Copper Interrupts and Blit Performance

[mcgeezer]

On Rygar I'm trying to get away with not double buffering the background blits that I'm doing (as I need the ram), currently I am using double buffering because each blit of a bitplane might occur during bitplane dma and I would get tearing.

If I were to use a copper interrupt to wait for a vertical position below the display area, then turn off bitplane DMA, turn on Blitter Nasty and then proceed to do by three straight A-D Blit copy operations.... would this yield me the maximum performance for rendering the background - and I would't have to double buffer the backgrounds?

I think I already know the answer to this but before I change the code it would be really great to get some insight or tips on further performance improvement.

Geezer

[DanScott]

You'll definitely get your best blitter perfomance during time where the bitplane DMA is off

Do you have enough time to get these 3 blits done after the display area ?

Is it possible also to blit in "chunks", from the top down... for example, the first 32 pixels of bitplane 1, then the first 32 of bitplane 2, then bitplane 3.... then the next 32 pixels of bitplane 1 etc... perhaps you could be blitting ahead of the beam with this method

[mcgeezer]

Originally Posted by DanScott

You'll definitely get your best blitter perfomance during time where the bitplane DMA is off

Yeah I thought that was the case.

Originally Posted by DanScott

Do you have enough time to get these 3 blits done after the display area ?

At the moment I'm unsure... so I take it you mean can I get all three blits done between say line 224 (after display) and 313 start of Vertical Blank?

Originally Posted by DanScott

Is it possible also to blit in "chunks", from the top down... for example, the first 32 pixels of bitplane 1, then the first 32 of bitplane 2, then bitplane 3.... then the next 32 pixels of bitplane 1 etc... perhaps you could be blitting ahead of the beam with this method

I never thought of doing that, but it would be fairly easy for me to implement and I can see why I would do it too!

[mcgeezer]

The other thought that has occurred to me is that I'm wondering if I will gain performance by setting FMODE to x4 instead of the normal x2 in the display area.

Given no bitplane dma is off I take it this would be pointless?

[ross]

Originally Posted by mcgeezer

The other thought that has occurred to me is that I'm wondering if I will gain performance by setting FMODE to x4 instead of the normal x2 in the display area.

Yep, you would get more performance.
But you can't in your actual engine because sprites loss (due to bigger BPLCON1 shift region and early DDFSTRT).
There is somewhere a thread where i've calculated FMODE and related screen position/sprite loss about this.

If you speak about static region, no problem to setup a 4x mode.

[hooverphonique]

Originally Posted by ross

Yep, you would get more performance.

Really? I'm pretty sure it makes no difference while bitplane dma is idle..

[hooverphonique]

Originally Posted by mcgeezer

If I were to use a copper interrupt to wait for a vertical position below the display area, then turn off bitplane DMA...

If you are below the display area, no bitplane dma will be taking place anyway...

[ross]

Originally Posted by hooverphonique

Really? I'm pretty sure it makes no difference while bitplane dma is idle..

Not read all the previous messages, so can be i've miss something.. but in this there is a reference to display area.

Originally Posted by mcgeezer

The other thought that has occurred to me is that I'm wondering if I will gain performance by setting FMODE to x4 instead of the normal x2 in the display area.

[hooverphonique]

Given no bitplane dma is off I take it this would be pointless?

We will have to wait for geezer to clarify, if he meant while bitplane is active, or not :-)

[ross]

Originally Posted by hooverphonique

We will have to wait for geezer to clarify, if he meant while bitplane is active, or not :-)

[mcgeezer]

I must apologise - I meant outside of the display area.



So bitplane dma is switched off and then I switch to x4 fetch mode from x2.

[roondar]

Originally Posted by mcgeezer

I must apologise - I meant outside of the display area.



So bitplane dma is switched off and then I switch to x4 fetch mode from x2.

Fetch mode only effects bitplane fetching so this would not make a difference in Blitter speed.

[mcgeezer]

Originally Posted by roondar

Fetch mode only effects bitplane fetching so this would not make a difference in Blitter speed.

Good stuff, so the blitter would just eat all the cycles it could undisturbed.

I’ll try it this weekend and provide some measurements.

[roondar]

Ok. Reading the thread again I did get a bit confused, so just to be 100% clear: fetch mode only affects bitplane fetching, so Blitter speed won't change for blits executing during rasterlines below (or above) the ones that fetch bitplanes.

For blits during the rasterlines that bitplanes are fetched, the Blitter will get more slots when running a higher fetch mode.

Sorry for the doubled up info - I got a bit confused on where (or rather when during the frame) you want to blit.

[mc6809e]

If you time it correctly, you should be able to start blitting a little before display fetch ends. Just make sure display fetch wins the race for access to the playfield.

And if you use interleaved bitplanes, you can begin even earlier.