[PowerPC] Emit dcbt and dcbtst in place of their extended mnemonics on AIX

[Conanap]

Summary

On AIX, the system assembler does not support the extended mnemonics
dcbtt and dcbtstt. This patch stops them from being emitted on
AIX and emits the base mnemonics instead, dcbt X, X, 16 and
dcbtstt X, X, 16 respectively.

[jsji]

Thanks for working on this.

llvm/lib/Target/PowerPC/MCTargetDesc/PPCInstPrinter.cpp

161

What do you mean by still requires dcbtt and dcbtstt implementation?

164

I think it is better to use the same condition as ModernAs Predicate , so that we may be able to enable it with feature bit if needed.

(!TT.isOSAIX() || STI.getFeatureBits()[PPC::HasModernAIXAs])

[nemanjai]

Please change the condition as Jinsong suggested and change the leading comment to:

// On AIX, only emit the extended mnemonics for dcbt and dcbtst if
// the "modern assembler" is available.

[Conanap]

Added mordern assembler check

[nemanjai]

LGTM. This is already approved, but just wanted to indicate that I think the comments were addressed adequately.

llvm/lib/Target/PowerPC/MCTargetDesc/PPCInstPrinter.cpp

void PPCInstPrinter::printInst(const MCInst *MI, uint64_t Address,
   // dcbt[st] is printed manually here because:
   //  1. The assembly syntax is different between embedded and server targets
   //  2. We must print the short mnemonics for TH == 0 because the
   //     embedded/server syntax default will not be stable across assemblers
   //  The syntax for dcbt is:
   //    dcbt ra, rb, th [server]
   //    dcbt th, ra, rb [embedded]
   //  where th can be omitted when it is 0. dcbtst is the same.
-  if (MI->getOpcode() == PPC::DCBT || MI->getOpcode() == PPC::DCBTST) {
+  // TODO: AIX still requires dcbtt and dcbtstt implementation,
+  // remove AIX OS check when dcbtt and dcbtstt become available.
+  if ((MI->getOpcode() == PPC::DCBT || MI->getOpcode() == PPC::DCBTST) &&
+      !TT.isOSAIX()) {
     unsigned char TH = MI->getOperand(0).getImm();
     O << "\tdcbt";
     if (MI->getOpcode() == PPC::DCBTST)
       O << "st";
     if (TH == 16)
       O << "t";
     O << " ";
 

llvm/test/CodeGen/PowerPC/builtins-ppc-xlcompat-prefetch.ll

 ; CHECK-NEXT:    ld 3, 0(3)
 ; CHECK-NEXT:    dcbtstt 0, 3
 ; CHECK-NEXT:    blr
 ;
 ; CHECK-AIX-LABEL: test_dcbtstt:
 ; CHECK-AIX:       # %bb.0: # %entry
 ; CHECK-AIX-NEXT:    lwz 3, L..C0(2) # @vpa
 ; CHECK-AIX-NEXT:    lwz 3, 0(3)
-; CHECK-AIX-NEXT:    dcbtstt 0, 3
+; CHECK-AIX-NEXT:    dcbtst 0, 3, 16
 ; CHECK-AIX-NEXT:    blr
 ;
 ; CHECK-AIX64-LABEL: test_dcbtstt:
 ; CHECK-AIX64:       # %bb.0: # %entry
 ; CHECK-AIX64-NEXT:    ld 3, L..C0(2) # @vpa
 ; CHECK-AIX64-NEXT:    ld 3, 0(3)
-; CHECK-AIX64-NEXT:    dcbtstt 0, 3
+; CHECK-AIX64-NEXT:    dcbtst 0, 3, 16
 ; CHECK-AIX64-NEXT:    blr
 entry:
   %0 = load i8*, i8** @vpa, align 8
   tail call void @llvm.ppc.dcbtstt(i8* %0)
   ret void
 }
 
 
 define dso_local void @test_dcbtt() {
 ; CHECK-LABEL: test_dcbtt:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    addis 3, 2, .LC0@toc@ha
 ; CHECK-NEXT:    ld 3, .LC0@toc@l(3)
 ; CHECK-NEXT:    ld 3, 0(3)
 ; CHECK-NEXT:    dcbtt 0, 3
 ; CHECK-NEXT:    blr
 ;
 ; CHECK-AIX-LABEL: test_dcbtt:
 ; CHECK-AIX:       # %bb.0: # %entry
 ; CHECK-AIX-NEXT:    lwz 3, L..C0(2) # @vpa
 ; CHECK-AIX-NEXT:    lwz 3, 0(3)
-; CHECK-AIX-NEXT:    dcbtt 0, 3
+; CHECK-AIX-NEXT:    dcbt 0, 3, 16
 ; CHECK-AIX-NEXT:    blr
 ;
 ; CHECK-AIX64-LABEL: test_dcbtt:
 ; CHECK-AIX64:       # %bb.0: # %entry
 ; CHECK-AIX64-NEXT:    ld 3, L..C0(2) # @vpa
 ; CHECK-AIX64-NEXT:    ld 3, 0(3)
-; CHECK-AIX64-NEXT:    dcbtt 0, 3
+; CHECK-AIX64-NEXT:    dcbt 0, 3, 16
 ; CHECK-AIX64-NEXT:    blr
 entry:
   %0 = load i8*, i8** @vpa, align 8
   tail call void @llvm.ppc.dcbtt(i8* %0)
   ret void
 }