Blitter nasty or not?

JackAsser · 18 February 2010, 06:06

Hi!

I'm sure this is all in the ref guide but I just wanted to hear your experiences with the blitter nasty bit. What gives most bang for the buck or is it doesn't matter in the end?

What I'm thinking is that using the blitter nasty bit in dmacon is nice - the blitter runs faster and I don't have to fiddle with waiting for the blitter to finish. Otoh, not using it gives me the opportunity to do CPU stuff during the blit.

However, if the blit AND the CPU stuff are both equally time critical, couldn't I just as well use the blitter nasty bit. Both must be finished ASAP anyway to produce the frame.

Regards, JackAsser

Leffmann · 18 February 2010, 11:57

Turn it off if you have some CPU code that can run in parallel, and turn it on while waiting for the blitter to finish.

If it's off then you allow the CPU access to chipmem regularly, providing means for it to run in parallel with a blitter operation. Lengthy instructions like multiplications and divisions that might need a single chipmem access to read the instruction (if it's stored in chipmem) can spend the next 50 cycles doing internal calculations while the blitter carries on working in chipmem. Theoretically you can overlap CPU and blitter operations and reduce the time some piece of CPU code needs to execute, down to the time the CPU needs to do the memory accesses alone.

Turning blitter nasty on only while waiting for the blitter to finish prevents the CPU from stealing bus cycles polling DMACONR. IIRC this could be up to 1/4 cycles and potentially a blitter operation could take 33% longer to finish.

Look into using 3 screen buffers as well if you're drawing using the CPU only. Displaying one buffer while letting the blitter clear the next and the CPU draw in the third will usually earn you a few rasterlines.

Toni Wilen · 18 February 2010, 12:03

Always wait for blitter. Only "acceptable" reason if you are doing A500 68000-only program that needs all available cycles and more and few cycles gained from removing blitter waits makes your effect better and faster

Quote:

Originally Posted by JackAsser

the blitter runs faster and I don't have to fiddle with waiting for the blitter to finish.

It isn't that simple. Check that blitter cycle diagram again. Most have idle cycles. And never ever do that if you want fast ram or 68020+ compatibility.

(EDIT: why do I always reply when someone has replied few moments ago?)

StingRay · 18 February 2010, 12:10

Quote:

Originally Posted by Toni Wilen

(EDIT: why do I always reply when someone has replied few moments ago?)

Because you're too slow, just like me. :P I was going to write exactly the same regarding "not necessary to wait for blitter to finish if blitter nasty is set".

JackAsser · 18 February 2010, 12:36

Noted, thanks again guys.

/JackAsser

Photon · 28 March 2010, 22:45

In response to the thread title: 'Yes. Yes, the blitter is very nasty indeed. But Paula is nastier, at least on weekday evenings.'

18 February 2010, 06:06	#1
JackAsser Registered User Join Date: Jul 2006 Location: Lund / Sweden Age: 45 Posts: 171	Blitter nasty or not? Hi! I'm sure this is all in the ref guide but I just wanted to hear your experiences with the blitter nasty bit. What gives most bang for the buck or is it doesn't matter in the end? What I'm thinking is that using the blitter nasty bit in dmacon is nice - the blitter runs faster and I don't have to fiddle with waiting for the blitter to finish. Otoh, not using it gives me the opportunity to do CPU stuff during the blit. However, if the blit AND the CPU stuff are both equally time critical, couldn't I just as well use the blitter nasty bit. Both must be finished ASAP anyway to produce the frame. Regards, JackAsser

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
CD32 nasty drive noise	Amiga1992	support.Hardware	11	31 August 2012 00:50
One "hole" in each scan line to turn off blitter nasty?	mc6809e	Coders. Asm / Hardware	1	03 July 2012 12:12
Found nasty bug in 68030 CPU!	Oliver_A	Coders. General	11	13 November 2010 15:39
Nasty sound problem in B.C.Kid	Hobbe	support.WinUAE	13	14 June 2009 22:22
Smarty And The Nasty Gluttons	mai	HOL contributions	1	03 February 2009 00:52

18 February 2010, 11:57	#2
Leffmann Join Date: Jul 2008 Location: Sweden Posts: 2,269	Turn it off if you have some CPU code that can run in parallel, and turn it on while waiting for the blitter to finish. If it's off then you allow the CPU access to chipmem regularly, providing means for it to run in parallel with a blitter operation. Lengthy instructions like multiplications and divisions that might need a single chipmem access to read the instruction (if it's stored in chipmem) can spend the next 50 cycles doing internal calculations while the blitter carries on working in chipmem. Theoretically you can overlap CPU and blitter operations and reduce the time some piece of CPU code needs to execute, down to the time the CPU needs to do the memory accesses alone. Turning blitter nasty on only while waiting for the blitter to finish prevents the CPU from stealing bus cycles polling DMACONR. IIRC this could be up to 1/4 cycles and potentially a blitter operation could take 33% longer to finish. Look into using 3 screen buffers as well if you're drawing using the CPU only. Displaying one buffer while letting the blitter clear the next and the CPU draw in the third will usually earn you a few rasterlines.

18 February 2010, 12:36	#5
JackAsser Registered User Join Date: Jul 2006 Location: Lund / Sweden Age: 45 Posts: 171	Noted, thanks again guys. /JackAsser

28 March 2010, 22:45	#6
Photon Moderator Join Date: Nov 2004 Location: Eksjö / Sweden Posts: 5,657	In response to the thread title: 'Yes. Yes, the blitter is very nasty indeed. But Paula is nastier, at least on weekday evenings.'

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