|
- `wip/ermsb`: **Extended rep-movsb (ERMSB) support for Intel processors**. Austin Seipp. Modern Intel processors feature extremely fast (yet simple!) string copy primitives, which are taken advantage of by simply using the existing `rep movsb` instruction. For many workloads, these copies are competetive with fast AVX-based copies (on my Haswell machine, the difference between the two was indistinguishable). This should result in superior copy performance for Ivy Bridge processors and later, which support the ERMSB addition. This also adds support for `-march` and `-mcpu` to the compiler.
|
|
- `wip/ermsb`: **Extended rep-movsb (ERMSB) support for Intel processors**. Austin Seipp. Modern Intel processors feature extremely fast (yet simple!) string copy primitives, which are taken advantage of by simply using the existing `rep movsb` instruction. For many workloads, these copies are competetive with fast AVX-based copies (on my Haswell machine, the difference between the two was indistinguishable). This should result in superior copy performance for Ivy Bridge processors and later, which support the ERMSB addition. This also adds support for `-march` and `-mcpu` to the compiler.
|