Whose version is controlled by `eigen_pin.txt`, but which will be installed only if BLAS providers could not be found.
Why this is good for CI: we don't really build with Eigen ever and gitlab can be down when github is up, which causes spurious CI failures in the past, for example.
Remove eigen submodule and replace it with eigen_pin.txt
Fixes https://github.com/pytorch/pytorch/issues/108773
Pull Request resolved: https://github.com/pytorch/pytorch/pull/155955
Approved by: https://github.com/atalman
Whose version is controlled by `eigen_pin.txt`, but which will be installed only if BLAS providers could not be found.
Why this is good for CI: we don't really build with Eigen ever and gitlab can be down when github is up, which causes spurious CI failures in the past, for example.
Remove eigen submodule and replace it with eigen_pin.txt
Fixes https://github.com/pytorch/pytorch/issues/108773
Pull Request resolved: https://github.com/pytorch/pytorch/pull/155955
Approved by: https://github.com/atalman
ghstack dependencies: #155947, #155954
There is a memory layout mismatching between `fft_r2c` XPU and Inductor meta deducing.
Original `fft_r2c` Inductor meta deducing for XPU backend is aligned with CPU (fallback). This PR is to correct the Inductor meta deducing and update the torch-xpu-ops commit to [intel/torch-xpu-ops@`3a9419c`](3a9419c8bb).
The XPU implementation first performs the R2C transform on the last dimension, followed by iterative C2C transforms on the remaining dimensions.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/156048
Approved by: https://github.com/guangyey, https://github.com/etaf, https://github.com/jansel
Update the torch-xpu-ops commit to [intel/torch-xpu-ops@`a3a196`](a3a196ccdb) includes:
- Enhanced Adaptive Average Pooling 2D Backward Kernel for performance and code simplification
- Group Norm Backward Optimization with vectorization and parallel reduction
- Support CL path for MaxUnpooling2d and MaxUnpooling3d
- Rename USE_ONEMKL as USE_ONEMKL_XPU and set it as default ON
- Refactor USE_XCCL & USE_C10D_XCCL option
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154962
Approved by: https://github.com/EikanWang
Updates cpp-httplib to 0.20.1. This mostly updates OSS with a bunch of CMake, CXX compiler errors, and bugfixes from upstream. It's a header only library so should be pretty straightforward to upgrade
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154825
Approved by: https://github.com/malfet
Last update to this submodule was 3 years ago, and the API is pretty stable and this is a minor version release update. Part of a bunch of PRs to eradicate low CMake required versions
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154796
Approved by: https://github.com/jansel
Update NVTX3 submodule to 3.2.1.
* Mostly improved compiler support, Python support, and better CMake and C++ support.
* Also has a few new APIs to support fancy new features.
* This is header only library so should be an easy non-invasive change.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154797
Approved by: https://github.com/jansel
Summary:
Update fbgemm pinned version in PyTroch.
Related update in fbgemm: D74434751
Included changes:
Update fbgemm external dependencies directory in setup.py
Add DISABLE_FBGEMM_AUTOVEC flag to disable fbgemm's autovec
Test Plan: PyTorch OSS CI
Differential Revision: D75073516
Pull Request resolved: https://github.com/pytorch/pytorch/pull/153950
Approved by: https://github.com/Skylion007, https://github.com/ngimel
This enables Gloo CUDA when used with a backend that supports GPUDirect which currently is only the IBVERBS backend.
This requires some changes to Gloo which are in https://github.com/pytorch/gloo/pull/441
Since we're now depending on gloo_cuda we need to split ProcessGroupGloo into two pieces, one with the CPU bits (libtorch_cpu) and one with CUDA kernels in libtorch_cuda. This unfortunately requires some major refactoring as some CPU code is shared across both.
The gloo submodule is updated to depend on the new Gloo changes
Test plan:
```py
import os
import time
transport = "TCP"
#transport = "IBVERBS"
os.environ["GLOO_DEVICE_TRANSPORT"] = transport
rank = int(os.environ["RANK"])
os.environ["CUDA_VISIBLE_DEVICES"] = str(rank)
ibv = "mlx5_0:1,mlx5_3:1,mlx5_4:1,mlx5_5:1,mlx5_6:1,mlx5_9:1,mlx5_10:1,mlx5_11:1".split(",")[rank]
ibv_name, ibv_port = ibv.split(":")
os.environ["TORCH_GLOO_IBV_NAME"] = ibv_name
os.environ["TORCH_GLOO_IBV_PORT"] = ibv_port
os.environ["TORCH_GLOO_IBV_INDEX"] = "3"
import torch
import torch.distributed as dist
dist.init_process_group("gloo")
rank = dist.get_rank()
# initial sanity check
#device = "cpu"
#t = torch.zeros(10, device=device)
#dist.all_reduce(t)
#print("sanity complete")
device = "cpu"
iters = 10
warmup_iters = 2
for nelem in [10, 100, 1000, 10000, 100000, 1000000, 10000000, 100000000]:
t = torch.zeros(nelem, device=device)
torch.cuda.current_stream().synchronize()
for i in range(warmup_iters):
dist.all_reduce(t)
torch.cuda.current_stream().synchronize()
start = time.perf_counter()
for i in range(iters):
dist.all_reduce(t)
torch.cuda.current_stream().synchronize()
dur = (time.perf_counter() - start)
qps = iters/dur
bandwidth_gb = t.nbytes * iters / dur / 1e9
gb = t.nbytes / 1e9
if rank == 0:
print(f"{transport=} {device=} {iters=} {nelem=} {qps=} {gb=} {bandwidth_gb=}\n", end="")
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/153406
Approved by: https://github.com/fduwjj
A lot of last minute bugfixes for CUTLASS blackwell that we should upstream. It's a header only library and a minor release so this should strictly improve compiler support and fix some bugs. Needed to update some instruction numbers in torch compile baselines for the new kernels
Pull Request resolved: https://github.com/pytorch/pytorch/pull/152779
Approved by: https://github.com/henrylhtsang
This adds lazy initialization support to ProcessGroupGloo via `TORCH_GLOO_LAZY_INIT` or via `create_device(..., lazy_init=True)`
This is still a draft PR as there's one race condition when doing coalesced operations that needs to be fixed upstream in Gloo first. Depends on https://github.com/facebookincubator/gloo/pull/427 landing first
This also updates the gloo submodule to include the required changes.
Test plan:
added lazy init test variants
```
pytest -v test/distributed/test_c10d_gloo.py -k Lazy
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/150801
Approved by: https://github.com/fduwjj
Summary:
Profiler side of memory snapshot.
1. Add API to actually do snapshot when client interface is called
2. Add ifdefs to builds so that kineto hooks snapshot correctly.
Design Philosophy: There is one interesting part of this implementation and it is during export. For export we are callign the python impl of the export rather than CPP even though we are already in CPP. This is because it is better to simply have one path of export rather than 2. Personally, I want there to be parity between auto-trace and on-demand so it if we can limit the side paths then we will have an easier time maintaining this relationship
Test Plan: {F1976563426}
Reviewed By: sanrise
Differential Revision: D70733247
Pull Request resolved: https://github.com/pytorch/pytorch/pull/150559
Approved by: https://github.com/sanrise
