IRC logs for #openrisc Tuesday, 2012-08-07

stekernjuliusb_: I've been thinking about the "flag register" we were speaking about, I think it perhaps would be a good idea that that would be a "normal" GPR and not some special register. The same goes for the PC.15:24
juliusb_yeah, i've seen that before, well, you could just make it any GPR, right?15:33
juliusb_having it as a special register is VERY handy though15:33
juliusb_do you mean you could have a GPR specified as "the flag register" and all compares end up there?15:34
stekernyes, the latter16:19
stekernpin down one reg as flag reg16:19
juliusb_do you like the idea of having arbitrary regs selectable as the one we use to compare?16:40
juliusb_i've just done another spin of this spreadsheet btw16:40
juliusb_will probably post it somewhere16:40
juliusb_another cool trick I've seen is that, I believe, the 8051 has this thing where part of its equivalent of the SPR space is a local SRAM16:41
juliusb_so single cycle guaranteed memory16:41
juliusb_sorta like the QMEM thing they had in the OR120016:42
juliusb_except it's an architectural specification16:42
juliusb_well, if you _want_ that is16:42
juliusb_but you could store away a few KB in SPR space for whatever you want16:43
juliusb_well, hook up a memory to the SPR bus and you're guaranteed no latency, single cycle access to it16:43
juliusb_but anyway, that's not somethign i'm considering, but as I'm coding the addressable space of the SPRs, it came to mind that there's a _lot_ of address space in the SPRs16:44
stekern"having arbitrary regs selectable..." <-hmm, I'm not sure I understand what you meant by that16:47
juliusb_like, if you go "sf rA, rB" and you want to set the flags for the comparison between rA and rB17:07
juliusb_the flags would normally go into the SPR, but if you had a version which put the flags result into an arbitrary register, so "sf, rA, rB, rC" and the flags results go into rC instead of the SPR17:08
juliusb_here I'm considering the flags reg to have individual results for "testing" registers (negative, zero etc.) and then comparison flags such as equal, gte, lte, lt, gt, gtu, ltu17:10
juliusb_all sorts of stuff17:10
juliusb_ideally the most important flags go into 15:017:10
stekernyeah, that sounds good17:13
juliusb_(because on some of the the 16-bit insns, there's only room for 3-bits of flag selection)17:13
juliusb_so actually, what I have so far is pretty bad\17:14
juliusb_not overly RISCy17:14
juliusb_the implementation won't be small I believe17:14
juliusb_although, as far as I can tell it won't be the end of the world17:14
juliusb_code density is the most important thing17:14
stekernyou could probably call that sf cmp17:14
juliusb_yeah or just cmp17:15
juliusb_will get my new spreadsheet out this arvo anyway17:15
stekernthat's what I meant, rename sf -> cmp ;)17:16
juliusb_i have a little bit of paid employment work todo :P17:16
stekernbut I still think I'm more found of a pinned down gpr flag than the option of spr or arbitrary reg though17:17
stekernbut individual flags sounds good17:19
juliusb_well it's good that you're fond of that, because the instructions you'll use will be smaller :)17:37

Generated by 2.15.2 by Marius Gedminas - find it at!