Refactored and extended tests for scan and distributed scan

src/TNL/Algorithms/detail/CudaScanKernel.h

@@ -373,6 +373,7 @@ struct CudaScanKernelLauncher
    static void resetMaxGridSize()
    {
       maxGridSize() = Cuda::getMaxGridSize();
+      gridsCount() = -1;
    }
 
    static int& gridsCount()

src/TNL/Algorithms/detail/Scan.hpp

@@ -107,6 +107,9 @@ perform( Vector& v,
    using ValueType = typename Vector::ValueType;
    using IndexType = typename Vector::IndexType;
 
+   if( end <= begin )
+      return;
+
    const IndexType size = end - begin;
    const int max_threads = Devices::Host::getMaxThreadsCount();
    const IndexType block_size = TNL::max( 1024, TNL::roundUpDivision( size, max_threads ) );
@@ -157,6 +160,12 @@ performFirstPhase( Vector& v,
    using ValueType = typename Vector::ValueType;
    using IndexType = typename Vector::IndexType;
 
+   if( end <= begin ) {
+      Containers::Array< typename Vector::ValueType, Devices::Sequential > block_results( 1 );
+      block_results.setValue( zero );
+      return block_results;
+   }
+
    const IndexType size = end - begin;
    const int max_threads = Devices::Host::getMaxThreadsCount();
    const IndexType block_size = TNL::max( 1024, TNL::roundUpDivision( size, max_threads ) );
@@ -204,6 +213,9 @@ performSecondPhase( Vector& v,
    using ValueType = typename Vector::ValueType;
    using IndexType = typename Vector::IndexType;
 
+   if( end <= begin )
+      return;
+
    const IndexType size = end - begin;
    const int max_threads = Devices::Host::getMaxThreadsCount();
    const IndexType block_size = TNL::max( 1024, TNL::roundUpDivision( size, max_threads ) );
@@ -241,6 +253,9 @@ perform( Vector& v,
    using ValueType = typename Vector::ValueType;
    using IndexType = typename Vector::IndexType;
 
+   if( end <= begin )
+      return;
+
    detail::CudaScanKernelLauncher< Type, ValueType, IndexType >::perform(
       end - begin,
       &v.getData()[ begin ],  // input
@@ -267,6 +282,12 @@ performFirstPhase( Vector& v,
    using ValueType = typename Vector::ValueType;
    using IndexType = typename Vector::IndexType;
 
+   if( end <= begin ) {
+      Containers::Array< typename Vector::ValueType, Devices::Cuda > block_results( 1 );
+      block_results.setValue( zero );
+      return block_results;
+   }
+
    return detail::CudaScanKernelLauncher< Type, ValueType, IndexType >::performFirstPhase(
       end - begin,
       &v.getData()[ begin ],  // input
@@ -295,6 +316,9 @@ performSecondPhase( Vector& v,
    using ValueType = typename Vector::ValueType;
    using IndexType = typename Vector::IndexType;
 
+   if( end <= begin )
+      return;
+
    detail::CudaScanKernelLauncher< Type, ValueType, IndexType >::performSecondPhase(
       end - begin,
       &v.getData()[ begin ],  // output

src/TNL/Algorithms/scan.h

@@ -64,6 +64,8 @@ inplaceInclusiveScan( Array& array,
                       Reduction&& reduction,
                       typename Array::ValueType zero )
 {
+   TNL_ASSERT_EQ( reduction( zero, zero ), zero,
+                  "zero is not an idempotent value of the reduction operation" );
    using Scan = detail::Scan< typename Array::DeviceType, detail::ScanType::Inclusive >;
    Scan::perform( array, begin, end, std::forward< Reduction >( reduction ), zero );
 }
@@ -134,6 +136,8 @@ inplaceExclusiveScan( Array& array,
                       Reduction&& reduction,
                       typename Array::ValueType zero )
 {
+   TNL_ASSERT_EQ( reduction( zero, zero ), zero,
+                  "zero is not an idempotent value of the reduction operation" );
    using Scan = detail::Scan< typename Array::DeviceType, detail::ScanType::Exclusive >;
    Scan::perform( array, begin, end, std::forward< Reduction >( reduction ), zero );
 }