Optimize matrix multiplication to be more cache-friendly.

Mnist-1000: 15s -> 7.74s.

1 files changed, 11 insertions, 9 deletions

diff --git a/src/lib/src/matrix.c b/src/lib/src/matrix.c
index 29cdec5..f937c01 100644
--- a/src/lib/src/matrix.c
+++ b/src/lib/src/matrix.c
@@ -131,21 +131,23 @@ void nnMatrixMul(const nnMatrix* left, const nnMatrix* right, nnMatrix* out) {
   assert(out->cols == right->cols);
 
   R* out_value = out->values;
+  for (int i = 0; i < out->rows * out->cols; ++i) {
+    *out_value++ = 0;
+  }
 
   for (int i = 0; i < left->rows; ++i) {
-    const R* left_row = &left->values[i * left->cols];
+    const R* p_left_value = &left->values[i * left->cols];
 
-    for (int j = 0; j < right->cols; ++j) {
-      const R* right_col = &right->values[j];
-      *out_value = 0;
+    for (int j = 0; j < left->cols; ++j) {
+      const R left_value = *p_left_value;
+      const R* right_value = &right->values[j * right->cols];
+      R* out_value = &out->values[i * out->cols];
 
-      // Vector dot product.
-      for (int k = 0; k < left->cols; ++k) {
-        *out_value += left_row[k] * right_col[0];
-        right_col += right->cols;  // Next row in the column.
+      for (int k = 0; k < right->cols; ++k) {
+        *out_value++ += left_value * *right_value++;
       }
 
-      out_value++;
+      p_left_value++;
     }
   }
 }