draw_pass.hh

Go to the documentation of this file.

/* SPDX-FileCopyrightText: 2022 Blender Authors
 *
 * SPDX-License-Identifier: GPL-2.0-or-later */
 
#pragma once

 
#include "BLI_listbase_wrapper.hh"
#include "BLI_vector.hh"
 
#include "BKE_image.hh"
 
#include "GPU_debug.hh"
#include "GPU_material.hh"
 
#include "DRW_gpu_wrapper.hh"
 
#include "draw_command.hh"
#include "draw_handle.hh"
#include "draw_manager.hh"
#include "draw_pass.hh"
#include "draw_shader_shared.hh"
#include "draw_state.hh"
 
#include "intern/gpu_codegen.hh"
 
#include <sstream>
 
namespace blender::draw {
using namespace blender::draw;
using namespace blender::draw::command;
 
class Manager;
 
namespace command {
class DrawCommandBuf;
}
 
/* -------------------------------------------------------------------- */
 
namespace detail {

template<typename T,
         int64_t block_size = 16>
class SubPassVector {
 private:
  Vector<std::unique_ptr<Vector<T, block_size>>, 0> blocks_;
 
 public:
  void clear()
  {
    blocks_.clear();
  }
 
  int64_t append_and_get_index(T &&elem)
  {
    /* Do not go over the inline size so that existing members never move. */
    if (blocks_.is_empty() || blocks_.last()->size() == block_size) {
      blocks_.append(std::make_unique<Vector<T, block_size>>());
    }
    return blocks_.last()->append_and_get_index(std::move(elem)) +
           (blocks_.size() - 1) * block_size;
  }
 
  T &operator[](int64_t index)
  {
    return (*blocks_[index / block_size])[index % block_size];
  }
 
  const T &operator[](int64_t index) const
  {
    return (*blocks_[index / block_size])[index % block_size];
  }
};

template<
    typename DrawCommandBufType>
class PassBase {
  friend Manager;
  friend DrawCommandBuf;

  static constexpr GPUSamplerState sampler_auto = GPUSamplerState::internal_sampler();
 
 protected:
  Vector<command::Header, 0> headers_;
  Vector<command::Undetermined, 0> commands_;
  /* Reference to draw commands buffer. Either own or from parent pass. */
  DrawCommandBufType &draw_commands_buf_;
  /* Reference to sub-pass commands buffer. Either own or from parent pass. */
  SubPassVector<PassBase<DrawCommandBufType>> &sub_passes_;
  GPUShader *shader_;
 
 public:
  const char *debug_name;
 
  bool use_custom_ids;
 
  PassBase(const char *name,
           DrawCommandBufType &draw_command_buf,
           SubPassVector<PassBase<DrawCommandBufType>> &sub_passes,
           GPUShader *shader = nullptr)
      : draw_commands_buf_(draw_command_buf),
        sub_passes_(sub_passes),
        shader_(shader),
        debug_name(name),
        use_custom_ids(false){};

  void init();

  PassBase<DrawCommandBufType> &sub(const char *name);

  void state_set(DRWState state, int clip_plane_count = 0);

  void clear_color(float4 color);
  void clear_depth(float depth);
  void clear_stencil(uint8_t stencil);
  void clear_depth_stencil(float depth, uint8_t stencil);
  void clear_color_depth_stencil(float4 color, float depth, uint8_t stencil);
  void clear_multi(Span<float4> colors);

  void state_stencil(uint8_t write_mask, uint8_t reference, uint8_t compare_mask);

  void shader_set(GPUShader *shader);

  void framebuffer_set(GPUFrameBuffer **framebuffer);

  void subpass_transition(GPUAttachmentState depth_attachment,
                          Span<GPUAttachmentState> color_attachments);

  void material_set(Manager &manager, GPUMaterial *material);

  void draw(gpu::Batch *batch,
            uint instance_len = -1,
            uint vertex_len = -1,
            uint vertex_first = -1,
            ResourceHandle handle = {0},
            uint custom_id = 0);

  void draw(gpu::Batch *batch, ResourceHandle handle, uint custom_id = 0);

  void draw_procedural(GPUPrimType primitive,
                       uint instance_len,
                       uint vertex_len,
                       uint vertex_first = -1,
                       ResourceHandle handle = {0},
                       uint custom_id = 0);

  void draw_expand(gpu::Batch *batch,
                   GPUPrimType primitive_type,
                   uint primitive_len,
                   uint instance_len,
                   uint vertex_len = -1,
                   uint vertex_first = -1,
                   ResourceHandle handle = {0},
                   uint custom_id = 0);

  void draw_expand(gpu::Batch *batch,
                   GPUPrimType primitive_type,
                   uint primitive_len,
                   uint instance_len,
                   ResourceHandle handle = {0},
                   uint custom_id = 0);

  void draw_indirect(gpu::Batch *batch,
                     StorageBuffer<DrawCommand, true> &indirect_buffer,
                     ResourceHandle handle = {0});
  void draw_procedural_indirect(GPUPrimType primitive,
                                StorageBuffer<DrawCommand, true> &indirect_buffer,
                                ResourceHandle handle = {0});

  void dispatch(int group_len);
  void dispatch(int2 group_len);
  void dispatch(int3 group_len);
  void dispatch(int3 *group_len);
  void dispatch(StorageBuffer<DispatchCommand> &indirect_buffer);

  void barrier(eGPUBarrier type);

  void bind_image(const char *name, GPUTexture *image);
  void bind_image(const char *name, GPUTexture **image);
  void bind_image(int slot, GPUTexture *image);
  void bind_image(int slot, GPUTexture **image);
  void bind_texture(const char *name, GPUTexture *texture, GPUSamplerState state = sampler_auto);
  void bind_texture(const char *name, GPUTexture **texture, GPUSamplerState state = sampler_auto);
  void bind_texture(const char *name, gpu::VertBuf *buffer);
  void bind_texture(const char *name, gpu::VertBuf **buffer);
  void bind_texture(int slot, GPUTexture *texture, GPUSamplerState state = sampler_auto);
  void bind_texture(int slot, GPUTexture **texture, GPUSamplerState state = sampler_auto);
  void bind_texture(int slot, gpu::VertBuf *buffer);
  void bind_texture(int slot, gpu::VertBuf **buffer);
  void bind_ssbo(const char *name, GPUStorageBuf *buffer);
  void bind_ssbo(const char *name, GPUStorageBuf **buffer);
  void bind_ssbo(int slot, GPUStorageBuf *buffer);
  void bind_ssbo(int slot, GPUStorageBuf **buffer);
  void bind_ssbo(const char *name, GPUUniformBuf *buffer);
  void bind_ssbo(const char *name, GPUUniformBuf **buffer);
  void bind_ssbo(int slot, GPUUniformBuf *buffer);
  void bind_ssbo(int slot, GPUUniformBuf **buffer);
  void bind_ssbo(const char *name, gpu::VertBuf *buffer);
  void bind_ssbo(const char *name, gpu::VertBuf **buffer);
  void bind_ssbo(int slot, gpu::VertBuf *buffer);
  void bind_ssbo(int slot, gpu::VertBuf **buffer);
  void bind_ssbo(const char *name, gpu::IndexBuf *buffer);
  void bind_ssbo(const char *name, gpu::IndexBuf **buffer);
  void bind_ssbo(int slot, gpu::IndexBuf *buffer);
  void bind_ssbo(int slot, gpu::IndexBuf **buffer);
  void bind_ubo(const char *name, GPUUniformBuf *buffer);
  void bind_ubo(const char *name, GPUUniformBuf **buffer);
  void bind_ubo(int slot, GPUUniformBuf *buffer);
  void bind_ubo(int slot, GPUUniformBuf **buffer);

  void push_constant(const char *name, const float &data);
  void push_constant(const char *name, const float2 &data);
  void push_constant(const char *name, const float3 &data);
  void push_constant(const char *name, const float4 &data);
  void push_constant(const char *name, const int &data);
  void push_constant(const char *name, const int2 &data);
  void push_constant(const char *name, const int3 &data);
  void push_constant(const char *name, const int4 &data);
  void push_constant(const char *name, const bool &data);
  void push_constant(const char *name, const float4x4 &data);
  void push_constant(const char *name, const float *data, int array_len = 1);
  void push_constant(const char *name, const float2 *data, int array_len = 1);
  void push_constant(const char *name, const float3 *data, int array_len = 1);
  void push_constant(const char *name, const float4 *data, int array_len = 1);
  void push_constant(const char *name, const int *data, int array_len = 1);
  void push_constant(const char *name, const int2 *data, int array_len = 1);
  void push_constant(const char *name, const int3 *data, int array_len = 1);
  void push_constant(const char *name, const int4 *data, int array_len = 1);
  void push_constant(const char *name, const float4x4 *data);

  void specialize_constant(GPUShader *shader, const char *name, const float &data);
  void specialize_constant(GPUShader *shader, const char *name, const int &data);
  void specialize_constant(GPUShader *shader, const char *name, const uint &data);
  void specialize_constant(GPUShader *shader, const char *name, const bool &data);
  void specialize_constant(GPUShader *shader, const char *name, const float *data);
  void specialize_constant(GPUShader *shader, const char *name, const int *data);
  void specialize_constant(GPUShader *shader, const char *name, const uint *data);
  void specialize_constant(GPUShader *shader, const char *name, const bool *data);

  template<class U> void bind_resources(U &resources)
  {
    resources.bind_resources(*this);
  }

  std::string serialize(std::string line_prefix = "") const;
 
  friend std::ostream &operator<<(std::ostream &stream, const PassBase &pass)
  {
    return stream << pass.serialize();
  }
 
 protected:
 
  int push_constant_offset(const char *name);
 
  void clear(eGPUFrameBufferBits planes, float4 color, float depth, uint8_t stencil);
 
  gpu::Batch *procedural_batch_get(GPUPrimType primitive);

  command::Undetermined &create_command(command::Type type);
 
  void submit(command::RecordingState &state) const;
};
 
template<typename DrawCommandBufType> class Pass : public detail::PassBase<DrawCommandBufType> {
 public:
  using Sub = detail::PassBase<DrawCommandBufType>;
 
 private:
  SubPassVector<detail::PassBase<DrawCommandBufType>> sub_passes_main_;
  DrawCommandBufType draw_commands_buf_main_;
 
 public:
  Pass(const char *name)
      : detail::PassBase<DrawCommandBufType>(name, draw_commands_buf_main_, sub_passes_main_){};
 
  void init()
  {
    this->headers_.clear();
    this->commands_.clear();
    this->sub_passes_.clear();
    this->draw_commands_buf_.clear();
  }
};  // namespace blender::draw
 
}  // namespace detail

 
/* -------------------------------------------------------------------- */

// using PassSimple = detail::Pass<DrawCommandBuf>;

// using PassMain = detail::Pass<DrawMultiBuf>;

class PassSortable : public PassMain {
  friend Manager;
 
 private:
  Vector<float> sorting_values_;
 
  bool sorted_ = false;
 
 public:
  PassSortable(const char *name_) : PassMain(name_){};
 
  void init()
  {
    sorting_values_.clear();
    sorted_ = false;
    PassMain::init();
  }
 
  PassMain::Sub &sub(const char *name, float sorting_value)
  {
    int64_t index = sub_passes_.append_and_get_index(
        PassBase(name, draw_commands_buf_, sub_passes_, shader_));
    headers_.append({Type::SubPass, uint(index)});
    /* Some sub-pass can also create sub-sub-passes (curve, point-clouds...) which will de-sync
     * the `sub_passes_.size()` and `sorting_values_.size()`, making the  `Header::index` not
     * reusable for the sorting value in the `sort()` function. To fix this, we flood the
     * `sorting_values_` to ensure the same index is valid for `sorting_values_` and
     * `sub_passes_`. */
    int64_t sorting_index;
    do {
      sorting_index = sorting_values_.append_and_get_index(sorting_value);
    } while (sorting_index != index);
    return sub_passes_[index];
  }
 
  std::string serialize(std::string line_prefix = "") const
  {
    if (sorted_ == false) {
      const_cast<PassSortable *>(this)->sort();
    }
    return PassMain::serialize(line_prefix);
  }
 
 protected:
  void sort()
  {
    if (sorted_ == false) {
      std::sort(headers_.begin(), headers_.end(), [&](Header &a, Header &b) {
        BLI_assert(a.type == Type::SubPass && b.type == Type::SubPass);
        float a_val = sorting_values_[a.index];
        float b_val = sorting_values_[b.index];
        return a_val < b_val || (a_val == b_val && a.index < b.index);
      });
      sorted_ = true;
    }
  }
};

 
namespace detail {
 
/* -------------------------------------------------------------------- */
 
template<class T> inline command::Undetermined &PassBase<T>::create_command(command::Type type)
{
  int64_t index = commands_.append_and_get_index({});
  headers_.append({type, uint(index)});
  return commands_[index];
}
 
template<class T>
inline void PassBase<T>::clear(eGPUFrameBufferBits planes,
                               float4 color,
                               float depth,
                               uint8_t stencil)
{
  create_command(command::Type::Clear).clear = {uint8_t(planes), stencil, depth, color};
}
 
template<class T> inline void PassBase<T>::clear_multi(Span<float4> colors)
{
  create_command(command::Type::ClearMulti).clear_multi = {colors.data(),
                                                           static_cast<int>(colors.size())};
}
 
template<class T> inline gpu::Batch *PassBase<T>::procedural_batch_get(GPUPrimType primitive)
{
  switch (primitive) {
    case GPU_PRIM_POINTS:
      return drw_cache_procedural_points_get();
    case GPU_PRIM_LINES:
      return drw_cache_procedural_lines_get();
    case GPU_PRIM_TRIS:
      return drw_cache_procedural_triangles_get();
    case GPU_PRIM_TRI_STRIP:
      return drw_cache_procedural_triangle_strips_get();
    default:
      /* Add new one as needed. */
      BLI_assert_unreachable();
      return nullptr;
  }
}
 
template<class T> inline PassBase<T> &PassBase<T>::sub(const char *name)
{
  int64_t index = sub_passes_.append_and_get_index(
      PassBase(name, draw_commands_buf_, sub_passes_, shader_));
  headers_.append({command::Type::SubPass, uint(index)});
  return sub_passes_[index];
}
 
template<class T> void PassBase<T>::submit(command::RecordingState &state) const
{
  GPU_debug_group_begin(debug_name);
 
  for (const command::Header &header : headers_) {
    switch (header.type) {
      default:
      case Type::None:
        break;
      case Type::SubPass:
        sub_passes_[header.index].submit(state);
        break;
      case command::Type::FramebufferBind:
        commands_[header.index].framebuffer_bind.execute();
        break;
      case command::Type::SubPassTransition:
        commands_[header.index].subpass_transition.execute();
        break;
      case command::Type::ShaderBind:
        commands_[header.index].shader_bind.execute(state);
        break;
      case command::Type::ResourceBind:
        commands_[header.index].resource_bind.execute();
        break;
      case command::Type::PushConstant:
        commands_[header.index].push_constant.execute(state);
        break;
      case command::Type::SpecializeConstant:
        commands_[header.index].specialize_constant.execute();
        break;
      case command::Type::Draw:
        commands_[header.index].draw.execute(state);
        break;
      case command::Type::DrawMulti:
        commands_[header.index].draw_multi.execute(state);
        break;
      case command::Type::DrawIndirect:
        commands_[header.index].draw_indirect.execute(state);
        break;
      case command::Type::Dispatch:
        commands_[header.index].dispatch.execute(state);
        break;
      case command::Type::DispatchIndirect:
        commands_[header.index].dispatch_indirect.execute(state);
        break;
      case command::Type::Barrier:
        commands_[header.index].barrier.execute();
        break;
      case command::Type::Clear:
        commands_[header.index].clear.execute();
        break;
      case command::Type::ClearMulti:
        commands_[header.index].clear_multi.execute();
        break;
      case command::Type::StateSet:
        commands_[header.index].state_set.execute(state);
        break;
      case command::Type::StencilSet:
        commands_[header.index].stencil_set.execute();
        break;
    }
  }
 
  GPU_debug_group_end();
}
 
template<class T> std::string PassBase<T>::serialize(std::string line_prefix) const
{
  std::stringstream ss;
  ss << line_prefix << "." << debug_name << std::endl;
  line_prefix += "  ";
  for (const command::Header &header : headers_) {
    switch (header.type) {
      default:
      case Type::None:
        break;
      case Type::SubPass:
        ss << sub_passes_[header.index].serialize(line_prefix);
        break;
      case Type::FramebufferBind:
        ss << line_prefix << commands_[header.index].framebuffer_bind.serialize() << std::endl;
        break;
      case Type::SubPassTransition:
        ss << line_prefix << commands_[header.index].subpass_transition.serialize() << std::endl;
        break;
      case Type::ShaderBind:
        ss << line_prefix << commands_[header.index].shader_bind.serialize() << std::endl;
        break;
      case Type::ResourceBind:
        ss << line_prefix << commands_[header.index].resource_bind.serialize() << std::endl;
        break;
      case Type::PushConstant:
        ss << line_prefix << commands_[header.index].push_constant.serialize() << std::endl;
        break;
      case Type::Draw:
        ss << line_prefix << commands_[header.index].draw.serialize() << std::endl;
        break;
      case Type::DrawMulti:
        ss << commands_[header.index].draw_multi.serialize(line_prefix);
        break;
      case Type::DrawIndirect:
        ss << line_prefix << commands_[header.index].draw_indirect.serialize() << std::endl;
        break;
      case Type::Dispatch:
        ss << line_prefix << commands_[header.index].dispatch.serialize() << std::endl;
        break;
      case Type::DispatchIndirect:
        ss << line_prefix << commands_[header.index].dispatch_indirect.serialize() << std::endl;
        break;
      case Type::Barrier:
        ss << line_prefix << commands_[header.index].barrier.serialize() << std::endl;
        break;
      case Type::Clear:
        ss << line_prefix << commands_[header.index].clear.serialize() << std::endl;
        break;
      case Type::ClearMulti:
        ss << line_prefix << commands_[header.index].clear_multi.serialize() << std::endl;
        break;
      case Type::StateSet:
        ss << line_prefix << commands_[header.index].state_set.serialize() << std::endl;
        break;
      case Type::StencilSet:
        ss << line_prefix << commands_[header.index].stencil_set.serialize() << std::endl;
        break;
    }
  }
  return ss.str();
}

 
/* -------------------------------------------------------------------- */
 
template<class T>
inline void PassBase<T>::draw(gpu::Batch *batch,
                              uint instance_len,
                              uint vertex_len,
                              uint vertex_first,
                              ResourceHandle handle,
                              uint custom_id)
{
  if (instance_len == 0 || vertex_len == 0) {
    return;
  }
  BLI_assert(shader_);
  draw_commands_buf_.append_draw(headers_,
                                 commands_,
                                 batch,
                                 instance_len,
                                 vertex_len,
                                 vertex_first,
                                 handle,
                                 custom_id,
#ifdef WITH_METAL_BACKEND
                                 GPU_shader_uses_ssbo_vertex_fetch(shader_) ? shader_ : nullptr,
#endif
                                 GPU_PRIM_NONE,
                                 0);
}
 
template<class T>
inline void PassBase<T>::draw(gpu::Batch *batch, ResourceHandle handle, uint custom_id)
{
  this->draw(batch, -1, -1, -1, handle, custom_id);
}
 
template<class T>
inline void PassBase<T>::draw_expand(gpu::Batch *batch,
                                     GPUPrimType primitive_type,
                                     uint primitive_len,
                                     uint instance_len,
                                     uint vertex_len,
                                     uint vertex_first,
                                     ResourceHandle handle,
                                     uint custom_id)
{
  if (instance_len == 0 || vertex_len == 0 || primitive_len == 0) {
    return;
  }
  BLI_assert(shader_);
  draw_commands_buf_.append_draw(headers_,
                                 commands_,
                                 batch,
                                 instance_len,
                                 vertex_len,
                                 vertex_first,
                                 handle,
                                 custom_id,
#ifdef WITH_METAL_BACKEND
                                 GPU_shader_uses_ssbo_vertex_fetch(shader_) ? shader_ : nullptr,
#endif
                                 primitive_type,
                                 primitive_len);
}
 
template<class T>
inline void PassBase<T>::draw_expand(gpu::Batch *batch,
                                     GPUPrimType primitive_type,
                                     uint primitive_len,
                                     uint instance_len,
                                     ResourceHandle handle,
                                     uint custom_id)
{
  this->draw_expand(batch, primitive_type, primitive_len, instance_len, -1, -1, handle, custom_id);
}
 
template<class T>
inline void PassBase<T>::draw_procedural(GPUPrimType primitive,
                                         uint instance_len,
                                         uint vertex_len,
                                         uint vertex_first,
                                         ResourceHandle handle,
                                         uint custom_id)
{
  this->draw(
      procedural_batch_get(primitive), instance_len, vertex_len, vertex_first, handle, custom_id);
}

 
/* -------------------------------------------------------------------- */
 
template<class T>
inline void PassBase<T>::draw_indirect(gpu::Batch *batch,
                                       StorageBuffer<DrawCommand, true> &indirect_buffer,
                                       ResourceHandle handle)
{
  BLI_assert(shader_);
  create_command(Type::DrawIndirect).draw_indirect = {batch, &indirect_buffer, handle};
}
 
template<class T>
inline void PassBase<T>::draw_procedural_indirect(
    GPUPrimType primitive,
    StorageBuffer<DrawCommand, true> &indirect_buffer,
    ResourceHandle handle)
{
  this->draw_indirect(procedural_batch_get(primitive), indirect_buffer, handle);
}

 
/* -------------------------------------------------------------------- */
 
template<class T> inline void PassBase<T>::dispatch(int group_len)
{
  BLI_assert(shader_);
  create_command(Type::Dispatch).dispatch = {int3(group_len, 1, 1)};
}
 
template<class T> inline void PassBase<T>::dispatch(int2 group_len)
{
  BLI_assert(shader_);
  create_command(Type::Dispatch).dispatch = {int3(group_len.x, group_len.y, 1)};
}
 
template<class T> inline void PassBase<T>::dispatch(int3 group_len)
{
  BLI_assert(shader_);
  create_command(Type::Dispatch).dispatch = {group_len};
}
 
template<class T> inline void PassBase<T>::dispatch(int3 *group_len)
{
  BLI_assert(shader_);
  create_command(Type::Dispatch).dispatch = {group_len};
}
 
template<class T>
inline void PassBase<T>::dispatch(StorageBuffer<DispatchCommand> &indirect_buffer)
{
  BLI_assert(shader_);
  create_command(Type::DispatchIndirect).dispatch_indirect = {&indirect_buffer};
}

 
/* -------------------------------------------------------------------- */
 
template<class T> inline void PassBase<T>::clear_color(float4 color)
{
  this->clear(GPU_COLOR_BIT, color, 0.0f, 0);
}
 
template<class T> inline void PassBase<T>::clear_depth(float depth)
{
  this->clear(GPU_DEPTH_BIT, float4(0.0f), depth, 0);
}
 
template<class T> inline void PassBase<T>::clear_stencil(uint8_t stencil)
{
  this->clear(GPU_STENCIL_BIT, float4(0.0f), 0.0f, stencil);
}
 
template<class T> inline void PassBase<T>::clear_depth_stencil(float depth, uint8_t stencil)
{
  this->clear(GPU_DEPTH_BIT | GPU_STENCIL_BIT, float4(0.0f), depth, stencil);
}
 
template<class T>
inline void PassBase<T>::clear_color_depth_stencil(float4 color, float depth, uint8_t stencil)
{
  this->clear(GPU_DEPTH_BIT | GPU_STENCIL_BIT | GPU_COLOR_BIT, color, depth, stencil);
}

 
/* -------------------------------------------------------------------- */
 
template<class T> inline void PassBase<T>::barrier(eGPUBarrier type)
{
  create_command(Type::Barrier).barrier = {type};
}

 
/* -------------------------------------------------------------------- */
 
template<class T> inline void PassBase<T>::state_set(DRWState state, int clip_plane_count)
{
  if (clip_plane_count > 0) {
    state |= DRW_STATE_CLIP_PLANES;
  }
  /* Assumed to always be enabled. */
  state |= DRW_STATE_PROGRAM_POINT_SIZE;
  create_command(Type::StateSet).state_set = {state, clip_plane_count};
}
 
template<class T>
inline void PassBase<T>::state_stencil(uint8_t write_mask, uint8_t reference, uint8_t compare_mask)
{
  create_command(Type::StencilSet).stencil_set = {write_mask, compare_mask, reference};
}
 
template<class T> inline void PassBase<T>::shader_set(GPUShader *shader)
{
  shader_ = shader;
  create_command(Type::ShaderBind).shader_bind = {shader};
}
 
template<class T> inline void PassBase<T>::framebuffer_set(GPUFrameBuffer **framebuffer)
{
  create_command(Type::FramebufferBind).framebuffer_bind = {framebuffer};
}
 
template<class T>
inline void PassBase<T>::subpass_transition(GPUAttachmentState depth_attachment,
                                            Span<GPUAttachmentState> color_attachments)
{
  uint8_t color_states[8] = {GPU_ATTACHMENT_IGNORE};
  for (auto i : color_attachments.index_range()) {
    color_states[i] = uint8_t(color_attachments[i]);
  }
  create_command(Type::SubPassTransition).subpass_transition = {uint8_t(depth_attachment),
                                                                {color_states[0],
                                                                 color_states[1],
                                                                 color_states[2],
                                                                 color_states[3],
                                                                 color_states[4],
                                                                 color_states[5],
                                                                 color_states[6],
                                                                 color_states[7]}};
}
 
template<class T> inline void PassBase<T>::material_set(Manager &manager, GPUMaterial *material)
{
  GPUPass *gpupass = GPU_material_get_pass(material);
  shader_set(GPU_pass_shader_get(gpupass));
 
  /* Bind all textures needed by the material. */
  ListBase textures = GPU_material_textures(material);
  for (GPUMaterialTexture *tex : ListBaseWrapper<GPUMaterialTexture>(textures)) {
    if (tex->ima) {
      /* Image */
      const bool use_tile_mapping = tex->tiled_mapping_name[0];
      ImageUser *iuser = tex->iuser_available ? &tex->iuser : nullptr;
      ImageGPUTextures gputex = BKE_image_get_gpu_material_texture(
          tex->ima, iuser, use_tile_mapping);
 
      manager.acquire_texture(gputex.texture);
      bind_texture(tex->sampler_name, gputex.texture, tex->sampler_state);
 
      if (gputex.tile_mapping) {
        manager.acquire_texture(gputex.tile_mapping);
        bind_texture(tex->tiled_mapping_name, gputex.tile_mapping, tex->sampler_state);
      }
    }
    else if (tex->colorband) {
      /* Color Ramp */
      bind_texture(tex->sampler_name, *tex->colorband);
    }
    else if (tex->sky) {
      /* Sky */
      bind_texture(tex->sampler_name, *tex->sky, tex->sampler_state);
    }
  }
 
  GPUUniformBuf *ubo = GPU_material_uniform_buffer_get(material);
  if (ubo != nullptr) {
    bind_ubo(GPU_NODE_TREE_UBO_SLOT, ubo);
  }
}

 
/* -------------------------------------------------------------------- */
 
template<class T> inline int PassBase<T>::push_constant_offset(const char *name)
{
  return GPU_shader_get_uniform(shader_, name);
}
 
template<class T> inline void PassBase<T>::bind_ssbo(const char *name, GPUStorageBuf *buffer)
{
  BLI_assert(buffer != nullptr);
  this->bind_ssbo(GPU_shader_get_ssbo_binding(shader_, name), buffer);
}
 
template<class T> inline void PassBase<T>::bind_ssbo(const char *name, GPUUniformBuf *buffer)
{
  BLI_assert(buffer != nullptr);
  this->bind_ssbo(GPU_shader_get_ssbo_binding(shader_, name), buffer);
}
 
template<class T> inline void PassBase<T>::bind_ssbo(const char *name, GPUUniformBuf **buffer)
{
  BLI_assert(buffer != nullptr);
  this->bind_ssbo(GPU_shader_get_ssbo_binding(shader_, name), buffer);
}
 
template<class T> inline void PassBase<T>::bind_ssbo(const char *name, gpu::VertBuf *buffer)
{
  BLI_assert(buffer != nullptr);
  this->bind_ssbo(GPU_shader_get_ssbo_binding(shader_, name), buffer);
}
 
template<class T> inline void PassBase<T>::bind_ssbo(const char *name, gpu::VertBuf **buffer)
{
  BLI_assert(buffer != nullptr);
  this->bind_ssbo(GPU_shader_get_ssbo_binding(shader_, name), buffer);
}
 
template<class T> inline void PassBase<T>::bind_ssbo(const char *name, gpu::IndexBuf *buffer)
{
  BLI_assert(buffer != nullptr);
  this->bind_ssbo(GPU_shader_get_ssbo_binding(shader_, name), buffer);
}
 
template<class T> inline void PassBase<T>::bind_ssbo(const char *name, gpu::IndexBuf **buffer)
{
  BLI_assert(buffer != nullptr);
  this->bind_ssbo(GPU_shader_get_ssbo_binding(shader_, name), buffer);
}
 
template<class T> inline void PassBase<T>::bind_ubo(const char *name, GPUUniformBuf *buffer)
{
  BLI_assert(buffer != nullptr);
  this->bind_ubo(GPU_shader_get_ubo_binding(shader_, name), buffer);
}
 
template<class T>
inline void PassBase<T>::bind_texture(const char *name, GPUTexture *texture, GPUSamplerState state)
{
  BLI_assert(texture != nullptr);
  this->bind_texture(GPU_shader_get_sampler_binding(shader_, name), texture, state);
}
 
template<class T> inline void PassBase<T>::bind_texture(const char *name, gpu::VertBuf *buffer)
{
  BLI_assert(buffer != nullptr);
  this->bind_texture(GPU_shader_get_sampler_binding(shader_, name), buffer);
}
 
template<class T> inline void PassBase<T>::bind_texture(const char *name, gpu::VertBuf **buffer)
{
  BLI_assert(buffer != nullptr);
  this->bind_texture(GPU_shader_get_sampler_binding(shader_, name), buffer);
}
 
template<class T> inline void PassBase<T>::bind_image(const char *name, GPUTexture *image)
{
  BLI_assert(image != nullptr);
  this->bind_image(GPU_shader_get_sampler_binding(shader_, name), image);
}
 
template<class T> inline void PassBase<T>::bind_ssbo(int slot, GPUStorageBuf *buffer)
{
  BLI_assert(buffer != nullptr);
  create_command(Type::ResourceBind).resource_bind = {slot, buffer};
}
 
template<class T> inline void PassBase<T>::bind_ssbo(int slot, GPUUniformBuf *buffer)
{
  BLI_assert(buffer != nullptr);
  create_command(Type::ResourceBind).resource_bind = {
      slot, buffer, ResourceBind::Type::UniformAsStorageBuf};
}
 
template<class T> inline void PassBase<T>::bind_ssbo(int slot, GPUUniformBuf **buffer)
{
  BLI_assert(buffer != nullptr);
  create_command(Type::ResourceBind).resource_bind = {
      slot, buffer, ResourceBind::Type::UniformAsStorageBuf};
}
 
template<class T> inline void PassBase<T>::bind_ssbo(int slot, gpu::VertBuf *buffer)
{
  BLI_assert(buffer != nullptr);
  create_command(Type::ResourceBind).resource_bind = {
      slot, buffer, ResourceBind::Type::VertexAsStorageBuf};
}
 
template<class T> inline void PassBase<T>::bind_ssbo(int slot, gpu::VertBuf **buffer)
{
  BLI_assert(buffer != nullptr);
  create_command(Type::ResourceBind).resource_bind = {
      slot, buffer, ResourceBind::Type::VertexAsStorageBuf};
}
 
template<class T> inline void PassBase<T>::bind_ssbo(int slot, gpu::IndexBuf *buffer)
{
  BLI_assert(buffer != nullptr);
  create_command(Type::ResourceBind).resource_bind = {
      slot, buffer, ResourceBind::Type::IndexAsStorageBuf};
}
 
template<class T> inline void PassBase<T>::bind_ssbo(int slot, gpu::IndexBuf **buffer)
{
  BLI_assert(buffer != nullptr);
  create_command(Type::ResourceBind).resource_bind = {
      slot, buffer, ResourceBind::Type::IndexAsStorageBuf};
}
 
template<class T> inline void PassBase<T>::bind_ubo(int slot, GPUUniformBuf *buffer)
{
  BLI_assert(buffer != nullptr);
  create_command(Type::ResourceBind).resource_bind = {slot, buffer};
}
 
template<class T>
inline void PassBase<T>::bind_texture(int slot, GPUTexture *texture, GPUSamplerState state)
{
  BLI_assert(texture != nullptr);
  create_command(Type::ResourceBind).resource_bind = {slot, texture, state};
}
 
template<class T> inline void PassBase<T>::bind_texture(int slot, gpu::VertBuf *buffer)
{
  BLI_assert(buffer != nullptr);
  create_command(Type::ResourceBind).resource_bind = {slot, buffer};
}
 
template<class T> inline void PassBase<T>::bind_texture(int slot, gpu::VertBuf **buffer)
{
  BLI_assert(buffer != nullptr);
  create_command(Type::ResourceBind).resource_bind = {slot, buffer};
}
 
template<class T> inline void PassBase<T>::bind_image(int slot, GPUTexture *image)
{
  BLI_assert(image != nullptr);
  create_command(Type::ResourceBind).resource_bind = {slot, as_image(image)};
}
 
template<class T> inline void PassBase<T>::bind_ssbo(const char *name, GPUStorageBuf **buffer)
{
  BLI_assert(buffer != nullptr);
  this->bind_ssbo(GPU_shader_get_ssbo_binding(shader_, name), buffer);
}
 
template<class T> inline void PassBase<T>::bind_ubo(const char *name, GPUUniformBuf **buffer)
{
  BLI_assert(buffer != nullptr);
  this->bind_ubo(GPU_shader_get_ubo_binding(shader_, name), buffer);
}
 
template<class T>
inline void PassBase<T>::bind_texture(const char *name,
                                      GPUTexture **texture,
                                      GPUSamplerState state)
{
  BLI_assert(texture != nullptr);
  this->bind_texture(GPU_shader_get_sampler_binding(shader_, name), texture, state);
}
 
template<class T> inline void PassBase<T>::bind_image(const char *name, GPUTexture **image)
{
  BLI_assert(image != nullptr);
  this->bind_image(GPU_shader_get_sampler_binding(shader_, name), image);
}
 
template<class T> inline void PassBase<T>::bind_ssbo(int slot, GPUStorageBuf **buffer)
{
 
  BLI_assert(buffer != nullptr);
  create_command(Type::ResourceBind).resource_bind = {slot, buffer};
}
 
template<class T> inline void PassBase<T>::bind_ubo(int slot, GPUUniformBuf **buffer)
{
  BLI_assert(buffer != nullptr);
  create_command(Type::ResourceBind).resource_bind = {slot, buffer};
}
 
template<class T>
inline void PassBase<T>::bind_texture(int slot, GPUTexture **texture, GPUSamplerState state)
{
  BLI_assert(texture != nullptr);
  create_command(Type::ResourceBind).resource_bind = {slot, texture, state};
}
 
template<class T> inline void PassBase<T>::bind_image(int slot, GPUTexture **image)
{
  BLI_assert(image != nullptr);
  create_command(Type::ResourceBind).resource_bind = {slot, as_image(image)};
}

 
/* -------------------------------------------------------------------- */
 
template<class T> inline void PassBase<T>::push_constant(const char *name, const float &data)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data};
}
 
template<class T> inline void PassBase<T>::push_constant(const char *name, const float2 &data)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data};
}
 
template<class T> inline void PassBase<T>::push_constant(const char *name, const float3 &data)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data};
}
 
template<class T> inline void PassBase<T>::push_constant(const char *name, const float4 &data)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data};
}
 
template<class T> inline void PassBase<T>::push_constant(const char *name, const int &data)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data};
}
 
template<class T> inline void PassBase<T>::push_constant(const char *name, const int2 &data)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data};
}
 
template<class T> inline void PassBase<T>::push_constant(const char *name, const int3 &data)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data};
}
 
template<class T> inline void PassBase<T>::push_constant(const char *name, const int4 &data)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data};
}
 
template<class T> inline void PassBase<T>::push_constant(const char *name, const bool &data)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data};
}
 
template<class T>
inline void PassBase<T>::push_constant(const char *name, const float *data, int array_len)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data, array_len};
}
 
template<class T>
inline void PassBase<T>::push_constant(const char *name, const float2 *data, int array_len)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data, array_len};
}
 
template<class T>
inline void PassBase<T>::push_constant(const char *name, const float3 *data, int array_len)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data, array_len};
}
 
template<class T>
inline void PassBase<T>::push_constant(const char *name, const float4 *data, int array_len)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data, array_len};
}
 
template<class T>
inline void PassBase<T>::push_constant(const char *name, const int *data, int array_len)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data, array_len};
}
 
template<class T>
inline void PassBase<T>::push_constant(const char *name, const int2 *data, int array_len)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data, array_len};
}
 
template<class T>
inline void PassBase<T>::push_constant(const char *name, const int3 *data, int array_len)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data, array_len};
}
 
template<class T>
inline void PassBase<T>::push_constant(const char *name, const int4 *data, int array_len)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data, array_len};
}
 
template<class T> inline void PassBase<T>::push_constant(const char *name, const float4x4 *data)
{
  create_command(Type::PushConstant).push_constant = {push_constant_offset(name), data};
}
 
template<class T> inline void PassBase<T>::push_constant(const char *name, const float4x4 &data)
{
  /* WORKAROUND: Push 3 consecutive commands to hold the 64 bytes of the float4x4.
   * This assumes that all commands are always stored in flat array of memory. */
  Undetermined commands[3];
 
  PushConstant &cmd = commands[0].push_constant;
  cmd.location = push_constant_offset(name);
  cmd.array_len = 1;
  cmd.comp_len = 16;
  cmd.type = PushConstant::Type::FloatValue;
  /* Copy overrides the next 2 commands. We append them as Type::None to not evaluate them. */
  *reinterpret_cast<float4x4 *>(&cmd.float4_value) = data;
 
  create_command(Type::PushConstant) = commands[0];
  create_command(Type::None) = commands[1];
  create_command(Type::None) = commands[2];
}

 
/* -------------------------------------------------------------------- */
 
template<class T>
inline void PassBase<T>::specialize_constant(GPUShader *shader,
                                             const char *constant_name,
                                             const int &constant_value)
{
  create_command(Type::SpecializeConstant).specialize_constant = {
      shader, GPU_shader_get_constant(shader, constant_name), constant_value};
}
 
template<class T>
inline void PassBase<T>::specialize_constant(GPUShader *shader,
                                             const char *constant_name,
                                             const uint &constant_value)
{
  create_command(Type::SpecializeConstant).specialize_constant = {
      shader, GPU_shader_get_constant(shader, constant_name), constant_value};
}
 
template<class T>
inline void PassBase<T>::specialize_constant(GPUShader *shader,
                                             const char *constant_name,
                                             const float &constant_value)
{
  create_command(Type::SpecializeConstant).specialize_constant = {
      shader, GPU_shader_get_constant(shader, constant_name), constant_value};
}
 
template<class T>
inline void PassBase<T>::specialize_constant(GPUShader *shader,
                                             const char *constant_name,
                                             const bool &constant_value)
{
  create_command(Type::SpecializeConstant).specialize_constant = {
      shader, GPU_shader_get_constant(shader, constant_name), constant_value};
}
 
template<class T>
inline void PassBase<T>::specialize_constant(GPUShader *shader,
                                             const char *constant_name,
                                             const int *constant_value)
{
  create_command(Type::SpecializeConstant).specialize_constant = {
      shader, GPU_shader_get_constant(shader, constant_name), constant_value};
}
 
template<class T>
inline void PassBase<T>::specialize_constant(GPUShader *shader,
                                             const char *constant_name,
                                             const uint *constant_value)
{
  create_command(Type::SpecializeConstant).specialize_constant = {
      shader, GPU_shader_get_constant(shader, constant_name), constant_value};
}
 
template<class T>
inline void PassBase<T>::specialize_constant(GPUShader *shader,
                                             const char *constant_name,
                                             const float *constant_value)
{
  create_command(Type::SpecializeConstant).specialize_constant = {
      shader, GPU_shader_get_constant(shader, constant_name), constant_value};
}
 
template<class T>
inline void PassBase<T>::specialize_constant(GPUShader *shader,
                                             const char *constant_name,
                                             const bool *constant_value)
{
  create_command(Type::SpecializeConstant).specialize_constant = {
      shader, GPU_shader_get_constant(shader, constant_name), constant_value};
}

 
}  // namespace detail
 
}  // namespace blender::draw