<div dir="ltr"><div><div>Ilia, thanks a lot for the info.<br><br></div>So basically if I submit to GPU's command stream:<br></div><div>perform occlusion query,<br></div><div>write X to Y.<br></div><div>I know that query is completed when after reading Y address I get X.<br></div><div><div><br></div>Regards,<br></div>Vyacheslav<br></div><div class="gmail_extra"><br><div class="gmail_quote">On Mon, Jul 6, 2015 at 9:13 PM, Ilia Mirkin <span dir="ltr"><<a href="mailto:imirkin@alum.mit.edu" target="_blank">imirkin@alum.mit.edu</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">I'm only really familiar with nouveau, but I think all GPU hardware<br>
works in roughly the same way. Basically you have some way of<br>
inserting "write X to address Y" into the command stream (aka a<br>
"fence"), after which you insert "write X+1 to address Y" and so on.<br>
If you want the CPU to wait on a given fence, you just do "while<br>
(*address < x);". If you have multiple GPU processing queues, you can<br>
usually also insert a "stall this queue until the value at address Y<br>
is at least X" command into the command stream.<br>
<br>
DRM uses implicit fences, so it knows which BOs are used for<br>
particular commands. So the flow goes something like "submit bunch of<br>
commands; submit fence write and attach that fence id to the BOs in<br>
the previous bunch of comands". Then to wait for a bo to become ready,<br>
you just wait until the GPU writes the appropriate number to memory<br>
address Y (from above).<br>
<br>
The mesa drivers can sometimes use clever tricks that avoid this<br>
sync'ing because it knows exactly how it emits the commands and<br>
perhaps it waits on something related earlier whereby it knows the<br>
other thing will be ready. No idea if that's the case here.<br>
<br>
Hope this helps,<br>
<br>
-ilia<br>
<div class="HOEnZb"><div class="h5"><br>
<br>
On Mon, Jul 6, 2015 at 1:05 PM, Vyacheslav Gonakhchyan<br>
<<a href="mailto:ytriffy@gmail.com">ytriffy@gmail.com</a>> wrote:<br>
> Ilia, thanks for the gallium link.<br>
> Do you know any links to high level info with broad strokes about how this<br>
> sync works? Frankly I do not know driver terminology and wanted to know more<br>
> about how this sync is performed for my research. I'm using mesa as a<br>
> reference because it has open implementation code. Occlusion query<br>
> functionality probably waits for z-buffer to become ready. Problem is that<br>
> usual synchronization techniques do not apply here. I'm thinking that driver<br>
> code gets notifications about state change. I want to know what kind of<br>
> notifications are available? Can query be performed in parallel with another<br>
> frame being processed or does it need complete GPU pipeline flush?<br>
><br>
> Thanks,<br>
> Vyacheslav<br>
><br>
> On Mon, Jul 6, 2015 at 8:32 PM, Ilia Mirkin <<a href="mailto:imirkin@alum.mit.edu">imirkin@alum.mit.edu</a>> wrote:<br>
>><br>
>> On Mon, Jul 6, 2015 at 11:29 AM, Vyacheslav Gonakhchyan<br>
>> <<a href="mailto:ytriffy@gmail.com">ytriffy@gmail.com</a>> wrote:<br>
>> > Hi, everyone.<br>
>> ><br>
>> > Trying to understand method radeonQueryGetResult (more broadly GPU-CPU<br>
>> > sync).<br>
>> ><br>
>> > static void radeonQueryGetResult(struct gl_context *ctx, struct<br>
>> > gl_query_object *q)<br>
>> > {<br>
>> > struct radeon_query_object *query = (struct radeon_query_object *)q;<br>
>> > uint32_t *result;<br>
>> > int i;<br>
>> ><br>
>> > radeon_print(RADEON_STATE, RADEON_VERBOSE,<br>
>> > "%s: query id %d, result %d\n",<br>
>> > __func__, query->Base.Id, (int) query->Base.Result);<br>
>> ><br>
>> > radeon_bo_map(query->bo, GL_FALSE);<br>
>> > result = query->bo->ptr;<br>
>> ><br>
>> > query->Base.Result = 0;<br>
>> > for (i = 0; i < query->curr_offset/sizeof(uint32_t); ++i) {<br>
>> > query->Base.Result += LE32_TO_CPU(result[i]);<br>
>> > radeon_print(RADEON_STATE, RADEON_TRACE, "result[%d] = %d\n", i,<br>
>> > LE32_TO_CPU(result[i]));<br>
>> > }<br>
>> ><br>
>> > radeon_bo_unmap(query->bo);<br>
>> > }<br>
>> ><br>
>> > I don't know which part is responsible for blocking behavior (waiting<br>
>> > for<br>
>> > response from GPU). I suspect that radeon_bo_map does this magic.<br>
>> > Can someone point in the right direction?<br>
>><br>
>> The radeon_bo_map defined in<br>
>> src/gallium/winsys/radeon/drm/radeon_drm_bo.c indeed has this magic.<br>
>> However the code in src/mesa/drivers/dri/radeon/radeon_queryobj.c<br>
>> references the radeon_bo_map in libdrm, which does not appear to wait.<br>
>><br>
>> FWIW for nouveau, nouveau_bo_map will also implicitly do a<br>
>> nouveau_bo_wait, but that does not appear to be the case for radeon.<br>
>><br>
>> Cheers,<br>
>><br>
>> -ilia<br>
><br>
><br>
</div></div></blockquote></div><br></div>