rocm-systems/rocclr/runtime/device/pal/palcounters.hpp

//
// Copyright (c) 2015 Advanced Micro Devices, Inc. All rights reserved.
//
#pragma once

#include "top.hpp"
#include "device/device.hpp"
#include "device/pal/paldevice.hpp"
#include "palPerfExperiment.h"

namespace pal {

enum class PCIndexSelect : uint {
  None = 0,                        ///< no index
  Instance,                        ///< index by block instance -- (1 to 1 mapping with PAL)
  ShaderEngine,                    ///< index by shader engine (sum = 1 value for all PAL Instances)
  ShaderArray,                     ///< index by shader array (sum = 1 value for each shader array)
  ComputeUnit,                     ///< index by compute unit
};

class VirtualGPU;

class PalCounterReference : public amd::ReferenceCountedObject {
 public:
  static PalCounterReference* Create(VirtualGPU& gpu);

  //! Default constructor
  PalCounterReference(VirtualGPU& gpu  //!< Virtual GPU device object
                      )
      : gpu_(gpu),
        perfExp_(nullptr),
        layout_(nullptr),
        memory_(nullptr),
        cpuAddr_(nullptr),
        numExpCounters_(0) {}

  //! Get PAL counter
  Pal::IPerfExperiment* iPerf() const { return perfExp_; }

  //! Returns the virtual GPU device
  const VirtualGPU& gpu() const { return gpu_; }

  //! Prepare for execution
  bool finalize();

  //! Returns the PAL counter results
  uint64_t result(const std::vector<int>& index);

  //! Get the latest Experiment Counter index
  uint getPalCounterIndex() { return numExpCounters_++; };

 protected:
  //! Default destructor
  ~PalCounterReference();

 private:
  //! Disable copy constructor
  PalCounterReference(const PalCounterReference&);

  //! Disable operator=
  PalCounterReference& operator=(const PalCounterReference&);

  VirtualGPU& gpu_;                   //!< The virtual GPU device object
  Pal::IPerfExperiment* perfExp_;     //!< PAL performance experiment object
  Pal::GlobalCounterLayout* layout_;  //!< Layout of the result
  Memory* memory_;                    //!< Memory used by PAL performance experiment
  void* cpuAddr_;                     //!< CPU address of memory_
  uint numExpCounters_;               //!< Number of Experiment Counter created
};

//! Performance counter implementation on GPU
class PerfCounter : public device::PerfCounter {
 public:
  //! The performance counter info
  struct Info : public amd::EmbeddedObject {
    uint blockIndex_;            //!< Index of the block to configure
    uint counterIndex_;          //!< Index of the hardware counter
    uint eventIndex_;            //!< Event you wish to count with the counter
    PCIndexSelect indexSelect_;  //!< IndexSelect type of the counter
  };

  //! Constructor for the GPU PerfCounter object
  PerfCounter(const Device& device,         //!< A GPU device object
              PalCounterReference* palRef,  //!< Counter Reference
              cl_uint blockIndex,           //!< HW block index
              cl_uint counterIndex,         //!< Counter index within the block
              cl_uint eventIndex)           //!< Event index for profiling
      : gpuDevice_(device), palRef_(palRef) {
    info_.blockIndex_ = blockIndex;
    info_.counterIndex_ = counterIndex;
    info_.eventIndex_ = eventIndex;
    convertInfo();
  }

  //! Destructor for the GPU PerfCounter object
  virtual ~PerfCounter();

  //! Creates the current object
  bool create();

  //! Returns the specific information about the counter
  uint64_t getInfo(uint64_t infoType  //!< The type of returned information
                   ) const;

  //! Returns the GPU device, associated with the current object
  const Device& dev() const { return gpuDevice_; }

  //! Returns the virtual GPU device
  const VirtualGPU& gpu() const { return palRef_->gpu(); }

  //! Returns the PAL performance counter descriptor
  const Info* info() const { return &info_; }

  //! Returns the Info structure for performance counter
  Pal::IPerfExperiment* iPerf() const { return palRef_->iPerf(); }

 private:
  //! Disable default copy constructor
  PerfCounter(const PerfCounter&);

  //! Disable default operator=
  PerfCounter& operator=(const PerfCounter&);

  //! Convert info from ORCA to PAL
  void convertInfo();

  const Device& gpuDevice_;      //!< The backend device
  PalCounterReference* palRef_;  //!< Reference counter
  Info info_;                    //!< The info structure for perfcounter
  std::vector<int> index_;       //!< Counter index in the PAL container
};

}  // namespace pal