Repository files navigation

hipBLAS is a Basic Linear Algebra Subprograms (BLAS) marshalling library with multiple supportedbackends. It sits between your application and a 'worker' BLAS library, where it marshals inputs to thebackend library and marshals results to your application. hipBLAS exports an interface that doesn'trequire the client to change, regardless of the chosen backend. Currently, hipBLAS supports rocBLASand cuBLAS backends.

To use hipBLAS, you must first install rocBLAS, rocSPARSE, and rocSOLVER or cuBLAS.

To build our documentation locally, use the following code:

cd docspip3 install -r sphinx/requirements.txtpython3 -m sphinx -T -E -b html -d _build/doctrees -D language=en. _build/html

Alternatively, build with CMake:

cmake -DBUILD_DOCS=ON ...

1. Download the hipBLAS source code (clone this repository):

       git clone https://github.com/ROCmSoftwarePlatform/hipBLAS.git

       hipBLAS requires specific versions of rocBLAS and rocSOLVER. Refer to    [CMakeLists.txt](https://github.com/ROCmSoftwarePlatform/hipBLAS/blob/develop/library/CMakeLists.txt)    for details.

2. Build hipBLAS and install it into/opt/rocm/hipblas:

   cd hipblas    ./install.sh -i

The hipBLAS interface is compatible with rocBLAS and cuBLAS-v2 APIs. Porting a CUDA applicationthat originally calls the cuBLAS API to an application that calls the hipBLAS API is relativelystraightforward. For example, the hipBLAS SGEMV interface is:

hipblasStatus_thipblasSgemv(hipblasHandle_thandle,hipblasOperation_ttrans,intm,intn,constfloat*alpha,constfloat*A,intlda,constfloat*x,intincx,constfloat*beta,float*y,intincy );

hipBLAS GEMM can process matrices in batches with regular strides by using the strided-batchedversion of the API:

hipblasStatus_thipblasSgemmStridedBatched(hipblasHandle_thandle,hipblasOperation_ttransa,hipblasOperation_ttransb,intm,intn,intk,constfloat*alpha,constfloat*A,intlda,long longbsa,constfloat*B,intldb,long longbsb,constfloat*beta,float*C,intldc,long longbsc,intbatchCount);

hipBLAS assumes matrix A and vectors x, y are allocated in GPU memory space filled with data. Youare responsible for copying data to and from the host and device memory.