Normal Quantization

opengl 에서 gpu로의 데이타 전송을 최소화 하는것이 성능상 이점이 있음.
특히 표면 노말의 경우는 x, y, z 3개의 float을 넘겨주는데.. 총 12바이트를 소요함.
이를 하나의 signed int (10.10.10.2) 로 표현하면 4바이트로 줄어듬.
비록 약간의 데이타 손실은 있지만 opengl에서 normal 로 표면을 표한할때는 차이가 거의 없음.
아래와 같이 약간의 손실을 감수하면서 opengl의 성능을 올릴때 사용하면 좋을것임.

노말말고 다른데 적용할부분이 어디가 있을까..

#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
 
static inline uint8_t   encode8_unorm(float    x) { return uint8_t( int (x * 255.f + 0.5f) ); }
static inline float     decode8_unorm(uint8_t  x) { return x / 255.f; }
static inline uint8_t   encode8_snorm(float    x) { return uint8_t( int (x * 127.f + (x > 0 ? 0.5f : -0.5f)) ); }
static inline float     decode8_snorm(uint8_t  x) { float f = x / 127.f; return f <= -1 ? -1.f : (f >= 1 ? 1.f : f); }
static inline uint8_t   encode8_snorm_gl2(float   x) { return uint8_t( int (x * 127.5f + (x >= 0.f ? 0.f : -1.f)) ); }
static inline float     decode8_snorm_gl2(uint8_t x) { return (2*x + 1) / 255.f; }
static inline uint16_t  encode10_unorm(float    x) { return uint16_t( int (x * 1023.f + 0.5f) ); }
static inline float     decode10_unorm(uint16_t x) { return x / 1023.f; }
static inline uint16_t  encode10_snorm(float    x) { return encode10_unorm( ( x + 1 ) * 0.5f ); }
static inline float     decode10_snorm(uint16_t x) { return ( decode10_unorm( x ) * 2 ) - 1 ; }
 
int main()
{
    float x, y, z;
    x = 0.001;
    y = -0.002;
    z = 0.005;
 
    int16_t xx = encode10_snorm(x);
    int16_t yy = encode10_snorm(y);
    int16_t zz = encode10_snorm(z);
 
    uint32_t v = uint32_t( (2<<30) | (encode10_snorm(x)<<20) | (encode10_snorm(y)<<10) | (encode10_snorm(z)<<0));
 
    float nx = decode10_snorm( ( v >> 20 ) & 0x3FF );
    float ny = decode10_snorm( ( v >> 10 ) & 0x3FF );
    float nz = decode10_snorm( ( v >>  0 ) & 0x3FF );
    printf("%f %f %f\n", x, y, z);
    printf("%f %f %f\n", nx, ny, nz);
 
    printf("%d %d\n", sizeof(float)*3, sizeof(uint32_t));
}

0.001000 -0.002000 0.005000
0.000978 -0.002933 0.004888
12 4