search

Quuxplusone / LLVMBugzillaTest

0 stars 0 forks source link

[X86] What AVX512 cpus should prefer 512-bit vectors? #47305

Open Quuxplusone opened 3 years ago

Quuxplusone commented 3 years ago
Bugzilla Link PR48336
Status NEW
Importance P enhancement
Reported by Simon Pilgrim (llvm-dev@redking.me.uk)
Reported on 2020-11-30 06:36:29 -0800
Last modified on 2021-10-06 09:18:22 -0700
Version trunk
Hardware PC Linux
CC craig.topper@gmail.com, llvm-bugs@lists.llvm.org, llvm-dev@redking.me.uk, matdzb@gmail.com, pengfei.wang@intel.com, peter@cordes.ca, spatel+llvm@rotateright.com
Fixed by commit(s)
Attachments
Blocks
Blocked by
See also

All avx512 capable Skylake CPUs and later currently prefer 256-bit vs 512-bit vector generation - only the Knights Landing family always prefers 512-bit (very different arch, no VLX support etc.).

AFAICT many of the frequency throttling issues with 512-bit ops were already addressed by/after Icelake - so do we need to alter the tuning for later cpus to let them vectorize to the full 512-bit alu width?

Quuxplusone commented 3 years ago

Candidate Patch: https://reviews.llvm.org/D111029