created by Vineet Jain, November 2013
This project involves the utilization of OpenMP, Intel SSE intrinsics, Loop Unrolling as well as various other optimization techniques to perform 2D Image Convolution using variable kernels on various different images.
Kernel Sizes (odd dimensions only): anywhere from 3-19, not necessarily square Image sizes: anything, must be 32-bit color image or 8-bit grayscale as well as in .bmp format
###Usage:
The program can be compiled using
$ make bench-extra
and then ran using the default script using
$ ./bench-extra
You can specify the size of an image using
$ ./bench-extra WIDTH HEIGHT
or an actual image using
$ ./bench-extra YOUR_IMAGE.bmp
To change the kernel, you can define it in benchmark.c. If you just change the dimensions a kernel will be randomly generated.
The output image resulted from 2D convolution can be found as a file called out_img.bmp in the same directory as the rest of your code.
Your speed output will also look like this
Kernel:
3 7 6 -3 5 2 -9 5 1
-3 -1 5 5 6 -9 -5 1 -7
4 -2 8 -4 -0 -4 -0 -0 8
-9 6 -7 3 -0 9 -4 -2 -5
-5 -2 8 8 -5 -3 -9 -1 3
4 -3 3 7 -9 9 4 -5 2
5 -6 3 0 -7 1 4 -8 3
3 4 -1 9 9 -7 2 -6 -9
3 6 -3 -2 3 6 -4 8 6
Image Dimemsions: x = 400, y = 400 Kernel Dimensions: x = 9, y = 9 Performance: 113.435 Gflop/s
Image Dimemsions: x = 400, y = 600 Kernel Dimensions: x = 9, y = 9 Performance: 121.088 Gflop/s
Image Dimemsions: x = 400, y = 800 Kernel Dimensions: x = 9, y = 9 Performance: 126.516 Gflop/s
Image Dimemsions: x = 400, y = 1000 Kernel Dimensions: x = 9, y = 9 Performance: 130.183 Gflop/s
Image Dimemsions: x = 400, y = 1200 Kernel Dimensions: x = 9, y = 9 Performance: 132.63 Gflop/s
Image Dimemsions: x = 600, y = 400 Kernel Dimensions: x = 9, y = 9 Performance: 135.365 Gflop/s
Image Dimemsions: x = 600, y = 600 Kernel Dimensions: x = 9, y = 9 Performance: 142.377 Gflop/s
Image Dimemsions: x = 600, y = 800 Kernel Dimensions: x = 9, y = 9 Performance: 142.217 Gflop/s
Image Dimemsions: x = 600, y = 1000 Kernel Dimensions: x = 9, y = 9 Performance: 143.209 Gflop/s
Image Dimemsions: x = 600, y = 1200 Kernel Dimensions: x = 9, y = 9 Performance: 144.962 Gflop/s
Image Dimemsions: x = 800, y = 400 Kernel Dimensions: x = 9, y = 9 Performance: 147.214 Gflop/s
Image Dimemsions: x = 800, y = 600 Kernel Dimensions: x = 9, y = 9 Performance: 145.402 Gflop/s
Image Dimemsions: x = 800, y = 800 Kernel Dimensions: x = 9, y = 9 Performance: 149.751 Gflop/s
Image Dimemsions: x = 800, y = 1000 Kernel Dimensions: x = 9, y = 9 Performance: 148.588 Gflop/s
Image Dimemsions: x = 800, y = 1200 Kernel Dimensions: x = 9, y = 9 Performance: 121.484 Gflop/s
Image Dimemsions: x = 1000, y = 400 Kernel Dimensions: x = 9, y = 9 Performance: 151.922 Gflop/s
Image Dimemsions: x = 1000, y = 600 Kernel Dimensions: x = 9, y = 9 Performance: 144.144 Gflop/s
Image Dimemsions: x = 1000, y = 800 Kernel Dimensions: x = 9, y = 9 Performance: 143.407 Gflop/s
Image Dimemsions: x = 1000, y = 1000 Kernel Dimensions: x = 9, y = 9 Performance: 150.839 Gflop/s
Image Dimemsions: x = 1000, y = 1200 Kernel Dimensions: x = 9, y = 9 Performance: 155.834 Gflop/s
Image Dimemsions: x = 1200, y = 400 Kernel Dimensions: x = 9, y = 9 Performance: 154.152 Gflop/s
Image Dimemsions: x = 1200, y = 600 Kernel Dimensions: x = 9, y = 9 Performance: 154.733 Gflop/s
Image Dimemsions: x = 1200, y = 800 Kernel Dimensions: x = 9, y = 9 Performance: 153.666 Gflop/s
Image Dimemsions: x = 1200, y = 1000 Kernel Dimensions: x = 9, y = 9 Performance: 157.075 Gflop/s
Image Dimemsions: x = 1200, y = 1200 Kernel Dimensions: x = 9, y = 9 Performance: 155.912 Gflop/s
Average Gflop/s = 142.644