gpu_shader_create_info.cc

Go to the documentation of this file.

/* SPDX-FileCopyrightText: 2021 Blender Authors
 *
 * SPDX-License-Identifier: GPL-2.0-or-later */

 
#include "BLI_map.hh"
#include "BLI_set.hh"
#include "BLI_string_ref.hh"
 
#include "BKE_global.hh"
 
#include "GPU_capabilities.hh"
#include "GPU_context.hh"
#include "GPU_platform.hh"
#include "GPU_shader.hh"
 
#include "gpu_shader_create_info.hh"
#include "gpu_shader_create_info_private.hh"
#include "gpu_shader_dependency_private.hh"
 
#undef GPU_SHADER_NAMED_INTERFACE_INFO
#undef GPU_SHADER_INTERFACE_INFO
#undef GPU_SHADER_CREATE_INFO
#undef GPU_SHADER_NAMED_INTERFACE_END
#undef GPU_SHADER_INTERFACE_END
#undef GPU_SHADER_CREATE_END
 
namespace blender::gpu::shader {
 
using CreateInfoDictionnary = Map<StringRef, ShaderCreateInfo *>;
using InterfaceDictionnary = Map<StringRef, StageInterfaceInfo *>;
 
static CreateInfoDictionnary *g_create_infos = nullptr;
static InterfaceDictionnary *g_interfaces = nullptr;
 
/* -------------------------------------------------------------------- */
 
static bool is_vulkan_compatible_interface(const StageInterfaceInfo &iface)
{
  if (iface.instance_name.is_empty()) {
    return true;
  }
 
  bool use_flat = false;
  bool use_smooth = false;
  bool use_noperspective = false;
  for (const StageInterfaceInfo::InOut &attr : iface.inouts) {
    switch (attr.interp) {
      case Interpolation::FLAT:
        use_flat = true;
        break;
      case Interpolation::SMOOTH:
        use_smooth = true;
        break;
      case Interpolation::NO_PERSPECTIVE:
        use_noperspective = true;
        break;
    }
  }
  int num_used_interpolation_types = (use_flat ? 1 : 0) + (use_smooth ? 1 : 0) +
                                     (use_noperspective ? 1 : 0);
 
#if 0
  if (num_used_interpolation_types > 1) {
    std::cout << "'" << iface.name << "' uses multiple interpolation types\n";
  }
#endif
 
  return num_used_interpolation_types <= 1;
}
 
bool ShaderCreateInfo::is_vulkan_compatible() const
{
  /* Vulkan doesn't support setting an interpolation mode per attribute in a struct. */
  for (const StageInterfaceInfo *iface : vertex_out_interfaces_) {
    if (!is_vulkan_compatible_interface(*iface)) {
      return false;
    }
  }
  for (const StageInterfaceInfo *iface : geometry_out_interfaces_) {
    if (!is_vulkan_compatible_interface(*iface)) {
      return false;
    }
  }
 
  return true;
}

 
void ShaderCreateInfo::finalize(const bool recursive)
{
  if (finalized_) {
    return;
  }
  finalized_ = true;
 
  Set<StringRefNull> deps_merged;
 
  validate_vertex_attributes();
 
  for (auto &info_name : additional_infos_) {
 
    /* Fetch create info. */
    const ShaderCreateInfo &info = *reinterpret_cast<const ShaderCreateInfo *>(
        gpu_shader_create_info_get(info_name.c_str()));
 
    if (recursive) {
      const_cast<ShaderCreateInfo &>(info).finalize(recursive);
    }
    else {
      BLI_assert(info.finalized_);
    }
 
    interface_names_size_ += info.interface_names_size_;
 
    /* NOTE: EEVEE Materials can result in nested includes. To avoid duplicate
     * shader resources, we need to avoid inserting duplicates.
     * TODO: Optimize create info preparation to include each individual "additional_info"
     * only a single time. */
    vertex_inputs_.extend_non_duplicates(info.vertex_inputs_);
    fragment_outputs_.extend_non_duplicates(info.fragment_outputs_);
    vertex_out_interfaces_.extend_non_duplicates(info.vertex_out_interfaces_);
    geometry_out_interfaces_.extend_non_duplicates(info.geometry_out_interfaces_);
    subpass_inputs_.extend_non_duplicates(info.subpass_inputs_);
    specialization_constants_.extend_non_duplicates(info.specialization_constants_);
    compilation_constants_.extend_non_duplicates(info.compilation_constants_);
 
    validate_vertex_attributes(&info);
 
    /* Insert with duplicate check. */
    push_constants_.extend_non_duplicates(info.push_constants_);
    defines_.extend_non_duplicates(info.defines_);
    batch_resources_.extend_non_duplicates(info.batch_resources_);
    pass_resources_.extend_non_duplicates(info.pass_resources_);
    geometry_resources_.extend_non_duplicates(info.geometry_resources_);
    typedef_sources_.extend_non_duplicates(info.typedef_sources_);
 
    /* API-specific parameters.
     * We will only copy API-specific parameters if they are otherwise unassigned. */
#ifdef WITH_METAL_BACKEND
    if (mtl_max_threads_per_threadgroup_ == 0) {
      mtl_max_threads_per_threadgroup_ = info.mtl_max_threads_per_threadgroup_;
    }
#endif
 
    if (info.early_fragment_test_) {
      early_fragment_test_ = true;
      depth_write_ = DepthWrite::UNCHANGED;
    }
    /* Modify depth write if has been changed from default.
     * `UNCHANGED` implies gl_FragDepth is not used at all. */
    if (info.depth_write_ != DepthWrite::UNCHANGED) {
      depth_write_ = info.depth_write_;
    }
 
    /* Inherit builtin bits from additional info. */
    builtins_ |= info.builtins_;
 
    validate_merge(info);
 
    auto assert_no_overlap = [&](const bool test, const StringRefNull error) {
      if (!test) {
        std::cout << name_ << ": Validation failed while merging " << info.name_ << " : ";
        std::cout << error << std::endl;
        BLI_assert(0);
      }
    };
 
    if (!deps_merged.add(info.name_)) {
      assert_no_overlap(false, "additional info already merged via another info");
    }
 
    if (info.compute_layout_.local_size_x != -1) {
      assert_no_overlap(compute_layout_.local_size_x == -1, "Compute layout already defined");
      compute_layout_ = info.compute_layout_;
    }
 
    if (!info.vertex_source_.is_empty()) {
      assert_no_overlap(vertex_source_.is_empty(), "Vertex source already existing");
      vertex_source_ = info.vertex_source_;
    }
    if (!info.geometry_source_.is_empty()) {
      assert_no_overlap(geometry_source_.is_empty(), "Geometry source already existing");
      geometry_source_ = info.geometry_source_;
      geometry_layout_ = info.geometry_layout_;
    }
    if (!info.fragment_source_.is_empty()) {
      assert_no_overlap(fragment_source_.is_empty(), "Fragment source already existing");
      fragment_source_ = info.fragment_source_;
    }
    if (!info.compute_source_.is_empty()) {
      assert_no_overlap(compute_source_.is_empty(), "Compute source already existing");
      compute_source_ = info.compute_source_;
    }
  }
 
  if (!geometry_source_.is_empty() && bool(builtins_ & BuiltinBits::LAYER)) {
    std::cout << name_
              << ": Validation failed. BuiltinBits::LAYER shouldn't be used with geometry shaders."
              << std::endl;
    BLI_assert(0);
  }
 
  if (auto_resource_location_) {
    int images = 0, samplers = 0, ubos = 0, ssbos = 0;
 
    auto set_resource_slot = [&](Resource &res) {
      switch (res.bind_type) {
        case Resource::BindType::UNIFORM_BUFFER:
          res.slot = ubos++;
          break;
        case Resource::BindType::STORAGE_BUFFER:
          res.slot = ssbos++;
          break;
        case Resource::BindType::SAMPLER:
          res.slot = samplers++;
          break;
        case Resource::BindType::IMAGE:
          res.slot = images++;
          break;
      }
    };
 
    for (auto &res : batch_resources_) {
      set_resource_slot(res);
    }
    for (auto &res : pass_resources_) {
      set_resource_slot(res);
    }
    for (auto &res : geometry_resources_) {
      set_resource_slot(res);
    }
  }
}
 
std::string ShaderCreateInfo::check_error() const
{
  std::string error;
 
  /* At least a vertex shader and a fragment shader are required, or only a compute shader. */
  if (this->compute_source_.is_empty()) {
    if (this->vertex_source_.is_empty()) {
      error += "Missing vertex shader in " + this->name_ + ".\n";
    }
    if (this->fragment_source_.is_empty()) {
      error += "Missing fragment shader in " + this->name_ + ".\n";
    }
  }
  else {
    if (!this->vertex_source_.is_empty()) {
      error += "Compute shader has vertex_source_ shader attached in " + this->name_ + ".\n";
    }
    if (!this->geometry_source_.is_empty()) {
      error += "Compute shader has geometry_source_ shader attached in " + this->name_ + ".\n";
    }
    if (!this->fragment_source_.is_empty()) {
      error += "Compute shader has fragment_source_ shader attached in " + this->name_ + ".\n";
    }
  }
 
  if (!this->geometry_source_.is_empty()) {
    if (bool(this->builtins_ & BuiltinBits::BARYCENTRIC_COORD)) {
      error += "Shader " + this->name_ +
               " has geometry stage and uses barycentric coordinates. This is not allowed as "
               "fallback injects a geometry stage.\n";
    }
    if (bool(this->builtins_ & BuiltinBits::VIEWPORT_INDEX)) {
      error += "Shader " + this->name_ +
               " has geometry stage and uses multi-viewport. This is not allowed as "
               "fallback injects a geometry stage.\n";
    }
    if (bool(this->builtins_ & BuiltinBits::LAYER)) {
      error += "Shader " + this->name_ +
               " has geometry stage and uses layer output. This is not allowed as "
               "fallback injects a geometry stage.\n";
    }
  }
 
  if ((G.debug & G_DEBUG_GPU) == 0) {
    return error;
  }
 
  if (bool(this->builtins_ &
           (BuiltinBits::BARYCENTRIC_COORD | BuiltinBits::VIEWPORT_INDEX | BuiltinBits::LAYER)))
  {
    for (const StageInterfaceInfo *interface : this->vertex_out_interfaces_) {
      if (interface->instance_name.is_empty()) {
        error += "Shader " + this->name_ + " uses interface " + interface->name +
                 " that doesn't contain an instance name, but is required for the fallback "
                 "geometry shader.\n";
      }
    }
  }
 
  if (!this->is_vulkan_compatible()) {
    error += this->name_ +
             " contains a stage interface using an instance name and mixed interpolation modes. "
             "This is not compatible with Vulkan and need to be adjusted.\n";
  }
 
  /* Validate specialization constants. */
  for (int i = 0; i < specialization_constants_.size(); i++) {
    for (int j = i + 1; j < specialization_constants_.size(); j++) {
      if (specialization_constants_[i].name == specialization_constants_[j].name) {
        error += this->name_ + " contains two specialization constants with the name: " +
                 std::string(specialization_constants_[i].name);
      }
    }
  }
 
  /* Validate compilation constants. */
  for (int i = 0; i < compilation_constants_.size(); i++) {
    for (int j = i + 1; j < compilation_constants_.size(); j++) {
      if (compilation_constants_[i].name == compilation_constants_[j].name) {
        error += this->name_ + " contains two compilation constants with the name: " +
                 std::string(compilation_constants_[i].name);
      }
    }
  }
 
  return error;
}
 
void ShaderCreateInfo::validate_merge(const ShaderCreateInfo &other_info)
{
  if (!auto_resource_location_) {
    /* Check same bind-points usage in OGL. */
    Set<int> images, samplers, ubos, ssbos;
 
    auto register_resource = [&](const Resource &res) -> bool {
      switch (res.bind_type) {
        case Resource::BindType::UNIFORM_BUFFER:
          return images.add(res.slot);
        case Resource::BindType::STORAGE_BUFFER:
          return samplers.add(res.slot);
        case Resource::BindType::SAMPLER:
          return ubos.add(res.slot);
        case Resource::BindType::IMAGE:
          return ssbos.add(res.slot);
        default:
          return false;
      }
    };
 
    auto print_error_msg = [&](const Resource &res, const Vector<Resource> &resources) {
      auto print_resource_name = [&](const Resource &res) {
        switch (res.bind_type) {
          case Resource::BindType::UNIFORM_BUFFER:
            std::cout << "Uniform Buffer " << res.uniformbuf.name;
            break;
          case Resource::BindType::STORAGE_BUFFER:
            std::cout << "Storage Buffer " << res.storagebuf.name;
            break;
          case Resource::BindType::SAMPLER:
            std::cout << "Sampler " << res.sampler.name;
            break;
          case Resource::BindType::IMAGE:
            std::cout << "Image " << res.image.name;
            break;
          default:
            std::cout << "Unknown Type";
            break;
        }
      };
 
      for (const Resource &_res : resources) {
        if (&res != &_res && res.bind_type == _res.bind_type && res.slot == _res.slot) {
          std::cout << name_ << ": Validation failed : Overlapping ";
          print_resource_name(res);
          std::cout << " and ";
          print_resource_name(_res);
          std::cout << " at (" << res.slot << ") while merging " << other_info.name_ << std::endl;
        }
      }
    };
 
    for (auto &res : batch_resources_) {
      if (register_resource(res) == false) {
        print_error_msg(res, resources_get_all_());
      }
    }
 
    for (auto &res : pass_resources_) {
      if (register_resource(res) == false) {
        print_error_msg(res, resources_get_all_());
      }
    }
 
    for (auto &res : geometry_resources_) {
      if (register_resource(res) == false) {
        print_error_msg(res, resources_get_all_());
      }
    }
  }
}
 
void ShaderCreateInfo::validate_vertex_attributes(const ShaderCreateInfo *other_info)
{
  uint32_t attr_bits = 0;
  for (auto &attr : vertex_inputs_) {
    if (attr.index >= 16 || attr.index < 0) {
      std::cout << name_ << ": \"" << attr.name
                << "\" : Type::float3x3_t unsupported as vertex attribute." << std::endl;
      BLI_assert(0);
    }
    if (attr.index >= 16 || attr.index < 0) {
      std::cout << name_ << ": Invalid index for attribute \"" << attr.name << "\"" << std::endl;
      BLI_assert(0);
    }
    uint32_t attr_new = 0;
    if (attr.type == Type::float4x4_t) {
      for (int i = 0; i < 4; i++) {
        attr_new |= 1 << (attr.index + i);
      }
    }
    else {
      attr_new |= 1 << attr.index;
    }
 
    if ((attr_bits & attr_new) != 0) {
      std::cout << name_ << ": Attribute \"" << attr.name
                << "\" overlap one or more index from another attribute."
                   " Note that mat4 takes up 4 indices.";
      if (other_info) {
        std::cout << " While merging " << other_info->name_ << std::endl;
      }
      std::cout << std::endl;
      BLI_assert(0);
    }
    attr_bits |= attr_new;
  }
}
 
}  // namespace blender::gpu::shader
 
using namespace blender::gpu::shader;
 
#ifdef _MSC_VER
/* Disable optimization for this function with MSVC. It does not like the fact
 * shaders info are declared in the same function (same basic block or not does
 * not change anything).
 * Since it is just a function called to register shaders (once),
 * the fact it's optimized or not does not matter, it's not on any hot
 * code path. */
#  pragma optimize("", off)
#endif
void gpu_shader_create_info_init()
{
  g_create_infos = new CreateInfoDictionnary();
  g_interfaces = new InterfaceDictionnary();
 
#define GPU_SHADER_NAMED_INTERFACE_INFO(_interface, _inst_name) \
  StageInterfaceInfo *ptr_##_interface = new StageInterfaceInfo(#_interface, #_inst_name); \
  StageInterfaceInfo &_interface = *ptr_##_interface; \
  g_interfaces->add_new(#_interface, ptr_##_interface); \
  _interface
 
#define GPU_SHADER_INTERFACE_INFO(_interface) \
  StageInterfaceInfo *ptr_##_interface = new StageInterfaceInfo(#_interface); \
  StageInterfaceInfo &_interface = *ptr_##_interface; \
  g_interfaces->add_new(#_interface, ptr_##_interface); \
  _interface
 
#define GPU_SHADER_CREATE_INFO(_info) \
  ShaderCreateInfo *ptr_##_info = new ShaderCreateInfo(#_info); \
  ShaderCreateInfo &_info = *ptr_##_info; \
  g_create_infos->add_new(#_info, ptr_##_info); \
  _info
 
#define GPU_SHADER_NAMED_INTERFACE_END(_inst_name) ;
#define GPU_SHADER_INTERFACE_END() ;
#define GPU_SHADER_CREATE_END() ;
 
/* Declare, register and construct the infos. */
#include "gpu_shader_create_info_list.hh"
 
  /* WORKAROUND: Replace the use of gpu_BaseInstance by an instance attribute. */
  if (GPU_shader_draw_parameters_support() == false) {
    draw_resource_id = draw_resource_id_fallback;
    draw_resource_with_custom_id = draw_resource_with_custom_id_fallback;
  }
 
  if (GPU_stencil_clasify_buffer_workaround()) {
    /* WORKAROUND: Adding a dummy buffer that isn't used fixes a bug inside the Qualcomm driver. */
    eevee_deferred_tile_classify.storage_buf(
        12, Qualifier::read_write, "uint", "dummy_workaround_buf[]");
  }
 
  for (ShaderCreateInfo *info : g_create_infos->values()) {
    info->builtins_ |= gpu_shader_dependency_get_builtins(info->vertex_source_);
    info->builtins_ |= gpu_shader_dependency_get_builtins(info->fragment_source_);
    info->builtins_ |= gpu_shader_dependency_get_builtins(info->geometry_source_);
    info->builtins_ |= gpu_shader_dependency_get_builtins(info->compute_source_);
 
#if GPU_SHADER_PRINTF_ENABLE
    const bool is_material_shader = info->name_.startswith("eevee_surf_");
    if ((info->builtins_ & BuiltinBits::USE_PRINTF) == BuiltinBits::USE_PRINTF ||
        (gpu_shader_dependency_force_gpu_print_injection() && is_material_shader))
    {
      info->additional_info("gpu_print");
    }
#endif
 
#ifndef NDEBUG
    /* Automatically amend the create info for ease of use of the debug feature. */
    if ((info->builtins_ & BuiltinBits::USE_DEBUG_DRAW) == BuiltinBits::USE_DEBUG_DRAW) {
      info->additional_info("draw_debug_draw");
    }
#endif
  }
 
  for (ShaderCreateInfo *info : g_create_infos->values()) {
    info->finalize(true);
  }
 
  /* TEST */
  // gpu_shader_create_info_compile(nullptr);
}
#ifdef _MSC_VER
#  pragma optimize("", on)
#endif
 
void gpu_shader_create_info_exit()
{
  for (auto *value : g_create_infos->values()) {
    delete value;
  }
  delete g_create_infos;
 
  for (auto *value : g_interfaces->values()) {
    delete value;
  }
  delete g_interfaces;
}
 
bool gpu_shader_create_info_compile(const char *name_starts_with_filter)
{
  using namespace blender;
  using namespace blender::gpu;
  int success = 0;
  int skipped_filter = 0;
  int skipped = 0;
  int total = 0;
 
  Vector<const GPUShaderCreateInfo *> infos;
 
  for (ShaderCreateInfo *info : g_create_infos->values()) {
    info->finalize();
    if (info->do_static_compilation_) {
      if (name_starts_with_filter &&
          !info->name_.startswith(blender::StringRefNull(name_starts_with_filter)))
      {
        skipped_filter++;
        continue;
      }
      if ((info->metal_backend_only_ && GPU_backend_get_type() != GPU_BACKEND_METAL) ||
          (GPU_geometry_shader_support() == false && info->geometry_source_ != nullptr))
      {
        skipped++;
        continue;
      }
      total++;
 
      infos.append(reinterpret_cast<const GPUShaderCreateInfo *>(info));
    }
  }
 
  BatchHandle batch = GPU_shader_batch_create_from_infos(infos);
  Vector<GPUShader *> result = GPU_shader_batch_finalize(batch);
 
  for (int i : result.index_range()) {
    const ShaderCreateInfo *info = reinterpret_cast<const ShaderCreateInfo *>(infos[i]);
    if (result[i] == nullptr) {
      std::cerr << "Compilation " << info->name_.c_str() << " Failed\n";
    }
    else {
      success++;
#if 0 /* TODO(fclem): This is too verbose for now. Make it a cmake option. */
        /* Test if any resource is optimized out and print a warning if that's the case. */
        /* TODO(fclem): Limit this to OpenGL backend. */
        const ShaderInterface *interface = unwrap(shader)->interface;
 
        blender::Vector<ShaderCreateInfo::Resource> all_resources = info->resources_get_all_();
 
        for (ShaderCreateInfo::Resource &res : all_resources) {
          blender::StringRefNull name = "";
          const ShaderInput *input = nullptr;
 
          switch (res.bind_type) {
            case ShaderCreateInfo::Resource::BindType::UNIFORM_BUFFER:
              input = interface->ubo_get(res.slot);
              name = res.uniformbuf.name;
              break;
            case ShaderCreateInfo::Resource::BindType::STORAGE_BUFFER:
              input = interface->ssbo_get(res.slot);
              name = res.storagebuf.name;
              break;
            case ShaderCreateInfo::Resource::BindType::SAMPLER:
              input = interface->texture_get(res.slot);
              name = res.sampler.name;
              break;
            case ShaderCreateInfo::Resource::BindType::IMAGE:
              input = interface->texture_get(res.slot);
              name = res.image.name;
              break;
          }
 
          if (input == nullptr) {
            std::cerr << "Error: " << info->name_;
            std::cerr << ": Resource « " << name << " » not found in the shader interface\n";
          }
          else if (input->location == -1) {
            std::cerr << "Warning: " << info->name_;
            std::cerr << ": Resource « " << name << " » is optimized out\n";
          }
        }
#endif
      GPU_shader_free(result[i]);
    }
  }
 
  printf("Shader Test compilation result: %d / %d passed", success, total);
  if (skipped_filter > 0) {
    printf(" (skipped %d when filtering)", skipped_filter);
  }
  if (skipped > 0) {
    printf(" (skipped %d for compatibility reasons)", skipped);
  }
  printf("\n");
  return success == total;
}
 
const GPUShaderCreateInfo *gpu_shader_create_info_get(const char *info_name)
{
  if (g_create_infos->contains(info_name) == false) {
    printf("Error: Cannot find shader create info named \"%s\"\n", info_name);
    return nullptr;
  }
  ShaderCreateInfo *info = g_create_infos->lookup(info_name);
  return reinterpret_cast<const GPUShaderCreateInfo *>(info);
}