ADD file via upload

neon优化加法及时间比较 (4).c

@@ -0,0 +1,62 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <time.h>
+#include <arm_neon.h>
+
+#define SIZE 1024
+
+void vector_add(float* A, float* B, float* C, int size) {
+	int i;
+    for (i = 0; i < size; i++) {
+        C[i] = A[i] + B[i];
+    }
+}
+
+void vector_add_optimized(float* A, float* B, float* C, int size) {
+    int i;
+    for (i = 0; i < size - 3; i += 4) {
+        float32x4_t a_vec = vld1q_f32(&A[i]);
+        float32x4_t b_vec = vld1q_f32(&B[i]);
+
+        float32x4_t c_vec = vaddq_f32(a_vec, b_vec);
+
+        vst1q_f32(&C[i], c_vec);
+    }
+
+    for (; i < size; i++) {
+        C[i] = A[i] + B[i];
+    }
+}
+
+int main() {
+    float A[SIZE];
+    float B[SIZE];
+    float C[SIZE];
+
+    srand((unsigned int)time(NULL));
+    int i;
+    for (i = 0; i < SIZE; i++) {
+        A[i] = (float)(rand() % 100);
+        B[i] = (float)(rand() % 100);
+    }
+
+    clock_t start, end;
+
+    start = clock();
+    vector_add(A, B, C, SIZE);
+    end = clock();
+    double time_taken_normal = ((double)(end - start)) / CLOCKS_PER_SEC;
+    printf("normal time: %lf s\n", time_taken_normal);
+
+    for (i = 0; i < SIZE; i++) {
+        C[i] = 0;
+    }
+
+    start = clock();
+    vector_add_optimized(A, B, C, SIZE);
+    end = clock();
+    double time_taken_optimized = ((double)(end - start)) / CLOCKS_PER_SEC;
+    printf("NEON time: %lf s\n", time_taken_optimized);
+
+    return 0;
+}