https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103640
Bug ID: 103640
Summary: asm goto w/ outputs numbering with tied outputs
differs from clang
Product: gcc
Version: 12.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: ndesaulniers at google dot com
CC: foom at fuhm dot net, segher at gcc dot gnu.org, vmakarov
at redhat dot com
Target Milestone: ---
Consider the following input:
void *x(void) {
void *p = &&foo;
asm goto ("# %0\n\t# %l1":"+r"(p):::foo);
foo:;
return p;
}
%0 refers to the 0th operand `p`.
%l1 refers to the 1st operand `foo`.
This code compiles in Clang, but not GCC.
If you change %l1 to %l2, that will compile in GCC but not clang.
Because +r constraint is being used, I suspect there's a hidden input being
tied to the output.
While we can change this in clang (https://reviews.llvm.org/D115471), I do
think it would be more ergonomic for developers if the order of operands for
the outputs templates matched the order of the asm operands. Otherwise,
developers need to understand hidden inputs, or use symbolic names like:
asm goto ("# %0\n\t# %l[foo]":"+r"(p):::foo);
I suspect there's not enough users yet of asm goto w/ outputs to notice such a
change, either way.
