Batch rendering using Vulkan's `VkDrawIndexedIndirectCommand` calls

[mrparmesan01]

Brief Task Overview

The goal of the VkDrawIndexedIndirectCommand is to allow us to reduce the draw calls that we need to make. This is also going to have a significant change to atlas::mesh_source.

Where there are consideration to redefining atlas::mesh_source:

struct mesh_source {
    uint32_t    index_count;    // Number of indices to draw
    uint32_t    instance_count; // Number of instances (usually 1)
    uint32_t    first_index;    // Offset into the index buffer
    int32_t     vertex_offset;  // Offset added to vertex index
    uint32_t    first_instance; // ID of the first instance (becomes gl_InstanceIndex)
};

Rendering with a single draw call

This is a brief example for performing this draw operation.

std::array<VkDrawIndexedIndirectCommand, 3> draw_commands{};

// Entity 1
drawCommands[0] = {
    .indexCount    = 1024;
    .instanceCount = 1,
    .firstIndex    = 0,
    .vertexOffset  = 0,
    .firstInstance = 0, // Points to Manifest[0]
};

// Entity 2
drawCommands[1] = {
    .indexCount = 512;
    .instanceCount = 1,
    .firstIndex    = 0,
    .vertexOffset  = 0,
    .firstInstance = 1, // Points to Manifest[1]
};

// Entity 3
drawCommands[2] = {
    .indexCount    = 2048,
    .instanceCount = 1,
    .firstIndex    = 0,
    .vertexOffset  = 0,
    .firstInstance = 2, // Points to Manifest[2]
};

std::span<const VkDrawIndexedIndirectCommand> draw_command_submissions = draw_commands;

const vk::buffer indirect_buffer = ...; // configured
uint32_t offset = 0;

// NOTE: these draw commands go to a VkBuffer handle
// must be configured with vk::buffer_usage::indirect_buffer_bit
vkCmdDrawIndexedIndirect(
  command,
  indirect_buffer,
  offset,
  static_cast<uint32_t>(draw_command_submission.size()),
  static_cast<uint32_t>(draw_command_submission.size_bytes()));