ADD file via upload

neon优化稠密矩阵乘法及时间比较 (1).c

@@ -0,0 +1,104 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <time.h>
+#include <arm_neon.h>
+
+void matmul(float** A, float** B, float** C, int n) {
+	int i,k,j; 
+    for (i = 0; i < n; i++) {
+        for (j = 0; j < n; j++) {
+            C[i][j] = 0;
+            for (k = 0; k < n; k++) {
+                C[i][j] += A[i][k] * B[k][j];
+            }
+        }
+    }
+}
+
+// NEON优化的矩阵乘法函数
+void matmul_optimized(float** A, float** B, float** C, int n) {
+	int i,j,k;
+    for (i = 0; i < n; i++) {
+        for (j = 0; j < n; j++) {
+            // 初始化结果矩阵C的当前元素为0
+            C[i][j] = 0;
+
+            // 用于累加的向量寄存器，初始化为0
+            float32x4_t vecC = vdupq_n_f32(0);
+
+            for (k = 0; k < n; k += 4) {
+                // 加载矩阵A的一行中的4个连续元素到向量寄存器
+                float32x4_t vecA = vld1q_f32(&A[i][k]);
+                // 加载矩阵B的一列中的4个连续元素到向量寄存器（注意这里要转置的逻辑，实际是按列取元素）
+                float32x4_t vecB = vld1q_f32(&B[k][j]);
+
+                // 对应元素相乘并累加到vecC
+                vecC = vmlaq_f32(vecC, vecA, vecB);
+            }
+
+            // 将累加结果从向量寄存器提取并累加到C[i][j]
+            C[i][j] += vgetq_lane_f32(vecC, 0) + vgetq_lane_f32(vecC, 1) + vgetq_lane_f32(vecC, 2) + vgetq_lane_f32(vecC, 3);
+        }
+    }
+}
+
+int main() {
+    // 定义矩阵大小
+    int n = 1024;
+
+    // 动态分配两个输入矩阵A和B，以及结果矩阵C的内存
+    float** A = (float**)malloc(n * sizeof(float*));
+    float** B = (float**)malloc(n * sizeof(float*));
+    float** C = (float**)malloc(n * sizeof(float*));
+
+	int i,j;
+    for (i = 0; i < n; i++) {
+        A[i] = (float*)malloc(n * sizeof(float));
+        B[i] = (float*)malloc(n * sizeof(float));
+        C[i] = (float*)malloc(n * sizeof(float));
+    }
+
+    // 初始化矩阵数据，将A和B矩阵的每个元素随机初始化
+    srand((unsigned int)time(NULL));
+    for (i = 0; i < n; i++) {
+        for (j = 0; j < n; j++) {
+            A[i][j] = (float)(rand() % 100);
+            B[i][j] = (float)(rand() % 100);
+        }
+    }
+
+    clock_t start, end;
+
+    // 测试普通矩阵乘法函数的运行时间
+    start = clock();
+    matmul(A, B, C, n);
+    end = clock();
+    double time_taken_normal = ((double)(end - start)) / CLOCKS_PER_SEC;
+    printf("normal time: %lf s\n", time_taken_normal);
+
+    // 重新初始化结果矩阵C为0，以便再次使用
+    for (i = 0; i < n; i++) {
+        for (j = 0; j < n; j++) {
+            C[i][j] = 0;
+        }
+    }
+
+    // 测试NEON优化矩阵乘法函数的运行时间
+    start = clock();
+    matmul_optimized(A, B, C, n);
+    end = clock();
+    double time_taken_optimized = ((double)(end - start)) / CLOCKS_PER_SEC;
+    printf("NEON time: %lf s\n", time_taken_optimized);
+
+    // 释放动态分配的内存空间
+    for ( i = 0; i < n; i++) {
+        free(A[i]);
+        free(B[i]);
+        free(C[i]);
+    }
+    free(A);
+    free(B);
+    free(C);
+
+    return 0;
+}