test-depth-fp.c

// reset && clear && clang test-depth-fp.c && ./a.out 123

#include <stdio.h>
#include <stdint.h>
#include <stdlib.h>
#include <assert.h>

/* Found these by checking values represented by 0xFFFF and 0xFEFFFF */
#define F16_MAX 511.9375f
#define F24_MAX 340279770772509196049560342183603601408.0f

/* [0.0, 511.9375] to 16 bit */
static uint16_t convert_double_to_f16(double d) {
  assert(d >= 0.0);
  if (d == 0.0) { return 0x0000; }
  uint64_t* i = (uint64_t*)&d;
  return (*i >> 40) - 0x3F8000;
}

/* 16 bit to [0.0, 511.9375] */
static double convert_f16_to_double(uint16_t f16) {
  if (f16 == 0x0000) { return 0.0; }
  uint64_t i = (f16 + 0x3F8000ULL) << 40;
  return *(double*)&i;
}

/* [0.0, 511.9375] to 16 bit */
static uint16_t convert_float_to_f16(float f) {
  assert(f >= 0.0);
  if (f == 0.0) { return 0x0000; }
  uint32_t* i = (uint32_t*)&f;
  return (*i >> 11) - 0x3F8000;
}

/* 16 bit to [0.0, 511.9375] */
static float convert_f16_to_float(uint16_t f16) {
  if (f16 == 0x0000) { return 0.0; }
  uint32_t i = (f16 << 11) + 0x3C000000;
  return *(float*)&i;
}

/* [0.0, F24_MAX] to 24 bit */
static uint32_t convert_double_to_f24(double d) {
  assert(d >= 0.0);
  if (d == 0.0) { return 0x000000; }
  uint64_t* i = (uint64_t*)&d;
  return ((*i >> 36) - 0x3800000) & 0xFFFFFF;
}

/* 24 bit to [0.0, F24_MAX] */
static double convert_f24_to_double(uint32_t f24) {
  assert(!(f24 >> 24));
  f24 &= 0xFFFFFF;
  if (f24 == 0x000000) { return 0.0; }
  uint64_t i = (f24 + 0x3800000ULL) << 36;
  return *(double*)&i;
}

/* [0.0, F24_MAX] to 24 bit */
static uint32_t convert_float_to_f24(float d) {
  assert(d >= 0.0);
  if (d == 0.0) { return 0x000000; }
  uint32_t* i = (uint32_t*)&d;
  return ((*i >> 7) - 0x3000000) & 0xFFFFFF;
}

/* 24 bit to [0.0, F24_MAX] */
static float convert_f24_to_float(uint32_t f24) {
  assert(!(f24 >> 24));
  f24 &= 0xFFFFFF;
  if (f24 == 0x000000) { return 0.0; }
  uint32_t i = f24 << 7;
  return *(float*)&i;
}

int main(int argc, char* argv[]) {
    double zd = atof(argv[1]);
    float zf = zd;

    uint16_t f16d = convert_double_to_f16(zd);
    double r16d = convert_f16_to_double(f16d);
    printf("XB16D: Converted %f to %04X to %f\n", zd, f16d, r16d);

    uint16_t f16f = convert_float_to_f16(zf);
    float r16f = convert_f16_to_float(f16f);
    printf("XB16F: Converted %f to %04X to %f\n", zf, f16f, r16f);

    uint32_t f24d = convert_double_to_f24(zd);
    double r24d = convert_f24_to_double(f24d);
    printf("XB24D: Converted %f to %06X to %f\n", zd, f24d, r24d);

    uint32_t f24f = convert_float_to_f24(zf);
    double r24f = convert_f24_to_float(f24f);
    printf("XB24F: Converted %f to %06X to %f\n", zf, f24f, r24f);

    return 0;
}