From: Peter D'Hoye <peter.dhoye_at_telenet.be>
Date: Mon, 12 Dec 2005 19:13:22 +0100

Hi all,

my first real mail on this list after I discovered that the mailing
list dropped my mails without complaining
(subscriber address != return address)

Sorry for putting them in one mail... Here we go:

============================================
patch 1378276 (H300 LCD: inlined some calls)
============================================

Inlined some calls and got a minor speed increase.

I also wrote a (very) small plugin to do some basic FPS
measurement. It basically does an lcd_xxx call for two
seconds and then calculates the amount of calls per second.

I wrote two tests:
- lcd_update()
- lcd_update_rect() for a centered rect 1/4 the area of the lcd

Some figures:

* inlined with Kaggen's fix:
        lcd_update(): 41,5 fps
        lcd_update_rect(): 162,0 fps

* inlined without Kaggen's fix:
        lcd_update(): 41,5 fps
        lcd_update_rect(): 160,5 fps

* not inlined with Kaggen's fix:
        lcd_update(): 41,5 fps
        lcd_update_rect(): 161,0 fps

* not inlined without Kaggen's fix:
        lcd_update(): 41,5 fps
        lcd_update_rect(): 158,0 fps

Maybe I should add an FPS counter to the box plugin?

=========================================================
patch 1377879 (H3xx lcd_update_rect() - already accepted)
=========================================================
 
Checked and tested code submitted here by Kaggen:
http://www.rockbox.org/mail/archive/rockbox-dev-archive-2005-11/0178.shtml

the H3xx lcd_update_rect() now uses windowed GRAM access.
The old code is still present...

If there is any speed gain, it is unnoticable.



==========================================
patch 1375787 (w32 sim cpu load and sound)
==========================================

this is a small but effective patch that makes the W32
simulator stop putting load on the CPU while having audio playback.

In the yield() implementation I've added a mechanism that
alternates between Sleep(0) and Sleep(1). This is because
Sleep(1) makes the app run too slow (no audio), and Sleep(0)
hogs the CPU (but gives sound).

I wrote it so that it calls Sleep(1) every 256 calls to
yield. All other times it does Sleep(0).
 
I've also added a Sleep() in sleep().

And can some core-dev remove patch _1373083_? thanks!
Received on 2005-12-12