On Fri, 2012-08-17 at 14:41 +0200, David Henningsson wrote:
> On 08/17/2012 02:39 PM, Tanu Kaskinen wrote:
> > On Fri, 2012-08-17 at 13:36 +0200, David Henningsson wrote:
> >> On 08/17/2012 01:30 PM, Tanu Kaskinen wrote:
> >>> The buffers in question should be properly aligned for any
> >>> integer size, so the warnings are false positives.
> >>
> >> Oh, so there were that many of them. Well, how about
> >>
> >> static inline void* pa_memblock_acquire_index(memblock, index)
> >> {
> >> return (void *) ((uint8_t *) pa_memblock_acquire(memblock) + index);
> >> }
> >>
> >> ?
> >
> > A very good idea. What would you think about having
> > pa_memchunk_acquire() instead? Almost all, if not all, instances where
> > this pattern is used, the parameters to pa_memblock_acquire_index()
> > would be "chunk->memblock, chunk->index", so it would be simpler to just
> > pass a chunk pointer.
>
> Sounds good to me, assuming you are correct in that all memblocks
> acquired are memchunks (which I haven't verified).
I did not say that all pa_memblock_acquire() calls would be for
memchunks, I said that (almost?) all pa_memblock_acquire_index() calls
would be memchunks. Is that a problem? The purpose wouldn't be to
replace pa_memblock_acquire(), the purpose would be to replace the
pa_memblock_acquire_index() helper.
--
Tanu
