mo@seismo.CSS.GOV (Mike O'Dell) (03/10/87)
The only problem with bit blitters is that none of the chips available actually do what you want in all the cases. Tom Duff tried to use the old SUN blackk/white hardware assist to do the equivalent of a Blit and discovered it took over 30 times as much code and ran much slower to really try and use the beast. A vanilla 68K can blit faster than any currently known hardware chip. Now that Apple has a Cray, maybe they will design some hardware which actually does what the software types need, but for now, it doesn't exist. Besides, around Bell Labs there are stories of a new Son-of-Blit based on a 68020 which can blit in color (something almost noone understands, much less implements in a useful, intuitive way) and greyscale at a speed which any custom chip in any technology could envy. So, who keeps perpetuaing this myth that you gotta have hardware or it ain't chic?? -Mike O'Dell
hadeishi@husc7.HARVARD.EDU (Mitsuharu Hadeishi) (03/10/87)
In <43140@beno.seismo.CSS.GOV> mo@seismo.CSS.GOV (Mike O'Dell) writes: >The only problem with bit blitters is that none of >the chips available actually do what you want in >all the cases. Tom Duff tried to use the >old SUN blackk/white hardware assist to do the >equivalent of a Blit and discovered it took >over 30 times as much code and ran much slower >to really try and use the beast. A vanilla 68K >can blit faster than any currently known hardware >chip. No no no no no. A vanilla 68K is a pretty poor blitter; even the Sun blitters can easily outstrip a 68K. You need at least a 68010 to get reasonable performance on memory moves (the 68010 in tight loop mode is reasonably fast.) However, the series of recently announced graphics powerhouses such as the TI graphics processor blow away a 68000 without blinking an eye. Even the Atari blitter in reported to speed graphics operations on the ST by a factor of ten (probably highly exaggerated, but a factor of two or three would not be surprising.) The Amiga custom chips, three of them, are a full-blown graphics/sound/IO coprocessor which achieve blit speeds at or exceeding that of a 68020 even on a 7 Mhz bus. And they of course do all that in color. All those blit calls, by the way, are easily accessed through the Amiga relocatable system libraries. >So, who keeps >perpetuaing this myth that you gotta have >hardware or it ain't chic?? Me. -Mitsu
ali@rocky.UUCP (03/10/87)
In article <43140@beno.seismo.CSS.GOV> mo@seismo.CSS.GOV (Mike O'Dell) writes: >The only problem with bit blitters is that none of >the chips available actually do what you want in >all the cases. ... A vanilla 68K >can blit faster than any currently known hardware >chip. Hmm, a vanilla Amiga (without the 68020 board), containing a vanilla 68K running at 7.2 MHz, can run the game of Life, written in C, at 1.2 Million cells a second. (20 generations a second on a 320 by 190 lo-res screen). Just recently someone posted a "super-fast" life game to comp.sys.atari.st, written in optimized assembler, and it can crank out 400,000 cells/sec. And the atari st is probably the fastest (in terms on raw computing) 68K micro around ---- Its clock is 8 Mhz (vs 7.2 Mhz for the Amiga) and it has no multitasking overhead (like the few percent, in the Amiga)... Now how can the Amiga do it 3 times faster? The blitter, of course. Amiga has a 3 source, 1 destination, multiple shift, multiple modulus blitter that can do almost anything, with the appropriate programming. I would never have thought you could run Life on a blitter, but, hey, it runs, and runs much faster! One more thing --- Because the game of Life is very blitter oriented on the Amiga, it leaves the 68000 free most of the time. Thus I can run my C compiler concurrently with the game of Life and both run fine, with minimal slowdown... Ali Ozer, ali@score.stanford.edu, decwrl!rocky.stanford.edu!ali
frankb@crash.UUCP (03/10/87)
> A vanilla 68K can blit faster than any currently known hardware chip. I think the boys at Texas Instruments would be interested in hearing that assertion. If true, it would mean that all the excitement about the TMS 34010 is for naught! Why use a 34010 when you can use a 68000? > ... there are stories of a new Son-of-Blit based on a 68020 which can > blit in color (something almost no one understands, much less implements > in a useful, intuitive way) and greyscale at a speed which any custom > chip in any technology could envy. Again, I'm sure TI would be interested in hearing this, since their TMS 34010 blits color in interesting and useful ways. I'm sure Apple would also be interested in this, since they based the transfer modes of the new Quickdraw's color CopyBits on the transfer modes native to the 34010--and have stated in documentation that they are working on a graphics accelerator card based on the 34010. Also, Levco has been publicly noted as working on adapting the 34010 to the Macintosh. ___________________________________________________________________________ Frank Boosman | "You fool, I'll take your Silicon Beach Software | soul anyway! Surrender to ARPA/UUCP: {sdcsvax, nosc}!crash!pnet01!frankb | my awesome power!" -- Jon MCI Mail: fboosman | Lovitz [as Satan], BIX: frankb | Saturday Night Live
ali@rocky.UUCP (03/10/87)
In article <43140@beno.seismo.CSS.GOV> mo@seismo.CSS.GOV (Mike O'Dell) writes: >The only problem with bit blitters is that none of >the chips available actually do what you want in >all the cases. ... A vanilla 68K >can blit faster than any currently known hardware >chip. One more important point about blitters. A blitter is essentially another processor, except it can do various things better than a CPU (for example, area fills, moves, boolean operations on bit planes, etc). But obviously there are some things it can't do (such as multiply, divide, various complicated addressing modes, etc). Thus a blitter and a CPU live together real well: They are good at different tasks, and, with a properly designed system (hardware and software), they should each work on the things they are good at without bothering the other. Thus you can expect a performance boost out of your machine even if you attach a wimpy blitter (ie, a blitter not better (in terms of speed at which it can move/fill memory, etc) than your CPU). Otherwise, you have a wonderful $7000 (you need a keyboard and a monitor, no?) system with a 68020 who spends most of its life redrawing windows. On the Amiga, while the 68000 is working on a CPU intensive program (such as compiling, TeXing, etc), I can go ahead and move windows around, resize them, click and view menu items, etc, without interrupting the computations. This is because the blitter is responsible for doing most of the windowing stuff, and except for the few clock cycles the 68000 spends on instructing the blitter what to do, it can go ahead with what it was doing. Sorry for laboring the point, but blitters are far more important and capable than you seem to think they are. And they are a necessity in any system destined for graphics work. Ali Ozer, ali@rocky.stanford.edu
kearns@tom.columbia.edu (Steve Kearns) (03/11/87)
Just hold your breath. What with the TI and Intel graphics chips recently announced, I dont think we will have to wait long to see video cards using these chips. Also witness a previous letter in which someone reported a company that had used the quickdraw bottleneck to implement quickdraw with the TI chip. -steve
jdm@gssc.UUCP (03/11/87)
In article <43140@beno.seismo.CSS.GOV> mo@seismo.CSS.GOV (Mike O'Dell) writes: >The only problem with bit blitters is that none of the chips available i >actually do what you want in all the cases.....A vanilla 68K can blit faster >than any currently known hardware chip. Now that Apple has a Cray, maybe >they will design some hardware which actually does what the software types >need, but for now, it doesn't exist..... > >So, who keeps perpetuaing this myth that you gotta have hardware or it ain't >chic?? > > -Mike O'Dell um, mike, i disagree. on several points. #1. There are some very good graphics chips available today. My favorite is the TI 34010, which is really a graphics CPU, rather than just a graphics turbocharger. It has it's own assembler, etc. and is as programmable as any well-rounded microprocessor. It handles ALL graphics - not just bitblts - in color, and is VERY fast. TI sells the chip as well as a development board for the PC. The chip can stand alone on a display card with oodles of memory, as opposed to a "graphics chip" such as the i80786, which typically requires external processor support of some kind. #2. A stock 68K - even running at, say, 10Mhz no-wait - is NOT very fast as a "graphics processor." I know that "not very fast" is a relative term, but I have worked with the Bell Labs bitblt code (originally by Rob Pike) and have written my own, using self-modifying code, etc. such that for a monochrome device, the inner blt loop was optimally less than 14 words (!). Believe me, it is not HALF as fast as the aforementioned 34010. Sorry. Admit it - scroll speed on the Mac is not as fast as say, a character-generator based terminal, is it? So there IS room for improvement. #3. The point of graphics accelerators is to off-load the main CPU so it can be doing "more important" things. The 'ole "two heads are better than one" syndrome. This becomes especially important in multi-tasking (read A/UX) environments. Ideally, tasks are distributed among specialized hardware that can expedite them, but in a real-world situation, cost trade-offs are usually involved, so the CPU is forced to do the dirty work because "slaves" are too expensive. -- jdm -- in real life: John D. Miller, Graphic Software Systems, Inc., Beaverton OR ...!{tektronix!verdix}!sequent!gssc!jdm (503) 641-2200