From a3b8695bf5927f0a43d295dfdfeafeef4da022ea Mon Sep 17 00:00:00 2001
From: Roman Lebedev <lebedev.ri@gmail.com>
Date: Tue, 1 Jun 2021 10:46:05 +0300
Subject: [PATCH] [X86] AMD Zen 3 has fast variable per-lane shuffles

... but lane-crossing shuffles are slow.
---
 llvm/lib/Target/X86/X86.td | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/llvm/lib/Target/X86/X86.td b/llvm/lib/Target/X86/X86.td
index c6dbbf5e961f..53bbd93798ac 100644
--- a/llvm/lib/Target/X86/X86.td
+++ b/llvm/lib/Target/X86/X86.td
@@ -1111,7 +1111,9 @@ def ProcessorFeatures {
                                                   FeaturePKU,
                                                   FeatureVAES,
                                                   FeatureVPCLMULQDQ];
-  list<SubtargetFeature> ZN3AdditionalTuning = [FeatureMacroFusion];
+  list<SubtargetFeature> ZN3AdditionalTuning =
+    [FeatureMacroFusion,
+     FeatureFastVariablePerLaneShuffle];
   list<SubtargetFeature> ZN3Tuning =
     !listconcat(ZNTuning, ZN3AdditionalTuning);
   list<SubtargetFeature> ZN3Features =