[PATCH] D152227: [X86] Prefer vmovmsk instead of vtest for alderlake.

Tue Jun 6 02:35:38 PDT 2023

RKSimon added inline comments.

================
Comment at: llvm/lib/Target/X86/X86.td:426
+def TuningSlowVtest : SubtargetFeature<"slow-vtest", "HasSlowVtest", "true",
+                                       "VTEST instruction is slow">;

----------------
pengfei wrote:
> LuoYuanke wrote:
> > goldstein.w.n wrote:
> > > Personally think slowvtest is a kind of confusing name b.c vtest also has a perf dropoff from SnB -> HSW.
> > > Maybe "PreferMovmskOverVTest" would be clearer?
> > I just follow the previous naming convention. I'm open to "PreferMovmskOverVTest". @RKSimon and @pengfei, what's your opinion?
> I think we should start from `Tuning`, but `TuningPreferMovmskOverVTest` looks verbose..
+1 TuningPreferMovmskOverVTest explains the purpose of the tuning flag better

================
Comment at: llvm/test/CodeGen/X86/combine-movmsk.ll:6
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=alderlake | FileCheck %s --check-prefixes=ADL

----------------
Add a variant that tests the tuning flag directly
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+slow-vtest | FileCheck %s --check-prefix=ADL

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D152227/new/

https://reviews.llvm.org/D152227