| 1 | // Ported from musl, which is licensed under the MIT license: |
| 2 | // https://git.musl-libc.org/cgit/musl/tree/COPYRIGHT |
| 3 | // |
| 4 | // https://git.musl-libc.org/cgit/musl/tree/src/math/__rem_pio2.c |
| 5 | |
| 6 | const std = @import("std"); |
| 7 | const rem_pio2_large = @import("rem_pio2_large.zig").rem_pio2_large; |
| 8 | const math = std.math; |
| 9 | |
| 10 | const toint = 1.5 / math.floatEps(f64); |
| 11 | // pi/4 |
| 12 | const pio4 = 0x1.921fb54442d18p-1; |
| 13 | // invpio2: 53 bits of 2/pi |
| 14 | const invpio2 = 6.36619772367581382433e-01; // 0x3FE45F30, 0x6DC9C883 |
| 15 | // pio2_1: first 33 bit of pi/2 |
| 16 | const pio2_1 = 1.57079632673412561417e+00; // 0x3FF921FB, 0x54400000 |
| 17 | // pio2_1t: pi/2 - pio2_1 |
| 18 | const pio2_1t = 6.07710050650619224932e-11; // 0x3DD0B461, 0x1A626331 |
| 19 | // pio2_2: second 33 bit of pi/2 |
| 20 | const pio2_2 = 6.07710050630396597660e-11; // 0x3DD0B461, 0x1A600000 |
| 21 | // pio2_2t: pi/2 - (pio2_1+pio2_2) |
| 22 | const pio2_2t = 2.02226624879595063154e-21; // 0x3BA3198A, 0x2E037073 |
| 23 | // pio2_3: third 33 bit of pi/2 |
| 24 | const pio2_3 = 2.02226624871116645580e-21; // 0x3BA3198A, 0x2E000000 |
| 25 | // pio2_3t: pi/2 - (pio2_1+pio2_2+pio2_3) |
| 26 | const pio2_3t = 8.47842766036889956997e-32; // 0x397B839A, 0x252049C1 |
| 27 | |
| 28 | fn medium(ix: u32, x: f64, y: *[2]f64) i32 { |
| 29 | var w: f64 = undefined; |
| 30 | var t: f64 = undefined; |
| 31 | var r: f64 = undefined; |
| 32 | var @"fn": f64 = undefined; |
| 33 | var n: i32 = undefined; |
| 34 | var ex: i32 = undefined; |
| 35 | var ey: i32 = undefined; |
| 36 | var ui: u64 = undefined; |
| 37 | |
| 38 | // rint(x/(pi/2)) |
| 39 | @"fn" = x * invpio2 + toint - toint; |
| 40 | n = @intFromFloat(@"fn"); |
| 41 | r = x - @"fn" * pio2_1; |
| 42 | w = @"fn" * pio2_1t; // 1st round, good to 85 bits |
| 43 | // Matters with directed rounding. |
| 44 | if (r - w < -pio4) { |
| 45 | n -= 1; |
| 46 | @"fn" -= 1; |
| 47 | r = x - @"fn" * pio2_1; |
| 48 | w = @"fn" * pio2_1t; |
| 49 | } else if (r - w > pio4) { |
| 50 | n += 1; |
| 51 | @"fn" += 1; |
| 52 | r = x - @"fn" * pio2_1; |
| 53 | w = @"fn" * pio2_1t; |
| 54 | } |
| 55 | y[0] = r - w; |
| 56 | ui = @bitCast(y[0]); |
| 57 | ey = @intCast((ui >> 52) & 0x7ff); |
| 58 | ex = @intCast(ix >> 20); |
| 59 | if (ex - ey > 16) { // 2nd round, good to 118 bits |
| 60 | t = r; |
| 61 | w = @"fn" * pio2_2; |
| 62 | r = t - w; |
| 63 | w = @"fn" * pio2_2t - ((t - r) - w); |
| 64 | y[0] = r - w; |
| 65 | ui = @bitCast(y[0]); |
| 66 | ey = @intCast((ui >> 52) & 0x7ff); |
| 67 | if (ex - ey > 49) { // 3rd round, good to 151 bits, covers all cases |
| 68 | t = r; |
| 69 | w = @"fn" * pio2_3; |
| 70 | r = t - w; |
| 71 | w = @"fn" * pio2_3t - ((t - r) - w); |
| 72 | y[0] = r - w; |
| 73 | } |
| 74 | } |
| 75 | y[1] = (r - y[0]) - w; |
| 76 | return n; |
| 77 | } |
| 78 | |
| 79 | // Returns the remainder of x rem pi/2 in y[0]+y[1] |
| 80 | // |
| 81 | // use rem_pio2_large() for large x |
| 82 | // |
| 83 | // caller must handle the case when reduction is not needed: |x| ~<= pi/4 */ |
| 84 | pub fn rem_pio2(x: f64, y: *[2]f64) i32 { |
| 85 | var z: f64 = undefined; |
| 86 | var tx: [3]f64 = undefined; |
| 87 | var ty: [2]f64 = undefined; |
| 88 | var n: i32 = undefined; |
| 89 | var ix: u32 = undefined; |
| 90 | var sign: bool = undefined; |
| 91 | var i: i32 = undefined; |
| 92 | var ui: u64 = undefined; |
| 93 | |
| 94 | ui = @bitCast(x); |
| 95 | sign = ui >> 63 != 0; |
| 96 | ix = @truncate((ui >> 32) & 0x7fffffff); |
| 97 | if (ix <= 0x400f6a7a) { // |x| ~<= 5pi/4 |
| 98 | if ((ix & 0xfffff) == 0x921fb) { // |x| ~= pi/2 or 2pi/2 |
| 99 | return medium(ix, x, y); |
| 100 | } |
| 101 | if (ix <= 0x4002d97c) { // |x| ~<= 3pi/4 |
| 102 | if (!sign) { |
| 103 | z = x - pio2_1; // one round good to 85 bits |
| 104 | y[0] = z - pio2_1t; |
| 105 | y[1] = (z - y[0]) - pio2_1t; |
| 106 | return 1; |
| 107 | } else { |
| 108 | z = x + pio2_1; |
| 109 | y[0] = z + pio2_1t; |
| 110 | y[1] = (z - y[0]) + pio2_1t; |
| 111 | return -1; |
| 112 | } |
| 113 | } else { |
| 114 | if (!sign) { |
| 115 | z = x - 2 * pio2_1; |
| 116 | y[0] = z - 2 * pio2_1t; |
| 117 | y[1] = (z - y[0]) - 2 * pio2_1t; |
| 118 | return 2; |
| 119 | } else { |
| 120 | z = x + 2 * pio2_1; |
| 121 | y[0] = z + 2 * pio2_1t; |
| 122 | y[1] = (z - y[0]) + 2 * pio2_1t; |
| 123 | return -2; |
| 124 | } |
| 125 | } |
| 126 | } |
| 127 | if (ix <= 0x401c463b) { // |x| ~<= 9pi/4 |
| 128 | if (ix <= 0x4015fdbc) { // |x| ~<= 7pi/4 |
| 129 | if (ix == 0x4012d97c) { // |x| ~= 3pi/2 |
| 130 | return medium(ix, x, y); |
| 131 | } |
| 132 | if (!sign) { |
| 133 | z = x - 3 * pio2_1; |
| 134 | y[0] = z - 3 * pio2_1t; |
| 135 | y[1] = (z - y[0]) - 3 * pio2_1t; |
| 136 | return 3; |
| 137 | } else { |
| 138 | z = x + 3 * pio2_1; |
| 139 | y[0] = z + 3 * pio2_1t; |
| 140 | y[1] = (z - y[0]) + 3 * pio2_1t; |
| 141 | return -3; |
| 142 | } |
| 143 | } else { |
| 144 | if (ix == 0x401921fb) { // |x| ~= 4pi/2 */ |
| 145 | return medium(ix, x, y); |
| 146 | } |
| 147 | if (!sign) { |
| 148 | z = x - 4 * pio2_1; |
| 149 | y[0] = z - 4 * pio2_1t; |
| 150 | y[1] = (z - y[0]) - 4 * pio2_1t; |
| 151 | return 4; |
| 152 | } else { |
| 153 | z = x + 4 * pio2_1; |
| 154 | y[0] = z + 4 * pio2_1t; |
| 155 | y[1] = (z - y[0]) + 4 * pio2_1t; |
| 156 | return -4; |
| 157 | } |
| 158 | } |
| 159 | } |
| 160 | if (ix < 0x413921fb) { // |x| ~< 2^20*(pi/2), medium size |
| 161 | return medium(ix, x, y); |
| 162 | } |
| 163 | // all other (large) arguments |
| 164 | if (ix >= 0x7ff00000) { // x is inf or NaN |
| 165 | y[0] = x - x; |
| 166 | y[1] = y[0]; |
| 167 | return 0; |
| 168 | } |
| 169 | // set z = scalbn(|x|,-ilogb(x)+23) |
| 170 | ui = @bitCast(x); |
| 171 | ui &= std.math.maxInt(u64) >> 12; |
| 172 | ui |= @as(u64, 0x3ff + 23) << 52; |
| 173 | z = @bitCast(ui); |
| 174 | |
| 175 | i = 0; |
| 176 | while (i < 2) : (i += 1) { |
| 177 | tx[@intCast(i)] = @floatFromInt(@as(i32, @intFromFloat(z))); |
| 178 | z = (z - tx[@intCast(i)]) * 0x1p24; |
| 179 | } |
| 180 | tx[@intCast(i)] = z; |
| 181 | // skip zero terms, first term is non-zero |
| 182 | while (tx[@intCast(i)] == 0.0) { |
| 183 | i -= 1; |
| 184 | } |
| 185 | n = rem_pio2_large(tx[0..], ty[0..], @as(i32, @intCast((ix >> 20))) - (0x3ff + 23), i + 1, 1); |
| 186 | if (sign) { |
| 187 | y[0] = -ty[0]; |
| 188 | y[1] = -ty[1]; |
| 189 | return -n; |
| 190 | } |
| 191 | y[0] = ty[0]; |
| 192 | y[1] = ty[1]; |
| 193 | return n; |
| 194 | } |