summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--COUNT/baselines/cpu/app_baseline.c289
-rw-r--r--COUNT/dpu/task.c138
-rw-r--r--COUNT/host/app.c529
-rwxr-xr-xCOUNT/support/common.h15
-rw-r--r--COUNT/support/params.h95
-rwxr-xr-xCOUNT/support/timer.h146
6 files changed, 664 insertions, 548 deletions
diff --git a/COUNT/baselines/cpu/app_baseline.c b/COUNT/baselines/cpu/app_baseline.c
index 4e96276..13e3f51 100644
--- a/COUNT/baselines/cpu/app_baseline.c
+++ b/COUNT/baselines/cpu/app_baseline.c
@@ -24,7 +24,7 @@
#include <numaif.h>
#include <numa.h>
-void* mp_pages[1];
+void *mp_pages[1];
int mp_status[1];
int mp_nodes[1];
int numa_node_in = -1;
@@ -43,71 +43,70 @@ volatile int total_count;
// Params ---------------------------------------------------------------------
typedef struct Params {
- char* dpu_type;
- int input_size;
- int n_warmup;
- int n_reps;
- int n_threads;
+ char *dpu_type;
+ int input_size;
+ int n_warmup;
+ int n_reps;
+ int n_threads;
#if NUMA
- struct bitmask* bitmask_in;
- struct bitmask* bitmask_out;
- int numa_node_cpu;
+ struct bitmask *bitmask_in;
+ struct bitmask *bitmask_out;
+ int numa_node_cpu;
#endif
-}Params;
+} Params;
struct Params p;
static T *A;
-bool pred(const T x){
- return (x % 2) == 0;
+bool pred(const T x)
+{
+ return (x % 2) == 0;
}
-
-void create_test_file(unsigned int nr_elements) {
- //srand(0);
+void create_test_file(unsigned int nr_elements)
+{
+ //srand(0);
#if NUMA
- if (p.bitmask_in) {
- numa_set_membind(p.bitmask_in);
- numa_free_nodemask(p.bitmask_in);
- }
- A = (T*) numa_alloc(nr_elements * sizeof(T));
+ if (p.bitmask_in) {
+ numa_set_membind(p.bitmask_in);
+ numa_free_nodemask(p.bitmask_in);
+ }
+ A = (T *) numa_alloc(nr_elements * sizeof(T));
#else
- A = (T*) malloc(nr_elements * sizeof(T));
+ A = (T *) malloc(nr_elements * sizeof(T));
#endif
#if NUMA
- struct bitmask *bitmask_all = numa_allocate_nodemask();
- numa_bitmask_setall(bitmask_all);
- numa_set_membind(bitmask_all);
- numa_free_nodemask(bitmask_all);
+ struct bitmask *bitmask_all = numa_allocate_nodemask();
+ numa_bitmask_setall(bitmask_all);
+ numa_set_membind(bitmask_all);
+ numa_free_nodemask(bitmask_all);
#endif
- for (unsigned int i = 0; i < nr_elements; i++) {
- //A[i] = (unsigned int) (rand());
- A[i] = i+1;
- }
+ for (unsigned int i = 0; i < nr_elements; i++) {
+ //A[i] = (unsigned int) (rand());
+ A[i] = i + 1;
+ }
#if NUMA
- mp_pages[0] = A;
- if (move_pages(0, 1, mp_pages, NULL, mp_status, 0) == -1) {
- perror("move_pages(A)");
- }
- else if (mp_status[0] < 0) {
- printf("move_pages error: %d", mp_status[0]);
- }
- else {
- numa_node_in = mp_status[0];
- }
-
- numa_node_cpu = p.numa_node_cpu;
- if (numa_node_cpu != -1) {
- if (numa_run_on_node(numa_node_cpu) == -1) {
- perror("numa_run_on_node");
- numa_node_cpu = -1;
- }
- }
+ mp_pages[0] = A;
+ if (move_pages(0, 1, mp_pages, NULL, mp_status, 0) == -1) {
+ perror("move_pages(A)");
+ } else if (mp_status[0] < 0) {
+ printf("move_pages error: %d", mp_status[0]);
+ } else {
+ numa_node_in = mp_status[0];
+ }
+
+ numa_node_cpu = p.numa_node_cpu;
+ if (numa_node_cpu != -1) {
+ if (numa_run_on_node(numa_node_cpu) == -1) {
+ perror("numa_run_on_node");
+ numa_node_cpu = -1;
+ }
+ }
#endif
}
@@ -115,132 +114,152 @@ void create_test_file(unsigned int nr_elements) {
/**
* @brief compute output in the host
*/
-static int count_host(int size, int t) {
- int count = 0;
-
- omp_set_num_threads(t);
- #pragma omp parallel for reduction(+:count)
- for(int my = 0; my < size; my++) {
- if(!pred(A[my])) {
- count++;
- }
- }
- return count;
+static int count_host(int size, int t)
+{
+ int count = 0;
+
+ omp_set_num_threads(t);
+#pragma omp parallel for reduction(+:count)
+ for (int my = 0; my < size; my++) {
+ if (!pred(A[my])) {
+ count++;
+ }
+ }
+ return count;
}
-void usage() {
- fprintf(stderr,
- "\nUsage: ./program [options]"
- "\n"
- "\nGeneral options:"
- "\n -h help"
- "\n -d <D> DPU type (default=fsim)"
- "\n -t <T> # of threads (default=8)"
- "\n -w <W> # of untimed warmup iterations (default=2)"
- "\n -e <E> # of timed repetition iterations (default=5)"
- "\n"
- "\nBenchmark-specific options:"
- "\n -i <I> input size (default=8M elements)"
- "\n");
+void usage()
+{
+ fprintf(stderr,
+ "\nUsage: ./program [options]"
+ "\n"
+ "\nGeneral options:"
+ "\n -h help"
+ "\n -d <D> DPU type (default=fsim)"
+ "\n -t <T> # of threads (default=8)"
+ "\n -w <W> # of untimed warmup iterations (default=2)"
+ "\n -e <E> # of timed repetition iterations (default=5)"
+ "\n"
+ "\nBenchmark-specific options:"
+ "\n -i <I> input size (default=8M elements)" "\n");
}
-void input_params(int argc, char **argv) {
- p.input_size = 16 << 20;
- p.n_warmup = 1;
- p.n_reps = 3;
- p.n_threads = 5;
+void input_params(int argc, char **argv)
+{
+ p.input_size = 16 << 20;
+ p.n_warmup = 1;
+ p.n_reps = 3;
+ p.n_threads = 5;
#if NUMA
- p.bitmask_in = NULL;
- p.bitmask_out = NULL;
- p.numa_node_cpu = -1;
+ p.bitmask_in = NULL;
+ p.bitmask_out = NULL;
+ p.numa_node_cpu = -1;
#endif
- int opt;
- while((opt = getopt(argc, argv, "hi:w:e:t:a:b:c:")) >= 0) {
- switch(opt) {
- case 'h':
- usage();
- exit(0);
- break;
- case 'i': p.input_size = atoi(optarg); break;
- case 'w': p.n_warmup = atoi(optarg); break;
- case 'e': p.n_reps = atoi(optarg); break;
- case 't': p.n_threads = atoi(optarg); break;
+ int opt;
+ while ((opt = getopt(argc, argv, "hi:w:e:t:a:b:c:")) >= 0) {
+ switch (opt) {
+ case 'h':
+ usage();
+ exit(0);
+ break;
+ case 'i':
+ p.input_size = atoi(optarg);
+ break;
+ case 'w':
+ p.n_warmup = atoi(optarg);
+ break;
+ case 'e':
+ p.n_reps = atoi(optarg);
+ break;
+ case 't':
+ p.n_threads = atoi(optarg);
+ break;
#if NUMA
- case 'a': p.bitmask_in = numa_parse_nodestring(optarg); break;
- case 'b': p.bitmask_out = numa_parse_nodestring(optarg); break;
- case 'c': p.numa_node_cpu = atoi(optarg); break;
+ case 'a':
+ p.bitmask_in = numa_parse_nodestring(optarg);
+ break;
+ case 'b':
+ p.bitmask_out = numa_parse_nodestring(optarg);
+ break;
+ case 'c':
+ p.numa_node_cpu = atoi(optarg);
+ break;
#endif
- default:
- fprintf(stderr, "\nUnrecognized option!\n");
- usage();
- exit(0);
- }
- }
- assert(p.n_threads > 0 && "Invalid # of ranks!");
+ default:
+ fprintf(stderr, "\nUnrecognized option!\n");
+ usage();
+ exit(0);
+ }
+ }
+ assert(p.n_threads > 0 && "Invalid # of ranks!");
}
/**
* @brief Main of the Host Application.
*/
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
- input_params(argc, argv);
+ input_params(argc, argv);
- const unsigned int file_size = p.input_size;
+ const unsigned int file_size = p.input_size;
- // Create an input file with arbitrary data.
- create_test_file(file_size);
+ // Create an input file with arbitrary data.
+ create_test_file(file_size);
#if WITH_BENCHMARK
- Timer timer;
+ Timer timer;
#endif
#if NOP_SYNC
- for(int rep = 0; rep < 200000; rep++) {
- asm volatile("nop" ::);
- }
+ for (int rep = 0; rep < 200000; rep++) {
+ asm volatile ("nop"::);
+ }
#endif
- for(int rep = 0; rep < p.n_warmup + p.n_reps; rep++) {
- start(&timer, 0, 0);
- total_count = count_host(file_size, p.n_threads);
- stop(&timer, 0);
+ for (int rep = 0; rep < p.n_warmup + p.n_reps; rep++) {
+ start(&timer, 0, 0);
+ total_count = count_host(file_size, p.n_threads);
+ stop(&timer, 0);
#if WITH_BENCHMARK
- unsigned int nr_threads = 0;
+ unsigned int nr_threads = 0;
#pragma omp parallel
#pragma omp atomic
- nr_threads++;
+ nr_threads++;
- if (rep >= p.n_warmup) {
- printf("[::] COUNT-CPU | n_threads=%d e_type=%s n_elements=%d"
+ if (rep >= p.n_warmup) {
+ printf
+ ("[::] COUNT-CPU | n_threads=%d e_type=%s n_elements=%d"
#if NUMA
- " numa_node_in=%d numa_node_out=%d numa_node_cpu=%d numa_distance_in_cpu=%d numa_distance_cpu_out=%d"
+ " numa_node_in=%d numa_node_out=%d numa_node_cpu=%d numa_distance_in_cpu=%d numa_distance_cpu_out=%d"
#endif
- " | throughput_MBps=%f",
- nr_threads, XSTR(T), file_size,
+ " | throughput_MBps=%f",
+ nr_threads, XSTR(T), file_size,
#if NUMA
- numa_node_in, numa_node_out, numa_node_cpu, numa_distance(numa_node_in, numa_node_cpu), numa_distance(numa_node_cpu, numa_node_out),
+ numa_node_in, numa_node_out, numa_node_cpu,
+ numa_distance(numa_node_in, numa_node_cpu),
+ numa_distance(numa_node_cpu, numa_node_out),
#endif
- file_size * 2 * sizeof(T) / timer.time[0]);
- printf(" throughput_MOpps=%f",
- file_size / timer.time[0]);
- printall(&timer, 0);
- }
-#endif // WITH_BENCHMARK
- }
+ file_size * 2 * sizeof(T) / timer.time[0]);
+ printf(" throughput_MOpps=%f",
+ file_size / timer.time[0]);
+ printall(&timer, 0);
+ }
+#endif // WITH_BENCHMARK
+ }
#if NOP_SYNC
- for(int rep = 0; rep < 200000; rep++) {
- asm volatile("nop" ::);
- }
+ for (int rep = 0; rep < 200000; rep++) {
+ asm volatile ("nop"::);
+ }
#endif
#if NUMA
- numa_free(A, file_size * sizeof(T));
+ numa_free(A, file_size * sizeof(T));
#else
- free(A);
+ free(A);
#endif
- return 0;
+ return 0;
}
diff --git a/COUNT/dpu/task.c b/COUNT/dpu/task.c
index b2ed79b..8ba6aaf 100644
--- a/COUNT/dpu/task.c
+++ b/COUNT/dpu/task.c
@@ -21,33 +21,36 @@ uint32_t message[NR_TASKLETS];
uint32_t message_partial_count;
// COUNT in each tasklet
-static unsigned int count(T *input){
- unsigned int cnt = 0;
- #pragma unroll
- for(unsigned int j = 0; j < REGS; j++) {
- if(!pred(input[j])) {
- cnt++;
- }
- }
- return cnt;
+static unsigned int count(T *input)
+{
+ unsigned int cnt = 0;
+#pragma unroll
+ for (unsigned int j = 0; j < REGS; j++) {
+ if (!pred(input[j])) {
+ cnt++;
+ }
+ }
+ return cnt;
}
// Handshake with adjacent tasklets
-static unsigned int handshake_sync(unsigned int l_count, unsigned int tasklet_id){
- unsigned int p_count;
- // Wait and read message
- if(tasklet_id != 0){
- handshake_wait_for(tasklet_id - 1);
- p_count = message[tasklet_id];
- } else {
- p_count = 0;
- }
- // Write message and notify
- if(tasklet_id < NR_TASKLETS - 1){
- message[tasklet_id + 1] = p_count + l_count;
- handshake_notify();
- }
- return p_count;
+static unsigned int handshake_sync(unsigned int l_count,
+ unsigned int tasklet_id)
+{
+ unsigned int p_count;
+ // Wait and read message
+ if (tasklet_id != 0) {
+ handshake_wait_for(tasklet_id - 1);
+ p_count = message[tasklet_id];
+ } else {
+ p_count = 0;
+ }
+ // Write message and notify
+ if (tasklet_id < NR_TASKLETS - 1) {
+ message[tasklet_id + 1] = p_count + l_count;
+ handshake_notify();
+ }
+ return p_count;
}
// Barrier
@@ -55,63 +58,70 @@ BARRIER_INIT(my_barrier, NR_TASKLETS);
extern int main_kernel1(void);
-int (*kernels[nr_kernels])(void) = {main_kernel1};
+int (*kernels[nr_kernels])(void) = { main_kernel1 };
-int main(void) {
- // Kernel
- return kernels[DPU_INPUT_ARGUMENTS.kernel]();
+int main(void)
+{
+ // Kernel
+ return kernels[DPU_INPUT_ARGUMENTS.kernel] ();
}
// main_kernel1
-int main_kernel1() {
- unsigned int tasklet_id = me();
+int main_kernel1()
+{
+ unsigned int tasklet_id = me();
#if PRINT
- printf("tasklet_id = %u\n", tasklet_id);
+ printf("tasklet_id = %u\n", tasklet_id);
#endif
- if (tasklet_id == 0){ // Initialize once the cycle counter
- mem_reset(); // Reset the heap
- }
- // Barrier
- barrier_wait(&my_barrier);
+ if (tasklet_id == 0) { // Initialize once the cycle counter
+ mem_reset(); // Reset the heap
+ }
+ // Barrier
+ barrier_wait(&my_barrier);
- dpu_results_t *result = &DPU_RESULTS[tasklet_id];
+ dpu_results_t *result = &DPU_RESULTS[tasklet_id];
- uint32_t input_size_dpu_bytes = DPU_INPUT_ARGUMENTS.size;
+ uint32_t input_size_dpu_bytes = DPU_INPUT_ARGUMENTS.size;
- // Address of the current processing block in MRAM
- uint32_t base_tasklet = tasklet_id << BLOCK_SIZE_LOG2;
- uint32_t mram_base_addr_A = (uint32_t)DPU_MRAM_HEAP_POINTER;
+ // Address of the current processing block in MRAM
+ uint32_t base_tasklet = tasklet_id << BLOCK_SIZE_LOG2;
+ uint32_t mram_base_addr_A = (uint32_t) DPU_MRAM_HEAP_POINTER;
- // Initialize a local cache to store the MRAM block
- T *cache_A = (T *) mem_alloc(BLOCK_SIZE);
+ // Initialize a local cache to store the MRAM block
+ T *cache_A = (T *) mem_alloc(BLOCK_SIZE);
- // Initialize shared variable
- if(tasklet_id == NR_TASKLETS - 1)
- message_partial_count = 0;
- // Barrier
- barrier_wait(&my_barrier);
+ // Initialize shared variable
+ if (tasklet_id == NR_TASKLETS - 1)
+ message_partial_count = 0;
+ // Barrier
+ barrier_wait(&my_barrier);
- for(unsigned int byte_index = base_tasklet; byte_index < input_size_dpu_bytes; byte_index += BLOCK_SIZE * NR_TASKLETS){
+ for (unsigned int byte_index = base_tasklet;
+ byte_index < input_size_dpu_bytes;
+ byte_index += BLOCK_SIZE * NR_TASKLETS) {
- // Load cache with current MRAM block
- mram_read((__mram_ptr void const*)(mram_base_addr_A + byte_index), cache_A, BLOCK_SIZE);
+ // Load cache with current MRAM block
+ mram_read((__mram_ptr void const *)(mram_base_addr_A +
+ byte_index), cache_A,
+ BLOCK_SIZE);
- // COUNT in each tasklet
- uint32_t l_count = count(cache_A);
+ // COUNT in each tasklet
+ uint32_t l_count = count(cache_A);
- // Sync with adjacent tasklets
- uint32_t p_count = handshake_sync(l_count, tasklet_id);
+ // Sync with adjacent tasklets
+ uint32_t p_count = handshake_sync(l_count, tasklet_id);
- // Barrier
- barrier_wait(&my_barrier);
+ // Barrier
+ barrier_wait(&my_barrier);
- // Total count in this DPU
- if(tasklet_id == NR_TASKLETS - 1){
- result->t_count = message_partial_count + p_count + l_count;
- message_partial_count = result->t_count;
- }
+ // Total count in this DPU
+ if (tasklet_id == NR_TASKLETS - 1) {
+ result->t_count =
+ message_partial_count + p_count + l_count;
+ message_partial_count = result->t_count;
+ }
- }
+ }
- return 0;
+ return 0;
}
diff --git a/COUNT/host/app.c b/COUNT/host/app.c
index 9ea6bea..dad674f 100644
--- a/COUNT/host/app.c
+++ b/COUNT/host/app.c
@@ -33,287 +33,350 @@
#include <dpu_target_macros.h>
// Pointer declaration
-static T* A;
+static T *A;
// Create input arrays
-static void read_input(T* A, unsigned int nr_elements, unsigned int nr_elements_round) {
- //srand(0);
- printf("nr_elements\t%u\t", nr_elements);
- for (unsigned int i = 0; i < nr_elements; i++) {
- //A[i] = (T) (rand());
- A[i] = i + 1;
- }
- for (unsigned int i = nr_elements; i < nr_elements_round; i++) { // Complete with removable elements
- A[i] = 0;
- }
+static void read_input(T *A, unsigned int nr_elements,
+ unsigned int nr_elements_round)
+{
+ //srand(0);
+ printf("nr_elements\t%u\t", nr_elements);
+ for (unsigned int i = 0; i < nr_elements; i++) {
+ //A[i] = (T) (rand());
+ A[i] = i + 1;
+ }
+ for (unsigned int i = nr_elements; i < nr_elements_round; i++) { // Complete with removable elements
+ A[i] = 0;
+ }
}
// Compute output in the host
-static unsigned int count_host(T* A, unsigned int nr_elements) {
- unsigned int count = 0;
- for (unsigned int i = 0; i < nr_elements; i++) {
- if(!pred(A[i])) {
- count++;
- }
- }
- return count;
+static unsigned int count_host(T *A, unsigned int nr_elements)
+{
+ unsigned int count = 0;
+ for (unsigned int i = 0; i < nr_elements; i++) {
+ if (!pred(A[i])) {
+ count++;
+ }
+ }
+ return count;
}
// Main of the Host Application
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
- struct Params p = input_params(argc, argv);
+ struct Params p = input_params(argc, argv);
- struct dpu_set_t dpu_set, dpu;
- uint32_t nr_of_dpus;
- uint32_t nr_of_ranks;
+ struct dpu_set_t dpu_set, dpu;
+ uint32_t nr_of_dpus;
+ uint32_t nr_of_ranks;
- // Timer declaration
- Timer timer;
+ // Timer declaration
+ Timer timer;
- int numa_node_rank = -2;
+ int numa_node_rank = -2;
- // Allocate DPUs and load binary
+ // Allocate DPUs and load binary
#if !WITH_ALLOC_OVERHEAD
- DPU_ASSERT(dpu_alloc(NR_DPUS, NULL, &dpu_set));
- timer.time[TMR_ALLOC] = 0; // alloc
+ DPU_ASSERT(dpu_alloc(NR_DPUS, NULL, &dpu_set));
+ timer.time[TMR_ALLOC] = 0; // alloc
#endif
#if !WITH_LOAD_OVERHEAD
- DPU_ASSERT(dpu_load(dpu_set, DPU_BINARY, NULL));
- DPU_ASSERT(dpu_get_nr_dpus(dpu_set, &nr_of_dpus));
- DPU_ASSERT(dpu_get_nr_ranks(dpu_set, &nr_of_ranks));
- assert(nr_of_dpus == NR_DPUS);
- timer.time[TMR_LOAD] = 0; // load
+ DPU_ASSERT(dpu_load(dpu_set, DPU_BINARY, NULL));
+ DPU_ASSERT(dpu_get_nr_dpus(dpu_set, &nr_of_dpus));
+ DPU_ASSERT(dpu_get_nr_ranks(dpu_set, &nr_of_ranks));
+ assert(nr_of_dpus == NR_DPUS);
+ timer.time[TMR_LOAD] = 0; // load
#endif
#if !WITH_FREE_OVERHEAD
- timer.time[TMR_FREE] = 0; // free
+ timer.time[TMR_FREE] = 0; // free
#endif
#if ENERGY
- struct dpu_probe_t probe;
- DPU_ASSERT(dpu_probe_init("energy_probe", &probe));
+ struct dpu_probe_t probe;
+ DPU_ASSERT(dpu_probe_init("energy_probe", &probe));
#endif
- unsigned int i = 0;
- uint32_t accum = 0;
- uint32_t total_count = 0;
+ unsigned int i = 0;
+ uint32_t accum = 0;
+ uint32_t total_count = 0;
- const unsigned int input_size = p.exp == 0 ? p.input_size * NR_DPUS : p.input_size; // Total input size (weak or strong scaling)
- const unsigned int input_size_dpu_ = divceil(input_size, NR_DPUS); // Input size per DPU (max.)
- const unsigned int input_size_dpu_round =
- (input_size_dpu_ % (NR_TASKLETS * REGS) != 0) ? roundup(input_size_dpu_, (NR_TASKLETS * REGS)) : input_size_dpu_; // Input size per DPU (max.), 8-byte aligned
+ const unsigned int input_size = p.exp == 0 ? p.input_size * NR_DPUS : p.input_size; // Total input size (weak or strong scaling)
+ const unsigned int input_size_dpu_ = divceil(input_size, NR_DPUS); // Input size per DPU (max.)
+ const unsigned int input_size_dpu_round = (input_size_dpu_ % (NR_TASKLETS * REGS) != 0) ? roundup(input_size_dpu_, (NR_TASKLETS * REGS)) : input_size_dpu_; // Input size per DPU (max.), 8-byte aligned
- // Input allocation
- A = malloc(input_size_dpu_round * NR_DPUS * sizeof(T));
- T *bufferA = A;
+ // Input allocation
+ A = malloc(input_size_dpu_round * NR_DPUS * sizeof(T));
+ T *bufferA = A;
- dpu_results_t* results_retrieve[NR_DPUS];
- for (i = 0; i < NR_DPUS; i++) {
- results_retrieve[i] = (dpu_results_t*)malloc(NR_TASKLETS * sizeof(dpu_results_t));
- }
+ dpu_results_t *results_retrieve[NR_DPUS];
+ for (i = 0; i < NR_DPUS; i++) {
+ results_retrieve[i] =
+ (dpu_results_t *) malloc(NR_TASKLETS *
+ sizeof(dpu_results_t));
+ }
- // Create an input file with arbitrary data
- read_input(A, input_size, input_size_dpu_round * NR_DPUS);
+ // Create an input file with arbitrary data
+ read_input(A, input_size, input_size_dpu_round * NR_DPUS);
- printf("NR_TASKLETS\t%d\tBL\t%d\n", NR_TASKLETS, BL);
+ printf("NR_TASKLETS\t%d\tBL\t%d\n", NR_TASKLETS, BL);
- // Loop over main kernel
- for(int rep = 0; rep < p.n_warmup + p.n_reps; rep++) {
+ // Loop over main kernel
+ for (int rep = 0; rep < p.n_warmup + p.n_reps; rep++) {
#if WITH_ALLOC_OVERHEAD
- if(rep >= p.n_warmup) {
- start(&timer, TMR_ALLOC, 0);
- }
- DPU_ASSERT(dpu_alloc(NR_DPUS, NULL, &dpu_set));
- if(rep >= p.n_warmup) {
- stop(&timer, TMR_ALLOC);
- }
+ if (rep >= p.n_warmup) {
+ start(&timer, TMR_ALLOC, 0);
+ }
+ DPU_ASSERT(dpu_alloc(NR_DPUS, NULL, &dpu_set));
+ if (rep >= p.n_warmup) {
+ stop(&timer, TMR_ALLOC);
+ }
#endif
#if WITH_LOAD_OVERHEAD
- if(rep >= p.n_warmup) {
- start(&timer, TMR_LOAD, 0);
- }
- DPU_ASSERT(dpu_load(dpu_set, DPU_BINARY, NULL));
- if(rep >= p.n_warmup) {
- stop(&timer, TMR_LOAD);
- }
- DPU_ASSERT(dpu_get_nr_dpus(dpu_set, &nr_of_dpus));
- DPU_ASSERT(dpu_get_nr_ranks(dpu_set, &nr_of_ranks));
- assert(nr_of_dpus == NR_DPUS);
+ if (rep >= p.n_warmup) {
+ start(&timer, TMR_LOAD, 0);
+ }
+ DPU_ASSERT(dpu_load(dpu_set, DPU_BINARY, NULL));
+ if (rep >= p.n_warmup) {
+ stop(&timer, TMR_LOAD);
+ }
+ DPU_ASSERT(dpu_get_nr_dpus(dpu_set, &nr_of_dpus));
+ DPU_ASSERT(dpu_get_nr_ranks(dpu_set, &nr_of_ranks));
+ assert(nr_of_dpus == NR_DPUS);
#endif
- // int prev_rank_id = -1;
- int rank_id = -1;
- DPU_FOREACH (dpu_set, dpu) {
- rank_id = dpu_get_rank_id(dpu_get_rank(dpu_from_set(dpu))) & DPU_TARGET_MASK;
- if ((numa_node_rank != -2) && numa_node_rank != dpu_get_rank_numa_node(dpu_get_rank(dpu_from_set(dpu)))) {
- numa_node_rank = -1;
- } else {
- numa_node_rank = dpu_get_rank_numa_node(dpu_get_rank(dpu_from_set(dpu)));
- }
- /*
- if (rank_id != prev_rank_id) {
- printf("/dev/dpu_rank%d @ NUMA node %d\n", rank_id, numa_node_rank);
- prev_rank_id = rank_id;
- }
- */
- }
-
- // Compute output on CPU (performance comparison and verification purposes)
- if(rep >= p.n_warmup)
- start(&timer, TMR_CPU, 0);
- total_count = count_host(A, input_size);
- if(rep >= p.n_warmup)
- stop(&timer, TMR_CPU);
-
- printf("Load input data\n");
- if(rep >= p.n_warmup)
- start(&timer, TMR_WRITE, 0);
- // Input arguments
- const unsigned int input_size_dpu = input_size_dpu_round;
- unsigned int kernel = 0;
- dpu_arguments_t input_arguments = {input_size_dpu * sizeof(T), kernel};
- // Copy input arrays
- i = 0;
- DPU_FOREACH(dpu_set, dpu, i) {
- DPU_ASSERT(dpu_prepare_xfer(dpu, &input_arguments));
- }
- DPU_ASSERT(dpu_push_xfer(dpu_set, DPU_XFER_TO_DPU, "DPU_INPUT_ARGUMENTS", 0, sizeof(input_arguments), DPU_XFER_DEFAULT));
- DPU_FOREACH(dpu_set, dpu, i) {
- DPU_ASSERT(dpu_prepare_xfer(dpu, bufferA + input_size_dpu * i));
- }
- DPU_ASSERT(dpu_push_xfer(dpu_set, DPU_XFER_TO_DPU, DPU_MRAM_HEAP_POINTER_NAME, 0, input_size_dpu * sizeof(T), DPU_XFER_DEFAULT));
- if(rep >= p.n_warmup)
- stop(&timer, TMR_WRITE);
-
- printf("Run program on DPU(s) \n");
- // Run DPU kernel
- if(rep >= p.n_warmup) {
- start(&timer, TMR_KERNEL, 0);
- #if ENERGY
- DPU_ASSERT(dpu_probe_start(&probe));
- #endif
- }
- DPU_ASSERT(dpu_launch(dpu_set, DPU_SYNCHRONOUS));
- if(rep >= p.n_warmup) {
- stop(&timer, TMR_KERNEL);
- #if ENERGY
- DPU_ASSERT(dpu_probe_stop(&probe));
- #endif
- }
-
+ // int prev_rank_id = -1;
+ int rank_id = -1;
+ DPU_FOREACH(dpu_set, dpu) {
+ rank_id =
+ dpu_get_rank_id(dpu_get_rank(dpu_from_set(dpu))) &
+ DPU_TARGET_MASK;
+ if ((numa_node_rank != -2)
+ && numa_node_rank !=
+ dpu_get_rank_numa_node(dpu_get_rank
+ (dpu_from_set(dpu)))) {
+ numa_node_rank = -1;
+ } else {
+ numa_node_rank =
+ dpu_get_rank_numa_node(dpu_get_rank
+ (dpu_from_set(dpu)));
+ }
+ /*
+ if (rank_id != prev_rank_id) {
+ printf("/dev/dpu_rank%d @ NUMA node %d\n", rank_id, numa_node_rank);
+ prev_rank_id = rank_id;
+ }
+ */
+ }
+
+ // Compute output on CPU (performance comparison and verification purposes)
+ if (rep >= p.n_warmup)
+ start(&timer, TMR_CPU, 0);
+ total_count = count_host(A, input_size);
+ if (rep >= p.n_warmup)
+ stop(&timer, TMR_CPU);
+
+ printf("Load input data\n");
+ if (rep >= p.n_warmup)
+ start(&timer, TMR_WRITE, 0);
+ // Input arguments
+ const unsigned int input_size_dpu = input_size_dpu_round;
+ unsigned int kernel = 0;
+ dpu_arguments_t input_arguments =
+ { input_size_dpu * sizeof(T), kernel };
+ // Copy input arrays
+ i = 0;
+ DPU_FOREACH(dpu_set, dpu, i) {
+ DPU_ASSERT(dpu_prepare_xfer(dpu, &input_arguments));
+ }
+ DPU_ASSERT(dpu_push_xfer
+ (dpu_set, DPU_XFER_TO_DPU, "DPU_INPUT_ARGUMENTS", 0,
+ sizeof(input_arguments), DPU_XFER_DEFAULT));
+ DPU_FOREACH(dpu_set, dpu, i) {
+ DPU_ASSERT(dpu_prepare_xfer
+ (dpu, bufferA + input_size_dpu * i));
+ }
+ DPU_ASSERT(dpu_push_xfer
+ (dpu_set, DPU_XFER_TO_DPU,
+ DPU_MRAM_HEAP_POINTER_NAME, 0,
+ input_size_dpu * sizeof(T), DPU_XFER_DEFAULT));
+ if (rep >= p.n_warmup)
+ stop(&timer, TMR_WRITE);
+
+ printf("Run program on DPU(s) \n");
+ // Run DPU kernel
+ if (rep >= p.n_warmup) {
+ start(&timer, TMR_KERNEL, 0);
+#if ENERGY
+ DPU_ASSERT(dpu_probe_start(&probe));
+#endif
+ }
+ DPU_ASSERT(dpu_launch(dpu_set, DPU_SYNCHRONOUS));
+ if (rep >= p.n_warmup) {
+ stop(&timer, TMR_KERNEL);
+#if ENERGY
+ DPU_ASSERT(dpu_probe_stop(&probe));
+#endif
+ }
#if PRINT
- {
- unsigned int each_dpu = 0;
- printf("Display DPU Logs\n");
- DPU_FOREACH (dpu_set, dpu) {
- printf("DPU#%d:\n", each_dpu);
- DPU_ASSERT(dpulog_read_for_dpu(dpu.dpu, stdout));
- each_dpu++;
- }
- }
+ {
+ unsigned int each_dpu = 0;
+ printf("Display DPU Logs\n");
+ DPU_FOREACH(dpu_set, dpu) {
+ printf("DPU#%d:\n", each_dpu);
+ DPU_ASSERT(dpulog_read_for_dpu
+ (dpu.dpu, stdout));
+ each_dpu++;
+ }
+ }
#endif
- printf("Retrieve results\n");
- dpu_results_t results[NR_DPUS];
- i = 0;
- accum = 0;
-
- if(rep >= p.n_warmup)
- start(&timer, TMR_READ, 0);
- // PARALLEL RETRIEVE TRANSFER
-
- DPU_FOREACH(dpu_set, dpu, i) {
- DPU_ASSERT(dpu_prepare_xfer(dpu, results_retrieve[i]));
- }
- DPU_ASSERT(dpu_push_xfer(dpu_set, DPU_XFER_FROM_DPU, "DPU_RESULTS", 0, NR_TASKLETS * sizeof(dpu_results_t), DPU_XFER_DEFAULT));
-
- DPU_FOREACH(dpu_set, dpu, i) {
- // Retrieve tasklet timings
- for (unsigned int each_tasklet = 0; each_tasklet < NR_TASKLETS; each_tasklet++) {
- // Count of this DPU
- if(each_tasklet == NR_TASKLETS - 1){
- results[i].t_count = results_retrieve[i][each_tasklet].t_count;
- }
- }
- // Sequential scan
- accum += results[i].t_count;
- }
- if(rep >= p.n_warmup)
- stop(&timer, TMR_READ);
-
- i = 0;
+ printf("Retrieve results\n");
+ dpu_results_t results[NR_DPUS];
+ i = 0;
+ accum = 0;
+
+ if (rep >= p.n_warmup)
+ start(&timer, TMR_READ, 0);
+ // PARALLEL RETRIEVE TRANSFER
+
+ DPU_FOREACH(dpu_set, dpu, i) {
+ DPU_ASSERT(dpu_prepare_xfer(dpu, results_retrieve[i]));
+ }
+ DPU_ASSERT(dpu_push_xfer
+ (dpu_set, DPU_XFER_FROM_DPU, "DPU_RESULTS", 0,
+ NR_TASKLETS * sizeof(dpu_results_t),
+ DPU_XFER_DEFAULT));
+
+ DPU_FOREACH(dpu_set, dpu, i) {
+ // Retrieve tasklet timings
+ for (unsigned int each_tasklet = 0;
+ each_tasklet < NR_TASKLETS; each_tasklet++) {
+ // Count of this DPU
+ if (each_tasklet == NR_TASKLETS - 1) {
+ results[i].t_count =
+ results_retrieve[i][each_tasklet].
+ t_count;
+ }
+ }
+ // Sequential scan
+ accum += results[i].t_count;
+ }
+ if (rep >= p.n_warmup)
+ stop(&timer, TMR_READ);
+
+ i = 0;
#if WITH_ALLOC_OVERHEAD
#if WITH_FREE_OVERHEAD
- if(rep >= p.n_warmup) {
- start(&timer, TMR_FREE, 0);
- }
+ if (rep >= p.n_warmup) {
+ start(&timer, TMR_FREE, 0);
+ }
#endif
- DPU_ASSERT(dpu_free(dpu_set));
+ DPU_ASSERT(dpu_free(dpu_set));
#if WITH_FREE_OVERHEAD
- if(rep >= p.n_warmup) {
- stop(&timer, TMR_FREE);
- }
+ if (rep >= p.n_warmup) {
+ stop(&timer, TMR_FREE);
+ }
#endif
#endif
- // Check output
- bool status = true;
- if(accum != total_count) status = false;
- if (status) {
- printf("[" ANSI_COLOR_GREEN "OK" ANSI_COLOR_RESET "] Outputs are equal\n");
- if (rep >= p.n_warmup) {
- printf("[::] COUNT-UPMEM | n_dpus=%d n_ranks=%d n_tasklets=%d e_type=%s block_size_B=%d n_elements=%d n_elements_per_dpu=%d",
- NR_DPUS, nr_of_ranks, NR_TASKLETS, XSTR(T), BLOCK_SIZE, input_size, input_size_dpu_round);
- printf(" b_with_alloc_overhead=%d b_with_load_overhead=%d b_with_free_overhead=%d numa_node_rank=%d ",
- WITH_ALLOC_OVERHEAD, WITH_LOAD_OVERHEAD, WITH_FREE_OVERHEAD, numa_node_rank);
- printf("| latency_alloc_us=%f latency_load_us=%f latency_cpu_us=%f latency_write_us=%f latency_kernel_us=%f latency_read_us=%f latency_free_us=%f",
- timer.time[0],
- timer.time[1],
- timer.time[2],
- timer.time[3], // write
- timer.time[4], // kernel
- timer.time[5], // read
- timer.time[8]);
- printf(" latency_total_us=%f",
- timer.time[0] + timer.time[1] + timer.time[3] + timer.time[4] + timer.time[5] + timer.time[8]);
- printf(" throughput_cpu_MBps=%f throughput_upmem_kernel_MBps=%f throughput_upmem_total_MBps=%f",
- input_size * sizeof(T) / timer.time[2],
- input_size * sizeof(T) / timer.time[4],
- input_size * sizeof(T) / (timer.time[0] + timer.time[1] + timer.time[3] + timer.time[4] + timer.time[5] + timer.time[8])
- );
- printf(" throughput_upmem_wxr_MBps=%f throughput_upmem_lwxr_MBps=%f throughput_upmem_alwxr_MBps=%f",
- input_size * sizeof(T) / (timer.time[3] + timer.time[4] + timer.time[5]),
- input_size * sizeof(T) / (timer.time[1] + timer.time[3] + timer.time[4] + timer.time[5]),
- input_size * sizeof(T) / (timer.time[0] + timer.time[1] + timer.time[3] + timer.time[4] + timer.time[5]));
- printf(" throughput_cpu_MOpps=%f throughput_upmem_kernel_MOpps=%f throughput_upmem_total_MOpps=%f",
- input_size / timer.time[2],
- input_size / timer.time[4],
- input_size / (timer.time[0] + timer.time[1] + timer.time[3] + timer.time[4] + timer.time[5] + timer.time[8])
- );
- printf(" throughput_upmem_wxr_MOpps=%f throughput_upmem_lwxr_MOpps=%f throughput_upmem_alwxr_MOpps=%f\n",
- input_size / (timer.time[3] + timer.time[4] + timer.time[5]),
- input_size / (timer.time[1] + timer.time[3] + timer.time[4] + timer.time[5]),
- input_size / (timer.time[0] + timer.time[1] + timer.time[3] + timer.time[4] + timer.time[5]));
- }
- } else {
- printf("[" ANSI_COLOR_RED "ERROR" ANSI_COLOR_RESET "] Outputs differ!\n");
- }
- }
-
- #if ENERGY
- double energy;
- DPU_ASSERT(dpu_probe_get(&probe, DPU_ENERGY, DPU_AVERAGE, &energy));
- printf("DPU Energy (J): %f\t", energy);
- #endif
-
- // Deallocation
- free(A);
+ // Check output
+ bool status = true;
+ if (accum != total_count)
+ status = false;
+ if (status) {
+ printf("[" ANSI_COLOR_GREEN "OK" ANSI_COLOR_RESET
+ "] Outputs are equal\n");
+ if (rep >= p.n_warmup) {
+ printf
+ ("[::] COUNT-UPMEM | n_dpus=%d n_ranks=%d n_tasklets=%d e_type=%s block_size_B=%d n_elements=%d n_elements_per_dpu=%d",
+ NR_DPUS, nr_of_ranks, NR_TASKLETS, XSTR(T),
+ BLOCK_SIZE, input_size,
+ input_size_dpu_round);
+ printf
+ (" b_with_alloc_overhead=%d b_with_load_overhead=%d b_with_free_overhead=%d numa_node_rank=%d ",
+ WITH_ALLOC_OVERHEAD, WITH_LOAD_OVERHEAD,
+ WITH_FREE_OVERHEAD, numa_node_rank);
+ printf("| latency_alloc_us=%f latency_load_us=%f latency_cpu_us=%f latency_write_us=%f latency_kernel_us=%f latency_read_us=%f latency_free_us=%f", timer.time[0], timer.time[1], timer.time[2], timer.time[3], // write
+ timer.time[4], // kernel
+ timer.time[5], // read
+ timer.time[8]);
+ printf(" latency_total_us=%f",
+ timer.time[0] + timer.time[1] +
+ timer.time[3] + timer.time[4] +
+ timer.time[5] + timer.time[8]);
+ printf
+ (" throughput_cpu_MBps=%f throughput_upmem_kernel_MBps=%f throughput_upmem_total_MBps=%f",
+ input_size * sizeof(T) / timer.time[2],
+ input_size * sizeof(T) / timer.time[4],
+ input_size * sizeof(T) / (timer.time[0] +
+ timer.time[1] +
+ timer.time[3] +
+ timer.time[4] +
+ timer.time[5] +
+ timer.time[8])
+ );
+ printf
+ (" throughput_upmem_wxr_MBps=%f throughput_upmem_lwxr_MBps=%f throughput_upmem_alwxr_MBps=%f",
+ input_size * sizeof(T) / (timer.time[3] +
+ timer.time[4] +
+ timer.time[5]),
+ input_size * sizeof(T) / (timer.time[1] +
+ timer.time[3] +
+ timer.time[4] +
+ timer.time[5]),
+ input_size * sizeof(T) / (timer.time[0] +
+ timer.time[1] +
+ timer.time[3] +
+ timer.time[4] +
+ timer.time[5]));
+ printf
+ (" throughput_cpu_MOpps=%f throughput_upmem_kernel_MOpps=%f throughput_upmem_total_MOpps=%f",
+ input_size / timer.time[2],
+ input_size / timer.time[4],
+ input_size / (timer.time[0] +
+ timer.time[1] +
+ timer.time[3] +
+ timer.time[4] +
+ timer.time[5] +
+ timer.time[8])
+ );
+ printf
+ (" throughput_upmem_wxr_MOpps=%f throughput_upmem_lwxr_MOpps=%f throughput_upmem_alwxr_MOpps=%f\n",
+ input_size / (timer.time[3] +
+ timer.time[4] +
+ timer.time[5]),
+ input_size / (timer.time[1] +
+ timer.time[3] +
+ timer.time[4] +
+ timer.time[5]),
+ input_size / (timer.time[0] +
+ timer.time[1] +
+ timer.time[3] +
+ timer.time[4] +
+ timer.time[5]));
+ }
+ } else {
+ printf("[" ANSI_COLOR_RED "ERROR" ANSI_COLOR_RESET
+ "] Outputs differ!\n");
+ }
+ }
+
+#if ENERGY
+ double energy;
+ DPU_ASSERT(dpu_probe_get(&probe, DPU_ENERGY, DPU_AVERAGE, &energy));
+ printf("DPU Energy (J): %f\t", energy);
+#endif
+
+ // Deallocation
+ free(A);
#if !WITH_ALLOC_OVERHEAD
- DPU_ASSERT(dpu_free(dpu_set));
+ DPU_ASSERT(dpu_free(dpu_set));
#endif
- return 0;
+ return 0;
}
diff --git a/COUNT/support/common.h b/COUNT/support/common.h
index 72270b0..afd5b2d 100755
--- a/COUNT/support/common.h
+++ b/COUNT/support/common.h
@@ -3,15 +3,15 @@
// Structures used by both the host and the dpu to communicate information
typedef struct {
- uint32_t size;
+ uint32_t size;
enum kernels {
- kernel1 = 0,
- nr_kernels = 1,
+ kernel1 = 0,
+ nr_kernels = 1,
} kernel;
} dpu_arguments_t;
typedef struct {
- uint32_t t_count;
+ uint32_t t_count;
} dpu_results_t;
// Transfer size between MRAM and WRAM
@@ -26,11 +26,12 @@ typedef struct {
// Data type
#define T uint64_t
-#define REGS (BLOCK_SIZE >> 3) // 64 bits
+#define REGS (BLOCK_SIZE >> 3) // 64 bits
// Sample predicate
-bool pred(const T x){
- return (x % 2) == 0;
+bool pred(const T x)
+{
+ return (x % 2) == 0;
}
#ifndef ENERGY
diff --git a/COUNT/support/params.h b/COUNT/support/params.h
index bb86211..dd1505e 100644
--- a/COUNT/support/params.h
+++ b/COUNT/support/params.h
@@ -4,53 +4,62 @@
#include "common.h"
typedef struct Params {
- unsigned int input_size;
- int n_warmup;
- int n_reps;
- int exp;
-}Params;
+ unsigned int input_size;
+ int n_warmup;
+ int n_reps;
+ int exp;
+} Params;
-static void usage() {
- fprintf(stderr,
- "\nUsage: ./program [options]"
- "\n"
- "\nGeneral options:"
- "\n -h help"
- "\n -w <W> # of untimed warmup iterations (default=1)"
- "\n -e <E> # of timed repetition iterations (default=3)"
- "\n -x <X> Weak (0) or strong (1) scaling (default=0)"
- "\n"
- "\nBenchmark-specific options:"
- "\n -i <I> input size (default=3932160 elements)"
- "\n");
+static void usage()
+{
+ fprintf(stderr,
+ "\nUsage: ./program [options]"
+ "\n"
+ "\nGeneral options:"
+ "\n -h help"
+ "\n -w <W> # of untimed warmup iterations (default=1)"
+ "\n -e <E> # of timed repetition iterations (default=3)"
+ "\n -x <X> Weak (0) or strong (1) scaling (default=0)"
+ "\n"
+ "\nBenchmark-specific options:"
+ "\n -i <I> input size (default=3932160 elements)" "\n");
}
-struct Params input_params(int argc, char **argv) {
- struct Params p;
- p.input_size = 3932160;
- p.n_warmup = 1;
- p.n_reps = 3;
- p.exp = 0;
+struct Params input_params(int argc, char **argv)
+{
+ struct Params p;
+ p.input_size = 3932160;
+ p.n_warmup = 1;
+ p.n_reps = 3;
+ p.exp = 0;
- int opt;
- while((opt = getopt(argc, argv, "hi:w:e:x:")) >= 0) {
- switch(opt) {
- case 'h':
- usage();
- exit(0);
- break;
- case 'i': p.input_size = atoi(optarg); break;
- case 'w': p.n_warmup = atoi(optarg); break;
- case 'e': p.n_reps = atoi(optarg); break;
- case 'x': p.exp = atoi(optarg); break;
- default:
- fprintf(stderr, "\nUnrecognized option!\n");
- usage();
- exit(0);
- }
- }
- assert(NR_DPUS > 0 && "Invalid # of dpus!");
+ int opt;
+ while ((opt = getopt(argc, argv, "hi:w:e:x:")) >= 0) {
+ switch (opt) {
+ case 'h':
+ usage();
+ exit(0);
+ break;
+ case 'i':
+ p.input_size = atoi(optarg);
+ break;
+ case 'w':
+ p.n_warmup = atoi(optarg);
+ break;
+ case 'e':
+ p.n_reps = atoi(optarg);
+ break;
+ case 'x':
+ p.exp = atoi(optarg);
+ break;
+ default:
+ fprintf(stderr, "\nUnrecognized option!\n");
+ usage();
+ exit(0);
+ }
+ }
+ assert(NR_DPUS > 0 && "Invalid # of dpus!");
- return p;
+ return p;
}
#endif
diff --git a/COUNT/support/timer.h b/COUNT/support/timer.h
index 3ec6d87..76fbcff 100755
--- a/COUNT/support/timer.h
+++ b/COUNT/support/timer.h
@@ -1,66 +1,80 @@
-/*
- * Copyright (c) 2016 University of Cordoba and University of Illinois
- * All rights reserved.
- *
- * Developed by: IMPACT Research Group
- * University of Cordoba and University of Illinois
- * http://impact.crhc.illinois.edu/
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * with the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * > Redistributions of source code must retain the above copyright notice,
- * this list of conditions and the following disclaimers.
- * > Redistributions in binary form must reproduce the above copyright
- * notice, this list of conditions and the following disclaimers in the
- * documentation and/or other materials provided with the distribution.
- * > Neither the names of IMPACT Research Group, University of Cordoba,
- * University of Illinois nor the names of its contributors may be used
- * to endorse or promote products derived from this Software without
- * specific prior written permission.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS WITH
- * THE SOFTWARE.
- *
- */
-
-#include <sys/time.h>
-
-typedef struct Timer{
-
- struct timeval startTime[9];
- struct timeval stopTime[9];
- double time[9];
-
-}Timer;
-
-void start(Timer *timer, int i, int rep) {
- if(rep == 0) {
- timer->time[i] = 0.0;
- }
- gettimeofday(&timer->startTime[i], NULL);
-}
-
-void stop(Timer *timer, int i) {
- gettimeofday(&timer->stopTime[i], NULL);
- timer->time[i] += (timer->stopTime[i].tv_sec - timer->startTime[i].tv_sec) * 1000000.0 +
- (timer->stopTime[i].tv_usec - timer->startTime[i].tv_usec);
-}
-
-void print(Timer *timer, int i, int REP) { printf("Time (ms): %f\t", timer->time[i] / (1000 * REP)); }
-
-void printall(Timer *timer, int maxt) {
- for (int i = 0; i <= maxt; i++) {
- printf(" timer%d_us=%f", i, timer->time[i]);
- }
- printf("\n");
-}
+/*
+ * Copyright (c) 2016 University of Cordoba and University of Illinois
+ * All rights reserved.
+ *
+ * Developed by: IMPACT Research Group
+ * University of Cordoba and University of Illinois
+ * http://impact.crhc.illinois.edu/
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * with the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * > Redistributions of source code must retain the above copyright notice,
+ * this list of conditions and the following disclaimers.
+ * > Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimers in the
+ * documentation and/or other materials provided with the distribution.
+ * > Neither the names of IMPACT Research Group, University of Cordoba,
+ * University of Illinois nor the names of its contributors may be used
+ * to endorse or promote products derived from this Software without
+ * specific prior written permission.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS WITH
+ * THE SOFTWARE.
+ *
+ */
+
+#include <sys/time.h>
+
+typedef struct Timer {
+ struct timeval startTime[7];
+ struct timeval stopTime[7];
+ double time[7];
+} Timer;
+
+#define TMR_ALLOC 0
+#define TMR_LOAD 1
+#define TMR_CPU 2
+#define TMR_WRITE 3
+#define TMR_KERNEL 4
+#define TMR_READ 5
+#define TMR_FREE 6
+
+void start(Timer *timer, int i, int rep)
+{
+ if (rep == 0) {
+ timer->time[i] = 0.0;
+ }
+ gettimeofday(&timer->startTime[i], NULL);
+}
+
+void stop(Timer *timer, int i)
+{
+ gettimeofday(&timer->stopTime[i], NULL);
+ timer->time[i] +=
+ (timer->stopTime[i].tv_sec -
+ timer->startTime[i].tv_sec) * 1000000.0 +
+ (timer->stopTime[i].tv_usec - timer->startTime[i].tv_usec);
+}
+
+void print(Timer *timer, int i, int REP)
+{
+ printf("Time (ms): %f\t", timer->time[i] / (1000 * REP));
+}
+
+void printall(Timer *timer, int maxt)
+{
+ for (int i = 0; i <= maxt; i++) {
+ printf(" timer%d_us=%f", i, timer->time[i]);
+ }
+ printf("\n");
+}
generated by cgit v1.2.3 (git 2.39.1) at 2025-01-19 03:31:04 +0000