Skip to main content

Table 1 Average time for each individual resolution step implemented as a convolution function of the band-limited kernel

From: Accelerating the computation for real-time application of the sinc function using graphics processing units

Matrix size

CPUref

CPUopt

GPU

64 × 64

0.173 ± 0.007

0.013 ± 0.001

0.295 ± 0.010

128 × 128

3.195 ± 0.164

0.107 ± 0.006

0.317 ± 0.009

192 × 192

15.733 ± 0.169

0.389 ± 0.007

0.385 ± 0.010

256 × 256

49.451 ± 1.050

1.133 ± 0.029

0.522 ± 0.023

320 × 320

116.072 ± 1.059

2.366 ± 0.016

0.749 ± 0.013

384 × 384

236.919 ± 3.379

4.811 ± 0.055

1.124 ± 0.017

448 × 448

432.670 ± 6.107

7.596 ± 0.018

1.750 ± 0.017

512 × 512

727.586 ± 8.031

12.479 ± 0.076

2.693 ± 0.023

  1. The reconstruction execution time is given in seconds. CPUref = matrix, matrix operation without optimization; CPUopt = matrix, matrix operation with BLAS optimization