Things the Amiga didn't get right from Day 1

[grond]

Yes, you are of course right about the details. The point remains that something workable with an 8 bit chunky mode could have been achieved at almost no cost. It would have cost a little more effort to arrive at a well-rounded chunky blitter (your points). Much the same a 32 bit blitter shouldn't have been an impossibility to add with AGA. Such little things would have made AGA look quite competitive at the low end for a couple of years.

[Korban]

If many of the responses here are an indication Id suggest what the Amiga lacked most was a userbase grounded in reality.

Lots of suggestions that people could have done much better than people like Jay Miner.
Knowing a buzz word or 2 doesn't make a person an engineer.

[Thomas Richter]

Originally Posted by Korban

Lots of suggestions that people could have done much better than people like Jay Miner. Knowing a buzz word or 2 doesn't make a person an engineer.

That's besides the point. The original chipset is what it is, no problem. However, there are certainly some minor fixes that would have made a difference, and I have no doubts engineering knew actually. It was rather a matter of lack of investment that blocked many ideas.


Look at ECS or FAT Agnus - the only thing CBM managed to get done is to add one (and later two) bits to the DMA channels of Agnus. Compare that to the development that happened before - creating an entire set of custom chips from scratch in the same timespan.

[Korban]

Not only is it not besides the point, its not possible form it to be more relevant to the point. The exact complete and utter opposite to not the point.


The engineers and Commodore knew what tweaks and improvements could be made, but they didn't, for various reasons. Financial, time, whatever.
It'd take an absolute moron to not know this.

And yet theres many armchair experts offering "revelations" (chortle), thinking themselves enlightened.

Ergo, couldn't be more on the point.
Its baffling that someone with the ability to tie their own laces doesn't understand this.
Velcro shoes?

[grond]

Originally Posted by Korban

Lots of suggestions that people could have done much better than people like Jay Miner.
Knowing a buzz word or 2 doesn't make a person an engineer.

I am not sure who you are referring to but Thomas certainly knows the Amiga inside out. I hold a doctorate degree in microelectronics. And Jay Miner wasn't even involved in the AGA chipset (and probably not even ECS). I am sure he would have done better even if tied by company resources.

[axilmar]

It's a fun exercise in API, software & hardware design, so even if things proposed were not really feasible, let us have this discussion. We all can learn a few things from it.

[Thomas Richter]

To make things a bit more technical:

The problem of creating a "chunky display" is not at all new to Amiga. The same problem appeared on the PC side as well because VGA is inherently a planar design, with four bitplanes. What happens there to create chunky bitplanes is to instruct the sequencer to read bits alternating from the bitplanes to form a chunky pixel (the "chain-4" mode), and combine that with a "memory crossbar" (aka aperture) to make the bits that are spread over multiple bitplanes accessible as single byte to the CPU.

So that wasn't at all a new problem, and it was solved on the PC side. Something similar could have been done on the Amiga as well, probably avoiding the crossbar on the CPU/memory side but rather interleave bitmap access and permute the bits before they enter Denise.

Concerning the blitter, the problem is here that for some combinations of blitter width and source and destination position, the source is one word wider or shorter than the destination (all depending on alignment), and in such a case, one would either need to fetch one additional word from one source, or have a LWM or FWM that are 32 bits in size.

The way how the Os solves that is that it pre-calculates a mask of the full line width, replacing the FWM and LWMs as blitter source A. Unfortunately, that means that only B is left as source, and C and D as destination, and as such, "cookie cut" blits such as BltMaskBitMapRastPort() have to be simulated by multiple blits as there is no blitter source left to provide the mask.

Again, this problem might have been foreseen, but at least it became eminent at the time the Os/graphics.library was written, but the hardware was never updated to handle it in a better way. PC blitters have the same problem, (three sources, pattern and source, and the destination), but they get away with it because they use byte-addressing (chunky!) rather than bit addressing, and thus a separate bit-mask was not needed.