clahe.cl 7.66 KB
Newer Older
wester committed
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188
/*M///////////////////////////////////////////////////////////////////////////////////////
//
//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
//  By downloading, copying, installing or using the software you agree to this license.
//  If you do not agree to this license, do not download, install,
//  copy or use the software.
//
//
//                           License Agreement
//                For Open Source Computer Vision Library
//
// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
//    Sen Liu, swjtuls1987@126.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
//   * Redistribution's of source code must retain the above copyright notice,
//     this list of conditions and the following disclaimer.
//
//   * Redistribution's in binary form must reproduce the above copyright notice,
//     this list of conditions and the following disclaimer in the documentation
//     and/or other materials provided with the distribution.
//
//   * The name of the copyright holders may not be used to endorse or promote products
//     derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/

#ifndef WAVE_SIZE
#define WAVE_SIZE 1
#endif

inline int calc_lut(__local int* smem, int val, int tid)
{
    smem[tid] = val;
    barrier(CLK_LOCAL_MEM_FENCE);

    if (tid == 0)
        for (int i = 1; i < 256; ++i)
            smem[i] += smem[i - 1];
    barrier(CLK_LOCAL_MEM_FENCE);

    return smem[tid];
}

#ifdef CPU
inline void reduce(volatile __local int* smem, int val, int tid)
{
    smem[tid] = val;
    barrier(CLK_LOCAL_MEM_FENCE);

    if (tid < 128)
        smem[tid] = val += smem[tid + 128];
    barrier(CLK_LOCAL_MEM_FENCE);

    if (tid < 64)
        smem[tid] = val += smem[tid + 64];
    barrier(CLK_LOCAL_MEM_FENCE);

    if (tid < 32)
        smem[tid] += smem[tid + 32];
    barrier(CLK_LOCAL_MEM_FENCE);

    if (tid < 16)
        smem[tid] += smem[tid + 16];
    barrier(CLK_LOCAL_MEM_FENCE);

    if (tid < 8)
        smem[tid] += smem[tid + 8];
    barrier(CLK_LOCAL_MEM_FENCE);

    if (tid < 4)
        smem[tid] += smem[tid + 4];
    barrier(CLK_LOCAL_MEM_FENCE);

    if (tid < 2)
        smem[tid] += smem[tid + 2];
    barrier(CLK_LOCAL_MEM_FENCE);

    if (tid < 1)
        smem[256] = smem[tid] + smem[tid + 1];
    barrier(CLK_LOCAL_MEM_FENCE);
}

#else

inline void reduce(__local volatile int* smem, int val, int tid)
{
    smem[tid] = val;
    barrier(CLK_LOCAL_MEM_FENCE);

    if (tid < 128)
        smem[tid] = val += smem[tid + 128];
    barrier(CLK_LOCAL_MEM_FENCE);

    if (tid < 64)
        smem[tid] = val += smem[tid + 64];
    barrier(CLK_LOCAL_MEM_FENCE);

    if (tid < 32)
    {
        smem[tid] += smem[tid + 32];
#if WAVE_SIZE < 32
    } barrier(CLK_LOCAL_MEM_FENCE);

    if (tid < 16)
    {
#endif
        smem[tid] += smem[tid + 16];
#if WAVE_SIZE < 16
    }
    barrier(CLK_LOCAL_MEM_FENCE);

    if (tid < 8)
    {
#endif
        smem[tid] += smem[tid + 8];
        smem[tid] += smem[tid + 4];
        smem[tid] += smem[tid + 2];
        smem[tid] += smem[tid + 1];
    }
}
#endif

__kernel void calcLut(__global __const uchar * src, const int srcStep,
                      const int src_offset, __global uchar * lut,
                      const int dstStep, const int dst_offset,
                      const int2 tileSize, const int tilesX,
                      const int clipLimit, const float lutScale)
{
    __local int smem[512];

    int tx = get_group_id(0);
    int ty = get_group_id(1);
    int tid = get_local_id(1) * get_local_size(0)
                             + get_local_id(0);
    smem[tid] = 0;
    barrier(CLK_LOCAL_MEM_FENCE);

    for (int i = get_local_id(1); i < tileSize.y; i += get_local_size(1))
    {
        __global const uchar* srcPtr = src + mad24(ty * tileSize.y + i, srcStep, tx * tileSize.x + src_offset);
        for (int j = get_local_id(0); j < tileSize.x; j += get_local_size(0))
        {
            const int data = srcPtr[j];
            atomic_inc(&smem[data]);
        }
    }
    barrier(CLK_LOCAL_MEM_FENCE);

    int tHistVal = smem[tid];
    barrier(CLK_LOCAL_MEM_FENCE);

    if (clipLimit > 0)
    {
        // clip histogram bar
        int clipped = 0;
        if (tHistVal > clipLimit)
        {
            clipped = tHistVal - clipLimit;
            tHistVal = clipLimit;
        }

        // find number of overall clipped samples
        reduce(smem, clipped, tid);
        barrier(CLK_LOCAL_MEM_FENCE);
#ifdef CPU
        clipped = smem[256];
#else
        clipped = smem[0];
#endif
a  
Kai Westerkamp committed
189 190 191 192 193 194 195

        // broadcast evaluated value

        __local int totalClipped;

        if (tid == 0)
            totalClipped = clipped;
wester committed
196 197 198
        barrier(CLK_LOCAL_MEM_FENCE);

        // redistribute clipped samples evenly
a  
Kai Westerkamp committed
199 200

        int redistBatch = totalClipped / 256;
wester committed
201 202
        tHistVal += redistBatch;

a  
Kai Westerkamp committed
203 204
        int residual = totalClipped - redistBatch * 256;
        if (tid < residual)
wester committed
205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252
            ++tHistVal;
    }

    const int lutVal = calc_lut(smem, tHistVal, tid);
    uint ires = (uint)convert_int_rte(lutScale * lutVal);
    lut[(ty * tilesX + tx) * dstStep + tid + dst_offset] =
        convert_uchar(clamp(ires, (uint)0, (uint)255));
}

__kernel void transform(__global __const uchar * src, const int srcStep, const int src_offset,
                        __global uchar * dst, const int dstStep, const int dst_offset,
                        __global uchar * lut, const int lutStep, int lut_offset,
                        const int cols, const int rows,
                        const int2 tileSize,
                        const int tilesX, const int tilesY)
{
    const int x = get_global_id(0);
    const int y = get_global_id(1);

    if (x >= cols || y >= rows)
        return;

    const float tyf = (convert_float(y) / tileSize.y) - 0.5f;
    int ty1 = convert_int_rtn(tyf);
    int ty2 = ty1 + 1;
    const float ya = tyf - ty1;
    ty1 = max(ty1, 0);
    ty2 = min(ty2, tilesY - 1);

    const float txf = (convert_float(x) / tileSize.x) - 0.5f;
    int tx1 = convert_int_rtn(txf);
    int tx2 = tx1 + 1;
    const float xa = txf - tx1;
    tx1 = max(tx1, 0);
    tx2 = min(tx2, tilesX - 1);

    const int srcVal = src[mad24(y, srcStep, x + src_offset)];

    float res = 0;

    res += lut[mad24(ty1 * tilesX + tx1, lutStep, srcVal + lut_offset)] * ((1.0f - xa) * (1.0f - ya));
    res += lut[mad24(ty1 * tilesX + tx2, lutStep, srcVal + lut_offset)] * ((xa) * (1.0f - ya));
    res += lut[mad24(ty2 * tilesX + tx1, lutStep, srcVal + lut_offset)] * ((1.0f - xa) * (ya));
    res += lut[mad24(ty2 * tilesX + tx2, lutStep, srcVal + lut_offset)] * ((xa) * (ya));

    uint ires = (uint)convert_int_rte(res);
    dst[mad24(y, dstStep, x + dst_offset)] = convert_uchar(clamp(ires, (uint)0, (uint)255));
}