Commit e4a5f7e1 authored by Doak, Peter W.'s avatar Doak, Peter W.
Browse files

changes to remove unecessary inheritance

parent 77c72a0a

include/dca/linalg/util/magma_queue.hpp

+16 −4
Original line number Diff line number Diff line
@@ -25,26 +25,25 @@ namespace linalg { 
namespace util {

// dca::linalg::util::



class MagmaQueue : public CudaStream {

class MagmaQueue {

public:

  MagmaQueue() {

    cublasCreate(&cublas_handle_);

    cusparseCreate(&cusparse_handle_);

    int device;

    cudaGetDevice(&device);

    magma_queue_create_from_cuda(device, static_cast<cudaStream_t>(*this), cublas_handle_,

    magma_queue_create_from_cuda(device, stream_, cublas_handle_,

                                 cusparse_handle_, &queue_);

  }



  MagmaQueue(const MagmaQueue& rhs) = delete;

  MagmaQueue& operator=(const MagmaQueue& rhs) = delete;



  MagmaQueue(MagmaQueue&& rhs) noexcept : CudaStream(std::move(rhs)) {

  MagmaQueue(MagmaQueue&& rhs) noexcept {

    swapMembers(rhs);

  }



  MagmaQueue& operator=(MagmaQueue&& rhs) noexcept {

    CudaStream::operator=(std::move(rhs));

    swapMembers(rhs);

    return *this;

  }
@@ -59,13 +58,26 @@ public: 
    return queue_;

  }



  // Allows a large number of calls that previously took a stream

  // take a MagmaQueue, this makes all this code less intelligible

  // but less verbose.  Consider this carefully.

  operator cudaStream_t() const {

    return stream_;

  }

  

  const CudaStream& getStream() const {

    return stream_;

  }

  

private:

  void swapMembers(MagmaQueue& rhs) noexcept {

    std::swap(stream_, rhs.stream_);

    std::swap(cublas_handle_, rhs.cublas_handle_);

    std::swap(cusparse_handle_, rhs.cusparse_handle_);

    std::swap(queue_, rhs.queue_);

  }



  CudaStream stream_;

  magma_queue_t queue_ = nullptr;

  cublasHandle_t cublas_handle_ = nullptr;

  cusparseHandle_t cusparse_handle_ = nullptr;

include/dca/math/function_transform/special_transforms/space_transform_2D_gpu.hpp

+2 −4
Original line number Diff line number Diff line
@@ -63,7 +63,7 @@ public: 


  // Returns: the stream associated with the magma queue.

  cudaStream_t get_stream() const {

    return stream_;

    return queue_.getStream();

  }



  std::size_t deviceFingerprint() const {
@@ -87,7 +87,6 @@ private: 
  const int nc_;



  const linalg::util::MagmaQueue& queue_;

  const linalg::util::CudaStream& stream_;



  std::shared_ptr<RMatrix> workspace_;
@@ -102,7 +101,6 @@ SpaceTransform2DGpu<RDmn, KDmn, Real>::SpaceTransform2DGpu(const int nw_pos, 
      nw_(2 * nw_pos),

      nc_(RDmn::dmn_size()),

      queue_(queue),

      stream_(queue),

      plan1_(queue_),

      plan2_(queue_) {

  workspace_ = std::make_shared<RMatrix>();
@@ -159,7 +157,7 @@ void SpaceTransform2DGpu<RDmn, KDmn, Real>::phaseFactorsAndRearrange(const RMatr 
      BaseClass::hasPhaseFactors() ? getPhaseFactors().ptr() : nullptr;

  details::phaseFactorsAndRearrange(in.ptr(), in.leadingDimension(), out.ptr(),

                                    out.leadingDimension(), n_bands_, nc_, nw_, phase_factors_ptr,

                                    stream_);

                                    queue_);

}



template <class RDmn, class KDmn, typename Real>

test/unit/math/function_transform/space_transform_2D_gpu_test.cpp

+1 −1
Original line number Diff line number Diff line
@@ -107,7 +107,7 @@ TYPED_TEST(SpaceTransform2DGpuTest, Execute) { 
  dca::math::transform::SpaceTransform2DGpu<RDmn, KDmn, Real> transform_obj(nw, queue);

  transform_obj.execute(M_dev);



  queue.sync();

  queue.getStream().sync();



  constexpr Real tolerance = std::numeric_limits<Real>::epsilon() * 500;