CLBlast.jl

Julia wrapper of CLBlast, a "tuned OpenCL BLAS library".
Author JuliaGPU
Popularity
8 Stars
Updated Last
4 Months Ago
Started In
June 2018

CLBlast

Build Status Build Status Build status Coverage Status codecov.io

Wrapper of CLBlast, a "tuned OpenCL BLAS library". This package has been inspired by CLBLAS.jl and the BLAS module of Julia and is designed similarly.

Current Status

Most low-level bindings and high-level wrappers of BLAS level 1, 2, and 3 routines are implemented.

Example

using CLBlast, OpenCL
@static if VERSION < v"0.7-"
    LA = LinAlg
else
    using Random, LinearAlgebra
    LA = LinearAlgebra
end

device, context, queue = cl.create_compute_context()

# setup data
α = 1.f0
β = 1.f0
A = rand(Float32, 10, 8)
B = rand(Float32, 8, 6)
C = zeros(Float32, 10, 6)

# transfer data
A_cl = cl.CLArray(queue, A)
B_cl = cl.CLArray(queue, B)
C_cl = cl.CLArray(queue, C)

# compute
LA.BLAS.gemm!('N', 'N', α, A, B, β, C)
CLBlast.gemm!('N', 'N', α, A_cl, B_cl, β, C_cl)

# compare results
@assert cl.to_host(C_cl)  C

Installation

Since this package is registered, you can add it using ] (activate package mode) and

(v0.7) pkg> add CLBlast

on Julia v0.7 or newer and using

julia> Pkg.add("CLBlast")

on Julia v0.6. During the build process, a suitable version of CLBlast will be downloaded and build. On Linux, you have to install clang, since the available binaries of CLBlast will fail to work with complex numbers from Julia.

Used By Packages

No packages found.