Merge branch 'multi-gpu-types' into 'master'

Support for multiple GPU types per system. Breaks API!

See merge request !12

libtcc/Correlator.cc

@@ -32,7 +32,8 @@ std::string Correlator::findNVRTCincludePath() const
 }
 
 
-Correlator::Correlator(unsigned nrBits,
+Correlator::Correlator(const cu::Device &device,
+		       unsigned nrBits,
 		       unsigned nrReceivers,
 		       unsigned nrChannels,
 		       unsigned nrSamplesPerChannel,
@@ -41,8 +42,7 @@ Correlator::Correlator(unsigned nrBits,
 		       const std::string &customStoreVisibility
 		      )
 :
-  capability([] { 
-    cu::Device device(0);
+  capability([&] { 
     return 10 * device.getAttribute<CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR>() + device.getAttribute<CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR>();
   } ()),
   nrReceiversPerBlock(nrReceiversPerBlock != 0 ? nrReceiversPerBlock : defaultNrReceiversPerBlock(nrReceivers)),

libtcc/Correlator.h

@@ -11,7 +11,8 @@
 namespace tcc {
   class Correlator {
     public:
-      Correlator(unsigned nrBits,
+      Correlator(const cu::Device &,
+		 unsigned nrBits,
 		 unsigned nrReceivers,
 		 unsigned nrChannels,
 		 unsigned nrSamplesPerChannel,

test/Benchmark/Benchmark.cc

@@ -59,7 +59,7 @@ template <typename SampleType, typename VisibilityType> void Benchmark::doTest(u
             unsigned nrTimesPerBlock     = 128 / nrBits;
 	    unsigned nrBaselines         = nrReceivers * (nrReceivers + 1) / 2;
 
-  tcc::Correlator correlator(nrBits, nrReceivers, nrChannels, nrSamplesPerChannel, nrPolarizations, nrReceiversPerBlock);
+  tcc::Correlator correlator(device, nrBits, nrReceivers, nrChannels, nrSamplesPerChannel, nrPolarizations, nrReceiversPerBlock);
   unsigned repeatCount;
 
   multi_array::extent<5> samplesExtent(multi_array::extents[nrChannels][nrSamplesPerChannel / nrTimesPerBlock][nrReceivers][nrPolarizations][nrTimesPerBlock]);

test/CorrelatorTest/CorrelatorTest.cc

@@ -20,7 +20,7 @@ CorrelatorTest::CorrelatorTest(const Options &options)
   UnitTest(options.deviceNumber),
   options(options),
   hasIntegratedMemory(device.getAttribute(CU_DEVICE_ATTRIBUTE_INTEGRATED)),
-  correlator(options.nrBits, options.nrReceivers, options.nrChannels, options.nrSamplesPerChannel, options.nrPolarizations, options.nrReceiversPerBlock)
+  correlator(device, options.nrBits, options.nrReceivers, options.nrChannels, options.nrSamplesPerChannel, options.nrPolarizations, options.nrReceiversPerBlock)
 {
 #if defined MEASURE_POWER
   Record start(*powerSensor), stop(*powerSensor);

test/SimpleExample/SimpleExample.cu

@@ -51,7 +51,7 @@ int main()
     checkCudaCall(cudaSetDevice(0)); // combine the CUDA runtime API and CUDA driver API
     checkCudaCall(cudaFree(0));
 
-    tcc::Correlator correlator(NR_BITS, NR_RECEIVERS, NR_CHANNELS, NR_SAMPLES_PER_CHANNEL, NR_POLARIZATIONS, NR_RECEIVERS_PER_BLOCK);
+    tcc::Correlator correlator(cu::Device(0), NR_BITS, NR_RECEIVERS, NR_CHANNELS, NR_SAMPLES_PER_CHANNEL, NR_POLARIZATIONS, NR_RECEIVERS_PER_BLOCK);
 
     cudaStream_t stream;
     Samples *samples;