freenode/#sbcl - IRC Chatlog
Search
16:48:52
karlosz
why does one need compute-lra-code at all? isnt it enough to have jalr give the link register at the call site and then add the label offest off of that?
17:40:18
minion
stassats, memo from scymtym: sb-sprof for non-x86 should be fixed but i didn't test it
17:43:03
stassats
karlosz: compute-lra-code does just that, take LRA and apply an offset from it to the CODE object
17:48:46
karlosz
stassats: i was going to keep the lra as a descriptor reg so i wouldn't need an extra non desrciptor for it
17:54:04
stassats
for pinning, i would need to recognize the point where the return address is loaded into a register and the stack frame is deallocated, but PC is still in the callee
17:56:44
stassats
interior reg pairing could work, but suppose i want to get rid of register pairing
18:04:30
stassats
but if the LR register is not participating in the register pairing protocol, it can go into the allocatable register pool
18:22:31
karlosz
i dont want to pass the return-pc on the stack , so ill have to allocate a descriptor reg for it and change lra to be a non descriptor. and it wont be interior if its pinned?
18:24:39
stassats
i'll just try walking up the stack and collecting the return addresses, will report if i hit any snags
18:25:09
stassats
then i'll try removing the lra object, and then stop making LRA an interior reg, removing all the uses from arrays
18:30:24
stassats
x86 is a really great candidate for being precise, with its unencumbered addressing offsets
18:34:06
karlosz
this has been bothering me for a bit, but is there a reason why so much code uses macrolets and defmacros with once only when it could just be a function?
18:55:00
Xof
I still have the belief that cheneygc is easier to implement; my cunning plan was to get something minimal vaguely running and then find someone to do the second 90%
18:59:06
pfdietz
Just as long as your cunning plan doesn't end up like the last season of that show.
19:36:48
Xof
pfdietz: Blackadder Goes Forth made my history teachers in secondary school cry. (Now that I am older, I understand why)
1:23:34
pkhuong
can't believe I lost an hour trying to debug something in slime when my issue was just using a non-monospace font
2:20:15
karlosz
the ocfp not having a tn offset assigned randomly was due to the call vops not having :save-p and :move-arg specified