Jenkins | 52ba29e | 2018-08-29 15:32:11 +0000 | [diff] [blame^] | 120 | <a href="gemmlowp_8cl.xhtml">Go to the documentation of this file.</a><div class="fragment"><div class="line"><a name="l00001"></a><span class="lineno"> 1</span> <span class="comment">/*</span></div><div class="line"><a name="l00002"></a><span class="lineno"> 2</span> <span class="comment"> * Copyright (c) 2017-2018 ARM Limited.</span></div><div class="line"><a name="l00003"></a><span class="lineno"> 3</span> <span class="comment"> *</span></div><div class="line"><a name="l00004"></a><span class="lineno"> 4</span> <span class="comment"> * SPDX-License-Identifier: MIT</span></div><div class="line"><a name="l00005"></a><span class="lineno"> 5</span> <span class="comment"> *</span></div><div class="line"><a name="l00006"></a><span class="lineno"> 6</span> <span class="comment"> * Permission is hereby granted, free of charge, to any person obtaining a copy</span></div><div class="line"><a name="l00007"></a><span class="lineno"> 7</span> <span class="comment"> * of this software and associated documentation files (the "Software"), to</span></div><div class="line"><a name="l00008"></a><span class="lineno"> 8</span> <span class="comment"> * deal in the Software without restriction, including without limitation the</span></div><div class="line"><a name="l00009"></a><span class="lineno"> 9</span> <span class="comment"> * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or</span></div><div class="line"><a name="l00010"></a><span class="lineno"> 10</span> <span class="comment"> * sell copies of the Software, and to permit persons to whom the Software is</span></div><div class="line"><a name="l00011"></a><span class="lineno"> 11</span> <span class="comment"> * furnished to do so, subject to the following conditions:</span></div><div class="line"><a name="l00012"></a><span class="lineno"> 12</span> <span class="comment"> *</span></div><div class="line"><a name="l00013"></a><span class="lineno"> 13</span> <span class="comment"> * The above copyright notice and this permission notice shall be included in all</span></div><div class="line"><a name="l00014"></a><span class="lineno"> 14</span> <span class="comment"> * copies or substantial portions of the Software.</span></div><div class="line"><a name="l00015"></a><span class="lineno"> 15</span> <span class="comment"> *</span></div><div class="line"><a name="l00016"></a><span class="lineno"> 16</span> <span class="comment"> * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR</span></div><div class="line"><a name="l00017"></a><span class="lineno"> 17</span> <span class="comment"> * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,</span></div><div class="line"><a name="l00018"></a><span class="lineno"> 18</span> <span class="comment"> * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE</span></div><div class="line"><a name="l00019"></a><span class="lineno"> 19</span> <span class="comment"> * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER</span></div><div class="line"><a name="l00020"></a><span class="lineno"> 20</span> <span class="comment"> * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,</span></div><div class="line"><a name="l00021"></a><span class="lineno"> 21</span> <span class="comment"> * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE</span></div><div class="line"><a name="l00022"></a><span class="lineno"> 22</span> <span class="comment"> * SOFTWARE.</span></div><div class="line"><a name="l00023"></a><span class="lineno"> 23</span> <span class="comment"> */</span></div><div class="line"><a name="l00024"></a><span class="lineno"> 24</span> <span class="preprocessor">#include "<a class="code" href="helpers_8h.xhtml">helpers.h</a>"</span></div><div class="line"><a name="l00025"></a><span class="lineno"> 25</span> <span class="preprocessor">#include "<a class="code" href="helpers__asymm_8h.xhtml">helpers_asymm.h</a>"</span></div><div class="line"><a name="l00026"></a><span class="lineno"> 26</span> </div><div class="line"><a name="l00027"></a><span class="lineno"> 27</span> <span class="preprocessor">#if defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED)</span></div><div class="line"><a name="l00028"></a><span class="lineno"> 28</span> <span class="preprocessor">#if defined(ARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED)</span></div><div class="line"><a name="l00029"></a><span class="lineno"> 29</span> <span class="preprocessor">#define ARM_DOT(x0, x1, x2, x3, y0, y1, y2, y3, val) val = arm_dot_acc((uchar4)(x0, x1, x2, x3), (uchar4)(y0, y1, y2, y3), val);</span></div><div class="line"><a name="l00030"></a><span class="lineno"> 30</span> <span class="preprocessor">#else // defined(ARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED)</span></div><div class="line"><a name="l00031"></a><span class="lineno"> 31</span> <span class="preprocessor">#define ARM_DOT(x0, x1, x2, x3, y0, y1, y2, y3, val) val += arm_dot((uchar4)(x0, x1, x2, x3), (uchar4)(y0, y1, y2, y3));</span></div><div class="line"><a name="l00032"></a><span class="lineno"> 32</span> <span class="preprocessor">#endif // defined(ARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED)</span></div><div class="line"><a name="l00033"></a><span class="lineno"> 33</span> <span class="preprocessor">#endif // defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED)</span></div><div class="line"><a name="l00034"></a><span class="lineno"> 34</span> </div><div class="line"><a name="l00035"></a><span class="lineno"> 35</span> <span class="preprocessor">#if defined(COLS_B) && defined(MULT_INTERLEAVE4X4_HEIGHT) && defined(TRANSPOSE1XW_WIDTH_STEP)</span></div><div class="line"><a name="l00036"></a><span class="lineno"> 36</span> </div><div class="line"><a name="l00062"></a><span class="lineno"> 62</span> __kernel <span class="keywordtype">void</span> gemmlowp_mm_interleaved_transposed_midgard(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div><div class="line"><a name="l00063"></a><span class="lineno"> 63</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div><div class="line"><a name="l00064"></a><span class="lineno"> 64</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l00065"></a><span class="lineno"> 65</span> {</div><div class="line"><a name="l00066"></a><span class="lineno"> 66</span>  <span class="keywordtype">int</span> x = get_global_id(0) / TRANSPOSE1XW_WIDTH_STEP;</div><div class="line"><a name="l00067"></a><span class="lineno"> 67</span>  <span class="keywordtype">int</span> y = get_global_id(1) / MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l00068"></a><span class="lineno"> 68</span> </div><div class="line"><a name="l00069"></a><span class="lineno"> 69</span>  <span class="comment">// Offset</span></div><div class="line"><a name="l00070"></a><span class="lineno"> 70</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_a = (get_global_id(1) % MULT_INTERLEAVE4X4_HEIGHT) * 4;</div><div class="line"><a name="l00071"></a><span class="lineno"> 71</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_b = (get_global_id(0) % TRANSPOSE1XW_WIDTH_STEP) * 4;</div><div class="line"><a name="l00072"></a><span class="lineno"> 72</span> </div><div class="line"><a name="l00073"></a><span class="lineno"> 73</span>  <span class="comment">// src_addr_a = address of matrix A</span></div><div class="line"><a name="l00074"></a><span class="lineno"> 74</span>  <span class="comment">// src_addr_b = address of matrix B</span></div><div class="line"><a name="l00075"></a><span class="lineno"> 75</span>  __global uchar *src_addr_a = (__global uchar *)(src0_ptr + y * src0_stride_y + src0_offset_first_element_in_bytes);</div><div class="line"><a name="l00076"></a><span class="lineno"> 76</span>  __global uchar *src_addr_b = (__global uchar *)(src1_ptr + x * src1_stride_y + src1_offset_first_element_in_bytes);</div><div class="line"><a name="l00077"></a><span class="lineno"> 77</span> </div><div class="line"><a name="l00078"></a><span class="lineno"> 78</span>  <span class="comment">// Compute end row address for matrix B</span></div><div class="line"><a name="l00079"></a><span class="lineno"> 79</span>  __global uchar *src_end_addr_b = src_addr_b + COLS_B;</div><div class="line"><a name="l00080"></a><span class="lineno"> 80</span> </div><div class="line"><a name="l00081"></a><span class="lineno"> 81</span>  src_addr_a += offset_row_a;</div><div class="line"><a name="l00082"></a><span class="lineno"> 82</span>  src_addr_b += offset_row_b;</div><div class="line"><a name="l00083"></a><span class="lineno"> 83</span> </div><div class="line"><a name="l00084"></a><span class="lineno"> 84</span>  <span class="comment">// Reset accumulators</span></div><div class="line"><a name="l00085"></a><span class="lineno"> 85</span>  int4 c00 = 0;</div><div class="line"><a name="l00086"></a><span class="lineno"> 86</span>  int4 c10 = 0;</div><div class="line"><a name="l00087"></a><span class="lineno"> 87</span>  int4 c20 = 0;</div><div class="line"><a name="l00088"></a><span class="lineno"> 88</span>  int4 c30 = 0;</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">for</span>(; src_addr_b <= (src_end_addr_b - (int)(8 * TRANSPOSE1XW_WIDTH_STEP)); src_addr_a += 8 * MULT_INTERLEAVE4X4_HEIGHT, src_addr_b += 8 * TRANSPOSE1XW_WIDTH_STEP)</div><div class="line"><a name="l00091"></a><span class="lineno"> 91</span>  {</div><div class="line"><a name="l00092"></a><span class="lineno"> 92</span>  <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l00093"></a><span class="lineno"> 93</span>  int4 a0 = convert_int4(vload4(0, src_addr_a));</div><div class="line"><a name="l00094"></a><span class="lineno"> 94</span>  int4 b0 = convert_int4(vload4(0, src_addr_b));</div><div class="line"><a name="l00095"></a><span class="lineno"> 95</span> </div><div class="line"><a name="l00096"></a><span class="lineno"> 96</span>  c00 += (int4)a0.s0 * b0;</div><div class="line"><a name="l00097"></a><span class="lineno"> 97</span>  c10 += (int4)a0.s1 * b0;</div><div class="line"><a name="l00098"></a><span class="lineno"> 98</span>  c20 += (int4)a0.s2 * b0;</div><div class="line"><a name="l00099"></a><span class="lineno"> 99</span>  c30 += (int4)a0.s3 * b0;</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>  a0 = convert_int4(vload4(0, src_addr_a + 4 * MULT_INTERLEAVE4X4_HEIGHT));</div><div class="line"><a name="l00102"></a><span class="lineno"> 102</span>  b0 = convert_int4(vload4(0, src_addr_b + 4 * TRANSPOSE1XW_WIDTH_STEP));</div><div class="line"><a name="l00103"></a><span class="lineno"> 103</span> </div><div class="line"><a name="l00104"></a><span class="lineno"> 104</span>  c00 += (int4)a0.s0 * b0;</div><div class="line"><a name="l00105"></a><span class="lineno"> 105</span>  c10 += (int4)a0.s1 * b0;</div><div class="line"><a name="l00106"></a><span class="lineno"> 106</span>  c20 += (int4)a0.s2 * b0;</div><div class="line"><a name="l00107"></a><span class="lineno"> 107</span>  c30 += (int4)a0.s3 * b0;</div><div class="line"><a name="l00108"></a><span class="lineno"> 108</span>  }</div><div class="line"><a name="l00109"></a><span class="lineno"> 109</span> </div><div class="line"><a name="l00110"></a><span class="lineno"> 110</span>  <span class="keywordflow">for</span>(; src_addr_b < src_end_addr_b; src_addr_a += (4 * MULT_INTERLEAVE4X4_HEIGHT), src_addr_b += (4 * TRANSPOSE1XW_WIDTH_STEP))</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>  <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l00113"></a><span class="lineno"> 113</span>  int4 a0 = convert_int4(vload4(0, src_addr_a));</div><div class="line"><a name="l00114"></a><span class="lineno"> 114</span>  int4 b0 = convert_int4(vload4(0, src_addr_b));</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>  c00 += (int4)a0.s0 * b0;</div><div class="line"><a name="l00117"></a><span class="lineno"> 117</span>  c10 += (int4)a0.s1 * b0;</div><div class="line"><a name="l00118"></a><span class="lineno"> 118</span>  c20 += (int4)a0.s2 * b0;</div><div class="line"><a name="l00119"></a><span class="lineno"> 119</span>  c30 += (int4)a0.s3 * b0;</div><div class="line"><a name="l00120"></a><span class="lineno"> 120</span>  }</div><div class="line"><a name="l00121"></a><span class="lineno"> 121</span> </div><div class="line"><a name="l00122"></a><span class="lineno"> 122</span>  <span class="comment">// Compute destination address</span></div><div class="line"><a name="l00123"></a><span class="lineno"> 123</span>  <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a> = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</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>  <span class="comment">// Store 4x4 block</span></div><div class="line"><a name="l00126"></a><span class="lineno"> 126</span>  vstore4(c00, 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div><div class="line"><a name="l00127"></a><span class="lineno"> 127</span>  vstore4(c10, 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 1)));</div><div class="line"><a name="l00128"></a><span class="lineno"> 128</span>  vstore4(c20, 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 2)));</div><div class="line"><a name="l00129"></a><span class="lineno"> 129</span>  vstore4(c30, 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 3)));</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> </div><div class="line"><a name="l00158"></a><span class="lineno"> 158</span> __kernel <span class="keywordtype">void</span> gemmlowp_mm_interleaved_transposed_bifrost(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div><div class="line"><a name="l00159"></a><span class="lineno"> 159</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div><div class="line"><a name="l00160"></a><span class="lineno"> 160</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div><div class="line"><a name="l00161"></a><span class="lineno"> 161</span> {</div><div class="line"><a name="l00162"></a><span class="lineno"> 162</span>  <span class="keywordtype">int</span> x = get_global_id(0) / TRANSPOSE1XW_WIDTH_STEP;</div><div class="line"><a name="l00163"></a><span class="lineno"> 163</span>  <span class="keywordtype">int</span> y = get_global_id(1) / MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l00164"></a><span class="lineno"> 164</span> </div><div class="line"><a name="l00165"></a><span class="lineno"> 165</span>  <span class="comment">// Offset</span></div><div class="line"><a name="l00166"></a><span class="lineno"> 166</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_a = (get_global_id(1) % MULT_INTERLEAVE4X4_HEIGHT) * 4;</div><div class="line"><a name="l00167"></a><span class="lineno"> 167</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_b = (get_global_id(0) % TRANSPOSE1XW_WIDTH_STEP) * 4;</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">// src_addr_a = address of matrix A</span></div><div class="line"><a name="l00170"></a><span class="lineno"> 170</span>  <span class="comment">// src_addr_b = address of matrix B</span></div><div class="line"><a name="l00171"></a><span class="lineno"> 171</span>  __global uchar *src_addr_a = (__global uchar *)(src0_ptr + y * src0_stride_y + src0_offset_first_element_in_bytes);</div><div class="line"><a name="l00172"></a><span class="lineno"> 172</span>  __global uchar *src_addr_b = (__global uchar *)(src1_ptr + x * src1_stride_y + src1_offset_first_element_in_bytes);</div><div class="line"><a name="l00173"></a><span class="lineno"> 173</span> </div><div class="line"><a name="l00174"></a><span class="lineno"> 174</span>  <span class="comment">// Compute end row address for matrix B</span></div><div class="line"><a name="l00175"></a><span class="lineno"> 175</span>  __global uchar *src_end_addr_b = src_addr_b + COLS_B;</div><div class="line"><a name="l00176"></a><span class="lineno"> 176</span> </div><div class="line"><a name="l00177"></a><span class="lineno"> 177</span>  src_addr_a += offset_row_a;</div><div class="line"><a name="l00178"></a><span class="lineno"> 178</span>  src_addr_b += offset_row_b;</div><div class="line"><a name="l00179"></a><span class="lineno"> 179</span> </div><div class="line"><a name="l00180"></a><span class="lineno"> 180</span>  <span class="comment">// Reset accumulators</span></div><div class="line"><a name="l00181"></a><span class="lineno"> 181</span>  uint c00 = 0;</div><div class="line"><a name="l00182"></a><span class="lineno"> 182</span>  uint c01 = 0;</div><div class="line"><a name="l00183"></a><span class="lineno"> 183</span>  uint c02 = 0;</div><div class="line"><a name="l00184"></a><span class="lineno"> 184</span>  uint c03 = 0;</div><div class="line"><a name="l00185"></a><span class="lineno"> 185</span>  uint c10 = 0;</div><div class="line"><a name="l00186"></a><span class="lineno"> 186</span>  uint c11 = 0;</div><div class="line"><a name="l00187"></a><span class="lineno"> 187</span>  uint c12 = 0;</div><div class="line"><a name="l00188"></a><span class="lineno"> 188</span>  uint c13 = 0;</div><div class="line"><a name="l00189"></a><span class="lineno"> 189</span>  uint c20 = 0;</div><div class="line"><a name="l00190"></a><span class="lineno"> 190</span>  uint c21 = 0;</div><div class="line"><a name="l00191"></a><span class="lineno"> 191</span>  uint c22 = 0;</div><div class="line"><a name="l00192"></a><span class="lineno"> 192</span>  uint c23 = 0;</div><div class="line"><a name="l00193"></a><span class="lineno"> 193</span>  uint c30 = 0;</div><div class="line"><a name="l00194"></a><span class="lineno"> 194</span>  uint c31 = 0;</div><div class="line"><a name="l00195"></a><span class="lineno"> 195</span>  uint c32 = 0;</div><div class="line"><a name="l00196"></a><span class="lineno"> 196</span>  uint c33 = 0;</div><div class="line"><a name="l00197"></a><span class="lineno"> 197</span> </div><div class="line"><a name="l00198"></a><span class="lineno"> 198</span> <span class="preprocessor">#if MULT_INTERLEAVE4X4_HEIGHT == 1</span></div><div class="line"><a name="l00199"></a><span class="lineno"> 199</span>  <span class="keywordflow">for</span>(; src_addr_b <= (src_end_addr_b - (int)(32 * TRANSPOSE1XW_WIDTH_STEP)); src_addr_a += (32 * MULT_INTERLEAVE4X4_HEIGHT), src_addr_b += (32 * TRANSPOSE1XW_WIDTH_STEP))</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">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l00202"></a><span class="lineno"> 202</span>  uchar16 a0 = vload16(0, src_addr_a);</div><div class="line"><a name="l00203"></a><span class="lineno"> 203</span>  uchar4 b0 = vload4(0, src_addr_b);</div><div class="line"><a name="l00204"></a><span class="lineno"> 204</span> </div><div class="line"><a name="l00205"></a><span class="lineno"> 205</span>  c00 += (ushort)a0.s0 * b0.s0;</div><div class="line"><a name="l00206"></a><span class="lineno"> 206</span>  c01 += (ushort)a0.s0 * b0.s1;</div><div class="line"><a name="l00207"></a><span class="lineno"> 207</span>  c02 += (ushort)a0.s0 * b0.s2;</div><div class="line"><a name="l00208"></a><span class="lineno"> 208</span>  c03 += (ushort)a0.s0 * b0.s3;</div><div class="line"><a name="l00209"></a><span class="lineno"> 209</span> </div><div class="line"><a name="l00210"></a><span class="lineno"> 210</span>  c10 += (ushort)a0.s1 * b0.s0;</div><div class="line"><a name="l00211"></a><span class="lineno"> 211</span>  c11 += (ushort)a0.s1 * b0.s1;</div><div class="line"><a name="l00212"></a><span class="lineno"> 212</span>  c12 += (ushort)a0.s1 * b0.s2;</div><div class="line"><a name="l00213"></a><span class="lineno"> 213</span>  c13 += (ushort)a0.s1 * b0.s3;</div><div class="line"><a name="l00214"></a><span class="lineno"> 214</span> </div><div class="line"><a name="l00215"></a><span class="lineno"> 215</span>  c20 += (ushort)a0.s2 * b0.s0;</div><div class="line"><a name="l00216"></a><span class="lineno"> 216</span>  c21 += (ushort)a0.s2 * b0.s1;</div><div class="line"><a name="l00217"></a><span class="lineno"> 217</span>  c22 += (ushort)a0.s2 * b0.s2;</div><div class="line"><a name="l00218"></a><span class="lineno"> 218</span>  c23 += (ushort)a0.s2 * b0.s3;</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>  c30 += (ushort)a0.s3 * b0.s0;</div><div class="line"><a name="l00221"></a><span class="lineno"> 221</span>  c31 += (ushort)a0.s3 * b0.s1;</div><div class="line"><a name="l00222"></a><span class="lineno"> 222</span>  c32 += (ushort)a0.s3 * b0.s2;</div><div class="line"><a name="l00223"></a><span class="lineno"> 223</span>  c33 += (ushort)a0.s3 * b0.s3;</div><div class="line"><a name="l00224"></a><span class="lineno"> 224</span> </div><div class="line"><a name="l00225"></a><span class="lineno"> 225</span>  <span class="comment">// Load values from matrix B (transposed)</span></div><div class="line"><a name="l00226"></a><span class="lineno"> 226</span>  b0 = vload4(0, src_addr_b + 4 * TRANSPOSE1XW_WIDTH_STEP);</div><div class="line"><a name="l00227"></a><span class="lineno"> 227</span> </div><div class="line"><a name="l00228"></a><span class="lineno"> 228</span>  c00 += (ushort)a0.s4 * b0.s0;</div><div class="line"><a name="l00229"></a><span class="lineno"> 229</span>  c01 += (ushort)a0.s4 * b0.s1;</div><div class="line"><a name="l00230"></a><span class="lineno"> 230</span>  c02 += (ushort)a0.s4 * b0.s2;</div><div class="line"><a name="l00231"></a><span class="lineno"> 231</span>  c03 += (ushort)a0.s4 * b0.s3;</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>  c10 += (ushort)a0.s5 * b0.s0;</div><div class="line"><a name="l00234"></a><span class="lineno"> 234</span>  c11 += (ushort)a0.s5 * b0.s1;</div><div class="line"><a name="l00235"></a><span class="lineno"> 235</span>  c12 += (ushort)a0.s5 * b0.s2;</div><div class="line"><a name="l00236"></a><span class="lineno"> 236</span>  c13 += (ushort)a0.s5 * b0.s3;</div><div class="line"><a name="l00237"></a><span class="lineno"> 237</span> </div><div class="line"><a name="l00238"></a><span class="lineno"> 238</span>  c20 += (ushort)a0.s6 * b0.s0;</div><div class="line"><a name="l00239"></a><span class="lineno"> 239</span>  c21 += (ushort)a0.s6 * b0.s1;</div><div class="line"><a name="l00240"></a><span class="lineno"> 240</span>  c22 += (ushort)a0.s6 * b0.s2;</div><div class="line"><a name="l00241"></a><span class="lineno"> 241</span>  c23 += (ushort)a0.s6 * b0.s3;</div><div class="line"><a name="l00242"></a><span class="lineno"> 242</span> </div><div class="line"><a name="l00243"></a><span class="lineno"> 243</span>  c30 += (ushort)a0.s7 * b0.s0;</div><div class="line"><a name="l00244"></a><span class="lineno"> 244</span>  c31 += (ushort)a0.s7 * b0.s1;</div><div class="line"><a name="l00245"></a><span class="lineno"> 245</span>  c32 += (ushort)a0.s7 * b0.s2;</div><div class="line"><a name="l00246"></a><span class="lineno"> 246</span>  c33 += (ushort)a0.s7 * b0.s3;</div><div class="line"><a name="l00247"></a><span class="lineno"> 247</span> </div><div class="line"><a name="l00248"></a><span class="lineno"> 248</span>  <span class="comment">// Load values from matrix B (transposed)</span></div><div class="line"><a name="l00249"></a><span class="lineno"> 249</span>  b0 = vload4(0, src_addr_b + 8 * TRANSPOSE1XW_WIDTH_STEP);</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>  c00 += (ushort)a0.s8 * b0.s0;</div><div class="line"><a name="l00252"></a><span class="lineno"> 252</span>  c01 += (ushort)a0.s8 * b0.s1;</div><div class="line"><a name="l00253"></a><span class="lineno"> 253</span>  c02 += (ushort)a0.s8 * b0.s2;</div><div class="line"><a name="l00254"></a><span class="lineno"> 254</span>  c03 += (ushort)a0.s8 * b0.s3;</div><div class="line"><a name="l00255"></a><span class="lineno"> 255</span> </div><div class="line"><a name="l00256"></a><span class="lineno"> 256</span>  c10 += (ushort)a0.s9 * b0.s0;</div><div class="line"><a name="l00257"></a><span class="lineno"> 257</span>  c11 += (ushort)a0.s9 * b0.s1;</div><div class="line"><a name="l00258"></a><span class="lineno"> 258</span>  c12 += (ushort)a0.s9 * b0.s2;</div><div class="line"><a name="l00259"></a><span class="lineno"> 259</span>  c13 += (ushort)a0.s9 * b0.s3;</div><div class="line"><a name="l00260"></a><span class="lineno"> 260</span> </div><div class="line"><a name="l00261"></a><span class="lineno"> 261</span>  c20 += (ushort)a0.sA * b0.s0;</div><div class="line"><a name="l00262"></a><span class="lineno"> 262</span>  c21 += (ushort)a0.sA * b0.s1;</div><div class="line"><a name="l00263"></a><span class="lineno"> 263</span>  c22 += (ushort)a0.sA * b0.s2;</div><div class="line"><a name="l00264"></a><span class="lineno"> 264</span>  c23 += (ushort)a0.sA * b0.s3;</div><div class="line"><a name="l00265"></a><span class="lineno"> 265</span> </div><div class="line"><a name="l00266"></a><span class="lineno"> 266</span>  c30 += (ushort)a0.sB * b0.s0;</div><div class="line"><a name="l00267"></a><span class="lineno"> 267</span>  c31 += (ushort)a0.sB * b0.s1;</div><div class="line"><a name="l00268"></a><span class="lineno"> 268</span>  c32 += (ushort)a0.sB * b0.s2;</div><div class="line"><a name="l00269"></a><span class="lineno"> 269</span>  c33 += (ushort)a0.sB * b0.s3;</div><div class="line"><a name="l00270"></a><span class="lineno"> 270</span> </div><div class="line"><a name="l00271"></a><span class="lineno"> 271</span>  <span class="comment">// Load values from matrix B (transposed)</span></div><div class="line"><a name="l00272"></a><span class="lineno"> 272</span>  b0 = vload4(0, src_addr_b + 12 * TRANSPOSE1XW_WIDTH_STEP);</div><div class="line"><a name="l00273"></a><span class="lineno"> 273</span> </div><div class="line"><a name="l00274"></a><span class="lineno"> 274</span>  c00 += (ushort)a0.sC * b0.s0;</div><div class="line"><a name="l00275"></a><span class="lineno"> 275</span>  c01 += (ushort)a0.sC * b0.s1;</div><div class="line"><a name="l00276"></a><span class="lineno"> 276</span>  c02 += (ushort)a0.sC * b0.s2;</div><div class="line"><a name="l00277"></a><span class="lineno"> 277</span>  c03 += (ushort)a0.sC * b0.s3;</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>  c10 += (ushort)a0.sD * b0.s0;</div><div class="line"><a name="l00280"></a><span class="lineno"> 280</span>  c11 += (ushort)a0.sD * b0.s1;</div><div class="line"><a name="l00281"></a><span class="lineno"> 281</span>  c12 += (ushort)a0.sD * b0.s2;</div><div class="line"><a name="l00282"></a><span class="lineno"> 282</span>  c13 += (ushort)a0.sD * b0.s3;</div><div class="line"><a name="l00283"></a><span class="lineno"> 283</span> </div><div class="line"><a name="l00284"></a><span class="lineno"> 284</span>  c20 += (ushort)a0.sE * b0.s0;</div><div class="line"><a name="l00285"></a><span class="lineno"> 285</span>  c21 += (ushort)a0.sE * b0.s1;</div><div class="line"><a name="l00286"></a><span class="lineno"> 286</span>  c22 += (ushort)a0.sE * b0.s2;</div><div class="line"><a name="l00287"></a><span class="lineno"> 287</span>  c23 += (ushort)a0.sE * b0.s3;</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>  c30 += (ushort)a0.sF * b0.s0;</div><div class="line"><a name="l00290"></a><span class="lineno"> 290</span>  c31 += (ushort)a0.sF * b0.s1;</div><div class="line"><a name="l00291"></a><span class="lineno"> 291</span>  c32 += (ushort)a0.sF * b0.s2;</div><div class="line"><a name="l00292"></a><span class="lineno"> 292</span>  c33 += (ushort)a0.sF * b0.s3;</div><div class="line"><a name="l00293"></a><span class="lineno"> 293</span> </div><div class="line"><a name="l00294"></a><span class="lineno"> 294</span>  <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l00295"></a><span class="lineno"> 295</span>  a0 = vload16(0, src_addr_a + 16);</div><div class="line"><a name="l00296"></a><span class="lineno"> 296</span>  b0 = vload4(0, src_addr_b + 16 * TRANSPOSE1XW_WIDTH_STEP);</div><div class="line"><a name="l00297"></a><span class="lineno"> 297</span> </div><div class="line"><a name="l00298"></a><span class="lineno"> 298</span>  c00 += (ushort)a0.s0 * b0.s0;</div><div class="line"><a name="l00299"></a><span class="lineno"> 299</span>  c01 += (ushort)a0.s0 * b0.s1;</div><div class="line"><a name="l00300"></a><span class="lineno"> 300</span>  c02 += (ushort)a0.s0 * b0.s2;</div><div class="line"><a name="l00301"></a><span class="lineno"> 301</span>  c03 += (ushort)a0.s0 * b0.s3;</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>  c10 += (ushort)a0.s1 * b0.s0;</div><div class="line"><a name="l00304"></a><span class="lineno"> 304</span>  c11 += (ushort)a0.s1 * b0.s1;</div><div class="line"><a name="l00305"></a><span class="lineno"> 305</span>  c12 += (ushort)a0.s1 * b0.s2;</div><div class="line"><a name="l00306"></a><span class="lineno"> 306</span>  c13 += (ushort)a0.s1 * b0.s3;</div><div class="line"><a name="l00307"></a><span class="lineno"> 307</span> </div><div class="line"><a name="l00308"></a><span class="lineno"> 308</span>  c20 += (ushort)a0.s2 * b0.s0;</div><div class="line"><a name="l00309"></a><span class="lineno"> 309</span>  c21 += (ushort)a0.s2 * b0.s1;</div><div class="line"><a name="l00310"></a><span class="lineno"> 310</span>  c22 += (ushort)a0.s2 * b0.s2;</div><div class="line"><a name="l00311"></a><span class="lineno"> 311</span>  c23 += (ushort)a0.s2 * b0.s3;</div><div class="line"><a name="l00312"></a><span class="lineno"> 312</span> </div><div class="line"><a name="l00313"></a><span class="lineno"> 313</span>  c30 += (ushort)a0.s3 * b0.s0;</div><div class="line"><a name="l00314"></a><span class="lineno"> 314</span>  c31 += (ushort)a0.s3 * b0.s1;</div><div class="line"><a name="l00315"></a><span class="lineno"> 315</span>  c32 += (ushort)a0.s3 * b0.s2;</div><div class="line"><a name="l00316"></a><span class="lineno"> 316</span>  c33 += (ushort)a0.s3 * b0.s3;</div><div class="line"><a name="l00317"></a><span class="lineno"> 317</span> </div><div class="line"><a name="l00318"></a><span class="lineno"> 318</span>  <span class="comment">// Load values from matrix B (transposed)</span></div><div class="line"><a name="l00319"></a><span class="lineno"> 319</span>  b0 = vload4(0, src_addr_b + 20 * TRANSPOSE1XW_WIDTH_STEP);</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>  c00 += (ushort)a0.s4 * b0.s0;</div><div class="line"><a name="l00322"></a><span class="lineno"> 322</span>  c01 += (ushort)a0.s4 * b0.s1;</div><div class="line"><a name="l00323"></a><span class="lineno"> 323</span>  c02 += (ushort)a0.s4 * b0.s2;</div><div class="line"><a name="l00324"></a><span class="lineno"> 324</span>  c03 += (ushort)a0.s4 * b0.s3;</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>  c10 += (ushort)a0.s5 * b0.s0;</div><div class="line"><a name="l00327"></a><span class="lineno"> 327</span>  c11 += (ushort)a0.s5 * b0.s1;</div><div class="line"><a name="l00328"></a><span class="lineno"> 328</span>  c12 += (ushort)a0.s5 * b0.s2;</div><div class="line"><a name="l00329"></a><span class="lineno"> 329</span>  c13 += (ushort)a0.s5 * b0.s3;</div><div class="line"><a name="l00330"></a><span class="lineno"> 330</span> </div><div class="line"><a name="l00331"></a><span class="lineno"> 331</span>  c20 += (ushort)a0.s6 * b0.s0;</div><div class="line"><a name="l00332"></a><span class="lineno"> 332</span>  c21 += (ushort)a0.s6 * b0.s1;</div><div class="line"><a name="l00333"></a><span class="lineno"> 333</span>  c22 += (ushort)a0.s6 * b0.s2;</div><div class="line"><a name="l00334"></a><span class="lineno"> 334</span>  c23 += (ushort)a0.s6 * b0.s3;</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>  c30 += (ushort)a0.s7 * b0.s0;</div><div class="line"><a name="l00337"></a><span class="lineno"> 337</span>  c31 += (ushort)a0.s7 * b0.s1;</div><div class="line"><a name="l00338"></a><span class="lineno"> 338</span>  c32 += (ushort)a0.s7 * b0.s2;</div><div class="line"><a name="l00339"></a><span class="lineno"> 339</span>  c33 += (ushort)a0.s7 * b0.s3;</div><div class="line"><a name="l00340"></a><span class="lineno"> 340</span> </div><div class="line"><a name="l00341"></a><span class="lineno"> 341</span>  <span class="comment">// Load values from matrix B (transposed)</span></div><div class="line"><a name="l00342"></a><span class="lineno"> 342</span>  b0 = vload4(0, src_addr_b + 24 * TRANSPOSE1XW_WIDTH_STEP);</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>  c00 += (ushort)a0.s8 * b0.s0;</div><div class="line"><a name="l00345"></a><span class="lineno"> 345</span>  c01 += (ushort)a0.s8 * b0.s1;</div><div class="line"><a name="l00346"></a><span class="lineno"> 346</span>  c02 += (ushort)a0.s8 * b0.s2;</div><div class="line"><a name="l00347"></a><span class="lineno"> 347</span>  c03 += (ushort)a0.s8 * b0.s3;</div><div class="line"><a name="l00348"></a><span class="lineno"> 348</span> </div><div class="line"><a name="l00349"></a><span class="lineno"> 349</span>  c10 += (ushort)a0.s9 * b0.s0;</div><div class="line"><a name="l00350"></a><span class="lineno"> 350</span>  c11 += (ushort)a0.s9 * b0.s1;</div><div class="line"><a name="l00351"></a><span class="lineno"> 351</span>  c12 += (ushort)a0.s9 * b0.s2;</div><div class="line"><a name="l00352"></a><span class="lineno"> 352</span>  c13 += (ushort)a0.s9 * b0.s3;</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>  c20 += (ushort)a0.sA * b0.s0;</div><div class="line"><a name="l00355"></a><span class="lineno"> 355</span>  c21 += (ushort)a0.sA * b0.s1;</div><div class="line"><a name="l00356"></a><span class="lineno"> 356</span>  c22 += (ushort)a0.sA * b0.s2;</div><div class="line"><a name="l00357"></a><span class="lineno"> 357</span>  c23 += (ushort)a0.sA * b0.s3;</div><div class="line"><a name="l00358"></a><span class="lineno"> 358</span> </div><div class="line"><a name="l00359"></a><span class="lineno"> 359</span>  c30 += (ushort)a0.sB * b0.s0;</div><div class="line"><a name="l00360"></a><span class="lineno"> 360</span>  c31 += (ushort)a0.sB * b0.s1;</div><div class="line"><a name="l00361"></a><span class="lineno"> 361</span>  c32 += (ushort)a0.sB * b0.s2;</div><div class="line"><a name="l00362"></a><span class="lineno"> 362</span>  c33 += (ushort)a0.sB * b0.s3;</div><div class="line"><a name="l00363"></a><span class="lineno"> 363</span> </div><div class="line"><a name="l00364"></a><span class="lineno"> 364</span>  <span class="comment">// Load values from matrix B (transposed)</span></div><div class="line"><a name="l00365"></a><span class="lineno"> 365</span>  b0 = vload4(0, src_addr_b + 28 * TRANSPOSE1XW_WIDTH_STEP);</div><div class="line"><a name="l00366"></a><span class="lineno"> 366</span> </div><div class="line"><a name="l00367"></a><span class="lineno"> 367</span>  c00 += (ushort)a0.sC * b0.s0;</div><div class="line"><a name="l00368"></a><span class="lineno"> 368</span>  c01 += (ushort)a0.sC * b0.s1;</div><div class="line"><a name="l00369"></a><span class="lineno"> 369</span>  c02 += (ushort)a0.sC * b0.s2;</div><div class="line"><a name="l00370"></a><span class="lineno"> 370</span>  c03 += (ushort)a0.sC * b0.s3;</div><div class="line"><a name="l00371"></a><span class="lineno"> 371</span> </div><div class="line"><a name="l00372"></a><span class="lineno"> 372</span>  c10 += (ushort)a0.sD * b0.s0;</div><div class="line"><a name="l00373"></a><span class="lineno"> 373</span>  c11 += (ushort)a0.sD * b0.s1;</div><div class="line"><a name="l00374"></a><span class="lineno"> 374</span>  c12 += (ushort)a0.sD * b0.s2;</div><div class="line"><a name="l00375"></a><span class="lineno"> 375</span>  c13 += (ushort)a0.sD * b0.s3;</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>  c20 += (ushort)a0.sE * b0.s0;</div><div class="line"><a name="l00378"></a><span class="lineno"> 378</span>  c21 += (ushort)a0.sE * b0.s1;</div><div class="line"><a name="l00379"></a><span class="lineno"> 379</span>  c22 += (ushort)a0.sE * b0.s2;</div><div class="line"><a name="l00380"></a><span class="lineno"> 380</span>  c23 += (ushort)a0.sE * b0.s3;</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>  c30 += (ushort)a0.sF * b0.s0;</div><div class="line"><a name="l00383"></a><span class="lineno"> 383</span>  c31 += (ushort)a0.sF * b0.s1;</div><div class="line"><a name="l00384"></a><span class="lineno"> 384</span>  c32 += (ushort)a0.sF * b0.s2;</div><div class="line"><a name="l00385"></a><span class="lineno"> 385</span>  c33 += (ushort)a0.sF * b0.s3;</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="preprocessor">#endif // MULT_INTERLEAVE4X4_HEIGHT == 1</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="keywordflow">for</span>(; src_addr_b < src_end_addr_b; src_addr_a += (4 * MULT_INTERLEAVE4X4_HEIGHT), src_addr_b += (4 * TRANSPOSE1XW_WIDTH_STEP))</div><div class="line"><a name="l00390"></a><span class="lineno"> 390</span>  {</div><div class="line"><a name="l00391"></a><span class="lineno"> 391</span>  <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l00392"></a><span class="lineno"> 392</span>  uchar4 a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l00393"></a><span class="lineno"> 393</span>  uchar4 b0 = vload4(0, src_addr_b);</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>  c00 += (ushort)a0.s0 * b0.s0;</div><div class="line"><a name="l00396"></a><span class="lineno"> 396</span>  c01 += (ushort)a0.s0 * b0.s1;</div><div class="line"><a name="l00397"></a><span class="lineno"> 397</span>  c02 += (ushort)a0.s0 * b0.s2;</div><div class="line"><a name="l00398"></a><span class="lineno"> 398</span>  c03 += (ushort)a0.s0 * b0.s3;</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>  c10 += (ushort)a0.s1 * b0.s0;</div><div class="line"><a name="l00401"></a><span class="lineno"> 401</span>  c11 += (ushort)a0.s1 * b0.s1;</div><div class="line"><a name="l00402"></a><span class="lineno"> 402</span>  c12 += (ushort)a0.s1 * b0.s2;</div><div class="line"><a name="l00403"></a><span class="lineno"> 403</span>  c13 += (ushort)a0.s1 * b0.s3;</div><div class="line"><a name="l00404"></a><span class="lineno"> 404</span> </div><div class="line"><a name="l00405"></a><span class="lineno"> 405</span>  c20 += (ushort)a0.s2 * b0.s0;</div><div class="line"><a name="l00406"></a><span class="lineno"> 406</span>  c21 += (ushort)a0.s2 * b0.s1;</div><div class="line"><a name="l00407"></a><span class="lineno"> 407</span>  c22 += (ushort)a0.s2 * b0.s2;</div><div class="line"><a name="l00408"></a><span class="lineno"> 408</span>  c23 += (ushort)a0.s2 * b0.s3;</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>  c30 += (ushort)a0.s3 * b0.s0;</div><div class="line"><a name="l00411"></a><span class="lineno"> 411</span>  c31 += (ushort)a0.s3 * b0.s1;</div><div class="line"><a name="l00412"></a><span class="lineno"> 412</span>  c32 += (ushort)a0.s3 * b0.s2;</div><div class="line"><a name="l00413"></a><span class="lineno"> 413</span>  c33 += (ushort)a0.s3 * b0.s3;</div><div class="line"><a name="l00414"></a><span class="lineno"> 414</span>  }</div><div class="line"><a name="l00415"></a><span class="lineno"> 415</span> </div><div class="line"><a name="l00416"></a><span class="lineno"> 416</span>  <span class="comment">// Compute destination address</span></div><div class="line"><a name="l00417"></a><span class="lineno"> 417</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l00418"></a><span class="lineno"> 418</span> </div><div class="line"><a name="l00419"></a><span class="lineno"> 419</span>  <span class="comment">// Store 4x4 block</span></div><div class="line"><a name="l00420"></a><span class="lineno"> 420</span>  vstore4((int4)(c00, c01, c02, c03), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div><div class="line"><a name="l00421"></a><span class="lineno"> 421</span>  vstore4((int4)(c10, c11, c12, c13), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 1)));</div><div class="line"><a name="l00422"></a><span class="lineno"> 422</span>  vstore4((int4)(c20, c21, c22, c23), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 2)));</div><div class="line"><a name="l00423"></a><span class="lineno"> 423</span>  vstore4((int4)(c30, c31, c32, c33), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 3)));</div><div class="line"><a name="l00424"></a><span class="lineno"> 424</span> }</div><div class="line"><a name="l00425"></a><span class="lineno"> 425</span> </div><div class="line"><a name="l00426"></a><span class="lineno"> 426</span> <span class="preprocessor">#if ARM_COMPUTE_OPENCL_DOT8_ENABLED</span></div><div class="line"><a name="l00427"></a><span class="lineno"> 427</span> </div><div class="line"><a name="l00453"></a><span class="lineno"> 453</span> __kernel <span class="keywordtype">void</span> gemmlowp_mm_interleaved_transposed_bifrost_dot8(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div><div class="line"><a name="l00454"></a><span class="lineno"> 454</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div><div class="line"><a name="l00455"></a><span class="lineno"> 455</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div><div class="line"><a name="l00456"></a><span class="lineno"> 456</span> {</div><div class="line"><a name="l00457"></a><span class="lineno"> 457</span>  <span class="keywordtype">int</span> x = get_global_id(0) / TRANSPOSE1XW_WIDTH_STEP;</div><div class="line"><a name="l00458"></a><span class="lineno"> 458</span>  <span class="keywordtype">int</span> y = get_global_id(1) / MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l00459"></a><span class="lineno"> 459</span> </div><div class="line"><a name="l00460"></a><span class="lineno"> 460</span>  <span class="comment">// Offset</span></div><div class="line"><a name="l00461"></a><span class="lineno"> 461</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_a = (get_global_id(1) % MULT_INTERLEAVE4X4_HEIGHT) * 4;</div><div class="line"><a name="l00462"></a><span class="lineno"> 462</span>  <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_b = (get_global_id(0) % TRANSPOSE1XW_WIDTH_STEP) * 4;</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="comment">// src_addr_a = address of matrix A</span></div><div class="line"><a name="l00465"></a><span class="lineno"> 465</span>  <span class="comment">// src_addr_b = address of matrix B</span></div><div class="line"><a name="l00466"></a><span class="lineno"> 466</span>  __global uchar *src_addr_a = (__global uchar *)(src0_ptr + y * src0_stride_y + src0_offset_first_element_in_bytes);</div><div class="line"><a name="l00467"></a><span class="lineno"> 467</span>  __global uchar *src_addr_b = (__global uchar *)(src1_ptr + x * src1_stride_y + src1_offset_first_element_in_bytes);</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">// Compute end row address for matrix B</span></div><div class="line"><a name="l00470"></a><span class="lineno"> 470</span>  __global uchar *src_end_addr_b = src_addr_b + COLS_B;</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>  src_addr_a += offset_row_a;</div><div class="line"><a name="l00473"></a><span class="lineno"> 473</span>  src_addr_b += offset_row_b;</div><div class="line"><a name="l00474"></a><span class="lineno"> 474</span> </div><div class="line"><a name="l00475"></a><span class="lineno"> 475</span>  <span class="comment">// Reset accumulators</span></div><div class="line"><a name="l00476"></a><span class="lineno"> 476</span>  uint c00 = 0;</div><div class="line"><a name="l00477"></a><span class="lineno"> 477</span>  uint c01 = 0;</div><div class="line"><a name="l00478"></a><span class="lineno"> 478</span>  uint c02 = 0;</div><div class="line"><a name="l00479"></a><span class="lineno"> 479</span>  uint c03 = 0;</div><div class="line"><a name="l00480"></a><span class="lineno"> 480</span>  uint c10 = 0;</div><div class="line"><a name="l00481"></a><span class="lineno"> 481</span>  uint c11 = 0;</div><div class="line"><a name="l00482"></a><span class="lineno"> 482</span>  uint c12 = 0;</div><div class="line"><a name="l00483"></a><span class="lineno"> 483</span>  uint c13 = 0;</div><div class="line"><a name="l00484"></a><span class="lineno"> 484</span>  uint c20 = 0;</div><div class="line"><a name="l00485"></a><span class="lineno"> 485</span>  uint c21 = 0;</div><div class="line"><a name="l00486"></a><span class="lineno"> 486</span>  uint c22 = 0;</div><div class="line"><a name="l00487"></a><span class="lineno"> 487</span>  uint c23 = 0;</div><div class="line"><a name="l00488"></a><span class="lineno"> 488</span>  uint c30 = 0;</div><div class="line"><a name="l00489"></a><span class="lineno"> 489</span>  uint c31 = 0;</div><div class="line"><a name="l00490"></a><span class="lineno"> 490</span>  uint c32 = 0;</div><div class="line"><a name="l00491"></a><span class="lineno"> 491</span>  uint c33 = 0;</div><div class="line"><a name="l00492"></a><span class="lineno"> 492</span> </div><div class="line"><a name="l00493"></a><span class="lineno"> 493</span> <span class="preprocessor">#if MULT_INTERLEAVE4X4_HEIGHT == 1</span></div><div class="line"><a name="l00494"></a><span class="lineno"> 494</span>  <span class="keywordflow">for</span>(; src_addr_b <= (src_end_addr_b - (int)(32 * TRANSPOSE1XW_WIDTH_STEP)); src_addr_a += (32 * MULT_INTERLEAVE4X4_HEIGHT), src_addr_b += (32 * TRANSPOSE1XW_WIDTH_STEP))</div><div class="line"><a name="l00495"></a><span class="lineno"> 495</span>  {</div><div class="line"><a name="l00496"></a><span class="lineno"> 496</span>  <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l00497"></a><span class="lineno"> 497</span>  uchar16 a0 = vload16(0, src_addr_a);</div><div class="line"><a name="l00498"></a><span class="lineno"> 498</span>  uchar4 b0 = vload4(0, src_addr_b);</div><div class="line"><a name="l00499"></a><span class="lineno"> 499</span>  uchar4 b1 = vload4(0, src_addr_b + 4 * TRANSPOSE1XW_WIDTH_STEP);</div><div class="line"><a name="l00500"></a><span class="lineno"> 500</span>  uchar4 b2 = vload4(0, src_addr_b + 8 * TRANSPOSE1XW_WIDTH_STEP);</div><div class="line"><a name="l00501"></a><span class="lineno"> 501</span>  uchar4 b3 = vload4(0, src_addr_b + 12 * TRANSPOSE1XW_WIDTH_STEP);</div><div class="line"><a name="l00502"></a><span class="lineno"> 502</span> </div><div class="line"><a name="l00503"></a><span class="lineno"> 503</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l00504"></a><span class="lineno"> 504</span>  ARM_DOT(a0.s0, a0.s4, a0.s8, a0.sC, b0.s0, b1.s0, b2.s0, b3.s0, c00);</div><div class="line"><a name="l00505"></a><span class="lineno"> 505</span>  ARM_DOT(a0.s0, a0.s4, a0.s8, a0.sC, b0.s1, b1.s1, b2.s1, b3.s1, c01);</div><div class="line"><a name="l00506"></a><span class="lineno"> 506</span>  ARM_DOT(a0.s0, a0.s4, a0.s8, a0.sC, b0.s2, b1.s2, b2.s2, b3.s2, c02);</div><div class="line"><a name="l00507"></a><span class="lineno"> 507</span>  ARM_DOT(a0.s0, a0.s4, a0.s8, a0.sC, b0.s3, b1.s3, b2.s3, b3.s3, c03);</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>  ARM_DOT(a0.s1, a0.s5, a0.s9, a0.sD, b0.s0, b1.s0, b2.s0, b3.s0, c10);</div><div class="line"><a name="l00510"></a><span class="lineno"> 510</span>  ARM_DOT(a0.s1, a0.s5, a0.s9, a0.sD, b0.s1, b1.s1, b2.s1, b3.s1, c11);</div><div class="line"><a name="l00511"></a><span class="lineno"> 511</span>  ARM_DOT(a0.s1, a0.s5, a0.s9, a0.sD, b0.s2, b1.s2, b2.s2, b3.s2, c12);</div><div class="line"><a name="l00512"></a><span class="lineno"> 512</span>  ARM_DOT(a0.s1, a0.s5, a0.s9, a0.sD, b0.s3, b1.s3, b2.s3, b3.s3, c13);</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>  ARM_DOT(a0.s2, a0.s6, a0.sA, a0.sE, b0.s0, b1.s0, b2.s0, b3.s0, c20);</div><div class="line"><a name="l00515"></a><span class="lineno"> 515</span>  ARM_DOT(a0.s2, a0.s6, a0.sA, a0.sE, b0.s1, b1.s1, b2.s1, b3.s1, c21);</div><div class="line"><a name="l00516"></a><span class="lineno"> 516</span>  ARM_DOT(a0.s2, a0.s6, a0.sA, a0.sE, b0.s2, b1.s2, b2.s2, b3.s2, c22);</div><div class="line"><a name="l00517"></a><span class="lineno"> 517</span>  ARM_DOT(a0.s2, a0.s6, a0.sA, a0.sE, b0.s3, b1.s3, b2.s3, b3.s3, c23);</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>  ARM_DOT(a0.s3, a0.s7, a0.sB, a0.sF, b0.s0, b1.s0, b2.s0, b3.s0, c30);</div><div class="line"><a name="l00520"></a><span class="lineno"> 520</span>  ARM_DOT(a0.s3, a0.s7, a0.sB, a0.sF, b0.s1, b1.s1, b2.s1, b3.s1, c31);</div><div class="line"><a name="l00521"></a><span class="lineno"> 521</span>  ARM_DOT(a0.s3, a0.s7, a0.sB, a0.sF, b0.s2, b1.s2, b2.s2, b3.s2, c32);</div><div class="line"><a name="l00522"></a><span class="lineno"> 522</span>  ARM_DOT(a0.s3, a0.s7, a0.sB, a0.sF, b0.s3, b1.s3, b2.s3, b3.s3, c33);</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">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l00525"></a><span class="lineno"> 525</span>  a0 = vload16(0, src_addr_a + 16);</div><div class="line"><a name="l00526"></a><span class="lineno"> 526</span>  b0 = vload4(0, src_addr_b + 16 * TRANSPOSE1XW_WIDTH_STEP);</div><div class="line"><a name="l00527"></a><span class="lineno"> 527</span>  b1 = vload4(0, src_addr_b + 20 * TRANSPOSE1XW_WIDTH_STEP);</div><div class="line"><a name="l00528"></a><span class="lineno"> 528</span>  b2 = vload4(0, src_addr_b + 24 * TRANSPOSE1XW_WIDTH_STEP);</div><div class="line"><a name="l00529"></a><span class="lineno"> 529</span>  b3 = vload4(0, src_addr_b + 28 * TRANSPOSE1XW_WIDTH_STEP);</div><div class="line"><a name="l00530"></a><span class="lineno"> 530</span> </div><div class="line"><a name="l00531"></a><span class="lineno"> 531</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l00532"></a><span class="lineno"> 532</span>  ARM_DOT(a0.s0, a0.s4, a0.s8, a0.sC, b0.s0, b1.s0, b2.s0, b3.s0, c00);</div><div class="line"><a name="l00533"></a><span class="lineno"> 533</span>  ARM_DOT(a0.s0, a0.s4, a0.s8, a0.sC, b0.s1, b1.s1, b2.s1, b3.s1, c01);</div><div class="line"><a name="l00534"></a><span class="lineno"> 534</span>  ARM_DOT(a0.s0, a0.s4, a0.s8, a0.sC, b0.s2, b1.s2, b2.s2, b3.s2, c02);</div><div class="line"><a name="l00535"></a><span class="lineno"> 535</span>  ARM_DOT(a0.s0, a0.s4, a0.s8, a0.sC, b0.s3, b1.s3, b2.s3, b3.s3, c03);</div><div class="line"><a name="l00536"></a><span class="lineno"> 536</span> </div><div class="line"><a name="l00537"></a><span class="lineno"> 537</span>  ARM_DOT(a0.s1, a0.s5, a0.s9, a0.sD, b0.s0, b1.s0, b2.s0, b3.s0, c10);</div><div class="line"><a name="l00538"></a><span class="lineno"> 538</span>  ARM_DOT(a0.s1, a0.s5, a0.s9, a0.sD, b0.s1, b1.s1, b2.s1, b3.s1, c11);</div><div class="line"><a name="l00539"></a><span class="lineno"> 539</span>  ARM_DOT(a0.s1, a0.s5, a0.s9, a0.sD, b0.s2, b1.s2, b2.s2, b3.s2, c12);</div><div class="line"><a name="l00540"></a><span class="lineno"> 540</span>  ARM_DOT(a0.s1, a0.s5, a0.s9, a0.sD, b0.s3, b1.s3, b2.s3, b3.s3, c13);</div><div class="line"><a name="l00541"></a><span class="lineno"> 541</span> </div><div class="line"><a name="l00542"></a><span class="lineno"> 542</span>  ARM_DOT(a0.s2, a0.s6, a0.sA, a0.sE, b0.s0, b1.s0, b2.s0, b3.s0, c20);</div><div class="line"><a name="l00543"></a><span class="lineno"> 543</span>  ARM_DOT(a0.s2, a0.s6, a0.sA, a0.sE, b0.s1, b1.s1, b2.s1, b3.s1, c21);</div><div class="line"><a name="l00544"></a><span class="lineno"> 544</span>  ARM_DOT(a0.s2, a0.s6, a0.sA, a0.sE, b0.s2, b1.s2, b2.s2, b3.s2, c22);</div><div class="line"><a name="l00545"></a><span class="lineno"> 545</span>  ARM_DOT(a0.s2, a0.s6, a0.sA, a0.sE, b0.s3, b1.s3, b2.s3, b3.s3, c23);</div><div class="line"><a name="l00546"></a><span class="lineno"> 546</span> </div><div class="line"><a name="l00547"></a><span class="lineno"> 547</span>  ARM_DOT(a0.s3, a0.s7, a0.sB, a0.sF, b0.s0, b1.s0, b2.s0, b3.s0, c30);</div><div class="line"><a name="l00548"></a><span class="lineno"> 548</span>  ARM_DOT(a0.s3, a0.s7, a0.sB, a0.sF, b0.s1, b1.s1, b2.s1, b3.s1, c31);</div><div class="line"><a name="l00549"></a><span class="lineno"> 549</span>  ARM_DOT(a0.s3, a0.s7, a0.sB, a0.sF, b0.s2, b1.s2, b2.s2, b3.s2, c32);</div><div class="line"><a name="l00550"></a><span class="lineno"> 550</span>  ARM_DOT(a0.s3, a0.s7, a0.sB, a0.sF, b0.s3, b1.s3, b2.s3, b3.s3, c33);</div><div class="line"><a name="l00551"></a><span class="lineno"> 551</span>  }</div><div class="line"><a name="l00552"></a><span class="lineno"> 552</span> <span class="preprocessor">#endif // MULT_INTERLEAVE4X4_HEIGHT == 1</span></div><div class="line"><a name="l00553"></a><span class="lineno"> 553</span> </div><div class="line"><a name="l00554"></a><span class="lineno"> 554</span>  <span class="keywordflow">for</span>(; src_addr_b < src_end_addr_b; src_addr_a += (4 * MULT_INTERLEAVE4X4_HEIGHT), src_addr_b += (4 * TRANSPOSE1XW_WIDTH_STEP))</div><div class="line"><a name="l00555"></a><span class="lineno"> 555</span>  {</div><div class="line"><a name="l00556"></a><span class="lineno"> 556</span>  <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l00557"></a><span class="lineno"> 557</span>  uchar4 a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l00558"></a><span class="lineno"> 558</span>  uchar4 b0 = vload4(0, src_addr_b);</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>  c00 += (ushort)a0.s0 * b0.s0;</div><div class="line"><a name="l00561"></a><span class="lineno"> 561</span>  c01 += (ushort)a0.s0 * b0.s1;</div><div class="line"><a name="l00562"></a><span class="lineno"> 562</span>  c02 += (ushort)a0.s0 * b0.s2;</div><div class="line"><a name="l00563"></a><span class="lineno"> 563</span>  c03 += (ushort)a0.s0 * b0.s3;</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>  c10 += (ushort)a0.s1 * b0.s0;</div><div class="line"><a name="l00566"></a><span class="lineno"> 566</span>  c11 += (ushort)a0.s1 * b0.s1;</div><div class="line"><a name="l00567"></a><span class="lineno"> 567</span>  c12 += (ushort)a0.s1 * b0.s2;</div><div class="line"><a name="l00568"></a><span class="lineno"> 568</span>  c13 += (ushort)a0.s1 * b0.s3;</div><div class="line"><a name="l00569"></a><span class="lineno"> 569</span> </div><div class="line"><a name="l00570"></a><span class="lineno"> 570</span>  c20 += (ushort)a0.s2 * b0.s0;</div><div class="line"><a name="l00571"></a><span class="lineno"> 571</span>  c21 += (ushort)a0.s2 * b0.s1;</div><div class="line"><a name="l00572"></a><span class="lineno"> 572</span>  c22 += (ushort)a0.s2 * b0.s2;</div><div class="line"><a name="l00573"></a><span class="lineno"> 573</span>  c23 += (ushort)a0.s2 * b0.s3;</div><div class="line"><a name="l00574"></a><span class="lineno"> 574</span> </div><div class="line"><a name="l00575"></a><span class="lineno"> 575</span>  c30 += (ushort)a0.s3 * b0.s0;</div><div class="line"><a name="l00576"></a><span class="lineno"> 576</span>  c31 += (ushort)a0.s3 * b0.s1;</div><div class="line"><a name="l00577"></a><span class="lineno"> 577</span>  c32 += (ushort)a0.s3 * b0.s2;</div><div class="line"><a name="l00578"></a><span class="lineno"> 578</span>  c33 += (ushort)a0.s3 * b0.s3;</div><div class="line"><a name="l00579"></a><span class="lineno"> 579</span>  }</div><div class="line"><a name="l00580"></a><span class="lineno"> 580</span> </div><div class="line"><a name="l00581"></a><span class="lineno"> 581</span>  <span class="comment">// Compute destination address</span></div><div class="line"><a name="l00582"></a><span class="lineno"> 582</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l00583"></a><span class="lineno"> 583</span> </div><div class="line"><a name="l00584"></a><span class="lineno"> 584</span>  <span class="comment">// Store 4x4 block</span></div><div class="line"><a name="l00585"></a><span class="lineno"> 585</span>  vstore4((int4)(c00, c01, c02, c03), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div><div class="line"><a name="l00586"></a><span class="lineno"> 586</span>  vstore4((int4)(c10, c11, c12, c13), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 1)));</div><div class="line"><a name="l00587"></a><span class="lineno"> 587</span>  vstore4((int4)(c20, c21, c22, c23), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 2)));</div><div class="line"><a name="l00588"></a><span class="lineno"> 588</span>  vstore4((int4)(c30, c31, c32, c33), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 3)));</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> <span class="preprocessor">#endif // ARM_COMPUTE_OPENCL_DOT8_ENABLED</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="preprocessor">#endif // defined(COLS_B) && defined(MULT_INTERLEAVE4X4_HEIGHT) && defined(TRANSPOSE1XW_WIDTH_STEP)</span></div><div class="line"><a name="l00593"></a><span class="lineno"> 593</span> </div><div class="line"><a name="l00594"></a><span class="lineno"> 594</span> <span class="preprocessor">#if defined(NUM_ELEMS_PROCESSED_PER_THREAD_X) && defined(NUM_ELEMS_PROCESSED_PER_THREAD_Y) && defined(COLS_A)</span></div><div class="line"><a name="l00595"></a><span class="lineno"> 595</span> <span class="preprocessor">#define VECTOR_UCHAR VEC_DATA_TYPE(uchar, NUM_ELEMS_PROCESSED_PER_THREAD_X)</span></div><div class="line"><a name="l00596"></a><span class="lineno"> 596</span> <span class="preprocessor">#define VECTOR_UINT VEC_DATA_TYPE(uint, NUM_ELEMS_PROCESSED_PER_THREAD_X)</span></div><div class="line"><a name="l00597"></a><span class="lineno"> 597</span> <span class="preprocessor">#define VECTOR_INT VEC_DATA_TYPE(int, NUM_ELEMS_PROCESSED_PER_THREAD_X)</span></div><div class="line"><a name="l00598"></a><span class="lineno"> 598</span> </div><div class="line"><a name="l00621"></a><span class="lineno"> 621</span> __kernel <span class="keywordtype">void</span> gemmlowp_mm_midgard(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div><div class="line"><a name="l00622"></a><span class="lineno"> 622</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div><div class="line"><a name="l00623"></a><span class="lineno"> 623</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div><div class="line"><a name="l00624"></a><span class="lineno"> 624</span> {</div><div class="line"><a name="l00625"></a><span class="lineno"> 625</span>  <span class="keywordtype">int</span> idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_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>  <span class="comment">// Compute starting address for matrix A and Matrix B</span></div><div class="line"><a name="l00628"></a><span class="lineno"> 628</span>  int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div><div class="line"><a name="l00629"></a><span class="lineno"> 629</span> </div><div class="line"><a name="l00630"></a><span class="lineno"> 630</span>  <span class="comment">// Update address for the matrix A</span></div><div class="line"><a name="l00631"></a><span class="lineno"> 631</span>  src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y;</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="comment">// Update address for the matrix B</span></div><div class="line"><a name="l00634"></a><span class="lineno"> 634</span>  src_addr.s1 += idx;</div><div class="line"><a name="l00635"></a><span class="lineno"> 635</span> </div><div class="line"><a name="l00636"></a><span class="lineno"> 636</span>  <span class="keywordtype">int</span> end_row_vec_a = src_addr.s0 + COLS_A;</div><div class="line"><a name="l00637"></a><span class="lineno"> 637</span> </div><div class="line"><a name="l00638"></a><span class="lineno"> 638</span>  VECTOR_UINT acc0 = 0;</div><div class="line"><a name="l00639"></a><span class="lineno"> 639</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00640"></a><span class="lineno"> 640</span>  VECTOR_UINT acc1 = 0;</div><div class="line"><a name="l00641"></a><span class="lineno"> 641</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00642"></a><span class="lineno"> 642</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00643"></a><span class="lineno"> 643</span>  VECTOR_UINT acc2 = 0;</div><div class="line"><a name="l00644"></a><span class="lineno"> 644</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00645"></a><span class="lineno"> 645</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00646"></a><span class="lineno"> 646</span>  VECTOR_UINT acc3 = 0;</div><div class="line"><a name="l00647"></a><span class="lineno"> 647</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00648"></a><span class="lineno"> 648</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00649"></a><span class="lineno"> 649</span>  VECTOR_UINT acc4 = 0;</div><div class="line"><a name="l00650"></a><span class="lineno"> 650</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00651"></a><span class="lineno"> 651</span> </div><div class="line"><a name="l00652"></a><span class="lineno"> 652</span>  <span class="keywordflow">for</span>(; src_addr.s0 <= (end_row_vec_a - 2); src_addr += (int2)(2, 2 * src1_stride_y))</div><div class="line"><a name="l00653"></a><span class="lineno"> 653</span>  {</div><div class="line"><a name="l00654"></a><span class="lineno"> 654</span>  <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l00655"></a><span class="lineno"> 655</span>  uchar2 a0 = vload2(0, src0_ptr + src_addr.s0 + 0 * src0_stride_y);</div><div class="line"><a name="l00656"></a><span class="lineno"> 656</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00657"></a><span class="lineno"> 657</span>  uchar2 a1 = vload2(0, src0_ptr + src_addr.s0 + 1 * src0_stride_y);</div><div class="line"><a name="l00658"></a><span class="lineno"> 658</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00659"></a><span class="lineno"> 659</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00660"></a><span class="lineno"> 660</span>  uchar2 a2 = vload2(0, src0_ptr + src_addr.s0 + 2 * src0_stride_y);</div><div class="line"><a name="l00661"></a><span class="lineno"> 661</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00662"></a><span class="lineno"> 662</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00663"></a><span class="lineno"> 663</span>  uchar2 a3 = vload2(0, src0_ptr + src_addr.s0 + 3 * src0_stride_y);</div><div class="line"><a name="l00664"></a><span class="lineno"> 664</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00665"></a><span class="lineno"> 665</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00666"></a><span class="lineno"> 666</span>  uchar2 a4 = vload2(0, src0_ptr + src_addr.s0 + 4 * src0_stride_y);</div><div class="line"><a name="l00667"></a><span class="lineno"> 667</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00668"></a><span class="lineno"> 668</span>  <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l00669"></a><span class="lineno"> 669</span>  VECTOR_UCHAR b0 = <a class="code" href="helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)(0, src1_ptr + src_addr.s1);</div><div class="line"><a name="l00670"></a><span class="lineno"> 670</span>  VECTOR_UCHAR b1 = <a class="code" href="helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)(0, src1_ptr + src_addr.s1 + src1_stride_y);</div><div class="line"><a name="l00671"></a><span class="lineno"> 671</span> </div><div class="line"><a name="l00672"></a><span class="lineno"> 672</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l00673"></a><span class="lineno"> 673</span>  acc0 += <a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a0.s0;</div><div class="line"><a name="l00674"></a><span class="lineno"> 674</span>  acc0 += <a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b1, VECTOR_UINT) * (VECTOR_UINT)a0.s1;</div><div class="line"><a name="l00675"></a><span class="lineno"> 675</span> #<span class="keywordflow">if</span> NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</div><div class="line"><a name="l00676"></a><span class="lineno"> 676</span>  acc1 += <a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a1.s0;</div><div class="line"><a name="l00677"></a><span class="lineno"> 677</span>  acc1 += <a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b1, VECTOR_UINT) * (VECTOR_UINT)a1.s1;</div><div class="line"><a name="l00678"></a><span class="lineno"> 678</span> #endif <span class="comment">// NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00679"></a><span class="lineno"> 679</span> #<span class="keywordflow">if</span> NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</div><div class="line"><a name="l00680"></a><span class="lineno"> 680</span>  acc2 += <a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a2.s0;</div><div class="line"><a name="l00681"></a><span class="lineno"> 681</span>  acc2 += <a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b1, VECTOR_UINT) * (VECTOR_UINT)a2.s1;</div><div class="line"><a name="l00682"></a><span class="lineno"> 682</span> #endif <span class="comment">// NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00683"></a><span class="lineno"> 683</span> #<span class="keywordflow">if</span> NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</div><div class="line"><a name="l00684"></a><span class="lineno"> 684</span>  acc3 += <a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a3.s0;</div><div class="line"><a name="l00685"></a><span class="lineno"> 685</span>  acc3 += <a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b1, VECTOR_UINT) * (VECTOR_UINT)a3.s1;</div><div class="line"><a name="l00686"></a><span class="lineno"> 686</span> #endif <span class="comment">// NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00687"></a><span class="lineno"> 687</span> #<span class="keywordflow">if</span> NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</div><div class="line"><a name="l00688"></a><span class="lineno"> 688</span>  acc4 += <a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a4.s0;</div><div class="line"><a name="l00689"></a><span class="lineno"> 689</span>  acc4 += <a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b1, VECTOR_UINT) * (VECTOR_UINT)a4.s1;</div><div class="line"><a name="l00690"></a><span class="lineno"> 690</span> #endif <span class="comment">// NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00691"></a><span class="lineno"> 691</span>  }</div><div class="line"><a name="l00692"></a><span class="lineno"> 692</span> </div><div class="line"><a name="l00693"></a><span class="lineno"> 693</span>  <span class="keywordflow">for</span>(; src_addr.s0 < end_row_vec_a; src_addr += (int2)(1, src1_stride_y))</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="comment">// Load values from matrix A</span></div><div class="line"><a name="l00696"></a><span class="lineno"> 696</span>  uchar a0 = *(src0_ptr + src_addr.s0 + 0 * src0_stride_y);</div><div class="line"><a name="l00697"></a><span class="lineno"> 697</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00698"></a><span class="lineno"> 698</span>  uchar a1 = *(src0_ptr + src_addr.s0 + 1 * src0_stride_y);</div><div class="line"><a name="l00699"></a><span class="lineno"> 699</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00700"></a><span class="lineno"> 700</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00701"></a><span class="lineno"> 701</span>  uchar a2 = *(src0_ptr + src_addr.s0 + 2 * src0_stride_y);</div><div class="line"><a name="l00702"></a><span class="lineno"> 702</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00703"></a><span class="lineno"> 703</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00704"></a><span class="lineno"> 704</span>  uchar a3 = *(src0_ptr + src_addr.s0 + 3 * src0_stride_y);</div><div class="line"><a name="l00705"></a><span class="lineno"> 705</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00706"></a><span class="lineno"> 706</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00707"></a><span class="lineno"> 707</span>  uchar a4 = *(src0_ptr + src_addr.s0 + 4 * src0_stride_y);</div><div class="line"><a name="l00708"></a><span class="lineno"> 708</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00709"></a><span class="lineno"> 709</span>  <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l00710"></a><span class="lineno"> 710</span>  VECTOR_UCHAR b0 = <a class="code" href="helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)(0, src1_ptr + src_addr.s1);</div><div class="line"><a name="l00711"></a><span class="lineno"> 711</span> </div><div class="line"><a name="l00712"></a><span class="lineno"> 712</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l00713"></a><span class="lineno"> 713</span>  acc0 += <a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a0;</div><div class="line"><a name="l00714"></a><span class="lineno"> 714</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00715"></a><span class="lineno"> 715</span>  acc1 += <a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a1;</div><div class="line"><a name="l00716"></a><span class="lineno"> 716</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00717"></a><span class="lineno"> 717</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00718"></a><span class="lineno"> 718</span>  acc2 += <a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a2;</div><div class="line"><a name="l00719"></a><span class="lineno"> 719</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00720"></a><span class="lineno"> 720</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00721"></a><span class="lineno"> 721</span>  acc3 += <a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a3;</div><div class="line"><a name="l00722"></a><span class="lineno"> 722</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00723"></a><span class="lineno"> 723</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00724"></a><span class="lineno"> 724</span>  acc4 += <a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a4;</div><div class="line"><a name="l00725"></a><span class="lineno"> 725</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00726"></a><span class="lineno"> 726</span>  }</div><div class="line"><a name="l00727"></a><span class="lineno"> 727</span> </div><div class="line"><a name="l00728"></a><span class="lineno"> 728</span>  <span class="comment">// Compute destination address</span></div><div class="line"><a name="l00729"></a><span class="lineno"> 729</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</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="comment">// Store the result</span></div><div class="line"><a name="l00732"></a><span class="lineno"> 732</span>  <a class="code" href="helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div><div class="line"><a name="l00733"></a><span class="lineno"> 733</span>  (<a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(acc0, VECTOR_INT), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div><div class="line"><a name="l00734"></a><span class="lineno"> 734</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00735"></a><span class="lineno"> 735</span>  <a class="code" href="helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div><div class="line"><a name="l00736"></a><span class="lineno"> 736</span>  (<a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(acc1, VECTOR_INT), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 1)));</div><div class="line"><a name="l00737"></a><span class="lineno"> 737</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00738"></a><span class="lineno"> 738</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00739"></a><span class="lineno"> 739</span>  <a class="code" href="helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div><div class="line"><a name="l00740"></a><span class="lineno"> 740</span>  (<a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(acc2, VECTOR_INT), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 2)));</div><div class="line"><a name="l00741"></a><span class="lineno"> 741</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00742"></a><span class="lineno"> 742</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00743"></a><span class="lineno"> 743</span>  <a class="code" href="helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div><div class="line"><a name="l00744"></a><span class="lineno"> 744</span>  (<a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(acc3, VECTOR_INT), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 3)));</div><div class="line"><a name="l00745"></a><span class="lineno"> 745</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00746"></a><span class="lineno"> 746</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00747"></a><span class="lineno"> 747</span>  <a class="code" href="helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div><div class="line"><a name="l00748"></a><span class="lineno"> 748</span>  (<a class="code" href="helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(acc4, VECTOR_INT), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 4)));</div><div class="line"><a name="l00749"></a><span class="lineno"> 749</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00750"></a><span class="lineno"> 750</span> }</div><div class="line"><a name="l00751"></a><span class="lineno"> 751</span> </div><div class="line"><a name="l00775"></a><span class="lineno"> 775</span> __kernel <span class="keywordtype">void</span> gemmlowp_mm_bifrost(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div><div class="line"><a name="l00776"></a><span class="lineno"> 776</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div><div class="line"><a name="l00777"></a><span class="lineno"> 777</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div><div class="line"><a name="l00778"></a><span class="lineno"> 778</span> {</div><div class="line"><a name="l00779"></a><span class="lineno"> 779</span>  <span class="keywordtype">int</span> idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X;</div><div class="line"><a name="l00780"></a><span class="lineno"> 780</span> </div><div class="line"><a name="l00781"></a><span class="lineno"> 781</span>  <span class="comment">// Compute starting address for matrix A and Matrix B</span></div><div class="line"><a name="l00782"></a><span class="lineno"> 782</span>  int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div><div class="line"><a name="l00783"></a><span class="lineno"> 783</span> </div><div class="line"><a name="l00784"></a><span class="lineno"> 784</span>  <span class="comment">// Update address for the matrix A</span></div><div class="line"><a name="l00785"></a><span class="lineno"> 785</span>  src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y;</div><div class="line"><a name="l00786"></a><span class="lineno"> 786</span> </div><div class="line"><a name="l00787"></a><span class="lineno"> 787</span>  <span class="comment">// Update address for the matrix B</span></div><div class="line"><a name="l00788"></a><span class="lineno"> 788</span>  src_addr.s1 += idx;</div><div class="line"><a name="l00789"></a><span class="lineno"> 789</span> </div><div class="line"><a name="l00790"></a><span class="lineno"> 790</span>  <span class="keywordtype">int</span> end_row_vec_a = src_addr.s0 + COLS_A;</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>  uint acc00 = 0;</div><div class="line"><a name="l00793"></a><span class="lineno"> 793</span>  uint acc01 = 0;</div><div class="line"><a name="l00794"></a><span class="lineno"> 794</span>  uint acc02 = 0;</div><div class="line"><a name="l00795"></a><span class="lineno"> 795</span>  uint acc03 = 0;</div><div class="line"><a name="l00796"></a><span class="lineno"> 796</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00797"></a><span class="lineno"> 797</span>  uint acc10 = 0;</div><div class="line"><a name="l00798"></a><span class="lineno"> 798</span>  uint acc11 = 0;</div><div class="line"><a name="l00799"></a><span class="lineno"> 799</span>  uint acc12 = 0;</div><div class="line"><a name="l00800"></a><span class="lineno"> 800</span>  uint acc13 = 0;</div><div class="line"><a name="l00801"></a><span class="lineno"> 801</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00802"></a><span class="lineno"> 802</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00803"></a><span class="lineno"> 803</span>  uint acc20 = 0;</div><div class="line"><a name="l00804"></a><span class="lineno"> 804</span>  uint acc21 = 0;</div><div class="line"><a name="l00805"></a><span class="lineno"> 805</span>  uint acc22 = 0;</div><div class="line"><a name="l00806"></a><span class="lineno"> 806</span>  uint acc23 = 0;</div><div class="line"><a name="l00807"></a><span class="lineno"> 807</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00808"></a><span class="lineno"> 808</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00809"></a><span class="lineno"> 809</span>  uint acc30 = 0;</div><div class="line"><a name="l00810"></a><span class="lineno"> 810</span>  uint acc31 = 0;</div><div class="line"><a name="l00811"></a><span class="lineno"> 811</span>  uint acc32 = 0;</div><div class="line"><a name="l00812"></a><span class="lineno"> 812</span>  uint acc33 = 0;</div><div class="line"><a name="l00813"></a><span class="lineno"> 813</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00814"></a><span class="lineno"> 814</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00815"></a><span class="lineno"> 815</span>  uint acc40 = 0;</div><div class="line"><a name="l00816"></a><span class="lineno"> 816</span>  uint acc41 = 0;</div><div class="line"><a name="l00817"></a><span class="lineno"> 817</span>  uint acc42 = 0;</div><div class="line"><a name="l00818"></a><span class="lineno"> 818</span>  uint acc43 = 0;</div><div class="line"><a name="l00819"></a><span class="lineno"> 819</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></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>  <span class="keywordflow">for</span>(; src_addr.s0 <= (end_row_vec_a - 4); src_addr += (int2)(4, 4 * src1_stride_y))</div><div class="line"><a name="l00822"></a><span class="lineno"> 822</span>  {</div><div class="line"><a name="l00823"></a><span class="lineno"> 823</span>  <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l00824"></a><span class="lineno"> 824</span>  uchar4 a0 = vload4(0, src0_ptr + src_addr.s0 + 0 * src0_stride_y);</div><div class="line"><a name="l00825"></a><span class="lineno"> 825</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00826"></a><span class="lineno"> 826</span>  uchar4 a1 = vload4(0, src0_ptr + src_addr.s0 + 1 * src0_stride_y);</div><div class="line"><a name="l00827"></a><span class="lineno"> 827</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00828"></a><span class="lineno"> 828</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00829"></a><span class="lineno"> 829</span>  uchar4 a2 = vload4(0, src0_ptr + src_addr.s0 + 2 * src0_stride_y);</div><div class="line"><a name="l00830"></a><span class="lineno"> 830</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00831"></a><span class="lineno"> 831</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00832"></a><span class="lineno"> 832</span>  uchar4 a3 = vload4(0, src0_ptr + src_addr.s0 + 3 * src0_stride_y);</div><div class="line"><a name="l00833"></a><span class="lineno"> 833</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00834"></a><span class="lineno"> 834</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00835"></a><span class="lineno"> 835</span>  uchar4 a4 = vload4(0, src0_ptr + src_addr.s0 + 4 * src0_stride_y);</div><div class="line"><a name="l00836"></a><span class="lineno"> 836</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00837"></a><span class="lineno"> 837</span>  <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l00838"></a><span class="lineno"> 838</span>  uchar4 b0 = vload4(0, src1_ptr + src_addr.s1 + 0 * src1_stride_y);</div><div class="line"><a name="l00839"></a><span class="lineno"> 839</span>  uchar4 b1 = vload4(0, src1_ptr + src_addr.s1 + 1 * src1_stride_y);</div><div class="line"><a name="l00840"></a><span class="lineno"> 840</span>  uchar4 b2 = vload4(0, src1_ptr + src_addr.s1 + 2 * src1_stride_y);</div><div class="line"><a name="l00841"></a><span class="lineno"> 841</span>  uchar4 b3 = vload4(0, src1_ptr + src_addr.s1 + 3 * src1_stride_y);</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="comment">// Accumulate</span></div><div class="line"><a name="l00845"></a><span class="lineno"> 845</span>  ushort tmp0 = (ushort)b0.s0 * (ushort)a0.s0;</div><div class="line"><a name="l00846"></a><span class="lineno"> 846</span>  ushort tmp1 = (ushort)b0.s1 * (ushort)a0.s0;</div><div class="line"><a name="l00847"></a><span class="lineno"> 847</span>  ushort tmp2 = (ushort)b0.s2 * (ushort)a0.s0;</div><div class="line"><a name="l00848"></a><span class="lineno"> 848</span>  ushort tmp3 = (ushort)b0.s3 * (ushort)a0.s0;</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>  ushort tmp4 = (ushort)b1.s0 * (ushort)a0.s1;</div><div class="line"><a name="l00851"></a><span class="lineno"> 851</span>  ushort tmp5 = (ushort)b1.s1 * (ushort)a0.s1;</div><div class="line"><a name="l00852"></a><span class="lineno"> 852</span>  ushort tmp6 = (ushort)b1.s2 * (ushort)a0.s1;</div><div class="line"><a name="l00853"></a><span class="lineno"> 853</span>  ushort tmp7 = (ushort)b1.s3 * (ushort)a0.s1;</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>  ushort tmp8 = (ushort)b2.s0 * (ushort)a0.s2;</div><div class="line"><a name="l00856"></a><span class="lineno"> 856</span>  ushort tmp9 = (ushort)b2.s1 * (ushort)a0.s2;</div><div class="line"><a name="l00857"></a><span class="lineno"> 857</span>  ushort tmpA = (ushort)b2.s2 * (ushort)a0.s2;</div><div class="line"><a name="l00858"></a><span class="lineno"> 858</span>  ushort tmpB = (ushort)b2.s3 * (ushort)a0.s2;</div><div class="line"><a name="l00859"></a><span class="lineno"> 859</span> </div><div class="line"><a name="l00860"></a><span class="lineno"> 860</span>  ushort tmpC = (ushort)b3.s0 * (ushort)a0.s3;</div><div class="line"><a name="l00861"></a><span class="lineno"> 861</span>  ushort tmpD = (ushort)b3.s1 * (ushort)a0.s3;</div><div class="line"><a name="l00862"></a><span class="lineno"> 862</span>  ushort tmpE = (ushort)b3.s2 * (ushort)a0.s3;</div><div class="line"><a name="l00863"></a><span class="lineno"> 863</span>  ushort tmpF = (ushort)b3.s3 * (ushort)a0.s3;</div><div class="line"><a name="l00864"></a><span class="lineno"> 864</span> </div><div class="line"><a name="l00865"></a><span class="lineno"> 865</span>  acc00 += ((uint)tmp0 + (uint)tmp4 + (uint)tmp8 + (uint)tmpC);</div><div class="line"><a name="l00866"></a><span class="lineno"> 866</span>  acc01 += ((uint)tmp1 + (uint)tmp5 + (uint)tmp9 + (uint)tmpD);</div><div class="line"><a name="l00867"></a><span class="lineno"> 867</span>  acc02 += ((uint)tmp2 + (uint)tmp6 + (uint)tmpA + (uint)tmpE);</div><div class="line"><a name="l00868"></a><span class="lineno"> 868</span>  acc03 += ((uint)tmp3 + (uint)tmp7 + (uint)tmpB + (uint)tmpF);</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> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00871"></a><span class="lineno"> 871</span>  {</div><div class="line"><a name="l00872"></a><span class="lineno"> 872</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l00873"></a><span class="lineno"> 873</span>  ushort tmp0 = (ushort)b0.s0 * (ushort)a1.s0;</div><div class="line"><a name="l00874"></a><span class="lineno"> 874</span>  ushort tmp1 = (ushort)b0.s1 * (ushort)a1.s0;</div><div class="line"><a name="l00875"></a><span class="lineno"> 875</span>  ushort tmp2 = (ushort)b0.s2 * (ushort)a1.s0;</div><div class="line"><a name="l00876"></a><span class="lineno"> 876</span>  ushort tmp3 = (ushort)b0.s3 * (ushort)a1.s0;</div><div class="line"><a name="l00877"></a><span class="lineno"> 877</span> </div><div class="line"><a name="l00878"></a><span class="lineno"> 878</span>  ushort tmp4 = (ushort)b1.s0 * (ushort)a1.s1;</div><div class="line"><a name="l00879"></a><span class="lineno"> 879</span>  ushort tmp5 = (ushort)b1.s1 * (ushort)a1.s1;</div><div class="line"><a name="l00880"></a><span class="lineno"> 880</span>  ushort tmp6 = (ushort)b1.s2 * (ushort)a1.s1;</div><div class="line"><a name="l00881"></a><span class="lineno"> 881</span>  ushort tmp7 = (ushort)b1.s3 * (ushort)a1.s1;</div><div class="line"><a name="l00882"></a><span class="lineno"> 882</span> </div><div class="line"><a name="l00883"></a><span class="lineno"> 883</span>  ushort tmp8 = (ushort)b2.s0 * (ushort)a1.s2;</div><div class="line"><a name="l00884"></a><span class="lineno"> 884</span>  ushort tmp9 = (ushort)b2.s1 * (ushort)a1.s2;</div><div class="line"><a name="l00885"></a><span class="lineno"> 885</span>  ushort tmpA = (ushort)b2.s2 * (ushort)a1.s2;</div><div class="line"><a name="l00886"></a><span class="lineno"> 886</span>  ushort tmpB = (ushort)b2.s3 * (ushort)a1.s2;</div><div class="line"><a name="l00887"></a><span class="lineno"> 887</span> </div><div class="line"><a name="l00888"></a><span class="lineno"> 888</span>  ushort tmpC = (ushort)b3.s0 * (ushort)a1.s3;</div><div class="line"><a name="l00889"></a><span class="lineno"> 889</span>  ushort tmpD = (ushort)b3.s1 * (ushort)a1.s3;</div><div class="line"><a name="l00890"></a><span class="lineno"> 890</span>  ushort tmpE = (ushort)b3.s2 * (ushort)a1.s3;</div><div class="line"><a name="l00891"></a><span class="lineno"> 891</span>  ushort tmpF = (ushort)b3.s3 * (ushort)a1.s3;</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>  acc10 += ((uint)tmp0 + (uint)tmp4 + (uint)tmp8 + (uint)tmpC);</div><div class="line"><a name="l00894"></a><span class="lineno"> 894</span>  acc11 += ((uint)tmp1 + (uint)tmp5 + (uint)tmp9 + (uint)tmpD);</div><div class="line"><a name="l00895"></a><span class="lineno"> 895</span>  acc12 += ((uint)tmp2 + (uint)tmp6 + (uint)tmpA + (uint)tmpE);</div><div class="line"><a name="l00896"></a><span class="lineno"> 896</span>  acc13 += ((uint)tmp3 + (uint)tmp7 + (uint)tmpB + (uint)tmpF);</div><div class="line"><a name="l00897"></a><span class="lineno"> 897</span>  }</div><div class="line"><a name="l00898"></a><span class="lineno"> 898</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00899"></a><span class="lineno"> 899</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00900"></a><span class="lineno"> 900</span>  {</div><div class="line"><a name="l00901"></a><span class="lineno"> 901</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l00902"></a><span class="lineno"> 902</span>  ushort tmp0 = (ushort)b0.s0 * (ushort)a2.s0;</div><div class="line"><a name="l00903"></a><span class="lineno"> 903</span>  ushort tmp1 = (ushort)b0.s1 * (ushort)a2.s0;</div><div class="line"><a name="l00904"></a><span class="lineno"> 904</span>  ushort tmp2 = (ushort)b0.s2 * (ushort)a2.s0;</div><div class="line"><a name="l00905"></a><span class="lineno"> 905</span>  ushort tmp3 = (ushort)b0.s3 * (ushort)a2.s0;</div><div class="line"><a name="l00906"></a><span class="lineno"> 906</span> </div><div class="line"><a name="l00907"></a><span class="lineno"> 907</span>  ushort tmp4 = (ushort)b1.s0 * (ushort)a2.s1;</div><div class="line"><a name="l00908"></a><span class="lineno"> 908</span>  ushort tmp5 = (ushort)b1.s1 * (ushort)a2.s1;</div><div class="line"><a name="l00909"></a><span class="lineno"> 909</span>  ushort tmp6 = (ushort)b1.s2 * (ushort)a2.s1;</div><div class="line"><a name="l00910"></a><span class="lineno"> 910</span>  ushort tmp7 = (ushort)b1.s3 * (ushort)a2.s1;</div><div class="line"><a name="l00911"></a><span class="lineno"> 911</span> </div><div class="line"><a name="l00912"></a><span class="lineno"> 912</span>  ushort tmp8 = (ushort)b2.s0 * (ushort)a2.s2;</div><div class="line"><a name="l00913"></a><span class="lineno"> 913</span>  ushort tmp9 = (ushort)b2.s1 * (ushort)a2.s2;</div><div class="line"><a name="l00914"></a><span class="lineno"> 914</span>  ushort tmpA = (ushort)b2.s2 * (ushort)a2.s2;</div><div class="line"><a name="l00915"></a><span class="lineno"> 915</span>  ushort tmpB = (ushort)b2.s3 * (ushort)a2.s2;</div><div class="line"><a name="l00916"></a><span class="lineno"> 916</span> </div><div class="line"><a name="l00917"></a><span class="lineno"> 917</span>  ushort tmpC = (ushort)b3.s0 * (ushort)a2.s3;</div><div class="line"><a name="l00918"></a><span class="lineno"> 918</span>  ushort tmpD = (ushort)b3.s1 * (ushort)a2.s3;</div><div class="line"><a name="l00919"></a><span class="lineno"> 919</span>  ushort tmpE = (ushort)b3.s2 * (ushort)a2.s3;</div><div class="line"><a name="l00920"></a><span class="lineno"> 920</span>  ushort tmpF = (ushort)b3.s3 * (ushort)a2.s3;</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>  acc20 += ((uint)tmp0 + (uint)tmp4 + (uint)tmp8 + (uint)tmpC);</div><div class="line"><a name="l00923"></a><span class="lineno"> 923</span>  acc21 += ((uint)tmp1 + (uint)tmp5 + (uint)tmp9 + (uint)tmpD);</div><div class="line"><a name="l00924"></a><span class="lineno"> 924</span>  acc22 += ((uint)tmp2 + (uint)tmp6 + (uint)tmpA + (uint)tmpE);</div><div class="line"><a name="l00925"></a><span class="lineno"> 925</span>  acc23 += ((uint)tmp3 + (uint)tmp7 + (uint)tmpB + (uint)tmpF);</div><div class="line"><a name="l00926"></a><span class="lineno"> 926</span>  }</div><div class="line"><a name="l00927"></a><span class="lineno"> 927</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00928"></a><span class="lineno"> 928</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00929"></a><span class="lineno"> 929</span>  {</div><div class="line"><a name="l00930"></a><span class="lineno"> 930</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l00931"></a><span class="lineno"> 931</span>  ushort tmp0 = (ushort)b0.s0 * (ushort)a3.s0;</div><div class="line"><a name="l00932"></a><span class="lineno"> 932</span>  ushort tmp1 = (ushort)b0.s1 * (ushort)a3.s0;</div><div class="line"><a name="l00933"></a><span class="lineno"> 933</span>  ushort tmp2 = (ushort)b0.s2 * (ushort)a3.s0;</div><div class="line"><a name="l00934"></a><span class="lineno"> 934</span>  ushort tmp3 = (ushort)b0.s3 * (ushort)a3.s0;</div><div class="line"><a name="l00935"></a><span class="lineno"> 935</span> </div><div class="line"><a name="l00936"></a><span class="lineno"> 936</span>  ushort tmp4 = (ushort)b1.s0 * (ushort)a3.s1;</div><div class="line"><a name="l00937"></a><span class="lineno"> 937</span>  ushort tmp5 = (ushort)b1.s1 * (ushort)a3.s1;</div><div class="line"><a name="l00938"></a><span class="lineno"> 938</span>  ushort tmp6 = (ushort)b1.s2 * (ushort)a3.s1;</div><div class="line"><a name="l00939"></a><span class="lineno"> 939</span>  ushort tmp7 = (ushort)b1.s3 * (ushort)a3.s1;</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>  ushort tmp8 = (ushort)b2.s0 * (ushort)a3.s2;</div><div class="line"><a name="l00942"></a><span class="lineno"> 942</span>  ushort tmp9 = (ushort)b2.s1 * (ushort)a3.s2;</div><div class="line"><a name="l00943"></a><span class="lineno"> 943</span>  ushort tmpA = (ushort)b2.s2 * (ushort)a3.s2;</div><div class="line"><a name="l00944"></a><span class="lineno"> 944</span>  ushort tmpB = (ushort)b2.s3 * (ushort)a3.s2;</div><div class="line"><a name="l00945"></a><span class="lineno"> 945</span> </div><div class="line"><a name="l00946"></a><span class="lineno"> 946</span>  ushort tmpC = (ushort)b3.s0 * (ushort)a3.s3;</div><div class="line"><a name="l00947"></a><span class="lineno"> 947</span>  ushort tmpD = (ushort)b3.s1 * (ushort)a3.s3;</div><div class="line"><a name="l00948"></a><span class="lineno"> 948</span>  ushort tmpE = (ushort)b3.s2 * (ushort)a3.s3;</div><div class="line"><a name="l00949"></a><span class="lineno"> 949</span>  ushort tmpF = (ushort)b3.s3 * (ushort)a3.s3;</div><div class="line"><a name="l00950"></a><span class="lineno"> 950</span> </div><div class="line"><a name="l00951"></a><span class="lineno"> 951</span>  acc30 += ((uint)tmp0 + (uint)tmp4 + (uint)tmp8 + (uint)tmpC);</div><div class="line"><a name="l00952"></a><span class="lineno"> 952</span>  acc31 += ((uint)tmp1 + (uint)tmp5 + (uint)tmp9 + (uint)tmpD);</div><div class="line"><a name="l00953"></a><span class="lineno"> 953</span>  acc32 += ((uint)tmp2 + (uint)tmp6 + (uint)tmpA + (uint)tmpE);</div><div class="line"><a name="l00954"></a><span class="lineno"> 954</span>  acc33 += ((uint)tmp3 + (uint)tmp7 + (uint)tmpB + (uint)tmpF);</div><div class="line"><a name="l00955"></a><span class="lineno"> 955</span>  }</div><div class="line"><a name="l00956"></a><span class="lineno"> 956</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00957"></a><span class="lineno"> 957</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00958"></a><span class="lineno"> 958</span>  {</div><div class="line"><a name="l00959"></a><span class="lineno"> 959</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l00960"></a><span class="lineno"> 960</span>  ushort tmp0 = (ushort)b0.s0 * (ushort)a4.s0;</div><div class="line"><a name="l00961"></a><span class="lineno"> 961</span>  ushort tmp1 = (ushort)b0.s1 * (ushort)a4.s0;</div><div class="line"><a name="l00962"></a><span class="lineno"> 962</span>  ushort tmp2 = (ushort)b0.s2 * (ushort)a4.s0;</div><div class="line"><a name="l00963"></a><span class="lineno"> 963</span>  ushort tmp3 = (ushort)b0.s3 * (ushort)a4.s0;</div><div class="line"><a name="l00964"></a><span class="lineno"> 964</span> </div><div class="line"><a name="l00965"></a><span class="lineno"> 965</span>  ushort tmp4 = (ushort)b1.s0 * (ushort)a4.s1;</div><div class="line"><a name="l00966"></a><span class="lineno"> 966</span>  ushort tmp5 = (ushort)b1.s1 * (ushort)a4.s1;</div><div class="line"><a name="l00967"></a><span class="lineno"> 967</span>  ushort tmp6 = (ushort)b1.s2 * (ushort)a4.s1;</div><div class="line"><a name="l00968"></a><span class="lineno"> 968</span>  ushort tmp7 = (ushort)b1.s3 * (ushort)a4.s1;</div><div class="line"><a name="l00969"></a><span class="lineno"> 969</span> </div><div class="line"><a name="l00970"></a><span class="lineno"> 970</span>  ushort tmp8 = (ushort)b2.s0 * (ushort)a4.s2;</div><div class="line"><a name="l00971"></a><span class="lineno"> 971</span>  ushort tmp9 = (ushort)b2.s1 * (ushort)a4.s2;</div><div class="line"><a name="l00972"></a><span class="lineno"> 972</span>  ushort tmpA = (ushort)b2.s2 * (ushort)a4.s2;</div><div class="line"><a name="l00973"></a><span class="lineno"> 973</span>  ushort tmpB = (ushort)b2.s3 * (ushort)a4.s2;</div><div class="line"><a name="l00974"></a><span class="lineno"> 974</span> </div><div class="line"><a name="l00975"></a><span class="lineno"> 975</span>  ushort tmpC = (ushort)b3.s0 * (ushort)a4.s3;</div><div class="line"><a name="l00976"></a><span class="lineno"> 976</span>  ushort tmpD = (ushort)b3.s1 * (ushort)a4.s3;</div><div class="line"><a name="l00977"></a><span class="lineno"> 977</span>  ushort tmpE = (ushort)b3.s2 * (ushort)a4.s3;</div><div class="line"><a name="l00978"></a><span class="lineno"> 978</span>  ushort tmpF = (ushort)b3.s3 * (ushort)a4.s3;</div><div class="line"><a name="l00979"></a><span class="lineno"> 979</span> </div><div class="line"><a name="l00980"></a><span class="lineno"> 980</span>  acc40 += ((uint)tmp0 + (uint)tmp4 + (uint)tmp8 + (uint)tmpC);</div><div class="line"><a name="l00981"></a><span class="lineno"> 981</span>  acc41 += ((uint)tmp1 + (uint)tmp5 + (uint)tmp9 + (uint)tmpD);</div><div class="line"><a name="l00982"></a><span class="lineno"> 982</span>  acc42 += ((uint)tmp2 + (uint)tmp6 + (uint)tmpA + (uint)tmpE);</div><div class="line"><a name="l00983"></a><span class="lineno"> 983</span>  acc43 += ((uint)tmp3 + (uint)tmp7 + (uint)tmpB + (uint)tmpF);</div><div class="line"><a name="l00984"></a><span class="lineno"> 984</span>  }</div><div class="line"><a name="l00985"></a><span class="lineno"> 985</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l00986"></a><span class="lineno"> 986</span>  }</div><div class="line"><a name="l00987"></a><span class="lineno"> 987</span> </div><div class="line"><a name="l00988"></a><span class="lineno"> 988</span>  <span class="keywordflow">for</span>(; src_addr.s0 < end_row_vec_a; src_addr += (int2)(1, src1_stride_y))</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>  <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l00991"></a><span class="lineno"> 991</span>  uchar a0 = *(src0_ptr + src_addr.s0 + 0 * src0_stride_y);</div><div class="line"><a name="l00992"></a><span class="lineno"> 992</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00993"></a><span class="lineno"> 993</span>  uchar a1 = *(src0_ptr + src_addr.s0 + 1 * src0_stride_y);</div><div class="line"><a name="l00994"></a><span class="lineno"> 994</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l00995"></a><span class="lineno"> 995</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00996"></a><span class="lineno"> 996</span>  uchar a2 = *(src0_ptr + src_addr.s0 + 2 * src0_stride_y);</div><div class="line"><a name="l00997"></a><span class="lineno"> 997</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l00998"></a><span class="lineno"> 998</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l00999"></a><span class="lineno"> 999</span>  uchar a3 = *(src0_ptr + src_addr.s0 + 3 * src0_stride_y);</div><div class="line"><a name="l01000"></a><span class="lineno"> 1000</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01001"></a><span class="lineno"> 1001</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01002"></a><span class="lineno"> 1002</span>  uchar a4 = *(src0_ptr + src_addr.s0 + 4 * src0_stride_y);</div><div class="line"><a name="l01003"></a><span class="lineno"> 1003</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01004"></a><span class="lineno"> 1004</span>  <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l01005"></a><span class="lineno"> 1005</span>  uchar4 b0 = vload4(0, src1_ptr + src_addr.s1);</div><div class="line"><a name="l01006"></a><span class="lineno"> 1006</span> </div><div class="line"><a name="l01007"></a><span class="lineno"> 1007</span>  <span class="comment">// Accumulate</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">// Accumulate</span></div><div class="line"><a name="l01010"></a><span class="lineno"> 1010</span>  ushort tmp0 = (ushort)b0.s0 * (ushort)a0;</div><div class="line"><a name="l01011"></a><span class="lineno"> 1011</span>  ushort tmp1 = (ushort)b0.s1 * (ushort)a0;</div><div class="line"><a name="l01012"></a><span class="lineno"> 1012</span>  ushort tmp2 = (ushort)b0.s2 * (ushort)a0;</div><div class="line"><a name="l01013"></a><span class="lineno"> 1013</span>  ushort tmp3 = (ushort)b0.s3 * (ushort)a0;</div><div class="line"><a name="l01014"></a><span class="lineno"> 1014</span> </div><div class="line"><a name="l01015"></a><span class="lineno"> 1015</span>  acc00 += ((uint)tmp0);</div><div class="line"><a name="l01016"></a><span class="lineno"> 1016</span>  acc01 += ((uint)tmp1);</div><div class="line"><a name="l01017"></a><span class="lineno"> 1017</span>  acc02 += ((uint)tmp2);</div><div class="line"><a name="l01018"></a><span class="lineno"> 1018</span>  acc03 += ((uint)tmp3);</div><div class="line"><a name="l01019"></a><span class="lineno"> 1019</span>  }</div><div class="line"><a name="l01020"></a><span class="lineno"> 1020</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01021"></a><span class="lineno"> 1021</span>  {</div><div class="line"><a name="l01022"></a><span class="lineno"> 1022</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l01023"></a><span class="lineno"> 1023</span>  ushort tmp0 = (ushort)b0.s0 * (ushort)a1;</div><div class="line"><a name="l01024"></a><span class="lineno"> 1024</span>  ushort tmp1 = (ushort)b0.s1 * (ushort)a1;</div><div class="line"><a name="l01025"></a><span class="lineno"> 1025</span>  ushort tmp2 = (ushort)b0.s2 * (ushort)a1;</div><div class="line"><a name="l01026"></a><span class="lineno"> 1026</span>  ushort tmp3 = (ushort)b0.s3 * (ushort)a1;</div><div class="line"><a name="l01027"></a><span class="lineno"> 1027</span> </div><div class="line"><a name="l01028"></a><span class="lineno"> 1028</span>  acc10 += ((uint)tmp0);</div><div class="line"><a name="l01029"></a><span class="lineno"> 1029</span>  acc11 += ((uint)tmp1);</div><div class="line"><a name="l01030"></a><span class="lineno"> 1030</span>  acc12 += ((uint)tmp2);</div><div class="line"><a name="l01031"></a><span class="lineno"> 1031</span>  acc13 += ((uint)tmp3);</div><div class="line"><a name="l01032"></a><span class="lineno"> 1032</span>  }</div><div class="line"><a name="l01033"></a><span class="lineno"> 1033</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01034"></a><span class="lineno"> 1034</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01035"></a><span class="lineno"> 1035</span>  {</div><div class="line"><a name="l01036"></a><span class="lineno"> 1036</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l01037"></a><span class="lineno"> 1037</span>  ushort tmp0 = (ushort)b0.s0 * (ushort)a2;</div><div class="line"><a name="l01038"></a><span class="lineno"> 1038</span>  ushort tmp1 = (ushort)b0.s1 * (ushort)a2;</div><div class="line"><a name="l01039"></a><span class="lineno"> 1039</span>  ushort tmp2 = (ushort)b0.s2 * (ushort)a2;</div><div class="line"><a name="l01040"></a><span class="lineno"> 1040</span>  ushort tmp3 = (ushort)b0.s3 * (ushort)a2;</div><div class="line"><a name="l01041"></a><span class="lineno"> 1041</span> </div><div class="line"><a name="l01042"></a><span class="lineno"> 1042</span>  acc20 += ((uint)tmp0);</div><div class="line"><a name="l01043"></a><span class="lineno"> 1043</span>  acc21 += ((uint)tmp1);</div><div class="line"><a name="l01044"></a><span class="lineno"> 1044</span>  acc22 += ((uint)tmp2);</div><div class="line"><a name="l01045"></a><span class="lineno"> 1045</span>  acc23 += ((uint)tmp3);</div><div class="line"><a name="l01046"></a><span class="lineno"> 1046</span>  }</div><div class="line"><a name="l01047"></a><span class="lineno"> 1047</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01048"></a><span class="lineno"> 1048</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01049"></a><span class="lineno"> 1049</span>  {</div><div class="line"><a name="l01050"></a><span class="lineno"> 1050</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l01051"></a><span class="lineno"> 1051</span>  ushort tmp0 = (ushort)b0.s0 * (ushort)a3;</div><div class="line"><a name="l01052"></a><span class="lineno"> 1052</span>  ushort tmp1 = (ushort)b0.s1 * (ushort)a3;</div><div class="line"><a name="l01053"></a><span class="lineno"> 1053</span>  ushort tmp2 = (ushort)b0.s2 * (ushort)a3;</div><div class="line"><a name="l01054"></a><span class="lineno"> 1054</span>  ushort tmp3 = (ushort)b0.s3 * (ushort)a3;</div><div class="line"><a name="l01055"></a><span class="lineno"> 1055</span> </div><div class="line"><a name="l01056"></a><span class="lineno"> 1056</span>  acc30 += ((uint)tmp0);</div><div class="line"><a name="l01057"></a><span class="lineno"> 1057</span>  acc31 += ((uint)tmp1);</div><div class="line"><a name="l01058"></a><span class="lineno"> 1058</span>  acc32 += ((uint)tmp2);</div><div class="line"><a name="l01059"></a><span class="lineno"> 1059</span>  acc33 += ((uint)tmp3);</div><div class="line"><a name="l01060"></a><span class="lineno"> 1060</span>  }</div><div class="line"><a name="l01061"></a><span class="lineno"> 1061</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01062"></a><span class="lineno"> 1062</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01063"></a><span class="lineno"> 1063</span>  {</div><div class="line"><a name="l01064"></a><span class="lineno"> 1064</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l01065"></a><span class="lineno"> 1065</span>  ushort tmp0 = (ushort)b0.s0 * (ushort)a4;</div><div class="line"><a name="l01066"></a><span class="lineno"> 1066</span>  ushort tmp1 = (ushort)b0.s1 * (ushort)a4;</div><div class="line"><a name="l01067"></a><span class="lineno"> 1067</span>  ushort tmp2 = (ushort)b0.s2 * (ushort)a4;</div><div class="line"><a name="l01068"></a><span class="lineno"> 1068</span>  ushort tmp3 = (ushort)b0.s3 * (ushort)a4;</div><div class="line"><a name="l01069"></a><span class="lineno"> 1069</span> </div><div class="line"><a name="l01070"></a><span class="lineno"> 1070</span>  acc40 += ((uint)tmp0);</div><div class="line"><a name="l01071"></a><span class="lineno"> 1071</span>  acc41 += ((uint)tmp1);</div><div class="line"><a name="l01072"></a><span class="lineno"> 1072</span>  acc42 += ((uint)tmp2);</div><div class="line"><a name="l01073"></a><span class="lineno"> 1073</span>  acc43 += ((uint)tmp3);</div><div class="line"><a name="l01074"></a><span class="lineno"> 1074</span>  }</div><div class="line"><a name="l01075"></a><span class="lineno"> 1075</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01076"></a><span class="lineno"> 1076</span>  }</div><div class="line"><a name="l01077"></a><span class="lineno"> 1077</span> </div><div class="line"><a name="l01078"></a><span class="lineno"> 1078</span>  <span class="comment">// Compute destination address</span></div><div class="line"><a name="l01079"></a><span class="lineno"> 1079</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l01080"></a><span class="lineno"> 1080</span> </div><div class="line"><a name="l01081"></a><span class="lineno"> 1081</span>  <span class="comment">// Store the result</span></div><div class="line"><a name="l01082"></a><span class="lineno"> 1082</span>  vstore4((int4)(acc00, acc01, acc02, acc03), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div><div class="line"><a name="l01083"></a><span class="lineno"> 1083</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01084"></a><span class="lineno"> 1084</span>  vstore4((int4)(acc10, acc11, acc12, acc13), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 1)));</div><div class="line"><a name="l01085"></a><span class="lineno"> 1085</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01086"></a><span class="lineno"> 1086</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01087"></a><span class="lineno"> 1087</span>  vstore4((int4)(acc20, acc21, acc22, acc23), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 2)));</div><div class="line"><a name="l01088"></a><span class="lineno"> 1088</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01089"></a><span class="lineno"> 1089</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01090"></a><span class="lineno"> 1090</span>  vstore4((int4)(acc30, acc31, acc32, acc33), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 3)));</div><div class="line"><a name="l01091"></a><span class="lineno"> 1091</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01092"></a><span class="lineno"> 1092</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01093"></a><span class="lineno"> 1093</span>  vstore4((int4)(acc40, acc41, acc42, acc43), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 4)));</div><div class="line"><a name="l01094"></a><span class="lineno"> 1094</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01095"></a><span class="lineno"> 1095</span> }</div><div class="line"><a name="l01096"></a><span class="lineno"> 1096</span> </div><div class="line"><a name="l01097"></a><span class="lineno"> 1097</span> <span class="preprocessor">#if ARM_COMPUTE_OPENCL_DOT8_ENABLED</span></div><div class="line"><a name="l01098"></a><span class="lineno"> 1098</span> </div><div class="line"><a name="l01121"></a><span class="lineno"> 1121</span> __kernel <span class="keywordtype">void</span> gemmlowp_mm_bifrost_dot8(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div><div class="line"><a name="l01122"></a><span class="lineno"> 1122</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div><div class="line"><a name="l01123"></a><span class="lineno"> 1123</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div><div class="line"><a name="l01124"></a><span class="lineno"> 1124</span> {</div><div class="line"><a name="l01125"></a><span class="lineno"> 1125</span>  <span class="keywordtype">int</span> idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X;</div><div class="line"><a name="l01126"></a><span class="lineno"> 1126</span> </div><div class="line"><a name="l01127"></a><span class="lineno"> 1127</span>  <span class="comment">// Compute starting address for matrix A and Matrix B</span></div><div class="line"><a name="l01128"></a><span class="lineno"> 1128</span>  int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div><div class="line"><a name="l01129"></a><span class="lineno"> 1129</span> </div><div class="line"><a name="l01130"></a><span class="lineno"> 1130</span>  <span class="comment">// Update address for the matrix A</span></div><div class="line"><a name="l01131"></a><span class="lineno"> 1131</span>  src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y;</div><div class="line"><a name="l01132"></a><span class="lineno"> 1132</span> </div><div class="line"><a name="l01133"></a><span class="lineno"> 1133</span>  <span class="comment">// Update address for the matrix B</span></div><div class="line"><a name="l01134"></a><span class="lineno"> 1134</span>  src_addr.s1 += idx;</div><div class="line"><a name="l01135"></a><span class="lineno"> 1135</span> </div><div class="line"><a name="l01136"></a><span class="lineno"> 1136</span>  <span class="keywordtype">int</span> end_row_vec_a = src_addr.s0 + COLS_A;</div><div class="line"><a name="l01137"></a><span class="lineno"> 1137</span> </div><div class="line"><a name="l01138"></a><span class="lineno"> 1138</span>  uint acc00 = 0;</div><div class="line"><a name="l01139"></a><span class="lineno"> 1139</span>  uint acc01 = 0;</div><div class="line"><a name="l01140"></a><span class="lineno"> 1140</span>  uint acc02 = 0;</div><div class="line"><a name="l01141"></a><span class="lineno"> 1141</span>  uint acc03 = 0;</div><div class="line"><a name="l01142"></a><span class="lineno"> 1142</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01143"></a><span class="lineno"> 1143</span>  uint acc10 = 0;</div><div class="line"><a name="l01144"></a><span class="lineno"> 1144</span>  uint acc11 = 0;</div><div class="line"><a name="l01145"></a><span class="lineno"> 1145</span>  uint acc12 = 0;</div><div class="line"><a name="l01146"></a><span class="lineno"> 1146</span>  uint acc13 = 0;</div><div class="line"><a name="l01147"></a><span class="lineno"> 1147</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01148"></a><span class="lineno"> 1148</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01149"></a><span class="lineno"> 1149</span>  uint acc20 = 0;</div><div class="line"><a name="l01150"></a><span class="lineno"> 1150</span>  uint acc21 = 0;</div><div class="line"><a name="l01151"></a><span class="lineno"> 1151</span>  uint acc22 = 0;</div><div class="line"><a name="l01152"></a><span class="lineno"> 1152</span>  uint acc23 = 0;</div><div class="line"><a name="l01153"></a><span class="lineno"> 1153</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01154"></a><span class="lineno"> 1154</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01155"></a><span class="lineno"> 1155</span>  uint acc30 = 0;</div><div class="line"><a name="l01156"></a><span class="lineno"> 1156</span>  uint acc31 = 0;</div><div class="line"><a name="l01157"></a><span class="lineno"> 1157</span>  uint acc32 = 0;</div><div class="line"><a name="l01158"></a><span class="lineno"> 1158</span>  uint acc33 = 0;</div><div class="line"><a name="l01159"></a><span class="lineno"> 1159</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01160"></a><span class="lineno"> 1160</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01161"></a><span class="lineno"> 1161</span>  uint acc40 = 0;</div><div class="line"><a name="l01162"></a><span class="lineno"> 1162</span>  uint acc41 = 0;</div><div class="line"><a name="l01163"></a><span class="lineno"> 1163</span>  uint acc42 = 0;</div><div class="line"><a name="l01164"></a><span class="lineno"> 1164</span>  uint acc43 = 0;</div><div class="line"><a name="l01165"></a><span class="lineno"> 1165</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01166"></a><span class="lineno"> 1166</span> </div><div class="line"><a name="l01167"></a><span class="lineno"> 1167</span>  <span class="keywordflow">for</span>(; src_addr.s0 <= (end_row_vec_a - 4); src_addr += (int2)(4, 4 * src1_stride_y))</div><div class="line"><a name="l01168"></a><span class="lineno"> 1168</span>  {</div><div class="line"><a name="l01169"></a><span class="lineno"> 1169</span>  <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l01170"></a><span class="lineno"> 1170</span>  uchar4 a0 = vload4(0, src0_ptr + src_addr.s0 + 0 * src0_stride_y);</div><div class="line"><a name="l01171"></a><span class="lineno"> 1171</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01172"></a><span class="lineno"> 1172</span>  uchar4 a1 = vload4(0, src0_ptr + src_addr.s0 + 1 * src0_stride_y);</div><div class="line"><a name="l01173"></a><span class="lineno"> 1173</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01174"></a><span class="lineno"> 1174</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01175"></a><span class="lineno"> 1175</span>  uchar4 a2 = vload4(0, src0_ptr + src_addr.s0 + 2 * src0_stride_y);</div><div class="line"><a name="l01176"></a><span class="lineno"> 1176</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01177"></a><span class="lineno"> 1177</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01178"></a><span class="lineno"> 1178</span>  uchar4 a3 = vload4(0, src0_ptr + src_addr.s0 + 3 * src0_stride_y);</div><div class="line"><a name="l01179"></a><span class="lineno"> 1179</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01180"></a><span class="lineno"> 1180</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01181"></a><span class="lineno"> 1181</span>  uchar4 a4 = vload4(0, src0_ptr + src_addr.s0 + 4 * src0_stride_y);</div><div class="line"><a name="l01182"></a><span class="lineno"> 1182</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01183"></a><span class="lineno"> 1183</span>  <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l01184"></a><span class="lineno"> 1184</span>  uchar4 b0 = vload4(0, src1_ptr + src_addr.s1 + 0 * src1_stride_y);</div><div class="line"><a name="l01185"></a><span class="lineno"> 1185</span>  uchar4 b1 = vload4(0, src1_ptr + src_addr.s1 + 1 * src1_stride_y);</div><div class="line"><a name="l01186"></a><span class="lineno"> 1186</span>  uchar4 b2 = vload4(0, src1_ptr + src_addr.s1 + 2 * src1_stride_y);</div><div class="line"><a name="l01187"></a><span class="lineno"> 1187</span>  uchar4 b3 = vload4(0, src1_ptr + src_addr.s1 + 3 * src1_stride_y);</div><div class="line"><a name="l01188"></a><span class="lineno"> 1188</span> </div><div class="line"><a name="l01189"></a><span class="lineno"> 1189</span>  {</div><div class="line"><a name="l01190"></a><span class="lineno"> 1190</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l01191"></a><span class="lineno"> 1191</span>  ARM_DOT(b0.s0, b1.s0, b2.s0, b3.s0, a0.s0, a0.s1, a0.s2, a0.s3, acc00);</div><div class="line"><a name="l01192"></a><span class="lineno"> 1192</span>  ARM_DOT(b0.s1, b1.s1, b2.s1, b3.s1, a0.s0, a0.s1, a0.s2, a0.s3, acc01);</div><div class="line"><a name="l01193"></a><span class="lineno"> 1193</span>  ARM_DOT(b0.s2, b1.s2, b2.s2, b3.s2, a0.s0, a0.s1, a0.s2, a0.s3, acc02);</div><div class="line"><a name="l01194"></a><span class="lineno"> 1194</span>  ARM_DOT(b0.s3, b1.s3, b2.s3, b3.s3, a0.s0, a0.s1, a0.s2, a0.s3, acc03);</div><div class="line"><a name="l01195"></a><span class="lineno"> 1195</span>  }</div><div class="line"><a name="l01196"></a><span class="lineno"> 1196</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01197"></a><span class="lineno"> 1197</span>  {</div><div class="line"><a name="l01198"></a><span class="lineno"> 1198</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l01199"></a><span class="lineno"> 1199</span>  ARM_DOT(b0.s0, b1.s0, b2.s0, b3.s0, a1.s0, a1.s1, a1.s2, a1.s3, acc10);</div><div class="line"><a name="l01200"></a><span class="lineno"> 1200</span>  ARM_DOT(b0.s1, b1.s1, b2.s1, b3.s1, a1.s0, a1.s1, a1.s2, a1.s3, acc11);</div><div class="line"><a name="l01201"></a><span class="lineno"> 1201</span>  ARM_DOT(b0.s2, b1.s2, b2.s2, b3.s2, a1.s0, a1.s1, a1.s2, a1.s3, acc12);</div><div class="line"><a name="l01202"></a><span class="lineno"> 1202</span>  ARM_DOT(b0.s3, b1.s3, b2.s3, b3.s3, a1.s0, a1.s1, a1.s2, a1.s3, acc13);</div><div class="line"><a name="l01203"></a><span class="lineno"> 1203</span>  }</div><div class="line"><a name="l01204"></a><span class="lineno"> 1204</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01205"></a><span class="lineno"> 1205</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01206"></a><span class="lineno"> 1206</span>  {</div><div class="line"><a name="l01207"></a><span class="lineno"> 1207</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l01208"></a><span class="lineno"> 1208</span>  ARM_DOT(b0.s0, b1.s0, b2.s0, b3.s0, a2.s0, a2.s1, a2.s2, a2.s3, acc20);</div><div class="line"><a name="l01209"></a><span class="lineno"> 1209</span>  ARM_DOT(b0.s1, b1.s1, b2.s1, b3.s1, a2.s0, a2.s1, a2.s2, a2.s3, acc21);</div><div class="line"><a name="l01210"></a><span class="lineno"> 1210</span>  ARM_DOT(b0.s2, b1.s2, b2.s2, b3.s2, a2.s0, a2.s1, a2.s2, a2.s3, acc22);</div><div class="line"><a name="l01211"></a><span class="lineno"> 1211</span>  ARM_DOT(b0.s3, b1.s3, b2.s3, b3.s3, a2.s0, a2.s1, a2.s2, a2.s3, acc23);</div><div class="line"><a name="l01212"></a><span class="lineno"> 1212</span>  }</div><div class="line"><a name="l01213"></a><span class="lineno"> 1213</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01214"></a><span class="lineno"> 1214</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01215"></a><span class="lineno"> 1215</span>  {</div><div class="line"><a name="l01216"></a><span class="lineno"> 1216</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l01217"></a><span class="lineno"> 1217</span>  ARM_DOT(b0.s0, b1.s0, b2.s0, b3.s0, a3.s0, a3.s1, a3.s2, a3.s3, acc30);</div><div class="line"><a name="l01218"></a><span class="lineno"> 1218</span>  ARM_DOT(b0.s1, b1.s1, b2.s1, b3.s1, a3.s0, a3.s1, a3.s2, a3.s3, acc31);</div><div class="line"><a name="l01219"></a><span class="lineno"> 1219</span>  ARM_DOT(b0.s2, b1.s2, b2.s2, b3.s2, a3.s0, a3.s1, a3.s2, a3.s3, acc32);</div><div class="line"><a name="l01220"></a><span class="lineno"> 1220</span>  ARM_DOT(b0.s3, b1.s3, b2.s3, b3.s3, a3.s0, a3.s1, a3.s2, a3.s3, acc33);</div><div class="line"><a name="l01221"></a><span class="lineno"> 1221</span>  }</div><div class="line"><a name="l01222"></a><span class="lineno"> 1222</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01223"></a><span class="lineno"> 1223</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01224"></a><span class="lineno"> 1224</span>  {</div><div class="line"><a name="l01225"></a><span class="lineno"> 1225</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l01226"></a><span class="lineno"> 1226</span>  ARM_DOT(b0.s0, b1.s0, b2.s0, b3.s0, a4.s0, a4.s1, a4.s2, a4.s3, acc40);</div><div class="line"><a name="l01227"></a><span class="lineno"> 1227</span>  ARM_DOT(b0.s1, b1.s1, b2.s1, b3.s1, a4.s0, a4.s1, a4.s2, a4.s3, acc41);</div><div class="line"><a name="l01228"></a><span class="lineno"> 1228</span>  ARM_DOT(b0.s2, b1.s2, b2.s2, b3.s2, a4.s0, a4.s1, a4.s2, a4.s3, acc42);</div><div class="line"><a name="l01229"></a><span class="lineno"> 1229</span>  ARM_DOT(b0.s3, b1.s3, b2.s3, b3.s3, a4.s0, a4.s1, a4.s2, a4.s3, acc43);</div><div class="line"><a name="l01230"></a><span class="lineno"> 1230</span>  }</div><div class="line"><a name="l01231"></a><span class="lineno"> 1231</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01232"></a><span class="lineno"> 1232</span>  }</div><div class="line"><a name="l01233"></a><span class="lineno"> 1233</span> </div><div class="line"><a name="l01234"></a><span class="lineno"> 1234</span>  <span class="keywordflow">for</span>(; src_addr.s0 < end_row_vec_a; src_addr += (int2)(1, src1_stride_y))</div><div class="line"><a name="l01235"></a><span class="lineno"> 1235</span>  {</div><div class="line"><a name="l01236"></a><span class="lineno"> 1236</span>  <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l01237"></a><span class="lineno"> 1237</span>  uchar a0 = *(src0_ptr + src_addr.s0 + 0 * src0_stride_y);</div><div class="line"><a name="l01238"></a><span class="lineno"> 1238</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01239"></a><span class="lineno"> 1239</span>  uchar a1 = *(src0_ptr + src_addr.s0 + 1 * src0_stride_y);</div><div class="line"><a name="l01240"></a><span class="lineno"> 1240</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01241"></a><span class="lineno"> 1241</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01242"></a><span class="lineno"> 1242</span>  uchar a2 = *(src0_ptr + src_addr.s0 + 2 * src0_stride_y);</div><div class="line"><a name="l01243"></a><span class="lineno"> 1243</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01244"></a><span class="lineno"> 1244</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01245"></a><span class="lineno"> 1245</span>  uchar a3 = *(src0_ptr + src_addr.s0 + 3 * src0_stride_y);</div><div class="line"><a name="l01246"></a><span class="lineno"> 1246</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01247"></a><span class="lineno"> 1247</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01248"></a><span class="lineno"> 1248</span>  uchar a4 = *(src0_ptr + src_addr.s0 + 4 * src0_stride_y);</div><div class="line"><a name="l01249"></a><span class="lineno"> 1249</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01250"></a><span class="lineno"> 1250</span>  <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l01251"></a><span class="lineno"> 1251</span>  uchar4 b0 = vload4(0, src1_ptr + src_addr.s1);</div><div class="line"><a name="l01252"></a><span class="lineno"> 1252</span> </div><div class="line"><a name="l01253"></a><span class="lineno"> 1253</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l01254"></a><span class="lineno"> 1254</span>  {</div><div class="line"><a name="l01255"></a><span class="lineno"> 1255</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l01256"></a><span class="lineno"> 1256</span>  ushort tmp0 = (ushort)b0.s0 * (ushort)a0;</div><div class="line"><a name="l01257"></a><span class="lineno"> 1257</span>  ushort tmp1 = (ushort)b0.s1 * (ushort)a0;</div><div class="line"><a name="l01258"></a><span class="lineno"> 1258</span>  ushort tmp2 = (ushort)b0.s2 * (ushort)a0;</div><div class="line"><a name="l01259"></a><span class="lineno"> 1259</span>  ushort tmp3 = (ushort)b0.s3 * (ushort)a0;</div><div class="line"><a name="l01260"></a><span class="lineno"> 1260</span> </div><div class="line"><a name="l01261"></a><span class="lineno"> 1261</span>  acc00 += ((uint)tmp0);</div><div class="line"><a name="l01262"></a><span class="lineno"> 1262</span>  acc01 += ((uint)tmp1);</div><div class="line"><a name="l01263"></a><span class="lineno"> 1263</span>  acc02 += ((uint)tmp2);</div><div class="line"><a name="l01264"></a><span class="lineno"> 1264</span>  acc03 += ((uint)tmp3);</div><div class="line"><a name="l01265"></a><span class="lineno"> 1265</span>  }</div><div class="line"><a name="l01266"></a><span class="lineno"> 1266</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01267"></a><span class="lineno"> 1267</span>  {</div><div class="line"><a name="l01268"></a><span class="lineno"> 1268</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l01269"></a><span class="lineno"> 1269</span>  ushort tmp0 = (ushort)b0.s0 * (ushort)a1;</div><div class="line"><a name="l01270"></a><span class="lineno"> 1270</span>  ushort tmp1 = (ushort)b0.s1 * (ushort)a1;</div><div class="line"><a name="l01271"></a><span class="lineno"> 1271</span>  ushort tmp2 = (ushort)b0.s2 * (ushort)a1;</div><div class="line"><a name="l01272"></a><span class="lineno"> 1272</span>  ushort tmp3 = (ushort)b0.s3 * (ushort)a1;</div><div class="line"><a name="l01273"></a><span class="lineno"> 1273</span> </div><div class="line"><a name="l01274"></a><span class="lineno"> 1274</span>  acc10 += ((uint)tmp0);</div><div class="line"><a name="l01275"></a><span class="lineno"> 1275</span>  acc11 += ((uint)tmp1);</div><div class="line"><a name="l01276"></a><span class="lineno"> 1276</span>  acc12 += ((uint)tmp2);</div><div class="line"><a name="l01277"></a><span class="lineno"> 1277</span>  acc13 += ((uint)tmp3);</div><div class="line"><a name="l01278"></a><span class="lineno"> 1278</span>  }</div><div class="line"><a name="l01279"></a><span class="lineno"> 1279</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01280"></a><span class="lineno"> 1280</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01281"></a><span class="lineno"> 1281</span>  {</div><div class="line"><a name="l01282"></a><span class="lineno"> 1282</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l01283"></a><span class="lineno"> 1283</span>  ushort tmp0 = (ushort)b0.s0 * (ushort)a2;</div><div class="line"><a name="l01284"></a><span class="lineno"> 1284</span>  ushort tmp1 = (ushort)b0.s1 * (ushort)a2;</div><div class="line"><a name="l01285"></a><span class="lineno"> 1285</span>  ushort tmp2 = (ushort)b0.s2 * (ushort)a2;</div><div class="line"><a name="l01286"></a><span class="lineno"> 1286</span>  ushort tmp3 = (ushort)b0.s3 * (ushort)a2;</div><div class="line"><a name="l01287"></a><span class="lineno"> 1287</span> </div><div class="line"><a name="l01288"></a><span class="lineno"> 1288</span>  acc20 += ((uint)tmp0);</div><div class="line"><a name="l01289"></a><span class="lineno"> 1289</span>  acc21 += ((uint)tmp1);</div><div class="line"><a name="l01290"></a><span class="lineno"> 1290</span>  acc22 += ((uint)tmp2);</div><div class="line"><a name="l01291"></a><span class="lineno"> 1291</span>  acc23 += ((uint)tmp3);</div><div class="line"><a name="l01292"></a><span class="lineno"> 1292</span>  }</div><div class="line"><a name="l01293"></a><span class="lineno"> 1293</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01294"></a><span class="lineno"> 1294</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01295"></a><span class="lineno"> 1295</span>  {</div><div class="line"><a name="l01296"></a><span class="lineno"> 1296</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l01297"></a><span class="lineno"> 1297</span>  ushort tmp0 = (ushort)b0.s0 * (ushort)a3;</div><div class="line"><a name="l01298"></a><span class="lineno"> 1298</span>  ushort tmp1 = (ushort)b0.s1 * (ushort)a3;</div><div class="line"><a name="l01299"></a><span class="lineno"> 1299</span>  ushort tmp2 = (ushort)b0.s2 * (ushort)a3;</div><div class="line"><a name="l01300"></a><span class="lineno"> 1300</span>  ushort tmp3 = (ushort)b0.s3 * (ushort)a3;</div><div class="line"><a name="l01301"></a><span class="lineno"> 1301</span> </div><div class="line"><a name="l01302"></a><span class="lineno"> 1302</span>  acc30 += ((uint)tmp0);</div><div class="line"><a name="l01303"></a><span class="lineno"> 1303</span>  acc31 += ((uint)tmp1);</div><div class="line"><a name="l01304"></a><span class="lineno"> 1304</span>  acc32 += ((uint)tmp2);</div><div class="line"><a name="l01305"></a><span class="lineno"> 1305</span>  acc33 += ((uint)tmp3);</div><div class="line"><a name="l01306"></a><span class="lineno"> 1306</span>  }</div><div class="line"><a name="l01307"></a><span class="lineno"> 1307</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01308"></a><span class="lineno"> 1308</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01309"></a><span class="lineno"> 1309</span>  {</div><div class="line"><a name="l01310"></a><span class="lineno"> 1310</span>  <span class="comment">// Accumulate</span></div><div class="line"><a name="l01311"></a><span class="lineno"> 1311</span>  ushort tmp0 = (ushort)b0.s0 * (ushort)a4;</div><div class="line"><a name="l01312"></a><span class="lineno"> 1312</span>  ushort tmp1 = (ushort)b0.s1 * (ushort)a4;</div><div class="line"><a name="l01313"></a><span class="lineno"> 1313</span>  ushort tmp2 = (ushort)b0.s2 * (ushort)a4;</div><div class="line"><a name="l01314"></a><span class="lineno"> 1314</span>  ushort tmp3 = (ushort)b0.s3 * (ushort)a4;</div><div class="line"><a name="l01315"></a><span class="lineno"> 1315</span> </div><div class="line"><a name="l01316"></a><span class="lineno"> 1316</span>  acc40 += ((uint)tmp0);</div><div class="line"><a name="l01317"></a><span class="lineno"> 1317</span>  acc41 += ((uint)tmp1);</div><div class="line"><a name="l01318"></a><span class="lineno"> 1318</span>  acc42 += ((uint)tmp2);</div><div class="line"><a name="l01319"></a><span class="lineno"> 1319</span>  acc43 += ((uint)tmp3);</div><div class="line"><a name="l01320"></a><span class="lineno"> 1320</span>  }</div><div class="line"><a name="l01321"></a><span class="lineno"> 1321</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01322"></a><span class="lineno"> 1322</span>  }</div><div class="line"><a name="l01323"></a><span class="lineno"> 1323</span> </div><div class="line"><a name="l01324"></a><span class="lineno"> 1324</span>  <span class="comment">// Compute destination address</span></div><div class="line"><a name="l01325"></a><span class="lineno"> 1325</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l01326"></a><span class="lineno"> 1326</span> </div><div class="line"><a name="l01327"></a><span class="lineno"> 1327</span>  <span class="comment">// Store the result</span></div><div class="line"><a name="l01328"></a><span class="lineno"> 1328</span>  vstore4((int4)(acc00, acc01, acc02, acc03), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div><div class="line"><a name="l01329"></a><span class="lineno"> 1329</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01330"></a><span class="lineno"> 1330</span>  vstore4((int4)(acc10, acc11, acc12, acc13), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 1)));</div><div class="line"><a name="l01331"></a><span class="lineno"> 1331</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div><div class="line"><a name="l01332"></a><span class="lineno"> 1332</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01333"></a><span class="lineno"> 1333</span>  vstore4((int4)(acc20, acc21, acc22, acc23), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 2)));</div><div class="line"><a name="l01334"></a><span class="lineno"> 1334</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div><div class="line"><a name="l01335"></a><span class="lineno"> 1335</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01336"></a><span class="lineno"> 1336</span>  vstore4((int4)(acc30, acc31, acc32, acc33), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 3)));</div><div class="line"><a name="l01337"></a><span class="lineno"> 1337</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div><div class="line"><a name="l01338"></a><span class="lineno"> 1338</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01339"></a><span class="lineno"> 1339</span>  vstore4((int4)(acc40, acc41, acc42, acc43), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 4)));</div><div class="line"><a name="l01340"></a><span class="lineno"> 1340</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 4</span></div><div class="line"><a name="l01341"></a><span class="lineno"> 1341</span> }</div><div class="line"><a name="l01342"></a><span class="lineno"> 1342</span> <span class="preprocessor">#endif // ARM_COMPUTE_OPENCL_DOT8_ENABLED</span></div><div class="line"><a name="l01343"></a><span class="lineno"> 1343</span> </div><div class="line"><a name="l01344"></a><span class="lineno"> 1344</span> <span class="preprocessor">#endif // defined(NUM_ELEMS_PROCESSED_PER_THREAD_X) && defined(NUM_ELEMS_PROCESSED_PER_THREAD_Y) && defined(COLS_A)</span></div><div class="line"><a name="l01345"></a><span class="lineno"> 1345</span> </div><div class="line"><a name="l01346"></a><span class="lineno"> 1346</span> <span class="preprocessor">#if defined(COLS_A)</span></div><div class="line"><a name="l01347"></a><span class="lineno"> 1347</span> </div><div class="line"><a name="l01369"></a><span class="lineno"> 1369</span> __kernel <span class="keywordtype">void</span> gemmlowp_matrix_a_reduction(<a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a6743f0a130e8311e6f5b1a23df102472">src</a>),</div><div class="line"><a name="l01370"></a><span class="lineno"> 1370</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div><div class="line"><a name="l01371"></a><span class="lineno"> 1371</span> {</div><div class="line"><a name="l01372"></a><span class="lineno"> 1372</span>  <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l01373"></a><span class="lineno"> 1373</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a6743f0a130e8311e6f5b1a23df102472">src</a> = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(src);</div><div class="line"><a name="l01374"></a><span class="lineno"> 1374</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l01375"></a><span class="lineno"> 1375</span> </div><div class="line"><a name="l01376"></a><span class="lineno"> 1376</span>  uint4 sum_row_u32 = (uint4)0;</div><div class="line"><a name="l01377"></a><span class="lineno"> 1377</span>  uint sum_row = 0;</div><div class="line"><a name="l01378"></a><span class="lineno"> 1378</span> </div><div class="line"><a name="l01379"></a><span class="lineno"> 1379</span>  __global <span class="keyword">const</span> uchar *matrix_a = (__global <span class="keyword">const</span> uchar *)(src.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + get_global_id(0) * src_stride_y + get_global_id(1) * src_stride_z);</div><div class="line"><a name="l01380"></a><span class="lineno"> 1380</span> </div><div class="line"><a name="l01381"></a><span class="lineno"> 1381</span>  <span class="keywordtype">int</span> i = 0;</div><div class="line"><a name="l01382"></a><span class="lineno"> 1382</span> </div><div class="line"><a name="l01383"></a><span class="lineno"> 1383</span>  <span class="comment">// This for loop performs 16 accumulations</span></div><div class="line"><a name="l01384"></a><span class="lineno"> 1384</span>  <span class="keywordflow">for</span>(; i <= ((int)COLS_A - 16); i += 16)</div><div class="line"><a name="l01385"></a><span class="lineno"> 1385</span>  {</div><div class="line"><a name="l01386"></a><span class="lineno"> 1386</span>  <span class="keyword">const</span> uchar16 a0_u8 = vload16(0, matrix_a + i);</div><div class="line"><a name="l01387"></a><span class="lineno"> 1387</span> </div><div class="line"><a name="l01388"></a><span class="lineno"> 1388</span>  sum_row_u32 += convert_uint4(a0_u8.s0123) + convert_uint4(a0_u8.s4567) + convert_uint4(a0_u8.s89AB) + convert_uint4(a0_u8.sCDEF);</div><div class="line"><a name="l01389"></a><span class="lineno"> 1389</span>  }</div><div class="line"><a name="l01390"></a><span class="lineno"> 1390</span> </div><div class="line"><a name="l01391"></a><span class="lineno"> 1391</span>  <span class="comment">// This for loop performs the leftover accumulations</span></div><div class="line"><a name="l01392"></a><span class="lineno"> 1392</span>  <span class="keywordflow">for</span>(; i < COLS_A; ++i)</div><div class="line"><a name="l01393"></a><span class="lineno"> 1393</span>  {</div><div class="line"><a name="l01394"></a><span class="lineno"> 1394</span>  sum_row += matrix_a[i];</div><div class="line"><a name="l01395"></a><span class="lineno"> 1395</span>  }</div><div class="line"><a name="l01396"></a><span class="lineno"> 1396</span> </div><div class="line"><a name="l01397"></a><span class="lineno"> 1397</span>  sum_row += sum_row_u32.s0 + sum_row_u32.s1 + sum_row_u32.s2 + sum_row_u32.s3;</div><div class="line"><a name="l01398"></a><span class="lineno"> 1398</span> </div><div class="line"><a name="l01399"></a><span class="lineno"> 1399</span>  *((__global <span class="keywordtype">int</span> *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) = (int)sum_row;</div><div class="line"><a name="l01400"></a><span class="lineno"> 1400</span> }</div><div class="line"><a name="l01401"></a><span class="lineno"> 1401</span> <span class="preprocessor">#endif // defined(COLS_A)</span></div><div class="line"><a name="l01402"></a><span class="lineno"> 1402</span> </div><div class="line"><a name="l01403"></a><span class="lineno"> 1403</span> <span class="preprocessor">#if defined(COLS_B) && defined(ROWS_B)</span></div><div class="line"><a name="l01404"></a><span class="lineno"> 1404</span> </div><div class="line"><a name="l01426"></a><span class="lineno"> 1426</span> __kernel <span class="keywordtype">void</span> gemmlowp_matrix_b_reduction(<a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l01427"></a><span class="lineno"> 1427</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div><div class="line"><a name="l01428"></a><span class="lineno"> 1428</span> {</div><div class="line"><a name="l01429"></a><span class="lineno"> 1429</span>  <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l01430"></a><span class="lineno"> 1430</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> src = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(src);</div><div class="line"><a name="l01431"></a><span class="lineno"> 1431</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l01432"></a><span class="lineno"> 1432</span> </div><div class="line"><a name="l01433"></a><span class="lineno"> 1433</span>  uint16 sum_col_u32 = (uint16)0;</div><div class="line"><a name="l01434"></a><span class="lineno"> 1434</span> </div><div class="line"><a name="l01435"></a><span class="lineno"> 1435</span>  __global <span class="keyword">const</span> uchar *matrix_b = (__global <span class="keyword">const</span> uchar *)(src.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + get_global_id(1) * src_stride_z);</div><div class="line"><a name="l01436"></a><span class="lineno"> 1436</span> </div><div class="line"><a name="l01437"></a><span class="lineno"> 1437</span>  <span class="keywordtype">int</span> i = 0;</div><div class="line"><a name="l01438"></a><span class="lineno"> 1438</span>  <span class="comment">// This for loop performs 4 accumulations</span></div><div class="line"><a name="l01439"></a><span class="lineno"> 1439</span>  <span class="keywordflow">for</span>(; i <= ((int)ROWS_B - 4); i += 4)</div><div class="line"><a name="l01440"></a><span class="lineno"> 1440</span>  {</div><div class="line"><a name="l01441"></a><span class="lineno"> 1441</span>  <span class="keyword">const</span> uchar16 b0_u8 = vload16(0, matrix_b + 0 * src_stride_y);</div><div class="line"><a name="l01442"></a><span class="lineno"> 1442</span>  <span class="keyword">const</span> uchar16 b1_u8 = vload16(0, matrix_b + 1 * src_stride_y);</div><div class="line"><a name="l01443"></a><span class="lineno"> 1443</span>  <span class="keyword">const</span> uchar16 b2_u8 = vload16(0, matrix_b + 2 * src_stride_y);</div><div class="line"><a name="l01444"></a><span class="lineno"> 1444</span>  <span class="keyword">const</span> uchar16 b3_u8 = vload16(0, matrix_b + 3 * src_stride_y);</div><div class="line"><a name="l01445"></a><span class="lineno"> 1445</span> </div><div class="line"><a name="l01446"></a><span class="lineno"> 1446</span>  sum_col_u32 += convert_uint16(b0_u8) + convert_uint16(b1_u8) + convert_uint16(b2_u8) + convert_uint16(b3_u8);</div><div class="line"><a name="l01447"></a><span class="lineno"> 1447</span> </div><div class="line"><a name="l01448"></a><span class="lineno"> 1448</span>  matrix_b += 4 * src_stride_y;</div><div class="line"><a name="l01449"></a><span class="lineno"> 1449</span>  }</div><div class="line"><a name="l01450"></a><span class="lineno"> 1450</span> </div><div class="line"><a name="l01451"></a><span class="lineno"> 1451</span>  <span class="comment">// This for loop perfoms the leftover accumulations</span></div><div class="line"><a name="l01452"></a><span class="lineno"> 1452</span>  <span class="keywordflow">for</span>(; i < (int)ROWS_B; ++i)</div><div class="line"><a name="l01453"></a><span class="lineno"> 1453</span>  {</div><div class="line"><a name="l01454"></a><span class="lineno"> 1454</span>  <span class="keyword">const</span> uchar16 b0_u8 = vload16(0, matrix_b);</div><div class="line"><a name="l01455"></a><span class="lineno"> 1455</span> </div><div class="line"><a name="l01456"></a><span class="lineno"> 1456</span>  sum_col_u32 += convert_uint16(b0_u8);</div><div class="line"><a name="l01457"></a><span class="lineno"> 1457</span> </div><div class="line"><a name="l01458"></a><span class="lineno"> 1458</span>  matrix_b += src_stride_y;</div><div class="line"><a name="l01459"></a><span class="lineno"> 1459</span>  }</div><div class="line"><a name="l01460"></a><span class="lineno"> 1460</span> </div><div class="line"><a name="l01461"></a><span class="lineno"> 1461</span>  vstore16(convert_int16(sum_col_u32), 0, (__global <span class="keywordtype">int</span> *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l01462"></a><span class="lineno"> 1462</span> }</div><div class="line"><a name="l01463"></a><span class="lineno"> 1463</span> <span class="preprocessor">#endif // defined(COLS_B) && defined(ROWS_B)</span></div><div class="line"><a name="l01464"></a><span class="lineno"> 1464</span> </div><div class="line"><a name="l01465"></a><span class="lineno"> 1465</span> <span class="preprocessor">#if defined(K_OFFSET)</span></div><div class="line"><a name="l01466"></a><span class="lineno"> 1466</span> <span class="comment">/* OpenCL kernel used to add the offset contribution after @ref CLGEMMLowpMatrixMultiplyKernel. The computation is performed in-place</span></div><div class="line"><a name="l01467"></a><span class="lineno"> 1467</span> <span class="comment"> *</span></div><div class="line"><a name="l01468"></a><span class="lineno"> 1468</span> <span class="comment"> * This kernel takes a final int32 accumulator value (the output of @CLGEMMLowpMatrixMultiplyKernel),</span></div><div class="line"><a name="l01469"></a><span class="lineno"> 1469</span> <span class="comment"> * and adds to it the offset contribution of matrix A and matrix B in-place.</span></div><div class="line"><a name="l01470"></a><span class="lineno"> 1470</span> <span class="comment"> *</span></div><div class="line"><a name="l01471"></a><span class="lineno"> 1471</span> <span class="comment"> * @attention The k_offset = a_offset * b_offset * k (where k is the number of matrix A columns) needs to be passed at compile time using -DK_OFFSET (i.e. -DK_OFFSET=1200)</span></div><div class="line"><a name="l01472"></a><span class="lineno"> 1472</span> <span class="comment"> * @note In case the offset contribution due to a_offset is required, a_offset needs to be passed at compile time using -DA_OFFSET (i.e. -DA_OFFSET=1)</span></div><div class="line"><a name="l01473"></a><span class="lineno"> 1473</span> <span class="comment"> * @note In case the offset contribution due to b_offset is required, b_offset needs to be passed at compile time using -DB_OFFSET (i.e. -DB_OFFSET=6)</span></div><div class="line"><a name="l01474"></a><span class="lineno"> 1474</span> <span class="comment"> * @note In case sum_col has batches, -DSUM_COL_HAS_BATCHES must be passed at compile time. Usually if gemmlowp is used to accelerate convolution layer, sum_col will not have batches</span></div><div class="line"><a name="l01475"></a><span class="lineno"> 1475</span> <span class="comment"> *</span></div><div class="line"><a name="l01476"></a><span class="lineno"> 1476</span> <span class="comment"> * The final result is:</span></div><div class="line"><a name="l01477"></a><span class="lineno"> 1477</span> <span class="comment"> *</span></div><div class="line"><a name="l01478"></a><span class="lineno"> 1478</span> <span class="comment"> * mm_result[i][k] = mm_result[i][k] +</span></div><div class="line"><a name="l01479"></a><span class="lineno"> 1479</span> <span class="comment"> * (sum_col[k] * A_OFFSET) +</span></div><div class="line"><a name="l01480"></a><span class="lineno"> 1480</span> <span class="comment"> * (sum_row[i] * B_OFFSET) +</span></div><div class="line"><a name="l01481"></a><span class="lineno"> 1481</span> <span class="comment"> * (K_OFFSET)</span></div><div class="line"><a name="l01482"></a><span class="lineno"> 1482</span> <span class="comment"> *</span></div><div class="line"><a name="l01483"></a><span class="lineno"> 1483</span> <span class="comment"> * @param[in] mm_result_ptr Pointer to the source tensor. Supported data type: S32</span></div><div class="line"><a name="l01484"></a><span class="lineno"> 1484</span> <span class="comment"> * @param[in] mm_result_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01485"></a><span class="lineno"> 1485</span> <span class="comment"> * @param[in] mm_result_step_x mm_result_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01486"></a><span class="lineno"> 1486</span> <span class="comment"> * @param[in] mm_result_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01487"></a><span class="lineno"> 1487</span> <span class="comment"> * @param[in] mm_result_step_y mm_result_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01488"></a><span class="lineno"> 1488</span> <span class="comment"> * @param[in] mm_result_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01489"></a><span class="lineno"> 1489</span> <span class="comment"> * @param[in] mm_result_step_z mm_result_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01490"></a><span class="lineno"> 1490</span> <span class="comment"> * @param[in] mm_result_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01491"></a><span class="lineno"> 1491</span> <span class="comment"> * @param[in] sum_col_result_ptr Pointer to the source tensor. Supported data type: same as @p mm_result_ptr</span></div><div class="line"><a name="l01492"></a><span class="lineno"> 1492</span> <span class="comment"> * @param[in] sum_col_result_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01493"></a><span class="lineno"> 1493</span> <span class="comment"> * @param[in] sum_col_result_step_x sum_col_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01494"></a><span class="lineno"> 1494</span> <span class="comment"> * @param[in] sum_col_result_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01495"></a><span class="lineno"> 1495</span> <span class="comment"> * @param[in] sum_col_result_step_y sum_col_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01496"></a><span class="lineno"> 1496</span> <span class="comment"> * @param[in] sum_col_result_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01497"></a><span class="lineno"> 1497</span> <span class="comment"> * @param[in] sum_row_result_ptr Pointer to the source tensor. Supported data type: same as @p mm_result_ptr</span></div><div class="line"><a name="l01498"></a><span class="lineno"> 1498</span> <span class="comment"> * @param[in] sum_row_result_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01499"></a><span class="lineno"> 1499</span> <span class="comment"> * @param[in] sum_row_result_step_x sum_row_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01500"></a><span class="lineno"> 1500</span> <span class="comment"> * @param[in] sum_row_result_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01501"></a><span class="lineno"> 1501</span> <span class="comment"> * @param[in] sum_row_result_step_y sum_row_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01502"></a><span class="lineno"> 1502</span> <span class="comment"> * @param[in] sum_row_result_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01503"></a><span class="lineno"> 1503</span> <span class="comment"> */</span></div><div class="line"><a name="l01504"></a><span class="lineno"> 1504</span> __kernel <span class="keywordtype">void</span> gemmlowp_offset_contribution(<a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(mm_result)</div><div class="line"><a name="l01505"></a><span class="lineno"> 1505</span> #<span class="keywordflow">if</span> defined(A_OFFSET)</div><div class="line"><a name="l01506"></a><span class="lineno"> 1506</span>  ,</div><div class="line"><a name="l01507"></a><span class="lineno"> 1507</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(sum_col)</div><div class="line"><a name="l01508"></a><span class="lineno"> 1508</span> #endif <span class="comment">// defined(A_OFFSET)</span></div><div class="line"><a name="l01509"></a><span class="lineno"> 1509</span> #<span class="keywordflow">if</span> defined(B_OFFSET)</div><div class="line"><a name="l01510"></a><span class="lineno"> 1510</span>  ,</div><div class="line"><a name="l01511"></a><span class="lineno"> 1511</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(sum_row)</div><div class="line"><a name="l01512"></a><span class="lineno"> 1512</span> #endif <span class="comment">// defined(B_OFFSET)</span></div><div class="line"><a name="l01513"></a><span class="lineno"> 1513</span>  )</div><div class="line"><a name="l01514"></a><span class="lineno"> 1514</span> {</div><div class="line"><a name="l01515"></a><span class="lineno"> 1515</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> mm_result = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(mm_result);</div><div class="line"><a name="l01516"></a><span class="lineno"> 1516</span> </div><div class="line"><a name="l01517"></a><span class="lineno"> 1517</span>  int4 a_offset_s32 = (int4)0;</div><div class="line"><a name="l01518"></a><span class="lineno"> 1518</span>  int4 b_offset_s32 = (int4)0;</div><div class="line"><a name="l01519"></a><span class="lineno"> 1519</span> </div><div class="line"><a name="l01520"></a><span class="lineno"> 1520</span> <span class="preprocessor">#if defined(A_OFFSET)</span></div><div class="line"><a name="l01521"></a><span class="lineno"> 1521</span>  <a class="code" href="struct_image.xhtml">Image</a> sum_col = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(sum_col);</div><div class="line"><a name="l01522"></a><span class="lineno"> 1522</span> </div><div class="line"><a name="l01523"></a><span class="lineno"> 1523</span>  <span class="comment">// Compute the offset contribution due to A_OFFSET</span></div><div class="line"><a name="l01524"></a><span class="lineno"> 1524</span> <span class="preprocessor">#if defined(SUM_COL_HAS_BATCHES)</span></div><div class="line"><a name="l01525"></a><span class="lineno"> 1525</span>  a_offset_s32 = vload4(0, (__global <span class="keywordtype">int</span> *)(sum_col.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + get_global_id(2) * sum_col_stride_y));</div><div class="line"><a name="l01526"></a><span class="lineno"> 1526</span> <span class="preprocessor">#else // defined(MATRIX_B_HAS_BATCHES)</span></div><div class="line"><a name="l01527"></a><span class="lineno"> 1527</span>  a_offset_s32 = vload4(0, (__global <span class="keywordtype">int</span> *)(sum_col.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>));</div><div class="line"><a name="l01528"></a><span class="lineno"> 1528</span> <span class="preprocessor">#endif // defined(MATRIX_B_HAS_BATCHES)</span></div><div class="line"><a name="l01529"></a><span class="lineno"> 1529</span> </div><div class="line"><a name="l01530"></a><span class="lineno"> 1530</span>  a_offset_s32 *= (int4)A_OFFSET;</div><div class="line"><a name="l01531"></a><span class="lineno"> 1531</span> <span class="preprocessor">#endif // defined(A_OFFSET)</span></div><div class="line"><a name="l01532"></a><span class="lineno"> 1532</span> </div><div class="line"><a name="l01533"></a><span class="lineno"> 1533</span> <span class="preprocessor">#if defined(B_OFFSET)</span></div><div class="line"><a name="l01534"></a><span class="lineno"> 1534</span>  <a class="code" href="struct_image.xhtml">Image</a> sum_row = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(sum_row);</div><div class="line"><a name="l01535"></a><span class="lineno"> 1535</span> </div><div class="line"><a name="l01536"></a><span class="lineno"> 1536</span>  <span class="comment">// Compute the offset contribution due to B_OFFSET</span></div><div class="line"><a name="l01537"></a><span class="lineno"> 1537</span>  b_offset_s32 = (int4) * (((__global <span class="keywordtype">int</span> *)(sum_row.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a> + get_global_id(2) * sum_row_stride_y)) + get_global_id(1));</div><div class="line"><a name="l01538"></a><span class="lineno"> 1538</span>  b_offset_s32 *= (int4)B_OFFSET;</div><div class="line"><a name="l01539"></a><span class="lineno"> 1539</span> <span class="preprocessor">#endif // defined(B_OFFSET)</span></div><div class="line"><a name="l01540"></a><span class="lineno"> 1540</span> </div><div class="line"><a name="l01541"></a><span class="lineno"> 1541</span>  <span class="keyword">const</span> int4 offset_term_s32 = (int4)K_OFFSET + a_offset_s32 + b_offset_s32;</div><div class="line"><a name="l01542"></a><span class="lineno"> 1542</span> </div><div class="line"><a name="l01543"></a><span class="lineno"> 1543</span>  int4 in_s32 = vload4(0, (__global <span class="keywordtype">int</span> *)mm_result.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l01544"></a><span class="lineno"> 1544</span> </div><div class="line"><a name="l01545"></a><span class="lineno"> 1545</span>  <span class="comment">// Add the offset terms to GEMM's result</span></div><div class="line"><a name="l01546"></a><span class="lineno"> 1546</span>  in_s32 += offset_term_s32;</div><div class="line"><a name="l01547"></a><span class="lineno"> 1547</span> </div><div class="line"><a name="l01548"></a><span class="lineno"> 1548</span>  <span class="comment">// Store the result with the offset contribution</span></div><div class="line"><a name="l01549"></a><span class="lineno"> 1549</span>  vstore4(in_s32, 0, (__global <span class="keywordtype">int</span> *)mm_result.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l01550"></a><span class="lineno"> 1550</span> }</div><div class="line"><a name="l01551"></a><span class="lineno"> 1551</span> <span class="preprocessor">#endif // defined(K_OFFSET)</span></div><div class="line"><a name="l01552"></a><span class="lineno"> 1552</span> </div><div class="line"><a name="l01553"></a><span class="lineno"> 1553</span> <span class="preprocessor">#if defined(RESULT_OFFSET) && defined(RESULT_MULT_INT) && defined(RESULT_SHIFT)</span></div><div class="line"><a name="l01554"></a><span class="lineno"> 1554</span> </div><div class="line"><a name="l01593"></a><span class="lineno"> 1593</span> __kernel <span class="keywordtype">void</span> gemmlowp_output_stage_quantize_down(<a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l01594"></a><span class="lineno"> 1594</span> #<span class="keywordflow">if</span> defined(ADD_BIAS)</div><div class="line"><a name="l01595"></a><span class="lineno"> 1595</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases),</div><div class="line"><a name="l01596"></a><span class="lineno"> 1596</span> #endif <span class="comment">// defined(ADD_BIAS)</span></div><div class="line"><a name="l01597"></a><span class="lineno"> 1597</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(dst))</div><div class="line"><a name="l01598"></a><span class="lineno"> 1598</span> {</div><div class="line"><a name="l01599"></a><span class="lineno"> 1599</span>  <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l01600"></a><span class="lineno"> 1600</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> src = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(src);</div><div class="line"><a name="l01601"></a><span class="lineno"> 1601</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> dst = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(dst);</div><div class="line"><a name="l01602"></a><span class="lineno"> 1602</span> <span class="preprocessor">#if defined(ADD_BIAS)</span></div><div class="line"><a name="l01603"></a><span class="lineno"> 1603</span>  <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="helpers_8h.xhtml#a527bfdf5eeb306f1cf01c4a8e29f38e0">CONVERT_TO_VECTOR_STRUCT</a>(biases);</div><div class="line"><a name="l01604"></a><span class="lineno"> 1604</span> <span class="preprocessor">#endif // defined(ADD_BIAS)</span></div><div class="line"><a name="l01605"></a><span class="lineno"> 1605</span> </div><div class="line"><a name="l01606"></a><span class="lineno"> 1606</span>  int16 input_values = vload16(0, (__global <span class="keywordtype">int</span> *)src.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l01607"></a><span class="lineno"> 1607</span> </div><div class="line"><a name="l01608"></a><span class="lineno"> 1608</span>  <span class="comment">// Add the offset terms to GEMM's result</span></div><div class="line"><a name="l01609"></a><span class="lineno"> 1609</span>  input_values += (int16)RESULT_OFFSET;</div><div class="line"><a name="l01610"></a><span class="lineno"> 1610</span> </div><div class="line"><a name="l01611"></a><span class="lineno"> 1611</span> <span class="preprocessor">#if defined(ADD_BIAS)</span></div><div class="line"><a name="l01612"></a><span class="lineno"> 1612</span>  <span class="comment">// Add bias</span></div><div class="line"><a name="l01613"></a><span class="lineno"> 1613</span>  <span class="keyword">const</span> int16 biases_values = vload16(0, (__global <span class="keywordtype">int</span> *)biases.<a class="code" href="struct_vector.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l01614"></a><span class="lineno"> 1614</span>  input_values += (int16)biases_values;</div><div class="line"><a name="l01615"></a><span class="lineno"> 1615</span> <span class="preprocessor">#endif // defined(ADD_BIAS)</span></div><div class="line"><a name="l01616"></a><span class="lineno"> 1616</span> </div><div class="line"><a name="l01617"></a><span class="lineno"> 1617</span>  <span class="comment">// Multiply by result_mult_int and shift</span></div><div class="line"><a name="l01618"></a><span class="lineno"> 1618</span>  input_values *= RESULT_MULT_INT;</div><div class="line"><a name="l01619"></a><span class="lineno"> 1619</span> </div><div class="line"><a name="l01620"></a><span class="lineno"> 1620</span>  input_values >>= RESULT_SHIFT;</div><div class="line"><a name="l01621"></a><span class="lineno"> 1621</span> </div><div class="line"><a name="l01622"></a><span class="lineno"> 1622</span>  uchar16 res = convert_uchar16_sat(input_values);</div><div class="line"><a name="l01623"></a><span class="lineno"> 1623</span> </div><div class="line"><a name="l01624"></a><span class="lineno"> 1624</span> <span class="preprocessor">#if defined(MIN_BOUND)</span></div><div class="line"><a name="l01625"></a><span class="lineno"> 1625</span>  res = max(res, (uchar16)MIN_BOUND);</div><div class="line"><a name="l01626"></a><span class="lineno"> 1626</span> <span class="preprocessor">#endif // defined(MIN_BOUND)</span></div><div class="line"><a name="l01627"></a><span class="lineno"> 1627</span> <span class="preprocessor">#if defined(MAX_BOUND)</span></div><div class="line"><a name="l01628"></a><span class="lineno"> 1628</span>  res = min(res, (uchar16)MAX_BOUND);</div><div class="line"><a name="l01629"></a><span class="lineno"> 1629</span> <span class="preprocessor">#endif // defined(MAX_BOUND)</span></div><div class="line"><a name="l01630"></a><span class="lineno"> 1630</span> </div><div class="line"><a name="l01631"></a><span class="lineno"> 1631</span>  <span class="comment">// Store the result</span></div><div class="line"><a name="l01632"></a><span class="lineno"> 1632</span>  vstore16(res, 0, dst.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l01633"></a><span class="lineno"> 1633</span> }</div><div class="line"><a name="l01634"></a><span class="lineno"> 1634</span> <span class="preprocessor">#endif // defined(RESULT_OFFSET) && defined(RESULT_MULT_INT) && defined(RESULT_SHIFT)</span></div><div class="line"><a name="l01635"></a><span class="lineno"> 1635</span> </div><div class="line"><a name="l01636"></a><span class="lineno"> 1636</span> <span class="preprocessor">#if defined(RESULT_OFFSET_AFTER_SHIFT) && defined(RESULT_FIXEDPOINT_MULTIPLIER) && defined(RESULT_SHIFT)</span></div><div class="line"><a name="l01637"></a><span class="lineno"> 1637</span> </div><div class="line"><a name="l01676"></a><span class="lineno"> 1676</span> __kernel <span class="keywordtype">void</span> gemmlowp_output_stage_quantize_down_fixedpoint(<a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l01677"></a><span class="lineno"> 1677</span> #<span class="keywordflow">if</span> defined(ADD_BIAS)</div><div class="line"><a name="l01678"></a><span class="lineno"> 1678</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases),</div><div class="line"><a name="l01679"></a><span class="lineno"> 1679</span> #endif <span class="comment">// defined(ADD_BIAS)</span></div><div class="line"><a name="l01680"></a><span class="lineno"> 1680</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(dst))</div><div class="line"><a name="l01681"></a><span class="lineno"> 1681</span> {</div><div class="line"><a name="l01682"></a><span class="lineno"> 1682</span>  <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l01683"></a><span class="lineno"> 1683</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> src = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(src);</div><div class="line"><a name="l01684"></a><span class="lineno"> 1684</span>  <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> dst = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(dst);</div><div class="line"><a name="l01685"></a><span class="lineno"> 1685</span> <span class="preprocessor">#if defined(ADD_BIAS)</span></div><div class="line"><a name="l01686"></a><span class="lineno"> 1686</span>  <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="helpers_8h.xhtml#a527bfdf5eeb306f1cf01c4a8e29f38e0">CONVERT_TO_VECTOR_STRUCT</a>(biases);</div><div class="line"><a name="l01687"></a><span class="lineno"> 1687</span> <span class="preprocessor">#endif // defined(ADD_BIAS)</span></div><div class="line"><a name="l01688"></a><span class="lineno"> 1688</span> </div><div class="line"><a name="l01689"></a><span class="lineno"> 1689</span>  int16 input_values = vload16(0, (__global <span class="keywordtype">int</span> *)src.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l01690"></a><span class="lineno"> 1690</span> </div><div class="line"><a name="l01691"></a><span class="lineno"> 1691</span> <span class="preprocessor">#if defined(ADD_BIAS)</span></div><div class="line"><a name="l01692"></a><span class="lineno"> 1692</span>  <span class="comment">// Add bias</span></div><div class="line"><a name="l01693"></a><span class="lineno"> 1693</span>  <span class="keyword">const</span> int16 biases_values = vload16(0, (__global <span class="keywordtype">int</span> *)biases.<a class="code" href="struct_vector.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l01694"></a><span class="lineno"> 1694</span>  input_values += (int16)biases_values;</div><div class="line"><a name="l01695"></a><span class="lineno"> 1695</span> <span class="preprocessor">#endif // defined(ADD_BIAS)</span></div><div class="line"><a name="l01696"></a><span class="lineno"> 1696</span> </div><div class="line"><a name="l01697"></a><span class="lineno"> 1697</span>  <span class="comment">// Multiply by result_mult_int and shift</span></div><div class="line"><a name="l01698"></a><span class="lineno"> 1698</span>  input_values = <a class="code" href="helpers__asymm_8h.xhtml#a86de0ffca367bfcf27a8ae5dd1fdef2d">ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE</a>(input_values, RESULT_FIXEDPOINT_MULTIPLIER, RESULT_SHIFT, 16);</div><div class="line"><a name="l01699"></a><span class="lineno"> 1699</span> </div><div class="line"><a name="l01700"></a><span class="lineno"> 1700</span>  <span class="comment">// Add the offset terms to GEMM's result</span></div><div class="line"><a name="l01701"></a><span class="lineno"> 1701</span>  input_values += (int16)RESULT_OFFSET_AFTER_SHIFT;</div><div class="line"><a name="l01702"></a><span class="lineno"> 1702</span> </div><div class="line"><a name="l01703"></a><span class="lineno"> 1703</span>  uchar16 res = convert_uchar16_sat(input_values);</div><div class="line"><a name="l01704"></a><span class="lineno"> 1704</span> </div><div class="line"><a name="l01705"></a><span class="lineno"> 1705</span> <span class="preprocessor">#if defined(MIN_BOUND)</span></div><div class="line"><a name="l01706"></a><span class="lineno"> 1706</span>  res = max(res, (uchar16)MIN_BOUND);</div><div class="line"><a name="l01707"></a><span class="lineno"> 1707</span> <span class="preprocessor">#endif // defined(MIN_BOUND)</span></div><div class="line"><a name="l01708"></a><span class="lineno"> 1708</span> <span class="preprocessor">#if defined(MAX_BOUND)</span></div><div class="line"><a name="l01709"></a><span class="lineno"> 1709</span>  res = min(res, (uchar16)MAX_BOUND);</div><div class="line"><a name="l01710"></a><span class="lineno"> 1710</span> <span class="preprocessor">#endif // defined(MAX_BOUND)</span></div><div class="line"><a name="l01711"></a><span class="lineno"> 1711</span> </div><div class="line"><a name="l01712"></a><span class="lineno"> 1712</span>  <span class="comment">// Store the result</span></div><div class="line"><a name="l01713"></a><span class="lineno"> 1713</span>  vstore16(res, 0, dst.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l01714"></a><span class="lineno"> 1714</span> }</div><div class="line"><a name="l01715"></a><span class="lineno"> 1715</span> <span class="preprocessor">#endif // defined(RESULT_OFFSET_AFTER_SHIFT) && defined(RESULT_FIXEDPOINT_MULTIPLIER) && defined(RESULT_SHIFT)</span></div><div class="ttc" id="struct_vector_xhtml"><div class="ttname"><a href="struct_vector.xhtml">Vector</a></div><div class="ttdoc">Structure to hold Vector information. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00142">helpers.h:142</a></div></div> |