|
00:31
lizmat_ joined
01:49
ilbot3 joined
05:06
geekosaur joined
05:41
brrt joined
05:42
domidumont joined
05:51
domidumont joined
06:39
brrt joined
07:30
brrt joined
07:38
brrt joined
07:43
domidumont joined
07:46
zakharyas joined,
domidumont1 joined
07:47
zakharyas joined
08:07
brrt joined
|
|||
| lizmat_ | And a new Perl 6 Weekly hits the Net: p6weekly.wordpress.com/2017/05/16/...-the-alps/ | 08:08 | |
| brrt | \o lizmat_ | ||
|
08:13
zakharyas joined
09:26
brrt joined
|
|||
| Geth | MoarVM/even-moar-jit: 45bf6e94eb | (Bart Wiegmans)++ | 3 files Disable expr compiling for throwish/invokish ops These ops need guarding, and that is not yet implemented. This fixes CORE.setting compilation for now. |
09:34 | |
| MoarVM/even-moar-jit: 9bc0ed2b9c | (Bart Wiegmans)++ | 2 files Split jit-dump from jit-bisect The jit-bisect debug tool was somewhat unreliable and frankly did too many things at once. I've split the code to get a specific JIT dump into its own script and refactored the bisecting code. Stuff is generally more reliable now. |
|||
| brrt | i'm still figuring out what is broken and why... | 09:46 | |
|
10:01
zakharyas joined
11:41
praisethemoon joined
|
|||
| Geth | MoarVM/even-moar-jit: 3f5211cb8c | (Bart Wiegmans)++ | src/jit/linear_scan.c tile->args is not tile->values Also, values and register_spec are +1 biased compared to refs because they include the output register which refs does not. It might be a good idea to change that (have tile->refs[0] == tile->node). |
12:54 | |
| MoarVM/even-moar-jit: e0b99d02bd | (Bart Wiegmans)++ | src/jit/linear_scan.c Invoke prepare_arglist_and_call for CALL tile It could potentially happen that the register used by CALL would be allocated after the ARGLIST was processed, e.g. when the associated value is spilled and loaded by a synthetic tile. This could then overwrite the register picked by prepare_arglist_and_call for resolving the conflict between the ARGLIST and CALL registers. Now that prepare_arglist_and_call is executed for the CALL tile this can no longer happen. |
|||
| MoarVM/even-moar-jit: 83e4976189 | (Bart Wiegmans)++ | 8 files Wrap invokish and throwish with a guard tile Because this means that control flow can be interrupted and resumed right after the compiled node, the expression assembler flushes all active values after the guard, forcing them to be reloaded after. The pre-throwish pseudotile (which is responsible for loading the JIT ... (7 more lines) |
|||
| brrt | ^ the last commit unblocks decont and similar invokish / throwish ops | ||
| can be improved (i want to do 'optimistic' flushing, i.e only flush when necessary) but yeah | 12:56 | ||
| nine | So settings compilation now works? | 13:04 | |
|
13:17
brrt joined
|
|||
| brrt | yes | 13:17 | |
| i intend to keep it working, but with ever larger pieces of code JITted :-) | 13:18 | ||
| nine++ nice blog post on the PTS | |||
| nine | Thanks :) | 13:21 | |
| brrt: that sounds as it's about time people joined the effort? | |||
| brrt | i'm guessing we're very near that point, yes | 13:23 | |
| Zoffix | \o/ | 13:25 | |
|
14:01
zakharyas joined
15:06
zakharyas joined
15:25
brrt joined
17:21
zakharyas joined,
AlexDaniel joined
17:23
samcv joined
17:36
praisethemoon joined
18:29
japhb joined
18:49
domidumont joined
18:57
Ven joined
19:33
domidumont joined
19:42
Ven_ joined
19:52
robertle joined
19:58
Ven joined
20:01
zakharyas joined
20:27
Ven_ joined
20:48
committable6 joined,
Ven_ joined
22:19
MasterDuke_ joined
|
|||