Optimizing math routines - GNU Tools Cauldron 2018





The interactive transcript could not be loaded.


Rating is available when the video has been rented.
This feature is not available right now. Please try again later.
Published on Oct 4, 2018

Presented by Szabolcs Nagy at GNU Tools Cauldron 2018

The new implementations are generic and the performance improvements should be portable across different targets. The main challenges were dealing with target variations, rounding errors, doing careful error analysis, following the requirements of various standards, testing and benchmarking different configurations, integrating the code into glibc and fighting with various tools. I plan to discuss these challenges and some of the design decisions behind the new code. The performance improvements are significant in most cases so it is also worth discussing the key ideas that made it possible.

Some of the patches were upstreamed last year, some are still under review, the list of optimized functions with ubenchmark performance improvement on common inputs: sinf (1.7x), cosf (1.7x), sincosf (1.7x), expf (3.5x), exp2f (3.1x), logf (2.3x), log2f (2.4x), powf (4.8x), sin (1.4x), cos (1.4x), sincos (1.7x), exp (1.9x), exp2 (2.0x), log (3.8x), log2 (2.0x) and pow (3.1x).


When autoplay is enabled, a suggested video will automatically play next.

Up next

to add this to Watch Later

Add to

Loading playlists...