[Mesa-dev] [PATCH 5/5] docs: describe the INTEL_* envvars that do exist

[Chris Forbes]

I've pushed these now, including dropping the old names for those debug bits.

I've left the stats stuff as-is for now, but suggested more useful
options in the docs.

On Mon, Nov 25, 2013 at 4:44 AM, Kenneth Graunke <kenneth at whitecape.org> wrote:
> On 11/23/2013 09:13 PM, Chris Forbes wrote:
>> Signed-off-by: Chris Forbes <chrisf at ijw.co.nz>
>> ---
>>  docs/envvars.html | 32 ++++++++++++++++++++++++++++++++
>>  1 file changed, 32 insertions(+)
>>
>> diff --git a/docs/envvars.html b/docs/envvars.html
>> index 81e74e6..d831826 100644
>> --- a/docs/envvars.html
>> +++ b/docs/envvars.html
>> @@ -121,6 +121,38 @@ See the <a href="xlibdriver.html">Xlib software driver page</a> for details.
>>  <h2>i945/i965 driver environment variables (non-Gallium)</h2>
>>
>>  <ul>
>> +<li>INTEL_NO_HW - if set to 1, prevents batches from being submitted to the hardware.
>> +   This is useful for debugging hangs, etc.</li>
>> +<li>INTEL_DEBUG - a comma-separated list of named flags, which do various things:
>> +<ul>
>> +   <li>tex - emit messages about textures.</li>
>> +   <li>state - emit messages about state flag tracking</li>
>> +   <li>blit - emit messages about blit operations</li>
>> +   <li>miptree - emit messages about miptrees</li>
>> +   <li>fall/perf - emit messages about performance issues</li>
>
> Not sure if the old names are worth documenting.  I guess if people find
> old text on wikis or something that says INTEL_DEBUG=fall, they'll know
> what it means.
>
> We might want to actually just drop the 'fall' name at this
> point...people seem to have moved over completely.
>
>> +   <li>perfmon - emit messages about AMD_performance_monitor</li>
>> +   <li>bat - emit batch information</li>
>> +   <li>pix - emit messages about pixel operations</li>
>> +   <li>buf - emit messages about buffer objects</li>
>> +   <li>reg - emit messages about regions</li>
>> +   <li>fbo - emit messages about framebuffers</li>
>> +   <li>fs/wm - dump shader assembly for fragment shaders</li>
>
> I've pretty much universally moved over to INTEL_DEBUG=fs too, but I
> don't know about others.
>
>> +   <li>gs - dump shader assembly for geometry shaders</li>
>> +   <li>sync - emit messages about synchronization</li>
>> +   <li>prim - emit messages about drawing primitives</li>
>> +   <li>vert - emit messages about vertex assembly</li>
>> +   <li>dri - emit messages about the DRI interface</li>
>> +   <li>sf - emit messages about the strips & fans unit (for old gens, includes the SF program)</li>
>> +   <li>stats - ?</li>
>
> This enables statistics counters for the vertex fetcher (on all
> generations), and for other units on Gen4-5.  That said, the counters
> aren't exposed other than reading registers, and on Gen6+ you can't even
> use intel_reg_read due to hardware contexts.
>
> Frankly, it seems pretty useless, and I think we ought to delete it.
>
>> +   <li>urb - emit messages about URB setup</li>
>> +   <li>vs - dump shader assembly for vertex shaders</li>
>> +   <li>clip - emit messages about the clip unit (for old gens, includes the CLIP program)</li>
>> +   <li>aub - dump batches into an AUB trace for use with simulation tools</li>
>> +   <li>shader_time - record how much GPU time is spent in each shader</li>
>> +   <li>no16 - suppress generation of 16-wide fragment shaders. useful for debugging broken shaders</li>
>> +   <li>blorp - emit messages about the blorp operations (blits & clears)</li>
>> +   <li>nodualobj - suppress generation of dual-object geometry shader code</li>
>> +</ul>
>>  </ul>
>
> Thanks for doing this, Chris.  For the series:
> Reviewed-by: Kenneth Graunke <kenneth at whitecape.org>