Update speed_filter2d_in_gray_image.cpp

Author: BBuf

Algorithm optimization/speed_filter2d_in_gray_image.cpp

@@ -10,7 +10,7 @@ using namespace cv;
 //RGB转化为灰度图
 Mat speed_rgb2gray(Mat src) {
 	Mat dst(src.rows, src.cols, CV_8UC1);
-#pragma omp parallel for num_threads(4)
+//#pragma omp parallel for num_threads(4)
 	for (int i = 0; i < src.rows; i++) {
 		for (int j = 0; j < src.cols; j++) {
 			dst.at<uchar>(i, j) = ((src.at<Vec3b>(i, j)[0] << 18) + (src.at<Vec3b>(i, j)[0] << 15) + (src.at<Vec3b>(i, j)[0] << 14) +
@@ -27,8 +27,8 @@ Mat speed_rgb2gray(Mat src) {
 
 //A增加Pad的运算
 void get_Pad(int pad_Height, int pad_Width, int row, int col, float *A_pad, float *A) {
-	int pad_x = pad_Height - row >> 1;
-	int pad_y = pad_Width - col >> 1;
+	int pad_x = (pad_Height - row) >> 1;
+	int pad_y = (pad_Width - col) >> 1;
 	printf("pad_x: %d pad_y: %d\n", pad_x, pad_y);
 	for (int i = 0; i < pad_Height; i++) {
 		for (int j = 0; j < pad_Width; j++) {
@@ -48,7 +48,45 @@ void get_Pad(int pad_Height, int pad_Width, int row, int col, float *A_pad, floa
 	}
 }
 
+//pad_A的转换，以适用于openblas，row2col的思想
+void convert_A(float *A_convert, const int OutHeight, const int OutWidth, const int pad_Height, const int pad_Width, float *A_pad) {
+	for (int i = 0; i < OutHeight; i++) {
+		for (int j = 0; j < OutWidth; j++) {
+			int index = i * OutHeight * pad_Height + j * pad_Width;
+			int col1 = i * pad_Height + j;
+			//row2col展开，这里是3*3卷积，展开9次
+			A_convert[index] = A_pad[col1];
+			A_convert[index + 1] = A_pad[col1 + 1];
+			A_convert[index + 2] = A_pad[col1 + 2];
 
+			int col2 = (i + 1) * pad_Height + j;
+			A_convert[index + 3] = A_pad[col2];
+			A_convert[index + 4] = A_pad[col2 + 1];
+			A_convert[index + 5] = A_pad[col2 + 2];
+
+			int col3 = (i + 2) * pad_Height + j;
+			A_convert[index + 6] = A_pad[col3];
+			A_convert[index + 7] = A_pad[col3 + 1];
+			A_convert[index + 8] = A_pad[col3 + 2];
+		}
+	}
+}
+//OpenBlas调用sgemm算法
+void Matrixmul_blas(const int convAh, const int convAw, float *A_convert, float *B, float *C) {
+	const enum CBLAS_ORDER Order = CblasRowMajor;
+	const enum CBLAS_TRANSPOSE TransA = CblasNoTrans;
+	const enum CBLAS_TRANSPOSE TransB = CblasNoTrans;
+	const int M = convAh;//A的行数，C的行数
+	const int N = 1;//B的列数，C的列数
+	const int K = convAw;//A的列数，B的行数
+	const float alpha = 1;
+	const float beta = 0;
+	const int lda = K;//A的列
+	const int ldb = N;//B的列
+	const int ldc = N;//C的列
+
+	cblas_sgemm(Order, TransA, TransB, M, N, K, alpha, A_convert, lda, B, ldb, beta, C, ldc);
+}
 
 int main() {
 	Mat src = cv::imread("F:\\1.jpg");
@@ -68,13 +106,32 @@ int main() {
 	//卷积核参数初始化为
 	const int pad = (KernelHeight - 1) / 2; //需要pad的长度
 	const int stride = 1; //卷积核滑动的步长
-	//计算卷积输出矩阵的长宽
+						  //计算卷积输出矩阵的长宽
 	const int OutHeight = (row - KernelHeight + 2 * pad) / stride + 1;
 	const int OutWidth = (col - KernelWidth + 2 * pad) / stride + 1;
 	//计算pad_A
 	const int pad_Height = row + 2 * pad;
 	const int pad_Width = col + 2 * pad;
 	float *A_pad = new float[pad_Height * pad_Width];
 	get_Pad(pad_Height, pad_Width, row, col, A_pad, A);
-
+	//定义被卷积矩阵宽高
+	const int convAw = KernelHeight * KernelWidth;
+	const int convAh = OutHeight * OutWidth;
+	//转换被卷积矩阵
+	float *A_convert = new float[convAh * convAw];
+	convert_A(A_convert, OutHeight, OutWidth, pad_Height, pad_Width, A_pad);
+	//定义卷积输出矩阵
+	float *C = new float[convAh * 1];
+	//sgemm算法计算输出矩阵
+	Matrixmul_blas(convAh, convAw, A_convert, B, C);
+	//输出验证
+	Mat dst(OutHeight, OutWidth, CV_32FC1);
+	for (int i = 0; i < OutHeight; i++) {
+		for (int j = 0; j < OutWidth; j++) {
+			dst.at<float>(i, j) = C[i * OutHeight + j];
+		}
+	}
+	cv::imshow("result", dst);
+	cv::waitKey(0);
+	return 0;
 }