Last active
October 22, 2022 10:49
-
-
Save mrbid/9a050ee747a9188bc0aa849385bef865 to your computer and use it in GitHub Desktop.
Benchmarking random float functions with box muller transformation.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
/* | |
James William Fletcher (github.com/mrbid) | |
August 2021 | |
Benchmarking random float functions using | |
the box muller transformation. | |
https://james-william-fletcher.medium.com/benchmarking-random-float-functions-in-c-13b9febb3d5b | |
*/ | |
#include <stdio.h> | |
#include <stdlib.h> | |
#include <stdint.h> | |
#include <math.h> | |
#include <string.h> | |
#include <locale.h> | |
#include <time.h> | |
#include <sys/time.h> | |
#include <unistd.h> | |
#include <fcntl.h> | |
#include <x86intrin.h> | |
#pragma GCC diagnostic ignored "-Wunused-result" | |
#define TEST_SQRTPS 1 | |
#define AVGITER 3000000 | |
#define rand_min -20 | |
#define rand_max 20 | |
//////// RANDOM FLOATS | |
// classic random | |
float rand_float1() | |
{ | |
static const float rmax = (float)RAND_MAX; | |
return (((float)rand())+1e-7f) / rmax; | |
} | |
float rand_float2() | |
{ | |
static const float FLOAT_UINT64_MAX = (float)UINT64_MAX; | |
int f = open("/dev/urandom", O_RDONLY | O_CLOEXEC); | |
uint64_t s = 0; | |
ssize_t result = read(f, &s, sizeof(uint64_t)); | |
close(f); | |
return (((float)s)+1e-7f) / FLOAT_UINT64_MAX; | |
} | |
// https://stackoverflow.com/questions/686353/random-float-number-generation | |
float rand_float3() | |
{ | |
uint_least32_t r = (rand() & 0xffff) + ((rand() & 0x00ff) << 16); | |
return (float)r * 5.9604645E-8f; | |
} | |
float rand_float4() | |
{ | |
uint32_t pattern = 0x3f800000; | |
uint32_t random23 = 0x7fffff & (rand() << 8 ^ rand()); | |
pattern |= random23; | |
char buffer[sizeof(float)]; | |
memcpy(buffer, &pattern, sizeof(float)); | |
float f; | |
memcpy(&f, buffer, sizeof(float)); | |
return f - 1.0f; | |
} | |
// adapted from ogre3d asm_math.h | |
// https://www.flipcode.com/archives/07-15-2002.shtml | |
// https://www.cs.cmu.edu/afs/andrew/scs/cs/oldfiles/15-494-sp09/dst/A/sw/ogre-1.6.4/OgreMain/include/asm_math.h | |
// https://gist.github.com/mrbid/9a050ee747a9188bc0aa849385bef865#file-rand_float_normal_bench-c-L63 | |
float rand_float5(const __int64_t seed) | |
{ | |
static const float FLOAT_MAX = 9223372036854775807.0f; | |
static __int64_t q = 8008135; | |
if(seed != 0) | |
q = seed; | |
__m64 mm0 = _mm_cvtsi64_m64(q); | |
__m64 mm1 = _m_pshufw(mm0, 0x1E); | |
mm0 = _mm_add_pi32(mm0, mm1); | |
q = _m_to_int64(mm0); | |
_m_empty(); | |
return ( fabsf(q+1e-7f) / FLOAT_MAX ); | |
} | |
//////// HELPERS | |
static inline float scale(const float normal, const float min, const float max) | |
{ | |
return ( normal * (max-min) ) + min; | |
} | |
static inline float rsqrtss(float f) | |
{ | |
return _mm_cvtss_f32(_mm_rsqrt_ss(_mm_set_ss(f))); | |
} | |
static inline float sqrtps(float f) | |
{ | |
return _mm_cvtss_f32(_mm_sqrt_ps(_mm_set_ss(f))); | |
} | |
void secure_random_srand() | |
{ | |
unsigned int s = 0; | |
int f = open("/dev/urandom", O_RDONLY | O_CLOEXEC); | |
read(f, &s, sizeof(unsigned int)); | |
close(f); | |
srand(s); | |
} | |
uint64_t secure_random_seed() | |
{ | |
uint64_t s = 0; | |
int f = open("/dev/urandom", O_RDONLY | O_CLOEXEC); | |
read(f, &s, sizeof(uint64_t)); | |
close(f); | |
return s; | |
} | |
//////// BOX MULLERS | |
float rand_normal_float1() | |
{ | |
float u = rand_float1() * 2.f - 1.f; | |
float v = rand_float1() * 2.f - 1.f; | |
float r = u * u + v * v; | |
while(r == 0.f || r > 1.f) | |
{ | |
u = rand_float1() * 2.f - 1.f; | |
v = rand_float1() * 2.f - 1.f; | |
r = u * u + v * v; | |
} | |
return u * sqrtf(-2.f * logf(r) / r); | |
} | |
float rand_normal_float2() | |
{ | |
float u = rand_float2() * 2.f - 1.f; | |
float v = rand_float2() * 2.f - 1.f; | |
float r = u * u + v * v; | |
while(r == 0.f || r > 1.f) | |
{ | |
u = rand_float2() * 2.f - 1.f; | |
v = rand_float2() * 2.f - 1.f; | |
r = u * u + v * v; | |
} | |
return u * sqrtf(-2.f * logf(r) / r); | |
} | |
float rand_normal_float3() | |
{ | |
float u = rand_float3() * 2.f - 1.f; | |
float v = rand_float3() * 2.f - 1.f; | |
float r = u * u + v * v; | |
while(r == 0.f || r > 1.f) | |
{ | |
u = rand_float3() * 2.f - 1.f; | |
v = rand_float3() * 2.f - 1.f; | |
r = u * u + v * v; | |
} | |
return u * sqrtf(-2.f * logf(r) / r); | |
} | |
float rand_normal_float4() | |
{ | |
float u = rand_float4() * 2.f - 1.f; | |
float v = rand_float4() * 2.f - 1.f; | |
float r = u * u + v * v; | |
while(r == 0.f || r > 1.f) | |
{ | |
u = rand_float4() * 2.f - 1.f; | |
v = rand_float4() * 2.f - 1.f; | |
r = u * u + v * v; | |
} | |
return u * sqrtf(-2.f * logf(r) / r); | |
} | |
float rand_normal_float5() | |
{ | |
float u = rand_float5(0) * 2.f - 1.f; | |
float v = rand_float5(0) * 2.f - 1.f; | |
float r = u * u + v * v; | |
while(r == 0.f || r > 1.f) | |
{ | |
u = rand_float5(0) * 2.f - 1.f; | |
v = rand_float5(0) * 2.f - 1.f; | |
r = u * u + v * v; | |
} | |
return u * sqrtf(-2.f * logf(r) / r); | |
} | |
//////// BOX MULLERS -- SQRTPS | |
float rand_normal_float1_sqrtps() | |
{ | |
float u = rand_float1() * 2.f - 1.f; | |
float v = rand_float1() * 2.f - 1.f; | |
float r = u * u + v * v; | |
while(r == 0.f || r > 1.f) | |
{ | |
u = rand_float1() * 2.f - 1.f; | |
v = rand_float1() * 2.f - 1.f; | |
r = u * u + v * v; | |
} | |
return u * sqrtps(-2.f * logf(r) / r); | |
} | |
float rand_normal_float2_sqrtps() | |
{ | |
float u = rand_float2() * 2.f - 1.f; | |
float v = rand_float2() * 2.f - 1.f; | |
float r = u * u + v * v; | |
while(r == 0.f || r > 1.f) | |
{ | |
u = rand_float2() * 2.f - 1.f; | |
v = rand_float2() * 2.f - 1.f; | |
r = u * u + v * v; | |
} | |
return u * sqrtps(-2.f * logf(r) / r); | |
} | |
float rand_normal_float3_sqrtps() | |
{ | |
float u = rand_float3() * 2.f - 1.f; | |
float v = rand_float3() * 2.f - 1.f; | |
float r = u * u + v * v; | |
while(r == 0.f || r > 1.f) | |
{ | |
u = rand_float3() * 2.f - 1.f; | |
v = rand_float3() * 2.f - 1.f; | |
r = u * u + v * v; | |
} | |
return u * sqrtps(-2.f * logf(r) / r); | |
} | |
float rand_normal_float4_sqrtps() | |
{ | |
float u = rand_float4() * 2.f - 1.f; | |
float v = rand_float4() * 2.f - 1.f; | |
float r = u * u + v * v; | |
while(r == 0.f || r > 1.f) | |
{ | |
u = rand_float4() * 2.f - 1.f; | |
v = rand_float4() * 2.f - 1.f; | |
r = u * u + v * v; | |
} | |
return u * sqrtps(-2.f * logf(r) / r); | |
} | |
float rand_normal_float5_sqrtps() | |
{ | |
float u = rand_float5(0) * 2.f - 1.f; | |
float v = rand_float5(0) * 2.f - 1.f; | |
float r = u * u + v * v; | |
while(r == 0.f || r > 1.f) | |
{ | |
u = rand_float5(0) * 2.f - 1.f; | |
v = rand_float5(0) * 2.f - 1.f; | |
r = u * u + v * v; | |
} | |
return u * sqrtps(-2.f * logf(r) / r); | |
} | |
int main() | |
{ | |
secure_random_srand(); | |
rand_float5(secure_random_seed()); | |
printf("\n"); | |
setlocale(LC_NUMERIC, ""); | |
float ret = 0.f; | |
uint64_t st = 0, et = 0, avg = 0; | |
avg = 0; | |
for(int i = 0; i < AVGITER; i++) | |
{ | |
st = __rdtsc(); | |
ret += rand_normal_float1(); | |
avg += __rdtsc()-st; | |
} | |
printf("rand_normal_float1() Cycles: %'lu\n", avg / AVGITER); | |
printf("0-1: %.20f\n", rand_float1()); | |
printf("norm: %.20f\n\n", rand_normal_float1()); | |
#if TEST_SQRTPS == 1 | |
avg = 0; | |
for(int i = 0; i < AVGITER; i++) | |
{ | |
st = __rdtsc(); | |
ret += rand_normal_float1_sqrtps(); | |
avg += __rdtsc()-st; | |
} | |
printf("rand_normal_float1_sqrtps() Cycles: %'lu\n", avg / AVGITER); | |
printf("0-1: %.20f\n", rand_float1()); | |
printf("norm: %.20f\n\n", rand_normal_float1_sqrtps()); | |
printf("\n--------\n\n"); | |
#endif | |
//////// | |
avg = 0; | |
for(int i = 0; i < AVGITER; i++) | |
{ | |
st = __rdtsc(); | |
ret += rand_normal_float2(); | |
avg += __rdtsc()-st; | |
} | |
printf("rand_normal_float2() Cycles: %'lu\n", avg / AVGITER); | |
printf("0-1: %.20f\n", rand_float2()); | |
printf("norm: %.20f\n\n", rand_normal_float2()); | |
#if TEST_SQRTPS == 1 | |
avg = 0; | |
for(int i = 0; i < AVGITER; i++) | |
{ | |
st = __rdtsc(); | |
ret += rand_normal_float2_sqrtps(); | |
avg += __rdtsc()-st; | |
} | |
printf("rand_normal_float2_sqrtps() Cycles: %'lu\n", avg / AVGITER); | |
printf("0-1: %.20f\n", rand_float2()); | |
printf("norm: %.20f\n\n", rand_normal_float2_sqrtps()); | |
printf("\n--------\n\n"); | |
#endif | |
//////// | |
avg = 0; | |
for(int i = 0; i < AVGITER; i++) | |
{ | |
st = __rdtsc(); | |
ret += rand_normal_float3(); | |
avg += __rdtsc()-st; | |
} | |
printf("rand_normal_float3() Cycles: %'lu\n", avg / AVGITER); | |
printf("0-1: %.20f\n", rand_float3()); | |
printf("norm: %.20f\n\n", rand_normal_float3()); | |
#if TEST_SQRTPS == 1 | |
avg = 0; | |
for(int i = 0; i < AVGITER; i++) | |
{ | |
st = __rdtsc(); | |
ret += rand_normal_float3_sqrtps(); | |
avg += __rdtsc()-st; | |
} | |
printf("rand_normal_float3_sqrtps() Cycles: %'lu\n", avg / AVGITER); | |
printf("0-1: %.20f\n", rand_float3()); | |
printf("norm: %.20f\n\n", rand_normal_float3_sqrtps()); | |
printf("\n--------\n\n"); | |
#endif | |
//////// | |
avg = 0; | |
for(int i = 0; i < AVGITER; i++) | |
{ | |
st = __rdtsc(); | |
ret += rand_normal_float4(); | |
avg += __rdtsc()-st; | |
} | |
printf("rand_normal_float4() Cycles: %'lu\n", avg / AVGITER); | |
printf("0-1: %.20f\n", rand_float4()); | |
printf("norm: %.20f\n\n", rand_normal_float4()); | |
#if TEST_SQRTPS == 1 | |
avg = 0; | |
for(int i = 0; i < AVGITER; i++) | |
{ | |
st = __rdtsc(); | |
ret += rand_normal_float4_sqrtps(); | |
avg += __rdtsc()-st; | |
} | |
printf("rand_normal_float4_sqrtps() Cycles: %'lu\n", avg / AVGITER); | |
printf("0-1: %.20f\n", rand_float4()); | |
printf("norm: %.20f\n\n", rand_normal_float4_sqrtps()); | |
printf("\n--------\n\n"); | |
#endif | |
//////// | |
avg = 0; | |
for(int i = 0; i < AVGITER; i++) | |
{ | |
st = __rdtsc(); | |
ret += rand_normal_float5(); | |
avg += __rdtsc()-st; | |
} | |
printf("rand_normal_float5() Cycles: %'lu\n", avg / AVGITER); | |
printf("0-1: %.20f\n", rand_float5(0)); | |
printf("norm: %.20f\n\n", rand_normal_float5()); | |
#if TEST_SQRTPS == 1 | |
avg = 0; | |
for(int i = 0; i < AVGITER; i++) | |
{ | |
st = __rdtsc(); | |
ret += rand_normal_float5_sqrtps(); | |
avg += __rdtsc()-st; | |
} | |
printf("rand_normal_float5_sqrtps() Cycles: %'lu\n", avg / AVGITER); | |
printf("0-1: %.20f\n", rand_float5(0)); | |
printf("norm: %.20f\n\n", rand_normal_float5_sqrtps()); | |
#endif | |
//////// | |
// done | |
printf("%c\n", (char)ret); // forces the compiler to not disregard the functions we are testing | |
return 0; | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment