11:15RSpliet: TIL NVIDIA apparently supports inline assembly in OpenCL kernels
16:19jamm: hakzsam, pmoreau: the rd barrier works! no more artifacts, yay :)
16:19jamm: thanks you guys! now i'll move on to fixing the other shaders (and will resend modified patch email to you, hakzsam)
16:19jamm: one question, for this snippet
16:19jamm: sched (st 0x1) (st 0x1) (st 0x1)
16:19jamm: fmul ftz $r3 $r3 $r7
16:19jamm: fmul ftz $r2 $r2 $r6
16:19jamm: fmul ftz $r1 $r1 $r5
16:21jamm: none of the registers are common across these instructions, and no dependency on the $r1-$r3 either, so is the stall count of 0x1 justified here, like in the case of mov ?
19:13mangix: RSpliet: check the hashcat source code. full of inline assembly
19:14mangix: mainly because the compiler is sub-optimal at times
19:22RSpliet: mangix: I see. inline ASM is not a part of the OpenCL standard though - for the obvious reason that there's no standard assembly language that can cover the wide variety of devices supported by OpenCL
19:22RSpliet: nonetheless it's cool to have this feature for research and REing purposes
19:48mangix: hashcat #ifdefs it away.
19:48mangix: i think LLVM supports inline assembly with OpenCL, not too sure
19:57tstellar: You can do OpenCL inline assembly with the AMDGPU backend.
21:01TheDrode: Good afternoon!
21:01TheDrode: i have a desktop PC with both a GTX950 and a gs8400 (used just for two extra monitors)
21:08TheDrode: sorry, looking for help getting my PC working with Nouveau. I have a GTX950 and a GS8400
21:08TheDrode: and no video in neither of them
21:25leberus: skeggsb: hi! I fixed the warnings in v8 ;)
21:28skeggsb: leberus: great! thanks, i'll take another look at them shortly
21:31karolherbst: skeggsb: by the way, mind taking a look over my PMU counter patches inside my github branch (pmu_counters_v2)? I didn't really changed much, but also wanted to have your review first before sending another version
21:31skeggsb: karolherbst: i'll add that to the "shortly" pile too
22:34RSpliet: tstellar: which assembly language? Presumably not PTX :-P
22:36karolherbst: RSpliet: I am sure it is this llvm assembly thing :p
22:37karolherbst: ohhh, it seems like some native AMD assembly indeed
22:38karolherbst: RSpliet: https://github.com/llvm-mirror/llvm/commit/a787066317cad918dcfc259c843ae93ff2734229
22:52tstellar: RSpliet: gcn assembly
23:39hakzsam: jamm: yes, it is
23:48pmoreau: jamm: As there is no dependency, the GPU can start running the next fmul as soon as possible. Therefore you want to stall as little as possible.