[profiler] Add stack trace to Fallback kernels for inductor lite mode

.ci/docker/almalinux/build.sh

@@ -36,7 +36,11 @@ case ${DOCKER_TAG_PREFIX} in
     ;;
   rocm*)
     BASE_TARGET=rocm
-    PYTORCH_ROCM_ARCH="gfx900;gfx906;gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201;gfx950;gfx1150;gfx1151"
+    PYTORCH_ROCM_ARCH="gfx900;gfx906;gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201"
+    # add gfx950, gfx115x conditionally starting in ROCm 7.0
+    if [[ "$ROCM_VERSION" == *"7.0"* ]]; then
+        PYTORCH_ROCM_ARCH="${PYTORCH_ROCM_ARCH};gfx950;gfx1150;gfx1151"
+    fi
     EXTRA_BUILD_ARGS="${EXTRA_BUILD_ARGS} --build-arg PYTORCH_ROCM_ARCH=${PYTORCH_ROCM_ARCH}"
     ;;
   *)

.ci/docker/build.sh

@@ -168,18 +168,6 @@ case "$tag" in
     VISION=yes
     TRITON=yes
     ;;
-  pytorch-linux-jammy-py3.11-clang12)
-    ANACONDA_PYTHON_VERSION=3.11
-    CLANG_VERSION=12
-    VISION=no
-    TRITON=no
-    ;;
-  pytorch-linux-jammy-py3.12-clang12)
-    ANACONDA_PYTHON_VERSION=3.12
-    CLANG_VERSION=12
-    VISION=no
-    TRITON=no
-    ;;
   pytorch-linux-jammy-rocm-n-py3 | pytorch-linux-jammy-rocm-n-py3-benchmarks | pytorch-linux-noble-rocm-n-py3)
     if [[ $tag =~ "jammy" ]]; then
       ANACONDA_PYTHON_VERSION=3.10
@@ -207,9 +195,9 @@ case "$tag" in
     NINJA_VERSION=1.9.0
     TRITON=yes
     ;;
-  pytorch-linux-noble-xpu-n-py3 | pytorch-linux-noble-xpu-n-py3-inductor-benchmarks)
+  pytorch-linux-jammy-xpu-n-py3 | pytorch-linux-jammy-xpu-n-py3-inductor-benchmarks)
     ANACONDA_PYTHON_VERSION=3.10
-    GCC_VERSION=13
+    GCC_VERSION=11
     VISION=yes
     XPU_VERSION=2025.2
     NINJA_VERSION=1.9.0
@@ -260,12 +248,6 @@ case "$tag" in
     HALIDE=yes
     TRITON=yes
     ;;
-  pytorch-linux-jammy-cuda12.8-py3.12-pallas)
-    CUDA_VERSION=12.8.1
-    ANACONDA_PYTHON_VERSION=3.12
-    GCC_VERSION=11
-    PALLAS=yes
-    ;;
   pytorch-linux-jammy-py3.12-triton-cpu)
     CUDA_VERSION=12.6
     ANACONDA_PYTHON_VERSION=3.12
@@ -387,7 +369,6 @@ docker build \
        --build-arg "INDUCTOR_BENCHMARKS=${INDUCTOR_BENCHMARKS}" \
        --build-arg "EXECUTORCH=${EXECUTORCH}" \
        --build-arg "HALIDE=${HALIDE}" \
-       --build-arg "PALLAS=${PALLAS}" \
        --build-arg "XPU_VERSION=${XPU_VERSION}" \
        --build-arg "UNINSTALL_DILL=${UNINSTALL_DILL}" \
        --build-arg "ACL=${ACL:-}" \

.ci/docker/ci_commit_pins/jax.txt

@@ -1 +0,0 @@
-0.8.0

.ci/docker/common/install_jax.sh

@@ -1,40 +0,0 @@
-#!/bin/bash
-
-set -ex
-
-source "$(dirname "${BASH_SOURCE[0]}")/common_utils.sh"
-
-# Get the pinned JAX version (same for all CUDA versions)
-JAX_VERSION=$(get_pinned_commit /ci_commit_pins/jax)
-
-function install_jax_12() {
-  echo "Installing JAX ${JAX_VERSION} with CUDA 12 support"
-  pip_install "jax[cuda12]==${JAX_VERSION}" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
-
-  # Verify installation
-  python -c "import jax"  # check for errors
-  echo "JAX ${JAX_VERSION} installation completed successfully for CUDA 12"
-}
-
-function install_jax_13() {
-  echo "Installing JAX ${JAX_VERSION} with CUDA 13 support"
-  pip_install "jax[cuda13]==${JAX_VERSION}" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
-
-  # Verify installation
-  python -c "import jax"  # check for errors
-  echo "JAX ${JAX_VERSION} installation completed successfully for CUDA 13"
-}
-
-# idiomatic parameter and option handling in sh
-while test $# -gt 0
-do
-    case "$1" in
-    12.4|12.6|12.6.*|12.8|12.8.*|12.9|12.9.*) install_jax_12;
-        ;;
-    13.0|13.0.*) install_jax_13;
-        ;;
-    *) echo "bad argument $1"; exit 1
-        ;;
-    esac
-    shift
-done

.ci/docker/common/install_xpu.sh

@@ -9,7 +9,7 @@ set -xe
 
 function install_ubuntu() {
     . /etc/os-release
-    if [[ ! " jammy noble " =~ " ${VERSION_CODENAME} " ]]; then
+    if [[ ! " jammy " =~ " ${VERSION_CODENAME} " ]]; then
         echo "Ubuntu version ${VERSION_CODENAME} not supported"
         exit
     fi
@@ -35,24 +35,25 @@ function install_ubuntu() {
     # The xpu-smi packages
     apt-get install -y flex bison xpu-smi
 
-    # Compute and Media Runtimes
-    if [[ " ${VERSION_CODENAME} " =~ " noble " ]]; then
+    if [[ "${XPU_DRIVER_TYPE,,}" == "lts" ]]; then
+        # Compute and Media Runtimes
         apt-get install -y \
-            intel-opencl-icd libze-intel-gpu1 libze1 \
-            intel-media-va-driver-non-free libmfx-gen1 libvpl2 \
-            libegl-mesa0 libegl1-mesa-dev libgbm1 libgl1-mesa-dev libgl1-mesa-dri \
+            intel-opencl-icd intel-level-zero-gpu level-zero \
+            intel-media-va-driver-non-free libmfx1 libmfxgen1 libvpl2 \
+            libegl-mesa0 libegl1-mesa libegl1-mesa-dev libgbm1 libgl1-mesa-dev libgl1-mesa-dri \
             libglapi-mesa libgles2-mesa-dev libglx-mesa0 libigdgmm12 libxatracker2 mesa-va-drivers \
-            mesa-vdpau-drivers mesa-vulkan-drivers va-driver-all vainfo hwinfo clinfo intel-ocloc
-    else # jammy
+            mesa-vdpau-drivers mesa-vulkan-drivers va-driver-all vainfo hwinfo clinfo
+        # Development Packages
+        apt-get install -y libigc-dev intel-igc-cm libigdfcl-dev libigfxcmrt-dev level-zero-dev
+    else # rolling driver
         apt-get install -y \
             intel-opencl-icd libze-intel-gpu1 libze1 \
             intel-media-va-driver-non-free libmfx-gen1 libvpl2 \
             libegl-mesa0 libegl1-mesa libegl1-mesa-dev libgbm1 libgl1-mesa-dev libgl1-mesa-dri \
             libglapi-mesa libglx-mesa0 libigdgmm12 libxatracker2 mesa-va-drivers \
             mesa-vdpau-drivers mesa-vulkan-drivers va-driver-all vainfo hwinfo clinfo intel-ocloc
+        apt-get install -y libigc-dev intel-igc-cm libigdfcl-dev libigfxcmrt-dev libze-dev
     fi
-    # Development Packages
-    apt-get install -y libigc-dev intel-igc-cm libigdfcl-dev libigfxcmrt-dev libze-dev
 
     # Install Intel Support Packages
     apt-get install -y ${XPU_PACKAGES}
@@ -65,7 +66,7 @@ function install_ubuntu() {
 function install_rhel() {
     . /etc/os-release
     if [[ "${ID}" == "rhel" ]]; then
-        if [[ ! " 8.8 8.10 9.0 9.2 9.3 " =~ " ${VERSION_ID} " ]]; then
+        if [[ ! " 8.8 8.9 9.0 9.2 9.3 " =~ " ${VERSION_ID} " ]]; then
             echo "RHEL version ${VERSION_ID} not supported"
             exit
         fi
@@ -146,7 +147,7 @@ function install_sles() {
 XPU_DRIVER_VERSION=""
 if [[ "${XPU_DRIVER_TYPE,,}" == "lts" ]]; then
     # Use GPU driver LTS releases
-    XPU_DRIVER_VERSION="/lts/2523"
+    XPU_DRIVER_VERSION="/lts/2350"
 fi
 
 # Default use Intel® oneAPI Deep Learning Essentials 2025.1

.ci/docker/libtorch/build.sh

@@ -49,7 +49,11 @@ case ${DOCKER_TAG_PREFIX} in
         fi
         BASE_TARGET=rocm
         GPU_IMAGE=rocm/dev-ubuntu-22.04:${GPU_ARCH_VERSION}-complete
-        PYTORCH_ROCM_ARCH="gfx900;gfx906;gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201;gfx950;gfx1150;gfx1151"
+        PYTORCH_ROCM_ARCH="gfx900;gfx906;gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201"
+        # add gfx950, gfx115x conditionally starting in ROCm 7.0
+        if [[ "$GPU_ARCH_VERSION" == *"7.0"* ]]; then
+            PYTORCH_ROCM_ARCH="${PYTORCH_ROCM_ARCH};gfx950;gfx1150;gfx1151"
+        fi
         DOCKER_GPU_BUILD_ARG="--build-arg PYTORCH_ROCM_ARCH=${PYTORCH_ROCM_ARCH} --build-arg ROCM_VERSION=${GPU_ARCH_VERSION}"
         ;;
     *)

.ci/docker/manywheel/build.sh

@@ -87,7 +87,11 @@ case ${image} in
         MANY_LINUX_VERSION="2_28"
         DEVTOOLSET_VERSION="11"
         GPU_IMAGE=rocm/dev-almalinux-8:${GPU_ARCH_VERSION}-complete
-        PYTORCH_ROCM_ARCH="gfx900;gfx906;gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201;gfx950;gfx1150;gfx1151"
+        PYTORCH_ROCM_ARCH="gfx900;gfx906;gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201"
+        # add gfx950, gfx115x conditionally starting in ROCm 7.0
+        if [[ "$GPU_ARCH_VERSION" == *"7.0"* ]]; then
+            PYTORCH_ROCM_ARCH="${PYTORCH_ROCM_ARCH};gfx950;gfx1150;gfx1151"
+        fi
         DOCKER_GPU_BUILD_ARG="--build-arg ROCM_VERSION=${GPU_ARCH_VERSION} --build-arg PYTORCH_ROCM_ARCH=${PYTORCH_ROCM_ARCH} --build-arg DEVTOOLSET_VERSION=${DEVTOOLSET_VERSION}"
         ;;
     manylinux2_28-builder:xpu)

.ci/docker/ubuntu/Dockerfile

@@ -143,15 +143,6 @@ COPY ci_commit_pins/halide.txt halide.txt
 RUN if [ -n "${HALIDE}" ]; then bash ./install_halide.sh; fi
 RUN rm install_halide.sh common_utils.sh halide.txt
 
-ARG PALLAS
-ARG CUDA_VERSION
-# Install JAX with CUDA support (for Pallas)
-COPY ./common/install_jax.sh install_jax.sh
-COPY ./common/common_utils.sh common_utils.sh
-COPY ./ci_commit_pins/jax.txt /ci_commit_pins/jax.txt
-RUN if [ -n "${PALLAS}" ]; then bash ./install_jax.sh ${CUDA_VERSION}; fi
-RUN rm -f install_jax.sh common_utils.sh /ci_commit_pins/jax.txt
-
 ARG ONNX
 # Install ONNX dependencies
 COPY ./common/install_onnx.sh ./common/common_utils.sh ./

.ci/lumen_cli/cli/lib/common/cli_helper.py

@@ -8,11 +8,9 @@ from abc import ABC, abstractmethod
 
 
 try:
-    from collections.abc import Callable  # Python 3.11+
-    from typing import Any, Required, TypedDict
+    from typing import Any, Callable, Required, TypedDict  # Python 3.11+
 except ImportError:
-    from collections.abc import Callable
-    from typing import Any, TypedDict
+    from typing import Any, Callable, TypedDict
 
     from typing_extensions import Required  # Fallback for Python <3.11
 

.ci/magma-rocm/README.md

@@ -30,6 +30,7 @@ into a tarball, with the following structure:
 More specifically, `build_magma.sh` copies over the relevant files from the `package_files` directory depending on the ROCm version.
 Outputted binaries should be in the `output` folder.
 
+
 ## Pushing
 
 Packages can be uploaded to an S3 bucket using:

.ci/onnx/common.sh

@@ -21,87 +21,3 @@ if [[ "${BUILD_ENVIRONMENT}" == *rocm* ]]; then
 fi
 
 mkdir -p "$pytest_reports_dir" || true
-
-##########################################
-# copied from .ci/pytorch/common_utils.sh
-##########################################
-
-function get_pinned_commit() {
-  cat .github/ci_commit_pins/"${1}".txt
-}
-
-function pip_install_whl() {
-  # This is used to install PyTorch and other build artifacts wheel locally
-  # without using any network connection
-
-  # Convert the input arguments into an array
-  local args=("$@")
-
-  # Check if the first argument contains multiple paths separated by spaces
-  if [[ "${args[0]}" == *" "* ]]; then
-    # Split the string by spaces into an array
-    IFS=' ' read -r -a paths <<< "${args[0]}"
-    # Loop through each path and install individually
-    for path in "${paths[@]}"; do
-      echo "Installing $path"
-      python3 -mpip install --no-index --no-deps "$path"
-    done
-  else
-    # Loop through each argument and install individually
-    for path in "${args[@]}"; do
-      echo "Installing $path"
-      python3 -mpip install --no-index --no-deps "$path"
-    done
-  fi
-}
-
-function pip_build_and_install() {
-  local build_target=$1
-  local wheel_dir=$2
-
-  local found_whl=0
-  for file in "${wheel_dir}"/*.whl
-  do
-    if [[ -f "${file}" ]]; then
-      found_whl=1
-      break
-    fi
-  done
-
-  # Build the wheel if it doesn't exist
-  if [ "${found_whl}" == "0" ]; then
-    python3 -m pip wheel \
-      --no-build-isolation \
-      --no-deps \
-      -w "${wheel_dir}" \
-      "${build_target}"
-  fi
-
-  for file in "${wheel_dir}"/*.whl
-  do
-    pip_install_whl "${file}"
-  done
-}
-
-function install_torchvision() {
-  local orig_preload
-  local commit
-  commit=$(get_pinned_commit vision)
-  orig_preload=${LD_PRELOAD}
-  if [ -n "${LD_PRELOAD}" ]; then
-    # Silence dlerror to work-around glibc ASAN bug, see https://sourceware.org/bugzilla/show_bug.cgi?id=27653#c9
-    echo 'char* dlerror(void) { return "";}'|gcc -fpic -shared -o "${HOME}/dlerror.so" -x c -
-    LD_PRELOAD=${orig_preload}:${HOME}/dlerror.so
-  fi
-
-  if [[ "${BUILD_ENVIRONMENT}" == *cuda* ]]; then
-    # Not sure if both are needed, but why not
-    export FORCE_CUDA=1
-    export WITH_CUDA=1
-  fi
-  pip_build_and_install "git+https://github.com/pytorch/vision.git@${commit}" dist/vision
-
-  if [ -n "${LD_PRELOAD}" ]; then
-    LD_PRELOAD=${orig_preload}
-  fi
-}

.ci/onnx/test.sh

@@ -19,7 +19,7 @@ git config --global --add safe.directory /var/lib/jenkins/workspace
 
 if [[ "$BUILD_ENVIRONMENT" == *onnx* ]]; then
   # TODO: This can be removed later once vision is also part of the Docker image
-  install_torchvision
+  pip install -q --no-use-pep517 "git+https://github.com/pytorch/vision.git@$(cat .github/ci_commit_pins/vision.txt)"
   # JIT C++ extensions require ninja, so put it into PATH.
   export PATH="/var/lib/jenkins/.local/bin:$PATH"
   # NB: ONNX test is fast (~15m) so it's ok to retry it few more times to avoid any flaky issue, we

.ci/pytorch/build.sh

@@ -168,16 +168,14 @@ if [[ "$BUILD_ENVIRONMENT" == *xpu* ]]; then
   # shellcheck disable=SC1091
   source /opt/intel/oneapi/compiler/latest/env/vars.sh
   # shellcheck disable=SC1091
-  source /opt/intel/oneapi/umf/latest/env/vars.sh
-  # shellcheck disable=SC1091
   source /opt/intel/oneapi/ccl/latest/env/vars.sh
   # shellcheck disable=SC1091
   source /opt/intel/oneapi/mpi/latest/env/vars.sh
-  # shellcheck disable=SC1091
-  source /opt/intel/oneapi/pti/latest/env/vars.sh
   # Enable XCCL build
   export USE_XCCL=1
   export USE_MPI=0
+  # XPU kineto feature dependencies are not fully ready, disable kineto build as temp WA
+  export USE_KINETO=0
   export TORCH_XPU_ARCH_LIST=pvc
 fi
 

.ci/pytorch/common_utils.sh

@@ -96,6 +96,7 @@ function pip_build_and_install() {
     python3 -m pip wheel \
       --no-build-isolation \
       --no-deps \
+      --no-use-pep517 \
       -w "${wheel_dir}" \
       "${build_target}"
   fi
@@ -307,28 +308,6 @@ function install_torchao() {
   pip_build_and_install "git+https://github.com/pytorch/ao.git@${commit}" dist/ao
 }
 
-function install_flash_attn_cute() {
-  echo "Installing FlashAttention CuTe from GitHub..."
-  # Grab latest main til we have a pinned commit
-  local flash_attn_commit
-  flash_attn_commit=$(git ls-remote https://github.com/Dao-AILab/flash-attention.git HEAD | cut -f1)
-
-  # Clone the repo to a temporary directory
-  rm -rf flash-attention-build
-  git clone --depth 1 --recursive https://github.com/Dao-AILab/flash-attention.git flash-attention-build
-
-  pushd flash-attention-build
-  git checkout "${flash_attn_commit}"
-
-  # Install only the 'cute' sub-directory
-  pip_install -e flash_attn/cute/
-  popd
-
-  # remove the local repo
-  rm -rf flash-attention-build
-  echo "FlashAttention CuTe installation complete."
-}
-
 function print_sccache_stats() {
   echo 'PyTorch Build Statistics'
   sccache --show-stats

.ci/pytorch/smoke_test/smoke_test.py

@@ -353,17 +353,6 @@ def test_linalg(device="cpu") -> None:
             torch.linalg.svd(A)
 
 
-def test_sdpa(device="cpu", dtype=torch.float16) -> None:
-    """Regression test for https://github.com/pytorch/pytorch/issues/167602
-    Without nvrtc_builtins on CuDNN-9.13 on CUDA-13 fails with ` No valid execution plans built.`
-    """
-    print(f"Testing SDPA on {device} using type {dtype}")
-    k, q, v = torch.rand(3, 1, 16, 77, 64, dtype=dtype, device=device).unbind(0)
-    attn = torch.rand(1, 1, 77, 77, dtype=dtype, device=device)
-    rc = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn)
-    assert rc.isnan().any().item() is False
-
-
 def smoke_test_compile(device: str = "cpu") -> None:
     supported_dtypes = [torch.float16, torch.float32, torch.float64]
 
@@ -500,12 +489,10 @@ def main() -> None:
     smoke_test_conv2d()
     test_linalg()
     test_numpy()
-    test_sdpa()
 
     if is_cuda_system:
         test_linalg("cuda")
         test_cuda_gds_errors_captured()
-        test_sdpa("cuda")
 
     if options.package == "all":
         smoke_test_modules()

.ci/pytorch/test.sh

@@ -208,8 +208,6 @@ if [[ "$BUILD_ENVIRONMENT" == *xpu* ]]; then
   source /opt/intel/oneapi/ccl/latest/env/vars.sh
   # shellcheck disable=SC1091
   source /opt/intel/oneapi/mpi/latest/env/vars.sh
-  # shellcheck disable=SC1091
-  source /opt/intel/oneapi/pti/latest/env/vars.sh
   # Check XPU status before testing
   timeout 30 xpu-smi discovery || true
 fi
@@ -344,18 +342,8 @@ test_python_smoke() {
 }
 
 test_python_smoke_b200() {
-  # Targeted smoke tests for B200 including FlashAttention CuTe coverage
-  install_flash_attn_cute
-  time python test/run_test.py \
-    --include \
-      test_matmul_cuda \
-      test_scaled_matmul_cuda \
-      inductor/test_fp8 \
-      nn/attention/test_fa4 \
-      nn/attention/test_open_registry \
-      inductor/test_flex_flash \
-    $PYTHON_TEST_EXTRA_OPTION \
-    --upload-artifacts-while-running
+  # Targeted smoke tests for B200 - staged approach to avoid too many failures
+  time python test/run_test.py --include test_matmul_cuda test_scaled_matmul_cuda inductor/test_fp8 $PYTHON_TEST_EXTRA_OPTION --upload-artifacts-while-running
   assert_git_not_dirty
 }
 
@@ -389,13 +377,6 @@ test_lazy_tensor_meta_reference_disabled() {
   export -n TORCH_DISABLE_FUNCTIONALIZATION_META_REFERENCE
 }
 
-test_dynamo_core() {
-  time python test/run_test.py \
-    --include-dynamo-core-tests \
-    --verbose \
-    --upload-artifacts-while-running
-  assert_git_not_dirty
-}
 
 test_dynamo_wrapped_shard() {
   if [[ -z "$NUM_TEST_SHARDS" ]]; then
@@ -843,11 +824,6 @@ test_inductor_halide() {
   assert_git_not_dirty
 }
 
-test_inductor_pallas() {
-  python test/run_test.py --include inductor/test_pallas.py --verbose
-  assert_git_not_dirty
-}
-
 test_inductor_triton_cpu() {
   python test/run_test.py --include inductor/test_triton_cpu_backend.py inductor/test_torchinductor_strided_blocks.py --verbose
   assert_git_not_dirty
@@ -1250,97 +1226,6 @@ test_custom_script_ops() {
   assert_git_not_dirty
 }
 
-test_libtorch_agnostic_targetting() {
-    echo "Testing libtorch_agnostic runs correctly on TORCH_TARGET_VERSION"
-
-    REPO_DIR=$(pwd)
-    WHEEL_DIR="${REPO_DIR}/test/cpp_extensions/.wheels"
-
-    # Build wheel with current PyTorch (this has TORCH_TARGET_VERSION 2_9_0)
-    echo "Building 2.9 extension wheel with current PyTorch..."
-    pushd test/cpp_extensions/libtorch_agnostic_2_9_extension
-    time python setup.py bdist_wheel
-
-    # Save the wheel
-    mkdir -p "$WHEEL_DIR"
-    cp dist/*.whl "$WHEEL_DIR/"
-    WHEEL_FILE=$(find "$WHEEL_DIR" -maxdepth 1 -name "*.whl" -type f | head -1)
-    echo "Built wheel: $(basename "$WHEEL_FILE")"
-    popd
-
-    # Create venv and install PyTorch 2.9
-    python -m venv venv_pytorch_2_9
-    # shellcheck disable=SC1091
-    . venv_pytorch_2_9/bin/activate
-
-    # Clear PYTHONPATH to avoid using the development PyTorch
-    echo "Clearing PYTHONPATH to use only venv packages..."
-    unset PYTHONPATH
-
-    # Upgrade pip to latest version
-    echo "Upgrading pip to latest version..."
-    pip install --upgrade pip
-    pip --version
-
-    echo "Installing PyTorch 2.9..."
-
-    # Install from release channel only
-    PYTORCH_VERSION="2.9.0"
-
-    # Extract CUDA version from BUILD_ENVIRONMENT (e.g., "cuda12.1" -> "cu121")
-    if [[ "$BUILD_ENVIRONMENT" =~ cuda([0-9]+)\.([0-9]+) ]]; then
-        CUDA_MAJOR="${BASH_REMATCH[1]}"
-        CUDA_MINOR="${BASH_REMATCH[2]}"
-        CUDA_VERSION="cu${CUDA_MAJOR}${CUDA_MINOR}"
-        echo "  Detected CUDA ${CUDA_MAJOR}.${CUDA_MINOR} from BUILD_ENVIRONMENT, using ${CUDA_VERSION}"
-    else
-        # Default to CPU build
-        CUDA_VERSION="cpu"
-        echo "  No CUDA detected in BUILD_ENVIRONMENT, using CPU build"
-    fi
-
-    if pip install torch=="${PYTORCH_VERSION}" --index-url https://download.pytorch.org/whl/${CUDA_VERSION}/; then
-        echo "Installed PyTorch ${PYTORCH_VERSION} from release channel (${CUDA_VERSION})"
-    else
-        echo "  FAILED to install PyTorch 2.9.0 from release channel"
-        echo "  URL: https://download.pytorch.org/whl/${CUDA_VERSION}/"
-        deactivate
-        rm -rf venv_pytorch_2_9
-        return 1
-    fi
-
-    INSTALLED_VERSION=$(python -c "import torch; print(torch.__version__)" 2>/dev/null || echo "unknown")
-    echo "  Installed version: $INSTALLED_VERSION"
-
-    # Install test dependencies
-    echo "Installing test dependencies..."
-    pip install expecttest numpy unittest-xml-reporting
-
-    # Install the pre-built wheel
-    echo ""
-    echo "Installing pre-built 2.9 extension wheel (built with PyTorch 2.10)..."
-    pip install "$WHEEL_FILE"
-    echo "Installed $(basename "$WHEEL_FILE") into PyTorch 2.9 environment"
-
-    # Run tests with PyTorch 2.9 runtime (2.10 tests will be skipped automatically)
-    echo ""
-    echo "Running tests with PyTorch 2.9 runtime (using wheel built on PyTorch 2.10)..."
-    if time python test/cpp_extensions/test_libtorch_agnostic.py -v; then
-        echo ""
-        echo "  Wheel built with current torch and TORCH_TARGET_VERSION 2_9_0 works with PyTorch 2.9 runtime!"
-    else
-        echo "targeting test failed"
-        deactivate
-        rm -rf venv_pytorch_2_9 "$WHEEL_DIR"
-        return 1
-    fi
-
-    deactivate
-    rm -rf venv_pytorch_2_9 "$WHEEL_DIR"
-
-    assert_git_not_dirty
-}
-
 test_jit_hooks() {
   echo "Testing jit hooks in cpp"
   HOOK_BUILD="${CUSTOM_TEST_ARTIFACT_BUILD_DIR}/jit-hook-build"
@@ -1778,22 +1663,6 @@ test_operator_microbenchmark() {
   done
 }
 
-test_attention_microbenchmark() {
-  TEST_REPORTS_DIR=$(pwd)/test/test-reports
-  mkdir -p "$TEST_REPORTS_DIR"
-  TEST_DIR=$(pwd)
-
-  # Install attention-gym dependency
-  echo "Installing attention-gym..."
-  python -m pip install git+https://github.com/meta-pytorch/attention-gym.git@main
-  pip show triton
-
-  cd "${TEST_DIR}"/benchmarks/transformer
-
-  $TASKSET python score_mod.py --config configs/config_basic.yaml \
-    --output-json-for-dashboard "${TEST_REPORTS_DIR}/attention_microbenchmark.json"
-}
-
 if ! [[ "${BUILD_ENVIRONMENT}" == *libtorch* || "${BUILD_ENVIRONMENT}" == *-bazel-* ]]; then
   (cd test && python -c "import torch; print(torch.__config__.show())")
   (cd test && python -c "import torch; print(torch.__config__.parallel_info())")
@@ -1813,8 +1682,6 @@ elif [[ "${BUILD_ENVIRONMENT}" == *aarch64* && "${TEST_CONFIG}" == 'default' ]];
 elif [[ "${TEST_CONFIG}" == *backward* ]]; then
   test_forward_backward_compatibility
   # Do NOT add tests after bc check tests, see its comment.
-elif [[ "${TEST_CONFIG}" == *libtorch_agnostic_targetting* ]]; then
-  test_libtorch_agnostic_targetting
 elif [[ "${TEST_CONFIG}" == *xla* ]]; then
   install_torchvision
   build_xla
@@ -1853,14 +1720,10 @@ elif [[ "${TEST_CONFIG}" == *operator_benchmark* ]]; then
   fi
 elif [[ "${TEST_CONFIG}" == *operator_microbenchmark* ]]; then
   test_operator_microbenchmark
-elif [[ "${TEST_CONFIG}" == *attention_microbenchmark* ]]; then
-  test_attention_microbenchmark
 elif [[ "${TEST_CONFIG}" == *inductor_distributed* ]]; then
   test_inductor_distributed
 elif [[ "${TEST_CONFIG}" == *inductor-halide* ]]; then
   test_inductor_halide
-elif [[ "${TEST_CONFIG}" == *inductor-pallas* ]]; then
-  test_inductor_pallas
 elif [[ "${TEST_CONFIG}" == *inductor-triton-cpu* ]]; then
   test_inductor_triton_cpu
 elif [[ "${TEST_CONFIG}" == *inductor-micro-benchmark* ]]; then
@@ -1914,8 +1777,6 @@ elif [[ "${TEST_CONFIG}" == *inductor* ]]; then
   test_inductor_shard "${SHARD_NUMBER}"
 elif [[ "${TEST_CONFIG}" == *einops* ]]; then
   test_einops
-elif [[ "${TEST_CONFIG}" == *dynamo_core* ]]; then
-  test_dynamo_core
 elif [[ "${TEST_CONFIG}" == *dynamo_wrapped* ]]; then
   install_torchvision
   test_dynamo_wrapped_shard "${SHARD_NUMBER}"

.github/actionlint.yaml

@@ -63,7 +63,7 @@ self-hosted-runner:
     - linux.rocm.gpu.gfx942.1
     - linux.rocm.gpu.gfx942.2
     - linux.rocm.gpu.gfx942.4
-    - linux.rocm.gfx942.docker-cache
+    - rocm-docker
     # Org wise AWS `mac2.metal` runners (2020 Mac mini hardware powered by Apple silicon M1 processors)
     - macos-m1-stable
     - macos-m1-14

.github/ci_commit_pins/audio.txt

@@ -1 +1 @@
-ee1a1350eb37804b94334768f328144f058f14e9
+ad5816f0eee1c873df1b7d371c69f1f811a89387

.github/ci_commit_pins/vision.txt

@@ -1 +1 @@
-2d82dc5caa336d179d9b46ac4a0fb8c43d84c5cc
+cfbc5c2f1c798991715a6b06bb3ce46478c4487c

.github/ci_commit_pins/xla.txt

@@ -1 +1 @@
-94631807d22c09723dd006f7be5beb649d5f88d0
+c8b09f5f77d6bf6fb7ed7a9aa83e5d8156b3a5e9

.github/labeler.yml

@@ -138,8 +138,7 @@
 - test/test_matmul_cuda.py
 - test/test_scaled_matmul_cuda.py
 - test/inductor/test_fp8.py
-- aten/src/ATen/native/cuda/*Blas.cpp
-- aten/src/ATen/cuda/CUDA*Blas.*
+- aten/src/ATen/native/cuda/Blas.cpp
 - torch/**/*cublas*
 - torch/_inductor/kernel/mm.py
 - test/inductor/test_max_autotune.py
@@ -149,8 +148,7 @@
 - test/test_matmul_cuda.py
 - test/test_scaled_matmul_cuda.py
 - test/inductor/test_fp8.py
-- aten/src/ATen/native/cuda/*Blas.cpp
-- aten/src/ATen/cuda/CUDA*Blas.*
+- aten/src/ATen/native/cuda/Blas.cpp
 - torch/**/*cublas*
 - torch/_inductor/kernel/mm.py
 - test/inductor/test_max_autotune.py
@@ -160,21 +158,7 @@
 - test/test_matmul_cuda.py
 - test/test_scaled_matmul_cuda.py
 - test/inductor/test_fp8.py
-- aten/src/ATen/native/cuda/*Blas.cpp
-- aten/src/ATen/cuda/CUDA*Blas.*
+- aten/src/ATen/native/cuda/Blas.cpp
 - torch/_inductor/kernel/mm.py
 - test/inductor/test_max_autotune.py
 - third_party/fbgemm
-
-"ciflow/mps":
-- aten/src/ATen/mps/**
-- aten/src/ATen/native/mps/**
-- torch/_inductor/codegen/mps.py
-- test/test_mps.py
-- test/inductor/test_mps_basic.py
-
-"ciflow/h100-symm-mem":
-- torch/csrc/distributed/c10d/symm_mem/**
-- torch/distributed/_symmetric_memory/**
-- test/distributed/**/*mem*
-- test/distributed/**/*mem*/**

.github/nitpicks.yml

@@ -10,4 +10,3 @@
   pathFilter:
     - 'torch/csrc/inductor/aoti_torch/c/*'
     - 'torch/csrc/inductor/aoti_torch/generated/*'
-    - 'torch/csrc/stable/c/*'

.github/pytorch-probot.yml

@@ -2,12 +2,11 @@ tracking_issue: 24422
 ciflow_tracking_issue: 64124
 ciflow_push_tags:
 - ciflow/b200
-- ciflow/b200-distributed
 - ciflow/b200-symm-mem
+- ciflow/b200-distributed
 - ciflow/binaries
 - ciflow/binaries_libtorch
 - ciflow/binaries_wheel
-- ciflow/dynamo
 - ciflow/h100
 - ciflow/h100-cutlass-backend
 - ciflow/h100-distributed
@@ -23,8 +22,6 @@ ciflow_push_tags:
 - ciflow/inductor-perf-test-nightly-xpu
 - ciflow/inductor-periodic
 - ciflow/inductor-rocm
-- ciflow/inductor-rocm-mi200
-- ciflow/inductor-rocm-mi300
 - ciflow/linux-aarch64
 - ciflow/mps
 - ciflow/nightly
@@ -36,13 +33,11 @@ ciflow_push_tags:
 - ciflow/quantization-periodic
 - ciflow/riscv64
 - ciflow/rocm
-- ciflow/rocm-mi200
 - ciflow/rocm-mi300
 - ciflow/rocm-mi355
 - ciflow/rocm-navi31
 - ciflow/s390
 - ciflow/slow
-- ciflow/slow-rocm-mi200
 - ciflow/torchbench
 - ciflow/triton_binaries
 - ciflow/trunk

.github/scripts/delete_old_branches.py

@@ -1,11 +1,10 @@
 # Delete old branches
 import os
 import re
-from collections.abc import Callable
 from datetime import datetime
 from functools import lru_cache
 from pathlib import Path
-from typing import Any
+from typing import Any, Callable
 
 from github_utils import gh_fetch_json_dict, gh_graphql
 from gitutils import GitRepo

.github/scripts/filter_test_configs.py

@@ -8,11 +8,10 @@ import re
 import subprocess
 import sys
 import warnings
-from collections.abc import Callable
 from enum import Enum
 from functools import cache
 from logging import info
-from typing import Any, Optional
+from typing import Any, Callable, Optional
 from urllib.request import Request, urlopen
 
 import yaml

.github/scripts/generate_pytorch_version.py

@@ -50,7 +50,7 @@ def get_tag() -> str:
 
 def get_base_version() -> str:
     root = get_pytorch_root()
-    dirty_version = Path(root / "version.txt").read_text().strip()
+    dirty_version = open(root / "version.txt").read().strip()
     # Strips trailing a0 from version.txt, not too sure why it's there in the
     # first place
     return re.sub(LEGACY_BASE_VERSION_SUFFIX_PATTERN, "", dirty_version)

.github/scripts/get_workflow_job_id.py

@@ -11,8 +11,7 @@ import sys
 import time
 import urllib
 import urllib.parse
-from collections.abc import Callable
-from typing import Any, Optional
+from typing import Any, Callable, Optional
 from urllib.request import Request, urlopen
 
 

.github/scripts/github_utils.py

@@ -3,9 +3,8 @@
 import json
 import os
 import warnings
-from collections.abc import Callable
 from dataclasses import dataclass
-from typing import Any, cast, Optional, Union
+from typing import Any, Callable, cast, Optional, Union
 from urllib.error import HTTPError
 from urllib.parse import quote
 from urllib.request import Request, urlopen

.github/scripts/gitutils.py

@@ -4,10 +4,10 @@ import os
 import re
 import tempfile
 from collections import defaultdict
-from collections.abc import Callable, Iterator
+from collections.abc import Iterator
 from datetime import datetime
 from functools import wraps
-from typing import Any, cast, Optional, TypeVar, Union
+from typing import Any, Callable, cast, Optional, TypeVar, Union
 
 
 T = TypeVar("T")

.github/scripts/lintrunner.sh

@@ -34,9 +34,6 @@ python3 torch/utils/data/datapipes/gen_pyi.py
 # Also check generated pyi files
 find torch -name '*.pyi' -exec git add --force -- "{}" +
 
-# Print current environment
-python3 -m pip freeze
-
 RC=0
 # Run lintrunner on all files
 if ! lintrunner --force-color --tee-json=lint.json ${ADDITIONAL_LINTRUNNER_ARGS} 2> /dev/null; then

.github/scripts/trymerge.py

@@ -17,12 +17,12 @@ import re
 import time
 import urllib.parse
 from collections import defaultdict
-from collections.abc import Callable, Iterable
+from collections.abc import Iterable
 from dataclasses import dataclass
 from functools import cache
 from pathlib import Path
 from re import Pattern
-from typing import Any, cast, NamedTuple, Optional
+from typing import Any, Callable, cast, NamedTuple, Optional
 from warnings import warn
 
 import yaml

.github/workflows/_linux-build.yml

@@ -330,6 +330,8 @@ jobs:
             -e PR_NUMBER \
             -e SHA1 \
             -e BRANCH \
+            -e SCCACHE_BUCKET \
+            -e SCCACHE_REGION \
             -e XLA_CUDA \
             -e XLA_CLANG_CACHE_S3_BUCKET_NAME \
             -e SKIP_SCCACHE_INITIALIZATION=1 \

.github/workflows/_linux-test.yml

@@ -326,7 +326,7 @@ jobs:
           SCCACHE_BUCKET: ${{ !contains(matrix.runner, 'b200') && 'ossci-compiler-cache-circleci-v2' || '' }}
           SCCACHE_REGION: ${{ !contains(matrix.runner, 'b200') && 'us-east-1' || '' }}
           SHM_SIZE: ${{ contains(inputs.build-environment, 'cuda') && '2g' || '1g' }}
-          DOCKER_IMAGE: ${{ steps.calculate-docker-image.outputs.docker-image }}
+          DOCKER_IMAGE: ${{ inputs.docker-image }}
           XLA_CUDA: ${{ contains(inputs.build-environment, 'xla') && '0' || '' }}
           XLA_CLANG_CACHE_S3_BUCKET_NAME: ossci-compiler-clang-cache-circleci-xla
           PYTORCH_TEST_CUDA_MEM_LEAK_CHECK: ${{ matrix.mem_leak_check && '1' || '0' }}

.github/workflows/attention_op_microbenchmark.yml

@@ -1,73 +0,0 @@
-name: attention_op_microbenchmark
-
-on:
-  push:
-    tags:
-      - ciflow/op-benchmark/*
-  workflow_dispatch:
-  schedule:
-    # Run at 06:00 UTC everyday
-    - cron: 0 7 * * *
-
-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:
-  id-token: write
-  contents: read
-
-jobs:
-  attn-microbenchmark-build:
-    if: github.repository_owner == 'pytorch'
-    uses: ./.github/workflows/_linux-build.yml
-    with:
-      runner: linux.12xlarge.memory
-      build-environment: linux-jammy-cuda12.8-py3.10-gcc9-sm80
-      docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11
-      cuda-arch-list: '8.0 9.0'
-      test-matrix: |
-        { include: [
-          { config: "attention_microbenchmark_test", shard: 1, num_shards: 1, runner: "linux.aws.a100" },
-          { config: "attention_microbenchmark_test", shard: 1, num_shards: 1, runner: "linux.aws.h100" },
-        ]}
-    secrets: inherit
-
-  attn-microbenchmark-test:
-    name: attn-microbenchmark-test
-    uses: ./.github/workflows/_linux-test.yml
-    needs: attn-microbenchmark-build
-    with:
-      timeout-minutes: 500
-      build-environment: linux-jammy-cuda12.8-py3.10-gcc9-sm80
-      docker-image: ${{ needs.attn-microbenchmark-build.outputs.docker-image }}
-      test-matrix: ${{ needs.attn-microbenchmark-build.outputs.test-matrix }}
-    secrets: inherit
-
-  # B200 runner
-  opmicrobenchmark-build-b200:
-    if: github.repository_owner == 'pytorch'
-    name: opmicrobenchmark-build-b200
-    uses: ./.github/workflows/_linux-build.yml
-    with:
-      runner: linux.12xlarge.memory
-      build-environment: linux-jammy-cuda12.8-py3.10-gcc9-sm100
-      docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11
-      cuda-arch-list: '10.0'
-      test-matrix: |
-        { include: [
-          { config: "operator_microbenchmark_test", shard: 1, num_shards: 1, runner: "linux.dgx.b200" },
-        ]}
-    secrets: inherit
-
-  opmicrobenchmark-test-b200:
-    name: opmicrobenchmark-test-b200
-    uses: ./.github/workflows/_linux-test.yml
-    needs: opmicrobenchmark-build-b200
-    with:
-      timeout-minutes: 500
-      build-environment: linux-jammy-cuda12.8-py3.10-gcc9-sm100
-      docker-image: ${{ needs.opmicrobenchmark-build-b200.outputs.docker-image }}
-      test-matrix: ${{ needs.opmicrobenchmark-build-b200.outputs.test-matrix }}
-      aws-role-to-assume: arn:aws:iam::308535385114:role/gha_workflow_s3_and_ecr_read_only
-    secrets: inherit

.github/workflows/docker-builds.yml

@@ -56,8 +56,6 @@ jobs:
           pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc9,
           pytorch-linux-jammy-cuda12.4-cudnn9-py3-gcc11,
           pytorch-linux-jammy-py3.10-clang12,
-          pytorch-linux-jammy-py3.11-clang12,
-          pytorch-linux-jammy-py3.12-clang12,
           pytorch-linux-jammy-py3.13-clang12,
           pytorch-linux-jammy-py3.14-clang12,
           pytorch-linux-jammy-rocm-n-py3,
@@ -67,10 +65,9 @@ jobs:
           pytorch-linux-jammy-py3.10-gcc11,
           pytorch-linux-jammy-py3-gcc11-inductor-benchmarks,
           pytorch-linux-jammy-py3.12-halide,
-          pytorch-linux-jammy-cuda12.8-py3.12-pallas,
           pytorch-linux-jammy-xpu-n-1-py3,
-          pytorch-linux-noble-xpu-n-py3,
-          pytorch-linux-noble-xpu-n-py3-inductor-benchmarks,
+          pytorch-linux-jammy-xpu-n-py3,
+          pytorch-linux-jammy-xpu-n-py3-inductor-benchmarks,
           pytorch-linux-jammy-py3-clang18-asan,
           pytorch-linux-jammy-py3-clang12-onnx,
           pytorch-linux-jammy-linter,
@@ -119,22 +116,6 @@ jobs:
         with:
           docker-image: ${{ steps.build-docker-image.outputs.docker-image }}
 
-      - name: Generate output
-        if: contains(matrix.docker-image-name, 'rocm')
-        id: generate_output
-        run: |
-          docker_image_name="${{ matrix.docker-image-name }}"
-          docker_image_tag="${{ steps.build-docker-image.outputs.docker-image }}"
-          echo "${docker_image_name}=${docker_image_tag}" >> docker-builds-output-${docker_image_name}.txt
-
-      - name: Upload artifacts
-        uses: actions/upload-artifact@v4.4.0
-        if: contains(matrix.docker-image-name, 'rocm')
-        with:
-          name: docker-builds-artifacts-${{ matrix.docker-image-name }}
-          retention-days: 14
-          path: ./docker-builds-output-${{ matrix.docker-image-name }}.txt
-
       - uses: nick-fields/retry@7152eba30c6575329ac0576536151aca5a72780e # v3.0.0
         name: Push to https://ghcr.io/
         id: push-to-ghcr-io

.github/workflows/docker-cache-mi300.yml

@@ -0,0 +1,55 @@
+name: docker-cache-mi300
+
+on:
+  # run every 6 hours
+  schedule:
+    - cron: 0 0,6,12,18 * * *
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name }}
+  cancel-in-progress: true
+
+permissions:
+  id-token: write
+  contents: read
+
+jobs:
+  docker-cache:
+    if: github.repository_owner == 'pytorch'
+    runs-on: rocm-docker
+    steps:
+      - name: Checkout PyTorch
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        with:
+          no-sudo: true
+
+      - name: configure aws credentials
+        id: aws_creds
+        uses: aws-actions/configure-aws-credentials@ececac1a45f3b08a01d2dd070d28d111c5fe6722 # v4.1.0
+        with:
+          role-to-assume: arn:aws:iam::308535385114:role/gha_workflow_s3_and_ecr_read_only
+          aws-region: us-east-1
+          role-duration-seconds: 18000
+
+      - name: Login to Amazon ECR
+        id: login-ecr
+        continue-on-error: false
+        uses: aws-actions/amazon-ecr-login@062b18b96a7aff071d4dc91bc00c4c1a7945b076 # v2.0.1
+
+      - name: Calculate docker image
+        id: calculate-docker-image
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        with:
+          docker-image-name: ci-image:pytorch-linux-jammy-rocm-n-py3
+          push: false
+
+      - name: Pull docker image
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        with:
+          docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
+
+      - name: Tar and upload to S3 bucket
+        run: |
+          sudo docker save -o ~/docker-data/pytorch/pytorch_docker_image.tar ${{ steps.calculate-docker-image.outputs.docker-image }}
+          sudo rclone copy -P --s3-upload-concurrency 64 --s3-chunk-size 200M --s3-upload-cutoff 300M ~/docker-data/pytorch/pytorch_docker_image.tar oci:pytorchbucket0002/pytorch_docker_image --progress

.github/workflows/docker-cache-rocm.yml

@@ -1,105 +0,0 @@
-name: docker-cache-rocm
-
-on:
-  workflow_run:
-    workflows: [docker-builds]
-    branches: [main, release]
-    types:
-      - completed
-  workflow_dispatch:
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name }}
-  cancel-in-progress: true
-
-permissions:
-  id-token: write
-  contents: read
-  actions: read
-
-jobs:
-  download-docker-builds-artifacts:
-    if: github.repository_owner == 'pytorch'
-    name: download-docker-builds-artifacts
-    runs-on: ubuntu-latest
-    outputs:
-      pytorch-linux-jammy-rocm-n-py3: ${{ steps.process-artifacts.outputs.pytorch-linux-jammy-rocm-n-py3 }}
-      pytorch-linux-noble-rocm-n-py3: ${{ steps.process-artifacts.outputs.pytorch-linux-noble-rocm-n-py3 }}
-      pytorch-linux-jammy-rocm-n-py3-benchmarks: ${{ steps.process-artifacts.outputs.pytorch-linux-jammy-rocm-n-py3-benchmarks }}
-    steps:
-      - name: Download artifacts
-        uses: actions/download-artifact@v4.1.7
-        with:
-          run-id: ${{ github.event.workflow_run.id }}
-          path: ./docker-builds-artifacts
-          merge-multiple: true
-          github-token: ${{ secrets.GITHUB_TOKEN }}
-
-      - name: Process artifacts
-        id: process-artifacts
-        run: |
-          ls -R ./docker-builds-artifacts
-          cat ./docker-builds-artifacts/*txt >> "${GITHUB_OUTPUT}"
-          cat "${GITHUB_OUTPUT}"
-
-  docker-cache:
-    if: github.repository_owner == 'pytorch'
-    needs: download-docker-builds-artifacts
-    strategy:
-      fail-fast: false
-      matrix:
-        runner: [linux.rocm.gfx942.docker-cache]
-        docker-image: [
-          "${{ needs.download-docker-builds-artifacts.outputs.pytorch-linux-jammy-rocm-n-py3 }}",
-          "${{ needs.download-docker-builds-artifacts.outputs.pytorch-linux-noble-rocm-n-py3 }}",
-          "${{ needs.download-docker-builds-artifacts.outputs.pytorch-linux-jammy-rocm-n-py3-benchmarks }}"
-        ]
-    runs-on: "${{ matrix.runner }}"
-    steps:
-      - name: debug
-        run: |
-          JSON_STRINGIFIED="${{ toJSON(needs.download-docker-builds-artifacts.outputs) }}"
-          echo "Outputs of download-docker-builds-artifacts job: ${JSON_STRINGIFIED}"
-
-      - name: configure aws credentials
-        id: aws_creds
-        uses: aws-actions/configure-aws-credentials@ececac1a45f3b08a01d2dd070d28d111c5fe6722 # v4.1.0
-        with:
-          role-to-assume: arn:aws:iam::308535385114:role/gha_workflow_s3_and_ecr_read_only
-          aws-region: us-east-1
-          role-duration-seconds: 18000
-
-      - name: Login to Amazon ECR
-        id: login-ecr
-        continue-on-error: false
-        uses: aws-actions/amazon-ecr-login@062b18b96a7aff071d4dc91bc00c4c1a7945b076 # v2.0.1
-
-      - name: Generate ghrc.io tag
-        id: ghcr-io-tag
-        run: |
-            ecr_image="${{ matrix.docker-image }}"
-            ghcr_image="ghcr.io/pytorch/ci-image:${ecr_image##*:}"
-            echo "ghcr_image=${ghcr_image}" >> "$GITHUB_OUTPUT"
-
-      - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
-        with:
-          docker-image: ${{ steps.ghcr-io-tag.outputs.ghcr_image }}
-
-      - name: Save as tarball
-        run: |
-          docker_image_tag=${{ matrix.docker-image }}
-          docker_image_tag="${docker_image_tag#*:}" # Remove everything before and including first ":"
-          docker_image_tag="${docker_image_tag%-*}" # Remove everything after and including last "-"
-          ref_name=${{ github.event.workflow_run.head_branch }}
-          if [[ $ref_name =~ "release/" ]]; then
-            ref_suffix="release"
-          elif [[ $ref_name == "main" ]]; then
-            ref_suffix="main"
-          else
-            echo "Unexpected branch in ref_name: ${ref_name}" && exit 1
-          fi
-          docker tag ${{ steps.ghcr-io-tag.outputs.ghcr_image }} ${{ matrix.docker-image }}
-          # mv is atomic operation, so we use intermediate tar.tmp file to prevent read-write contention
-          docker save -o ~/pytorch-data/docker/${docker_image_tag}.tar.tmp ${{ matrix.docker-image }}
-          mv ~/pytorch-data/docker/${docker_image_tag}.tar.tmp ~/pytorch-data/docker/${docker_image_tag}_${ref_suffix}.tar

.github/workflows/dynamo-unittest.yml

@@ -1,70 +0,0 @@
-# Workflow: Dynamo Unit Test
-# runs unit tests for dynamo.
-name: dynamo-unittest
-
-on:
-  push:
-    tags:
-      - ciflow/dynamo/*
-  workflow_call:
-  schedule:
-    - cron: 29 8 * * * # about 1:29am PDT
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
-  cancel-in-progress: true
-
-permissions:
-  id-token: write
-  contents: read
-
-jobs:
-  get-label-type:
-    name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
-    if: ${{ (github.event_name != 'schedule' || github.repository == 'pytorch/pytorch') && github.repository_owner == 'pytorch' }}
-    with:
-      triggering_actor: ${{ github.triggering_actor }}
-      issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
-      curr_branch: ${{ github.head_ref || github.ref_name }}
-      curr_ref_type: ${{ github.ref_type }}
-      opt_out_experiments: lf
-
-  dynamo-build:
-    name: dynamo-build
-    uses: ./.github/workflows/_linux-build.yml
-    needs: get-label-type
-    strategy:
-      matrix:
-        python-version: ['3.11', '3.12']
-    with:
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build-environment: linux-jammy-py${{ matrix.python-version }}-clang12
-      docker-image-name: ci-image:pytorch-linux-jammy-py${{ matrix.python-version }}-clang12
-      test-matrix: |
-        { include: [
-          { config: "dynamo_core", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.c7i.2xlarge" },
-          { config: "dynamo_wrapped", shard: 1, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.c7i.2xlarge" },
-          { config: "dynamo_wrapped", shard: 2, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.c7i.2xlarge" },
-          { config: "dynamo_wrapped", shard: 3, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.c7i.2xlarge" },
-        ]}
-    secrets: inherit
-
-  dynamo-test:
-    name: dynamo-test
-    uses: ./.github/workflows/_linux-test.yml
-    needs: [get-label-type, dynamo-build]
-    strategy:
-      matrix:
-        python-version: ['3.11', '3.12']
-    with:
-      build-environment: linux-jammy-py${{ matrix.python-version }}-clang12
-      docker-image: ci-image:pytorch-linux-jammy-py${{ matrix.python-version }}-clang12
-      test-matrix: |
-        { include: [
-          { config: "dynamo_core", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.c7i.2xlarge" },
-          { config: "dynamo_wrapped", shard: 1, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.c7i.2xlarge" },
-          { config: "dynamo_wrapped", shard: 2, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.c7i.2xlarge" },
-          { config: "dynamo_wrapped", shard: 3, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.c7i.2xlarge" },
-        ]}
-    secrets: inherit

.github/workflows/h100-distributed.yml

@@ -37,6 +37,7 @@ jobs:
     needs: get-label-type
     with:
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      runner: "linux.c7i.12xlarge"
       build-environment: linux-jammy-cuda12.8-py3.10-gcc11-sm90-dist
       docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11
       cuda-arch-list: '9.0'

.github/workflows/inductor-perf-test-nightly-xpu.yml

@@ -83,9 +83,9 @@ jobs:
     needs: get-label-type
     with:
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build-environment: linux-noble-xpu-n-py3.10
-      docker-image-name: ci-image:pytorch-linux-noble-xpu-n-py3-inductor-benchmarks
-      runner: linux.c7i.2xlarge
+      build-environment: linux-jammy-xpu-n-py3.10
+      docker-image-name: ci-image:pytorch-linux-jammy-xpu-n-py3-inductor-benchmarks
+      runner: linux.c7i.12xlarge
       test-matrix: |
         { include: [
           { config: "inductor_huggingface_perf_xpu", shard: 1, num_shards: 5, runner: "linux.idc.xpu" },
@@ -108,41 +108,41 @@ jobs:
         ]}
     secrets: inherit
 
-  # xpu-n-py3_10-inductor-benchmark-test-nightly:
-  #   permissions:
-  #     id-token: write
-  #     contents: read
-  #   if: github.event_name != 'workflow_dispatch'
-  #   name: xpu-n-py3.10-inductor-benchmark
-  #   uses: ./.github/workflows/_xpu-test.yml
-  #   needs: xpu-n-py3_10-inductor-benchmark-build
-  #   with:
-  #     build-environment: linux-noble-xpu-n-py3.10
-  #     dashboard-tag: training-true-inference-true-default-true-dynamic-true-cudagraphs-false-cppwrapper-true-aotinductor-true-freezing_cudagraphs-false-cudagraphs_low_precision-false
-  #     docker-image: ${{ needs.xpu-n-py3_10-inductor-benchmark-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.xpu-n-py3_10-inductor-benchmark-build.outputs.test-matrix }}
-  #     timeout-minutes: 720
-  #     # Disable monitor in perf tests for more investigation
-  #     disable-monitor: true
-  #     monitor-log-interval: 10
-  #     monitor-data-collect-interval: 2
-  #   secrets: inherit
+  xpu-n-py3_10-inductor-benchmark-test-nightly:
+    permissions:
+      id-token: write
+      contents: read
+    if: github.event_name != 'workflow_dispatch'
+    name: xpu-n-py3.10-inductor-benchmark
+    uses: ./.github/workflows/_xpu-test.yml
+    needs: xpu-n-py3_10-inductor-benchmark-build
+    with:
+      build-environment: linux-jammy-xpu-n-py3.10
+      dashboard-tag: training-true-inference-true-default-true-dynamic-true-cudagraphs-false-cppwrapper-true-aotinductor-true-freezing_cudagraphs-false-cudagraphs_low_precision-false
+      docker-image: ${{ needs.xpu-n-py3_10-inductor-benchmark-build.outputs.docker-image }}
+      test-matrix: ${{ needs.xpu-n-py3_10-inductor-benchmark-build.outputs.test-matrix }}
+      timeout-minutes: 720
+      # Disable monitor in perf tests for more investigation
+      disable-monitor: true
+      monitor-log-interval: 10
+      monitor-data-collect-interval: 2
+    secrets: inherit
 
-  # xpu-n-py3_10-inductor-benchmark-test:
-  #   permissions:
-  #     id-token: write
-  #     contents: read
-  #   if: github.event_name == 'workflow_dispatch'
-  #   name: xpu-n-py3.10-inductor-test
-  #   uses: ./.github/workflows/_xpu-test.yml
-  #   needs: xpu-n-py3_10-inductor-benchmark-build
-  #   with:
-  #     build-environment: linux-noble-xpu-n-py3.10
-  #     dashboard-tag: training-${{ inputs.training }}-inference-${{ inputs.inference }}-default-${{ inputs.default }}-dynamic-${{ inputs.dynamic }}-cudagraphs-${{ inputs.cudagraphs }}-cppwrapper-${{ inputs.cppwrapper }}-aotinductor-${{ inputs.aotinductor }}-maxautotune-${{ inputs.maxautotune }}-freezing_cudagraphs-${{ inputs.freezing_cudagraphs }}-cudagraphs_low_precision-${{ inputs.cudagraphs }}
-  #     docker-image: ${{ needs.xpu-n-py3_10-inductor-benchmark-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.xpu-n-py3_10-inductor-benchmark-build.outputs.test-matrix }}
-  #     timeout-minutes: 720
-  #     disable-monitor: false
-  #     monitor-log-interval: 15
-  #     monitor-data-collect-interval: 4
-  #   secrets: inherit
+  xpu-n-py3_10-inductor-benchmark-test:
+    permissions:
+      id-token: write
+      contents: read
+    if: github.event_name == 'workflow_dispatch'
+    name: xpu-n-py3.10-inductor-test
+    uses: ./.github/workflows/_xpu-test.yml
+    needs: xpu-n-py3_10-inductor-benchmark-build
+    with:
+      build-environment: linux-jammy-xpu-n-py3.10
+      dashboard-tag: training-${{ inputs.training }}-inference-${{ inputs.inference }}-default-${{ inputs.default }}-dynamic-${{ inputs.dynamic }}-cudagraphs-${{ inputs.cudagraphs }}-cppwrapper-${{ inputs.cppwrapper }}-aotinductor-${{ inputs.aotinductor }}-maxautotune-${{ inputs.maxautotune }}-freezing_cudagraphs-${{ inputs.freezing_cudagraphs }}-cudagraphs_low_precision-${{ inputs.cudagraphs }}
+      docker-image: ${{ needs.xpu-n-py3_10-inductor-benchmark-build.outputs.docker-image }}
+      test-matrix: ${{ needs.xpu-n-py3_10-inductor-benchmark-build.outputs.test-matrix }}
+      timeout-minutes: 720
+      disable-monitor: false
+      monitor-log-interval: 15
+      monitor-data-collect-interval: 4
+    secrets: inherit

.github/workflows/inductor-rocm-mi300.yml

@@ -7,7 +7,6 @@ on:
       - release/*
     tags:
       - ciflow/inductor-rocm/*
-      - ciflow/inductor-rocm-mi300/*
   workflow_dispatch:
 
 concurrency:

.github/workflows/inductor-rocm.yml

@@ -1,4 +1,4 @@
-name: inductor-rocm-mi200
+name: inductor-rocm
 
 on:
   schedule:
@@ -7,7 +7,7 @@ on:
     branches:
       - release/*
     tags:
-      - ciflow/inductor-rocm-mi200/*
+      - ciflow/inductor-rocm/*
   workflow_dispatch:
 
 concurrency:

.github/workflows/inductor-unittest.yml

@@ -81,32 +81,6 @@ jobs:
       test-matrix: ${{ needs.inductor-halide-build.outputs.test-matrix }}
     secrets: inherit
 
-  inductor-pallas-build:
-    name: inductor-pallas-build
-    uses: ./.github/workflows/_linux-build.yml
-    needs: get-label-type
-    with:
-      build-environment: linux-jammy-cuda12.8-py3.12-gcc11
-      docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-py3.12-pallas
-      cuda-arch-list: '8.9'
-      runner: linux.8xlarge.memory
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      test-matrix: |
-        { include: [
-          { config: "inductor-pallas", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.g5.12xlarge.nvidia.gpu" },
-        ]}
-    secrets: inherit
-
-  inductor-pallas-test:
-    name: inductor-pallas-test
-    uses: ./.github/workflows/_linux-test.yml
-    needs: inductor-pallas-build
-    with:
-      build-environment: linux-jammy-py3.12-gcc11
-      docker-image: ${{ needs.inductor-pallas-build.outputs.docker-image }}
-      test-matrix: ${{ needs.inductor-pallas-build.outputs.test-matrix }}
-    secrets: inherit
-
   inductor-triton-cpu-build:
     name: inductor-triton-cpu-build
     uses: ./.github/workflows/_linux-build.yml

.github/workflows/nightly.yml

@@ -5,11 +5,9 @@ on:
     - cron: 0 0 * * *
   push:
     tags:
-      # NOTE: Doc build pipelines should only get triggered on:
-      # Major or minor release candidates builds
-      - v[0-9]+.[0-9]+.0+-rc[0-9]+
-      # Final RC for major, minor and patch releases
-      - v[0-9]+.[0-9]+.[0-9]+
+      # NOTE: Doc build pipelines should only get triggered on release candidate builds
+      # Release candidate tags look like: v1.11.0-rc1
+      - v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
       - ciflow/nightly/*
   workflow_dispatch:
 

.github/workflows/operator_microbenchmark.yml

@@ -24,7 +24,7 @@ jobs:
     name: opmicrobenchmark-build
     uses: ./.github/workflows/_linux-build.yml
     with:
-      runner: linux.c7i.2xlarge
+      runner: linux.12xlarge.memory
       build-environment: linux-jammy-cuda12.8-py3.10-gcc9-sm80
       docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11
       cuda-arch-list: '8.0 9.0'
@@ -35,16 +35,16 @@ jobs:
         ]}
     secrets: inherit
 
-  # opmicrobenchmark-test:
-  #   name: opmicrobenchmark-test
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs: opmicrobenchmark-build
-  #   with:
-  #     timeout-minutes: 500
-  #     build-environment: linux-jammy-cuda12.8-py3.10-gcc9-sm80
-  #     docker-image: ${{ needs.opmicrobenchmark-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.opmicrobenchmark-build.outputs.test-matrix }}
-  #   secrets: inherit
+  opmicrobenchmark-test:
+    name: opmicrobenchmark-test
+    uses: ./.github/workflows/_linux-test.yml
+    needs: opmicrobenchmark-build
+    with:
+      timeout-minutes: 500
+      build-environment: linux-jammy-cuda12.8-py3.10-gcc9-sm80
+      docker-image: ${{ needs.opmicrobenchmark-build.outputs.docker-image }}
+      test-matrix: ${{ needs.opmicrobenchmark-build.outputs.test-matrix }}
+    secrets: inherit
 
   # B200 runner
   opmicrobenchmark-build-b200:
@@ -62,17 +62,17 @@ jobs:
         ]}
     secrets: inherit
 
-  # opmicrobenchmark-test-b200:
-  #   name: opmicrobenchmark-test-b200
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs: opmicrobenchmark-build-b200
-  #   with:
-  #     timeout-minutes: 500
-  #     build-environment: linux-jammy-cuda12.8-py3.10-gcc9-sm100
-  #     docker-image: ${{ needs.opmicrobenchmark-build-b200.outputs.docker-image }}
-  #     test-matrix: ${{ needs.opmicrobenchmark-build-b200.outputs.test-matrix }}
-  #     aws-role-to-assume: arn:aws:iam::308535385114:role/gha_workflow_s3_and_ecr_read_only
-  #   secrets: inherit
+  opmicrobenchmark-test-b200:
+    name: opmicrobenchmark-test-b200
+    uses: ./.github/workflows/_linux-test.yml
+    needs: opmicrobenchmark-build-b200
+    with:
+      timeout-minutes: 500
+      build-environment: linux-jammy-cuda12.8-py3.10-gcc9-sm100
+      docker-image: ${{ needs.opmicrobenchmark-build-b200.outputs.docker-image }}
+      test-matrix: ${{ needs.opmicrobenchmark-build-b200.outputs.test-matrix }}
+      aws-role-to-assume: arn:aws:iam::308535385114:role/gha_workflow_s3_and_ecr_read_only
+    secrets: inherit
 
   # ROCM MI300 runner
   opmicrobenchmark-build-rocm:
@@ -88,13 +88,13 @@ jobs:
         ]}
     secrets: inherit
 
-  # opmicrobenchmark-test-rocm:
-  #   name: opmicrobenchmark-test-rocm
-  #   uses: ./.github/workflows/_rocm-test.yml
-  #   needs: opmicrobenchmark-build-rocm
-  #   with:
-  #     timeout-minutes: 500
-  #     build-environment: linux-jammy-rocm-py3_10
-  #     docker-image: ${{ needs.opmicrobenchmark-build-rocm.outputs.docker-image }}
-  #     test-matrix: ${{ needs.opmicrobenchmark-build-rocm.outputs.test-matrix }}
-  #   secrets: inherit
+  opmicrobenchmark-test-rocm:
+    name: opmicrobenchmark-test-rocm
+    uses: ./.github/workflows/_rocm-test.yml
+    needs: opmicrobenchmark-build-rocm
+    with:
+      timeout-minutes: 500
+      build-environment: linux-jammy-rocm-py3_10
+      docker-image: ${{ needs.opmicrobenchmark-build-rocm.outputs.docker-image }}
+      test-matrix: ${{ needs.opmicrobenchmark-build-rocm.outputs.test-matrix }}
+    secrets: inherit

.github/workflows/periodic-rocm-mi200.yml

@@ -11,6 +11,7 @@ on:
     - cron: 29 8 * * *  # about 1:29am PDT, for mem leak check and rerun disabled tests
   push:
     tags:
+      - ciflow/periodic/*
       - ciflow/periodic-rocm-mi200/*
     branches:
       - release/*

.github/workflows/periodic-rocm-mi300.yml

@@ -11,7 +11,6 @@ on:
     - cron: 29 8 * * *  # about 1:29am PDT, for mem leak check and rerun disabled tests
   push:
     tags:
-      - ciflow/periodic/*
       - ciflow/periodic-rocm-mi300/*
     branches:
       - release/*

.github/workflows/pull.yml

@@ -70,21 +70,20 @@ jobs:
           { config: "distributed", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
           { config: "distributed", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
           { config: "numpy_2_x", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.c7i.2xlarge" },
-          { config: "libtorch_agnostic_targetting", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
         ]}
     secrets: inherit
 
-  # linux-jammy-py3_10-gcc11-test:
-  #   name: linux-jammy-py3.10-gcc11
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs:
-  #     - linux-jammy-py3_10-gcc11-build
-  #     - target-determination
-  #   with:
-  #     build-environment: linux-jammy-py3.10-gcc11
-  #     docker-image: ${{ needs.linux-jammy-py3_10-gcc11-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.linux-jammy-py3_10-gcc11-build.outputs.test-matrix }}
-  #   secrets: inherit
+  linux-jammy-py3_10-gcc11-test:
+    name: linux-jammy-py3.10-gcc11
+    uses: ./.github/workflows/_linux-test.yml
+    needs:
+      - linux-jammy-py3_10-gcc11-build
+      - target-determination
+    with:
+      build-environment: linux-jammy-py3.10-gcc11
+      docker-image: ${{ needs.linux-jammy-py3_10-gcc11-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-py3_10-gcc11-build.outputs.test-matrix }}
+    secrets: inherit
 
   linux-docs:
     name: linux-docs
@@ -128,7 +127,7 @@ jobs:
     uses: ./.github/workflows/_linux-build.yml
     needs: get-label-type
     with:
-      runner: linux.c7i.2xlarge
+      runner: linux.2xlarge.memory
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       build-environment: linux-jammy-py3.10-clang18-asan
       docker-image-name: ci-image:pytorch-linux-jammy-py3-clang18-asan
@@ -145,18 +144,18 @@ jobs:
       sync-tag: asan-build
     secrets: inherit
 
-  # linux-jammy-py3_10-clang18-asan-test:
-  #   name: linux-jammy-py3.10-clang18-asan
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs:
-  #     - linux-jammy-py3_10-clang18-asan-build
-  #     - target-determination
-  #   with:
-  #     build-environment: linux-jammy-py3.10-clang18-asan
-  #     docker-image: ${{ needs.linux-jammy-py3_10-clang18-asan-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.linux-jammy-py3_10-clang18-asan-build.outputs.test-matrix }}
-  #     sync-tag: asan-test
-  #   secrets: inherit
+  linux-jammy-py3_10-clang18-asan-test:
+    name: linux-jammy-py3.10-clang18-asan
+    uses: ./.github/workflows/_linux-test.yml
+    needs:
+      - linux-jammy-py3_10-clang18-asan-build
+      - target-determination
+    with:
+      build-environment: linux-jammy-py3.10-clang18-asan
+      docker-image: ${{ needs.linux-jammy-py3_10-clang18-asan-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-py3_10-clang18-asan-build.outputs.test-matrix }}
+      sync-tag: asan-test
+    secrets: inherit
 
   linux-jammy-py3_10-clang12-onnx-build:
     name: linux-jammy-py3.10-clang12-onnx
@@ -173,17 +172,17 @@ jobs:
         ]}
     secrets: inherit
 
-  # linux-jammy-py3_10-clang12-onnx-test:
-  #   name: linux-jammy-py3.10-clang12-onnx
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs:
-  #     - linux-jammy-py3_10-clang12-onnx-build
-  #     - target-determination
-  #   with:
-  #     build-environment: linux-jammy-py3.10-clang12-onnx
-  #     docker-image: ${{ needs.linux-jammy-py3_10-clang12-onnx-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.linux-jammy-py3_10-clang12-onnx-build.outputs.test-matrix }}
-  #   secrets: inherit
+  linux-jammy-py3_10-clang12-onnx-test:
+    name: linux-jammy-py3.10-clang12-onnx
+    uses: ./.github/workflows/_linux-test.yml
+    needs:
+      - linux-jammy-py3_10-clang12-onnx-build
+      - target-determination
+    with:
+      build-environment: linux-jammy-py3.10-clang12-onnx
+      docker-image: ${{ needs.linux-jammy-py3_10-clang12-onnx-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-py3_10-clang12-onnx-build.outputs.test-matrix }}
+    secrets: inherit
 
   linux-jammy-py3_10-clang12-build:
     name: linux-jammy-py3.10-clang12
@@ -209,17 +208,17 @@ jobs:
         ]}
     secrets: inherit
 
-  # linux-jammy-py3_10-clang12-test:
-  #   name: linux-jammy-py3.10-clang12
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs:
-  #     - linux-jammy-py3_10-clang12-build
-  #     - target-determination
-  #   with:
-  #     build-environment: linux-jammy-py3.10-clang12
-  #     docker-image: ${{ needs.linux-jammy-py3_10-clang12-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.linux-jammy-py3_10-clang12-build.outputs.test-matrix }}
-  #   secrets: inherit
+  linux-jammy-py3_10-clang12-test:
+    name: linux-jammy-py3.10-clang12
+    uses: ./.github/workflows/_linux-test.yml
+    needs:
+      - linux-jammy-py3_10-clang12-build
+      - target-determination
+    with:
+      build-environment: linux-jammy-py3.10-clang12
+      docker-image: ${{ needs.linux-jammy-py3_10-clang12-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-py3_10-clang12-build.outputs.test-matrix }}
+    secrets: inherit
 
   linux-jammy-py3_13-clang12-build:
     name: linux-jammy-py3.13-clang12
@@ -245,15 +244,15 @@ jobs:
         ]}
     secrets: inherit
 
-  # linux-jammy-py3_13-clang12-test:
-  #   name: linux-jammy-py3.13-clang12
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs: linux-jammy-py3_13-clang12-build
-  #   with:
-  #     build-environment: linux-jammy-py3.13-clang12
-  #     docker-image: ${{ needs.linux-jammy-py3_13-clang12-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.linux-jammy-py3_13-clang12-build.outputs.test-matrix }}
-  #   secrets: inherit
+  linux-jammy-py3_13-clang12-test:
+    name: linux-jammy-py3.13-clang12
+    uses: ./.github/workflows/_linux-test.yml
+    needs: linux-jammy-py3_13-clang12-build
+    with:
+      build-environment: linux-jammy-py3.13-clang12
+      docker-image: ${{ needs.linux-jammy-py3_13-clang12-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-py3_13-clang12-build.outputs.test-matrix }}
+    secrets: inherit
 
   linux-jammy-cuda12_8-cudnn9-py3_10-clang12-build:
     name: linux-jammy-cuda12.8-cudnn9-py3.10-clang12
@@ -269,20 +268,20 @@ jobs:
         ]}
     secrets: inherit
 
-  # linux-jammy-cpu-py3_10-gcc11-bazel-test:
-  #   name: linux-jammy-cpu-py3.10-gcc11-bazel-test
-  #   uses: ./.github/workflows/_bazel-build-test.yml
-  #   needs: get-label-type
-  #   with:
-  #     runner: "${{ needs.get-label-type.outputs.label-type }}linux.large"
-  #     build-environment: linux-jammy-cuda12.8-py3.10-gcc11-bazel-test
-  #     docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11
-  #     cuda-version: cpu
-  #     test-matrix: |
-  #       { include: [
-  #         { config: "default", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
-  #       ]}
-  #   secrets: inherit
+  linux-jammy-cpu-py3_10-gcc11-bazel-test:
+    name: linux-jammy-cpu-py3.10-gcc11-bazel-test
+    uses: ./.github/workflows/_bazel-build-test.yml
+    needs: get-label-type
+    with:
+      runner: "${{ needs.get-label-type.outputs.label-type }}linux.large"
+      build-environment: linux-jammy-cuda12.8-py3.10-gcc11-bazel-test
+      docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11
+      cuda-version: cpu
+      test-matrix: |
+        { include: [
+          { config: "default", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" },
+        ]}
+    secrets: inherit
 
   linux-jammy-py3_10-gcc11-mobile-lightweight-dispatch-build:
     name: linux-jammy-py3.10-gcc11-mobile-lightweight-dispatch-build
@@ -333,26 +332,26 @@ jobs:
         ]}
     secrets: inherit
 
-  # linux-jammy-cuda12_8-py3_10-gcc9-inductor-test:
-  #   name: cuda12.8-py3.10-gcc9-sm75
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs: linux-jammy-cuda12_8-py3_10-gcc9-inductor-build
-  #   with:
-  #     build-environment: linux-jammy-cuda12.8-py3.10-gcc9-sm75
-  #     docker-image: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc9-inductor-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc9-inductor-build.outputs.test-matrix }}
-  #   secrets: inherit
+  linux-jammy-cuda12_8-py3_10-gcc9-inductor-test:
+    name: cuda12.8-py3.10-gcc9-sm75
+    uses: ./.github/workflows/_linux-test.yml
+    needs: linux-jammy-cuda12_8-py3_10-gcc9-inductor-build
+    with:
+      build-environment: linux-jammy-cuda12.8-py3.10-gcc9-sm75
+      docker-image: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc9-inductor-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc9-inductor-build.outputs.test-matrix }}
+    secrets: inherit
 
-  linux-noble-xpu-n-py3_10-build:
-    name: linux-noble-xpu-n-py3.10
+  linux-jammy-xpu-n-py3_10-build:
+    name: linux-jammy-xpu-n-py3.10
     uses: ./.github/workflows/_linux-build.yml
     needs: get-label-type
     with:
       # This should sync with the build in xpu.yml but xpu uses a larger runner
       # sync-tag: linux-xpu-n-build
       runner_prefix: ${{ needs.get-label-type.outputs.label-type }}
-      build-environment: linux-noble-xpu-n-py3.10
-      docker-image-name: ci-image:pytorch-linux-noble-xpu-n-py3
+      build-environment: linux-jammy-xpu-n-py3.10
+      docker-image-name: ci-image:pytorch-linux-jammy-xpu-n-py3
       test-matrix: |
         { include: [
           { config: "default", shard: 1, num_shards: 4, runner: "linux.idc.xpu" },

.github/workflows/riscv64.yml

@@ -20,5 +20,5 @@ jobs:
     with:
       build-environment: linux-noble-riscv64-py3.12-gcc14
       docker-image-name: pytorch-linux-noble-riscv64-py3.12-gcc14
-      runner: linux.c7i.2xlarge
+      runner: linux.2xlarge
     secrets: inherit

.github/workflows/rocm-mi300.yml

@@ -6,7 +6,6 @@ on:
       - main
       - release/*
     tags:
-      - ciflow/rocm/*
       - ciflow/rocm-mi300/*
   workflow_dispatch:
   schedule:

.github/workflows/rocm.yml

@@ -1,11 +1,11 @@
-name: rocm-mi200
+name: rocm
 
 on:
   push:
     branches:
       - release/*
     tags:
-      - ciflow/rocm-mi200/*
+      - ciflow/rocm/*
   workflow_dispatch:
   schedule:
     - cron: 29 8 * * *  # about 1:29am PDT

.github/workflows/slow-rocm-mi200.yml

@@ -1,81 +0,0 @@
-# This workflow is dedicated to host slow jobs that are run only periodically because
-# they are too slow to run in every commit.  The list of slow tests can be found in
-# https://github.com/pytorch/test-infra/blob/generated-stats/stats/slow-tests.json
-name: slow-rocm-mi200
-
-on:
-  push:
-    branches:
-      - release/*
-    tags:
-      - ciflow/slow/*
-      - ciflow/slow-rocm-mi200/*
-  schedule:
-    - cron: 0 */3 * * *
-  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' }}-${{ github.event_name == 'schedule' }}-${{ github.event.schedule }}
-  cancel-in-progress: true
-
-permissions:
-  id-token: write
-  contents: read
-
-jobs:
-  llm-td:
-    if: github.repository_owner == 'pytorch'
-    name: before-test
-    uses: ./.github/workflows/llm_td_retrieval.yml
-    permissions:
-      id-token: write
-      contents: read
-
-  target-determination:
-    name: before-test
-    uses: ./.github/workflows/target_determination.yml
-    needs: llm-td
-    permissions:
-      id-token: write
-      contents: read
-
-  get-label-type:
-    name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
-    if: ${{ (github.event_name != 'schedule' || github.repository == 'pytorch/pytorch') && github.repository_owner == 'pytorch' }}
-    with:
-      triggering_actor: ${{ github.triggering_actor }}
-      issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
-      curr_branch: ${{ github.head_ref || github.ref_name }}
-      curr_ref_type: ${{ github.ref_type }}
-
-  linux-jammy-rocm-py3_10-build:
-    name: linux-jammy-rocm-py3.10
-    uses: ./.github/workflows/_linux-build.yml
-    needs: get-label-type
-    with:
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build-environment: linux-jammy-rocm-py3.10
-      docker-image-name: ci-image:pytorch-linux-jammy-rocm-n-py3
-      sync-tag: rocm-build
-      test-matrix: |
-        { include: [
-          { config: "slow", shard: 1, num_shards: 2, runner: "linux.rocm.gpu.2", owners: ["module:rocm"] },
-          { config: "slow", shard: 2, num_shards: 2, runner: "linux.rocm.gpu.2", owners: ["module:rocm"] },
-        ]}
-    secrets: inherit
-
-  linux-jammy-rocm-py3_10-test:
-    permissions:
-      id-token: write
-      contents: read
-    name: linux-jammy-rocm-py3.10
-    uses: ./.github/workflows/_rocm-test.yml
-    needs:
-      - linux-jammy-rocm-py3_10-build
-      - target-determination
-    with:
-      build-environment: linux-jammy-rocm-py3.10
-      docker-image: ${{ needs.linux-jammy-rocm-py3_10-build.outputs.docker-image }}
-      test-matrix: ${{ needs.linux-jammy-rocm-py3_10-build.outputs.test-matrix }}
-    secrets: inherit

.github/workflows/slow.yml

@@ -49,68 +49,98 @@ jobs:
       curr_branch: ${{ github.head_ref || github.ref_name }}
       curr_ref_type: ${{ github.ref_type }}
 
-  # linux-jammy-cuda12_8-py3_10-gcc11-sm86-build:
-  #   name: linux-jammy-cuda12.8-py3.10-gcc11-sm86
-  #   uses: ./.github/workflows/_linux-build.yml
-  #   needs: get-label-type
-  #   with:
-  #     runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-  #     build-environment: linux-jammy-cuda12.8-py3.10-gcc11-sm86
-  #     docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11
-  #     cuda-arch-list: 8.6
-  #     test-matrix: |
-  #       { include: [
-  #         { config: "slow", shard: 1, num_shards: 3, runner: "linux.g5.4xlarge.nvidia.gpu" },
-  #         { config: "slow", shard: 2, num_shards: 3, runner: "linux.g5.4xlarge.nvidia.gpu" },
-  #         { config: "slow", shard: 3, num_shards: 3, runner: "linux.g5.4xlarge.nvidia.gpu" },
-  #       ]}
-  #   secrets: inherit
+  linux-jammy-cuda12_8-py3_10-gcc11-sm86-build:
+    name: linux-jammy-cuda12.8-py3.10-gcc11-sm86
+    uses: ./.github/workflows/_linux-build.yml
+    needs: get-label-type
+    with:
+      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      build-environment: linux-jammy-cuda12.8-py3.10-gcc11-sm86
+      docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11
+      cuda-arch-list: 8.6
+      test-matrix: |
+        { include: [
+          { config: "slow", shard: 1, num_shards: 3, runner: "linux.g5.4xlarge.nvidia.gpu" },
+          { config: "slow", shard: 2, num_shards: 3, runner: "linux.g5.4xlarge.nvidia.gpu" },
+          { config: "slow", shard: 3, num_shards: 3, runner: "linux.g5.4xlarge.nvidia.gpu" },
+        ]}
+    secrets: inherit
 
-  # linux-jammy-cuda12_8-py3_10-gcc11-sm86-test:
-  #   name: linux-jammy-cuda12.8-py3.10-gcc11-sm86
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs:
-  #     - linux-jammy-cuda12_8-py3_10-gcc11-sm86-build
-  #     - target-determination
-  #   with:
-  #     build-environment: linux-jammy-cuda12.8-py3.10-gcc11-sm86
-  #     docker-image: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-sm86-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-sm86-build.outputs.test-matrix }}
-  #   secrets: inherit
+  linux-jammy-cuda12_8-py3_10-gcc11-sm86-test:
+    name: linux-jammy-cuda12.8-py3.10-gcc11-sm86
+    uses: ./.github/workflows/_linux-test.yml
+    needs:
+      - linux-jammy-cuda12_8-py3_10-gcc11-sm86-build
+      - target-determination
+    with:
+      build-environment: linux-jammy-cuda12.8-py3.10-gcc11-sm86
+      docker-image: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-sm86-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-sm86-build.outputs.test-matrix }}
+    secrets: inherit
 
-  # linux-jammy-py3_10-clang12-build:
-  #   name: linux-jammy-py3.10-clang12
-  #   uses: ./.github/workflows/_linux-build.yml
-  #   needs: get-label-type
-  #   with:
-  #     runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-  #     build-environment: linux-jammy-py3.10-clang12
-  #     docker-image-name: ci-image:pytorch-linux-jammy-py3.10-clang12
-  #     test-matrix: |
-  #       { include: [
-  #         { config: "slow", shard: 1, num_shards: 2, runner: "linux.2xlarge" },
-  #         { config: "slow", shard: 2, num_shards: 2, runner: "linux.2xlarge" },
-  #       ]}
-  #   secrets: inherit
+  linux-jammy-py3_10-clang12-build:
+    name: linux-jammy-py3.10-clang12
+    uses: ./.github/workflows/_linux-build.yml
+    needs: get-label-type
+    with:
+      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      build-environment: linux-jammy-py3.10-clang12
+      docker-image-name: ci-image:pytorch-linux-jammy-py3.10-clang12
+      test-matrix: |
+        { include: [
+          { config: "slow", shard: 1, num_shards: 2, runner: "linux.2xlarge" },
+          { config: "slow", shard: 2, num_shards: 2, runner: "linux.2xlarge" },
+        ]}
+    secrets: inherit
 
-  # linux-jammy-py3_10-clang12-test:
-  #   name: linux-jammy-py3.10-clang12
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs:
-  #     - linux-jammy-py3_10-clang12-build
-  #     - target-determination
-  #   with:
-  #     build-environment: linux-jammy-py3.10-clang12
-  #     docker-image: ${{ needs.linux-jammy-py3_10-clang12-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.linux-jammy-py3_10-clang12-build.outputs.test-matrix }}
-  #   secrets: inherit
+  linux-jammy-py3_10-clang12-test:
+    name: linux-jammy-py3.10-clang12
+    uses: ./.github/workflows/_linux-test.yml
+    needs:
+      - linux-jammy-py3_10-clang12-build
+      - target-determination
+    with:
+      build-environment: linux-jammy-py3.10-clang12
+      docker-image: ${{ needs.linux-jammy-py3_10-clang12-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-py3_10-clang12-build.outputs.test-matrix }}
+    secrets: inherit
+
+  linux-jammy-rocm-py3_10-build:
+    name: linux-jammy-rocm-py3.10
+    uses: ./.github/workflows/_linux-build.yml
+    needs: get-label-type
+    with:
+      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      build-environment: linux-jammy-rocm-py3.10
+      docker-image-name: ci-image:pytorch-linux-jammy-rocm-n-py3
+      test-matrix: |
+        { include: [
+          { config: "slow", shard: 1, num_shards: 2, runner: "linux.rocm.gpu.2", owners: ["module:rocm"] },
+          { config: "slow", shard: 2, num_shards: 2, runner: "linux.rocm.gpu.2", owners: ["module:rocm"] },
+        ]}
+    secrets: inherit
+
+  linux-jammy-rocm-py3_10-test:
+    permissions:
+      id-token: write
+      contents: read
+    name: linux-jammy-rocm-py3.10
+    uses: ./.github/workflows/_rocm-test.yml
+    needs:
+      - linux-jammy-rocm-py3_10-build
+      - target-determination
+    with:
+      build-environment: linux-jammy-rocm-py3.10
+      docker-image: ${{ needs.linux-jammy-rocm-py3_10-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-rocm-py3_10-build.outputs.test-matrix }}
+    secrets: inherit
 
   linux-jammy-py3_10-clang18-asan-build:
     name: linux-jammy-py3.10-clang18-asan
     uses: ./.github/workflows/_linux-build.yml
     needs: get-label-type
     with:
-      runner: linux.c7i.2xlarge
+      runner: linux.2xlarge.memory
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       build-environment: linux-jammy-py3.10-clang18-asan
       docker-image-name: ci-image:pytorch-linux-jammy-py3-clang18-asan
@@ -123,15 +153,15 @@ jobs:
       sync-tag: asan-build
     secrets: inherit
 
-  # linux-jammy-py3_10-clang18-asan-test:
-  #   name: linux-jammy-py3.10-clang18-asan
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs:
-  #     - linux-jammy-py3_10-clang18-asan-build
-  #     - target-determination
-  #   with:
-  #     build-environment: linux-jammy-py3.10-clang18-asan
-  #     docker-image: ${{ needs.linux-jammy-py3_10-clang18-asan-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.linux-jammy-py3_10-clang18-asan-build.outputs.test-matrix }}
-  #     sync-tag: asan-test
-  #   secrets: inherit
+  linux-jammy-py3_10-clang18-asan-test:
+    name: linux-jammy-py3.10-clang18-asan
+    uses: ./.github/workflows/_linux-test.yml
+    needs:
+      - linux-jammy-py3_10-clang18-asan-build
+      - target-determination
+    with:
+      build-environment: linux-jammy-py3.10-clang18-asan
+      docker-image: ${{ needs.linux-jammy-py3_10-clang18-asan-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-py3_10-clang18-asan-build.outputs.test-matrix }}
+      sync-tag: asan-test
+    secrets: inherit

.github/workflows/test-b200.yml

@@ -5,9 +5,7 @@
 # Flow:
 # 1. Builds PyTorch with CUDA 12.8+ and sm100 architecture for B200
 # 2. Runs smoke tests on linux.dgx.b200 runner
-# 3. Tests executed are defined in .ci/pytorch/test.sh -> test_python_smoke_b200() function
-#    - Includes matmul, scaled_matmul, FP8, and FlashAttention CuTe tests
-#    - FlashAttention CuTe DSL is installed as part of test execution
+# 3. Tests executed are defined in .ci/pytorch/test.sh -> test_python_smoke() function
 #
 # Triggered by:
 # - Pull requests modifying this workflow file
@@ -54,7 +52,7 @@ jobs:
     needs: get-label-type
     with:
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runner: linux.r7i.4xlarge
+      runner: linux.12xlarge.memory
       build-environment: linux-jammy-cuda12.8-py3.10-gcc11-sm100
       docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11
       cuda-arch-list: '10.0'
@@ -65,14 +63,14 @@ jobs:
       # config: "smoke_b200" maps to test_python_smoke_b200() in .ci/pytorch/test.sh
     secrets: inherit
 
-  # linux-jammy-cuda12_8-py3_10-gcc11-sm100-test:
-  #   name: linux-jammy-cuda12.8-py3.10-gcc11-sm100
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs:
-  #     - linux-jammy-cuda12_8-py3_10-gcc11-sm100-build
-  #   with:
-  #     build-environment: linux-jammy-cuda12.8-py3.10-gcc11-sm100
-  #     docker-image: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-sm100-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-sm100-build.outputs.test-matrix }}
-  #     aws-role-to-assume: arn:aws:iam::308535385114:role/gha_workflow_s3_and_ecr_read_only
-  #   secrets: inherit
+  linux-jammy-cuda12_8-py3_10-gcc11-sm100-test:
+    name: linux-jammy-cuda12.8-py3.10-gcc11-sm100
+    uses: ./.github/workflows/_linux-test.yml
+    needs:
+      - linux-jammy-cuda12_8-py3_10-gcc11-sm100-build
+    with:
+      build-environment: linux-jammy-cuda12.8-py3.10-gcc11-sm100
+      docker-image: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-sm100-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-sm100-build.outputs.test-matrix }}
+      aws-role-to-assume: arn:aws:iam::308535385114:role/gha_workflow_s3_and_ecr_read_only
+    secrets: inherit

.github/workflows/test-h100.yml

@@ -41,6 +41,7 @@ jobs:
     needs: get-label-type
     with:
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      runner: linux.12xlarge.memory
       build-environment: linux-jammy-cuda12.8-py3.10-gcc11-sm90
       docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11
       cuda-arch-list: '9.0'

.github/workflows/trunk-rocm-mi300.yml

@@ -1,83 +0,0 @@
-name: trunk-rocm-mi300
-
-on:
-  push:
-    branches:
-      - main
-      - release/*
-  workflow_dispatch:
-  schedule:
-    - cron: 29 8 * * *  # about 1:29am PDT
-
-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:
-  id-token: write
-  contents: read
-
-jobs:
-  llm-td:
-    if: github.repository_owner == 'pytorch'
-    name: before-test
-    uses: ./.github/workflows/llm_td_retrieval.yml
-    permissions:
-      id-token: write
-      contents: read
-
-  target-determination:
-    name: before-test
-    uses: ./.github/workflows/target_determination.yml
-    needs: llm-td
-    permissions:
-      id-token: write
-      contents: read
-
-  get-label-type:
-    name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
-    if: ${{ (github.event_name != 'schedule' || github.repository == 'pytorch/pytorch') && github.repository_owner == 'pytorch' }}
-    with:
-      triggering_actor: ${{ github.triggering_actor }}
-      issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
-      curr_branch: ${{ github.head_ref || github.ref_name }}
-      curr_ref_type: ${{ github.ref_type }}
-
-  linux-jammy-rocm-py3_10-build:
-    name: linux-jammy-rocm-py3.10
-    uses: ./.github/workflows/_linux-build.yml
-    needs: get-label-type
-    with:
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build-environment: linux-jammy-rocm-py3.10
-      docker-image-name: ci-image:pytorch-linux-jammy-rocm-n-py3
-      sync-tag: rocm-build
-      test-matrix: |
-        { include: [
-          { config: "default", shard: 1, num_shards: 6, runner: "linux.rocm.gpu.gfx942.1.b" },
-          { config: "default", shard: 2, num_shards: 6, runner: "linux.rocm.gpu.gfx942.1.b" },
-          { config: "default", shard: 3, num_shards: 6, runner: "linux.rocm.gpu.gfx942.1.b" },
-          { config: "default", shard: 4, num_shards: 6, runner: "linux.rocm.gpu.gfx942.1.b" },
-          { config: "default", shard: 5, num_shards: 6, runner: "linux.rocm.gpu.gfx942.1.b" },
-          { config: "default", shard: 6, num_shards: 6, runner: "linux.rocm.gpu.gfx942.1.b" },
-          { config: "distributed", shard: 1, num_shards: 3, runner: "linux.rocm.gpu.gfx942.4.b" },
-          { config: "distributed", shard: 2, num_shards: 3, runner: "linux.rocm.gpu.gfx942.4.b" },
-          { config: "distributed", shard: 3, num_shards: 3, runner: "linux.rocm.gpu.gfx942.4.b" },
-        ]}
-    secrets: inherit
-
-  linux-jammy-rocm-py3_10-test:
-    permissions:
-      id-token: write
-      contents: read
-    name: linux-jammy-rocm-py3.10
-    uses: ./.github/workflows/_rocm-test.yml
-    needs:
-      - linux-jammy-rocm-py3_10-build
-      - target-determination
-    with:
-      build-environment: linux-jammy-rocm-py3.10
-      docker-image: ${{ needs.linux-jammy-rocm-py3_10-build.outputs.docker-image }}
-      test-matrix: ${{ needs.linux-jammy-rocm-py3_10-build.outputs.test-matrix }}
-    secrets: inherit

.github/workflows/trunk.yml

@@ -56,7 +56,7 @@ jobs:
       docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11
       build-generates-artifacts: false
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runner: "linux.c7i.2xlarge"
+      runner: "linux.c7i.4xlarge"
       test-matrix: |
         { include: [
           { config: "default", shard: 1, num_shards: 1 },
@@ -83,22 +83,21 @@ jobs:
           { config: "distributed", shard: 2, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.g4dn.12xlarge.nvidia.gpu" },
           { config: "distributed", shard: 3, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}linux.g4dn.12xlarge.nvidia.gpu" },
           { config: "pr_time_benchmarks", shard: 1, num_shards: 1, runner: "linux.g4dn.metal.nvidia.gpu" },
-          { config: "libtorch_agnostic_targetting", shard: 1, num_shards: 1, runner: "linux.g4dn.metal.nvidia.gpu" },
         ]}
     secrets: inherit
 
-  # linux-jammy-cuda12_8-py3_10-gcc11-test:
-  #   name: linux-jammy-cuda12.8-py3.10-gcc11
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs:
-  #     - linux-jammy-cuda12_8-py3_10-gcc11-build
-  #     - target-determination
-  #   with:
-  #     timeout-minutes: 360
-  #     build-environment: linux-jammy-cuda12.8-py3.10-gcc11
-  #     docker-image: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-build.outputs.test-matrix }}
-  #   secrets: inherit
+  linux-jammy-cuda12_8-py3_10-gcc11-test:
+    name: linux-jammy-cuda12.8-py3.10-gcc11
+    uses: ./.github/workflows/_linux-test.yml
+    needs:
+      - linux-jammy-cuda12_8-py3_10-gcc11-build
+      - target-determination
+    with:
+      timeout-minutes: 360
+      build-environment: linux-jammy-cuda12.8-py3.10-gcc11
+      docker-image: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-build.outputs.test-matrix }}
+    secrets: inherit
 
 
   # no-ops builds test USE_PER_OPERATOR_HEADERS=0 where ATen/ops is not generated
@@ -137,19 +136,19 @@ jobs:
         ]}
     secrets: inherit
 
-  # macos-py3-arm64-test:
-  #   name: macos-py3-arm64
-  #   uses: ./.github/workflows/_mac-test.yml
-  #   needs:
-  #     - macos-py3-arm64-build
-  #     - target-determination
-  #   with:
-  #     build-environment: macos-py3-arm64
-  #     # Same as the build job
-  #     python-version: 3.12.7
-  #     test-matrix: ${{ needs.macos-py3-arm64-build.outputs.test-matrix }}
-  #     disable-monitor: false
-  #   secrets: inherit
+  macos-py3-arm64-test:
+    name: macos-py3-arm64
+    uses: ./.github/workflows/_mac-test.yml
+    needs:
+      - macos-py3-arm64-build
+      - target-determination
+    with:
+      build-environment: macos-py3-arm64
+      # Same as the build job
+      python-version: 3.12.7
+      test-matrix: ${{ needs.macos-py3-arm64-build.outputs.test-matrix }}
+      disable-monitor: false
+    secrets: inherit
 
   win-vs2022-cpu-py3-build:
     name: win-vs2022-cpu-py3
@@ -168,18 +167,18 @@ jobs:
         ]}
     secrets: inherit
 
-  # win-vs2022-cpu-py3-test:
-  #   name: win-vs2022-cpu-py3
-  #   uses: ./.github/workflows/_win-test.yml
-  #   needs:
-  #     - win-vs2022-cpu-py3-build
-  #     - target-determination
-  #   with:
-  #     build-environment: win-vs2022-cpu-py3
-  #     cuda-version: cpu
-  #     test-matrix: ${{ needs.win-vs2022-cpu-py3-build.outputs.test-matrix }}
-  #     disable-monitor: false
-  #   secrets: inherit
+  win-vs2022-cpu-py3-test:
+    name: win-vs2022-cpu-py3
+    uses: ./.github/workflows/_win-test.yml
+    needs:
+      - win-vs2022-cpu-py3-build
+      - target-determination
+    with:
+      build-environment: win-vs2022-cpu-py3
+      cuda-version: cpu
+      test-matrix: ${{ needs.win-vs2022-cpu-py3-build.outputs.test-matrix }}
+      disable-monitor: false
+    secrets: inherit
 
   win-vs2022-cuda12_8-py3-build:
     name: win-vs2022-cuda12.8-py3
@@ -209,22 +208,22 @@ jobs:
         ]}
     secrets: inherit
 
-  # linux-jammy-rocm-py3_10-test:
-  #   if: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/trunk') }}
-  #   permissions:
-  #     id-token: write
-  #     contents: read
-  #   name: linux-jammy-rocm-py3.10
-  #   uses: ./.github/workflows/_rocm-test.yml
-  #   needs:
-  #     - linux-jammy-rocm-py3_10-build
-  #     - target-determination
-  #   with:
-  #     build-environment: linux-jammy-rocm-py3.10
-  #     docker-image: ${{ needs.linux-jammy-rocm-py3_10-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.linux-jammy-rocm-py3_10-build.outputs.test-matrix }}
-  #     tests-to-include: "test_nn test_torch test_cuda test_ops test_unary_ufuncs test_binary_ufuncs test_autograd inductor/test_torchinductor distributed/test_c10d_common distributed/test_c10d_nccl"
-  #   secrets: inherit
+  linux-jammy-rocm-py3_10-test:
+    if: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/trunk') }}
+    permissions:
+      id-token: write
+      contents: read
+    name: linux-jammy-rocm-py3.10
+    uses: ./.github/workflows/_rocm-test.yml
+    needs:
+      - linux-jammy-rocm-py3_10-build
+      - target-determination
+    with:
+      build-environment: linux-jammy-rocm-py3.10
+      docker-image: ${{ needs.linux-jammy-rocm-py3_10-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-rocm-py3_10-build.outputs.test-matrix }}
+      tests-to-include: "test_nn test_torch test_cuda test_ops test_unary_ufuncs test_binary_ufuncs test_autograd inductor/test_torchinductor distributed/test_c10d_common distributed/test_c10d_nccl"
+    secrets: inherit
 
   inductor-build:
     name: inductor-build
@@ -237,21 +236,21 @@ jobs:
     secrets: inherit
 
   # Test cross-compiled models with Windows libs extracted from wheel
-  # cross-compile-linux-test:
-  #   name: cross-compile-linux-test
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs:
-  #     - linux-jammy-cuda12_8-py3_10-gcc11-build
-  #     - get-label-type
-  #     - win-vs2022-cuda12_8-py3-build
-  #   with:
-  #     build-environment: linux-jammy-cuda12.8-py3.10-gcc11
-  #     docker-image: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-build.outputs.docker-image }}
-  #     test-matrix: |
-  #       { include: [
-  #         { config: "aoti_cross_compile_for_windows", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.g6.4xlarge.experimental.nvidia.gpu", win_torch_wheel_artifact: "win-vs2022-cuda12.8-py3" },
-  #       ]}
-  #   secrets: inherit
+  cross-compile-linux-test:
+    name: cross-compile-linux-test
+    uses: ./.github/workflows/_linux-test.yml
+    needs:
+      - linux-jammy-cuda12_8-py3_10-gcc11-build
+      - get-label-type
+      - win-vs2022-cuda12_8-py3-build
+    with:
+      build-environment: linux-jammy-cuda12.8-py3.10-gcc11
+      docker-image: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-build.outputs.docker-image }}
+      test-matrix: |
+        { include: [
+          { config: "aoti_cross_compile_for_windows", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.g6.4xlarge.experimental.nvidia.gpu", win_torch_wheel_artifact: "win-vs2022-cuda12.8-py3" },
+        ]}
+    secrets: inherit
 
   verify-cachebench-cpu-build:
     name: verify-cachebench-cpu-build
@@ -267,17 +266,17 @@ jobs:
         ]}
     secrets: inherit
 
-  # verify-cachebench-cpu-test:
-  #   name: verify-cachebench-cpu-test
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs:
-  #     - verify-cachebench-cpu-build
-  #     - target-determination
-  #   with:
-  #     build-environment: linux-jammy-py3.10-gcc11
-  #     docker-image: ${{ needs.verify-cachebench-cpu-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.verify-cachebench-cpu-build.outputs.test-matrix }}
-  #   secrets: inherit
+  verify-cachebench-cpu-test:
+    name: verify-cachebench-cpu-test
+    uses: ./.github/workflows/_linux-test.yml
+    needs:
+      - verify-cachebench-cpu-build
+      - target-determination
+    with:
+      build-environment: linux-jammy-py3.10-gcc11
+      docker-image: ${{ needs.verify-cachebench-cpu-build.outputs.docker-image }}
+      test-matrix: ${{ needs.verify-cachebench-cpu-build.outputs.test-matrix }}
+    secrets: inherit
 
   linux-jammy-py3-clang12-executorch-build:
     name: linux-jammy-py3-clang12-executorch
@@ -293,15 +292,15 @@ jobs:
         ]}
     secrets: inherit
 
-  # linux-jammy-py3-clang12-executorch-test:
-  #   name: linux-jammy-py3-clang12-executorch
-  #   uses: ./.github/workflows/_linux-test.yml
-  #   needs: linux-jammy-py3-clang12-executorch-build
-  #   with:
-  #     build-environment: linux-jammy-py3-clang12-executorch
-  #     docker-image: ${{ needs.linux-jammy-py3-clang12-executorch-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.linux-jammy-py3-clang12-executorch-build.outputs.test-matrix }}
-  #   secrets: inherit
+  linux-jammy-py3-clang12-executorch-test:
+    name: linux-jammy-py3-clang12-executorch
+    uses: ./.github/workflows/_linux-test.yml
+    needs: linux-jammy-py3-clang12-executorch-build
+    with:
+      build-environment: linux-jammy-py3-clang12-executorch
+      docker-image: ${{ needs.linux-jammy-py3-clang12-executorch-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-py3-clang12-executorch-build.outputs.test-matrix }}
+    secrets: inherit
 
   linux-jammy-py3_10-gcc11-full-debug-build-only:
     name: linux-jammy-py3.10-gcc11-full-debug-build-only
@@ -309,7 +308,7 @@ jobs:
     needs: get-label-type
     with:
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runner: linux.c7i.2xlarge
+      runner: linux.2xlarge.memory
       build-environment: linux-jammy-py3.10-gcc11-full-debug-build-only
       docker-image-name: ci-image:pytorch-linux-jammy-py3.10-gcc11
     secrets: inherit

.github/workflows/upload-test-stats.yml

@@ -5,23 +5,21 @@ on:
     workflows:
       - pull
       - trunk
-      - trunk-rocm-mi300
       - periodic
       - periodic-rocm-mi200
       - periodic-rocm-mi300
       - inductor
       - unstable
       - slow
-      - slow-rocm-mi200
       - unstable-periodic
       - inductor-periodic
-      - rocm-mi200
+      - rocm
       - rocm-mi300
       - rocm-mi355
       - inductor-micro-benchmark
       - inductor-micro-benchmark-x86
       - inductor-cu124
-      - inductor-rocm-mi200
+      - inductor-rocm
       - inductor-rocm-mi300
       - mac-mps
       - linux-aarch64

.github/workflows/vllm.yml

@@ -43,7 +43,7 @@ jobs:
       build-environment: linux-jammy-cuda12.8-py3.12-gcc11
       docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3.12-gcc11-vllm
       cuda-arch-list: '8.0 8.9 9.0'
-      runner: linux.c7i.2xlarge
+      runner: linux.24xlarge.memory
       test-matrix: |
         { include: [
           { config: "vllm_basic_correctness_test", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" },
@@ -65,14 +65,14 @@ jobs:
         ]}
     secrets: inherit
 
-  # vllm-test-sm89:
-  #     name: ci-vllm-test
-  #     uses: ./.github/workflows/_linux-test.yml
-  #     needs: [
-  #       torch-build,
-  #     ]
-  #     with:
-  #       build-environment: linux-jammy-cuda12.8-py3.12-gcc11
-  #       docker-image: ${{ needs.torch-build.outputs.docker-image }}
-  #       test-matrix: ${{ needs.torch-build.outputs.test-matrix }}
-  #     secrets: inherit
+  vllm-test-sm89:
+      name: ci-vllm-test
+      uses: ./.github/workflows/_linux-test.yml
+      needs: [
+        torch-build,
+      ]
+      with:
+        build-environment: linux-jammy-cuda12.8-py3.12-gcc11
+        docker-image: ${{ needs.torch-build.outputs.docker-image }}
+        test-matrix: ${{ needs.torch-build.outputs.test-matrix }}
+      secrets: inherit

.github/workflows/xpu.yml

@@ -35,7 +35,7 @@ jobs:
       runner_prefix: ${{ needs.get-label-type.outputs.label-type }}
       build-environment: linux-jammy-xpu-n-1-py3.10
       docker-image-name: ci-image:pytorch-linux-jammy-xpu-n-1-py3
-      runner: linux.c7i.2xlarge
+      runner: linux.c7i.12xlarge
       test-matrix: |
         { include: [
           { config: "default", shard: 1, num_shards: 6, runner: "linux.idc.xpu" },
@@ -47,16 +47,16 @@ jobs:
         ]}
     secrets: inherit
 
-  linux-noble-xpu-n-py3_10-build:
-    name: linux-noble-xpu-n-py3.10
+  linux-jammy-xpu-n-py3_10-build:
+    name: linux-jammy-xpu-n-py3.10
     uses: ./.github/workflows/_linux-build.yml
     needs: get-label-type
     with:
       sync-tag: linux-xpu-n-build
       runner_prefix: ${{ needs.get-label-type.outputs.label-type }}
-      build-environment: linux-noble-xpu-n-py3.10
-      docker-image-name: ci-image:pytorch-linux-noble-xpu-n-py3
-      runner: linux.c7i.2xlarge
+      build-environment: linux-jammy-xpu-n-py3.10
+      docker-image-name: ci-image:pytorch-linux-jammy-xpu-n-py3
+      runner: linux.c7i.12xlarge
       test-matrix: |
         { include: [
           { config: "default", shard: 1, num_shards: 12, runner: "linux.idc.xpu" },
@@ -74,39 +74,39 @@ jobs:
         ]}
     secrets: inherit
 
-  # linux-noble-xpu-n-py3_10-test:
-  #   name: linux-noble-xpu-n-py3.10
-  #   uses: ./.github/workflows/_xpu-test.yml
-  #   needs: linux-noble-xpu-n-py3_10-build
-  #   permissions:
-  #     id-token: write
-  #     contents: read
-  #   with:
-  #     build-environment: linux-noble-xpu-n-py3.10
-  #     docker-image: ${{ needs.linux-noble-xpu-n-py3_10-build.outputs.docker-image }}
-  #     test-matrix: ${{ needs.linux-noble-xpu-n-py3_10-build.outputs.test-matrix }}
-  #   secrets: inherit
+  linux-jammy-xpu-n-py3_10-test:
+    name: linux-jammy-xpu-n-py3.10
+    uses: ./.github/workflows/_xpu-test.yml
+    needs: linux-jammy-xpu-n-py3_10-build
+    permissions:
+      id-token: write
+      contents: read
+    with:
+      build-environment: linux-jammy-xpu-n-py3.10
+      docker-image: ${{ needs.linux-jammy-xpu-n-py3_10-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-xpu-n-py3_10-build.outputs.test-matrix }}
+    secrets: inherit
 
-  # windows-xpu-n-1-build:
-  #   if: github.repository_owner == 'pytorch'
-  #   name: win-vs2022-xpu-n-1-py3
-  #   uses: ./.github/workflows/_win-build.yml
-  #   with:
-  #     build-environment: win-vs2022-xpu-n-1-py3
-  #     cuda-version: cpu
-  #     use-xpu: true
-  #     xpu-version: '2025.1'
-  #     vc-year: '2022'
-  #   secrets: inherit
+  windows-xpu-n-1-build:
+    if: github.repository_owner == 'pytorch'
+    name: win-vs2022-xpu-n-1-py3
+    uses: ./.github/workflows/_win-build.yml
+    with:
+      build-environment: win-vs2022-xpu-n-1-py3
+      cuda-version: cpu
+      use-xpu: true
+      xpu-version: '2025.1'
+      vc-year: '2022'
+    secrets: inherit
 
-  # windows-xpu-n-build:
-  #   if: github.repository_owner == 'pytorch'
-  #   name: win-vs2022-xpu-n-py3
-  #   uses: ./.github/workflows/_win-build.yml
-  #   with:
-  #     build-environment: win-vs2022-xpu-n-py3
-  #     cuda-version: cpu
-  #     use-xpu: true
-  #     xpu-version: '2025.2'
-  #     vc-year: '2022'
-  #   secrets: inherit
+  windows-xpu-n-build:
+    if: github.repository_owner == 'pytorch'
+    name: win-vs2022-xpu-n-py3
+    uses: ./.github/workflows/_win-build.yml
+    with:
+      build-environment: win-vs2022-xpu-n-py3
+      cuda-version: cpu
+      use-xpu: true
+      xpu-version: '2025.2'
+      vc-year: '2022'
+    secrets: inherit

.lintrunner.toml

@@ -143,8 +143,7 @@ init_command = [
     'tools/linter/adapters/pip_init.py',
     '--dry-run={{DRYRUN}}',
     'numpy==1.26.4 ; python_version >= "3.10" and python_version <= "3.11"',
-    'numpy==2.1.0 ; python_version >= "3.12" and python_version <= "3.13"',
-    'numpy==2.3.4 ; python_version >= "3.14"',
+    'numpy==2.1.0 ; python_version >= "3.12"',
     'expecttest==0.3.0',
     'pyrefly==0.36.2',
     'sympy==1.13.3',
@@ -186,8 +185,6 @@ include_patterns = [
     'aten/src/ATen/native/nested/cuda/*.h',
     'aten/src/ATen/native/nested/*.cpp',
     'aten/src/ATen/native/nested/*.h',
-    'aten/src/ATen/xpu/**/*.h',
-    'aten/src/ATen/xpu/**/*.cpp',
     'c10/**/*.cpp',
     'c10/**/*.h',
     'torch/*.h',
@@ -1404,7 +1401,7 @@ init_command = [
     '--dry-run={{DRYRUN}}',
     'usort==1.0.8.post1',
     'isort==6.0.1',
-    'ruff==0.14.4',  # sync with RUFF
+    'ruff==0.13.1',  # sync with RUFF
 ]
 is_formatter = true
 
@@ -1539,7 +1536,7 @@ init_command = [
     'python3',
     'tools/linter/adapters/pip_init.py',
     '--dry-run={{DRYRUN}}',
-    'ruff==0.14.4',  # sync with PYFMT
+    'ruff==0.13.1',  # sync with PYFMT
 ]
 is_formatter = true
 

.spin/cmds.py

@@ -1,330 +0,0 @@
-import hashlib
-import subprocess
-import sys
-from pathlib import Path
-
-import click
-import spin
-
-
-def file_digest(file, algorithm: str):
-    try:
-        return hashlib.file_digest(file, algorithm)
-    except AttributeError:
-        pass  # Fallback to manual implementation below
-    hash = hashlib.new(algorithm)
-    while chunk := file.read(8192):
-        hash.update(chunk)
-    return hash
-
-
-def _hash_file(file):
-    with open(file, "rb") as f:
-        hash = file_digest(f, "sha256")
-    return hash.hexdigest()
-
-
-def _hash_files(files):
-    hashes = {file: _hash_file(file) for file in files}
-    return hashes
-
-
-def _read_hashes(hash_file: Path):
-    if not hash_file.exists():
-        return {}
-    with hash_file.open("r") as f:
-        lines = f.readlines()
-    hashes = {}
-    for line in lines:
-        hash = line[:64]
-        file = line[66:].strip()
-        hashes[file] = hash
-    return hashes
-
-
-def _updated_hashes(hash_file, files_to_hash):
-    old_hashes = _read_hashes(hash_file)
-    new_hashes = _hash_files(files_to_hash)
-    if new_hashes != old_hashes:
-        return new_hashes
-    return None
-
-
-@click.command()
-def regenerate_version():
-    """Regenerate version.py."""
-    cmd = [
-        sys.executable,
-        "-m",
-        "tools.generate_torch_version",
-        "--is-debug=false",
-    ]
-    spin.util.run(cmd)
-
-
-TYPE_STUBS = [
-    (
-        "Pytorch type stubs",
-        Path(".lintbin/.pytorch-type-stubs.sha256"),
-        [
-            "aten/src/ATen/native/native_functions.yaml",
-            "aten/src/ATen/native/tags.yaml",
-            "tools/autograd/deprecated.yaml",
-        ],
-        [
-            sys.executable,
-            "-m",
-            "tools.pyi.gen_pyi",
-            "--native-functions-path",
-            "aten/src/ATen/native/native_functions.yaml",
-            "--tags-path",
-            "aten/src/ATen/native/tags.yaml",
-            "--deprecated-functions-path",
-            "tools/autograd/deprecated.yaml",
-        ],
-    ),
-    (
-        "Datapipes type stubs",
-        None,
-        [],
-        [
-            sys.executable,
-            "torch/utils/data/datapipes/gen_pyi.py",
-        ],
-    ),
-]
-
-
-@click.command()
-def regenerate_type_stubs():
-    """Regenerate type stubs."""
-    for name, hash_file, files_to_hash, cmd in TYPE_STUBS:
-        if hash_file:
-            if hashes := _updated_hashes(hash_file, files_to_hash):
-                click.echo(
-                    f"Changes detected in type stub files for {name}. Regenerating..."
-                )
-                spin.util.run(cmd)
-                hash_file.parent.mkdir(parents=True, exist_ok=True)
-                with hash_file.open("w") as f:
-                    for file, hash in hashes.items():
-                        f.write(f"{hash}  {file}\n")
-                click.echo("Type stubs and hashes updated.")
-            else:
-                click.echo(f"No changes detected in type stub files for {name}.")
-        else:
-            click.echo(f"No hash file for {name}. Regenerating...")
-            spin.util.run(cmd)
-            click.echo("Type stubs regenerated.")
-
-
-@click.command()
-def regenerate_clangtidy_files():
-    """Regenerate clang-tidy files."""
-    cmd = [
-        sys.executable,
-        "-m",
-        "tools.linter.clang_tidy.generate_build_files",
-    ]
-    spin.util.run(cmd)
-
-
-#: These linters are expected to need less than 3s cpu time total
-VERY_FAST_LINTERS = {
-    "ATEN_CPU_GPU_AGNOSTIC",
-    "BAZEL_LINTER",
-    "C10_NODISCARD",
-    "C10_UNUSED",
-    "CALL_ONCE",
-    "CMAKE_MINIMUM_REQUIRED",
-    "CONTEXT_DECORATOR",
-    "COPYRIGHT",
-    "CUBINCLUDE",
-    "DEPLOY_DETECTION",
-    "ERROR_PRONE_ISINSTANCE",
-    "EXEC",
-    "HEADER_ONLY_LINTER",
-    "IMPORT_LINTER",
-    "INCLUDE",
-    "LINTRUNNER_VERSION",
-    "MERGE_CONFLICTLESS_CSV",
-    "META_NO_CREATE_UNBACKED",
-    "NEWLINE",
-    "NOQA",
-    "NO_WORKFLOWS_ON_FORK",
-    "ONCE_FLAG",
-    "PYBIND11_INCLUDE",
-    "PYBIND11_SPECIALIZATION",
-    "PYPIDEP",
-    "PYPROJECT",
-    "RAWCUDA",
-    "RAWCUDADEVICE",
-    "ROOT_LOGGING",
-    "TABS",
-    "TESTOWNERS",
-    "TYPEIGNORE",
-    "TYPENOSKIP",
-    "WORKFLOWSYNC",
-}
-
-
-#: These linters are expected to take a few seconds, but less than 10s cpu time total
-FAST_LINTERS = {
-    "CMAKE",
-    "DOCSTRING_LINTER",
-    "GHA",
-    "NATIVEFUNCTIONS",
-    "RUFF",
-    "SET_LINTER",
-    "SHELLCHECK",
-    "SPACES",
-}
-
-
-#: These linters are expected to take more than 10s cpu time total;
-#: some need more than 1 hour.
-SLOW_LINTERS = {
-    "ACTIONLINT",
-    "CLANGFORMAT",
-    "CLANGTIDY",
-    "CODESPELL",
-    "FLAKE8",
-    "GB_REGISTRY",
-    "PYFMT",
-    "PYREFLY",
-    "TEST_DEVICE_BIAS",
-    "TEST_HAS_MAIN",
-}
-
-
-ALL_LINTERS = VERY_FAST_LINTERS | FAST_LINTERS | SLOW_LINTERS
-
-
-LINTRUNNER_CACHE_INFO = (
-    Path(".lintbin/.lintrunner.sha256"),
-    [
-        "requirements.txt",
-        "pyproject.toml",
-        ".lintrunner.toml",
-    ],
-)
-
-
-LINTRUNNER_BASE_CMD = [
-    "uvx",
-    "--python",
-    "3.10",
-    "lintrunner@0.12.7",
-]
-
-
-@click.command()
-def setup_lint():
-    """Set up lintrunner with current CI version."""
-    cmd = LINTRUNNER_BASE_CMD + ["init"]
-    subprocess.run(cmd, check=True, capture_output=True, text=True)
-
-
-def _check_linters():
-    cmd = LINTRUNNER_BASE_CMD + ["list"]
-    ret = spin.util.run(cmd, output=False, stderr=subprocess.PIPE)
-    linters = {l.strip() for l in ret.stdout.decode().strip().split("\n")[1:]}
-    unknown_linters = linters - ALL_LINTERS
-    missing_linters = ALL_LINTERS - linters
-    if unknown_linters:
-        click.secho(
-            f"Unknown linters found; please add them to the correct category "
-            f"in .spin/cmds.py: {', '.join(unknown_linters)}",
-            fg="yellow",
-        )
-    if missing_linters:
-        click.secho(
-            f"Missing linters found; please update the corresponding category "
-            f"in .spin/cmds.py: {', '.join(missing_linters)}",
-            fg="yellow",
-        )
-    return unknown_linters, missing_linters
-
-
-@spin.util.extend_command(
-    setup_lint,
-    doc=f"""
-        If configuration has changed, update lintrunner.
-
-        Compares the stored old hashes of configuration files with new ones and
-        performs setup via setup-lint if the hashes have changed.
-        Hashes are stored in {LINTRUNNER_CACHE_INFO[0]}; the following files are
-        considered: {", ".join(LINTRUNNER_CACHE_INFO[1])}.
-        """,
-)
-@click.pass_context
-def lazy_setup_lint(ctx, parent_callback, **kwargs):
-    if hashes := _updated_hashes(*LINTRUNNER_CACHE_INFO):
-        click.echo(
-            "Changes detected in lint configuration files. Setting up linting tools..."
-        )
-        parent_callback(**kwargs)
-        hash_file = LINTRUNNER_CACHE_INFO[0]
-        hash_file.parent.mkdir(parents=True, exist_ok=True)
-        with hash_file.open("w") as f:
-            for file, hash in hashes.items():
-                f.write(f"{hash}  {file}\n")
-        click.echo("Linting tools set up and hashes updated.")
-    else:
-        click.echo("No changes detected in lint configuration files. Skipping setup.")
-    click.echo("Regenerating version...")
-    ctx.invoke(regenerate_version)
-    click.echo("Regenerating type stubs...")
-    ctx.invoke(regenerate_type_stubs)
-    click.echo("Done.")
-    _check_linters()
-
-
-@click.command()
-@click.option("-a", "--apply-patches", is_flag=True)
-@click.pass_context
-def lint(ctx, apply_patches, **kwargs):
-    """Lint all files."""
-    ctx.invoke(lazy_setup_lint)
-    all_files_linters = VERY_FAST_LINTERS | FAST_LINTERS
-    changed_files_linters = SLOW_LINTERS
-    cmd = LINTRUNNER_BASE_CMD
-    if apply_patches:
-        cmd += ["--apply-patches"]
-    all_files_cmd = cmd + [
-        "--take",
-        ",".join(all_files_linters),
-        "--all-files",
-    ]
-    spin.util.run(all_files_cmd)
-    changed_files_cmd = cmd + [
-        "--take",
-        ",".join(changed_files_linters),
-    ]
-    spin.util.run(changed_files_cmd)
-
-
-@click.command()
-@click.pass_context
-def fixlint(ctx, **kwargs):
-    """Autofix all files."""
-    ctx.invoke(lint, apply_patches=True)
-
-
-@click.command()
-@click.option("-a", "--apply-patches", is_flag=True)
-@click.pass_context
-def quicklint(ctx, apply_patches, **kwargs):
-    """Lint changed files."""
-    ctx.invoke(lazy_setup_lint)
-    cmd = LINTRUNNER_BASE_CMD
-    if apply_patches:
-        cmd += ["--apply-patches"]
-    spin.util.run(cmd)
-
-
-@click.command()
-@click.pass_context
-def quickfix(ctx, **kwargs):
-    """Autofix changed files."""
-    ctx.invoke(quicklint, apply_patches=True)

CMakeLists.txt

@@ -736,44 +736,6 @@ if(NOT DEFINED USE_BLAS)
   set(USE_BLAS ON)
 endif()
 
-# Prioritized Text Linker Optimization
-if(USE_PRIORITIZED_TEXT_FOR_LD)
-
-  set(LINKER_SCRIPT_FILE_IN "${CMAKE_SOURCE_DIR}/cmake/prioritized_text.txt")
-  set(LINKER_SCRIPT_FILE_OUT "${CMAKE_SOURCE_DIR}/cmake/linker_script.ld")
-
-  execute_process(
-    COMMAND ${Python_EXECUTABLE}
-            ${CMAKE_SOURCE_DIR}/tools/setup_helpers/generate_linker_script.py
-            --filein "${LINKER_SCRIPT_FILE_IN}"
-            --fout  "${LINKER_SCRIPT_FILE_OUT}"
-    RESULT_VARIABLE _gen_result
-    OUTPUT_VARIABLE _gen_output
-    ERROR_VARIABLE  _gen_error
-  )
-
-  if(NOT _gen_result EQUAL 0)
-    message(FATAL_ERROR
-      "Failed to generate linker script:\n${_gen_output}\n${_gen_error}")
-  endif()
-
-  append_cxx_flag_if_supported("-ffunction-sections" CMAKE_CXX_FLAGS)
-  append_cxx_flag_if_supported("-fdata-sections" CMAKE_CXX_FLAGS)
-  append_c_flag_if_supported("-ffunction-sections" CMAKE_C_FLAGS)
-  append_c_flag_if_supported("-fdata-sections" CMAKE_C_FLAGS)
-
-  set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -T${LINKER_SCRIPT_FILE_OUT}")
-  set(CMAKE_MODULE_LINKER_FLAGS "${CMAKE_MODULE_LINKER_FLAGS} -T${LINKER_SCRIPT_FILE_OUT}")
-
-else()
-  if(LINUX AND CPU_AARCH64)
-    message(WARNING [[
-    It is strongly recommend to enable linker script optimization for all AArch64 Linux builds.
-    To do so please export USE_PRIORITIZED_TEXT_FOR_LD=1
-    ]])
-  endif()
-endif()
-
 # Build libtorch mobile library, which contains ATen/TH ops and native support
 # for TorchScript model, but doesn't contain not-yet-unified caffe2 ops;
 if(INTERN_BUILD_MOBILE)
@@ -1440,6 +1402,9 @@ if(BUILD_JNI)
   add_subdirectory(android/pytorch_android)
 endif()
 
+include(cmake/Summary.cmake)
+caffe2_print_configuration_summary()
+
 # Parse custom debug info
 if(DEFINED USE_CUSTOM_DEBINFO)
   string(REPLACE ";" " " SOURCE_FILES "${USE_CUSTOM_DEBINFO}")
@@ -1479,5 +1444,56 @@ if(BUILD_BUNDLE_PTXAS AND USE_CUDA)
           DESTINATION "${CMAKE_INSTALL_BINDIR}")
 endif()
 
-include(cmake/Summary.cmake)
-caffe2_print_configuration_summary()
+if(USE_PRIORITIZED_TEXT_FOR_LD)
+  add_compile_options(
+    $<$<COMPILE_LANGUAGE:C,CXX>:-ffunction-sections>
+    $<$<COMPILE_LANGUAGE:C,CXX>:-fdata-sections>
+  )
+  set(LINKER_SCRIPT_FILE_OUT "${CMAKE_SOURCE_DIR}/cmake/linker_script.ld")
+  set(LINKER_SCRIPT_FILE_IN "${CMAKE_SOURCE_DIR}/cmake/prioritized_text.txt")
+
+  add_custom_command(
+    OUTPUT "${LINKER_SCRIPT_FILE_OUT}"
+    COMMAND ${Python_EXECUTABLE} ${CMAKE_SOURCE_DIR}/tools/setup_helpers/generate_linker_script.py --filein "${LINKER_SCRIPT_FILE_IN}" --fout "${LINKER_SCRIPT_FILE_OUT}"
+    DEPENDS ${CMAKE_SOURCE_DIR}/tools/setup_helpers/generate_linker_script.py "${LINKER_SCRIPT_FILE_IN}"
+    COMMENT "Generating prioritized text linker files"
+    VERBATIM
+  )
+
+  add_custom_target(generate_linker_script DEPENDS "${LINKER_SCRIPT_FILE_OUT}")
+
+  if(BUILD_PYTHON)
+    set(LINKER_OPT_TARGETS torch_python)
+  endif()
+
+  if(NOT BUILD_LIBTORCHLESS)
+    list(APPEND LINKER_OPT_TARGETS torch_cpu c10)
+    if(USE_CUDA)
+      list(APPEND LINKER_OPT_TARGETS torch_cuda c10_cuda)
+    endif()
+    if(USE_XPU)
+      list(APPEND LINKER_OPT_TARGETS torch_xpu c10_xpu)
+    endif()
+    if(USE_ROCM)
+      list(APPEND LINKER_OPT_TARGETS torch_hip c10_hip)
+    endif()
+  endif()
+
+  foreach(tgt IN LISTS LINKER_OPT_TARGETS)
+    if(TARGET ${tgt})
+      add_dependencies("${tgt}" generate_linker_script)
+      target_link_options_if_supported(${tgt} "-T,${LINKER_SCRIPT_FILE_OUT}")
+      set_property(TARGET ${tgt} APPEND PROPERTY LINK_DEPENDS "${LINKER_SCRIPT_FILE_OUT}")
+    else()
+       message(WARNING "Requested target '${tgt}' for linker script optimization was not found.")
+    endif()
+  endforeach()
+
+else()
+  if(LINUX AND CPU_AARCH64)
+    message(WARNING [[
+    It is strongly recommend to enable linker script optimization for all AArch64 Linux builds.
+    To do so please export USE_PRIORITIZED_TEXT_FOR_LD=1
+    ]])
+  endif()
+endif()

CODEOWNERS

@@ -210,12 +210,8 @@ torch/backends/cudnn/ @eqy @syed-ahmed @Aidyn-A
 /test/inductor/test_flex_attention.py @drisspg
 /test/inductor/test_flex_decoding.py @drisspg
 
-# Low Precision & Grouped GEMMs
+# Low Precision GEMMs
 /aten/src/ATen/native/cuda/Blas.cpp @drisspg @slayton58
-/aten/src/ATen/native/cuda/GroupedBlas.cpp @drisspg @slayton58
-/aten/src/ATen/native/cuda/ScaledBlas.cpp @drisspg @slayton58
 /aten/src/ATen/cuda/CUDABlas.cpp @drisspg @slayton58
 /aten/src/ATen/cuda/CUDABlas.h @drisspg @slayton58
-/aten/src/ATen/cuda/CUDAScaledBlas.cpp @drisspg @slayton58
-/aten/src/ATen/cuda/CUDAScaledBlas.h @drisspg @slayton58
 /test/test_scaled_matmul_cuda.py @drisspg @slayton58

LICENSE

@@ -37,7 +37,7 @@ Copyright (c) 2024 Tri Dao.
 All rights reserved.
 
 All contributions by Arm:
-Copyright (c) 2021, 2023-2025 Arm Limited and/or its affiliates
+Copyright (c) 2021, 2023-2024 Arm Limited and/or its affiliates
 
 All contributions from Caffe:
 Copyright(c) 2013, 2014, 2015, the respective contributors

SECURITY.md

@@ -18,8 +18,6 @@ Please report security issues using https://github.com/pytorch/pytorch/security/
 
 All reports submitted through the security advisories mechanism would **either be made public or dismissed by the team within 90 days of the submission**. If advisory has been closed on the grounds that it is not a security issue, please do not hesitate to create an [new issue](https://github.com/pytorch/pytorch/issues/new?template=bug-report.yml) as it is still likely a valid issue within the framework.
 
-**Note on crashes and out of bounds access**: PyTorch is a computational framework that performs operations on behalf of the caller. Like many low-level libraries, PyTorch generally does not validate all inputs to every function—the responsibility for providing valid arguments lies with the calling code. While crashes and out of bounds memory access should be reported as bugs, they are generally not considered security vulnerabilities in PyTorch's threat model.
-
 Please refer to the following page for our responsible disclosure policy, reward guidelines, and those things that should not be reported:
 
 https://www.facebook.com/whitehat

aten/src/ATen/Context.h

@@ -174,12 +174,6 @@ class TORCH_API Context {
   static long versionCuDNN() {
     return detail::getCUDAHooks().versionCuDNN();
   }
-  static long versionRuntimeCuDNN() {
-    return detail::getCUDAHooks().versionRuntimeCuDNN();
-  }
-  static long versionCuDNNFrontend() {
-    return detail::getCUDAHooks().versionCuDNNFrontend();
-  }
   static bool hasCuSOLVER() {
     return detail::getCUDAHooks().hasCuSOLVER();
   }

aten/src/ATen/DeviceAccelerator.h

@@ -94,11 +94,6 @@ TORCH_API inline void resetPeakStats(c10::DeviceIndex device_index) {
   at::getDeviceAllocator(device_type)->resetPeakStats(device_index);
 }
 
-TORCH_API inline std::pair<size_t, size_t> getMemoryInfo(
-    c10::DeviceIndex device_index) {
-  const auto device_type = getAccelerator(true).value();
-  return at::getDeviceAllocator(device_type)->getMemoryInfo(device_index);
-}
 } // namespace at::accelerator
 
 namespace at {

aten/src/ATen/Dispatch.h

@@ -6,7 +6,6 @@
 #include <c10/util/Half.h>
 #include <c10/util/Metaprogramming.h>
 #include <c10/util/complex.h>
-#include <torch/headeronly/core/Dispatch.h>
 
 #ifdef __CUDACC__
 #include <cuda.h> // For CUDA_VERSION
@@ -62,9 +61,12 @@ TORCH_API void record_kernel_function_dtype(std::string name);
     }                                                 \
   } while (0)
 
-#define AT_PRIVATE_CASE_TYPE_USING_HINT(enum_type, HINT, ...) \
-  THO_PRIVATE_CASE_TYPE_USING_HINT_TMPL(                      \
-      AT_PRIVATE_CHECK_SELECTIVE_BUILD, enum_type, HINT, __VA_ARGS__)
+#define AT_PRIVATE_CASE_TYPE_USING_HINT(enum_type, HINT, ...)                 \
+  case enum_type: {                                                           \
+    AT_PRIVATE_CHECK_SELECTIVE_BUILD(enum_type);                              \
+    using HINT [[maybe_unused]] = c10::impl::ScalarTypeToCPPTypeT<enum_type>; \
+    return __VA_ARGS__();                                                     \
+  }
 
 #define AT_DISPATCH_CASE(enum_type, ...) \
   AT_PRIVATE_CASE_TYPE_USING_HINT(enum_type, scalar_t, __VA_ARGS__)
@@ -93,6 +95,14 @@ TORCH_API void record_kernel_function_dtype(std::string name);
     return __VA_ARGS__();                                                   \
   }
 
+namespace detail {
+
+inline at::ScalarType scalar_type(at::ScalarType s) {
+  return s;
+}
+
+} // namespace detail
+
 // The AT_DISPATCH_* family of macros provides the ability to
 // conveniently generate specializations of a kernel over all of the
 // dtypes we care about in PyTorch.  We call it "dispatch" because
@@ -180,13 +190,27 @@ TORCH_API void record_kernel_function_dtype(std::string name);
 // but we're just being safe (and it doesn't hurt.)  Note we must
 // use it to shut up warnings about unused store.
 
-#define AT_DISPATCH_SWITCH(TYPE, NAME, ...) \
-  THO_DISPATCH_SWITCH_TMPL(                 \
-      RECORD_KERNEL_FUNCTION_DTYPE,         \
-      TORCH_CHECK_NOT_IMPLEMENTED,          \
-      TYPE,                                 \
-      NAME,                                 \
-      __VA_ARGS__)
+#define AT_DISPATCH_SWITCH(TYPE, NAME, ...)                                 \
+  [&] {                                                                     \
+    const auto& the_type = TYPE;                                            \
+    constexpr const char* at_dispatch_name = NAME;                          \
+    /* don't use TYPE again in case it is an expensive or side-effect op */ \
+    at::ScalarType _st = ::detail::scalar_type(the_type);                   \
+    RECORD_KERNEL_FUNCTION_DTYPE(at_dispatch_name, _st);                    \
+    C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wswitch-enum")             \
+    switch (_st) {                                                          \
+      __VA_ARGS__                                                           \
+      default:                                                              \
+        TORCH_CHECK_NOT_IMPLEMENTED(                                        \
+            false,                                                          \
+            '"',                                                            \
+            at_dispatch_name,                                               \
+            "\" not implemented for '",                                     \
+            toString(_st),                                                  \
+            "'");                                                           \
+    }                                                                       \
+    C10_DIAGNOSTIC_POP()                                                    \
+  }()
 
 #define AT_DISPATCH_CASE_FLOATING_TYPES(...)            \
   AT_DISPATCH_CASE(at::ScalarType::Double, __VA_ARGS__) \

aten/src/ATen/Dispatch_v2.h

@@ -1,8 +1,3 @@
-#pragma once
-
-#include <torch/headeronly/core/Dispatch_v2.h>
-
-// Get AT_DISPATCH_SWITCH and AT_DISPATCH_CASE:
 #include <ATen/Dispatch.h>
 
 // This is a new implementation of the AT_DISPATCH macro family from
@@ -79,19 +74,41 @@
 // macro expansion occurs, mediated with AT_EXPAND and AT_GUARD.  I mostly
 // relied on GPT4 to help me get it right.
 
+// Public API macros
+
 // See documentation above
 #define AT_DISPATCH_V2(TYPE, NAME, BODY, ...) \
-  THO_DISPATCH_V2_TMPL(                       \
-      AT_DISPATCH_SWITCH,                     \
-      AT_DISPATCH_CASE,                       \
-      TYPE,                                   \
-      NAME,                                   \
-      AT_WRAP(BODY),                          \
-      __VA_ARGS__)
+  AT_DISPATCH_SWITCH(TYPE, NAME, AT_AP_VAR(AT_WRAP(BODY), TYPE, __VA_ARGS__))
+
+// This macro lets you pass an arbitrary expression that may contain internal
+// commas to another macro without having the commas causing the expression
+// to be interpreted as being multiple arguments
+#define AT_WRAP(...) __VA_ARGS__
+
+#define AT_FLOAT8_TYPES                                          \
+  c10::kFloat8_e5m2, c10::kFloat8_e5m2fnuz, c10::kFloat8_e4m3fn, \
+      c10::kFloat8_e4m3fnuz, c10::kFloat8_e8m0fnu
+
+#define AT_INTEGRAL_TYPES \
+  c10::kByte, c10::kChar, c10::kInt, c10::kLong, c10::kShort
+#define AT_FLOATING_TYPES c10::kDouble, c10::kFloat
+#define AT_BAREBONES_UNSIGNED_TYPES c10::kUInt16, c10::kUInt32, c10::kUInt64
+#define AT_INTEGRAL_TYPES_V2 \
+  AT_EXPAND(AT_INTEGRAL_TYPES), AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES)
+#define AT_COMPLEX_TYPES c10::kComplexDouble, c10::kComplexFloat
+#define AT_QINT_TYPES c10::kQInt8, c10::kQUInt8, c10::kQInt32
+// NB: not *actually* all types
+#define AT_ALL_TYPES AT_EXPAND(AT_INTEGRAL_TYPES), AT_EXPAND(AT_FLOATING_TYPES)
+#define AT_ALL_TYPES_AND_COMPLEX \
+  AT_EXPAND(AT_ALL_TYPES), AT_EXPAND(AT_COMPLEX_TYPES)
+
+// Helper macros
 
-// Unused helper macros, kept for BC:
 #define AT_AP_VAR(N, T, ...) \
   AT_EXPAND(AT_CONCAT(AT_AP, AT_NUM_ARGS(__VA_ARGS__))(AT_WRAP(N), __VA_ARGS__))
+#define AT_CONCAT(a, b) AT_CONCAT_AUX(a, b)
+#define AT_CONCAT_AUX(a, b) a##b
+#define AT_EXPAND(X) X
 
 // Ensure we never have too many scalar types for the expansion here to
 // support.  To bump this, you must regenerate the macros below.
@@ -102,6 +119,12 @@ static_assert(static_cast<int>(c10::ScalarType::NumOptions) < 60);
 
 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))
+
+print(f'#define AT_NUM_ARGS(...) AT_EXPAND(AT_NUM_ARGS_AUX(__VA_ARGS__, {nums}))')
+print(f'#define AT_NUM_ARGS_AUX({args}, N, ...) N')
+
 for i in range(1, num_args+1):
     args = ', '.join(f'_{i}' for i in range(1, i+1))
     cases = ' '.join([f'AT_DISPATCH_CASE(_{j}, N)' for j in range(1, i+1)])
@@ -112,6 +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__, 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)

aten/src/ATen/LegacyBatchedTensorImpl.h

@@ -144,7 +144,7 @@ inline std::bitset<kVmapNumLevels> createVmapLevelsBitset(BatchDimsRef bdims) {
 }
 
 inline std::ostream& operator<<(std::ostream& out, const BatchDim& bdim) {
-  out << "(lvl=" << bdim.level() << ", dim=" << bdim.dim() << ')';
+  out << "(lvl=" << bdim.level() << ", dim=" << bdim.dim() << ")";
   return out;
 }
 

aten/src/ATen/TensorIndexing.cpp

@@ -9,7 +9,7 @@ namespace indexing {
 const EllipsisIndexType Ellipsis = EllipsisIndexType();
 
 std::ostream& operator<<(std::ostream& stream, const Slice& slice) {
-  stream << slice.start() << ':' << slice.stop() << ':' << slice.step();
+  stream << slice.start() << ":" << slice.stop() << ":" << slice.step();
   return stream;
 }
 
@@ -31,12 +31,12 @@ std::ostream& operator<<(std::ostream& stream, const TensorIndex& tensor_index)
 }
 
 std::ostream& operator<<(std::ostream& stream, const std::vector<TensorIndex>& tensor_indices) {
-  stream << '(';
+  stream << "(";
   for (const auto i : c10::irange(tensor_indices.size())) {
     stream << tensor_indices[i];
     if (i < tensor_indices.size() - 1) stream << ", ";
   }
-  stream << ')';
+  stream << ")";
   return stream;
 }
 

aten/src/ATen/TensorNames.cpp

@@ -113,7 +113,7 @@ void TensorNames::checkUnique(const char* op_name) const {
 std::ostream& operator<<(std::ostream& out, const TensorName& tensorname) {
   out << tensorname.name_ << " (index ";
   out << tensorname.origin_idx_ << " of ";
-  out << tensorname.origin_ << ')';
+  out << tensorname.origin_ << ")";
   return out;
 }
 

aten/src/ATen/TensorUtils.cpp

@@ -13,9 +13,9 @@ std::ostream& operator<<(std::ostream & out, const TensorGeometryArg& t) {
   if (t.pos == 0) {
     // 0 is distinguished; it usually indicates 'self' or the return
     // tensor
-    out << '\'' << t.name << '\'';
+    out << "'" << t.name << "'";
   } else {
-    out << "argument #" << t.pos << " '" << t.name << '\'';
+    out << "argument #" << t.pos << " '" << t.name << "'";
   }
   return out;
 }
@@ -154,7 +154,7 @@ void checkSameGPU(CheckedFrom c, const TensorArg& t1, const TensorArg& t2) {
       oss << "Tensor for " << t2 << " is on CPU, ";
     }
     oss << "but expected " << ((!t1->is_cpu() && !t2->is_cpu()) ? "them" : "it")
-        << " to be on GPU (while checking arguments for " << c << ')';
+        << " to be on GPU (while checking arguments for " << c << ")";
     TORCH_CHECK(false, oss.str());
   }
   TORCH_CHECK(
@@ -199,7 +199,7 @@ void checkScalarTypes(CheckedFrom c, const TensorArg& t,
         i++;
       }
       oss << "; but got " << t->toString()
-          << " instead (while checking arguments for " << c << ')';
+          << " instead (while checking arguments for " << c << ")";
       TORCH_CHECK(false, oss.str());
     }
 }

aten/src/ATen/Version.cpp

@@ -43,8 +43,8 @@ std::string get_mkldnn_version() {
     // https://github.com/intel/ideep/issues/29
     {
       const dnnl_version_t* ver = dnnl_version();
-      ss << "Intel(R) MKL-DNN v" << ver->major << '.' << ver->minor << '.' << ver->patch
-         << " (Git Hash " << ver->hash << ')';
+      ss << "Intel(R) MKL-DNN v" << ver->major << "." << ver->minor << "." << ver->patch
+         << " (Git Hash " << ver->hash << ")";
     }
   #else
     ss << "MKLDNN not found";
@@ -81,7 +81,7 @@ std::string get_openmp_version() {
           break;
       }
       if (ver_str) {
-        ss << " (a.k.a. OpenMP " << ver_str << ')';
+        ss << " (a.k.a. OpenMP " << ver_str << ")";
       }
     }
   #else
@@ -135,38 +135,38 @@ std::string show_config() {
 
 #if defined(__GNUC__)
   {
-    ss << "  - GCC " << __GNUC__ << '.' << __GNUC_MINOR__ << '\n';
+    ss << "  - GCC " << __GNUC__ << "." << __GNUC_MINOR__ << "\n";
   }
 #endif
 
 #if defined(__cplusplus)
   {
-    ss << "  - C++ Version: " << __cplusplus << '\n';
+    ss << "  - C++ Version: " << __cplusplus << "\n";
   }
 #endif
 
 #if defined(__clang_major__)
   {
-    ss << "  - clang " << __clang_major__ << '.' << __clang_minor__ << '.' << __clang_patchlevel__ << '\n';
+    ss << "  - clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__ << "\n";
   }
 #endif
 
 #if defined(_MSC_VER)
   {
-    ss << "  - MSVC " << _MSC_FULL_VER << '\n';
+    ss << "  - MSVC " << _MSC_FULL_VER << "\n";
   }
 #endif
 
 #if AT_MKL_ENABLED()
-  ss << "  - " << get_mkl_version() << '\n';
+  ss << "  - " << get_mkl_version() << "\n";
 #endif
 
 #if AT_MKLDNN_ENABLED()
-  ss << "  - " << get_mkldnn_version() << '\n';
+  ss << "  - " << get_mkldnn_version() << "\n";
 #endif
 
 #ifdef _OPENMP
-  ss << "  - " << get_openmp_version() << '\n';
+  ss << "  - " << get_openmp_version() << "\n";
 #endif
 
 #if AT_BUILD_WITH_LAPACK()
@@ -183,7 +183,7 @@ std::string show_config() {
   ss << "  - Cross compiling on MacOSX\n";
 #endif
 
-  ss << "  - "<< used_cpu_capability() << '\n';
+  ss << "  - "<< used_cpu_capability() << "\n";
 
   if (hasCUDA()) {
     ss << detail::getCUDAHooks().showConfig();
@@ -200,10 +200,10 @@ std::string show_config() {
   ss << "  - Build settings: ";
   for (const auto& pair : caffe2::GetBuildOptions()) {
     if (!pair.second.empty()) {
-      ss << pair.first << '=' << pair.second << ", ";
+      ss << pair.first << "=" << pair.second << ", ";
     }
   }
-  ss << '\n';
+  ss << "\n";
 
   // TODO: do HIP
   // TODO: do XLA

aten/src/ATen/code_template.h

@@ -209,7 +209,7 @@ struct CodeTemplate {
   // to indent correctly in the context.
   void emitIndent(std::ostream& out, size_t indent) const {
     for ([[maybe_unused]] const auto i : c10::irange(indent)) {
-      out << ' ';
+      out << " ";
     }
   }
   void emitStringWithIndents(

aten/src/ATen/core/CachingHostAllocator.h

@@ -226,8 +226,8 @@ template <
     typename B = HostBlock<S>>
 struct CachingHostAllocatorImpl {
   virtual ~CachingHostAllocatorImpl() {
-    if (active_) {
-      active_ = false;
+    active_ = false;
+    if (pinned_use_background_threads()) {
       getBackgroundThreadPool()->waitWorkComplete();
     }
   }
@@ -260,7 +260,6 @@ struct CachingHostAllocatorImpl {
     if (pinned_use_background_threads()) {
       // Launch the background thread and process events in a loop.
       static bool background_thread_flag [[maybe_unused]] = [this] {
-        active_ = true;
         getBackgroundThreadPool()->run([&]() {
           while (active_) {
             process_events();
@@ -684,9 +683,9 @@ struct CachingHostAllocatorImpl {
   alignas(hardware_destructive_interference_size) std::mutex events_mutex_;
   std::deque<std::pair<E, B*>> events_; // event queue paired with block
 
-  // Indicates whether the event-processing thread pool is active.
+  // Indicates whether the object is active.
   // Set to false in the destructor to signal background threads to stop.
-  std::atomic<bool> active_{false};
+  std::atomic<bool> active_{true};
 protected:
   alignas(hardware_destructive_interference_size) HostStatsStaged stats_;
 };

aten/src/ATen/core/Dimname.cpp

@@ -10,7 +10,7 @@ std::ostream& operator<<(std::ostream& out, const Dimname& dimname) {
   if (dimname.type() == NameType::WILDCARD) {
     out << "None";
   } else {
-    out << '\'' << dimname.symbol().toUnqualString() << '\'';
+    out << "'" << dimname.symbol().toUnqualString() << "'";
   }
   return out;
 }

aten/src/ATen/core/Range.cpp

@@ -5,7 +5,7 @@
 namespace at {
 
 std::ostream& operator<<(std::ostream& out, const Range& range) {
-  out << "Range[" << range.begin << ", " << range.end << ']';
+  out << "Range[" << range.begin << ", " << range.end << "]";
   return out;
 }
 

aten/src/ATen/core/Tensor.cpp

@@ -71,7 +71,7 @@ void TensorBase::enforce_invariants() {
 
 void TensorBase::print() const {
   if (defined()) {
-    std::cerr << '[' << toString() << ' ' << sizes() << ']' << '\n';
+    std::cerr << "[" << toString() << " " << sizes() << "]" << '\n';
   } else {
     std::cerr << "[UndefinedTensor]" << '\n';
   }

aten/src/ATen/core/TensorAccessor.h

@@ -1,6 +1,5 @@
 #pragma once
 
-#include <torch/headeronly/core/TensorAccessor.h>
 #include <c10/macros/Macros.h>
 #include <c10/util/ArrayRef.h>
 #include <c10/util/Deprecated.h>
@@ -12,37 +11,252 @@
 
 namespace at {
 
-using torch::headeronly::DefaultPtrTraits;
+// The PtrTraits argument to the TensorAccessor/GenericPackedTensorAccessor
+// is used to enable the __restrict__ keyword/modifier for the data
+// passed to cuda.
+template <typename T>
+struct DefaultPtrTraits {
+  typedef T* PtrType;
+};
+
 #if defined(__CUDACC__) || defined(__HIPCC__)
-  using torch::headeronly::RestrictPtrTraits;
+template <typename T>
+struct RestrictPtrTraits {
+  typedef T* __restrict__ PtrType;
+};
 #endif
 
+// TensorAccessorBase and TensorAccessor are used for both CPU and CUDA tensors.
+// For CUDA tensors it is used in device code (only). This means that we restrict ourselves
+// to functions and types available there (e.g. IntArrayRef isn't).
+
+// The PtrTraits argument is only relevant to cuda to support `__restrict__` pointers.
 template<typename T, size_t N, template <typename U> class PtrTraits = DefaultPtrTraits, typename index_t = int64_t>
-using TensorAccessorBase = torch::headeronly::detail::TensorAccessorBase<c10::IntArrayRef, T, N, PtrTraits, index_t>;
+class TensorAccessorBase {
+public:
+  typedef typename PtrTraits<T>::PtrType PtrType;
 
+  C10_HOST_DEVICE TensorAccessorBase(
+      PtrType data_,
+      const index_t* sizes_,
+      const index_t* strides_)
+      : data_(data_), sizes_(sizes_), strides_(strides_) {}
+  C10_HOST IntArrayRef sizes() const {
+    return IntArrayRef(sizes_,N);
+  }
+  C10_HOST IntArrayRef strides() const {
+    return IntArrayRef(strides_,N);
+  }
+  C10_HOST_DEVICE index_t stride(index_t i) const {
+    return strides_[i];
+  }
+  C10_HOST_DEVICE index_t size(index_t i) const {
+    return sizes_[i];
+  }
+  C10_HOST_DEVICE PtrType data() {
+    return data_;
+  }
+  C10_HOST_DEVICE const PtrType data() const {
+    return data_;
+  }
+protected:
+  PtrType data_;
+  const index_t* sizes_;
+  const index_t* strides_;
+};
+
+// The `TensorAccessor` is typically instantiated for CPU `Tensor`s using
+// `Tensor.accessor<T, N>()`.
+// For CUDA `Tensor`s, `GenericPackedTensorAccessor` is used on the host and only
+// indexing on the device uses `TensorAccessor`s.
 template<typename T, size_t N, template <typename U> class PtrTraits = DefaultPtrTraits, typename index_t = int64_t>
-using TensorAccessor = torch::headeronly::detail::TensorAccessor<c10::IntArrayRef, T, N, PtrTraits, index_t>;
+class TensorAccessor : public TensorAccessorBase<T,N,PtrTraits,index_t> {
+public:
+  typedef typename PtrTraits<T>::PtrType PtrType;
 
-namespace detail {
+  C10_HOST_DEVICE TensorAccessor(
+      PtrType data_,
+      const index_t* sizes_,
+      const index_t* strides_)
+      : TensorAccessorBase<T, N, PtrTraits, index_t>(data_,sizes_,strides_) {}
 
-template <size_t N, typename index_t>
-struct IndexBoundsCheck {
-    IndexBoundsCheck(index_t i) {
-      TORCH_CHECK_INDEX(
+  C10_HOST_DEVICE TensorAccessor<T, N - 1, PtrTraits, index_t> operator[](index_t i) {
+    return TensorAccessor<T,N-1,PtrTraits,index_t>(this->data_ + this->strides_[0]*i,this->sizes_+1,this->strides_+1);
+  }
+
+  C10_HOST_DEVICE const TensorAccessor<T, N-1, PtrTraits, index_t> operator[](index_t i) const {
+    return TensorAccessor<T,N-1,PtrTraits,index_t>(this->data_ + this->strides_[0]*i,this->sizes_+1,this->strides_+1);
+  }
+};
+
+template<typename T, template <typename U> class PtrTraits, typename index_t>
+class TensorAccessor<T,1,PtrTraits,index_t> : public TensorAccessorBase<T,1,PtrTraits,index_t> {
+public:
+  typedef typename PtrTraits<T>::PtrType PtrType;
+
+  C10_HOST_DEVICE TensorAccessor(
+      PtrType data_,
+      const index_t* sizes_,
+      const index_t* strides_)
+      : TensorAccessorBase<T, 1, PtrTraits, index_t>(data_,sizes_,strides_) {}
+  C10_HOST_DEVICE T & operator[](index_t i) {
+    // NOLINTNEXTLINE(clang-analyzer-core.NullDereference)
+    return this->data_[this->strides_[0]*i];
+  }
+  C10_HOST_DEVICE const T & operator[](index_t i) const {
+    return this->data_[this->strides_[0]*i];
+  }
+};
+
+
+// GenericPackedTensorAccessorBase and GenericPackedTensorAccessor are used on for CUDA `Tensor`s on the host
+// and as
+// In contrast to `TensorAccessor`s, they copy the strides and sizes on instantiation (on the host)
+// in order to transfer them on the device when calling kernels.
+// On the device, indexing of multidimensional tensors gives to `TensorAccessor`s.
+// Use RestrictPtrTraits as PtrTraits if you want the tensor's data pointer to be marked as __restrict__.
+// Instantiation from data, sizes, strides is only needed on the host and std::copy isn't available
+// on the device, so those functions are host only.
+template<typename T, size_t N, template <typename U> class PtrTraits = DefaultPtrTraits, typename index_t = int64_t>
+class GenericPackedTensorAccessorBase {
+public:
+  typedef typename PtrTraits<T>::PtrType PtrType;
+  C10_HOST GenericPackedTensorAccessorBase(
+      PtrType data_,
+      const index_t* sizes_,
+      const index_t* strides_)
+      : data_(data_) {
+    std::copy(sizes_, sizes_ + N, std::begin(this->sizes_));
+    std::copy(strides_, strides_ + N, std::begin(this->strides_));
+  }
+
+  // if index_t is not int64_t, we want to have an int64_t constructor
+  template <typename source_index_t, class = std::enable_if_t<std::is_same_v<source_index_t, int64_t>>>
+  C10_HOST GenericPackedTensorAccessorBase(
+      PtrType data_,
+      const source_index_t* sizes_,
+      const source_index_t* strides_)
+      : data_(data_) {
+    for (const auto i : c10::irange(N)) {
+      this->sizes_[i] = sizes_[i];
+      this->strides_[i] = strides_[i];
+    }
+  }
+
+  C10_HOST_DEVICE index_t stride(index_t i) const {
+    return strides_[i];
+  }
+  C10_HOST_DEVICE index_t size(index_t i) const {
+    return sizes_[i];
+  }
+  C10_HOST_DEVICE PtrType data() {
+    return data_;
+  }
+  C10_HOST_DEVICE const PtrType data() const {
+    return data_;
+  }
+protected:
+  PtrType data_;
+  // NOLINTNEXTLINE(*c-arrays*)
+  index_t sizes_[N];
+  // NOLINTNEXTLINE(*c-arrays*)
+  index_t strides_[N];
+  C10_HOST void bounds_check_(index_t i) const {
+    TORCH_CHECK_INDEX(
         0 <= i && i < index_t{N},
         "Index ",
         i,
         " is not within bounds of a tensor of dimension ",
         N);
-    }
+  }
 };
-}  // namespace detail
 
 template<typename T, size_t N, template <typename U> class PtrTraits = DefaultPtrTraits, typename index_t = int64_t>
-using GenericPackedTensorAccessorBase = torch::headeronly::detail::GenericPackedTensorAccessorBase<detail::IndexBoundsCheck<N, index_t>, T, N, PtrTraits, index_t>;
+class GenericPackedTensorAccessor : public GenericPackedTensorAccessorBase<T,N,PtrTraits,index_t> {
+public:
+  typedef typename PtrTraits<T>::PtrType PtrType;
+
+  C10_HOST GenericPackedTensorAccessor(
+      PtrType data_,
+      const index_t* sizes_,
+      const index_t* strides_)
+      : GenericPackedTensorAccessorBase<T, N, PtrTraits, index_t>(data_, sizes_, strides_) {}
+
+  // if index_t is not int64_t, we want to have an int64_t constructor
+  template <typename source_index_t, class = std::enable_if_t<std::is_same_v<source_index_t, int64_t>>>
+  C10_HOST GenericPackedTensorAccessor(
+      PtrType data_,
+      const source_index_t* sizes_,
+      const source_index_t* strides_)
+      : GenericPackedTensorAccessorBase<T, N, PtrTraits, index_t>(data_, sizes_, strides_) {}
+
+  C10_DEVICE TensorAccessor<T, N - 1, PtrTraits, index_t> operator[](index_t i) {
+    index_t* new_sizes = this->sizes_ + 1;
+    index_t* new_strides = this->strides_ + 1;
+    return TensorAccessor<T,N-1,PtrTraits,index_t>(this->data_ + this->strides_[0]*i, new_sizes, new_strides);
+  }
+
+  C10_DEVICE const TensorAccessor<T, N - 1, PtrTraits, index_t> operator[](index_t i) const {
+    const index_t* new_sizes = this->sizes_ + 1;
+    const index_t* new_strides = this->strides_ + 1;
+    return TensorAccessor<T,N-1,PtrTraits,index_t>(this->data_ + this->strides_[0]*i, new_sizes, new_strides);
+  }
+
+  /// Returns a PackedTensorAccessor of the same dimension after transposing the
+  /// two dimensions given. Does not actually move elements; transposition is
+  /// made by permuting the size/stride arrays. If the dimensions are not valid,
+  /// asserts.
+  C10_HOST GenericPackedTensorAccessor<T, N, PtrTraits, index_t> transpose(
+      index_t dim1,
+      index_t dim2) const {
+    this->bounds_check_(dim1);
+    this->bounds_check_(dim2);
+    GenericPackedTensorAccessor<T, N, PtrTraits, index_t> result(
+        this->data_, this->sizes_, this->strides_);
+    std::swap(result.strides_[dim1], result.strides_[dim2]);
+    std::swap(result.sizes_[dim1], result.sizes_[dim2]);
+    return result;
+  }
+};
+
+template<typename T, template <typename U> class PtrTraits, typename index_t>
+class GenericPackedTensorAccessor<T,1,PtrTraits,index_t> : public GenericPackedTensorAccessorBase<T,1,PtrTraits,index_t> {
+public:
+  typedef typename PtrTraits<T>::PtrType PtrType;
+  C10_HOST GenericPackedTensorAccessor(
+      PtrType data_,
+      const index_t* sizes_,
+      const index_t* strides_)
+      : GenericPackedTensorAccessorBase<T, 1, PtrTraits, index_t>(data_, sizes_, strides_) {}
+
+  // if index_t is not int64_t, we want to have an int64_t constructor
+  template <typename source_index_t, class = std::enable_if_t<std::is_same_v<source_index_t, int64_t>>>
+  C10_HOST GenericPackedTensorAccessor(
+      PtrType data_,
+      const source_index_t* sizes_,
+      const source_index_t* strides_)
+      : GenericPackedTensorAccessorBase<T, 1, PtrTraits, index_t>(data_, sizes_, strides_) {}
+
+  C10_DEVICE T & operator[](index_t i) {
+    return this->data_[this->strides_[0] * i];
+  }
+  C10_DEVICE const T& operator[](index_t i) const {
+    return this->data_[this->strides_[0]*i];
+  }
+
+  // Same as in the general N-dimensional case, but note that in the
+  // 1-dimensional case the returned PackedTensorAccessor will always be an
+  // identical copy of the original
+  C10_HOST GenericPackedTensorAccessor<T, 1, PtrTraits, index_t> transpose(
+      index_t dim1,
+      index_t dim2) const {
+    this->bounds_check_(dim1);
+    this->bounds_check_(dim2);
+    return GenericPackedTensorAccessor<T, 1, PtrTraits, index_t>(
+        this->data_, this->sizes_, this->strides_);
+  }
+};
 
-template<typename T, size_t N, template <typename U> class PtrTraits = DefaultPtrTraits, typename index_t = int64_t>
-using GenericPackedTensorAccessor = torch::headeronly::detail::GenericPackedTensorAccessor<TensorAccessor<T, N-1, PtrTraits, index_t>, detail::IndexBoundsCheck<N, index_t>, T, N, PtrTraits, index_t>;
 
 // Can't put this directly into the macro function args because of commas
 #define AT_X GenericPackedTensorAccessor<T, N, PtrTraits, index_t>

aten/src/ATen/core/TensorBase.h

@@ -245,9 +245,6 @@ class TORCH_API TensorBase {
   size_t weak_use_count() const noexcept {
     return impl_.weak_use_count();
   }
-  bool is_uniquely_owned() const noexcept {
-    return impl_.is_uniquely_owned();
-  }
 
   std::string toString() const;
 

aten/src/ATen/core/Vitals.cpp

@@ -9,8 +9,8 @@ APIVitals VitalsAPI;
 
 std::ostream& operator<<(std::ostream& os, TorchVital const& tv) {
   for (const auto& m : tv.attrs) {
-    os << "[TORCH_VITAL] " << tv.name << '.' << m.first << "\t\t "
-       << m.second.value << '\n';
+    os << "[TORCH_VITAL] " << tv.name << "." << m.first << "\t\t "
+       << m.second.value << "\n";
   }
   return os;
 }

aten/src/ATen/core/alias_info.h

@@ -100,18 +100,18 @@ inline bool operator==(const AliasInfo& lhs, const AliasInfo& rhs) {
 
 // this does match the way things are represented in the schema
 inline std::ostream& operator<<(std::ostream& out, const AliasInfo& aliasInfo) {
-  out << '(';
+  out << "(";
   bool first = true;
   for (const auto& set : aliasInfo.beforeSets()) {
     if (first) {
       first = false;
     } else {
-      out << '|';
+      out << "|";
     }
     out << set.toUnqualString();
   }
   if (aliasInfo.isWrite()) {
-    out << '!';
+    out << "!";
   }
   if (aliasInfo.beforeSets() != aliasInfo.afterSets()) {
     out << " -> ";
@@ -120,12 +120,12 @@ inline std::ostream& operator<<(std::ostream& out, const AliasInfo& aliasInfo) {
       if (first) {
         first = false;
       } else {
-        out << '|';
+        out << "|";
       }
       out << set.toUnqualString();
     }
   }
-  out << ')';
+  out << ")";
   return out;
 }
 } // namespace c10

aten/src/ATen/core/blob.h

@@ -198,7 +198,7 @@ inline void swap(Blob& lhs, Blob& rhs)  noexcept {
 }
 
 inline std::ostream& operator<<(std::ostream& out, const Blob& v) {
-  return out << "Blob[" << v.TypeName() << ']';
+  return out << "Blob[" << v.TypeName() << "]";
 }
 
 } // namespace caffe2

aten/src/ATen/core/class_type.cpp

@@ -456,8 +456,8 @@ bool ClassType::isSubtypeOfExt(const Type& rhs, std::ostream* why_not) const {
           *why_not << "Method on class '" << repr_str()
                    << "' (1) is not compatible with interface '"
                    << rhs.repr_str() << "' (2)\n"
-                   << "  (1) " << self_method->getSchema() << '\n'
-                   << "  (2) " << schema << '\n';
+                   << "  (1) " << self_method->getSchema() << "\n"
+                   << "  (2) " << schema << "\n";
         }
         return false;
       }

aten/src/ATen/core/class_type.h

@@ -100,7 +100,7 @@ struct TORCH_API ClassType : public NamedType {
   std::string repr_str() const override {
     std::stringstream ss;
     ss << str()
-       << " (of Python compilation unit at: " << compilation_unit().get() << ')';
+       << " (of Python compilation unit at: " << compilation_unit().get() << ")";
     return ss.str();
   }
 

aten/src/ATen/core/dispatch/DispatchKeyExtractor.cpp

@@ -58,12 +58,12 @@ std::string DispatchKeyExtractor::dumpState() const {
   std::ostringstream oss;
   for (const auto i : c10::irange(c10::utils::bitset::NUM_BITS())) {
     if (dispatch_arg_indices_reverse_.get(i)) {
-      oss << '1';
+      oss << "1";
     } else {
-      oss << '0';
+      oss << "0";
     }
   }
-  oss << ' ' << nonFallthroughKeys_ << '\n';
+  oss << " " << nonFallthroughKeys_ << "\n";
   return oss.str();
 }
 

aten/src/ATen/core/dispatch/Dispatcher.cpp

@@ -69,8 +69,8 @@ private:
 
 void _print_dispatch_trace(const std::string& label, const std::string& op_name, const DispatchKeySet& dispatchKeySet) {
   auto nesting_value = dispatch_trace_nesting_value();
-  for (int64_t i = 0; i < nesting_value; ++i) std::cerr << ' ';
-  std::cerr << label << " op=[" << op_name << "], key=[" << toString(dispatchKeySet.highestPriorityTypeId()) << ']' << std::endl;
+  for (int64_t i = 0; i < nesting_value; ++i) std::cerr << " ";
+  std::cerr << label << " op=[" << op_name << "], key=[" << toString(dispatchKeySet.highestPriorityTypeId()) << "]" << std::endl;
 }
 } // namespace detail
 

aten/src/ATen/core/dispatch/OperatorEntry.cpp

@@ -570,7 +570,7 @@ void OperatorEntry::checkInvariants() const {
 
 std::string OperatorEntry::listAllDispatchKeys() const {
   std::ostringstream str;
-  str << '[';
+  str << "[";
 
   bool has_kernels = false;
   for (auto k : allDispatchKeysInFullSet()) {
@@ -584,7 +584,7 @@ std::string OperatorEntry::listAllDispatchKeys() const {
     str << k;
     has_kernels = true;
   }
-  str << ']';
+  str << "]";
   return str.str();
 }
 
@@ -683,12 +683,12 @@ void OperatorEntry::setReportErrorCallback_(std::unique_ptr<c10::SafePyObject> c
 // This WON'T report backend fallbacks.
 std::string OperatorEntry::dumpState() const {
   std::ostringstream oss;
-  oss << "name: " << name_ << '\n';
+  oss << "name: " << name_ << "\n";
   if (schema_) {
-    oss << "schema: " << schema_->schema << '\n';
-    oss << "debug: " << schema_->debug << '\n';
+    oss << "schema: " << schema_->schema << "\n";
+    oss << "debug: " << schema_->debug << "\n";
     oss << "alias analysis kind: " << toString(schema_->schema.aliasAnalysis())
-        << (schema_->schema.isDefaultAliasAnalysisKind() ? " (default)" : "") << '\n';
+        << (schema_->schema.isDefaultAliasAnalysisKind() ? " (default)" : "") << "\n";
   } else {
     oss << "schema: (none)\n";
   }

aten/src/ATen/core/function_schema.cpp

@@ -7,7 +7,7 @@
 namespace c10 {
 
 void FunctionSchema::dump() const {
-  std::cout << *this << '\n';
+  std::cout << *this << "\n";
 }
 
 const std::vector<Argument>& FunctionSchema::getCorrectList(SchemaArgType type) const {
@@ -210,9 +210,9 @@ std::ostream& operator<<(std::ostream& out, const FunctionSchema& schema) {
 
   out << schema.name();
   if (!schema.overload_name().empty()) {
-    out << '.' << schema.overload_name();
+    out << "." << schema.overload_name();
   }
-  out << '(';
+  out << "(";
 
   bool seen_kwarg_only = false;
   for (const auto i : c10::irange(schema.arguments().size())) {
@@ -273,7 +273,7 @@ std::ostream& operator<<(std::ostream& out, const FunctionSchema& schema) {
   }
 
   if (need_paren) {
-    out << '(';
+    out << "(";
   }
   for (const auto i : c10::irange(returns.size())) {
     if (i > 0) {
@@ -288,7 +288,7 @@ std::ostream& operator<<(std::ostream& out, const FunctionSchema& schema) {
     out << "...";
   }
   if (need_paren) {
-    out << ')';
+    out << ")";
   }
   return out;
 }
@@ -471,7 +471,7 @@ bool FunctionSchema::isForwardCompatibleWith(
     if (!arguments().at(i).isForwardCompatibleWith(old.arguments().at(i))) {
       if (why_not) {
         why_not
-            << '\'' << arguments().at(i).name() << '\''
+            << "'" << arguments().at(i).name() << "'"
             << " is not forward compatible with the older version of the schema";
       }
       return false;
@@ -511,7 +511,7 @@ bool FunctionSchema::isForwardCompatibleWith(
              .isForwardCompatibleWith(old.arguments().at(i))) {
       if (why_not) {
         why_not << "Out argument '"
-                << '\'' << arguments().at(i).name()
+                << "'" << arguments().at(i).name()
                 << " is not FC with the older version of the schema";
       }
       return false;

aten/src/ATen/core/function_schema.h

@@ -571,7 +571,7 @@ inline std::ostream& operator<<(std::ostream& out, const Argument& arg) {
     if (arg.N()) {
         N = std::to_string(*arg.N());
     }
-    out << '[' << N << ']';
+    out << "[" << N << "]";
   } else {
     out << unopt_type->str();
   }
@@ -582,15 +582,15 @@ inline std::ostream& operator<<(std::ostream& out, const Argument& arg) {
   }
 
   if (is_opt) {
-    out << '?';
+    out << "?";
   }
 
   if (!arg.name().empty()) {
-    out << ' ' << arg.name();
+    out << " " << arg.name();
   }
 
   if (arg.default_value()) {
-    out << '=';
+    out << "=";
     if ((type->kind() == c10::TypeKind::StringType ||
         unopt_type->kind() == c10::TypeKind::StringType) &&
         arg.default_value().value().isString()) {

aten/src/ATen/core/ivalue.cpp

@@ -66,7 +66,7 @@ bool operator==(const ivalue::Tuple& lhs, const ivalue::Tuple& rhs) {
 }
 
 std::ostream& operator<<(std::ostream& out, const ivalue::EnumHolder& v) {
-  out << v.qualifiedClassName() << '.' << v.name();
+  out << v.qualifiedClassName() << "." << v.name();
   return out;
 }
 
@@ -526,7 +526,7 @@ std::ostream& printMaybeAnnotatedList(
       !elementTypeCanBeInferredFromMembers(list_elem_type)) {
     out << "annotate(" << the_list.type<c10::Type>()->annotation_str() << ", ";
     printList(out, the_list.toListRef(), "[", "]", formatter);
-    out << ')';
+    out << ")";
     return out;
   } else {
     return printList(out, the_list.toListRef(), "[", "]", formatter);
@@ -538,7 +538,7 @@ std::ostream& printDict(
     std::ostream& out,
     const Dict& v,
     const IValueFormatter& formatter) {
-  out << '{';
+  out << "{";
 
   bool first = true;
   for (const auto& pair : v) {
@@ -552,7 +552,7 @@ std::ostream& printDict(
     first = false;
   }
 
-  out << '}';
+  out << "}";
   return out;
 }
 }
@@ -565,8 +565,8 @@ static std::ostream& printMaybeAnnotatedDict(
   auto value_type = the_dict.type()->castRaw<DictType>()->getValueType();
   if (the_dict.toGenericDict().empty() ||
       !elementTypeCanBeInferredFromMembers(value_type)) {
-    out << "annotate(" << the_dict.type<c10::Type>()->annotation_str() << ',';
-    printDict(out, the_dict.toGenericDict(), formatter) << ')';
+    out << "annotate(" << the_dict.type<c10::Type>()->annotation_str() << ",";
+    printDict(out, the_dict.toGenericDict(), formatter) << ")";
   } else {
     return printDict(out, the_dict.toGenericDict(), formatter);
   }
@@ -577,7 +577,7 @@ static std::ostream& printComplex(std::ostream & out, const IValue & v) {
   c10::complex<double> d = v.toComplexDouble();
   IValue real(d.real()), imag(std::abs(d.imag()));
   auto sign = d.imag() >= 0 ? '+' : '-';
-  return out << real << sign << imag << 'j';
+  return out << real << sign << imag << "j";
 }
 
 std::ostream& IValue::repr(
@@ -605,9 +605,9 @@ std::ostream& IValue::repr(
         if (static_cast<double>(i) == d) {
           // -0.0 (signed zero) needs to be parsed as -0.
           if (i == 0 && std::signbit(d)) {
-            return out << '-' << i << '.';
+            return out << "-" << i << ".";
           }
-          return out << i << '.';
+          return out << i << ".";
         }
       }
       auto orig_prec = out.precision();
@@ -643,20 +643,20 @@ std::ostream& IValue::repr(
       device_stream << v.toDevice();
       out << "torch.device(";
       c10::printQuotedString(out, device_stream.str());
-      return out << ')';
+      return out << ")";
     }
     case IValue::Tag::Generator: {
       auto generator = v.toGenerator();
       out << "torch.Generator(device=";
       c10::printQuotedString(out, generator.device().str());
-      out << ", seed=" << generator.current_seed() << ')';
+      out << ", seed=" << generator.current_seed() << ")";
       return out;
     }
     case IValue::Tag::GenericDict:
       return printMaybeAnnotatedDict(out, v, formatter);
     case IValue::Tag::Enum: {
       auto enum_holder = v.toEnumHolder();
-      return out << enum_holder->qualifiedClassName() << '.' <<
+      return out << enum_holder->qualifiedClassName() << "." <<
           enum_holder->name();
     }
     case IValue::Tag::Object: {
@@ -801,7 +801,7 @@ std::ostream& operator<<(std::ostream & out, const IValue & v) {
       if (c == FP_NORMAL || c == FP_ZERO) {
         int64_t i = static_cast<int64_t>(d);
         if (static_cast<double>(i) == d) {
-          return out << i << '.';
+          return out << i << ".";
         }
       }
       auto orig_prec = out.precision();
@@ -852,7 +852,7 @@ std::ostream& operator<<(std::ostream & out, const IValue & v) {
       return printDict(out, v.toGenericDict(), formatter);
     case IValue::Tag::PyObject: {
       auto py_obj = v.toPyObject();
-      return out << "<PyObject at" << py_obj << '>';
+      return out << "<PyObject at" << py_obj << ">";
     }
     case IValue::Tag::Generator:
       return out << "Generator";
@@ -862,22 +862,22 @@ std::ostream& operator<<(std::ostream & out, const IValue & v) {
       // TODO we should attempt to call __str__ if the object defines it.
       auto obj = v.toObject();
       // print this out the way python would do it
-      return out << '<' << obj->name() << " object at " << obj.get() << '>';
+      return out << "<" << obj->name() << " object at " << obj.get() << ">";
     }
     case IValue::Tag::Enum: {
       auto enum_holder = v.toEnumHolder();
-      return out << "Enum<" << enum_holder->unqualifiedClassName() << '.' <<
-          enum_holder->name() << '>';
+      return out << "Enum<" << enum_holder->unqualifiedClassName() << "." <<
+          enum_holder->name() << ">";
     }
 
   }
-  return out << "<Invalid IValue tag=" << std::to_string(static_cast<uint32_t>(v.tag)) << '>';
+  return out << "<Invalid IValue tag=" << std::to_string(static_cast<uint32_t>(v.tag)) << ">";
 }
 
 #undef TORCH_FORALL_TAGS
 
 void IValue::dump() const {
-  std::cout << *this << '\n';
+  std::cout << *this << "\n";
 }
 
 std::shared_ptr<ClassType> ivalue::Object::type() const {
@@ -1050,7 +1050,7 @@ c10::intrusive_ptr<ivalue::Object> ivalue::Object::deepcopy(
       std::stringstream err;
       err << "Cannot serialize custom bound C++ class";
       if (auto qualname = type()->name()) {
-        err << ' ' << qualname->qualifiedName();
+        err << " " << qualname->qualifiedName();
       }
       err << ". Please define serialization methods via def_pickle() for "
             "this class.";

aten/src/ATen/core/ivalue.h

@@ -18,8 +18,6 @@
 #include <unordered_set>
 #include <utility>
 
-C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wswitch-default")
-
 namespace torch {
 class TORCH_API CustomClassHolder : public c10::intrusive_ptr_target {};
 namespace jit {
@@ -1632,6 +1630,4 @@ struct TORCH_API WeakOrStrongTypePtr {
 
 } // namespace c10
 
-C10_DIAGNOSTIC_POP()
-
 #include <ATen/core/ivalue_inl.h> // IWYU pragma: keep

aten/src/ATen/core/ivalue_inl.h

@@ -29,8 +29,6 @@
 #include <c10/util/intrusive_ptr.h>
 #include <c10/util/irange.h>
 
-C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wswitch-default")
-
 namespace torch {
 namespace jit {
 struct Function;
@@ -2569,5 +2567,3 @@ TypePtr IValue::type() const {
 }
 
 } // namespace c10
-
-C10_DIAGNOSTIC_POP()

aten/src/ATen/core/jit_type.h

@@ -211,7 +211,7 @@ struct TORCH_API OptionalType : public UnionType {
 
   std::string str() const override {
     std::stringstream ss;
-    ss << getElementType()->str() << '?';
+    ss << getElementType()->str() << "?";
     return ss.str();
   }
 
@@ -240,7 +240,7 @@ struct TORCH_API OptionalType : public UnionType {
 
   std::string annotation_str_impl(const TypePrinter& printer = nullptr) const override {
     std::stringstream ss;
-    ss << "Optional[" << getElementType()->annotation_str(printer) << ']';
+    ss << "Optional[" << getElementType()->annotation_str(printer) << "]";
     return ss.str();
   }
 };
@@ -906,7 +906,7 @@ struct TORCH_API ListType
 
   std::string annotation_str_impl(const TypePrinter& printer = nullptr) const override {
     std::stringstream ss;
-    ss << "List[" << getElementType()->annotation_str(printer) << ']';
+    ss << "List[" << getElementType()->annotation_str(printer) << "]";
     return ss.str();
   }
 };
@@ -946,7 +946,7 @@ struct TORCH_API DictType : public SharedType {
   std::string str() const override {
     std::stringstream ss;
     ss << "Dict(" << getKeyType()->str() << ", " << getValueType()->str()
-       << ')';
+       << ")";
     return ss.str();
   }
 
@@ -1018,7 +1018,7 @@ struct TORCH_API FutureType
 
   std::string str() const override {
     std::stringstream ss;
-    ss << "Future(" << getElementType()->str() << ')';
+    ss << "Future(" << getElementType()->str() << ")";
     return ss.str();
   }
   TypePtr createWithContained(
@@ -1041,7 +1041,7 @@ struct TORCH_API FutureType
 
   std::string annotation_str_impl(const TypePrinter& printer = nullptr) const override {
     std::stringstream ss;
-    ss << "Future[" << getElementType()->annotation_str(printer) << ']';
+    ss << "Future[" << getElementType()->annotation_str(printer) << "]";
     return ss.str();
   }
 };
@@ -1060,7 +1060,7 @@ struct TORCH_API AwaitType
 
   std::string str() const override {
     std::stringstream ss;
-    ss << "Await(" << getElementType()->str() << ')';
+    ss << "Await(" << getElementType()->str() << ")";
     return ss.str();
   }
   TypePtr createWithContained(
@@ -1083,7 +1083,7 @@ struct TORCH_API AwaitType
 
   std::string annotation_str_impl(const TypePrinter& printer = nullptr) const override {
     std::stringstream ss;
-    ss << "Await[" << getElementType()->annotation_str(printer) << ']';
+    ss << "Await[" << getElementType()->annotation_str(printer) << "]";
     return ss.str();
   }
 };
@@ -1102,7 +1102,7 @@ struct TORCH_API RRefType
 
   std::string str() const override {
     std::stringstream ss;
-    ss << "RRef(" << getElementType()->str() << ')';
+    ss << "RRef(" << getElementType()->str() << ")";
     return ss.str();
   }
   TypePtr createWithContained(
@@ -1115,7 +1115,7 @@ struct TORCH_API RRefType
 
   std::string annotation_str_impl(const TypePrinter& printer = nullptr) const override {
     std::stringstream ss;
-    ss << "RRef[" << getElementType()->annotation_str(printer) << ']';
+    ss << "RRef[" << getElementType()->annotation_str(printer) << "]";
     return ss.str();
   }
 };

aten/src/ATen/core/operator_name.cpp

@@ -11,7 +11,7 @@ std::string toString(const OperatorName& opName) {
 std::ostream& operator<<(std::ostream& os, const OperatorName& opName) {
   os << opName.name;
   if (!opName.overload_name.empty()) {
-    os << '.' << opName.overload_name;
+    os << "." << opName.overload_name;
   }
   return os;
 }

aten/src/ATen/core/tensor_type.cpp

@@ -65,7 +65,7 @@ VaryingShape<T> VaryingShape<T>::merge(const VaryingShape<T>& other) const {
 
 template <typename T>
 std::ostream& operator<<(std::ostream& out, const VaryingShape<T>& vs) {
-  out << '(';
+  out << "(";
   if (!vs.size()) {
     out << "*)";
     return out;
@@ -79,10 +79,10 @@ std::ostream& operator<<(std::ostream& out, const VaryingShape<T>& vs) {
     if (v.has_value()) {
       out << v.value();
     } else {
-      out << '*';
+      out << "*";
     }
   }
-  out << ')';
+  out << ")";
   return out;
 }
 
@@ -105,7 +105,7 @@ std::ostream& operator<<(
   }
   auto sizes_opt = ss.sizes();
 
-  os << '(';
+  os << "(";
   for (size_t i = 0; i < rank_opt.value(); i++) {
     if (i > 0) {
       os << ", ";
@@ -113,10 +113,10 @@ std::ostream& operator<<(
     if(sizes_opt.has_value() && sizes_opt.value()[i].is_static()) {
       os << sizes_opt.value()[i];
     } else {
-      os << '*';
+      os << "*";
     }
   }
-  os << ')';
+  os << ")";
 
   return os;
 }
@@ -131,17 +131,17 @@ std::ostream& operator<<(std::ostream& os, const ShapeSymbol& s) {
 }
 
 std::ostream& operator<<(std::ostream& os, const Stride& s) {
-  os << '{';
+  os << "{";
   if (s.stride_index_.has_value()) {
     os << *s.stride_index_;
   } else {
-    os << '*';
+    os << "*";
   }
-  os << ':';
+  os << ":";
   if (s.stride_.has_value()) {
     os << *s.stride_;
   } else {
-    os << '*';
+    os << "*";
   }
   os << '}';
   return os;

aten/src/ATen/core/type.cpp

@@ -67,7 +67,7 @@ std::ostream& operator<<(std::ostream & out, const Type & t) {
       bool has_valid_strides_info = ndim > 0 &&
           value->strides().isComplete() && value->strides().size() == ndim;
 
-      out << '(';
+      out << "(";
       size_t i = 0;
       bool symbolic = type_verbosity() == TypeVerbosity::Symbolic;
       for (i = 0; i < *ndim; ++i) {
@@ -79,7 +79,7 @@ std::ostream& operator<<(std::ostream & out, const Type & t) {
         } else if (symbolic) {
           out << value->symbolic_sizes().at(i);
         } else {
-          out << '*';
+          out << "*";
         }
       }
       if (has_valid_strides_info &&
@@ -91,7 +91,7 @@ std::ostream& operator<<(std::ostream & out, const Type & t) {
           }
           out << value->strides()[i].value();
         }
-        out << ']';
+        out << "]";
       }
       if (type_verbosity() >= TypeVerbosity::Full) {
         if (value->requiresGrad()) {
@@ -107,12 +107,12 @@ std::ostream& operator<<(std::ostream & out, const Type & t) {
           out << "device=" << *value->device();
         }
       }
-      out << ')';
+      out << ")";
     } else {
       if (type_verbosity() >= TypeVerbosity::Full) {
         size_t i = 0;
         if (value->requiresGrad()) {
-          out << '('
+          out << "("
               << "requires_grad=" << *value->requiresGrad();
           i++;
         }
@@ -120,7 +120,7 @@ std::ostream& operator<<(std::ostream & out, const Type & t) {
           out << ((i++ > 0) ? ", " : "(") << "device=" << *value->device();
         }
         if (i > 0) {
-          out << ')';
+          out << ")";
         }
       }
     }
@@ -133,18 +133,18 @@ std::ostream& operator<<(std::ostream & out, const Type & t) {
     out << *prim << "[]";
   } else if (t.kind() == TypeKind::OptionalType) {
     auto prim = t.castRaw<OptionalType>()->getElementType();
-    out << *prim << '?';
+    out << *prim << "?";
   } else if(t.kind() == TypeKind::FutureType) {
     auto elem = t.castRaw<FutureType>()->getElementType();
-    out << "Future[" << *elem << ']';
+    out << "Future[" << *elem << "]";
   } else if(t.kind() == TypeKind::RRefType) {
     auto elem = t.castRaw<RRefType>()->getElementType();
-    out << "RRef[" << *elem << ']';
+    out << "RRef[" << *elem << "]";
   } else if(auto tup = t.cast<TupleType>()) {
     if (tup->schema()) {
       out << "NamedTuple";
     }
-    out << '(';
+    out << "(";
     for(size_t i = 0; i < tup->elements().size(); ++i) {
       if(i > 0)
         out << ", ";
@@ -160,7 +160,7 @@ std::ostream& operator<<(std::ostream & out, const Type & t) {
         out << *(tup->elements()[i]);
       }
     }
-    out << ')';
+    out << ")";
   } else if (t.kind() == TypeKind::FunctionType) {
     out << "Function";
   } else {
@@ -475,7 +475,7 @@ std::optional<TypePtr> unifyTypeList(
       why_not << "Could not unify type list since element " << i << " of type "
               << elements.at(i)->repr_str()
               << " did not match the types before it ("
-              << ret_type->repr_str() << ')';
+              << ret_type->repr_str() << ")";
       return std::nullopt;
     }
     ret_type = *maybe_unified;
@@ -907,13 +907,13 @@ std::string TupleType::str() const {
     // NOLINTNEXTLINE(bugprone-unchecked-optional-access)
     ss << name()->qualifiedName();
   } else {
-    ss << '(';
+    ss << "(";
     for(size_t i = 0; i < elements().size(); ++i) {
       if(i > 0)
         ss << ", ";
       ss << elements()[i]->str();
     }
-    ss << ')';
+    ss << ")";
   }
   return ss.str();
 }
@@ -1003,8 +1003,8 @@ bool InterfaceType::isSubTypeImpl(
           *why_not << "Method on interface '" << lhs.repr_str()
                    << "' (1) is not compatible with interface '"
                    << rhs.repr_str() << "' (2)\n"
-                   << "  (1) " << *self_schema << '\n'
-                   << "  (2) " << schema << '\n';
+                   << "  (1) " << *self_schema << "\n"
+                   << "  (2) " << schema << "\n";
           return false;
         }
         return false;
@@ -1078,7 +1078,7 @@ SymbolicShape SymbolicShape::merge(const SymbolicShape& other) const {
 }
 
 void SymbolicShape::dump() const {
-  std::cout << *this << '\n';
+  std::cout << *this << "\n";
 }
 
 bool EnumType::isSubtypeOfExt(const Type& rhs, std::ostream* why_not) const {

aten/src/ATen/core/union_type.cpp

@@ -205,9 +205,9 @@ UnionType::UnionType(std::vector<TypePtr> reference, TypeKind kind) : SharedType
     for (const auto i : c10::irange(reference.size())) {
       msg << reference[i]->repr_str();
       if (i > 0) {
-        msg << ',';
+        msg << ",";
       }
-      msg << ' ';
+      msg << " ";
     }
     msg << "} has the single type " << types_[0]->repr_str()
          << ". Use the common supertype instead of creating a Union"