github/ROCm
1
1
Fork 0
mirror of https://github.com/ROCm/ROCm.git synced 2026-01-08 22:28:06 -05:00
Code Issues Packages Projects Releases Wiki Activity

gpu-aware MPI changes (#2311)

Browse Source 
- simplify the configure arguments of UCX to only provide
flags absolutely required

- add the UCC compatibility matrix to the docs
This commit is contained in:
Edgar Gabriel
2023-07-06 10:17:56 -05:00
committed by GitHub
parent 26935408e0
commit 2721042eac
2 changed files with 17 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
docs/how_to/gpu_aware_mpi.md
View File

@@ -66,11 +66,8 @@ cd ucx
./autogen.sh
mkdir build
cd build
../contrib/configure-release -prefix=$UCX_DIR \
--with-rocm=/opt/rocm \
--without-cuda -enable-optimizations -disable-logging \
--disable-debug -disable-assertions \
--disable-params-check -without-java
../configure -prefix=$UCX_DIR \
--with-rocm=/opt/rocm
make -j $(nproc)
make -j $(nproc) install
```
@@ -93,9 +90,7 @@ cd ompi
mkdir build
cd build
../configure --prefix=$OMPI_DIR --with-ucx=$UCX_DIR \
--with-rocm=/opt/rocm \
--enable-mca-no-build=btl-uct --enable-mpi1-compatibility \
CC=clang CXX=clang++ FC=flang
--with-rocm=/opt/rocm
make -j $(nproc)
make -j $(nproc) install
```
@@ -165,7 +160,12 @@ Inter-GPU bandwidth with various payload sizes.
Collective Operations on GPU buffers are best handled through the
Unified Collective Communication Library (UCC) component in Open MPI.
For this, the UCC library has to be configured and compiled with ROCm
support. An example for configuring UCC and Open MPI with ROCm support
support.
Please note the compatibility [table](../release/3rd_party_support_matrix.md#communication-libraries)
for UCC versions with the various ROCm versions.
An example for configuring UCC and Open MPI with ROCm support
is shown below:
```shell

8
docs/release/3rd_party_support_matrix.md
View File

@@ -31,6 +31,14 @@ UCX version | ROCm 5.4 and older | ROCm 5.5 and newer |
| -1.14.0 | COMPATIBLE | INCOMPATIBLE |
| 1.14.1+ | COMPATIBLE | COMPATIBLE |
The Unified Collective Communication Library [UCC](https://https://github.com/openucx/ucc)
also has support for ROCm devices.
UCC version | ROCm 5.5 and older | ROCm 5.6 and newer |
|:----------|:------------------:|:------------------:|
| -1.1.0 | COMPATIBLE | INCOMPATIBLE |
| 1.2.0+ | COMPATIBLE | COMPATIBLE |
## Algorithm libraries
ROCm releases provide algorithm libraries with interfaces compatible with