arm_compute v18.05
diff --git a/documentation/depthwise__convolution_8cl_source.xhtml b/documentation/depthwise__convolution_8cl_source.xhtml
index 3e0eeaf..58f3684 100644
--- a/documentation/depthwise__convolution_8cl_source.xhtml
+++ b/documentation/depthwise__convolution_8cl_source.xhtml
@@ -40,7 +40,7 @@
<tr style="height: 56px;">
<td style="padding-left: 0.5em;">
<div id="projectname">Compute Library
-  <span id="projectnumber">18.03</span>
+  <span id="projectnumber">18.05</span>
</div>
</td>
</tr>
@@ -117,31 +117,27 @@
<div class="title">depthwise_convolution.cl</div> </div>
</div><!--header-->
<div class="contents">
-<a href="depthwise__convolution_8cl.xhtml">Go to the documentation of this file.</a><div class="fragment"><div class="line"><a name="l00001"></a><span class="lineno"> 1</span> <span class="comment">/*</span></div><div class="line"><a name="l00002"></a><span class="lineno"> 2</span> <span class="comment"> * Copyright (c) 2017-2018 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> </div><div class="line"><a name="l00025"></a><span class="lineno"> 25</span> <span class="preprocessor">#include "<a class="code" href="helpers_8h.xhtml">helpers.h</a>"</span></div><div class="line"><a name="l00026"></a><span class="lineno"> 26</span> </div><div class="line"><a name="l00027"></a><span class="lineno"> 27</span> <span class="preprocessor">#if defined(CONV_STRIDE_X)</span></div><div class="line"><a name="l00028"></a><span class="lineno"> 28</span> </div><div class="line"><a name="l00029"></a><span class="lineno"> 29</span> <span class="preprocessor">#if CONV_STRIDE_X == 1</span></div><div class="line"><a name="l00030"></a><span class="lineno"> 30</span> <span class="preprocessor">#define convolution1x3 convolution1x3_stride_1</span></div><div class="line"><a name="l00031"></a><span class="lineno"> 31</span> <span class="preprocessor">#elif CONV_STRIDE_X == 2</span></div><div class="line"><a name="l00032"></a><span class="lineno"> 32</span> <span class="preprocessor">#define convolution1x3 convolution1x3_stride_2</span></div><div class="line"><a name="l00033"></a><span class="lineno"> 33</span> <span class="preprocessor">#elif CONV_STRIDE_X == 3</span></div><div class="line"><a name="l00034"></a><span class="lineno"> 34</span> <span class="preprocessor">#define convolution1x3 convolution1x3_stride_3</span></div><div class="line"><a name="l00035"></a><span class="lineno"> 35</span> <span class="preprocessor">#else </span><span class="comment">/* CONV_STRIDE_X */</span><span class="preprocessor"></span></div><div class="line"><a name="l00036"></a><span class="lineno"> 36</span> <span class="preprocessor">#error "Stride not supported"</span></div><div class="line"><a name="l00037"></a><span class="lineno"> 37</span> <span class="preprocessor">#endif </span><span class="comment">/* CONV_STRIDE_X */</span><span class="preprocessor"></span></div><div class="line"><a name="l00038"></a><span class="lineno"> 38</span> </div><div class="line"><a name="l00048"></a><span class="lineno"> 48</span> <span class="keyword">inline</span> float2 convolution1x3_stride_1(__global <span class="keyword">const</span> uchar *left_pixel,</div><div class="line"><a name="l00049"></a><span class="lineno"> 49</span>  <span class="keyword">const</span> <span class="keywordtype">float</span> left_coeff,</div><div class="line"><a name="l00050"></a><span class="lineno"> 50</span>  <span class="keyword">const</span> <span class="keywordtype">float</span> middle_coeff,</div><div class="line"><a name="l00051"></a><span class="lineno"> 51</span>  <span class="keyword">const</span> <span class="keywordtype">float</span> right_coeff)</div><div class="line"><a name="l00052"></a><span class="lineno"> 52</span> {</div><div class="line"><a name="l00053"></a><span class="lineno"> 53</span>  float4 temp = vload4(0, (__global <span class="keywordtype">float</span> *)left_pixel);</div><div class="line"><a name="l00054"></a><span class="lineno"> 54</span> </div><div class="line"><a name="l00055"></a><span class="lineno"> 55</span>  float2 left = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp.s01, float2);</div><div class="line"><a name="l00056"></a><span class="lineno"> 56</span>  float2 middle = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp.s12, float2);</div><div class="line"><a name="l00057"></a><span class="lineno"> 57</span>  float2 right = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp.s23, float2);</div><div class="line"><a name="l00058"></a><span class="lineno"> 58</span> </div><div class="line"><a name="l00059"></a><span class="lineno"> 59</span>  <span class="keywordflow">return</span> left * (float2)left_coeff + middle * (float2)middle_coeff + right * (float2)right_coeff;</div><div class="line"><a name="l00060"></a><span class="lineno"> 60</span> }</div><div class="line"><a name="l00061"></a><span class="lineno"> 61</span> </div><div class="line"><a name="l00071"></a><span class="lineno"> 71</span> inline float2 convolution1x3_stride_2(__global const uchar *left_pixel,</div><div class="line"><a name="l00072"></a><span class="lineno"> 72</span>  const <span class="keywordtype">float</span> left_coeff,</div><div class="line"><a name="l00073"></a><span class="lineno"> 73</span>  const <span class="keywordtype">float</span> middle_coeff,</div><div class="line"><a name="l00074"></a><span class="lineno"> 74</span>  const <span class="keywordtype">float</span> right_coeff)</div><div class="line"><a name="l00075"></a><span class="lineno"> 75</span> {</div><div class="line"><a name="l00076"></a><span class="lineno"> 76</span>  float4 temp0 = vload4(0, (__global <span class="keywordtype">float</span> *)left_pixel);</div><div class="line"><a name="l00077"></a><span class="lineno"> 77</span>  <span class="keywordtype">float</span> temp1 = *((__global <span class="keywordtype">float</span> *)(left_pixel + 4 * <span class="keyword">sizeof</span>(<span class="keywordtype">float</span>)));</div><div class="line"><a name="l00078"></a><span class="lineno"> 78</span> </div><div class="line"><a name="l00079"></a><span class="lineno"> 79</span>  float2 left = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s02, float2);</div><div class="line"><a name="l00080"></a><span class="lineno"> 80</span>  float2 middle = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s13, float2);</div><div class="line"><a name="l00081"></a><span class="lineno"> 81</span>  float2 right = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>((float2)(temp0.s2, temp1), float2);</div><div class="line"><a name="l00082"></a><span class="lineno"> 82</span> </div><div class="line"><a name="l00083"></a><span class="lineno"> 83</span>  <span class="keywordflow">return</span> left * (float2)left_coeff + middle * (float2)middle_coeff + right * (float2)right_coeff;</div><div class="line"><a name="l00084"></a><span class="lineno"> 84</span> }</div><div class="line"><a name="l00085"></a><span class="lineno"> 85</span> </div><div class="line"><a name="l00095"></a><span class="lineno"> 95</span> inline float2 convolution1x3_stride_3(__global const uchar *left_pixel,</div><div class="line"><a name="l00096"></a><span class="lineno"> 96</span>  const <span class="keywordtype">float</span> left_coeff,</div><div class="line"><a name="l00097"></a><span class="lineno"> 97</span>  const <span class="keywordtype">float</span> middle_coeff,</div><div class="line"><a name="l00098"></a><span class="lineno"> 98</span>  const <span class="keywordtype">float</span> right_coeff)</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>  float4 temp0 = vload4(0, (__global <span class="keywordtype">float</span> *)left_pixel);</div><div class="line"><a name="l00101"></a><span class="lineno"> 101</span>  float2 temp1 = vload2(0, (__global <span class="keywordtype">float</span> *)(left_pixel + 4 * <span class="keyword">sizeof</span>(<span class="keywordtype">float</span>)));</div><div class="line"><a name="l00102"></a><span class="lineno"> 102</span> </div><div class="line"><a name="l00103"></a><span class="lineno"> 103</span>  float2 left = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s03, float2);</div><div class="line"><a name="l00104"></a><span class="lineno"> 104</span>  float2 middle = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>((float2)(temp0.s1, temp1.s0), float2);</div><div class="line"><a name="l00105"></a><span class="lineno"> 105</span>  float2 right = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>((float2)(temp0.s2, temp1.s1), float2);</div><div class="line"><a name="l00106"></a><span class="lineno"> 106</span> </div><div class="line"><a name="l00107"></a><span class="lineno"> 107</span>  <span class="keywordflow">return</span> left * (float2)left_coeff + middle * (float2)middle_coeff + right * (float2)right_coeff;</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> </div><div class="line"><a name="l00132"></a><span class="lineno"> 132</span> inline float2 <a class="code" href="convolution3x3_8cl.xhtml#afc5fefe72e66f0ae5191fd5b708fade9">convolution3x3</a>(</div><div class="line"><a name="l00133"></a><span class="lineno"> 133</span>  <a class="code" href="struct_image.xhtml">Image</a> *<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a6743f0a130e8311e6f5b1a23df102472">src</a>,</div><div class="line"><a name="l00134"></a><span class="lineno"> 134</span>  const <span class="keywordtype">float</span> mat0, const <span class="keywordtype">float</span> mat1, const <span class="keywordtype">float</span> mat2,</div><div class="line"><a name="l00135"></a><span class="lineno"> 135</span>  const <span class="keywordtype">float</span> mat3, const <span class="keywordtype">float</span> mat4, const <span class="keywordtype">float</span> mat5,</div><div class="line"><a name="l00136"></a><span class="lineno"> 136</span>  const <span class="keywordtype">float</span> mat6, const <span class="keywordtype">float</span> mat7, const <span class="keywordtype">float</span> mat8)</div><div class="line"><a name="l00137"></a><span class="lineno"> 137</span> {</div><div class="line"><a name="l00138"></a><span class="lineno"> 138</span>  float2 pixels;</div><div class="line"><a name="l00139"></a><span class="lineno"> 139</span> </div><div class="line"><a name="l00140"></a><span class="lineno"> 140</span>  pixels = <a class="code" href="convolution3x3_8cl.xhtml#a92702074338198e81a46c3e309d9b04f">convolution1x3</a>(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(src, 0, 0), mat0, mat1, mat2);</div><div class="line"><a name="l00141"></a><span class="lineno"> 141</span>  pixels += <a class="code" href="convolution3x3_8cl.xhtml#a92702074338198e81a46c3e309d9b04f">convolution1x3</a>(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(src, 0, 1), mat3, mat4, mat5);</div><div class="line"><a name="l00142"></a><span class="lineno"> 142</span>  pixels += <a class="code" href="convolution3x3_8cl.xhtml#a92702074338198e81a46c3e309d9b04f">convolution1x3</a>(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(src, 0, 2), mat6, mat7, mat8);</div><div class="line"><a name="l00143"></a><span class="lineno"> 143</span> </div><div class="line"><a name="l00144"></a><span class="lineno"> 144</span>  <span class="keywordflow">return</span> pixels;</div><div class="line"><a name="l00145"></a><span class="lineno"> 145</span> }</div><div class="line"><a name="l00146"></a><span class="lineno"> 146</span> </div><div class="line"><a name="l00178"></a><span class="lineno"> 178</span> __kernel <span class="keywordtype">void</span> depthwise_convolution_3x3(</div><div class="line"><a name="l00179"></a><span class="lineno"> 179</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00180"></a><span class="lineno"> 180</span>  <a class="code" href="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="l00181"></a><span class="lineno"> 181</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(weights)</div><div class="line"><a name="l00182"></a><span class="lineno"> 182</span> #<span class="keywordflow">if</span> defined(HAS_BIAS)</div><div class="line"><a name="l00183"></a><span class="lineno"> 183</span>  ,</div><div class="line"><a name="l00184"></a><span class="lineno"> 184</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases)</div><div class="line"><a name="l00185"></a><span class="lineno"> 185</span> #endif <span class="comment">//defined(HAS_BIAS)</span></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> {</div><div class="line"><a name="l00188"></a><span class="lineno"> 188</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(src);</div><div class="line"><a name="l00189"></a><span class="lineno"> 189</span>  <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a> = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l00190"></a><span class="lineno"> 190</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> weights = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(weights);</div><div class="line"><a name="l00191"></a><span class="lineno"> 191</span> <span class="preprocessor">#if defined(HAS_BIAS)</span></div><div class="line"><a name="l00192"></a><span class="lineno"> 192</span>  <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00193"></a><span class="lineno"> 193</span> <span class="preprocessor">#endif //defined(HAS_BIAS)</span></div><div class="line"><a name="l00194"></a><span class="lineno"> 194</span> </div><div class="line"><a name="l00195"></a><span class="lineno"> 195</span>  uchar3 <a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a> = (uchar3)(0, 1, 2) * (uchar3)weights_stride_y;</div><div class="line"><a name="l00196"></a><span class="lineno"> 196</span>  float3 weights_values0 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + offset.s0));</div><div class="line"><a name="l00197"></a><span class="lineno"> 197</span>  float3 weights_values1 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + offset.s1));</div><div class="line"><a name="l00198"></a><span class="lineno"> 198</span>  float3 weights_values2 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + offset.s2));</div><div class="line"><a name="l00199"></a><span class="lineno"> 199</span> </div><div class="line"><a name="l00200"></a><span class="lineno"> 200</span>  float2 pixels = <a class="code" href="convolution3x3_8cl.xhtml#afc5fefe72e66f0ae5191fd5b708fade9">convolution3x3</a>(&src, weights_values0.s0, weights_values0.s1, weights_values0.s2,</div><div class="line"><a name="l00201"></a><span class="lineno"> 201</span>  weights_values1.s0, weights_values1.s1, weights_values1.s2,</div><div class="line"><a name="l00202"></a><span class="lineno"> 202</span>  weights_values2.s0, weights_values2.s1, weights_values2.s2);</div><div class="line"><a name="l00203"></a><span class="lineno"> 203</span> <span class="preprocessor">#if defined(HAS_BIAS)</span></div><div class="line"><a name="l00204"></a><span class="lineno"> 204</span>  pixels += (float2)(*((__global <span class="keywordtype">float</span> *)(biases.<a class="code" href="struct_vector.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + get_global_id(2) * biases_stride_x)));</div><div class="line"><a name="l00205"></a><span class="lineno"> 205</span> <span class="preprocessor">#endif //defined(HAS_BIAS)</span></div><div class="line"><a name="l00206"></a><span class="lineno"> 206</span> </div><div class="line"><a name="l00207"></a><span class="lineno"> 207</span>  vstore2(pixels, 0, (__global <span class="keywordtype">float</span> *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</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="preprocessor">#endif //defined(CONV_STRIDE_X)</span></div><div class="line"><a name="l00210"></a><span class="lineno"> 210</span> </div><div class="line"><a name="l00211"></a><span class="lineno"><a class="line" href="depthwise__convolution_8cl.xhtml#aa5ab4cf85fe56de3821727cfcef327d9"> 211</a></span> <span class="preprocessor">#define CONVOLUTION1x3_BIFROST2X1_STRIDE1(acc, src0, weights_row0) \</span></div><div class="line"><a name="l00212"></a><span class="lineno"> 212</span> <span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00213"></a><span class="lineno"> 213</span> <span class="preprocessor"> acc.s0 = fma(src0.s0, weights_row0.s0, acc.s0); \</span></div><div class="line"><a name="l00214"></a><span class="lineno"> 214</span> <span class="preprocessor"> acc.s0 = fma(src0.s1, weights_row0.s1, acc.s0); \</span></div><div class="line"><a name="l00215"></a><span class="lineno"> 215</span> <span class="preprocessor"> acc.s0 = fma(src0.s2, weights_row0.s2, acc.s0); \</span></div><div class="line"><a name="l00216"></a><span class="lineno"> 216</span> <span class="preprocessor"> acc.s1 = fma(src0.s1, weights_row0.s0, acc.s1); \</span></div><div class="line"><a name="l00217"></a><span class="lineno"> 217</span> <span class="preprocessor"> acc.s1 = fma(src0.s2, weights_row0.s1, acc.s1); \</span></div><div class="line"><a name="l00218"></a><span class="lineno"> 218</span> <span class="preprocessor"> acc.s1 = fma(src0.s3, weights_row0.s2, acc.s1); \</span></div><div class="line"><a name="l00219"></a><span class="lineno"> 219</span> <span class="preprocessor"> })</span></div><div class="line"><a name="l00220"></a><span class="lineno"> 220</span> </div><div class="line"><a name="l00221"></a><span class="lineno"><a class="line" href="depthwise__convolution_8cl.xhtml#a31cbdd6b28ee6331839432f69d87fdcb"> 221</a></span> <span class="preprocessor">#define CONVOLUTION1x3_BIFROST2X1_STRIDE2(acc, src0, src1, weights_row0) \</span></div><div class="line"><a name="l00222"></a><span class="lineno"> 222</span> <span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00223"></a><span class="lineno"> 223</span> <span class="preprocessor"> acc.s0 = fma(src0.s0, weights_row0.s0, acc.s0); \</span></div><div class="line"><a name="l00224"></a><span class="lineno"> 224</span> <span class="preprocessor"> acc.s0 = fma(src0.s1, weights_row0.s1, acc.s0); \</span></div><div class="line"><a name="l00225"></a><span class="lineno"> 225</span> <span class="preprocessor"> acc.s0 = fma(src0.s2, weights_row0.s2, acc.s0); \</span></div><div class="line"><a name="l00226"></a><span class="lineno"> 226</span> <span class="preprocessor"> acc.s1 = fma(src0.s2, weights_row0.s0, acc.s1); \</span></div><div class="line"><a name="l00227"></a><span class="lineno"> 227</span> <span class="preprocessor"> acc.s1 = fma(src0.s3, weights_row0.s1, acc.s1); \</span></div><div class="line"><a name="l00228"></a><span class="lineno"> 228</span> <span class="preprocessor"> acc.s1 = fma(src1.s0, weights_row0.s2, acc.s1); \</span></div><div class="line"><a name="l00229"></a><span class="lineno"> 229</span> <span class="preprocessor"> })</span></div><div class="line"><a name="l00230"></a><span class="lineno"> 230</span> </div><div class="line"><a name="l00263"></a><span class="lineno"><a class="line" href="depthwise__convolution_8cl.xhtml#a2fbbf5b73260b66b310d4d857db5289e"> 263</a></span> __kernel <span class="keywordtype">void</span> <a class="code" href="depthwise__convolution_8cl.xhtml#a2fbbf5b73260b66b310d4d857db5289e">depthwise_convolution_3x3_stridex1_stridey1_bifrost</a>(</div><div class="line"><a name="l00264"></a><span class="lineno"> 264</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00265"></a><span class="lineno"> 265</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(dst),</div><div class="line"><a name="l00266"></a><span class="lineno"> 266</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(weights)</div><div class="line"><a name="l00267"></a><span class="lineno"> 267</span> #<span class="keywordflow">if</span> defined(HAS_BIAS)</div><div class="line"><a name="l00268"></a><span class="lineno"> 268</span>  ,</div><div class="line"><a name="l00269"></a><span class="lineno"> 269</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases)</div><div class="line"><a name="l00270"></a><span class="lineno"> 270</span> #endif <span class="comment">//defined(HAS_BIAS)</span></div><div class="line"><a name="l00271"></a><span class="lineno"> 271</span> )</div><div class="line"><a name="l00272"></a><span class="lineno"> 272</span> {</div><div class="line"><a name="l00273"></a><span class="lineno"> 273</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(src);</div><div class="line"><a name="l00274"></a><span class="lineno"> 274</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l00275"></a><span class="lineno"> 275</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> weights = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(weights);</div><div class="line"><a name="l00276"></a><span class="lineno"> 276</span> </div><div class="line"><a name="l00277"></a><span class="lineno"> 277</span>  float2 pixels0 = 0.0f;</div><div class="line"><a name="l00278"></a><span class="lineno"> 278</span>  float2 pixels1 = 0.0f;</div><div class="line"><a name="l00279"></a><span class="lineno"> 279</span>  float2 pixels2 = 0.0f;</div><div class="line"><a name="l00280"></a><span class="lineno"> 280</span>  float2 pixels3 = 0.0f;</div><div class="line"><a name="l00281"></a><span class="lineno"> 281</span> </div><div class="line"><a name="l00282"></a><span class="lineno"> 282</span>  __global uchar *weights_addr = (__global uchar *)weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>;</div><div class="line"><a name="l00283"></a><span class="lineno"> 283</span>  __global uchar *src_addr = (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 0);</div><div class="line"><a name="l00284"></a><span class="lineno"> 284</span> </div><div class="line"><a name="l00285"></a><span class="lineno"> 285</span>  <span class="comment">// Load the weights</span></div><div class="line"><a name="l00286"></a><span class="lineno"> 286</span>  float3 weights_row0 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights_addr + 0 * weights_stride_y));</div><div class="line"><a name="l00287"></a><span class="lineno"> 287</span>  float3 weights_row1 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights_addr + 1 * weights_stride_y));</div><div class="line"><a name="l00288"></a><span class="lineno"> 288</span>  float3 weights_row2 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights_addr + 2 * weights_stride_y));</div><div class="line"><a name="l00289"></a><span class="lineno"> 289</span> </div><div class="line"><a name="l00290"></a><span class="lineno"> 290</span>  <span class="comment">// Note: Since each work-item computes 4x2 elements, we need to load 4 rows from the input tensor</span></div><div class="line"><a name="l00291"></a><span class="lineno"> 291</span>  float4 src00 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 0 * src_stride_y)); <span class="comment">// Row0</span></div><div class="line"><a name="l00292"></a><span class="lineno"> 292</span>  float4 src10 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 1 * src_stride_y)); <span class="comment">// Row1</span></div><div class="line"><a name="l00293"></a><span class="lineno"> 293</span>  float4 src20 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 2 * src_stride_y)); <span class="comment">// Row2</span></div><div class="line"><a name="l00294"></a><span class="lineno"> 294</span>  float4 src30 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 3 * src_stride_y)); <span class="comment">// Row3</span></div><div class="line"><a name="l00295"></a><span class="lineno"> 295</span>  float4 src40 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 4 * src_stride_y)); <span class="comment">// Row3</span></div><div class="line"><a name="l00296"></a><span class="lineno"> 296</span>  float4 src50 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 5 * src_stride_y)); <span class="comment">// Row3</span></div><div class="line"><a name="l00297"></a><span class="lineno"> 297</span> </div><div class="line"><a name="l00298"></a><span class="lineno"> 298</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#aa5ab4cf85fe56de3821727cfcef327d9">CONVOLUTION1x3_BIFROST2X1_STRIDE1</a>(pixels0, src00, weights_row0);</div><div class="line"><a name="l00299"></a><span class="lineno"> 299</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#aa5ab4cf85fe56de3821727cfcef327d9">CONVOLUTION1x3_BIFROST2X1_STRIDE1</a>(pixels0, src10, weights_row1);</div><div class="line"><a name="l00300"></a><span class="lineno"> 300</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#aa5ab4cf85fe56de3821727cfcef327d9">CONVOLUTION1x3_BIFROST2X1_STRIDE1</a>(pixels0, src20, weights_row2);</div><div class="line"><a name="l00301"></a><span class="lineno"> 301</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#aa5ab4cf85fe56de3821727cfcef327d9">CONVOLUTION1x3_BIFROST2X1_STRIDE1</a>(pixels1, src10, weights_row0);</div><div class="line"><a name="l00302"></a><span class="lineno"> 302</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#aa5ab4cf85fe56de3821727cfcef327d9">CONVOLUTION1x3_BIFROST2X1_STRIDE1</a>(pixels1, src20, weights_row1);</div><div class="line"><a name="l00303"></a><span class="lineno"> 303</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#aa5ab4cf85fe56de3821727cfcef327d9">CONVOLUTION1x3_BIFROST2X1_STRIDE1</a>(pixels1, src30, weights_row2);</div><div class="line"><a name="l00304"></a><span class="lineno"> 304</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#aa5ab4cf85fe56de3821727cfcef327d9">CONVOLUTION1x3_BIFROST2X1_STRIDE1</a>(pixels2, src20, weights_row0);</div><div class="line"><a name="l00305"></a><span class="lineno"> 305</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#aa5ab4cf85fe56de3821727cfcef327d9">CONVOLUTION1x3_BIFROST2X1_STRIDE1</a>(pixels2, src30, weights_row1);</div><div class="line"><a name="l00306"></a><span class="lineno"> 306</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#aa5ab4cf85fe56de3821727cfcef327d9">CONVOLUTION1x3_BIFROST2X1_STRIDE1</a>(pixels2, src40, weights_row2);</div><div class="line"><a name="l00307"></a><span class="lineno"> 307</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#aa5ab4cf85fe56de3821727cfcef327d9">CONVOLUTION1x3_BIFROST2X1_STRIDE1</a>(pixels3, src30, weights_row0);</div><div class="line"><a name="l00308"></a><span class="lineno"> 308</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#aa5ab4cf85fe56de3821727cfcef327d9">CONVOLUTION1x3_BIFROST2X1_STRIDE1</a>(pixels3, src40, weights_row1);</div><div class="line"><a name="l00309"></a><span class="lineno"> 309</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#aa5ab4cf85fe56de3821727cfcef327d9">CONVOLUTION1x3_BIFROST2X1_STRIDE1</a>(pixels3, src50, weights_row2);</div><div class="line"><a name="l00310"></a><span class="lineno"> 310</span> </div><div class="line"><a name="l00311"></a><span class="lineno"> 311</span> <span class="preprocessor">#ifdef HAS_BIAS</span></div><div class="line"><a name="l00312"></a><span class="lineno"> 312</span>  <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00313"></a><span class="lineno"> 313</span> </div><div class="line"><a name="l00314"></a><span class="lineno"> 314</span>  <span class="keywordtype">float</span> bias = *((__global <span class="keywordtype">float</span> *)(<a class="code" href="helpers_8h.xhtml#a7e4940407322d6f0ccb8b6b86b856019">vector_offset</a>(&biases, get_global_id(2))));</div><div class="line"><a name="l00315"></a><span class="lineno"> 315</span> </div><div class="line"><a name="l00316"></a><span class="lineno"> 316</span>  pixels0 += (float2)bias;</div><div class="line"><a name="l00317"></a><span class="lineno"> 317</span>  pixels1 += (float2)bias;</div><div class="line"><a name="l00318"></a><span class="lineno"> 318</span>  pixels2 += (float2)bias;</div><div class="line"><a name="l00319"></a><span class="lineno"> 319</span>  pixels3 += (float2)bias;</div><div class="line"><a name="l00320"></a><span class="lineno"> 320</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(HAS_BIAS) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00321"></a><span class="lineno"> 321</span> </div><div class="line"><a name="l00322"></a><span class="lineno"> 322</span>  vstore2(pixels0, 0, (__global <span class="keywordtype">float</span> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 0 * dst_stride_y));</div><div class="line"><a name="l00323"></a><span class="lineno"> 323</span>  vstore2(pixels1, 0, (__global <span class="keywordtype">float</span> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 1 * dst_stride_y));</div><div class="line"><a name="l00324"></a><span class="lineno"> 324</span>  vstore2(pixels2, 0, (__global <span class="keywordtype">float</span> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 2 * dst_stride_y));</div><div class="line"><a name="l00325"></a><span class="lineno"> 325</span>  vstore2(pixels3, 0, (__global <span class="keywordtype">float</span> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 3 * dst_stride_y));</div><div class="line"><a name="l00326"></a><span class="lineno"> 326</span> }</div><div class="line"><a name="l00327"></a><span class="lineno"> 327</span> </div><div class="line"><a name="l00360"></a><span class="lineno"><a class="line" href="depthwise__convolution_8cl.xhtml#a328ab548c536089e8238fb33de617bec"> 360</a></span> __kernel <span class="keywordtype">void</span> <a class="code" href="depthwise__convolution_8cl.xhtml#a328ab548c536089e8238fb33de617bec">depthwise_convolution_3x3_stridex2_stridey2_bifrost</a>(</div><div class="line"><a name="l00361"></a><span class="lineno"> 361</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00362"></a><span class="lineno"> 362</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(dst),</div><div class="line"><a name="l00363"></a><span class="lineno"> 363</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(weights)</div><div class="line"><a name="l00364"></a><span class="lineno"> 364</span> #<span class="keywordflow">if</span> defined(HAS_BIAS)</div><div class="line"><a name="l00365"></a><span class="lineno"> 365</span>  ,</div><div class="line"><a name="l00366"></a><span class="lineno"> 366</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases)</div><div class="line"><a name="l00367"></a><span class="lineno"> 367</span> #endif <span class="comment">//defined(HAS_BIAS)</span></div><div class="line"><a name="l00368"></a><span class="lineno"> 368</span> )</div><div class="line"><a name="l00369"></a><span class="lineno"> 369</span> {</div><div class="line"><a name="l00370"></a><span class="lineno"> 370</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(src);</div><div class="line"><a name="l00371"></a><span class="lineno"> 371</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l00372"></a><span class="lineno"> 372</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> weights = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(weights);</div><div class="line"><a name="l00373"></a><span class="lineno"> 373</span> </div><div class="line"><a name="l00374"></a><span class="lineno"> 374</span>  float2 pixels0 = 0.0f;</div><div class="line"><a name="l00375"></a><span class="lineno"> 375</span>  float2 pixels1 = 0.0f;</div><div class="line"><a name="l00376"></a><span class="lineno"> 376</span> </div><div class="line"><a name="l00377"></a><span class="lineno"> 377</span>  __global uchar *weights_addr = (__global uchar *)weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>;</div><div class="line"><a name="l00378"></a><span class="lineno"> 378</span>  __global uchar *src_addr = (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 0);</div><div class="line"><a name="l00379"></a><span class="lineno"> 379</span> </div><div class="line"><a name="l00380"></a><span class="lineno"> 380</span>  <span class="comment">// Load the weights</span></div><div class="line"><a name="l00381"></a><span class="lineno"> 381</span>  float3 weights_row0 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights_addr + 0 * weights_stride_y));</div><div class="line"><a name="l00382"></a><span class="lineno"> 382</span>  float3 weights_row1 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights_addr + 1 * weights_stride_y));</div><div class="line"><a name="l00383"></a><span class="lineno"> 383</span>  float3 weights_row2 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights_addr + 2 * weights_stride_y));</div><div class="line"><a name="l00384"></a><span class="lineno"> 384</span> </div><div class="line"><a name="l00385"></a><span class="lineno"> 385</span>  <span class="comment">// Note: Since each work-item computes 4x2 elements, we need to load 5 rows from the input tensor</span></div><div class="line"><a name="l00386"></a><span class="lineno"> 386</span>  float4 src00 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 0 * src_stride_y)); <span class="comment">// Row0</span></div><div class="line"><a name="l00387"></a><span class="lineno"> 387</span>  float2 src01 = vload2(2, (__global <span class="keywordtype">float</span> *)(src_addr + 0 * src_stride_y)); <span class="comment">// Row0</span></div><div class="line"><a name="l00388"></a><span class="lineno"> 388</span>  float4 src10 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 1 * src_stride_y)); <span class="comment">// Row1</span></div><div class="line"><a name="l00389"></a><span class="lineno"> 389</span>  float2 src11 = vload2(2, (__global <span class="keywordtype">float</span> *)(src_addr + 1 * src_stride_y)); <span class="comment">// Row1</span></div><div class="line"><a name="l00390"></a><span class="lineno"> 390</span>  float4 src20 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 2 * src_stride_y)); <span class="comment">// Row2</span></div><div class="line"><a name="l00391"></a><span class="lineno"> 391</span>  float2 src21 = vload2(2, (__global <span class="keywordtype">float</span> *)(src_addr + 2 * src_stride_y)); <span class="comment">// Row2</span></div><div class="line"><a name="l00392"></a><span class="lineno"> 392</span>  float4 src30 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 3 * src_stride_y)); <span class="comment">// Row3</span></div><div class="line"><a name="l00393"></a><span class="lineno"> 393</span>  float2 src31 = vload2(2, (__global <span class="keywordtype">float</span> *)(src_addr + 3 * src_stride_y)); <span class="comment">// Row3</span></div><div class="line"><a name="l00394"></a><span class="lineno"> 394</span>  float4 src40 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 4 * src_stride_y)); <span class="comment">// Row4</span></div><div class="line"><a name="l00395"></a><span class="lineno"> 395</span>  float2 src41 = vload2(2, (__global <span class="keywordtype">float</span> *)(src_addr + 4 * src_stride_y)); <span class="comment">// Row4</span></div><div class="line"><a name="l00396"></a><span class="lineno"> 396</span> </div><div class="line"><a name="l00397"></a><span class="lineno"> 397</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#a31cbdd6b28ee6331839432f69d87fdcb">CONVOLUTION1x3_BIFROST2X1_STRIDE2</a>(pixels0, src00, src01, weights_row0);</div><div class="line"><a name="l00398"></a><span class="lineno"> 398</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#a31cbdd6b28ee6331839432f69d87fdcb">CONVOLUTION1x3_BIFROST2X1_STRIDE2</a>(pixels0, src10, src11, weights_row1);</div><div class="line"><a name="l00399"></a><span class="lineno"> 399</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#a31cbdd6b28ee6331839432f69d87fdcb">CONVOLUTION1x3_BIFROST2X1_STRIDE2</a>(pixels0, src20, src21, weights_row2);</div><div class="line"><a name="l00400"></a><span class="lineno"> 400</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#a31cbdd6b28ee6331839432f69d87fdcb">CONVOLUTION1x3_BIFROST2X1_STRIDE2</a>(pixels1, src20, src21, weights_row0);</div><div class="line"><a name="l00401"></a><span class="lineno"> 401</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#a31cbdd6b28ee6331839432f69d87fdcb">CONVOLUTION1x3_BIFROST2X1_STRIDE2</a>(pixels1, src30, src31, weights_row1);</div><div class="line"><a name="l00402"></a><span class="lineno"> 402</span>  <a class="code" href="depthwise__convolution_8cl.xhtml#a31cbdd6b28ee6331839432f69d87fdcb">CONVOLUTION1x3_BIFROST2X1_STRIDE2</a>(pixels1, src40, src41, weights_row2);</div><div class="line"><a name="l00403"></a><span class="lineno"> 403</span> </div><div class="line"><a name="l00404"></a><span class="lineno"> 404</span> <span class="preprocessor">#ifdef HAS_BIAS</span></div><div class="line"><a name="l00405"></a><span class="lineno"> 405</span>  <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00406"></a><span class="lineno"> 406</span> </div><div class="line"><a name="l00407"></a><span class="lineno"> 407</span>  <span class="keywordtype">float</span> bias = *((__global <span class="keywordtype">float</span> *)(<a class="code" href="helpers_8h.xhtml#a7e4940407322d6f0ccb8b6b86b856019">vector_offset</a>(&biases, get_global_id(2))));</div><div class="line"><a name="l00408"></a><span class="lineno"> 408</span> </div><div class="line"><a name="l00409"></a><span class="lineno"> 409</span>  pixels0 += (float2)bias;</div><div class="line"><a name="l00410"></a><span class="lineno"> 410</span>  pixels1 += (float2)bias;</div><div class="line"><a name="l00411"></a><span class="lineno"> 411</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(HAS_BIAS) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00412"></a><span class="lineno"> 412</span> </div><div class="line"><a name="l00413"></a><span class="lineno"> 413</span>  vstore2(pixels0, 0, (__global <span class="keywordtype">float</span> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 0 * dst_stride_y));</div><div class="line"><a name="l00414"></a><span class="lineno"> 414</span>  vstore2(pixels1, 0, (__global <span class="keywordtype">float</span> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 1 * dst_stride_y));</div><div class="line"><a name="l00415"></a><span class="lineno"> 415</span> }</div><div class="line"><a name="l00416"></a><span class="lineno"> 416</span> </div><div class="line"><a name="l00417"></a><span class="lineno"> 417</span> <span class="preprocessor">#if defined(SRC_WIDTH) && defined(DATA_TYPE)</span></div><div class="line"><a name="l00418"></a><span class="lineno"> 418</span> </div><div class="line"><a name="l00441"></a><span class="lineno"> 441</span> __kernel <span class="keywordtype">void</span> depthwise_weights_reshape(</div><div class="line"><a name="l00442"></a><span class="lineno"> 442</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00443"></a><span class="lineno"> 443</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst)</div><div class="line"><a name="l00444"></a><span class="lineno"> 444</span> #ifdef HAS_BIAS</div><div class="line"><a name="l00445"></a><span class="lineno"> 445</span>  ,</div><div class="line"><a name="l00446"></a><span class="lineno"> 446</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases)</div><div class="line"><a name="l00447"></a><span class="lineno"> 447</span> #endif <span class="comment">/* HAS_BIAS */</span></div><div class="line"><a name="l00448"></a><span class="lineno"> 448</span> )</div><div class="line"><a name="l00449"></a><span class="lineno"> 449</span> {</div><div class="line"><a name="l00450"></a><span class="lineno"> 450</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> src = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(src);</div><div class="line"><a name="l00451"></a><span class="lineno"> 451</span> <span class="preprocessor">#ifdef HAS_BIAS</span></div><div class="line"><a name="l00452"></a><span class="lineno"> 452</span>  <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00453"></a><span class="lineno"> 453</span> <span class="preprocessor">#endif </span><span class="comment">/* HAS_BIAS */</span><span class="preprocessor"></span></div><div class="line"><a name="l00454"></a><span class="lineno"> 454</span> </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> *input_ptr = (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)src.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>;</div><div class="line"><a name="l00456"></a><span class="lineno"> 456</span>  __global uchar *output_ptr = dst_ptr + dst_offset_first_element_in_bytes + get_global_id(1) * SRC_WIDTH * dst_stride_x + get_global_id(2) * dst_stride_y;</div><div class="line"><a name="l00457"></a><span class="lineno"> 457</span> </div><div class="line"><a name="l00458"></a><span class="lineno"> 458</span>  <span class="keywordflow">for</span>(<span class="keywordtype">int</span> i = 0; i < SRC_WIDTH; ++i, ++input_ptr)</div><div class="line"><a name="l00459"></a><span class="lineno"> 459</span>  {</div><div class="line"><a name="l00460"></a><span class="lineno"> 460</span>  *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + i * dst_stride_x)) = *input_ptr;</div><div class="line"><a name="l00461"></a><span class="lineno"> 461</span>  }</div><div class="line"><a name="l00462"></a><span class="lineno"> 462</span> </div><div class="line"><a name="l00463"></a><span class="lineno"> 463</span> <span class="preprocessor">#if defined(HAS_BIAS)</span></div><div class="line"><a name="l00464"></a><span class="lineno"> 464</span>  <span class="keywordflow">if</span>(get_global_id(1) == 0)</div><div class="line"><a name="l00465"></a><span class="lineno"> 465</span>  {</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> *)(output_ptr + SRC_WIDTH * get_global_size(1) * dst_stride_x)) = *((__global <span class="keywordtype">float</span> *)(biases.<a class="code" href="struct_vector.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + get_global_id(2) * biases_stride_x));</div><div class="line"><a name="l00467"></a><span class="lineno"> 467</span>  }</div><div class="line"><a name="l00468"></a><span class="lineno"> 468</span> <span class="preprocessor">#endif // defined(HAS_BIAS)</span></div><div class="line"><a name="l00469"></a><span class="lineno"> 469</span> }</div><div class="line"><a name="l00470"></a><span class="lineno"> 470</span> <span class="preprocessor">#endif //defined(SRC_WIDTH) && defined(DATA_TYPE)</span></div><div class="line"><a name="l00471"></a><span class="lineno"> 471</span> </div><div class="line"><a name="l00472"></a><span class="lineno"> 472</span> <span class="preprocessor">#if defined(STRIDE_X) && defined(STRIDE_Y) && defined(PAD_LEFT) && defined(PAD_TOP) && defined(PAD_RIGHT) && defined(PAD_BOTTOM) && defined(KERNEL_WIDTH) && defined(KERNEL_HEIGHT) && defined(SRC_WIDTH) && defined(SRC_HEIGHT) && defined(DATA_TYPE) && defined(PAD_VALUE)</span></div><div class="line"><a name="l00473"></a><span class="lineno"> 473</span> </div><div class="line"><a name="l00495"></a><span class="lineno"> 495</span> __kernel <span class="keywordtype">void</span> depthwise_im2col(<a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src), <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(dst))</div><div class="line"><a name="l00496"></a><span class="lineno"> 496</span> {</div><div class="line"><a name="l00497"></a><span class="lineno"> 497</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> dst = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(dst);</div><div class="line"><a name="l00498"></a><span class="lineno"> 498</span> </div><div class="line"><a name="l00499"></a><span class="lineno"> 499</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> src_pixel_linear = get_global_id(1) * STRIDE_X;</div><div class="line"><a name="l00500"></a><span class="lineno"> 500</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> full_length = SRC_WIDTH + PAD_LEFT + PAD_RIGHT;</div><div class="line"><a name="l00501"></a><span class="lineno"> 501</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> max_initial_x = STRIDE_X * (((full_length - KERNEL_WIDTH) / STRIDE_X) + 1);</div><div class="line"><a name="l00502"></a><span class="lineno"> 502</span> </div><div class="line"><a name="l00503"></a><span class="lineno"> 503</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> src_x = -PAD_LEFT + src_pixel_linear % max_initial_x;</div><div class="line"><a name="l00504"></a><span class="lineno"> 504</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> src_y = -PAD_TOP + src_pixel_linear / max_initial_x * STRIDE_Y;</div><div class="line"><a name="l00505"></a><span class="lineno"> 505</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> src_z = get_global_id(2);</div><div class="line"><a name="l00506"></a><span class="lineno"> 506</span> </div><div class="line"><a name="l00507"></a><span class="lineno"> 507</span>  __global uchar *input_ptr = src_ptr + src_offset_first_element_in_bytes + src_z * src_stride_z;</div><div class="line"><a name="l00508"></a><span class="lineno"> 508</span>  __global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *output_ptr = ((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>));</div><div class="line"><a name="l00509"></a><span class="lineno"> 509</span> </div><div class="line"><a name="l00510"></a><span class="lineno"> 510</span>  <span class="keywordflow">for</span>(<span class="keywordtype">int</span> y = src_y; y < src_y + KERNEL_HEIGHT; ++y)</div><div class="line"><a name="l00511"></a><span class="lineno"> 511</span>  {</div><div class="line"><a name="l00512"></a><span class="lineno"> 512</span>  <span class="keywordflow">for</span>(<span class="keywordtype">int</span> x = src_x; x < src_x + KERNEL_WIDTH; ++x, ++output_ptr)</div><div class="line"><a name="l00513"></a><span class="lineno"> 513</span>  {</div><div class="line"><a name="l00514"></a><span class="lineno"> 514</span>  <span class="keywordflow">if</span>(x < 0 || x >= SRC_WIDTH || y < 0 || y >= SRC_HEIGHT)</div><div class="line"><a name="l00515"></a><span class="lineno"> 515</span>  {</div><div class="line"><a name="l00516"></a><span class="lineno"> 516</span>  *output_ptr = PAD_VALUE;</div><div class="line"><a name="l00517"></a><span class="lineno"> 517</span>  }</div><div class="line"><a name="l00518"></a><span class="lineno"> 518</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00519"></a><span class="lineno"> 519</span>  {</div><div class="line"><a name="l00520"></a><span class="lineno"> 520</span>  *output_ptr = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + x * src_stride_x + y * src_stride_y));</div><div class="line"><a name="l00521"></a><span class="lineno"> 521</span>  }</div><div class="line"><a name="l00522"></a><span class="lineno"> 522</span>  }</div><div class="line"><a name="l00523"></a><span class="lineno"> 523</span>  }</div><div class="line"><a name="l00524"></a><span class="lineno"> 524</span> <span class="preprocessor">#if defined(HAS_BIAS)</span></div><div class="line"><a name="l00525"></a><span class="lineno"> 525</span>  *output_ptr = (<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)(1);</div><div class="line"><a name="l00526"></a><span class="lineno"> 526</span> <span class="preprocessor">#endif // defined(HAS_BIAS)</span></div><div class="line"><a name="l00527"></a><span class="lineno"> 527</span> }</div><div class="line"><a name="l00528"></a><span class="lineno"> 528</span> </div><div class="line"><a name="l00529"></a><span class="lineno"> 529</span> <span class="preprocessor">#endif //defined(STRIDE_X) && defined(STRIDE_Y) && defined(PAD_LEFT) && defined(PAD_TOP) && defined(PAD_RIGHT) && defined(PAD_BOTTOM) && defined(KERNEL_WIDTH) && defined(KERNEL_HEIGHT) && defined(SRC_WIDTH) && defined(DATA_TYPE) && defined(PAD_VALUE)</span></div><div class="line"><a name="l00530"></a><span class="lineno"> 530</span> </div><div class="line"><a name="l00531"></a><span class="lineno"> 531</span> <span class="preprocessor">#if defined(CONV_WIDTH) && defined(CONV_HEIGHT) && defined(DATA_TYPE)</span></div><div class="line"><a name="l00532"></a><span class="lineno"> 532</span> </div><div class="line"><a name="l00551"></a><span class="lineno"> 551</span> __kernel <span class="keywordtype">void</span> depthwise_vector_to_tensor(</div><div class="line"><a name="l00552"></a><span class="lineno"> 552</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(src),</div><div class="line"><a name="l00553"></a><span class="lineno"> 553</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(dst))</div><div class="line"><a name="l00554"></a><span class="lineno"> 554</span> {</div><div class="line"><a name="l00555"></a><span class="lineno"> 555</span>  <a class="code" href="struct_vector.xhtml">Vector</a> src = <a class="code" href="helpers_8h.xhtml#a527bfdf5eeb306f1cf01c4a8e29f38e0">CONVERT_TO_VECTOR_STRUCT</a>(src);</div><div class="line"><a name="l00556"></a><span class="lineno"> 556</span> </div><div class="line"><a name="l00557"></a><span class="lineno"> 557</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> patch_size = CONV_WIDTH * CONV_HEIGHT;</div><div class="line"><a name="l00558"></a><span class="lineno"> 558</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> id0 = get_global_id(0);</div><div class="line"><a name="l00559"></a><span class="lineno"> 559</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> z = id0 / patch_size;</div><div class="line"><a name="l00560"></a><span class="lineno"> 560</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> index2D = id0 - z * patch_size;</div><div class="line"><a name="l00561"></a><span class="lineno"> 561</span> </div><div class="line"><a name="l00562"></a><span class="lineno"> 562</span>  __global uchar *out_ptr = dst_ptr + dst_offset_first_element_in_bytes + index2D % CONV_WIDTH * dst_stride_x + index2D / CONV_WIDTH * dst_stride_y + z * dst_stride_z;</div><div class="line"><a name="l00563"></a><span class="lineno"> 563</span>  *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)out_ptr) = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)src.<a class="code" href="struct_vector.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l00564"></a><span class="lineno"> 564</span> }</div><div class="line"><a name="l00565"></a><span class="lineno"> 565</span> </div><div class="line"><a name="l00566"></a><span class="lineno"> 566</span> <span class="preprocessor">#endif //defined(CONV_WIDTH) && defined(CONV_HEIGHT) && defined(DATA_TYPE)</span></div><div class="line"><a name="l00567"></a><span class="lineno"> 567</span> </div><div class="line"><a name="l00568"></a><span class="lineno"> 568</span> <span class="preprocessor">#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED)</span></div><div class="line"><a name="l00569"></a><span class="lineno"> 569</span> <span class="preprocessor">#if defined(CONV_STRIDE_X)</span></div><div class="line"><a name="l00570"></a><span class="lineno"> 570</span> <span class="preprocessor">#if CONV_STRIDE_X == 1</span></div><div class="line"><a name="l00571"></a><span class="lineno"> 571</span> <span class="preprocessor">#define convolution1x3_f16 convolution1x3_stride_1_f16</span></div><div class="line"><a name="l00572"></a><span class="lineno"> 572</span> <span class="preprocessor">#elif CONV_STRIDE_X == 2</span></div><div class="line"><a name="l00573"></a><span class="lineno"> 573</span> <span class="preprocessor">#define convolution1x3_f16 convolution1x3_stride_2_f16</span></div><div class="line"><a name="l00574"></a><span class="lineno"> 574</span> <span class="preprocessor">#elif CONV_STRIDE_X == 3</span></div><div class="line"><a name="l00575"></a><span class="lineno"> 575</span> <span class="preprocessor">#define convolution1x3_f16 convolution1x3_stride_3_f16</span></div><div class="line"><a name="l00576"></a><span class="lineno"> 576</span> <span class="preprocessor">#else </span><span class="comment">/* CONV_STRIDE_X */</span><span class="preprocessor"></span></div><div class="line"><a name="l00577"></a><span class="lineno"> 577</span> <span class="preprocessor">#error "Stride not supported"</span></div><div class="line"><a name="l00578"></a><span class="lineno"> 578</span> <span class="preprocessor">#endif </span><span class="comment">/* CONV_STRIDE_X */</span><span class="preprocessor"></span></div><div class="line"><a name="l00579"></a><span class="lineno"> 579</span> </div><div class="line"><a name="l00589"></a><span class="lineno"> 589</span> <span class="keyword">inline</span> half4 convolution1x3_stride_1_f16(__global <span class="keyword">const</span> uchar *left_pixel,</div><div class="line"><a name="l00590"></a><span class="lineno"> 590</span>  <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> left_coeff,</div><div class="line"><a name="l00591"></a><span class="lineno"> 591</span>  <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> middle_coeff,</div><div class="line"><a name="l00592"></a><span class="lineno"> 592</span>  <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> right_coeff)</div><div class="line"><a name="l00593"></a><span class="lineno"> 593</span> {</div><div class="line"><a name="l00594"></a><span class="lineno"> 594</span>  half8 temp = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)left_pixel);</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>  half4 left = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp.s0123, half4);</div><div class="line"><a name="l00597"></a><span class="lineno"> 597</span>  half4 middle = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp.s1234, half4);</div><div class="line"><a name="l00598"></a><span class="lineno"> 598</span>  half4 right = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp.s2345, half4);</div><div class="line"><a name="l00599"></a><span class="lineno"> 599</span> </div><div class="line"><a name="l00600"></a><span class="lineno"> 600</span>  <span class="keywordflow">return</span> left * (half4)left_coeff + middle * (half4)middle_coeff + right * (half4)right_coeff;</div><div class="line"><a name="l00601"></a><span class="lineno"> 601</span> }</div><div class="line"><a name="l00602"></a><span class="lineno"> 602</span> </div><div class="line"><a name="l00612"></a><span class="lineno"> 612</span> <span class="keyword">inline</span> half4 convolution1x3_stride_2_f16(__global <span class="keyword">const</span> uchar *left_pixel,</div><div class="line"><a name="l00613"></a><span class="lineno"> 613</span>  <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> left_coeff,</div><div class="line"><a name="l00614"></a><span class="lineno"> 614</span>  <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> middle_coeff,</div><div class="line"><a name="l00615"></a><span class="lineno"> 615</span>  <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> right_coeff)</div><div class="line"><a name="l00616"></a><span class="lineno"> 616</span> {</div><div class="line"><a name="l00617"></a><span class="lineno"> 617</span>  half8 temp0 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)left_pixel);</div><div class="line"><a name="l00618"></a><span class="lineno"> 618</span>  <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> temp1 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(left_pixel + 8 * <span class="keyword">sizeof</span>(<a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a>)));</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>  half4 left = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s0246, half4);</div><div class="line"><a name="l00621"></a><span class="lineno"> 621</span>  half4 middle = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s1357, half4);</div><div class="line"><a name="l00622"></a><span class="lineno"> 622</span>  half4 right = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>((half4)(temp0.s246, temp1), half4);</div><div class="line"><a name="l00623"></a><span class="lineno"> 623</span> </div><div class="line"><a name="l00624"></a><span class="lineno"> 624</span>  <span class="keywordflow">return</span> left * (half4)left_coeff + middle * (half4)middle_coeff + right * (half4)right_coeff;</div><div class="line"><a name="l00625"></a><span class="lineno"> 625</span> }</div><div class="line"><a name="l00626"></a><span class="lineno"> 626</span> </div><div class="line"><a name="l00636"></a><span class="lineno"> 636</span> <span class="keyword">inline</span> half4 convolution1x3_stride_3_f16(__global <span class="keyword">const</span> uchar *left_pixel,</div><div class="line"><a name="l00637"></a><span class="lineno"> 637</span>  <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> left_coeff,</div><div class="line"><a name="l00638"></a><span class="lineno"> 638</span>  <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> middle_coeff,</div><div class="line"><a name="l00639"></a><span class="lineno"> 639</span>  <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> right_coeff)</div><div class="line"><a name="l00640"></a><span class="lineno"> 640</span> {</div><div class="line"><a name="l00641"></a><span class="lineno"> 641</span>  half16 temp0 = vload16(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)left_pixel);</div><div class="line"><a name="l00642"></a><span class="lineno"> 642</span> </div><div class="line"><a name="l00643"></a><span class="lineno"> 643</span>  half4 left = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s0369, half4);</div><div class="line"><a name="l00644"></a><span class="lineno"> 644</span>  half4 middle = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s147A, half4);</div><div class="line"><a name="l00645"></a><span class="lineno"> 645</span>  half4 right = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s258B, half4);</div><div class="line"><a name="l00646"></a><span class="lineno"> 646</span> </div><div class="line"><a name="l00647"></a><span class="lineno"> 647</span>  <span class="keywordflow">return</span> left * (half4)left_coeff + middle * (half4)middle_coeff + right * (half4)right_coeff;</div><div class="line"><a name="l00648"></a><span class="lineno"> 648</span> }</div><div class="line"><a name="l00649"></a><span class="lineno"> 649</span> </div><div class="line"><a name="l00672"></a><span class="lineno"> 672</span> <span class="keyword">inline</span> half4 convolution3x3_f16(</div><div class="line"><a name="l00673"></a><span class="lineno"> 673</span>  <a class="code" href="struct_image.xhtml">Image</a> *src,</div><div class="line"><a name="l00674"></a><span class="lineno"> 674</span>  <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat0, <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat1, <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat2,</div><div class="line"><a name="l00675"></a><span class="lineno"> 675</span>  <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat3, <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat4, <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat5,</div><div class="line"><a name="l00676"></a><span class="lineno"> 676</span>  <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat6, <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat7, <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat8)</div><div class="line"><a name="l00677"></a><span class="lineno"> 677</span> {</div><div class="line"><a name="l00678"></a><span class="lineno"> 678</span>  half4 pixels;</div><div class="line"><a name="l00679"></a><span class="lineno"> 679</span> </div><div class="line"><a name="l00680"></a><span class="lineno"> 680</span>  pixels = convolution1x3_f16(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(src, 0, 0), mat0, mat1, mat2);</div><div class="line"><a name="l00681"></a><span class="lineno"> 681</span>  pixels += convolution1x3_f16(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(src, 0, 1), mat3, mat4, mat5);</div><div class="line"><a name="l00682"></a><span class="lineno"> 682</span>  pixels += convolution1x3_f16(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(src, 0, 2), mat6, mat7, mat8);</div><div class="line"><a name="l00683"></a><span class="lineno"> 683</span> </div><div class="line"><a name="l00684"></a><span class="lineno"> 684</span>  <span class="keywordflow">return</span> pixels;</div><div class="line"><a name="l00685"></a><span class="lineno"> 685</span> }</div><div class="line"><a name="l00686"></a><span class="lineno"> 686</span> </div><div class="line"><a name="l00718"></a><span class="lineno"> 718</span> __kernel <span class="keywordtype">void</span> depthwise_convolution_3x3_f16(</div><div class="line"><a name="l00719"></a><span class="lineno"> 719</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00720"></a><span class="lineno"> 720</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(dst),</div><div class="line"><a name="l00721"></a><span class="lineno"> 721</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(weights)</div><div class="line"><a name="l00722"></a><span class="lineno"> 722</span> #<span class="keywordflow">if</span> defined(HAS_BIAS)</div><div class="line"><a name="l00723"></a><span class="lineno"> 723</span>  ,</div><div class="line"><a name="l00724"></a><span class="lineno"> 724</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases)</div><div class="line"><a name="l00725"></a><span class="lineno"> 725</span> #endif <span class="comment">//defined(HAS_BIAS)</span></div><div class="line"><a name="l00726"></a><span class="lineno"> 726</span> )</div><div class="line"><a name="l00727"></a><span class="lineno"> 727</span> {</div><div class="line"><a name="l00728"></a><span class="lineno"> 728</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(src);</div><div class="line"><a name="l00729"></a><span class="lineno"> 729</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l00730"></a><span class="lineno"> 730</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> weights = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(weights);</div><div class="line"><a name="l00731"></a><span class="lineno"> 731</span> <span class="preprocessor">#if defined(HAS_BIAS)</span></div><div class="line"><a name="l00732"></a><span class="lineno"> 732</span>  <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00733"></a><span class="lineno"> 733</span> <span class="preprocessor">#endif //defined(HAS_BIAS)</span></div><div class="line"><a name="l00734"></a><span class="lineno"> 734</span> </div><div class="line"><a name="l00735"></a><span class="lineno"> 735</span>  uchar3 offset = (uchar3)(0, 1, 2) * (uchar3)weights_stride_y;</div><div class="line"><a name="l00736"></a><span class="lineno"> 736</span>  half3 weights_values0 = vload3(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + offset.s0));</div><div class="line"><a name="l00737"></a><span class="lineno"> 737</span>  half3 weights_values1 = vload3(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + offset.s1));</div><div class="line"><a name="l00738"></a><span class="lineno"> 738</span>  half3 weights_values2 = vload3(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + offset.s2));</div><div class="line"><a name="l00739"></a><span class="lineno"> 739</span> </div><div class="line"><a name="l00740"></a><span class="lineno"> 740</span>  half4 pixels = convolution3x3_f16(&src, weights_values0.s0, weights_values0.s1, weights_values0.s2,</div><div class="line"><a name="l00741"></a><span class="lineno"> 741</span>  weights_values1.s0, weights_values1.s1, weights_values1.s2,</div><div class="line"><a name="l00742"></a><span class="lineno"> 742</span>  weights_values2.s0, weights_values2.s1, weights_values2.s2);</div><div class="line"><a name="l00743"></a><span class="lineno"> 743</span> <span class="preprocessor">#if defined(HAS_BIAS)</span></div><div class="line"><a name="l00744"></a><span class="lineno"> 744</span>  pixels += (half4)(*((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(biases.<a class="code" href="struct_vector.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + get_global_id(2) * biases_stride_x)));</div><div class="line"><a name="l00745"></a><span class="lineno"> 745</span> <span class="preprocessor">#endif //defined(HAS_BIAS)</span></div><div class="line"><a name="l00746"></a><span class="lineno"> 746</span> </div><div class="line"><a name="l00747"></a><span class="lineno"> 747</span>  vstore4(pixels, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l00748"></a><span class="lineno"> 748</span> }</div><div class="line"><a name="l00749"></a><span class="lineno"> 749</span> <span class="preprocessor">#endif // defined(CONV_STRIDE_X)</span></div><div class="line"><a name="l00750"></a><span class="lineno"> 750</span> <span class="preprocessor">#endif // defined(ARM_COMPUTE_OPENCL_FP16_ENABLED)</span></div><div class="ttc" id="struct_vector_xhtml"><div class="ttname"><a href="struct_vector.xhtml">Vector</a></div><div class="ttdoc">Structure to hold Vector information. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00132">helpers.h:132</a></div></div>
+<a href="depthwise__convolution_8cl.xhtml">Go to the documentation of this file.</a><div class="fragment"><div class="line"><a name="l00001"></a><span class="lineno"> 1</span> <span class="comment">/*</span></div><div class="line"><a name="l00002"></a><span class="lineno"> 2</span> <span class="comment"> * Copyright (c) 2017-2018 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> </div><div class="line"><a name="l00025"></a><span class="lineno"> 25</span> <span class="preprocessor">#include "<a class="code" href="helpers_8h.xhtml">helpers.h</a>"</span></div><div class="line"><a name="l00026"></a><span class="lineno"> 26</span> </div><div class="line"><a name="l00027"></a><span class="lineno"> 27</span> <span class="preprocessor">#if defined(DEPTH_MULTIPLIER)</span></div><div class="line"><a name="l00028"></a><span class="lineno"> 28</span> <span class="preprocessor">#if defined(CONV_STRIDE_X)</span></div><div class="line"><a name="l00029"></a><span class="lineno"> 29</span> </div><div class="line"><a name="l00030"></a><span class="lineno"> 30</span> <span class="preprocessor">#if CONV_STRIDE_X == 1</span></div><div class="line"><a name="l00031"></a><span class="lineno"> 31</span> <span class="preprocessor">#define convolution1x3 convolution1x3_stride_1</span></div><div class="line"><a name="l00032"></a><span class="lineno"> 32</span> <span class="preprocessor">#elif CONV_STRIDE_X == 2</span></div><div class="line"><a name="l00033"></a><span class="lineno"> 33</span> <span class="preprocessor">#define convolution1x3 convolution1x3_stride_2</span></div><div class="line"><a name="l00034"></a><span class="lineno"> 34</span> <span class="preprocessor">#elif CONV_STRIDE_X == 3</span></div><div class="line"><a name="l00035"></a><span class="lineno"> 35</span> <span class="preprocessor">#define convolution1x3 convolution1x3_stride_3</span></div><div class="line"><a name="l00036"></a><span class="lineno"> 36</span> <span class="preprocessor">#else </span><span class="comment">/* CONV_STRIDE_X */</span><span class="preprocessor"></span></div><div class="line"><a name="l00037"></a><span class="lineno"> 37</span> <span class="preprocessor">#error "Stride not supported"</span></div><div class="line"><a name="l00038"></a><span class="lineno"> 38</span> <span class="preprocessor">#endif </span><span class="comment">/* CONV_STRIDE_X */</span><span class="preprocessor"></span></div><div class="line"><a name="l00039"></a><span class="lineno"> 39</span> </div><div class="line"><a name="l00049"></a><span class="lineno"> 49</span> <span class="keyword">inline</span> float2 convolution1x3_stride_1(__global <span class="keyword">const</span> uchar *left_pixel,</div><div class="line"><a name="l00050"></a><span class="lineno"> 50</span>  <span class="keyword">const</span> <span class="keywordtype">float</span> left_coeff,</div><div class="line"><a name="l00051"></a><span class="lineno"> 51</span>  <span class="keyword">const</span> <span class="keywordtype">float</span> middle_coeff,</div><div class="line"><a name="l00052"></a><span class="lineno"> 52</span>  <span class="keyword">const</span> <span class="keywordtype">float</span> right_coeff)</div><div class="line"><a name="l00053"></a><span class="lineno"> 53</span> {</div><div class="line"><a name="l00054"></a><span class="lineno"> 54</span>  float4 temp = vload4(0, (__global <span class="keywordtype">float</span> *)left_pixel);</div><div class="line"><a name="l00055"></a><span class="lineno"> 55</span> </div><div class="line"><a name="l00056"></a><span class="lineno"> 56</span>  float2 left = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp.s01, float2);</div><div class="line"><a name="l00057"></a><span class="lineno"> 57</span>  float2 middle = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp.s12, float2);</div><div class="line"><a name="l00058"></a><span class="lineno"> 58</span>  float2 right = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp.s23, float2);</div><div class="line"><a name="l00059"></a><span class="lineno"> 59</span> </div><div class="line"><a name="l00060"></a><span class="lineno"> 60</span>  <span class="keywordflow">return</span> left * (float2)left_coeff + middle * (float2)middle_coeff + right * (float2)right_coeff;</div><div class="line"><a name="l00061"></a><span class="lineno"> 61</span> }</div><div class="line"><a name="l00062"></a><span class="lineno"> 62</span> </div><div class="line"><a name="l00072"></a><span class="lineno"> 72</span> inline float2 convolution1x3_stride_2(__global const uchar *left_pixel,</div><div class="line"><a name="l00073"></a><span class="lineno"> 73</span>  const <span class="keywordtype">float</span> left_coeff,</div><div class="line"><a name="l00074"></a><span class="lineno"> 74</span>  const <span class="keywordtype">float</span> middle_coeff,</div><div class="line"><a name="l00075"></a><span class="lineno"> 75</span>  const <span class="keywordtype">float</span> right_coeff)</div><div class="line"><a name="l00076"></a><span class="lineno"> 76</span> {</div><div class="line"><a name="l00077"></a><span class="lineno"> 77</span>  float4 temp0 = vload4(0, (__global <span class="keywordtype">float</span> *)left_pixel);</div><div class="line"><a name="l00078"></a><span class="lineno"> 78</span>  <span class="keywordtype">float</span> temp1 = *((__global <span class="keywordtype">float</span> *)(left_pixel + 4 * <span class="keyword">sizeof</span>(<span class="keywordtype">float</span>)));</div><div class="line"><a name="l00079"></a><span class="lineno"> 79</span> </div><div class="line"><a name="l00080"></a><span class="lineno"> 80</span>  float2 left = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s02, float2);</div><div class="line"><a name="l00081"></a><span class="lineno"> 81</span>  float2 middle = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s13, float2);</div><div class="line"><a name="l00082"></a><span class="lineno"> 82</span>  float2 right = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>((float2)(temp0.s2, temp1), float2);</div><div class="line"><a name="l00083"></a><span class="lineno"> 83</span> </div><div class="line"><a name="l00084"></a><span class="lineno"> 84</span>  <span class="keywordflow">return</span> left * (float2)left_coeff + middle * (float2)middle_coeff + right * (float2)right_coeff;</div><div class="line"><a name="l00085"></a><span class="lineno"> 85</span> }</div><div class="line"><a name="l00086"></a><span class="lineno"> 86</span> </div><div class="line"><a name="l00096"></a><span class="lineno"> 96</span> inline float2 convolution1x3_stride_3(__global const uchar *left_pixel,</div><div class="line"><a name="l00097"></a><span class="lineno"> 97</span>  const <span class="keywordtype">float</span> left_coeff,</div><div class="line"><a name="l00098"></a><span class="lineno"> 98</span>  const <span class="keywordtype">float</span> middle_coeff,</div><div class="line"><a name="l00099"></a><span class="lineno"> 99</span>  const <span class="keywordtype">float</span> right_coeff)</div><div class="line"><a name="l00100"></a><span class="lineno"> 100</span> {</div><div class="line"><a name="l00101"></a><span class="lineno"> 101</span>  float4 temp0 = vload4(0, (__global <span class="keywordtype">float</span> *)left_pixel);</div><div class="line"><a name="l00102"></a><span class="lineno"> 102</span>  float2 temp1 = vload2(0, (__global <span class="keywordtype">float</span> *)(left_pixel + 4 * <span class="keyword">sizeof</span>(<span class="keywordtype">float</span>)));</div><div class="line"><a name="l00103"></a><span class="lineno"> 103</span> </div><div class="line"><a name="l00104"></a><span class="lineno"> 104</span>  float2 left = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s03, float2);</div><div class="line"><a name="l00105"></a><span class="lineno"> 105</span>  float2 middle = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>((float2)(temp0.s1, temp1.s0), float2);</div><div class="line"><a name="l00106"></a><span class="lineno"> 106</span>  float2 right = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>((float2)(temp0.s2, temp1.s1), float2);</div><div class="line"><a name="l00107"></a><span class="lineno"> 107</span> </div><div class="line"><a name="l00108"></a><span class="lineno"> 108</span>  <span class="keywordflow">return</span> left * (float2)left_coeff + middle * (float2)middle_coeff + right * (float2)right_coeff;</div><div class="line"><a name="l00109"></a><span class="lineno"> 109</span> }</div><div class="line"><a name="l00110"></a><span class="lineno"> 110</span> </div><div class="line"><a name="l00133"></a><span class="lineno"> 133</span> inline float2 <a class="code" href="convolution3x3_8cl.xhtml#afc5fefe72e66f0ae5191fd5b708fade9">convolution3x3</a>(</div><div class="line"><a name="l00134"></a><span class="lineno"> 134</span>  <a class="code" href="struct_image.xhtml">Image</a> *<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a6743f0a130e8311e6f5b1a23df102472">src</a>,</div><div class="line"><a name="l00135"></a><span class="lineno"> 135</span>  const <span class="keywordtype">float</span> mat0, const <span class="keywordtype">float</span> mat1, const <span class="keywordtype">float</span> mat2,</div><div class="line"><a name="l00136"></a><span class="lineno"> 136</span>  const <span class="keywordtype">float</span> mat3, const <span class="keywordtype">float</span> mat4, const <span class="keywordtype">float</span> mat5,</div><div class="line"><a name="l00137"></a><span class="lineno"> 137</span>  const <span class="keywordtype">float</span> mat6, const <span class="keywordtype">float</span> mat7, const <span class="keywordtype">float</span> mat8)</div><div class="line"><a name="l00138"></a><span class="lineno"> 138</span> {</div><div class="line"><a name="l00139"></a><span class="lineno"> 139</span>  float2 pixels;</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>  pixels = <a class="code" href="convolution3x3_8cl.xhtml#a92702074338198e81a46c3e309d9b04f">convolution1x3</a>(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(src, 0, 0), mat0, mat1, mat2);</div><div class="line"><a name="l00142"></a><span class="lineno"> 142</span>  pixels += <a class="code" href="convolution3x3_8cl.xhtml#a92702074338198e81a46c3e309d9b04f">convolution1x3</a>(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(src, 0, 1), mat3, mat4, mat5);</div><div class="line"><a name="l00143"></a><span class="lineno"> 143</span>  pixels += <a class="code" href="convolution3x3_8cl.xhtml#a92702074338198e81a46c3e309d9b04f">convolution1x3</a>(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(src, 0, 2), mat6, mat7, mat8);</div><div class="line"><a name="l00144"></a><span class="lineno"> 144</span> </div><div class="line"><a name="l00145"></a><span class="lineno"> 145</span>  <span class="keywordflow">return</span> pixels;</div><div class="line"><a name="l00146"></a><span class="lineno"> 146</span> }</div><div class="line"><a name="l00147"></a><span class="lineno"> 147</span> </div><div class="line"><a name="l00179"></a><span class="lineno"> 179</span> __kernel <span class="keywordtype">void</span> depthwise_convolution_3x3(</div><div class="line"><a name="l00180"></a><span class="lineno"> 180</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00181"></a><span class="lineno"> 181</span>  <a class="code" href="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="l00182"></a><span class="lineno"> 182</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(weights)</div><div class="line"><a name="l00183"></a><span class="lineno"> 183</span> #<span class="keywordflow">if</span> defined(HAS_BIAS)</div><div class="line"><a name="l00184"></a><span class="lineno"> 184</span>  ,</div><div class="line"><a name="l00185"></a><span class="lineno"> 185</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases)</div><div class="line"><a name="l00186"></a><span class="lineno"> 186</span> #endif <span class="comment">//defined(HAS_BIAS)</span></div><div class="line"><a name="l00187"></a><span class="lineno"> 187</span> )</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>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(src);</div><div class="line"><a name="l00190"></a><span class="lineno"> 190</span>  <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a> = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l00191"></a><span class="lineno"> 191</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> weights = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(weights);</div><div class="line"><a name="l00192"></a><span class="lineno"> 192</span> <span class="preprocessor">#if defined(HAS_BIAS)</span></div><div class="line"><a name="l00193"></a><span class="lineno"> 193</span>  <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00194"></a><span class="lineno"> 194</span> <span class="preprocessor">#endif //defined(HAS_BIAS)</span></div><div class="line"><a name="l00195"></a><span class="lineno"> 195</span> </div><div class="line"><a name="l00196"></a><span class="lineno"> 196</span>  src.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> -= (get_global_id(2) - get_global_id(2) / DEPTH_MULTIPLIER) * src_step_z;</div><div class="line"><a name="l00197"></a><span class="lineno"> 197</span> </div><div class="line"><a name="l00198"></a><span class="lineno"> 198</span>  uchar3 <a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a> = (uchar3)(0, 1, 2) * (uchar3)weights_stride_y;</div><div class="line"><a name="l00199"></a><span class="lineno"> 199</span>  float3 weights_values0 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + offset.s0));</div><div class="line"><a name="l00200"></a><span class="lineno"> 200</span>  float3 weights_values1 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + offset.s1));</div><div class="line"><a name="l00201"></a><span class="lineno"> 201</span>  float3 weights_values2 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + offset.s2));</div><div class="line"><a name="l00202"></a><span class="lineno"> 202</span> </div><div class="line"><a name="l00203"></a><span class="lineno"> 203</span>  float2 pixels = <a class="code" href="convolution3x3_8cl.xhtml#afc5fefe72e66f0ae5191fd5b708fade9">convolution3x3</a>(&src, weights_values0.s0, weights_values0.s1, weights_values0.s2,</div><div class="line"><a name="l00204"></a><span class="lineno"> 204</span>  weights_values1.s0, weights_values1.s1, weights_values1.s2,</div><div class="line"><a name="l00205"></a><span class="lineno"> 205</span>  weights_values2.s0, weights_values2.s1, weights_values2.s2);</div><div class="line"><a name="l00206"></a><span class="lineno"> 206</span> <span class="preprocessor">#if defined(HAS_BIAS)</span></div><div class="line"><a name="l00207"></a><span class="lineno"> 207</span>  pixels += (float2)(*((__global <span class="keywordtype">float</span> *)(biases.<a class="code" href="struct_vector.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + get_global_id(2) * biases_stride_x)));</div><div class="line"><a name="l00208"></a><span class="lineno"> 208</span> <span class="preprocessor">#endif //defined(HAS_BIAS)</span></div><div class="line"><a name="l00209"></a><span class="lineno"> 209</span> </div><div class="line"><a name="l00210"></a><span class="lineno"> 210</span>  vstore2(pixels, 0, (__global <span class="keywordtype">float</span> *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l00211"></a><span class="lineno"> 211</span> }</div><div class="line"><a name="l00212"></a><span class="lineno"> 212</span> <span class="preprocessor">#endif //defined(CONV_STRIDE_X)</span></div><div class="line"><a name="l00213"></a><span class="lineno"> 213</span> </div><div class="line"><a name="l00214"></a><span class="lineno"> 214</span> <span class="preprocessor">#define CONVOLUTION1x3_BIFROST2X1_STRIDE1(acc, src0, weights_row0) \</span></div><div class="line"><a name="l00215"></a><span class="lineno"> 215</span> <span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00216"></a><span class="lineno"> 216</span> <span class="preprocessor"> acc.s0 = fma(src0.s0, weights_row0.s0, acc.s0); \</span></div><div class="line"><a name="l00217"></a><span class="lineno"> 217</span> <span class="preprocessor"> acc.s0 = fma(src0.s1, weights_row0.s1, acc.s0); \</span></div><div class="line"><a name="l00218"></a><span class="lineno"> 218</span> <span class="preprocessor"> acc.s0 = fma(src0.s2, weights_row0.s2, acc.s0); \</span></div><div class="line"><a name="l00219"></a><span class="lineno"> 219</span> <span class="preprocessor"> acc.s1 = fma(src0.s1, weights_row0.s0, acc.s1); \</span></div><div class="line"><a name="l00220"></a><span class="lineno"> 220</span> <span class="preprocessor"> acc.s1 = fma(src0.s2, weights_row0.s1, acc.s1); \</span></div><div class="line"><a name="l00221"></a><span class="lineno"> 221</span> <span class="preprocessor"> acc.s1 = fma(src0.s3, weights_row0.s2, acc.s1); \</span></div><div class="line"><a name="l00222"></a><span class="lineno"> 222</span> <span class="preprocessor"> })</span></div><div class="line"><a name="l00223"></a><span class="lineno"> 223</span> </div><div class="line"><a name="l00224"></a><span class="lineno"> 224</span> <span class="preprocessor">#define CONVOLUTION1x3_BIFROST4X1_STRIDE1(acc, src0, weights_row0) \</span></div><div class="line"><a name="l00225"></a><span class="lineno"> 225</span> <span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00226"></a><span class="lineno"> 226</span> <span class="preprocessor"> acc.s0 = fma(src0.s0, weights_row0.s0, acc.s0); \</span></div><div class="line"><a name="l00227"></a><span class="lineno"> 227</span> <span class="preprocessor"> acc.s0 = fma(src0.s1, weights_row0.s1, acc.s0); \</span></div><div class="line"><a name="l00228"></a><span class="lineno"> 228</span> <span class="preprocessor"> acc.s0 = fma(src0.s2, weights_row0.s2, acc.s0); \</span></div><div class="line"><a name="l00229"></a><span class="lineno"> 229</span> <span class="preprocessor"> acc.s1 = fma(src0.s1, weights_row0.s0, acc.s1); \</span></div><div class="line"><a name="l00230"></a><span class="lineno"> 230</span> <span class="preprocessor"> acc.s1 = fma(src0.s2, weights_row0.s1, acc.s1); \</span></div><div class="line"><a name="l00231"></a><span class="lineno"> 231</span> <span class="preprocessor"> acc.s1 = fma(src0.s3, weights_row0.s2, acc.s1); \</span></div><div class="line"><a name="l00232"></a><span class="lineno"> 232</span> <span class="preprocessor"> acc.s2 = fma(src0.s2, weights_row0.s0, acc.s2); \</span></div><div class="line"><a name="l00233"></a><span class="lineno"> 233</span> <span class="preprocessor"> acc.s2 = fma(src0.s3, weights_row0.s1, acc.s2); \</span></div><div class="line"><a name="l00234"></a><span class="lineno"> 234</span> <span class="preprocessor"> acc.s2 = fma(src0.s4, weights_row0.s2, acc.s2); \</span></div><div class="line"><a name="l00235"></a><span class="lineno"> 235</span> <span class="preprocessor"> acc.s3 = fma(src0.s3, weights_row0.s0, acc.s3); \</span></div><div class="line"><a name="l00236"></a><span class="lineno"> 236</span> <span class="preprocessor"> acc.s3 = fma(src0.s4, weights_row0.s1, acc.s3); \</span></div><div class="line"><a name="l00237"></a><span class="lineno"> 237</span> <span class="preprocessor"> acc.s3 = fma(src0.s5, weights_row0.s2, acc.s3); \</span></div><div class="line"><a name="l00238"></a><span class="lineno"> 238</span> <span class="preprocessor"> })</span></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="preprocessor">#define CONVOLUTION1x3_BIFROST2X1_STRIDE2(acc, src0, src1, weights_row0) \</span></div><div class="line"><a name="l00241"></a><span class="lineno"> 241</span> <span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00242"></a><span class="lineno"> 242</span> <span class="preprocessor"> acc.s0 = fma(src0.s0, weights_row0.s0, acc.s0); \</span></div><div class="line"><a name="l00243"></a><span class="lineno"> 243</span> <span class="preprocessor"> acc.s0 = fma(src0.s1, weights_row0.s1, acc.s0); \</span></div><div class="line"><a name="l00244"></a><span class="lineno"> 244</span> <span class="preprocessor"> acc.s0 = fma(src0.s2, weights_row0.s2, acc.s0); \</span></div><div class="line"><a name="l00245"></a><span class="lineno"> 245</span> <span class="preprocessor"> acc.s1 = fma(src0.s2, weights_row0.s0, acc.s1); \</span></div><div class="line"><a name="l00246"></a><span class="lineno"> 246</span> <span class="preprocessor"> acc.s1 = fma(src0.s3, weights_row0.s1, acc.s1); \</span></div><div class="line"><a name="l00247"></a><span class="lineno"> 247</span> <span class="preprocessor"> acc.s1 = fma(src1.s0, weights_row0.s2, acc.s1); \</span></div><div class="line"><a name="l00248"></a><span class="lineno"> 248</span> <span class="preprocessor"> })</span></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="preprocessor">#define CONVOLUTION1x3_BIFROST4X1_STRIDE2(acc, src0, src1, weights_row0) \</span></div><div class="line"><a name="l00251"></a><span class="lineno"> 251</span> <span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00252"></a><span class="lineno"> 252</span> <span class="preprocessor"> acc.s0 = fma(src0.s0, weights_row0.s0, acc.s0); \</span></div><div class="line"><a name="l00253"></a><span class="lineno"> 253</span> <span class="preprocessor"> acc.s0 = fma(src0.s1, weights_row0.s1, acc.s0); \</span></div><div class="line"><a name="l00254"></a><span class="lineno"> 254</span> <span class="preprocessor"> acc.s0 = fma(src0.s2, weights_row0.s2, acc.s0); \</span></div><div class="line"><a name="l00255"></a><span class="lineno"> 255</span> <span class="preprocessor"> acc.s1 = fma(src0.s2, weights_row0.s0, acc.s1); \</span></div><div class="line"><a name="l00256"></a><span class="lineno"> 256</span> <span class="preprocessor"> acc.s1 = fma(src0.s3, weights_row0.s1, acc.s1); \</span></div><div class="line"><a name="l00257"></a><span class="lineno"> 257</span> <span class="preprocessor"> acc.s1 = fma(src0.s4, weights_row0.s2, acc.s1); \</span></div><div class="line"><a name="l00258"></a><span class="lineno"> 258</span> <span class="preprocessor"> acc.s2 = fma(src0.s4, weights_row0.s0, acc.s2); \</span></div><div class="line"><a name="l00259"></a><span class="lineno"> 259</span> <span class="preprocessor"> acc.s2 = fma(src0.s5, weights_row0.s1, acc.s2); \</span></div><div class="line"><a name="l00260"></a><span class="lineno"> 260</span> <span class="preprocessor"> acc.s2 = fma(src0.s6, weights_row0.s2, acc.s2); \</span></div><div class="line"><a name="l00261"></a><span class="lineno"> 261</span> <span class="preprocessor"> acc.s3 = fma(src0.s6, weights_row0.s0, acc.s3); \</span></div><div class="line"><a name="l00262"></a><span class="lineno"> 262</span> <span class="preprocessor"> acc.s3 = fma(src0.s7, weights_row0.s1, acc.s3); \</span></div><div class="line"><a name="l00263"></a><span class="lineno"> 263</span> <span class="preprocessor"> acc.s3 = fma(src1.s0, weights_row0.s2, acc.s3); \</span></div><div class="line"><a name="l00264"></a><span class="lineno"> 264</span> <span class="preprocessor"> })</span></div><div class="line"><a name="l00265"></a><span class="lineno"> 265</span> </div><div class="line"><a name="l00298"></a><span class="lineno"> 298</span> __kernel <span class="keywordtype">void</span> depthwise_convolution_3x3_stridex1_stridey1_bifrost_f32(</div><div class="line"><a name="l00299"></a><span class="lineno"> 299</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00300"></a><span class="lineno"> 300</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(dst),</div><div class="line"><a name="l00301"></a><span class="lineno"> 301</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(weights)</div><div class="line"><a name="l00302"></a><span class="lineno"> 302</span> #<span class="keywordflow">if</span> defined(HAS_BIAS)</div><div class="line"><a name="l00303"></a><span class="lineno"> 303</span>  ,</div><div class="line"><a name="l00304"></a><span class="lineno"> 304</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases)</div><div class="line"><a name="l00305"></a><span class="lineno"> 305</span> #endif <span class="comment">//defined(HAS_BIAS)</span></div><div class="line"><a name="l00306"></a><span class="lineno"> 306</span> )</div><div class="line"><a name="l00307"></a><span class="lineno"> 307</span> {</div><div class="line"><a name="l00308"></a><span class="lineno"> 308</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(src);</div><div class="line"><a name="l00309"></a><span class="lineno"> 309</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l00310"></a><span class="lineno"> 310</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> weights = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(weights);</div><div class="line"><a name="l00311"></a><span class="lineno"> 311</span> </div><div class="line"><a name="l00312"></a><span class="lineno"> 312</span>  float2 pixels0 = 0.0f;</div><div class="line"><a name="l00313"></a><span class="lineno"> 313</span>  float2 pixels1 = 0.0f;</div><div class="line"><a name="l00314"></a><span class="lineno"> 314</span>  float2 pixels2 = 0.0f;</div><div class="line"><a name="l00315"></a><span class="lineno"> 315</span>  float2 pixels3 = 0.0f;</div><div class="line"><a name="l00316"></a><span class="lineno"> 316</span> </div><div class="line"><a name="l00317"></a><span class="lineno"> 317</span>  __global uchar *weights_addr = (__global uchar *)weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>;</div><div class="line"><a name="l00318"></a><span class="lineno"> 318</span>  __global uchar *src_addr = src.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> - (get_global_id(2) - get_global_id(2) / DEPTH_MULTIPLIER) * src_step_z;</div><div class="line"><a name="l00319"></a><span class="lineno"> 319</span> </div><div class="line"><a name="l00320"></a><span class="lineno"> 320</span>  <span class="comment">// Load the weights</span></div><div class="line"><a name="l00321"></a><span class="lineno"> 321</span>  float3 weights_row0 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights_addr + 0 * weights_stride_y));</div><div class="line"><a name="l00322"></a><span class="lineno"> 322</span>  float3 weights_row1 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights_addr + 1 * weights_stride_y));</div><div class="line"><a name="l00323"></a><span class="lineno"> 323</span>  float3 weights_row2 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights_addr + 2 * weights_stride_y));</div><div class="line"><a name="l00324"></a><span class="lineno"> 324</span> </div><div class="line"><a name="l00325"></a><span class="lineno"> 325</span>  <span class="comment">// Note: Since each work-item computes 4x2 elements, we need to load 6 rows from the input tensor</span></div><div class="line"><a name="l00326"></a><span class="lineno"> 326</span>  float4 src00 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 0 * src_stride_y)); <span class="comment">// Row0</span></div><div class="line"><a name="l00327"></a><span class="lineno"> 327</span>  float4 src10 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 1 * src_stride_y)); <span class="comment">// Row1</span></div><div class="line"><a name="l00328"></a><span class="lineno"> 328</span>  float4 src20 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 2 * src_stride_y)); <span class="comment">// Row2</span></div><div class="line"><a name="l00329"></a><span class="lineno"> 329</span>  float4 src30 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 3 * src_stride_y)); <span class="comment">// Row3</span></div><div class="line"><a name="l00330"></a><span class="lineno"> 330</span>  float4 src40 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 4 * src_stride_y)); <span class="comment">// Row4</span></div><div class="line"><a name="l00331"></a><span class="lineno"> 331</span>  float4 src50 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 5 * src_stride_y)); <span class="comment">// Row5</span></div><div class="line"><a name="l00332"></a><span class="lineno"> 332</span> </div><div class="line"><a name="l00333"></a><span class="lineno"> 333</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE1(pixels0, src00, weights_row0);</div><div class="line"><a name="l00334"></a><span class="lineno"> 334</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE1(pixels0, src10, weights_row1);</div><div class="line"><a name="l00335"></a><span class="lineno"> 335</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE1(pixels0, src20, weights_row2);</div><div class="line"><a name="l00336"></a><span class="lineno"> 336</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE1(pixels1, src10, weights_row0);</div><div class="line"><a name="l00337"></a><span class="lineno"> 337</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE1(pixels1, src20, weights_row1);</div><div class="line"><a name="l00338"></a><span class="lineno"> 338</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE1(pixels1, src30, weights_row2);</div><div class="line"><a name="l00339"></a><span class="lineno"> 339</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE1(pixels2, src20, weights_row0);</div><div class="line"><a name="l00340"></a><span class="lineno"> 340</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE1(pixels2, src30, weights_row1);</div><div class="line"><a name="l00341"></a><span class="lineno"> 341</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE1(pixels2, src40, weights_row2);</div><div class="line"><a name="l00342"></a><span class="lineno"> 342</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE1(pixels3, src30, weights_row0);</div><div class="line"><a name="l00343"></a><span class="lineno"> 343</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE1(pixels3, src40, weights_row1);</div><div class="line"><a name="l00344"></a><span class="lineno"> 344</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE1(pixels3, src50, weights_row2);</div><div class="line"><a name="l00345"></a><span class="lineno"> 345</span> </div><div class="line"><a name="l00346"></a><span class="lineno"> 346</span> <span class="preprocessor">#ifdef HAS_BIAS</span></div><div class="line"><a name="l00347"></a><span class="lineno"> 347</span>  <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00348"></a><span class="lineno"> 348</span> </div><div class="line"><a name="l00349"></a><span class="lineno"> 349</span>  <span class="keywordtype">float</span> bias = *((__global <span class="keywordtype">float</span> *)(<a class="code" href="helpers_8h.xhtml#a7e4940407322d6f0ccb8b6b86b856019">vector_offset</a>(&biases, get_global_id(2))));</div><div class="line"><a name="l00350"></a><span class="lineno"> 350</span> </div><div class="line"><a name="l00351"></a><span class="lineno"> 351</span>  pixels0 += (float2)bias;</div><div class="line"><a name="l00352"></a><span class="lineno"> 352</span>  pixels1 += (float2)bias;</div><div class="line"><a name="l00353"></a><span class="lineno"> 353</span>  pixels2 += (float2)bias;</div><div class="line"><a name="l00354"></a><span class="lineno"> 354</span>  pixels3 += (float2)bias;</div><div class="line"><a name="l00355"></a><span class="lineno"> 355</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(HAS_BIAS) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00356"></a><span class="lineno"> 356</span> </div><div class="line"><a name="l00357"></a><span class="lineno"> 357</span>  vstore2(pixels0, 0, (__global <span class="keywordtype">float</span> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 0 * dst_stride_y));</div><div class="line"><a name="l00358"></a><span class="lineno"> 358</span>  vstore2(pixels1, 0, (__global <span class="keywordtype">float</span> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 1 * dst_stride_y));</div><div class="line"><a name="l00359"></a><span class="lineno"> 359</span>  vstore2(pixels2, 0, (__global <span class="keywordtype">float</span> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 2 * dst_stride_y));</div><div class="line"><a name="l00360"></a><span class="lineno"> 360</span>  vstore2(pixels3, 0, (__global <span class="keywordtype">float</span> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 3 * dst_stride_y));</div><div class="line"><a name="l00361"></a><span class="lineno"> 361</span> }</div><div class="line"><a name="l00362"></a><span class="lineno"> 362</span> </div><div class="line"><a name="l00395"></a><span class="lineno"> 395</span> __kernel <span class="keywordtype">void</span> depthwise_convolution_3x3_stridex2_stridey2_bifrost_f32(</div><div class="line"><a name="l00396"></a><span class="lineno"> 396</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00397"></a><span class="lineno"> 397</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(dst),</div><div class="line"><a name="l00398"></a><span class="lineno"> 398</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(weights)</div><div class="line"><a name="l00399"></a><span class="lineno"> 399</span> #<span class="keywordflow">if</span> defined(HAS_BIAS)</div><div class="line"><a name="l00400"></a><span class="lineno"> 400</span>  ,</div><div class="line"><a name="l00401"></a><span class="lineno"> 401</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases)</div><div class="line"><a name="l00402"></a><span class="lineno"> 402</span> #endif <span class="comment">//defined(HAS_BIAS)</span></div><div class="line"><a name="l00403"></a><span class="lineno"> 403</span> )</div><div class="line"><a name="l00404"></a><span class="lineno"> 404</span> {</div><div class="line"><a name="l00405"></a><span class="lineno"> 405</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(src);</div><div class="line"><a name="l00406"></a><span class="lineno"> 406</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l00407"></a><span class="lineno"> 407</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> weights = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(weights);</div><div class="line"><a name="l00408"></a><span class="lineno"> 408</span> </div><div class="line"><a name="l00409"></a><span class="lineno"> 409</span>  float2 pixels0 = 0.0f;</div><div class="line"><a name="l00410"></a><span class="lineno"> 410</span>  float2 pixels1 = 0.0f;</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>  __global uchar *weights_addr = (__global uchar *)weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>;</div><div class="line"><a name="l00413"></a><span class="lineno"> 413</span>  __global uchar *src_addr = src.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> - (get_global_id(2) - get_global_id(2) / DEPTH_MULTIPLIER) * src_step_z;</div><div class="line"><a name="l00414"></a><span class="lineno"> 414</span> </div><div class="line"><a name="l00415"></a><span class="lineno"> 415</span>  <span class="comment">// Load the weights</span></div><div class="line"><a name="l00416"></a><span class="lineno"> 416</span>  float3 weights_row0 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights_addr + 0 * weights_stride_y));</div><div class="line"><a name="l00417"></a><span class="lineno"> 417</span>  float3 weights_row1 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights_addr + 1 * weights_stride_y));</div><div class="line"><a name="l00418"></a><span class="lineno"> 418</span>  float3 weights_row2 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights_addr + 2 * weights_stride_y));</div><div class="line"><a name="l00419"></a><span class="lineno"> 419</span> </div><div class="line"><a name="l00420"></a><span class="lineno"> 420</span>  <span class="comment">// Note: Since each work-item computes 4x2 elements, we need to load 5 rows from the input tensor</span></div><div class="line"><a name="l00421"></a><span class="lineno"> 421</span>  float4 src00 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 0 * src_stride_y)); <span class="comment">// Row0</span></div><div class="line"><a name="l00422"></a><span class="lineno"> 422</span>  float2 src01 = vload2(2, (__global <span class="keywordtype">float</span> *)(src_addr + 0 * src_stride_y)); <span class="comment">// Row0</span></div><div class="line"><a name="l00423"></a><span class="lineno"> 423</span>  float4 src10 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 1 * src_stride_y)); <span class="comment">// Row1</span></div><div class="line"><a name="l00424"></a><span class="lineno"> 424</span>  float2 src11 = vload2(2, (__global <span class="keywordtype">float</span> *)(src_addr + 1 * src_stride_y)); <span class="comment">// Row1</span></div><div class="line"><a name="l00425"></a><span class="lineno"> 425</span>  float4 src20 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 2 * src_stride_y)); <span class="comment">// Row2</span></div><div class="line"><a name="l00426"></a><span class="lineno"> 426</span>  float2 src21 = vload2(2, (__global <span class="keywordtype">float</span> *)(src_addr + 2 * src_stride_y)); <span class="comment">// Row2</span></div><div class="line"><a name="l00427"></a><span class="lineno"> 427</span>  float4 src30 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 3 * src_stride_y)); <span class="comment">// Row3</span></div><div class="line"><a name="l00428"></a><span class="lineno"> 428</span>  float2 src31 = vload2(2, (__global <span class="keywordtype">float</span> *)(src_addr + 3 * src_stride_y)); <span class="comment">// Row3</span></div><div class="line"><a name="l00429"></a><span class="lineno"> 429</span>  float4 src40 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 4 * src_stride_y)); <span class="comment">// Row4</span></div><div class="line"><a name="l00430"></a><span class="lineno"> 430</span>  float2 src41 = vload2(2, (__global <span class="keywordtype">float</span> *)(src_addr + 4 * src_stride_y)); <span class="comment">// Row4</span></div><div class="line"><a name="l00431"></a><span class="lineno"> 431</span> </div><div class="line"><a name="l00432"></a><span class="lineno"> 432</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE2(pixels0, src00, src01, weights_row0);</div><div class="line"><a name="l00433"></a><span class="lineno"> 433</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE2(pixels0, src10, src11, weights_row1);</div><div class="line"><a name="l00434"></a><span class="lineno"> 434</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE2(pixels0, src20, src21, weights_row2);</div><div class="line"><a name="l00435"></a><span class="lineno"> 435</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE2(pixels1, src20, src21, weights_row0);</div><div class="line"><a name="l00436"></a><span class="lineno"> 436</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE2(pixels1, src30, src31, weights_row1);</div><div class="line"><a name="l00437"></a><span class="lineno"> 437</span>  CONVOLUTION1x3_BIFROST2X1_STRIDE2(pixels1, src40, src41, weights_row2);</div><div class="line"><a name="l00438"></a><span class="lineno"> 438</span> </div><div class="line"><a name="l00439"></a><span class="lineno"> 439</span> <span class="preprocessor">#ifdef HAS_BIAS</span></div><div class="line"><a name="l00440"></a><span class="lineno"> 440</span>  <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00441"></a><span class="lineno"> 441</span> </div><div class="line"><a name="l00442"></a><span class="lineno"> 442</span>  <span class="keywordtype">float</span> bias = *((__global <span class="keywordtype">float</span> *)(<a class="code" href="helpers_8h.xhtml#a7e4940407322d6f0ccb8b6b86b856019">vector_offset</a>(&biases, get_global_id(2))));</div><div class="line"><a name="l00443"></a><span class="lineno"> 443</span> </div><div class="line"><a name="l00444"></a><span class="lineno"> 444</span>  pixels0 += (float2)bias;</div><div class="line"><a name="l00445"></a><span class="lineno"> 445</span>  pixels1 += (float2)bias;</div><div class="line"><a name="l00446"></a><span class="lineno"> 446</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(HAS_BIAS) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00447"></a><span class="lineno"> 447</span> </div><div class="line"><a name="l00448"></a><span class="lineno"> 448</span>  vstore2(pixels0, 0, (__global <span class="keywordtype">float</span> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 0 * dst_stride_y));</div><div class="line"><a name="l00449"></a><span class="lineno"> 449</span>  vstore2(pixels1, 0, (__global <span class="keywordtype">float</span> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 1 * dst_stride_y));</div><div class="line"><a name="l00450"></a><span class="lineno"> 450</span> }</div><div class="line"><a name="l00451"></a><span class="lineno"> 451</span> </div><div class="line"><a name="l00452"></a><span class="lineno"> 452</span> <span class="preprocessor">#endif // defined(DEPTH_MULTIPLIER)</span></div><div class="line"><a name="l00453"></a><span class="lineno"> 453</span> </div><div class="line"><a name="l00454"></a><span class="lineno"> 454</span> <span class="preprocessor">#if defined(SRC_WIDTH) && defined(DATA_TYPE)</span></div><div class="line"><a name="l00455"></a><span class="lineno"> 455</span> </div><div class="line"><a name="l00478"></a><span class="lineno"> 478</span> __kernel <span class="keywordtype">void</span> depthwise_weights_reshape(</div><div class="line"><a name="l00479"></a><span class="lineno"> 479</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00480"></a><span class="lineno"> 480</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst)</div><div class="line"><a name="l00481"></a><span class="lineno"> 481</span> #ifdef HAS_BIAS</div><div class="line"><a name="l00482"></a><span class="lineno"> 482</span>  ,</div><div class="line"><a name="l00483"></a><span class="lineno"> 483</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases)</div><div class="line"><a name="l00484"></a><span class="lineno"> 484</span> #endif <span class="comment">/* HAS_BIAS */</span></div><div class="line"><a name="l00485"></a><span class="lineno"> 485</span> )</div><div class="line"><a name="l00486"></a><span class="lineno"> 486</span> {</div><div class="line"><a name="l00487"></a><span class="lineno"> 487</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> src = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(src);</div><div class="line"><a name="l00488"></a><span class="lineno"> 488</span> <span class="preprocessor">#ifdef HAS_BIAS</span></div><div class="line"><a name="l00489"></a><span class="lineno"> 489</span>  <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00490"></a><span class="lineno"> 490</span> <span class="preprocessor">#endif </span><span class="comment">/* HAS_BIAS */</span><span class="preprocessor"></span></div><div class="line"><a name="l00491"></a><span class="lineno"> 491</span> </div><div class="line"><a name="l00492"></a><span class="lineno"> 492</span>  __global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *input_ptr = (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)src.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>;</div><div class="line"><a name="l00493"></a><span class="lineno"> 493</span>  __global uchar *output_ptr = dst_ptr + dst_offset_first_element_in_bytes + get_global_id(1) * SRC_WIDTH * dst_stride_x + get_global_id(2) * dst_stride_y;</div><div class="line"><a name="l00494"></a><span class="lineno"> 494</span> </div><div class="line"><a name="l00495"></a><span class="lineno"> 495</span>  <span class="keywordflow">for</span>(<span class="keywordtype">int</span> i = 0; i < SRC_WIDTH; ++i, ++input_ptr)</div><div class="line"><a name="l00496"></a><span class="lineno"> 496</span>  {</div><div class="line"><a name="l00497"></a><span class="lineno"> 497</span>  *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + i * dst_stride_x)) = *input_ptr;</div><div class="line"><a name="l00498"></a><span class="lineno"> 498</span>  }</div><div class="line"><a name="l00499"></a><span class="lineno"> 499</span> </div><div class="line"><a name="l00500"></a><span class="lineno"> 500</span> <span class="preprocessor">#if defined(HAS_BIAS)</span></div><div class="line"><a name="l00501"></a><span class="lineno"> 501</span>  <span class="keywordflow">if</span>(get_global_id(1) == 0)</div><div class="line"><a name="l00502"></a><span class="lineno"> 502</span>  {</div><div class="line"><a name="l00503"></a><span class="lineno"> 503</span>  *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + SRC_WIDTH * get_global_size(1) * dst_stride_x)) = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(biases.<a class="code" href="struct_vector.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + get_global_id(2) * biases_stride_x));</div><div class="line"><a name="l00504"></a><span class="lineno"> 504</span>  }</div><div class="line"><a name="l00505"></a><span class="lineno"> 505</span> <span class="preprocessor">#endif // defined(HAS_BIAS)</span></div><div class="line"><a name="l00506"></a><span class="lineno"> 506</span> }</div><div class="line"><a name="l00507"></a><span class="lineno"> 507</span> <span class="preprocessor">#endif //defined(SRC_WIDTH) && defined(DATA_TYPE)</span></div><div class="line"><a name="l00508"></a><span class="lineno"> 508</span> </div><div class="line"><a name="l00509"></a><span class="lineno"> 509</span> <span class="preprocessor">#if defined(STRIDE_X) && defined(STRIDE_Y) && defined(PAD_LEFT) && defined(PAD_TOP) && defined(PAD_RIGHT) && defined(PAD_BOTTOM) && defined(KERNEL_WIDTH) && defined(KERNEL_HEIGHT) && defined(SRC_WIDTH) && defined(SRC_HEIGHT) && defined(DATA_TYPE) && defined(PAD_VALUE) && defined(DEPTH_MULTIPLIER)</span></div><div class="line"><a name="l00510"></a><span class="lineno"> 510</span> </div><div class="line"><a name="l00532"></a><span class="lineno"> 532</span> __kernel <span class="keywordtype">void</span> depthwise_im2col(<a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src), <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(dst))</div><div class="line"><a name="l00533"></a><span class="lineno"> 533</span> {</div><div class="line"><a name="l00534"></a><span class="lineno"> 534</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> dst = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(dst);</div><div class="line"><a name="l00535"></a><span class="lineno"> 535</span> </div><div class="line"><a name="l00536"></a><span class="lineno"> 536</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> src_pixel_linear = get_global_id(1) * STRIDE_X;</div><div class="line"><a name="l00537"></a><span class="lineno"> 537</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> full_length = SRC_WIDTH + PAD_LEFT + PAD_RIGHT;</div><div class="line"><a name="l00538"></a><span class="lineno"> 538</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> max_initial_x = STRIDE_X * (((full_length - KERNEL_WIDTH) / STRIDE_X) + 1);</div><div class="line"><a name="l00539"></a><span class="lineno"> 539</span> </div><div class="line"><a name="l00540"></a><span class="lineno"> 540</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> src_x = -PAD_LEFT + src_pixel_linear % max_initial_x;</div><div class="line"><a name="l00541"></a><span class="lineno"> 541</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> src_y = -PAD_TOP + src_pixel_linear / max_initial_x * STRIDE_Y;</div><div class="line"><a name="l00542"></a><span class="lineno"> 542</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> src_z = get_global_id(2) / DEPTH_MULTIPLIER;</div><div class="line"><a name="l00543"></a><span class="lineno"> 543</span> </div><div class="line"><a name="l00544"></a><span class="lineno"> 544</span>  __global uchar *input_ptr = src_ptr + src_offset_first_element_in_bytes + src_z * src_stride_z;</div><div class="line"><a name="l00545"></a><span class="lineno"> 545</span>  __global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *output_ptr = ((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>));</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="keywordflow">for</span>(<span class="keywordtype">int</span> y = src_y; y < src_y + KERNEL_HEIGHT; ++y)</div><div class="line"><a name="l00548"></a><span class="lineno"> 548</span>  {</div><div class="line"><a name="l00549"></a><span class="lineno"> 549</span>  <span class="keywordflow">for</span>(<span class="keywordtype">int</span> x = src_x; x < src_x + KERNEL_WIDTH; ++x, ++output_ptr)</div><div class="line"><a name="l00550"></a><span class="lineno"> 550</span>  {</div><div class="line"><a name="l00551"></a><span class="lineno"> 551</span>  <span class="keywordflow">if</span>(x < 0 || x >= SRC_WIDTH || y < 0 || y >= SRC_HEIGHT)</div><div class="line"><a name="l00552"></a><span class="lineno"> 552</span>  {</div><div class="line"><a name="l00553"></a><span class="lineno"> 553</span>  *output_ptr = PAD_VALUE;</div><div class="line"><a name="l00554"></a><span class="lineno"> 554</span>  }</div><div class="line"><a name="l00555"></a><span class="lineno"> 555</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00556"></a><span class="lineno"> 556</span>  {</div><div class="line"><a name="l00557"></a><span class="lineno"> 557</span>  *output_ptr = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + x * src_stride_x + y * src_stride_y));</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>  }</div><div class="line"><a name="l00560"></a><span class="lineno"> 560</span>  }</div><div class="line"><a name="l00561"></a><span class="lineno"> 561</span> <span class="preprocessor">#if defined(HAS_BIAS)</span></div><div class="line"><a name="l00562"></a><span class="lineno"> 562</span>  *output_ptr = (<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)(1);</div><div class="line"><a name="l00563"></a><span class="lineno"> 563</span> <span class="preprocessor">#endif // defined(HAS_BIAS)</span></div><div class="line"><a name="l00564"></a><span class="lineno"> 564</span> }</div><div class="line"><a name="l00565"></a><span class="lineno"> 565</span> </div><div class="line"><a name="l00566"></a><span class="lineno"> 566</span> <span class="preprocessor">#endif //defined(STRIDE_X) && defined(STRIDE_Y) && defined(PAD_LEFT) && defined(PAD_TOP) && defined(PAD_RIGHT) && defined(PAD_BOTTOM) && defined(KERNEL_WIDTH) && defined(KERNEL_HEIGHT) && defined(SRC_WIDTH) && defined(DATA_TYPE) && defined(PAD_VALUE) && defined(DEPTH_MULTIPLIER)</span></div><div class="line"><a name="l00567"></a><span class="lineno"> 567</span> </div><div class="line"><a name="l00568"></a><span class="lineno"> 568</span> <span class="preprocessor">#if defined(CONV_WIDTH) && defined(CONV_HEIGHT) && defined(DATA_TYPE)</span></div><div class="line"><a name="l00569"></a><span class="lineno"> 569</span> </div><div class="line"><a name="l00588"></a><span class="lineno"> 588</span> __kernel <span class="keywordtype">void</span> depthwise_vector_to_tensor(</div><div class="line"><a name="l00589"></a><span class="lineno"> 589</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(src),</div><div class="line"><a name="l00590"></a><span class="lineno"> 590</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(dst))</div><div class="line"><a name="l00591"></a><span class="lineno"> 591</span> {</div><div class="line"><a name="l00592"></a><span class="lineno"> 592</span>  <a class="code" href="struct_vector.xhtml">Vector</a> src = <a class="code" href="helpers_8h.xhtml#a527bfdf5eeb306f1cf01c4a8e29f38e0">CONVERT_TO_VECTOR_STRUCT</a>(src);</div><div class="line"><a name="l00593"></a><span class="lineno"> 593</span> </div><div class="line"><a name="l00594"></a><span class="lineno"> 594</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> patch_size = CONV_WIDTH * CONV_HEIGHT;</div><div class="line"><a name="l00595"></a><span class="lineno"> 595</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> id0 = get_global_id(0);</div><div class="line"><a name="l00596"></a><span class="lineno"> 596</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> z = id0 / patch_size;</div><div class="line"><a name="l00597"></a><span class="lineno"> 597</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> index2D = id0 - z * patch_size;</div><div class="line"><a name="l00598"></a><span class="lineno"> 598</span> </div><div class="line"><a name="l00599"></a><span class="lineno"> 599</span>  __global uchar *out_ptr = dst_ptr + dst_offset_first_element_in_bytes + index2D % CONV_WIDTH * dst_stride_x + index2D / CONV_WIDTH * dst_stride_y + z * dst_stride_z;</div><div class="line"><a name="l00600"></a><span class="lineno"> 600</span>  *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)out_ptr) = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)src.<a class="code" href="struct_vector.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l00601"></a><span class="lineno"> 601</span> }</div><div class="line"><a name="l00602"></a><span class="lineno"> 602</span> </div><div class="line"><a name="l00603"></a><span class="lineno"> 603</span> <span class="preprocessor">#endif //defined(CONV_WIDTH) && defined(CONV_HEIGHT) && defined(DATA_TYPE)</span></div><div class="line"><a name="l00604"></a><span class="lineno"> 604</span> </div><div class="line"><a name="l00605"></a><span class="lineno"> 605</span> <span class="preprocessor">#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(DEPTH_MULTIPLIER)</span></div><div class="line"><a name="l00606"></a><span class="lineno"> 606</span> <span class="preprocessor">#if defined(CONV_STRIDE_X)</span></div><div class="line"><a name="l00607"></a><span class="lineno"> 607</span> <span class="preprocessor">#if CONV_STRIDE_X == 1</span></div><div class="line"><a name="l00608"></a><span class="lineno"> 608</span> <span class="preprocessor">#define convolution1x3_f16 convolution1x3_stride_1_f16</span></div><div class="line"><a name="l00609"></a><span class="lineno"> 609</span> <span class="preprocessor">#elif CONV_STRIDE_X == 2</span></div><div class="line"><a name="l00610"></a><span class="lineno"> 610</span> <span class="preprocessor">#define convolution1x3_f16 convolution1x3_stride_2_f16</span></div><div class="line"><a name="l00611"></a><span class="lineno"> 611</span> <span class="preprocessor">#elif CONV_STRIDE_X == 3</span></div><div class="line"><a name="l00612"></a><span class="lineno"> 612</span> <span class="preprocessor">#define convolution1x3_f16 convolution1x3_stride_3_f16</span></div><div class="line"><a name="l00613"></a><span class="lineno"> 613</span> <span class="preprocessor">#else </span><span class="comment">/* CONV_STRIDE_X */</span><span class="preprocessor"></span></div><div class="line"><a name="l00614"></a><span class="lineno"> 614</span> <span class="preprocessor">#error "Stride not supported"</span></div><div class="line"><a name="l00615"></a><span class="lineno"> 615</span> <span class="preprocessor">#endif </span><span class="comment">/* CONV_STRIDE_X */</span><span class="preprocessor"></span></div><div class="line"><a name="l00616"></a><span class="lineno"> 616</span> </div><div class="line"><a name="l00626"></a><span class="lineno"> 626</span> <span class="keyword">inline</span> half4 convolution1x3_stride_1_f16(__global <span class="keyword">const</span> uchar *left_pixel,</div><div class="line"><a name="l00627"></a><span class="lineno"> 627</span>  <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> left_coeff,</div><div class="line"><a name="l00628"></a><span class="lineno"> 628</span>  <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> middle_coeff,</div><div class="line"><a name="l00629"></a><span class="lineno"> 629</span>  <span class="keyword">const</span> <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> right_coeff)</div><div class="line"><a name="l00630"></a><span class="lineno"> 630</span> {</div><div class="line"><a name="l00631"></a><span class="lineno"> 631</span>  half8 temp = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)left_pixel);</div><div class="line"><a name="l00632"></a><span class="lineno"> 632</span> </div><div class="line"><a name="l00633"></a><span class="lineno"> 633</span>  half4 left = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp.s0123, half4);</div><div class="line"><a name="l00634"></a><span class="lineno"> 634</span>  half4 middle = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp.s1234, half4);</div><div class="line"><a name="l00635"></a><span class="lineno"> 635</span>  half4 right = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp.s2345, half4);</div><div class="line"><a name="l00636"></a><span class="lineno"> 636</span> </div><div class="line"><a name="l00637"></a><span class="lineno"> 637</span>  <span class="keywordflow">return</span> left * (half4)left_coeff + middle * (half4)middle_coeff + right * (half4)right_coeff;</div><div class="line"><a name="l00638"></a><span class="lineno"> 638</span> }</div><div class="line"><a name="l00639"></a><span class="lineno"> 639</span> </div><div class="line"><a name="l00649"></a><span class="lineno"> 649</span> inline half4 convolution1x3_stride_2_f16(__global const uchar *left_pixel,</div><div class="line"><a name="l00650"></a><span class="lineno"> 650</span>  const <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> left_coeff,</div><div class="line"><a name="l00651"></a><span class="lineno"> 651</span>  const <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> middle_coeff,</div><div class="line"><a name="l00652"></a><span class="lineno"> 652</span>  const <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> right_coeff)</div><div class="line"><a name="l00653"></a><span class="lineno"> 653</span> {</div><div class="line"><a name="l00654"></a><span class="lineno"> 654</span>  half8 temp0 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)left_pixel);</div><div class="line"><a name="l00655"></a><span class="lineno"> 655</span>  <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> temp1 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(left_pixel + 8 * <span class="keyword">sizeof</span>(<a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a>)));</div><div class="line"><a name="l00656"></a><span class="lineno"> 656</span> </div><div class="line"><a name="l00657"></a><span class="lineno"> 657</span>  half4 left = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s0246, half4);</div><div class="line"><a name="l00658"></a><span class="lineno"> 658</span>  half4 middle = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s1357, half4);</div><div class="line"><a name="l00659"></a><span class="lineno"> 659</span>  half4 right = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>((half4)(temp0.s246, temp1), half4);</div><div class="line"><a name="l00660"></a><span class="lineno"> 660</span> </div><div class="line"><a name="l00661"></a><span class="lineno"> 661</span>  <span class="keywordflow">return</span> left * (half4)left_coeff + middle * (half4)middle_coeff + right * (half4)right_coeff;</div><div class="line"><a name="l00662"></a><span class="lineno"> 662</span> }</div><div class="line"><a name="l00663"></a><span class="lineno"> 663</span> </div><div class="line"><a name="l00673"></a><span class="lineno"> 673</span> inline half4 convolution1x3_stride_3_f16(__global const uchar *left_pixel,</div><div class="line"><a name="l00674"></a><span class="lineno"> 674</span>  const <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> left_coeff,</div><div class="line"><a name="l00675"></a><span class="lineno"> 675</span>  const <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> middle_coeff,</div><div class="line"><a name="l00676"></a><span class="lineno"> 676</span>  const <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> right_coeff)</div><div class="line"><a name="l00677"></a><span class="lineno"> 677</span> {</div><div class="line"><a name="l00678"></a><span class="lineno"> 678</span>  half16 temp0 = vload16(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)left_pixel);</div><div class="line"><a name="l00679"></a><span class="lineno"> 679</span> </div><div class="line"><a name="l00680"></a><span class="lineno"> 680</span>  half4 left = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s0369, half4);</div><div class="line"><a name="l00681"></a><span class="lineno"> 681</span>  half4 middle = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s147A, half4);</div><div class="line"><a name="l00682"></a><span class="lineno"> 682</span>  half4 right = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(temp0.s258B, half4);</div><div class="line"><a name="l00683"></a><span class="lineno"> 683</span> </div><div class="line"><a name="l00684"></a><span class="lineno"> 684</span>  <span class="keywordflow">return</span> left * (half4)left_coeff + middle * (half4)middle_coeff + right * (half4)right_coeff;</div><div class="line"><a name="l00685"></a><span class="lineno"> 685</span> }</div><div class="line"><a name="l00686"></a><span class="lineno"> 686</span> </div><div class="line"><a name="l00709"></a><span class="lineno"> 709</span> inline half4 convolution3x3_f16(</div><div class="line"><a name="l00710"></a><span class="lineno"> 710</span>  <a class="code" href="struct_image.xhtml">Image</a> *src,</div><div class="line"><a name="l00711"></a><span class="lineno"> 711</span>  const <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat0, const <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat1, const <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat2,</div><div class="line"><a name="l00712"></a><span class="lineno"> 712</span>  const <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat3, const <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat4, const <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat5,</div><div class="line"><a name="l00713"></a><span class="lineno"> 713</span>  const <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat6, const <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat7, const <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> mat8)</div><div class="line"><a name="l00714"></a><span class="lineno"> 714</span> {</div><div class="line"><a name="l00715"></a><span class="lineno"> 715</span>  half4 pixels;</div><div class="line"><a name="l00716"></a><span class="lineno"> 716</span> </div><div class="line"><a name="l00717"></a><span class="lineno"> 717</span>  pixels = convolution1x3_f16(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(src, 0, 0), mat0, mat1, mat2);</div><div class="line"><a name="l00718"></a><span class="lineno"> 718</span>  pixels += convolution1x3_f16(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(src, 0, 1), mat3, mat4, mat5);</div><div class="line"><a name="l00719"></a><span class="lineno"> 719</span>  pixels += convolution1x3_f16(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(src, 0, 2), mat6, mat7, mat8);</div><div class="line"><a name="l00720"></a><span class="lineno"> 720</span> </div><div class="line"><a name="l00721"></a><span class="lineno"> 721</span>  <span class="keywordflow">return</span> pixels;</div><div class="line"><a name="l00722"></a><span class="lineno"> 722</span> }</div><div class="line"><a name="l00723"></a><span class="lineno"> 723</span> </div><div class="line"><a name="l00724"></a><span class="lineno"> 724</span> <span class="preprocessor">#if defined(DEPTH_MULTIPLIER)</span></div><div class="line"><a name="l00725"></a><span class="lineno"> 725</span> </div><div class="line"><a name="l00757"></a><span class="lineno"> 757</span> __kernel <span class="keywordtype">void</span> depthwise_convolution_3x3_f16(</div><div class="line"><a name="l00758"></a><span class="lineno"> 758</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00759"></a><span class="lineno"> 759</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(dst),</div><div class="line"><a name="l00760"></a><span class="lineno"> 760</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(weights)</div><div class="line"><a name="l00761"></a><span class="lineno"> 761</span> #<span class="keywordflow">if</span> defined(HAS_BIAS)</div><div class="line"><a name="l00762"></a><span class="lineno"> 762</span>  ,</div><div class="line"><a name="l00763"></a><span class="lineno"> 763</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases)</div><div class="line"><a name="l00764"></a><span class="lineno"> 764</span> #endif <span class="comment">//defined(HAS_BIAS)</span></div><div class="line"><a name="l00765"></a><span class="lineno"> 765</span> )</div><div class="line"><a name="l00766"></a><span class="lineno"> 766</span> {</div><div class="line"><a name="l00767"></a><span class="lineno"> 767</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(src);</div><div class="line"><a name="l00768"></a><span class="lineno"> 768</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l00769"></a><span class="lineno"> 769</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> weights = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(weights);</div><div class="line"><a name="l00770"></a><span class="lineno"> 770</span> <span class="preprocessor">#if defined(HAS_BIAS)</span></div><div class="line"><a name="l00771"></a><span class="lineno"> 771</span>  <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00772"></a><span class="lineno"> 772</span> <span class="preprocessor">#endif //defined(HAS_BIAS)</span></div><div class="line"><a name="l00773"></a><span class="lineno"> 773</span> </div><div class="line"><a name="l00774"></a><span class="lineno"> 774</span>  src.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> -= (get_global_id(2) - get_global_id(2) / DEPTH_MULTIPLIER) * src_step_z;</div><div class="line"><a name="l00775"></a><span class="lineno"> 775</span> </div><div class="line"><a name="l00776"></a><span class="lineno"> 776</span>  uchar3 offset = (uchar3)(0, 1, 2) * (uchar3)weights_stride_y;</div><div class="line"><a name="l00777"></a><span class="lineno"> 777</span>  half3 weights_values0 = vload3(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + offset.s0));</div><div class="line"><a name="l00778"></a><span class="lineno"> 778</span>  half3 weights_values1 = vload3(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + offset.s1));</div><div class="line"><a name="l00779"></a><span class="lineno"> 779</span>  half3 weights_values2 = vload3(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + offset.s2));</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>  half4 pixels = convolution3x3_f16(&src, weights_values0.s0, weights_values0.s1, weights_values0.s2,</div><div class="line"><a name="l00782"></a><span class="lineno"> 782</span>  weights_values1.s0, weights_values1.s1, weights_values1.s2,</div><div class="line"><a name="l00783"></a><span class="lineno"> 783</span>  weights_values2.s0, weights_values2.s1, weights_values2.s2);</div><div class="line"><a name="l00784"></a><span class="lineno"> 784</span> <span class="preprocessor">#if defined(HAS_BIAS)</span></div><div class="line"><a name="l00785"></a><span class="lineno"> 785</span>  pixels += (half4)(*((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(biases.<a class="code" href="struct_vector.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + get_global_id(2) * biases_stride_x)));</div><div class="line"><a name="l00786"></a><span class="lineno"> 786</span> <span class="preprocessor">#endif //defined(HAS_BIAS)</span></div><div class="line"><a name="l00787"></a><span class="lineno"> 787</span> </div><div class="line"><a name="l00788"></a><span class="lineno"> 788</span>  vstore4(pixels, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l00789"></a><span class="lineno"> 789</span> }</div><div class="line"><a name="l00790"></a><span class="lineno"> 790</span> <span class="preprocessor">#endif // defined(DEPTH_MULTIPLIER)</span></div><div class="line"><a name="l00791"></a><span class="lineno"> 791</span> <span class="preprocessor">#endif // defined(CONV_STRIDE_X)</span></div><div class="line"><a name="l00792"></a><span class="lineno"> 792</span> </div><div class="line"><a name="l00825"></a><span class="lineno"> 825</span> __kernel <span class="keywordtype">void</span> depthwise_convolution_3x3_stridex1_stridey1_bifrost_f16(</div><div class="line"><a name="l00826"></a><span class="lineno"> 826</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00827"></a><span class="lineno"> 827</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(dst),</div><div class="line"><a name="l00828"></a><span class="lineno"> 828</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(weights)</div><div class="line"><a name="l00829"></a><span class="lineno"> 829</span> #<span class="keywordflow">if</span> defined(HAS_BIAS)</div><div class="line"><a name="l00830"></a><span class="lineno"> 830</span>  ,</div><div class="line"><a name="l00831"></a><span class="lineno"> 831</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases)</div><div class="line"><a name="l00832"></a><span class="lineno"> 832</span> #endif <span class="comment">//defined(HAS_BIAS)</span></div><div class="line"><a name="l00833"></a><span class="lineno"> 833</span> )</div><div class="line"><a name="l00834"></a><span class="lineno"> 834</span> {</div><div class="line"><a name="l00835"></a><span class="lineno"> 835</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(src);</div><div class="line"><a name="l00836"></a><span class="lineno"> 836</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l00837"></a><span class="lineno"> 837</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> weights = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(weights);</div><div class="line"><a name="l00838"></a><span class="lineno"> 838</span> </div><div class="line"><a name="l00839"></a><span class="lineno"> 839</span> <span class="preprocessor">#ifdef HAS_BIAS</span></div><div class="line"><a name="l00840"></a><span class="lineno"> 840</span>  <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00841"></a><span class="lineno"> 841</span> </div><div class="line"><a name="l00842"></a><span class="lineno"> 842</span>  <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> bias = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(<a class="code" href="helpers_8h.xhtml#a7e4940407322d6f0ccb8b6b86b856019">vector_offset</a>(&biases, get_global_id(2))));</div><div class="line"><a name="l00843"></a><span class="lineno"> 843</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(HAS_BIAS) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00844"></a><span class="lineno"> 844</span> </div><div class="line"><a name="l00845"></a><span class="lineno"> 845</span>  half4 pixels0 = 0.0f;</div><div class="line"><a name="l00846"></a><span class="lineno"> 846</span>  half4 pixels1 = 0.0f;</div><div class="line"><a name="l00847"></a><span class="lineno"> 847</span>  half4 pixels2 = 0.0f;</div><div class="line"><a name="l00848"></a><span class="lineno"> 848</span>  half4 pixels3 = 0.0f;</div><div class="line"><a name="l00849"></a><span class="lineno"> 849</span> </div><div class="line"><a name="l00850"></a><span class="lineno"> 850</span>  __global uchar *weights_addr = (__global uchar *)weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>;</div><div class="line"><a name="l00851"></a><span class="lineno"> 851</span>  __global uchar *src_addr = (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 0) - (get_global_id(2) - get_global_id(2) / DEPTH_MULTIPLIER) * src_step_z;</div><div class="line"><a name="l00852"></a><span class="lineno"> 852</span> </div><div class="line"><a name="l00853"></a><span class="lineno"> 853</span>  <span class="comment">// Load the weights</span></div><div class="line"><a name="l00854"></a><span class="lineno"> 854</span>  half3 weights_row0 = vload3(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(weights_addr + 0 * weights_stride_y));</div><div class="line"><a name="l00855"></a><span class="lineno"> 855</span>  half3 weights_row1 = vload3(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(weights_addr + 1 * weights_stride_y));</div><div class="line"><a name="l00856"></a><span class="lineno"> 856</span>  half3 weights_row2 = vload3(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(weights_addr + 2 * weights_stride_y));</div><div class="line"><a name="l00857"></a><span class="lineno"> 857</span> </div><div class="line"><a name="l00858"></a><span class="lineno"> 858</span>  <span class="comment">// Note: Since each work-item computes 4x4 elements, we need to load 6 rows from the input tensor</span></div><div class="line"><a name="l00859"></a><span class="lineno"> 859</span>  half8 src00 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 0 * src_stride_y)); <span class="comment">// Row0</span></div><div class="line"><a name="l00860"></a><span class="lineno"> 860</span>  half8 src10 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 1 * src_stride_y)); <span class="comment">// Row1</span></div><div class="line"><a name="l00861"></a><span class="lineno"> 861</span>  half8 src20 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 2 * src_stride_y)); <span class="comment">// Row2</span></div><div class="line"><a name="l00862"></a><span class="lineno"> 862</span>  half8 src30 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 3 * src_stride_y)); <span class="comment">// Row3</span></div><div class="line"><a name="l00863"></a><span class="lineno"> 863</span>  half8 src40 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 4 * src_stride_y)); <span class="comment">// Row4</span></div><div class="line"><a name="l00864"></a><span class="lineno"> 864</span>  half8 src50 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 5 * src_stride_y)); <span class="comment">// Row5</span></div><div class="line"><a name="l00865"></a><span class="lineno"> 865</span> </div><div class="line"><a name="l00866"></a><span class="lineno"> 866</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE1(pixels0, src00, weights_row0);</div><div class="line"><a name="l00867"></a><span class="lineno"> 867</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE1(pixels0, src10, weights_row1);</div><div class="line"><a name="l00868"></a><span class="lineno"> 868</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE1(pixels0, src20, weights_row2);</div><div class="line"><a name="l00869"></a><span class="lineno"> 869</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE1(pixels1, src10, weights_row0);</div><div class="line"><a name="l00870"></a><span class="lineno"> 870</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE1(pixels1, src20, weights_row1);</div><div class="line"><a name="l00871"></a><span class="lineno"> 871</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE1(pixels1, src30, weights_row2);</div><div class="line"><a name="l00872"></a><span class="lineno"> 872</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE1(pixels2, src20, weights_row0);</div><div class="line"><a name="l00873"></a><span class="lineno"> 873</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE1(pixels2, src30, weights_row1);</div><div class="line"><a name="l00874"></a><span class="lineno"> 874</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE1(pixels2, src40, weights_row2);</div><div class="line"><a name="l00875"></a><span class="lineno"> 875</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE1(pixels3, src30, weights_row0);</div><div class="line"><a name="l00876"></a><span class="lineno"> 876</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE1(pixels3, src40, weights_row1);</div><div class="line"><a name="l00877"></a><span class="lineno"> 877</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE1(pixels3, src50, weights_row2);</div><div class="line"><a name="l00878"></a><span class="lineno"> 878</span> </div><div class="line"><a name="l00879"></a><span class="lineno"> 879</span> <span class="preprocessor">#ifdef HAS_BIAS</span></div><div class="line"><a name="l00880"></a><span class="lineno"> 880</span>  pixels0 += (half4)bias;</div><div class="line"><a name="l00881"></a><span class="lineno"> 881</span>  pixels1 += (half4)bias;</div><div class="line"><a name="l00882"></a><span class="lineno"> 882</span>  pixels2 += (half4)bias;</div><div class="line"><a name="l00883"></a><span class="lineno"> 883</span>  pixels3 += (half4)bias;</div><div class="line"><a name="l00884"></a><span class="lineno"> 884</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(HAS_BIAS) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00885"></a><span class="lineno"> 885</span> </div><div class="line"><a name="l00886"></a><span class="lineno"> 886</span>  vstore4(pixels0, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 0 * dst_stride_y));</div><div class="line"><a name="l00887"></a><span class="lineno"> 887</span>  vstore4(pixels1, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 1 * dst_stride_y));</div><div class="line"><a name="l00888"></a><span class="lineno"> 888</span>  vstore4(pixels2, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 2 * dst_stride_y));</div><div class="line"><a name="l00889"></a><span class="lineno"> 889</span>  vstore4(pixels3, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 3 * dst_stride_y));</div><div class="line"><a name="l00890"></a><span class="lineno"> 890</span> }</div><div class="line"><a name="l00891"></a><span class="lineno"> 891</span> </div><div class="line"><a name="l00924"></a><span class="lineno"> 924</span> __kernel <span class="keywordtype">void</span> depthwise_convolution_3x3_stridex2_stridey2_bifrost_f16(</div><div class="line"><a name="l00925"></a><span class="lineno"> 925</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00926"></a><span class="lineno"> 926</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(dst),</div><div class="line"><a name="l00927"></a><span class="lineno"> 927</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(weights)</div><div class="line"><a name="l00928"></a><span class="lineno"> 928</span> #<span class="keywordflow">if</span> defined(HAS_BIAS)</div><div class="line"><a name="l00929"></a><span class="lineno"> 929</span>  ,</div><div class="line"><a name="l00930"></a><span class="lineno"> 930</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases)</div><div class="line"><a name="l00931"></a><span class="lineno"> 931</span> #endif <span class="comment">//defined(HAS_BIAS)</span></div><div class="line"><a name="l00932"></a><span class="lineno"> 932</span> )</div><div class="line"><a name="l00933"></a><span class="lineno"> 933</span> {</div><div class="line"><a name="l00934"></a><span class="lineno"> 934</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(src);</div><div class="line"><a name="l00935"></a><span class="lineno"> 935</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l00936"></a><span class="lineno"> 936</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> weights = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(weights);</div><div class="line"><a name="l00937"></a><span class="lineno"> 937</span> </div><div class="line"><a name="l00938"></a><span class="lineno"> 938</span> <span class="preprocessor">#ifdef HAS_BIAS</span></div><div class="line"><a name="l00939"></a><span class="lineno"> 939</span>  <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00940"></a><span class="lineno"> 940</span> </div><div class="line"><a name="l00941"></a><span class="lineno"> 941</span>  <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> bias = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(<a class="code" href="helpers_8h.xhtml#a7e4940407322d6f0ccb8b6b86b856019">vector_offset</a>(&biases, get_global_id(2))));</div><div class="line"><a name="l00942"></a><span class="lineno"> 942</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(HAS_BIAS) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00943"></a><span class="lineno"> 943</span> </div><div class="line"><a name="l00944"></a><span class="lineno"> 944</span>  half4 pixels0 = 0.0f;</div><div class="line"><a name="l00945"></a><span class="lineno"> 945</span>  half4 pixels1 = 0.0f;</div><div class="line"><a name="l00946"></a><span class="lineno"> 946</span> </div><div class="line"><a name="l00947"></a><span class="lineno"> 947</span>  __global uchar *weights_addr = (__global uchar *)weights.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>;</div><div class="line"><a name="l00948"></a><span class="lineno"> 948</span>  __global uchar *src_addr = (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 0) - (get_global_id(2) - get_global_id(2) / DEPTH_MULTIPLIER) * src_step_z;</div><div class="line"><a name="l00949"></a><span class="lineno"> 949</span> </div><div class="line"><a name="l00950"></a><span class="lineno"> 950</span>  <span class="comment">// Load the weights</span></div><div class="line"><a name="l00951"></a><span class="lineno"> 951</span>  half3 weights_row0 = vload3(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(weights_addr + 0 * weights_stride_y));</div><div class="line"><a name="l00952"></a><span class="lineno"> 952</span>  half3 weights_row1 = vload3(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(weights_addr + 1 * weights_stride_y));</div><div class="line"><a name="l00953"></a><span class="lineno"> 953</span>  half3 weights_row2 = vload3(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(weights_addr + 2 * weights_stride_y));</div><div class="line"><a name="l00954"></a><span class="lineno"> 954</span> </div><div class="line"><a name="l00955"></a><span class="lineno"> 955</span>  <span class="comment">// Note: Since each work-item computes 2x4 elements, we need to load 5 rows from the input tensor</span></div><div class="line"><a name="l00956"></a><span class="lineno"> 956</span>  half8 src00 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 0 * src_stride_y)); <span class="comment">// Row0</span></div><div class="line"><a name="l00957"></a><span class="lineno"> 957</span>  half2 src01 = vload2(4, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 0 * src_stride_y)); <span class="comment">// Row0</span></div><div class="line"><a name="l00958"></a><span class="lineno"> 958</span>  half8 src10 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 1 * src_stride_y)); <span class="comment">// Row1</span></div><div class="line"><a name="l00959"></a><span class="lineno"> 959</span>  half2 src11 = vload2(4, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 1 * src_stride_y)); <span class="comment">// Row1</span></div><div class="line"><a name="l00960"></a><span class="lineno"> 960</span>  half8 src20 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 2 * src_stride_y)); <span class="comment">// Row2</span></div><div class="line"><a name="l00961"></a><span class="lineno"> 961</span>  half2 src21 = vload2(4, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 2 * src_stride_y)); <span class="comment">// Row2</span></div><div class="line"><a name="l00962"></a><span class="lineno"> 962</span>  half8 src30 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 3 * src_stride_y)); <span class="comment">// Row3</span></div><div class="line"><a name="l00963"></a><span class="lineno"> 963</span>  half2 src31 = vload2(4, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 3 * src_stride_y)); <span class="comment">// Row3</span></div><div class="line"><a name="l00964"></a><span class="lineno"> 964</span>  half8 src40 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 4 * src_stride_y)); <span class="comment">// Row4</span></div><div class="line"><a name="l00965"></a><span class="lineno"> 965</span>  half2 src41 = vload2(4, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src_addr + 4 * src_stride_y)); <span class="comment">// Row4</span></div><div class="line"><a name="l00966"></a><span class="lineno"> 966</span> </div><div class="line"><a name="l00967"></a><span class="lineno"> 967</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE2(pixels0, src00, src01, weights_row0);</div><div class="line"><a name="l00968"></a><span class="lineno"> 968</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE2(pixels0, src10, src11, weights_row1);</div><div class="line"><a name="l00969"></a><span class="lineno"> 969</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE2(pixels0, src20, src21, weights_row2);</div><div class="line"><a name="l00970"></a><span class="lineno"> 970</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE2(pixels1, src20, src21, weights_row0);</div><div class="line"><a name="l00971"></a><span class="lineno"> 971</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE2(pixels1, src30, src31, weights_row1);</div><div class="line"><a name="l00972"></a><span class="lineno"> 972</span>  CONVOLUTION1x3_BIFROST4X1_STRIDE2(pixels1, src40, src41, weights_row2);</div><div class="line"><a name="l00973"></a><span class="lineno"> 973</span> </div><div class="line"><a name="l00974"></a><span class="lineno"> 974</span> <span class="preprocessor">#ifdef HAS_BIAS</span></div><div class="line"><a name="l00975"></a><span class="lineno"> 975</span>  pixels0 += (half4)bias;</div><div class="line"><a name="l00976"></a><span class="lineno"> 976</span>  pixels1 += (half4)bias;</div><div class="line"><a name="l00977"></a><span class="lineno"> 977</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(HAS_BIAS) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00978"></a><span class="lineno"> 978</span> </div><div class="line"><a name="l00979"></a><span class="lineno"> 979</span>  vstore4(pixels0, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 0 * dst_stride_y));</div><div class="line"><a name="l00980"></a><span class="lineno"> 980</span>  vstore4(pixels1, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + 1 * dst_stride_y));</div><div class="line"><a name="l00981"></a><span class="lineno"> 981</span> }</div><div class="line"><a name="l00982"></a><span class="lineno"> 982</span> <span class="preprocessor">#endif // defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(DEPTH_MULTIPLIER)</span></div><div class="ttc" id="struct_vector_xhtml"><div class="ttname"><a href="struct_vector.xhtml">Vector</a></div><div class="ttdoc">Structure to hold Vector information. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00134">helpers.h:134</a></div></div>
<div class="ttc" id="fixed__point_8h_xhtml_aa8d95ba04fc73845abc6045952cae5be"><div class="ttname"><a href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a></div><div class="ttdeci">#define CONVERT(x, type)</div><div class="ttdef"><b>Definition:</b> <a href="fixed__point_8h_source.xhtml#l00098">fixed_point.h:98</a></div></div>
-<div class="ttc" id="helpers_8h_xhtml_a31c8c760f08fb1a331b16b7c204321dc"><div class="ttname"><a href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a></div><div class="ttdeci">#define CONVERT_TO_TENSOR3D_STRUCT(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00117">helpers.h:117</a></div></div>
-<div class="ttc" id="helpers_8h_xhtml_a527bfdf5eeb306f1cf01c4a8e29f38e0"><div class="ttname"><a href="helpers_8h.xhtml#a527bfdf5eeb306f1cf01c4a8e29f38e0">CONVERT_TO_VECTOR_STRUCT</a></div><div class="ttdeci">#define CONVERT_TO_VECTOR_STRUCT(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00096">helpers.h:96</a></div></div>
+<div class="ttc" id="helpers_8h_xhtml_a31c8c760f08fb1a331b16b7c204321dc"><div class="ttname"><a href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a></div><div class="ttdeci">#define CONVERT_TO_TENSOR3D_STRUCT(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00119">helpers.h:119</a></div></div>
+<div class="ttc" id="helpers_8h_xhtml_a527bfdf5eeb306f1cf01c4a8e29f38e0"><div class="ttname"><a href="helpers_8h.xhtml#a527bfdf5eeb306f1cf01c4a8e29f38e0">CONVERT_TO_VECTOR_STRUCT</a></div><div class="ttdeci">#define CONVERT_TO_VECTOR_STRUCT(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00098">helpers.h:98</a></div></div>
<div class="ttc" id="convolution3x3_8cl_xhtml_afb8c72ce35c4a1f4a2588d6573e54aa1"><div class="ttname"><a href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a></div><div class="ttdeci">#define DATA_TYPE</div><div class="ttdef"><b>Definition:</b> <a href="convolution3x3_8cl_source.xhtml#l00027">convolution3x3.cl:27</a></div></div>
<div class="ttc" id="namespacearm__compute_xhtml_a73e2825fd61d349c5ca2f5313e3c8ea1"><div class="ttname"><a href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">arm_compute::half</a></div><div class="ttdeci">half_float::half half</div><div class="ttdoc">16-bit floating point type </div><div class="ttdef"><b>Definition:</b> <a href="arm__compute_2core_2_types_8h_source.xhtml#l00044">Types.h:44</a></div></div>
-<div class="ttc" id="helpers_8h_xhtml_a22f42fcf2077d951271df83b55c1a71a"><div class="ttname"><a href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a></div><div class="ttdeci">#define IMAGE_DECLARATION(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00066">helpers.h:66</a></div></div>
-<div class="ttc" id="struct_tensor3_d_xhtml"><div class="ttname"><a href="struct_tensor3_d.xhtml">Tensor3D</a></div><div class="ttdoc">Structure to hold 3D tensor information. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00149">helpers.h:149</a></div></div>
-<div class="ttc" id="depthwise__convolution_8cl_xhtml_a31cbdd6b28ee6331839432f69d87fdcb"><div class="ttname"><a href="depthwise__convolution_8cl.xhtml#a31cbdd6b28ee6331839432f69d87fdcb">CONVOLUTION1x3_BIFROST2X1_STRIDE2</a></div><div class="ttdeci">#define CONVOLUTION1x3_BIFROST2X1_STRIDE2(acc, src0, src1, weights_row0)</div><div class="ttdef"><b>Definition:</b> <a href="depthwise__convolution_8cl_source.xhtml#l00221">depthwise_convolution.cl:221</a></div></div>
-<div class="ttc" id="helpers_8h_xhtml_a009469e4d9b8fce3b6d5e97d2077827d"><div class="ttname"><a href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a></div><div class="ttdeci">__global uchar * offset(const Image *img, int x, int y)</div><div class="ttdoc">Get the pointer position of a Image. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00301">helpers.h:301</a></div></div>
-<div class="ttc" id="depthwise__convolution_8cl_xhtml_a2fbbf5b73260b66b310d4d857db5289e"><div class="ttname"><a href="depthwise__convolution_8cl.xhtml#a2fbbf5b73260b66b310d4d857db5289e">depthwise_convolution_3x3_stridex1_stridey1_bifrost</a></div><div class="ttdeci">__kernel void depthwise_convolution_3x3_stridex1_stridey1_bifrost(__global uchar *src_ptr, uint src_stride_x, uint src_step_x, uint src_stride_y, uint src_step_y, uint src_stride_z, uint src_step_z, uint src_offset_first_element_in_bytes, __global uchar *dst_ptr, uint dst_stride_x, uint dst_step_x, uint dst_stride_y, uint dst_step_y, uint dst_stride_z, uint dst_step_z, uint dst_offset_first_element_in_bytes, __global uchar *weights_ptr, uint weights_stride_x, uint weights_step_x, uint weights_stride_y, uint weights_step_y, uint weights_stride_z, uint weights_step_z, uint weights_offset_first_element_in_bytes, __global uchar *biases_ptr, uint biases_stride_x, uint biases_step_x, uint biases_offset_first_element_in_bytes)</div><div class="ttdoc">This OpenCL kernel is optimized for Bifrost architectures and computes the depthwise convolution 3x3 ...</div><div class="ttdef"><b>Definition:</b> <a href="depthwise__convolution_8cl_source.xhtml#l00263">depthwise_convolution.cl:263</a></div></div>
+<div class="ttc" id="helpers_8h_xhtml_a22f42fcf2077d951271df83b55c1a71a"><div class="ttname"><a href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a></div><div class="ttdeci">#define IMAGE_DECLARATION(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00068">helpers.h:68</a></div></div>
+<div class="ttc" id="struct_tensor3_d_xhtml"><div class="ttname"><a href="struct_tensor3_d.xhtml">Tensor3D</a></div><div class="ttdoc">Structure to hold 3D tensor information. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00151">helpers.h:151</a></div></div>
+<div class="ttc" id="helpers_8h_xhtml_a009469e4d9b8fce3b6d5e97d2077827d"><div class="ttname"><a href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a></div><div class="ttdeci">__global uchar * offset(const Image *img, int x, int y)</div><div class="ttdoc">Get the pointer position of a Image. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00303">helpers.h:303</a></div></div>
<div class="ttc" id="helpers_8h_xhtml"><div class="ttname"><a href="helpers_8h.xhtml">helpers.h</a></div></div>
<div class="ttc" id="convolution3x3_8cl_xhtml_afc5fefe72e66f0ae5191fd5b708fade9"><div class="ttname"><a href="convolution3x3_8cl.xhtml#afc5fefe72e66f0ae5191fd5b708fade9">convolution3x3</a></div><div class="ttdeci">short8 convolution3x3(Image *src, const short mat0, const short mat1, const short mat2, const short mat3, const short mat4, const short mat5, const short mat6, const short mat7, const short mat8, uint scale)</div><div class="ttdoc">Apply a 3x3 convolution matrix to a single channel U8 input image and return the result. </div><div class="ttdef"><b>Definition:</b> <a href="convolution3x3_8cl_source.xhtml#l00082">convolution3x3.cl:82</a></div></div>
-<div class="ttc" id="struct_vector_xhtml_acf52c23cbd7424606c10a606524e3e32"><div class="ttname"><a href="struct_vector.xhtml#acf52c23cbd7424606c10a606524e3e32">Vector::ptr</a></div><div class="ttdeci">__global uchar * ptr</div><div class="ttdoc">Pointer to the starting postion of the buffer. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00134">helpers.h:134</a></div></div>
-<div class="ttc" id="helpers_8h_xhtml_a40a6eb9f2a7712f08d6bb8ff6c9e6ca7"><div class="ttname"><a href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a></div><div class="ttdeci">#define VECTOR_DECLARATION(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00060">helpers.h:60</a></div></div>
+<div class="ttc" id="struct_vector_xhtml_acf52c23cbd7424606c10a606524e3e32"><div class="ttname"><a href="struct_vector.xhtml#acf52c23cbd7424606c10a606524e3e32">Vector::ptr</a></div><div class="ttdeci">__global uchar * ptr</div><div class="ttdoc">Pointer to the starting postion of the buffer. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00136">helpers.h:136</a></div></div>
+<div class="ttc" id="helpers_8h_xhtml_a40a6eb9f2a7712f08d6bb8ff6c9e6ca7"><div class="ttname"><a href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a></div><div class="ttdeci">#define VECTOR_DECLARATION(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00062">helpers.h:62</a></div></div>
<div class="ttc" id="namespacearm__compute_1_1test_1_1validation_xhtml_adbf67dcee294e673cf796f1ed8aeb6a4"><div class="ttname"><a href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">arm_compute::test::validation::dst</a></div><div class="ttdeci">CLTensor dst</div><div class="ttdef"><b>Definition:</b> <a href="_c_l_2_convolution_8cpp_source.xhtml#l00137">Convolution.cpp:137</a></div></div>
-<div class="ttc" id="struct_image_xhtml"><div class="ttname"><a href="struct_image.xhtml">Image</a></div><div class="ttdoc">Structure to hold Image information. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00140">helpers.h:140</a></div></div>
-<div class="ttc" id="helpers_8h_xhtml_a6b83038822d1ae7ab619b684ed3b7fc0"><div class="ttname"><a href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a></div><div class="ttdeci">#define TENSOR3D_DECLARATION(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00074">helpers.h:74</a></div></div>
-<div class="ttc" id="depthwise__convolution_8cl_xhtml_aa5ab4cf85fe56de3821727cfcef327d9"><div class="ttname"><a href="depthwise__convolution_8cl.xhtml#aa5ab4cf85fe56de3821727cfcef327d9">CONVOLUTION1x3_BIFROST2X1_STRIDE1</a></div><div class="ttdeci">#define CONVOLUTION1x3_BIFROST2X1_STRIDE1(acc, src0, weights_row0)</div><div class="ttdef"><b>Definition:</b> <a href="depthwise__convolution_8cl_source.xhtml#l00211">depthwise_convolution.cl:211</a></div></div>
-<div class="ttc" id="struct_image_xhtml_acf52c23cbd7424606c10a606524e3e32"><div class="ttname"><a href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">Image::ptr</a></div><div class="ttdeci">__global uchar * ptr</div><div class="ttdoc">Pointer to the starting postion of the buffer. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00142">helpers.h:142</a></div></div>
-<div class="ttc" id="depthwise__convolution_8cl_xhtml_a328ab548c536089e8238fb33de617bec"><div class="ttname"><a href="depthwise__convolution_8cl.xhtml#a328ab548c536089e8238fb33de617bec">depthwise_convolution_3x3_stridex2_stridey2_bifrost</a></div><div class="ttdeci">__kernel void depthwise_convolution_3x3_stridex2_stridey2_bifrost(__global uchar *src_ptr, uint src_stride_x, uint src_step_x, uint src_stride_y, uint src_step_y, uint src_stride_z, uint src_step_z, uint src_offset_first_element_in_bytes, __global uchar *dst_ptr, uint dst_stride_x, uint dst_step_x, uint dst_stride_y, uint dst_step_y, uint dst_stride_z, uint dst_step_z, uint dst_offset_first_element_in_bytes, __global uchar *weights_ptr, uint weights_stride_x, uint weights_step_x, uint weights_stride_y, uint weights_step_y, uint weights_stride_z, uint weights_step_z, uint weights_offset_first_element_in_bytes, __global uchar *biases_ptr, uint biases_stride_x, uint biases_step_x, uint biases_offset_first_element_in_bytes)</div><div class="ttdoc">This OpenCL kernel is optimized for Bifrost architectures and computes the depthwise convolution 3x3 ...</div><div class="ttdef"><b>Definition:</b> <a href="depthwise__convolution_8cl_source.xhtml#l00360">depthwise_convolution.cl:360</a></div></div>
-<div class="ttc" id="helpers_8h_xhtml_a64d779f80eeb923e0ab2313433f7b40b"><div class="ttname"><a href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a></div><div class="ttdeci">#define CONVERT_TO_VECTOR_STRUCT_NO_STEP(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00099">helpers.h:99</a></div></div>
-<div class="ttc" id="helpers_8h_xhtml_a7e4940407322d6f0ccb8b6b86b856019"><div class="ttname"><a href="helpers_8h.xhtml#a7e4940407322d6f0ccb8b6b86b856019">vector_offset</a></div><div class="ttdeci">__global const uchar * vector_offset(const Vector *vec, int x)</div><div class="ttdoc">Get the pointer position of a Vector. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00290">helpers.h:290</a></div></div>
-<div class="ttc" id="helpers_8h_xhtml_a541f8db866a0fa93ee67d58ea31a7d0c"><div class="ttname"><a href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a></div><div class="ttdeci">#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00114">helpers.h:114</a></div></div>
-<div class="ttc" id="struct_tensor3_d_xhtml_acf52c23cbd7424606c10a606524e3e32"><div class="ttname"><a href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">Tensor3D::ptr</a></div><div class="ttdeci">__global uchar * ptr</div><div class="ttdoc">Pointer to the starting postion of the buffer. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00151">helpers.h:151</a></div></div>
+<div class="ttc" id="struct_image_xhtml"><div class="ttname"><a href="struct_image.xhtml">Image</a></div><div class="ttdoc">Structure to hold Image information. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00142">helpers.h:142</a></div></div>
+<div class="ttc" id="helpers_8h_xhtml_a6b83038822d1ae7ab619b684ed3b7fc0"><div class="ttname"><a href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a></div><div class="ttdeci">#define TENSOR3D_DECLARATION(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00076">helpers.h:76</a></div></div>
+<div class="ttc" id="struct_image_xhtml_acf52c23cbd7424606c10a606524e3e32"><div class="ttname"><a href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">Image::ptr</a></div><div class="ttdeci">__global uchar * ptr</div><div class="ttdoc">Pointer to the starting postion of the buffer. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00144">helpers.h:144</a></div></div>
+<div class="ttc" id="helpers_8h_xhtml_a64d779f80eeb923e0ab2313433f7b40b"><div class="ttname"><a href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a></div><div class="ttdeci">#define CONVERT_TO_VECTOR_STRUCT_NO_STEP(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00101">helpers.h:101</a></div></div>
+<div class="ttc" id="helpers_8h_xhtml_a7e4940407322d6f0ccb8b6b86b856019"><div class="ttname"><a href="helpers_8h.xhtml#a7e4940407322d6f0ccb8b6b86b856019">vector_offset</a></div><div class="ttdeci">__global const uchar * vector_offset(const Vector *vec, int x)</div><div class="ttdoc">Get the pointer position of a Vector. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00292">helpers.h:292</a></div></div>
+<div class="ttc" id="helpers_8h_xhtml_a541f8db866a0fa93ee67d58ea31a7d0c"><div class="ttname"><a href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a></div><div class="ttdeci">#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00116">helpers.h:116</a></div></div>
+<div class="ttc" id="struct_tensor3_d_xhtml_acf52c23cbd7424606c10a606524e3e32"><div class="ttname"><a href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">Tensor3D::ptr</a></div><div class="ttdeci">__global uchar * ptr</div><div class="ttdoc">Pointer to the starting postion of the buffer. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00153">helpers.h:153</a></div></div>
<div class="ttc" id="namespacearm__compute_1_1test_1_1validation_xhtml_a6743f0a130e8311e6f5b1a23df102472"><div class="ttname"><a href="namespacearm__compute_1_1test_1_1validation.xhtml#a6743f0a130e8311e6f5b1a23df102472">arm_compute::test::validation::src</a></div><div class="ttdeci">convolution configure & src</div><div class="ttdef"><b>Definition:</b> <a href="_c_l_2_convolution_8cpp_source.xhtml#l00147">Convolution.cpp:147</a></div></div>
<div class="ttc" id="convolution3x3_8cl_xhtml_a92702074338198e81a46c3e309d9b04f"><div class="ttname"><a href="convolution3x3_8cl.xhtml#a92702074338198e81a46c3e309d9b04f">convolution1x3</a></div><div class="ttdeci">short8 convolution1x3(__global const uchar *left_pixel, const short left_coeff, const short middle_coeff, const short right_coeff)</div><div class="ttdoc">Compute a 1D horizontal convolution of size 3 for 8 bytes assuming the input is made of 1 channel of ...</div><div class="ttdef"><b>Definition:</b> <a href="convolution3x3_8cl_source.xhtml#l00043">convolution3x3.cl:43</a></div></div>
</div><!-- fragment --></div><!-- contents -->
@@ -150,7 +146,7 @@
<div id="nav-path" class="navpath"><!-- id is needed for treeview function! -->
<ul>
<li class="navelem"><a class="el" href="dir_68267d1309a1af8e8297ef4c3efbcdba.xhtml">src</a></li><li class="navelem"><a class="el" href="dir_aebb8dcc11953d78e620bbef0b9e2183.xhtml">core</a></li><li class="navelem"><a class="el" href="dir_8c278f79c760e5c5fbd911f9870614c1.xhtml">CL</a></li><li class="navelem"><a class="el" href="dir_25885286e9dad4fa105b7b25a8031bbf.xhtml">cl_kernels</a></li><li class="navelem"><a class="el" href="depthwise__convolution_8cl.xhtml">depthwise_convolution.cl</a></li>
- <li class="footer">Generated on Fri Mar 2 2018 12:37:53 for Compute Library by
+ <li class="footer">Generated on Wed May 23 2018 11:36:36 for Compute Library by
<a href="http://www.doxygen.org/index.html">
<img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.8.11 </li>
</ul>