| 1 | /* @title: Fixed Point Arithmetic */ |
| 2 | #pragma once |
| 3 | #include <kassert.h> |
| 4 | #include <types/types.h> |
| 5 | |
| 6 | #define FX_ONE ((fx32_32_t) (1LL << 32)) |
| 7 | #define FX_HALF ((fx32_32_t) (1LL << 31)) |
| 8 | |
| 9 | /* Cast truncates towards zero, so we need to round */ |
| 10 | #define FX(x) ((fx32_32_t) ((x) * 4294967296.0 + ((x) < 0 ? -0.5 : 0.5))) |
| 11 | #define FX_FROM_RATIO(n, d) ((fx32_32_t) (((__int128) (n) << 32) / (d))) |
| 12 | |
| 13 | /* strtol style */ |
| 14 | fx32_32_t fx_parse(const char *str, char **endptr); |
| 15 | |
| 16 | static inline fx32_32_t fx_add(fx32_32_t a, fx32_32_t b) { |
| 17 | return (fx32_32_t) (a + b); |
| 18 | } |
| 19 | |
| 20 | static inline fx32_32_t fx_sub(fx32_32_t a, fx32_32_t b) { |
| 21 | return (fx32_32_t) (a - b); |
| 22 | } |
| 23 | |
| 24 | static inline fx32_32_t fx_mul(fx32_32_t a, fx32_32_t b) { |
| 25 | bool neg = (a ^ b) < 0; |
| 26 | uint64_t ua = (uint64_t) (a < 0 ? -a : a); |
| 27 | uint64_t ub = (uint64_t) (b < 0 ? -b : b); |
| 28 | |
| 29 | uint64_t a_hi = ua >> 32; |
| 30 | uint64_t a_lo = ua & 0xFFFFFFFFULL; |
| 31 | uint64_t b_hi = ub >> 32; |
| 32 | uint64_t b_lo = ub & 0xFFFFFFFFULL; |
| 33 | |
| 34 | uint64_t p0 = a_lo * b_lo; |
| 35 | uint64_t p1 = a_lo * b_hi; |
| 36 | uint64_t p2 = a_hi * b_lo; |
| 37 | uint64_t p3 = a_hi * b_hi; |
| 38 | |
| 39 | uint64_t mid = (p0 >> 32) + (p1 & 0xFFFFFFFFULL) + (p2 & 0xFFFFFFFFULL); |
| 40 | uint64_t high = p3 + (p1 >> 32) + (p2 >> 32) + (mid >> 32); |
| 41 | |
| 42 | uint64_t q = (high << 32) | (mid & 0xFFFFFFFFULL); |
| 43 | |
| 44 | return (fx32_32_t) (neg ? -(int64_t) q : (int64_t) q); |
| 45 | } |
| 46 | |
| 47 | static inline fx32_32_t fx_div(fx32_32_t a, fx32_32_t b) { |
| 48 | kassert(b); |
| 49 | bool neg = (a ^ b) < 0; |
| 50 | uint64_t ua = (uint64_t) (a < 0 ? -a : a); |
| 51 | uint64_t ub = (uint64_t) (b < 0 ? -b : b); |
| 52 | |
| 53 | uint64_t num_hi = ua >> 32; |
| 54 | uint64_t num_lo = ua << 32; |
| 55 | uint64_t q = 0; |
| 56 | |
| 57 | for (int i = 63; i >= 0; i--) { |
| 58 | uint64_t bit = num_hi >> 63; |
| 59 | num_hi = (num_hi << 1) | (num_lo >> 63); |
| 60 | num_lo <<= 1; |
| 61 | |
| 62 | if (bit || num_hi >= ub) { |
| 63 | num_hi -= ub; |
| 64 | q |= (1ULL << i); |
| 65 | } |
| 66 | } |
| 67 | |
| 68 | return (fx32_32_t) (neg ? -(int64_t) q : (int64_t) q); |
| 69 | } |
| 70 | |
| 71 | static inline fx32_32_t fx_from_int(int64_t x) { |
| 72 | return x << 32; |
| 73 | } |
| 74 | |
| 75 | static inline int64_t fx_to_int(fx32_32_t x) { |
| 76 | return x >> 32; |
| 77 | } |
| 78 | |
| 79 | static inline fx32_32_t fx_min(fx32_32_t a, fx32_32_t b) { |
| 80 | return a < b ? a : b; |
| 81 | } |
| 82 | |
| 83 | static inline fx32_32_t fx_max(fx32_32_t a, fx32_32_t b) { |
| 84 | return a > b ? a : b; |
| 85 | } |
| 86 | |
| 87 | static inline fx32_32_t fx_clamp(fx32_32_t x, fx32_32_t lo, fx32_32_t hi) { |
| 88 | return fx_max(a: lo, b: fx_min(a: x, b: hi)); |
| 89 | } |
| 90 | |
| 91 | static inline fx32_32_t fx_pow_i32(fx32_32_t base, int exp) { |
| 92 | fx32_32_t result = FX_ONE; |
| 93 | if (exp < 0) { |
| 94 | exp = -exp; |
| 95 | base = fx_div(FX_ONE, b: base); |
| 96 | } |
| 97 | while (exp) { |
| 98 | if (exp & 1) |
| 99 | result = fx_mul(a: result, b: base); |
| 100 | base = fx_mul(a: base, b: base); |
| 101 | exp >>= 1; |
| 102 | } |
| 103 | return result; |
| 104 | } |
| 105 | |
| 106 | static inline fx32_32_t fx_sqrt(fx32_32_t x) { |
| 107 | if (x <= 0) |
| 108 | return 0; |
| 109 | fx32_32_t r = x; |
| 110 | for (int i = 0; i < 8; i++) { |
| 111 | r = (r + fx_div(a: x, b: (fx32_32_t) r)) >> 1; |
| 112 | } |
| 113 | return r; |
| 114 | } |
| 115 | |
| 116 | /* Use two's complement to branchlessly floor for both signs */ |
| 117 | static inline fx32_32_t fx_ceil(fx32_32_t x) { |
| 118 | return (x + FX_ONE - 1) & ~(FX_ONE - 1); |
| 119 | } |
| 120 | |
| 121 | static inline fx32_32_t fx_floor(fx32_32_t x) { |
| 122 | return x & ~(FX_ONE - 1); |
| 123 | } |
| 124 | |
| 125 | static inline fx32_32_t fx_map(fx32_32_t value, fx32_32_t from_low, |
| 126 | fx32_32_t from_high, fx32_32_t to_low, |
| 127 | fx32_32_t to_high) { |
| 128 | fx32_32_t dist = value - from_low; |
| 129 | fx32_32_t to_range = to_high - to_low; |
| 130 | fx32_32_t from_range = from_high - from_low; |
| 131 | fx32_32_t num = to_low + fx_mul(a: dist, b: to_range); |
| 132 | return fx_div(a: num, b: from_range); |
| 133 | } |
| 134 | |
| 135 | static inline fx32_32_t fx_lerp(fx32_32_t a, fx32_32_t b, fx32_32_t t) { |
| 136 | return a + fx_mul(a: t, b: b - a); |
| 137 | } |
| 138 | |
| 139 | static inline fx32_32_t fx_round_up(fx32_32_t x, fx32_32_t multiple) { |
| 140 | return fx_mul(a: fx_div(a: x + multiple - FX_ONE, b: multiple), b: multiple); |
| 141 | } |
| 142 | |
| 143 | static inline fx32_32_t fx_round_down(fx32_32_t x, fx32_32_t multiple) { |
| 144 | return fx_mul(a: fx_div(a: x, b: multiple), b: multiple); |
| 145 | } |
| 146 | |