<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 Mar 12, 2013, at 9:25 AM, Duncan Sands <<a href="mailto:baldrick@free.fr">baldrick@free.fr</a>> wrote:</div>
<br class="Apple-interchange-newline">
<blockquote type="cite">
<div style="letter-spacing: normal; orphans: auto; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: auto; word-spacing: 0px; -webkit-text-size-adjust: auto; -webkit-text-stroke-width: 0px;">
Hi Richard,<br>
<br>
On 12/03/13 16:00, Relph, Richard wrote:<br>
<blockquote type="cite">DAGTypeLegalizer::SplitRes_SELECT() currently does an assert if the condition is not of type MVT::i1. This patch uses the existing type of the condition to generate the type for the split conditions instead of a fixed i1.<br>
This is an issue for us when trying to legalize a 8x or 16x vector to a 4x type… it isn't the case that we always have an MVT::i1 condition coming in.<br>
</blockquote>
<br>
if the select condition is a vector then you should be using a vselect node<br>
rather than a select node. So I think this patch is wrong and the real problem<br>
is that the wrong select node type is being generated somewhere.<br>
</div>
</blockquote>
<div><br>
</div>
<div>That makes sense… except you'll note that the existing code seems to expect a vector condition, which (as I understand it) is the difference between SELECT and VSELECT... why would it do that for a SELECT?</div>
<div>Can end up in this code for VSELECT, despite the name of the function? If not, what's the point of splitting a scalar condition, or testing it's type, or testing that it is a vector?</div>
<div><br>
</div>
<div>There was a wrong select node type being generated in visitSIGN_EXTEND(), which was generating a SELECT with a vector (see my separate email on that.) My first attempt at fixing that was to generate a VSELECT if the sign extend was against a vector. With
that fix, I had this problem, so it would seem to indicate that the source was 'elsewhere'. However, when I implemented Nadav's solution to the visitSIGN_EXTEND problem (see that same separate email), this problem seems to have disappeared. So color me confused
and I'll drop this commit request… though I would appreciate education on the questions above.</div>
<div><br>
</div>
<div>Thanks,</div>
<div>Richard</div>
<div><br>
</div>
<div><br>
</div>
<div><br>
</div>
<blockquote type="cite">
<div style="letter-spacing: normal; orphans: auto; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: auto; word-spacing: 0px; -webkit-text-size-adjust: auto; -webkit-text-stroke-width: 0px;">
<br>
Ciao, Duncan.<br>
<br>
<blockquote type="cite"><br>
Index: lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp<br>
===================================================================<br>
--- lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp<span class="Apple-tab-span" style="white-space: pre;">
</span>(revision 176837)<br>
+++ lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp<span class="Apple-tab-span" style="white-space: pre;">
</span>(working copy)<br>
@@ -490,10 +490,10 @@<br>
SDValue Cond = N->getOperand(0);<br>
CL = CH = Cond;<br>
if (Cond.getValueType().isVector()) {<br>
- assert(Cond.getValueType().getVectorElementType() == MVT::i1 &&<br>
- "Condition legalized before result?");<br>
unsigned NumElements = Cond.getValueType().getVectorNumElements();<br>
- EVT VCondTy = EVT::getVectorVT(*DAG.getContext(), MVT::i1, NumElements / 2);<br>
+ EVT VCondTy = EVT::getVectorVT(*DAG.getContext(),<br>
+ Cond.getValueType().getVectorElementType(),<br>
+ NumElements / 2);<br>
CL = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VCondTy, Cond,<br>
DAG.getIntPtrConstant(0));<br>
CH = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VCondTy, Cond,<br>
<br>
<br>
<br>
_______________________________________________<br>
llvm-commits mailing list<br>
<a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
<a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
<br>
</blockquote>
<br>
_______________________________________________<br>
llvm-commits mailing list<br>
<a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
<a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a></div>
</blockquote>
</div>
<br>
</body>
</html>