[X86] Allow 8-bit INC/DEC to be converted to LEA.

We already do this for 16/32/64 as well as 8-bit add with register/immediate. Might as well do it for 8-bit INC/DEC too.

Differential Revision: https://reviews.llvm.org/D58869

llvm-svn: 355424

llvm/lib/Target/X86/X86InstrArithmetic.td

@@ -435,11 +435,10 @@ def X86sub_flag_nocf : PatFrag<(ops node:$lhs, node:$rhs),
 // TODO: inc/dec is slow for P4, but fast for Pentium-M.
 let Defs = [EFLAGS] in {
 let Constraints = "$src1 = $dst", SchedRW = [WriteALU] in {
-let CodeSize = 2 in
+let isConvertibleToThreeAddress = 1, CodeSize = 2 in { // Can xform into LEA.
 def INC8r  : I<0xFE, MRM0r, (outs GR8 :$dst), (ins GR8 :$src1),
                "inc{b}\t$dst",
                [(set GR8:$dst, EFLAGS, (X86add_flag_nocf GR8:$src1, 1))]>;
-let isConvertibleToThreeAddress = 1, CodeSize = 2 in { // Can xform into LEA.
 def INC16r : I<0xFF, MRM0r, (outs GR16:$dst), (ins GR16:$src1),
                "inc{w}\t$dst",
                [(set GR16:$dst, EFLAGS, (X86add_flag_nocf GR16:$src1, 1))]>,
@@ -483,11 +482,10 @@ let Predicates = [UseIncDec, In64BitMode] in {
 } // CodeSize = 2, SchedRW
 
 let Constraints = "$src1 = $dst", SchedRW = [WriteALU] in {
-let CodeSize = 2 in
+let isConvertibleToThreeAddress = 1, CodeSize = 2 in { // Can xform into LEA.
 def DEC8r  : I<0xFE, MRM1r, (outs GR8 :$dst), (ins GR8 :$src1),
                "dec{b}\t$dst",
                [(set GR8:$dst, EFLAGS, (X86sub_flag_nocf GR8:$src1, 1))]>;
-let isConvertibleToThreeAddress = 1, CodeSize = 2 in { // Can xform into LEA.
 def DEC16r : I<0xFF, MRM1r, (outs GR16:$dst), (ins GR16:$src1),
                "dec{w}\t$dst",
                [(set GR16:$dst, EFLAGS, (X86sub_flag_nocf GR16:$src1, 1))]>,

llvm/lib/Target/X86/X86InstrInfo.cpp

@@ -761,9 +761,11 @@ MachineInstr *X86InstrInfo::convertToThreeAddressWithLEA(
        .addReg(InRegLEA, RegState::Kill).addImm(0).addReg(0);
     break;
   }
+  case X86::INC8r:
   case X86::INC16r:
     addRegOffset(MIB, InRegLEA, true, 1);
     break;
+  case X86::DEC8r:
   case X86::DEC16r:
     addRegOffset(MIB, InRegLEA, true, -1);
     break;
@@ -945,8 +947,6 @@ X86InstrInfo::convertToThreeAddress(MachineFunction::iterator &MFI,
     NewMI = addOffset(MIB, 1);
     break;
   }
-  case X86::INC16r:
-    return convertToThreeAddressWithLEA(MIOpc, MFI, MI, LV, Is8BitOp);
   case X86::DEC64r:
   case X86::DEC32r: {
     assert(MI.getNumOperands() >= 2 && "Unknown dec instruction!");
@@ -970,7 +970,12 @@ X86InstrInfo::convertToThreeAddress(MachineFunction::iterator &MFI,
 
     break;
   }
+  case X86::DEC8r:
+  case X86::INC8r:
+    Is8BitOp = true;
+    LLVM_FALLTHROUGH;
   case X86::DEC16r:
+  case X86::INC16r:
     return convertToThreeAddressWithLEA(MIOpc, MFI, MI, LV, Is8BitOp);
   case X86::ADD64rr:
   case X86::ADD64rr_DB:

llvm/test/CodeGen/X86/MergeConsecutiveStores.ll

@@ -632,7 +632,7 @@ define void @loadStoreBaseIndexOffsetSextNoSex(i8* %a, i8* %b, i8* %c, i32 %n) {
 ; BWON-NEXT:  .LBB12_1: # =>This Inner Loop Header: Depth=1
 ; BWON-NEXT:    movsbq (%rdi,%rcx), %rax
 ; BWON-NEXT:    movzbl (%rdx,%rax), %r9d
-; BWON-NEXT:    incb %al
+; BWON-NEXT:    leal 1(%rax), %eax
 ; BWON-NEXT:    movsbq %al, %rax
 ; BWON-NEXT:    movzbl (%rdx,%rax), %eax
 ; BWON-NEXT:    movb %r9b, (%rsi,%rcx,2)
@@ -651,7 +651,7 @@ define void @loadStoreBaseIndexOffsetSextNoSex(i8* %a, i8* %b, i8* %c, i32 %n) {
 ; BWOFF-NEXT:  .LBB12_1: # =>This Inner Loop Header: Depth=1
 ; BWOFF-NEXT:    movsbq (%rdi,%rcx), %rax
 ; BWOFF-NEXT:    movb (%rdx,%rax), %r9b
-; BWOFF-NEXT:    incb %al
+; BWOFF-NEXT:    leal 1(%rax), %eax
 ; BWOFF-NEXT:    movsbq %al, %rax
 ; BWOFF-NEXT:    movb (%rdx,%rax), %al
 ; BWOFF-NEXT:    movb %r9b, (%rsi,%rcx,2)

llvm/test/CodeGen/X86/copy-eflags.ll

@@ -43,19 +43,17 @@ define i32 @test1() nounwind {
 ;
 ; X64-LABEL: test1:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movb {{.*}}(%rip), %dil
-; X64-NEXT:    movl %edi, %eax
-; X64-NEXT:    incb %al
+; X64-NEXT:    movb {{.*}}(%rip), %cl
+; X64-NEXT:    leal 1(%rcx), %eax
 ; X64-NEXT:    movb %al, {{.*}}(%rip)
 ; X64-NEXT:    incl {{.*}}(%rip)
-; X64-NEXT:    sete %sil
-; X64-NEXT:    movb {{.*}}(%rip), %cl
-; X64-NEXT:    movl %ecx, %edx
-; X64-NEXT:    incb %dl
-; X64-NEXT:    cmpb %dil, %cl
+; X64-NEXT:    sete %dl
+; X64-NEXT:    movb {{.*}}(%rip), %sil
+; X64-NEXT:    leal 1(%rsi), %edi
+; X64-NEXT:    cmpb %cl, %sil
 ; X64-NEXT:    sete {{.*}}(%rip)
-; X64-NEXT:    movb %dl, {{.*}}(%rip)
-; X64-NEXT:    testb %sil, %sil
+; X64-NEXT:    movb %dil, {{.*}}(%rip)
+; X64-NEXT:    testb %dl, %dl
 ; X64-NEXT:    jne .LBB0_2
 ; X64-NEXT:  # %bb.1: # %if.then
 ; X64-NEXT:    pushq %rax