Codegen signed divides by 2 and -2 more efficiently. In particular
instead of:
s: ;; X / 2
movl 4(%esp), %eax
movl %eax, %ecx
shrl $31, %ecx
movl %eax, %edx
addl %ecx, %edx
sarl $1, %eax
ret
t: ;; X / -2
movl 4(%esp), %eax
movl %eax, %ecx
shrl $31, %ecx
movl %eax, %edx
addl %ecx, %edx
sarl $1, %eax
negl %eax
ret
Emit:
s:
movl 4(%esp), %eax
cmpl $-2147483648, %eax
sbbl $-1, %eax
sarl $1, %eax
ret
t:
movl 4(%esp), %eax
cmpl $-2147483648, %eax
sbbl $-1, %eax
sarl $1, %eax
negl %eax
ret
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@16760 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/X86/X86ISelSimple.cpp b/lib/Target/X86/X86ISelSimple.cpp
index d65b677..a47bcc3 100644
--- a/lib/Target/X86/X86ISelSimple.cpp
+++ b/lib/Target/X86/X86ISelSimple.cpp
@@ -2695,6 +2695,28 @@
return;
}
+ if (V == 2 || V == -2) { // X /s 2
+ static const unsigned CMPOpcode[] = {
+ X86::CMP8ri, X86::CMP16ri, X86::CMP32ri
+ };
+ static const unsigned SBBOpcode[] = {
+ X86::SBB8ri, X86::SBB16ri, X86::SBB32ri
+ };
+ unsigned Op0Reg = getReg(Op0, BB, IP);
+ unsigned SignBit = 1 << (CI->getType()->getPrimitiveSize()*8-1);
+ BuildMI(*BB, IP, CMPOpcode[Class], 2).addReg(Op0Reg).addImm(SignBit);
+
+ unsigned TmpReg = makeAnotherReg(Op0->getType());
+ BuildMI(*BB, IP, SBBOpcode[Class], 2, TmpReg).addReg(Op0Reg).addImm(-1);
+
+ unsigned TmpReg2 = V == 2 ? ResultReg : makeAnotherReg(Op0->getType());
+ BuildMI(*BB, IP, SAROpcode[Class], 2, TmpReg2).addReg(TmpReg).addImm(1);
+ if (V == -2) {
+ BuildMI(*BB, IP, NEGOpcode[Class], 1, ResultReg).addReg(TmpReg2);
+ }
+ return;
+ }
+
bool isNeg = false;
if (V < 0) { // Not a positive power of 2?
V = -V;