| <a href="winograd__filter__transform_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) 2018-2019 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(SRC_DIM_Z)</span></div><div class="line"><a name="l00027"></a><span class="lineno"> 27</span> </div><div class="line"><a name="l00028"></a><span class="lineno"> 28</span> <span class="preprocessor">#define OUTPUT_ROW_2x2_7x7(out, tmp) \</span></div><div class="line"><a name="l00029"></a><span class="lineno"> 29</span> <span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00030"></a><span class="lineno"> 30</span> <span class="preprocessor"> out.s0 = -tmp.s0 / 36.f; \</span></div><div class="line"><a name="l00031"></a><span class="lineno"> 31</span> <span class="preprocessor"> out.s1 = (tmp.s0 - tmp.s1 + tmp.s2 - tmp.s3 + tmp.s4 - tmp.s5 + tmp.s6) / 48.f; \</span></div><div class="line"><a name="l00032"></a><span class="lineno"> 32</span> <span class="preprocessor"> out.s2 = (tmp.s0 + tmp.s1 + tmp.s2 + tmp.s3 + tmp.s4 + tmp.s5 + tmp.s6) / 48.f; \</span></div><div class="line"><a name="l00033"></a><span class="lineno"> 33</span> <span class="preprocessor"> out.s3 = (-tmp.s0 + 2.f * tmp.s1 - 4.f * tmp.s2 + 8.f * tmp.s3 - 16.f * tmp.s4 + 32.f * tmp.s5 - 64.f * tmp.s6) / 120.f; \</span></div><div class="line"><a name="l00034"></a><span class="lineno"> 34</span> <span class="preprocessor"> out.s4 = (-tmp.s0 - 2.f * tmp.s1 - 4.f * tmp.s2 - 8.f * tmp.s3 - 16.f * tmp.s4 - 32.f * tmp.s5 - 64.f * tmp.s6) / 120.f; \</span></div><div class="line"><a name="l00035"></a><span class="lineno"> 35</span> <span class="preprocessor"> out.s5 = (tmp.s0 - 3.f * tmp.s1 + 9.f * tmp.s2 - 27.f * tmp.s3 + 81.f * tmp.s4 - 243.f * tmp.s5 + 729.f * tmp.s6) / 720.f; \</span></div><div class="line"><a name="l00036"></a><span class="lineno"> 36</span> <span class="preprocessor"> out.s6 = (tmp.s0 + 3.f * tmp.s1 + 9.f * tmp.s2 + 27.f * tmp.s3 + 81.f * tmp.s4 + 243.f * tmp.s5 + 729.f * tmp.s6) / 720.f; \</span></div><div class="line"><a name="l00037"></a><span class="lineno"> 37</span> <span class="preprocessor"> out.s7 = tmp.s6; \</span></div><div class="line"><a name="l00038"></a><span class="lineno"> 38</span> <span class="preprocessor"> })</span></div><div class="line"><a name="l00039"></a><span class="lineno"> 39</span> <span class="comment"></span></div><div class="line"><a name="l00040"></a><span class="lineno"> 40</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 3x3/3x1/1x3 when the data layout is NCHW and the output tile is 2x2/2x1/1x2</span></div><div class="line"><a name="l00041"></a><span class="lineno"> 41</span> <span class="comment"> *</span></div><div class="line"><a name="l00042"></a><span class="lineno"> 42</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l00043"></a><span class="lineno"> 43</span> <span class="comment"> * @note If this kernel is used to perform Winograd filter transform 3x1, -DWINOGRAD_FILTER_TRANSFORM_HORIZONTAL has to be passed at compile time</span></div><div class="line"><a name="l00044"></a><span class="lineno"> 44</span> <span class="comment"> * @note If this kernel is used to perform Winograd filter transform 1x3, -DWINOGRAD_FILTER_TRANSFORM_VERTICAL has to be passed at compile time</span></div><div class="line"><a name="l00045"></a><span class="lineno"> 45</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l00046"></a><span class="lineno"> 46</span> <span class="comment"> *</span></div><div class="line"><a name="l00047"></a><span class="lineno"> 47</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l00048"></a><span class="lineno"> 48</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00049"></a><span class="lineno"> 49</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00050"></a><span class="lineno"> 50</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00051"></a><span class="lineno"> 51</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00052"></a><span class="lineno"> 52</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00053"></a><span class="lineno"> 53</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00054"></a><span class="lineno"> 54</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l00055"></a><span class="lineno"> 55</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l00056"></a><span class="lineno"> 56</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l00057"></a><span class="lineno"> 57</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l00058"></a><span class="lineno"> 58</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00059"></a><span class="lineno"> 59</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00060"></a><span class="lineno"> 60</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00061"></a><span class="lineno"> 61</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00062"></a><span class="lineno"> 62</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00063"></a><span class="lineno"> 63</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00064"></a><span class="lineno"> 64</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l00065"></a><span class="lineno"> 65</span> <span class="comment"> */</span></div><div class="line"><a name="l00066"></a><span class="lineno"> 66</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_2x2_3x3_nchw(</div><div class="line"><a name="l00067"></a><span class="lineno"> 67</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l00068"></a><span class="lineno"> 68</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l00069"></a><span class="lineno"> 69</span> {</div><div class="line"><a name="l00070"></a><span class="lineno"> 70</span>  <a class="code" href="struct_tensor4_d.xhtml">Tensor4D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</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#a989ab3e96426615bb98e04e0235088ca">src</a>, SRC_DIM_Z);</div><div class="line"><a name="l00071"></a><span class="lineno"> 71</span> </div><div class="line"><a name="l00072"></a><span class="lineno"> 72</span>  <span class="keyword">const</span> __global uchar *src_addr = <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#a989ab3e96426615bb98e04e0235088ca">src</a>, 0, 0, 0, 0);</div><div class="line"><a name="l00073"></a><span class="lineno"> 73</span> </div><div class="line"><a name="l00074"></a><span class="lineno"> 74</span>  <span class="comment">// Load the values from the input tensor</span></div><div class="line"><a name="l00075"></a><span class="lineno"> 75</span> <span class="preprocessor">#if defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL)</span></div><div class="line"><a name="l00076"></a><span class="lineno"> 76</span>  <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>, 3)</div><div class="line"><a name="l00077"></a><span class="lineno"> 77</span>  w0 = vload3(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr));</div><div class="line"><a name="l00078"></a><span class="lineno"> 78</span> <span class="preprocessor">#elif defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00079"></a><span class="lineno"> 79</span>  <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>, 3)</div><div class="line"><a name="l00080"></a><span class="lineno"> 80</span>  w0 = (<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>, 3))(*((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_y)),</div><div class="line"><a name="l00081"></a><span class="lineno"> 81</span>  *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_y)),</div><div class="line"><a name="l00082"></a><span class="lineno"> 82</span>  *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_y)));</div><div class="line"><a name="l00083"></a><span class="lineno"> 83</span> <span class="preprocessor">#else // defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00084"></a><span class="lineno"> 84</span>  <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>, 3)</div><div class="line"><a name="l00085"></a><span class="lineno"> 85</span>  w0 = vload3(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_y));</div><div class="line"><a name="l00086"></a><span class="lineno"> 86</span>  <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>, 3)</div><div class="line"><a name="l00087"></a><span class="lineno"> 87</span>  w1 = vload3(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_y));</div><div class="line"><a name="l00088"></a><span class="lineno"> 88</span>  <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>, 3)</div><div class="line"><a name="l00089"></a><span class="lineno"> 89</span>  w2 = vload3(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_y));</div><div class="line"><a name="l00090"></a><span class="lineno"> 90</span> <span class="preprocessor">#endif // defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL)</span></div><div class="line"><a name="l00091"></a><span class="lineno"> 91</span> </div><div class="line"><a name="l00092"></a><span class="lineno"> 92</span>  <span class="comment">// Row 0</span></div><div class="line"><a name="l00093"></a><span class="lineno"> 93</span>  <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)</div><div class="line"><a name="l00094"></a><span class="lineno"> 94</span>  out0 = 0.0f;</div><div class="line"><a name="l00095"></a><span class="lineno"> 95</span>  out0.s0 = (w0.s0);</div><div class="line"><a name="l00096"></a><span class="lineno"> 96</span>  out0.s1 = (w0.s0 + w0.s1 + w0.s2) * 0.5f;</div><div class="line"><a name="l00097"></a><span class="lineno"> 97</span>  out0.s2 = (w0.s0 + w0.s2 - w0.s1) * 0.5f;</div><div class="line"><a name="l00098"></a><span class="lineno"> 98</span>  out0.s3 = (w0.s2);</div><div class="line"><a name="l00099"></a><span class="lineno"> 99</span> </div><div class="line"><a name="l00100"></a><span class="lineno"> 100</span> <span class="preprocessor">#if !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00101"></a><span class="lineno"> 101</span>  <span class="comment">// Row 1</span></div><div class="line"><a name="l00102"></a><span class="lineno"> 102</span>  <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)</div><div class="line"><a name="l00103"></a><span class="lineno"> 103</span>  out1 = 0.0f;</div><div class="line"><a name="l00104"></a><span class="lineno"> 104</span>  out1.s0 = (w0.s0 + w1.s0 + w2.s0) * 0.5f;</div><div class="line"><a name="l00105"></a><span class="lineno"> 105</span>  out1.s1 = (w0.s0 + w1.s0 + w2.s0 + w0.s1 + w1.s1 + w2.s1 + w0.s2 + w1.s2 + w2.s2) * 0.25f;</div><div class="line"><a name="l00106"></a><span class="lineno"> 106</span>  out1.s2 = (w0.s0 + w1.s0 + w2.s0 + w0.s2 + w1.s2 + w2.s2 - w0.s1 - w1.s1 - w2.s1) * 0.25f;</div><div class="line"><a name="l00107"></a><span class="lineno"> 107</span>  out1.s3 = (w0.s2 + w1.s2 + w2.s2) * 0.5f;</div><div class="line"><a name="l00108"></a><span class="lineno"> 108</span> </div><div class="line"><a name="l00109"></a><span class="lineno"> 109</span>  <span class="comment">// Row 2</span></div><div class="line"><a name="l00110"></a><span class="lineno"> 110</span>  <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)</div><div class="line"><a name="l00111"></a><span class="lineno"> 111</span>  out2 = 0.0f;</div><div class="line"><a name="l00112"></a><span class="lineno"> 112</span>  out2.s0 = (w0.s0 + w2.s0 - w1.s0) * 0.5f;</div><div class="line"><a name="l00113"></a><span class="lineno"> 113</span>  out2.s1 = (w0.s0 + w2.s0 + w0.s1 + w2.s1 + w0.s2 + w2.s2 - w1.s0 - w1.s1 - w1.s2) * 0.25f;</div><div class="line"><a name="l00114"></a><span class="lineno"> 114</span>  out2.s2 = (w0.s0 + w2.s0 + w1.s1 + w0.s2 + w2.s2 - w1.s0 - w0.s1 - w2.s1 - w1.s2) * 0.25f;</div><div class="line"><a name="l00115"></a><span class="lineno"> 115</span>  out2.s3 = (w0.s2 + w2.s2 - w1.s2) * 0.5f;</div><div class="line"><a name="l00116"></a><span class="lineno"> 116</span> </div><div class="line"><a name="l00117"></a><span class="lineno"> 117</span>  <span class="comment">// Row 3</span></div><div class="line"><a name="l00118"></a><span class="lineno"> 118</span>  <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)</div><div class="line"><a name="l00119"></a><span class="lineno"> 119</span>  out3 = 0.0f;</div><div class="line"><a name="l00120"></a><span class="lineno"> 120</span>  out3.s0 = (w2.s0);</div><div class="line"><a name="l00121"></a><span class="lineno"> 121</span>  out3.s1 = (w2.s0 + w2.s1 + w2.s2) * 0.5f;</div><div class="line"><a name="l00122"></a><span class="lineno"> 122</span>  out3.s2 = (w2.s0 + w2.s2 - w2.s1) * 0.5f;</div><div class="line"><a name="l00123"></a><span class="lineno"> 123</span>  out3.s3 = (w2.s2);</div><div class="line"><a name="l00124"></a><span class="lineno"> 124</span> <span class="preprocessor">#endif // !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00125"></a><span class="lineno"> 125</span> </div><div class="line"><a name="l00126"></a><span class="lineno"> 126</span>  <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l00127"></a><span class="lineno"> 127</span>  <span class="keywordtype">int</span> x0 = z / SRC_DIM_Z; <span class="comment">// idx filter</span></div><div class="line"><a name="l00128"></a><span class="lineno"> 128</span>  <span class="keywordtype">int</span> y0 = z % SRC_DIM_Z; <span class="comment">// idx channel</span></div><div class="line"><a name="l00129"></a><span class="lineno"> 129</span> </div><div class="line"><a name="l00130"></a><span class="lineno"> 130</span>  <span class="comment">// Get output address</span></div><div class="line"><a name="l00131"></a><span class="lineno"> 131</span>  __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x0 * dst_stride_x + y0 * dst_stride_y;</div><div class="line"><a name="l00132"></a><span class="lineno"> 132</span> </div><div class="line"><a name="l00133"></a><span class="lineno"> 133</span>  <span class="comment">// Store the values across the channels</span></div><div class="line"><a name="l00134"></a><span class="lineno"> 134</span>  <span class="comment">// 16 channels for 3x3 kernels</span></div><div class="line"><a name="l00135"></a><span class="lineno"> 135</span>  <span class="comment">// 4 channels for 3x1 or 1x3 kernels</span></div><div class="line"><a name="l00136"></a><span class="lineno"> 136</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 0 * dst_stride_z) = out0.s0;</div><div class="line"><a name="l00137"></a><span class="lineno"> 137</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 1 * dst_stride_z) = out0.s1;</div><div class="line"><a name="l00138"></a><span class="lineno"> 138</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 2 * dst_stride_z) = out0.s2;</div><div class="line"><a name="l00139"></a><span class="lineno"> 139</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 3 * dst_stride_z) = out0.s3;</div><div class="line"><a name="l00140"></a><span class="lineno"> 140</span> </div><div class="line"><a name="l00141"></a><span class="lineno"> 141</span> <span class="preprocessor">#if !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00142"></a><span class="lineno"> 142</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 4 * dst_stride_z) = out1.s0;</div><div class="line"><a name="l00143"></a><span class="lineno"> 143</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 5 * dst_stride_z) = out1.s1;</div><div class="line"><a name="l00144"></a><span class="lineno"> 144</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 6 * dst_stride_z) = out1.s2;</div><div class="line"><a name="l00145"></a><span class="lineno"> 145</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 7 * dst_stride_z) = out1.s3;</div><div class="line"><a name="l00146"></a><span class="lineno"> 146</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 8 * dst_stride_z) = out2.s0;</div><div class="line"><a name="l00147"></a><span class="lineno"> 147</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 9 * dst_stride_z) = out2.s1;</div><div class="line"><a name="l00148"></a><span class="lineno"> 148</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 10 * dst_stride_z) = out2.s2;</div><div class="line"><a name="l00149"></a><span class="lineno"> 149</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 11 * dst_stride_z) = out2.s3;</div><div class="line"><a name="l00150"></a><span class="lineno"> 150</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 12 * dst_stride_z) = out3.s0;</div><div class="line"><a name="l00151"></a><span class="lineno"> 151</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 13 * dst_stride_z) = out3.s1;</div><div class="line"><a name="l00152"></a><span class="lineno"> 152</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 14 * dst_stride_z) = out3.s2;</div><div class="line"><a name="l00153"></a><span class="lineno"> 153</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 15 * dst_stride_z) = out3.s3;</div><div class="line"><a name="l00154"></a><span class="lineno"> 154</span> <span class="preprocessor">#endif // !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00155"></a><span class="lineno"> 155</span> }</div><div class="line"><a name="l00156"></a><span class="lineno"> 156</span> <span class="comment"></span></div><div class="line"><a name="l00157"></a><span class="lineno"> 157</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 3x3/3x1/1x3 when the data layout is NCHW and the output tile is 4x4/4x1/1x4</span></div><div class="line"><a name="l00158"></a><span class="lineno"> 158</span> <span class="comment"> *</span></div><div class="line"><a name="l00159"></a><span class="lineno"> 159</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l00160"></a><span class="lineno"> 160</span> <span class="comment"> * @note If this kernel is used to perform Winograd filter transform 3x1, -DWINOGRAD_FILTER_TRANSFORM_HORIZONTAL has to be passed at compile time</span></div><div class="line"><a name="l00161"></a><span class="lineno"> 161</span> <span class="comment"> * @note If this kernel is used to perform Winograd filter transform 1x3, -DWINOGRAD_FILTER_TRANSFORM_VERTICAL has to be passed at compile time</span></div><div class="line"><a name="l00162"></a><span class="lineno"> 162</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l00163"></a><span class="lineno"> 163</span> <span class="comment"> *</span></div><div class="line"><a name="l00164"></a><span class="lineno"> 164</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l00165"></a><span class="lineno"> 165</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00166"></a><span class="lineno"> 166</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00167"></a><span class="lineno"> 167</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00168"></a><span class="lineno"> 168</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00169"></a><span class="lineno"> 169</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00170"></a><span class="lineno"> 170</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00171"></a><span class="lineno"> 171</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l00172"></a><span class="lineno"> 172</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l00173"></a><span class="lineno"> 173</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l00174"></a><span class="lineno"> 174</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l00175"></a><span class="lineno"> 175</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00176"></a><span class="lineno"> 176</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00177"></a><span class="lineno"> 177</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00178"></a><span class="lineno"> 178</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00179"></a><span class="lineno"> 179</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00180"></a><span class="lineno"> 180</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00181"></a><span class="lineno"> 181</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l00182"></a><span class="lineno"> 182</span> <span class="comment"> */</span></div><div class="line"><a name="l00183"></a><span class="lineno"> 183</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_4x4_3x3_nchw(</div><div class="line"><a name="l00184"></a><span class="lineno"> 184</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l00185"></a><span class="lineno"> 185</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l00186"></a><span class="lineno"> 186</span> {</div><div class="line"><a name="l00187"></a><span class="lineno"> 187</span>  <a class="code" href="struct_tensor4_d.xhtml">Tensor4D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</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#a989ab3e96426615bb98e04e0235088ca">src</a>, SRC_DIM_Z);</div><div class="line"><a name="l00188"></a><span class="lineno"> 188</span> </div><div class="line"><a name="l00189"></a><span class="lineno"> 189</span>  <span class="keyword">const</span> __global uchar *src_addr = <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#a989ab3e96426615bb98e04e0235088ca">src</a>, 0, 0, 0, 0);</div><div class="line"><a name="l00190"></a><span class="lineno"> 190</span> </div><div class="line"><a name="l00191"></a><span class="lineno"> 191</span>  <span class="comment">// Load the values from the input tensor</span></div><div class="line"><a name="l00192"></a><span class="lineno"> 192</span> <span class="preprocessor">#if defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL)</span></div><div class="line"><a name="l00193"></a><span class="lineno"> 193</span>  <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>, 3)</div><div class="line"><a name="l00194"></a><span class="lineno"> 194</span>  w0 = vload3(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr));</div><div class="line"><a name="l00195"></a><span class="lineno"> 195</span> <span class="preprocessor">#elif defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00196"></a><span class="lineno"> 196</span>  <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>, 3)</div><div class="line"><a name="l00197"></a><span class="lineno"> 197</span>  w0 = (<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>, 3))(*((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_y)),</div><div class="line"><a name="l00198"></a><span class="lineno"> 198</span>  *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_y)),</div><div class="line"><a name="l00199"></a><span class="lineno"> 199</span>  *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_y)));</div><div class="line"><a name="l00200"></a><span class="lineno"> 200</span> <span class="preprocessor">#else // defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00201"></a><span class="lineno"> 201</span>  <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>, 3)</div><div class="line"><a name="l00202"></a><span class="lineno"> 202</span>  w0 = vload3(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_y));</div><div class="line"><a name="l00203"></a><span class="lineno"> 203</span>  <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>, 3)</div><div class="line"><a name="l00204"></a><span class="lineno"> 204</span>  w1 = vload3(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_y));</div><div class="line"><a name="l00205"></a><span class="lineno"> 205</span>  <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>, 3)</div><div class="line"><a name="l00206"></a><span class="lineno"> 206</span>  w2 = vload3(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_y));</div><div class="line"><a name="l00207"></a><span class="lineno"> 207</span> <span class="preprocessor">#endif // defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL)</span></div><div class="line"><a name="l00208"></a><span class="lineno"> 208</span> </div><div class="line"><a name="l00209"></a><span class="lineno"> 209</span>  <span class="comment">// Row 0</span></div><div class="line"><a name="l00210"></a><span class="lineno"> 210</span>  <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)</div><div class="line"><a name="l00211"></a><span class="lineno"> 211</span>  out0 = 0.0f;</div><div class="line"><a name="l00212"></a><span class="lineno"> 212</span>  out0.s0 = (w0.s0) / 16.f;</div><div class="line"><a name="l00213"></a><span class="lineno"> 213</span>  out0.s1 = (-w0.s0 - w0.s1 - w0.s2) / 24.f;</div><div class="line"><a name="l00214"></a><span class="lineno"> 214</span>  out0.s2 = (-w0.s0 + w0.s1 - w0.s2) / 24.f;</div><div class="line"><a name="l00215"></a><span class="lineno"> 215</span>  out0.s3 = (w0.s0 + 2.f * w0.s1 + 4.f * w0.s2) / 96.f;</div><div class="line"><a name="l00216"></a><span class="lineno"> 216</span>  out0.s4 = (w0.s0 - 2.f * w0.s1 + 4.f * w0.s2) / 96.f;</div><div class="line"><a name="l00217"></a><span class="lineno"> 217</span>  out0.s5 = (w0.s2) / 4.f;</div><div class="line"><a name="l00218"></a><span class="lineno"> 218</span> </div><div class="line"><a name="l00219"></a><span class="lineno"> 219</span> <span class="preprocessor">#if !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00220"></a><span class="lineno"> 220</span>  <span class="comment">// Row 1</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>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 8)</div><div class="line"><a name="l00222"></a><span class="lineno"> 222</span>  out1 = 0.0f;</div><div class="line"><a name="l00223"></a><span class="lineno"> 223</span>  out1.s0 = (-w0.s0 - w1.s0 - w2.s0) / 24.f;</div><div class="line"><a name="l00224"></a><span class="lineno"> 224</span>  out1.s1 = (w0.s0 + w1.s0 + w2.s0 + w0.s1 + w1.s1 + w2.s1 + w0.s2 + w1.s2 + w2.s2) / 36.f;</div><div class="line"><a name="l00225"></a><span class="lineno"> 225</span>  out1.s2 = (w0.s0 + w1.s0 + w2.s0 - w0.s1 - w1.s1 - w2.s1 + w0.s2 + w1.s2 + w2.s2) / 36.f;</div><div class="line"><a name="l00226"></a><span class="lineno"> 226</span>  out1.s3 = (-w0.s0 - w1.s0 - w2.s0 + 2.f * (-w0.s1 - w1.s1 - w2.s1) + 4.f * (-w0.s2 - w1.s2 - w2.s2)) / 144.f;</div><div class="line"><a name="l00227"></a><span class="lineno"> 227</span>  out1.s4 = (-w0.s0 - w1.s0 - w2.s0 + 2.f * (w0.s1 + w1.s1 + w2.s1) + 4.f * (-w0.s2 - w1.s2 - w2.s2)) / 144.f;</div><div class="line"><a name="l00228"></a><span class="lineno"> 228</span>  out1.s5 = (-w0.s2 - w1.s2 - w2.s2) / 6.f;</div><div class="line"><a name="l00229"></a><span class="lineno"> 229</span> </div><div class="line"><a name="l00230"></a><span class="lineno"> 230</span>  <span class="comment">// Row 2</span></div><div class="line"><a name="l00231"></a><span class="lineno"> 231</span>  <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)</div><div class="line"><a name="l00232"></a><span class="lineno"> 232</span>  out2 = 0.0f;</div><div class="line"><a name="l00233"></a><span class="lineno"> 233</span>  out2.s0 = (-w0.s0 + w1.s0 - w2.s0) / 24.f;</div><div class="line"><a name="l00234"></a><span class="lineno"> 234</span>  out2.s1 = (w0.s0 - w1.s0 + w2.s0 + w0.s1 - w1.s1 + w2.s1 + w0.s2 - w1.s2 + w2.s2) / 36.f;</div><div class="line"><a name="l00235"></a><span class="lineno"> 235</span>  out2.s2 = (w0.s0 - w1.s0 + w2.s0 - w0.s1 + w1.s1 - w2.s1 + w0.s2 - w1.s2 + w2.s2) / 36.f;</div><div class="line"><a name="l00236"></a><span class="lineno"> 236</span>  out2.s3 = (-w0.s0 + w1.s0 - w2.s0 + 2.f * (-w0.s1 + w1.s1 - w2.s1) + 4.f * (-w0.s2 + w1.s2 - w2.s2)) / 144.f;</div><div class="line"><a name="l00237"></a><span class="lineno"> 237</span>  out2.s4 = (-w0.s0 + w1.s0 - w2.s0 + 2.f * (w0.s1 - w1.s1 + w2.s1) + 4.f * (-w0.s2 + w1.s2 - w2.s2)) / 144.f;</div><div class="line"><a name="l00238"></a><span class="lineno"> 238</span>  out2.s5 = (-w0.s2 + w1.s2 - w2.s2) / 6.f;</div><div class="line"><a name="l00239"></a><span class="lineno"> 239</span> </div><div class="line"><a name="l00240"></a><span class="lineno"> 240</span>  <span class="comment">// Row 3</span></div><div class="line"><a name="l00241"></a><span class="lineno"> 241</span>  <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)</div><div class="line"><a name="l00242"></a><span class="lineno"> 242</span>  out3 = 0.0f;</div><div class="line"><a name="l00243"></a><span class="lineno"> 243</span>  out3.s0 = (w0.s0 + 2.f * w1.s0 + 4.f * w2.s0) / 96.f;</div><div class="line"><a name="l00244"></a><span class="lineno"> 244</span>  out3.s1 = (-w0.s0 - 2.f * w1.s0 - 4.f * w2.s0 - w0.s1 - 2.f * w1.s1 - 4.f * w2.s1 - w0.s2 - 2.f * w1.s2 - 4.f * w2.s2) / 144.f;</div><div class="line"><a name="l00245"></a><span class="lineno"> 245</span>  out3.s2 = (-w0.s0 - 2.f * w1.s0 - 4.f * w2.s0 + w0.s1 + 2.f * w1.s1 + 4.f * w2.s1 - w0.s2 - 2.f * w1.s2 - 4.f * w2.s2) / 144.f;</div><div class="line"><a name="l00246"></a><span class="lineno"> 246</span>  out3.s3 = ((w0.s0 + 2.f * w1.s0 + 4.f * w2.s0) + 2.f * (w0.s1 + 2.f * w1.s1 + 4.f * w2.s1) + 4.f * (w0.s2 + 2.f * w1.s2 + 4.f * w2.s2)) / 576.f;</div><div class="line"><a name="l00247"></a><span class="lineno"> 247</span>  out3.s4 = ((w0.s0 + 2.f * w1.s0 + 4.f * w2.s0) + 2.f * (-w0.s1 - 2.f * w1.s1 - 4.f * w2.s1) + 4.f * (w0.s2 + 2.f * w1.s2 + 4.f * w2.s2)) / 576.f;</div><div class="line"><a name="l00248"></a><span class="lineno"> 248</span>  out3.s5 = (w0.s2 + 2.f * w1.s2 + 4.f * w2.s2) / 24.f;</div><div class="line"><a name="l00249"></a><span class="lineno"> 249</span> </div><div class="line"><a name="l00250"></a><span class="lineno"> 250</span>  <span class="comment">// Row 4</span></div><div class="line"><a name="l00251"></a><span class="lineno"> 251</span>  <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)</div><div class="line"><a name="l00252"></a><span class="lineno"> 252</span>  out4 = 0.0f;</div><div class="line"><a name="l00253"></a><span class="lineno"> 253</span>  out4.s0 = (w0.s0 - 2.f * w1.s0 + 4.f * w2.s0) / 96.f;</div><div class="line"><a name="l00254"></a><span class="lineno"> 254</span>  out4.s1 = (-w0.s0 + 2.f * w1.s0 - 4.f * w2.s0 - w0.s1 + 2.f * w1.s1 - 4.f * w2.s1 - w0.s2 + 2.f * w1.s2 - 4.f * w2.s2) / 144.f;</div><div class="line"><a name="l00255"></a><span class="lineno"> 255</span>  out4.s2 = (-w0.s0 + 2.f * w1.s0 - 4.f * w2.s0 + w0.s1 - 2.f * w1.s1 + 4.f * w2.s1 - w0.s2 + 2.f * w1.s2 - 4.f * w2.s2) / 144.f;</div><div class="line"><a name="l00256"></a><span class="lineno"> 256</span>  out4.s3 = ((w0.s0 - 2.f * w1.s0 + 4.f * w2.s0) + 2.f * (w0.s1 - 2.f * w1.s1 + 4.f * w2.s1) + 4.f * (w0.s2 - 2.f * w1.s2 + 4.f * w2.s2)) / 576.f;</div><div class="line"><a name="l00257"></a><span class="lineno"> 257</span>  out4.s4 = ((w0.s0 - 2.f * w1.s0 + 4.f * w2.s0) + 2.f * (-w0.s1 + 2.f * w1.s1 - 4.f * w2.s1) + 4.f * (w0.s2 - 2.f * w1.s2 + 4.f * w2.s2)) / 576.f;</div><div class="line"><a name="l00258"></a><span class="lineno"> 258</span>  out4.s5 = (w0.s2 - 2.f * w1.s2 + 4.f * w2.s2) / 24.f;</div><div class="line"><a name="l00259"></a><span class="lineno"> 259</span> </div><div class="line"><a name="l00260"></a><span class="lineno"> 260</span>  <span class="comment">// Row 5</span></div><div class="line"><a name="l00261"></a><span class="lineno"> 261</span>  <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)</div><div class="line"><a name="l00262"></a><span class="lineno"> 262</span>  out5 = 0.0f;</div><div class="line"><a name="l00263"></a><span class="lineno"> 263</span>  out5.s0 = (w2.s0) / 4.f;</div><div class="line"><a name="l00264"></a><span class="lineno"> 264</span>  out5.s1 = (-w2.s0 - w2.s1 - w2.s2) / 6.f;</div><div class="line"><a name="l00265"></a><span class="lineno"> 265</span>  out5.s2 = (-w2.s0 + w2.s1 - w2.s2) / 6.f;</div><div class="line"><a name="l00266"></a><span class="lineno"> 266</span>  out5.s3 = (w2.s0 + 2.f * w2.s1 + 4.f * w2.s2) / 24.f;</div><div class="line"><a name="l00267"></a><span class="lineno"> 267</span>  out5.s4 = (w2.s0 - 2.f * w2.s1 + 4.f * w2.s2) / 24.f;</div><div class="line"><a name="l00268"></a><span class="lineno"> 268</span>  out5.s5 = (w2.s2);</div><div class="line"><a name="l00269"></a><span class="lineno"> 269</span> <span class="preprocessor">#endif // !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00270"></a><span class="lineno"> 270</span> </div><div class="line"><a name="l00271"></a><span class="lineno"> 271</span>  <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l00272"></a><span class="lineno"> 272</span>  <span class="keywordtype">int</span> x0 = z / SRC_DIM_Z; <span class="comment">// idx filter</span></div><div class="line"><a name="l00273"></a><span class="lineno"> 273</span>  <span class="keywordtype">int</span> y0 = z % SRC_DIM_Z; <span class="comment">// idx channel</span></div><div class="line"><a name="l00274"></a><span class="lineno"> 274</span> </div><div class="line"><a name="l00275"></a><span class="lineno"> 275</span>  <span class="comment">// Get output address</span></div><div class="line"><a name="l00276"></a><span class="lineno"> 276</span>  __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x0 * dst_stride_x + y0 * dst_stride_y;</div><div class="line"><a name="l00277"></a><span class="lineno"> 277</span> </div><div class="line"><a name="l00278"></a><span class="lineno"> 278</span>  <span class="comment">// Store the values across the channels</span></div><div class="line"><a name="l00279"></a><span class="lineno"> 279</span>  <span class="comment">// 36 channels for 3x3 kernels</span></div><div class="line"><a name="l00280"></a><span class="lineno"> 280</span>  <span class="comment">// 6 channels for 3x1 or 1x3 kernels</span></div><div class="line"><a name="l00281"></a><span class="lineno"> 281</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 0 * dst_stride_z) = out0.s0;</div><div class="line"><a name="l00282"></a><span class="lineno"> 282</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 1 * dst_stride_z) = out0.s1;</div><div class="line"><a name="l00283"></a><span class="lineno"> 283</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 2 * dst_stride_z) = out0.s2;</div><div class="line"><a name="l00284"></a><span class="lineno"> 284</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 3 * dst_stride_z) = out0.s3;</div><div class="line"><a name="l00285"></a><span class="lineno"> 285</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 4 * dst_stride_z) = out0.s4;</div><div class="line"><a name="l00286"></a><span class="lineno"> 286</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 5 * dst_stride_z) = out0.s5;</div><div class="line"><a name="l00287"></a><span class="lineno"> 287</span> </div><div class="line"><a name="l00288"></a><span class="lineno"> 288</span> <span class="preprocessor">#if !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00289"></a><span class="lineno"> 289</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 6 * dst_stride_z) = out1.s0;</div><div class="line"><a name="l00290"></a><span class="lineno"> 290</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 7 * dst_stride_z) = out1.s1;</div><div class="line"><a name="l00291"></a><span class="lineno"> 291</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 8 * dst_stride_z) = out1.s2;</div><div class="line"><a name="l00292"></a><span class="lineno"> 292</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 9 * dst_stride_z) = out1.s3;</div><div class="line"><a name="l00293"></a><span class="lineno"> 293</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 10 * dst_stride_z) = out1.s4;</div><div class="line"><a name="l00294"></a><span class="lineno"> 294</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 11 * dst_stride_z) = out1.s5;</div><div class="line"><a name="l00295"></a><span class="lineno"> 295</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 12 * dst_stride_z) = out2.s0;</div><div class="line"><a name="l00296"></a><span class="lineno"> 296</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 13 * dst_stride_z) = out2.s1;</div><div class="line"><a name="l00297"></a><span class="lineno"> 297</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 14 * dst_stride_z) = out2.s2;</div><div class="line"><a name="l00298"></a><span class="lineno"> 298</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 15 * dst_stride_z) = out2.s3;</div><div class="line"><a name="l00299"></a><span class="lineno"> 299</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 16 * dst_stride_z) = out2.s4;</div><div class="line"><a name="l00300"></a><span class="lineno"> 300</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 17 * dst_stride_z) = out2.s5;</div><div class="line"><a name="l00301"></a><span class="lineno"> 301</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 18 * dst_stride_z) = out3.s0;</div><div class="line"><a name="l00302"></a><span class="lineno"> 302</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 19 * dst_stride_z) = out3.s1;</div><div class="line"><a name="l00303"></a><span class="lineno"> 303</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 20 * dst_stride_z) = out3.s2;</div><div class="line"><a name="l00304"></a><span class="lineno"> 304</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 21 * dst_stride_z) = out3.s3;</div><div class="line"><a name="l00305"></a><span class="lineno"> 305</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 22 * dst_stride_z) = out3.s4;</div><div class="line"><a name="l00306"></a><span class="lineno"> 306</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 23 * dst_stride_z) = out3.s5;</div><div class="line"><a name="l00307"></a><span class="lineno"> 307</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 24 * dst_stride_z) = out4.s0;</div><div class="line"><a name="l00308"></a><span class="lineno"> 308</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 25 * dst_stride_z) = out4.s1;</div><div class="line"><a name="l00309"></a><span class="lineno"> 309</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 26 * dst_stride_z) = out4.s2;</div><div class="line"><a name="l00310"></a><span class="lineno"> 310</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 27 * dst_stride_z) = out4.s3;</div><div class="line"><a name="l00311"></a><span class="lineno"> 311</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 28 * dst_stride_z) = out4.s4;</div><div class="line"><a name="l00312"></a><span class="lineno"> 312</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 29 * dst_stride_z) = out4.s5;</div><div class="line"><a name="l00313"></a><span class="lineno"> 313</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 30 * dst_stride_z) = out5.s0;</div><div class="line"><a name="l00314"></a><span class="lineno"> 314</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 31 * dst_stride_z) = out5.s1;</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> *)(dst_addr + 32 * dst_stride_z) = out5.s2;</div><div class="line"><a name="l00316"></a><span class="lineno"> 316</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 33 * dst_stride_z) = out5.s3;</div><div class="line"><a name="l00317"></a><span class="lineno"> 317</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 34 * dst_stride_z) = out5.s4;</div><div class="line"><a name="l00318"></a><span class="lineno"> 318</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 35 * dst_stride_z) = out5.s5;</div><div class="line"><a name="l00319"></a><span class="lineno"> 319</span> <span class="preprocessor">#endif // !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00320"></a><span class="lineno"> 320</span> }</div><div class="line"><a name="l00321"></a><span class="lineno"> 321</span> <span class="comment"></span></div><div class="line"><a name="l00322"></a><span class="lineno"> 322</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 3x3/3x1/1x3 when the data layout is NHWC and the output tile is 4x4/4x1/1x4</span></div><div class="line"><a name="l00323"></a><span class="lineno"> 323</span> <span class="comment"> *</span></div><div class="line"><a name="l00324"></a><span class="lineno"> 324</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l00325"></a><span class="lineno"> 325</span> <span class="comment"> * @note If this kernel is used to perform Winograd filter transform 3x1, -DWINOGRAD_FILTER_TRANSFORM_HORIZONTAL has to be passed at compile time</span></div><div class="line"><a name="l00326"></a><span class="lineno"> 326</span> <span class="comment"> * @note If this kernel is used to perform Winograd filter transform 1x3, -DWINOGRAD_FILTER_TRANSFORM_VERTICAL has to be passed at compile time</span></div><div class="line"><a name="l00327"></a><span class="lineno"> 327</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l00328"></a><span class="lineno"> 328</span> <span class="comment"> *</span></div><div class="line"><a name="l00329"></a><span class="lineno"> 329</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l00330"></a><span class="lineno"> 330</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00331"></a><span class="lineno"> 331</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00332"></a><span class="lineno"> 332</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00333"></a><span class="lineno"> 333</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00334"></a><span class="lineno"> 334</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00335"></a><span class="lineno"> 335</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00336"></a><span class="lineno"> 336</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l00337"></a><span class="lineno"> 337</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l00338"></a><span class="lineno"> 338</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l00339"></a><span class="lineno"> 339</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l00340"></a><span class="lineno"> 340</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00341"></a><span class="lineno"> 341</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00342"></a><span class="lineno"> 342</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00343"></a><span class="lineno"> 343</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00344"></a><span class="lineno"> 344</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00345"></a><span class="lineno"> 345</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00346"></a><span class="lineno"> 346</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l00347"></a><span class="lineno"> 347</span> <span class="comment"> */</span></div><div class="line"><a name="l00348"></a><span class="lineno"> 348</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_4x4_3x3_nhwc(</div><div class="line"><a name="l00349"></a><span class="lineno"> 349</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l00350"></a><span class="lineno"> 350</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l00351"></a><span class="lineno"> 351</span> {</div><div class="line"><a name="l00352"></a><span class="lineno"> 352</span>  <a class="code" href="struct_tensor4_d.xhtml">Tensor4D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</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#a989ab3e96426615bb98e04e0235088ca">src</a>, SRC_DIM_Z);</div><div class="line"><a name="l00353"></a><span class="lineno"> 353</span> </div><div class="line"><a name="l00354"></a><span class="lineno"> 354</span>  <span class="keyword">const</span> __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + get_global_id(0) * src_step_x + get_global_id(1) * src_step_y + get_global_id(2) * src_step_w;</div><div class="line"><a name="l00355"></a><span class="lineno"> 355</span> </div><div class="line"><a name="l00356"></a><span class="lineno"> 356</span>  <span class="comment">// Load the values from the input tensor</span></div><div class="line"><a name="l00357"></a><span class="lineno"> 357</span> <span class="preprocessor">#if defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00358"></a><span class="lineno"> 358</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w00 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_z));</div><div class="line"><a name="l00359"></a><span class="lineno"> 359</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w01 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z));</div><div class="line"><a name="l00360"></a><span class="lineno"> 360</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w02 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z));</div><div class="line"><a name="l00361"></a><span class="lineno"> 361</span> <span class="preprocessor">#else // defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00362"></a><span class="lineno"> 362</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w00 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_z + 0 * src_stride_y));</div><div class="line"><a name="l00363"></a><span class="lineno"> 363</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w01 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_z + 1 * src_stride_y));</div><div class="line"><a name="l00364"></a><span class="lineno"> 364</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w02 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_z + 2 * src_stride_y));</div><div class="line"><a name="l00365"></a><span class="lineno"> 365</span> <span class="preprocessor">#if !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL)</span></div><div class="line"><a name="l00366"></a><span class="lineno"> 366</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w10 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z + 0 * src_stride_y));</div><div class="line"><a name="l00367"></a><span class="lineno"> 367</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w11 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z + 1 * src_stride_y));</div><div class="line"><a name="l00368"></a><span class="lineno"> 368</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w12 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z + 2 * src_stride_y));</div><div class="line"><a name="l00369"></a><span class="lineno"> 369</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w20 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z + 0 * src_stride_y));</div><div class="line"><a name="l00370"></a><span class="lineno"> 370</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w21 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z + 1 * src_stride_y));</div><div class="line"><a name="l00371"></a><span class="lineno"> 371</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w22 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z + 2 * src_stride_y));</div><div class="line"><a name="l00372"></a><span class="lineno"> 372</span> <span class="preprocessor">#endif // !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL)</span></div><div class="line"><a name="l00373"></a><span class="lineno"> 373</span> <span class="preprocessor">#endif // defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></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">// Row 0</span></div><div class="line"><a name="l00376"></a><span class="lineno"> 376</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> out00, out01, out02, out03, out04, out05;</div><div class="line"><a name="l00377"></a><span class="lineno"> 377</span>  out00 = (w00) / 16.f;</div><div class="line"><a name="l00378"></a><span class="lineno"> 378</span>  out01 = (-w00 - w01 - w02) / 24.f;</div><div class="line"><a name="l00379"></a><span class="lineno"> 379</span>  out02 = (-w00 + w01 - w02) / 24.f;</div><div class="line"><a name="l00380"></a><span class="lineno"> 380</span>  out03 = (w00 + 2.f * w01 + 4.f * w02) / 96.f;</div><div class="line"><a name="l00381"></a><span class="lineno"> 381</span>  out04 = (w00 - 2.f * w01 + 4.f * w02) / 96.f;</div><div class="line"><a name="l00382"></a><span class="lineno"> 382</span>  out05 = (w02) / 4.f;</div><div class="line"><a name="l00383"></a><span class="lineno"> 383</span> </div><div class="line"><a name="l00384"></a><span class="lineno"> 384</span> <span class="preprocessor">#if !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00385"></a><span class="lineno"> 385</span>  <span class="comment">// Row 1</span></div><div class="line"><a name="l00386"></a><span class="lineno"> 386</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> out10, out11, out12, out13, out14, out15;</div><div class="line"><a name="l00387"></a><span class="lineno"> 387</span>  out10 = (-w00 - w10 - w20) / 24.f;</div><div class="line"><a name="l00388"></a><span class="lineno"> 388</span>  out11 = (w00 + w10 + w20 + w01 + w11 + w21 + w02 + w12 + w22) / 36.f;</div><div class="line"><a name="l00389"></a><span class="lineno"> 389</span>  out12 = (w00 + w10 + w20 - w01 - w11 - w21 + w02 + w12 + w22) / 36.f;</div><div class="line"><a name="l00390"></a><span class="lineno"> 390</span>  out13 = (-w00 - w10 - w20 + 2.f * (-w01 - w11 - w21) + 4.f * (-w02 - w12 - w22)) / 144.f;</div><div class="line"><a name="l00391"></a><span class="lineno"> 391</span>  out14 = (-w00 - w10 - w20 + 2.f * (w01 + w11 + w21) + 4.f * (-w02 - w12 - w22)) / 144.f;</div><div class="line"><a name="l00392"></a><span class="lineno"> 392</span>  out15 = (-w02 - w12 - w22) / 6.f;</div><div class="line"><a name="l00393"></a><span class="lineno"> 393</span> </div><div class="line"><a name="l00394"></a><span class="lineno"> 394</span>  <span class="comment">// Row 2</span></div><div class="line"><a name="l00395"></a><span class="lineno"> 395</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> out20, out21, out22, out23, out24, out25;</div><div class="line"><a name="l00396"></a><span class="lineno"> 396</span>  out20 = (-w00 + w10 - w20) / 24.f;</div><div class="line"><a name="l00397"></a><span class="lineno"> 397</span>  out21 = (w00 - w10 + w20 + w01 - w11 + w21 + w02 - w12 + w22) / 36.f;</div><div class="line"><a name="l00398"></a><span class="lineno"> 398</span>  out22 = (w00 - w10 + w20 - w01 + w11 - w21 + w02 - w12 + w22) / 36.f;</div><div class="line"><a name="l00399"></a><span class="lineno"> 399</span>  out23 = (-w00 + w10 - w20 + 2.f * (-w01 + w11 - w21) + 4.f * (-w02 + w12 - w22)) / 144.f;</div><div class="line"><a name="l00400"></a><span class="lineno"> 400</span>  out24 = (-w00 + w10 - w20 + 2.f * (w01 - w11 + w21) + 4.f * (-w02 + w12 - w22)) / 144.f;</div><div class="line"><a name="l00401"></a><span class="lineno"> 401</span>  out25 = (-w02 + w12 - w22) / 6.f;</div><div class="line"><a name="l00402"></a><span class="lineno"> 402</span> </div><div class="line"><a name="l00403"></a><span class="lineno"> 403</span>  <span class="comment">// Row 3</span></div><div class="line"><a name="l00404"></a><span class="lineno"> 404</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> out30, out31, out32, out33, out34, out35;</div><div class="line"><a name="l00405"></a><span class="lineno"> 405</span>  out30 = (w00 + 2.f * w10 + 4.f * w20) / 96.f;</div><div class="line"><a name="l00406"></a><span class="lineno"> 406</span>  out31 = (-w00 - 2.f * w10 - 4.f * w20 - w01 - 2.f * w11 - 4.f * w21 - w02 - 2.f * w12 - 4.f * w22) / 144.f;</div><div class="line"><a name="l00407"></a><span class="lineno"> 407</span>  out32 = (-w00 - 2.f * w10 - 4.f * w20 + w01 + 2.f * w11 + 4.f * w21 - w02 - 2.f * w12 - 4.f * w22) / 144.f;</div><div class="line"><a name="l00408"></a><span class="lineno"> 408</span>  out33 = ((w00 + 2.f * w10 + 4.f * w20) + 2.f * (w01 + 2.f * w11 + 4.f * w21) + 4.f * (w02 + 2.f * w12 + 4.f * w22)) / 576.f;</div><div class="line"><a name="l00409"></a><span class="lineno"> 409</span>  out34 = ((w00 + 2.f * w10 + 4.f * w20) + 2.f * (-w01 - 2.f * w11 - 4.f * w21) + 4.f * (w02 + 2.f * w12 + 4.f * w22)) / 576.f;</div><div class="line"><a name="l00410"></a><span class="lineno"> 410</span>  out35 = (w02 + 2.f * w12 + 4.f * w22) / 24.f;</div><div class="line"><a name="l00411"></a><span class="lineno"> 411</span> </div><div class="line"><a name="l00412"></a><span class="lineno"> 412</span>  <span class="comment">// Row 4</span></div><div class="line"><a name="l00413"></a><span class="lineno"> 413</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> out40, out41, out42, out43, out44, out45;</div><div class="line"><a name="l00414"></a><span class="lineno"> 414</span>  out40 = (w00 - 2.f * w10 + 4.f * w20) / 96.f;</div><div class="line"><a name="l00415"></a><span class="lineno"> 415</span>  out41 = (-w00 + 2.f * w10 - 4.f * w20 - w01 + 2.f * w11 - 4.f * w21 - w02 + 2.f * w12 - 4.f * w22) / 144.f;</div><div class="line"><a name="l00416"></a><span class="lineno"> 416</span>  out42 = (-w00 + 2.f * w10 - 4.f * w20 + w01 - 2.f * w11 + 4.f * w21 - w02 + 2.f * w12 - 4.f * w22) / 144.f;</div><div class="line"><a name="l00417"></a><span class="lineno"> 417</span>  out43 = ((w00 - 2.f * w10 + 4.f * w20) + 2.f * (w01 - 2.f * w11 + 4.f * w21) + 4.f * (w02 - 2.f * w12 + 4.f * w22)) / 576.f;</div><div class="line"><a name="l00418"></a><span class="lineno"> 418</span>  out44 = ((w00 - 2.f * w10 + 4.f * w20) + 2.f * (-w01 + 2.f * w11 - 4.f * w21) + 4.f * (w02 - 2.f * w12 + 4.f * w22)) / 576.f;</div><div class="line"><a name="l00419"></a><span class="lineno"> 419</span>  out45 = (w02 - 2.f * w12 + 4.f * w22) / 24.f;</div><div class="line"><a name="l00420"></a><span class="lineno"> 420</span> </div><div class="line"><a name="l00421"></a><span class="lineno"> 421</span>  <span class="comment">// Row 5</span></div><div class="line"><a name="l00422"></a><span class="lineno"> 422</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> out50, out51, out52, out53, out54, out55;</div><div class="line"><a name="l00423"></a><span class="lineno"> 423</span>  out50 = (w20) / 4.f;</div><div class="line"><a name="l00424"></a><span class="lineno"> 424</span>  out51 = (-w20 - w21 - w22) / 6.f;</div><div class="line"><a name="l00425"></a><span class="lineno"> 425</span>  out52 = (-w20 + w21 - w22) / 6.f;</div><div class="line"><a name="l00426"></a><span class="lineno"> 426</span>  out53 = (w20 + 2.f * w21 + 4.f * w22) / 24.f;</div><div class="line"><a name="l00427"></a><span class="lineno"> 427</span>  out54 = (w20 - 2.f * w21 + 4.f * w22) / 24.f;</div><div class="line"><a name="l00428"></a><span class="lineno"> 428</span>  out55 = (w22);</div><div class="line"><a name="l00429"></a><span class="lineno"> 429</span> <span class="preprocessor">#endif // !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00430"></a><span class="lineno"> 430</span> </div><div class="line"><a name="l00431"></a><span class="lineno"> 431</span>  <span class="keywordtype">int</span> x0 = get_global_id(2); <span class="comment">// idx filter</span></div><div class="line"><a name="l00432"></a><span class="lineno"> 432</span>  <span class="keywordtype">int</span> y0 = get_global_id(0); <span class="comment">// idx channel</span></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">// Get output address</span></div><div class="line"><a name="l00435"></a><span class="lineno"> 435</span>  __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x0 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>) + y0 * dst_stride_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="comment">// Store the values across the channels</span></div><div class="line"><a name="l00438"></a><span class="lineno"> 438</span>  <span class="comment">// 36 channels for 3x3 kernels</span></div><div class="line"><a name="l00439"></a><span class="lineno"> 439</span>  <span class="comment">// 6 channels for 3x1 or 1x3 kernels</span></div><div class="line"><a name="l00440"></a><span class="lineno"> 440</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 0 * dst_stride_z) = out00;</div><div class="line"><a name="l00441"></a><span class="lineno"> 441</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 1 * dst_stride_z) = out01;</div><div class="line"><a name="l00442"></a><span class="lineno"> 442</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 2 * dst_stride_z) = out02;</div><div class="line"><a name="l00443"></a><span class="lineno"> 443</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 3 * dst_stride_z) = out03;</div><div class="line"><a name="l00444"></a><span class="lineno"> 444</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 4 * dst_stride_z) = out04;</div><div class="line"><a name="l00445"></a><span class="lineno"> 445</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 5 * dst_stride_z) = out05;</div><div class="line"><a name="l00446"></a><span class="lineno"> 446</span> <span class="preprocessor">#if !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00447"></a><span class="lineno"> 447</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 6 * dst_stride_z) = out10;</div><div class="line"><a name="l00448"></a><span class="lineno"> 448</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 7 * dst_stride_z) = out11;</div><div class="line"><a name="l00449"></a><span class="lineno"> 449</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 8 * dst_stride_z) = out12;</div><div class="line"><a name="l00450"></a><span class="lineno"> 450</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 9 * dst_stride_z) = out13;</div><div class="line"><a name="l00451"></a><span class="lineno"> 451</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 10 * dst_stride_z) = out14;</div><div class="line"><a name="l00452"></a><span class="lineno"> 452</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 11 * dst_stride_z) = out15;</div><div class="line"><a name="l00453"></a><span class="lineno"> 453</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 12 * dst_stride_z) = out20;</div><div class="line"><a name="l00454"></a><span class="lineno"> 454</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 13 * dst_stride_z) = out21;</div><div class="line"><a name="l00455"></a><span class="lineno"> 455</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 14 * dst_stride_z) = out22;</div><div class="line"><a name="l00456"></a><span class="lineno"> 456</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 15 * dst_stride_z) = out23;</div><div class="line"><a name="l00457"></a><span class="lineno"> 457</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 16 * dst_stride_z) = out24;</div><div class="line"><a name="l00458"></a><span class="lineno"> 458</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 17 * dst_stride_z) = out25;</div><div class="line"><a name="l00459"></a><span class="lineno"> 459</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 18 * dst_stride_z) = out30;</div><div class="line"><a name="l00460"></a><span class="lineno"> 460</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 19 * dst_stride_z) = out31;</div><div class="line"><a name="l00461"></a><span class="lineno"> 461</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 20 * dst_stride_z) = out32;</div><div class="line"><a name="l00462"></a><span class="lineno"> 462</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 21 * dst_stride_z) = out33;</div><div class="line"><a name="l00463"></a><span class="lineno"> 463</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 22 * dst_stride_z) = out34;</div><div class="line"><a name="l00464"></a><span class="lineno"> 464</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 23 * dst_stride_z) = out35;</div><div class="line"><a name="l00465"></a><span class="lineno"> 465</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 24 * dst_stride_z) = out40;</div><div class="line"><a name="l00466"></a><span class="lineno"> 466</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 25 * dst_stride_z) = out41;</div><div class="line"><a name="l00467"></a><span class="lineno"> 467</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 26 * dst_stride_z) = out42;</div><div class="line"><a name="l00468"></a><span class="lineno"> 468</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 27 * dst_stride_z) = out43;</div><div class="line"><a name="l00469"></a><span class="lineno"> 469</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 28 * dst_stride_z) = out44;</div><div class="line"><a name="l00470"></a><span class="lineno"> 470</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 29 * dst_stride_z) = out45;</div><div class="line"><a name="l00471"></a><span class="lineno"> 471</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 30 * dst_stride_z) = out50;</div><div class="line"><a name="l00472"></a><span class="lineno"> 472</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 31 * dst_stride_z) = out51;</div><div class="line"><a name="l00473"></a><span class="lineno"> 473</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 32 * dst_stride_z) = out52;</div><div class="line"><a name="l00474"></a><span class="lineno"> 474</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 33 * dst_stride_z) = out53;</div><div class="line"><a name="l00475"></a><span class="lineno"> 475</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 34 * dst_stride_z) = out54;</div><div class="line"><a name="l00476"></a><span class="lineno"> 476</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 35 * dst_stride_z) = out55;</div><div class="line"><a name="l00477"></a><span class="lineno"> 477</span> <span class="preprocessor">#endif // !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</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"></span></div><div class="line"><a name="l00480"></a><span class="lineno"> 480</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 5x5/5x1 or 1x5 when the data layout is NCHW and the output tile is 4x4/4x1 or 1x4</span></div><div class="line"><a name="l00481"></a><span class="lineno"> 481</span> <span class="comment"> *</span></div><div class="line"><a name="l00482"></a><span class="lineno"> 482</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l00483"></a><span class="lineno"> 483</span> <span class="comment"> *</span></div><div class="line"><a name="l00484"></a><span class="lineno"> 484</span> <span class="comment"> * @note If this kernel is used to perform Winograd filter transform 5x1, -DWINOGRAD_FILTER_TRANSFORM_HORIZONTAL has to be passed at compile time</span></div><div class="line"><a name="l00485"></a><span class="lineno"> 485</span> <span class="comment"> * @note If this kernel is used to perform Winograd filter transform 1x5, -DWINOGRAD_FILTER_TRANSFORM_VERTICAL has to be passed at compile time</span></div><div class="line"><a name="l00486"></a><span class="lineno"> 486</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l00487"></a><span class="lineno"> 487</span> <span class="comment"> *</span></div><div class="line"><a name="l00488"></a><span class="lineno"> 488</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l00489"></a><span class="lineno"> 489</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00490"></a><span class="lineno"> 490</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00491"></a><span class="lineno"> 491</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00492"></a><span class="lineno"> 492</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00493"></a><span class="lineno"> 493</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00494"></a><span class="lineno"> 494</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00495"></a><span class="lineno"> 495</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l00496"></a><span class="lineno"> 496</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l00497"></a><span class="lineno"> 497</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l00498"></a><span class="lineno"> 498</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l00499"></a><span class="lineno"> 499</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00500"></a><span class="lineno"> 500</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00501"></a><span class="lineno"> 501</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00502"></a><span class="lineno"> 502</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00503"></a><span class="lineno"> 503</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00504"></a><span class="lineno"> 504</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00505"></a><span class="lineno"> 505</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</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> __kernel <span class="keywordtype">void</span> winograd_filter_transform_4x4_5x5_nchw(</div><div class="line"><a name="l00508"></a><span class="lineno"> 508</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l00509"></a><span class="lineno"> 509</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l00510"></a><span class="lineno"> 510</span> {</div><div class="line"><a name="l00511"></a><span class="lineno"> 511</span>  <a class="code" href="struct_tensor4_d.xhtml">Tensor4D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</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#a989ab3e96426615bb98e04e0235088ca">src</a>, SRC_DIM_Z);</div><div class="line"><a name="l00512"></a><span class="lineno"> 512</span> </div><div class="line"><a name="l00513"></a><span class="lineno"> 513</span>  <span class="keyword">const</span> __global uchar *src_addr = <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#a989ab3e96426615bb98e04e0235088ca">src</a>, 0, 0, 0, 0);</div><div class="line"><a name="l00514"></a><span class="lineno"> 514</span> </div><div class="line"><a name="l00515"></a><span class="lineno"> 515</span>  <span class="comment">// Load the values from the input tensor</span></div><div class="line"><a name="l00516"></a><span class="lineno"> 516</span> <span class="preprocessor">#if defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL)</span></div><div class="line"><a name="l00517"></a><span class="lineno"> 517</span>  <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)</div><div class="line"><a name="l00518"></a><span class="lineno"> 518</span>  w00 = vload4(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_y));</div><div class="line"><a name="l00519"></a><span class="lineno"> 519</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w01 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_y) + 4);</div><div class="line"><a name="l00520"></a><span class="lineno"> 520</span> <span class="preprocessor">#elif defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00521"></a><span class="lineno"> 521</span>  <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)</div><div class="line"><a name="l00522"></a><span class="lineno"> 522</span>  w00 = (<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))(*((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_y)),</div><div class="line"><a name="l00523"></a><span class="lineno"> 523</span>  *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_y)),</div><div class="line"><a name="l00524"></a><span class="lineno"> 524</span>  *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_y)),</div><div class="line"><a name="l00525"></a><span class="lineno"> 525</span>  *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_y)));</div><div class="line"><a name="l00526"></a><span class="lineno"> 526</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w01 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_y));</div><div class="line"><a name="l00527"></a><span class="lineno"> 527</span> <span class="preprocessor">#else // defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00528"></a><span class="lineno"> 528</span>  <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)</div><div class="line"><a name="l00529"></a><span class="lineno"> 529</span>  w00 = vload4(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_y));</div><div class="line"><a name="l00530"></a><span class="lineno"> 530</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w01 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_y) + 4);</div><div class="line"><a name="l00531"></a><span class="lineno"> 531</span>  <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)</div><div class="line"><a name="l00532"></a><span class="lineno"> 532</span>  w10 = vload4(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_y));</div><div class="line"><a name="l00533"></a><span class="lineno"> 533</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w11 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_y) + 4);</div><div class="line"><a name="l00534"></a><span class="lineno"> 534</span>  <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)</div><div class="line"><a name="l00535"></a><span class="lineno"> 535</span>  w20 = vload4(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_y));</div><div class="line"><a name="l00536"></a><span class="lineno"> 536</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w21 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_y) + 4);</div><div class="line"><a name="l00537"></a><span class="lineno"> 537</span>  <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)</div><div class="line"><a name="l00538"></a><span class="lineno"> 538</span>  w30 = vload4(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_y));</div><div class="line"><a name="l00539"></a><span class="lineno"> 539</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w31 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_y) + 4);</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#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 4)</div><div class="line"><a name="l00541"></a><span class="lineno"> 541</span>  w40 = vload4(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_y));</div><div class="line"><a name="l00542"></a><span class="lineno"> 542</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w41 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_y) + 4);</div><div class="line"><a name="l00543"></a><span class="lineno"> 543</span> <span class="preprocessor">#endif // defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL)</span></div><div class="line"><a name="l00544"></a><span class="lineno"> 544</span> </div><div class="line"><a name="l00545"></a><span class="lineno"> 545</span>  <span class="comment">// Transform the input tile</span></div><div class="line"><a name="l00546"></a><span class="lineno"> 546</span> </div><div class="line"><a name="l00547"></a><span class="lineno"> 547</span>  <span class="comment">// Row 0</span></div><div class="line"><a name="l00548"></a><span class="lineno"> 548</span>  <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)</div><div class="line"><a name="l00549"></a><span class="lineno"> 549</span>  out0 = 0.0f;</div><div class="line"><a name="l00550"></a><span class="lineno"> 550</span>  out0.s0 = w00.s0;</div><div class="line"><a name="l00551"></a><span class="lineno"> 551</span>  out0.s1 = -2.f * (w00.s0 + w00.s1 + w00.s2 + w00.s3 + w01) / 9.f;</div><div class="line"><a name="l00552"></a><span class="lineno"> 552</span>  out0.s2 = -2.f * (w00.s0 - w00.s1 + w00.s2 - w00.s3 + w01) / 9.f;</div><div class="line"><a name="l00553"></a><span class="lineno"> 553</span>  out0.s3 = (w00.s0 + 2.f * w00.s1 + 4.f * w00.s2 + 8.f * w00.s3 + 16.f * w01) / 90.f;</div><div class="line"><a name="l00554"></a><span class="lineno"> 554</span>  out0.s4 = (w00.s0 - 2.f * w00.s1 + 4.f * w00.s2 - 8.f * w00.s3 + 16.f * w01) / 90.f;</div><div class="line"><a name="l00555"></a><span class="lineno"> 555</span>  out0.s5 = (16.f * w00.s0 + 8.f * w00.s1 + 4.f * w00.s2 + 2.f * w00.s3 + w01) / 180.f;</div><div class="line"><a name="l00556"></a><span class="lineno"> 556</span>  out0.s6 = (16.f * w00.s0 - 8.f * w00.s1 + 4.f * w00.s2 - 2.f * w00.s3 + w01) / 180.f;</div><div class="line"><a name="l00557"></a><span class="lineno"> 557</span>  out0.s7 = w01;</div><div class="line"><a name="l00558"></a><span class="lineno"> 558</span> </div><div class="line"><a name="l00559"></a><span class="lineno"> 559</span> <span class="preprocessor">#if !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00560"></a><span class="lineno"> 560</span>  <span class="comment">// Row 1</span></div><div class="line"><a name="l00561"></a><span class="lineno"> 561</span>  <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)</div><div class="line"><a name="l00562"></a><span class="lineno"> 562</span>  out1 = 0.0f;</div><div class="line"><a name="l00563"></a><span class="lineno"> 563</span>  out1.s0 = -2.f * (w00.s0 + w10.s0 + w20.s0 + w30.s0 + w40.s0) / 9.f;</div><div class="line"><a name="l00564"></a><span class="lineno"> 564</span>  out1.s1 = 4.f * ((w00.s0 + w10.s0 + w20.s0 + w30.s0 + w40.s0) + (w00.s1 + w10.s1 + w20.s1 + w30.s1 + w40.s1) + (w00.s2 + w10.s2 + w20.s2 + w30.s2 + w40.s2) +</div><div class="line"><a name="l00565"></a><span class="lineno"> 565</span>  (w00.s3 + w10.s3 + w20.s3 + w30.s3 + w40.s3) + (w01 + w11 + w21 + w31 + w41)) / 81.f;</div><div class="line"><a name="l00566"></a><span class="lineno"> 566</span>  out1.s2 = 4.f * ((w00.s0 + w10.s0 + w20.s0 + w30.s0 + w40.s0) - (w00.s1 + w10.s1 + w20.s1 + w30.s1 + w40.s1) + (w00.s2 + w10.s2 + w20.s2 + w30.s2 + w40.s2) -</div><div class="line"><a name="l00567"></a><span class="lineno"> 567</span>  (w00.s3 + w10.s3 + w20.s3 + w30.s3 + w40.s3) + (w01 + w11 + w21 + w31 + w41)) / 81.f;</div><div class="line"><a name="l00568"></a><span class="lineno"> 568</span>  out1.s3 = -((w00.s0 + w10.s0 + w20.s0 + w30.s0 + w40.s0) + 2.f * (w00.s1 + w10.s1 + w20.s1 + w30.s1 + w40.s1) + 4.f * (w00.s2 + w10.s2 + w20.s2 + w30.s2 + w40.s2) + 8.f *</div><div class="line"><a name="l00569"></a><span class="lineno"> 569</span>  (w00.s3 + w10.s3 + w20.s3 + w30.s3 + w40.s3) + 16.f * (w01 + w11 + w21 + w31 + w41)) / 405.f;</div><div class="line"><a name="l00570"></a><span class="lineno"> 570</span>  out1.s4 = -((w00.s0 + w10.s0 + w20.s0 + w30.s0 + w40.s0) - 2.f * (w00.s1 + w10.s1 + w20.s1 + w30.s1 + w40.s1) + 4.f * (w00.s2 + w10.s2 + w20.s2 + w30.s2 + w40.s2) - 8.f *</div><div class="line"><a name="l00571"></a><span class="lineno"> 571</span>  (w00.s3 + w10.s3 + w20.s3 + w30.s3 + w40.s3) + 16.f * (w01 + w11 + w21 + w31 + w41)) / 405.f;</div><div class="line"><a name="l00572"></a><span class="lineno"> 572</span>  out1.s5 = -(16.f * (w00.s0 + w10.s0 + w20.s0 + w30.s0 + w40.s0) + 8.f * (w00.s1 + w10.s1 + w20.s1 + w30.s1 + w40.s1) + 4.f * (w00.s2 + w10.s2 + w20.s2 + w30.s2 + w40.s2) + 2.f *</div><div class="line"><a name="l00573"></a><span class="lineno"> 573</span>  (w00.s3 + w10.s3 + w20.s3 + w30.s3 + w40.s3) + (w01 + w11 + w21 + w31 + w41)) / 810.f;</div><div class="line"><a name="l00574"></a><span class="lineno"> 574</span>  out1.s6 = -(16.f * (w00.s0 + w10.s0 + w20.s0 + w30.s0 + w40.s0) - 8.f * (w00.s1 + w10.s1 + w20.s1 + w30.s1 + w40.s1) + 4.f * (w00.s2 + w10.s2 + w20.s2 + w30.s2 + w40.s2) - 2.f *</div><div class="line"><a name="l00575"></a><span class="lineno"> 575</span>  (w00.s3 + w10.s3 + w20.s3 + w30.s3 + w40.s3) + (w01 + w11 + w21 + w31 + w41)) / 810.f;</div><div class="line"><a name="l00576"></a><span class="lineno"> 576</span>  out1.s7 = -2.f * (w01 + w11 + w21 + w31 + w41) / 9.f;</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="comment">// Row 2</span></div><div class="line"><a name="l00579"></a><span class="lineno"> 579</span>  <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)</div><div class="line"><a name="l00580"></a><span class="lineno"> 580</span>  out2 = 0.0f;</div><div class="line"><a name="l00581"></a><span class="lineno"> 581</span>  out2.s0 = -2.f * (w00.s0 - w10.s0 + w20.s0 - w30.s0 + w40.s0) / 9.f;</div><div class="line"><a name="l00582"></a><span class="lineno"> 582</span>  out2.s1 = 4.f * ((w00.s0 - w10.s0 + w20.s0 - w30.s0 + w40.s0) + (w00.s1 - w10.s1 + w20.s1 - w30.s1 + w40.s1) + (w00.s2 - w10.s2 + w20.s2 - w30.s2 + w40.s2) +</div><div class="line"><a name="l00583"></a><span class="lineno"> 583</span>  (w00.s3 - w10.s3 + w20.s3 - w30.s3 + w40.s3) + (w01 - w11 + w21 - w31 + w41)) / 81.f;</div><div class="line"><a name="l00584"></a><span class="lineno"> 584</span>  out2.s2 = 4.f * ((w00.s0 - w10.s0 + w20.s0 - w30.s0 + w40.s0) - (w00.s1 - w10.s1 + w20.s1 - w30.s1 + w40.s1) + (w00.s2 - w10.s2 + w20.s2 - w30.s2 + w40.s2) -</div><div class="line"><a name="l00585"></a><span class="lineno"> 585</span>  (w00.s3 - w10.s3 + w20.s3 - w30.s3 + w40.s3) + (w01 - w11 + w21 - w31 + w41)) / 81.f;</div><div class="line"><a name="l00586"></a><span class="lineno"> 586</span>  out2.s3 = -((w00.s0 - w10.s0 + w20.s0 - w30.s0 + w40.s0) + 2.f * (w00.s1 - w10.s1 + w20.s1 - w30.s1 + w40.s1) + 4.f * (w00.s2 - w10.s2 + w20.s2 - w30.s2 + w40.s2) + 8.f *</div><div class="line"><a name="l00587"></a><span class="lineno"> 587</span>  (w00.s3 - w10.s3 + w20.s3 - w30.s3 + w40.s3) + 16.f * (w01 - w11 + w21 - w31 + w41)) / 405.f;</div><div class="line"><a name="l00588"></a><span class="lineno"> 588</span>  out2.s4 = -((w00.s0 - w10.s0 + w20.s0 - w30.s0 + w40.s0) - 2.f * (w00.s1 - w10.s1 + w20.s1 - w30.s1 + w40.s1) + 4.f * (w00.s2 - w10.s2 + w20.s2 - w30.s2 + w40.s2) - 8.f *</div><div class="line"><a name="l00589"></a><span class="lineno"> 589</span>  (w00.s3 - w10.s3 + w20.s3 - w30.s3 + w40.s3) + 16.f * (w01 - w11 + w21 - w31 + w41)) / 405.f;</div><div class="line"><a name="l00590"></a><span class="lineno"> 590</span>  out2.s5 = -(16.f * (w00.s0 - w10.s0 + w20.s0 - w30.s0 + w40.s0) + 8.f * (w00.s1 - w10.s1 + w20.s1 - w30.s1 + w40.s1) + 4.f * (w00.s2 - w10.s2 + w20.s2 - w30.s2 + w40.s2) + 2.f *</div><div class="line"><a name="l00591"></a><span class="lineno"> 591</span>  (w00.s3 - w10.s3 + w20.s3 - w30.s3 + w40.s3) + (w01 - w11 + w21 - w31 + w41)) / 810.f;</div><div class="line"><a name="l00592"></a><span class="lineno"> 592</span>  out2.s6 = -(16.f * (w00.s0 - w10.s0 + w20.s0 - w30.s0 + w40.s0) - 8.f * (w00.s1 - w10.s1 + w20.s1 - w30.s1 + w40.s1) + 4.f * (w00.s2 - w10.s2 + w20.s2 - w30.s2 + w40.s2) - 2.f *</div><div class="line"><a name="l00593"></a><span class="lineno"> 593</span>  (w00.s3 - w10.s3 + w20.s3 - w30.s3 + w40.s3) + (w01 - w11 + w21 - w31 + w41)) / 810.f;</div><div class="line"><a name="l00594"></a><span class="lineno"> 594</span>  out2.s7 = -2.f * (w01 - w11 + w21 - w31 + w41) / 9.f;</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">// Row 3</span></div><div class="line"><a name="l00597"></a><span class="lineno"> 597</span>  <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)</div><div class="line"><a name="l00598"></a><span class="lineno"> 598</span>  out3 = 0.0f;</div><div class="line"><a name="l00599"></a><span class="lineno"> 599</span>  out3.s0 = (w00.s0 + 2.f * w10.s0 + 4.f * w20.s0 + 8.f * w30.s0 + 16.f * w40.s0) / 90.f;</div><div class="line"><a name="l00600"></a><span class="lineno"> 600</span>  out3.s1 = -((w00.s0 + 2.f * w10.s0 + 4.f * w20.s0 + 8.f * w30.s0 + 16.f * w40.s0) + (w00.s1 + 2.f * w10.s1 + 4.f * w20.s1 + 8.f * w30.s1 + 16.f * w40.s1) +</div><div class="line"><a name="l00601"></a><span class="lineno"> 601</span>  (w00.s2 + 2.f * w10.s2 + 4.f * w20.s2 + 8.f * w30.s2 + 16.f * w40.s2) + (w00.s3 + 2.f * w10.s3 + 4.f * w20.s3 + 8.f * w30.s3 + 16.f * w40.s3) +</div><div class="line"><a name="l00602"></a><span class="lineno"> 602</span>  (w01 + 2.f * w11 + 4.f * w21 + 8.f * w31 + 16.f * w41)) / 405.f;</div><div class="line"><a name="l00603"></a><span class="lineno"> 603</span>  out3.s2 = -((w00.s0 + 2.f * w10.s0 + 4.f * w20.s0 + 8.f * w30.s0 + 16.f * w40.s0) - (w00.s1 + 2.f * w10.s1 + 4.f * w20.s1 + 8.f * w30.s1 + 16.f * w40.s1) +</div><div class="line"><a name="l00604"></a><span class="lineno"> 604</span>  (w00.s2 + 2.f * w10.s2 + 4.f * w20.s2 + 8.f * w30.s2 + 16.f * w40.s2) - (w00.s3 + 2.f * w10.s3 + 4.f * w20.s3 + 8.f * w30.s3 + 16.f * w40.s3) +</div><div class="line"><a name="l00605"></a><span class="lineno"> 605</span>  (w01 + 2.f * w11 + 4.f * w21 + 8.f * w31 + 16.f * w41)) / 405.f;</div><div class="line"><a name="l00606"></a><span class="lineno"> 606</span>  out3.s3 = ((w00.s0 + 2.f * w10.s0 + 4.f * w20.s0 + 8.f * w30.s0 + 16.f * w40.s0) + 2.f * (w00.s1 + 2.f * w10.s1 + 4.f * w20.s1 + 8.f * w30.s1 + 16.f * w40.s1) + 4.f *</div><div class="line"><a name="l00607"></a><span class="lineno"> 607</span>  (w00.s2 + 2.f * w10.s2 + 4.f * w20.s2 + 8.f * w30.s2 + 16.f * w40.s2) + 8.f * (w00.s3 + 2.f * w10.s3 + 4.f * w20.s3 + 8.f * w30.s3 + 16.f * w40.s3) + 16.f *</div><div class="line"><a name="l00608"></a><span class="lineno"> 608</span>  (w01 + 2.f * w11 + 4.f * w21 + 8.f * w31 + 16.f * w41)) / 8100.f;</div><div class="line"><a name="l00609"></a><span class="lineno"> 609</span>  out3.s4 = ((w00.s0 + 2.f * w10.s0 + 4.f * w20.s0 + 8.f * w30.s0 + 16.f * w40.s0) - 2.f * (w00.s1 + 2.f * w10.s1 + 4.f * w20.s1 + 8.f * w30.s1 + 16.f * w40.s1) + 4.f *</div><div class="line"><a name="l00610"></a><span class="lineno"> 610</span>  (w00.s2 + 2.f * w10.s2 + 4.f * w20.s2 + 8.f * w30.s2 + 16.f * w40.s2) - 8.f * (w00.s3 + 2.f * w10.s3 + 4.f * w20.s3 + 8.f * w30.s3 + 16.f * w40.s3) + 16.f *</div><div class="line"><a name="l00611"></a><span class="lineno"> 611</span>  (w01 + 2.f * w11 + 4.f * w21 + 8.f * w31 + 16.f * w41)) / 8100.f;</div><div class="line"><a name="l00612"></a><span class="lineno"> 612</span>  out3.s5 = (16.f * (w00.s0 + 2.f * w10.s0 + 4.f * w20.s0 + 8.f * w30.s0 + 16.f * w40.s0) + 8.f * (w00.s1 + 2.f * w10.s1 + 4.f * w20.s1 + 8.f * w30.s1 + 16.f * w40.s1) + 4.f *</div><div class="line"><a name="l00613"></a><span class="lineno"> 613</span>  (w00.s2 + 2.f * w10.s2 + 4.f * w20.s2 + 8.f * w30.s2 + 16.f * w40.s2) + 2.f * (w00.s3 + 2.f * w10.s3 + 4.f * w20.s3 + 8.f * w30.s3 + 16.f * w40.s3) +</div><div class="line"><a name="l00614"></a><span class="lineno"> 614</span>  (w01 + 2.f * w11 + 4.f * w21 + 8.f * w31 + 16.f * w41)) / 16200.f;</div><div class="line"><a name="l00615"></a><span class="lineno"> 615</span>  out3.s6 = (16.f * (w00.s0 + 2.f * w10.s0 + 4.f * w20.s0 + 8.f * w30.s0 + 16.f * w40.s0) - 8.f * (w00.s1 + 2.f * w10.s1 + 4.f * w20.s1 + 8.f * w30.s1 + 16.f * w40.s1) + 4.f *</div><div class="line"><a name="l00616"></a><span class="lineno"> 616</span>  (w00.s2 + 2.f * w10.s2 + 4.f * w20.s2 + 8.f * w30.s2 + 16.f * w40.s2) - 2.f * (w00.s3 + 2.f * w10.s3 + 4.f * w20.s3 + 8.f * w30.s3 + 16.f * w40.s3) +</div><div class="line"><a name="l00617"></a><span class="lineno"> 617</span>  (w01 + 2.f * w11 + 4.f * w21 + 8.f * w31 + 16.f * w41)) / 16200.f;</div><div class="line"><a name="l00618"></a><span class="lineno"> 618</span>  out3.s7 = (w01 + 2.f * w11 + 4.f * w21 + 8.f * w31 + 16.f * w41) / 90.f;</div><div class="line"><a name="l00619"></a><span class="lineno"> 619</span> </div><div class="line"><a name="l00620"></a><span class="lineno"> 620</span>  <span class="comment">// Row 4</span></div><div class="line"><a name="l00621"></a><span class="lineno"> 621</span>  <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)</div><div class="line"><a name="l00622"></a><span class="lineno"> 622</span>  out4 = 0.0f;</div><div class="line"><a name="l00623"></a><span class="lineno"> 623</span>  out4.s0 = (w00.s0 - 2.f * w10.s0 + 4.f * w20.s0 - 8.f * w30.s0 + 16.f * w40.s0) / 90.f;</div><div class="line"><a name="l00624"></a><span class="lineno"> 624</span>  out4.s1 = -((w00.s0 - 2.f * w10.s0 + 4.f * w20.s0 - 8.f * w30.s0 + 16.f * w40.s0) + (w00.s1 - 2.f * w10.s1 + 4.f * w20.s1 - 8.f * w30.s1 + 16.f * w40.s1) +</div><div class="line"><a name="l00625"></a><span class="lineno"> 625</span>  (w00.s2 - 2.f * w10.s2 + 4.f * w20.s2 - 8.f * w30.s2 + 16.f * w40.s2) + (w00.s3 - 2.f * w10.s3 + 4.f * w20.s3 - 8.f * w30.s3 + 16.f * w40.s3) +</div><div class="line"><a name="l00626"></a><span class="lineno"> 626</span>  (w01 - 2.f * w11 + 4.f * w21 - 8.f * w31 + 16.f * w41)) / 405.f;</div><div class="line"><a name="l00627"></a><span class="lineno"> 627</span>  out4.s2 = -((w00.s0 - 2.f * w10.s0 + 4.f * w20.s0 - 8.f * w30.s0 + 16.f * w40.s0) - (w00.s1 - 2.f * w10.s1 + 4.f * w20.s1 - 8.f * w30.s1 + 16.f * w40.s1) +</div><div class="line"><a name="l00628"></a><span class="lineno"> 628</span>  (w00.s2 - 2.f * w10.s2 + 4.f * w20.s2 - 8.f * w30.s2 + 16.f * w40.s2) - (w00.s3 - 2.f * w10.s3 + 4.f * w20.s3 - 8.f * w30.s3 + 16.f * w40.s3) +</div><div class="line"><a name="l00629"></a><span class="lineno"> 629</span>  (w01 - 2.f * w11 + 4.f * w21 - 8.f * w31 + 16.f * w41)) / 405.f;</div><div class="line"><a name="l00630"></a><span class="lineno"> 630</span>  out4.s3 = ((w00.s0 - 2.f * w10.s0 + 4.f * w20.s0 - 8.f * w30.s0 + 16.f * w40.s0) + 2.f * (w00.s1 - 2.f * w10.s1 + 4.f * w20.s1 - 8.f * w30.s1 + 16.f * w40.s1) + 4.f *</div><div class="line"><a name="l00631"></a><span class="lineno"> 631</span>  (w00.s2 - 2.f * w10.s2 + 4.f * w20.s2 - 8.f * w30.s2 + 16.f * w40.s2) + 8.f * (w00.s3 - 2.f * w10.s3 + 4.f * w20.s3 - 8.f * w30.s3 + 16.f * w40.s3) + 16.f *</div><div class="line"><a name="l00632"></a><span class="lineno"> 632</span>  (w01 - 2.f * w11 + 4.f * w21 - 8.f * w31 + 16.f * w41)) / 8100.f;</div><div class="line"><a name="l00633"></a><span class="lineno"> 633</span>  out4.s4 = ((w00.s0 - 2.f * w10.s0 + 4.f * w20.s0 - 8.f * w30.s0 + 16.f * w40.s0) - 2.f * (w00.s1 - 2.f * w10.s1 + 4.f * w20.s1 - 8.f * w30.s1 + 16.f * w40.s1) + 4.f *</div><div class="line"><a name="l00634"></a><span class="lineno"> 634</span>  (w00.s2 - 2.f * w10.s2 + 4.f * w20.s2 - 8.f * w30.s2 + 16.f * w40.s2) - 8.f * (w00.s3 - 2.f * w10.s3 + 4.f * w20.s3 - 8.f * w30.s3 + 16.f * w40.s3) + 16.f *</div><div class="line"><a name="l00635"></a><span class="lineno"> 635</span>  (w01 - 2.f * w11 + 4.f * w21 - 8.f * w31 + 16.f * w41)) / 8100.f;</div><div class="line"><a name="l00636"></a><span class="lineno"> 636</span>  out4.s5 = (16.f * (w00.s0 - 2.f * w10.s0 + 4.f * w20.s0 - 8.f * w30.s0 + 16.f * w40.s0) + 8.f * (w00.s1 - 2.f * w10.s1 + 4.f * w20.s1 - 8.f * w30.s1 + 16.f * w40.s1) + 4.f *</div><div class="line"><a name="l00637"></a><span class="lineno"> 637</span>  (w00.s2 - 2.f * w10.s2 + 4.f * w20.s2 - 8.f * w30.s2 + 16.f * w40.s2) + 2.f * (w00.s3 - 2.f * w10.s3 + 4.f * w20.s3 - 8.f * w30.s3 + 16.f * w40.s3) +</div><div class="line"><a name="l00638"></a><span class="lineno"> 638</span>  (w01 - 2.f * w11 + 4.f * w21 - 8.f * w31 + 16.f * w41)) / 16200.f;</div><div class="line"><a name="l00639"></a><span class="lineno"> 639</span>  out4.s6 = (16.f * (w00.s0 - 2.f * w10.s0 + 4.f * w20.s0 - 8.f * w30.s0 + 16.f * w40.s0) - 8.f * (w00.s1 - 2.f * w10.s1 + 4.f * w20.s1 - 8.f * w30.s1 + 16.f * w40.s1) + 4.f *</div><div class="line"><a name="l00640"></a><span class="lineno"> 640</span>  (w00.s2 - 2.f * w10.s2 + 4.f * w20.s2 - 8.f * w30.s2 + 16.f * w40.s2) - 2.f * (w00.s3 - 2.f * w10.s3 + 4.f * w20.s3 - 8.f * w30.s3 + 16.f * w40.s3) +</div><div class="line"><a name="l00641"></a><span class="lineno"> 641</span>  (w01 - 2.f * w11 + 4.f * w21 - 8.f * w31 + 16.f * w41)) / 16200.f;</div><div class="line"><a name="l00642"></a><span class="lineno"> 642</span>  out4.s7 = (w01 - 2.f * w11 + 4.f * w21 - 8.f * w31 + 16.f * w41) / 90.f;</div><div class="line"><a name="l00643"></a><span class="lineno"> 643</span> </div><div class="line"><a name="l00644"></a><span class="lineno"> 644</span>  <span class="comment">// Row 5</span></div><div class="line"><a name="l00645"></a><span class="lineno"> 645</span>  <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)</div><div class="line"><a name="l00646"></a><span class="lineno"> 646</span>  out5 = 0.0f;</div><div class="line"><a name="l00647"></a><span class="lineno"> 647</span>  out5.s0 = (16.f * w00.s0 + 8.f * w10.s0 + 4.f * w20.s0 + 2.f * w30.s0 + w40.s0) / 180.f;</div><div class="line"><a name="l00648"></a><span class="lineno"> 648</span>  out5.s1 = -((16.f * w00.s0 + 8.f * w10.s0 + 4.f * w20.s0 + 2.f * w30.s0 + w40.s0) + (16.f * w00.s1 + 8.f * w10.s1 + 4.f * w20.s1 + 2.f * w30.s1 + w40.s1) +</div><div class="line"><a name="l00649"></a><span class="lineno"> 649</span>  (16.f * w00.s2 + 8.f * w10.s2 + 4.f * w20.s2 + 2.f * w30.s2 + w40.s2) + (16.f * w00.s3 + 8.f * w10.s3 + 4.f * w20.s3 + 2.f * w30.s3 + w40.s3) +</div><div class="line"><a name="l00650"></a><span class="lineno"> 650</span>  (16.f * w01 + 8.f * w11 + 4.f * w21 + 2.f * w31 + w41)) / 810.f;</div><div class="line"><a name="l00651"></a><span class="lineno"> 651</span>  out5.s2 = -((16.f * w00.s0 + 8.f * w10.s0 + 4.f * w20.s0 + 2.f * w30.s0 + w40.s0) - (16.f * w00.s1 + 8.f * w10.s1 + 4.f * w20.s1 + 2.f * w30.s1 + w40.s1) +</div><div class="line"><a name="l00652"></a><span class="lineno"> 652</span>  (16.f * w00.s2 + 8.f * w10.s2 + 4.f * w20.s2 + 2.f * w30.s2 + w40.s2) - (16.f * w00.s3 + 8.f * w10.s3 + 4.f * w20.s3 + 2.f * w30.s3 + w40.s3) +</div><div class="line"><a name="l00653"></a><span class="lineno"> 653</span>  (16.f * w01 + 8.f * w11 + 4.f * w21 + 2.f * w31 + w41)) / 810.f;</div><div class="line"><a name="l00654"></a><span class="lineno"> 654</span>  out5.s3 = ((16.f * w00.s0 + 8.f * w10.s0 + 4.f * w20.s0 + 2.f * w30.s0 + w40.s0) + 2.f * (16.f * w00.s1 + 8.f * w10.s1 + 4.f * w20.s1 + 2.f * w30.s1 + w40.s1) + 4.f *</div><div class="line"><a name="l00655"></a><span class="lineno"> 655</span>  (16.f * w00.s2 + 8.f * w10.s2 + 4.f * w20.s2 + 2.f * w30.s2 + w40.s2) + 8.f * (16.f * w00.s3 + 8.f * w10.s3 + 4.f * w20.s3 + 2.f * w30.s3 + w40.s3) + 16.f *</div><div class="line"><a name="l00656"></a><span class="lineno"> 656</span>  (16.f * w01 + 8.f * w11 + 4.f * w21 + 2.f * w31 + w41)) / 16200.f;</div><div class="line"><a name="l00657"></a><span class="lineno"> 657</span>  out5.s4 = ((16.f * w00.s0 + 8.f * w10.s0 + 4.f * w20.s0 + 2.f * w30.s0 + w40.s0) - 2.f * (16.f * w00.s1 + 8.f * w10.s1 + 4.f * w20.s1 + 2.f * w30.s1 + w40.s1) + 4.f *</div><div class="line"><a name="l00658"></a><span class="lineno"> 658</span>  (16.f * w00.s2 + 8.f * w10.s2 + 4.f * w20.s2 + 2.f * w30.s2 + w40.s2) - 8.f * (16.f * w00.s3 + 8.f * w10.s3 + 4.f * w20.s3 + 2.f * w30.s3 + w40.s3) + 16.f *</div><div class="line"><a name="l00659"></a><span class="lineno"> 659</span>  (16.f * w01 + 8.f * w11 + 4.f * w21 + 2.f * w31 + w41)) / 16200.f;</div><div class="line"><a name="l00660"></a><span class="lineno"> 660</span>  out5.s5 = (16.f * (16.f * w00.s0 + 8.f * w10.s0 + 4.f * w20.s0 + 2.f * w30.s0 + w40.s0) + 8.f * (16.f * w00.s1 + 8.f * w10.s1 + 4.f * w20.s1 + 2.f * w30.s1 + w40.s1) + 4.f *</div><div class="line"><a name="l00661"></a><span class="lineno"> 661</span>  (16.f * w00.s2 + 8.f * w10.s2 + 4.f * w20.s2 + 2.f * w30.s2 + w40.s2) + 2.f * (16.f * w00.s3 + 8.f * w10.s3 + 4.f * w20.s3 + 2.f * w30.s3 + w40.s3) +</div><div class="line"><a name="l00662"></a><span class="lineno"> 662</span>  (16.f * w01 + 8.f * w11 + 4.f * w21 + 2.f * w31 + w41)) / 32400.f;</div><div class="line"><a name="l00663"></a><span class="lineno"> 663</span>  out5.s6 = (16.f * (16.f * w00.s0 + 8.f * w10.s0 + 4.f * w20.s0 + 2.f * w30.s0 + w40.s0) - 8.f * (16.f * w00.s1 + 8.f * w10.s1 + 4.f * w20.s1 + 2.f * w30.s1 + w40.s1) + 4.f *</div><div class="line"><a name="l00664"></a><span class="lineno"> 664</span>  (16.f * w00.s2 + 8.f * w10.s2 + 4.f * w20.s2 + 2.f * w30.s2 + w40.s2) - 2.f * (16.f * w00.s3 + 8.f * w10.s3 + 4.f * w20.s3 + 2.f * w30.s3 + w40.s3) +</div><div class="line"><a name="l00665"></a><span class="lineno"> 665</span>  (16.f * w01 + 8.f * w11 + 4.f * w21 + 2.f * w31 + w41)) / 32400.f;</div><div class="line"><a name="l00666"></a><span class="lineno"> 666</span>  out5.s7 = (16.f * w01 + 8.f * w11 + 4.f * w21 + 2.f * w31 + w41) / 180.f;</div><div class="line"><a name="l00667"></a><span class="lineno"> 667</span> </div><div class="line"><a name="l00668"></a><span class="lineno"> 668</span>  <span class="comment">// Row 6</span></div><div class="line"><a name="l00669"></a><span class="lineno"> 669</span>  <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)</div><div class="line"><a name="l00670"></a><span class="lineno"> 670</span>  out6 = 0.0f;</div><div class="line"><a name="l00671"></a><span class="lineno"> 671</span>  out6.s0 = (16.f * w00.s0 - 8.f * w10.s0 + 4.f * w20.s0 - 2.f * w30.s0 + w40.s0) / 180.f;</div><div class="line"><a name="l00672"></a><span class="lineno"> 672</span>  out6.s1 = -((16.f * w00.s0 - 8.f * w10.s0 + 4.f * w20.s0 - 2.f * w30.s0 + w40.s0) + (16.f * w00.s1 - 8.f * w10.s1 + 4.f * w20.s1 - 2.f * w30.s1 + w40.s1) +</div><div class="line"><a name="l00673"></a><span class="lineno"> 673</span>  (16.f * w00.s2 - 8.f * w10.s2 + 4.f * w20.s2 - 2.f * w30.s2 + w40.s2) + (16.f * w00.s3 - 8.f * w10.s3 + 4.f * w20.s3 - 2.f * w30.s3 + w40.s3) +</div><div class="line"><a name="l00674"></a><span class="lineno"> 674</span>  (16.f * w01 - 8.f * w11 + 4.f * w21 - 2.f * w31 + w41)) / 810.f;</div><div class="line"><a name="l00675"></a><span class="lineno"> 675</span>  out6.s2 = -((16.f * w00.s0 - 8.f * w10.s0 + 4.f * w20.s0 - 2.f * w30.s0 + w40.s0) - (16.f * w00.s1 - 8.f * w10.s1 + 4.f * w20.s1 - 2.f * w30.s1 + w40.s1) +</div><div class="line"><a name="l00676"></a><span class="lineno"> 676</span>  (16.f * w00.s2 - 8.f * w10.s2 + 4.f * w20.s2 - 2.f * w30.s2 + w40.s2) - (16.f * w00.s3 - 8.f * w10.s3 + 4.f * w20.s3 - 2.f * w30.s3 + w40.s3) +</div><div class="line"><a name="l00677"></a><span class="lineno"> 677</span>  (16.f * w01 - 8.f * w11 + 4.f * w21 - 2.f * w31 + w41)) / 810.f;</div><div class="line"><a name="l00678"></a><span class="lineno"> 678</span>  out6.s3 = ((16.f * w00.s0 - 8.f * w10.s0 + 4.f * w20.s0 - 2.f * w30.s0 + w40.s0) + 2.f * (16.f * w00.s1 - 8.f * w10.s1 + 4.f * w20.s1 - 2.f * w30.s1 + w40.s1) + 4.f *</div><div class="line"><a name="l00679"></a><span class="lineno"> 679</span>  (16.f * w00.s2 - 8.f * w10.s2 + 4.f * w20.s2 - 2.f * w30.s2 + w40.s2) + 8.f * (16.f * w00.s3 - 8.f * w10.s3 + 4.f * w20.s3 - 2.f * w30.s3 + w40.s3) + 16.f *</div><div class="line"><a name="l00680"></a><span class="lineno"> 680</span>  (16.f * w01 - 8.f * w11 + 4.f * w21 - 2.f * w31 + w41)) / 16200.f;</div><div class="line"><a name="l00681"></a><span class="lineno"> 681</span>  out6.s4 = ((16.f * w00.s0 - 8.f * w10.s0 + 4.f * w20.s0 - 2.f * w30.s0 + w40.s0) - 2.f * (16.f * w00.s1 - 8.f * w10.s1 + 4.f * w20.s1 - 2.f * w30.s1 + w40.s1) + 4.f *</div><div class="line"><a name="l00682"></a><span class="lineno"> 682</span>  (16.f * w00.s2 - 8.f * w10.s2 + 4.f * w20.s2 - 2.f * w30.s2 + w40.s2) - 8.f * (16.f * w00.s3 - 8.f * w10.s3 + 4.f * w20.s3 - 2.f * w30.s3 + w40.s3) + 16.f *</div><div class="line"><a name="l00683"></a><span class="lineno"> 683</span>  (16.f * w01 - 8.f * w11 + 4.f * w21 - 2.f * w31 + w41)) / 16200.f;</div><div class="line"><a name="l00684"></a><span class="lineno"> 684</span>  out6.s5 = (16.f * (16.f * w00.s0 - 8.f * w10.s0 + 4.f * w20.s0 - 2.f * w30.s0 + w40.s0) + 8.f * (16.f * w00.s1 - 8.f * w10.s1 + 4.f * w20.s1 - 2.f * w30.s1 + w40.s1) + 4.f *</div><div class="line"><a name="l00685"></a><span class="lineno"> 685</span>  (16.f * w00.s2 - 8.f * w10.s2 + 4.f * w20.s2 - 2.f * w30.s2 + w40.s2) + 2.f * (16.f * w00.s3 - 8.f * w10.s3 + 4.f * w20.s3 - 2.f * w30.s3 + w40.s3) +</div><div class="line"><a name="l00686"></a><span class="lineno"> 686</span>  (16.f * w01 - 8.f * w11 + 4.f * w21 - 2.f * w31 + w41)) / 32400.f;</div><div class="line"><a name="l00687"></a><span class="lineno"> 687</span>  out6.s6 = (16.f * (16.f * w00.s0 - 8.f * w10.s0 + 4.f * w20.s0 - 2.f * w30.s0 + w40.s0) - 8.f * (16.f * w00.s1 - 8.f * w10.s1 + 4.f * w20.s1 - 2.f * w30.s1 + w40.s1) + 4.f *</div><div class="line"><a name="l00688"></a><span class="lineno"> 688</span>  (16.f * w00.s2 - 8.f * w10.s2 + 4.f * w20.s2 - 2.f * w30.s2 + w40.s2) - 2.f * (16.f * w00.s3 - 8.f * w10.s3 + 4.f * w20.s3 - 2.f * w30.s3 + w40.s3) +</div><div class="line"><a name="l00689"></a><span class="lineno"> 689</span>  (16.f * w01 - 8.f * w11 + 4.f * w21 - 2.f * w31 + w41)) / 32400.f;</div><div class="line"><a name="l00690"></a><span class="lineno"> 690</span>  out6.s7 = (16.f * w01 - 8.f * w11 + 4.f * w21 - 2.f * w31 + w41) / 180.f;</div><div class="line"><a name="l00691"></a><span class="lineno"> 691</span> </div><div class="line"><a name="l00692"></a><span class="lineno"> 692</span>  <span class="comment">// Row 7</span></div><div class="line"><a name="l00693"></a><span class="lineno"> 693</span>  <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)</div><div class="line"><a name="l00694"></a><span class="lineno"> 694</span>  out7 = 0.0f;</div><div class="line"><a name="l00695"></a><span class="lineno"> 695</span>  out7.s0 = w40.s0;</div><div class="line"><a name="l00696"></a><span class="lineno"> 696</span>  out7.s1 = -2.f * (w40.s0 + w40.s1 + w40.s2 + w40.s3 + w41) / 9.f;</div><div class="line"><a name="l00697"></a><span class="lineno"> 697</span>  out7.s2 = -2.f * (w40.s0 - w40.s1 + w40.s2 - w40.s3 + w41) / 9.f;</div><div class="line"><a name="l00698"></a><span class="lineno"> 698</span>  out7.s3 = (w40.s0 + 2.f * w40.s1 + 4.f * w40.s2 + 8.f * w40.s3 + 16.f * w41) / 90.f;</div><div class="line"><a name="l00699"></a><span class="lineno"> 699</span>  out7.s4 = (w40.s0 - 2.f * w40.s1 + 4.f * w40.s2 - 8.f * w40.s3 + 16.f * w41) / 90.f;</div><div class="line"><a name="l00700"></a><span class="lineno"> 700</span>  out7.s5 = (16.f * w40.s0 + 8.f * w40.s1 + 4.f * w40.s2 + 2.f * w40.s3 + w41) / 180.f;</div><div class="line"><a name="l00701"></a><span class="lineno"> 701</span>  out7.s6 = (16.f * w40.s0 - 8.f * w40.s1 + 4.f * w40.s2 - 2.f * w40.s3 + w41) / 180.f;</div><div class="line"><a name="l00702"></a><span class="lineno"> 702</span>  out7.s7 = w41;</div><div class="line"><a name="l00703"></a><span class="lineno"> 703</span> <span class="preprocessor">#endif // !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00704"></a><span class="lineno"> 704</span> </div><div class="line"><a name="l00705"></a><span class="lineno"> 705</span>  <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l00706"></a><span class="lineno"> 706</span>  <span class="keywordtype">int</span> x0 = z / SRC_DIM_Z; <span class="comment">// idx filter</span></div><div class="line"><a name="l00707"></a><span class="lineno"> 707</span>  <span class="keywordtype">int</span> y0 = z % SRC_DIM_Z; <span class="comment">// idx channel</span></div><div class="line"><a name="l00708"></a><span class="lineno"> 708</span> </div><div class="line"><a name="l00709"></a><span class="lineno"> 709</span>  <span class="comment">// Get output address</span></div><div class="line"><a name="l00710"></a><span class="lineno"> 710</span>  __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x0 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>) + y0 * dst_stride_y;</div><div class="line"><a name="l00711"></a><span class="lineno"> 711</span> </div><div class="line"><a name="l00712"></a><span class="lineno"> 712</span>  <span class="comment">// Store the values across the channels</span></div><div class="line"><a name="l00713"></a><span class="lineno"> 713</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 0 * dst_stride_z) = out0.s0;</div><div class="line"><a name="l00714"></a><span class="lineno"> 714</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 1 * dst_stride_z) = out0.s1;</div><div class="line"><a name="l00715"></a><span class="lineno"> 715</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 2 * dst_stride_z) = out0.s2;</div><div class="line"><a name="l00716"></a><span class="lineno"> 716</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 3 * dst_stride_z) = out0.s3;</div><div class="line"><a name="l00717"></a><span class="lineno"> 717</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 4 * dst_stride_z) = out0.s4;</div><div class="line"><a name="l00718"></a><span class="lineno"> 718</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 5 * dst_stride_z) = out0.s5;</div><div class="line"><a name="l00719"></a><span class="lineno"> 719</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 6 * dst_stride_z) = out0.s6;</div><div class="line"><a name="l00720"></a><span class="lineno"> 720</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 7 * dst_stride_z) = out0.s7;</div><div class="line"><a name="l00721"></a><span class="lineno"> 721</span> </div><div class="line"><a name="l00722"></a><span class="lineno"> 722</span> <span class="preprocessor">#if !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00723"></a><span class="lineno"> 723</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 8 * dst_stride_z) = out1.s0;</div><div class="line"><a name="l00724"></a><span class="lineno"> 724</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 9 * dst_stride_z) = out1.s1;</div><div class="line"><a name="l00725"></a><span class="lineno"> 725</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 10 * dst_stride_z) = out1.s2;</div><div class="line"><a name="l00726"></a><span class="lineno"> 726</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 11 * dst_stride_z) = out1.s3;</div><div class="line"><a name="l00727"></a><span class="lineno"> 727</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 12 * dst_stride_z) = out1.s4;</div><div class="line"><a name="l00728"></a><span class="lineno"> 728</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 13 * dst_stride_z) = out1.s5;</div><div class="line"><a name="l00729"></a><span class="lineno"> 729</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 14 * dst_stride_z) = out1.s6;</div><div class="line"><a name="l00730"></a><span class="lineno"> 730</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 15 * dst_stride_z) = out1.s7;</div><div class="line"><a name="l00731"></a><span class="lineno"> 731</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 16 * dst_stride_z) = out2.s0;</div><div class="line"><a name="l00732"></a><span class="lineno"> 732</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 17 * dst_stride_z) = out2.s1;</div><div class="line"><a name="l00733"></a><span class="lineno"> 733</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 18 * dst_stride_z) = out2.s2;</div><div class="line"><a name="l00734"></a><span class="lineno"> 734</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 19 * dst_stride_z) = out2.s3;</div><div class="line"><a name="l00735"></a><span class="lineno"> 735</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 20 * dst_stride_z) = out2.s4;</div><div class="line"><a name="l00736"></a><span class="lineno"> 736</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 21 * dst_stride_z) = out2.s5;</div><div class="line"><a name="l00737"></a><span class="lineno"> 737</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 22 * dst_stride_z) = out2.s6;</div><div class="line"><a name="l00738"></a><span class="lineno"> 738</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 23 * dst_stride_z) = out2.s7;</div><div class="line"><a name="l00739"></a><span class="lineno"> 739</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 24 * dst_stride_z) = out3.s0;</div><div class="line"><a name="l00740"></a><span class="lineno"> 740</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 25 * dst_stride_z) = out3.s1;</div><div class="line"><a name="l00741"></a><span class="lineno"> 741</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 26 * dst_stride_z) = out3.s2;</div><div class="line"><a name="l00742"></a><span class="lineno"> 742</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 27 * dst_stride_z) = out3.s3;</div><div class="line"><a name="l00743"></a><span class="lineno"> 743</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 28 * dst_stride_z) = out3.s4;</div><div class="line"><a name="l00744"></a><span class="lineno"> 744</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 29 * dst_stride_z) = out3.s5;</div><div class="line"><a name="l00745"></a><span class="lineno"> 745</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 30 * dst_stride_z) = out3.s6;</div><div class="line"><a name="l00746"></a><span class="lineno"> 746</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 31 * dst_stride_z) = out3.s7;</div><div class="line"><a name="l00747"></a><span class="lineno"> 747</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 32 * dst_stride_z) = out4.s0;</div><div class="line"><a name="l00748"></a><span class="lineno"> 748</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 33 * dst_stride_z) = out4.s1;</div><div class="line"><a name="l00749"></a><span class="lineno"> 749</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 34 * dst_stride_z) = out4.s2;</div><div class="line"><a name="l00750"></a><span class="lineno"> 750</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 35 * dst_stride_z) = out4.s3;</div><div class="line"><a name="l00751"></a><span class="lineno"> 751</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 36 * dst_stride_z) = out4.s4;</div><div class="line"><a name="l00752"></a><span class="lineno"> 752</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 37 * dst_stride_z) = out4.s5;</div><div class="line"><a name="l00753"></a><span class="lineno"> 753</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 38 * dst_stride_z) = out4.s6;</div><div class="line"><a name="l00754"></a><span class="lineno"> 754</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 39 * dst_stride_z) = out4.s7;</div><div class="line"><a name="l00755"></a><span class="lineno"> 755</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 40 * dst_stride_z) = out5.s0;</div><div class="line"><a name="l00756"></a><span class="lineno"> 756</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 41 * dst_stride_z) = out5.s1;</div><div class="line"><a name="l00757"></a><span class="lineno"> 757</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 42 * dst_stride_z) = out5.s2;</div><div class="line"><a name="l00758"></a><span class="lineno"> 758</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 43 * dst_stride_z) = out5.s3;</div><div class="line"><a name="l00759"></a><span class="lineno"> 759</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 44 * dst_stride_z) = out5.s4;</div><div class="line"><a name="l00760"></a><span class="lineno"> 760</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 45 * dst_stride_z) = out5.s5;</div><div class="line"><a name="l00761"></a><span class="lineno"> 761</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 46 * dst_stride_z) = out5.s6;</div><div class="line"><a name="l00762"></a><span class="lineno"> 762</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 47 * dst_stride_z) = out5.s7;</div><div class="line"><a name="l00763"></a><span class="lineno"> 763</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 48 * dst_stride_z) = out6.s0;</div><div class="line"><a name="l00764"></a><span class="lineno"> 764</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 49 * dst_stride_z) = out6.s1;</div><div class="line"><a name="l00765"></a><span class="lineno"> 765</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 50 * dst_stride_z) = out6.s2;</div><div class="line"><a name="l00766"></a><span class="lineno"> 766</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 51 * dst_stride_z) = out6.s3;</div><div class="line"><a name="l00767"></a><span class="lineno"> 767</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 52 * dst_stride_z) = out6.s4;</div><div class="line"><a name="l00768"></a><span class="lineno"> 768</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 53 * dst_stride_z) = out6.s5;</div><div class="line"><a name="l00769"></a><span class="lineno"> 769</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 54 * dst_stride_z) = out6.s6;</div><div class="line"><a name="l00770"></a><span class="lineno"> 770</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 55 * dst_stride_z) = out6.s7;</div><div class="line"><a name="l00771"></a><span class="lineno"> 771</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 56 * dst_stride_z) = out7.s0;</div><div class="line"><a name="l00772"></a><span class="lineno"> 772</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 57 * dst_stride_z) = out7.s1;</div><div class="line"><a name="l00773"></a><span class="lineno"> 773</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 58 * dst_stride_z) = out7.s2;</div><div class="line"><a name="l00774"></a><span class="lineno"> 774</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 59 * dst_stride_z) = out7.s3;</div><div class="line"><a name="l00775"></a><span class="lineno"> 775</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 60 * dst_stride_z) = out7.s4;</div><div class="line"><a name="l00776"></a><span class="lineno"> 776</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 61 * dst_stride_z) = out7.s5;</div><div class="line"><a name="l00777"></a><span class="lineno"> 777</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 62 * dst_stride_z) = out7.s6;</div><div class="line"><a name="l00778"></a><span class="lineno"> 778</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 63 * dst_stride_z) = out7.s7;</div><div class="line"><a name="l00779"></a><span class="lineno"> 779</span> <span class="preprocessor">#endif // !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00780"></a><span class="lineno"> 780</span> }</div><div class="line"><a name="l00781"></a><span class="lineno"> 781</span> <span class="comment"></span></div><div class="line"><a name="l00782"></a><span class="lineno"> 782</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 5x5/5x1 or 1x5 when the data layout is NHWC and the output tile is 4x4/4x1 or 1x4</span></div><div class="line"><a name="l00783"></a><span class="lineno"> 783</span> <span class="comment"> *</span></div><div class="line"><a name="l00784"></a><span class="lineno"> 784</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l00785"></a><span class="lineno"> 785</span> <span class="comment"> * @note If this kernel is used to perform Winograd filter transform 5x1, -DWINOGRAD_FILTER_TRANSFORM_HORIZONTAL has to be passed at compile time</span></div><div class="line"><a name="l00786"></a><span class="lineno"> 786</span> <span class="comment"> * @note If this kernel is used to perform Winograd filter transform 1x5, -DWINOGRAD_FILTER_TRANSFORM_VERTICAL has to be passed at compile time</span></div><div class="line"><a name="l00787"></a><span class="lineno"> 787</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l00788"></a><span class="lineno"> 788</span> <span class="comment"> *</span></div><div class="line"><a name="l00789"></a><span class="lineno"> 789</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l00790"></a><span class="lineno"> 790</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00791"></a><span class="lineno"> 791</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00792"></a><span class="lineno"> 792</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00793"></a><span class="lineno"> 793</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00794"></a><span class="lineno"> 794</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00795"></a><span class="lineno"> 795</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00796"></a><span class="lineno"> 796</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l00797"></a><span class="lineno"> 797</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l00798"></a><span class="lineno"> 798</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l00799"></a><span class="lineno"> 799</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l00800"></a><span class="lineno"> 800</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00801"></a><span class="lineno"> 801</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00802"></a><span class="lineno"> 802</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00803"></a><span class="lineno"> 803</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00804"></a><span class="lineno"> 804</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00805"></a><span class="lineno"> 805</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00806"></a><span class="lineno"> 806</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l00807"></a><span class="lineno"> 807</span> <span class="comment"> */</span></div><div class="line"><a name="l00808"></a><span class="lineno"> 808</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_4x4_5x5_nhwc(</div><div class="line"><a name="l00809"></a><span class="lineno"> 809</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l00810"></a><span class="lineno"> 810</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l00811"></a><span class="lineno"> 811</span> {</div><div class="line"><a name="l00812"></a><span class="lineno"> 812</span>  <a class="code" href="struct_tensor4_d.xhtml">Tensor4D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</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#a989ab3e96426615bb98e04e0235088ca">src</a>, SRC_DIM_Z);</div><div class="line"><a name="l00813"></a><span class="lineno"> 813</span> </div><div class="line"><a name="l00814"></a><span class="lineno"> 814</span>  <span class="keyword">const</span> __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + get_global_id(0) * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>) + get_global_id(1) * src_step_y + get_global_id(2) * src_step_w;</div><div class="line"><a name="l00815"></a><span class="lineno"> 815</span> </div><div class="line"><a name="l00816"></a><span class="lineno"> 816</span> <span class="preprocessor">#if defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00817"></a><span class="lineno"> 817</span>  <span class="comment">// Load the values from the input tensor</span></div><div class="line"><a name="l00818"></a><span class="lineno"> 818</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w00 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_z));</div><div class="line"><a name="l00819"></a><span class="lineno"> 819</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w01 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z));</div><div class="line"><a name="l00820"></a><span class="lineno"> 820</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w02 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z));</div><div class="line"><a name="l00821"></a><span class="lineno"> 821</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w03 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_z));</div><div class="line"><a name="l00822"></a><span class="lineno"> 822</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w04 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_z));</div><div class="line"><a name="l00823"></a><span class="lineno"> 823</span> <span class="preprocessor">#else // defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00824"></a><span class="lineno"> 824</span>  <span class="comment">// Load the values from the input tensor</span></div><div class="line"><a name="l00825"></a><span class="lineno"> 825</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w00 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_y));</div><div class="line"><a name="l00826"></a><span class="lineno"> 826</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w01 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_y));</div><div class="line"><a name="l00827"></a><span class="lineno"> 827</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w02 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_y));</div><div class="line"><a name="l00828"></a><span class="lineno"> 828</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w03 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_y));</div><div class="line"><a name="l00829"></a><span class="lineno"> 829</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w04 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_y));</div><div class="line"><a name="l00830"></a><span class="lineno"> 830</span> <span class="preprocessor">#endif // defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00831"></a><span class="lineno"> 831</span> </div><div class="line"><a name="l00832"></a><span class="lineno"> 832</span> <span class="preprocessor">#if !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00833"></a><span class="lineno"> 833</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w10 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z + 0 * src_stride_y));</div><div class="line"><a name="l00834"></a><span class="lineno"> 834</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w11 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z + 1 * src_stride_y));</div><div class="line"><a name="l00835"></a><span class="lineno"> 835</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w12 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z + 2 * src_stride_y));</div><div class="line"><a name="l00836"></a><span class="lineno"> 836</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w13 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z + 3 * src_stride_y));</div><div class="line"><a name="l00837"></a><span class="lineno"> 837</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w14 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z + 4 * src_stride_y));</div><div class="line"><a name="l00838"></a><span class="lineno"> 838</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w20 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z + 0 * src_stride_y));</div><div class="line"><a name="l00839"></a><span class="lineno"> 839</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w21 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z + 1 * src_stride_y));</div><div class="line"><a name="l00840"></a><span class="lineno"> 840</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w22 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z + 2 * src_stride_y));</div><div class="line"><a name="l00841"></a><span class="lineno"> 841</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w23 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z + 3 * src_stride_y));</div><div class="line"><a name="l00842"></a><span class="lineno"> 842</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w24 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z + 4 * src_stride_y));</div><div class="line"><a name="l00843"></a><span class="lineno"> 843</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w30 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_z + 0 * src_stride_y));</div><div class="line"><a name="l00844"></a><span class="lineno"> 844</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w31 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_z + 1 * src_stride_y));</div><div class="line"><a name="l00845"></a><span class="lineno"> 845</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w32 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_z + 2 * src_stride_y));</div><div class="line"><a name="l00846"></a><span class="lineno"> 846</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w33 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_z + 3 * src_stride_y));</div><div class="line"><a name="l00847"></a><span class="lineno"> 847</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w34 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_z + 4 * src_stride_y));</div><div class="line"><a name="l00848"></a><span class="lineno"> 848</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w40 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_z + 0 * src_stride_y));</div><div class="line"><a name="l00849"></a><span class="lineno"> 849</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w41 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_z + 1 * src_stride_y));</div><div class="line"><a name="l00850"></a><span class="lineno"> 850</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w42 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_z + 2 * src_stride_y));</div><div class="line"><a name="l00851"></a><span class="lineno"> 851</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w43 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_z + 3 * src_stride_y));</div><div class="line"><a name="l00852"></a><span class="lineno"> 852</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w44 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_z + 4 * src_stride_y));</div><div class="line"><a name="l00853"></a><span class="lineno"> 853</span> <span class="preprocessor">#endif // !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00854"></a><span class="lineno"> 854</span> </div><div class="line"><a name="l00855"></a><span class="lineno"> 855</span>  <span class="comment">// Row 0</span></div><div class="line"><a name="l00856"></a><span class="lineno"> 856</span>  <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)</div><div class="line"><a name="l00857"></a><span class="lineno"> 857</span>  out0 = 0.0f;</div><div class="line"><a name="l00858"></a><span class="lineno"> 858</span>  out0.s0 = w00;</div><div class="line"><a name="l00859"></a><span class="lineno"> 859</span>  out0.s1 = -2.f * (w00 + w01 + w02 + w03 + w04) / 9.f;</div><div class="line"><a name="l00860"></a><span class="lineno"> 860</span>  out0.s2 = -2.f * (w00 - w01 + w02 - w03 + w04) / 9.f;</div><div class="line"><a name="l00861"></a><span class="lineno"> 861</span>  out0.s3 = (w00 + 2.f * w01 + 4.f * w02 + 8.f * w03 + 16.f * w04) / 90.f;</div><div class="line"><a name="l00862"></a><span class="lineno"> 862</span>  out0.s4 = (w00 - 2.f * w01 + 4.f * w02 - 8.f * w03 + 16.f * w04) / 90.f;</div><div class="line"><a name="l00863"></a><span class="lineno"> 863</span>  out0.s5 = (16.f * w00 + 8.f * w01 + 4.f * w02 + 2.f * w03 + w04) / 180.f;</div><div class="line"><a name="l00864"></a><span class="lineno"> 864</span>  out0.s6 = (16.f * w00 - 8.f * w01 + 4.f * w02 - 2.f * w03 + w04) / 180.f;</div><div class="line"><a name="l00865"></a><span class="lineno"> 865</span>  out0.s7 = w04;</div><div class="line"><a name="l00866"></a><span class="lineno"> 866</span> </div><div class="line"><a name="l00867"></a><span class="lineno"> 867</span> <span class="preprocessor">#if !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00868"></a><span class="lineno"> 868</span>  <span class="comment">// Row 1</span></div><div class="line"><a name="l00869"></a><span class="lineno"> 869</span>  <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)</div><div class="line"><a name="l00870"></a><span class="lineno"> 870</span>  out1 = 0.0f;</div><div class="line"><a name="l00871"></a><span class="lineno"> 871</span>  out1.s0 = -2.f * (w00 + w10 + w20 + w30 + w40) / 9.f;</div><div class="line"><a name="l00872"></a><span class="lineno"> 872</span>  out1.s1 = 4.f * ((w00 + w10 + w20 + w30 + w40) + (w01 + w11 + w21 + w31 + w41) + (w02 + w12 + w22 + w32 + w42) + (w03 + w13 + w23 + w33 + w43) + (w04 + w14 + w24 + w34 + w44)) / 81.f;</div><div class="line"><a name="l00873"></a><span class="lineno"> 873</span>  out1.s2 = 4.f * ((w00 + w10 + w20 + w30 + w40) - (w01 + w11 + w21 + w31 + w41) + (w02 + w12 + w22 + w32 + w42) - (w03 + w13 + w23 + w33 + w43) + (w04 + w14 + w24 + w34 + w44)) / 81.f;</div><div class="line"><a name="l00874"></a><span class="lineno"> 874</span>  out1.s3 = -((w00 + w10 + w20 + w30 + w40) + 2.f * (w01 + w11 + w21 + w31 + w41) + 4.f * (w02 + w12 + w22 + w32 + w42) + 8.f * (w03 + w13 + w23 + w33 + w43) + 16.f *</div><div class="line"><a name="l00875"></a><span class="lineno"> 875</span>  (w04 + w14 + w24 + w34 + w44)) / 405.f;</div><div class="line"><a name="l00876"></a><span class="lineno"> 876</span>  out1.s4 = -((w00 + w10 + w20 + w30 + w40) - 2.f * (w01 + w11 + w21 + w31 + w41) + 4.f * (w02 + w12 + w22 + w32 + w42) - 8.f * (w03 + w13 + w23 + w33 + w43) + 16.f *</div><div class="line"><a name="l00877"></a><span class="lineno"> 877</span>  (w04 + w14 + w24 + w34 + w44)) / 405.f;</div><div class="line"><a name="l00878"></a><span class="lineno"> 878</span>  out1.s5 = -(16.f * (w00 + w10 + w20 + w30 + w40) + 8.f * (w01 + w11 + w21 + w31 + w41) + 4.f * (w02 + w12 + w22 + w32 + w42) + 2.f * (w03 + w13 + w23 + w33 + w43) +</div><div class="line"><a name="l00879"></a><span class="lineno"> 879</span>  (w04 + w14 + w24 + w34 + w44)) / 810.f;</div><div class="line"><a name="l00880"></a><span class="lineno"> 880</span>  out1.s6 = -(16.f * (w00 + w10 + w20 + w30 + w40) - 8.f * (w01 + w11 + w21 + w31 + w41) + 4.f * (w02 + w12 + w22 + w32 + w42) - 2.f * (w03 + w13 + w23 + w33 + w43) +</div><div class="line"><a name="l00881"></a><span class="lineno"> 881</span>  (w04 + w14 + w24 + w34 + w44)) / 810.f;</div><div class="line"><a name="l00882"></a><span class="lineno"> 882</span>  out1.s7 = -2.f * (w04 + w14 + w24 + w34 + w44) / 9.f;</div><div class="line"><a name="l00883"></a><span class="lineno"> 883</span> </div><div class="line"><a name="l00884"></a><span class="lineno"> 884</span>  <span class="comment">// Row 2</span></div><div class="line"><a name="l00885"></a><span class="lineno"> 885</span>  <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)</div><div class="line"><a name="l00886"></a><span class="lineno"> 886</span>  out2 = 0.0f;</div><div class="line"><a name="l00887"></a><span class="lineno"> 887</span>  out2.s0 = -2.f * (w00 - w10 + w20 - w30 + w40) / 9.f;</div><div class="line"><a name="l00888"></a><span class="lineno"> 888</span>  out2.s1 = 4.f * ((w00 - w10 + w20 - w30 + w40) + (w01 - w11 + w21 - w31 + w41) + (w02 - w12 + w22 - w32 + w42) + (w03 - w13 + w23 - w33 + w43) + (w04 - w14 + w24 - w34 + w44)) / 81.f;</div><div class="line"><a name="l00889"></a><span class="lineno"> 889</span>  out2.s2 = 4.f * ((w00 - w10 + w20 - w30 + w40) - (w01 - w11 + w21 - w31 + w41) + (w02 - w12 + w22 - w32 + w42) - (w03 - w13 + w23 - w33 + w43) + (w04 - w14 + w24 - w34 + w44)) / 81.f;</div><div class="line"><a name="l00890"></a><span class="lineno"> 890</span>  out2.s3 = -((w00 - w10 + w20 - w30 + w40) + 2.f * (w01 - w11 + w21 - w31 + w41) + 4.f * (w02 - w12 + w22 - w32 + w42) + 8.f * (w03 - w13 + w23 - w33 + w43) + 16.f *</div><div class="line"><a name="l00891"></a><span class="lineno"> 891</span>  (w04 - w14 + w24 - w34 + w44)) / 405.f;</div><div class="line"><a name="l00892"></a><span class="lineno"> 892</span>  out2.s4 = -((w00 - w10 + w20 - w30 + w40) - 2.f * (w01 - w11 + w21 - w31 + w41) + 4.f * (w02 - w12 + w22 - w32 + w42) - 8.f * (w03 - w13 + w23 - w33 + w43) + 16.f *</div><div class="line"><a name="l00893"></a><span class="lineno"> 893</span>  (w04 - w14 + w24 - w34 + w44)) / 405.f;</div><div class="line"><a name="l00894"></a><span class="lineno"> 894</span>  out2.s5 = -(16.f * (w00 - w10 + w20 - w30 + w40) + 8.f * (w01 - w11 + w21 - w31 + w41) + 4.f * (w02 - w12 + w22 - w32 + w42) + 2.f * (w03 - w13 + w23 - w33 + w43) +</div><div class="line"><a name="l00895"></a><span class="lineno"> 895</span>  (w04 - w14 + w24 - w34 + w44)) / 810.f;</div><div class="line"><a name="l00896"></a><span class="lineno"> 896</span>  out2.s6 = -(16.f * (w00 - w10 + w20 - w30 + w40) - 8.f * (w01 - w11 + w21 - w31 + w41) + 4.f * (w02 - w12 + w22 - w32 + w42) - 2.f * (w03 - w13 + w23 - w33 + w43) +</div><div class="line"><a name="l00897"></a><span class="lineno"> 897</span>  (w04 - w14 + w24 - w34 + w44)) / 810.f;</div><div class="line"><a name="l00898"></a><span class="lineno"> 898</span>  out2.s7 = -2.f * (w04 - w14 + w24 - w34 + w44) / 9.f;</div><div class="line"><a name="l00899"></a><span class="lineno"> 899</span> </div><div class="line"><a name="l00900"></a><span class="lineno"> 900</span>  <span class="comment">// Row 3</span></div><div class="line"><a name="l00901"></a><span class="lineno"> 901</span>  <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)</div><div class="line"><a name="l00902"></a><span class="lineno"> 902</span>  out3 = 0.0f;</div><div class="line"><a name="l00903"></a><span class="lineno"> 903</span>  out3.s0 = (w00 + 2.f * w10 + 4.f * w20 + 8.f * w30 + 16.f * w40) / 90.f;</div><div class="line"><a name="l00904"></a><span class="lineno"> 904</span>  out3.s1 = -((w00 + 2.f * w10 + 4.f * w20 + 8.f * w30 + 16.f * w40) + (w01 + 2.f * w11 + 4.f * w21 + 8.f * w31 + 16.f * w41) + (w02 + 2.f * w12 + 4.f * w22 + 8.f * w32 + 16.f * w42) +</div><div class="line"><a name="l00905"></a><span class="lineno"> 905</span>  (w03 + 2.f * w13 + 4.f * w23 + 8.f * w33 + 16.f * w43) + (w04 + 2.f * w14 + 4.f * w24 + 8.f * w34 + 16.f * w44)) / 405.f;</div><div class="line"><a name="l00906"></a><span class="lineno"> 906</span>  out3.s2 = -((w00 + 2.f * w10 + 4.f * w20 + 8.f * w30 + 16.f * w40) - (w01 + 2.f * w11 + 4.f * w21 + 8.f * w31 + 16.f * w41) + (w02 + 2.f * w12 + 4.f * w22 + 8.f * w32 + 16.f * w42) -</div><div class="line"><a name="l00907"></a><span class="lineno"> 907</span>  (w03 + 2.f * w13 + 4.f * w23 + 8.f * w33 + 16.f * w43) + (w04 + 2.f * w14 + 4.f * w24 + 8.f * w34 + 16.f * w44)) / 405.f;</div><div class="line"><a name="l00908"></a><span class="lineno"> 908</span>  out3.s3 = ((w00 + 2.f * w10 + 4.f * w20 + 8.f * w30 + 16.f * w40) + 2.f * (w01 + 2.f * w11 + 4.f * w21 + 8.f * w31 + 16.f * w41) + 4.f * (w02 + 2.f * w12 + 4.f * w22 + 8.f * w32 + 16.f * w42) + 8.f</div><div class="line"><a name="l00909"></a><span class="lineno"> 909</span>  * (w03 + 2.f * w13 + 4.f * w23 + 8.f * w33 + 16.f * w43) + 16.f * (w04 + 2.f * w14 + 4.f * w24 + 8.f * w34 + 16.f * w44)) / 8100.f;</div><div class="line"><a name="l00910"></a><span class="lineno"> 910</span>  out3.s4 = ((w00 + 2.f * w10 + 4.f * w20 + 8.f * w30 + 16.f * w40) - 2.f * (w01 + 2.f * w11 + 4.f * w21 + 8.f * w31 + 16.f * w41) + 4.f * (w02 + 2.f * w12 + 4.f * w22 + 8.f * w32 + 16.f * w42) - 8.f</div><div class="line"><a name="l00911"></a><span class="lineno"> 911</span>  * (w03 + 2.f * w13 + 4.f * w23 + 8.f * w33 + 16.f * w43) + 16.f * (w04 + 2.f * w14 + 4.f * w24 + 8.f * w34 + 16.f * w44)) / 8100.f;</div><div class="line"><a name="l00912"></a><span class="lineno"> 912</span>  out3.s5 = (16.f * (w00 + 2.f * w10 + 4.f * w20 + 8.f * w30 + 16.f * w40) + 8.f * (w01 + 2.f * w11 + 4.f * w21 + 8.f * w31 + 16.f * w41) + 4.f *</div><div class="line"><a name="l00913"></a><span class="lineno"> 913</span>  (w02 + 2.f * w12 + 4.f * w22 + 8.f * w32 + 16.f * w42) + 2.f * (w03 + 2.f * w13 + 4.f * w23 + 8.f * w33 + 16.f * w43) + (w04 + 2.f * w14 + 4.f * w24 + 8.f * w34 + 16.f * w44)) / 16200.f;</div><div class="line"><a name="l00914"></a><span class="lineno"> 914</span>  out3.s6 = (16.f * (w00 + 2.f * w10 + 4.f * w20 + 8.f * w30 + 16.f * w40) - 8.f * (w01 + 2.f * w11 + 4.f * w21 + 8.f * w31 + 16.f * w41) + 4.f *</div><div class="line"><a name="l00915"></a><span class="lineno"> 915</span>  (w02 + 2.f * w12 + 4.f * w22 + 8.f * w32 + 16.f * w42) - 2.f * (w03 + 2.f * w13 + 4.f * w23 + 8.f * w33 + 16.f * w43) + (w04 + 2.f * w14 + 4.f * w24 + 8.f * w34 + 16.f * w44)) / 16200.f;</div><div class="line"><a name="l00916"></a><span class="lineno"> 916</span>  out3.s7 = (w04 + 2.f * w14 + 4.f * w24 + 8.f * w34 + 16.f * w44) / 90.f;</div><div class="line"><a name="l00917"></a><span class="lineno"> 917</span> </div><div class="line"><a name="l00918"></a><span class="lineno"> 918</span>  <span class="comment">// Row 4</span></div><div class="line"><a name="l00919"></a><span class="lineno"> 919</span>  <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)</div><div class="line"><a name="l00920"></a><span class="lineno"> 920</span>  out4 = 0.0f;</div><div class="line"><a name="l00921"></a><span class="lineno"> 921</span>  out4.s0 = (w00 - 2.f * w10 + 4.f * w20 - 8.f * w30 + 16.f * w40) / 90.f;</div><div class="line"><a name="l00922"></a><span class="lineno"> 922</span>  out4.s1 = -((w00 - 2.f * w10 + 4.f * w20 - 8.f * w30 + 16.f * w40) + (w01 - 2.f * w11 + 4.f * w21 - 8.f * w31 + 16.f * w41) + (w02 - 2.f * w12 + 4.f * w22 - 8.f * w32 + 16.f * w42) +</div><div class="line"><a name="l00923"></a><span class="lineno"> 923</span>  (w03 - 2.f * w13 + 4.f * w23 - 8.f * w33 + 16.f * w43) + (w04 - 2.f * w14 + 4.f * w24 - 8.f * w34 + 16.f * w44)) / 405.f;</div><div class="line"><a name="l00924"></a><span class="lineno"> 924</span>  out4.s2 = -((w00 - 2.f * w10 + 4.f * w20 - 8.f * w30 + 16.f * w40) - (w01 - 2.f * w11 + 4.f * w21 - 8.f * w31 + 16.f * w41) + (w02 - 2.f * w12 + 4.f * w22 - 8.f * w32 + 16.f * w42) -</div><div class="line"><a name="l00925"></a><span class="lineno"> 925</span>  (w03 - 2.f * w13 + 4.f * w23 - 8.f * w33 + 16.f * w43) + (w04 - 2.f * w14 + 4.f * w24 - 8.f * w34 + 16.f * w44)) / 405.f;</div><div class="line"><a name="l00926"></a><span class="lineno"> 926</span>  out4.s3 = ((w00 - 2.f * w10 + 4.f * w20 - 8.f * w30 + 16.f * w40) + 2.f * (w01 - 2.f * w11 + 4.f * w21 - 8.f * w31 + 16.f * w41) + 4.f * (w02 - 2.f * w12 + 4.f * w22 - 8.f * w32 + 16.f * w42) + 8.f</div><div class="line"><a name="l00927"></a><span class="lineno"> 927</span>  * (w03 - 2.f * w13 + 4.f * w23 - 8.f * w33 + 16.f * w43) + 16.f * (w04 - 2.f * w14 + 4.f * w24 - 8.f * w34 + 16.f * w44)) / 8100.f;</div><div class="line"><a name="l00928"></a><span class="lineno"> 928</span>  out4.s4 = ((w00 - 2.f * w10 + 4.f * w20 - 8.f * w30 + 16.f * w40) - 2.f * (w01 - 2.f * w11 + 4.f * w21 - 8.f * w31 + 16.f * w41) + 4.f * (w02 - 2.f * w12 + 4.f * w22 - 8.f * w32 + 16.f * w42) - 8.f</div><div class="line"><a name="l00929"></a><span class="lineno"> 929</span>  * (w03 - 2.f * w13 + 4.f * w23 - 8.f * w33 + 16.f * w43) + 16.f * (w04 - 2.f * w14 + 4.f * w24 - 8.f * w34 + 16.f * w44)) / 8100.f;</div><div class="line"><a name="l00930"></a><span class="lineno"> 930</span>  out4.s5 = (16.f * (w00 - 2.f * w10 + 4.f * w20 - 8.f * w30 + 16.f * w40) + 8.f * (w01 - 2.f * w11 + 4.f * w21 - 8.f * w31 + 16.f * w41) + 4.f *</div><div class="line"><a name="l00931"></a><span class="lineno"> 931</span>  (w02 - 2.f * w12 + 4.f * w22 - 8.f * w32 + 16.f * w42) + 2.f * (w03 - 2.f * w13 + 4.f * w23 - 8.f * w33 + 16.f * w43) + (w04 - 2.f * w14 + 4.f * w24 - 8.f * w34 + 16.f * w44)) / 16200.f;</div><div class="line"><a name="l00932"></a><span class="lineno"> 932</span>  out4.s6 = (16.f * (w00 - 2.f * w10 + 4.f * w20 - 8.f * w30 + 16.f * w40) - 8.f * (w01 - 2.f * w11 + 4.f * w21 - 8.f * w31 + 16.f * w41) + 4.f *</div><div class="line"><a name="l00933"></a><span class="lineno"> 933</span>  (w02 - 2.f * w12 + 4.f * w22 - 8.f * w32 + 16.f * w42) - 2.f * (w03 - 2.f * w13 + 4.f * w23 - 8.f * w33 + 16.f * w43) + (w04 - 2.f * w14 + 4.f * w24 - 8.f * w34 + 16.f * w44)) / 16200.f;</div><div class="line"><a name="l00934"></a><span class="lineno"> 934</span>  out4.s7 = (w04 - 2.f * w14 + 4.f * w24 - 8.f * w34 + 16.f * w44) / 90.f;</div><div class="line"><a name="l00935"></a><span class="lineno"> 935</span> </div><div class="line"><a name="l00936"></a><span class="lineno"> 936</span>  <span class="comment">// Row 5</span></div><div class="line"><a name="l00937"></a><span class="lineno"> 937</span>  <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)</div><div class="line"><a name="l00938"></a><span class="lineno"> 938</span>  out5 = 0.0f;</div><div class="line"><a name="l00939"></a><span class="lineno"> 939</span>  out5.s0 = (16.f * w00 + 8.f * w10 + 4.f * w20 + 2.f * w30 + w40) / 180.f;</div><div class="line"><a name="l00940"></a><span class="lineno"> 940</span>  out5.s1 = -((16.f * w00 + 8.f * w10 + 4.f * w20 + 2.f * w30 + w40) + (16.f * w01 + 8.f * w11 + 4.f * w21 + 2.f * w31 + w41) + (16.f * w02 + 8.f * w12 + 4.f * w22 + 2.f * w32 + w42) +</div><div class="line"><a name="l00941"></a><span class="lineno"> 941</span>  (16.f * w03 + 8.f * w13 + 4.f * w23 + 2.f * w33 + w43) + (16.f * w04 + 8.f * w14 + 4.f * w24 + 2.f * w34 + w44)) / 810.f;</div><div class="line"><a name="l00942"></a><span class="lineno"> 942</span>  out5.s2 = -((16.f * w00 + 8.f * w10 + 4.f * w20 + 2.f * w30 + w40) - (16.f * w01 + 8.f * w11 + 4.f * w21 + 2.f * w31 + w41) + (16.f * w02 + 8.f * w12 + 4.f * w22 + 2.f * w32 + w42) -</div><div class="line"><a name="l00943"></a><span class="lineno"> 943</span>  (16.f * w03 + 8.f * w13 + 4.f * w23 + 2.f * w33 + w43) + (16.f * w04 + 8.f * w14 + 4.f * w24 + 2.f * w34 + w44)) / 810.f;</div><div class="line"><a name="l00944"></a><span class="lineno"> 944</span>  out5.s3 = ((16.f * w00 + 8.f * w10 + 4.f * w20 + 2.f * w30 + w40) + 2.f * (16.f * w01 + 8.f * w11 + 4.f * w21 + 2.f * w31 + w41) + 4.f * (16.f * w02 + 8.f * w12 + 4.f * w22 + 2.f * w32 + w42) + 8.f</div><div class="line"><a name="l00945"></a><span class="lineno"> 945</span>  * (16.f * w03 + 8.f * w13 + 4.f * w23 + 2.f * w33 + w43) + 16.f * (16.f * w04 + 8.f * w14 + 4.f * w24 + 2.f * w34 + w44)) / 16200.f;</div><div class="line"><a name="l00946"></a><span class="lineno"> 946</span>  out5.s4 = ((16.f * w00 + 8.f * w10 + 4.f * w20 + 2.f * w30 + w40) - 2.f * (16.f * w01 + 8.f * w11 + 4.f * w21 + 2.f * w31 + w41) + 4.f * (16.f * w02 + 8.f * w12 + 4.f * w22 + 2.f * w32 + w42) - 8.f</div><div class="line"><a name="l00947"></a><span class="lineno"> 947</span>  * (16.f * w03 + 8.f * w13 + 4.f * w23 + 2.f * w33 + w43) + 16.f * (16.f * w04 + 8.f * w14 + 4.f * w24 + 2.f * w34 + w44)) / 16200.f;</div><div class="line"><a name="l00948"></a><span class="lineno"> 948</span>  out5.s5 = (16.f * (16.f * w00 + 8.f * w10 + 4.f * w20 + 2.f * w30 + w40) + 8.f * (16.f * w01 + 8.f * w11 + 4.f * w21 + 2.f * w31 + w41) + 4.f *</div><div class="line"><a name="l00949"></a><span class="lineno"> 949</span>  (16.f * w02 + 8.f * w12 + 4.f * w22 + 2.f * w32 + w42) + 2.f * (16.f * w03 + 8.f * w13 + 4.f * w23 + 2.f * w33 + w43) + (16.f * w04 + 8.f * w14 + 4.f * w24 + 2.f * w34 + w44)) / 32400.f;</div><div class="line"><a name="l00950"></a><span class="lineno"> 950</span>  out5.s6 = (16.f * (16.f * w00 + 8.f * w10 + 4.f * w20 + 2.f * w30 + w40) - 8.f * (16.f * w01 + 8.f * w11 + 4.f * w21 + 2.f * w31 + w41) + 4.f *</div><div class="line"><a name="l00951"></a><span class="lineno"> 951</span>  (16.f * w02 + 8.f * w12 + 4.f * w22 + 2.f * w32 + w42) - 2.f * (16.f * w03 + 8.f * w13 + 4.f * w23 + 2.f * w33 + w43) + (16.f * w04 + 8.f * w14 + 4.f * w24 + 2.f * w34 + w44)) / 32400.f;</div><div class="line"><a name="l00952"></a><span class="lineno"> 952</span>  out5.s7 = (16.f * w04 + 8.f * w14 + 4.f * w24 + 2.f * w34 + w44) / 180.f;</div><div class="line"><a name="l00953"></a><span class="lineno"> 953</span> </div><div class="line"><a name="l00954"></a><span class="lineno"> 954</span>  <span class="comment">// Row 6</span></div><div class="line"><a name="l00955"></a><span class="lineno"> 955</span>  <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)</div><div class="line"><a name="l00956"></a><span class="lineno"> 956</span>  out6 = 0.0f;</div><div class="line"><a name="l00957"></a><span class="lineno"> 957</span>  out6.s0 = (16.f * w00 - 8.f * w10 + 4.f * w20 - 2.f * w30 + w40) / 180.f;</div><div class="line"><a name="l00958"></a><span class="lineno"> 958</span>  out6.s1 = -((16.f * w00 - 8.f * w10 + 4.f * w20 - 2.f * w30 + w40) + (16.f * w01 - 8.f * w11 + 4.f * w21 - 2.f * w31 + w41) + (16.f * w02 - 8.f * w12 + 4.f * w22 - 2.f * w32 + w42) +</div><div class="line"><a name="l00959"></a><span class="lineno"> 959</span>  (16.f * w03 - 8.f * w13 + 4.f * w23 - 2.f * w33 + w43) + (16.f * w04 - 8.f * w14 + 4.f * w24 - 2.f * w34 + w44)) / 810.f;</div><div class="line"><a name="l00960"></a><span class="lineno"> 960</span>  out6.s2 = -((16.f * w00 - 8.f * w10 + 4.f * w20 - 2.f * w30 + w40) - (16.f * w01 - 8.f * w11 + 4.f * w21 - 2.f * w31 + w41) + (16.f * w02 - 8.f * w12 + 4.f * w22 - 2.f * w32 + w42) -</div><div class="line"><a name="l00961"></a><span class="lineno"> 961</span>  (16.f * w03 - 8.f * w13 + 4.f * w23 - 2.f * w33 + w43) + (16.f * w04 - 8.f * w14 + 4.f * w24 - 2.f * w34 + w44)) / 810.f;</div><div class="line"><a name="l00962"></a><span class="lineno"> 962</span>  out6.s3 = ((16.f * w00 - 8.f * w10 + 4.f * w20 - 2.f * w30 + w40) + 2.f * (16.f * w01 - 8.f * w11 + 4.f * w21 - 2.f * w31 + w41) + 4.f * (16.f * w02 - 8.f * w12 + 4.f * w22 - 2.f * w32 + w42) + 8.f</div><div class="line"><a name="l00963"></a><span class="lineno"> 963</span>  * (16.f * w03 - 8.f * w13 + 4.f * w23 - 2.f * w33 + w43) + 16.f * (16.f * w04 - 8.f * w14 + 4.f * w24 - 2.f * w34 + w44)) / 16200.f;</div><div class="line"><a name="l00964"></a><span class="lineno"> 964</span>  out6.s4 = ((16.f * w00 - 8.f * w10 + 4.f * w20 - 2.f * w30 + w40) - 2.f * (16.f * w01 - 8.f * w11 + 4.f * w21 - 2.f * w31 + w41) + 4.f * (16.f * w02 - 8.f * w12 + 4.f * w22 - 2.f * w32 + w42) - 8.f</div><div class="line"><a name="l00965"></a><span class="lineno"> 965</span>  * (16.f * w03 - 8.f * w13 + 4.f * w23 - 2.f * w33 + w43) + 16.f * (16.f * w04 - 8.f * w14 + 4.f * w24 - 2.f * w34 + w44)) / 16200.f;</div><div class="line"><a name="l00966"></a><span class="lineno"> 966</span>  out6.s5 = (16.f * (16.f * w00 - 8.f * w10 + 4.f * w20 - 2.f * w30 + w40) + 8.f * (16.f * w01 - 8.f * w11 + 4.f * w21 - 2.f * w31 + w41) + 4.f *</div><div class="line"><a name="l00967"></a><span class="lineno"> 967</span>  (16.f * w02 - 8.f * w12 + 4.f * w22 - 2.f * w32 + w42) + 2.f * (16.f * w03 - 8.f * w13 + 4.f * w23 - 2.f * w33 + w43) + (16.f * w04 - 8.f * w14 + 4.f * w24 - 2.f * w34 + w44)) / 32400.f;</div><div class="line"><a name="l00968"></a><span class="lineno"> 968</span>  out6.s6 = (16.f * (16.f * w00 - 8.f * w10 + 4.f * w20 - 2.f * w30 + w40) - 8.f * (16.f * w01 - 8.f * w11 + 4.f * w21 - 2.f * w31 + w41) + 4.f *</div><div class="line"><a name="l00969"></a><span class="lineno"> 969</span>  (16.f * w02 - 8.f * w12 + 4.f * w22 - 2.f * w32 + w42) - 2.f * (16.f * w03 - 8.f * w13 + 4.f * w23 - 2.f * w33 + w43) + (16.f * w04 - 8.f * w14 + 4.f * w24 - 2.f * w34 + w44)) / 32400.f;</div><div class="line"><a name="l00970"></a><span class="lineno"> 970</span>  out6.s7 = (16.f * w04 - 8.f * w14 + 4.f * w24 - 2.f * w34 + w44) / 180.f;</div><div class="line"><a name="l00971"></a><span class="lineno"> 971</span> </div><div class="line"><a name="l00972"></a><span class="lineno"> 972</span>  <span class="comment">// Row 7</span></div><div class="line"><a name="l00973"></a><span class="lineno"> 973</span>  <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)</div><div class="line"><a name="l00974"></a><span class="lineno"> 974</span>  out7 = 0.0f;</div><div class="line"><a name="l00975"></a><span class="lineno"> 975</span>  out7.s0 = w40;</div><div class="line"><a name="l00976"></a><span class="lineno"> 976</span>  out7.s1 = -2.f * (w40 + w41 + w42 + w43 + w44) / 9.f;</div><div class="line"><a name="l00977"></a><span class="lineno"> 977</span>  out7.s2 = -2.f * (w40 - w41 + w42 - w43 + w44) / 9.f;</div><div class="line"><a name="l00978"></a><span class="lineno"> 978</span>  out7.s3 = (w40 + 2.f * w41 + 4.f * w42 + 8.f * w43 + 16.f * w44) / 90.f;</div><div class="line"><a name="l00979"></a><span class="lineno"> 979</span>  out7.s4 = (w40 - 2.f * w41 + 4.f * w42 - 8.f * w43 + 16.f * w44) / 90.f;</div><div class="line"><a name="l00980"></a><span class="lineno"> 980</span>  out7.s5 = (16.f * w40 + 8.f * w41 + 4.f * w42 + 2.f * w43 + w44) / 180.f;</div><div class="line"><a name="l00981"></a><span class="lineno"> 981</span>  out7.s6 = (16.f * w40 - 8.f * w41 + 4.f * w42 - 2.f * w43 + w44) / 180.f;</div><div class="line"><a name="l00982"></a><span class="lineno"> 982</span>  out7.s7 = w44;</div><div class="line"><a name="l00983"></a><span class="lineno"> 983</span> <span class="preprocessor">#endif // !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l00984"></a><span class="lineno"> 984</span> </div><div class="line"><a name="l00985"></a><span class="lineno"> 985</span>  <span class="keywordtype">int</span> x0 = get_global_id(2); <span class="comment">// idx filter</span></div><div class="line"><a name="l00986"></a><span class="lineno"> 986</span>  <span class="keywordtype">int</span> y0 = get_global_id(0); <span class="comment">// idx channel</span></div><div class="line"><a name="l00987"></a><span class="lineno"> 987</span> </div><div class="line"><a name="l00988"></a><span class="lineno"> 988</span>  <span class="comment">// Get output address</span></div><div class="line"><a name="l00989"></a><span class="lineno"> 989</span>  __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x0 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>) + y0 * dst_stride_y;</div><div class="line"><a name="l00990"></a><span class="lineno"> 990</span> </div><div class="line"><a name="l00991"></a><span class="lineno"> 991</span>  <span class="comment">// Store the values across the channels</span></div><div class="line"><a name="l00992"></a><span class="lineno"> 992</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 0 * dst_stride_z) = out0.s0;</div><div class="line"><a name="l00993"></a><span class="lineno"> 993</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 1 * dst_stride_z) = out0.s1;</div><div class="line"><a name="l00994"></a><span class="lineno"> 994</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 2 * dst_stride_z) = out0.s2;</div><div class="line"><a name="l00995"></a><span class="lineno"> 995</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 3 * dst_stride_z) = out0.s3;</div><div class="line"><a name="l00996"></a><span class="lineno"> 996</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 4 * dst_stride_z) = out0.s4;</div><div class="line"><a name="l00997"></a><span class="lineno"> 997</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 5 * dst_stride_z) = out0.s5;</div><div class="line"><a name="l00998"></a><span class="lineno"> 998</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 6 * dst_stride_z) = out0.s6;</div><div class="line"><a name="l00999"></a><span class="lineno"> 999</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 7 * dst_stride_z) = out0.s7;</div><div class="line"><a name="l01000"></a><span class="lineno"> 1000</span> </div><div class="line"><a name="l01001"></a><span class="lineno"> 1001</span> <span class="preprocessor">#if !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l01002"></a><span class="lineno"> 1002</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 8 * dst_stride_z) = out1.s0;</div><div class="line"><a name="l01003"></a><span class="lineno"> 1003</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 9 * dst_stride_z) = out1.s1;</div><div class="line"><a name="l01004"></a><span class="lineno"> 1004</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 10 * dst_stride_z) = out1.s2;</div><div class="line"><a name="l01005"></a><span class="lineno"> 1005</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 11 * dst_stride_z) = out1.s3;</div><div class="line"><a name="l01006"></a><span class="lineno"> 1006</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 12 * dst_stride_z) = out1.s4;</div><div class="line"><a name="l01007"></a><span class="lineno"> 1007</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 13 * dst_stride_z) = out1.s5;</div><div class="line"><a name="l01008"></a><span class="lineno"> 1008</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 14 * dst_stride_z) = out1.s6;</div><div class="line"><a name="l01009"></a><span class="lineno"> 1009</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 15 * dst_stride_z) = out1.s7;</div><div class="line"><a name="l01010"></a><span class="lineno"> 1010</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 16 * dst_stride_z) = out2.s0;</div><div class="line"><a name="l01011"></a><span class="lineno"> 1011</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 17 * dst_stride_z) = out2.s1;</div><div class="line"><a name="l01012"></a><span class="lineno"> 1012</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 18 * dst_stride_z) = out2.s2;</div><div class="line"><a name="l01013"></a><span class="lineno"> 1013</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 19 * dst_stride_z) = out2.s3;</div><div class="line"><a name="l01014"></a><span class="lineno"> 1014</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 20 * dst_stride_z) = out2.s4;</div><div class="line"><a name="l01015"></a><span class="lineno"> 1015</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 21 * dst_stride_z) = out2.s5;</div><div class="line"><a name="l01016"></a><span class="lineno"> 1016</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 22 * dst_stride_z) = out2.s6;</div><div class="line"><a name="l01017"></a><span class="lineno"> 1017</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 23 * dst_stride_z) = out2.s7;</div><div class="line"><a name="l01018"></a><span class="lineno"> 1018</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 24 * dst_stride_z) = out3.s0;</div><div class="line"><a name="l01019"></a><span class="lineno"> 1019</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 25 * dst_stride_z) = out3.s1;</div><div class="line"><a name="l01020"></a><span class="lineno"> 1020</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 26 * dst_stride_z) = out3.s2;</div><div class="line"><a name="l01021"></a><span class="lineno"> 1021</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 27 * dst_stride_z) = out3.s3;</div><div class="line"><a name="l01022"></a><span class="lineno"> 1022</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 28 * dst_stride_z) = out3.s4;</div><div class="line"><a name="l01023"></a><span class="lineno"> 1023</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 29 * dst_stride_z) = out3.s5;</div><div class="line"><a name="l01024"></a><span class="lineno"> 1024</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 30 * dst_stride_z) = out3.s6;</div><div class="line"><a name="l01025"></a><span class="lineno"> 1025</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 31 * dst_stride_z) = out3.s7;</div><div class="line"><a name="l01026"></a><span class="lineno"> 1026</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 32 * dst_stride_z) = out4.s0;</div><div class="line"><a name="l01027"></a><span class="lineno"> 1027</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 33 * dst_stride_z) = out4.s1;</div><div class="line"><a name="l01028"></a><span class="lineno"> 1028</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 34 * dst_stride_z) = out4.s2;</div><div class="line"><a name="l01029"></a><span class="lineno"> 1029</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 35 * dst_stride_z) = out4.s3;</div><div class="line"><a name="l01030"></a><span class="lineno"> 1030</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 36 * dst_stride_z) = out4.s4;</div><div class="line"><a name="l01031"></a><span class="lineno"> 1031</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 37 * dst_stride_z) = out4.s5;</div><div class="line"><a name="l01032"></a><span class="lineno"> 1032</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 38 * dst_stride_z) = out4.s6;</div><div class="line"><a name="l01033"></a><span class="lineno"> 1033</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 39 * dst_stride_z) = out4.s7;</div><div class="line"><a name="l01034"></a><span class="lineno"> 1034</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 40 * dst_stride_z) = out5.s0;</div><div class="line"><a name="l01035"></a><span class="lineno"> 1035</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 41 * dst_stride_z) = out5.s1;</div><div class="line"><a name="l01036"></a><span class="lineno"> 1036</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 42 * dst_stride_z) = out5.s2;</div><div class="line"><a name="l01037"></a><span class="lineno"> 1037</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 43 * dst_stride_z) = out5.s3;</div><div class="line"><a name="l01038"></a><span class="lineno"> 1038</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 44 * dst_stride_z) = out5.s4;</div><div class="line"><a name="l01039"></a><span class="lineno"> 1039</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 45 * dst_stride_z) = out5.s5;</div><div class="line"><a name="l01040"></a><span class="lineno"> 1040</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 46 * dst_stride_z) = out5.s6;</div><div class="line"><a name="l01041"></a><span class="lineno"> 1041</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 47 * dst_stride_z) = out5.s7;</div><div class="line"><a name="l01042"></a><span class="lineno"> 1042</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 48 * dst_stride_z) = out6.s0;</div><div class="line"><a name="l01043"></a><span class="lineno"> 1043</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 49 * dst_stride_z) = out6.s1;</div><div class="line"><a name="l01044"></a><span class="lineno"> 1044</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 50 * dst_stride_z) = out6.s2;</div><div class="line"><a name="l01045"></a><span class="lineno"> 1045</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 51 * dst_stride_z) = out6.s3;</div><div class="line"><a name="l01046"></a><span class="lineno"> 1046</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 52 * dst_stride_z) = out6.s4;</div><div class="line"><a name="l01047"></a><span class="lineno"> 1047</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 53 * dst_stride_z) = out6.s5;</div><div class="line"><a name="l01048"></a><span class="lineno"> 1048</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 54 * dst_stride_z) = out6.s6;</div><div class="line"><a name="l01049"></a><span class="lineno"> 1049</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 55 * dst_stride_z) = out6.s7;</div><div class="line"><a name="l01050"></a><span class="lineno"> 1050</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 56 * dst_stride_z) = out7.s0;</div><div class="line"><a name="l01051"></a><span class="lineno"> 1051</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 57 * dst_stride_z) = out7.s1;</div><div class="line"><a name="l01052"></a><span class="lineno"> 1052</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 58 * dst_stride_z) = out7.s2;</div><div class="line"><a name="l01053"></a><span class="lineno"> 1053</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 59 * dst_stride_z) = out7.s3;</div><div class="line"><a name="l01054"></a><span class="lineno"> 1054</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 60 * dst_stride_z) = out7.s4;</div><div class="line"><a name="l01055"></a><span class="lineno"> 1055</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 61 * dst_stride_z) = out7.s5;</div><div class="line"><a name="l01056"></a><span class="lineno"> 1056</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 62 * dst_stride_z) = out7.s6;</div><div class="line"><a name="l01057"></a><span class="lineno"> 1057</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 63 * dst_stride_z) = out7.s7;</div><div class="line"><a name="l01058"></a><span class="lineno"> 1058</span> <span class="preprocessor">#endif // !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l01059"></a><span class="lineno"> 1059</span> }<span class="comment"></span></div><div class="line"><a name="l01060"></a><span class="lineno"> 1060</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 7x7/7x1 or 1x7 when the data layout is NHWC and the output tile is 2x2/2x1 or 1x2</span></div><div class="line"><a name="l01061"></a><span class="lineno"> 1061</span> <span class="comment"> *</span></div><div class="line"><a name="l01062"></a><span class="lineno"> 1062</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l01063"></a><span class="lineno"> 1063</span> <span class="comment"> * @note If this kernel is used to perform Winograd filter transform 7x1, -DWINOGRAD_FILTER_TRANSFORM_HORIZONTAL has to be passed at compile time</span></div><div class="line"><a name="l01064"></a><span class="lineno"> 1064</span> <span class="comment"> * @note If this kernel is used to perform Winograd filter transform 1x7, -DWINOGRAD_FILTER_TRANSFORM_VERTICAL has to be passed at compile time</span></div><div class="line"><a name="l01065"></a><span class="lineno"> 1065</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l01066"></a><span class="lineno"> 1066</span> <span class="comment"> *</span></div><div class="line"><a name="l01067"></a><span class="lineno"> 1067</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l01068"></a><span class="lineno"> 1068</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01069"></a><span class="lineno"> 1069</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01070"></a><span class="lineno"> 1070</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01071"></a><span class="lineno"> 1071</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01072"></a><span class="lineno"> 1072</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01073"></a><span class="lineno"> 1073</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01074"></a><span class="lineno"> 1074</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l01075"></a><span class="lineno"> 1075</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l01076"></a><span class="lineno"> 1076</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01077"></a><span class="lineno"> 1077</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l01078"></a><span class="lineno"> 1078</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01079"></a><span class="lineno"> 1079</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01080"></a><span class="lineno"> 1080</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01081"></a><span class="lineno"> 1081</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01082"></a><span class="lineno"> 1082</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01083"></a><span class="lineno"> 1083</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01084"></a><span class="lineno"> 1084</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l01085"></a><span class="lineno"> 1085</span> <span class="comment"> */</span></div><div class="line"><a name="l01086"></a><span class="lineno"> 1086</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_2x2_7x7_nhwc(</div><div class="line"><a name="l01087"></a><span class="lineno"> 1087</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01088"></a><span class="lineno"> 1088</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01089"></a><span class="lineno"> 1089</span> {</div><div class="line"><a name="l01090"></a><span class="lineno"> 1090</span>  <a class="code" href="struct_tensor4_d.xhtml">Tensor4D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</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#a989ab3e96426615bb98e04e0235088ca">src</a>, SRC_DIM_Z);</div><div class="line"><a name="l01091"></a><span class="lineno"> 1091</span> </div><div class="line"><a name="l01092"></a><span class="lineno"> 1092</span>  <span class="keyword">const</span> __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + get_global_id(0) * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>) + get_global_id(1) * src_step_y + get_global_id(2) * src_step_w;</div><div class="line"><a name="l01093"></a><span class="lineno"> 1093</span> </div><div class="line"><a name="l01094"></a><span class="lineno"> 1094</span> <span class="preprocessor">#if defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l01095"></a><span class="lineno"> 1095</span>  <span class="comment">// Load the values from the input tensor</span></div><div class="line"><a name="l01096"></a><span class="lineno"> 1096</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w00 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_z));</div><div class="line"><a name="l01097"></a><span class="lineno"> 1097</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w01 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z));</div><div class="line"><a name="l01098"></a><span class="lineno"> 1098</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w02 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z));</div><div class="line"><a name="l01099"></a><span class="lineno"> 1099</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w03 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_z));</div><div class="line"><a name="l01100"></a><span class="lineno"> 1100</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w04 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_z));</div><div class="line"><a name="l01101"></a><span class="lineno"> 1101</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w05 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 5 * src_stride_z));</div><div class="line"><a name="l01102"></a><span class="lineno"> 1102</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w06 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 6 * src_stride_z));</div><div class="line"><a name="l01103"></a><span class="lineno"> 1103</span> <span class="preprocessor">#else // defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l01104"></a><span class="lineno"> 1104</span>  <span class="comment">// Load the values from the input tensor</span></div><div class="line"><a name="l01105"></a><span class="lineno"> 1105</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w00 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_y));</div><div class="line"><a name="l01106"></a><span class="lineno"> 1106</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w01 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_y));</div><div class="line"><a name="l01107"></a><span class="lineno"> 1107</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w02 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_y));</div><div class="line"><a name="l01108"></a><span class="lineno"> 1108</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w03 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_y));</div><div class="line"><a name="l01109"></a><span class="lineno"> 1109</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w04 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_y));</div><div class="line"><a name="l01110"></a><span class="lineno"> 1110</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w05 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 5 * src_stride_y));</div><div class="line"><a name="l01111"></a><span class="lineno"> 1111</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w06 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 6 * src_stride_y));</div><div class="line"><a name="l01112"></a><span class="lineno"> 1112</span> <span class="preprocessor">#endif // defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l01113"></a><span class="lineno"> 1113</span> </div><div class="line"><a name="l01114"></a><span class="lineno"> 1114</span> <span class="preprocessor">#if !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l01115"></a><span class="lineno"> 1115</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w10 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z + 0 * src_stride_y));</div><div class="line"><a name="l01116"></a><span class="lineno"> 1116</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w11 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z + 1 * src_stride_y));</div><div class="line"><a name="l01117"></a><span class="lineno"> 1117</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w12 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z + 2 * src_stride_y));</div><div class="line"><a name="l01118"></a><span class="lineno"> 1118</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w13 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z + 3 * src_stride_y));</div><div class="line"><a name="l01119"></a><span class="lineno"> 1119</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w14 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z + 4 * src_stride_y));</div><div class="line"><a name="l01120"></a><span class="lineno"> 1120</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w15 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z + 5 * src_stride_y));</div><div class="line"><a name="l01121"></a><span class="lineno"> 1121</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w16 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_z + 6 * src_stride_y));</div><div class="line"><a name="l01122"></a><span class="lineno"> 1122</span> </div><div class="line"><a name="l01123"></a><span class="lineno"> 1123</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w20 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z + 0 * src_stride_y));</div><div class="line"><a name="l01124"></a><span class="lineno"> 1124</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w21 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z + 1 * src_stride_y));</div><div class="line"><a name="l01125"></a><span class="lineno"> 1125</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w22 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z + 2 * src_stride_y));</div><div class="line"><a name="l01126"></a><span class="lineno"> 1126</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w23 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z + 3 * src_stride_y));</div><div class="line"><a name="l01127"></a><span class="lineno"> 1127</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w24 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z + 4 * src_stride_y));</div><div class="line"><a name="l01128"></a><span class="lineno"> 1128</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w25 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z + 5 * src_stride_y));</div><div class="line"><a name="l01129"></a><span class="lineno"> 1129</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w26 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_z + 6 * src_stride_y));</div><div class="line"><a name="l01130"></a><span class="lineno"> 1130</span> </div><div class="line"><a name="l01131"></a><span class="lineno"> 1131</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w30 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_z + 0 * src_stride_y));</div><div class="line"><a name="l01132"></a><span class="lineno"> 1132</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w31 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_z + 1 * src_stride_y));</div><div class="line"><a name="l01133"></a><span class="lineno"> 1133</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w32 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_z + 2 * src_stride_y));</div><div class="line"><a name="l01134"></a><span class="lineno"> 1134</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w33 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_z + 3 * src_stride_y));</div><div class="line"><a name="l01135"></a><span class="lineno"> 1135</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w34 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_z + 4 * src_stride_y));</div><div class="line"><a name="l01136"></a><span class="lineno"> 1136</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w35 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_z + 5 * src_stride_y));</div><div class="line"><a name="l01137"></a><span class="lineno"> 1137</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w36 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 3 * src_stride_z + 6 * src_stride_y));</div><div class="line"><a name="l01138"></a><span class="lineno"> 1138</span> </div><div class="line"><a name="l01139"></a><span class="lineno"> 1139</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w40 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_z + 0 * src_stride_y));</div><div class="line"><a name="l01140"></a><span class="lineno"> 1140</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w41 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_z + 1 * src_stride_y));</div><div class="line"><a name="l01141"></a><span class="lineno"> 1141</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w42 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_z + 2 * src_stride_y));</div><div class="line"><a name="l01142"></a><span class="lineno"> 1142</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w43 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_z + 3 * src_stride_y));</div><div class="line"><a name="l01143"></a><span class="lineno"> 1143</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w44 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_z + 4 * src_stride_y));</div><div class="line"><a name="l01144"></a><span class="lineno"> 1144</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w45 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_z + 5 * src_stride_y));</div><div class="line"><a name="l01145"></a><span class="lineno"> 1145</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w46 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 4 * src_stride_z + 6 * src_stride_y));</div><div class="line"><a name="l01146"></a><span class="lineno"> 1146</span> </div><div class="line"><a name="l01147"></a><span class="lineno"> 1147</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w50 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 5 * src_stride_z + 0 * src_stride_y));</div><div class="line"><a name="l01148"></a><span class="lineno"> 1148</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w51 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 5 * src_stride_z + 1 * src_stride_y));</div><div class="line"><a name="l01149"></a><span class="lineno"> 1149</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w52 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 5 * src_stride_z + 2 * src_stride_y));</div><div class="line"><a name="l01150"></a><span class="lineno"> 1150</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w53 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 5 * src_stride_z + 3 * src_stride_y));</div><div class="line"><a name="l01151"></a><span class="lineno"> 1151</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w54 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 5 * src_stride_z + 4 * src_stride_y));</div><div class="line"><a name="l01152"></a><span class="lineno"> 1152</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w55 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 5 * src_stride_z + 5 * src_stride_y));</div><div class="line"><a name="l01153"></a><span class="lineno"> 1153</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w56 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 5 * src_stride_z + 6 * src_stride_y));</div><div class="line"><a name="l01154"></a><span class="lineno"> 1154</span> </div><div class="line"><a name="l01155"></a><span class="lineno"> 1155</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w60 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 6 * src_stride_z + 0 * src_stride_y));</div><div class="line"><a name="l01156"></a><span class="lineno"> 1156</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w61 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 6 * src_stride_z + 1 * src_stride_y));</div><div class="line"><a name="l01157"></a><span class="lineno"> 1157</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w62 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 6 * src_stride_z + 2 * src_stride_y));</div><div class="line"><a name="l01158"></a><span class="lineno"> 1158</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w63 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 6 * src_stride_z + 3 * src_stride_y));</div><div class="line"><a name="l01159"></a><span class="lineno"> 1159</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w64 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 6 * src_stride_z + 4 * src_stride_y));</div><div class="line"><a name="l01160"></a><span class="lineno"> 1160</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w65 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 6 * src_stride_z + 5 * src_stride_y));</div><div class="line"><a name="l01161"></a><span class="lineno"> 1161</span>  <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> w66 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 6 * src_stride_z + 6 * src_stride_y));</div><div class="line"><a name="l01162"></a><span class="lineno"> 1162</span> </div><div class="line"><a name="l01163"></a><span class="lineno"> 1163</span> <span class="preprocessor">#endif // !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l01164"></a><span class="lineno"> 1164</span> </div><div class="line"><a name="l01165"></a><span class="lineno"> 1165</span>  <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)</div><div class="line"><a name="l01166"></a><span class="lineno"> 1166</span>  tmp = 0.0f;</div><div class="line"><a name="l01167"></a><span class="lineno"> 1167</span> </div><div class="line"><a name="l01168"></a><span class="lineno"> 1168</span>  <span class="comment">// Row 0</span></div><div class="line"><a name="l01169"></a><span class="lineno"> 1169</span>  <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)</div><div class="line"><a name="l01170"></a><span class="lineno"> 1170</span>  out0 = 0.0f;</div><div class="line"><a name="l01171"></a><span class="lineno"> 1171</span> </div><div class="line"><a name="l01172"></a><span class="lineno"> 1172</span>  out0.s0 = -w00 / 36.0f;</div><div class="line"><a name="l01173"></a><span class="lineno"> 1173</span>  out0.s1 = (w00 - w01 + w02 - w03 + w04 - w05 + w06) / 48.f;</div><div class="line"><a name="l01174"></a><span class="lineno"> 1174</span>  out0.s2 = (w00 + w01 + w02 + w03 + w04 + w05 + w06) / 48.f;</div><div class="line"><a name="l01175"></a><span class="lineno"> 1175</span>  out0.s3 = (-w00 + 2.f * w01 - 4.f * w02 + 8.f * w03 - 16.f * w04 + 32.f * w05 - 64.f * w06) / 120.f;</div><div class="line"><a name="l01176"></a><span class="lineno"> 1176</span>  out0.s4 = (-w00 - 2.f * w01 - 4.f * w02 - 8.f * w03 - 16.f * w04 - 32.f * w05 - 64.f * w06) / 120.f;</div><div class="line"><a name="l01177"></a><span class="lineno"> 1177</span>  out0.s5 = (w00 - 3.f * w01 + 9.f * w02 - 27.f * w03 + 81.f * w04 - 243.f * w05 + 729.f * w06) / 720.f;</div><div class="line"><a name="l01178"></a><span class="lineno"> 1178</span>  out0.s6 = (w00 + 3.f * w01 + 9.f * w02 + 27.f * w03 + 81.f * w04 + 243.f * w05 + 729.f * w06) / 720.f;</div><div class="line"><a name="l01179"></a><span class="lineno"> 1179</span>  out0.s7 = w06;</div><div class="line"><a name="l01180"></a><span class="lineno"> 1180</span> </div><div class="line"><a name="l01181"></a><span class="lineno"> 1181</span>  out0 /= (<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)) - 36.f;</div><div class="line"><a name="l01182"></a><span class="lineno"> 1182</span> </div><div class="line"><a name="l01183"></a><span class="lineno"> 1183</span> <span class="preprocessor">#if !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l01184"></a><span class="lineno"> 1184</span> </div><div class="line"><a name="l01185"></a><span class="lineno"> 1185</span>  <span class="comment">// Row 1</span></div><div class="line"><a name="l01186"></a><span class="lineno"> 1186</span>  <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)</div><div class="line"><a name="l01187"></a><span class="lineno"> 1187</span>  out1 = 0.0f;</div><div class="line"><a name="l01188"></a><span class="lineno"> 1188</span> </div><div class="line"><a name="l01189"></a><span class="lineno"> 1189</span>  tmp.s0 = (w00 - w10 + w20 - w30 + w40 - w50 + w60) / 48.f;</div><div class="line"><a name="l01190"></a><span class="lineno"> 1190</span>  tmp.s1 = (w01 - w11 + w21 - w31 + w41 - w51 + w61) / 48.f;</div><div class="line"><a name="l01191"></a><span class="lineno"> 1191</span>  tmp.s2 = (w02 - w12 + w22 - w32 + w42 - w52 + w62) / 48.f;</div><div class="line"><a name="l01192"></a><span class="lineno"> 1192</span>  tmp.s3 = (w03 - w13 + w23 - w33 + w43 - w53 + w63) / 48.f;</div><div class="line"><a name="l01193"></a><span class="lineno"> 1193</span>  tmp.s4 = (w04 - w14 + w24 - w34 + w44 - w54 + w64) / 48.f;</div><div class="line"><a name="l01194"></a><span class="lineno"> 1194</span>  tmp.s5 = (w05 - w15 + w25 - w35 + w45 - w55 + w65) / 48.f;</div><div class="line"><a name="l01195"></a><span class="lineno"> 1195</span>  tmp.s6 = (w06 - w16 + w26 - w36 + w46 - w56 + w66) / 48.f;</div><div class="line"><a name="l01196"></a><span class="lineno"> 1196</span> </div><div class="line"><a name="l01197"></a><span class="lineno"> 1197</span>  <a class="code" href="winograd__input__transform_8cl.xhtml#a4b01af9651d23ef289e82a4db19aac7f">OUTPUT_ROW_2x2_7x7</a>(out1, tmp);</div><div class="line"><a name="l01198"></a><span class="lineno"> 1198</span> </div><div class="line"><a name="l01199"></a><span class="lineno"> 1199</span>  <span class="comment">// Row 2</span></div><div class="line"><a name="l01200"></a><span class="lineno"> 1200</span>  <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)</div><div class="line"><a name="l01201"></a><span class="lineno"> 1201</span>  out2 = 0.0f;</div><div class="line"><a name="l01202"></a><span class="lineno"> 1202</span> </div><div class="line"><a name="l01203"></a><span class="lineno"> 1203</span>  tmp.s0 = (w00 + w10 + w20 + w30 + w40 + w50 + w60) / 48.f;</div><div class="line"><a name="l01204"></a><span class="lineno"> 1204</span>  tmp.s1 = (w01 + w11 + w21 + w31 + w41 + w51 + w61) / 48.f;</div><div class="line"><a name="l01205"></a><span class="lineno"> 1205</span>  tmp.s2 = (w02 + w12 + w22 + w32 + w42 + w52 + w62) / 48.f;</div><div class="line"><a name="l01206"></a><span class="lineno"> 1206</span>  tmp.s3 = (w03 + w13 + w23 + w33 + w43 + w53 + w63) / 48.f;</div><div class="line"><a name="l01207"></a><span class="lineno"> 1207</span>  tmp.s4 = (w04 + w14 + w24 + w34 + w44 + w54 + w64) / 48.f;</div><div class="line"><a name="l01208"></a><span class="lineno"> 1208</span>  tmp.s5 = (w05 + w15 + w25 + w35 + w45 + w55 + w65) / 48.f;</div><div class="line"><a name="l01209"></a><span class="lineno"> 1209</span>  tmp.s6 = (w06 + w16 + w26 + w36 + w46 + w56 + w66) / 48.f;</div><div class="line"><a name="l01210"></a><span class="lineno"> 1210</span> </div><div class="line"><a name="l01211"></a><span class="lineno"> 1211</span>  <a class="code" href="winograd__input__transform_8cl.xhtml#a4b01af9651d23ef289e82a4db19aac7f">OUTPUT_ROW_2x2_7x7</a>(out2, tmp);</div><div class="line"><a name="l01212"></a><span class="lineno"> 1212</span> </div><div class="line"><a name="l01213"></a><span class="lineno"> 1213</span>  <span class="comment">// Row 3</span></div><div class="line"><a name="l01214"></a><span class="lineno"> 1214</span>  <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)</div><div class="line"><a name="l01215"></a><span class="lineno"> 1215</span>  out3 = 0.0f;</div><div class="line"><a name="l01216"></a><span class="lineno"> 1216</span> </div><div class="line"><a name="l01217"></a><span class="lineno"> 1217</span>  tmp.s0 = (-w00 + 2.f * w10 - 4.f * w20 + 8.f * w30 - 16.f * w40 + 32.f * w50 - 64.f * w60) / 120.f;</div><div class="line"><a name="l01218"></a><span class="lineno"> 1218</span>  tmp.s1 = (-w01 + 2.f * w11 - 4.f * w21 + 8.f * w31 - 16.f * w41 + 32.f * w51 - 64.f * w61) / 120.f;</div><div class="line"><a name="l01219"></a><span class="lineno"> 1219</span>  tmp.s2 = (-w02 + 2.f * w12 - 4.f * w22 + 8.f * w32 - 16.f * w42 + 32.f * w52 - 64.f * w62) / 120.f;</div><div class="line"><a name="l01220"></a><span class="lineno"> 1220</span>  tmp.s3 = (-w03 + 2.f * w13 - 4.f * w23 + 8.f * w33 - 16.f * w43 + 32.f * w53 - 64.f * w63) / 120.f;</div><div class="line"><a name="l01221"></a><span class="lineno"> 1221</span>  tmp.s4 = (-w04 + 2.f * w14 - 4.f * w24 + 8.f * w34 - 16.f * w44 + 32.f * w54 - 64.f * w64) / 120.f;</div><div class="line"><a name="l01222"></a><span class="lineno"> 1222</span>  tmp.s5 = (-w05 + 2.f * w15 - 4.f * w25 + 8.f * w35 - 16.f * w45 + 32.f * w55 - 64.f * w65) / 120.f;</div><div class="line"><a name="l01223"></a><span class="lineno"> 1223</span>  tmp.s6 = (-w06 + 2.f * w16 - 4.f * w26 + 8.f * w36 - 16.f * w46 + 32.f * w56 - 64.f * w66) / 120.f;</div><div class="line"><a name="l01224"></a><span class="lineno"> 1224</span> </div><div class="line"><a name="l01225"></a><span class="lineno"> 1225</span>  <a class="code" href="winograd__input__transform_8cl.xhtml#a4b01af9651d23ef289e82a4db19aac7f">OUTPUT_ROW_2x2_7x7</a>(out3, tmp);</div><div class="line"><a name="l01226"></a><span class="lineno"> 1226</span> </div><div class="line"><a name="l01227"></a><span class="lineno"> 1227</span>  <span class="comment">// Row 4</span></div><div class="line"><a name="l01228"></a><span class="lineno"> 1228</span>  <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)</div><div class="line"><a name="l01229"></a><span class="lineno"> 1229</span>  out4 = 0.0f;</div><div class="line"><a name="l01230"></a><span class="lineno"> 1230</span> </div><div class="line"><a name="l01231"></a><span class="lineno"> 1231</span>  tmp.s0 = (-w00 - 2.f * w10 - 4.f * w20 - 8.f * w30 - 16.f * w40 - 32.f * w50 - 64.f * w60) / 120.f;</div><div class="line"><a name="l01232"></a><span class="lineno"> 1232</span>  tmp.s1 = (-w01 - 2.f * w11 - 4.f * w21 - 8.f * w31 - 16.f * w41 - 32.f * w51 - 64.f * w61) / 120.f;</div><div class="line"><a name="l01233"></a><span class="lineno"> 1233</span>  tmp.s2 = (-w02 - 2.f * w12 - 4.f * w22 - 8.f * w32 - 16.f * w42 - 32.f * w52 - 64.f * w62) / 120.f;</div><div class="line"><a name="l01234"></a><span class="lineno"> 1234</span>  tmp.s3 = (-w03 - 2.f * w13 - 4.f * w23 - 8.f * w33 - 16.f * w43 - 32.f * w53 - 64.f * w63) / 120.f;</div><div class="line"><a name="l01235"></a><span class="lineno"> 1235</span>  tmp.s4 = (-w04 - 2.f * w14 - 4.f * w24 - 8.f * w34 - 16.f * w44 - 32.f * w54 - 64.f * w64) / 120.f;</div><div class="line"><a name="l01236"></a><span class="lineno"> 1236</span>  tmp.s5 = (-w05 - 2.f * w15 - 4.f * w25 - 8.f * w35 - 16.f * w45 - 32.f * w55 - 64.f * w65) / 120.f;</div><div class="line"><a name="l01237"></a><span class="lineno"> 1237</span>  tmp.s6 = (-w06 - 2.f * w16 - 4.f * w26 - 8.f * w36 - 16.f * w46 - 32.f * w56 - 64.f * w66) / 120.f;</div><div class="line"><a name="l01238"></a><span class="lineno"> 1238</span> </div><div class="line"><a name="l01239"></a><span class="lineno"> 1239</span>  <a class="code" href="winograd__input__transform_8cl.xhtml#a4b01af9651d23ef289e82a4db19aac7f">OUTPUT_ROW_2x2_7x7</a>(out4, tmp);</div><div class="line"><a name="l01240"></a><span class="lineno"> 1240</span> </div><div class="line"><a name="l01241"></a><span class="lineno"> 1241</span>  <span class="comment">// Row 5</span></div><div class="line"><a name="l01242"></a><span class="lineno"> 1242</span>  <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)</div><div class="line"><a name="l01243"></a><span class="lineno"> 1243</span>  out5 = 0.0f;</div><div class="line"><a name="l01244"></a><span class="lineno"> 1244</span> </div><div class="line"><a name="l01245"></a><span class="lineno"> 1245</span>  tmp.s0 = (w00 - 3.f * w10 + 9.f * w20 - 27.f * w30 + 81.f * w40 - 243.f * w50 + 729.f * w60) / 720.f;</div><div class="line"><a name="l01246"></a><span class="lineno"> 1246</span>  tmp.s1 = (w01 - 3.f * w11 + 9.f * w21 - 27.f * w31 + 81.f * w41 - 243.f * w51 + 729.f * w61) / 720.f;</div><div class="line"><a name="l01247"></a><span class="lineno"> 1247</span>  tmp.s2 = (w02 - 3.f * w12 + 9.f * w22 - 27.f * w32 + 81.f * w42 - 243.f * w52 + 729.f * w62) / 720.f;</div><div class="line"><a name="l01248"></a><span class="lineno"> 1248</span>  tmp.s3 = (w03 - 3.f * w13 + 9.f * w23 - 27.f * w33 + 81.f * w43 - 243.f * w53 + 729.f * w63) / 720.f;</div><div class="line"><a name="l01249"></a><span class="lineno"> 1249</span>  tmp.s4 = (w04 - 3.f * w14 + 9.f * w24 - 27.f * w34 + 81.f * w44 - 243.f * w54 + 729.f * w64) / 720.f;</div><div class="line"><a name="l01250"></a><span class="lineno"> 1250</span>  tmp.s5 = (w05 - 3.f * w15 + 9.f * w25 - 27.f * w35 + 81.f * w45 - 243.f * w55 + 729.f * w65) / 720.f;</div><div class="line"><a name="l01251"></a><span class="lineno"> 1251</span>  tmp.s6 = (w06 - 3.f * w16 + 9.f * w26 - 27.f * w36 + 81.f * w46 - 243.f * w56 + 729.f * w66) / 720.f;</div><div class="line"><a name="l01252"></a><span class="lineno"> 1252</span> </div><div class="line"><a name="l01253"></a><span class="lineno"> 1253</span>  <a class="code" href="winograd__input__transform_8cl.xhtml#a4b01af9651d23ef289e82a4db19aac7f">OUTPUT_ROW_2x2_7x7</a>(out5, tmp);</div><div class="line"><a name="l01254"></a><span class="lineno"> 1254</span> </div><div class="line"><a name="l01255"></a><span class="lineno"> 1255</span>  <span class="comment">// Row 6</span></div><div class="line"><a name="l01256"></a><span class="lineno"> 1256</span>  <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)</div><div class="line"><a name="l01257"></a><span class="lineno"> 1257</span>  out6 = 0.0f;</div><div class="line"><a name="l01258"></a><span class="lineno"> 1258</span> </div><div class="line"><a name="l01259"></a><span class="lineno"> 1259</span>  tmp.s0 = (w00 + 3.f * w10 + 9.f * w20 + 27.f * w30 + 81.f * w40 + 243.f * w50 + 729.f * w60) / 720.f;</div><div class="line"><a name="l01260"></a><span class="lineno"> 1260</span>  tmp.s1 = (w01 + 3.f * w11 + 9.f * w21 + 27.f * w31 + 81.f * w41 + 243.f * w51 + 729.f * w61) / 720.f;</div><div class="line"><a name="l01261"></a><span class="lineno"> 1261</span>  tmp.s2 = (w02 + 3.f * w12 + 9.f * w22 + 27.f * w32 + 81.f * w42 + 243.f * w52 + 729.f * w62) / 720.f;</div><div class="line"><a name="l01262"></a><span class="lineno"> 1262</span>  tmp.s3 = (w03 + 3.f * w13 + 9.f * w23 + 27.f * w33 + 81.f * w43 + 243.f * w53 + 729.f * w63) / 720.f;</div><div class="line"><a name="l01263"></a><span class="lineno"> 1263</span>  tmp.s4 = (w04 + 3.f * w14 + 9.f * w24 + 27.f * w34 + 81.f * w44 + 243.f * w54 + 729.f * w64) / 720.f;</div><div class="line"><a name="l01264"></a><span class="lineno"> 1264</span>  tmp.s5 = (w05 + 3.f * w15 + 9.f * w25 + 27.f * w35 + 81.f * w45 + 243.f * w55 + 729.f * w65) / 720.f;</div><div class="line"><a name="l01265"></a><span class="lineno"> 1265</span>  tmp.s6 = (w06 + 3.f * w16 + 9.f * w26 + 27.f * w36 + 81.f * w46 + 243.f * w56 + 729.f * w66) / 720.f;</div><div class="line"><a name="l01266"></a><span class="lineno"> 1266</span> </div><div class="line"><a name="l01267"></a><span class="lineno"> 1267</span>  <a class="code" href="winograd__input__transform_8cl.xhtml#a4b01af9651d23ef289e82a4db19aac7f">OUTPUT_ROW_2x2_7x7</a>(out6, tmp);</div><div class="line"><a name="l01268"></a><span class="lineno"> 1268</span> </div><div class="line"><a name="l01269"></a><span class="lineno"> 1269</span>  <span class="comment">// Row 7</span></div><div class="line"><a name="l01270"></a><span class="lineno"> 1270</span>  <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)</div><div class="line"><a name="l01271"></a><span class="lineno"> 1271</span>  out7 = 0.0f;</div><div class="line"><a name="l01272"></a><span class="lineno"> 1272</span> </div><div class="line"><a name="l01273"></a><span class="lineno"> 1273</span>  tmp.s0 = w60;</div><div class="line"><a name="l01274"></a><span class="lineno"> 1274</span>  tmp.s1 = w61;</div><div class="line"><a name="l01275"></a><span class="lineno"> 1275</span>  tmp.s2 = w62;</div><div class="line"><a name="l01276"></a><span class="lineno"> 1276</span>  tmp.s3 = w63;</div><div class="line"><a name="l01277"></a><span class="lineno"> 1277</span>  tmp.s4 = w64;</div><div class="line"><a name="l01278"></a><span class="lineno"> 1278</span>  tmp.s5 = w65;</div><div class="line"><a name="l01279"></a><span class="lineno"> 1279</span>  tmp.s6 = w66;</div><div class="line"><a name="l01280"></a><span class="lineno"> 1280</span> </div><div class="line"><a name="l01281"></a><span class="lineno"> 1281</span>  <a class="code" href="winograd__input__transform_8cl.xhtml#a4b01af9651d23ef289e82a4db19aac7f">OUTPUT_ROW_2x2_7x7</a>(out7, tmp);</div><div class="line"><a name="l01282"></a><span class="lineno"> 1282</span> </div><div class="line"><a name="l01283"></a><span class="lineno"> 1283</span> <span class="preprocessor">#endif // !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l01284"></a><span class="lineno"> 1284</span> </div><div class="line"><a name="l01285"></a><span class="lineno"> 1285</span>  <span class="keywordtype">int</span> x0 = get_global_id(2); <span class="comment">// idx filter</span></div><div class="line"><a name="l01286"></a><span class="lineno"> 1286</span>  <span class="keywordtype">int</span> y0 = get_global_id(0); <span class="comment">// idx channel</span></div><div class="line"><a name="l01287"></a><span class="lineno"> 1287</span> </div><div class="line"><a name="l01288"></a><span class="lineno"> 1288</span>  <span class="comment">// Get output address</span></div><div class="line"><a name="l01289"></a><span class="lineno"> 1289</span>  __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x0 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>) + y0 * dst_stride_y;</div><div class="line"><a name="l01290"></a><span class="lineno"> 1290</span> </div><div class="line"><a name="l01291"></a><span class="lineno"> 1291</span>  <span class="comment">// Store the values across the channels</span></div><div class="line"><a name="l01292"></a><span class="lineno"> 1292</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 0 * dst_stride_z) = out0.s0;</div><div class="line"><a name="l01293"></a><span class="lineno"> 1293</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 1 * dst_stride_z) = out0.s1;</div><div class="line"><a name="l01294"></a><span class="lineno"> 1294</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 2 * dst_stride_z) = out0.s2;</div><div class="line"><a name="l01295"></a><span class="lineno"> 1295</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 3 * dst_stride_z) = out0.s3;</div><div class="line"><a name="l01296"></a><span class="lineno"> 1296</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 4 * dst_stride_z) = out0.s4;</div><div class="line"><a name="l01297"></a><span class="lineno"> 1297</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 5 * dst_stride_z) = out0.s5;</div><div class="line"><a name="l01298"></a><span class="lineno"> 1298</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 6 * dst_stride_z) = out0.s6;</div><div class="line"><a name="l01299"></a><span class="lineno"> 1299</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 7 * dst_stride_z) = out0.s7;</div><div class="line"><a name="l01300"></a><span class="lineno"> 1300</span> </div><div class="line"><a name="l01301"></a><span class="lineno"> 1301</span> <span class="preprocessor">#if !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l01302"></a><span class="lineno"> 1302</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 8 * dst_stride_z) = out1.s0;</div><div class="line"><a name="l01303"></a><span class="lineno"> 1303</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 9 * dst_stride_z) = out1.s1;</div><div class="line"><a name="l01304"></a><span class="lineno"> 1304</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 10 * dst_stride_z) = out1.s2;</div><div class="line"><a name="l01305"></a><span class="lineno"> 1305</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 11 * dst_stride_z) = out1.s3;</div><div class="line"><a name="l01306"></a><span class="lineno"> 1306</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 12 * dst_stride_z) = out1.s4;</div><div class="line"><a name="l01307"></a><span class="lineno"> 1307</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 13 * dst_stride_z) = out1.s5;</div><div class="line"><a name="l01308"></a><span class="lineno"> 1308</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 14 * dst_stride_z) = out1.s6;</div><div class="line"><a name="l01309"></a><span class="lineno"> 1309</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 15 * dst_stride_z) = out1.s7;</div><div class="line"><a name="l01310"></a><span class="lineno"> 1310</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 16 * dst_stride_z) = out2.s0;</div><div class="line"><a name="l01311"></a><span class="lineno"> 1311</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 17 * dst_stride_z) = out2.s1;</div><div class="line"><a name="l01312"></a><span class="lineno"> 1312</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 18 * dst_stride_z) = out2.s2;</div><div class="line"><a name="l01313"></a><span class="lineno"> 1313</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 19 * dst_stride_z) = out2.s3;</div><div class="line"><a name="l01314"></a><span class="lineno"> 1314</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 20 * dst_stride_z) = out2.s4;</div><div class="line"><a name="l01315"></a><span class="lineno"> 1315</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 21 * dst_stride_z) = out2.s5;</div><div class="line"><a name="l01316"></a><span class="lineno"> 1316</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 22 * dst_stride_z) = out2.s6;</div><div class="line"><a name="l01317"></a><span class="lineno"> 1317</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 23 * dst_stride_z) = out2.s7;</div><div class="line"><a name="l01318"></a><span class="lineno"> 1318</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 24 * dst_stride_z) = out3.s0;</div><div class="line"><a name="l01319"></a><span class="lineno"> 1319</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 25 * dst_stride_z) = out3.s1;</div><div class="line"><a name="l01320"></a><span class="lineno"> 1320</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 26 * dst_stride_z) = out3.s2;</div><div class="line"><a name="l01321"></a><span class="lineno"> 1321</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 27 * dst_stride_z) = out3.s3;</div><div class="line"><a name="l01322"></a><span class="lineno"> 1322</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 28 * dst_stride_z) = out3.s4;</div><div class="line"><a name="l01323"></a><span class="lineno"> 1323</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 29 * dst_stride_z) = out3.s5;</div><div class="line"><a name="l01324"></a><span class="lineno"> 1324</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 30 * dst_stride_z) = out3.s6;</div><div class="line"><a name="l01325"></a><span class="lineno"> 1325</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 31 * dst_stride_z) = out3.s7;</div><div class="line"><a name="l01326"></a><span class="lineno"> 1326</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 32 * dst_stride_z) = out4.s0;</div><div class="line"><a name="l01327"></a><span class="lineno"> 1327</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 33 * dst_stride_z) = out4.s1;</div><div class="line"><a name="l01328"></a><span class="lineno"> 1328</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 34 * dst_stride_z) = out4.s2;</div><div class="line"><a name="l01329"></a><span class="lineno"> 1329</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 35 * dst_stride_z) = out4.s3;</div><div class="line"><a name="l01330"></a><span class="lineno"> 1330</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 36 * dst_stride_z) = out4.s4;</div><div class="line"><a name="l01331"></a><span class="lineno"> 1331</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 37 * dst_stride_z) = out4.s5;</div><div class="line"><a name="l01332"></a><span class="lineno"> 1332</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 38 * dst_stride_z) = out4.s6;</div><div class="line"><a name="l01333"></a><span class="lineno"> 1333</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 39 * dst_stride_z) = out4.s7;</div><div class="line"><a name="l01334"></a><span class="lineno"> 1334</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 40 * dst_stride_z) = out5.s0;</div><div class="line"><a name="l01335"></a><span class="lineno"> 1335</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 41 * dst_stride_z) = out5.s1;</div><div class="line"><a name="l01336"></a><span class="lineno"> 1336</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 42 * dst_stride_z) = out5.s2;</div><div class="line"><a name="l01337"></a><span class="lineno"> 1337</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 43 * dst_stride_z) = out5.s3;</div><div class="line"><a name="l01338"></a><span class="lineno"> 1338</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 44 * dst_stride_z) = out5.s4;</div><div class="line"><a name="l01339"></a><span class="lineno"> 1339</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 45 * dst_stride_z) = out5.s5;</div><div class="line"><a name="l01340"></a><span class="lineno"> 1340</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 46 * dst_stride_z) = out5.s6;</div><div class="line"><a name="l01341"></a><span class="lineno"> 1341</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 47 * dst_stride_z) = out5.s7;</div><div class="line"><a name="l01342"></a><span class="lineno"> 1342</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 48 * dst_stride_z) = out6.s0;</div><div class="line"><a name="l01343"></a><span class="lineno"> 1343</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 49 * dst_stride_z) = out6.s1;</div><div class="line"><a name="l01344"></a><span class="lineno"> 1344</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 50 * dst_stride_z) = out6.s2;</div><div class="line"><a name="l01345"></a><span class="lineno"> 1345</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 51 * dst_stride_z) = out6.s3;</div><div class="line"><a name="l01346"></a><span class="lineno"> 1346</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 52 * dst_stride_z) = out6.s4;</div><div class="line"><a name="l01347"></a><span class="lineno"> 1347</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 53 * dst_stride_z) = out6.s5;</div><div class="line"><a name="l01348"></a><span class="lineno"> 1348</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 54 * dst_stride_z) = out6.s6;</div><div class="line"><a name="l01349"></a><span class="lineno"> 1349</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 55 * dst_stride_z) = out6.s7;</div><div class="line"><a name="l01350"></a><span class="lineno"> 1350</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 56 * dst_stride_z) = out7.s0;</div><div class="line"><a name="l01351"></a><span class="lineno"> 1351</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 57 * dst_stride_z) = out7.s1;</div><div class="line"><a name="l01352"></a><span class="lineno"> 1352</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 58 * dst_stride_z) = out7.s2;</div><div class="line"><a name="l01353"></a><span class="lineno"> 1353</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 59 * dst_stride_z) = out7.s3;</div><div class="line"><a name="l01354"></a><span class="lineno"> 1354</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 60 * dst_stride_z) = out7.s4;</div><div class="line"><a name="l01355"></a><span class="lineno"> 1355</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 61 * dst_stride_z) = out7.s5;</div><div class="line"><a name="l01356"></a><span class="lineno"> 1356</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 62 * dst_stride_z) = out7.s6;</div><div class="line"><a name="l01357"></a><span class="lineno"> 1357</span>  *(__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 63 * dst_stride_z) = out7.s7;</div><div class="line"><a name="l01358"></a><span class="lineno"> 1358</span> <span class="preprocessor">#endif // !defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL) && !defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l01359"></a><span class="lineno"> 1359</span> }</div><div class="line"><a name="l01360"></a><span class="lineno"> 1360</span> <span class="preprocessor">#endif // defined(SRC_DIM_Z)</span></div><div class="line"><a name="l01361"></a><span class="lineno"> 1361</span> </div><div class="line"><a name="l01362"></a><span class="lineno"> 1362</span> <span class="preprocessor">#if defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL)</span></div><div class="line"><a name="l01363"></a><span class="lineno"> 1363</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 3x1 when the data layout is NCHW and the output tile is 2x1</span></div><div class="line"><a name="l01364"></a><span class="lineno"> 1364</span> <span class="comment"> *</span></div><div class="line"><a name="l01365"></a><span class="lineno"> 1365</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l01366"></a><span class="lineno"> 1366</span> <span class="comment"> * @note -DWINOGRAD_FILTER_TRANSFORM_HORIZONTAL has to be passed at compile time to perform Winograd Filter Transform</span></div><div class="line"><a name="l01367"></a><span class="lineno"> 1367</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l01368"></a><span class="lineno"> 1368</span> <span class="comment"> *</span></div><div class="line"><a name="l01369"></a><span class="lineno"> 1369</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l01370"></a><span class="lineno"> 1370</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01371"></a><span class="lineno"> 1371</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01372"></a><span class="lineno"> 1372</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01373"></a><span class="lineno"> 1373</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01374"></a><span class="lineno"> 1374</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01375"></a><span class="lineno"> 1375</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01376"></a><span class="lineno"> 1376</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l01377"></a><span class="lineno"> 1377</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l01378"></a><span class="lineno"> 1378</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01379"></a><span class="lineno"> 1379</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l01380"></a><span class="lineno"> 1380</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01381"></a><span class="lineno"> 1381</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01382"></a><span class="lineno"> 1382</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01383"></a><span class="lineno"> 1383</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01384"></a><span class="lineno"> 1384</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01385"></a><span class="lineno"> 1385</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01386"></a><span class="lineno"> 1386</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l01387"></a><span class="lineno"> 1387</span> <span class="comment"> */</span></div><div class="line"><a name="l01388"></a><span class="lineno"> 1388</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_2x1_3x1_nchw(</div><div class="line"><a name="l01389"></a><span class="lineno"> 1389</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01390"></a><span class="lineno"> 1390</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01391"></a><span class="lineno"> 1391</span> {</div><div class="line"><a name="l01392"></a><span class="lineno"> 1392</span>  winograd_filter_transform_2x2_3x3_nchw(src_ptr,</div><div class="line"><a name="l01393"></a><span class="lineno"> 1393</span>  src_stride_x,</div><div class="line"><a name="l01394"></a><span class="lineno"> 1394</span>  src_step_x,</div><div class="line"><a name="l01395"></a><span class="lineno"> 1395</span>  src_stride_y,</div><div class="line"><a name="l01396"></a><span class="lineno"> 1396</span>  src_step_y,</div><div class="line"><a name="l01397"></a><span class="lineno"> 1397</span>  src_stride_z,</div><div class="line"><a name="l01398"></a><span class="lineno"> 1398</span>  src_step_z,</div><div class="line"><a name="l01399"></a><span class="lineno"> 1399</span>  src_stride_w,</div><div class="line"><a name="l01400"></a><span class="lineno"> 1400</span>  src_step_w,</div><div class="line"><a name="l01401"></a><span class="lineno"> 1401</span>  src_offset_first_element_in_bytes,</div><div class="line"><a name="l01402"></a><span class="lineno"> 1402</span>  dst_ptr,</div><div class="line"><a name="l01403"></a><span class="lineno"> 1403</span>  dst_stride_x,</div><div class="line"><a name="l01404"></a><span class="lineno"> 1404</span>  dst_step_x,</div><div class="line"><a name="l01405"></a><span class="lineno"> 1405</span>  dst_stride_y,</div><div class="line"><a name="l01406"></a><span class="lineno"> 1406</span>  dst_step_y,</div><div class="line"><a name="l01407"></a><span class="lineno"> 1407</span>  dst_stride_z,</div><div class="line"><a name="l01408"></a><span class="lineno"> 1408</span>  dst_step_z,</div><div class="line"><a name="l01409"></a><span class="lineno"> 1409</span>  dst_offset_first_element_in_bytes);</div><div class="line"><a name="l01410"></a><span class="lineno"> 1410</span> }</div><div class="line"><a name="l01411"></a><span class="lineno"> 1411</span> <span class="comment"></span></div><div class="line"><a name="l01412"></a><span class="lineno"> 1412</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 3x1 when the data layout is NCHW and the output tile is 4x1</span></div><div class="line"><a name="l01413"></a><span class="lineno"> 1413</span> <span class="comment"> *</span></div><div class="line"><a name="l01414"></a><span class="lineno"> 1414</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l01415"></a><span class="lineno"> 1415</span> <span class="comment"> * @note -DWINOGRAD_FILTER_TRANSFORM_HORIZONTAL has to be passed at compile time to perform Winograd Filter Transform</span></div><div class="line"><a name="l01416"></a><span class="lineno"> 1416</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l01417"></a><span class="lineno"> 1417</span> <span class="comment"> *</span></div><div class="line"><a name="l01418"></a><span class="lineno"> 1418</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l01419"></a><span class="lineno"> 1419</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01420"></a><span class="lineno"> 1420</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01421"></a><span class="lineno"> 1421</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01422"></a><span class="lineno"> 1422</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01423"></a><span class="lineno"> 1423</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01424"></a><span class="lineno"> 1424</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01425"></a><span class="lineno"> 1425</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l01426"></a><span class="lineno"> 1426</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l01427"></a><span class="lineno"> 1427</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01428"></a><span class="lineno"> 1428</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l01429"></a><span class="lineno"> 1429</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01430"></a><span class="lineno"> 1430</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01431"></a><span class="lineno"> 1431</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01432"></a><span class="lineno"> 1432</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01433"></a><span class="lineno"> 1433</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01434"></a><span class="lineno"> 1434</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01435"></a><span class="lineno"> 1435</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l01436"></a><span class="lineno"> 1436</span> <span class="comment"> */</span></div><div class="line"><a name="l01437"></a><span class="lineno"> 1437</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_4x1_3x1_nchw(</div><div class="line"><a name="l01438"></a><span class="lineno"> 1438</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01439"></a><span class="lineno"> 1439</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01440"></a><span class="lineno"> 1440</span> {</div><div class="line"><a name="l01441"></a><span class="lineno"> 1441</span>  winograd_filter_transform_4x4_3x3_nchw(src_ptr,</div><div class="line"><a name="l01442"></a><span class="lineno"> 1442</span>  src_stride_x,</div><div class="line"><a name="l01443"></a><span class="lineno"> 1443</span>  src_step_x,</div><div class="line"><a name="l01444"></a><span class="lineno"> 1444</span>  src_stride_y,</div><div class="line"><a name="l01445"></a><span class="lineno"> 1445</span>  src_step_y,</div><div class="line"><a name="l01446"></a><span class="lineno"> 1446</span>  src_stride_z,</div><div class="line"><a name="l01447"></a><span class="lineno"> 1447</span>  src_step_z,</div><div class="line"><a name="l01448"></a><span class="lineno"> 1448</span>  src_stride_w,</div><div class="line"><a name="l01449"></a><span class="lineno"> 1449</span>  src_step_w,</div><div class="line"><a name="l01450"></a><span class="lineno"> 1450</span>  src_offset_first_element_in_bytes,</div><div class="line"><a name="l01451"></a><span class="lineno"> 1451</span>  dst_ptr,</div><div class="line"><a name="l01452"></a><span class="lineno"> 1452</span>  dst_stride_x,</div><div class="line"><a name="l01453"></a><span class="lineno"> 1453</span>  dst_step_x,</div><div class="line"><a name="l01454"></a><span class="lineno"> 1454</span>  dst_stride_y,</div><div class="line"><a name="l01455"></a><span class="lineno"> 1455</span>  dst_step_y,</div><div class="line"><a name="l01456"></a><span class="lineno"> 1456</span>  dst_stride_z,</div><div class="line"><a name="l01457"></a><span class="lineno"> 1457</span>  dst_step_z,</div><div class="line"><a name="l01458"></a><span class="lineno"> 1458</span>  dst_offset_first_element_in_bytes);</div><div class="line"><a name="l01459"></a><span class="lineno"> 1459</span> }</div><div class="line"><a name="l01460"></a><span class="lineno"> 1460</span> <span class="comment"></span></div><div class="line"><a name="l01461"></a><span class="lineno"> 1461</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 5x1 when the data layout is NCHW and the output tile is 4x1</span></div><div class="line"><a name="l01462"></a><span class="lineno"> 1462</span> <span class="comment"> *</span></div><div class="line"><a name="l01463"></a><span class="lineno"> 1463</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l01464"></a><span class="lineno"> 1464</span> <span class="comment"> * @note -DWINOGRAD_FILTER_TRANSFORM_HORIZONTAL has to be passed at compile time to perform Winograd Filter Transform</span></div><div class="line"><a name="l01465"></a><span class="lineno"> 1465</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l01466"></a><span class="lineno"> 1466</span> <span class="comment"> *</span></div><div class="line"><a name="l01467"></a><span class="lineno"> 1467</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l01468"></a><span class="lineno"> 1468</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01469"></a><span class="lineno"> 1469</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01470"></a><span class="lineno"> 1470</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01471"></a><span class="lineno"> 1471</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01472"></a><span class="lineno"> 1472</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01473"></a><span class="lineno"> 1473</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01474"></a><span class="lineno"> 1474</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l01475"></a><span class="lineno"> 1475</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l01476"></a><span class="lineno"> 1476</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01477"></a><span class="lineno"> 1477</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l01478"></a><span class="lineno"> 1478</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01479"></a><span class="lineno"> 1479</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01480"></a><span class="lineno"> 1480</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01481"></a><span class="lineno"> 1481</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01482"></a><span class="lineno"> 1482</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01483"></a><span class="lineno"> 1483</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01484"></a><span class="lineno"> 1484</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l01485"></a><span class="lineno"> 1485</span> <span class="comment"> */</span></div><div class="line"><a name="l01486"></a><span class="lineno"> 1486</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_4x1_5x1_nchw(</div><div class="line"><a name="l01487"></a><span class="lineno"> 1487</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01488"></a><span class="lineno"> 1488</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01489"></a><span class="lineno"> 1489</span> {</div><div class="line"><a name="l01490"></a><span class="lineno"> 1490</span>  winograd_filter_transform_4x4_5x5_nchw(src_ptr,</div><div class="line"><a name="l01491"></a><span class="lineno"> 1491</span>  src_stride_x,</div><div class="line"><a name="l01492"></a><span class="lineno"> 1492</span>  src_step_x,</div><div class="line"><a name="l01493"></a><span class="lineno"> 1493</span>  src_stride_y,</div><div class="line"><a name="l01494"></a><span class="lineno"> 1494</span>  src_step_y,</div><div class="line"><a name="l01495"></a><span class="lineno"> 1495</span>  src_stride_z,</div><div class="line"><a name="l01496"></a><span class="lineno"> 1496</span>  src_step_z,</div><div class="line"><a name="l01497"></a><span class="lineno"> 1497</span>  src_stride_w,</div><div class="line"><a name="l01498"></a><span class="lineno"> 1498</span>  src_step_w,</div><div class="line"><a name="l01499"></a><span class="lineno"> 1499</span>  src_offset_first_element_in_bytes,</div><div class="line"><a name="l01500"></a><span class="lineno"> 1500</span>  dst_ptr,</div><div class="line"><a name="l01501"></a><span class="lineno"> 1501</span>  dst_stride_x,</div><div class="line"><a name="l01502"></a><span class="lineno"> 1502</span>  dst_step_x,</div><div class="line"><a name="l01503"></a><span class="lineno"> 1503</span>  dst_stride_y,</div><div class="line"><a name="l01504"></a><span class="lineno"> 1504</span>  dst_step_y,</div><div class="line"><a name="l01505"></a><span class="lineno"> 1505</span>  dst_stride_z,</div><div class="line"><a name="l01506"></a><span class="lineno"> 1506</span>  dst_step_z,</div><div class="line"><a name="l01507"></a><span class="lineno"> 1507</span>  dst_offset_first_element_in_bytes);</div><div class="line"><a name="l01508"></a><span class="lineno"> 1508</span> }</div><div class="line"><a name="l01509"></a><span class="lineno"> 1509</span> <span class="comment"></span></div><div class="line"><a name="l01510"></a><span class="lineno"> 1510</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 3x1 when the data layout is NHWC and the output tile is 4x1</span></div><div class="line"><a name="l01511"></a><span class="lineno"> 1511</span> <span class="comment"> *</span></div><div class="line"><a name="l01512"></a><span class="lineno"> 1512</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l01513"></a><span class="lineno"> 1513</span> <span class="comment"> * @note -DWINOGRAD_FILTER_TRANSFORM_HORIZONTAL has to be passed at compile time to perform Winograd Filter Transform</span></div><div class="line"><a name="l01514"></a><span class="lineno"> 1514</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l01515"></a><span class="lineno"> 1515</span> <span class="comment"> *</span></div><div class="line"><a name="l01516"></a><span class="lineno"> 1516</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l01517"></a><span class="lineno"> 1517</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01518"></a><span class="lineno"> 1518</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01519"></a><span class="lineno"> 1519</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01520"></a><span class="lineno"> 1520</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01521"></a><span class="lineno"> 1521</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01522"></a><span class="lineno"> 1522</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01523"></a><span class="lineno"> 1523</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l01524"></a><span class="lineno"> 1524</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l01525"></a><span class="lineno"> 1525</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01526"></a><span class="lineno"> 1526</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l01527"></a><span class="lineno"> 1527</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01528"></a><span class="lineno"> 1528</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01529"></a><span class="lineno"> 1529</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01530"></a><span class="lineno"> 1530</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01531"></a><span class="lineno"> 1531</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01532"></a><span class="lineno"> 1532</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01533"></a><span class="lineno"> 1533</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l01534"></a><span class="lineno"> 1534</span> <span class="comment"> */</span></div><div class="line"><a name="l01535"></a><span class="lineno"> 1535</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_4x1_3x1_nhwc(</div><div class="line"><a name="l01536"></a><span class="lineno"> 1536</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01537"></a><span class="lineno"> 1537</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01538"></a><span class="lineno"> 1538</span> {</div><div class="line"><a name="l01539"></a><span class="lineno"> 1539</span>  winograd_filter_transform_4x4_3x3_nhwc(src_ptr,</div><div class="line"><a name="l01540"></a><span class="lineno"> 1540</span>  src_stride_x,</div><div class="line"><a name="l01541"></a><span class="lineno"> 1541</span>  src_step_x,</div><div class="line"><a name="l01542"></a><span class="lineno"> 1542</span>  src_stride_y,</div><div class="line"><a name="l01543"></a><span class="lineno"> 1543</span>  src_step_y,</div><div class="line"><a name="l01544"></a><span class="lineno"> 1544</span>  src_stride_z,</div><div class="line"><a name="l01545"></a><span class="lineno"> 1545</span>  src_step_z,</div><div class="line"><a name="l01546"></a><span class="lineno"> 1546</span>  src_stride_w,</div><div class="line"><a name="l01547"></a><span class="lineno"> 1547</span>  src_step_w,</div><div class="line"><a name="l01548"></a><span class="lineno"> 1548</span>  src_offset_first_element_in_bytes,</div><div class="line"><a name="l01549"></a><span class="lineno"> 1549</span>  dst_ptr,</div><div class="line"><a name="l01550"></a><span class="lineno"> 1550</span>  dst_stride_x,</div><div class="line"><a name="l01551"></a><span class="lineno"> 1551</span>  dst_step_x,</div><div class="line"><a name="l01552"></a><span class="lineno"> 1552</span>  dst_stride_y,</div><div class="line"><a name="l01553"></a><span class="lineno"> 1553</span>  dst_step_y,</div><div class="line"><a name="l01554"></a><span class="lineno"> 1554</span>  dst_stride_z,</div><div class="line"><a name="l01555"></a><span class="lineno"> 1555</span>  dst_step_z,</div><div class="line"><a name="l01556"></a><span class="lineno"> 1556</span>  dst_offset_first_element_in_bytes);</div><div class="line"><a name="l01557"></a><span class="lineno"> 1557</span> }</div><div class="line"><a name="l01558"></a><span class="lineno"> 1558</span> <span class="comment"></span></div><div class="line"><a name="l01559"></a><span class="lineno"> 1559</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 5x1 when the data layout is NHWC and the output tile is 4x1</span></div><div class="line"><a name="l01560"></a><span class="lineno"> 1560</span> <span class="comment"> *</span></div><div class="line"><a name="l01561"></a><span class="lineno"> 1561</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l01562"></a><span class="lineno"> 1562</span> <span class="comment"> * @note -DWINOGRAD_FILTER_TRANSFORM_HORIZONTAL has to be passed at compile time to perform Winograd Filter Transform</span></div><div class="line"><a name="l01563"></a><span class="lineno"> 1563</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l01564"></a><span class="lineno"> 1564</span> <span class="comment"> *</span></div><div class="line"><a name="l01565"></a><span class="lineno"> 1565</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l01566"></a><span class="lineno"> 1566</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01567"></a><span class="lineno"> 1567</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01568"></a><span class="lineno"> 1568</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01569"></a><span class="lineno"> 1569</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01570"></a><span class="lineno"> 1570</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01571"></a><span class="lineno"> 1571</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01572"></a><span class="lineno"> 1572</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l01573"></a><span class="lineno"> 1573</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l01574"></a><span class="lineno"> 1574</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01575"></a><span class="lineno"> 1575</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l01576"></a><span class="lineno"> 1576</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01577"></a><span class="lineno"> 1577</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01578"></a><span class="lineno"> 1578</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01579"></a><span class="lineno"> 1579</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01580"></a><span class="lineno"> 1580</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01581"></a><span class="lineno"> 1581</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01582"></a><span class="lineno"> 1582</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l01583"></a><span class="lineno"> 1583</span> <span class="comment"> */</span></div><div class="line"><a name="l01584"></a><span class="lineno"> 1584</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_4x1_5x1_nhwc(</div><div class="line"><a name="l01585"></a><span class="lineno"> 1585</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01586"></a><span class="lineno"> 1586</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01587"></a><span class="lineno"> 1587</span> {</div><div class="line"><a name="l01588"></a><span class="lineno"> 1588</span>  winograd_filter_transform_4x4_5x5_nhwc(src_ptr,</div><div class="line"><a name="l01589"></a><span class="lineno"> 1589</span>  src_stride_x,</div><div class="line"><a name="l01590"></a><span class="lineno"> 1590</span>  src_step_x,</div><div class="line"><a name="l01591"></a><span class="lineno"> 1591</span>  src_stride_y,</div><div class="line"><a name="l01592"></a><span class="lineno"> 1592</span>  src_step_y,</div><div class="line"><a name="l01593"></a><span class="lineno"> 1593</span>  src_stride_z,</div><div class="line"><a name="l01594"></a><span class="lineno"> 1594</span>  src_step_z,</div><div class="line"><a name="l01595"></a><span class="lineno"> 1595</span>  src_stride_w,</div><div class="line"><a name="l01596"></a><span class="lineno"> 1596</span>  src_step_w,</div><div class="line"><a name="l01597"></a><span class="lineno"> 1597</span>  src_offset_first_element_in_bytes,</div><div class="line"><a name="l01598"></a><span class="lineno"> 1598</span>  dst_ptr,</div><div class="line"><a name="l01599"></a><span class="lineno"> 1599</span>  dst_stride_x,</div><div class="line"><a name="l01600"></a><span class="lineno"> 1600</span>  dst_step_x,</div><div class="line"><a name="l01601"></a><span class="lineno"> 1601</span>  dst_stride_y,</div><div class="line"><a name="l01602"></a><span class="lineno"> 1602</span>  dst_step_y,</div><div class="line"><a name="l01603"></a><span class="lineno"> 1603</span>  dst_stride_z,</div><div class="line"><a name="l01604"></a><span class="lineno"> 1604</span>  dst_step_z,</div><div class="line"><a name="l01605"></a><span class="lineno"> 1605</span>  dst_offset_first_element_in_bytes);</div><div class="line"><a name="l01606"></a><span class="lineno"> 1606</span> }</div><div class="line"><a name="l01607"></a><span class="lineno"> 1607</span> <span class="comment"></span></div><div class="line"><a name="l01608"></a><span class="lineno"> 1608</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 7x1 when the data layout is NHWC and the output tile is 2x1</span></div><div class="line"><a name="l01609"></a><span class="lineno"> 1609</span> <span class="comment"> *</span></div><div class="line"><a name="l01610"></a><span class="lineno"> 1610</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l01611"></a><span class="lineno"> 1611</span> <span class="comment"> * @note -DWINOGRAD_FILTER_TRANSFORM_HORIZONTAL has to be passed at compile time to perform Winograd Filter Transform</span></div><div class="line"><a name="l01612"></a><span class="lineno"> 1612</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float.</span></div><div class="line"><a name="l01613"></a><span class="lineno"> 1613</span> <span class="comment"> *</span></div><div class="line"><a name="l01614"></a><span class="lineno"> 1614</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l01615"></a><span class="lineno"> 1615</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01616"></a><span class="lineno"> 1616</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01617"></a><span class="lineno"> 1617</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01618"></a><span class="lineno"> 1618</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01619"></a><span class="lineno"> 1619</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01620"></a><span class="lineno"> 1620</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01621"></a><span class="lineno"> 1621</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l01622"></a><span class="lineno"> 1622</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l01623"></a><span class="lineno"> 1623</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01624"></a><span class="lineno"> 1624</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l01625"></a><span class="lineno"> 1625</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01626"></a><span class="lineno"> 1626</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01627"></a><span class="lineno"> 1627</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01628"></a><span class="lineno"> 1628</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01629"></a><span class="lineno"> 1629</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01630"></a><span class="lineno"> 1630</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01631"></a><span class="lineno"> 1631</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l01632"></a><span class="lineno"> 1632</span> <span class="comment"> */</span></div><div class="line"><a name="l01633"></a><span class="lineno"> 1633</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_2x1_7x1_nhwc(</div><div class="line"><a name="l01634"></a><span class="lineno"> 1634</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01635"></a><span class="lineno"> 1635</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01636"></a><span class="lineno"> 1636</span> {</div><div class="line"><a name="l01637"></a><span class="lineno"> 1637</span>  winograd_filter_transform_2x2_7x7_nhwc(src_ptr,</div><div class="line"><a name="l01638"></a><span class="lineno"> 1638</span>  src_stride_x,</div><div class="line"><a name="l01639"></a><span class="lineno"> 1639</span>  src_step_x,</div><div class="line"><a name="l01640"></a><span class="lineno"> 1640</span>  src_stride_y,</div><div class="line"><a name="l01641"></a><span class="lineno"> 1641</span>  src_step_y,</div><div class="line"><a name="l01642"></a><span class="lineno"> 1642</span>  src_stride_z,</div><div class="line"><a name="l01643"></a><span class="lineno"> 1643</span>  src_step_z,</div><div class="line"><a name="l01644"></a><span class="lineno"> 1644</span>  src_stride_w,</div><div class="line"><a name="l01645"></a><span class="lineno"> 1645</span>  src_step_w,</div><div class="line"><a name="l01646"></a><span class="lineno"> 1646</span>  src_offset_first_element_in_bytes,</div><div class="line"><a name="l01647"></a><span class="lineno"> 1647</span>  dst_ptr,</div><div class="line"><a name="l01648"></a><span class="lineno"> 1648</span>  dst_stride_x,</div><div class="line"><a name="l01649"></a><span class="lineno"> 1649</span>  dst_step_x,</div><div class="line"><a name="l01650"></a><span class="lineno"> 1650</span>  dst_stride_y,</div><div class="line"><a name="l01651"></a><span class="lineno"> 1651</span>  dst_step_y,</div><div class="line"><a name="l01652"></a><span class="lineno"> 1652</span>  dst_stride_z,</div><div class="line"><a name="l01653"></a><span class="lineno"> 1653</span>  dst_step_z,</div><div class="line"><a name="l01654"></a><span class="lineno"> 1654</span>  dst_offset_first_element_in_bytes);</div><div class="line"><a name="l01655"></a><span class="lineno"> 1655</span> }</div><div class="line"><a name="l01656"></a><span class="lineno"> 1656</span> <span class="preprocessor">#endif // defined(WINOGRAD_FILTER_TRANSFORM_HORIZONTAL)</span></div><div class="line"><a name="l01657"></a><span class="lineno"> 1657</span> </div><div class="line"><a name="l01658"></a><span class="lineno"> 1658</span> <span class="preprocessor">#if defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="line"><a name="l01659"></a><span class="lineno"> 1659</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 1x3 when the data layout is NCHW and the output tile is 1x2</span></div><div class="line"><a name="l01660"></a><span class="lineno"> 1660</span> <span class="comment"> *</span></div><div class="line"><a name="l01661"></a><span class="lineno"> 1661</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l01662"></a><span class="lineno"> 1662</span> <span class="comment"> * @note -DWINOGRAD_FILTER_TRANSFORM_VERTICAL has to be passed at compile time to perform Winograd Filter Transform</span></div><div class="line"><a name="l01663"></a><span class="lineno"> 1663</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l01664"></a><span class="lineno"> 1664</span> <span class="comment"> *</span></div><div class="line"><a name="l01665"></a><span class="lineno"> 1665</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l01666"></a><span class="lineno"> 1666</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01667"></a><span class="lineno"> 1667</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01668"></a><span class="lineno"> 1668</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01669"></a><span class="lineno"> 1669</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01670"></a><span class="lineno"> 1670</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01671"></a><span class="lineno"> 1671</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01672"></a><span class="lineno"> 1672</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l01673"></a><span class="lineno"> 1673</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l01674"></a><span class="lineno"> 1674</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01675"></a><span class="lineno"> 1675</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l01676"></a><span class="lineno"> 1676</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01677"></a><span class="lineno"> 1677</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01678"></a><span class="lineno"> 1678</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01679"></a><span class="lineno"> 1679</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01680"></a><span class="lineno"> 1680</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01681"></a><span class="lineno"> 1681</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01682"></a><span class="lineno"> 1682</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l01683"></a><span class="lineno"> 1683</span> <span class="comment"> */</span></div><div class="line"><a name="l01684"></a><span class="lineno"> 1684</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_1x2_1x3_nchw(</div><div class="line"><a name="l01685"></a><span class="lineno"> 1685</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01686"></a><span class="lineno"> 1686</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01687"></a><span class="lineno"> 1687</span> {</div><div class="line"><a name="l01688"></a><span class="lineno"> 1688</span>  winograd_filter_transform_2x2_3x3_nchw(src_ptr,</div><div class="line"><a name="l01689"></a><span class="lineno"> 1689</span>  src_stride_x,</div><div class="line"><a name="l01690"></a><span class="lineno"> 1690</span>  src_step_x,</div><div class="line"><a name="l01691"></a><span class="lineno"> 1691</span>  src_stride_y,</div><div class="line"><a name="l01692"></a><span class="lineno"> 1692</span>  src_step_y,</div><div class="line"><a name="l01693"></a><span class="lineno"> 1693</span>  src_stride_z,</div><div class="line"><a name="l01694"></a><span class="lineno"> 1694</span>  src_step_z,</div><div class="line"><a name="l01695"></a><span class="lineno"> 1695</span>  src_stride_w,</div><div class="line"><a name="l01696"></a><span class="lineno"> 1696</span>  src_step_w,</div><div class="line"><a name="l01697"></a><span class="lineno"> 1697</span>  src_offset_first_element_in_bytes,</div><div class="line"><a name="l01698"></a><span class="lineno"> 1698</span>  dst_ptr,</div><div class="line"><a name="l01699"></a><span class="lineno"> 1699</span>  dst_stride_x,</div><div class="line"><a name="l01700"></a><span class="lineno"> 1700</span>  dst_step_x,</div><div class="line"><a name="l01701"></a><span class="lineno"> 1701</span>  dst_stride_y,</div><div class="line"><a name="l01702"></a><span class="lineno"> 1702</span>  dst_step_y,</div><div class="line"><a name="l01703"></a><span class="lineno"> 1703</span>  dst_stride_z,</div><div class="line"><a name="l01704"></a><span class="lineno"> 1704</span>  dst_step_z,</div><div class="line"><a name="l01705"></a><span class="lineno"> 1705</span>  dst_offset_first_element_in_bytes);</div><div class="line"><a name="l01706"></a><span class="lineno"> 1706</span> }</div><div class="line"><a name="l01707"></a><span class="lineno"> 1707</span> <span class="comment"></span></div><div class="line"><a name="l01708"></a><span class="lineno"> 1708</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 1x3 when the data layout is NCHW and the output tile is 1x4</span></div><div class="line"><a name="l01709"></a><span class="lineno"> 1709</span> <span class="comment"> *</span></div><div class="line"><a name="l01710"></a><span class="lineno"> 1710</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l01711"></a><span class="lineno"> 1711</span> <span class="comment"> * @note -DWINOGRAD_FILTER_TRANSFORM_VERTICAL has to be passed at compile time to perform Winograd Filter Transform</span></div><div class="line"><a name="l01712"></a><span class="lineno"> 1712</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l01713"></a><span class="lineno"> 1713</span> <span class="comment"> *</span></div><div class="line"><a name="l01714"></a><span class="lineno"> 1714</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l01715"></a><span class="lineno"> 1715</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01716"></a><span class="lineno"> 1716</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01717"></a><span class="lineno"> 1717</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01718"></a><span class="lineno"> 1718</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01719"></a><span class="lineno"> 1719</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01720"></a><span class="lineno"> 1720</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01721"></a><span class="lineno"> 1721</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l01722"></a><span class="lineno"> 1722</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l01723"></a><span class="lineno"> 1723</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01724"></a><span class="lineno"> 1724</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l01725"></a><span class="lineno"> 1725</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01726"></a><span class="lineno"> 1726</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01727"></a><span class="lineno"> 1727</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01728"></a><span class="lineno"> 1728</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01729"></a><span class="lineno"> 1729</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01730"></a><span class="lineno"> 1730</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01731"></a><span class="lineno"> 1731</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l01732"></a><span class="lineno"> 1732</span> <span class="comment"> */</span></div><div class="line"><a name="l01733"></a><span class="lineno"> 1733</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_1x4_1x3_nchw(</div><div class="line"><a name="l01734"></a><span class="lineno"> 1734</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01735"></a><span class="lineno"> 1735</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01736"></a><span class="lineno"> 1736</span> {</div><div class="line"><a name="l01737"></a><span class="lineno"> 1737</span>  winograd_filter_transform_4x4_3x3_nchw(src_ptr,</div><div class="line"><a name="l01738"></a><span class="lineno"> 1738</span>  src_stride_x,</div><div class="line"><a name="l01739"></a><span class="lineno"> 1739</span>  src_step_x,</div><div class="line"><a name="l01740"></a><span class="lineno"> 1740</span>  src_stride_y,</div><div class="line"><a name="l01741"></a><span class="lineno"> 1741</span>  src_step_y,</div><div class="line"><a name="l01742"></a><span class="lineno"> 1742</span>  src_stride_z,</div><div class="line"><a name="l01743"></a><span class="lineno"> 1743</span>  src_step_z,</div><div class="line"><a name="l01744"></a><span class="lineno"> 1744</span>  src_stride_w,</div><div class="line"><a name="l01745"></a><span class="lineno"> 1745</span>  src_step_w,</div><div class="line"><a name="l01746"></a><span class="lineno"> 1746</span>  src_offset_first_element_in_bytes,</div><div class="line"><a name="l01747"></a><span class="lineno"> 1747</span>  dst_ptr,</div><div class="line"><a name="l01748"></a><span class="lineno"> 1748</span>  dst_stride_x,</div><div class="line"><a name="l01749"></a><span class="lineno"> 1749</span>  dst_step_x,</div><div class="line"><a name="l01750"></a><span class="lineno"> 1750</span>  dst_stride_y,</div><div class="line"><a name="l01751"></a><span class="lineno"> 1751</span>  dst_step_y,</div><div class="line"><a name="l01752"></a><span class="lineno"> 1752</span>  dst_stride_z,</div><div class="line"><a name="l01753"></a><span class="lineno"> 1753</span>  dst_step_z,</div><div class="line"><a name="l01754"></a><span class="lineno"> 1754</span>  dst_offset_first_element_in_bytes);</div><div class="line"><a name="l01755"></a><span class="lineno"> 1755</span> }</div><div class="line"><a name="l01756"></a><span class="lineno"> 1756</span> <span class="comment"></span></div><div class="line"><a name="l01757"></a><span class="lineno"> 1757</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 1x5 when the data layout is NCHW and the output tile is 1x4</span></div><div class="line"><a name="l01758"></a><span class="lineno"> 1758</span> <span class="comment"> *</span></div><div class="line"><a name="l01759"></a><span class="lineno"> 1759</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l01760"></a><span class="lineno"> 1760</span> <span class="comment"> * @note -DWINOGRAD_FILTER_TRANSFORM_VERTICAL has to be passed at compile time to perform Winograd Filter Transform</span></div><div class="line"><a name="l01761"></a><span class="lineno"> 1761</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l01762"></a><span class="lineno"> 1762</span> <span class="comment"> *</span></div><div class="line"><a name="l01763"></a><span class="lineno"> 1763</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l01764"></a><span class="lineno"> 1764</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01765"></a><span class="lineno"> 1765</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01766"></a><span class="lineno"> 1766</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01767"></a><span class="lineno"> 1767</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01768"></a><span class="lineno"> 1768</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01769"></a><span class="lineno"> 1769</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01770"></a><span class="lineno"> 1770</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l01771"></a><span class="lineno"> 1771</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l01772"></a><span class="lineno"> 1772</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01773"></a><span class="lineno"> 1773</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l01774"></a><span class="lineno"> 1774</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01775"></a><span class="lineno"> 1775</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01776"></a><span class="lineno"> 1776</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01777"></a><span class="lineno"> 1777</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01778"></a><span class="lineno"> 1778</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01779"></a><span class="lineno"> 1779</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01780"></a><span class="lineno"> 1780</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l01781"></a><span class="lineno"> 1781</span> <span class="comment"> */</span></div><div class="line"><a name="l01782"></a><span class="lineno"> 1782</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_1x4_1x5_nchw(</div><div class="line"><a name="l01783"></a><span class="lineno"> 1783</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01784"></a><span class="lineno"> 1784</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01785"></a><span class="lineno"> 1785</span> {</div><div class="line"><a name="l01786"></a><span class="lineno"> 1786</span>  winograd_filter_transform_4x4_5x5_nchw(src_ptr,</div><div class="line"><a name="l01787"></a><span class="lineno"> 1787</span>  src_stride_x,</div><div class="line"><a name="l01788"></a><span class="lineno"> 1788</span>  src_step_x,</div><div class="line"><a name="l01789"></a><span class="lineno"> 1789</span>  src_stride_y,</div><div class="line"><a name="l01790"></a><span class="lineno"> 1790</span>  src_step_y,</div><div class="line"><a name="l01791"></a><span class="lineno"> 1791</span>  src_stride_z,</div><div class="line"><a name="l01792"></a><span class="lineno"> 1792</span>  src_step_z,</div><div class="line"><a name="l01793"></a><span class="lineno"> 1793</span>  src_stride_w,</div><div class="line"><a name="l01794"></a><span class="lineno"> 1794</span>  src_step_w,</div><div class="line"><a name="l01795"></a><span class="lineno"> 1795</span>  src_offset_first_element_in_bytes,</div><div class="line"><a name="l01796"></a><span class="lineno"> 1796</span>  dst_ptr,</div><div class="line"><a name="l01797"></a><span class="lineno"> 1797</span>  dst_stride_x,</div><div class="line"><a name="l01798"></a><span class="lineno"> 1798</span>  dst_step_x,</div><div class="line"><a name="l01799"></a><span class="lineno"> 1799</span>  dst_stride_y,</div><div class="line"><a name="l01800"></a><span class="lineno"> 1800</span>  dst_step_y,</div><div class="line"><a name="l01801"></a><span class="lineno"> 1801</span>  dst_stride_z,</div><div class="line"><a name="l01802"></a><span class="lineno"> 1802</span>  dst_step_z,</div><div class="line"><a name="l01803"></a><span class="lineno"> 1803</span>  dst_offset_first_element_in_bytes);</div><div class="line"><a name="l01804"></a><span class="lineno"> 1804</span> }</div><div class="line"><a name="l01805"></a><span class="lineno"> 1805</span> <span class="comment"></span></div><div class="line"><a name="l01806"></a><span class="lineno"> 1806</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 1x3 when the data layout is NHWC and the output tile is 1x4</span></div><div class="line"><a name="l01807"></a><span class="lineno"> 1807</span> <span class="comment"> *</span></div><div class="line"><a name="l01808"></a><span class="lineno"> 1808</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l01809"></a><span class="lineno"> 1809</span> <span class="comment"> * @note -DWINOGRAD_FILTER_TRANSFORM_VERTICAL has to be passed at compile time to perform Winograd Filter Transform</span></div><div class="line"><a name="l01810"></a><span class="lineno"> 1810</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l01811"></a><span class="lineno"> 1811</span> <span class="comment"> *</span></div><div class="line"><a name="l01812"></a><span class="lineno"> 1812</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l01813"></a><span class="lineno"> 1813</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01814"></a><span class="lineno"> 1814</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01815"></a><span class="lineno"> 1815</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01816"></a><span class="lineno"> 1816</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01817"></a><span class="lineno"> 1817</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01818"></a><span class="lineno"> 1818</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01819"></a><span class="lineno"> 1819</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l01820"></a><span class="lineno"> 1820</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l01821"></a><span class="lineno"> 1821</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01822"></a><span class="lineno"> 1822</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l01823"></a><span class="lineno"> 1823</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01824"></a><span class="lineno"> 1824</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01825"></a><span class="lineno"> 1825</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01826"></a><span class="lineno"> 1826</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01827"></a><span class="lineno"> 1827</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01828"></a><span class="lineno"> 1828</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01829"></a><span class="lineno"> 1829</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l01830"></a><span class="lineno"> 1830</span> <span class="comment"> */</span></div><div class="line"><a name="l01831"></a><span class="lineno"> 1831</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_1x4_1x3_nhwc(</div><div class="line"><a name="l01832"></a><span class="lineno"> 1832</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01833"></a><span class="lineno"> 1833</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01834"></a><span class="lineno"> 1834</span> {</div><div class="line"><a name="l01835"></a><span class="lineno"> 1835</span>  winograd_filter_transform_4x4_3x3_nhwc(src_ptr,</div><div class="line"><a name="l01836"></a><span class="lineno"> 1836</span>  src_stride_x,</div><div class="line"><a name="l01837"></a><span class="lineno"> 1837</span>  src_step_x,</div><div class="line"><a name="l01838"></a><span class="lineno"> 1838</span>  src_stride_y,</div><div class="line"><a name="l01839"></a><span class="lineno"> 1839</span>  src_step_y,</div><div class="line"><a name="l01840"></a><span class="lineno"> 1840</span>  src_stride_z,</div><div class="line"><a name="l01841"></a><span class="lineno"> 1841</span>  src_step_z,</div><div class="line"><a name="l01842"></a><span class="lineno"> 1842</span>  src_stride_w,</div><div class="line"><a name="l01843"></a><span class="lineno"> 1843</span>  src_step_w,</div><div class="line"><a name="l01844"></a><span class="lineno"> 1844</span>  src_offset_first_element_in_bytes,</div><div class="line"><a name="l01845"></a><span class="lineno"> 1845</span>  dst_ptr,</div><div class="line"><a name="l01846"></a><span class="lineno"> 1846</span>  dst_stride_x,</div><div class="line"><a name="l01847"></a><span class="lineno"> 1847</span>  dst_step_x,</div><div class="line"><a name="l01848"></a><span class="lineno"> 1848</span>  dst_stride_y,</div><div class="line"><a name="l01849"></a><span class="lineno"> 1849</span>  dst_step_y,</div><div class="line"><a name="l01850"></a><span class="lineno"> 1850</span>  dst_stride_z,</div><div class="line"><a name="l01851"></a><span class="lineno"> 1851</span>  dst_step_z,</div><div class="line"><a name="l01852"></a><span class="lineno"> 1852</span>  dst_offset_first_element_in_bytes);</div><div class="line"><a name="l01853"></a><span class="lineno"> 1853</span> }</div><div class="line"><a name="l01854"></a><span class="lineno"> 1854</span> <span class="comment"></span></div><div class="line"><a name="l01855"></a><span class="lineno"> 1855</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 1x5 when the data layout is NHWC and the output tile is 1x4</span></div><div class="line"><a name="l01856"></a><span class="lineno"> 1856</span> <span class="comment"> *</span></div><div class="line"><a name="l01857"></a><span class="lineno"> 1857</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l01858"></a><span class="lineno"> 1858</span> <span class="comment"> * @note -DWINOGRAD_FILTER_TRANSFORM_VERTICAL has to be passed at compile time to perform Winograd Filter Transform</span></div><div class="line"><a name="l01859"></a><span class="lineno"> 1859</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float/half.</span></div><div class="line"><a name="l01860"></a><span class="lineno"> 1860</span> <span class="comment"> *</span></div><div class="line"><a name="l01861"></a><span class="lineno"> 1861</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l01862"></a><span class="lineno"> 1862</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01863"></a><span class="lineno"> 1863</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01864"></a><span class="lineno"> 1864</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01865"></a><span class="lineno"> 1865</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01866"></a><span class="lineno"> 1866</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01867"></a><span class="lineno"> 1867</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01868"></a><span class="lineno"> 1868</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l01869"></a><span class="lineno"> 1869</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l01870"></a><span class="lineno"> 1870</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01871"></a><span class="lineno"> 1871</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l01872"></a><span class="lineno"> 1872</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01873"></a><span class="lineno"> 1873</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01874"></a><span class="lineno"> 1874</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01875"></a><span class="lineno"> 1875</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01876"></a><span class="lineno"> 1876</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01877"></a><span class="lineno"> 1877</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01878"></a><span class="lineno"> 1878</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l01879"></a><span class="lineno"> 1879</span> <span class="comment"> */</span></div><div class="line"><a name="l01880"></a><span class="lineno"> 1880</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_1x4_1x5_nhwc(</div><div class="line"><a name="l01881"></a><span class="lineno"> 1881</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01882"></a><span class="lineno"> 1882</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01883"></a><span class="lineno"> 1883</span> {</div><div class="line"><a name="l01884"></a><span class="lineno"> 1884</span>  winograd_filter_transform_4x4_5x5_nhwc(src_ptr,</div><div class="line"><a name="l01885"></a><span class="lineno"> 1885</span>  src_stride_x,</div><div class="line"><a name="l01886"></a><span class="lineno"> 1886</span>  src_step_x,</div><div class="line"><a name="l01887"></a><span class="lineno"> 1887</span>  src_stride_y,</div><div class="line"><a name="l01888"></a><span class="lineno"> 1888</span>  src_step_y,</div><div class="line"><a name="l01889"></a><span class="lineno"> 1889</span>  src_stride_z,</div><div class="line"><a name="l01890"></a><span class="lineno"> 1890</span>  src_step_z,</div><div class="line"><a name="l01891"></a><span class="lineno"> 1891</span>  src_stride_w,</div><div class="line"><a name="l01892"></a><span class="lineno"> 1892</span>  src_step_w,</div><div class="line"><a name="l01893"></a><span class="lineno"> 1893</span>  src_offset_first_element_in_bytes,</div><div class="line"><a name="l01894"></a><span class="lineno"> 1894</span>  dst_ptr,</div><div class="line"><a name="l01895"></a><span class="lineno"> 1895</span>  dst_stride_x,</div><div class="line"><a name="l01896"></a><span class="lineno"> 1896</span>  dst_step_x,</div><div class="line"><a name="l01897"></a><span class="lineno"> 1897</span>  dst_stride_y,</div><div class="line"><a name="l01898"></a><span class="lineno"> 1898</span>  dst_step_y,</div><div class="line"><a name="l01899"></a><span class="lineno"> 1899</span>  dst_stride_z,</div><div class="line"><a name="l01900"></a><span class="lineno"> 1900</span>  dst_step_z,</div><div class="line"><a name="l01901"></a><span class="lineno"> 1901</span>  dst_offset_first_element_in_bytes);</div><div class="line"><a name="l01902"></a><span class="lineno"> 1902</span> }</div><div class="line"><a name="l01903"></a><span class="lineno"> 1903</span> <span class="comment"></span></div><div class="line"><a name="l01904"></a><span class="lineno"> 1904</span> <span class="comment">/** This OpenCL kernel performs Winograd filter transform 1x7 when the data layout is NHWC and the output tile is 1x2</span></div><div class="line"><a name="l01905"></a><span class="lineno"> 1905</span> <span class="comment"> *</span></div><div class="line"><a name="l01906"></a><span class="lineno"> 1906</span> <span class="comment"> * @note In order to correctly split the input tensor in batches, its dimension across the Z axis (channels for NCHW, height for NHWC) must be passed at compile time using -DSRC_DIM_Z: e.g. -DSRC_DIM_Z=64</span></div><div class="line"><a name="l01907"></a><span class="lineno"> 1907</span> <span class="comment"> * @note -DWINOGRAD_FILTER_TRANSFORM_VERTICAL has to be passed at compile time to perform Winograd Filter Transform</span></div><div class="line"><a name="l01908"></a><span class="lineno"> 1908</span> <span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types: float.</span></div><div class="line"><a name="l01909"></a><span class="lineno"> 1909</span> <span class="comment"> *</span></div><div class="line"><a name="l01910"></a><span class="lineno"> 1910</span> <span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32/F16</span></div><div class="line"><a name="l01911"></a><span class="lineno"> 1911</span> <span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01912"></a><span class="lineno"> 1912</span> <span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01913"></a><span class="lineno"> 1913</span> <span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01914"></a><span class="lineno"> 1914</span> <span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01915"></a><span class="lineno"> 1915</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01916"></a><span class="lineno"> 1916</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01917"></a><span class="lineno"> 1917</span> <span class="comment"> * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes)</span></div><div class="line"><a name="l01918"></a><span class="lineno"> 1918</span> <span class="comment"> * @param[in] src_step_w src_stride_w * number of elements along W processed per workitem(in bytes)</span></div><div class="line"><a name="l01919"></a><span class="lineno"> 1919</span> <span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01920"></a><span class="lineno"> 1920</span> <span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l01921"></a><span class="lineno"> 1921</span> <span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01922"></a><span class="lineno"> 1922</span> <span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01923"></a><span class="lineno"> 1923</span> <span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01924"></a><span class="lineno"> 1924</span> <span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01925"></a><span class="lineno"> 1925</span> <span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01926"></a><span class="lineno"> 1926</span> <span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01927"></a><span class="lineno"> 1927</span> <span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l01928"></a><span class="lineno"> 1928</span> <span class="comment"> */</span></div><div class="line"><a name="l01929"></a><span class="lineno"> 1929</span> __kernel <span class="keywordtype">void</span> winograd_filter_transform_1x2_1x7_nhwc(</div><div class="line"><a name="l01930"></a><span class="lineno"> 1930</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01931"></a><span class="lineno"> 1931</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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01932"></a><span class="lineno"> 1932</span> {</div><div class="line"><a name="l01933"></a><span class="lineno"> 1933</span>  winograd_filter_transform_2x2_7x7_nhwc(src_ptr,</div><div class="line"><a name="l01934"></a><span class="lineno"> 1934</span>  src_stride_x,</div><div class="line"><a name="l01935"></a><span class="lineno"> 1935</span>  src_step_x,</div><div class="line"><a name="l01936"></a><span class="lineno"> 1936</span>  src_stride_y,</div><div class="line"><a name="l01937"></a><span class="lineno"> 1937</span>  src_step_y,</div><div class="line"><a name="l01938"></a><span class="lineno"> 1938</span>  src_stride_z,</div><div class="line"><a name="l01939"></a><span class="lineno"> 1939</span>  src_step_z,</div><div class="line"><a name="l01940"></a><span class="lineno"> 1940</span>  src_stride_w,</div><div class="line"><a name="l01941"></a><span class="lineno"> 1941</span>  src_step_w,</div><div class="line"><a name="l01942"></a><span class="lineno"> 1942</span>  src_offset_first_element_in_bytes,</div><div class="line"><a name="l01943"></a><span class="lineno"> 1943</span>  dst_ptr,</div><div class="line"><a name="l01944"></a><span class="lineno"> 1944</span>  dst_stride_x,</div><div class="line"><a name="l01945"></a><span class="lineno"> 1945</span>  dst_step_x,</div><div class="line"><a name="l01946"></a><span class="lineno"> 1946</span>  dst_stride_y,</div><div class="line"><a name="l01947"></a><span class="lineno"> 1947</span>  dst_step_y,</div><div class="line"><a name="l01948"></a><span class="lineno"> 1948</span>  dst_stride_z,</div><div class="line"><a name="l01949"></a><span class="lineno"> 1949</span>  dst_step_z,</div><div class="line"><a name="l01950"></a><span class="lineno"> 1950</span>  dst_offset_first_element_in_bytes);</div><div class="line"><a name="l01951"></a><span class="lineno"> 1951</span> }</div><div class="line"><a name="l01952"></a><span class="lineno"> 1952</span> <span class="preprocessor">#endif // defined(WINOGRAD_FILTER_TRANSFORM_VERTICAL)</span></div><div class="ttc" id="winograd__input__transform_8cl_xhtml_a4b01af9651d23ef289e82a4db19aac7f"><div class="ttname"><a href="winograd__input__transform_8cl.xhtml#a4b01af9651d23ef289e82a4db19aac7f">OUTPUT_ROW_2x2_7x7</a></div><div class="ttdeci">#define OUTPUT_ROW_2x2_7x7(out, tmp, comm_fact)</div><div class="ttdef"><b>Definition:</b> <a href="winograd__input__transform_8cl_source.xhtml#l00046">winograd_input_transform.cl:46</a></div></div> |
