From b1c7331483cc85cd5e9ae236c2b53f522b68416f Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Sat, 4 Jan 2020 15:43:55 +0300
Subject: [PATCH] Added fast random functions

---
 src/dropout_layer_kernels.cu |  8 +++---
 src/utils.c                  | 51 +++++++++++++++++++++++++++++++++---
 src/utils.h                  |  3 +++
 3 files changed, 55 insertions(+), 7 deletions(-)

diff --git a/src/dropout_layer_kernels.cu b/src/dropout_layer_kernels.cu
index 79ed37d9..09493bb0 100644
--- a/src/dropout_layer_kernels.cu
+++ b/src/dropout_layer_kernels.cu
@@ -59,12 +59,14 @@ void forward_dropout_layer_gpu(dropout_layer l, network_state state)
             for (k = 0; k < l.c; k++) {
                 for (i = 0; i < max_blocks_per_channel; i++) {
                     float rnd = random_float();
-                    //printf(" rnd = %f \n", rnd);
+                    printf(" rnd = %f \n", rnd);
+                    //int rn = rand_int_fast(1, 7);
+                    //printf(" rnd = %d \n", rn);
                     if (rnd < prob_place_block) {
                         //count_ones += block_width  *block_height;
                         const int pre_index = k*l.w*l.h + b*l.w*l.h*l.c;
-                        const int x_block = rand_int(0, l.w - block_width - 1);
-                        const int y_block = rand_int(0, l.h - block_height - 1);
+                        const int x_block = rand_int_fast(0, l.w - block_width - 1);
+                        const int y_block = rand_int_fast(0, l.h - block_height - 1);
                         for (y = y_block; y < (y_block + block_height); y++) {
                             memset(&l.rand[x_block + y*l.w + pre_index], 1, block_width * sizeof(float));
                             //for (x = x_block; x < (x_block + block_width); x++) {
diff --git a/src/utils.c b/src/utils.c
index 4d8e439d..0e3a76fc 100644
--- a/src/utils.c
+++ b/src/utils.c
@@ -813,6 +813,40 @@ float **one_hot_encode(float *a, int n, int k)
     return t;
 }
 
+static unsigned int x = 123456789, y = 362436069, z = 521288629;
+
+// Marsaglia's xorshf96 generator: period 2^96-1
+unsigned int random_gen_fast(void)
+{
+    unsigned int t;
+    x ^= x << 16;
+    x ^= x >> 5;
+    x ^= x << 1;
+
+    t = x;
+    x = y;
+    y = z;
+    z = t ^ x ^ y;
+
+    return z;
+}
+
+float random_float_fast()
+{
+    return ((float)random_gen_fast() / (float)UINT_MAX);
+}
+
+int rand_int_fast(int min, int max)
+{
+    if (max < min) {
+        int s = min;
+        min = max;
+        max = s;
+    }
+    int r = (random_gen_fast() % (max - min + 1)) + min;
+    return r;
+}
+
 unsigned int random_gen()
 {
     unsigned int rnd = 0;
@@ -829,11 +863,20 @@ unsigned int random_gen()
 
 float random_float()
 {
+    unsigned int rnd = 0;
 #ifdef WIN32
-    return ((float)random_gen() / (float)UINT_MAX);
-#else
-    return ((float)random_gen() / (float)RAND_MAX);
-#endif
+    rand_s(&rnd);
+    return ((float)rnd / (float)UINT_MAX);
+#else   // WIN32
+
+    rnd = rand();
+#if (RAND_MAX < 65536)
+    rnd = rand()*(RAND_MAX + 1) + rnd;
+    return((float)rnd / (float)(RAND_MAX*RAND_MAX));
+#endif  //(RAND_MAX < 65536)
+    return ((float)rnd / (float)RAND_MAX);
+
+#endif  // WIN32
 }
 
 float rand_uniform_strong(float min, float max)
diff --git a/src/utils.h b/src/utils.h
index 421a313d..284e4556 100644
--- a/src/utils.h
+++ b/src/utils.h
@@ -73,6 +73,9 @@ char *find_char_arg(int argc, char **argv, char *arg, char *def);
 int sample_array(float *a, int n);
 int sample_array_custom(float *a, int n);
 void print_statistics(float *a, int n);
+unsigned int random_gen_fast(void);
+float random_float_fast();
+int rand_int_fast(int min, int max);
 unsigned int random_gen();
 float random_float();
 float rand_uniform_strong(float min, float max);