C++ 17 Parallel Sum

[AlexCantor]

Following web site has small program to parallel sum elements of Vector:

<
http://en.cppreference.com/w/cpp/experimental/reduce
>

<

#include <iostream>
#include <chrono>
#include <vector>
#include <numeric>
// #include <experimental/execution_policy>
// #include <experimental/numeric>
#include <experimental/execution_policy>
#include <experimental/numeric>

int main()
{
std::vector<double> v(10'000'007, 0.5);

{
auto t1 = std::chrono::high_resolution_clock::now();
double result = std::accumulate(v.begin(), v.end(), 0.0);
auto t2 = std::chrono::high_resolution_clock::now();
std::chrono::duration<double, std::milli> ms = t2 - t1;
std::cout << std::fixed << "std::accumulate result " << result
<< " took " << ms.count() << " ms\n";
}

{
auto t1 = std::chrono::high_resolution_clock::now();
double result = std::experimental::parallel::reduce(
std::experimental::parallel::par,
v.begin(), v.end());
auto t2 = std::chrono::high_resolution_clock::now();
std::chrono::duration<double, std::milli> ms = t2 - t1;
std::cout << "parallel::reduce result "
<< result << " took " << ms.count() << " ms\n";
}
}

>

Above program output will be:

<
std::accumulate result 5000003.50000 took 12.7365 ms
parallel::reduce result 5000003.50000 took 5.06423 ms
>

I compiled this program using Microsoft Visual C++ 2017 Professional.

Compiler is giving following errors:

cannot open source file "experimental/execution_policy"

cannot open source file "experimental/numeric"

Any information to resolve these compilation errors please?

[Cubbi]

That program uses the Parallelism TS (Technical specification), which was a sort of a beta branch of the C++ standard, which had since been merged into C++17.

You can wait for when C++17 is implemented by your compiler, at which point this program will work, except you'd have to rewrite it as was done on http://en.cppreference.com/w/cpp/algorithm/reduce#Example

Or you can find one of the half a dozen implementations of the Parallelism TS, and read its documentation on how exactly to plug it in. As I recall, HPX had the best implementation.

[AlexCantor]

Cubbi,
Thank you very much.
Microsoft VC++ 2017 has not implemented parallelism? Correct?
When may be Microsoft planning to release next C++ version (2018?) and will they include parallel C++features in 2018?

I have gcc.exe and g++.exe. Both are of version 8.0.0 and can be run from DOS prompt.
Looks like g++ 8.0.0 has not yet implemented parallel C++? correct?

Is there a later gcc/g++ version that implements parallel C++ please?
If yes, which version of gcc/g++? And is there an easy way to install the gcc/g++ with parallel C++?

Thanks

[AlexCantor]

Cubbi,
Is HPX freely downloadable and usable software?
Is there a tutorial website, with HPX installation and running steps please?

[AlexCantor]

I have downloaded and extracted hpx_1.0.0. I did not see any HPX installation and usage instructions in the HPX folder.

May I know any information about HPX installation and usage please?

[AlexCantor]

Is there a webpage tutorial or instructions for plugging HPX into Visual C++ Studio 2017 Professional please?

[JLBorges]

HPX is integrated into VcPkg. See: http://stellar-group.org/2017/05/using-hpx-on-windows/

[Cubbi]

Microsoft VC++ 2017 has not implemented parallelism? Correct?

Microsoft's take on that technical specification was http://parallelstl.codeplex.com/

Looks like g++ 8.0.0 has not yet implemented parallel C++? correct?

GCC's take on parallelism was https://gcc.gnu.org/onlinedocs/libstdc++/manual/parallel_mode.html - that approach is older than the technical specification and is not compatible with it

Note I am using past tense. It doesn't matter who implemented Parallelism TS. Those features (after a number of modifications) were all merged into C++17, next compiler to support that revision of the standard will have std::reduce (not std::experimental::parallel::reduce)

[jonnin]

If you need to stick to visual studio, you can thread your own easily, and divide up the work then sum the results of that. Split it 4 ways maybe, for typical desktop cpus.

You might also see if cuda is an option for it, this sort of thing would do really well on a graphics card.

If the data changes slowly, you might be able to keep a running total in a real example. That is, say you get an initial chunk of 10k and add 10 here and 10 there, just adjust the running total by the 10 changes ...

[AlexCantor]

Is there a freely downloadable latest compiler such as g++ that compiles on Microsoft Windows 10 computer, all C++17 features that includes Parallelism TS (Technical specification) features please?

If yes, how to download and install (any website link please)?

[JLBorges]

> Is there a freely downloadable latest compiler such as g++ with all C++17 features

No.

Though the GNU compiler does make a spurious claim by defining __cplusplus as 201703L
For example, GCC 8.0 (snapshot): https://godbolt.org/g/F2mrna
(Uncomment the #include of the C++17 header to call the bluff)