Blitter filling speed, how much?

[sandruzzo]

Hy all

I've found that blitter can move 4 mb per second and draw 1 milion pixels per second. What about filling? How many pixels it can draw per second?

[roondar]

Hi Sandruzzo, I've read your Rygar thread with interest so I thought I'd add what I know.

The blitter clear/copies about ~3.5MB/second (but only if the CPU is blocked and there is no bitplane DMA). OCS Bitplane DMA loses you ~0.24MB/sec for a 320x256 PAL bitplane (or ~0,97MB/sec less for a 4 bitplane screen).

It draws line pixels at about 1 million per second, again lowered by Bitplane DMA.

Filling is 'free' in that it is done during the copy and costs no extra cycles.

Edit: accidently doubled DMA contention figures

[Toni Wilen]

Filling unfortunately is not always free. D=A and 3 other less commonly used channel combinations add 1 extra cycle to blit's cycle diagram. (for example D=A copy with fill becomes 3 cycle blit).

[sandruzzo]

so filling is going near to 4 mb per second

[roondar]

Originally Posted by Toni Wilen

Filling unfortunately is not always free. D=A and 3 other less commonly used channel combinations add 1 extra cycle to blit's cycle diagram. (for example D=A copy with fill becomes 3 cycle blit).

Ok, I did not know that, the HRM seems to suggest it is free (or has hidden the information in a place I didn't find).

Thanks for the information, that might save me some headaches if I start using fills on bigger objects!

Originally Posted by sandruzzo

so filling is going near to 4 mb per second

If I understand Toni correctly, it's going to be more like ~2,3MB/sec because the blitter is slowed down while filling in the most commonly used mode (D=A).

[Toni Wilen]

Yeah, max real world rate is about 2.33Mbytes/s (3.5M DMA slots in 1 second ((227-4)*312.5*50), 3.5M/3*(16/8)=2.33Mbytes/s)

Ok, I did not know that, the HRM seems to suggest it is free (or has hidden the information in a place I didn't find).

It was only documented in some early HRM revision, it had blitter cycle diagram that included fill mode special cases. I didn't even know about that HRM until few years ago, I originally found out it by experimenting (I am sure most Amiga coders only had later "blue" or "black" HRM revision), then later it was 100% confirmed using logic analyzer.

Note that all blitter idle cycles are special, it needs to be free cycle (not used by any higher Agnus DMA channels) but blitter does not "use" it, it is also free for the CPU!

[sandruzzo]

not so bad! It's fast!!

[Vot]

Now thats some good advice some info he can use!