[gfx]2D Performance of menlow Gfx dirver far fell behind that of Q1U

[Jing Wang]

Donald did a basic test on 2D driver on Crown Beach to collect the information of our 2D HW acceleration driver
I ran a benchmark utility- x11perf on three environments.

Command line: x11perf -repeat 1 -time 1 all

a) Crown Beach with HW acceleration driver-Poulsbo enabled + Alpha release image
The log is X11perf_psb.log.

b) Crown Beach with software driver- vesa + Alpha release image
The log is X11perf_vesa.log.

c) Q1 Ultra + Alpha release image
The log is X11perf_q1.log.

Take some test items from the log to make a table:

                 10-pixel ellipse 10-pixel partial ellipse Scroll 10x10 pixels
a) Poulsbo 346000.0/sec 256000.0/sec 27600.0/sec
b) vesa 172000.0/sec 117000.0/sec 30000.0/sec
c) Q1 414000.0/sec 467000.0/sec 306000.0/sec

Summary:
For pixel operation, Poulsbo increases about 200% to software driver solution.
For Scroll operation, no significant gain to software driver solution

Rob: What this is showing is that Menlow is currently worse than McCaslin w.r.t. to Gfx performance.

[rob rhoads (rob-rhoads)]

This is NOT a kernel bug. This is a bug in the PSB Gfx drivers which is both kernel and X Windows. We should be tracking all Gfx bugs in it's own category.

I'm assigning this bug to Jacob as he own's integrating the Gfx driver, but in reality this should really be assigned to Calvin in UMG.

[Chang Liu (chang-l-liu)]

on UME beta4 crownbeach, xmark for psb is 13.66, xmark for vesa is 14.43

[Chang Liu (chang-l-liu)]

sorry, i made a mistake, it is in inuka beta5 crownbeach