<html>
<head>
<meta content="text/html; charset=utf-8" http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
<br>
<br>
<div class="moz-cite-prefix">On 18/09/15 20:00, Jason Ekstrand
wrote:<br>
</div>
<blockquote
cite="mid:CAOFGe96_6xBw6JossiWjtjG1FYgz+0Se1k2upnkLY9HKLop0kQ@mail.gmail.com"
type="cite">
<p dir="ltr"><br>
On Sep 18, 2015 10:09 AM, "Alejandro Piñeiro" <<a
moz-do-not-send="true" href="mailto:apinheiro@igalia.com"><a class="moz-txt-link-abbreviated" href="mailto:apinheiro@igalia.com">apinheiro@igalia.com</a></a>>
wrote:<br>
><br>
> SEL and MOV instructions, as long as they don't have source
modifiers, are<br>
> just copying bits around. So those kind of instruction
could be propagated<br>
> even if there are type mismatches. This is needed because
NIR generates<br>
> integer SEL and MOV instructions whenever it doesn't know
what else to<br>
> generate.<br>
><br>
> This commit adds support for copy propagation using current
instruction<br>
> as reference.<br>
><br>
> v2: include check for saturate, as Jason Ekstrand suggested<br>
> v3: check that the dst.type and the src type are the same,
in order to<br>
> solve (among others) the following deqp regression with
v2:<br>
>
dEQP-GLES3.functional.shaders.operator.unary_operator.minus.lowp_uint_vertex</p>
<p dir="ltr">Reviewed-by: Jason Ekstrand <<a
moz-do-not-send="true" href="mailto:jason.ekstrand@intel.com"><a class="moz-txt-link-abbreviated" href="mailto:jason.ekstrand@intel.com">jason.ekstrand@intel.com</a></a>></p>
<p dir="ltr">The other will take more thinking.</p>
</blockquote>
<br>
Taking into account that the second patch is more complex, and
probably a new update, I changed my mind, and I think that it would
be better to push them individually.<br>
<br>
I have just pushed this one. Thanks for the review and patience.<br>
<br>
BR<br>
<br>
<pre class="moz-signature" cols="72">--
Alejandro Piñeiro (<a class="moz-txt-link-abbreviated" href="mailto:apinheiro@igalia.com">apinheiro@igalia.com</a>)</pre>
</body>
</html>