|
1 |
| -See pytorch PR https://github.com/pytorch/pytorch/pull/53104 |
2 |
| - |
3 |
| -before master commit 1.9.0a0+git68b6249 |
4 |
| - |
5 |
| -after commit 1.9.0a0+gitdce610d |
6 |
| - |
7 |
| -This page was tested on E5-2680 v3 and RTX 2070 Super. Libraries: intel-mkl 2020.2.254-1 and cuda 11.2 |
8 |
| - |
9 |
| -- before: magma |
10 |
| -- after: potrf + potrfBatched (when batch_size >= 2) **[this PR]** |
11 |
| - |
12 | 1 | time is in **us** (10^-6 s)
|
13 | 2 |
|
14 |
| -|shape|cpu|before_magma|after_potrf_and_batched| |
| 3 | +|shape|cpu|before_magma|after_potrf64bit_and_magma| |
15 | 4 | |---:|---:|---:|---:|
|
16 |
| -| [] 2 torch.float32 | 19.055 | 2280.221 | 35.030 | |
17 |
| -| [] 4 torch.float32 | 14.152 | 2303.458 | 35.046 | |
18 |
| -| [] 8 torch.float32 | 17.151 | 2326.320 | 35.175 | |
19 |
| -| [] 16 torch.float32 | 19.888 | 2358.930 | 35.133 | |
20 |
| -| [] 32 torch.float32 | 22.837 | 2341.613 | 35.418 | |
21 |
| -| [] 64 torch.float32 | 49.388 | 2383.232 | 42.917 | |
22 |
| -| [] 128 torch.float32 | 168.919 | 2431.619 | 82.173 | |
23 |
| -| [] 256 torch.float32 | 415.945 | 2884.719 | 171.727 | |
24 |
| -| [] 512 torch.float32 | 1567.114 | 3219.271 | 397.378 | |
25 |
| -| [] 1024 torch.float32 | 6100.324 | 3948.443 | 953.525 | |
26 |
| -| [] 2048 torch.float32 | 43855.028 | 6239.588 | 2755.491 | |
27 |
| -| [1] 2 torch.float32 | 17.567 | 55.177 | 35.341 | |
28 |
| -| [1] 4 torch.float32 | 17.678 | 55.194 | 35.539 | |
29 |
| -| [1] 8 torch.float32 | 18.053 | 56.533 | 35.596 | |
30 |
| -| [1] 16 torch.float32 | 19.512 | 62.417 | 35.927 | |
31 |
| -| [1] 32 torch.float32 | 22.422 | 74.961 | 35.769 | |
32 |
| -| [1] 64 torch.float32 | 54.413 | 102.557 | 42.919 | |
33 |
| -| [1] 128 torch.float32 | 169.238 | 170.707 | 70.264 | |
34 |
| -| [1] 256 torch.float32 | 432.289 | 374.914 | 146.682 | |
35 |
| -| [1] 512 torch.float32 | 1205.813 | 1147.444 | 337.876 | |
36 |
| -| [1] 1024 torch.float32 | 4601.706 | 4157.900 | 954.473 | |
37 |
| -| [1] 2048 torch.float32 | 28580.566 | 6390.159 | 2754.500 | |
38 |
| -| [2] 2 torch.float32 | 20.761 | 54.883 | 59.240 | |
39 |
| -| [2] 4 torch.float32 | 20.363 | 55.135 | 59.264 | |
40 |
| -| [2] 8 torch.float32 | 21.775 | 56.541 | 59.102 | |
41 |
| -| [2] 16 torch.float32 | 24.421 | 61.926 | 59.221 | |
42 |
| -| [2] 32 torch.float32 | 29.667 | 74.519 | 67.042 | |
43 |
| -| [2] 64 torch.float32 | 76.576 | 101.504 | 82.712 | |
44 |
| -| [2] 128 torch.float32 | 254.089 | 168.731 | 111.827 | |
45 |
| -| [2] 256 torch.float32 | 599.307 | 369.279 | 205.888 | |
46 |
| -| [2] 512 torch.float32 | 1637.425 | 1204.858 | 522.455 | |
47 |
| -| [2] 1024 torch.float32 | 5790.532 | 4448.121 | 1589.960 | |
48 |
| -| [2] 2048 torch.float32 | 40138.750 | 7395.678 | 5928.248 | |
49 |
| -| [4] 2 torch.float32 | 19.804 | 41.088 | 42.454 | |
50 |
| -| [4] 4 torch.float32 | 19.767 | 41.878 | 42.386 | |
51 |
| -| [4] 8 torch.float32 | 21.751 | 42.951 | 42.989 | |
52 |
| -| [4] 16 torch.float32 | 27.984 | 48.655 | 42.655 | |
53 |
| -| [4] 32 torch.float32 | 42.049 | 61.765 | 50.501 | |
54 |
| -| [4] 64 torch.float32 | 136.775 | 91.196 | 63.519 | |
55 |
| -| [4] 128 torch.float32 | 576.380 | 166.014 | 95.640 | |
56 |
| -| [4] 256 torch.float32 | 848.511 | 385.509 | 221.888 | |
57 |
| -| [4] 512 torch.float32 | 2480.692 | 1304.622 | 613.503 | |
58 |
| -| [4] 1024 torch.float32 | 7707.767 | 4886.211 | 2107.344 | |
59 |
| -| [4] 2048 torch.float32 | 55563.672 | 9442.459 | 8907.055 | |
60 |
| -| [8] 2 torch.float32 | 18.787 | 41.769 | 42.939 | |
61 |
| -| [8] 4 torch.float32 | 18.123 | 43.375 | 42.443 | |
62 |
| -| [8] 8 torch.float32 | 21.067 | 44.121 | 42.948 | |
63 |
| -| [8] 16 torch.float32 | 29.974 | 50.869 | 42.498 | |
64 |
| -| [8] 32 torch.float32 | 61.718 | 64.079 | 51.111 | |
65 |
| -| [8] 64 torch.float32 | 174.167 | 96.937 | 63.341 | |
66 |
| -| [8] 128 torch.float32 | 745.488 | 176.674 | 103.029 | |
67 |
| -| [8] 256 torch.float32 | 1331.958 | 432.471 | 258.526 | |
68 |
| -| [8] 512 torch.float32 | 4076.162 | 1462.709 | 790.236 | |
69 |
| -| [8] 1024 torch.float32 | 17392.765 | 5786.699 | 3151.025 | |
70 |
| -| [16] 2 torch.float32 | 20.866 | 42.434 | 43.010 | |
71 |
| -| [16] 4 torch.float32 | 21.590 | 43.252 | 42.318 | |
72 |
| -| [16] 8 torch.float32 | 27.167 | 44.373 | 42.324 | |
73 |
| -| [16] 16 torch.float32 | 49.028 | 51.100 | 42.846 | |
74 |
| -| [16] 32 torch.float32 | 116.404 | 64.965 | 49.931 | |
75 |
| -| [16] 64 torch.float32 | 388.556 | 98.290 | 63.181 | |
76 |
| -| [16] 128 torch.float32 | 1330.891 | 185.933 | 117.122 | |
77 |
| -| [16] 256 torch.float32 | 2376.293 | 492.626 | 335.315 | |
78 |
| -| [16] 512 torch.float32 | 7307.411 | 1754.295 | 1171.981 | |
79 |
| -| [32] 2 torch.float32 | 22.934 | 42.380 | 42.502 | |
80 |
| -| [32] 4 torch.float32 | 23.678 | 43.250 | 42.658 | |
81 |
| -| [32] 8 torch.float32 | 31.927 | 44.699 | 42.715 | |
82 |
| -| [32] 16 torch.float32 | 94.794 | 50.902 | 43.057 | |
83 |
| -| [32] 32 torch.float32 | 135.281 | 65.271 | 50.708 | |
84 |
| -| [32] 64 torch.float32 | 529.833 | 99.046 | 62.631 | |
85 |
| -| [32] 128 torch.float32 | 2911.740 | 214.874 | 153.772 | |
86 |
| -| [32] 256 torch.float32 | 4753.231 | 610.045 | 518.030 | |
87 |
| -| [64] 2 torch.float32 | 29.902 | 42.636 | 43.020 | |
88 |
| -| [64] 4 torch.float32 | 35.429 | 43.522 | 42.179 | |
89 |
| -| [64] 8 torch.float32 | 49.895 | 44.746 | 43.572 | |
90 |
| -| [64] 16 torch.float32 | 135.721 | 50.989 | 42.832 | |
91 |
| -| [64] 32 torch.float32 | 224.029 | 65.572 | 49.696 | |
92 |
| -| [64] 64 torch.float32 | 961.395 | 104.305 | 76.027 | |
93 |
| -| [64] 128 torch.float32 | 5609.256 | 271.441 | 225.720 | |
94 |
| -| [128] 2 torch.float32 | 52.542 | 42.892 | 42.747 | |
95 |
| -| [128] 4 torch.float32 | 62.965 | 43.840 | 43.114 | |
96 |
| -| [128] 8 torch.float32 | 96.781 | 44.832 | 42.930 | |
97 |
| -| [128] 16 torch.float32 | 222.146 | 51.180 | 42.766 | |
98 |
| -| [128] 32 torch.float32 | 376.536 | 66.161 | 49.907 | |
99 |
| -| [128] 64 torch.float32 | 1880.561 | 133.194 | 114.067 | |
100 |
| -| [256] 2 torch.float32 | 66.206 | 44.754 | 42.242 | |
101 |
| -| [256] 4 torch.float32 | 81.555 | 45.447 | 42.391 | |
102 |
| -| [256] 8 torch.float32 | 147.877 | 47.612 | 42.688 | |
103 |
| -| [256] 16 torch.float32 | 336.137 | 54.235 | 42.174 | |
104 |
| -| [256] 32 torch.float32 | 634.189 | 74.383 | 64.845 | |
105 |
| -| [512] 2 torch.float32 | 112.168 | 46.233 | 42.259 | |
106 |
| -| [512] 4 torch.float32 | 140.019 | 47.107 | 42.090 | |
107 |
| -| [512] 8 torch.float32 | 268.559 | 49.280 | 42.298 | |
108 |
| -| [512] 16 torch.float32 | 573.994 | 57.851 | 42.053 | |
109 |
| -| [1024] 2 torch.float32 | 203.898 | 48.189 | 41.716 | |
110 |
| -| [1024] 4 torch.float32 | 258.988 | 48.978 | 41.807 | |
111 |
| -| [1024] 8 torch.float32 | 438.113 | 53.205 | 41.791 | |
| 5 | +| [] 2 torch.float32 | 66.774 | 2280.221 | 58.959 | |
| 6 | +| [] 4 torch.float32 | 14.175 | 2303.458 | 59.223 | |
| 7 | +| [] 8 torch.float32 | 18.633 | 2326.320 | 59.109 | |
| 8 | +| [] 16 torch.float32 | 20.404 | 2358.930 | 59.451 | |
| 9 | +| [] 32 torch.float32 | 23.551 | 2341.613 | 59.565 | |
| 10 | +| [] 64 torch.float32 | 50.053 | 2383.232 | 83.184 | |
| 11 | +| [] 128 torch.float32 | 173.814 | 2431.619 | 120.336 | |
| 12 | +| [] 256 torch.float32 | 437.671 | 2884.719 | 208.042 | |
| 13 | +| [] 512 torch.float32 | 1672.962 | 3219.271 | 400.172 | |
| 14 | +| [] 1024 torch.float32 | 6052.655 | 3948.443 | 972.376 | |
| 15 | +| [] 2048 torch.float32 | 41786.490 | 6239.588 | 2771.738 | |
| 16 | +| [1] 2 torch.float32 | 16.018 | 55.177 | 58.924 | |
| 17 | +| [1] 4 torch.float32 | 15.890 | 55.194 | 64.280 | |
| 18 | +| [1] 8 torch.float32 | 16.391 | 56.533 | 59.009 | |
| 19 | +| [1] 16 torch.float32 | 17.996 | 62.417 | 59.372 | |
| 20 | +| [1] 32 torch.float32 | 21.377 | 74.961 | 59.963 | |
| 21 | +| [1] 64 torch.float32 | 54.598 | 102.557 | 75.136 | |
| 22 | +| [1] 128 torch.float32 | 153.839 | 170.707 | 112.467 | |
| 23 | +| [1] 256 torch.float32 | 411.058 | 374.914 | 186.747 | |
| 24 | +| [1] 512 torch.float32 | 1176.441 | 1147.444 | 359.730 | |
| 25 | +| [1] 1024 torch.float32 | 4504.277 | 4157.900 | 972.515 | |
| 26 | +| [1] 2048 torch.float32 | 28452.310 | 6390.159 | 2773.273 | |
| 27 | +| [2] 2 torch.float32 | 17.597 | 54.883 | 39.668 | |
| 28 | +| [2] 4 torch.float32 | 18.118 | 55.135 | 40.283 | |
| 29 | +| [2] 8 torch.float32 | 19.303 | 56.541 | 41.505 | |
| 30 | +| [2] 16 torch.float32 | 22.752 | 61.926 | 47.370 | |
| 31 | +| [2] 32 torch.float32 | 28.764 | 74.519 | 59.570 | |
| 32 | +| [2] 64 torch.float32 | 71.314 | 101.504 | 86.426 | |
| 33 | +| [2] 128 torch.float32 | 292.647 | 168.731 | 154.935 | |
| 34 | +| [2] 256 torch.float32 | 552.208 | 369.279 | 354.285 | |
| 35 | +| [2] 512 torch.float32 | 1600.543 | 1204.858 | 1201.538 | |
| 36 | +| [2] 1024 torch.float32 | 5972.632 | 4448.121 | 4521.013 | |
| 37 | +| [2] 2048 torch.float32 | 39196.011 | 7395.678 | 7465.609 | |
| 38 | +| [4] 2 torch.float32 | 18.258 | 41.088 | 40.741 | |
| 39 | +| [4] 4 torch.float32 | 18.169 | 41.878 | 40.869 | |
| 40 | +| [4] 8 torch.float32 | 20.840 | 42.951 | 43.534 | |
| 41 | +| [4] 16 torch.float32 | 27.483 | 48.655 | 48.096 | |
| 42 | +| [4] 32 torch.float32 | 41.822 | 61.765 | 60.854 | |
| 43 | +| [4] 64 torch.float32 | 138.791 | 91.196 | 91.095 | |
| 44 | +| [4] 128 torch.float32 | 574.721 | 166.014 | 165.793 | |
| 45 | +| [4] 256 torch.float32 | 757.351 | 385.509 | 384.977 | |
| 46 | +| [4] 512 torch.float32 | 2342.804 | 1304.622 | 1304.067 | |
| 47 | +| [4] 1024 torch.float32 | 9071.045 | 4886.211 | 4929.194 | |
| 48 | +| [4] 2048 torch.float32 | 57826.840 | 9442.459 | 9631.141 | |
| 49 | +| [8] 2 torch.float32 | 17.141 | 41.769 | 42.027 | |
| 50 | +| [8] 4 torch.float32 | 16.745 | 43.375 | 41.145 | |
| 51 | +| [8] 8 torch.float32 | 19.215 | 44.121 | 42.852 | |
| 52 | +| [8] 16 torch.float32 | 27.425 | 50.869 | 48.796 | |
| 53 | +| [8] 32 torch.float32 | 46.709 | 64.079 | 62.668 | |
| 54 | +| [8] 64 torch.float32 | 174.225 | 96.937 | 95.206 | |
| 55 | +| [8] 128 torch.float32 | 762.877 | 176.674 | 175.077 | |
| 56 | +| [8] 256 torch.float32 | 1386.097 | 432.471 | 429.544 | |
| 57 | +| [8] 512 torch.float32 | 4195.243 | 1462.709 | 1457.487 | |
| 58 | +| [8] 1024 torch.float32 | 19305.512 | 5786.699 | 5868.911 | |
| 59 | +| [16] 2 torch.float32 | 19.890 | 42.434 | 45.466 | |
| 60 | +| [16] 4 torch.float32 | 20.660 | 43.252 | 43.327 | |
| 61 | +| [16] 8 torch.float32 | 26.371 | 44.373 | 46.226 | |
| 62 | +| [16] 16 torch.float32 | 48.751 | 51.100 | 50.815 | |
| 63 | +| [16] 32 torch.float32 | 94.477 | 64.965 | 63.677 | |
| 64 | +| [16] 64 torch.float32 | 377.310 | 98.290 | 96.308 | |
| 65 | +| [16] 128 torch.float32 | 1474.216 | 185.933 | 184.481 | |
| 66 | +| [16] 256 torch.float32 | 2405.352 | 492.626 | 490.029 | |
| 67 | +| [16] 512 torch.float32 | 7828.434 | 1754.295 | 1753.410 | |
| 68 | +| [32] 2 torch.float32 | 21.078 | 42.380 | 40.783 | |
| 69 | +| [32] 4 torch.float32 | 22.185 | 43.250 | 42.487 | |
| 70 | +| [32] 8 torch.float32 | 31.124 | 44.699 | 43.080 | |
| 71 | +| [32] 16 torch.float32 | 87.727 | 50.902 | 49.999 | |
| 72 | +| [32] 32 torch.float32 | 131.084 | 65.271 | 64.570 | |
| 73 | +| [32] 64 torch.float32 | 539.154 | 99.046 | 97.589 | |
| 74 | +| [32] 128 torch.float32 | 2753.116 | 214.874 | 215.273 | |
| 75 | +| [32] 256 torch.float32 | 4813.017 | 610.045 | 610.006 | |
| 76 | +| [64] 2 torch.float32 | 27.769 | 42.636 | 40.654 | |
| 77 | +| [64] 4 torch.float32 | 29.520 | 43.522 | 41.364 | |
| 78 | +| [64] 8 torch.float32 | 47.020 | 44.746 | 43.315 | |
| 79 | +| [64] 16 torch.float32 | 132.615 | 50.989 | 49.638 | |
| 80 | +| [64] 32 torch.float32 | 217.310 | 65.572 | 63.921 | |
| 81 | +| [64] 64 torch.float32 | 986.923 | 104.305 | 104.182 | |
| 82 | +| [64] 128 torch.float32 | 5390.051 | 271.441 | 271.572 | |
| 83 | +| [128] 2 torch.float32 | 50.669 | 42.892 | 42.474 | |
| 84 | +| [128] 4 torch.float32 | 60.421 | 43.840 | 41.955 | |
| 85 | +| [128] 8 torch.float32 | 96.448 | 44.832 | 42.927 | |
| 86 | +| [128] 16 torch.float32 | 219.459 | 51.180 | 50.108 | |
| 87 | +| [128] 32 torch.float32 | 364.438 | 66.161 | 64.927 | |
| 88 | +| [128] 64 torch.float32 | 1903.152 | 133.194 | 133.470 | |
| 89 | +| [256] 2 torch.float32 | 67.254 | 44.754 | 42.825 | |
| 90 | +| [256] 4 torch.float32 | 80.547 | 45.447 | 43.048 | |
| 91 | +| [256] 8 torch.float32 | 146.505 | 47.612 | 46.427 | |
| 92 | +| [256] 16 torch.float32 | 332.615 | 54.235 | 51.697 | |
| 93 | +| [256] 32 torch.float32 | 660.817 | 74.383 | 73.969 | |
| 94 | +| [512] 2 torch.float32 | 118.439 | 46.233 | 43.343 | |
| 95 | +| [512] 4 torch.float32 | 144.591 | 47.107 | 45.033 | |
| 96 | +| [512] 8 torch.float32 | 267.354 | 49.280 | 47.363 | |
| 97 | +| [512] 16 torch.float32 | 581.125 | 57.851 | 56.217 | |
| 98 | +| [1024] 2 torch.float32 | 218.353 | 48.189 | 45.818 | |
| 99 | +| [1024] 4 torch.float32 | 276.394 | 48.978 | 49.839 | |
| 100 | +| [1024] 8 torch.float32 | 451.058 | 53.205 | 50.737 | |
0 commit comments