Add task3_3.cpp

task3_3.cpp

@@ -0,0 +1,97 @@
+#include <stdio.h>
+#include <ctime>
+#include <stdlib.h>
+#include<arm_neon.h>
+
+#define H 5
+#define W 500
+
+float kx[3]={0.25,0.5,0.25};
+float ky[3]={0.25,0.5,0.25};
+
+float32x4_t kx_0={kx[0],kx[0],kx[0],kx[0]};
+float32x4_t kx_1={kx[1],kx[1],kx[1],kx[1]};
+float32x4_t kx_2={kx[2],kx[2],kx[2],kx[2]};
+
+float32x4_t ky_0={ky[0],ky[0],ky[0],ky[0]};
+float32x4_t ky_1={ky[1],ky[1],ky[1],ky[1]};
+float32x4_t ky_2={ky[2],ky[2],ky[2],ky[2]};
+
+void applySeparableGaussianBlur(float src[H][W], float dst[H][W], int h, int w, float kx[3], float ky[3])
+{
+        float buf[H][W-1]={0};
+
+         for(int k = 0; k< h; ++k)
+        {
+            for (int j = 1; j < w - 1;j+=4)
+            {
+
+                float32x4_t left =vld1q_f32(&src[k][j-1]);
+                float32x4_t mid =vld1q_f32(&src[k][j]);
+                float32x4_t right =vld1q_f32(&src[k][j+1]);
+
+                float32x4_t result =vmulq_f32(left, kx_0);
+                result=vmlaq_f32(result,mid,kx_1);
+                result=vmlaq_f32(result,right,kx_2);
+
+                vst1q_f32(&buf[k][j],result);              
+                }
+        }       
+        //开始进行可分离卷积 
+        for (int i = 2; i < h ; i++) 
+        {
+            // 计算当前行的行内卷积 
+            for (int j = 1; j < w - 1; j+=4) 
+            {    
+
+                float32x4_t left1 =vld1q_f32(&src[i][j-1]);
+                float32x4_t mid1 =vld1q_f32(&src[i][j]);
+                float32x4_t right1 =vld1q_f32(&src[i][j+1]);
+
+                float32x4_t result1 =vmulq_f32(left1, ky_0);
+                result1=vmlaq_f32(result1,mid1,ky_1);
+                result1=vmlaq_f32(result1,right1,ky_2);
+
+            vst1q_f32(&dst[i-1][j],result1);  
+            }
+        }
+
+}
+int main()
+{
+    float inputImage[H][W]={0};
+     float dst[H][W]={0};
+    float a[H][W]={0};
+
+    for(int i=0;i<H;i++)
+    {
+        for(int j=0;j<W;j++)
+        {
+            inputImage[i][j]=W*i+j+1;
+            dst[i][j]=W*i+j+1;
+            a[i][j]=W*i+j+1;
+        }
+    }
+
+    clock_t start = clock();
+    applySeparableGaussianBlur(inputImage, dst, H, W, kx, ky);
+    clock_t end = clock();
+    double time_spent = double(end - start) / CLOCKS_PER_SEC;
+    printf("运行时间：%lf秒\n",time_spent);
+    for(int i=0;i<H;i++)
+    {
+        for(int j=0;j<W;j++)
+        {
+            if(i==0||i==H-1||j==0||j==W-1)
+            {
+                printf("%.1f ",a[i][j]);
+            }else
+            {
+                printf("%.1f ",dst[i][j]);
+
+            }
+        }
+        printf("\n");
+    }
+}
+