On Sat, Jul 11, 2015 at 11:08 AM, Chris Wilson <ch...@chris-wilson.co.uk> wrote:
> On Fri, Jul 10, 2015 at 11:44:58AM -0700, Matt Turner wrote:
>> So that everything writing to the batch between BEGIN_BATCH() and
>> ADVANCE_BATCH() goes through OUT_BATCH.
>
> Reviewed-by: Chris Wilson <ch...@chris-wilson.co.uk>
>
>> +#define OUT_RELOC64(buf, read_domains, write_domain, delta) do { \
>> + uint64_t reloc64 = intel_batchbuffer_reloc64(brw, buf, read_domains, \
>> + write_domain, delta); \
>> + OUT_BATCH(reloc64); \
>> + OUT_BATCH(reloc64 >> 32); \
>> } while (0)
>
> Is the compiler smart enough to convert this to movq?
It is not, but I'm not totally sure that's better -- an unaligned movq
might cross a cache boundary which is apparently pretty expensive.
> The pointer emission variant at least makes it trivial to use
> *((uint64_t *)map)++ = reloc64
> if so desired.
I'll do some performance tests on top of my v3 series.
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
