http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58805
--- Comment #13 from Richard Biener <rguenth at gcc dot gnu.org> --- (In reply to Richard Biener from comment #12) > (In reply to vries from comment #10) > > I've tried out the example from comment 9. > > > > Tree-tail-merge considers the statements without effect because: > > ... > > (gdb) call debug_gimple_stmt (stmt) > > # .MEM_10 = VDEF <.MEM_3(D)> > > __asm__("movq $42, %0 > > movq %0, %1 > > " : "t" "=&r" t_8, "r" "=r" *x_4(D)); > > (gdb) call gimple_has_side_effects (stmt) > > $2 = false > > ... > > and tree-ssa-tail-merge.c:stmt_local_def will return true. > > > > The side effect of the asm is in the expression for the output operand, so > > gimple_has_side_effects should return true. > > No it shouldn't. It should return true if the stmt has side-effects that > are _not_ explicit in the statement. This side-effect is explicitely there. That said, it doesn't return true for the call foo(x) in struct X { int a[256]; }; struct X __attribute__((const)) foo (); void bar (void) { struct X x; x = foo (); } either, even though the call has a (explicit) store.