Richard Earnshaw via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
> On 22/11/2022 09:01, Christophe Lyon via Gcc-patches wrote:
>> gcc.target/aarch64/aapcs64/test_dfp_17.c has been failing on
>> big-endian, because the _Decimal32 on-stack argument is not padded in
>> the same direction depending on endianness.
>>
>> This patch fixes the testcase so that it expects the argument in the
>> right stack location, similarly to what other tests do in the same
>> directory.
>>
>> gcc/testsuite/ChangeLog:
>>
>> PR target/107604
>> * gcc.target/aarch64/aapcs64/test_dfp_17.c: Fix for big-endian.
>> ---
>> gcc/testsuite/gcc.target/aarch64/aapcs64/test_dfp_17.c | 4 ++++
>> 1 file changed, 4 insertions(+)
>>
>> diff --git a/gcc/testsuite/gcc.target/aarch64/aapcs64/test_dfp_17.c
>> b/gcc/testsuite/gcc.target/aarch64/aapcs64/test_dfp_17.c
>> index 22dc462bf7c..3c45f715cf7 100644
>> --- a/gcc/testsuite/gcc.target/aarch64/aapcs64/test_dfp_17.c
>> +++ b/gcc/testsuite/gcc.target/aarch64/aapcs64/test_dfp_17.c
>> @@ -32,6 +32,10 @@ struct z b = { 9.0dd, 10.0dd, 11.0dd, 12.0dd };
>> ANON(struct z, a, D1)
>> ANON(struct z, b, STACK)
>> ANON(int , 5, W0)
>> +#ifndef __AAPCS64_BIG_ENDIAN__
>> ANON(_Decimal32, f1, STACK+32) /* Note: no promotion to _Decimal64. */
>> +#else
>> + ANON(_Decimal32, f1, STACK+36) /* Note: no promotion to _Decimal64. */
>> +#endif
>> LAST_ANON(_Decimal64, 0.5dd, STACK+40)
>> #endif
>
> Why would a Decimal32 change stack placement based on the endianness?
> Isn't it a 4-byte object?
Yes, but PARM_BOUNDARY (64) sets a minimum alignment for all stack arguments.
Richard
