Re: Funny results with long double denorms on m68k

[Laurent Vivier <laurent@vivier.eu>]

Le 21/08/2023 à 10:34, Laurent Vivier a écrit :
> Le 20/08/2023 à 22:55, Keith Packard a écrit :
>> #include <stdio.h>
>> #include <stdint.h>
>>
>> #define X       0x1p+16383l
>> #define Y       0x1p-16446l
>>
>> static long double build_mul = X * Y;
>> static volatile long double x = X;
>> static volatile long double y = Y;
>>
>> static void
>> dump_ld(const char *label, long double ld)
>> {
>>      union {
>>          long double     d;
>>          struct {
>>              uint32_t    exp:16;
>>              uint32_t    space:16;
>>              uint32_t    h;
>>              uint32_t    l;
>>          };
>>      } u;
>>
>>      u.d = ld;
>>      printf("%12s: % -27La 0x%04x 0x%08x 0x%08x\n", label, u.d, u.exp, u.h, u.l);
>> }
>>
>> int main(void)
>> {
>>      long double runtime_mul = x * y;
>>
>>      dump_ld("x", x);
>>      dump_ld("y", y);
>>      dump_ld("build_mul", build_mul);
>>      dump_ld("runtime_mul", runtime_mul);
>>      return 0;
>> }
> 
> FYI, result of this program on a real hardware (q800/m68040) is:
> 
>             x:  0x8p+16380                 0x7ffe 0x80000000 0x00000000
>             y:  0x0.000000000000001p-16386 0x0000 0x00000000 0x00000001
>     build_mul:  0x8p-66                    0x3fc0 0x80000000 0x00000000
>   runtime_mul:  0x8p-66                    0x3fc0 0x80000000 0x00000000
> 

When I developped the FPU emulation I compared the result of QEMU and a real hardware using 
https://github.com/vivier/m68k-testfloat and https://github.com/vivier/m68k-softfloat

Thanks,
Laurent