<tr style="height: 56px;">
<td style="padding-left: 0.5em;">
<div id="projectname">Compute Library
-  <span id="projectnumber">17.12</span>
+  <span id="projectnumber">18.02</span>
</div>
</td>
</tr>
<div class="title">softmax_layer_quantized.cl</div> </div>
</div><!--header-->
<div class="contents">
-<a href="softmax__layer__quantized_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 ARM Limited.</span></div><div class="line"><a name="l00003"></a><span class="lineno"> 3</span> <span class="comment"> *</span></div><div class="line"><a name="l00004"></a><span class="lineno"> 4</span> <span class="comment"> * SPDX-License-Identifier: MIT</span></div><div class="line"><a name="l00005"></a><span class="lineno"> 5</span> <span class="comment"> *</span></div><div class="line"><a name="l00006"></a><span class="lineno"> 6</span> <span class="comment"> * Permission is hereby granted, free of charge, to any person obtaining a copy</span></div><div class="line"><a name="l00007"></a><span class="lineno"> 7</span> <span class="comment"> * of this software and associated documentation files (the "Software"), to</span></div><div class="line"><a name="l00008"></a><span class="lineno"> 8</span> <span class="comment"> * deal in the Software without restriction, including without limitation the</span></div><div class="line"><a name="l00009"></a><span class="lineno"> 9</span> <span class="comment"> * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or</span></div><div class="line"><a name="l00010"></a><span class="lineno"> 10</span> <span class="comment"> * sell copies of the Software, and to permit persons to whom the Software is</span></div><div class="line"><a name="l00011"></a><span class="lineno"> 11</span> <span class="comment"> * furnished to do so, subject to the following conditions:</span></div><div class="line"><a name="l00012"></a><span class="lineno"> 12</span> <span class="comment"> *</span></div><div class="line"><a name="l00013"></a><span class="lineno"> 13</span> <span class="comment"> * The above copyright notice and this permission notice shall be included in all</span></div><div class="line"><a name="l00014"></a><span class="lineno"> 14</span> <span class="comment"> * copies or substantial portions of the Software.</span></div><div class="line"><a name="l00015"></a><span class="lineno"> 15</span> <span class="comment"> *</span></div><div class="line"><a name="l00016"></a><span class="lineno"> 16</span> <span class="comment"> * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR</span></div><div class="line"><a name="l00017"></a><span class="lineno"> 17</span> <span class="comment"> * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,</span></div><div class="line"><a name="l00018"></a><span class="lineno"> 18</span> <span class="comment"> * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE</span></div><div class="line"><a name="l00019"></a><span class="lineno"> 19</span> <span class="comment"> * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER</span></div><div class="line"><a name="l00020"></a><span class="lineno"> 20</span> <span class="comment"> * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,</span></div><div class="line"><a name="l00021"></a><span class="lineno"> 21</span> <span class="comment"> * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE</span></div><div class="line"><a name="l00022"></a><span class="lineno"> 22</span> <span class="comment"> * SOFTWARE.</span></div><div class="line"><a name="l00023"></a><span class="lineno"> 23</span> <span class="comment"> */</span></div><div class="line"><a name="l00024"></a><span class="lineno"> 24</span> <span class="preprocessor">#include "<a class="code" href="asymm__helper_8h.xhtml">asymm_helper.h</a>"</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"><a class="line" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01"> 27</a></span> <span class="preprocessor">#define MAX_OP(x, y, type, size) max((x), (y))</span></div><div class="line"><a name="l00028"></a><span class="lineno"><a class="line" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88"> 28</a></span> <span class="preprocessor">#define ADD_OP(x, y, type, size) ((x) + (y))</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"><a class="line" href="softmax__layer__quantized_8cl.xhtml#a4701bcd6d0c0472d7d3e8017a24f2be6"> 30</a></span> __constant uchar16 <a class="code" href="softmax__layer__quantized_8cl.xhtml#a4701bcd6d0c0472d7d3e8017a24f2be6">type_min</a> = 0;</div><div class="line"><a name="l00031"></a><span class="lineno"><a class="line" href="softmax__layer__quantized_8cl.xhtml#a0712735973f172ac9efc7d48a31e47ad"> 31</a></span> __constant uint16 <a class="code" href="softmax__layer__quantized_8cl.xhtml#a0712735973f172ac9efc7d48a31e47ad">idx16</a> = (uint16)(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15);</div><div class="line"><a name="l00032"></a><span class="lineno"> 32</span> </div><div class="line"><a name="l00055"></a><span class="lineno"><a class="line" href="softmax__layer__quantized_8cl.xhtml#a79bd9092e1a23c550d4dc1aeb9b746ad"> 55</a></span> __kernel <span class="keywordtype">void</span> <a class="code" href="softmax__layer__quantized_8cl.xhtml#a79bd9092e1a23c550d4dc1aeb9b746ad">softmax_layer_max_quantized</a>(</div><div class="line"><a name="l00056"></a><span class="lineno"> 56</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00057"></a><span class="lineno"> 57</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="l00058"></a><span class="lineno"> 58</span>  uint width)</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>  <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="l00061"></a><span class="lineno"> 61</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="l00062"></a><span class="lineno"> 62</span> </div><div class="line"><a name="l00063"></a><span class="lineno"> 63</span>  <span class="comment">// Initialize local maximum</span></div><div class="line"><a name="l00064"></a><span class="lineno"> 64</span>  uchar16 max_val = 0;</div><div class="line"><a name="l00065"></a><span class="lineno"> 65</span> </div><div class="line"><a name="l00066"></a><span class="lineno"> 66</span>  <span class="comment">// Calculate max of row</span></div><div class="line"><a name="l00067"></a><span class="lineno"> 67</span>  <span class="keyword">const</span> uint width4 = width >> 4;</div><div class="line"><a name="l00068"></a><span class="lineno"> 68</span>  <span class="keywordflow">for</span>(uint i = 0; i < width4; i++)</div><div class="line"><a name="l00069"></a><span class="lineno"> 69</span>  {</div><div class="line"><a name="l00070"></a><span class="lineno"> 70</span>  uchar16 data = vload16(0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, i << 4, 0));</div><div class="line"><a name="l00071"></a><span class="lineno"> 71</span>  max_val = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(data, max_val, uchar, 16);</div><div class="line"><a name="l00072"></a><span class="lineno"> 72</span>  }</div><div class="line"><a name="l00073"></a><span class="lineno"> 73</span> </div><div class="line"><a name="l00074"></a><span class="lineno"> 74</span> <span class="preprocessor">#ifdef NON_MULTIPLE_OF_16</span></div><div class="line"><a name="l00075"></a><span class="lineno"> 75</span>  <span class="comment">// Handle non multiple of 16</span></div><div class="line"><a name="l00076"></a><span class="lineno"> 76</span>  uchar16 data = vload16(0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, width4 << 4, 0));</div><div class="line"><a name="l00077"></a><span class="lineno"> 77</span>  uchar16 widx = convert_uchar16(((uint16)(width4 << 4) + <a class="code" href="softmax__layer__quantized_8cl.xhtml#a0712735973f172ac9efc7d48a31e47ad">idx16</a>) < width);</div><div class="line"><a name="l00078"></a><span class="lineno"> 78</span>  max_val = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(max_val, select(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a4701bcd6d0c0472d7d3e8017a24f2be6">type_min</a>, data, widx), uchar, 16);</div><div class="line"><a name="l00079"></a><span class="lineno"> 79</span> <span class="preprocessor">#endif </span><span class="comment">/* NON_MULTIPLE_OF_16 */</span><span class="preprocessor"></span></div><div class="line"><a name="l00080"></a><span class="lineno"> 80</span> </div><div class="line"><a name="l00081"></a><span class="lineno"> 81</span>  <span class="comment">// Perform max reduction</span></div><div class="line"><a name="l00082"></a><span class="lineno"> 82</span>  max_val.s01234567 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(max_val.s01234567, max_val.s89ABCDEF, uchar, 8);</div><div class="line"><a name="l00083"></a><span class="lineno"> 83</span>  max_val.s0123 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(max_val.s0123, max_val.s4567, uchar, 4);</div><div class="line"><a name="l00084"></a><span class="lineno"> 84</span>  max_val.s01 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(max_val.s01, max_val.s23, uchar, 2);</div><div class="line"><a name="l00085"></a><span class="lineno"> 85</span>  max_val.s0 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(max_val.s0, max_val.s1, uchar, 1);</div><div class="line"><a name="l00086"></a><span class="lineno"> 86</span> </div><div class="line"><a name="l00087"></a><span class="lineno"> 87</span>  <span class="comment">// Store result</span></div><div class="line"><a name="l00088"></a><span class="lineno"> 88</span>  *((__global uchar *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) = max_val.s0;</div><div class="line"><a name="l00089"></a><span class="lineno"> 89</span> }</div><div class="line"><a name="l00090"></a><span class="lineno"> 90</span> </div><div class="line"><a name="l00091"></a><span class="lineno"> 91</span> <span class="preprocessor">#if defined(DIFF_MIN)</span></div><div class="line"><a name="l00092"></a><span class="lineno"> 92</span> </div><div class="line"><a name="l00093"></a><span class="lineno"> 93</span> int16 mult_by_quantized_multiplier(int16 data)</div><div class="line"><a name="l00094"></a><span class="lineno"> 94</span> {</div><div class="line"><a name="l00095"></a><span class="lineno"> 95</span> <span class="preprocessor">#if defined(INPUT_BETA_MULTIPLIER) && defined(INPUT_BETA_LEFT_SHIFT)</span></div><div class="line"><a name="l00096"></a><span class="lineno"> 96</span>  <span class="keywordflow">if</span>(INPUT_BETA_MULTIPLIER > 1)</div><div class="line"><a name="l00097"></a><span class="lineno"> 97</span>  {</div><div class="line"><a name="l00098"></a><span class="lineno"> 98</span>  <span class="keywordflow">return</span> <a class="code" href="asymm__helper_8h.xhtml#ab98c50e16a18dd439b2b7ecdd1ecd81c">asymm_mult</a>(data * (1 << INPUT_BETA_LEFT_SHIFT), INPUT_BETA_MULTIPLIER);</div><div class="line"><a name="l00099"></a><span class="lineno"> 99</span>  }</div><div class="line"><a name="l00100"></a><span class="lineno"> 100</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(INPUT_BETA_MULTIPLIER) && defined(INPUT_BETA_LEFT_SHIFT) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00101"></a><span class="lineno"> 101</span>  <span class="keywordflow">return</span> data;</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> </div><div class="line"><a name="l00145"></a><span class="lineno"> 145</span> __kernel <span class="keywordtype">void</span> softmax_layer_shift_exp_sum_quantized(</div><div class="line"><a name="l00146"></a><span class="lineno"> 146</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00147"></a><span class="lineno"> 147</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1fixed__point__arithmetic_1_1detail.xhtml#ad91bb73431b4de1f4946ed949d444849">max</a>),</div><div class="line"><a name="l00148"></a><span class="lineno"> 148</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="l00149"></a><span class="lineno"> 149</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="reduction__operation_8cl.xhtml#ab0df00f5333da51860deb93deb44a782">sum</a>),</div><div class="line"><a name="l00150"></a><span class="lineno"> 150</span>  uint width)</div><div class="line"><a name="l00151"></a><span class="lineno"> 151</span> {</div><div class="line"><a name="l00152"></a><span class="lineno"> 152</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="l00153"></a><span class="lineno"> 153</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="l00154"></a><span class="lineno"> 154</span>  <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="namespacearm__compute_1_1test_1_1fixed__point__arithmetic_1_1detail.xhtml#ad91bb73431b4de1f4946ed949d444849">max</a> = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(max);</div><div class="line"><a name="l00155"></a><span class="lineno"> 155</span>  <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="reduction__operation_8cl.xhtml#ab0df00f5333da51860deb93deb44a782">sum</a> = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(sum);</div><div class="line"><a name="l00156"></a><span class="lineno"> 156</span> </div><div class="line"><a name="l00157"></a><span class="lineno"> 157</span>  <span class="comment">// Load max value of 1D logits vector (row)</span></div><div class="line"><a name="l00158"></a><span class="lineno"> 158</span>  <span class="keywordtype">int</span> max_val = convert_int(*((__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&max, 0, 0)));</div><div class="line"><a name="l00159"></a><span class="lineno"> 159</span> </div><div class="line"><a name="l00160"></a><span class="lineno"> 160</span>  <span class="comment">// Set sum vector, Q(EXP_ACCUMULATION_INT_BITS)</span></div><div class="line"><a name="l00161"></a><span class="lineno"> 161</span>  int16 sum1D = 0;</div><div class="line"><a name="l00162"></a><span class="lineno"> 162</span> </div><div class="line"><a name="l00163"></a><span class="lineno"> 163</span>  <span class="comment">// Shift values, exp and sum</span></div><div class="line"><a name="l00164"></a><span class="lineno"> 164</span>  <span class="keyword">const</span> uint width4 = width >> 4;</div><div class="line"><a name="l00165"></a><span class="lineno"> 165</span>  <span class="keywordflow">for</span>(uint i = 0; i < width4; i++)</div><div class="line"><a name="l00166"></a><span class="lineno"> 166</span>  {</div><div class="line"><a name="l00167"></a><span class="lineno"> 167</span>  uchar16 data = vload16(0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, i << 4, 0));</div><div class="line"><a name="l00168"></a><span class="lineno"> 168</span>  int16 data_fp = convert_int16(data);</div><div class="line"><a name="l00169"></a><span class="lineno"> 169</span>  int16 data_diff = data_fp - max_val;</div><div class="line"><a name="l00170"></a><span class="lineno"> 170</span>  int16 data_diff_mult = mult_by_quantized_multiplier(data_diff);</div><div class="line"><a name="l00171"></a><span class="lineno"> 171</span>  data_fp = <a class="code" href="asymm__helper_8h.xhtml#ad1aa37a99125a65c4f4f3741011677e2">asymm_exp_on_negative_values</a>(data_diff_mult, SCALED_DIFF_INT_BITS);</div><div class="line"><a name="l00172"></a><span class="lineno"> 172</span>  data_fp = <a class="code" href="asymm__helper_8h.xhtml#a4496e3d044dda78318d27aa7f9c0e810">asymm_rescale</a>(data_fp, 0, EXP_ACCUMULATION_INT_BITS);</div><div class="line"><a name="l00173"></a><span class="lineno"> 173</span>  vstore16(data_diff, 0, (__global <span class="keywordtype">int</span> *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, i << 4, 0));</div><div class="line"><a name="l00174"></a><span class="lineno"> 174</span>  sum1D = sum1D + select(0, data_fp, data_diff >= (int16)(DIFF_MIN));</div><div class="line"><a name="l00175"></a><span class="lineno"> 175</span>  }</div><div class="line"><a name="l00176"></a><span class="lineno"> 176</span> </div><div class="line"><a name="l00177"></a><span class="lineno"> 177</span> <span class="preprocessor">#ifdef NON_MULTIPLE_OF_16</span></div><div class="line"><a name="l00178"></a><span class="lineno"> 178</span>  <span class="comment">// Handle non multiple of 16</span></div><div class="line"><a name="l00179"></a><span class="lineno"> 179</span>  uchar16 data = vload16(0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, width4 << 4, 0));</div><div class="line"><a name="l00180"></a><span class="lineno"> 180</span>  int16 data_fp = convert_int16(data);</div><div class="line"><a name="l00181"></a><span class="lineno"> 181</span>  int16 data_diff = data_fp - max_val;</div><div class="line"><a name="l00182"></a><span class="lineno"> 182</span>  int16 data_diff_mult = mult_by_quantized_multiplier(data_diff);</div><div class="line"><a name="l00183"></a><span class="lineno"> 183</span>  data_fp = <a class="code" href="asymm__helper_8h.xhtml#ad1aa37a99125a65c4f4f3741011677e2">asymm_exp_on_negative_values</a>(data_diff_mult, SCALED_DIFF_INT_BITS);</div><div class="line"><a name="l00184"></a><span class="lineno"> 184</span>  data_fp = <a class="code" href="asymm__helper_8h.xhtml#a4496e3d044dda78318d27aa7f9c0e810">asymm_rescale</a>(data_fp, 0, EXP_ACCUMULATION_INT_BITS);</div><div class="line"><a name="l00185"></a><span class="lineno"> 185</span>  int16 widx = convert_int16(((uint16)(width4 << 4) + <a class="code" href="softmax__layer__quantized_8cl.xhtml#a0712735973f172ac9efc7d48a31e47ad">idx16</a>) < width);</div><div class="line"><a name="l00186"></a><span class="lineno"> 186</span>  vstore16(data_diff, 0, (__global <span class="keywordtype">int</span> *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, width4 << 4, 0));</div><div class="line"><a name="l00187"></a><span class="lineno"> 187</span>  data_fp = select(0, data_fp, data_diff >= (int16)(DIFF_MIN));</div><div class="line"><a name="l00188"></a><span class="lineno"> 188</span>  sum1D = sum1D + select(0, data_fp, widx);</div><div class="line"><a name="l00189"></a><span class="lineno"> 189</span> <span class="preprocessor">#endif </span><span class="comment">/* NON_MULTIPLE_OF_16 */</span><span class="preprocessor"></span></div><div class="line"><a name="l00190"></a><span class="lineno"> 190</span> </div><div class="line"><a name="l00191"></a><span class="lineno"> 191</span>  <span class="comment">// Perform min/max reduction</span></div><div class="line"><a name="l00192"></a><span class="lineno"> 192</span>  sum1D.s01234567 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(sum1D.s01234567, sum1D.s89ABCDEF, <a class="code" href="fixed__point_8h.xhtml#a26a8ca855cd14b1867173d301baf6c4f">qs16</a>, 8);</div><div class="line"><a name="l00193"></a><span class="lineno"> 193</span>  sum1D.s0123 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(sum1D.s0123, sum1D.s4567, <a class="code" href="fixed__point_8h.xhtml#a26a8ca855cd14b1867173d301baf6c4f">qs16</a>, 4);</div><div class="line"><a name="l00194"></a><span class="lineno"> 194</span>  sum1D.s01 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(sum1D.s01, sum1D.s23, <a class="code" href="fixed__point_8h.xhtml#a26a8ca855cd14b1867173d301baf6c4f">qs16</a>, 2);</div><div class="line"><a name="l00195"></a><span class="lineno"> 195</span>  sum1D.s0 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(sum1D.s0, sum1D.s1, <a class="code" href="fixed__point_8h.xhtml#a26a8ca855cd14b1867173d301baf6c4f">qs16</a>, 1);</div><div class="line"><a name="l00196"></a><span class="lineno"> 196</span> </div><div class="line"><a name="l00197"></a><span class="lineno"> 197</span>  <span class="comment">// Calculate and store result</span></div><div class="line"><a name="l00198"></a><span class="lineno"> 198</span>  *((__global <span class="keywordtype">int</span> *)sum.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) = sum1D.s0;</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> </div><div class="line"><a name="l00232"></a><span class="lineno"> 232</span> __kernel <span class="keywordtype">void</span> softmax_layer_norm_quantized(</div><div class="line"><a name="l00233"></a><span class="lineno"> 233</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00234"></a><span class="lineno"> 234</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="reduction__operation_8cl.xhtml#ab0df00f5333da51860deb93deb44a782">sum</a>),</div><div class="line"><a name="l00235"></a><span class="lineno"> 235</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="l00236"></a><span class="lineno"> 236</span> {</div><div class="line"><a name="l00237"></a><span class="lineno"> 237</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="l00238"></a><span class="lineno"> 238</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="l00239"></a><span class="lineno"> 239</span>  <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="reduction__operation_8cl.xhtml#ab0df00f5333da51860deb93deb44a782">sum</a> = <a class="code" href="helpers_8h.xhtml#a071aa45af973feac43b14f62e54a6fce">CONVERT_TENSOR3D_TO_IMAGE_STRUCT_NO_STEP</a>(sum);</div><div class="line"><a name="l00240"></a><span class="lineno"> 240</span> </div><div class="line"><a name="l00241"></a><span class="lineno"> 241</span>  <span class="comment">// Load max value of 1D logits vector (row)</span></div><div class="line"><a name="l00242"></a><span class="lineno"> 242</span>  <span class="keywordtype">int</span> sum_val = *((__global <span class="keywordtype">int</span> *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&sum, 0, get_global_id(1)));</div><div class="line"><a name="l00243"></a><span class="lineno"> 243</span> </div><div class="line"><a name="l00244"></a><span class="lineno"> 244</span>  <span class="comment">// It will be better to calculate this in prev layer and pass here as parameter</span></div><div class="line"><a name="l00245"></a><span class="lineno"> 245</span>  uint sum_val_u = convert_uint(sum_val);</div><div class="line"><a name="l00246"></a><span class="lineno"> 246</span>  <span class="keywordtype">int</span> headroom_plus_one = <a class="code" href="namespacearm__compute_1_1test_1_1fixed__point__arithmetic_1_1detail.xhtml#aa43c67a536c661ec28398d06bc8ba5af">clz</a>(sum_val_u);</div><div class="line"><a name="l00247"></a><span class="lineno"> 247</span>  <span class="keywordtype">int</span> num_bits_over_unit = EXP_ACCUMULATION_INT_BITS - headroom_plus_one;</div><div class="line"><a name="l00248"></a><span class="lineno"> 248</span>  <span class="keywordtype">int</span> shifted_sum_minus_one_1 = convert_int((sum_val_u << headroom_plus_one) - (1u << 31));</div><div class="line"><a name="l00249"></a><span class="lineno"> 249</span>  int16 shifted_sum_minus_one = shifted_sum_minus_one_1;</div><div class="line"><a name="l00250"></a><span class="lineno"> 250</span>  int16 shifted_scale = <a class="code" href="asymm__helper_8h.xhtml#adae82358534f6f4a2bcbdd833413ccb8">asymm_one_over_one_plus_x_for_x_in_0_1</a>(shifted_sum_minus_one);</div><div class="line"><a name="l00251"></a><span class="lineno"> 251</span> </div><div class="line"><a name="l00252"></a><span class="lineno"> 252</span>  <span class="comment">// It was already calculated in prev layer, should be stored into tmp output and reused</span></div><div class="line"><a name="l00253"></a><span class="lineno"> 253</span>  int16 data_diff = vload16(0, (__global <span class="keywordtype">int</span> *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 0));</div><div class="line"><a name="l00254"></a><span class="lineno"> 254</span>  int16 data_diff_mult = mult_by_quantized_multiplier(data_diff);</div><div class="line"><a name="l00255"></a><span class="lineno"> 255</span>  int16 data = <a class="code" href="asymm__helper_8h.xhtml#ad1aa37a99125a65c4f4f3741011677e2">asymm_exp_on_negative_values</a>(data_diff_mult, SCALED_DIFF_INT_BITS);</div><div class="line"><a name="l00256"></a><span class="lineno"> 256</span> </div><div class="line"><a name="l00257"></a><span class="lineno"> 257</span>  data = <a class="code" href="asymm__helper_8h.xhtml#ab98c50e16a18dd439b2b7ecdd1ecd81c">asymm_mult</a>(shifted_scale, data);</div><div class="line"><a name="l00258"></a><span class="lineno"> 258</span>  data = <a class="code" href="asymm__helper_8h.xhtml#a766e0a81889256cfa161f7c08951c0b8">asymm_rounding_divide_by_pow2</a>(data, num_bits_over_unit + 31 - 8);</div><div class="line"><a name="l00259"></a><span class="lineno"> 259</span>  data = select(0, data, data_diff >= (int16)(DIFF_MIN));</div><div class="line"><a name="l00260"></a><span class="lineno"> 260</span>  vstore16(convert_uchar16_sat(data), 0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0));</div><div class="line"><a name="l00261"></a><span class="lineno"> 261</span> }</div><div class="line"><a name="l00262"></a><span class="lineno"> 262</span> </div><div class="line"><a name="l00263"></a><span class="lineno"> 263</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(DIFF_MIN) */</span><span class="preprocessor"></span></div><div class="ttc" id="helpers_8h_xhtml_a071aa45af973feac43b14f62e54a6fce"><div class="ttname"><a href="helpers_8h.xhtml#a071aa45af973feac43b14f62e54a6fce">CONVERT_TENSOR3D_TO_IMAGE_STRUCT_NO_STEP</a></div><div class="ttdeci">#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT_NO_STEP(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00111">helpers.h:111</a></div></div>
-<div class="ttc" id="asymm__helper_8h_xhtml_ad1aa37a99125a65c4f4f3741011677e2"><div class="ttname"><a href="asymm__helper_8h.xhtml#ad1aa37a99125a65c4f4f3741011677e2">asymm_exp_on_negative_values</a></div><div class="ttdeci">int16 asymm_exp_on_negative_values(int16 a, int k_integer_bits)</div><div class="ttdoc">Calculates for x < 0. </div><div class="ttdef"><b>Definition:</b> <a href="asymm__helper_8h_source.xhtml#l00200">asymm_helper.h:200</a></div></div>
-<div class="ttc" id="softmax__layer__quantized_8cl_xhtml_a4701bcd6d0c0472d7d3e8017a24f2be6"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#a4701bcd6d0c0472d7d3e8017a24f2be6">type_min</a></div><div class="ttdeci">__constant uchar16 type_min</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00030">softmax_layer_quantized.cl:30</a></div></div>
-<div class="ttc" id="asymm__helper_8h_xhtml_adae82358534f6f4a2bcbdd833413ccb8"><div class="ttname"><a href="asymm__helper_8h.xhtml#adae82358534f6f4a2bcbdd833413ccb8">asymm_one_over_one_plus_x_for_x_in_0_1</a></div><div class="ttdeci">int16 asymm_one_over_one_plus_x_for_x_in_0_1(int16 a)</div><div class="ttdoc">Calculates for x in (0, 1). </div><div class="ttdef"><b>Definition:</b> <a href="asymm__helper_8h_source.xhtml#l00243">asymm_helper.h:243</a></div></div>
-<div class="ttc" id="softmax__layer__quantized_8cl_xhtml_abaa48ad818c44e415fd3f9dd0f27bf01"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a></div><div class="ttdeci">#define MAX_OP(x, y, type, size)</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00027">softmax_layer_quantized.cl:27</a></div></div>
+<a href="softmax__layer__quantized_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> <span class="preprocessor">#include "<a class="code" href="helpers__asymm_8h.xhtml">helpers_asymm.h</a>"</span></div><div class="line"><a name="l00025"></a><span class="lineno"> 25</span> </div><div class="line"><a name="l00026"></a><span class="lineno"><a class="line" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01"> 26</a></span> <span class="preprocessor">#define MAX_OP(x, y, type, size) max((x), (y))</span></div><div class="line"><a name="l00027"></a><span class="lineno"><a class="line" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88"> 27</a></span> <span class="preprocessor">#define ADD_OP(x, y, type, size) ((x) + (y))</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="comment">/* Number of workitems in dimension 0. */</span></div><div class="line"><a name="l00030"></a><span class="lineno"> 30</span> <span class="preprocessor">#if !defined(GRID_SIZE)</span></div><div class="line"><a name="l00031"></a><span class="lineno"><a class="line" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f"> 31</a></span> <span class="preprocessor">#define GRID_SIZE 1</span></div><div class="line"><a name="l00032"></a><span class="lineno"> 32</span> <span class="preprocessor">#endif </span><span class="comment">/* !defined(GRID_SIZE) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00033"></a><span class="lineno"> 33</span> </div><div class="line"><a name="l00034"></a><span class="lineno"> 34</span> <span class="preprocessor">#if VECTOR_SIZE == 2</span></div><div class="line"><a name="l00035"></a><span class="lineno"> 35</span> __constant uint2 <a class="code" href="softmax__layer__quantized_8cl.xhtml#aa1dd94b8d98f1c6d790bdf0fc5de29e9">idx__</a> = (uint2)(0, 1);</div><div class="line"><a name="l00036"></a><span class="lineno"> 36</span> <span class="preprocessor">#define asymm_mult(a, b) ASYMM_MULT(a, b, 2)</span></div><div class="line"><a name="l00037"></a><span class="lineno"> 37</span> <span class="preprocessor">#define asymm_exp_on_negative_values(a, k_integer_bits) ASYMM_EXP_ON_NEGATIVE_VALUES(a, k_integer_bits, 2)</span></div><div class="line"><a name="l00038"></a><span class="lineno"> 38</span> <span class="preprocessor">#define asymm_rescale(value, src_integer_bits, dst_integer_bits) ASYMM_RESCALE(value, src_integer_bits, dst_integer_bits, 2)</span></div><div class="line"><a name="l00039"></a><span class="lineno"> 39</span> </div><div class="line"><a name="l00040"></a><span class="lineno"> 40</span> <span class="preprocessor">#elif VECTOR_SIZE == 4</span></div><div class="line"><a name="l00041"></a><span class="lineno"> 41</span> __constant uint4 <a class="code" href="softmax__layer__quantized_8cl.xhtml#aa1dd94b8d98f1c6d790bdf0fc5de29e9">idx__</a> = (uint4)(0, 1, 2, 3);</div><div class="line"><a name="l00042"></a><span class="lineno"> 42</span> <span class="preprocessor">#define asymm_mult(a, b) ASYMM_MULT(a, b, 4)</span></div><div class="line"><a name="l00043"></a><span class="lineno"> 43</span> <span class="preprocessor">#define asymm_exp_on_negative_values(a, k_integer_bits) ASYMM_EXP_ON_NEGATIVE_VALUES(a, k_integer_bits, 4)</span></div><div class="line"><a name="l00044"></a><span class="lineno"> 44</span> <span class="preprocessor">#define asymm_rescale(value, src_integer_bits, dst_integer_bits) ASYMM_RESCALE(value, src_integer_bits, dst_integer_bits, 4)</span></div><div class="line"><a name="l00045"></a><span class="lineno"> 45</span> </div><div class="line"><a name="l00046"></a><span class="lineno"> 46</span> <span class="preprocessor">#elif VECTOR_SIZE == 8</span></div><div class="line"><a name="l00047"></a><span class="lineno"> 47</span> __constant uint8 idx__ = (uint8)(0, 1, 2, 3, 4, 5, 6, 7);</div><div class="line"><a name="l00048"></a><span class="lineno"> 48</span> <span class="preprocessor">#define asymm_mult(a, b) ASYMM_MULT(a, b, 8)</span></div><div class="line"><a name="l00049"></a><span class="lineno"> 49</span> <span class="preprocessor">#define asymm_exp_on_negative_values(a, k_integer_bits) ASYMM_EXP_ON_NEGATIVE_VALUES(a, k_integer_bits, 8)</span></div><div class="line"><a name="l00050"></a><span class="lineno"> 50</span> <span class="preprocessor">#define asymm_rescale(value, src_integer_bits, dst_integer_bits) ASYMM_RESCALE(value, src_integer_bits, dst_integer_bits, 8)</span></div><div class="line"><a name="l00051"></a><span class="lineno"> 51</span> </div><div class="line"><a name="l00052"></a><span class="lineno"> 52</span> <span class="preprocessor">#else </span><span class="comment">/* VECTOR_SIZE DEFAULT */</span><span class="preprocessor"></span></div><div class="line"><a name="l00053"></a><span class="lineno"><a class="line" href="softmax__layer__quantized_8cl.xhtml#a7c78836761fa3b5b124efea237dac70f"> 53</a></span> <span class="preprocessor">#define VECTOR_SIZE 16</span></div><div class="line"><a name="l00054"></a><span class="lineno"><a class="line" href="softmax__layer__quantized_8cl.xhtml#a372393c380805985b813dbb16d589a64"> 54</a></span> <span class="preprocessor">#define LOG_VECTOR_SIZE 4</span></div><div class="line"><a name="l00055"></a><span class="lineno"><a class="line" href="softmax__layer__quantized_8cl.xhtml#aa1dd94b8d98f1c6d790bdf0fc5de29e9"> 55</a></span> __constant uint16 idx__ = (uint16)(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15);</div><div class="line"><a name="l00056"></a><span class="lineno"><a class="line" href="softmax__layer__quantized_8cl.xhtml#a525a42d38133b1051b8924b456add4a1"> 56</a></span> <span class="preprocessor">#define asymm_mult(a, b) ASYMM_MULT(a, b, 16)</span></div><div class="line"><a name="l00057"></a><span class="lineno"><a class="line" href="softmax__layer__quantized_8cl.xhtml#a54aedfa17c5ac2567107d5f488b0f4af"> 57</a></span> <span class="preprocessor">#define asymm_exp_on_negative_values(a, k_integer_bits) ASYMM_EXP_ON_NEGATIVE_VALUES(a, k_integer_bits, 16)</span></div><div class="line"><a name="l00058"></a><span class="lineno"><a class="line" href="softmax__layer__quantized_8cl.xhtml#ad57ea340cdcfeb2e1375b70c3ae59bae"> 58</a></span> <span class="preprocessor">#define asymm_rescale(value, src_integer_bits, dst_integer_bits) ASYMM_RESCALE(value, src_integer_bits, dst_integer_bits, 16)</span></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="preprocessor">#endif </span><span class="comment">/* VECTOR_SIZE END */</span><span class="preprocessor"></span></div><div class="line"><a name="l00061"></a><span class="lineno"> 61</span> </div><div class="line"><a name="l00062"></a><span class="lineno"><a class="line" href="softmax__layer__quantized_8cl.xhtml#af5987b09a234231612b2b1eded343025"> 62</a></span> <span class="preprocessor">#define VEC_UCHAR VEC_DATA_TYPE(uchar, VECTOR_SIZE)</span></div><div class="line"><a name="l00063"></a><span class="lineno"><a class="line" href="softmax__layer__quantized_8cl.xhtml#a16110bd2b92003141dbaf8a44498ff82"> 63</a></span> <span class="preprocessor">#define VEC_UINT VEC_DATA_TYPE(uint, VECTOR_SIZE)</span></div><div class="line"><a name="l00064"></a><span class="lineno"><a class="line" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd"> 64</a></span> <span class="preprocessor">#define VEC_INT VEC_DATA_TYPE(int, VECTOR_SIZE)</span></div><div class="line"><a name="l00065"></a><span class="lineno"> 65</span> </div><div class="line"><a name="l00066"></a><span class="lineno"> 66</span> <span class="preprocessor">#if defined(DIFF_MIN)</span></div><div class="line"><a name="l00067"></a><span class="lineno"> 67</span> </div><div class="line"><a name="l00068"></a><span class="lineno"> 68</span> <a class="code" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a> mult_by_quantized_multiplier_serial(<a class="code" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a> data)</div><div class="line"><a name="l00069"></a><span class="lineno"> 69</span> {</div><div class="line"><a name="l00070"></a><span class="lineno"> 70</span> <span class="preprocessor">#if defined(INPUT_BETA_MULTIPLIER) && defined(INPUT_BETA_LEFT_SHIFT)</span></div><div class="line"><a name="l00071"></a><span class="lineno"> 71</span>  <span class="keywordflow">if</span>(INPUT_BETA_MULTIPLIER > 1)</div><div class="line"><a name="l00072"></a><span class="lineno"> 72</span>  {</div><div class="line"><a name="l00073"></a><span class="lineno"> 73</span>  <span class="keywordflow">return</span> <a class="code" href="softmax__layer__quantized_8cl.xhtml#a525a42d38133b1051b8924b456add4a1">asymm_mult</a>(data * (1 << INPUT_BETA_LEFT_SHIFT), INPUT_BETA_MULTIPLIER);</div><div class="line"><a name="l00074"></a><span class="lineno"> 74</span>  }</div><div class="line"><a name="l00075"></a><span class="lineno"> 75</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(INPUT_BETA_MULTIPLIER) && defined(INPUT_BETA_LEFT_SHIFT) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00076"></a><span class="lineno"> 76</span>  <span class="keywordflow">return</span> data;</div><div class="line"><a name="l00077"></a><span class="lineno"> 77</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> int4 mult_by_quantized_multiplier_parallel(int4 data)</div><div class="line"><a name="l00080"></a><span class="lineno"> 80</span> {</div><div class="line"><a name="l00081"></a><span class="lineno"> 81</span> <span class="preprocessor">#if defined(INPUT_BETA_MULTIPLIER) && defined(INPUT_BETA_LEFT_SHIFT)</span></div><div class="line"><a name="l00082"></a><span class="lineno"> 82</span>  <span class="keywordflow">if</span>(INPUT_BETA_MULTIPLIER > 1)</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> <a class="code" href="helpers__asymm_8h.xhtml#a5483aefd5e07244661178bfd3f434448">ASYMM_MULT</a>(data * (1 << INPUT_BETA_LEFT_SHIFT), INPUT_BETA_MULTIPLIER, 4);</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> <span class="preprocessor">#endif </span><span class="comment">/* defined(INPUT_BETA_MULTIPLIER) && defined(INPUT_BETA_LEFT_SHIFT) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00087"></a><span class="lineno"> 87</span>  <span class="keywordflow">return</span> data;</div><div class="line"><a name="l00088"></a><span class="lineno"> 88</span> }</div><div class="line"><a name="l00089"></a><span class="lineno"> 89</span> </div><div class="line"><a name="l00131"></a><span class="lineno"> 131</span> __kernel <span class="keywordtype">void</span> softmax_layer_max_shift_exp_sum_quantized_serial(</div><div class="line"><a name="l00132"></a><span class="lineno"> 132</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a6743f0a130e8311e6f5b1a23df102472">src</a>),</div><div class="line"><a name="l00133"></a><span class="lineno"> 133</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(maxo),</div><div class="line"><a name="l00134"></a><span class="lineno"> 134</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="l00135"></a><span class="lineno"> 135</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="reduction__operation_8cl.xhtml#ab0df00f5333da51860deb93deb44a782">sum</a>),</div><div class="line"><a name="l00136"></a><span class="lineno"> 136</span>  uint width)</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>  <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a6743f0a130e8311e6f5b1a23df102472">src</a> = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(src);</div><div class="line"><a name="l00139"></a><span class="lineno"> 139</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="l00140"></a><span class="lineno"> 140</span>  <a class="code" href="struct_image.xhtml">Image</a> maxo = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(maxo);</div><div class="line"><a name="l00141"></a><span class="lineno"> 141</span>  <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="reduction__operation_8cl.xhtml#ab0df00f5333da51860deb93deb44a782">sum</a> = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(sum);</div><div class="line"><a name="l00142"></a><span class="lineno"> 142</span> </div><div class="line"><a name="l00143"></a><span class="lineno"> 143</span>  <a class="code" href="softmax__layer__quantized_8cl.xhtml#af5987b09a234231612b2b1eded343025">VEC_UCHAR</a> max_val_vec = 0;</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="comment">// Calculate max of row</span></div><div class="line"><a name="l00146"></a><span class="lineno"> 146</span>  <span class="keyword">const</span> uint width4 = width >> <a class="code" href="softmax__layer__quantized_8cl.xhtml#a372393c380805985b813dbb16d589a64">LOG_VECTOR_SIZE</a>;</div><div class="line"><a name="l00147"></a><span class="lineno"> 147</span>  <span class="keywordflow">for</span>(uint i = 0; i < width4; i++)</div><div class="line"><a name="l00148"></a><span class="lineno"> 148</span>  {</div><div class="line"><a name="l00149"></a><span class="lineno"> 149</span>  <a class="code" href="softmax__layer__quantized_8cl.xhtml#af5987b09a234231612b2b1eded343025">VEC_UCHAR</a> data = <a class="code" href="helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a7c78836761fa3b5b124efea237dac70f">VECTOR_SIZE</a>)(0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, i << LOG_VECTOR_SIZE, 0));</div><div class="line"><a name="l00150"></a><span class="lineno"> 150</span>  max_val_vec = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(data, max_val_vec, uchar, 16);</div><div class="line"><a name="l00151"></a><span class="lineno"> 151</span>  }</div><div class="line"><a name="l00152"></a><span class="lineno"> 152</span> </div><div class="line"><a name="l00153"></a><span class="lineno"> 153</span> <span class="preprocessor">#ifdef NON_MULTIPLE_OF_VECTOR_SIZE</span></div><div class="line"><a name="l00154"></a><span class="lineno"> 154</span>  <span class="comment">// Handle non multiple of 16</span></div><div class="line"><a name="l00155"></a><span class="lineno"> 155</span>  <a class="code" href="softmax__layer__quantized_8cl.xhtml#af5987b09a234231612b2b1eded343025">VEC_UCHAR</a> uchar_min = (<a class="code" href="softmax__layer__quantized_8cl.xhtml#af5987b09a234231612b2b1eded343025">VEC_UCHAR</a>)0;</div><div class="line"><a name="l00156"></a><span class="lineno"> 156</span>  <a class="code" href="softmax__layer__quantized_8cl.xhtml#af5987b09a234231612b2b1eded343025">VEC_UCHAR</a> data = <a class="code" href="helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a7c78836761fa3b5b124efea237dac70f">VECTOR_SIZE</a>)(0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, width4 << LOG_VECTOR_SIZE, 0));</div><div class="line"><a name="l00157"></a><span class="lineno"> 157</span>  <a class="code" href="softmax__layer__quantized_8cl.xhtml#af5987b09a234231612b2b1eded343025">VEC_UCHAR</a> widx = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(((<a class="code" href="softmax__layer__quantized_8cl.xhtml#a16110bd2b92003141dbaf8a44498ff82">VEC_UINT</a>)(width4 << LOG_VECTOR_SIZE) + idx__) < width, <a class="code" href="softmax__layer__quantized_8cl.xhtml#af5987b09a234231612b2b1eded343025">VEC_UCHAR</a>);</div><div class="line"><a name="l00158"></a><span class="lineno"> 158</span>  max_val_vec = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(max_val_vec, select(uchar_min, data, widx), uchar, 16);</div><div class="line"><a name="l00159"></a><span class="lineno"> 159</span> <span class="preprocessor">#endif </span><span class="comment">/* NON_MULTIPLE_OF_VECTOR_SIZE */</span><span class="preprocessor"></span></div><div class="line"><a name="l00160"></a><span class="lineno"> 160</span> </div><div class="line"><a name="l00161"></a><span class="lineno"> 161</span>  <span class="comment">// Perform max reduction</span></div><div class="line"><a name="l00162"></a><span class="lineno"> 162</span> <span class="preprocessor">#if VECTOR_SIZE == 16</span></div><div class="line"><a name="l00163"></a><span class="lineno"> 163</span>  max_val_vec.s01234567 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(max_val_vec.s01234567, max_val_vec.s89ABCDEF, uchar, 8);</div><div class="line"><a name="l00164"></a><span class="lineno"> 164</span> <span class="preprocessor">#endif </span><span class="comment">/* VECTOR SIZE 16 END */</span><span class="preprocessor"></span></div><div class="line"><a name="l00165"></a><span class="lineno"> 165</span> <span class="preprocessor">#if VECTOR_SIZE >= 8</span></div><div class="line"><a name="l00166"></a><span class="lineno"> 166</span>  max_val_vec.s0123 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(max_val_vec.s0123, max_val_vec.s4567, uchar, 4);</div><div class="line"><a name="l00167"></a><span class="lineno"> 167</span> <span class="preprocessor">#endif </span><span class="comment">/* VECTOR SIZE 8 END */</span><span class="preprocessor"></span></div><div class="line"><a name="l00168"></a><span class="lineno"> 168</span> <span class="preprocessor">#if VECTOR_SIZE >= 4</span></div><div class="line"><a name="l00169"></a><span class="lineno"> 169</span>  max_val_vec.s01 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(max_val_vec.s01, max_val_vec.s23, uchar, 2);</div><div class="line"><a name="l00170"></a><span class="lineno"> 170</span> <span class="preprocessor">#endif </span><span class="comment">/* VECTOR SIZE 4 END */</span><span class="preprocessor"></span></div><div class="line"><a name="l00171"></a><span class="lineno"> 171</span>  max_val_vec.s0 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(max_val_vec.s0, max_val_vec.s1, uchar, 1);</div><div class="line"><a name="l00172"></a><span class="lineno"> 172</span> </div><div class="line"><a name="l00173"></a><span class="lineno"> 173</span>  <span class="comment">// Store result</span></div><div class="line"><a name="l00174"></a><span class="lineno"> 174</span>  *((__global uchar *)maxo.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) = max_val_vec.s0;</div><div class="line"><a name="l00175"></a><span class="lineno"> 175</span> </div><div class="line"><a name="l00176"></a><span class="lineno"> 176</span>  <span class="comment">// Second part</span></div><div class="line"><a name="l00177"></a><span class="lineno"> 177</span> </div><div class="line"><a name="l00178"></a><span class="lineno"> 178</span>  <span class="comment">// Load max value of 1D logits vector (row)</span></div><div class="line"><a name="l00179"></a><span class="lineno"> 179</span>  <span class="keywordtype">int</span> max_val = convert_int(*((__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&maxo, 0, 0)));</div><div class="line"><a name="l00180"></a><span class="lineno"> 180</span> </div><div class="line"><a name="l00181"></a><span class="lineno"> 181</span>  <span class="comment">// Set sum vector, Q(EXP_ACCUMULATION_INT_BITS)</span></div><div class="line"><a name="l00182"></a><span class="lineno"> 182</span>  <a class="code" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a> sum1D = 0;</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>  <span class="comment">// Shift values, exp and sum</span></div><div class="line"><a name="l00185"></a><span class="lineno"> 185</span>  <span class="keywordflow">for</span>(uint i = 0; i < width4; i++)</div><div class="line"><a name="l00186"></a><span class="lineno"> 186</span>  {</div><div class="line"><a name="l00187"></a><span class="lineno"> 187</span>  <a class="code" href="softmax__layer__quantized_8cl.xhtml#af5987b09a234231612b2b1eded343025">VEC_UCHAR</a> data = <a class="code" href="helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a7c78836761fa3b5b124efea237dac70f">VECTOR_SIZE</a>)(0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, i << LOG_VECTOR_SIZE, 0));</div><div class="line"><a name="l00188"></a><span class="lineno"> 188</span>  <a class="code" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a> data_fp = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(data, <a class="code" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a>);</div><div class="line"><a name="l00189"></a><span class="lineno"> 189</span>  <a class="code" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a> data_diff = data_fp - max_val;</div><div class="line"><a name="l00190"></a><span class="lineno"> 190</span>  <a class="code" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a> data_diff_mult = mult_by_quantized_multiplier_serial(data_diff);</div><div class="line"><a name="l00191"></a><span class="lineno"> 191</span>  data_fp = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a54aedfa17c5ac2567107d5f488b0f4af">asymm_exp_on_negative_values</a>(data_diff_mult, SCALED_DIFF_INT_BITS);</div><div class="line"><a name="l00192"></a><span class="lineno"> 192</span>  data_fp = <a class="code" href="softmax__layer__quantized_8cl.xhtml#ad57ea340cdcfeb2e1375b70c3ae59bae">asymm_rescale</a>(data_fp, 0, EXP_ACCUMULATION_INT_BITS);</div><div class="line"><a name="l00193"></a><span class="lineno"> 193</span>  <a class="code" href="helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a7c78836761fa3b5b124efea237dac70f">VECTOR_SIZE</a>)</div><div class="line"><a name="l00194"></a><span class="lineno"> 194</span>  (data_diff, 0, (__global <span class="keywordtype">int</span> *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, i << LOG_VECTOR_SIZE, 0));</div><div class="line"><a name="l00195"></a><span class="lineno"> 195</span>  sum1D = sum1D + select(0, data_fp, data_diff >= (<a class="code" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a>)(DIFF_MIN));</div><div class="line"><a name="l00196"></a><span class="lineno"> 196</span>  }</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> <span class="preprocessor">#ifdef NON_MULTIPLE_OF_VECTOR_SIZE</span></div><div class="line"><a name="l00199"></a><span class="lineno"> 199</span>  <span class="comment">// Handle non multiple of 16</span></div><div class="line"><a name="l00200"></a><span class="lineno"> 200</span>  data = <a class="code" href="helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a7c78836761fa3b5b124efea237dac70f">VECTOR_SIZE</a>)(0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, width4 << LOG_VECTOR_SIZE, 0));</div><div class="line"><a name="l00201"></a><span class="lineno"> 201</span>  <a class="code" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a> data_fp = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(data, <a class="code" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a>);</div><div class="line"><a name="l00202"></a><span class="lineno"> 202</span>  <a class="code" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a> data_diff = data_fp - max_val;</div><div class="line"><a name="l00203"></a><span class="lineno"> 203</span>  <a class="code" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a> data_diff_mult = mult_by_quantized_multiplier_serial(data_diff);</div><div class="line"><a name="l00204"></a><span class="lineno"> 204</span>  data_fp = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a54aedfa17c5ac2567107d5f488b0f4af">asymm_exp_on_negative_values</a>(data_diff_mult, SCALED_DIFF_INT_BITS);</div><div class="line"><a name="l00205"></a><span class="lineno"> 205</span>  data_fp = <a class="code" href="softmax__layer__quantized_8cl.xhtml#ad57ea340cdcfeb2e1375b70c3ae59bae">asymm_rescale</a>(data_fp, 0, EXP_ACCUMULATION_INT_BITS);</div><div class="line"><a name="l00206"></a><span class="lineno"> 206</span>  <a class="code" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a> widx_ = <a class="code" href="fixed__point_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(((<a class="code" href="softmax__layer__quantized_8cl.xhtml#a16110bd2b92003141dbaf8a44498ff82">VEC_UINT</a>)(width4 << LOG_VECTOR_SIZE) + idx__) < width, <a class="code" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a>);</div><div class="line"><a name="l00207"></a><span class="lineno"> 207</span>  <a class="code" href="helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a7c78836761fa3b5b124efea237dac70f">VECTOR_SIZE</a>)</div><div class="line"><a name="l00208"></a><span class="lineno"> 208</span>  (data_diff, 0, (__global <span class="keywordtype">int</span> *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, width4 << LOG_VECTOR_SIZE, 0));</div><div class="line"><a name="l00209"></a><span class="lineno"> 209</span>  data_fp = select(0, data_fp, data_diff >= (<a class="code" href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a>)(DIFF_MIN));</div><div class="line"><a name="l00210"></a><span class="lineno"> 210</span>  sum1D = sum1D + select(0, data_fp, widx_);</div><div class="line"><a name="l00211"></a><span class="lineno"> 211</span> <span class="preprocessor">#endif </span><span class="comment">/* NON_MULTIPLE_OF_VECTOR_SIZE */</span><span class="preprocessor"></span></div><div class="line"><a name="l00212"></a><span class="lineno"> 212</span> </div><div class="line"><a name="l00213"></a><span class="lineno"> 213</span>  <span class="comment">// Perform sum reduction</span></div><div class="line"><a name="l00214"></a><span class="lineno"> 214</span> <span class="preprocessor">#if VECTOR_SIZE == 16</span></div><div class="line"><a name="l00215"></a><span class="lineno"> 215</span>  sum1D.s01234567 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(sum1D.s01234567, sum1D.s89ABCDEF, uchar, 8);</div><div class="line"><a name="l00216"></a><span class="lineno"> 216</span> <span class="preprocessor">#endif </span><span class="comment">/* VECTOR SIZE 16 END */</span><span class="preprocessor"></span></div><div class="line"><a name="l00217"></a><span class="lineno"> 217</span> <span class="preprocessor">#if VECTOR_SIZE >= 8</span></div><div class="line"><a name="l00218"></a><span class="lineno"> 218</span>  sum1D.s0123 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(sum1D.s0123, sum1D.s4567, uchar, 4);</div><div class="line"><a name="l00219"></a><span class="lineno"> 219</span> <span class="preprocessor">#endif </span><span class="comment">/* VECTOR SIZE 8 END */</span><span class="preprocessor"></span></div><div class="line"><a name="l00220"></a><span class="lineno"> 220</span> <span class="preprocessor">#if VECTOR_SIZE >= 4</span></div><div class="line"><a name="l00221"></a><span class="lineno"> 221</span>  sum1D.s01 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(sum1D.s01, sum1D.s23, uchar, 2);</div><div class="line"><a name="l00222"></a><span class="lineno"> 222</span> <span class="preprocessor">#endif </span><span class="comment">/* VECTOR SIZE 4 END */</span><span class="preprocessor"></span></div><div class="line"><a name="l00223"></a><span class="lineno"> 223</span>  sum1D.s0 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(sum1D.s0, sum1D.s1, uchar, 1);</div><div class="line"><a name="l00224"></a><span class="lineno"> 224</span> </div><div class="line"><a name="l00225"></a><span class="lineno"> 225</span>  <span class="comment">// Calculate and store result</span></div><div class="line"><a name="l00226"></a><span class="lineno"> 226</span>  *((__global <span class="keywordtype">int</span> *)sum.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) = sum1D.s0;</div><div class="line"><a name="l00227"></a><span class="lineno"> 227</span> }</div><div class="line"><a name="l00228"></a><span class="lineno"> 228</span> </div><div class="line"><a name="l00270"></a><span class="lineno"> 270</span> __kernel <span class="keywordtype">void</span> softmax_layer_max_shift_exp_sum_quantized_parallel(</div><div class="line"><a name="l00271"></a><span class="lineno"> 271</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a6743f0a130e8311e6f5b1a23df102472">src</a>),</div><div class="line"><a name="l00272"></a><span class="lineno"> 272</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(maxo),</div><div class="line"><a name="l00273"></a><span class="lineno"> 273</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="l00274"></a><span class="lineno"> 274</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="reduction__operation_8cl.xhtml#ab0df00f5333da51860deb93deb44a782">sum</a>),</div><div class="line"><a name="l00275"></a><span class="lineno"> 275</span>  uint width)</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>  <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a6743f0a130e8311e6f5b1a23df102472">src</a> = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(src);</div><div class="line"><a name="l00278"></a><span class="lineno"> 278</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="l00279"></a><span class="lineno"> 279</span>  <a class="code" href="struct_image.xhtml">Image</a> maxo = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(maxo);</div><div class="line"><a name="l00280"></a><span class="lineno"> 280</span>  <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="reduction__operation_8cl.xhtml#ab0df00f5333da51860deb93deb44a782">sum</a> = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(sum);</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>  <span class="keyword">const</span> uint4 <a class="code" href="softmax__layer_8cl.xhtml#a4884a666a1e93fbf8c27bd7d2da3c8bb">idx4</a> = (uint4)(0, 1, 2, 3);</div><div class="line"><a name="l00283"></a><span class="lineno"> 283</span>  <span class="keyword">const</span> uint lid = get_local_id(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">// Define one temporary vector per work-item.</span></div><div class="line"><a name="l00286"></a><span class="lineno"> 286</span>  __local int4 tmp_local[<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a>];</div><div class="line"><a name="l00287"></a><span class="lineno"> 287</span>  __local uchar max_local;</div><div class="line"><a name="l00288"></a><span class="lineno"> 288</span> </div><div class="line"><a name="l00289"></a><span class="lineno"> 289</span>  uchar4 uchar_min = (uchar4)0;</div><div class="line"><a name="l00290"></a><span class="lineno"> 290</span>  uchar4 max_val_vec = uchar_min;</div><div class="line"><a name="l00291"></a><span class="lineno"> 291</span> </div><div class="line"><a name="l00292"></a><span class="lineno"> 292</span>  <span class="comment">// Number of elements per work-item.</span></div><div class="line"><a name="l00293"></a><span class="lineno"> 293</span>  <span class="keyword">const</span> uint row = width / <a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a>;</div><div class="line"><a name="l00294"></a><span class="lineno"> 294</span>  <span class="comment">// Number of iterations per work-item.</span></div><div class="line"><a name="l00295"></a><span class="lineno"> 295</span>  <span class="keyword">const</span> uint width_ = row >> 2;</div><div class="line"><a name="l00296"></a><span class="lineno"> 296</span>  <span class="comment">// Calculate max of row</span></div><div class="line"><a name="l00297"></a><span class="lineno"> 297</span>  uint i = 0;</div><div class="line"><a name="l00298"></a><span class="lineno"> 298</span>  <span class="keywordflow">for</span>(; i < width_; i++)</div><div class="line"><a name="l00299"></a><span class="lineno"> 299</span>  {</div><div class="line"><a name="l00300"></a><span class="lineno"> 300</span>  uchar4 data_max = vload4(0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, i * <a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> * 4, 0));</div><div class="line"><a name="l00301"></a><span class="lineno"> 301</span>  max_val_vec = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(data_max, max_val_vec, uchar, 4);</div><div class="line"><a name="l00302"></a><span class="lineno"> 302</span>  }</div><div class="line"><a name="l00303"></a><span class="lineno"> 303</span> <span class="preprocessor">#ifdef NON_MULTIPLE_OF_GRID_SIZE</span></div><div class="line"><a name="l00304"></a><span class="lineno"> 304</span>  <span class="comment">// How many work-items needed to complete the computation.</span></div><div class="line"><a name="l00305"></a><span class="lineno"> 305</span>  <span class="keywordtype">int</span> boundary_workitems = (width % (<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> * 4)) / 4;</div><div class="line"><a name="l00306"></a><span class="lineno"> 306</span>  <span class="keywordflow">if</span>(lid < boundary_workitems)</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>  uchar4 data_max = vload4(0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, i * <a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> * 4, 0));</div><div class="line"><a name="l00309"></a><span class="lineno"> 309</span>  max_val_vec = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(data_max, max_val_vec, uchar, 4);</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 NON_MULTIPLE_OF_VECTOR_SIZE</span></div><div class="line"><a name="l00312"></a><span class="lineno"> 312</span>  <span class="keywordflow">if</span>(boundary_workitems == 0)</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>  boundary_workitems = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a>;</div><div class="line"><a name="l00315"></a><span class="lineno"> 315</span>  i--;</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>  <span class="keywordflow">if</span>(lid == (boundary_workitems - 1))</div><div class="line"><a name="l00318"></a><span class="lineno"> 318</span>  {</div><div class="line"><a name="l00319"></a><span class="lineno"> 319</span>  <span class="comment">// Handle non multiple of 4</span></div><div class="line"><a name="l00320"></a><span class="lineno"> 320</span>  uchar4 data_max = vload4(0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, (<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> * i * 4) + 4, 0));</div><div class="line"><a name="l00321"></a><span class="lineno"> 321</span>  uchar4 widx = convert_uchar4(((uint4)(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> * i * 4) + boundary_workitems * 4 + idx4) < width);</div><div class="line"><a name="l00322"></a><span class="lineno"> 322</span>  max_val_vec = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(max_val_vec, select(uchar_min, data_max, widx), uchar, 4);</div><div class="line"><a name="l00323"></a><span class="lineno"> 323</span>  }</div><div class="line"><a name="l00324"></a><span class="lineno"> 324</span> <span class="preprocessor">#endif </span><span class="comment">/* NON_MULTIPLE_OF_VECTOR_SIZE */</span><span class="preprocessor"></span></div><div class="line"><a name="l00325"></a><span class="lineno"> 325</span> <span class="preprocessor">#endif </span><span class="comment">/* NON_MULTIPLE_OF_GRID_SIZE */</span><span class="preprocessor"></span></div><div class="line"><a name="l00326"></a><span class="lineno"> 326</span>  tmp_local[lid] = convert_int4(max_val_vec);</div><div class="line"><a name="l00327"></a><span class="lineno"> 327</span> </div><div class="line"><a name="l00328"></a><span class="lineno"> 328</span>  barrier(CLK_LOCAL_MEM_FENCE);</div><div class="line"><a name="l00329"></a><span class="lineno"> 329</span> </div><div class="line"><a name="l00330"></a><span class="lineno"> 330</span>  <span class="keywordflow">if</span>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> >= 256)</div><div class="line"><a name="l00331"></a><span class="lineno"> 331</span>  {</div><div class="line"><a name="l00332"></a><span class="lineno"> 332</span>  <span class="keywordflow">if</span>(lid < 128)</div><div class="line"><a name="l00333"></a><span class="lineno"> 333</span>  {</div><div class="line"><a name="l00334"></a><span class="lineno"> 334</span>  tmp_local[lid] = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(tmp_local[lid + 128], tmp_local[lid], <span class="keywordtype">int</span>, 4);</div><div class="line"><a name="l00335"></a><span class="lineno"> 335</span>  }</div><div class="line"><a name="l00336"></a><span class="lineno"> 336</span>  barrier(CLK_LOCAL_MEM_FENCE);</div><div class="line"><a name="l00337"></a><span class="lineno"> 337</span>  }</div><div class="line"><a name="l00338"></a><span class="lineno"> 338</span>  <span class="keywordflow">if</span>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> >= 128)</div><div class="line"><a name="l00339"></a><span class="lineno"> 339</span>  {</div><div class="line"><a name="l00340"></a><span class="lineno"> 340</span>  <span class="keywordflow">if</span>(lid < 64)</div><div class="line"><a name="l00341"></a><span class="lineno"> 341</span>  {</div><div class="line"><a name="l00342"></a><span class="lineno"> 342</span>  tmp_local[lid] = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(tmp_local[lid + 64], tmp_local[lid], <span class="keywordtype">int</span>, 4);</div><div class="line"><a name="l00343"></a><span class="lineno"> 343</span>  }</div><div class="line"><a name="l00344"></a><span class="lineno"> 344</span>  barrier(CLK_LOCAL_MEM_FENCE);</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="keywordflow">if</span>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> >= 64)</div><div class="line"><a name="l00347"></a><span class="lineno"> 347</span>  {</div><div class="line"><a name="l00348"></a><span class="lineno"> 348</span>  <span class="keywordflow">if</span>(lid < 32)</div><div class="line"><a name="l00349"></a><span class="lineno"> 349</span>  {</div><div class="line"><a name="l00350"></a><span class="lineno"> 350</span>  tmp_local[lid] = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(tmp_local[lid + 32], tmp_local[lid], <span class="keywordtype">int</span>, 4);</div><div class="line"><a name="l00351"></a><span class="lineno"> 351</span>  }</div><div class="line"><a name="l00352"></a><span class="lineno"> 352</span>  barrier(CLK_LOCAL_MEM_FENCE);</div><div class="line"><a name="l00353"></a><span class="lineno"> 353</span>  }</div><div class="line"><a name="l00354"></a><span class="lineno"> 354</span>  <span class="keywordflow">if</span>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> >= 32)</div><div class="line"><a name="l00355"></a><span class="lineno"> 355</span>  {</div><div class="line"><a name="l00356"></a><span class="lineno"> 356</span>  <span class="keywordflow">if</span>(lid < 16)</div><div class="line"><a name="l00357"></a><span class="lineno"> 357</span>  {</div><div class="line"><a name="l00358"></a><span class="lineno"> 358</span>  tmp_local[lid] = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(tmp_local[lid + 16], tmp_local[lid], <span class="keywordtype">int</span>, 4);</div><div class="line"><a name="l00359"></a><span class="lineno"> 359</span>  }</div><div class="line"><a name="l00360"></a><span class="lineno"> 360</span>  barrier(CLK_LOCAL_MEM_FENCE);</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>  <span class="keywordflow">if</span>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> >= 16)</div><div class="line"><a name="l00363"></a><span class="lineno"> 363</span>  {</div><div class="line"><a name="l00364"></a><span class="lineno"> 364</span>  <span class="keywordflow">if</span>(lid < 8)</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>  tmp_local[lid] = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(tmp_local[lid + 8], tmp_local[lid], <span class="keywordtype">int</span>, 4);</div><div class="line"><a name="l00367"></a><span class="lineno"> 367</span>  }</div><div class="line"><a name="l00368"></a><span class="lineno"> 368</span>  barrier(CLK_LOCAL_MEM_FENCE);</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>  <span class="keywordflow">if</span>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> >= 8)</div><div class="line"><a name="l00371"></a><span class="lineno"> 371</span>  {</div><div class="line"><a name="l00372"></a><span class="lineno"> 372</span>  <span class="keywordflow">if</span>(lid < 4)</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>  tmp_local[lid] = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(tmp_local[lid + 4], tmp_local[lid], <span class="keywordtype">int</span>, 4);</div><div class="line"><a name="l00375"></a><span class="lineno"> 375</span>  }</div><div class="line"><a name="l00376"></a><span class="lineno"> 376</span>  barrier(CLK_LOCAL_MEM_FENCE);</div><div class="line"><a name="l00377"></a><span class="lineno"> 377</span>  }</div><div class="line"><a name="l00378"></a><span class="lineno"> 378</span>  <span class="keywordflow">if</span>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> >= 4)</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="keywordflow">if</span>(lid < 2)</div><div class="line"><a name="l00381"></a><span class="lineno"> 381</span>  {</div><div class="line"><a name="l00382"></a><span class="lineno"> 382</span>  tmp_local[lid] = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(tmp_local[lid + 2], tmp_local[lid], <span class="keywordtype">int</span>, 4);</div><div class="line"><a name="l00383"></a><span class="lineno"> 383</span>  }</div><div class="line"><a name="l00384"></a><span class="lineno"> 384</span>  barrier(CLK_LOCAL_MEM_FENCE);</div><div class="line"><a name="l00385"></a><span class="lineno"> 385</span>  }</div><div class="line"><a name="l00386"></a><span class="lineno"> 386</span>  <span class="keywordflow">if</span>(lid == 0)</div><div class="line"><a name="l00387"></a><span class="lineno"> 387</span>  {</div><div class="line"><a name="l00388"></a><span class="lineno"> 388</span>  max_val_vec = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(convert_uchar4(tmp_local[lid + 1]), convert_uchar4(tmp_local[lid]), uchar, 4);</div><div class="line"><a name="l00389"></a><span class="lineno"> 389</span>  max_val_vec.s01 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(max_val_vec.s01, max_val_vec.s23, uchar, 2);</div><div class="line"><a name="l00390"></a><span class="lineno"> 390</span>  max_val_vec.s0 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a>(max_val_vec.s0, max_val_vec.s1, uchar, 1);</div><div class="line"><a name="l00391"></a><span class="lineno"> 391</span>  max_local = max_val_vec.s0;</div><div class="line"><a name="l00392"></a><span class="lineno"> 392</span>  }</div><div class="line"><a name="l00393"></a><span class="lineno"> 393</span>  barrier(CLK_LOCAL_MEM_FENCE);</div><div class="line"><a name="l00394"></a><span class="lineno"> 394</span> </div><div class="line"><a name="l00395"></a><span class="lineno"> 395</span>  <span class="comment">/* Second section */</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>  <span class="comment">// Set sum vector</span></div><div class="line"><a name="l00398"></a><span class="lineno"> 398</span>  int4 sum1D = 0;</div><div class="line"><a name="l00399"></a><span class="lineno"> 399</span>  <span class="keywordtype">int</span> max_val = convert_int(max_local);</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>  <span class="comment">// Shift values, exp and sum</span></div><div class="line"><a name="l00402"></a><span class="lineno"> 402</span>  <span class="keywordflow">for</span>(i = 0; i < width_; i++)</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>  uchar4 data = vload4(0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, i * <a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> * 4, 0));</div><div class="line"><a name="l00405"></a><span class="lineno"> 405</span>  int4 data_fp = convert_int4(data);</div><div class="line"><a name="l00406"></a><span class="lineno"> 406</span>  int4 data_diff = data_fp - max_val;</div><div class="line"><a name="l00407"></a><span class="lineno"> 407</span>  int4 data_diff_mult = mult_by_quantized_multiplier_parallel(data_diff);</div><div class="line"><a name="l00408"></a><span class="lineno"> 408</span>  data_fp = <a class="code" href="helpers__asymm_8h.xhtml#a3a4f1b5d8f1cd67ac31bc62c9a6f4aa8">ASYMM_EXP_ON_NEGATIVE_VALUES</a>(data_diff_mult, SCALED_DIFF_INT_BITS, 4);</div><div class="line"><a name="l00409"></a><span class="lineno"> 409</span>  data_fp = <a class="code" href="helpers__asymm_8h.xhtml#a98585f1bb84dea90aecbf59785c46151">ASYMM_RESCALE</a>(data_fp, 0, EXP_ACCUMULATION_INT_BITS, 4);</div><div class="line"><a name="l00410"></a><span class="lineno"> 410</span>  vstore4(data_diff, 0, (__global <span class="keywordtype">int</span> *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, i * GRID_SIZE * 4, 0));</div><div class="line"><a name="l00411"></a><span class="lineno"> 411</span>  sum1D = sum1D + select(0, data_fp, data_diff >= (int4)(DIFF_MIN));</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> <span class="preprocessor">#ifdef NON_MULTIPLE_OF_GRID_SIZE</span></div><div class="line"><a name="l00414"></a><span class="lineno"> 414</span>  boundary_workitems = (width % (<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> * 4)) / 4;</div><div class="line"><a name="l00415"></a><span class="lineno"> 415</span>  <span class="keywordflow">if</span>(lid < boundary_workitems)</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>  uchar4 data = vload4(0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, i * <a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> * 4, 0));</div><div class="line"><a name="l00418"></a><span class="lineno"> 418</span>  int4 data_fp = convert_int4(data);</div><div class="line"><a name="l00419"></a><span class="lineno"> 419</span>  int4 data_diff = data_fp - max_val;</div><div class="line"><a name="l00420"></a><span class="lineno"> 420</span>  int4 data_diff_mult = mult_by_quantized_multiplier_parallel(data_diff);</div><div class="line"><a name="l00421"></a><span class="lineno"> 421</span>  data_fp = <a class="code" href="helpers__asymm_8h.xhtml#a3a4f1b5d8f1cd67ac31bc62c9a6f4aa8">ASYMM_EXP_ON_NEGATIVE_VALUES</a>(data_diff_mult, SCALED_DIFF_INT_BITS, 4);</div><div class="line"><a name="l00422"></a><span class="lineno"> 422</span>  data_fp = <a class="code" href="helpers__asymm_8h.xhtml#a98585f1bb84dea90aecbf59785c46151">ASYMM_RESCALE</a>(data_fp, 0, EXP_ACCUMULATION_INT_BITS, 4);</div><div class="line"><a name="l00423"></a><span class="lineno"> 423</span>  vstore4(data_diff, 0, (__global <span class="keywordtype">int</span> *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, i * GRID_SIZE * 4, 0));</div><div class="line"><a name="l00424"></a><span class="lineno"> 424</span>  sum1D = sum1D + select(0, data_fp, data_diff >= (int4)(DIFF_MIN));</div><div class="line"><a name="l00425"></a><span class="lineno"> 425</span>  }</div><div class="line"><a name="l00426"></a><span class="lineno"> 426</span> <span class="preprocessor">#ifdef NON_MULTIPLE_OF_VECTOR_SIZE</span></div><div class="line"><a name="l00427"></a><span class="lineno"> 427</span>  <span class="keywordflow">if</span>(boundary_workitems == 0)</div><div class="line"><a name="l00428"></a><span class="lineno"> 428</span>  {</div><div class="line"><a name="l00429"></a><span class="lineno"> 429</span>  boundary_workitems = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a>;</div><div class="line"><a name="l00430"></a><span class="lineno"> 430</span>  i--;</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>  <span class="keywordflow">if</span>(lid == (boundary_workitems - 1))</div><div class="line"><a name="l00433"></a><span class="lineno"> 433</span>  {</div><div class="line"><a name="l00434"></a><span class="lineno"> 434</span>  <span class="comment">// Handle non multiple of vector size ((GRID_SIZE * i * 4) + 4, 0); move 4 float positions ahead, *4 is due to the stride</span></div><div class="line"><a name="l00435"></a><span class="lineno"> 435</span>  uchar4 data = vload4(0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, i * <a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> * 4 + 4, 0));</div><div class="line"><a name="l00436"></a><span class="lineno"> 436</span>  int4 data_fp = convert_int4(data);</div><div class="line"><a name="l00437"></a><span class="lineno"> 437</span>  int4 data_diff = data_fp - max_val;</div><div class="line"><a name="l00438"></a><span class="lineno"> 438</span>  int4 data_diff_mult = mult_by_quantized_multiplier_parallel(data_diff);</div><div class="line"><a name="l00439"></a><span class="lineno"> 439</span>  data_fp = <a class="code" href="helpers__asymm_8h.xhtml#a3a4f1b5d8f1cd67ac31bc62c9a6f4aa8">ASYMM_EXP_ON_NEGATIVE_VALUES</a>(data_diff_mult, SCALED_DIFF_INT_BITS, 4);</div><div class="line"><a name="l00440"></a><span class="lineno"> 440</span>  data_fp = <a class="code" href="helpers__asymm_8h.xhtml#a98585f1bb84dea90aecbf59785c46151">ASYMM_RESCALE</a>(data_fp, 0, EXP_ACCUMULATION_INT_BITS, 4);</div><div class="line"><a name="l00441"></a><span class="lineno"> 441</span>  int4 widx = convert_int4(((uint4)(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> * i * 4) + boundary_workitems * 4 + idx4) < width);</div><div class="line"><a name="l00442"></a><span class="lineno"> 442</span>  data_fp = select(0, data_fp, widx);</div><div class="line"><a name="l00443"></a><span class="lineno"> 443</span>  vstore4(data_diff, 0, (__global <span class="keywordtype">int</span> *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, i * <a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> * 4 + 4, 0));</div><div class="line"><a name="l00444"></a><span class="lineno"> 444</span>  sum1D = sum1D + select(0, data_fp, data_diff >= (int4)(DIFF_MIN));</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> <span class="preprocessor">#endif </span><span class="comment">/* NON_MULTIPLE_OF_VECTOR_SIZE */</span><span class="preprocessor"></span></div><div class="line"><a name="l00447"></a><span class="lineno"> 447</span> <span class="preprocessor">#endif </span><span class="comment">/* NON_MULTIPLE_OF_GRID_SIZE */</span><span class="preprocessor"></span></div><div class="line"><a name="l00448"></a><span class="lineno"> 448</span>  tmp_local[lid] = sum1D;</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>  barrier(CLK_LOCAL_MEM_FENCE);</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="keywordflow">if</span>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> >= 256)</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="keywordflow">if</span>(lid < 128)</div><div class="line"><a name="l00455"></a><span class="lineno"> 455</span>  {</div><div class="line"><a name="l00456"></a><span class="lineno"> 456</span>  tmp_local[lid] = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(tmp_local[lid + 128], tmp_local[lid], <span class="keywordtype">int</span>, 4);</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>  barrier(CLK_LOCAL_MEM_FENCE);</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>  <span class="keywordflow">if</span>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> >= 128)</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>  <span class="keywordflow">if</span>(lid < 64)</div><div class="line"><a name="l00463"></a><span class="lineno"> 463</span>  {</div><div class="line"><a name="l00464"></a><span class="lineno"> 464</span>  tmp_local[lid] = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(tmp_local[lid + 64], tmp_local[lid], <span class="keywordtype">int</span>, 4);</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>  barrier(CLK_LOCAL_MEM_FENCE);</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="keywordflow">if</span>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> >= 64)</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="keywordflow">if</span>(lid < 32)</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>  tmp_local[lid] = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(tmp_local[lid + 32], tmp_local[lid], <span class="keywordtype">int</span>, 4);</div><div class="line"><a name="l00473"></a><span class="lineno"> 473</span>  }</div><div class="line"><a name="l00474"></a><span class="lineno"> 474</span>  barrier(CLK_LOCAL_MEM_FENCE);</div><div class="line"><a name="l00475"></a><span class="lineno"> 475</span>  }</div><div class="line"><a name="l00476"></a><span class="lineno"> 476</span>  <span class="keywordflow">if</span>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> >= 32)</div><div class="line"><a name="l00477"></a><span class="lineno"> 477</span>  {</div><div class="line"><a name="l00478"></a><span class="lineno"> 478</span>  <span class="keywordflow">if</span>(lid < 16)</div><div class="line"><a name="l00479"></a><span class="lineno"> 479</span>  {</div><div class="line"><a name="l00480"></a><span class="lineno"> 480</span>  tmp_local[lid] = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(tmp_local[lid + 16], tmp_local[lid], <span class="keywordtype">int</span>, 4);</div><div class="line"><a name="l00481"></a><span class="lineno"> 481</span>  }</div><div class="line"><a name="l00482"></a><span class="lineno"> 482</span>  barrier(CLK_LOCAL_MEM_FENCE);</div><div class="line"><a name="l00483"></a><span class="lineno"> 483</span>  }</div><div class="line"><a name="l00484"></a><span class="lineno"> 484</span>  <span class="keywordflow">if</span>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> >= 16)</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>  <span class="keywordflow">if</span>(lid < 8)</div><div class="line"><a name="l00487"></a><span class="lineno"> 487</span>  {</div><div class="line"><a name="l00488"></a><span class="lineno"> 488</span>  tmp_local[lid] = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(tmp_local[lid + 8], tmp_local[lid], <span class="keywordtype">int</span>, 4);</div><div class="line"><a name="l00489"></a><span class="lineno"> 489</span>  }</div><div class="line"><a name="l00490"></a><span class="lineno"> 490</span>  barrier(CLK_LOCAL_MEM_FENCE);</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>  <span class="keywordflow">if</span>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> >= 8)</div><div class="line"><a name="l00493"></a><span class="lineno"> 493</span>  {</div><div class="line"><a name="l00494"></a><span class="lineno"> 494</span>  <span class="keywordflow">if</span>(lid < 4)</div><div class="line"><a name="l00495"></a><span class="lineno"> 495</span>  {</div><div class="line"><a name="l00496"></a><span class="lineno"> 496</span>  tmp_local[lid] = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(tmp_local[lid + 4], tmp_local[lid], <span class="keywordtype">int</span>, 4);</div><div class="line"><a name="l00497"></a><span class="lineno"> 497</span>  }</div><div class="line"><a name="l00498"></a><span class="lineno"> 498</span>  barrier(CLK_LOCAL_MEM_FENCE);</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="keywordflow">if</span>(<a class="code" href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a> >= 4)</div><div class="line"><a name="l00501"></a><span class="lineno"> 501</span>  {</div><div class="line"><a name="l00502"></a><span class="lineno"> 502</span>  <span class="keywordflow">if</span>(lid < 2)</div><div class="line"><a name="l00503"></a><span class="lineno"> 503</span>  {</div><div class="line"><a name="l00504"></a><span class="lineno"> 504</span>  tmp_local[lid] = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(tmp_local[lid + 2], tmp_local[lid], <span class="keywordtype">int</span>, 4);</div><div class="line"><a name="l00505"></a><span class="lineno"> 505</span>  }</div><div class="line"><a name="l00506"></a><span class="lineno"> 506</span>  barrier(CLK_LOCAL_MEM_FENCE);</div><div class="line"><a name="l00507"></a><span class="lineno"> 507</span>  }</div><div class="line"><a name="l00508"></a><span class="lineno"> 508</span>  <span class="keywordflow">if</span>(lid == 0)</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>  sum1D = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(tmp_local[lid + 1], tmp_local[lid], <span class="keywordtype">int</span>, 4);</div><div class="line"><a name="l00511"></a><span class="lineno"> 511</span>  <span class="comment">// Perform max reduction</span></div><div class="line"><a name="l00512"></a><span class="lineno"> 512</span>  sum1D.s01 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(sum1D.s01, sum1D.s23, <span class="keywordtype">int</span>, 2);</div><div class="line"><a name="l00513"></a><span class="lineno"> 513</span>  sum1D.s0 = <a class="code" href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a>(sum1D.s0, sum1D.s1, <span class="keywordtype">int</span>, 1);</div><div class="line"><a name="l00514"></a><span class="lineno"> 514</span>  *((__global <span class="keywordtype">int</span> *)sum.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) = sum1D.s0;</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> }</div><div class="line"><a name="l00517"></a><span class="lineno"> 517</span> </div><div class="line"><a name="l00549"></a><span class="lineno"> 549</span> __kernel <span class="keywordtype">void</span> softmax_layer_norm_quantized(</div><div class="line"><a name="l00550"></a><span class="lineno"> 550</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a6743f0a130e8311e6f5b1a23df102472">src</a>),</div><div class="line"><a name="l00551"></a><span class="lineno"> 551</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="reduction__operation_8cl.xhtml#ab0df00f5333da51860deb93deb44a782">sum</a>),</div><div class="line"><a name="l00552"></a><span class="lineno"> 552</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="l00553"></a><span class="lineno"> 553</span> {</div><div class="line"><a name="l00554"></a><span class="lineno"> 554</span>  <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a6743f0a130e8311e6f5b1a23df102472">src</a> = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(src);</div><div class="line"><a name="l00555"></a><span class="lineno"> 555</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="l00556"></a><span class="lineno"> 556</span>  <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="reduction__operation_8cl.xhtml#ab0df00f5333da51860deb93deb44a782">sum</a> = <a class="code" href="helpers_8h.xhtml#a071aa45af973feac43b14f62e54a6fce">CONVERT_TENSOR3D_TO_IMAGE_STRUCT_NO_STEP</a>(sum);</div><div class="line"><a name="l00557"></a><span class="lineno"> 557</span> </div><div class="line"><a name="l00558"></a><span class="lineno"> 558</span>  <span class="comment">// Load max value of 1D logits vector (row)</span></div><div class="line"><a name="l00559"></a><span class="lineno"> 559</span>  <span class="keywordtype">int</span> sum_val = *((__global <span class="keywordtype">int</span> *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&sum, 0, get_global_id(1)));</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="comment">// It will be better to calculate this in prev layer and pass here as parameter</span></div><div class="line"><a name="l00562"></a><span class="lineno"> 562</span>  uint sum_val_u = convert_uint(sum_val);</div><div class="line"><a name="l00563"></a><span class="lineno"> 563</span>  <span class="keywordtype">int</span> headroom_plus_one = <a class="code" href="namespacearm__compute_1_1test_1_1fixed__point__arithmetic_1_1detail.xhtml#aa43c67a536c661ec28398d06bc8ba5af">clz</a>(sum_val_u);</div><div class="line"><a name="l00564"></a><span class="lineno"> 564</span>  <span class="keywordtype">int</span> num_bits_over_unit = EXP_ACCUMULATION_INT_BITS - headroom_plus_one;</div><div class="line"><a name="l00565"></a><span class="lineno"> 565</span>  <span class="keywordtype">int</span> shifted_sum_minus_one_1 = convert_int((sum_val_u << headroom_plus_one) - (1u << 31));</div><div class="line"><a name="l00566"></a><span class="lineno"> 566</span>  int16 shifted_sum_minus_one = shifted_sum_minus_one_1;</div><div class="line"><a name="l00567"></a><span class="lineno"> 567</span>  int16 shifted_scale = <a class="code" href="helpers__asymm_8h.xhtml#ae77f34e1316d52c1ee84c35be9efb0d8">ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1</a>(shifted_sum_minus_one, 16);</div><div class="line"><a name="l00568"></a><span class="lineno"> 568</span> </div><div class="line"><a name="l00569"></a><span class="lineno"> 569</span>  <span class="comment">// It was already calculated in prev layer, should be stored into tmp output and reused</span></div><div class="line"><a name="l00570"></a><span class="lineno"> 570</span>  int16 data_diff = vload16(0, (__global <span class="keywordtype">int</span> *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 0));</div><div class="line"><a name="l00571"></a><span class="lineno"> 571</span>  int16 data_diff_mult = data_diff;</div><div class="line"><a name="l00572"></a><span class="lineno"> 572</span> <span class="preprocessor">#if defined(INPUT_BETA_MULTIPLIER) && defined(INPUT_BETA_LEFT_SHIFT)</span></div><div class="line"><a name="l00573"></a><span class="lineno"> 573</span>  <span class="keywordflow">if</span>(INPUT_BETA_MULTIPLIER > 1)</div><div class="line"><a name="l00574"></a><span class="lineno"> 574</span>  {</div><div class="line"><a name="l00575"></a><span class="lineno"> 575</span>  data_diff_mult = <a class="code" href="helpers__asymm_8h.xhtml#a5483aefd5e07244661178bfd3f434448">ASYMM_MULT</a>(data_diff * (1 << INPUT_BETA_LEFT_SHIFT), INPUT_BETA_MULTIPLIER, 16);</div><div class="line"><a name="l00576"></a><span class="lineno"> 576</span>  }</div><div class="line"><a name="l00577"></a><span class="lineno"> 577</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(INPUT_BETA_MULTIPLIER) && defined(INPUT_BETA_LEFT_SHIFT) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00578"></a><span class="lineno"> 578</span>  int16 data = <a class="code" href="helpers__asymm_8h.xhtml#a3a4f1b5d8f1cd67ac31bc62c9a6f4aa8">ASYMM_EXP_ON_NEGATIVE_VALUES</a>(data_diff_mult, SCALED_DIFF_INT_BITS, 16);</div><div class="line"><a name="l00579"></a><span class="lineno"> 579</span> </div><div class="line"><a name="l00580"></a><span class="lineno"> 580</span>  data = <a class="code" href="helpers__asymm_8h.xhtml#a5483aefd5e07244661178bfd3f434448">ASYMM_MULT</a>(shifted_scale, data, 16);</div><div class="line"><a name="l00581"></a><span class="lineno"> 581</span>  data = <a class="code" href="helpers__asymm_8h.xhtml#aa43fc359dea64362f3016384f4269845">ASYMM_ROUNDING_DIVIDE_BY_POW2</a>(data, num_bits_over_unit + 31 - 8, 16);</div><div class="line"><a name="l00582"></a><span class="lineno"> 582</span>  data = select(0, data, data_diff >= (int16)(DIFF_MIN));</div><div class="line"><a name="l00583"></a><span class="lineno"> 583</span>  vstore16(convert_uchar16_sat(data), 0, (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0));</div><div class="line"><a name="l00584"></a><span class="lineno"> 584</span> }</div><div class="line"><a name="l00585"></a><span class="lineno"> 585</span> </div><div class="line"><a name="l00586"></a><span class="lineno"> 586</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(DIFF_MIN) */</span><span class="preprocessor"></span></div><div class="ttc" id="helpers_8h_xhtml_a071aa45af973feac43b14f62e54a6fce"><div class="ttname"><a href="helpers_8h.xhtml#a071aa45af973feac43b14f62e54a6fce">CONVERT_TENSOR3D_TO_IMAGE_STRUCT_NO_STEP</a></div><div class="ttdeci">#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT_NO_STEP(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00111">helpers.h:111</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_a287e2fc366c312b468382c95bb90f91f"><div class="ttname"><a href="helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a></div><div class="ttdeci">#define VLOAD(size)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00040">helpers.h:40</a></div></div>
+<div class="ttc" id="softmax__layer__quantized_8cl_xhtml_a16110bd2b92003141dbaf8a44498ff82"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#a16110bd2b92003141dbaf8a44498ff82">VEC_UINT</a></div><div class="ttdeci">#define VEC_UINT</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00063">softmax_layer_quantized.cl:63</a></div></div>
+<div class="ttc" id="softmax__layer__quantized_8cl_xhtml_a525a42d38133b1051b8924b456add4a1"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#a525a42d38133b1051b8924b456add4a1">asymm_mult</a></div><div class="ttdeci">#define asymm_mult(a, b)</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00056">softmax_layer_quantized.cl:56</a></div></div>
+<div class="ttc" id="softmax__layer__quantized_8cl_xhtml_aa1dd94b8d98f1c6d790bdf0fc5de29e9"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#aa1dd94b8d98f1c6d790bdf0fc5de29e9">idx__</a></div><div class="ttdeci">__constant uint16 idx__</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00055">softmax_layer_quantized.cl:55</a></div></div>
+<div class="ttc" id="softmax__layer__quantized_8cl_xhtml_a7c78836761fa3b5b124efea237dac70f"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#a7c78836761fa3b5b124efea237dac70f">VECTOR_SIZE</a></div><div class="ttdeci">#define VECTOR_SIZE</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00053">softmax_layer_quantized.cl:53</a></div></div>
+<div class="ttc" id="helpers__asymm_8h_xhtml_a5483aefd5e07244661178bfd3f434448"><div class="ttname"><a href="helpers__asymm_8h.xhtml#a5483aefd5e07244661178bfd3f434448">ASYMM_MULT</a></div><div class="ttdeci">#define ASYMM_MULT(a, b, size)</div><div class="ttdef"><b>Definition:</b> <a href="helpers__asymm_8h_source.xhtml#l00295">helpers_asymm.h:295</a></div></div>
+<div class="ttc" id="softmax__layer__quantized_8cl_xhtml_abaa48ad818c44e415fd3f9dd0f27bf01"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#abaa48ad818c44e415fd3f9dd0f27bf01">MAX_OP</a></div><div class="ttdeci">#define MAX_OP(x, y, type, size)</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00026">softmax_layer_quantized.cl:26</a></div></div>
<div class="ttc" id="reduction__operation_8cl_xhtml_ab0df00f5333da51860deb93deb44a782"><div class="ttname"><a href="reduction__operation_8cl.xhtml#ab0df00f5333da51860deb93deb44a782">sum</a></div><div class="ttdeci">DATA_TYPE sum(__global const DATA_TYPE *input)</div><div class="ttdoc">Calculate sum of a vector. </div><div class="ttdef"><b>Definition:</b> <a href="reduction__operation_8cl_source.xhtml#l00052">reduction_operation.cl:52</a></div></div>
-<div class="ttc" id="softmax__layer__quantized_8cl_xhtml_a44206a4e5783c7aabacec88aad878c88"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a></div><div class="ttdeci">#define ADD_OP(x, y, type, size)</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00028">softmax_layer_quantized.cl:28</a></div></div>
+<div class="ttc" id="helpers__asymm_8h_xhtml_aa43fc359dea64362f3016384f4269845"><div class="ttname"><a href="helpers__asymm_8h.xhtml#aa43fc359dea64362f3016384f4269845">ASYMM_ROUNDING_DIVIDE_BY_POW2</a></div><div class="ttdeci">#define ASYMM_ROUNDING_DIVIDE_BY_POW2(x, exponent, size)</div><div class="ttdef"><b>Definition:</b> <a href="helpers__asymm_8h_source.xhtml#l00294">helpers_asymm.h:294</a></div></div>
+<div class="ttc" id="softmax__layer__quantized_8cl_xhtml_a44206a4e5783c7aabacec88aad878c88"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#a44206a4e5783c7aabacec88aad878c88">ADD_OP</a></div><div class="ttdeci">#define ADD_OP(x, y, type, size)</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00027">softmax_layer_quantized.cl:27</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="softmax__layer__quantized_8cl_xhtml_a79bd9092e1a23c550d4dc1aeb9b746ad"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#a79bd9092e1a23c550d4dc1aeb9b746ad">softmax_layer_max_quantized</a></div><div class="ttdeci">__kernel void softmax_layer_max_quantized(__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, uint width)</div><div class="ttdoc">Identifies the maximum value across the 1st dimension. </div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00055">softmax_layer_quantized.cl:55</a></div></div>
-<div class="ttc" id="asymm__helper_8h_xhtml_a4496e3d044dda78318d27aa7f9c0e810"><div class="ttname"><a href="asymm__helper_8h.xhtml#a4496e3d044dda78318d27aa7f9c0e810">asymm_rescale</a></div><div class="ttdeci">int16 asymm_rescale(int16 value, int src_integer_bits, int dst_integer_bits)</div><div class="ttdoc">Considering the integer value as fixed-point, change the number of integer bits and update value acco...</div><div class="ttdef"><b>Definition:</b> <a href="asymm__helper_8h_source.xhtml#l00269">asymm_helper.h:269</a></div></div>
-<div class="ttc" id="fixed__point_8h_xhtml_a26a8ca855cd14b1867173d301baf6c4f"><div class="ttname"><a href="fixed__point_8h.xhtml#a26a8ca855cd14b1867173d301baf6c4f">qs16</a></div><div class="ttdeci">short qs16</div><div class="ttdef"><b>Definition:</b> <a href="fixed__point_8h_source.xhtml#l00037">fixed_point.h:37</a></div></div>
-<div class="ttc" id="asymm__helper_8h_xhtml_ab98c50e16a18dd439b2b7ecdd1ecd81c"><div class="ttname"><a href="asymm__helper_8h.xhtml#ab98c50e16a18dd439b2b7ecdd1ecd81c">asymm_mult</a></div><div class="ttdeci">int16 asymm_mult(int16 a, int16 b)</div><div class="ttdoc">Fixed-point multiplication. </div><div class="ttdef"><b>Definition:</b> <a href="asymm__helper_8h_source.xhtml#l00167">asymm_helper.h:167</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="asymm__helper_8h_xhtml_a766e0a81889256cfa161f7c08951c0b8"><div class="ttname"><a href="asymm__helper_8h.xhtml#a766e0a81889256cfa161f7c08951c0b8">asymm_rounding_divide_by_pow2</a></div><div class="ttdeci">int16 asymm_rounding_divide_by_pow2(int16 x, int exponent)</div><div class="ttdoc">Correctly rounded to nearest division by a power of two. </div><div class="ttdef"><b>Definition:</b> <a href="asymm__helper_8h_source.xhtml#l00082">asymm_helper.h:82</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">Tensor dst</div><div class="ttdef"><b>Definition:</b> <a href="validation_2_n_e_o_n_2_g_e_m_m_8cpp_source.xhtml#l00118">GEMM.cpp:118</a></div></div>
+<div class="ttc" id="helpers_8h_xhtml_acb282042d1edeeaa3cc979a206f78b54"><div class="ttname"><a href="helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a></div><div class="ttdeci">#define VSTORE(size)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00043">helpers.h:43</a></div></div>
+<div class="ttc" id="helpers__asymm_8h_xhtml_a98585f1bb84dea90aecbf59785c46151"><div class="ttname"><a href="helpers__asymm_8h.xhtml#a98585f1bb84dea90aecbf59785c46151">ASYMM_RESCALE</a></div><div class="ttdeci">#define ASYMM_RESCALE(value, src_integer_bits, dst_integer_bits, size)</div><div class="ttdef"><b>Definition:</b> <a href="helpers__asymm_8h_source.xhtml#l00307">helpers_asymm.h:307</a></div></div>
+<div class="ttc" id="softmax__layer_8cl_xhtml_a4884a666a1e93fbf8c27bd7d2da3c8bb"><div class="ttname"><a href="softmax__layer_8cl.xhtml#a4884a666a1e93fbf8c27bd7d2da3c8bb">idx4</a></div><div class="ttdeci">__constant uint4 idx4</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer_8cl_source.xhtml#l00088">softmax_layer.cl:88</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="softmax__layer__quantized_8cl_xhtml_ad57ea340cdcfeb2e1375b70c3ae59bae"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#ad57ea340cdcfeb2e1375b70c3ae59bae">asymm_rescale</a></div><div class="ttdeci">#define asymm_rescale(value, src_integer_bits, dst_integer_bits)</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00058">softmax_layer_quantized.cl:58</a></div></div>
+<div class="ttc" id="softmax__layer__quantized_8cl_xhtml_aee190caf3b3571e939ac129e12c368cd"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#aee190caf3b3571e939ac129e12c368cd">VEC_INT</a></div><div class="ttdeci">#define VEC_INT</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00064">softmax_layer_quantized.cl:64</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="softmax__layer__quantized_8cl_xhtml_a372393c380805985b813dbb16d589a64"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#a372393c380805985b813dbb16d589a64">LOG_VECTOR_SIZE</a></div><div class="ttdeci">#define LOG_VECTOR_SIZE</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00054">softmax_layer_quantized.cl:54</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="helpers__asymm_8h_xhtml"><div class="ttname"><a href="helpers__asymm_8h.xhtml">helpers_asymm.h</a></div></div>
+<div class="ttc" id="softmax__layer__quantized_8cl_xhtml_af5987b09a234231612b2b1eded343025"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#af5987b09a234231612b2b1eded343025">VEC_UCHAR</a></div><div class="ttdeci">#define VEC_UCHAR</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00062">softmax_layer_quantized.cl:62</a></div></div>
+<div class="ttc" id="softmax__layer__quantized_8cl_xhtml_a08246606c233e7785a497c09672f366f"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#a08246606c233e7785a497c09672f366f">GRID_SIZE</a></div><div class="ttdeci">#define GRID_SIZE</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00031">softmax_layer_quantized.cl:31</a></div></div>
+<div class="ttc" id="softmax__layer__quantized_8cl_xhtml_a54aedfa17c5ac2567107d5f488b0f4af"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#a54aedfa17c5ac2567107d5f488b0f4af">asymm_exp_on_negative_values</a></div><div class="ttdeci">#define asymm_exp_on_negative_values(a, k_integer_bits)</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00057">softmax_layer_quantized.cl:57</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="namespacearm__compute_1_1test_1_1fixed__point__arithmetic_1_1detail_xhtml_aa43c67a536c661ec28398d06bc8ba5af"><div class="ttname"><a href="namespacearm__compute_1_1test_1_1fixed__point__arithmetic_1_1detail.xhtml#aa43c67a536c661ec28398d06bc8ba5af">arm_compute::test::fixed_point_arithmetic::detail::clz</a></div><div class="ttdeci">constexpr int clz(T value)</div><div class="ttdoc">Count the number of leading zero bits in the given value. </div><div class="ttdef"><b>Definition:</b> <a href="tests_2validation_2_fixed_point_8h_source.xhtml#l00276">FixedPoint.h:276</a></div></div>
-<div class="ttc" id="namespacearm__compute_1_1test_1_1fixed__point__arithmetic_1_1detail_xhtml_ad91bb73431b4de1f4946ed949d444849"><div class="ttname"><a href="namespacearm__compute_1_1test_1_1fixed__point__arithmetic_1_1detail.xhtml#ad91bb73431b4de1f4946ed949d444849">arm_compute::test::fixed_point_arithmetic::detail::max</a></div><div class="ttdeci">fixed_point< T > max(fixed_point< T > x, fixed_point< T > y)</div><div class="ttdef"><b>Definition:</b> <a href="tests_2validation_2_fixed_point_8h_source.xhtml#l00889">FixedPoint.h:889</a></div></div>
-<div class="ttc" id="asymm__helper_8h_xhtml"><div class="ttname"><a href="asymm__helper_8h.xhtml">asymm_helper.h</a></div></div>
-<div class="ttc" id="softmax__layer__quantized_8cl_xhtml_a0712735973f172ac9efc7d48a31e47ad"><div class="ttname"><a href="softmax__layer__quantized_8cl.xhtml#a0712735973f172ac9efc7d48a31e47ad">idx16</a></div><div class="ttdeci">__constant uint16 idx16</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer__quantized_8cl_source.xhtml#l00031">softmax_layer_quantized.cl:31</a></div></div>
+<div class="ttc" id="helpers__asymm_8h_xhtml_ae77f34e1316d52c1ee84c35be9efb0d8"><div class="ttname"><a href="helpers__asymm_8h.xhtml#ae77f34e1316d52c1ee84c35be9efb0d8">ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1</a></div><div class="ttdeci">#define ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1(a, size)</div><div class="ttdef"><b>Definition:</b> <a href="helpers__asymm_8h_source.xhtml#l00304">helpers_asymm.h:304</a></div></div>
+<div class="ttc" id="helpers__asymm_8h_xhtml_a3a4f1b5d8f1cd67ac31bc62c9a6f4aa8"><div class="ttname"><a href="helpers__asymm_8h.xhtml#a3a4f1b5d8f1cd67ac31bc62c9a6f4aa8">ASYMM_EXP_ON_NEGATIVE_VALUES</a></div><div class="ttdeci">#define ASYMM_EXP_ON_NEGATIVE_VALUES(a, k_integer_bits, size)</div><div class="ttdef"><b>Definition:</b> <a href="helpers__asymm_8h_source.xhtml#l00303">helpers_asymm.h:303</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><!-- fragment --></div><!-- contents -->
</div><!-- doc-content -->
<!-- start footer part -->
<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="softmax__layer__quantized_8cl.xhtml">softmax_layer_quantized.cl</a></li>
- <li class="footer">Generated on Thu Dec 14 2017 23:48:32 for Compute Library by
+ <li class="footer">Generated on Thu Feb 22 2018 15:45:19 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>
projects / platform / upstream / armcl.git / blobdiff