SVE: Reuse ptrue value instead of rematerializing it everytime

Open

#114,286 opened on Apr 4, 2025 6 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Assessment

Difficulty: 4/5
Estimated time: 3-5 days
Newbie friendliness: 42/100

Issue type: Refactor
Clarity: Needs clarification
Activity status: Stale
Domain: compilers

Research direction

Start from the SVE code-generation path that emits the ptrue instructions shown in the issue and trace how predicate values are represented across uses and call boundaries. Define completion as reusing compatible ptrue values without changing predicate types or generated-code semantics, then validate the resulting assembly and relevant compiler tests.

Written by the indexing model from the issue text.

Description

area-CodeGen-coreclr arm-sve

Noticed the below piece of code, where we materialize p0 before every usage instead of just saving it and reusing it.

G_M34563_IG04:        ; bbWeight=4, gcrefRegs=0000 {}, byrefRegs=0000 {}, byref, isz
            ldp     q18, q19, [x0]
            ptrue   p0.d
            fabs    z18.d, p0/m, z18.d
            fadd    v0.2d, v0.2d, v18.2d
            ptrue   p0.d
            fabs    z18.d, p0/m, z19.d
            fadd    v0.2d, v0.2d, v18.2d
            ldp     q18, q19, [x0, #0x20]
            ptrue   p0.d
            fabs    z18.d, p0/m, z18.d
            fadd    v0.2d, v0.2d, v18.2d
            ptrue   p0.d
            fabs    z18.d, p0/m, z19.d
            fadd    v0.2d, v0.2d, v18.2d
            ldp     q18, q19, [x0, #0x40]
            ptrue   p0.d
            fabs    z18.d, p0/m, z18.d
            fadd    v0.2d, v0.2d, v18.2d
            ptrue   p0.d
            fabs    z18.d, p0/m, z19.d
            fadd    v0.2d, v0.2d, v18.2d
            ldp     q18, q19, [x0, #0x60]
            ptrue   p0.d
            fabs    z18.d, p0/m, z18.d
            fadd    v0.2d, v0.2d, v18.2d
            ptrue   p0.d
            fabs    z18.d, p0/m, z19.d
            fadd    v0.2d, v0.2d, v18.2d
            add     x0, x0, #128
            sub     x1, x1, #16
            cmp     x1, #16
            bhs     G_M34563_IG04

Either we can track the number of usages (probably count separate groups that do not cross call boundary) and then store them as LclVar. We could have a dedicated predicate register to represent ptrue value, but we would need to have different registers depending on the type (ptrue p0.b and ptrue p0.d are different).

Dominant language: C#
Stars: 18.3k
Forks: 5.6k
PR merge metrics: PR metrics pending

Contributor guide

Open the contributing guide

First steps

Read the whole issue, then the project's contributing guide.
Comment on the issue to say you are picking it up — it saves two people doing the same work.
Fork the repository and make your change on a branch.
Open a pull request that references the issue number.

Similar issues

Tsavorite.test fails to compile on main: TestBase.cs references Garnet.test.TestUtils, which is not linked into Tsavorite.test.csproj Open

Difficulty 2/5 1-3 hours Newbie friendliness 86/100

microsoft/garnet#2161 · 2 comments · Opened Sep 20, 2026
v11 update: Validation attributes are no longer experimental Open

:watch: Not Triaged 11.0 fundamentals/subsvc

Difficulty 2/5 1-3 hours Newbie friendliness 92/100

dotnet/AspNetCore.Docs#37699 · Opened Sep 20, 2026
SE 5.3.0 beta 7: In Settings Video player download buttons aren't together Open

Difficulty 2/5 1-3 hours Newbie friendliness 72/100

SubtitleEdit/subtitleedit#15108 · 1 comment · Opened Sep 20, 2026
Kubernetes get-started create-component.md: Go and C# silently default isMinikube instead of requiring it, unlike TS/Python/Java/HCL Open

area/docs-content Bug pulumi/docs

Difficulty 1/5 1-3 hours Newbie friendliness 94/100

pulumi/docs#21762 · Opened Sep 20, 2026
Create parent directories only after the containment check in InstallHelper.TryExtractToDirectory Open

Difficulty 2/5 1-3 hours Newbie friendliness 78/100

PowerShell/PSResourceGet#2056 · Opened Sep 20, 2026

More C# issues