<div dir="ltr"><div>New patch looks good to me.</div><div><br></div>It sounds like we have two cases of size mismatch:<div>- The output operand lvalue is smaller than the constraint, meaning the store will write out of bounds. Your patch adds this.</div>
<div>- The output operand lvalue is bigger than the constraint, meaning the whole value won't be initialized. We currently warn here via validateConstraintModifier.</div><div><br></div><div>This code probably deserves some cleanup, but your patch is consistent with what we do for input operands, so let's go with that.</div>
</div><div class="gmail_extra"><br><br><div class="gmail_quote">On Wed, Aug 27, 2014 at 12:35 PM, Akira Hatanaka <span dir="ltr"><<a href="mailto:ahatanak@gmail.com" target="_blank">ahatanak@gmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr">The commit log in r166737 doesn't say much about why this is a warning instead of an error, but I know there are cases where warnings are needed. For example, clang has to issue warnings instead of errors for the inline-asm statements in the test case committed in r216260. If it's not desirable to change validateConstraintModifier, we can add a function which checks the output size that is similar to validateInputSize in r167717 (see attached patch), which was suggested in the post-commit review.<div>
<div><br></div><div><a href="http://lists.cs.uiuc.edu/pipermail/cfe-commits/Week-of-Mon-20121112/067945.html" target="_blank">http://lists.cs.uiuc.edu/pipermail/cfe-commits/Week-of-Mon-20121112/067945.html</a><br></div><div>
<br></div><div>
I am not sure whether we can use fixit in this case. Fixit hints should be used only if we know the user's intent and it's very clear that applying the fixit hint is the right thing to do. Changing the type of variable "r" to a 32-bit int will avoid crashing, but it doesn't look like that's what the user wants.</div>
<div><br></div></div><div><div class="h5"><div class="gmail_extra"><div class="gmail_quote">On Mon, Aug 25, 2014 at 6:20 PM, Reid Kleckner <span dir="ltr"><<a href="mailto:rnk@google.com" target="_blank">rnk@google.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr">Can you investigate why we are warning in the first place? I think we should either only warn or only error. Currently we have a warning with a fixit but we don't recover as though we had applied the fixit. If we did that, we would not crash.<div>
<br></div><div>In addition to the Clang-side changes, LLVM should probably be returning an error or reporting a fatal error instead of hitting unreachable.</div></div><div class="gmail_extra"><br><br><div class="gmail_quote">
<div><div>
On Mon, Aug 25, 2014 at 2:10 PM, Akira Hatanaka <span dir="ltr"><<a href="mailto:ahatanak@gmail.com" target="_blank">ahatanak@gmail.com</a>></span> wrote:<br></div></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div><div>
<div dir="ltr">Rebased patches attached.<div><br></div><div>I also made changes to the clang patch so that clang can error-out after a size mismatch is found as soon as possible.TargetInfo::validateConstraintModifier has an extra parameter IsError, which is set when it decides there is no point in continuing compilation and it should stop compilation immediately. The error message clang prints looks better than lllvm's message, but if it isn't right to change the warning to an error, then I guess we have to detect the error later just before isel, as is done in the llvm patch.</div>
<div><div>
<div><br></div><div><div class="gmail_extra"><div class="gmail_quote">On Wed, Jul 23, 2014 at 3:56 PM, Akira Hatanaka <span dir="ltr"><<a href="mailto:ahatanak@gmail.com" target="_blank">ahatanak@gmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr">llvm should error-out when a 64-bit variable is bound to a single register in x86 32-bit mode, but ToT clang/llvm fails to detect this error and continues compilation until it crashes in type-legalization:<div>
<br></div><div><div><p style="margin:0px;font-size:11px;font-family:Menlo">$ llc test/CodeGen/X86/inline-asm-regsize.ll -O3 -mtriple=i386-apple-darwin -o -</p></div><div><br></div><div><p style="margin:0px;font-size:11px;font-family:Menlo">
inline-asm-regsize.ll -O3 -mtriple=i386-apple-darwin -o -</p>
<p style="margin:0px;font-size:11px;font-family:Menlo"><span style="white-space:pre-wrap"> </span>.section<span style="white-space:pre-wrap"> </span>__TEXT,__text,regular,pure_instructions</p>
<p style="margin:0px;font-size:11px;font-family:Menlo">ExpandIntegerResult #0: 0x7fa2d1041728: i64 = Register %RCX [ID=0]</p>
<p style="margin:0px;font-size:11px;font-family:Menlo;min-height:13px"><br></p>
<p style="margin:0px;font-size:11px;font-family:Menlo">Do not know how to expand the result of this operator!</p>
<p style="margin:0px;font-size:11px;font-family:Menlo">UNREACHABLE executed at /Users/ahatanaka/projects/llvm/git/llvm3/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp:1116!</p></div><div><br></div><div>The attached patch fixes llvm to error-out and print this error message:</div>
<div><br></div><div><p style="margin:0px;font-size:11px;font-family:Menlo">error: Cannot bind a variable larger than 32-bit to a single register in 32-bit mode</p><p style="margin:0px;font-size:11px;font-family:Menlo"><br>
</p><div>My initial solution was to have clang detect this error in TargetInfo::validateConstraintModifier. However, the code in SemaStmtAsm.cpp has to be changed to error-out instead of issuing a warning, which I wasn't sure was the right thing to do. I am attaching this patch too in case someone has a suggestion or an opinion on it.</div>
<div><br></div></div><div><rdar://problem/17476970><br></div><div><br></div></div></div>
</blockquote></div><br></div></div></div></div></div>
<br></div></div>_______________________________________________<br>
cfe-commits mailing list<br>
<a href="mailto:cfe-commits@cs.uiuc.edu" target="_blank">cfe-commits@cs.uiuc.edu</a><br>
<a href="http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits</a><br>
<br></blockquote></div><br></div>
</blockquote></div><br></div></div></div></div>
</blockquote></div><br></div>