<marex>
austriancoder: btw I got as far as command stream execution on the MP2 it seems, but I never got interrupt on command stream completion it seems
<marex>
austriancoder: that made any attempt at running command stream time out, I think
<marex>
austriancoder: AXI fault interrupts did however get generated, so it isn't the interrupt itself, it might be something in the command stream that should raise the interrupt on completion that is missing , maybe ?
<austriancoder>
marex: ah yes .. I got that far, that I added two boards to my ci-farm - nothing more :/ want to finish some other work before looking into it.
<marex>
austriancoder: is my assumption correct, that there has to be something in the command stream which would trigger the interrupt (some interrupt?) at the end of execution of the command stream ?
<marex>
(is there any trivial command stream test I can use to investigate this further?)
<marex>
I also tried the etna-viv interposer on the ST blob, but the MP2 blob is very different from the previous blobs
<cmeissl[m]>
marex: yeah, I did some tests with 25.1.3 and NV12 where it worked just fine. Other YUV formats seem to still work in 25.2.2.
<marex>
cmeissl[m]: OK
<austriancoder>
marex: yes - in the cmd stream you do: CMD_LOAD_STATE(buffer, VIVS_GL_EVENT, VIVS_GL_EVENT_EVENT_ID(event) | VIVS_GL_EVENT_FROM_PE);
<marex>
austriancoder: and I do that, and the etnaviv kernel driver consider that command stream done successfully ?
<austriancoder>
marex: the kernel driver add this part on its own - the simplest cmd stream to send would be one nop
<marex>
austriancoder: is there a test program which would do this for me , or do I need to write one ?
<cmeissl[m]>
erle: I currently have no support for input in my compositor. I have to check if weston is included, and if not, it will take some time.
<marex>
austriancoder: I was hoping for an existing test tool, because if I write test tool, then I don't know if the test tool works ... and I don't know if the GPU works, so now I will have two unknown variables
<marex>
austriancoder: but if I do only ... etna_set_state(stream, VIVS_GL_FLUSH_CACHE, VIVS_GL_FLUSH_CACHE_PE2D); ... that should be enough ?
<austriancoder>
marex: yes .. maybe be safe and use VIVS_GL_FLUSH_CACHE_COLOR for the 3d core
<marex>
all right, I'll try that tomorrow, thanks
<austriancoder>
marex: maybe the hwdb has some information what we might be missing
<marex>
austriancoder: there is already an HWDB entry for this GPU it seems
<marex>
btw the GPU is misdetected by mesa as NPU, I had to patch that out
<austriancoder>
marex: had a quick look at the feature values and there is nothing that jumps out.
<marex>
austriancoder: I know, also considering that there should be another GC8000 that is also supported already