Skip to content

Commit

Permalink
Merge d1d847a into 027a073
Browse files Browse the repository at this point in the history
  • Loading branch information
@bashtage
bashtage committed May 26, 2019
2 parents 027a073 + d1d847a commit f749835
Show file tree
Hide file tree
Showing 7 changed files with 119 additions and 95 deletions.
67 changes: 39 additions & 28 deletions benchmark.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -27,8 +27,8 @@
else:
scale_64 = 2

PRNGS = ['DSFMT', 'PCG64', 'PCG32', 'MT19937', 'Xoroshiro128', 'Xorshift1024',
'Xoshiro256', 'Xoshiro512', 'Philox', 'ThreeFry',
PRNGS = ['DSFMT', 'PCG64', 'PCG32', 'MT19937', 'MT64', 'Xoroshiro128',
'Xorshift1024', 'Xoshiro256', 'Xoshiro512', 'Philox', 'ThreeFry',
'ThreeFry32', 'numpy']


Expand Down Expand Up @@ -109,9 +109,11 @@ def timer_bounded(bits=8, max=95, use_masked=True):
minimum = 0

if use_masked: # Use masking & rejection.
command = 'rg.randint({min}, {max}+1, 1000000, dtype=np.uint{bits}, use_masked=True)'
command = 'rg.randint({min}, {max}+1, 1000000, dtype=np.uint{bits},' \
' use_masked=True)'
else: # Use Lemire's algo.
command = 'rg.randint({min}, {max}+1, 1000000, dtype=np.uint{bits}, use_masked=False)'
command = 'rg.randint({min}, {max}+1, 1000000, dtype=np.uint{bits},' \
' use_masked=False)'

command = command.format(min=minimum, max=max, bits=bits)

Expand All @@ -120,7 +122,8 @@ def timer_bounded(bits=8, max=95, use_masked=True):

run_timer(command, command_numpy, SETUP,
'{bits}-bit bounded unsigned integers (max={max}, '
'use_masked={use_masked})'.format(max=max, use_masked=use_masked, bits=bits))
'use_masked={use_masked})'.format(max=max, use_masked=use_masked,
bits=bits))


def timer_32bit():
Expand Down Expand Up @@ -152,9 +155,9 @@ def timer_normal_zig():

parser = argparse.ArgumentParser()
parser.add_argument('-f', '--full',
help='Run benchmarks for a wide range of distributions.'
' If not provided, only tests the production of '
'uniform values.',
help='Run benchmarks for a wide range of '
'distributions. If not provided, only tests the '
'production of uniform values.',
dest='full', action='store_true')
parser.add_argument('-bi', '--bounded-ints',
help='Included benchmark coverage of the bounded '
Expand All @@ -165,31 +168,39 @@ def timer_normal_zig():
timer_uniform()
if args.full:
timer_raw()
timer_32bit()
timer_64bit()
timer_normal_zig()

if args.bounded_ints:
timer_bounded(use_masked=True)
timer_bounded(max=64, use_masked=False) # Worst case for Numpy.
timer_bounded(max=95, use_masked=False) # Typ. avrg. case for Numpy.
timer_bounded(max=127, use_masked=False) # Best case for Numpy.
# Worst case for Numpy.
timer_bounded(max=64, use_masked=False)
# Typ. avrg. case for Numpy.
timer_bounded(max=95, use_masked=False)
# Best case for Numpy.
timer_bounded(max=127, use_masked=False)

timer_bounded(16, use_masked=True)
timer_bounded(16, max=1024, use_masked=False) # Worst case for Numpy.
timer_bounded(16, max=1535, use_masked=False) # Typ. avrg. case for Numpy.
timer_bounded(16, max=2047, use_masked=False) # Best case for Numpy.

timer_32bit()
# Worst case for Numpy.
timer_bounded(16, max=1024, use_masked=False)
# Typ. avrg. case for Numpy.
timer_bounded(16, max=1535, use_masked=False)
# Best case for Numpy.
timer_bounded(16, max=2047, use_masked=False)

if args.bounded_ints:
timer_bounded(32, use_masked=True)
timer_bounded(32, max=1024, use_masked=False) # Worst case for Numpy.
timer_bounded(32, max=1535, use_masked=False) # Typ. avrg. case for Numpy.
timer_bounded(32, max=2047, use_masked=False) # Best case for Numpy.

timer_64bit()
# Worst case for Numpy.
timer_bounded(32, max=1024, use_masked=False)
# Typ. avrg. case for Numpy.
timer_bounded(32, max=1535, use_masked=False)
# Best case for Numpy.
timer_bounded(32, max=2047, use_masked=False)

if args.bounded_ints:
timer_bounded(64, use_masked=True)
timer_bounded(64, max=1024, use_masked=False) # Worst case for Numpy.
timer_bounded(64, max=1535, use_masked=False) # Typ. avrg. case for Numpy.
timer_bounded(64, max=2047, use_masked=False) # Best case for Numpy.

timer_normal_zig()
# Worst case for Numpy.
timer_bounded(64, max=1024, use_masked=False)
# Typ. avrg. case for Numpy.
timer_bounded(64, max=1535, use_masked=False)
# Best case for Numpy.
timer_bounded(64, max=2047, use_masked=False)
32 changes: 18 additions & 14 deletions doc/source/performance.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -12,18 +12,17 @@
SIZE = 25000
PRNGS = [DSFMT, MT19937, MT64, Philox, PCG64, ThreeFry, Xoshiro256, Xoshiro512]

funcs = {'32-bit Unsigned Ints':
f'integers(2**32, dtype="uint32", size={SIZE})',
'64-bit Unsigned Ints':
f'integers(2**64, dtype="uint64", size={SIZE})',
'Uniforms': f'random(size={SIZE})',
'Complex Normals': f'complex_normal(size={SIZE})',
'Normals': f'standard_normal(size={SIZE})',
'Exponentials': f'standard_exponential(size={SIZE})',
'Gammas': f'standard_gamma(3.0,size={SIZE})',
'Binomials': f'binomial(9, .1, size={SIZE})',
'Laplaces': f'laplace(size={SIZE})',
'Poissons': f'poisson(3.0, size={SIZE})', }
funcs = OrderedDict()
funcs['32-bit Unsigned Ints'] = f'integers(2**32, dtype="uint32", size={SIZE})'
funcs['64-bit Unsigned Ints'] = f'integers(2**64, dtype="uint64", size={SIZE})'
funcs['Uniform'] = f'random(size={SIZE})'
funcs['Exponential'] = f'standard_exponential(size={SIZE})'
funcs['Normal'] = f'standard_normal(size={SIZE})'
funcs['Gamma'] = f'standard_gamma(3.0,size={SIZE})'
funcs['Complex Normal'] = f'complex_normal(size={SIZE})'
funcs['Binomial'] = f'binomial(9, .1, size={SIZE})'
funcs['Laplace'] = f'laplace(size={SIZE})'
funcs['Poisson'] = f'poisson(3.0, size={SIZE})'

setup = """
from randomgen import {prng}, Generator
Expand Down Expand Up @@ -73,11 +72,16 @@
table = table.T
table = table.reindex(order)
table = table.T
table.pop('Xoshiro512')
table.pop('DSFMT')
table = 1000000 * table / (SIZE * NUMBER)
print(table.to_csv(float_format='%0.1f'))

try:
from tabulate import tabulate

print(tabulate(table, headers='keys', tablefmt='psql'))
except ImportError:
pass

rel = table.loc[:, ['NumPy']].values @ np.ones((1, table.shape[1])) / table
rel.pop('NumPy')
rel = rel.T
Expand Down
6 changes: 3 additions & 3 deletions randomgen/src/mt19937/mt19937.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -98,10 +98,10 @@ void mt19937_gen(mt19937_state *state) {
state->pos = 0;
}

extern inline uint64_t mt19937_next64(mt19937_state *state);
extern INLINE uint64_t mt19937_next64(mt19937_state *state);

extern inline uint32_t mt19937_next32(mt19937_state *state);
extern INLINE uint32_t mt19937_next32(mt19937_state *state);

extern inline double mt19937_next_double(mt19937_state *state);
extern INLINE double mt19937_next_double(mt19937_state *state);

void mt19937_jump(mt19937_state *state) { mt19937_jump_state(state, poly); }
22 changes: 11 additions & 11 deletions randomgen/src/mt19937/mt19937.h
View file
Original file line number Diff line number Diff line change
@@ -1,17 +1,17 @@
#pragma once
#include <math.h>

#ifdef _WIN32
#if _MSC_VER == 1500
#include "../common/stdint.h"
#include "../common/inttypes.h"
#define INLINE __forceinline
#else
#include <stdint.h>
#include <inttypes.h>
#define INLINE __inline __forceinline
#endif
#else
#include <stdint.h>
#endif

#ifdef _WIN32
#define inline __forceinline
#include <inttypes.h>
#define INLINE inline
#endif

#define RK_STATE_LEN 624
Expand All @@ -32,7 +32,7 @@ extern void mt19937_seed(mt19937_state *state, uint32_t seed);
extern void mt19937_gen(mt19937_state *state);

/* Slightly optimized reference implementation of the Mersenne Twister */
static inline uint32_t mt19937_next(mt19937_state *state) {
static INLINE uint32_t mt19937_next(mt19937_state *state) {
uint32_t y;

if (state->pos == RK_STATE_LEN) {
Expand All @@ -53,15 +53,15 @@ static inline uint32_t mt19937_next(mt19937_state *state) {
extern void mt19937_init_by_array(mt19937_state *state, uint32_t *init_key,
int key_length);

static inline uint64_t mt19937_next64(mt19937_state *state) {
static INLINE uint64_t mt19937_next64(mt19937_state *state) {
return (uint64_t)mt19937_next(state) << 32 | mt19937_next(state);
}

static inline uint32_t mt19937_next32(mt19937_state *state) {
static INLINE uint32_t mt19937_next32(mt19937_state *state) {
return mt19937_next(state);
}

static inline double mt19937_next_double(mt19937_state *state) {
static INLINE double mt19937_next_double(mt19937_state *state) {
int32_t a = mt19937_next(state) >> 5, b = mt19937_next(state) >> 6;
return (a * 67108864.0 + b) / 9007199254740992.0;
}
Expand Down
16 changes: 8 additions & 8 deletions randomgen/src/pcg64/pcg64.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -24,17 +24,17 @@

#include "pcg64.h"

extern inline void pcg_setseq_128_step_r(pcg_state_setseq_128 *rng);
extern inline uint64_t pcg_output_xsl_rr_128_64(pcg128_t state);
extern inline void pcg_setseq_128_srandom_r(pcg_state_setseq_128 *rng,
extern INLINE void pcg_setseq_128_step_r(pcg_state_setseq_128 *rng);
extern INLINE uint64_t pcg_output_xsl_rr_128_64(pcg128_t state);
extern INLINE void pcg_setseq_128_srandom_r(pcg_state_setseq_128 *rng,
pcg128_t initstate,
pcg128_t initseq);
extern inline uint64_t
extern INLINE uint64_t
pcg_setseq_128_xsl_rr_64_random_r(pcg_state_setseq_128 *rng);
extern inline uint64_t
extern INLINE uint64_t
pcg_setseq_128_xsl_rr_64_boundedrand_r(pcg_state_setseq_128 *rng,
uint64_t bound);
extern inline void pcg_setseq_128_advance_r(pcg_state_setseq_128 *rng,
extern INLINE void pcg_setseq_128_advance_r(pcg_state_setseq_128 *rng,
pcg128_t delta);

/* Multi-step advance functions (jump-ahead, jump-back)
Expand Down Expand Up @@ -89,8 +89,8 @@ pcg128_t pcg_advance_lcg_128(pcg128_t state, pcg128_t delta, pcg128_t cur_mult,

#endif

extern inline uint64_t pcg64_next64(pcg64_state *state);
extern inline uint32_t pcg64_next32(pcg64_state *state);
extern INLINE uint64_t pcg64_next64(pcg64_state *state);
extern INLINE uint32_t pcg64_next32(pcg64_state *state);

extern void pcg64_advance(pcg64_state *state, uint64_t *step) {
pcg128_t delta;
Expand Down

0 comments on commit f749835

Please sign in to comment.