JJSierraM/squiggle.c
1
0
Fork 0
forked from personal/squiggle.c
Code Issues Pull Requests Packages Projects Releases Wiki Activity

seed cache box name, make "quotient" private var

Browse Source
This commit is contained in:
NunoSempere 2024-01-12 00:23:01 +01:00
parent 73ea6f32c2
commit 2b5b496c25
1 changed files with 11 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
squiggle_more.c
View File

@ -10,10 +10,11 @@
/* Parallel sampler */ /* Parallel sampler */
#define CACHE_LINE_SIZE 64 #define CACHE_LINE_SIZE 64
typedef struct padded_seed_t { typedef struct seed_cache_box_t {
uint64_t* seed; uint64_t* seed;
char padding[CACHE_LINE_SIZE - sizeof(uint64_t*)]; char padding[CACHE_LINE_SIZE - sizeof(uint64_t*)];
} padded_seed; } 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) void sampler_parallel(double (*sampler)(uint64_t* seed), double* results, int n_threads, int n_samples)
{ {
@ -36,14 +37,14 @@ void sampler_parallel(double (*sampler)(uint64_t* seed), double* results, int n_
int divisor_multiple = quotient * n_threads; int divisor_multiple = quotient * n_threads;
// uint64_t** seeds = malloc((size_t)n_threads * sizeof(uint64_t*)); // uint64_t** seeds = malloc((size_t)n_threads * sizeof(uint64_t*));
padded_seed* seeds = (padded_seed*) malloc(sizeof(padded_seed) * (size_t)n_threads); seed_cache_box* cache_box = (seed_cache_box*) malloc(sizeof(seed_cache_box) * (size_t)n_threads);
srand(1); srand(1);
for (int i = 0; i < n_threads; i++) { for (int i = 0; i < n_threads; i++) {
seeds[i].seed = malloc(sizeof(uint64_t*)); cache_box[i].seed = malloc(sizeof(uint64_t*));
// Constraints: // Constraints:
// - xorshift can't start with 0 // - xorshift can't start with 0
// - the seeds should be reasonably separated and not correlated // - the seeds should be reasonably separated and not correlated
*(seeds[i].seed) = (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]); // printf("#%ld: %lu\n",i, *seeds[i]);
// Other initializations tried: // Other initializations tried:
@ -53,26 +54,26 @@ void sampler_parallel(double (*sampler)(uint64_t* seed), double* results, int n_
} }
int i; int i;
#pragma omp parallel private(i) #pragma omp parallel private(i, quotient)
{ {
#pragma omp for #pragma omp for
for (i = 0; i < n_threads; i++) { for (i = 0; i < n_threads; i++) {
int lower_bound_inclusive = i * quotient; int lower_bound_inclusive = i * quotient;
int upper_bound_not_inclusive = ((i + 1) * quotient); // note the < in the for loop below, int upper_bound_not_inclusive = ((i + 1) * quotient); // note the < in the for loop below,
for (int j = lower_bound_inclusive; j < upper_bound_not_inclusive; j++) { for (int j = lower_bound_inclusive; j < upper_bound_not_inclusive; j++) {
results[j] = sampler(seeds[i].seed); results[j] = sampler(cache_box[i].seed);
} }
} }
} }
for (int j = divisor_multiple; j < n_samples; j++) { for (int j = divisor_multiple; j < n_samples; j++) {
results[j] = sampler(seeds[0].seed); results[j] = sampler(cache_box[0].seed);
// we can just reuse a seed, this isn't problematic because we are not doing multithreading // we can just reuse a seed, this isn't problematic because we are not doing multithreading
} }
for (int i = 0; i < n_threads; i++) { for (int i = 0; i < n_threads; i++) {
free(seeds[i].seed); free(cache_box[i].seed);
} }
free(seeds); free(cache_box);
} }
/* Get confidence intervals, given a sampler */ /* Get confidence intervals, given a sampler */