Search
Tuesday, 11th of February 2020, 18:01:25 UTC
20:15:02
Xach
Hmm, I don't understand how declaims for defuns make the function name have a :declaration source-definition.
20:29:41
flip214
slyrus__: which and how many cores?
20:38:04
stassats
flip214: i build in 50 flat on some old four cores
20:40:46
flip214
stassats: including CFLAGS=-g __LDFLAGS__=-g? Any tricks beside SBCL_MAKE_PARALLEL and SBCL_MAKE_JOBS, perhaps running only make-target-2.sh instead of make.sh or so?
20:41:23
stassats
make-target-2.sh wouldn't have anything to run
20:42:08
stassats
the host sbcl is compiled with speed 3, safety 0, without threads
20:43:57
stassats
and without sb-unicode
20:44:44
stassats
but that's not worth a lot
20:45:25
stassats
(but is for me, when i have to redo all the time and every second counts)
20:48:03
flip214
I understand what you're getting at! For me, recompiling twice a week, that's not a real difference, but for you it is
20:48:58
stassats
i have compiled more times since my last message
20:49:12
flip214
201.33user 25.47system 1:27.79elapsed
20:51:23
flip214
parallel QL or ASDF would still be very nice
20:51:59
dlowe
ITA made a parallel ASDF called pcvb and it was fairly complicated.
20:52:36
dlowe
I can't keep them straight anymore
20:52:52
dlowe
piou is the parallel one, I think. xcvb is the asdf replacement that didn't take off
20:52:53
flip214
but parallel QL should be much easier
20:53:50
stassats
trying the latest sbcl as a build host, let's see what the recent compiler changes bring
20:54:01
stassats
(expecting a slow down / no change)
20:54:35
stassats
right, just some measurement noise
20:57:09
stassats
but my laptop fast-sbcl had sb-thread enabled, without it goes 2m13.440s => 1m51.990s
20:57:19
stassats
also with sb-fasteval for parallel make-host-1
20:59:35
stassats
except make-host-1 is no different, parallel or not
22:15:09
karlosz
wow, sb-thread really adds that much time?
22:15:23
karlosz
i didn't expect threading to affect build time that much
22:15:37
karlosz
sb-unicode i understand because there's a lot of unicode data to slurp
22:15:50
karlosz
but just having slower specials access seems like a lot
22:15:56
stassats
and it's a newer checkout
22:16:00
stassats
so, not validated results
22:17:00
stassats
it all may be due to sb-fasteval
Wednesday, 12th of February 2020, 6:01:25 UTC