Michele Di Giorgio | 56dd726 | 2017-07-27 09:53:49 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 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 | /** This performs the quantization of floating point inputs to 8-bit unsigned integers. |
| 27 | * |
| 28 | * @param[in] input_ptr Pointer to the source image. Supported data types: F32 |
| 29 | * @param[in] input_stride_x Stride of the source image in X dimension (in bytes) |
| 30 | * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes) |
| 31 | * @param[in] input_stride_y Stride of the source image in Y dimension (in bytes) |
| 32 | * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes) |
| 33 | * @param[in] input_stride_z Stride of the source tensor in Z dimension (in bytes) |
| 34 | * @param[in] input_step_z input_stride_z * number of elements along Z processed per workitem(in bytes) |
| 35 | * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image |
| 36 | * @param[out] output_ptr Pointer to the destination image. Supported data types: U8 |
| 37 | * @param[in] output_stride_x Stride of the destination image in X dimension (in bytes) |
| 38 | * @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes) |
| 39 | * @param[in] output_stride_y Stride of the destination image in Y dimension (in bytes) |
| 40 | * @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes) |
| 41 | * @param[in] output_stride_z Stride of the source tensor in Z dimension (in bytes) |
| 42 | * @param[in] output_step_z output_stride_z * number of elements along Z processed per workitem(in bytes) |
| 43 | * @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination image |
| 44 | * @param[in] min_max_ptr Pointer to the min/max vector. Minimum value in position 0, maximum value in position 1. Supported data types: F32. |
| 45 | * @param[in] min_max_stride_x Stride of the min/max vector in X dimension (in bytes) |
| 46 | * @param[in] min_max_step_x min_max_stride_x * number of elements along X processed per workitem(in bytes) |
| 47 | * @param[in] min_max_offset_first_element_in_bytes The offset of the first element in the min/max vector |
| 48 | */ |
| 49 | __kernel void quantization_layer( |
| 50 | TENSOR3D_DECLARATION(input), |
| 51 | TENSOR3D_DECLARATION(output), |
| 52 | VECTOR_DECLARATION(min_max)) |
| 53 | { |
| 54 | // Get pixels pointer |
| 55 | Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT(input); |
| 56 | Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT(output); |
| 57 | |
| 58 | // min_max_value.s0 = min, min_max_value.s1 = max |
| 59 | const float2 min_max_value = vload2(0, (__global float *)(min_max_ptr + min_max_offset_first_element_in_bytes)); |
| 60 | |
| 61 | const float4 vmin = (float4)min_max_value.s0; |
| 62 | const float4 vrange = (float4)(min_max_value.s1 - min_max_value.s0); |
| 63 | |
| 64 | // Load data |
| 65 | float4 data = vload4(0, (__global float *)input.ptr); |
| 66 | |
| 67 | // Map float values to range [0.0, 1.0] |
| 68 | data = (data - vmin) / vrange; |
| 69 | |
| 70 | // Quantize and saturate |
| 71 | uchar4 res = convert_uchar4_sat(data * 256.0f); |
| 72 | |
| 73 | // Store result |
| 74 | vstore4(res, 0, (__global uchar *)output.ptr); |
| 75 | } |