On Wed, 18 Dec 2013 17:06:29 +0100, Mark Wielaard wrote:
I thought I understood
the casting, but apparently the compiler (gcc 4.8.2) doesn't...
One has to run the compiler on s390* machine as I forgot first.
--- a/backends/s390_initreg.c
+++ b/backends/s390_initreg.c
@@ -68,8 +68,15 @@ s390_set_initial_registers_tid (pid_t tid __attribute__ ((unused)),
eu_static_assert (sizeof user_regs.regs.fp_regs.fprs[0]
== sizeof dwarf_regs[0]);
for (unsigned u = 0; u < 16; u++)
- dwarf_regs[u] = *((const __typeof (dwarf_regs[0]) *)
- &user_regs.regs.fp_regs.fprs[u]);
+ {
+ // Store the double bits as is in the Dwarf_Word without conversion.
+ union
+ {
+ double d;
+ Dwarf_Word w;
+ } fpr = { .d = user_regs.regs.fp_regs.fprs[u] };
+ dwarf_regs[u] = fpr.w;
+ }
if (! setfunc (16, 16, dwarf_regs, arg))
return false;
dwarf_regs[0] = user_regs.regs.psw.addr;
Yes, I am fine with the patch, please check it in.
Thanks,
Jan