Can Burst Apply Function Stitching To Boost Performance?

Arowx · Sep 11, 2019

Intel article on functional stitching that achieved 1.4x-1.9x performance improvements on blocks of function calls... https://pdfs.semanticscholar.org/8928/6c76c6f17fd50921584467fe44f793de3d8a.pdf

We introduced a novel method to improve computing performance called function stitching, and demonstrated its benefits on the example of cryptographic functions. It unleashes the full potential of the modern superscalar microarchitecture of Intel processors. Stitching is the source-level fine grained interleaving of the processing of independent instruction streams.

We stitched the most common pairs of cryptographic functions for increased performance, such as RC4-MD5, AES128-SHA1 and AES256-SHA1, achieving excellent performance improvement with the speedups in the 1.4X-1.9X range over the optimized individual functions called sequentially. Stitched functions are able to execute very efficiently on IA cores, closely approaching IA core peak execution limits.

Stitching also allowed us to take popular pairs of block ciphers and hash algorithms such as RC4-MD5, AES128-SHA1 and AES256-SHA1 to the same highly improved performance level; now the choice of cryptography strengths comes without any performance trade-off.
Click to expand...

Joachim_Ante · Sep 11, 2019

Hey man. You seem to ask a lot of random questions / suggestions that are not really about concrete things in DOTS, based on finding random articles on the internet.

Those questions take up my own time and other peoples time on the forum. They don't help the general understanding of DOTS or solve any specific problems.

Maybe you could either tone it down or alternatively just make a single thread and keep those questions there?

Search Unity

Can Burst Apply Function Stitching To Boost Performance?

Arowx

Joachim_Ante

Unity Technologies

Search Unity

Unity ID

Useful Searches

Can Burst Apply Function Stitching To Boost Performance?

Arowx

Joachim_Ante

Unity Technologies