review sw_saturate

This commit is contained in:
Bigfoot71 2025-05-18 18:51:17 +02:00
parent aead157626
commit dd48df432e

60
src/external/rlsw.h vendored
View file

@ -753,59 +753,31 @@ static inline void sw_vec4_transform(float dst[4], const float v[4], const sw_ma
static inline float sw_saturate(float x) static inline float sw_saturate(float x)
{ {
// Clamps a floating point value between 0.0 and 1.0 union { float f; uint32_t u; } fb;
fb.f = x;
// This implementation uses IEEE 754 bit manipulation: const uint32_t ZERO_BITS = 0x00000000; // Bit pattern of 0.0f
// - Uses the sign bit to detect negative values const uint32_t ONE_BITS = 0x3F800000; // Bit pattern of 1.0f (exp = 127, mantissa = 0)
// - Directly compares with binary representation of 1.0f to detect values > 1.0
// Use union to access the bits of the float as an unsigned int // Check if x < 0.0f
union { float f; uint32_t u; } v; // If sign bit is set (MSB), x is negative
v.f = x; if ((fb.u & 0x80000000) != 0) {
return 0.0f;
}
// Check sign bit (bit 31): if set, x is negative, return 0.0f // Check if x > 1.0f
if (v.u & 0x80000000) return 0.0f; // Works for positive floats: IEEE 754 ordering matches integer ordering
if (fb.u > ONE_BITS) {
return 1.0f;
}
// Extract the unsigned magnitude (exponent + mantissa bits) // x is in [0.0f, 1.0f]
uint32_t expMantissa = v.u & 0x7FFFFFFF;
// If magnitude > binary representation of 1.0f (0x3F800000), return 1.0f
// This efficiently handles all values > 1.0f without additional computation
if (expMantissa > 0x3F800000) return 1.0f;
// Value is between 0.0f and 1.0f inclusive, return unchanged
return x; return x;
} }
static inline float sw_fract(float x) static inline float sw_fract(float x)
{ {
// Computes the positive fractional part of a float. return x - floorf(x);
// Equivalent to fabs(x) - floorf(fabs(x)).
// Uses IEEE 754 bit tricks for efficiency and edge case handling.
union { float f; uint32_t u; } v;
v.f = x;
// Get absolute value bits (clear sign bit)
uint32_t abs_bits = v.u & 0x7FFFFFFF;
// Case 1: |x| < 1.0f -> integer part is 0, return |x|
if (abs_bits < 0x3F800000) {
v.u = abs_bits; // Ensure positive result
return v.f;
}
// Case 2: |x| ≥ 2^24 -> float is an exact integer, return 0.0f
// Also handles Inf and NaN as 0.0f
if (abs_bits >= 0x4B000000) {
return 0.0f;
}
// Case 3: 1.0f ≤ |x| < 2^24 -> compute |x| - floor(|x|)
v.u = abs_bits;
float abs_x = v.f;
return abs_x - floorf(abs_x);
} }
static inline int sw_clampi(int v, int min, int max) static inline int sw_clampi(int v, int min, int max)