Timing With Events

Overview

This application shows how to use events to measure the time taken to process VPI tasks. It does an image downsampling composed of two tasks: image blurring to avoid aliasing, followed by actual downsampling. Using a 1920x1080, 8bpp single channel input image, it record events before and after processing it, and also in between the tasks. At the end it shows the time taken for each task and the total time.

This sample shows the following VPI features:

• Creating and destroying a VPI device.
• Creating a VPI-managed 2D images.
• Create a GaussianFilter and Resample algorithms and submit them.
• Creation of VPI events.
• Use events to record the state of a VPI device.
• Measure elapsed time between events.
• Simple device synchronization.
• Error handling.
• Environment clean up.

Instructions

The usage is:

./vpi_sample_05_timing <backend>

where

• backend: either cpu, cuda or pva; it defines the backend that will perform the processing.

Here's one example:

./vpi_sample_05_timing cuda

This is using the CUDA backend. Try other backends to see how the processing time differs between them.

Source code

For convenience, here's the code that is also installed in the samples directory.

/*
* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*  * Redistributions of source code must retain the above copyright
*    notice, this list of conditions and the following disclaimer.
*  * Redistributions in binary form must reproduce the above copyright
*    notice, this list of conditions and the following disclaimer in the
*    documentation and/or other materials provided with the distribution.
*  * Neither the name of NVIDIA CORPORATION nor the names of its
*    contributors may be used to endorse or promote products derived
*    from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
 
#include <vpi/Event.h>
#include <vpi/Image.h>
#include <vpi/Pyramid.h>
#include <vpi/Stream.h>
#include <vpi/algo/GaussianImageFilter.h>
#include <vpi/algo/GaussianPyramidGenerator.h>
 
#include <iostream>
 
#define CHECK_STATUS(STMT)                                      \
    do                                                          \
    {                                                           \
        VPIStatus status = (STMT);                              \
        if (status != VPI_SUCCESS)                              \
        {                                                       \
            throw std::runtime_error(vpiStatusGetName(status)); \
        }                                                       \
    } while (0);
 
int main(int argc, char *argv[])
{
    VPIImage image    = nullptr;
    VPIImage blurred  = nullptr;
    VPIPyramid output = nullptr;
    VPIStream stream  = nullptr;
 
    VPIEvent evStart   = nullptr;
    VPIEvent evBlur    = nullptr;
    VPIEvent evPyramid = nullptr;
 
    int retval = 0;
 
    try
    {
        // 1. Initialization stage ----------------------
 
        if (argc != 2)
        {
            throw std::runtime_error(std::string("Usage: ") + argv[0] + " <cpu|pva|cuda>");
        }
 
        std::string strDevType = argv[1];
 
        // Process the device type
        VPIDeviceType devType;
 
        if (strDevType == "cpu")
        {
            devType = VPI_DEVICE_TYPE_CPU;
        }
        else if (strDevType == "cuda")
        {
            devType = VPI_DEVICE_TYPE_CUDA;
        }
        else if (strDevType == "pva")
        {
            devType = VPI_DEVICE_TYPE_PVA;
        }
        else
        {
            throw std::runtime_error("Backend '" + strDevType +
                                     "' not recognized, it must be either cpu, cuda or pva.");
        }
 
        // Create the stream for the given backend.
        CHECK_STATUS(vpiStreamCreate(devType, &stream));
 
        int width = 1920, height = 1080;
        VPIImageType imgType = VPI_IMAGE_TYPE_Y16;
 
        std::cout << "Input size: " << width << " x " << height << '\n';
 
        // Create image with zero content
        CHECK_STATUS(vpiImageCreate(width, height, imgType, 0, &image));
 
        // Now create the output image
        CHECK_STATUS(vpiPyramidCreate(width, height, imgType, 3, 0.5, 0, &output));
 
        // Create a temporary image convolved with a low-pass filter.
        CHECK_STATUS(vpiImageCreate(width, height, imgType, 0, &blurred));
 
        // Create the events we'll need to get timing info
        CHECK_STATUS(vpiEventCreate(0, &evStart));
        CHECK_STATUS(vpiEventCreate(0, &evBlur));
        CHECK_STATUS(vpiEventCreate(0, &evPyramid));
 
        // 2. Processing ----------------------
 
        // Record stream queue when we start processing
        CHECK_STATUS(vpiEventRecord(evStart, stream));
 
        // First we apply a low-pass filter using gaussian to avoid aliasing
        CHECK_STATUS(vpiSubmitGaussianImageFilter(stream, image, blurred, 5, 5, 1, 1, VPI_BOUNDARY_COND_ZERO));
 
        // Record stream queue just after blurring
        CHECK_STATUS(vpiEventRecord(evBlur, stream));
 
        // Now we create a gaussian pyramid out of it.
        CHECK_STATUS(vpiSubmitGaussianPyramidGenerator(stream, blurred, output));
 
        // Record stream queue just after pyramid generation, thus capturing the whole processing
        CHECK_STATUS(vpiEventRecord(evPyramid, stream));
 
        // Wait until the gaussian pyramid finishes processing
        CHECK_STATUS(vpiEventSync(evPyramid));
 
        // 3. Timing analysis ----------------------
 
        float elapsedBlurMS, elapsedPyramidMS, elapsedTotalMS;
        CHECK_STATUS(vpiEventElapsedTime(evStart, evBlur, &elapsedBlurMS));
        CHECK_STATUS(vpiEventElapsedTime(evBlur, evPyramid, &elapsedPyramidMS));
        CHECK_STATUS(vpiEventElapsedTime(evStart, evPyramid, &elapsedTotalMS));
 
        printf("Blurring elapsed time: %f ms\n", elapsedBlurMS);
        printf("Gaussian pyramid elapsed time: %f ms\n", elapsedPyramidMS);
        printf("Total elapsed time: %f ms\n", elapsedTotalMS);
    }
    catch (std::exception &e)
    {
        std::cerr << e.what() << std::endl;
        retval = 1;
    }
 
    // 4. Clean up -----------------------------------
    vpiImageDestroy(image);
    vpiPyramidDestroy(output);
    vpiImageDestroy(blurred);
    vpiEventDestroy(evStart);
    vpiEventDestroy(evBlur);
    vpiEventDestroy(evPyramid);
    vpiStreamDestroy(stream);
 
    return retval;
}

Results

The input image is 1920x1080, 8bpp single channel. Image downsample currently isn't supported on PVA, so no data for this backend.

Note

You should not interpret these elapsed times as a performance benchmark. For this reason, we're not specifying here the hardware used to measure them.

CPU Backend

Input size: 1920 x 1080
Blurring elapsed time: 703.280823 ms
Gaussian pyramid elapsed time: 97.995987 ms
Total elapsed time: 801.276855 ms

CUDA Backend

Input size: 1920 x 1080
Blurring elapsed time: 0.531488 ms
Gaussian pyramid elapsed time: 0.313856 ms
Total elapsed time: 0.845344 ms