[Minor] Avoid some warnings

2 files changed, 206 insertions, 121 deletions

diff --git a/contrib/kann/kann.c b/contrib/kann/kann.c
index 70d1f02d6..658f98a44 100644
--- a/contrib/kann/kann.c
+++ b/contrib/kann/kann.c
@@ -1,3 +1,19 @@
+/*
+ * Copyright 2024 Vsevolod Stakhov
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
 #include "config.h"
 
 #include <math.h>
@@ -19,9 +35,9 @@ static void kad_ext_collate(int n, kad_node_t **a, float **_x, float **_g, float
 	int i, j, k, l, n_var;
 	float *x, *g, *c;
 	n_var = kad_size_var(n, a);
-	x = *_x = (float*)realloc(*_x, n_var * sizeof(float));
-	g = *_g = (float*)realloc(*_g, n_var * sizeof(float));
-	c = *_c = (float*)realloc(*_c, kad_size_const(n, a) * sizeof(float));
+	x = *_x = (float *) realloc(*_x, n_var * sizeof(float));
+	g = *_g = (float *) realloc(*_g, n_var * sizeof(float));
+	c = *_c = (float *) realloc(*_c, kad_size_const(n, a) * sizeof(float));
 	memset(g, 0, n_var * sizeof(float));
 	for (i = j = k = 0; i < n; ++i) {
 		kad_node_t *v = a[i];
@@ -32,7 +48,8 @@ static void kad_ext_collate(int n, kad_node_t **a, float **_x, float **_g, float
 			v->x = &x[j];
 			v->g = &g[j];
 			j += l;
-		} else if (kad_is_const(v)) {
+		}
+		else if (kad_is_const(v)) {
 			l = kad_len(v);
 			memcpy(&c[k], v->x, l * sizeof(float));
 			free(v->x);
@@ -51,7 +68,8 @@ static void kad_ext_sync(int n, kad_node_t **a, float *x, float *g, float *c)
 			v->x = &x[j];
 			v->g = &g[j];
 			j += kad_len(v);
-		} else if (kad_is_const(v)) {
+		}
+		else if (kad_is_const(v)) {
 			v->x = &c[k];
 			k += kad_len(v);
 		}
@@ -68,14 +86,14 @@ kann_t *kann_new(kad_node_t *cost, int n_rest, ...)
 	if (cost->n_d != 0) return 0;
 
 	va_start(ap, n_rest);
-	roots = (kad_node_t**)malloc((n_roots + 1) * sizeof(kad_node_t*));
+	roots = (kad_node_t **) malloc((n_roots + 1) * sizeof(kad_node_t *));
 	for (i = 0; i < n_rest; ++i)
-		roots[i] = va_arg(ap, kad_node_t*);
+		roots[i] = va_arg(ap, kad_node_t *);
 	roots[i++] = cost;
 	va_end(ap);
 
 	cost->ext_flag |= KANN_F_COST;
-	a = (kann_t*)calloc(1, sizeof(kann_t));
+	a = (kann_t *) calloc(1, sizeof(kann_t));
 	a->v = kad_compile_array(&a->n, n_roots, roots);
 
 	for (i = 0; i < a->n; ++i) {
@@ -84,7 +102,7 @@ kann_t *kann_new(kad_node_t *cost, int n_rest, ...)
 	}
 	if (has_recur && !has_pivot) { /* an RNN that doesn't have a pivot; then add a pivot on top of cost and recompile */
 		cost->ext_flag &= ~KANN_F_COST;
-		roots[n_roots-1] = cost = kad_avg(1, &cost), cost->ext_flag |= KANN_F_COST;
+		roots[n_roots - 1] = cost = kad_avg(1, &cost), cost->ext_flag |= KANN_F_COST;
 		free(a->v);
 		a->v = kad_compile_array(&a->n, n_roots, roots);
 	}
@@ -96,7 +114,7 @@ kann_t *kann_new(kad_node_t *cost, int n_rest, ...)
 kann_t *kann_clone(kann_t *a, int batch_size)
 {
 	kann_t *b;
-	b = (kann_t*)calloc(1, sizeof(kann_t));
+	b = (kann_t *) calloc(1, sizeof(kann_t));
 	b->n = a->n;
 	b->v = kad_clone(a->n, a->v, batch_size);
 	kad_ext_collate(b->n, b->v, &b->x, &b->g, &b->c);
@@ -106,7 +124,7 @@ kann_t *kann_clone(kann_t *a, int batch_size)
 kann_t *kann_unroll_array(kann_t *a, int *len)
 {
 	kann_t *b;
-	b = (kann_t*)calloc(1, sizeof(kann_t));
+	b = (kann_t *) calloc(1, sizeof(kann_t));
 	b->x = a->x, b->g = a->g, b->c = a->c; /* these arrays are shared */
 	b->v = kad_unroll(a->n, a->v, &b->n, len);
 	return b;
@@ -118,7 +136,7 @@ kann_t *kann_unroll(kann_t *a, ...)
 	va_list ap;
 	int i, n_pivots, *len;
 	n_pivots = kad_n_pivots(a->n, a->v);
-	len = (int*)calloc(n_pivots, sizeof(int));
+	len = (int *) calloc(n_pivots, sizeof(int));
 	va_start(ap, a);
 	for (i = 0; i < n_pivots; ++i) len[i] = va_arg(ap, int);
 	va_end(ap);
@@ -137,7 +155,9 @@ void kann_delete_unrolled(kann_t *a)
 void kann_delete(kann_t *a)
 {
 	if (a == 0) return;
-	free(a->x); free(a->g); free(a->c);
+	free(a->x);
+	free(a->g);
+	free(a->c);
 	kann_delete_unrolled(a);
 }
 
@@ -146,7 +166,7 @@ static void kann_switch_core(kann_t *a, int is_train)
 	int i;
 	for (i = 0; i < a->n; ++i)
 		if (a->v[i]->op == 12 && a->v[i]->n_child == 2)
-			*(int32_t*)a->v[i]->ptr = !!is_train;
+			*(int32_t *) a->v[i]->ptr = !!is_train;
 }
 
 #define chk_flg(flag, mask) ((mask) == 0 || ((flag) & (mask)))
@@ -158,7 +178,8 @@ int kann_find(const kann_t *a, uint32_t ext_flag, int32_t ext_label)
 	for (i = k = 0; i < a->n; ++i)
 		if (chk_flg(a->v[i]->ext_flag, ext_flag) && chk_lbl(a->v[i]->ext_label, ext_label))
 			++k, r = i;
-	return k == 1? r : k == 0? -1 : -2;
+	return k == 1 ? r : k == 0 ? -1
+							   : -2;
 }
 
 int kann_feed_bind(kann_t *a, uint32_t ext_flag, int32_t ext_label, float **x)
@@ -176,8 +197,10 @@ int kann_feed_dim(const kann_t *a, uint32_t ext_flag, int32_t ext_label)
 	int i, k, n = 0;
 	for (i = k = 0; i < a->n; ++i)
 		if (kad_is_feed(a->v[i]) && chk_flg(a->v[i]->ext_flag, ext_flag) && chk_lbl(a->v[i]->ext_label, ext_label))
-			++k, n = a->v[i]->n_d > 1? kad_len(a->v[i]) / a->v[i]->d[0] : a->v[i]->n_d == 1? a->v[i]->d[0] : 1;
-	return k == 1? n : k == 0? -1 : -2;
+			++k, n = a->v[i]->n_d > 1 ? kad_len(a->v[i]) / a->v[i]->d[0] : a->v[i]->n_d == 1 ? a->v[i]->d[0]
+																							 : 1;
+	return k == 1 ? n : k == 0 ? -1
+							   : -2;
 }
 
 static float kann_cost_core(kann_t *a, int cost_label, int cal_grad)
@@ -210,7 +233,8 @@ void kann_rnn_start(kann_t *a)
 		if (p->pre) { /* NB: BE CAREFUL of the interaction between kann_rnn_start() and kann_set_batch_size() */
 			kad_node_t *q = p->pre;
 			if (q->x) memcpy(p->x, q->x, kad_len(p) * sizeof(float));
-			else memset(p->x, 0, kad_len(p) * sizeof(float));
+			else
+				memset(p->x, 0, kad_len(p) * sizeof(float));
 			if (q->n_child > 0) free(q->x);
 			q->x = p->x;
 		}
@@ -223,7 +247,7 @@ void kann_rnn_end(kann_t *a)
 	kad_ext_sync(a->n, a->v, a->x, a->g, a->c);
 	for (i = 0; i < a->n; ++i)
 		if (a->v[i]->pre && a->v[i]->pre->n_child > 0)
-			a->v[i]->pre->x = (float*)calloc(kad_len(a->v[i]->pre), sizeof(float));
+			a->v[i]->pre->x = (float *) calloc(kad_len(a->v[i]->pre), sizeof(float));
 }
 
 static int kann_class_error_core(const kann_t *ann, int *base)
@@ -238,10 +262,10 @@ static int kann_class_error_core(const kann_t *ann, int *base)
 				float t_sum = 0.0f, t_min = 1.0f, t_max = 0.0f, x_max = 0.0f, x_min = 1.0f;
 				int x_max_k = -1, t_max_k = -1;
 				for (k = 0; k < n; ++k) {
-					float xk = x->x[off+k], tk = t->x[off+k];
+					float xk = x->x[off + k], tk = t->x[off + k];
 					t_sum += tk;
-					t_min = t_min < tk? t_min : tk;
-					x_min = x_min < xk? x_min : xk;
+					t_min = t_min < tk ? t_min : tk;
+					x_min = x_min < xk ? x_min : xk;
 					if (t_max < tk) t_max = tk, t_max_k = k;
 					if (x_max < xk) x_max = xk, x_max_k = k;
 				}
@@ -283,7 +307,7 @@ typedef struct mtaux_t { /* cross-worker data */
 
 static void *mt_worker(void *data) /* pthread worker */
 {
-	mtaux1_t *mt1 = (mtaux1_t*)data;
+	mtaux1_t *mt1 = (mtaux1_t *) data;
 	mtaux_t *mt = mt1->g;
 	for (;;) {
 		int action;
@@ -297,7 +321,8 @@ static void *mt_worker(void *data) /* pthread worker */
 		if (action == -1) break;
 
 		if (mt->eval_out) kann_eval(mt1->a, KANN_F_OUT, 0);
-		else mt1->cost = kann_cost_core(mt1->a, mt->cost_label, mt->cal_grad);
+		else
+			mt1->cost = kann_cost_core(mt1->a, mt->cost_label, mt->cal_grad);
 	}
 	pthread_exit(0);
 }
@@ -324,18 +349,18 @@ void kann_mt(kann_t *ann, int n_threads, int max_batch_size)
 	int i, k;
 
 	if (n_threads <= 1) {
-		if (ann->mt) mt_destroy((mtaux_t*)ann->mt);
+		if (ann->mt) mt_destroy((mtaux_t *) ann->mt);
 		ann->mt = 0;
 		return;
 	}
 	if (n_threads > max_batch_size) n_threads = max_batch_size;
 	if (n_threads <= 1) return;
 
-	mt = (mtaux_t*)calloc(1, sizeof(mtaux_t));
+	mt = (mtaux_t *) calloc(1, sizeof(mtaux_t));
 	mt->n_threads = n_threads, mt->max_batch_size = max_batch_size;
 	pthread_mutex_init(&mt->mtx, 0);
 	pthread_cond_init(&mt->cv, 0);
-	mt->mt = (mtaux1_t*)calloc(n_threads, sizeof(mtaux1_t));
+	mt->mt = (mtaux1_t *) calloc(n_threads, sizeof(mtaux1_t));
 	for (i = k = 0; i < n_threads; ++i) {
 		int size = (max_batch_size - k) / (n_threads - i);
 		mt->mt[i].a = kann_clone(ann, size);
@@ -350,11 +375,11 @@ void kann_mt(kann_t *ann, int n_threads, int max_batch_size)
 
 static void mt_kickoff(kann_t *a, int cost_label, int cal_grad, int eval_out)
 {
-	mtaux_t *mt = (mtaux_t*)a->mt;
+	mtaux_t *mt = (mtaux_t *) a->mt;
 	int i, j, k, B, n_var;
 
 	B = kad_sync_dim(a->n, a->v, -1); /* get the current batch size */
-	assert(B <= mt->max_batch_size); /* TODO: can be relaxed */
+	assert(B <= mt->max_batch_size);  /* TODO: can be relaxed */
 	n_var = kann_size_var(a);
 
 	pthread_mutex_lock(&mt->mtx);
@@ -376,7 +401,7 @@ static void mt_kickoff(kann_t *a, int cost_label, int cal_grad, int eval_out)
 
 float kann_cost(kann_t *a, int cost_label, int cal_grad)
 {
-	mtaux_t *mt = (mtaux_t*)a->mt;
+	mtaux_t *mt = (mtaux_t *) a->mt;
 	int i, j, B, k, n_var;
 	float cost;
 
@@ -392,7 +417,7 @@ float kann_cost(kann_t *a, int cost_label, int cal_grad)
 	for (i = k = 0, cost = 0.0f; i < mt->n_threads; ++i) {
 		int size = (B - k) / (mt->n_threads - i);
 		cost += mt->mt[i].cost * size / B;
-		kad_saxpy(n_var, (float)size / B, mt->mt[i].a->g, a->g);
+		kad_saxpy(n_var, (float) size / B, mt->mt[i].a->g, a->g);
 		k += size;
 	}
 	for (j = 0; j < a->n; ++j) { /* copy values back at recurrent nodes (needed by textgen; TODO: temporary solution) */
@@ -410,14 +435,14 @@ float kann_cost(kann_t *a, int cost_label, int cal_grad)
 
 int kann_eval_out(kann_t *a)
 {
-	mtaux_t *mt = (mtaux_t*)a->mt;
+	mtaux_t *mt = (mtaux_t *) a->mt;
 	int j, B, n_eval;
 	if (mt == 0) return kann_eval(a, KANN_F_OUT, 0);
 	B = kad_sync_dim(a->n, a->v, -1); /* get the current batch size */
 	mt_kickoff(a, 0, 0, 1);
 	n_eval = kann_eval(mt->mt[0].a, KANN_F_OUT, 0);
 	while (mt->n_idle < mt->n_threads - 1); /* busy waiting until all threads in sync */
-	for (j = 0; j < a->n; ++j) { /* copy output values back */
+	for (j = 0; j < a->n; ++j) {            /* copy output values back */
 		kad_node_t *p = a->v[j];
 		if (p->ext_flag & KANN_F_OUT) {
 			int i, t, k, d0 = p->d[0] / B, d1 = 1; /* for RNN, p->d[0] may equal unroll_len * batch_size */
@@ -438,7 +463,7 @@ int kann_eval_out(kann_t *a)
 
 int kann_class_error(const kann_t *ann, int *base)
 {
-	mtaux_t *mt = (mtaux_t*)ann->mt;
+	mtaux_t *mt = (mtaux_t *) ann->mt;
 	int i, n_err = 0, b = 0;
 	if (mt == 0) return kann_class_error_core(ann, base);
 	for (i = 0; i < mt->n_threads; ++i) {
@@ -450,7 +475,7 @@ int kann_class_error(const kann_t *ann, int *base)
 
 void kann_switch(kann_t *ann, int is_train)
 {
-	mtaux_t *mt = (mtaux_t*)ann->mt;
+	mtaux_t *mt = (mtaux_t *) ann->mt;
 	int i;
 	if (mt == 0) {
 		kann_switch_core(ann, is_train);
@@ -460,11 +485,25 @@ void kann_switch(kann_t *ann, int is_train)
 		kann_switch_core(mt->mt[i].a, is_train);
 }
 #else
-void kann_mt(kann_t *ann, int n_threads, int max_batch_size) {}
-float kann_cost(kann_t *a, int cost_label, int cal_grad) { return kann_cost_core(a, cost_label, cal_grad); }
-int kann_eval_out(kann_t *a) { return kann_eval(a, KANN_F_OUT, 0); }
-int kann_class_error(const kann_t *a, int *base) { return kann_class_error_core(a, base); }
-void kann_switch(kann_t *ann, int is_train) { return kann_switch_core(ann, is_train); }
+void kann_mt(kann_t *ann, int n_threads, int max_batch_size)
+{
+}
+float kann_cost(kann_t *a, int cost_label, int cal_grad)
+{
+	return kann_cost_core(a, cost_label, cal_grad);
+}
+int kann_eval_out(kann_t *a)
+{
+	return kann_eval(a, KANN_F_OUT, 0);
+}
+int kann_class_error(const kann_t *a, int *base)
+{
+	return kann_class_error_core(a, base);
+}
+void kann_switch(kann_t *ann, int is_train)
+{
+	return kann_switch_core(ann, is_train);
+}
 #endif
 
 /***********************
@@ -485,7 +524,7 @@ void kann_save_fp(FILE *fp, kann_t *ann)
 void kann_save(const char *fn, kann_t *ann)
 {
 	FILE *fp;
-	fp = fn && strcmp(fn, "-")? fopen(fn, "wb") : stdout;
+	fp = fn && strcmp(fn, "-") ? fopen(fn, "wb") : stdout;
 	kann_save_fp(fp, ann);
 	fclose(fp);
 }
@@ -500,13 +539,13 @@ kann_t *kann_load_fp(FILE *fp)
 	if (strncmp(magic, KANN_MAGIC, 4) != 0) {
 		return 0;
 	}
-	ann = (kann_t*)calloc(1, sizeof(kann_t));
+	ann = (kann_t *) calloc(1, sizeof(kann_t));
 	ann->v = kad_load(fp, &ann->n);
 	n_var = kad_size_var(ann->n, ann->v);
 	n_const = kad_size_const(ann->n, ann->v);
-	ann->x = (float*)malloc(n_var * sizeof(float));
-	ann->g = (float*)calloc(n_var, sizeof(float));
-	ann->c = (float*)malloc(n_const * sizeof(float));
+	ann->x = (float *) malloc(n_var * sizeof(float));
+	ann->g = (float *) calloc(n_var, sizeof(float));
+	ann->c = (float *) malloc(n_const * sizeof(float));
 	(void) !fread(ann->x, sizeof(float), n_var, fp);
 	(void) !fread(ann->c, sizeof(float), n_const, fp);
 	kad_ext_sync(ann->n, ann->v, ann->x, ann->g, ann->c);
@@ -517,7 +556,7 @@ kann_t *kann_load(const char *fn)
 {
 	FILE *fp;
 	kann_t *ann;
-	fp = fn && strcmp(fn, "-")? fopen(fn, "rb") : stdin;
+	fp = fn && strcmp(fn, "-") ? fopen(fn, "rb") : stdin;
 	ann = kann_load_fp(fp);
 	fclose(fp);
 	return ann;
@@ -531,23 +570,24 @@ kann_t *kann_load(const char *fn)
 
 kad_node_t *kann_new_leaf_array(int *offset, kad_node_p *par, uint8_t flag, float x0_01, int n_d, int32_t d[KAD_MAX_DIM])
 {
-	int i, len, off = offset && par? *offset : -1;
+	int i, len, off = offset && par ? *offset : -1;
 	kad_node_t *p;
 
 	if (off >= 0 && par[off]) return par[(*offset)++];
-	p = (kad_node_t*)calloc(1, sizeof(kad_node_t));
+	p = (kad_node_t *) calloc(1, sizeof(kad_node_t));
 	p->n_d = n_d, p->flag = flag;
 	memcpy(p->d, d, n_d * sizeof(int32_t));
 	len = kad_len(p);
-	p->x = (float*)calloc(len, sizeof(float));
+	p->x = (float *) calloc(len, sizeof(float));
 	if (p->n_d <= 1) {
 		for (i = 0; i < len; ++i)
 			p->x[i] = x0_01;
-	} else {
+	}
+	else {
 		double sdev_inv;
-		sdev_inv = 1.0 / sqrt((double)len / p->d[0]);
+		sdev_inv = 1.0 / sqrt((double) len / p->d[0]);
 		for (i = 0; i < len; ++i)
-			p->x[i] = (float)(kad_drand_normal(0) * sdev_inv);
+			p->x[i] = (float) (kad_drand_normal(0) * sdev_inv);
 	}
 	if (off >= 0) par[off] = p, ++(*offset);
 	return p;
@@ -557,7 +597,9 @@ kad_node_t *kann_new_leaf2(int *offset, kad_node_p *par, uint8_t flag, float x0_
 {
 	int32_t i, d[KAD_MAX_DIM];
 	va_list ap;
-	va_start(ap, n_d); for (i = 0; i < n_d; ++i) d[i] = va_arg(ap, int); va_end(ap);
+	va_start(ap, n_d);
+	for (i = 0; i < n_d; ++i) d[i] = va_arg(ap, int);
+	va_end(ap);
 	return kann_new_leaf_array(offset, par, flag, x0_01, n_d, d);
 }
 
@@ -565,7 +607,7 @@ kad_node_t *kann_layer_dense2(int *offset, kad_node_p *par, kad_node_t *in, int
 {
 	int n0;
 	kad_node_t *w, *b;
-	n0 = in->n_d >= 2? kad_len(in) / in->d[0] : kad_len(in);
+	n0 = in->n_d >= 2 ? kad_len(in) / in->d[0] : kad_len(in);
 	w = kann_new_leaf2(offset, par, KAD_VAR, 0.0f, 2, n1, n0);
 	b = kann_new_leaf2(offset, par, KAD_VAR, 0.0f, 1, n1);
 	return kad_add(kad_cmul(in, w), b);
@@ -583,27 +625,27 @@ kad_node_t *kann_layer_layernorm2(int *offset, kad_node_t **par, kad_node_t *in)
 {
 	int n0;
 	kad_node_t *alpha, *beta;
-	n0 = in->n_d >= 2? kad_len(in) / in->d[0] : kad_len(in);
+	n0 = in->n_d >= 2 ? kad_len(in) / in->d[0] : kad_len(in);
 	alpha = kann_new_leaf2(offset, par, KAD_VAR, 1.0f, 1, n0);
-	beta  = kann_new_leaf2(offset, par, KAD_VAR, 0.0f, 1, n0);
+	beta = kann_new_leaf2(offset, par, KAD_VAR, 0.0f, 1, n0);
 	return kad_add(kad_mul(kad_stdnorm(in), alpha), beta);
 }
 
 static inline kad_node_t *cmul_norm2(int *offset, kad_node_t **par, kad_node_t *x, kad_node_t *w, int use_norm)
 {
-	return use_norm? kann_layer_layernorm2(offset, par, kad_cmul(x, w)) : kad_cmul(x, w);
+	return use_norm ? kann_layer_layernorm2(offset, par, kad_cmul(x, w)) : kad_cmul(x, w);
 }
 
 kad_node_t *kann_layer_rnn2(int *offset, kad_node_t **par, kad_node_t *in, kad_node_t *h0, int rnn_flag)
 {
-	int n0, n1 = h0->d[h0->n_d-1], use_norm = !!(rnn_flag & KANN_RNN_NORM);
+	int n0, n1 = h0->d[h0->n_d - 1], use_norm = !!(rnn_flag & KANN_RNN_NORM);
 	kad_node_t *t, *w, *u, *b, *out;
 
 	u = kann_new_leaf2(offset, par, KAD_VAR, 0.0f, 2, n1, n1);
 	b = kann_new_leaf2(offset, par, KAD_VAR, 0.0f, 1, n1);
 	t = cmul_norm2(offset, par, h0, u, use_norm);
 	if (in) {
-		n0 = in->n_d >= 2? kad_len(in) / in->d[0] : kad_len(in);
+		n0 = in->n_d >= 2 ? kad_len(in) / in->d[0] : kad_len(in);
 		w = kann_new_leaf2(offset, par, KAD_VAR, 0.0f, 2, n1, n0);
 		t = kad_add(cmul_norm2(offset, par, in, w, use_norm), t);
 	}
@@ -614,10 +656,10 @@ kad_node_t *kann_layer_rnn2(int *offset, kad_node_t **par, kad_node_t *in, kad_n
 
 kad_node_t *kann_layer_gru2(int *offset, kad_node_t **par, kad_node_t *in, kad_node_t *h0, int rnn_flag)
 {
-	int n0 = 0, n1 = h0->d[h0->n_d-1], use_norm = !!(rnn_flag & KANN_RNN_NORM);
+	int n0 = 0, n1 = h0->d[h0->n_d - 1], use_norm = !!(rnn_flag & KANN_RNN_NORM);
 	kad_node_t *t, *r, *z, *w, *u, *b, *s, *out;
 
-	if (in) n0 = in->n_d >= 2? kad_len(in) / in->d[0] : kad_len(in);
+	if (in) n0 = in->n_d >= 2 ? kad_len(in) / in->d[0] : kad_len(in);
 	/* z = sigm(x_t * W_z + h_{t-1} * U_z + b_z) */
 	u = kann_new_leaf2(offset, par, KAD_VAR, 0.0f, 2, n1, n1);
 	b = kann_new_leaf2(offset, par, KAD_VAR, 0.0f, 1, n1);
@@ -657,16 +699,36 @@ kad_node_t *kann_new_leaf(uint8_t flag, float x0_01, int n_d, ...)
 {
 	int32_t i, d[KAD_MAX_DIM];
 	va_list ap;
-	va_start(ap, n_d); for (i = 0; i < n_d; ++i) d[i] = va_arg(ap, int); va_end(ap);
+	va_start(ap, n_d);
+	for (i = 0; i < n_d; ++i) d[i] = va_arg(ap, int);
+	va_end(ap);
 	return kann_new_leaf_array(0, 0, flag, x0_01, n_d, d);
 }
 
-kad_node_t *kann_new_scalar(uint8_t flag, float x) { return kann_new_leaf(flag, x, 0); }
-kad_node_t *kann_new_weight(int n_row, int n_col) { return kann_new_leaf(KAD_VAR, 0.0f, 2, n_row, n_col); }
-kad_node_t *kann_new_vec(int n, float x) { return kann_new_leaf(KAD_VAR, x, 1, n); }
-kad_node_t *kann_new_bias(int n) { return kann_new_vec(n, 0.0f); }
-kad_node_t *kann_new_weight_conv2d(int n_out, int n_in, int k_row, int k_col) { return kann_new_leaf(KAD_VAR, 0.0f, 4, n_out, n_in, k_row, k_col); }
-kad_node_t *kann_new_weight_conv1d(int n_out, int n_in, int kernel_len) { return kann_new_leaf(KAD_VAR, 0.0f, 3, n_out, n_in, kernel_len); }
+kad_node_t *kann_new_scalar(uint8_t flag, float x)
+{
+	return kann_new_leaf(flag, x, 0);
+}
+kad_node_t *kann_new_weight(int n_row, int n_col)
+{
+	return kann_new_leaf(KAD_VAR, 0.0f, 2, n_row, n_col);
+}
+kad_node_t *kann_new_vec(int n, float x)
+{
+	return kann_new_leaf(KAD_VAR, x, 1, n);
+}
+kad_node_t *kann_new_bias(int n)
+{
+	return kann_new_vec(n, 0.0f);
+}
+kad_node_t *kann_new_weight_conv2d(int n_out, int n_in, int k_row, int k_col)
+{
+	return kann_new_leaf(KAD_VAR, 0.0f, 4, n_out, n_in, k_row, k_col);
+}
+kad_node_t *kann_new_weight_conv1d(int n_out, int n_in, int kernel_len)
+{
+	return kann_new_leaf(KAD_VAR, 0.0f, 3, n_out, n_in, kernel_len);
+}
 
 kad_node_t *kann_layer_input(int n1)
 {
@@ -676,23 +738,32 @@ kad_node_t *kann_layer_input(int n1)
 	return t;
 }
 
-kad_node_t *kann_layer_dense(kad_node_t *in, int n1) { return kann_layer_dense2(0, 0, in, n1); }
-kad_node_t *kann_layer_dropout(kad_node_t *t, float r) { return kann_layer_dropout2(0, 0, t, r); }
-kad_node_t *kann_layer_layernorm(kad_node_t *in) { return kann_layer_layernorm2(0, 0, in); }
+kad_node_t *kann_layer_dense(kad_node_t *in, int n1)
+{
+	return kann_layer_dense2(0, 0, in, n1);
+}
+kad_node_t *kann_layer_dropout(kad_node_t *t, float r)
+{
+	return kann_layer_dropout2(0, 0, t, r);
+}
+kad_node_t *kann_layer_layernorm(kad_node_t *in)
+{
+	return kann_layer_layernorm2(0, 0, in);
+}
 
 kad_node_t *kann_layer_rnn(kad_node_t *in, int n1, int rnn_flag)
 {
 	kad_node_t *h0;
-	h0 = (rnn_flag & KANN_RNN_VAR_H0)? kad_var(0, 0, 2, 1, n1) : kad_const(0, 2, 1, n1);
-	h0->x = (float*)calloc(n1, sizeof(float));
+	h0 = (rnn_flag & KANN_RNN_VAR_H0) ? kad_var(0, 0, 2, 1, n1) : kad_const(0, 2, 1, n1);
+	h0->x = (float *) calloc(n1, sizeof(float));
 	return kann_layer_rnn2(0, 0, in, h0, rnn_flag);
 }
 
 kad_node_t *kann_layer_gru(kad_node_t *in, int n1, int rnn_flag)
 {
 	kad_node_t *h0;
-	h0 = (rnn_flag & KANN_RNN_VAR_H0)? kad_var(0, 0, 2, 1, n1) : kad_const(0, 2, 1, n1);
-	h0->x = (float*)calloc(n1, sizeof(float));
+	h0 = (rnn_flag & KANN_RNN_VAR_H0) ? kad_var(0, 0, 2, 1, n1) : kad_const(0, 2, 1, n1);
+	h0->x = (float *) calloc(n1, sizeof(float));
 	return kann_layer_gru2(0, 0, in, h0, rnn_flag);
 }
 
@@ -705,13 +776,13 @@ kad_node_t *kann_layer_lstm(kad_node_t *in, int n1, int rnn_flag)
 {
 	int n0;
 	kad_node_t *i, *f, *o, *g, *w, *u, *b, *h0, *c0, *c, *out;
-	kad_node_t *(*cmul)(kad_node_t*, kad_node_t*) = (rnn_flag & KANN_RNN_NORM)? kann_cmul_norm : kad_cmul;
+	kad_node_t *(*cmul)(kad_node_t *, kad_node_t *) = (rnn_flag & KANN_RNN_NORM) ? kann_cmul_norm : kad_cmul;
 
-	n0 = in->n_d >= 2? kad_len(in) / in->d[0] : kad_len(in);
-	h0 = (rnn_flag & KANN_RNN_VAR_H0)? kad_var(0, 0, 2, 1, n1) : kad_const(0, 2, 1, n1);
-	h0->x = (float*)calloc(n1, sizeof(float));
-	c0 = (rnn_flag & KANN_RNN_VAR_H0)? kad_var(0, 0, 2, 1, n1) : kad_const(0, 2, 1, n1);
-	c0->x = (float*)calloc(n1, sizeof(float));
+	n0 = in->n_d >= 2 ? kad_len(in) / in->d[0] : kad_len(in);
+	h0 = (rnn_flag & KANN_RNN_VAR_H0) ? kad_var(0, 0, 2, 1, n1) : kad_const(0, 2, 1, n1);
+	h0->x = (float *) calloc(n1, sizeof(float));
+	c0 = (rnn_flag & KANN_RNN_VAR_H0) ? kad_var(0, 0, 2, 1, n1) : kad_const(0, 2, 1, n1);
+	c0->x = (float *) calloc(n1, sizeof(float));
 
 	/* i = sigm(x_t * W_i + h_{t-1} * U_i + b_i) */
 	w = kann_new_weight(n1, n0);
@@ -766,18 +837,21 @@ kad_node_t *kann_layer_cost(kad_node_t *t, int n_out, int cost_type)
 
 	if (cost_type == KANN_C_MSE) {
 		cost = kad_mse(t, truth);
-	} else if (cost_type == KANN_C_CEB) {
+	}
+	else if (cost_type == KANN_C_CEB) {
 		t = kad_sigm(t);
 		cost = kad_ce_bin(t, truth);
-	} else if (cost_type == KANN_C_CEB_NEG) {
+	}
+	else if (cost_type == KANN_C_CEB_NEG) {
 		t = kad_tanh(t);
 		cost = kad_ce_bin_neg(t, truth);
-	} else if (cost_type == KANN_C_CEM) {
+	}
+	else if (cost_type == KANN_C_CEM) {
 		t = kad_softmax(t);
 		cost = kad_ce_multi(t, truth);
 	}
 	else {
-		assert (0);
+		assert(0);
 	}
 
 	t->ext_flag |= KANN_F_OUT;
@@ -791,8 +865,8 @@ void kann_shuffle(int n, int *s)
 	int i, j, t;
 	for (i = 0; i < n; ++i) s[i] = i;
 	for (i = n; i > 0; --i) {
-		j = (int)(i * kad_drand(0));
-		t = s[j], s[j] = s[i-1], s[i-1] = t;
+		j = (int) (i * kad_drand(0));
+		t = s[j], s[j] = s[i - 1], s[i - 1] = t;
 	}
 }
 
@@ -805,7 +879,7 @@ void kann_shuffle(int n, int *s)
 
 void kann_RMSprop(int n, float h0, const float *h, float decay, const float *g, float *t, float *r)
 {
-	int i, n4 = n>>2<<2;
+	int i, n4 = n >> 2 << 2;
 	__m128 vh, vg, vr, vt, vd, vd1, tmp, vtiny;
 	vh = _mm_set1_ps(h0);
 	vd = _mm_set1_ps(decay);
@@ -823,7 +897,7 @@ void kann_RMSprop(int n, float h0, const float *h, float decay, const float *g,
 	}
 	for (; i < n; ++i) {
 		r[i] = (1. - decay) * g[i] * g[i] + decay * r[i];
-		t[i] -= (h? h[i] : h0) / sqrtf(1e-6f + r[i]) * g[i];
+		t[i] -= (h ? h[i] : h0) / sqrtf(1e-6f + r[i]) * g[i];
 	}
 }
 #else
@@ -831,7 +905,7 @@ void kann_RMSprop(int n, float h0, const float *h, float decay, const float *g,
 {
 	int i;
 	for (i = 0; i < n; ++i) {
-		float lr = h? h[i] : h0;
+		float lr = h ? h[i] : h0;
 		r[i] = (1.0f - decay) * g[i] * g[i] + decay * r[i];
 		t[i] -= lr / sqrtf(1e-6f + r[i]) * g[i];
 	}
@@ -847,8 +921,8 @@ float kann_grad_clip(float thres, int n, float *g)
 	s2 = sqrt(s2);
 	if (s2 > thres)
 		for (i = 0, s2 = 1.0 / s2; i < n; ++i)
-			g[i] *= (float)s2;
-	return (float)s2 / thres;
+			g[i] *= (float) s2;
+	return (float) s2 / thres;
 }
 
 /****************************************************************
@@ -868,21 +942,21 @@ int kann_train_fnn1(kann_t *ann, float lr, int mini_size, int max_epoch,
 	if (n_in < 0 || n_out < 0) return -1;
 	n_var = kann_size_var(ann);
 	n_const = kann_size_const(ann);
-	r = (float*)calloc(n_var, sizeof(float));
-	shuf = (int*)malloc(n * sizeof(int));
-	x = (float**)malloc(n * sizeof(float*));
-	y = (float**)malloc(n * sizeof(float*));
+	r = (float *) calloc(n_var, sizeof(float));
+	shuf = (int *) malloc(n * sizeof(int));
+	x = (float **) malloc(n * sizeof(float *));
+	y = (float **) malloc(n * sizeof(float *));
 	kann_shuffle(n, shuf);
 	for (j = 0; j < n; ++j)
 		x[j] = _x[shuf[j]], y[j] = _y[shuf[j]];
-	n_val = (int)(n * frac_val);
+	n_val = (int) (n * frac_val);
 	n_train = n - n_val;
-	min_x = (float*)malloc(n_var * sizeof(float));
-	min_c = (float*)malloc(n_const * sizeof(float));
+	min_x = (float *) malloc(n_var * sizeof(float));
+	min_c = (float *) malloc(n_const * sizeof(float));
 
-	x1 = (float*)malloc(n_in  * mini_size * sizeof(float));
-	y1 = (float*)malloc(n_out * mini_size * sizeof(float));
-	kann_feed_bind(ann, KANN_F_IN,    0, &x1);
+	x1 = (float *) malloc(n_in * mini_size * sizeof(float));
+	y1 = (float *) malloc(n_out * mini_size * sizeof(float));
+	kann_feed_bind(ann, KANN_F_IN, 0, &x1);
 	kann_feed_bind(ann, KANN_F_TRUTH, 0, &y1);
 
 	for (i = 0; i < max_epoch; ++i) {
@@ -891,10 +965,10 @@ int kann_train_fnn1(kann_t *ann, float lr, int mini_size, int max_epoch,
 		kann_shuffle(n_train, shuf);
 		kann_switch(ann, 1);
 		while (n_proc < n_train) {
-			int b, c, ms = n_train - n_proc < mini_size? n_train - n_proc : mini_size;
+			int b, c, ms = n_train - n_proc < mini_size ? n_train - n_proc : mini_size;
 			for (b = 0; b < ms; ++b) {
-				memcpy(&x1[b*n_in],  x[shuf[n_proc+b]], n_in  * sizeof(float));
-				memcpy(&y1[b*n_out], y[shuf[n_proc+b]], n_out * sizeof(float));
+				memcpy(&x1[b * n_in], x[shuf[n_proc + b]], n_in * sizeof(float));
+				memcpy(&y1[b * n_out], y[shuf[n_proc + b]], n_out * sizeof(float));
 			}
 			kann_set_batch_size(ann, ms);
 			train_cost += kann_cost(ann, 0, 1) * ms;
@@ -907,10 +981,10 @@ int kann_train_fnn1(kann_t *ann, float lr, int mini_size, int max_epoch,
 		kann_switch(ann, 0);
 		n_proc = 0;
 		while (n_proc < n_val) {
-			int b, c, ms = n_val - n_proc < mini_size? n_val - n_proc : mini_size;
+			int b, c, ms = n_val - n_proc < mini_size ? n_val - n_proc : mini_size;
 			for (b = 0; b < ms; ++b) {
-				memcpy(&x1[b*n_in],  x[n_train+n_proc+b], n_in  * sizeof(float));
-				memcpy(&y1[b*n_out], y[n_train+n_proc+b], n_out * sizeof(float));
+				memcpy(&x1[b * n_in], x[n_train + n_proc + b], n_in * sizeof(float));
+				memcpy(&y1[b * n_out], y[n_train + n_proc + b], n_out * sizeof(float));
 			}
 			kann_set_batch_size(ann, ms);
 			val_cost += kann_cost(ann, 0, 0) * ms;
@@ -919,6 +993,8 @@ int kann_train_fnn1(kann_t *ann, float lr, int mini_size, int max_epoch,
 			n_proc += ms;
 		}
 		if (n_val > 0) val_cost /= n_val;
+		(void) (n_train_err);
+		(void) (n_val_err);
 		if (cb) {
 			cb(i + 1, train_cost, val_cost, ud);
 #if 0
@@ -937,8 +1013,9 @@ int kann_train_fnn1(kann_t *ann, float lr, int mini_size, int max_epoch,
 				memcpy(min_x, ann->x, n_var * sizeof(float));
 				memcpy(min_c, ann->c, n_const * sizeof(float));
 				drop_streak = 0;
-				min_val_cost = (float)val_cost;
-			} else if (++drop_streak >= max_drop_streak)
+				min_val_cost = (float) val_cost;
+			}
+			else if (++drop_streak >= max_drop_streak)
 				break;
 		}
 	}
@@ -947,13 +1024,20 @@ int kann_train_fnn1(kann_t *ann, float lr, int mini_size, int max_epoch,
 		memcpy(ann->c, min_c, n_const * sizeof(float));
 	}
 
-	free(min_c); free(min_x); free(y1); free(x1); free(y); free(x); free(shuf); free(r);
+	free(min_c);
+	free(min_x);
+	free(y1);
+	free(x1);
+	free(y);
+	free(x);
+	free(shuf);
+	free(r);
 	return i;
 }
 
 float kann_cost_fnn1(kann_t *ann, int n, float **x, float **y)
 {
-	int n_in, n_out, n_proc = 0, mini_size = 64 < n? 64 : n;
+	int n_in, n_out, n_proc = 0, mini_size = 64 < n ? 64 : n;
 	float *x1, *y1;
 	double cost = 0.0;
 
@@ -961,23 +1045,24 @@ float kann_cost_fnn1(kann_t *ann, int n, float **x, float **y)
 	n_out = kann_dim_out(ann);
 	if (n <= 0 || n_in < 0 || n_out < 0) return 0.0;
 
-	x1 = (float*)malloc(n_in  * mini_size * sizeof(float));
-	y1 = (float*)malloc(n_out * mini_size * sizeof(float));
-	kann_feed_bind(ann, KANN_F_IN,    0, &x1);
+	x1 = (float *) malloc(n_in * mini_size * sizeof(float));
+	y1 = (float *) malloc(n_out * mini_size * sizeof(float));
+	kann_feed_bind(ann, KANN_F_IN, 0, &x1);
 	kann_feed_bind(ann, KANN_F_TRUTH, 0, &y1);
 	kann_switch(ann, 0);
 	while (n_proc < n) {
-		int b, ms = n - n_proc < mini_size? n - n_proc : mini_size;
+		int b, ms = n - n_proc < mini_size ? n - n_proc : mini_size;
 		for (b = 0; b < ms; ++b) {
-			memcpy(&x1[b*n_in],  x[n_proc+b], n_in  * sizeof(float));
-			memcpy(&y1[b*n_out], y[n_proc+b], n_out * sizeof(float));
+			memcpy(&x1[b * n_in], x[n_proc + b], n_in * sizeof(float));
+			memcpy(&y1[b * n_out], y[n_proc + b], n_out * sizeof(float));
 		}
 		kann_set_batch_size(ann, ms);
 		cost += kann_cost(ann, 0, 0) * ms;
 		n_proc += ms;
 	}
-	free(y1); free(x1);
-	return (float)(cost / n);
+	free(y1);
+	free(x1);
+	return (float) (cost / n);
 }
 
 const float *kann_apply1(kann_t *a, float *x)
diff --git a/contrib/zstd/CMakeLists.txt b/contrib/zstd/CMakeLists.txt
index 4601e53ce..9fba28fa4 100644
--- a/contrib/zstd/CMakeLists.txt
+++ b/contrib/zstd/CMakeLists.txt
@@ -24,4 +24,4 @@ SET(ZSTDSRC
 		zstd_opt.c)
 
 ADD_LIBRARY(rspamd-zstd STATIC ${ZSTDSRC})
-ADD_DEFINITIONS(-DZSTD_DISABLE_ASM)
\ No newline at end of file
+ADD_DEFINITIONS(-DZSTD_DISABLE_ASM -DZSTD_DISABLE_DEPRECATE_WARNINGS)
\ No newline at end of file