1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534
|
// Copyright 2014-5 Randall Farmer. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
package sorts
import (
"bytes"
"sort"
)
const radix = 8
const mask = (1 << radix) - 1
// qSortCutoff is when we bail out to a quicksort. It's changed to 1 for
// certain tests so we can more easily exercise the radix sorting. This was
// around the break-even point in some sloppy tests.
var qSortCutoff = 1 << 7
const keyPanicMessage = "sort failed: Key and Less aren't consistent with each other"
const keyUint64Help = " (for float data, sortutil Key functions may help resolve this)"
const panicMessage = "sort failed: could be a data race, a bug in package sorts, or a subtle bug in the interface implementation"
// maxRadixDepth limits how deeply the radix part of string sorts can
// recurse before we bail to quicksort. Each recursion uses 2KB stack.
const maxRadixDepth = 32
// task describes a range of data to be sorted and additional
// information the sorter needs: bitshift in a numeric sort, byte offset in
// a string sort, or maximum depth (expressed as -maxDepth-1) for a
// quicksort.
type task struct{ offs, pos, end int }
// ByUint64 sorts data by a uint64 key.
func ByUint64(data Uint64Interface) {
l := data.Len()
if l < qSortCutoff {
qSort(data, 0, l)
return
}
shift := guessIntShift(data, l)
parallelSort(data, radixSortUint64, task{offs: int(shift), end: l})
// check results if we radix sorted!
for i := 1; i < l; i++ {
if data.Less(i, i-1) {
if data.Key(i) > data.Key(i-1) {
panic(keyPanicMessage + keyUint64Help)
}
panic(panicMessage)
}
}
}
// int64Key generates a uint64 from an int64
func int64Key(i int64) uint64 { return uint64(i) ^ 1<<63 }
// intwrapper tunrs an Int64Interface into a Uint64Interface for
// guessIntShift
type intwrapper struct{ Int64Interface }
func (iw intwrapper) Key(i int) uint64 {
return int64Key(iw.Int64Interface.Key(i))
}
// ByInt64 sorts data by an int64 key.
func ByInt64(data Int64Interface) {
l := data.Len()
if l < qSortCutoff {
qSort(data, 0, l)
return
}
shift := guessIntShift(intwrapper{data}, l)
parallelSort(data, radixSortInt64, task{offs: int(shift), end: l})
// check results!
for i := 1; i < l; i++ {
if data.Less(i, i-1) {
if data.Key(i) > data.Key(i-1) {
panic(keyPanicMessage + keyUint64Help)
}
panic(panicMessage)
}
}
}
// ByString sorts data by a string key.
func ByString(data StringInterface) {
l := data.Len()
if l < qSortCutoff {
qSort(data, 0, l)
return
}
parallelSort(data, radixSortString, task{end: l})
// check results if we radix sorted!
for i := 1; i < l; i++ {
if data.Less(i, i-1) {
if data.Key(i) > data.Key(i-1) {
panic(keyPanicMessage)
}
panic(panicMessage)
}
}
}
// ByBytes sorts data by a []byte key.
func ByBytes(data BytesInterface) {
l := data.Len()
if l < qSortCutoff {
qSort(data, 0, l)
return
}
parallelSort(data, radixSortBytes, task{end: l})
// check results if we radix sorted!
for i := 1; i < l; i++ {
if data.Less(i, i-1) {
if bytes.Compare(data.Key(i), data.Key(i-1)) > 0 {
panic(keyPanicMessage)
}
panic(panicMessage)
}
}
}
// guessIntShift saves a pass when the data is distributed roughly uniformly
// in a small range (think shuffled indices into a small array), and rarely
// hurts much otherwise: either it just returns 64-radix quickly, or it
// returns too small a shift and the sort notices after one useless counting
// pass.
func guessIntShift(data Uint64Interface, l int) uint {
step := l >> 5
if l > 1<<16 {
step = l >> 8
}
if step == 0 { // only for tests w/qSortCutoff lowered
step = 1
}
min := data.Key(l - 1)
max := min
for i := 0; i < l; i += step {
k := data.Key(i)
if k < min {
min = k
}
if k > max {
max = k
}
}
diff := min ^ max
log2diff := 0
for diff != 0 {
log2diff++
diff >>= 1
}
shiftGuess := log2diff - radix
if shiftGuess < 0 {
return 0
}
return uint(shiftGuess)
}
/*
Thanks to (and please refer to):
Victor J. Duvanenko, "Parallel In-Place Radix Sort Simplified", 2011, at
http://www.drdobbs.com/parallel/parallel-in-place-radix-sort-simplified/229000734
for lots of practical discussion of performance
Michael Herf, "Radix Tricks", 2001, at
http://stereopsis.com/radix.html
for the idea for Float32Key()/Float64Key() (via Pierre Tardiman, "Radix Sort
Revisited", 2000, at http://codercorner.com/RadixSortRevisited.htm) and more
performance talk.
A handy slide deck summarizing Robert Sedgewick and Kevin Wayne's Algorithms
on string sorts:
http://algs4.cs.princeton.edu/lectures/51StringSorts.pdf
for a grounding in string sorts and pointer to American flag sort
McIlroy, Bostic, and McIlroy, "Engineering Radix Sort", 1993 at
http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.22.6990
for laying out American flag sort
- We're not using American flag sort's trick of keeping our own stack. It
might help on some data, but just bailing to qsort after 32 bytes is
enough to keep stack use from exploding.
- I suspect the quicksort phase could be sped up, especially for strings.
If you collected the next, say, eight bytes of each string in an array,
sorted those, and only compared full strings as a tiebreaker, you could
likely avoid following a lot of pointers and use cache better. That's a
lot of work and a lot of code, though.
- I'm sure with a radically different approach--like with a type like this:
type Index struct { Indices, Keys uint64 }
you could do a bunch of other cool things.
*/
// All three radixSort functions below do a counting pass and a swapping
// pass, then recurse. They fall back to comparison sort for small buckets
// and equal ranges, and the int sorts try to skip bits that are identical
// across the whole range being sorted.
func radixSortUint64(dataI sort.Interface, t task, sortRange func(task)) {
data := dataI.(Uint64Interface)
shift, a, b := uint(t.offs), t.pos, t.end
if b-a < qSortCutoff {
qSort(data, a, b)
return
}
// use a single pass over the keys to bucket data and find min/max
// (for skipping over bits that are always identical)
var bucketStarts, bucketEnds [1 << radix]int
min := data.Key(a)
max := min
for i := a; i < b; i++ {
k := data.Key(i)
bucketStarts[(k>>shift)&mask]++
if k < min {
min = k
}
if k > max {
max = k
}
}
// skip past common prefixes, bail if all keys equal
diff := min ^ max
if diff == 0 {
qSortEqualKeyRange(data, a, b)
return
}
if diff>>shift == 0 || diff>>(shift+radix) != 0 {
// find highest 1 bit in diff
log2diff := 0
for diff != 0 {
log2diff++
diff >>= 1
}
nextShift := log2diff - radix
if nextShift < 0 {
nextShift = 0
}
sortRange(task{nextShift, a, b})
return
}
pos := a
for i, c := range bucketStarts {
bucketStarts[i] = pos
pos += c
bucketEnds[i] = pos
}
for curBucket, bucketEnd := range bucketEnds {
i := bucketStarts[curBucket]
for i < bucketEnd {
destBucket := (data.Key(i) >> shift) & mask
if destBucket == uint64(curBucket) {
i++
bucketStarts[destBucket]++
continue
}
data.Swap(i, bucketStarts[destBucket])
bucketStarts[destBucket]++
}
}
if shift == 0 {
pos = a
for _, end := range bucketEnds {
if end > pos+1 {
qSortEqualKeyRange(data, pos, end)
}
pos = end
}
return
}
nextShift := shift - radix
if shift < radix {
nextShift = 0
}
pos = a
for _, end := range bucketEnds {
if end > pos+1 {
sortRange(task{int(nextShift), pos, end})
}
pos = end
}
}
func radixSortInt64(dataI sort.Interface, t task, sortRange func(task)) {
data := dataI.(Int64Interface)
shift, a, b := uint(t.offs), t.pos, t.end
if b-a < qSortCutoff {
qSort(data, a, b)
return
}
// use a single pass over the keys to bucket data and find min/max
// (for skipping over bits that are always identical)
var bucketStarts, bucketEnds [1 << radix]int
min := int64Key(data.Key(a))
max := min
for i := a; i < b; i++ {
k := int64Key(data.Key(i))
bucketStarts[(k>>shift)&mask]++
if k < min {
min = k
}
if k > max {
max = k
}
}
// skip past common prefixes, bail if all keys equal
diff := min ^ max
if diff == 0 {
qSortEqualKeyRange(data, a, b)
return
}
if diff>>shift == 0 || diff>>(shift+radix) != 0 {
// find highest 1 bit in diff
log2diff := 0
for diff != 0 {
log2diff++
diff >>= 1
}
nextShift := log2diff - radix
if nextShift < 0 {
nextShift = 0
}
sortRange(task{nextShift, a, b})
return
}
pos := a
for i, c := range bucketStarts {
bucketStarts[i] = pos
pos += c
bucketEnds[i] = pos
}
for curBucket, bucketEnd := range bucketEnds {
i := bucketStarts[curBucket]
for i < bucketEnd {
destBucket := (int64Key(data.Key(i)) >> shift) & mask
if destBucket == uint64(curBucket) {
i++
bucketStarts[destBucket]++
continue
}
data.Swap(i, bucketStarts[destBucket])
bucketStarts[destBucket]++
}
}
if shift == 0 {
// each bucket is a unique key
pos = a
for _, end := range bucketEnds {
if end > pos+1 {
qSortEqualKeyRange(data, pos, end)
}
pos = end
}
return
}
nextShift := shift - radix
if shift < radix {
nextShift = 0
}
pos = a
for _, end := range bucketEnds {
if end > pos+1 {
sortRange(task{int(nextShift), pos, end})
}
pos = end
}
}
func radixSortString(dataI sort.Interface, t task, sortRange func(task)) {
data := dataI.(StringInterface)
offset, a, b := t.offs, t.pos, t.end
if offset < 0 {
// in a parallel quicksort of items w/long common key prefix
quickSortWorker(data, t, sortRange)
return
}
if b-a < qSortCutoff {
qSort(data, a, b)
return
}
if offset == maxRadixDepth {
qSortPar(data, t, sortRange)
return
}
// swap too-short strings to start and count bucket sizes
bucketStarts, bucketEnds := [256]int{}, [256]int{}
aInitial := a
for i := a; i < b; i++ {
k := data.Key(i)
if len(k) <= offset {
// swap too-short strings to start
data.Swap(a, i)
a++
continue
}
bucketStarts[k[offset]]++
}
if a > aInitial+1 {
qSortEqualKeyRange(data, aInitial, a)
}
pos := a
for i, c := range bucketStarts {
bucketStarts[i] = pos
pos += c
bucketEnds[i] = pos
if bucketStarts[i] == a && bucketEnds[i] == b {
// everything was in the same bucket
sortRange(task{offset + 1, a, b})
return
}
}
i := a
for curBucket, bucketEnd := range bucketEnds {
start := i
i = bucketStarts[curBucket]
for i < bucketEnd {
destBucket := data.Key(i)[offset]
if destBucket == byte(curBucket) {
i++
bucketStarts[destBucket]++
continue
}
data.Swap(i, bucketStarts[destBucket])
bucketStarts[destBucket]++
}
if i > start+1 {
sortRange(task{offset + 1, start, i})
}
}
}
func radixSortBytes(dataI sort.Interface, t task, sortRange func(task)) {
data := dataI.(BytesInterface)
offset, a, b := t.offs, t.pos, t.end
if offset < 0 {
// in a parallel quicksort of items w/long common key prefix
quickSortWorker(data, t, sortRange)
return
}
if b-a < qSortCutoff {
qSort(data, a, b)
return
}
if offset == maxRadixDepth {
qSortPar(data, t, sortRange)
return
}
// swap too-short strings to start and count bucket sizes
bucketStarts, bucketEnds := [256]int{}, [256]int{}
aInitial := a
for i := a; i < b; i++ {
k := data.Key(i)
if len(k) <= offset {
// swap too-short strings to start
data.Swap(a, i)
a++
continue
}
bucketStarts[k[offset]]++
}
if a > aInitial+1 {
qSortEqualKeyRange(data, aInitial, a)
}
pos := a
for i, c := range bucketStarts {
bucketStarts[i] = pos
pos += c
bucketEnds[i] = pos
if bucketStarts[i] == a && bucketEnds[i] == b {
// everything was in the same bucket
sortRange(task{offset + 1, a, b})
return
}
}
i := a
for curBucket, bucketEnd := range bucketEnds {
start := i
i = bucketStarts[curBucket]
for i < bucketEnd {
destBucket := data.Key(i)[offset]
if destBucket == byte(curBucket) {
i++
bucketStarts[destBucket]++
continue
}
data.Swap(i, bucketStarts[destBucket])
bucketStarts[destBucket]++
}
if i > start+1 {
sortRange(task{offset + 1, start, i})
}
}
}
// qSortEqualKeyRange qSorts data[a:b] if it is not already sorted
func qSortEqualKeyRange(data sort.Interface, a, b int) {
for i := a; i < b-1; i++ {
if data.Less(i+1, i) {
qSort(data, a, b)
return
}
}
return
}
|