Search
14:49:57
stassats
why can't we divide via multiplication for signed fixnums?
15:49:15
Krystof
stassats: I think it "just" takes a bit of work to implement it
15:49:36
stassats
the original paper has it, trying to digest
15:52:30
stassats
spotted it in get-internal-real-time, that divides by 1000, but the whole thing could be done in C
15:53:09
stassats
but making our own gen-signed-truncate-by-constant-expr would benefit more things
15:55:46
stassats
and lisp_init_time has to be dereferenced twice, and sb-alien is too primitive to handle using a variable twice
17:10:35
stassats
ok, got signed-truncate
17:11:05
stassats
had to add a signed-multiply-high vop, wonder if there's something portable
17:16:02
stassats
well, i can do arm64 and x86-64
17:24:14
fiddlerwoaroof
stassats: I managed to get the build to complete on my mac (up to make-target-2.sh)
17:24:27
fiddlerwoaroof
However, I'm getting this error: ensure_space: failed to allocate 1048576 bytes at 0xf0000000
17:25:54
fiddlerwoaroof
I can find the code that prints this message, but I'm wondering what the normal cause of this message is? mmap failing?
17:26:34
stassats
sorry, no time to deal with that
18:26:55
flip214
with run-program, can I have arbitrary file handles passed on? Ie. create a socketpair or pipe and pass one end as FD 17 or whatever?
19:21:32
stassats
the floor code in the paper doesn't make sense to me, as it overflows
19:22:21
stassats
flip214: crate a new fd-stream
19:23:02
stassats
and didn't get it right in time for the freeze, oh well
19:28:43
stassats
ok, no floor then, truncate still helps with idiv, but there's two imuls for floor