small self-contained low-precision GEMM library

 This is not a full linear algebra library, only a GEMM library: it only does
 general matrix multiplication ("GEMM").
 .
 Its performance goals differ from typical GEMM performance goals in the
 following ways:
 1. It cares not only about speed, but also about minimizing power usage.
    It specifically cares about charge usage in mobile/embedded devices.
 2. Most GEMMs are optimized primarily for large dense matrix sizes (>= 1000).
    It does care about large sizes, but it also cares specifically about the
    typically smaller matrix sizes encountered in various mobile applications.
 .
 Keep in mind (previous section) that gemmlowp itself is a pure-headers-only
 library.