ROCm with gfx803

This repo upload some archieves for ROCm with gfx803 after ROCm-4.0.

https://github.com/xuhuisheng/rocm-build/tree/master/gfx803

My environments

OS	Python	ROCm	GPU
Ubuntu-20.04.2	3.8.10	4.3.0	RX580

ROCm-4.3.0

https://github.com/xuhuisheng/rocm-gfx803/releases/tag/rocm43

Install ROCm First https://rocmdocs.amd.com/en/latest/Installation_Guide/Installation-Guide.html#ubuntu

component	version	size	link
pytorch	1.9.0	206M	https://github.com/xuhuisheng/rocm-gfx803/releases/download/rocm43/torch-1.9.0a0+gitd69c22d-cp38-cp38-linux_x86_64.whl
torchvision	0.10.0	21M	https://github.com/xuhuisheng/rocm-gfx803/releases/download/rocm43/torchvision-0.10.0a0+300a8a4-cp38-cp38-linux_x86_64.whl
rocblas	2.39.0	7.4M	https://github.com/xuhuisheng/rocm-gfx803/releases/download/rocm43/rocblas_2.39.0-8328dcce~dirty_amd64.deb

1. sudo apt install rocm-dkms rocm-libs
2. sudo dpkg -i rocblas_2.39.0-8328dcce~dirty_amd64.deb
3. pip3 install torch-1.9.0a0+gitd69c22d-cp38-cp38-linux_x86_64.whl
4. pip3 install torchvision-0.10.0a0+300a8a4-cp38-cp38-linux_x86_64.whl

ROCm-3.5.1

https://github.com/xuhuisheng/rocm-gfx803/releases/tag/rocm35

Install ROCm-3.5.1 First https://github.com/boriswinner/RX580-rocM-tensorflow-ubuntu20.4-guide

component	version	size	link
pytorch	1.7.0	173M	https://github.com/xuhuisheng/rocm-gfx803/releases/download/rocm35/torch-1.7.0a0-cp38-cp38-linux_x86_64.whl
torchvision	0.8.0	6.4M	https://github.com/xuhuisheng/rocm-gfx803/releases/download/rocm35/torchvision-0.8.0a0+2f40a48-cp38-cp38-linux_x86_64.whl

1. sudo apt install rocm-dkms rocm-libs
2. export LB_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/rocm/hip/lib
3. pip3 install torch-1.7.0a0-cp38-cp38-linux_x86_64.whl
4. pip3 install torchvision-0.8.0a0+2f40a48-cp38-cp38-linux_x86_64.whl