[Pixman] [PATCH 2/3] Moving helper functions in pixman-sse2.c to pixman-sse2.h.
chris at chris-wilson.co.uk
Tue Jan 22 10:06:56 UTC 2019
Quoting Matt Turner (2019-01-22 00:49:39)
> On Thu, Jan 17, 2019 at 12:27 AM Chris Wilson <chris at chris-wilson.co.uk> wrote:
> > Quoting Raghuveer Devulapalli (2019-01-17 00:59:59)
> > > From: raghuveer devulapalli <raghuveer.devulapalli at intel.com>
> > >
> > > These helper function will be reused in pixman-avx2.c implementations in
> > > the future.
> > Are we ever going to run into a naming conflict in the future? Is it
> > worth prefixing all the inlines with sse2_? Probably makes sense so that
> > we can see the instruction set used when mixing later.
> > -Chris
> The SSE2 intrinsics will actually be compiled into VEX-prefixed (AVX)
> instructions operating on xmm registers when -mavx2 is used.
> I can't think of a reason the lack of a prefix would cause any
> confusion for the ones that already have "128" in the name. For
> unpack_565_to_8888, etc, maybe it would be best to add a _128 suffix.
> We have functions (e.g., pack_565_2x128_128) that look like that
Ok, that seems like a reasonable means to distinguish these helpers from
possible future ones; as width is only likely to go up, and an
incompatible ia will then not include the sse2/avx2 inlines.
More information about the Pixman