| <a href="_n_e_g_e_m_m_lowp_offset_contribution_output_stage_kernel_8cpp.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) 2019-2020 ARM Limited.</span></div><div class="line"><a name="l00003"></a><span class="lineno"> 3</span> <span class="comment"> *</span></div><div class="line"><a name="l00004"></a><span class="lineno"> 4</span> <span class="comment"> * SPDX-License-Identifier: MIT</span></div><div class="line"><a name="l00005"></a><span class="lineno"> 5</span> <span class="comment"> *</span></div><div class="line"><a name="l00006"></a><span class="lineno"> 6</span> <span class="comment"> * Permission is hereby granted, free of charge, to any person obtaining a copy</span></div><div class="line"><a name="l00007"></a><span class="lineno"> 7</span> <span class="comment"> * of this software and associated documentation files (the "Software"), to</span></div><div class="line"><a name="l00008"></a><span class="lineno"> 8</span> <span class="comment"> * deal in the Software without restriction, including without limitation the</span></div><div class="line"><a name="l00009"></a><span class="lineno"> 9</span> <span class="comment"> * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or</span></div><div class="line"><a name="l00010"></a><span class="lineno"> 10</span> <span class="comment"> * sell copies of the Software, and to permit persons to whom the Software is</span></div><div class="line"><a name="l00011"></a><span class="lineno"> 11</span> <span class="comment"> * furnished to do so, subject to the following conditions:</span></div><div class="line"><a name="l00012"></a><span class="lineno"> 12</span> <span class="comment"> *</span></div><div class="line"><a name="l00013"></a><span class="lineno"> 13</span> <span class="comment"> * The above copyright notice and this permission notice shall be included in all</span></div><div class="line"><a name="l00014"></a><span class="lineno"> 14</span> <span class="comment"> * copies or substantial portions of the Software.</span></div><div class="line"><a name="l00015"></a><span class="lineno"> 15</span> <span class="comment"> *</span></div><div class="line"><a name="l00016"></a><span class="lineno"> 16</span> <span class="comment"> * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR</span></div><div class="line"><a name="l00017"></a><span class="lineno"> 17</span> <span class="comment"> * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,</span></div><div class="line"><a name="l00018"></a><span class="lineno"> 18</span> <span class="comment"> * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE</span></div><div class="line"><a name="l00019"></a><span class="lineno"> 19</span> <span class="comment"> * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER</span></div><div class="line"><a name="l00020"></a><span class="lineno"> 20</span> <span class="comment"> * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,</span></div><div class="line"><a name="l00021"></a><span class="lineno"> 21</span> <span class="comment"> * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE</span></div><div class="line"><a name="l00022"></a><span class="lineno"> 22</span> <span class="comment"> * SOFTWARE.</span></div><div class="line"><a name="l00023"></a><span class="lineno"> 23</span> <span class="comment"> */</span></div><div class="line"><a name="l00024"></a><span class="lineno"> 24</span> <span class="preprocessor">#include "<a class="code" href="_n_e_g_e_m_m_lowp_offset_contribution_output_stage_kernel_8h.xhtml">arm_compute/core/NEON/kernels/NEGEMMLowpOffsetContributionOutputStageKernel.h</a>"</span></div><div class="line"><a name="l00025"></a><span class="lineno"> 25</span> </div><div class="line"><a name="l00026"></a><span class="lineno"> 26</span> <span class="preprocessor">#include "<a class="code" href="_access_window_static_8h.xhtml">arm_compute/core/AccessWindowStatic.h</a>"</span></div><div class="line"><a name="l00027"></a><span class="lineno"> 27</span> <span class="preprocessor">#include "<a class="code" href="_error_8h.xhtml">arm_compute/core/Error.h</a>"</span></div><div class="line"><a name="l00028"></a><span class="lineno"> 28</span> <span class="preprocessor">#include "<a class="code" href="arm__compute_2core_2_helpers_8h.xhtml">arm_compute/core/Helpers.h</a>"</span></div><div class="line"><a name="l00029"></a><span class="lineno"> 29</span> <span class="preprocessor">#include "<a class="code" href="_i_tensor_8h.xhtml">arm_compute/core/ITensor.h</a>"</span></div><div class="line"><a name="l00030"></a><span class="lineno"> 30</span> <span class="preprocessor">#include "<a class="code" href="_n_e_asymm_8h.xhtml">arm_compute/core/NEON/NEAsymm.h</a>"</span></div><div class="line"><a name="l00031"></a><span class="lineno"> 31</span> <span class="preprocessor">#include "<a class="code" href="wrapper_8h.xhtml">arm_compute/core/NEON/wrapper/wrapper.h</a>"</span></div><div class="line"><a name="l00032"></a><span class="lineno"> 32</span> <span class="preprocessor">#include "<a class="code" href="_tensor_info_8h.xhtml">arm_compute/core/TensorInfo.h</a>"</span></div><div class="line"><a name="l00033"></a><span class="lineno"> 33</span> <span class="preprocessor">#include "<a class="code" href="arm__compute_2core_2_types_8h.xhtml">arm_compute/core/Types.h</a>"</span></div><div class="line"><a name="l00034"></a><span class="lineno"> 34</span> <span class="preprocessor">#include "<a class="code" href="arm__compute_2core_2_utils_8h.xhtml">arm_compute/core/Utils.h</a>"</span></div><div class="line"><a name="l00035"></a><span class="lineno"> 35</span> <span class="preprocessor">#include "<a class="code" href="_validate_8h.xhtml">arm_compute/core/Validate.h</a>"</span></div><div class="line"><a name="l00036"></a><span class="lineno"> 36</span> <span class="preprocessor">#include "<a class="code" href="_window_8h.xhtml">arm_compute/core/Window.h</a>"</span></div><div class="line"><a name="l00037"></a><span class="lineno"> 37</span> </div><div class="line"><a name="l00038"></a><span class="lineno"> 38</span> <span class="preprocessor">#include <arm_neon.h></span></div><div class="line"><a name="l00039"></a><span class="lineno"> 39</span> <span class="preprocessor">#include <cstddef></span></div><div class="line"><a name="l00040"></a><span class="lineno"> 40</span> <span class="preprocessor">#include <cstdint></span></div><div class="line"><a name="l00041"></a><span class="lineno"> 41</span> <span class="preprocessor">#include <map></span></div><div class="line"><a name="l00042"></a><span class="lineno"> 42</span> </div><div class="line"><a name="l00043"></a><span class="lineno"> 43</span> <span class="keyword">namespace </span><a class="code" href="namespacearm__compute.xhtml">arm_compute</a></div><div class="line"><a name="l00044"></a><span class="lineno"> 44</span> {</div><div class="line"><a name="l00045"></a><span class="lineno"> 45</span> <span class="keyword">class </span>Coordinates;</div><div class="line"><a name="l00046"></a><span class="lineno"> 46</span> </div><div class="line"><a name="l00047"></a><span class="lineno"> 47</span> <span class="keyword">namespace</span></div><div class="line"><a name="l00048"></a><span class="lineno"> 48</span> {</div><div class="line"><a name="l00049"></a><span class="lineno"> 49</span> <span class="keyword">inline</span> int32x4x4_t load_results_input(<span class="keyword">const</span> Iterator &mm_result_it, int32_t x)</div><div class="line"><a name="l00050"></a><span class="lineno"> 50</span> {</div><div class="line"><a name="l00051"></a><span class="lineno"> 51</span>  <span class="keywordflow">return</span></div><div class="line"><a name="l00052"></a><span class="lineno"> 52</span>  {</div><div class="line"><a name="l00053"></a><span class="lineno"> 53</span>  {</div><div class="line"><a name="l00054"></a><span class="lineno"> 54</span>  vld1q_s32(reinterpret_cast<const int32_t *>(mm_result_it.ptr()) + x + 0),</div><div class="line"><a name="l00055"></a><span class="lineno"> 55</span>  vld1q_s32(reinterpret_cast<const int32_t *>(mm_result_it.ptr()) + x + 4),</div><div class="line"><a name="l00056"></a><span class="lineno"> 56</span>  vld1q_s32(reinterpret_cast<const int32_t *>(mm_result_it.ptr()) + x + 8),</div><div class="line"><a name="l00057"></a><span class="lineno"> 57</span>  vld1q_s32(reinterpret_cast<const int32_t *>(mm_result_it.ptr()) + x + 12)</div><div class="line"><a name="l00058"></a><span class="lineno"> 58</span>  }</div><div class="line"><a name="l00059"></a><span class="lineno"> 59</span>  };</div><div class="line"><a name="l00060"></a><span class="lineno"> 60</span> }</div><div class="line"><a name="l00061"></a><span class="lineno"> 61</span> </div><div class="line"><a name="l00062"></a><span class="lineno"> 62</span> <span class="keyword">inline</span> int32x4x4_t load(<span class="keyword">const</span> int32_t *ptr, int32_t x)</div><div class="line"><a name="l00063"></a><span class="lineno"> 63</span> {</div><div class="line"><a name="l00064"></a><span class="lineno"> 64</span>  <span class="keywordflow">return</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>  {</div><div class="line"><a name="l00067"></a><span class="lineno"> 67</span>  vld1q_s32(ptr + x + 0),</div><div class="line"><a name="l00068"></a><span class="lineno"> 68</span>  vld1q_s32(ptr + x + 4),</div><div class="line"><a name="l00069"></a><span class="lineno"> 69</span>  vld1q_s32(ptr + x + 8),</div><div class="line"><a name="l00070"></a><span class="lineno"> 70</span>  vld1q_s32(ptr + x + 12)</div><div class="line"><a name="l00071"></a><span class="lineno"> 71</span>  }</div><div class="line"><a name="l00072"></a><span class="lineno"> 72</span>  };</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> </div><div class="line"><a name="l00075"></a><span class="lineno"> 75</span> <span class="keyword">inline</span> int32x4x4_t add_s32(int32x4x4_t a, int32x4_t <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>)</div><div class="line"><a name="l00076"></a><span class="lineno"> 76</span> {</div><div class="line"><a name="l00077"></a><span class="lineno"> 77</span>  <span class="keywordflow">return</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>  {</div><div class="line"><a name="l00080"></a><span class="lineno"> 80</span>  vaddq_s32(a.val[0], <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>),</div><div class="line"><a name="l00081"></a><span class="lineno"> 81</span>  vaddq_s32(a.val[1], <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>),</div><div class="line"><a name="l00082"></a><span class="lineno"> 82</span>  vaddq_s32(a.val[2], <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>),</div><div class="line"><a name="l00083"></a><span class="lineno"> 83</span>  vaddq_s32(a.val[3], <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>)</div><div class="line"><a name="l00084"></a><span class="lineno"> 84</span>  }</div><div class="line"><a name="l00085"></a><span class="lineno"> 85</span>  };</div><div class="line"><a name="l00086"></a><span class="lineno"> 86</span> }</div><div class="line"><a name="l00087"></a><span class="lineno"> 87</span> </div><div class="line"><a name="l00088"></a><span class="lineno"> 88</span> <span class="keyword">inline</span> int32x4x4_t add_s32(int32x4x4_t a, int32x4x4_t <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>)</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>  <span class="keywordflow">return</span></div><div class="line"><a name="l00091"></a><span class="lineno"> 91</span>  {</div><div class="line"><a name="l00092"></a><span class="lineno"> 92</span>  {</div><div class="line"><a name="l00093"></a><span class="lineno"> 93</span>  vaddq_s32(a.val[0], <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>.val[0]),</div><div class="line"><a name="l00094"></a><span class="lineno"> 94</span>  vaddq_s32(a.val[1], <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>.val[1]),</div><div class="line"><a name="l00095"></a><span class="lineno"> 95</span>  vaddq_s32(a.val[2], <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>.val[2]),</div><div class="line"><a name="l00096"></a><span class="lineno"> 96</span>  vaddq_s32(a.val[3], <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>.val[3])</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>  };</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> </div><div class="line"><a name="l00101"></a><span class="lineno"> 101</span> <span class="keyword">inline</span> int32x4x4_t mul_s32(int32x4x4_t &a, int32_t mul_scalar)</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>  <span class="keywordflow">return</span></div><div class="line"><a name="l00104"></a><span class="lineno"> 104</span>  {</div><div class="line"><a name="l00105"></a><span class="lineno"> 105</span>  {</div><div class="line"><a name="l00106"></a><span class="lineno"> 106</span>  vmulq_n_s32(a.val[0], mul_scalar),</div><div class="line"><a name="l00107"></a><span class="lineno"> 107</span>  vmulq_n_s32(a.val[1], mul_scalar),</div><div class="line"><a name="l00108"></a><span class="lineno"> 108</span>  vmulq_n_s32(a.val[2], mul_scalar),</div><div class="line"><a name="l00109"></a><span class="lineno"> 109</span>  vmulq_n_s32(a.val[3], mul_scalar)</div><div class="line"><a name="l00110"></a><span class="lineno"> 110</span>  }</div><div class="line"><a name="l00111"></a><span class="lineno"> 111</span>  };</div><div class="line"><a name="l00112"></a><span class="lineno"> 112</span> }</div><div class="line"><a name="l00113"></a><span class="lineno"> 113</span> </div><div class="line"><a name="l00114"></a><span class="lineno"> 114</span> <span class="keyword">inline</span> int32x4x4_t mul_s32(int32x4x4_t &a, <span class="keyword">const</span> int32_t *multilpier)</div><div class="line"><a name="l00115"></a><span class="lineno"> 115</span> {</div><div class="line"><a name="l00116"></a><span class="lineno"> 116</span>  <span class="keywordflow">return</span></div><div class="line"><a name="l00117"></a><span class="lineno"> 117</span>  {</div><div class="line"><a name="l00118"></a><span class="lineno"> 118</span>  {</div><div class="line"><a name="l00119"></a><span class="lineno"> 119</span>  vmulq_s32(a.val[0], vld1q_s32(multilpier)),</div><div class="line"><a name="l00120"></a><span class="lineno"> 120</span>  vmulq_s32(a.val[1], vld1q_s32(multilpier + 4)),</div><div class="line"><a name="l00121"></a><span class="lineno"> 121</span>  vmulq_s32(a.val[2], vld1q_s32(multilpier + 8)),</div><div class="line"><a name="l00122"></a><span class="lineno"> 122</span>  vmulq_s32(a.val[3], vld1q_s32(multilpier + 12))</div><div class="line"><a name="l00123"></a><span class="lineno"> 123</span>  }</div><div class="line"><a name="l00124"></a><span class="lineno"> 124</span>  };</div><div class="line"><a name="l00125"></a><span class="lineno"> 125</span> }</div><div class="line"><a name="l00126"></a><span class="lineno"> 126</span> </div><div class="line"><a name="l00127"></a><span class="lineno"> 127</span> <span class="keyword">inline</span> int32x4x4_t get_a_offset(<span class="keyword">const</span> int32_t *vector_sum_col_ptr, int32_t a_offset, int32_t x)</div><div class="line"><a name="l00128"></a><span class="lineno"> 128</span> {</div><div class="line"><a name="l00129"></a><span class="lineno"> 129</span>  int32x4x4_t a_offset_term_s32 = load(vector_sum_col_ptr, x);</div><div class="line"><a name="l00130"></a><span class="lineno"> 130</span> </div><div class="line"><a name="l00131"></a><span class="lineno"> 131</span>  a_offset_term_s32.val[0] = vmulq_n_s32(a_offset_term_s32.val[0], a_offset);</div><div class="line"><a name="l00132"></a><span class="lineno"> 132</span>  a_offset_term_s32.val[1] = vmulq_n_s32(a_offset_term_s32.val[1], a_offset);</div><div class="line"><a name="l00133"></a><span class="lineno"> 133</span>  a_offset_term_s32.val[2] = vmulq_n_s32(a_offset_term_s32.val[2], a_offset);</div><div class="line"><a name="l00134"></a><span class="lineno"> 134</span>  a_offset_term_s32.val[3] = vmulq_n_s32(a_offset_term_s32.val[3], a_offset);</div><div class="line"><a name="l00135"></a><span class="lineno"> 135</span>  <span class="keywordflow">return</span> a_offset_term_s32;</div><div class="line"><a name="l00136"></a><span class="lineno"> 136</span> }</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> <span class="keyword">inline</span> int32x4_t get_b_offset(<span class="keyword">const</span> int32_t *vector_sum_row_ptr, int32_t b_offset)</div><div class="line"><a name="l00139"></a><span class="lineno"> 139</span> {</div><div class="line"><a name="l00140"></a><span class="lineno"> 140</span>  int32x4_t b_offset_term_s32 = vld1q_dup_s32(vector_sum_row_ptr);</div><div class="line"><a name="l00141"></a><span class="lineno"> 141</span>  b_offset_term_s32 = vmulq_n_s32(b_offset_term_s32, b_offset);</div><div class="line"><a name="l00142"></a><span class="lineno"> 142</span>  <span class="keywordflow">return</span> b_offset_term_s32;</div><div class="line"><a name="l00143"></a><span class="lineno"> 143</span> }</div><div class="line"><a name="l00144"></a><span class="lineno"> 144</span> </div><div class="line"><a name="l00145"></a><span class="lineno"> 145</span> <span class="keyword">inline</span> int32x4x4_t get_k_offset(int32_t k_offset)</div><div class="line"><a name="l00146"></a><span class="lineno"> 146</span> {</div><div class="line"><a name="l00147"></a><span class="lineno"> 147</span>  <span class="keywordflow">return</span></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>  {</div><div class="line"><a name="l00150"></a><span class="lineno"> 150</span>  vdupq_n_s32(k_offset),</div><div class="line"><a name="l00151"></a><span class="lineno"> 151</span>  vdupq_n_s32(k_offset),</div><div class="line"><a name="l00152"></a><span class="lineno"> 152</span>  vdupq_n_s32(k_offset),</div><div class="line"><a name="l00153"></a><span class="lineno"> 153</span>  vdupq_n_s32(k_offset)</div><div class="line"><a name="l00154"></a><span class="lineno"> 154</span>  }</div><div class="line"><a name="l00155"></a><span class="lineno"> 155</span>  };</div><div class="line"><a name="l00156"></a><span class="lineno"> 156</span> }</div><div class="line"><a name="l00157"></a><span class="lineno"> 157</span> </div><div class="line"><a name="l00158"></a><span class="lineno"> 158</span> <span class="keyword">template</span> <<span class="keywordtype">bool</span> is_bounded_relu></div><div class="line"><a name="l00159"></a><span class="lineno"> 159</span> <span class="keyword">inline</span> uint8x16_t finalize_quantization_floating_point(int32x4x4_t &in_s32, int32x4_t result_shift_s32, uint8x16_t min_u8, uint8x16_t max_u8)</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="keyword">const</span> <span class="keyword">static</span> int32x4_t zero_s32 = vdupq_n_s32(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 final result (negative value shift right)</span></div><div class="line"><a name="l00164"></a><span class="lineno"> 164</span>  in_s32.val[0] = vshlq_s32(in_s32.val[0], result_shift_s32);</div><div class="line"><a name="l00165"></a><span class="lineno"> 165</span>  in_s32.val[1] = vshlq_s32(in_s32.val[1], result_shift_s32);</div><div class="line"><a name="l00166"></a><span class="lineno"> 166</span>  in_s32.val[2] = vshlq_s32(in_s32.val[2], result_shift_s32);</div><div class="line"><a name="l00167"></a><span class="lineno"> 167</span>  in_s32.val[3] = vshlq_s32(in_s32.val[3], result_shift_s32);</div><div class="line"><a name="l00168"></a><span class="lineno"> 168</span> </div><div class="line"><a name="l00169"></a><span class="lineno"> 169</span>  <span class="comment">// Saturate negative values</span></div><div class="line"><a name="l00170"></a><span class="lineno"> 170</span>  in_s32.val[0] = vmaxq_s32(in_s32.val[0], zero_s32);</div><div class="line"><a name="l00171"></a><span class="lineno"> 171</span>  in_s32.val[1] = vmaxq_s32(in_s32.val[1], zero_s32);</div><div class="line"><a name="l00172"></a><span class="lineno"> 172</span>  in_s32.val[2] = vmaxq_s32(in_s32.val[2], zero_s32);</div><div class="line"><a name="l00173"></a><span class="lineno"> 173</span>  in_s32.val[3] = vmaxq_s32(in_s32.val[3], zero_s32);</div><div class="line"><a name="l00174"></a><span class="lineno"> 174</span> </div><div class="line"><a name="l00175"></a><span class="lineno"> 175</span>  <span class="comment">// Convert S32 to S16</span></div><div class="line"><a name="l00176"></a><span class="lineno"> 176</span>  <span class="keyword">const</span> int16x8x2_t in_s16 =</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>  {</div><div class="line"><a name="l00179"></a><span class="lineno"> 179</span>  vcombine_s16(vqmovn_s32(in_s32.val[0]), vqmovn_s32(in_s32.val[1])),</div><div class="line"><a name="l00180"></a><span class="lineno"> 180</span>  vcombine_s16(vqmovn_s32(in_s32.val[2]), vqmovn_s32(in_s32.val[3]))</div><div class="line"><a name="l00181"></a><span class="lineno"> 181</span>  }</div><div class="line"><a name="l00182"></a><span class="lineno"> 182</span>  };</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">// Convert S16 to U8</span></div><div class="line"><a name="l00185"></a><span class="lineno"> 185</span>  uint8x16_t out_u8 = vcombine_u8(vqmovun_s16(in_s16.val[0]), vqmovun_s16(in_s16.val[1]));</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>  <span class="keywordflow">if</span>(is_bounded_relu)</div><div class="line"><a name="l00188"></a><span class="lineno"> 188</span>  {</div><div class="line"><a name="l00189"></a><span class="lineno"> 189</span>  out_u8 = vmaxq_u8(out_u8, min_u8);</div><div class="line"><a name="l00190"></a><span class="lineno"> 190</span>  out_u8 = vminq_u8(out_u8, max_u8);</div><div class="line"><a name="l00191"></a><span class="lineno"> 191</span>  }</div><div class="line"><a name="l00192"></a><span class="lineno"> 192</span> </div><div class="line"><a name="l00193"></a><span class="lineno"> 193</span>  <span class="keywordflow">return</span> out_u8;</div><div class="line"><a name="l00194"></a><span class="lineno"> 194</span> }</div><div class="line"><a name="l00195"></a><span class="lineno"> 195</span> </div><div class="line"><a name="l00196"></a><span class="lineno"> 196</span> <span class="keyword">template</span> <<span class="keywordtype">bool</span> is_bounded_relu></div><div class="line"><a name="l00197"></a><span class="lineno"> 197</span> <span class="keyword">inline</span> int8x16_t finalize_quantization_floating_point(int32x4x4_t &in_s32, int32x4_t result_shift_s32, int8x16_t min_s8, int8x16_t max_s8)</div><div class="line"><a name="l00198"></a><span class="lineno"> 198</span> {</div><div class="line"><a name="l00199"></a><span class="lineno"> 199</span>  <span class="keyword">const</span> <span class="keyword">static</span> int32x4_t zero_s32 = vdupq_n_s32(0);</div><div class="line"><a name="l00200"></a><span class="lineno"> 200</span> </div><div class="line"><a name="l00201"></a><span class="lineno"> 201</span>  <span class="comment">// Shift final result (negative value shift right)</span></div><div class="line"><a name="l00202"></a><span class="lineno"> 202</span>  in_s32.val[0] = vshlq_s32(in_s32.val[0], result_shift_s32);</div><div class="line"><a name="l00203"></a><span class="lineno"> 203</span>  in_s32.val[1] = vshlq_s32(in_s32.val[1], result_shift_s32);</div><div class="line"><a name="l00204"></a><span class="lineno"> 204</span>  in_s32.val[2] = vshlq_s32(in_s32.val[2], result_shift_s32);</div><div class="line"><a name="l00205"></a><span class="lineno"> 205</span>  in_s32.val[3] = vshlq_s32(in_s32.val[3], result_shift_s32);</div><div class="line"><a name="l00206"></a><span class="lineno"> 206</span> </div><div class="line"><a name="l00207"></a><span class="lineno"> 207</span>  <span class="comment">// Saturate negative values</span></div><div class="line"><a name="l00208"></a><span class="lineno"> 208</span>  in_s32.val[0] = vmaxq_s32(in_s32.val[0], zero_s32);</div><div class="line"><a name="l00209"></a><span class="lineno"> 209</span>  in_s32.val[1] = vmaxq_s32(in_s32.val[1], zero_s32);</div><div class="line"><a name="l00210"></a><span class="lineno"> 210</span>  in_s32.val[2] = vmaxq_s32(in_s32.val[2], zero_s32);</div><div class="line"><a name="l00211"></a><span class="lineno"> 211</span>  in_s32.val[3] = vmaxq_s32(in_s32.val[3], zero_s32);</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">// Convert S32 to S16</span></div><div class="line"><a name="l00214"></a><span class="lineno"> 214</span>  <span class="keyword">const</span> int16x8x2_t in_s16 =</div><div class="line"><a name="l00215"></a><span class="lineno"> 215</span>  {</div><div class="line"><a name="l00216"></a><span class="lineno"> 216</span>  {</div><div class="line"><a name="l00217"></a><span class="lineno"> 217</span>  vcombine_s16(vqmovn_s32(in_s32.val[0]), vqmovn_s32(in_s32.val[1])),</div><div class="line"><a name="l00218"></a><span class="lineno"> 218</span>  vcombine_s16(vqmovn_s32(in_s32.val[2]), vqmovn_s32(in_s32.val[3]))</div><div class="line"><a name="l00219"></a><span class="lineno"> 219</span>  }</div><div class="line"><a name="l00220"></a><span class="lineno"> 220</span>  };</div><div class="line"><a name="l00221"></a><span class="lineno"> 221</span> </div><div class="line"><a name="l00222"></a><span class="lineno"> 222</span>  <span class="comment">// Convert S16 to S8</span></div><div class="line"><a name="l00223"></a><span class="lineno"> 223</span>  int8x16_t out_s8 = vcombine_s8(vqmovn_s16(in_s16.val[0]), vqmovn_s16(in_s16.val[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="keywordflow">if</span>(is_bounded_relu)</div><div class="line"><a name="l00226"></a><span class="lineno"> 226</span>  {</div><div class="line"><a name="l00227"></a><span class="lineno"> 227</span>  out_s8 = vmaxq_s8(out_s8, min_s8);</div><div class="line"><a name="l00228"></a><span class="lineno"> 228</span>  out_s8 = vminq_s8(out_s8, max_s8);</div><div class="line"><a name="l00229"></a><span class="lineno"> 229</span>  }</div><div class="line"><a name="l00230"></a><span class="lineno"> 230</span> </div><div class="line"><a name="l00231"></a><span class="lineno"> 231</span>  <span class="keywordflow">return</span> out_s8;</div><div class="line"><a name="l00232"></a><span class="lineno"> 232</span> }</div><div class="line"><a name="l00233"></a><span class="lineno"> 233</span> </div><div class="line"><a name="l00234"></a><span class="lineno"> 234</span> <span class="keyword">template</span> <<span class="keywordtype">bool</span> is_bounded_relu></div><div class="line"><a name="l00235"></a><span class="lineno"> 235</span> <span class="keyword">inline</span> int8x16_t finalize_quantization_floating_point(int32x4x4_t &in_s32, int32x4x4_t result_shift_s32, int8x16_t min_s8, int8x16_t max_s8)</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>  <span class="keyword">const</span> <span class="keyword">static</span> int32x4_t zero_s32 = vdupq_n_s32(0);</div><div class="line"><a name="l00238"></a><span class="lineno"> 238</span> </div><div class="line"><a name="l00239"></a><span class="lineno"> 239</span>  <span class="comment">// Shift final result (negative value shift right)</span></div><div class="line"><a name="l00240"></a><span class="lineno"> 240</span>  in_s32.val[0] = vshlq_s32(in_s32.val[0], vnegq_s32(result_shift_s32.val[0]));</div><div class="line"><a name="l00241"></a><span class="lineno"> 241</span>  in_s32.val[1] = vshlq_s32(in_s32.val[1], vnegq_s32(result_shift_s32.val[1]));</div><div class="line"><a name="l00242"></a><span class="lineno"> 242</span>  in_s32.val[2] = vshlq_s32(in_s32.val[2], vnegq_s32(result_shift_s32.val[2]));</div><div class="line"><a name="l00243"></a><span class="lineno"> 243</span>  in_s32.val[3] = vshlq_s32(in_s32.val[3], vnegq_s32(result_shift_s32.val[3]));</div><div class="line"><a name="l00244"></a><span class="lineno"> 244</span> </div><div class="line"><a name="l00245"></a><span class="lineno"> 245</span>  <span class="comment">// Saturate negative values</span></div><div class="line"><a name="l00246"></a><span class="lineno"> 246</span>  in_s32.val[0] = vmaxq_s32(in_s32.val[0], zero_s32);</div><div class="line"><a name="l00247"></a><span class="lineno"> 247</span>  in_s32.val[1] = vmaxq_s32(in_s32.val[1], zero_s32);</div><div class="line"><a name="l00248"></a><span class="lineno"> 248</span>  in_s32.val[2] = vmaxq_s32(in_s32.val[2], zero_s32);</div><div class="line"><a name="l00249"></a><span class="lineno"> 249</span>  in_s32.val[3] = vmaxq_s32(in_s32.val[3], zero_s32);</div><div class="line"><a name="l00250"></a><span class="lineno"> 250</span> </div><div class="line"><a name="l00251"></a><span class="lineno"> 251</span>  <span class="comment">// Convert S32 to S16</span></div><div class="line"><a name="l00252"></a><span class="lineno"> 252</span>  <span class="keyword">const</span> int16x8x2_t in_s16 =</div><div class="line"><a name="l00253"></a><span class="lineno"> 253</span>  {</div><div class="line"><a name="l00254"></a><span class="lineno"> 254</span>  {</div><div class="line"><a name="l00255"></a><span class="lineno"> 255</span>  vcombine_s16(vqmovn_s32(in_s32.val[0]), vqmovn_s32(in_s32.val[1])),</div><div class="line"><a name="l00256"></a><span class="lineno"> 256</span>  vcombine_s16(vqmovn_s32(in_s32.val[2]), vqmovn_s32(in_s32.val[3]))</div><div class="line"><a name="l00257"></a><span class="lineno"> 257</span>  }</div><div class="line"><a name="l00258"></a><span class="lineno"> 258</span>  };</div><div class="line"><a name="l00259"></a><span class="lineno"> 259</span> </div><div class="line"><a name="l00260"></a><span class="lineno"> 260</span>  <span class="comment">// Convert S16 to S8</span></div><div class="line"><a name="l00261"></a><span class="lineno"> 261</span>  int8x16_t out_s8 = vcombine_s8(vqmovn_s16(in_s16.val[0]), vqmovn_s16(in_s16.val[1]));</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="keywordflow">if</span>(is_bounded_relu)</div><div class="line"><a name="l00264"></a><span class="lineno"> 264</span>  {</div><div class="line"><a name="l00265"></a><span class="lineno"> 265</span>  out_s8 = vmaxq_s8(out_s8, min_s8);</div><div class="line"><a name="l00266"></a><span class="lineno"> 266</span>  out_s8 = vminq_s8(out_s8, max_s8);</div><div class="line"><a name="l00267"></a><span class="lineno"> 267</span>  }</div><div class="line"><a name="l00268"></a><span class="lineno"> 268</span> </div><div class="line"><a name="l00269"></a><span class="lineno"> 269</span>  <span class="keywordflow">return</span> out_s8;</div><div class="line"><a name="l00270"></a><span class="lineno"> 270</span> }</div><div class="line"><a name="l00271"></a><span class="lineno"> 271</span> </div><div class="line"><a name="l00272"></a><span class="lineno"> 272</span> <span class="keyword">template</span> <<span class="keyword">typename</span> T></div><div class="line"><a name="l00273"></a><span class="lineno"> 273</span> <span class="keyword">struct </span>VectorTyper</div><div class="line"><a name="l00274"></a><span class="lineno"> 274</span> {</div><div class="line"><a name="l00275"></a><span class="lineno"> 275</span>  <span class="keyword">using</span> stype = T;</div><div class="line"><a name="l00276"></a><span class="lineno"> 276</span>  <span class="keyword">using</span> vtype = <span class="keyword">typename</span> wrapper::traits::neon_bitvector_t<T, wrapper::traits::BitWidth::W128>;</div><div class="line"><a name="l00277"></a><span class="lineno"> 277</span> };</div><div class="line"><a name="l00278"></a><span class="lineno"> 278</span> </div><div class="line"><a name="l00279"></a><span class="lineno"> 279</span> <span class="keyword">inline</span> Window get_win_vector_sum(<span class="keyword">const</span> Window &window)</div><div class="line"><a name="l00280"></a><span class="lineno"> 280</span> {</div><div class="line"><a name="l00281"></a><span class="lineno"> 281</span>  Window win_vector_sum(window);</div><div class="line"><a name="l00282"></a><span class="lineno"> 282</span>  win_vector_sum.set(<a class="code" href="classarm__compute_1_1_window.xhtml#ad2d402364fa822b0b7775081291eeca9">Window::DimY</a>, Window::Dimension(0, 0, 0));</div><div class="line"><a name="l00283"></a><span class="lineno"> 283</span>  win_vector_sum.set(<a class="code" href="classarm__compute_1_1_window.xhtml#a893d17b56b9abc4423ce26e9a24ac5dc">Window::DimZ</a>, Window::Dimension(0, 0, 0));</div><div class="line"><a name="l00284"></a><span class="lineno"> 284</span>  <span class="keywordflow">return</span> win_vector_sum;</div><div class="line"><a name="l00285"></a><span class="lineno"> 285</span> }</div><div class="line"><a name="l00286"></a><span class="lineno"> 286</span> </div><div class="line"><a name="l00287"></a><span class="lineno"> 287</span> <span class="keyword">inline</span> Iterator get_vector_sum_col_it(<span class="keyword">const</span> Window &window, <span class="keyword">const</span> ITensor *vector_sum_col)</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>  Iterator vector_sum_col_it(vector_sum_col, get_win_vector_sum(window));</div><div class="line"><a name="l00290"></a><span class="lineno"> 290</span>  <span class="keywordflow">return</span> vector_sum_col_it;</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> </div><div class="line"><a name="l00293"></a><span class="lineno"> 293</span> <span class="keyword">inline</span> Iterator get_vector_sum_row_it(<span class="keyword">const</span> Window &window, <span class="keyword">const</span> ITensor *vector_sum_row)</div><div class="line"><a name="l00294"></a><span class="lineno"> 294</span> {</div><div class="line"><a name="l00295"></a><span class="lineno"> 295</span>  Window win_vector_sum_row = get_win_vector_sum(window);</div><div class="line"><a name="l00296"></a><span class="lineno"> 296</span>  win_vector_sum_row.set(<a class="code" href="classarm__compute_1_1_window.xhtml#aa96e81276ee4f87ab386cd05a5539a7d">Window::DimX</a>, Window::Dimension(0, 0, 0));</div><div class="line"><a name="l00297"></a><span class="lineno"> 297</span>  Iterator vector_sum_row_it(vector_sum_row, win_vector_sum_row);</div><div class="line"><a name="l00298"></a><span class="lineno"> 298</span>  <span class="keywordflow">return</span> vector_sum_row_it;</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> </div><div class="line"><a name="l00301"></a><span class="lineno"> 301</span> <span class="keyword">inline</span> Iterator get_bias_it(<span class="keyword">const</span> Window &window, <span class="keyword">const</span> ITensor *<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>)</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>  Window win_bias(window);</div><div class="line"><a name="l00304"></a><span class="lineno"> 304</span>  win_bias.set(<a class="code" href="classarm__compute_1_1_window.xhtml#ad2d402364fa822b0b7775081291eeca9">Window::DimY</a>, Window::Dimension(0, 1, 1));</div><div class="line"><a name="l00305"></a><span class="lineno"> 305</span>  win_bias.set(<a class="code" href="classarm__compute_1_1_window.xhtml#a893d17b56b9abc4423ce26e9a24ac5dc">Window::DimZ</a>, Window::Dimension(0, 1, 1));</div><div class="line"><a name="l00306"></a><span class="lineno"> 306</span>  Iterator bias_it(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>, win_bias);</div><div class="line"><a name="l00307"></a><span class="lineno"> 307</span>  <span class="keywordflow">return</span> bias_it;</div><div class="line"><a name="l00308"></a><span class="lineno"> 308</span> }</div><div class="line"><a name="l00309"></a><span class="lineno"> 309</span> </div><div class="line"><a name="l00310"></a><span class="lineno"> 310</span> <span class="keyword">template</span> <<span class="keyword">typename</span> VT, <span class="keywordtype">bool</span> has_a_offset, <span class="keywordtype">bool</span> has_b_offset, <span class="keywordtype">bool</span> has_bias, <span class="keywordtype">bool</span> is_bounded_relu, <span class="keywordtype">bool</span> is_fixed_po<span class="keywordtype">int</span>></div><div class="line"><a name="l00311"></a><span class="lineno"> 311</span> <span class="keyword">inline</span> <span class="keywordtype">void</span> run_offset_contribution_output_stage_window(<span class="keyword">const</span> int32_t *vector_sum_col_ptr, <span class="keyword">const</span> int32_t *vector_sum_row_ptr, <span class="keyword">const</span> int32_t *bias_ptr, Iterator mm_result_it, Iterator out_it,</div><div class="line"><a name="l00312"></a><span class="lineno"> 312</span>  <span class="keyword">const</span> int32x4_t result_offset_s32, <span class="keyword">const</span> int32x4_t result_shift_s32,</div><div class="line"><a name="l00313"></a><span class="lineno"> 313</span>  <span class="keyword">typename</span> VT::vtype min_vec, <span class="keyword">typename</span> VT::vtype max_vec,</div><div class="line"><a name="l00314"></a><span class="lineno"> 314</span>  int32_t a_offset, int32_t b_offset, int32_t k_offset,</div><div class="line"><a name="l00315"></a><span class="lineno"> 315</span>  int32_t multiplier, int32_t <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a979a54caef6e77ce0259e427136847e8">shift</a>, int32_t <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>, int32_t min_bound, int32_t max_bound,</div><div class="line"><a name="l00316"></a><span class="lineno"> 316</span>  <span class="keywordtype">int</span> window_step_x, <span class="keywordtype">int</span> window_start_x, <span class="keywordtype">int</span> window_end_x)</div><div class="line"><a name="l00317"></a><span class="lineno"> 317</span> {</div><div class="line"><a name="l00318"></a><span class="lineno"> 318</span>  int32x4x4_t offset_term_s32 = { 0, 0, 0, 0 };</div><div class="line"><a name="l00319"></a><span class="lineno"> 319</span>  <span class="keywordflow">if</span>(!is_fixed_point)</div><div class="line"><a name="l00320"></a><span class="lineno"> 320</span>  {</div><div class="line"><a name="l00321"></a><span class="lineno"> 321</span>  <span class="comment">// Combine quantization offset with other offsets.</span></div><div class="line"><a name="l00322"></a><span class="lineno"> 322</span>  offset_term_s32 = add_s32(offset_term_s32, result_offset_s32);</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="keywordflow">if</span>(has_a_offset && has_b_offset)</div><div class="line"><a name="l00325"></a><span class="lineno"> 325</span>  {</div><div class="line"><a name="l00326"></a><span class="lineno"> 326</span>  offset_term_s32 = add_s32(offset_term_s32, get_k_offset(k_offset));</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>  <span class="keywordflow">if</span>(has_b_offset)</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>  offset_term_s32 = add_s32(offset_term_s32, get_b_offset(vector_sum_row_ptr, b_offset));</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> </div><div class="line"><a name="l00333"></a><span class="lineno"> 333</span>  <span class="keywordtype">int</span> x = window_start_x;</div><div class="line"><a name="l00334"></a><span class="lineno"> 334</span>  <span class="keywordflow">for</span>(; x <= (window_end_x - window_step_x); x += window_step_x)</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>  int32x4x4_t in_s32 = load_results_input(mm_result_it, x);</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>(has_a_offset)</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>  in_s32 = add_s32(in_s32, get_a_offset(vector_sum_col_ptr, a_offset, x));</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>  <span class="keywordflow">if</span>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a9aeced5a5128f60a31ea3e327a45ee21">has_bias</a>)</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>  in_s32 = add_s32(in_s32, load(bias_ptr, x));</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>(!is_fixed_point || has_b_offset)</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>  in_s32 = add_s32(in_s32, offset_term_s32);</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>  <span class="keywordflow">if</span>(!is_fixed_point)</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>  in_s32 = mul_s32(in_s32, multiplier);</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> </div><div class="line"><a name="l00355"></a><span class="lineno"> 355</span>  <span class="keywordflow">if</span>(is_fixed_point)</div><div class="line"><a name="l00356"></a><span class="lineno"> 356</span>  {</div><div class="line"><a name="l00357"></a><span class="lineno"> 357</span>  <a class="code" href="namespacearm__compute_1_1wrapper.xhtml#ae7943ea9c1f74dc72c62d4cc3966a459">wrapper::vstore</a>(reinterpret_cast<typename VT::stype *>(out_it.ptr() + x),</div><div class="line"><a name="l00358"></a><span class="lineno"> 358</span>  finalize_quantization<is_bounded_relu>(in_s32, multiplier, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a979a54caef6e77ce0259e427136847e8">shift</a>, result_offset_s32, min_vec, max_vec));</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>  <span class="keywordflow">else</span></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>  <a class="code" href="namespacearm__compute_1_1wrapper.xhtml#ae7943ea9c1f74dc72c62d4cc3966a459">wrapper::vstore</a>(reinterpret_cast<typename VT::stype *>(out_it.ptr() + x),</div><div class="line"><a name="l00363"></a><span class="lineno"> 363</span>  finalize_quantization_floating_point<is_bounded_relu>(in_s32, result_shift_s32, min_vec, max_vec));</div><div class="line"><a name="l00364"></a><span class="lineno"> 364</span>  }</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>  <span class="comment">// Compute left-over elements</span></div><div class="line"><a name="l00367"></a><span class="lineno"> 367</span>  <span class="keywordflow">for</span>(; x < window_end_x; ++x)</div><div class="line"><a name="l00368"></a><span class="lineno"> 368</span>  {</div><div class="line"><a name="l00369"></a><span class="lineno"> 369</span>  int32_t in_value = *(reinterpret_cast<const int32_t *>(mm_result_it.ptr()) + x) + <a class="code" href="namespacearm__compute_1_1wrapper.xhtml#aa16ace001ab8287faa46d6962f369219">wrapper::vgetlane</a>(offset_term_s32.val[0], 0);</div><div class="line"><a name="l00370"></a><span class="lineno"> 370</span> </div><div class="line"><a name="l00371"></a><span class="lineno"> 371</span>  <span class="keywordflow">if</span>(has_a_offset)</div><div class="line"><a name="l00372"></a><span class="lineno"> 372</span>  {</div><div class="line"><a name="l00373"></a><span class="lineno"> 373</span>  in_value += (*(vector_sum_col_ptr + x) * a_offset);</div><div class="line"><a name="l00374"></a><span class="lineno"> 374</span>  }</div><div class="line"><a name="l00375"></a><span class="lineno"> 375</span>  <span class="keywordflow">if</span>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a9aeced5a5128f60a31ea3e327a45ee21">has_bias</a>)</div><div class="line"><a name="l00376"></a><span class="lineno"> 376</span>  {</div><div class="line"><a name="l00377"></a><span class="lineno"> 377</span>  in_value += *(bias_ptr + x);</div><div class="line"><a name="l00378"></a><span class="lineno"> 378</span>  }</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>(is_fixed_point)</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>  <span class="comment">// Finalize and store the result</span></div><div class="line"><a name="l00383"></a><span class="lineno"> 383</span>  *reinterpret_cast<typename VT::stype *>(out_it.ptr() + x) = finalize_quantization<is_bounded_relu>(in_value, multiplier, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a979a54caef6e77ce0259e427136847e8">shift</a>, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>,</div><div class="line"><a name="l00384"></a><span class="lineno"> 384</span>  static_cast<typename VT::stype>(min_bound),</div><div class="line"><a name="l00385"></a><span class="lineno"> 385</span>  static_cast<typename VT::stype>(max_bound));</div><div class="line"><a name="l00386"></a><span class="lineno"> 386</span>  }</div><div class="line"><a name="l00387"></a><span class="lineno"> 387</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00388"></a><span class="lineno"> 388</span>  {</div><div class="line"><a name="l00389"></a><span class="lineno"> 389</span>  <span class="comment">// Finalize quantization</span></div><div class="line"><a name="l00390"></a><span class="lineno"> 390</span>  in_value = (in_value * multiplier) >> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a979a54caef6e77ce0259e427136847e8">shift</a>;</div><div class="line"><a name="l00391"></a><span class="lineno"> 391</span> </div><div class="line"><a name="l00392"></a><span class="lineno"> 392</span>  <span class="comment">// Bound and store the result</span></div><div class="line"><a name="l00393"></a><span class="lineno"> 393</span>  <span class="keywordflow">if</span>(is_bounded_relu)</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>  in_value = static_cast<typename VT::stype>(std::max<int32_t>(min_bound, std::min<int32_t>(max_bound, in_value)));</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>  *reinterpret_cast<typename VT::stype *>(out_it.ptr() + x) = static_cast<typename VT::stype>(std::max<int32_t>(static_cast<int32_t>(<a class="code" href="namespacearm__compute_1_1support_1_1cpp11.xhtml#a73e352c61baaf9c1178da2d30105b04e">std::numeric_limits<typename VT::stype>::lowest</a>()),</div><div class="line"><a name="l00398"></a><span class="lineno"> 398</span>  std::min<int32_t>(static_cast<int32_t>(std::numeric_limits<typename VT::stype>::max()), in_value)));</div><div class="line"><a name="l00399"></a><span class="lineno"> 399</span>  }</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> }</div><div class="line"><a name="l00402"></a><span class="lineno"> 402</span> </div><div class="line"><a name="l00403"></a><span class="lineno"> 403</span> <span class="keyword">template</span> <<span class="keywordtype">bool</span> has_a_offset, <span class="keywordtype">bool</span> has_bias, <span class="keywordtype">bool</span> is_bounded_relu, <span class="keywordtype">bool</span> is_fixed_po<span class="keywordtype">int</span>></div><div class="line"><a name="l00404"></a><span class="lineno"> 404</span> <span class="keyword">inline</span> <span class="keywordtype">void</span> run_offset_contribution_output_stage_window_symm(<span class="keyword">const</span> int32_t *vector_sum_col_ptr, <span class="keyword">const</span> int32_t *bias_ptr, Iterator mm_result_it, Iterator out_it,</div><div class="line"><a name="l00405"></a><span class="lineno"> 405</span>  <span class="keyword">const</span> int32_t *result_multipliers, <span class="keyword">const</span> int32_t *result_shifts,</div><div class="line"><a name="l00406"></a><span class="lineno"> 406</span>  <span class="keyword">const</span> int32x4_t result_offset, int8x16_t min_s8, int8x16_t max_s8,</div><div class="line"><a name="l00407"></a><span class="lineno"> 407</span>  int32_t a_offset, int32_t <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>, int32_t min_bound, int32_t max_bound,</div><div class="line"><a name="l00408"></a><span class="lineno"> 408</span>  <span class="keywordtype">int</span> window_step_x, <span class="keywordtype">int</span> window_start_x, <span class="keywordtype">int</span> window_end_x)</div><div class="line"><a name="l00409"></a><span class="lineno"> 409</span> {</div><div class="line"><a name="l00410"></a><span class="lineno"> 410</span>  int32x4x4_t offset_term_s32 = { 0, 0, 0, 0 };</div><div class="line"><a name="l00411"></a><span class="lineno"> 411</span>  <span class="keywordflow">if</span>(!is_fixed_point)</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="comment">// Combine quantization offset with other offsets.</span></div><div class="line"><a name="l00414"></a><span class="lineno"> 414</span>  offset_term_s32 = add_s32(offset_term_s32, result_offset);</div><div class="line"><a name="l00415"></a><span class="lineno"> 415</span>  }</div><div class="line"><a name="l00416"></a><span class="lineno"> 416</span> </div><div class="line"><a name="l00417"></a><span class="lineno"> 417</span>  <span class="keywordtype">int</span> x = window_start_x;</div><div class="line"><a name="l00418"></a><span class="lineno"> 418</span>  <span class="keywordflow">for</span>(; x <= (window_end_x - window_step_x); x += window_step_x)</div><div class="line"><a name="l00419"></a><span class="lineno"> 419</span>  {</div><div class="line"><a name="l00420"></a><span class="lineno"> 420</span>  int32x4x4_t in_s32 = load_results_input(mm_result_it, x);</div><div class="line"><a name="l00421"></a><span class="lineno"> 421</span> </div><div class="line"><a name="l00422"></a><span class="lineno"> 422</span>  <span class="keywordflow">if</span>(has_a_offset)</div><div class="line"><a name="l00423"></a><span class="lineno"> 423</span>  {</div><div class="line"><a name="l00424"></a><span class="lineno"> 424</span>  in_s32 = add_s32(in_s32, get_a_offset(vector_sum_col_ptr, a_offset, x));</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="keywordflow">if</span>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a9aeced5a5128f60a31ea3e327a45ee21">has_bias</a>)</div><div class="line"><a name="l00427"></a><span class="lineno"> 427</span>  {</div><div class="line"><a name="l00428"></a><span class="lineno"> 428</span>  in_s32 = add_s32(in_s32, load(bias_ptr, x));</div><div class="line"><a name="l00429"></a><span class="lineno"> 429</span>  }</div><div class="line"><a name="l00430"></a><span class="lineno"> 430</span>  <span class="keywordflow">if</span>(!is_fixed_point)</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>  in_s32 = add_s32(in_s32, offset_term_s32);</div><div class="line"><a name="l00433"></a><span class="lineno"> 433</span>  in_s32 = mul_s32(in_s32, result_multipliers + x);</div><div class="line"><a name="l00434"></a><span class="lineno"> 434</span>  }</div><div class="line"><a name="l00435"></a><span class="lineno"> 435</span> </div><div class="line"><a name="l00436"></a><span class="lineno"> 436</span>  <span class="keywordflow">if</span>(is_fixed_point)</div><div class="line"><a name="l00437"></a><span class="lineno"> 437</span>  {</div><div class="line"><a name="l00438"></a><span class="lineno"> 438</span>  vst1q_s8(reinterpret_cast<int8_t *>(out_it.ptr() + x), finalize_quantization_symm<is_bounded_relu>(in_s32, load(result_multipliers, x), load(result_shifts, x), result_offset, min_s8, max_s8));</div><div class="line"><a name="l00439"></a><span class="lineno"> 439</span>  }</div><div class="line"><a name="l00440"></a><span class="lineno"> 440</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00441"></a><span class="lineno"> 441</span>  {</div><div class="line"><a name="l00442"></a><span class="lineno"> 442</span>  vst1q_s8(reinterpret_cast<int8_t *>(out_it.ptr() + x), finalize_quantization_floating_point<is_bounded_relu>(in_s32, load(result_shifts, x), min_s8, max_s8));</div><div class="line"><a name="l00443"></a><span class="lineno"> 443</span>  }</div><div class="line"><a name="l00444"></a><span class="lineno"> 444</span>  }</div><div class="line"><a name="l00445"></a><span class="lineno"> 445</span>  <span class="comment">// Compute left-over elements</span></div><div class="line"><a name="l00446"></a><span class="lineno"> 446</span>  <span class="keywordflow">for</span>(; x < window_end_x; ++x)</div><div class="line"><a name="l00447"></a><span class="lineno"> 447</span>  {</div><div class="line"><a name="l00448"></a><span class="lineno"> 448</span>  int32_t in_value = *(reinterpret_cast<const int32_t *>(mm_result_it.ptr()) + x) + <a class="code" href="namespacearm__compute_1_1wrapper.xhtml#aa16ace001ab8287faa46d6962f369219">wrapper::vgetlane</a>(offset_term_s32.val[0], 0);</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>  <span class="keywordflow">if</span>(has_a_offset)</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>  in_value += (*(vector_sum_col_ptr + x) * a_offset);</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>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a9aeced5a5128f60a31ea3e327a45ee21">has_bias</a>)</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>  in_value += *(bias_ptr + x);</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> </div><div class="line"><a name="l00459"></a><span class="lineno"> 459</span>  <span class="keywordflow">if</span>(is_fixed_point)</div><div class="line"><a name="l00460"></a><span class="lineno"> 460</span>  {</div><div class="line"><a name="l00461"></a><span class="lineno"> 461</span>  <span class="comment">// Finalize and store the result</span></div><div class="line"><a name="l00462"></a><span class="lineno"> 462</span>  *(out_it.ptr() + x) = finalize_quantization<is_bounded_relu>(in_value, result_multipliers[x], result_shifts[x], <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>, static_cast<int8_t>(min_bound), static_cast<int8_t>(max_bound));</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>  <span class="keywordflow">else</span></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>  <span class="comment">// Finalize quantization</span></div><div class="line"><a name="l00467"></a><span class="lineno"> 467</span>  in_value = (in_value * result_multipliers[x]) >> (-result_shifts[x]);</div><div class="line"><a name="l00468"></a><span class="lineno"> 468</span> </div><div class="line"><a name="l00469"></a><span class="lineno"> 469</span>  <span class="comment">// Bound and store the result</span></div><div class="line"><a name="l00470"></a><span class="lineno"> 470</span>  <span class="keywordflow">if</span>(is_bounded_relu)</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>  in_value = static_cast<int8_t>(std::max<int32_t>(min_bound, std::min<int32_t>(max_bound, in_value)));</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>  *(out_it.ptr() + x) = static_cast<int8_t>(std::max<int32_t>(-128, std::min<int32_t>(127, in_value)));</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>  }</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> </div><div class="line"><a name="l00479"></a><span class="lineno"> 479</span> <span class="keyword">template</span> <<span class="keyword">typename</span> T, <span class="keywordtype">bool</span> is_gemm3d, <span class="keywordtype">bool</span> is_bounded_relu, <span class="keywordtype">bool</span> is_fixed_po<span class="keywordtype">int</span>></div><div class="line"><a name="l00480"></a><span class="lineno"> 480</span> <span class="keywordtype">void</span> run_offset_contribution_output_stage(<span class="keyword">const</span> Window &window,</div><div class="line"><a name="l00481"></a><span class="lineno"> 481</span>  <span class="keyword">const</span> ITensor *mm_result, <span class="keyword">const</span> ITensor *vector_sum_col, <span class="keyword">const</span> ITensor *vector_sum_row, <span class="keyword">const</span> ITensor *<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>, ITensor *output,</div><div class="line"><a name="l00482"></a><span class="lineno"> 482</span>  int32_t a_offset, int32_t b_offset, int32_t k_offset, <span class="keywordtype">bool</span> slide_vector_sum_col,</div><div class="line"><a name="l00483"></a><span class="lineno"> 483</span>  GEMMLowpOutputStageInfo output_stage)</div><div class="line"><a name="l00484"></a><span class="lineno"> 484</span> {</div><div class="line"><a name="l00485"></a><span class="lineno"> 485</span>  <span class="keyword">using</span> ExactTagType = <span class="keyword">typename</span> wrapper::traits::neon_bitvector_tag_t<T, wrapper::traits::BitWidth::W128>;</div><div class="line"><a name="l00486"></a><span class="lineno"> 486</span>  <span class="keyword">using</span> Typer = VectorTyper<T>;</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>  <span class="keyword">const</span> <span class="keywordtype">int</span> height_input = is_gemm3d ? mm_result->info()->dimension(1) : 0;</div><div class="line"><a name="l00489"></a><span class="lineno"> 489</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> depth_input = is_gemm3d ? mm_result->info()->dimension(2) : 1;</div><div class="line"><a name="l00490"></a><span class="lineno"> 490</span> </div><div class="line"><a name="l00491"></a><span class="lineno"> 491</span>  <span class="keyword">const</span> int32_t multiplier = output_stage.gemmlowp_multiplier;</div><div class="line"><a name="l00492"></a><span class="lineno"> 492</span>  <span class="keyword">const</span> int32_t <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a979a54caef6e77ce0259e427136847e8">shift</a> = output_stage.gemmlowp_shift;</div><div class="line"><a name="l00493"></a><span class="lineno"> 493</span>  <span class="keyword">const</span> int32_t <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a> = output_stage.gemmlowp_offset;</div><div class="line"><a name="l00494"></a><span class="lineno"> 494</span>  <span class="keyword">const</span> int32_t min_bound = output_stage.gemmlowp_min_bound;</div><div class="line"><a name="l00495"></a><span class="lineno"> 495</span>  <span class="keyword">const</span> int32_t max_bound = output_stage.gemmlowp_max_bound;</div><div class="line"><a name="l00496"></a><span class="lineno"> 496</span> </div><div class="line"><a name="l00497"></a><span class="lineno"> 497</span>  <span class="keyword">const</span> int32x4_t result_offset_s32 = vdupq_n_s32(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>);</div><div class="line"><a name="l00498"></a><span class="lineno"> 498</span>  <span class="keyword">const</span> int32x4_t result_shift_s32 = vdupq_n_s32(is_fixed_point ? <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a979a54caef6e77ce0259e427136847e8">shift</a> : -<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a979a54caef6e77ce0259e427136847e8">shift</a>);</div><div class="line"><a name="l00499"></a><span class="lineno"> 499</span>  <span class="keyword">const</span> <span class="keyword">auto</span> min_vec = <a class="code" href="namespacearm__compute_1_1wrapper.xhtml#a39e87435be178fba49b76f49426ef873">wrapper::vdup_n</a>(static_cast<T>(min_bound), ExactTagType{});</div><div class="line"><a name="l00500"></a><span class="lineno"> 500</span>  <span class="keyword">const</span> <span class="keyword">auto</span> max_vec = <a class="code" href="namespacearm__compute_1_1wrapper.xhtml#a39e87435be178fba49b76f49426ef873">wrapper::vdup_n</a>(static_cast<T>(max_bound), ExactTagType{});</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="keyword">const</span> <span class="keywordtype">int</span> window_step_x = 16;</div><div class="line"><a name="l00503"></a><span class="lineno"> 503</span>  <span class="keyword">const</span> <span class="keyword">auto</span> window_start_x = static_cast<int>(window.x().start());</div><div class="line"><a name="l00504"></a><span class="lineno"> 504</span>  <span class="keyword">const</span> <span class="keyword">auto</span> window_end_x = static_cast<int>(window.x().end());</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>  Window win(window);</div><div class="line"><a name="l00507"></a><span class="lineno"> 507</span>  win.set(<a class="code" href="classarm__compute_1_1_window.xhtml#aa96e81276ee4f87ab386cd05a5539a7d">Window::DimX</a>, Window::Dimension(0, 1, 1));</div><div class="line"><a name="l00508"></a><span class="lineno"> 508</span> </div><div class="line"><a name="l00509"></a><span class="lineno"> 509</span>  Window collapsed_window = win.collapse_if_possible(win, <a class="code" href="classarm__compute_1_1_window.xhtml#a893d17b56b9abc4423ce26e9a24ac5dc">Window::DimZ</a>);</div><div class="line"><a name="l00510"></a><span class="lineno"> 510</span> </div><div class="line"><a name="l00511"></a><span class="lineno"> 511</span>  Iterator mm_result_it(mm_result, win);</div><div class="line"><a name="l00512"></a><span class="lineno"> 512</span>  Iterator out_it(output, win);</div><div class="line"><a name="l00513"></a><span class="lineno"> 513</span> </div><div class="line"><a name="l00514"></a><span class="lineno"> 514</span>  <span class="keywordflow">if</span>((a_offset != 0) && (b_offset != 0))</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>  <a class="code" href="_validate_8h.xhtml#a921b705e9e3e0fe928928447869e62a5">ARM_COMPUTE_ERROR_ON_NULLPTR</a>(vector_sum_col);</div><div class="line"><a name="l00517"></a><span class="lineno"> 517</span>  <a class="code" href="_validate_8h.xhtml#a921b705e9e3e0fe928928447869e62a5">ARM_COMPUTE_ERROR_ON_NULLPTR</a>(vector_sum_row);</div><div class="line"><a name="l00518"></a><span class="lineno"> 518</span> </div><div class="line"><a name="l00519"></a><span class="lineno"> 519</span>  Iterator vector_sum_col_it = get_vector_sum_col_it(collapsed_window, vector_sum_col);</div><div class="line"><a name="l00520"></a><span class="lineno"> 520</span>  Iterator vector_sum_row_it = get_vector_sum_row_it(collapsed_window, vector_sum_row);</div><div class="line"><a name="l00521"></a><span class="lineno"> 521</span> </div><div class="line"><a name="l00522"></a><span class="lineno"> 522</span>  <span class="keyword">const</span> <span class="keywordtype">size_t</span> sum_row_stride_y = vector_sum_row->info()->strides_in_bytes().y();</div><div class="line"><a name="l00523"></a><span class="lineno"> 523</span> </div><div class="line"><a name="l00524"></a><span class="lineno"> 524</span>  <span class="comment">// Offset in case vector_sum_col is batched</span></div><div class="line"><a name="l00525"></a><span class="lineno"> 525</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> vector_sum_col_batch_offset = slide_vector_sum_col ? vector_sum_col->info()->strides_in_bytes().z() : 0;</div><div class="line"><a name="l00526"></a><span class="lineno"> 526</span> </div><div class="line"><a name="l00527"></a><span class="lineno"> 527</span>  <span class="keywordflow">if</span>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a> != <span class="keyword">nullptr</span>)</div><div class="line"><a name="l00528"></a><span class="lineno"> 528</span>  {</div><div class="line"><a name="l00529"></a><span class="lineno"> 529</span>  Iterator bias_it = get_bias_it(collapsed_window, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>);</div><div class="line"><a name="l00530"></a><span class="lineno"> 530</span>  <a class="code" href="namespacearm__compute.xhtml#a5002bf7ec46d52971f9526e94172cfee">execute_window_loop</a>(collapsed_window, [&](<span class="keyword">const</span> Coordinates & <span class="keywordtype">id</span>)</div><div class="line"><a name="l00531"></a><span class="lineno"> 531</span>  {</div><div class="line"><a name="l00532"></a><span class="lineno"> 532</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> batch_id = <span class="keywordtype">id</span>.z() / depth_input;</div><div class="line"><a name="l00533"></a><span class="lineno"> 533</span>  <span class="keyword">const</span> <span class="keyword">auto</span> vector_sum_col_ptr = reinterpret_cast<const int32_t *>(vector_sum_col_it.ptr() + batch_id * vector_sum_col_batch_offset);</div><div class="line"><a name="l00534"></a><span class="lineno"> 534</span>  <span class="keyword">const</span> <span class="keyword">auto</span> vector_sum_row_ptr = reinterpret_cast<const int32_t *>(vector_sum_row_it.ptr() + batch_id * sum_row_stride_y)</div><div class="line"><a name="l00535"></a><span class="lineno"> 535</span>  + <span class="keywordtype">id</span>.y() + (<span class="keywordtype">id</span>.z() % depth_input) * height_input;</div><div class="line"><a name="l00536"></a><span class="lineno"> 536</span>  run_offset_contribution_output_stage_window<Typer, true, true, true, is_bounded_relu, is_fixed_point>(vector_sum_col_ptr, vector_sum_row_ptr, reinterpret_cast<const int32_t *>(bias_it.ptr()),</div><div class="line"><a name="l00537"></a><span class="lineno"> 537</span>  mm_result_it,</div><div class="line"><a name="l00538"></a><span class="lineno"> 538</span>  out_it,</div><div class="line"><a name="l00539"></a><span class="lineno"> 539</span>  result_offset_s32, result_shift_s32,</div><div class="line"><a name="l00540"></a><span class="lineno"> 540</span>  min_vec, max_vec, a_offset, b_offset, k_offset,</div><div class="line"><a name="l00541"></a><span class="lineno"> 541</span>  multiplier, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a979a54caef6e77ce0259e427136847e8">shift</a>, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>, min_bound, max_bound,</div><div class="line"><a name="l00542"></a><span class="lineno"> 542</span>  window_step_x, window_start_x, window_end_x);</div><div class="line"><a name="l00543"></a><span class="lineno"> 543</span>  },</div><div class="line"><a name="l00544"></a><span class="lineno"> 544</span>  vector_sum_col_it, vector_sum_row_it, bias_it, mm_result_it, out_it);</div><div class="line"><a name="l00545"></a><span class="lineno"> 545</span>  }</div><div class="line"><a name="l00546"></a><span class="lineno"> 546</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00547"></a><span class="lineno"> 547</span>  {</div><div class="line"><a name="l00548"></a><span class="lineno"> 548</span>  <a class="code" href="namespacearm__compute.xhtml#a5002bf7ec46d52971f9526e94172cfee">execute_window_loop</a>(collapsed_window, [&](<span class="keyword">const</span> Coordinates & <span class="keywordtype">id</span>)</div><div class="line"><a name="l00549"></a><span class="lineno"> 549</span>  {</div><div class="line"><a name="l00550"></a><span class="lineno"> 550</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> batch_id = <span class="keywordtype">id</span>.z() / depth_input;</div><div class="line"><a name="l00551"></a><span class="lineno"> 551</span>  <span class="keyword">const</span> <span class="keyword">auto</span> vector_sum_col_ptr = reinterpret_cast<const int32_t *>(vector_sum_col_it.ptr() + batch_id * vector_sum_col_batch_offset);</div><div class="line"><a name="l00552"></a><span class="lineno"> 552</span>  <span class="keyword">const</span> <span class="keyword">auto</span> vector_sum_row_ptr = reinterpret_cast<const int32_t *>(vector_sum_row_it.ptr() + batch_id * sum_row_stride_y)</div><div class="line"><a name="l00553"></a><span class="lineno"> 553</span>  + <span class="keywordtype">id</span>.y() + (<span class="keywordtype">id</span>.z() % depth_input) * height_input;</div><div class="line"><a name="l00554"></a><span class="lineno"> 554</span>  run_offset_contribution_output_stage_window<Typer, true, true, false, is_bounded_relu, is_fixed_point>(vector_sum_col_ptr, vector_sum_row_ptr, <span class="keyword">nullptr</span>, mm_result_it, out_it,</div><div class="line"><a name="l00555"></a><span class="lineno"> 555</span>  result_offset_s32, result_shift_s32,</div><div class="line"><a name="l00556"></a><span class="lineno"> 556</span>  min_vec, max_vec, a_offset, b_offset, k_offset,</div><div class="line"><a name="l00557"></a><span class="lineno"> 557</span>  multiplier, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a979a54caef6e77ce0259e427136847e8">shift</a>, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>, min_bound, max_bound,</div><div class="line"><a name="l00558"></a><span class="lineno"> 558</span>  window_step_x, window_start_x, window_end_x);</div><div class="line"><a name="l00559"></a><span class="lineno"> 559</span>  },</div><div class="line"><a name="l00560"></a><span class="lineno"> 560</span>  vector_sum_col_it, vector_sum_row_it, mm_result_it, out_it);</div><div class="line"><a name="l00561"></a><span class="lineno"> 561</span>  }</div><div class="line"><a name="l00562"></a><span class="lineno"> 562</span>  }</div><div class="line"><a name="l00563"></a><span class="lineno"> 563</span>  <span class="keywordflow">else</span> <span class="keywordflow">if</span>((a_offset == 0) && (b_offset != 0))</div><div class="line"><a name="l00564"></a><span class="lineno"> 564</span>  {</div><div class="line"><a name="l00565"></a><span class="lineno"> 565</span>  <a class="code" href="_validate_8h.xhtml#a921b705e9e3e0fe928928447869e62a5">ARM_COMPUTE_ERROR_ON_NULLPTR</a>(vector_sum_row);</div><div class="line"><a name="l00566"></a><span class="lineno"> 566</span> </div><div class="line"><a name="l00567"></a><span class="lineno"> 567</span>  Iterator vector_sum_row_it = get_vector_sum_row_it(collapsed_window, vector_sum_row);</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="keyword">const</span> <span class="keywordtype">size_t</span> sum_row_stride_y = vector_sum_row->info()->strides_in_bytes().y();</div><div class="line"><a name="l00570"></a><span class="lineno"> 570</span> </div><div class="line"><a name="l00571"></a><span class="lineno"> 571</span>  <span class="keywordflow">if</span>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a> != <span class="keyword">nullptr</span>)</div><div class="line"><a name="l00572"></a><span class="lineno"> 572</span>  {</div><div class="line"><a name="l00573"></a><span class="lineno"> 573</span>  Iterator bias_it = get_bias_it(collapsed_window, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>);</div><div class="line"><a name="l00574"></a><span class="lineno"> 574</span>  <a class="code" href="namespacearm__compute.xhtml#a5002bf7ec46d52971f9526e94172cfee">execute_window_loop</a>(collapsed_window, [&](<span class="keyword">const</span> Coordinates & <span class="keywordtype">id</span>)</div><div class="line"><a name="l00575"></a><span class="lineno"> 575</span>  {</div><div class="line"><a name="l00576"></a><span class="lineno"> 576</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> batch_id = <span class="keywordtype">id</span>.z() / depth_input;</div><div class="line"><a name="l00577"></a><span class="lineno"> 577</span>  <span class="keyword">const</span> <span class="keyword">auto</span> vector_sum_row_ptr = reinterpret_cast<const int32_t *>(vector_sum_row_it.ptr() + batch_id * sum_row_stride_y)</div><div class="line"><a name="l00578"></a><span class="lineno"> 578</span>  + <span class="keywordtype">id</span>.y() + (<span class="keywordtype">id</span>.z() % depth_input) * height_input;</div><div class="line"><a name="l00579"></a><span class="lineno"> 579</span>  run_offset_contribution_output_stage_window<Typer, false, true, true, is_bounded_relu, is_fixed_point>(<span class="keyword">nullptr</span>, vector_sum_row_ptr, reinterpret_cast<const int32_t *>(bias_it.ptr()), mm_result_it,</div><div class="line"><a name="l00580"></a><span class="lineno"> 580</span>  out_it,</div><div class="line"><a name="l00581"></a><span class="lineno"> 581</span>  result_offset_s32, result_shift_s32,</div><div class="line"><a name="l00582"></a><span class="lineno"> 582</span>  min_vec, max_vec, a_offset, b_offset, k_offset,</div><div class="line"><a name="l00583"></a><span class="lineno"> 583</span>  multiplier, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a979a54caef6e77ce0259e427136847e8">shift</a>, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>, min_bound, max_bound,</div><div class="line"><a name="l00584"></a><span class="lineno"> 584</span>  window_step_x, window_start_x, window_end_x);</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>  vector_sum_row_it, bias_it, mm_result_it, out_it);</div><div class="line"><a name="l00587"></a><span class="lineno"> 587</span>  }</div><div class="line"><a name="l00588"></a><span class="lineno"> 588</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00589"></a><span class="lineno"> 589</span>  {</div><div class="line"><a name="l00590"></a><span class="lineno"> 590</span>  <a class="code" href="namespacearm__compute.xhtml#a5002bf7ec46d52971f9526e94172cfee">execute_window_loop</a>(collapsed_window, [&](<span class="keyword">const</span> Coordinates & <span class="keywordtype">id</span>)</div><div class="line"><a name="l00591"></a><span class="lineno"> 591</span>  {</div><div class="line"><a name="l00592"></a><span class="lineno"> 592</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> batch_id = <span class="keywordtype">id</span>.z() / depth_input;</div><div class="line"><a name="l00593"></a><span class="lineno"> 593</span>  <span class="keyword">const</span> <span class="keyword">auto</span> vector_sum_row_ptr = reinterpret_cast<const int32_t *>(vector_sum_row_it.ptr() + batch_id * sum_row_stride_y)</div><div class="line"><a name="l00594"></a><span class="lineno"> 594</span>  + <span class="keywordtype">id</span>.y() + (<span class="keywordtype">id</span>.z() % depth_input) * height_input;</div><div class="line"><a name="l00595"></a><span class="lineno"> 595</span>  run_offset_contribution_output_stage_window<Typer, false, true, false, is_bounded_relu, is_fixed_point>(<span class="keyword">nullptr</span>, vector_sum_row_ptr, <span class="keyword">nullptr</span>, mm_result_it, out_it,</div><div class="line"><a name="l00596"></a><span class="lineno"> 596</span>  result_offset_s32, result_shift_s32,</div><div class="line"><a name="l00597"></a><span class="lineno"> 597</span>  min_vec, max_vec, a_offset, b_offset, k_offset,</div><div class="line"><a name="l00598"></a><span class="lineno"> 598</span>  multiplier, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a979a54caef6e77ce0259e427136847e8">shift</a>, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>, min_bound, max_bound,</div><div class="line"><a name="l00599"></a><span class="lineno"> 599</span>  window_step_x, window_start_x, window_end_x);</div><div class="line"><a name="l00600"></a><span class="lineno"> 600</span>  },</div><div class="line"><a name="l00601"></a><span class="lineno"> 601</span>  vector_sum_row_it, mm_result_it, out_it);</div><div class="line"><a name="l00602"></a><span class="lineno"> 602</span>  }</div><div class="line"><a name="l00603"></a><span class="lineno"> 603</span>  }</div><div class="line"><a name="l00604"></a><span class="lineno"> 604</span>  <span class="keywordflow">else</span> <span class="keywordflow">if</span>((a_offset != 0) && (b_offset == 0))</div><div class="line"><a name="l00605"></a><span class="lineno"> 605</span>  {</div><div class="line"><a name="l00606"></a><span class="lineno"> 606</span>  <a class="code" href="_validate_8h.xhtml#a921b705e9e3e0fe928928447869e62a5">ARM_COMPUTE_ERROR_ON_NULLPTR</a>(vector_sum_col);</div><div class="line"><a name="l00607"></a><span class="lineno"> 607</span> </div><div class="line"><a name="l00608"></a><span class="lineno"> 608</span>  Iterator vector_sum_col_it = get_vector_sum_col_it(collapsed_window, vector_sum_col);</div><div class="line"><a name="l00609"></a><span class="lineno"> 609</span> </div><div class="line"><a name="l00610"></a><span class="lineno"> 610</span>  <span class="comment">// Offset in case vector_sum_col is batched</span></div><div class="line"><a name="l00611"></a><span class="lineno"> 611</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> vector_sum_col_batch_offset = slide_vector_sum_col ? vector_sum_col->info()->strides_in_bytes().z() : 0;</div><div class="line"><a name="l00612"></a><span class="lineno"> 612</span> </div><div class="line"><a name="l00613"></a><span class="lineno"> 613</span>  <span class="keywordflow">if</span>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a> != <span class="keyword">nullptr</span>)</div><div class="line"><a name="l00614"></a><span class="lineno"> 614</span>  {</div><div class="line"><a name="l00615"></a><span class="lineno"> 615</span>  Iterator bias_it = get_bias_it(collapsed_window, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>);</div><div class="line"><a name="l00616"></a><span class="lineno"> 616</span>  <a class="code" href="namespacearm__compute.xhtml#a5002bf7ec46d52971f9526e94172cfee">execute_window_loop</a>(collapsed_window, [&](<span class="keyword">const</span> Coordinates & <span class="keywordtype">id</span>)</div><div class="line"><a name="l00617"></a><span class="lineno"> 617</span>  {</div><div class="line"><a name="l00618"></a><span class="lineno"> 618</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> batch_id = <span class="keywordtype">id</span>.z() / depth_input;</div><div class="line"><a name="l00619"></a><span class="lineno"> 619</span>  <span class="keyword">const</span> <span class="keyword">auto</span> vector_sum_col_ptr = reinterpret_cast<const int32_t *>(vector_sum_col_it.ptr() + batch_id * vector_sum_col_batch_offset);</div><div class="line"><a name="l00620"></a><span class="lineno"> 620</span>  run_offset_contribution_output_stage_window<Typer, true, false, true, is_bounded_relu, is_fixed_point>(vector_sum_col_ptr, <span class="keyword">nullptr</span>, reinterpret_cast<const int32_t *>(bias_it.ptr()), mm_result_it,</div><div class="line"><a name="l00621"></a><span class="lineno"> 621</span>  out_it,</div><div class="line"><a name="l00622"></a><span class="lineno"> 622</span>  result_offset_s32, result_shift_s32,</div><div class="line"><a name="l00623"></a><span class="lineno"> 623</span>  min_vec, max_vec, a_offset, b_offset, k_offset,</div><div class="line"><a name="l00624"></a><span class="lineno"> 624</span>  multiplier, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a979a54caef6e77ce0259e427136847e8">shift</a>, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>, min_bound, max_bound,</div><div class="line"><a name="l00625"></a><span class="lineno"> 625</span>  window_step_x, window_start_x, window_end_x);</div><div class="line"><a name="l00626"></a><span class="lineno"> 626</span>  },</div><div class="line"><a name="l00627"></a><span class="lineno"> 627</span>  vector_sum_col_it, bias_it, mm_result_it, out_it);</div><div class="line"><a name="l00628"></a><span class="lineno"> 628</span>  }</div><div class="line"><a name="l00629"></a><span class="lineno"> 629</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00630"></a><span class="lineno"> 630</span>  {</div><div class="line"><a name="l00631"></a><span class="lineno"> 631</span>  <a class="code" href="namespacearm__compute.xhtml#a5002bf7ec46d52971f9526e94172cfee">execute_window_loop</a>(collapsed_window, [&](<span class="keyword">const</span> Coordinates & <span class="keywordtype">id</span>)</div><div class="line"><a name="l00632"></a><span class="lineno"> 632</span>  {</div><div class="line"><a name="l00633"></a><span class="lineno"> 633</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> batch_id = <span class="keywordtype">id</span>.z() / depth_input;</div><div class="line"><a name="l00634"></a><span class="lineno"> 634</span>  <span class="keyword">const</span> <span class="keyword">auto</span> vector_sum_col_ptr = reinterpret_cast<const int32_t *>(vector_sum_col_it.ptr() + batch_id * vector_sum_col_batch_offset);</div><div class="line"><a name="l00635"></a><span class="lineno"> 635</span>  run_offset_contribution_output_stage_window<Typer, true, false, false, is_bounded_relu, is_fixed_point>(vector_sum_col_ptr, <span class="keyword">nullptr</span>, <span class="keyword">nullptr</span>, mm_result_it, out_it,</div><div class="line"><a name="l00636"></a><span class="lineno"> 636</span>  result_offset_s32, result_shift_s32,</div><div class="line"><a name="l00637"></a><span class="lineno"> 637</span>  min_vec, max_vec, a_offset, b_offset, k_offset,</div><div class="line"><a name="l00638"></a><span class="lineno"> 638</span>  multiplier, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a979a54caef6e77ce0259e427136847e8">shift</a>, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>, min_bound, max_bound,</div><div class="line"><a name="l00639"></a><span class="lineno"> 639</span>  window_step_x, window_start_x, window_end_x);</div><div class="line"><a name="l00640"></a><span class="lineno"> 640</span>  },</div><div class="line"><a name="l00641"></a><span class="lineno"> 641</span>  vector_sum_col_it, mm_result_it, out_it);</div><div class="line"><a name="l00642"></a><span class="lineno"> 642</span>  }</div><div class="line"><a name="l00643"></a><span class="lineno"> 643</span>  }</div><div class="line"><a name="l00644"></a><span class="lineno"> 644</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00645"></a><span class="lineno"> 645</span>  {</div><div class="line"><a name="l00646"></a><span class="lineno"> 646</span>  <span class="keywordflow">if</span>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a> != <span class="keyword">nullptr</span>)</div><div class="line"><a name="l00647"></a><span class="lineno"> 647</span>  {</div><div class="line"><a name="l00648"></a><span class="lineno"> 648</span>  Iterator bias_it = get_bias_it(collapsed_window, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>);</div><div class="line"><a name="l00649"></a><span class="lineno"> 649</span>  <a class="code" href="namespacearm__compute.xhtml#a5002bf7ec46d52971f9526e94172cfee">execute_window_loop</a>(collapsed_window, [&](<span class="keyword">const</span> Coordinates &)</div><div class="line"><a name="l00650"></a><span class="lineno"> 650</span>  {</div><div class="line"><a name="l00651"></a><span class="lineno"> 651</span>  run_offset_contribution_output_stage_window<Typer, false, false, true, is_bounded_relu, is_fixed_point>(<span class="keyword">nullptr</span>, <span class="keyword">nullptr</span>, reinterpret_cast<const int32_t *>(bias_it.ptr()), mm_result_it, out_it,</div><div class="line"><a name="l00652"></a><span class="lineno"> 652</span>  result_offset_s32, result_shift_s32,</div><div class="line"><a name="l00653"></a><span class="lineno"> 653</span>  min_vec, max_vec, a_offset, b_offset, k_offset,</div><div class="line"><a name="l00654"></a><span class="lineno"> 654</span>  multiplier, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a979a54caef6e77ce0259e427136847e8">shift</a>, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>, min_bound, max_bound,</div><div class="line"><a name="l00655"></a><span class="lineno"> 655</span>  window_step_x, window_start_x, window_end_x);</div><div class="line"><a name="l00656"></a><span class="lineno"> 656</span>  },</div><div class="line"><a name="l00657"></a><span class="lineno"> 657</span>  bias_it, mm_result_it, out_it);</div><div class="line"><a name="l00658"></a><span class="lineno"> 658</span>  }</div><div class="line"><a name="l00659"></a><span class="lineno"> 659</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00660"></a><span class="lineno"> 660</span>  {</div><div class="line"><a name="l00661"></a><span class="lineno"> 661</span>  <a class="code" href="namespacearm__compute.xhtml#a5002bf7ec46d52971f9526e94172cfee">execute_window_loop</a>(collapsed_window, [&](<span class="keyword">const</span> Coordinates &)</div><div class="line"><a name="l00662"></a><span class="lineno"> 662</span>  {</div><div class="line"><a name="l00663"></a><span class="lineno"> 663</span>  run_offset_contribution_output_stage_window<Typer, false, false, false, is_bounded_relu, is_fixed_point>(<span class="keyword">nullptr</span>, <span class="keyword">nullptr</span>, <span class="keyword">nullptr</span>, mm_result_it, out_it,</div><div class="line"><a name="l00664"></a><span class="lineno"> 664</span>  result_offset_s32, result_shift_s32,</div><div class="line"><a name="l00665"></a><span class="lineno"> 665</span>  min_vec, max_vec, a_offset, b_offset, k_offset,</div><div class="line"><a name="l00666"></a><span class="lineno"> 666</span>  multiplier, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a979a54caef6e77ce0259e427136847e8">shift</a>, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>, min_bound, max_bound,</div><div class="line"><a name="l00667"></a><span class="lineno"> 667</span>  window_step_x, window_start_x, window_end_x);</div><div class="line"><a name="l00668"></a><span class="lineno"> 668</span>  },</div><div class="line"><a name="l00669"></a><span class="lineno"> 669</span>  mm_result_it, out_it);</div><div class="line"><a name="l00670"></a><span class="lineno"> 670</span>  }</div><div class="line"><a name="l00671"></a><span class="lineno"> 671</span>  <span class="keywordflow">return</span>;</div><div class="line"><a name="l00672"></a><span class="lineno"> 672</span>  }</div><div class="line"><a name="l00673"></a><span class="lineno"> 673</span> }</div><div class="line"><a name="l00674"></a><span class="lineno"> 674</span> </div><div class="line"><a name="l00675"></a><span class="lineno"> 675</span> <span class="keyword">template</span> <<span class="keywordtype">bool</span> is_gemm3d, <span class="keywordtype">bool</span> is_bounded_relu, <span class="keywordtype">bool</span> is_fixed_po<span class="keywordtype">int</span>></div><div class="line"><a name="l00676"></a><span class="lineno"> 676</span> <span class="keywordtype">void</span> run_offset_contribution_output_stage_symm(<span class="keyword">const</span> Window &window,</div><div class="line"><a name="l00677"></a><span class="lineno"> 677</span>  <span class="keyword">const</span> ITensor *mm_result, <span class="keyword">const</span> ITensor *vector_sum_col, <span class="keyword">const</span> ITensor *vector_sum_row, <span class="keyword">const</span> ITensor *<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>, ITensor *output,</div><div class="line"><a name="l00678"></a><span class="lineno"> 678</span>  int32_t a_offset, int32_t b_offset, int32_t k_offset, <span class="keywordtype">bool</span> slide_vector_sum_col,</div><div class="line"><a name="l00679"></a><span class="lineno"> 679</span>  GEMMLowpOutputStageInfo output_stage)</div><div class="line"><a name="l00680"></a><span class="lineno"> 680</span> {</div><div class="line"><a name="l00681"></a><span class="lineno"> 681</span>  <a class="code" href="_error_8h.xhtml#a6dc630a6ae9cc063b3924bcea8dee9d6">ARM_COMPUTE_UNUSED</a>(vector_sum_row, b_offset, k_offset);</div><div class="line"><a name="l00682"></a><span class="lineno"> 682</span> </div><div class="line"><a name="l00683"></a><span class="lineno"> 683</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> depth_input = is_gemm3d ? mm_result->info()->dimension(2) : 1;</div><div class="line"><a name="l00684"></a><span class="lineno"> 684</span> </div><div class="line"><a name="l00685"></a><span class="lineno"> 685</span>  <span class="keyword">const</span> int32_t <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a> = output_stage.gemmlowp_offset;</div><div class="line"><a name="l00686"></a><span class="lineno"> 686</span>  <span class="keyword">const</span> int32_t min_bound = output_stage.gemmlowp_min_bound;</div><div class="line"><a name="l00687"></a><span class="lineno"> 687</span>  <span class="keyword">const</span> int32_t max_bound = output_stage.gemmlowp_max_bound;</div><div class="line"><a name="l00688"></a><span class="lineno"> 688</span> </div><div class="line"><a name="l00689"></a><span class="lineno"> 689</span>  <span class="keyword">const</span> int32_t *result_multipliers = output_stage.gemmlowp_multipliers.data();</div><div class="line"><a name="l00690"></a><span class="lineno"> 690</span>  <span class="keyword">const</span> int32_t *result_shifts = output_stage.gemmlowp_shifts.data();</div><div class="line"><a name="l00691"></a><span class="lineno"> 691</span>  <span class="keyword">const</span> int32x4_t result_offset_s32 = vdupq_n_s32(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>);</div><div class="line"><a name="l00692"></a><span class="lineno"> 692</span>  <span class="keyword">const</span> int8x16_t min_s8 = vdupq_n_s8(static_cast<int8_t>(min_bound));</div><div class="line"><a name="l00693"></a><span class="lineno"> 693</span>  <span class="keyword">const</span> int8x16_t max_s8 = vdupq_n_s8(static_cast<int8_t>(max_bound));</div><div class="line"><a name="l00694"></a><span class="lineno"> 694</span> </div><div class="line"><a name="l00695"></a><span class="lineno"> 695</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> window_step_x = 16;</div><div class="line"><a name="l00696"></a><span class="lineno"> 696</span>  <span class="keyword">const</span> <span class="keyword">auto</span> window_start_x = static_cast<int>(window.x().start());</div><div class="line"><a name="l00697"></a><span class="lineno"> 697</span>  <span class="keyword">const</span> <span class="keyword">auto</span> window_end_x = static_cast<int>(window.x().end());</div><div class="line"><a name="l00698"></a><span class="lineno"> 698</span> </div><div class="line"><a name="l00699"></a><span class="lineno"> 699</span>  Window win(window);</div><div class="line"><a name="l00700"></a><span class="lineno"> 700</span>  win.set(<a class="code" href="classarm__compute_1_1_window.xhtml#aa96e81276ee4f87ab386cd05a5539a7d">Window::DimX</a>, Window::Dimension(0, 1, 1));</div><div class="line"><a name="l00701"></a><span class="lineno"> 701</span> </div><div class="line"><a name="l00702"></a><span class="lineno"> 702</span>  Window collapsed_window = win.collapse_if_possible(win, <a class="code" href="classarm__compute_1_1_window.xhtml#a893d17b56b9abc4423ce26e9a24ac5dc">Window::DimZ</a>);</div><div class="line"><a name="l00703"></a><span class="lineno"> 703</span> </div><div class="line"><a name="l00704"></a><span class="lineno"> 704</span>  Iterator mm_result_it(mm_result, win);</div><div class="line"><a name="l00705"></a><span class="lineno"> 705</span>  Iterator out_it(output, win);</div><div class="line"><a name="l00706"></a><span class="lineno"> 706</span> </div><div class="line"><a name="l00707"></a><span class="lineno"> 707</span>  <span class="keywordflow">if</span>(a_offset != 0)</div><div class="line"><a name="l00708"></a><span class="lineno"> 708</span>  {</div><div class="line"><a name="l00709"></a><span class="lineno"> 709</span>  <a class="code" href="_validate_8h.xhtml#a921b705e9e3e0fe928928447869e62a5">ARM_COMPUTE_ERROR_ON_NULLPTR</a>(vector_sum_col);</div><div class="line"><a name="l00710"></a><span class="lineno"> 710</span> </div><div class="line"><a name="l00711"></a><span class="lineno"> 711</span>  Iterator vector_sum_col_it = get_vector_sum_col_it(collapsed_window, vector_sum_col);</div><div class="line"><a name="l00712"></a><span class="lineno"> 712</span> </div><div class="line"><a name="l00713"></a><span class="lineno"> 713</span>  <span class="comment">// Offset in case vector_sum_col is batched</span></div><div class="line"><a name="l00714"></a><span class="lineno"> 714</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> vector_sum_col_batch_offset = slide_vector_sum_col ? vector_sum_col->info()->strides_in_bytes().z() : 0;</div><div class="line"><a name="l00715"></a><span class="lineno"> 715</span> </div><div class="line"><a name="l00716"></a><span class="lineno"> 716</span>  <span class="keywordflow">if</span>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a> != <span class="keyword">nullptr</span>)</div><div class="line"><a name="l00717"></a><span class="lineno"> 717</span>  {</div><div class="line"><a name="l00718"></a><span class="lineno"> 718</span>  Iterator bias_it = get_bias_it(collapsed_window, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>);</div><div class="line"><a name="l00719"></a><span class="lineno"> 719</span>  <a class="code" href="namespacearm__compute.xhtml#a5002bf7ec46d52971f9526e94172cfee">execute_window_loop</a>(collapsed_window, [&](<span class="keyword">const</span> Coordinates & <span class="keywordtype">id</span>)</div><div class="line"><a name="l00720"></a><span class="lineno"> 720</span>  {</div><div class="line"><a name="l00721"></a><span class="lineno"> 721</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> batch_id = <span class="keywordtype">id</span>.z() / depth_input;</div><div class="line"><a name="l00722"></a><span class="lineno"> 722</span>  <span class="keyword">const</span> <span class="keyword">auto</span> vector_sum_col_ptr = reinterpret_cast<const int32_t *>(vector_sum_col_it.ptr() + batch_id * vector_sum_col_batch_offset);</div><div class="line"><a name="l00723"></a><span class="lineno"> 723</span>  run_offset_contribution_output_stage_window_symm<true, true, is_bounded_relu, is_fixed_point>(vector_sum_col_ptr, reinterpret_cast<const int32_t *>(bias_it.ptr()), mm_result_it, out_it,</div><div class="line"><a name="l00724"></a><span class="lineno"> 724</span>  result_multipliers, result_shifts,</div><div class="line"><a name="l00725"></a><span class="lineno"> 725</span>  result_offset_s32, min_s8, max_s8,</div><div class="line"><a name="l00726"></a><span class="lineno"> 726</span>  a_offset, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>, min_bound, max_bound,</div><div class="line"><a name="l00727"></a><span class="lineno"> 727</span>  window_step_x, window_start_x, window_end_x);</div><div class="line"><a name="l00728"></a><span class="lineno"> 728</span>  },</div><div class="line"><a name="l00729"></a><span class="lineno"> 729</span>  vector_sum_col_it, bias_it, mm_result_it, out_it);</div><div class="line"><a name="l00730"></a><span class="lineno"> 730</span>  }</div><div class="line"><a name="l00731"></a><span class="lineno"> 731</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00732"></a><span class="lineno"> 732</span>  {</div><div class="line"><a name="l00733"></a><span class="lineno"> 733</span>  <a class="code" href="namespacearm__compute.xhtml#a5002bf7ec46d52971f9526e94172cfee">execute_window_loop</a>(collapsed_window, [&](<span class="keyword">const</span> Coordinates & <span class="keywordtype">id</span>)</div><div class="line"><a name="l00734"></a><span class="lineno"> 734</span>  {</div><div class="line"><a name="l00735"></a><span class="lineno"> 735</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> batch_id = <span class="keywordtype">id</span>.z() / depth_input;</div><div class="line"><a name="l00736"></a><span class="lineno"> 736</span>  <span class="keyword">const</span> <span class="keyword">auto</span> vector_sum_col_ptr = reinterpret_cast<const int32_t *>(vector_sum_col_it.ptr() + batch_id * vector_sum_col_batch_offset);</div><div class="line"><a name="l00737"></a><span class="lineno"> 737</span>  run_offset_contribution_output_stage_window_symm<true, false, is_bounded_relu, is_fixed_point>(vector_sum_col_ptr, <span class="keyword">nullptr</span>, mm_result_it, out_it,</div><div class="line"><a name="l00738"></a><span class="lineno"> 738</span>  result_multipliers, result_shifts,</div><div class="line"><a name="l00739"></a><span class="lineno"> 739</span>  result_offset_s32, min_s8, max_s8,</div><div class="line"><a name="l00740"></a><span class="lineno"> 740</span>  a_offset, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>, min_bound, max_bound,</div><div class="line"><a name="l00741"></a><span class="lineno"> 741</span>  window_step_x, window_start_x, window_end_x);</div><div class="line"><a name="l00742"></a><span class="lineno"> 742</span>  },</div><div class="line"><a name="l00743"></a><span class="lineno"> 743</span>  vector_sum_col_it, mm_result_it, out_it);</div><div class="line"><a name="l00744"></a><span class="lineno"> 744</span>  }</div><div class="line"><a name="l00745"></a><span class="lineno"> 745</span>  }</div><div class="line"><a name="l00746"></a><span class="lineno"> 746</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00747"></a><span class="lineno"> 747</span>  {</div><div class="line"><a name="l00748"></a><span class="lineno"> 748</span>  <span class="keywordflow">if</span>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a> != <span class="keyword">nullptr</span>)</div><div class="line"><a name="l00749"></a><span class="lineno"> 749</span>  {</div><div class="line"><a name="l00750"></a><span class="lineno"> 750</span>  Iterator bias_it = get_bias_it(collapsed_window, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>);</div><div class="line"><a name="l00751"></a><span class="lineno"> 751</span>  <a class="code" href="namespacearm__compute.xhtml#a5002bf7ec46d52971f9526e94172cfee">execute_window_loop</a>(collapsed_window, [&](<span class="keyword">const</span> Coordinates &)</div><div class="line"><a name="l00752"></a><span class="lineno"> 752</span>  {</div><div class="line"><a name="l00753"></a><span class="lineno"> 753</span>  run_offset_contribution_output_stage_window_symm<false, true, is_bounded_relu, is_fixed_point>(<span class="keyword">nullptr</span>, reinterpret_cast<const int32_t *>(bias_it.ptr()), mm_result_it, out_it,</div><div class="line"><a name="l00754"></a><span class="lineno"> 754</span>  result_multipliers, result_shifts,</div><div class="line"><a name="l00755"></a><span class="lineno"> 755</span>  result_offset_s32, min_s8, max_s8,</div><div class="line"><a name="l00756"></a><span class="lineno"> 756</span>  a_offset, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>, min_bound, max_bound,</div><div class="line"><a name="l00757"></a><span class="lineno"> 757</span>  window_step_x, window_start_x, window_end_x);</div><div class="line"><a name="l00758"></a><span class="lineno"> 758</span>  },</div><div class="line"><a name="l00759"></a><span class="lineno"> 759</span>  bias_it, mm_result_it, out_it);</div><div class="line"><a name="l00760"></a><span class="lineno"> 760</span>  }</div><div class="line"><a name="l00761"></a><span class="lineno"> 761</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00762"></a><span class="lineno"> 762</span>  {</div><div class="line"><a name="l00763"></a><span class="lineno"> 763</span>  <a class="code" href="namespacearm__compute.xhtml#a5002bf7ec46d52971f9526e94172cfee">execute_window_loop</a>(collapsed_window, [&](<span class="keyword">const</span> Coordinates &)</div><div class="line"><a name="l00764"></a><span class="lineno"> 764</span>  {</div><div class="line"><a name="l00765"></a><span class="lineno"> 765</span>  run_offset_contribution_output_stage_window_symm<false, false, is_bounded_relu, is_fixed_point>(<span class="keyword">nullptr</span>, <span class="keyword">nullptr</span>, mm_result_it, out_it,</div><div class="line"><a name="l00766"></a><span class="lineno"> 766</span>  result_multipliers, result_shifts,</div><div class="line"><a name="l00767"></a><span class="lineno"> 767</span>  result_offset_s32, min_s8, max_s8,</div><div class="line"><a name="l00768"></a><span class="lineno"> 768</span>  a_offset, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>, min_bound, max_bound,</div><div class="line"><a name="l00769"></a><span class="lineno"> 769</span>  window_step_x, window_start_x, window_end_x);</div><div class="line"><a name="l00770"></a><span class="lineno"> 770</span>  },</div><div class="line"><a name="l00771"></a><span class="lineno"> 771</span>  mm_result_it, out_it);</div><div class="line"><a name="l00772"></a><span class="lineno"> 772</span>  }</div><div class="line"><a name="l00773"></a><span class="lineno"> 773</span>  <span class="keywordflow">return</span>;</div><div class="line"><a name="l00774"></a><span class="lineno"> 774</span>  }</div><div class="line"><a name="l00775"></a><span class="lineno"> 775</span> }</div><div class="line"><a name="l00776"></a><span class="lineno"> 776</span> </div><div class="line"><a name="l00777"></a><span class="lineno"> 777</span> Status validate_arguments(<span class="keyword">const</span> ITensorInfo *mm_result, <span class="keyword">const</span> ITensorInfo *vector_sum_col, <span class="keyword">const</span> ITensorInfo *vector_sum_row, <span class="keyword">const</span> ITensorInfo *<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>, <span class="keyword">const</span> ITensorInfo *output,</div><div class="line"><a name="l00778"></a><span class="lineno"> 778</span>  int32_t a_offset, int32_t b_offset, GEMMLowpOutputStageInfo output_stage)</div><div class="line"><a name="l00779"></a><span class="lineno"> 779</span> {</div><div class="line"><a name="l00780"></a><span class="lineno"> 780</span>  <a class="code" href="_validate_8h.xhtml#ae7eed178dac535c6e727061b1f5bc6eb">ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN</a>(mm_result, 1, <a class="code" href="namespacearm__compute.xhtml#ab4e88c89b3b7ea1735996cc4def22d58aa1e28eee0339658d39a8b4d325b56e9c">DataType::S32</a>);</div><div class="line"><a name="l00781"></a><span class="lineno"> 781</span>  <span class="keywordflow">if</span>(output->data_type() == <a class="code" href="namespacearm__compute.xhtml#ad8ed01ff3ff33333d8e19db4d2818bb6af14462d71aa842202c3e4b272c7ec924">DataType::QASYMM8</a>)</div><div class="line"><a name="l00782"></a><span class="lineno"> 782</span>  {</div><div class="line"><a name="l00783"></a><span class="lineno"> 783</span>  <a class="code" href="_error_8h.xhtml#a206d6e247e0957ac3dee45d27756fc25">ARM_COMPUTE_RETURN_ERROR_ON</a>(output_stage.gemmlowp_max_bound > 255);</div><div class="line"><a name="l00784"></a><span class="lineno"> 784</span>  <a class="code" href="_error_8h.xhtml#a206d6e247e0957ac3dee45d27756fc25">ARM_COMPUTE_RETURN_ERROR_ON</a>(output_stage.gemmlowp_min_bound < 0);</div><div class="line"><a name="l00785"></a><span class="lineno"> 785</span>  }</div><div class="line"><a name="l00786"></a><span class="lineno"> 786</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00787"></a><span class="lineno"> 787</span>  {</div><div class="line"><a name="l00788"></a><span class="lineno"> 788</span>  <a class="code" href="_error_8h.xhtml#a206d6e247e0957ac3dee45d27756fc25">ARM_COMPUTE_RETURN_ERROR_ON</a>(output_stage.gemmlowp_max_bound > 127);</div><div class="line"><a name="l00789"></a><span class="lineno"> 789</span>  <a class="code" href="_error_8h.xhtml#a206d6e247e0957ac3dee45d27756fc25">ARM_COMPUTE_RETURN_ERROR_ON</a>(output_stage.gemmlowp_min_bound < -128);</div><div class="line"><a name="l00790"></a><span class="lineno"> 790</span>  <a class="code" href="_error_8h.xhtml#a206d6e247e0957ac3dee45d27756fc25">ARM_COMPUTE_RETURN_ERROR_ON</a>(mm_result->dimension(0) > 1 && output_stage.gemmlowp_multipliers.size() > 1 && b_offset != 0);</div><div class="line"><a name="l00791"></a><span class="lineno"> 791</span>  }</div><div class="line"><a name="l00792"></a><span class="lineno"> 792</span>  <a class="code" href="_error_8h.xhtml#a206d6e247e0957ac3dee45d27756fc25">ARM_COMPUTE_RETURN_ERROR_ON</a>(output_stage.gemmlowp_min_bound > output_stage.gemmlowp_max_bound);</div><div class="line"><a name="l00793"></a><span class="lineno"> 793</span>  <a class="code" href="_error_8h.xhtml#a206d6e247e0957ac3dee45d27756fc25">ARM_COMPUTE_RETURN_ERROR_ON</a>(output_stage.type != <a class="code" href="namespacearm__compute.xhtml#a5558e2cc22f7f4771653d992c8ad8864a079e2ddc95b344b5cb0188bed9a80d8b">GEMMLowpOutputStageType::QUANTIZE_DOWN</a> && output_stage.type != <a class="code" href="namespacearm__compute.xhtml#a5558e2cc22f7f4771653d992c8ad8864ab300cae200f67712c1eb9234e28158ca">GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT</a>);</div><div class="line"><a name="l00794"></a><span class="lineno"> 794</span> </div><div class="line"><a name="l00795"></a><span class="lineno"> 795</span>  <span class="keywordflow">if</span>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a> != <span class="keyword">nullptr</span>)</div><div class="line"><a name="l00796"></a><span class="lineno"> 796</span>  {</div><div class="line"><a name="l00797"></a><span class="lineno"> 797</span>  <a class="code" href="_validate_8h.xhtml#ae7eed178dac535c6e727061b1f5bc6eb">ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>, 1, <a class="code" href="namespacearm__compute.xhtml#ab4e88c89b3b7ea1735996cc4def22d58aa1e28eee0339658d39a8b4d325b56e9c">DataType::S32</a>);</div><div class="line"><a name="l00798"></a><span class="lineno"> 798</span>  <a class="code" href="_error_8h.xhtml#a206d6e247e0957ac3dee45d27756fc25">ARM_COMPUTE_RETURN_ERROR_ON</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>->num_dimensions() > 1);</div><div class="line"><a name="l00799"></a><span class="lineno"> 799</span>  <a class="code" href="_error_8h.xhtml#a206d6e247e0957ac3dee45d27756fc25">ARM_COMPUTE_RETURN_ERROR_ON</a>(mm_result->dimension(0) != <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>->dimension(0));</div><div class="line"><a name="l00800"></a><span class="lineno"> 800</span>  }</div><div class="line"><a name="l00801"></a><span class="lineno"> 801</span> </div><div class="line"><a name="l00802"></a><span class="lineno"> 802</span>  <span class="comment">// If a_offset == 0, vector_sum_col can be a nullptr</span></div><div class="line"><a name="l00803"></a><span class="lineno"> 803</span>  <span class="keywordflow">if</span>(a_offset != 0)</div><div class="line"><a name="l00804"></a><span class="lineno"> 804</span>  {</div><div class="line"><a name="l00805"></a><span class="lineno"> 805</span>  <a class="code" href="_validate_8h.xhtml#ae7eed178dac535c6e727061b1f5bc6eb">ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN</a>(vector_sum_col, 1, <a class="code" href="namespacearm__compute.xhtml#ab4e88c89b3b7ea1735996cc4def22d58aa1e28eee0339658d39a8b4d325b56e9c">DataType::S32</a>);</div><div class="line"><a name="l00806"></a><span class="lineno"> 806</span>  <a class="code" href="_error_8h.xhtml#a206d6e247e0957ac3dee45d27756fc25">ARM_COMPUTE_RETURN_ERROR_ON</a>(vector_sum_col->dimension(0) != mm_result->dimension(0));</div><div class="line"><a name="l00807"></a><span class="lineno"> 807</span>  }</div><div class="line"><a name="l00808"></a><span class="lineno"> 808</span> </div><div class="line"><a name="l00809"></a><span class="lineno"> 809</span>  <span class="comment">// If b_offset == 0, vector_sum_row can be a nullptr</span></div><div class="line"><a name="l00810"></a><span class="lineno"> 810</span>  <span class="keywordflow">if</span>(b_offset != 0)</div><div class="line"><a name="l00811"></a><span class="lineno"> 811</span>  {</div><div class="line"><a name="l00812"></a><span class="lineno"> 812</span>  <a class="code" href="_validate_8h.xhtml#ae7eed178dac535c6e727061b1f5bc6eb">ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN</a>(vector_sum_row, 1, <a class="code" href="namespacearm__compute.xhtml#ab4e88c89b3b7ea1735996cc4def22d58aa1e28eee0339658d39a8b4d325b56e9c">DataType::S32</a>);</div><div class="line"><a name="l00813"></a><span class="lineno"> 813</span> </div><div class="line"><a name="l00814"></a><span class="lineno"> 814</span>  <span class="comment">// Check if input is a 3D reinterpretation</span></div><div class="line"><a name="l00815"></a><span class="lineno"> 815</span>  <span class="keyword">const</span> <span class="keywordtype">bool</span> reinterpret_as_3d = mm_result->num_dimensions() > 1 && mm_result->tensor_shape().y() != vector_sum_row->tensor_shape().x();</div><div class="line"><a name="l00816"></a><span class="lineno"> 816</span> </div><div class="line"><a name="l00817"></a><span class="lineno"> 817</span>  <span class="comment">// Validate input</span></div><div class="line"><a name="l00818"></a><span class="lineno"> 818</span>  <a class="code" href="_error_8h.xhtml#a206d6e247e0957ac3dee45d27756fc25">ARM_COMPUTE_RETURN_ERROR_ON</a>(reinterpret_as_3d && vector_sum_row->dimension(0) != (mm_result->dimension(1) * mm_result->dimension(2)));</div><div class="line"><a name="l00819"></a><span class="lineno"> 819</span>  <a class="code" href="_error_8h.xhtml#a206d6e247e0957ac3dee45d27756fc25">ARM_COMPUTE_RETURN_ERROR_ON</a>(!reinterpret_as_3d && vector_sum_row->dimension(0) != mm_result->dimension(1));</div><div class="line"><a name="l00820"></a><span class="lineno"> 820</span> </div><div class="line"><a name="l00821"></a><span class="lineno"> 821</span>  TensorShape <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#ab1806bf0c5a41f674fb9d2dc6af644f5">output_shape</a> = output->tensor_shape();</div><div class="line"><a name="l00822"></a><span class="lineno"> 822</span>  <span class="keywordflow">if</span>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#ab1806bf0c5a41f674fb9d2dc6af644f5">output_shape</a>.num_dimensions() > 1)</div><div class="line"><a name="l00823"></a><span class="lineno"> 823</span>  {</div><div class="line"><a name="l00824"></a><span class="lineno"> 824</span>  <span class="keyword">const</span> <span class="keywordtype">unsigned</span> <span class="keywordtype">int</span> output_batch_idx = reinterpret_as_3d ? 3 : 2;</div><div class="line"><a name="l00825"></a><span class="lineno"> 825</span> </div><div class="line"><a name="l00826"></a><span class="lineno"> 826</span>  TensorShape vector_sum_row_shape = vector_sum_row->tensor_shape();</div><div class="line"><a name="l00827"></a><span class="lineno"> 827</span>  vector_sum_row_shape.collapse_from(1);</div><div class="line"><a name="l00828"></a><span class="lineno"> 828</span>  <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#ab1806bf0c5a41f674fb9d2dc6af644f5">output_shape</a>.collapse_from(output_batch_idx);</div><div class="line"><a name="l00829"></a><span class="lineno"> 829</span> </div><div class="line"><a name="l00830"></a><span class="lineno"> 830</span>  <a class="code" href="_error_8h.xhtml#a1c69762a42ab8add645d0a949b6f4b1f">ARM_COMPUTE_RETURN_ERROR_ON_MSG</a>(vector_sum_row_shape[1] != <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#ab1806bf0c5a41f674fb9d2dc6af644f5">output_shape</a>[output_batch_idx],</div><div class="line"><a name="l00831"></a><span class="lineno"> 831</span>  <span class="stringliteral">"mm_result tensor must have the same number of batches of output tensor"</span>);</div><div class="line"><a name="l00832"></a><span class="lineno"> 832</span> </div><div class="line"><a name="l00833"></a><span class="lineno"> 833</span>  <span class="keywordflow">if</span>(a_offset != 0)</div><div class="line"><a name="l00834"></a><span class="lineno"> 834</span>  {</div><div class="line"><a name="l00835"></a><span class="lineno"> 835</span>  TensorShape vector_sum_col_shape = vector_sum_col->tensor_shape();</div><div class="line"><a name="l00836"></a><span class="lineno"> 836</span>  vector_sum_col_shape.collapse_from(1);</div><div class="line"><a name="l00837"></a><span class="lineno"> 837</span> </div><div class="line"><a name="l00838"></a><span class="lineno"> 838</span>  <a class="code" href="_error_8h.xhtml#a1c69762a42ab8add645d0a949b6f4b1f">ARM_COMPUTE_RETURN_ERROR_ON_MSG</a>(vector_sum_col_shape[1] != 1 && vector_sum_col_shape[1] != vector_sum_row_shape[1],</div><div class="line"><a name="l00839"></a><span class="lineno"> 839</span>  <span class="stringliteral">"vector_sum_col tensor must have the same number of batches of vector_sum_row_shape or the number of batches must be set to 1"</span>);</div><div class="line"><a name="l00840"></a><span class="lineno"> 840</span>  }</div><div class="line"><a name="l00841"></a><span class="lineno"> 841</span>  }</div><div class="line"><a name="l00842"></a><span class="lineno"> 842</span>  }</div><div class="line"><a name="l00843"></a><span class="lineno"> 843</span> </div><div class="line"><a name="l00844"></a><span class="lineno"> 844</span>  <span class="keywordflow">if</span>(output->total_size() != 0)</div><div class="line"><a name="l00845"></a><span class="lineno"> 845</span>  {</div><div class="line"><a name="l00846"></a><span class="lineno"> 846</span>  <a class="code" href="_validate_8h.xhtml#ae7eed178dac535c6e727061b1f5bc6eb">ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN</a>(output, 1, <a class="code" href="namespacearm__compute.xhtml#ad8ed01ff3ff33333d8e19db4d2818bb6af14462d71aa842202c3e4b272c7ec924">DataType::QASYMM8</a>, <a class="code" href="namespacearm__compute.xhtml#ad8ed01ff3ff33333d8e19db4d2818bb6a329f5d0c4b0c80e3474951d2c4435dd9">DataType::QASYMM8_SIGNED</a>);</div><div class="line"><a name="l00847"></a><span class="lineno"> 847</span>  <a class="code" href="_validate_8h.xhtml#a27e4638546c88b8916f967e6e54480a9">ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES</a>(mm_result, output);</div><div class="line"><a name="l00848"></a><span class="lineno"> 848</span>  }</div><div class="line"><a name="l00849"></a><span class="lineno"> 849</span> </div><div class="line"><a name="l00850"></a><span class="lineno"> 850</span>  <span class="keywordflow">return</span> Status{};</div><div class="line"><a name="l00851"></a><span class="lineno"> 851</span> }</div><div class="line"><a name="l00852"></a><span class="lineno"> 852</span> </div><div class="line"><a name="l00853"></a><span class="lineno"> 853</span> std::pair<Status, Window> validate_and_configure_window(ITensorInfo *mm_result, ITensorInfo *output)</div><div class="line"><a name="l00854"></a><span class="lineno"> 854</span> {</div><div class="line"><a name="l00855"></a><span class="lineno"> 855</span>  <span class="comment">// Output auto inizialitation if not yet initialized</span></div><div class="line"><a name="l00856"></a><span class="lineno"> 856</span>  <a class="code" href="namespacearm__compute.xhtml#a47be6fa38308d0003c25b60b7dbc45ce">auto_init_if_empty</a>(*output, mm_result->clone()->set_data_type(<a class="code" href="namespacearm__compute.xhtml#ad8ed01ff3ff33333d8e19db4d2818bb6af14462d71aa842202c3e4b272c7ec924">DataType::QASYMM8</a>));</div><div class="line"><a name="l00857"></a><span class="lineno"> 857</span> </div><div class="line"><a name="l00858"></a><span class="lineno"> 858</span>  <span class="comment">// Configure kernel window</span></div><div class="line"><a name="l00859"></a><span class="lineno"> 859</span>  Window win = <a class="code" href="namespacearm__compute.xhtml#ab7980fa5ee693e3282a76da047a1c3b5">calculate_max_window</a>(*mm_result, Steps());</div><div class="line"><a name="l00860"></a><span class="lineno"> 860</span> </div><div class="line"><a name="l00861"></a><span class="lineno"> 861</span>  <span class="comment">// Note: This kernel performs 16 elements per iteration.</span></div><div class="line"><a name="l00862"></a><span class="lineno"> 862</span>  <span class="comment">// However, since we use a left-over for loop, we cannot have any read or write out of memory</span></div><div class="line"><a name="l00863"></a><span class="lineno"> 863</span>  <span class="comment">// For this reason num_elems_processed_per_iteration is 1 and so update_window_and_padding() can be skipped</span></div><div class="line"><a name="l00864"></a><span class="lineno"> 864</span>  Coordinates coord;</div><div class="line"><a name="l00865"></a><span class="lineno"> 865</span>  coord.set_num_dimensions(output->num_dimensions());</div><div class="line"><a name="l00866"></a><span class="lineno"> 866</span>  output->set_valid_region(ValidRegion(coord, output->tensor_shape()));</div><div class="line"><a name="l00867"></a><span class="lineno"> 867</span> </div><div class="line"><a name="l00868"></a><span class="lineno"> 868</span>  <span class="keywordflow">return</span> std::make_pair(Status{}, win);</div><div class="line"><a name="l00869"></a><span class="lineno"> 869</span> }</div><div class="line"><a name="l00870"></a><span class="lineno"> 870</span> </div><div class="line"><a name="l00871"></a><span class="lineno"> 871</span> <a class="code" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_offset_contribution_output_stage_kernel.xhtml#adb79bb3ad15444f00b55b35f1d6e16b7">NEGEMMLowpOffsetContributionOutputStageKernel::NEGEMMLowpOffsetContributionOutputStageFunction</a></div><div class="line"><a name="l00872"></a><span class="lineno"> 872</span> get_configured_function(<span class="keyword">const</span> ITensor *mm_result, <span class="keyword">const</span> ITensor *vector_sum_row, <span class="keyword">const</span> ITensor *output, GEMMLowpOutputStageInfo output_stage)</div><div class="line"><a name="l00873"></a><span class="lineno"> 873</span> {</div><div class="line"><a name="l00874"></a><span class="lineno"> 874</span>  <span class="keyword">static</span> std::map<uint8_t, NEGEMMLowpOffsetContributionOutputStageKernel::NEGEMMLowpOffsetContributionOutputStageFunction> map_function_qasymm =</div><div class="line"><a name="l00875"></a><span class="lineno"> 875</span>  {</div><div class="line"><a name="l00876"></a><span class="lineno"> 876</span>  { 0, &run_offset_contribution_output_stage<uint8_t, false, false, false> },</div><div class="line"><a name="l00877"></a><span class="lineno"> 877</span>  { 1, &run_offset_contribution_output_stage<uint8_t, true, false, false> },</div><div class="line"><a name="l00878"></a><span class="lineno"> 878</span>  { 2, &run_offset_contribution_output_stage<uint8_t, false, true, false> },</div><div class="line"><a name="l00879"></a><span class="lineno"> 879</span>  { 3, &run_offset_contribution_output_stage<uint8_t, true, true, false> },</div><div class="line"><a name="l00880"></a><span class="lineno"> 880</span>  { 4, &run_offset_contribution_output_stage<uint8_t, false, false, true> },</div><div class="line"><a name="l00881"></a><span class="lineno"> 881</span>  { 5, &run_offset_contribution_output_stage<uint8_t, true, false, true> },</div><div class="line"><a name="l00882"></a><span class="lineno"> 882</span>  { 6, &run_offset_contribution_output_stage<uint8_t, false, true, true> },</div><div class="line"><a name="l00883"></a><span class="lineno"> 883</span>  { 7, &run_offset_contribution_output_stage<uint8_t, true, true, true> },</div><div class="line"><a name="l00884"></a><span class="lineno"> 884</span>  { 8, &run_offset_contribution_output_stage<int8_t, false, false, false> },</div><div class="line"><a name="l00885"></a><span class="lineno"> 885</span>  { 9, &run_offset_contribution_output_stage<int8_t, true, false, false> },</div><div class="line"><a name="l00886"></a><span class="lineno"> 886</span>  { 10, &run_offset_contribution_output_stage<int8_t, false, true, false> },</div><div class="line"><a name="l00887"></a><span class="lineno"> 887</span>  { 11, &run_offset_contribution_output_stage<int8_t, true, true, false> },</div><div class="line"><a name="l00888"></a><span class="lineno"> 888</span>  { 12, &run_offset_contribution_output_stage<int8_t, false, false, true> },</div><div class="line"><a name="l00889"></a><span class="lineno"> 889</span>  { 13, &run_offset_contribution_output_stage<int8_t, true, false, true> },</div><div class="line"><a name="l00890"></a><span class="lineno"> 890</span>  { 14, &run_offset_contribution_output_stage<int8_t, false, true, true> },</div><div class="line"><a name="l00891"></a><span class="lineno"> 891</span>  { 15, &run_offset_contribution_output_stage<int8_t, true, true, true> },</div><div class="line"><a name="l00892"></a><span class="lineno"> 892</span>  };</div><div class="line"><a name="l00893"></a><span class="lineno"> 893</span> </div><div class="line"><a name="l00894"></a><span class="lineno"> 894</span>  <span class="keyword">static</span> std::map<uint8_t, NEGEMMLowpOffsetContributionOutputStageKernel::NEGEMMLowpOffsetContributionOutputStageFunction> map_function_qsymm =</div><div class="line"><a name="l00895"></a><span class="lineno"> 895</span>  {</div><div class="line"><a name="l00896"></a><span class="lineno"> 896</span>  { 0, &run_offset_contribution_output_stage_symm<false, false, false> },</div><div class="line"><a name="l00897"></a><span class="lineno"> 897</span>  { 1, &run_offset_contribution_output_stage_symm<true, false, false> },</div><div class="line"><a name="l00898"></a><span class="lineno"> 898</span>  { 2, &run_offset_contribution_output_stage_symm<false, true, false> },</div><div class="line"><a name="l00899"></a><span class="lineno"> 899</span>  { 3, &run_offset_contribution_output_stage_symm<true, true, false> },</div><div class="line"><a name="l00900"></a><span class="lineno"> 900</span>  { 4, &run_offset_contribution_output_stage_symm<false, false, true> },</div><div class="line"><a name="l00901"></a><span class="lineno"> 901</span>  { 5, &run_offset_contribution_output_stage_symm<true, false, true> },</div><div class="line"><a name="l00902"></a><span class="lineno"> 902</span>  { 6, &run_offset_contribution_output_stage_symm<false, true, true> },</div><div class="line"><a name="l00903"></a><span class="lineno"> 903</span>  { 7, &run_offset_contribution_output_stage_symm<true, true, true> }</div><div class="line"><a name="l00904"></a><span class="lineno"> 904</span>  };</div><div class="line"><a name="l00905"></a><span class="lineno"> 905</span> </div><div class="line"><a name="l00906"></a><span class="lineno"> 906</span>  <span class="comment">// Check if input is a 3D reinterpretation</span></div><div class="line"><a name="l00907"></a><span class="lineno"> 907</span>  <span class="keyword">const</span> <span class="keywordtype">bool</span> reinterpret_as_3d = vector_sum_row != <span class="keyword">nullptr</span></div><div class="line"><a name="l00908"></a><span class="lineno"> 908</span>  && mm_result->info()->num_dimensions() > 1</div><div class="line"><a name="l00909"></a><span class="lineno"> 909</span>  && mm_result->info()->tensor_shape().y() != vector_sum_row->info()->tensor_shape().x();</div><div class="line"><a name="l00910"></a><span class="lineno"> 910</span> </div><div class="line"><a name="l00911"></a><span class="lineno"> 911</span>  <span class="comment">// Check if we need to clamp the result using min and max</span></div><div class="line"><a name="l00912"></a><span class="lineno"> 912</span>  PixelValue <a class="code" href="minmaxloc_8cl.xhtml#a538b4b63f40e7b12891774e03a4f0dec">type_min</a>{};</div><div class="line"><a name="l00913"></a><span class="lineno"> 913</span>  PixelValue <a class="code" href="minmaxloc_8cl.xhtml#a4464d6f922ea17b4a9ca6a2cec7ddb75">type_max</a>{};</div><div class="line"><a name="l00914"></a><span class="lineno"> 914</span>  std::tie(<a class="code" href="minmaxloc_8cl.xhtml#a538b4b63f40e7b12891774e03a4f0dec">type_min</a>, <a class="code" href="minmaxloc_8cl.xhtml#a4464d6f922ea17b4a9ca6a2cec7ddb75">type_max</a>) = <a class="code" href="namespacearm__compute.xhtml#ae69217acf0f0b5d4de030a09ad50a0bc">get_min_max</a>(output->info()->data_type());</div><div class="line"><a name="l00915"></a><span class="lineno"> 915</span>  int32_t type_min_int = <a class="code" href="minmaxloc_8cl.xhtml#a538b4b63f40e7b12891774e03a4f0dec">type_min</a>.get<int32_t>();</div><div class="line"><a name="l00916"></a><span class="lineno"> 916</span>  int32_t type_max_int = <a class="code" href="minmaxloc_8cl.xhtml#a4464d6f922ea17b4a9ca6a2cec7ddb75">type_max</a>.get<int32_t>();</div><div class="line"><a name="l00917"></a><span class="lineno"> 917</span>  <span class="keyword">const</span> <span class="keywordtype">bool</span> is_bounded_relu = !(output_stage.gemmlowp_min_bound == type_min_int && output_stage.gemmlowp_max_bound == type_max_int);</div><div class="line"><a name="l00918"></a><span class="lineno"> 918</span> </div><div class="line"><a name="l00919"></a><span class="lineno"> 919</span>  <span class="comment">// Check if we need to perform fixed point requantization</span></div><div class="line"><a name="l00920"></a><span class="lineno"> 920</span>  <span class="keyword">const</span> <span class="keywordtype">bool</span> is_fixed_point = output_stage.type != <a class="code" href="namespacearm__compute.xhtml#a5558e2cc22f7f4771653d992c8ad8864a079e2ddc95b344b5cb0188bed9a80d8b">GEMMLowpOutputStageType::QUANTIZE_DOWN</a>;</div><div class="line"><a name="l00921"></a><span class="lineno"> 921</span> </div><div class="line"><a name="l00922"></a><span class="lineno"> 922</span>  <span class="comment">// Check if symmetric per-channel execution</span></div><div class="line"><a name="l00923"></a><span class="lineno"> 923</span>  <span class="keyword">const</span> <span class="keywordtype">bool</span> is_signed = output->info()->data_type() == <a class="code" href="namespacearm__compute.xhtml#ad8ed01ff3ff33333d8e19db4d2818bb6a329f5d0c4b0c80e3474951d2c4435dd9">DataType::QASYMM8_SIGNED</a>;</div><div class="line"><a name="l00924"></a><span class="lineno"> 924</span> </div><div class="line"><a name="l00925"></a><span class="lineno"> 925</span>  <span class="comment">// Check if symmetric per-channel execution</span></div><div class="line"><a name="l00926"></a><span class="lineno"> 926</span>  <span class="keyword">const</span> <span class="keywordtype">bool</span> is_symm = output_stage.is_quantized_per_channel;</div><div class="line"><a name="l00927"></a><span class="lineno"> 927</span> </div><div class="line"><a name="l00928"></a><span class="lineno"> 928</span>  <span class="comment">// key acts as a bitset, setting the first bit on reinterpret_as_3d,</span></div><div class="line"><a name="l00929"></a><span class="lineno"> 929</span>  <span class="comment">// the second on is_bounded_relu, and the third on is_fixed_point.</span></div><div class="line"><a name="l00930"></a><span class="lineno"> 930</span>  uint8_t key = (reinterpret_as_3d ? 1UL : 0UL) | ((is_bounded_relu ? 1UL : 0UL) << 1) | ((is_fixed_point ? 1UL : 0UL) << 2);</div><div class="line"><a name="l00931"></a><span class="lineno"> 931</span>  <span class="keywordflow">if</span>(is_symm)</div><div class="line"><a name="l00932"></a><span class="lineno"> 932</span>  {</div><div class="line"><a name="l00933"></a><span class="lineno"> 933</span>  <span class="keywordflow">return</span> map_function_qsymm.find(key)->second;</div><div class="line"><a name="l00934"></a><span class="lineno"> 934</span>  }</div><div class="line"><a name="l00935"></a><span class="lineno"> 935</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00936"></a><span class="lineno"> 936</span>  {</div><div class="line"><a name="l00937"></a><span class="lineno"> 937</span>  key |= ((is_signed ? 1UL : 0UL) << 3);</div><div class="line"><a name="l00938"></a><span class="lineno"> 938</span>  <span class="keywordflow">return</span> map_function_qasymm.find(key)->second;</div><div class="line"><a name="l00939"></a><span class="lineno"> 939</span>  }</div><div class="line"><a name="l00940"></a><span class="lineno"> 940</span> }</div><div class="line"><a name="l00941"></a><span class="lineno"> 941</span> } <span class="comment">// namespace</span></div><div class="line"><a name="l00942"></a><span class="lineno"> 942</span> </div><div class="line"><a name="l00943"></a><span class="lineno"><a class="line" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_offset_contribution_output_stage_kernel.xhtml#a2851f631e9660a4dd9644cb749282723"> 943</a></span> <a class="code" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_offset_contribution_output_stage_kernel.xhtml#a2851f631e9660a4dd9644cb749282723">NEGEMMLowpOffsetContributionOutputStageKernel::NEGEMMLowpOffsetContributionOutputStageKernel</a>()</div><div class="line"><a name="l00944"></a><span class="lineno"> 944</span>  : _function(nullptr), _vector_sum_col(nullptr), _vector_sum_row(nullptr), _bias(nullptr), _mm_result(nullptr), _output(nullptr), _a_offset(0), _b_offset(0), _k_offset(0), _slide_vector_sum_col(true),</div><div class="line"><a name="l00945"></a><span class="lineno"> 945</span>  _output_stage(<a class="code" href="structarm__compute_1_1_g_e_m_m_lowp_output_stage_info.xhtml">GEMMLowpOutputStageInfo</a>())</div><div class="line"><a name="l00946"></a><span class="lineno"> 946</span> </div><div class="line"><a name="l00947"></a><span class="lineno"> 947</span> {</div><div class="line"><a name="l00948"></a><span class="lineno"> 948</span> }</div><div class="line"><a name="l00949"></a><span class="lineno"> 949</span> </div><div class="line"><a name="l00950"></a><span class="lineno"><a class="line" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_offset_contribution_output_stage_kernel.xhtml#a97ebe5c0444a53d58d9b9f079ebe2d0f"> 950</a></span> <span class="keywordtype">void</span> <a class="code" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_offset_contribution_output_stage_kernel.xhtml#a97ebe5c0444a53d58d9b9f079ebe2d0f">NEGEMMLowpOffsetContributionOutputStageKernel::configure</a>(<span class="keyword">const</span> <a class="code" href="classarm__compute_1_1_i_tensor.xhtml">ITensor</a> *mm_result, <span class="keyword">const</span> <a class="code" href="classarm__compute_1_1_i_tensor.xhtml">ITensor</a> *vector_sum_col,</div><div class="line"><a name="l00951"></a><span class="lineno"> 951</span>  <span class="keyword">const</span> <a class="code" href="classarm__compute_1_1_i_tensor.xhtml">ITensor</a> *vector_sum_row, <span class="keyword">const</span> <a class="code" href="classarm__compute_1_1_i_tensor.xhtml">ITensor</a> *<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>, <a class="code" href="classarm__compute_1_1_i_tensor.xhtml">ITensor</a> *output,</div><div class="line"><a name="l00952"></a><span class="lineno"> 952</span>  int32_t k, int32_t a_offset, int32_t b_offset,</div><div class="line"><a name="l00953"></a><span class="lineno"> 953</span>  <a class="code" href="structarm__compute_1_1_g_e_m_m_lowp_output_stage_info.xhtml">GEMMLowpOutputStageInfo</a> output_stage)</div><div class="line"><a name="l00954"></a><span class="lineno"> 954</span> {</div><div class="line"><a name="l00955"></a><span class="lineno"> 955</span>  <span class="comment">// Perform validate step</span></div><div class="line"><a name="l00956"></a><span class="lineno"> 956</span>  <a class="code" href="_validate_8h.xhtml#a921b705e9e3e0fe928928447869e62a5">ARM_COMPUTE_ERROR_ON_NULLPTR</a>(mm_result, output);</div><div class="line"><a name="l00957"></a><span class="lineno"> 957</span> </div><div class="line"><a name="l00958"></a><span class="lineno"> 958</span>  <a class="code" href="_error_8h.xhtml#a938dcd406ce611ef5345ad2531cdb948">ARM_COMPUTE_ERROR_THROW_ON</a>(validate_arguments(mm_result-><a class="code" href="classarm__compute_1_1_i_tensor.xhtml#a0e95dc1e53c361348314873b168ae237">info</a>(),</div><div class="line"><a name="l00959"></a><span class="lineno"> 959</span>  vector_sum_col != <span class="keyword">nullptr</span> ? vector_sum_col-><a class="code" href="classarm__compute_1_1_i_tensor.xhtml#a0e95dc1e53c361348314873b168ae237">info</a>() : <span class="keyword">nullptr</span>, <span class="comment">// NOLINT</span></div><div class="line"><a name="l00960"></a><span class="lineno"> 960</span>  vector_sum_row != <span class="keyword">nullptr</span> ? vector_sum_row-><a class="code" href="classarm__compute_1_1_i_tensor.xhtml#a0e95dc1e53c361348314873b168ae237">info</a>() : <span class="keyword">nullptr</span>, <span class="comment">// NOLINT</span></div><div class="line"><a name="l00961"></a><span class="lineno"> 961</span>  <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a> != <span class="keyword">nullptr</span> ? <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>-><a class="code" href="classarm__compute_1_1_c_l_tensor.xhtml#ad45f0c01a0713dfb6bd7232c7f396fc4">info</a>() : <span class="keyword">nullptr</span>, <span class="comment">// NOLINT</span></div><div class="line"><a name="l00962"></a><span class="lineno"> 962</span>  output-><a class="code" href="classarm__compute_1_1_i_tensor.xhtml#a0e95dc1e53c361348314873b168ae237">info</a>(), a_offset, b_offset, output_stage)); <span class="comment">// NOLINT</span></div><div class="line"><a name="l00963"></a><span class="lineno"> 963</span> </div><div class="line"><a name="l00964"></a><span class="lineno"> 964</span>  _vector_sum_col = vector_sum_col;</div><div class="line"><a name="l00965"></a><span class="lineno"> 965</span>  _vector_sum_row = vector_sum_row;</div><div class="line"><a name="l00966"></a><span class="lineno"> 966</span>  _bias = <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>;</div><div class="line"><a name="l00967"></a><span class="lineno"> 967</span>  _mm_result = mm_result;</div><div class="line"><a name="l00968"></a><span class="lineno"> 968</span>  _output = output;</div><div class="line"><a name="l00969"></a><span class="lineno"> 969</span>  _a_offset = a_offset;</div><div class="line"><a name="l00970"></a><span class="lineno"> 970</span>  _b_offset = b_offset;</div><div class="line"><a name="l00971"></a><span class="lineno"> 971</span>  _k_offset = a_offset * b_offset * k;</div><div class="line"><a name="l00972"></a><span class="lineno"> 972</span>  _output_stage = output_stage;</div><div class="line"><a name="l00973"></a><span class="lineno"> 973</span> </div><div class="line"><a name="l00974"></a><span class="lineno"> 974</span>  <span class="comment">// If a_offset == 0, vector_sum_col can be a nullptr</span></div><div class="line"><a name="l00975"></a><span class="lineno"> 975</span>  <span class="keywordflow">if</span>(a_offset != 0)</div><div class="line"><a name="l00976"></a><span class="lineno"> 976</span>  {</div><div class="line"><a name="l00977"></a><span class="lineno"> 977</span>  <span class="comment">// Check if vector_sum_col_shape should be slidden or not</span></div><div class="line"><a name="l00978"></a><span class="lineno"> 978</span>  <span class="comment">// Don't slide vector_sum_col_shape along the y dimension if vector_sum_col_shape has just 1 dimension and vector_sum_row_shape more than 1</span></div><div class="line"><a name="l00979"></a><span class="lineno"> 979</span>  <span class="comment">// This scenario can happen when the the matrix multiplication is used to perform a convolution operation</span></div><div class="line"><a name="l00980"></a><span class="lineno"> 980</span>  _slide_vector_sum_col = vector_sum_col-><a class="code" href="classarm__compute_1_1_i_tensor.xhtml#a0e95dc1e53c361348314873b168ae237">info</a>()-><a class="code" href="classarm__compute_1_1_i_tensor_info.xhtml#a7c66505457d00ece3aa4b34cab80757d">tensor_shape</a>().<a class="code" href="classarm__compute_1_1_dimensions.xhtml#a80a5f2d6e3a697c9aad893a3b4242615">num_dimensions</a>() > 1;</div><div class="line"><a name="l00981"></a><span class="lineno"> 981</span>  }</div><div class="line"><a name="l00982"></a><span class="lineno"> 982</span> </div><div class="line"><a name="l00983"></a><span class="lineno"> 983</span>  <span class="comment">// Configure kernel window</span></div><div class="line"><a name="l00984"></a><span class="lineno"> 984</span>  <span class="keyword">auto</span> win_config = validate_and_configure_window(mm_result-><a class="code" href="classarm__compute_1_1_i_tensor.xhtml#a0e95dc1e53c361348314873b168ae237">info</a>(), output-><a class="code" href="classarm__compute_1_1_i_tensor.xhtml#a0e95dc1e53c361348314873b168ae237">info</a>());</div><div class="line"><a name="l00985"></a><span class="lineno"> 985</span>  <a class="code" href="_error_8h.xhtml#a938dcd406ce611ef5345ad2531cdb948">ARM_COMPUTE_ERROR_THROW_ON</a>(win_config.first);</div><div class="line"><a name="l00986"></a><span class="lineno"> 986</span>  INEKernel::configure(win_config.second);</div><div class="line"><a name="l00987"></a><span class="lineno"> 987</span> </div><div class="line"><a name="l00988"></a><span class="lineno"> 988</span>  _function = get_configured_function(mm_result, vector_sum_row, output, output_stage);</div><div class="line"><a name="l00989"></a><span class="lineno"> 989</span> }</div><div class="line"><a name="l00990"></a><span class="lineno"> 990</span> </div><div class="line"><a name="l00991"></a><span class="lineno"><a class="line" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_offset_contribution_output_stage_kernel.xhtml#a6296f2754011b2221b343ccedfc0ba35"> 991</a></span> <a class="code" href="classarm__compute_1_1_status.xhtml">Status</a> <a class="code" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_offset_contribution_output_stage_kernel.xhtml#a6296f2754011b2221b343ccedfc0ba35">NEGEMMLowpOffsetContributionOutputStageKernel::validate</a>(<span class="keyword">const</span> <a class="code" href="classarm__compute_1_1_i_tensor_info.xhtml">ITensorInfo</a> *mm_result, <span class="keyword">const</span> <a class="code" href="classarm__compute_1_1_i_tensor_info.xhtml">ITensorInfo</a> *vector_sum_col,</div><div class="line"><a name="l00992"></a><span class="lineno"> 992</span>  <span class="keyword">const</span> <a class="code" href="classarm__compute_1_1_i_tensor_info.xhtml">ITensorInfo</a> *vector_sum_row, <span class="keyword">const</span> <a class="code" href="classarm__compute_1_1_i_tensor_info.xhtml">ITensorInfo</a> *<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>, <span class="keyword">const</span> <a class="code" href="classarm__compute_1_1_i_tensor_info.xhtml">ITensorInfo</a> *output,</div><div class="line"><a name="l00993"></a><span class="lineno"> 993</span>  int32_t a_offset, int32_t b_offset, <a class="code" href="structarm__compute_1_1_g_e_m_m_lowp_output_stage_info.xhtml">GEMMLowpOutputStageInfo</a> output_stage)</div><div class="line"><a name="l00994"></a><span class="lineno"> 994</span> {</div><div class="line"><a name="l00995"></a><span class="lineno"> 995</span>  <a class="code" href="_validate_8h.xhtml#a921b705e9e3e0fe928928447869e62a5">ARM_COMPUTE_ERROR_ON_NULLPTR</a>(mm_result, output);</div><div class="line"><a name="l00996"></a><span class="lineno"> 996</span>  <a class="code" href="_error_8h.xhtml#a8a1e1c105f0bdaf37db408c7cfcb77a4">ARM_COMPUTE_RETURN_ON_ERROR</a>(validate_arguments(mm_result, vector_sum_col, vector_sum_row, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>, output, a_offset, b_offset, output_stage));</div><div class="line"><a name="l00997"></a><span class="lineno"> 997</span>  <a class="code" href="_error_8h.xhtml#a8a1e1c105f0bdaf37db408c7cfcb77a4">ARM_COMPUTE_RETURN_ON_ERROR</a>(validate_and_configure_window(mm_result-><a class="code" href="classarm__compute_1_1misc_1_1_i_cloneable.xhtml#a4d10e5012a872e7f78f2b539b673049d">clone</a>().get(), output-><a class="code" href="classarm__compute_1_1misc_1_1_i_cloneable.xhtml#a4d10e5012a872e7f78f2b539b673049d">clone</a>().get()).first);</div><div class="line"><a name="l00998"></a><span class="lineno"> 998</span>  <span class="keywordflow">return</span> <a class="code" href="classarm__compute_1_1_status.xhtml">Status</a>{};</div><div class="line"><a name="l00999"></a><span class="lineno"> 999</span> }</div><div class="line"><a name="l01000"></a><span class="lineno"> 1000</span> </div><div class="line"><a name="l01001"></a><span class="lineno"><a class="line" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_offset_contribution_output_stage_kernel.xhtml#a112b35dd205c62ea6ed1447ef226da82"> 1001</a></span> <span class="keywordtype">void</span> <a class="code" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_offset_contribution_output_stage_kernel.xhtml#a112b35dd205c62ea6ed1447ef226da82">NEGEMMLowpOffsetContributionOutputStageKernel::run</a>(<span class="keyword">const</span> <a class="code" href="classarm__compute_1_1_window.xhtml">Window</a> &window, <span class="keyword">const</span> <a class="code" href="structarm__compute_1_1_thread_info.xhtml">ThreadInfo</a> &<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a4f4125dba5283887b34f889b1c615c0c">info</a>)</div><div class="line"><a name="l01002"></a><span class="lineno"> 1002</span> {</div><div class="line"><a name="l01003"></a><span class="lineno"> 1003</span>  <a class="code" href="_error_8h.xhtml#a6dc630a6ae9cc063b3924bcea8dee9d6">ARM_COMPUTE_UNUSED</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a4f4125dba5283887b34f889b1c615c0c">info</a>);</div><div class="line"><a name="l01004"></a><span class="lineno"> 1004</span>  <a class="code" href="_validate_8h.xhtml#a1b35b0d258183cf9ef36adf684d0b88c">ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL</a>(<span class="keyword">this</span>);</div><div class="line"><a name="l01005"></a><span class="lineno"> 1005</span>  <a class="code" href="_validate_8h.xhtml#a6eb9ce82815fe429250189da7592ba75">ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW</a>(<a class="code" href="classarm__compute_1_1_i_kernel.xhtml#ad34a46f53686c12a5c5e717cc9617fb6">INEKernel::window</a>(), <a class="code" href="classarm__compute_1_1_i_kernel.xhtml#ad34a46f53686c12a5c5e717cc9617fb6">window</a>);</div><div class="line"><a name="l01006"></a><span class="lineno"> 1006</span>  _function(<a class="code" href="classarm__compute_1_1_i_kernel.xhtml#ad34a46f53686c12a5c5e717cc9617fb6">window</a>, _mm_result, _vector_sum_col, _vector_sum_row, _bias, _output, _a_offset, _b_offset, _k_offset, _slide_vector_sum_col, _output_stage);</div><div class="line"><a name="l01007"></a><span class="lineno"> 1007</span> }</div><div class="line"><a name="l01008"></a><span class="lineno"> 1008</span> </div><div class="line"><a name="l01009"></a><span class="lineno"> 1009</span> } <span class="comment">// namespace arm_compute</span></div><div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_a009469e4d9b8fce3b6d5e97d2077827d"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_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="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00510">helpers.h:510</a></div></div> |
