hughperkins/tf-coriander

OpenCL 1.2 implementation for Tensorflow

Tensorflow-cl

Run Tensorflow on OpenCL™ 1.2 devices

• tested on:
  • Mac Sierra using Radeon Pro 450 GPU (thank you ASAPP :-) )
  • Ubuntu 16.04, using NVIDIA K520
• should work theoretically on any OpenCL 1.2 GPU

Piccie

On a Mac:

Execution speed

• Execution speed

What's working

• What's working

Installation

• Installation

Tests

• Tests

Design/architecture

• tensorflow code stays 100% NVIDIA® CUDA™
• Coriander compiles the NVIDIA® CUDA™ code into OpenCL
• Cedric Nugteren's CLBlast provides BLAS (matrix multiplications)

Presentation on Coriander at this year's IWOCL 2017

Related projects

• OpenCL Torch
• DeepCL

News

• June 11 2017:
  • set up Jenkins build, which makes build logs and a Ubuntu 16.04 wheel available for certain commits, https://github.com/hughperkins/tf-coriander/commits/example-jenkins-builds (click on one of the green ticks)
• June 7 2017:
  • created v0.18.3 release:
    • tf.split enabled
    • following examples from Aymeric Damien's Tensorflow-Examples run now:
      • recurrent_network.py
      • bidirectional_rnn.py
      • dynamic_rnn.py