/**************************************************************************/ /* rendering_device_vulkan.cpp */ /**************************************************************************/ /* This file is part of: */ /* GODOT ENGINE */ /* https://godotengine.org */ /**************************************************************************/ /* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */ /* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */ /* */ /* Permission is hereby granted, free of charge, to any person obtaining */ /* a copy of this software and associated documentation files (the */ /* "Software"), to deal in the Software without restriction, including */ /* without limitation the rights to use, copy, modify, merge, publish, */ /* distribute, sublicense, and/or sell copies of the Software, and to */ /* permit persons to whom the Software is furnished to do so, subject to */ /* the following conditions: */ /* */ /* The above copyright notice and this permission notice shall be */ /* included in all copies or substantial portions of the Software. */ /* */ /* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */ /* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */ /* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. */ /* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */ /* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */ /* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */ /* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */ /**************************************************************************/ #include "rendering_device_vulkan.h" #include "core/config/project_settings.h" #include "core/io/compression.h" #include "core/io/dir_access.h" #include "core/io/file_access.h" #include "core/io/marshalls.h" #include "core/os/os.h" #include "core/templates/hashfuncs.h" #include "drivers/vulkan/vulkan_context.h" #include "thirdparty/misc/smolv.h" //#define FORCE_FULL_BARRIER static const uint32_t SMALL_ALLOCATION_MAX_SIZE = 4096; // Get the Vulkan object information and possible stage access types (bitwise OR'd with incoming values). RenderingDeviceVulkan::Buffer *RenderingDeviceVulkan::_get_buffer_from_owner(RID p_buffer, VkPipelineStageFlags &r_stage_mask, VkAccessFlags &r_access_mask, BitField<BarrierMask> p_post_barrier) { Buffer *buffer = nullptr; if (vertex_buffer_owner.owns(p_buffer)) { buffer = vertex_buffer_owner.get_or_null(p_buffer); r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT; r_access_mask |= VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT; if (buffer->usage & VK_BUFFER_USAGE_STORAGE_BUFFER_BIT) { if (p_post_barrier.has_flag(BARRIER_MASK_VERTEX)) { r_access_mask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_FRAGMENT)) { r_access_mask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; r_stage_mask |= VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_COMPUTE)) { r_access_mask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; r_stage_mask |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; } } } else if (index_buffer_owner.owns(p_buffer)) { r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT; r_access_mask |= VK_ACCESS_INDEX_READ_BIT; buffer = index_buffer_owner.get_or_null(p_buffer); } else if (uniform_buffer_owner.owns(p_buffer)) { if (p_post_barrier.has_flag(BARRIER_MASK_VERTEX)) { r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_FRAGMENT)) { r_stage_mask |= VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_COMPUTE)) { r_stage_mask |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; } r_access_mask |= VK_ACCESS_UNIFORM_READ_BIT; buffer = uniform_buffer_owner.get_or_null(p_buffer); } else if (texture_buffer_owner.owns(p_buffer)) { if (p_post_barrier.has_flag(BARRIER_MASK_VERTEX)) { r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT; r_access_mask |= VK_ACCESS_SHADER_READ_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_FRAGMENT)) { r_stage_mask |= VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; r_access_mask |= VK_ACCESS_SHADER_READ_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_COMPUTE)) { r_stage_mask |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; r_access_mask |= VK_ACCESS_SHADER_READ_BIT; } buffer = &texture_buffer_owner.get_or_null(p_buffer)->buffer; } else if (storage_buffer_owner.owns(p_buffer)) { buffer = storage_buffer_owner.get_or_null(p_buffer); if (p_post_barrier.has_flag(BARRIER_MASK_VERTEX)) { r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT; r_access_mask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_FRAGMENT)) { r_stage_mask |= VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; r_access_mask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_COMPUTE)) { r_stage_mask |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; r_access_mask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (buffer->usage & VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT) { r_stage_mask |= VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT; r_access_mask |= VK_ACCESS_INDIRECT_COMMAND_READ_BIT; } } return buffer; } static void update_external_dependency_for_store(VkSubpassDependency2KHR &dependency, bool is_sampled, bool is_storage, bool is_depth) { // Transitioning from write to read, protect the shaders that may use this next. // Allow for copies/image layout transitions. dependency.dstStageMask |= VK_PIPELINE_STAGE_TRANSFER_BIT; dependency.dstAccessMask |= VK_ACCESS_TRANSFER_READ_BIT; if (is_sampled) { dependency.dstStageMask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; dependency.dstAccessMask |= VK_ACCESS_SHADER_READ_BIT; } else if (is_storage) { dependency.dstStageMask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; dependency.dstAccessMask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } else { dependency.dstStageMask |= VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT; dependency.dstAccessMask |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT; } if (is_depth) { // Depth resources have additional stages that may be interested in them. dependency.dstStageMask |= VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT; dependency.dstAccessMask |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; } } void RenderingDeviceVulkan::_add_dependency(RID p_id, RID p_depends_on) { if (!dependency_map.has(p_depends_on)) { dependency_map[p_depends_on] = HashSet<RID>(); } dependency_map[p_depends_on].insert(p_id); if (!reverse_dependency_map.has(p_id)) { reverse_dependency_map[p_id] = HashSet<RID>(); } reverse_dependency_map[p_id].insert(p_depends_on); } void RenderingDeviceVulkan::_free_dependencies(RID p_id) { // Direct dependencies must be freed. HashMap<RID, HashSet<RID>>::Iterator E = dependency_map.find(p_id); if (E) { while (E->value.size()) { free(*E->value.begin()); } dependency_map.remove(E); } // Reverse dependencies must be unreferenced. E = reverse_dependency_map.find(p_id); if (E) { for (const RID &F : E->value) { HashMap<RID, HashSet<RID>>::Iterator G = dependency_map.find(F); ERR_CONTINUE(!G); ERR_CONTINUE(!G->value.has(p_id)); G->value.erase(p_id); } reverse_dependency_map.remove(E); } } const VkFormat RenderingDeviceVulkan::vulkan_formats[RenderingDevice::DATA_FORMAT_MAX] = { VK_FORMAT_R4G4_UNORM_PACK8, VK_FORMAT_R4G4B4A4_UNORM_PACK16, VK_FORMAT_B4G4R4A4_UNORM_PACK16, VK_FORMAT_R5G6B5_UNORM_PACK16, VK_FORMAT_B5G6R5_UNORM_PACK16, VK_FORMAT_R5G5B5A1_UNORM_PACK16, VK_FORMAT_B5G5R5A1_UNORM_PACK16, VK_FORMAT_A1R5G5B5_UNORM_PACK16, VK_FORMAT_R8_UNORM, VK_FORMAT_R8_SNORM, VK_FORMAT_R8_USCALED, VK_FORMAT_R8_SSCALED, VK_FORMAT_R8_UINT, VK_FORMAT_R8_SINT, VK_FORMAT_R8_SRGB, VK_FORMAT_R8G8_UNORM, VK_FORMAT_R8G8_SNORM, VK_FORMAT_R8G8_USCALED, VK_FORMAT_R8G8_SSCALED, VK_FORMAT_R8G8_UINT, VK_FORMAT_R8G8_SINT, VK_FORMAT_R8G8_SRGB, VK_FORMAT_R8G8B8_UNORM, VK_FORMAT_R8G8B8_SNORM, VK_FORMAT_R8G8B8_USCALED, VK_FORMAT_R8G8B8_SSCALED, VK_FORMAT_R8G8B8_UINT, VK_FORMAT_R8G8B8_SINT, VK_FORMAT_R8G8B8_SRGB, VK_FORMAT_B8G8R8_UNORM, VK_FORMAT_B8G8R8_SNORM, VK_FORMAT_B8G8R8_USCALED, VK_FORMAT_B8G8R8_SSCALED, VK_FORMAT_B8G8R8_UINT, VK_FORMAT_B8G8R8_SINT, VK_FORMAT_B8G8R8_SRGB, VK_FORMAT_R8G8B8A8_UNORM, VK_FORMAT_R8G8B8A8_SNORM, VK_FORMAT_R8G8B8A8_USCALED, VK_FORMAT_R8G8B8A8_SSCALED, VK_FORMAT_R8G8B8A8_UINT, VK_FORMAT_R8G8B8A8_SINT, VK_FORMAT_R8G8B8A8_SRGB, VK_FORMAT_B8G8R8A8_UNORM, VK_FORMAT_B8G8R8A8_SNORM, VK_FORMAT_B8G8R8A8_USCALED, VK_FORMAT_B8G8R8A8_SSCALED, VK_FORMAT_B8G8R8A8_UINT, VK_FORMAT_B8G8R8A8_SINT, VK_FORMAT_B8G8R8A8_SRGB, VK_FORMAT_A8B8G8R8_UNORM_PACK32, VK_FORMAT_A8B8G8R8_SNORM_PACK32, VK_FORMAT_A8B8G8R8_USCALED_PACK32, VK_FORMAT_A8B8G8R8_SSCALED_PACK32, VK_FORMAT_A8B8G8R8_UINT_PACK32, VK_FORMAT_A8B8G8R8_SINT_PACK32, VK_FORMAT_A8B8G8R8_SRGB_PACK32, VK_FORMAT_A2R10G10B10_UNORM_PACK32, VK_FORMAT_A2R10G10B10_SNORM_PACK32, VK_FORMAT_A2R10G10B10_USCALED_PACK32, VK_FORMAT_A2R10G10B10_SSCALED_PACK32, VK_FORMAT_A2R10G10B10_UINT_PACK32, VK_FORMAT_A2R10G10B10_SINT_PACK32, VK_FORMAT_A2B10G10R10_UNORM_PACK32, VK_FORMAT_A2B10G10R10_SNORM_PACK32, VK_FORMAT_A2B10G10R10_USCALED_PACK32, VK_FORMAT_A2B10G10R10_SSCALED_PACK32, VK_FORMAT_A2B10G10R10_UINT_PACK32, VK_FORMAT_A2B10G10R10_SINT_PACK32, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_SNORM, VK_FORMAT_R16_USCALED, VK_FORMAT_R16_SSCALED, VK_FORMAT_R16_UINT, VK_FORMAT_R16_SINT, VK_FORMAT_R16_SFLOAT, VK_FORMAT_R16G16_UNORM, VK_FORMAT_R16G16_SNORM, VK_FORMAT_R16G16_USCALED, VK_FORMAT_R16G16_SSCALED, VK_FORMAT_R16G16_UINT, VK_FORMAT_R16G16_SINT, VK_FORMAT_R16G16_SFLOAT, VK_FORMAT_R16G16B16_UNORM, VK_FORMAT_R16G16B16_SNORM, VK_FORMAT_R16G16B16_USCALED, VK_FORMAT_R16G16B16_SSCALED, VK_FORMAT_R16G16B16_UINT, VK_FORMAT_R16G16B16_SINT, VK_FORMAT_R16G16B16_SFLOAT, VK_FORMAT_R16G16B16A16_UNORM, VK_FORMAT_R16G16B16A16_SNORM, VK_FORMAT_R16G16B16A16_USCALED, VK_FORMAT_R16G16B16A16_SSCALED, VK_FORMAT_R16G16B16A16_UINT, VK_FORMAT_R16G16B16A16_SINT, VK_FORMAT_R16G16B16A16_SFLOAT, VK_FORMAT_R32_UINT, VK_FORMAT_R32_SINT, VK_FORMAT_R32_SFLOAT, VK_FORMAT_R32G32_UINT, VK_FORMAT_R32G32_SINT, VK_FORMAT_R32G32_SFLOAT, VK_FORMAT_R32G32B32_UINT, VK_FORMAT_R32G32B32_SINT, VK_FORMAT_R32G32B32_SFLOAT, VK_FORMAT_R32G32B32A32_UINT, VK_FORMAT_R32G32B32A32_SINT, VK_FORMAT_R32G32B32A32_SFLOAT, VK_FORMAT_R64_UINT, VK_FORMAT_R64_SINT, VK_FORMAT_R64_SFLOAT, VK_FORMAT_R64G64_UINT, VK_FORMAT_R64G64_SINT, VK_FORMAT_R64G64_SFLOAT, VK_FORMAT_R64G64B64_UINT, VK_FORMAT_R64G64B64_SINT, VK_FORMAT_R64G64B64_SFLOAT, VK_FORMAT_R64G64B64A64_UINT, VK_FORMAT_R64G64B64A64_SINT, VK_FORMAT_R64G64B64A64_SFLOAT, VK_FORMAT_B10G11R11_UFLOAT_PACK32, VK_FORMAT_E5B9G9R9_UFLOAT_PACK32, VK_FORMAT_D16_UNORM, VK_FORMAT_X8_D24_UNORM_PACK32, VK_FORMAT_D32_SFLOAT, VK_FORMAT_S8_UINT, VK_FORMAT_D16_UNORM_S8_UINT, VK_FORMAT_D24_UNORM_S8_UINT, VK_FORMAT_D32_SFLOAT_S8_UINT, VK_FORMAT_BC1_RGB_UNORM_BLOCK, VK_FORMAT_BC1_RGB_SRGB_BLOCK, VK_FORMAT_BC1_RGBA_UNORM_BLOCK, VK_FORMAT_BC1_RGBA_SRGB_BLOCK, VK_FORMAT_BC2_UNORM_BLOCK, VK_FORMAT_BC2_SRGB_BLOCK, VK_FORMAT_BC3_UNORM_BLOCK, VK_FORMAT_BC3_SRGB_BLOCK, VK_FORMAT_BC4_UNORM_BLOCK, VK_FORMAT_BC4_SNORM_BLOCK, VK_FORMAT_BC5_UNORM_BLOCK, VK_FORMAT_BC5_SNORM_BLOCK, VK_FORMAT_BC6H_UFLOAT_BLOCK, VK_FORMAT_BC6H_SFLOAT_BLOCK, VK_FORMAT_BC7_UNORM_BLOCK, VK_FORMAT_BC7_SRGB_BLOCK, VK_FORMAT_ETC2_R8G8B8_UNORM_BLOCK, VK_FORMAT_ETC2_R8G8B8_SRGB_BLOCK, VK_FORMAT_ETC2_R8G8B8A1_UNORM_BLOCK, VK_FORMAT_ETC2_R8G8B8A1_SRGB_BLOCK, VK_FORMAT_ETC2_R8G8B8A8_UNORM_BLOCK, VK_FORMAT_ETC2_R8G8B8A8_SRGB_BLOCK, VK_FORMAT_EAC_R11_UNORM_BLOCK, VK_FORMAT_EAC_R11_SNORM_BLOCK, VK_FORMAT_EAC_R11G11_UNORM_BLOCK, VK_FORMAT_EAC_R11G11_SNORM_BLOCK, VK_FORMAT_ASTC_4x4_UNORM_BLOCK, VK_FORMAT_ASTC_4x4_SRGB_BLOCK, VK_FORMAT_ASTC_5x4_UNORM_BLOCK, VK_FORMAT_ASTC_5x4_SRGB_BLOCK, VK_FORMAT_ASTC_5x5_UNORM_BLOCK, VK_FORMAT_ASTC_5x5_SRGB_BLOCK, VK_FORMAT_ASTC_6x5_UNORM_BLOCK, VK_FORMAT_ASTC_6x5_SRGB_BLOCK, VK_FORMAT_ASTC_6x6_UNORM_BLOCK, VK_FORMAT_ASTC_6x6_SRGB_BLOCK, VK_FORMAT_ASTC_8x5_UNORM_BLOCK, VK_FORMAT_ASTC_8x5_SRGB_BLOCK, VK_FORMAT_ASTC_8x6_UNORM_BLOCK, VK_FORMAT_ASTC_8x6_SRGB_BLOCK, VK_FORMAT_ASTC_8x8_UNORM_BLOCK, VK_FORMAT_ASTC_8x8_SRGB_BLOCK, VK_FORMAT_ASTC_10x5_UNORM_BLOCK, VK_FORMAT_ASTC_10x5_SRGB_BLOCK, VK_FORMAT_ASTC_10x6_UNORM_BLOCK, VK_FORMAT_ASTC_10x6_SRGB_BLOCK, VK_FORMAT_ASTC_10x8_UNORM_BLOCK, VK_FORMAT_ASTC_10x8_SRGB_BLOCK, VK_FORMAT_ASTC_10x10_UNORM_BLOCK, VK_FORMAT_ASTC_10x10_SRGB_BLOCK, VK_FORMAT_ASTC_12x10_UNORM_BLOCK, VK_FORMAT_ASTC_12x10_SRGB_BLOCK, VK_FORMAT_ASTC_12x12_UNORM_BLOCK, VK_FORMAT_ASTC_12x12_SRGB_BLOCK, VK_FORMAT_G8B8G8R8_422_UNORM, VK_FORMAT_B8G8R8G8_422_UNORM, VK_FORMAT_G8_B8_R8_3PLANE_420_UNORM, VK_FORMAT_G8_B8R8_2PLANE_420_UNORM, VK_FORMAT_G8_B8_R8_3PLANE_422_UNORM, VK_FORMAT_G8_B8R8_2PLANE_422_UNORM, VK_FORMAT_G8_B8_R8_3PLANE_444_UNORM, VK_FORMAT_R10X6_UNORM_PACK16, VK_FORMAT_R10X6G10X6_UNORM_2PACK16, VK_FORMAT_R10X6G10X6B10X6A10X6_UNORM_4PACK16, VK_FORMAT_G10X6B10X6G10X6R10X6_422_UNORM_4PACK16, VK_FORMAT_B10X6G10X6R10X6G10X6_422_UNORM_4PACK16, VK_FORMAT_G10X6_B10X6_R10X6_3PLANE_420_UNORM_3PACK16, VK_FORMAT_G10X6_B10X6R10X6_2PLANE_420_UNORM_3PACK16, VK_FORMAT_G10X6_B10X6_R10X6_3PLANE_422_UNORM_3PACK16, VK_FORMAT_G10X6_B10X6R10X6_2PLANE_422_UNORM_3PACK16, VK_FORMAT_G10X6_B10X6_R10X6_3PLANE_444_UNORM_3PACK16, VK_FORMAT_R12X4_UNORM_PACK16, VK_FORMAT_R12X4G12X4_UNORM_2PACK16, VK_FORMAT_R12X4G12X4B12X4A12X4_UNORM_4PACK16, VK_FORMAT_G12X4B12X4G12X4R12X4_422_UNORM_4PACK16, VK_FORMAT_B12X4G12X4R12X4G12X4_422_UNORM_4PACK16, VK_FORMAT_G12X4_B12X4_R12X4_3PLANE_420_UNORM_3PACK16, VK_FORMAT_G12X4_B12X4R12X4_2PLANE_420_UNORM_3PACK16, VK_FORMAT_G12X4_B12X4_R12X4_3PLANE_422_UNORM_3PACK16, VK_FORMAT_G12X4_B12X4R12X4_2PLANE_422_UNORM_3PACK16, VK_FORMAT_G12X4_B12X4_R12X4_3PLANE_444_UNORM_3PACK16, VK_FORMAT_G16B16G16R16_422_UNORM, VK_FORMAT_B16G16R16G16_422_UNORM, VK_FORMAT_G16_B16_R16_3PLANE_420_UNORM, VK_FORMAT_G16_B16R16_2PLANE_420_UNORM, VK_FORMAT_G16_B16_R16_3PLANE_422_UNORM, VK_FORMAT_G16_B16R16_2PLANE_422_UNORM, VK_FORMAT_G16_B16_R16_3PLANE_444_UNORM, }; const char *RenderingDeviceVulkan::named_formats[RenderingDevice::DATA_FORMAT_MAX] = { "R4G4_Unorm_Pack8", "R4G4B4A4_Unorm_Pack16", "B4G4R4A4_Unorm_Pack16", "R5G6B5_Unorm_Pack16", "B5G6R5_Unorm_Pack16", "R5G5B5A1_Unorm_Pack16", "B5G5R5A1_Unorm_Pack16", "A1R5G5B5_Unorm_Pack16", "R8_Unorm", "R8_Snorm", "R8_Uscaled", "R8_Sscaled", "R8_Uint", "R8_Sint", "R8_Srgb", "R8G8_Unorm", "R8G8_Snorm", "R8G8_Uscaled", "R8G8_Sscaled", "R8G8_Uint", "R8G8_Sint", "R8G8_Srgb", "R8G8B8_Unorm", "R8G8B8_Snorm", "R8G8B8_Uscaled", "R8G8B8_Sscaled", "R8G8B8_Uint", "R8G8B8_Sint", "R8G8B8_Srgb", "B8G8R8_Unorm", "B8G8R8_Snorm", "B8G8R8_Uscaled", "B8G8R8_Sscaled", "B8G8R8_Uint", "B8G8R8_Sint", "B8G8R8_Srgb", "R8G8B8A8_Unorm", "R8G8B8A8_Snorm", "R8G8B8A8_Uscaled", "R8G8B8A8_Sscaled", "R8G8B8A8_Uint", "R8G8B8A8_Sint", "R8G8B8A8_Srgb", "B8G8R8A8_Unorm", "B8G8R8A8_Snorm", "B8G8R8A8_Uscaled", "B8G8R8A8_Sscaled", "B8G8R8A8_Uint", "B8G8R8A8_Sint", "B8G8R8A8_Srgb", "A8B8G8R8_Unorm_Pack32", "A8B8G8R8_Snorm_Pack32", "A8B8G8R8_Uscaled_Pack32", "A8B8G8R8_Sscaled_Pack32", "A8B8G8R8_Uint_Pack32", "A8B8G8R8_Sint_Pack32", "A8B8G8R8_Srgb_Pack32", "A2R10G10B10_Unorm_Pack32", "A2R10G10B10_Snorm_Pack32", "A2R10G10B10_Uscaled_Pack32", "A2R10G10B10_Sscaled_Pack32", "A2R10G10B10_Uint_Pack32", "A2R10G10B10_Sint_Pack32", "A2B10G10R10_Unorm_Pack32", "A2B10G10R10_Snorm_Pack32", "A2B10G10R10_Uscaled_Pack32", "A2B10G10R10_Sscaled_Pack32", "A2B10G10R10_Uint_Pack32", "A2B10G10R10_Sint_Pack32", "R16_Unorm", "R16_Snorm", "R16_Uscaled", "R16_Sscaled", "R16_Uint", "R16_Sint", "R16_Sfloat", "R16G16_Unorm", "R16G16_Snorm", "R16G16_Uscaled", "R16G16_Sscaled", "R16G16_Uint", "R16G16_Sint", "R16G16_Sfloat", "R16G16B16_Unorm", "R16G16B16_Snorm", "R16G16B16_Uscaled", "R16G16B16_Sscaled", "R16G16B16_Uint", "R16G16B16_Sint", "R16G16B16_Sfloat", "R16G16B16A16_Unorm", "R16G16B16A16_Snorm", "R16G16B16A16_Uscaled", "R16G16B16A16_Sscaled", "R16G16B16A16_Uint", "R16G16B16A16_Sint", "R16G16B16A16_Sfloat", "R32_Uint", "R32_Sint", "R32_Sfloat", "R32G32_Uint", "R32G32_Sint", "R32G32_Sfloat", "R32G32B32_Uint", "R32G32B32_Sint", "R32G32B32_Sfloat", "R32G32B32A32_Uint", "R32G32B32A32_Sint", "R32G32B32A32_Sfloat", "R64_Uint", "R64_Sint", "R64_Sfloat", "R64G64_Uint", "R64G64_Sint", "R64G64_Sfloat", "R64G64B64_Uint", "R64G64B64_Sint", "R64G64B64_Sfloat", "R64G64B64A64_Uint", "R64G64B64A64_Sint", "R64G64B64A64_Sfloat", "B10G11R11_Ufloat_Pack32", "E5B9G9R9_Ufloat_Pack32", "D16_Unorm", "X8_D24_Unorm_Pack32", "D32_Sfloat", "S8_Uint", "D16_Unorm_S8_Uint", "D24_Unorm_S8_Uint", "D32_Sfloat_S8_Uint", "Bc1_Rgb_Unorm_Block", "Bc1_Rgb_Srgb_Block", "Bc1_Rgba_Unorm_Block", "Bc1_Rgba_Srgb_Block", "Bc2_Unorm_Block", "Bc2_Srgb_Block", "Bc3_Unorm_Block", "Bc3_Srgb_Block", "Bc4_Unorm_Block", "Bc4_Snorm_Block", "Bc5_Unorm_Block", "Bc5_Snorm_Block", "Bc6H_Ufloat_Block", "Bc6H_Sfloat_Block", "Bc7_Unorm_Block", "Bc7_Srgb_Block", "Etc2_R8G8B8_Unorm_Block", "Etc2_R8G8B8_Srgb_Block", "Etc2_R8G8B8A1_Unorm_Block", "Etc2_R8G8B8A1_Srgb_Block", "Etc2_R8G8B8A8_Unorm_Block", "Etc2_R8G8B8A8_Srgb_Block", "Eac_R11_Unorm_Block", "Eac_R11_Snorm_Block", "Eac_R11G11_Unorm_Block", "Eac_R11G11_Snorm_Block", "Astc_4X4_Unorm_Block", "Astc_4X4_Srgb_Block", "Astc_5X4_Unorm_Block", "Astc_5X4_Srgb_Block", "Astc_5X5_Unorm_Block", "Astc_5X5_Srgb_Block", "Astc_6X5_Unorm_Block", "Astc_6X5_Srgb_Block", "Astc_6X6_Unorm_Block", "Astc_6X6_Srgb_Block", "Astc_8X5_Unorm_Block", "Astc_8X5_Srgb_Block", "Astc_8X6_Unorm_Block", "Astc_8X6_Srgb_Block", "Astc_8X8_Unorm_Block", "Astc_8X8_Srgb_Block", "Astc_10X5_Unorm_Block", "Astc_10X5_Srgb_Block", "Astc_10X6_Unorm_Block", "Astc_10X6_Srgb_Block", "Astc_10X8_Unorm_Block", "Astc_10X8_Srgb_Block", "Astc_10X10_Unorm_Block", "Astc_10X10_Srgb_Block", "Astc_12X10_Unorm_Block", "Astc_12X10_Srgb_Block", "Astc_12X12_Unorm_Block", "Astc_12X12_Srgb_Block", "G8B8G8R8_422_Unorm", "B8G8R8G8_422_Unorm", "G8_B8_R8_3Plane_420_Unorm", "G8_B8R8_2Plane_420_Unorm", "G8_B8_R8_3Plane_422_Unorm", "G8_B8R8_2Plane_422_Unorm", "G8_B8_R8_3Plane_444_Unorm", "R10X6_Unorm_Pack16", "R10X6G10X6_Unorm_2Pack16", "R10X6G10X6B10X6A10X6_Unorm_4Pack16", "G10X6B10X6G10X6R10X6_422_Unorm_4Pack16", "B10X6G10X6R10X6G10X6_422_Unorm_4Pack16", "G10X6_B10X6_R10X6_3Plane_420_Unorm_3Pack16", "G10X6_B10X6R10X6_2Plane_420_Unorm_3Pack16", "G10X6_B10X6_R10X6_3Plane_422_Unorm_3Pack16", "G10X6_B10X6R10X6_2Plane_422_Unorm_3Pack16", "G10X6_B10X6_R10X6_3Plane_444_Unorm_3Pack16", "R12X4_Unorm_Pack16", "R12X4G12X4_Unorm_2Pack16", "R12X4G12X4B12X4A12X4_Unorm_4Pack16", "G12X4B12X4G12X4R12X4_422_Unorm_4Pack16", "B12X4G12X4R12X4G12X4_422_Unorm_4Pack16", "G12X4_B12X4_R12X4_3Plane_420_Unorm_3Pack16", "G12X4_B12X4R12X4_2Plane_420_Unorm_3Pack16", "G12X4_B12X4_R12X4_3Plane_422_Unorm_3Pack16", "G12X4_B12X4R12X4_2Plane_422_Unorm_3Pack16", "G12X4_B12X4_R12X4_3Plane_444_Unorm_3Pack16", "G16B16G16R16_422_Unorm", "B16G16R16G16_422_Unorm", "G16_B16_R16_3Plane_420_Unorm", "G16_B16R16_2Plane_420_Unorm", "G16_B16_R16_3Plane_422_Unorm", "G16_B16R16_2Plane_422_Unorm", "G16_B16_R16_3Plane_444_Unorm", }; int RenderingDeviceVulkan::get_format_vertex_size(DataFormat p_format) { switch (p_format) { case DATA_FORMAT_R8_UNORM: case DATA_FORMAT_R8_SNORM: case DATA_FORMAT_R8_UINT: case DATA_FORMAT_R8_SINT: case DATA_FORMAT_R8G8_UNORM: case DATA_FORMAT_R8G8_SNORM: case DATA_FORMAT_R8G8_UINT: case DATA_FORMAT_R8G8_SINT: case DATA_FORMAT_R8G8B8_UNORM: case DATA_FORMAT_R8G8B8_SNORM: case DATA_FORMAT_R8G8B8_UINT: case DATA_FORMAT_R8G8B8_SINT: case DATA_FORMAT_B8G8R8_UNORM: case DATA_FORMAT_B8G8R8_SNORM: case DATA_FORMAT_B8G8R8_UINT: case DATA_FORMAT_B8G8R8_SINT: case DATA_FORMAT_R8G8B8A8_UNORM: case DATA_FORMAT_R8G8B8A8_SNORM: case DATA_FORMAT_R8G8B8A8_UINT: case DATA_FORMAT_R8G8B8A8_SINT: case DATA_FORMAT_B8G8R8A8_UNORM: case DATA_FORMAT_B8G8R8A8_SNORM: case DATA_FORMAT_B8G8R8A8_UINT: case DATA_FORMAT_B8G8R8A8_SINT: case DATA_FORMAT_A2B10G10R10_UNORM_PACK32: return 4; case DATA_FORMAT_R16_UNORM: case DATA_FORMAT_R16_SNORM: case DATA_FORMAT_R16_UINT: case DATA_FORMAT_R16_SINT: case DATA_FORMAT_R16_SFLOAT: return 4; case DATA_FORMAT_R16G16_UNORM: case DATA_FORMAT_R16G16_SNORM: case DATA_FORMAT_R16G16_UINT: case DATA_FORMAT_R16G16_SINT: case DATA_FORMAT_R16G16_SFLOAT: return 4; case DATA_FORMAT_R16G16B16_UNORM: case DATA_FORMAT_R16G16B16_SNORM: case DATA_FORMAT_R16G16B16_UINT: case DATA_FORMAT_R16G16B16_SINT: case DATA_FORMAT_R16G16B16_SFLOAT: return 8; case DATA_FORMAT_R16G16B16A16_UNORM: case DATA_FORMAT_R16G16B16A16_SNORM: case DATA_FORMAT_R16G16B16A16_UINT: case DATA_FORMAT_R16G16B16A16_SINT: case DATA_FORMAT_R16G16B16A16_SFLOAT: return 8; case DATA_FORMAT_R32_UINT: case DATA_FORMAT_R32_SINT: case DATA_FORMAT_R32_SFLOAT: return 4; case DATA_FORMAT_R32G32_UINT: case DATA_FORMAT_R32G32_SINT: case DATA_FORMAT_R32G32_SFLOAT: return 8; case DATA_FORMAT_R32G32B32_UINT: case DATA_FORMAT_R32G32B32_SINT: case DATA_FORMAT_R32G32B32_SFLOAT: return 12; case DATA_FORMAT_R32G32B32A32_UINT: case DATA_FORMAT_R32G32B32A32_SINT: case DATA_FORMAT_R32G32B32A32_SFLOAT: return 16; case DATA_FORMAT_R64_UINT: case DATA_FORMAT_R64_SINT: case DATA_FORMAT_R64_SFLOAT: return 8; case DATA_FORMAT_R64G64_UINT: case DATA_FORMAT_R64G64_SINT: case DATA_FORMAT_R64G64_SFLOAT: return 16; case DATA_FORMAT_R64G64B64_UINT: case DATA_FORMAT_R64G64B64_SINT: case DATA_FORMAT_R64G64B64_SFLOAT: return 24; case DATA_FORMAT_R64G64B64A64_UINT: case DATA_FORMAT_R64G64B64A64_SINT: case DATA_FORMAT_R64G64B64A64_SFLOAT: return 32; default: return 0; } } uint32_t RenderingDeviceVulkan::get_image_format_pixel_size(DataFormat p_format) { switch (p_format) { case DATA_FORMAT_R4G4_UNORM_PACK8: return 1; case DATA_FORMAT_R4G4B4A4_UNORM_PACK16: case DATA_FORMAT_B4G4R4A4_UNORM_PACK16: case DATA_FORMAT_R5G6B5_UNORM_PACK16: case DATA_FORMAT_B5G6R5_UNORM_PACK16: case DATA_FORMAT_R5G5B5A1_UNORM_PACK16: case DATA_FORMAT_B5G5R5A1_UNORM_PACK16: case DATA_FORMAT_A1R5G5B5_UNORM_PACK16: return 2; case DATA_FORMAT_R8_UNORM: case DATA_FORMAT_R8_SNORM: case DATA_FORMAT_R8_USCALED: case DATA_FORMAT_R8_SSCALED: case DATA_FORMAT_R8_UINT: case DATA_FORMAT_R8_SINT: case DATA_FORMAT_R8_SRGB: return 1; case DATA_FORMAT_R8G8_UNORM: case DATA_FORMAT_R8G8_SNORM: case DATA_FORMAT_R8G8_USCALED: case DATA_FORMAT_R8G8_SSCALED: case DATA_FORMAT_R8G8_UINT: case DATA_FORMAT_R8G8_SINT: case DATA_FORMAT_R8G8_SRGB: return 2; case DATA_FORMAT_R8G8B8_UNORM: case DATA_FORMAT_R8G8B8_SNORM: case DATA_FORMAT_R8G8B8_USCALED: case DATA_FORMAT_R8G8B8_SSCALED: case DATA_FORMAT_R8G8B8_UINT: case DATA_FORMAT_R8G8B8_SINT: case DATA_FORMAT_R8G8B8_SRGB: case DATA_FORMAT_B8G8R8_UNORM: case DATA_FORMAT_B8G8R8_SNORM: case DATA_FORMAT_B8G8R8_USCALED: case DATA_FORMAT_B8G8R8_SSCALED: case DATA_FORMAT_B8G8R8_UINT: case DATA_FORMAT_B8G8R8_SINT: case DATA_FORMAT_B8G8R8_SRGB: return 3; case DATA_FORMAT_R8G8B8A8_UNORM: case DATA_FORMAT_R8G8B8A8_SNORM: case DATA_FORMAT_R8G8B8A8_USCALED: case DATA_FORMAT_R8G8B8A8_SSCALED: case DATA_FORMAT_R8G8B8A8_UINT: case DATA_FORMAT_R8G8B8A8_SINT: case DATA_FORMAT_R8G8B8A8_SRGB: case DATA_FORMAT_B8G8R8A8_UNORM: case DATA_FORMAT_B8G8R8A8_SNORM: case DATA_FORMAT_B8G8R8A8_USCALED: case DATA_FORMAT_B8G8R8A8_SSCALED: case DATA_FORMAT_B8G8R8A8_UINT: case DATA_FORMAT_B8G8R8A8_SINT: case DATA_FORMAT_B8G8R8A8_SRGB: return 4; case DATA_FORMAT_A8B8G8R8_UNORM_PACK32: case DATA_FORMAT_A8B8G8R8_SNORM_PACK32: case DATA_FORMAT_A8B8G8R8_USCALED_PACK32: case DATA_FORMAT_A8B8G8R8_SSCALED_PACK32: case DATA_FORMAT_A8B8G8R8_UINT_PACK32: case DATA_FORMAT_A8B8G8R8_SINT_PACK32: case DATA_FORMAT_A8B8G8R8_SRGB_PACK32: case DATA_FORMAT_A2R10G10B10_UNORM_PACK32: case DATA_FORMAT_A2R10G10B10_SNORM_PACK32: case DATA_FORMAT_A2R10G10B10_USCALED_PACK32: case DATA_FORMAT_A2R10G10B10_SSCALED_PACK32: case DATA_FORMAT_A2R10G10B10_UINT_PACK32: case DATA_FORMAT_A2R10G10B10_SINT_PACK32: case DATA_FORMAT_A2B10G10R10_UNORM_PACK32: case DATA_FORMAT_A2B10G10R10_SNORM_PACK32: case DATA_FORMAT_A2B10G10R10_USCALED_PACK32: case DATA_FORMAT_A2B10G10R10_SSCALED_PACK32: case DATA_FORMAT_A2B10G10R10_UINT_PACK32: case DATA_FORMAT_A2B10G10R10_SINT_PACK32: return 4; case DATA_FORMAT_R16_UNORM: case DATA_FORMAT_R16_SNORM: case DATA_FORMAT_R16_USCALED: case DATA_FORMAT_R16_SSCALED: case DATA_FORMAT_R16_UINT: case DATA_FORMAT_R16_SINT: case DATA_FORMAT_R16_SFLOAT: return 2; case DATA_FORMAT_R16G16_UNORM: case DATA_FORMAT_R16G16_SNORM: case DATA_FORMAT_R16G16_USCALED: case DATA_FORMAT_R16G16_SSCALED: case DATA_FORMAT_R16G16_UINT: case DATA_FORMAT_R16G16_SINT: case DATA_FORMAT_R16G16_SFLOAT: return 4; case DATA_FORMAT_R16G16B16_UNORM: case DATA_FORMAT_R16G16B16_SNORM: case DATA_FORMAT_R16G16B16_USCALED: case DATA_FORMAT_R16G16B16_SSCALED: case DATA_FORMAT_R16G16B16_UINT: case DATA_FORMAT_R16G16B16_SINT: case DATA_FORMAT_R16G16B16_SFLOAT: return 6; case DATA_FORMAT_R16G16B16A16_UNORM: case DATA_FORMAT_R16G16B16A16_SNORM: case DATA_FORMAT_R16G16B16A16_USCALED: case DATA_FORMAT_R16G16B16A16_SSCALED: case DATA_FORMAT_R16G16B16A16_UINT: case DATA_FORMAT_R16G16B16A16_SINT: case DATA_FORMAT_R16G16B16A16_SFLOAT: return 8; case DATA_FORMAT_R32_UINT: case DATA_FORMAT_R32_SINT: case DATA_FORMAT_R32_SFLOAT: return 4; case DATA_FORMAT_R32G32_UINT: case DATA_FORMAT_R32G32_SINT: case DATA_FORMAT_R32G32_SFLOAT: return 8; case DATA_FORMAT_R32G32B32_UINT: case DATA_FORMAT_R32G32B32_SINT: case DATA_FORMAT_R32G32B32_SFLOAT: return 12; case DATA_FORMAT_R32G32B32A32_UINT: case DATA_FORMAT_R32G32B32A32_SINT: case DATA_FORMAT_R32G32B32A32_SFLOAT: return 16; case DATA_FORMAT_R64_UINT: case DATA_FORMAT_R64_SINT: case DATA_FORMAT_R64_SFLOAT: return 8; case DATA_FORMAT_R64G64_UINT: case DATA_FORMAT_R64G64_SINT: case DATA_FORMAT_R64G64_SFLOAT: return 16; case DATA_FORMAT_R64G64B64_UINT: case DATA_FORMAT_R64G64B64_SINT: case DATA_FORMAT_R64G64B64_SFLOAT: return 24; case DATA_FORMAT_R64G64B64A64_UINT: case DATA_FORMAT_R64G64B64A64_SINT: case DATA_FORMAT_R64G64B64A64_SFLOAT: return 32; case DATA_FORMAT_B10G11R11_UFLOAT_PACK32: case DATA_FORMAT_E5B9G9R9_UFLOAT_PACK32: return 4; case DATA_FORMAT_D16_UNORM: return 2; case DATA_FORMAT_X8_D24_UNORM_PACK32: return 4; case DATA_FORMAT_D32_SFLOAT: return 4; case DATA_FORMAT_S8_UINT: return 1; case DATA_FORMAT_D16_UNORM_S8_UINT: return 4; case DATA_FORMAT_D24_UNORM_S8_UINT: return 4; case DATA_FORMAT_D32_SFLOAT_S8_UINT: return 5; // ? case DATA_FORMAT_BC1_RGB_UNORM_BLOCK: case DATA_FORMAT_BC1_RGB_SRGB_BLOCK: case DATA_FORMAT_BC1_RGBA_UNORM_BLOCK: case DATA_FORMAT_BC1_RGBA_SRGB_BLOCK: case DATA_FORMAT_BC2_UNORM_BLOCK: case DATA_FORMAT_BC2_SRGB_BLOCK: case DATA_FORMAT_BC3_UNORM_BLOCK: case DATA_FORMAT_BC3_SRGB_BLOCK: case DATA_FORMAT_BC4_UNORM_BLOCK: case DATA_FORMAT_BC4_SNORM_BLOCK: case DATA_FORMAT_BC5_UNORM_BLOCK: case DATA_FORMAT_BC5_SNORM_BLOCK: case DATA_FORMAT_BC6H_UFLOAT_BLOCK: case DATA_FORMAT_BC6H_SFLOAT_BLOCK: case DATA_FORMAT_BC7_UNORM_BLOCK: case DATA_FORMAT_BC7_SRGB_BLOCK: return 1; case DATA_FORMAT_ETC2_R8G8B8_UNORM_BLOCK: case DATA_FORMAT_ETC2_R8G8B8_SRGB_BLOCK: case DATA_FORMAT_ETC2_R8G8B8A1_UNORM_BLOCK: case DATA_FORMAT_ETC2_R8G8B8A1_SRGB_BLOCK: case DATA_FORMAT_ETC2_R8G8B8A8_UNORM_BLOCK: case DATA_FORMAT_ETC2_R8G8B8A8_SRGB_BLOCK: return 1; case DATA_FORMAT_EAC_R11_UNORM_BLOCK: case DATA_FORMAT_EAC_R11_SNORM_BLOCK: case DATA_FORMAT_EAC_R11G11_UNORM_BLOCK: case DATA_FORMAT_EAC_R11G11_SNORM_BLOCK: return 1; case DATA_FORMAT_ASTC_4x4_UNORM_BLOCK: case DATA_FORMAT_ASTC_4x4_SRGB_BLOCK: case DATA_FORMAT_ASTC_5x4_UNORM_BLOCK: case DATA_FORMAT_ASTC_5x4_SRGB_BLOCK: case DATA_FORMAT_ASTC_5x5_UNORM_BLOCK: case DATA_FORMAT_ASTC_5x5_SRGB_BLOCK: case DATA_FORMAT_ASTC_6x5_UNORM_BLOCK: case DATA_FORMAT_ASTC_6x5_SRGB_BLOCK: case DATA_FORMAT_ASTC_6x6_UNORM_BLOCK: case DATA_FORMAT_ASTC_6x6_SRGB_BLOCK: case DATA_FORMAT_ASTC_8x5_UNORM_BLOCK: case DATA_FORMAT_ASTC_8x5_SRGB_BLOCK: case DATA_FORMAT_ASTC_8x6_UNORM_BLOCK: case DATA_FORMAT_ASTC_8x6_SRGB_BLOCK: case DATA_FORMAT_ASTC_8x8_UNORM_BLOCK: case DATA_FORMAT_ASTC_8x8_SRGB_BLOCK: case DATA_FORMAT_ASTC_10x5_UNORM_BLOCK: case DATA_FORMAT_ASTC_10x5_SRGB_BLOCK: case DATA_FORMAT_ASTC_10x6_UNORM_BLOCK: case DATA_FORMAT_ASTC_10x6_SRGB_BLOCK: case DATA_FORMAT_ASTC_10x8_UNORM_BLOCK: case DATA_FORMAT_ASTC_10x8_SRGB_BLOCK: case DATA_FORMAT_ASTC_10x10_UNORM_BLOCK: case DATA_FORMAT_ASTC_10x10_SRGB_BLOCK: case DATA_FORMAT_ASTC_12x10_UNORM_BLOCK: case DATA_FORMAT_ASTC_12x10_SRGB_BLOCK: case DATA_FORMAT_ASTC_12x12_UNORM_BLOCK: case DATA_FORMAT_ASTC_12x12_SRGB_BLOCK: return 1; case DATA_FORMAT_G8B8G8R8_422_UNORM: case DATA_FORMAT_B8G8R8G8_422_UNORM: return 4; case DATA_FORMAT_G8_B8_R8_3PLANE_420_UNORM: case DATA_FORMAT_G8_B8R8_2PLANE_420_UNORM: case DATA_FORMAT_G8_B8_R8_3PLANE_422_UNORM: case DATA_FORMAT_G8_B8R8_2PLANE_422_UNORM: case DATA_FORMAT_G8_B8_R8_3PLANE_444_UNORM: return 4; case DATA_FORMAT_R10X6_UNORM_PACK16: case DATA_FORMAT_R10X6G10X6_UNORM_2PACK16: case DATA_FORMAT_R10X6G10X6B10X6A10X6_UNORM_4PACK16: case DATA_FORMAT_G10X6B10X6G10X6R10X6_422_UNORM_4PACK16: case DATA_FORMAT_B10X6G10X6R10X6G10X6_422_UNORM_4PACK16: case DATA_FORMAT_G10X6_B10X6_R10X6_3PLANE_420_UNORM_3PACK16: case DATA_FORMAT_G10X6_B10X6R10X6_2PLANE_420_UNORM_3PACK16: case DATA_FORMAT_G10X6_B10X6_R10X6_3PLANE_422_UNORM_3PACK16: case DATA_FORMAT_G10X6_B10X6R10X6_2PLANE_422_UNORM_3PACK16: case DATA_FORMAT_G10X6_B10X6_R10X6_3PLANE_444_UNORM_3PACK16: case DATA_FORMAT_R12X4_UNORM_PACK16: case DATA_FORMAT_R12X4G12X4_UNORM_2PACK16: case DATA_FORMAT_R12X4G12X4B12X4A12X4_UNORM_4PACK16: case DATA_FORMAT_G12X4B12X4G12X4R12X4_422_UNORM_4PACK16: case DATA_FORMAT_B12X4G12X4R12X4G12X4_422_UNORM_4PACK16: case DATA_FORMAT_G12X4_B12X4_R12X4_3PLANE_420_UNORM_3PACK16: case DATA_FORMAT_G12X4_B12X4R12X4_2PLANE_420_UNORM_3PACK16: case DATA_FORMAT_G12X4_B12X4_R12X4_3PLANE_422_UNORM_3PACK16: case DATA_FORMAT_G12X4_B12X4R12X4_2PLANE_422_UNORM_3PACK16: case DATA_FORMAT_G12X4_B12X4_R12X4_3PLANE_444_UNORM_3PACK16: return 2; case DATA_FORMAT_G16B16G16R16_422_UNORM: case DATA_FORMAT_B16G16R16G16_422_UNORM: case DATA_FORMAT_G16_B16_R16_3PLANE_420_UNORM: case DATA_FORMAT_G16_B16R16_2PLANE_420_UNORM: case DATA_FORMAT_G16_B16_R16_3PLANE_422_UNORM: case DATA_FORMAT_G16_B16R16_2PLANE_422_UNORM: case DATA_FORMAT_G16_B16_R16_3PLANE_444_UNORM: return 8; default: { ERR_PRINT("Format not handled, bug"); } } return 1; } // https://www.khronos.org/registry/DataFormat/specs/1.1/dataformat.1.1.pdf void RenderingDeviceVulkan::get_compressed_image_format_block_dimensions(DataFormat p_format, uint32_t &r_w, uint32_t &r_h) { switch (p_format) { case DATA_FORMAT_BC1_RGB_UNORM_BLOCK: case DATA_FORMAT_BC1_RGB_SRGB_BLOCK: case DATA_FORMAT_BC1_RGBA_UNORM_BLOCK: case DATA_FORMAT_BC1_RGBA_SRGB_BLOCK: case DATA_FORMAT_BC2_UNORM_BLOCK: case DATA_FORMAT_BC2_SRGB_BLOCK: case DATA_FORMAT_BC3_UNORM_BLOCK: case DATA_FORMAT_BC3_SRGB_BLOCK: case DATA_FORMAT_BC4_UNORM_BLOCK: case DATA_FORMAT_BC4_SNORM_BLOCK: case DATA_FORMAT_BC5_UNORM_BLOCK: case DATA_FORMAT_BC5_SNORM_BLOCK: case DATA_FORMAT_BC6H_UFLOAT_BLOCK: case DATA_FORMAT_BC6H_SFLOAT_BLOCK: case DATA_FORMAT_BC7_UNORM_BLOCK: case DATA_FORMAT_BC7_SRGB_BLOCK: case DATA_FORMAT_ETC2_R8G8B8_UNORM_BLOCK: case DATA_FORMAT_ETC2_R8G8B8_SRGB_BLOCK: case DATA_FORMAT_ETC2_R8G8B8A1_UNORM_BLOCK: case DATA_FORMAT_ETC2_R8G8B8A1_SRGB_BLOCK: case DATA_FORMAT_ETC2_R8G8B8A8_UNORM_BLOCK: case DATA_FORMAT_ETC2_R8G8B8A8_SRGB_BLOCK: case DATA_FORMAT_EAC_R11_UNORM_BLOCK: case DATA_FORMAT_EAC_R11_SNORM_BLOCK: case DATA_FORMAT_EAC_R11G11_UNORM_BLOCK: case DATA_FORMAT_EAC_R11G11_SNORM_BLOCK: case DATA_FORMAT_ASTC_4x4_UNORM_BLOCK: // Again, not sure about astc. case DATA_FORMAT_ASTC_4x4_SRGB_BLOCK: { r_w = 4; r_h = 4; } break; case DATA_FORMAT_ASTC_5x4_UNORM_BLOCK: // Unsupported case DATA_FORMAT_ASTC_5x4_SRGB_BLOCK: case DATA_FORMAT_ASTC_5x5_UNORM_BLOCK: case DATA_FORMAT_ASTC_5x5_SRGB_BLOCK: case DATA_FORMAT_ASTC_6x5_UNORM_BLOCK: case DATA_FORMAT_ASTC_6x5_SRGB_BLOCK: case DATA_FORMAT_ASTC_6x6_UNORM_BLOCK: case DATA_FORMAT_ASTC_6x6_SRGB_BLOCK: case DATA_FORMAT_ASTC_8x5_UNORM_BLOCK: case DATA_FORMAT_ASTC_8x5_SRGB_BLOCK: case DATA_FORMAT_ASTC_8x6_UNORM_BLOCK: case DATA_FORMAT_ASTC_8x6_SRGB_BLOCK: { r_w = 4; r_h = 4; } break; case DATA_FORMAT_ASTC_8x8_UNORM_BLOCK: case DATA_FORMAT_ASTC_8x8_SRGB_BLOCK: { r_w = 8; r_h = 8; } break; case DATA_FORMAT_ASTC_10x5_UNORM_BLOCK: // Unsupported case DATA_FORMAT_ASTC_10x5_SRGB_BLOCK: case DATA_FORMAT_ASTC_10x6_UNORM_BLOCK: case DATA_FORMAT_ASTC_10x6_SRGB_BLOCK: case DATA_FORMAT_ASTC_10x8_UNORM_BLOCK: case DATA_FORMAT_ASTC_10x8_SRGB_BLOCK: case DATA_FORMAT_ASTC_10x10_UNORM_BLOCK: case DATA_FORMAT_ASTC_10x10_SRGB_BLOCK: case DATA_FORMAT_ASTC_12x10_UNORM_BLOCK: case DATA_FORMAT_ASTC_12x10_SRGB_BLOCK: case DATA_FORMAT_ASTC_12x12_UNORM_BLOCK: case DATA_FORMAT_ASTC_12x12_SRGB_BLOCK: r_w = 4; r_h = 4; return; default: { r_w = 1; r_h = 1; } } } uint32_t RenderingDeviceVulkan::get_compressed_image_format_block_byte_size(DataFormat p_format) { switch (p_format) { case DATA_FORMAT_BC1_RGB_UNORM_BLOCK: case DATA_FORMAT_BC1_RGB_SRGB_BLOCK: case DATA_FORMAT_BC1_RGBA_UNORM_BLOCK: case DATA_FORMAT_BC1_RGBA_SRGB_BLOCK: return 8; case DATA_FORMAT_BC2_UNORM_BLOCK: case DATA_FORMAT_BC2_SRGB_BLOCK: return 16; case DATA_FORMAT_BC3_UNORM_BLOCK: case DATA_FORMAT_BC3_SRGB_BLOCK: return 16; case DATA_FORMAT_BC4_UNORM_BLOCK: case DATA_FORMAT_BC4_SNORM_BLOCK: return 8; case DATA_FORMAT_BC5_UNORM_BLOCK: case DATA_FORMAT_BC5_SNORM_BLOCK: return 16; case DATA_FORMAT_BC6H_UFLOAT_BLOCK: case DATA_FORMAT_BC6H_SFLOAT_BLOCK: return 16; case DATA_FORMAT_BC7_UNORM_BLOCK: case DATA_FORMAT_BC7_SRGB_BLOCK: return 16; case DATA_FORMAT_ETC2_R8G8B8_UNORM_BLOCK: case DATA_FORMAT_ETC2_R8G8B8_SRGB_BLOCK: return 8; case DATA_FORMAT_ETC2_R8G8B8A1_UNORM_BLOCK: case DATA_FORMAT_ETC2_R8G8B8A1_SRGB_BLOCK: return 8; case DATA_FORMAT_ETC2_R8G8B8A8_UNORM_BLOCK: case DATA_FORMAT_ETC2_R8G8B8A8_SRGB_BLOCK: return 16; case DATA_FORMAT_EAC_R11_UNORM_BLOCK: case DATA_FORMAT_EAC_R11_SNORM_BLOCK: return 8; case DATA_FORMAT_EAC_R11G11_UNORM_BLOCK: case DATA_FORMAT_EAC_R11G11_SNORM_BLOCK: return 16; case DATA_FORMAT_ASTC_4x4_UNORM_BLOCK: // Again, not sure about astc. case DATA_FORMAT_ASTC_4x4_SRGB_BLOCK: case DATA_FORMAT_ASTC_5x4_UNORM_BLOCK: case DATA_FORMAT_ASTC_5x4_SRGB_BLOCK: case DATA_FORMAT_ASTC_5x5_UNORM_BLOCK: case DATA_FORMAT_ASTC_5x5_SRGB_BLOCK: case DATA_FORMAT_ASTC_6x5_UNORM_BLOCK: case DATA_FORMAT_ASTC_6x5_SRGB_BLOCK: case DATA_FORMAT_ASTC_6x6_UNORM_BLOCK: case DATA_FORMAT_ASTC_6x6_SRGB_BLOCK: case DATA_FORMAT_ASTC_8x5_UNORM_BLOCK: case DATA_FORMAT_ASTC_8x5_SRGB_BLOCK: case DATA_FORMAT_ASTC_8x6_UNORM_BLOCK: case DATA_FORMAT_ASTC_8x6_SRGB_BLOCK: case DATA_FORMAT_ASTC_8x8_UNORM_BLOCK: case DATA_FORMAT_ASTC_8x8_SRGB_BLOCK: case DATA_FORMAT_ASTC_10x5_UNORM_BLOCK: case DATA_FORMAT_ASTC_10x5_SRGB_BLOCK: case DATA_FORMAT_ASTC_10x6_UNORM_BLOCK: case DATA_FORMAT_ASTC_10x6_SRGB_BLOCK: case DATA_FORMAT_ASTC_10x8_UNORM_BLOCK: case DATA_FORMAT_ASTC_10x8_SRGB_BLOCK: case DATA_FORMAT_ASTC_10x10_UNORM_BLOCK: case DATA_FORMAT_ASTC_10x10_SRGB_BLOCK: case DATA_FORMAT_ASTC_12x10_UNORM_BLOCK: case DATA_FORMAT_ASTC_12x10_SRGB_BLOCK: case DATA_FORMAT_ASTC_12x12_UNORM_BLOCK: case DATA_FORMAT_ASTC_12x12_SRGB_BLOCK: return 16; default: { } } return 1; } uint32_t RenderingDeviceVulkan::get_compressed_image_format_pixel_rshift(DataFormat p_format) { switch (p_format) { case DATA_FORMAT_BC1_RGB_UNORM_BLOCK: // These formats are half byte size, so rshift is 1. case DATA_FORMAT_BC1_RGB_SRGB_BLOCK: case DATA_FORMAT_BC1_RGBA_UNORM_BLOCK: case DATA_FORMAT_BC1_RGBA_SRGB_BLOCK: case DATA_FORMAT_BC4_UNORM_BLOCK: case DATA_FORMAT_BC4_SNORM_BLOCK: case DATA_FORMAT_ETC2_R8G8B8_UNORM_BLOCK: case DATA_FORMAT_ETC2_R8G8B8_SRGB_BLOCK: case DATA_FORMAT_ETC2_R8G8B8A1_UNORM_BLOCK: case DATA_FORMAT_ETC2_R8G8B8A1_SRGB_BLOCK: case DATA_FORMAT_EAC_R11_UNORM_BLOCK: case DATA_FORMAT_EAC_R11_SNORM_BLOCK: return 1; case DATA_FORMAT_ASTC_8x8_SRGB_BLOCK: case DATA_FORMAT_ASTC_8x8_UNORM_BLOCK: { return 2; } default: { } } return 0; } bool RenderingDeviceVulkan::format_has_stencil(DataFormat p_format) { switch (p_format) { case DATA_FORMAT_S8_UINT: case DATA_FORMAT_D16_UNORM_S8_UINT: case DATA_FORMAT_D24_UNORM_S8_UINT: case DATA_FORMAT_D32_SFLOAT_S8_UINT: { return true; } default: { } } return false; } uint32_t RenderingDeviceVulkan::get_image_format_required_size(DataFormat p_format, uint32_t p_width, uint32_t p_height, uint32_t p_depth, uint32_t p_mipmaps, uint32_t *r_blockw, uint32_t *r_blockh, uint32_t *r_depth) { ERR_FAIL_COND_V(p_mipmaps == 0, 0); uint32_t w = p_width; uint32_t h = p_height; uint32_t d = p_depth; uint32_t size = 0; uint32_t pixel_size = get_image_format_pixel_size(p_format); uint32_t pixel_rshift = get_compressed_image_format_pixel_rshift(p_format); uint32_t blockw, blockh; get_compressed_image_format_block_dimensions(p_format, blockw, blockh); for (uint32_t i = 0; i < p_mipmaps; i++) { uint32_t bw = w % blockw != 0 ? w + (blockw - w % blockw) : w; uint32_t bh = h % blockh != 0 ? h + (blockh - h % blockh) : h; uint32_t s = bw * bh; s *= pixel_size; s >>= pixel_rshift; size += s * d; if (r_blockw) { *r_blockw = bw; } if (r_blockh) { *r_blockh = bh; } if (r_depth) { *r_depth = d; } w = MAX(blockw, w >> 1); h = MAX(blockh, h >> 1); d = MAX(1u, d >> 1); } return size; } uint32_t RenderingDeviceVulkan::get_image_required_mipmaps(uint32_t p_width, uint32_t p_height, uint32_t p_depth) { // Formats and block size don't really matter here since they can all go down to 1px (even if block is larger). uint32_t w = p_width; uint32_t h = p_height; uint32_t d = p_depth; uint32_t mipmaps = 1; while (true) { if (w == 1 && h == 1 && d == 1) { break; } w = MAX(1u, w >> 1); h = MAX(1u, h >> 1); d = MAX(1u, d >> 1); mipmaps++; } return mipmaps; } /////////////////////// const VkCompareOp RenderingDeviceVulkan::compare_operators[RenderingDevice::COMPARE_OP_MAX] = { VK_COMPARE_OP_NEVER, VK_COMPARE_OP_LESS, VK_COMPARE_OP_EQUAL, VK_COMPARE_OP_LESS_OR_EQUAL, VK_COMPARE_OP_GREATER, VK_COMPARE_OP_NOT_EQUAL, VK_COMPARE_OP_GREATER_OR_EQUAL, VK_COMPARE_OP_ALWAYS }; const VkStencilOp RenderingDeviceVulkan::stencil_operations[RenderingDevice::STENCIL_OP_MAX] = { VK_STENCIL_OP_KEEP, VK_STENCIL_OP_ZERO, VK_STENCIL_OP_REPLACE, VK_STENCIL_OP_INCREMENT_AND_CLAMP, VK_STENCIL_OP_DECREMENT_AND_CLAMP, VK_STENCIL_OP_INVERT, VK_STENCIL_OP_INCREMENT_AND_WRAP, VK_STENCIL_OP_DECREMENT_AND_WRAP }; const VkSampleCountFlagBits RenderingDeviceVulkan::rasterization_sample_count[RenderingDevice::TEXTURE_SAMPLES_MAX] = { VK_SAMPLE_COUNT_1_BIT, VK_SAMPLE_COUNT_2_BIT, VK_SAMPLE_COUNT_4_BIT, VK_SAMPLE_COUNT_8_BIT, VK_SAMPLE_COUNT_16_BIT, VK_SAMPLE_COUNT_32_BIT, VK_SAMPLE_COUNT_64_BIT, }; const VkLogicOp RenderingDeviceVulkan::logic_operations[RenderingDevice::LOGIC_OP_MAX] = { VK_LOGIC_OP_CLEAR, VK_LOGIC_OP_AND, VK_LOGIC_OP_AND_REVERSE, VK_LOGIC_OP_COPY, VK_LOGIC_OP_AND_INVERTED, VK_LOGIC_OP_NO_OP, VK_LOGIC_OP_XOR, VK_LOGIC_OP_OR, VK_LOGIC_OP_NOR, VK_LOGIC_OP_EQUIVALENT, VK_LOGIC_OP_INVERT, VK_LOGIC_OP_OR_REVERSE, VK_LOGIC_OP_COPY_INVERTED, VK_LOGIC_OP_OR_INVERTED, VK_LOGIC_OP_NAND, VK_LOGIC_OP_SET }; const VkBlendFactor RenderingDeviceVulkan::blend_factors[RenderingDevice::BLEND_FACTOR_MAX] = { VK_BLEND_FACTOR_ZERO, VK_BLEND_FACTOR_ONE, VK_BLEND_FACTOR_SRC_COLOR, VK_BLEND_FACTOR_ONE_MINUS_SRC_COLOR, VK_BLEND_FACTOR_DST_COLOR, VK_BLEND_FACTOR_ONE_MINUS_DST_COLOR, VK_BLEND_FACTOR_SRC_ALPHA, VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA, VK_BLEND_FACTOR_DST_ALPHA, VK_BLEND_FACTOR_ONE_MINUS_DST_ALPHA, VK_BLEND_FACTOR_CONSTANT_COLOR, VK_BLEND_FACTOR_ONE_MINUS_CONSTANT_COLOR, VK_BLEND_FACTOR_CONSTANT_ALPHA, VK_BLEND_FACTOR_ONE_MINUS_CONSTANT_ALPHA, VK_BLEND_FACTOR_SRC_ALPHA_SATURATE, VK_BLEND_FACTOR_SRC1_COLOR, VK_BLEND_FACTOR_ONE_MINUS_SRC1_COLOR, VK_BLEND_FACTOR_SRC1_ALPHA, VK_BLEND_FACTOR_ONE_MINUS_SRC1_ALPHA }; const VkBlendOp RenderingDeviceVulkan::blend_operations[RenderingDevice::BLEND_OP_MAX] = { VK_BLEND_OP_ADD, VK_BLEND_OP_SUBTRACT, VK_BLEND_OP_REVERSE_SUBTRACT, VK_BLEND_OP_MIN, VK_BLEND_OP_MAX }; const VkSamplerAddressMode RenderingDeviceVulkan::address_modes[RenderingDevice::SAMPLER_REPEAT_MODE_MAX] = { VK_SAMPLER_ADDRESS_MODE_REPEAT, VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT, VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE, VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER, VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE }; const VkBorderColor RenderingDeviceVulkan::sampler_border_colors[RenderingDevice::SAMPLER_BORDER_COLOR_MAX] = { VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK, VK_BORDER_COLOR_INT_TRANSPARENT_BLACK, VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK, VK_BORDER_COLOR_INT_OPAQUE_BLACK, VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE, VK_BORDER_COLOR_INT_OPAQUE_WHITE }; const VkImageType RenderingDeviceVulkan::vulkan_image_type[RenderingDevice::TEXTURE_TYPE_MAX] = { VK_IMAGE_TYPE_1D, VK_IMAGE_TYPE_2D, VK_IMAGE_TYPE_3D, VK_IMAGE_TYPE_2D, VK_IMAGE_TYPE_1D, VK_IMAGE_TYPE_2D, VK_IMAGE_TYPE_2D }; /***************************/ /**** BUFFER MANAGEMENT ****/ /***************************/ Error RenderingDeviceVulkan::_buffer_allocate(Buffer *p_buffer, uint32_t p_size, uint32_t p_usage, VmaMemoryUsage p_mem_usage, VmaAllocationCreateFlags p_mem_flags) { VkBufferCreateInfo bufferInfo; bufferInfo.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO; bufferInfo.pNext = nullptr; bufferInfo.flags = 0; bufferInfo.size = p_size; bufferInfo.usage = p_usage; bufferInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE; bufferInfo.queueFamilyIndexCount = 0; bufferInfo.pQueueFamilyIndices = nullptr; VmaAllocationCreateInfo allocInfo; allocInfo.flags = p_mem_flags; allocInfo.usage = p_mem_usage; allocInfo.requiredFlags = 0; allocInfo.preferredFlags = 0; allocInfo.memoryTypeBits = 0; allocInfo.pool = nullptr; allocInfo.pUserData = nullptr; if (p_size <= SMALL_ALLOCATION_MAX_SIZE) { uint32_t mem_type_index = 0; vmaFindMemoryTypeIndexForBufferInfo(allocator, &bufferInfo, &allocInfo, &mem_type_index); allocInfo.pool = _find_or_create_small_allocs_pool(mem_type_index); } VkResult err = vmaCreateBuffer(allocator, &bufferInfo, &allocInfo, &p_buffer->buffer, &p_buffer->allocation, nullptr); ERR_FAIL_COND_V_MSG(err, ERR_CANT_CREATE, "Can't create buffer of size: " + itos(p_size) + ", error " + itos(err) + "."); p_buffer->size = p_size; p_buffer->buffer_info.buffer = p_buffer->buffer; p_buffer->buffer_info.offset = 0; p_buffer->buffer_info.range = p_size; p_buffer->usage = p_usage; buffer_memory += p_size; return OK; } Error RenderingDeviceVulkan::_buffer_free(Buffer *p_buffer) { ERR_FAIL_COND_V(p_buffer->size == 0, ERR_INVALID_PARAMETER); buffer_memory -= p_buffer->size; vmaDestroyBuffer(allocator, p_buffer->buffer, p_buffer->allocation); p_buffer->buffer = VK_NULL_HANDLE; p_buffer->allocation = nullptr; p_buffer->size = 0; return OK; } Error RenderingDeviceVulkan::_insert_staging_block() { VkBufferCreateInfo bufferInfo; bufferInfo.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO; bufferInfo.pNext = nullptr; bufferInfo.flags = 0; bufferInfo.size = staging_buffer_block_size; bufferInfo.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT; bufferInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE; bufferInfo.queueFamilyIndexCount = 0; bufferInfo.pQueueFamilyIndices = nullptr; VmaAllocationCreateInfo allocInfo; allocInfo.flags = VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT; allocInfo.usage = VMA_MEMORY_USAGE_AUTO_PREFER_HOST; allocInfo.requiredFlags = 0; allocInfo.preferredFlags = 0; allocInfo.memoryTypeBits = 0; allocInfo.pool = nullptr; allocInfo.pUserData = nullptr; StagingBufferBlock block; VkResult err = vmaCreateBuffer(allocator, &bufferInfo, &allocInfo, &block.buffer, &block.allocation, nullptr); ERR_FAIL_COND_V_MSG(err, ERR_CANT_CREATE, "vmaCreateBuffer failed with error " + itos(err) + "."); block.frame_used = 0; block.fill_amount = 0; staging_buffer_blocks.insert(staging_buffer_current, block); return OK; } Error RenderingDeviceVulkan::_staging_buffer_allocate(uint32_t p_amount, uint32_t p_required_align, uint32_t &r_alloc_offset, uint32_t &r_alloc_size, bool p_can_segment) { // Determine a block to use. r_alloc_size = p_amount; while (true) { r_alloc_offset = 0; // See if we can use current block. if (staging_buffer_blocks[staging_buffer_current].frame_used == frames_drawn) { // We used this block this frame, let's see if there is still room. uint32_t write_from = staging_buffer_blocks[staging_buffer_current].fill_amount; { uint32_t align_remainder = write_from % p_required_align; if (align_remainder != 0) { write_from += p_required_align - align_remainder; } } int32_t available_bytes = int32_t(staging_buffer_block_size) - int32_t(write_from); if ((int32_t)p_amount < available_bytes) { // All is good, we should be ok, all will fit. r_alloc_offset = write_from; } else if (p_can_segment && available_bytes >= (int32_t)p_required_align) { // Ok all won't fit but at least we can fit a chunkie. // All is good, update what needs to be written to. r_alloc_offset = write_from; r_alloc_size = available_bytes - (available_bytes % p_required_align); } else { // Can't fit it into this buffer. // Will need to try next buffer. staging_buffer_current = (staging_buffer_current + 1) % staging_buffer_blocks.size(); // Before doing anything, though, let's check that we didn't manage to fill all blocks. // Possible in a single frame. if (staging_buffer_blocks[staging_buffer_current].frame_used == frames_drawn) { // Guess we did.. ok, let's see if we can insert a new block. if ((uint64_t)staging_buffer_blocks.size() * staging_buffer_block_size < staging_buffer_max_size) { // We can, so we are safe. Error err = _insert_staging_block(); if (err) { return err; } // Claim for this frame. staging_buffer_blocks.write[staging_buffer_current].frame_used = frames_drawn; } else { // Ok, worst case scenario, all the staging buffers belong to this frame // and this frame is not even done. // If this is the main thread, it means the user is likely loading a lot of resources at once,. // Otherwise, the thread should just be blocked until the next frame (currently unimplemented). if (false) { // Separate thread from render. //block_until_next_frame() continue; } else { // Flush EVERYTHING including setup commands. IF not immediate, also need to flush the draw commands. _flush(true); // Clear the whole staging buffer. for (int i = 0; i < staging_buffer_blocks.size(); i++) { staging_buffer_blocks.write[i].frame_used = 0; staging_buffer_blocks.write[i].fill_amount = 0; } // Claim current. staging_buffer_blocks.write[staging_buffer_current].frame_used = frames_drawn; } } } else { // Not from current frame, so continue and try again. continue; } } } else if (staging_buffer_blocks[staging_buffer_current].frame_used <= frames_drawn - frame_count) { // This is an old block, which was already processed, let's reuse. staging_buffer_blocks.write[staging_buffer_current].frame_used = frames_drawn; staging_buffer_blocks.write[staging_buffer_current].fill_amount = 0; } else { // This block may still be in use, let's not touch it unless we have to, so.. can we create a new one? if ((uint64_t)staging_buffer_blocks.size() * staging_buffer_block_size < staging_buffer_max_size) { // We are still allowed to create a new block, so let's do that and insert it for current pos. Error err = _insert_staging_block(); if (err) { return err; } // Claim for this frame. staging_buffer_blocks.write[staging_buffer_current].frame_used = frames_drawn; } else { // Oops, we are out of room and we can't create more. // Let's flush older frames. // The logic here is that if a game is loading a lot of data from the main thread, it will need to be stalled anyway. // If loading from a separate thread, we can block that thread until next frame when more room is made (not currently implemented, though). if (false) { // Separate thread from render. //block_until_next_frame() continue; // And try again. } else { _flush(false); for (int i = 0; i < staging_buffer_blocks.size(); i++) { // Clear all blocks but the ones from this frame. int block_idx = (i + staging_buffer_current) % staging_buffer_blocks.size(); if (staging_buffer_blocks[block_idx].frame_used == frames_drawn) { break; // Ok, we reached something from this frame, abort. } staging_buffer_blocks.write[block_idx].frame_used = 0; staging_buffer_blocks.write[block_idx].fill_amount = 0; } // Claim for current frame. staging_buffer_blocks.write[staging_buffer_current].frame_used = frames_drawn; } } } // All was good, break. break; } staging_buffer_used = true; return OK; } Error RenderingDeviceVulkan::_buffer_update(Buffer *p_buffer, size_t p_offset, const uint8_t *p_data, size_t p_data_size, bool p_use_draw_command_buffer, uint32_t p_required_align) { // Submitting may get chunked for various reasons, so convert this to a task. size_t to_submit = p_data_size; size_t submit_from = 0; while (to_submit > 0) { uint32_t block_write_offset; uint32_t block_write_amount; Error err = _staging_buffer_allocate(MIN(to_submit, staging_buffer_block_size), p_required_align, block_write_offset, block_write_amount); if (err) { return err; } // Map staging buffer (It's CPU and coherent). void *data_ptr = nullptr; { VkResult vkerr = vmaMapMemory(allocator, staging_buffer_blocks[staging_buffer_current].allocation, &data_ptr); ERR_FAIL_COND_V_MSG(vkerr, ERR_CANT_CREATE, "vmaMapMemory failed with error " + itos(vkerr) + "."); } // Copy to staging buffer. memcpy(((uint8_t *)data_ptr) + block_write_offset, p_data + submit_from, block_write_amount); // Unmap. vmaUnmapMemory(allocator, staging_buffer_blocks[staging_buffer_current].allocation); // Insert a command to copy this. VkBufferCopy region; region.srcOffset = block_write_offset; region.dstOffset = submit_from + p_offset; region.size = block_write_amount; vkCmdCopyBuffer(p_use_draw_command_buffer ? frames[frame].draw_command_buffer : frames[frame].setup_command_buffer, staging_buffer_blocks[staging_buffer_current].buffer, p_buffer->buffer, 1, ®ion); staging_buffer_blocks.write[staging_buffer_current].fill_amount = block_write_offset + block_write_amount; to_submit -= block_write_amount; submit_from += block_write_amount; } return OK; } void RenderingDeviceVulkan::_memory_barrier(VkPipelineStageFlags p_src_stage_mask, VkPipelineStageFlags p_dst_stage_mask, VkAccessFlags p_src_access, VkAccessFlags p_dst_access, bool p_sync_with_draw) { VkMemoryBarrier mem_barrier; mem_barrier.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER; mem_barrier.pNext = nullptr; mem_barrier.srcAccessMask = p_src_access; mem_barrier.dstAccessMask = p_dst_access; if (p_src_stage_mask == 0 || p_dst_stage_mask == 0) { return; // No barrier, since this is invalid. } vkCmdPipelineBarrier(p_sync_with_draw ? frames[frame].draw_command_buffer : frames[frame].setup_command_buffer, p_src_stage_mask, p_dst_stage_mask, 0, 1, &mem_barrier, 0, nullptr, 0, nullptr); } void RenderingDeviceVulkan::_full_barrier(bool p_sync_with_draw) { // Used for debug. _memory_barrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_ACCESS_INDIRECT_COMMAND_READ_BIT | VK_ACCESS_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_UNIFORM_READ_BIT | VK_ACCESS_INPUT_ATTACHMENT_READ_BIT | VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_HOST_READ_BIT | VK_ACCESS_HOST_WRITE_BIT, VK_ACCESS_INDIRECT_COMMAND_READ_BIT | VK_ACCESS_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_UNIFORM_READ_BIT | VK_ACCESS_INPUT_ATTACHMENT_READ_BIT | VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_HOST_READ_BIT | VK_ACCESS_HOST_WRITE_BIT, p_sync_with_draw); } void RenderingDeviceVulkan::_buffer_memory_barrier(VkBuffer buffer, uint64_t p_from, uint64_t p_size, VkPipelineStageFlags p_src_stage_mask, VkPipelineStageFlags p_dst_stage_mask, VkAccessFlags p_src_access, VkAccessFlags p_dst_access, bool p_sync_with_draw) { VkBufferMemoryBarrier buffer_mem_barrier; buffer_mem_barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER; buffer_mem_barrier.pNext = nullptr; buffer_mem_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; buffer_mem_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; buffer_mem_barrier.srcAccessMask = p_src_access; buffer_mem_barrier.dstAccessMask = p_dst_access; buffer_mem_barrier.buffer = buffer; buffer_mem_barrier.offset = p_from; buffer_mem_barrier.size = p_size; vkCmdPipelineBarrier(p_sync_with_draw ? frames[frame].draw_command_buffer : frames[frame].setup_command_buffer, p_src_stage_mask, p_dst_stage_mask, 0, 0, nullptr, 1, &buffer_mem_barrier, 0, nullptr); } /*****************/ /**** TEXTURE ****/ /*****************/ RID RenderingDeviceVulkan::texture_create(const TextureFormat &p_format, const TextureView &p_view, const Vector<Vector<uint8_t>> &p_data) { _THREAD_SAFE_METHOD_ VkImageCreateInfo image_create_info; image_create_info.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO; image_create_info.pNext = nullptr; image_create_info.flags = 0; VkImageFormatListCreateInfoKHR format_list_create_info; // Keep out of the if, needed for creation. Vector<VkFormat> allowed_formats; // Keep out of the if, needed for creation. if (p_format.shareable_formats.size()) { image_create_info.flags |= VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT; if (context->is_device_extension_enabled(VK_KHR_IMAGE_FORMAT_LIST_EXTENSION_NAME)) { for (int i = 0; i < p_format.shareable_formats.size(); i++) { allowed_formats.push_back(vulkan_formats[p_format.shareable_formats[i]]); } format_list_create_info.sType = VK_STRUCTURE_TYPE_IMAGE_FORMAT_LIST_CREATE_INFO_KHR; format_list_create_info.pNext = nullptr; format_list_create_info.viewFormatCount = allowed_formats.size(); format_list_create_info.pViewFormats = allowed_formats.ptr(); image_create_info.pNext = &format_list_create_info; ERR_FAIL_COND_V_MSG(p_format.shareable_formats.find(p_format.format) == -1, RID(), "If supplied a list of shareable formats, the current format must be present in the list"); ERR_FAIL_COND_V_MSG(p_view.format_override != DATA_FORMAT_MAX && p_format.shareable_formats.find(p_view.format_override) == -1, RID(), "If supplied a list of shareable formats, the current view format override must be present in the list"); } } if (p_format.texture_type == TEXTURE_TYPE_CUBE || p_format.texture_type == TEXTURE_TYPE_CUBE_ARRAY) { image_create_info.flags |= VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT; } /*if (p_format.type == TEXTURE_TYPE_2D || p_format.type == TEXTURE_TYPE_2D_ARRAY) { image_create_info.flags |= VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT; }*/ ERR_FAIL_INDEX_V(p_format.texture_type, TEXTURE_TYPE_MAX, RID()); image_create_info.imageType = vulkan_image_type[p_format.texture_type]; ERR_FAIL_COND_V_MSG(p_format.width < 1, RID(), "Width must be equal or greater than 1 for all textures"); image_create_info.format = vulkan_formats[p_format.format]; image_create_info.extent.width = p_format.width; if (image_create_info.imageType == VK_IMAGE_TYPE_3D || image_create_info.imageType == VK_IMAGE_TYPE_2D) { ERR_FAIL_COND_V_MSG(p_format.height < 1, RID(), "Height must be equal or greater than 1 for 2D and 3D textures"); image_create_info.extent.height = p_format.height; } else { image_create_info.extent.height = 1; } if (image_create_info.imageType == VK_IMAGE_TYPE_3D) { ERR_FAIL_COND_V_MSG(p_format.depth < 1, RID(), "Depth must be equal or greater than 1 for 3D textures"); image_create_info.extent.depth = p_format.depth; } else { image_create_info.extent.depth = 1; } ERR_FAIL_COND_V(p_format.mipmaps < 1, RID()); image_create_info.mipLevels = p_format.mipmaps; if (p_format.texture_type == TEXTURE_TYPE_1D_ARRAY || p_format.texture_type == TEXTURE_TYPE_2D_ARRAY || p_format.texture_type == TEXTURE_TYPE_CUBE_ARRAY || p_format.texture_type == TEXTURE_TYPE_CUBE) { ERR_FAIL_COND_V_MSG(p_format.array_layers < 1, RID(), "Amount of layers must be equal or greater than 1 for arrays and cubemaps."); ERR_FAIL_COND_V_MSG((p_format.texture_type == TEXTURE_TYPE_CUBE_ARRAY || p_format.texture_type == TEXTURE_TYPE_CUBE) && (p_format.array_layers % 6) != 0, RID(), "Cubemap and cubemap array textures must provide a layer number that is multiple of 6"); image_create_info.arrayLayers = p_format.array_layers; } else { image_create_info.arrayLayers = 1; } ERR_FAIL_INDEX_V(p_format.samples, TEXTURE_SAMPLES_MAX, RID()); image_create_info.samples = _ensure_supported_sample_count(p_format.samples); image_create_info.tiling = (p_format.usage_bits & TEXTURE_USAGE_CPU_READ_BIT) ? VK_IMAGE_TILING_LINEAR : VK_IMAGE_TILING_OPTIMAL; // Usage. image_create_info.usage = 0; if (p_format.usage_bits & TEXTURE_USAGE_SAMPLING_BIT) { image_create_info.usage |= VK_IMAGE_USAGE_SAMPLED_BIT; } if (p_format.usage_bits & TEXTURE_USAGE_STORAGE_BIT) { image_create_info.usage |= VK_IMAGE_USAGE_STORAGE_BIT; } if (p_format.usage_bits & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { image_create_info.usage |= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT; } if (p_format.usage_bits & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { image_create_info.usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT; } if (p_format.usage_bits & TEXTURE_USAGE_INPUT_ATTACHMENT_BIT) { image_create_info.usage |= VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT; } if (p_format.usage_bits & TEXTURE_USAGE_VRS_ATTACHMENT_BIT) { image_create_info.usage |= VK_IMAGE_USAGE_FRAGMENT_SHADING_RATE_ATTACHMENT_BIT_KHR; } if (p_format.usage_bits & TEXTURE_USAGE_CAN_UPDATE_BIT) { image_create_info.usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT; } if (p_format.usage_bits & TEXTURE_USAGE_CAN_COPY_FROM_BIT) { image_create_info.usage |= VK_IMAGE_USAGE_TRANSFER_SRC_BIT; } if (p_format.usage_bits & TEXTURE_USAGE_CAN_COPY_TO_BIT) { image_create_info.usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT; } image_create_info.sharingMode = VK_SHARING_MODE_EXCLUSIVE; image_create_info.queueFamilyIndexCount = 0; image_create_info.pQueueFamilyIndices = nullptr; image_create_info.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; uint32_t required_mipmaps = get_image_required_mipmaps(image_create_info.extent.width, image_create_info.extent.height, image_create_info.extent.depth); ERR_FAIL_COND_V_MSG(required_mipmaps < image_create_info.mipLevels, RID(), "Too many mipmaps requested for texture format and dimensions (" + itos(image_create_info.mipLevels) + "), maximum allowed: (" + itos(required_mipmaps) + ")."); if (p_data.size()) { ERR_FAIL_COND_V_MSG(!(p_format.usage_bits & TEXTURE_USAGE_CAN_UPDATE_BIT), RID(), "Texture needs the TEXTURE_USAGE_CAN_UPDATE_BIT usage flag in order to be updated at initialization or later"); int expected_images = image_create_info.arrayLayers; ERR_FAIL_COND_V_MSG(p_data.size() != expected_images, RID(), "Default supplied data for image format is of invalid length (" + itos(p_data.size()) + "), should be (" + itos(expected_images) + ")."); for (uint32_t i = 0; i < image_create_info.arrayLayers; i++) { uint32_t required_size = get_image_format_required_size(p_format.format, image_create_info.extent.width, image_create_info.extent.height, image_create_info.extent.depth, image_create_info.mipLevels); ERR_FAIL_COND_V_MSG((uint32_t)p_data[i].size() != required_size, RID(), "Data for slice index " + itos(i) + " (mapped to layer " + itos(i) + ") differs in size (supplied: " + itos(p_data[i].size()) + ") than what is required by the format (" + itos(required_size) + ")."); } } { // Validate that this image is supported for the intended use. VkFormatProperties properties; vkGetPhysicalDeviceFormatProperties(context->get_physical_device(), image_create_info.format, &properties); VkFormatFeatureFlags flags; String format_text = "'" + String(named_formats[p_format.format]) + "'"; if (p_format.usage_bits & TEXTURE_USAGE_CPU_READ_BIT) { flags = properties.linearTilingFeatures; format_text += " (with CPU read bit)"; } else { flags = properties.optimalTilingFeatures; } if (p_format.usage_bits & TEXTURE_USAGE_SAMPLING_BIT && !(flags & VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT)) { ERR_FAIL_V_MSG(RID(), "Format " + format_text + " does not support usage as sampling texture."); } if (p_format.usage_bits & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT && !(flags & VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BIT)) { ERR_FAIL_V_MSG(RID(), "Format " + format_text + " does not support usage as color attachment."); } if (p_format.usage_bits & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT && !(flags & VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT)) { ERR_FAIL_V_MSG(RID(), "Format " + format_text + " does not support usage as depth-stencil attachment."); } if (p_format.usage_bits & TEXTURE_USAGE_STORAGE_BIT && !(flags & VK_FORMAT_FEATURE_STORAGE_IMAGE_BIT)) { ERR_FAIL_V_MSG(RID(), "Format " + format_text + " does not support usage as storage image."); } if (p_format.usage_bits & TEXTURE_USAGE_STORAGE_ATOMIC_BIT && !(flags & VK_FORMAT_FEATURE_STORAGE_IMAGE_ATOMIC_BIT)) { ERR_FAIL_V_MSG(RID(), "Format " + format_text + " does not support usage as atomic storage image."); } // Validation via VK_FORMAT_FEATURE_FRAGMENT_SHADING_RATE_ATTACHMENT_BIT_KHR fails if VRS attachment is not supported. if (p_format.usage_bits & TEXTURE_USAGE_VRS_ATTACHMENT_BIT && p_format.format != DATA_FORMAT_R8_UINT) { ERR_FAIL_V_MSG(RID(), "Format " + format_text + " does not support usage as VRS attachment."); } } // Some view validation. if (p_view.format_override != DATA_FORMAT_MAX) { ERR_FAIL_INDEX_V(p_view.format_override, DATA_FORMAT_MAX, RID()); } ERR_FAIL_INDEX_V(p_view.swizzle_r, TEXTURE_SWIZZLE_MAX, RID()); ERR_FAIL_INDEX_V(p_view.swizzle_g, TEXTURE_SWIZZLE_MAX, RID()); ERR_FAIL_INDEX_V(p_view.swizzle_b, TEXTURE_SWIZZLE_MAX, RID()); ERR_FAIL_INDEX_V(p_view.swizzle_a, TEXTURE_SWIZZLE_MAX, RID()); // Allocate memory. uint32_t width, height; uint32_t image_size = get_image_format_required_size(p_format.format, p_format.width, p_format.height, p_format.depth, p_format.mipmaps, &width, &height); VmaAllocationCreateInfo allocInfo; allocInfo.flags = (p_format.usage_bits & TEXTURE_USAGE_CPU_READ_BIT) ? VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT : 0; allocInfo.pool = nullptr; allocInfo.usage = VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE; allocInfo.requiredFlags = 0; allocInfo.preferredFlags = 0; allocInfo.memoryTypeBits = 0; allocInfo.pUserData = nullptr; if (image_size <= SMALL_ALLOCATION_MAX_SIZE) { uint32_t mem_type_index = 0; vmaFindMemoryTypeIndexForImageInfo(allocator, &image_create_info, &allocInfo, &mem_type_index); allocInfo.pool = _find_or_create_small_allocs_pool(mem_type_index); } Texture texture; VkResult err = vmaCreateImage(allocator, &image_create_info, &allocInfo, &texture.image, &texture.allocation, &texture.allocation_info); ERR_FAIL_COND_V_MSG(err, RID(), "vmaCreateImage failed with error " + itos(err) + "."); image_memory += texture.allocation_info.size; texture.type = p_format.texture_type; texture.format = p_format.format; texture.width = image_create_info.extent.width; texture.height = image_create_info.extent.height; texture.depth = image_create_info.extent.depth; texture.layers = image_create_info.arrayLayers; texture.mipmaps = image_create_info.mipLevels; texture.base_mipmap = 0; texture.base_layer = 0; texture.is_resolve_buffer = p_format.is_resolve_buffer; texture.usage_flags = p_format.usage_bits; texture.samples = p_format.samples; texture.allowed_shared_formats = p_format.shareable_formats; // Set base layout based on usage priority. if (p_format.usage_bits & TEXTURE_USAGE_SAMPLING_BIT) { // First priority, readable. texture.layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; } else if (p_format.usage_bits & TEXTURE_USAGE_STORAGE_BIT) { // Second priority, storage. texture.layout = VK_IMAGE_LAYOUT_GENERAL; } else if (p_format.usage_bits & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { // Third priority, color or depth. texture.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; } else if (p_format.usage_bits & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { texture.layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; } else { texture.layout = VK_IMAGE_LAYOUT_GENERAL; } if (p_format.usage_bits & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { texture.read_aspect_mask = VK_IMAGE_ASPECT_DEPTH_BIT; texture.barrier_aspect_mask = VK_IMAGE_ASPECT_DEPTH_BIT; if (format_has_stencil(p_format.format)) { texture.barrier_aspect_mask |= VK_IMAGE_ASPECT_STENCIL_BIT; } } else { texture.read_aspect_mask = VK_IMAGE_ASPECT_COLOR_BIT; texture.barrier_aspect_mask = VK_IMAGE_ASPECT_COLOR_BIT; } texture.bound = false; // Create view. VkImageViewCreateInfo image_view_create_info; image_view_create_info.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO; image_view_create_info.pNext = nullptr; image_view_create_info.flags = 0; image_view_create_info.image = texture.image; static const VkImageViewType view_types[TEXTURE_TYPE_MAX] = { VK_IMAGE_VIEW_TYPE_1D, VK_IMAGE_VIEW_TYPE_2D, VK_IMAGE_VIEW_TYPE_3D, VK_IMAGE_VIEW_TYPE_CUBE, VK_IMAGE_VIEW_TYPE_1D_ARRAY, VK_IMAGE_VIEW_TYPE_2D_ARRAY, VK_IMAGE_VIEW_TYPE_CUBE_ARRAY, }; image_view_create_info.viewType = view_types[p_format.texture_type]; if (p_view.format_override == DATA_FORMAT_MAX) { image_view_create_info.format = image_create_info.format; } else { image_view_create_info.format = vulkan_formats[p_view.format_override]; } static const VkComponentSwizzle component_swizzles[TEXTURE_SWIZZLE_MAX] = { VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_ZERO, VK_COMPONENT_SWIZZLE_ONE, VK_COMPONENT_SWIZZLE_R, VK_COMPONENT_SWIZZLE_G, VK_COMPONENT_SWIZZLE_B, VK_COMPONENT_SWIZZLE_A }; image_view_create_info.components.r = component_swizzles[p_view.swizzle_r]; image_view_create_info.components.g = component_swizzles[p_view.swizzle_g]; image_view_create_info.components.b = component_swizzles[p_view.swizzle_b]; image_view_create_info.components.a = component_swizzles[p_view.swizzle_a]; image_view_create_info.subresourceRange.baseMipLevel = 0; image_view_create_info.subresourceRange.levelCount = image_create_info.mipLevels; image_view_create_info.subresourceRange.baseArrayLayer = 0; image_view_create_info.subresourceRange.layerCount = image_create_info.arrayLayers; if (p_format.usage_bits & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { image_view_create_info.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT; } else { image_view_create_info.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; } err = vkCreateImageView(device, &image_view_create_info, nullptr, &texture.view); if (err) { vmaDestroyImage(allocator, texture.image, texture.allocation); ERR_FAIL_V_MSG(RID(), "vkCreateImageView failed with error " + itos(err) + "."); } // Barrier to set layout. { VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = 0; image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT; image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED; image_memory_barrier.newLayout = texture.layout; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = texture.image; image_memory_barrier.subresourceRange.aspectMask = texture.barrier_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = 0; image_memory_barrier.subresourceRange.levelCount = image_create_info.mipLevels; image_memory_barrier.subresourceRange.baseArrayLayer = 0; image_memory_barrier.subresourceRange.layerCount = image_create_info.arrayLayers; vkCmdPipelineBarrier(frames[frame].setup_command_buffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } RID id = texture_owner.make_rid(texture); #ifdef DEV_ENABLED set_resource_name(id, "RID:" + itos(id.get_id())); #endif if (p_data.size()) { for (uint32_t i = 0; i < image_create_info.arrayLayers; i++) { _texture_update(id, i, p_data[i], RD::BARRIER_MASK_ALL_BARRIERS, true); } } return id; } RID RenderingDeviceVulkan::texture_create_shared(const TextureView &p_view, RID p_with_texture) { _THREAD_SAFE_METHOD_ Texture *src_texture = texture_owner.get_or_null(p_with_texture); ERR_FAIL_NULL_V(src_texture, RID()); if (src_texture->owner.is_valid()) { // Ahh this is a share. p_with_texture = src_texture->owner; src_texture = texture_owner.get_or_null(src_texture->owner); ERR_FAIL_NULL_V(src_texture, RID()); // This is a bug. } // Create view. Texture texture = *src_texture; VkImageViewCreateInfo image_view_create_info; image_view_create_info.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO; image_view_create_info.pNext = nullptr; image_view_create_info.flags = 0; image_view_create_info.image = texture.image; static const VkImageViewType view_types[TEXTURE_TYPE_MAX] = { VK_IMAGE_VIEW_TYPE_1D, VK_IMAGE_VIEW_TYPE_2D, VK_IMAGE_VIEW_TYPE_3D, VK_IMAGE_VIEW_TYPE_CUBE, VK_IMAGE_VIEW_TYPE_1D_ARRAY, VK_IMAGE_VIEW_TYPE_2D_ARRAY, VK_IMAGE_VIEW_TYPE_CUBE_ARRAY, }; image_view_create_info.viewType = view_types[texture.type]; if (p_view.format_override == DATA_FORMAT_MAX || p_view.format_override == texture.format) { image_view_create_info.format = vulkan_formats[texture.format]; } else { ERR_FAIL_INDEX_V(p_view.format_override, DATA_FORMAT_MAX, RID()); ERR_FAIL_COND_V_MSG(texture.allowed_shared_formats.find(p_view.format_override) == -1, RID(), "Format override is not in the list of allowed shareable formats for original texture."); image_view_create_info.format = vulkan_formats[p_view.format_override]; } static const VkComponentSwizzle component_swizzles[TEXTURE_SWIZZLE_MAX] = { VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_ZERO, VK_COMPONENT_SWIZZLE_ONE, VK_COMPONENT_SWIZZLE_R, VK_COMPONENT_SWIZZLE_G, VK_COMPONENT_SWIZZLE_B, VK_COMPONENT_SWIZZLE_A }; image_view_create_info.components.r = component_swizzles[p_view.swizzle_r]; image_view_create_info.components.g = component_swizzles[p_view.swizzle_g]; image_view_create_info.components.b = component_swizzles[p_view.swizzle_b]; image_view_create_info.components.a = component_swizzles[p_view.swizzle_a]; image_view_create_info.subresourceRange.baseMipLevel = 0; image_view_create_info.subresourceRange.levelCount = texture.mipmaps; image_view_create_info.subresourceRange.layerCount = texture.layers; image_view_create_info.subresourceRange.baseArrayLayer = 0; if (texture.usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { image_view_create_info.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT; } else { image_view_create_info.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; } VkImageViewUsageCreateInfo usage_info; if (context->is_device_extension_enabled(VK_KHR_MAINTENANCE_2_EXTENSION_NAME)) { // May need to make VK_KHR_maintenance2 manditory and thus has Vulkan 1.1 be our minimum supported version // if we require setting this information. Vulkan 1.0 may simply not care.. usage_info.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO; usage_info.pNext = nullptr; if (p_view.format_override != DATA_FORMAT_MAX) { // Need to validate usage with vulkan. usage_info.usage = 0; if (texture.usage_flags & TEXTURE_USAGE_SAMPLING_BIT) { usage_info.usage |= VK_IMAGE_USAGE_SAMPLED_BIT; } if (texture.usage_flags & TEXTURE_USAGE_STORAGE_BIT) { if (texture_is_format_supported_for_usage(p_view.format_override, TEXTURE_USAGE_STORAGE_BIT)) { usage_info.usage |= VK_IMAGE_USAGE_STORAGE_BIT; } } if (texture.usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { if (texture_is_format_supported_for_usage(p_view.format_override, TEXTURE_USAGE_COLOR_ATTACHMENT_BIT)) { usage_info.usage |= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT; } } if (texture.usage_flags & TEXTURE_USAGE_INPUT_ATTACHMENT_BIT) { usage_info.usage |= VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT; } if (texture.usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { usage_info.usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT; } if (texture.usage_flags & TEXTURE_USAGE_CAN_UPDATE_BIT) { usage_info.usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT; } if (texture.usage_flags & TEXTURE_USAGE_CAN_COPY_FROM_BIT) { usage_info.usage |= VK_IMAGE_USAGE_TRANSFER_SRC_BIT; } if (texture.usage_flags & TEXTURE_USAGE_CAN_COPY_TO_BIT) { usage_info.usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT; } image_view_create_info.pNext = &usage_info; } } VkResult err = vkCreateImageView(device, &image_view_create_info, nullptr, &texture.view); ERR_FAIL_COND_V_MSG(err, RID(), "vkCreateImageView failed with error " + itos(err) + "."); texture.owner = p_with_texture; RID id = texture_owner.make_rid(texture); #ifdef DEV_ENABLED set_resource_name(id, "RID:" + itos(id.get_id())); #endif _add_dependency(id, p_with_texture); return id; } RID RenderingDeviceVulkan::texture_create_from_extension(TextureType p_type, DataFormat p_format, TextureSamples p_samples, uint64_t p_flags, uint64_t p_image, uint64_t p_width, uint64_t p_height, uint64_t p_depth, uint64_t p_layers) { _THREAD_SAFE_METHOD_ // This method creates a texture object using a VkImage created by an extension, module or other external source (OpenXR uses this). VkImage image = (VkImage)p_image; Texture texture; texture.image = image; // If we leave texture.allocation as a nullptr, would that be enough to detect we don't "own" the image? // Also leave texture.allocation_info alone. // We'll set texture.view later on. texture.type = p_type; texture.format = p_format; texture.samples = p_samples; texture.width = p_width; texture.height = p_height; texture.depth = p_depth; texture.layers = p_layers; texture.mipmaps = 1; texture.usage_flags = p_flags; texture.base_mipmap = 0; texture.base_layer = 0; texture.allowed_shared_formats.push_back(RD::DATA_FORMAT_R8G8B8A8_UNORM); texture.allowed_shared_formats.push_back(RD::DATA_FORMAT_R8G8B8A8_SRGB); // Set base layout based on usage priority. if (texture.usage_flags & TEXTURE_USAGE_SAMPLING_BIT) { // First priority, readable. texture.layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; } else if (texture.usage_flags & TEXTURE_USAGE_STORAGE_BIT) { // Second priority, storage. texture.layout = VK_IMAGE_LAYOUT_GENERAL; } else if (texture.usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { // Third priority, color or depth. texture.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; } else if (texture.usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { texture.layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; } else { texture.layout = VK_IMAGE_LAYOUT_GENERAL; } if (texture.usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { texture.read_aspect_mask = VK_IMAGE_ASPECT_DEPTH_BIT; texture.barrier_aspect_mask = VK_IMAGE_ASPECT_DEPTH_BIT; // if (format_has_stencil(p_format.format)) { // texture.barrier_aspect_mask |= VK_IMAGE_ASPECT_STENCIL_BIT; // } } else { texture.read_aspect_mask = VK_IMAGE_ASPECT_COLOR_BIT; texture.barrier_aspect_mask = VK_IMAGE_ASPECT_COLOR_BIT; } // Create a view for us to use. VkImageViewCreateInfo image_view_create_info; image_view_create_info.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO; image_view_create_info.pNext = nullptr; image_view_create_info.flags = 0; image_view_create_info.image = texture.image; static const VkImageViewType view_types[TEXTURE_TYPE_MAX] = { VK_IMAGE_VIEW_TYPE_1D, VK_IMAGE_VIEW_TYPE_2D, VK_IMAGE_VIEW_TYPE_3D, VK_IMAGE_VIEW_TYPE_CUBE, VK_IMAGE_VIEW_TYPE_1D_ARRAY, VK_IMAGE_VIEW_TYPE_2D_ARRAY, VK_IMAGE_VIEW_TYPE_CUBE_ARRAY, }; image_view_create_info.viewType = view_types[texture.type]; image_view_create_info.format = vulkan_formats[texture.format]; static const VkComponentSwizzle component_swizzles[TEXTURE_SWIZZLE_MAX] = { VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_ZERO, VK_COMPONENT_SWIZZLE_ONE, VK_COMPONENT_SWIZZLE_R, VK_COMPONENT_SWIZZLE_G, VK_COMPONENT_SWIZZLE_B, VK_COMPONENT_SWIZZLE_A }; // Hardcode for now, maybe make this settable from outside. image_view_create_info.components.r = component_swizzles[TEXTURE_SWIZZLE_R]; image_view_create_info.components.g = component_swizzles[TEXTURE_SWIZZLE_G]; image_view_create_info.components.b = component_swizzles[TEXTURE_SWIZZLE_B]; image_view_create_info.components.a = component_swizzles[TEXTURE_SWIZZLE_A]; image_view_create_info.subresourceRange.baseMipLevel = 0; image_view_create_info.subresourceRange.levelCount = texture.mipmaps; image_view_create_info.subresourceRange.baseArrayLayer = 0; image_view_create_info.subresourceRange.layerCount = texture.layers; if (texture.usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { image_view_create_info.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT; } else { image_view_create_info.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; } VkResult err = vkCreateImageView(device, &image_view_create_info, nullptr, &texture.view); if (err) { // vmaDestroyImage(allocator, texture.image, texture.allocation); ERR_FAIL_V_MSG(RID(), "vkCreateImageView failed with error " + itos(err) + "."); } // Barrier to set layout. { VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = 0; image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT; image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED; image_memory_barrier.newLayout = texture.layout; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = texture.image; image_memory_barrier.subresourceRange.aspectMask = texture.barrier_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = 0; image_memory_barrier.subresourceRange.levelCount = texture.mipmaps; image_memory_barrier.subresourceRange.baseArrayLayer = 0; image_memory_barrier.subresourceRange.layerCount = texture.layers; vkCmdPipelineBarrier(frames[frame].setup_command_buffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } RID id = texture_owner.make_rid(texture); #ifdef DEV_ENABLED set_resource_name(id, "RID:" + itos(id.get_id())); #endif return id; } RID RenderingDeviceVulkan::texture_create_shared_from_slice(const TextureView &p_view, RID p_with_texture, uint32_t p_layer, uint32_t p_mipmap, uint32_t p_mipmaps, TextureSliceType p_slice_type, uint32_t p_layers) { _THREAD_SAFE_METHOD_ Texture *src_texture = texture_owner.get_or_null(p_with_texture); ERR_FAIL_NULL_V(src_texture, RID()); if (src_texture->owner.is_valid()) { // Ahh this is a share. p_with_texture = src_texture->owner; src_texture = texture_owner.get_or_null(src_texture->owner); ERR_FAIL_NULL_V(src_texture, RID()); // This is a bug. } ERR_FAIL_COND_V_MSG(p_slice_type == TEXTURE_SLICE_CUBEMAP && (src_texture->type != TEXTURE_TYPE_CUBE && src_texture->type != TEXTURE_TYPE_CUBE_ARRAY), RID(), "Can only create a cubemap slice from a cubemap or cubemap array mipmap"); ERR_FAIL_COND_V_MSG(p_slice_type == TEXTURE_SLICE_3D && src_texture->type != TEXTURE_TYPE_3D, RID(), "Can only create a 3D slice from a 3D texture"); ERR_FAIL_COND_V_MSG(p_slice_type == TEXTURE_SLICE_2D_ARRAY && (src_texture->type != TEXTURE_TYPE_2D_ARRAY), RID(), "Can only create an array slice from a 2D array mipmap"); // Create view. ERR_FAIL_UNSIGNED_INDEX_V(p_mipmap, src_texture->mipmaps, RID()); ERR_FAIL_COND_V(p_mipmap + p_mipmaps > src_texture->mipmaps, RID()); ERR_FAIL_UNSIGNED_INDEX_V(p_layer, src_texture->layers, RID()); int slice_layers = 1; if (p_layers != 0) { ERR_FAIL_COND_V_MSG(p_layers > 1 && p_slice_type != TEXTURE_SLICE_2D_ARRAY, RID(), "layer slicing only supported for 2D arrays"); ERR_FAIL_COND_V_MSG(p_layer + p_layers > src_texture->layers, RID(), "layer slice is out of bounds"); slice_layers = p_layers; } else if (p_slice_type == TEXTURE_SLICE_2D_ARRAY) { ERR_FAIL_COND_V_MSG(p_layer != 0, RID(), "layer must be 0 when obtaining a 2D array mipmap slice"); slice_layers = src_texture->layers; } else if (p_slice_type == TEXTURE_SLICE_CUBEMAP) { slice_layers = 6; } Texture texture = *src_texture; get_image_format_required_size(texture.format, texture.width, texture.height, texture.depth, p_mipmap + 1, &texture.width, &texture.height); texture.mipmaps = p_mipmaps; texture.layers = slice_layers; texture.base_mipmap = p_mipmap; texture.base_layer = p_layer; VkImageViewCreateInfo image_view_create_info; image_view_create_info.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO; image_view_create_info.pNext = nullptr; image_view_create_info.flags = 0; image_view_create_info.image = texture.image; static const VkImageViewType view_types[TEXTURE_TYPE_MAX] = { VK_IMAGE_VIEW_TYPE_1D, VK_IMAGE_VIEW_TYPE_2D, VK_IMAGE_VIEW_TYPE_2D, VK_IMAGE_VIEW_TYPE_2D, VK_IMAGE_VIEW_TYPE_1D, VK_IMAGE_VIEW_TYPE_2D, VK_IMAGE_VIEW_TYPE_2D, }; image_view_create_info.viewType = view_types[texture.type]; if (p_slice_type == TEXTURE_SLICE_CUBEMAP) { image_view_create_info.viewType = VK_IMAGE_VIEW_TYPE_CUBE; } else if (p_slice_type == TEXTURE_SLICE_3D) { image_view_create_info.viewType = VK_IMAGE_VIEW_TYPE_3D; } else if (p_slice_type == TEXTURE_SLICE_2D_ARRAY) { image_view_create_info.viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY; } if (p_slice_type == TEXTURE_SLICE_2D) { texture.type = TEXTURE_TYPE_2D; } else if (p_slice_type == TEXTURE_SLICE_3D) { texture.type = TEXTURE_TYPE_3D; } if (p_view.format_override == DATA_FORMAT_MAX || p_view.format_override == texture.format) { image_view_create_info.format = vulkan_formats[texture.format]; } else { ERR_FAIL_INDEX_V(p_view.format_override, DATA_FORMAT_MAX, RID()); ERR_FAIL_COND_V_MSG(texture.allowed_shared_formats.find(p_view.format_override) == -1, RID(), "Format override is not in the list of allowed shareable formats for original texture."); image_view_create_info.format = vulkan_formats[p_view.format_override]; } static const VkComponentSwizzle component_swizzles[TEXTURE_SWIZZLE_MAX] = { VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_ZERO, VK_COMPONENT_SWIZZLE_ONE, VK_COMPONENT_SWIZZLE_R, VK_COMPONENT_SWIZZLE_G, VK_COMPONENT_SWIZZLE_B, VK_COMPONENT_SWIZZLE_A }; image_view_create_info.components.r = component_swizzles[p_view.swizzle_r]; image_view_create_info.components.g = component_swizzles[p_view.swizzle_g]; image_view_create_info.components.b = component_swizzles[p_view.swizzle_b]; image_view_create_info.components.a = component_swizzles[p_view.swizzle_a]; if (p_slice_type == TEXTURE_SLICE_CUBEMAP) { ERR_FAIL_COND_V_MSG(p_layer >= src_texture->layers, RID(), "Specified layer is invalid for cubemap"); ERR_FAIL_COND_V_MSG((p_layer % 6) != 0, RID(), "Specified layer must be a multiple of 6."); } image_view_create_info.subresourceRange.baseMipLevel = p_mipmap; image_view_create_info.subresourceRange.levelCount = p_mipmaps; image_view_create_info.subresourceRange.layerCount = slice_layers; image_view_create_info.subresourceRange.baseArrayLayer = p_layer; if (texture.usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { image_view_create_info.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT; } else { image_view_create_info.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; } VkResult err = vkCreateImageView(device, &image_view_create_info, nullptr, &texture.view); ERR_FAIL_COND_V_MSG(err, RID(), "vkCreateImageView failed with error " + itos(err) + "."); texture.owner = p_with_texture; RID id = texture_owner.make_rid(texture); #ifdef DEV_ENABLED set_resource_name(id, "RID:" + itos(id.get_id())); #endif _add_dependency(id, p_with_texture); return id; } Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, const Vector<uint8_t> &p_data, BitField<BarrierMask> p_post_barrier) { return _texture_update(p_texture, p_layer, p_data, p_post_barrier, false); } static _ALWAYS_INLINE_ void _copy_region(uint8_t const *__restrict p_src, uint8_t *__restrict p_dst, uint32_t p_src_x, uint32_t p_src_y, uint32_t p_src_w, uint32_t p_src_h, uint32_t p_src_full_w, uint32_t p_unit_size) { uint32_t src_offset = (p_src_y * p_src_full_w + p_src_x) * p_unit_size; uint32_t dst_offset = 0; for (uint32_t y = p_src_h; y > 0; y--) { uint8_t const *__restrict src = p_src + src_offset; uint8_t *__restrict dst = p_dst + dst_offset; for (uint32_t x = p_src_w * p_unit_size; x > 0; x--) { *dst = *src; src++; dst++; } src_offset += p_src_full_w * p_unit_size; dst_offset += p_src_w * p_unit_size; } } Error RenderingDeviceVulkan::_texture_update(RID p_texture, uint32_t p_layer, const Vector<uint8_t> &p_data, BitField<BarrierMask> p_post_barrier, bool p_use_setup_queue) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG((draw_list || compute_list) && !p_use_setup_queue, ERR_INVALID_PARAMETER, "Updating textures is forbidden during creation of a draw or compute list"); Texture *texture = texture_owner.get_or_null(p_texture); ERR_FAIL_NULL_V(texture, ERR_INVALID_PARAMETER); if (texture->owner != RID()) { p_texture = texture->owner; texture = texture_owner.get_or_null(texture->owner); ERR_FAIL_NULL_V(texture, ERR_BUG); // This is a bug. } ERR_FAIL_COND_V_MSG(texture->bound, ERR_CANT_ACQUIRE_RESOURCE, "Texture can't be updated while a draw list that uses it as part of a framebuffer is being created. Ensure the draw list is finalized (and that the color/depth texture using it is not set to `RenderingDevice.FINAL_ACTION_CONTINUE`) to update this texture."); ERR_FAIL_COND_V_MSG(!(texture->usage_flags & TEXTURE_USAGE_CAN_UPDATE_BIT), ERR_INVALID_PARAMETER, "Texture requires the `RenderingDevice.TEXTURE_USAGE_CAN_UPDATE_BIT` to be set to be updatable."); uint32_t layer_count = texture->layers; if (texture->type == TEXTURE_TYPE_CUBE || texture->type == TEXTURE_TYPE_CUBE_ARRAY) { layer_count *= 6; } ERR_FAIL_COND_V(p_layer >= layer_count, ERR_INVALID_PARAMETER); uint32_t width, height; uint32_t image_size = get_image_format_required_size(texture->format, texture->width, texture->height, texture->depth, texture->mipmaps, &width, &height); uint32_t required_size = image_size; uint32_t required_align = get_compressed_image_format_block_byte_size(texture->format); if (required_align == 1) { required_align = get_image_format_pixel_size(texture->format); } if ((required_align % 4) != 0) { // Alignment rules are really strange. required_align *= 4; } ERR_FAIL_COND_V_MSG(required_size != (uint32_t)p_data.size(), ERR_INVALID_PARAMETER, "Required size for texture update (" + itos(required_size) + ") does not match data supplied size (" + itos(p_data.size()) + ")."); uint32_t region_size = texture_upload_region_size_px; const uint8_t *r = p_data.ptr(); VkCommandBuffer command_buffer = p_use_setup_queue ? frames[frame].setup_command_buffer : frames[frame].draw_command_buffer; // Barrier to transfer. { VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = 0; image_memory_barrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; image_memory_barrier.oldLayout = texture->layout; image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = texture->image; image_memory_barrier.subresourceRange.aspectMask = texture->barrier_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = 0; image_memory_barrier.subresourceRange.levelCount = texture->mipmaps; image_memory_barrier.subresourceRange.baseArrayLayer = p_layer; image_memory_barrier.subresourceRange.layerCount = 1; vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } uint32_t mipmap_offset = 0; uint32_t logic_width = texture->width; uint32_t logic_height = texture->height; for (uint32_t mm_i = 0; mm_i < texture->mipmaps; mm_i++) { uint32_t depth; uint32_t image_total = get_image_format_required_size(texture->format, texture->width, texture->height, texture->depth, mm_i + 1, &width, &height, &depth); const uint8_t *read_ptr_mipmap = r + mipmap_offset; image_size = image_total - mipmap_offset; for (uint32_t z = 0; z < depth; z++) { // For 3D textures, depth may be > 0. const uint8_t *read_ptr = read_ptr_mipmap + (image_size / depth) * z; for (uint32_t y = 0; y < height; y += region_size) { for (uint32_t x = 0; x < width; x += region_size) { uint32_t region_w = MIN(region_size, width - x); uint32_t region_h = MIN(region_size, height - y); uint32_t region_logic_w = MIN(region_size, logic_width - x); uint32_t region_logic_h = MIN(region_size, logic_height - y); uint32_t pixel_size = get_image_format_pixel_size(texture->format); uint32_t to_allocate = region_w * region_h * pixel_size; to_allocate >>= get_compressed_image_format_pixel_rshift(texture->format); uint32_t alloc_offset, alloc_size; Error err = _staging_buffer_allocate(to_allocate, required_align, alloc_offset, alloc_size, false); ERR_FAIL_COND_V(err, ERR_CANT_CREATE); uint8_t *write_ptr; { // Map. void *data_ptr = nullptr; VkResult vkerr = vmaMapMemory(allocator, staging_buffer_blocks[staging_buffer_current].allocation, &data_ptr); ERR_FAIL_COND_V_MSG(vkerr, ERR_CANT_CREATE, "vmaMapMemory failed with error " + itos(vkerr) + "."); write_ptr = (uint8_t *)data_ptr; write_ptr += alloc_offset; } uint32_t block_w, block_h; get_compressed_image_format_block_dimensions(texture->format, block_w, block_h); ERR_FAIL_COND_V(region_w % block_w, ERR_BUG); ERR_FAIL_COND_V(region_h % block_h, ERR_BUG); if (block_w != 1 || block_h != 1) { // Compressed image (blocks). // Must copy a block region. uint32_t block_size = get_compressed_image_format_block_byte_size(texture->format); // Re-create current variables in blocky format. uint32_t xb = x / block_w; uint32_t yb = y / block_h; uint32_t wb = width / block_w; //uint32_t hb = height / block_h; uint32_t region_wb = region_w / block_w; uint32_t region_hb = region_h / block_h; _copy_region(read_ptr, write_ptr, xb, yb, region_wb, region_hb, wb, block_size); } else { // Regular image (pixels). // Must copy a pixel region. _copy_region(read_ptr, write_ptr, x, y, region_w, region_h, width, pixel_size); } { // Unmap. vmaUnmapMemory(allocator, staging_buffer_blocks[staging_buffer_current].allocation); } VkBufferImageCopy buffer_image_copy; buffer_image_copy.bufferOffset = alloc_offset; buffer_image_copy.bufferRowLength = 0; // Tightly packed. buffer_image_copy.bufferImageHeight = 0; // Tightly packed. buffer_image_copy.imageSubresource.aspectMask = texture->read_aspect_mask; buffer_image_copy.imageSubresource.mipLevel = mm_i; buffer_image_copy.imageSubresource.baseArrayLayer = p_layer; buffer_image_copy.imageSubresource.layerCount = 1; buffer_image_copy.imageOffset.x = x; buffer_image_copy.imageOffset.y = y; buffer_image_copy.imageOffset.z = z; buffer_image_copy.imageExtent.width = region_logic_w; buffer_image_copy.imageExtent.height = region_logic_h; buffer_image_copy.imageExtent.depth = 1; vkCmdCopyBufferToImage(command_buffer, staging_buffer_blocks[staging_buffer_current].buffer, texture->image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &buffer_image_copy); staging_buffer_blocks.write[staging_buffer_current].fill_amount = alloc_offset + alloc_size; } } } mipmap_offset = image_total; logic_width = MAX(1u, logic_width >> 1); logic_height = MAX(1u, logic_height >> 1); } // Barrier to restore layout. { uint32_t barrier_flags = 0; uint32_t access_flags = 0; if (p_post_barrier.has_flag(BARRIER_MASK_COMPUTE)) { barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_VERTEX)) { barrier_flags |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_FRAGMENT)) { barrier_flags |= VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_TRANSFER)) { barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT; } if (barrier_flags == 0) { barrier_flags = VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT; } VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; image_memory_barrier.dstAccessMask = access_flags; image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL; image_memory_barrier.newLayout = texture->layout; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = texture->image; image_memory_barrier.subresourceRange.aspectMask = texture->barrier_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = 0; image_memory_barrier.subresourceRange.levelCount = texture->mipmaps; image_memory_barrier.subresourceRange.baseArrayLayer = p_layer; image_memory_barrier.subresourceRange.layerCount = 1; vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, barrier_flags, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } if (texture->used_in_frame != frames_drawn) { texture->used_in_raster = false; texture->used_in_compute = false; texture->used_in_frame = frames_drawn; } texture->used_in_transfer = true; return OK; } Vector<uint8_t> RenderingDeviceVulkan::_texture_get_data_from_image(Texture *tex, VkImage p_image, VmaAllocation p_allocation, uint32_t p_layer, bool p_2d) { uint32_t width, height, depth; uint32_t image_size = get_image_format_required_size(tex->format, tex->width, tex->height, p_2d ? 1 : tex->depth, tex->mipmaps, &width, &height, &depth); Vector<uint8_t> image_data; image_data.resize(image_size); void *img_mem; vmaMapMemory(allocator, p_allocation, &img_mem); uint32_t blockw, blockh; get_compressed_image_format_block_dimensions(tex->format, blockw, blockh); uint32_t block_size = get_compressed_image_format_block_byte_size(tex->format); uint32_t pixel_size = get_image_format_pixel_size(tex->format); { uint8_t *w = image_data.ptrw(); uint32_t mipmap_offset = 0; for (uint32_t mm_i = 0; mm_i < tex->mipmaps; mm_i++) { uint32_t image_total = get_image_format_required_size(tex->format, tex->width, tex->height, p_2d ? 1 : tex->depth, mm_i + 1, &width, &height, &depth); uint8_t *write_ptr_mipmap = w + mipmap_offset; image_size = image_total - mipmap_offset; VkImageSubresource image_sub_resorce; image_sub_resorce.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; image_sub_resorce.arrayLayer = p_layer; image_sub_resorce.mipLevel = mm_i; VkSubresourceLayout layout; vkGetImageSubresourceLayout(device, p_image, &image_sub_resorce, &layout); for (uint32_t z = 0; z < depth; z++) { uint8_t *write_ptr = write_ptr_mipmap + z * image_size / depth; const uint8_t *slice_read_ptr = ((uint8_t *)img_mem) + layout.offset + z * layout.depthPitch; if (block_size > 1) { // Compressed. uint32_t line_width = (block_size * (width / blockw)); for (uint32_t y = 0; y < height / blockh; y++) { const uint8_t *rptr = slice_read_ptr + y * layout.rowPitch; uint8_t *wptr = write_ptr + y * line_width; memcpy(wptr, rptr, line_width); } } else { // Uncompressed. for (uint32_t y = 0; y < height; y++) { const uint8_t *rptr = slice_read_ptr + y * layout.rowPitch; uint8_t *wptr = write_ptr + y * pixel_size * width; memcpy(wptr, rptr, (uint64_t)pixel_size * width); } } } mipmap_offset = image_total; } } vmaUnmapMemory(allocator, p_allocation); return image_data; } Vector<uint8_t> RenderingDeviceVulkan::texture_get_data(RID p_texture, uint32_t p_layer) { _THREAD_SAFE_METHOD_ Texture *tex = texture_owner.get_or_null(p_texture); ERR_FAIL_NULL_V(tex, Vector<uint8_t>()); ERR_FAIL_COND_V_MSG(tex->bound, Vector<uint8_t>(), "Texture can't be retrieved while a draw list that uses it as part of a framebuffer is being created. Ensure the draw list is finalized (and that the color/depth texture using it is not set to `RenderingDevice.FINAL_ACTION_CONTINUE`) to retrieve this texture."); ERR_FAIL_COND_V_MSG(!(tex->usage_flags & TEXTURE_USAGE_CAN_COPY_FROM_BIT), Vector<uint8_t>(), "Texture requires the `RenderingDevice.TEXTURE_USAGE_CAN_COPY_FROM_BIT` to be set to be retrieved."); uint32_t layer_count = tex->layers; if (tex->type == TEXTURE_TYPE_CUBE || tex->type == TEXTURE_TYPE_CUBE_ARRAY) { layer_count *= 6; } ERR_FAIL_COND_V(p_layer >= layer_count, Vector<uint8_t>()); if (tex->usage_flags & TEXTURE_USAGE_CPU_READ_BIT) { // Does not need anything fancy, map and read. return _texture_get_data_from_image(tex, tex->image, tex->allocation, p_layer); } else { // Compute total image size. uint32_t width, height, depth; uint32_t buffer_size = get_image_format_required_size(tex->format, tex->width, tex->height, tex->depth, tex->mipmaps, &width, &height, &depth); // Allocate buffer. VkCommandBuffer command_buffer = frames[frame].draw_command_buffer; // Makes more sense to retrieve. Buffer tmp_buffer; _buffer_allocate(&tmp_buffer, buffer_size, VK_BUFFER_USAGE_TRANSFER_DST_BIT, VMA_MEMORY_USAGE_AUTO_PREFER_HOST, VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT); { // Source image barrier. VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = 0; image_memory_barrier.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT; image_memory_barrier.oldLayout = tex->layout; image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = tex->image; image_memory_barrier.subresourceRange.aspectMask = tex->barrier_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = 0; image_memory_barrier.subresourceRange.levelCount = tex->mipmaps; image_memory_barrier.subresourceRange.baseArrayLayer = p_layer; image_memory_barrier.subresourceRange.layerCount = 1; vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } uint32_t computed_w = tex->width; uint32_t computed_h = tex->height; uint32_t computed_d = tex->depth; uint32_t prev_size = 0; uint32_t offset = 0; for (uint32_t i = 0; i < tex->mipmaps; i++) { VkBufferImageCopy buffer_image_copy; uint32_t image_size = get_image_format_required_size(tex->format, tex->width, tex->height, tex->depth, i + 1); uint32_t size = image_size - prev_size; prev_size = image_size; buffer_image_copy.bufferOffset = offset; buffer_image_copy.bufferImageHeight = 0; buffer_image_copy.bufferRowLength = 0; buffer_image_copy.imageSubresource.aspectMask = tex->read_aspect_mask; buffer_image_copy.imageSubresource.baseArrayLayer = p_layer; buffer_image_copy.imageSubresource.layerCount = 1; buffer_image_copy.imageSubresource.mipLevel = i; buffer_image_copy.imageOffset.x = 0; buffer_image_copy.imageOffset.y = 0; buffer_image_copy.imageOffset.z = 0; buffer_image_copy.imageExtent.width = computed_w; buffer_image_copy.imageExtent.height = computed_h; buffer_image_copy.imageExtent.depth = computed_d; vkCmdCopyImageToBuffer(command_buffer, tex->image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, tmp_buffer.buffer, 1, &buffer_image_copy); computed_w = MAX(1u, computed_w >> 1); computed_h = MAX(1u, computed_h >> 1); computed_d = MAX(1u, computed_d >> 1); offset += size; } { // Restore src. VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT; image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT; if (tex->usage_flags & TEXTURE_USAGE_STORAGE_BIT) { image_memory_barrier.dstAccessMask |= VK_ACCESS_SHADER_WRITE_BIT; } image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL; image_memory_barrier.newLayout = tex->layout; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = tex->image; image_memory_barrier.subresourceRange.aspectMask = tex->barrier_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = 0; image_memory_barrier.subresourceRange.levelCount = tex->mipmaps; image_memory_barrier.subresourceRange.baseArrayLayer = p_layer; image_memory_barrier.subresourceRange.layerCount = 1; vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } _flush(true); void *buffer_mem; VkResult vkerr = vmaMapMemory(allocator, tmp_buffer.allocation, &buffer_mem); ERR_FAIL_COND_V_MSG(vkerr, Vector<uint8_t>(), "vmaMapMemory failed with error " + itos(vkerr) + "."); Vector<uint8_t> buffer_data; { buffer_data.resize(buffer_size); uint8_t *w = buffer_data.ptrw(); memcpy(w, buffer_mem, buffer_size); } vmaUnmapMemory(allocator, tmp_buffer.allocation); _buffer_free(&tmp_buffer); return buffer_data; } } bool RenderingDeviceVulkan::texture_is_shared(RID p_texture) { _THREAD_SAFE_METHOD_ Texture *tex = texture_owner.get_or_null(p_texture); ERR_FAIL_NULL_V(tex, false); return tex->owner.is_valid(); } bool RenderingDeviceVulkan::texture_is_valid(RID p_texture) { return texture_owner.owns(p_texture); } RD::TextureFormat RenderingDeviceVulkan::texture_get_format(RID p_texture) { _THREAD_SAFE_METHOD_ Texture *tex = texture_owner.get_or_null(p_texture); ERR_FAIL_NULL_V(tex, TextureFormat()); TextureFormat tf; tf.format = tex->format; tf.width = tex->width; tf.height = tex->height; tf.depth = tex->depth; tf.array_layers = tex->layers; tf.mipmaps = tex->mipmaps; tf.texture_type = tex->type; tf.samples = tex->samples; tf.usage_bits = tex->usage_flags; tf.shareable_formats = tex->allowed_shared_formats; tf.is_resolve_buffer = tex->is_resolve_buffer; return tf; } Size2i RenderingDeviceVulkan::texture_size(RID p_texture) { _THREAD_SAFE_METHOD_ Texture *tex = texture_owner.get_or_null(p_texture); ERR_FAIL_NULL_V(tex, Size2i()); return Size2i(tex->width, tex->height); } uint64_t RenderingDeviceVulkan::texture_get_native_handle(RID p_texture) { _THREAD_SAFE_METHOD_ Texture *tex = texture_owner.get_or_null(p_texture); ERR_FAIL_NULL_V(tex, 0); return (uint64_t)tex->image; } Error RenderingDeviceVulkan::texture_copy(RID p_from_texture, RID p_to_texture, const Vector3 &p_from, const Vector3 &p_to, const Vector3 &p_size, uint32_t p_src_mipmap, uint32_t p_dst_mipmap, uint32_t p_src_layer, uint32_t p_dst_layer, BitField<BarrierMask> p_post_barrier) { _THREAD_SAFE_METHOD_ Texture *src_tex = texture_owner.get_or_null(p_from_texture); ERR_FAIL_NULL_V(src_tex, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V_MSG(src_tex->bound, ERR_INVALID_PARAMETER, "Source texture can't be copied while a draw list that uses it as part of a framebuffer is being created. Ensure the draw list is finalized (and that the color/depth texture using it is not set to `RenderingDevice.FINAL_ACTION_CONTINUE`) to copy this texture."); ERR_FAIL_COND_V_MSG(!(src_tex->usage_flags & TEXTURE_USAGE_CAN_COPY_FROM_BIT), ERR_INVALID_PARAMETER, "Source texture requires the `RenderingDevice.TEXTURE_USAGE_CAN_COPY_FROM_BIT` to be set to be retrieved."); uint32_t src_layer_count = src_tex->layers; uint32_t src_width, src_height, src_depth; get_image_format_required_size(src_tex->format, src_tex->width, src_tex->height, src_tex->depth, p_src_mipmap + 1, &src_width, &src_height, &src_depth); if (src_tex->type == TEXTURE_TYPE_CUBE || src_tex->type == TEXTURE_TYPE_CUBE_ARRAY) { src_layer_count *= 6; } ERR_FAIL_COND_V(p_from.x < 0 || p_from.x + p_size.x > src_width, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V(p_from.y < 0 || p_from.y + p_size.y > src_height, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V(p_from.z < 0 || p_from.z + p_size.z > src_depth, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V(p_src_mipmap >= src_tex->mipmaps, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V(p_src_layer >= src_layer_count, ERR_INVALID_PARAMETER); Texture *dst_tex = texture_owner.get_or_null(p_to_texture); ERR_FAIL_NULL_V(dst_tex, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V_MSG(dst_tex->bound, ERR_INVALID_PARAMETER, "Destination texture can't be copied while a draw list that uses it as part of a framebuffer is being created. Ensure the draw list is finalized (and that the color/depth texture using it is not set to `RenderingDevice.FINAL_ACTION_CONTINUE`) to copy this texture."); ERR_FAIL_COND_V_MSG(!(dst_tex->usage_flags & TEXTURE_USAGE_CAN_COPY_TO_BIT), ERR_INVALID_PARAMETER, "Destination texture requires the `RenderingDevice.TEXTURE_USAGE_CAN_COPY_TO_BIT` to be set to be retrieved."); uint32_t dst_layer_count = dst_tex->layers; uint32_t dst_width, dst_height, dst_depth; get_image_format_required_size(dst_tex->format, dst_tex->width, dst_tex->height, dst_tex->depth, p_dst_mipmap + 1, &dst_width, &dst_height, &dst_depth); if (dst_tex->type == TEXTURE_TYPE_CUBE || dst_tex->type == TEXTURE_TYPE_CUBE_ARRAY) { dst_layer_count *= 6; } ERR_FAIL_COND_V(p_to.x < 0 || p_to.x + p_size.x > dst_width, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V(p_to.y < 0 || p_to.y + p_size.y > dst_height, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V(p_to.z < 0 || p_to.z + p_size.z > dst_depth, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V(p_dst_mipmap >= dst_tex->mipmaps, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V(p_dst_layer >= dst_layer_count, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V_MSG(src_tex->read_aspect_mask != dst_tex->read_aspect_mask, ERR_INVALID_PARAMETER, "Source and destination texture must be of the same type (color or depth)."); VkCommandBuffer command_buffer = frames[frame].draw_command_buffer; { // PRE Copy the image. { // Source. VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = 0; image_memory_barrier.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT; image_memory_barrier.oldLayout = src_tex->layout; image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = src_tex->image; image_memory_barrier.subresourceRange.aspectMask = src_tex->barrier_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = p_src_mipmap; image_memory_barrier.subresourceRange.levelCount = 1; image_memory_barrier.subresourceRange.baseArrayLayer = p_src_layer; image_memory_barrier.subresourceRange.layerCount = 1; vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } { // Dest. VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = 0; image_memory_barrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; image_memory_barrier.oldLayout = dst_tex->layout; image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = dst_tex->image; image_memory_barrier.subresourceRange.aspectMask = dst_tex->read_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = p_dst_mipmap; image_memory_barrier.subresourceRange.levelCount = 1; image_memory_barrier.subresourceRange.baseArrayLayer = p_dst_layer; image_memory_barrier.subresourceRange.layerCount = 1; vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } // COPY. { VkImageCopy image_copy_region; image_copy_region.srcSubresource.aspectMask = src_tex->read_aspect_mask; image_copy_region.srcSubresource.baseArrayLayer = p_src_layer; image_copy_region.srcSubresource.layerCount = 1; image_copy_region.srcSubresource.mipLevel = p_src_mipmap; image_copy_region.srcOffset.x = p_from.x; image_copy_region.srcOffset.y = p_from.y; image_copy_region.srcOffset.z = p_from.z; image_copy_region.dstSubresource.aspectMask = dst_tex->read_aspect_mask; image_copy_region.dstSubresource.baseArrayLayer = p_dst_layer; image_copy_region.dstSubresource.layerCount = 1; image_copy_region.dstSubresource.mipLevel = p_dst_mipmap; image_copy_region.dstOffset.x = p_to.x; image_copy_region.dstOffset.y = p_to.y; image_copy_region.dstOffset.z = p_to.z; image_copy_region.extent.width = p_size.x; image_copy_region.extent.height = p_size.y; image_copy_region.extent.depth = p_size.z; vkCmdCopyImage(command_buffer, src_tex->image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_tex->image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &image_copy_region); } // RESTORE LAYOUT for SRC and DST. uint32_t barrier_flags = 0; uint32_t access_flags = 0; if (p_post_barrier.has_flag(BARRIER_MASK_COMPUTE)) { barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_VERTEX)) { barrier_flags |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_FRAGMENT)) { barrier_flags |= VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_TRANSFER)) { barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT; } if (barrier_flags == 0) { barrier_flags = VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT; } { // Restore src. VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT; image_memory_barrier.dstAccessMask = access_flags; image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL; image_memory_barrier.newLayout = src_tex->layout; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = src_tex->image; image_memory_barrier.subresourceRange.aspectMask = src_tex->barrier_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = p_src_mipmap; image_memory_barrier.subresourceRange.levelCount = src_tex->mipmaps; image_memory_barrier.subresourceRange.baseArrayLayer = p_src_layer; image_memory_barrier.subresourceRange.layerCount = 1; vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, barrier_flags, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } { // Make dst readable. VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; image_memory_barrier.dstAccessMask = access_flags; image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL; image_memory_barrier.newLayout = dst_tex->layout; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = dst_tex->image; image_memory_barrier.subresourceRange.aspectMask = dst_tex->read_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = p_src_mipmap; image_memory_barrier.subresourceRange.levelCount = 1; image_memory_barrier.subresourceRange.baseArrayLayer = p_src_layer; image_memory_barrier.subresourceRange.layerCount = 1; vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, barrier_flags, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } } if (dst_tex->used_in_frame != frames_drawn) { dst_tex->used_in_raster = false; dst_tex->used_in_compute = false; dst_tex->used_in_frame = frames_drawn; } dst_tex->used_in_transfer = true; return OK; } Error RenderingDeviceVulkan::texture_resolve_multisample(RID p_from_texture, RID p_to_texture, BitField<BarrierMask> p_post_barrier) { _THREAD_SAFE_METHOD_ Texture *src_tex = texture_owner.get_or_null(p_from_texture); ERR_FAIL_NULL_V(src_tex, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V_MSG(src_tex->bound, ERR_INVALID_PARAMETER, "Source texture can't be copied while a draw list that uses it as part of a framebuffer is being created. Ensure the draw list is finalized (and that the color/depth texture using it is not set to `RenderingDevice.FINAL_ACTION_CONTINUE`) to copy this texture."); ERR_FAIL_COND_V_MSG(!(src_tex->usage_flags & TEXTURE_USAGE_CAN_COPY_FROM_BIT), ERR_INVALID_PARAMETER, "Source texture requires the `RenderingDevice.TEXTURE_USAGE_CAN_COPY_FROM_BIT` to be set to be retrieved."); ERR_FAIL_COND_V_MSG(src_tex->type != TEXTURE_TYPE_2D, ERR_INVALID_PARAMETER, "Source texture must be 2D (or a slice of a 3D/Cube texture)"); ERR_FAIL_COND_V_MSG(src_tex->samples == TEXTURE_SAMPLES_1, ERR_INVALID_PARAMETER, "Source texture must be multisampled."); Texture *dst_tex = texture_owner.get_or_null(p_to_texture); ERR_FAIL_NULL_V(dst_tex, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V_MSG(dst_tex->bound, ERR_INVALID_PARAMETER, "Destination texture can't be copied while a draw list that uses it as part of a framebuffer is being created. Ensure the draw list is finalized (and that the color/depth texture using it is not set to `RenderingDevice.FINAL_ACTION_CONTINUE`) to copy this texture."); ERR_FAIL_COND_V_MSG(!(dst_tex->usage_flags & TEXTURE_USAGE_CAN_COPY_TO_BIT), ERR_INVALID_PARAMETER, "Destination texture requires the `RenderingDevice.TEXTURE_USAGE_CAN_COPY_TO_BIT` to be set to be retrieved."); ERR_FAIL_COND_V_MSG(dst_tex->type != TEXTURE_TYPE_2D, ERR_INVALID_PARAMETER, "Destination texture must be 2D (or a slice of a 3D/Cube texture)."); ERR_FAIL_COND_V_MSG(dst_tex->samples != TEXTURE_SAMPLES_1, ERR_INVALID_PARAMETER, "Destination texture must not be multisampled."); ERR_FAIL_COND_V_MSG(src_tex->format != dst_tex->format, ERR_INVALID_PARAMETER, "Source and Destination textures must be the same format."); ERR_FAIL_COND_V_MSG(src_tex->width != dst_tex->width && src_tex->height != dst_tex->height && src_tex->depth != dst_tex->depth, ERR_INVALID_PARAMETER, "Source and Destination textures must have the same dimensions."); ERR_FAIL_COND_V_MSG(src_tex->read_aspect_mask != dst_tex->read_aspect_mask, ERR_INVALID_PARAMETER, "Source and destination texture must be of the same type (color or depth)."); VkCommandBuffer command_buffer = frames[frame].draw_command_buffer; { // PRE Copy the image. { // Source. VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = 0; image_memory_barrier.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT; image_memory_barrier.oldLayout = src_tex->layout; image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = src_tex->image; image_memory_barrier.subresourceRange.aspectMask = src_tex->barrier_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = src_tex->base_mipmap; image_memory_barrier.subresourceRange.levelCount = 1; image_memory_barrier.subresourceRange.baseArrayLayer = src_tex->base_layer; image_memory_barrier.subresourceRange.layerCount = 1; vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } { // Dest. VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = 0; image_memory_barrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; image_memory_barrier.oldLayout = dst_tex->layout; image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = dst_tex->image; image_memory_barrier.subresourceRange.aspectMask = dst_tex->read_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = dst_tex->base_mipmap; image_memory_barrier.subresourceRange.levelCount = 1; image_memory_barrier.subresourceRange.baseArrayLayer = dst_tex->base_layer; image_memory_barrier.subresourceRange.layerCount = 1; vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } // COPY. { VkImageResolve image_copy_region; image_copy_region.srcSubresource.aspectMask = src_tex->read_aspect_mask; image_copy_region.srcSubresource.baseArrayLayer = src_tex->base_layer; image_copy_region.srcSubresource.layerCount = 1; image_copy_region.srcSubresource.mipLevel = src_tex->base_mipmap; image_copy_region.srcOffset.x = 0; image_copy_region.srcOffset.y = 0; image_copy_region.srcOffset.z = 0; image_copy_region.dstSubresource.aspectMask = dst_tex->read_aspect_mask; image_copy_region.dstSubresource.baseArrayLayer = dst_tex->base_layer; image_copy_region.dstSubresource.layerCount = 1; image_copy_region.dstSubresource.mipLevel = dst_tex->base_mipmap; image_copy_region.dstOffset.x = 0; image_copy_region.dstOffset.y = 0; image_copy_region.dstOffset.z = 0; image_copy_region.extent.width = src_tex->width; image_copy_region.extent.height = src_tex->height; image_copy_region.extent.depth = src_tex->depth; vkCmdResolveImage(command_buffer, src_tex->image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_tex->image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &image_copy_region); } // RESTORE LAYOUT for SRC and DST. uint32_t barrier_flags = 0; uint32_t access_flags = 0; if (p_post_barrier.has_flag(BARRIER_MASK_COMPUTE)) { barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_VERTEX)) { barrier_flags |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_FRAGMENT)) { barrier_flags |= VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_TRANSFER)) { barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT; } if (barrier_flags == 0) { barrier_flags = VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT; } { // Restore src. VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT; image_memory_barrier.dstAccessMask = access_flags; image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL; image_memory_barrier.newLayout = src_tex->layout; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = src_tex->image; image_memory_barrier.subresourceRange.aspectMask = src_tex->barrier_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = src_tex->base_mipmap; image_memory_barrier.subresourceRange.levelCount = 1; image_memory_barrier.subresourceRange.baseArrayLayer = src_tex->base_layer; image_memory_barrier.subresourceRange.layerCount = 1; vkCmdPipelineBarrier(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, barrier_flags, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } { // Make dst readable. VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; image_memory_barrier.dstAccessMask = access_flags; image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL; image_memory_barrier.newLayout = dst_tex->layout; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = dst_tex->image; image_memory_barrier.subresourceRange.aspectMask = dst_tex->read_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = dst_tex->base_mipmap; image_memory_barrier.subresourceRange.levelCount = 1; image_memory_barrier.subresourceRange.baseArrayLayer = dst_tex->base_layer; image_memory_barrier.subresourceRange.layerCount = 1; vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, barrier_flags, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } } return OK; } Error RenderingDeviceVulkan::texture_clear(RID p_texture, const Color &p_color, uint32_t p_base_mipmap, uint32_t p_mipmaps, uint32_t p_base_layer, uint32_t p_layers, BitField<BarrierMask> p_post_barrier) { _THREAD_SAFE_METHOD_ Texture *src_tex = texture_owner.get_or_null(p_texture); ERR_FAIL_NULL_V(src_tex, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V_MSG(src_tex->bound, ERR_INVALID_PARAMETER, "Source texture can't be cleared while a draw list that uses it as part of a framebuffer is being created. Ensure the draw list is finalized (and that the color/depth texture using it is not set to `RenderingDevice.FINAL_ACTION_CONTINUE`) to clear this texture."); ERR_FAIL_COND_V(p_layers == 0, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V(p_mipmaps == 0, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V_MSG(!(src_tex->usage_flags & TEXTURE_USAGE_CAN_COPY_TO_BIT), ERR_INVALID_PARAMETER, "Source texture requires the `RenderingDevice.TEXTURE_USAGE_CAN_COPY_TO_BIT` to be set to be cleared."); uint32_t src_layer_count = src_tex->layers; if (src_tex->type == TEXTURE_TYPE_CUBE || src_tex->type == TEXTURE_TYPE_CUBE_ARRAY) { src_layer_count *= 6; } ERR_FAIL_COND_V(p_base_mipmap + p_mipmaps > src_tex->mipmaps, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V(p_base_layer + p_layers > src_layer_count, ERR_INVALID_PARAMETER); VkCommandBuffer command_buffer = frames[frame].draw_command_buffer; VkImageLayout clear_layout = (src_tex->layout == VK_IMAGE_LAYOUT_GENERAL) ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL; // NOTE: Perhaps the valid stages/accesses for a given owner should be a property of the owner. (Here and places like _get_buffer_from_owner.) const VkPipelineStageFlags valid_texture_stages = VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; constexpr VkAccessFlags read_access = VK_ACCESS_SHADER_READ_BIT; constexpr VkAccessFlags read_write_access = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; const VkAccessFlags valid_texture_access = (src_tex->usage_flags & TEXTURE_USAGE_STORAGE_BIT) ? read_write_access : read_access; { // Barrier from previous access with optional layout change (see clear_layout logic above). VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = valid_texture_access; image_memory_barrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; image_memory_barrier.oldLayout = src_tex->layout; image_memory_barrier.newLayout = clear_layout; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = src_tex->image; image_memory_barrier.subresourceRange.aspectMask = src_tex->read_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = src_tex->base_mipmap + p_base_mipmap; image_memory_barrier.subresourceRange.levelCount = p_mipmaps; image_memory_barrier.subresourceRange.baseArrayLayer = src_tex->base_layer + p_base_layer; image_memory_barrier.subresourceRange.layerCount = p_layers; vkCmdPipelineBarrier(command_buffer, valid_texture_stages, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } VkClearColorValue clear_color; clear_color.float32[0] = p_color.r; clear_color.float32[1] = p_color.g; clear_color.float32[2] = p_color.b; clear_color.float32[3] = p_color.a; VkImageSubresourceRange range; range.aspectMask = src_tex->read_aspect_mask; range.baseArrayLayer = src_tex->base_layer + p_base_layer; range.layerCount = p_layers; range.baseMipLevel = src_tex->base_mipmap + p_base_mipmap; range.levelCount = p_mipmaps; vkCmdClearColorImage(command_buffer, src_tex->image, clear_layout, &clear_color, 1, &range); { // Barrier to post clear accesses (changing back the layout if needed). uint32_t barrier_flags = 0; uint32_t access_flags = 0; if (p_post_barrier.has_flag(BARRIER_MASK_COMPUTE)) { barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_VERTEX)) { barrier_flags |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_FRAGMENT)) { barrier_flags |= VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_TRANSFER)) { barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT; } if (barrier_flags == 0) { barrier_flags = VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT; } VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; image_memory_barrier.dstAccessMask = access_flags; image_memory_barrier.oldLayout = clear_layout; image_memory_barrier.newLayout = src_tex->layout; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = src_tex->image; image_memory_barrier.subresourceRange.aspectMask = src_tex->read_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = src_tex->base_mipmap + p_base_mipmap; image_memory_barrier.subresourceRange.levelCount = p_mipmaps; image_memory_barrier.subresourceRange.baseArrayLayer = src_tex->base_layer + p_base_layer; image_memory_barrier.subresourceRange.layerCount = p_layers; vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, barrier_flags, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } if (src_tex->used_in_frame != frames_drawn) { src_tex->used_in_raster = false; src_tex->used_in_compute = false; src_tex->used_in_frame = frames_drawn; } src_tex->used_in_transfer = true; return OK; } bool RenderingDeviceVulkan::texture_is_format_supported_for_usage(DataFormat p_format, BitField<RenderingDevice::TextureUsageBits> p_usage) const { ERR_FAIL_INDEX_V(p_format, DATA_FORMAT_MAX, false); _THREAD_SAFE_METHOD_ // Validate that this image is supported for the intended use. VkFormatProperties properties; vkGetPhysicalDeviceFormatProperties(context->get_physical_device(), vulkan_formats[p_format], &properties); VkFormatFeatureFlags flags; if (p_usage.has_flag(TEXTURE_USAGE_CPU_READ_BIT)) { flags = properties.linearTilingFeatures; } else { flags = properties.optimalTilingFeatures; } if (p_usage.has_flag(TEXTURE_USAGE_SAMPLING_BIT) && !(flags & VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT)) { return false; } if (p_usage.has_flag(TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) && !(flags & VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BIT)) { return false; } if (p_usage.has_flag(TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) && !(flags & VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT)) { return false; } if (p_usage.has_flag(TEXTURE_USAGE_STORAGE_BIT) && !(flags & VK_FORMAT_FEATURE_STORAGE_IMAGE_BIT)) { return false; } if (p_usage.has_flag(TEXTURE_USAGE_STORAGE_ATOMIC_BIT) && !(flags & VK_FORMAT_FEATURE_STORAGE_IMAGE_ATOMIC_BIT)) { return false; } // Validation via VK_FORMAT_FEATURE_FRAGMENT_SHADING_RATE_ATTACHMENT_BIT_KHR fails if VRS attachment is not supported. if (p_usage.has_flag(TEXTURE_USAGE_VRS_ATTACHMENT_BIT) && p_format != DATA_FORMAT_R8_UINT) { return false; } return true; } /********************/ /**** ATTACHMENT ****/ /********************/ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentFormat> &p_attachments, const Vector<FramebufferPass> &p_passes, InitialAction p_initial_action, FinalAction p_final_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, uint32_t p_view_count, Vector<TextureSamples> *r_samples) { // Set up dependencies from/to external equivalent to the default (implicit) one, and then amend them. const VkPipelineStageFlags default_access_mask = VK_ACCESS_INPUT_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | // From Section 7.1 of Vulkan API Spec v1.1.148. VK_ACCESS_FRAGMENT_SHADING_RATE_ATTACHMENT_READ_BIT_KHR; VkPipelineStageFlags reading_stages = VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT; VkSubpassDependency2KHR dependencies[2] = { { VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_2_KHR, nullptr, VK_SUBPASS_EXTERNAL, 0, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, 0, default_access_mask, 0, 0 }, { VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_2_KHR, nullptr, 0, VK_SUBPASS_EXTERNAL, VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, default_access_mask, 0, 0, 0 } }; VkSubpassDependency2KHR &dependency_from_external = dependencies[0]; VkSubpassDependency2KHR &dependency_to_external = dependencies[1]; LocalVector<int32_t> attachment_last_pass; attachment_last_pass.resize(p_attachments.size()); if (p_view_count > 1) { const VulkanContext::MultiviewCapabilities capabilities = context->get_multiview_capabilities(); // This only works with multiview! ERR_FAIL_COND_V_MSG(!capabilities.is_supported, VK_NULL_HANDLE, "Multiview not supported"); // Make sure we limit this to the number of views we support. ERR_FAIL_COND_V_MSG(p_view_count > capabilities.max_view_count, VK_NULL_HANDLE, "Hardware does not support requested number of views for Multiview render pass"); } // These are only used if we use multiview but we need to define them in scope. const uint32_t view_mask = (1 << p_view_count) - 1; const uint32_t correlation_mask = (1 << p_view_count) - 1; Vector<VkAttachmentDescription2KHR> attachments; Vector<int> attachment_remap; for (int i = 0; i < p_attachments.size(); i++) { if (p_attachments[i].usage_flags == AttachmentFormat::UNUSED_ATTACHMENT) { attachment_remap.push_back(VK_ATTACHMENT_UNUSED); continue; } ERR_FAIL_INDEX_V(p_attachments[i].format, DATA_FORMAT_MAX, VK_NULL_HANDLE); ERR_FAIL_INDEX_V(p_attachments[i].samples, TEXTURE_SAMPLES_MAX, VK_NULL_HANDLE); ERR_FAIL_COND_V_MSG(!(p_attachments[i].usage_flags & (TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | TEXTURE_USAGE_INPUT_ATTACHMENT_BIT | TEXTURE_USAGE_VRS_ATTACHMENT_BIT)), VK_NULL_HANDLE, "Texture format for index (" + itos(i) + ") requires an attachment (color, depth-stencil, input or VRS) bit set."); VkAttachmentDescription2KHR description = {}; description.sType = VK_STRUCTURE_TYPE_ATTACHMENT_DESCRIPTION_2_KHR; description.pNext = nullptr; description.flags = 0; description.format = vulkan_formats[p_attachments[i].format]; description.samples = _ensure_supported_sample_count(p_attachments[i].samples); bool is_sampled = p_attachments[i].usage_flags & TEXTURE_USAGE_SAMPLING_BIT; bool is_storage = p_attachments[i].usage_flags & TEXTURE_USAGE_STORAGE_BIT; bool is_depth = p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT; // We can setup a framebuffer where we write to our VRS texture to set it up. // We make the assumption here that if our texture is actually used as our VRS attachment. // It is used as such for each subpass. This is fairly certain seeing the restrictions on subpasses. bool is_vrs = p_attachments[i].usage_flags & TEXTURE_USAGE_VRS_ATTACHMENT_BIT && i == p_passes[0].vrs_attachment; if (is_vrs) { // For VRS we only read, there is no writing to this texture. description.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD; description.initialLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD; } else { // For each UNDEFINED, assume the prior use was a *read*, as we'd be discarding the output of a write. // Also, each UNDEFINED will do an immediate layout transition (write), s.t. we must ensure execution synchronization vs // the read. If this is a performance issue, one could track the actual last accessor of each resource, adding only that // stage. switch (is_depth ? p_initial_depth_action : p_initial_action) { case INITIAL_ACTION_CLEAR_REGION: case INITIAL_ACTION_CLEAR: { if (p_attachments[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { description.loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR; description.initialLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL); description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; } else if (p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { description.loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR; description.initialLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL); description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_CLEAR; dependency_from_external.srcStageMask |= reading_stages; } else { description.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; description.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; // Don't care what is there. dependency_from_external.srcStageMask |= reading_stages; } } break; case INITIAL_ACTION_KEEP: { if (p_attachments[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { description.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD; description.initialLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL); description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; } else if (p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { description.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD; description.initialLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL); description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD; dependency_from_external.srcStageMask |= reading_stages; } else { description.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; description.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; // Don't care what is there. dependency_from_external.srcStageMask |= reading_stages; } } break; case INITIAL_ACTION_DROP: { if (p_attachments[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { description.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; description.initialLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL); description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; } else if (p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { description.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; description.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; // Don't care what is there. description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; dependency_from_external.srcStageMask |= reading_stages; } else { description.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; description.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; // Don't care what is there. dependency_from_external.srcStageMask |= reading_stages; } } break; case INITIAL_ACTION_CLEAR_REGION_CONTINUE: case INITIAL_ACTION_CONTINUE: { if (p_attachments[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { description.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD; description.initialLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; } else if (p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { description.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD; description.initialLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD; } else { description.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; description.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; // Don't care what is there. dependency_from_external.srcStageMask |= reading_stages; } } break; default: { ERR_FAIL_V(VK_NULL_HANDLE); // Should never reach here. } } } bool used_last = false; { int last_pass = p_passes.size() - 1; if (is_depth) { // Likely missing depth resolve? if (p_passes[last_pass].depth_attachment == i) { used_last = true; } } else if (is_vrs) { if (p_passes[last_pass].vrs_attachment == i) { used_last = true; } } else { if (p_passes[last_pass].resolve_attachments.size()) { // If using resolve attachments, check resolve attachments. for (int j = 0; j < p_passes[last_pass].resolve_attachments.size(); j++) { if (p_passes[last_pass].resolve_attachments[j] == i) { used_last = true; break; } } } if (!used_last) { for (int j = 0; j < p_passes[last_pass].color_attachments.size(); j++) { if (p_passes[last_pass].color_attachments[j] == i) { used_last = true; break; } } } } if (!used_last) { for (int j = 0; j < p_passes[last_pass].preserve_attachments.size(); j++) { if (p_passes[last_pass].preserve_attachments[j] == i) { used_last = true; break; } } } } FinalAction final_action = p_final_action; FinalAction final_depth_action = p_final_depth_action; if (!used_last) { if (is_depth) { final_depth_action = FINAL_ACTION_DISCARD; } else { final_action = FINAL_ACTION_DISCARD; } } if (is_vrs) { // We don't change our VRS texture during this process. description.storeOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.finalLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; // TODO: Do we need to update our external dependency? // update_external_dependency_for_store(dependency_to_external, is_sampled, is_storage, false); } else { switch (is_depth ? final_depth_action : final_action) { case FINAL_ACTION_READ: { if (p_attachments[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { description.storeOp = VK_ATTACHMENT_STORE_OP_STORE; description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.finalLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL); update_external_dependency_for_store(dependency_to_external, is_sampled, is_storage, false); } else if (p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { description.storeOp = VK_ATTACHMENT_STORE_OP_STORE; description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE; description.finalLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL); update_external_dependency_for_store(dependency_to_external, is_sampled, is_storage, true); } else { description.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; description.finalLayout = VK_IMAGE_LAYOUT_UNDEFINED; // Don't care what is there. // TODO: What does this mean about the next usage (and thus appropriate dependency masks. } } break; case FINAL_ACTION_DISCARD: { if (p_attachments[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { description.storeOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.finalLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL); } else if (p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { description.storeOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.finalLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL); } else { description.storeOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.finalLayout = VK_IMAGE_LAYOUT_UNDEFINED; // Don't care what is there. } } break; case FINAL_ACTION_CONTINUE: { if (p_attachments[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { description.storeOp = VK_ATTACHMENT_STORE_OP_STORE; description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.finalLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; } else if (p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { description.storeOp = VK_ATTACHMENT_STORE_OP_STORE; description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE; description.finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; } else { description.storeOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.finalLayout = VK_IMAGE_LAYOUT_UNDEFINED; // Don't care what is there. } } break; default: { ERR_FAIL_V(VK_NULL_HANDLE); // Should never reach here. } } } attachment_last_pass[i] = -1; attachment_remap.push_back(attachments.size()); attachments.push_back(description); } LocalVector<VkSubpassDescription2KHR> subpasses; LocalVector<LocalVector<VkAttachmentReference2KHR>> color_reference_array; LocalVector<LocalVector<VkAttachmentReference2KHR>> input_reference_array; LocalVector<LocalVector<VkAttachmentReference2KHR>> resolve_reference_array; LocalVector<LocalVector<uint32_t>> preserve_reference_array; LocalVector<VkAttachmentReference2KHR> depth_reference_array; LocalVector<VkAttachmentReference2KHR> vrs_reference_array; LocalVector<VkFragmentShadingRateAttachmentInfoKHR> vrs_attachment_info_array; subpasses.resize(p_passes.size()); color_reference_array.resize(p_passes.size()); input_reference_array.resize(p_passes.size()); resolve_reference_array.resize(p_passes.size()); preserve_reference_array.resize(p_passes.size()); depth_reference_array.resize(p_passes.size()); vrs_reference_array.resize(p_passes.size()); vrs_attachment_info_array.resize(p_passes.size()); LocalVector<VkSubpassDependency2KHR> subpass_dependencies; for (int i = 0; i < p_passes.size(); i++) { const FramebufferPass *pass = &p_passes[i]; LocalVector<VkAttachmentReference2KHR> &color_references = color_reference_array[i]; TextureSamples texture_samples = TEXTURE_SAMPLES_1; bool is_multisample_first = true; void *subpass_nextptr = nullptr; for (int j = 0; j < pass->color_attachments.size(); j++) { int32_t attachment = pass->color_attachments[j]; VkAttachmentReference2KHR reference; reference.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2_KHR; reference.pNext = nullptr; if (attachment == FramebufferPass::ATTACHMENT_UNUSED) { reference.attachment = VK_ATTACHMENT_UNUSED; reference.layout = VK_IMAGE_LAYOUT_UNDEFINED; } else { ERR_FAIL_INDEX_V_MSG(attachment, p_attachments.size(), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), color attachment (" + itos(j) + ")."); ERR_FAIL_COND_V_MSG(!(p_attachments[attachment].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it's marked as depth, but it's not usable as color attachment."); ERR_FAIL_COND_V_MSG(attachment_last_pass[attachment] == i, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it already was used for something else before in this pass."); if (is_multisample_first) { texture_samples = p_attachments[attachment].samples; is_multisample_first = false; } else { ERR_FAIL_COND_V_MSG(texture_samples != p_attachments[attachment].samples, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), if an attachment is marked as multisample, all of them should be multisample and use the same number of samples."); } reference.attachment = attachment_remap[attachment]; reference.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; attachment_last_pass[attachment] = i; } reference.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; color_references.push_back(reference); } LocalVector<VkAttachmentReference2KHR> &input_references = input_reference_array[i]; for (int j = 0; j < pass->input_attachments.size(); j++) { int32_t attachment = pass->input_attachments[j]; VkAttachmentReference2KHR reference; reference.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2_KHR; reference.pNext = nullptr; if (attachment == FramebufferPass::ATTACHMENT_UNUSED) { reference.attachment = VK_ATTACHMENT_UNUSED; reference.layout = VK_IMAGE_LAYOUT_UNDEFINED; } else { ERR_FAIL_INDEX_V_MSG(attachment, p_attachments.size(), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), input attachment (" + itos(j) + ")."); ERR_FAIL_COND_V_MSG(!(p_attachments[attachment].usage_flags & TEXTURE_USAGE_INPUT_ATTACHMENT_BIT), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it isn't marked as an input texture."); ERR_FAIL_COND_V_MSG(attachment_last_pass[attachment] == i, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it already was used for something else before in this pass."); reference.attachment = attachment_remap[attachment]; reference.layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; attachment_last_pass[attachment] = i; } reference.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; input_references.push_back(reference); } LocalVector<VkAttachmentReference2KHR> &resolve_references = resolve_reference_array[i]; if (pass->resolve_attachments.size() > 0) { ERR_FAIL_COND_V_MSG(pass->resolve_attachments.size() != pass->color_attachments.size(), VK_NULL_HANDLE, "The amount of resolve attachments (" + itos(pass->resolve_attachments.size()) + ") must match the number of color attachments (" + itos(pass->color_attachments.size()) + ")."); ERR_FAIL_COND_V_MSG(texture_samples == TEXTURE_SAMPLES_1, VK_NULL_HANDLE, "Resolve attachments specified, but color attachments are not multisample."); } for (int j = 0; j < pass->resolve_attachments.size(); j++) { int32_t attachment = pass->resolve_attachments[j]; VkAttachmentReference2KHR reference; reference.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2_KHR; reference.pNext = nullptr; if (attachment == FramebufferPass::ATTACHMENT_UNUSED) { reference.attachment = VK_ATTACHMENT_UNUSED; reference.layout = VK_IMAGE_LAYOUT_UNDEFINED; } else { ERR_FAIL_INDEX_V_MSG(attachment, p_attachments.size(), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), resolve attachment (" + itos(j) + ")."); ERR_FAIL_COND_V_MSG(pass->color_attachments[j] == FramebufferPass::ATTACHMENT_UNUSED, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), resolve attachment (" + itos(j) + "), the respective color attachment is marked as unused."); ERR_FAIL_COND_V_MSG(!(p_attachments[attachment].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), resolve attachment, it isn't marked as a color texture."); ERR_FAIL_COND_V_MSG(attachment_last_pass[attachment] == i, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it already was used for something else before in this pass."); bool multisample = p_attachments[attachment].samples > TEXTURE_SAMPLES_1; ERR_FAIL_COND_V_MSG(multisample, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), resolve attachments can't be multisample."); reference.attachment = attachment_remap[attachment]; reference.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; // VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL attachment_last_pass[attachment] = i; } reference.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; resolve_references.push_back(reference); } VkAttachmentReference2KHR &depth_stencil_reference = depth_reference_array[i]; depth_stencil_reference.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2_KHR; depth_stencil_reference.pNext = nullptr; if (pass->depth_attachment != FramebufferPass::ATTACHMENT_UNUSED) { int32_t attachment = pass->depth_attachment; ERR_FAIL_INDEX_V_MSG(attachment, p_attachments.size(), VK_NULL_HANDLE, "Invalid framebuffer depth format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), depth attachment."); ERR_FAIL_COND_V_MSG(!(p_attachments[attachment].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT), VK_NULL_HANDLE, "Invalid framebuffer depth format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it's marked as depth, but it's not a depth attachment."); ERR_FAIL_COND_V_MSG(attachment_last_pass[attachment] == i, VK_NULL_HANDLE, "Invalid framebuffer depth format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it already was used for something else before in this pass."); depth_stencil_reference.attachment = attachment_remap[attachment]; depth_stencil_reference.layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; depth_stencil_reference.aspectMask = VK_IMAGE_ASPECT_NONE; attachment_last_pass[attachment] = i; if (is_multisample_first) { texture_samples = p_attachments[attachment].samples; is_multisample_first = false; } else { ERR_FAIL_COND_V_MSG(texture_samples != p_attachments[attachment].samples, VK_NULL_HANDLE, "Invalid framebuffer depth format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), if an attachment is marked as multisample, all of them should be multisample and use the same number of samples including the depth."); } } else { depth_stencil_reference.attachment = VK_ATTACHMENT_UNUSED; depth_stencil_reference.layout = VK_IMAGE_LAYOUT_UNDEFINED; } if (context->get_vrs_capabilities().attachment_vrs_supported && pass->vrs_attachment != FramebufferPass::ATTACHMENT_UNUSED) { int32_t attachment = pass->vrs_attachment; ERR_FAIL_INDEX_V_MSG(attachment, p_attachments.size(), VK_NULL_HANDLE, "Invalid framebuffer VRS format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), VRS attachment."); ERR_FAIL_COND_V_MSG(!(p_attachments[attachment].usage_flags & TEXTURE_USAGE_VRS_ATTACHMENT_BIT), VK_NULL_HANDLE, "Invalid framebuffer VRS format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it's marked as VRS, but it's not a VRS attachment."); ERR_FAIL_COND_V_MSG(attachment_last_pass[attachment] == i, VK_NULL_HANDLE, "Invalid framebuffer VRS attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it already was used for something else before in this pass."); VkAttachmentReference2KHR &vrs_reference = vrs_reference_array[i]; vrs_reference.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2_KHR; vrs_reference.pNext = nullptr; vrs_reference.attachment = attachment_remap[attachment]; vrs_reference.layout = VK_IMAGE_LAYOUT_FRAGMENT_SHADING_RATE_ATTACHMENT_OPTIMAL_KHR; vrs_reference.aspectMask = VK_IMAGE_ASPECT_NONE; Size2i texel_size = context->get_vrs_capabilities().texel_size; VkFragmentShadingRateAttachmentInfoKHR &vrs_attachment_info = vrs_attachment_info_array[i]; vrs_attachment_info.sType = VK_STRUCTURE_TYPE_FRAGMENT_SHADING_RATE_ATTACHMENT_INFO_KHR; vrs_attachment_info.pNext = nullptr; vrs_attachment_info.pFragmentShadingRateAttachment = &vrs_reference; vrs_attachment_info.shadingRateAttachmentTexelSize = { uint32_t(texel_size.x), uint32_t(texel_size.y) }; attachment_last_pass[attachment] = i; subpass_nextptr = &vrs_attachment_info; } LocalVector<uint32_t> &preserve_references = preserve_reference_array[i]; for (int j = 0; j < pass->preserve_attachments.size(); j++) { int32_t attachment = pass->preserve_attachments[j]; ERR_FAIL_COND_V_MSG(attachment == FramebufferPass::ATTACHMENT_UNUSED, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), preserve attachment (" + itos(j) + "). Preserve attachments can't be unused."); ERR_FAIL_INDEX_V_MSG(attachment, p_attachments.size(), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), preserve attachment (" + itos(j) + ")."); if (attachment_last_pass[attachment] != i) { // Preserve can still be used to keep depth or color from being discarded after use. attachment_last_pass[attachment] = i; preserve_references.push_back(attachment); } } VkSubpassDescription2KHR &subpass = subpasses[i]; subpass.sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_2_KHR; subpass.pNext = subpass_nextptr; subpass.flags = 0; subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS; if (p_view_count == 1) { // VUID-VkSubpassDescription2-multiview-06558: If the multiview feature is not enabled, viewMask must be 0. subpass.viewMask = 0; } else { subpass.viewMask = view_mask; } subpass.inputAttachmentCount = input_references.size(); if (input_references.size()) { subpass.pInputAttachments = input_references.ptr(); } else { subpass.pInputAttachments = nullptr; } subpass.colorAttachmentCount = color_references.size(); if (color_references.size()) { subpass.pColorAttachments = color_references.ptr(); } else { subpass.pColorAttachments = nullptr; } if (depth_stencil_reference.attachment != VK_ATTACHMENT_UNUSED) { subpass.pDepthStencilAttachment = &depth_stencil_reference; } else { subpass.pDepthStencilAttachment = nullptr; } if (resolve_references.size()) { subpass.pResolveAttachments = resolve_references.ptr(); } else { subpass.pResolveAttachments = nullptr; } subpass.preserveAttachmentCount = preserve_references.size(); if (preserve_references.size()) { subpass.pPreserveAttachments = preserve_references.ptr(); } else { subpass.pPreserveAttachments = nullptr; } if (r_samples) { r_samples->push_back(texture_samples); } if (i > 0) { VkSubpassDependency2KHR dependency; dependency.sType = VK_STRUCTURE_TYPE_SUBPASS_DEPENDENCY_2_KHR; dependency.pNext = nullptr; dependency.srcSubpass = i - 1; dependency.dstSubpass = i; dependency.srcStageMask = 0; dependency.srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT; dependency.dstStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; dependency.srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; dependency.dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_INPUT_ATTACHMENT_READ_BIT; dependency.dependencyFlags = VK_DEPENDENCY_BY_REGION_BIT; dependency.viewOffset = 0; subpass_dependencies.push_back(dependency); } /* // NOTE: Big Mallet Approach -- any layout transition causes a full barrier. if (reference.layout != description.initialLayout) { // NOTE: This should be smarter based on the texture's knowledge of its previous role. dependency_from_external.srcStageMask |= VK_PIPELINE_STAGE_ALL_COMMANDS_BIT; dependency_from_external.srcAccessMask |= VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT; } if (reference.layout != description.finalLayout) { // NOTE: This should be smarter based on the texture's knowledge of its subsequent role. dependency_to_external.dstStageMask |= VK_PIPELINE_STAGE_ALL_COMMANDS_BIT; dependency_to_external.dstAccessMask |= VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT; } */ } VkRenderPassCreateInfo2KHR render_pass_create_info; render_pass_create_info.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2_KHR; render_pass_create_info.pNext = nullptr; render_pass_create_info.flags = 0; render_pass_create_info.attachmentCount = attachments.size(); render_pass_create_info.pAttachments = attachments.ptr(); render_pass_create_info.subpassCount = subpasses.size(); render_pass_create_info.pSubpasses = subpasses.ptr(); // Commenting this because it seems it just avoids raster and compute to work at the same time. // Other barriers seem to be protecting the render pass fine. // render_pass_create_info.dependencyCount = 2; // render_pass_create_info.pDependencies = dependencies; render_pass_create_info.dependencyCount = subpass_dependencies.size(); if (subpass_dependencies.size()) { render_pass_create_info.pDependencies = subpass_dependencies.ptr(); } else { render_pass_create_info.pDependencies = nullptr; } if (p_view_count == 1) { // VUID-VkRenderPassCreateInfo2-viewMask-03057: If the VkSubpassDescription2::viewMask member of all elements of pSubpasses is 0, correlatedViewMaskCount must be 0. render_pass_create_info.correlatedViewMaskCount = 0; render_pass_create_info.pCorrelatedViewMasks = nullptr; } else { render_pass_create_info.correlatedViewMaskCount = 1; render_pass_create_info.pCorrelatedViewMasks = &correlation_mask; } Vector<uint32_t> view_masks; VkRenderPassMultiviewCreateInfo render_pass_multiview_create_info; if ((p_view_count > 1) && !context->supports_renderpass2()) { // This is only required when using vkCreateRenderPass, we add it if vkCreateRenderPass2KHR is not supported // resulting this in being passed to our vkCreateRenderPass fallback. // Set view masks for each subpass. for (uint32_t i = 0; i < subpasses.size(); i++) { view_masks.push_back(view_mask); } render_pass_multiview_create_info.sType = VK_STRUCTURE_TYPE_RENDER_PASS_MULTIVIEW_CREATE_INFO; render_pass_multiview_create_info.pNext = nullptr; render_pass_multiview_create_info.subpassCount = subpasses.size(); render_pass_multiview_create_info.pViewMasks = view_masks.ptr(); render_pass_multiview_create_info.dependencyCount = 0; render_pass_multiview_create_info.pViewOffsets = nullptr; render_pass_multiview_create_info.correlationMaskCount = 1; render_pass_multiview_create_info.pCorrelationMasks = &correlation_mask; render_pass_create_info.pNext = &render_pass_multiview_create_info; } VkRenderPass render_pass; VkResult res = context->vkCreateRenderPass2KHR(device, &render_pass_create_info, nullptr, &render_pass); ERR_FAIL_COND_V_MSG(res, VK_NULL_HANDLE, "vkCreateRenderPass2KHR failed with error " + itos(res) + "."); return render_pass; } RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_create(const Vector<AttachmentFormat> &p_format, uint32_t p_view_count) { FramebufferPass pass; for (int i = 0; i < p_format.size(); i++) { if (p_format[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { pass.depth_attachment = i; } else { pass.color_attachments.push_back(i); } } Vector<FramebufferPass> passes; passes.push_back(pass); return framebuffer_format_create_multipass(p_format, passes, p_view_count); } RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_create_multipass(const Vector<AttachmentFormat> &p_attachments, const Vector<FramebufferPass> &p_passes, uint32_t p_view_count) { _THREAD_SAFE_METHOD_ FramebufferFormatKey key; key.attachments = p_attachments; key.passes = p_passes; key.view_count = p_view_count; const RBMap<FramebufferFormatKey, FramebufferFormatID>::Element *E = framebuffer_format_cache.find(key); if (E) { // Exists, return. return E->get(); } Vector<TextureSamples> samples; VkRenderPass render_pass = _render_pass_create(p_attachments, p_passes, INITIAL_ACTION_CLEAR, FINAL_ACTION_READ, INITIAL_ACTION_CLEAR, FINAL_ACTION_READ, p_view_count, &samples); // Actions don't matter for this use case. if (render_pass == VK_NULL_HANDLE) { // Was likely invalid. return INVALID_ID; } FramebufferFormatID id = FramebufferFormatID(framebuffer_format_cache.size()) | (FramebufferFormatID(ID_TYPE_FRAMEBUFFER_FORMAT) << FramebufferFormatID(ID_BASE_SHIFT)); E = framebuffer_format_cache.insert(key, id); FramebufferFormat fb_format; fb_format.E = E; fb_format.render_pass = render_pass; fb_format.pass_samples = samples; fb_format.view_count = p_view_count; framebuffer_formats[id] = fb_format; return id; } RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_create_empty(TextureSamples p_samples) { FramebufferFormatKey key; key.passes.push_back(FramebufferPass()); const RBMap<FramebufferFormatKey, FramebufferFormatID>::Element *E = framebuffer_format_cache.find(key); if (E) { // Exists, return. return E->get(); } VkSubpassDescription2KHR subpass; subpass.sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_2_KHR; subpass.pNext = nullptr; subpass.flags = 0; subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS; subpass.viewMask = 0; subpass.inputAttachmentCount = 0; // Unsupported for now. subpass.pInputAttachments = nullptr; subpass.colorAttachmentCount = 0; subpass.pColorAttachments = nullptr; subpass.pDepthStencilAttachment = nullptr; subpass.pResolveAttachments = nullptr; subpass.preserveAttachmentCount = 0; subpass.pPreserveAttachments = nullptr; VkRenderPassCreateInfo2KHR render_pass_create_info; render_pass_create_info.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2_KHR; render_pass_create_info.pNext = nullptr; render_pass_create_info.flags = 0; render_pass_create_info.attachmentCount = 0; render_pass_create_info.pAttachments = nullptr; render_pass_create_info.subpassCount = 1; render_pass_create_info.pSubpasses = &subpass; render_pass_create_info.dependencyCount = 0; render_pass_create_info.pDependencies = nullptr; render_pass_create_info.correlatedViewMaskCount = 0; render_pass_create_info.pCorrelatedViewMasks = nullptr; VkRenderPass render_pass; VkResult res = context->vkCreateRenderPass2KHR(device, &render_pass_create_info, nullptr, &render_pass); ERR_FAIL_COND_V_MSG(res, 0, "vkCreateRenderPass2KHR for empty fb failed with error " + itos(res) + "."); if (render_pass == VK_NULL_HANDLE) { // Was likely invalid. return INVALID_ID; } FramebufferFormatID id = FramebufferFormatID(framebuffer_format_cache.size()) | (FramebufferFormatID(ID_TYPE_FRAMEBUFFER_FORMAT) << FramebufferFormatID(ID_BASE_SHIFT)); E = framebuffer_format_cache.insert(key, id); FramebufferFormat fb_format; fb_format.E = E; fb_format.render_pass = render_pass; fb_format.pass_samples.push_back(p_samples); framebuffer_formats[id] = fb_format; return id; } RenderingDevice::TextureSamples RenderingDeviceVulkan::framebuffer_format_get_texture_samples(FramebufferFormatID p_format, uint32_t p_pass) { HashMap<FramebufferFormatID, FramebufferFormat>::Iterator E = framebuffer_formats.find(p_format); ERR_FAIL_COND_V(!E, TEXTURE_SAMPLES_1); ERR_FAIL_COND_V(p_pass >= uint32_t(E->value.pass_samples.size()), TEXTURE_SAMPLES_1); return E->value.pass_samples[p_pass]; } /***********************/ /**** RENDER TARGET ****/ /***********************/ RID RenderingDeviceVulkan::framebuffer_create_empty(const Size2i &p_size, TextureSamples p_samples, FramebufferFormatID p_format_check) { _THREAD_SAFE_METHOD_ Framebuffer framebuffer; framebuffer.format_id = framebuffer_format_create_empty(p_samples); ERR_FAIL_COND_V(p_format_check != INVALID_FORMAT_ID && framebuffer.format_id != p_format_check, RID()); framebuffer.size = p_size; framebuffer.view_count = 1; RID id = framebuffer_owner.make_rid(framebuffer); #ifdef DEV_ENABLED set_resource_name(id, "RID:" + itos(id.get_id())); #endif return id; } RID RenderingDeviceVulkan::framebuffer_create(const Vector<RID> &p_texture_attachments, FramebufferFormatID p_format_check, uint32_t p_view_count) { _THREAD_SAFE_METHOD_ FramebufferPass pass; for (int i = 0; i < p_texture_attachments.size(); i++) { Texture *texture = texture_owner.get_or_null(p_texture_attachments[i]); ERR_FAIL_COND_V_MSG(texture && texture->layers != p_view_count, RID(), "Layers of our texture doesn't match view count for this framebuffer"); if (texture && texture->usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { pass.depth_attachment = i; } else if (texture && texture->usage_flags & TEXTURE_USAGE_VRS_ATTACHMENT_BIT) { pass.vrs_attachment = i; } else { if (texture && texture->is_resolve_buffer) { pass.resolve_attachments.push_back(i); } else { pass.color_attachments.push_back(texture ? i : FramebufferPass::ATTACHMENT_UNUSED); } } } Vector<FramebufferPass> passes; passes.push_back(pass); return framebuffer_create_multipass(p_texture_attachments, passes, p_format_check, p_view_count); } RID RenderingDeviceVulkan::framebuffer_create_multipass(const Vector<RID> &p_texture_attachments, const Vector<FramebufferPass> &p_passes, FramebufferFormatID p_format_check, uint32_t p_view_count) { _THREAD_SAFE_METHOD_ Vector<AttachmentFormat> attachments; attachments.resize(p_texture_attachments.size()); Size2i size; bool size_set = false; for (int i = 0; i < p_texture_attachments.size(); i++) { AttachmentFormat af; Texture *texture = texture_owner.get_or_null(p_texture_attachments[i]); if (!texture) { af.usage_flags = AttachmentFormat::UNUSED_ATTACHMENT; } else { ERR_FAIL_COND_V_MSG(texture->layers != p_view_count, RID(), "Layers of our texture doesn't match view count for this framebuffer"); if (!size_set) { size.width = texture->width; size.height = texture->height; size_set = true; } else if (texture->usage_flags & TEXTURE_USAGE_VRS_ATTACHMENT_BIT) { // If this is not the first attachment we assume this is used as the VRS attachment. // In this case this texture will be 1/16th the size of the color attachment. // So we skip the size check. } else { ERR_FAIL_COND_V_MSG((uint32_t)size.width != texture->width || (uint32_t)size.height != texture->height, RID(), "All textures in a framebuffer should be the same size."); } af.format = texture->format; af.samples = texture->samples; af.usage_flags = texture->usage_flags; } attachments.write[i] = af; } ERR_FAIL_COND_V_MSG(!size_set, RID(), "All attachments unused."); FramebufferFormatID format_id = framebuffer_format_create_multipass(attachments, p_passes, p_view_count); if (format_id == INVALID_ID) { return RID(); } ERR_FAIL_COND_V_MSG(p_format_check != INVALID_ID && format_id != p_format_check, RID(), "The format used to check this framebuffer differs from the intended framebuffer format."); Framebuffer framebuffer; framebuffer.format_id = format_id; framebuffer.texture_ids = p_texture_attachments; framebuffer.size = size; framebuffer.view_count = p_view_count; RID id = framebuffer_owner.make_rid(framebuffer); #ifdef DEV_ENABLED set_resource_name(id, "RID:" + itos(id.get_id())); #endif for (int i = 0; i < p_texture_attachments.size(); i++) { if (p_texture_attachments[i].is_valid()) { _add_dependency(id, p_texture_attachments[i]); } } return id; } RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_get_format(RID p_framebuffer) { _THREAD_SAFE_METHOD_ Framebuffer *framebuffer = framebuffer_owner.get_or_null(p_framebuffer); ERR_FAIL_NULL_V(framebuffer, INVALID_ID); return framebuffer->format_id; } bool RenderingDeviceVulkan::framebuffer_is_valid(RID p_framebuffer) const { _THREAD_SAFE_METHOD_ return framebuffer_owner.owns(p_framebuffer); } void RenderingDeviceVulkan::framebuffer_set_invalidation_callback(RID p_framebuffer, InvalidationCallback p_callback, void *p_userdata) { _THREAD_SAFE_METHOD_ Framebuffer *framebuffer = framebuffer_owner.get_or_null(p_framebuffer); ERR_FAIL_NULL(framebuffer); framebuffer->invalidated_callback = p_callback; framebuffer->invalidated_callback_userdata = p_userdata; } /*****************/ /**** SAMPLER ****/ /*****************/ RID RenderingDeviceVulkan::sampler_create(const SamplerState &p_state) { _THREAD_SAFE_METHOD_ VkSamplerCreateInfo sampler_create_info; sampler_create_info.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO; sampler_create_info.pNext = nullptr; sampler_create_info.flags = 0; sampler_create_info.magFilter = p_state.mag_filter == SAMPLER_FILTER_LINEAR ? VK_FILTER_LINEAR : VK_FILTER_NEAREST; sampler_create_info.minFilter = p_state.min_filter == SAMPLER_FILTER_LINEAR ? VK_FILTER_LINEAR : VK_FILTER_NEAREST; sampler_create_info.mipmapMode = p_state.mip_filter == SAMPLER_FILTER_LINEAR ? VK_SAMPLER_MIPMAP_MODE_LINEAR : VK_SAMPLER_MIPMAP_MODE_NEAREST; ERR_FAIL_INDEX_V(p_state.repeat_u, SAMPLER_REPEAT_MODE_MAX, RID()); sampler_create_info.addressModeU = address_modes[p_state.repeat_u]; ERR_FAIL_INDEX_V(p_state.repeat_v, SAMPLER_REPEAT_MODE_MAX, RID()); sampler_create_info.addressModeV = address_modes[p_state.repeat_v]; ERR_FAIL_INDEX_V(p_state.repeat_w, SAMPLER_REPEAT_MODE_MAX, RID()); sampler_create_info.addressModeW = address_modes[p_state.repeat_w]; sampler_create_info.mipLodBias = p_state.lod_bias; sampler_create_info.anisotropyEnable = p_state.use_anisotropy && context->get_physical_device_features().samplerAnisotropy; sampler_create_info.maxAnisotropy = p_state.anisotropy_max; sampler_create_info.compareEnable = p_state.enable_compare; ERR_FAIL_INDEX_V(p_state.compare_op, COMPARE_OP_MAX, RID()); sampler_create_info.compareOp = compare_operators[p_state.compare_op]; sampler_create_info.minLod = p_state.min_lod; sampler_create_info.maxLod = p_state.max_lod; ERR_FAIL_INDEX_V(p_state.border_color, SAMPLER_BORDER_COLOR_MAX, RID()); sampler_create_info.borderColor = sampler_border_colors[p_state.border_color]; sampler_create_info.unnormalizedCoordinates = p_state.unnormalized_uvw; VkSampler sampler; VkResult res = vkCreateSampler(device, &sampler_create_info, nullptr, &sampler); ERR_FAIL_COND_V_MSG(res, RID(), "vkCreateSampler failed with error " + itos(res) + "."); RID id = sampler_owner.make_rid(sampler); #ifdef DEV_ENABLED set_resource_name(id, "RID:" + itos(id.get_id())); #endif return id; } bool RenderingDeviceVulkan::sampler_is_format_supported_for_filter(DataFormat p_format, SamplerFilter p_sampler_filter) const { ERR_FAIL_INDEX_V(p_format, DATA_FORMAT_MAX, false); _THREAD_SAFE_METHOD_ // Validate that this image is supported for the intended filtering. VkFormatProperties properties; vkGetPhysicalDeviceFormatProperties(context->get_physical_device(), vulkan_formats[p_format], &properties); return p_sampler_filter == RD::SAMPLER_FILTER_NEAREST || (p_sampler_filter == RD::SAMPLER_FILTER_LINEAR && (properties.optimalTilingFeatures & VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_LINEAR_BIT)); } /**********************/ /**** VERTEX ARRAY ****/ /**********************/ RID RenderingDeviceVulkan::vertex_buffer_create(uint32_t p_size_bytes, const Vector<uint8_t> &p_data, bool p_use_as_storage) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V(p_data.size() && (uint32_t)p_data.size() != p_size_bytes, RID()); uint32_t usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_VERTEX_BUFFER_BIT; if (p_use_as_storage) { usage |= VK_BUFFER_USAGE_STORAGE_BUFFER_BIT; } Buffer buffer; _buffer_allocate(&buffer, p_size_bytes, usage, VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE, 0); if (p_data.size()) { uint64_t data_size = p_data.size(); const uint8_t *r = p_data.ptr(); _buffer_update(&buffer, 0, r, data_size); _buffer_memory_barrier(buffer.buffer, 0, data_size, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT, false); } RID id = vertex_buffer_owner.make_rid(buffer); #ifdef DEV_ENABLED set_resource_name(id, "RID:" + itos(id.get_id())); #endif return id; } // Internally reference counted, this ID is warranted to be unique for the same description, but needs to be freed as many times as it was allocated. RenderingDevice::VertexFormatID RenderingDeviceVulkan::vertex_format_create(const Vector<VertexAttribute> &p_vertex_formats) { _THREAD_SAFE_METHOD_ VertexDescriptionKey key; key.vertex_formats = p_vertex_formats; VertexFormatID *idptr = vertex_format_cache.getptr(key); if (idptr) { return *idptr; } // Does not exist, create one and cache it. VertexDescriptionCache vdcache; vdcache.bindings = memnew_arr(VkVertexInputBindingDescription, p_vertex_formats.size()); vdcache.attributes = memnew_arr(VkVertexInputAttributeDescription, p_vertex_formats.size()); HashSet<int> used_locations; for (int i = 0; i < p_vertex_formats.size(); i++) { ERR_CONTINUE(p_vertex_formats[i].format >= DATA_FORMAT_MAX); ERR_FAIL_COND_V(used_locations.has(p_vertex_formats[i].location), INVALID_ID); ERR_FAIL_COND_V_MSG(get_format_vertex_size(p_vertex_formats[i].format) == 0, INVALID_ID, "Data format for attachment (" + itos(i) + "), '" + named_formats[p_vertex_formats[i].format] + "', is not valid for a vertex array."); vdcache.bindings[i].binding = i; vdcache.bindings[i].stride = p_vertex_formats[i].stride; vdcache.bindings[i].inputRate = p_vertex_formats[i].frequency == VERTEX_FREQUENCY_INSTANCE ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX; vdcache.attributes[i].binding = i; vdcache.attributes[i].location = p_vertex_formats[i].location; vdcache.attributes[i].format = vulkan_formats[p_vertex_formats[i].format]; vdcache.attributes[i].offset = p_vertex_formats[i].offset; used_locations.insert(p_vertex_formats[i].location); } vdcache.create_info.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO; vdcache.create_info.pNext = nullptr; vdcache.create_info.flags = 0; vdcache.create_info.vertexAttributeDescriptionCount = p_vertex_formats.size(); vdcache.create_info.pVertexAttributeDescriptions = vdcache.attributes; vdcache.create_info.vertexBindingDescriptionCount = p_vertex_formats.size(); vdcache.create_info.pVertexBindingDescriptions = vdcache.bindings; vdcache.vertex_formats = p_vertex_formats; VertexFormatID id = VertexFormatID(vertex_format_cache.size()) | (VertexFormatID(ID_TYPE_VERTEX_FORMAT) << ID_BASE_SHIFT); vertex_format_cache[key] = id; vertex_formats[id] = vdcache; return id; } RID RenderingDeviceVulkan::vertex_array_create(uint32_t p_vertex_count, VertexFormatID p_vertex_format, const Vector<RID> &p_src_buffers, const Vector<uint64_t> &p_offsets) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V(!vertex_formats.has(p_vertex_format), RID()); const VertexDescriptionCache &vd = vertex_formats[p_vertex_format]; ERR_FAIL_COND_V(vd.vertex_formats.size() != p_src_buffers.size(), RID()); for (int i = 0; i < p_src_buffers.size(); i++) { ERR_FAIL_COND_V(!vertex_buffer_owner.owns(p_src_buffers[i]), RID()); } VertexArray vertex_array; if (p_offsets.is_empty()) { vertex_array.offsets.resize_zeroed(p_src_buffers.size()); } else { ERR_FAIL_COND_V(p_offsets.size() != p_src_buffers.size(), RID()); vertex_array.offsets = p_offsets; } vertex_array.vertex_count = p_vertex_count; vertex_array.description = p_vertex_format; vertex_array.max_instances_allowed = 0xFFFFFFFF; // By default as many as you want. for (int i = 0; i < p_src_buffers.size(); i++) { Buffer *buffer = vertex_buffer_owner.get_or_null(p_src_buffers[i]); // Validate with buffer. { const VertexAttribute &atf = vd.vertex_formats[i]; uint32_t element_size = get_format_vertex_size(atf.format); ERR_FAIL_COND_V(element_size == 0, RID()); // Should never happens since this was prevalidated. if (atf.frequency == VERTEX_FREQUENCY_VERTEX) { // Validate size for regular drawing. uint64_t total_size = uint64_t(atf.stride) * (p_vertex_count - 1) + atf.offset + element_size; ERR_FAIL_COND_V_MSG(total_size > buffer->size, RID(), "Attachment (" + itos(i) + ") will read past the end of the buffer."); } else { // Validate size for instances drawing. uint64_t available = buffer->size - atf.offset; ERR_FAIL_COND_V_MSG(available < element_size, RID(), "Attachment (" + itos(i) + ") uses instancing, but it's just too small."); uint32_t instances_allowed = available / atf.stride; vertex_array.max_instances_allowed = MIN(instances_allowed, vertex_array.max_instances_allowed); } } vertex_array.buffers.push_back(buffer->buffer); } RID id = vertex_array_owner.make_rid(vertex_array); for (int i = 0; i < p_src_buffers.size(); i++) { _add_dependency(id, p_src_buffers[i]); } return id; } RID RenderingDeviceVulkan::index_buffer_create(uint32_t p_index_count, IndexBufferFormat p_format, const Vector<uint8_t> &p_data, bool p_use_restart_indices) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V(p_index_count == 0, RID()); IndexBuffer index_buffer; index_buffer.index_type = (p_format == INDEX_BUFFER_FORMAT_UINT16) ? VK_INDEX_TYPE_UINT16 : VK_INDEX_TYPE_UINT32; index_buffer.supports_restart_indices = p_use_restart_indices; index_buffer.index_count = p_index_count; uint32_t size_bytes = p_index_count * ((p_format == INDEX_BUFFER_FORMAT_UINT16) ? 2 : 4); #ifdef DEBUG_ENABLED if (p_data.size()) { index_buffer.max_index = 0; ERR_FAIL_COND_V_MSG((uint32_t)p_data.size() != size_bytes, RID(), "Default index buffer initializer array size (" + itos(p_data.size()) + ") does not match format required size (" + itos(size_bytes) + ")."); const uint8_t *r = p_data.ptr(); if (p_format == INDEX_BUFFER_FORMAT_UINT16) { const uint16_t *index16 = (const uint16_t *)r; for (uint32_t i = 0; i < p_index_count; i++) { if (p_use_restart_indices && index16[i] == 0xFFFF) { continue; // Restart index, ignore. } index_buffer.max_index = MAX(index16[i], index_buffer.max_index); } } else { const uint32_t *index32 = (const uint32_t *)r; for (uint32_t i = 0; i < p_index_count; i++) { if (p_use_restart_indices && index32[i] == 0xFFFFFFFF) { continue; // Restart index, ignore. } index_buffer.max_index = MAX(index32[i], index_buffer.max_index); } } } else { index_buffer.max_index = 0xFFFFFFFF; } #else index_buffer.max_index = 0xFFFFFFFF; #endif _buffer_allocate(&index_buffer, size_bytes, VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT, VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE, 0); if (p_data.size()) { uint64_t data_size = p_data.size(); const uint8_t *r = p_data.ptr(); _buffer_update(&index_buffer, 0, r, data_size); _buffer_memory_barrier(index_buffer.buffer, 0, data_size, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_INDEX_READ_BIT, false); } RID id = index_buffer_owner.make_rid(index_buffer); #ifdef DEV_ENABLED set_resource_name(id, "RID:" + itos(id.get_id())); #endif return id; } RID RenderingDeviceVulkan::index_array_create(RID p_index_buffer, uint32_t p_index_offset, uint32_t p_index_count) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V(!index_buffer_owner.owns(p_index_buffer), RID()); IndexBuffer *index_buffer = index_buffer_owner.get_or_null(p_index_buffer); ERR_FAIL_COND_V(p_index_count == 0, RID()); ERR_FAIL_COND_V(p_index_offset + p_index_count > index_buffer->index_count, RID()); IndexArray index_array; index_array.max_index = index_buffer->max_index; index_array.buffer = index_buffer->buffer; index_array.offset = p_index_offset; index_array.indices = p_index_count; index_array.index_type = index_buffer->index_type; index_array.supports_restart_indices = index_buffer->supports_restart_indices; RID id = index_array_owner.make_rid(index_array); _add_dependency(id, p_index_buffer); return id; } /****************/ /**** SHADER ****/ /****************/ static const char *shader_uniform_names[RenderingDevice::UNIFORM_TYPE_MAX] = { "Sampler", "CombinedSampler", "Texture", "Image", "TextureBuffer", "SamplerTextureBuffer", "ImageBuffer", "UniformBuffer", "StorageBuffer", "InputAttachment" }; static VkShaderStageFlagBits shader_stage_masks[RenderingDevice::SHADER_STAGE_MAX] = { VK_SHADER_STAGE_VERTEX_BIT, VK_SHADER_STAGE_FRAGMENT_BIT, VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT, VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT, VK_SHADER_STAGE_COMPUTE_BIT, }; String RenderingDeviceVulkan::_shader_uniform_debug(RID p_shader, int p_set) { String ret; const Shader *shader = shader_owner.get_or_null(p_shader); ERR_FAIL_NULL_V(shader, String()); for (int i = 0; i < shader->sets.size(); i++) { if (p_set >= 0 && i != p_set) { continue; } for (int j = 0; j < shader->sets[i].uniform_info.size(); j++) { const UniformInfo &ui = shader->sets[i].uniform_info[j]; if (!ret.is_empty()) { ret += "\n"; } ret += "Set: " + itos(i) + " Binding: " + itos(ui.binding) + " Type: " + shader_uniform_names[ui.type] + " Writable: " + (ui.writable ? "Y" : "N") + " Length: " + itos(ui.length); } } return ret; } // Version 1: initial. // Version 2: Added shader name. // Version 3: Added writable. #define SHADER_BINARY_VERSION 3 String RenderingDeviceVulkan::shader_get_binary_cache_key() const { return "Vulkan-SV" + itos(SHADER_BINARY_VERSION); } struct RenderingDeviceVulkanShaderBinaryDataBinding { uint32_t type; uint32_t binding; uint32_t stages; uint32_t length; // Size of arrays (in total elements), or ubos (in bytes * total elements). uint32_t writable; }; struct RenderingDeviceVulkanShaderBinarySpecializationConstant { uint32_t type; uint32_t constant_id; union { uint32_t int_value; float float_value; bool bool_value; }; uint32_t stage_flags; }; struct RenderingDeviceVulkanShaderBinaryData { uint32_t vertex_input_mask; uint32_t fragment_output_mask; uint32_t specialization_constants_count; uint32_t is_compute; uint32_t compute_local_size[3]; uint32_t set_count; uint32_t push_constant_size; uint32_t push_constant_vk_stages_mask; uint32_t stage_count; uint32_t shader_name_len; }; Vector<uint8_t> RenderingDeviceVulkan::shader_compile_binary_from_spirv(const Vector<ShaderStageSPIRVData> &p_spirv, const String &p_shader_name) { SpirvReflectionData spirv_data; if (_reflect_spirv(p_spirv, spirv_data) != OK) { return Vector<uint8_t>(); } ERR_FAIL_COND_V_MSG((uint32_t)spirv_data.uniforms.size() > limits.maxBoundDescriptorSets, Vector<uint8_t>(), "Number of uniform sets is larger than what is supported by the hardware (" + itos(limits.maxBoundDescriptorSets) + ")."); // Collect reflection data into binary data. RenderingDeviceVulkanShaderBinaryData binary_data{}; Vector<Vector<RenderingDeviceVulkanShaderBinaryDataBinding>> uniform_info; // Set bindings. Vector<RenderingDeviceVulkanShaderBinarySpecializationConstant> specialization_constants; { binary_data.vertex_input_mask = spirv_data.vertex_input_mask; binary_data.fragment_output_mask = spirv_data.fragment_output_mask; binary_data.specialization_constants_count = spirv_data.specialization_constants.size(); binary_data.is_compute = spirv_data.is_compute; binary_data.compute_local_size[0] = spirv_data.compute_local_size[0]; binary_data.compute_local_size[1] = spirv_data.compute_local_size[1]; binary_data.compute_local_size[2] = spirv_data.compute_local_size[2]; binary_data.set_count = spirv_data.uniforms.size(); binary_data.push_constant_size = spirv_data.push_constant_size; for (uint32_t i = 0; i < SHADER_STAGE_MAX; i++) { if (spirv_data.push_constant_stages_mask.has_flag((ShaderStage)(1 << i))) { binary_data.push_constant_vk_stages_mask |= shader_stage_masks[i]; } } for (const Vector<SpirvReflectionData::Uniform> &spirv_set : spirv_data.uniforms) { Vector<RenderingDeviceVulkanShaderBinaryDataBinding> set_bindings; for (const SpirvReflectionData::Uniform &spirv_uniform : spirv_set) { RenderingDeviceVulkanShaderBinaryDataBinding binding{}; binding.type = (uint32_t)spirv_uniform.type; binding.binding = spirv_uniform.binding; binding.stages = (uint32_t)spirv_uniform.stages_mask; binding.length = spirv_uniform.length; binding.writable = (uint32_t)spirv_uniform.writable; set_bindings.push_back(binding); } uniform_info.push_back(set_bindings); } for (const SpirvReflectionData::SpecializationConstant &spirv_sc : spirv_data.specialization_constants) { RenderingDeviceVulkanShaderBinarySpecializationConstant spec_constant{}; spec_constant.type = (uint32_t)spirv_sc.type; spec_constant.constant_id = spirv_sc.constant_id; spec_constant.int_value = spirv_sc.int_value; spec_constant.stage_flags = (uint32_t)spirv_sc.stages_mask; specialization_constants.push_back(spec_constant); } } Vector<Vector<uint8_t>> compressed_stages; Vector<uint32_t> smolv_size; Vector<uint32_t> zstd_size; // If 0, zstd not used. uint32_t stages_binary_size = 0; bool strip_debug = false; for (int i = 0; i < p_spirv.size(); i++) { smolv::ByteArray smolv; if (!smolv::Encode(p_spirv[i].spir_v.ptr(), p_spirv[i].spir_v.size(), smolv, strip_debug ? smolv::kEncodeFlagStripDebugInfo : 0)) { ERR_FAIL_V_MSG(Vector<uint8_t>(), "Error compressing shader stage :" + String(shader_stage_names[p_spirv[i].shader_stage])); } else { smolv_size.push_back(smolv.size()); { // zstd. Vector<uint8_t> zstd; zstd.resize(Compression::get_max_compressed_buffer_size(smolv.size(), Compression::MODE_ZSTD)); int dst_size = Compression::compress(zstd.ptrw(), &smolv[0], smolv.size(), Compression::MODE_ZSTD); if (dst_size > 0 && (uint32_t)dst_size < smolv.size()) { zstd_size.push_back(dst_size); zstd.resize(dst_size); compressed_stages.push_back(zstd); } else { Vector<uint8_t> smv; smv.resize(smolv.size()); memcpy(smv.ptrw(), &smolv[0], smolv.size()); zstd_size.push_back(0); // Not using zstd. compressed_stages.push_back(smv); } } } uint32_t s = compressed_stages[i].size(); if (s % 4 != 0) { s += 4 - (s % 4); } stages_binary_size += s; } binary_data.specialization_constants_count = specialization_constants.size(); binary_data.set_count = uniform_info.size(); binary_data.stage_count = p_spirv.size(); CharString shader_name_utf = p_shader_name.utf8(); binary_data.shader_name_len = shader_name_utf.length(); uint32_t total_size = sizeof(uint32_t) * 3; // Header + version + main datasize;. total_size += sizeof(RenderingDeviceVulkanShaderBinaryData); total_size += binary_data.shader_name_len; if ((binary_data.shader_name_len % 4) != 0) { // Alignment rules are really strange. total_size += 4 - (binary_data.shader_name_len % 4); } for (int i = 0; i < uniform_info.size(); i++) { total_size += sizeof(uint32_t); total_size += uniform_info[i].size() * sizeof(RenderingDeviceVulkanShaderBinaryDataBinding); } total_size += sizeof(RenderingDeviceVulkanShaderBinarySpecializationConstant) * specialization_constants.size(); total_size += compressed_stages.size() * sizeof(uint32_t) * 3; // Sizes. total_size += stages_binary_size; Vector<uint8_t> ret; ret.resize(total_size); { uint32_t offset = 0; uint8_t *binptr = ret.ptrw(); binptr[0] = 'G'; binptr[1] = 'S'; binptr[2] = 'B'; binptr[3] = 'D'; // Godot Shader Binary Data. offset += 4; encode_uint32(SHADER_BINARY_VERSION, binptr + offset); offset += sizeof(uint32_t); encode_uint32(sizeof(RenderingDeviceVulkanShaderBinaryData), binptr + offset); offset += sizeof(uint32_t); memcpy(binptr + offset, &binary_data, sizeof(RenderingDeviceVulkanShaderBinaryData)); offset += sizeof(RenderingDeviceVulkanShaderBinaryData); if (binary_data.shader_name_len > 0) { memcpy(binptr + offset, shader_name_utf.ptr(), binary_data.shader_name_len); offset += binary_data.shader_name_len; if ((binary_data.shader_name_len % 4) != 0) { // Alignment rules are really strange. offset += 4 - (binary_data.shader_name_len % 4); } } for (int i = 0; i < uniform_info.size(); i++) { int count = uniform_info[i].size(); encode_uint32(count, binptr + offset); offset += sizeof(uint32_t); if (count > 0) { memcpy(binptr + offset, uniform_info[i].ptr(), sizeof(RenderingDeviceVulkanShaderBinaryDataBinding) * count); offset += sizeof(RenderingDeviceVulkanShaderBinaryDataBinding) * count; } } if (specialization_constants.size()) { memcpy(binptr + offset, specialization_constants.ptr(), sizeof(RenderingDeviceVulkanShaderBinarySpecializationConstant) * specialization_constants.size()); offset += sizeof(RenderingDeviceVulkanShaderBinarySpecializationConstant) * specialization_constants.size(); } for (int i = 0; i < compressed_stages.size(); i++) { encode_uint32(p_spirv[i].shader_stage, binptr + offset); offset += sizeof(uint32_t); encode_uint32(smolv_size[i], binptr + offset); offset += sizeof(uint32_t); encode_uint32(zstd_size[i], binptr + offset); offset += sizeof(uint32_t); memcpy(binptr + offset, compressed_stages[i].ptr(), compressed_stages[i].size()); uint32_t s = compressed_stages[i].size(); if (s % 4 != 0) { s += 4 - (s % 4); } offset += s; } ERR_FAIL_COND_V(offset != (uint32_t)ret.size(), Vector<uint8_t>()); } return ret; } RID RenderingDeviceVulkan::shader_create_from_bytecode(const Vector<uint8_t> &p_shader_binary, RID p_placeholder) { const uint8_t *binptr = p_shader_binary.ptr(); uint32_t binsize = p_shader_binary.size(); uint32_t read_offset = 0; // Consistency check. ERR_FAIL_COND_V(binsize < sizeof(uint32_t) * 3 + sizeof(RenderingDeviceVulkanShaderBinaryData), RID()); ERR_FAIL_COND_V(binptr[0] != 'G' || binptr[1] != 'S' || binptr[2] != 'B' || binptr[3] != 'D', RID()); uint32_t bin_version = decode_uint32(binptr + 4); ERR_FAIL_COND_V(bin_version != SHADER_BINARY_VERSION, RID()); uint32_t bin_data_size = decode_uint32(binptr + 8); const RenderingDeviceVulkanShaderBinaryData &binary_data = *(reinterpret_cast<const RenderingDeviceVulkanShaderBinaryData *>(binptr + 12)); Shader::PushConstant push_constant; push_constant.size = binary_data.push_constant_size; push_constant.vk_stages_mask = binary_data.push_constant_vk_stages_mask; uint32_t vertex_input_mask = binary_data.vertex_input_mask; uint32_t fragment_output_mask = binary_data.fragment_output_mask; bool is_compute = binary_data.is_compute; const uint32_t compute_local_size[3] = { binary_data.compute_local_size[0], binary_data.compute_local_size[1], binary_data.compute_local_size[2] }; read_offset += sizeof(uint32_t) * 3 + bin_data_size; String name; if (binary_data.shader_name_len) { name.parse_utf8((const char *)(binptr + read_offset), binary_data.shader_name_len); read_offset += binary_data.shader_name_len; if ((binary_data.shader_name_len % 4) != 0) { // Alignment rules are really strange. read_offset += 4 - (binary_data.shader_name_len % 4); } } Vector<Vector<VkDescriptorSetLayoutBinding>> set_bindings; Vector<Vector<UniformInfo>> uniform_info; set_bindings.resize(binary_data.set_count); uniform_info.resize(binary_data.set_count); for (uint32_t i = 0; i < binary_data.set_count; i++) { ERR_FAIL_COND_V(read_offset + sizeof(uint32_t) >= binsize, RID()); uint32_t set_count = decode_uint32(binptr + read_offset); read_offset += sizeof(uint32_t); const RenderingDeviceVulkanShaderBinaryDataBinding *set_ptr = reinterpret_cast<const RenderingDeviceVulkanShaderBinaryDataBinding *>(binptr + read_offset); uint32_t set_size = set_count * sizeof(RenderingDeviceVulkanShaderBinaryDataBinding); ERR_FAIL_COND_V(read_offset + set_size >= binsize, RID()); for (uint32_t j = 0; j < set_count; j++) { UniformInfo info; info.type = UniformType(set_ptr[j].type); info.writable = set_ptr[j].writable; info.length = set_ptr[j].length; info.binding = set_ptr[j].binding; info.stages = set_ptr[j].stages; VkDescriptorSetLayoutBinding layout_binding; layout_binding.pImmutableSamplers = nullptr; layout_binding.binding = set_ptr[j].binding; layout_binding.descriptorCount = 1; layout_binding.stageFlags = 0; for (uint32_t k = 0; k < SHADER_STAGE_MAX; k++) { if (set_ptr[j].stages & (1 << k)) { layout_binding.stageFlags |= shader_stage_masks[k]; } } switch (info.type) { case UNIFORM_TYPE_SAMPLER: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLER; layout_binding.descriptorCount = set_ptr[j].length; } break; case UNIFORM_TYPE_SAMPLER_WITH_TEXTURE: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; layout_binding.descriptorCount = set_ptr[j].length; } break; case UNIFORM_TYPE_TEXTURE: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE; layout_binding.descriptorCount = set_ptr[j].length; } break; case UNIFORM_TYPE_IMAGE: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE; layout_binding.descriptorCount = set_ptr[j].length; } break; case UNIFORM_TYPE_TEXTURE_BUFFER: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; layout_binding.descriptorCount = set_ptr[j].length; } break; case UNIFORM_TYPE_IMAGE_BUFFER: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER; } break; case UNIFORM_TYPE_UNIFORM_BUFFER: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; } break; case UNIFORM_TYPE_STORAGE_BUFFER: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER; } break; case UNIFORM_TYPE_INPUT_ATTACHMENT: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT; } break; default: { ERR_FAIL_V(RID()); } } set_bindings.write[i].push_back(layout_binding); uniform_info.write[i].push_back(info); } read_offset += set_size; } ERR_FAIL_COND_V(read_offset + binary_data.specialization_constants_count * sizeof(RenderingDeviceVulkanShaderBinarySpecializationConstant) >= binsize, RID()); Vector<Shader::SpecializationConstant> specialization_constants; for (uint32_t i = 0; i < binary_data.specialization_constants_count; i++) { const RenderingDeviceVulkanShaderBinarySpecializationConstant &src_sc = *(reinterpret_cast<const RenderingDeviceVulkanShaderBinarySpecializationConstant *>(binptr + read_offset)); Shader::SpecializationConstant sc; sc.constant.int_value = src_sc.int_value; sc.constant.type = PipelineSpecializationConstantType(src_sc.type); sc.constant.constant_id = src_sc.constant_id; sc.stage_flags = src_sc.stage_flags; specialization_constants.push_back(sc); read_offset += sizeof(RenderingDeviceVulkanShaderBinarySpecializationConstant); } Vector<Vector<uint8_t>> stage_spirv_data; Vector<ShaderStage> stage_type; for (uint32_t i = 0; i < binary_data.stage_count; i++) { ERR_FAIL_COND_V(read_offset + sizeof(uint32_t) * 3 >= binsize, RID()); uint32_t stage = decode_uint32(binptr + read_offset); read_offset += sizeof(uint32_t); uint32_t smolv_size = decode_uint32(binptr + read_offset); read_offset += sizeof(uint32_t); uint32_t zstd_size = decode_uint32(binptr + read_offset); read_offset += sizeof(uint32_t); uint32_t buf_size = (zstd_size > 0) ? zstd_size : smolv_size; Vector<uint8_t> smolv; const uint8_t *src_smolv = nullptr; if (zstd_size > 0) { // Decompress to smolv. smolv.resize(smolv_size); int dec_smolv_size = Compression::decompress(smolv.ptrw(), smolv.size(), binptr + read_offset, zstd_size, Compression::MODE_ZSTD); ERR_FAIL_COND_V(dec_smolv_size != (int32_t)smolv_size, RID()); src_smolv = smolv.ptr(); } else { src_smolv = binptr + read_offset; } Vector<uint8_t> spirv; uint32_t spirv_size = smolv::GetDecodedBufferSize(src_smolv, smolv_size); spirv.resize(spirv_size); if (!smolv::Decode(src_smolv, smolv_size, spirv.ptrw(), spirv_size)) { ERR_FAIL_V_MSG(RID(), "Malformed smolv input uncompressing shader stage:" + String(shader_stage_names[stage])); } stage_spirv_data.push_back(spirv); stage_type.push_back(ShaderStage(stage)); if (buf_size % 4 != 0) { buf_size += 4 - (buf_size % 4); } ERR_FAIL_COND_V(read_offset + buf_size > binsize, RID()); read_offset += buf_size; } ERR_FAIL_COND_V(read_offset != binsize, RID()); // All good, let's create modules. _THREAD_SAFE_METHOD_ RID id; if (p_placeholder.is_null()) { id = shader_owner.make_rid(); } else { id = p_placeholder; } Shader *shader = shader_owner.get_or_null(id); shader->vertex_input_mask = vertex_input_mask; shader->fragment_output_mask = fragment_output_mask; shader->push_constant = push_constant; shader->is_compute = is_compute; shader->compute_local_size[0] = compute_local_size[0]; shader->compute_local_size[1] = compute_local_size[1]; shader->compute_local_size[2] = compute_local_size[2]; shader->specialization_constants = specialization_constants; shader->name = name; String error_text; bool success = true; for (int i = 0; i < stage_spirv_data.size(); i++) { VkShaderModuleCreateInfo shader_module_create_info; shader_module_create_info.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO; shader_module_create_info.pNext = nullptr; shader_module_create_info.flags = 0; shader_module_create_info.codeSize = stage_spirv_data[i].size(); const uint8_t *r = stage_spirv_data[i].ptr(); shader_module_create_info.pCode = (const uint32_t *)r; VkShaderModule module; VkResult res = vkCreateShaderModule(device, &shader_module_create_info, nullptr, &module); if (res) { success = false; error_text = "Error (" + itos(res) + ") creating shader module for stage: " + String(shader_stage_names[stage_type[i]]); break; } VkPipelineShaderStageCreateInfo shader_stage; shader_stage.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO; shader_stage.pNext = nullptr; shader_stage.flags = 0; shader_stage.stage = shader_stage_masks[stage_type[i]]; shader_stage.module = module; shader_stage.pName = "main"; shader_stage.pSpecializationInfo = nullptr; shader->pipeline_stages.push_back(shader_stage); } // Proceed to create descriptor sets. if (success) { for (int i = 0; i < set_bindings.size(); i++) { // Empty ones are fine if they were not used according to spec (binding count will be 0). VkDescriptorSetLayoutCreateInfo layout_create_info; layout_create_info.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO; layout_create_info.pNext = nullptr; layout_create_info.flags = 0; layout_create_info.bindingCount = set_bindings[i].size(); layout_create_info.pBindings = set_bindings[i].ptr(); VkDescriptorSetLayout layout; VkResult res = vkCreateDescriptorSetLayout(device, &layout_create_info, nullptr, &layout); if (res) { error_text = "Error (" + itos(res) + ") creating descriptor set layout for set " + itos(i); success = false; break; } Shader::Set set; set.descriptor_set_layout = layout; set.uniform_info = uniform_info[i]; // Sort and hash. set.uniform_info.sort(); uint32_t format = 0; // No format, default. if (set.uniform_info.size()) { // Has data, needs an actual format. UniformSetFormat usformat; usformat.uniform_info = set.uniform_info; RBMap<UniformSetFormat, uint32_t>::Element *E = uniform_set_format_cache.find(usformat); if (E) { format = E->get(); } else { format = uniform_set_format_cache.size() + 1; uniform_set_format_cache.insert(usformat, format); } } shader->sets.push_back(set); shader->set_formats.push_back(format); } } if (success) { // Create pipeline layout. VkPipelineLayoutCreateInfo pipeline_layout_create_info; pipeline_layout_create_info.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO; pipeline_layout_create_info.pNext = nullptr; pipeline_layout_create_info.flags = 0; pipeline_layout_create_info.setLayoutCount = shader->sets.size(); Vector<VkDescriptorSetLayout> layouts; layouts.resize(shader->sets.size()); for (int i = 0; i < layouts.size(); i++) { layouts.write[i] = shader->sets[i].descriptor_set_layout; } pipeline_layout_create_info.pSetLayouts = layouts.ptr(); // Needs to be declared in this outer scope, otherwise it may not outlive its assignment // to pipeline_layout_create_info. VkPushConstantRange push_constant_range; if (push_constant.size) { push_constant_range.stageFlags = push_constant.vk_stages_mask; push_constant_range.offset = 0; push_constant_range.size = push_constant.size; pipeline_layout_create_info.pushConstantRangeCount = 1; pipeline_layout_create_info.pPushConstantRanges = &push_constant_range; } else { pipeline_layout_create_info.pushConstantRangeCount = 0; pipeline_layout_create_info.pPushConstantRanges = nullptr; } VkResult err = vkCreatePipelineLayout(device, &pipeline_layout_create_info, nullptr, &shader->pipeline_layout); if (err) { error_text = "Error (" + itos(err) + ") creating pipeline layout."; success = false; } } if (!success) { // Clean up if failed. for (int i = 0; i < shader->pipeline_stages.size(); i++) { vkDestroyShaderModule(device, shader->pipeline_stages[i].module, nullptr); } for (int i = 0; i < shader->sets.size(); i++) { vkDestroyDescriptorSetLayout(device, shader->sets[i].descriptor_set_layout, nullptr); } shader_owner.free(id); ERR_FAIL_V_MSG(RID(), error_text); } #ifdef DEV_ENABLED set_resource_name(id, "RID:" + itos(id.get_id())); #endif return id; } RID RenderingDeviceVulkan::shader_create_placeholder() { Shader shader; return shader_owner.make_rid(shader); } uint32_t RenderingDeviceVulkan::shader_get_vertex_input_attribute_mask(RID p_shader) { _THREAD_SAFE_METHOD_ const Shader *shader = shader_owner.get_or_null(p_shader); ERR_FAIL_NULL_V(shader, 0); return shader->vertex_input_mask; } /******************/ /**** UNIFORMS ****/ /******************/ RID RenderingDeviceVulkan::uniform_buffer_create(uint32_t p_size_bytes, const Vector<uint8_t> &p_data) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V(p_data.size() && (uint32_t)p_data.size() != p_size_bytes, RID()); Buffer buffer; Error err = _buffer_allocate(&buffer, p_size_bytes, VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE, 0); ERR_FAIL_COND_V(err != OK, RID()); if (p_data.size()) { uint64_t data_size = p_data.size(); const uint8_t *r = p_data.ptr(); _buffer_update(&buffer, 0, r, data_size); _buffer_memory_barrier(buffer.buffer, 0, data_size, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_UNIFORM_READ_BIT, false); } RID id = uniform_buffer_owner.make_rid(buffer); #ifdef DEV_ENABLED set_resource_name(id, "RID:" + itos(id.get_id())); #endif return id; } RID RenderingDeviceVulkan::storage_buffer_create(uint32_t p_size_bytes, const Vector<uint8_t> &p_data, BitField<StorageBufferUsage> p_usage) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V(p_data.size() && (uint32_t)p_data.size() != p_size_bytes, RID()); Buffer buffer; uint32_t flags = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT; if (p_usage.has_flag(STORAGE_BUFFER_USAGE_DISPATCH_INDIRECT)) { flags |= VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT; } Error err = _buffer_allocate(&buffer, p_size_bytes, flags, VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE, 0); ERR_FAIL_COND_V(err != OK, RID()); if (p_data.size()) { uint64_t data_size = p_data.size(); const uint8_t *r = p_data.ptr(); _buffer_update(&buffer, 0, r, data_size); _buffer_memory_barrier(buffer.buffer, 0, data_size, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, false); } return storage_buffer_owner.make_rid(buffer); } RID RenderingDeviceVulkan::texture_buffer_create(uint32_t p_size_elements, DataFormat p_format, const Vector<uint8_t> &p_data) { _THREAD_SAFE_METHOD_ uint32_t element_size = get_format_vertex_size(p_format); ERR_FAIL_COND_V_MSG(element_size == 0, RID(), "Format requested is not supported for texture buffers"); uint64_t size_bytes = uint64_t(element_size) * p_size_elements; ERR_FAIL_COND_V(p_data.size() && (uint32_t)p_data.size() != size_bytes, RID()); TextureBuffer texture_buffer; Error err = _buffer_allocate(&texture_buffer.buffer, size_bytes, VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT, VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE, 0); ERR_FAIL_COND_V(err != OK, RID()); if (p_data.size()) { uint64_t data_size = p_data.size(); const uint8_t *r = p_data.ptr(); _buffer_update(&texture_buffer.buffer, 0, r, data_size); _buffer_memory_barrier(texture_buffer.buffer.buffer, 0, data_size, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_SHADER_READ_BIT, false); } VkBufferViewCreateInfo view_create_info; view_create_info.sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO; view_create_info.pNext = nullptr; view_create_info.flags = 0; view_create_info.buffer = texture_buffer.buffer.buffer; view_create_info.format = vulkan_formats[p_format]; view_create_info.offset = 0; view_create_info.range = size_bytes; texture_buffer.view = VK_NULL_HANDLE; VkResult res = vkCreateBufferView(device, &view_create_info, nullptr, &texture_buffer.view); if (res) { _buffer_free(&texture_buffer.buffer); ERR_FAIL_V_MSG(RID(), "Unable to create buffer view, error " + itos(res) + "."); } // Allocate the view. RID id = texture_buffer_owner.make_rid(texture_buffer); #ifdef DEV_ENABLED set_resource_name(id, "RID:" + itos(id.get_id())); #endif return id; } RenderingDeviceVulkan::DescriptorPool *RenderingDeviceVulkan::_descriptor_pool_allocate(const DescriptorPoolKey &p_key) { if (!descriptor_pools.has(p_key)) { descriptor_pools[p_key] = HashSet<DescriptorPool *>(); } DescriptorPool *pool = nullptr; for (DescriptorPool *E : descriptor_pools[p_key]) { if (E->usage < max_descriptors_per_pool) { pool = E; break; } } if (!pool) { // Create a new one. pool = memnew(DescriptorPool); pool->usage = 0; VkDescriptorPoolCreateInfo descriptor_pool_create_info; descriptor_pool_create_info.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO; descriptor_pool_create_info.pNext = nullptr; descriptor_pool_create_info.flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT; // Can't think how somebody may NOT need this flag. descriptor_pool_create_info.maxSets = max_descriptors_per_pool; Vector<VkDescriptorPoolSize> sizes; // Here comes more vulkan API strangeness. if (p_key.uniform_type[UNIFORM_TYPE_SAMPLER]) { VkDescriptorPoolSize s; s.type = VK_DESCRIPTOR_TYPE_SAMPLER; s.descriptorCount = p_key.uniform_type[UNIFORM_TYPE_SAMPLER] * max_descriptors_per_pool; sizes.push_back(s); } if (p_key.uniform_type[UNIFORM_TYPE_SAMPLER_WITH_TEXTURE]) { VkDescriptorPoolSize s; s.type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; s.descriptorCount = p_key.uniform_type[UNIFORM_TYPE_SAMPLER_WITH_TEXTURE] * max_descriptors_per_pool; sizes.push_back(s); } if (p_key.uniform_type[UNIFORM_TYPE_TEXTURE]) { VkDescriptorPoolSize s; s.type = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE; s.descriptorCount = p_key.uniform_type[UNIFORM_TYPE_TEXTURE] * max_descriptors_per_pool; sizes.push_back(s); } if (p_key.uniform_type[UNIFORM_TYPE_IMAGE]) { VkDescriptorPoolSize s; s.type = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE; s.descriptorCount = p_key.uniform_type[UNIFORM_TYPE_IMAGE] * max_descriptors_per_pool; sizes.push_back(s); } if (p_key.uniform_type[UNIFORM_TYPE_TEXTURE_BUFFER] || p_key.uniform_type[UNIFORM_TYPE_SAMPLER_WITH_TEXTURE_BUFFER]) { VkDescriptorPoolSize s; s.type = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; s.descriptorCount = (p_key.uniform_type[UNIFORM_TYPE_TEXTURE_BUFFER] + p_key.uniform_type[UNIFORM_TYPE_SAMPLER_WITH_TEXTURE_BUFFER]) * max_descriptors_per_pool; sizes.push_back(s); } if (p_key.uniform_type[UNIFORM_TYPE_IMAGE_BUFFER]) { VkDescriptorPoolSize s; s.type = VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER; s.descriptorCount = p_key.uniform_type[UNIFORM_TYPE_IMAGE_BUFFER] * max_descriptors_per_pool; sizes.push_back(s); } if (p_key.uniform_type[UNIFORM_TYPE_UNIFORM_BUFFER]) { VkDescriptorPoolSize s; s.type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; s.descriptorCount = p_key.uniform_type[UNIFORM_TYPE_UNIFORM_BUFFER] * max_descriptors_per_pool; sizes.push_back(s); } if (p_key.uniform_type[UNIFORM_TYPE_STORAGE_BUFFER]) { VkDescriptorPoolSize s; s.type = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER; s.descriptorCount = p_key.uniform_type[UNIFORM_TYPE_STORAGE_BUFFER] * max_descriptors_per_pool; sizes.push_back(s); } if (p_key.uniform_type[UNIFORM_TYPE_INPUT_ATTACHMENT]) { VkDescriptorPoolSize s; s.type = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT; s.descriptorCount = p_key.uniform_type[UNIFORM_TYPE_INPUT_ATTACHMENT] * max_descriptors_per_pool; sizes.push_back(s); } descriptor_pool_create_info.poolSizeCount = sizes.size(); descriptor_pool_create_info.pPoolSizes = sizes.ptr(); VkResult res = vkCreateDescriptorPool(device, &descriptor_pool_create_info, nullptr, &pool->pool); if (res) { memdelete(pool); ERR_FAIL_COND_V_MSG(res, nullptr, "vkCreateDescriptorPool failed with error " + itos(res) + "."); } descriptor_pools[p_key].insert(pool); } pool->usage++; return pool; } void RenderingDeviceVulkan::_descriptor_pool_free(const DescriptorPoolKey &p_key, DescriptorPool *p_pool) { #ifdef DEBUG_ENABLED ERR_FAIL_COND(!descriptor_pools[p_key].has(p_pool)); #endif ERR_FAIL_COND(p_pool->usage == 0); p_pool->usage--; if (p_pool->usage == 0) { vkDestroyDescriptorPool(device, p_pool->pool, nullptr); descriptor_pools[p_key].erase(p_pool); memdelete(p_pool); if (descriptor_pools[p_key].is_empty()) { descriptor_pools.erase(p_key); } } } RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, RID p_shader, uint32_t p_shader_set) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V(p_uniforms.size() == 0, RID()); Shader *shader = shader_owner.get_or_null(p_shader); ERR_FAIL_NULL_V(shader, RID()); ERR_FAIL_COND_V_MSG(p_shader_set >= (uint32_t)shader->sets.size() || shader->sets[p_shader_set].uniform_info.size() == 0, RID(), "Desired set (" + itos(p_shader_set) + ") not used by shader."); // See that all sets in shader are satisfied. const Shader::Set &set = shader->sets[p_shader_set]; uint32_t uniform_count = p_uniforms.size(); const Uniform *uniforms = p_uniforms.ptr(); uint32_t set_uniform_count = set.uniform_info.size(); const UniformInfo *set_uniforms = set.uniform_info.ptr(); Vector<VkWriteDescriptorSet> writes; DescriptorPoolKey pool_key; // To keep them alive until update call. List<Vector<VkDescriptorBufferInfo>> buffer_infos; List<Vector<VkBufferView>> buffer_views; List<Vector<VkDescriptorImageInfo>> image_infos; // Used for verification to make sure a uniform set does not use a framebuffer bound texture. LocalVector<UniformSet::AttachableTexture> attachable_textures; Vector<Texture *> mutable_sampled_textures; Vector<Texture *> mutable_storage_textures; for (uint32_t i = 0; i < set_uniform_count; i++) { const UniformInfo &set_uniform = set_uniforms[i]; int uniform_idx = -1; for (int j = 0; j < (int)uniform_count; j++) { if (uniforms[j].binding == set_uniform.binding) { uniform_idx = j; } } ERR_FAIL_COND_V_MSG(uniform_idx == -1, RID(), "All the shader bindings for the given set must be covered by the uniforms provided. Binding (" + itos(set_uniform.binding) + "), set (" + itos(p_shader_set) + ") was not provided."); const Uniform &uniform = uniforms[uniform_idx]; ERR_FAIL_COND_V_MSG(uniform.uniform_type != set_uniform.type, RID(), "Mismatch uniform type for binding (" + itos(set_uniform.binding) + "), set (" + itos(p_shader_set) + "). Expected '" + shader_uniform_names[set_uniform.type] + "', supplied: '" + shader_uniform_names[uniform.uniform_type] + "'."); VkWriteDescriptorSet write; // Common header. write.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET; write.pNext = nullptr; write.dstSet = VK_NULL_HANDLE; // Will assign afterwards when everything is valid. write.dstBinding = set_uniform.binding; write.dstArrayElement = 0; write.descriptorCount = 0; write.descriptorType = VK_DESCRIPTOR_TYPE_MAX_ENUM; // Invalid value. write.pImageInfo = nullptr; write.pBufferInfo = nullptr; write.pTexelBufferView = nullptr; uint32_t type_size = 1; switch (uniform.uniform_type) { case UNIFORM_TYPE_SAMPLER: { if (uniform.get_id_count() != (uint32_t)set_uniform.length) { if (set_uniform.length > 1) { ERR_FAIL_V_MSG(RID(), "Sampler (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") sampler elements, so it should be provided equal number of sampler IDs to satisfy it (IDs provided: " + itos(uniform.get_id_count()) + ")."); } else { ERR_FAIL_V_MSG(RID(), "Sampler (binding: " + itos(uniform.binding) + ") should provide one ID referencing a sampler (IDs provided: " + itos(uniform.get_id_count()) + ")."); } } Vector<VkDescriptorImageInfo> image_info; for (uint32_t j = 0; j < uniform.get_id_count(); j++) { VkSampler *sampler = sampler_owner.get_or_null(uniform.get_id(j)); ERR_FAIL_NULL_V_MSG(sampler, RID(), "Sampler (binding: " + itos(uniform.binding) + ", index " + itos(j) + ") is not a valid sampler."); VkDescriptorImageInfo img_info; img_info.sampler = *sampler; img_info.imageView = VK_NULL_HANDLE; img_info.imageLayout = VK_IMAGE_LAYOUT_UNDEFINED; image_info.push_back(img_info); } write.dstArrayElement = 0; write.descriptorCount = uniform.get_id_count(); write.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLER; write.pImageInfo = image_infos.push_back(image_info)->get().ptr(); write.pBufferInfo = nullptr; write.pTexelBufferView = nullptr; type_size = uniform.get_id_count(); } break; case UNIFORM_TYPE_SAMPLER_WITH_TEXTURE: { if (uniform.get_id_count() != (uint32_t)set_uniform.length * 2) { if (set_uniform.length > 1) { ERR_FAIL_V_MSG(RID(), "SamplerTexture (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") sampler&texture elements, so it should provided twice the amount of IDs (sampler,texture pairs) to satisfy it (IDs provided: " + itos(uniform.get_id_count()) + ")."); } else { ERR_FAIL_V_MSG(RID(), "SamplerTexture (binding: " + itos(uniform.binding) + ") should provide two IDs referencing a sampler and then a texture (IDs provided: " + itos(uniform.get_id_count()) + ")."); } } Vector<VkDescriptorImageInfo> image_info; for (uint32_t j = 0; j < uniform.get_id_count(); j += 2) { VkSampler *sampler = sampler_owner.get_or_null(uniform.get_id(j + 0)); ERR_FAIL_NULL_V_MSG(sampler, RID(), "SamplerBuffer (binding: " + itos(uniform.binding) + ", index " + itos(j + 1) + ") is not a valid sampler."); Texture *texture = texture_owner.get_or_null(uniform.get_id(j + 1)); ERR_FAIL_NULL_V_MSG(texture, RID(), "Texture (binding: " + itos(uniform.binding) + ", index " + itos(j) + ") is not a valid texture."); ERR_FAIL_COND_V_MSG(!(texture->usage_flags & TEXTURE_USAGE_SAMPLING_BIT), RID(), "Texture (binding: " + itos(uniform.binding) + ", index " + itos(j) + ") needs the TEXTURE_USAGE_SAMPLING_BIT usage flag set in order to be used as uniform."); VkDescriptorImageInfo img_info; img_info.sampler = *sampler; img_info.imageView = texture->view; if (texture->usage_flags & (TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | TEXTURE_USAGE_INPUT_ATTACHMENT_BIT)) { UniformSet::AttachableTexture attachable_texture; attachable_texture.bind = set_uniform.binding; attachable_texture.texture = texture->owner.is_valid() ? texture->owner : uniform.get_id(j + 1); attachable_textures.push_back(attachable_texture); } if (texture->usage_flags & TEXTURE_USAGE_STORAGE_BIT) { // Can also be used as storage, add to mutable sampled. mutable_sampled_textures.push_back(texture); } DEV_ASSERT(!texture->owner.is_valid() || texture_owner.get_or_null(texture->owner)); img_info.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; image_info.push_back(img_info); } write.dstArrayElement = 0; write.descriptorCount = uniform.get_id_count() / 2; write.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; write.pImageInfo = image_infos.push_back(image_info)->get().ptr(); write.pBufferInfo = nullptr; write.pTexelBufferView = nullptr; type_size = uniform.get_id_count() / 2; } break; case UNIFORM_TYPE_TEXTURE: { if (uniform.get_id_count() != (uint32_t)set_uniform.length) { if (set_uniform.length > 1) { ERR_FAIL_V_MSG(RID(), "Texture (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") textures, so it should be provided equal number of texture IDs to satisfy it (IDs provided: " + itos(uniform.get_id_count()) + ")."); } else { ERR_FAIL_V_MSG(RID(), "Texture (binding: " + itos(uniform.binding) + ") should provide one ID referencing a texture (IDs provided: " + itos(uniform.get_id_count()) + ")."); } } Vector<VkDescriptorImageInfo> image_info; for (uint32_t j = 0; j < uniform.get_id_count(); j++) { Texture *texture = texture_owner.get_or_null(uniform.get_id(j)); ERR_FAIL_NULL_V_MSG(texture, RID(), "Texture (binding: " + itos(uniform.binding) + ", index " + itos(j) + ") is not a valid texture."); ERR_FAIL_COND_V_MSG(!(texture->usage_flags & TEXTURE_USAGE_SAMPLING_BIT), RID(), "Texture (binding: " + itos(uniform.binding) + ", index " + itos(j) + ") needs the TEXTURE_USAGE_SAMPLING_BIT usage flag set in order to be used as uniform."); VkDescriptorImageInfo img_info; img_info.sampler = VK_NULL_HANDLE; img_info.imageView = texture->view; if (texture->usage_flags & (TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | TEXTURE_USAGE_INPUT_ATTACHMENT_BIT)) { UniformSet::AttachableTexture attachable_texture; attachable_texture.bind = set_uniform.binding; attachable_texture.texture = texture->owner.is_valid() ? texture->owner : uniform.get_id(j); attachable_textures.push_back(attachable_texture); } if (texture->usage_flags & TEXTURE_USAGE_STORAGE_BIT) { // Can also be used as storage, add to mutable sampled. mutable_sampled_textures.push_back(texture); } DEV_ASSERT(!texture->owner.is_valid() || texture_owner.get_or_null(texture->owner)); img_info.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; image_info.push_back(img_info); } write.dstArrayElement = 0; write.descriptorCount = uniform.get_id_count(); write.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE; write.pImageInfo = image_infos.push_back(image_info)->get().ptr(); write.pBufferInfo = nullptr; write.pTexelBufferView = nullptr; type_size = uniform.get_id_count(); } break; case UNIFORM_TYPE_IMAGE: { if (uniform.get_id_count() != (uint32_t)set_uniform.length) { if (set_uniform.length > 1) { ERR_FAIL_V_MSG(RID(), "Image (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") textures, so it should be provided equal number of texture IDs to satisfy it (IDs provided: " + itos(uniform.get_id_count()) + ")."); } else { ERR_FAIL_V_MSG(RID(), "Image (binding: " + itos(uniform.binding) + ") should provide one ID referencing a texture (IDs provided: " + itos(uniform.get_id_count()) + ")."); } } Vector<VkDescriptorImageInfo> image_info; for (uint32_t j = 0; j < uniform.get_id_count(); j++) { Texture *texture = texture_owner.get_or_null(uniform.get_id(j)); ERR_FAIL_NULL_V_MSG(texture, RID(), "Image (binding: " + itos(uniform.binding) + ", index " + itos(j) + ") is not a valid texture."); ERR_FAIL_COND_V_MSG(!(texture->usage_flags & TEXTURE_USAGE_STORAGE_BIT), RID(), "Image (binding: " + itos(uniform.binding) + ", index " + itos(j) + ") needs the TEXTURE_USAGE_STORAGE_BIT usage flag set in order to be used as uniform."); VkDescriptorImageInfo img_info; img_info.sampler = VK_NULL_HANDLE; img_info.imageView = texture->view; if (texture->usage_flags & TEXTURE_USAGE_SAMPLING_BIT) { // Can also be used as storage, add to mutable sampled. mutable_storage_textures.push_back(texture); } DEV_ASSERT(!texture->owner.is_valid() || texture_owner.get_or_null(texture->owner)); img_info.imageLayout = VK_IMAGE_LAYOUT_GENERAL; image_info.push_back(img_info); } write.dstArrayElement = 0; write.descriptorCount = uniform.get_id_count(); write.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE; write.pImageInfo = image_infos.push_back(image_info)->get().ptr(); write.pBufferInfo = nullptr; write.pTexelBufferView = nullptr; type_size = uniform.get_id_count(); } break; case UNIFORM_TYPE_TEXTURE_BUFFER: { if (uniform.get_id_count() != (uint32_t)set_uniform.length) { if (set_uniform.length > 1) { ERR_FAIL_V_MSG(RID(), "Buffer (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") texture buffer elements, so it should be provided equal number of texture buffer IDs to satisfy it (IDs provided: " + itos(uniform.get_id_count()) + ")."); } else { ERR_FAIL_V_MSG(RID(), "Buffer (binding: " + itos(uniform.binding) + ") should provide one ID referencing a texture buffer (IDs provided: " + itos(uniform.get_id_count()) + ")."); } } Vector<VkDescriptorBufferInfo> buffer_info; Vector<VkBufferView> buffer_view; for (uint32_t j = 0; j < uniform.get_id_count(); j++) { TextureBuffer *buffer = texture_buffer_owner.get_or_null(uniform.get_id(j)); ERR_FAIL_NULL_V_MSG(buffer, RID(), "Texture Buffer (binding: " + itos(uniform.binding) + ", index " + itos(j) + ") is not a valid texture buffer."); buffer_info.push_back(buffer->buffer.buffer_info); buffer_view.push_back(buffer->view); } write.dstArrayElement = 0; write.descriptorCount = uniform.get_id_count(); write.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; write.pImageInfo = nullptr; write.pBufferInfo = buffer_infos.push_back(buffer_info)->get().ptr(); write.pTexelBufferView = buffer_views.push_back(buffer_view)->get().ptr(); type_size = uniform.get_id_count(); } break; case UNIFORM_TYPE_SAMPLER_WITH_TEXTURE_BUFFER: { if (uniform.get_id_count() != (uint32_t)set_uniform.length * 2) { if (set_uniform.length > 1) { ERR_FAIL_V_MSG(RID(), "SamplerBuffer (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") sampler buffer elements, so it should provided twice the amount of IDs (sampler,buffer pairs) to satisfy it (IDs provided: " + itos(uniform.get_id_count()) + ")."); } else { ERR_FAIL_V_MSG(RID(), "SamplerBuffer (binding: " + itos(uniform.binding) + ") should provide two IDs referencing a sampler and then a texture buffer (IDs provided: " + itos(uniform.get_id_count()) + ")."); } } Vector<VkDescriptorImageInfo> image_info; Vector<VkDescriptorBufferInfo> buffer_info; Vector<VkBufferView> buffer_view; for (uint32_t j = 0; j < uniform.get_id_count(); j += 2) { VkSampler *sampler = sampler_owner.get_or_null(uniform.get_id(j + 0)); ERR_FAIL_NULL_V_MSG(sampler, RID(), "SamplerBuffer (binding: " + itos(uniform.binding) + ", index " + itos(j + 1) + ") is not a valid sampler."); TextureBuffer *buffer = texture_buffer_owner.get_or_null(uniform.get_id(j + 1)); VkDescriptorImageInfo img_info; img_info.sampler = *sampler; img_info.imageView = VK_NULL_HANDLE; img_info.imageLayout = VK_IMAGE_LAYOUT_UNDEFINED; image_info.push_back(img_info); ERR_FAIL_NULL_V_MSG(buffer, RID(), "SamplerBuffer (binding: " + itos(uniform.binding) + ", index " + itos(j + 1) + ") is not a valid texture buffer."); buffer_info.push_back(buffer->buffer.buffer_info); buffer_view.push_back(buffer->view); } write.dstArrayElement = 0; write.descriptorCount = uniform.get_id_count() / 2; write.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; write.pImageInfo = image_infos.push_back(image_info)->get().ptr(); write.pBufferInfo = buffer_infos.push_back(buffer_info)->get().ptr(); write.pTexelBufferView = buffer_views.push_back(buffer_view)->get().ptr(); type_size = uniform.get_id_count() / 2; } break; case UNIFORM_TYPE_IMAGE_BUFFER: { // Todo. } break; case UNIFORM_TYPE_UNIFORM_BUFFER: { ERR_FAIL_COND_V_MSG(uniform.get_id_count() != 1, RID(), "Uniform buffer supplied (binding: " + itos(uniform.binding) + ") must provide one ID (" + itos(uniform.get_id_count()) + " provided)."); Buffer *buffer = uniform_buffer_owner.get_or_null(uniform.get_id(0)); ERR_FAIL_NULL_V_MSG(buffer, RID(), "Uniform buffer supplied (binding: " + itos(uniform.binding) + ") is invalid."); ERR_FAIL_COND_V_MSG(buffer->size != (uint32_t)set_uniform.length, RID(), "Uniform buffer supplied (binding: " + itos(uniform.binding) + ") size (" + itos(buffer->size) + " does not match size of shader uniform: (" + itos(set_uniform.length) + ")."); write.dstArrayElement = 0; write.descriptorCount = 1; write.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; write.pImageInfo = nullptr; write.pBufferInfo = &buffer->buffer_info; write.pTexelBufferView = nullptr; } break; case UNIFORM_TYPE_STORAGE_BUFFER: { ERR_FAIL_COND_V_MSG(uniform.get_id_count() != 1, RID(), "Storage buffer supplied (binding: " + itos(uniform.binding) + ") must provide one ID (" + itos(uniform.get_id_count()) + " provided)."); Buffer *buffer = nullptr; if (storage_buffer_owner.owns(uniform.get_id(0))) { buffer = storage_buffer_owner.get_or_null(uniform.get_id(0)); } else if (vertex_buffer_owner.owns(uniform.get_id(0))) { buffer = vertex_buffer_owner.get_or_null(uniform.get_id(0)); ERR_FAIL_COND_V_MSG(!(buffer->usage & VK_BUFFER_USAGE_STORAGE_BUFFER_BIT), RID(), "Vertex buffer supplied (binding: " + itos(uniform.binding) + ") was not created with storage flag."); } ERR_FAIL_NULL_V_MSG(buffer, RID(), "Storage buffer supplied (binding: " + itos(uniform.binding) + ") is invalid."); // If 0, then it's sized on link time. ERR_FAIL_COND_V_MSG(set_uniform.length > 0 && buffer->size != (uint32_t)set_uniform.length, RID(), "Storage buffer supplied (binding: " + itos(uniform.binding) + ") size (" + itos(buffer->size) + " does not match size of shader uniform: (" + itos(set_uniform.length) + ")."); write.dstArrayElement = 0; write.descriptorCount = 1; write.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER; write.pImageInfo = nullptr; write.pBufferInfo = &buffer->buffer_info; write.pTexelBufferView = nullptr; } break; case UNIFORM_TYPE_INPUT_ATTACHMENT: { ERR_FAIL_COND_V_MSG(shader->is_compute, RID(), "InputAttachment (binding: " + itos(uniform.binding) + ") supplied for compute shader (this is not allowed)."); if (uniform.get_id_count() != (uint32_t)set_uniform.length) { if (set_uniform.length > 1) { ERR_FAIL_V_MSG(RID(), "InputAttachment (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") textures, so it should be provided equal number of texture IDs to satisfy it (IDs provided: " + itos(uniform.get_id_count()) + ")."); } else { ERR_FAIL_V_MSG(RID(), "InputAttachment (binding: " + itos(uniform.binding) + ") should provide one ID referencing a texture (IDs provided: " + itos(uniform.get_id_count()) + ")."); } } Vector<VkDescriptorImageInfo> image_info; for (uint32_t j = 0; j < uniform.get_id_count(); j++) { Texture *texture = texture_owner.get_or_null(uniform.get_id(j)); ERR_FAIL_NULL_V_MSG(texture, RID(), "InputAttachment (binding: " + itos(uniform.binding) + ", index " + itos(j) + ") is not a valid texture."); ERR_FAIL_COND_V_MSG(!(texture->usage_flags & TEXTURE_USAGE_SAMPLING_BIT), RID(), "InputAttachment (binding: " + itos(uniform.binding) + ", index " + itos(j) + ") needs the TEXTURE_USAGE_SAMPLING_BIT usage flag set in order to be used as uniform."); VkDescriptorImageInfo img_info; img_info.sampler = VK_NULL_HANDLE; img_info.imageView = texture->view; DEV_ASSERT(!texture->owner.is_valid() || texture_owner.get_or_null(texture->owner)); img_info.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; image_info.push_back(img_info); } write.dstArrayElement = 0; write.descriptorCount = uniform.get_id_count(); write.descriptorType = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT; write.pImageInfo = image_infos.push_back(image_info)->get().ptr(); write.pBufferInfo = nullptr; write.pTexelBufferView = nullptr; type_size = uniform.get_id_count(); } break; default: { } } writes.push_back(write); ERR_FAIL_COND_V_MSG(pool_key.uniform_type[set_uniform.type] == MAX_DESCRIPTOR_POOL_ELEMENT, RID(), "Uniform set reached the limit of bindings for the same type (" + itos(MAX_DESCRIPTOR_POOL_ELEMENT) + ")."); pool_key.uniform_type[set_uniform.type] += type_size; } // Need a descriptor pool. DescriptorPool *pool = _descriptor_pool_allocate(pool_key); ERR_FAIL_NULL_V(pool, RID()); VkDescriptorSetAllocateInfo descriptor_set_allocate_info; descriptor_set_allocate_info.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO; descriptor_set_allocate_info.pNext = nullptr; descriptor_set_allocate_info.descriptorPool = pool->pool; descriptor_set_allocate_info.descriptorSetCount = 1; descriptor_set_allocate_info.pSetLayouts = &shader->sets[p_shader_set].descriptor_set_layout; VkDescriptorSet descriptor_set; VkResult res = vkAllocateDescriptorSets(device, &descriptor_set_allocate_info, &descriptor_set); if (res) { _descriptor_pool_free(pool_key, pool); // Meh. ERR_FAIL_V_MSG(RID(), "Cannot allocate descriptor sets, error " + itos(res) + "."); } UniformSet uniform_set; uniform_set.pool = pool; uniform_set.pool_key = pool_key; uniform_set.descriptor_set = descriptor_set; uniform_set.format = shader->set_formats[p_shader_set]; uniform_set.attachable_textures = attachable_textures; uniform_set.mutable_sampled_textures = mutable_sampled_textures; uniform_set.mutable_storage_textures = mutable_storage_textures; uniform_set.shader_set = p_shader_set; uniform_set.shader_id = p_shader; RID id = uniform_set_owner.make_rid(uniform_set); #ifdef DEV_ENABLED set_resource_name(id, "RID:" + itos(id.get_id())); #endif // Add dependencies. _add_dependency(id, p_shader); for (uint32_t i = 0; i < uniform_count; i++) { const Uniform &uniform = uniforms[i]; int id_count = uniform.get_id_count(); for (int j = 0; j < id_count; j++) { _add_dependency(id, uniform.get_id(j)); } } // Write the contents. if (writes.size()) { for (int i = 0; i < writes.size(); i++) { writes.write[i].dstSet = descriptor_set; } vkUpdateDescriptorSets(device, writes.size(), writes.ptr(), 0, nullptr); } return id; } bool RenderingDeviceVulkan::uniform_set_is_valid(RID p_uniform_set) { return uniform_set_owner.owns(p_uniform_set); } void RenderingDeviceVulkan::uniform_set_set_invalidation_callback(RID p_uniform_set, InvalidationCallback p_callback, void *p_userdata) { UniformSet *us = uniform_set_owner.get_or_null(p_uniform_set); ERR_FAIL_NULL(us); us->invalidated_callback = p_callback; us->invalidated_callback_userdata = p_userdata; } Error RenderingDeviceVulkan::buffer_copy(RID p_src_buffer, RID p_dst_buffer, uint32_t p_src_offset, uint32_t p_dst_offset, uint32_t p_size, BitField<BarrierMask> p_post_barrier) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG(draw_list, ERR_INVALID_PARAMETER, "Copying buffers is forbidden during creation of a draw list"); ERR_FAIL_COND_V_MSG(compute_list, ERR_INVALID_PARAMETER, "Copying buffers is forbidden during creation of a compute list"); // This method assumes the barriers have been pushed prior to being called, therefore no barriers are pushed // for the source or destination buffers before performing the copy. These masks are effectively ignored. VkPipelineStageFlags src_stage_mask = 0; VkAccessFlags src_access_mask = 0; Buffer *src_buffer = _get_buffer_from_owner(p_src_buffer, src_stage_mask, src_access_mask, BARRIER_MASK_NO_BARRIER); if (!src_buffer) { ERR_FAIL_V_MSG(ERR_INVALID_PARAMETER, "Source buffer argument is not a valid buffer of any type."); } VkPipelineStageFlags dst_stage_mask = 0; VkAccessFlags dst_access = 0; if (p_post_barrier.has_flag(BARRIER_MASK_TRANSFER)) { // If the post barrier mask defines it, we indicate the destination buffer will require a barrier with these flags set // after the copy command is queued. dst_stage_mask = VK_PIPELINE_STAGE_TRANSFER_BIT; dst_access = VK_ACCESS_TRANSFER_WRITE_BIT; } Buffer *dst_buffer = _get_buffer_from_owner(p_dst_buffer, dst_stage_mask, dst_access, p_post_barrier); if (!dst_buffer) { ERR_FAIL_V_MSG(ERR_INVALID_PARAMETER, "Destination buffer argument is not a valid buffer of any type."); } // Validate the copy's dimensions for both buffers. ERR_FAIL_COND_V_MSG((p_size + p_src_offset) > src_buffer->size, ERR_INVALID_PARAMETER, "Size is larger than the source buffer."); ERR_FAIL_COND_V_MSG((p_size + p_dst_offset) > dst_buffer->size, ERR_INVALID_PARAMETER, "Size is larger than the destination buffer."); // Perform the copy. VkBufferCopy region; region.srcOffset = p_src_offset; region.dstOffset = p_dst_offset; region.size = p_size; vkCmdCopyBuffer(frames[frame].draw_command_buffer, src_buffer->buffer, dst_buffer->buffer, 1, ®ion); #ifdef FORCE_FULL_BARRIER _full_barrier(true); #else if (dst_stage_mask == 0) { dst_stage_mask = VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT; } // As indicated by the post barrier mask, push a new barrier. if (p_post_barrier != RD::BARRIER_MASK_NO_BARRIER) { _buffer_memory_barrier(dst_buffer->buffer, p_dst_offset, p_size, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stage_mask, VK_ACCESS_TRANSFER_WRITE_BIT, dst_access, true); } #endif return OK; } Error RenderingDeviceVulkan::buffer_update(RID p_buffer, uint32_t p_offset, uint32_t p_size, const void *p_data, BitField<BarrierMask> p_post_barrier) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG(draw_list, ERR_INVALID_PARAMETER, "Updating buffers is forbidden during creation of a draw list"); ERR_FAIL_COND_V_MSG(compute_list, ERR_INVALID_PARAMETER, "Updating buffers is forbidden during creation of a compute list"); VkPipelineStageFlags dst_stage_mask = 0; VkAccessFlags dst_access = 0; if (p_post_barrier.has_flag(BARRIER_MASK_TRANSFER)) { // Protect subsequent updates. dst_stage_mask = VK_PIPELINE_STAGE_TRANSFER_BIT; dst_access = VK_ACCESS_TRANSFER_WRITE_BIT; } Buffer *buffer = _get_buffer_from_owner(p_buffer, dst_stage_mask, dst_access, p_post_barrier); if (!buffer) { ERR_FAIL_V_MSG(ERR_INVALID_PARAMETER, "Buffer argument is not a valid buffer of any type."); } ERR_FAIL_COND_V_MSG(p_offset + p_size > buffer->size, ERR_INVALID_PARAMETER, "Attempted to write buffer (" + itos((p_offset + p_size) - buffer->size) + " bytes) past the end."); // No barrier should be needed here. // _buffer_memory_barrier(buffer->buffer, p_offset, p_size, dst_stage_mask, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_access, VK_ACCESS_TRANSFER_WRITE_BIT, true); Error err = _buffer_update(buffer, p_offset, (uint8_t *)p_data, p_size, p_post_barrier); if (err) { return err; } #ifdef FORCE_FULL_BARRIER _full_barrier(true); #else if (dst_stage_mask == 0) { dst_stage_mask = VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT; } if (p_post_barrier != RD::BARRIER_MASK_NO_BARRIER) { _buffer_memory_barrier(buffer->buffer, p_offset, p_size, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stage_mask, VK_ACCESS_TRANSFER_WRITE_BIT, dst_access, true); } #endif return err; } Error RenderingDeviceVulkan::buffer_clear(RID p_buffer, uint32_t p_offset, uint32_t p_size, BitField<BarrierMask> p_post_barrier) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG((p_size % 4) != 0, ERR_INVALID_PARAMETER, "Size must be a multiple of four"); ERR_FAIL_COND_V_MSG(draw_list, ERR_INVALID_PARAMETER, "Updating buffers in is forbidden during creation of a draw list"); ERR_FAIL_COND_V_MSG(compute_list, ERR_INVALID_PARAMETER, "Updating buffers is forbidden during creation of a compute list"); VkPipelineStageFlags dst_stage_mask = 0; VkAccessFlags dst_access = 0; if (p_post_barrier.has_flag(BARRIER_MASK_TRANSFER)) { // Protect subsequent updates. dst_stage_mask = VK_PIPELINE_STAGE_TRANSFER_BIT; dst_access = VK_ACCESS_TRANSFER_WRITE_BIT; } Buffer *buffer = _get_buffer_from_owner(p_buffer, dst_stage_mask, dst_access, p_post_barrier); if (!buffer) { ERR_FAIL_V_MSG(ERR_INVALID_PARAMETER, "Buffer argument is not a valid buffer of any type."); } ERR_FAIL_COND_V_MSG(p_offset + p_size > buffer->size, ERR_INVALID_PARAMETER, "Attempted to write buffer (" + itos((p_offset + p_size) - buffer->size) + " bytes) past the end."); vkCmdFillBuffer(frames[frame].draw_command_buffer, buffer->buffer, p_offset, p_size, 0); #ifdef FORCE_FULL_BARRIER _full_barrier(true); #else if (dst_stage_mask == 0) { dst_stage_mask = VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT; } _buffer_memory_barrier(buffer->buffer, p_offset, p_size, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stage_mask, VK_ACCESS_TRANSFER_WRITE_BIT, dst_access, true); #endif return OK; } Vector<uint8_t> RenderingDeviceVulkan::buffer_get_data(RID p_buffer, uint32_t p_offset, uint32_t p_size) { _THREAD_SAFE_METHOD_ // It could be this buffer was just created. VkPipelineShaderStageCreateFlags src_stage_mask = VK_PIPELINE_STAGE_TRANSFER_BIT; VkAccessFlags src_access_mask = VK_ACCESS_TRANSFER_WRITE_BIT; // Get the vulkan buffer and the potential stage/access possible. Buffer *buffer = _get_buffer_from_owner(p_buffer, src_stage_mask, src_access_mask, BARRIER_MASK_ALL_BARRIERS); if (!buffer) { ERR_FAIL_V_MSG(Vector<uint8_t>(), "Buffer is either invalid or this type of buffer can't be retrieved. Only Index and Vertex buffers allow retrieving."); } // Make sure no one is using the buffer -- the "true" gets us to the same command buffer as below. _buffer_memory_barrier(buffer->buffer, 0, buffer->size, src_stage_mask, VK_PIPELINE_STAGE_TRANSFER_BIT, src_access_mask, VK_ACCESS_TRANSFER_READ_BIT, true); VkCommandBuffer command_buffer = frames[frame].draw_command_buffer; // Size of buffer to retrieve. if (!p_size) { p_size = buffer->size; } else { ERR_FAIL_COND_V_MSG(p_size + p_offset > buffer->size, Vector<uint8_t>(), "Size is larger than the buffer."); } Buffer tmp_buffer; _buffer_allocate(&tmp_buffer, p_size, VK_BUFFER_USAGE_TRANSFER_DST_BIT, VMA_MEMORY_USAGE_AUTO_PREFER_HOST, VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT); VkBufferCopy region; region.srcOffset = p_offset; region.dstOffset = 0; region.size = p_size; vkCmdCopyBuffer(command_buffer, buffer->buffer, tmp_buffer.buffer, 1, ®ion); // Dst buffer is in CPU, but I wonder if src buffer needs a barrier for this. // Flush everything so memory can be safely mapped. _flush(true); void *buffer_mem; VkResult vkerr = vmaMapMemory(allocator, tmp_buffer.allocation, &buffer_mem); ERR_FAIL_COND_V_MSG(vkerr, Vector<uint8_t>(), "vmaMapMemory failed with error " + itos(vkerr) + "."); Vector<uint8_t> buffer_data; { buffer_data.resize(p_size); uint8_t *w = buffer_data.ptrw(); memcpy(w, buffer_mem, p_size); } vmaUnmapMemory(allocator, tmp_buffer.allocation); _buffer_free(&tmp_buffer); return buffer_data; } /*************************/ /**** RENDER PIPELINE ****/ /*************************/ RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferFormatID p_framebuffer_format, VertexFormatID p_vertex_format, RenderPrimitive p_render_primitive, const PipelineRasterizationState &p_rasterization_state, const PipelineMultisampleState &p_multisample_state, const PipelineDepthStencilState &p_depth_stencil_state, const PipelineColorBlendState &p_blend_state, BitField<PipelineDynamicStateFlags> p_dynamic_state_flags, uint32_t p_for_render_pass, const Vector<PipelineSpecializationConstant> &p_specialization_constants) { _THREAD_SAFE_METHOD_ // Needs a shader. Shader *shader = shader_owner.get_or_null(p_shader); ERR_FAIL_NULL_V(shader, RID()); ERR_FAIL_COND_V_MSG(shader->is_compute, RID(), "Compute shaders can't be used in render pipelines"); if (p_framebuffer_format == INVALID_ID) { // If nothing provided, use an empty one (no attachments). p_framebuffer_format = framebuffer_format_create(Vector<AttachmentFormat>()); } ERR_FAIL_COND_V(!framebuffer_formats.has(p_framebuffer_format), RID()); const FramebufferFormat &fb_format = framebuffer_formats[p_framebuffer_format]; { // Validate shader vs framebuffer. ERR_FAIL_COND_V_MSG(p_for_render_pass >= uint32_t(fb_format.E->key().passes.size()), RID(), "Render pass requested for pipeline creation (" + itos(p_for_render_pass) + ") is out of bounds"); const FramebufferPass &pass = fb_format.E->key().passes[p_for_render_pass]; uint32_t output_mask = 0; for (int i = 0; i < pass.color_attachments.size(); i++) { if (pass.color_attachments[i] != FramebufferPass::ATTACHMENT_UNUSED) { output_mask |= 1 << i; } } ERR_FAIL_COND_V_MSG(shader->fragment_output_mask != output_mask, RID(), "Mismatch fragment shader output mask (" + itos(shader->fragment_output_mask) + ") and framebuffer color output mask (" + itos(output_mask) + ") when binding both in render pipeline."); } // Vertex. VkPipelineVertexInputStateCreateInfo pipeline_vertex_input_state_create_info; if (p_vertex_format != INVALID_ID) { // Uses vertices, else it does not. ERR_FAIL_COND_V(!vertex_formats.has(p_vertex_format), RID()); const VertexDescriptionCache &vd = vertex_formats[p_vertex_format]; pipeline_vertex_input_state_create_info = vd.create_info; // Validate with inputs. for (uint32_t i = 0; i < 32; i++) { if (!(shader->vertex_input_mask & (1UL << i))) { continue; } bool found = false; for (int j = 0; j < vd.vertex_formats.size(); j++) { if (vd.vertex_formats[j].location == i) { found = true; } } ERR_FAIL_COND_V_MSG(!found, RID(), "Shader vertex input location (" + itos(i) + ") not provided in vertex input description for pipeline creation."); } } else { // Does not use vertices. pipeline_vertex_input_state_create_info.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO; pipeline_vertex_input_state_create_info.pNext = nullptr; pipeline_vertex_input_state_create_info.flags = 0; pipeline_vertex_input_state_create_info.vertexBindingDescriptionCount = 0; pipeline_vertex_input_state_create_info.pVertexBindingDescriptions = nullptr; pipeline_vertex_input_state_create_info.vertexAttributeDescriptionCount = 0; pipeline_vertex_input_state_create_info.pVertexAttributeDescriptions = nullptr; ERR_FAIL_COND_V_MSG(shader->vertex_input_mask != 0, RID(), "Shader contains vertex inputs, but no vertex input description was provided for pipeline creation."); } // Input assembly. ERR_FAIL_INDEX_V(p_render_primitive, RENDER_PRIMITIVE_MAX, RID()); VkPipelineInputAssemblyStateCreateInfo input_assembly_create_info; input_assembly_create_info.sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO; input_assembly_create_info.pNext = nullptr; input_assembly_create_info.flags = 0; static const VkPrimitiveTopology topology_list[RENDER_PRIMITIVE_MAX] = { VK_PRIMITIVE_TOPOLOGY_POINT_LIST, VK_PRIMITIVE_TOPOLOGY_LINE_LIST, VK_PRIMITIVE_TOPOLOGY_LINE_LIST_WITH_ADJACENCY, VK_PRIMITIVE_TOPOLOGY_LINE_STRIP, VK_PRIMITIVE_TOPOLOGY_LINE_STRIP_WITH_ADJACENCY, VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST, VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY, VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP, VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP_WITH_ADJACENCY, VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP, VK_PRIMITIVE_TOPOLOGY_PATCH_LIST }; input_assembly_create_info.topology = topology_list[p_render_primitive]; input_assembly_create_info.primitiveRestartEnable = (p_render_primitive == RENDER_PRIMITIVE_TRIANGLE_STRIPS_WITH_RESTART_INDEX); // Tessellation. VkPipelineTessellationStateCreateInfo tessellation_create_info; tessellation_create_info.sType = VK_STRUCTURE_TYPE_PIPELINE_TESSELLATION_STATE_CREATE_INFO; tessellation_create_info.pNext = nullptr; tessellation_create_info.flags = 0; ERR_FAIL_COND_V(limits.maxTessellationPatchSize > 0 && (p_rasterization_state.patch_control_points < 1 || p_rasterization_state.patch_control_points > limits.maxTessellationPatchSize), RID()); tessellation_create_info.patchControlPoints = p_rasterization_state.patch_control_points; VkPipelineViewportStateCreateInfo viewport_state_create_info; viewport_state_create_info.sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO; viewport_state_create_info.pNext = nullptr; viewport_state_create_info.flags = 0; viewport_state_create_info.viewportCount = 1; // If VR extensions are supported at some point, this will have to be customizable in the framebuffer format. viewport_state_create_info.pViewports = nullptr; viewport_state_create_info.scissorCount = 1; viewport_state_create_info.pScissors = nullptr; // Rasterization. VkPipelineRasterizationStateCreateInfo rasterization_state_create_info; rasterization_state_create_info.sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO; rasterization_state_create_info.pNext = nullptr; rasterization_state_create_info.flags = 0; rasterization_state_create_info.depthClampEnable = p_rasterization_state.enable_depth_clamp; rasterization_state_create_info.rasterizerDiscardEnable = p_rasterization_state.discard_primitives; rasterization_state_create_info.polygonMode = (p_rasterization_state.wireframe ? VK_POLYGON_MODE_LINE : VK_POLYGON_MODE_FILL); static const VkCullModeFlags cull_mode[3] = { VK_CULL_MODE_NONE, VK_CULL_MODE_FRONT_BIT, VK_CULL_MODE_BACK_BIT }; ERR_FAIL_INDEX_V(p_rasterization_state.cull_mode, 3, RID()); rasterization_state_create_info.cullMode = cull_mode[p_rasterization_state.cull_mode]; rasterization_state_create_info.frontFace = (p_rasterization_state.front_face == POLYGON_FRONT_FACE_CLOCKWISE ? VK_FRONT_FACE_CLOCKWISE : VK_FRONT_FACE_COUNTER_CLOCKWISE); rasterization_state_create_info.depthBiasEnable = p_rasterization_state.depth_bias_enabled; rasterization_state_create_info.depthBiasConstantFactor = p_rasterization_state.depth_bias_constant_factor; rasterization_state_create_info.depthBiasClamp = p_rasterization_state.depth_bias_clamp; rasterization_state_create_info.depthBiasSlopeFactor = p_rasterization_state.depth_bias_slope_factor; rasterization_state_create_info.lineWidth = p_rasterization_state.line_width; // Multisample. VkPipelineMultisampleStateCreateInfo multisample_state_create_info; multisample_state_create_info.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO; multisample_state_create_info.pNext = nullptr; multisample_state_create_info.flags = 0; multisample_state_create_info.rasterizationSamples = _ensure_supported_sample_count(p_multisample_state.sample_count); multisample_state_create_info.sampleShadingEnable = p_multisample_state.enable_sample_shading; multisample_state_create_info.minSampleShading = p_multisample_state.min_sample_shading; Vector<VkSampleMask> sample_mask; if (p_multisample_state.sample_mask.size()) { // Use sample mask. const int rasterization_sample_mask_expected_size[TEXTURE_SAMPLES_MAX] = { 1, 2, 4, 8, 16, 32, 64 }; ERR_FAIL_COND_V(rasterization_sample_mask_expected_size[p_multisample_state.sample_count] != p_multisample_state.sample_mask.size(), RID()); sample_mask.resize(p_multisample_state.sample_mask.size()); for (int i = 0; i < p_multisample_state.sample_mask.size(); i++) { VkSampleMask mask = p_multisample_state.sample_mask[i]; sample_mask.push_back(mask); } multisample_state_create_info.pSampleMask = sample_mask.ptr(); } else { multisample_state_create_info.pSampleMask = nullptr; } multisample_state_create_info.alphaToCoverageEnable = p_multisample_state.enable_alpha_to_coverage; multisample_state_create_info.alphaToOneEnable = p_multisample_state.enable_alpha_to_one; // Depth stencil. VkPipelineDepthStencilStateCreateInfo depth_stencil_state_create_info; depth_stencil_state_create_info.sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO; depth_stencil_state_create_info.pNext = nullptr; depth_stencil_state_create_info.flags = 0; depth_stencil_state_create_info.depthTestEnable = p_depth_stencil_state.enable_depth_test; depth_stencil_state_create_info.depthWriteEnable = p_depth_stencil_state.enable_depth_write; ERR_FAIL_INDEX_V(p_depth_stencil_state.depth_compare_operator, COMPARE_OP_MAX, RID()); depth_stencil_state_create_info.depthCompareOp = compare_operators[p_depth_stencil_state.depth_compare_operator]; depth_stencil_state_create_info.depthBoundsTestEnable = p_depth_stencil_state.enable_depth_range; depth_stencil_state_create_info.stencilTestEnable = p_depth_stencil_state.enable_stencil; ERR_FAIL_INDEX_V(p_depth_stencil_state.front_op.fail, STENCIL_OP_MAX, RID()); depth_stencil_state_create_info.front.failOp = stencil_operations[p_depth_stencil_state.front_op.fail]; ERR_FAIL_INDEX_V(p_depth_stencil_state.front_op.pass, STENCIL_OP_MAX, RID()); depth_stencil_state_create_info.front.passOp = stencil_operations[p_depth_stencil_state.front_op.pass]; ERR_FAIL_INDEX_V(p_depth_stencil_state.front_op.depth_fail, STENCIL_OP_MAX, RID()); depth_stencil_state_create_info.front.depthFailOp = stencil_operations[p_depth_stencil_state.front_op.depth_fail]; ERR_FAIL_INDEX_V(p_depth_stencil_state.front_op.compare, COMPARE_OP_MAX, RID()); depth_stencil_state_create_info.front.compareOp = compare_operators[p_depth_stencil_state.front_op.compare]; depth_stencil_state_create_info.front.compareMask = p_depth_stencil_state.front_op.compare_mask; depth_stencil_state_create_info.front.writeMask = p_depth_stencil_state.front_op.write_mask; depth_stencil_state_create_info.front.reference = p_depth_stencil_state.front_op.reference; ERR_FAIL_INDEX_V(p_depth_stencil_state.back_op.fail, STENCIL_OP_MAX, RID()); depth_stencil_state_create_info.back.failOp = stencil_operations[p_depth_stencil_state.back_op.fail]; ERR_FAIL_INDEX_V(p_depth_stencil_state.back_op.pass, STENCIL_OP_MAX, RID()); depth_stencil_state_create_info.back.passOp = stencil_operations[p_depth_stencil_state.back_op.pass]; ERR_FAIL_INDEX_V(p_depth_stencil_state.back_op.depth_fail, STENCIL_OP_MAX, RID()); depth_stencil_state_create_info.back.depthFailOp = stencil_operations[p_depth_stencil_state.back_op.depth_fail]; ERR_FAIL_INDEX_V(p_depth_stencil_state.back_op.compare, COMPARE_OP_MAX, RID()); depth_stencil_state_create_info.back.compareOp = compare_operators[p_depth_stencil_state.back_op.compare]; depth_stencil_state_create_info.back.compareMask = p_depth_stencil_state.back_op.compare_mask; depth_stencil_state_create_info.back.writeMask = p_depth_stencil_state.back_op.write_mask; depth_stencil_state_create_info.back.reference = p_depth_stencil_state.back_op.reference; depth_stencil_state_create_info.minDepthBounds = p_depth_stencil_state.depth_range_min; depth_stencil_state_create_info.maxDepthBounds = p_depth_stencil_state.depth_range_max; // Blend state. VkPipelineColorBlendStateCreateInfo color_blend_state_create_info; color_blend_state_create_info.sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO; color_blend_state_create_info.pNext = nullptr; color_blend_state_create_info.flags = 0; color_blend_state_create_info.logicOpEnable = p_blend_state.enable_logic_op; ERR_FAIL_INDEX_V(p_blend_state.logic_op, LOGIC_OP_MAX, RID()); color_blend_state_create_info.logicOp = logic_operations[p_blend_state.logic_op]; Vector<VkPipelineColorBlendAttachmentState> attachment_states; { const FramebufferPass &pass = fb_format.E->key().passes[p_for_render_pass]; attachment_states.resize(pass.color_attachments.size()); ERR_FAIL_COND_V(p_blend_state.attachments.size() < pass.color_attachments.size(), RID()); for (int i = 0; i < pass.color_attachments.size(); i++) { VkPipelineColorBlendAttachmentState state; if (pass.color_attachments[i] == FramebufferPass::ATTACHMENT_UNUSED) { state.blendEnable = false; state.srcColorBlendFactor = VK_BLEND_FACTOR_ZERO; state.dstColorBlendFactor = VK_BLEND_FACTOR_ZERO; state.colorBlendOp = VK_BLEND_OP_ADD; state.srcAlphaBlendFactor = VK_BLEND_FACTOR_ZERO; state.dstAlphaBlendFactor = VK_BLEND_FACTOR_ZERO; state.alphaBlendOp = VK_BLEND_OP_ADD; state.colorWriteMask = 0; } else { state.blendEnable = p_blend_state.attachments[i].enable_blend; ERR_FAIL_INDEX_V(p_blend_state.attachments[i].src_color_blend_factor, BLEND_FACTOR_MAX, RID()); state.srcColorBlendFactor = blend_factors[p_blend_state.attachments[i].src_color_blend_factor]; ERR_FAIL_INDEX_V(p_blend_state.attachments[i].dst_color_blend_factor, BLEND_FACTOR_MAX, RID()); state.dstColorBlendFactor = blend_factors[p_blend_state.attachments[i].dst_color_blend_factor]; ERR_FAIL_INDEX_V(p_blend_state.attachments[i].color_blend_op, BLEND_OP_MAX, RID()); state.colorBlendOp = blend_operations[p_blend_state.attachments[i].color_blend_op]; ERR_FAIL_INDEX_V(p_blend_state.attachments[i].src_alpha_blend_factor, BLEND_FACTOR_MAX, RID()); state.srcAlphaBlendFactor = blend_factors[p_blend_state.attachments[i].src_alpha_blend_factor]; ERR_FAIL_INDEX_V(p_blend_state.attachments[i].dst_alpha_blend_factor, BLEND_FACTOR_MAX, RID()); state.dstAlphaBlendFactor = blend_factors[p_blend_state.attachments[i].dst_alpha_blend_factor]; ERR_FAIL_INDEX_V(p_blend_state.attachments[i].alpha_blend_op, BLEND_OP_MAX, RID()); state.alphaBlendOp = blend_operations[p_blend_state.attachments[i].alpha_blend_op]; state.colorWriteMask = 0; if (p_blend_state.attachments[i].write_r) { state.colorWriteMask |= VK_COLOR_COMPONENT_R_BIT; } if (p_blend_state.attachments[i].write_g) { state.colorWriteMask |= VK_COLOR_COMPONENT_G_BIT; } if (p_blend_state.attachments[i].write_b) { state.colorWriteMask |= VK_COLOR_COMPONENT_B_BIT; } if (p_blend_state.attachments[i].write_a) { state.colorWriteMask |= VK_COLOR_COMPONENT_A_BIT; } } attachment_states.write[i] = state; } } color_blend_state_create_info.attachmentCount = attachment_states.size(); color_blend_state_create_info.pAttachments = attachment_states.ptr(); color_blend_state_create_info.blendConstants[0] = p_blend_state.blend_constant.r; color_blend_state_create_info.blendConstants[1] = p_blend_state.blend_constant.g; color_blend_state_create_info.blendConstants[2] = p_blend_state.blend_constant.b; color_blend_state_create_info.blendConstants[3] = p_blend_state.blend_constant.a; // Dynamic state. VkPipelineDynamicStateCreateInfo dynamic_state_create_info; dynamic_state_create_info.sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO; dynamic_state_create_info.pNext = nullptr; dynamic_state_create_info.flags = 0; Vector<VkDynamicState> dynamic_states; // Vulkan is weird. dynamic_states.push_back(VK_DYNAMIC_STATE_VIEWPORT); // Viewport and scissor are always dynamic. dynamic_states.push_back(VK_DYNAMIC_STATE_SCISSOR); if (p_dynamic_state_flags.has_flag(DYNAMIC_STATE_LINE_WIDTH)) { dynamic_states.push_back(VK_DYNAMIC_STATE_LINE_WIDTH); } if (p_dynamic_state_flags.has_flag(DYNAMIC_STATE_DEPTH_BIAS)) { dynamic_states.push_back(VK_DYNAMIC_STATE_DEPTH_BIAS); } if (p_dynamic_state_flags.has_flag(DYNAMIC_STATE_BLEND_CONSTANTS)) { dynamic_states.push_back(VK_DYNAMIC_STATE_BLEND_CONSTANTS); } if (p_dynamic_state_flags.has_flag(DYNAMIC_STATE_DEPTH_BOUNDS)) { dynamic_states.push_back(VK_DYNAMIC_STATE_DEPTH_BOUNDS); } if (p_dynamic_state_flags.has_flag(DYNAMIC_STATE_STENCIL_COMPARE_MASK)) { dynamic_states.push_back(VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK); } if (p_dynamic_state_flags.has_flag(DYNAMIC_STATE_STENCIL_WRITE_MASK)) { dynamic_states.push_back(VK_DYNAMIC_STATE_STENCIL_WRITE_MASK); } if (p_dynamic_state_flags.has_flag(DYNAMIC_STATE_STENCIL_REFERENCE)) { dynamic_states.push_back(VK_DYNAMIC_STATE_STENCIL_REFERENCE); } dynamic_state_create_info.dynamicStateCount = dynamic_states.size(); dynamic_state_create_info.pDynamicStates = dynamic_states.ptr(); void *graphics_pipeline_nextptr = nullptr; VkPipelineFragmentShadingRateStateCreateInfoKHR vrs_create_info; if (context->get_vrs_capabilities().attachment_vrs_supported) { // If VRS is used, this defines how the different VRS types are combined. // combinerOps[0] decides how we use the output of pipeline and primitive (drawcall) VRS. // combinerOps[1] decides how we use the output of combinerOps[0] and our attachment VRS. vrs_create_info.sType = VK_STRUCTURE_TYPE_PIPELINE_FRAGMENT_SHADING_RATE_STATE_CREATE_INFO_KHR; vrs_create_info.pNext = nullptr; vrs_create_info.fragmentSize = { 4, 4 }; vrs_create_info.combinerOps[0] = VK_FRAGMENT_SHADING_RATE_COMBINER_OP_KEEP_KHR; // We don't use pipeline/primitive VRS so this really doesn't matter. vrs_create_info.combinerOps[1] = VK_FRAGMENT_SHADING_RATE_COMBINER_OP_REPLACE_KHR; // Always use the outcome of attachment VRS if enabled. graphics_pipeline_nextptr = &vrs_create_info; } // Finally, pipeline create info. VkGraphicsPipelineCreateInfo graphics_pipeline_create_info; graphics_pipeline_create_info.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO; graphics_pipeline_create_info.pNext = graphics_pipeline_nextptr; graphics_pipeline_create_info.flags = 0; Vector<VkPipelineShaderStageCreateInfo> pipeline_stages = shader->pipeline_stages; Vector<VkSpecializationInfo> specialization_info; Vector<Vector<VkSpecializationMapEntry>> specialization_map_entries; Vector<uint32_t> specialization_constant_data; if (shader->specialization_constants.size()) { specialization_constant_data.resize(shader->specialization_constants.size()); uint32_t *data_ptr = specialization_constant_data.ptrw(); specialization_info.resize(pipeline_stages.size()); specialization_map_entries.resize(pipeline_stages.size()); for (int i = 0; i < shader->specialization_constants.size(); i++) { // See if overridden. const Shader::SpecializationConstant &sc = shader->specialization_constants[i]; data_ptr[i] = sc.constant.int_value; // Just copy the 32 bits. for (int j = 0; j < p_specialization_constants.size(); j++) { const PipelineSpecializationConstant &psc = p_specialization_constants[j]; if (psc.constant_id == sc.constant.constant_id) { ERR_FAIL_COND_V_MSG(psc.type != sc.constant.type, RID(), "Specialization constant provided for id (" + itos(sc.constant.constant_id) + ") is of the wrong type."); data_ptr[i] = psc.int_value; break; } } VkSpecializationMapEntry entry; entry.constantID = sc.constant.constant_id; entry.offset = i * sizeof(uint32_t); entry.size = sizeof(uint32_t); for (int j = 0; j < SHADER_STAGE_MAX; j++) { if (sc.stage_flags & (1 << j)) { VkShaderStageFlagBits stage = shader_stage_masks[j]; for (int k = 0; k < pipeline_stages.size(); k++) { if (pipeline_stages[k].stage == stage) { specialization_map_entries.write[k].push_back(entry); } } } } } for (int i = 0; i < pipeline_stages.size(); i++) { if (specialization_map_entries[i].size()) { specialization_info.write[i].dataSize = specialization_constant_data.size() * sizeof(uint32_t); specialization_info.write[i].pData = data_ptr; specialization_info.write[i].mapEntryCount = specialization_map_entries[i].size(); specialization_info.write[i].pMapEntries = specialization_map_entries[i].ptr(); pipeline_stages.write[i].pSpecializationInfo = specialization_info.ptr() + i; } } } graphics_pipeline_create_info.stageCount = pipeline_stages.size(); graphics_pipeline_create_info.pStages = pipeline_stages.ptr(); graphics_pipeline_create_info.pVertexInputState = &pipeline_vertex_input_state_create_info; graphics_pipeline_create_info.pInputAssemblyState = &input_assembly_create_info; graphics_pipeline_create_info.pTessellationState = &tessellation_create_info; graphics_pipeline_create_info.pViewportState = &viewport_state_create_info; graphics_pipeline_create_info.pRasterizationState = &rasterization_state_create_info; graphics_pipeline_create_info.pMultisampleState = &multisample_state_create_info; graphics_pipeline_create_info.pDepthStencilState = &depth_stencil_state_create_info; graphics_pipeline_create_info.pColorBlendState = &color_blend_state_create_info; graphics_pipeline_create_info.pDynamicState = &dynamic_state_create_info; graphics_pipeline_create_info.layout = shader->pipeline_layout; graphics_pipeline_create_info.renderPass = fb_format.render_pass; graphics_pipeline_create_info.subpass = p_for_render_pass; graphics_pipeline_create_info.basePipelineHandle = VK_NULL_HANDLE; graphics_pipeline_create_info.basePipelineIndex = 0; RenderPipeline pipeline; VkResult err = vkCreateGraphicsPipelines(device, pipelines_cache.cache_object, 1, &graphics_pipeline_create_info, nullptr, &pipeline.pipeline); ERR_FAIL_COND_V_MSG(err, RID(), "vkCreateGraphicsPipelines failed with error " + itos(err) + " for shader '" + shader->name + "'."); if (pipelines_cache.cache_object != VK_NULL_HANDLE) { _update_pipeline_cache(); } pipeline.set_formats = shader->set_formats; pipeline.push_constant_stages_mask = shader->push_constant.vk_stages_mask; pipeline.pipeline_layout = shader->pipeline_layout; pipeline.shader = p_shader; pipeline.push_constant_size = shader->push_constant.size; #ifdef DEBUG_ENABLED pipeline.validation.dynamic_state = p_dynamic_state_flags; pipeline.validation.framebuffer_format = p_framebuffer_format; pipeline.validation.render_pass = p_for_render_pass; pipeline.validation.vertex_format = p_vertex_format; pipeline.validation.uses_restart_indices = input_assembly_create_info.primitiveRestartEnable; static const uint32_t primitive_divisor[RENDER_PRIMITIVE_MAX] = { 1, 2, 1, 1, 1, 3, 1, 1, 1, 1, 1 }; pipeline.validation.primitive_divisor = primitive_divisor[p_render_primitive]; static const uint32_t primitive_minimum[RENDER_PRIMITIVE_MAX] = { 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 1, }; pipeline.validation.primitive_minimum = primitive_minimum[p_render_primitive]; #endif // Create ID to associate with this pipeline. RID id = render_pipeline_owner.make_rid(pipeline); #ifdef DEV_ENABLED set_resource_name(id, "RID:" + itos(id.get_id())); #endif // Now add all the dependencies. _add_dependency(id, p_shader); return id; } bool RenderingDeviceVulkan::render_pipeline_is_valid(RID p_pipeline) { _THREAD_SAFE_METHOD_ return render_pipeline_owner.owns(p_pipeline); } /**************************/ /**** COMPUTE PIPELINE ****/ /**************************/ RID RenderingDeviceVulkan::compute_pipeline_create(RID p_shader, const Vector<PipelineSpecializationConstant> &p_specialization_constants) { _THREAD_SAFE_METHOD_ // Needs a shader. Shader *shader = shader_owner.get_or_null(p_shader); ERR_FAIL_NULL_V(shader, RID()); ERR_FAIL_COND_V_MSG(!shader->is_compute, RID(), "Non-compute shaders can't be used in compute pipelines"); // Finally, pipeline create info. VkComputePipelineCreateInfo compute_pipeline_create_info; compute_pipeline_create_info.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO; compute_pipeline_create_info.pNext = nullptr; compute_pipeline_create_info.flags = 0; compute_pipeline_create_info.stage = shader->pipeline_stages[0]; compute_pipeline_create_info.layout = shader->pipeline_layout; compute_pipeline_create_info.basePipelineHandle = VK_NULL_HANDLE; compute_pipeline_create_info.basePipelineIndex = 0; VkSpecializationInfo specialization_info; Vector<VkSpecializationMapEntry> specialization_map_entries; Vector<uint32_t> specialization_constant_data; if (shader->specialization_constants.size()) { specialization_constant_data.resize(shader->specialization_constants.size()); uint32_t *data_ptr = specialization_constant_data.ptrw(); for (int i = 0; i < shader->specialization_constants.size(); i++) { // See if overridden. const Shader::SpecializationConstant &sc = shader->specialization_constants[i]; data_ptr[i] = sc.constant.int_value; // Just copy the 32 bits. for (int j = 0; j < p_specialization_constants.size(); j++) { const PipelineSpecializationConstant &psc = p_specialization_constants[j]; if (psc.constant_id == sc.constant.constant_id) { ERR_FAIL_COND_V_MSG(psc.type != sc.constant.type, RID(), "Specialization constant provided for id (" + itos(sc.constant.constant_id) + ") is of the wrong type."); data_ptr[i] = psc.int_value; break; } } VkSpecializationMapEntry entry; entry.constantID = sc.constant.constant_id; entry.offset = i * sizeof(uint32_t); entry.size = sizeof(uint32_t); specialization_map_entries.push_back(entry); } specialization_info.dataSize = specialization_constant_data.size() * sizeof(uint32_t); specialization_info.pData = data_ptr; specialization_info.mapEntryCount = specialization_map_entries.size(); specialization_info.pMapEntries = specialization_map_entries.ptr(); compute_pipeline_create_info.stage.pSpecializationInfo = &specialization_info; } ComputePipeline pipeline; VkResult err = vkCreateComputePipelines(device, pipelines_cache.cache_object, 1, &compute_pipeline_create_info, nullptr, &pipeline.pipeline); ERR_FAIL_COND_V_MSG(err, RID(), "vkCreateComputePipelines failed with error " + itos(err) + "."); if (pipelines_cache.cache_object != VK_NULL_HANDLE) { _update_pipeline_cache(); } pipeline.set_formats = shader->set_formats; pipeline.push_constant_stages_mask = shader->push_constant.vk_stages_mask; pipeline.pipeline_layout = shader->pipeline_layout; pipeline.shader = p_shader; pipeline.push_constant_size = shader->push_constant.size; pipeline.local_group_size[0] = shader->compute_local_size[0]; pipeline.local_group_size[1] = shader->compute_local_size[1]; pipeline.local_group_size[2] = shader->compute_local_size[2]; // Create ID to associate with this pipeline. RID id = compute_pipeline_owner.make_rid(pipeline); #ifdef DEV_ENABLED set_resource_name(id, "RID:" + itos(id.get_id())); #endif // Now add all the dependencies. _add_dependency(id, p_shader); return id; } bool RenderingDeviceVulkan::compute_pipeline_is_valid(RID p_pipeline) { return compute_pipeline_owner.owns(p_pipeline); } /****************/ /**** SCREEN ****/ /****************/ int RenderingDeviceVulkan::screen_get_width(DisplayServer::WindowID p_screen) const { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG(local_device.is_valid(), -1, "Local devices have no screen"); return context->window_get_width(p_screen); } int RenderingDeviceVulkan::screen_get_height(DisplayServer::WindowID p_screen) const { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG(local_device.is_valid(), -1, "Local devices have no screen"); return context->window_get_height(p_screen); } RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::screen_get_framebuffer_format() const { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG(local_device.is_valid(), INVALID_ID, "Local devices have no screen"); // Very hacky, but not used often per frame so I guess ok. VkFormat vkformat = context->get_screen_format(); DataFormat format = DATA_FORMAT_MAX; for (int i = 0; i < DATA_FORMAT_MAX; i++) { if (vkformat == vulkan_formats[i]) { format = DataFormat(i); break; } } ERR_FAIL_COND_V(format == DATA_FORMAT_MAX, INVALID_ID); AttachmentFormat attachment; attachment.format = format; attachment.samples = TEXTURE_SAMPLES_1; attachment.usage_flags = TEXTURE_USAGE_COLOR_ATTACHMENT_BIT; Vector<AttachmentFormat> screen_attachment; screen_attachment.push_back(attachment); return const_cast<RenderingDeviceVulkan *>(this)->framebuffer_format_create(screen_attachment); } /*******************/ /**** DRAW LIST ****/ /*******************/ RenderingDevice::DrawListID RenderingDeviceVulkan::draw_list_begin_for_screen(DisplayServer::WindowID p_screen, const Color &p_clear_color) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG(local_device.is_valid(), INVALID_ID, "Local devices have no screen"); ERR_FAIL_COND_V_MSG(draw_list != nullptr, INVALID_ID, "Only one draw list can be active at the same time."); ERR_FAIL_COND_V_MSG(compute_list != nullptr, INVALID_ID, "Only one draw/compute list can be active at the same time."); VkCommandBuffer command_buffer = frames[frame].draw_command_buffer; if (!context->window_is_valid_swapchain(p_screen)) { return INVALID_ID; } Size2i size = Size2i(context->window_get_width(p_screen), context->window_get_height(p_screen)); _draw_list_allocate(Rect2i(Vector2i(), size), 0, 0); #ifdef DEBUG_ENABLED draw_list_framebuffer_format = screen_get_framebuffer_format(); #endif draw_list_subpass_count = 1; VkRenderPassBeginInfo render_pass_begin; render_pass_begin.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO; render_pass_begin.pNext = nullptr; render_pass_begin.renderPass = context->window_get_render_pass(p_screen); render_pass_begin.framebuffer = context->window_get_framebuffer(p_screen); render_pass_begin.renderArea.extent.width = size.width; render_pass_begin.renderArea.extent.height = size.height; render_pass_begin.renderArea.offset.x = 0; render_pass_begin.renderArea.offset.y = 0; render_pass_begin.clearValueCount = 1; VkClearValue clear_value; clear_value.color.float32[0] = p_clear_color.r; clear_value.color.float32[1] = p_clear_color.g; clear_value.color.float32[2] = p_clear_color.b; clear_value.color.float32[3] = p_clear_color.a; render_pass_begin.pClearValues = &clear_value; vkCmdBeginRenderPass(command_buffer, &render_pass_begin, VK_SUBPASS_CONTENTS_INLINE); uint32_t size_x = screen_get_width(p_screen); uint32_t size_y = screen_get_height(p_screen); VkViewport viewport; viewport.x = 0; viewport.y = 0; viewport.width = size_x; viewport.height = size_y; viewport.minDepth = 0; viewport.maxDepth = 1.0; vkCmdSetViewport(command_buffer, 0, 1, &viewport); VkRect2D scissor; scissor.offset.x = 0; scissor.offset.y = 0; scissor.extent.width = size_x; scissor.extent.height = size_y; vkCmdSetScissor(command_buffer, 0, 1, &scissor); return int64_t(ID_TYPE_DRAW_LIST) << ID_BASE_SHIFT; } Error RenderingDeviceVulkan::_draw_list_setup_framebuffer(Framebuffer *p_framebuffer, InitialAction p_initial_color_action, FinalAction p_final_color_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, VkFramebuffer *r_framebuffer, VkRenderPass *r_render_pass, uint32_t *r_subpass_count) { Framebuffer::VersionKey vk; vk.initial_color_action = p_initial_color_action; vk.final_color_action = p_final_color_action; vk.initial_depth_action = p_initial_depth_action; vk.final_depth_action = p_final_depth_action; vk.view_count = p_framebuffer->view_count; if (!p_framebuffer->framebuffers.has(vk)) { // Need to create this version. Framebuffer::Version version; version.render_pass = _render_pass_create(framebuffer_formats[p_framebuffer->format_id].E->key().attachments, framebuffer_formats[p_framebuffer->format_id].E->key().passes, p_initial_color_action, p_final_color_action, p_initial_depth_action, p_final_depth_action, p_framebuffer->view_count); VkFramebufferCreateInfo framebuffer_create_info; framebuffer_create_info.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO; framebuffer_create_info.pNext = nullptr; framebuffer_create_info.flags = 0; framebuffer_create_info.renderPass = version.render_pass; Vector<VkImageView> attachments; for (int i = 0; i < p_framebuffer->texture_ids.size(); i++) { Texture *texture = texture_owner.get_or_null(p_framebuffer->texture_ids[i]); if (texture) { attachments.push_back(texture->view); if (!(texture->usage_flags & TEXTURE_USAGE_VRS_ATTACHMENT_BIT)) { // VRS attachment will be a different size. ERR_FAIL_COND_V(texture->width != p_framebuffer->size.width, ERR_BUG); ERR_FAIL_COND_V(texture->height != p_framebuffer->size.height, ERR_BUG); } } } framebuffer_create_info.attachmentCount = attachments.size(); framebuffer_create_info.pAttachments = attachments.ptr(); framebuffer_create_info.width = p_framebuffer->size.width; framebuffer_create_info.height = p_framebuffer->size.height; framebuffer_create_info.layers = 1; VkResult err = vkCreateFramebuffer(device, &framebuffer_create_info, nullptr, &version.framebuffer); ERR_FAIL_COND_V_MSG(err, ERR_CANT_CREATE, "vkCreateFramebuffer failed with error " + itos(err) + "."); version.subpass_count = framebuffer_formats[p_framebuffer->format_id].E->key().passes.size(); p_framebuffer->framebuffers.insert(vk, version); } const Framebuffer::Version &version = p_framebuffer->framebuffers[vk]; *r_framebuffer = version.framebuffer; *r_render_pass = version.render_pass; *r_subpass_count = version.subpass_count; return OK; } Error RenderingDeviceVulkan::_draw_list_render_pass_begin(Framebuffer *framebuffer, InitialAction p_initial_color_action, FinalAction p_final_color_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, const Vector<Color> &p_clear_colors, float p_clear_depth, uint32_t p_clear_stencil, Point2i viewport_offset, Point2i viewport_size, VkFramebuffer vkframebuffer, VkRenderPass render_pass, VkCommandBuffer command_buffer, VkSubpassContents subpass_contents, const Vector<RID> &p_storage_textures) { VkRenderPassBeginInfo render_pass_begin; render_pass_begin.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO; render_pass_begin.pNext = nullptr; render_pass_begin.renderPass = render_pass; render_pass_begin.framebuffer = vkframebuffer; /* * Given how API works, it makes sense to always fully operate on the whole framebuffer. * This allows better continue operations for operations like shadowmapping. render_pass_begin.renderArea.extent.width = viewport_size.width; render_pass_begin.renderArea.extent.height = viewport_size.height; render_pass_begin.renderArea.offset.x = viewport_offset.x; render_pass_begin.renderArea.offset.y = viewport_offset.y; */ render_pass_begin.renderArea.extent.width = framebuffer->size.width; render_pass_begin.renderArea.extent.height = framebuffer->size.height; render_pass_begin.renderArea.offset.x = 0; render_pass_begin.renderArea.offset.y = 0; Vector<VkClearValue> clear_values; clear_values.resize(framebuffer->texture_ids.size()); int clear_values_count = 0; { int color_index = 0; for (int i = 0; i < framebuffer->texture_ids.size(); i++) { VkClearValue clear_value; Texture *texture = texture_owner.get_or_null(framebuffer->texture_ids[i]); if (!texture) { color_index++; continue; } if (color_index < p_clear_colors.size() && texture->usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { ERR_FAIL_INDEX_V(color_index, p_clear_colors.size(), ERR_BUG); // A bug. Color clear_color = p_clear_colors[color_index]; clear_value.color.float32[0] = clear_color.r; clear_value.color.float32[1] = clear_color.g; clear_value.color.float32[2] = clear_color.b; clear_value.color.float32[3] = clear_color.a; color_index++; } else if (texture->usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { clear_value.depthStencil.depth = p_clear_depth; clear_value.depthStencil.stencil = p_clear_stencil; } else { clear_value.color.float32[0] = 0; clear_value.color.float32[1] = 0; clear_value.color.float32[2] = 0; clear_value.color.float32[3] = 0; } clear_values.write[clear_values_count++] = clear_value; } } render_pass_begin.clearValueCount = clear_values_count; render_pass_begin.pClearValues = clear_values.ptr(); for (int i = 0; i < p_storage_textures.size(); i++) { Texture *texture = texture_owner.get_or_null(p_storage_textures[i]); if (!texture) { continue; } ERR_CONTINUE_MSG(!(texture->usage_flags & TEXTURE_USAGE_STORAGE_BIT), "Supplied storage texture " + itos(i) + " for draw list is not set to be used for storage."); if (texture->usage_flags & TEXTURE_USAGE_SAMPLING_BIT) { // Must change layout to general. VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; image_memory_barrier.oldLayout = texture->layout; image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_GENERAL; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = texture->image; image_memory_barrier.subresourceRange.aspectMask = texture->read_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = texture->base_mipmap; image_memory_barrier.subresourceRange.levelCount = texture->mipmaps; image_memory_barrier.subresourceRange.baseArrayLayer = texture->base_layer; image_memory_barrier.subresourceRange.layerCount = texture->layers; vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); texture->layout = VK_IMAGE_LAYOUT_GENERAL; draw_list_storage_textures.push_back(p_storage_textures[i]); } } vkCmdBeginRenderPass(command_buffer, &render_pass_begin, subpass_contents); // Mark textures as bound. draw_list_bound_textures.clear(); draw_list_unbind_color_textures = p_final_color_action != FINAL_ACTION_CONTINUE; draw_list_unbind_depth_textures = p_final_depth_action != FINAL_ACTION_CONTINUE; for (int i = 0; i < framebuffer->texture_ids.size(); i++) { Texture *texture = texture_owner.get_or_null(framebuffer->texture_ids[i]); if (!texture) { continue; } texture->bound = true; draw_list_bound_textures.push_back(framebuffer->texture_ids[i]); } return OK; } void RenderingDeviceVulkan::_draw_list_insert_clear_region(DrawList *p_draw_list, Framebuffer *p_framebuffer, Point2i p_viewport_offset, Point2i p_viewport_size, bool p_clear_color, const Vector<Color> &p_clear_colors, bool p_clear_depth, float p_depth, uint32_t p_stencil) { Vector<VkClearAttachment> clear_attachments; int color_index = 0; int texture_index = 0; for (int i = 0; i < p_framebuffer->texture_ids.size(); i++) { Texture *texture = texture_owner.get_or_null(p_framebuffer->texture_ids[i]); if (!texture) { texture_index++; continue; } VkClearAttachment clear_at = {}; if (p_clear_color && texture->usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { Color clear_color = p_clear_colors[texture_index++]; clear_at.clearValue.color.float32[0] = clear_color.r; clear_at.clearValue.color.float32[1] = clear_color.g; clear_at.clearValue.color.float32[2] = clear_color.b; clear_at.clearValue.color.float32[3] = clear_color.a; clear_at.colorAttachment = color_index++; clear_at.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; } else if (p_clear_depth && texture->usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { clear_at.clearValue.depthStencil.depth = p_depth; clear_at.clearValue.depthStencil.stencil = p_stencil; clear_at.colorAttachment = 0; clear_at.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT; if (format_has_stencil(texture->format)) { clear_at.aspectMask |= VK_IMAGE_ASPECT_STENCIL_BIT; } } else { ERR_CONTINUE(true); } clear_attachments.push_back(clear_at); } VkClearRect cr; cr.baseArrayLayer = 0; cr.layerCount = 1; cr.rect.offset.x = p_viewport_offset.x; cr.rect.offset.y = p_viewport_offset.y; cr.rect.extent.width = p_viewport_size.width; cr.rect.extent.height = p_viewport_size.height; vkCmdClearAttachments(p_draw_list->command_buffer, clear_attachments.size(), clear_attachments.ptr(), 1, &cr); } RenderingDevice::DrawListID RenderingDeviceVulkan::draw_list_begin(RID p_framebuffer, InitialAction p_initial_color_action, FinalAction p_final_color_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, const Vector<Color> &p_clear_color_values, float p_clear_depth, uint32_t p_clear_stencil, const Rect2 &p_region, const Vector<RID> &p_storage_textures) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG(draw_list != nullptr, INVALID_ID, "Only one draw list can be active at the same time."); ERR_FAIL_COND_V_MSG(compute_list != nullptr && !compute_list->state.allow_draw_overlap, INVALID_ID, "Only one draw/compute list can be active at the same time."); Framebuffer *framebuffer = framebuffer_owner.get_or_null(p_framebuffer); ERR_FAIL_NULL_V(framebuffer, INVALID_ID); Point2i viewport_offset; Point2i viewport_size = framebuffer->size; bool needs_clear_color = false; bool needs_clear_depth = false; if (p_region != Rect2() && p_region != Rect2(Vector2(), viewport_size)) { // Check custom region. Rect2i viewport(viewport_offset, viewport_size); Rect2i regioni = p_region; if (!(regioni.position.x >= viewport.position.x) && (regioni.position.y >= viewport.position.y) && ((regioni.position.x + regioni.size.x) <= (viewport.position.x + viewport.size.x)) && ((regioni.position.y + regioni.size.y) <= (viewport.position.y + viewport.size.y))) { ERR_FAIL_V_MSG(INVALID_ID, "When supplying a custom region, it must be contained within the framebuffer rectangle"); } viewport_offset = regioni.position; viewport_size = regioni.size; if (p_initial_color_action == INITIAL_ACTION_CLEAR_REGION_CONTINUE) { needs_clear_color = true; p_initial_color_action = INITIAL_ACTION_CONTINUE; } if (p_initial_depth_action == INITIAL_ACTION_CLEAR_REGION_CONTINUE) { needs_clear_depth = true; p_initial_depth_action = INITIAL_ACTION_CONTINUE; } if (p_initial_color_action == INITIAL_ACTION_CLEAR_REGION) { needs_clear_color = true; p_initial_color_action = INITIAL_ACTION_KEEP; } if (p_initial_depth_action == INITIAL_ACTION_CLEAR_REGION) { needs_clear_depth = true; p_initial_depth_action = INITIAL_ACTION_KEEP; } } if (p_initial_color_action == INITIAL_ACTION_CLEAR || needs_clear_color) { // Check clear values. int color_count = 0; for (int i = 0; i < framebuffer->texture_ids.size(); i++) { Texture *texture = texture_owner.get_or_null(framebuffer->texture_ids[i]); // We only check for our VRS usage bit if this is not the first texture id. // If it is the first we're likely populating our VRS texture. // Bit dirty but... if (!texture || (!(texture->usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) && !(i != 0 && texture->usage_flags & TEXTURE_USAGE_VRS_ATTACHMENT_BIT))) { if (!texture || !texture->is_resolve_buffer) { color_count++; } } } ERR_FAIL_COND_V_MSG(p_clear_color_values.size() != color_count, INVALID_ID, "Clear color values supplied (" + itos(p_clear_color_values.size()) + ") differ from the amount required for framebuffer color attachments (" + itos(color_count) + ")."); } VkFramebuffer vkframebuffer; VkRenderPass render_pass; Error err = _draw_list_setup_framebuffer(framebuffer, p_initial_color_action, p_final_color_action, p_initial_depth_action, p_final_depth_action, &vkframebuffer, &render_pass, &draw_list_subpass_count); ERR_FAIL_COND_V(err != OK, INVALID_ID); VkCommandBuffer command_buffer = frames[frame].draw_command_buffer; err = _draw_list_render_pass_begin(framebuffer, p_initial_color_action, p_final_color_action, p_initial_depth_action, p_final_depth_action, p_clear_color_values, p_clear_depth, p_clear_stencil, viewport_offset, viewport_size, vkframebuffer, render_pass, command_buffer, VK_SUBPASS_CONTENTS_INLINE, p_storage_textures); if (err != OK) { return INVALID_ID; } draw_list_render_pass = render_pass; draw_list_vkframebuffer = vkframebuffer; _draw_list_allocate(Rect2i(viewport_offset, viewport_size), 0, 0); #ifdef DEBUG_ENABLED draw_list_framebuffer_format = framebuffer->format_id; #endif draw_list_current_subpass = 0; if (needs_clear_color || needs_clear_depth) { _draw_list_insert_clear_region(draw_list, framebuffer, viewport_offset, viewport_size, needs_clear_color, p_clear_color_values, needs_clear_depth, p_clear_depth, p_clear_stencil); } VkViewport viewport; viewport.x = viewport_offset.x; viewport.y = viewport_offset.y; viewport.width = viewport_size.width; viewport.height = viewport_size.height; viewport.minDepth = 0; viewport.maxDepth = 1.0; vkCmdSetViewport(command_buffer, 0, 1, &viewport); VkRect2D scissor; scissor.offset.x = viewport_offset.x; scissor.offset.y = viewport_offset.y; scissor.extent.width = viewport_size.width; scissor.extent.height = viewport_size.height; vkCmdSetScissor(command_buffer, 0, 1, &scissor); return int64_t(ID_TYPE_DRAW_LIST) << ID_BASE_SHIFT; } Error RenderingDeviceVulkan::draw_list_begin_split(RID p_framebuffer, uint32_t p_splits, DrawListID *r_split_ids, InitialAction p_initial_color_action, FinalAction p_final_color_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, const Vector<Color> &p_clear_color_values, float p_clear_depth, uint32_t p_clear_stencil, const Rect2 &p_region, const Vector<RID> &p_storage_textures) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG(draw_list != nullptr, ERR_BUSY, "Only one draw list can be active at the same time."); ERR_FAIL_COND_V_MSG(compute_list != nullptr && !compute_list->state.allow_draw_overlap, ERR_BUSY, "Only one draw/compute list can be active at the same time."); ERR_FAIL_COND_V(p_splits < 1, ERR_INVALID_DECLARATION); Framebuffer *framebuffer = framebuffer_owner.get_or_null(p_framebuffer); ERR_FAIL_NULL_V(framebuffer, ERR_INVALID_DECLARATION); Point2i viewport_offset; Point2i viewport_size = framebuffer->size; bool needs_clear_color = false; bool needs_clear_depth = false; if (p_region != Rect2() && p_region != Rect2(Vector2(), viewport_size)) { // Check custom region. Rect2i viewport(viewport_offset, viewport_size); Rect2i regioni = p_region; if (!(regioni.position.x >= viewport.position.x) && (regioni.position.y >= viewport.position.y) && ((regioni.position.x + regioni.size.x) <= (viewport.position.x + viewport.size.x)) && ((regioni.position.y + regioni.size.y) <= (viewport.position.y + viewport.size.y))) { ERR_FAIL_V_MSG(ERR_INVALID_PARAMETER, "When supplying a custom region, it must be contained within the framebuffer rectangle"); } viewport_offset = regioni.position; viewport_size = regioni.size; if (p_initial_color_action == INITIAL_ACTION_CLEAR_REGION) { needs_clear_color = true; p_initial_color_action = INITIAL_ACTION_KEEP; } if (p_initial_depth_action == INITIAL_ACTION_CLEAR_REGION) { needs_clear_depth = true; p_initial_depth_action = INITIAL_ACTION_KEEP; } } if (p_initial_color_action == INITIAL_ACTION_CLEAR || needs_clear_color) { // Check clear values. int color_count = 0; for (int i = 0; i < framebuffer->texture_ids.size(); i++) { Texture *texture = texture_owner.get_or_null(framebuffer->texture_ids[i]); if (!texture || !(texture->usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT)) { color_count++; } } ERR_FAIL_COND_V_MSG(p_clear_color_values.size() != color_count, ERR_INVALID_PARAMETER, "Clear color values supplied (" + itos(p_clear_color_values.size()) + ") differ from the amount required for framebuffer (" + itos(color_count) + ")."); } VkFramebuffer vkframebuffer; VkRenderPass render_pass; Error err = _draw_list_setup_framebuffer(framebuffer, p_initial_color_action, p_final_color_action, p_initial_depth_action, p_final_depth_action, &vkframebuffer, &render_pass, &draw_list_subpass_count); ERR_FAIL_COND_V(err != OK, ERR_CANT_CREATE); VkCommandBuffer frame_command_buffer = frames[frame].draw_command_buffer; err = _draw_list_render_pass_begin(framebuffer, p_initial_color_action, p_final_color_action, p_initial_depth_action, p_final_depth_action, p_clear_color_values, p_clear_depth, p_clear_stencil, viewport_offset, viewport_size, vkframebuffer, render_pass, frame_command_buffer, VK_SUBPASS_CONTENTS_SECONDARY_COMMAND_BUFFERS, p_storage_textures); if (err != OK) { return ERR_CANT_CREATE; } draw_list_current_subpass = 0; #ifdef DEBUG_ENABLED draw_list_framebuffer_format = framebuffer->format_id; #endif draw_list_render_pass = render_pass; draw_list_vkframebuffer = vkframebuffer; err = _draw_list_allocate(Rect2i(viewport_offset, viewport_size), p_splits, 0); if (err != OK) { return err; } if (needs_clear_color || needs_clear_depth) { _draw_list_insert_clear_region(&draw_list[0], framebuffer, viewport_offset, viewport_size, needs_clear_color, p_clear_color_values, needs_clear_depth, p_clear_depth, p_clear_stencil); } for (uint32_t i = 0; i < p_splits; i++) { VkViewport viewport; viewport.x = viewport_offset.x; viewport.y = viewport_offset.y; viewport.width = viewport_size.width; viewport.height = viewport_size.height; viewport.minDepth = 0; viewport.maxDepth = 1.0; vkCmdSetViewport(draw_list[i].command_buffer, 0, 1, &viewport); VkRect2D scissor; scissor.offset.x = viewport_offset.x; scissor.offset.y = viewport_offset.y; scissor.extent.width = viewport_size.width; scissor.extent.height = viewport_size.height; vkCmdSetScissor(draw_list[i].command_buffer, 0, 1, &scissor); r_split_ids[i] = (int64_t(ID_TYPE_SPLIT_DRAW_LIST) << ID_BASE_SHIFT) + i; } return OK; } RenderingDeviceVulkan::DrawList *RenderingDeviceVulkan::_get_draw_list_ptr(DrawListID p_id) { if (p_id < 0) { return nullptr; } if (!draw_list) { return nullptr; } else if (p_id == (int64_t(ID_TYPE_DRAW_LIST) << ID_BASE_SHIFT)) { if (draw_list_split) { return nullptr; } return draw_list; } else if (p_id >> DrawListID(ID_BASE_SHIFT) == ID_TYPE_SPLIT_DRAW_LIST) { if (!draw_list_split) { return nullptr; } uint64_t index = p_id & ((DrawListID(1) << DrawListID(ID_BASE_SHIFT)) - 1); // Mask. if (index >= draw_list_count) { return nullptr; } return &draw_list[index]; } else { return nullptr; } } void RenderingDeviceVulkan::draw_list_set_blend_constants(DrawListID p_list, const Color &p_color) { DrawList *dl = _get_draw_list_ptr(p_list); ERR_FAIL_NULL(dl); #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!dl->validation.active, "Submitted Draw Lists can no longer be modified."); #endif vkCmdSetBlendConstants(dl->command_buffer, p_color.components); } void RenderingDeviceVulkan::draw_list_bind_render_pipeline(DrawListID p_list, RID p_render_pipeline) { DrawList *dl = _get_draw_list_ptr(p_list); ERR_FAIL_NULL(dl); #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!dl->validation.active, "Submitted Draw Lists can no longer be modified."); #endif const RenderPipeline *pipeline = render_pipeline_owner.get_or_null(p_render_pipeline); ERR_FAIL_NULL(pipeline); #ifdef DEBUG_ENABLED ERR_FAIL_COND(pipeline->validation.framebuffer_format != draw_list_framebuffer_format && pipeline->validation.render_pass != draw_list_current_subpass); #endif if (p_render_pipeline == dl->state.pipeline) { return; // Redundant state, return. } dl->state.pipeline = p_render_pipeline; dl->state.pipeline_layout = pipeline->pipeline_layout; vkCmdBindPipeline(dl->command_buffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline->pipeline); if (dl->state.pipeline_shader != pipeline->shader) { // Shader changed, so descriptor sets may become incompatible. // Go through ALL sets, and unbind them (and all those above) if the format is different. uint32_t pcount = pipeline->set_formats.size(); // Formats count in this pipeline. dl->state.set_count = MAX(dl->state.set_count, pcount); const uint32_t *pformats = pipeline->set_formats.ptr(); // Pipeline set formats. bool sets_valid = true; // Once invalid, all above become invalid. for (uint32_t i = 0; i < pcount; i++) { // If a part of the format is different, invalidate it (and the rest). if (!sets_valid || dl->state.sets[i].pipeline_expected_format != pformats[i]) { dl->state.sets[i].bound = false; dl->state.sets[i].pipeline_expected_format = pformats[i]; sets_valid = false; } } for (uint32_t i = pcount; i < dl->state.set_count; i++) { // Unbind the ones above (not used) if exist. dl->state.sets[i].bound = false; } dl->state.set_count = pcount; // Update set count. if (pipeline->push_constant_size) { dl->state.pipeline_push_constant_stages = pipeline->push_constant_stages_mask; #ifdef DEBUG_ENABLED dl->validation.pipeline_push_constant_supplied = false; #endif } dl->state.pipeline_shader = pipeline->shader; } #ifdef DEBUG_ENABLED // Update render pass pipeline info. dl->validation.pipeline_active = true; dl->validation.pipeline_dynamic_state = pipeline->validation.dynamic_state; dl->validation.pipeline_vertex_format = pipeline->validation.vertex_format; dl->validation.pipeline_uses_restart_indices = pipeline->validation.uses_restart_indices; dl->validation.pipeline_primitive_divisor = pipeline->validation.primitive_divisor; dl->validation.pipeline_primitive_minimum = pipeline->validation.primitive_minimum; dl->validation.pipeline_push_constant_size = pipeline->push_constant_size; #endif } void RenderingDeviceVulkan::draw_list_bind_uniform_set(DrawListID p_list, RID p_uniform_set, uint32_t p_index) { #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(p_index >= limits.maxBoundDescriptorSets || p_index >= MAX_UNIFORM_SETS, "Attempting to bind a descriptor set (" + itos(p_index) + ") greater than what the hardware supports (" + itos(limits.maxBoundDescriptorSets) + ")."); #endif DrawList *dl = _get_draw_list_ptr(p_list); ERR_FAIL_NULL(dl); #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!dl->validation.active, "Submitted Draw Lists can no longer be modified."); #endif const UniformSet *uniform_set = uniform_set_owner.get_or_null(p_uniform_set); ERR_FAIL_NULL(uniform_set); if (p_index > dl->state.set_count) { dl->state.set_count = p_index; } dl->state.sets[p_index].descriptor_set = uniform_set->descriptor_set; // Update set pointer. dl->state.sets[p_index].bound = false; // Needs rebind. dl->state.sets[p_index].uniform_set_format = uniform_set->format; dl->state.sets[p_index].uniform_set = p_uniform_set; uint32_t mst_count = uniform_set->mutable_storage_textures.size(); if (mst_count) { Texture **mst_textures = const_cast<UniformSet *>(uniform_set)->mutable_storage_textures.ptrw(); for (uint32_t i = 0; i < mst_count; i++) { if (mst_textures[i]->used_in_frame != frames_drawn) { mst_textures[i]->used_in_frame = frames_drawn; mst_textures[i]->used_in_transfer = false; mst_textures[i]->used_in_compute = false; } mst_textures[i]->used_in_raster = true; } } #ifdef DEBUG_ENABLED { // Validate that textures bound are not attached as framebuffer bindings. uint32_t attachable_count = uniform_set->attachable_textures.size(); const UniformSet::AttachableTexture *attachable_ptr = uniform_set->attachable_textures.ptr(); uint32_t bound_count = draw_list_bound_textures.size(); const RID *bound_ptr = draw_list_bound_textures.ptr(); for (uint32_t i = 0; i < attachable_count; i++) { for (uint32_t j = 0; j < bound_count; j++) { ERR_FAIL_COND_MSG(attachable_ptr[i].texture == bound_ptr[j], "Attempted to use the same texture in framebuffer attachment and a uniform (set: " + itos(p_index) + ", binding: " + itos(attachable_ptr[i].bind) + "), this is not allowed."); } } } #endif } void RenderingDeviceVulkan::draw_list_bind_vertex_array(DrawListID p_list, RID p_vertex_array) { DrawList *dl = _get_draw_list_ptr(p_list); ERR_FAIL_NULL(dl); #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!dl->validation.active, "Submitted Draw Lists can no longer be modified."); #endif const VertexArray *vertex_array = vertex_array_owner.get_or_null(p_vertex_array); ERR_FAIL_NULL(vertex_array); if (dl->state.vertex_array == p_vertex_array) { return; // Already set. } dl->state.vertex_array = p_vertex_array; #ifdef DEBUG_ENABLED dl->validation.vertex_format = vertex_array->description; dl->validation.vertex_max_instances_allowed = vertex_array->max_instances_allowed; #endif dl->validation.vertex_array_size = vertex_array->vertex_count; vkCmdBindVertexBuffers(dl->command_buffer, 0, vertex_array->buffers.size(), vertex_array->buffers.ptr(), vertex_array->offsets.ptr()); } void RenderingDeviceVulkan::draw_list_bind_index_array(DrawListID p_list, RID p_index_array) { DrawList *dl = _get_draw_list_ptr(p_list); ERR_FAIL_NULL(dl); #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!dl->validation.active, "Submitted Draw Lists can no longer be modified."); #endif const IndexArray *index_array = index_array_owner.get_or_null(p_index_array); ERR_FAIL_NULL(index_array); if (dl->state.index_array == p_index_array) { return; // Already set. } dl->state.index_array = p_index_array; #ifdef DEBUG_ENABLED dl->validation.index_array_max_index = index_array->max_index; #endif dl->validation.index_array_size = index_array->indices; dl->validation.index_array_offset = index_array->offset; vkCmdBindIndexBuffer(dl->command_buffer, index_array->buffer, 0, index_array->index_type); } void RenderingDeviceVulkan::draw_list_set_line_width(DrawListID p_list, float p_width) { DrawList *dl = _get_draw_list_ptr(p_list); ERR_FAIL_NULL(dl); #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!dl->validation.active, "Submitted Draw Lists can no longer be modified."); #endif vkCmdSetLineWidth(dl->command_buffer, p_width); } void RenderingDeviceVulkan::draw_list_set_push_constant(DrawListID p_list, const void *p_data, uint32_t p_data_size) { DrawList *dl = _get_draw_list_ptr(p_list); ERR_FAIL_NULL(dl); #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!dl->validation.active, "Submitted Draw Lists can no longer be modified."); #endif #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(p_data_size != dl->validation.pipeline_push_constant_size, "This render pipeline requires (" + itos(dl->validation.pipeline_push_constant_size) + ") bytes of push constant data, supplied: (" + itos(p_data_size) + ")"); #endif vkCmdPushConstants(dl->command_buffer, dl->state.pipeline_layout, dl->state.pipeline_push_constant_stages, 0, p_data_size, p_data); #ifdef DEBUG_ENABLED dl->validation.pipeline_push_constant_supplied = true; #endif } void RenderingDeviceVulkan::draw_list_draw(DrawListID p_list, bool p_use_indices, uint32_t p_instances, uint32_t p_procedural_vertices) { DrawList *dl = _get_draw_list_ptr(p_list); ERR_FAIL_NULL(dl); #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!dl->validation.active, "Submitted Draw Lists can no longer be modified."); #endif #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!dl->validation.pipeline_active, "No render pipeline was set before attempting to draw."); if (dl->validation.pipeline_vertex_format != INVALID_ID) { // Pipeline uses vertices, validate format. ERR_FAIL_COND_MSG(dl->validation.vertex_format == INVALID_ID, "No vertex array was bound, and render pipeline expects vertices."); // Make sure format is right. ERR_FAIL_COND_MSG(dl->validation.pipeline_vertex_format != dl->validation.vertex_format, "The vertex format used to create the pipeline does not match the vertex format bound."); // Make sure number of instances is valid. ERR_FAIL_COND_MSG(p_instances > dl->validation.vertex_max_instances_allowed, "Number of instances requested (" + itos(p_instances) + " is larger than the maximum number supported by the bound vertex array (" + itos(dl->validation.vertex_max_instances_allowed) + ")."); } if (dl->validation.pipeline_push_constant_size > 0) { // Using push constants, check that they were supplied. ERR_FAIL_COND_MSG(!dl->validation.pipeline_push_constant_supplied, "The shader in this pipeline requires a push constant to be set before drawing, but it's not present."); } #endif // Bind descriptor sets. for (uint32_t i = 0; i < dl->state.set_count; i++) { if (dl->state.sets[i].pipeline_expected_format == 0) { continue; // Nothing expected by this pipeline. } #ifdef DEBUG_ENABLED if (dl->state.sets[i].pipeline_expected_format != dl->state.sets[i].uniform_set_format) { if (dl->state.sets[i].uniform_set_format == 0) { ERR_FAIL_MSG("Uniforms were never supplied for set (" + itos(i) + ") at the time of drawing, which are required by the pipeline"); } else if (uniform_set_owner.owns(dl->state.sets[i].uniform_set)) { UniformSet *us = uniform_set_owner.get_or_null(dl->state.sets[i].uniform_set); ERR_FAIL_MSG("Uniforms supplied for set (" + itos(i) + "):\n" + _shader_uniform_debug(us->shader_id, us->shader_set) + "\nare not the same format as required by the pipeline shader. Pipeline shader requires the following bindings:\n" + _shader_uniform_debug(dl->state.pipeline_shader)); } else { ERR_FAIL_MSG("Uniforms supplied for set (" + itos(i) + ", which was was just freed) are not the same format as required by the pipeline shader. Pipeline shader requires the following bindings:\n" + _shader_uniform_debug(dl->state.pipeline_shader)); } } #endif if (!dl->state.sets[i].bound) { // All good, see if this requires re-binding. vkCmdBindDescriptorSets(dl->command_buffer, VK_PIPELINE_BIND_POINT_GRAPHICS, dl->state.pipeline_layout, i, 1, &dl->state.sets[i].descriptor_set, 0, nullptr); dl->state.sets[i].bound = true; } } if (p_use_indices) { #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(p_procedural_vertices > 0, "Procedural vertices can't be used together with indices."); ERR_FAIL_COND_MSG(!dl->validation.index_array_size, "Draw command requested indices, but no index buffer was set."); ERR_FAIL_COND_MSG(dl->validation.pipeline_uses_restart_indices != dl->validation.index_buffer_uses_restart_indices, "The usage of restart indices in index buffer does not match the render primitive in the pipeline."); #endif uint32_t to_draw = dl->validation.index_array_size; #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(to_draw < dl->validation.pipeline_primitive_minimum, "Too few indices (" + itos(to_draw) + ") for the render primitive set in the render pipeline (" + itos(dl->validation.pipeline_primitive_minimum) + ")."); ERR_FAIL_COND_MSG((to_draw % dl->validation.pipeline_primitive_divisor) != 0, "Index amount (" + itos(to_draw) + ") must be a multiple of the amount of indices required by the render primitive (" + itos(dl->validation.pipeline_primitive_divisor) + ")."); #endif vkCmdDrawIndexed(dl->command_buffer, to_draw, p_instances, dl->validation.index_array_offset, 0, 0); } else { uint32_t to_draw; if (p_procedural_vertices > 0) { #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(dl->validation.pipeline_vertex_format != INVALID_ID, "Procedural vertices requested, but pipeline expects a vertex array."); #endif to_draw = p_procedural_vertices; } else { #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(dl->validation.pipeline_vertex_format == INVALID_ID, "Draw command lacks indices, but pipeline format does not use vertices."); #endif to_draw = dl->validation.vertex_array_size; } #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(to_draw < dl->validation.pipeline_primitive_minimum, "Too few vertices (" + itos(to_draw) + ") for the render primitive set in the render pipeline (" + itos(dl->validation.pipeline_primitive_minimum) + ")."); ERR_FAIL_COND_MSG((to_draw % dl->validation.pipeline_primitive_divisor) != 0, "Vertex amount (" + itos(to_draw) + ") must be a multiple of the amount of vertices required by the render primitive (" + itos(dl->validation.pipeline_primitive_divisor) + ")."); #endif vkCmdDraw(dl->command_buffer, to_draw, p_instances, 0, 0); } } void RenderingDeviceVulkan::draw_list_enable_scissor(DrawListID p_list, const Rect2 &p_rect) { DrawList *dl = _get_draw_list_ptr(p_list); ERR_FAIL_NULL(dl); #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!dl->validation.active, "Submitted Draw Lists can no longer be modified."); #endif Rect2i rect = p_rect; rect.position += dl->viewport.position; rect = dl->viewport.intersection(rect); if (rect.get_area() == 0) { return; } VkRect2D scissor; scissor.offset.x = rect.position.x; scissor.offset.y = rect.position.y; scissor.extent.width = rect.size.width; scissor.extent.height = rect.size.height; vkCmdSetScissor(dl->command_buffer, 0, 1, &scissor); } void RenderingDeviceVulkan::draw_list_disable_scissor(DrawListID p_list) { DrawList *dl = _get_draw_list_ptr(p_list); ERR_FAIL_NULL(dl); #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!dl->validation.active, "Submitted Draw Lists can no longer be modified."); #endif VkRect2D scissor; scissor.offset.x = dl->viewport.position.x; scissor.offset.y = dl->viewport.position.y; scissor.extent.width = dl->viewport.size.width; scissor.extent.height = dl->viewport.size.height; vkCmdSetScissor(dl->command_buffer, 0, 1, &scissor); } uint32_t RenderingDeviceVulkan::draw_list_get_current_pass() { return draw_list_current_subpass; } RenderingDevice::DrawListID RenderingDeviceVulkan::draw_list_switch_to_next_pass() { _THREAD_SAFE_METHOD_ ERR_FAIL_NULL_V(draw_list, INVALID_ID); ERR_FAIL_COND_V(draw_list_current_subpass >= draw_list_subpass_count - 1, INVALID_FORMAT_ID); draw_list_current_subpass++; Rect2i viewport; _draw_list_free(&viewport); vkCmdNextSubpass(frames[frame].draw_command_buffer, VK_SUBPASS_CONTENTS_INLINE); _draw_list_allocate(viewport, 0, draw_list_current_subpass); return int64_t(ID_TYPE_DRAW_LIST) << ID_BASE_SHIFT; } Error RenderingDeviceVulkan::draw_list_switch_to_next_pass_split(uint32_t p_splits, DrawListID *r_split_ids) { _THREAD_SAFE_METHOD_ ERR_FAIL_NULL_V(draw_list, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V(draw_list_current_subpass >= draw_list_subpass_count - 1, ERR_INVALID_PARAMETER); draw_list_current_subpass++; Rect2i viewport; _draw_list_free(&viewport); vkCmdNextSubpass(frames[frame].draw_command_buffer, VK_SUBPASS_CONTENTS_INLINE); _draw_list_allocate(viewport, p_splits, draw_list_current_subpass); for (uint32_t i = 0; i < p_splits; i++) { r_split_ids[i] = (int64_t(ID_TYPE_SPLIT_DRAW_LIST) << ID_BASE_SHIFT) + i; } return OK; } Error RenderingDeviceVulkan::_draw_list_allocate(const Rect2i &p_viewport, uint32_t p_splits, uint32_t p_subpass) { // Lock while draw_list is active. _THREAD_SAFE_LOCK_ if (p_splits == 0) { draw_list = memnew(DrawList); draw_list->command_buffer = frames[frame].draw_command_buffer; draw_list->viewport = p_viewport; draw_list_count = 0; draw_list_split = false; } else { if (p_splits > (uint32_t)split_draw_list_allocators.size()) { uint32_t from = split_draw_list_allocators.size(); split_draw_list_allocators.resize(p_splits); for (uint32_t i = from; i < p_splits; i++) { VkCommandPoolCreateInfo cmd_pool_info; cmd_pool_info.sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO; cmd_pool_info.pNext = nullptr; cmd_pool_info.queueFamilyIndex = context->get_graphics_queue_family_index(); cmd_pool_info.flags = VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT; VkResult res = vkCreateCommandPool(device, &cmd_pool_info, nullptr, &split_draw_list_allocators.write[i].command_pool); ERR_FAIL_COND_V_MSG(res, ERR_CANT_CREATE, "vkCreateCommandPool failed with error " + itos(res) + "."); for (int j = 0; j < frame_count; j++) { VkCommandBuffer command_buffer; VkCommandBufferAllocateInfo cmdbuf; // No command buffer exists, create it. cmdbuf.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO; cmdbuf.pNext = nullptr; cmdbuf.commandPool = split_draw_list_allocators[i].command_pool; cmdbuf.level = VK_COMMAND_BUFFER_LEVEL_SECONDARY; cmdbuf.commandBufferCount = 1; VkResult err = vkAllocateCommandBuffers(device, &cmdbuf, &command_buffer); ERR_FAIL_COND_V_MSG(err, ERR_CANT_CREATE, "vkAllocateCommandBuffers failed with error " + itos(err) + "."); split_draw_list_allocators.write[i].command_buffers.push_back(command_buffer); } } } draw_list = memnew_arr(DrawList, p_splits); draw_list_count = p_splits; draw_list_split = true; for (uint32_t i = 0; i < p_splits; i++) { // Take a command buffer and initialize it. VkCommandBuffer command_buffer = split_draw_list_allocators[i].command_buffers[frame]; VkCommandBufferInheritanceInfo inheritance_info; inheritance_info.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_INHERITANCE_INFO; inheritance_info.pNext = nullptr; inheritance_info.renderPass = draw_list_render_pass; inheritance_info.subpass = p_subpass; inheritance_info.framebuffer = draw_list_vkframebuffer; inheritance_info.occlusionQueryEnable = false; inheritance_info.queryFlags = 0; // ? inheritance_info.pipelineStatistics = 0; VkCommandBufferBeginInfo cmdbuf_begin; cmdbuf_begin.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO; cmdbuf_begin.pNext = nullptr; cmdbuf_begin.flags = VK_COMMAND_BUFFER_USAGE_SIMULTANEOUS_USE_BIT | VK_COMMAND_BUFFER_USAGE_RENDER_PASS_CONTINUE_BIT; cmdbuf_begin.pInheritanceInfo = &inheritance_info; VkResult res = vkResetCommandBuffer(command_buffer, 0); if (res) { memdelete_arr(draw_list); draw_list = nullptr; ERR_FAIL_V_MSG(ERR_CANT_CREATE, "vkResetCommandBuffer failed with error " + itos(res) + "."); } res = vkBeginCommandBuffer(command_buffer, &cmdbuf_begin); if (res) { memdelete_arr(draw_list); draw_list = nullptr; ERR_FAIL_V_MSG(ERR_CANT_CREATE, "vkBeginCommandBuffer failed with error " + itos(res) + "."); } draw_list[i].command_buffer = command_buffer; draw_list[i].viewport = p_viewport; } } return OK; } void RenderingDeviceVulkan::_draw_list_free(Rect2i *r_last_viewport) { if (draw_list_split) { // Send all command buffers. VkCommandBuffer *command_buffers = (VkCommandBuffer *)alloca(sizeof(VkCommandBuffer) * draw_list_count); for (uint32_t i = 0; i < draw_list_count; i++) { vkEndCommandBuffer(draw_list[i].command_buffer); command_buffers[i] = draw_list[i].command_buffer; if (r_last_viewport) { if (i == 0 || draw_list[i].viewport_set) { *r_last_viewport = draw_list[i].viewport; } } } vkCmdExecuteCommands(frames[frame].draw_command_buffer, draw_list_count, command_buffers); memdelete_arr(draw_list); draw_list = nullptr; } else { if (r_last_viewport) { *r_last_viewport = draw_list->viewport; } // Just end the list. memdelete(draw_list); draw_list = nullptr; } // Draw_list is no longer active. _THREAD_SAFE_UNLOCK_ } void RenderingDeviceVulkan::draw_list_end(BitField<BarrierMask> p_post_barrier) { _THREAD_SAFE_METHOD_ ERR_FAIL_NULL_MSG(draw_list, "Immediate draw list is already inactive."); _draw_list_free(); vkCmdEndRenderPass(frames[frame].draw_command_buffer); for (int i = 0; i < draw_list_bound_textures.size(); i++) { Texture *texture = texture_owner.get_or_null(draw_list_bound_textures[i]); ERR_CONTINUE(!texture); // Wtf. if (draw_list_unbind_color_textures && (texture->usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT)) { texture->bound = false; } if (draw_list_unbind_depth_textures && (texture->usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT)) { texture->bound = false; } } uint32_t barrier_flags = 0; uint32_t access_flags = 0; if (p_post_barrier.has_flag(BARRIER_MASK_COMPUTE)) { barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_VERTEX)) { barrier_flags |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_VERTEX_SHADER_BIT /*| VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT*/; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT /*| VK_ACCESS_INDIRECT_COMMAND_READ_BIT*/; } if (p_post_barrier.has_flag(BARRIER_MASK_FRAGMENT)) { barrier_flags |= VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT /*| VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT*/; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT /*| VK_ACCESS_INDIRECT_COMMAND_READ_BIT*/; } if (p_post_barrier.has_flag(BARRIER_MASK_TRANSFER)) { barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_TRANSFER_READ_BIT; } if (barrier_flags == 0) { barrier_flags = VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT; } draw_list_bound_textures.clear(); VkImageMemoryBarrier *image_barriers = nullptr; uint32_t image_barrier_count = draw_list_storage_textures.size(); if (image_barrier_count) { image_barriers = (VkImageMemoryBarrier *)alloca(sizeof(VkImageMemoryBarrier) * draw_list_storage_textures.size()); } uint32_t src_stage = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT; uint32_t src_access = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT; if (image_barrier_count) { src_stage |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; src_access |= VK_ACCESS_SHADER_WRITE_BIT; } for (uint32_t i = 0; i < image_barrier_count; i++) { Texture *texture = texture_owner.get_or_null(draw_list_storage_textures[i]); VkImageMemoryBarrier &image_memory_barrier = image_barriers[i]; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = src_access; image_memory_barrier.dstAccessMask = access_flags; image_memory_barrier.oldLayout = texture->layout; image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = texture->image; image_memory_barrier.subresourceRange.aspectMask = texture->read_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = texture->base_mipmap; image_memory_barrier.subresourceRange.levelCount = texture->mipmaps; image_memory_barrier.subresourceRange.baseArrayLayer = texture->base_layer; image_memory_barrier.subresourceRange.layerCount = texture->layers; texture->layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; } draw_list_storage_textures.clear(); // To ensure proper synchronization, we must make sure rendering is done before: // * Some buffer is copied. // * Another render pass happens (since we may be done). VkMemoryBarrier mem_barrier; mem_barrier.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER; mem_barrier.pNext = nullptr; mem_barrier.srcAccessMask = src_access; mem_barrier.dstAccessMask = access_flags; if (image_barrier_count > 0 || p_post_barrier != BARRIER_MASK_NO_BARRIER) { vkCmdPipelineBarrier(frames[frame].draw_command_buffer, src_stage, barrier_flags, 0, 1, &mem_barrier, 0, nullptr, image_barrier_count, image_barriers); } #ifdef FORCE_FULL_BARRIER _full_barrier(true); #endif } /***********************/ /**** COMPUTE LISTS ****/ /***********************/ RenderingDevice::ComputeListID RenderingDeviceVulkan::compute_list_begin(bool p_allow_draw_overlap) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG(!p_allow_draw_overlap && draw_list != nullptr, INVALID_ID, "Only one draw list can be active at the same time."); ERR_FAIL_COND_V_MSG(compute_list != nullptr, INVALID_ID, "Only one draw/compute list can be active at the same time."); // Lock while compute_list is active. _THREAD_SAFE_LOCK_ compute_list = memnew(ComputeList); compute_list->command_buffer = frames[frame].draw_command_buffer; compute_list->state.allow_draw_overlap = p_allow_draw_overlap; return ID_TYPE_COMPUTE_LIST; } void RenderingDeviceVulkan::compute_list_bind_compute_pipeline(ComputeListID p_list, RID p_compute_pipeline) { // Must be called within a compute list, the class mutex is locked during that time ERR_FAIL_COND(p_list != ID_TYPE_COMPUTE_LIST); ERR_FAIL_NULL(compute_list); ComputeList *cl = compute_list; const ComputePipeline *pipeline = compute_pipeline_owner.get_or_null(p_compute_pipeline); ERR_FAIL_NULL(pipeline); if (p_compute_pipeline == cl->state.pipeline) { return; // Redundant state, return. } cl->state.pipeline = p_compute_pipeline; cl->state.pipeline_layout = pipeline->pipeline_layout; vkCmdBindPipeline(cl->command_buffer, VK_PIPELINE_BIND_POINT_COMPUTE, pipeline->pipeline); if (cl->state.pipeline_shader != pipeline->shader) { // Shader changed, so descriptor sets may become incompatible. // Go through ALL sets, and unbind them (and all those above) if the format is different. uint32_t pcount = pipeline->set_formats.size(); // Formats count in this pipeline. cl->state.set_count = MAX(cl->state.set_count, pcount); const uint32_t *pformats = pipeline->set_formats.ptr(); // Pipeline set formats. bool sets_valid = true; // Once invalid, all above become invalid. for (uint32_t i = 0; i < pcount; i++) { // If a part of the format is different, invalidate it (and the rest). if (!sets_valid || cl->state.sets[i].pipeline_expected_format != pformats[i]) { cl->state.sets[i].bound = false; cl->state.sets[i].pipeline_expected_format = pformats[i]; sets_valid = false; } } for (uint32_t i = pcount; i < cl->state.set_count; i++) { // Unbind the ones above (not used) if exist. cl->state.sets[i].bound = false; } cl->state.set_count = pcount; // Update set count. if (pipeline->push_constant_size) { cl->state.pipeline_push_constant_stages = pipeline->push_constant_stages_mask; #ifdef DEBUG_ENABLED cl->validation.pipeline_push_constant_supplied = false; #endif } cl->state.pipeline_shader = pipeline->shader; cl->state.local_group_size[0] = pipeline->local_group_size[0]; cl->state.local_group_size[1] = pipeline->local_group_size[1]; cl->state.local_group_size[2] = pipeline->local_group_size[2]; } #ifdef DEBUG_ENABLED // Update compute pass pipeline info. cl->validation.pipeline_active = true; cl->validation.pipeline_push_constant_size = pipeline->push_constant_size; #endif } void RenderingDeviceVulkan::compute_list_bind_uniform_set(ComputeListID p_list, RID p_uniform_set, uint32_t p_index) { // Must be called within a compute list, the class mutex is locked during that time ERR_FAIL_COND(p_list != ID_TYPE_COMPUTE_LIST); ERR_FAIL_NULL(compute_list); ComputeList *cl = compute_list; #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(p_index >= limits.maxBoundDescriptorSets || p_index >= MAX_UNIFORM_SETS, "Attempting to bind a descriptor set (" + itos(p_index) + ") greater than what the hardware supports (" + itos(limits.maxBoundDescriptorSets) + ")."); #endif #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!cl->validation.active, "Submitted Compute Lists can no longer be modified."); #endif UniformSet *uniform_set = uniform_set_owner.get_or_null(p_uniform_set); ERR_FAIL_NULL(uniform_set); if (p_index > cl->state.set_count) { cl->state.set_count = p_index; } cl->state.sets[p_index].descriptor_set = uniform_set->descriptor_set; // Update set pointer. cl->state.sets[p_index].bound = false; // Needs rebind. cl->state.sets[p_index].uniform_set_format = uniform_set->format; cl->state.sets[p_index].uniform_set = p_uniform_set; uint32_t textures_to_sampled_count = uniform_set->mutable_sampled_textures.size(); uint32_t textures_to_storage_count = uniform_set->mutable_storage_textures.size(); Texture **textures_to_sampled = uniform_set->mutable_sampled_textures.ptrw(); VkImageMemoryBarrier *texture_barriers = nullptr; if (textures_to_sampled_count + textures_to_storage_count) { texture_barriers = (VkImageMemoryBarrier *)alloca(sizeof(VkImageMemoryBarrier) * (textures_to_sampled_count + textures_to_storage_count)); } uint32_t texture_barrier_count = 0; uint32_t src_stage_flags = 0; for (uint32_t i = 0; i < textures_to_sampled_count; i++) { if (textures_to_sampled[i]->layout != VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL) { src_stage_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; VkImageMemoryBarrier &image_memory_barrier = texture_barriers[texture_barrier_count++]; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; image_memory_barrier.oldLayout = textures_to_sampled[i]->layout; image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = textures_to_sampled[i]->image; image_memory_barrier.subresourceRange.aspectMask = textures_to_sampled[i]->read_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = textures_to_sampled[i]->base_mipmap; image_memory_barrier.subresourceRange.levelCount = textures_to_sampled[i]->mipmaps; image_memory_barrier.subresourceRange.baseArrayLayer = textures_to_sampled[i]->base_layer; image_memory_barrier.subresourceRange.layerCount = textures_to_sampled[i]->layers; textures_to_sampled[i]->layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; cl->state.textures_to_sampled_layout.erase(textures_to_sampled[i]); } if (textures_to_sampled[i]->used_in_frame != frames_drawn) { textures_to_sampled[i]->used_in_frame = frames_drawn; textures_to_sampled[i]->used_in_transfer = false; textures_to_sampled[i]->used_in_raster = false; } textures_to_sampled[i]->used_in_compute = true; } Texture **textures_to_storage = uniform_set->mutable_storage_textures.ptrw(); for (uint32_t i = 0; i < textures_to_storage_count; i++) { if (textures_to_storage[i]->layout != VK_IMAGE_LAYOUT_GENERAL) { uint32_t src_access_flags = 0; if (textures_to_storage[i]->used_in_frame == frames_drawn) { if (textures_to_storage[i]->used_in_compute) { src_stage_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; src_access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (textures_to_storage[i]->used_in_raster) { src_stage_flags |= VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_VERTEX_SHADER_BIT; src_access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (textures_to_storage[i]->used_in_transfer) { src_stage_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; src_access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_TRANSFER_READ_BIT; } textures_to_storage[i]->used_in_compute = false; textures_to_storage[i]->used_in_raster = false; textures_to_storage[i]->used_in_transfer = false; } else { src_access_flags = 0; textures_to_storage[i]->used_in_compute = false; textures_to_storage[i]->used_in_raster = false; textures_to_storage[i]->used_in_transfer = false; textures_to_storage[i]->used_in_frame = frames_drawn; } VkImageMemoryBarrier &image_memory_barrier = texture_barriers[texture_barrier_count++]; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = src_access_flags; image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; image_memory_barrier.oldLayout = textures_to_storage[i]->layout; image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_GENERAL; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = textures_to_storage[i]->image; image_memory_barrier.subresourceRange.aspectMask = textures_to_storage[i]->read_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = textures_to_storage[i]->base_mipmap; image_memory_barrier.subresourceRange.levelCount = textures_to_storage[i]->mipmaps; image_memory_barrier.subresourceRange.baseArrayLayer = textures_to_storage[i]->base_layer; image_memory_barrier.subresourceRange.layerCount = textures_to_storage[i]->layers; textures_to_storage[i]->layout = VK_IMAGE_LAYOUT_GENERAL; cl->state.textures_to_sampled_layout.insert(textures_to_storage[i]); // Needs to go back to sampled layout afterwards. } } if (texture_barrier_count) { if (src_stage_flags == 0) { src_stage_flags = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT; } vkCmdPipelineBarrier(cl->command_buffer, src_stage_flags, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, texture_barrier_count, texture_barriers); } #if 0 { // Validate that textures bound are not attached as framebuffer bindings. uint32_t attachable_count = uniform_set->attachable_textures.size(); const RID *attachable_ptr = uniform_set->attachable_textures.ptr(); uint32_t bound_count = draw_list_bound_textures.size(); const RID *bound_ptr = draw_list_bound_textures.ptr(); for (uint32_t i = 0; i < attachable_count; i++) { for (uint32_t j = 0; j < bound_count; j++) { ERR_FAIL_COND_MSG(attachable_ptr[i] == bound_ptr[j], "Attempted to use the same texture in framebuffer attachment and a uniform set, this is not allowed."); } } } #endif } void RenderingDeviceVulkan::compute_list_set_push_constant(ComputeListID p_list, const void *p_data, uint32_t p_data_size) { ERR_FAIL_COND(p_list != ID_TYPE_COMPUTE_LIST); ERR_FAIL_NULL(compute_list); ComputeList *cl = compute_list; #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!cl->validation.active, "Submitted Compute Lists can no longer be modified."); #endif #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(p_data_size != cl->validation.pipeline_push_constant_size, "This compute pipeline requires (" + itos(cl->validation.pipeline_push_constant_size) + ") bytes of push constant data, supplied: (" + itos(p_data_size) + ")"); #endif vkCmdPushConstants(cl->command_buffer, cl->state.pipeline_layout, cl->state.pipeline_push_constant_stages, 0, p_data_size, p_data); #ifdef DEBUG_ENABLED cl->validation.pipeline_push_constant_supplied = true; #endif } void RenderingDeviceVulkan::compute_list_dispatch(ComputeListID p_list, uint32_t p_x_groups, uint32_t p_y_groups, uint32_t p_z_groups) { // Must be called within a compute list, the class mutex is locked during that time ERR_FAIL_COND(p_list != ID_TYPE_COMPUTE_LIST); ERR_FAIL_NULL(compute_list); ComputeList *cl = compute_list; #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(p_x_groups == 0, "Dispatch amount of X compute groups (" + itos(p_x_groups) + ") is zero."); ERR_FAIL_COND_MSG(p_z_groups == 0, "Dispatch amount of Z compute groups (" + itos(p_z_groups) + ") is zero."); ERR_FAIL_COND_MSG(p_y_groups == 0, "Dispatch amount of Y compute groups (" + itos(p_y_groups) + ") is zero."); ERR_FAIL_COND_MSG(p_x_groups > limits.maxComputeWorkGroupCount[0], "Dispatch amount of X compute groups (" + itos(p_x_groups) + ") is larger than device limit (" + itos(limits.maxComputeWorkGroupCount[0]) + ")"); ERR_FAIL_COND_MSG(p_y_groups > limits.maxComputeWorkGroupCount[1], "Dispatch amount of Y compute groups (" + itos(p_y_groups) + ") is larger than device limit (" + itos(limits.maxComputeWorkGroupCount[1]) + ")"); ERR_FAIL_COND_MSG(p_z_groups > limits.maxComputeWorkGroupCount[2], "Dispatch amount of Z compute groups (" + itos(p_z_groups) + ") is larger than device limit (" + itos(limits.maxComputeWorkGroupCount[2]) + ")"); ERR_FAIL_COND_MSG(!cl->validation.active, "Submitted Compute Lists can no longer be modified."); #endif #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!cl->validation.pipeline_active, "No compute pipeline was set before attempting to draw."); if (cl->validation.pipeline_push_constant_size > 0) { // Using push constants, check that they were supplied. ERR_FAIL_COND_MSG(!cl->validation.pipeline_push_constant_supplied, "The shader in this pipeline requires a push constant to be set before drawing, but it's not present."); } #endif // Bind descriptor sets. for (uint32_t i = 0; i < cl->state.set_count; i++) { if (cl->state.sets[i].pipeline_expected_format == 0) { continue; // Nothing expected by this pipeline. } #ifdef DEBUG_ENABLED if (cl->state.sets[i].pipeline_expected_format != cl->state.sets[i].uniform_set_format) { if (cl->state.sets[i].uniform_set_format == 0) { ERR_FAIL_MSG("Uniforms were never supplied for set (" + itos(i) + ") at the time of drawing, which are required by the pipeline"); } else if (uniform_set_owner.owns(cl->state.sets[i].uniform_set)) { UniformSet *us = uniform_set_owner.get_or_null(cl->state.sets[i].uniform_set); ERR_FAIL_MSG("Uniforms supplied for set (" + itos(i) + "):\n" + _shader_uniform_debug(us->shader_id, us->shader_set) + "\nare not the same format as required by the pipeline shader. Pipeline shader requires the following bindings:\n" + _shader_uniform_debug(cl->state.pipeline_shader)); } else { ERR_FAIL_MSG("Uniforms supplied for set (" + itos(i) + ", which was was just freed) are not the same format as required by the pipeline shader. Pipeline shader requires the following bindings:\n" + _shader_uniform_debug(cl->state.pipeline_shader)); } } #endif if (!cl->state.sets[i].bound) { // All good, see if this requires re-binding. vkCmdBindDescriptorSets(cl->command_buffer, VK_PIPELINE_BIND_POINT_COMPUTE, cl->state.pipeline_layout, i, 1, &cl->state.sets[i].descriptor_set, 0, nullptr); cl->state.sets[i].bound = true; } } vkCmdDispatch(cl->command_buffer, p_x_groups, p_y_groups, p_z_groups); } void RenderingDeviceVulkan::compute_list_dispatch_threads(ComputeListID p_list, uint32_t p_x_threads, uint32_t p_y_threads, uint32_t p_z_threads) { ERR_FAIL_COND(p_list != ID_TYPE_COMPUTE_LIST); ERR_FAIL_NULL(compute_list); #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(p_x_threads == 0, "Dispatch amount of X compute threads (" + itos(p_x_threads) + ") is zero."); ERR_FAIL_COND_MSG(p_y_threads == 0, "Dispatch amount of Y compute threads (" + itos(p_y_threads) + ") is zero."); ERR_FAIL_COND_MSG(p_z_threads == 0, "Dispatch amount of Z compute threads (" + itos(p_z_threads) + ") is zero."); #endif ComputeList *cl = compute_list; #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!cl->validation.pipeline_active, "No compute pipeline was set before attempting to draw."); if (cl->validation.pipeline_push_constant_size > 0) { // Using push constants, check that they were supplied. ERR_FAIL_COND_MSG(!cl->validation.pipeline_push_constant_supplied, "The shader in this pipeline requires a push constant to be set before drawing, but it's not present."); } #endif compute_list_dispatch(p_list, (p_x_threads - 1) / cl->state.local_group_size[0] + 1, (p_y_threads - 1) / cl->state.local_group_size[1] + 1, (p_z_threads - 1) / cl->state.local_group_size[2] + 1); } void RenderingDeviceVulkan::compute_list_dispatch_indirect(ComputeListID p_list, RID p_buffer, uint32_t p_offset) { ERR_FAIL_COND(p_list != ID_TYPE_COMPUTE_LIST); ERR_FAIL_NULL(compute_list); ComputeList *cl = compute_list; Buffer *buffer = storage_buffer_owner.get_or_null(p_buffer); ERR_FAIL_NULL(buffer); ERR_FAIL_COND_MSG(!(buffer->usage & VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT), "Buffer provided was not created to do indirect dispatch."); ERR_FAIL_COND_MSG(p_offset + 12 > buffer->size, "Offset provided (+12) is past the end of buffer."); #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!cl->validation.active, "Submitted Compute Lists can no longer be modified."); #endif #ifdef DEBUG_ENABLED ERR_FAIL_COND_MSG(!cl->validation.pipeline_active, "No compute pipeline was set before attempting to draw."); if (cl->validation.pipeline_push_constant_size > 0) { // Using push constants, check that they were supplied. ERR_FAIL_COND_MSG(!cl->validation.pipeline_push_constant_supplied, "The shader in this pipeline requires a push constant to be set before drawing, but it's not present."); } #endif // Bind descriptor sets. for (uint32_t i = 0; i < cl->state.set_count; i++) { if (cl->state.sets[i].pipeline_expected_format == 0) { continue; // Nothing expected by this pipeline. } #ifdef DEBUG_ENABLED if (cl->state.sets[i].pipeline_expected_format != cl->state.sets[i].uniform_set_format) { if (cl->state.sets[i].uniform_set_format == 0) { ERR_FAIL_MSG("Uniforms were never supplied for set (" + itos(i) + ") at the time of drawing, which are required by the pipeline"); } else if (uniform_set_owner.owns(cl->state.sets[i].uniform_set)) { UniformSet *us = uniform_set_owner.get_or_null(cl->state.sets[i].uniform_set); ERR_FAIL_MSG("Uniforms supplied for set (" + itos(i) + "):\n" + _shader_uniform_debug(us->shader_id, us->shader_set) + "\nare not the same format as required by the pipeline shader. Pipeline shader requires the following bindings:\n" + _shader_uniform_debug(cl->state.pipeline_shader)); } else { ERR_FAIL_MSG("Uniforms supplied for set (" + itos(i) + ", which was was just freed) are not the same format as required by the pipeline shader. Pipeline shader requires the following bindings:\n" + _shader_uniform_debug(cl->state.pipeline_shader)); } } #endif if (!cl->state.sets[i].bound) { // All good, see if this requires re-binding. vkCmdBindDescriptorSets(cl->command_buffer, VK_PIPELINE_BIND_POINT_COMPUTE, cl->state.pipeline_layout, i, 1, &cl->state.sets[i].descriptor_set, 0, nullptr); cl->state.sets[i].bound = true; } } vkCmdDispatchIndirect(cl->command_buffer, buffer->buffer, p_offset); } void RenderingDeviceVulkan::compute_list_add_barrier(ComputeListID p_list) { // Must be called within a compute list, the class mutex is locked during that time uint32_t barrier_flags = VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; uint32_t access_flags = VK_ACCESS_SHADER_READ_BIT; _compute_list_add_barrier(BARRIER_MASK_COMPUTE, barrier_flags, access_flags); } void RenderingDeviceVulkan::_compute_list_add_barrier(BitField<BarrierMask> p_post_barrier, uint32_t p_barrier_flags, uint32_t p_access_flags) { ERR_FAIL_NULL(compute_list); VkImageMemoryBarrier *image_barriers = nullptr; uint32_t image_barrier_count = compute_list->state.textures_to_sampled_layout.size(); if (image_barrier_count) { image_barriers = (VkImageMemoryBarrier *)alloca(sizeof(VkImageMemoryBarrier) * image_barrier_count); } image_barrier_count = 0; // We'll count how many we end up issuing. for (Texture *E : compute_list->state.textures_to_sampled_layout) { if (E->layout != VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL) { VkImageMemoryBarrier &image_memory_barrier = image_barriers[image_barrier_count++]; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; image_memory_barrier.dstAccessMask = p_access_flags; image_memory_barrier.oldLayout = E->layout; image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.image = E->image; image_memory_barrier.subresourceRange.aspectMask = E->read_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = E->base_mipmap; image_memory_barrier.subresourceRange.levelCount = E->mipmaps; image_memory_barrier.subresourceRange.baseArrayLayer = E->base_layer; image_memory_barrier.subresourceRange.layerCount = E->layers; E->layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; } if (E->used_in_frame != frames_drawn) { E->used_in_transfer = false; E->used_in_raster = false; E->used_in_compute = false; E->used_in_frame = frames_drawn; } } if (p_barrier_flags) { VkMemoryBarrier mem_barrier; mem_barrier.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER; mem_barrier.pNext = nullptr; mem_barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT; mem_barrier.dstAccessMask = p_access_flags; vkCmdPipelineBarrier(compute_list->command_buffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, p_barrier_flags, 0, 1, &mem_barrier, 0, nullptr, image_barrier_count, image_barriers); } else if (image_barrier_count) { vkCmdPipelineBarrier(compute_list->command_buffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, 0, 0, nullptr, 0, nullptr, image_barrier_count, image_barriers); } #ifdef FORCE_FULL_BARRIER _full_barrier(true); #endif } void RenderingDeviceVulkan::compute_list_end(BitField<BarrierMask> p_post_barrier) { ERR_FAIL_NULL(compute_list); uint32_t barrier_flags = 0; uint32_t access_flags = 0; if (p_post_barrier.has_flag(BARRIER_MASK_COMPUTE)) { barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_VERTEX)) { barrier_flags |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_FRAGMENT)) { barrier_flags |= VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT; } if (p_post_barrier.has_flag(BARRIER_MASK_TRANSFER)) { barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_TRANSFER_READ_BIT; } _compute_list_add_barrier(p_post_barrier, barrier_flags, access_flags); memdelete(compute_list); compute_list = nullptr; // Compute_list is no longer active. _THREAD_SAFE_UNLOCK_ } void RenderingDeviceVulkan::barrier(BitField<BarrierMask> p_from, BitField<BarrierMask> p_to) { uint32_t src_barrier_flags = 0; uint32_t src_access_flags = 0; if (p_from == 0) { src_barrier_flags = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT; } else { if (p_from.has_flag(BARRIER_MASK_COMPUTE)) { src_barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; src_access_flags |= VK_ACCESS_SHADER_WRITE_BIT; } if (p_from.has_flag(BARRIER_MASK_FRAGMENT)) { src_barrier_flags |= VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT; src_access_flags |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT; } if (p_from.has_flag(BARRIER_MASK_TRANSFER)) { src_barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; src_access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT; } } uint32_t dst_barrier_flags = 0; uint32_t dst_access_flags = 0; if (p_to == 0) { dst_barrier_flags = VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT; } else { if (p_to.has_flag(BARRIER_MASK_COMPUTE)) { dst_barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; dst_access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_to.has_flag(BARRIER_MASK_VERTEX)) { dst_barrier_flags |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT; dst_access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT; } if (p_to.has_flag(BARRIER_MASK_FRAGMENT)) { dst_barrier_flags |= VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT; dst_access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT; } if (p_to.has_flag(BARRIER_MASK_TRANSFER)) { dst_barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; dst_access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_TRANSFER_READ_BIT; } } _memory_barrier(src_barrier_flags, dst_barrier_flags, src_access_flags, dst_access_flags, true); } void RenderingDeviceVulkan::full_barrier() { #ifndef DEBUG_ENABLED ERR_PRINT("Full barrier is debug-only, should not be used in production"); #endif _full_barrier(true); } #if 0 void RenderingDeviceVulkan::draw_list_render_secondary_to_framebuffer(ID p_framebuffer, ID *p_draw_lists, uint32_t p_draw_list_count, InitialAction p_initial_action, FinalAction p_final_action, const Vector<Variant> &p_clear_colors) { VkCommandBuffer frame_cmdbuf = frames[frame].frame_buffer; ERR_FAIL_NULL(frame_cmdbuf); VkRenderPassBeginInfo render_pass_begin; render_pass_begin.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO; render_pass_begin.pNext = nullptr; render_pass_begin.renderPass = context->get_render_pass(); render_pass_begin.framebuffer = context->get_frame_framebuffer(frame); render_pass_begin.renderArea.extent.width = context->get_screen_width(p_screen); render_pass_begin.renderArea.extent.height = context->get_screen_height(p_screen); render_pass_begin.renderArea.offset.x = 0; render_pass_begin.renderArea.offset.y = 0; render_pass_begin.clearValueCount = 1; VkClearValue clear_value; clear_value.color.float32[0] = p_clear_color.r; clear_value.color.float32[1] = p_clear_color.g; clear_value.color.float32[2] = p_clear_color.b; clear_value.color.float32[3] = p_clear_color.a; render_pass_begin.pClearValues = &clear_value; vkCmdBeginRenderPass(frame_cmdbuf, &render_pass_begin, VK_SUBPASS_CONTENTS_SECONDARY_COMMAND_BUFFERS); ID screen_format = screen_get_framebuffer_format(); { VkCommandBuffer *command_buffers = (VkCommandBuffer *)alloca(sizeof(VkCommandBuffer) * p_draw_list_count); uint32_t command_buffer_count = 0; for (uint32_t i = 0; i < p_draw_list_count; i++) { DrawList *dl = _get_draw_list_ptr(p_draw_lists[i]); ERR_CONTINUE_MSG(!dl, "Draw list index (" + itos(i) + ") is not a valid draw list ID."); ERR_CONTINUE_MSG(dl->validation.framebuffer_format != p_format_check, "Draw list index (" + itos(i) + ") is created with a framebuffer format incompatible with this render pass."); if (dl->validation.active) { // Needs to be closed, so close it. vkEndCommandBuffer(dl->command_buffer); dl->validation.active = false; } command_buffers[command_buffer_count++] = dl->command_buffer; } print_line("to draw: " + itos(command_buffer_count)); vkCmdExecuteCommands(p_primary, command_buffer_count, command_buffers); } vkCmdEndRenderPass(frame_cmdbuf); } #endif void RenderingDeviceVulkan::_free_internal(RID p_id) { #ifdef DEV_ENABLED String resource_name; if (resource_names.has(p_id)) { resource_name = resource_names[p_id]; resource_names.erase(p_id); } #endif // Push everything so it's disposed of next time this frame index is processed (means, it's safe to do it). if (texture_owner.owns(p_id)) { Texture *texture = texture_owner.get_or_null(p_id); frames[frame].textures_to_dispose_of.push_back(*texture); texture_owner.free(p_id); } else if (framebuffer_owner.owns(p_id)) { Framebuffer *framebuffer = framebuffer_owner.get_or_null(p_id); frames[frame].framebuffers_to_dispose_of.push_back(*framebuffer); if (framebuffer->invalidated_callback != nullptr) { framebuffer->invalidated_callback(framebuffer->invalidated_callback_userdata); } framebuffer_owner.free(p_id); } else if (sampler_owner.owns(p_id)) { VkSampler *sampler = sampler_owner.get_or_null(p_id); frames[frame].samplers_to_dispose_of.push_back(*sampler); sampler_owner.free(p_id); } else if (vertex_buffer_owner.owns(p_id)) { Buffer *vertex_buffer = vertex_buffer_owner.get_or_null(p_id); frames[frame].buffers_to_dispose_of.push_back(*vertex_buffer); vertex_buffer_owner.free(p_id); } else if (vertex_array_owner.owns(p_id)) { vertex_array_owner.free(p_id); } else if (index_buffer_owner.owns(p_id)) { IndexBuffer *index_buffer = index_buffer_owner.get_or_null(p_id); Buffer b; b.allocation = index_buffer->allocation; b.buffer = index_buffer->buffer; b.size = index_buffer->size; b.buffer_info = {}; frames[frame].buffers_to_dispose_of.push_back(b); index_buffer_owner.free(p_id); } else if (index_array_owner.owns(p_id)) { index_array_owner.free(p_id); } else if (shader_owner.owns(p_id)) { Shader *shader = shader_owner.get_or_null(p_id); frames[frame].shaders_to_dispose_of.push_back(*shader); shader_owner.free(p_id); } else if (uniform_buffer_owner.owns(p_id)) { Buffer *uniform_buffer = uniform_buffer_owner.get_or_null(p_id); frames[frame].buffers_to_dispose_of.push_back(*uniform_buffer); uniform_buffer_owner.free(p_id); } else if (texture_buffer_owner.owns(p_id)) { TextureBuffer *texture_buffer = texture_buffer_owner.get_or_null(p_id); frames[frame].buffers_to_dispose_of.push_back(texture_buffer->buffer); frames[frame].buffer_views_to_dispose_of.push_back(texture_buffer->view); texture_buffer_owner.free(p_id); } else if (storage_buffer_owner.owns(p_id)) { Buffer *storage_buffer = storage_buffer_owner.get_or_null(p_id); frames[frame].buffers_to_dispose_of.push_back(*storage_buffer); storage_buffer_owner.free(p_id); } else if (uniform_set_owner.owns(p_id)) { UniformSet *uniform_set = uniform_set_owner.get_or_null(p_id); frames[frame].uniform_sets_to_dispose_of.push_back(*uniform_set); uniform_set_owner.free(p_id); if (uniform_set->invalidated_callback != nullptr) { uniform_set->invalidated_callback(uniform_set->invalidated_callback_userdata); } } else if (render_pipeline_owner.owns(p_id)) { RenderPipeline *pipeline = render_pipeline_owner.get_or_null(p_id); frames[frame].render_pipelines_to_dispose_of.push_back(*pipeline); render_pipeline_owner.free(p_id); } else if (compute_pipeline_owner.owns(p_id)) { ComputePipeline *pipeline = compute_pipeline_owner.get_or_null(p_id); frames[frame].compute_pipelines_to_dispose_of.push_back(*pipeline); compute_pipeline_owner.free(p_id); } else { #ifdef DEV_ENABLED ERR_PRINT("Attempted to free invalid ID: " + itos(p_id.get_id()) + " " + resource_name); #else ERR_PRINT("Attempted to free invalid ID: " + itos(p_id.get_id())); #endif } } void RenderingDeviceVulkan::free(RID p_id) { _THREAD_SAFE_METHOD_ _free_dependencies(p_id); // Recursively erase dependencies first, to avoid potential API problems. _free_internal(p_id); } // The full list of resources that can be named is in the VkObjectType enum. // We just expose the resources that are owned and can be accessed easily. void RenderingDeviceVulkan::set_resource_name(RID p_id, const String p_name) { if (texture_owner.owns(p_id)) { Texture *texture = texture_owner.get_or_null(p_id); if (texture->owner.is_null()) { // Don't set the source texture's name when calling on a texture view. context->set_object_name(VK_OBJECT_TYPE_IMAGE, uint64_t(texture->image), p_name); } context->set_object_name(VK_OBJECT_TYPE_IMAGE_VIEW, uint64_t(texture->view), p_name + " View"); } else if (framebuffer_owner.owns(p_id)) { //Framebuffer *framebuffer = framebuffer_owner.get_or_null(p_id); // Not implemented for now as the relationship between Framebuffer and RenderPass is very complex. } else if (sampler_owner.owns(p_id)) { VkSampler *sampler = sampler_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_SAMPLER, uint64_t(*sampler), p_name); } else if (vertex_buffer_owner.owns(p_id)) { Buffer *vertex_buffer = vertex_buffer_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_BUFFER, uint64_t(vertex_buffer->buffer), p_name); } else if (index_buffer_owner.owns(p_id)) { IndexBuffer *index_buffer = index_buffer_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_BUFFER, uint64_t(index_buffer->buffer), p_name); } else if (shader_owner.owns(p_id)) { Shader *shader = shader_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_PIPELINE_LAYOUT, uint64_t(shader->pipeline_layout), p_name + " Pipeline Layout"); for (int i = 0; i < shader->sets.size(); i++) { context->set_object_name(VK_OBJECT_TYPE_DESCRIPTOR_SET_LAYOUT, uint64_t(shader->sets[i].descriptor_set_layout), p_name); } } else if (uniform_buffer_owner.owns(p_id)) { Buffer *uniform_buffer = uniform_buffer_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_BUFFER, uint64_t(uniform_buffer->buffer), p_name); } else if (texture_buffer_owner.owns(p_id)) { TextureBuffer *texture_buffer = texture_buffer_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_BUFFER, uint64_t(texture_buffer->buffer.buffer), p_name); context->set_object_name(VK_OBJECT_TYPE_BUFFER_VIEW, uint64_t(texture_buffer->view), p_name + " View"); } else if (storage_buffer_owner.owns(p_id)) { Buffer *storage_buffer = storage_buffer_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_BUFFER, uint64_t(storage_buffer->buffer), p_name); } else if (uniform_set_owner.owns(p_id)) { UniformSet *uniform_set = uniform_set_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_DESCRIPTOR_SET, uint64_t(uniform_set->descriptor_set), p_name); } else if (render_pipeline_owner.owns(p_id)) { RenderPipeline *pipeline = render_pipeline_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_PIPELINE, uint64_t(pipeline->pipeline), p_name); context->set_object_name(VK_OBJECT_TYPE_PIPELINE_LAYOUT, uint64_t(pipeline->pipeline_layout), p_name + " Layout"); } else if (compute_pipeline_owner.owns(p_id)) { ComputePipeline *pipeline = compute_pipeline_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_PIPELINE, uint64_t(pipeline->pipeline), p_name); context->set_object_name(VK_OBJECT_TYPE_PIPELINE_LAYOUT, uint64_t(pipeline->pipeline_layout), p_name + " Layout"); } else { ERR_PRINT("Attempted to name invalid ID: " + itos(p_id.get_id())); return; } #ifdef DEV_ENABLED resource_names[p_id] = p_name; #endif } void RenderingDeviceVulkan::draw_command_begin_label(String p_label_name, const Color p_color) { _THREAD_SAFE_METHOD_ context->command_begin_label(frames[frame].draw_command_buffer, p_label_name, p_color); } void RenderingDeviceVulkan::draw_command_insert_label(String p_label_name, const Color p_color) { _THREAD_SAFE_METHOD_ context->command_insert_label(frames[frame].draw_command_buffer, p_label_name, p_color); } void RenderingDeviceVulkan::draw_command_end_label() { _THREAD_SAFE_METHOD_ context->command_end_label(frames[frame].draw_command_buffer); } String RenderingDeviceVulkan::get_device_vendor_name() const { return context->get_device_vendor_name(); } String RenderingDeviceVulkan::get_device_name() const { return context->get_device_name(); } RenderingDevice::DeviceType RenderingDeviceVulkan::get_device_type() const { return context->get_device_type(); } String RenderingDeviceVulkan::get_device_api_version() const { return context->get_device_api_version(); } String RenderingDeviceVulkan::get_device_pipeline_cache_uuid() const { return context->get_device_pipeline_cache_uuid(); } void RenderingDeviceVulkan::_finalize_command_bufers() { if (draw_list) { ERR_PRINT("Found open draw list at the end of the frame, this should never happen (further drawing will likely not work)."); } if (compute_list) { ERR_PRINT("Found open compute list at the end of the frame, this should never happen (further compute will likely not work)."); } { // Complete the setup buffer (that needs to be processed before anything else). vkEndCommandBuffer(frames[frame].setup_command_buffer); vkEndCommandBuffer(frames[frame].draw_command_buffer); } } void RenderingDeviceVulkan::_begin_frame() { // Erase pending resources. _free_pending_resources(frame); // Create setup command buffer and set as the setup buffer. { VkCommandBufferBeginInfo cmdbuf_begin; cmdbuf_begin.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO; cmdbuf_begin.pNext = nullptr; cmdbuf_begin.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT; cmdbuf_begin.pInheritanceInfo = nullptr; VkResult err = vkResetCommandBuffer(frames[frame].setup_command_buffer, 0); ERR_FAIL_COND_MSG(err, "vkResetCommandBuffer failed with error " + itos(err) + "."); err = vkBeginCommandBuffer(frames[frame].setup_command_buffer, &cmdbuf_begin); ERR_FAIL_COND_MSG(err, "vkBeginCommandBuffer failed with error " + itos(err) + "."); err = vkBeginCommandBuffer(frames[frame].draw_command_buffer, &cmdbuf_begin); ERR_FAIL_COND_MSG(err, "vkBeginCommandBuffer failed with error " + itos(err) + "."); if (local_device.is_null()) { context->append_command_buffer(frames[frame].draw_command_buffer); context->set_setup_buffer(frames[frame].setup_command_buffer); // Append now so it's added before everything else. } } // Advance current frame. frames_drawn++; // Advance staging buffer if used. if (staging_buffer_used) { staging_buffer_current = (staging_buffer_current + 1) % staging_buffer_blocks.size(); staging_buffer_used = false; } if (frames[frame].timestamp_count) { vkGetQueryPoolResults(device, frames[frame].timestamp_pool, 0, frames[frame].timestamp_count, sizeof(uint64_t) * max_timestamp_query_elements, frames[frame].timestamp_result_values.ptr(), sizeof(uint64_t), VK_QUERY_RESULT_64_BIT); vkCmdResetQueryPool(frames[frame].setup_command_buffer, frames[frame].timestamp_pool, 0, frames[frame].timestamp_count); SWAP(frames[frame].timestamp_names, frames[frame].timestamp_result_names); SWAP(frames[frame].timestamp_cpu_values, frames[frame].timestamp_cpu_result_values); } frames[frame].timestamp_result_count = frames[frame].timestamp_count; frames[frame].timestamp_count = 0; frames[frame].index = Engine::get_singleton()->get_frames_drawn(); } VkSampleCountFlagBits RenderingDeviceVulkan::_ensure_supported_sample_count(TextureSamples p_requested_sample_count) const { VkSampleCountFlags sample_count_flags = limits.framebufferColorSampleCounts & limits.framebufferDepthSampleCounts; if (sample_count_flags & rasterization_sample_count[p_requested_sample_count]) { // The requested sample count is supported. return rasterization_sample_count[p_requested_sample_count]; } else { // Find the closest lower supported sample count. VkSampleCountFlagBits sample_count = rasterization_sample_count[p_requested_sample_count]; while (sample_count > VK_SAMPLE_COUNT_1_BIT) { if (sample_count_flags & sample_count) { return sample_count; } sample_count = (VkSampleCountFlagBits)(sample_count >> 1); } } return VK_SAMPLE_COUNT_1_BIT; } void RenderingDeviceVulkan::swap_buffers() { ERR_FAIL_COND_MSG(local_device.is_valid(), "Local devices can't swap buffers."); _THREAD_SAFE_METHOD_ _finalize_command_bufers(); screen_prepared = false; // Swap buffers. context->swap_buffers(); frame = (frame + 1) % frame_count; _begin_frame(); } void RenderingDeviceVulkan::submit() { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_MSG(local_device.is_null(), "Only local devices can submit and sync."); ERR_FAIL_COND_MSG(local_device_processing, "device already submitted, call sync to wait until done."); _finalize_command_bufers(); VkCommandBuffer command_buffers[2] = { frames[frame].setup_command_buffer, frames[frame].draw_command_buffer }; context->local_device_push_command_buffers(local_device, command_buffers, 2); local_device_processing = true; } void RenderingDeviceVulkan::sync() { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_MSG(local_device.is_null(), "Only local devices can submit and sync."); ERR_FAIL_COND_MSG(!local_device_processing, "sync can only be called after a submit"); context->local_device_sync(local_device); _begin_frame(); local_device_processing = false; } VmaPool RenderingDeviceVulkan::_find_or_create_small_allocs_pool(uint32_t p_mem_type_index) { if (small_allocs_pools.has(p_mem_type_index)) { return small_allocs_pools[p_mem_type_index]; } print_verbose("Creating VMA small objects pool for memory type index " + itos(p_mem_type_index)); VmaPoolCreateInfo pci; pci.memoryTypeIndex = p_mem_type_index; pci.flags = 0; pci.blockSize = 0; pci.minBlockCount = 0; pci.maxBlockCount = SIZE_MAX; pci.priority = 0.5f; pci.minAllocationAlignment = 0; pci.pMemoryAllocateNext = nullptr; VmaPool pool = VK_NULL_HANDLE; VkResult res = vmaCreatePool(allocator, &pci, &pool); small_allocs_pools[p_mem_type_index] = pool; // Don't try to create it again if failed the first time. ERR_FAIL_COND_V_MSG(res, pool, "vmaCreatePool failed with error " + itos(res) + "."); return pool; } void RenderingDeviceVulkan::_free_pending_resources(int p_frame) { // Free in dependency usage order, so nothing weird happens. // Pipelines. while (frames[p_frame].render_pipelines_to_dispose_of.front()) { RenderPipeline *pipeline = &frames[p_frame].render_pipelines_to_dispose_of.front()->get(); vkDestroyPipeline(device, pipeline->pipeline, nullptr); frames[p_frame].render_pipelines_to_dispose_of.pop_front(); } while (frames[p_frame].compute_pipelines_to_dispose_of.front()) { ComputePipeline *pipeline = &frames[p_frame].compute_pipelines_to_dispose_of.front()->get(); vkDestroyPipeline(device, pipeline->pipeline, nullptr); frames[p_frame].compute_pipelines_to_dispose_of.pop_front(); } // Uniform sets. while (frames[p_frame].uniform_sets_to_dispose_of.front()) { UniformSet *uniform_set = &frames[p_frame].uniform_sets_to_dispose_of.front()->get(); vkFreeDescriptorSets(device, uniform_set->pool->pool, 1, &uniform_set->descriptor_set); _descriptor_pool_free(uniform_set->pool_key, uniform_set->pool); frames[p_frame].uniform_sets_to_dispose_of.pop_front(); } // Buffer views. while (frames[p_frame].buffer_views_to_dispose_of.front()) { VkBufferView buffer_view = frames[p_frame].buffer_views_to_dispose_of.front()->get(); vkDestroyBufferView(device, buffer_view, nullptr); frames[p_frame].buffer_views_to_dispose_of.pop_front(); } // Shaders. while (frames[p_frame].shaders_to_dispose_of.front()) { Shader *shader = &frames[p_frame].shaders_to_dispose_of.front()->get(); // Descriptor set layout for each set. for (int i = 0; i < shader->sets.size(); i++) { vkDestroyDescriptorSetLayout(device, shader->sets[i].descriptor_set_layout, nullptr); } // Pipeline layout. vkDestroyPipelineLayout(device, shader->pipeline_layout, nullptr); // Shaders themselves. for (int i = 0; i < shader->pipeline_stages.size(); i++) { vkDestroyShaderModule(device, shader->pipeline_stages[i].module, nullptr); } frames[p_frame].shaders_to_dispose_of.pop_front(); } // Samplers. while (frames[p_frame].samplers_to_dispose_of.front()) { VkSampler sampler = frames[p_frame].samplers_to_dispose_of.front()->get(); vkDestroySampler(device, sampler, nullptr); frames[p_frame].samplers_to_dispose_of.pop_front(); } // Framebuffers. while (frames[p_frame].framebuffers_to_dispose_of.front()) { Framebuffer *framebuffer = &frames[p_frame].framebuffers_to_dispose_of.front()->get(); for (const KeyValue<Framebuffer::VersionKey, Framebuffer::Version> &E : framebuffer->framebuffers) { // First framebuffer, then render pass because it depends on it. vkDestroyFramebuffer(device, E.value.framebuffer, nullptr); vkDestroyRenderPass(device, E.value.render_pass, nullptr); } frames[p_frame].framebuffers_to_dispose_of.pop_front(); } // Textures. while (frames[p_frame].textures_to_dispose_of.front()) { Texture *texture = &frames[p_frame].textures_to_dispose_of.front()->get(); if (texture->bound) { WARN_PRINT("Deleted a texture while it was bound."); } vkDestroyImageView(device, texture->view, nullptr); if (texture->owner.is_null()) { // Actually owns the image and the allocation too. image_memory -= texture->allocation_info.size; vmaDestroyImage(allocator, texture->image, texture->allocation); } frames[p_frame].textures_to_dispose_of.pop_front(); } // Buffers. while (frames[p_frame].buffers_to_dispose_of.front()) { _buffer_free(&frames[p_frame].buffers_to_dispose_of.front()->get()); frames[p_frame].buffers_to_dispose_of.pop_front(); } } void RenderingDeviceVulkan::prepare_screen_for_drawing() { _THREAD_SAFE_METHOD_ context->prepare_buffers(); screen_prepared = true; } uint32_t RenderingDeviceVulkan::get_frame_delay() const { return frame_count; } uint64_t RenderingDeviceVulkan::get_memory_usage(MemoryType p_type) const { if (p_type == MEMORY_BUFFERS) { return buffer_memory; } else if (p_type == MEMORY_TEXTURES) { return image_memory; } else { VmaTotalStatistics stats; vmaCalculateStatistics(allocator, &stats); return stats.total.statistics.allocationBytes; } } void RenderingDeviceVulkan::_flush(bool p_current_frame) { if (local_device.is_valid() && !p_current_frame) { return; // Flushing previous frames has no effect with local device. } // Not doing this crashes RADV (undefined behavior). if (p_current_frame) { vkEndCommandBuffer(frames[frame].setup_command_buffer); vkEndCommandBuffer(frames[frame].draw_command_buffer); } if (local_device.is_valid()) { VkCommandBuffer command_buffers[2] = { frames[frame].setup_command_buffer, frames[frame].draw_command_buffer }; context->local_device_push_command_buffers(local_device, command_buffers, 2); context->local_device_sync(local_device); VkCommandBufferBeginInfo cmdbuf_begin; cmdbuf_begin.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO; cmdbuf_begin.pNext = nullptr; cmdbuf_begin.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT; cmdbuf_begin.pInheritanceInfo = nullptr; VkResult err = vkBeginCommandBuffer(frames[frame].setup_command_buffer, &cmdbuf_begin); ERR_FAIL_COND_MSG(err, "vkBeginCommandBuffer failed with error " + itos(err) + "."); err = vkBeginCommandBuffer(frames[frame].draw_command_buffer, &cmdbuf_begin); ERR_FAIL_COND_MSG(err, "vkBeginCommandBuffer failed with error " + itos(err) + "."); } else { context->flush(p_current_frame, p_current_frame); // Re-create the setup command. if (p_current_frame) { VkCommandBufferBeginInfo cmdbuf_begin; cmdbuf_begin.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO; cmdbuf_begin.pNext = nullptr; cmdbuf_begin.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT; cmdbuf_begin.pInheritanceInfo = nullptr; VkResult err = vkBeginCommandBuffer(frames[frame].setup_command_buffer, &cmdbuf_begin); ERR_FAIL_COND_MSG(err, "vkBeginCommandBuffer failed with error " + itos(err) + "."); context->set_setup_buffer(frames[frame].setup_command_buffer); // Append now so it's added before everything else. } if (p_current_frame) { VkCommandBufferBeginInfo cmdbuf_begin; cmdbuf_begin.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO; cmdbuf_begin.pNext = nullptr; cmdbuf_begin.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT; cmdbuf_begin.pInheritanceInfo = nullptr; VkResult err = vkBeginCommandBuffer(frames[frame].draw_command_buffer, &cmdbuf_begin); ERR_FAIL_COND_MSG(err, "vkBeginCommandBuffer failed with error " + itos(err) + "."); context->append_command_buffer(frames[frame].draw_command_buffer); } } } void RenderingDeviceVulkan::initialize(VulkanContext *p_context, bool p_local_device) { // Get our device capabilities. { device_capabilities.version_major = p_context->get_vulkan_major(); device_capabilities.version_minor = p_context->get_vulkan_minor(); } context = p_context; device = p_context->get_device(); if (p_local_device) { frame_count = 1; local_device = p_context->local_device_create(); device = p_context->local_device_get_vk_device(local_device); } else { frame_count = p_context->get_swapchain_image_count() + 1; // Always need one extra to ensure it's unused at any time, without having to use a fence for this. } limits = p_context->get_device_limits(); max_timestamp_query_elements = 256; { // Initialize allocator. VmaAllocatorCreateInfo allocatorInfo; memset(&allocatorInfo, 0, sizeof(VmaAllocatorCreateInfo)); allocatorInfo.physicalDevice = p_context->get_physical_device(); allocatorInfo.device = device; allocatorInfo.instance = p_context->get_instance(); vmaCreateAllocator(&allocatorInfo, &allocator); } frames.resize(frame_count); frame = 0; // Create setup and frame buffers. for (int i = 0; i < frame_count; i++) { frames[i].index = 0; { // Create command pool, one per frame is recommended. VkCommandPoolCreateInfo cmd_pool_info; cmd_pool_info.sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO; cmd_pool_info.pNext = nullptr; cmd_pool_info.queueFamilyIndex = p_context->get_graphics_queue_family_index(); cmd_pool_info.flags = VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT; VkResult res = vkCreateCommandPool(device, &cmd_pool_info, nullptr, &frames[i].command_pool); ERR_FAIL_COND_MSG(res, "vkCreateCommandPool failed with error " + itos(res) + "."); } { // Create command buffers. VkCommandBufferAllocateInfo cmdbuf; // No command buffer exists, create it. cmdbuf.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO; cmdbuf.pNext = nullptr; cmdbuf.commandPool = frames[i].command_pool; cmdbuf.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY; cmdbuf.commandBufferCount = 1; VkResult err = vkAllocateCommandBuffers(device, &cmdbuf, &frames[i].setup_command_buffer); ERR_CONTINUE_MSG(err, "vkAllocateCommandBuffers failed with error " + itos(err) + "."); err = vkAllocateCommandBuffers(device, &cmdbuf, &frames[i].draw_command_buffer); ERR_CONTINUE_MSG(err, "vkAllocateCommandBuffers failed with error " + itos(err) + "."); } { // Create query pool. VkQueryPoolCreateInfo query_pool_create_info; query_pool_create_info.sType = VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO; query_pool_create_info.flags = 0; query_pool_create_info.pNext = nullptr; query_pool_create_info.queryType = VK_QUERY_TYPE_TIMESTAMP; query_pool_create_info.queryCount = max_timestamp_query_elements; query_pool_create_info.pipelineStatistics = 0; vkCreateQueryPool(device, &query_pool_create_info, nullptr, &frames[i].timestamp_pool); frames[i].timestamp_names.resize(max_timestamp_query_elements); frames[i].timestamp_cpu_values.resize(max_timestamp_query_elements); frames[i].timestamp_count = 0; frames[i].timestamp_result_names.resize(max_timestamp_query_elements); frames[i].timestamp_cpu_result_values.resize(max_timestamp_query_elements); frames[i].timestamp_result_values.resize(max_timestamp_query_elements); frames[i].timestamp_result_count = 0; } } { // Begin the first command buffer for the first frame, so // setting up things can be done in the meantime until swap_buffers(), which is called before advance. VkCommandBufferBeginInfo cmdbuf_begin; cmdbuf_begin.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO; cmdbuf_begin.pNext = nullptr; cmdbuf_begin.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT; cmdbuf_begin.pInheritanceInfo = nullptr; VkResult err = vkBeginCommandBuffer(frames[0].setup_command_buffer, &cmdbuf_begin); ERR_FAIL_COND_MSG(err, "vkBeginCommandBuffer failed with error " + itos(err) + "."); err = vkBeginCommandBuffer(frames[0].draw_command_buffer, &cmdbuf_begin); ERR_FAIL_COND_MSG(err, "vkBeginCommandBuffer failed with error " + itos(err) + "."); if (local_device.is_null()) { context->set_setup_buffer(frames[0].setup_command_buffer); // Append now so it's added before everything else. context->append_command_buffer(frames[0].draw_command_buffer); } } for (int i = 0; i < frame_count; i++) { //Reset all queries in a query pool before doing any operations with them. vkCmdResetQueryPool(frames[0].setup_command_buffer, frames[i].timestamp_pool, 0, max_timestamp_query_elements); } staging_buffer_block_size = GLOBAL_GET("rendering/rendering_device/staging_buffer/block_size_kb"); staging_buffer_block_size = MAX(4u, staging_buffer_block_size); staging_buffer_block_size *= 1024; // Kb -> bytes. staging_buffer_max_size = GLOBAL_GET("rendering/rendering_device/staging_buffer/max_size_mb"); staging_buffer_max_size = MAX(1u, staging_buffer_max_size); staging_buffer_max_size *= 1024 * 1024; if (staging_buffer_max_size < staging_buffer_block_size * 4) { // Validate enough blocks. staging_buffer_max_size = staging_buffer_block_size * 4; } texture_upload_region_size_px = GLOBAL_GET("rendering/rendering_device/staging_buffer/texture_upload_region_size_px"); texture_upload_region_size_px = nearest_power_of_2_templated(texture_upload_region_size_px); frames_drawn = frame_count; // Start from frame count, so everything else is immediately old. // Ensure current staging block is valid and at least one per frame exists. staging_buffer_current = 0; staging_buffer_used = false; for (int i = 0; i < frame_count; i++) { // Staging was never used, create a block. Error err = _insert_staging_block(); ERR_CONTINUE(err != OK); } max_descriptors_per_pool = GLOBAL_GET("rendering/rendering_device/vulkan/max_descriptors_per_pool"); // Check to make sure DescriptorPoolKey is good. static_assert(sizeof(uint64_t) * 3 >= UNIFORM_TYPE_MAX * sizeof(uint16_t)); draw_list = nullptr; draw_list_count = 0; draw_list_split = false; compute_list = nullptr; pipelines_cache.file_path = "user://vulkan/pipelines"; pipelines_cache.file_path += "." + context->get_device_name().validate_filename().replace(" ", "_").to_lower(); if (Engine::get_singleton()->is_editor_hint()) { pipelines_cache.file_path += ".editor"; } pipelines_cache.file_path += ".cache"; // Prepare most fields now. VkPhysicalDeviceProperties props; vkGetPhysicalDeviceProperties(context->get_physical_device(), &props); pipelines_cache.header.magic = 868 + VK_PIPELINE_CACHE_HEADER_VERSION_ONE; pipelines_cache.header.device_id = props.deviceID; pipelines_cache.header.vendor_id = props.vendorID; pipelines_cache.header.driver_version = props.driverVersion; memcpy(pipelines_cache.header.uuid, props.pipelineCacheUUID, VK_UUID_SIZE); pipelines_cache.header.driver_abi = sizeof(void *); _load_pipeline_cache(); print_verbose(vformat("Startup PSO cache (%.1f MiB)", pipelines_cache.buffer.size() / (1024.0f * 1024.0f))); VkPipelineCacheCreateInfo cache_info = {}; cache_info.sType = VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO; cache_info.pNext = nullptr; if (context->get_pipeline_cache_control_support()) { cache_info.flags = VK_PIPELINE_CACHE_CREATE_EXTERNALLY_SYNCHRONIZED_BIT; } cache_info.initialDataSize = pipelines_cache.buffer.size(); cache_info.pInitialData = pipelines_cache.buffer.ptr(); VkResult err = vkCreatePipelineCache(device, &cache_info, nullptr, &pipelines_cache.cache_object); if (err != VK_SUCCESS) { WARN_PRINT("vkCreatePipelinecache failed with error " + itos(err) + "."); } } void RenderingDeviceVulkan::_load_pipeline_cache() { DirAccess::make_dir_recursive_absolute(pipelines_cache.file_path.get_base_dir()); if (FileAccess::exists(pipelines_cache.file_path)) { Error file_error; Vector<uint8_t> file_data = FileAccess::get_file_as_bytes(pipelines_cache.file_path, &file_error); if (file_error != OK || file_data.size() <= (int)sizeof(PipelineCacheHeader)) { WARN_PRINT("Invalid/corrupt pipelines cache."); return; } const PipelineCacheHeader *header = reinterpret_cast<const PipelineCacheHeader *>(file_data.ptr()); if (header->magic != 868 + VK_PIPELINE_CACHE_HEADER_VERSION_ONE) { WARN_PRINT("Invalid pipelines cache magic number."); return; } const uint8_t *loaded_buffer_start = file_data.ptr() + sizeof(PipelineCacheHeader); uint32_t loaded_buffer_size = file_data.size() - sizeof(PipelineCacheHeader); if (header->data_hash != hash_murmur3_buffer(loaded_buffer_start, loaded_buffer_size) || header->data_size != loaded_buffer_size || header->vendor_id != pipelines_cache.header.vendor_id || header->device_id != pipelines_cache.header.device_id || header->driver_version != pipelines_cache.header.driver_version || memcmp(header->uuid, pipelines_cache.header.uuid, VK_UUID_SIZE) != 0 || header->driver_abi != pipelines_cache.header.driver_abi) { WARN_PRINT("Invalid pipelines cache header."); pipelines_cache.current_size = 0; pipelines_cache.buffer.clear(); } else { pipelines_cache.current_size = loaded_buffer_size; pipelines_cache.buffer.resize(loaded_buffer_size); memcpy(pipelines_cache.buffer.ptr(), loaded_buffer_start, pipelines_cache.buffer.size()); } } } void RenderingDeviceVulkan::_update_pipeline_cache(bool p_closing) { { bool still_saving = pipelines_cache_save_task != WorkerThreadPool::INVALID_TASK_ID && !WorkerThreadPool::get_singleton()->is_task_completed(pipelines_cache_save_task); if (still_saving) { if (p_closing) { WorkerThreadPool::get_singleton()->wait_for_task_completion(pipelines_cache_save_task); pipelines_cache_save_task = WorkerThreadPool::INVALID_TASK_ID; } else { // We can't save until the currently running save is done. We'll retry next time; worst case, we'll save when exiting. return; } } } { // FIXME: // We're letting the cache grow unboundedly. We may want to set at limit and see if implementations use LRU or the like. // If we do, we won't be able to assume any longer that the cache is dirty if, and only if, it has grown. size_t pso_blob_size = 0; VkResult vr = vkGetPipelineCacheData(device, pipelines_cache.cache_object, &pso_blob_size, nullptr); ERR_FAIL_COND(vr); size_t difference = pso_blob_size - pipelines_cache.current_size; bool must_save = false; if (p_closing) { must_save = difference > 0; } else { float save_interval = GLOBAL_GET("rendering/rendering_device/pipeline_cache/save_chunk_size_mb"); must_save = difference > 0 && difference / (1024.0f * 1024.0f) >= save_interval; } if (must_save) { pipelines_cache.current_size = pso_blob_size; } else { return; } } if (p_closing) { _save_pipeline_cache(this); } else { pipelines_cache_save_task = WorkerThreadPool::get_singleton()->add_native_task(&_save_pipeline_cache, this, false, "PipelineCacheSave"); } } void RenderingDeviceVulkan::_save_pipeline_cache(void *p_data) { RenderingDeviceVulkan *self = static_cast<RenderingDeviceVulkan *>(p_data); self->pipelines_cache.buffer.resize(self->pipelines_cache.current_size); self->_thread_safe_.lock(); VkResult vr = vkGetPipelineCacheData(self->device, self->pipelines_cache.cache_object, &self->pipelines_cache.current_size, self->pipelines_cache.buffer.ptr()); self->_thread_safe_.unlock(); ERR_FAIL_COND(vr != VK_SUCCESS && vr != VK_INCOMPLETE); // Incomplete is OK because the cache may have grown since the size was queried (unless when exiting). print_verbose(vformat("Updated PSO cache (%.1f MiB)", self->pipelines_cache.current_size / (1024.0f * 1024.0f))); // The real buffer size may now be bigger than the updated current_size. // We take into account the new size but keep the buffer resized in a worst-case fashion. self->pipelines_cache.header.data_size = self->pipelines_cache.current_size; self->pipelines_cache.header.data_hash = hash_murmur3_buffer(self->pipelines_cache.buffer.ptr(), self->pipelines_cache.current_size); Ref<FileAccess> f = FileAccess::open(self->pipelines_cache.file_path, FileAccess::WRITE, nullptr); if (f.is_valid()) { f->store_buffer((const uint8_t *)&self->pipelines_cache.header, sizeof(PipelineCacheHeader)); f->store_buffer(self->pipelines_cache.buffer.ptr(), self->pipelines_cache.current_size); } } template <class T> void RenderingDeviceVulkan::_free_rids(T &p_owner, const char *p_type) { List<RID> owned; p_owner.get_owned_list(&owned); if (owned.size()) { if (owned.size() == 1) { WARN_PRINT(vformat("1 RID of type \"%s\" was leaked.", p_type)); } else { WARN_PRINT(vformat("%d RIDs of type \"%s\" were leaked.", owned.size(), p_type)); } for (const RID &E : owned) { #ifdef DEV_ENABLED if (resource_names.has(E)) { print_line(String(" - ") + resource_names[E]); } #endif free(E); } } } void RenderingDeviceVulkan::capture_timestamp(const String &p_name) { ERR_FAIL_COND_MSG(draw_list != nullptr, "Capturing timestamps during draw list creation is not allowed. Offending timestamp was: " + p_name); ERR_FAIL_COND(frames[frame].timestamp_count >= max_timestamp_query_elements); // This should be optional for profiling, else it will slow things down. { VkMemoryBarrier memoryBarrier; memoryBarrier.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER; memoryBarrier.pNext = nullptr; memoryBarrier.srcAccessMask = VK_ACCESS_INDIRECT_COMMAND_READ_BIT | VK_ACCESS_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_UNIFORM_READ_BIT | VK_ACCESS_INPUT_ATTACHMENT_READ_BIT | VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_HOST_READ_BIT | VK_ACCESS_HOST_WRITE_BIT; memoryBarrier.dstAccessMask = VK_ACCESS_INDIRECT_COMMAND_READ_BIT | VK_ACCESS_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_UNIFORM_READ_BIT | VK_ACCESS_INPUT_ATTACHMENT_READ_BIT | VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_HOST_READ_BIT | VK_ACCESS_HOST_WRITE_BIT; vkCmdPipelineBarrier(frames[frame].draw_command_buffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, 1, &memoryBarrier, 0, nullptr, 0, nullptr); } vkCmdWriteTimestamp(frames[frame].draw_command_buffer, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, frames[frame].timestamp_pool, frames[frame].timestamp_count); frames[frame].timestamp_names[frames[frame].timestamp_count] = p_name; frames[frame].timestamp_cpu_values[frames[frame].timestamp_count] = OS::get_singleton()->get_ticks_usec(); frames[frame].timestamp_count++; } uint64_t RenderingDeviceVulkan::get_driver_resource(DriverResource p_resource, RID p_rid, uint64_t p_index) { _THREAD_SAFE_METHOD_ switch (p_resource) { case DRIVER_RESOURCE_VULKAN_DEVICE: { return (uint64_t)context->get_device(); } break; case DRIVER_RESOURCE_VULKAN_PHYSICAL_DEVICE: { return (uint64_t)context->get_physical_device(); } break; case DRIVER_RESOURCE_VULKAN_INSTANCE: { return (uint64_t)context->get_instance(); } break; case DRIVER_RESOURCE_VULKAN_QUEUE: { return (uint64_t)context->get_graphics_queue(); } break; case DRIVER_RESOURCE_VULKAN_QUEUE_FAMILY_INDEX: { return context->get_graphics_queue_family_index(); } break; case DRIVER_RESOURCE_VULKAN_IMAGE: { Texture *tex = texture_owner.get_or_null(p_rid); ERR_FAIL_NULL_V(tex, 0); return (uint64_t)tex->image; } break; case DRIVER_RESOURCE_VULKAN_IMAGE_VIEW: { Texture *tex = texture_owner.get_or_null(p_rid); ERR_FAIL_NULL_V(tex, 0); return (uint64_t)tex->view; } break; case DRIVER_RESOURCE_VULKAN_IMAGE_NATIVE_TEXTURE_FORMAT: { Texture *tex = texture_owner.get_or_null(p_rid); ERR_FAIL_NULL_V(tex, 0); return vulkan_formats[tex->format]; } break; case DRIVER_RESOURCE_VULKAN_SAMPLER: { VkSampler *sampler = sampler_owner.get_or_null(p_rid); ERR_FAIL_NULL_V(sampler, 0); return uint64_t(*sampler); } break; case DRIVER_RESOURCE_VULKAN_DESCRIPTOR_SET: { UniformSet *uniform_set = uniform_set_owner.get_or_null(p_rid); ERR_FAIL_NULL_V(uniform_set, 0); return uint64_t(uniform_set->descriptor_set); } break; case DRIVER_RESOURCE_VULKAN_BUFFER: { Buffer *buffer = nullptr; if (vertex_buffer_owner.owns(p_rid)) { buffer = vertex_buffer_owner.get_or_null(p_rid); } else if (index_buffer_owner.owns(p_rid)) { buffer = index_buffer_owner.get_or_null(p_rid); } else if (uniform_buffer_owner.owns(p_rid)) { buffer = uniform_buffer_owner.get_or_null(p_rid); } else if (texture_buffer_owner.owns(p_rid)) { buffer = &texture_buffer_owner.get_or_null(p_rid)->buffer; } else if (storage_buffer_owner.owns(p_rid)) { buffer = storage_buffer_owner.get_or_null(p_rid); } ERR_FAIL_NULL_V(buffer, 0); return uint64_t(buffer->buffer); } break; case DRIVER_RESOURCE_VULKAN_COMPUTE_PIPELINE: { ComputePipeline *compute_pipeline = compute_pipeline_owner.get_or_null(p_rid); ERR_FAIL_NULL_V(compute_pipeline, 0); return uint64_t(compute_pipeline->pipeline); } break; case DRIVER_RESOURCE_VULKAN_RENDER_PIPELINE: { RenderPipeline *render_pipeline = render_pipeline_owner.get_or_null(p_rid); ERR_FAIL_NULL_V(render_pipeline, 0); return uint64_t(render_pipeline->pipeline); } break; default: { // Not supported for this driver. return 0; } break; } } uint32_t RenderingDeviceVulkan::get_captured_timestamps_count() const { return frames[frame].timestamp_result_count; } uint64_t RenderingDeviceVulkan::get_captured_timestamps_frame() const { return frames[frame].index; } static void mult64to128(uint64_t u, uint64_t v, uint64_t &h, uint64_t &l) { uint64_t u1 = (u & 0xffffffff); uint64_t v1 = (v & 0xffffffff); uint64_t t = (u1 * v1); uint64_t w3 = (t & 0xffffffff); uint64_t k = (t >> 32); u >>= 32; t = (u * v1) + k; k = (t & 0xffffffff); uint64_t w1 = (t >> 32); v >>= 32; t = (u1 * v) + k; k = (t >> 32); h = (u * v) + w1 + k; l = (t << 32) + w3; } uint64_t RenderingDeviceVulkan::get_captured_timestamp_gpu_time(uint32_t p_index) const { ERR_FAIL_UNSIGNED_INDEX_V(p_index, frames[frame].timestamp_result_count, 0); // This sucks because timestampPeriod multiplier is a float, while the timestamp is 64 bits nanosecs. // So, in cases like nvidia which give you enormous numbers and 1 as multiplier, multiplying is next to impossible. // Need to do 128 bits fixed point multiplication to get the right value. uint64_t shift_bits = 16; uint64_t h, l; mult64to128(frames[frame].timestamp_result_values[p_index], uint64_t(double(limits.timestampPeriod) * double(1 << shift_bits)), h, l); l >>= shift_bits; l |= h << (64 - shift_bits); return l; } uint64_t RenderingDeviceVulkan::get_captured_timestamp_cpu_time(uint32_t p_index) const { ERR_FAIL_UNSIGNED_INDEX_V(p_index, frames[frame].timestamp_result_count, 0); return frames[frame].timestamp_cpu_result_values[p_index]; } String RenderingDeviceVulkan::get_captured_timestamp_name(uint32_t p_index) const { ERR_FAIL_UNSIGNED_INDEX_V(p_index, frames[frame].timestamp_result_count, String()); return frames[frame].timestamp_result_names[p_index]; } uint64_t RenderingDeviceVulkan::limit_get(Limit p_limit) const { switch (p_limit) { case LIMIT_MAX_BOUND_UNIFORM_SETS: return limits.maxBoundDescriptorSets; case LIMIT_MAX_FRAMEBUFFER_COLOR_ATTACHMENTS: return limits.maxColorAttachments; case LIMIT_MAX_TEXTURES_PER_UNIFORM_SET: return limits.maxDescriptorSetSampledImages; case LIMIT_MAX_SAMPLERS_PER_UNIFORM_SET: return limits.maxDescriptorSetSamplers; case LIMIT_MAX_STORAGE_BUFFERS_PER_UNIFORM_SET: return limits.maxDescriptorSetStorageBuffers; case LIMIT_MAX_STORAGE_IMAGES_PER_UNIFORM_SET: return limits.maxDescriptorSetStorageImages; case LIMIT_MAX_UNIFORM_BUFFERS_PER_UNIFORM_SET: return limits.maxDescriptorSetUniformBuffers; case LIMIT_MAX_DRAW_INDEXED_INDEX: return limits.maxDrawIndexedIndexValue; case LIMIT_MAX_FRAMEBUFFER_HEIGHT: return limits.maxFramebufferHeight; case LIMIT_MAX_FRAMEBUFFER_WIDTH: return limits.maxFramebufferWidth; case LIMIT_MAX_TEXTURE_ARRAY_LAYERS: return limits.maxImageArrayLayers; case LIMIT_MAX_TEXTURE_SIZE_1D: return limits.maxImageDimension1D; case LIMIT_MAX_TEXTURE_SIZE_2D: return limits.maxImageDimension2D; case LIMIT_MAX_TEXTURE_SIZE_3D: return limits.maxImageDimension3D; case LIMIT_MAX_TEXTURE_SIZE_CUBE: return limits.maxImageDimensionCube; case LIMIT_MAX_TEXTURES_PER_SHADER_STAGE: return limits.maxPerStageDescriptorSampledImages; case LIMIT_MAX_SAMPLERS_PER_SHADER_STAGE: return limits.maxPerStageDescriptorSamplers; case LIMIT_MAX_STORAGE_BUFFERS_PER_SHADER_STAGE: return limits.maxPerStageDescriptorStorageBuffers; case LIMIT_MAX_STORAGE_IMAGES_PER_SHADER_STAGE: return limits.maxPerStageDescriptorStorageImages; case LIMIT_MAX_UNIFORM_BUFFERS_PER_SHADER_STAGE: return limits.maxPerStageDescriptorUniformBuffers; case LIMIT_MAX_PUSH_CONSTANT_SIZE: return limits.maxPushConstantsSize; case LIMIT_MAX_UNIFORM_BUFFER_SIZE: return limits.maxUniformBufferRange; case LIMIT_MAX_VERTEX_INPUT_ATTRIBUTE_OFFSET: return limits.maxVertexInputAttributeOffset; case LIMIT_MAX_VERTEX_INPUT_ATTRIBUTES: return limits.maxVertexInputAttributes; case LIMIT_MAX_VERTEX_INPUT_BINDINGS: return limits.maxVertexInputBindings; case LIMIT_MAX_VERTEX_INPUT_BINDING_STRIDE: return limits.maxVertexInputBindingStride; case LIMIT_MIN_UNIFORM_BUFFER_OFFSET_ALIGNMENT: return limits.minUniformBufferOffsetAlignment; case LIMIT_MAX_COMPUTE_WORKGROUP_COUNT_X: return limits.maxComputeWorkGroupCount[0]; case LIMIT_MAX_COMPUTE_WORKGROUP_COUNT_Y: return limits.maxComputeWorkGroupCount[1]; case LIMIT_MAX_COMPUTE_WORKGROUP_COUNT_Z: return limits.maxComputeWorkGroupCount[2]; case LIMIT_MAX_COMPUTE_WORKGROUP_INVOCATIONS: return limits.maxComputeWorkGroupInvocations; case LIMIT_MAX_COMPUTE_WORKGROUP_SIZE_X: return limits.maxComputeWorkGroupSize[0]; case LIMIT_MAX_COMPUTE_WORKGROUP_SIZE_Y: return limits.maxComputeWorkGroupSize[1]; case LIMIT_MAX_COMPUTE_WORKGROUP_SIZE_Z: return limits.maxComputeWorkGroupSize[2]; case LIMIT_MAX_VIEWPORT_DIMENSIONS_X: return limits.maxViewportDimensions[0]; case LIMIT_MAX_VIEWPORT_DIMENSIONS_Y: return limits.maxViewportDimensions[1]; case LIMIT_SUBGROUP_SIZE: { VulkanContext::SubgroupCapabilities subgroup_capabilities = context->get_subgroup_capabilities(); return subgroup_capabilities.size; } case LIMIT_SUBGROUP_IN_SHADERS: { VulkanContext::SubgroupCapabilities subgroup_capabilities = context->get_subgroup_capabilities(); return subgroup_capabilities.supported_stages_flags_rd(); } case LIMIT_SUBGROUP_OPERATIONS: { VulkanContext::SubgroupCapabilities subgroup_capabilities = context->get_subgroup_capabilities(); return subgroup_capabilities.supported_operations_flags_rd(); } case LIMIT_VRS_TEXEL_WIDTH: { return context->get_vrs_capabilities().texel_size.x; } case LIMIT_VRS_TEXEL_HEIGHT: { return context->get_vrs_capabilities().texel_size.y; } default: ERR_FAIL_V(0); } return 0; } void RenderingDeviceVulkan::finalize() { // Free all resources. _flush(false); _free_rids(render_pipeline_owner, "Pipeline"); _free_rids(compute_pipeline_owner, "Compute"); _free_rids(uniform_set_owner, "UniformSet"); _free_rids(texture_buffer_owner, "TextureBuffer"); _free_rids(storage_buffer_owner, "StorageBuffer"); _free_rids(uniform_buffer_owner, "UniformBuffer"); _free_rids(shader_owner, "Shader"); _free_rids(index_array_owner, "IndexArray"); _free_rids(index_buffer_owner, "IndexBuffer"); _free_rids(vertex_array_owner, "VertexArray"); _free_rids(vertex_buffer_owner, "VertexBuffer"); _free_rids(framebuffer_owner, "Framebuffer"); _free_rids(sampler_owner, "Sampler"); { // For textures it's a bit more difficult because they may be shared. List<RID> owned; texture_owner.get_owned_list(&owned); if (owned.size()) { if (owned.size() == 1) { WARN_PRINT("1 RID of type \"Texture\" was leaked."); } else { WARN_PRINT(vformat("%d RIDs of type \"Texture\" were leaked.", owned.size())); } // Free shared first. for (List<RID>::Element *E = owned.front(); E;) { List<RID>::Element *N = E->next(); if (texture_is_shared(E->get())) { #ifdef DEV_ENABLED if (resource_names.has(E->get())) { print_line(String(" - ") + resource_names[E->get()]); } #endif free(E->get()); owned.erase(E); } E = N; } // Free non shared second, this will avoid an error trying to free unexisting textures due to dependencies. for (const RID &E : owned) { #ifdef DEV_ENABLED if (resource_names.has(E)) { print_line(String(" - ") + resource_names[E]); } #endif free(E); } } } // Free everything pending. for (int i = 0; i < frame_count; i++) { int f = (frame + i) % frame_count; _free_pending_resources(f); vkDestroyCommandPool(device, frames[i].command_pool, nullptr); vkDestroyQueryPool(device, frames[i].timestamp_pool, nullptr); } _update_pipeline_cache(true); vkDestroyPipelineCache(device, pipelines_cache.cache_object, nullptr); for (int i = 0; i < split_draw_list_allocators.size(); i++) { vkDestroyCommandPool(device, split_draw_list_allocators[i].command_pool, nullptr); } frames.clear(); for (int i = 0; i < staging_buffer_blocks.size(); i++) { vmaDestroyBuffer(allocator, staging_buffer_blocks[i].buffer, staging_buffer_blocks[i].allocation); } while (small_allocs_pools.size()) { HashMap<uint32_t, VmaPool>::Iterator E = small_allocs_pools.begin(); vmaDestroyPool(allocator, E->value); small_allocs_pools.remove(E); } vmaDestroyAllocator(allocator); while (vertex_formats.size()) { HashMap<VertexFormatID, VertexDescriptionCache>::Iterator temp = vertex_formats.begin(); memdelete_arr(temp->value.bindings); memdelete_arr(temp->value.attributes); vertex_formats.remove(temp); } for (KeyValue<FramebufferFormatID, FramebufferFormat> &E : framebuffer_formats) { vkDestroyRenderPass(device, E.value.render_pass, nullptr); } framebuffer_formats.clear(); // All these should be clear at this point. ERR_FAIL_COND(descriptor_pools.size()); ERR_FAIL_COND(dependency_map.size()); ERR_FAIL_COND(reverse_dependency_map.size()); } RenderingDevice *RenderingDeviceVulkan::create_local_device() { RenderingDeviceVulkan *rd = memnew(RenderingDeviceVulkan); rd->initialize(context, true); return rd; } bool RenderingDeviceVulkan::has_feature(const Features p_feature) const { switch (p_feature) { case SUPPORTS_MULTIVIEW: { VulkanContext::MultiviewCapabilities multiview_capabilies = context->get_multiview_capabilities(); return multiview_capabilies.is_supported && multiview_capabilies.max_view_count > 1; } break; case SUPPORTS_FSR_HALF_FLOAT: { return context->get_shader_capabilities().shader_float16_is_supported && context->get_physical_device_features().shaderInt16 && context->get_storage_buffer_capabilities().storage_buffer_16_bit_access_is_supported; } break; case SUPPORTS_ATTACHMENT_VRS: { VulkanContext::VRSCapabilities vrs_capabilities = context->get_vrs_capabilities(); return vrs_capabilities.attachment_vrs_supported && context->get_physical_device_features().shaderStorageImageExtendedFormats; } break; case SUPPORTS_FRAGMENT_SHADER_WITH_ONLY_SIDE_EFFECTS: { return true; } break; default: { return false; } } } RenderingDeviceVulkan::RenderingDeviceVulkan() { device_capabilities.device_family = DEVICE_VULKAN; } RenderingDeviceVulkan::~RenderingDeviceVulkan() { if (local_device.is_valid()) { finalize(); context->local_device_free(local_device); } }