The NPPIRESIZE function does not output a value in different cases

[1446561717]

I tried to compress my images using the NPP library, which I developed on Jetson ORIN64G, using cuda-11.4.

I tried to use the RESZIE function, initially using 1024 * 768 data and changing it to 512 * 768. It works, but I actually need to compress the 32000 * 121 data to 2048 * 121, but it outputs a value of 0, and it doesn’t. My code is as follows

#include <stdio.h>
#include <stdlib.h>
#include <npp.h>
#include <cuda_runtime.h>
#include <nppi.h>
#include <nppdefs.h>
#include <iostream>


using namespace std;
#define NPP_INTER_LINEAR 2
//#include<nppdef.h>
// Function to resize a 2D matrix using NPP
    float* decimate_cuda(float* readbuff, int nSrcH, int nSrcW, int nDstH, int nDstW)
    {
            size_t  srcStep;
            size_t  dstStep;

            NppiSize oSrcSize = {nSrcW, nSrcH};
            NppiRect oSrcROI = {0, 0, nSrcW, nSrcH};
            float *devSrc;
            cudaMallocPitch((void**)&devSrc, &srcStep, nSrcW * sizeof(float), nSrcH);
            cudaMemcpy2D(devSrc, srcStep,readbuff, nSrcW * sizeof(Npp32f), nSrcW * sizeof(Npp32f), nSrcH, cudaMemcpyHostToDevice);

            NppiSize oDstSize = {nDstW, nDstH};
            NppiRect oDstROI = {0, 0, nDstW, nDstH};
            float *devDst;
            cudaMallocPitch((void**)&devDst, &dstStep, nDstW * sizeof(float), nDstH);
	    printf("nSrcW=%d srcStep=%d\n",nSrcW,srcStep);
            NppStatus result = nppiResize_32f_C1R(devSrc,srcStep,oSrcSize,oSrcROI,devDst,dstStep,oDstSize,oDstROI,NPPI_INTER_SUPER);
            if (result != NPP_SUCCESS) {
                std::cout << "Unable to run decimate_cuda, error " << result << std::endl;
                exit(0);
            }

            Npp64s                 writesize;
            Npp32f                 *hostDst;
            writesize = (Npp64s)   nDstW * nDstH;         // Y
            if(NULL == (hostDst = (Npp32f *)malloc(writesize * sizeof(Npp32f)))){
                printf("Error : Unable to alloctae hostDst in decimate_cuda, exiting...\n");
                exit(1);
            }

            cudaMemcpy2D(hostDst, nDstW * sizeof(Npp32f),devDst, dstStep, nDstW * sizeof(Npp32f),nDstH, cudaMemcpyDeviceToHost);
            cudaFree(devSrc);
            cudaFree(devDst);
            return(hostDst);
        }     

int main() {
    const int width = 1030;
    const int height = 513;

    // Allocate memory for the input and output matrices
    float* inputMatrix = (float*)malloc(width * height * sizeof(float));
    float* outputMatrix = (float*)malloc(width * height * sizeof(float));

    // Initialize the input matrix with some data
    // (You can replace this with your own data)
    for (int i = 0; i < height; ++i) 
    {
    	for(int j=0;j<width;j++)
        inputMatrix[i*width+j] = j;
    }


    // Specify the desired output size (e.g., half of the input size)
    const int outputWidth = width / 2;
    const int outputHeight = height ;
    

    // Call the resize function with OpenACC
    #pragma acc enter data copyin(inputMatrix[0:width*height])
    outputMatrix=decimate_cuda(inputMatrix, height, width,outputWidth, outputHeight);


    for(int i=0;i<20;i++)
    printf("input[%d]=%f, outputMatrix[%d]=%f\n",i,inputMatrix[i],i,outputMatrix[i]);

    // Free allocated memory
    free(inputMatrix);
    free(outputMatrix);

    return 0;
}

[Robert_Crovella]

when I run your code I get an error -23 NPP_RESIZE_FACTOR_ERROR

If I switch from NPP_INTER_SUPER to NPP_INTER_LINEAR, or NPP_INTER_LANCZOS, the error -23 goes away (and the output matrix has non-zero values)

NPP_INTER_SUPER cannot be used when the resize factor is exactly 1.0. See here Your vertical resize factor going from 32000x121 to 2048x121 is exactly 1.0.

Also see here:

• NPP_RESIZE_FACTOR_ERROR Indicates an error condition if either nXFactor or
  nYFactor is less than or equal to zero or in the case of NPPI_INTER_SUPER are not both downscaling.

Also, you should not be doing this:

@1446561717

#define NPP_INTER_LINEAR 2

NPP defines that already for you.

As a side note, this is creating a possible memory leak:

float* outputMatrix = (float*)malloc(width * height * sizeof(float));
...
outputMatrix=decimate_cuda(inputMatrix, height, width,outputWidth, outputHeight);

Finally, I imagine this is leftover junk, but you cannot use this method with OpenACC parallelization:

@1446561717

    // Call the resize function with OpenACC
    #pragma acc enter data copyin(inputMatrix[0:width*height])

[1446561717]

Thank you very much for your answer. The problem has been successfully resolved