This is the developer repository for ISAAC, a new approach to BLAS implementations. ISAAC uses Machine-Learning techniques to achieve input-specific and architecture-aware computations, thereby outperforming cuBLAS and clBLAS on many applications for NVidia, Intel and AMD GPUs.

License

ISAAC is distributed under the GNU LGPL v2.1 License.

Installation

ISAAC is dependency-free, and will load either OpenCL or CUDA 7.0+ dynamically depending on which GPUs are detected at runtime.

Installation requires CMake 2.8.7+ and a C++11 compliant compiler:

git clone https://github.com/ptillet/isaac.git
mkdir -p isaac/build && cd isaac/build
cmake ../ && make -j4

Link against libisaac.so instead of libcublas.so or libclblas.so, and you're good to go!

The C++ and Python API does some kernel fusion, but is not entirely stable. It works well to compose element-wise operations, though.

Benchmark

./bench/bench-blas OP

where OP is axpy, gemv or gemm

If you have multiple devices

./bench/bench-blas DEVICE_ID OP

BLAS routines supported

Currently supported functions are:

BLAS1	BLAS2	BLAS3
xAXPY	xGEMV	xGEMM
xCOPY	xGER
xSCAL
xDOT
xASUM

for x = {S, D}