diff options
| author | Isabella Gottardi <isabella.gottardi@arm.com> | 2017-08-29 13:47:03 +0100 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:35:24 +0000 |
| commit | 83be745adba7a9928c03beda65a6a83f14846475 (patch) | |
| tree | 3a95d8ec8dc853256e469dadba4dd57360db57d7 /src/core/CL | |
| parent | ab60fe8cd25232fdefef95b86e18482343f9d05f (diff) | |
| download | ComputeLibrary-83be745adba7a9928c03beda65a6a83f14846475.tar.gz | |
COMPMID-424 Implemented reference implementation and tests for WarpAffine
Change-Id: I4924ab1de17adc3b880a5cc22f2497abbc8e221b
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/85820
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Steven Niu <steven.niu@arm.com>
Diffstat (limited to 'src/core/CL')
| -rw-r--r-- | src/core/CL/cl_kernels/warp_affine.cl | 4 | ||||
| -rw-r--r-- | src/core/CL/kernels/CLWarpAffineKernel.cpp | 7 |
2 files changed, 7 insertions, 4 deletions
diff --git a/src/core/CL/cl_kernels/warp_affine.cl b/src/core/CL/cl_kernels/warp_affine.cl index 0a4748f452..f41821cdca 100644 --- a/src/core/CL/cl_kernels/warp_affine.cl +++ b/src/core/CL/cl_kernels/warp_affine.cl @@ -84,7 +84,7 @@ __kernel void warp_affine_nearest_neighbour( { Image in = CONVERT_TO_IMAGE_STRUCT_NO_STEP(in); Image out = CONVERT_TO_IMAGE_STRUCT(out); - vstore4(read_texels4(&in, convert_int8(clamp_to_border(apply_affine_transform(get_current_coords(), build_affine_mtx()), width, height))), 0, out.ptr); + vstore4(read_texels4(&in, convert_int8_rtn(clamp_to_border(apply_affine_transform(get_current_coords(), build_affine_mtx()), width, height))), 0, out.ptr); } /** Performs an affine transform on an image interpolating with the BILINEAR method. Input and output are single channel U8. @@ -116,5 +116,5 @@ __kernel void warp_affine_bilinear( { Image in = CONVERT_TO_IMAGE_STRUCT_NO_STEP(in); Image out = CONVERT_TO_IMAGE_STRUCT(out); - vstore4(bilinear_interpolate(&in, clamp_to_border(apply_affine_transform(get_current_coords(), build_affine_mtx()), width, height), width, height), 0, out.ptr); + vstore4(bilinear_interpolate(&in, apply_affine_transform(get_current_coords(), build_affine_mtx()), width, height), 0, out.ptr); } diff --git a/src/core/CL/kernels/CLWarpAffineKernel.cpp b/src/core/CL/kernels/CLWarpAffineKernel.cpp index e549dbc258..be095f2c34 100644 --- a/src/core/CL/kernels/CLWarpAffineKernel.cpp +++ b/src/core/CL/kernels/CLWarpAffineKernel.cpp @@ -88,8 +88,11 @@ void CLWarpAffineKernel::configure(const ICLTensor *input, ICLTensor *output, co Window win = calculate_max_window(*output->info(), Steps(num_elems_processed_per_iteration)); - AccessWindowHorizontal input_access(input->info(), 0, num_elems_processed_per_iteration); - AccessWindowStatic output_access(output->info(), 0, 0, output->info()->dimension(0), output->info()->dimension(1)); + int total_right = ceil_to_multiple(input->info()->dimension(0), num_elems_processed_per_iteration); + const int access_right = total_right + (((total_right - input->info()->dimension(0)) == 0) ? border_size().right : 0); + + AccessWindowStatic input_access(input->info(), -border_size().left, -border_size().top, access_right, input->info()->dimension(1) + border_size().bottom); + AccessWindowHorizontal output_access(output->info(), 0, num_elems_processed_per_iteration); update_window_and_padding(win, input_access, output_access); |