diff options
author | cnugteren <web@cedricnugteren.nl> | 2016-04-20 22:12:51 -0600 |
---|---|---|
committer | cnugteren <web@cedricnugteren.nl> | 2016-04-20 22:12:51 -0600 |
commit | 16a048f1ac8102ad4bcce9cf9fc320f791323e45 (patch) | |
tree | bd7df539c92b92cb8841ff27fd83c4e14e5958bc /src/routines/level1 | |
parent | 894983fc3c7c57ffc48c21523641694cde318eca (diff) |
Added support for the iSAMAX/iDAMAX/iCAMAX/iZAMAX routines
Diffstat (limited to 'src/routines/level1')
-rw-r--r-- | src/routines/level1/xamax.cc | 112 |
1 files changed, 112 insertions, 0 deletions
diff --git a/src/routines/level1/xamax.cc b/src/routines/level1/xamax.cc new file mode 100644 index 00000000..ffdfa496 --- /dev/null +++ b/src/routines/level1/xamax.cc @@ -0,0 +1,112 @@ + +// ================================================================================================= +// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This +// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max- +// width of 100 characters per line. +// +// Author(s): +// Cedric Nugteren <www.cedricnugteren.nl> +// +// This file implements the Xamax class (see the header for information about the class). +// +// ================================================================================================= + +#include "internal/routines/level1/xamax.h" + +#include <string> +#include <vector> + +namespace clblast { +// ================================================================================================= + +// Specific implementations to get the memory-type based on a template argument +template <> const Precision Xamax<float>::precision_ = Precision::kSingle; +template <> const Precision Xamax<double>::precision_ = Precision::kDouble; +template <> const Precision Xamax<float2>::precision_ = Precision::kComplexSingle; +template <> const Precision Xamax<double2>::precision_ = Precision::kComplexDouble; + +// ================================================================================================= + +// Constructor: forwards to base class constructor +template <typename T> +Xamax<T>::Xamax(Queue &queue, EventPointer event, const std::string &name): + Routine<T>(queue, event, name, {"Xdot"}, precision_) { + source_string_ = + #include "../../kernels/level1/xamax.opencl" + ; +} + +// ================================================================================================= + +// The main routine +template <typename T> +StatusCode Xamax<T>::DoAmax(const size_t n, + const Buffer<T> &imax_buffer, const size_t imax_offset, + const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc) { + + // Makes sure all dimensions are larger than zero + if (n == 0) { return StatusCode::kInvalidDimension; } + + // Tests the vectors for validity + auto status = TestVectorX(n, x_buffer, x_offset, x_inc, sizeof(T)); + if (ErrorIn(status)) { return status; } + status = TestVectorDot(1, imax_buffer, imax_offset, sizeof(T)); + if (ErrorIn(status)) { return status; } + + // Retrieves the Xamax kernels from the compiled binary + try { + auto& program = GetProgramFromCache(); + auto kernel1 = Kernel(program, "Xamax"); + auto kernel2 = Kernel(program, "XamaxEpilogue"); + + // Creates the buffer for intermediate values + auto temp_size = 2*db_["WGS2"]; + auto temp_buffer1 = Buffer<T>(context_, temp_size); + auto temp_buffer2 = Buffer<unsigned int>(context_, temp_size); + + // Sets the kernel arguments + kernel1.SetArgument(0, static_cast<int>(n)); + kernel1.SetArgument(1, x_buffer()); + kernel1.SetArgument(2, static_cast<int>(x_offset)); + kernel1.SetArgument(3, static_cast<int>(x_inc)); + kernel1.SetArgument(4, temp_buffer1()); + kernel1.SetArgument(5, temp_buffer2()); + + // Event waiting list + auto eventWaitList = std::vector<Event>(); + + // Launches the main kernel + auto global1 = std::vector<size_t>{db_["WGS1"]*temp_size}; + auto local1 = std::vector<size_t>{db_["WGS1"]}; + auto kernelEvent = Event(); + status = RunKernel(kernel1, global1, local1, kernelEvent.pointer()); + if (ErrorIn(status)) { return status; } + eventWaitList.push_back(kernelEvent); + + // Sets the arguments for the epilogue kernel + kernel2.SetArgument(0, temp_buffer1()); + kernel2.SetArgument(1, temp_buffer2()); + kernel2.SetArgument(2, imax_buffer()); + kernel2.SetArgument(3, static_cast<int>(imax_offset)); + + // Launches the epilogue kernel + auto global2 = std::vector<size_t>{db_["WGS2"]}; + auto local2 = std::vector<size_t>{db_["WGS2"]}; + status = RunKernel(kernel2, global2, local2, event_, eventWaitList); + if (ErrorIn(status)) { return status; } + + // Succesfully finished the computation + return StatusCode::kSuccess; + } catch (...) { return StatusCode::kInvalidKernel; } +} + +// ================================================================================================= + +// Compiles the templated class +template class Xamax<float>; +template class Xamax<double>; +template class Xamax<float2>; +template class Xamax<double2>; + +// ================================================================================================= +} // namespace clblast |