On Tue, Jul 26, 2022 at 12:55 AM Anurag Thakur <
[email protected]> wrote:

> I cleaned up the code a bit and used `ftbench` to get performance numbers
> for the `smooth` and `dense` rasterizers.
> There is much room for improvement in the ported code, as can be clearly
> seen in the performance graph.
> [image: Image]
>
> (ftbench output for both renderers at 20ppem attached)
> Phase-2 will be focused on profiling and improving the performance of the
> ported code.
>

Hi Anurag,

This looks encouraging, within reach after some more optimizations,
although the square curvature is probably unavoidable for the dense arrays.
I suggest you use 'ftbench -bc' to focus on rendering and avoid other
benchmarks. You can also add '-f 0x2' to apply FT_LOAD_NO_HINTING and '-i
0-100' or so, to use some portion of good glyphs and avoid simple
diacritics.

I also strongly recommend 'perf record ftbench -bc...' followed by 'perf
report' to find functions that take most of the time. Fedora distributes
perf in kernel-tools.

Good luck,
Alexei

Reply via email to