Search
Monday, 9th of October 2017, 14:41:47 UTC
15:18:18
stassats
got rid of the weird JNE L1 JMP L5 L1: ...
15:18:29
stassats
which come in a lot from optional dispatch
15:27:29
stylewarning
stassats: do you see any value in peephole optimization for SBCL
15:27:44
stylewarning
I know it’s been thought about before
15:27:52
stassats
stylewarning: sure, but not any road to it
15:30:21
stassats
but shouldn't the stupidity of representation selection behind a peephole optimizer in any case
15:31:12
stassats
which is the source for the most "damn, i could write better assembly by hand"
15:36:00
stassats
in other news, i overtook nikodemus by the number of commits, now sitting behind dougk_
15:41:46
stylewarning
stassats: But will you overtake nikodemus by the number of cool features?
15:43:53
stassats
is the no cost (declare (ignore &rest-arg)) a cool feature?
15:50:28
stylewarning
CONSTANTLY should be so fast now
16:05:38
stassats
stylewarning: it is faster, but only if you pass to it a lot of arguments
16:06:38
stassats
faster either way, but that case is noticeable
16:09:50
stassats
constantly could be improved further, by not issuing any stack allocation code and just returning the closed-over value
16:11:11
stassats
done in a proper fashion would bring me closer to specialized entry points, for floats or words
16:12:09
stassats
but already have a pretty good angle on that, might just do it one day
16:21:36
stylewarning
That sounds useful
16:24:55
slyrus
sb-introspect appears to be broken again
16:26:11
slyrus
only this time it's a test failure, instead of a compilation failure
16:49:20
slyrus
dougk_: gc bignum fix broke sb-introspect
18:06:59
_death
hmm, latest commit has an sb-introspect test, allocation-information.7, which fails.. also seems to have silly code: (and (let ... (eq ...) (eq ...)) ...) .. and since it fails make install skips the module silently, may want to warn instead
Tuesday, 10th of October 2017, 2:41:47 UTC