| jonibo|laptop | what would be the consequence of making r0 just another register... i.e. not constant zero at all... when you need a 0 zero register, you can just use any register really... | 17:37 |
|---|---|---|
| jonibo|laptop | what's the concrete advantage of a constant zero register? | 17:38 |
| derRichard | danm, i forgot to resend you the r0-patch :-\ | 17:45 |
| derRichard | AFAIK a zero register is useful because you don't need to write 0 to it all the time... | 17:45 |
| derRichard | mips has also such an register | 17:46 |
| jonibo|laptop | yeah, I can see that argument... but what the uses of a register that is known to contain 0... seems a lot of operations can be optimized better if you already know that one of the operands is 0 | 17:48 |
| derRichard | i really don't know. i'm a pure software guy :) | 17:48 |
| jonibo|laptop | x && r0 = 0, might as well just load 0 to the result register | 17:48 |
| jonibo|laptop | x OR r0 = x, might as well just load x to the result reg | 17:49 |
| jonibo|laptop | SPR accesses use r0 a lot... but seems a small cost to load r0 with 0 as needed | 17:49 |
| juliusb | well it's probably the same argument for having 1, 2 or 0xffffffff preloaded in a register | 18:29 |
| juliusb | convenience | 18:29 |
| stekern | well, or1k being a 3-register instruction set, you kind of need the r0 to load the x into result reg | 22:51 |
| stekern | never mind that, you don't.... | 22:53 |
Generated by irclog2html.py 2.15.2 by Marius Gedminas - find it at mg.pov.lt!