I am finding that blitting with a mask varies from fast to horribly slow on an Efika with a Radeon, and I'd appreciate if some people could please run the following test to confirm it (on different hardware) :
http://cshandley.co.uk/temp/MOS_MaskSpeedTest.lha

Just reboot, then unpack it somewhere, then run MOS_MaskSpeedTest, and do nothing until the new window disappears (may take seconds or minutes). Then please post your results. Here are mine:

Quote:

The first two results are about as expected; BltMaskBitMapRastPort() is around 10 times slower than BltBitMapRastPort(). This also happens on OS4, although on emulated VESA AROS it is no slower.

But the remaining results are strange & pretty bad; when the mask contains some transparent pixels, it can take up to 20 times *longer* to blit, meaning up to 200 times worse than without a mask! Such a masked blit is taking up to 0.3 seconds, which is clearly ridiculous. (OS4 & AROS do not slow down at all for transparent pixels.)

Interestingly the speed degredation is proportional to the number of transparent pixels, suggesting that the "blit" is CPU dependant. If other peoples' results confirm this, then either MOS has a bug, or I'm doing something wrong. If anyone can suggest my mistake, that would be great.

The source code is included, but you can also read it here:
http://cshandley.co.uk/temp/MOS_MaskSpeedTest.cpp


[ Edited by ChrisH 24.10.2011 - 22:23 ]

> Just reboot, then unpack it somewhere, then run MOS_MaskSpeedTest, and
> do nothing until the new window disappears (may take seconds or minutes).
> Then please post your results. Here are mine: [...]

Mine (Mac mini, MorphOS 2.7):

Code:

> Interestingly the speed degredation is proportional to the number of
> transparent pixels, suggesting that the "blit" is CPU dependant.

CPU is clearly running at maximum during execution of your speed test.

@Andreas_Wolf
Thanks. Not exactly the same as me, but it does show that masked blits with transparent pixels are up to 100 times slower than without such pixels (or mask), and slow-down is also proportional to the number of transparent pixels.

Test using a 640 by 480 window, bitmap & mask.

BltBitMapRastPort() took 565 uS

With a clear mask:
BltMaskBitMapRastPort() took 2868 uS (5 times slower than no mask)

With a set mask:
BltMaskBitMapRastPort() took 153361 uS (271 times slower than no mask)

With an alternating-line mask:
BltMaskBitMapRastPort() took 79492 uS (141 times slower than no mask)

With a chequered mask:
BltMaskBitMapRastPort() took 87351 uS (154 times slower than no mask)

With an alternating-thick-line mask:
BltMaskBitMapRastPort() took 79446 uS (140 times slower than no mask)


PowerMac MDD G4 @ 867MHz

@ChrisH

Change AllocBitMap()'s friend bitmap to be NULL and your blits wil be around 100 times faster.
Reading from video memory using CPU is ultra slow.


edit: on a second thought, maybe just clearing the BFM_DISPLAYABLE flag could be a better idea.

[ Edited by Henes 22.10.2011 - 18:22 ]

Clearing BMF_DISPLAYBLE does it.

Here are results from my Mac mini with original flags:

Quote:

And results with BMF_DISPLAYABLE cleared:

Quote:

itix,
Quote:
Thanks, that does indeed fix it . This should hopefully mean that 'my' Shadow Of The Beast demo will now run very fast...

EDIT: Yup, now running very smoothly :)))

[ Edited by ChrisH 24.10.2011 - 22:24 ]