[Pixman] MIPS bilinear fast paths (src_8888_8_8888, src_8888_8_0565, src_0565_8_x888, src_0565_8_0565, add_8888_8_8888).
nlukic at mips.com
Sun Jul 1 08:17:46 PDT 2012
Sure, I'll make necessary changes and resubmit patches.
From: Siarhei Siamashka [mailto:siarhei.siamashka at gmail.com]
Sent: Saturday, June 30, 2012 8:28 PM
To: Lukic, Nemanja
Cc: pixman at lists.freedesktop.org
Subject: Re: [Pixman] MIPS bilinear fast paths (src_8888_8_8888, src_8888_8_0565, src_0565_8_x888, src_0565_8_0565, add_8888_8_8888).
On Mon, Jun 25, 2012 at 10:08 PM, Nemanja Lukic <nlukic at mips.com> wrote:
> Added optimizations for several bilinear fast paths:
> - src_8888_8_8888
> - src_8888_8_0565
> - src_0565_8_x888
> - src_0565_8_0565
> - add_8888_8_8888
> Benchmark results (using tweaked version of the lowlevel-blt-bench which does
> bilinear scaling using almost identity matrix) on Malta board (@1Ghz) are
> included in the log message.
> These routines still use old BILINEAR_INTERPOLATE_SINGLE_PIXEL macro which has
> potential to be more efficient (as pointed out in previous code review). This
> improvement will be part of the future MIPS bilinear fast-path commits.
> Any comments to this patch are welcome.
The MIPS DSP ASE performance patches are always welcome. Just the
timing is a bit bad this time. Could you please update your patchset
and make it compatible with
So that the assembly code respects the interpolation precision
settings defined as BILINEAR_INTERPOLATION_BITS constant and "make
check" tests pass when setting it to 7 and 8. Thanks.
More information about the Pixman