| <<<Back 1 day (to 2016/05/26) | 20160527 |
tor8 | Robin_Watts: pam fix LGTM. | 08:56.41 |
| Robin_Watts: usage_track, track_usage ... very mixed naming order in that commit, but given the lack of a fz_ prefix I take it that's only temporary code to measure once? | 08:57.56 |
Robin_Watts | tor8: It's intended as a standalone mechanism like memento. | 10:33.33 |
| it could be used in other projects. | 10:33.42 |
| Couldn't think of a snappy name for it though. | 10:33.50 |
tor8 | you could just grep for the plotter names in a regular profiling run, though? | 10:34.33 |
Robin_Watts | tor8: Not as it currently is, because the profiler can't tell between the different static inline instantiations. | 10:36.51 |
| I have a new commit though... | 10:36.58 |
| Up now for your consideration. Not necessarily finished, cos I need to test it, but it'll show what I'm thinking. | 10:37.30 |
tor8 | will look after lunch | 10:37.47 |
Robin_Watts | And the point of the usage stuff is so I can enable it in a cluster build and then grep the logs. | 10:37.54 |
tor8 | Robin_Watts: fz_paint_affine_near etc are starting to look like they should just use a lookup table of function pointers | 12:29.39 |
| and maybe a program to generate the code; in which case we can do optimized versions directly and not mess with the static inline specialization tricks. | 12:30.37 |
Robin_Watts | tor8: I'm not convinced that a program to generate the code would be any more maintainable. | 12:40.49 |
| And a table of function pointers is tricky if you want to start eliding some in some cases - especially when you start having different special cases available. | 12:41.40 |
tor8 | with lots of manually maintained almost duplicates, there's a significant risk of typos | 12:41.45 |
Robin_Watts | Yes, but every one of these is tested by something in our cluster I believe. | 12:42.08 |
tor8 | we can always have a 'generic' function to fill the function pointer table with | 12:42.34 |
Robin_Watts | SO works by using such a table. | 12:42.59 |
| but plotters are smaller things in the SO world. | 12:43.07 |
tor8 | SO only has one output format right? | 12:43.24 |
Robin_Watts | Noooo. | 12:43.32 |
| Many. | 12:43.35 |
tor8 | oh, I was under the impression you only did 16-bit color | 12:43.46 |
Robin_Watts | g8, r5g6b5, b5g6r5, xr5g5b5, some 444 stuff etc. | 12:44.11 |
Robin_Watts | runs to lunch. | 12:44.24 |
tor8 | ah, right. just no >16 bit :) | 12:44.26 |
Robin_Watts | yeah. | 12:44.31 |
| Ok, tor8:, new commits online. | 14:41.32 |
| cluster shows no diffs. | 14:41.45 |
| I'm open to ideas for improving the usage tracking stuff. | 14:42.38 |
| Do you prefer USAGE_TRACK or TRACK_USAGE ? | 14:43.10 |
tor8 | Robin_Watts: I prefer TRACK_USAGE by far | 15:30.24 |
Robin_Watts | Ok, I can do a global rename. | 15:30.40 |
| How do you feel about the other 'fix' commits there? | 15:31.19 |
tor8 | they look fine | 15:32.20 |
| dereferencing function pointers like (*fn)(dp, da, sp, sa, mp, n, w) hasn't been necessary since I started programming in C :) | 15:33.00 |
Robin_Watts | It's a personal taste thing. | 15:33.55 |
| I'll change if it really offends you. | 15:34.05 |
| Do we believe that compilers are smart enough to spot that for (i=3; i < x; i++) blah is just blah if x== 4 when static inlining? | 15:35.25 |
| Was the "move halftoning into render threads" one of the "fix" ones that you thought was fine ? | 15:36.11 |
tor8 | Robin_Watts: depending on optimization settings, it *ought* to be able to realize the loop can be unrolled | 16:12.35 |
| yes, the halftoning to render threads LGTM | 16:13.09 |
| I don't really care re: the function pointer thing, just thought it odd that you use such archaic syntax :) | 16:13.33 |
Robin_Watts | Thanks. | 16:14.15 |
| I've had problems treating functions and function pointers interchangably in the (recent) past. | 16:15.00 |
| Hence I always mentally use them as pointers to avoid pratfalls. | 16:15.18 |
| tor8: OK, updated commits with the renamed usage stuff. Are you happier with that? | 16:19.33 |
| Forward 1 day (to 2016/05/28)>>> | |