GPU610/TeamEh
Team Eh
Team Members
- Benjamin Snively, Some responsibility
- Brad Hoover, Some other responsibility
- Balint Czunyi, Some other responsibility
- ...
Progress
Assignment 1
Benjamin Snively's Results
Introduction
This image processing program was found on github. It processes and manipulates images using convolutions matrices (kernels). It has several different functions including aligning and sharpening images.
To convolve an image the kernel is applied to each pixel. Using the kernel, the pixel's value is combined with that of its neighbors to create a new pixel value. This program implements the filter using two loops to loop over each pixel in sequence. For a given an image convolution is an O(rows x columns) function. As blurring operation on each pixel is independent of the others, therefore it is a perfect candidate for parallelization.
To profile the application, I created a large bitmap file (about 800 x 800, 2MB) and ran it through three different operations. To conserve space, I have not included a profile of all of the available operations.
Gassian Blur
Command: --gassian 5
Each sample counts as 0.01 seconds.
% cumulative self self total time seconds seconds calls s/call s/call name 44.81 88.57 88.57 _mcount_private 31.92 151.66 63.09 __fentry__ 4.85 161.25 9.59 1 9.59 45.84 Gauss_filter::smooth_ord(Matrix<std::tuple<unsigned int, unsigned int, unsigned int> >&) 1.92 165.05 3.80 633231640 0.00 0.00 Matrix<std::tuple<unsigned int, unsigned int, unsigned int> >::operator()(unsigned int, unsigned int) 1.43 167.88 2.83 633887736 0.00 0.00 std::__shared_ptr<std::tuple<unsigned int, unsigned int, unsigned int>, (__gnu_cxx::_Lock_policy)2>::get() const 1.37 170.58 2.70 630508256 0.00 0.00 std::_Tuple_impl<0ul, int&, int&, int&>& std::_Tuple_impl<0ul, int&, int&, int&>::operator=<unsigned int, unsigned int, unsigned int>(std::_Tuple_impl<0ul, unsigned int, unsigned int, unsigned int> const&) 0.92 172.40 1.82 630508256 0.00 0.00 std::_Head_base<0ul, int&, false>::_Head_base(int&) 0.87 174.12 1.72 630508256 0.00 0.00 std::_Tuple_impl<2ul, int&>& std::_Tuple_impl<2ul, int&>::operator=<unsigned int>(std::_Tuple_impl<2ul, unsigned int> const&) 0.86 175.81 1.69 630508256 0.00 0.00 std::_Head_base<1ul, int&, false>::_Head_base(int&) 0.84 177.47 1.66 630508256 0.00 0.00 std::_Tuple_impl<1ul, int&, int&>& std::_Tuple_impl<1ul, int&, int&>::operator=<unsigned int, unsigned int>(std::_Tuple_impl<1ul, unsigned int, unsigned int> const&) 0.78 179.02 1.55 630508256 0.00 0.00 std::_Head_base<2ul, int&, false>::_Head_base(int&) 0.77 180.54 1.52 630508256 0.00 0.00 std::tuple<int&, int&, int&> std::tie<int, int, int>(int&, int&, int&) 0.74 182.00 1.46 630508256 0.00 0.00 std::_Tuple_impl<2ul, int&>::_Tuple_impl(int&) 0.65 183.28 1.28 630508256 0.00 0.00 std::tuple<int&, int&, int&>& std::tuple<int&, int&, int&>::operator=<unsigned int, unsigned int, unsigned int, void>(std::tuple<unsigned int, unsigned int, unsigned int> const&) 0.57 184.41 1.13 630508256 0.00 0.00 std::tuple<int&, int&, int&>::tuple(int&, int&, int&) 0.55 185.50 1.09 630508256 0.00 0.00 std::_Head_base<0ul, int&, false>::_M_head(std::_Head_base<0ul, int&, false>&) 0.52 186.53 1.03 630508256 0.00 0.00 std::_Tuple_impl<0ul, int&, int&, int&>::_Tuple_impl(int&, int&, int&)
Sharpen
Command: --unsharp
Each sample counts as 0.01 seconds.
% cumulative self self total time seconds seconds calls ms/call ms/call name 44.44 0.96 0.96 _mcount_private 27.31 1.55 0.59 __fentry__ 7.41 1.71 0.16 1 160.00 458.44 unsharp(Matrix<std::tuple<unsigned int, unsigned int, unsigned int> >) 5.56 1.83 0.12 20345464 0.00 0.00 Matrix<std::tuple<unsigned int, unsigned int, unsigned int> >::operator()(unsigned int, unsigned int) 1.39 1.86 0.03 21001560 0.00 0.00 std::__shared_ptr<std::tuple<unsigned int, unsigned int, unsigned int>, (__gnu_cxx::_Lock_policy)2>::get() const 1.39 1.89 0.03 7876396 0.00 0.00 std::_Tuple_impl<0ul, unsigned int, unsigned int, unsigned int>::_M_head(std::_Tuple_impl<0ul, unsigned int, unsigned int, unsigned int>&) 1.39 1.92 0.03 656096 0.00 0.00 std::_Tuple_impl<2ul, unsigned int>& std::_Tuple_impl<2ul, unsigned int>::operator=<unsigned char>(std::_Tuple_impl<2ul, unsigned char>&&) 0.93 1.94 0.02 7876396 0.00 0.00 std::_Tuple_impl<1ul, unsigned int, unsigned int>::_M_head(std::_Tuple_impl<1ul, unsigned int, unsigned int>&) 0.93 1.96 0.02 1968288 0.00 0.00 unsigned char&& std::forward<unsigned char>(std::remove_reference<unsigned char>::type&) 0.93 1.98 0.02 656096 0.00 0.00 std::_Head_base<0ul, unsigned char, false>::_M_head(std::_Head_base<0ul, unsigned char, false>&)
Identity
command: --custom '0,0,0,0,1,0,0,0,0'
Each sample counts as 0.01 seconds.
% cumulative self self total time seconds seconds calls ms/call ms/call name 53.61 1.71 1.71 _mcount_private 28.21 2.61 0.90 __fentry__ 4.39 2.75 0.14 2 70.00 218.45 Use_kernel::new_im() 1.88 2.81 0.06 8542240 0.00 0.00 Matrix<std::tuple<unsigned int, unsigned int, unsigned int> >::operator()(unsigned int, unsigned int) 0.94 2.84 0.03 5904864 0.00 0.00 std::_Tuple_impl<0ul, int&, int&, int&>::_Tuple_impl(int&, int&, int&) 0.63 2.86 0.02 13126802 0.00 0.00 __gnu_cxx::__enable_if<std::__is_integer<int>::__value, double>::__type std::floor<int>(int) 0.63 2.88 0.02 9841440 0.00 0.00 double& std::forward<double&>(std::remove_reference<double&>::type&) 0.63 2.90 0.02 7223552 0.00 0.00 std::_Head_base<0ul, unsigned int, false>::_M_head(std::_Head_base<0ul, unsigned int, false> const&) 0.63 2.92 0.02 5904864 0.00 0.00 std::_Tuple_impl<1ul, int&, int&>& std::_Tuple_impl<1ul, int&, int&>::operator=<unsigned int, unsigned int>(std::_Tuple_impl<1ul, unsigned int, unsigned int> const&) 0.63 2.94 0.02 5904864 0.00 0.00 std::_Tuple_impl<2ul, int&>::_Tuple_impl(int&)
Summary
The functions that perform the filtering are Gauss_filter::smooth_ord, unsharp and Use_kernel::new_im(). These functions are all O(r x c) with respect to image dimensions and thus where the biggest gains from parallelization will be found.