conv1d_gemm_?

General 1D convolution interface implemented by the GEMM algorithm.

Interface Definition

C interface:

void conv1d_gemm_fp32(const float *input, const int batch, const int inputChannels, const int inputLength, const float* kernel, const int kernelLength, const int stride, const int padLength, const int dilation, const float *bias, float *output, const int outputChannels);

void conv1d_gemm_fp16(const __fp16 *input, const int batch, const int inputChannels, const int inputLength, const __fp16 *kernel, const int kernelLength, const int stride, const int padLength, const int dilation, const __fp16 *bias, __fp16 *output, const int outputChannels);

Parameters

Parameter	Type	Description	Input/Output
input	Float type in conv1d_gemm_fp32 __fp16 type in conv1d_gemm_fp16	Input data.	Input
batch	Int	Number of batches of input data.	Input
inputChannels	Int	Number of input channels.	Input
inputLength	Int	Length of the input data.	Input
kernel	Float type in conv1d_gemm_fp32 __fp16 type in conv1d_gemm_fp16	Convolution kernel.	Input
kernelLength	Int	Convolution kernel length.	Input
stride	Int	Stride.	Input
padLength	Int	Zero-padding length at both ends of the original input data.	Input
dilation	Int	Dilation rate.	Input
bias	Float type in conv1d_gemm_fp32 __fp16 type in conv1d_gemm_fp16	Bias. The value NULL indicates that there is no bias.	Input
output	Float type in conv1d_gemm_fp32 __fp16 type in conv1d_gemm_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 inputLength = 10;
    int kernelLength = 3;
    int stride = 1;
    int padLength = 0;
    int dilation = 1;
    int outputChannels = 1;
    float input[10] = {1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0};
    float kernel[3] = {1.0, 2.0, 3.0};
    float *bias = NULL;

    int outputLength = (inputLength + 2 * padLength - dilation * (kernelLength - 1) - 1) / stride + 1;
    /*
     *   outputLength = 8
     */
    float output[8] = {0.0};

    conv1d_gemm_fp32(input, batch, inputChannels, inputLength, kernel, kernelLength, stride, padLength, dilation, bias, output, outputChannels);
    /*
     * output = [14.0 20.0 26.0 32.0 38.0 44.0 50.0 56.0]
     */

Parent topic: General Convolution Interfaces