00:11 fdobridge: <j​ekstrand> The caselist sync thing is because of the `VK_ERROR_INITIALIZATION_FAILED`, I'm pretty sure. And that's likely because the kernel has just stopped accepting new connections or something like that. Better than the whole machine keeling over but still probably requires a rebot.
00:26 fdobridge: <j​ekstrand> rmckeever ^^
00:41 rmckeever: Okay, that makes sense.
12:50 fdobridge: <p​homes> I do not see the rebase of nvk/main in the nouveau/mesa repo. Did you forget to push it there?
13:48 fdobridge: <g​ouz> I am trying to support resuming with counter buffers for transform feedback.
13:48 fdobridge: <g​ouz> I followed something like this but i cannot yet make it work:
13:48 fdobridge: <g​ouz> - begin transform feedback without a counter buffer i set "SET_STREAM_OUT_BUFFER_LOAD_WRITE_POINTER" to zero
13:48 fdobridge: <g​ouz> - at begin transform feedback with a counter buffer i run an mme that loads from the buffer to the SET_STREAM_OUT_BUFFER_LOAD_WRITE_POINTER
13:50 fdobridge: <g​ouz> - at end transform feedback with a counter buffer i suspect we need to somehow store "SET_REPORT_SEMAPHORE_D_REPORT_STREAMING_BYTE_COUNT" to the counter buffer
13:52 fdobridge: <g​ouz> Can i store this value directly to the counter buffer? or should i do this indirectly (e.g. a query / temp cmd buffer memory)
13:52 fdobridge: <g​ouz> I am trying to support resuming with counter buffers for transform feedback.
13:52 fdobridge: <g​ouz> I followed something like this but i cannot yet make it work:
13:52 fdobridge: <g​ouz> - at begin transform feedback without a counter buffer i set "SET_STREAM_OUT_BUFFER_LOAD_WRITE_POINTER" to zero (edited)
16:25 fdobridge: <j​ekstrand> Drp. Pushed to the wrong remote. Fixed now.
16:27 fdobridge: <j​ekstrand> If you need a temporary, use a scratch state, not memory. Take a look at nvk_mme.h and specifically nvk_mme_scratch and look at what we do for CS invocations.
16:27 fdobridge: <j​ekstrand> But I don't remember if you need actual scratch or can write straight to memory. It's been a while since I looked at the XFB API.
16:29 fdobridge: <g​ouz> I tried to write streaming byte count directly to the buffer but without correct results,
16:29 fdobridge: <g​ouz> thanks for the feedback, will search a bit more
18:45 fdobridge: <p​homes> thanks. I will rebase my MR's. Btw the old nvk/main had 2fbad0728203011023fc750fb4c8879595adf62e and the new one does not. Not sure if that was intentional
20:25 fdobridge: <g​ouz> I've done it! The only thing missing was a INVALIDATE_SHADER_CACHES_NO_WFI after the write to the counter buffer.
20:25 fdobridge: <g​ouz> We do not need temporary storage at all.
21:43 fdobridge: <g​ouz> At last all transform feedback tests now pass
21:44 fdobridge: <g​ouz> (excluding the not supported ones😅)