update squiggle.c with speedup after avoiding conflicting cache hits

This commit is contained in:
NunoSempere 2024-01-12 00:25:58 +01:00
parent fb2c471172
commit b208879e45
7 changed files with 112 additions and 95 deletions

View File

@ -25,7 +25,7 @@ The name of this repository is a pun on two meanings of "time to": "how much tim
| Language | Time | Lines of code |
|-----------------------------|-----------|---------------|
| C | 5.6ms | 252 |
| squiggle.c | 12.7ms | 29* |
| squiggle.c | 10.5ms | 29* |
| Nim | 40.8ms | 84 |
| Lua (LuaJIT) | 69.9ms | 82 |
| OCaml (flambda) | 187.9ms | 123 |

Binary file not shown.

View File

@ -9,8 +9,8 @@ int main()
double p_b = 0.5;
double p_c = p_a * p_b;
double sample_0(uint64_t * seed) { return 0; }
double sample_1(uint64_t * seed) { return 1; }
double sample_0(uint64_t * seed) { UNUSED(seed); return 0; }
double sample_1(uint64_t * seed) { UNUSED(seed); return 1; }
double sample_few(uint64_t * seed) { return sample_to(1, 3, seed); }
double sample_many(uint64_t * seed) { return sample_to(2, 10, seed); }
@ -22,8 +22,8 @@ int main()
return sample_mixture(samplers, weights, n_dists, seed);
}
int n_samples = 1000000, n_threads = 16;
double* results = malloc(n_samples * sizeof(double));
int n_samples = 1000 * 1000, n_threads = 16;
double* results = malloc((size_t)n_samples * sizeof(double));
sampler_parallel(sampler_result, results, n_threads, n_samples);
printf("Avg: %f\n", array_sum(results, n_samples) / n_samples);
free(results);

View File

@ -3,12 +3,14 @@
#include <stdint.h>
#include <stdlib.h>
// math constants
// Defs
#define PI 3.14159265358979323846 // M_PI in gcc gnu99
#define NORMAL90CONFIDENCE 1.6448536269514727
#define UNUSED(x) (void)(x)
// ^ https://stackoverflow.com/questions/3599160/how-can-i-suppress-unused-parameter-warnings-in-c
// Pseudo Random number generator
static uint64_t xorshift32(uint32_t* seed)
// Pseudo Random number generators
static uint64_t xorshift64(uint64_t* seed)
{
// Algorithm "xor" from p. 4 of Marsaglia, "Xorshift RNGs"
// See:
@ -19,19 +21,20 @@ static uint64_t xorshift32(uint32_t* seed)
// <https://prng.di.unimi.it/>
uint64_t x = *seed;
x ^= x << 13;
x ^= x >> 17;
x ^= x << 5;
return *seed = x;
}
static uint64_t xorshift64(uint64_t* seed)
{
// same as above, but for generating doubles instead of floats
uint64_t x = *seed;
x ^= x << 13;
x ^= x >> 7;
x ^= x << 17;
return *seed = x;
/*
// if one wanted to generate 32 bit ints,
// from which to generate floats,
// one could do the following:
uint32_t x = *seed;
x ^= x << 13;
x ^= x >> 17;
x ^= x << 5;
return *seed = x;
*/
}
// Distribution & sampling functions
@ -47,7 +50,7 @@ double sample_unit_normal(uint64_t* seed)
// // See: <https://en.wikipedia.org/wiki/Box%E2%80%93Muller_transform>
double u1 = sample_unit_uniform(seed);
double u2 = sample_unit_uniform(seed);
double z = sqrtf(-2.0 * log(u1)) * sin(2 * PI * u2);
double z = sqrt(-2.0 * log(u1)) * sin(2 * PI * u2);
return z;
}
@ -67,7 +70,7 @@ double sample_lognormal(double logmean, double logstd, uint64_t* seed)
return exp(sample_normal(logmean, logstd, seed));
}
inline double sample_normal_from_90_confidence_interval(double low, double high, uint64_t* seed)
double sample_normal_from_90_ci(double low, double high, uint64_t* seed)
{
// Explanation of key idea:
// 1. We know that the 90% confidence interval of the unit normal is
@ -98,10 +101,10 @@ double sample_to(double low, double high, uint64_t* seed)
// returns a sample from a lognorma with a matching 90% c.i.
// Key idea: If we want a lognormal with 90% confidence interval [a, b]
// we need but get a normal with 90% confidence interval [log(a), log(b)].
// Then see code for sample_normal_from_90_confidence_interval
double loglow = logf(low);
double loghigh = logf(high);
return exp(sample_normal_from_90_confidence_interval(loglow, loghigh, seed));
// Then see code for sample_normal_from_90_ci
double loglow = log(low);
double loghigh = log(high);
return exp(sample_normal_from_90_ci(loglow, loghigh, seed));
}
double sample_gamma(double alpha, uint64_t* seed)
@ -201,7 +204,7 @@ double sample_mixture(double (*samplers[])(uint64_t*), double* weights, int n_di
{
// Sample from samples with frequency proportional to their weights.
double sum_weights = array_sum(weights, n_dists);
double* cumsummed_normalized_weights = (double*)malloc(n_dists * sizeof(double));
double* cumsummed_normalized_weights = (double*)malloc((size_t)n_dists * sizeof(double));
cumsummed_normalized_weights[0] = weights[0] / sum_weights;
for (int i = 1; i < n_dists; i++) {
cumsummed_normalized_weights[i] = cumsummed_normalized_weights[i - 1] + weights[i] / sum_weights;

View File

@ -15,6 +15,7 @@ double sample_unit_normal(uint64_t* seed);
double sample_uniform(double start, double end, uint64_t* seed);
double sample_normal(double mean, double sigma, uint64_t* seed);
double sample_lognormal(double logmean, double logsigma, uint64_t* seed);
double sample_normal_from_90_ci(double low, double high, uint64_t* seed);
double sample_to(double low, double high, uint64_t* seed);
double sample_gamma(double alpha, uint64_t* seed);
@ -30,4 +31,7 @@ double array_std(double* array, int length);
// Mixture function
double sample_mixture(double (*samplers[])(uint64_t*), double* weights, int n_dists, uint64_t* seed);
// Macro to mute "unused variable" warning when -Wall -Wextra is enabled. Useful for nested functions
#define UNUSED(x) (void)(x)
#endif

View File

@ -6,19 +6,27 @@
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h> // memcpy
/* Parallel sampler */
#define CACHE_LINE_SIZE 64
typedef struct seed_cache_box_t {
uint64_t* seed;
char padding[CACHE_LINE_SIZE - sizeof(uint64_t*)];
} seed_cache_box;
// This avoid false sharing. Dealing with this shaves ~2ms.
void sampler_parallel(double (*sampler)(uint64_t* seed), double* results, int n_threads, int n_samples)
{
// Division terminology:
// a = b * quotient + reminder
// a = (a/b)*b + (a%b)
// Terms of the division:
// a = b * quotient + reminder
// a = b * (a/b) + (a%b)
// dividend: a
// divisor: b
// quotient = a / b
// reminder = a % b
// "divisor's multiple" := (a/b)*b
// quotient = a/b
// reminder = a%b
// "divisor's multiple" := b*(a/b)
// now, we have n_samples and n_threads
// to make our life easy, each thread will have a number of samples of: a/b (quotient)
@ -26,18 +34,17 @@ void sampler_parallel(double (*sampler)(uint64_t* seed), double* results, int n_
// to possibly do by Jorge: improve so that the remainder is included in the threads
int quotient = n_samples / n_threads;
/* int remainder = n_samples % n_threads; // not used, comment to avoid lint warning */
int divisor_multiple = quotient * n_threads;
uint64_t** seeds = malloc(n_threads * sizeof(uint64_t*));
// printf("UINT64_MAX: %lu\n", UINT64_MAX);
// uint64_t** seeds = malloc((size_t)n_threads * sizeof(uint64_t*));
seed_cache_box* cache_box = (seed_cache_box*) malloc(sizeof(seed_cache_box) * (size_t)n_threads);
srand(1);
for (uint64_t i = 0; i < n_threads; i++) {
seeds[i] = malloc(sizeof(uint64_t));
for (int i = 0; i < n_threads; i++) {
cache_box[i].seed = malloc(sizeof(uint64_t*));
// Constraints:
// - xorshift can't start with 0
// - the seeds should be reasonably separated and not correlated
*seeds[i] = (uint64_t)rand() * (UINT64_MAX / RAND_MAX);
*(cache_box[i].seed) = (uint64_t)rand() * (UINT64_MAX / RAND_MAX);
// printf("#%ld: %lu\n",i, *seeds[i]);
// Other initializations tried:
@ -47,33 +54,29 @@ void sampler_parallel(double (*sampler)(uint64_t* seed), double* results, int n_
}
int i;
#pragma omp parallel private(i)
#pragma omp parallel private(i, quotient)
{
#pragma omp for
for (i = 0; i < n_threads; i++) {
int lower_bound_inclusive = i * quotient;
int upper_bound_not_inclusive = ((i + 1) * quotient); // note the < in the for loop below,
// printf("Lower bound: %d, upper bound: %d\n", lower_bound, upper_bound);
for (int j = lower_bound_inclusive; j < upper_bound_not_inclusive; j++) {
results[j] = sampler(seeds[i]);
results[j] = sampler(cache_box[i].seed);
}
}
}
for (int j = divisor_multiple; j < n_samples; j++) {
results[j] = sampler(seeds[0]);
results[j] = sampler(cache_box[0].seed);
// we can just reuse a seed, this isn't problematic because we are not doing multithreading
}
for (uint64_t i = 0; i < n_threads; i++) {
free(seeds[i]);
for (int i = 0; i < n_threads; i++) {
free(cache_box[i].seed);
}
free(seeds);
free(cache_box);
}
/* Get confidence intervals, given a sampler */
// Not in core yet because I'm not sure how much I like the struct
// and the built-in 100k samples
// to do: add n to function parameters and document
typedef struct ci_t {
double low;
@ -89,10 +92,12 @@ static void swp(int i, int j, double xs[])
static int partition(int low, int high, double xs[], int length)
{
// To understand this function:
// - see the note after gt variable definition
// - go to commit 578bfa27 and the scratchpad/ folder in it, which has printfs sprinkled throughout
int pivot = low + floor((high - low) / 2);
if (low > high || high >= length) {
printf("Invariant violated for function partition in %s (%d)", __FILE__, __LINE__);
exit(1);
}
// Note: the scratchpad/ folder in commit 578bfa27 has printfs sprinkled throughout
int pivot = low + (int)floor((high - low) / 2);
double pivot_value = xs[pivot];
swp(pivot, high, xs);
int gt = low; /* This pointer will iterate until finding an element which is greater than the pivot. Then it will move elements that are smaller before it--more specifically, it will move elements to its position and then increment. As a result all elements between gt and i will be greater than the pivot. */
@ -109,15 +114,24 @@ static int partition(int low, int high, double xs[], int length)
static double quickselect(int k, double xs[], int n)
{
// https://en.wikipedia.org/wiki/Quickselect
double *ys = malloc((size_t)n * sizeof(double));
memcpy(ys, xs, (size_t)n * sizeof(double));
// ^: don't rearrange item order in the original array
int low = 0;
int high = n - 1;
for (;;) {
if (low == high) {
return xs[low];
double result = ys[low];
free(ys);
return result;
}
int pivot = partition(low, high, xs, n);
int pivot = partition(low, high, ys, n);
if (pivot == k) {
return xs[pivot];
double result = ys[pivot];
free(ys);
return result;
} else if (k < pivot) {
high = pivot - 1;
} else {
@ -129,8 +143,8 @@ static double quickselect(int k, double xs[], int n)
ci array_get_ci(ci interval, double* xs, int n)
{
int low_k = floor(interval.low * n);
int high_k = ceil(interval.high * n);
int low_k = (int)floor(interval.low * n);
int high_k = (int)ceil(interval.high * n);
ci result = {
.low = quickselect(low_k, xs, n),
.high = quickselect(high_k, xs, n),
@ -144,10 +158,8 @@ ci array_get_90_ci(double xs[], int n)
ci sampler_get_ci(ci interval, double (*sampler)(uint64_t*), int n, uint64_t* seed)
{
double* xs = malloc(n * sizeof(double));
/*for (int i = 0; i < n; i++) {
xs[i] = sampler(seed);
}*/
UNUSED(seed); // don't want to use it right now, but want to preserve ability to do so (e.g., remove parallelism from internals). Also nicer for consistency.
double* xs = malloc((size_t)n * sizeof(double));
sampler_parallel(sampler, xs, 16, n);
ci result = array_get_ci(interval, xs, n);
free(xs);
@ -159,9 +171,6 @@ ci sampler_get_90_ci(double (*sampler)(uint64_t*), int n, uint64_t* seed)
}
/* Algebra manipulations */
// here I discover named structs,
// which mean that I don't have to be typing
// struct blah all the time.
#define NORMAL90CONFIDENCE 1.6448536269514727
@ -195,8 +204,8 @@ lognormal_params algebra_product_lognormals(lognormal_params a, lognormal_params
lognormal_params convert_ci_to_lognormal_params(ci x)
{
double loghigh = logf(x.high);
double loglow = logf(x.low);
double loghigh = log(x.high);
double loglow = log(x.low);
double logmean = (loghigh + loglow) / 2.0;
double logstd = (loghigh - loglow) / (2.0 * NORMAL90CONFIDENCE);
lognormal_params result = { .logmean = logmean, .logstd = logstd };
@ -221,21 +230,21 @@ ci convert_lognormal_params_to_ci(lognormal_params y)
#define EXIT_ON_ERROR 0
#define PROCESS_ERROR(error_msg) process_error(error_msg, EXIT_ON_ERROR, __FILE__, __LINE__)
struct box {
typedef struct box_t {
int empty;
double content;
char* error_msg;
};
} box;
struct box process_error(const char* error_msg, int should_exit, char* file, int line)
box process_error(const char* error_msg, int should_exit, char* file, int line)
{
if (should_exit) {
printf("@, in %s (%d)", file, line);
printf("%s, @, in %s (%d)", error_msg, file, line);
exit(1);
} else {
char error_msg[MAX_ERROR_LENGTH];
snprintf(error_msg, MAX_ERROR_LENGTH, "@, in %s (%d)", file, line); // NOLINT: We are being carefull here by considering MAX_ERROR_LENGTH explicitly.
struct box error = { .empty = 1, .error_msg = error_msg };
box error = { .empty = 1, .error_msg = error_msg };
return error;
}
}
@ -244,7 +253,7 @@ struct box process_error(const char* error_msg, int should_exit, char* file, int
// Version #1:
// - input: (cdf: double => double, p)
// - output: Box(number|error)
struct box inverse_cdf_double(double cdf(double), double p)
box inverse_cdf_double(double cdf(double), double p)
{
// given a cdf: [-Inf, Inf] => [0,1]
// returns a box with either
@ -257,8 +266,9 @@ struct box inverse_cdf_double(double cdf(double), double p)
// 1. Make sure that cdf(low) < p < cdf(high)
int interval_found = 0;
while ((!interval_found) && (low > -FLT_MAX / 4) && (high < FLT_MAX / 4)) {
// ^ Using FLT_MIN and FLT_MAX is overkill
while ((!interval_found) && (low > -DBL_MAX / 4) && (high < DBL_MAX / 4)) {
// for floats, use FLT_MAX instead
// Note that this approach is overkill
// but it's also the *correct* thing to do.
int low_condition = (cdf(low) < p);
@ -299,7 +309,7 @@ struct box inverse_cdf_double(double cdf(double), double p)
}
if (convergence_condition) {
struct box result = { .empty = 0, .content = low };
box result = { .empty = 0, .content = low };
return result;
} else {
return PROCESS_ERROR("Search process did not converge, in function inverse_cdf");
@ -310,7 +320,7 @@ struct box inverse_cdf_double(double cdf(double), double p)
// Version #2:
// - input: (cdf: double => Box(number|error), p)
// - output: Box(number|error)
struct box inverse_cdf_box(struct box cdf_box(double), double p)
box inverse_cdf_box(box cdf_box(double), double p)
{
// given a cdf: [-Inf, Inf] => Box([0,1])
// returns a box with either
@ -323,15 +333,16 @@ struct box inverse_cdf_box(struct box cdf_box(double), double p)
// 1. Make sure that cdf(low) < p < cdf(high)
int interval_found = 0;
while ((!interval_found) && (low > -FLT_MAX / 4) && (high < FLT_MAX / 4)) {
// ^ Using FLT_MIN and FLT_MAX is overkill
while ((!interval_found) && (low > -DBL_MAX / 4) && (high < DBL_MAX / 4)) {
// for floats, use FLT_MAX instead
// Note that this approach is overkill
// but it's also the *correct* thing to do.
struct box cdf_low = cdf_box(low);
box cdf_low = cdf_box(low);
if (cdf_low.empty) {
return PROCESS_ERROR(cdf_low.error_msg);
}
struct box cdf_high = cdf_box(high);
box cdf_high = cdf_box(high);
if (cdf_high.empty) {
return PROCESS_ERROR(cdf_low.error_msg);
}
@ -361,7 +372,7 @@ struct box inverse_cdf_box(struct box cdf_box(double), double p)
// if ((width < 1e-8) || mid_not_new){
convergence_condition = 1;
} else {
struct box cdf_mid = cdf_box(mid);
box cdf_mid = cdf_box(mid);
if (cdf_mid.empty) {
return PROCESS_ERROR(cdf_mid.error_msg);
}
@ -378,7 +389,7 @@ struct box inverse_cdf_box(struct box cdf_box(double), double p)
}
if (convergence_condition) {
struct box result = { .empty = 0, .content = low };
box result = { .empty = 0, .content = low };
return result;
} else {
return PROCESS_ERROR("Search process did not converge, in function inverse_cdf");
@ -389,22 +400,22 @@ struct box inverse_cdf_box(struct box cdf_box(double), double p)
/* Sample from an arbitrary cdf */
// Before: invert an arbitrary cdf at a point
// Now: from an arbitrary cdf, get a sample
struct box sampler_cdf_box(struct box cdf(double), uint64_t* seed)
box sampler_cdf_box(box cdf(double), uint64_t* seed)
{
double p = sample_unit_uniform(seed);
struct box result = inverse_cdf_box(cdf, p);
box result = inverse_cdf_box(cdf, p);
return result;
}
struct box sampler_cdf_double(double cdf(double), uint64_t* seed)
box sampler_cdf_double(double cdf(double), uint64_t* seed)
{
double p = sample_unit_uniform(seed);
struct box result = inverse_cdf_double(cdf, p);
box result = inverse_cdf_double(cdf, p);
return result;
}
double sampler_cdf_danger(struct box cdf(double), uint64_t* seed)
double sampler_cdf_danger(box cdf(double), uint64_t* seed)
{
double p = sample_unit_uniform(seed);
struct box result = inverse_cdf_box(cdf, p);
box result = inverse_cdf_box(cdf, p);
if (result.empty) {
exit(1);
} else {
@ -413,7 +424,6 @@ double sampler_cdf_danger(struct box cdf(double), uint64_t* seed)
}
/* array print: potentially useful for debugging */
void array_print(double xs[], int n)
{
printf("[");

View File

@ -32,23 +32,23 @@ lognormal_params convert_ci_to_lognormal_params(ci x);
ci convert_lognormal_params_to_ci(lognormal_params y);
/* Error handling */
struct box {
typedef struct box_t {
int empty;
double content;
char* error_msg;
};
} box;
#define MAX_ERROR_LENGTH 500
#define EXIT_ON_ERROR 0
#define PROCESS_ERROR(error_msg) process_error(error_msg, EXIT_ON_ERROR, __FILE__, __LINE__)
struct box process_error(const char* error_msg, int should_exit, char* file, int line);
box process_error(const char* error_msg, int should_exit, char* file, int line);
void array_print(double* array, int length);
/* Inverse cdf */
struct box inverse_cdf_double(double cdf(double), double p);
struct box inverse_cdf_box(struct box cdf_box(double), double p);
box inverse_cdf_double(double cdf(double), double p);
box inverse_cdf_box(box cdf_box(double), double p);
/* Samplers from cdf */
struct box sampler_cdf_double(double cdf(double), uint64_t* seed);
struct box sampler_cdf_box(struct box cdf(double), uint64_t* seed);
box sampler_cdf_double(double cdf(double), uint64_t* seed);
box sampler_cdf_box(box cdf(double), uint64_t* seed);
#endif