<html><head><meta http-equiv="Content-Type" content="text/html charset=windows-1252"></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space;"><br><div><div>On Aug 19, 2013, at 2:23 PM, Redmond, Paul <<a href="mailto:paul.redmond@intel.com">paul.redmond@intel.com</a>> wrote:</div><br class="Apple-interchange-newline"><blockquote type="cite"><div style="font-size: 12px; font-style: normal; font-variant: normal; font-weight: normal; letter-spacing: normal; line-height: normal; orphans: auto; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: auto; word-spacing: 0px; -webkit-text-stroke-width: 0px;"><br><br>On 2013-08-19 3:04 PM, "Nadav Rotem" <<a href="mailto:nrotem@apple.com">nrotem@apple.com</a>> wrote:<br><br><blockquote type="cite"><br>On Aug 19, 2013, at 11:30 AM, Redmond, Paul <<a href="mailto:paul.redmond@intel.com">paul.redmond@intel.com</a>><br>wrote:<br><br><blockquote type="cite">Hi Nadav,<br><br>On 2013-08-19 1:00 PM, "Nadav Rotem" <<a href="mailto:nrotem@apple.com">nrotem@apple.com</a>> wrote:<br><br><blockquote type="cite">Hi Paul,<span class="Apple-converted-space"> </span><br><br>This patch looks good, but I am a little worried about denormals. With<br>this patch we will execute vector operations on garbage at the fourth<br>vector element. One possible solution would be to mask out the last<br>element. Does that sound right to you ?<br></blockquote><br>That's a good question. I don't have a good answer right now. What about<br>handling the floating point operations in WidenVecRes_BinaryCanTrap for<br>now? (only the integral types are widened)<br><br></blockquote><br>I am okay with this solution.<br></blockquote><br>I've committed this in r188699.<br><br>However, I just looked at this code again and it seems the denormal<br>problem existed before (or perhaps isn't a problem?). For example,<br>consider a v3f32 add on x86: In WidenVecRes_BinaryCanTrap WidenVT will be<br>v4f32 and operation will be widened (since canOpTrap() is false) There<br>seems to be an assumption that unused elements are already masked out.<br><br></div></blockquote><div><br></div><div>Hi Paul, </div><div><br></div><div>I looked at the code and I agree that the problem existed before this patch and I think its a bug :) I suspect that we don’t handle denormals because no one needed to optimize non-power-of-two vector code before. </div><div><br></div><div>Thanks,</div><div>Nadav </div><br><blockquote type="cite"><div style="font-size: 12px; font-style: normal; font-variant: normal; font-weight: normal; letter-spacing: normal; line-height: normal; orphans: auto; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: auto; word-spacing: 0px; -webkit-text-stroke-width: 0px;">Thoughts?<br><br>paul<br><br><blockquote type="cite"><br>Thanks,<br>Nadav</blockquote></div></blockquote></div><br></body></html>