<div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Fri, Oct 24, 2014 at 1:00 PM, Matt Turner <span dir="ltr"><<a href="mailto:mattst88@gmail.com" target="_blank">mattst88@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="HOEnZb"><div class="h5">On Fri, Oct 24, 2014 at 12:55 PM, Jason Ekstrand <<a href="mailto:jason@jlekstrand.net">jason@jlekstrand.net</a>> wrote:<br>
> On Oct 24, 2014 12:51 PM, "Matt Turner" <<a href="mailto:mattst88@gmail.com">mattst88@gmail.com</a>> wrote:<br>
>> On Fri, Oct 24, 2014 at 12:25 PM, Jason Ekstrand <<a href="mailto:jason@jlekstrand.net">jason@jlekstrand.net</a>><br>
>> wrote:<br>
>> > I thought this would be a clever way to make spilling less expensive.<br>
>> > However, it appears that the oword read/write messages we are using for<br>
>> > spilling ignore the execution size and assume SIMD16 whenever working<br>
>> > with<br>
>> > more than one register.<br>
>><br>
>> What do you mean by "assume SIMD16"?<br>
><br>
> From reading the PRM it looks like it ignores the exec size and treats it as<br>
> 16-wide. Also, this makes sense given my experiments.<br>
<br>
</div></div>That sentence seems to mean something to you that it doesn't to me.<br>
Doing the operation on 16 elements seems like the intention.<br>
<br>
Any idea what's actually wrong? Is it using bogus high-8 channel<br>
enable bits or something?<br></blockquote><div> </div></div> Yup. Which ends up with half your data is bogus.<br></div></div>