I am trying to use the execution policies in the standard algorithm library. However, when I try to compile I get the following error message
c:\mingw\lib\gcc\mingw32\9.2.0\include\c++\pstl\parallel_backend_tbb.h:19:10: fatal error: tbb/blocked_range.h: No such file or directory
After looking at various other related questions such as this or this, I understand that the execution library depends upon a software called tbb. Moreover in order to compile code which uses <execution> one has to manually link to tbb. My issue is precisely with how to download and link tbb to a script that uses <execution>.
I believe I have some serious gaps in my understanding in terms of how one downloads the correct files and then links to them. I will first make a list with my understanding of the linking process and then I will explain what I have tried to fix the issue. I have chosen this format so that it is faster for the one to answer my question to point at the issue at fault. I will attempt to keep this as concise as possible.
cpp files, where the former usually only contain the interface to the software and the later the implementationcpp files can be pre-compiled and grouped into a single library file#include the header/s in their script and also tell the compiler where the header files as well as the library file, are located-I for the headers and -L, -l for the library file-L provides the location of the library files, the -l specifies which libraries to useThe script I try to compile is:
#include <execution>
int main() {
std::execution::par;
return 0;
}
with
g++ script.cpp -o out -I C:(path to the headers) -L C:(path to the library) -l (name of library) -std=c++17
I should also mention I am trying to do this on Windows 10
I had a particularly hard time understanding where to find the header and library files for tbb.
In the Intel getting started with TBB webpage, this github repository is listed as "TBB being available at". As I am used to header-only libraries I thought everything would be in the include directory but no .dll files where there. It is now my understanding that I have to compile the DLLs myself for my specific system which makes sense. I followed the following process using cmake:
# Do our experiments in /tmp
cd /tmp
# Clone oneTBB repository
git clone https://github.com/oneapi-src/oneTBB.git
cd oneTBB
# Create binary directory for out-of-source build
mkdir build && cd build
# Configure: customize CMAKE_INSTALL_PREFIX and disable TBB_TEST to avoid tests build
cmake -DCMAKE_INSTALL_PREFIX=/tmp/my_installed_onetbb -DTBB_TEST=OFF ..
# Build
cmake --build
# Install
cmake --install .
# Well done! Your installed oneTBB is in /tmp/my_installed_onetbb
However at the cmake --build step, cmake does not accept the command but requests more options. One of them is the dir option, which for which I made another directory and supplied it but then the error message Error: could not load cache printed out.
In any case, some files had been created so I searched for the .dll file but could not find it.
I downloaded the Intel oneAPI Base Toolkit as is suggested here. After the installation at ../Program Files (x86)/Intel/oneAPI I found the specific tbb tool at C:\Program Files (x86)\Intel\oneAPI\tbb and I used this address for the -I and -L flags but initial error message persists.
I also copied the directory C:\Program Files (x86)\Intel\oneAPI\tbb\2021.6.0 to the local directory of the script so I could link with -flag tbb\2021.6.0 but no luck
Many thanks
Yes you were correct. By using the package mingw-w64-tbb. You can use -ltbb12 instead of -ltbb. As the library files are related to ltbb12.
For using -ltbb option, you should set the Intel oneAPI environment. It can be used by downloading Intel oneAPI Base Toolkit.You can set the environment by sourcing setvars.sh file using the below command.
source /opt/intel/oneapi/setvars.sh