I am using Watcom C compiler (wcc) version 2.0. I compile this simple code with many optimizations enabled, but the resulting ASM seems very unoptimized to me.
int test(int x) {
return x ? 1 : 2;
}
Compiling with 8086 as target, fastest possible optimizations (-otexan).
wcc test.c -i="C:\Data\Projects\WATCOM/h" -otexan -d2 -bt=dos -fo=.obj -mc
Then disassembling with:
wdis test.obj -s > test.dasm
Resulting assembler looks like this:
...
return x ? 1 : 2;
02C7 83 7E FA 00 cmp word ptr -0x6[bp],0x0000
02CB 74 03 je L$39
02CD E9 02 00 jmp L$40
02D0 L$39:
02D0 EB 07 jmp L$41
02D2 L$40:
02D2 C7 46 FE 01 00 mov word ptr -0x2[bp],0x0001
02D7 EB 05 jmp L$42
02D9 L$41:
02D9 C7 46 FE 02 00 mov word ptr -0x2[bp],0x0002
02DE L$42:
02DE 8B 46 FE mov ax,word ptr -0x2[bp]
02E1 89 46 FC mov word ptr -0x4[bp],ax
02E4 8B 46 FC mov ax,word ptr -0x4[bp]
...
These jumps look heavily un-optimized to me. I would expect less unnecessary jumps and maybe putting result directly to AX without putting it under BP location there and back (last two lines).
cmp word ptr -0x6[bp],0x0000
jz L$39
mov ax,0x0001
jmp L$40
L$39:
mov ax,0x0002
L$40:
...
Am I missing something there? Is wcc ignoring my switches for some reason? Thanks.
Problem is the -d2 switch that generates detailed debug info. It probably inserts unnecessary lines to correspond to original C file's lines (to be able to put hardware breakpoints there, maybe?). I put -d1 instead and voila:
01A0 test_:
01A0 85 C0 test ax,ax
01A2 74 04 je L$15
01A4 B8 01 00 mov ax,0x0001
01A7 C3 ret
01A8 L$15:
01A8 B8 02 00 mov ax,0x0002
01AB C3 ret