Add double-pumped neon-i8x32 and neon-i16x16 ISAs, new Cortex and Apple CPUs
#42
Conversation
|
https://github.com/ispc/ispc/releases/tag/v1.27.0 seems to have received some new AVX targets as well, perhaps those should be added. |
… Apple CPUs These were added in [the 1.26 release]. Also improve some existing deprecation notices to list _when_ they were deprecated, now that we know the actual version in which that happened. Some "pending" deprecations weren't pushed through in 1.26 nor 1.27 though. [the 1.26 release]: https://github.com/ispc/ispc/releases/tag/v1.26.0
|
We should most likely also add all the new |
There was a problem hiding this comment.
Thanks @MarijnS95 ! If you want to add the other ISAs or flags I'm happy to review PRs, my wife and I just had our first baby recently though so I likely won't be making direct contributions to most side projects for a bit
|
@Twinklebear congrats, awesome! Enjoy your precious time! I'll happily PR more flags, but want it to be driven by some benchmarking on our end. We're seeing heavy slowdowns on Apple devices and believe it is because of missing these flags but I need a more digestible benchmark in https://github.com/Traverse-Research/intel-tex-rs-2 similar to what we have in https://github.com/Traverse-Research/ispc-downsampler. Then I can also experiment with migrating to those generic flags :) |
|
Using generic arguments appears to not really be implemented yet, and/or makes it impossible to correctly select what CPU and ISA to target (https://github.com/ispc/ispc/blob/381723518e6673f21d86a8c411f2be6b86449b9e/src/ispc.cpp#L2840-L2854): This only works when selecting one specific Target ISA and setting a CPU. |
These were added in the 1.26 release. Also improve some existing deprecation notices to list when they were deprecated, now that we know the actual version in which that happened. Some "pending" deprecations weren't pushed through in 1.26 nor 1.27 though.