jerinphilip/MozIntGemm

Wrapper around intgemm (x86_64) and ruy (ARM) to switch between both based on architecture and provide a fast matrix multiplication backend for Mozilla Firefox's translation feature.

C++MIT

Issues

Optimize MaxAbsolute with SIMD
#30 opened 3 years ago
0
Tagged releases
#29 opened 3 years ago
0
Connect ARM to the gecko-dev import script
#28 opened 3 years ago
0
Import fallback and WASM portable matmul here
#27 opened 3 years ago
0
Do something about std::memcpy from a nullptr
#26 opened 3 years ago
0
Use reduced version of cpuinfo
#24 opened 3 years ago
0
auto lambda not supported on termux compiler
#23 opened 3 years ago
0
Update comments to reflect optimized transpose
#20 opened 3 years ago
0
Optimize for latency in memory loads
#17 opened 3 years ago
0
Brute-force best multiply path
#8 opened 3 years ago
0
Optimize and verify emitted quantize on ARM
#7 opened 3 years ago
0