Re: [AD] About the speed of flat polygon primitives

["Eric Botcazou" <ebotcazou@xxxxxxxxxx>]

> DGA2, 800x600 16 bpp, hlines per second:
>
> Accel:   241630 (unclipped), 607677 (clipped)
> Unaccel: 573807 (unclipped), 768050 (clipped)

Hmm... looks like it's worse than expected as even *individual* hlines are
slower in hardware drawing mode than in software drawing mode.

Peter, could you do a complete screen profiling with the test program and
use the PERL script misc/cmplog.pl to generate a diff file ? (misc/cmplog.pl
doesn't work for me with PERL 5.6.1 so I've attached a little patch that
should fix the problems).

Here are my results under Windows:

Comparing test profile logs 800x600x15_accel.log and 800x600x15_soft.log

DRAW_MODE_SOLID results:

 putpixel()                       = 100%
 hline()                          = 82%
 vline()                          = 75%
 line()                           = 100%
 rectfill()                       = 17%
 circle()                         = 99%
 circlefill()                     = 736%
 ellipse()                        = 99%
 ellipsefill()                    = 654%
 arc()                            = 100%
 triangle()                       = 746%


hline() is slower in software drawing mode on my box. The filled primitives
(circlefill, ellipsefill, triangle) that use multiple calls to hline() are
much faster, contrary of course to rectfill() that uses one single call to
the hardware-accelerated primitive in hardware drawing mode.

> I think that's enough to justify your proposal.

Yes, I'm going to take a look at the implementation details.

--
Eric Botcazou
ebotcazou@xxxxxxxxxx

Attachment: cmplog.diff
Description: Binary data