GPUStridedCopy.hpp

#pragma once
#include <Gfx/Graph/RenderState.hpp>
 
#include <score_plugin_gfx_export.h>
 
#include <private/qrhi_p.h>
 
namespace score::gfx
{
 
class SCORE_PLUGIN_GFX_TEST_EXPORT GPUStridedCopy
{
public:
  struct Params
  {
    QRhiBuffer* src{};
    QRhiBuffer* dst{};
    uint32_t count{};             // number of elements
    uint32_t element_bytes{};     // bytes copied per element
    uint32_t src_offset_bytes{};
    uint32_t src_stride_bytes{};
    uint32_t dst_offset_bytes{};
    uint32_t dst_stride_bytes{};
  };
 
  struct PreparedOp
  {
    QRhiShaderResourceBindings* srb{};
    QRhiBuffer* paramsUBO{};
    // The buffers srb was last created against, by globalResourceId: a buffer
    // freed and reallocated can come back at the same address.
    quint64 src_id{};
    quint64 dst_id{};
  };
 
  static bool eligible(const Params& p) noexcept;
 
  bool init(RenderState& state);
  void release();
 
  PreparedOp prepare(QRhi& rhi, const Params& p);
  static void releaseOp(PreparedOp& op);
 
  void updateParams(QRhiResourceUpdateBatch& res, PreparedOp& op, const Params& p);
  void dispatch(QRhiCommandBuffer& cb, const PreparedOp& op, const Params& p);
 
  static constexpr int LocalSize = 256;
 
private:
  struct Dims
  {
    int x{}, y{};
  };
  Dims dims(uint64_t invocations) const noexcept;
 
  QRhiComputePipeline* m_pipeline{};
  QShader m_shader;
  int m_maxWorkgroupsPerDim{65535};
};
 
}