Optimization opportunity

[mheyer32]

Consider: http://franke.ms/cex/z/oYqroh

The compiler will do a lot of extra work, while I was expecting to see just to swap the first argument and move.w the second argument into the lower word.

[bebbo]

not worth the effort. please use some asm instead.

[bebbo]

#define MKWORD(hi,lo) \
    asm ("swap %0" : "=r"(hi) : "r"(hi)); \
    asm ("move.w %2,%0": "=r"(hi) : "r"(hi), "rm"(lo))

[mheyer32]

Fair enough. I was just surprised that this seemingly simple operation turned into relatively many instructions.

[bebbo]

Fair enough. I was just surprised that this seemingly simple operation turned into relatively many instructions.

because the short's need an extension to int to apply shift and or.

[mheyer32]

For completeness sake, after lots of trial and error, I have no settled on:

static inline int makeDWORD(short hi, short lo)
{
    int res;
    __asm __volatile ("swap %0 \n"
                      "move.w %2,%0"
                      : "=&r"(res)
                      : "0"(hi), "g"(lo)
                      : "cc");
    return res;
}

Feel free to critique, I'm really not well versed in GCC inline assembly