codegen_acc_visitor.hpp

Go to the documentation of this file.

/*
 * Copyright 2023 Blue Brain Project, EPFL.
 * See the top-level LICENSE file for details.
 *
 * SPDX-License-Identifier: Apache-2.0
 */
 
#pragma once
 
/**
 * \file
 * \brief \copybrief nmodl::codegen::CodegenAccVisitor
 */
 
#include "codegen/codegen_coreneuron_cpp_visitor.hpp"
 
 
namespace nmodl {
namespace codegen {
 
/**
 * \addtogroup codegen_backends
 * \{
 */
 
/**
 * \class CodegenAccVisitor
 * \brief %Visitor for printing C++ code with OpenACC backend
 */
class CodegenAccVisitor: public CodegenCoreneuronCppVisitor {
  public:
    using CodegenCoreneuronCppVisitor::CodegenCoreneuronCppVisitor;
 
  protected:
    /// name of the code generation backend
    std::string backend_name() const override;
 
 
    /// common includes : standard c++, coreneuron and backend specific
    void print_backend_includes() override;
 
 
    /// ivdep like annotation for channel iterations
    void print_parallel_iteration_hint(BlockType type, const ast::Block* block) override;
 
 
    /// atomic update pragma for reduction statements
    void print_atomic_reduction_pragma() override;
 
 
    /// memory allocation routine
    void print_memory_allocation_routine() const override;
 
 
    /// abort routine
    void print_abort_routine() const override;
 
 
    /// annotations like "acc enter data present(...)" for main kernel
    void print_kernel_data_present_annotation_block_begin() override;
 
 
    /// end of annotation like "acc enter data"
    void print_kernel_data_present_annotation_block_end() override;
 
 
    /// start of annotation "acc kernels" for net_init kernel
    void print_net_init_acc_serial_annotation_block_begin() override;
 
 
    /// end of annotation "acc kernels" for net_init kernel
    void print_net_init_acc_serial_annotation_block_end() override;
 
 
    /// update to matrix elements with/without shadow vectors
    void print_nrn_cur_matrix_shadow_update() override;
 
 
    /// reduction to matrix elements from shadow vectors
    void print_nrn_cur_matrix_shadow_reduction() override;
 
    /// fast membrane current calculation
    void print_fast_imem_calculation() override;
 
    /// setup method for setting matrix shadow vectors
    void print_rhs_d_shadow_variables() override;
 
 
    /// if reduction block in nrn_cur required
    bool nrn_cur_reduction_loop_required() override;
 
    /// update global variable from host to the device
    void print_global_variable_device_update_annotation() override;
 
    /// transfer newtonspace structure to device
    void print_newtonspace_transfer_to_device() const override;
 
    /// declare helper functions for copying the instance struct to the device
    void print_instance_struct_transfer_routine_declarations() override;
 
    /// define helper functions for copying the instance struct to the device
    void print_instance_struct_transfer_routines(
        const std::vector<std::string>& ptr_members) override;
 
    /// call helper function for copying the instance struct to the device
    void print_instance_struct_copy_to_device() override;
 
    /// call helper function that deletes the instance struct from the device
    void print_instance_struct_delete_from_device() override;
 
    /// update derivimplicit advance flag on the gpu device
    void print_deriv_advance_flag_transfer_to_device() const override;
 
    /// update NetSendBuffer_t count from device to host
    void print_net_send_buf_count_update_to_host() const override;
 
    /// update NetSendBuffer_t from device to host
    void print_net_send_buf_update_to_host() const override;
 
    /// update NetSendBuffer_t count from host to device
    void print_net_send_buf_count_update_to_device() const override;
 
    /// update dt from host to device
    void print_dt_update_to_device() const override;
 
    // synchronise/wait on stream specific to NrnThread
    void print_device_stream_wait() const override;
 
    /// print atomic capture pragma
    void print_device_atomic_capture_annotation() const override;
 
    /// print atomic update of NetSendBuffer_t cnt
    void print_net_send_buffering_cnt_update() const override;
 
    /// Replace default implementation by a no-op
    /// since the buffer cannot be grown up during gpu execution
    void print_net_send_buffering_grow() override;
};
 
/** \} */  // end of codegen_backends
 
}  // namespace codegen
}  // namespace nmodl