Benchmarking F(2 ** 32) Arithmetics with CUDA Backend DEVICE=gpu make cuda && ./run running on Tesla V100-SXM2-16GB
hilbert matrix generation with F(2 ** 32) elements 👇
dimension total
128 x 128 1232 us
256 x 256 696 us
512 x 512 1647 us
1024 x 1024 5555 us
addition on F(2 ** 32) elements 👇
dimension iterations total avg
128 x 128 1024 40712 ns 0.00242662 ns
256 x 256 1024 19506 ns 0.000290662 ns
512 x 512 1024 18193 ns 6.77742e-05 ns
1024 x 1024 1024 27604 ns 2.57082e-05 ns
subtraction on F(2 ** 32) elements 👇
dimension iterations total avg
128 x 128 1024 24565 ns 0.00146419 ns
256 x 256 1024 18235 ns 0.000271723 ns
512 x 512 1024 17838 ns 6.64517e-05 ns
1024 x 1024 1024 27123 ns 2.52603e-05 ns
multiplication on F(2 ** 32) elements 👇
dimension iterations total avg
128 x 128 1024 24633 ns 0.00146824 ns
256 x 256 1024 18355 ns 0.000273511 ns
512 x 512 1024 19659 ns 7.32355e-05 ns
1024 x 1024 1024 28587 ns 2.66237e-05 ns
division on F(2 ** 32) elements 👇
dimension iterations total avg
128 x 128 1024 23520 ns 0.0014019 ns
256 x 256 1024 17529 ns 0.000261202 ns
512 x 512 1024 20052 ns 7.46995e-05 ns
1024 x 1024 1024 29490 ns 2.74647e-05 ns
inversion on F(2 ** 32) elements 👇
dimension iterations total avg
128 x 128 1024 26186 ns 0.00156081 ns
256 x 256 1024 17737 ns 0.000264302 ns
512 x 512 1024 19427 ns 7.23712e-05 ns
1024 x 1024 1024 30040 ns 2.79769e-05 ns
exponentiation on F(2 ** 32) elements 👇
dimension iterations total avg
128 x 128 1024 25294 ns 0.00150764 ns
256 x 256 1024 18120 ns 0.000270009 ns
512 x 512 1024 19715 ns 7.34441e-05 ns
1024 x 1024 1024 29630 ns 2.75951e-05 ns