API tools faq
paste
Advertisement
Guest User

Untitled

a guest
Jul 24th, 2020
57
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
C++ 6.25 KB | None | 0 0
raw download clone embed print report
  1. #include "conv.h"
  2.  
  3. F_DATA data_buf[3][20][30];
  4. F_WEIGHT weight_buf[3][3][3];
  5. F_ACC bias_buf[18][28];
  6. F_ACC ans_buf[18][28];
  7.  
  8.  
  9. void conv(u8 data[3][20][30], u8 weight[3][3][3], u32 bias[18][28], u32 ans[18][28]) {
  10. #pragma HLS INTERFACE m_axi depth=512 port=ans bundle=DATA
  11. #pragma HLS INTERFACE s_axilite depth=512 port=ans bundle=CTRL
  12. #pragma HLS INTERFACE m_axi depth=512 port=bias bundle=DATA
  13. #pragma HLS INTERFACE s_axilite depth=512 port=bias bundle=CTRL
  14. #pragma HLS INTERFACE m_axi depth=512 port=weight bundle=DATA
  15. #pragma HLS INTERFACE s_axilite depth=512 port=weight bundle=CTRL
  16. #pragma HLS INTERFACE s_axilite register port=data bundle=CTRL
  17. #pragma HLS INTERFACE m_axi depth=512 port=data bundle=DATA
  18. #pragma HLS INTERFACE s_axilite register port=return bundle=CTRL
  19.     data_to_buf(data, data_buf);
  20.     weight_to_buf(weight, weight_buf);
  21.     bias_to_buf(bias, bias_buf);
  22.     set_bias(bias_buf, ans_buf);
  23.     compute(data_buf, weight_buf, ans_buf);
  24.     ans_to_ddr(ans_buf, ans);
  25.     return;
  26. }
  27.  
  28. void data_to_buf(u8 data[3][20][30], F_DATA data_buf[3][20][30]) {
  29.  
  30.     for (int i = 0; i < 3; ++i) {
  31.         for (int j = 0; j < 20; ++j) {
  32. #pragma HLS PIPELINE
  33.             for (int k = 0; k < 30; ++k) {
  34. #pragma HLS UNROLL
  35.                 data_buf[i][j][k] = data[i][j][k];
  36.             }
  37.         }
  38.     }
  39. }
  40.  
  41. void weight_to_buf(u8 weight[3][3][3], F_WEIGHT weight_buf[3][3][3]) {
  42.     for (int i = 0; i < 3; ++i) {
  43.         for (int j = 0; j < 3; ++j) {
  44. #pragma HLS PIPELINE
  45.             for (int k = 0; k < 3; ++k) {
  46. #pragma HLS UNROLL
  47.                 weight_buf[i][j][k] = weight[i][j][k];
  48.             }
  49.         }
  50.     }
  51. }
  52.  
  53. void bias_to_buf(u32 bias[18][28], F_ACC bias_buf[18][28]) {
  54.     for (int i = 0; i < 18; ++i) {
  55. #pragma HLS PIPELINE
  56.         for (int j = 0; j < 28; ++j) {
  57. #pragma HLS UNROLL
  58.             bias_buf[i][j] = bias[i][j];
  59.         }
  60.     }
  61. }
  62.  
  63. void set_bias(F_ACC bias_buf[18][28], F_ACC ans_buf[18][28]) {
  64. #pragma HLS ARRAY_PARTITION variable=ans_buf complete dim=1
  65. #pragma HLS ARRAY_PARTITION variable=bias_buf complete dim=1
  66.     for (int i = 0; i < 18; ++i) {
  67. #pragma HLS PIPELINE
  68.         for (int j = 0; j < 28; ++j) {
  69. #pragma HLS UNROLL
  70.             ans_buf[i][j] = bias_buf[i][j];
  71.         }
  72.     }
  73. }
  74.  
  75. F_ACC compute_element(
  76.     F_DATA data0, F_WEIGHT weight0,
  77.     F_DATA data1, F_WEIGHT weight1,
  78.     F_DATA data2, F_WEIGHT weight2,
  79.     F_DATA data3, F_WEIGHT weight3,
  80.     F_DATA data4, F_WEIGHT weight4,
  81.     F_DATA data5, F_WEIGHT weight5,
  82.     F_DATA data6, F_WEIGHT weight6,
  83.     F_DATA data7, F_WEIGHT weight7,
  84.     F_DATA data8, F_WEIGHT weight8,
  85.     F_DATA data9, F_WEIGHT weight9,
  86.     F_DATA data10, F_WEIGHT weight10,
  87.     F_DATA data11, F_WEIGHT weight11,
  88.     F_DATA data12, F_WEIGHT weight12,
  89.     F_DATA data13, F_WEIGHT weight13,
  90.     F_DATA data14, F_WEIGHT weight14,
  91.     F_DATA data15, F_WEIGHT weight15,
  92.     F_DATA data16, F_WEIGHT weight16,
  93.     F_DATA data17, F_WEIGHT weight17,
  94.     F_DATA data18, F_WEIGHT weight18,
  95.     F_DATA data19, F_WEIGHT weight19,
  96.     F_DATA data20, F_WEIGHT weight20,
  97.     F_DATA data21, F_WEIGHT weight21,
  98.     F_DATA data22, F_WEIGHT weight22,
  99.     F_DATA data23, F_WEIGHT weight23,
  100.     F_DATA data24, F_WEIGHT weight24,
  101.     F_DATA data25, F_WEIGHT weight25,
  102.     F_DATA data26, F_WEIGHT weight26
  103. ) {
  104.     F_ACC ans = data0 * weight0 + data1 * weight1 + data2 * weight2 +
  105.      data3 * weight3 + data4 * weight4 + data5 * weight5 +
  106.      data6 * weight6 + data7 * weight7 + data8 * weight8 +
  107.      data9 * weight9 + data10 * weight10 + data11 * weight11 +
  108.      data12 * weight12 + data13 * weight13 + data14 * weight14 +
  109.      data15 * weight15 + data16 * weight16 + data17 * weight17 +
  110.      data18 * weight18 + data19 * weight19 + data20 * weight20 +
  111.      data21 * weight21 + data22 * weight22 + data23 * weight23 +
  112.      data24 * weight24 + data25 * weight25 + data26 * weight26;
  113.      return ans;
  114.  
  115.  
  116. }
  117.  
  118. void compute(F_DATA data_buf[3][20][30], F_WEIGHT weight_buf[3][3][3], F_ACC ans_buf[18][28]) {
  119. #pragma HLS ARRAY_PARTITION variable=data_buf complete dim=1
  120. #pragma HLS ARRAY_PARTITION variable=weight_buf complete dim=1
  121. #pragma HLS ARRAY_PARTITION variable=ans_buf complete dim=1
  122.     for (int i = 0; i < 18; ++i) {
  123. #pragma HLS PIPELINE
  124.         for (int j = 0; j < 28; ++j) {
  125. #pragma HLS UNROLL
  126.             ans_buf[i][j] += compute_element(
  127.                 data_buf[0][i+0][j+0] , weight_buf[0][0][0],
  128.                 data_buf[0][i+0][j+1] , weight_buf[0][0][1],
  129.                 data_buf[0][i+0][j+2] , weight_buf[0][0][2],
  130.                 data_buf[0][i+1][j+0] , weight_buf[0][1][0],
  131.                 data_buf[0][i+1][j+1] , weight_buf[0][1][1],
  132.                 data_buf[0][i+1][j+2] , weight_buf[0][1][2],
  133.                 data_buf[0][i+2][j+0] , weight_buf[0][2][0],
  134.                 data_buf[0][i+2][j+1] , weight_buf[0][2][1],
  135.                 data_buf[0][i+2][j+2] , weight_buf[0][2][2],
  136.                 data_buf[1][i+0][j+0] , weight_buf[1][0][0],
  137.                 data_buf[1][i+0][j+1] , weight_buf[1][0][1],
  138.                 data_buf[1][i+0][j+2] , weight_buf[1][0][2],
  139.                 data_buf[1][i+1][j+0] , weight_buf[1][1][0],
  140.                 data_buf[1][i+1][j+1] , weight_buf[1][1][1],
  141.                 data_buf[1][i+1][j+2] , weight_buf[1][1][2],
  142.                 data_buf[1][i+2][j+0] , weight_buf[1][2][0],
  143.                 data_buf[1][i+2][j+1] , weight_buf[1][2][1],
  144.                 data_buf[1][i+2][j+2] , weight_buf[1][2][2],
  145.                 data_buf[2][i+0][j+0] , weight_buf[2][0][0],
  146.                 data_buf[2][i+0][j+1] , weight_buf[2][0][1],
  147.                 data_buf[2][i+0][j+2] , weight_buf[2][0][2],
  148.                 data_buf[2][i+1][j+0] , weight_buf[2][1][0],
  149.                 data_buf[2][i+1][j+1] , weight_buf[2][1][1],
  150.                 data_buf[2][i+1][j+2] , weight_buf[2][1][2],
  151.                 data_buf[2][i+2][j+0] , weight_buf[2][2][0],
  152.                 data_buf[2][i+2][j+1] , weight_buf[2][2][1],
  153.                 data_buf[2][i+2][j+2] , weight_buf[2][2][2]
  154.             );
  155.         }
  156.     }
  157. }
  158.  
  159. void ans_to_ddr(F_ACC ans_buf[18][28], u32 ans[18][28]) {
  160.     for (int i = 0; i < 18; ++i) {
  161. #pragma HLS PIPELINE
  162.         for (int j = 0; j < 28; ++j) {
  163. #pragma HLS UNROLL
  164.             ans[i][j] = ans_buf[i][j];
  165.         }
  166.     }
  167. }
  168.  
  169.  
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
Public Pastes
Advertisement
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!