-
Notifications
You must be signed in to change notification settings - Fork 0
Dev Notes: 2017_01_27
Improving the performance of the code clearly means reducing some of the memory management overheads.
The use of std::vector has been a great way to start things, and it's always a good idea to avoid explicit uses of new and delete, but in this case any small overheads are a problem because we tend to create a lot of objects. In addition, being able to use a customized approach will allow alternative approaches that will significantly optimize things for smaller numbers. Using an array-based approach also means being able to directly access result digit arrays.
All of the current operations are 3-operand forms: A = B op C, in which B and C are constant, but this means that we have to allocate a new object, A, for the result. We have 2-operand forms, A op= B, but they're expressed in terms of the 3-operand variants. Defining explicit versions allows us to avoid this extra overhead.
After the initial set of changes we get the following results for the natural_check test:
cons 0 | 316 | pass | 0
cons 1 | 376 | pass | 123456789abc
cons 2 | 604 | pass | 0
cons 3 | 17320 | pass | 3837439787487386792386728abcd88379dc
cons 4 | 34236 | pass | 3897894117580750151618270927682762897697428275427542907478758957487582700682675349287325097
cons 5 | 5356 | pass | 115415157637671751
cons 6 | 5453013 | pass | exception thrown: invalid digit
cons 7 | 5651 | pass | 100000000000000000000000
count 0 | 66 | pass | 0
count 1 | 48 | pass | 64
count 2 | 48 | pass | 17
count 3 | 67 | pass | 185
add 0 | 224 | pass | 73
add 1 | 179 | pass | 42
add 2 | 206 | pass | 10000000000000001
add 3 | 269 | pass | 98888880000000000000000000000000000000000000000000000000000001000000789
sub 0 | 354 | pass | 50
sub 1 | 324 | pass | 5872488729698595999749602411500766185722239445613509099777952305512191704320129156897500143
sub 2 | 342 | pass | 897
sub 3 | 16131 | pass | exception thrown: not a number
comp 0a | 324 | pass | 0
comp 0b | 70 | pass | 1
comp 0c | 76 | pass | 1
comp 0d | 67 | pass | 1
comp 0e | 48 | pass | 0
comp 0f | 45 | pass | 0
comp 1a | 94 | pass | 0
comp 1b | 85 | pass | 1
comp 1c | 39 | pass | 1
comp 1d | 58 | pass | 1
comp 1e | 58 | pass | 0
comp 1f | 39 | pass | 0
comp 2a | 100 | pass | 0
comp 2b | 106 | pass | 1
comp 2c | 39 | pass | 0
comp 2d | 42 | pass | 0
comp 2e | 39 | pass | 1
comp 2f | 58 | pass | 1
comp 3a | 191 | pass | 1
comp 3b | 106 | pass | 0
comp 3c | 106 | pass | 0
comp 3d | 82 | pass | 1
comp 3e | 82 | pass | 0
comp 3f | 82 | pass | 1
lsh 0a | 376 | pass | 349f
lsh 0b | 163 | pass | 693e
lsh 0c | 118 | pass | d27c0000
lsh 0d | 221 | pass | 1a4f80000000000000000000000000000000000000000000000
lsh 1a | 293 | pass | 693e5306ea64730b5f79d306f250f30f13bffdffdd30ecf0d0ecf0ceceacac400000000000000000
rsh 0a | 272 | pass | 23490000000000000000000000000000000000000000000000000000
rsh 0b | 136 | pass | 11a48000000000000000000000000000000000000000000000000000
rsh 0c | 139 | pass | 469200000000000000000000000000000000000000000000000
rsh 0d | 148 | pass | 11a4800
rsh 1a | 158 | pass | d27ca60dd4c8e616bef3a60de4a1e61e277ffbffba61d9e1a
mul 0 | 161 | pass | 66
mul 1 | 291 | pass | 9999999999999999999000000000000000000
mul 2 | 236 | pass | 8000000000000000000000000000000
mul 3 | 1165 | pass | 15241578753238836750495351562566681945008382873376009755225118122311263526910001371743100137174310012193273126047859425087639153757049236500533455762536198787501905199875019052100
mul 4 | 281 | pass | abcdef1200000000abcdef120000000abcdef1200000000
div 0a | 2755 | pass | 10
div 0b | 2755 | pass | 10
div 1a | 8951 | pass | 78292387927518758972102054472775487212767983201652300846
div 1b | 8951 | pass | 35600667362958008
div 2a | 6145 | pass | ffffffffffffffff000000000000000
div 2b | 6145 | pass | 100000000000000000000000
div 3 | 21141 | pass | exception thrown: divide by zero
div 4a | 2770 | pass | 100000
div 4b | 2770 | pass | 0
div 5a | 2982 | pass | 10000000000000001000000000000000100000000
div 5b | 2982 | pass | 0
gcd 0 | 6054 | pass | 8
gcd 1 | 31539 | pass | 1
gcd 2 | 1429 | pass | 8888888
gcd 3 | 6336 | pass | 20181732873032947492728336135378088830674353623374417329043358630878748833567
toull 0 | 103 | pass | 0
toull 1 | 66 | pass | 2000
toull 2 | 10655 | pass | exception thrown: overflow error
toull 3 | 85 | pass | 123456789a
toull 4 | 9526 | pass | exception thrown: overflow error
prn 0 | 32779 | pass | 4701397401952099592073
prn 1 | 25515 | pass | fedcfedc0123456789
prn 2 | 25197 | pass | FEDCFEDC0123456789
prn 3 | 34139 | pass | 775563766700044321263611
prn 4 | 31665 | pass | 4701397401952099592073
prn 5 | 25760 | pass | 0xfedcfedc0123456789
prn 6 | 25333 | pass | 0XFEDCFEDC0123456789
prn 7 | 33812 | pass | 0775563766700044321263611
The numbers for divide-heavy operations (some of the constructors, and print routines are included) are anywhere from 8% to 20% faster.