_wrap0(): (amd64) indexing off of %rsp rather than %rbp seems to avoid compiler differences.

coro/swap.c

@@ -170,13 +170,7 @@ _wrap0 (void)
   void * co;
   // x86_64 passes args in registers.  but coro.__create() puts
   // the coroutine on the stack.  fetch it from there.
-  // llvm does the prologue differently...
-#ifdef __llvm__
-  __asm__ ("movq 16(%%rbp), %[co]" : [co] "=r" (co));
-#else
-  __asm__ ("movq 8(%%rbp), %[co]" : [co] "=r" (co));
-#endif
-
+  __asm__ ("movq 16(%%rsp), %[co]" : [co] "=r" (co));
   _wrap1 (co);
   __yield();
 }