[PATCH] D123306: [X86] Enable fast variable per-lane shuffle tuning on all Ryzen targets

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon created this revision.
RKSimon added reviewers: lebedev.ri, craig.topper, GGanesh, spatel.
Herald added subscribers: StephenFan, pengfei, hiraditya.
Herald added a project: All.
RKSimon requested review of this revision.
Herald added a project: LLVM.

rGa3b8695bf592 <https://reviews.llvm.org/rGa3b8695bf5927f0a43d295dfdfeafeef4da022ea> enabled this for znver3, but AMD SoG, Agner and uops.info all agree that even znver1 has a fast per-lane shuffle op (VPSHUFB), but cross-lane shuffles seem to be slow (PERMPS etc.)

Fixes PR44795


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D123306

Files:
  llvm/lib/Target/X86/X86.td
  llvm/test/CodeGen/X86/vector-shuffle-fast-per-lane.ll


Index: llvm/test/CodeGen/X86/vector-shuffle-fast-per-lane.ll
===================================================================
--- llvm/test/CodeGen/X86/vector-shuffle-fast-per-lane.ll
+++ llvm/test/CodeGen/X86/vector-shuffle-fast-per-lane.ll
@@ -4,8 +4,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx512f,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=FAST
 ;
-; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=znver1 | FileCheck %s --check-prefixes=SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=znver2 | FileCheck %s --check-prefixes=SLOW
+; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=znver1 | FileCheck %s --check-prefixes=FAST
+; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=znver2 | FileCheck %s --check-prefixes=FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=znver3 | FileCheck %s --check-prefixes=FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=haswell | FileCheck %s --check-prefixes=FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=skx | FileCheck %s --check-prefixes=FAST
Index: llvm/lib/Target/X86/X86.td
===================================================================
--- llvm/lib/Target/X86/X86.td
+++ llvm/lib/Target/X86/X86.td
@@ -1193,6 +1193,7 @@
                                      TuningFastScalarFSQRT,
                                      TuningFastVectorFSQRT,
                                      TuningFastScalarShiftMasks,
+                                     TuningFastVariablePerLaneShuffle,
                                      TuningFastMOVBE,
                                      TuningSlowSHLD,
                                      TuningSBBDepBreaking,
@@ -1208,9 +1209,7 @@
                                                   FeaturePKU,
                                                   FeatureVAES,
                                                   FeatureVPCLMULQDQ];
-  list<SubtargetFeature> ZN3AdditionalTuning =
-    [TuningMacroFusion,
-     TuningFastVariablePerLaneShuffle];
+  list<SubtargetFeature> ZN3AdditionalTuning = [TuningMacroFusion];
   list<SubtargetFeature> ZN3Tuning =
     !listconcat(ZN2Tuning, ZN3AdditionalTuning);
   list<SubtargetFeature> ZN3Features =


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D123306.421181.patch
Type: text/x-patch
Size: 2292 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220407/1910e894/attachment.bin>