<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Thu, Jun 16, 2016 at 11:26 AM, Chad Versace <span dir="ltr"><<a href="mailto:chad.versace@intel.com" target="_blank">chad.versace@intel.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><span class="">On Sat 11 Jun 2016, Jason Ekstrand wrote:<br>
> This fixes 688 Vulkan CTS tests on Haswell.<br>
<br>
</span>Whoo! Bugfixes!<br>
<br>
The PRM states that, on HSW, you also need to program<br>
SAMPLER_BORDER_COLOR_STATE correctly for integer formats. I suspect that<br>
the CTS tests may be passing by luck.<br>
<br>
See this encouraging comment in brw_sampler_state.c:<br>
<br>
/* Haswell's integer border color support is completely insane:<br>
* SAMPLER_BORDER_COLOR_STATE is 20 DWords. The first four are<br>
* for float colors. The next 12 DWords are MBZ and only exist to<br>
* pad it out to a 64 byte cacheline boundary. DWords 16-19 then<br>
* contain integer colors; these are only used if SURFACE_STATE<br>
* has the "Integer Surface Format" bit set. Even then, the<br>
* arrangement of the RGBA data devolves into madness.<br>
*/<br></blockquote><div><br></div><div>Oh, I'm well aware of the insanity there! I reviewed those patches.<br></div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<br>
Anyway, this patch is<br>
Reviewed-by: Chad Versace <<a href="mailto:chad.versace@intel.com">chad.versace@intel.com</a>><br>
and please add a TODO somewhere for SAMPLER_BORDER_COLOR_STATE.<br>
</blockquote></div><br></div><div class="gmail_extra">I'm not quite sure where to put that TODO... Sadly, integer border color will probably never work properly on Haswell :-(<br></div><div class="gmail_extra">--Jason<br></div></div>