| <<<Back 1 day (to 2018/06/06) | 20180607 |
moolc | sebras: few minutes ago i've glimpsed into the future.. it's chinese.. bloody football | 08:52.51 |
sebras | moolc: :) | 09:15.22 |
moolc | sebras: and i'm only half joking.. buy groceries became hell.. dozens of chinese with shopping carts trurn 1min procedure into a 20min torture | 09:16.39 |
| s;buy;buying | 09:16.45 |
| tor8: any thoughts about build=native thing? | 11:42.59 |
tor8 | what's x64 specific about -march=native? | 11:43.44 |
| I know we have an intel specific thing with the SSE math option | 11:44.01 |
moolc | tor8: yep that's one of them | 13:23.00 |
| tor8: making it impossible to make build=native on ARM | 13:23.11 |
| correction, the only one (after a quick glance) | 13:24.14 |
tor8 | moolc: something like the top commit on tor/master then? | 13:26.17 |
moolc | tor8: that certainly would work... a bit cannon shooting sparrows like but an improvement nevertheless (besides fpmath sse only makes a difference for 32bit x86 svr abi) | 13:28.18 |
tor8 | moolc: the -mfpmath=sse was added before 64-bit intel machines were the common install | 13:29.12 |
moolc | tor8: sure. apparently nobody builds much on (non android)arm/ppc/whatever | 13:30.35 |
tor8 | most probably aren't even aware the build=native option was there | 13:31.03 |
| is there ... freudian slip? ;) | 13:31.27 |
moolc | haha :) | 13:31.45 |
tor8 | Robin_Watts: two approaches for handling VS2015 limitations and fontdumping, on tor/bin2coff and tor/winres | 14:23.54 |
| both work, but I think we need to consider which poison is the most attractive | 14:24.22 |
| I'm leaning towards bin2coff; it has fewer surprises and gotchas | 14:24.41 |
| but the resource file based one lets us get rid of the extra make generate step altogether | 14:25.17 |
| with the caveat that we need to maintain the windows font file list separately, with all the TOFU ifdefs, etc. | 14:25.48 |
Robin_Watts | tor8: I prefer bin2coff, but I'd prefer it more if we could avoid the script. | 14:26.38 |
tor8 | Robin_Watts: right now it's an nmake makefile to invoke the bin2coff | 14:27.01 |
Robin_Watts | tor8: Right. | 14:27.10 |
| Let me pull it and have a play. | 14:27.24 |
tor8 | maybe there's a way to tell VS to do a custom command line to go directly from .ttf to .obj? | 14:27.28 |
| we could also get rid of the libresources.vcproj if we put the additional dependencies on the libmupdf.vcproj | 14:28.08 |
| but I wanted to keep the patch small | 14:28.12 |
| and that would get rid of the pointless dummy.c I had to add | 14:29.13 |
| if it helps, I don't mind checking in a win32 executable 'scripts/bin2coff.exe' | 14:30.26 |
Robin_Watts | http://git.ghostscript.com/?p=user/tor/mupdf.git;a=blame;f=source/tools/pdfshow.c;h=68feb9f871db27b3e1a43aa176ebb708fa5f6f32;hb=refs/heads/bin2coff | 14:33.25 |
| Line 147ish... | 14:33.31 |
| shouldn't the showstream be outside that if (tight) thing ? | 14:33.46 |
| Oh, I see. | 14:35.38 |
tor8 | yeah, 'tight' should really be 'grep' | 14:53.01 |
Robin_Watts | I almost have a version working just using msvc stuff. | 15:48.32 |
| tor8: ^ | 15:48.39 |
| So you're choosing to put "__binary_resources_fonts" in front of each name? | 15:48.56 |
| Bah. symbols with '-' in them are bad. | 16:27.49 |
| tor8: OK, so updated robin/bin2coff for your consideration. | 17:32.33 |
| sebras; You awake? | 18:41.26 |
tor8 | Robin_Watts: yeah, just for symmetry with objcopy which puts "_binary_" in front of each name | 21:28.26 |
Robin_Watts | tor8: So look over my commits when you get a mo. I think it works out OK. | 22:49.48 |
| There is a bit of nastiness in the way we do 64/32bit, but it's not bad. | 22:50.06 |
| Forward 1 day (to 2018/06/08)>>> | |