From bf8c2b0bb55a417415145618aaab086f4698aa92 Mon Sep 17 00:00:00 2001
From: Bienert <bienert.andreas@siemens.com>
Date: Thu, 20 Nov 2025 14:37:56 +0100
Subject: [PATCH] Test3

---
 matrix.c             | 55 +++++++++++++++++++++++++++++++++++++++-----
 neuralNetworkTests.c | 45 +++++++++++++++++++++++++++++++++++-
 2 files changed, 93 insertions(+), 7 deletions(-)

diff --git a/matrix.c b/matrix.c
index d3ea248..38bbbbc 100644
--- a/matrix.c
+++ b/matrix.c
@@ -60,15 +60,58 @@ Matrix add(const Matrix matrix1, const Matrix matrix2)
     result.rows = result.cols = 0;
     result.buffer = NULL;
 
-    if (matrix1.rows != matrix2.rows || matrix1.cols != matrix2.cols) return result;
+    /* einfache Validierung */
+    if (matrix1.rows == 0 || matrix1.cols == 0 || matrix1.buffer == NULL) return result;
+    if (matrix2.rows == 0 || matrix2.cols == 0 || matrix2.buffer == NULL) return result;
 
-    result = createMatrix(matrix1.rows, matrix1.cols);
-    if (!result.buffer) return result;
+    /* Fall 1: gleiche Dimensionen -> elementweise Addition */
+    if (matrix1.rows == matrix2.rows && matrix1.cols == matrix2.cols)
+    {
+        result = createMatrix(matrix1.rows, matrix1.cols);
+        if (!result.buffer) return result;
 
-    size_t count = (size_t)matrix1.rows * matrix1.cols;
-    for (size_t i = 0; i < count; ++i) {
-        result.buffer[i] = matrix1.buffer[i] + matrix2.buffer[i];
+        size_t count = (size_t)matrix1.rows * matrix1.cols;
+        for (size_t i = 0; i < count; ++i)
+            result.buffer[i] = matrix1.buffer[i] + matrix2.buffer[i];
+
+        return result;
     }
+
+    /* Fall 2: matrix2 ist Bias (rows x 1) -> broadcast über Spalten von matrix1 */
+    if (matrix1.rows == matrix2.rows && matrix2.cols == 1)
+    {
+        result = createMatrix(matrix1.rows, matrix1.cols);
+        if (!result.buffer) return result;
+
+        for (unsigned int c = 0; c < matrix1.cols; ++c)
+        {
+            for (unsigned int r = 0; r < matrix1.rows; ++r)
+            {
+                size_t idx = (size_t)r * matrix1.cols + c;
+                result.buffer[idx] = matrix1.buffer[idx] + matrix2.buffer[r];
+            }
+        }
+        return result;
+    }
+
+    /* Fall 3: matrix1 ist Bias (rows x 1) -> broadcast über Spalten von matrix2 */
+    if (matrix2.rows == matrix1.rows && matrix1.cols == 1)
+    {
+        result = createMatrix(matrix2.rows, matrix2.cols);
+        if (!result.buffer) return result;
+
+        for (unsigned int c = 0; c < matrix2.cols; ++c)
+        {
+            for (unsigned int r = 0; r < matrix2.rows; ++r)
+            {
+                size_t idx = (size_t)r * matrix2.cols + c;
+                result.buffer[idx] = matrix2.buffer[idx] + matrix1.buffer[r];
+            }
+        }
+        return result;
+    }
+
+    /* keine kompatible Form -> leere Matrix zurückgeben */
     return result;
 }
 
diff --git a/neuralNetworkTests.c b/neuralNetworkTests.c
index 21ab370..2fe5834 100644
--- a/neuralNetworkTests.c
+++ b/neuralNetworkTests.c
@@ -8,7 +8,50 @@
 
 static void prepareNeuralNetworkFile(const char *path, const NeuralNetwork nn)
 {
-    // TODO
+    if (!path) return;
+
+    FILE *file = fopen(path, "wb");
+    if (!file) return;
+
+    /* Datei-Header */
+    const char *fileTag = "__info2_neural_network_file_format__";
+    fwrite(fileTag, sizeof(char), strlen(fileTag), file);
+
+    /* Falls kein Layer vorhanden -> schreibe zwei Nullen und beenden */
+    if (nn.numberOfLayers == 0) {
+        int zero = 0;
+        fwrite(&zero, sizeof(int), 1, file);
+        fwrite(&zero, sizeof(int), 1, file);
+        fclose(file);
+        return;
+    }
+
+    /* erste Dimensionen: input = cols der ersten Weight-Matrix, output = rows der ersten Weight-Matrix */
+    int inputDim = (int)nn.layers[0].weights.cols;
+    int outputDim = (int)nn.layers[0].weights.rows;
+    fwrite(&inputDim, sizeof(int), 1, file);
+    fwrite(&outputDim, sizeof(int), 1, file);
+
+    /* für jede Schicht: weights (rows*cols floats), biases (rows*1 floats)
+       danach die nächste output-dimension (oder 0 zum Beenden) */
+    for (unsigned int i = 0; i < nn.numberOfLayers; ++i) {
+        Layer layer = nn.layers[i];
+
+        size_t wcount = (size_t)layer.weights.rows * layer.weights.cols;
+        if (wcount > 0 && layer.weights.buffer != NULL) {
+            fwrite(layer.weights.buffer, sizeof(MatrixType), wcount, file);
+        }
+
+        size_t bcount = (size_t)layer.biases.rows * layer.biases.cols;
+        if (bcount > 0 && layer.biases.buffer != NULL) {
+            fwrite(layer.biases.buffer, sizeof(MatrixType), bcount, file);
+        }
+
+        int nextOutput = (i + 1 < nn.numberOfLayers) ? (int)nn.layers[i + 1].weights.rows : 0;
+        fwrite(&nextOutput, sizeof(int), 1, file);
+    }
+
+    fclose(file);
 }
 
 void test_loadModelReturnsCorrectNumberOfLayers(void)