The Math Vector Library GPU brings our flagship mathematical subroutine library to a new level of performance. Accelerate computations by offloading to General-Purpose GPU (GPGPU) or SIMD accelerators without the need to deal with special considerations or tools.
The Math Vector Library GPU can be used as a drop-in for existing applications using the Math Vector Library, no code or ABI changes needed. This maintains our goal of providing a true write-once execute anywhere paradigm.
Math Vector Library GPU is currently in an experimental state and is available for customers having purchased Math Vector Library Linux amd64/x86_64 (SA-00002-01X) runtime licenses free of charge until reaching a stable state at which point a new runtime license will be introduced including both the standard contents of the Linux amd64/x86_64 runtime as well as the GPU version.
For general information, please visit the Math Vector Library product page.
Supported Platforms:
- Linux on AMD64/x86_64
The following offloading targets are supported as of the current version:
- AMD GCN HSA
In the future we will add support for NVIDIA PTX targets as well.