frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 21:14:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

update build guide to use mkl-static. (#116946)

Browse Source 
# Background:
We found current build guide use mkl dynamic link. It may trigger a mkl link issue.

Detailed:
In build environment, libtorch_cpu.so will dynamic link to system mkl binaries by default.
If users install another version mkl library, it may lead to mkl symbol conflict.

I also checked released pytorch binary it use static mkl link. The build script shows it: https://github.com/pytorch/builder/blob/main/common/install_mkl.sh#L10

# Solution:
Update build guide to use mkl static link. And it is aligned to build script.

Conda install command docs:
https://anaconda.org/intel/mkl-static
https://anaconda.org/intel/mkl-include

# Validation
No mkl libraries dependencing, after use `conda install intel::mkl-static intel::mkl-include`.
## Windows
![image](https://github.com/pytorch/pytorch/assets/8433590/cc554ded-d827-4de5-81c6-cc3039155580)

## Linux
<img width="959" alt="image" src="https://github.com/pytorch/pytorch/assets/8433590/79766ad8-4ba2-4ff1-adc9-63affd8d419a">

Pull Request resolved: https://github.com/pytorch/pytorch/pull/116946
Approved by: https://github.com/jgong5, https://github.com/malfet
This commit is contained in:
Xu Han
2024-01-10 01:34:58 +00:00
committed by PyTorch MergeBot
parent b4f1ab4505
commit 4a10e9eed4
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
README.md
View File

@ -194,7 +194,7 @@ pip install -r requirements.txt
**On Linux**
```bash
conda install mkl mkl-include
conda install intel::mkl-static intel::mkl-include
# CUDA only: Add LAPACK support for the GPU if needed
conda install -c pytorch magma-cuda110 # or the magma-cuda* that matches your CUDA version from https://anaconda.org/pytorch/repo
@ -207,7 +207,7 @@ make triton
```bash
# Add this package on intel x86 processor machines only
conda install mkl mkl-include
conda install intel::mkl-static intel::mkl-include
# Add these packages if torch.distributed is needed
conda install pkg-config libuv
```
@ -215,7 +215,7 @@ conda install pkg-config libuv
**On Windows**
```bash
conda install mkl mkl-include
conda install intel::mkl-static intel::mkl-include
# Add these packages if torch.distributed is needed.
# Distributed package support on Windows is a prototype feature and is subject to changes.
conda install -c conda-forge libuv=1.39