[LLVM][CodeGen][X86] Support larger NOOPs on x86 when available

The mnop-mcount attribute benefits from using a single no-op
instruction instead of 3 when supported.

Update codegen to support single instruction 3-10 byte NOOPs.

Author: pmur

llvm/lib/Target/X86/X86MCInstLower.cpp

@@ -664,9 +664,8 @@ static unsigned emitNop(MCStreamer &OS, unsigned NumBytes,
   // target cpu.  15-bytes is the longest single NOP instruction, but some
   // platforms can't decode the longest forms efficiently.
   unsigned MaxNopLength = 1;
+  unsigned BaseReg = X86::RAX;
   if (Subtarget->is64Bit()) {
-    // FIXME: We can use NOOPL on 32-bit targets with FeatureNOPL, but the
-    // IndexReg/BaseReg below need to be updated.
     if (Subtarget->hasFeature(X86::TuningFast7ByteNOP))
       MaxNopLength = 7;
     else if (Subtarget->hasFeature(X86::TuningFast15ByteNOP))
@@ -675,16 +674,19 @@ static unsigned emitNop(MCStreamer &OS, unsigned NumBytes,
       MaxNopLength = 11;
     else
       MaxNopLength = 10;
-  } if (Subtarget->is32Bit())
+  } else if (Subtarget->is32Bit() && Subtarget->getFeatureBits()[X86::FeatureNOPL]) {
+    BaseReg = X86::EAX;
+    MaxNopLength = 10;
+  }
+  else if (Subtarget->is32Bit())
     MaxNopLength = 2;
 
   // Cap a single nop emission at the profitable value for the target
   NumBytes = std::min(NumBytes, MaxNopLength);
 
   unsigned NopSize;
-  unsigned Opc, BaseReg, ScaleVal, IndexReg, Displacement, SegmentReg;
+  unsigned Opc, ScaleVal, IndexReg, Displacement, SegmentReg;
   IndexReg = Displacement = SegmentReg = 0;
-  BaseReg = X86::RAX;
   ScaleVal = 1;
   switch (NumBytes) {
   case 0:
@@ -711,13 +713,13 @@ static unsigned emitNop(MCStreamer &OS, unsigned NumBytes,
     NopSize = 5;
     Opc = X86::NOOPL;
     Displacement = 8;
-    IndexReg = X86::RAX;
+    IndexReg = BaseReg;
     break;
   case 6:
     NopSize = 6;
     Opc = X86::NOOPW;
     Displacement = 8;
-    IndexReg = X86::RAX;
+    IndexReg = BaseReg;
     break;
   case 7:
     NopSize = 7;
@@ -728,19 +730,19 @@ static unsigned emitNop(MCStreamer &OS, unsigned NumBytes,
     NopSize = 8;
     Opc = X86::NOOPL;
     Displacement = 512;
-    IndexReg = X86::RAX;
+    IndexReg = BaseReg;
     break;
   case 9:
     NopSize = 9;
     Opc = X86::NOOPW;
     Displacement = 512;
-    IndexReg = X86::RAX;
+    IndexReg = BaseReg;
     break;
   default:
     NopSize = 10;
     Opc = X86::NOOPW;
     Displacement = 512;
-    IndexReg = X86::RAX;
+    IndexReg = BaseReg;
     SegmentReg = X86::CS;
     break;
   }

llvm/test/CodeGen/X86/mnop-mcount-01.ll

@@ -2,6 +2,8 @@
 ; RUN:   | FileCheck --check-prefixes=X64 %s
 ; RUN: llc %s -mtriple=i686-linux-gnu -o - -verify-machineinstrs \
 ; RUN:   | FileCheck --check-prefix=X86 %s
+; RUN: llc %s -mtriple=i686-linux-gnu -mcpu=pentium_pro -o - -verify-machineinstrs \
+; RUN:   | FileCheck --check-prefix=PPRO %s
 
 define void @test1() #0 {
 entry:
@@ -13,6 +15,9 @@ entry:
 ; X86-LABEL: @test1
 ; X86: calll __fentry__
 ; X86: retl
+; PPRO-LABEL: @test1
+; PPRO: calll __fentry__
+; PPRO: retl
 }
 
 define void @test2() #1 {
@@ -29,6 +34,10 @@ entry:
 ; X86: xchgw %ax, %ax
 ; X86: nop
 ; X86: retl
+; PPRO-LABEL: @test2
+; PPRO: nopl 8(%eax,%eax)
+; PPRO-NOT: calll __fentry__
+; PPRO: retl
 }
 
 attributes #0 = { "fentry-call"="true" }

llvm/test/CodeGen/X86/mrecord-mcount-01.ll

@@ -2,6 +2,8 @@
 ; RUN:   | FileCheck --check-prefix=X64 %s
 ; RUN: llc %s -mtriple=i686-linux-gnu -o - -verify-machineinstrs \
 ; RUN:   | FileCheck --check-prefix=X86 %s
+; RUN: llc %s -mtriple=i686-linux-gnu -mcpu=pentium_pro -o - -verify-machineinstrs \
+; RUN:   | FileCheck --check-prefix=PPRO %s
 
 define void @test1() #0 {
 entry:
@@ -21,6 +23,13 @@ entry:
 ; X86: .Ltmp0:
 ; X86: calll __fentry__
 ; X86: retl
+; PPRO-LABEL: test1:
+; PPRO: .section __mcount_loc,"a",@progbits
+; PPRO: .long .Ltmp0
+; PPRO: .text
+; PPRO: .Ltmp0:
+; PPRO: calll __fentry__
+; PPRO: retl
 }
 
 define void @test2() #1 {
@@ -45,6 +54,14 @@ entry:
 ; X86: nop
 ; X86-NOT: calll __fentry__
 ; X86: retl
+; PPRO-LABEL: test2:
+; PPRO: .section __mcount_loc,"a",@progbits
+; PPRO: .long .Ltmp1
+; PPRO: .text
+; PPRO: .Ltmp1:
+; PPRO: nopl 8(%eax,%eax)
+; PPRO-NOT: calll __fentry__
+; PPRO: retl
 }
 
 attributes #0 = { "fentry-call"="true" "mrecord-mcount" }