| japhb | Anyone know offhand what hash algorithm we're using to hash keys in MoarVM's map/hash implementations? | 00:58 | |
| Currently deep diving some info about relatively recent hash algorithms, so curious which (if any) we're using. | 00:59 | ||
| ugexe | rapidhash github.com/MoarVM/MoarVM/commit/62...192812b7c8 | 01:05 | |
|
02:25
sourceable6 left,
notable6 left,
shareable6 left,
notable6 joined,
quotable6 left,
unicodable6 left,
committable6 left,
greppable6 left
02:27
committable6 joined,
bloatable6 left,
linkable6 left
02:28
shareable6 joined,
sourceable6 joined,
greppable6 joined,
unicodable6 joined,
quotable6 joined
02:29
bloatable6 joined,
linkable6 joined
|
|||
| japhb | thx | 02:52 | |
| Voldenet | huh, rapidhash benchmarks suggest that xxh3 is faster on most common PCs | 10:37 | |
| and gxhash is a lot faster | 10:39 | ||
| in fact xxh3 is so standard that it's packaged into .net8 | 10:45 | ||
|
11:41
librasteve_ joined
|
|||
| Geth | MoarVM: MasterDuke17++ created pull request #1959: Bump rapidhash to v3 |
11:55 | |
|
13:58
librasteve_ left
|
|||
| japhb | It's frustrating how benchmarks age, such that comparisons are only a snapshot in time, with the most active implementations switching leadership positions on the regular. | 16:22 | |
| This could of course be solved by someone benchmarking All The Things at a consistent interval, but most of the people who do that sort of thing don't do an all-inclusive suite, or have an axe to grind about certain algorithms and rather than just marking them in the results, don't test them at all .... | 16:23 | ||
| Witness the number of benchmarks that don't include anything (like gxhash) that's dependent on particular CPU features, or don't include any hash that doesn't have seeds of a certain size, or what have you. | 16:25 | ||
| FWIW Voldenet I *suspect* xxh3 is getting special treatment not just because it's a good performant hash (which AFAICT it is), but also because it was written by the same guy who created LZ4, zstd, and I suspect OpenZL (I know he's a contributor, but I don't know if he wrote the core) | 16:29 | ||
| Geth | MoarVM/main: 204a73098b | MasterDuke17++ (committed using GitHub Web editor) | 2 files Bump rapidhash to v3 |
||
| Voldenet | I would bet that the lib gets special treatment, it just feels funny to me that rapidhash presented benchmarks that show it as slower | 16:35 | |
| [Coke] | GAH | 16:36 | |
| Voldenet | while smhasher benchmarks show rapidhash as slightly faster | ||
| [Coke] | Folks, are we really pushing a change to moarvm with a week to go AFTER I've started the blin run? | ||
| Voldenet | and on amd ryzen cpus no less | ||
| [Coke] | .seen masterduke | 16:37 | |
| tellable6 | [Coke], I saw masterduke 2025-09-18T21:16:20Z in #raku-dev: <MasterDuke> dinner & | ||
| [Coke] kills the blin run | |||
| [Coke] shuts down the VM so he's not paying for it. :| | 16:38 | ||
| Voldenet | otoh on intel CPUs rapidhash has half the throughput than xxh3 for some reason | 16:39 | |
| Geth | MoarVM/revert-1959-update_rapidhash_to_v3: 9d79b310f2 | (Elizabeth Mattijsen)++ (committed using GitHub Web editor) | 2 files Revert "Bump rapidhash to v3" This reverts commit 204a73098bc5d3e34db467c85cb2d6b3ac7991de. |
16:42 | |
| MoarVM: lizmat++ created pull request #1960: Revert "Bump rapidhash to v3" |
|||
| MoarVM/main: 70225c6b5e | (Elizabeth Mattijsen)++ (committed using GitHub Web editor) | 2 files Revert "Bump rapidhash to v3" (#1960) This reverts commit 204a73098bc5d3e34db467c85cb2d6b3ac7991de. |
|||
| lizmat | [Coke]: sorry, was a little over eager doing something :-( | 16:43 | |
| Geth | MoarVM/revert-1960-revert-1959-update_rapidhash_to_v3: a61094c65c | (Elizabeth Mattijsen)++ (committed using GitHub Web editor) | 2 files Revert "Revert "Bump rapidhash to v3" (#1960)" This reverts commit 70225c6b5efd10aa2c930982f2cfebf4e7073ba7. |
||
| MoarVM: lizmat++ created pull request #1961: Bump rapidhash to v3 |
16:44 | ||
| lizmat | [Coke]: sorry for the noise :-( | ||
| [Coke] | no worries! | 16:48 | |
| Geth | MoarVM: MasterDuke17++ created pull request #1962: Bump libtommath to 1.3.0 |
17:56 | |
| japhb | [Coke]: Apologies for starting something; I was just studying and pontificating .... | 21:26 | |
| [Coke] | heee. no worries, everything is fine. | 21:56 | |