[LLVMdev] Help me improve two-address code

[Evan Cheng]

On Apr 16, 2009, at 3:17 PM, Greg McGary wrote:

> I have my new port limping enough to compile a very basic function:
>
> int
> foo (int a, int b, int c, int d)
> {
>  return a + b - c + d;
> }
>
> clang-cc -O2 yields:
>
> define i32 @foo(i32 %a, i32 %b, i32 %c, i32 %d) nounwind readnone {
> entry:
>    %add = add i32 %b, %a        ; <i32> [#uses=1]
>    %sub = sub i32 %add, %c        ; <i32> [#uses=1]
>    %add4 = add i32 %sub, %d        ; <i32> [#uses=1]
>    ret i32 %add4
> }
>
> which lowers to this assembler code (note: args arrive in r1..r12, and
> results are returned in r1..r3.):
>
> foo:
>    add r2,r1      ### add r1,r2 is better
>    sub r2,r3
>    mov r1,r2      ### unnecessary!!
>    add r1,r4
>    jmp [r30]
>    .end foo
>
> The mov insn would be unnecessary if the operand order for the first  
> add
> were reversed.  For this function, GCC does the right thing.
>
> Is there some optimizer knob I'm not turning properly?  In more  
> complex
> cases, GCC does poorly with two-address operand choices and so bloats
> the code with unnecessary register moves.  I have high hopes LLVM  
> can do
> better, so this result for a simple case is bothersome.

Are you marking add as commutable? Are you making mov as a copy  
instruction?

Evan

>
> G
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev