07:01 karolherbst: glehmann: so.. I found a correctness issue on asahi with lower_fminmax_signed_zero. When the numbers are denorms, the feq returns true, but the integer max/min could screw things up and I can fix this with this: https://gitlab.freedesktop.org/karolherbst/mesa/-/commit/37d68f953d04d63bc396a52a956101f1866ee460 any better ideas?
07:01 karolherbst: ehh
07:01 karolherbst: it returns false
07:02 karolherbst: ehh no, true
07:02 karolherbst: Seeing issues like this in the CL CTS: ERROR: fmax: -8386560.000000 ulp error at {-0x1.ffcp-139 (0x800007ff), -0x1.fffffcp-127 (0x807fffff)}: *-0x1.ffcp-139 vs. -0x1.fffffcp-127 (0x807fffff) at index: 3763
07:20 karolherbst: probably better to fcanonicalize the result actually
07:55 glehmann: yeah I think either solution makes sense
07:55 glehmann: karolherbst: but the whole lowering is garbage anyway: https://gitlab.freedesktop.org/mesa/mesa/-/work_items/14937
08:12 karolherbst: glehmann: okay... well at least with my fix the CL CTS is happy enough about this