Re: [Tinycc-devel] The inline assembly clobber for floating point register stack `%st' not implemented?

[Roy Tam]

Hello,


2015-01-21 1:14 GMT+08:00 Sergey Korshunoff <address@hidden>:
> There is patch for the win32/include/math.h which removes "t", "st"
>

I wonder why not just have %t,%st implemented?
I wrote a small hack for that about half year ago.
http://lists.gnu.org/archive/html/tinycc-devel/2014-08/msg00024.html

> 2015-01-20 18:25 GMT+03:00, Sergey Korshunoff <address@hidden>:
>> Because I don't know how to handle a "t" constraint right, a math.h
>> can be rewritten
>> like:
>> long lrint (double x)
>> {
>>     long retval;
>>     __asm__ __volatile__ (
>>         "fldl   %1\n"
>>         "fistpl %0\n"  : "=m" (retval) : "m" (x) );
>>     return retval;
>> }
>> insteed of the
>> long lrint (double x)
>> {
>>      long retval;
>>      __asm__ __volatile__ ("fistpl %0"  : "=m" (retval) : "t" (x) : "st");
>>        return retval;
>> }
>>
>> 2015-01-20 17:43 GMT+03:00, Sergey Korshunoff <address@hidden>:
>>> Handling of the "t" spec needs to be improved. I take patch for it
>>> from the this mail list (2014)
>>> tcc don't load a function parameter into the float stack.
>>> There is asm of the gcc and tcc:
>>>
>>> long lrint (double x)
>>> {
>>>     long retval;
>>>     __asm__ __volatile__ ("fistpl %0"  : "=m" (retval) : "t" (x) : "st");
>>>       return retval;
>>> }
>>>
>>> gcc:
>>>   push   %ebp
>>>   mov    %esp,%ebp
>>>   sub    $0x4,%esp
>>>   fldl      0x8(%ebp)    # !!!
>>>   fistpl  -0x4(%ebp)
>>>   mov   -0x4(%ebp),%eax
>>>   leave
>>>   ret
>>>
>>> tcc:
>>>   push   %ebp
>>>   mov    %esp,%ebp
>>>   sub    $0x4,%esp
>>>   fistpl -0x4(%ebp)    # where is fldl ???
>>>   mov  -0x4(%ebp),%eax
>>>   jmp    14 <lrint+0x14>
>>>   leave
>>>   ret
>>>
>>
>
> _______________________________________________
> Tinycc-devel mailing list
> address@hidden
> https://lists.nongnu.org/mailman/listinfo/tinycc-devel
>