[X86] Add the FSRM feature (Fast Short Rep Mov) to Zen3.

[hjyamauchi]

Summary

Note -x86-use-fsrm-for-memcpy is still disabled by default and there's no
default behavior change.

[GGanesh]

LGTM!

The enhancement applies to specific string lengths. I will check it and if need be will submit changes with respect to this prerequisite.

[davidxl]

lgtm

llvm/lib/Target/X86/X86.td

list<SubtargetFeature> ZNTuning = [FeatureFastLZCNT,
                                      FeatureSlowSHLD,
                                      FeatureInsertVZEROUPPER];
   list<SubtargetFeature> ZN2AdditionalFeatures = [FeatureCLWB,
                                                   FeatureRDPID,
                                                   FeatureWBNOINVD];
   list<SubtargetFeature> ZN2Tuning = ZNTuning;
   list<SubtargetFeature> ZN2Features =
     !listconcat(ZNFeatures, ZN2AdditionalFeatures);
-  list<SubtargetFeature> ZN3AdditionalFeatures = [FeatureINVPCID,
+  list<SubtargetFeature> ZN3AdditionalFeatures = [FeatureFSRM,
+                                                  FeatureINVPCID,
                                                   FeaturePKU,
                                                   FeatureVAES,
                                                   FeatureVPCLMULQDQ];
   list<SubtargetFeature> ZN3Tuning = ZNTuning;
   list<SubtargetFeature> ZN3Features =
     !listconcat(ZN2Features, ZN3AdditionalFeatures);
 }
 

llvm/test/CodeGen/X86/memcpy-inline-fsrm.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=x86_64-linux-gnu -x86-use-fsrm-for-memcpy -mattr=-fsrm < %s -o - | FileCheck %s --check-prefix=NOFSRM
 ; RUN: llc -mtriple=x86_64-linux-gnu -x86-use-fsrm-for-memcpy -mattr=+fsrm < %s -o - | FileCheck %s --check-prefix=FSRM
 ; RUN: llc -mtriple=x86_64-linux-gnu -x86-use-fsrm-for-memcpy -mcpu=haswell < %s | FileCheck %s --check-prefix=NOFSRM
 ; RUN: llc -mtriple=x86_64-linux-gnu -x86-use-fsrm-for-memcpy -mcpu=icelake-client < %s | FileCheck %s --check-prefix=FSRM
 ; RUN: llc -mtriple=x86_64-linux-gnu -x86-use-fsrm-for-memcpy -mcpu=icelake-server < %s | FileCheck %s --check-prefix=FSRM
+; RUN: llc -mtriple=x86_64-linux-gnu -x86-use-fsrm-for-memcpy -mcpu=znver3 < %s | FileCheck %s --check-prefix=FSRM
 
 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i1) nounwind
 
 define void @test1(i8* %a, i8* %b, i64 %s) nounwind {
 ; NOFSRM-LABEL: test1
 ; NOFSRM:       # %bb.0:
 ; NOFSRM:         jmp memcpy
 ;