| 1 |
use crate::constants::{MAX_I32_SCALE, MAX_PRECISION_I32, POWERS_10}; |
| 2 |
use crate::decimal::{CalculationResult, Decimal}; |
| 3 |
use crate::ops::common::{Buf12, Buf16, Buf24, Dec64}; |
| 4 |
|
| 5 |
pub(crate) fn rem_impl(d1: &Decimal, d2: &Decimal) -> CalculationResult { |
| 6 |
if d2.is_zero() { |
| 7 |
return CalculationResult::DivByZero; |
| 8 |
} |
| 9 |
if d1.is_zero() { |
| 10 |
return CalculationResult::Ok(Decimal::ZERO); |
| 11 |
} |
| 12 |
|
| 13 |
// We handle the structs a bit different here. Firstly, we ignore both the sign/scale of d2. |
| 14 |
// This is because during a remainder operation we do not care about the sign of the divisor |
| 15 |
// and only concern ourselves with that of the dividend. |
| 16 |
let mut d1 = Dec64::new(d1); |
| 17 |
let d2_scale = d2.scale(); |
| 18 |
let mut d2 = Buf12::from_decimal(d2); |
| 19 |
|
| 20 |
let cmp = crate::ops::cmp::cmp_internal( |
| 21 |
&d1, |
| 22 |
&Dec64 { |
| 23 |
negative: d1.negative, |
| 24 |
scale: d2_scale, |
| 25 |
hi: d2.hi(), |
| 26 |
low64: d2.low64(), |
| 27 |
}, |
| 28 |
); |
| 29 |
match cmp { |
| 30 |
core::cmp::Ordering::Equal => { |
| 31 |
// Same numbers meaning that remainder is zero |
| 32 |
return CalculationResult::Ok(Decimal::ZERO); |
| 33 |
} |
| 34 |
core::cmp::Ordering::Less => { |
| 35 |
// d1 < d2, e.g. 1/2. This means that the result is the value of d1 |
| 36 |
return CalculationResult::Ok(d1.to_decimal()); |
| 37 |
} |
| 38 |
core::cmp::Ordering::Greater => {} |
| 39 |
} |
| 40 |
|
| 41 |
// At this point we know that the dividend > divisor and that they are both non-zero. |
| 42 |
let mut scale = d1.scale as i32 - d2_scale as i32; |
| 43 |
if scale > 0 { |
| 44 |
// Scale up the divisor |
| 45 |
loop { |
| 46 |
let power = if scale >= MAX_I32_SCALE { |
| 47 |
POWERS_10[9] |
| 48 |
} else { |
| 49 |
POWERS_10[scale as usize] |
| 50 |
} as u64; |
| 51 |
|
| 52 |
let mut tmp = d2.lo() as u64 * power; |
| 53 |
d2.set_lo(tmp as u32); |
| 54 |
tmp >>= 32; |
| 55 |
tmp = tmp.wrapping_add((d2.mid() as u64 + ((d2.hi() as u64) << 32)) * power); |
| 56 |
d2.set_mid(tmp as u32); |
| 57 |
d2.set_hi((tmp >> 32) as u32); |
| 58 |
|
| 59 |
// Keep scaling if there is more to go |
| 60 |
scale -= MAX_I32_SCALE; |
| 61 |
if scale <= 0 { |
| 62 |
break; |
| 63 |
} |
| 64 |
} |
| 65 |
scale = 0; |
| 66 |
} |
| 67 |
|
| 68 |
loop { |
| 69 |
// If the dividend is smaller than the divisor then try to scale that up first |
| 70 |
if scale < 0 { |
| 71 |
let mut quotient = Buf12 { |
| 72 |
data: [d1.lo(), d1.mid(), d1.hi], |
| 73 |
}; |
| 74 |
loop { |
| 75 |
// Figure out how much we can scale by |
| 76 |
let power_scale; |
| 77 |
if let Some(u) = quotient.find_scale(MAX_PRECISION_I32 + scale) { |
| 78 |
if u >= POWERS_10.len() { |
| 79 |
power_scale = 9; |
| 80 |
} else { |
| 81 |
power_scale = u; |
| 82 |
} |
| 83 |
} else { |
| 84 |
return CalculationResult::Overflow; |
| 85 |
}; |
| 86 |
if power_scale == 0 { |
| 87 |
break; |
| 88 |
} |
| 89 |
let power = POWERS_10[power_scale] as u64; |
| 90 |
scale += power_scale as i32; |
| 91 |
|
| 92 |
let mut tmp = quotient.data[0] as u64 * power; |
| 93 |
quotient.data[0] = tmp as u32; |
| 94 |
tmp >>= 32; |
| 95 |
quotient.set_high64(tmp.wrapping_add(quotient.high64().wrapping_mul(power))); |
| 96 |
if power_scale != 9 { |
| 97 |
break; |
| 98 |
} |
| 99 |
if scale >= 0 { |
| 100 |
break; |
| 101 |
} |
| 102 |
} |
| 103 |
d1.low64 = quotient.low64(); |
| 104 |
d1.hi = quotient.data[2]; |
| 105 |
d1.scale = d2_scale; |
| 106 |
} |
| 107 |
|
| 108 |
// if the high portion is empty then return the modulus of the bottom portion |
| 109 |
if d1.hi == 0 { |
| 110 |
d1.low64 %= d2.low64(); |
| 111 |
return CalculationResult::Ok(d1.to_decimal()); |
| 112 |
} else if (d2.mid() | d2.hi()) == 0 { |
| 113 |
let mut tmp = d1.high64(); |
| 114 |
tmp = ((tmp % d2.lo() as u64) << 32) | (d1.lo() as u64); |
| 115 |
d1.low64 = tmp % d2.lo() as u64; |
| 116 |
d1.hi = 0; |
| 117 |
} else { |
| 118 |
// Divisor is > 32 bits |
| 119 |
return rem_full(&d1, &d2, scale); |
| 120 |
} |
| 121 |
|
| 122 |
if scale >= 0 { |
| 123 |
break; |
| 124 |
} |
| 125 |
} |
| 126 |
|
| 127 |
CalculationResult::Ok(d1.to_decimal()) |
| 128 |
} |
| 129 |
|
| 130 |
fn rem_full(d1: &Dec64, d2: &Buf12, scale: i32) -> CalculationResult { |
| 131 |
let mut scale = scale; |
| 132 |
|
| 133 |
// First normalize the divisor |
| 134 |
let shift = if d2.hi() == 0 { |
| 135 |
d2.mid().leading_zeros() |
| 136 |
} else { |
| 137 |
d2.hi().leading_zeros() |
| 138 |
}; |
| 139 |
|
| 140 |
let mut buffer = Buf24::zero(); |
| 141 |
let mut overflow = 0u32; |
| 142 |
buffer.set_low64(d1.low64 << shift); |
| 143 |
buffer.set_mid64(((d1.mid() as u64).wrapping_add((d1.hi as u64) << 32)) >> (32 - shift)); |
| 144 |
let mut upper = 3; // We start at 3 due to bit shifting |
| 145 |
|
| 146 |
while scale < 0 { |
| 147 |
let power = if -scale >= MAX_I32_SCALE { |
| 148 |
POWERS_10[9] |
| 149 |
} else { |
| 150 |
POWERS_10[-scale as usize] |
| 151 |
} as u64; |
| 152 |
let mut tmp64 = buffer.data[0] as u64 * power; |
| 153 |
buffer.data[0] = tmp64 as u32; |
| 154 |
|
| 155 |
for (index, part) in buffer.data.iter_mut().enumerate().skip(1) { |
| 156 |
if index > upper { |
| 157 |
break; |
| 158 |
} |
| 159 |
tmp64 >>= 32; |
| 160 |
tmp64 = tmp64.wrapping_add((*part as u64).wrapping_mul(power)); |
| 161 |
*part = tmp64 as u32; |
| 162 |
} |
| 163 |
// If we have overflow then also process that |
| 164 |
if upper == 6 { |
| 165 |
tmp64 >>= 32; |
| 166 |
tmp64 = tmp64.wrapping_add((overflow as u64).wrapping_mul(power)); |
| 167 |
overflow = tmp64 as u32; |
| 168 |
} |
| 169 |
|
| 170 |
// Make sure the high bit is not set |
| 171 |
if tmp64 > 0x7FFF_FFFF { |
| 172 |
upper += 1; |
| 173 |
if upper > 5 { |
| 174 |
overflow = (tmp64 >> 32) as u32; |
| 175 |
} else { |
| 176 |
buffer.data[upper] = (tmp64 >> 32) as u32; |
| 177 |
} |
| 178 |
} |
| 179 |
scale += MAX_I32_SCALE; |
| 180 |
} |
| 181 |
|
| 182 |
// TODO: Optimize slice logic |
| 183 |
|
| 184 |
let mut tmp = Buf16::zero(); |
| 185 |
let divisor = d2.low64() << shift; |
| 186 |
if d2.hi() == 0 { |
| 187 |
// Do some division |
| 188 |
if upper == 6 { |
| 189 |
upper -= 1; |
| 190 |
|
| 191 |
tmp.data = [buffer.data[4], buffer.data[5], overflow, 0]; |
| 192 |
tmp.partial_divide_64(divisor); |
| 193 |
buffer.data[4] = tmp.data[0]; |
| 194 |
buffer.data[5] = tmp.data[1]; |
| 195 |
} |
| 196 |
if upper == 5 { |
| 197 |
upper -= 1; |
| 198 |
tmp.data = [buffer.data[3], buffer.data[4], buffer.data[5], 0]; |
| 199 |
tmp.partial_divide_64(divisor); |
| 200 |
buffer.data[3] = tmp.data[0]; |
| 201 |
buffer.data[4] = tmp.data[1]; |
| 202 |
buffer.data[5] = tmp.data[2]; |
| 203 |
} |
| 204 |
if upper == 4 { |
| 205 |
tmp.data = [buffer.data[2], buffer.data[3], buffer.data[4], 0]; |
| 206 |
tmp.partial_divide_64(divisor); |
| 207 |
buffer.data[2] = tmp.data[0]; |
| 208 |
buffer.data[3] = tmp.data[1]; |
| 209 |
buffer.data[4] = tmp.data[2]; |
| 210 |
} |
| 211 |
|
| 212 |
tmp.data = [buffer.data[1], buffer.data[2], buffer.data[3], 0]; |
| 213 |
tmp.partial_divide_64(divisor); |
| 214 |
buffer.data[1] = tmp.data[0]; |
| 215 |
buffer.data[2] = tmp.data[1]; |
| 216 |
buffer.data[3] = tmp.data[2]; |
| 217 |
|
| 218 |
tmp.data = [buffer.data[0], buffer.data[1], buffer.data[2], 0]; |
| 219 |
tmp.partial_divide_64(divisor); |
| 220 |
buffer.data[0] = tmp.data[0]; |
| 221 |
buffer.data[1] = tmp.data[1]; |
| 222 |
buffer.data[2] = tmp.data[2]; |
| 223 |
|
| 224 |
let low64 = buffer.low64() >> shift; |
| 225 |
CalculationResult::Ok(Decimal::from_parts( |
| 226 |
low64 as u32, |
| 227 |
(low64 >> 32) as u32, |
| 228 |
0, |
| 229 |
d1.negative, |
| 230 |
d1.scale, |
| 231 |
)) |
| 232 |
} else { |
| 233 |
let divisor_low64 = divisor; |
| 234 |
let divisor = Buf12 { |
| 235 |
data: [ |
| 236 |
divisor_low64 as u32, |
| 237 |
(divisor_low64 >> 32) as u32, |
| 238 |
(((d2.mid() as u64) + ((d2.hi() as u64) << 32)) >> (32 - shift)) as u32, |
| 239 |
], |
| 240 |
}; |
| 241 |
|
| 242 |
// Do some division |
| 243 |
if upper == 6 { |
| 244 |
upper -= 1; |
| 245 |
tmp.data = [buffer.data[3], buffer.data[4], buffer.data[5], overflow]; |
| 246 |
tmp.partial_divide_96(&divisor); |
| 247 |
buffer.data[3] = tmp.data[0]; |
| 248 |
buffer.data[4] = tmp.data[1]; |
| 249 |
buffer.data[5] = tmp.data[2]; |
| 250 |
} |
| 251 |
if upper == 5 { |
| 252 |
upper -= 1; |
| 253 |
tmp.data = [buffer.data[2], buffer.data[3], buffer.data[4], buffer.data[5]]; |
| 254 |
tmp.partial_divide_96(&divisor); |
| 255 |
buffer.data[2] = tmp.data[0]; |
| 256 |
buffer.data[3] = tmp.data[1]; |
| 257 |
buffer.data[4] = tmp.data[2]; |
| 258 |
buffer.data[5] = tmp.data[3]; |
| 259 |
} |
| 260 |
if upper == 4 { |
| 261 |
tmp.data = [buffer.data[1], buffer.data[2], buffer.data[3], buffer.data[4]]; |
| 262 |
tmp.partial_divide_96(&divisor); |
| 263 |
buffer.data[1] = tmp.data[0]; |
| 264 |
buffer.data[2] = tmp.data[1]; |
| 265 |
buffer.data[3] = tmp.data[2]; |
| 266 |
buffer.data[4] = tmp.data[3]; |
| 267 |
} |
| 268 |
|
| 269 |
tmp.data = [buffer.data[0], buffer.data[1], buffer.data[2], buffer.data[3]]; |
| 270 |
tmp.partial_divide_96(&divisor); |
| 271 |
buffer.data[0] = tmp.data[0]; |
| 272 |
buffer.data[1] = tmp.data[1]; |
| 273 |
buffer.data[2] = tmp.data[2]; |
| 274 |
buffer.data[3] = tmp.data[3]; |
| 275 |
|
| 276 |
let low64 = (buffer.low64() >> shift) + ((buffer.data[2] as u64) << (32 - shift) << 32); |
| 277 |
CalculationResult::Ok(Decimal::from_parts( |
| 278 |
low64 as u32, |
| 279 |
(low64 >> 32) as u32, |
| 280 |
buffer.data[2] >> shift, |
| 281 |
d1.negative, |
| 282 |
d1.scale, |
| 283 |
)) |
| 284 |
} |
| 285 |
} |