Adding vim modeline for .c files

oysteijo · Nov 21, 2023 · b5d5db7 · b5d5db7
1 parent ea7868c
commit b5d5db7
Show file tree

Hide file tree

Showing 29 changed files with 184 additions and 96 deletions.
diff --git a/examples/custom_activation/README.md b/examples/custom_activation/README.md
@@ -1,6 +1,6 @@
 # Implementing custom activation functions.
 
-Here is an example og how too implement and use a custom implemented activation function.
+Here is an example of how to implement and use a custom implemented activation function.
 This example will implement a simple sigmoid function with a pre scale of the input.
 
 $$ f(x) = \frac{1}{1+e^{-\alpha x}} $$
@@ -48,5 +48,3 @@ You should now be able to specify the custom activation in a neural network like
 When setting a loss function with `neuralnet_set_loss()` the derivative will be set correctly if it
 is implemented and present in the same shared object file as the forward activation.
 
-
-
diff --git a/src/activation.c b/src/activation.c
@@ -1,3 +1,7 @@
+/* activation.c - Øystein Schønning-Johansen 2013 - 2023 */
+/* 
+ vim: ts=4 sw=4 softtabstop=4 expandtab 
+*/
 #include "activation.h"
 #include "simd.h"
 

diff --git a/src/activation.h b/src/activation.h
@@ -1,4 +1,7 @@
-/* activation.h - Øystein Schønning-Johansen 2013 */
+/* activation.h - Øystein Schønning-Johansen 2013 - 2023 */
+/*
+  vim: ts=4 sw=4 softtabstop=4 expandtab 
+ */
 #ifndef __ACTIVATION_H__
 #define __ACTIVATION_H__
 

diff --git a/src/callback.h b/src/callback.h
@@ -1,4 +1,6 @@
-/* callback.h - oysteijo@gmail.com  
+/* callback.h - Øystein Schønning-Johansen 2020 - 2023 */
+/*
+  vim: ts=4 sw=4 softtabstop=4 expandtab 
  */
 
 /** \struct _callback_t
@@ -37,18 +39,18 @@
 
 typedef struct _callback_t callback_t;
 struct _callback_t {
-	void (*callback_run) ( callback_t *self, optimizer_t *opt, const float *result, bool has_valid );
-	void (*free) (callback_t *self);
+    void (*callback_run) ( callback_t *self, optimizer_t *opt, const float *result, bool has_valid );
+    void (*free) (callback_t *self);
 };
 
 static inline void callback_run( callback_t *self, optimizer_t *opt, const float *result, bool has_valid )
 {
-	self->callback_run( self, opt, result, has_valid );
+    self->callback_run( self, opt, result, has_valid );
 }
 
 static inline void callback_free( callback_t *self)
 {
-	self->free( self );
+    self->free( self );
 }
 
 #if defined(__GNUC__)
@@ -66,16 +68,16 @@ static inline void callback_free( callback_t *self)
 #define CALLBACK_DEFINE(name,...) \
 static void name ## _callback_run( callback_t *cb, optimizer_t *opt, const float *result, bool has_valid ); \
 DLLEXPORT name ## _t * name ## _new( void * UNUSED(config)) \
-{	\
-	name ## _t *newcb = malloc( sizeof( name ## _t ) ); \
-	if ( !newcb ) {\
-		fprintf( stderr ,"Can't allocate memory for '" #name "_t' callback type.\n"); \
-		return NULL; \
-	} \
-	newcb->cb.callback_run = name ## _callback_run; \
-	newcb->cb.free = (void(*)(callback_t*)) free; \
-	__VA_ARGS__ ; \
-	return newcb; \
+{   \
+    name ## _t *newcb = malloc( sizeof( name ## _t ) ); \
+    if ( !newcb ) {\
+        fprintf( stderr ,"Can't allocate memory for '" #name "_t' callback type.\n"); \
+        return NULL; \
+    } \
+    newcb->cb.callback_run = name ## _callback_run; \
+    newcb->cb.free = (void(*)(callback_t*)) free; \
+    __VA_ARGS__ ; \
+    return newcb; \
 }
 
 #define CALLBACK_DECLARE(name) \

diff --git a/src/earlystopping.c b/src/earlystopping.c
@@ -1,3 +1,7 @@
+/* earlystopping.c - Øystein Schønning-Johansen 2013 - 2023 */
+/* 
+ vim: ts=4 sw=4 softtabstop=4 expandtab 
+*/
 #include "earlystopping.h"
 #include <stdio.h>
 #include <float.h>

diff --git a/src/earlystopping.h b/src/earlystopping.h
@@ -1,3 +1,7 @@
+/* earlystopping.h - Øystein Schønning-Johansen 2020 - 2023 */
+/*
+  vim: ts=4 sw=4 softtabstop=4 expandtab 
+ */
 #ifndef __EARLYSTOPPING_H__
 #define __EARLYSTOPPING_H__
 #include "callback.h"

diff --git a/src/evaluate.c b/src/evaluate.c
@@ -1,3 +1,7 @@
+/* evaluate.c - Øystein Schønning-Johansen 2013 - 2023 */
+/* 
+ vim: ts=4 sw=4 softtabstop=4 expandtab 
+*/
 #include "evaluate.h"
 #include "simd.h"
 #include <string.h>

diff --git a/src/evaluate.h b/src/evaluate.h
@@ -1,3 +1,7 @@
+/* evaluate.h - Øystein Schønning-Johansen 2019 - 2023 */
+/*
+  vim: ts=4 sw=4 softtabstop=4 expandtab 
+ */
 #ifndef __EVALUATE_H__
 #define __EVALUATE_H__
 #include "neuralnet.h"

diff --git a/src/logger.c b/src/logger.c
@@ -1,3 +1,7 @@
+/* logger.c - Øystein Schønning-Johansen 2013 - 2023 */
+/* 
+ vim: ts=4 sw=4 softtabstop=4 expandtab 
+*/
 #include "logger.h"
 #include "metrics.h"
 

diff --git a/src/logger.h b/src/logger.h
@@ -1,3 +1,7 @@
+/* logger.h - Øystein Schønning-Johansen 2020 - 2023 */
+/*
+  vim: ts=4 sw=4 softtabstop=4 expandtab 
+ */
 #ifndef __LOGGER_H__
 #define __LOGGER_H__
 #include "callback.h"

diff --git a/src/loss.c b/src/loss.c
@@ -1,3 +1,7 @@
+/* loss.c - Øystein Schønning-Johansen 2013 - 2023 */
+/* 
+ vim: ts=4 sw=4 softtabstop=4 expandtab 
+*/
 #include "loss.h"
 
 #include <stdint.h> 

diff --git a/src/loss.h b/src/loss.h
@@ -1,4 +1,7 @@
-/* loss.h - Øystein Schønning-Johansen 2013-2019 */
+/* loss.h - Øystein Schønning-Johansen 2013 - 2023 */
+/*
+  vim: ts=4 sw=4 softtabstop=4 expandtab 
+ */
 #ifndef __LOSS_H__
 #define __LOSS_H__
 

diff --git a/src/matrix_operations.c b/src/matrix_operations.c
@@ -1,3 +1,7 @@
+/* matrix_operations.c - Øystein Schønning-Johansen 2013 - 2023 */
+/* 
+ vim: ts=4 sw=4 softtabstop=4 expandtab 
+*/
 #include "matrix_operations.h"
 #include "simd.h"
 #include <assert.h>
@@ -14,11 +18,11 @@
 #ifdef __AVX__  
 static inline float horizontalsum_avx( __m256 x )
 {
-	float sumAVX = 0.0f;
-	__m256 hsum = _mm256_hadd_ps(x, x);
-	hsum = _mm256_add_ps(hsum, _mm256_permute2f128_ps(hsum, hsum, 0x1));
-	_mm_store_ss(&sumAVX, _mm_hadd_ps( _mm256_castps256_ps128(hsum), _mm256_castps256_ps128(hsum) ) );
-	return sumAVX;
+    float sumAVX = 0.0f;
+    __m256 hsum = _mm256_hadd_ps(x, x);
+    hsum = _mm256_add_ps(hsum, _mm256_permute2f128_ps(hsum, hsum, 0x1));
+    _mm_store_ss(&sumAVX, _mm_hadd_ps( _mm256_castps256_ps128(hsum), _mm256_castps256_ps128(hsum) ) );
+    return sumAVX;
 }
 #endif
 
@@ -42,24 +46,26 @@ void matrix_vector_multiply( int n_rows, int n_cols, const float *matrix, const
         int j = 0;
 
 #ifdef __AVX512F__
-		__m512 sums = _mm512_setzero_ps ();
-		for (; j <= ((n_cols)-16); j += 16, m_ptr += 16, v_ptr += 16) /* Check if faster: unroll w prefetch */
-   #if defined(__FMA__)
-			sums = _mm512_fmadd_ps( _mm512_loadu_ps(v_ptr), _mm512_load_ps(m_ptr), sums);
-   #else
-			sums = _mm512_add_ps (sums, _mm512_mul_ps(_mm512_loadu_ps(v_ptr), _mm512_load_ps(m_ptr)));
-   #endif
-		y[i] = _mm512_reduce_add_ps( sums );
+        __m512 sums = _mm512_setzero_ps ();
+        for (; j <= ((n_cols)-16); j += 16, m_ptr += 16, v_ptr += 16){ /* Check if faster: unroll w prefetch */
+#if defined(__FMA__)
+            sums = _mm512_fmadd_ps( _mm512_loadu_ps(v_ptr), _mm512_load_ps(m_ptr), sums);
+#else
+            sums = _mm512_add_ps (sums, _mm512_mul_ps(_mm512_loadu_ps(v_ptr), _mm512_load_ps(m_ptr)));
+#endif
+        }
+        y[i] = _mm512_reduce_add_ps( sums );
 #endif
 #ifdef __AVX__
-		__m256 sum = _mm256_setzero_ps ();
-		for (; j <= ((n_cols)-8); j += 8, m_ptr += 8, v_ptr += 8) /* Check if faster: unroll w prefetch */
-   #if defined(__FMA__)
-			sum = _mm256_fmadd_ps( _mm256_load_ps(v_ptr), _mm256_load_ps(m_ptr), sum);
-   #else
-			sum = _mm256_add_ps (sum, _mm256_mul_ps(_mm256_load_ps(v_ptr), _mm256_load_ps(m_ptr)));
-   #endif
-		y[i] += horizontalsum_avx( sum );
+        __m256 sum = _mm256_setzero_ps ();
+        for (; j <= ((n_cols)-8); j += 8, m_ptr += 8, v_ptr += 8){ /* Check if faster: unroll w prefetch */
+#if defined(__FMA__)
+            sum = _mm256_fmadd_ps( _mm256_load_ps(v_ptr), _mm256_load_ps(m_ptr), sum);
+#else
+            sum = _mm256_add_ps (sum, _mm256_mul_ps(_mm256_load_ps(v_ptr), _mm256_load_ps(m_ptr)));
+#endif
+        }
+        y[i] += horizontalsum_avx( sum );
 #endif
         for(; j < n_cols; j++ )
             y[i] += *v_ptr++ * *m_ptr++;
@@ -116,68 +122,66 @@ void vector_matrix_multiply( int n, int m, const float *weight, const float *bia
     assert( is_aligned( y ));
     */
     const float *bias_ptr = bias;
-	float *y_ptr = y; 
+    float *y_ptr = y; 
     int i = 0;
 
 #ifdef __AVX512F__
-	for (; i <= ((m)-16) ; i += 16, bias_ptr +=16, y_ptr +=16 ){
-		_mm512_store_ps( y_ptr, _mm512_load_ps( bias_ptr ));
+    for (; i <= ((m)-16) ; i += 16, bias_ptr +=16, y_ptr +=16 ){
+        _mm512_store_ps( y_ptr, _mm512_load_ps( bias_ptr ));
     }
 #endif
 #ifdef __AVX__
-	for (; i <= ((m)-8) ; i += 8, bias_ptr +=8, y_ptr +=8 ){
-		_mm256_store_ps( y_ptr, _mm256_load_ps( bias_ptr ));
+    for (; i <= ((m)-8) ; i += 8, bias_ptr +=8, y_ptr +=8 ){
+        _mm256_store_ps( y_ptr, _mm256_load_ps( bias_ptr ));
     }
 #endif
     for( ; i < m; i++ ){
         *y_ptr++ = *bias_ptr++;
     }
 
-	for (int i = 0; i < n; i++) {
-		float const inp = input[i];
-		const float *weight_ptr = weight + ( i * m );  /* Argh! if m is not a multiple of ALIGN_SIZE, the pointer wil be unaligned! :-( */
-		if (inp) {
-			float  *y_ptr = y;  /* same goes for this */
-			if (inp == 1.0f){
+    for (int i = 0; i < n; i++) {
+        float const inp = input[i];
+        const float *weight_ptr = weight + ( i * m );  /* Argh! if m is not a multiple of ALIGN_SIZE, the pointer wil be unaligned! :-( */
+        if (inp) {
+            float  *y_ptr = y;  /* same goes for this */
+            if (inp == 1.0f){
                 int j = 0;
 #ifdef __AVX512F__
-				for (; j <= ((m)-16) ; j += 16, y_ptr += 16, weight_ptr += 16) 
-					_mm512_storeu_ps(y_ptr, _mm512_add_ps (_mm512_loadu_ps(y_ptr), _mm512_loadu_ps( weight_ptr )));
+                for (; j <= ((m)-16) ; j += 16, y_ptr += 16, weight_ptr += 16) 
+                    _mm512_storeu_ps(y_ptr, _mm512_add_ps (_mm512_loadu_ps(y_ptr), _mm512_loadu_ps( weight_ptr )));
 #endif /*  __AVX512F__ */
 #ifdef __AVX__
-				for (; j <= ((m)-8) ; j += 8, y_ptr += 8, weight_ptr += 8) 
-					_mm256_storeu_ps(y_ptr, _mm256_add_ps (_mm256_loadu_ps(y_ptr), _mm256_loadu_ps( weight_ptr )));
+                for (; j <= ((m)-8) ; j += 8, y_ptr += 8, weight_ptr += 8) 
+                    _mm256_storeu_ps(y_ptr, _mm256_add_ps (_mm256_loadu_ps(y_ptr), _mm256_loadu_ps( weight_ptr )));
 #endif /*  __AVX__ */
                 for( ; j < m; j++ )
                     *y_ptr++ += *weight_ptr++;
-            }
-
-			else {
+            } else {
                 int j = 0;
 #ifdef __AVX512F__
-				for (; j < ((m)-16) ; j += 16, y_ptr += 16, weight_ptr += 16){
-   #if defined(__FMA__)
-					_mm512_storeu_ps(y_ptr, _mm512_fmadd_ps( _mm512_loadu_ps(weight_ptr), _mm512_set1_ps(inp), _mm512_loadu_ps(y_ptr)));
-   #else
-					_mm512_storeu_ps(y_ptr, _mm512_add_ps(_mm512_loadu_ps(y_ptr), _mm512_mul_ps(_mm512_loadu_ps(weight_ptr), _mm512_set1_ps(inp))));
-   #endif  
+                for (; j < ((m)-16) ; j += 16, y_ptr += 16, weight_ptr += 16){
+#if defined(__FMA__)
+                    _mm512_storeu_ps(y_ptr, _mm512_fmadd_ps( _mm512_loadu_ps(weight_ptr), _mm512_set1_ps(inp), _mm512_loadu_ps(y_ptr)));
+#else
+                    _mm512_storeu_ps(y_ptr, _mm512_add_ps(_mm512_loadu_ps(y_ptr), _mm512_mul_ps(_mm512_loadu_ps(weight_ptr), _mm512_set1_ps(inp))));
+#endif  
                 }
 #endif
 #ifdef __AVX__
-				__m256 scalevec = _mm256_set1_ps(inp);
-				for (; j < ((m)-8) ; j += 8, y_ptr += 8, weight_ptr += 8){
-   #if defined(__FMA__)
-					_mm256_storeu_ps(y_ptr, _mm256_fmadd_ps( _mm256_loadu_ps(weight_ptr), scalevec, _mm256_loadu_ps(y_ptr)));
-   #else
-					_mm256_storeu_ps(y_ptr, _mm256_add_ps(_mm256_loadu_ps(y_ptr), _mm256_mul_ps(_mm256_loadu_ps(weight_ptr), scalevec)));
-   #endif  
+                __m256 scalevec = _mm256_set1_ps(inp);
+                for (; j < ((m)-8) ; j += 8, y_ptr += 8, weight_ptr += 8){
+#if defined(__FMA__)
+                    _mm256_storeu_ps(y_ptr, _mm256_fmadd_ps( _mm256_loadu_ps(weight_ptr), scalevec, _mm256_loadu_ps(y_ptr)));
+#else
+                    _mm256_storeu_ps(y_ptr, _mm256_add_ps(_mm256_loadu_ps(y_ptr), _mm256_mul_ps(_mm256_loadu_ps(weight_ptr), scalevec)));
+#endif  
                 }
 #endif
                 for(; j < m; j++ )
                     *y_ptr++ += inp * *weight_ptr++;
-			}
-		}
-	}
+            }
+        }
+    }
 #endif /* USE_CBLAS */
 }
 

diff --git a/src/matrix_operations.h b/src/matrix_operations.h
@@ -1,3 +1,7 @@
+/* matrix_operations.h - Øystein Schønning-Johansen 2019 - 2023 */
+/*
+  vim: ts=4 sw=4 softtabstop=4 expandtab 
+ */
 #ifndef __MATRIX_OPERATIONS_H__
 #define __MATRIX_OPERATIONS_H__
 

diff --git a/src/metrics.c b/src/metrics.c
@@ -1,3 +1,7 @@
+/* metrics.c - Øystein Schønning-Johansen 2013 - 2023 */
+/* 
+ vim: ts=4 sw=4 softtabstop=4 expandtab 
+*/
 #include "metrics.h"
 
 #include <stdint.h> 

diff --git a/src/metrics.h b/src/metrics.h
@@ -1,4 +1,7 @@
-/* metric.h - Øystein Schønning-Johansen 2013 */
+/* metrics.h - Øystein Schønning-Johansen 2019 - 2023 */
+/*
+  vim: ts=4 sw=4 softtabstop=4 expandtab 
+ */
 #ifndef __METRICS_H__
 #define __METRICS_H__
 

diff --git a/src/modelcheckpoint.c b/src/modelcheckpoint.c
@@ -1,3 +1,7 @@
+/* modelcheckpoint.c - Øystein Schønning-Johansen 2013 - 2023 */
+/* 
+ vim: ts=4 sw=4 softtabstop=4 expandtab 
+*/
 #include "modelcheckpoint.h"
 #include "neuralnet.h"
 #include "metrics.h"

diff --git a/src/modelcheckpoint.h b/src/modelcheckpoint.h
@@ -1,3 +1,7 @@
+/* modelcheckpoint.h - Øystein Schønning-Johansen 2020 - 2023 */
+/*
+  vim: ts=4 sw=4 softtabstop=4 expandtab 
+ */
 #ifndef __MODELCHECKPOINT_H__
 #define __MODELCHECKPOINT_H__
 #include "callback.h"

diff --git a/src/neuralnet.c b/src/neuralnet.c
@@ -1,5 +1,7 @@
-/* Copyright -- Øystein Schønning-Johansen 2007-2023 */
-
+/* neuralnet.c - Øystein Schønning-Johansen 2007 - 2023 */
+/* 
+ vim: ts=4 sw=4 softtabstop=4 expandtab 
+*/
 #include "neuralnet.h"
 #include "simd.h"
 #include "activation.h"

diff --git a/src/neuralnet.h b/src/neuralnet.h
@@ -1,3 +1,7 @@
+/* neuralnet.h - Øystein Schønning-Johansen 2007 - 2023 */
+/* 
+ vim: ts=4 sw=4 softtabstop=4 expandtab 
+*/
 #ifndef __NN_NEURALNET_H__
 #define __NN_NEURALNET_H__