Compiler/TMS320F28377S: Inaccurate floating-point constants

Tool/software: TI C/C++ Compiler

Hi,

I am trying to use function macro to convert Q15.16 number to floating-point number by converting integer value to float and multiplying the result by inverse of 65536.0.

#define Fix16ToFloat(x)   ( (float)(x) * 0x1p-16 )

However, the compiler (15.2.12) list file shows that the exact hexadecimal value has been changed to 0.000015F.

Hexadecimal value of 0x1p-16 is exactly 0x37800000 but value of 0.000015F is 0x377ba882.

For example, the following program compiled with GCC shows the problem:

  float u = 1.0f/65536.0f;
  float v = 0x1p-16;
  float t = 0.000015f;
  printf("\n%08x %f %g\n%08x %f %g\n%08x %f %g\n", *(uint32_t *)&u, u, u, *(uint32_t *)&v, v, v, *(uint32_t *)&t, t, t );

which prints out:

37800000 0.000015 1.52588e-005
37800000 0.000015 1.52588e-005
377ba882 0.000015 1.5e-005

Compiler uses 0.000015F Even if I explicitly use 0x1p-16 in calculation.

Now, if I compile the following program with C2000

float u = 1.0f/65536.0f;
float v = 0x1p-16;
float t = 0.000015f;

the lst file shows

    4607              ;*** 639        -----------------------    u = 0.000015F;
    4608              ;*** 640        -----------------------    v = u;
    4609              ;*** 641        -----------------------    t = 0.000015F;

EDIT: It seems that the memory contents are still ok...

u == 37800000
v == 37800000
t == 377BA882

So I guess nothing after all except that lst/asm file could show the exact floating-point value.

Juhamatti