On Sun, Mar 4, 2012 at 11:38 PM, H.J. Lu <hjl.to...@gmail.com> wrote:
> On Sun, Mar 4, 2012 at 2:12 PM, Uros Bizjak <ubiz...@gmail.com> wrote:
>> On Fri, Mar 2, 2012 at 9:36 PM, H.J. Lu <hongjiu...@intel.com> wrote:
>>
>>> X86-64 linker optimizes TLS_MODEL_INITIAL_EXEC to TLS_MODEL_LOCAL_EXEC
>>> by checking
>>>
>>> movq foo@gottpoff(%rip), %reg
>>>
>>> and
>>>
>>> addq foo@gottpoff(%rip), %reg
>>>
>>> It uses the REX prefix to avoid the last byte of the previous
>>> instruction. With 32bit Pmode, we may not have the REX prefix and
>>> the last byte of the previous instruction may be an offset, which
>>> may look like a REX prefix. IE->LE optimization will generate corrupted
>>> binary. This patch makes sure we always output an REX pfrefix for
>>> UNSPEC_GOTNTPOFF. OK for trunk?
>>
>> No, please implement this using UNSPEC in the same way as
>> tls_initial_exec_64_sun implements Sun linker quirk.
>>
>
> I am not sure how it can be done with UNSPEC cleanly.
> Unlike the Sun linker issue, this is an instruction encoding
> issue. At instruction pattern level, there is no difference
> between x32 and x86-64. I need to make sure that there is
> always one and only one REX prefix with UNSPEC_GOTNTPOFF.
> If REG is r8-r15, we shouldn't add another REX prefix.
(define_insn "tls_initial_exec_x32"
[(set (match_operand:SI 0 "register_operand" "=r")
(unspec:SI
[(match_operand:SI 1 "tls_symbolic_operand" "")]
UNSPEC_TLS_IE_X32))
(clobber (reg:CC FLAGS_REG))]
"TARGET_X32"
{
if (!REX_INT_REG_P (operands[0])
fputs (\trex\n", asm_out_file);
output_asm_insn
("mov{l}\t{%%fs:0, %0|%0, QWORD PTR fs:0}", operands);
if (!REX_INT_REG_P (operands[0])
fputs (\trex\n", asm_out_file);
return "add{l}\t{%a1@gottpoff(%%rip), %0|%0, %a1@gottpoff[rip]}";
}
[(set_attr "type" "multi")])
rex or rex64 prefix, whatever you wish.
Then generate this pattern from legitimize_tls_address,
<TLS_MODEL_INITIAL_EXEC>, see TARGET_SUN_TLS part.
Uros.
