From 57a150ac8a0238b29ba08ea0018396adfa3bc1b8 Mon Sep 17 00:00:00 2001 From: Frank Lei Date: Tue, 19 Dec 2017 10:14:57 +0800 Subject: APPBROWSER-341: Add ResizeNearest operator support Change-Id: I1eefcb89b2185514f6eba8f3d0e7b126b20f7d8f Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/113856 Reviewed-by: Joel Liang Tested-by: Jenkins Reviewed-by: Pablo Tello --- src/core/GLES_COMPUTE/cs_shaders/scale.cs | 130 ++++++++++++++++++++++++++++++ 1 file changed, 130 insertions(+) create mode 100644 src/core/GLES_COMPUTE/cs_shaders/scale.cs (limited to 'src/core/GLES_COMPUTE/cs_shaders/scale.cs') diff --git a/src/core/GLES_COMPUTE/cs_shaders/scale.cs b/src/core/GLES_COMPUTE/cs_shaders/scale.cs new file mode 100644 index 0000000000..08233abaa2 --- /dev/null +++ b/src/core/GLES_COMPUTE/cs_shaders/scale.cs @@ -0,0 +1,130 @@ +/* + * Copyright (c) 2016, 2017 ARM Limited. + * + * SPDX-License-Identifier: MIT + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to + * deal in the Software without restriction, including without limitation the + * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or + * sell copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in all + * copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +layout(local_size_x = LOCAL_SIZE_X, local_size_y = LOCAL_SIZE_Y, local_size_z = LOCAL_SIZE_Z) in; + +#include "helpers_cs.h" + +/** Performs an affine transformation on an image interpolating with the NEAREAST NEIGHBOUR method. Input and output are single channel FP16. + * + * @param[in] src_ptr Pointer to the source tensor. Supported data types: FP16. + * @param[in] src_attrs The attributes of the source tensor + * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: FP16. (Must be the same as the input) + * @param[in] dst_attrs The attributes of the destination tensor + * @param[in] input_width Input image width + * @param[in] input_height Input image height + * @param[in] scale_x The scale factor along x dimension + * @param[in] scale_y The scale factor along y dimension + */ +SHADER_PARAMS_DECLARATION +{ + ImageAttributes src_attrs; + ImageAttributes dst_attrs; + float input_width; + float input_height; + float scale_x; + float scale_y; +}; + +#if defined(DATA_TYPE_FP16) +#if defined(SCALE_NEAREST_GENERIC) +TENSOR_DECLARATION(1, srcBuffer, uint, src_ptr, src_shift, 2, readonly); +TENSOR_DECLARATION(2, dstBuffer, uvec2, dst_ptr, dst_shift, 3, writeonly); + +vec4[2] transform_nearest(vec2 coord, vec2 scale) +{ + vec4 in_x_coords = vec4(coord.x, 1.f + coord.x, 2.f + coord.x, 3.f + coord.x); + + vec4[2] t; + t[0] = (in_x_coords + (vec4(0.5f))) * scale.x; + t[1] = vec4((coord.y + 0.5f) * scale.y); + + return t; +} + +vec4[2] clamp_to_border_with_size(vec4[2] coords, float width, float height, float border_size) +{ + vec4[2] c; + c[0] = clamp(coords[0], 0.0f - border_size, width - 1.f + border_size); + c[1] = clamp(coords[1], 0.0f - border_size, height - 1.f + border_size); + + return c; +} + +void main() +{ + ImageIterator src_iter = CONVERT_TO_IMAGE_ITERATOR_NO_STEP(src_attrs, src_shift); + ImageIterator dst_iter = CONVERT_TO_IMAGE_ITERATOR(dst_attrs, dst_shift); + + vec2 r = vec2(scale_x, scale_y); + vec4[2] tc = clamp_to_border_with_size(transform_nearest(vec2(gl_GlobalInvocationID.x << uint(2), gl_GlobalInvocationID.y), r), input_width, input_height, float(BORDER_SIZE)); + + mediump vec2 s = vec2(0.0f); + mediump vec4 d = vec4(0.0f); + + for(int i = 0; i < 4; i++) + { + uint offset = image_offset_in_bytes(src_iter, int(tc[0][i]), int(tc[1][i])); + + s = LOAD_UNPACK2_HALF(src_ptr, uint(offset >> src_shift)); + + if(offset % uint(4) == uint(0)) + { + d[i] = s.x; + } + else + { + d[i] = s.y; + } + } + + STORE_PACK4_CURRENT_ITEM_HALF(dst_ptr, dst_iter, d); +} +#elif defined(SCALE_NEAREST_8X) /* SCALE_NEAREST_GENERIC */ +TENSOR_DECLARATION(1, srcBuffer, uvec2, src_ptr, src_shift, 3, readonly); +TENSOR_DECLARATION(2, dstBuffer, uvec4, dst_ptr, dst_shift, 4, writeonly); + +void main() +{ + ImageIterator src_iter = CONVERT_TO_IMAGE_ITERATOR_NO_STEP(src_attrs, src_shift); + ImageIterator dst_iter = CONVERT_TO_IMAGE_ITERATOR(dst_attrs, dst_shift); + + uvec2 tc = uvec2(gl_GlobalInvocationID.x << uint(2), gl_GlobalInvocationID.y >> uint(1)); + + mediump vec4 s = vec4(0.0f); + mediump vec4[2] d; + + uint offset = image_offset_in_bytes(src_iter, int(tc[0]), int(tc[1])); + s = LOAD_UNPACK4_HALF(src_ptr, uint(offset >> src_shift)); + + d[0] = vec4(s.x, s.x, s.y, s.y); + d[1] = vec4(s.z, s.z, s.w, s.w); + + STORE_PACK8_CURRENT_ITEM_HALF(dst_ptr, dst_iter, d); +} +#endif /* SCALE_NEAREST_GENERIC */ + +#else /* DATA_TYPE_FP16 */ +#error Data type not supported +#endif /* DATA_TYPE_FP16 */ -- cgit v1.2.1