llvm/llvm-project

[AVX-512] Look for masked `vpmovqb` opportunities on Zen 4+

Open

#156,883 opened on Sep 4, 2025

View on GitHub
 (9 comments) (0 reactions) (1 assignee)C++ (10,782 forks)batch import
backend:X86good first issuellvm:SelectionDAGmissed-optimization

Repository metrics

Stars
 (26,378 stars)
PR merge metrics
 (Avg merge 1d 2h) (1,000 merged PRs in 30d)

Description

Zig version: (Godbolt)

export fn foo(x: @Vector(8, u64), y: @Vector(8, u64)) @Vector(8, u8) {
    return @select(u8, x == y, @popCount(x), @as(@Vector(8, u8), @splat(0)));
}

LLVM version: (Godbolt)

define dso_local range(i8 0, 65) <8 x i8> @foo(<8 x i64> %0, <8 x i64> %1) local_unnamed_addr {
Entry:
  %2 = icmp eq <8 x i64> %0, %1
  %3 = tail call range(i64 0, 65) <8 x i64> @llvm.ctpop.v8i64(<8 x i64> %0)
  %4 = trunc nuw nsw <8 x i64> %3 to <8 x i8>
  %5 = select <8 x i1> %2, <8 x i8> %4, <8 x i8> zeroinitializer
  ret <8 x i8> %5
}

declare <8 x i64> @llvm.ctpop.v8i64(<8 x i64>) #1

Emits:

foo:
        vpcmpeqq        k1, zmm0, zmm1
        vpopcntq        zmm0, zmm0
        vpmovqb xmm0, zmm0
        vmovdqu8        xmm0 {k1} {z}, xmm0
        vzeroupper
        ret

Should be:

foo:
        vpcmpeqq        k1, zmm0, zmm1
        vpopcntq        zmm0, zmm0
        vpmovqb xmm0 {k1}, zmm0
        vzeroupper
        ret

Contributor guide

Research direction
Investigate LLVM's X86 instruction selection and DAG combining for masked vpmovqb. The current code generates a separate vpmovqb and then a masked move, but it should combine into a single masked vpmovqb. Look at how LLVM handles masked operations and find where the optimization is missing.
Tech stack
cpp
Domain
performance
Issue type
Feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
4
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
3-5 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Clear
Prerequisites
LLVM backendAVX 512
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
20