<html><head><meta http-equiv="Content-Type" content="text/html charset=iso-8859-1"></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; ">The "opt" run is a better isolation of this change. If we were curious, we could do an opt run with non-overloadable intrinsics to see if that regressed at all. It's clearly a massive win for overloadable intrinsics, which were O(n).<div><br><div><div>On Mar 2, 2013, at 2:48 AM, Renato Golin <<a href="mailto:renato.golin@linaro.org">renato.golin@linaro.org</a>> wrote:</div><br class="Apple-interchange-newline"><blockquote type="cite"><div dir="ltr">On 1 March 2013 19:41, Jean-Luc Duprat <span dir="ltr"><<a href="mailto:jduprat@apple.com" target="_blank">jduprat@apple.com</a>></span> wrote:<br><div class="gmail_extra"><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
After this change:<br>
$ time ./bin/llc -march=arm -mattr=+neon vsub_JLD2.ll -O0<br>
real 0m10.440s (-5%)<br>
$ time ./bin/llc -march=arm -mattr=+neon vsub_JLD2.ll -O2<br>
real 0m11.383s (-5%)<br></blockquote><div><br></div><div style="">Wow, that's a big difference from the original 10%, but the times are also different, so I wonder how much it is really the changes and how much was noise...</div>
<div style=""><br></div><div style="">--renato</div></div></div></div>
</blockquote></div><br></div></body></html>