DMA memory to memory copy

[BlueAchenar]

How do I copy x bytes from address A to address B using DMA transfers or something similar (custom chips)?

[alexh]

Remember that with a DMA copy you'll be limited to CHIP RAM and that a DMA copy will be slower than a CPU fast RAM copy on accelerated Amiga's.

Google for BLITTER

http://pub.elowar.com/AmigaDev/Hardw.../node0121.html

[BlueAchenar]

Only chip memory... I remembered that right after I wrote the post. And that's a problem...

[musashi5150]

I guess it'll be a tight movem loop for you then

[Photon]

Remember to check for non-longword aligned start/end addresses for source and destination, and to copy from end to start instead of from start to end, depending on how the areas overlap. If they don't overlap all is fine, and if source and destination start address is the same, you just skip the copy.

[BlueAchenar]

Originally Posted by Photon

Remember to check for non-longword aligned start/end addresses for source and destination, and to copy from end to start instead of from start to end, depending on how the areas overlap. If they don't overlap all is fine, and if source and destination start address is the same, you just skip the copy.

I'll remember that when I need it (I opted not to try anything DMA related right now). Thanks

[Photon]

I has nothing to do with DMA and everything to do with copying areas without overwriting the original.

[Ray Norrish]

I remember using the blitter to copy bits of a soundtracker module around for the mix-e-loader on Sonic Attack. If you can't find the source in PD, or want to see, let me know.

[BlueAchenar]

Originally Posted by Ray Norrish

I remember using the blitter to copy bits of a soundtracker module around for the mix-e-loader on Sonic Attack. If you can't find the source in PD, or want to see, let me know.

Thanks. I'd love to know how to use the blitter well.

[Ray Norrish]

This was a rather unusual way to use the blitter. Usually, it would be used to copy screen memory around.
Fortunately, I noted that the ST note track is 256 bytes (or something) and was blitter freindly.

A simple blitter routine can used to copy memory from a to b for any number of purposes, but restricted to certain block sizes (binary restrictions)

[BlueAchenar]

Originally Posted by Ray Norrish

This was a rather unusual way to use the blitter. Usually, it would be used to copy screen memory around.
Fortunately, I noted that the ST note track is 256 bytes (or something) and was blitter freindly.

A simple blitter routine can used to copy memory from a to b for any number of purposes, but restricted to certain block sizes (binary restrictions)

I saw somewhere that the blitter could be used to encode/decode MFM disk tracks. Is this really possible?

[Ray Norrish]

Originally Posted by BlueAchenar

I saw somewhere that the blitter could be used to encode/decode MFM disk tracks. Is this really possible?

of course. Remember that the blitter worked on odd cycles, so it could double throughput. It was a powerful chip in its own right

[BlueAchenar]

Originally Posted by Ray Norrish

of course. Remember that the blitter worked on odd cycles, so it could double throughput. It was a powerful chip in its own right

That would be a very useful trick for me to learn!

[Ray Norrish]

I posted the code a long time ago. No probs to post it again if you don't have.
Check out the demo first

[Kalms]

Originally Posted by BlueAchenar

I saw somewhere that the blitter could be used to encode/decode MFM disk tracks. Is this really possible?

Certainly. The logical operation that you perform to decode a sector is, in C:

char rawSectorData[512*2];
char decodedSectorData[512];

// fill rawSectorData with contents first

for (int i = 0; i < 512; i++)
   decodedSectorData[i] = ((rawSectorData[i] & 0x55) << 1)
    | (rawSectorData[i + 512] & 0x55);

In order to perform that operation using the blitter:
- rewrite it using word operations
- the blitter can only shift left when in descending mode, so make loop go backwards
- the blitter shifts before masking together channels, so move shift before masking

short rawSectorData[256*2];
short decodedSectorData[256];
// fill rawSectorData with contents first

for (int i = 0; i < 256; i++)
   decodedSectorData[255 - i] = ((rawSectorData[255 - i] << 1) & ~0x5555)
    | (rawSectorData[511 - i] & 0x5555);

So in order to decode with a blit operation, you do something like:
BLTAPT = rawSectorData + 255 words;
BLTBPT = rawSectorData + 511 words;
BLTDPT = decodedSectorData + 255 words;
BPLAMOD..BPLDMOD = 0
BLTCDAT = $5555
minterm = Ac + BC = ABc + Abc + ABC + aBC = $40 + $10 + $80 + $08 = $D8
Channel A-shift = 1
Channel B-shift = 0
Enabled DMA channels = ABD
Descending = 1
=> BLTCON0 = $1DD8
=> BLTCON1 = $0002
BLTAFWM,BLTALWM = $FFFF
BLTSIZE = 256 * 64 + 1