
           set -x
@@ -248,7 +236,6 @@ jobs:
           # shellcheck disable=SC2086,SC2090
           container_name=$(docker run \
             ${GPU_FLAG:-} \
-            ${SCCACHE_SERVER_PORT_DOCKER_FLAG:-} \
             -e BUILD_ENVIRONMENT \
             -e PR_NUMBER \
             -e GITHUB_ACTIONS \
@@ -278,7 +265,6 @@ jobs:
             -e PR_LABELS \
             -e MAX_JOBS="$(nproc --ignore=2)" \
             -e SCCACHE_BUCKET \
-            -e SCCACHE_REGION \
             -e SCCACHE_S3_KEY_PREFIX \
             -e XLA_CUDA \
             -e XLA_CLANG_CACHE_S3_BUCKET_NAME \
@@ -288,7 +274,6 @@ jobs:
             -e HUGGING_FACE_HUB_TOKEN \
             -e SCRIBE_GRAPHQL_ACCESS_TOKEN \
             -e DASHBOARD_TAG \
-            -e IS_A100_RUNNER \
             --env-file="/tmp/github_env_${GITHUB_RUN_ID}" \
             --security-opt seccomp=unconfined \
             --cap-add=SYS_PTRACE \
@@ -358,7 +343,7 @@ jobs:
 
       - name: Teardown Linux
         uses: pytorch/test-infra/.github/actions/teardown-linux@main
-        if: always() && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false'
+        if: always()
 
       # NB: We are currently having an intermittent GPU-related issue on G5 runners with
       # A10G GPU. Once this happens, trying to reset the GPU as done in setup-nvidia does

.github/workflows/_runner-determinator.yml

@@ -59,10 +59,6 @@ jobs:
           cat <<EOF > runner_determinator.py
           # flake8: noqa: G004
 
-          # Note: Copies of this script in runner_determinator.py and _runner-determinator.yml
-          #       must be kept in sync. You can do it easily by running the following command:
-          #           python .github/scripts/update_runner_determinator.py
-
           """
           This runner determinator is used to determine which set of runners to run a
           GitHub job on. It uses the first comment of a GitHub issue (by default
@@ -142,9 +138,6 @@ jobs:
               rollout_perc: float = (
                   0  # Percentage of workflows to experiment on when user is not opted-in.
               )
-              all_branches: bool = (
-                  False  # If True, the experiment is also enabled on the exception branches
-              )
 
               # Add more fields as needed
 
@@ -278,7 +271,7 @@ jobs:
 
           def is_exception_branch(branch: str) -> bool:
               """
-              Branches that get opted out of experiments by default, until they're explicitly enabled.
+              Branches that get opted out of all experiments and should always use Meta runners
               """
               return branch.split("/")[0] in {"main", "nightly", "release", "landchecks"}
 
@@ -404,10 +397,7 @@ jobs:
 
 
           def get_runner_prefix(
-              rollout_state: str,
-              workflow_requestors: Iterable[str],
-              branch: str,
-              is_canary: bool = False,
+              rollout_state: str, workflow_requestors: Iterable[str], is_canary: bool = False
           ) -> str:
               settings = parse_settings(rollout_state)
               user_optins = parse_users(rollout_state)
@@ -417,12 +407,6 @@ jobs:
               for experiment_name, experiment_settings in settings.experiments.items():
                   enabled = False
 
-                  if not experiment_settings.all_branches and is_exception_branch(branch):
-                      log.info(
-                          f"Branch {branch} is an exception branch. Not enabling experiment {experiment_name}."
-                      )
-                      continue
-
                   # Is any workflow_requestor opted in to this experiment?
                   opted_in_users = [
                       requestor
@@ -482,34 +466,35 @@ jobs:
           def main() -> None:
               args = parse_args()
 
-              runner_label_prefix = DEFAULT_LABEL_PREFIX
-
-              try:
-                  rollout_state = get_rollout_state_from_issue(
-                      args.github_token, args.github_issue_repo, args.github_issue
+              if args.github_ref_type == "branch" and is_exception_branch(args.github_branch):
+                  log.info(
+                      f"Exception branch: '{args.github_branch}', using Meta runners and no experiments."
                   )
+                  runner_label_prefix = DEFAULT_LABEL_PREFIX
+              else:
+                  try:
+                      rollout_state = get_rollout_state_from_issue(
+                          args.github_token, args.github_issue_repo, args.github_issue
+                      )
 
-                  username = get_potential_pr_author(
-                      args.github_token,
-                      args.github_repo,
-                      args.github_actor,
-                      args.github_ref_type,
-                      args.github_branch,
-                  )
+                      username = get_potential_pr_author(
+                          args.github_token,
+                          args.github_repo,
+                          args.github_actor,
+                          args.github_ref_type,
+                          args.github_branch,
+                      )
 
-                  is_canary = args.github_repo == "pytorch/pytorch-canary"
+                      is_canary = args.github_repo == "pytorch/pytorch-canary"
 
-                  runner_label_prefix = get_runner_prefix(
-                      rollout_state,
-                      (args.github_issue_owner, username),
-                      args.github_branch,
-                      is_canary,
-                  )
+                      runner_label_prefix = get_runner_prefix(
+                          rollout_state, (args.github_issue_owner, username), is_canary
+                      )
 
-              except Exception as e:
-                  log.error(
-                      f"Failed to get issue. Defaulting to Meta runners and no experiments. Exception: {e}"
-                  )
+                  except Exception as e:
+                      log.error(
+                          f"Failed to get issue. Defaulting to Meta runners and no experiments. Exception: {e}"
+                      )
 
               set_github_output(GH_OUTPUT_KEY_LABEL_TYPE, runner_label_prefix)
 

.github/workflows/_win-test.yml

@@ -189,7 +189,7 @@ jobs:
         run: |
           pushd "${PYTORCH_FINAL_PACKAGE_DIR}"
           # shellcheck disable=SC2046,SC2102
-          python3 -mpip install $(echo *.whl)[opt-einsum,optree] optree==0.13.0
+          python3 -mpip install $(echo *.whl)[opt-einsum,optree] optree==0.12.1
           popd
 
           .ci/pytorch/win-test.sh

.github/workflows/build-triton-wheel.yml

@@ -43,7 +43,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        py_vers: [ "3.9", "3.10", "3.11", "3.12" ]
+        py_vers: [ "3.8", "3.9", "3.10", "3.11", "3.12" ]
         device: ["cuda", "rocm", "xpu"]
         include:
           - device: "rocm"
@@ -91,6 +91,9 @@ jobs:
 
           # Determine python executable for given version
           case $PY_VERS in
+          3.8)
+            PYTHON_EXECUTABLE=/opt/python/cp38-cp38/bin/python
+            ;;
           3.9)
             PYTHON_EXECUTABLE=/opt/python/cp39-cp39/bin/python
             ;;
@@ -211,7 +214,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        py_vers: [ "3.9", "3.10", "3.11", "3.12" ]
+        py_vers: [ "3.8", "3.9", "3.10", "3.11", "3.12" ]
     timeout-minutes: 40
     env:
       DOCKER_IMAGE: pytorch/conda-builder:cpu

.github/workflows/docker-builds.yml

@@ -67,7 +67,6 @@ jobs:
           pytorch-linux-jammy-py3.12-halide,
           pytorch-linux-jammy-xpu-2024.0-py3,
           pytorch-linux-jammy-py3-clang15-asan,
-          pytorch-linux-jammy-py3-clang18-asan,
           pytorch-linux-focal-py3-clang10-onnx,
           pytorch-linux-focal-linter,
           pytorch-linux-jammy-cuda11.8-cudnn9-py3.9-linter,
@@ -79,9 +78,7 @@ jobs:
           - docker-image-name: pytorch-linux-jammy-aarch64-py3.10-gcc11-inductor-benchmarks
             runner: linux.arm64.m7g.4xlarge
             timeout-minutes: 600
-    # Docker uploads fail from LF runners, see https://github.com/pytorch/pytorch/pull/137358
-    # runs-on: "${{ needs.get-label-type.outputs.label-type }}${{ matrix.runner }}"
-    runs-on: "${{ matrix.runner }}"
+    runs-on: "${{ needs.get-label-type.outputs.label-type }}${{ matrix.runner }}"
     env:
       DOCKER_IMAGE_BASE: 308535385114.dkr.ecr.us-east-1.amazonaws.com/pytorch/${{ matrix.docker-image-name }}
     steps:

.github/workflows/generated-linux-aarch64-binary-manywheel-nightly.yml

@@ -60,7 +60,6 @@ jobs:
       DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
       use_split_build: False
       DESIRED_PYTHON: "3.9"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_9-cpu-aarch64
@@ -87,7 +86,6 @@ jobs:
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
     secrets:
@@ -132,7 +130,6 @@ jobs:
       DESIRED_DEVTOOLSET: cxx11-abi
       use_split_build: False
       DESIRED_PYTHON: "3.9"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_9-cuda-aarch64
@@ -180,7 +177,6 @@ jobs:
       DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
       use_split_build: False
       DESIRED_PYTHON: "3.10"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_10-cpu-aarch64
@@ -207,7 +203,6 @@ jobs:
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
     secrets:
@@ -252,7 +247,6 @@ jobs:
       DESIRED_DEVTOOLSET: cxx11-abi
       use_split_build: False
       DESIRED_PYTHON: "3.10"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_10-cuda-aarch64
@@ -300,7 +294,6 @@ jobs:
       DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
       use_split_build: False
       DESIRED_PYTHON: "3.11"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_11-cpu-aarch64
@@ -327,7 +320,6 @@ jobs:
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
     secrets:
@@ -372,7 +364,6 @@ jobs:
       DESIRED_DEVTOOLSET: cxx11-abi
       use_split_build: False
       DESIRED_PYTHON: "3.11"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_11-cuda-aarch64
@@ -420,7 +411,6 @@ jobs:
       DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
       use_split_build: False
       DESIRED_PYTHON: "3.12"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_12-cpu-aarch64
@@ -447,7 +437,6 @@ jobs:
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
     secrets:
@@ -492,7 +481,6 @@ jobs:
       DESIRED_DEVTOOLSET: cxx11-abi
       use_split_build: False
       DESIRED_PYTHON: "3.12"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_12-cuda-aarch64

.github/workflows/generated-linux-binary-manywheel-nightly.yml

@@ -3324,353 +3324,3 @@ jobs:
       conda-pytorchbot-token: ${{ secrets.CONDA_PYTORCHBOT_TOKEN }}
       conda-pytorchbot-token-test: ${{ secrets.CONDA_PYTORCHBOT_TOKEN_TEST }}
     uses: ./.github/workflows/_binary-upload.yml
-
-  manywheel-py3_13t-cpu-build:
-    if: ${{ github.repository_owner == 'pytorch' }}
-    uses: ./.github/workflows/_binary-build-linux.yml
-    needs: get-label-type
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cpu
-      GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
-      use_split_build: False
-      DESIRED_PYTHON: "3.13t"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build_name: manywheel-py3_13t-cpu
-      build_environment: linux-binary-manywheel
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cpu-test:  # Testing
-    if: ${{ github.repository_owner == 'pytorch' }}
-    needs:
-      - manywheel-py3_13t-cpu-build
-      - get-label-type
-    uses: ./.github/workflows/_binary-test-linux.yml
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cpu
-      GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
-      use_split_build: False
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cpu
-      build_environment: linux-binary-manywheel
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.4xlarge
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cpu-upload:  # Uploading
-    if: ${{ github.repository_owner == 'pytorch' }}
-    permissions:
-      id-token: write
-      contents: read
-    needs: manywheel-py3_13t-cpu-test
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cpu
-      GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
-      use_split_build: False
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cpu
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-      conda-pytorchbot-token: ${{ secrets.CONDA_PYTORCHBOT_TOKEN }}
-      conda-pytorchbot-token-test: ${{ secrets.CONDA_PYTORCHBOT_TOKEN_TEST }}
-    uses: ./.github/workflows/_binary-upload.yml
-
-  manywheel-py3_13t-cpu-cxx11-abi-build:
-    if: ${{ github.repository_owner == 'pytorch' }}
-    uses: ./.github/workflows/_binary-build-linux.yml
-    needs: get-label-type
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cpu-cxx11-abi
-      GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
-      DESIRED_DEVTOOLSET: cxx11-abi
-      use_split_build: False
-      DESIRED_PYTHON: "3.13t"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build_name: manywheel-py3_13t-cpu-cxx11-abi
-      build_environment: linux-binary-manywheel
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cpu-cxx11-abi-test:  # Testing
-    if: ${{ github.repository_owner == 'pytorch' }}
-    needs:
-      - manywheel-py3_13t-cpu-cxx11-abi-build
-      - get-label-type
-    uses: ./.github/workflows/_binary-test-linux.yml
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cpu-cxx11-abi
-      GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
-      DESIRED_DEVTOOLSET: cxx11-abi
-      use_split_build: False
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cpu-cxx11-abi
-      build_environment: linux-binary-manywheel
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.4xlarge
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cpu-cxx11-abi-upload:  # Uploading
-    if: ${{ github.repository_owner == 'pytorch' }}
-    permissions:
-      id-token: write
-      contents: read
-    needs: manywheel-py3_13t-cpu-cxx11-abi-test
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cpu-cxx11-abi
-      GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
-      DESIRED_DEVTOOLSET: cxx11-abi
-      use_split_build: False
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cpu-cxx11-abi
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-      conda-pytorchbot-token: ${{ secrets.CONDA_PYTORCHBOT_TOKEN }}
-      conda-pytorchbot-token-test: ${{ secrets.CONDA_PYTORCHBOT_TOKEN_TEST }}
-    uses: ./.github/workflows/_binary-upload.yml
-
-  manywheel-py3_13t-cuda11_8-build:
-    if: ${{ github.repository_owner == 'pytorch' }}
-    uses: ./.github/workflows/_binary-build-linux.yml
-    needs: get-label-type
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu118
-      GPU_ARCH_VERSION: 11.8
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-main
-      use_split_build: False
-      DESIRED_PYTHON: "3.13t"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build_name: manywheel-py3_13t-cuda11_8
-      build_environment: linux-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu11==11.8.89; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu11==11.8.89; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu11==11.8.87; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu11==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu11==11.11.3.6; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu11==10.9.0.58; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu11==10.3.0.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu11==11.4.1.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu11==11.7.5.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu11==2.21.5; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu11==11.8.86; platform_system == 'Linux' and platform_machine == 'x86_64'
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cuda11_8-test:  # Testing
-    if: ${{ github.repository_owner == 'pytorch' }}
-    needs:
-      - manywheel-py3_13t-cuda11_8-build
-      - get-label-type
-    uses: ./.github/workflows/_binary-test-linux.yml
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu118
-      GPU_ARCH_VERSION: 11.8
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-main
-      use_split_build: False
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cuda11_8
-      build_environment: linux-binary-manywheel
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.4xlarge.nvidia.gpu
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cuda11_8-upload:  # Uploading
-    if: ${{ github.repository_owner == 'pytorch' }}
-    permissions:
-      id-token: write
-      contents: read
-    needs: manywheel-py3_13t-cuda11_8-test
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu118
-      GPU_ARCH_VERSION: 11.8
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-main
-      use_split_build: False
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cuda11_8
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-      conda-pytorchbot-token: ${{ secrets.CONDA_PYTORCHBOT_TOKEN }}
-      conda-pytorchbot-token-test: ${{ secrets.CONDA_PYTORCHBOT_TOKEN_TEST }}
-    uses: ./.github/workflows/_binary-upload.yml
-
-  manywheel-py3_13t-cuda12_1-build:
-    if: ${{ github.repository_owner == 'pytorch' }}
-    uses: ./.github/workflows/_binary-build-linux.yml
-    needs: get-label-type
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu121
-      GPU_ARCH_VERSION: 12.1
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-main
-      use_split_build: False
-      DESIRED_PYTHON: "3.13t"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build_name: manywheel-py3_13t-cuda12_1
-      build_environment: linux-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.21.5; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64'
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cuda12_1-test:  # Testing
-    if: ${{ github.repository_owner == 'pytorch' }}
-    needs:
-      - manywheel-py3_13t-cuda12_1-build
-      - get-label-type
-    uses: ./.github/workflows/_binary-test-linux.yml
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu121
-      GPU_ARCH_VERSION: 12.1
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-main
-      use_split_build: False
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cuda12_1
-      build_environment: linux-binary-manywheel
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.4xlarge.nvidia.gpu
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cuda12_1-upload:  # Uploading
-    if: ${{ github.repository_owner == 'pytorch' }}
-    permissions:
-      id-token: write
-      contents: read
-    needs: manywheel-py3_13t-cuda12_1-test
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu121
-      GPU_ARCH_VERSION: 12.1
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-main
-      use_split_build: False
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cuda12_1
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-      conda-pytorchbot-token: ${{ secrets.CONDA_PYTORCHBOT_TOKEN }}
-      conda-pytorchbot-token-test: ${{ secrets.CONDA_PYTORCHBOT_TOKEN_TEST }}
-    uses: ./.github/workflows/_binary-upload.yml
-
-  manywheel-py3_13t-cuda12_4-build:
-    if: ${{ github.repository_owner == 'pytorch' }}
-    uses: ./.github/workflows/_binary-build-linux.yml
-    needs: get-label-type
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu124
-      GPU_ARCH_VERSION: 12.4
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-main
-      use_split_build: False
-      DESIRED_PYTHON: "3.13t"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build_name: manywheel-py3_13t-cuda12_4
-      build_environment: linux-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.21.5; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cuda12_4-test:  # Testing
-    if: ${{ github.repository_owner == 'pytorch' }}
-    needs:
-      - manywheel-py3_13t-cuda12_4-build
-      - get-label-type
-    uses: ./.github/workflows/_binary-test-linux.yml
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu124
-      GPU_ARCH_VERSION: 12.4
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-main
-      use_split_build: False
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cuda12_4
-      build_environment: linux-binary-manywheel
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.4xlarge.nvidia.gpu
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cuda12_4-upload:  # Uploading
-    if: ${{ github.repository_owner == 'pytorch' }}
-    permissions:
-      id-token: write
-      contents: read
-    needs: manywheel-py3_13t-cuda12_4-test
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu124
-      GPU_ARCH_VERSION: 12.4
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-main
-      use_split_build: False
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cuda12_4
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-      conda-pytorchbot-token: ${{ secrets.CONDA_PYTORCHBOT_TOKEN }}
-      conda-pytorchbot-token-test: ${{ secrets.CONDA_PYTORCHBOT_TOKEN_TEST }}
-    uses: ./.github/workflows/_binary-upload.yml

.github/workflows/generated-linux-binary-manywheel-split-nightly.yml

@@ -1514,283 +1514,3 @@ jobs:
       conda-pytorchbot-token: ${{ secrets.CONDA_PYTORCHBOT_TOKEN }}
       conda-pytorchbot-token-test: ${{ secrets.CONDA_PYTORCHBOT_TOKEN_TEST }}
     uses: ./.github/workflows/_binary-upload.yml
-
-  manywheel-py3_13t-cuda11_8-build:
-    if: ${{ github.repository_owner == 'pytorch' }}
-    uses: ./.github/workflows/_binary-build-linux.yml
-    needs: get-label-type
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu118
-      GPU_ARCH_VERSION: 11.8
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-main
-      use_split_build: True
-      DESIRED_PYTHON: "3.13t"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build_name: manywheel-py3_13t-cuda11_8
-      build_environment: linux-binary-manywheel-split
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu11==11.8.89; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu11==11.8.89; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu11==11.8.87; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu11==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu11==11.11.3.6; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu11==10.9.0.58; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu11==10.3.0.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu11==11.4.1.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu11==11.7.5.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu11==2.21.5; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu11==11.8.86; platform_system == 'Linux' and platform_machine == 'x86_64'
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cuda11_8-test:  # Testing
-    if: ${{ github.repository_owner == 'pytorch' }}
-    needs:
-      - manywheel-py3_13t-cuda11_8-build
-      - get-label-type
-    uses: ./.github/workflows/_binary-test-linux.yml
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu118
-      GPU_ARCH_VERSION: 11.8
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-main
-      use_split_build: True
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cuda11_8
-      build_environment: linux-binary-manywheel-split
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.4xlarge.nvidia.gpu
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cuda11_8-upload:  # Uploading
-    if: ${{ github.repository_owner == 'pytorch' }}
-    permissions:
-      id-token: write
-      contents: read
-    needs: manywheel-py3_13t-cuda11_8-test
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu118
-      GPU_ARCH_VERSION: 11.8
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-main
-      use_split_build: True
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cuda11_8
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-      conda-pytorchbot-token: ${{ secrets.CONDA_PYTORCHBOT_TOKEN }}
-      conda-pytorchbot-token-test: ${{ secrets.CONDA_PYTORCHBOT_TOKEN_TEST }}
-    uses: ./.github/workflows/_binary-upload.yml
-
-  manywheel-py3_13t-cuda12_1-build:
-    if: ${{ github.repository_owner == 'pytorch' }}
-    uses: ./.github/workflows/_binary-build-linux.yml
-    needs: get-label-type
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu121
-      GPU_ARCH_VERSION: 12.1
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-main
-      use_split_build: True
-      DESIRED_PYTHON: "3.13t"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build_name: manywheel-py3_13t-cuda12_1
-      build_environment: linux-binary-manywheel-split
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.21.5; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64'
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cuda12_1-test:  # Testing
-    if: ${{ github.repository_owner == 'pytorch' }}
-    needs:
-      - manywheel-py3_13t-cuda12_1-build
-      - get-label-type
-    uses: ./.github/workflows/_binary-test-linux.yml
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu121
-      GPU_ARCH_VERSION: 12.1
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-main
-      use_split_build: True
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cuda12_1
-      build_environment: linux-binary-manywheel-split
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.4xlarge.nvidia.gpu
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cuda12_1-upload:  # Uploading
-    if: ${{ github.repository_owner == 'pytorch' }}
-    permissions:
-      id-token: write
-      contents: read
-    needs: manywheel-py3_13t-cuda12_1-test
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu121
-      GPU_ARCH_VERSION: 12.1
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-main
-      use_split_build: True
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cuda12_1
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-      conda-pytorchbot-token: ${{ secrets.CONDA_PYTORCHBOT_TOKEN }}
-      conda-pytorchbot-token-test: ${{ secrets.CONDA_PYTORCHBOT_TOKEN_TEST }}
-    uses: ./.github/workflows/_binary-upload.yml
-
-  manywheel-py3_13t-cuda12_4-build:
-    if: ${{ github.repository_owner == 'pytorch' }}
-    uses: ./.github/workflows/_binary-build-linux.yml
-    needs: get-label-type
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu124
-      GPU_ARCH_VERSION: 12.4
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-main
-      use_split_build: True
-      DESIRED_PYTHON: "3.13t"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build_name: manywheel-py3_13t-cuda12_4
-      build_environment: linux-binary-manywheel-split
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.21.5; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cuda12_4-test:  # Testing
-    if: ${{ github.repository_owner == 'pytorch' }}
-    needs:
-      - manywheel-py3_13t-cuda12_4-build
-      - get-label-type
-    uses: ./.github/workflows/_binary-test-linux.yml
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu124
-      GPU_ARCH_VERSION: 12.4
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-main
-      use_split_build: True
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cuda12_4
-      build_environment: linux-binary-manywheel-split
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.4xlarge.nvidia.gpu
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cuda12_4-upload:  # Uploading
-    if: ${{ github.repository_owner == 'pytorch' }}
-    permissions:
-      id-token: write
-      contents: read
-    needs: manywheel-py3_13t-cuda12_4-test
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu124
-      GPU_ARCH_VERSION: 12.4
-      GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-main
-      use_split_build: True
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cuda12_4
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-      conda-pytorchbot-token: ${{ secrets.CONDA_PYTORCHBOT_TOKEN }}
-      conda-pytorchbot-token-test: ${{ secrets.CONDA_PYTORCHBOT_TOKEN_TEST }}
-    uses: ./.github/workflows/_binary-upload.yml
-
-  manywheel-py3_13t-cpu-build:
-    if: ${{ github.repository_owner == 'pytorch' }}
-    uses: ./.github/workflows/_binary-build-linux.yml
-    needs: get-label-type
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cpu
-      GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
-      use_split_build: True
-      DESIRED_PYTHON: "3.13t"
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build_name: manywheel-py3_13t-cpu
-      build_environment: linux-binary-manywheel-split
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cpu-test:  # Testing
-    if: ${{ github.repository_owner == 'pytorch' }}
-    needs:
-      - manywheel-py3_13t-cpu-build
-      - get-label-type
-    uses: ./.github/workflows/_binary-test-linux.yml
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cpu
-      GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
-      use_split_build: True
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cpu
-      build_environment: linux-binary-manywheel-split
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.4xlarge
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cpu-upload:  # Uploading
-    if: ${{ github.repository_owner == 'pytorch' }}
-    permissions:
-      id-token: write
-      contents: read
-    needs: manywheel-py3_13t-cpu-test
-    with:
-      PYTORCH_ROOT: /pytorch
-      BUILDER_ROOT: /builder
-      PACKAGE_TYPE: manywheel
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cpu
-      GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
-      use_split_build: True
-      DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cpu
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-      conda-pytorchbot-token: ${{ secrets.CONDA_PYTORCHBOT_TOKEN }}
-      conda-pytorchbot-token-test: ${{ secrets.CONDA_PYTORCHBOT_TOKEN_TEST }}
-    uses: ./.github/workflows/_binary-upload.yml

.github/workflows/inductor.yml

@@ -120,28 +120,6 @@ jobs:
       test-matrix: ${{ needs.linux-jammy-cpu-py3_12-inductor-halide-build.outputs.test-matrix }}
     secrets: inherit
 
-  linux-jammy-cpu-py3_12-inductor-triton-cpu-build:
-    name: linux-jammy-cpu-py3.12-gcc11-inductor-triton-cpu
-    uses: ./.github/workflows/_linux-build.yml
-    needs: get-label-type
-    with:
-      build-environment: linux-jammy-py3.12-gcc11
-      docker-image-name: pytorch-linux-jammy-py3.12-triton-cpu
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      test-matrix: |
-        { include: [
-          { config: "inductor-triton-cpu", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.12xlarge" },
-        ]}
-
-  linux-jammy-cpu-py3_12-inductor-triton-cpu-test:
-    name: linux-jammy-cpu-py3.12-gcc11-inductor-triton-cpu
-    uses: ./.github/workflows/_linux-test.yml
-    needs: linux-jammy-cpu-py3_12-inductor-triton-cpu-build
-    with:
-      build-environment: linux-jammy-py3.12-gcc11
-      docker-image: ${{ needs.linux-jammy-cpu-py3_12-inductor-triton-cpu-build.outputs.docker-image }}
-      test-matrix: ${{ needs.linux-jammy-cpu-py3_12-inductor-triton-cpu-build.outputs.test-matrix }}
-
   linux-focal-cuda12_4-py3_10-gcc9-inductor-build:
     # Should be synced with the one in inductor-periodic.yml but this only runs inductor_timm
     name: cuda12.4-py3.10-gcc9-sm86

.github/workflows/lint-autoformat.yml

@@ -11,6 +11,7 @@ jobs:
       contents: read
       pull-requests: write
     runs-on: lf.linux.2xlarge
+    continue-on-error: true
     if: ${{ github.repository_owner == 'pytorch' }}
     steps:
       - name: Checkout pytorch
@@ -30,12 +31,10 @@ jobs:
           bash .github/scripts/lintrunner.sh
       - name: Check for changes
         id: git-check
-        continue-on-error: true
         run: |
           git diff --exit-code || echo "changes=true" >> "$GITHUB_OUTPUT"
       - name: Suggest changes
         if: steps.git-check.outputs.changes == 'true'
-        continue-on-error: true
         uses: parkerbxyz/suggest-changes@v1
         with:
           comment: "Please commit the suggested changes from pytorch's linter."

.github/workflows/lint.yml

@@ -223,7 +223,6 @@ jobs:
           cache: pip
       - name: Install dependencies
         run: |
-          python3 -m pip install --upgrade pip
           pip install pytest-rerunfailures==11.1.* pytest-flakefinder==1.1.* pytest-xdist==3.3.* expecttest==0.2.* fbscribelogger==0.1.* numpy==1.24.*
           pip install torch --pre --index-url https://download.pytorch.org/whl/nightly/cpu/
       - name: Run run_test.py (nonretryable)

.github/workflows/periodic.yml

@@ -57,10 +57,10 @@ jobs:
       docker-image-name: pytorch-linux-focal-cuda12.1-cudnn9-py3-gcc9
       test-matrix: |
         { include: [
-          { config: "nogpu_AVX512", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "nogpu_AVX512", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "nogpu_NO_AVX2", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "nogpu_NO_AVX2", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
+          { config: "nogpu_AVX512", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "nogpu_AVX512", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "nogpu_NO_AVX2", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "nogpu_NO_AVX2", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
           { config: "jit_legacy", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge.nvidia.gpu" },
         ]}
   linux-focal-cuda12_1-py3_10-gcc9-test:
@@ -89,10 +89,10 @@ jobs:
           { config: "default", shard: 3, num_shards: 5, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge.nvidia.gpu" },
           { config: "default", shard: 4, num_shards: 5, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge.nvidia.gpu" },
           { config: "default", shard: 5, num_shards: 5, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge.nvidia.gpu" },
-          { config: "nogpu_AVX512", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "nogpu_AVX512", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "nogpu_NO_AVX2", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "nogpu_NO_AVX2", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
+          { config: "nogpu_AVX512", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "nogpu_AVX512", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "nogpu_NO_AVX2", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "nogpu_NO_AVX2", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
           { config: "jit_legacy", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge.nvidia.gpu" },
         ]}
 
@@ -118,10 +118,9 @@ jobs:
       docker-image-name: pytorch-linux-jammy-py3.9-gcc11
       test-matrix: |
         { include: [
-          { config: "default", shard: 1, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
-          { config: "default", shard: 2, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
-          { config: "default", shard: 3, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
-          { config: "default", shard: 4, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "default", shard: 1, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "default", shard: 2, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "default", shard: 3, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
         ]}
 
   parallelnative-linux-jammy-py3_9-gcc11-test:
@@ -340,10 +339,10 @@ jobs:
       docker-image-name: pytorch-linux-focal-cuda12.1-cudnn9-py3-gcc9
       test-matrix: |
         { include: [
-          { config: "nogpu_AVX512", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "nogpu_AVX512", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "nogpu_NO_AVX2", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "nogpu_NO_AVX2", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
+          { config: "nogpu_AVX512", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "nogpu_AVX512", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "nogpu_NO_AVX2", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "nogpu_NO_AVX2", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
           { config: "jit_legacy", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge.nvidia.gpu" },
         ]}
 

.github/workflows/pull.yml

@@ -185,10 +185,10 @@ jobs:
       docker-image-name: pytorch-linux-focal-py3.9-clang10
       test-matrix: |
         { include: [
-          { config: "default", shard: 1, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "default", shard: 2, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "default", shard: 3, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "default", shard: 4, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
+          { config: "default", shard: 1, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "default", shard: 2, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "default", shard: 3, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "default", shard: 4, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
           { config: "crossref", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
           { config: "crossref", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
           { config: "dynamo", shard: 1, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
@@ -217,10 +217,10 @@ jobs:
       docker-image-name: pytorch-linux-focal-py3.11-clang10
       test-matrix: |
         { include: [
-          { config: "default", shard: 1, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "default", shard: 2, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "default", shard: 3, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "default", shard: 4, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
+          { config: "default", shard: 1, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "default", shard: 2, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "default", shard: 3, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "default", shard: 4, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
           { config: "crossref", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
           { config: "crossref", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
           { config: "dynamo", shard: 1, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
@@ -251,10 +251,10 @@ jobs:
       docker-image-name: pytorch-linux-focal-py3.12-clang10
       test-matrix: |
         { include: [
-          { config: "default", shard: 1, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "default", shard: 2, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "default", shard: 3, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "default", shard: 4, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
+          { config: "default", shard: 1, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "default", shard: 2, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "default", shard: 3, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "default", shard: 4, num_shards: 4, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
           { config: "dynamo", shard: 1, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
           { config: "dynamo", shard: 2, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
           { config: "dynamo", shard: 3, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
@@ -588,9 +588,9 @@ jobs:
       docker-image-name: pytorch-linux-focal-py3.12-clang10
       test-matrix: |
         { include: [
-          { config: "default", shard: 1, num_shards: 3, runner: "linux.4xlarge" },
-          { config: "default", shard: 2, num_shards: 3, runner: "linux.4xlarge" },
-          { config: "default", shard: 3, num_shards: 3, runner: "linux.4xlarge" },
+          { config: "default", shard: 1, num_shards: 3, runner: "linux.2xlarge" },
+          { config: "default", shard: 2, num_shards: 3, runner: "linux.2xlarge" },
+          { config: "default", shard: 3, num_shards: 3, runner: "linux.2xlarge" },
           { config: "dynamo", shard: 1, num_shards: 3, runner: "linux.2xlarge" },
           { config: "dynamo", shard: 2, num_shards: 3, runner: "linux.2xlarge" },
           { config: "dynamo", shard: 3, num_shards: 3, runner: "linux.2xlarge" },

.github/workflows/s390.yml

@@ -1,24 +0,0 @@
-name: s390
-
-on:
-  push:
-    branches:
-      - main
-    tags:
-      - ciflow/s390/*
-  workflow_dispatch:
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }}
-  cancel-in-progress: true
-
-permissions: read-all
-
-jobs:
-  linux-manylinux-2_28-py3-cpu-s390x-build:
-    name: linux-manylinux-2_28-py3-cpu-s390x
-    uses: ./.github/workflows/_linux-build.yml
-    with:
-      build-environment: linux-s390x-binary-manywheel
-      docker-image-name: pytorch/manylinuxs390x-builder:cpu-s390x-main
-      runner: linux.s390x

.github/workflows/trunk.yml

@@ -266,10 +266,10 @@ jobs:
       docker-image-name: pytorch-linux-focal-cuda12.4-cudnn9-py3-gcc9
       test-matrix: |
         { include: [
-          { config: "nogpu_AVX512", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "nogpu_AVX512", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "nogpu_NO_AVX2", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-          { config: "nogpu_NO_AVX2", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
+          { config: "nogpu_AVX512", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "nogpu_AVX512", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "nogpu_NO_AVX2", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
+          { config: "nogpu_NO_AVX2", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
           { config: "jit_legacy", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge.nvidia.gpu" },
           { config: "default", shard: 1, num_shards: 5, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge.nvidia.gpu" },
           { config: "default", shard: 2, num_shards: 5, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge.nvidia.gpu" },
@@ -316,3 +316,11 @@ jobs:
       build-environment: linux-focal-cuda11.8-py3.10-gcc9-experimental-split-build
       docker-image: ${{ needs.linux-focal-cuda11_8-py3_10-gcc9-experimental-split-build.outputs.docker-image }}
       test-matrix: ${{ needs.linux-focal-cuda11_8-py3_10-gcc9-experimental-split-build.outputs.test-matrix }}
+
+  linux-manylinux-2_28-py3-cpu-s390x-build:
+    name: linux-manylinux-2_28-py3-cpu-s390x
+    uses: ./.github/workflows/_linux-build.yml
+    with:
+      build-environment: linux-s390x-binary-manywheel
+      docker-image-name: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      runner: linux.s390x

.github/workflows/trymerge.yml

@@ -28,7 +28,7 @@ jobs:
           check-latest: false
           cache: pip
           architecture: x64
-      - run: pip install pyyaml==6.0
+      - run: pip install pyyaml==6.0 rockset==1.0.3
 
       - name: Setup committer id
         run: |
@@ -43,6 +43,7 @@ jobs:
           COMMENT_ID: ${{ github.event.client_payload.comment_id }}
           REBASE: ${{ github.event.client_payload.rebase }}
           IGNORE_CURRENT: ${{ github.event.client_payload.ignore_current }}
+          ROCKSET_API_KEY: ${{ secrets.ROCKSET_API_KEY }}
           DRCI_BOT_KEY: ${{ secrets.DRCI_BOT_KEY }}
           GITHUB_RUN_ID: ${{ github.run_id }}
         run: |

.lintrunner.toml

@@ -153,7 +153,7 @@ init_command = [
     'junitparser==2.1.1',
     'rich==10.9.0',
     'pyyaml==6.0.1',
-    'optree==0.13.0',
+    'optree==0.12.1',
 ]
 
 [[linter]]
@@ -216,10 +216,6 @@ include_patterns = [
     'torch/csrc/*.cpp',
     'torch/csrc/**/*.h',
     'torch/csrc/**/*.cpp',
-    'torch/csrc/distributed/autograd/**/*.cpp',
-    'torch/csrc/distributed/autograd/**/*.h',
-    'torch/csrc/distributed/rpc/**/*.cpp',
-    'torch/csrc/distributed/rpc/**/*.h',
     'torch/csrc/jit/serialization/*.h',
     'torch/csrc/jit/serialization/*.cpp',
 ]

CMakeLists.txt

@@ -1083,16 +1083,8 @@ if(NOT MSVC)
   append_cxx_flag_if_supported("-Wno-unused-but-set-variable" CMAKE_CXX_FLAGS)
   append_cxx_flag_if_supported("-Wno-maybe-uninitialized" CMAKE_CXX_FLAGS)
   append_cxx_flag_if_supported("-fstandalone-debug" CMAKE_CXX_FLAGS_DEBUG)
-  if(CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64" AND CMAKE_CXX_COMPILER_ID MATCHES "GNU")
-    if(CMAKE_BUILD_TYPE MATCHES Debug)
-      message(Warning "Applying -Og optimization for aarch64 GCC debug build to workaround ICE")
-    endif()
-    string(APPEND CMAKE_CXX_FLAGS_DEBUG " -fno-omit-frame-pointer -Og")
-    string(APPEND CMAKE_LINKER_FLAGS_DEBUG " -fno-omit-frame-pointer -Og")
-  else()
-    string(APPEND CMAKE_CXX_FLAGS_DEBUG " -fno-omit-frame-pointer -O0")
-    string(APPEND CMAKE_LINKER_FLAGS_DEBUG " -fno-omit-frame-pointer -O0")
-  endif()
+  string(APPEND CMAKE_CXX_FLAGS_DEBUG " -fno-omit-frame-pointer -O0")
+  string(APPEND CMAKE_LINKER_FLAGS_DEBUG " -fno-omit-frame-pointer -O0")
   append_cxx_flag_if_supported("-fno-math-errno" CMAKE_CXX_FLAGS)
   append_cxx_flag_if_supported("-fno-trapping-math" CMAKE_CXX_FLAGS)
   append_cxx_flag_if_supported("-Werror=format" CMAKE_CXX_FLAGS)

CODEOWNERS

@@ -121,7 +121,7 @@ torch/profiler/ @aaronenyeshi @sraikund16
 test/functorch/test_aotdispatch.py @ezyang @Chillee
 
 # Dataloader
-torch/utils/data/ @andrewkho @divyanshk
+torch/utils/data/ @andrewkho @gokulavasan
 
 # hipify
 torch/utils/hipify/ @jeffdaily @jithunnair-amd

aten/src/ATen/Context.h

@@ -39,16 +39,25 @@ class TORCH_API Context {
 
   const Generator& defaultGenerator(Device device) {
     c10::DeviceType device_type = device.type();
-    lazyInitDevice(device_type);
-
+    initCUDAIfNeeded(device_type);
+    initHIPIfNeeded(device_type);
     if (device_type == at::kCPU) {
       return at::detail::getDefaultCPUGenerator();
+    } else if (device_type == at::kCUDA) {
+      return at::detail::getCUDAHooks().getDefaultCUDAGenerator(device.index());
+    } else if (device_type == at::kMPS) {
+      return at::detail::getMPSHooks().getDefaultMPSGenerator();
+    } else if (device_type == at::kXPU) {
+      return at::detail::getXPUHooks().getDefaultXPUGenerator(device.index());
+    } else if (device_type == at::kIPU) {
+      return at::detail::getIPUHooks().getDefaultIPUGenerator(device.index());
+    } else if (device_type == at::kPrivateUse1) {
+      return at::detail::getPrivateUse1Hooks().getDefaultGenerator(
+          device.index());
     } else {
-      return getAcceleratorHooksInterface(device_type)
-          .getDefaultGenerator(device.index());
+      AT_ERROR(c10::DeviceTypeName(device_type), " device type not enabled.");
     }
   }
-
   const AcceleratorHooksInterface& getAcceleratorHooksInterface(
       std::optional<c10::DeviceType> opt_device_type = std::nullopt) {
     c10::DeviceType device_type = opt_device_type.has_value()
@@ -71,10 +80,10 @@ class TORCH_API Context {
           c10::DeviceTypeName(device_type), " device type not an accelerator.");
     }
   }
-
   Device getDeviceFromPtr(void* data, c10::DeviceType device_type) {
-    lazyInitDevice(device_type);
-
+    initCUDAIfNeeded(device_type);
+    initHIPIfNeeded(device_type);
+    initXPUIfNeeded(device_type);
     if (device_type == at::kCPU) {
       return c10::DeviceType::CPU;
     } else if (device_type == at::kCUDA) {
@@ -87,7 +96,6 @@ class TORCH_API Context {
       AT_ERROR(c10::DeviceTypeName(device_type), " device type not enabled.");
     }
   }
-
   bool isPinnedPtr(
       const void* data,
       std::optional<c10::DeviceType> device_type = std::nullopt) {
@@ -98,22 +106,13 @@ class TORCH_API Context {
             opt_device_type.value())) { // passed device not an accelerator
       return false;
     }
-    return getAcceleratorHooksInterface(opt_device_type).isPinnedPtr(data);
+    return getAcceleratorHooksInterface(opt_device_type.value())
+        .isPinnedPtr(data);
   }
-
   Allocator* getPinnedMemoryAllocator(
       std::optional<c10::DeviceType> device_type = std::nullopt) {
     return getAcceleratorHooksInterface(device_type).getPinnedMemoryAllocator();
   }
-
-  void lazyInitDevice(c10::DeviceType device_type) {
-    if (device_type != at::kCPU) {
-      c10::call_once(init_[static_cast<int8_t>(device_type)], [&] {
-        getAcceleratorHooksInterface(device_type).init();
-      });
-    }
-  }
-
   static bool hasOpenMP();
   static bool hasMKL();
   static bool hasLAPACK();
@@ -166,6 +165,27 @@ class TORCH_API Context {
   static bool hasMAIA() {
     return c10::impl::hasDeviceGuardImpl(c10::DeviceType::MAIA);
   }
+  // defined in header so that getNonVariableType has ability to inline
+  // call_once check. getNonVariableType is called fairly frequently
+  void lazyInitCUDA() {
+    c10::call_once(thc_init, [&] { detail::getCUDAHooks().initCUDA(); });
+  }
+  void lazyInitHIP() {
+    c10::call_once(thh_init, [&] { detail::getHIPHooks().initHIP(); });
+  }
+  void lazyInitXPU() {
+    c10::call_once(thx_init, [&] { detail::getXPUHooks().initXPU(); });
+  }
+  void lazyInitMTIA() {
+    c10::call_once(th_mtia_init, [&] { detail::getMTIAHooks().initMTIA(); });
+  }
+  void lazyInitPrivateUse1() {
+    c10::call_once(thp_init, [&] {
+      if (isPrivateUse1HooksRegistered()) {
+        at::detail::getPrivateUse1Hooks().initPrivateUse1();
+      }
+    });
+  }
   static const at::cuda::NVRTC& getNVRTC() {
     return detail::getCUDAHooks().nvrtc();
   }
@@ -341,8 +361,27 @@ class TORCH_API Context {
   void setAllowFP16ReductionCPU(bool);
 
  private:
+  void initCUDAIfNeeded(c10::DeviceType p) {
+    if (p == c10::DeviceType::CUDA) {
+      lazyInitCUDA();
+    }
+  }
+  void initHIPIfNeeded(c10::DeviceType p) {
+    if (p == c10::DeviceType::HIP) {
+      lazyInitHIP();
+    }
+  }
+  void initXPUIfNeeded(c10::DeviceType p) {
+    if (p == c10::DeviceType::XPU) {
+      lazyInitXPU();
+    }
+  }
   static bool checkCuBLASConfigDeterministic();
-  std::array<c10::once_flag, at::COMPILE_TIME_MAX_DEVICE_TYPES> init_;
+  c10::once_flag thc_init;
+  c10::once_flag thh_init;
+  c10::once_flag thx_init;
+  c10::once_flag th_mtia_init;
+  c10::once_flag thp_init;
   bool enabled_cudnn = true;
   bool deterministic_cudnn = false;
   bool deterministic_mkldnn = false;
@@ -474,7 +513,7 @@ inline size_t getNumGPUs() {
         "to be CUDA (e.g., when you say CUDA, on a HIP build of ATen, this actually "
         "means HIP.  Rebuild PyTorch with one or the other disabled.");
   } else if (hasCUDA()) {
-    return detail::getCUDAHooks().deviceCount();
+    return detail::getCUDAHooks().getNumGPUs();
   } else if (hasHIP()) {
     return detail::getHIPHooks().getNumGPUs();
   } else {
@@ -511,7 +550,7 @@ inline void manual_seed(uint64_t seed) {
   }
   // NB: Sometimes we build with CUDA, but we don't have any GPUs
   // available. In that case, we must not seed CUDA; it will fail!
-  const auto cuda_num_gpus = detail::getCUDAHooks().deviceCount();
+  const auto cuda_num_gpus = detail::getCUDAHooks().getNumGPUs();
   if (hasCUDA() && cuda_num_gpus > 0) {
     for (const auto i : c10::irange(cuda_num_gpus)) {
       auto cuda_gen = globalContext().defaultGenerator(
@@ -524,7 +563,7 @@ inline void manual_seed(uint64_t seed) {
     }
   }
 
-  const auto xpu_num_gpus = detail::getXPUHooks().deviceCount();
+  const auto xpu_num_gpus = detail::getXPUHooks().getNumGPUs();
   if (hasXPU() && xpu_num_gpus) {
     for (const auto i : c10::irange(xpu_num_gpus)) {
       auto xpu_gen = globalContext().defaultGenerator(

aten/src/ATen/DLConvertor.cpp

@@ -22,6 +22,13 @@ DLDataType getDLDataType(const Tensor& t) {
     case ScalarType::UInt64:
       dtype.code = DLDataTypeCode::kDLUInt;
       break;
+    case ScalarType::Int1:
+    case ScalarType::Int2:
+    case ScalarType::Int3:
+    case ScalarType::Int4:
+    case ScalarType::Int5:
+    case ScalarType::Int6:
+    case ScalarType::Int7:
     case ScalarType::Char:
       dtype.code = DLDataTypeCode::kDLInt;
       break;

aten/src/ATen/Dispatch_v2.h

@@ -112,12 +112,12 @@
 
 // Ensure we never have too many scalar types for the expansion here to
 // support.  To bump this, you must regenerate the macros below.
-static_assert(static_cast<int>(c10::ScalarType::NumOptions) < 45);
+static_assert(static_cast<int>(c10::ScalarType::NumOptions) < 60);
 
 // Python code to regenerate generate code below:
 #if 0
 
-num_args = 45
+num_args = 60
 
 nums = ', '.join(str(i) for i in reversed(range(num_args+1)))
 args = ', '.join(f'_{i}' for i in range(1, num_args+1))
@@ -135,8 +135,8 @@ for i in range(1, num_args+1):
 // Begin generated code
 // clang-format off
 
-#define AT_NUM_ARGS(...) AT_EXPAND(AT_NUM_ARGS_AUX(__VA_ARGS__, 45, 44, 43, 42, 41, 40, 39, 38, 37, 36, 35, 34, 33, 32, 31, 30, 29, 28, 27, 26, 25, 24, 23, 22, 21, 20, 19, 18, 17, 16, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0))
-#define AT_NUM_ARGS_AUX(_1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, N, ...) N
+#define AT_NUM_ARGS(...) AT_EXPAND(AT_NUM_ARGS_AUX(__VA_ARGS__, 60, 59, 58, 57, 56, 55, 54, 53, 52, 51, 50, 49, 48, 47, 46, 45, 44, 43, 42, 41, 40, 39, 38, 37, 36, 35, 34, 33, 32, 31, 30, 29, 28, 27, 26, 25, 24, 23, 22, 21, 20, 19, 18, 17, 16, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0))
+#define AT_NUM_ARGS_AUX(_1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54, _55, _56, _57, _58, _59, _60, N, ...) N
 #define AT_AP1(N, _1) AT_DISPATCH_CASE(_1, N)
 #define AT_AP2(N, _1, _2) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N)
 #define AT_AP3(N, _1, _2, _3) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N)
@@ -182,5 +182,21 @@ for i in range(1, num_args+1):
 #define AT_AP43(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N)
 #define AT_AP44(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N)
 #define AT_AP45(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N)
+#define AT_AP46(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N)
+#define AT_AP47(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N)
+#define AT_AP48(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N)
+#define AT_AP49(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N)
+#define AT_AP50(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N)
+#define AT_AP51(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N)
+#define AT_AP52(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N)
+#define AT_AP53(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N)
+#define AT_AP54(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N) AT_DISPATCH_CASE(_54, N)
+#define AT_AP55(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54, _55) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N) AT_DISPATCH_CASE(_54, N) AT_DISPATCH_CASE(_55, N)
+#define AT_AP56(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54, _55, _56) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N) AT_DISPATCH_CASE(_54, N) AT_DISPATCH_CASE(_55, N) AT_DISPATCH_CASE(_56, N)
+#define AT_AP57(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54, _55, _56, _57) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N) AT_DISPATCH_CASE(_54, N) AT_DISPATCH_CASE(_55, N) AT_DISPATCH_CASE(_56, N) AT_DISPATCH_CASE(_57, N)
+#define AT_AP58(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54, _55, _56, _57, _58) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N) AT_DISPATCH_CASE(_54, N) AT_DISPATCH_CASE(_55, N) AT_DISPATCH_CASE(_56, N) AT_DISPATCH_CASE(_57, N) AT_DISPATCH_CASE(_58, N)
+#define AT_AP59(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54, _55, _56, _57, _58, _59) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N) AT_DISPATCH_CASE(_54, N) AT_DISPATCH_CASE(_55, N) AT_DISPATCH_CASE(_56, N) AT_DISPATCH_CASE(_57, N) AT_DISPATCH_CASE(_58, N) AT_DISPATCH_CASE(_59, N)
+#define AT_AP60(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54, _55, _56, _57, _58, _59, _60) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N) AT_DISPATCH_CASE(_54, N) AT_DISPATCH_CASE(_55, N) AT_DISPATCH_CASE(_56, N) AT_DISPATCH_CASE(_57, N) AT_DISPATCH_CASE(_58, N) AT_DISPATCH_CASE(_59, N) AT_DISPATCH_CASE(_60, N)
+
 // End generated code
 // clang-format on

aten/src/ATen/EmptyTensor.cpp

@@ -18,8 +18,6 @@ c10::Allocator* GetCPUAllocatorMaybePinned(bool pin_memory) {
     // To properly support this, see https://github.com/pytorch/pytorch/issues/14560
     if (at::globalContext().hasCUDA()) {
       return at::detail::getCUDAHooks().getPinnedMemoryAllocator();
-    } else if (at::globalContext().hasMTIA()) {
-      return at::detail::getMTIAHooks().getPinnedMemoryAllocator();
     } else if (at::globalContext().hasXPU()) {
       return at::detail::getXPUHooks().getPinnedMemoryAllocator();
     } else if(at::isPrivateUse1HooksRegistered()) {

aten/src/ATen/ExpandUtils.h

@@ -420,15 +420,15 @@ inline c10::MaybeOwned<Tensor> expand_size(
 inline std::vector<Tensor> expand_outplace(TensorList to_expand) {
   // expands a list of Tensors; ignores undefined (null) tensors
   bool first = true;
-  SymDimVector sizes;
+  DimVector sizes;
   for (const auto i : c10::irange(to_expand.size())) {
     if (!to_expand[i].defined()) {
       continue;
     } else if (first) {
-      sizes = to_expand[i].sym_sizes();
+      sizes = to_expand[i].sizes();
       first = false;
     } else {
-      sizes = infer_size_symdimvector(sizes, to_expand[i].sym_sizes());
+      sizes = infer_size_dimvector(sizes, to_expand[i].sizes());
     }
   }
 
@@ -436,10 +436,10 @@ inline std::vector<Tensor> expand_outplace(TensorList to_expand) {
   for (const auto i : c10::irange(to_expand.size())) {
     if (!to_expand[i].defined()) {
       continue;
-    } else if (to_expand[i].sym_sizes().equals(sizes)) {
+    } else if (to_expand[i].sizes().equals(sizes)) {
       result[i] = to_expand[i];
     } else {
-      result[i] = to_expand[i].expand_symint(sizes);
+      result[i] = to_expand[i].expand(sizes);
     }
   }
   return result;

aten/src/ATen/ParallelNative.cpp

@@ -209,8 +209,8 @@ void init_num_threads() {
 }
 
 void set_num_threads(int nthreads) {
-#ifndef C10_MOBILE
   TORCH_CHECK(nthreads > 0, "Expected positive number of threads");
+#ifndef C10_MOBILE
   int no_value = NOT_SET;
   if (!num_intraop_threads.compare_exchange_strong(no_value, nthreads)) {
     // num_intraop_threads either stores a positive integer or CONSUMED,
@@ -229,9 +229,8 @@ void set_num_threads(int nthreads) {
     }
   }
 #else
-  caffe2::PThreadPool* const pool = caffe2::pthreadpool();
+  caffe2::PThreadPool* const pool = caffe2::pthreadpool(nthreads);
   TORCH_INTERNAL_ASSERT(pool, "Invalid thread pool!");
-  pool->set_thread_count(nthreads);
 #endif // C10_MOBILE
 }
 

aten/src/ATen/ScalarOps.cpp

@@ -19,7 +19,7 @@ Tensor& scalar_fill(Tensor& self, const Scalar& value) {
   AT_DISPATCH_V2(
       self.scalar_type(), "fill_out", AT_WRAP([&]() {
         fill_inplace<scalar_t>(self, value);
-      }), kComplexHalf, kHalf, kBool, kBFloat16, AT_EXPAND(AT_ALL_TYPES_AND_COMPLEX), AT_EXPAND(AT_FLOAT8_TYPES), AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES));
+      }), kComplexHalf, kHalf, kBool, kBFloat16, AT_EXPAND(AT_ALL_TYPES_AND_COMPLEX), AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES));
   return self;
 }
 

aten/src/ATen/SparseCsrTensorUtils.h

@@ -144,8 +144,8 @@ class CheckSparseTensorInvariants {
   bool old_state;
 
  public:
-  CheckSparseTensorInvariants(bool state)
-      : old_state(at::globalContext().checkSparseTensorInvariants()) {
+  CheckSparseTensorInvariants(bool state) {
+    old_state = at::globalContext().checkSparseTensorInvariants();
     at::globalContext().setCheckSparseTensorInvariants(state);
   }
 

aten/src/ATen/TensorIndexing.h

@@ -255,9 +255,7 @@ inline Tensor applySelect(
     // the other hand, indexing wraping is valid for all negative int64_t
     // values, as x[INT64_MIN] is the same as x[INT64_MAX]
     TORCH_CHECK_INDEX(
-        size.sym_gt(-1 - index)
-            .sym_and(size.sym_gt(index))
-            .expect_true(__FILE__, __LINE__),
+        size > -1 - index && size > index,
         "index ",
         index,
         " is out of bounds for dimension ",

aten/src/ATen/ThreadLocalState.h

@@ -82,7 +82,7 @@ class TORCH_API ThreadLocalState {
     !defined(BUILD_LITE_INTERPRETER)
   // TLS for autocast dtypes
   std::array<at::ScalarType, at::COMPILE_TIME_MAX_DEVICE_TYPES>
-      autocast_dtypes_{};
+      autocast_dtypes_;
 #endif
 
   friend class ThreadLocalStateGuard;

aten/src/ATen/core/PhiloxRNGEngine.h

@@ -13,6 +13,8 @@
 
 #include <ATen/core/Array.h>
 #include <c10/macros/Macros.h>
+#include <c10/util/Exception.h>
+#include <c10/util/Half.h>
 #include <cmath>
 #include <cstdint>
 

aten/src/ATen/cpu/vec/vec256/vec256_convert.h

@@ -306,10 +306,11 @@ struct VecConvert<float, 1, BFloat16, 1> {
       const VectorizedN<BFloat16, 1>& src) {
     VectorizedN<float, 1> result;
     uint16x8_t u16_8 = vld1q_u16(reinterpret_cast<const uint16_t*>(&src[0]));
+    int32x4_t shift = vdupq_n_s32(16);
     auto u16_low1 = vget_low_u16(u16_8);
     auto u16_high1 = vget_high_u16(u16_8);
-    float32x4_t f32x4_0 = vreinterpretq_f32_u32(vshlq_n_u32(vmovl_u16(u16_low1), 16));
-    float32x4_t f32x4_1 = vreinterpretq_f32_u32(vshlq_n_u32(vmovl_u16(u16_high1), 16));
+    float32x4_t f32x4_0 = vreinterpretq_f32_u32(vshlq_u32(vmovl_u16(u16_low1), shift));
+    float32x4_t f32x4_1 = vreinterpretq_f32_u32(vshlq_u32(vmovl_u16(u16_high1), shift));
     result[0] = {f32x4_0, f32x4_1};
     return result;
   }

aten/src/ATen/cpu/vec/vec256/vec256_qint.h

@@ -75,7 +75,7 @@ inline __m256i pack_saturate_and_clamp<int32_t>(
     int32_t /*min_val*/,
     int32_t /*max_val*/) {
   // This function is for linkage only, will not be used
-  TORCH_CHECK(false, "pack_saturate_and_clamp<int32_t> is not supported");
+  AT_ERROR("pack_saturate_and_clamp<int32_t> is not supported");
 }
 
 template <>

aten/src/ATen/cpu/vec/vec512/vec512_qint.h

@@ -77,7 +77,7 @@ inline __m512i pack_saturate_and_clamp<int32_t>(
     int32_t min_val [[maybe_unused]],
     int32_t max_val [[maybe_unused]]) {
   // This function is for linkage only, will not be used
-  TORCH_CHECK(false, "pack_saturate_and_clamp<int32_t> is not supported");
+  AT_ERROR("pack_saturate_and_clamp<int32_t> is not supported");
   return __m512i{};
 }
 

aten/src/ATen/cuda/CUDAGraph.cpp

@@ -125,7 +125,7 @@ void CUDAGraph::capture_begin(MempoolId_t pool/*=0*/, cudaStreamCaptureMode capt
   // due to the capture status being updated _after_ a capture had already started.
   c10::cuda::CUDACachingAllocator::beginAllocateToPool(capture_dev_, mempool_id_, [this](cudaStream_t stream) {
       cudaStreamCaptureStatus status;
-      CaptureId_t stream_capture_id = 0;
+      CaptureId_t stream_capture_id;
       AT_CUDA_CHECK(cudaStreamGetCaptureInfo(stream, &status, &stream_capture_id));
       return status == cudaStreamCaptureStatus::cudaStreamCaptureStatusActive && stream_capture_id == capture_id_;
   });

aten/src/ATen/cuda/EmptyTensor.cpp

@@ -10,7 +10,7 @@ TensorBase empty_cuda(
     ScalarType dtype,
     std::optional<Device> device_opt,
     std::optional<c10::MemoryFormat> memory_format_opt) {
-  at::globalContext().lazyInitDevice(c10::DeviceType::CUDA);
+  at::globalContext().lazyInitCUDA();
   const auto device = device_or_default(device_opt);
   TORCH_INTERNAL_ASSERT(device.is_cuda());
   const DeviceGuard device_guard(device);
@@ -50,7 +50,7 @@ TensorBase empty_strided_cuda(
     IntArrayRef stride,
     ScalarType dtype,
     std::optional<Device> device_opt) {
-  at::globalContext().lazyInitDevice(c10::DeviceType::CUDA);
+  at::globalContext().lazyInitCUDA();
   const auto device = device_or_default(device_opt);
   TORCH_INTERNAL_ASSERT(device.is_cuda());
   const DeviceGuard device_guard(device);

aten/src/ATen/cuda/PeerToPeerAccess.cpp

@@ -34,7 +34,7 @@ void init_p2p_access_cache(int64_t num_devices) {
 }  // namespace detail
 
 bool get_p2p_access(int dev, int dev_to_access) {
-  at::globalContext().lazyInitDevice(c10::DeviceType::CUDA);
+  at::globalContext().lazyInitCUDA();
 
   TORCH_CHECK(dev >= 0 || dev < num_devices_,
               dev, " is not a device");

aten/src/ATen/cuda/detail/CUDAHooks.cpp

@@ -14,7 +14,6 @@
 #include <ATen/detail/CUDAHooksInterface.h>
 #include <ATen/native/cuda/CuFFTPlanCache.h>
 #include <c10/util/Exception.h>
-#include <c10/util/env.h>
 #include <c10/cuda/CUDACachingAllocator.h>
 #include <c10/cuda/CUDAFunctions.h>
 #include <c10/util/irange.h>
@@ -80,19 +79,30 @@ struct _Initializer {
 } initializer;
 } // anonymous namespace
 
+// Sets the CUDA_MODULE_LOADING environment variable
+// if it's not set by the user.
+void maybe_set_cuda_module_loading(const std::string &def_value) {
+  auto value = std::getenv("CUDA_MODULE_LOADING");
+  if (!value) {
+#ifdef _WIN32
+    auto env_var = "CUDA_MODULE_LOADING=" + def_value;
+    _putenv(env_var.c_str());
+#else
+    setenv("CUDA_MODULE_LOADING", def_value.c_str(), 1);
+#endif
+  }
+}
 
 // NB: deleter is dynamic, because we need it to live in a separate
 // compilation unit (alt is to have another method in hooks, but
 // let's not if we don't need to!)
-void CUDAHooks::init() const {
+void CUDAHooks::initCUDA() const {
   C10_LOG_API_USAGE_ONCE("aten.init.cuda");
   // Force the update to enable unit testing. This code get executed before unit tests
   // have a chance to enable vitals.
   at::vitals::VitalsAPI.setVital("CUDA", "used", "true", /* force = */ true);
 
-  // Sets the CUDA_MODULE_LOADING environment variable
-  // if it's not set by the user.
-  c10::utils::set_env("CUDA_MODULE_LOADING", "LAZY", false);
+  maybe_set_cuda_module_loading("LAZY");
   const auto num_devices = c10::cuda::device_count_ensure_non_zero();
   c10::cuda::CUDACachingAllocator::init(num_devices);
   at::cuda::detail::init_p2p_access_cache(num_devices);
@@ -103,7 +113,7 @@ void CUDAHooks::init() const {
 #endif
 }
 
-const Generator& CUDAHooks::getDefaultGenerator(DeviceIndex device_index) const {
+const Generator& CUDAHooks::getDefaultCUDAGenerator(DeviceIndex device_index) const {
   return at::cuda::detail::getDefaultCUDAGenerator(device_index);
 }
 
@@ -231,9 +241,6 @@ DeviceIndex current_device() {
   return -1;
 }
 
-/**
- * DEPRECATED: use getCurrentDevice() instead
- */
 DeviceIndex CUDAHooks::current_device() const {
   return at::cuda::detail::current_device();
 }
@@ -429,21 +436,10 @@ void CUDAHooks::cuFFTClearPlanCache(DeviceIndex device_index) const {
   at::native::detail::cufft_clear_plan_cache_impl(device_index);
 }
 
-/**
- * DEPRECATED: use deviceCount() instead
- */
 int CUDAHooks::getNumGPUs() const {
   return at::cuda::device_count();
 }
 
-DeviceIndex CUDAHooks::deviceCount() const {
-  return at::cuda::device_count();
-}
-
-DeviceIndex CUDAHooks::getCurrentDevice() const {
-  return at::cuda::detail::current_device();
-}
-
 #ifdef USE_ROCM
 bool CUDAHooks::isGPUArch(DeviceIndex device_index, const std::vector<std::string>& archs) const {
   hipDeviceProp_t* prop = at::cuda::getDeviceProperties(device_index);

aten/src/ATen/cuda/detail/CUDAHooks.h

@@ -19,11 +19,10 @@ TORCH_CUDA_CPP_API void set_magma_init_fn(void (*magma_init_fn)());
 // The real implementation of CUDAHooksInterface
 struct CUDAHooks : public at::CUDAHooksInterface {
   CUDAHooks(at::CUDAHooksArgs) {}
-  void init() const override;
+  void initCUDA() const override;
   Device getDeviceFromPtr(void* data) const override;
   bool isPinnedPtr(const void* data) const override;
-  const Generator& getDefaultGenerator(
-      DeviceIndex device_index = -1) const override;
+  const Generator& getDefaultCUDAGenerator(DeviceIndex device_index = -1) const override;
   bool hasCUDA() const override;
   bool hasMAGMA() const override;
   bool hasCuDNN() const override;
@@ -50,9 +49,6 @@ struct CUDAHooks : public at::CUDAHooksInterface {
   int64_t cuFFTGetPlanCacheSize(DeviceIndex device_index) const override;
   void cuFFTClearPlanCache(DeviceIndex device_index) const override;
   int getNumGPUs() const override;
-  DeviceIndex deviceCount() const override;
-  DeviceIndex getCurrentDevice() const override;
-
 #ifdef USE_ROCM
   bool isGPUArch(DeviceIndex device_index, const std::vector<std::string>& archs) const override;
 #endif

aten/src/ATen/cuda/tunable/README.md

@@ -77,31 +77,6 @@ default, now called through TunableOp. Any call to at::cuda::blas::gemm() or ::b
 when enabled. Calling gemm() for a given set of input arguments (transa, transb, m, n, k) will attempt to use the
 fastest available implementation across both rocblas and hipblaslt.
 
-## Offline Tuning
-
-### Motivation
-Basically it is used for workload with high-memory utilization where one might run out of memory with regular tuning.
-
-### Workflow
-There are basically two steps:
-1) Set the environment variables to collect the untuned GEMM and this will generate `tunableop_untuned?.csv` ("?" is placeholder for the GPU ID), like:
-```
-PYTORCH_TUNABLEOP_ENABLED=1
-PYTORCH_TUNABLEOP_TUNING=0
-PYTORCH_TUNABLEOP_RECORD_UNTUNED=1
-...
-```
-2) Run a Python script that reads the `tunableop_untuned?.csv` and generates the `tunableop_results?.csv`, like:
-```
-import torch.cuda.tunable as tunable
-import os
-
-os.putenv('PYTORCH_TUNABLEOP_ENABLED', '1')
-os.putenv('PYTORCH_TUNABLEOP_TUNING', '1')
-os.putenv('PYTORCH_TUNABLEOP_RECORD_UNTUNED', '0')
-tunable.tune_gemm_in_file("tunableop_results?.csv")
-```
-
 ## Tuning Context
 The behavior of TunableOp is currently manipulated through environment variables, the C++ interface of
 at::cuda::tunable::getTuningContext(), or the `torch.cuda.tunable` python interfaces. The environment variables take
@@ -115,8 +90,6 @@ programmatically since the settings become fixed. Use the C++ or Python APIs ins
 | -------------------- | ----------- |
 | PYTORCH_TUNABLEOP_ENABLED | Default is 0. Set to 1 to enable. |
 | PYTORCH_TUNABLEOP_TUNING | Default is 1. Set to 0 to disable. |
-| PYTORCH_TUNABLEOP_RECORD_UNTUNED | Default is 0. Set to 1 to enable. |
-| PYTORCH_TUNABLEOP_UNTUNED_FILENAME | Default is 'tunableop_untuned.csv'. |
 | PYTORCH_TUNABLEOP_VERBOSE | Default is 0. Set to 1 to enable basic logging. 2 for basic tuning status. 3 for full trace. |
 | PYTORCH_TUNABLEOP_VERBOSE_FILENAME | Default is "err" for stderr. Set to "out" for stdout or a filename for capturing verbose logging. |
 | PYTORCH_TUNABLEOP_FILENAME | Default is 'tunableop_results.csv'. |
@@ -139,8 +112,6 @@ All python APIs exist in the `torch.cuda.tunable` module.
 | is_enabled() -> bool | |
 | tuning_enable(val: bool = True) -> None | Default is True. |
 | tuning_is_enabled() -> bool | |
-| record_untuned_enable(val: bool = True) -> None | Default is True. |
-| record_untuned_is_enabled() -> bool | |
 | set_max_tuning_duration(duration: int) -> None | |
 | get_max_tuning_duration() -> int | |
 | set_max_tuning_iterations(iterations: int) -> None | |
@@ -152,7 +123,6 @@ All python APIs exist in the `torch.cuda.tunable` module.
 | write_file_on_exit(val: bool) -> None | Default is True. |
 | write_file(filename: Optional[str] = None) -> None | If filename not given, it will call get_filename(). |
 | read_file(filename: Optional[str] = None) -> None | If filename not given, it will call get_filename(). |
-| tune_gemm_in_file(filename: str) -> None | read an untuned file and tune GEMMs in it. |
 
 ### C++ Interface
 Example:

aten/src/ATen/cuda/tunable/Tunable.cpp

@@ -112,32 +112,6 @@ void TuningResultsManager::Add(const std::string& op_signature, const std::strin
   AddImpl(op_signature, params_signature, best, it->second);
 }
 
-void TuningResultsManager::RecordUntuned( std::ofstream& untuned_file, const std::string& op_signature, const std::string& params_signature) {
-  std::scoped_lock l{lock_};
-  if (!untuned_file.good()) {
-    TORCH_WARN_ONCE("failed to open file for writing; untuned gemm will not be saved");
-    return;
-  } else {
-    bool isNew = false;
-    auto it = untuned_results_.find(op_signature);
-    if (it == untuned_results_.end()) {
-      it = untuned_results_.insert({op_signature, {}}).first;
-      isNew = true;
-    }
-
-    auto it_kernel_map = it->second.find(params_signature);
-    if (it_kernel_map == it->second.end()) {
-      it->second.insert(params_signature);
-      isNew = true;
-    }
-
-    if (isNew) {
-      untuned_file << op_signature << "," << params_signature << std::endl;
-      TUNABLE_LOG3("Untuned,", op_signature, ",", params_signature);
-    }
-  }
-}
-
 void TuningResultsManager::Delete(const std::string& op_signature, const std::string& params_signature) {
   std::scoped_lock l{lock_};
 
@@ -385,7 +359,6 @@ TuningStatus TuningResultsValidator::ValidatePyTorchVersion(const std::string& v
 TuningContext::TuningContext() :
     enable_{false},
     tuning_enable_{true},
-    record_untuned_enable_{false},
     manager_initialized_{false},
     write_file_on_exit_{true},
     numerics_check_enable_{false},
@@ -396,7 +369,6 @@ TuningContext::TuningContext() :
     icache_flush_{true},
     rotating_buffer_size_{-1},
     filename_{},
-    untuned_file_{},
     results_count_from_input_file_{0}
 {
 }
@@ -422,10 +394,6 @@ TuningContext::~TuningContext() {
       }
     }
   }
-
-  if (untuned_file_.good()) {
-    untuned_file_.close();
-  }
 }
 
 void TuningContext::EnableTunableOp(bool value) {
@@ -456,15 +424,6 @@ void TuningContext::EnableTuning(bool value) {
   }
 }
 
-void TuningContext::EnableRecordUntuned(bool value) {
-  record_untuned_enable_ = value;
-  if (value) {
-    TUNABLE_LOG1("Enable Record Untuned for TunableOp");
-  } else {
-    TUNABLE_LOG1("Disable Record Untuned for TunableOp");
-  }
-}
-
 bool TuningContext::IsTuningEnabled() const {
   static const char *env = std::getenv("PYTORCH_TUNABLEOP_TUNING");
   if (env != nullptr && strcmp(env, "0") == 0) {
@@ -473,33 +432,6 @@ bool TuningContext::IsTuningEnabled() const {
   return tuning_enable_;
 }
 
-bool TuningContext::IsRecordUntunedEnabled() const {
-  static const char *env = std::getenv("PYTORCH_TUNABLEOP_RECORD_UNTUNED");
-  if (env != nullptr && strcmp(env, "1") == 0) {
-    return true;
-  }
-  return record_untuned_enable_;
-}
-
-std::ofstream& TuningContext::GetUntunedFile(){
-  if (!untuned_file_.is_open()) {
-    const char *env = std::getenv("PYTORCH_TUNABLEOP_UNTUNED_FILENAME");
-    std::string filename = (env == nullptr) ? "tunableop_untuned.csv" : env;
-
-    std::string device = c10::str(int(c10::cuda::current_device()));
-    std::size_t found = filename.rfind(".");
-    if (found != std::string::npos) {
-      filename.insert(found, device);
-    } else {
-      // all else fails, just append
-      filename.append(device);
-    }
-
-    untuned_file_ = std::ofstream(filename, std::ios::out | std::ios::trunc);
-  }
-  return untuned_file_;
-}
-
 void TuningContext::WriteFileOnExit(bool value) {
   write_file_on_exit_ = value;
 }
@@ -613,7 +545,7 @@ TuningResultsManager& TuningContext::GetTuningResultsManager() {
       SetFilename(filename, true);
     }
     auto filename = GetFilename();
-    if (!filename.empty() && !IsRecordUntunedEnabled()) {
+    if (!filename.empty()) {
       ReadFile(filename);
       // attempt immediately to open file for writing to catch errors early
       std::ofstream file(filename, std::ios::out | std::ios::app);

aten/src/ATen/cuda/tunable/Tunable.h

@@ -19,7 +19,6 @@
 #include <string>
 #include <type_traits>
 #include <unordered_map>
-#include <unordered_set>
 #include <utility>
 #include <vector>
 
@@ -88,7 +87,6 @@ class TORCH_CUDA_CPP_API ResultEntry {
 
 typedef std::unordered_map<std::string, ResultEntry> KernelMap;
 typedef std::unordered_map<std::string, KernelMap> ResultsMap;
-typedef std::unordered_map<std::string, std::unordered_set<std::string>> UntunedMap;
 
 struct TORCH_CUDA_CPP_API TuningResults {
   // Validates if these results are compatible with the libraries
@@ -131,12 +129,9 @@ class TORCH_CUDA_CPP_API TuningResultsManager {
 
     size_t GetSize();
 
-    void RecordUntuned( std::ofstream& untuned_file, const std::string& op_signature, const std::string& params_signature);
   private:
     std::mutex lock_;
     ResultsMap results_;
-    UntunedMap untuned_results_;
-
 };
 
 class TORCH_CUDA_CPP_API TuningResultsValidator {
@@ -178,10 +173,6 @@ class TORCH_CUDA_CPP_API TuningContext {
     void EnableTuning(bool value);
     bool IsTuningEnabled() const;
 
-    void EnableRecordUntuned(bool value);
-    bool IsRecordUntunedEnabled() const;
-    std::ofstream& GetUntunedFile();
-
     void EnableNumericsCheck(bool value);
     bool IsNumericsCheckEnabled() const;
 
@@ -222,7 +213,6 @@ class TORCH_CUDA_CPP_API TuningContext {
   private:
     bool enable_;
     bool tuning_enable_;
-    bool record_untuned_enable_;
     bool manager_initialized_;
     bool write_file_on_exit_;
     bool numerics_check_enable_;
@@ -236,7 +226,6 @@ class TORCH_CUDA_CPP_API TuningContext {
     mutable c10::once_flag manager_init_once_;
     TuningResultsValidator validator_;
     std::string filename_;
-    std::ofstream untuned_file_;
     size_t results_count_from_input_file_;
 };
 

aten/src/ATen/cuda/tunable/TunableOp.h

@@ -54,15 +54,9 @@ class TunableOp {
         auto params_sig = params->Signature();
         result = mgr.Lookup(op_sig, params_sig);
         // If there is not previous tuning result been found, we do the tuning iff tuning is enabled
-        if (result == ResultEntry::Null()) {
-          if (ctx->IsTuningEnabled()) {
-            result = FindFastest(params);
-            mgr.Add(op_sig, params_sig, result);
-          }
-          else if (ctx->IsRecordUntunedEnabled()) {
-            // or record the gemm into file
-            mgr.RecordUntuned(ctx->GetUntunedFile(), op_sig, params_sig);
-          }
+        if (result == ResultEntry::Null() && ctx->IsTuningEnabled()) {
+          result = FindFastest(params);
+          mgr.Add(op_sig, params_sig, result);
         }
       }
       else {

aten/src/ATen/detail/AcceleratorHooksInterface.h

@@ -1,13 +1,9 @@
 #pragma once
 
-#include <ATen/core/Generator.h>
-
-#include <c10/core/Allocator.h>
 #include <c10/core/Device.h>
 #include <c10/core/Stream.h>
-
+#include <c10/core/Allocator.h>
 C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wunused-parameter")
-
 namespace at {
 
 // AcceleratorHooksInterface is a shared interface provided by all
@@ -23,10 +19,6 @@ struct TORCH_API AcceleratorHooksInterface {
   // Whether the device at device_index is fully initialized or not.
   virtual bool hasPrimaryContext(DeviceIndex device_index) const = 0;
 
-  virtual void init() const {
-    TORCH_CHECK(false, "Backend doesn`t support init()");
-  }
-
   virtual DeviceIndex deviceCount() const {
     return 0;
   }
@@ -58,18 +50,7 @@ struct TORCH_API AcceleratorHooksInterface {
     TORCH_CHECK(false, "Backend doesn't support getPinnedMemoryAllocator()");
     return nullptr;
   }
-
-  virtual const Generator& getDefaultGenerator(
-      C10_UNUSED DeviceIndex device_index = -1) const {
-    TORCH_CHECK(false, "Backend doesn`t support getDefaultGenerator()");
-  }
-
-  virtual Generator getNewGenerator(
-      C10_UNUSED DeviceIndex device_index = -1) const {
-    TORCH_CHECK(false, "Backend doesn`t support getNewGenerator()");
-  }
 };
 
 } // namespace at
-
 C10_DIAGNOSTIC_POP()

aten/src/ATen/detail/CUDAHooksInterface.h

@@ -6,13 +6,16 @@
 
 #include <ATen/detail/AcceleratorHooksInterface.h>
 
-// NB: Class must live in `at` due to limitations of Registry.h.
+// Forward-declares at::Generator and at::cuda::NVRTC
 namespace at {
-
-// Forward-declares at::cuda::NVRTC
+struct Generator;
 namespace cuda {
 struct NVRTC;
 } // namespace cuda
+} // namespace at
+
+// NB: Class must live in `at` due to limitations of Registry.h.
+namespace at {
 
 #ifdef _MSC_VER
 constexpr const char* CUDA_HELP =
@@ -62,16 +65,12 @@ struct TORCH_API CUDAHooksInterface : AcceleratorHooksInterface {
   ~CUDAHooksInterface() override = default;
 
   // Initialize THCState and, transitively, the CUDA state
-  void init() const override {
+  virtual void initCUDA() const {
     TORCH_CHECK(false, "Cannot initialize CUDA without ATen_cuda library. ", CUDA_HELP);
   }
 
-  const Generator& getDefaultGenerator(
-      C10_UNUSED DeviceIndex device_index = -1) const override {
-    TORCH_CHECK(
-        false,
-        "Cannot get default CUDA generator without ATen_cuda library. ",
-        CUDA_HELP);
+  virtual const Generator& getDefaultCUDAGenerator(C10_UNUSED DeviceIndex device_index = -1) const {
+    TORCH_CHECK(false, "Cannot get default CUDA generator without ATen_cuda library. ", CUDA_HELP);
   }
 
   virtual Device getDeviceFromPtr(void* /*data*/) const {

aten/src/ATen/detail/HIPHooksInterface.h

@@ -1,13 +1,19 @@
 #pragma once
 
 #include <c10/core/Allocator.h>
+#include <c10/core/GeneratorImpl.h>
 #include <c10/util/Exception.h>
+
 #include <c10/util/Registry.h>
 
 #include <ATen/detail/AcceleratorHooksInterface.h>
 
 #include <memory>
 
+namespace at {
+class Context;
+}
+
 // NB: Class must live in `at` due to limitations of Registry.h.
 namespace at {
 
@@ -20,13 +26,13 @@ struct TORCH_API HIPHooksInterface : AcceleratorHooksInterface {
   // squelch -Werror=non-virtual-dtor
   ~HIPHooksInterface() override = default;
 
-  void init() const override {
-    TORCH_CHECK(false, "Cannot initialize HIP without ATen_hip library.");
+  // Initialize the HIP library state
+  virtual void initHIP() const {
+    AT_ERROR("Cannot initialize HIP without ATen_hip library.");
   }
 
-  const Generator& getDefaultGenerator(
-      C10_UNUSED DeviceIndex device_index = -1) const override {
-    TORCH_CHECK(false, "Cannot initialize HIP without ATen_hip library.");
+  virtual std::unique_ptr<c10::GeneratorImpl> initHIPGenerator(Context*) const {
+    AT_ERROR("Cannot initialize HIP generator without ATen_hip library.");
   }
 
   virtual bool hasHIP() const {
@@ -45,6 +51,10 @@ struct TORCH_API HIPHooksInterface : AcceleratorHooksInterface {
     AT_ERROR("Pinned memory requires HIP.");
   }
 
+  virtual void registerHIPTypes(Context*) const {
+    AT_ERROR("Cannot registerHIPTypes() without ATen_hip library.");
+  }
+
   virtual int getNumGPUs() const {
     return 0;
   }

aten/src/ATen/detail/IPUHooksInterface.h

@@ -1,33 +1,25 @@
 #pragma once
 
-#include <ATen/detail/AcceleratorHooksInterface.h>
-
+#include <ATen/core/Generator.h>
 #include <c10/core/Allocator.h>
 #include <c10/util/Exception.h>
 #include <c10/util/Registry.h>
 
 namespace at {
 
-struct TORCH_API IPUHooksInterface : AcceleratorHooksInterface {
-  ~IPUHooksInterface() override = default;
+struct TORCH_API IPUHooksInterface {
+  virtual ~IPUHooksInterface() = default;
 
-  void init() const override {
-    TORCH_CHECK(false, "Cannot initialize IPU without ATen_ipu library.");
+  virtual const Generator& getDefaultIPUGenerator(
+      DeviceIndex device_index [[maybe_unused]] = -1) const {
+    AT_ERROR(
+        "Cannot get the default IPU generator: the IPU backend is not "
+        "available.");
   }
 
-  bool hasPrimaryContext(DeviceIndex device_index) const override {
-    TORCH_CHECK(false, "Cannot initialize IPU without ATen_ipu library.");
-    return false;
-  }
-
-  const Generator& getDefaultGenerator(
-      C10_UNUSED DeviceIndex device_index = -1) const override {
-    TORCH_CHECK(false, "Cannot initialize IPU without ATen_ipu library.");
-  }
-
-  Generator getNewGenerator(
-      DeviceIndex device_index [[maybe_unused]] = -1) const override {
-    TORCH_CHECK(false, "Cannot initialize IPU without ATen_ipu library.");
+  virtual Generator newIPUGenerator(DeviceIndex device_index [[maybe_unused]] = -1) const {
+    AT_ERROR(
+        "Cannot create a new IPU generator: the IPU backend is not available.");
   }
 };
 

aten/src/ATen/detail/MAIAHooksInterface.h

@@ -3,24 +3,13 @@
 #include <c10/util/Exception.h>
 #include <c10/util/Registry.h>
 
-#include <ATen/detail/AcceleratorHooksInterface.h>
-
 // NB: Class must live in `at` due to limitations of Registry.h.
 namespace at {
 
-struct TORCH_API MAIAHooksInterface : AcceleratorHooksInterface {
+struct TORCH_API MAIAHooksInterface {
   // This should never actually be implemented, but it is used to
   // squelch -Werror=non-virtual-dtor
-  ~MAIAHooksInterface() override = default;
-
-  void init() const override {
-    TORCH_CHECK(false, "Cannot initialize MAIA without ATen_maia library.");
-  }
-
-  bool hasPrimaryContext(DeviceIndex device_index) const override {
-    TORCH_CHECK(false, "Cannot initialize MAIA without ATen_maia library.");
-    return false;
-  }
+  virtual ~MAIAHooksInterface() = default;
 
   virtual std::string showConfig() const {
     TORCH_CHECK(false, "Cannot query detailed MAIA version information.");

aten/src/ATen/detail/MPSHooksInterface.h

@@ -2,9 +2,9 @@
 
 #pragma once
 
-#include <ATen/detail/AcceleratorHooksInterface.h>
-
 #include <c10/core/Allocator.h>
+#include <ATen/core/Generator.h>
+#include <ATen/detail/AcceleratorHooksInterface.h>
 #include <c10/util/Exception.h>
 #include <c10/util/Registry.h>
 
@@ -22,7 +22,7 @@ struct TORCH_API MPSHooksInterface : AcceleratorHooksInterface {
   ~MPSHooksInterface() override = default;
 
   // Initialize the MPS library state
-  void init() const override {
+  virtual void initMPS() const {
     FAIL_MPSHOOKS_FUNC(__func__);
   }
   virtual bool hasMPS() const {
@@ -31,8 +31,7 @@ struct TORCH_API MPSHooksInterface : AcceleratorHooksInterface {
   virtual bool isOnMacOSorNewer(unsigned major = 13, unsigned minor = 0) const {
     FAIL_MPSHOOKS_FUNC(__func__);
   }
-  const Generator& getDefaultGenerator(
-      C10_UNUSED DeviceIndex device_index = -1) const override {
+  virtual const Generator& getDefaultMPSGenerator() const {
     FAIL_MPSHOOKS_FUNC(__func__);
   }
   virtual Allocator* getMPSDeviceAllocator() const {

aten/src/ATen/detail/MTIAHooksInterface.h

@@ -31,7 +31,7 @@ struct TORCH_API MTIAHooksInterface : AcceleratorHooksInterface {
 
   ~MTIAHooksInterface() override = default;
 
-  void init() const override {
+  virtual void initMTIA() const {
     // Avoid logging here, since MTIA needs init devices first then it will know
     // how many devices are available. Make it as no-op if mtia extension is not
     // dynamically loaded.

aten/src/ATen/detail/PrivateUse1HooksInterface.h

@@ -1,20 +1,18 @@
 #pragma once
 
+#include <ATen/core/Generator.h>
 #include <ATen/detail/AcceleratorHooksInterface.h>
 #include <c10/core/Allocator.h>
 #include <c10/core/Device.h>
 #include <c10/core/Storage.h>
 #include <c10/util/Exception.h>
-
 C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wunused-parameter")
-
 namespace at {
 
 struct TORCH_API PrivateUse1HooksInterface : AcceleratorHooksInterface {
   ~PrivateUse1HooksInterface() override = default;
-
-  const at::Generator& getDefaultGenerator(
-      c10::DeviceIndex device_index) const override {
+  virtual const at::Generator& getDefaultGenerator(
+      c10::DeviceIndex device_index) const {
     TORCH_CHECK_NOT_IMPLEMENTED(
         false,
         "You should register `PrivateUse1HooksInterface` for PrivateUse1 before call `getDefaultGenerator`.");
@@ -26,23 +24,23 @@ struct TORCH_API PrivateUse1HooksInterface : AcceleratorHooksInterface {
         "You should register `PrivateUse1HooksInterface` for PrivateUse1 before call `getDeviceFromPtr`.");
   }
 
-  bool isPinnedPtr(const void* data) const override {
+  virtual bool isPinnedPtr(const void* data) const override {
     return false;
   }
 
-  Allocator* getPinnedMemoryAllocator() const override {
+  virtual Allocator* getPinnedMemoryAllocator() const override {
     TORCH_CHECK(
         false,
         "You should register `PrivateUse1HooksInterface` for PrivateUse1 before call `getPinnedMemoryAllocator`.");
   }
 
-  bool hasPrimaryContext(DeviceIndex device_index) const override {
+  virtual bool hasPrimaryContext(DeviceIndex device_index) const override {
     TORCH_CHECK_NOT_IMPLEMENTED(
         false,
         "You should register `PrivateUse1HooksInterface` for PrivateUse1 before call `hasPrimaryContext`.");
   }
 
-  void init() const override {}
+  virtual void initPrivateUse1() const {}
   virtual void resizePrivateUse1Bytes(
       const c10::Storage& storage,
       size_t newsize) const {

aten/src/ATen/detail/XPUHooksInterface.h

@@ -4,6 +4,7 @@
 #include <c10/util/Exception.h>
 #include <c10/util/Registry.h>
 
+#include <ATen/core/Generator.h>
 #include <ATen/detail/AcceleratorHooksInterface.h>
 
 C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wunused-parameter")
@@ -13,8 +14,10 @@ namespace at {
 struct TORCH_API XPUHooksInterface : AcceleratorHooksInterface{
   ~XPUHooksInterface() override = default;
 
-  void init() const override {
-    TORCH_CHECK(false, "Cannot initialize XPU without ATen_xpu library.");
+  virtual void initXPU() const {
+    TORCH_CHECK(
+        false,
+        "Cannot initialize XPU without ATen_xpu library.");
   }
 
   virtual bool hasXPU() const {
@@ -31,15 +34,12 @@ struct TORCH_API XPUHooksInterface : AcceleratorHooksInterface{
     TORCH_CHECK(false, "Cannot get XPU global device index without ATen_xpu library.");
   }
 
-  const Generator& getDefaultGenerator(
-      C10_UNUSED DeviceIndex device_index = -1) const override {
-    TORCH_CHECK(
-        false, "Cannot get default XPU generator without ATen_xpu library.");
+  virtual Generator getXPUGenerator(C10_UNUSED DeviceIndex device_index = -1) const {
+    TORCH_CHECK(false, "Cannot get XPU generator without ATen_xpu library.");
   }
 
-  Generator getNewGenerator(
-      C10_UNUSED DeviceIndex device_index = -1) const override {
-    TORCH_CHECK(false, "Cannot get XPU generator without ATen_xpu library.");
+  virtual const Generator& getDefaultXPUGenerator(C10_UNUSED DeviceIndex device_index = -1) const {
+    TORCH_CHECK(false, "Cannot get default XPU generator without ATen_xpu library.");
   }
 
   virtual DeviceIndex getNumGPUs() const {

aten/src/ATen/functorch/BatchRulesConvolution.cpp

@@ -362,7 +362,6 @@ static std::tuple<Tensor,Tensor,Tensor> convolution_backward_plumbing(
     const Tensor& grad_output_, const Tensor& input_, const Tensor& weight_,
     const c10::OptionalArrayRef<SymInt> bias_sizes_opt,
     c10::SymIntArrayRef stride, c10::SymIntArrayRef padding, c10::SymIntArrayRef dilation, bool transposed,
-    // NOLINTNEXTLINE(performance-unnecessary-value-param)
     c10::SymIntArrayRef output_padding, c10::SymInt groups, std::array<bool, 3> output_mask) {
   const auto maybe_layer = maybeCurrentDynamicLayer();
   vmap_check_escaped(maybe_layer, "convolution_backward_plumbing");

aten/src/ATen/functorch/BatchRulesHelper.h

@@ -458,16 +458,6 @@ inline int64_t get_bdim_size2(
   TORCH_INTERNAL_ASSERT(false);
 }
 
-inline c10::SymInt get_bdim_size2_symint(
-    const Tensor& a_value, std::optional<int64_t> a_bdim,
-    const Tensor& b_value, std::optional<int64_t> b_bdim) {
-  if (a_bdim)
-    return a_value.sym_size(*a_bdim);
-  if (b_bdim)
-    return b_value.sym_size(*b_bdim);
-  TORCH_INTERNAL_ASSERT(false);
-}
-
 // [start, start + 1, ..., stop - 1]
 inline VmapDimVector range(int64_t start, int64_t stop) {
   TORCH_INTERNAL_ASSERT(stop >= start);

aten/src/ATen/functorch/BatchRulesIndexing.cpp

@@ -8,7 +8,7 @@
 #include <ATen/core/dispatch/Dispatcher.h>
 #include <ATen/functorch/BatchRulesHelper.h>
 
-namespace at::functorch {
+namespace at { namespace functorch {
 
 #define OP_DECOMPOSE(op)  m.impl(#op, static_cast<decltype(&ATEN_FN(op))>(native::op));
 #define OP_DECOMPOSE2(op, overload)  m.impl(#op"."#overload, static_cast<decltype(&ATEN_FN2(op, overload))>(native::op));
@@ -20,4 +20,4 @@ TORCH_LIBRARY_IMPL(aten, FuncTorchBatched, m) {
   OP_DECOMPOSE(_unsafe_masked_index_put_accumulate);
 }
 
-}
+}}

aten/src/ATen/functorch/BatchRulesModules.cpp

@@ -226,7 +226,7 @@ static Tensor one_hot_decomposition_hack(const Tensor &self, int64_t num_classes
         if (num_classes <= 0) {
             AT_ERROR("Can not infer total number of classes from empty tensor.");
         } else {
-            shape.emplace_back(num_classes);
+            shape.push_back(num_classes);
             return at::empty_symint(shape, self.options());
         }
     }
@@ -246,7 +246,7 @@ static Tensor one_hot_decomposition_hack(const Tensor &self, int64_t num_classes
     //   TORCH_CHECK(num_classes > self.max().item().toLong(), "Class values must be smaller than num_classes.");
     // }
 
-    shape.emplace_back(num_classes);
+    shape.push_back(num_classes);
     Tensor ret = at::zeros_symint(shape, self.options());
     return ret.scatter(-1, self.unsqueeze(-1), 1);
 }

aten/src/ATen/functorch/BatchRulesRandomness.cpp

@@ -213,7 +213,7 @@ static std::tuple<Tensor,Tensor> native_dropout_batching_rule(const Tensor& tens
   return std::make_tuple(output, mask);
 }
 
-static Tensor multinomial_batching_rule(const Tensor& self, const int64_t num_samples, const bool replacement, std::optional<Generator> generator) {
+static Tensor multinomial_batching_rule(const Tensor& self, const int64_t num_samples, const bool replacement, const std::optional<Generator> generator) {
   c10::impl::ExcludeDispatchKeyGuard guard(DispatchKey::FuncTorchVmapMode);
   auto maybe_layer = maybeCurrentDynamicLayer();
   const auto cur_level = maybe_layer->layerId();
@@ -237,7 +237,7 @@ static Tensor multinomial_batching_rule(const Tensor& self, const int64_t num_sa
     if (is_2D_case) {
       self_value = reshape_dim_into(0, 0, self_value);
     }
-    auto out = multinomial(self_value, num_samples, replacement, std::move(generator));
+    auto out = multinomial(self_value, num_samples, replacement, generator);
     if (is_2D_case) {
       out = reshape_dim_outof_symint(0, maybe_layer->batchSize(), out);
     }
@@ -249,7 +249,7 @@ static Tensor multinomial_batching_rule(const Tensor& self, const int64_t num_sa
   // Must be same randomness with unbatched input
   // 1D case: S -> multinomial(S) -> S
   // 2D case: MS -> multinomial(MS) -> MS
-  return multinomial(self_value, num_samples, replacement, std::move(generator));
+  return multinomial(self_value, num_samples, replacement, generator);
 }
 
 template <typename A, A a, typename C>

aten/src/ATen/functorch/BatchRulesScatterOps.cpp

@@ -58,7 +58,7 @@ static int64_t get_max_index_logical_dim(
 static std::vector<std::optional<Tensor>> batchIndices(
   ArrayRef<std::optional<Tensor>> indices,
   ArrayRef<std::optional<int64_t>> indices_bdims,
-  const c10::SymInt& batch_size,
+  int64_t batch_size,
   std::optional<int64_t> self_bdim,
   std::optional<int64_t> values_bdim = std::nullopt) {
   // There are 3 main cases:
@@ -89,7 +89,7 @@ static std::vector<std::optional<Tensor>> batchIndices(
 
   for (size_t i = 0; i < indices.size(); i++) {
     auto index = indices[i];
-    if (index.has_value() && index->sym_numel() != 0) {
+    if (index.has_value() && index->numel() != 0) {
       const auto idx_bdim = indices_bdims[i];
       indices_.emplace_back(maybePadToLogicalRank(moveBatchDimToFront(index.value(), idx_bdim), idx_bdim, maxLogicalRank));
       if (index.value().dtype() == kBool && indices_bdims[i].has_value()) {
@@ -346,10 +346,10 @@ namespace {
   // Code is mostly duplicated from
   // https://github.com/pytorch/pytorch/blob/fb0e27d38a8fdab4e1c14d6378c9e41cb30fd6a3
   // /aten/src/ATen/native/TensorAdvancedIndexing.cpp#L294-L312
-  VmapSymDimVector compute_indexed_shape(const Tensor &src, TensorList indices_list)
+  VmapDimVector compute_indexed_shape(const Tensor &src, TensorList indices_list)
   {
     int64_t dims_before = 0, dims_indexed = 0;
-    SymIntArrayRef replacement_shape;
+    IntArrayRef replacement_shape;
     for (const auto dim : c10::irange(indices_list.size())) {
       if (!indices_list[dim].defined()) {
         if (dims_indexed == 0) {
@@ -357,7 +357,7 @@ namespace {
         }
       } else {
         dims_indexed++;
-        replacement_shape = indices_list[dim].sym_sizes();
+        replacement_shape = indices_list[dim].sizes();
       }
     }
 
@@ -365,7 +365,7 @@ namespace {
     // The offset in these dimensions is computed by the kernel using the index tensor's
     // values and the stride of src. The new shape is not meaningful. It's used to make
     // the shape compatible with the result tensor.
-    auto shape = VmapSymDimVector(src.sym_sizes());
+    auto shape = VmapDimVector(src.sizes());
     int64_t end = dims_before + dims_indexed;
     shape.erase(shape.begin() + dims_before, shape.begin() + end);
     shape.insert(shape.begin() + dims_before, replacement_shape.begin(), replacement_shape.end());
@@ -375,7 +375,7 @@ namespace {
   // Code is mostly duplicated from
   // https://github.com/pytorch/pytorch/blob/fb0e27d38a8fdab4e1c14d6378c9e41cb30fd6a3
   // /aten/src/ATen/native/TensorAdvancedIndexing.cpp#L379-L405
-  VmapSymDimVector get_indexed_shape(Tensor self, const torch::List<std::optional<at::Tensor>> &orig)
+  VmapDimVector get_indexed_shape(Tensor self, const torch::List<std::optional<at::Tensor>> &orig)
   {
     at::native::checkIndexTensorTypes(orig, /*allow_int*/ true);
     // first expand BoolTensor (masks) or ByteTensor (masks) into 1 or more LongTensors
@@ -406,13 +406,13 @@ namespace {
                               ArrayRef<std::optional<int64_t>> indices_bdims,
                               const Tensor &values,
                               std::optional<int64_t> values_bdim,
-                              std::optional<c10::SymInt> opt_batch_size = {}) {
+                              std::optional<int64_t> opt_batch_size = {}) {
 
     Tensor self_ = moveBatchDimToFront(self, self_bdim);
     Tensor values_ = moveBatchDimToFront(values, values_bdim);
     // for inplace variants `index_put_` and `_index_put_impl_` we find the batch_size
     // here while for `index_put` does it outside of this function.
-    const auto batch_size = opt_batch_size ? opt_batch_size.value() : self_.sym_size(0);
+    const auto batch_size = opt_batch_size ? opt_batch_size.value() : self_.size(0);
     self_ = ensure_has_bdim(self_, self_bdim.has_value(), batch_size);
     values_ = ensure_has_bdim(values_, values_bdim.has_value(), batch_size);
     TORCH_INTERNAL_ASSERT(indices.size() == indices_bdims.size());
@@ -431,7 +431,7 @@ namespace {
 
       // number of unit dims (for broadcasting value to indexed_shape)
       auto n_unit_dims = indexed_shape.size() - values_sizes.size();
-      VmapSymDimVector new_values_shape(values_sizes.size() + n_unit_dims);
+      VmapDimVector new_values_shape(values_sizes.size() + n_unit_dims);
 
       // add the batch-dim
       new_values_shape[0] = batch_size;
@@ -445,7 +445,7 @@ namespace {
         // since batch and unit dims are already be filled.
         new_values_shape[idx + n_unit_dims] = values_sizes[idx];
       }
-      values_ = values_.view_symint(new_values_shape);
+      values_ = values_.view(new_values_shape);
     }
 
     return std::make_tuple(self_, indices_, values_);
@@ -613,14 +613,14 @@ std::tuple<Tensor, std::optional<int64_t>> index_put_batch_rule(
   TORCH_INTERNAL_ASSERT(indices.size() == indices_bdims.size());
 
   // find the batch_size
-  c10::SymInt batch_size = 0;
+  int64_t batch_size = 0;
   if (self_bdim || values_bdim) {
-    batch_size = get_bdim_size2_symint(self, self_bdim, values, values_bdim);
+    batch_size = get_bdim_size2(self, self_bdim, values, values_bdim);
   } else {
     // one or more of the indices is batched.
     for (size_t i = 0; i < indices.size(); i++) {
       if (indices_bdims[i] && indices[i].has_value()) {
-        batch_size = indices[i].value().sym_size(*indices_bdims[i]);
+        batch_size = indices[i].value().size(*indices_bdims[i]);
         break;
       }
     }

aten/src/ATen/functorch/LegacyVmapTransforms.cpp

@@ -102,7 +102,7 @@ static Tensor moveDimToFrontAndExpand(Tensor tensor, std::optional<int64_t> dim,
   } else {
     tensor = tensor.unsqueeze(0);
     auto expanded_sizes = tensor.sym_sizes().vec();
-    expanded_sizes[0] = std::move(size);
+    expanded_sizes[0] = size;
     tensor = tensor.expand_symint(expanded_sizes);
   }
   return tensor;

aten/src/ATen/functorch/PyTorchOperatorHacks.cpp

@@ -4,6 +4,7 @@
 #include <ATen/WrapDimUtils.h>
 #include <ATen/functorch/TensorWrapper.h>
 #include <ATen/functorch/BatchedTensorImpl.h>
+#include <ATen/ATen.h>
 #include <ATen/Dispatch.h>
 #include <c10/util/irange.h>
 #include <ATen/NamedTensorUtils.h>

aten/src/ATen/mps/MPSHooks.h

@@ -12,15 +12,14 @@ namespace at::mps {
 // The real implementation of MPSHooksInterface
 struct MPSHooks : public at::MPSHooksInterface {
   MPSHooks(at::MPSHooksArgs) {}
-  void init() const override;
+  void initMPS() const override;
 
   // MPSDevice interface
   bool hasMPS() const override;
   bool isOnMacOSorNewer(unsigned major, unsigned minor) const override;
 
   // MPSGeneratorImpl interface
-  const Generator& getDefaultGenerator(
-      DeviceIndex device_index = -1) const override;
+  const Generator& getDefaultMPSGenerator() const override;
 
   // MPSStream interface
   void deviceSynchronize() const override;

aten/src/ATen/mps/MPSHooks.mm

@@ -10,7 +10,7 @@
 
 namespace at::mps {
 
-void MPSHooks::init() const {
+void MPSHooks::initMPS() const {
   C10_LOG_API_USAGE_ONCE("aten.init.mps");
   // TODO: initialize MPS devices and streams here
 }
@@ -59,7 +59,7 @@ Allocator* MPSHooks::getMPSDeviceAllocator() const {
   return at::mps::GetMPSAllocator();
 }
 
-const Generator& MPSHooks::getDefaultGenerator([[maybe_unused]] DeviceIndex device_index) const {
+const Generator& MPSHooks::getDefaultMPSGenerator() const {
   return at::mps::detail::getDefaultMPSGenerator();
 }
 

aten/src/ATen/native/LinearAlgebra.cpp

@@ -207,7 +207,6 @@ TORCH_META_FUNC(mm)(const Tensor & self, const Tensor & mat2) {
 
 TORCH_META_FUNC(linalg_vector_norm)(const Tensor& self, const Scalar& scalar_ord, OptionalIntArrayRef opt_dim, bool keepdim, std::optional<ScalarType> opt_dtype) {
   at::native::checkFloatingOrComplex(self, "linalg.vector_norm");
-  TORCH_CHECK(!at::isComplexType(scalar_ord.type()), "linalg.vector_norm: Expected a non-complex scalar as the order of norm.");
 
   auto dim = opt_dim.value_or(IntArrayRef{});
   // Casting a large integer to a double will just introduce an error for
@@ -2893,7 +2892,6 @@ Tensor linalg_matrix_norm(
     bool keepdim,
     std::optional<ScalarType> opt_dtype) {
   // Check ord first as it will be used in the dtype check of A
-  TORCH_CHECK(!at::isComplexType(scalar_ord.type()), "linalg.matrix_norm: Expected a non-complex scalar as the order of norm.");
   auto ord = scalar_ord.toDouble();
   auto abs_ord = std::abs(ord);
   TORCH_CHECK(abs_ord == 2. || abs_ord == 1. || abs_ord == INFINITY, "linalg.matrix_norm: Order ", ord, " not supported.");

aten/src/ATen/native/ReduceOps.cpp

@@ -2291,7 +2291,7 @@ bool cpu_equal(const Tensor& self, const Tensor& other) {
         other_data += strides[1];
       }
     });
-  }), kBool, kBFloat16, kHalf, AT_EXPAND(AT_ALL_TYPES_AND_COMPLEX), AT_EXPAND(AT_FLOAT8_TYPES), AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES));
+  }), kBool, kBFloat16, kHalf, AT_EXPAND(AT_ALL_TYPES_AND_COMPLEX), AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES));
   return result.load();
 }
 

aten/src/ATen/native/SummaryOps.cpp

@@ -5,8 +5,6 @@
 #include <ATen/Dispatch.h>
 #include <c10/util/irange.h>
 
-#include <limits>
-
 #ifndef AT_PER_OPERATOR_HEADERS
 #include <ATen/Functions.h>
 #include <ATen/NativeFunctions.h>
@@ -35,17 +33,6 @@ Tensor _bincount_cpu_template(
     AT_ERROR("bincount only supports 1-d non-negative integral inputs.");
   }
 
-  // Ensure max_val < 2 ^ 63 - 1 (9223372036854775807)
-  auto max_val = *self.max().data_ptr<input_t>();
-  if (max_val >= std::numeric_limits<int64_t>::max()) {
-    AT_ERROR(
-        "maximum value of input overflowed, it should be < ",
-        std::numeric_limits<int64_t>::max(),
-        " but got ",
-        max_val
-    );
-  }
-
   bool has_weights = weights.defined();
   if (has_weights && (weights.dim() != 1 || weights.size(0) != self.size(0))) {
     AT_ERROR("weights should be 1-d and have the same length as input");
@@ -53,7 +40,7 @@ Tensor _bincount_cpu_template(
 
   Tensor output;
   int64_t self_size = self.size(0);
-  int64_t nbins = static_cast<int64_t>(max_val) + 1L;
+  int64_t nbins = static_cast<int64_t>(*self.max().data_ptr<input_t>()) + 1L;
   nbins = std::max(nbins, minlength); // at least minlength # of bins
 
   const input_t* self_p = self.const_data_ptr<input_t>();

aten/src/ATen/native/TensorAdvancedIndexing.cpp

@@ -1435,8 +1435,8 @@ Tensor & index_select_out_cpu_(const Tensor & self, int64_t dim, const Tensor &
         });
       });
     } else {
-      AT_DISPATCH_V2(
-        self.scalar_type(), "index_select", AT_WRAP([&index_contig, &self, &result, &dim, &numel] {
+      AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND4(ScalarType::ComplexHalf, ScalarType::Half, ScalarType::Bool, ScalarType::BFloat16,
+        self.scalar_type(), "index_select", [&index_contig, &self, &result, &dim, &numel] {
         auto self_stride = self.dim() == 0 ? 1 : self.stride(dim);
         auto result_stride = result.dim() == 0 ? 1 : result.stride(dim);
 
@@ -1453,7 +1453,7 @@ Tensor & index_select_out_cpu_(const Tensor & self, int64_t dim, const Tensor &
             *(result_data_ptr + i * result_stride) = *self_ip;
           }
         });
-        }), AT_EXPAND(AT_ALL_TYPES_AND_COMPLEX), ScalarType::ComplexHalf, ScalarType::Half, ScalarType::Bool, ScalarType::BFloat16, AT_EXPAND(AT_FLOAT8_TYPES));
+      });
     }
   }
 

aten/src/ATen/native/TensorFactories.h

@@ -106,7 +106,7 @@ inline Tensor& fill_empty_deterministic_(Tensor& tensor) {
     AT_DISPATCH_V2(
       tensor.scalar_type(), "fill_empty_deterministic_", AT_WRAP([&]() {
         tensor.fill_(std::numeric_limits<scalar_t>::quiet_NaN());
-    }), AT_EXPAND(AT_FLOATING_TYPES), AT_EXPAND(AT_COMPLEX_TYPES), AT_EXPAND(AT_FLOAT8_TYPES), kBFloat16, kHalf, kComplexHalf);
+    }), AT_EXPAND(AT_FLOATING_TYPES), AT_EXPAND(AT_COMPLEX_TYPES), AT_EXPAND(AT_FLOAT8_TYPES), kBFloat16, kHalf);
   } else {
     AT_DISPATCH_V2(
       tensor.scalar_type(), "fill_empty_deterministic_", AT_WRAP([&]() {

aten/src/ATen/native/cuda/CUDAScalar.cu

@@ -44,7 +44,7 @@ Scalar _local_scalar_dense_cuda(const Tensor& self) {
           cudaStream_t stream = at::cuda::getCurrentCUDAStream();
           at::cuda::memcpy_and_sync((void *)value.const_data_ptr<scalar_t>(), self.const_data_ptr<scalar_t>(), sizeof(scalar_t), cudaMemcpyDeviceToHost, stream);
           r = Scalar(*value.const_data_ptr<scalar_t>());
-        }), AT_EXPAND(AT_ALL_TYPES_AND_COMPLEX), kComplexHalf, kHalf, kBool, kBFloat16, AT_EXPAND(AT_FLOAT8_TYPES), AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES));
+        }), AT_EXPAND(AT_ALL_TYPES_AND_COMPLEX), kComplexHalf, kHalf, kBool, kBFloat16, AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES));
 #if defined(USE_ROCM)
   } else {
     auto cpu_self = self.cpu();

aten/src/ATen/native/cuda/ForeachBinaryOpList.cu

@@ -285,64 +285,44 @@ struct Copy<dst_t, c10::complex<float>> {
   }
 };
 
-#define AT_DISPATCH_SOURCE_TYPES(TYPE, NAME, ...)                                                \
-  AT_DISPATCH_SWITCH(                                                                            \
-      TYPE,                                                                                      \
-      NAME,                                                                                      \
-      AT_PRIVATE_CASE_TYPE_USING_HINT(                                                           \
-          at::ScalarType::Byte,                                                                  \
-          src_t,                                                                                 \
-          __VA_ARGS__) AT_PRIVATE_CASE_TYPE_USING_HINT(at::ScalarType::Char, src_t, __VA_ARGS__) \
-          AT_PRIVATE_CASE_TYPE_USING_HINT(                                                       \
-              at::ScalarType::Long, src_t, __VA_ARGS__)                                          \
-              AT_PRIVATE_CASE_TYPE_USING_HINT(                                                   \
-                  at::ScalarType::Short, src_t, __VA_ARGS__)                                     \
-                  AT_PRIVATE_CASE_TYPE_USING_HINT(                                               \
-                      at::ScalarType::Int, src_t, __VA_ARGS__)                                   \
-                      AT_PRIVATE_CASE_TYPE_USING_HINT(                                           \
-                          at::ScalarType::Double, src_t, __VA_ARGS__)                            \
-                          AT_PRIVATE_CASE_TYPE_USING_HINT(                                       \
-                              at::ScalarType::Float, src_t, __VA_ARGS__)                         \
-                              AT_PRIVATE_CASE_TYPE_USING_HINT(                                   \
-                                  at::ScalarType::ComplexDouble,                                 \
-                                  src_t,                                                         \
-                                  __VA_ARGS__)                                                   \
-                                  AT_PRIVATE_CASE_TYPE_USING_HINT(                               \
-                                      at::ScalarType::ComplexFloat,                              \
-                                      src_t,                                                     \
-                                      __VA_ARGS__)                                               \
-                                      AT_PRIVATE_CASE_TYPE_USING_HINT(                           \
-                                          at::ScalarType::Half,                                  \
-                                          src_t,                                                 \
-                                          __VA_ARGS__)                                           \
-                                          AT_PRIVATE_CASE_TYPE_USING_HINT(                       \
-                                              at::ScalarType::BFloat16,                          \
-                                              src_t,                                             \
-                                              __VA_ARGS__)                                       \
-                                              AT_PRIVATE_CASE_TYPE_USING_HINT(                   \
-                                                  at::ScalarType::Bool,                          \
-                                                  src_t,                                         \
-                                                  __VA_ARGS__)                                   \
-                                                  AT_PRIVATE_CASE_TYPE_USING_HINT(               \
-                                                      at::ScalarType::                           \
-                                                          Float8_e4m3fn,                         \
-                                                      src_t,                                     \
-                                                      __VA_ARGS__)                               \
-                                                      AT_PRIVATE_CASE_TYPE_USING_HINT(           \
-                                                          at::ScalarType::                       \
-                                                              Float8_e4m3fnuz,                   \
-                                                          src_t,                                 \
-                                                          __VA_ARGS__)                           \
-                                                          AT_PRIVATE_CASE_TYPE_USING_HINT(       \
-                                                              at::ScalarType::                   \
-                                                                  Float8_e5m2,                   \
-                                                              src_t,                             \
-                                                              __VA_ARGS__)                       \
-                                                              AT_PRIVATE_CASE_TYPE_USING_HINT(   \
-                                                                  at::ScalarType::               \
-                                                                      Float8_e5m2fnuz,           \
-                                                                  src_t,                         \
-                                                                  __VA_ARGS__))
+#define AT_DISPATCH_SOURCE_TYPES(TYPE, NAME, ...)                                  \
+  AT_DISPATCH_SWITCH(                                                              \
+      TYPE,                                                                        \
+      NAME,                                                                        \
+      AT_PRIVATE_CASE_TYPE_USING_HINT(                                             \
+          at::ScalarType::Byte, src_t, __VA_ARGS__)                                \
+          AT_PRIVATE_CASE_TYPE_USING_HINT(                                         \
+              at::ScalarType::Char, src_t, __VA_ARGS__)                            \
+              AT_PRIVATE_CASE_TYPE_USING_HINT(                                     \
+                  at::ScalarType::Long, src_t, __VA_ARGS__)                        \
+                  AT_PRIVATE_CASE_TYPE_USING_HINT(                                 \
+                      at::ScalarType::Short, src_t, __VA_ARGS__)                   \
+                      AT_PRIVATE_CASE_TYPE_USING_HINT(                             \
+                          at::ScalarType::Int, src_t, __VA_ARGS__)                 \
+                          AT_PRIVATE_CASE_TYPE_USING_HINT(                         \
+                              at::ScalarType::Double, src_t, __VA_ARGS__)          \
+                              AT_PRIVATE_CASE_TYPE_USING_HINT(                     \
+                                  at::ScalarType::Float, src_t, __VA_ARGS__)       \
+                                  AT_PRIVATE_CASE_TYPE_USING_HINT(                 \
+                                      at::ScalarType::ComplexDouble,               \
+                                      src_t,                                       \
+                                      __VA_ARGS__)                                 \
+                                      AT_PRIVATE_CASE_TYPE_USING_HINT(             \
+                                          at::ScalarType::ComplexFloat,            \
+                                          src_t,                                   \
+                                          __VA_ARGS__)                             \
+                                          AT_PRIVATE_CASE_TYPE_USING_HINT(         \
+                                              at::ScalarType::Half,                \
+                                              src_t,                               \
+                                              __VA_ARGS__)                         \
+                                              AT_PRIVATE_CASE_TYPE_USING_HINT(     \
+                                                  at::ScalarType::BFloat16,        \
+                                                  src_t,                           \
+                                                  __VA_ARGS__)                     \
+                                                  AT_PRIVATE_CASE_TYPE_USING_HINT( \
+                                                      at::ScalarType::Bool,        \
+                                                      src_t,                       \
+                                                      __VA_ARGS__))
 
 namespace {
 
@@ -430,14 +410,10 @@ void foreach_tensor_copy_list_kernel_cuda_(
 
   std::vector<std::vector<at::Tensor>> tensor_lists{src.vec(), self.vec()};
 
-  AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND7(
+  AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND3(
       ScalarType::Half,
       ScalarType::BFloat16,
       ScalarType::Bool,
-      ScalarType::Float8_e4m3fn,
-      ScalarType::Float8_e4m3fnuz,
-      ScalarType::Float8_e5m2,
-      ScalarType::Float8_e5m2fnuz,
       self[0].scalar_type(),
       "foreach_tensor_copy",
       [&]() {