12:56mannerov: orbea: it's simple, all the fix requested weren't merged
15:11orbea: mannerov: huh, i thought it was merged?
15:12orbea: mannerov: was there another fix beyond this? https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4980
20:21mannerov: orbea: yes
20:21mannerov: I had analysed ways to reduce memory usage a lot by disabling one of the three radeonsi memory shader cache which isn't useful enough
20:22mannerov: Marek proposed a merge request for that, but it was dropped
20:23mannerov: And I think there was a way to remove from ram the third cache (which has disk backing) or something like that, but maybe I'm mistaken
20:24mannerov: Also there was a third way which was to reduce nir and llvm duplication
20:24mannerov: but the work wasn't done for that
20:25mannerov: To No surprise dxvk doesn't suffer from that as it doesn't hit all these shader caches and nir duplication
20:25mannerov: wined3d doesn't suffer either as the shaders are compiled on demand, not when created
20:26mannerov: The thing is the game creates a huge amount of shaders
20:28mannerov: orbea: check the bug report, isn't there a way to disable some of the caches with an env var or something ?
21:43orbea: mannerov: there was: NINE_SHADER=no_nir_ps,no_nir_vs
21:43orbea: is that still supported?
21:43orbea: i dont see anything else in https://github.com/iXit/Mesa-3D/issues/349
21:45orbea: to be clear I am trying this on a second computer now which has more ram and faster hardware than the first one.
21:45orbea: and it seems to eat more ram on it than the first...
21:51iive: mannerov, the merge request been dropped, does it mean the patch is rejected or just delayed?
21:54orbea: iive: where is it dropped? https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4980
21:54orbea: if I understand right, it fixed part of the problem, but apparently not enough of it?
22:33mannerov: iive: rejected
22:33mannerov: not enough gains
22:33mannerov: there was regression on some titles
22:33mannerov: the patches removed that cache entirely
22:34mannerov: instead it should have disabled it for 32 bits I think
22:34mannerov: but as only nine suffers from that, not GL... well
22:34mannerov: If GL games had that too it would have been different
22:36orbea: mannerov: so it was reverted?
22:38orbea: no, its still the most recent commit in the file https://gitlab.freedesktop.org/mesa/mesa/-/commits/master/src/gallium/auxiliary/util/u_live_shader_cache.c
22:59mannerov: orbea: the corresponding merge request by mareko was dropped