[X86] Make ADD*_DB post-RA pseudos and expand them in expandPostRAPseudo.

These are used to help convert OR->LEA when needed to avoid avoid a copy. They
aren't need after register allocation.

Happens to remove an ugly goto from X86MCCodeEmitter.cpp

llvm-svn: 356356

diff --git a/llvm/lib/Target/X86/X86InstrCompiler.td b/llvm/lib/Target/X86/X86InstrCompiler.td
index e4f324e..44f2ac4 100644 (file)
--- a/llvm/lib/Target/X86/X86InstrCompiler.td
+++ b/llvm/lib/Target/X86/X86InstrCompiler.td
@@ -1378,7 +1378,7 @@ def or_is_add : PatFrag<(ops node:$lhs, node:$rhs), (or node:$lhs, node:$rhs),[{
 // Try this before the selecting to OR.
 let SchedRW = [WriteALU] in {
 
-let isConvertibleToThreeAddress = 1,
+let isConvertibleToThreeAddress = 1, isPseudo = 1,
     Constraints = "$src1 = $dst", Defs = [EFLAGS] in {
 let isCommutable = 1 in {
 def ADD8rr_DB   : I<0, Pseudo, (outs GR8:$dst), (ins GR8:$src1, GR8:$src2),

diff --git a/llvm/lib/Target/X86/X86InstrFoldTables.cpp b/llvm/lib/Target/X86/X86InstrFoldTables.cpp
index 5cd327f..d8c21f6 100644 (file)
--- a/llvm/lib/Target/X86/X86InstrFoldTables.cpp
+++ b/llvm/lib/Target/X86/X86InstrFoldTables.cpp
@@ -33,6 +33,17 @@ using namespace llvm;
 // tables that would be incorrect. The manual review process allows us a chance
 // to catch these before they become observable bugs.
 static const X86MemoryFoldTableEntry MemoryFoldTable2Addr[] = {
+  { X86::ADD16ri8_DB, X86::ADD16mi8,   TB_NO_REVERSE },
+  { X86::ADD16ri_DB,  X86::ADD16mi,    TB_NO_REVERSE },
+  { X86::ADD16rr_DB,  X86::ADD16mr,    TB_NO_REVERSE },
+  { X86::ADD32ri8_DB, X86::ADD32mi8,   TB_NO_REVERSE },
+  { X86::ADD32ri_DB,  X86::ADD32mi,    TB_NO_REVERSE },
+  { X86::ADD32rr_DB,  X86::ADD32mr,    TB_NO_REVERSE },
+  { X86::ADD64ri32_DB,X86::ADD64mi32,  TB_NO_REVERSE },
+  { X86::ADD64ri8_DB, X86::ADD64mi8,   TB_NO_REVERSE },
+  { X86::ADD64rr_DB,  X86::ADD64mr,    TB_NO_REVERSE },
+  { X86::ADD8ri_DB,   X86::ADD8mi,     TB_NO_REVERSE },
+  { X86::ADD8rr_DB,   X86::ADD8mr,     TB_NO_REVERSE },
   { X86::ADC16ri,     X86::ADC16mi,    0 },
   { X86::ADC16ri8,    X86::ADC16mi8,   0 },
   { X86::ADC16rr,     X86::ADC16mr,    0 },
@@ -47,27 +58,16 @@ static const X86MemoryFoldTableEntry MemoryFoldTable2Addr[] = {
   { X86::ADC8rr,      X86::ADC8mr,     0 },
   { X86::ADD16ri,     X86::ADD16mi,    0 },
   { X86::ADD16ri8,    X86::ADD16mi8,   0 },
-  { X86::ADD16ri8_DB, X86::ADD16mi8,   TB_NO_REVERSE },
-  { X86::ADD16ri_DB,  X86::ADD16mi,    TB_NO_REVERSE },
   { X86::ADD16rr,     X86::ADD16mr,    0 },
-  { X86::ADD16rr_DB,  X86::ADD16mr,    TB_NO_REVERSE },
   { X86::ADD32ri,     X86::ADD32mi,    0 },
   { X86::ADD32ri8,    X86::ADD32mi8,   0 },
-  { X86::ADD32ri8_DB, X86::ADD32mi8,   TB_NO_REVERSE },
-  { X86::ADD32ri_DB,  X86::ADD32mi,    TB_NO_REVERSE },
   { X86::ADD32rr,     X86::ADD32mr,    0 },
-  { X86::ADD32rr_DB,  X86::ADD32mr,    TB_NO_REVERSE },
   { X86::ADD64ri32,   X86::ADD64mi32,  0 },
-  { X86::ADD64ri32_DB,X86::ADD64mi32,  TB_NO_REVERSE },
   { X86::ADD64ri8,    X86::ADD64mi8,   0 },
-  { X86::ADD64ri8_DB, X86::ADD64mi8,   TB_NO_REVERSE },
   { X86::ADD64rr,     X86::ADD64mr,    0 },
-  { X86::ADD64rr_DB,  X86::ADD64mr,    TB_NO_REVERSE },
   { X86::ADD8ri,      X86::ADD8mi,     0 },
   { X86::ADD8ri8,     X86::ADD8mi8,    0 },
-  { X86::ADD8ri_DB,   X86::ADD8mi,     TB_NO_REVERSE },
   { X86::ADD8rr,      X86::ADD8mr,     0 },
-  { X86::ADD8rr_DB,   X86::ADD8mr,     TB_NO_REVERSE },
   { X86::AND16ri,     X86::AND16mi,    0 },
   { X86::AND16ri8,    X86::AND16mi8,   0 },
   { X86::AND16rr,     X86::AND16mr,    0 },
@@ -1207,6 +1207,10 @@ static const X86MemoryFoldTableEntry MemoryFoldTable1[] = {
 };
 
 static const X86MemoryFoldTableEntry MemoryFoldTable2[] = {
+  { X86::ADD16rr_DB,               X86::ADD16rm,                  TB_NO_REVERSE },
+  { X86::ADD32rr_DB,               X86::ADD32rm,                  TB_NO_REVERSE },
+  { X86::ADD64rr_DB,               X86::ADD64rm,                  TB_NO_REVERSE },
+  { X86::ADD8rr_DB,                X86::ADD8rm,                   TB_NO_REVERSE },
   { X86::ADC16rr,                  X86::ADC16rm,                  0 },
   { X86::ADC32rr,                  X86::ADC32rm,                  0 },
   { X86::ADC64rr,                  X86::ADC64rm,                  0 },
@@ -1214,13 +1218,9 @@ static const X86MemoryFoldTableEntry MemoryFoldTable2[] = {
   { X86::ADCX32rr,                 X86::ADCX32rm,                 0 },
   { X86::ADCX64rr,                 X86::ADCX64rm,                 0 },
   { X86::ADD16rr,                  X86::ADD16rm,                  0 },
-  { X86::ADD16rr_DB,               X86::ADD16rm,                  TB_NO_REVERSE },
   { X86::ADD32rr,                  X86::ADD32rm,                  0 },
-  { X86::ADD32rr_DB,               X86::ADD32rm,                  TB_NO_REVERSE },
   { X86::ADD64rr,                  X86::ADD64rm,                  0 },
-  { X86::ADD64rr_DB,               X86::ADD64rm,                  TB_NO_REVERSE },
   { X86::ADD8rr,                   X86::ADD8rm,                   0 },
-  { X86::ADD8rr_DB,                X86::ADD8rm,                   TB_NO_REVERSE },
   { X86::ADDPDrr,                  X86::ADDPDrm,                  TB_ALIGN_16 },
   { X86::ADDPSrr,                  X86::ADDPSrm,                  TB_ALIGN_16 },
   { X86::ADDSDrr,                  X86::ADDSDrm,                  0 },

diff --git a/llvm/lib/Target/X86/X86InstrInfo.cpp b/llvm/lib/Target/X86/X86InstrInfo.cpp
index 21db322..c03b602 100644 (file)
--- a/llvm/lib/Target/X86/X86InstrInfo.cpp
+++ b/llvm/lib/Target/X86/X86InstrInfo.cpp
@@ -4237,6 +4237,17 @@ bool X86InstrInfo::expandPostRAPseudo(MachineInstr &MI) const {
   case X86::XOR64_FP:
   case X86::XOR32_FP:
     return expandXorFP(MIB, *this);
+  case X86::ADD8rr_DB:    MIB->setDesc(get(X86::OR8rr));    break;
+  case X86::ADD16rr_DB:   MIB->setDesc(get(X86::OR16rr));   break;
+  case X86::ADD32rr_DB:   MIB->setDesc(get(X86::OR32rr));   break;
+  case X86::ADD64rr_DB:   MIB->setDesc(get(X86::OR64rr));   break;
+  case X86::ADD8ri_DB:    MIB->setDesc(get(X86::OR8ri));    break;
+  case X86::ADD16ri_DB:   MIB->setDesc(get(X86::OR16ri));   break;
+  case X86::ADD32ri_DB:   MIB->setDesc(get(X86::OR32ri));   break;
+  case X86::ADD64ri32_DB: MIB->setDesc(get(X86::OR64ri32)); break;
+  case X86::ADD16ri8_DB:  MIB->setDesc(get(X86::OR16ri8));  break;
+  case X86::ADD32ri8_DB:  MIB->setDesc(get(X86::OR32ri8));  break;
+  case X86::ADD64ri8_DB:  MIB->setDesc(get(X86::OR64ri8));  break;
   }
   return false;
 }

diff --git a/llvm/lib/Target/X86/X86MCInstLower.cpp b/llvm/lib/Target/X86/X86MCInstLower.cpp
index 4b62085..cd1dcf3 100644 (file)
--- a/llvm/lib/Target/X86/X86MCInstLower.cpp
+++ b/llvm/lib/Target/X86/X86MCInstLower.cpp
@@ -435,7 +435,6 @@ void X86MCInstLower::Lower(const MachineInstr *MI, MCInst &OutMI) const {
       OutMI.addOperand(MaybeMCOp.getValue());
 
   // Handle a few special cases to eliminate operand modifiers.
-ReSimplify:
   switch (OutMI.getOpcode()) {
   case X86::LEA64_32r:
   case X86::LEA64r:
@@ -583,21 +582,6 @@ ReSimplify:
     }
     break;
 
-  // These are pseudo-ops for OR to help with the OR->ADD transformation.  We do
-  // this with an ugly goto in case the resultant OR uses EAX and needs the
-  // short form.
-  case X86::ADD8rr_DB:    OutMI.setOpcode(X86::OR8rr);    goto ReSimplify;
-  case X86::ADD16rr_DB:   OutMI.setOpcode(X86::OR16rr);   goto ReSimplify;
-  case X86::ADD32rr_DB:   OutMI.setOpcode(X86::OR32rr);   goto ReSimplify;
-  case X86::ADD64rr_DB:   OutMI.setOpcode(X86::OR64rr);   goto ReSimplify;
-  case X86::ADD8ri_DB:    OutMI.setOpcode(X86::OR8ri);    goto ReSimplify;
-  case X86::ADD16ri_DB:   OutMI.setOpcode(X86::OR16ri);   goto ReSimplify;
-  case X86::ADD32ri_DB:   OutMI.setOpcode(X86::OR32ri);   goto ReSimplify;
-  case X86::ADD64ri32_DB: OutMI.setOpcode(X86::OR64ri32); goto ReSimplify;
-  case X86::ADD16ri8_DB:  OutMI.setOpcode(X86::OR16ri8);  goto ReSimplify;
-  case X86::ADD32ri8_DB:  OutMI.setOpcode(X86::OR32ri8);  goto ReSimplify;
-  case X86::ADD64ri8_DB:  OutMI.setOpcode(X86::OR64ri8);  goto ReSimplify;
-
   // We don't currently select the correct instruction form for instructions
   // which have a short %eax, etc. form. Handle this by custom lowering, for
   // now.