commit | be760b63149d8205dfb3ca66d78a049dc1ab7772 | [log] [tgz] |
---|---|---|
author | Georgios Pinitas <georgios.pinitas@arm.com> | Tue Feb 09 10:49:17 2021 -0800 |
committer | Copybara-Service <copybara-worker@google.com> | Tue Feb 09 10:49:37 2021 -0800 |
tree | 45b033783360ed59b9efbfd8bd2dfdb4fc8bdb72 | |
parent | 287015c8ea2b2bbc7780f85650263a92518dcd37 [diff] |
Simplify quantized multiplier Alter sequence to a single rounded scaling with normal rounded shift. Double rounding and symmetric rounding are removed compared to reference. Double rounding seems unnecessary and can complicate implementations. Moreover, symmetric rounding also adds implementation complexity. For NEON the new sequence can be translated to VQDMULH + VRSHR. Closes https://github.com/google/ruy/pull/227 COPYBARA_INTEGRATE_REVIEW=https://github.com/google/ruy/pull/227 from GeorgeARM:mul_pr dec00bd87a8815fdad79d302494430aa63522752 PiperOrigin-RevId: 356539687
This is not an officially supported Google product.
ruy is a matrix multiplication library. Its focus is to cover the matrix multiplication needs of neural network inference engines. Its initial user has been TensorFlow Lite, where it is used by default on the ARM CPU architecture.
ruy supports both floating-point and 8bit-integer-quantized matrices.
ruy is designed to achieve high performance not just on very large sizes, as is the focus of many established libraries, but on whatever are the actual sizes and shapes of matrices most critical in current TensorFlow Lite applications. This often means quite small sizes, e.g. 100x100 or even 50x50, and all sorts of rectangular shapes. It's not as fast as completely specialized code for each shape, but it aims to offer a good compromise of speed across all shapes and a small binary size.
Some documentation will eventually be available in the doc/ directory, see doc/README.md.