The following code from ConstantIteratorTest should use bolt::cl::device_vector, but uses std::vector insted

TYPED_TEST_P( CountingIterator, DeviceTransformVector )

// initialize the data vector to be sequential numbers
std::vector< TypeParam > devVec( 3 );
bolt::cl::transform( devVec.begin( ), devVec.end( ), bolt::cl::make_counting_iterator( 42 ), devVec.begin( ),
bolt::cl::plus< TypeParam >( ) );
EXPECT_EQ( 42, devVec[ 0 ] );
EXPECT_EQ( 43, devVec[ 1 ] );
EXPECT_EQ( 44, devVec[ 2 ] );
}

why not linux?

The download URL for gtest in superbuild/ExternalGtest.cmake is not working. It begins with "https://...", after deleting the "s" character, it works fine.

Bolt 1.2, file include/bolt/cl/detail/transform_reduce.inl, lines 446-447:

dblog->CodePathTaken(BOLTLOG::BOLT_TRANSFORMREDUCE,BOLTLOG::BOLT_MULTICORE_CPU,"

::Transform_Reduce::MULTICORE_CPU");

Clearly the string markers (") are located in two lines, which makes the compiler issue unnecessary warnings

Z Koza

This issue can be observed only with the bolt i,e, when calling bolt::min_element and bolt::max_element on device_vector. For std::min_element and std::max_element it is working fine.

CODE:

//code for BOLT_MIN_ELEMENT:

TEST(sanity_min_element_2bolt_cl_device_vect_loop, ints_loop){
int size = 10;
bolt::cl::device_vector intStdVect (size);
bolt::cl::device_vector intBoltVect (size);

for (int i = 0 ; i < size; i++){
    intBoltVect[i] = (int)std::rand()%65535 ;
     intStdVect[i] = intBoltVect[i];
}

bolt::cl::device_vector<int>::iterator std_min_ele; 
for (int i = 0 ; i < 1000; i++){
            std_min_ele = std::min_element (intStdVect.begin(), intStdVect.end());
}

bolt::cl::device_vector<int>::iterator bolt_min_ele;  
for (int i = 0 ; i < 1000; i++){
    bolt_min_ele =  bolt::cl::min_element(intBoltVect.begin(), intBoltVect.end());
}
EXPECT_EQ(*std_min_ele, *bolt_min_ele)<<std::endl;

}

// code for BOLT_MAX_ELEMENT:

TEST(sanity_max_element_2bolt_cl_device_vect_loop, ints_loop){
int size = 10;
bolt::cl::device_vector intStdVect (size);
bolt::cl::device_vector intBoltVect (size);

for (int i = 0 ; i < size; i++){
    intBoltVect[i] = (int)std::rand()%65535 ;
     intStdVect[i] = intBoltVect[i];
}

bolt::cl::device_vector<int>::iterator std_min_ele; 
for (int i = 0 ; i < 1000; i++){
            std_max_ele = std::max_element (intStdVect.begin(), intStdVect.end());
}

bolt::cl::device_vector<int>::iterator bolt_min_ele;  
for (int i = 0 ; i < 1000; i++){
    bolt_max_ele =  bolt::cl::max_element(intBoltVect.begin(), intBoltVect.end());
}
EXPECT_EQ(*std_max_ele, *bolt_max_ele)<<std::endl;

}

boost dependency is not needed and creates some compiler issues with different versions of boost out there.

You should use C++11 directly. And please update CMakefiles to support C++11 flags for gcc.

After downloading the binary tarball for Linux and unpacking it, I find a directory with some stuff in it, but no instructions on installation. For example, should one copy include, lib and lib64 to /usr/local? Or is it intended that Bolt applications should set up -I and -L compiler flags to wherever one unpacked Bolt? Is any special care needed if one already has Boost (and no doubt a different version of Boost) installed to avoid version conflicts?

When we try to build bolt samples of AMD APP SDK by linking to the bolt 1.2 package, two samples( BoxFilterSAT and Stockdataanalysis,) are failed to build by throwing compilation error.

This is the issue with VS2012 compiler. The solution is to simplify the code which compiler doesn’t list where to simplify. This issue is fixed with VS2013.

Hard coding the path is creating issues on Ubuntu. I think this will cause issues on OSX as well.

I need bolt library for 32 bit Windows. But when i am getting above error.
Kindly help me in fixing this.

Hi ,
clone the bolt codes, compile on rocm1,9 and opencl-runtime, build the project with cmake commands as "cmake -DBOOST_LIBRARYDIR=/home/qcxie/software/boost_1_65_1/stage/lib -DBOOST_ROOT=/home/qcxie/software/boost_1_65_1 -DGTEST_ROOT=/home/qcxie/software/boost_1_65_1 -DCMAKE_BUILD_TYPE=Debug -DBolt_BUILD64=1 -DCMAKE_CXX_FLAGS="-std =c++14 -fpermissive -I /opt/rocm/opencl/include -L/opt/rocm/opencl/lib/x86_64 -lOpenCL" ../" successfully,
but, it throws cl kernels error in running test case, for example clBolt.Test.StableSort
" error: unknown type name 'namespace' namespace bolt { namespace cl { "
how i should do to configure or set buildprogram optimons to fix this issue? thanks very much.

When I try to build Bolt from source, I get numerous errors about lock_guard because <boost/thread/lock_guard.hpp> is not included anywhere. This could be because I'm using my system's version of Boost (1.53) rather than using the superbuild.

If we run bolt::cl::sort for 1000 iterations by having higher odd buffer sizes like 2 power 23, 2 power 25,.. for double and float data type it is hanging. No issues with non powers of 2 and even buffer sizes like 2 power 24, 2 power 26..

CODE:

int get_global_id(int i);

int global_id;

BOLT_FUNCTOR(UDD_trans,
struct UDD_trans
{
int i ;
float f ;

UDD_trans ()  
    {
    }; 

UDD_trans (int val1)  
    {
        i =  val1 ;

    }; 
bool operator == (const UDD_trans& other) const {
    return ((i+f) == (other.i+other.f));
}

UDD_trans operator() ()  const 
    { 
        UDD_trans temp ;
        temp.i =  get_global_id(0);
        return temp ;
        //return get_global_id(0); 
    }

};
);

int get_global_id(int i)
{
return global_id++;
}

BOLT_FUNCTOR(add_UDD,
struct add_UDD
{
int operator() (const UDD_trans x) const { return x.i + 3; }
typedef int result_type;
};
);

BOLT_TEMPLATE_REGISTER_NEW_ITERATOR( bolt::cl::device_vector, int, UDD_trans);
BOLT_TEMPLATE_REGISTER_NEW_TRANSFORM_ITERATOR( bolt::cl::transform_iterator, add_UDD, UDD_trans);

int main()
{

int length =  5;

std::vector< UDD_trans > svInVec1( length ); 

std::vector< int > stlOut(length);

add_UDD add1;
UDD_trans gen_udd(0) ;


// ADD
bolt::cl::transform_iterator< add_UDD, std::vector< UDD_trans >::const_iterator>        sv_trf_begin1 (svInVec1.begin(), add1) ;
bolt::cl::transform_iterator< add_UDD, std::vector< UDD_trans >::const_iterator>            sv_trf_end1   (svInVec1.end(),   add1) ;

    global_id = 0;
std::generate(svInVec1.begin(), svInVec1.end(), gen_udd);

        bolt::cl::plus<int> pls;

        bolt::cl::control ctrl = bolt::cl::control::getDefault();

    global_id = 0;

 //STD_PARTIAL_SUM
std::partial_sum(sv_trf_begin1, sv_trf_end1, stlOut.begin(), pls);

for (int i=0; i<length;  i++ )
{
    std::cout << "Val = "  << stlOut[i] <<  "\n" ;

}

return 0;
}

This issue is with respect to only for Opensuse 12.3 with gcc 4.7.2 when building bolt from github directly. The same issue working fine with ubuntu 13.10 with gcc 4.8.1.

Does Bolt have a recommended style checker to use? I noticed that even though Bolt has coding style guideline, not all of it is used/enforced. In particular, use of tab character and indentation seem wrong and seems to be different depending who checks in the code.

The guideline states "Use only spaces, and indent 2 spaces at a time", it seems most of the code is indented with 4 spaces and uses tab characters as well as spaces.

Hello,

I'm testing new functionality of Bolt 1.3 library on GPU.

I've followed example from follwong page: http://developer.amd.com/community/blog/2013/04/26/details-of-the-bolt-beta/

I found iterator functionality very attractive but the example from the mentioned webpage seems not to work correctly, at least under linux OS:

#include "bolt/cl/device_vector.h"
#include "bolt/cl/iterator/counting_iterator.h"
#include "bolt/cl/copy.h"

int main( int argc, char* argv[] )
{
    bolt::cl::device_vector< int > devV( 100 );
    bolt::cl::copy( bolt::cl::make_counting_iterator< int >( 10 ),
                    bolt::cl::make_counting_iterator< int >( 10 + devV.size( ) ),
                    devV.begin( ) );
}

The input vector is not changed at all. I've executed it using bolt::cl::transform instead of bolt::cl::copy and it is working correctly.

Hi,
I'm studying Bolt and wanted to implement an example program that needs scatter_if operation (http://thrust.github.io/doc/group__scattering.html#ga1079bc05bcb3d4b5080f1e07444fee37). I started to port thrust scatter_if code, which uses permutation_iterator but came across this (https://groups.google.com/forum/#!topic/thrust-users/Xe2JkFy_hUk). The Google Group post claims that permutation_iterator in AMP kernel is not possible because of the restriction AMP put on the use of pointer in kernel (ie. restrict(amp)). Is this true? If so, is it possible to implement permutation_iterator in bolt::cl?

BTW, Bolt forum in AMD Dev Central does not seem to work correctly. It is set to private and my post there does not seem to go through. :(

Just as the title says, I'm having trouble convincing CMake to let go of the -m32 compiler flag. Everything is working except for that one crucial hangup.

I'm using CMake-3.0.

I've tried to manually edit the CMAKE_CXX_COMPILER and CMAKE_EXE_LINKER variables from the command line with little success.

Any assistance would be most well received.

In developing OpenCV(Open-source Computer Vision library)'s OpenCL module I intended to import Bolt library and use its sorting and scan APIs, but by browsing the source code I noticed that AMD-only Static OpenCL C++ templates is heavily used in OpenCL kernels.

For OpenCV, we must ensure that it can run OpenCL on most platforms, not only AMD's. Can I ask if you have any plans to make Bolt available on non-amdappsdk platforms, such as nvidia's and intel's OpenCL SDK?

To fix this issue, I have adapted some of the codes from Bolt and use macros to simulate templates. What I added in this branch is sort_by_key using OpenCL. By the way, I included radix sort with float type types, which is not supported yet for Bolt.

Please see the following links:
Host code
Radix sort kernel file

Thanks!

Observed test case failures for bolt cl scan families like exclusive_scan, inclusive_scan, transform_exclusive_scan, transform_inclusive_scan only for the OpenCL CPU path. For other paths like Gpu, Automatic, MultiCoreCpu, and Serial Cpu same test cases are passing.

EXAMPLE:
TEST (sanity_exclusive_scan_stdVectVsDeviceVectWithIters, floatSameValuesSerialRange){

int size = 1000;
TAKE_THIS_CONTROL_PATH
bolt::cl::device_vector< float > boltInput( size, 1.125f );
bolt::cl::device_vector< float >::iterator boltEnd = bolt::cl::exclusive_scan( my_ctl, boltInput.begin( ), boltInput.end( ), boltInput.begin( ), 2.0f);

std::vector< float > stdInput( size, 1.125f);
std::vector< float >::iterator stdEnd  =    bolt::cl::exclusive_scan( stdInput.begin( ), stdInput.end( ), stdInput.begin( ), 2.0f );

EXPECT_FLOAT_EQ((*(boltEnd-1)), (*(stdEnd-1)))<<std::endl;

}

Using std::merge/std::stable_sort function with device_vector as a inputs( running function on CPU when data present on GPU) doesn't compile .

Linux systems do not have 32 bit headers and libraries installed by default. Debugging the errors that arise because of this cause a lot of overhead on the users end.

#include <iostream>
#include <vector>
#include <bolt/cl/iterator/counting_iterator.h>
#include <bolt/cl/iterator/transform_iterator.h>
#include <bolt/cl/functional.h>
#include <bolt/cl/device_vector.h>
#include <bolt/cl/copy.h>

BOLT_FUNCTOR(GetSquare,
struct GetSquare
{
public:
    int operator()(const int& globalId) const
    {
        return globalId*globalId;
    }
};);

int main()
{
    const std::size_t n=10;

    bolt::cl::control ctrl = bolt::cl::control::getDefault();

    bolt::cl::device_vector<int> debug(n);

    auto globalId = bolt::cl::make_counting_iterator(0);

    // This is OK
    // bolt::cl::transform(globalId, globalId + n, debug.begin(), GetSquare());

    // This causes compilation error
    auto square = bolt::cl::make_transform_iterator(globalId, GetSquare());
    bolt::cl::copy(square, square + n, debug.begin());

    for(int i = 0; i < n; i++)
    {
        std::cout << i << ": " << debug[i] << std::endl;
    }

    return 0;
}

This problem seems to be because bolt::cl::transform_iterator have the method getContainer() only with template type

template<typename Container >
Container& getContainer() const
{
    return this->base().getContainer( );
}

but bolt::cl::copy needs ITERATOR::getContainer() without any template type

V_OPENCL( kernels[whichKernel].setArg( 0, first.getContainer().getBuffer()), "Error setArg kernels[ 0 ]" );

This can be solved with C++11

auto getContainer() const -> decltype(base().getContainer())
{
    return this->base().getContainer();
}

I don't have any idea in C++03 (boost::result_of?).

when std::sort function call on device_vector, means running function on CPU when we have data on GPU. These are working fine on Windows 32/64 and linux 64bit. We are seeing this only on 32 bit linux which may be because of compiler restriction.

I noticed that there are some differences between develop branch and master/v1.0 branch. Is that intentional? Most of the differences are minor documentation differences that probably won't affect much. But it seems like something was missed when syncing v1.0/master/develop branches in preparation of v1.0 release.

I'm just getting started with Git way of doing thing. I've only been using SVN. So, it's possible that I'm missing something...

Also, to submit an entry for Bolt Sample Code Contest, I should open a pull request to develop branch, right? Or does it not matter?

Enhancement:
Some JNA or JNI binding for a Bolt wrapper, to get Bolt-like API with GPU and performance access from Java.

Since Aparapi seems to be on hold, experimental and solving other issues that bolt does. And Sumatra may not even be available with Java9 in 2016, it would be very nice to get some Java access to GPU reduce, sort, and similar stuff

Hi,
sorry as for sure that's not the best place to post but I remember seeing at APU13 a presentation saying AMD would open source his "OpenCL Static C++ Kernel Language Extension" so all OpenCL implementations could take advantage of it.. I think it was in Bolt presentation so asking here as I also remember it's was said either Q1 2014 or H1 2014 so must be coming soon right?
Also Bolt takes advantage of it right now in OCL path right?