Coriander

Build applications written in NVIDIA® CUDA™ code for OpenCL™ 1.2 devices.

Concept

• leave applications in NVIDIA® CUDA™
• compile into OpenCL 1.2
• run on any OpenCL 1.2 GPU

How to use

• Write an NVIDIA® CUDA™ sourcecode file, or find an existing one
• Let's use cuda_sample.cu
• Compile, using cocl:

$ cocl_py cuda_sample.cu
   ...
   ... (bunch of compily stuff) ...
   ...

    ./cuda_sample.cu compiled into ./cuda_sample

Run:

$ ./cuda_sample
Using Intel , OpenCL platform: Intel Gen OCL Driver
Using OpenCL device: Intel(R) HD Graphics 5500 BroadWell U-Processor GT2
hostFloats[2] 123
hostFloats[2] 222
hostFloats[2] 444

Options

What Coriander provides

• compiler for host-side code, including memory allocation, copy, streams, kernel launches
• compiler for device-side code, handling templated C++ code, converting it into bog-standard OpenCL 1.2 code
• cuBLAS API implementations for GEMM, GEMV, SCAL, SAXPY (using Cedric Nugteren's CLBlast)
• cuDNN API implementations for: convolutions (using im2col algorithm over Cedric Nugteren's CLBlast, pooling, ReLU, tanh, and sigmoid

How Coriander works

Kernel compilation proceeds in two steps:

Slides on the IWOCL website, here

Installation

Coriander development is carried out using the following platforms:

• Ubuntu 16.04, with:
  • NVIDIA K80 GPU and/or NVIDIA K520 GPU (via aws)
• Mac Book Pro 4th generation (thank you ASAPP :-) ), with:
  • Intel HD Graphics 530
  • Radeon Pro 450
  • Sierra OS

Other systems should work too, ideally. You will need at a minimum at least one OpenCL-enabled GPU, and appropriate OpenCL drivers installed, for the GPU. Both linux and Mac systems stand a reasonable chance of working ok.

For installation, please see installation

Plugins

You can install the following plugins:

• Coriander-clblast: just do cocl_plugins.py install --repo-url https://github.com/hughperkins/coriander-clblast
• Coriander-dnn: just do cocl_plugins.py install --repo-url https://github.com/hughperkins/coriander-dnn
• Your plugin here?

How to create a plugin

Add to your own cmake project

• use cocl_add_executable and cocl_add_library
• see cmake usage

Testing

See testing

Assumptions/relaxations made by Coriander

See assumptions

Roadmap

Libraries

Coriander uses the following libraries:

• clang/llvm: c/c++ parser/compiler; many contributors
• thrust: parallel GPU library, from NVIDIA®
• yaml-cpp: yaml for c++, by Jesse Beder
• EasyCL: wrapper for OpenCL 1.2 boilerplate
• argparsecpp: command-line parser for c++
• gtest: unit tests for c++, from Google

Related projects

• Eigen-CL: Minimally-tweaked fork of Eigen, for OpenCL 1.2
• tf-coriander: Tensorflow for OpenCL-1.2

How to Cite

Please cite: CUDA-on-CL: a compiler and runtime for running NVIDIA® CUDA™ C++11 applications on OpenCL™ 1.2 Devices

License

Apache 2.0

News

• June 23:
  • factorized CLBlast implementation of NVIDIA® CUDA™ cuBLAS API, into new plugin coriander-clblast
• June 21:
  • created a new release v6.0.0, that marks a bunch of changes:
    • incorporates of course the earlier changes:
      • took some big steps towards portability and Windows compilation, ie using python 2.7 scripts, rather than bash scripts, and fixing many Windows-related compilation issues
      • the plugin architecture
      • factorizing the partial NVIDIA® CUDA™ cuDNN API implementation into a new plugin coriander-dnn
    • moved the default installation directory from /usr/local to ~/coriander
      • this means that plugins can be installed without sudo
      • it also makes it relatively easy to wipe and reinstall, for more effective jenkins testing
    • install_distro.py is now considerably more tested than a few days ago, and handles downloading llvm-4.0 automatically
• Older news