conv2d_?

General 2D convolution interface that automatically selects a high-performance implementation based on the input features.

Interface Definition

C interface:

void conv2d_fp32(const float *input, const int batch, const int inputChannels, const int inputHeight, const int inputWidth, const float *kernel, const int kernelHeight, const int kernelWidth, const int strideY, const int strideX, const int padHeight, const int padWidth, const int dilationY, const int dilationX, const float *bias, float *output, const int outputChannels);

void conv2d_fp16(const __fp16 *input, const int batch, const int inputChannels, const int inputHeight, const int inputWidth, const __fp16 *kernel, const int kernelHeight, const int kernelWidth, const int strideY, const int strideX, const int padHeight, const int padWidth, const int dilationY, const int dilationX, const __fp16 *bias, __fp16 *output, const int outputChannels);

Parameters

Parameter	Type	Description	Input/Output
input	Float type in conv2d_fp32 __fp16 type in conv2d_fp16	Input data.	Input
batch	Int	Number of batches of input data.	Input
inputChannels	Int	Number of input channels.	Input
inputHeight	Int	Height of the input data.	Input
inputWidth	Int	Width of the input data.	Input
kernel	Float type in conv2d_fp32 __fp16 type in conv2d_fp16	Convolution kernel.	Input
kernelHeight	Int	Height of the convolution kernel.	Input
kernelWidth	Int	Width of the convolution kernel.	Input
strideY	Int	Stride of the convolution kernel in the height direction.	Input
strideX	Int	Stride of the convolution kernel in the width direction.	Input
padHeight	Int	Zero-padding length at both ends of the original input data's height.	Input
padWidth	Int	Zero-padding length at both ends of the original input data's width.	Input
dilationY	Int	Dilation rate in the height direction.	Input
dilationX	Int	Dilation rate in the width direction.	Input
bias	Float type in conv2d_fp32 __fp16 type in conv2d_fp16	Bias. The value NULL indicates that there is no bias.	Input
output	Float type in conv2d_fp32 __fp16 type in conv2d_fp16	Output result data.	Output
outputChannels	Int	Number of output channels.	Input

Dependencies

#include "conv.h"

Examples

C interface:

    int batch = 1;
    int inputChannels = 1;
    int inputHeight = 6;
    int inputWidth = 6;
    int kernelHeight = 3;
    int kernelWidth = 3;
    int strideY = 1;
    int strideX = 1;
    int padHeight = 0;
    int padWidth = 0;
    int dilationY = 1;
    int dilationX = 1;
    int outputChannels = 1;
    float input[36] = {1.0, 2.0, 3.0, 4.0, 5.0, 6.0,
                       7.0, 8.0, 9.0, 10.0, 11.0, 12.0,
                       13.0, 14.0, 15.0, 16.0, 17.0, 18.0,
                       19.0, 20.0, 21.0, 22.0, 23.0, 24.0,
                       25.0, 26.0, 27.0, 28.0, 29.0, 30.0,
                       31.0, 32.0, 33.0, 34.0, 35.0, 36.0};
    float kernel[9] = {1.0, 2.0, 3.0,
                       4.0, 5.0, 6.0,
                       7.0, 8.0, 9.0};

    float *bias = NULL;
    int outputHeight = (inputHeight + 2 * padHeight - dilationY * (kernelHeight - 1) - 1) / strideY + 1;
    int outputWidth = (inputWidth + 2 * padWidth - dilationX * (kernelWidth - 1) - 1) / strideX + 1;
    /*
     *   outputHeight x outputWidth = 4 x 4
     */

    float output[16] = {0.0};
    conv2d_fp32(input, batch, inputChannels, inputHeight, inputWidth, kernel, kernelHeight, kernelWidth, strideY, strideX, padHeight, padWidth, dilationY, dilationX, bias, output, outputChannels);
    /*
     * output = [474.0 519.0 564.0 609.0,
     *           744.0 789.0 834.0 879.0, 
     *           1014.0 1059.0 1104.0 1149.0, 
     *           1284.0 1329.0 1374.0 1419.0]
     */

Parent topic: General Convolution Interfaces