This project aims to provide cross platform OpenCL tools for deep learning and inference.
Today, most of deep learning training is done on NVidia GPUs using closed source CUDA and CUDNN libraries. It is either challenging or virtually impossible to use AMD or Intel GPUs. For example: AMD provides ROCm platform, but there is no support of RDNA platforms yet (more than a year since a release), there is no support of APUs and no support of any operating systems other than Linux.
- Create an open source, cross platform deep learning primitives library similar to cuDNN or MIOpen that supports multiple GPU architectures.
- Create an inference library with minimal dependencies for efficient inference on any modern GPU, similar to TensorRT or MIGraphX.
- Create minimalistic deep-learning framework as POC of capabilities and performance.
- Long Shot: Integrate to existing large scale deep learing projects like PyTorch, TF, MXNet such that vendor independent open-source OpenCL API will be first class citizen for deep learning.
Please note this is only work in progress - first and preliminary stages.
Is published under http://dlprimitives.org/docs/
| Operator | Features | Computation |
|---|---|---|
| Softmax | Fwd | |
| SoftmaxWithLoss | Fwd,Bwd | |
| Elementwise | ax+by, max(ax,by), ax*y | Fwd,Bwd |
| Concat | Fwd,Bwd | |
| Slice | Fwd,Bwd | |
| MaxPool2d | Fwd,Bwd | |
| AvgPool2d | Fwd,Bwd | |
| GlobalMaxPool2d | Fwd,Bwd | |
| GlobalAvgPool2d | Fwd,Bwd | |
| Inner Product | Fwd,Bwd | |
| BatchNorm2D | Fwd,Bwd | |
| Conv2d | GEMM, Winograd, Depthwise Separable | Fwd,Bwd |
| TransposedConv2d | GEMM, Winograd, Depthwise Separable | Fwd,Bwd |
| Activation | relu, sigmoid, tanh, relu6 | Fwd,Bwd |
Solvers: SGD, Adam
| Network | Source of model | Operation |
|---|---|---|
| AlexNet | torchvision.models | Inference |
| VGG16 | torchvision.models | Inference |
| ResNet50 | torchvision.models | Inference |
| ResNet18 | torchvision.models | Inference |
| MobileNet v2 | torchvision.models | Inference |
The networks were exported from pytorch to ONNX and imported for DLPrimitives. Results compared with sample images. Note currently only inference validated, backpropogation is convered by per-layer regression.
| Device | Vendor | Notes |
|---|---|---|
| RX 6600XT | AMD | ROCr |
| RX 560 | AMD | 16cu model, ROCm, PAL, Clover |
| HD 530 | Intel | i5-6600, NEO driver |
| GTX 960 | NVidia | |
| GTX 1080 | NVidia | |
| RTX 2060S | NVidia | |
| MaliG52 MC2 | ARM | performance not optimised yet |
Devices Tested on Windows: AMD RX 560, NVidia GTX 960.
- Network object for inference
- ONNX to DlPrimitives model converter