Add COUNT benchmark (based on SEL)

6 files changed, 645 insertions, 0 deletions

diff --git a/COUNT/Makefile b/COUNT/Makefile
new file mode 100644
index 0000000..ed40aac
--- /dev/null
+++ b/COUNT/Makefile
@@ -0,0 +1,40 @@
+NR_TASKLETS ?= 16
+BL ?= 10
+NR_DPUS ?= 1
+ENERGY ?= 0
+WITH_ALLOC_OVERHEAD ?= 0
+WITH_LOAD_OVERHEAD ?= 0
+WITH_FREE_OVERHEAD ?= 0
+
+COMMON_INCLUDES := support
+HOST_SOURCES := $(wildcard host/*.c)
+DPU_SOURCES := $(wildcard dpu/*.c)
+
+COMMON_FLAGS := -Wall -Wextra -g -I${COMMON_INCLUDES}
+HOST_FLAGS := ${COMMON_FLAGS} -std=c11 -O3 `dpu-pkg-config --cflags --libs dpu` -DNR_TASKLETS=${NR_TASKLETS} -DNR_DPUS=${NR_DPUS} -DBL=${BL} -DENERGY=${ENERGY}  -DWITH_ALLOC_OVERHEAD=${WITH_ALLOC_OVERHEAD} -DWITH_LOAD_OVERHEAD=${WITH_LOAD_OVERHEAD} -DWITH_FREE_OVERHEAD=${WITH_FREE_OVERHEAD}
+DPU_FLAGS := ${COMMON_FLAGS} -O2 -DNR_TASKLETS=${NR_TASKLETS} -DBL=${BL}
+
+QUIET = @
+
+ifdef verbose
+	QUIET =
+endif
+
+all: bin/host_code bin/dpu_code
+
+bin:
+	${QUIET}mkdir -p bin
+
+bin/host_code: ${HOST_SOURCES} ${COMMON_INCLUDES} bin
+	${QUIET}${CC} -o $@ ${HOST_SOURCES} ${HOST_FLAGS}
+
+bin/dpu_code: ${DPU_SOURCES} ${COMMON_INCLUDES} bin
+	${QUIET}dpu-upmem-dpurte-clang ${DPU_FLAGS} -o $@ ${DPU_SOURCES}
+
+clean:
+	${QUIET}rm -rf bin
+
+test: all
+	${QUIET}bin/host_code
+
+.PHONY: all clean test
diff --git a/COUNT/dpu/task.c b/COUNT/dpu/task.c
new file mode 100644
index 0000000..b2ed79b
--- /dev/null
+++ b/COUNT/dpu/task.c
@@ -0,0 +1,117 @@
+/*
+* Count with multiple tasklets
+*
+*/
+#include <stdint.h>
+#include <stdio.h>
+#include <defs.h>
+#include <mram.h>
+#include <alloc.h>
+#include <perfcounter.h>
+#include <handshake.h>
+#include <barrier.h>
+
+#include "../support/common.h"
+
+__host dpu_arguments_t DPU_INPUT_ARGUMENTS;
+__host dpu_results_t DPU_RESULTS[NR_TASKLETS];
+
+// Array for communication between adjacent tasklets
+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;
+}
+
+// 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;
+}
+
+// Barrier
+BARRIER_INIT(my_barrier, NR_TASKLETS);
+
+extern int main_kernel1(void);
+
+int (*kernels[nr_kernels])(void) = {main_kernel1};
+
+int main(void) { 
+    // Kernel
+    return kernels[DPU_INPUT_ARGUMENTS.kernel](); 
+}
+
+// main_kernel1
+int main_kernel1() {
+    unsigned int tasklet_id = me();
+#if PRINT
+    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);
+
+    dpu_results_t *result = &DPU_RESULTS[tasklet_id];
+
+    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;
+
+    // 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);
+
+    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);
+
+        // 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);
+
+        // 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;
+        }
+
+    }
+
+    return 0;
+}
diff --git a/COUNT/host/app.c b/COUNT/host/app.c
new file mode 100644
index 0000000..7708f6d
--- /dev/null
+++ b/COUNT/host/app.c
@@ -0,0 +1,319 @@
+/**
+* app.c
+* COUNT Host Application Source File
+*
+*/
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdbool.h>
+#include <string.h>
+#include <dpu.h>
+#include <dpu_log.h>
+#include <unistd.h>
+#include <getopt.h>
+#include <assert.h>
+
+#include "../support/common.h"
+#include "../support/timer.h"
+#include "../support/params.h"
+
+// Define the DPU Binary path as DPU_BINARY here
+#ifndef DPU_BINARY
+#define DPU_BINARY "./bin/dpu_code"
+#endif
+
+#define XSTR(x) STR(x)
+#define STR(x) #x
+
+#if ENERGY
+#include <dpu_probe.h>
+#endif
+
+#include <dpu_management.h>
+#include <dpu_target_macros.h>
+
+// Pointer declaration
+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;
+    }
+}
+
+// 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;
+}
+
+// Main of the Host Application
+int main(int argc, char **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;
+
+    // Timer declaration
+    Timer timer;
+
+    int numa_node_rank = -2;
+
+    // Allocate DPUs and load binary
+#if !WITH_ALLOC_OVERHEAD
+    DPU_ASSERT(dpu_alloc(NR_DPUS, NULL, &dpu_set));
+    timer.time[0] = 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[1] = 0; // load
+#endif
+#if !WITH_FREE_OVERHEAD
+    timer.time[6] = 0; // free
+#endif
+
+#if ENERGY
+    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;
+
+    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;
+
+    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);
+
+    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++) {
+
+#if WITH_ALLOC_OVERHEAD
+        if(rep >= p.n_warmup) {
+            start(&timer, 0, 0);
+        }
+        DPU_ASSERT(dpu_alloc(NR_DPUS, NULL, &dpu_set));
+        if(rep >= p.n_warmup) {
+            stop(&timer, 0);
+        }
+#endif
+#if WITH_LOAD_OVERHEAD
+        if(rep >= p.n_warmup) {
+            start(&timer, 1, 0);
+        }
+        DPU_ASSERT(dpu_load(dpu_set, DPU_BINARY, NULL));
+        if(rep >= p.n_warmup) {
+            stop(&timer, 1);
+        }
+        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, 2, 0);
+        total_count = count_host(A, input_size);
+        if(rep >= p.n_warmup)
+            stop(&timer, 2);
+
+        printf("Load input data\n");
+        if(rep >= p.n_warmup)
+            start(&timer, 3, 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, 3);
+
+        printf("Run program on DPU(s) \n");
+        // Run DPU kernel
+        if(rep >= p.n_warmup) {
+            start(&timer, 4, 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, 4);
+            #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++;
+            }
+        }
+#endif
+
+        printf("Retrieve results\n");
+        dpu_results_t results[NR_DPUS];
+        i = 0;
+        accum = 0;
+
+        if(rep >= p.n_warmup)
+		    start(&timer, 5, 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, 5);
+
+        i = 0;
+
+#if WITH_ALLOC_OVERHEAD
+#if WITH_FREE_OVERHEAD
+        if(rep >= p.n_warmup) {
+            start(&timer, 8, 0);
+        }
+#endif
+        DPU_ASSERT(dpu_free(dpu_set));
+#if WITH_FREE_OVERHEAD
+        if(rep >= p.n_warmup) {
+            stop(&timer, 8);
+        }
+#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);
+
+#if !WITH_ALLOC_OVERHEAD
+    DPU_ASSERT(dpu_free(dpu_set));
+#endif
+
+    return 0;
+}
diff --git a/COUNT/support/common.h b/COUNT/support/common.h
new file mode 100755
index 0000000..72270b0
--- /dev/null
+++ b/COUNT/support/common.h
@@ -0,0 +1,47 @@
+#ifndef _COMMON_H_
+#define _COMMON_H_
+
+// Structures used by both the host and the dpu to communicate information 
+typedef struct {
+    uint32_t size;
+	enum kernels {
+	    kernel1 = 0,
+	    nr_kernels = 1,
+	} kernel;
+} dpu_arguments_t;
+
+typedef struct {
+    uint32_t t_count;
+} dpu_results_t;
+
+// Transfer size between MRAM and WRAM
+#ifdef BL
+#define BLOCK_SIZE_LOG2 BL
+#define BLOCK_SIZE (1 << BLOCK_SIZE_LOG2)
+#else
+#define BLOCK_SIZE_LOG2 8
+#define BLOCK_SIZE (1 << BLOCK_SIZE_LOG2)
+#define BL BLOCK_SIZE_LOG2
+#endif
+
+// Data type
+#define T uint64_t
+#define REGS (BLOCK_SIZE >> 3) // 64 bits
+
+// Sample predicate
+bool pred(const T x){
+  return (x % 2) == 0;
+}
+
+#ifndef ENERGY
+#define ENERGY 0
+#endif
+#define PRINT 0
+
+#define ANSI_COLOR_RED     "\x1b[31m"
+#define ANSI_COLOR_GREEN   "\x1b[32m"
+#define ANSI_COLOR_RESET   "\x1b[0m"
+
+#define divceil(n, m) (((n)-1) / (m) + 1)
+#define roundup(n, m) ((n / m) * m + m)
+#endif
diff --git a/COUNT/support/params.h b/COUNT/support/params.h
new file mode 100644
index 0000000..bb86211
--- /dev/null
+++ b/COUNT/support/params.h
@@ -0,0 +1,56 @@
+#ifndef _PARAMS_H_
+#define _PARAMS_H_
+
+#include "common.h"
+
+typedef struct 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");
+}
+
+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!");
+
+    return p;
+}
+#endif
diff --git a/COUNT/support/timer.h b/COUNT/support/timer.h
new file mode 100755
index 0000000..3ec6d87
--- /dev/null
+++ b/COUNT/support/timer.h
@@ -0,0 +1,66 @@
+/*

+ * 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");

+}