*** Bug llvm/llvm-bugzilla-archive#1490 has been marked as a duplicate of this bug. ***

*** Bug llvm/llvm-bugzilla-archive#1640 has been marked as a duplicate of this bug. ***

Probably not that useful to comment, but I just ran into this attempting to compile ffmpeg.

Eli, can you include the C source for the inline asm expression that causes the problem?

code from ffmpeg
Okay, here's the preprocessed source of the relevant function. It's in the file dsputil_mmx_avg.h in ffmpeg/libavcodec.

Filed Bug 2019... we can work out what's going on with this testcase there.

Reduced testcase here:

int pixels;
int h;

void test() {
__asm __volatile(" %0 %1 %2 %3"
:"+imr"(h), "+S"(pixels)
: :"eax", "memory");
}
void test2() {
__asm __volatile(
" %0 %1 %2 %3"
:"+r"(h), "+S"(pixels)
: :"eax", "memory");
}

llvm-gcc rejects the former, but is fine with the later.

Hmm, so it is actually this bug... I guess that makes bug 2019 a clang bug.

Patch
As far as I can tell, this fix is correct. Of course, this should be reviewed by someone familiar with the relevant code.

Your patch looks great to me, applied:
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20080218/058625.html

Thanks Eli!

mentioned in issue llvm/llvm-bugzilla-archive#1490

mentioned in issue llvm/llvm-bugzilla-archive#1640

mentioned in issue llvm/llvm-bugzilla-archive#1682

mentioned in issue #1234