00:04 karolherbst: anholt: yeah soo.. I think that never hit any crashes with TGSI as you can't specify a lod and a comparator, so we were always capped at 4 sources :)
00:04 karolherbst: and then everything is fine
00:05 karolherbst: or something
00:06 anholt: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16156
00:29 karolherbst: anholt: I wouldn't be surprised if that even improves the overall shader-db stats
00:34 karolherbst: anholt: although I guess we could also check simply within nouveau if the lod is zero and set the bit, but... if that MR helps ntt as well, so be it
01:48 imirkin: 5 or 6 values into a quad is not going to end well
01:48 imirkin: quad = 4, so ... ;)
01:49 imirkin: the rast limit points thing is ES point rast vs GL point rast
01:49 imirkin: we could get the ES point rast by not clipping
01:49 imirkin: iirc that's the i965 way
01:49 imirkin: and instead letting scissors do the talking...
01:50 imirkin: there's a weird #ifdef which totally doesn't work in the nv50 or nvc0_state which messes with whether scissors are used
01:50 imirkin: i think its intent was something around that
01:50 imirkin: by you couldn't undef it by the time i got my hands on the code, so you know whatever intent behind it is _real_ old
18:09 anholt: imirkin: can I still keep the frontend levelZero thing, or do you want the proposed handleTXL change to replace it?
18:09 anholt: note that handleTXB only does CUBE_SHADOW, while for TXL we need 2D_ARRAY_SHADOW too. should I extend the txb case for 2d_array_shadow?