...to emulate if-then-else constructs.

...for efficiency. The iteration via get() isn't fast.

...as suggested in this thread on the Boost mailing list: http://lists.boost.org/Archives/boost/2016/10/231136.php

Moved from https://bitbucket.org/gentryx/libflatarray/issues/13/adopt-boost-directory-layout

What needs to be done first?

• directory layout #9
• comprehensive list of examples
• documentation: 5min how-to
• documentation: design rationale, comparison with existing vectorization libs

moved from https://bitbucket.org/gentryx/libflatarray/issues/15/submit-to-boost-boost-incubator

Hi,

Is it really necessary to keep this link inside src?
If library is header only maybe it's better to put all files under include/libflatarray/*.
It's more common layout.

Of course if you have good reasons, feel free to keep it as is.

I've gotten following error:

/home/evansl/dwnlds/llvm/3.8/prebuilt/clang+llvm-3.8.0-x86_64-linux-gnu-ubuntu-14.04/bin/clang++ -I../github/libflatarray/include -DHAVE_LIBFLATARRAY -O0 -std=c++14 -o /tmp/build/clangxx3_8_pkg/clang/struct_of_arrays/work/soa_compare.benchmark.optim0.exe soa_compare.benchmark.cpp
In file included from soa_compare.benchmark.cpp:51:
In file included from ../github/libflatarray/include/libflatarray/flat_array.hpp:20:
In file included from ../github/libflatarray/include/libflatarray/estimate_optimum_short_vec_type.hpp:11:
In file included from ../github/libflatarray/include/libflatarray/detail/streaming_short_vec_switch.hpp:11:
In file included from ../github/libflatarray/include/libflatarray/short_vec.hpp:203:
../github/libflatarray/include/libflatarray/detail/short_vec_scalar_double_1.hpp:13:10: fatal error:
'libflatarray/config.h' file not found

include <libflatarray/config.h>

     ^

Maybe you've not pushed it yet?

...like begin(), end(), or a range based for loop.

Moved from https://bitbucket.org/gentryx/libflatarray/issues/20/add-element-iteration-facilities-to-short

This is available from similar libraries (like Vc or Boost.Simd).

I think this library needs to be ported to ARM arch. As 0.2.0 failed to build from source.

https://kojipkgs.fedoraproject.org//work/tasks/4820/8904820/build.log

Thoughts?

soa_grid_tests fail in TestDefaultSizesFor1D2D3D when creating the 1D grid.

e.g. template<typename CARGO, std::size_t ARITY mask_type which then wraps a short_vec's internal mask

Moved from https://bitbucket.org/gentryx/libflatarray/issues/18/implement-short_vec

Moved from https://bitbucket.org/gentryx/libflatarray/issues/21/add-gathering-c-tor-to-short_vec

Reference: http://rrsd.com/blincubator.com/requirements_documentation/

Moved from https://bitbucket.org/gentryx/libflatarray/issues/14/add-basic-documentation

I notice there is a workaround:
https://github.com/gentryx/libflatarray/blob/master/include/libflatarray/aligned_allocator.hpp

    /**
     * Added due to compiling for Intel MIC with CPP14=TRUE
     * GCC Bug Report: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=51626
     */
    void construct(pointer p)
    {
        std::allocator<T>().construct(p, value_type());
    }

This compiler bug is marked as fixed: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=51626
Shall the workaround be removed? Is the compiler bug actually fixed?

• compilation with MSVC
• run tests

Reason: during initialization no c-tor is run and when copying one soa_grid to another a bitwise copy is performed.

Resolution: add tests for soa_grid and soa_array with non-trivial member types (e.g. std::map<int, std::vector >), then copy containers and read/write individual cells.

Hi,

i tried to use LibFlatArray with a class template which has more than one template parameter and it didn't compile. A simple example can be found here: https://gist.github.com/anonymous/b188cd1459c1b6d5ca6f

The error message is: error: macro "LIBFLATARRAY_REGISTER_SOA" passed 3 arguments, but takes just 2 LIBFLATARRAY_REGISTER_SOA(Cell<4, 1>, ((double)(sum))((double)(value)))

Since it works fine with just one template parameter, the issue might be related to the comma inside Cell<4, 1>.

Originally reported by @shifty91

Moved from https://bitbucket.org/gentryx/libflatarray/issues/16/libflatarray_register_soa-does-not-work

...to utilize float, float2, float4 (artity: WARP_SIZE * 4), double (arity WARP_SIZE), double2 (arity WARP_SIZE * 2) and corresponding load/store operations. needs benchmarks, obviously.

bonus points for using warp shuffle operations.

moved from https://bitbucket.org/gentryx/libflatarray/issues/12/add-short_vec-implementation-for-cuda