Jump to content

Chilly Willy

Members
  • Content Count

    842
  • Joined

  • Last visited

  • Days Won

    1

Chilly Willy last won the day on February 28 2013

Chilly Willy had the most liked content!

Community Reputation

353 Excellent

About Chilly Willy

  • Rank
    Dragonstomper
  • Birthday 08/24/1965

Profile Information

  • Gender
    Male
  • Location
    The Heart of NASCAR
  • Interests
    Tennis
    Bowling
    Programming

Recent Profile Visitors

The recent visitors block is disabled and is not being shown to other users.

  1. Yeah, he's clearly not using vasm. 😄 It's generating something and silently dropping the errors. That can be a pain if you aren't fluent in 680x0 and don't realize what you're doing wrong.
  2. The destination EA cannot be PC relative, only the source. No telling what the assembler is generating for the opcodes.
  3. No, what you're not doing is reading the thread completely. I always called for replacing JUST THE CUSTOM RISC processors with mature ones. When I said 3 ARMs would kick ass, it was following the rest of the thread in that you would be replacing the 68000, the GPU RISC, the DSP RISC, and NOTHING ELSE with ARM processors. You'd still have the blitter and object processor and all the other hardware, just not the processors Atari used. My original post left the 68000, only replacing the GPU and DSP RISC processors. Someone else pointed out that replacing the 68000 would also boost performance. That was the basis for that part of the thread, not using ONLY RISC processors and nothing else.
  4. The GPU and Blitter are only "tightly coupled" in that they share silicon in Tom. Many games use the 68000 rather than the RISC to drive the blitter. Most of the description of the GPU RISC itself is applicable to virtually any CPU of the time and nothing special. The 4KB of fast internal ram wasn't anything special, either. The SH2 has 4KB of internal ram that can be used as 4KB of 4 way set associative cache, 2KB of 2 way set associative cache + 2KB fast scratch ram, or 4KB of fast scratch ram. The GPU can only use its ram as scratch ram. ARM chips might also have fast scratch ram as we saw in things like the GBA. The rest of the description has nothing at all to do with the GPU RISC and is the part I say they should have kept. In case you're wondering, the original SH1s had 1KB of ram for fast scratch. Later versions had 2KB. ARM was kinda the oddball of the time in that it didn't come with internal ram standard, but could be added to the design if desired for a particular use. Various MIPS chips had different cache sizes internally.
  5. BSEE. I've designed computer cards and input adapters and other such devices. I also write software. It's my opinion (and just an opinion) that taking hardware classes makes for better programmers than just software classes. Those two quotes go together in this discussion. Tom has some extra hardware associated with handling graphics alongside the RISC. The RISC itself has nothing special for graphics other than being fast and having a very limited matrix operation. My own thought is that they should have kept all the stuff in Tom other than the RISC processor, which would have made Tom cheaper and easier to debug. Ditto for Jerry - they had some limited operations to make saturated operations work on audio summing, but that's hardly an issue with audio rates. Again, rip out the RISC processor and keep the rest making Jerry cheaper and easier to debug again. Maybe it wouldn't have had the serial bug (for example) if they hadn't had to spend all their time working on RISC bugs instead. So I wasn't saying get rid of Tom and Jerry, just get rid of the RISC processors inside them. Keep the Object Processor and the Blitter and all the rest, just not the custom RISC cores. That would have left them plenty of time to work the bugs out of the rest of the hardware and given them mature programming tools to work with. The SuperH is not hard to program at all - it's much easier to be more effective than x86, for example. That's talking about assembly, of course, as C is the same for everybody - it's all on the compiler to turn the C into good assembly. The Saturn was much more capable then the 32X for a number of reasons: the SH2s were clocked faster and had 32-bit access to 1MB of ram (where the 32X SH2s just had 16-bit access to 256KB of ram), and the Saturn had a dedicated video rendering processor (VDP1) while the 32X only had a simple line fill function in its VDP. The 32X would have been far more capable if Sega had given it only one SH2 along with the VDP1 from the Saturn, but SOJ didn't want the 32X competing with the Saturn... they didn't want it at all, to be honest.
  6. Yeah, three ARM processors would have kicked some serious arse. Three MIPS probably would have been too expensive. Three SuperH would have worked... given when it was developed, that would have been three SH1 processors.
  7. Sorry about the delay... had a hectic schedule at work last week. Subdivided affine was a common technique used on better games to overcome (most of) the issues with affine mapping. Instead of rendering the entire raster line of a triangle using the same parameters, you cut the line into pieces... often 8 or 16 pixels at a time. This minimized the warping at the expense of a little more overhead. One of the best explanations of old-school 3D rendering can be found in the old articles Chris Hecker wrote for Dr Dobbs WAAAAAAAAY back in the stone-age. You'll find those articles on his site, here: https://www.chrishecker.com/Miscellaneous_Technical_Articles Another great thing covered by Chris - getting the fill conventions decided on at the start so that you avoid issues like the slight gaps between triangles you sometimes saw on PS1 games. I'd have the GPU rasterize the triangles, doing subdivided affine calculations to feed line segments to the blitter. More work, slightly lower speed, but much less warping. Perhaps make a decision on doing the whole raster line as affine vs doing subdivided affine depending on the depth of the triangle. The further away a triangle is, the less affine mapping matters. That's why fish-eye like on various PS1 games was always worse when you walked right up to the wall. And that's pretty much how later PS1 games dealt with the issue - tessellating triangles close to the camera to avoid a large span of pixels that show the warping.
  8. From the POV of an engineer/programmer, I'd say they should have gone with a more mature, existing RISC processor instead of trying to roll their own. The buggy state of Tom and Jerry combined with the lack of stable programming tools for them killed the Jaguar more than almost anything else. The 68000 was a good choice; a custom RISC was not. They'd have been better off with MIPS, ARM, or SuperH instead. Those processors were cheap, fast, and had stable compilers/assemblers/debuggers.
  9. The general idea for 3D on the Jaguar can go like this (note that there's quite a few ways games can handle this): 68000: Run the game logic. Once per frame, tell Tom to render the display. Tom: Go through the game level and entities and compute triangles from game world coords to screen coords, generating a list of triangles. Go through the triangles and rasterize them, generating a list of blitter commands to draw each line (or simply command the blitter right then for each line... depending on your game, one might work better than the other). Blitter: Draw each raster line for each line of every triangle rendered. This can be controlled by the 68000 or Tom. Rendering can be solid color or textured, with or without z control, with or without gouraud shading, and always affine (there is no provision for perspective correct rendering on the blitter).
  10. I'm an old Myst fan. I played the hell out of the Mac edition back in the day. I bought an Elansar cart specifically because it reminded me of Myst. So Myst would be my pick for best of this bunch. Iron Soldier 2 is pretty good, too. It's kind of reminds me of Metal Head for the 32X, but better. I liked playing Metal Head back in the day - it was one of the better 32X titles. Hover Strike? Reminds me of Battlecorps, but not as good. Battlecorps is a great game on the Sega CD - Hover Strike makes me want to turn off the Jaguar and play Battlecorps instead. 😄
  11. In some cases, they're hoping you don't have the time, so you simply pay your way through. That's what Bethesda's doing with the upcoming Fallout 76 Seasons - you can spend 100 to 200 hours every season to grind your way to the objective, or you can pay your way to it instead. They're also reducing the ability to earn atoms in-game, so guess what they're hoping most people will do.
  12. He's just going through the games alphabetically and seeing what people think of different titles. It makes some sense. You could also go through by genre, by publisher, by publishing date, etc. The problem is that going alphabetically leads to some pretty weird match ups, like this one. The games are almost unrelated except for being Jaguar games. Xenon 2 is a decent enough shooter. Pretty standard. Val d'Isere Skiing is line racer, just with skiers instead of cars. I don't really care for White Men Can't Jump - never could get into basketball games. Wolf3D is Wolf3D. Not much to say there, but it's a decent enough port of it. Of the batch, the one I think I like the best is the skiing game.
  13. They're all almost completely different game categories, so they could reasonably all be called the best - for the list given for their category. You could argue that the Skiing game and Basketball game are both sports, but most people separate winter sports like skiing from regular sports. As to best in the list, they're so different that most folks will all give different answers. It's more a question of which ones (if any) are considered BAD rather than which is the best.
  14. I think this is the crux of the matter. Any kind of Jaguar item is simply rare enough to make it valuable regardless of its state. It's like the Dreamcast BBA - it's so rare that even a used one in poor condition is worth more than everything else Dreamcast related put together.
×
×
  • Create New...