> No test for popcnt, seeing as there's a bug in m3?
Originally popcnt was not in the task list, it was added later.
> You can't split these two asm, lest the ltgr and sel not be adjacent, and the
> flags not
> having the correct value when we arrive at the sel.
This was tested, both gcc and clang assemble multiple 'asm' statements
into a single block as long as there are no C statements between.
I'm happy to change it.
On Wed, Feb 23, 2022 at 2:45 PM Richard Henderson
<richard.hender...@linaro.org> wrote:
>
> On 2/17/22 13:17, David Miller wrote:
> > +#define F_PRO asm ( \
> > + "lg %%r2, %[a]\n" \
> > + "lg %%r3, %[b]\n" \
> > + "lg %%r0, %[c]\n" \
> > + "ltgr %%r0, %%r0" \
> > + : : [a] "m" (a), \
> > + [b] "m" (b), \
> > + [c] "m" (c) \
> > + : "r0", "r2", "r3", "r4")
> > +
> > +
> > +
> > +#define Fi3(S, ASM) uint64_t S(uint64_t a, uint64_t b, uint64_t c) \
> > +{ uint64_t res = 0; F_PRO ; ASM ; return res; }
> > +
> > +
> > +Fi3 (_selre, asm("selre %%r0, %%r3, %%r2\n" F_EPI))
> > +Fi3 (_selgrz, asm("selgrz %%r0, %%r3, %%r2\n" F_EPI))
> > +Fi3 (_selfhrnz, asm("selfhrnz %%r0, %%r3, %%r2\n" F_EPI))
>
> You can't split these two asm, lest the ltgr and sel not be adjacent, and the
> flags not
> having the correct value when we arrive at the sel.
>
> No test for popcnt, seeing as there's a bug in m3?
>
>
> r~
