Added __device__ flag to the Shared union constructors in... (aa2ab8b0) · Commits · TNL / tnl-dev

src/TNL/Algorithms/detail/CudaReductionKernel.h

+2 −2

Original line number	Diff line number	Diff line
		@@ -393,7 +393,7 @@ CudaReductionKernel( DataFetcher dataFetcher,

		// initialization is not allowed for __shared__ variables, so we need to
		// disable initialization in the implicit default constructor
		Shared() {}
		__device__ Shared() {}
		};
		__shared__ Shared storage;

		@@ -455,7 +455,7 @@ CudaReductionWithArgumentKernel( DataFetcher dataFetcher,

		// initialization is not allowed for __shared__ variables, so we need to
		// disable initialization in the implicit default constructor
		Shared() {}
		__device__ Shared() {}
		};
		__shared__ Shared storage;

+4 −4

Original line number	Diff line number	Diff line
		@@ -433,7 +433,7 @@ CudaScanKernelUpsweep( const InputView input,

		// initialization is not allowed for __shared__ variables, so we need to
		// disable initialization in the implicit default constructor
		Shared() {}
		__device__ Shared() {}
		};
		__shared__ Shared storage;

		@@ -510,7 +510,7 @@ CudaScanKernelDownsweep( const InputView input,

		// initialization is not allowed for __shared__ variables, so we need to
		// disable initialization in the implicit default constructor
		Shared() {}
		__device__ Shared() {}
		};
		__shared__ Shared storage;

		@@ -550,7 +550,7 @@ CudaScanKernelParallel( const InputView input,

		// initialization is not allowed for __shared__ variables, so we need to
		// disable initialization in the implicit default constructor
		Shared() {}
		__device__ Shared() {}
		};
		__shared__ Shared storage;

		@@ -588,7 +588,7 @@ CudaScanKernelUniformShift( OutputView output,

		// initialization is not allowed for __shared__ variables, so we need to
		// disable initialization in the implicit default constructor
		Shared() {}
		__device__ Shared() {}
		};
		__shared__ Shared storage;
		if( threadIdx.x == 0 )