Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2016, 2017 ARM Limited. |
| 3 | * |
| 4 | * SPDX-License-Identifier: MIT |
| 5 | * |
| 6 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
| 7 | * of this software and associated documentation files (the "Software"), to |
| 8 | * deal in the Software without restriction, including without limitation the |
| 9 | * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or |
| 10 | * sell copies of the Software, and to permit persons to whom the Software is |
| 11 | * furnished to do so, subject to the following conditions: |
| 12 | * |
| 13 | * The above copyright notice and this permission notice shall be included in all |
| 14 | * copies or substantial portions of the Software. |
| 15 | * |
| 16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 19 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
| 22 | * SOFTWARE. |
| 23 | */ |
| 24 | #include "helpers.h" |
| 25 | |
| 26 | /** Calculates L1 normalization between two inputs. |
| 27 | * |
| 28 | * @param[in] a First input. Supported data types: S16, S32 |
| 29 | * @param[in] b Second input. Supported data types: S16, S32 |
| 30 | * |
| 31 | * @return L1 normalization magnitude result. Supported data types: S16, S32 |
| 32 | */ |
| 33 | inline VEC_DATA_TYPE(DATA_TYPE, 16) magnitude_l1(VEC_DATA_TYPE(DATA_TYPE, 16) a, VEC_DATA_TYPE(DATA_TYPE, 16) b) |
| 34 | { |
| 35 | return CONVERT_SAT(add_sat(abs(a), abs(b)), VEC_DATA_TYPE(DATA_TYPE, 16)); |
| 36 | } |
| 37 | |
| 38 | /** Calculates L2 normalization between two inputs. |
| 39 | * |
| 40 | * @param[in] a First input. Supported data types: S16, S32 |
| 41 | * @param[in] b Second input. Supported data types: S16, S32 |
| 42 | * |
| 43 | * @return L2 normalization magnitude result. Supported data types: S16, S32 |
| 44 | */ |
| 45 | inline VEC_DATA_TYPE(DATA_TYPE, 16) magnitude_l2(int16 a, int16 b) |
| 46 | { |
| 47 | return CONVERT_SAT((sqrt(convert_float16((convert_uint16(a * a) + convert_uint16(b * b)))) + 0.5f), |
| 48 | VEC_DATA_TYPE(DATA_TYPE, 16)); |
| 49 | } |
| 50 | |
| 51 | /** Calculates unsigned phase between two inputs. |
| 52 | * |
| 53 | * @param[in] a First input. Supported data types: S16, S32 |
| 54 | * @param[in] b Second input. Supported data types: S16, S32 |
| 55 | * |
| 56 | * @return Unsigned phase mapped in the interval [0, 180]. Supported data types: U8 |
| 57 | */ |
| 58 | inline uchar16 phase_unsigned(VEC_DATA_TYPE(DATA_TYPE, 16) a, VEC_DATA_TYPE(DATA_TYPE, 16) b) |
| 59 | { |
| 60 | float16 angle_deg_f32 = atan2pi(convert_float16(b), convert_float16(a)) * (float16)180.0f; |
| 61 | angle_deg_f32 = select(angle_deg_f32, (float16)180.0f + angle_deg_f32, angle_deg_f32 < (float16)0.0f); |
| 62 | return convert_uchar16(angle_deg_f32); |
| 63 | } |
| 64 | |
| 65 | /** Calculates signed phase between two inputs. |
| 66 | * |
| 67 | * @param[in] a First input. Supported data types: S16, S32 |
| 68 | * @param[in] b Second input. Supported data types: S16, S32 |
| 69 | * |
| 70 | * @return Signed phase mapped in the interval [0, 256). Supported data types: U8 |
| 71 | */ |
| 72 | inline uchar16 phase_signed(VEC_DATA_TYPE(DATA_TYPE, 16) a, VEC_DATA_TYPE(DATA_TYPE, 16) b) |
| 73 | { |
| 74 | float16 arct = atan2pi(convert_float16(b), convert_float16(a)); |
| 75 | arct = select(arct, arct + 2, arct < 0.0f); |
| 76 | |
| 77 | return convert_uchar16(convert_int16(mad(arct, 128, 0.5f)) & 0xFFu); |
| 78 | } |
| 79 | |
| 80 | #if(1 == MAGNITUDE) |
| 81 | #define MAGNITUDE_OP(x, y) magnitude_l1((x), (y)) |
| 82 | #elif(2 == MAGNITUDE) |
| 83 | #define MAGNITUDE_OP(x, y) magnitude_l2(convert_int16(x), convert_int16(y)) |
Anthony Barbier | ac69aa1 | 2017-07-03 17:39:37 +0100 | [diff] [blame] | 84 | #else /* MAGNITUDE */ |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 85 | #define MAGNITUDE_OP(x, y) |
Anthony Barbier | ac69aa1 | 2017-07-03 17:39:37 +0100 | [diff] [blame] | 86 | #endif /* MAGNITUDE */ |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 87 | |
| 88 | #if(1 == PHASE) |
| 89 | #define PHASE_OP(x, y) phase_unsigned((x), (y)) |
| 90 | #elif(2 == PHASE) |
| 91 | #define PHASE_OP(x, y) phase_signed((x), (y)) |
Anthony Barbier | ac69aa1 | 2017-07-03 17:39:37 +0100 | [diff] [blame] | 92 | #else /* PHASE */ |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 93 | #define PHASE_OP(x, y) |
Anthony Barbier | ac69aa1 | 2017-07-03 17:39:37 +0100 | [diff] [blame] | 94 | #endif /* PHASE */ |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 95 | |
| 96 | /** Calculate the magnitude and phase of given the gradients of an image. |
| 97 | * |
| 98 | * @note Magnitude calculation supported: L1 normalization(type = 1) and L2 normalization(type = 2). |
| 99 | * @note Phase calculation supported: Unsigned(type = 1) [0,128] and Signed(type = 2) [0,256). |
| 100 | * |
| 101 | * @attention To enable phase calculation -DPHASE="phase_calculation_type_id" must be provided at compile time. eg -DPHASE=1 |
| 102 | * @attention To enable magnitude calculation -DMAGNITUDE="magnitude_calculation_type_id" must be provided at compile time. eg -DMAGNITUDE=1 |
| 103 | * @attention Datatype of the two inputs is passed at compile time using -DDATA_TYPE. e.g -DDATA_TYPE=short. Supported data_types are: short and int |
| 104 | * |
| 105 | * @param[in] gx_ptr Pointer to the first source image (gradient X). Supported data types: S16, S32 |
| 106 | * @param[in] gx_stride_x Stride of the source image in X dimension (in bytes) |
| 107 | * @param[in] gx_step_x gx_stride_x * number of elements along X processed per workitem(in bytes) |
| 108 | * @param[in] gx_stride_y Stride of the source image in Y dimension (in bytes) |
| 109 | * @param[in] gx_step_y gx_stride_y * number of elements along Y processed per workitem(in bytes) |
| 110 | * @param[in] gx_offset_first_element_in_bytes The offset of the first element in the source image |
| 111 | * @param[in] gy_ptr Pointer to the second source image (gradient Y) . Supported data types: S16, S32 |
| 112 | * @param[in] gy_stride_x Stride of the destination image in X dimension (in bytes) |
| 113 | * @param[in] gy_step_x gy_stride_x * number of elements along X processed per workitem(in bytes) |
| 114 | * @param[in] gy_stride_y Stride of the destination image in Y dimension (in bytes) |
| 115 | * @param[in] gy_step_y gy_stride_y * number of elements along Y processed per workitem(in bytes) |
| 116 | * @param[in] gy_offset_first_element_in_bytes The offset of the first element in the destination image |
| 117 | * @param[out] magnitude_ptr Pointer to the magnitude destination image. Supported data types: S16, S32 |
| 118 | * @param[in] magnitude_stride_x Stride of the source image in X dimension (in bytes) |
| 119 | * @param[in] magnitude_step_x magnitude_stride_x * number of elements along X processed per workitem(in bytes) |
| 120 | * @param[in] magnitude_stride_y Stride of the source image in Y dimension (in bytes) |
| 121 | * @param[in] magnitude_step_y magnitude_stride_y * number of elements along Y processed per workitem(in bytes) |
| 122 | * @param[in] magnitude_offset_first_element_in_bytes The offset of the first element in the source image |
| 123 | * @param[out] phase_ptr Pointer to the phase destination image. Supported data types: U8 |
| 124 | * @param[in] phase_stride_x Stride of the destination image in X dimension (in bytes) |
| 125 | * @param[in] phase_step_x phase_stride_x * number of elements along X processed per workitem(in bytes) |
| 126 | * @param[in] phase_stride_y Stride of the destination image in Y dimension (in bytes) |
| 127 | * @param[in] phase_step_y phase_stride_y * number of elements along Y processed per workitem(in bytes) |
| 128 | * @param[in] phase_offset_first_element_in_bytes The offset of the first element in the destination image |
| 129 | * */ |
| 130 | __kernel void magnitude_phase( |
| 131 | IMAGE_DECLARATION(gx), |
| 132 | IMAGE_DECLARATION(gy) |
| 133 | #ifdef MAGNITUDE |
| 134 | , |
| 135 | IMAGE_DECLARATION(magnitude) |
Anthony Barbier | ac69aa1 | 2017-07-03 17:39:37 +0100 | [diff] [blame] | 136 | #endif /* MAGNITUDE */ |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 137 | #ifdef PHASE |
| 138 | , |
| 139 | IMAGE_DECLARATION(phase) |
Anthony Barbier | ac69aa1 | 2017-07-03 17:39:37 +0100 | [diff] [blame] | 140 | #endif /* PHASE */ |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 141 | ) |
| 142 | { |
| 143 | // Get pixels pointer |
| 144 | Image gx = CONVERT_TO_IMAGE_STRUCT(gx); |
| 145 | Image gy = CONVERT_TO_IMAGE_STRUCT(gy); |
| 146 | |
| 147 | // Load values |
| 148 | VEC_DATA_TYPE(DATA_TYPE, 16) |
| 149 | in_a = vload16(0, (__global DATA_TYPE *)gx.ptr); |
| 150 | VEC_DATA_TYPE(DATA_TYPE, 16) |
| 151 | in_b = vload16(0, (__global DATA_TYPE *)gy.ptr); |
| 152 | |
| 153 | // Calculate and store the results |
| 154 | #ifdef MAGNITUDE |
| 155 | Image magnitude = CONVERT_TO_IMAGE_STRUCT(magnitude); |
| 156 | vstore16(MAGNITUDE_OP(in_a, in_b), 0, (__global DATA_TYPE *)magnitude.ptr); |
Anthony Barbier | ac69aa1 | 2017-07-03 17:39:37 +0100 | [diff] [blame] | 157 | #endif /* MAGNITUDE */ |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 158 | #ifdef PHASE |
| 159 | Image phase = CONVERT_TO_IMAGE_STRUCT(phase); |
| 160 | vstore16(PHASE_OP(in_a, in_b), 0, phase.ptr); |
Anthony Barbier | ac69aa1 | 2017-07-03 17:39:37 +0100 | [diff] [blame] | 161 | #endif /* PHASE */ |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 162 | } |