A Zip iterator which enables iterating over multiple iterable containers of the same size. Performance of the iterator was a major concern; tests indicate it's as good as manually iterating over all of the containers simultaneously with index iteration. Note that the zip iterator defaults to assuming all of the iterable containers are of the Random Access type, this can be changed by specifying the type as the first template parameter. This requires C++17, and should be compatible with CUDA code
Basic usage:
#include "zip.hpp"
std::vector<std::array<double, 3>> pos, vel;
// ...
for(auto [p, v] : zip::make_zip(pos, vel)) {
// ...
for(auto [p_x, v_x] : zip::make_zip(p, v)) {
p_x += v_x * dt;
}
}
// With the iterator tag specified
for(auto [pos, vel] : zip::make_zip<std::random_access_iterator_tag>(pos, vel)) {
// ...
}
auto z = zip::make_zip(pos, vel);
std::sort(z.begin(), z.end(),
[](const std::tuple<std::array<double, 3> &, std::array<double, 3> &> &lhs,
const std::tuple<std::array<double, 3> &, std::array<double, 3> &> &rhs) {
return std::get<0>(lhs)[0] < std::get<0>(rhs);
});- Processor:
Intel(R) Core(TM) i7-6700K CPU @ 4.00GHz, 8192 KB cache, 4 cores, 8 hyperthreads - Compiler:
g++ (GCC) 10.2.1 20200723 (Red Hat 10.2.1-1) - Compilation Flags:
-march=native -fstrict-aliasing -ffast-math -O3 - Command line:
./performance --benchmark_repetitions=30 --benchmark_report_aggregates_only=true
------------------------------------------------------------------------------
Benchmark Time CPU Iterations
------------------------------------------------------------------------------
BM_Small_PosVel_Index_mean 111 ns 111 ns 30
BM_Small_PosVel_Index_median 115 ns 115 ns 30
BM_Small_PosVel_Index_stddev 9.16 ns 9.10 ns 30
BM_Small_PosVel_Zip_mean 103 ns 103 ns 30
BM_Small_PosVel_Zip_median 101 ns 101 ns 30
BM_Small_PosVel_Zip_stddev 8.29 ns 8.27 ns 30
BM_Large_PosVel_Index_mean 237347 ns 236161 ns 30
BM_Large_PosVel_Index_median 238537 ns 237313 ns 30
BM_Large_PosVel_Index_stddev 16817 ns 16710 ns 30
BM_Large_PosVel_Zip_mean 239222 ns 238073 ns 30
BM_Large_PosVel_Zip_median 237924 ns 236708 ns 30
BM_Large_PosVel_Zip_stddev 19898 ns 19750 ns 30
BM_1D_Small_Index_Iterate_mean 22.4 ns 22.4 ns 30
BM_1D_Small_Index_Iterate_median 22.7 ns 22.6 ns 30
BM_1D_Small_Index_Iterate_stddev 3.73 ns 3.72 ns 30
BM_1D_Small_Zip_Iterate_mean 21.9 ns 21.8 ns 30
BM_1D_Small_Zip_Iterate_median 21.9 ns 21.8 ns 30
BM_1D_Small_Zip_Iterate_stddev 3.35 ns 3.34 ns 30
BM_1D_Small_Index_Initialize_mean 39.6 ns 39.5 ns 30
BM_1D_Small_Index_Initialize_median 40.5 ns 40.4 ns 30
BM_1D_Small_Index_Initialize_stddev 3.16 ns 3.15 ns 30
BM_1D_Small_Zip_Initialize_mean 44.1 ns 44.0 ns 30
BM_1D_Small_Zip_Initialize_median 45.1 ns 44.9 ns 30
BM_1D_Small_Zip_Initialize_stddev 4.43 ns 4.41 ns 30
* BM_2D_Small_Index_Iterate_mean 13.6 ns 13.6 ns 30
BM_2D_Small_Index_Iterate_median 13.6 ns 13.6 ns 30
BM_2D_Small_Index_Iterate_stddev 2.05 ns 2.04 ns 30
BM_2D_Small_Zip_Iterate_mean 59.8 ns 59.7 ns 30
BM_2D_Small_Zip_Iterate_median 61.3 ns 61.1 ns 30
BM_2D_Small_Zip_Iterate_stddev 6.23 ns 6.22 ns 30
BM_2D_Small_Index_Initialize_mean 119 ns 119 ns 30
BM_2D_Small_Index_Initialize_median 119 ns 119 ns 30
BM_2D_Small_Index_Initialize_stddev 10.1 ns 10.1 ns 30
BM_2D_Small_Zip_Initialize_mean 112 ns 111 ns 30
BM_2D_Small_Zip_Initialize_median 111 ns 111 ns 30
BM_2D_Small_Zip_Initialize_stddev 7.90 ns 7.89 ns 30
BM_1D_Large_Index_Iterate_mean 100420 ns 99985 ns 30
BM_1D_Large_Index_Iterate_median 100691 ns 100279 ns 30
BM_1D_Large_Index_Iterate_stddev 4935 ns 4905 ns 30
BM_1D_Large_Zip_Iterate_mean 103519 ns 103127 ns 30
BM_1D_Large_Zip_Iterate_median 103785 ns 103359 ns 30
BM_1D_Large_Zip_Iterate_stddev 2029 ns 2017 ns 30
BM_1D_Large_Index_Initialize_mean 147467 ns 147023 ns 30
BM_1D_Large_Index_Initialize_median 147093 ns 146600 ns 30
BM_1D_Large_Index_Initialize_stddev 6346 ns 6315 ns 30
BM_1D_Large_Zip_Initialize_mean 143974 ns 143518 ns 30
BM_1D_Large_Zip_Initialize_median 144939 ns 144486 ns 30
BM_1D_Large_Zip_Initialize_stddev 6778 ns 6750 ns 30
* BM_2D_Large_Index_Iterate_mean 27348 ns 27280 ns 30
BM_2D_Large_Index_Iterate_median 30076 ns 30004 ns 30
BM_2D_Large_Index_Iterate_stddev 5369 ns 5358 ns 30
BM_2D_Large_Zip_Iterate_mean 305916 ns 303949 ns 30
BM_2D_Large_Zip_Iterate_median 309785 ns 307707 ns 30
BM_2D_Large_Zip_Iterate_stddev 18784 ns 18568 ns 30
BM_2D_Large_Index_Initialize_mean 545104 ns 541892 ns 30
BM_2D_Large_Index_Initialize_median 546444 ns 543156 ns 30
BM_2D_Large_Index_Initialize_stddev 13655 ns 13447 ns 30
BM_2D_Large_Zip_Initialize_mean 552160 ns 549212 ns 30
BM_2D_Large_Zip_Initialize_median 553067 ns 550062 ns 30
BM_2D_Large_Zip_Initialize_stddev 12430 ns 12214 ns 30
*The BM_2D_*_Index_Iterate benchmark is almost certainly wrong and needs to be fixed
- Support taking iterator pairs {begin, end} in the constructor
- Find out how to avoid implementing std::swap(std::tuple<_Elements...> &&, std::tuple<_Elements...> &&);
- Add checks for attempts to dereference past the end of one of the iterator limits
- Understand and fix the BM_2D_Large_Index_Iterate issue
- Add more performance tests