commit 14a18276c01a650893383457712cbb83189da3b6 parent bbe0116381f797bb29b92d3f2b1429583f4bcf0f Author: NunoSempere <nuno.sempere@protonmail.com> Date: Fri, 12 Jan 2024 16:52:56 +0100 tweak: don't use unneeded pointers to get cache math right Diffstat:
16 files changed, 7 insertions(+), 7 deletions(-)
diff --git a/examples/more/00_example_template/example b/examples/more/00_example_template/example Binary files differ. diff --git a/examples/more/01_sample_from_cdf/example b/examples/more/01_sample_from_cdf/example Binary files differ. diff --git a/examples/more/02_sample_from_cdf_beta/example b/examples/more/02_sample_from_cdf_beta/example Binary files differ. diff --git a/examples/more/03_ci_beta/example b/examples/more/03_ci_beta/example Binary files differ. diff --git a/examples/more/04_nuclear_war/example b/examples/more/04_nuclear_war/example Binary files differ. diff --git a/examples/more/05_burn_10kg_fat/example b/examples/more/05_burn_10kg_fat/example Binary files differ. diff --git a/examples/more/06_nuclear_recovery/example b/examples/more/06_nuclear_recovery/example Binary files differ. diff --git a/examples/more/07_algebra/example b/examples/more/07_algebra/example Binary files differ. diff --git a/examples/more/08_algebra_and_conversion/example b/examples/more/08_algebra_and_conversion/example Binary files differ. diff --git a/examples/more/09_ergonomic_algebra/example b/examples/more/09_ergonomic_algebra/example Binary files differ. diff --git a/examples/more/10_twitter_thread_example/example b/examples/more/10_twitter_thread_example/example Binary files differ. diff --git a/examples/more/11_billion_lognormals_paralell/example b/examples/more/11_billion_lognormals_paralell/example Binary files differ. diff --git a/examples/more/12_time_to_botec_parallel/example b/examples/more/12_time_to_botec_parallel/example Binary files differ. diff --git a/examples/more/13_parallelize_min/example b/examples/more/13_parallelize_min/example Binary files differ. diff --git a/examples/more/14_check_confidence_interval/example b/examples/more/14_check_confidence_interval/example Binary files differ. diff --git a/squiggle_more.c b/squiggle_more.c @@ -11,7 +11,7 @@ /* Parallel sampler */ #define CACHE_LINE_SIZE 64 typedef struct seed_cache_box_t { - uint64_t* seed; + uint64_t seed; char padding[CACHE_LINE_SIZE - sizeof(uint64_t*)]; } seed_cache_box; // This avoids "false sharing", i.e., different threads competing for the same cache line @@ -43,11 +43,10 @@ void sampler_parallel(double (*sampler)(uint64_t* seed), double* results, int n_ seed_cache_box* cache_box = (seed_cache_box*) malloc(sizeof(seed_cache_box) * (size_t)n_threads); srand(1); 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 - *(cache_box[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]); // Other initializations tried: @@ -57,25 +56,26 @@ void sampler_parallel(double (*sampler)(uint64_t* seed), double* results, int n_ } int i; -#pragma omp parallel private(i, quotient) +#pragma omp parallel private(i) { #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, for (int j = lower_bound_inclusive; j < upper_bound_not_inclusive; j++) { - results[j] = sampler(cache_box[i].seed); + results[j] = sampler(&(cache_box[i].seed)); } } } for (int j = divisor_multiple; j < n_samples; j++) { - results[j] = sampler(cache_box[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 } - + /* for (int i = 0; i < n_threads; i++) { free(cache_box[i].seed); } + */ free(cache_box); }