PerfMPL

PerfMPL, an independently developed mathematical computation library by PerfXLab, has successfully constructed source codes of key mathematical computation libraries, including BLAS, FFT, MATH/VML, Sparse, IPP, DNN, etc., realizing commercialization with close to 6000 APIs. PerfMPL holds a leading domestic position in the completeness and performance optimization of mathematical computation libraries. It aims ultimately to match the international level libraries, such as Intel MKL and CUDA computation libraries.


...

Standard Version

PerfMPL for ARM, PerfMPL for PerfMPL for OpenCL GPUPanel content

Domain Version

Providing targeted optimization for application fields, such as EDA , CFD , Signal Processing, CV , AI, and so on.

Customized Version

Domestic GPU/NPU manufacturers offer customized services of PerfMPL, benchmarking against cuBLAS, cuFFT, and cu Math, providing targeted optimization for various application fields such as EDA, CFD, Signal Processing, CV, AI, and so on.
PerfMPL on Arm CPU
...
...
...
...
PerfMPL
...
...
...
...
PerfMPL on OpenCL-GPU
...
...
...
...

Feature Highlights

1 Performance benchmarked against international standards;

2 Ultimate precision;

3 Safe, reliable, and stable.

ADDRESS

Building 9, Yard 55, Haidian, Zique Road, Beijing, China
Room 511, DoBe, Yuelu District, Changsha, China

E-MAIL

[email protected]

TELEPHONE

010-62600621