diff --git a/RTCP/Cobalt/GPUProc/src/CMakeLists.txt b/RTCP/Cobalt/GPUProc/src/CMakeLists.txt
index 8bd1194b58b177ef45e8127e37a127dd075288d1..36a70dd791a5b48cb5ab48c44d161b56bfb6c876 100644
--- a/RTCP/Cobalt/GPUProc/src/CMakeLists.txt
+++ b/RTCP/Cobalt/GPUProc/src/CMakeLists.txt
@@ -13,7 +13,7 @@ set(_gpuproc_sources
BlockID.cc
FilterBank.cc
global_defines.cc
- Kernels/CompileDefinitions.cc
+ KernelCompiler.cc
Storage/SSH.cc
Storage/StorageProcess.cc
Storage/StorageProcesses.cc
@@ -24,6 +24,7 @@ if(USE_CUDA)
cuda/gpu_wrapper.cc
cuda/gpu_utils.cc
cuda/CudaRuntimeCompiler.cc
+ cuda/KernelCompiler.cc
cuda/PerformanceCounter.cc
cuda/Kernels/Kernel.cc
cuda/Kernels/BeamFormerKernel.cc
diff --git a/RTCP/Cobalt/GPUProc/src/Kernels/Kernel.h b/RTCP/Cobalt/GPUProc/src/Kernels/Kernel.h
index f4cd48822e86c8a5aef2177614fe2408ba8f60a7..1ba3d8edd7405dc20829ed541d638f5cb1702ff2 100644
--- a/RTCP/Cobalt/GPUProc/src/Kernels/Kernel.h
+++ b/RTCP/Cobalt/GPUProc/src/Kernels/Kernel.h
@@ -28,7 +28,7 @@
#include <CoInterface/Parset.h>
#include <GPUProc/gpu_wrapper.h>
-#include <GPUProc/Kernels/CompileDefinitions.h>
+#include <GPUProc/KernelCompiler.h>
//#include <GPUProc/PerformanceCounter.h>
namespace LOFAR
@@ -43,7 +43,7 @@ namespace LOFAR
void enqueue(gpu::Stream &queue/*, PerformanceCounter &counter*/);
// Return required compile definitions given the Parset \a ps.
- static CompileDefinitions compileDefinitions(const Parset& ps);
+ static const CompileDefinitions& compileDefinitions(const Parset& ps);
protected:
gpu::Event event;
diff --git a/RTCP/Cobalt/GPUProc/src/cuda/CudaRuntimeCompiler.cc b/RTCP/Cobalt/GPUProc/src/cuda/CudaRuntimeCompiler.cc
index eb534673c1e3ba24714bd441625354c1e1709ca2..d07c01283168e8158f8df10a1c49725e06997f87 100644
--- a/RTCP/Cobalt/GPUProc/src/cuda/CudaRuntimeCompiler.cc
+++ b/RTCP/Cobalt/GPUProc/src/cuda/CudaRuntimeCompiler.cc
@@ -93,7 +93,7 @@ namespace LOFAR
// of definitions. Use this command to call nvcc and compile the file at input path to a ptx file
// which content is returned as a string
std::string compileToPtx(const std::string& pathToCuFile,
- const flags_type& flags,
+ const CompileFlags& flags,
const CompileDefinitions& definitions)
{
const string cudaCompiler = "nvcc";
@@ -102,10 +102,7 @@ namespace LOFAR
cmd << " " << pathToCuFile ;
cmd << " --ptx";
cmd << definitions;
-
- // add the set of flags
- for (flags_type::const_iterator it=flags.begin(); it!=flags.end(); ++it)
- cmd << " --" << *it; // flags should be prepended with a space and a minus
+ cmd << flags;
// output to stdout
cmd << " -o -";
diff --git a/RTCP/Cobalt/GPUProc/src/cuda/CudaRuntimeCompiler.h b/RTCP/Cobalt/GPUProc/src/cuda/CudaRuntimeCompiler.h
index 41741f617e23e58293c6462c12a485377dd84c6f..e673eef4d43c28e2660761c815ffc83ec1ec9f5a 100644
--- a/RTCP/Cobalt/GPUProc/src/cuda/CudaRuntimeCompiler.h
+++ b/RTCP/Cobalt/GPUProc/src/cuda/CudaRuntimeCompiler.h
@@ -30,7 +30,7 @@
#include <set>
#include <CoInterface/Parset.h>
-#include <GPUProc/Kernels/CompileDefinitions.h>
+#include <GPUProc/KernelCompiler.h>
#include <GPUProc/gpu_utils.h>
#include "gpu_incl.h"
@@ -41,29 +41,18 @@ namespace LOFAR
namespace Cobalt
{
- // // flags
- // typedef std::set<std::string> flags_type;
-
- // Return the set of default flags for the nvcc compilation of a cuda kernel in Cobalt
- flags_type defaultFlags();
-
- // // Return the set of default definitions for the nvcc compilation of a cuda kernel in Cobalt
- // // DEFINITION=0 results in a definition and cannot for used for undef. Do not insert parameters in this case
- // definitions_type defaultDefinitions();
-
// Performs a 'system' call of nvcc. Return the stdout of the command
// on error no stdout is created and an exception is thrown
std::string runNVCC(const std::string &cmd);
- // Create a nvcc command line string based on the input path, a set of flags and a map
- // of definitions. Use this command to call nvcc and compile the file at input path to a ptx file
- // which content is returned as a string
+ // Create a nvcc command line string based on the input path, a set of flags
+ // and a map of definitions. Use this command to call nvcc and compile the
+ // file at input path to a ptx file which content is returned as a string
std::string
compileToPtx(const std::string& pathToCuFile,
- const flags_type& flags = defaultFlags(),
+ const CompileFlags& flags = CompileFlags(),
const CompileDefinitions& definitions = CompileDefinitions());
- // overloaded function. Use the path and default flags and definitions to call nvcc
}
}
diff --git a/RTCP/Cobalt/GPUProc/src/cuda/Kernels/Kernel.cc b/RTCP/Cobalt/GPUProc/src/cuda/Kernels/Kernel.cc
index 74404c1f9f9eb25ec68d29e16078da3ef442df9f..a8f2da23826f116f039a6e29ac2835897d9d3974 100644
--- a/RTCP/Cobalt/GPUProc/src/cuda/Kernels/Kernel.cc
+++ b/RTCP/Cobalt/GPUProc/src/cuda/Kernels/Kernel.cc
@@ -59,42 +59,43 @@ namespace LOFAR
// counter.doOperation(event, nrOperations, nrBytesRead, nrBytesWritten);
}
- CompileDefinitions Kernel::compileDefinitions(const Parset& ps)
+ const CompileDefinitions& Kernel::compileDefinitions(const Parset& ps)
{
- CompileDefinitions defs;
+ static CompileDefinitions defs;
using boost::format;
- defs["NVIDIA_CUDA"] = ""; // left-over from OpenCL for Correlator.cl/.cu
-
- // TODO: support device specific defs somehow (createPTX() knows about targets, but may be kernel and target specific)
- //if (devices[0].getInfo<CL_DEVICE_NAME>() == "GeForce GTX 680")
- // defs["USE_FLOAT4_IN_CORRELATOR"] = "";
-
- // TODO: kernel-specific defs should be specified in the XXXKernel class
- defs["COMPLEX"] = "2";
-
- defs["NR_BITS_PER_SAMPLE"] = str(format("%u") % ps.nrBitsPerSample());
- defs["SUBBAND_BANDWIDTH"] = str(format("%.7ff") % ps.subbandBandwidth()); // returns double, so rounding issue?
- defs["NR_SUBBANDS"] = str(format("%u") % ps.nrSubbands()); // size_t, but %zu not supp
- defs["NR_CHANNELS"] = str(format("%u") % ps.nrChannelsPerSubband());
- defs["NR_STATIONS"] = str(format("%u") % ps.nrStations());
- defs["NR_SAMPLES_PER_CHANNEL"] = str(format("%u") % ps.nrSamplesPerChannel());
- defs["NR_SAMPLES_PER_SUBBAND"] = str(format("%u") % ps.nrSamplesPerSubband());
- defs["NR_BEAMS"] = str(format("%u") % ps.nrBeams());
- defs["NR_TABS"] = str(format("%u") % ps.nrTABs(0)); // TODO: 0 should be dep on #beams
- defs["NR_COHERENT_STOKES"] = str(format("%u") % ps.nrCoherentStokes()); // size_t
- defs["NR_INCOHERENT_STOKES"] = str(format("%u") % ps.nrIncoherentStokes()); // size_t
- defs["COHERENT_STOKES_TIME_INTEGRATION_FACTOR"] = str(format("%u") % ps.coherentStokesTimeIntegrationFactor());
- defs["INCOHERENT_STOKES_TIME_INTEGRATION_FACTOR"] = str(format("%u") % ps.incoherentStokesTimeIntegrationFactor());
- defs["NR_POLARIZATIONS"] = str(format("%u") % NR_POLARIZATIONS);
- defs["NR_TAPS"] = str(format("%u") % NR_TAPS);
- defs["NR_STATION_FILTER_TAPS"] = str(format("%u") % NR_STATION_FILTER_TAPS);
- if (ps.delayCompensation())
- defs["DELAY_COMPENSATION"] = "";
- if (ps.correctBandPass())
- defs["BANDPASS_CORRECTION"] = "";
- defs["DEDISPERSION_FFT_SIZE"] = str(format("%u") % ps.dedispersionFFTsize()); // size_t
+ if (defs.empty()) {
+ defs["NVIDIA_CUDA"] = ""; // left-over from OpenCL for Correlator.cl/.cu
+ // TODO: support device specific defs somehow (createPTX() knows about targets, but may be kernel and target specific)
+ //if (devices[0].getInfo<CL_DEVICE_NAME>() == "GeForce GTX 680")
+ // defs["USE_FLOAT4_IN_CORRELATOR"] = "";
+
+ // TODO: kernel-specific defs should be specified in the XXXKernel class
+ defs["COMPLEX"] = "2";
+
+ defs["NR_BITS_PER_SAMPLE"] = str(format("%u") % ps.nrBitsPerSample());
+ defs["SUBBAND_BANDWIDTH"] = str(format("%.7ff") % ps.subbandBandwidth()); // returns double, so rounding issue?
+ defs["NR_SUBBANDS"] = str(format("%u") % ps.nrSubbands()); // size_t, but %zu not supp
+ defs["NR_CHANNELS"] = str(format("%u") % ps.nrChannelsPerSubband());
+ defs["NR_STATIONS"] = str(format("%u") % ps.nrStations());
+ defs["NR_SAMPLES_PER_CHANNEL"] = str(format("%u") % ps.nrSamplesPerChannel());
+ defs["NR_SAMPLES_PER_SUBBAND"] = str(format("%u") % ps.nrSamplesPerSubband());
+ defs["NR_BEAMS"] = str(format("%u") % ps.nrBeams());
+ defs["NR_TABS"] = str(format("%u") % ps.nrTABs(0)); // TODO: 0 should be dep on #beams
+ defs["NR_COHERENT_STOKES"] = str(format("%u") % ps.nrCoherentStokes()); // size_t
+ defs["NR_INCOHERENT_STOKES"] = str(format("%u") % ps.nrIncoherentStokes()); // size_t
+ defs["COHERENT_STOKES_TIME_INTEGRATION_FACTOR"] = str(format("%u") % ps.coherentStokesTimeIntegrationFactor());
+ defs["INCOHERENT_STOKES_TIME_INTEGRATION_FACTOR"] = str(format("%u") % ps.incoherentStokesTimeIntegrationFactor());
+ defs["NR_POLARIZATIONS"] = str(format("%u") % NR_POLARIZATIONS);
+ defs["NR_TAPS"] = str(format("%u") % NR_TAPS);
+ defs["NR_STATION_FILTER_TAPS"] = str(format("%u") % NR_STATION_FILTER_TAPS);
+ if (ps.delayCompensation())
+ defs["DELAY_COMPENSATION"] = "";
+ if (ps.correctBandPass())
+ defs["BANDPASS_CORRECTION"] = "";
+ defs["DEDISPERSION_FFT_SIZE"] = str(format("%u") % ps.dedispersionFFTsize()); // size_t
+ }
return defs;
}
diff --git a/RTCP/Cobalt/GPUProc/src/cuda/Pipelines/Pipeline.cc b/RTCP/Cobalt/GPUProc/src/cuda/Pipelines/Pipeline.cc
index 3e5f4207b0c5dee89b377e3855147130bfb8e1dd..8401f81b02e556cff484f0e3d20c78dbaba8e8c3 100644
--- a/RTCP/Cobalt/GPUProc/src/cuda/Pipelines/Pipeline.cc
+++ b/RTCP/Cobalt/GPUProc/src/cuda/Pipelines/Pipeline.cc
@@ -46,7 +46,7 @@ namespace LOFAR
std::string Pipeline::createPTX(const string &srcFilename)
{
- flags_type flags(defaultFlags());
+ CompileFlags flags(defaultCompileFlags());
CompileDefinitions definitions(Kernel::compileDefinitions(ps));
return LOFAR::Cobalt::createPTX(devices, srcFilename, flags, definitions);
diff --git a/RTCP/Cobalt/GPUProc/src/cuda/gpu_utils.cc b/RTCP/Cobalt/GPUProc/src/cuda/gpu_utils.cc
index 086e4dbee4f63ae8abcc4637c55aed4397f2f925..944ead2b63318ada8ca87aabf620ec7d94c99490 100644
--- a/RTCP/Cobalt/GPUProc/src/cuda/gpu_utils.cc
+++ b/RTCP/Cobalt/GPUProc/src/cuda/gpu_utils.cc
@@ -39,7 +39,6 @@
#include <Stream/FileStream.h>
#include <GPUProc/global_defines.h>
-#include <GPUProc/Kernels/CompileDefinitions.h>
#include "CudaRuntimeCompiler.h"
#define BUILD_MAX_LOG_SIZE 4095
@@ -51,22 +50,6 @@ namespace LOFAR
using namespace std;
using boost::format;
- flags_type defaultFlags()
- {
- flags_type flags;
-
- using boost::format;
-
- //flags.insert("device-debug");
-
- // TODO: If this simplifies trig funcs, verify effect on the
- // BPDelayKernel. Ideally, a Kernel specifies when to enable build flags.
- flags.insert("use_fast_math");
-
- return flags;
- };
-
-
namespace {
// Return the highest compute target supported by the given device
@@ -214,14 +197,14 @@ namespace LOFAR
std::string createPTX(const vector<gpu::Device> &devices,
const std::string &srcFilename,
- flags_type &flags,
+ CompileFlags &flags,
const CompileDefinitions &definitions )
{
// The CUDA code is assumed to be written for the architecture of the
// oldest device.
#if CUDA_VERSION >= 5000
CUjit_target commonTarget = computeTarget(devices);
- flags.insert(str(format("gpu-architecture %s") % get_virtarch(commonTarget)));
+ flags.add(str(format("--gpu-architecture %s") % get_virtarch(commonTarget)));
#endif
#if 0
@@ -230,18 +213,18 @@ namespace LOFAR
set<CUjit_target> allTargets;
for (vector<gpu::Device>::const_iterator i = devices.begin(); i != devices.end(); ++i) {
- allTargets.insert(computeTarget(*i));
+ allTargets.add(computeTarget(*i));
}
for (set<CUjit_target>::const_iterator i = allTargets.begin(); i != allTargets.end(); ++i) {
- flags.insert(str(format("gpu-code %s") % get_gpuarch(*i)));
+ flags.add(str(format("--gpu-code %s") % get_gpuarch(*i)));
}
#endif
// Add $LOFARROOT/include to include path, if $LOFARROOT is set.
const char* lofarroot = getenv("LOFARROOT");
if (lofarroot) {
- flags.insert(str(format("include-path %s/include") % lofarroot));
+ flags.add(str(format("--include-path %s/include") % lofarroot));
}
// Prefix the CUDA kernel filename with $LOFARROOT/share/gpu/kernels
diff --git a/RTCP/Cobalt/GPUProc/src/gpu_utils.h b/RTCP/Cobalt/GPUProc/src/gpu_utils.h
index 45bd1fae4c258a0cb46fc583a2f27fedc70b1225..d7901eba4df93420dcd49f28f2937cfb6e03c79d 100644
--- a/RTCP/Cobalt/GPUProc/src/gpu_utils.h
+++ b/RTCP/Cobalt/GPUProc/src/gpu_utils.h
@@ -25,6 +25,7 @@
#include <vector>
#include <CoInterface/Parset.h>
+#include <GPUProc/KernelCompiler.h>
#include "gpu_wrapper.h"
// #include "CudaRuntimeCompiler.h"
@@ -32,14 +33,6 @@ namespace LOFAR
{
namespace Cobalt
{
- // flags
- typedef std::set<std::string> flags_type;
-
- // Return default flags to be used for kernel compilation. The
- // implementation of this method is CUDA/OpenCL specific.
- flags_type defaultFlags();
-
- class CompileDefinitions;
/*
* If no devices are given, the program is compiled for the latest
* architecture.
@@ -48,7 +41,7 @@ namespace LOFAR
*/
std::string createPTX( const std::vector<gpu::Device> &devices,
const std::string &srcFilename,
- flags_type &flags,
+ CompileFlags &flags,
const CompileDefinitions &definitions );
/*
* Create a Module from a PTX (string).
diff --git a/RTCP/Cobalt/GPUProc/test/cuda/tCorrelator.cc b/RTCP/Cobalt/GPUProc/test/cuda/tCorrelator.cc
index 7ada5b32d1880fca82d540f6d73f1494a1673ecf..286fc6ce49c25decf9986859b5036fa73fbe2316 100644
--- a/RTCP/Cobalt/GPUProc/test/cuda/tCorrelator.cc
+++ b/RTCP/Cobalt/GPUProc/test/cuda/tCorrelator.cc
@@ -31,7 +31,7 @@
#include <GPUProc/gpu_wrapper.h>
#include <GPUProc/gpu_utils.h>
-#include <GPUProc/Kernels/CompileDefinitions.h>
+#include <GPUProc/KernelCompiler.h>
#include <GPUProc/cuda/CudaRuntimeCompiler.h>
#include "TestUtil.h"
@@ -61,7 +61,7 @@ HostMemory runTest(gpu::Context ctx,
// Get an instantiation of the default parameters
CompileDefinitions definitions = CompileDefinitions();
- flags_type flags = defaultFlags();
+ CompileFlags flags = defaultCompileFlags();
// ****************************************
// Compile to ptx
diff --git a/RTCP/Cobalt/GPUProc/test/cuda/tCorrelatorWorkQueueProcessSb.cc b/RTCP/Cobalt/GPUProc/test/cuda/tCorrelatorWorkQueueProcessSb.cc
index a575a464340f842fd68978d5e9230d56eb6cf418..fb8f5328d8b3f802c556ddef3e0896f92d6a124f 100644
--- a/RTCP/Cobalt/GPUProc/test/cuda/tCorrelatorWorkQueueProcessSb.cc
+++ b/RTCP/Cobalt/GPUProc/test/cuda/tCorrelatorWorkQueueProcessSb.cc
@@ -25,7 +25,7 @@
#include <Common/LofarLogger.h>
#include <CoInterface/Parset.h>
#include <GPUProc/gpu_utils.h>
-#include <GPUProc/Kernels/Kernel.h>
+#include <GPUProc/KernelCompiler.h>
#include <GPUProc/WorkQueues/CorrelatorWorkQueue.h>
using namespace std;
@@ -61,7 +61,7 @@ int main() {
kernels.push_back(kfilenameCor);
map<string, string> ptx;
- flags_type flags(defaultFlags());
+ CompileFlags flags(defaultCompileFlags());
CompileDefinitions definitions(Kernel::compileDefinitions(ps));
ptx[kfilenameFIR] = createPTX(devices, kfilenameFIR, flags, definitions);
ptx[kfilenameDBP] = createPTX(devices, kfilenameDBP, flags, definitions);
diff --git a/RTCP/Cobalt/GPUProc/test/cuda/tCudaRuntimeCompiler.cc b/RTCP/Cobalt/GPUProc/test/cuda/tCudaRuntimeCompiler.cc
index d041e3a27f9b95b129edad04798d81cbfecc930e..92e04e5296f1fdface1b5d59ad4298e138129e91 100644
--- a/RTCP/Cobalt/GPUProc/test/cuda/tCudaRuntimeCompiler.cc
+++ b/RTCP/Cobalt/GPUProc/test/cuda/tCudaRuntimeCompiler.cc
@@ -21,6 +21,7 @@
#include <lofar_config.h>
#include <string>
+#include <GPUProc/KernelCompiler.h>
#include <GPUProc/cuda/CudaRuntimeCompiler.h>
#include <Common/LofarLogger.h>
@@ -40,10 +41,10 @@ int main()
// override the default with a magic number
definitions["NVIDIA_CUDA"] = "123456";
- string ptx1 = compileToPtx(kernelPath, defaultFlags(), definitions);
+ string ptx1 = compileToPtx(kernelPath, defaultCompileFlags(), definitions);
definitions["NVIDIA_CUDA"] = "654321";
- string ptx2 = compileToPtx(kernelPath, defaultFlags(), definitions);
+ string ptx2 = compileToPtx(kernelPath, defaultCompileFlags(), definitions);
// tests if the magic numbers are inserted into the ptx files
if ((std::string::npos != ptx1.find("123456")) ||
diff --git a/RTCP/Cobalt/GPUProc/test/cuda/tDelayAndBandPass.cc b/RTCP/Cobalt/GPUProc/test/cuda/tDelayAndBandPass.cc
index d7b101cb0f7e1a9f502524e7d66bc65f46dce1eb..1319e625b097a59df4a51311466cc676ebcfb9d2 100644
--- a/RTCP/Cobalt/GPUProc/test/cuda/tDelayAndBandPass.cc
+++ b/RTCP/Cobalt/GPUProc/test/cuda/tDelayAndBandPass.cc
@@ -32,6 +32,7 @@
#include <GPUProc/gpu_wrapper.h>
#include <GPUProc/gpu_utils.h>
+#include <GPUProc/KernelCompiler.h>
#include <GPUProc/cuda/CudaRuntimeCompiler.h>
#include <UnitTest++.h>
@@ -68,7 +69,7 @@ float * runTest(float bandPassFactor,
// Get an instantiation of the default parameters
CompileDefinitions definitions;
- flags_type flags = defaultFlags();
+ CompileFlags flags = defaultCompileFlags();
// ****************************************
// Compile to ptx
diff --git a/RTCP/Cobalt/GPUProc/test/cuda/tDelayAndBandPassKernel.cc b/RTCP/Cobalt/GPUProc/test/cuda/tDelayAndBandPassKernel.cc
index f887237a9f53fd34faf7c5df1f935922b2a35e50..3e4dcb386d8da1eaf87ebf33aab3a33d98a79c9d 100644
--- a/RTCP/Cobalt/GPUProc/test/cuda/tDelayAndBandPassKernel.cc
+++ b/RTCP/Cobalt/GPUProc/test/cuda/tDelayAndBandPassKernel.cc
@@ -25,6 +25,7 @@
#include <GPUProc/gpu_wrapper.h>
#include <GPUProc/gpu_utils.h>
#include <GPUProc/BandPass.h>
+#include <GPUProc/KernelCompiler.h>
#include <GPUProc/Kernels/DelayAndBandPassKernel.h>
#include <GPUProc/WorkQueues/CorrelatorWorkQueue.h>
@@ -51,7 +52,7 @@ int main() {
string srcFilename("DelayAndBandPass.cu");
// Get default parameters for the compiler
- flags_type flags = defaultFlags();
+ CompileFlags flags = defaultCompileFlags();
CompileDefinitions definitions(Kernel::compileDefinitions(ps));
string ptx = createPTX(devices, srcFilename, flags, definitions);
diff --git a/RTCP/Cobalt/GPUProc/test/cuda/tIntToFloat.cc b/RTCP/Cobalt/GPUProc/test/cuda/tIntToFloat.cc
index 5513c09032bfbd82237bf31fd3c600b081802cf2..0f0a59dd9ce9798e9207bef5d40118e4f5f4913f 100644
--- a/RTCP/Cobalt/GPUProc/test/cuda/tIntToFloat.cc
+++ b/RTCP/Cobalt/GPUProc/test/cuda/tIntToFloat.cc
@@ -32,6 +32,7 @@
#include <GPUProc/gpu_wrapper.h>
#include <GPUProc/gpu_utils.h>
+#include <GPUProc/KernelCompiler.h>
#include <GPUProc/cuda/CudaRuntimeCompiler.h>
#include <UnitTest++.h>
@@ -66,7 +67,7 @@ float * runTest(unsigned NR_BITS_PER_SAMPLE = 16,
// Compile to ptx
// Get an instantiation of the default parameters
CompileDefinitions definitions;
- flags_type flags = defaultFlags();
+ CompileFlags flags = defaultCompileFlags();
// Set op string string pairs to be provided to the compiler as defines
definitions["NR_STATIONS"] = "2";
diff --git a/RTCP/Cobalt/GPUProc/test/cuda/tIntToFloatKernel.cc b/RTCP/Cobalt/GPUProc/test/cuda/tIntToFloatKernel.cc
index 584de7f4f347d4d7c53ee5a5e4e6ee0343e91258..1ac55859717cff57992295f9257b31364b197546 100644
--- a/RTCP/Cobalt/GPUProc/test/cuda/tIntToFloatKernel.cc
+++ b/RTCP/Cobalt/GPUProc/test/cuda/tIntToFloatKernel.cc
@@ -25,6 +25,7 @@
#include <GPUProc/gpu_wrapper.h>
#include <GPUProc/gpu_utils.h>
#include <GPUProc/BandPass.h>
+#include <GPUProc/KernelCompiler.h>
#include <GPUProc/Kernels/IntToFloatKernel.h>
#include <GPUProc/WorkQueues/CorrelatorWorkQueue.h>
@@ -52,7 +53,7 @@ int main() {
string srcFilename("IntToFloat.cu");
// Get default parameters for the compiler
- flags_type flags = defaultFlags();
+ CompileFlags flags = defaultCompileFlags();
CompileDefinitions definitions(Kernel::compileDefinitions(ps));
string ptx = createPTX(devices, srcFilename, flags, definitions);
diff --git a/RTCP/Cobalt/GPUProc/test/cuda/tKernel.cc b/RTCP/Cobalt/GPUProc/test/cuda/tKernel.cc
index 11f19cccfa1d671b56637b6472667fb16bccb572..d86bc73f79fc652475b42e93cdc047941f439e5b 100644
--- a/RTCP/Cobalt/GPUProc/test/cuda/tKernel.cc
+++ b/RTCP/Cobalt/GPUProc/test/cuda/tKernel.cc
@@ -28,6 +28,7 @@
#include <CoInterface/Parset.h>
#include <GPUProc/Kernels/Kernel.h>
#include <GPUProc/gpu_utils.h>
+#include <GPUProc/KernelCompiler.h>
#include <GPUProc/cuda/CudaRuntimeCompiler.h>
#include <GPUProc/global_defines.h>
#include <Common/LofarLogger.h>
@@ -53,7 +54,7 @@ int main() {
Parset ps("tKernel.parset.in");
// Get default parameters for the compiler
- flags_type flags = defaultFlags();
+ CompileFlags flags = defaultCompileFlags();
CompileDefinitions definitions(Kernel::compileDefinitions(ps));
string ptx = createPTX(devices, srcFilename, flags, definitions);
diff --git a/RTCP/Cobalt/GPUProc/test/cuda/tcreateProgram.cc b/RTCP/Cobalt/GPUProc/test/cuda/tcreateProgram.cc
index e25be71c55e7877dba839bd53e89aee45eaab0cd..8f696325e0374873f703ae6319a0c6b6e03972dd 100644
--- a/RTCP/Cobalt/GPUProc/test/cuda/tcreateProgram.cc
+++ b/RTCP/Cobalt/GPUProc/test/cuda/tcreateProgram.cc
@@ -28,6 +28,7 @@
#include <CoInterface/Parset.h>
#include <GPUProc/gpu_utils.h>
#include <GPUProc/Kernels/Kernel.h>
+#include <GPUProc/KernelCompiler.h>
#include <GPUProc/cuda/CudaRuntimeCompiler.h>
#include <GPUProc/global_defines.h>
@@ -59,7 +60,7 @@ int main(int argc, char *argv[]) {
// Collect inputs from the parste and assign them to CudaRuntimeCompiler
// input_types.
- flags_type flags = defaultFlags();
+ CompileFlags flags = defaultCompileFlags();
CompileDefinitions definitions(Kernel::compileDefinitions(ps));
string ptx = createPTX(devices, srcFilename, flags, definitions);