Skip to content

Commit bce5012

Browse files
authored
Ccl/2.5 (#226)
* update oneccl to 2021.14 (#224) Use oneccl from https://github.com/oneapi-src/oneCCL/releases/tag/2021.14 - 3afa1bb7936f57683a2503c34b29c0daca6a9cc * update readme (#223) * restore known issue
1 parent a756fd4 commit bce5012

File tree

2 files changed

+20
-17
lines changed

2 files changed

+20
-17
lines changed

README.md

Lines changed: 19 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -36,6 +36,7 @@ We recommend using Anaconda as Python package management system. The followings
3636
| `torch` | `oneccl_bindings_for_pytorch` |
3737
| :-------------------------------------------------------------: | :-----------------------------------------------------------------------: |
3838
| `master` | `master` |
39+
| [v2.5.0](https://github.com/pytorch/pytorch/tree/v2.5.0) | [ccl_torch2.5.0](https://github.com/intel/torch-ccl/tree/ccl_torch2.5.0+xpu) |
3940
| [v2.3.1](https://github.com/pytorch/pytorch/tree/v2.3.1) | [ccl_torch2.3.100](https://github.com/intel/torch-ccl/tree/ccl_torch2.3.100+xpu) |
4041
| [v2.1.0](https://github.com/pytorch/pytorch/tree/v2.1.0) | [ccl_torch2.1.400](https://github.com/intel/torch-ccl/tree/ccl_torch2.1.400+xpu) |
4142
| [v2.1.0](https://github.com/pytorch/pytorch/tree/v2.1.0) | [ccl_torch2.1.300](https://github.com/intel/torch-ccl/tree/ccl_torch2.1.300+xpu) |
@@ -59,7 +60,7 @@ The usage details can be found in the README of corresponding branch.
5960

6061
- Python 3.8 or later and a C++17 compiler
6162

62-
- PyTorch v2.3.1
63+
- PyTorch v2.5.1
6364

6465
## Build Option List
6566

@@ -93,6 +94,7 @@ The following launch options are supported in Intel® oneCCL Bindings for PyTorc
9394

9495
```bash
9596
git clone https://github.com/intel/torch-ccl.git && cd torch-ccl
97+
git checkout ccl_torch2.5.0+xpu
9698
git submodule sync
9799
git submodule update --init --recursive
98100
```
@@ -114,22 +116,23 @@ The following launch options are supported in Intel® oneCCL Bindings for PyTorc
114116

115117
Wheel files are available for the following Python versions. Please always use the latest release to get started.
116118

117-
| Extension Version | Python 3.6 | Python 3.7 | Python 3.8 | Python 3.9 | Python 3.10 | Python 3.11 |
118-
| :---------------: | :--------: | :--------: | :--------: | :--------: | :---------: | :---------: |
119-
| 2.3.100 | | |||||
120-
| 2.1.400 | | |||||
121-
| 2.1.300 | | |||||
122-
| 2.1.200 | | |||||
123-
| 2.1.100 | | |||||
124-
| 2.0.100 | | |||||
125-
| 1.13 | ||||| |
126-
| 1.12.100 | ||||| |
127-
| 1.12.0 | ||||| |
128-
| 1.11.0 | ||||| |
129-
| 1.10.0 ||||| | |
119+
| Extension Version | Python 3.6 | Python 3.7 | Python 3.8 | Python 3.9 | Python 3.10 | Python 3.11 | Python 3.12 |
120+
| :---------------: | :--------: | :--------: | :--------: | :--------: | :---------: | :---------: | :---------: |
121+
| 2.5.1 | | | |||||
122+
| 2.3.100 | | ||||| |
123+
| 2.1.400 | | ||||| |
124+
| 2.1.300 | | ||||| |
125+
| 2.1.200 | | ||||| |
126+
| 2.1.100 | | ||||| |
127+
| 2.0.100 | | ||||| |
128+
| 1.13 | ||||| | |
129+
| 1.12.100 | ||||| | |
130+
| 1.12.0 | ||||| | |
131+
| 1.11.0 | ||||| | |
132+
| 1.10.0 ||||| | | |
130133

131134
```bash
132-
python -m pip install oneccl_bind_pt==2.3.100 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
135+
python -m pip install oneccl_bind_pt==2.5.0 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
133136
```
134137

135138
**Note:** Please set proxy or update URL address to https://pytorch-extension.intel.com/release-whl/stable/xpu/cn/ if you meet connection issue.
@@ -270,7 +273,7 @@ mpirun -n 2 -l python profiling.py
270273

271274
## Known Issues
272275

273-
For Point-to-point communication, directly call dist.send/recv after initializing the process group in launch script will trigger runtime error. Because all ranks of the group are expected to participate in this call to create communicators in our current implementation, while dist.send/recv only has a pair of ranks' participation. As a result, dist.send/recv should be used after collective call, which ensures all ranks' participation. The further solution for supporting directly call dist.send/recv after initializing the process group is still under investigation.
276+
For Point-to-point communication, directly call dist.send/recv after initializing the process group in launch script will trigger runtime error. Because all ranks of the group are expected to participate in this call to create communicators in our current implementation, while dist.send/recv only has a pair of ranks' participation. As a result, dist.send/recv should be used after collective call, which ensures all ranks' participation.
274277

275278
## License
276279

third_party/oneCCL

0 commit comments

Comments
 (0)