|
00:08
ggoebel joined
01:49
ilbot3 joined
06:29
domidumont joined
06:36
domidumont joined
07:14
stmuk_ joined
07:58
geekosaur joined
08:19
brrt joined
|
|||
| brrt | good * #moarvm | 08:21 | |
|
08:39
praisethemoon joined
|
|||
| timotimo | yo brrt | 09:13 | |
| brrt | \o timotimo | 09:14 | |
| i'm reasonably confident that a bitmap is the right choice for my set-of-live-ranges-per-block | |||
| timotimo | so for each block you'll have as many bits as there are ranges? | 09:16 | |
| brrt | yes | 09:18 | |
| so with 64 bit pointers, or 32 bit keys, you're going to need a pretty large number of live ranges and a pretty low 'fill' per block to beat the memory usage of the bitmap | 09:21 | ||
| and in terms of compuation efficiency it's no contest | |||
| timotimo | mhm | 09:22 | |
|
09:48
lizmat joined
|
|||
| Geth | MoarVM: 3e4ed6bffe | (Samantha McVey)++ | src/strings/ops.c Add can_fit_into_8bit funct, put logic used many places into one funct There are 7 places we use checks for this within src/strings/ops.c Writing out the conditionals which sometimes use || and sometimes use &&. They also sometimes use >= sign and other times > signs. Make it more consistent for ease of reading and clarity. |
10:37 | |
| MoarVM: 957759c84b | (Samantha McVey)++ (committed using GitHub Web editor) | src/strings/ops.c Merge pull request #591 from samcv/master Add can_fit_into_8bit funct, put logic used many places into one funct |
|||
|
10:37
brrt joined
11:35
Ven joined
11:41
Ven_ joined
12:25
Ven_ joined
12:33
Ven joined
12:50
Ven_ joined
13:04
Ven_ joined
13:15
colomon_ joined
13:17
brrt joined
13:21
Ven_ joined
13:33
Ven joined
13:34
lizmat joined
13:40
Ven_ joined
14:03
Ven_ joined
14:10
Ven_ joined
14:31
Ven joined
14:38
Ven_ joined
14:43
ZofBot joined
14:44
Ven_ joined
14:45
ZofBot joined,
Ven_ joined
|
|||
| jnthn | Nice. From scratch MoarVM build on my new machine is 8s. :) | 14:51 | |
| And spectest in 106s | 14:59 | ||
| With TEST_JOBS=16 | |||
| TEST_JOBS=20 shaves a few more seconds off, but not much | 15:01 | ||
| Though with that a make test of Rakudo (with NativeCall already being precomp'd) takes 2s | 15:02 | ||
| Heh, with --jobs=20 I can get NQP to spectest in 2s also | 15:03 | ||
| (hardware improvements)++ :) | |||
| MasterDuke_ | nice. what is the new hardware? | 15:27 | |
| jnthn | Has a Xeon 6-core (12 virtual) CPU, 32 GB RAM and a fast SSD :) | 15:30 | |
| dogbert17 | congrats, nice HW | ||
| MasterDuke_ | nice. i've gotten kind of spoiled by 20-core (40 virtual) xeons at work and have thought about going the xeon route at home | 15:33 | |
| dogbert17 | I thought you're looking into AMD as well | ||
| MasterDuke_ | but nine's numbers on his new ryzen are also pretty good, and that's a much cheaper option | ||
|
15:48
lizmat joined
16:12
Ven joined
16:36
Ven_ joined
17:25
praisethemoon_ joined
18:10
AlexDani` joined
18:11
AlexDani` joined
18:27
ZofBot joined,
buggable joined
18:28
Geth joined
18:42
zakharyas joined
19:17
AlexDaniel joined
19:58
geekosaur joined
20:04
praisethemoon joined
|
|||