| <a href="pooling__layer_8cl.xhtml">Go to the documentation of this file.</a><div class="fragment"><div class="line"><a name="l00001"></a><span class="lineno"> 1</span> <span class="comment">/*</span></div><div class="line"><a name="l00002"></a><span class="lineno"> 2</span> <span class="comment"> * Copyright (c) 2017-2020 ARM Limited.</span></div><div class="line"><a name="l00003"></a><span class="lineno"> 3</span> <span class="comment"> *</span></div><div class="line"><a name="l00004"></a><span class="lineno"> 4</span> <span class="comment"> * SPDX-License-Identifier: MIT</span></div><div class="line"><a name="l00005"></a><span class="lineno"> 5</span> <span class="comment"> *</span></div><div class="line"><a name="l00006"></a><span class="lineno"> 6</span> <span class="comment"> * Permission is hereby granted, free of charge, to any person obtaining a copy</span></div><div class="line"><a name="l00007"></a><span class="lineno"> 7</span> <span class="comment"> * of this software and associated documentation files (the "Software"), to</span></div><div class="line"><a name="l00008"></a><span class="lineno"> 8</span> <span class="comment"> * deal in the Software without restriction, including without limitation the</span></div><div class="line"><a name="l00009"></a><span class="lineno"> 9</span> <span class="comment"> * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or</span></div><div class="line"><a name="l00010"></a><span class="lineno"> 10</span> <span class="comment"> * sell copies of the Software, and to permit persons to whom the Software is</span></div><div class="line"><a name="l00011"></a><span class="lineno"> 11</span> <span class="comment"> * furnished to do so, subject to the following conditions:</span></div><div class="line"><a name="l00012"></a><span class="lineno"> 12</span> <span class="comment"> *</span></div><div class="line"><a name="l00013"></a><span class="lineno"> 13</span> <span class="comment"> * The above copyright notice and this permission notice shall be included in all</span></div><div class="line"><a name="l00014"></a><span class="lineno"> 14</span> <span class="comment"> * copies or substantial portions of the Software.</span></div><div class="line"><a name="l00015"></a><span class="lineno"> 15</span> <span class="comment"> *</span></div><div class="line"><a name="l00016"></a><span class="lineno"> 16</span> <span class="comment"> * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR</span></div><div class="line"><a name="l00017"></a><span class="lineno"> 17</span> <span class="comment"> * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,</span></div><div class="line"><a name="l00018"></a><span class="lineno"> 18</span> <span class="comment"> * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE</span></div><div class="line"><a name="l00019"></a><span class="lineno"> 19</span> <span class="comment"> * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER</span></div><div class="line"><a name="l00020"></a><span class="lineno"> 20</span> <span class="comment"> * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,</span></div><div class="line"><a name="l00021"></a><span class="lineno"> 21</span> <span class="comment"> * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE</span></div><div class="line"><a name="l00022"></a><span class="lineno"> 22</span> <span class="comment"> * SOFTWARE.</span></div><div class="line"><a name="l00023"></a><span class="lineno"> 23</span> <span class="comment"> */</span></div><div class="line"><a name="l00024"></a><span class="lineno"> 24</span> <span class="preprocessor">#include "<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml">helpers.h</a>"</span></div><div class="line"><a name="l00025"></a><span class="lineno"> 25</span> </div><div class="line"><a name="l00026"></a><span class="lineno"> 26</span> <span class="preprocessor">#if defined(POOL_AVG) || defined(POOL_L2)</span></div><div class="line"><a name="l00027"></a><span class="lineno"><a class="line" href="pooling__layer_8cl.xhtml#a482ef7d59a5f474ca126e737c7f0978a"> 27</a></span> <span class="preprocessor">#define POOL_OP(x, y) ((x) + (y))</span></div><div class="line"><a name="l00028"></a><span class="lineno"> 28</span> <span class="preprocessor">#else </span><span class="comment">/* defined(POOL_AVG) || defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00029"></a><span class="lineno"> 29</span> <span class="preprocessor">#define POOL_OP(x, y) (fmax((x), (y)))</span></div><div class="line"><a name="l00030"></a><span class="lineno"> 30</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_AVG) || defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00031"></a><span class="lineno"> 31</span> </div><div class="line"><a name="l00032"></a><span class="lineno"> 32</span> <span class="preprocessor">#if defined(POOL_L2)</span></div><div class="line"><a name="l00033"></a><span class="lineno"> 33</span> <span class="preprocessor">#define POW2_OP(x, vec_size) ((x) * (x))</span></div><div class="line"><a name="l00034"></a><span class="lineno"> 34</span> <span class="preprocessor">#else </span><span class="comment">/* defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00035"></a><span class="lineno"><a class="line" href="pooling__layer_8cl.xhtml#a03898439d164d74f8c35bafb67262d95"> 35</a></span> <span class="preprocessor">#define POW2_OP(x, vec_size) (x)</span></div><div class="line"><a name="l00036"></a><span class="lineno"> 36</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00037"></a><span class="lineno"> 37</span> </div><div class="line"><a name="l00038"></a><span class="lineno"><a class="line" href="pooling__layer_8cl.xhtml#a5db17889d824975fefb2ce2f4690637f"> 38</a></span> <span class="preprocessor">#define DIV_OP(x, y) (x * (1.f / y))</span></div><div class="line"><a name="l00039"></a><span class="lineno"><a class="line" href="pooling__layer_8cl.xhtml#ac9af19bec38fe50b4b9585c0e5c0ccca"> 39</a></span> <span class="preprocessor">#define SQRT_OP(x) sqrt((x))</span></div><div class="line"><a name="l00040"></a><span class="lineno"> 40</span> </div><div class="line"><a name="l00041"></a><span class="lineno"><a class="line" href="pooling__layer_8cl.xhtml#a6c01fa98d360a9d52926dc6a5a599711"> 41</a></span> <span class="preprocessor">#define DIV_OP_NHWC(x, y) (x * (VEC_DATA_TYPE(ACC_DATA_TYPE, 8))(1.f / y))</span></div><div class="line"><a name="l00042"></a><span class="lineno"> 42</span> </div><div class="line"><a name="l00043"></a><span class="lineno"> 43</span> <span class="preprocessor">#if STRIDE_X == 1</span></div><div class="line"><a name="l00044"></a><span class="lineno"> 44</span> <span class="preprocessor">#define POOLING3x3(res, input, output) POOLING3x3_STRIDE1(res, input, output)</span></div><div class="line"><a name="l00045"></a><span class="lineno"> 45</span> <span class="preprocessor">#elif STRIDE_X == 2 </span><span class="comment">/* STRIDE_X == 1 */</span><span class="preprocessor"></span></div><div class="line"><a name="l00046"></a><span class="lineno"> 46</span> <span class="preprocessor">#define POOLING3x3(res, input, output) POOLING3x3_STRIDE2(res, input, output)</span></div><div class="line"><a name="l00047"></a><span class="lineno"> 47</span> <span class="preprocessor">#elif STRIDE_X == 3 </span><span class="comment">/* STRIDE_X not equals 1 or 2 */</span><span class="preprocessor"></span></div><div class="line"><a name="l00048"></a><span class="lineno"> 48</span> <span class="preprocessor">#define POOLING3x3(res, input, output) POOLING3x3_STRIDE3(res, input, output)</span></div><div class="line"><a name="l00049"></a><span class="lineno"> 49</span> <span class="preprocessor">#endif </span><span class="comment">/* STRIDE_X == 3 */</span><span class="preprocessor"></span></div><div class="line"><a name="l00050"></a><span class="lineno"> 50</span> </div><div class="line"><a name="l00051"></a><span class="lineno"> 51</span> <span class="preprocessor">#if defined(FP_MIXED_PRECISION)</span></div><div class="line"><a name="l00052"></a><span class="lineno"> 52</span> <span class="preprocessor">#define CONVERT_TO_ACC_DATA_TYPE(x, n) CONVERT(x, VEC_DATA_TYPE(ACC_DATA_TYPE, n))</span></div><div class="line"><a name="l00053"></a><span class="lineno"> 53</span> <span class="preprocessor">#define VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(n, offset, ptr) \</span></div><div class="line"><a name="l00054"></a><span class="lineno"> 54</span> <span class="preprocessor"> CONVERT_TO_ACC_DATA_TYPE(vload##n(offset, ptr), n)</span></div><div class="line"><a name="l00055"></a><span class="lineno"> 55</span> <span class="preprocessor">#else </span><span class="comment">/* defined(FP_MIXED_PRECISION) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00056"></a><span class="lineno"><a class="line" href="pooling__layer_8cl.xhtml#af6c2e106d0b5d1cb755bbc79d0b09d52"> 56</a></span> <span class="preprocessor">#define VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(n, offset, ptr) vload##n(offset, ptr)</span></div><div class="line"><a name="l00057"></a><span class="lineno"> 57</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(FP_MIXED_PRECISION) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00058"></a><span class="lineno"> 58</span> </div><div class="line"><a name="l00059"></a><span class="lineno"><a class="line" href="pooling__layer_8cl.xhtml#a6b8e66069b8cd3e743141d7f024a7d76"> 59</a></span> <span class="preprocessor">#define POOLING3x3_STRIDE1(res, input, output) \</span></div><div class="line"><a name="l00060"></a><span class="lineno"> 60</span> <span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00061"></a><span class="lineno"> 61</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 4) \</span></div><div class="line"><a name="l00062"></a><span class="lineno"> 62</span> <span class="preprocessor"> data00 = VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(4, 0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 0, 0)); \</span></div><div class="line"><a name="l00063"></a><span class="lineno"> 63</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 2) \</span></div><div class="line"><a name="l00064"></a><span class="lineno"> 64</span> <span class="preprocessor"> data01 = VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(2, 0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 0, 0) + 4); \</span></div><div class="line"><a name="l00065"></a><span class="lineno"> 65</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 4) \</span></div><div class="line"><a name="l00066"></a><span class="lineno"> 66</span> <span class="preprocessor"> data10 = VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(4, 0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 1, 0)); \</span></div><div class="line"><a name="l00067"></a><span class="lineno"> 67</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 2) \</span></div><div class="line"><a name="l00068"></a><span class="lineno"> 68</span> <span class="preprocessor"> data11 = VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(2, 0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 1, 0) + 4); \</span></div><div class="line"><a name="l00069"></a><span class="lineno"> 69</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 4) \</span></div><div class="line"><a name="l00070"></a><span class="lineno"> 70</span> <span class="preprocessor"> data20 = VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(4, 0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 2, 0)); \</span></div><div class="line"><a name="l00071"></a><span class="lineno"> 71</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 2) \</span></div><div class="line"><a name="l00072"></a><span class="lineno"> 72</span> <span class="preprocessor"> data21 = VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(2, 0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 2, 0) + 4); \</span></div><div class="line"><a name="l00073"></a><span class="lineno"> 73</span> <span class="preprocessor"> data00 = POW2_OP(data00, 4); \</span></div><div class="line"><a name="l00074"></a><span class="lineno"> 74</span> <span class="preprocessor"> data01 = POW2_OP(data01, 2); \</span></div><div class="line"><a name="l00075"></a><span class="lineno"> 75</span> <span class="preprocessor"> data10 = POW2_OP(data10, 4); \</span></div><div class="line"><a name="l00076"></a><span class="lineno"> 76</span> <span class="preprocessor"> data11 = POW2_OP(data11, 2); \</span></div><div class="line"><a name="l00077"></a><span class="lineno"> 77</span> <span class="preprocessor"> data20 = POW2_OP(data20, 4); \</span></div><div class="line"><a name="l00078"></a><span class="lineno"> 78</span> <span class="preprocessor"> data21 = POW2_OP(data21, 2); \</span></div><div class="line"><a name="l00079"></a><span class="lineno"> 79</span> <span class="preprocessor"> \</span></div><div class="line"><a name="l00080"></a><span class="lineno"> 80</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 8) \</span></div><div class="line"><a name="l00081"></a><span class="lineno"> 81</span> <span class="preprocessor"> values00 = (VEC_DATA_TYPE(ACC_DATA_TYPE, 8))(data00.s01212323); \</span></div><div class="line"><a name="l00082"></a><span class="lineno"> 82</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 4) \</span></div><div class="line"><a name="l00083"></a><span class="lineno"> 83</span> <span class="preprocessor"> values01 = (VEC_DATA_TYPE(ACC_DATA_TYPE, 4))(data01.s0, data00.s3, data01.s01); \</span></div><div class="line"><a name="l00084"></a><span class="lineno"> 84</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 8) \</span></div><div class="line"><a name="l00085"></a><span class="lineno"> 85</span> <span class="preprocessor"> values10 = (VEC_DATA_TYPE(ACC_DATA_TYPE, 8))(data10.s01212323); \</span></div><div class="line"><a name="l00086"></a><span class="lineno"> 86</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 4) \</span></div><div class="line"><a name="l00087"></a><span class="lineno"> 87</span> <span class="preprocessor"> values11 = (VEC_DATA_TYPE(ACC_DATA_TYPE, 4))(data11.s0, data10.s3, data11.s01); \</span></div><div class="line"><a name="l00088"></a><span class="lineno"> 88</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 8) \</span></div><div class="line"><a name="l00089"></a><span class="lineno"> 89</span> <span class="preprocessor"> values20 = (VEC_DATA_TYPE(ACC_DATA_TYPE, 8))(data20.s01212323); \</span></div><div class="line"><a name="l00090"></a><span class="lineno"> 90</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 4) \</span></div><div class="line"><a name="l00091"></a><span class="lineno"> 91</span> <span class="preprocessor"> values21 = (VEC_DATA_TYPE(ACC_DATA_TYPE, 4))(data21.s0, data20.s3, data21.s01); \</span></div><div class="line"><a name="l00092"></a><span class="lineno"> 92</span> <span class="preprocessor"> \</span></div><div class="line"><a name="l00093"></a><span class="lineno"> 93</span> <span class="preprocessor"> values00 = POOL_OP(values00, values10); \</span></div><div class="line"><a name="l00094"></a><span class="lineno"> 94</span> <span class="preprocessor"> values01 = POOL_OP(values01, values11); \</span></div><div class="line"><a name="l00095"></a><span class="lineno"> 95</span> <span class="preprocessor"> values00 = POOL_OP(values00, values20); \</span></div><div class="line"><a name="l00096"></a><span class="lineno"> 96</span> <span class="preprocessor"> values01 = POOL_OP(values01, values21); \</span></div><div class="line"><a name="l00097"></a><span class="lineno"> 97</span> <span class="preprocessor"> \</span></div><div class="line"><a name="l00098"></a><span class="lineno"> 98</span> <span class="preprocessor"> res = POOL_OP((VEC_DATA_TYPE(ACC_DATA_TYPE, 4))(values00.s036, values01.s1), (VEC_DATA_TYPE(ACC_DATA_TYPE, 4))(values00.s147, values01.s2)); \</span></div><div class="line"><a name="l00099"></a><span class="lineno"> 99</span> <span class="preprocessor"> res = POOL_OP(res, (VEC_DATA_TYPE(ACC_DATA_TYPE, 4))(values00.s25, values01.s03)); \</span></div><div class="line"><a name="l00100"></a><span class="lineno"> 100</span> <span class="preprocessor"> })</span></div><div class="line"><a name="l00101"></a><span class="lineno"> 101</span> </div><div class="line"><a name="l00102"></a><span class="lineno"><a class="line" href="pooling__layer_8cl.xhtml#abb8f7128361a6a1965b1b2a5b3a719b2"> 102</a></span> <span class="preprocessor">#define POOLING3x3_STRIDE2(res, input, output) \</span></div><div class="line"><a name="l00103"></a><span class="lineno"> 103</span> <span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00104"></a><span class="lineno"> 104</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 8) \</span></div><div class="line"><a name="l00105"></a><span class="lineno"> 105</span> <span class="preprocessor"> data00 = VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(8, 0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 0, 0)); \</span></div><div class="line"><a name="l00106"></a><span class="lineno"> 106</span> <span class="preprocessor"> ACC_DATA_TYPE data01 = (ACC_DATA_TYPE)(*((__global DATA_TYPE *)tensor3D_offset(&input, 0, 0, 0) + 8)); \</span></div><div class="line"><a name="l00107"></a><span class="lineno"> 107</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 8) \</span></div><div class="line"><a name="l00108"></a><span class="lineno"> 108</span> <span class="preprocessor"> data10 = VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(8, 0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 1, 0)); \</span></div><div class="line"><a name="l00109"></a><span class="lineno"> 109</span> <span class="preprocessor"> ACC_DATA_TYPE data11 = (ACC_DATA_TYPE)(*((__global DATA_TYPE *)tensor3D_offset(&input, 0, 1, 0) + 8)); \</span></div><div class="line"><a name="l00110"></a><span class="lineno"> 110</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 8) \</span></div><div class="line"><a name="l00111"></a><span class="lineno"> 111</span> <span class="preprocessor"> data20 = VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(8, 0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 2, 0)); \</span></div><div class="line"><a name="l00112"></a><span class="lineno"> 112</span> <span class="preprocessor"> ACC_DATA_TYPE data21 = (ACC_DATA_TYPE)(*((__global DATA_TYPE *)tensor3D_offset(&input, 0, 2, 0) + 8)); \</span></div><div class="line"><a name="l00113"></a><span class="lineno"> 113</span> <span class="preprocessor"> data00 = POW2_OP(data00, 8); \</span></div><div class="line"><a name="l00114"></a><span class="lineno"> 114</span> <span class="preprocessor"> data01 = POW2_OP(data01, 1); \</span></div><div class="line"><a name="l00115"></a><span class="lineno"> 115</span> <span class="preprocessor"> data10 = POW2_OP(data10, 8); \</span></div><div class="line"><a name="l00116"></a><span class="lineno"> 116</span> <span class="preprocessor"> data11 = POW2_OP(data11, 1); \</span></div><div class="line"><a name="l00117"></a><span class="lineno"> 117</span> <span class="preprocessor"> data20 = POW2_OP(data20, 8); \</span></div><div class="line"><a name="l00118"></a><span class="lineno"> 118</span> <span class="preprocessor"> data21 = POW2_OP(data21, 1); \</span></div><div class="line"><a name="l00119"></a><span class="lineno"> 119</span> <span class="preprocessor"> \</span></div><div class="line"><a name="l00120"></a><span class="lineno"> 120</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 8) \</span></div><div class="line"><a name="l00121"></a><span class="lineno"> 121</span> <span class="preprocessor"> values00 = (VEC_DATA_TYPE(ACC_DATA_TYPE, 8))(data00.s01223445); \</span></div><div class="line"><a name="l00122"></a><span class="lineno"> 122</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 4) \</span></div><div class="line"><a name="l00123"></a><span class="lineno"> 123</span> <span class="preprocessor"> values01 = (VEC_DATA_TYPE(ACC_DATA_TYPE, 4))(data00.s667, data01); \</span></div><div class="line"><a name="l00124"></a><span class="lineno"> 124</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 8) \</span></div><div class="line"><a name="l00125"></a><span class="lineno"> 125</span> <span class="preprocessor"> values10 = (VEC_DATA_TYPE(ACC_DATA_TYPE, 8))(data10.s01223445); \</span></div><div class="line"><a name="l00126"></a><span class="lineno"> 126</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 4) \</span></div><div class="line"><a name="l00127"></a><span class="lineno"> 127</span> <span class="preprocessor"> values11 = (VEC_DATA_TYPE(ACC_DATA_TYPE, 4))(data10.s667, data11); \</span></div><div class="line"><a name="l00128"></a><span class="lineno"> 128</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 8) \</span></div><div class="line"><a name="l00129"></a><span class="lineno"> 129</span> <span class="preprocessor"> values20 = (VEC_DATA_TYPE(ACC_DATA_TYPE, 8))(data20.s01223445); \</span></div><div class="line"><a name="l00130"></a><span class="lineno"> 130</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 4) \</span></div><div class="line"><a name="l00131"></a><span class="lineno"> 131</span> <span class="preprocessor"> values21 = (VEC_DATA_TYPE(ACC_DATA_TYPE, 4))(data20.s667, data21); \</span></div><div class="line"><a name="l00132"></a><span class="lineno"> 132</span> <span class="preprocessor"> \</span></div><div class="line"><a name="l00133"></a><span class="lineno"> 133</span> <span class="preprocessor"> values00 = POOL_OP(values00, values10); \</span></div><div class="line"><a name="l00134"></a><span class="lineno"> 134</span> <span class="preprocessor"> values01 = POOL_OP(values01, values11); \</span></div><div class="line"><a name="l00135"></a><span class="lineno"> 135</span> <span class="preprocessor"> values00 = POOL_OP(values00, values20); \</span></div><div class="line"><a name="l00136"></a><span class="lineno"> 136</span> <span class="preprocessor"> values01 = POOL_OP(values01, values21); \</span></div><div class="line"><a name="l00137"></a><span class="lineno"> 137</span> <span class="preprocessor"> \</span></div><div class="line"><a name="l00138"></a><span class="lineno"> 138</span> <span class="preprocessor"> res = POOL_OP((VEC_DATA_TYPE(ACC_DATA_TYPE, 4))(values00.s036, values01.s1), (VEC_DATA_TYPE(ACC_DATA_TYPE, 4))(values00.s147, values01.s2)); \</span></div><div class="line"><a name="l00139"></a><span class="lineno"> 139</span> <span class="preprocessor"> res = POOL_OP(res, (VEC_DATA_TYPE(ACC_DATA_TYPE, 4))(values00.s25, values01.s03)); \</span></div><div class="line"><a name="l00140"></a><span class="lineno"> 140</span> <span class="preprocessor"> })</span></div><div class="line"><a name="l00141"></a><span class="lineno"> 141</span> </div><div class="line"><a name="l00142"></a><span class="lineno"><a class="line" href="pooling__layer_8cl.xhtml#aba30215e4df370ff8935f83046e696ea"> 142</a></span> <span class="preprocessor">#define POOLING3x3_STRIDE3(res, input, output) \</span></div><div class="line"><a name="l00143"></a><span class="lineno"> 143</span> <span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00144"></a><span class="lineno"> 144</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 8) \</span></div><div class="line"><a name="l00145"></a><span class="lineno"> 145</span> <span class="preprocessor"> data00 = VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(8, 0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 0, 0)); \</span></div><div class="line"><a name="l00146"></a><span class="lineno"> 146</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 4) \</span></div><div class="line"><a name="l00147"></a><span class="lineno"> 147</span> <span class="preprocessor"> data01 = VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(4, 0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 0, 0) + 8); \</span></div><div class="line"><a name="l00148"></a><span class="lineno"> 148</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 8) \</span></div><div class="line"><a name="l00149"></a><span class="lineno"> 149</span> <span class="preprocessor"> data10 = VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(8, 0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 1, 0)); \</span></div><div class="line"><a name="l00150"></a><span class="lineno"> 150</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 4) \</span></div><div class="line"><a name="l00151"></a><span class="lineno"> 151</span> <span class="preprocessor"> data11 = VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(4, 0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 1, 0) + 8); \</span></div><div class="line"><a name="l00152"></a><span class="lineno"> 152</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 8) \</span></div><div class="line"><a name="l00153"></a><span class="lineno"> 153</span> <span class="preprocessor"> data20 = VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(8, 0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 2, 0)); \</span></div><div class="line"><a name="l00154"></a><span class="lineno"> 154</span> <span class="preprocessor"> VEC_DATA_TYPE(ACC_DATA_TYPE, 4) \</span></div><div class="line"><a name="l00155"></a><span class="lineno"> 155</span> <span class="preprocessor"> data21 = VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE(4, 0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 2, 0) + 8); \</span></div><div class="line"><a name="l00156"></a><span class="lineno"> 156</span> <span class="preprocessor"> data00 = POW2_OP(data00, 8); \</span></div><div class="line"><a name="l00157"></a><span class="lineno"> 157</span> <span class="preprocessor"> data01 = POW2_OP(data01, 4); \</span></div><div class="line"><a name="l00158"></a><span class="lineno"> 158</span> <span class="preprocessor"> data10 = POW2_OP(data10, 8); \</span></div><div class="line"><a name="l00159"></a><span class="lineno"> 159</span> <span class="preprocessor"> data11 = POW2_OP(data11, 4); \</span></div><div class="line"><a name="l00160"></a><span class="lineno"> 160</span> <span class="preprocessor"> data20 = POW2_OP(data20, 8); \</span></div><div class="line"><a name="l00161"></a><span class="lineno"> 161</span> <span class="preprocessor"> data21 = POW2_OP(data21, 4); \</span></div><div class="line"><a name="l00162"></a><span class="lineno"> 162</span> <span class="preprocessor"> \</span></div><div class="line"><a name="l00163"></a><span class="lineno"> 163</span> <span class="preprocessor"> data00 = POOL_OP(data00, data10); \</span></div><div class="line"><a name="l00164"></a><span class="lineno"> 164</span> <span class="preprocessor"> data01 = POOL_OP(data01, data11); \</span></div><div class="line"><a name="l00165"></a><span class="lineno"> 165</span> <span class="preprocessor"> data00 = POOL_OP(data00, data20); \</span></div><div class="line"><a name="l00166"></a><span class="lineno"> 166</span> <span class="preprocessor"> data01 = POOL_OP(data01, data21); \</span></div><div class="line"><a name="l00167"></a><span class="lineno"> 167</span> <span class="preprocessor"> \</span></div><div class="line"><a name="l00168"></a><span class="lineno"> 168</span> <span class="preprocessor"> res = POOL_OP((VEC_DATA_TYPE(ACC_DATA_TYPE, 4))(data00.s036, data01.s1), (VEC_DATA_TYPE(ACC_DATA_TYPE, 4))(data00.s147, data01.s2)); \</span></div><div class="line"><a name="l00169"></a><span class="lineno"> 169</span> <span class="preprocessor"> res = POOL_OP(res, (VEC_DATA_TYPE(ACC_DATA_TYPE, 4))(data00.s25, data01.s03)); \</span></div><div class="line"><a name="l00170"></a><span class="lineno"> 170</span> <span class="preprocessor"> })</span></div><div class="line"><a name="l00171"></a><span class="lineno"> 171</span> </div><div class="line"><a name="l00172"></a><span class="lineno"><a class="line" href="pooling__layer_8cl.xhtml#ae9df7602479b001f6280f2f528f10a92"> 172</a></span> ACC_DATA_TYPE <a class="code" href="pooling__layer_8cl.xhtml#ae9df7602479b001f6280f2f528f10a92">calculate_avg_scale</a>(<span class="keyword">const</span> <span class="keywordtype">int</span> pool_size_x, <span class="keyword">const</span> <span class="keywordtype">int</span> pool_size_y, <span class="keyword">const</span> <span class="keywordtype">int</span> upper_bound_w, <span class="keyword">const</span> <span class="keywordtype">int</span> upper_bound_h,</div><div class="line"><a name="l00173"></a><span class="lineno"> 173</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> pad_x, <span class="keyword">const</span> <span class="keywordtype">int</span> pad_y, <span class="keyword">const</span> <span class="keywordtype">int</span> stride_x, <span class="keyword">const</span> <span class="keywordtype">int</span> stride_y)</div><div class="line"><a name="l00174"></a><span class="lineno"> 174</span> {</div><div class="line"><a name="l00175"></a><span class="lineno"> 175</span>  <span class="keywordtype">int</span> start_x = get_global_id(0) * stride_x - pad_x;</div><div class="line"><a name="l00176"></a><span class="lineno"> 176</span>  <span class="keywordtype">int</span> start_y = get_global_id(1) * stride_y - pad_y;</div><div class="line"><a name="l00177"></a><span class="lineno"> 177</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> end_x = min(start_x + pool_size_x, upper_bound_w);</div><div class="line"><a name="l00178"></a><span class="lineno"> 178</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> end_y = min(start_y + pool_size_y, upper_bound_h);</div><div class="line"><a name="l00179"></a><span class="lineno"> 179</span> <span class="preprocessor">#if defined(EXCLUDE_PADDING)</span></div><div class="line"><a name="l00180"></a><span class="lineno"> 180</span>  start_x = max(0, start_x);</div><div class="line"><a name="l00181"></a><span class="lineno"> 181</span>  start_y = max(0, start_y);</div><div class="line"><a name="l00182"></a><span class="lineno"> 182</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(EXCLUDE_PADDING) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00183"></a><span class="lineno"> 183</span>  <span class="keywordflow">return</span> ((end_y - start_y) * (end_x - start_x));</div><div class="line"><a name="l00184"></a><span class="lineno"> 184</span> }</div><div class="line"><a name="l00185"></a><span class="lineno"> 185</span> <span class="comment"></span></div><div class="line"><a name="l00186"></a><span class="lineno"> 186</span> <span class="comment">/** Performs a pooling function of pool size equal to 2.</span></div><div class="line"><a name="l00187"></a><span class="lineno"> 187</span> <span class="comment"> *</span></div><div class="line"><a name="l00188"></a><span class="lineno"> 188</span> <span class="comment"> * @note Datatype must be passed using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types are F16/F32;</span></div><div class="line"><a name="l00189"></a><span class="lineno"> 189</span> <span class="comment"> * @note In case of average pooling the following information must be passed at compile time:</span></div><div class="line"><a name="l00190"></a><span class="lineno"> 190</span> <span class="comment"> * -DPOOL_AVG or -DPOOL_L2 must be provided otherwise max pooling will be performed.</span></div><div class="line"><a name="l00191"></a><span class="lineno"> 191</span> <span class="comment"> * -DMAX_WIDTH and -DMAX_HEIGHT which are the maximum accessible indeces in x and y dimensions (width + pad)</span></div><div class="line"><a name="l00192"></a><span class="lineno"> 192</span> <span class="comment"> * -DSTRIDE_X and -DSTRIDE_Y which are the steps of the window along the x and y directions</span></div><div class="line"><a name="l00193"></a><span class="lineno"> 193</span> <span class="comment"> * -DPAD_X and -DPAD_Y which are the pooling paddings in x and y dimension</span></div><div class="line"><a name="l00194"></a><span class="lineno"> 194</span> <span class="comment"> *</span></div><div class="line"><a name="l00195"></a><span class="lineno"> 195</span> <span class="comment"> * @param[in] input_ptr Pointer to the source image. Supported data types: F16/F32</span></div><div class="line"><a name="l00196"></a><span class="lineno"> 196</span> <span class="comment"> * @param[in] input_stride_x Stride of the source image in X dimension (in bytes)</span></div><div class="line"><a name="l00197"></a><span class="lineno"> 197</span> <span class="comment"> * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00198"></a><span class="lineno"> 198</span> <span class="comment"> * @param[in] input_stride_y Stride of the source image in Y dimension (in bytes)</span></div><div class="line"><a name="l00199"></a><span class="lineno"> 199</span> <span class="comment"> * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00200"></a><span class="lineno"> 200</span> <span class="comment"> * @param[in] input_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00201"></a><span class="lineno"> 201</span> <span class="comment"> * @param[in] input_step_z input_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00202"></a><span class="lineno"> 202</span> <span class="comment"> * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image</span></div><div class="line"><a name="l00203"></a><span class="lineno"> 203</span> <span class="comment"> * @param[out] output_ptr Pointer to the destination image. Supported data types: same as @p input_ptr</span></div><div class="line"><a name="l00204"></a><span class="lineno"> 204</span> <span class="comment"> * @param[in] output_stride_x Stride of the destination image in X dimension (in bytes)</span></div><div class="line"><a name="l00205"></a><span class="lineno"> 205</span> <span class="comment"> * @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00206"></a><span class="lineno"> 206</span> <span class="comment"> * @param[in] output_stride_y Stride of the destination image in Y dimension (in bytes)</span></div><div class="line"><a name="l00207"></a><span class="lineno"> 207</span> <span class="comment"> * @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00208"></a><span class="lineno"> 208</span> <span class="comment"> * @param[in] output_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00209"></a><span class="lineno"> 209</span> <span class="comment"> * @param[in] output_step_z output_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00210"></a><span class="lineno"> 210</span> <span class="comment"> * @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination image</span></div><div class="line"><a name="l00211"></a><span class="lineno"> 211</span> <span class="comment"> */</span></div><div class="line"><a name="l00212"></a><span class="lineno"><a class="line" href="pooling__layer_8cl.xhtml#a2d95de36199fd06803ffb62f5ff1df08"> 212</a></span> __kernel <span class="keywordtype">void</span> <a class="code" href="pooling__layer_8cl.xhtml#a2d95de36199fd06803ffb62f5ff1df08">pooling_layer_2</a>(</div><div class="line"><a name="l00213"></a><span class="lineno"> 213</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>),</div><div class="line"><a name="l00214"></a><span class="lineno"> 214</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(output))</div><div class="line"><a name="l00215"></a><span class="lineno"> 215</span> {</div><div class="line"><a name="l00216"></a><span class="lineno"> 216</span>  <span class="comment">// Get pixels pointer</span></div><div class="line"><a name="l00217"></a><span class="lineno"> 217</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>);</div><div class="line"><a name="l00218"></a><span class="lineno"> 218</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> output = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(output);</div><div class="line"><a name="l00219"></a><span class="lineno"> 219</span> </div><div class="line"><a name="l00220"></a><span class="lineno"> 220</span>  <span class="comment">// Load data</span></div><div class="line"><a name="l00221"></a><span class="lineno"> 221</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 2)</div><div class="line"><a name="l00222"></a><span class="lineno"> 222</span>  data0 = <a class="code" href="pooling__layer_8cl.xhtml#af6c2e106d0b5d1cb755bbc79d0b09d52">VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE</a>(2, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a2101b2fe0193ce227ae4e0945e321d85">tensor3D_offset</a>(&<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>, 0, 0, 0));</div><div class="line"><a name="l00223"></a><span class="lineno"> 223</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 2)</div><div class="line"><a name="l00224"></a><span class="lineno"> 224</span>  data1 = <a class="code" href="pooling__layer_8cl.xhtml#af6c2e106d0b5d1cb755bbc79d0b09d52">VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE</a>(2, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a2101b2fe0193ce227ae4e0945e321d85">tensor3D_offset</a>(&<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>, 0, 1, 0));</div><div class="line"><a name="l00225"></a><span class="lineno"> 225</span> </div><div class="line"><a name="l00226"></a><span class="lineno"> 226</span> <span class="preprocessor">#if defined(POOL_L2)</span></div><div class="line"><a name="l00227"></a><span class="lineno"> 227</span>  <span class="comment">// Raise to power of 2 for L2 Pooling</span></div><div class="line"><a name="l00228"></a><span class="lineno"> 228</span>  data0 = <a class="code" href="pooling__layer_8cl.xhtml#a03898439d164d74f8c35bafb67262d95">POW2_OP</a>(data0, 2);</div><div class="line"><a name="l00229"></a><span class="lineno"> 229</span>  data1 = <a class="code" href="pooling__layer_8cl.xhtml#a03898439d164d74f8c35bafb67262d95">POW2_OP</a>(data1, 2);</div><div class="line"><a name="l00230"></a><span class="lineno"> 230</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00231"></a><span class="lineno"> 231</span> </div><div class="line"><a name="l00232"></a><span class="lineno"> 232</span>  <span class="comment">// Perform calculations</span></div><div class="line"><a name="l00233"></a><span class="lineno"> 233</span>  data0 = <a class="code" href="pooling__layer_8cl.xhtml#a482ef7d59a5f474ca126e737c7f0978a">POOL_OP</a>(data0, data1);</div><div class="line"><a name="l00234"></a><span class="lineno"> 234</span>  ACC_DATA_TYPE res = <a class="code" href="pooling__layer_8cl.xhtml#a482ef7d59a5f474ca126e737c7f0978a">POOL_OP</a>(data0.s0, data0.s1);</div><div class="line"><a name="l00235"></a><span class="lineno"> 235</span> </div><div class="line"><a name="l00236"></a><span class="lineno"> 236</span> <span class="preprocessor">#if defined(POOL_AVG) || defined(POOL_L2)</span></div><div class="line"><a name="l00237"></a><span class="lineno"> 237</span>  <span class="comment">// Divide by pool region in case of average or l2 pooling</span></div><div class="line"><a name="l00238"></a><span class="lineno"> 238</span>  res = <a class="code" href="pooling__layer_8cl.xhtml#a5db17889d824975fefb2ce2f4690637f">DIV_OP</a>(res, <a class="code" href="pooling__layer_8cl.xhtml#ae9df7602479b001f6280f2f528f10a92">calculate_avg_scale</a>(2, 2, MAX_WIDTH, MAX_HEIGHT, PAD_X, PAD_Y, STRIDE_X, STRIDE_Y));</div><div class="line"><a name="l00239"></a><span class="lineno"> 239</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_AVG) || defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00240"></a><span class="lineno"> 240</span> </div><div class="line"><a name="l00241"></a><span class="lineno"> 241</span> <span class="preprocessor">#if defined(POOL_L2)</span></div><div class="line"><a name="l00242"></a><span class="lineno"> 242</span>  <span class="comment">// Take square root of the result in L2 pooling</span></div><div class="line"><a name="l00243"></a><span class="lineno"> 243</span>  res = <a class="code" href="pooling__layer_8cl.xhtml#ac9af19bec38fe50b4b9585c0e5c0ccca">SQRT_OP</a>(res);</div><div class="line"><a name="l00244"></a><span class="lineno"> 244</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00245"></a><span class="lineno"> 245</span> </div><div class="line"><a name="l00246"></a><span class="lineno"> 246</span>  <span class="comment">// Store result</span></div><div class="line"><a name="l00247"></a><span class="lineno"> 247</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)output.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> = (<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)res;</div><div class="line"><a name="l00248"></a><span class="lineno"> 248</span> }</div><div class="line"><a name="l00249"></a><span class="lineno"> 249</span> <span class="comment"></span></div><div class="line"><a name="l00250"></a><span class="lineno"> 250</span> <span class="comment">/** Performs a pooling function of pool size equal to 3</span></div><div class="line"><a name="l00251"></a><span class="lineno"> 251</span> <span class="comment"> *</span></div><div class="line"><a name="l00252"></a><span class="lineno"> 252</span> <span class="comment"> * @note Datatype must be passed using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types are F16/F32;</span></div><div class="line"><a name="l00253"></a><span class="lineno"> 253</span> <span class="comment"> * @note In case of average pooling the following information must be passed at compile time:</span></div><div class="line"><a name="l00254"></a><span class="lineno"> 254</span> <span class="comment"> * -DPOOL_AVG or -DPOOL_L2 must be provided otherwise max pooling will be performed.</span></div><div class="line"><a name="l00255"></a><span class="lineno"> 255</span> <span class="comment"> * -DMAX_WIDTH and -DMAX_HEIGHT which are the maximum accessible indeces in x and y dimensions (width + pad)</span></div><div class="line"><a name="l00256"></a><span class="lineno"> 256</span> <span class="comment"> * -DSTRIDE_X and -DSTRIDE_Y which are the steps of the window along the x and y directions</span></div><div class="line"><a name="l00257"></a><span class="lineno"> 257</span> <span class="comment"> * -DPAD_X and -DPAD_Y which are the pooling paddings in x and y dimension</span></div><div class="line"><a name="l00258"></a><span class="lineno"> 258</span> <span class="comment"> *</span></div><div class="line"><a name="l00259"></a><span class="lineno"> 259</span> <span class="comment"> * @param[in] input_ptr Pointer to the source image. Supported data types: F16/F32</span></div><div class="line"><a name="l00260"></a><span class="lineno"> 260</span> <span class="comment"> * @param[in] input_stride_x Stride of the source image in X dimension (in bytes)</span></div><div class="line"><a name="l00261"></a><span class="lineno"> 261</span> <span class="comment"> * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00262"></a><span class="lineno"> 262</span> <span class="comment"> * @param[in] input_stride_y Stride of the source image in Y dimension (in bytes)</span></div><div class="line"><a name="l00263"></a><span class="lineno"> 263</span> <span class="comment"> * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00264"></a><span class="lineno"> 264</span> <span class="comment"> * @param[in] input_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00265"></a><span class="lineno"> 265</span> <span class="comment"> * @param[in] input_step_z input_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00266"></a><span class="lineno"> 266</span> <span class="comment"> * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image</span></div><div class="line"><a name="l00267"></a><span class="lineno"> 267</span> <span class="comment"> * @param[out] output_ptr Pointer to the destination image. Supported data types: same as @p input_ptr</span></div><div class="line"><a name="l00268"></a><span class="lineno"> 268</span> <span class="comment"> * @param[in] output_stride_x Stride of the destination image in X dimension (in bytes)</span></div><div class="line"><a name="l00269"></a><span class="lineno"> 269</span> <span class="comment"> * @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00270"></a><span class="lineno"> 270</span> <span class="comment"> * @param[in] output_stride_y Stride of the destination image in Y dimension (in bytes)</span></div><div class="line"><a name="l00271"></a><span class="lineno"> 271</span> <span class="comment"> * @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00272"></a><span class="lineno"> 272</span> <span class="comment"> * @param[in] output_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00273"></a><span class="lineno"> 273</span> <span class="comment"> * @param[in] output_step_z output_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00274"></a><span class="lineno"> 274</span> <span class="comment"> * @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination image</span></div><div class="line"><a name="l00275"></a><span class="lineno"> 275</span> <span class="comment"> */</span></div><div class="line"><a name="l00276"></a><span class="lineno"><a class="line" href="pooling__layer_8cl.xhtml#af5751970a4d8c62febdc6c63d6d4fd1d"> 276</a></span> __kernel <span class="keywordtype">void</span> <a class="code" href="pooling__layer_8cl.xhtml#af5751970a4d8c62febdc6c63d6d4fd1d">pooling_layer_3</a>(</div><div class="line"><a name="l00277"></a><span class="lineno"> 277</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>),</div><div class="line"><a name="l00278"></a><span class="lineno"> 278</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(output))</div><div class="line"><a name="l00279"></a><span class="lineno"> 279</span> {</div><div class="line"><a name="l00280"></a><span class="lineno"> 280</span>  <span class="comment">// Get pixels pointer</span></div><div class="line"><a name="l00281"></a><span class="lineno"> 281</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>);</div><div class="line"><a name="l00282"></a><span class="lineno"> 282</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> output = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(output);</div><div class="line"><a name="l00283"></a><span class="lineno"> 283</span> </div><div class="line"><a name="l00284"></a><span class="lineno"> 284</span>  <span class="comment">// Load data</span></div><div class="line"><a name="l00285"></a><span class="lineno"> 285</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 3)</div><div class="line"><a name="l00286"></a><span class="lineno"> 286</span>  data0 = <a class="code" href="pooling__layer_8cl.xhtml#af6c2e106d0b5d1cb755bbc79d0b09d52">VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE</a>(3, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a2101b2fe0193ce227ae4e0945e321d85">tensor3D_offset</a>(&<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>, 0, 0, 0));</div><div class="line"><a name="l00287"></a><span class="lineno"> 287</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 3)</div><div class="line"><a name="l00288"></a><span class="lineno"> 288</span>  data1 = <a class="code" href="pooling__layer_8cl.xhtml#af6c2e106d0b5d1cb755bbc79d0b09d52">VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE</a>(3, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a2101b2fe0193ce227ae4e0945e321d85">tensor3D_offset</a>(&<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>, 0, 1, 0));</div><div class="line"><a name="l00289"></a><span class="lineno"> 289</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 3)</div><div class="line"><a name="l00290"></a><span class="lineno"> 290</span>  data2 = <a class="code" href="pooling__layer_8cl.xhtml#af6c2e106d0b5d1cb755bbc79d0b09d52">VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE</a>(3, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a2101b2fe0193ce227ae4e0945e321d85">tensor3D_offset</a>(&<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>, 0, 2, 0));</div><div class="line"><a name="l00291"></a><span class="lineno"> 291</span> </div><div class="line"><a name="l00292"></a><span class="lineno"> 292</span> <span class="preprocessor">#if defined(POOL_L2)</span></div><div class="line"><a name="l00293"></a><span class="lineno"> 293</span>  <span class="comment">// Raise to power of 2 for L2 Pooling</span></div><div class="line"><a name="l00294"></a><span class="lineno"> 294</span>  data0 = <a class="code" href="pooling__layer_8cl.xhtml#a03898439d164d74f8c35bafb67262d95">POW2_OP</a>(data0, 3);</div><div class="line"><a name="l00295"></a><span class="lineno"> 295</span>  data1 = <a class="code" href="pooling__layer_8cl.xhtml#a03898439d164d74f8c35bafb67262d95">POW2_OP</a>(data1, 3);</div><div class="line"><a name="l00296"></a><span class="lineno"> 296</span>  data2 = <a class="code" href="pooling__layer_8cl.xhtml#a03898439d164d74f8c35bafb67262d95">POW2_OP</a>(data2, 3);</div><div class="line"><a name="l00297"></a><span class="lineno"> 297</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00298"></a><span class="lineno"> 298</span> </div><div class="line"><a name="l00299"></a><span class="lineno"> 299</span>  <span class="comment">// Perform calculations</span></div><div class="line"><a name="l00300"></a><span class="lineno"> 300</span>  data0 = <a class="code" href="pooling__layer_8cl.xhtml#a482ef7d59a5f474ca126e737c7f0978a">POOL_OP</a>(data0, data1);</div><div class="line"><a name="l00301"></a><span class="lineno"> 301</span>  data0 = <a class="code" href="pooling__layer_8cl.xhtml#a482ef7d59a5f474ca126e737c7f0978a">POOL_OP</a>(data0, data2);</div><div class="line"><a name="l00302"></a><span class="lineno"> 302</span>  ACC_DATA_TYPE res = <a class="code" href="pooling__layer_8cl.xhtml#a482ef7d59a5f474ca126e737c7f0978a">POOL_OP</a>(<a class="code" href="pooling__layer_8cl.xhtml#a482ef7d59a5f474ca126e737c7f0978a">POOL_OP</a>(data0.s0, data0.s1), data0.s2);</div><div class="line"><a name="l00303"></a><span class="lineno"> 303</span> </div><div class="line"><a name="l00304"></a><span class="lineno"> 304</span> <span class="preprocessor">#if defined(POOL_AVG) || defined(POOL_L2)</span></div><div class="line"><a name="l00305"></a><span class="lineno"> 305</span>  <span class="comment">// Divide by pool region in case of average pooling</span></div><div class="line"><a name="l00306"></a><span class="lineno"> 306</span>  res = <a class="code" href="pooling__layer_8cl.xhtml#a5db17889d824975fefb2ce2f4690637f">DIV_OP</a>(res, <a class="code" href="pooling__layer_8cl.xhtml#ae9df7602479b001f6280f2f528f10a92">calculate_avg_scale</a>(3, 3, MAX_WIDTH, MAX_HEIGHT, PAD_X, PAD_Y, STRIDE_X, STRIDE_Y));</div><div class="line"><a name="l00307"></a><span class="lineno"> 307</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_AVG) || defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00308"></a><span class="lineno"> 308</span> </div><div class="line"><a name="l00309"></a><span class="lineno"> 309</span> <span class="preprocessor">#if defined(POOL_L2)</span></div><div class="line"><a name="l00310"></a><span class="lineno"> 310</span>  <span class="comment">// Take square root of the result in L2 pooling</span></div><div class="line"><a name="l00311"></a><span class="lineno"> 311</span>  res = <a class="code" href="pooling__layer_8cl.xhtml#ac9af19bec38fe50b4b9585c0e5c0ccca">SQRT_OP</a>(res);</div><div class="line"><a name="l00312"></a><span class="lineno"> 312</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00313"></a><span class="lineno"> 313</span> </div><div class="line"><a name="l00314"></a><span class="lineno"> 314</span>  <span class="comment">// Store result</span></div><div class="line"><a name="l00315"></a><span class="lineno"> 315</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)output.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> = (<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)res;</div><div class="line"><a name="l00316"></a><span class="lineno"> 316</span> }</div><div class="line"><a name="l00317"></a><span class="lineno"> 317</span> </div><div class="line"><a name="l00318"></a><span class="lineno"> 318</span> <span class="preprocessor">#if defined(POOLING3x3)</span></div><div class="line"><a name="l00319"></a><span class="lineno"> 319</span> </div><div class="line"><a name="l00320"></a><span class="lineno"> 320</span> <span class="preprocessor">#define CONVERT_OP(data_type) convert_##data_type##4</span></div><div class="line"><a name="l00321"></a><span class="lineno"> 321</span> <span class="preprocessor">#define CONVERT_VECTOR4(data_type) CONVERT_OP(data_type)</span></div><div class="line"><a name="l00322"></a><span class="lineno"> 322</span> </div><div class="line"><a name="l00323"></a><span class="lineno"> 323</span> <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 4)</div><div class="line"><a name="l00324"></a><span class="lineno"> 324</span> calculate_avg_scale4(const <span class="keywordtype">int</span> pool_size, const <span class="keywordtype">int</span> upper_bound_w, const <span class="keywordtype">int</span> upper_bound_h,</div><div class="line"><a name="l00325"></a><span class="lineno"> 325</span>  const <span class="keywordtype">int</span> pad_x, const <span class="keywordtype">int</span> pad_y, const <span class="keywordtype">int</span> stride_x, const <span class="keywordtype">int</span> stride_y)</div><div class="line"><a name="l00326"></a><span class="lineno"> 326</span> {</div><div class="line"><a name="l00327"></a><span class="lineno"> 327</span>  int4 start_x = ((int4)get_global_id(0) * 4 + (int4)(0, 1, 2, 3)) * (int4)stride_x - (int4)pad_x;</div><div class="line"><a name="l00328"></a><span class="lineno"> 328</span>  <span class="keywordtype">int</span> start_y = get_global_id(1) * stride_y - pad_y;</div><div class="line"><a name="l00329"></a><span class="lineno"> 329</span>  <span class="keyword">const</span> int4 end_x = min(start_x + (int4)pool_size, (int4)upper_bound_w);</div><div class="line"><a name="l00330"></a><span class="lineno"> 330</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> end_y = min(start_y + pool_size, upper_bound_h);</div><div class="line"><a name="l00331"></a><span class="lineno"> 331</span> <span class="preprocessor">#if defined(EXCLUDE_PADDING)</span></div><div class="line"><a name="l00332"></a><span class="lineno"> 332</span>  start_x = max((int4)0, start_x);</div><div class="line"><a name="l00333"></a><span class="lineno"> 333</span>  start_y = max(0, start_y);</div><div class="line"><a name="l00334"></a><span class="lineno"> 334</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(EXCLUDE_PADDING) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00335"></a><span class="lineno"> 335</span>  <span class="keywordflow">return</span> (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 4))(1.f) / CONVERT_VECTOR4(ACC_DATA_TYPE)(((int4)(end_y - start_y)) * (end_x - start_x));</div><div class="line"><a name="l00336"></a><span class="lineno"> 336</span> }</div><div class="line"><a name="l00337"></a><span class="lineno"> 337</span> <span class="comment"></span></div><div class="line"><a name="l00338"></a><span class="lineno"> 338</span> <span class="comment">/** Performs an optimized pooling function of pool size equal to 3 when the stride_x is less equal than 3</span></div><div class="line"><a name="l00339"></a><span class="lineno"> 339</span> <span class="comment"> *</span></div><div class="line"><a name="l00340"></a><span class="lineno"> 340</span> <span class="comment"> * @note Datatype must be passed using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types are F16/F32;</span></div><div class="line"><a name="l00341"></a><span class="lineno"> 341</span> <span class="comment"> * @note In case of average pooling the following information must be passed at compile time:</span></div><div class="line"><a name="l00342"></a><span class="lineno"> 342</span> <span class="comment"> * -DPOOL_AVG or -DPOOL_L2 must be provided otherwise max pooling will be performed.</span></div><div class="line"><a name="l00343"></a><span class="lineno"> 343</span> <span class="comment"> * -DMAX_WIDTH and -DMAX_HEIGHT which are the maximum accessible indeces in x and y dimensions (width + pad)</span></div><div class="line"><a name="l00344"></a><span class="lineno"> 344</span> <span class="comment"> * -DSTRIDE_X and -DSTRIDE_Y which are the steps of the window along the x and y directions</span></div><div class="line"><a name="l00345"></a><span class="lineno"> 345</span> <span class="comment"> * -DPAD_X and -DPAD_Y which are the pooling paddings in x and y dimension</span></div><div class="line"><a name="l00346"></a><span class="lineno"> 346</span> <span class="comment"> *</span></div><div class="line"><a name="l00347"></a><span class="lineno"> 347</span> <span class="comment"> * @param[in] input_ptr Pointer to the source image. Supported data types: F16/F32</span></div><div class="line"><a name="l00348"></a><span class="lineno"> 348</span> <span class="comment"> * @param[in] input_stride_x Stride of the source image in X dimension (in bytes)</span></div><div class="line"><a name="l00349"></a><span class="lineno"> 349</span> <span class="comment"> * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00350"></a><span class="lineno"> 350</span> <span class="comment"> * @param[in] input_stride_y Stride of the source image in Y dimension (in bytes)</span></div><div class="line"><a name="l00351"></a><span class="lineno"> 351</span> <span class="comment"> * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00352"></a><span class="lineno"> 352</span> <span class="comment"> * @param[in] input_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00353"></a><span class="lineno"> 353</span> <span class="comment"> * @param[in] input_step_z input_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00354"></a><span class="lineno"> 354</span> <span class="comment"> * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image</span></div><div class="line"><a name="l00355"></a><span class="lineno"> 355</span> <span class="comment"> * @param[out] output_ptr Pointer to the destination image. Supported data types: same as @p input_ptr</span></div><div class="line"><a name="l00356"></a><span class="lineno"> 356</span> <span class="comment"> * @param[in] output_stride_x Stride of the destination image in X dimension (in bytes)</span></div><div class="line"><a name="l00357"></a><span class="lineno"> 357</span> <span class="comment"> * @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00358"></a><span class="lineno"> 358</span> <span class="comment"> * @param[in] output_stride_y Stride of the destination image in Y dimension (in bytes)</span></div><div class="line"><a name="l00359"></a><span class="lineno"> 359</span> <span class="comment"> * @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00360"></a><span class="lineno"> 360</span> <span class="comment"> * @param[in] output_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00361"></a><span class="lineno"> 361</span> <span class="comment"> * @param[in] output_step_z output_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00362"></a><span class="lineno"> 362</span> <span class="comment"> * @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination image</span></div><div class="line"><a name="l00363"></a><span class="lineno"> 363</span> <span class="comment"> */</span></div><div class="line"><a name="l00364"></a><span class="lineno"> 364</span> __kernel <span class="keywordtype">void</span> pooling_layer_optimized_3(</div><div class="line"><a name="l00365"></a><span class="lineno"> 365</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>),</div><div class="line"><a name="l00366"></a><span class="lineno"> 366</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(output))</div><div class="line"><a name="l00367"></a><span class="lineno"> 367</span> {</div><div class="line"><a name="l00368"></a><span class="lineno"> 368</span>  <span class="comment">// Get pixels pointer</span></div><div class="line"><a name="l00369"></a><span class="lineno"> 369</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>);</div><div class="line"><a name="l00370"></a><span class="lineno"> 370</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> output = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(output);</div><div class="line"><a name="l00371"></a><span class="lineno"> 371</span> </div><div class="line"><a name="l00372"></a><span class="lineno"> 372</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 4)</div><div class="line"><a name="l00373"></a><span class="lineno"> 373</span>  res;</div><div class="line"><a name="l00374"></a><span class="lineno"> 374</span> </div><div class="line"><a name="l00375"></a><span class="lineno"> 375</span>  <span class="comment">// Perform pooling 3x3 for 4 output elements</span></div><div class="line"><a name="l00376"></a><span class="lineno"> 376</span>  POOLING3x3(res, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>, output);</div><div class="line"><a name="l00377"></a><span class="lineno"> 377</span> </div><div class="line"><a name="l00378"></a><span class="lineno"> 378</span> <span class="preprocessor">#if defined(POOL_AVG) || defined(POOL_L2)</span></div><div class="line"><a name="l00379"></a><span class="lineno"> 379</span>  <span class="comment">// Divide by pool region in case of average pooling</span></div><div class="line"><a name="l00380"></a><span class="lineno"> 380</span>  res *= calculate_avg_scale4(3, MAX_WIDTH, MAX_HEIGHT, PAD_X, PAD_Y, STRIDE_X, STRIDE_Y);</div><div class="line"><a name="l00381"></a><span class="lineno"> 381</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_AVG) || defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00382"></a><span class="lineno"> 382</span> </div><div class="line"><a name="l00383"></a><span class="lineno"> 383</span> <span class="preprocessor">#if defined(POOL_L2)</span></div><div class="line"><a name="l00384"></a><span class="lineno"> 384</span>  <span class="comment">// Take square root of the result in L2 pooling</span></div><div class="line"><a name="l00385"></a><span class="lineno"> 385</span>  res = <a class="code" href="pooling__layer_8cl.xhtml#ac9af19bec38fe50b4b9585c0e5c0ccca">SQRT_OP</a>(res);</div><div class="line"><a name="l00386"></a><span class="lineno"> 386</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00387"></a><span class="lineno"> 387</span> </div><div class="line"><a name="l00388"></a><span class="lineno"> 388</span>  vstore4(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(res, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 4)), 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)output.ptr);</div><div class="line"><a name="l00389"></a><span class="lineno"> 389</span> }</div><div class="line"><a name="l00390"></a><span class="lineno"> 390</span> <span class="preprocessor">#endif // defined(POOLING3x3)</span></div><div class="line"><a name="l00391"></a><span class="lineno"> 391</span> </div><div class="line"><a name="l00392"></a><span class="lineno"> 392</span> <span class="preprocessor">#if defined(POOL_SIZE_X) && defined(POOL_SIZE_Y)</span></div><div class="line"><a name="l00393"></a><span class="lineno"> 393</span> <span class="comment"></span></div><div class="line"><a name="l00394"></a><span class="lineno"> 394</span> <span class="comment">/** Performs a pooling function of pool size equal to N (NCHW)</span></div><div class="line"><a name="l00395"></a><span class="lineno"> 395</span> <span class="comment"> *</span></div><div class="line"><a name="l00396"></a><span class="lineno"> 396</span> <span class="comment"> * @note Datatype must be passed using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types are F16/F32;</span></div><div class="line"><a name="l00397"></a><span class="lineno"> 397</span> <span class="comment"> * @note Pool sizes must be passed using -DPOOL_SIZE_X and -DPOOL_SIZE_Y e.g. -DPOOL_SIZE_X=13;</span></div><div class="line"><a name="l00398"></a><span class="lineno"> 398</span> <span class="comment"> * @note In case of average pooling the following information must be passed at compile time:</span></div><div class="line"><a name="l00399"></a><span class="lineno"> 399</span> <span class="comment"> * -DPOOL_AVG must be provided otherwise max pooling will be performed.</span></div><div class="line"><a name="l00400"></a><span class="lineno"> 400</span> <span class="comment"> * -DMAX_WIDTH and -DMAX_HEIGHT which are the maximum accessible indeces in x and y dimensions (width + pad)</span></div><div class="line"><a name="l00401"></a><span class="lineno"> 401</span> <span class="comment"> * -DSTRIDE_X and -DSTRIDE_Y which are the steps of the window along the x and y directions</span></div><div class="line"><a name="l00402"></a><span class="lineno"> 402</span> <span class="comment"> * -DPAD_X and -DPAD_Y which are the pooling paddings in x and y dimension</span></div><div class="line"><a name="l00403"></a><span class="lineno"> 403</span> <span class="comment"> * @note The initial value for the pooling operation must be passed at compile time using -DINITIAL_VALUE e.g. -DINITIAL_VALUE=0</span></div><div class="line"><a name="l00404"></a><span class="lineno"> 404</span> <span class="comment"> *</span></div><div class="line"><a name="l00405"></a><span class="lineno"> 405</span> <span class="comment"> * @param[in] input_ptr Pointer to the source image. Supported data types: F16/F32</span></div><div class="line"><a name="l00406"></a><span class="lineno"> 406</span> <span class="comment"> * @param[in] input_stride_x Stride of the source image in X dimension (in bytes)</span></div><div class="line"><a name="l00407"></a><span class="lineno"> 407</span> <span class="comment"> * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00408"></a><span class="lineno"> 408</span> <span class="comment"> * @param[in] input_stride_y Stride of the source image in Y dimension (in bytes)</span></div><div class="line"><a name="l00409"></a><span class="lineno"> 409</span> <span class="comment"> * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00410"></a><span class="lineno"> 410</span> <span class="comment"> * @param[in] input_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00411"></a><span class="lineno"> 411</span> <span class="comment"> * @param[in] input_step_z input_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00412"></a><span class="lineno"> 412</span> <span class="comment"> * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image</span></div><div class="line"><a name="l00413"></a><span class="lineno"> 413</span> <span class="comment"> * @param[out] output_ptr Pointer to the destination image. Supported data types: same as @p input_ptr</span></div><div class="line"><a name="l00414"></a><span class="lineno"> 414</span> <span class="comment"> * @param[in] output_stride_x Stride of the destination image in X dimension (in bytes)</span></div><div class="line"><a name="l00415"></a><span class="lineno"> 415</span> <span class="comment"> * @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00416"></a><span class="lineno"> 416</span> <span class="comment"> * @param[in] output_stride_y Stride of the destination image in Y dimension (in bytes)</span></div><div class="line"><a name="l00417"></a><span class="lineno"> 417</span> <span class="comment"> * @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00418"></a><span class="lineno"> 418</span> <span class="comment"> * @param[in] output_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00419"></a><span class="lineno"> 419</span> <span class="comment"> * @param[in] output_step_z output_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00420"></a><span class="lineno"> 420</span> <span class="comment"> * @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination image</span></div><div class="line"><a name="l00421"></a><span class="lineno"> 421</span> <span class="comment"> */</span></div><div class="line"><a name="l00422"></a><span class="lineno"> 422</span> __kernel <span class="keywordtype">void</span> pooling_layer_MxN_nchw(</div><div class="line"><a name="l00423"></a><span class="lineno"> 423</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>),</div><div class="line"><a name="l00424"></a><span class="lineno"> 424</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(output))</div><div class="line"><a name="l00425"></a><span class="lineno"> 425</span> {</div><div class="line"><a name="l00426"></a><span class="lineno"> 426</span>  <span class="comment">// Get pixels pointer</span></div><div class="line"><a name="l00427"></a><span class="lineno"> 427</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>);</div><div class="line"><a name="l00428"></a><span class="lineno"> 428</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> output = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(output);</div><div class="line"><a name="l00429"></a><span class="lineno"> 429</span> </div><div class="line"><a name="l00430"></a><span class="lineno"> 430</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 8)</div><div class="line"><a name="l00431"></a><span class="lineno"> 431</span>  vdata = INITIAL_VALUE;</div><div class="line"><a name="l00432"></a><span class="lineno"> 432</span>  ACC_DATA_TYPE sdata = INITIAL_VALUE;</div><div class="line"><a name="l00433"></a><span class="lineno"> 433</span> </div><div class="line"><a name="l00434"></a><span class="lineno"> 434</span>  <span class="comment">// Load data</span></div><div class="line"><a name="l00435"></a><span class="lineno"> 435</span>  <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a0886942393a3ba0dfefaa7516b159784">for</a>(<span class="keywordtype">int</span> y = 0; y < POOL_SIZE_Y; y++)</div><div class="line"><a name="l00436"></a><span class="lineno"> 436</span>  {</div><div class="line"><a name="l00437"></a><span class="lineno"> 437</span>  <span class="keywordtype">int</span> x = 0;</div><div class="line"><a name="l00438"></a><span class="lineno"> 438</span>  <span class="keywordflow">for</span>(; x <= ((int)POOL_SIZE_X - 8); x += 8)</div><div class="line"><a name="l00439"></a><span class="lineno"> 439</span>  {</div><div class="line"><a name="l00440"></a><span class="lineno"> 440</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 8)</div><div class="line"><a name="l00441"></a><span class="lineno"> 441</span>  data0 = <a class="code" href="pooling__layer_8cl.xhtml#af6c2e106d0b5d1cb755bbc79d0b09d52">VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE</a>(8, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a2101b2fe0193ce227ae4e0945e321d85">tensor3D_offset</a>(&<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>, x, y, 0));</div><div class="line"><a name="l00442"></a><span class="lineno"> 442</span> <span class="preprocessor">#if defined(POOL_L2)</span></div><div class="line"><a name="l00443"></a><span class="lineno"> 443</span>  <span class="comment">// Raise to power of 2 for L2 Pooling</span></div><div class="line"><a name="l00444"></a><span class="lineno"> 444</span>  data0 *= data0;</div><div class="line"><a name="l00445"></a><span class="lineno"> 445</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00446"></a><span class="lineno"> 446</span>  vdata = <a class="code" href="pooling__layer_8cl.xhtml#a482ef7d59a5f474ca126e737c7f0978a">POOL_OP</a>(vdata, data0);</div><div class="line"><a name="l00447"></a><span class="lineno"> 447</span>  }</div><div class="line"><a name="l00448"></a><span class="lineno"> 448</span> </div><div class="line"><a name="l00449"></a><span class="lineno"> 449</span>  <span class="comment">// Leftover</span></div><div class="line"><a name="l00450"></a><span class="lineno"> 450</span>  <span class="keywordflow">for</span>(; x < (int)POOL_SIZE_X; ++x)</div><div class="line"><a name="l00451"></a><span class="lineno"> 451</span>  {</div><div class="line"><a name="l00452"></a><span class="lineno"> 452</span>  ACC_DATA_TYPE data0 = (ACC_DATA_TYPE)(*((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a2101b2fe0193ce227ae4e0945e321d85">tensor3D_offset</a>(&<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>, x, y, 0)));</div><div class="line"><a name="l00453"></a><span class="lineno"> 453</span> <span class="preprocessor">#if defined(POOL_L2)</span></div><div class="line"><a name="l00454"></a><span class="lineno"> 454</span>  <span class="comment">// Raise to power of 2 for L2 Pooling</span></div><div class="line"><a name="l00455"></a><span class="lineno"> 455</span>  data0 *= data0;</div><div class="line"><a name="l00456"></a><span class="lineno"> 456</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00457"></a><span class="lineno"> 457</span>  sdata = <a class="code" href="pooling__layer_8cl.xhtml#a482ef7d59a5f474ca126e737c7f0978a">POOL_OP</a>(sdata, data0);</div><div class="line"><a name="l00458"></a><span class="lineno"> 458</span>  }</div><div class="line"><a name="l00459"></a><span class="lineno"> 459</span>  }</div><div class="line"><a name="l00460"></a><span class="lineno"> 460</span> </div><div class="line"><a name="l00461"></a><span class="lineno"> 461</span>  <span class="comment">// Reduce result</span></div><div class="line"><a name="l00462"></a><span class="lineno"> 462</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 4)</div><div class="line"><a name="l00463"></a><span class="lineno"> 463</span>  reduce4 = <a class="code" href="pooling__layer_8cl.xhtml#a482ef7d59a5f474ca126e737c7f0978a">POOL_OP</a>(vdata.s0123, vdata.s4567);</div><div class="line"><a name="l00464"></a><span class="lineno"> 464</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 2)</div><div class="line"><a name="l00465"></a><span class="lineno"> 465</span>  reduce2 = <a class="code" href="pooling__layer_8cl.xhtml#a482ef7d59a5f474ca126e737c7f0978a">POOL_OP</a>(reduce4.s01, reduce4.s23);</div><div class="line"><a name="l00466"></a><span class="lineno"> 466</span>  ACC_DATA_TYPE res = <a class="code" href="pooling__layer_8cl.xhtml#a482ef7d59a5f474ca126e737c7f0978a">POOL_OP</a>(reduce2.s0, reduce2.s1);</div><div class="line"><a name="l00467"></a><span class="lineno"> 467</span>  res = <a class="code" href="pooling__layer_8cl.xhtml#a482ef7d59a5f474ca126e737c7f0978a">POOL_OP</a>(res, sdata);</div><div class="line"><a name="l00468"></a><span class="lineno"> 468</span> </div><div class="line"><a name="l00469"></a><span class="lineno"> 469</span> <span class="preprocessor">#if defined(POOL_AVG) || defined(POOL_L2)</span></div><div class="line"><a name="l00470"></a><span class="lineno"> 470</span>  <span class="comment">// Divide by pool region in case of average pooling</span></div><div class="line"><a name="l00471"></a><span class="lineno"> 471</span>  res = <a class="code" href="pooling__layer_8cl.xhtml#a5db17889d824975fefb2ce2f4690637f">DIV_OP</a>(res, <a class="code" href="pooling__layer_8cl.xhtml#ae9df7602479b001f6280f2f528f10a92">calculate_avg_scale</a>(POOL_SIZE_X, POOL_SIZE_Y, MAX_WIDTH, MAX_HEIGHT, PAD_X, PAD_Y, STRIDE_X, STRIDE_Y));</div><div class="line"><a name="l00472"></a><span class="lineno"> 472</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_AVG) || defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00473"></a><span class="lineno"> 473</span> </div><div class="line"><a name="l00474"></a><span class="lineno"> 474</span> <span class="preprocessor">#if defined(POOL_L2)</span></div><div class="line"><a name="l00475"></a><span class="lineno"> 475</span>  <span class="comment">// Take square root of the result in L2 pooling</span></div><div class="line"><a name="l00476"></a><span class="lineno"> 476</span>  res = <a class="code" href="pooling__layer_8cl.xhtml#ac9af19bec38fe50b4b9585c0e5c0ccca">SQRT_OP</a>(res);</div><div class="line"><a name="l00477"></a><span class="lineno"> 477</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00478"></a><span class="lineno"> 478</span> </div><div class="line"><a name="l00479"></a><span class="lineno"> 479</span>  <span class="comment">// Store result</span></div><div class="line"><a name="l00480"></a><span class="lineno"> 480</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)output.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> = (<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)res;</div><div class="line"><a name="l00481"></a><span class="lineno"> 481</span> }</div><div class="line"><a name="l00482"></a><span class="lineno"> 482</span> <span class="preprocessor">#endif // defined(POOL_SIZE_X) && defined(POOL_SIZE_Y)</span></div><div class="line"><a name="l00483"></a><span class="lineno"> 483</span> </div><div class="line"><a name="l00484"></a><span class="lineno"><a class="line" href="pooling__layer_8cl.xhtml#a3c82c9d144f57ed5694299ec248baad6"> 484</a></span> ACC_DATA_TYPE <a class="code" href="pooling__layer_8cl.xhtml#a3c82c9d144f57ed5694299ec248baad6">calculate_avg_scale_nhwc</a>(<span class="keyword">const</span> <span class="keywordtype">int</span> pool_size_x, <span class="keyword">const</span> <span class="keywordtype">int</span> pool_size_y, <span class="keywordtype">int</span> upper_bound_w, <span class="keywordtype">int</span> upper_bound_h,</div><div class="line"><a name="l00485"></a><span class="lineno"> 485</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> pad_x, <span class="keyword">const</span> <span class="keywordtype">int</span> pad_y, <span class="keyword">const</span> <span class="keywordtype">int</span> stride_x, <span class="keyword">const</span> <span class="keywordtype">int</span> stride_y)</div><div class="line"><a name="l00486"></a><span class="lineno"> 486</span> {</div><div class="line"><a name="l00487"></a><span class="lineno"> 487</span>  <span class="keywordtype">int</span> start_x = get_global_id(1) * stride_x - pad_x;</div><div class="line"><a name="l00488"></a><span class="lineno"> 488</span> <span class="preprocessor">#if defined(DST_DEPTH)</span></div><div class="line"><a name="l00489"></a><span class="lineno"> 489</span>  <span class="keywordtype">int</span> start_y = (get_global_id(2) % DST_DEPTH) * stride_y - pad_y;</div><div class="line"><a name="l00490"></a><span class="lineno"> 490</span> <span class="preprocessor">#else </span><span class="comment">/* defined(DST_DEPTH) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00491"></a><span class="lineno"> 491</span>  <span class="keywordtype">int</span> start_y = get_global_id(2) * stride_y - pad_y;</div><div class="line"><a name="l00492"></a><span class="lineno"> 492</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(DST_DEPTH) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00493"></a><span class="lineno"> 493</span> </div><div class="line"><a name="l00494"></a><span class="lineno"> 494</span> <span class="preprocessor">#if !defined(EXCLUDE_PADDING)</span></div><div class="line"><a name="l00495"></a><span class="lineno"> 495</span>  upper_bound_w += pad_x;</div><div class="line"><a name="l00496"></a><span class="lineno"> 496</span>  upper_bound_h += pad_y;</div><div class="line"><a name="l00497"></a><span class="lineno"> 497</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(EXCLUDE_PADDING) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00498"></a><span class="lineno"> 498</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> end_x = min(start_x + pool_size_x, upper_bound_w);</div><div class="line"><a name="l00499"></a><span class="lineno"> 499</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> end_y = min(start_y + pool_size_y, upper_bound_h);</div><div class="line"><a name="l00500"></a><span class="lineno"> 500</span> <span class="preprocessor">#if defined(EXCLUDE_PADDING)</span></div><div class="line"><a name="l00501"></a><span class="lineno"> 501</span>  start_x = max(0, start_x);</div><div class="line"><a name="l00502"></a><span class="lineno"> 502</span>  start_y = max(0, start_y);</div><div class="line"><a name="l00503"></a><span class="lineno"> 503</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(EXCLUDE_PADDING) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00504"></a><span class="lineno"> 504</span>  <span class="keywordflow">return</span> ((end_y - start_y) * (end_x - start_x));</div><div class="line"><a name="l00505"></a><span class="lineno"> 505</span> }</div><div class="line"><a name="l00506"></a><span class="lineno"> 506</span> <span class="comment"></span></div><div class="line"><a name="l00507"></a><span class="lineno"> 507</span> <span class="comment">/** Performs a pooling function of pool size equal to N (NHWC)</span></div><div class="line"><a name="l00508"></a><span class="lineno"> 508</span> <span class="comment"> *</span></div><div class="line"><a name="l00509"></a><span class="lineno"> 509</span> <span class="comment"> * @note Datatype must be passed using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types are F16/F32</span></div><div class="line"><a name="l00510"></a><span class="lineno"> 510</span> <span class="comment"> * @note Pool sizes must be passed using -DPOOL_SIZE_X and -DPOOL_SIZE_Y e.g. -DPOOL_SIZE_X=13;</span></div><div class="line"><a name="l00511"></a><span class="lineno"> 511</span> <span class="comment"> * @note Tensors width and height must be passed at compile time using -DMAX_WIDTH and -DMAX_HEIGHT</span></div><div class="line"><a name="l00512"></a><span class="lineno"> 512</span> <span class="comment"> * @note Strides must be passed at compile time using -DSTRIDE_X and -DSTRIDE_Y which are the steps of the window along the x and y directions</span></div><div class="line"><a name="l00513"></a><span class="lineno"> 513</span> <span class="comment"> * @note Pad values must be passed at compile time using -DPAD_X and -DPAD_Y which are the pooling paddings in x and y dimension</span></div><div class="line"><a name="l00514"></a><span class="lineno"> 514</span> <span class="comment"> * @note In case of average pooling the following information must be passed at compile time:</span></div><div class="line"><a name="l00515"></a><span class="lineno"> 515</span> <span class="comment"> * -DPOOL_AVG must be provided otherwise max pooling will be performed.</span></div><div class="line"><a name="l00516"></a><span class="lineno"> 516</span> <span class="comment"> * @note The initial value for the pooling operation must be passed at compile time using -DINITIAL_VALUE e.g. -DINITIAL_VALUE=0</span></div><div class="line"><a name="l00517"></a><span class="lineno"> 517</span> <span class="comment"> *</span></div><div class="line"><a name="l00518"></a><span class="lineno"> 518</span> <span class="comment"> * @param[in] input_ptr Pointer to the source image. Supported data types: F16/F32</span></div><div class="line"><a name="l00519"></a><span class="lineno"> 519</span> <span class="comment"> * @param[in] input_stride_x Stride of the source image in X dimension (in bytes)</span></div><div class="line"><a name="l00520"></a><span class="lineno"> 520</span> <span class="comment"> * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00521"></a><span class="lineno"> 521</span> <span class="comment"> * @param[in] input_stride_y Stride of the source image in Y dimension (in bytes)</span></div><div class="line"><a name="l00522"></a><span class="lineno"> 522</span> <span class="comment"> * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00523"></a><span class="lineno"> 523</span> <span class="comment"> * @param[in] input_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00524"></a><span class="lineno"> 524</span> <span class="comment"> * @param[in] input_step_z input_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00525"></a><span class="lineno"> 525</span> <span class="comment"> * @param[in] input_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l00526"></a><span class="lineno"> 526</span> <span class="comment"> * @param[in] input_step_w input_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l00527"></a><span class="lineno"> 527</span> <span class="comment"> * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image</span></div><div class="line"><a name="l00528"></a><span class="lineno"> 528</span> <span class="comment"> * @param[out] output_ptr Pointer to the destination image. Supported data types: same as @p input_ptr</span></div><div class="line"><a name="l00529"></a><span class="lineno"> 529</span> <span class="comment"> * @param[in] output_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00530"></a><span class="lineno"> 530</span> <span class="comment"> * @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00531"></a><span class="lineno"> 531</span> <span class="comment"> * @param[in] output_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00532"></a><span class="lineno"> 532</span> <span class="comment"> * @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00533"></a><span class="lineno"> 533</span> <span class="comment"> * @param[in] output_stride_z Stride of the destination tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00534"></a><span class="lineno"> 534</span> <span class="comment"> * @param[in] output_step_z output_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00535"></a><span class="lineno"> 535</span> <span class="comment"> * @param[in] output_stride_w Stride of the destination tensor in W dimension (in bytes)</span></div><div class="line"><a name="l00536"></a><span class="lineno"> 536</span> <span class="comment"> * @param[in] output_step_w output_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l00537"></a><span class="lineno"> 537</span> <span class="comment"> * @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination image</span></div><div class="line"><a name="l00538"></a><span class="lineno"> 538</span> <span class="comment"> */</span></div><div class="line"><a name="l00539"></a><span class="lineno"><a class="line" href="pooling__layer_8cl.xhtml#a074db9113f7fb9fc3f5e389892b38d32"> 539</a></span> __kernel <span class="keywordtype">void</span> <a class="code" href="pooling__layer_8cl.xhtml#a074db9113f7fb9fc3f5e389892b38d32">pooling_layer_MxN_nhwc</a>(</div><div class="line"><a name="l00540"></a><span class="lineno"> 540</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a481bdc6d61b3df9dcdbdb244f0f97790">TENSOR4D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>),</div><div class="line"><a name="l00541"></a><span class="lineno"> 541</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a481bdc6d61b3df9dcdbdb244f0f97790">TENSOR4D_DECLARATION</a>(output))</div><div class="line"><a name="l00542"></a><span class="lineno"> 542</span> {</div><div class="line"><a name="l00543"></a><span class="lineno"> 543</span>  <span class="comment">// Get pixels pointer</span></div><div class="line"><a name="l00544"></a><span class="lineno"> 544</span> <span class="preprocessor">#if defined(DST_DEPTH)</span></div><div class="line"><a name="l00545"></a><span class="lineno"> 545</span>  <a class="code" href="struct_tensor4_d.xhtml">Tensor4D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a23b9032d1b9d59547545e457f82ee478">CONVERT_TO_TENSOR4D_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>, DST_DEPTH);</div><div class="line"><a name="l00546"></a><span class="lineno"> 546</span>  <a class="code" href="struct_tensor4_d.xhtml">Tensor4D</a> output = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a23b9032d1b9d59547545e457f82ee478">CONVERT_TO_TENSOR4D_STRUCT</a>(output, DST_DEPTH);</div><div class="line"><a name="l00547"></a><span class="lineno"> 547</span> <span class="preprocessor">#else </span><span class="comment">/* defined(DST_DEPTH) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00548"></a><span class="lineno"> 548</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>);</div><div class="line"><a name="l00549"></a><span class="lineno"> 549</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> output = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(output);</div><div class="line"><a name="l00550"></a><span class="lineno"> 550</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(DST_DEPTH) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00551"></a><span class="lineno"> 551</span> </div><div class="line"><a name="l00552"></a><span class="lineno"> 552</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 8)</div><div class="line"><a name="l00553"></a><span class="lineno"> 553</span>  vdata = INITIAL_VALUE;</div><div class="line"><a name="l00554"></a><span class="lineno"> 554</span> </div><div class="line"><a name="l00555"></a><span class="lineno"> 555</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> idx_width = get_global_id(1) * STRIDE_X;</div><div class="line"><a name="l00556"></a><span class="lineno"> 556</span> <span class="preprocessor">#if defined(DST_DEPTH)</span></div><div class="line"><a name="l00557"></a><span class="lineno"> 557</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> idx_height = (get_global_id(2) % DST_DEPTH) * STRIDE_Y;</div><div class="line"><a name="l00558"></a><span class="lineno"> 558</span> <span class="preprocessor">#else </span><span class="comment">/* defined(DST_DEPTH) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00559"></a><span class="lineno"> 559</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> idx_height = get_global_id(2) * STRIDE_Y;</div><div class="line"><a name="l00560"></a><span class="lineno"> 560</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(DST_DEPTH) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00561"></a><span class="lineno"> 561</span> </div><div class="line"><a name="l00562"></a><span class="lineno"> 562</span>  <span class="keywordflow">for</span>(<span class="keywordtype">int</span> y = 0; y < POOL_SIZE_Y; ++y)</div><div class="line"><a name="l00563"></a><span class="lineno"> 563</span>  {</div><div class="line"><a name="l00564"></a><span class="lineno"> 564</span>  <span class="keywordtype">int</span> y1 = <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#af77145fbdc6b0c8931148f5597d9de53">select</a>(y, PAD_Y - idx_height, y + idx_height - PAD_Y < 0 || y + idx_height - PAD_Y >= MAX_HEIGHT);</div><div class="line"><a name="l00565"></a><span class="lineno"> 565</span>  <span class="keywordflow">for</span>(<span class="keywordtype">int</span> x = 0; x < POOL_SIZE_X; ++x)</div><div class="line"><a name="l00566"></a><span class="lineno"> 566</span>  {</div><div class="line"><a name="l00567"></a><span class="lineno"> 567</span>  <span class="keywordtype">int</span> x1 = <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#af77145fbdc6b0c8931148f5597d9de53">select</a>(x, PAD_X - idx_width - 1, x + idx_width - PAD_X < 0 || x + idx_width - PAD_X >= MAX_WIDTH);</div><div class="line"><a name="l00568"></a><span class="lineno"> 568</span>  x1 = <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#af77145fbdc6b0c8931148f5597d9de53">select</a>(x1, PAD_X - idx_width - 1, y != y1);</div><div class="line"><a name="l00569"></a><span class="lineno"> 569</span> </div><div class="line"><a name="l00570"></a><span class="lineno"> 570</span> <span class="preprocessor">#if defined(DST_DEPTH)</span></div><div class="line"><a name="l00571"></a><span class="lineno"> 571</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 8)</div><div class="line"><a name="l00572"></a><span class="lineno"> 572</span>  data0 = <a class="code" href="pooling__layer_8cl.xhtml#af6c2e106d0b5d1cb755bbc79d0b09d52">VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE</a>(8, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#ad442fb5ec8be1fff97f543150de5d822">tensor4D_offset</a>(&<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>, 0, x1 - PAD_X, y1 - PAD_Y, 0));</div><div class="line"><a name="l00573"></a><span class="lineno"> 573</span> <span class="preprocessor">#else </span><span class="comment">/* defined(DST_DEPTH) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00574"></a><span class="lineno"> 574</span>  <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 8)</div><div class="line"><a name="l00575"></a><span class="lineno"> 575</span>  data0 = <a class="code" href="pooling__layer_8cl.xhtml#af6c2e106d0b5d1cb755bbc79d0b09d52">VLOAD_AND_CONVERT_TO_ACC_DATA_TYPE</a>(8, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a2101b2fe0193ce227ae4e0945e321d85">tensor3D_offset</a>(&<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a8fcf2ddd9a1d58b1b280f5c0aed71845">input</a>, 0, x1 - PAD_X, y1 - PAD_Y));</div><div class="line"><a name="l00576"></a><span class="lineno"> 576</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(DST_DEPTH) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00577"></a><span class="lineno"> 577</span> </div><div class="line"><a name="l00578"></a><span class="lineno"> 578</span> <span class="preprocessor">#if defined(POOL_L2)</span></div><div class="line"><a name="l00579"></a><span class="lineno"> 579</span>  <span class="comment">// Raise to power of 2 for L2 Pooling</span></div><div class="line"><a name="l00580"></a><span class="lineno"> 580</span>  data0 *= data0;</div><div class="line"><a name="l00581"></a><span class="lineno"> 581</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00582"></a><span class="lineno"> 582</span>  vdata = <a class="code" href="pooling__layer_8cl.xhtml#a482ef7d59a5f474ca126e737c7f0978a">POOL_OP</a>(vdata, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(data0, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(ACC_DATA_TYPE, 8)));</div><div class="line"><a name="l00583"></a><span class="lineno"> 583</span>  }</div><div class="line"><a name="l00584"></a><span class="lineno"> 584</span>  }</div><div class="line"><a name="l00585"></a><span class="lineno"> 585</span> </div><div class="line"><a name="l00586"></a><span class="lineno"> 586</span> <span class="preprocessor">#if defined(POOL_AVG) || defined(POOL_L2)</span></div><div class="line"><a name="l00587"></a><span class="lineno"> 587</span>  <span class="comment">// Divide by pool region in case of average pooling</span></div><div class="line"><a name="l00588"></a><span class="lineno"> 588</span>  vdata = <a class="code" href="pooling__layer_8cl.xhtml#a6c01fa98d360a9d52926dc6a5a599711">DIV_OP_NHWC</a>(vdata, <a class="code" href="pooling__layer_8cl.xhtml#a3c82c9d144f57ed5694299ec248baad6">calculate_avg_scale_nhwc</a>(POOL_SIZE_X, POOL_SIZE_Y, MAX_WIDTH, MAX_HEIGHT, PAD_X, PAD_Y, STRIDE_X, STRIDE_Y));</div><div class="line"><a name="l00589"></a><span class="lineno"> 589</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_AVG) || defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00590"></a><span class="lineno"> 590</span> </div><div class="line"><a name="l00591"></a><span class="lineno"> 591</span> <span class="preprocessor">#if defined(POOL_L2)</span></div><div class="line"><a name="l00592"></a><span class="lineno"> 592</span>  <span class="comment">// Take square root of the result in L2 pooling</span></div><div class="line"><a name="l00593"></a><span class="lineno"> 593</span>  vdata = <a class="code" href="pooling__layer_8cl.xhtml#ac9af19bec38fe50b4b9585c0e5c0ccca">SQRT_OP</a>(vdata);</div><div class="line"><a name="l00594"></a><span class="lineno"> 594</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(POOL_L2) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00595"></a><span class="lineno"> 595</span> </div><div class="line"><a name="l00596"></a><span class="lineno"> 596</span>  <span class="comment">// Store result</span></div><div class="line"><a name="l00597"></a><span class="lineno"> 597</span>  vstore8(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(vdata, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 8)), 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)output.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l00598"></a><span class="lineno"> 598</span> }</div><div class="ttc" id="pooling__layer_8cl_xhtml_a5db17889d824975fefb2ce2f4690637f"><div class="ttname"><a href="pooling__layer_8cl.xhtml#a5db17889d824975fefb2ce2f4690637f">DIV_OP</a></div><div class="ttdeci">#define DIV_OP(x, y)</div><div class="ttdef"><b>Definition:</b> <a href="pooling__layer_8cl_source.xhtml#l00038">pooling_layer.cl:38</a></div></div> |
