Draft: Aarch64 popcount
Some initial, possibly controversial, work to add support for popcount using the built in cnt
instruction. This instruction requires access to vector registers, so this PR adds some initial support for them. AArch64's vector registers are a little strange, they are both integer and float registers, but are really just the float registers renamed.
The pop count is very simple, the source is moved into the low bytes of a vector reg, cnt (which counts the set bits in each byte and returns the result in each byte) modifies that register, the bytes are summed, and the result is moved into the dest register (which is what clang produces for __builtin_popcountll etc.).
Please take a few moments to address the following points:
-
if your MR may break existing programs (e.g. touches base
or causes the compiler to reject programs), please describe the expected breakage and add the user-facing label. This will run ghc/head.hackage> to characterise the effect of your change on Hackage. -
ensure that your commits are either individually buildable or squashed -
ensure that your commit messages describe what they do (referring to tickets using #NNNN
syntax when appropriate) -
have added source comments describing your change. For larger changes you likely should add a Note and cross-reference it from the relevant places. -
add a testcase to the testsuite. -
updates the users guide if applicable -
mentions new features in the release notes for the next release
If you have any questions don't hesitate to open your merge request and inquire
in a comment. If your patch isn't quite done yet please do add prefix your MR
title with WIP:
.
By default a minimal validation pipeline is run on each merge request, the full-ci label can be applied to perform additional validation checks if your MR affects a more unusual configuration.
Once your change is ready please remove the WIP:
tag and wait for review. If
no one has offered a review in a few days then please leave a comment mentioning
@triagers and apply the Blocked on Review label.