Add a CUDA viewcopy benchmark

[bernhardmgruber]

This example only shows variations of naive, layout oblivious copy kernels between different data layouts. No specialized layout aware copy routine is implemented.

[bernhardmgruber]

Dataset size:    1.5GB (x2)
GMemory size:    6.2GB
Max bandwidth: 313.0GiB/s
SMs: 30
src        -> dst        alg                ms      GiB/s hash
byte[]     -> byte[]     cudaMemcpy      9.930    275.374   OK
A AoS      -> SoA MB     naive          10.362    263.874   OK
A AoS      -> SoA MB     naive GS1D     13.594    201.142   OK
SoA MB     -> A AoS      naive          14.575    187.610   OK
SoA MB     -> A AoS      naive GS1D     17.755    154.007   OK
SoA MB     -> AoSoA32    naive          10.401    262.894   OK
SoA MB     -> AoSoA32    naive GS1D     13.684    199.822   OK
AoSoA32    -> SoA MB     naive          10.240    267.021   OK
AoSoA32    -> SoA MB     naive GS1D     14.054    194.559   OK
AoSoA8     -> AoSoA32    naive          10.508    260.210   OK
AoSoA8     -> AoSoA32    naive GS1D     13.841    197.561   OK
AoSoA32    -> AoSoA8     naive          10.261    266.482   OK
AoSoA32    -> AoSoA8     naive GS1D     13.597    201.105   OK
AoSoA8     -> AoSoA64    naive          10.288    265.782   OK
AoSoA8     -> AoSoA64    naive GS1D     13.678    199.904   OK
AoSoA64    -> AoSoA8     naive          10.252    266.726   OK
AoSoA64    -> AoSoA8     naive GS1D     13.763    198.683   OK

[codecov]

Codecov Report

Merging #721 (b658a1b) into develop (829b57d) will decrease coverage by 0.01%.
The diff coverage is n/a.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #721      +/-   ##
===========================================
- Coverage    98.79%   98.79%   -0.01%     
===========================================
  Files           75       75              
  Lines         7283     7282       -1     
===========================================
- Hits          7195     7194       -1     
  Misses          88       88              

[bernhardmgruber]

V100:

Dataset size:    7.3GB (x2)
GMemory size:   34.1GB
Max bandwidth: 836.4GiB/s
SMs: 80
Max threads per SM: 2048

src        -> dst        alg                 ms      GiB/s hash
byte[]     -> byte[]     cudaMemcpy      18.805    727.044   OK
A AoS      -> SoA MB     naive 1D        25.266    541.110   OK
A AoS      -> SoA MB     naive 3D        27.566    495.960   OK
A AoS      -> SoA MB     naive GS 1D     29.514    463.232   OK
A AoS      -> SoA MB     naive GS 3D     93.297    146.541   OK
SoA MB     -> A AoS      naive 1D        47.524    287.683   OK
SoA MB     -> A AoS      naive 3D        52.509    260.371   OK
SoA MB     -> A AoS      naive GS 1D     76.178    179.472   OK
SoA MB     -> A AoS      naive GS 3D     83.028    164.665   OK
SoA MB     -> AoSoA32    naive 1D        19.911    686.659   OK
SoA MB     -> AoSoA32    naive 3D        22.057    619.854   OK
SoA MB     -> AoSoA32    naive GS 1D     28.440    480.728   OK
SoA MB     -> AoSoA32    naive GS 3D     73.554    185.875   OK
AoSoA32    -> SoA MB     naive 1D        20.664    661.638   OK
AoSoA32    -> SoA MB     naive 3D        22.065    619.607   OK
AoSoA32    -> SoA MB     naive GS 1D     30.250    451.959   OK
AoSoA32    -> SoA MB     naive GS 3D    105.816    129.205   OK
AoSoA8     -> AoSoA32    naive 1D        20.525    666.101   OK
AoSoA8     -> AoSoA32    naive 3D        22.680    602.809   OK
AoSoA8     -> AoSoA32    naive GS 1D     33.616    406.702   OK
AoSoA8     -> AoSoA32    naive GS 3D     94.332    144.933   OK
AoSoA32    -> AoSoA8     naive 1D        20.290    673.835   OK
AoSoA32    -> AoSoA8     naive 3D        25.422    537.788   OK
AoSoA32    -> AoSoA8     naive GS 1D     45.214    302.379   OK
AoSoA32    -> AoSoA8     naive GS 3D     55.649    245.681   OK
AoSoA8     -> AoSoA64    naive 1D        20.603    663.594   OK
AoSoA8     -> AoSoA64    naive 3D        22.759    600.733   OK
AoSoA8     -> AoSoA64    naive GS 1D     33.637    406.453   OK
AoSoA8     -> AoSoA64    naive GS 3D     97.774    139.832   OK
AoSoA64    -> AoSoA8     naive 1D        20.498    666.981   OK
AoSoA64    -> AoSoA8     naive 3D        24.875    549.616   OK
AoSoA64    -> AoSoA8     naive GS 1D     44.755    305.483   OK
AoSoA64    -> AoSoA8     naive GS 3D     54.623    250.296   OK

[bernhardmgruber]

Nice, nvcc 11.3 gets an ICE ..

[bernhardmgruber]

Nice, nvcc 11.3 gets an ICE ..

Dropped: #766