test/inductor/test_flex_attention.py | 5 5 + 0 - 0 !
torch/_inductor/kernel/flex_attention.py | 11 5 + 6 - 0 !
2 files changed, 10 insertions(+), 6 deletions(-)

 [patch 1/5] update

[ghstack-poisoned]

aten/src/ATen/native/quantized/cpu/qnnpack/CMakeLists.txt | 11 11 + 0 - 0 !
cmake/Dependencies.cmake | 41 36 + 5 - 0 !
cmake/External/nnpack.cmake | 7 7 + 0 - 0 !
cmake/ProtoBuf.cmake | 9 8 + 1 - 0 !
test/custom_backend/CMakeLists.txt | 2 1 + 1 - 0 !
test/custom_operator/CMakeLists.txt | 2 1 + 1 - 0 !
test/edge/CMakeLists.txt | 2 1 + 1 - 0 !
test/jit_hooks/CMakeLists.txt | 2 1 + 1 - 0 !
test/mobile/custom_build/CMakeLists.txt | 2 1 + 1 - 0 !
test/mobile/lightweight_dispatch/CMakeLists.txt | 2 1 + 1 - 0 !
10 files changed, 68 insertions(+), 12 deletions(-)

 [patch] [cherry-pick] make pytorch buildable with cmake-4 (#150460)

* [Cmake] Make PyTorch buildable by CMake-4.x (#150203)

By turning on compatibility mode for protobuf, nnpack, PSimd and FP16, ittapi, TensorPipe and Gloo
Update CMake requirements

 Revert 0ece461ccafe5649d2d0f058ff5477765fd56499 and b0901d62ae2c2e909f91401eacebf3731df20cbe to test that it actually works

TODO:
  - Update/get rid of those libraries

Fixes https://github.com/pytorch/pytorch/issues/150149

Pull Request resolved: https://github.com/pytorch/pytorch/pull/150203
Approved by: https://github.com/clee2000

(cherry picked from commit 493c7fa66f82cf781ee0f9d0cc9e305688f0a286)

* Make PyTorch buildable by CMake-4.x on s390x (#150294)

This is a continuation of
https://github.com/pytorch/pytorch/pull/150203
that fixes nightly build on s390x.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/150294
Approved by: https://github.com/malfet

(cherry picked from commit ab342d3793472c65aaa0b007ca13a98fc9206dc5)

torch/csrc/distributed/c10d/ProcessGroupGloo.cpp | 19 17 + 2 - 0 !
torch/csrc/distributed/c10d/ProcessGroupGloo.hpp | 4 2 + 2 - 0 !
2 files changed, 19 insertions(+), 4 deletions(-)

 [patch] gloo: use shared stores (#150230)

Summary:
X-link: https://github.com/facebookincubator/gloo/pull/423

This modifies `connectFullMesh` to take in a shared_ptr<IStore> instead of a reference. This is an API breaking change but fairly easy to work around.

To have backwards compatibility in PyTorch during the commit phase we add a new ifdef `GLOO_SHARED_STORE` which can provide backwards compatibility until we update the pinned Gloo version in pytorch OSS repo.

This also adds a new `wait_get` method to `IStore` which will allow us to do a more efficient operation in PyTorch TCPStore. PyTorch's `Store::get` automatically waits so we want to make sure we can avoid waiting twice to reduce network traffic.

This change will land simultaneously in PyTorch and Gloo repos.

Test Plan:
```
buck2 test //gloo/... //caffe2/caffe2/contrib/gloo:
```

CMakeLists.txt | 13 8 + 5 - 0 !
aten/src/ATen/native/quantized/cpu/qnnpack/CMakeLists.txt | 11 9 + 2 - 0 !
aten/src/ATen/native/quantized/cpu/qnnpack/deps/clog/CMakeLists.txt | 2 1 + 1 - 0 !
c10/cuda/CMakeLists.txt | 4 4 + 0 - 0 !
caffe2/CMakeLists.txt | 35 29 + 6 - 0 !
cmake/Dependencies.cmake | 46 24 + 22 - 0 !
test/cpp/c10d/CMakeLists.txt | 4 4 + 0 - 0 !
test/cpp/jit/CMakeLists.txt | 12 12 + 0 - 0 !
test/cpp/tensorexpr/CMakeLists.txt | 2 1 + 1 - 0 !
tools/setup_helpers/cmake.py | 2 1 + 1 - 0 !
torch/CMakeLists.txt | 14 12 + 2 - 0 !
11 files changed, 105 insertions(+), 40 deletions(-)

 the elegant patching work is based on the master branch
    https://github.com/pytorch/pytorch/issues/14699
  And we will be able to use that solution in the next upstream release.
  I don't want to rebase my patches back to this version, so let's go with a fast, yet dirty hack.

cmake/Modules/FindMKLDNN.cmake | 15 7 + 8 - 0 !
cmake/public/mkldnn.cmake | 2 2 + 0 - 0 !
2 files changed, 9 insertions(+), 8 deletions(-)

 enable dnnl/ideep integration

caffe2/CMakeLists.txt | 10 0 + 10 - 0 !
1 file changed, 10 deletions(-)

 patch cmake to strip third-party directory includes/checks.

caffe2/CMakeLists.txt | 1 1 + 0 - 0 !
cmake/Dependencies.cmake | 5 3 + 2 - 0 !
2 files changed, 4 insertions(+), 2 deletions(-)

 use system cpp-httplib instead.

test/run_test.py | 7 4 + 3 - 0 !
1 file changed, 4 insertions(+), 3 deletions(-)

 remove use of deprecated pkg_resources module.
 Replace with importlib.metadata from stdlib

tools/generate_torch_version.py | 12 1 + 11 - 0 !
version.txt | 2 1 + 1 - 0 !
2 files changed, 2 insertions(+), 12 deletions(-)

 make the "torch.__version__" and wheel version correct. the upstream use env vars instead.
 debian specific. Never forward this.

c10/util/Logging.cpp | 12 1 + 11 - 0 !
1 file changed, 1 insertion(+), 11 deletions(-)

 move isgooglelogginginitialized() to public api
 It was an internal function and project used hacks to reach it. Now it's part
 of the public API.

torch/distributed/elastic/__init__.py | 2 1 + 1 - 0 !
torch/distributed/elastic/events/__init__.py | 2 1 + 1 - 0 !
torch/distributed/elastic/metrics/__init__.py | 2 1 + 1 - 0 !
torch/distributed/launcher/__init__.py | 2 1 + 1 - 0 !
4 files changed, 4 insertions(+), 4 deletions(-)

 fix wrong shebang

aten/src/ATen/ATenConfig.cmake.in | 2 1 + 1 - 0 !
1 file changed, 1 insertion(+), 1 deletion(-)

 avoid including the build path in the cmake configuration file
Fixes: #1031687

cmake/Dependencies.cmake | 10 0 + 10 - 0 !
torch/CMakeLists.txt | 1 0 + 1 - 0 !
2 files changed, 11 deletions(-)

 get rid of opentelemetry dependency which is not yet in debian

caffe2/CMakeLists.txt | 2 1 + 1 - 0 !
1 file changed, 1 insertion(+), 1 deletion(-)

---

benchmarks/dynamo/Makefile | 2 1 + 1 - 0 !
pyproject.toml | 1 0 + 1 - 0 !
requirements.txt | 1 0 + 1 - 0 !
torch/jit/frontend.py | 10 2 + 8 - 0 !
4 files changed, 3 insertions(+), 11 deletions(-)

 remove dependency on astunparse
 It has been replaced with ast.unparse since Python 3.9.

CMakeLists.txt | 2 2 + 0 - 0 !
1 file changed, 2 insertions(+)

 disable -fno-math-errno on arm64 to workaroun gcc 14 bug (#1094828)

cmake/Modules/FindARM.cmake | 4 3 + 1 - 0 !
1 file changed, 3 insertions(+), 1 deletion(-)

 disable use of sve on arm64

caffe2/CMakeLists.txt | 6 1 + 5 - 0 !
cmake/public/cuda.cmake | 11 2 + 9 - 0 !
2 files changed, 3 insertions(+), 14 deletions(-)

 let it fallback to use the old nvtx3 headers. grep the code with
 TORCH_CUDA_USE_NVTX3 can you will find the simple fallback logic. The new
 header needs new nvtx packages. The old one is already in nvidia-cuda-dev.
Forward: no need.

torch/_inductor/cpp_builder.py | 7 0 + 7 - 0 !
1 file changed, 7 deletions(-)

 disable non-portable flags