-
Notifications
You must be signed in to change notification settings - Fork 11
/
USAGE.txt
814 lines (572 loc) · 32.9 KB
/
USAGE.txt
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
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
PerfectHashBulkCreate.exe Usage:
<KeysDirectory> <OutputDirectory>
<Algorithm> <HashFunction> <MaskFunction>
<MaximumConcurrency>
[BulkCreateFlags] [KeysLoadFlags] [TableCreateFlags]
[TableCompileFlags] [TableCreateParameters]
PerfectHashCreate.exe Usage:
<KeysPath> <OutputDirectory>
<Algorithm> <HashFunction> <MaskFunction>
<MaximumConcurrency>
[CreateFlags] [KeysLoadFlags] [TableCreateFlags]
[TableCompileFlags] [TableCreateParameters]
Create and Bulk Create Flags:
--SkipTestAfterCreate
Normally, after a table has been successfully created, it is tested.
Setting this flag disables this behavior.
N.B. This will also disable benchmarking, so no performance information
will be present in the .csv output file.
--Compile
Compiles the table after creation.
N.B. msbuild.exe must be on the PATH environment variable for this
to work. (The current error message if this isn't the case is
quite cryptic.)
Keys Load Flags:
--TryLargePagesForKeysData
Tries to allocate the keys buffer using large pages.
--SkipKeysVerification
Skips the logic that enumerates all keys after loading and a) ensures
they are sorted, and b) constructs a keys bitmap. If you can be certain
the keys are sorted, specifying this flag can provide a speedup when
loading large key sets.
--DisableImplicitKeyDownsizing
When loading keys that are 64-bit (8 bytes), a bitmap is kept that
tracks whether or not a given bit was seen across the entire key set.
After enumerating the set, the number of zeros in the bitmap are
counted; if this number is less than or equal to 32, it means that the
entire key set can be compressed into 32-bit values with some parallel
bit extraction logic (i.e. _pext_u64()). As this has beneficial size
and performance implications, when detected, the key load operation will
implicitly heap-allocate another array and convert all the 64-bit keys
into their unique 32-bit equivalent. Specifying this flag will disable
this behavior.
--TryInferKeySizeFromKeysFilename
The default key size is 32-bit (4 bytes). When this flag is present,
if the keys file name ends with "64.keys" (e.g. "foo64.keys"), the key
size will be interpreted as 64-bit (8 bytes). This flag takes
precedence over the table create parameter --KeySizeInBytes.
Table Create Flags:
--Silent
Disables console printing of the dots, dashes and other characters used
to (crudely) visualize the result of individual table create operations,
and disable best graph information from being printed to the console.
N.B. Implies --Quiet.
--Quiet
Disables printing best graph information to the console; only the dots
and dashes etc. will be printed.
N.B. Incompatible with --Silent.
--NoFileIo
Disables writing of all files when a perfect hash solution has been
found. The only time you would use this flag from the console
application is to observe the performance of table creation without
performing any file I/O operations.
--Paranoid
Enables redundant checks in the routine that determines whether or not
a generated graph is acyclic.
--FirstGraphWins | --FindBestGraph
--FirstGraphWins [default]
This is the default behavior. When searching for solutions in
parallel, the first graph to be found, "wins". i.e. it's the
solution that is subsequently written to disk.
--FindBestGraph
Requires the following two table create parameters to be present:
--BestCoverageAttempts=N
--BestCoverageType=<CoverageType>
The table create routine will then run until it finds the number of
best coverage attempts specified. At that point, the graph that was
found to be the "best" based on the coverage type predicate "wins",
and is subsequently saved to disk.
N.B. This option is significantly more CPU intensive than the
--FirstGraphWins mode, however, it is highly probable that it
will find a graph that is better (based on the predicate) than
when in first graph wins mode.
N.B. See also: --TargetNumberOfSolutions, --FixedAttempts.
--SkipMemoryCoverageInFirstGraphWinsMode
Skips calculating memory coverage information when in "first graph wins"
mode. This will result in the corresponding fields in the .csv output
indicating 0.
--SkipGraphVerification
When present, skips the internal graph verification check that ensures
a valid perfect hash solution has been found (i.e. with no collisions
across the entire key set).
--DisableCsvOutputFile
When present, disables writing the .csv output file. This is required
when running multiple instances of the tool against the same output
directory in parallel.
--OmitCsvRowIfTableCreateFailed
When present, omits writing a row in the .csv output file if table
creation fails for a given keys file. Ignored if --DisableCsvOutputFile
is speficied.
--OmitCsvRowIfTableCreateSucceeded
When present, omits writing a row in the .csv output file if table
creation succeeded for a given keys file. Ignored if
--DisableCsvOutputFile is specified.
--IndexOnly
When set, affects the generated C files by defining the C preprocessor
macro CPH_INDEX_ONLY, which results in omitting the compiled perfect
hash routines that deal with the underlying table values array (i.e.
any routine other than Index(); e.g. Insert(), Lookup(), Delete() etc),
as well as the array itself. This results in a size reduction of the
final compiled perfect hash binary. Additionally, only the .dll and
BenchmarkIndex projects will be built, as the BenchmarkFull and Test
projects require access to a table values array. This flag is intended
to be used if you only need the Index() routine and will be managing the
table values array independently.
--UseRwsSectionForTableValues [default]
--DoNotUseRwsSectionForTableValues
When set, tells the linker to use a shared read-write section for the
table values array, e.g.: #pragma comment(linker,"/section:.cphval,rws")
This will result in the table values array being accessible across
multiple processes. Thus, the array will persist as long as one process
maintains an open section (mapping); i.e. keeps the .dll loaded.
--UseNonTemporalAvx2Routines
When set, uses implementations of RtlCopyPages and RtlFillPages that
use non-temporal hints.
--ClampNumberOfEdges
When present, clamps the number of edges to always be equal to the
number of keys, rounded up to a power of two, regardless of the
number of table resizes currently in effect. Normally, when a table
is resized, the number of vertices are doubled, and the number of
edges are set to the number of vertices shifted right once (divided
by two). When this flag is set, the vertex doubling stays the same,
however, the number of edges is always clamped to be equal to the
number of keys rounded up to a power of two. This is a research
option used to evaluate the impact of the number of edges on the
graph solving probability for a given key set. Only applies to
And masking (i.e. not modulus masking).
--UseOriginalSeededHashRoutines
When set, uses the original (slower) seeded hash routines (the ones
that return an HRESULT return code and write the hash value to an
output parameter) -- as opposed to using the newer, faster, "Ex"
version of the hash routines.
N.B. This flag is incompatible with --HashAllKeysFirst.
--HashAllKeysFirst [default]
--DoNotHashAllKeysFirst
When set, changes the graph solving logic such that vertices (i.e.
hash values) are generated for all keys up-front, prior to graph
construction. The hashed keys are stored in a "vertex pair" array.
The page table type and page protection applied to this array can be
further refined by the following flags.
N.B. This flag is incompatible with --UseOriginalSeededHashRoutines.
--EnableWriteCombineForVertexPairs
When set, allocates the memory for the vertex pairs array with
write-combine page protection.
N.B. Only applies when --HashAllKeysFirst is set. Incompatible with
--TryLargePagesForVertexPairs.
--RemoveWriteCombineAfterSuccessfulHashKeys
When set, automatically changes the page protection of the vertex
pairs array (after successful hashing of all keys without any vertex
collisions) from PAGE_READWRITE|PAGE_WRITECOMBINE to PAGE_READONLY.
N.B. Only applies when the flags --EnableWriteCombineForVertexPairs
and --HashAllKeysFirst is set.
--TryLargePagesForVertexPairs
When set, tries to allocate the array for vertex pairs using large
pages.
N.B. Only applies when HashAllKeysFirst is set. Incompatible with
EnableWriteCombineForVertexPairs.
--TryLargePagesForGraphEdgeAndVertexArrays
When set, tries to allocate the edge and vertex arrays used by graphs
during solving using large pages.
--TryLargePagesForGraphTableData
When set, tries to allocate the table data used by graphs during
solving using large pages.
--UsePreviousTableSize
When set, uses any previously-recorded table sizes associated with
the keys file for the given algorithm, hash function and masking type.
N.B. To forcibly delete all previously-recorded table sizes from all
keys in a directory, the following PowerShell snippet can be used:
PS C:\Temp\keys> Get-Item -Path *.keys -Stream *.TableSize | Remove-Item
--IncludeNumberOfTableResizeEventsInOutputPath
When set, incorporates the number of table resize events encountered
whilst searching for a perfect hash solution into the final output
names, e.g.:
C:\Temp\output\KernelBase_2485_1_Chm01_Crc32Rotate_And\...
^
Number of resize events.
--IncludeNumberOfTableElementsInOutputPath
When set, incorporates the number of table elements (i.e. the final
table size) into the output path, e.g.:
C:\Temp\output\KernelBase_2485_16384_Chm01_Crc32Rotate_And\...
^
Number of table elements.
N.B. These two flags can be combined, yielding a path as follows:
C:\Temp\output\KernelBase_2485_1_16384_Chm01_Crc32Rotate_And\...
N.B. It is important to understand how table resize events impact the
behavior of this program if one or both of these flags are present.
Using the example above, the initial path that will be used for
the solution will be:
C:\Temp\output\KernelBase_2485_0_8192_Chm01_Crc32Rotate_And\...
After the maximum number of attempts are reached, a table resize
event occurs; the new path component will be:
C:\Temp\output\KernelBase_2485_1_16384_Chm01_Crc32Rotate_And\...
However, the actual renaming of the directory is not done until
a solution has been found and all files have been written. If
this program is being run repeatedly, then the target directory
will already exist. This complicates things, as, unlike files,
we can't just replace an existing directory with a new one.
There are two ways this could be handled: a) delete all the
existing files under the target path, then delete the directory,
then perform the rename, or b) move the target directory somewhere
else first, preserving the existing contents, then proceed with
the rename.
This program takes the latter approach. The existing directory
will be moved to:
C:\Temp\output\old\KernelBase_1_16384_Chm01_Crc32Rotate_And_2018-11-19-011023-512\...
The timestamp appended to the directory name is derived from the
existing directory's creation time, which should ensure uniqueness.
(In the unlikely situation the target directory already exists in
the old subdirectory, the whole operation is aborted and the table
create routine returns a failure.)
The point of mentioning all of this is the following: when one or
both of these flags are routinely specified, the number of output
files rooted in the output directory's 'old' subdirectory will grow
rapidly, consuming a lot of disk space. Thus, if the old files are
not required, it is recommended to regularly delete them manually.
--RngUseRandomStartSeed
Used in conjunction with --Rng. If present, initializes the random
number generator with a random seed (obtained via the operating system).
If not present, the default seed 0x2019090319811025 will be used.
N.B. If you're benchmarking performance, omit this flag, as starting
from the same default seed is required to get comparable runs.
See Also:
--Rng
--RngSeed
--RngSubsequence
--RngOffset
--TryUseAvx2HashFunction [default]
--DoNotTryUseAvx2HashFunction
When set, tries to use optimized AVX2 routines for hashing keys, if
applicable.
N.B. Only applies when HashAllKeysFirst is set.
--TryUseAvx512HashFunction
When set, tries to use optimized AVX512 routines for hashing keys, if
applicable.
N.B. Only applies when HashAllKeysFirst is set.
--DoNotTryUseAvx2MemoryCoverageFunction
When set, disables automatically using the AVX2 memory coverage
calculation routine when the CPU supports the AVX2 instruction set.
--IncludeKeysInCompiledDll [default]
--DoNotIncludeKeysInCompiledDll
When set, includes the keys in the compiled DLL file. If you want to
benchmark a compiled perfect hash table DLL's index routine against a
normal binary search routine (i.e. IndexBsearch()), you'll need to
supply this flag to ensure the keys get built into the binary. We
don't do this by default as they're not needed for a normal perfect
hash table binary.
--DisableSavingCallbackTableValues
When set, does not attempt to save the runtime table values when running
with a _penter-hooked binary.
--DoNotTryUseHash16Impl
By default, if the following conditions exist, the library will
automatically switch to using the USHORT, 16-bit implementations
of hash functions and assigned table data seamlessly during graph
solving:
- Algorithm is Chm01.
- GraphImpl is 3.
- Number of vertices is <= 65,534 (i.e. MAX_USHORT-1).
This provides significant performance improvements, which is why it's
the default. To disable this behavior, set this flag. This flag is
intended to be used during debugging and performance comparisons when
benchmarking -- you shouldn't need to use it in normal use.
N.B. This only affects the solving graph and table instances; the
compiled perfect hash table generated files will still use the
appropriate USHORT C-types if applicable (number of vertices less
than or equal to 65,534).
Table Compile Flags:
N/A
Table Create Parameters:
--GraphImpl=1|2|3 [default: 3]
Selects the backend version of the graph assignment step. Version 1
matches the original CHM algorithm, version 2 is faster and was derived
from NetBSD's nbperf module, version 3 is even faster and was derived
from additional improvements to NetBSD's nbperf module in 2020.
--ValueSizeInBytes=4|8
Sets the size, in bytes, of the value element that will be stored in the
compiled perfect hash table via Insert(). Defaults to 4 bytes (ULONG).
--MainWorkThreadpoolPriority=<High|Normal|Low> [default: Normal]
--FileWorkThreadpoolPriority=<High|Normal|Low> [default: Normal]
Sets the main work (i.e. the CPU-intensive graph solving) threadpool
priority, or the file work threadpool priority, to the given value.
--AttemptsBeforeTableResize=N [default = 4,294,967,295 ]
Specifies the number of attempts at solving the graph that will be made
before a table resize event will occur (assuming that resize events are
permitted, as per the following flag).
--MaxNumberOfTableResizes=N [default = 5]
Maximum number of table resizes that will be permitted before giving up.
--InitialNumberOfTableResizes=N [default = 0]
Initial number of table resizes to simulate before attempting graph
solving. Each table resize doubles the number of vertices used to
solve the graph, which lowers the keys-to-vertices ratio, which will
improve graph solving probability.
--AutoResizeWhenKeysToEdgesRatioExceeds=D
Supplies a keys to edges ratio that, if exceeded, results in an auto
resize, i.e. the equivalent of --InitialNumberOfTableResizes=1. Valid
values are above 0.0 and less than 1.0. Typical values would be 0.8 to
0.9 depending on the hash function being used.
This will result in much faster solving rates for "nearly full" key sets
(i.e., when the number of keys is very close to a power of two, e.g.
HydrogenWorld-31016.keys).
--BestCoverageAttempts=N
Where N is a positive integer, and represents the number of attempts
that will be made at finding a "best" graph (based on the best coverage
type requested below) before the create table routine returns. For
example, if this value is set to 5, the solving will stop when the 5th
new best graph is found. (A graph is considered a "new best" if its
coverage type predicate (below) is the highest/lowest seen before; see
also, --MaxNumberOfEqualBestGraphs.)
--BestCoverageType=<CoverageType>
Indicates the predicate to determine what constitutes the best graph.
N.B. The terms "best graph" and "best coverage" mean the same thing.
You're either in "first graph wins" mode, or "find best graph"
mode. When the latter applies, we're looking for the best graph,
and that means the one with the winning coverage predicate.
Valid coverage types:
HighestNumberOfEmptyPages
LowestNumberOfEmptyPages
HighestNumberOfEmptyLargePages
LowestNumberOfEmptyLargePages
HighestNumberOfEmptyCacheLines
LowestNumberOfEmptyCacheLines
HighestNumberOfUsedPages
LowestNumberOfUsedPages
HighestNumberOfUsedLargePages
LowestNumberOfUsedLargePages
HighestNumberOfUsedCacheLines
LowestNumberOfUsedCacheLines
HighestMaxGraphTraversalDepth
LowestMaxGraphTraversalDepth
HighestTotalGraphTraversals
LowestTotalGraphTraversals
HighestNumberOfEmptyVertices
LowestNumberOfEmptyVertices
HighestNumberOfCollisionsDuringAssignment
LowestNumberOfCollisionsDuringAssignment
HighestMaxAssignedPerCacheLineCount
LowestMaxAssignedPerCacheLineCount
HighestPredictedNumberOfFilledCacheLines
LowestPredictedNumberOfFilledCacheLines
HighestSlope
LowestSlope
HighestScore
LowestScore
HighestRank
LowestRank
The following predicates must be used in conjunction with --KeysSubset
(only applies to PerfectHashCreate.exe):
HighestMaxAssignedPerCacheLineCountForKeysSubset
LowestMaxAssignedPerCacheLineCountForKeysSubset
HighestNumberOfCacheLinesUsedByKeysSubset
LowestNumberOfCacheLinesUsedByKeysSubset
HighestNumberOfLargePagesUsedByKeysSubset
LowestNumberOfLargePagesUsedByKeysSubset
HighestNumberOfPagesUsedByKeysSubset
LowestNumberOfPagesUsedByKeysSubset
--MaxNumberOfEqualBestGraphs=N
Where N is a positive integer, and represents the number of times an
"equal" best graph is encountered (based on the best coverage type)
before stopping further solving attempts for this graph. For example,
let's say you're using --BestCoverageType=HighestNumberOfEmptyCacheLines
--BestCoverageAttempts=5, and that 4th new best graph encountered had a
value of 8 for this coverage type; subsequent graphs that also have a
value of 8 get classed as an "equal" best graph (as we've already found
one with 8). If we supply --MaxNumberOfEqualBestGraphs=10, then we'll
stop further solving attempts once we see the 10th graph that has 8
empty cache lines.
This parameter is particularly useful for the "highest" predicates that
aren't restricted by page or cache line quantities, e.g.:
HighestMaxGraphTraversalDepth
HighestTotalGraphTraversals
HighestNumberOfCollisionsDuringAssignment
However, it's still useful for all other predicates as a mechanism for
avoiding never solving a graph (because you never hit the Nth best graph
attempt).
--MinNumberOfKeysForFindBestGraph=N [default: 512]
Where N is a positive integer, and represents the minimum number of keys
that need to be present before --FindBestGraph mode is honored (with the
default being 512). (There's very little to no benefit in performing
extensive best graph searches for such small key sets, as there's not
going to be enough variation in assigned value cache line occupancy to
yield runtime performance differences.)
--BestCoverageTargetValue=N
Where N is a floating point number if the best coverage type uses
doubles (i.e., HighestRank, LowestSlope), otherwise, a positive integer.
When provided, graph solving will be stopped if a best graph's coverage
value meets the target value provided by this parameter. The type of
comparison is derived from the coverage type, e.g., if the following
params are provided:
--BestCoverageType=HighestRank --BestCoverageTargetValue=0.5
Then graph solving will stop when a solution is found that has a rank
greater than or equal to 0.5. If LowestRank was specified, the reverse
applies: we'd stop solving as soon as we see a solution with a rank
less than or equal to 0.5.
In bulk create mode, the most useful predicate is rank, as it is a
normalized score between [0.0, 1.0), and a rank of 0.5 or greater is
usually indicative of a tightly-packed assigned table (which we want).
Other predicates use absolute values, which aren't useful in bulk create
context when you have many differing key sizes (e.g. HighestScore and
--BestCoverageTargetValue=200000 does not make sense for bulk create as
a table has to be a certain size in order to achieve that score).
This parameter can be used in conjunction with other parameters like
--FixedAttempts=N or --TargetNumberOfSolutions=N. However, note that
whichever limit is reached first will terminate the solving; i.e. if
you use --BestCoverageType=HighestRank --BestCoverageTargetValue=0.5
and --FixedAttempts=10, then solving will stop after 10 attempts,
regardless of whether or not the target value is reached.
Also note that this behavior, as with all "find best graph" behavior,
is trumped by the logic that skips finding a best graph if there are
less than the minimum number of keys available (default: 512). This
can be altered via --MinNumberOfKeysForFindBestGraph.
In general, this parameter is useful for finding a balance between
solving time and solution quality; some key sets may take a lot of
attempts to break a rank of 0.39-0.40, but in general, most keys (at
least in the venerable sys32 set) will eventually yield tables with
a Rank of 0.5 or greater within a few seconds to a few minutes.
--KeysSubset=N,N+1[,N+2,N+3,...] (e.g. --KeysSubset=10,50,123,600,670)
Supplies a comma-separated list of keys in ascending key-value order.
Must contain two or more elements.
N.B. Only applies to PerfectHashCreate.exe.
--TargetNumberOfSolutions=N
Where N is a positive integer and represents a target number of
solutions to find before stopping graph solving. Typically only useful
for benchmarking.
--FixedAttempts=N
Where N is a positive integer and represents a fixed number of attempts
that will be made (irrespective of whether or not a solution was found)
before graph solving stops. Typically only useful for benchmarking.
--Seeds=<n1,...n8>
Supplies an optional comma-separated list of up to 8 integers that
represent the seed values to use for every graph solving attempt.
Each value may be zero, which tells the algorithm to use a random
seed for this position as per normal.
The logic is also cognizant of the hash function's seed masks, e.g.
MultiplyShiftR has a seed mask of 0x1f1f for seed 3 (which is used to
control the final right shift amount), so, if we use the following:
--Seeds=0,0,0x1000
It will use random bytes for the first two seeds. For the second byte
of the third seed, it'll use 0x10 (as 4096 is 0x1000), but will use a
random byte for the first byte. (If we were to use --Seeds=0,0,16,
then the first byte will be locked to 0x10 and the second byte will
be random.)
This has proven useful for the hash function MultiplyShiftR when using
--InitialNumberOfTableResizes=1 --Seeds=0,0,0x1010 as it forces all
vertices to be constrained to the first half of the assigned array
(thus negating the overhead of a table resize). It may be useful in
other contexts, too.
N.B. Either hex or decimal can be used for the seed values.
--Seed3Byte1MaskCounts=<n1,...n31>
--Seed3Byte2MaskCounts=<n1,...n31>
Supplies a comma-separated list of 32 integers that represent weighted
counts of seed mask's byte values. (Experimental.)
--SolutionsFoundRatio=<double>
--TryUsePredictedAttemptsToLimitMaxConcurrency
Supplies a double (64-bit) floating point number indicating the ratio
of solutions found (obtained from a prior run). This is then used to
calculate the predicted number of attempts required to solve a given
graph; when combined with --TryUsePredictedAttemptsToLimitMaxConcurrency
the maximum concurrency used when solving will be the minimum of the
predicted attempts and the maximum concurrency indicated on the command
line.
N.B. These parameters are typically less useful for bulk-create options
as each table will have different solving characteristics.
--Rng=<RNG name>
Supplies the name of a random number generator to use for obtaining the
random bytes needed as part of graph solving. Valid values:
Philox43210
Uses the Philox 4x32 10-round pseudo-RNG. This is the default.
This should be used when benchmarking creation performance, as
it ensures the random numbers fed to each graph solving attempt
are identical between runs, resulting in consistent runtimes
across subsequent runs. It may result in slower solving times
versus the System RNG, depending on your key set.
System
Uses the standard operating system facilities for obtaining
random data. All other --Rng* parameters are ignored. This
should be used when attempting to find legitimate solutions,
however, due to the inherent randomness, will result in varying
runtimes across subsequent runs.
--RngSeed=<Seed>
Supplies a 64-bit seed used to initialize the RNG. Defaults to
0x2019090319811025, unless --RngUseRandomStartSeed is supplied (in which
case, a random seed will be used, obtained via the operating system).
--RngSubsequence=<Subsequence>
Supplies the initial subsequence used by the RNG. The first graph will
use this sequence, with each additional graph adding 1 to this value for
their subsequence. This ensures parallel graphs generate different
random numbers (even if the seed is identical) when solving. (Defaults
to 0.)
--RngOffset=<Offset>
Supplies the initial offset used by the RNG. (Defaults to 0.)
--Remark="Additional description about run"
Supplies a remark to be associated with the run that will be included
in the .csv output files under the 'Remark' column. An error will
be returned if the provided string contains commas (as this will
break the .csv output).
--MaxSolveTimeInSeconds=<Seconds>
Supplies the maximum number of seconds to try and solve an individual
graph.
--FunctionHookCallbackDllPath=<Path>
Supplies a fully-qualified path to a .dll file that will be used as the
callback handler for hooked functions.
--FunctionHookCallbackFunctionName=<ExportedFunctionName>
Supplies the exported function name to resolve from the callback module
(above) and use as the callback for hooked functions. The default is
InterlockedIncrement.
--FunctionHookCallbackIgnoreRip=<RelativeRIP>
Supplies a relative RIP to ignore during function callback. That is,
if a caller matches the supplied relative RIP, the function callback
will not be executed.
Console Output Character Legend
N.B. You can limit console output to *just* the characters via the --Quiet
command line parameter. (--Silent will disable all console output.)
Char | Meaning
. Table created successfully.
+ Table resize event occured.
x Failed to create a table. The maximum number of attempts at trying to
solve the table at a given size was reached, and no more resize attempts
were possible (due to the maximum resize limit also being hit).
F Failed to create a table due to a target not being reached by a specific
number of attempts.
* None of the worker threads were able to allocate sufficient memory to
attempt solving the graph.
! The system is out of memory.
L The system is running low on memory (a low memory event is triggered
at about 90% RAM usage). In certain situations we can detect this
situation prior to actually running out of memory; in these cases,
we abort the current table creation attempt (which will instantly
relieve system memory pressure).
V The graph was created successfully, however, we weren't able to allocate
enough memory for the table values array in order for the array to be
used after creation. This can be avoided by supplying the command line
parameter --SkipTestAfterCreate.
T The requested number of table elements was too large.
S A shutdown event was received. This shouldn't be seen unless externally
signaling the named shutdown event associated with a context.
t The solve timeout was reached before a solution was found.
? The error code isn't recognized! E-mail [email protected] with details.
Algorithms:
ID | Name
1 Chm01
All Hash Functions:
ID | Name (Number of Seeds)
2 Jenkins (2)
12 Fnv (2)
14 Crc32RotateX (3)
17 RotateMultiplyXorRotate (3)
18 ShiftMultiplyXorShift (3)
19 ShiftMultiplyXorShift2 (6)
20 RotateMultiplyXorRotate2 (6)
21 MultiplyRotateR (3)
22 MultiplyRotateLR (3)
23 MultiplyShiftR (3)
24 MultiplyShiftLR (3)
27 MultiplyRotateRMultiply (5)
28 MultiplyRotateR2 (5)
29 MultiplyShiftRMultiply (5)
30 MultiplyShiftR2 (5)
32 RotateRMultiplyRotateR (3)
33 Multiply643ShiftR (3)
33 Multiply644ShiftR (3)
34 MultiplyShiftRX (3)
Mask Functions:
ID | Name
2 And