<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
<div class="moz-cite-prefix">On 6/7/13 4:39 PM, Eli Friedman wrote:<br>
</div>
<blockquote
cite="mid:CAJdarcGpDp-1QTwvhk+KASo_bAj4Jg=vnjkqxUGRke1P7uaSwQ@mail.gmail.com"
type="cite">
<div dir="ltr">On Fri, Jun 7, 2013 at 3:45 PM, Shuxin Yang <span
dir="ltr"><<a moz-do-not-send="true"
href="mailto:shuxin.llvm@gmail.com" target="_blank">shuxin.llvm@gmail.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">Author:
shuxin_yang<br>
Date: Fri Jun 7 17:45:21 2013<br>
New Revision: 183584<br>
<br>
URL: <a moz-do-not-send="true"
href="http://llvm.org/viewvc/llvm-project?rev=183584&view=rev"
target="_blank">http://llvm.org/viewvc/llvm-project?rev=183584&view=rev</a><br>
Log:<br>
Fix an assertion in MemCpyOpt pass.<br>
<br>
The MemCpyOpt pass is capable of optimizing:<br>
callee(&S); copy N bytes from S to D.<br>
into:<br>
callee(&D);<br>
subject to some legality constraints.<br>
<br>
Assertion is triggered when the compiler tries to
evalute "sizeof(typeof(D))",<br>
while D is an opaque-typed, 'sret' formal argument of
function being compiled.<br>
i.e. the signature of the func being compiled is something
like this:<br>
T caller(...,%opaque* noalias nocapture sret %D, ...)<br>
<br>
The fix is that when come across such situation, instead
of calling some<br>
utility functions to get the size of D's type (which will
crash), we simply<br>
assume D has at least N bytes as implified by the
copy-instruction.<br>
<br>
rdar://14073661<br>
<br>
Modified:<br>
llvm/trunk/lib/Transforms/Scalar/MemCpyOptimizer.cpp<br>
llvm/trunk/test/Transforms/MemCpyOpt/memcpy.ll<br>
<br>
Modified:
llvm/trunk/lib/Transforms/Scalar/MemCpyOptimizer.cpp<br>
URL: <a moz-do-not-send="true"
href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/MemCpyOptimizer.cpp?rev=183584&r1=183583&r2=183584&view=diff"
target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/MemCpyOptimizer.cpp?rev=183584&r1=183583&r2=183584&view=diff</a><br>
==============================================================================<br>
--- llvm/trunk/lib/Transforms/Scalar/MemCpyOptimizer.cpp
(original)<br>
+++ llvm/trunk/lib/Transforms/Scalar/MemCpyOptimizer.cpp
Fri Jun 7 17:45:21 2013<br>
@@ -626,8 +626,10 @@ bool
MemCpyOpt::performCallSlotOptzn(Ins<br>
return false;<br>
<br>
Type *StructTy =
cast<PointerType>(A->getType())->getElementType();<br>
- uint64_t destSize =
TD->getTypeAllocSize(StructTy);<br>
-<br>
+ // If StructTy is an opaque type, it should have at
least <cpyLen> bytes,<br>
+ // as implified by the copy-instruction.<br>
+ uint64_t destSize = StructTy->isSized() ?<br>
+ TD->getTypeAllocSize(StructTy)
: cpyLen;<br>
if (destSize < srcSize)<br>
return false;<br>
</blockquote>
<div><br>
</div>
<div>Why is it safe to assume StructTy is at least cpyLen
bytes? <br>
</div>
</div>
</div>
</div>
</blockquote>
If not, why is it possible copying cpyLen byte from (part of) src to
Dest? <br>
<br>
<blockquote
cite="mid:CAJdarcGpDp-1QTwvhk+KASo_bAj4Jg=vnjkqxUGRke1P7uaSwQ@mail.gmail.com"
type="cite">
<div dir="ltr">
<div class="gmail_extra">
<div class="gmail_quote">
<div>And if it is safe, why is this check necessary at all?</div>
<div><br>
</div>
</div>
</div>
</div>
</blockquote>
I guess you confuse cpyLen with SrcLen. <br>
I asked the same question when I read the code. After the 2nd
thought, I think it is necessary. <br>
You never know how the callee access the Src, it could access the
last byte (the SrcSize-th byte) of the object <br>
corresponding to the Src, and cpyLen < SrcSize. <br>
<br>
<br>
<br>
<br>
</body>
</html>