P a r s i n g J S O N R e a l l y Q u i c k l y : L e s s o n s L e a - - PowerPoint PPT Presentation

P a r s i n g J S O N R e a l l y Q u i c k l y : Le s s o n s Le a r n e d

H o w fa s t c a n yo u re a d a l a rg e f i l e ?

A re yo u l i m i te d by yo u r d i s k o r A re yo u l i m i te d by yo u r C P U ?

A n i Ma c d i s k : 2 . 2 G B / s , Fa s te r S S Ds ( e .g . , 5 G B / s ) a re ava i l a b l e

Re a d i n g tex t l i n e s ( C P U o n l y ) ~ 0.6 G B / s o n 3. 4 G H z S k y l a ke i n J ava

void parseLine(String s) { volume += s.length(); } void readString(StringReader data) { BufferedReader bf = new BufferedReader(data); bf.lines().forEach(s -> parseLine(s)); }

• 8 2 2 9 0 2 2

Re a d i n g tex t l i n e s ( C P U o n l y )

size_t sum_line_lengths(char * data, size_t length) { std::stringstream is; is.rdbuf()->pubsetbuf(data, length); std::string line; size_t sumofalllinelengths{0}; while(getline(is, line)) { sumofalllinelengths += line.size(); } return sumofalllinelengths; }

J S O N

{"Image": {"Width": 800,"Height": 600, "Title": "View from 15th Floor", "Thumbnail": { "Url": "http://www.example.com/81989943", "Height": 125,"Width": 100} }

t w i t te r . j s o n : 0.3 5 G B / s o n 3. 4 G H z S k y l a ke

• 0. 3 5 G B / s

• 1. 5 G B / s

t w i t te r . j s o n : 0.6 5 0 G B / s o n 3. 4 G H z S k y l a ke

• 0. 6 5 G B / s

• 0. 3 5 G B / s

• 1. 5 G B / s

s i m d j s o n s p e e d ( C+ + )

• 0. 6 5 G B / s

• 0. 3 5 G B / s

• 1. 5 G B / s

2 . 4 G B / s o n a 3. 4 G H z ( + t u r b o ) p ro c e s s o r i s ~ 1. 5 c yc l e s p e r i n p u t by te

T r i c k # 1 : a vo i d h a rd - to - p re d i c t b ra n c h e s

while (howmany != 0) {

• ut[index] = random();

index += 1; howmany--; }

while (howmany != 0) { val = random(); if( val is odd) { // <=== new

• ut[index] = val;

index += 1; } howmany--; }

F ro m 3 c yc l e s to 1 5 c yc l e s p e r va l u e !

G o b ra n c h l e s s !

while (howmany != 0) { val = random();

• ut[index] = val;

index += (val bitand 1); howmany--; }

W h a t i f I ke e p r u n n i n g t h e s a m e b e n c h m a r k ?

T r i c k # 2 : U s e w i d e " wo rd s "

D o n' t p ro c e s s by te by by te

W h e n p o s s i b l e , u s e S I M D

• 5 1 2

T r i c k # 3 : a vo i d m e m o r y / o b j e c t a l l o c a t i o n

T r i c k # 4 : m e a s u re t h e p e r fo r m a n c e !

C o n t i n u o u s I n te g ra t i o n Pe r fo r m a n c e te s t s

P ro c e s s o r f re q u e n c i e s a re n o t c o n s ta n t

S p e c i f i c exa m p l e s

E xa m p l e 1. U T F - 8

if (byte1 < 0x80) { return true; // ASCII } if (byte1 < 0xE0) { if (byte1 < 0xC2 || byte2 > 0xBF) { return false; } } else if (byte1 < 0xF0) { // Three-byte form. if (byte2 > 0xBF || (byte1 == 0xE0 && byte2 < 0xA0) || (byte1 == 0xED && 0xA0 <= byte2) blablabla ) blablabla } else { // Four-byte form. .... blabla }

U s i n g S I M D

_mm256_subs_epu8(current_bytes, 244 );

p ro c e s s i n g ra n d o m U T F - 8

• 0. 5

20 x fa s te r !

E xa m p l e 2 . C l a s s i f y i n g c h a ra c te r s

• t h e r s

A R M N E O N a n d x 6 4 p ro c e s s o r s h ave i n s t r u c t i o n s to l o o k u p 1 6 - by te ta b l e s i n a ve c to r i ze d m a n n e r ( 1 6 va l u e s a t a t i m e ) : p s h u f b, t b l

H1(low(c)) & H2(high(c))

• t h e r s : 0

const uint8x16_t low_nibble_mask = (uint8x16_t){16, 0, 0, 0, 0, 0, 0, 0, 0, 8, 12, 1, 2, 9, 0, 0}; const uint8x16_t high_nibble_mask = (uint8x16_t){8, 0, 18, 4, 0, 1, 0, 1, 0, 0, 0, 3, 2, 1, 0, 0}; const uint8x16_t low_nib_and_mask = vmovq_n_u8(0xf);

uint8x16_t nib_lo = vandq_u8(chunk, low_nib_and_mask); uint8x16_t nib_hi = vshrq_n_u8(chunk, 4); uint8x16_t shuf_lo = vqtbl1q_u8(low_nibble_mask, nib_lo); uint8x16_t shuf_hi = vqtbl1q_u8(high_nibble_mask, nib_hi); return vandq_u8(shuf_lo, shuf_hi);

E xa m p l e 3. D e te c t i n g e s c a p e d c h a ra c te r s

" \ " \ \ \ \ " \ \ \ "

Ca n yo u te l l w h e re t h e s t r i n g s s ta r t a n d e n d ?

{ "\\\"Nam[{": [ 116,"\\\\"

. . . W i t h o u t b ra n c h i n g ?

E s c a p e c h a ra c te r s fo l l ow a n o d d s e q u e n c e o f b a c ks l a s h e s !

{ "\\\"Nam[{": [ 116,"\\\\" ___111________________1111_

1_1_1_1_1_1_1_1_1_1_1_1_1_1

_1_1_1_1_1_1_1_1_1_1_1_1_1_

(((B + (B &~(B << 1)& E))& ~B)& ~E) | (((B + ((B &~(B << 1))& O))& ~B)& E)

{ "\\\"Nam[{": [ 116,"\\\\"

______1____________________

Re m ove t h e e s c a p e d q u o te s , a n d

t h e re m a i n i n g q u o te s te l l yo u w h e re t h e s t r i n g s a re !

{ "\\\"Nam[{": [ 116,"\\\\" __1___1_____1________1____1

______1____________________

__1_________1________1____1

F i n d t h e s p a n o f t h e s t r i n g

mask = quote xor (quote << 1); mask = mask xor (mask << 2); mask = mask xor (mask << 4); mask = mask xor (mask << 8); mask = mask xor (mask << 16); ... __1_________1________1____1

__1111111111_________11111_

E n t i re s t r u c t u re o f t h e J S O N d o c u m e n t c a n b e i d e n t i f i e d ( a s a b i t s e t ) w i t h o u t a ny b ra n c h !

E xa m p l e 4 . D e c o d e b i t i n d exe s

while (word != 0) { result[i] = trailingzeroes(word); word = word & (word - 1); i++; }

while (word != 0) { result[i] = trailingzeroes(word); word = word & (word - 1); result[i+1] = trailingzeroes(word); word = word & (word - 1); result[i+2] = trailingzeroes(word); word = word & (word - 1); result[i+3] = trailingzeroes(word); word = word & (word - 1); i+=4; }

bitCount

E xa m p l e 5 . N u m b e r p a r s i n g i s ex p e n s i ve

strtod

bool is_made_of_eight_digits_fast(const char *chars) { uint64_t val; memcpy(&val, chars, 8); return (((val & 0xF0F0F0F0F0F0F0F0) | (((val + 0x0606060606060606) & 0xF0F0F0F0F0F0F0F0) >> 4)) == 0x3333333333333333); }

uint32_t parse_eight_digits_unrolled(const char *chars) { uint64_t val; memcpy(&val, chars, sizeof(uint64_t)); val = (val & 0x0F0F0F0F0F0F0F0F) * 2561 >> 8; val = (val & 0x00FF00FF00FF00FF) * 6553601 >> 16; return (val & 0x0000FFFF0000FFFF) * 42949672960001 >> 32; }

R u n t i m e d i s p a tc h

int json_parse_dispatch(...) { Architecture best_implementation = find_best_supported_implementation(); // Selecting the best implementation switch (best_implementation) { case Architecture::HASWELL: json_parse_ptr = &json_parse_implementation<Architecture::HASWELL>; break; case Architecture::WESTMERE: json_parse_ptr= &json_parse_implementation<Architecture::WESTMERE>; break; default: return UNEXPECTED_ERROR; } return json_parse_ptr(....); }

• l f, T

• j c i e c h M u ł a , G e o rg i o s F l o ro s , D o n g X i e , Na n X i a o, E g o r

• u te r B o l s te r l e e , A n i s h Ka r a n d i k a r

• m D y s o n , I h o r D o t s e n ko, A l e x e y M i l ov i d ov