Fix a minor code-quality issue. When passing 8 and 16-bit integer constants

to function calls, we would emit dead code, like this:

int Y(int, short, double);
int X() {
  Y(4, 123, 4);
}

--- Old
X:
        sub %ESP, 20
        mov %EAX, 4
        mov DWORD PTR [%ESP], %EAX
***     mov %AX, 123
        mov %AX, 123
        movsx %EAX, %AX
        mov DWORD PTR [%ESP + 4], %EAX
        fld QWORD PTR [.CPIX_0]
        fstp QWORD PTR [%ESP + 8]
        call Y
        mov %EAX, 0
        # IMPLICIT_USE %EAX %ESP
        add %ESP, 20
        ret

Now we emit:
X:
        sub %ESP, 20
        mov %EAX, 4
        mov DWORD PTR [%ESP], %EAX
        mov %AX, 123
        movsx %EAX, %AX
        mov DWORD PTR [%ESP + 4], %EAX
        fld QWORD PTR [.CPIX_0]
        fstp QWORD PTR [%ESP + 8]
        call Y
        mov %EAX, 0
        # IMPLICIT_USE %EAX %ESP
        add %ESP, 20
        ret

Next up, eliminate the mov AX and movsx entirely!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@12026 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/InstSelectSimple.cpp b/lib/Target/X86/InstSelectSimple.cpp
index 0aa4ef9..b779664 100644 (file)
--- a/lib/Target/X86/InstSelectSimple.cpp
+++ b/lib/Target/X86/InstSelectSimple.cpp
@@ -1098,7 +1098,7 @@ void ISel::doCall(const ValueRecord &Ret, MachineInstr *CallMI,
     // Arguments go on the stack in reverse order, as specified by the ABI.
     unsigned ArgOffset = 0;
     for (unsigned i = 0, e = Args.size(); i != e; ++i) {
-      unsigned ArgReg = Args[i].Val ? getReg(Args[i].Val) : Args[i].Reg;
+      unsigned ArgReg;
       switch (getClassB(Args[i].Ty)) {
       case cByte:
       case cShort: {
@@ -1110,10 +1110,12 @@ void ISel::doCall(const ValueRecord &Ret, MachineInstr *CallMI,
         break;
       }
       case cInt:
+        ArgReg = Args[i].Val ? getReg(Args[i].Val) : Args[i].Reg;
         addRegOffset(BuildMI(BB, X86::MOV32mr, 5),
                      X86::ESP, ArgOffset).addReg(ArgReg);
         break;
       case cLong:
+        ArgReg = Args[i].Val ? getReg(Args[i].Val) : Args[i].Reg;
         addRegOffset(BuildMI(BB, X86::MOV32mr, 5),
                      X86::ESP, ArgOffset).addReg(ArgReg);
         addRegOffset(BuildMI(BB, X86::MOV32mr, 5),
@@ -1122,6 +1124,7 @@ void ISel::doCall(const ValueRecord &Ret, MachineInstr *CallMI,
         break;
         
       case cFP:
+        ArgReg = Args[i].Val ? getReg(Args[i].Val) : Args[i].Reg;
         if (Args[i].Ty == Type::FloatTy) {
           addRegOffset(BuildMI(BB, X86::FST32m, 5),
                        X86::ESP, ArgOffset).addReg(ArgReg);

diff --git a/lib/Target/X86/X86ISelSimple.cpp b/lib/Target/X86/X86ISelSimple.cpp
index 0aa4ef9..b779664 100644 (file)
--- a/lib/Target/X86/X86ISelSimple.cpp
+++ b/lib/Target/X86/X86ISelSimple.cpp
@@ -1098,7 +1098,7 @@ void ISel::doCall(const ValueRecord &Ret, MachineInstr *CallMI,
     // Arguments go on the stack in reverse order, as specified by the ABI.
     unsigned ArgOffset = 0;
     for (unsigned i = 0, e = Args.size(); i != e; ++i) {
-      unsigned ArgReg = Args[i].Val ? getReg(Args[i].Val) : Args[i].Reg;
+      unsigned ArgReg;
       switch (getClassB(Args[i].Ty)) {
       case cByte:
       case cShort: {
@@ -1110,10 +1110,12 @@ void ISel::doCall(const ValueRecord &Ret, MachineInstr *CallMI,
         break;
       }
       case cInt:
+        ArgReg = Args[i].Val ? getReg(Args[i].Val) : Args[i].Reg;
         addRegOffset(BuildMI(BB, X86::MOV32mr, 5),
                      X86::ESP, ArgOffset).addReg(ArgReg);
         break;
       case cLong:
+        ArgReg = Args[i].Val ? getReg(Args[i].Val) : Args[i].Reg;
         addRegOffset(BuildMI(BB, X86::MOV32mr, 5),
                      X86::ESP, ArgOffset).addReg(ArgReg);
         addRegOffset(BuildMI(BB, X86::MOV32mr, 5),
@@ -1122,6 +1124,7 @@ void ISel::doCall(const ValueRecord &Ret, MachineInstr *CallMI,
         break;
         
       case cFP:
+        ArgReg = Args[i].Val ? getReg(Args[i].Val) : Args[i].Reg;
         if (Args[i].Ty == Type::FloatTy) {
           addRegOffset(BuildMI(BB, X86::FST32m, 5),
                        X86::ESP, ArgOffset).addReg(ArgReg);