Table 1
Performance comparison with other implementations at lmax = 2047, ncore = 1.
| Code | Version | Grid | Spin | nSHT | RAVX | RSSE2 | Rscalar |
|
|
|||||||
| shtns | 2.31 | Gauss-Legendre | 0 | 1 | 0.84 | 0.88 | 0.91 |
| Wavemoth (brute-force) | Nov. 2011 | HEALPix (Nside = 1024) | 0 | 1 | 1.63 | 0.98 | – |
| ′′ | ′′ | ′′ | 0 | 5 | 1.59 | 1.09 | – |
| Wavemoth (butterfly) | Nov. 2011 | HEALPix (Nside = 1024) | 0 | 5 | 0.96 | 0.66 | – |
| libpsht | Jan. 2011 | Gauss-Legendre | 0 | 1 | 4.06 | 2.30 | 2.30 |
| ′′ | ′′ | ′′ | 0 | 5 | 2.66 | 1.75 | 1.62 |
| ′′ | ′′ | ′′ | 2 | 1 | 2.50 | 1.48 | 1.20 |
| ′′ | ′′ | ′′ | 2 | 5 | 2.15 | 1.44 | 1.08 |
| spinsfast | r104 | ECP (Clenshaw-Curtis) | 0 | 1 | 57.04 | 32.12 | 15.31 |
| ′′ | ′′ | ′′ | 0 | 5 | 28.39 | 18.72 | 9.38 |
| ′′ | ′′ | ′′ | 2 | 1 | 16.99 | 10.20 | 4.73 |
| ′′ | ′′ | ′′ | 2 | 5 | 8.60 | 5.66 | 2.56 |
| SSHT | 1.0b1 | MW sampling theorem | 0 | 1 | 20.91 | 15.60 | 9.46 |
| ′′ | ′′ | ′′ | 2 | 1 | 13.40 | 9.29 | 4.99 |
| S2HAT | 2.55beta | HEALPix (Nside = 1024) | 0 | 1 | 12.33 | 7.33 | 3.60 |
| Glesp | 2 | Gauss-Legendre | 0 | 1 | 55.32 | 31.26 | 14.95 |
Notes.All tests had a band limit of lmax = 2047 and were carried out on a single core. The grids used by libsharp and the respective comparison code were identical in each run. RAVX denotes the quotient of wall-clock times for the respective code and libsharp in the presence of the AVX instruction set, RSSE2 is the quotient when SSE2 (but not AVX) is supported, and Rscalar was measured with both SSE2 and AVX disabled. The libsharp support for the MW sampling theorem used for the SSHT comparisons is experimental. For Wavemoth, butterfly matrix compression can optionally be enabled. In the benchmark given we requested an accuracy of 10-4, which led to an extra requirement of 4 GB of precomputed data in memory. Note that when running on a single core, Wavemoth is at an advantage compared to the normal situation where the memory bus is shared between multiple cores.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.