On Thu, Jan 15, 2015 at 2:00 PM, H.J. Lu <hjl.to...@gmail.com> wrote:
>>> gcc/testsuite/gcc.target/i386/pr54445-2.c | 2 +-
>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> 2015-01-11 H.J. Lu <hongjiu...@intel.com>
>>>
>>> * gcc.target/i386/pr54445-2.c: Add -fno-pie.
>>>
>>> diff --git a/gcc/testsuite/gcc.target/i386/pr54445-2.c
>>> b/gcc/testsuite/gcc.target/i386/pr54445-2.c
>>> index 5151c13..4d49816 100644
>>> --- a/gcc/testsuite/gcc.target/i386/pr54445-2.c
>>> +++ b/gcc/testsuite/gcc.target/i386/pr54445-2.c
>>> @@ -1,5 +1,5 @@
>>> /* { dg-do compile { target { *-*-linux* && { ! { ia32 } } } } } */
>>> -/* { dg-options "-O2 -fno-pic" } */
>>> +/* { dg-options "-O2 -fno-pic -fno-pie" } */
>>>
>>> __thread unsigned char tls_array[64];
>>
>> No. PI *executable* has nothing to do with the access to data.
>>
>> Can you rather investigate why -fpie is generating:
>>
>> movzbl %fs:-1+tls_array@tpoff(%rdi), %eax
>>
>> instead of (expected)
>>
>> movzbl %fs:tls_array@tpoff-1(%rdi), %eax
>>
>> Uros.
>
> They have the same encoding. The differences are in the order of address
> in the first operand due to different output orders of PLUS between
> output_addr_const and output_pic_addr_const.
Then we should improve scan string to accept both expressions.
Uros.
