mirror of
https://github.com/pytorch/pytorch.git
synced 2025-11-04 08:00:58 +08:00
Compare commits
1 Commits
try-removi
...
sdpa-bs-ze
| Author | SHA1 | Date | |
|---|---|---|---|
| e3e06d9e4d |
@ -13,4 +13,3 @@ exclude:
|
||||
- "**/benchmarks/**"
|
||||
- "**/test_*.py"
|
||||
- "**/*_test.py"
|
||||
- "tools/**"
|
||||
|
||||
@ -195,16 +195,13 @@ case "$tag" in
|
||||
NINJA_VERSION=1.9.0
|
||||
TRITON=yes
|
||||
;;
|
||||
pytorch-linux-jammy-xpu-n-py3 | pytorch-linux-jammy-xpu-n-py3-inductor-benchmarks)
|
||||
pytorch-linux-jammy-xpu-n-py3)
|
||||
ANACONDA_PYTHON_VERSION=3.10
|
||||
GCC_VERSION=11
|
||||
VISION=yes
|
||||
XPU_VERSION=2025.2
|
||||
NINJA_VERSION=1.9.0
|
||||
TRITON=yes
|
||||
if [[ $tag =~ "benchmarks" ]]; then
|
||||
INDUCTOR_BENCHMARKS=yes
|
||||
fi
|
||||
;;
|
||||
pytorch-linux-jammy-py3-gcc11-inductor-benchmarks)
|
||||
ANACONDA_PYTHON_VERSION=3.10
|
||||
|
||||
@ -3,7 +3,7 @@
|
||||
|
||||
set -eux
|
||||
|
||||
ACL_VERSION=${ACL_VERSION:-"v52.6.0"}
|
||||
ACL_VERSION=${ACL_VERSION:-"v25.02"}
|
||||
ACL_INSTALL_DIR="/acl"
|
||||
|
||||
# Clone ACL
|
||||
|
||||
@ -49,20 +49,12 @@ if [ -n "$ANACONDA_PYTHON_VERSION" ]; then
|
||||
export SYSROOT_DEP="sysroot_linux-64=2.17"
|
||||
fi
|
||||
|
||||
# Install correct Python version
|
||||
# Also ensure sysroot is using a modern GLIBC to match system compilers
|
||||
if [ "$ANACONDA_PYTHON_VERSION" = "3.14" ]; then
|
||||
as_jenkins conda create -n py_$ANACONDA_PYTHON_VERSION -y\
|
||||
python="3.14.0" \
|
||||
${SYSROOT_DEP} \
|
||||
-c conda-forge
|
||||
else
|
||||
# Install correct Python version
|
||||
# Also ensure sysroot is using a modern GLIBC to match system compilers
|
||||
as_jenkins conda create -n py_$ANACONDA_PYTHON_VERSION -y\
|
||||
python="$ANACONDA_PYTHON_VERSION" \
|
||||
${SYSROOT_DEP}
|
||||
fi
|
||||
|
||||
# libstdcxx from conda default channels are too old, we need GLIBCXX_3.4.30
|
||||
# which is provided in libstdcxx 12 and up.
|
||||
conda_install libstdcxx-ng=12.3.0 --update-deps -c conda-forge
|
||||
|
||||
@ -10,7 +10,7 @@ else
|
||||
arch_path='sbsa'
|
||||
fi
|
||||
|
||||
NVSHMEM_VERSION=3.4.5
|
||||
NVSHMEM_VERSION=3.3.24
|
||||
|
||||
function install_cuda {
|
||||
version=$1
|
||||
@ -150,7 +150,7 @@ function install_130 {
|
||||
CUDNN_VERSION=9.13.0.50
|
||||
echo "Installing CUDA 13.0 and cuDNN ${CUDNN_VERSION} and NVSHMEM and NCCL and cuSparseLt-0.7.1"
|
||||
# install CUDA 13.0 in the same container
|
||||
install_cuda 13.0.2 cuda_13.0.2_580.95.05_linux
|
||||
install_cuda 13.0.0 cuda_13.0.0_580.65.06_linux
|
||||
|
||||
# cuDNN license: https://developer.nvidia.com/cudnn/license_agreement
|
||||
install_cudnn 13 $CUDNN_VERSION
|
||||
|
||||
@ -40,7 +40,11 @@ EOF
|
||||
|
||||
# Default url values
|
||||
rocm_baseurl="http://repo.radeon.com/rocm/apt/${ROCM_VERSION}"
|
||||
amdgpu_baseurl="https://repo.radeon.com/amdgpu/${ROCM_VERSION}/ubuntu"
|
||||
|
||||
# Add amdgpu repository
|
||||
UBUNTU_VERSION_NAME=`cat /etc/os-release | grep UBUNTU_CODENAME | awk -F= '{print $2}'`
|
||||
echo "deb [arch=amd64] ${amdgpu_baseurl} ${UBUNTU_VERSION_NAME} main" > /etc/apt/sources.list.d/amdgpu.list
|
||||
|
||||
# Add rocm repository
|
||||
wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add -
|
||||
|
||||
@ -12,8 +12,8 @@ function do_install() {
|
||||
|
||||
rocm_version_nodot=${rocm_version//./}
|
||||
|
||||
# post merge of https://github.com/icl-utk-edu/magma/pull/65
|
||||
MAGMA_VERSION=c0792ae825fb36872784892ea643dd6f3456bc5f
|
||||
# https://github.com/icl-utk-edu/magma/pull/65
|
||||
MAGMA_VERSION=d6e4117bc88e73f06d26c6c2e14f064e8fc3d1ec
|
||||
magma_archive="magma-rocm${rocm_version_nodot}-${MAGMA_VERSION}-1.tar.bz2"
|
||||
|
||||
rocm_dir="/opt/rocm"
|
||||
|
||||
@ -97,7 +97,7 @@ case ${image} in
|
||||
manylinux2_28-builder:xpu)
|
||||
TARGET=xpu_final
|
||||
GPU_IMAGE=amd64/almalinux:8
|
||||
DOCKER_GPU_BUILD_ARG=" --build-arg DEVTOOLSET_VERSION=13"
|
||||
DOCKER_GPU_BUILD_ARG=" --build-arg DEVTOOLSET_VERSION=11"
|
||||
MANY_LINUX_VERSION="2_28"
|
||||
;;
|
||||
*)
|
||||
|
||||
@ -138,12 +138,10 @@ numba==0.60.0 ; python_version == "3.12" and platform_machine != "s390x"
|
||||
#test_binary_ufuncs.py
|
||||
numpy==1.22.4; python_version == "3.10"
|
||||
numpy==1.26.2; python_version == "3.11" or python_version == "3.12"
|
||||
numpy==2.1.2; python_version >= "3.13" and python_version < "3.14"
|
||||
numpy==2.3.4; python_version >= "3.14"
|
||||
numpy==2.1.2; python_version >= "3.13"
|
||||
|
||||
pandas==2.0.3; python_version < "3.13"
|
||||
pandas==2.2.3; python_version >= "3.13" and python_version < "3.14"
|
||||
pandas==2.3.3; python_version >= "3.14"
|
||||
pandas==2.2.3; python_version >= "3.13"
|
||||
|
||||
#onnxruntime
|
||||
#Description: scoring engine for Open Neural Network Exchange (ONNX) models
|
||||
@ -155,8 +153,7 @@ opt-einsum==3.3
|
||||
#Pinned versions: 3.3
|
||||
#test that import: test_linalg.py
|
||||
|
||||
optree==0.13.0 ; python_version < "3.14"
|
||||
optree==0.17.0 ; python_version >= "3.14"
|
||||
optree==0.13.0
|
||||
#Description: A library for tree manipulation
|
||||
#Pinned versions: 0.13.0
|
||||
#test that import: test_vmap.py, test_aotdispatch.py, test_dynamic_shapes.py,
|
||||
@ -255,8 +252,7 @@ scikit-image==0.22.0
|
||||
#test that import:
|
||||
|
||||
scipy==1.10.1 ; python_version <= "3.11"
|
||||
scipy==1.14.1 ; python_version > "3.11" and python_version < "3.14"
|
||||
scipy==1.16.2 ; python_version >= "3.14"
|
||||
scipy==1.14.1 ; python_version >= "3.12"
|
||||
# Pin SciPy because of failing distribution tests (see #60347)
|
||||
#Description: scientific python
|
||||
#Pinned versions: 1.10.1
|
||||
@ -328,8 +324,7 @@ pywavelets==1.7.0 ; python_version >= "3.12"
|
||||
#Pinned versions: 1.4.1
|
||||
#test that import:
|
||||
|
||||
lxml==5.3.0 ; python_version < "3.14"
|
||||
lxml==6.0.2 ; python_version >= "3.14"
|
||||
lxml==5.3.0
|
||||
#Description: This is a requirement of unittest-xml-reporting
|
||||
|
||||
PyGithub==2.3.0
|
||||
@ -339,9 +334,7 @@ sympy==1.13.3
|
||||
#Pinned versions:
|
||||
#test that import:
|
||||
|
||||
onnx==1.19.1 ; python_version < "3.14"
|
||||
# Unpin once Python 3.14 is supported. See onnxruntime issue 26309.
|
||||
onnx==1.18.0 ; python_version == "3.14"
|
||||
onnx==1.19.1
|
||||
#Description: Required by onnx tests, and mypy and test_public_bindings.py when checking torch.onnx._internal
|
||||
#Pinned versions:
|
||||
#test that import:
|
||||
@ -366,7 +359,7 @@ pwlf==2.2.1
|
||||
#test that import: test_sac_estimator.py
|
||||
|
||||
# To build PyTorch itself
|
||||
pyyaml==6.0.3
|
||||
pyyaml==6.0.2
|
||||
pyzstd
|
||||
setuptools==78.1.1
|
||||
packaging==23.1
|
||||
|
||||
@ -54,15 +54,12 @@ ENV OPENSSL_DIR /opt/openssl
|
||||
RUN rm install_openssl.sh
|
||||
|
||||
ARG INDUCTOR_BENCHMARKS
|
||||
ARG ANACONDA_PYTHON_VERSION
|
||||
ENV ANACONDA_PYTHON_VERSION=$ANACONDA_PYTHON_VERSION
|
||||
COPY ./common/install_inductor_benchmark_deps.sh install_inductor_benchmark_deps.sh
|
||||
COPY ./common/common_utils.sh common_utils.sh
|
||||
COPY ci_commit_pins/huggingface-requirements.txt huggingface-requirements.txt
|
||||
COPY ci_commit_pins/timm.txt timm.txt
|
||||
COPY ci_commit_pins/torchbench.txt torchbench.txt
|
||||
RUN if [ -n "${INDUCTOR_BENCHMARKS}" ]; then bash ./install_inductor_benchmark_deps.sh; fi
|
||||
RUN rm install_inductor_benchmark_deps.sh common_utils.sh timm.txt huggingface-requirements.txt torchbench.txt
|
||||
RUN rm install_inductor_benchmark_deps.sh common_utils.sh timm.txt huggingface-requirements.txt
|
||||
|
||||
# Install XPU Dependencies
|
||||
ARG XPU_VERSION
|
||||
|
||||
@ -100,8 +100,6 @@ COPY ./common/common_utils.sh common_utils.sh
|
||||
COPY ci_commit_pins/huggingface-requirements.txt huggingface-requirements.txt
|
||||
COPY ci_commit_pins/timm.txt timm.txt
|
||||
COPY ci_commit_pins/torchbench.txt torchbench.txt
|
||||
# Only build aoti cpp tests when INDUCTOR_BENCHMARKS is set to True
|
||||
ENV BUILD_AOT_INDUCTOR_TEST ${INDUCTOR_BENCHMARKS}
|
||||
RUN if [ -n "${INDUCTOR_BENCHMARKS}" ]; then bash ./install_inductor_benchmark_deps.sh; fi
|
||||
RUN rm install_inductor_benchmark_deps.sh common_utils.sh timm.txt huggingface-requirements.txt torchbench.txt
|
||||
|
||||
|
||||
@ -6,7 +6,7 @@ dependencies = [
|
||||
"GitPython==3.1.45",
|
||||
"docker==7.1.0",
|
||||
"pytest==7.3.2",
|
||||
"uv==0.9.6"
|
||||
"uv==0.9.5"
|
||||
]
|
||||
|
||||
[tool.setuptools]
|
||||
|
||||
@ -1,7 +1,7 @@
|
||||
SHELL=/usr/bin/env bash
|
||||
|
||||
DOCKER_CMD ?= docker
|
||||
DESIRED_ROCM ?= 7.1
|
||||
DESIRED_ROCM ?= 7.0
|
||||
DESIRED_ROCM_SHORT = $(subst .,,$(DESIRED_ROCM))
|
||||
PACKAGE_NAME = magma-rocm
|
||||
# inherit this from underlying docker image, do not pass this env var to docker
|
||||
@ -16,7 +16,6 @@ DOCKER_RUN = set -eou pipefail; ${DOCKER_CMD} run --rm -i \
|
||||
magma-rocm/build_magma.sh
|
||||
|
||||
.PHONY: all
|
||||
all: magma-rocm71
|
||||
all: magma-rocm70
|
||||
all: magma-rocm64
|
||||
|
||||
@ -25,11 +24,6 @@ clean:
|
||||
$(RM) -r magma-*
|
||||
$(RM) -r output
|
||||
|
||||
.PHONY: magma-rocm71
|
||||
magma-rocm71: DESIRED_ROCM := 7.1
|
||||
magma-rocm71:
|
||||
$(DOCKER_RUN)
|
||||
|
||||
.PHONY: magma-rocm70
|
||||
magma-rocm70: DESIRED_ROCM := 7.0
|
||||
magma-rocm70:
|
||||
|
||||
@ -6,8 +6,8 @@ set -eou pipefail
|
||||
# The script expects DESIRED_CUDA and PACKAGE_NAME to be set
|
||||
ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
|
||||
|
||||
# post merge of https://github.com/icl-utk-edu/magma/pull/65
|
||||
MAGMA_VERSION=c0792ae825fb36872784892ea643dd6f3456bc5f
|
||||
# https://github.com/icl-utk-edu/magma/pull/65
|
||||
MAGMA_VERSION=d6e4117bc88e73f06d26c6c2e14f064e8fc3d1ec
|
||||
|
||||
# Folders for the build
|
||||
PACKAGE_FILES=${ROOT_DIR}/magma-rocm/package_files # metadata
|
||||
@ -20,7 +20,7 @@ mkdir -p ${PACKAGE_DIR} ${PACKAGE_OUTPUT}/linux-64 ${PACKAGE_BUILD} ${PACKAGE_RE
|
||||
|
||||
# Fetch magma sources and verify checksum
|
||||
pushd ${PACKAGE_DIR}
|
||||
git clone https://github.com/icl-utk-edu/magma
|
||||
git clone https://github.com/jeffdaily/magma
|
||||
pushd magma
|
||||
git checkout ${MAGMA_VERSION}
|
||||
popd
|
||||
|
||||
@ -426,7 +426,7 @@ fi
|
||||
if [[ "$BUILD_ENVIRONMENT" != *libtorch* && "$BUILD_ENVIRONMENT" != *bazel* ]]; then
|
||||
# export test times so that potential sharded tests that'll branch off this build will use consistent data
|
||||
# don't do this for libtorch as libtorch is C++ only and thus won't have python tests run on its build
|
||||
PYTHONPATH=. python tools/stats/export_test_times.py
|
||||
python tools/stats/export_test_times.py
|
||||
fi
|
||||
# don't do this for bazel or s390x or riscv64 as they don't use sccache
|
||||
if [[ "$BUILD_ENVIRONMENT" != *s390x* && "$BUILD_ENVIRONMENT" != *riscv64* && "$BUILD_ENVIRONMENT" != *-bazel-* ]]; then
|
||||
|
||||
@ -460,18 +460,28 @@ test_inductor_shard() {
|
||||
--verbose
|
||||
}
|
||||
|
||||
test_inductor_aoti_cpp() {
|
||||
test_inductor_aoti() {
|
||||
# docker build uses bdist_wheel which does not work with test_aot_inductor
|
||||
# TODO: need a faster way to build
|
||||
if [[ "$BUILD_ENVIRONMENT" == *rocm* ]]; then
|
||||
# We need to hipify before building again
|
||||
python3 tools/amd_build/build_amd.py
|
||||
fi
|
||||
if [[ "$BUILD_ENVIRONMENT" == *sm86* ]]; then
|
||||
BUILD_COMMAND=(TORCH_CUDA_ARCH_LIST=8.6 USE_FLASH_ATTENTION=OFF python -m pip install --no-build-isolation -v -e .)
|
||||
# TODO: Replace me completely, as one should not use conda libstdc++, nor need special path to TORCH_LIB
|
||||
TEST_ENVS=(CPP_TESTS_DIR="${BUILD_BIN_DIR}" LD_LIBRARY_PATH="/opt/conda/envs/py_3.10/lib:${TORCH_LIB_DIR}:${LD_LIBRARY_PATH}")
|
||||
else
|
||||
BUILD_COMMAND=(python -m pip install --no-build-isolation -v -e .)
|
||||
TEST_ENVS=(CPP_TESTS_DIR="${BUILD_BIN_DIR}" LD_LIBRARY_PATH="${TORCH_LIB_DIR}")
|
||||
fi
|
||||
|
||||
# aoti cmake custom command requires `torch` to be installed
|
||||
# initialize the cmake build cache and install torch
|
||||
/usr/bin/env "${BUILD_COMMAND[@]}"
|
||||
# rebuild with the build cache with `BUILD_AOT_INDUCTOR_TEST` enabled
|
||||
/usr/bin/env CMAKE_FRESH=1 BUILD_AOT_INDUCTOR_TEST=1 "${BUILD_COMMAND[@]}"
|
||||
|
||||
/usr/bin/env "${TEST_ENVS[@]}" python test/run_test.py --cpp --verbose -i cpp/test_aoti_abi_check cpp/test_aoti_inference cpp/test_vec_half_AVX2 -dist=loadfile
|
||||
}
|
||||
|
||||
@ -572,8 +582,6 @@ fi
|
||||
|
||||
if [[ "${TEST_CONFIG}" == *cpu* ]]; then
|
||||
DYNAMO_BENCHMARK_FLAGS+=(--device cpu)
|
||||
elif [[ "${TEST_CONFIG}" == *xpu* ]]; then
|
||||
DYNAMO_BENCHMARK_FLAGS+=(--device xpu)
|
||||
else
|
||||
DYNAMO_BENCHMARK_FLAGS+=(--device cuda)
|
||||
fi
|
||||
@ -667,8 +675,6 @@ test_perf_for_dashboard() {
|
||||
device=cuda_b200
|
||||
elif [[ "${TEST_CONFIG}" == *rocm* ]]; then
|
||||
device=rocm
|
||||
elif [[ "${TEST_CONFIG}" == *xpu* ]]; then
|
||||
device=xpu
|
||||
fi
|
||||
|
||||
for mode in "${modes[@]}"; do
|
||||
@ -1761,7 +1767,7 @@ elif [[ "${TEST_CONFIG}" == *torchbench* ]]; then
|
||||
else
|
||||
# Do this after checkout_install_torchbench to ensure we clobber any
|
||||
# nightlies that torchbench may pull in
|
||||
if [[ "${TEST_CONFIG}" != *cpu* && "${TEST_CONFIG}" != *xpu* ]]; then
|
||||
if [[ "${TEST_CONFIG}" != *cpu* ]]; then
|
||||
install_torchrec_and_fbgemm
|
||||
fi
|
||||
PYTHONPATH=/torchbench test_dynamo_benchmark torchbench "$id"
|
||||
@ -1770,7 +1776,7 @@ elif [[ "${TEST_CONFIG}" == *inductor_cpp_wrapper* ]]; then
|
||||
install_torchvision
|
||||
PYTHONPATH=/torchbench test_inductor_cpp_wrapper_shard "$SHARD_NUMBER"
|
||||
if [[ "$SHARD_NUMBER" -eq "1" ]]; then
|
||||
test_inductor_aoti_cpp
|
||||
test_inductor_aoti
|
||||
fi
|
||||
elif [[ "${TEST_CONFIG}" == *inductor* ]]; then
|
||||
install_torchvision
|
||||
|
||||
@ -7,9 +7,12 @@ if "%DESIRED_PYTHON%" == "3.13t" (
|
||||
set "PYTHON_INSTALLER_URL=https://www.python.org/ftp/python/3.13.0/python-3.13.0-amd64.exe"
|
||||
set ADDITIONAL_OPTIONS="Include_freethreaded=1"
|
||||
set PYTHON_EXEC="python3.13t"
|
||||
) else if "%DESIRED_PYTHON%"=="3.14" (
|
||||
echo Python version is set to 3.14 or 3.14t
|
||||
set "PYTHON_INSTALLER_URL=https://www.python.org/ftp/python/3.14.0/python-3.14.0rc1-amd64.exe"
|
||||
) else if "%DESIRED_PYTHON%"=="3.14t" (
|
||||
echo Python version is set to 3.14 or 3.14t
|
||||
set "PYTHON_INSTALLER_URL=https://www.python.org/ftp/python/3.14.0/python-3.14.0-amd64.exe"
|
||||
set "PYTHON_INSTALLER_URL=https://www.python.org/ftp/python/3.14.0/python-3.14.0rc1-amd64.exe"
|
||||
set ADDITIONAL_OPTIONS="Include_freethreaded=1"
|
||||
set PYTHON_EXEC="python3.14t"
|
||||
) else (
|
||||
|
||||
@ -60,11 +60,9 @@ performance-*,
|
||||
readability-container-size-empty,
|
||||
readability-delete-null-pointer,
|
||||
readability-duplicate-include,
|
||||
readability-named-parameter,
|
||||
readability-misplaced-array-index,
|
||||
readability-redundant*,
|
||||
readability-simplify-subscript-expr,
|
||||
readability-static-definition-in-anonymous-namespace
|
||||
readability-string-compare,
|
||||
-readability-redundant-access-specifiers,
|
||||
-readability-redundant-control-flow,
|
||||
|
||||
@ -1,319 +0,0 @@
|
||||
---
|
||||
name: add-uint-support
|
||||
description: Add unsigned integer (uint) type support to PyTorch operators by updating AT_DISPATCH macros. Use when adding support for uint16, uint32, uint64 types to operators, kernels, or when user mentions enabling unsigned types, barebones unsigned types, or uint support.
|
||||
---
|
||||
|
||||
# Add Unsigned Integer (uint) Support to Operators
|
||||
|
||||
This skill helps add support for unsigned integer types (uint16, uint32, uint64) to PyTorch operators by updating their AT_DISPATCH macros.
|
||||
|
||||
## When to use this skill
|
||||
|
||||
Use this skill when:
|
||||
- Adding uint16, uint32, or uint64 support to an operator
|
||||
- User mentions "unsigned types", "uint support", "barebones unsigned types"
|
||||
- Enabling support for kUInt16, kUInt32, kUInt64 in kernels
|
||||
- Working with operator implementations that need expanded type coverage
|
||||
|
||||
## Quick reference
|
||||
|
||||
**Add unsigned types to existing dispatch:**
|
||||
```cpp
|
||||
// Before
|
||||
AT_DISPATCH_V2(dtype, "op", AT_WRAP([&]() {
|
||||
kernel<scalar_t>();
|
||||
}), AT_EXPAND(AT_ALL_TYPES));
|
||||
|
||||
// After (method 1: add unsigned types explicitly)
|
||||
AT_DISPATCH_V2(dtype, "op", AT_WRAP([&]() {
|
||||
kernel<scalar_t>();
|
||||
}), AT_EXPAND(AT_ALL_TYPES), AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES));
|
||||
|
||||
// After (method 2: use V2 integral types if AT_INTEGRAL_TYPES present)
|
||||
AT_DISPATCH_V2(dtype, "op", AT_WRAP([&]() {
|
||||
kernel<scalar_t>();
|
||||
}), AT_EXPAND(AT_INTEGRAL_TYPES_V2), AT_EXPAND(AT_FLOATING_TYPES));
|
||||
```
|
||||
|
||||
## Type group reference
|
||||
|
||||
**Unsigned type groups:**
|
||||
- `AT_BAREBONES_UNSIGNED_TYPES`: kUInt16, kUInt32, kUInt64
|
||||
- `AT_INTEGRAL_TYPES_V2`: AT_INTEGRAL_TYPES + AT_BAREBONES_UNSIGNED_TYPES
|
||||
|
||||
**Relationship:**
|
||||
```cpp
|
||||
AT_INTEGRAL_TYPES // kByte, kChar, kInt, kLong, kShort
|
||||
AT_BAREBONES_UNSIGNED_TYPES // kUInt16, kUInt32, kUInt64
|
||||
AT_INTEGRAL_TYPES_V2 // INTEGRAL_TYPES + BAREBONES_UNSIGNED_TYPES
|
||||
```
|
||||
|
||||
## Instructions
|
||||
|
||||
### Step 1: Determine if conversion to V2 is needed
|
||||
|
||||
Check if the file uses AT_DISPATCH_V2:
|
||||
|
||||
**If using old AT_DISPATCH:**
|
||||
- First convert to AT_DISPATCH_V2 using the at-dispatch-v2 skill
|
||||
- Then proceed with adding uint support
|
||||
|
||||
**If already using AT_DISPATCH_V2:**
|
||||
- Proceed directly to Step 2
|
||||
|
||||
### Step 2: Analyze the current dispatch macro
|
||||
|
||||
Identify what type groups are currently in use:
|
||||
|
||||
```cpp
|
||||
AT_DISPATCH_V2(dtype, "op", AT_WRAP([&]() {
|
||||
// body
|
||||
}), AT_EXPAND(AT_ALL_TYPES), kHalf, kBFloat16);
|
||||
^^^^^^^^^^^^^^^^^^^^^^^^^
|
||||
Current type coverage
|
||||
```
|
||||
|
||||
Common patterns:
|
||||
- `AT_EXPAND(AT_ALL_TYPES)` → includes AT_INTEGRAL_TYPES + AT_FLOATING_TYPES
|
||||
- `AT_EXPAND(AT_INTEGRAL_TYPES)` → signed integers only
|
||||
- `AT_EXPAND(AT_FLOATING_TYPES)` → floating point types
|
||||
|
||||
### Step 3: Choose the uint addition method
|
||||
|
||||
Two approaches:
|
||||
|
||||
**Method 1: Add AT_BAREBONES_UNSIGNED_TYPES explicitly**
|
||||
- Use when: You want to be explicit about adding uint support
|
||||
- Add `AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES)` to the type list
|
||||
|
||||
**Method 2: Substitute AT_INTEGRAL_TYPES with AT_INTEGRAL_TYPES_V2**
|
||||
- Use when: The dispatch already uses `AT_EXPAND(AT_INTEGRAL_TYPES)`
|
||||
- More concise: replaces one type group with its superset
|
||||
- Only applicable if AT_INTEGRAL_TYPES is present
|
||||
|
||||
### Step 4: Apply the transformation
|
||||
|
||||
**Method 1 example:**
|
||||
```cpp
|
||||
// Before
|
||||
AT_DISPATCH_V2(
|
||||
dtype,
|
||||
"min_values_cuda",
|
||||
AT_WRAP([&]() {
|
||||
kernel_impl<scalar_t>(iter);
|
||||
}),
|
||||
AT_EXPAND(AT_ALL_TYPES),
|
||||
kBFloat16, kHalf, kBool
|
||||
);
|
||||
|
||||
// After (add unsigned types)
|
||||
AT_DISPATCH_V2(
|
||||
dtype,
|
||||
"min_values_cuda",
|
||||
AT_WRAP([&]() {
|
||||
kernel_impl<scalar_t>(iter);
|
||||
}),
|
||||
AT_EXPAND(AT_ALL_TYPES),
|
||||
AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES),
|
||||
kBFloat16, kHalf, kBool
|
||||
);
|
||||
```
|
||||
|
||||
**Method 2 example:**
|
||||
```cpp
|
||||
// Before
|
||||
AT_DISPATCH_V2(
|
||||
dtype,
|
||||
"integral_op",
|
||||
AT_WRAP([&]() {
|
||||
kernel<scalar_t>();
|
||||
}),
|
||||
AT_EXPAND(AT_INTEGRAL_TYPES)
|
||||
);
|
||||
|
||||
// After (substitute with V2)
|
||||
AT_DISPATCH_V2(
|
||||
dtype,
|
||||
"integral_op",
|
||||
AT_WRAP([&]() {
|
||||
kernel<scalar_t>();
|
||||
}),
|
||||
AT_EXPAND(AT_INTEGRAL_TYPES_V2)
|
||||
);
|
||||
```
|
||||
|
||||
### Step 5: Handle AT_ALL_TYPES vs individual type groups
|
||||
|
||||
If the dispatch uses `AT_EXPAND(AT_ALL_TYPES)`:
|
||||
- `AT_ALL_TYPES` = `AT_INTEGRAL_TYPES` + `AT_FLOATING_TYPES`
|
||||
- To add uint: add `AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES)` to the list
|
||||
|
||||
If the dispatch separately lists INTEGRAL and FLOATING:
|
||||
```cpp
|
||||
// Before
|
||||
AT_EXPAND(AT_INTEGRAL_TYPES), AT_EXPAND(AT_FLOATING_TYPES)
|
||||
|
||||
// After (Method 2 preferred)
|
||||
AT_EXPAND(AT_INTEGRAL_TYPES_V2), AT_EXPAND(AT_FLOATING_TYPES)
|
||||
```
|
||||
|
||||
### Step 6: Verify all dispatch sites
|
||||
|
||||
Check the file for ALL dispatch macros that need uint support:
|
||||
- Some operators have multiple dispatch sites (CPU, CUDA, different functions)
|
||||
- Apply the transformation consistently across all sites
|
||||
- Ensure each gets the same type coverage updates
|
||||
|
||||
### Step 7: Validate the changes
|
||||
|
||||
Check that:
|
||||
- [ ] AT_DISPATCH_V2 format is used (not old AT_DISPATCH)
|
||||
- [ ] Unsigned types are added via one of the two methods
|
||||
- [ ] All relevant dispatch sites in the file are updated
|
||||
- [ ] Type groups use `AT_EXPAND()`
|
||||
- [ ] Arguments are properly formatted and comma-separated
|
||||
|
||||
## Common patterns
|
||||
|
||||
### Pattern 1: AT_ALL_TYPES + extras
|
||||
|
||||
```cpp
|
||||
// Before
|
||||
AT_DISPATCH_V2(dtype, "op", AT_WRAP([&]() {
|
||||
kernel<scalar_t>();
|
||||
}), AT_EXPAND(AT_ALL_TYPES), kHalf, kBFloat16);
|
||||
|
||||
// After
|
||||
AT_DISPATCH_V2(dtype, "op", AT_WRAP([&]() {
|
||||
kernel<scalar_t>();
|
||||
}), AT_EXPAND(AT_ALL_TYPES), AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES), kHalf, kBFloat16);
|
||||
```
|
||||
|
||||
### Pattern 2: Separate INTEGRAL + FLOATING
|
||||
|
||||
```cpp
|
||||
// Before
|
||||
AT_DISPATCH_V2(dtype, "op", AT_WRAP([&]() {
|
||||
kernel<scalar_t>();
|
||||
}), AT_EXPAND(AT_INTEGRAL_TYPES), AT_EXPAND(AT_FLOATING_TYPES));
|
||||
|
||||
// After
|
||||
AT_DISPATCH_V2(dtype, "op", AT_WRAP([&]() {
|
||||
kernel<scalar_t>();
|
||||
}), AT_EXPAND(AT_INTEGRAL_TYPES_V2), AT_EXPAND(AT_FLOATING_TYPES));
|
||||
```
|
||||
|
||||
### Pattern 3: Old dispatch needs conversion first
|
||||
|
||||
```cpp
|
||||
// Before (needs v2 conversion first)
|
||||
AT_DISPATCH_ALL_TYPES_AND2(kHalf, kBFloat16, dtype, "op", [&]() {
|
||||
kernel<scalar_t>();
|
||||
});
|
||||
|
||||
// After v2 conversion
|
||||
AT_DISPATCH_V2(dtype, "op", AT_WRAP([&]() {
|
||||
kernel<scalar_t>();
|
||||
}), AT_EXPAND(AT_ALL_TYPES), kHalf, kBFloat16);
|
||||
|
||||
// After adding uint support
|
||||
AT_DISPATCH_V2(dtype, "op", AT_WRAP([&]() {
|
||||
kernel<scalar_t>();
|
||||
}), AT_EXPAND(AT_ALL_TYPES), AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES), kHalf, kBFloat16);
|
||||
```
|
||||
|
||||
## Multiple dispatch sites example
|
||||
|
||||
For a file with multiple functions:
|
||||
|
||||
```cpp
|
||||
void min_values_kernel_cuda(TensorIterator& iter) {
|
||||
AT_DISPATCH_V2(iter.dtype(), "min_values_cuda", AT_WRAP([&]() {
|
||||
impl<scalar_t>(iter);
|
||||
}), AT_EXPAND(AT_ALL_TYPES), AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES), kBFloat16, kHalf);
|
||||
// ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
||||
// Added uint support
|
||||
}
|
||||
|
||||
void min_launch_kernel(TensorIterator &iter) {
|
||||
AT_DISPATCH_V2(iter.input_dtype(), "min_cuda", AT_WRAP([&]() {
|
||||
gpu_reduce_kernel<scalar_t>(iter);
|
||||
}), AT_EXPAND(AT_ALL_TYPES), AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES), kBFloat16, kHalf);
|
||||
// ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
||||
// Added uint support here too
|
||||
}
|
||||
```
|
||||
|
||||
## Decision tree
|
||||
|
||||
Use this decision tree to determine the approach:
|
||||
|
||||
```
|
||||
Is the file using AT_DISPATCH_V2?
|
||||
├─ No → Use at-dispatch-v2 skill first, then continue
|
||||
└─ Yes
|
||||
└─ Does it use AT_EXPAND(AT_INTEGRAL_TYPES)?
|
||||
├─ Yes → Replace with AT_EXPAND(AT_INTEGRAL_TYPES_V2)
|
||||
└─ No → Add AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES) to type list
|
||||
```
|
||||
|
||||
## Edge cases
|
||||
|
||||
### Case 1: Dispatch with only floating types
|
||||
|
||||
If the operator only supports floating point types, don't add uint support:
|
||||
|
||||
```cpp
|
||||
// Leave as-is - floating point only operator
|
||||
AT_DISPATCH_V2(dtype, "float_op", AT_WRAP([&]() {
|
||||
kernel<scalar_t>();
|
||||
}), AT_EXPAND(AT_FLOATING_TYPES), kHalf);
|
||||
```
|
||||
|
||||
### Case 2: Complex types present
|
||||
|
||||
Unsigned types work alongside complex types:
|
||||
|
||||
```cpp
|
||||
AT_DISPATCH_V2(dtype, "op", AT_WRAP([&]() {
|
||||
kernel<scalar_t>();
|
||||
}), AT_EXPAND(AT_ALL_TYPES),
|
||||
AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES),
|
||||
AT_EXPAND(AT_COMPLEX_TYPES),
|
||||
kHalf, kBFloat16);
|
||||
```
|
||||
|
||||
### Case 3: Already has uint support
|
||||
|
||||
Check if uint types are already present:
|
||||
- If `AT_INTEGRAL_TYPES_V2` is used → already has uint support
|
||||
- If `AT_BAREBONES_UNSIGNED_TYPES` is already in list → already has uint support
|
||||
- Skip the file if uint support is already present
|
||||
|
||||
## Workflow
|
||||
|
||||
When asked to add uint support:
|
||||
|
||||
1. Read the target file
|
||||
2. Check if using AT_DISPATCH_V2:
|
||||
- If not → use at-dispatch-v2 skill first
|
||||
3. Identify all dispatch macro sites
|
||||
4. For each dispatch:
|
||||
- Analyze current type groups
|
||||
- Choose method (add BAREBONES_UNSIGNED or upgrade to V2)
|
||||
- Apply transformation with Edit tool
|
||||
5. Show the user the changes
|
||||
6. Explain what was modified
|
||||
|
||||
## Important notes
|
||||
|
||||
- Always check if v2 conversion is needed first
|
||||
- Apply changes consistently across all dispatch sites in the file
|
||||
- Method 2 (AT_INTEGRAL_TYPES_V2) is cleaner when applicable
|
||||
- Method 1 (explicit AT_BAREBONES_UNSIGNED_TYPES) is more explicit
|
||||
- Unsigned types are: kUInt16, kUInt32, kUInt64 (not kByte which is uint8)
|
||||
- Some operators may not semantically support unsigned types - use judgment
|
||||
|
||||
## Testing
|
||||
|
||||
After adding uint support, the operator should accept uint16, uint32, and uint64 tensors. The user is responsible for functional testing.
|
||||
@ -1,305 +0,0 @@
|
||||
---
|
||||
name: at-dispatch-v2
|
||||
description: Convert PyTorch AT_DISPATCH macros to AT_DISPATCH_V2 format in ATen C++ code. Use when porting AT_DISPATCH_ALL_TYPES_AND*, AT_DISPATCH_FLOATING_TYPES*, or other dispatch macros to the new v2 API. For ATen kernel files, CUDA kernels, and native operator implementations.
|
||||
---
|
||||
|
||||
# AT_DISPATCH to AT_DISPATCH_V2 Converter
|
||||
|
||||
This skill helps convert PyTorch's legacy AT_DISPATCH macros to the new AT_DISPATCH_V2 format, as defined in `aten/src/ATen/Dispatch_v2.h`.
|
||||
|
||||
## When to use this skill
|
||||
|
||||
Use this skill when:
|
||||
- Converting AT_DISPATCH_* macros to AT_DISPATCH_V2
|
||||
- Porting ATen kernels to use the new dispatch API
|
||||
- Working with files in `aten/src/ATen/native/` that use dispatch macros
|
||||
- User mentions "AT_DISPATCH", "dispatch v2", "Dispatch_v2.h", or macro conversion
|
||||
|
||||
## Quick reference
|
||||
|
||||
**Old format:**
|
||||
```cpp
|
||||
AT_DISPATCH_ALL_TYPES_AND3(kBFloat16, kHalf, kBool, dtype, "kernel_name", [&]() {
|
||||
// lambda body
|
||||
});
|
||||
```
|
||||
|
||||
**New format:**
|
||||
```cpp
|
||||
AT_DISPATCH_V2(dtype, "kernel_name", AT_WRAP([&]() {
|
||||
// lambda body
|
||||
}), AT_EXPAND(AT_ALL_TYPES), kBFloat16, kHalf, kBool);
|
||||
```
|
||||
|
||||
## Key transformations
|
||||
|
||||
1. **Reorder arguments**: `scalar_type` and `name` come first, then lambda, then types
|
||||
2. **Wrap the lambda**: Use `AT_WRAP(lambda)` to handle internal commas
|
||||
3. **Expand type groups**: Use `AT_EXPAND(AT_ALL_TYPES)` instead of implicit expansion
|
||||
4. **List individual types**: Add extra types (kHalf, kBFloat16, etc.) after expanded groups
|
||||
5. **Add include**: `#include <ATen/Dispatch_v2.h>` near other Dispatch includes
|
||||
|
||||
## Instructions
|
||||
|
||||
### Step 1: Add the Dispatch_v2.h include
|
||||
|
||||
Add the v2 header near the existing `#include <ATen/Dispatch.h>`:
|
||||
|
||||
```cpp
|
||||
#include <ATen/Dispatch.h>
|
||||
#include <ATen/Dispatch_v2.h>
|
||||
```
|
||||
|
||||
Keep the old Dispatch.h include for now (other code may still need it).
|
||||
|
||||
### Step 2: Identify the old dispatch pattern
|
||||
|
||||
Common patterns to convert:
|
||||
|
||||
- `AT_DISPATCH_ALL_TYPES_AND{2,3,4}(type1, type2, ..., scalar_type, name, lambda)`
|
||||
- `AT_DISPATCH_FLOATING_TYPES_AND{2,3}(type1, type2, ..., scalar_type, name, lambda)`
|
||||
- `AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND{2,3}(type1, ..., scalar_type, name, lambda)`
|
||||
- `AT_DISPATCH_FLOATING_AND_COMPLEX_TYPES_AND{2,3}(type1, ..., scalar_type, name, lambda)`
|
||||
|
||||
### Step 3: Map the old macro to type groups
|
||||
|
||||
Identify which type group macro corresponds to the base types:
|
||||
|
||||
| Old macro base | AT_DISPATCH_V2 type group |
|
||||
|----------------|---------------------------|
|
||||
| `ALL_TYPES` | `AT_EXPAND(AT_ALL_TYPES)` |
|
||||
| `FLOATING_TYPES` | `AT_EXPAND(AT_FLOATING_TYPES)` |
|
||||
| `INTEGRAL_TYPES` | `AT_EXPAND(AT_INTEGRAL_TYPES)` |
|
||||
| `COMPLEX_TYPES` | `AT_EXPAND(AT_COMPLEX_TYPES)` |
|
||||
| `ALL_TYPES_AND_COMPLEX` | `AT_EXPAND(AT_ALL_TYPES_AND_COMPLEX)` |
|
||||
|
||||
For combined patterns, use multiple `AT_EXPAND()` entries:
|
||||
```cpp
|
||||
// Old: AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND2(...)
|
||||
// New: AT_EXPAND(AT_ALL_TYPES), AT_EXPAND(AT_COMPLEX_TYPES), type1, type2
|
||||
```
|
||||
|
||||
### Step 4: Extract the individual types
|
||||
|
||||
From `AT_DISPATCH_*_AND2(type1, type2, ...)` or `AT_DISPATCH_*_AND3(type1, type2, type3, ...)`, extract the individual types (type1, type2, etc.).
|
||||
|
||||
These become the trailing arguments after the type group:
|
||||
```cpp
|
||||
AT_DISPATCH_V2(..., AT_EXPAND(AT_ALL_TYPES), kBFloat16, kHalf, kBool)
|
||||
^^^^^^^^^^^^^^^^^^^^^^^^
|
||||
Individual types from AND3
|
||||
```
|
||||
|
||||
### Step 5: Transform to AT_DISPATCH_V2
|
||||
|
||||
Apply the transformation:
|
||||
|
||||
**Pattern:**
|
||||
```cpp
|
||||
AT_DISPATCH_V2(
|
||||
scalar_type, // 1st: The dtype expression
|
||||
"name", // 2nd: The debug string
|
||||
AT_WRAP(lambda), // 3rd: The lambda wrapped in AT_WRAP
|
||||
type_groups, // 4th+: Type groups with AT_EXPAND()
|
||||
individual_types // Last: Individual types
|
||||
)
|
||||
```
|
||||
|
||||
**Example transformation:**
|
||||
```cpp
|
||||
// BEFORE
|
||||
AT_DISPATCH_ALL_TYPES_AND3(
|
||||
kBFloat16, kHalf, kBool,
|
||||
iter.dtype(),
|
||||
"min_values_cuda",
|
||||
[&]() {
|
||||
min_values_kernel_cuda_impl<scalar_t>(iter);
|
||||
}
|
||||
);
|
||||
|
||||
// AFTER
|
||||
AT_DISPATCH_V2(
|
||||
iter.dtype(),
|
||||
"min_values_cuda",
|
||||
AT_WRAP([&]() {
|
||||
min_values_kernel_cuda_impl<scalar_t>(iter);
|
||||
}),
|
||||
AT_EXPAND(AT_ALL_TYPES),
|
||||
kBFloat16, kHalf, kBool
|
||||
);
|
||||
```
|
||||
|
||||
### Step 6: Handle multi-line lambdas
|
||||
|
||||
For lambdas with internal commas or complex expressions, AT_WRAP is essential:
|
||||
|
||||
```cpp
|
||||
AT_DISPATCH_V2(
|
||||
dtype,
|
||||
"complex_kernel",
|
||||
AT_WRAP([&]() {
|
||||
gpu_reduce_kernel<scalar_t, scalar_t>(
|
||||
iter,
|
||||
MinOps<scalar_t>{},
|
||||
thrust::pair<scalar_t, int64_t>(upper_bound(), 0) // Commas inside!
|
||||
);
|
||||
}),
|
||||
AT_EXPAND(AT_ALL_TYPES)
|
||||
);
|
||||
```
|
||||
|
||||
### Step 7: Verify the conversion
|
||||
|
||||
Check that:
|
||||
- [ ] `AT_WRAP()` wraps the entire lambda
|
||||
- [ ] Type groups use `AT_EXPAND()`
|
||||
- [ ] Individual types don't have `AT_EXPAND()` (just `kBFloat16`, not `AT_EXPAND(kBFloat16)`)
|
||||
- [ ] Argument order is: scalar_type, name, lambda, types
|
||||
- [ ] Include added: `#include <ATen/Dispatch_v2.h>`
|
||||
|
||||
## Type group reference
|
||||
|
||||
Available type group macros (use with `AT_EXPAND()`):
|
||||
|
||||
```cpp
|
||||
AT_INTEGRAL_TYPES // kByte, kChar, kInt, kLong, kShort
|
||||
AT_FLOATING_TYPES // kDouble, kFloat
|
||||
AT_COMPLEX_TYPES // kComplexDouble, kComplexFloat
|
||||
AT_QINT_TYPES // kQInt8, kQUInt8, kQInt32
|
||||
AT_ALL_TYPES // INTEGRAL_TYPES + FLOATING_TYPES
|
||||
AT_ALL_TYPES_AND_COMPLEX // ALL_TYPES + COMPLEX_TYPES
|
||||
AT_INTEGRAL_TYPES_V2 // INTEGRAL_TYPES + unsigned types
|
||||
AT_BAREBONES_UNSIGNED_TYPES // kUInt16, kUInt32, kUInt64
|
||||
AT_FLOAT8_TYPES // Float8 variants
|
||||
```
|
||||
|
||||
## Common patterns
|
||||
|
||||
### Pattern: AT_DISPATCH_ALL_TYPES_AND2
|
||||
|
||||
```cpp
|
||||
// Before
|
||||
AT_DISPATCH_ALL_TYPES_AND2(kHalf, kBFloat16, dtype, "op", [&]() {
|
||||
kernel<scalar_t>(data);
|
||||
});
|
||||
|
||||
// After
|
||||
AT_DISPATCH_V2(dtype, "op", AT_WRAP([&]() {
|
||||
kernel<scalar_t>(data);
|
||||
}), AT_EXPAND(AT_ALL_TYPES), kHalf, kBFloat16);
|
||||
```
|
||||
|
||||
### Pattern: AT_DISPATCH_FLOATING_TYPES_AND3
|
||||
|
||||
```cpp
|
||||
// Before
|
||||
AT_DISPATCH_FLOATING_TYPES_AND3(kHalf, kBFloat16, kFloat8_e4m3fn,
|
||||
tensor.scalar_type(), "float_op", [&] {
|
||||
process<scalar_t>(tensor);
|
||||
});
|
||||
|
||||
// After
|
||||
AT_DISPATCH_V2(tensor.scalar_type(), "float_op", AT_WRAP([&] {
|
||||
process<scalar_t>(tensor);
|
||||
}), AT_EXPAND(AT_FLOATING_TYPES), kHalf, kBFloat16, kFloat8_e4m3fn);
|
||||
```
|
||||
|
||||
### Pattern: AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND2
|
||||
|
||||
```cpp
|
||||
// Before
|
||||
AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND2(
|
||||
kComplexHalf, kHalf,
|
||||
self.scalar_type(),
|
||||
"complex_op",
|
||||
[&] {
|
||||
result = compute<scalar_t>(self);
|
||||
}
|
||||
);
|
||||
|
||||
// After
|
||||
AT_DISPATCH_V2(
|
||||
self.scalar_type(),
|
||||
"complex_op",
|
||||
AT_WRAP([&] {
|
||||
result = compute<scalar_t>(self);
|
||||
}),
|
||||
AT_EXPAND(AT_ALL_TYPES),
|
||||
AT_EXPAND(AT_COMPLEX_TYPES),
|
||||
kComplexHalf,
|
||||
kHalf
|
||||
);
|
||||
```
|
||||
|
||||
## Edge cases
|
||||
|
||||
### Case 1: No extra types (rare)
|
||||
|
||||
```cpp
|
||||
// Before
|
||||
AT_DISPATCH_ALL_TYPES(dtype, "op", [&]() { kernel<scalar_t>(); });
|
||||
|
||||
// After
|
||||
AT_DISPATCH_V2(dtype, "op", AT_WRAP([&]() {
|
||||
kernel<scalar_t>();
|
||||
}), AT_EXPAND(AT_ALL_TYPES));
|
||||
```
|
||||
|
||||
### Case 2: Many individual types (AND4, AND5, etc.)
|
||||
|
||||
```cpp
|
||||
// Before
|
||||
AT_DISPATCH_FLOATING_TYPES_AND4(kHalf, kBFloat16, kFloat8_e4m3fn, kFloat8_e5m2,
|
||||
dtype, "float8_op", [&]() { kernel<scalar_t>(); });
|
||||
|
||||
// After
|
||||
AT_DISPATCH_V2(dtype, "float8_op", AT_WRAP([&]() {
|
||||
kernel<scalar_t>();
|
||||
}), AT_EXPAND(AT_FLOATING_TYPES), kHalf, kBFloat16, kFloat8_e4m3fn, kFloat8_e5m2);
|
||||
```
|
||||
|
||||
### Case 3: Lambda with no captures
|
||||
|
||||
```cpp
|
||||
// Before
|
||||
AT_DISPATCH_ALL_TYPES_AND2(kHalf, kBool, dtype, "op", []() {
|
||||
static_kernel<scalar_t>();
|
||||
});
|
||||
|
||||
// After
|
||||
AT_DISPATCH_V2(dtype, "op", AT_WRAP([]() {
|
||||
static_kernel<scalar_t>();
|
||||
}), AT_EXPAND(AT_ALL_TYPES), kHalf, kBool);
|
||||
```
|
||||
|
||||
## Benefits of AT_DISPATCH_V2
|
||||
|
||||
1. **No arity in macro name**: Don't need different macros for AND2, AND3, AND4
|
||||
2. **Composable type sets**: Mix and match type groups with `AT_EXPAND()`
|
||||
3. **Extensible**: Easy to add more types without hitting macro limits
|
||||
4. **Clearer**: Type groups are explicit, not implicit in macro name
|
||||
|
||||
## Important notes
|
||||
|
||||
- Keep `#include <ATen/Dispatch.h>` - other code may need it
|
||||
- The `AT_WRAP()` is mandatory - prevents comma parsing issues in the lambda
|
||||
- Type groups need `AT_EXPAND()`, individual types don't
|
||||
- The v2 API is in `aten/src/ATen/Dispatch_v2.h` - refer to it for full docs
|
||||
- See the header file for the Python script to regenerate the macro implementation
|
||||
|
||||
## Workflow
|
||||
|
||||
When asked to convert AT_DISPATCH macros:
|
||||
|
||||
1. Read the file to identify all AT_DISPATCH uses
|
||||
2. Add `#include <ATen/Dispatch_v2.h>` if not present
|
||||
3. For each dispatch macro:
|
||||
- Identify the pattern and extract components
|
||||
- Map the base type group
|
||||
- Extract individual types
|
||||
- Construct the AT_DISPATCH_V2 call
|
||||
- Apply with Edit tool
|
||||
4. Show the user the complete converted file
|
||||
5. Explain what was changed
|
||||
|
||||
Do NOT compile or test the code - focus on accurate conversion only.
|
||||
@ -1,8 +1,3 @@
|
||||
---
|
||||
name: docstring
|
||||
description: Write docstrings for PyTorch functions and methods following PyTorch conventions. Use when writing or updating docstrings in PyTorch code.
|
||||
---
|
||||
|
||||
# PyTorch Docstring Writing Guide
|
||||
|
||||
This skill describes how to write docstrings for functions and methods in the PyTorch project, following the conventions in `torch/_tensor_docs.py` and `torch/nn/functional.py`.
|
||||
@ -1,385 +0,0 @@
|
||||
---
|
||||
name: skill-writer
|
||||
description: Guide users through creating Agent Skills for Claude Code. Use when the user wants to create, write, author, or design a new Skill, or needs help with SKILL.md files, frontmatter, or skill structure.
|
||||
---
|
||||
|
||||
# Skill Writer
|
||||
|
||||
This Skill helps you create well-structured Agent Skills for Claude Code that follow best practices and validation requirements.
|
||||
|
||||
## When to use this Skill
|
||||
|
||||
Use this Skill when:
|
||||
- Creating a new Agent Skill
|
||||
- Writing or updating SKILL.md files
|
||||
- Designing skill structure and frontmatter
|
||||
- Troubleshooting skill discovery issues
|
||||
- Converting existing prompts or workflows into Skills
|
||||
|
||||
## Instructions
|
||||
|
||||
### Step 1: Determine Skill scope
|
||||
|
||||
First, understand what the Skill should do:
|
||||
|
||||
1. **Ask clarifying questions**:
|
||||
- What specific capability should this Skill provide?
|
||||
- When should Claude use this Skill?
|
||||
- What tools or resources does it need?
|
||||
- Is this for personal use or team sharing?
|
||||
|
||||
2. **Keep it focused**: One Skill = one capability
|
||||
- Good: "PDF form filling", "Excel data analysis"
|
||||
- Too broad: "Document processing", "Data tools"
|
||||
|
||||
### Step 2: Choose Skill location
|
||||
|
||||
Determine where to create the Skill:
|
||||
|
||||
**Personal Skills** (`~/.claude/skills/`):
|
||||
- Individual workflows and preferences
|
||||
- Experimental Skills
|
||||
- Personal productivity tools
|
||||
|
||||
**Project Skills** (`.claude/skills/`):
|
||||
- Team workflows and conventions
|
||||
- Project-specific expertise
|
||||
- Shared utilities (committed to git)
|
||||
|
||||
### Step 3: Create Skill structure
|
||||
|
||||
Create the directory and files:
|
||||
|
||||
```bash
|
||||
# Personal
|
||||
mkdir -p ~/.claude/skills/skill-name
|
||||
|
||||
# Project
|
||||
mkdir -p .claude/skills/skill-name
|
||||
```
|
||||
|
||||
For multi-file Skills:
|
||||
```
|
||||
skill-name/
|
||||
├── SKILL.md (required)
|
||||
├── reference.md (optional)
|
||||
├── examples.md (optional)
|
||||
├── scripts/
|
||||
│ └── helper.py (optional)
|
||||
└── templates/
|
||||
└── template.txt (optional)
|
||||
```
|
||||
|
||||
### Step 4: Write SKILL.md frontmatter
|
||||
|
||||
Create YAML frontmatter with required fields:
|
||||
|
||||
```yaml
|
||||
---
|
||||
name: skill-name
|
||||
description: Brief description of what this does and when to use it
|
||||
---
|
||||
```
|
||||
|
||||
**Field requirements**:
|
||||
|
||||
- **name**:
|
||||
- Lowercase letters, numbers, hyphens only
|
||||
- Max 64 characters
|
||||
- Must match directory name
|
||||
- Good: `pdf-processor`, `git-commit-helper`
|
||||
- Bad: `PDF_Processor`, `Git Commits!`
|
||||
|
||||
- **description**:
|
||||
- Max 1024 characters
|
||||
- Include BOTH what it does AND when to use it
|
||||
- Use specific trigger words users would say
|
||||
- Mention file types, operations, and context
|
||||
|
||||
**Optional frontmatter fields**:
|
||||
|
||||
- **allowed-tools**: Restrict tool access (comma-separated list)
|
||||
```yaml
|
||||
allowed-tools: Read, Grep, Glob
|
||||
```
|
||||
Use for:
|
||||
- Read-only Skills
|
||||
- Security-sensitive workflows
|
||||
- Limited-scope operations
|
||||
|
||||
### Step 5: Write effective descriptions
|
||||
|
||||
The description is critical for Claude to discover your Skill.
|
||||
|
||||
**Formula**: `[What it does] + [When to use it] + [Key triggers]`
|
||||
|
||||
**Examples**:
|
||||
|
||||
✅ **Good**:
|
||||
```yaml
|
||||
description: Extract text and tables from PDF files, fill forms, merge documents. Use when working with PDF files or when the user mentions PDFs, forms, or document extraction.
|
||||
```
|
||||
|
||||
✅ **Good**:
|
||||
```yaml
|
||||
description: Analyze Excel spreadsheets, create pivot tables, and generate charts. Use when working with Excel files, spreadsheets, or analyzing tabular data in .xlsx format.
|
||||
```
|
||||
|
||||
❌ **Too vague**:
|
||||
```yaml
|
||||
description: Helps with documents
|
||||
description: For data analysis
|
||||
```
|
||||
|
||||
**Tips**:
|
||||
- Include specific file extensions (.pdf, .xlsx, .json)
|
||||
- Mention common user phrases ("analyze", "extract", "generate")
|
||||
- List concrete operations (not generic verbs)
|
||||
- Add context clues ("Use when...", "For...")
|
||||
|
||||
### Step 6: Structure the Skill content
|
||||
|
||||
Use clear Markdown sections:
|
||||
|
||||
```markdown
|
||||
# Skill Name
|
||||
|
||||
Brief overview of what this Skill does.
|
||||
|
||||
## Quick start
|
||||
|
||||
Provide a simple example to get started immediately.
|
||||
|
||||
## Instructions
|
||||
|
||||
Step-by-step guidance for Claude:
|
||||
1. First step with clear action
|
||||
2. Second step with expected outcome
|
||||
3. Handle edge cases
|
||||
|
||||
## Examples
|
||||
|
||||
Show concrete usage examples with code or commands.
|
||||
|
||||
## Best practices
|
||||
|
||||
- Key conventions to follow
|
||||
- Common pitfalls to avoid
|
||||
- When to use vs. not use
|
||||
|
||||
## Requirements
|
||||
|
||||
List any dependencies or prerequisites:
|
||||
```bash
|
||||
pip install package-name
|
||||
```
|
||||
|
||||
## Advanced usage
|
||||
|
||||
For complex scenarios, see [reference.md](reference.md).
|
||||
```
|
||||
|
||||
### Step 7: Add supporting files (optional)
|
||||
|
||||
Create additional files for progressive disclosure:
|
||||
|
||||
**reference.md**: Detailed API docs, advanced options
|
||||
**examples.md**: Extended examples and use cases
|
||||
**scripts/**: Helper scripts and utilities
|
||||
**templates/**: File templates or boilerplate
|
||||
|
||||
Reference them from SKILL.md:
|
||||
```markdown
|
||||
For advanced usage, see [reference.md](reference.md).
|
||||
|
||||
Run the helper script:
|
||||
\`\`\`bash
|
||||
python scripts/helper.py input.txt
|
||||
\`\`\`
|
||||
```
|
||||
|
||||
### Step 8: Validate the Skill
|
||||
|
||||
Check these requirements:
|
||||
|
||||
✅ **File structure**:
|
||||
- [ ] SKILL.md exists in correct location
|
||||
- [ ] Directory name matches frontmatter `name`
|
||||
|
||||
✅ **YAML frontmatter**:
|
||||
- [ ] Opening `---` on line 1
|
||||
- [ ] Closing `---` before content
|
||||
- [ ] Valid YAML (no tabs, correct indentation)
|
||||
- [ ] `name` follows naming rules
|
||||
- [ ] `description` is specific and < 1024 chars
|
||||
|
||||
✅ **Content quality**:
|
||||
- [ ] Clear instructions for Claude
|
||||
- [ ] Concrete examples provided
|
||||
- [ ] Edge cases handled
|
||||
- [ ] Dependencies listed (if any)
|
||||
|
||||
✅ **Testing**:
|
||||
- [ ] Description matches user questions
|
||||
- [ ] Skill activates on relevant queries
|
||||
- [ ] Instructions are clear and actionable
|
||||
|
||||
### Step 9: Test the Skill
|
||||
|
||||
1. **Restart Claude Code** (if running) to load the Skill
|
||||
|
||||
2. **Ask relevant questions** that match the description:
|
||||
```
|
||||
Can you help me extract text from this PDF?
|
||||
```
|
||||
|
||||
3. **Verify activation**: Claude should use the Skill automatically
|
||||
|
||||
4. **Check behavior**: Confirm Claude follows the instructions correctly
|
||||
|
||||
### Step 10: Debug if needed
|
||||
|
||||
If Claude doesn't use the Skill:
|
||||
|
||||
1. **Make description more specific**:
|
||||
- Add trigger words
|
||||
- Include file types
|
||||
- Mention common user phrases
|
||||
|
||||
2. **Check file location**:
|
||||
```bash
|
||||
ls ~/.claude/skills/skill-name/SKILL.md
|
||||
ls .claude/skills/skill-name/SKILL.md
|
||||
```
|
||||
|
||||
3. **Validate YAML**:
|
||||
```bash
|
||||
cat SKILL.md | head -n 10
|
||||
```
|
||||
|
||||
4. **Run debug mode**:
|
||||
```bash
|
||||
claude --debug
|
||||
```
|
||||
|
||||
## Common patterns
|
||||
|
||||
### Read-only Skill
|
||||
|
||||
```yaml
|
||||
---
|
||||
name: code-reader
|
||||
description: Read and analyze code without making changes. Use for code review, understanding codebases, or documentation.
|
||||
allowed-tools: Read, Grep, Glob
|
||||
---
|
||||
```
|
||||
|
||||
### Script-based Skill
|
||||
|
||||
```yaml
|
||||
---
|
||||
name: data-processor
|
||||
description: Process CSV and JSON data files with Python scripts. Use when analyzing data files or transforming datasets.
|
||||
---
|
||||
|
||||
# Data Processor
|
||||
|
||||
## Instructions
|
||||
|
||||
1. Use the processing script:
|
||||
\`\`\`bash
|
||||
python scripts/process.py input.csv --output results.json
|
||||
\`\`\`
|
||||
|
||||
2. Validate output with:
|
||||
\`\`\`bash
|
||||
python scripts/validate.py results.json
|
||||
\`\`\`
|
||||
```
|
||||
|
||||
### Multi-file Skill with progressive disclosure
|
||||
|
||||
```yaml
|
||||
---
|
||||
name: api-designer
|
||||
description: Design REST APIs following best practices. Use when creating API endpoints, designing routes, or planning API architecture.
|
||||
---
|
||||
|
||||
# API Designer
|
||||
|
||||
Quick start: See [examples.md](examples.md)
|
||||
|
||||
Detailed reference: See [reference.md](reference.md)
|
||||
|
||||
## Instructions
|
||||
|
||||
1. Gather requirements
|
||||
2. Design endpoints (see examples.md)
|
||||
3. Document with OpenAPI spec
|
||||
4. Review against best practices (see reference.md)
|
||||
```
|
||||
|
||||
## Best practices for Skill authors
|
||||
|
||||
1. **One Skill, one purpose**: Don't create mega-Skills
|
||||
2. **Specific descriptions**: Include trigger words users will say
|
||||
3. **Clear instructions**: Write for Claude, not humans
|
||||
4. **Concrete examples**: Show real code, not pseudocode
|
||||
5. **List dependencies**: Mention required packages in description
|
||||
6. **Test with teammates**: Verify activation and clarity
|
||||
7. **Version your Skills**: Document changes in content
|
||||
8. **Use progressive disclosure**: Put advanced details in separate files
|
||||
|
||||
## Validation checklist
|
||||
|
||||
Before finalizing a Skill, verify:
|
||||
|
||||
- [ ] Name is lowercase, hyphens only, max 64 chars
|
||||
- [ ] Description is specific and < 1024 chars
|
||||
- [ ] Description includes "what" and "when"
|
||||
- [ ] YAML frontmatter is valid
|
||||
- [ ] Instructions are step-by-step
|
||||
- [ ] Examples are concrete and realistic
|
||||
- [ ] Dependencies are documented
|
||||
- [ ] File paths use forward slashes
|
||||
- [ ] Skill activates on relevant queries
|
||||
- [ ] Claude follows instructions correctly
|
||||
|
||||
## Troubleshooting
|
||||
|
||||
**Skill doesn't activate**:
|
||||
- Make description more specific with trigger words
|
||||
- Include file types and operations in description
|
||||
- Add "Use when..." clause with user phrases
|
||||
|
||||
**Multiple Skills conflict**:
|
||||
- Make descriptions more distinct
|
||||
- Use different trigger words
|
||||
- Narrow the scope of each Skill
|
||||
|
||||
**Skill has errors**:
|
||||
- Check YAML syntax (no tabs, proper indentation)
|
||||
- Verify file paths (use forward slashes)
|
||||
- Ensure scripts have execute permissions
|
||||
- List all dependencies
|
||||
|
||||
## Examples
|
||||
|
||||
See the documentation for complete examples:
|
||||
- Simple single-file Skill (commit-helper)
|
||||
- Skill with tool permissions (code-reviewer)
|
||||
- Multi-file Skill (pdf-processing)
|
||||
|
||||
## Output format
|
||||
|
||||
When creating a Skill, I will:
|
||||
|
||||
1. Ask clarifying questions about scope and requirements
|
||||
2. Suggest a Skill name and location
|
||||
3. Create the SKILL.md file with proper frontmatter
|
||||
4. Include clear instructions and examples
|
||||
5. Add supporting files if needed
|
||||
6. Provide testing instructions
|
||||
7. Validate against all requirements
|
||||
|
||||
The result will be a complete, working Skill that follows all best practices and validation rules.
|
||||
4
.github/actions/diskspace-cleanup/action.yml
vendored
4
.github/actions/diskspace-cleanup/action.yml
vendored
@ -27,9 +27,7 @@ runs:
|
||||
docker system prune -af
|
||||
diskspace_new=$(df -H --output=pcent ${docker_root_dir} | sed -n 2p | sed 's/%//' | sed 's/ //')
|
||||
if [[ "$diskspace_new" -gt "$diskspace_cutoff" ]] ; then
|
||||
diskspace_cutoff_int=$((diskspace_cutoff + 0))
|
||||
difference=$((100 - diskspace_cutoff_int))
|
||||
echo "Error: Available diskspace is less than $difference percent. Not enough diskspace."
|
||||
echo "Error: Available diskspace is less than $diskspace_cutoff percent. Not enough diskspace."
|
||||
echo "$msg"
|
||||
exit 1
|
||||
else
|
||||
|
||||
2
.github/ci_commit_pins/audio.txt
vendored
2
.github/ci_commit_pins/audio.txt
vendored
@ -1 +1 @@
|
||||
3b0e7a6f192ca2715e7e6cbe5db007aea7165fe2
|
||||
69bbe7363897764f9e758d851cd0340147d27f94
|
||||
|
||||
2
.github/ci_commit_pins/vision.txt
vendored
2
.github/ci_commit_pins/vision.txt
vendored
@ -1 +1 @@
|
||||
cfbc5c2f1c798991715a6b06bb3ce46478c4487c
|
||||
1752fe6809b74921644866275ab80244b96e80bc
|
||||
|
||||
2
.github/ci_commit_pins/xla.txt
vendored
2
.github/ci_commit_pins/xla.txt
vendored
@ -1 +1 @@
|
||||
c8b09f5f77d6bf6fb7ed7a9aa83e5d8156b3a5e9
|
||||
df6798dfb931ce7c7fe5bed2447cd1092a5981af
|
||||
|
||||
20
.github/merge_rules.yaml
vendored
20
.github/merge_rules.yaml
vendored
@ -540,26 +540,6 @@
|
||||
- Lint
|
||||
- pull
|
||||
|
||||
- name: PrivateUse1
|
||||
patterns:
|
||||
- torch/accelerator/**
|
||||
- torch/utils/backend_registration.py
|
||||
- torch/csrc/acc/**
|
||||
- torch/csrc/DeviceAccelerator.*
|
||||
- torch/csrc/profiler/standalone/privateuse1_observer.*
|
||||
- aten/src/ATen/DeviceAccelerator.*
|
||||
- aten/src/ATen/core/GeneratorForPrivateuseone.*
|
||||
- aten/src/ATen/detail/PrivateUse1HooksInterface.*
|
||||
- docs/source/accelerator/**
|
||||
- test/cpp_extensions/open_registration_extension/torch_openreg/**
|
||||
approved_by:
|
||||
- albanD
|
||||
- fffrog
|
||||
mandatory_checks_name:
|
||||
- EasyCLA
|
||||
- Lint
|
||||
- pull
|
||||
|
||||
- name: superuser
|
||||
patterns:
|
||||
- '*'
|
||||
|
||||
2
.github/pytorch-probot.yml
vendored
2
.github/pytorch-probot.yml
vendored
@ -19,7 +19,6 @@ ciflow_push_tags:
|
||||
- ciflow/inductor-perf-test-nightly-rocm-mi300
|
||||
- ciflow/inductor-perf-test-nightly-rocm-mi355
|
||||
- ciflow/inductor-perf-test-nightly-x86-zen
|
||||
- ciflow/inductor-perf-test-nightly-xpu
|
||||
- ciflow/inductor-periodic
|
||||
- ciflow/inductor-rocm
|
||||
- ciflow/linux-aarch64
|
||||
@ -27,7 +26,6 @@ ciflow_push_tags:
|
||||
- ciflow/nightly
|
||||
- ciflow/op-benchmark
|
||||
- ciflow/periodic
|
||||
- ciflow/periodic-rocm-mi200
|
||||
- ciflow/periodic-rocm-mi300
|
||||
- ciflow/pull
|
||||
- ciflow/quantization-periodic
|
||||
|
||||
119
.github/scripts/generate_binary_build_matrix.py
vendored
119
.github/scripts/generate_binary_build_matrix.py
vendored
@ -11,24 +11,18 @@ architectures:
|
||||
* Latest XPU
|
||||
"""
|
||||
|
||||
import json
|
||||
import os
|
||||
import re
|
||||
from pathlib import Path
|
||||
from typing import Optional
|
||||
|
||||
|
||||
SCRIPT_DIR = Path(__file__).absolute().parent
|
||||
REPO_ROOT = SCRIPT_DIR.parent.parent
|
||||
|
||||
|
||||
# NOTE: Please also update the CUDA sources in `PIP_SOURCES` in tools/nightly.py when changing this
|
||||
CUDA_ARCHES = ["12.6", "12.8", "12.9", "13.0"]
|
||||
CUDA_STABLE = "12.8"
|
||||
CUDA_ARCHES_FULL_VERSION = {
|
||||
"12.6": "12.6.3",
|
||||
"12.8": "12.8.1",
|
||||
"12.9": "12.9.1",
|
||||
"13.0": "13.0.2",
|
||||
"13.0": "13.0.0",
|
||||
}
|
||||
CUDA_ARCHES_CUDNN_VERSION = {
|
||||
"12.6": "9",
|
||||
@ -37,7 +31,8 @@ CUDA_ARCHES_CUDNN_VERSION = {
|
||||
"13.0": "9",
|
||||
}
|
||||
|
||||
ROCM_ARCHES = ["7.0", "7.1"]
|
||||
# NOTE: Please also update the ROCm sources in `PIP_SOURCES` in tools/nightly.py when changing this
|
||||
ROCM_ARCHES = ["6.4", "7.0"]
|
||||
|
||||
XPU_ARCHES = ["xpu"]
|
||||
|
||||
@ -61,7 +56,7 @@ PYTORCH_EXTRA_INSTALL_REQUIREMENTS = {
|
||||
"nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | "
|
||||
"nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | "
|
||||
"nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | "
|
||||
"nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | "
|
||||
"nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | "
|
||||
"nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | "
|
||||
"nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | "
|
||||
"nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'"
|
||||
@ -78,7 +73,7 @@ PYTORCH_EXTRA_INSTALL_REQUIREMENTS = {
|
||||
"nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | "
|
||||
"nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | "
|
||||
"nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | "
|
||||
"nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | "
|
||||
"nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | "
|
||||
"nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | "
|
||||
"nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | "
|
||||
"nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'"
|
||||
@ -95,27 +90,27 @@ PYTORCH_EXTRA_INSTALL_REQUIREMENTS = {
|
||||
"nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' | "
|
||||
"nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | "
|
||||
"nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | "
|
||||
"nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | "
|
||||
"nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | "
|
||||
"nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' | "
|
||||
"nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' | "
|
||||
"nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux'"
|
||||
),
|
||||
"13.0": (
|
||||
"nvidia-cuda-nvrtc==13.0.88; platform_system == 'Linux' | "
|
||||
"nvidia-cuda-runtime==13.0.96; platform_system == 'Linux' | "
|
||||
"nvidia-cuda-cupti==13.0.85; platform_system == 'Linux' | "
|
||||
"nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | "
|
||||
"nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | "
|
||||
"nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | "
|
||||
"nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | "
|
||||
"nvidia-cublas==13.1.0.3; platform_system == 'Linux' | "
|
||||
"nvidia-cufft==12.0.0.61; platform_system == 'Linux' | "
|
||||
"nvidia-cublas==13.0.0.19; platform_system == 'Linux' | "
|
||||
"nvidia-cufft==12.0.0.15; platform_system == 'Linux' | "
|
||||
"nvidia-curand==10.4.0.35; platform_system == 'Linux' | "
|
||||
"nvidia-cusolver==12.0.4.66; platform_system == 'Linux' | "
|
||||
"nvidia-cusparse==12.6.3.3; platform_system == 'Linux' | "
|
||||
"nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | "
|
||||
"nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | "
|
||||
"nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | "
|
||||
"nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | "
|
||||
"nvidia-nvshmem-cu13==3.4.5; platform_system == 'Linux' | "
|
||||
"nvidia-nvtx==13.0.85; platform_system == 'Linux' | "
|
||||
"nvidia-nvjitlink==13.0.88; platform_system == 'Linux' | "
|
||||
"nvidia-cufile==1.15.1.6; platform_system == 'Linux'"
|
||||
"nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | "
|
||||
"nvidia-nvtx==13.0.39; platform_system == 'Linux' | "
|
||||
"nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | "
|
||||
"nvidia-cufile==1.15.0.42; platform_system == 'Linux'"
|
||||
),
|
||||
"xpu": (
|
||||
"intel-cmplr-lib-rt==2025.2.1 | "
|
||||
@ -142,48 +137,9 @@ PYTORCH_EXTRA_INSTALL_REQUIREMENTS = {
|
||||
}
|
||||
|
||||
|
||||
# Used by tools/nightly.py
|
||||
PYTORCH_NIGHTLY_PIP_INDEX_URL = "https://download.pytorch.org/whl/nightly"
|
||||
NIGHTLY_SOURCE_MATRIX = {
|
||||
"cpu": dict(
|
||||
name="cpu",
|
||||
index_url=f"{PYTORCH_NIGHTLY_PIP_INDEX_URL}/cpu",
|
||||
supported_platforms=["Linux", "macOS", "Windows"],
|
||||
accelerator="cpu",
|
||||
)
|
||||
}
|
||||
CUDA_NIGHTLY_SOURCE_MATRIX = {
|
||||
f"cuda-{major}.{minor}": dict(
|
||||
name=f"cuda-{major}.{minor}",
|
||||
index_url=f"{PYTORCH_NIGHTLY_PIP_INDEX_URL}/cu{major}{minor}",
|
||||
supported_platforms=["Linux", "Windows"],
|
||||
accelerator="cuda",
|
||||
)
|
||||
for major, minor in (map(int, version.split(".")) for version in CUDA_ARCHES)
|
||||
}
|
||||
ROCM_NIGHTLY_SOURCE_MATRIX = {
|
||||
f"rocm-{major}.{minor}": dict(
|
||||
name=f"rocm-{major}.{minor}",
|
||||
index_url=f"{PYTORCH_NIGHTLY_PIP_INDEX_URL}/rocm{major}.{minor}",
|
||||
supported_platforms=["Linux"],
|
||||
accelerator="rocm",
|
||||
)
|
||||
for major, minor in (map(int, version.split(".")) for version in ROCM_ARCHES)
|
||||
}
|
||||
XPU_NIGHTLY_SOURCE_MATRIX = {
|
||||
"xpu": dict(
|
||||
name="xpu",
|
||||
index_url=f"{PYTORCH_NIGHTLY_PIP_INDEX_URL}/xpu",
|
||||
supported_platforms=["Linux"],
|
||||
accelerator="xpu",
|
||||
)
|
||||
}
|
||||
NIGHTLY_SOURCE_MATRIX.update(CUDA_NIGHTLY_SOURCE_MATRIX)
|
||||
NIGHTLY_SOURCE_MATRIX.update(ROCM_NIGHTLY_SOURCE_MATRIX)
|
||||
NIGHTLY_SOURCE_MATRIX.update(XPU_NIGHTLY_SOURCE_MATRIX)
|
||||
|
||||
|
||||
def get_nccl_wheel_version(arch_version: str) -> str:
|
||||
import re
|
||||
|
||||
requirements = map(
|
||||
str.strip, re.split("[;|]", PYTORCH_EXTRA_INSTALL_REQUIREMENTS[arch_version])
|
||||
)
|
||||
@ -191,14 +147,17 @@ def get_nccl_wheel_version(arch_version: str) -> str:
|
||||
|
||||
|
||||
def read_nccl_pin(arch_version: str) -> str:
|
||||
nccl_pin_path = (
|
||||
REPO_ROOT
|
||||
/ ".ci"
|
||||
/ "docker"
|
||||
/ "ci_commit_pins"
|
||||
/ f"nccl-cu{arch_version[:2]}.txt"
|
||||
from pathlib import Path
|
||||
|
||||
nccl_pin_path = os.path.join(
|
||||
Path(__file__).absolute().parents[2],
|
||||
".ci",
|
||||
"docker",
|
||||
"ci_commit_pins",
|
||||
f"nccl-cu{arch_version[:2]}.txt",
|
||||
)
|
||||
return nccl_pin_path.read_text().strip()
|
||||
with open(nccl_pin_path) as f:
|
||||
return f.read().strip()
|
||||
|
||||
|
||||
def validate_nccl_dep_consistency(arch_version: str) -> None:
|
||||
@ -206,8 +165,7 @@ def validate_nccl_dep_consistency(arch_version: str) -> None:
|
||||
wheel_ver = get_nccl_wheel_version(arch_version)
|
||||
if not nccl_release_tag.startswith(f"v{wheel_ver}"):
|
||||
raise RuntimeError(
|
||||
f"{arch_version} NCCL release tag version {nccl_release_tag} "
|
||||
f"does not correspond to wheel version {wheel_ver}"
|
||||
f"{arch_version} NCCL release tag version {nccl_release_tag} does not correspond to wheel version {wheel_ver}"
|
||||
)
|
||||
|
||||
|
||||
@ -454,14 +412,7 @@ def generate_wheels_matrix(
|
||||
return ret
|
||||
|
||||
|
||||
arch_version = ""
|
||||
for arch_version in CUDA_ARCHES:
|
||||
validate_nccl_dep_consistency(arch_version)
|
||||
del arch_version
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
# Used by tools/nightly.py
|
||||
(SCRIPT_DIR / "nightly_source_matrix.json").write_text(
|
||||
json.dumps(NIGHTLY_SOURCE_MATRIX, indent=4) + "\n"
|
||||
)
|
||||
validate_nccl_dep_consistency("13.0")
|
||||
validate_nccl_dep_consistency("12.9")
|
||||
validate_nccl_dep_consistency("12.8")
|
||||
validate_nccl_dep_consistency("12.6")
|
||||
|
||||
13
.github/workflows/_xpu-test.yml
vendored
13
.github/workflows/_xpu-test.yml
vendored
@ -38,10 +38,6 @@ on:
|
||||
default: ""
|
||||
description: |
|
||||
List of tests to include (empty string implies default list)
|
||||
dashboard-tag:
|
||||
required: false
|
||||
type: string
|
||||
default: ""
|
||||
disable-monitor:
|
||||
description: |
|
||||
[Experimental] Disable utilization monitoring for tests.
|
||||
@ -62,11 +58,6 @@ on:
|
||||
required: false
|
||||
type: number
|
||||
default: 1
|
||||
secrets:
|
||||
HUGGING_FACE_HUB_TOKEN:
|
||||
required: false
|
||||
description: |
|
||||
HF Auth token to avoid rate limits when downloading models or datasets from hub
|
||||
permissions:
|
||||
id-token: write
|
||||
contents: read
|
||||
@ -205,8 +196,6 @@ jobs:
|
||||
PYTORCH_TEST_CUDA_MEM_LEAK_CHECK: ${{ matrix.mem_leak_check && '1' || '0' }}
|
||||
PYTORCH_TEST_RERUN_DISABLED_TESTS: ${{ matrix.rerun_disabled_tests && '1' || '0' }}
|
||||
TESTS_TO_INCLUDE: ${{ inputs.tests-to-include }}
|
||||
DASHBOARD_TAG: ${{ inputs.dashboard-tag }}
|
||||
HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
|
||||
timeout-minutes: ${{ fromJson(steps.test-timeout.outputs.timeout) }}
|
||||
run: |
|
||||
# Fetch aws credential from IMDs
|
||||
@ -257,8 +246,6 @@ jobs:
|
||||
-e PYTORCH_TEST_RERUN_DISABLED_TESTS \
|
||||
-e TESTS_TO_INCLUDE \
|
||||
-e ZE_AFFINITY_MASK \
|
||||
-e HUGGING_FACE_HUB_TOKEN \
|
||||
-e DASHBOARD_TAG \
|
||||
--env-file="/tmp/github_env_${GITHUB_RUN_ID}" \
|
||||
--ulimit stack=10485760:83886080 \
|
||||
--ulimit core=0 \
|
||||
|
||||
2
.github/workflows/build-almalinux-images.yml
vendored
2
.github/workflows/build-almalinux-images.yml
vendored
@ -36,7 +36,7 @@ jobs:
|
||||
runs-on: linux.9xlarge.ephemeral
|
||||
strategy:
|
||||
matrix:
|
||||
tag: ["cuda12.6", "cuda12.8", "cuda12.9", "cuda13.0", "rocm7.0", "rocm7.1", "cpu"]
|
||||
tag: ["cuda12.6", "cuda12.8", "cuda12.9", "cuda13.0", "rocm6.4", "rocm7.0", "cpu"]
|
||||
steps:
|
||||
- name: Build docker image
|
||||
uses: pytorch/pytorch/.github/actions/binary-docker-build@main
|
||||
|
||||
2
.github/workflows/build-libtorch-images.yml
vendored
2
.github/workflows/build-libtorch-images.yml
vendored
@ -52,8 +52,8 @@ jobs:
|
||||
{ tag: "cuda12.9" },
|
||||
{ tag: "cuda12.8" },
|
||||
{ tag: "cuda12.6" },
|
||||
{ tag: "rocm6.4" },
|
||||
{ tag: "rocm7.0" },
|
||||
{ tag: "rocm7.1" },
|
||||
{ tag: "cpu" },
|
||||
]
|
||||
steps:
|
||||
|
||||
2
.github/workflows/build-magma-rocm-linux.yml
vendored
2
.github/workflows/build-magma-rocm-linux.yml
vendored
@ -34,7 +34,7 @@ jobs:
|
||||
id-token: write
|
||||
strategy:
|
||||
matrix:
|
||||
rocm_version: ["71", "70"]
|
||||
rocm_version: ["70", "64"]
|
||||
steps:
|
||||
- name: Checkout PyTorch
|
||||
uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
|
||||
|
||||
2
.github/workflows/build-manywheel-images.yml
vendored
2
.github/workflows/build-manywheel-images.yml
vendored
@ -54,8 +54,8 @@ jobs:
|
||||
{ name: "manylinuxaarch64-builder", tag: "cuda12.9", runner: "linux.arm64.2xlarge.ephemeral" },
|
||||
{ name: "manylinuxaarch64-builder", tag: "cuda12.8", runner: "linux.arm64.2xlarge.ephemeral" },
|
||||
{ name: "manylinuxaarch64-builder", tag: "cuda12.6", runner: "linux.arm64.2xlarge.ephemeral" },
|
||||
{ name: "manylinux2_28-builder", tag: "rocm6.4", runner: "linux.9xlarge.ephemeral" },
|
||||
{ name: "manylinux2_28-builder", tag: "rocm7.0", runner: "linux.9xlarge.ephemeral" },
|
||||
{ name: "manylinux2_28-builder", tag: "rocm7.1", runner: "linux.9xlarge.ephemeral" },
|
||||
{ name: "manylinux2_28-builder", tag: "cpu", runner: "linux.9xlarge.ephemeral" },
|
||||
{ name: "manylinux2_28_aarch64-builder", tag: "cpu-aarch64", runner: "linux.arm64.2xlarge.ephemeral" },
|
||||
{ name: "manylinux2_28-builder", tag: "xpu", runner: "linux.9xlarge.ephemeral" },
|
||||
|
||||
9
.github/workflows/build-triton-wheel.yml
vendored
9
.github/workflows/build-triton-wheel.yml
vendored
@ -55,7 +55,7 @@ jobs:
|
||||
docker-image: ["pytorch/manylinux2_28-builder:cpu"]
|
||||
include:
|
||||
- device: "rocm"
|
||||
rocm_version: "7.1"
|
||||
rocm_version: "7.0"
|
||||
runs_on: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge"
|
||||
- device: "cuda"
|
||||
rocm_version: ""
|
||||
@ -159,7 +159,12 @@ jobs:
|
||||
WITH_CLANG_LDD="--with-clang-ldd"
|
||||
fi
|
||||
|
||||
docker exec -t "${container_name}" bash -c "${PYTHON_EXECUTABLE} /pytorch/.github/scripts/build_triton_wheel.py --device=$BUILD_DEVICE $RELEASE $WITH_CLANG_LDD"
|
||||
if [[ "${BUILD_DEVICE}" == xpu ]]; then
|
||||
docker exec -t "${container_name}" bash -c "dnf install -y gcc-toolset-13-gcc-c++"
|
||||
docker exec -t "${container_name}" bash -c "source /opt/rh/gcc-toolset-13/enable && ${PYTHON_EXECUTABLE} /pytorch/.github/scripts/build_triton_wheel.py --device=$BUILD_DEVICE $RELEASE"
|
||||
else
|
||||
docker exec -t "${container_name}" bash -c "${PYTHON_EXECUTABLE} /pytorch/.github/scripts/build_triton_wheel.py --device=$BUILD_DEVICE $RELEASE $WITH_CLANG_LDD"
|
||||
fi
|
||||
|
||||
if [[ ("${{ matrix.device }}" == "cuda" || "${{ matrix.device }}" == "xpu") ]]; then
|
||||
docker exec -t "${container_name}" bash -c "auditwheel repair --plat ${PLATFORM} //artifacts/*.whl"
|
||||
|
||||
2
.github/workflows/docker-builds.yml
vendored
2
.github/workflows/docker-builds.yml
vendored
@ -57,7 +57,6 @@ jobs:
|
||||
pytorch-linux-jammy-cuda12.4-cudnn9-py3-gcc11,
|
||||
pytorch-linux-jammy-py3.10-clang12,
|
||||
pytorch-linux-jammy-py3.13-clang12,
|
||||
pytorch-linux-jammy-py3.14-clang12,
|
||||
pytorch-linux-jammy-rocm-n-py3,
|
||||
pytorch-linux-noble-rocm-n-py3,
|
||||
pytorch-linux-jammy-rocm-n-py3-benchmarks,
|
||||
@ -67,7 +66,6 @@ jobs:
|
||||
pytorch-linux-jammy-py3.12-halide,
|
||||
pytorch-linux-jammy-xpu-n-1-py3,
|
||||
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,
|
||||
|
||||
56
.github/workflows/generated-linux-aarch64-binary-manywheel-nightly.yml
generated
vendored
56
.github/workflows/generated-linux-aarch64-binary-manywheel-nightly.yml
generated
vendored
@ -132,7 +132,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_10-cuda-aarch64-12_6
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -178,7 +178,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_10-cuda-aarch64-12_8
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -224,7 +224,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_10-cuda-aarch64-12_9
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -270,7 +270,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_10-cuda-aarch64-13_0
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.88; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.96; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.85; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.1.0.3; platform_system == 'Linux' | nvidia-cufft==12.0.0.61; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.4.66; platform_system == 'Linux' | nvidia-cusparse==12.6.3.3; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.4.5; platform_system == 'Linux' | nvidia-nvtx==13.0.85; platform_system == 'Linux' | nvidia-nvjitlink==13.0.88; platform_system == 'Linux' | nvidia-cufile==1.15.1.6; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | nvidia-nvtx==13.0.39; platform_system == 'Linux' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | nvidia-cufile==1.15.0.42; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -381,7 +381,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_11-cuda-aarch64-12_6
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -427,7 +427,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_11-cuda-aarch64-12_8
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -473,7 +473,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_11-cuda-aarch64-12_9
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -519,7 +519,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_11-cuda-aarch64-13_0
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.88; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.96; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.85; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.1.0.3; platform_system == 'Linux' | nvidia-cufft==12.0.0.61; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.4.66; platform_system == 'Linux' | nvidia-cusparse==12.6.3.3; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.4.5; platform_system == 'Linux' | nvidia-nvtx==13.0.85; platform_system == 'Linux' | nvidia-nvjitlink==13.0.88; platform_system == 'Linux' | nvidia-cufile==1.15.1.6; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | nvidia-nvtx==13.0.39; platform_system == 'Linux' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | nvidia-cufile==1.15.0.42; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -630,7 +630,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_12-cuda-aarch64-12_6
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -676,7 +676,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_12-cuda-aarch64-12_8
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -722,7 +722,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_12-cuda-aarch64-12_9
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -768,7 +768,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_12-cuda-aarch64-13_0
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.88; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.96; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.85; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.1.0.3; platform_system == 'Linux' | nvidia-cufft==12.0.0.61; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.4.66; platform_system == 'Linux' | nvidia-cusparse==12.6.3.3; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.4.5; platform_system == 'Linux' | nvidia-nvtx==13.0.85; platform_system == 'Linux' | nvidia-nvjitlink==13.0.88; platform_system == 'Linux' | nvidia-cufile==1.15.1.6; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | nvidia-nvtx==13.0.39; platform_system == 'Linux' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | nvidia-cufile==1.15.0.42; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -879,7 +879,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_13-cuda-aarch64-12_6
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -925,7 +925,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_13-cuda-aarch64-12_8
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -971,7 +971,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_13-cuda-aarch64-12_9
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -1017,7 +1017,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_13-cuda-aarch64-13_0
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.88; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.96; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.85; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.1.0.3; platform_system == 'Linux' | nvidia-cufft==12.0.0.61; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.4.66; platform_system == 'Linux' | nvidia-cusparse==12.6.3.3; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.4.5; platform_system == 'Linux' | nvidia-nvtx==13.0.85; platform_system == 'Linux' | nvidia-nvjitlink==13.0.88; platform_system == 'Linux' | nvidia-cufile==1.15.1.6; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | nvidia-nvtx==13.0.39; platform_system == 'Linux' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | nvidia-cufile==1.15.0.42; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -1128,7 +1128,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_13t-cuda-aarch64-12_6
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -1174,7 +1174,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_13t-cuda-aarch64-12_8
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -1220,7 +1220,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_13t-cuda-aarch64-12_9
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -1266,7 +1266,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_13t-cuda-aarch64-13_0
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.88; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.96; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.85; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.1.0.3; platform_system == 'Linux' | nvidia-cufft==12.0.0.61; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.4.66; platform_system == 'Linux' | nvidia-cusparse==12.6.3.3; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.4.5; platform_system == 'Linux' | nvidia-nvtx==13.0.85; platform_system == 'Linux' | nvidia-nvjitlink==13.0.88; platform_system == 'Linux' | nvidia-cufile==1.15.1.6; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | nvidia-nvtx==13.0.39; platform_system == 'Linux' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | nvidia-cufile==1.15.0.42; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -1377,7 +1377,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_14-cuda-aarch64-12_6
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -1423,7 +1423,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_14-cuda-aarch64-12_8
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -1469,7 +1469,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_14-cuda-aarch64-12_9
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -1515,7 +1515,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_14-cuda-aarch64-13_0
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.88; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.96; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.85; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.1.0.3; platform_system == 'Linux' | nvidia-cufft==12.0.0.61; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.4.66; platform_system == 'Linux' | nvidia-cusparse==12.6.3.3; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.4.5; platform_system == 'Linux' | nvidia-nvtx==13.0.85; platform_system == 'Linux' | nvidia-nvjitlink==13.0.88; platform_system == 'Linux' | nvidia-cufile==1.15.1.6; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | nvidia-nvtx==13.0.39; platform_system == 'Linux' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | nvidia-cufile==1.15.0.42; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -1626,7 +1626,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_14t-cuda-aarch64-12_6
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -1672,7 +1672,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_14t-cuda-aarch64-12_8
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -1718,7 +1718,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_14t-cuda-aarch64-12_9
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.4.5; platform_system == 'Linux' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
@ -1764,7 +1764,7 @@ jobs:
|
||||
ALPINE_IMAGE: "arm64v8/alpine"
|
||||
build_name: manywheel-py3_14t-cuda-aarch64-13_0
|
||||
build_environment: linux-aarch64-binary-manywheel
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.88; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.96; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.85; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.1.0.3; platform_system == 'Linux' | nvidia-cufft==12.0.0.61; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.4.66; platform_system == 'Linux' | nvidia-cusparse==12.6.3.3; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.4.5; platform_system == 'Linux' | nvidia-nvtx==13.0.85; platform_system == 'Linux' | nvidia-nvjitlink==13.0.88; platform_system == 'Linux' | nvidia-cufile==1.15.1.6; platform_system == 'Linux'
|
||||
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | nvidia-nvtx==13.0.39; platform_system == 'Linux' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | nvidia-cufile==1.15.0.42; platform_system == 'Linux'
|
||||
timeout-minutes: 420
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
|
||||
236
.github/workflows/generated-linux-binary-libtorch-nightly.yml
generated
vendored
236
.github/workflows/generated-linux-binary-libtorch-nightly.yml
generated
vendored
@ -384,6 +384,124 @@ jobs:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
uses: ./.github/workflows/_binary-upload.yml
|
||||
|
||||
libtorch-rocm6_4-shared-with-deps-release-build:
|
||||
if: ${{ github.repository_owner == 'pytorch' }}
|
||||
uses: ./.github/workflows/_binary-build-linux.yml
|
||||
needs: get-label-type
|
||||
with:
|
||||
PYTORCH_ROOT: /pytorch
|
||||
PACKAGE_TYPE: libtorch
|
||||
# TODO: This is a legacy variable that we eventually want to get rid of in
|
||||
# favor of GPU_ARCH_VERSION
|
||||
DESIRED_CUDA: rocm6.4
|
||||
GPU_ARCH_VERSION: "6.4"
|
||||
GPU_ARCH_TYPE: rocm
|
||||
DOCKER_IMAGE: libtorch-cxx11-builder
|
||||
DOCKER_IMAGE_TAG_PREFIX: rocm6.4
|
||||
LIBTORCH_CONFIG: release
|
||||
LIBTORCH_VARIANT: shared-with-deps
|
||||
runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
|
||||
timeout-minutes: 300
|
||||
build_name: libtorch-rocm6_4-shared-with-deps-release
|
||||
build_environment: linux-binary-libtorch
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
libtorch-rocm6_4-shared-with-deps-release-test: # Testing
|
||||
if: ${{ github.repository_owner == 'pytorch' }}
|
||||
needs:
|
||||
- libtorch-rocm6_4-shared-with-deps-release-build
|
||||
- get-label-type
|
||||
runs-on: linux.rocm.gpu.mi250
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
PYTORCH_ROOT: /pytorch
|
||||
PACKAGE_TYPE: libtorch
|
||||
# TODO: This is a legacy variable that we eventually want to get rid of in
|
||||
# favor of GPU_ARCH_VERSION
|
||||
DESIRED_CUDA: rocm6.4
|
||||
GPU_ARCH_VERSION: "6.4"
|
||||
GPU_ARCH_TYPE: rocm
|
||||
SKIP_ALL_TESTS: 1
|
||||
DOCKER_IMAGE: libtorch-cxx11-builder
|
||||
DOCKER_IMAGE_TAG_PREFIX: rocm6.4
|
||||
LIBTORCH_CONFIG: release
|
||||
LIBTORCH_VARIANT: shared-with-deps
|
||||
permissions:
|
||||
id-token: write
|
||||
contents: read
|
||||
steps:
|
||||
- name: Setup ROCm
|
||||
uses: ./.github/actions/setup-rocm
|
||||
- uses: actions/download-artifact@v4.1.7
|
||||
name: Download Build Artifacts
|
||||
with:
|
||||
name: libtorch-rocm6_4-shared-with-deps-release
|
||||
path: "${{ runner.temp }}/artifacts/"
|
||||
- name: Checkout PyTorch
|
||||
uses: actions/checkout@v4
|
||||
with:
|
||||
ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
|
||||
submodules: recursive
|
||||
path: pytorch
|
||||
show-progress: false
|
||||
- name: Clean PyTorch checkout
|
||||
run: |
|
||||
# Remove any artifacts from the previous checkouts
|
||||
git clean -fxd
|
||||
working-directory: pytorch
|
||||
- name: ROCm set GPU_FLAG
|
||||
run: |
|
||||
echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}"
|
||||
- name: configure aws credentials
|
||||
id: aws_creds
|
||||
if: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') }}
|
||||
uses: aws-actions/configure-aws-credentials@v4
|
||||
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: Calculate docker image
|
||||
id: calculate-docker-image
|
||||
uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
|
||||
with:
|
||||
docker-registry: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') && '308535385114.dkr.ecr.us-east-1.amazonaws.com' || 'docker.io' }}
|
||||
docker-image-name: libtorch-cxx11-builder
|
||||
custom-tag-prefix: rocm6.4
|
||||
docker-build-dir: .ci/docker
|
||||
working-directory: pytorch
|
||||
- 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: Test Pytorch binary
|
||||
uses: ./pytorch/.github/actions/test-pytorch-binary
|
||||
env:
|
||||
DOCKER_IMAGE: ${{ steps.calculate-docker-image.outputs.docker-image }}
|
||||
- name: Teardown ROCm
|
||||
uses: ./.github/actions/teardown-rocm
|
||||
libtorch-rocm6_4-shared-with-deps-release-upload: # Uploading
|
||||
if: ${{ github.repository_owner == 'pytorch' }}
|
||||
permissions:
|
||||
id-token: write
|
||||
contents: read
|
||||
needs: libtorch-rocm6_4-shared-with-deps-release-test
|
||||
with:
|
||||
PYTORCH_ROOT: /pytorch
|
||||
PACKAGE_TYPE: libtorch
|
||||
# TODO: This is a legacy variable that we eventually want to get rid of in
|
||||
# favor of GPU_ARCH_VERSION
|
||||
DESIRED_CUDA: rocm6.4
|
||||
GPU_ARCH_VERSION: "6.4"
|
||||
GPU_ARCH_TYPE: rocm
|
||||
DOCKER_IMAGE: libtorch-cxx11-builder
|
||||
DOCKER_IMAGE_TAG_PREFIX: rocm6.4
|
||||
LIBTORCH_CONFIG: release
|
||||
LIBTORCH_VARIANT: shared-with-deps
|
||||
build_name: libtorch-rocm6_4-shared-with-deps-release
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
uses: ./.github/workflows/_binary-upload.yml
|
||||
|
||||
libtorch-rocm7_0-shared-with-deps-release-build:
|
||||
if: ${{ github.repository_owner == 'pytorch' }}
|
||||
uses: ./.github/workflows/_binary-build-linux.yml
|
||||
@ -501,121 +619,3 @@ jobs:
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
uses: ./.github/workflows/_binary-upload.yml
|
||||
|
||||
libtorch-rocm7_1-shared-with-deps-release-build:
|
||||
if: ${{ github.repository_owner == 'pytorch' }}
|
||||
uses: ./.github/workflows/_binary-build-linux.yml
|
||||
needs: get-label-type
|
||||
with:
|
||||
PYTORCH_ROOT: /pytorch
|
||||
PACKAGE_TYPE: libtorch
|
||||
# TODO: This is a legacy variable that we eventually want to get rid of in
|
||||
# favor of GPU_ARCH_VERSION
|
||||
DESIRED_CUDA: rocm7.1
|
||||
GPU_ARCH_VERSION: "7.1"
|
||||
GPU_ARCH_TYPE: rocm
|
||||
DOCKER_IMAGE: libtorch-cxx11-builder
|
||||
DOCKER_IMAGE_TAG_PREFIX: rocm7.1
|
||||
LIBTORCH_CONFIG: release
|
||||
LIBTORCH_VARIANT: shared-with-deps
|
||||
runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
|
||||
timeout-minutes: 300
|
||||
build_name: libtorch-rocm7_1-shared-with-deps-release
|
||||
build_environment: linux-binary-libtorch
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
libtorch-rocm7_1-shared-with-deps-release-test: # Testing
|
||||
if: ${{ github.repository_owner == 'pytorch' }}
|
||||
needs:
|
||||
- libtorch-rocm7_1-shared-with-deps-release-build
|
||||
- get-label-type
|
||||
runs-on: linux.rocm.gpu.mi250
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
PYTORCH_ROOT: /pytorch
|
||||
PACKAGE_TYPE: libtorch
|
||||
# TODO: This is a legacy variable that we eventually want to get rid of in
|
||||
# favor of GPU_ARCH_VERSION
|
||||
DESIRED_CUDA: rocm7.1
|
||||
GPU_ARCH_VERSION: "7.1"
|
||||
GPU_ARCH_TYPE: rocm
|
||||
SKIP_ALL_TESTS: 1
|
||||
DOCKER_IMAGE: libtorch-cxx11-builder
|
||||
DOCKER_IMAGE_TAG_PREFIX: rocm7.1
|
||||
LIBTORCH_CONFIG: release
|
||||
LIBTORCH_VARIANT: shared-with-deps
|
||||
permissions:
|
||||
id-token: write
|
||||
contents: read
|
||||
steps:
|
||||
- name: Setup ROCm
|
||||
uses: ./.github/actions/setup-rocm
|
||||
- uses: actions/download-artifact@v4.1.7
|
||||
name: Download Build Artifacts
|
||||
with:
|
||||
name: libtorch-rocm7_1-shared-with-deps-release
|
||||
path: "${{ runner.temp }}/artifacts/"
|
||||
- name: Checkout PyTorch
|
||||
uses: actions/checkout@v4
|
||||
with:
|
||||
ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
|
||||
submodules: recursive
|
||||
path: pytorch
|
||||
show-progress: false
|
||||
- name: Clean PyTorch checkout
|
||||
run: |
|
||||
# Remove any artifacts from the previous checkouts
|
||||
git clean -fxd
|
||||
working-directory: pytorch
|
||||
- name: ROCm set GPU_FLAG
|
||||
run: |
|
||||
echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}"
|
||||
- name: configure aws credentials
|
||||
id: aws_creds
|
||||
if: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') }}
|
||||
uses: aws-actions/configure-aws-credentials@v4
|
||||
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: Calculate docker image
|
||||
id: calculate-docker-image
|
||||
uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
|
||||
with:
|
||||
docker-registry: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') && '308535385114.dkr.ecr.us-east-1.amazonaws.com' || 'docker.io' }}
|
||||
docker-image-name: libtorch-cxx11-builder
|
||||
custom-tag-prefix: rocm7.1
|
||||
docker-build-dir: .ci/docker
|
||||
working-directory: pytorch
|
||||
- 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: Test Pytorch binary
|
||||
uses: ./pytorch/.github/actions/test-pytorch-binary
|
||||
env:
|
||||
DOCKER_IMAGE: ${{ steps.calculate-docker-image.outputs.docker-image }}
|
||||
- name: Teardown ROCm
|
||||
uses: ./.github/actions/teardown-rocm
|
||||
libtorch-rocm7_1-shared-with-deps-release-upload: # Uploading
|
||||
if: ${{ github.repository_owner == 'pytorch' }}
|
||||
permissions:
|
||||
id-token: write
|
||||
contents: read
|
||||
needs: libtorch-rocm7_1-shared-with-deps-release-test
|
||||
with:
|
||||
PYTORCH_ROOT: /pytorch
|
||||
PACKAGE_TYPE: libtorch
|
||||
# TODO: This is a legacy variable that we eventually want to get rid of in
|
||||
# favor of GPU_ARCH_VERSION
|
||||
DESIRED_CUDA: rocm7.1
|
||||
GPU_ARCH_VERSION: "7.1"
|
||||
GPU_ARCH_TYPE: rocm
|
||||
DOCKER_IMAGE: libtorch-cxx11-builder
|
||||
DOCKER_IMAGE_TAG_PREFIX: rocm7.1
|
||||
LIBTORCH_CONFIG: release
|
||||
LIBTORCH_VARIANT: shared-with-deps
|
||||
build_name: libtorch-rocm7_1-shared-with-deps-release
|
||||
secrets:
|
||||
github-token: ${{ secrets.GITHUB_TOKEN }}
|
||||
uses: ./.github/workflows/_binary-upload.yml
|
||||
|
||||
1666
.github/workflows/generated-linux-binary-manywheel-nightly.yml
generated
vendored
1666
.github/workflows/generated-linux-binary-manywheel-nightly.yml
generated
vendored
File diff suppressed because it is too large
Load Diff
148
.github/workflows/inductor-perf-test-nightly-xpu.yml
vendored
148
.github/workflows/inductor-perf-test-nightly-xpu.yml
vendored
@ -1,148 +0,0 @@
|
||||
name: inductor-perf-nightly-xpu
|
||||
|
||||
on:
|
||||
push:
|
||||
tags:
|
||||
- ciflow/inductor-perf-test-nightly-xpu/*
|
||||
schedule:
|
||||
- cron: 30 17 * * *
|
||||
workflow_dispatch:
|
||||
inputs:
|
||||
training:
|
||||
description: Run training (on by default)?
|
||||
required: false
|
||||
type: boolean
|
||||
default: true
|
||||
inference:
|
||||
description: Run inference (on by default)?
|
||||
required: false
|
||||
type: boolean
|
||||
default: true
|
||||
default:
|
||||
description: Run inductor_default?
|
||||
required: false
|
||||
type: boolean
|
||||
default: false
|
||||
dynamic:
|
||||
description: Run inductor_dynamic_shapes?
|
||||
required: false
|
||||
type: boolean
|
||||
default: false
|
||||
cppwrapper:
|
||||
description: Run inductor_cpp_wrapper?
|
||||
required: false
|
||||
type: boolean
|
||||
default: false
|
||||
cudagraphs:
|
||||
description: Run inductor_cudagraphs?
|
||||
required: false
|
||||
type: boolean
|
||||
default: false
|
||||
freezing_cudagraphs:
|
||||
description: Run inductor_cudagraphs with freezing for inference?
|
||||
required: false
|
||||
type: boolean
|
||||
default: false
|
||||
aotinductor:
|
||||
description: Run aot_inductor for inference?
|
||||
required: false
|
||||
type: boolean
|
||||
default: false
|
||||
maxautotune:
|
||||
description: Run inductor_max_autotune?
|
||||
required: false
|
||||
type: boolean
|
||||
default: false
|
||||
benchmark_configs:
|
||||
description: The list of configs used the benchmark
|
||||
required: false
|
||||
type: string
|
||||
default: inductor_huggingface_perf,inductor_timm_perf,inductor_torchbench_perf,cachebench
|
||||
|
||||
concurrency:
|
||||
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }}
|
||||
cancel-in-progress: true
|
||||
|
||||
permissions: read-all
|
||||
|
||||
jobs:
|
||||
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: ${{ github.ref_type }}
|
||||
opt_out_experiments: lf
|
||||
|
||||
xpu-n-py3_10-inductor-benchmark-build:
|
||||
name: xpu-n-py3.10-inductor-benchmark
|
||||
uses: ./.github/workflows/_linux-build.yml
|
||||
needs: get-label-type
|
||||
with:
|
||||
runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
|
||||
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" },
|
||||
{ config: "inductor_huggingface_perf_xpu", shard: 2, num_shards: 5, runner: "linux.idc.xpu" },
|
||||
{ config: "inductor_huggingface_perf_xpu", shard: 3, num_shards: 5, runner: "linux.idc.xpu" },
|
||||
{ config: "inductor_huggingface_perf_xpu", shard: 4, num_shards: 5, runner: "linux.idc.xpu" },
|
||||
{ config: "inductor_huggingface_perf_xpu", shard: 5, num_shards: 5, runner: "linux.idc.xpu" },
|
||||
{ config: "inductor_timm_perf_xpu", shard: 1, num_shards: 6, runner: "linux.idc.xpu" },
|
||||
{ config: "inductor_timm_perf_xpu", shard: 2, num_shards: 6, runner: "linux.idc.xpu" },
|
||||
{ config: "inductor_timm_perf_xpu", shard: 3, num_shards: 6, runner: "linux.idc.xpu" },
|
||||
{ config: "inductor_timm_perf_xpu", shard: 4, num_shards: 6, runner: "linux.idc.xpu" },
|
||||
{ config: "inductor_timm_perf_xpu", shard: 5, num_shards: 6, runner: "linux.idc.xpu" },
|
||||
{ config: "inductor_timm_perf_xpu", shard: 6, num_shards: 6, runner: "linux.idc.xpu" },
|
||||
{ config: "inductor_torchbench_perf_xpu", shard: 1, num_shards: 6, runner: "linux.idc.xpu" },
|
||||
{ config: "inductor_torchbench_perf_xpu", shard: 2, num_shards: 6, runner: "linux.idc.xpu" },
|
||||
{ config: "inductor_torchbench_perf_xpu", shard: 3, num_shards: 6, runner: "linux.idc.xpu" },
|
||||
{ config: "inductor_torchbench_perf_xpu", shard: 4, num_shards: 6, runner: "linux.idc.xpu" },
|
||||
{ config: "inductor_torchbench_perf_xpu", shard: 5, num_shards: 6, runner: "linux.idc.xpu" },
|
||||
{ config: "inductor_torchbench_perf_xpu", shard: 6, num_shards: 6, runner: "linux.idc.xpu" },
|
||||
]}
|
||||
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: ${{ needs.xpu-n-py3_10-inductor-benchmark-build.outputs.test-matrix }}
|
||||
timeout-minutes: 720
|
||||
# Disable monitor in perf tests for more investigation
|
||||
disable-monitor: true
|
||||
monitor-log-interval: 10
|
||||
monitor-data-collect-interval: 2
|
||||
secrets: inherit
|
||||
|
||||
xpu-n-py3_10-inductor-benchmark-test:
|
||||
permissions:
|
||||
id-token: write
|
||||
contents: read
|
||||
if: github.event_name == 'workflow_dispatch'
|
||||
name: xpu-n-py3.10-inductor-test
|
||||
uses: ./.github/workflows/_xpu-test.yml
|
||||
needs: xpu-n-py3_10-inductor-benchmark-build
|
||||
with:
|
||||
build-environment: linux-jammy-xpu-n-py3.10
|
||||
dashboard-tag: training-${{ 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: ${{ needs.xpu-n-py3_10-inductor-benchmark-build.outputs.test-matrix }}
|
||||
timeout-minutes: 720
|
||||
disable-monitor: false
|
||||
monitor-log-interval: 15
|
||||
monitor-data-collect-interval: 4
|
||||
secrets: inherit
|
||||
3
.github/workflows/inductor-rocm.yml
vendored
3
.github/workflows/inductor-rocm.yml
vendored
@ -1,10 +1,9 @@
|
||||
name: inductor-rocm
|
||||
|
||||
on:
|
||||
schedule:
|
||||
- cron: 0 * * * *
|
||||
push:
|
||||
branches:
|
||||
- main
|
||||
- release/*
|
||||
tags:
|
||||
- ciflow/inductor-rocm/*
|
||||
|
||||
15
.github/workflows/lint.yml
vendored
15
.github/workflows/lint.yml
vendored
@ -76,12 +76,11 @@ jobs:
|
||||
|
||||
# NOTE: mypy needs its own job because it depends on --all-files, without assessing all files it sometimes
|
||||
# fails to find types when it should
|
||||
# NOTE: We should be able to disable this and consolidate with Pyrefly
|
||||
lintrunner-pyrefly:
|
||||
lintrunner-mypy:
|
||||
uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
|
||||
name: lintrunner-pyrefly-${{ needs.get-changed-files.outputs.changed-files == '*' && 'all' || 'partial' }}
|
||||
name: lintrunner-mypy-${{ needs.get-changed-files.outputs.changed-files == '*' && 'all' || 'partial' }}
|
||||
needs: [get-label-type, get-changed-files]
|
||||
# Only run if there are changed files relevant to pyrefly
|
||||
# Only run if there are changed files relevant to mypy
|
||||
if: |
|
||||
github.repository_owner == 'pytorch' && (
|
||||
needs.get-changed-files.outputs.changed-files == '*' ||
|
||||
@ -99,8 +98,8 @@ jobs:
|
||||
ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
|
||||
script: |
|
||||
CHANGED_FILES="${{ needs.get-changed-files.outputs.changed-files }}"
|
||||
echo "Running pyrefly"
|
||||
ADDITIONAL_LINTRUNNER_ARGS="--take PYREFLY --all-files" .github/scripts/lintrunner.sh
|
||||
echo "Running mypy"
|
||||
ADDITIONAL_LINTRUNNER_ARGS="--take MYPY,MYPYSTRICT --all-files" .github/scripts/lintrunner.sh
|
||||
|
||||
lintrunner-noclang:
|
||||
uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
|
||||
@ -119,9 +118,9 @@ jobs:
|
||||
CHANGED_FILES="${{ needs.get-changed-files.outputs.changed-files }}"
|
||||
echo "Running all other linters"
|
||||
if [ "$CHANGED_FILES" = '*' ]; then
|
||||
ADDITIONAL_LINTRUNNER_ARGS="--skip CLANGTIDY,CLANGFORMAT,PYREFLY --all-files" .github/scripts/lintrunner.sh
|
||||
ADDITIONAL_LINTRUNNER_ARGS="--skip CLANGTIDY,CLANGFORMAT,MYPY,MYPYSTRICT,PYREFLY --all-files" .github/scripts/lintrunner.sh
|
||||
else
|
||||
ADDITIONAL_LINTRUNNER_ARGS="--skip CLANGTIDY,CLANGFORMAT,PYREFLY ${CHANGED_FILES}" .github/scripts/lintrunner.sh
|
||||
ADDITIONAL_LINTRUNNER_ARGS="--skip CLANGTIDY,CLANGFORMAT,MYPY,MYPYSTRICT,PYREFLY ${CHANGED_FILES}" .github/scripts/lintrunner.sh
|
||||
fi
|
||||
|
||||
quick-checks:
|
||||
|
||||
2
.github/workflows/nightly.yml
vendored
2
.github/workflows/nightly.yml
vendored
@ -41,7 +41,7 @@ jobs:
|
||||
uses: ./.github/workflows/_linux-build.yml
|
||||
needs: get-label-type
|
||||
with:
|
||||
runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
|
||||
runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge"
|
||||
build-environment: linux-jammy-py3.10-gcc11
|
||||
docker-image-name: ci-image:pytorch-linux-jammy-py3.10-gcc11
|
||||
secrets: inherit
|
||||
|
||||
84
.github/workflows/periodic-rocm-mi200.yml
vendored
84
.github/workflows/periodic-rocm-mi200.yml
vendored
@ -1,84 +0,0 @@
|
||||
name: periodic-rocm-mi200
|
||||
|
||||
on:
|
||||
schedule:
|
||||
# We have several schedules so jobs can check github.event.schedule to activate only for a fraction of the runs.
|
||||
# Also run less frequently on weekends.
|
||||
- cron: 45 0,8,16 * * 1-5
|
||||
- cron: 45 4 * * 0,6
|
||||
- cron: 45 4,12,20 * * 1-5
|
||||
- cron: 45 12 * * 0,6
|
||||
- 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/*
|
||||
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
|
||||
test-matrix: |
|
||||
{ include: [
|
||||
{ config: "distributed", shard: 1, num_shards: 3, runner: "linux.rocm.gpu.4", owners: ["module:rocm", "oncall:distributed"] },
|
||||
{ config: "distributed", shard: 2, num_shards: 3, runner: "linux.rocm.gpu.4", owners: ["module:rocm", "oncall:distributed"] },
|
||||
{ config: "distributed", shard: 3, num_shards: 3, runner: "linux.rocm.gpu.4", owners: ["module:rocm", "oncall:distributed"] },
|
||||
]}
|
||||
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
|
||||
31
.github/workflows/periodic.yml
vendored
31
.github/workflows/periodic.yml
vendored
@ -204,6 +204,37 @@ jobs:
|
||||
test-matrix: ${{ needs.linux-jammy-cuda13_0-py3_10-gcc11-build.outputs.test-matrix }}
|
||||
secrets: inherit
|
||||
|
||||
linux-jammy-rocm-py3_10-build:
|
||||
name: linux-jammy-rocm-py3.10
|
||||
uses: ./.github/workflows/_linux-build.yml
|
||||
needs: get-label-type
|
||||
with:
|
||||
runner_prefix: "${{ 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: "distributed", shard: 1, num_shards: 3, runner: "linux.rocm.gpu.mi250.4", owners: ["module:rocm", "oncall:distributed"] },
|
||||
{ config: "distributed", shard: 2, num_shards: 3, runner: "linux.rocm.gpu.mi250.4", owners: ["module:rocm", "oncall:distributed"] },
|
||||
{ config: "distributed", shard: 3, num_shards: 3, runner: "linux.rocm.gpu.mi250.4", owners: ["module:rocm", "oncall:distributed"] },
|
||||
]}
|
||||
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: ${{ needs.linux-jammy-rocm-py3_10-build.outputs.test-matrix }}
|
||||
secrets: inherit
|
||||
|
||||
linux-jammy-cuda12_8-py3-gcc11-slow-gradcheck-build:
|
||||
name: linux-jammy-cuda12.8-py3-gcc11-slow-gradcheck
|
||||
uses: ./.github/workflows/_linux-build.yml
|
||||
|
||||
8
.github/workflows/pull.yml
vendored
8
.github/workflows/pull.yml
vendored
@ -66,10 +66,10 @@ jobs:
|
||||
{ config: "default", shard: 5, num_shards: 5, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
|
||||
{ config: "docs_test", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
|
||||
{ config: "jit_legacy", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
|
||||
{ config: "backwards_compat", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.c7i.2xlarge" },
|
||||
{ config: "backwards_compat", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
|
||||
{ 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: "${{ needs.get-label-type.outputs.label-type }}linux.c7i.2xlarge" },
|
||||
{ config: "numpy_2_x", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
|
||||
]}
|
||||
secrets: inherit
|
||||
|
||||
@ -167,8 +167,8 @@ jobs:
|
||||
docker-image-name: ci-image:pytorch-linux-jammy-py3-clang12-onnx
|
||||
test-matrix: |
|
||||
{ include: [
|
||||
{ config: "default", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.c7i.2xlarge" },
|
||||
{ config: "default", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.c7i.2xlarge" },
|
||||
{ config: "default", shard: 1, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
|
||||
{ config: "default", shard: 2, num_shards: 2, runner: "${{ needs.get-label-type.outputs.label-type }}linux.2xlarge" },
|
||||
]}
|
||||
secrets: inherit
|
||||
|
||||
|
||||
2
.github/workflows/rocm.yml
vendored
2
.github/workflows/rocm.yml
vendored
@ -3,13 +3,13 @@ name: rocm
|
||||
on:
|
||||
push:
|
||||
branches:
|
||||
- main
|
||||
- release/*
|
||||
tags:
|
||||
- ciflow/rocm/*
|
||||
workflow_dispatch:
|
||||
schedule:
|
||||
- cron: 29 8 * * * # about 1:29am PDT
|
||||
- cron: 0 * * * *
|
||||
|
||||
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' }}
|
||||
|
||||
1
.github/workflows/upload-test-stats.yml
vendored
1
.github/workflows/upload-test-stats.yml
vendored
@ -6,7 +6,6 @@ on:
|
||||
- pull
|
||||
- trunk
|
||||
- periodic
|
||||
- periodic-rocm-mi200
|
||||
- periodic-rocm-mi300
|
||||
- inductor
|
||||
- unstable
|
||||
|
||||
20
.github/workflows/xpu.yml
vendored
20
.github/workflows/xpu.yml
vendored
@ -59,18 +59,14 @@ jobs:
|
||||
runner: linux.c7i.12xlarge
|
||||
test-matrix: |
|
||||
{ include: [
|
||||
{ config: "default", shard: 1, num_shards: 12, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 2, num_shards: 12, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 3, num_shards: 12, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 4, num_shards: 12, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 5, num_shards: 12, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 6, num_shards: 12, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 7, num_shards: 12, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 8, num_shards: 12, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 9, num_shards: 12, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 10, num_shards: 12, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 11, num_shards: 12, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 12, num_shards: 12, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 1, num_shards: 8, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 2, num_shards: 8, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 3, num_shards: 8, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 4, num_shards: 8, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 5, num_shards: 8, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 6, num_shards: 8, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 7, num_shards: 8, runner: "linux.idc.xpu" },
|
||||
{ config: "default", shard: 8, num_shards: 8, runner: "linux.idc.xpu" },
|
||||
]}
|
||||
secrets: inherit
|
||||
|
||||
|
||||
2
.gitignore
vendored
2
.gitignore
vendored
@ -143,7 +143,6 @@ scripts/release_notes/*.json
|
||||
sccache-stats*.json
|
||||
lint.json
|
||||
merge_record.json
|
||||
.github/scripts/nightly_source_matrix.json
|
||||
|
||||
# These files get copied over on invoking setup.py
|
||||
torchgen/packaged/*
|
||||
@ -398,4 +397,3 @@ CLAUDE.local.md
|
||||
/test_*.py
|
||||
/debug_*.py
|
||||
CLAUDE_CONTEXT/
|
||||
/.claude/settings.local.json
|
||||
|
||||
@ -121,6 +121,94 @@ command = [
|
||||
]
|
||||
is_formatter = true
|
||||
|
||||
[[linter]]
|
||||
code = 'MYPY'
|
||||
include_patterns = [
|
||||
'setup.py',
|
||||
'functorch/dim/**/*.py',
|
||||
'torch/**/*.py',
|
||||
'torch/**/*.pyi',
|
||||
'caffe2/**/*.py',
|
||||
'caffe2/**/*.pyi',
|
||||
'test/test_bundled_images.py',
|
||||
'test/test_bundled_inputs.py',
|
||||
'test/test_complex.py',
|
||||
'test/test_datapipe.py',
|
||||
'test/test_futures.py',
|
||||
'test/test_numpy_interop.py',
|
||||
'test/test_torch.py',
|
||||
'test/test_type_hints.py',
|
||||
'test/test_type_info.py',
|
||||
'test/test_utils.py',
|
||||
]
|
||||
exclude_patterns = [
|
||||
'**/fb/**',
|
||||
]
|
||||
command = [
|
||||
'python3',
|
||||
'tools/linter/adapters/mypy_linter.py',
|
||||
'--config=mypy.ini',
|
||||
'--',
|
||||
'@{{PATHSFILE}}'
|
||||
]
|
||||
init_command = [
|
||||
'python3',
|
||||
'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"',
|
||||
'expecttest==0.3.0',
|
||||
'mypy==1.16.0',
|
||||
'sympy==1.13.3',
|
||||
'types-requests==2.27.25',
|
||||
'types-pyyaml==6.0.2',
|
||||
'types-tabulate==0.8.8',
|
||||
'types-protobuf==5.29.1.20250403',
|
||||
'types-setuptools==79.0.0.20250422',
|
||||
'types-jinja2==2.11.9',
|
||||
'types-colorama==0.4.6',
|
||||
'filelock==3.18.0',
|
||||
'junitparser==2.1.1',
|
||||
'rich==14.1.0',
|
||||
'pyyaml==6.0.2',
|
||||
'optree==0.13.0',
|
||||
'dataclasses-json==0.6.7',
|
||||
'pandas==2.2.3',
|
||||
]
|
||||
|
||||
[[linter]]
|
||||
code = 'MYPYSTRICT'
|
||||
include_patterns = [
|
||||
'.github/**/*.py',
|
||||
'benchmarks/instruction_counts/**/*.py',
|
||||
'tools/**/*.py',
|
||||
'torchgen/**/*.py',
|
||||
'torch/utils/_pytree.py',
|
||||
'torch/utils/_cxx_pytree.py',
|
||||
'torch/utils/benchmark/utils/common.py',
|
||||
'torch/utils/benchmark/utils/timer.py',
|
||||
'torch/utils/benchmark/utils/valgrind_wrapper/**/*.py',
|
||||
]
|
||||
exclude_patterns = [
|
||||
# (linbinyu) copied from internal repo
|
||||
'**/fb/**',
|
||||
'tools/code_analyzer/gen_operators_yaml.py',
|
||||
'tools/dynamo/verify_dynamo.py',
|
||||
'tools/gen_vulkan_spv.py',
|
||||
'tools/test/gen_operators_yaml_test.py',
|
||||
'tools/test/gen_oplist_test.py',
|
||||
'tools/test/test_selective_build.py',
|
||||
'tools/experimental/torchfuzz/**',
|
||||
]
|
||||
command = [
|
||||
'python3',
|
||||
'tools/linter/adapters/mypy_linter.py',
|
||||
'--config=mypy-strict.ini',
|
||||
'--code=MYPYSTRICT',
|
||||
'--',
|
||||
'@{{PATHSFILE}}'
|
||||
]
|
||||
|
||||
|
||||
[[linter]]
|
||||
code = 'PYREFLY'
|
||||
@ -142,7 +230,6 @@ init_command = [
|
||||
'python3',
|
||||
'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"',
|
||||
'expecttest==0.3.0',
|
||||
'pyrefly==0.36.2',
|
||||
|
||||
@ -374,7 +374,7 @@ cmake_dependent_option(
|
||||
"Build the lazy Torchscript backend, not compatible with mobile builds" ON
|
||||
"NOT INTERN_BUILD_MOBILE" OFF)
|
||||
cmake_dependent_option(BUILD_FUNCTORCH "Build Functorch" ON "BUILD_PYTHON" OFF)
|
||||
cmake_dependent_option(BUILD_BUNDLE_PTXAS "Bundle PTX into torch/bin folder"
|
||||
cmake_dependent_option(BUILD_BUNDLE_PTXAS "Bundle PTX into torch/bin fodler"
|
||||
OFF "USE_CUDA" OFF)
|
||||
cmake_dependent_option(USE_KLEIDIAI "Use KleidiAI for the ARM CPU & AARCH64 architecture." ON
|
||||
"CPU_AARCH64" OFF)
|
||||
|
||||
@ -11,6 +11,7 @@ aspects of contributing to PyTorch.
|
||||
<!-- toc -->
|
||||
|
||||
- [Developing PyTorch](#developing-pytorch)
|
||||
- [Setup the development environment](#setup-the-development-environment)
|
||||
- [Tips and Debugging](#tips-and-debugging)
|
||||
- [Nightly Checkout & Pull](#nightly-checkout--pull)
|
||||
- [Codebase structure](#codebase-structure)
|
||||
@ -66,6 +67,23 @@ aspects of contributing to PyTorch.
|
||||
|
||||
Follow the instructions for [installing PyTorch from source](https://github.com/pytorch/pytorch#from-source). If you get stuck when developing PyTorch on your machine, check out the [tips and debugging](#tips-and-debugging) section below for common solutions.
|
||||
|
||||
### Setup the development environment
|
||||
|
||||
First, you need to [fork the PyTorch project on GitHub](https://github.com/pytorch/pytorch/fork) and follow the instructions at [Connecting to GitHub with SSH](https://docs.github.com/en/authentication/connecting-to-github-with-ssh) to setup your SSH authentication credentials.
|
||||
|
||||
Then clone the PyTorch project and setup the development environment:
|
||||
|
||||
```bash
|
||||
git clone git@github.com:<USERNAME>/pytorch.git
|
||||
cd pytorch
|
||||
git remote add upstream git@github.com:pytorch/pytorch.git
|
||||
|
||||
make setup-env
|
||||
# Or run `make setup-env-cuda` for pre-built CUDA binaries
|
||||
# Or run `make setup-env-rocm` for pre-built ROCm binaries
|
||||
source venv/bin/activate # or `. .\venv\Scripts\activate` on Windows
|
||||
```
|
||||
|
||||
### Tips and Debugging
|
||||
|
||||
* If you want to have no-op incremental rebuilds (which are fast), see [Make no-op build fast](#make-no-op-build-fast) below.
|
||||
|
||||
@ -1,4 +1,4 @@
|
||||
|
||||
|
||||
|
||||
--------------------------------------------------------------------------------
|
||||
|
||||
@ -72,7 +72,7 @@ Elaborating Further:
|
||||
|
||||
If you use NumPy, then you have used Tensors (a.k.a. ndarray).
|
||||
|
||||
|
||||
|
||||
|
||||
PyTorch provides Tensors that can live either on the CPU or the GPU and accelerates the
|
||||
computation by a huge amount.
|
||||
@ -99,7 +99,7 @@ from several research papers on this topic, as well as current and past work suc
|
||||
While this technique is not unique to PyTorch, it's one of the fastest implementations of it to date.
|
||||
You get the best of speed and flexibility for your crazy research.
|
||||
|
||||
|
||||
|
||||
|
||||
### Python First
|
||||
|
||||
|
||||
@ -31,9 +31,9 @@ Be careful when running untrusted models. This classification includes models cr
|
||||
|
||||
**Prefer to execute untrusted models within a secure, isolated environment such as a sandbox** (e.g., containers, virtual machines). This helps protect your system from potentially malicious code. You can find further details and instructions in [this page](https://developers.google.com/code-sandboxing).
|
||||
|
||||
**Be mindful of risky model formats**. Give preference to share and load weights with the appropriate format for your use case. [safetensors](https://huggingface.co/docs/safetensors/en/index) gives the most safety but is the most restricted in what it supports. [`torch.load`](https://pytorch.org/docs/stable/generated/torch.load.html#torch.load) has a significantly larger surface of attack but is more flexible in what it can serialize. See the documentation for more details.
|
||||
**Be mindful of risky model formats**. Give preference to share and load weights with the appropriate format for your use case. [safetensors](https://huggingface.co/docs/safetensors/en/index) gives the most safety but is the most restricted in what it supports. [`torch.load`](https://pytorch.org/docs/stable/generated/torch.load.html#torch.load) with `weights_only=True` is also secure to our knowledge even though it offers significantly larger surface of attack. Loading un-trusted checkpoint with `weights_only=False` MUST never be done.
|
||||
|
||||
|
||||
Even for more secure serialization formats, unexpected inputs to the downstream system can cause diverse security threats (e.g. denial of service, out of bound reads/writes) and thus we recommend extensive validation of any untrusted inputs.
|
||||
|
||||
Important Note: The trustworthiness of a model is not binary. You must always determine the proper level of caution depending on the specific model and how it matches your use case and risk tolerance.
|
||||
|
||||
|
||||
@ -38,7 +38,7 @@ set_bool(AT_HIPSPARSELT_ENABLED CAFFE2_USE_HIPSPARSELT)
|
||||
|
||||
configure_file(Config.h.in "${CMAKE_CURRENT_SOURCE_DIR}/Config.h")
|
||||
# TODO: Do not generate CUDAConfig.h for ROCm BUILDS
|
||||
# At the moment, `jit_macros.h` include CUDAConfig.h for both CUDA and HIP builds
|
||||
# At the moment, `jit_macors.h` include CUDAConfig.h for both CUDA and HIP builds
|
||||
if(USE_CUDA OR USE_ROCM)
|
||||
configure_file(cuda/CUDAConfig.h.in "${CMAKE_CURRENT_SOURCE_DIR}/cuda/CUDAConfig.h")
|
||||
endif()
|
||||
@ -260,7 +260,7 @@ IF(USE_FBGEMM_GENAI)
|
||||
if(USE_CUDA)
|
||||
# To avoid increasing the build time/binary size unnecessarily, use an allow-list of kernels to build.
|
||||
# If you want to integrate a kernel from FBGEMM into torch, you have to add it here.
|
||||
set(FBGEMM_CUTLASS_KERNELS_REGEX ".*(mx8mx8bf16_grouped|f4f4bf16_grouped).*")
|
||||
set(FBGEMM_CUTLASS_KERNELS_REGEX ".*mx8mx8bf16_grouped.*")
|
||||
file(GLOB_RECURSE fbgemm_genai_native_cuda_cu
|
||||
"${FBGEMM_GENAI_SRCS}/cutlass_extensions/*.cu"
|
||||
"${FBGEMM_GENAI_SRCS}/cutlass_extensions/**/*.cu")
|
||||
@ -291,7 +291,6 @@ IF(USE_FBGEMM_GENAI)
|
||||
|
||||
set(fbgemm_genai_cuh
|
||||
"${FBGEMM_GENAI_SRCS}/cutlass_extensions/mx8mx8bf16_grouped/"
|
||||
"${FBGEMM_GENAI_SRCS}/cutlass_extensions/f4f4bf16_grouped/"
|
||||
"${FBGEMM_GENAI_SRCS}/"
|
||||
)
|
||||
|
||||
|
||||
@ -181,7 +181,7 @@ c10::intrusive_ptr<c10::TensorImpl> CPUGeneratorImpl::get_state() const {
|
||||
static const size_t size = sizeof(CPUGeneratorImplState);
|
||||
static_assert(std::is_standard_layout_v<CPUGeneratorImplState>, "CPUGeneratorImplState is not a PODType");
|
||||
|
||||
auto state_tensor = at::detail::empty_cpu({static_cast<int64_t>(size)}, ScalarType::Byte, std::nullopt, std::nullopt, std::nullopt, std::nullopt);
|
||||
auto state_tensor = at::detail::empty_cpu({(int64_t)size}, ScalarType::Byte, std::nullopt, std::nullopt, std::nullopt, std::nullopt);
|
||||
auto rng_state = state_tensor.data_ptr();
|
||||
|
||||
// accumulate generator data to be copied into byte tensor
|
||||
|
||||
@ -223,7 +223,7 @@ void Context::setSDPPriorityOrder(const std::vector<int64_t>& order) {
|
||||
"setSDPPriority order expected ", sdp_priority_order.size() - 1, " but got ",
|
||||
at::num_sdp_backends, " unique backends specified in priority order.");
|
||||
for (uint32_t i = 0; i < order.size(); i++) {
|
||||
sdp_priority_order[i] = static_cast<at::SDPBackend>(order[i]);
|
||||
sdp_priority_order[i] = (at::SDPBackend) order[i];
|
||||
}
|
||||
}
|
||||
|
||||
@ -825,14 +825,6 @@ void Context::setDisplayVmapFallbackWarnings(bool enabled) {
|
||||
display_vmap_fallback_warnings_ = enabled;
|
||||
}
|
||||
|
||||
bool Context::warnOnAccumulateGradStreamMismatch() const {
|
||||
return warn_on_accumulate_grad_stream_mismatch_;
|
||||
}
|
||||
|
||||
void Context::setWarnOnAccumulateGradStreamMismatch(bool enabled) {
|
||||
warn_on_accumulate_grad_stream_mismatch_ = enabled;
|
||||
}
|
||||
|
||||
bool Context::isDefaultMobileCPUAllocatorSet() {
|
||||
return prev_allocator_ptr_ != nullptr;
|
||||
}
|
||||
|
||||
@ -404,9 +404,6 @@ class TORCH_API Context {
|
||||
void setDisplayVmapFallbackWarnings(bool enabled);
|
||||
bool areVmapFallbackWarningsEnabled() const;
|
||||
|
||||
void setWarnOnAccumulateGradStreamMismatch(bool enabled);
|
||||
bool warnOnAccumulateGradStreamMismatch() const;
|
||||
|
||||
bool isDefaultMobileCPUAllocatorSet();
|
||||
void setDefaultMobileCPUAllocator();
|
||||
void unsetDefaultMobileCPUAllocator();
|
||||
@ -497,7 +494,6 @@ class TORCH_API Context {
|
||||
bool release_original_weights = false;
|
||||
#endif
|
||||
bool display_vmap_fallback_warnings_ = false;
|
||||
bool warn_on_accumulate_grad_stream_mismatch_ = true;
|
||||
std::atomic<at::QEngine> quantized_engine = at::QEngine::NoQEngine;
|
||||
bool enable_sparse_tensor_invariant_checks = false;
|
||||
bool allow_fp16_reduction_cpu = false;
|
||||
|
||||
@ -197,7 +197,6 @@ inline at::ScalarType scalar_type(at::ScalarType s) {
|
||||
/* 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: \
|
||||
@ -209,7 +208,6 @@ inline at::ScalarType scalar_type(at::ScalarType s) {
|
||||
toString(_st), \
|
||||
"'"); \
|
||||
} \
|
||||
C10_DIAGNOSTIC_POP() \
|
||||
}()
|
||||
|
||||
#define AT_DISPATCH_CASE_FLOATING_TYPES(...) \
|
||||
|
||||
@ -122,7 +122,7 @@ void FunctionalTensorWrapper::freeze_storage() const {
|
||||
// | have their own storages, but backends like functorch |
|
||||
// \/ are allowed to re-alias underneath the pass \/
|
||||
// . - - - - - - - - - - - - - . . - - - - - - - - - - - - - - - .
|
||||
// | underlying_storage | | underlying_storage |
|
||||
// | underyling_storage | | underyling_storage |
|
||||
// . - - - - - - - - - - - - - . . - - - - - - - - - - - - - - - .
|
||||
//
|
||||
// This constructor is only used by view ops.
|
||||
|
||||
@ -252,13 +252,13 @@ MapAllocator::MapAllocator(WithFd /*unused*/, std::string_view filename, int fd,
|
||||
if (!(flags_ & ALLOCATOR_MAPPED_FROMFD)) {
|
||||
if (flags_ & ALLOCATOR_MAPPED_SHARED) {
|
||||
// NOLINTNEXTLINE(bugprone-assignment-in-if-condition)
|
||||
if ((fd = open(filename_.c_str(), flags, static_cast<mode_t>(0600))) == -1) {
|
||||
if ((fd = open(filename_.c_str(), flags, (mode_t)0600)) == -1) {
|
||||
TORCH_CHECK(false, "unable to open file <", filename_, "> in read-write mode: ", c10::utils::str_error(errno), " (", errno, ")");
|
||||
}
|
||||
} else if (flags_ & ALLOCATOR_MAPPED_SHAREDMEM) {
|
||||
#ifdef HAVE_SHM_OPEN
|
||||
// NOLINTNEXTLINE(bugprone-assignment-in-if-condition)
|
||||
if((fd = shm_open(filename_.c_str(), flags, static_cast<mode_t>(0600))) == -1) {
|
||||
if((fd = shm_open(filename_.c_str(), flags, (mode_t)0600)) == -1) {
|
||||
TORCH_CHECK(false, "unable to open shared memory object <", filename_, "> in read-write mode: ", c10::utils::str_error(errno), " (", errno, ")");
|
||||
}
|
||||
#else
|
||||
@ -503,7 +503,7 @@ RefcountedMapAllocator::RefcountedMapAllocator(WithFd /*unused*/, const char *fi
|
||||
|
||||
void RefcountedMapAllocator::initializeAlloc() {
|
||||
TORCH_CHECK(base_ptr_, "base_ptr_ is null");
|
||||
MapInfo *map_info = static_cast<MapInfo*>(base_ptr_);
|
||||
MapInfo *map_info = (MapInfo*)base_ptr_;
|
||||
|
||||
#ifdef _WIN32
|
||||
ReleaseContext* r_ctx = new ReleaseContext;
|
||||
@ -539,7 +539,7 @@ void RefcountedMapAllocator::close() {
|
||||
}
|
||||
#else /* _WIN32 */
|
||||
|
||||
MapInfo *info = static_cast<MapInfo*>(data);
|
||||
MapInfo *info = (MapInfo*)(data);
|
||||
if (--info->refcount == 0) {
|
||||
#ifdef HAVE_SHM_UNLINK
|
||||
if (shm_unlink(filename_.c_str()) == -1) {
|
||||
|
||||
@ -862,7 +862,7 @@ void TensorIteratorBase::narrow(int dim, int64_t start, int64_t size) {
|
||||
shape_[dim] = size;
|
||||
view_offsets_[dim] += start;
|
||||
for (auto& op : operands_) {
|
||||
op.data = (static_cast<char*>(op.data)) + op.stride_bytes[dim] * start;
|
||||
op.data = ((char*)op.data) + op.stride_bytes[dim] * start;
|
||||
}
|
||||
if (size == 1 && !is_reduction_) {
|
||||
coalesce_dimensions();
|
||||
@ -873,7 +873,7 @@ void TensorIteratorBase::select_all_keeping_dim(int start_dim, IntArrayRef indic
|
||||
TORCH_INTERNAL_ASSERT(start_dim <= ndim());
|
||||
for (const auto i : c10::irange(start_dim, ndim())) {
|
||||
for (auto& op : operands_) {
|
||||
op.data = (static_cast<char*>(op.data)) + op.stride_bytes[i] * indices[i - start_dim];
|
||||
op.data = ((char*)op.data) + op.stride_bytes[i] * indices[i - start_dim];
|
||||
}
|
||||
shape_[i] = 1;
|
||||
}
|
||||
@ -1534,7 +1534,7 @@ void TensorIteratorBase::build(TensorIteratorConfig& config) {
|
||||
|
||||
// XLA and lazy tensors don't have storage, so they don't have an underlying data pointer.
|
||||
// Nothing beyond this point is important for meta functions, so it's fine to exit early here.
|
||||
// Extend the condition to MAIA tensors as MAIA tensors also don't have storage.
|
||||
// Extend the condition to MAIA tesnors as MAIA tensors also don't have storage.
|
||||
if (privateuse1_without_storage ||
|
||||
common_device_.type() == DeviceType::XLA ||
|
||||
common_device_.type() == DeviceType::IPU ||
|
||||
|
||||
@ -41,7 +41,7 @@ inline void serial_for_each(
|
||||
IntArrayRef strides,
|
||||
char** base_ptrs,
|
||||
size_t ntensors,
|
||||
TensorIteratorBase::loop2d_t loop,
|
||||
typename TensorIteratorBase::loop2d_t loop,
|
||||
Range range) {
|
||||
const auto ndim = shape.size();
|
||||
TORCH_INTERNAL_ASSERT_DEBUG_ONLY(
|
||||
|
||||
@ -94,11 +94,11 @@ struct PinnedReserveSegment {
|
||||
struct TORCH_API HostStats {
|
||||
// COUNT: total allocations (active)
|
||||
Stat active_requests;
|
||||
// SUM: bytes allocated/reserved by this memory allocator. (active)
|
||||
// SUM: bytes allocated/reserved by this memory alocator. (active)
|
||||
Stat active_bytes;
|
||||
// COUNT: total allocations (active + free)
|
||||
Stat allocations;
|
||||
// SUM: bytes allocated/reserved by this memory allocator. This accounts
|
||||
// SUM: bytes allocated/reserved by this memory alocator. This accounts
|
||||
// for both free and in-use blocks.
|
||||
Stat allocated_bytes;
|
||||
|
||||
@ -127,7 +127,7 @@ struct alignas(hardware_destructive_interference_size) HostStatsStaged {
|
||||
// COUNT: total allocations (active + free)
|
||||
// LOCK: access to this stat is protected by the allocator's blocks_mutex_
|
||||
Stat allocations;
|
||||
// SUM: bytes allocated/reserved by this memory allocator. This accounts
|
||||
// SUM: bytes allocated/reserved by this memory alocator. This accounts
|
||||
// for both free and in-use blocks.
|
||||
Stat allocated_bytes;
|
||||
// COUNT: number of allocations per bucket (active)
|
||||
@ -455,7 +455,7 @@ struct CachingHostAllocatorImpl {
|
||||
}
|
||||
|
||||
void resetAccumulatedStats() {
|
||||
// Resetting accumulated memory stats requires concurrently holding both the
|
||||
// Reseting accumulated memory stats requires concurrently holding both the
|
||||
// free list mutexes and the blocks mutex. Previously, this was only done in
|
||||
// empty_cache function.
|
||||
for (size_t i = 0; i < free_list_.size(); ++i) {
|
||||
@ -482,7 +482,7 @@ struct CachingHostAllocatorImpl {
|
||||
}
|
||||
|
||||
void resetPeakStats() {
|
||||
// Resetting peak memory stats requires concurrently holding both the
|
||||
// Reseting peak memory stats requires concurrently holding both the
|
||||
// free list mutexes and the blocks mutex. Previously, this was only done in
|
||||
// empty_cache function.
|
||||
for (size_t i = 0; i < free_list_.size(); ++i) {
|
||||
@ -677,8 +677,8 @@ struct CachingHostAllocatorImpl {
|
||||
// size. This allows us to quickly find a free block of the right size.
|
||||
// We use deque to store per size free list and guard the list with its own
|
||||
// mutex.
|
||||
alignas(hardware_destructive_interference_size) std::vector<FreeBlockList<B>>
|
||||
free_list_{MAX_SIZE_INDEX};
|
||||
alignas(hardware_destructive_interference_size) std::vector<FreeBlockList<B>> free_list_ =
|
||||
std::vector<FreeBlockList<B>>(MAX_SIZE_INDEX);
|
||||
|
||||
alignas(hardware_destructive_interference_size) std::mutex events_mutex_;
|
||||
std::deque<std::pair<E, B*>> events_; // event queue paired with block
|
||||
|
||||
@ -190,14 +190,12 @@ class IListRef;
|
||||
* it to a function (e.g. `ImplT::<dispatch-function>(this_)`).
|
||||
*/
|
||||
#define TORCH_ILISTREF_UNWRAP(TAG, BODY) \
|
||||
C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wswitch-enum") \
|
||||
switch (TAG) { \
|
||||
TORCH_ILISTREF_FORALL_TAGS(TORCH_ILISTREF_UNWRAP_CASE, BODY) \
|
||||
break; \
|
||||
default: \
|
||||
TORCH_INTERNAL_ASSERT(false, "invalid IListRef tag."); \
|
||||
} \
|
||||
C10_DIAGNOSTIC_POP()
|
||||
}
|
||||
|
||||
enum class IListRefTag {
|
||||
#define DEFINE_TAG(tag, ...) tag,
|
||||
|
||||
@ -56,7 +56,7 @@ C10_HOST_DEVICE inline T uniform_int_full_range(V val) {
|
||||
* in this overloaded version
|
||||
*/
|
||||
template <typename T, typename V>
|
||||
C10_HOST_DEVICE inline std::enable_if_t<!std::is_floating_point_v<T>, T>uniform_int(V val) {
|
||||
C10_HOST_DEVICE inline std::enable_if_t<!(std::is_floating_point_v<T>), T>uniform_int(V val) {
|
||||
if constexpr (std::is_same_v<T, bool>) {
|
||||
return static_cast<bool>(val & 1);
|
||||
} else if constexpr (std::is_same_v<T, int64_t>) {
|
||||
|
||||
@ -114,25 +114,25 @@ inline typename remove_symint<T>::type unpackSymInt(T x) {
|
||||
}
|
||||
|
||||
template <>
|
||||
inline remove_symint<c10::SymInt>::type unpackSymInt(c10::SymInt x) {
|
||||
inline typename remove_symint<c10::SymInt>::type unpackSymInt(c10::SymInt x) {
|
||||
return x.guard_int(__FILE__, __LINE__);
|
||||
}
|
||||
|
||||
template <>
|
||||
inline remove_symint<c10::SymIntArrayRef>::type unpackSymInt(
|
||||
inline typename remove_symint<c10::SymIntArrayRef>::type unpackSymInt(
|
||||
c10::SymIntArrayRef x) {
|
||||
return C10_AS_INTARRAYREF_SLOW(x);
|
||||
}
|
||||
|
||||
template <>
|
||||
inline remove_symint<std::optional<c10::SymInt>>::type unpackSymInt(
|
||||
inline typename remove_symint<std::optional<c10::SymInt>>::type unpackSymInt(
|
||||
std::optional<c10::SymInt> x) {
|
||||
return x.has_value() ? std::make_optional(x->guard_int(__FILE__, __LINE__))
|
||||
: std::nullopt;
|
||||
}
|
||||
|
||||
template <>
|
||||
inline remove_symint<at::OptionalSymIntArrayRef>::type unpackSymInt(
|
||||
inline typename remove_symint<at::OptionalSymIntArrayRef>::type unpackSymInt(
|
||||
at::OptionalSymIntArrayRef x) {
|
||||
return x.has_value() ? std::make_optional(C10_AS_INTARRAYREF_SLOW(*x))
|
||||
: std::nullopt;
|
||||
|
||||
@ -631,8 +631,8 @@ call_functor_with_args_from_stack_(
|
||||
Stack* stack,
|
||||
std::index_sequence<ivalue_arg_indices...> /*unused*/,
|
||||
guts::typelist::typelist<ArgTypes...>* /*unused*/) {
|
||||
(void)stack; // when sizeof...(ivalue_arg_indices) == 0, this argument would
|
||||
// be unused and we have to silence the compiler warning.
|
||||
(void)(stack); // when sizeof...(ivalue_arg_indices) == 0, this argument would
|
||||
// be unused and we have to silence the compiler warning.
|
||||
|
||||
// We're explicitly filtering out DispatchKeySet from the argument list.
|
||||
// Some kernels take a DispatchKeySet as their first argument in order to
|
||||
|
||||
@ -148,7 +148,7 @@ struct TORCH_API ClassType : public NamedType {
|
||||
|
||||
void checkNotExist(const std::string& name, const std::string& what) const;
|
||||
|
||||
// Attributes are stored in a specific slot at runtime for efficiency.
|
||||
// Attributes are stored in a specific slot at runtime for effiency.
|
||||
// When emitting instructions we specify the slot so that attribute access is
|
||||
// a constant lookup
|
||||
std::optional<size_t> findAttributeSlot(const std::string& name) const {
|
||||
@ -412,7 +412,7 @@ struct TORCH_API ClassType : public NamedType {
|
||||
// Holds method attributes
|
||||
std::weak_ptr<CompilationUnit> compilation_unit_;
|
||||
|
||||
// Holds all attributes, attribute details are found on ClassAttribute
|
||||
// Holds all atrributes, attribute details are found on ClassAttribute
|
||||
std::vector<ClassAttribute> attributes_;
|
||||
// Construct mirroring attributes_, only around due to the fact that `containedTypes()` method returns an ArrayRef.
|
||||
// Never fill this without using the appropriate provideNewClassAttribute method
|
||||
|
||||
@ -537,7 +537,7 @@ int64_t Dispatcher::sequenceNumberForRunningRecordFunction(DispatchKey dispatchK
|
||||
|
||||
// Note: this records a sequence number for both Autograd keys, and for
|
||||
// non-Autograd keys where the dispatchKeySet still contains an autograd key.
|
||||
// This means that we might collect the same sequence number two different
|
||||
// This means that we might collect the same sequence nubmer two different
|
||||
// events if they all occurred above Autograd and still had the Autograd
|
||||
// dispatch key in the dispatch key set.
|
||||
// However, this usually doesn't happen: normally the first call will
|
||||
|
||||
@ -585,7 +585,7 @@ class TORCH_API OperatorHandle {
|
||||
|
||||
// We need to store this iterator in order to make
|
||||
// Dispatcher::cleanup() fast -- it runs a lot on program
|
||||
// termination (and presumably library unloading).
|
||||
// termination (and presuambly library unloading).
|
||||
std::list<Dispatcher::OperatorDef>::iterator operatorIterator_;
|
||||
};
|
||||
|
||||
|
||||
@ -365,7 +365,7 @@ std::pair<const AnnotatedKernel&, const char*> OperatorEntry::computeDispatchTab
|
||||
// For autograd keys, we only use kernel from CompositeImplicitAutograd when there's no direct registration
|
||||
// to its corresponding backend key or CompositeExplicitAutograd. See Note [CompositeExplicitAutograd and CompositeImplicitAutograd].
|
||||
// For AutogradOther, we eagerly return ambiguousAutogradOtherKernel() if there's registration to any of
|
||||
// its backends and ask backend extender to request a dedicated Autograd key for the backend.
|
||||
// its backends and ask backend extender to request a decicated Autograd key for the backend.
|
||||
// See Note [Ambiguity in AutogradOther kernel] for more details.
|
||||
// A CompositeExplicitAutograd kernel prevents CompositeImplicitAutograd kernel being used for Autograd keys, but it doesn't
|
||||
// cause confusion for AutogradOther. It's pretty straightforward to use Autograd (if available)
|
||||
|
||||
@ -18,7 +18,6 @@ struct TORCH_API EnumType : public NamedType {
|
||||
TypePtr value,
|
||||
std::vector<EnumNameValue> enum_names_values,
|
||||
std::weak_ptr<::torch::jit::CompilationUnit> cu) {
|
||||
C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wswitch-enum")
|
||||
switch (value->kind()) {
|
||||
case TypeKind::IntType:
|
||||
case TypeKind::FloatType:
|
||||
@ -35,7 +34,6 @@ struct TORCH_API EnumType : public NamedType {
|
||||
value->str(),
|
||||
"', only int, float and string are supported");
|
||||
}
|
||||
C10_DIAGNOSTIC_POP()
|
||||
}
|
||||
|
||||
std::string str() const override {
|
||||
|
||||
@ -261,7 +261,7 @@ std::ostream& operator<<(std::ostream& out, const FunctionSchema& schema) {
|
||||
//
|
||||
// There are 2 cases
|
||||
// 1. something like 'aten::items.str(Dict(str, t) self) -> ((str, t)[])'.
|
||||
// without the extra parenthesis, the c++ scheme parser can not parse it.
|
||||
// without the extra parenthesis, the c++ schem parser can not parse it.
|
||||
// 2. something like '-> ((str, str))'. Need extra parenthesis so the return
|
||||
// type is a single tuple rather than two strings.
|
||||
// PR (https://github.com/pytorch/pytorch/pull/23204) has more context about
|
||||
|
||||
@ -601,8 +601,8 @@ std::ostream& IValue::repr(
|
||||
double d = v.toDouble();
|
||||
int c = std::fpclassify(d);
|
||||
if ((c == FP_NORMAL || c == FP_ZERO ) && std::abs(d) < 1e10) {
|
||||
int64_t i = static_cast<int64_t>(d);
|
||||
if (static_cast<double>(i) == d) {
|
||||
int64_t i = int64_t(d);
|
||||
if (double(i) == d) {
|
||||
// -0.0 (signed zero) needs to be parsed as -0.
|
||||
if (i == 0 && std::signbit(d)) {
|
||||
return out << "-" << i << ".";
|
||||
@ -799,8 +799,8 @@ std::ostream& operator<<(std::ostream & out, const IValue & v) {
|
||||
double d = v.toDouble();
|
||||
int c = std::fpclassify(d);
|
||||
if (c == FP_NORMAL || c == FP_ZERO) {
|
||||
int64_t i = static_cast<int64_t>(d);
|
||||
if (static_cast<double>(i) == d) {
|
||||
int64_t i = int64_t(d);
|
||||
if (double(i) == d) {
|
||||
return out << i << ".";
|
||||
}
|
||||
}
|
||||
|
||||
@ -1176,7 +1176,7 @@ struct TORCH_API IValue final {
|
||||
using HashIdentityIValueMap =
|
||||
std::unordered_map<IValue, IValue, HashIdentityIValue, CompIdentityIValues>;
|
||||
|
||||
// Checks if this and rhs has a subvalues in common.
|
||||
// Chechs if this and rhs has a subvalues in common.
|
||||
// [t1,t2] and [t2, t3] returns true.
|
||||
bool overlaps(const IValue& rhs) const;
|
||||
|
||||
|
||||
@ -1501,7 +1501,7 @@ struct C10_EXPORT ivalue::Object final : c10::intrusive_ptr_target {
|
||||
// However, the CompilationUnit holds ownership of the type's graphs, so
|
||||
// inserting a constant object into a Graph would create a reference cycle if
|
||||
// that constant object held a shared_ptr to its CU. For these objects we
|
||||
// instantiate them with non-owning references to its CU
|
||||
// instatiate them with non-owning references to its CU
|
||||
Object(WeakOrStrongTypePtr type, size_t numSlots) : type_(std::move(type)) {
|
||||
slots_.resize(numSlots);
|
||||
}
|
||||
|
||||
@ -41,7 +41,7 @@ void standardizeVectorForUnion(std::vector<TypePtr>* to_flatten);
|
||||
inline bool is_contiguous_strides(
|
||||
const IntArrayRef sizes,
|
||||
const IntArrayRef strides) {
|
||||
size_t n_dim = sizes.size();
|
||||
int n_dim = static_cast<int>(sizes.size());
|
||||
if (n_dim == 0) {
|
||||
return true;
|
||||
}
|
||||
@ -50,7 +50,7 @@ inline bool is_contiguous_strides(
|
||||
return false;
|
||||
}
|
||||
|
||||
for (int i = static_cast<int>(n_dim) - 2; i >= 0; i--) {
|
||||
for (int i = n_dim - 2; i >= 0; i--) {
|
||||
if (strides[i] != strides[i + 1] * sizes[i + 1]) {
|
||||
return false;
|
||||
}
|
||||
@ -373,7 +373,7 @@ struct TORCH_API SymbolicShape {
|
||||
// Unranked shape constructor.
|
||||
SymbolicShape() : dims_(std::nullopt) {}
|
||||
|
||||
// Known rank but unknown dimensions.
|
||||
// Known rank but unknown dimentions.
|
||||
SymbolicShape(std::optional<size_t> rank) : dims_(std::nullopt) {
|
||||
if(!rank) {
|
||||
return;
|
||||
@ -884,9 +884,9 @@ struct TORCH_API ListType
|
||||
|
||||
// global singleton
|
||||
// Given an inner type T and an identifier,
|
||||
// this function will return the global singleton type pointer
|
||||
// this function wil return the global singleton type pointer
|
||||
// the type List<T>.
|
||||
// The extra "identifier" argument is needed because we have multiple container types
|
||||
// The extra "identifier" argument is needed beccause we have multiple container types
|
||||
// that all re-use this function (List<T>, array<T, N>, etc.)
|
||||
static TypePtr get(const std::string& identifier, TypePtr inner);
|
||||
|
||||
@ -922,7 +922,6 @@ struct TORCH_API DictType : public SharedType {
|
||||
if (auto dyn = key->castRaw<DynamicType>()) {
|
||||
kind = dyn->dynamicKind();
|
||||
}
|
||||
C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wswitch-enum")
|
||||
switch (kind) {
|
||||
case TypeKind::AnyType:
|
||||
case TypeKind::IntType:
|
||||
@ -939,7 +938,6 @@ struct TORCH_API DictType : public SharedType {
|
||||
key->str(),
|
||||
"', only int, float, complex, Tensor, device and string keys are supported");
|
||||
}
|
||||
C10_DIAGNOSTIC_POP()
|
||||
}
|
||||
|
||||
// aligned with the format in FunctionSchema
|
||||
@ -2373,7 +2371,7 @@ private:
|
||||
};
|
||||
|
||||
template<>
|
||||
inline detail::CastReturnType<NamedType>::type Type::cast<NamedType>() {
|
||||
inline typename detail::CastReturnType<NamedType>::type Type::cast<NamedType>() {
|
||||
if (kind() == TypeKind::TupleType || kind() == TypeKind::FunctionType ||
|
||||
kind() == TypeKind::ClassType || kind() == TypeKind::InterfaceType) {
|
||||
return std::static_pointer_cast<NamedType>(static_cast<NamedType *>(this)->shared_from_this());
|
||||
@ -2382,7 +2380,7 @@ inline detail::CastReturnType<NamedType>::type Type::cast<NamedType>() {
|
||||
}
|
||||
|
||||
template<>
|
||||
inline detail::CastConstReturnType<NamedType>::type Type::cast<NamedType>() const {
|
||||
inline typename detail::CastConstReturnType<NamedType>::type Type::cast<NamedType>() const {
|
||||
if (kind() == TypeKind::TupleType || kind() == TypeKind::FunctionType ||
|
||||
kind() == TypeKind::ClassType || kind() == TypeKind::InterfaceType) {
|
||||
return std::static_pointer_cast<const NamedType>(static_cast<const NamedType *>(this)->shared_from_this());
|
||||
|
||||
@ -21,7 +21,7 @@ namespace c10 {
|
||||
|
||||
namespace detail {
|
||||
// The first argument of the schema might be of type DispatchKeySet, in which case we remove it.
|
||||
// We do this because every argument in a function schema is expected to be convertible
|
||||
// We do this because every argument in a function schema is expected to be convertable
|
||||
// to an ivalue, but DispatchKeySet is not a type we want the jit to be aware of.
|
||||
// See Note [Plumbing Keys Through The Dispatcher]
|
||||
template<class KernelFunctor>
|
||||
|
||||
@ -251,7 +251,7 @@ TEST(OperatorRegistrationTest, whenRegisteringCPUTensorType_thenCanOnlyCallUnbox
|
||||
callOpUnboxedWithPrecomputedDispatchKeySet<void, Tensor>(*op, c10::DispatchKeySet(c10::DispatchKey::CPU), dummyTensor(c10::DispatchKey::CUDA));
|
||||
EXPECT_TRUE(called_kernel_cpu);
|
||||
|
||||
// Ensure that dispatch key from tensor is not used here.
|
||||
// Ensure that disptach key from tensor is not used here.
|
||||
called_kernel_cpu = false;
|
||||
expectThrows<c10::Error>([&] {
|
||||
callOpUnboxedWithPrecomputedDispatchKeySet<void, Tensor>(*op, c10::DispatchKeySet(c10::DispatchKey::CUDA), dummyTensor(c10::DispatchKey::CPU));
|
||||
|
||||
@ -172,7 +172,7 @@ VaryingShape<Stride> TensorType::computeStrideProps(
|
||||
// The logic below follows what TensorIterator uses in its logic:
|
||||
// 1. Fast_set_up is the short-cut to identify a. channels_last and
|
||||
// b. contiguous format, which is what we have in the below logic.
|
||||
// 2. In more general cases, it does best effort to preserve permutatoin.
|
||||
// 2. In more generla cases, it does best effort to preserve permutatoin.
|
||||
if (is_channels_last_strides_2d(sizes, strides) || is_channels_last_strides_3d(sizes, strides)) {
|
||||
// case 1.a. short cut channels last
|
||||
std::iota(stride_indices.rbegin() + 1, stride_indices.rend() - 1, 2);
|
||||
|
||||
@ -19,13 +19,6 @@ inline namespace CPU_CAPABILITY {
|
||||
#error "Big endian is not supported."
|
||||
#endif
|
||||
|
||||
// GCC does not properly optimize bf16 operators
|
||||
#if defined(__ARM_FEATURE_BF16) && (__clang_major__ >= 19)
|
||||
#define BF16_ARITHMETIC_SUPPORTED() 1
|
||||
#else
|
||||
#define BF16_ARITHMETIC_SUPPORTED() 0
|
||||
#endif
|
||||
|
||||
// Unlike the float16_t family of types, bfloat16_t is not available
|
||||
// when we're not targeting bfloat16 hardware support on some
|
||||
// platforms (but not Mac, so we have to be careful not to shadow the
|
||||
@ -359,35 +352,18 @@ class Vectorized<c10::BFloat16> : public Vectorized16<
|
||||
other, &Vectorized<float>::name); \
|
||||
}
|
||||
|
||||
DEFINE_UNARY_ELEMENTWISE_FUNC_VIA_FLOAT_METHOD(abs)
|
||||
Vectorized frac() const;
|
||||
DEFINE_UNARY_ELEMENTWISE_FUNC_VIA_FLOAT_METHOD(trunc)
|
||||
DEFINE_UNARY_ELEMENTWISE_FUNC_VIA_FLOAT_METHOD(sqrt)
|
||||
|
||||
#ifdef __ARM_FEATURE_BF16
|
||||
// Flip sign bit
|
||||
Vectorized<c10::BFloat16> neg() const {
|
||||
return vreinterpretq_bf16_s16(vreinterpretq_s16_bf16(values) ^ (-32768));
|
||||
return -values;
|
||||
}
|
||||
// Fast reciprocal is fine because we are truncating results
|
||||
Vectorized<c10::BFloat16> reciprocal() const {
|
||||
auto x = vcvtq_low_f32_bf16(values);
|
||||
auto y = vcvtq_high_f32_bf16(values);
|
||||
x = vrecpeq_f32(x);
|
||||
y = vrecpeq_f32(y);
|
||||
return vcvtq_high_bf16_f32(vcvtq_low_bf16_f32(x), y);
|
||||
return 1.0f / values;
|
||||
}
|
||||
// Clearing the sign bit
|
||||
Vectorized<c10::BFloat16> abs() const {
|
||||
return vreinterpretq_bf16_u16(vreinterpretq_u16_bf16(values) & 0x7FFF);
|
||||
}
|
||||
#else
|
||||
DEFINE_UNARY_ELEMENTWISE_FUNC_VIA_FLOAT_METHOD(abs)
|
||||
DEFINE_UNARY_ELEMENTWISE_FUNC_VIA_FLOAT_METHOD(neg)
|
||||
DEFINE_UNARY_ELEMENTWISE_FUNC_VIA_FLOAT_METHOD(reciprocal)
|
||||
#endif
|
||||
|
||||
// These functions are optimized on clang-21+
|
||||
#if BF16_ARITHMETIC_SUPPORTED() && (__clang_major__ >= 21)
|
||||
Vectorized<c10::BFloat16> operator==(
|
||||
const Vectorized<c10::BFloat16>& other) const {
|
||||
return values == other.values;
|
||||
@ -418,6 +394,8 @@ class Vectorized<c10::BFloat16> : public Vectorized16<
|
||||
return values >= other.values;
|
||||
}
|
||||
#else
|
||||
DEFINE_UNARY_ELEMENTWISE_FUNC_VIA_FLOAT_METHOD(neg)
|
||||
DEFINE_UNARY_ELEMENTWISE_FUNC_VIA_FLOAT_METHOD(reciprocal)
|
||||
DEFINE_BINARY_COMPARISON_OPERATOR_VIA_FLOAT_METHOD(operator==)
|
||||
DEFINE_BINARY_COMPARISON_OPERATOR_VIA_FLOAT_METHOD(operator!=)
|
||||
DEFINE_BINARY_COMPARISON_OPERATOR_VIA_FLOAT_METHOD(operator<)
|
||||
@ -473,7 +451,7 @@ template <>
|
||||
Vectorized<c10::BFloat16> inline operator+(
|
||||
const Vectorized<c10::BFloat16>& a,
|
||||
const Vectorized<c10::BFloat16>& b) {
|
||||
#if BF16_ARITHMETIC_SUPPORTED()
|
||||
#ifdef __ARM_FEATURE_BF16
|
||||
bfloat16x8_t x = a;
|
||||
bfloat16x8_t y = b;
|
||||
return x + y;
|
||||
@ -486,7 +464,7 @@ template <>
|
||||
Vectorized<c10::BFloat16> inline operator-(
|
||||
const Vectorized<c10::BFloat16>& a,
|
||||
const Vectorized<c10::BFloat16>& b) {
|
||||
#if BF16_ARITHMETIC_SUPPORTED()
|
||||
#ifdef __ARM_FEATURE_BF16
|
||||
bfloat16x8_t x = a;
|
||||
bfloat16x8_t y = b;
|
||||
return x - y;
|
||||
@ -499,7 +477,7 @@ template <>
|
||||
Vectorized<c10::BFloat16> inline operator*(
|
||||
const Vectorized<c10::BFloat16>& a,
|
||||
const Vectorized<c10::BFloat16>& b) {
|
||||
#if BF16_ARITHMETIC_SUPPORTED()
|
||||
#ifdef __ARM_FEATURE_BF16
|
||||
bfloat16x8_t x = a;
|
||||
bfloat16x8_t y = b;
|
||||
return x * y;
|
||||
@ -512,7 +490,7 @@ template <>
|
||||
Vectorized<c10::BFloat16> inline operator/(
|
||||
const Vectorized<c10::BFloat16>& a,
|
||||
const Vectorized<c10::BFloat16>& b) {
|
||||
#if BF16_ARITHMETIC_SUPPORTED()
|
||||
#ifdef __ARM_FEATURE_BF16
|
||||
bfloat16x8_t x = a;
|
||||
bfloat16x8_t y = b;
|
||||
return x / y;
|
||||
@ -629,7 +607,7 @@ Vectorized<c10::BFloat16> inline fmadd(
|
||||
const Vectorized<c10::BFloat16>& a,
|
||||
const Vectorized<c10::BFloat16>& b,
|
||||
const Vectorized<c10::BFloat16>& c) {
|
||||
#if BF16_ARITHMETIC_SUPPORTED()
|
||||
#ifdef __ARM_FEATURE_BF16
|
||||
bfloat16x8_t x = a;
|
||||
bfloat16x8_t y = b;
|
||||
bfloat16x8_t z = c;
|
||||
@ -649,7 +627,7 @@ Vectorized<c10::BFloat16> inline fnmadd(
|
||||
const Vectorized<c10::BFloat16>& a,
|
||||
const Vectorized<c10::BFloat16>& b,
|
||||
const Vectorized<c10::BFloat16>& c) {
|
||||
#if BF16_ARITHMETIC_SUPPORTED()
|
||||
#ifdef __ARM_FEATURE_BF16
|
||||
bfloat16x8_t x = a;
|
||||
bfloat16x8_t y = b;
|
||||
bfloat16x8_t z = c;
|
||||
@ -665,7 +643,7 @@ Vectorized<c10::BFloat16> inline fmsub(
|
||||
const Vectorized<c10::BFloat16>& a,
|
||||
const Vectorized<c10::BFloat16>& b,
|
||||
const Vectorized<c10::BFloat16>& c) {
|
||||
#if BF16_ARITHMETIC_SUPPORTED()
|
||||
#ifdef __ARM_FEATURE_BF16
|
||||
bfloat16x8_t x = a;
|
||||
bfloat16x8_t y = b;
|
||||
bfloat16x8_t z = c;
|
||||
@ -681,7 +659,7 @@ Vectorized<c10::BFloat16> inline fnmsub(
|
||||
const Vectorized<c10::BFloat16>& a,
|
||||
const Vectorized<c10::BFloat16>& b,
|
||||
const Vectorized<c10::BFloat16>& c) {
|
||||
#if BF16_ARITHMETIC_SUPPORTED()
|
||||
#ifdef __ARM_FEATURE_BF16
|
||||
bfloat16x8_t x = a;
|
||||
bfloat16x8_t y = b;
|
||||
bfloat16x8_t z = c;
|
||||
|
||||
@ -6,9 +6,9 @@ namespace at::vec {
|
||||
inline namespace CPU_CAPABILITY {
|
||||
#if (defined(__aarch64__) && !defined(CPU_CAPABILITY_SVE256))
|
||||
|
||||
// Enable auto-vectorization for clang-17+
|
||||
// Enable auto-vectorization for GCC-13+ and clang-17+
|
||||
// GCC-12 has a bug: gcc.gnu.org/bugzilla/show_bug.cgi?id=117001
|
||||
#if defined(__clang__) && (__clang_major__ >= 17)
|
||||
#if __GNUC__ > 12 || (defined(__clang__) && (__clang_major__ >= 17))
|
||||
|
||||
template <typename from_type, typename to_type>
|
||||
inline void convertImpl(
|
||||
@ -21,46 +21,12 @@ inline void convertImpl(
|
||||
}
|
||||
}
|
||||
|
||||
template <typename to_type>
|
||||
inline void convertFromBool(
|
||||
const bool* __restrict src,
|
||||
to_type* __restrict dst,
|
||||
int64_t n) {
|
||||
const uint8_t* srcPtr = reinterpret_cast<const uint8_t*>(src);
|
||||
uint64_t len = static_cast<uint64_t>(n);
|
||||
for (uint64_t i = 0; i < len; i++) {
|
||||
dst[i] = srcPtr[i] != 0 ? static_cast<to_type>(1) : static_cast<to_type>(0);
|
||||
}
|
||||
}
|
||||
|
||||
template <typename from_type>
|
||||
inline void convertToBool(
|
||||
const from_type* __restrict src,
|
||||
bool* __restrict dst,
|
||||
int64_t n) {
|
||||
uint8_t* dstPtr = reinterpret_cast<uint8_t*>(dst);
|
||||
uint64_t len = static_cast<uint64_t>(n);
|
||||
for (uint64_t i = 0; i < len; i++) {
|
||||
dstPtr[i] = src[i] != static_cast<from_type>(0) ? 1 : 0;
|
||||
}
|
||||
}
|
||||
|
||||
#define CONVERT_TEMPLATE(from_type, to_type) \
|
||||
template <> \
|
||||
inline void convert(const from_type* src, to_type* dst, int64_t n) { \
|
||||
return convertImpl<from_type, to_type>(src, dst, n); \
|
||||
}
|
||||
|
||||
#define CONVERT_FROM_BOOL_TEMPLATE(to_type) \
|
||||
inline void convert(const bool* src, to_type* dst, int64_t n) { \
|
||||
return convertFromBool<to_type>(src, dst, n); \
|
||||
}
|
||||
|
||||
#define CONVERT_TO_BOOL_TEMPLATE(from_type) \
|
||||
inline void convert(const from_type* src, bool* dst, int64_t n) { \
|
||||
return convertToBool<from_type>(src, dst, n); \
|
||||
}
|
||||
|
||||
CONVERT_TEMPLATE(uint8_t, uint8_t)
|
||||
CONVERT_TEMPLATE(uint8_t, int8_t)
|
||||
CONVERT_TEMPLATE(uint8_t, int16_t)
|
||||
@ -68,7 +34,6 @@ CONVERT_TEMPLATE(uint8_t, int32_t)
|
||||
CONVERT_TEMPLATE(uint8_t, int64_t)
|
||||
CONVERT_TEMPLATE(uint8_t, float)
|
||||
CONVERT_TEMPLATE(uint8_t, double)
|
||||
CONVERT_TO_BOOL_TEMPLATE(uint8_t)
|
||||
CONVERT_TEMPLATE(int8_t, uint8_t)
|
||||
CONVERT_TEMPLATE(int8_t, int8_t)
|
||||
CONVERT_TEMPLATE(int8_t, int16_t)
|
||||
@ -76,7 +41,6 @@ CONVERT_TEMPLATE(int8_t, int32_t)
|
||||
CONVERT_TEMPLATE(int8_t, int64_t)
|
||||
CONVERT_TEMPLATE(int8_t, float)
|
||||
CONVERT_TEMPLATE(int8_t, double)
|
||||
CONVERT_TO_BOOL_TEMPLATE(int8_t)
|
||||
CONVERT_TEMPLATE(int16_t, uint8_t)
|
||||
CONVERT_TEMPLATE(int16_t, int8_t)
|
||||
CONVERT_TEMPLATE(int16_t, int16_t)
|
||||
@ -84,7 +48,6 @@ CONVERT_TEMPLATE(int16_t, int32_t)
|
||||
CONVERT_TEMPLATE(int16_t, int64_t)
|
||||
CONVERT_TEMPLATE(int16_t, float)
|
||||
CONVERT_TEMPLATE(int16_t, double)
|
||||
CONVERT_TO_BOOL_TEMPLATE(int16_t)
|
||||
CONVERT_TEMPLATE(int32_t, uint8_t)
|
||||
CONVERT_TEMPLATE(int32_t, int8_t)
|
||||
CONVERT_TEMPLATE(int32_t, int16_t)
|
||||
@ -92,7 +55,6 @@ CONVERT_TEMPLATE(int32_t, int32_t)
|
||||
CONVERT_TEMPLATE(int32_t, int64_t)
|
||||
CONVERT_TEMPLATE(int32_t, float)
|
||||
CONVERT_TEMPLATE(int32_t, double)
|
||||
CONVERT_TO_BOOL_TEMPLATE(int32_t)
|
||||
CONVERT_TEMPLATE(int64_t, uint8_t)
|
||||
CONVERT_TEMPLATE(int64_t, int8_t)
|
||||
CONVERT_TEMPLATE(int64_t, int16_t)
|
||||
@ -100,7 +62,6 @@ CONVERT_TEMPLATE(int64_t, int32_t)
|
||||
CONVERT_TEMPLATE(int64_t, int64_t)
|
||||
CONVERT_TEMPLATE(int64_t, float)
|
||||
CONVERT_TEMPLATE(int64_t, double)
|
||||
CONVERT_TO_BOOL_TEMPLATE(int64_t)
|
||||
CONVERT_TEMPLATE(float, uint8_t)
|
||||
CONVERT_TEMPLATE(float, int8_t)
|
||||
CONVERT_TEMPLATE(float, int16_t)
|
||||
@ -108,7 +69,6 @@ CONVERT_TEMPLATE(float, int32_t)
|
||||
CONVERT_TEMPLATE(float, int64_t)
|
||||
CONVERT_TEMPLATE(float, float)
|
||||
CONVERT_TEMPLATE(float, double)
|
||||
CONVERT_TO_BOOL_TEMPLATE(float)
|
||||
CONVERT_TEMPLATE(double, uint8_t)
|
||||
CONVERT_TEMPLATE(double, int8_t)
|
||||
CONVERT_TEMPLATE(double, int16_t)
|
||||
@ -116,80 +76,22 @@ CONVERT_TEMPLATE(double, int32_t)
|
||||
CONVERT_TEMPLATE(double, int64_t)
|
||||
CONVERT_TEMPLATE(double, float)
|
||||
CONVERT_TEMPLATE(double, double)
|
||||
CONVERT_TO_BOOL_TEMPLATE(double)
|
||||
CONVERT_FROM_BOOL_TEMPLATE(uint8_t)
|
||||
CONVERT_FROM_BOOL_TEMPLATE(int8_t)
|
||||
CONVERT_FROM_BOOL_TEMPLATE(int16_t)
|
||||
CONVERT_FROM_BOOL_TEMPLATE(int32_t)
|
||||
CONVERT_FROM_BOOL_TEMPLATE(int64_t)
|
||||
CONVERT_FROM_BOOL_TEMPLATE(float)
|
||||
CONVERT_FROM_BOOL_TEMPLATE(double)
|
||||
#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
|
||||
|
||||
#define CONVERT_FROM_FP16_TEMPLATE(to_type) \
|
||||
template <> \
|
||||
inline void convert(const at::Half* src, to_type* dst, int64_t n) { \
|
||||
const float16_t* srcPtr = reinterpret_cast<const float16_t*>(src); \
|
||||
return convertImpl<float16_t, to_type>(srcPtr, dst, n); \
|
||||
}
|
||||
|
||||
#define CONVERT_TO_FP16_TEMPLATE(from_type) \
|
||||
template <> \
|
||||
inline void convert(const from_type* src, at::Half* dst, int64_t n) { \
|
||||
float16_t* dstPtr = reinterpret_cast<float16_t*>(dst); \
|
||||
return convertImpl<from_type, float16_t>(src, dstPtr, n); \
|
||||
}
|
||||
|
||||
CONVERT_FROM_FP16_TEMPLATE(uint8_t)
|
||||
CONVERT_FROM_FP16_TEMPLATE(int8_t)
|
||||
CONVERT_FROM_FP16_TEMPLATE(int16_t)
|
||||
CONVERT_FROM_FP16_TEMPLATE(int32_t)
|
||||
CONVERT_FROM_FP16_TEMPLATE(int64_t)
|
||||
CONVERT_FROM_FP16_TEMPLATE(float16_t)
|
||||
CONVERT_FROM_FP16_TEMPLATE(float)
|
||||
CONVERT_FROM_FP16_TEMPLATE(double)
|
||||
CONVERT_TO_FP16_TEMPLATE(uint8_t)
|
||||
CONVERT_TO_FP16_TEMPLATE(int8_t)
|
||||
CONVERT_TO_FP16_TEMPLATE(int16_t)
|
||||
CONVERT_TO_FP16_TEMPLATE(int32_t)
|
||||
CONVERT_TO_FP16_TEMPLATE(int64_t)
|
||||
CONVERT_TO_FP16_TEMPLATE(float)
|
||||
CONVERT_TO_FP16_TEMPLATE(double)
|
||||
|
||||
inline void convertBoolToFp16Impl(
|
||||
const bool* __restrict src,
|
||||
at::Half* __restrict dst,
|
||||
int64_t n) {
|
||||
const uint8_t* srcPtr = reinterpret_cast<const uint8_t*>(src);
|
||||
float16_t* dstPtr = reinterpret_cast<float16_t*>(dst);
|
||||
uint64_t len = static_cast<uint64_t>(n);
|
||||
for (uint64_t i = 0; i < len; i++) {
|
||||
dstPtr[i] = srcPtr[i] != 0 ? 1.0 : 0;
|
||||
}
|
||||
}
|
||||
|
||||
template <>
|
||||
inline void convert(const bool* src, at::Half* dst, int64_t n) {
|
||||
return convertBoolToFp16Impl(src, dst, n);
|
||||
}
|
||||
|
||||
inline void convertFp16ToBoolImpl(
|
||||
const at::Half* __restrict src,
|
||||
bool* __restrict dst,
|
||||
int64_t n) {
|
||||
const float16_t* srcPtr = reinterpret_cast<const float16_t*>(src);
|
||||
uint8_t* dstPtr = reinterpret_cast<uint8_t*>(dst);
|
||||
uint64_t len = static_cast<uint64_t>(n);
|
||||
for (uint64_t i = 0; i < len; i++) {
|
||||
dstPtr[i] = srcPtr[i] != 0.0 ? 1 : 0;
|
||||
}
|
||||
}
|
||||
|
||||
template <>
|
||||
inline void convert(const at::Half* src, bool* dst, int64_t n) {
|
||||
return convertFp16ToBoolImpl(src, dst, n);
|
||||
}
|
||||
|
||||
CONVERT_TEMPLATE(float16_t, uint8_t)
|
||||
CONVERT_TEMPLATE(float16_t, int8_t)
|
||||
CONVERT_TEMPLATE(float16_t, int16_t)
|
||||
CONVERT_TEMPLATE(float16_t, int32_t)
|
||||
CONVERT_TEMPLATE(float16_t, int64_t)
|
||||
CONVERT_TEMPLATE(float16_t, float16_t)
|
||||
CONVERT_TEMPLATE(float16_t, float)
|
||||
CONVERT_TEMPLATE(float16_t, double)
|
||||
CONVERT_TEMPLATE(uint8_t, float16_t)
|
||||
CONVERT_TEMPLATE(int8_t, float16_t)
|
||||
CONVERT_TEMPLATE(int16_t, float16_t)
|
||||
CONVERT_TEMPLATE(int32_t, float16_t)
|
||||
CONVERT_TEMPLATE(int64_t, float16_t)
|
||||
CONVERT_TEMPLATE(float, float16_t)
|
||||
CONVERT_TEMPLATE(double, float16_t)
|
||||
#endif
|
||||
#ifdef __ARM_FEATURE_BF16
|
||||
CONVERT_TEMPLATE(bfloat16_t, uint8_t)
|
||||
@ -207,44 +109,6 @@ CONVERT_TEMPLATE(int32_t, bfloat16_t)
|
||||
CONVERT_TEMPLATE(int64_t, bfloat16_t)
|
||||
CONVERT_TEMPLATE(float, bfloat16_t)
|
||||
CONVERT_TEMPLATE(double, bfloat16_t)
|
||||
|
||||
inline void convertBoolToBfloat16Impl(
|
||||
const bool* __restrict src,
|
||||
c10::BFloat16* __restrict dst,
|
||||
int64_t n) {
|
||||
const uint8_t* srcPtr = reinterpret_cast<const uint8_t*>(src);
|
||||
uint16_t* dstPtr = reinterpret_cast<uint16_t*>(dst);
|
||||
uint64_t len = static_cast<uint64_t>(n);
|
||||
constexpr uint16_t kBf16One = 0x3f80; // 1.0 in bfloat16
|
||||
for (uint64_t i = 0; i < len; i++) {
|
||||
dstPtr[i] = srcPtr[i] != 0 ? kBf16One : 0;
|
||||
}
|
||||
}
|
||||
|
||||
template <>
|
||||
inline void convert(const bool* src, c10::BFloat16* dst, int64_t n) {
|
||||
return convertBoolToBfloat16Impl(src, dst, n);
|
||||
}
|
||||
|
||||
inline void convertBfloat16ToBoolImpl(
|
||||
const c10::BFloat16* __restrict src,
|
||||
bool* __restrict dst,
|
||||
int64_t n) {
|
||||
uint8_t* dstPtr = reinterpret_cast<uint8_t*>(dst);
|
||||
const uint16_t* srcPtr = reinterpret_cast<const uint16_t*>(src);
|
||||
uint64_t len = static_cast<uint64_t>(n);
|
||||
for (uint64_t i = 0; i < len; i++) {
|
||||
// Check if all non-sign bits are 0
|
||||
bool isBf16Zero = (srcPtr[i] & 0x7fff) == 0;
|
||||
dstPtr[i] = isBf16Zero ? 0 : 1;
|
||||
}
|
||||
}
|
||||
|
||||
template <>
|
||||
inline void convert(const c10::BFloat16* src, bool* dst, int64_t n) {
|
||||
return convertBfloat16ToBoolImpl(src, dst, n);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
#endif
|
||||
|
||||
@ -309,7 +309,7 @@ class Vectorized<float> {
|
||||
DEFINE_SLEEF_COMPATIBLE_UNARY_ELEMENTWISE_FUNC(expm1)
|
||||
// Implementation copied from Arm Optimized Routine
|
||||
// https://github.com/ARM-software/optimized-routines/blob/master/math/aarch64/advsimd/expf.c
|
||||
inline Vectorized<float> vexpq_f32_u20() const {
|
||||
Vectorized<float> exp_u20() const {
|
||||
// bail out to sleef if it's a special case:
|
||||
// i.e. there's an input s.t. |input| > 87.3....
|
||||
const float32x4_t special_bound = vdupq_n_f32(0x1.5d5e2ap+6f);
|
||||
@ -348,9 +348,6 @@ class Vectorized<float> {
|
||||
|
||||
return vfmaq_f32(scale, poly, scale);
|
||||
}
|
||||
Vectorized<float> exp_u20() const {
|
||||
return vexpq_f32_u20();
|
||||
}
|
||||
Vectorized<float> fexp_u20() const {
|
||||
return exp_u20();
|
||||
}
|
||||
@ -637,7 +634,8 @@ inline Vectorized<float> Vectorized<float>::erf() const {
|
||||
// - exp(- x * x)
|
||||
auto pow_2 = (*this) * (*this);
|
||||
auto neg_pow_2 = pow_2 ^ neg_zero_vec;
|
||||
auto tmp4 = neg_pow_2.vexpq_f32_u20();
|
||||
auto tmp4 = neg_pow_2.map(
|
||||
std::exp); // This can be swapped for a faster implementation of exp.
|
||||
auto tmp5 = tmp4 ^ neg_zero_vec;
|
||||
// erf(x) = sign(x) * (1 - r * t * exp(- x * x))
|
||||
auto tmp6 = t * tmp5;
|
||||
|
||||
@ -234,7 +234,7 @@ class Vectorized<c10::Half> : public Vectorized16<
|
||||
vshlq_u16(vandq_u16(is_zero_vec, vdupq_n_u16(1)), shift);
|
||||
return vaddvq_u16(bits_vec);
|
||||
#else // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
|
||||
// use known working implementation.
|
||||
// use known working implmentation.
|
||||
__at_align__ value_type tmp[size()];
|
||||
store(tmp);
|
||||
int mask = 0;
|
||||
|
||||
@ -1740,7 +1740,7 @@ Vectorized<int16_t> inline shift_256_16(
|
||||
|
||||
// Control masks for shuffle operation, treating 256 bits as an
|
||||
// array of 16-bit elements, and considering pairs of neighboring
|
||||
// elements. Specifically, a mask named "ctl_M_N" (M,N in [0,1], and
|
||||
// elements. Specifially, a mask named "ctl_M_N" (M,N in [0,1], and
|
||||
// M!=N) is set so that shuffle will move element with index M from
|
||||
// input pair into element with index N in output pair, and element
|
||||
// with index M in output pair will be set to all 0s.
|
||||
@ -1875,7 +1875,7 @@ Vectorized<T> inline shift_256_8(
|
||||
|
||||
// Control masks for shuffle operation, treating 256 bits as an
|
||||
// array of 8-bit elements, and considering quadruples of
|
||||
// neighboring elements. Specifically, a mask named "ctl_M_N" (M,N
|
||||
// neighboring elements. Specifially, a mask named "ctl_M_N" (M,N
|
||||
// in [0,1,2,3], and M!=N) is set so that shuffle will move element
|
||||
// with index M from input quadruple into element with index N in
|
||||
// output quadruple, and other elements in output quadruple will be
|
||||
|
||||
@ -514,7 +514,7 @@ struct Vectorized<c10::qint8> : public Vectorizedqi {
|
||||
|
||||
using float_vec_return_type = std::array<Vectorized<float>, kFloatNumVecs>;
|
||||
using int_vec_return_type = std::array<Vectorized<c10::qint32>, kIntNumVecs>;
|
||||
using value_type = c10::qint8::underlying;
|
||||
using value_type = typename c10::qint8::underlying;
|
||||
|
||||
public:
|
||||
using Vectorizedqi::Vectorizedqi;
|
||||
@ -727,7 +727,7 @@ struct Vectorized<c10::quint8> : public Vectorizedqi {
|
||||
|
||||
using float_vec_return_type = std::array<Vectorized<float>, kFloatNumVecs>;
|
||||
using int_vec_return_type = std::array<Vectorized<c10::qint32>, kIntNumVecs>;
|
||||
using value_type = c10::quint8::underlying;
|
||||
using value_type = typename c10::quint8::underlying;
|
||||
|
||||
public:
|
||||
using Vectorizedqi::Vectorizedqi;
|
||||
|
||||
@ -143,7 +143,7 @@ class Vectorized<double> {
|
||||
const Vectorized<double>& a,
|
||||
const Vectorized<double>& b,
|
||||
const Vectorized<double>& mask) {
|
||||
// the mask used here returned by comparison of vec256
|
||||
// the mask used here returned by comparision of vec256
|
||||
|
||||
return {
|
||||
vec_sel(a._vec0, b._vec0, mask._vecb0),
|
||||
|
||||
@ -142,7 +142,7 @@ class Vectorized<float> {
|
||||
const Vectorized<float>& a,
|
||||
const Vectorized<float>& b,
|
||||
const Vectorized<float>& mask) {
|
||||
// the mask used here returned by comparison of vec256
|
||||
// the mask used here returned by comparision of vec256
|
||||
// assuming this we can use the same mask directly with vec_sel
|
||||
return {
|
||||
vec_sel(a._vec0, b._vec0, mask._vecb0),
|
||||
|
||||
@ -202,7 +202,7 @@ class Vectorized<int16_t> {
|
||||
const Vectorized<int16_t>& a,
|
||||
const Vectorized<int16_t>& b,
|
||||
const Vectorized<int16_t>& mask) {
|
||||
// the mask used here returned by comparison of vec256
|
||||
// the mask used here returned by comparision of vec256
|
||||
// assuming this we can use the same mask directly with vec_sel
|
||||
// warning intel style mask will not work properly
|
||||
return {
|
||||
|
||||
@ -155,7 +155,7 @@ class Vectorized<int32_t> {
|
||||
const Vectorized<int32_t>& a,
|
||||
const Vectorized<int32_t>& b,
|
||||
const Vectorized<int32_t>& mask) {
|
||||
// the mask used here returned by comparison of vec256
|
||||
// the mask used here returned by comparision of vec256
|
||||
// assuming this we can use the same mask directly with vec_sel
|
||||
// warning intel style mask will not work properly
|
||||
return {
|
||||
|
||||
@ -119,7 +119,7 @@ class Vectorized<int64_t> {
|
||||
const Vectorized<int64_t>& a,
|
||||
const Vectorized<int64_t>& b,
|
||||
const Vectorized<int64_t>& mask) {
|
||||
// the mask used here returned by comparison of vec256
|
||||
// the mask used here returned by comparision of vec256
|
||||
|
||||
return {
|
||||
vec_sel(a._vec0, b._vec0, mask._vecb0),
|
||||
|
||||
@ -397,7 +397,7 @@ inline Vectorized<bool> operator&&(
|
||||
const __m512i* other_ = reinterpret_cast<const __m512i*>(other.as_bytes());
|
||||
__m512i out = _mm512_and_si512(*self_, *other_);
|
||||
Vectorized<bool> ret;
|
||||
// We do not have a constructor that takes __m512i, so we need to memcpy
|
||||
// We do not have a constructer that takes __m512i, so we need to memcpy
|
||||
std::memcpy(ret, &out, ret.size() * sizeof(bool));
|
||||
return ret;
|
||||
}
|
||||
|
||||
@ -1852,7 +1852,7 @@ Vectorized<T> inline shift_512_8(
|
||||
|
||||
// Control masks for shuffle operation, treating 512 bits as an
|
||||
// array of 8-bit elements, and considering pairs of neighboring
|
||||
// elements. Specifically, a mask named "ctl_M_N" (M,N in [0,1], and
|
||||
// elements. Specifially, a mask named "ctl_M_N" (M,N in [0,1], and
|
||||
// M!=N) is set so that shuffle will move element with index M from
|
||||
// input pair into element with index N in output pair, and element
|
||||
// with index M in output pair will be set to all 0s.
|
||||
|
||||
@ -567,7 +567,7 @@ struct Vectorized<c10::qint8> : public Vectorizedqi {
|
||||
|
||||
using float_vec_return_type = std::array<Vectorized<float>, 4>;
|
||||
using int_vec_return_type = std::array<Vectorized<c10::qint32>, 4>;
|
||||
using value_type = c10::qint8::underlying;
|
||||
using value_type = typename c10::qint8::underlying;
|
||||
|
||||
public:
|
||||
using Vectorizedqi::Vectorizedqi;
|
||||
@ -804,7 +804,7 @@ struct Vectorized<c10::quint8> : public Vectorizedqi {
|
||||
|
||||
using float_vec_return_type = std::array<Vectorized<float>, 4>;
|
||||
using int_vec_return_type = std::array<Vectorized<c10::qint32>, 4>;
|
||||
using value_type = c10::quint8::underlying;
|
||||
using value_type = typename c10::quint8::underlying;
|
||||
|
||||
public:
|
||||
using Vectorizedqi::Vectorizedqi;
|
||||
|
||||
@ -634,7 +634,7 @@ struct Vectorized {
|
||||
}
|
||||
Vectorized<T> neg() const {
|
||||
// NB: the trailing return type is needed because we need to coerce the
|
||||
// return value back to T in the case of unary operator- incurring a
|
||||
// return value back to T in the case of unary operator- incuring a
|
||||
// promotion
|
||||
return map([](T x) -> T { return -x; });
|
||||
}
|
||||
@ -672,7 +672,7 @@ struct Vectorized {
|
||||
return map(std::sqrt);
|
||||
}
|
||||
Vectorized<T> reciprocal() const {
|
||||
return map([](T x) { return (T)1 / x; });
|
||||
return map([](T x) { return (T)(1) / x; });
|
||||
}
|
||||
Vectorized<T> rsqrt() const {
|
||||
return map([](T x) { return (T)1 / std::sqrt(x); });
|
||||
|
||||
@ -46,7 +46,7 @@ inline void vrsqrt(scalar_t* out, scalar_t* in, int64_t size) {
|
||||
parallel_for(0, size, 2048, [out, in](int64_t begin, int64_t end) {
|
||||
map(
|
||||
[](const Vectorized<scalar_t>& x) {
|
||||
return Vectorized<scalar_t>((scalar_t)1) / x.sqrt();
|
||||
return Vectorized<scalar_t>((scalar_t)(1)) / x.sqrt();
|
||||
},
|
||||
out + begin,
|
||||
in + begin,
|
||||
|
||||
@ -1958,7 +1958,7 @@ void scaled_gemm(
|
||||
ScalarType result_dtype,
|
||||
bool use_fast_accum,
|
||||
const std::optional<Tensor>& alpha) {
|
||||
// Note: see `cublasCommonArgs` for various non-intuitive manipulations
|
||||
// Note: see `cublasCommonArgs` for various non-intuitive manupulations
|
||||
// of input arguments to this function.
|
||||
const auto computeType = CUBLAS_COMPUTE_32F;
|
||||
const auto scaleType = CUDA_R_32F;
|
||||
|
||||
@ -2,10 +2,10 @@
|
||||
|
||||
#include <ATen/cuda/ATenCUDAGeneral.h>
|
||||
#include <ATen/cuda/CUDAContext.h>
|
||||
#include <ATen/cuda/Exceptions.h>
|
||||
#include <c10/core/impl/GPUTrace.h>
|
||||
#include <c10/cuda/CUDAGuard.h>
|
||||
#include <c10/cuda/CUDAStream.h>
|
||||
#include <c10/cuda/CUDAGuard.h>
|
||||
#include <ATen/cuda/Exceptions.h>
|
||||
#include <c10/util/Exception.h>
|
||||
|
||||
#include <cuda_runtime_api.h>
|
||||
@ -246,79 +246,4 @@ private:
|
||||
}
|
||||
};
|
||||
|
||||
// EventPool - Thread-safe pool of CUDA events to avoid expensive cudaEventCreate
|
||||
// calls. cudaEventCreate when concurrently invoked from multiple threads can be
|
||||
// very expensive (especially on certain device/driver combinations).
|
||||
using CUDAEventPtr =
|
||||
std::unique_ptr<CUDAEvent, std::function<void(CUDAEvent*)>>;
|
||||
|
||||
class EventPool {
|
||||
public:
|
||||
EventPool() : pools_(at::cuda::device_count()) {}
|
||||
|
||||
CUDAEventPtr get(const DeviceIndex device) {
|
||||
// If the device is invalid, return a default event and no pooling
|
||||
if (device < 0 || device >= (DeviceIndex)pools_.size()) {
|
||||
auto deleter = [](CUDAEvent* event) {
|
||||
delete event;
|
||||
};
|
||||
return CUDAEventPtr(
|
||||
std::make_unique<CUDAEvent>(cudaEventDisableTiming).release(), deleter);
|
||||
}
|
||||
|
||||
auto& pool = pools_[device];
|
||||
|
||||
// Create a destructor that returns the event to the appropriate device pool
|
||||
auto destructor = [&pool](CUDAEvent* event) noexcept {
|
||||
if (event != nullptr) {
|
||||
std::lock_guard<std::mutex> lock(pool.mutex_);
|
||||
pool.event_pool_.emplace_back(event);
|
||||
}
|
||||
};
|
||||
|
||||
{
|
||||
std::lock_guard<std::mutex> lock(pool.mutex_);
|
||||
if (!pool.event_pool_.empty()) {
|
||||
auto event = std::move(pool.event_pool_.back());
|
||||
pool.event_pool_.pop_back();
|
||||
return CUDAEventPtr(event.release(), destructor);
|
||||
}
|
||||
}
|
||||
|
||||
return CUDAEventPtr(
|
||||
std::make_unique<CUDAEvent>(cudaEventDisableTiming).release(),
|
||||
destructor);
|
||||
}
|
||||
|
||||
void empty_cache() {
|
||||
for (auto& pool : pools_) {
|
||||
std::lock_guard<std::mutex> lock(pool.mutex_);
|
||||
pool.event_pool_.clear();
|
||||
}
|
||||
}
|
||||
|
||||
void init_num_events(const size_t num_events) {
|
||||
for (DeviceIndex device_idx = 0; device_idx < at::cuda::device_count(); ++device_idx) {
|
||||
CUDAGuard device_guard(device_idx);
|
||||
std::vector<CUDAEventPtr> temp_events;
|
||||
temp_events.reserve(num_events);
|
||||
for (size_t i = 0; i < num_events; ++i) {
|
||||
auto event = get(device_idx);
|
||||
// Record the event to ensure it's properly initialized
|
||||
event->record();
|
||||
temp_events.emplace_back(std::move(event));
|
||||
}
|
||||
// Events will be returned to pool when temp_events is destroyed
|
||||
}
|
||||
}
|
||||
|
||||
private:
|
||||
struct alignas(64) PerDevicePool {
|
||||
alignas(64) std::mutex mutex_;
|
||||
std::vector<std::unique_ptr<CUDAEvent>> event_pool_;
|
||||
};
|
||||
|
||||
std::vector<PerDevicePool> pools_;
|
||||
};
|
||||
|
||||
} // namespace at::cuda
|
||||
|
||||
Some files were not shown because too many files have changed in this diff Show More
Reference in New Issue
Block a user