Search
7:28:43
karlosz
does ppc64 really work? it assumes that n-fixnum-tag-bits is the same as word-shift in a lot of places, unlike arm64
7:28:44
minion
karlosz, memo from stassats: it's going to go into the thread object anyway, so don't bother too much
7:29:35
karlosz
so in places like %more-arg-values or the array reffers the difference seems crucial as well
10:03:58
stassats
minion: memo for emaczen: don't leave the main thread, execute it there from the beginning
10:03:59
minion
Remembered. I'll tell emaczen when he/she/it next speaks.
10:04:34
stassats
karlosz: no, ppc64 doesn't really work
10:04:49
karlosz
got it, that explains a lot
10:05:17
stassats
>unlike arm64, yeah, that word-shift thing was a source of major pain
10:06:58
karlosz
right, but its probably way less brittle to assume that n-fixnum-tag-bits != word-shift
10:08:08
karlosz
better is to write (unless (= tag bits word-shift) (inst slli ...))
10:08:21
karlosz
that way its tag agnostic without generating a spurious instruction on 32-bit
10:08:43
stassats
n-fixnum-tag-bits is kinda configurable
10:08:49
karlosz
i think i can manage with keeping 32bit and 64bit united for the time bein
10:09:01
stassats
so look at how it's handled in some x86-64 vops
10:09:29
karlosz
yah, i already wrote in the word-shift - tag bits thing that the arm64 backend does
11:32:18
stassats
i wonder if i can still break inlining
11:42:01
stassats
the binaries for sparc or mips can't be used to build current HEAD
11:42:19
stassats
i can build in steps, and i could make new binaries, but will they work on old glibcs?
11:44:16
stassats
2.28 on sparc and 2.19 on mips
12:49:31
stassats
The function SB-EXT:PURIFY is undefined.
16:12:54
stassats
i guess inline-expansion-limit should be per inline function, not for the whole component
16:13:20
stassats
or it shouldn't exist at all and properly deal with recursive inlining
16:18:40
stassats
the problem of SOME/EVERY and inline expansion is that it introduces an inline function after the IR has been optimized
16:18:53
stassats
now, can i induce that behavior without define a transform
16:21:22
stassats
i can even delay it for a long time until constraint propagation
16:25:25
stassats
and i can break my fix
16:25:40
stassats
"couldn't inline expand because expansion calls this LET-converted local function:"
16:25:50
stassats
not an error, but not something i want
16:39:01
stassats
it's going to happen rarely, unlike the SOME/EVERY stuff
16:39:28
stassats
but inline copying is the answer, and i want to use for some automatic inlining as well