00:00:00 --- log: started forth/01.09.26 00:14:59 * aaronl is away: homework 01:13:41 --- quit: aaronl (The name's X. Bitch X.) 06:19:52 --- join: Fare (Fare.LISPM@samaris.tunes.org) joined #forth 06:23:35 --- quit: Fare (Excess Flood) 06:23:38 --- join: Fare (Fare.LISPM@samaris.tunes.org) joined #forth 08:49:21 --- join: I440r (mark4@A010-0087.BLMG.splitrock.net) joined #forth 08:50:06 --- mode: ChanServ set mode: +o I440r 08:50:11 --- mode: I440r set mode: +o Fare 08:50:13 --- mode: I440r set mode: +o clog 09:27:11 --- join: edrx (edrx@200.240.18.107) joined #forth 09:53:11 --- quit: edrx (Ping timeout for edrx[200.240.18.107]) 11:27:48 --- quit: I440r (Read error to I440r[A010-0087.BLMG.splitrock.net]: Connection reset by peer) 12:45:54 --- join: downix (downix@user-38ldc24.dialup.mindspring.com) joined #forth 12:47:04 man, empty 12:57:18 --- join: downix_ (downix@user-38ldeou.dialup.mindspring.com) joined #forth 12:58:08 --- quit: downix (Ping timeout for downix[user-38ldc24.dialup.mindspring.com]) 12:59:09 --- nick: downix_ -> Downix 13:00:31 downix! 13:00:37 What becomes of you? 13:00:50 working hard 13:00:57 trying to research forth CPU's 13:01:06 I need to replace the CPU in my chipset with something much smaller 13:01:41 I have only 130k gates to fit a processor that can handle 3D 13:01:50 figured a forth CPU might be just what the doctor ordered 13:02:29 wouldn't an ARM7 processor be << 130k gates? 13:02:50 ARM7 is horrid for 3D 13:03:01 nowhere near fast enough 13:03:39 Downix: FORTH is not that great, but I guess you can put several FORTH processors and parallelize in 130k gates 13:03:53 that was the thought 13:04:12 I've heard of forth CPU's fitting into 20k gates 13:04:15 however, for optimal performance, these processors depend on programs running in local storage 13:04:20 which means I could fit 7 of them 13:04:22 * Downix nods 13:04:27 which was the goal 13:04:29 and then you have to spend some of these 130k gates in caches. 13:04:47 I have the old CPU cache still there 13:04:57 enough to fit your tight loops, their data, and some extra space. 13:05:00 48k worth 13:05:33 48k is quite nice for 2 or 3 processors. might be tight for 7 of them. 13:05:41 not 48k gates 13:05:50 48kilobytes of cache 13:06:09 130,000 gates, roughly, to fit a processor into 13:06:33 so for your 3D target applications what's the size of the working set? 13:07:34 * Downix shrugs. I was letting someone else handle that before, when I was using a MIPS cpu 13:08:08 If it wasn't expected, I wouldn't have bothered with 3D 13:09:24 What does "3D" mean, in practice? 13:09:36 typically games, like Quake 13:10:15 the Gameboy Advance has an ARM and seems ok with it. 13:10:44 it's also got a 3D co-processor 13:10:49 oh 13:12:45 the CPU in the system is a PPC, I'm trying to make the 3D co-processor 13:12:52 * Downix didn't choose the CPU, his client did 13:13:34 does that mean doing floating-point stuff, or fixed-point integer things? 13:13:45 floating-point 13:14:10 efficient FPU can eat lots of gates 13:14:27 * Downix nods 13:14:46 doesn't need to be terribly efficient, just scalable for when I can give it more gates in later designs 13:15:00 I considered a M2M vector unit 13:15:08 M2M ? 13:15:14 memory to memory, no registers 13:15:18 * Fare is reminded of transputers 13:15:29 * Downix doesn't know transputers well 13:16:15 problem is: there are lots of great things, but because of IP reasons, even if noone uses them anymore, you can't reuse them -- it's DEAD 13:16:21 * Downix nods 13:16:28 right 13:17:47 hence one of the reasons why I was considering an 8-pipeline Forth unit 13:18:30 8-pipeline??? 13:18:44 yes, 8 units in parallell I figured would fit in the space 13:19:02 beware of interlocking problems, with that many units 13:19:27 all the more since 6KB of cache per processor is not that much 13:19:55 * Downix nods 13:20:29 btw, what CAD tools are you using? What do they output? VHDL? 13:20:40 I'm using VHDL 13:20:47 directly? 13:20:58 and using WebTools from Xilinx to output to FPGA 13:20:59 * Downix nods 13:21:02 of course 13:22:29 beware that you'll also have to fit the FP library in your cache. 13:22:37 hmm 13:22:55 too cramped for 8 13:23:20 doesn't the PPC already have a FPU unit? 13:23:30 or is it an embedded version w/o FPU? 13:24:03 a stack-based FPU, perhaps 13:24:03 the PPC has it's own FPU 13:24:03 but I want a 3D unit, to save the CPU some work 13:24:52 I'm not convinced you can do a 3D unit that can rival with Altivec et al with 130kgates on a tight schedule. 13:25:04 not planning on rivaling 13:25:04 hardware AND software. 13:25:10 and the PPC here doesn't have Altivec 13:25:37 G3 based (possible G4/G5 in future systems) 13:26:04 although it's nice to have a coprocessor, anyway 13:27:13 * Downix nods 13:27:22 the other thought was to make a DSP-port 13:29:12 and use an external DSP for 3D work 13:30:28 with it's own memory bank 13:35:47 hmm 13:36:01 a bit complex 13:39:17 --- quit: Downix (Read error to Downix[user-38ldeou.dialup.mindspring.com]: Connection reset by peer) 13:52:49 --- join: downix (downix@user-38lddr2.dialup.mindspring.com) joined #forth 13:53:21 anyways 13:53:53 a top of the line LISP processor was 500k transistors. 13:54:09 including FPU (dunno about cache) 13:54:12 * downix nods 13:54:22 so that would fit 13:54:27 since a gate is 8 transistors 13:54:50 how much would that do in FPGA gates, I don't know, because there's lossage in the gate->transistor count when using FPGA. 13:54:56 * downix nods 13:54:57 I know 13:55:09 likely that would end up around 120k FPGA gates 13:55:25 and a LISP CPU wouldn't be very useful for 3D work 13:55:42 LISP processors were designed with LISP tools 13:55:42 great for the 2D work I'm doing, however 13:55:53 I suppose one could dump VHDL from that 13:56:09 I have a LISP 2 VHDL tool 13:56:13 I used that in a few areas 13:57:30 downix: LISP processors with a FPU coprocessor were used for 3D CGI in 1980's pictures 13:57:36 Hrm? 13:57:38 hmm 13:57:47 well, where could I find one for study? 14:03:20 the software from that became Mirai from wingededge.com 14:03:47 downix: formerly the N-world from Nichimen Graphics 14:04:03 hmm 14:04:07 formerly the S-world from Symbolics 14:23:41 --- join: Downix_ (downix@user-38lddo8.dialup.mindspring.com) joined #forth 14:23:45 --- quit: Downix_ (Read error to Downix_[user-38lddo8.dialup.mindspring.com]: EOF from client) 14:24:08 --- join: downix_ (downix@user-38lddo8.dialup.mindspring.com) joined #forth 14:24:13 --- quit: downix (Killed (NickServ (GHOST command used by downix_))) 14:24:38 --- nick: downix_ -> Downix 15:25:32 --- quit: Downix (BitchX: need we say more?) 16:20:27 --- join: aaronl (aaronl@vitelus.com) joined #forth 16:48:43 --- join: edrx (edrx@200.240.18.110) joined #forth 18:24:21 --- quit: edrx ([x]chat) 18:41:49 --- log: started forth/01.09.26 18:41:49 --- join: clog (nef@bespin.org) joined #forth 18:41:49 --- names: list (clog aaronl @Fare) 19:09:05 --- join: Speuler (l@c38038.upc-c.chello.nl) joined #forth 19:09:37 g'day 19:19:44 --- quit: Fare (Connection reset by pear) 20:06:56 * aaronl is away: enterprise 23:15:25 * aaronl is away: sleep 23:59:59 --- log: ended forth/01.09.26