From 9a60392849d6560b5a87ff9473ece37386dcd9d2 Mon Sep 17 00:00:00 2001 From: NunoSempere Date: Sat, 18 Nov 2023 19:22:28 +0000 Subject: [PATCH] reorg C experiments, and start factoring out paralellization functions --- C/{C-01-simple => alt/01-simple}/makefile | 0 C/{C-01-simple => alt/01-simple}/samples | Bin C/{C-01-simple => alt/01-simple}/samples.c | 0 .../02-better-algorithm-one-thread}/makefile | 0 .../out/samples-one-thread | Bin .../samples-one-thread.c | 0 C/alt/03-factor-out-parallelization/makefile | 106 ++++++ C/alt/03-factor-out-parallelization/samples.c | 309 ++++++++++++++++++ C/{ => alt}/xorshift-scratchpad/makefile | 0 C/{ => alt}/xorshift-scratchpad/notes.md | 0 .../xorshift-scratchpad/out/xorshift-pointer | Bin .../xorshift-scratchpad/out/xorshift-struct | Bin .../xorshift-scratchpad/xorshift-pointer.c | 0 .../xorshift-scratchpad/xorshift-struct.c | 0 C/samples.c | 83 ++--- 15 files changed, 457 insertions(+), 41 deletions(-) rename C/{C-01-simple => alt/01-simple}/makefile (100%) rename C/{C-01-simple => alt/01-simple}/samples (100%) rename C/{C-01-simple => alt/01-simple}/samples.c (100%) rename C/{C-02-better-algorithm-one-thread => alt/02-better-algorithm-one-thread}/makefile (100%) rename C/{C-02-better-algorithm-one-thread => alt/02-better-algorithm-one-thread}/out/samples-one-thread (100%) rename C/{C-02-better-algorithm-one-thread => alt/02-better-algorithm-one-thread}/samples-one-thread.c (100%) create mode 100644 C/alt/03-factor-out-parallelization/makefile create mode 100644 C/alt/03-factor-out-parallelization/samples.c rename C/{ => alt}/xorshift-scratchpad/makefile (100%) rename C/{ => alt}/xorshift-scratchpad/notes.md (100%) rename C/{ => alt}/xorshift-scratchpad/out/xorshift-pointer (100%) rename C/{ => alt}/xorshift-scratchpad/out/xorshift-struct (100%) rename C/{ => alt}/xorshift-scratchpad/xorshift-pointer.c (100%) rename C/{ => alt}/xorshift-scratchpad/xorshift-struct.c (100%) diff --git a/C/C-01-simple/makefile b/C/alt/01-simple/makefile similarity index 100% rename from C/C-01-simple/makefile rename to C/alt/01-simple/makefile diff --git a/C/C-01-simple/samples b/C/alt/01-simple/samples similarity index 100% rename from C/C-01-simple/samples rename to C/alt/01-simple/samples diff --git a/C/C-01-simple/samples.c b/C/alt/01-simple/samples.c similarity index 100% rename from C/C-01-simple/samples.c rename to C/alt/01-simple/samples.c diff --git a/C/C-02-better-algorithm-one-thread/makefile b/C/alt/02-better-algorithm-one-thread/makefile similarity index 100% rename from C/C-02-better-algorithm-one-thread/makefile rename to C/alt/02-better-algorithm-one-thread/makefile diff --git a/C/C-02-better-algorithm-one-thread/out/samples-one-thread b/C/alt/02-better-algorithm-one-thread/out/samples-one-thread similarity index 100% rename from C/C-02-better-algorithm-one-thread/out/samples-one-thread rename to C/alt/02-better-algorithm-one-thread/out/samples-one-thread diff --git a/C/C-02-better-algorithm-one-thread/samples-one-thread.c b/C/alt/02-better-algorithm-one-thread/samples-one-thread.c similarity index 100% rename from C/C-02-better-algorithm-one-thread/samples-one-thread.c rename to C/alt/02-better-algorithm-one-thread/samples-one-thread.c diff --git a/C/alt/03-factor-out-parallelization/makefile b/C/alt/03-factor-out-parallelization/makefile new file mode 100644 index 00000000..149b8827 --- /dev/null +++ b/C/alt/03-factor-out-parallelization/makefile @@ -0,0 +1,106 @@ +# Interface: +# make +# make build +# make format +# make run + +# Compiler +CC=gcc +# CC=tcc # <= faster compilation + +# Main file +SRC=samples.c +OUTPUT=out/samples + +SRC_ONE_THREAD=./samples-one-thread.c +OUTPUT_ONE_THREAD=out/samples-one-thread + +## Dependencies +# Has no dependencies +MATH=-lm + +## Flags +DEBUG= #'-g' +STANDARD=-std=c99 +WARNINGS=-Wall +OPTIMIZED=-O3 #-O3 actually gives better performance than -Ofast, at least for this version +OPENMP=-fopenmp + +## Formatter +STYLE_BLUEPRINT=webkit +FORMATTER=clang-format -i -style=$(STYLE_BLUEPRINT) + +## make build +build: $(SRC) + $(CC) $(OPTIMIZED) $(DEBUG) $(SRC) $(OPENMP) $(MATH) -o $(OUTPUT) + +static: + $(CC) $(OPTIMIZED) $(DEBUG) $(SRC) $(OPENMP) $(MATH) -o $(OUTPUT) + +format: $(SRC) + $(FORMATTER) $(SRC) + +run: $(SRC) $(OUTPUT) + OMP_NUM_THREADS=1 ./$(OUTPUT) && echo + +multi: + OMP_NUM_THREADS=1 ./$(OUTPUT) && echo + OMP_NUM_THREADS=2 ./$(OUTPUT) && echo + OMP_NUM_THREADS=4 ./$(OUTPUT) && echo + OMP_NUM_THREADS=8 ./$(OUTPUT) && echo + OMP_NUM_THREADS=16 ./$(OUTPUT) && echo + +## Timing + +time-linux: + @echo "Requires /bin/time, found on GNU/Linux systems" && echo + + @echo "Running 100x and taking avg time: OMP_NUM_THREADS=1 $(OUTPUT)" + @t=$$(/usr/bin/time -f "%e" -p bash -c 'for i in {1..100}; do OMP_NUM_THREADS=1 $(OUTPUT); done' 2>&1 >/dev/null | grep real | awk '{print $$2}' ); echo "scale=2; 1000 * $$t / 100" | bc | sed "s|^|Time using 1 thread: |" | sed 's|$$|ms|' && echo + + @echo "Running 100x and taking avg time: OMP_NUM_THREADS=2 $(OUTPUT)" + @t=$$(/usr/bin/time -f "%e" -p bash -c 'for i in {1..100}; do OMP_NUM_THREADS=2 $(OUTPUT); done' 2>&1 >/dev/null | grep real | awk '{print $$2}' ); echo "scale=2; 1000 * $$t / 100" | bc | sed "s|^|Time using 2 threads: |" | sed 's|$$|ms|' && echo + + @echo "Running 100x and taking avg time: OMP_NUM_THREADS=4 $(OUTPUT)" + @t=$$(/usr/bin/time -f "%e" -p bash -c 'for i in {1..100}; do OMP_NUM_THREADS=4 $(OUTPUT); done' 2>&1 >/dev/null | grep real | awk '{print $$2}' ); echo "scale=2; 1000 * $$t / 100" | bc | sed "s|^|Time for 4 threads: |" | sed 's|$$|ms|' && echo + + @echo "Running 100x and taking avg time: OMP_NUM_THREADS=8 $(OUTPUT)" + @t=$$(/usr/bin/time -f "%e" -p bash -c 'for i in {1..100}; do OMP_NUM_THREADS=8 $(OUTPUT); done' 2>&1 >/dev/null | grep real | awk '{print $$2}' ); echo "scale=2; 1000 * $$t / 100" | bc | sed "s|^|Time using 8 threads: |" | sed 's|$$|ms|' && echo + + @echo "Running 100x and taking avg time: OMP_NUM_THREADS=16 $(OUTPUT)" + @t=$$(/usr/bin/time -f "%e" -p bash -c 'for i in {1..100}; do OMP_NUM_THREADS=16 $(OUTPUT); done' 2>&1 >/dev/null | grep real | awk '{print $$2}' ); echo "scale=2; 1000 * $$t / 100" | bc | sed "s|^|Time using 16 threads: |" | sed 's|$$|ms|' && echo + +time-linux-fastest: + @echo "Running 100x and taking avg time: OMP_NUM_THREADS=16 $(OUTPUT)" + @t=$$(/usr/bin/time -f "%e" -p bash -c 'for i in {1..100}; do OMP_NUM_THREADS=16 $(OUTPUT); done' 2>&1 >/dev/null | grep real | awk '{print $$2}' ); echo "scale=2; 1000 * $$t / 100" | bc | sed "s|^|Time using 16 threads: |" | sed 's|$$|ms|' && echo + +time-linux-simple: + @echo "Requires /bin/time, found on GNU/Linux systems" && echo + OMP_NUM_THREADS=1 /bin/time -f "Time: %es" ./$(OUTPUT) && echo + OMP_NUM_THREADS=2 /bin/time -f "Time: %es" ./$(OUTPUT) && echo + OMP_NUM_THREADS=4 /bin/time -f "Time: %es" ./$(OUTPUT) && echo + OMP_NUM_THREADS=8 /bin/time -f "Time: %es" ./$(OUTPUT) && echo + OMP_NUM_THREADS=16 /bin/time -f "Time: %es" ./$(OUTPUT) && echo + +## Profiling + +profile-linux: + echo "Requires perf, which depends on the kernel version, and might be in linux-tools package or similar" + echo "Must be run as sudo" + $(CC) $(SRC) $(OPENMP) $(MATH) -o $(OUTPUT) + # ./$(OUTPUT) + # gprof: + # gprof $(OUTPUT) gmon.out > analysis.txt + # rm gmon.out + # vim analysis.txt + # rm analysis.txt + # perf: + OMP_NUM_THREADS=16 sudo perf record $(OUTPUT) + sudo perf report + rm perf.data + + +## Install +debian-install-dependencies: + sudo apt-get install libomp-dev + diff --git a/C/alt/03-factor-out-parallelization/samples.c b/C/alt/03-factor-out-parallelization/samples.c new file mode 100644 index 00000000..a130045e --- /dev/null +++ b/C/alt/03-factor-out-parallelization/samples.c @@ -0,0 +1,309 @@ +#include +#include +#include +#include +#include + +const float PI = 3.14159265358979323846; + +#define N 1000000 + +//Array helpers + +void array_print(float* array, int length) +{ + for (int i = 0; i < length; i++) { + printf("item[%d] = %f\n", i, array[i]); + } + printf("\n"); +} + +float array_sum(float* array, int length) +{ + float output = 0.0; + for (int i = 0; i < length; i++) { + output += array[i]; + } + return output; +} + +void array_cumsum(float* array_to_sum, float* array_cumsummed, int length) +{ + array_cumsummed[0] = array_to_sum[0]; + for (int i = 1; i < length; i++) { + array_cumsummed[i] = array_cumsummed[i - 1] + array_to_sum[i]; + } +} + +// Split array helpers +int split_array_get_length(int index, int total_length, int n_threads) +{ + return (total_length % n_threads > index ? total_length / n_threads + 1 : total_length / n_threads); +} + +void split_array_allocate(float** meta_array, int length, int divide_into) +{ + int split_array_length; + + for (int i = 0; i < divide_into; i++) { + split_array_length = split_array_get_length(i, length, divide_into); + meta_array[i] = malloc(split_array_length * sizeof(float)); + } +} + +void split_array_free(float** meta_array, int divided_into) +{ + for (int i = 0; i < divided_into; i++) { + free(meta_array[i]); + } + free(meta_array); +} + +float split_array_sum(float** meta_array, int length, int divided_into) +{ + int i; + float output = 0; + +#pragma omp parallel for reduction(+ \ + : output) + for (int i = 0; i < divided_into; i++) { + float own_partial_sum = 0; + int split_array_length = split_array_get_length(i, length, divided_into); + for (int j = 0; j < split_array_length; j++) { + own_partial_sum += meta_array[i][j]; + } + output += own_partial_sum; + } + return output; +} + +// Pseudo Random number generator + +uint32_t xorshift32(uint32_t* seed) +{ + // Algorithm "xor" from p. 4 of Marsaglia, "Xorshift RNGs" + // See + // https://en.wikipedia.org/wiki/Xorshift + // Also some drama: , + + uint32_t x = *seed; + x ^= x << 13; + x ^= x >> 17; + x ^= x << 5; + return *seed = x; +} + +// Distribution & sampling functions + +float rand_0_to_1(uint32_t* seed) +{ + return ((float)xorshift32(seed)) / ((float)UINT32_MAX); + /* + uint32_t x = *seed; + x ^= x << 13; + x ^= x >> 17; + x ^= x << 5; + return ((float)(*seed = x))/((float) UINT32_MAX); + */ + // previously: + // ((float)rand_r(seed) / (float)RAND_MAX) + // and before that: rand, but it wasn't thread-safe. + // See: for why to use rand_r: + // rand() is not thread-safe, as it relies on (shared) hidden seed. +} + +float rand_float(float max, uint32_t* seed) +{ + return rand_0_to_1(seed) * max; +} + +float ur_normal(uint32_t* seed) +{ + float u1 = rand_0_to_1(seed); + float u2 = rand_0_to_1(seed); + float z = sqrtf(-2.0 * log(u1)) * sin(2 * PI * u2); + return z; +} + +float random_uniform(float from, float to, uint32_t* seed) +{ + return rand_0_to_1(seed) * (to - from) + from; +} + +float random_normal(float mean, float sigma, uint32_t* seed) +{ + return (mean + sigma * ur_normal(seed)); +} + +float random_lognormal(float logmean, float logsigma, uint32_t* seed) +{ + return expf(random_normal(logmean, logsigma, seed)); +} + +float random_to(float low, float high, uint32_t* seed) +{ + const float NORMAL95CONFIDENCE = 1.6448536269514722; + float loglow = logf(low); + float loghigh = logf(high); + float logmean = (loglow + loghigh) / 2; + float logsigma = (loghigh - loglow) / (2.0 * NORMAL95CONFIDENCE); + return random_lognormal(logmean, logsigma, seed); +} + +// Mixture function + +float mixture_one_thread(float (*samplers[])(uint32_t*), float* weights, int n_dists, uint32_t* seed) +{ + + // You can see a slightly simpler version of this function in the git history + // or in alt/C-02-better-algorithm-one-thread/ + float sum_weights = array_sum(weights, n_dists); + float* cumsummed_normalized_weights = malloc(n_dists * sizeof(float)); + 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; + } + + //create var holders + float p1, result; + int sample_index, i, own_length; + p1 = random_uniform(0, 1); + for (int i = 0; i < n_dists; i++) { + if (p1 < cummulative_weights[i]) { + result = samplers[i](); + break; + } + } + free(normalized_weights); + free(cummulative_weights); + return result; +} + +// mixture paralellized +void mixture_paralell(float (*samplers[])(uint32_t*), float* weights, int n_dists, float** results, int n_threads) +{ + // You can see a simpler version of this function in the git history + // or in alt/C-02-better-algorithm-one-thread/ + float sum_weights = array_sum(weights, n_dists); + float* cumsummed_normalized_weights = malloc(n_dists * sizeof(float)); + 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; + } + + //create var holders + float p1; + int sample_index, i, split_array_length; + + // uint32_t* seeds[n_threads]; + uint32_t** seeds = malloc(n_threads * sizeof(uint32_t*)); + for (uint32_t i = 0; i < n_threads; i++) { + seeds[i] = malloc(sizeof(uint32_t)); + *seeds[i] = i + 1; // xorshift can't start with 0 + } + +#pragma omp parallel private(i, p1, sample_index, split_array_length) + { +#pragma omp for + for (i = 0; i < n_threads; i++) { + split_array_length = split_array_get_length(i, N, n_threads); + for (int j = 0; j < split_array_length; j++) { + p1 = random_uniform(0, 1, seeds[i]); + for (int k = 0; k < n_dists; k++) { + if (p1 < cumsummed_normalized_weights[k]) { + results[i][j] = samplers[k](seeds[i]); + break; + } + } + } + } + } + // free(normalized_weights); + // free(cummulative_weights); + free(cumsummed_normalized_weights); + for (uint32_t i = 0; i < n_threads; i++) { + free(seeds[i]); + } + free(seeds); +} + +// Parallization function +void paralellize(float *sampler(uint32_t* seed), float** results, int n_threads){ + + int sample_index, i, split_array_length; + uint32_t** seeds = malloc(n_threads * sizeof(uint32_t*)); + for (uint32_t i = 0; i < n_threads; i++) { + seeds[i] = malloc(sizeof(uint32_t)); + *seeds[i] = i + 1; // xorshift can't start with 0 + } + + #pragma omp parallel private(i, p1, sample_index, split_array_length) + { + #pragma omp for + for (i = 0; i < n_threads; i++) { + split_array_length = split_array_get_length(i, N, n_threads); + for (int j = 0; j < split_array_length; j++) { + results[i][j] = sampler(seeds[i]); + break; + } + } + } + free(cumsummed_normalized_weights); + for (uint32_t i = 0; i < n_threads; i++) { + free(seeds[i]); + } + free(seeds); +} + +// Functions used for the BOTEC. +// Their type has to be the same, as we will be passing them around. + +float sample_0(uint32_t* seed) +{ + return 0; +} + +float sample_1(uint32_t* seed) +{ + return 1; +} + +float sample_few(uint32_t* seed) +{ + return random_to(1, 3, seed); +} + +float sample_many(uint32_t* seed) +{ + return random_to(2, 10, seed); +} + +int main() +{ + + // Toy example + // Declare variables in play + float p_a, p_b, p_c; + int n_threads = omp_get_max_threads(); + // printf("Max threads: %d\n", n_threads); + // omp_set_num_threads(n_threads); + float** dist_mixture = malloc(n_threads * sizeof(float*)); + split_array_allocate(dist_mixture, N, n_threads); + + // Initialize variables + p_a = 0.8; + p_b = 0.5; + p_c = p_a * p_b; + + // Generate mixture + int n_dists = 4; + float weights[] = { 1 - p_c, p_c / 2, p_c / 4, p_c / 4 }; + float (*samplers[])(uint32_t*) = { sample_0, sample_1, sample_few, sample_many }; + + mixture(samplers, weights, n_dists, dist_mixture, n_threads); + printf("Sum(dist_mixture, N)/N = %f\n", split_array_sum(dist_mixture, N, n_threads) / N); + // array_print(dist_mixture[0], N); + split_array_free(dist_mixture, n_threads); + + return 0; +} diff --git a/C/xorshift-scratchpad/makefile b/C/alt/xorshift-scratchpad/makefile similarity index 100% rename from C/xorshift-scratchpad/makefile rename to C/alt/xorshift-scratchpad/makefile diff --git a/C/xorshift-scratchpad/notes.md b/C/alt/xorshift-scratchpad/notes.md similarity index 100% rename from C/xorshift-scratchpad/notes.md rename to C/alt/xorshift-scratchpad/notes.md diff --git a/C/xorshift-scratchpad/out/xorshift-pointer b/C/alt/xorshift-scratchpad/out/xorshift-pointer similarity index 100% rename from C/xorshift-scratchpad/out/xorshift-pointer rename to C/alt/xorshift-scratchpad/out/xorshift-pointer diff --git a/C/xorshift-scratchpad/out/xorshift-struct b/C/alt/xorshift-scratchpad/out/xorshift-struct similarity index 100% rename from C/xorshift-scratchpad/out/xorshift-struct rename to C/alt/xorshift-scratchpad/out/xorshift-struct diff --git a/C/xorshift-scratchpad/xorshift-pointer.c b/C/alt/xorshift-scratchpad/xorshift-pointer.c similarity index 100% rename from C/xorshift-scratchpad/xorshift-pointer.c rename to C/alt/xorshift-scratchpad/xorshift-pointer.c diff --git a/C/xorshift-scratchpad/xorshift-struct.c b/C/alt/xorshift-scratchpad/xorshift-struct.c similarity index 100% rename from C/xorshift-scratchpad/xorshift-struct.c rename to C/alt/xorshift-scratchpad/xorshift-struct.c diff --git a/C/samples.c b/C/samples.c index e9f13702..ab9a0e55 100644 --- a/C/samples.c +++ b/C/samples.c @@ -1,7 +1,7 @@ -#include #include -#include +#include #include +#include #include const float PI = 3.14159265358979323846; @@ -64,51 +64,52 @@ float split_array_sum(float** meta_array, int length, int divided_into) int i; float output = 0; - #pragma omp parallel for reduction(+:output) - for (int i = 0; i < divided_into; i++) { - float own_partial_sum = 0; - int split_array_length = split_array_get_length(i, length, divided_into); - for (int j = 0; j < split_array_length; j++) { - own_partial_sum += meta_array[i][j]; - } - output += own_partial_sum; - } - return output; - +#pragma omp parallel for reduction(+ \ + : output) + for (int i = 0; i < divided_into; i++) { + float own_partial_sum = 0; + int split_array_length = split_array_get_length(i, length, divided_into); + for (int j = 0; j < split_array_length; j++) { + own_partial_sum += meta_array[i][j]; + } + output += own_partial_sum; + } + return output; } // Pseudo Random number generator uint32_t xorshift32(uint32_t* seed) { - // Algorithm "xor" from p. 4 of Marsaglia, "Xorshift RNGs" - // See - // https://en.wikipedia.org/wiki/Xorshift - // Also some drama: , + // Algorithm "xor" from p. 4 of Marsaglia, "Xorshift RNGs" + // See + // https://en.wikipedia.org/wiki/Xorshift + // Also some drama: , - uint32_t x = *seed; - x ^= x << 13; - x ^= x >> 17; - x ^= x << 5; - return *seed = x; + uint32_t x = *seed; + x ^= x << 13; + x ^= x >> 17; + x ^= x << 5; + return *seed = x; } // Distribution & sampling functions -float rand_0_to_1(uint32_t* seed){ - return ((float) xorshift32(seed)) / ((float) UINT32_MAX); - /* +float rand_0_to_1(uint32_t* seed) +{ + return ((float)xorshift32(seed)) / ((float)UINT32_MAX); + /* uint32_t x = *seed; x ^= x << 13; x ^= x >> 17; x ^= x << 5; return ((float)(*seed = x))/((float) UINT32_MAX); */ - // previously: - // ((float)rand_r(seed) / (float)RAND_MAX) - // and before that: rand, but it wasn't thread-safe. - // See: for why to use rand_r: - // rand() is not thread-safe, as it relies on (shared) hidden seed. + // previously: + // ((float)rand_r(seed) / (float)RAND_MAX) + // and before that: rand, but it wasn't thread-safe. + // See: for why to use rand_r: + // rand() is not thread-safe, as it relies on (shared) hidden seed. } float rand_float(float max, uint32_t* seed) @@ -153,28 +154,28 @@ float random_to(float low, float high, uint32_t* seed) void mixture(float (*samplers[])(uint32_t*), float* weights, int n_dists, float** results, int n_threads) { // You can see a simpler version of this function in the git history - // or in C-02-better-algorithm-one-thread/ + // or in alt/C-02-better-algorithm-one-thread/ float sum_weights = array_sum(weights, n_dists); - float* cumsummed_normalized_weights = malloc(n_dists * sizeof(float)); - cumsummed_normalized_weights[0] = weights[0]/sum_weights; + float* cumsummed_normalized_weights = malloc(n_dists * sizeof(float)); + 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; + cumsummed_normalized_weights[i] = cumsummed_normalized_weights[i - 1] + weights[i] / sum_weights; } //create var holders float p1; int sample_index, i, split_array_length; - - // uint32_t* seeds[n_threads]; + + // uint32_t* seeds[n_threads]; uint32_t** seeds = malloc(n_threads * sizeof(uint32_t*)); - for (uint32_t i = 0; i < n_threads; i++) { + for (uint32_t i = 0; i < n_threads; i++) { seeds[i] = malloc(sizeof(uint32_t)); *seeds[i] = i + 1; // xorshift can't start with 0 } - #pragma omp parallel private(i, p1, sample_index, split_array_length) +#pragma omp parallel private(i, p1, sample_index, split_array_length) { - #pragma omp for +#pragma omp for for (i = 0; i < n_threads; i++) { split_array_length = split_array_get_length(i, N, n_threads); for (int j = 0; j < split_array_length; j++) { @@ -190,11 +191,11 @@ void mixture(float (*samplers[])(uint32_t*), float* weights, int n_dists, float* } // free(normalized_weights); // free(cummulative_weights); - free(cumsummed_normalized_weights); + free(cumsummed_normalized_weights); for (uint32_t i = 0; i < n_threads; i++) { free(seeds[i]); } - free(seeds); + free(seeds); } // Functions used for the BOTEC.