difference in blitting uae vs. hw

yoki · 15 July 2009, 13:48

i coded some stuff in uae, works perfect... but when i run it on my a1200/030 50mhz all my bobs have random dots with color0 on random places making the background come through.. i run on a doublebuffered screen (same thing is happening without DB) and the screen does not take more than approx 1/10th of the rastertime.. any ideas?

edit:

seems this is a bug in the 030 cpu cache when fetching data from chipmem.. anyone knows about workarounds?

Toni Wilen · 15 July 2009, 14:01

UAE in 68020+ modes is not very accurate (and it isn't meant to be accurate either)

but.. AFAIK data cache should never be enabled in chip ram due to DMA bypassing it. (normally MMU is used to enable cacheing in fast ram but not in chip ram)

Leffmann · 15 July 2009, 14:09

Any screenshots?

yoki · 15 July 2009, 14:23

i dont even know how to make a screenshot on my 1200, but i can take a photo later today when my gf is back with the camera

this seems to be a bit blitzrelated since the standard procedure in blitz is to copy bobs from a bitmap allocated in chip to shape-structures (which might be allocated in fast).. it seems that its the chipmem->fastmem copy (which i do in the init of the program) is the one that is broken so that 0:s are copied instead of the actual data which applies that color-0 noise on my bobs

Asman · 15 July 2009, 15:20

Can you post your source ( upload to the zone ) ?

Photon · 15 July 2009, 15:21

If the 0-dots follow the bobs around the screen perfectly, it's the bob mask initialization, if the bobs seem to "move through a veil of dots" or is completely random it's in the frameloop.

15 July 2009, 13:48	#1
yoki Registered User Join Date: Jan 2009 Location: Göteborg / Sweden Posts: 25	difference in blitting uae vs. hw i coded some stuff in uae, works perfect... but when i run it on my a1200/030 50mhz all my bobs have random dots with color0 on random places making the background come through.. i run on a doublebuffered screen (same thing is happening without DB) and the screen does not take more than approx 1/10th of the rastertime.. any ideas? edit: seems this is a bug in the 030 cpu cache when fetching data from chipmem.. anyone knows about workarounds?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
I need quick transparent blitting	meynaf	Coders. General	15	13 November 2020 16:08
What's wrong with my blitting code?	MagerValp	Coders. System	6	22 April 2013 15:00
DIfference between WIN UAE and Amiga Forever	nineoc	New to Emulation or Amiga scene	19	01 April 2013 22:27
Blitting question	sandruzzo	Coders. General	30	06 April 2011 11:29
Bitmaps in FASTRAM and blitting - C++	NovaCoder	Coders. General	22	01 December 2010 23:40

15 July 2009, 14:01	#2
Toni Wilen WinUAE developer Join Date: Aug 2001 Location: Hämeenlinna/Finland Age: 49 Posts: 26,567	UAE in 68020+ modes is not very accurate (and it isn't meant to be accurate either) but.. AFAIK data cache should never be enabled in chip ram due to DMA bypassing it. (normally MMU is used to enable cacheing in fast ram but not in chip ram)

15 July 2009, 14:09	#3
Leffmann Join Date: Jul 2008 Location: Sweden Posts: 2,269	Any screenshots?

15 July 2009, 14:23	#4
yoki Registered User Join Date: Jan 2009 Location: Göteborg / Sweden Posts: 25	i dont even know how to make a screenshot on my 1200, but i can take a photo later today when my gf is back with the camera this seems to be a bit blitzrelated since the standard procedure in blitz is to copy bobs from a bitmap allocated in chip to shape-structures (which might be allocated in fast).. it seems that its the chipmem->fastmem copy (which i do in the init of the program) is the one that is broken so that 0:s are copied instead of the actual data which applies that color-0 noise on my bobs

15 July 2009, 15:20	#5
Asman 68k Join Date: Sep 2005 Location: Somewhere Posts: 829	Can you post your source ( upload to the zone ) ?

15 July 2009, 15:21	#6
Photon Moderator Join Date: Nov 2004 Location: Eksjö / Sweden Posts: 5,652	If the 0-dots follow the bobs around the screen perfectly, it's the bob mask initialization, if the bobs seem to "move through a veil of dots" or is completely random it's in the frameloop.

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)