IRC Logs of #dri-devel on for 2016-10-28

Previous dayChoose dateNext day Show menu

00:51 #dri-devel: < tarceri> airlied: I'm adding shader_info to ac_shader_variant_info for lack of a better name are you ok with info? We would have things like
00:52 #dri-devel: < tarceri> I don't want to have to change it on review :P
00:57 #dri-devel: < tarceri> actually I'm not sure how well that will work. I'll keep the separate for now
01:55 #dri-devel: < tarceri> anholt: fairly deep into refactoring and I think I spotted the cause of the problem you were hitting. Patch sent :)
03:51 #dri-devel: < tarceri> robclark: freedreno seems to leak the nir_shader passed to it by st_program, or at least I can't find the code where it is freed
04:59 #dri-devel: < tarceri> airlied: same with radv meta when using nir_builder_init_simple_shader() I don't see nir_shader getting freed anywhere
05:33 #dri-devel: < tarceri> jekstrand: looks like it happens with intel/blorp too. For example we create nir_shader via brw_blorp_build_nir_shader() then it get stolen by a mem_ctx in brw_blorp_compile_nir_shader() it then gets cloned in brw_compile_fs() but that whole ctx never gets released
05:40 #dri-devel: < jekstrand> tarceri: Yeah, I've got a patch that fixes that :/
05:40 #dri-devel: < jekstrand> It's reviewed. I could push it
05:41 #dri-devel: < jekstrand> tarceri: Unless it would conflict with your refactor :)
05:41 #dri-devel: < jekstrand> tarceri:
05:42 #dri-devel: < tarceri> jekstrand: cool, thats what I was hoping it would do. Feel free to push. You can add my r-b too if you like
05:53 #dri-devel: < jekstrand> tarceri: Pushed.
08:21 #dri-devel: < jekstrand> tarceri: Thanks for all your hard refactoring work! I know it's been painful but I think the end result will be pretty fantastic.
08:41 #dri-devel: < tarceri> jekstrand: no problem. I can't say I'm not regretting opening this can of worms ... but I agree the results are looking good :)
08:42 #dri-devel: < jekstrand> tarceri: Well, Sometimes the code needs cleaning. It's never pleasent but sometimes it just needs to happen.
09:03 #dri-devel: < tarceri> jekstrand: yeah agreed I just didn't realise how much re-factoring the change would make possible. There is a whole other side still left where we can replace most of gl_linked_shader:
09:03 #dri-devel: < tarceri> well alot of gl_linker_shader maybe not most
09:07 #dri-devel: < jekstrand> yeah
09:07 #dri-devel: < jekstrand> it'll be nice
09:08 #dri-devel: < tarceri> jekstrand: by the way my thoughts on sending a part 1 was that Eric might want master working again
09:16 #dri-devel: < jekstrand> heh, sure
09:16 #dri-devel: < jekstrand> I'll have some anv cleanups once the refactors land but they're not pressing so I can wait.
09:17 #dri-devel: < jekstrand> I mostly realized after looking at your patches, that anv is being stupid about ralloc contexts
11:06 #dri-devel: < tarceri> anholt: does vc4 work out of the box on any distro at the moment or do I need to build a special kernel? I've got a pi2 that I've never used. I should probably test some of these refactoring changes
11:21 #dri-devel: < FireBurn|Work> tarceri:
11:26 #dri-devel: < padovan> danvet: stark3y: thanks for the review. v7 will have proper IGT tests.
12:00 #dri-devel: < robclark> tarceri, ir3_shader_destroy() should be the spot

Written by Christoph Brill © 2007-2016

Valid XHTML 1.0 Transitional

Available in Android Market