ETF_gpu.m~

%% ETF on GPU
function [tx_new, ty_new, tmag] = ETF_gpu(tx,ty,gmag_norm,im,kern)

% set kernel file to be loaded
% cudaFilename = '/home/bee/TRY/kernel.cu';
% ptxFilename = '/home/bee/TRY/kernel.ptx';
cudaFilename = '/home/bee/TRY/etfStraight.cu';
ptxFilename = '/home/bee/TRY/etfStraight.ptx';
kernel = parallel.gpu.CUDAKernel(ptxFilename, cudaFilename);

% host to device memcpy
[r,c] = size(im);
im = gpuArray(im)';
gmag = gpuArray(gmag_norm)';
tmag = zeros(size(gmag),'gpuArray');
tx = gpuArray(tx)';
ty = gpuArray(ty)';
tx_new = zeros(size(tx),'gpuArray');
ty_new = zeros(size(tx),'gpuArray');

% initialize kernel dimensions
blockDim_x = ceil(c/32);
blockDim_y = ceil(r/32);
kernel.ThreadBlockSize = [32,32,1];
kernel.GridSize = [blockDim_x,blockDim_y,1];

% kernel function call
[tx_new, ty_new, tmag] = feval(kernel, tx_new,ty_new,tmag, tx,ty,im,gmag,r,c);

% device to host memcpy
tx_new = gather(tx_new);
ty_new = gather(ty_new);