int
Lucas::fft_setup (int length)
{
clAmdFftSetupData fftSetupData;
OPENCL_V_THROW(clAmdFftInitSetupData (&fftSetupData),"Failed to clAmdFftInitSetupData.");
fftSetupData.debugFlags = CLFFT_DUMP_PROGRAMS; // Dumps the FFT kernels
// Setup the AMD FFT library.
OPENCL_V_THROW(clAmdFftSetup (&fftSetupData),"Failed to clAmdFftSetup.");
// Create FFT Plan
const size_t logicalDimensions[1] = { length };
// Create default plan.
OPENCL_V_THROW(clAmdFftCreateDefaultPlan(&plan, context(), CLFFT_1D, logicalDimensions),"Failed to clAmdFftCreateDefaultPlan.");
// Set double precision.
OPENCL_V_THROW(clAmdFftSetPlanPrecision(plan, CLFFT_DOUBLE),"Failed to clAmdFftSetPlanPrecision.");
// Set layout.
OPENCL_V_THROW(clAmdFftSetLayout(plan, CLFFT_COMPLEX_INTERLEAVED,CLFFT_COMPLEX_INTERLEAVED),"Failed to clAmdFftSetLayout.");
// Normalize forward transformation.
OPENCL_V_THROW(clAmdFftSetPlanScale(plan, CLFFT_FORWARD, 1.0f/static_cast<cl_float>(length)),"Failed to clAmdFftSetPlanScale.");
// Normalize backward transformation.
OPENCL_V_THROW(clAmdFftSetPlanScale(plan, CLFFT_BACKWARD, 1.0f),"Failed to clAmdFftSetPlanScale.");
// In-place FFT.
OPENCL_V_THROW(clAmdFftSetResultLocation(plan, CLFFT_INPLACE),"Failed to clAmdFftSetResultLocation.");
// Set number of transformations per plan.
OPENCL_V_THROW(clAmdFftSetPlanBatchSize(plan, 1),"Failed to clAmdFftSetPlanBatchSize.");
// BakePlan
OPENCL_V_THROW(clAmdFftBakePlan(plan, 1, &commandQueue(), NULL, NULL),"Failed to clAmdFftBakePlan.");
}
virtual ~clblasFunc()
{
clblasTeardown();
OPENCL_V_THROW( clReleaseCommandQueue(queue_),
"releasing command queue" );
OPENCL_V_THROW( clReleaseContext(ctx_), "releasing context" );
}
clblasFunc(StatisticalTimer& _timer, cl_device_type devType)
: timer(_timer)
{
cl_int err;
/* Setup OpenCL environment. */
OPENCL_V_THROW(clGetPlatformIDs(1, &platform_, NULL),
"getting platform IDs");
OPENCL_V_THROW(clGetDeviceIDs(platform_, devType, 1,
&device_, NULL), "getting device IDs");
props_[0] = CL_CONTEXT_PLATFORM;
props_[1] = (cl_context_properties)platform_;
props_[2] = 0;
ctx_ = clCreateContext(props_, 1, &device_, NULL, NULL, &err);
OPENCL_V_THROW(err, "creating context");
queue_ = clCreateCommandQueue(ctx_, device_, 0, &err);
timer_id = timer.getUniqueID( "clfunc", 0 );
maxMemAllocSize = queryMemAllocSize( device_ );
/* Setup clblas. */
err = clblasSetup();
if (err != CL_SUCCESS) {
std::cerr << "clblasSetup() failed with %d\n";
clReleaseCommandQueue(queue_);
clReleaseContext(ctx_);
}
}
~xGeru()
{
delete buffer.cpuA;
delete buffer.cpuX;
OPENCL_V_THROW( clReleaseMemObject(buffer.A), "releasing buffer A");
OPENCL_V_THROW( clReleaseMemObject(buffer.X), "releasing buffer C");
}
~xTrmv()
{
delete buffer_.a_;
delete buffer_.x_;
OPENCL_V_THROW( clReleaseMemObject(buffer_.buf_a_), "releasing buffer A");
OPENCL_V_THROW( clReleaseMemObject(buffer_.buf_x_), "releasing buffer X");
OPENCL_V_THROW( clReleaseMemObject(buffer_.scratch_), "releasing buffer X");
}
~xSyrk()
{
delete buffer_.a_;
delete buffer_.c_;
OPENCL_V_THROW( clReleaseMemObject(buffer_.buf_a_),
"releasing buffer A");
OPENCL_V_THROW( clReleaseMemObject(buffer_.buf_c_),
"releasing buffer C");
}
void releaseGPUBuffer_deleteCPUBuffer()
{
//this is necessary since we are running a iteration of tests and calculate the average time. (in client.cpp)
//need to do this before we eventually hit the destructor
delete buffer_.a_;
delete buffer_.c_;
OPENCL_V_THROW( clReleaseMemObject(buffer_.buf_a_),
"releasing buffer A");
OPENCL_V_THROW( clReleaseMemObject(buffer_.buf_c_),
"releasing buffer C");
}
~xSymv()
{
delete buffer_.a_;
delete buffer_.x_;
delete buffer_.y_;
OPENCL_V_THROW( clReleaseMemObject(buffer_.buf_a_),
"releasing buffer A");
OPENCL_V_THROW( clReleaseMemObject(buffer_.buf_x_),
"releasing buffer X");
OPENCL_V_THROW( clReleaseMemObject(buffer_.buf_y_),
"releasing buffer Y");
}
size_t max_mem_available_on_cl_device(size_t device_index) {
std::vector< cl_device_id > device_id;
cl_context tempContext = NULL;
device_id = initializeCL(
g_device_type,
(cl_int)device_index,
g_platform_id,
tempContext,
false
);
cl_ulong device_max_to_allocate = 0;
if( device_id.size() == 0 || device_index > device_id.size() )
{
}
else
{
OPENCL_V_THROW( ::clGetDeviceInfo( device_id[device_index], CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof( cl_ulong ), &device_max_to_allocate, NULL ),
"Getting CL_DEVICE_MAX_MEM_ALLOC_SIZE device info ( ::clGetDeviceInfo() )" );
}
cl_command_queue tempQueue = NULL;
cl_event tempEvent = NULL;
::cleanupCL( &tempContext, &tempQueue, 0, NULL, 0, NULL, &tempEvent );
return static_cast<size_t>(device_max_to_allocate);
}
size_t max_mem_available_on_cl_device(size_t device_index) {
static size_t g_device_max_mem_size = 0;
// this is not thread-safe using globals, it is just quick fix for now, todo proper fix
if (g_device_max_mem_size == 0)
{
std::vector< cl_device_id > device_id;
cl_context tempContext = NULL;
device_id = initializeCL(
g_device_type,
(cl_int)device_index,
g_platform_id,
tempContext,
false
);
cl_ulong device_max_to_allocate = 0;
if (device_id.size() == 0 || device_index > device_id.size())
{
}
else
{
OPENCL_V_THROW(::clGetDeviceInfo(device_id[device_index], CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(cl_ulong), &device_max_to_allocate, NULL),
"Getting CL_DEVICE_MAX_MEM_ALLOC_SIZE device info ( ::clGetDeviceInfo() )");
}
cl_command_queue tempQueue = NULL;
cl_event tempEvent = NULL;
::cleanupCL(&tempContext, &tempQueue, 0, NULL, 0, NULL, &tempEvent);
g_device_max_mem_size = static_cast<size_t>(device_max_to_allocate);
}
return g_device_max_mem_size;
}
int
Lucas::fft_close ()
{
OPENCL_V_THROW(clAmdFftDestroyPlan (&plan),"Failed to clAmdFftDestroyPlan.");
OPENCL_V_THROW(clAmdFftTeardown (),"Failed to clAmdFftTeardown.");
}
int _tmain( int argc, _TCHAR* argv[] )
{
// This helps with mixing output of both wide and narrow characters to the screen
std::ios::sync_with_stdio( false );
// Define MEMORYREPORT on windows platfroms to enable debug memory heap checking
#if defined( MEMORYREPORT ) && defined( _WIN32 )
TCHAR logPath[ MAX_PATH ];
::GetCurrentDirectory( MAX_PATH, logPath );
::_tcscat_s( logPath, _T( "\\MemoryReport.txt") );
// We leak the handle to this file, on purpose, so that the ::_CrtSetReportFile() can output it's memory
// statistics on app shutdown
HANDLE hLogFile;
hLogFile = ::CreateFile( logPath, GENERIC_WRITE,
FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, CREATE_ALWAYS, FILE_ATTRIBUTE_NORMAL, NULL );
::_CrtSetReportMode( _CRT_ASSERT, _CRTDBG_MODE_FILE | _CRTDBG_MODE_WNDW | _CRTDBG_MODE_DEBUG );
::_CrtSetReportMode( _CRT_ERROR, _CRTDBG_MODE_FILE | _CRTDBG_MODE_WNDW | _CRTDBG_MODE_DEBUG );
::_CrtSetReportMode( _CRT_WARN, _CRTDBG_MODE_FILE | _CRTDBG_MODE_DEBUG );
::_CrtSetReportFile( _CRT_ASSERT, hLogFile );
::_CrtSetReportFile( _CRT_ERROR, hLogFile );
::_CrtSetReportFile( _CRT_WARN, hLogFile );
int tmp = ::_CrtSetDbgFlag( _CRTDBG_REPORT_FLAG );
tmp |= _CRTDBG_LEAK_CHECK_DF | _CRTDBG_ALLOC_MEM_DF | _CRTDBG_CHECK_ALWAYS_DF;
::_CrtSetDbgFlag( tmp );
// By looking at the memory leak report that is generated by this debug heap, there is a number with
// {} brackets that indicates the incremental allocation number of that block. If you wish to set
// a breakpoint on that allocation number, put it in the _CrtSetBreakAlloc() call below, and the heap
// will issue a bp on the request, allowing you to look at the call stack
// ::_CrtSetBreakAlloc( 1833 );
#endif /* MEMORYREPORT */
// OpenCL state
cl_device_type deviceType = CL_DEVICE_TYPE_ALL;
cl_int deviceId = 0;
cl_int platformId = 0;
// FFT state
clfftResultLocation place = CLFFT_INPLACE;
clfftLayout inLayout = CLFFT_COMPLEX_INTERLEAVED;
clfftLayout outLayout = CLFFT_COMPLEX_INTERLEAVED;
clfftPrecision precision = CLFFT_SINGLE;
clfftDirection dir = CLFFT_FORWARD;
size_t lengths[ 3 ] = {1,1,1};
size_t iStrides[ 4 ] = {0,0,0,0};
size_t oStrides[ 4 ] = {0,0,0,0};
cl_uint profile_count = 0;
cl_uint command_queue_flags = 0;
size_t batchSize = 1;
// Initialize flags for FFT library
std::auto_ptr< clfftSetupData > setupData( new clfftSetupData );
OPENCL_V_THROW( clfftInitSetupData( setupData.get( ) ),
"clfftInitSetupData failed" );
try
{
// Declare the supported options.
po::options_description desc( "clFFT client command line options" );
desc.add_options()
( "help,h", "produces this help message" )
( "version,v", "Print queryable version information from the clFFT library" )
( "clinfo,i", "Print queryable information of all the OpenCL runtimes and devices" )
( "printChosen", "Print queryable information of the selected OpenCL runtime and device" )
( "gpu,g", "Force selection of OpenCL GPU devices only" )
( "cpu,c", "Force selection of OpenCL CPU devices only" )
( "all,a", "Force selection of all OpenCL devices (default)" )
( "platform", po::value< cl_int >( &platformId )->default_value( 0 ), "Select a specific OpenCL platform id as it is reported by clinfo" )
( "device", po::value< cl_int >( &deviceId )->default_value( 0 ), "Select a specific OpenCL device id as it is reported by clinfo" )
( "outPlace,o", "Out of place FFT transform (default: in place)" )
( "double", "Double precision transform (default: single)" )
( "inv", "Backward transform (default: forward)" )
( "dumpKernels,d", "FFT engine will dump generated OpenCL FFT kernels to disk (default: dump off)" )
( "lenX,x", po::value< size_t >( &lengths[ 0 ] )->default_value( 1024 ), "Specify the length of the 1st dimension of a test array" )
( "lenY,y", po::value< size_t >( &lengths[ 1 ] )->default_value( 1 ), "Specify the length of the 2nd dimension of a test array" )
( "lenZ,z", po::value< size_t >( &lengths[ 2 ] )->default_value( 1 ), "Specify the length of the 3rd dimension of a test array" )
( "isX", po::value< size_t >( &iStrides[ 0 ] )->default_value( 1 ), "Specify the input stride of the 1st dimension of a test array" )
( "isY", po::value< size_t >( &iStrides[ 1 ] )->default_value( 0 ), "Specify the input stride of the 2nd dimension of a test array" )
( "isZ", po::value< size_t >( &iStrides[ 2 ] )->default_value( 0 ), "Specify the input stride of the 3rd dimension of a test array" )
( "iD", po::value< size_t >( &iStrides[ 3 ] )->default_value( 0 ), "input distance between subsequent sets of data when batch size > 1" )
( "osX", po::value< size_t >( &oStrides[ 0 ] )->default_value( 1 ), "Specify the output stride of the 1st dimension of a test array" )
( "osY", po::value< size_t >( &oStrides[ 1 ] )->default_value( 0 ), "Specify the output stride of the 2nd dimension of a test array" )
( "osZ", po::value< size_t >( &oStrides[ 2 ] )->default_value( 0 ), "Specify the output stride of the 3rd dimension of a test array" )
( "oD", po::value< size_t >( &oStrides[ 3 ] )->default_value( 0 ), "output distance between subsequent sets of data when batch size > 1" )
( "batchSize,b", po::value< size_t >( &batchSize )->default_value( 1 ), "If this value is greater than one, arrays will be used " )
( "profile,p", po::value< cl_uint >( &profile_count )->default_value( 1 ), "Time and report the kernel speed of the FFT (default: profiling off)" )
( "inLayout", po::value< clfftLayout >( &inLayout )->default_value( CLFFT_COMPLEX_INTERLEAVED ), "Layout of input data:\n1) interleaved\n2) planar\n3) hermitian interleaved\n4) hermitian planar\n5) real" )
( "outLayout", po::value< clfftLayout >( &outLayout )->default_value( CLFFT_COMPLEX_INTERLEAVED ), "Layout of input data:\n1) interleaved\n2) planar\n3) hermitian interleaved\n4) hermitian planar\n5) real" )
;
po::variables_map vm;
po::store( po::parse_command_line( argc, argv, desc ), vm );
po::notify( vm );
if( vm.count( "version" ) )
{
const int indent = countOf( "clFFT client API version: " );
tout << std::left << std::setw( indent ) << _T( "clFFT client API version: " )
<< clfftVersionMajor << _T( "." )
<< clfftVersionMinor << _T( "." )
<< clfftVersionPatch << std::endl;
cl_uint libMajor, libMinor, libPatch;
clfftGetVersion( &libMajor, &libMinor, &libPatch );
tout << std::left << std::setw( indent ) << _T( "clFFT runtime version: " )
<< libMajor << _T( "." )
<< libMinor << _T( "." )
<< libPatch << std::endl << std::endl;
}
if( vm.count( "help" ) )
{
// This needs to be 'cout' as program-options does not support wcout yet
std::cout << desc << std::endl;
return 0;
}
size_t mutex = ((vm.count( "gpu" ) > 0) ? 1 : 0)
| ((vm.count( "cpu" ) > 0) ? 2 : 0)
| ((vm.count( "all" ) > 0) ? 4 : 0);
if ((mutex & (mutex-1)) != 0) {
terr << _T("You have selected mutually-exclusive OpenCL device options:") << std::endl;
if (vm.count ( "gpu" ) > 0) terr << _T(" gpu,g Force selection of OpenCL GPU devices only" ) << std::endl;
if (vm.count ( "cpu" ) > 0) terr << _T(" cpu,c Force selection of OpenCL CPU devices only" ) << std::endl;
if (vm.count ( "all" ) > 0) terr << _T(" all,a Force selection of all OpenCL devices (default)" ) << std::endl;
return 1;
}
if( vm.count( "gpu" ) )
{
deviceType = CL_DEVICE_TYPE_GPU;
}
if( vm.count( "cpu" ) )
{
deviceType = CL_DEVICE_TYPE_CPU;
}
if( vm.count( "all" ) )
{
deviceType = CL_DEVICE_TYPE_ALL;
}
if( vm.count( "clinfo" ) )
{
std::vector< cl_platform_id > platformInfos;
std::vector< std::vector< cl_device_id > > deviceInfos;
discoverCLPlatforms( deviceType, platformInfos, deviceInfos );
prettyPrintCLPlatforms(platformInfos, deviceInfos);
return 0;
}
bool printInfo = false;
if( vm.count( "printChosen" ) )
{
printInfo = true;
}
if( vm.count( "outPlace" ) )
{
place = CLFFT_OUTOFPLACE;
}
if( vm.count( "double" ) )
{
precision = CLFFT_DOUBLE;
}
if( vm.count( "inv" ) )
{
dir = CLFFT_BACKWARD;
}
if( profile_count > 1 )
{
command_queue_flags |= CL_QUEUE_PROFILING_ENABLE;
}
if( vm.count( "dumpKernels" ) )
{
setupData->debugFlags |= CLFFT_DUMP_PROGRAMS;
}
int inL = (int)inLayout;
int otL = (int)outLayout;
// input output layout support matrix
int ioLayoutSupport[5][5] = {
{ 1, 1, 0, 0, 1 },
{ 1, 1, 0, 0, 1 },
{ 0, 0, 0, 0, 1 },
{ 0, 0, 0, 0, 1 },
{ 1, 1, 1, 1, 0 },
};
if((inL < 1) || (inL > 5)) throw std::runtime_error( "Invalid Input layout format" );
if((otL < 1) || (otL > 5)) throw std::runtime_error( "Invalid Output layout format" );
if(ioLayoutSupport[inL-1][otL-1] == 0) throw std::runtime_error( "Invalid combination of Input/Output layout formats" );
if( ((inL == 1) || (inL == 2)) && ((otL == 1) || (otL == 2)) ) // Complex-Complex cases
{
iStrides[1] = iStrides[1] ? iStrides[1] : lengths[0] * iStrides[0];
iStrides[2] = iStrides[2] ? iStrides[2] : lengths[1] * iStrides[1];
iStrides[3] = iStrides[3] ? iStrides[3] : lengths[2] * iStrides[2];
if(place == CLFFT_INPLACE)
{
oStrides[0] = iStrides[0];
oStrides[1] = iStrides[1];
oStrides[2] = iStrides[2];
oStrides[3] = iStrides[3];
}
else
{
oStrides[1] = oStrides[1] ? oStrides[1] : lengths[0] * oStrides[0];
oStrides[2] = oStrides[2] ? oStrides[2] : lengths[1] * oStrides[1];
oStrides[3] = oStrides[3] ? oStrides[3] : lengths[2] * oStrides[2];
}
}
else // Real-Complex and Complex-Real cases
{
size_t *rst, *cst;
size_t N = lengths[0];
size_t Nt = 1 + lengths[0]/2;
bool iflag = false;
bool rcFull = (inL == 1) || (inL == 2) || (otL == 1) || (otL == 2);
if(inLayout == CLFFT_REAL) { iflag = true; rst = iStrides; }
else { rst = oStrides; } // either in or out should be REAL
// Set either in or out strides whichever is real
if(place == CLFFT_INPLACE)
{
if(rcFull) { rst[1] = rst[1] ? rst[1] : N * 2 * rst[0]; }
else { rst[1] = rst[1] ? rst[1] : Nt * 2 * rst[0]; }
rst[2] = rst[2] ? rst[2] : lengths[1] * rst[1];
rst[3] = rst[3] ? rst[3] : lengths[2] * rst[2];
}
else
{
rst[1] = rst[1] ? rst[1] : lengths[0] * rst[0];
rst[2] = rst[2] ? rst[2] : lengths[1] * rst[1];
rst[3] = rst[3] ? rst[3] : lengths[2] * rst[2];
}
// Set the remaining of in or out strides that is not real
if(iflag) { cst = oStrides; }
else { cst = iStrides; }
if(rcFull) { cst[1] = cst[1] ? cst[1] : N * cst[0]; }
else { cst[1] = cst[1] ? cst[1] : Nt * cst[0]; }
cst[2] = cst[2] ? cst[2] : lengths[1] * cst[1];
cst[3] = cst[3] ? cst[3] : lengths[2] * cst[2];
}
if( precision == CLFFT_SINGLE )
transform<float>( lengths, iStrides, oStrides, batchSize, inLayout, outLayout, place, precision, dir, deviceType, deviceId, platformId, printInfo, command_queue_flags, profile_count, setupData );
else
transform<double>( lengths, iStrides, oStrides, batchSize, inLayout, outLayout, place, precision, dir, deviceType, deviceId, platformId, printInfo, command_queue_flags, profile_count, setupData );
}
catch( std::exception& e )
{
terr << _T( "clFFT error condition reported:" ) << std::endl << e.what() << std::endl;
return 1;
}
return 0;
}
int transform( size_t* lengths, const size_t *inStrides, const size_t *outStrides, size_t batch_size,
clfftLayout in_layout, clfftLayout out_layout,
clfftResultLocation place, clfftPrecision precision, clfftDirection dir,
cl_device_type deviceType, cl_int deviceId, cl_int platformId, bool printInfo,
cl_uint command_queue_flags, cl_uint profile_count,
std::auto_ptr< clfftSetupData > setupData )
{
// Our command line does not specify what dimension FFT we wish to transform; we decode
// this from the lengths that the user specifies for X, Y, Z. A length of one means that
// The user does not want that dimension.
const size_t max_dimensions = 3;
size_t strides[ 4 ];
size_t o_strides[ 4 ];
size_t fftVectorSize = 0;
size_t fftVectorSizePadded = 0;
size_t fftBatchSize = 0;
size_t outfftVectorSize = 0;
size_t outfftVectorSizePadded = 0;
size_t outfftBatchSize = 0;
size_t size_of_input_buffers_in_bytes = 0;
size_t size_of_output_buffers_in_bytes = 0;
cl_uint number_of_output_buffers = 0;
clfftDim dim = CLFFT_1D;
cl_mem input_cl_mem_buffers [2] = { NULL, NULL };
cl_mem output_cl_mem_buffers[2] = { NULL, NULL };
std::vector< cl_device_id > device_id;
cl_context context;
cl_command_queue queue;
cl_event outEvent = NULL;
clfftPlanHandle plan_handle;
for (unsigned u = 0; u < max_dimensions; ++u) {
if (0 != lengths[u])
continue;
lengths[u] = 1;
}
if( lengths[ 1 ] > 1 )
{
dim = CLFFT_2D;
}
if( lengths[ 2 ] > 1 )
{
dim = CLFFT_3D;
}
strides[ 0 ] = inStrides[0];
strides[ 1 ] = inStrides[1];
strides[ 2 ] = inStrides[2];
strides[ 3 ] = inStrides[3];
o_strides[ 0 ] = outStrides[0];
o_strides[ 1 ] = outStrides[1];
o_strides[ 2 ] = outStrides[2];
o_strides[ 3 ] = outStrides[3];
fftVectorSize = lengths[0] * lengths[1] * lengths[2];
fftVectorSizePadded = strides[3];
fftBatchSize = fftVectorSizePadded * batch_size;
size_t Nt = 1 + lengths[0]/2;
if(place == CLFFT_INPLACE)
{
outfftVectorSize = fftVectorSize;
outfftVectorSizePadded = fftVectorSizePadded;
outfftBatchSize = fftBatchSize;
}
else
{
outfftVectorSize = lengths[0] * lengths[1] * lengths[2];
outfftVectorSizePadded = o_strides[3];
outfftBatchSize = outfftVectorSizePadded * batch_size;
}
// Real to complex case
if( (in_layout == CLFFT_REAL) || (out_layout == CLFFT_REAL) )
{
fftVectorSizePadded = strides[3];
fftBatchSize = fftVectorSizePadded * batch_size;
outfftVectorSizePadded = o_strides[3];
outfftBatchSize = outfftVectorSizePadded * batch_size;
fftVectorSize = lengths[0] * lengths[1] * lengths[2];
outfftVectorSize = fftVectorSize;
}
switch( out_layout )
{
case CLFFT_COMPLEX_INTERLEAVED:
number_of_output_buffers = 1;
size_of_output_buffers_in_bytes = outfftBatchSize * sizeof( std::complex< T > );
break;
case CLFFT_COMPLEX_PLANAR:
number_of_output_buffers = 2;
size_of_output_buffers_in_bytes = outfftBatchSize * sizeof(T);
break;
case CLFFT_HERMITIAN_INTERLEAVED:
number_of_output_buffers = 1;
size_of_output_buffers_in_bytes = outfftBatchSize * sizeof( std::complex< T > );
break;
case CLFFT_HERMITIAN_PLANAR:
number_of_output_buffers = 2;
size_of_output_buffers_in_bytes = outfftBatchSize * sizeof(T);
break;
case CLFFT_REAL:
number_of_output_buffers = 1;
size_of_output_buffers_in_bytes = outfftBatchSize * sizeof(T);
break;
}
// Fill the input buffers
switch( in_layout )
{
case CLFFT_COMPLEX_INTERLEAVED:
{
// This call creates our openCL context and sets up our devices; expected to throw on error
size_of_input_buffers_in_bytes = fftBatchSize * sizeof( std::complex< T > );
device_id = initializeCL( deviceType, deviceId, platformId, context, printInfo );
createOpenCLCommandQueue( context,
command_queue_flags, queue,
device_id,
size_of_input_buffers_in_bytes, 1, input_cl_mem_buffers,
size_of_output_buffers_in_bytes, number_of_output_buffers, output_cl_mem_buffers);
std::vector< std::complex< T > > input( fftBatchSize );
// set zero
for( cl_uint i = 0; i < fftBatchSize; ++i )
{
input[ i ] = 0;
}
// impulse test case
for(size_t b = 0; b < batch_size; b++)
{
size_t p3 = b * strides[3];
for(size_t k = 0; k < lengths[2]; k++)
{
size_t p2 = p3 + k * strides[2];
for(size_t j = 0; j < lengths[1]; j++)
{
size_t p1 = p2 + j * strides[1];
for(size_t i = 0; i < lengths[0]; i++)
{
size_t p0 = p1 + i * strides[0];
input[p0] = 1;
}
}
}
}
OPENCL_V_THROW( clEnqueueWriteBuffer( queue, input_cl_mem_buffers[ 0 ], CL_TRUE, 0, size_of_input_buffers_in_bytes, &input[ 0 ],
0, NULL, &outEvent ),
"clEnqueueWriteBuffer failed" );
}
break;
case CLFFT_COMPLEX_PLANAR:
{
// This call creates our openCL context and sets up our devices; expected to throw on error
size_of_input_buffers_in_bytes = fftBatchSize * sizeof( T );
device_id = initializeCL( deviceType, deviceId, platformId, context, printInfo );
createOpenCLCommandQueue( context,
command_queue_flags, queue,
device_id,
size_of_input_buffers_in_bytes, 2, input_cl_mem_buffers,
size_of_output_buffers_in_bytes, number_of_output_buffers, output_cl_mem_buffers);
std::vector< T > real( fftBatchSize );
std::vector< T > imag( fftBatchSize );
// set zero
for( cl_uint i = 0; i < fftBatchSize; ++i )
{
real[ i ] = 0;
imag[ i ] = 0;
}
// impulse test case
for(size_t b = 0; b < batch_size; b++)
{
size_t p3 = b * strides[3];
for(size_t k = 0; k < lengths[2]; k++)
{
size_t p2 = p3 + k * strides[2];
for(size_t j = 0; j < lengths[1]; j++)
{
size_t p1 = p2 + j * strides[1];
for(size_t i = 0; i < lengths[0]; i++)
{
size_t p0 = p1 + i * strides[0];
real[p0] = 1;
}
}
}
}
OPENCL_V_THROW( clEnqueueWriteBuffer( queue, input_cl_mem_buffers[ 0 ], CL_TRUE, 0, size_of_input_buffers_in_bytes, &real[ 0 ],
0, NULL, &outEvent ),
"clEnqueueWriteBuffer failed" );
OPENCL_V_THROW( clEnqueueWriteBuffer( queue, input_cl_mem_buffers[ 1 ], CL_TRUE, 0, size_of_input_buffers_in_bytes, &imag[ 0 ],
0, NULL, &outEvent ),
"clEnqueueWriteBuffer failed" );
}
break;
case CLFFT_HERMITIAN_INTERLEAVED:
{
// This call creates our openCL context and sets up our devices; expected to throw on error
size_of_input_buffers_in_bytes = fftBatchSize * sizeof( std::complex< T > );
device_id = initializeCL( deviceType, deviceId, platformId, context, printInfo );
createOpenCLCommandQueue( context,
command_queue_flags, queue,
device_id,
size_of_input_buffers_in_bytes, 1, input_cl_mem_buffers,
size_of_output_buffers_in_bytes, number_of_output_buffers, output_cl_mem_buffers);
std::vector< std::complex< T > > input( fftBatchSize );
// set zero
for( cl_uint i = 0; i < fftBatchSize; ++i )
{
input[ i ] = 0;
}
// impulse test case
for(size_t b = 0; b < batch_size; b++)
{
size_t p3 = b * strides[3];
input[p3] = static_cast<T>(outfftVectorSize);
}
OPENCL_V_THROW( clEnqueueWriteBuffer( queue, input_cl_mem_buffers[ 0 ], CL_TRUE, 0, size_of_input_buffers_in_bytes, &input[ 0 ],
0, NULL, &outEvent ),
"clEnqueueWriteBuffer failed" );
}
break;
case CLFFT_HERMITIAN_PLANAR:
{
// This call creates our openCL context and sets up our devices; expected to throw on error
size_of_input_buffers_in_bytes = fftBatchSize * sizeof( T );
device_id = initializeCL( deviceType, deviceId, platformId, context, printInfo );
createOpenCLCommandQueue( context,
command_queue_flags, queue,
device_id,
size_of_input_buffers_in_bytes, 2, input_cl_mem_buffers,
size_of_output_buffers_in_bytes, number_of_output_buffers, output_cl_mem_buffers);
std::vector< T > real( fftBatchSize );
std::vector< T > imag( fftBatchSize );
// set zero
for( cl_uint i = 0; i < fftBatchSize; ++i )
{
real[ i ] = 0;
imag[ i ] = 0;
}
// impulse test case
for(size_t b = 0; b < batch_size; b++)
{
size_t p3 = b * strides[3];
real[p3] = static_cast<T>(outfftVectorSize);
}
OPENCL_V_THROW( clEnqueueWriteBuffer( queue, input_cl_mem_buffers[ 0 ], CL_TRUE, 0, size_of_input_buffers_in_bytes, &real[ 0 ],
0, NULL, &outEvent ),
"clEnqueueWriteBuffer failed" );
OPENCL_V_THROW( clEnqueueWriteBuffer( queue, input_cl_mem_buffers[ 1 ], CL_TRUE, 0, size_of_input_buffers_in_bytes, &imag[ 0 ],
0, NULL, &outEvent ),
"clEnqueueWriteBuffer failed" );
}
break;
case CLFFT_REAL:
{
// This call creates our openCL context and sets up our devices; expected to throw on error
size_of_input_buffers_in_bytes = fftBatchSize * sizeof( T );
device_id = initializeCL( deviceType, deviceId, platformId, context, printInfo );
createOpenCLCommandQueue( context,
command_queue_flags, queue,
device_id,
size_of_input_buffers_in_bytes, 1, input_cl_mem_buffers,
size_of_output_buffers_in_bytes, number_of_output_buffers, output_cl_mem_buffers);
std::vector< T > real( fftBatchSize );
// set zero
for( cl_uint i = 0; i < fftBatchSize; ++i )
{
real[ i ] = 0;
}
// impulse test case
for(size_t b = 0; b < batch_size; b++)
{
size_t p3 = b * strides[3];
for(size_t k = 0; k < lengths[2]; k++)
{
size_t p2 = p3 + k * strides[2];
for(size_t j = 0; j < lengths[1]; j++)
{
size_t p1 = p2 + j * strides[1];
for(size_t i = 0; i < lengths[0]; i++)
{
size_t p0 = p1 + i * strides[0];
real[p0] = 1;
}
}
}
}
OPENCL_V_THROW( clEnqueueWriteBuffer( queue, input_cl_mem_buffers[ 0 ], CL_TRUE, 0, size_of_input_buffers_in_bytes, &real[ 0 ],
0, NULL, &outEvent ),
"clEnqueueWriteBuffer failed" );
}
break;
default:
{
throw std::runtime_error( "Input layout format not yet supported" );
}
break;
}
// Discover and load the timer module if present
void* timerLibHandle = LoadSharedLibrary( "lib", "StatTimer", false );
if( timerLibHandle == NULL )
{
terr << _T( "Could not find the external timing library; timings disabled" ) << std::endl;
}
// Timer module discovered and loaded successfully
// Initialize function pointers to call into the shared module
PFGETSTATTIMER get_timer = reinterpret_cast< PFGETSTATTIMER > ( LoadFunctionAddr( timerLibHandle, "getStatTimer" ) );
// Create and initialize our timer class, if the external timer shared library loaded
baseStatTimer* timer = NULL;
size_t clFFTID = 0;
if( get_timer )
{
timer = get_timer( CLFFT_GPU );
timer->Reserve( 1, profile_count );
timer->setNormalize( true );
clFFTID = timer->getUniqueID( "clFFT", 0 );
}
OPENCL_V_THROW( clfftSetup( setupData.get( ) ), "clfftSetup failed" );
OPENCL_V_THROW( clfftCreateDefaultPlan( &plan_handle, context, dim, lengths ), "clfftCreateDefaultPlan failed" );
// Default plan creates a plan that expects an inPlace transform with interleaved complex numbers
OPENCL_V_THROW( clfftSetResultLocation( plan_handle, place ), "clfftSetResultLocation failed" );
OPENCL_V_THROW( clfftSetLayout( plan_handle, in_layout, out_layout ), "clfftSetLayout failed" );
OPENCL_V_THROW( clfftSetPlanBatchSize( plan_handle, batch_size ), "clfftSetPlanBatchSize failed" );
OPENCL_V_THROW( clfftSetPlanPrecision( plan_handle, precision ), "clfftSetPlanPrecision failed" );
OPENCL_V_THROW (clfftSetPlanInStride ( plan_handle, dim, strides ), "clfftSetPlanInStride failed" );
OPENCL_V_THROW (clfftSetPlanOutStride ( plan_handle, dim, o_strides ), "clfftSetPlanOutStride failed" );
OPENCL_V_THROW (clfftSetPlanDistance ( plan_handle, strides[ 3 ], o_strides[ 3 ]), "clfftSetPlanDistance failed" );
// Set backward scale factor to 1.0 for non real FFTs to do correct output checks
if(dir == CLFFT_BACKWARD && in_layout != CLFFT_REAL && out_layout != CLFFT_REAL)
OPENCL_V_THROW (clfftSetPlanScale( plan_handle, CLFFT_BACKWARD, (cl_float)1.0f ), "clfftSetPlanScale failed" );
OPENCL_V_THROW( clfftBakePlan( plan_handle, 1, &queue, NULL, NULL ), "clfftBakePlan failed" );
//get the buffersize
size_t buffersize=0;
OPENCL_V_THROW( clfftGetTmpBufSize(plan_handle, &buffersize ), "clfftGetTmpBufSize failed" );
//allocate the intermediate buffer
cl_mem clMedBuffer=NULL;
if (buffersize)
{
cl_int medstatus;
clMedBuffer = clCreateBuffer ( context, CL_MEM_READ_WRITE, buffersize, 0, &medstatus);
OPENCL_V_THROW( medstatus, "Creating intmediate Buffer failed" );
}
switch( in_layout )
{
case CLFFT_COMPLEX_INTERLEAVED:
case CLFFT_COMPLEX_PLANAR:
case CLFFT_HERMITIAN_INTERLEAVED:
case CLFFT_HERMITIAN_PLANAR:
case CLFFT_REAL:
break;
default:
// Don't recognize input layout
return CLFFT_INVALID_ARG_VALUE;
}
switch( out_layout )
{
case CLFFT_COMPLEX_INTERLEAVED:
case CLFFT_COMPLEX_PLANAR:
case CLFFT_HERMITIAN_INTERLEAVED:
case CLFFT_HERMITIAN_PLANAR:
case CLFFT_REAL:
break;
default:
// Don't recognize output layout
return CLFFT_INVALID_ARG_VALUE;
}
if (( place == CLFFT_INPLACE )
&& ( in_layout != out_layout )) {
switch( in_layout )
{
case CLFFT_COMPLEX_INTERLEAVED:
{
if( (out_layout == CLFFT_COMPLEX_PLANAR) || (out_layout == CLFFT_HERMITIAN_PLANAR) )
{
throw std::runtime_error( "Cannot use the same buffer for interleaved->planar in-place transforms" );
}
break;
}
case CLFFT_COMPLEX_PLANAR:
{
if( (out_layout == CLFFT_COMPLEX_INTERLEAVED) || (out_layout == CLFFT_HERMITIAN_INTERLEAVED) )
{
throw std::runtime_error( "Cannot use the same buffer for planar->interleaved in-place transforms" );
}
break;
}
case CLFFT_HERMITIAN_INTERLEAVED:
{
if( out_layout != CLFFT_REAL )
{
throw std::runtime_error( "Cannot use the same buffer for interleaved->planar in-place transforms" );
}
break;
}
case CLFFT_HERMITIAN_PLANAR:
{
throw std::runtime_error( "Cannot use the same buffer for planar->interleaved in-place transforms" );
break;
}
case CLFFT_REAL:
{
if( (out_layout == CLFFT_COMPLEX_PLANAR) || (out_layout == CLFFT_HERMITIAN_PLANAR) )
{
throw std::runtime_error( "Cannot use the same buffer for interleaved->planar in-place transforms" );
}
break;
}
}
}
// Loop as many times as the user specifies to average out the timings
//
cl_mem * BuffersOut = ( place == CLFFT_INPLACE ) ? NULL : &output_cl_mem_buffers[ 0 ];
Timer tr;
tr.Start();
for( cl_uint i = 0; i < profile_count; ++i )
{
if( timer ) timer->Start( clFFTID );
OPENCL_V_THROW( clfftEnqueueTransform( plan_handle, dir, 1, &queue, 0, NULL, &outEvent,
&input_cl_mem_buffers[ 0 ], BuffersOut, clMedBuffer ),
"clfftEnqueueTransform failed" );
if( timer ) timer->Stop( clFFTID );
}
OPENCL_V_THROW( clFinish( queue ), "clFinish failed" );
if(clMedBuffer) clReleaseMemObject(clMedBuffer);
double wtime = tr.Sample()/((double)profile_count);
size_t totalLen = 1;
for(int i=0; i<dim; i++) totalLen *= lengths[i];
double opsconst = 5.0 * (double)totalLen * log((double)totalLen) / log(2.0);
if(profile_count > 1)
{
tout << "\nExecution wall time: " << 1000.0*wtime << " ms" << std::endl;
tout << "Execution gflops: " << ((double)batch_size * opsconst)/(1000000000.0*wtime) << std::endl;
}
if( timer && (command_queue_flags & CL_QUEUE_PROFILING_ENABLE) )
{
// Remove all timings that are outside of 2 stddev (keep 65% of samples); we ignore outliers to get a more consistent result
timer->pruneOutliers( 2.0 );
timer->Print( );
timer->Reset( );
}
/*****************/
FreeSharedLibrary( timerLibHandle );
// Read and check output data
// This check is not valid if the FFT is executed multiple times inplace.
//
if (( place == CLFFT_OUTOFPLACE )
|| ( profile_count == 1))
{
bool checkflag= false;
switch( out_layout )
{
case CLFFT_HERMITIAN_INTERLEAVED:
case CLFFT_COMPLEX_INTERLEAVED:
{
std::vector< std::complex< T > > output( outfftBatchSize );
if( place == CLFFT_INPLACE )
{
OPENCL_V_THROW( clEnqueueReadBuffer( queue, input_cl_mem_buffers[ 0 ], CL_TRUE, 0, size_of_input_buffers_in_bytes, &output[ 0 ],
0, NULL, NULL ),
"Reading the result buffer failed" );
}
else
{
OPENCL_V_THROW( clEnqueueReadBuffer( queue, BuffersOut[ 0 ], CL_TRUE, 0, size_of_output_buffers_in_bytes, &output[ 0 ],
0, NULL, NULL ),
"Reading the result buffer failed" );
}
//check output data
for( cl_uint i = 0; i < outfftBatchSize; ++i )
{
if (0 == (i % outfftVectorSizePadded))
{
if (output[i].real() != outfftVectorSize)
{
checkflag = true;
break;
}
}
else
{
if (output[ i ].real() != 0)
{
checkflag = true;
break;
}
}
if (output[ i ].imag() != 0)
{
checkflag = true;
break;
}
}
}
break;
case CLFFT_HERMITIAN_PLANAR:
case CLFFT_COMPLEX_PLANAR:
{
std::valarray< T > real( outfftBatchSize );
std::valarray< T > imag( outfftBatchSize );
if( place == CLFFT_INPLACE )
{
OPENCL_V_THROW( clEnqueueReadBuffer( queue, input_cl_mem_buffers[ 0 ], CL_TRUE, 0, size_of_input_buffers_in_bytes, &real[ 0 ],
0, NULL, NULL ),
"Reading the result buffer failed" );
OPENCL_V_THROW( clEnqueueReadBuffer( queue, input_cl_mem_buffers[ 1 ], CL_TRUE, 0, size_of_input_buffers_in_bytes, &imag[ 0 ],
0, NULL, NULL ),
"Reading the result buffer failed" );
}
else
{
OPENCL_V_THROW( clEnqueueReadBuffer( queue, BuffersOut[ 0 ], CL_TRUE, 0, size_of_output_buffers_in_bytes, &real[ 0 ],
0, NULL, NULL ),
"Reading the result buffer failed" );
OPENCL_V_THROW( clEnqueueReadBuffer( queue, BuffersOut[ 1 ], CL_TRUE, 0, size_of_output_buffers_in_bytes, &imag[ 0 ],
0, NULL, NULL ),
"Reading the result buffer failed" );
}
// Check output data
for( cl_uint i = 0; i < outfftBatchSize; ++i )
{
if (0 == (i % outfftVectorSizePadded))
{
if (real[i] != outfftVectorSize)
{
checkflag = true;
break;
}
}
else
{
if (real[i] != 0)
{
checkflag = true;
break;
}
}
if (imag[i] != 0)
{
checkflag = true;
break;
}
}
}
break;
case CLFFT_REAL:
{
std::valarray< T > real( outfftBatchSize );
if( place == CLFFT_INPLACE )
{
OPENCL_V_THROW( clEnqueueReadBuffer( queue, input_cl_mem_buffers[ 0 ], CL_TRUE, 0, size_of_input_buffers_in_bytes, &real[ 0 ],
0, NULL, NULL ),
"Reading the result buffer failed" );
}
else
{
OPENCL_V_THROW( clEnqueueReadBuffer( queue, BuffersOut[ 0 ], CL_TRUE, 0, size_of_output_buffers_in_bytes, &real[ 0 ],
0, NULL, NULL ),
"Reading the result buffer failed" );
}
////check output data
for(size_t b = 0; b < batch_size; b++)
{
size_t p3 = b * o_strides[3];
for(size_t k = 0; k < lengths[2]; k++)
{
size_t p2 = p3 + k * o_strides[2];
for(size_t j = 0; j < lengths[1]; j++)
{
size_t p1 = p2 + j * o_strides[1];
for(size_t i = 0; i < lengths[0]; i++)
{
size_t p0 = p1 + i * o_strides[0];
if (real[p0] != 1)
{
checkflag = true;
break;
}
}
}
}
}
}
break;
default:
{
throw std::runtime_error( "Input layout format not yet supported" );
}
break;
}
if (checkflag)
{
std::cout << "\n\n\t\tInternal Client Test *****FAIL*****" << std::endl;
}
else
{
std::cout << "\n\n\t\tInternal Client Test *****PASS*****" << std::endl;
}
}
OPENCL_V_THROW( clfftDestroyPlan( &plan_handle ), "clfftDestroyPlan failed" );
OPENCL_V_THROW( clfftTeardown( ), "clfftTeardown failed" );
cleanupCL( &context, &queue, countOf( input_cl_mem_buffers ), input_cl_mem_buffers, countOf( output_cl_mem_buffers ), output_cl_mem_buffers, &outEvent );
return 0;
}
int main()
{
//Control Variables
bool showStartInput=false;// Setting it to true shows the original Input
bool showFftOutput=false;// Shows the output after the FFT but before the Reshuffle
bool showReshuffleOutput=false;// Shows the output after the reshuffle
bool showFinalResult=false; // Shows final result after cross-correlation
bool showGemmInput=false; // Shows output after the reshuffle but before the matrix multiplication
bool showReformatOutputAfterReshuffle=false; // Shows output after it has been reformatted after the reshuffling
//openCL State
cl_platform_id platform_id=NULL;
cl_device_id device_id=NULL;
cl_context context=NULL;
cl_command_queue queue=NULL;
cl_program program=NULL;
cl_kernel kernel=NULL;
cl_uint ret_num_devices;
cl_uint ret_num_platforms;
cl_int ret=0; // Stores the error values retuned by many functions
cl_event event = NULL;
cl_event events[10];
cl_kernel clKernel;
//FFT state
clAmdFftPlanHandle plHandle;
clAmdFftResultLocation place = CLFFT_OUTOFPLACE; //Alternative CLFFT_INPLACE
clAmdFftLayout inLayout = CLFFT_COMPLEX_INTERLEAVED;
clAmdFftLayout outLayout = CLFFT_COMPLEX_INTERLEAVED;
clAmdFftDim dim = CLFFT_1D;
size_t clStrides[3]={0,0,0};
size_t clLengths[3];
clLengths[0]=(MEM_SIZE/2);//Length of first dimension of fft
clLengths[1]=1;//length of second dimension of fft
clLengths[2]=1;
clStrides[ 0 ] = 1;
clStrides[ 1 ] = clStrides[ 0 ] * clLengths[ 0 ];
clStrides[ 2 ] = clStrides[ 1 ] * clLengths[ 1 ];
clStrides[ 3 ] = clStrides[ 2 ] * clLengths[ 2 ];
size_t batchSize=CHANSIZE;//number of discreet fft's to be calculated simultaneously
//Initialise openCL
OPENCL_V_THROW(clGetPlatformIDs(1, &platform_id, &ret_num_platforms),"clGetPlatformIDs Failed");
OPENCL_V_THROW(clGetDeviceIDs(platform_id, CL_DEVICE_TYPE_GPU, 1, &device_id,&ret_num_devices),"clGetDeviceIDs Failed");
context = clCreateContext(NULL, 1, &device_id, NULL, NULL, &ret);
OPENCL_V_THROW(ret, "Creating Context failed" );
queue = clCreateCommandQueue(context, device_id, 0, &ret);
OPENCL_V_THROW(ret, "Creating command queue failed" );
//===========Initialise the host buffers======================================
/*
* The functions sgenerate2darray(), screate2darray() and sgenerate2darrayout() are defined and declared in definition.h
*/
float** src_a_h=sgenerate2darray(NO_INPUTS,MEM_SIZE);//To be used to store the original input
float** answer=screate2darray(NO_INPUTS,MEM_SIZE);//To be used to store the answer after the reshuffling
float** corr_h=sgenerate2darrayout(NO_INPUTS,CHANSIZE << 1,CHANNELNO);// To be used to store the final answer
if(showStartInput){
cout << "Initial Input Buffer" << "\n";
for(int j=0;j<NO_INPUTS;j++){
for(int i=0;i<MEM_SIZE;i++){
cout << src_a_h[j][i] << " ";
}cout << "\n";
}printf("\n");
}
//===================================================================
//Calculation of facs for reshuffling
complex <float>* facs_h=(complex <float>*) malloc(sizeof(complex <float>)*(MEM_SIZE/2));
complex<float> I=1.0i;
complex <float> xx=2.0*PI;
for(int i=0;i<MEM_SIZE/2;i++){
facs_h[i]=(1.0*i)/(1.0*MEM_SIZE);
facs_h[i]=exp(xx*(-I*facs_h[i]));
}
//===================================================================
//Initialise GPU memory buffers
size_t sizeofgpumem=NO_INPUTS*MEM_SIZE*sizeof(float);
size_t sizeoffacsmem=MEM_SIZE*sizeof(float);
cl_mem clMemBuffersIn = clCreateBuffer(context,CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR,sizeofgpumem,src_a_h[0],&ret);
OPENCL_V_THROW( ret, "Creating clMemBuffersIn Buffer failed" );
cl_mem clMemBuffersOut = clCreateBuffer(context,CL_MEM_READ_WRITE,sizeofgpumem,NULL,&ret);
OPENCL_V_THROW (ret, "Creating fft output Buffer failed");
cl_mem facs = clCreateBuffer(context,CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,sizeoffacsmem,facs_h,&ret);
OPENCL_V_THROW (ret, "Creating facs Buffer failed");
//===========================Starting the fft=============================//
clAmdFftSetupData setupData;
OPENCL_V_THROW( clAmdFftInitSetupData( &setupData ),"clAmdFftInitSetupData failed" );
OPENCL_V_THROW( clAmdFftSetup( &setupData ), "clAmdFftSetup failed" );
OPENCL_V_THROW( clAmdFftCreateDefaultPlan( &plHandle, context, dim, clLengths ), "clAmdFftCreateDefaultPlan failed" );
OPENCL_V_THROW (clAmdFftSetPlanBatchSize (plHandle, batchSize),"Setting BatchSize Failed");
OPENCL_V_THROW (clAmdFftSetResultLocation( plHandle, place ), "clAmdFftSetResultLocation failed" );
OPENCL_V_THROW (clAmdFftSetPlanInStride ( plHandle, dim, clStrides ), "clAmdFftSetPlanInStride failed" );
OPENCL_V_THROW (clAmdFftSetPlanOutStride ( plHandle, dim, clStrides ), "clAmdFftSetPlanOutStride failed" );
OPENCL_V_THROW (clAmdFftSetPlanDistance ( plHandle, clStrides[ dim ], clStrides[ dim ]), "clAmdFftSetPlanDistance failed" );
OPENCL_V_THROW( clAmdFftBakePlan( plHandle, 1, &queue, NULL, NULL ), "clAmdFftBakePlan failed" );
size_t tempbuffersize=0;
OPENCL_V_THROW( clAmdFftGetTmpBufSize(plHandle, &tempbuffersize ), "clAmdFftGetTmpBufSize failed" );
//allocate the intermediate buffer
cl_mem clMedBuffer=NULL;
if (tempbuffersize)
{
cl_int medstatus;
clMedBuffer = clCreateBuffer ( context, CL_MEM_READ_WRITE | CL_MEM_ALLOC_HOST_PTR,tempbuffersize, 0, &medstatus);
OPENCL_V_THROW( medstatus, "Creating fft intermediate Buffer failed" );
}
if (( place == CLFFT_INPLACE )&& ( inLayout != outLayout )) {
switch( inLayout )
{
case CLFFT_COMPLEX_INTERLEAVED:
{
assert (CLFFT_COMPLEX_PLANAR == outLayout);
throw std::runtime_error( "Cannot use the same buffer for interleaved->planar in-place transforms" );
break;
}
case CLFFT_COMPLEX_PLANAR:
{
assert (CLFFT_COMPLEX_INTERLEAVED == outLayout);
throw std::runtime_error( "Cannot use the same buffer for planar->interleaved in-place transforms" );
break;
}
}
}
cl_mem * BuffersOut = ( place == CLFFT_INPLACE ) ? NULL : &clMemBuffersOut;
//========Timimg fft============//
double time_fft_start=omp_get_wtime();
for(int i=0;i<ITER_FFT;i++){
OPENCL_V_THROW( clAmdFftEnqueueTransform( plHandle, CLFFT_FORWARD, 1,&queue,0,NULL,&event,&clMemBuffersIn,BuffersOut,clMedBuffer ),"clAmdFftEnqueueTransform failed" );
}
ret=clWaitForEvents(1,&event);
double time_fft_end=omp_get_wtime();
//Cleaning up fft
OPENCL_V_THROW( clAmdFftDestroyPlan( &plHandle ), "clAmdFftDestroyPlan failed" );
OPENCL_V_THROW( clAmdFftTeardown( ), "clAmdFftTeardown failed" );
//displaying results
if(showFftOutput){
OPENCL_V_THROW( clEnqueueReadBuffer( queue, clMemBuffersOut, CL_TRUE, 0, sizeofgpumem,answer [0], 0, NULL, NULL ),"Reading the result buffer failed" );
cout << "**FFT Output**" << endl;
for(int j=0;j<NO_INPUTS;j++){
for(int i=0;i<MEM_SIZE;i++){
cout << answer[j][i] << " ";
} printf("\n");
}printf("\n");
}
//==================End of FFT=============================================//
//==================Start the Reshuffling==================================//
FILE *fp;
char fileName[]="./reshuffle.cl";
char* source_str=NULL;
size_t source_size;
//Load the source code containing the kernel/
fp = fopen(fileName, "r");
if (!fp) {
fprintf(stderr, "Failed to load reshuffle kernel.¥n");
exit(1);
}
source_str = (char*)malloc(MAX_SOURCE_SIZE);
source_size = fread(source_str, 1, MAX_SOURCE_SIZE, fp);
fclose(fp);
//Preparation for building the Kernel
program = clCreateProgramWithSource(context, 1, (const char **)&source_str,(const size_t *)&source_size, &ret);
OPENCL_V_THROW( ret, "Creating program with source failed for Reshuffle" );
OPENCL_V_THROW( clBuildProgram(program, 1, &device_id, NULL, NULL, NULL),"Build Program Failed for Reshuffle");
kernel = clCreateKernel(program, "reshuffle", &ret);
OPENCL_V_THROW( ret, "Creating kernel failed for Reshuffle" );
//Set kernel parameters
const int num=NO_INPUTS*MEM_SIZE;
const int block=MEM_SIZE;
OPENCL_V_THROW(clSetKernelArg(kernel, 0, sizeof(cl_mem), (float *)&clMemBuffersIn),"Passing argument 0 of reshuffle failed");
OPENCL_V_THROW(clSetKernelArg(kernel, 1, sizeof(cl_mem), (float *)&facs),"Passing arg 1 of reshuffle failed");
OPENCL_V_THROW(clSetKernelArg(kernel, 2, sizeof(cl_mem), (float *)&clMemBuffersOut),"Passing arg2 of reshuffle failed");
OPENCL_V_THROW(clSetKernelArg(kernel, 3, sizeof(int), (int *)&num),"Passing arg3 of reshuffle failed");
OPENCL_V_THROW(clSetKernelArg(kernel, 4, sizeof(int), (int *)&block),"Passing arg4 of reshuffle failed");
// Execute OpenCL Kernel //
const size_t local_ws=NO_INPUTS*MEM_SIZE;
const size_t global_ws=min(NO_THREAD_PER_BLOCK,MEM_SIZE);//ceil(MEM_SIZE/local_ws);
//===========timing the reshuffle===============//
double time_reshuffle_start=omp_get_wtime();
for(int i=0;i<ITER_FFT;i++){
OPENCL_V_THROW(clEnqueueNDRangeKernel(queue,kernel, 1, NULL,&local_ws,&global_ws, 0, NULL, NULL),"Reshuffle Kernel execution failed");
}
double time_reshuffle_end=omp_get_wtime();
//Read back data
OPENCL_V_THROW(clEnqueueReadBuffer(queue, clMemBuffersOut, CL_TRUE, 0, sizeofgpumem,answer[0], 0, NULL, NULL),"Reading back reshuffled data failed");
//====================Finish the reshuffling================================//
if(showReshuffleOutput){
cout << "Output after reshuffling" << endl;
for(int j=0;j<NO_INPUTS;j++){
for(int i=0;i<MEM_SIZE;i++){
cout << answer[j][i] << " ";
} printf("\n");
}printf("\n");
}
//=================Reformatting the input given to the matrix multiply===================================//
float** answer_final=screate2darray(NO_INPUTS*2,MEM_SIZE/2);
for(int i=0;i<NO_INPUTS;i++){
for(int j=0;j<MEM_SIZE;j++){
if(j&1)
answer_final[(i<<1)+1][j >> 1]=answer[i][j];
else
answer_final[(i<<1)][j >> 1]=answer[i][j];
}
}
