JOSS paper comments
-
instructions to reproduce the benchmarks should exist somewhere -
is lapacke integrated with cmake? find_lapack(LAPACKE)
? I seem to have to include the headers directly in CXXFLAGS when trying to build with-DCPPDUALS_BENCHMARK
. -
the hardware on which the benchmarks were run should be given in the paper -
openblas version used in the benchmark should be given in the paper. In fact, does it have a doi? -
eigen version used in the benchmark should be given in the paper. In fact, does it have a doi? -
how are the benchmarks performed? E.g. which framework? Anything special to add, such as warming up, number of samples, etc? -
assuming there is no threading of dualcpp itself, were the benchmarks run with or without threading enabled, whether by eigen or by openblas -
unless I missed it, it is unclear from reading the docs what vectorization is available (SSE, AVX...), though it is present in the paper. Does the user have to do anything to enable anyone vectorization? Can the user explicitly enable or disable a vectorization? -
"The integration lets us achieve near optimum performance for forward first-order differentiation of matrix expressions". Optimal in what sense? How is it measured? -
the installation instructions are quite ad-hoc. Any chance cppduals could be installed as per any other CMake project? -
the installation instructions are really quite ad-hoc. Any chance something could be done, say a cmake file provided with an ExternalProject
setup to download and then find cppduals, or even integration with a package manager like conan? -
typo? "The template specializations map the dual arithmetic" -> "The template specializations [which map][mapping] the dual arithmetic" -
typo? "we can use Eigen cache and vector optimized [...]" -> "we can use Eigen's cache [...]"
Edited by Michael Tesch