/*************************************************************************/ /* rendering_device_vulkan.cpp */ /*************************************************************************/ /* This file is part of: */ /* GODOT ENGINE */ /* https://godotengine.org */ /*************************************************************************/ /* Copyright (c) 2007-2021 Juan Linietsky, Ariel Manzur. */ /* Copyright (c) 2014-2021 Godot Engine contributors (cf. AUTHORS.md). */ /* */ /* 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/os/file_access.h" #include "core/os/os.h" #include "core/templates/hashfuncs.h" #include "drivers/vulkan/vulkan_context.h" #include "thirdparty/spirv-reflect/spirv_reflect.h" //#define FORCE_FULL_BARRIER // 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, uint32_t p_post_barrier) { Buffer *buffer = nullptr; if (vertex_buffer_owner.owns(p_buffer)) { buffer = vertex_buffer_owner.getornull(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 & BARRIER_MASK_RASTER) { r_access_mask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; } if (p_post_barrier & 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.getornull(p_buffer); } else if (uniform_buffer_owner.owns(p_buffer)) { if (p_post_barrier & BARRIER_MASK_RASTER) { r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; } if (p_post_barrier & BARRIER_MASK_COMPUTE) { r_stage_mask |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; } r_access_mask |= VK_ACCESS_UNIFORM_READ_BIT; buffer = uniform_buffer_owner.getornull(p_buffer); } else if (texture_buffer_owner.owns(p_buffer)) { if (p_post_barrier & BARRIER_MASK_RASTER) { r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; r_access_mask |= VK_ACCESS_SHADER_READ_BIT; } if (p_post_barrier & BARRIER_MASK_COMPUTE) { r_stage_mask |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; r_access_mask |= VK_ACCESS_SHADER_READ_BIT; } buffer = &texture_buffer_owner.getornull(p_buffer)->buffer; } else if (storage_buffer_owner.owns(p_buffer)) { buffer = storage_buffer_owner.getornull(p_buffer); if (p_post_barrier & BARRIER_MASK_RASTER) { r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; r_access_mask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier & 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(VkSubpassDependency &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] = Set(); } dependency_map[p_depends_on].insert(p_id); if (!reverse_dependency_map.has(p_id)) { reverse_dependency_map[p_id] = Set(); } reverse_dependency_map[p_id].insert(p_depends_on); } void RenderingDeviceVulkan::_free_dependencies(RID p_id) { //direct dependencies must be freed Map>::Element *E = dependency_map.find(p_id); if (E) { while (E->get().size()) { free(E->get().front()->get()); } dependency_map.erase(E); } //reverse dependencies must be unreferenced E = reverse_dependency_map.find(p_id); if (E) { for (Set::Element *F = E->get().front(); F; F = F->next()) { Map>::Element *G = dependency_map.find(F->get()); ERR_CONTINUE(!G); ERR_CONTINUE(!G->get().has(p_id)); G->get().erase(p_id); } reverse_dependency_map.erase(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, VK_FORMAT_PVRTC1_2BPP_UNORM_BLOCK_IMG, VK_FORMAT_PVRTC1_4BPP_UNORM_BLOCK_IMG, VK_FORMAT_PVRTC2_2BPP_UNORM_BLOCK_IMG, VK_FORMAT_PVRTC2_4BPP_UNORM_BLOCK_IMG, VK_FORMAT_PVRTC1_2BPP_SRGB_BLOCK_IMG, VK_FORMAT_PVRTC1_4BPP_SRGB_BLOCK_IMG, VK_FORMAT_PVRTC2_2BPP_SRGB_BLOCK_IMG, VK_FORMAT_PVRTC2_4BPP_SRGB_BLOCK_IMG, }; 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", "Pvrtc1_2Bpp_Unorm_Block_Img", "Pvrtc1_4Bpp_Unorm_Block_Img", "Pvrtc2_2Bpp_Unorm_Block_Img", "Pvrtc2_4Bpp_Unorm_Block_Img", "Pvrtc1_2Bpp_Srgb_Block_Img", "Pvrtc1_4Bpp_Srgb_Block_Img", "Pvrtc2_2Bpp_Srgb_Block_Img", "Pvrtc2_4Bpp_Srgb_Block_Img" }; 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; case DATA_FORMAT_PVRTC1_2BPP_UNORM_BLOCK_IMG: case DATA_FORMAT_PVRTC1_4BPP_UNORM_BLOCK_IMG: case DATA_FORMAT_PVRTC2_2BPP_UNORM_BLOCK_IMG: case DATA_FORMAT_PVRTC2_4BPP_UNORM_BLOCK_IMG: case DATA_FORMAT_PVRTC1_2BPP_SRGB_BLOCK_IMG: case DATA_FORMAT_PVRTC1_4BPP_SRGB_BLOCK_IMG: case DATA_FORMAT_PVRTC2_2BPP_SRGB_BLOCK_IMG: case DATA_FORMAT_PVRTC2_4BPP_SRGB_BLOCK_IMG: return 1; 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: 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: r_w = 4; r_h = 4; return; case DATA_FORMAT_PVRTC1_4BPP_UNORM_BLOCK_IMG: case DATA_FORMAT_PVRTC2_4BPP_UNORM_BLOCK_IMG: case DATA_FORMAT_PVRTC1_4BPP_SRGB_BLOCK_IMG: case DATA_FORMAT_PVRTC2_4BPP_SRGB_BLOCK_IMG: r_w = 4; r_h = 4; return; case DATA_FORMAT_PVRTC1_2BPP_UNORM_BLOCK_IMG: case DATA_FORMAT_PVRTC2_2BPP_UNORM_BLOCK_IMG: case DATA_FORMAT_PVRTC1_2BPP_SRGB_BLOCK_IMG: case DATA_FORMAT_PVRTC2_2BPP_SRGB_BLOCK_IMG: r_w = 8; 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 8; //wrong case DATA_FORMAT_PVRTC1_4BPP_UNORM_BLOCK_IMG: case DATA_FORMAT_PVRTC2_4BPP_UNORM_BLOCK_IMG: case DATA_FORMAT_PVRTC1_4BPP_SRGB_BLOCK_IMG: case DATA_FORMAT_PVRTC2_4BPP_SRGB_BLOCK_IMG: case DATA_FORMAT_PVRTC1_2BPP_UNORM_BLOCK_IMG: case DATA_FORMAT_PVRTC2_2BPP_UNORM_BLOCK_IMG: case DATA_FORMAT_PVRTC1_2BPP_SRGB_BLOCK_IMG: case DATA_FORMAT_PVRTC2_2BPP_SRGB_BLOCK_IMG: return 8; //what varies is resolution 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: case DATA_FORMAT_PVRTC1_4BPP_UNORM_BLOCK_IMG: case DATA_FORMAT_PVRTC2_4BPP_UNORM_BLOCK_IMG: case DATA_FORMAT_PVRTC1_4BPP_SRGB_BLOCK_IMG: case DATA_FORMAT_PVRTC2_4BPP_SRGB_BLOCK_IMG: return 1; case DATA_FORMAT_PVRTC1_2BPP_UNORM_BLOCK_IMG: //these formats are quarter byte size, so rshift is 1 case DATA_FORMAT_PVRTC2_2BPP_UNORM_BLOCK_IMG: case DATA_FORMAT_PVRTC1_2BPP_SRGB_BLOCK_IMG: case DATA_FORMAT_PVRTC2_2BPP_SRGB_BLOCK_IMG: 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(1, 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) int w = p_width; int h = p_height; int d = p_depth; int mipmaps = 1; while (true) { if (w == 1 && h == 1 && d == 1) { break; } w = MAX(1, w >> 1); h = MAX(1, h >> 1); d = MAX(1, 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_mapping) { 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 = 0; allocInfo.usage = p_mapping; allocInfo.requiredFlags = 0; allocInfo.preferredFlags = 0; allocInfo.memoryTypeBits = 0; allocInfo.pool = nullptr; allocInfo.pUserData = nullptr; 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; return OK; } Error RenderingDeviceVulkan::_buffer_free(Buffer *p_buffer) { ERR_FAIL_COND_V(p_buffer->size == 0, ERR_INVALID_PARAMETER); 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 = 0; allocInfo.usage = VMA_MEMORY_USAGE_CPU_ONLY; 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, bool p_on_draw_command_buffer) { //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 (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 if (staging_buffer_blocks[staging_buffer_current].frame_used > frames_drawn - frame_count) { //this block may still be in use, let's not touch it unless we have to, so.. can we create a new one? if (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, p_use_draw_command_buffer); 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 copymem(((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_sccess, 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_sccess; 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_sccess, 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_sccess; 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> &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; #ifndef _MSC_VER #warning TODO check for support via RenderingDevice to enable on mobile when possible #endif #ifndef ANDROID_ENABLED // vkCreateImage fails with format list on Android (VK_ERROR_OUT_OF_HOST_MEMORY) VkImageFormatListCreateInfoKHR format_list_create_info; //keep out of the if, needed for creation Vector allowed_formats; //keep out of the if, needed for creation #endif if (p_format.shareable_formats.size()) { image_create_info.flags |= VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT; #ifndef ANDROID_ENABLED 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"); #endif } 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 = rasterization_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_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)) { printf("vkformat: %x\n", image_create_info.format); 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."); } } //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 VmaAllocationCreateInfo allocInfo; allocInfo.flags = 0; allocInfo.usage = p_format.usage_bits & TEXTURE_USAGE_CPU_READ_BIT ? VMA_MEMORY_USAGE_CPU_ONLY : VMA_MEMORY_USAGE_GPU_ONLY; allocInfo.requiredFlags = 0; allocInfo.preferredFlags = 0; allocInfo.memoryTypeBits = 0; allocInfo.pool = nullptr; allocInfo.pUserData = nullptr; 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) + "."); 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.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); if (p_data.size()) { for (uint32_t i = 0; i < image_create_info.arrayLayers; i++) { texture_update(id, i, p_data[i]); } } return id; } RID RenderingDeviceVulkan::texture_create_shared(const TextureView &p_view, RID p_with_texture) { _THREAD_SAFE_METHOD_ Texture *src_texture = texture_owner.getornull(p_with_texture); ERR_FAIL_COND_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.getornull(src_texture->owner); ERR_FAIL_COND_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; 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_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); _add_dependency(id, p_with_texture); 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, TextureSliceType p_slice_type) { _THREAD_SAFE_METHOD_ Texture *src_texture = texture_owner.getornull(p_with_texture); ERR_FAIL_COND_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.getornull(src_texture->owner); ERR_FAIL_COND_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_UNSIGNED_INDEX_V(p_layer, src_texture->layers, RID()); int slice_layers = 1; 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 = 1; 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_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 = 1; 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); _add_dependency(id, p_with_texture); return id; } Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, const Vector &p_data, uint32_t p_post_barrier) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG(draw_list || compute_list, ERR_INVALID_PARAMETER, "Updating textures in is forbidden during creation of a draw or compute list"); Texture *texture = texture_owner.getornull(p_texture); ERR_FAIL_COND_V(!texture, ERR_INVALID_PARAMETER); if (texture->owner != RID()) { p_texture = texture->owner; texture = texture_owner.getornull(texture->owner); ERR_FAIL_COND_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 render pass that uses it is being created. Ensure render pass is finalized (and that it was created with RENDER_PASS_CONTENTS_FINISH) to unbind this texture."); ERR_FAIL_COND_V_MSG(!(texture->usage_flags & TEXTURE_USAGE_CAN_UPDATE_BIT), ERR_INVALID_PARAMETER, "Texture requires the TEXTURE_USAGE_CAN_UPDATE_BIT in order 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_post_barrier ? frames[frame].draw_command_buffer : frames[frame].setup_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 * z / depth; for (uint32_t x = 0; x < width; x += region_size) { for (uint32_t y = 0; y < height; y += 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, p_post_barrier); 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; for (uint32_t xr = 0; xr < region_wb; xr++) { for (uint32_t yr = 0; yr < region_hb; yr++) { uint32_t src_offset = ((yr + yb) * wb + xr + xb) * block_size; uint32_t dst_offset = (yr * region_wb + xr) * block_size; //copy block for (uint32_t i = 0; i < block_size; i++) { write_ptr[dst_offset + i] = read_ptr[src_offset + i]; } } } } else { //regular image (pixels) //must copy a pixel region for (uint32_t xr = 0; xr < region_w; xr++) { for (uint32_t yr = 0; yr < region_h; yr++) { uint32_t src_offset = ((yr + y) * width + xr + x) * pixel_size; uint32_t dst_offset = (yr * region_w + xr) * pixel_size; //copy block for (uint32_t i = 0; i < pixel_size; i++) { write_ptr[dst_offset + i] = read_ptr[src_offset + i]; } } } } { //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_size; } } } mipmap_offset = image_total; logic_width = MAX(1, logic_width >> 1); logic_height = MAX(1, logic_height >> 1); } //barrier to restore layout { uint32_t barrier_flags = 0; uint32_t access_flags = 0; if (p_post_barrier & BARRIER_MASK_COMPUTE) { barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT; } if (p_post_barrier & BARRIER_MASK_RASTER) { barrier_flags |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT; } if (p_post_barrier & 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_ACCESS_TRANSFER_WRITE_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 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 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; copymem(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; copymem(wptr, rptr, pixel_size * width); } } } mipmap_offset = image_total; } } vmaUnmapMemory(allocator, p_allocation); return image_data; } Vector RenderingDeviceVulkan::texture_get_data(RID p_texture, uint32_t p_layer) { _THREAD_SAFE_METHOD_ Texture *tex = texture_owner.getornull(p_texture); ERR_FAIL_COND_V(!tex, Vector()); ERR_FAIL_COND_V_MSG(tex->bound, Vector(), "Texture can't be retrieved while a render pass that uses it is being created. Ensure render pass is finalized (and that it was created with RENDER_PASS_CONTENTS_FINISH) to unbind this texture."); ERR_FAIL_COND_V_MSG(!(tex->usage_flags & TEXTURE_USAGE_CAN_COPY_FROM_BIT), Vector(), "Texture requires the TEXTURE_USAGE_CAN_COPY_FROM_BIT in order 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()); 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_CPU_ONLY); { //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(1, computed_w >> 1); computed_h = MAX(1, computed_h >> 1); computed_d = MAX(1, 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(), "vmaMapMemory failed with error " + itos(vkerr) + "."); Vector buffer_data; { buffer_data.resize(buffer_size); uint8_t *w = buffer_data.ptrw(); copymem(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.getornull(p_texture); ERR_FAIL_COND_V(!tex, false); return tex->owner.is_valid(); } bool RenderingDeviceVulkan::texture_is_valid(RID p_texture) { return texture_owner.owns(p_texture); } 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, uint32_t p_post_barrier) { _THREAD_SAFE_METHOD_ Texture *src_tex = texture_owner.getornull(p_from_texture); ERR_FAIL_COND_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 render pass that uses it is being created. Ensure render pass is finalized (and that it was created with RENDER_PASS_CONTENTS_FINISH) to unbind this texture."); ERR_FAIL_COND_V_MSG(!(src_tex->usage_flags & TEXTURE_USAGE_CAN_COPY_FROM_BIT), ERR_INVALID_PARAMETER, "Source texture requires the TEXTURE_USAGE_CAN_COPY_FROM_BIT in order 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.getornull(p_to_texture); ERR_FAIL_COND_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 render pass that uses it is being created. Ensure render pass is finalized (and that it was created with RENDER_PASS_CONTENTS_FINISH) to unbind this texture."); ERR_FAIL_COND_V_MSG(!(dst_tex->usage_flags & TEXTURE_USAGE_CAN_COPY_TO_BIT), ERR_INVALID_PARAMETER, "Destination texture requires the TEXTURE_USAGE_CAN_COPY_TO_BIT in order 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 & 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 & BARRIER_MASK_RASTER) { barrier_flags |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier & 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_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 = VK_IMAGE_ASPECT_COLOR_BIT; 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); } } return OK; } Error RenderingDeviceVulkan::texture_resolve_multisample(RID p_from_texture, RID p_to_texture, uint32_t p_post_barrier) { _THREAD_SAFE_METHOD_ Texture *src_tex = texture_owner.getornull(p_from_texture); ERR_FAIL_COND_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 render pass that uses it is being created. Ensure render pass is finalized (and that it was created with RENDER_PASS_CONTENTS_FINISH) to unbind this texture."); ERR_FAIL_COND_V_MSG(!(src_tex->usage_flags & TEXTURE_USAGE_CAN_COPY_FROM_BIT), ERR_INVALID_PARAMETER, "Source texture requires the TEXTURE_USAGE_CAN_COPY_FROM_BIT in order 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.getornull(p_to_texture); ERR_FAIL_COND_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 render pass that uses it is being created. Ensure render pass is finalized (and that it was created with RENDER_PASS_CONTENTS_FINISH) to unbind this texture."); ERR_FAIL_COND_V_MSG(!(dst_tex->usage_flags & TEXTURE_USAGE_CAN_COPY_TO_BIT), ERR_INVALID_PARAMETER, "Destination texture requires the TEXTURE_USAGE_CAN_COPY_TO_BIT in order 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 & 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 & BARRIER_MASK_RASTER) { barrier_flags |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier & 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 = VK_IMAGE_ASPECT_COLOR_BIT; 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, uint32_t p_post_barrier) { _THREAD_SAFE_METHOD_ Texture *src_tex = texture_owner.getornull(p_texture); ERR_FAIL_COND_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 render pass that uses it is being created. Ensure render pass is finalized (and that it was created with RENDER_PASS_CONTENTS_FINISH) to unbind 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 TEXTURE_USAGE_CAN_COPY_TO_BIT in order 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 & 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 & BARRIER_MASK_RASTER) { barrier_flags |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; } if (p_post_barrier & 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, uint32_t 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 & TEXTURE_USAGE_CPU_READ_BIT) { flags = properties.linearTilingFeatures; } else { flags = properties.optimalTilingFeatures; } if (p_usage & TEXTURE_USAGE_SAMPLING_BIT && !(flags & VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT)) { return false; } if (p_usage & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT && !(flags & VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BIT)) { return false; } if (p_usage & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT && !(flags & VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT)) { return false; } if (p_usage & TEXTURE_USAGE_STORAGE_BIT && !(flags & VK_FORMAT_FEATURE_STORAGE_IMAGE_BIT)) { return false; } if (p_usage & TEXTURE_USAGE_STORAGE_ATOMIC_BIT && !(flags & VK_FORMAT_FEATURE_STORAGE_IMAGE_ATOMIC_BIT)) { return false; } return true; } /********************/ /**** ATTACHMENT ****/ /********************/ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector &p_format, InitialAction p_initial_color_action, FinalAction p_final_color_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, int *r_color_attachment_count) { Vector attachments; Vector color_references; Vector depth_stencil_references; Vector resolve_references; // 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 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; VkSubpassDependency dependencies[2] = { { VK_SUBPASS_EXTERNAL, 0, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, 0, default_access_mask, 0 }, { 0, VK_SUBPASS_EXTERNAL, VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, default_access_mask, 0, 0 } }; VkSubpassDependency &dependency_from_external = dependencies[0]; VkSubpassDependency &dependency_to_external = dependencies[1]; for (int i = 0; i < p_format.size(); i++) { ERR_FAIL_INDEX_V(p_format[i].format, DATA_FORMAT_MAX, VK_NULL_HANDLE); ERR_FAIL_INDEX_V(p_format[i].samples, TEXTURE_SAMPLES_MAX, VK_NULL_HANDLE); ERR_FAIL_COND_V_MSG(!(p_format[i].usage_flags & (TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | TEXTURE_USAGE_RESOLVE_ATTACHMENT_BIT)), VK_NULL_HANDLE, "Texture format for index (" + itos(i) + ") requires an attachment (depth, stencil or resolve) bit set."); VkAttachmentDescription description = {}; description.flags = 0; description.format = vulkan_formats[p_format[i].format]; description.samples = rasterization_sample_count[p_format[i].samples]; bool is_depth_stencil = p_format[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT; bool is_sampled = p_format[i].usage_flags & TEXTURE_USAGE_SAMPLING_BIT; bool is_storage = p_format[i].usage_flags & TEXTURE_USAGE_STORAGE_BIT; // 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_stencil ? p_initial_depth_action : p_initial_color_action) { case INITIAL_ACTION_CLEAR_REGION: case INITIAL_ACTION_CLEAR: { description.loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR; description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_CLEAR; 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_format[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_format[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_format[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_format[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_format[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_format[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 } } switch (is_depth_stencil ? p_final_depth_action : p_final_color_action) { case FINAL_ACTION_READ: { if (p_format[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_format[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.initialLayout = 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_format[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_format[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.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 } } break; case FINAL_ACTION_CONTINUE: { if (p_format[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_format[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.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 } } break; default: { ERR_FAIL_V(VK_NULL_HANDLE); //should never reach here } } attachments.push_back(description); VkAttachmentReference reference; reference.attachment = i; if (p_format[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { reference.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; color_references.push_back(reference); } else if (p_format[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { reference.layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; depth_stencil_references.push_back(reference); } else if (p_format[i].usage_flags & TEXTURE_USAGE_RESOLVE_ATTACHMENT_BIT) { reference.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; resolve_references.push_back(reference); // if resolves are done, we need to ensure the copy is safe dependency_to_external.dstStageMask |= VK_PIPELINE_STAGE_TRANSFER_BIT; dependency_to_external.dstAccessMask |= VK_ACCESS_TRANSFER_READ_BIT; } else { ERR_FAIL_V_MSG(VK_NULL_HANDLE, "Texture index " + itos(i) + " is neither color, depth stencil or resolve so it can't be used as attachment."); } // 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; } } ERR_FAIL_COND_V_MSG(depth_stencil_references.size() > 1, VK_NULL_HANDLE, "Formats can only have one depth/stencil attachment, supplied (" + itos(depth_stencil_references.size()) + ")."); ERR_FAIL_COND_V_MSG(resolve_references.size() > 1, VK_NULL_HANDLE, "Formats can only have one resolve attachment, supplied (" + itos(resolve_references.size()) + ")."); VkSubpassDescription subpass; subpass.flags = 0; subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS; subpass.inputAttachmentCount = 0; //unsupported for now subpass.pInputAttachments = nullptr; subpass.colorAttachmentCount = color_references.size(); subpass.pColorAttachments = color_references.ptr(); subpass.pDepthStencilAttachment = depth_stencil_references.ptr(); subpass.pResolveAttachments = resolve_references.ptr(); subpass.preserveAttachmentCount = 0; subpass.pPreserveAttachments = nullptr; VkRenderPassCreateInfo render_pass_create_info; render_pass_create_info.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO; 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 = 1; render_pass_create_info.pSubpasses = &subpass; // 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 = 0; render_pass_create_info.pDependencies = nullptr; VkRenderPass render_pass; VkResult res = vkCreateRenderPass(device, &render_pass_create_info, nullptr, &render_pass); ERR_FAIL_COND_V_MSG(res, VK_NULL_HANDLE, "vkCreateRenderPass failed with error " + itos(res) + "."); if (r_color_attachment_count) { *r_color_attachment_count = color_references.size(); } return render_pass; } RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_create(const Vector &p_format) { _THREAD_SAFE_METHOD_ FramebufferFormatKey key; key.attachments = p_format; const Map::Element *E = framebuffer_format_cache.find(key); if (E) { //exists, return return E->get(); } int color_references; VkRenderPass render_pass = _render_pass_create(p_format, INITIAL_ACTION_CLEAR, FINAL_ACTION_READ, INITIAL_ACTION_CLEAR, FINAL_ACTION_READ, &color_references); //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.color_attachments = color_references; fb_format.render_pass = render_pass; fb_format.samples = p_format[0].samples; framebuffer_formats[id] = fb_format; return id; } RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_create_empty(TextureSamples p_samples) { FramebufferFormatKey key; const Map::Element *E = framebuffer_format_cache.find(key); if (E) { //exists, return return E->get(); } VkSubpassDescription subpass; subpass.flags = 0; subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS; 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; VkRenderPassCreateInfo render_pass_create_info; render_pass_create_info.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO; 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; VkRenderPass render_pass; VkResult res = vkCreateRenderPass(device, &render_pass_create_info, nullptr, &render_pass); ERR_FAIL_COND_V_MSG(res, VK_NULL_HANDLE, "vkCreateRenderPass 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.color_attachments = 0; fb_format.render_pass = render_pass; fb_format.samples = p_samples; framebuffer_formats[id] = fb_format; return id; } RenderingDevice::TextureSamples RenderingDeviceVulkan::framebuffer_format_get_texture_samples(FramebufferFormatID p_format) { Map::Element *E = framebuffer_formats.find(p_format); ERR_FAIL_COND_V(!E, TEXTURE_SAMPLES_1); return E->get().samples; } /***********************/ /**** 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; return framebuffer_owner.make_rid(framebuffer); } RID RenderingDeviceVulkan::framebuffer_create(const Vector &p_texture_attachments, FramebufferFormatID p_format_check) { _THREAD_SAFE_METHOD_ Vector attachments; Size2i size; for (int i = 0; i < p_texture_attachments.size(); i++) { Texture *texture = texture_owner.getornull(p_texture_attachments[i]); ERR_FAIL_COND_V_MSG(!texture, RID(), "Texture index supplied for framebuffer (" + itos(i) + ") is not a valid texture."); if (i == 0) { size.width = texture->width; size.height = texture->height; } 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."); } AttachmentFormat af; af.format = texture->format; af.samples = texture->samples; af.usage_flags = texture->usage_flags; attachments.push_back(af); } FramebufferFormatID format_id = framebuffer_format_create(attachments); 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; RID id = framebuffer_owner.make_rid(framebuffer); for (int i = 0; i < p_texture_attachments.size(); i++) { _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.getornull(p_framebuffer); ERR_FAIL_COND_V(!framebuffer, INVALID_ID); return framebuffer->format_id; } /*****************/ /**** 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; 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) + "."); return sampler_owner.make_rid(sampler); } /**********************/ /**** VERTEX ARRAY ****/ /**********************/ RID RenderingDeviceVulkan::vertex_buffer_create(uint32_t p_size_bytes, const Vector &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()); ERR_FAIL_COND_V_MSG(draw_list != nullptr && p_data.size(), RID(), "Creating buffers with data is forbidden during creation of a draw list"); ERR_FAIL_COND_V_MSG(compute_list != nullptr && p_data.size(), RID(), "Creating buffers with data is forbidden during creation of a draw list"); 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_GPU_ONLY); 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); } return vertex_buffer_owner.make_rid(buffer); } // 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 &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()); Set 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 &p_src_buffers) { _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; 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.getornull(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); vertex_array.offsets.push_back(0); //offset unused, but passing anyway } 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 &p_data, bool p_use_restart_indices) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG(draw_list != nullptr && p_data.size(), RID(), "Creating buffers with data is forbidden during creation of a draw list"); ERR_FAIL_COND_V_MSG(compute_list != nullptr && p_data.size(), RID(), "Creating buffers with data is forbidden during creation of a draw list"); 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_GPU_ONLY); 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); } return index_buffer_owner.make_rid(index_buffer); } 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.getornull(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_stage_names[RenderingDevice::SHADER_STAGE_MAX] = { "Vertex", "Fragment", "TesselationControl", "TesselationEvaluation", "Compute" }; 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.getornull(p_shader); ERR_FAIL_COND_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 != String()) { ret += "\n"; } ret += "Set: " + itos(i) + " Binding: " + itos(ui.binding) + " Type: " + shader_uniform_names[ui.type] + " Length: " + itos(ui.length); } } return ret; } #if 0 bool RenderingDeviceVulkan::_uniform_add_binding(Vector > &bindings, Vector > &uniform_infos, const glslang::TObjectReflection &reflection, RenderingDevice::ShaderStage p_stage, Shader::PushConstant &push_constant, String *r_error) { VkDescriptorSetLayoutBinding layout_binding; UniformInfo info; switch (reflection.getType()->getBasicType()) { case glslang::EbtSampler: { //print_line("DEBUG: IsSampler"); if (reflection.getType()->getSampler().dim == glslang::EsdBuffer) { //texture buffers if (reflection.getType()->getSampler().isCombined()) { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; info.type = UNIFORM_TYPE_SAMPLER_WITH_TEXTURE_BUFFER; //print_line("DEBUG: SAMPLER: texel combined"); } else if (reflection.getType()->getSampler().isTexture()) { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; info.type = UNIFORM_TYPE_TEXTURE_BUFFER; //print_line("DEBUG: SAMPLER: texel alone"); } else if (reflection.getType()->getSampler().isImage()) { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER; info.type = UNIFORM_TYPE_IMAGE_BUFFER; //print_line("DEBUG: SAMPLER: texel buffer"); } else { if (r_error) { *r_error = "On shader stage '" + String(shader_stage_names[p_stage]) + "', uniform '" + reflection.name + "' is of unsupported buffer type."; } return false; } } else if (reflection.getType()->getSampler().isCombined()) { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; info.type = UNIFORM_TYPE_SAMPLER_WITH_TEXTURE; //print_line("DEBUG: SAMPLER: combined"); } else if (reflection.getType()->getSampler().isPureSampler()) { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLER; info.type = UNIFORM_TYPE_SAMPLER; //print_line("DEBUG: SAMPLER: sampler"); } else if (reflection.getType()->getSampler().isTexture()) { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE; info.type = UNIFORM_TYPE_TEXTURE; //print_line("DEBUG: SAMPLER: image"); } else if (reflection.getType()->getSampler().isImage()) { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE; info.type = UNIFORM_TYPE_IMAGE; //print_line("DEBUG: SAMPLER: storage image"); } else { //print_line("DEBUG: sampler unknown"); if (r_error) { *r_error = "On shader stage '" + String(shader_stage_names[p_stage]) + "', uniform '" + reflection.name + "' is of unsupported sampler type."; } return false; } if (reflection.getType()->isArray()) { layout_binding.descriptorCount = reflection.getType()->getArraySizes()->getCumulativeSize(); //print_line("DEBUG: array of size: " + itos(layout_binding.descriptorCount)); } else { layout_binding.descriptorCount = 1; } info.length = layout_binding.descriptorCount; } break; /*case glslang::EbtStruct: { print_line("DEBUG: Struct"); } break;*/ case glslang::EbtBlock: { //print_line("DEBUG: Block"); if (reflection.getType()->getQualifier().storage == glslang::EvqUniform) { if (reflection.getType()->getQualifier().layoutPushConstant) { uint32_t len = reflection.size; if (push_constant.push_constant_size != 0 && push_constant.push_constant_size != len) { *r_error = "On shader stage '" + String(shader_stage_names[p_stage]) + "', uniform '" + reflection.name + "' push constants for different stages should all be the same size."; return false; } push_constant.push_constant_size = len; push_constant.push_constants_vk_stage |= shader_stage_masks[p_stage]; return true; } //print_line("DEBUG: Uniform buffer"); layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; info.type = UNIFORM_TYPE_UNIFORM_BUFFER; } else if (reflection.getType()->getQualifier().storage == glslang::EvqBuffer) { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER; info.type = UNIFORM_TYPE_STORAGE_BUFFER; //print_line("DEBUG: Storage buffer"); } else { if (r_error) { *r_error = "On shader stage '" + String(shader_stage_names[p_stage]) + "', uniform '" + reflection.name + "' is of unsupported block type: (" + itos(reflection.getType()->getQualifier().storage) + ")."; } return false; } if (reflection.getType()->isArray()) { layout_binding.descriptorCount = reflection.getType()->getArraySizes()->getCumulativeSize(); //print_line("DEBUG: array of size: " + itos(layout_binding.descriptorCount)); } else { layout_binding.descriptorCount = 1; } info.length = reflection.size; } break; /*case glslang::EbtReference: { } break;*/ /*case glslang::EbtAtomicUint: { } break;*/ default: { if (reflection.getType()->getQualifier().hasOffset() || reflection.name.find(".") != std::string::npos) { //member of uniform block? return true; } if (r_error) { *r_error = "On shader stage '" + String(shader_stage_names[p_stage]) + "', uniform '" + reflection.name + "' unsupported uniform type."; } return false; } } if (!reflection.getType()->getQualifier().hasBinding()) { if (r_error) { *r_error = "On shader stage '" + String(shader_stage_names[p_stage]) + "', uniform '" + reflection.name + "' lacks a binding number."; } return false; } uint32_t set = reflection.getType()->getQualifier().hasSet() ? reflection.getType()->getQualifier().layoutSet : 0; if (set >= MAX_UNIFORM_SETS) { if (r_error) { *r_error = "On shader stage '" + String(shader_stage_names[p_stage]) + "', uniform '" + reflection.name + "' uses a set (" + itos(set) + ") index larger than what is supported (" + itos(MAX_UNIFORM_SETS) + ")."; } return false; } if (set >= limits.maxBoundDescriptorSets) { if (r_error) { *r_error = "On shader stage '" + String(shader_stage_names[p_stage]) + "', uniform '" + reflection.name + "' uses a set (" + itos(set) + ") index larger than what is supported by the hardware (" + itos(limits.maxBoundDescriptorSets) + ")."; } return false; } uint32_t binding = reflection.getType()->getQualifier().layoutBinding; if (set < (uint32_t)bindings.size()) { //check if this already exists for (int i = 0; i < bindings[set].size(); i++) { if (bindings[set][i].binding == binding) { //already exists, verify that it's the same type if (bindings[set][i].descriptorType != layout_binding.descriptorType) { if (r_error) { *r_error = "On shader stage '" + String(shader_stage_names[p_stage]) + "', uniform '" + reflection.name + "' trying to re-use location for set=" + itos(set) + ", binding=" + itos(binding) + " with different uniform type."; } return false; } //also, verify that it's the same size if (bindings[set][i].descriptorCount != layout_binding.descriptorCount || uniform_infos[set][i].length != info.length) { if (r_error) { *r_error = "On shader stage '" + String(shader_stage_names[p_stage]) + "', uniform '" + reflection.name + "' trying to re-use location for set=" + itos(set) + ", binding=" + itos(binding) + " with different uniform size."; } return false; } //just append stage mask and return bindings.write[set].write[i].stageFlags |= shader_stage_masks[p_stage]; uniform_infos.write[set].write[i].stages |= 1 << p_stage; return true; } } } layout_binding.binding = binding; layout_binding.stageFlags = shader_stage_masks[p_stage]; layout_binding.pImmutableSamplers = nullptr; //no support for this yet info.stages = 1 << p_stage; info.binding = binding; if (set >= (uint32_t)bindings.size()) { bindings.resize(set + 1); uniform_infos.resize(set + 1); } #if 0 print_line("stage: " + String(shader_stage_names[p_stage]) + " set: " + itos(set) + " binding: " + itos(info.binding) + " type:" + shader_uniform_names[info.type] + " length: " + itos(info.length)); #endif bindings.write[set].push_back(layout_binding); uniform_infos.write[set].push_back(info); return true; } #endif RID RenderingDeviceVulkan::shader_create(const Vector &p_stages) { //descriptor layouts Vector> set_bindings; Vector> uniform_info; Shader::PushConstant push_constant; push_constant.push_constant_size = 0; push_constant.push_constants_vk_stage = 0; uint32_t vertex_input_mask = 0; uint32_t fragment_outputs = 0; uint32_t stages_processed = 0; bool is_compute = false; uint32_t compute_local_size[3] = { 0, 0, 0 }; for (int i = 0; i < p_stages.size(); i++) { if (p_stages[i].shader_stage == SHADER_STAGE_COMPUTE) { is_compute = true; ERR_FAIL_COND_V_MSG(p_stages.size() != 1, RID(), "Compute shaders can only receive one stage, dedicated to compute."); } ERR_FAIL_COND_V_MSG(stages_processed & (1 << p_stages[i].shader_stage), RID(), "Stage " + String(shader_stage_names[p_stages[i].shader_stage]) + " submitted more than once."); { SpvReflectShaderModule module; const uint8_t *spirv = p_stages[i].spir_v.ptr(); SpvReflectResult result = spvReflectCreateShaderModule(p_stages[i].spir_v.size(), spirv, &module); ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed parsing shader."); if (is_compute) { compute_local_size[0] = module.entry_points->local_size.x; compute_local_size[1] = module.entry_points->local_size.y; compute_local_size[2] = module.entry_points->local_size.z; } uint32_t binding_count = 0; result = spvReflectEnumerateDescriptorBindings(&module, &binding_count, nullptr); ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed enumerating descriptor bindings."); uint32_t stage = p_stages[i].shader_stage; if (binding_count > 0) { //Parse bindings Vector bindings; bindings.resize(binding_count); result = spvReflectEnumerateDescriptorBindings(&module, &binding_count, bindings.ptrw()); ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed getting descriptor bindings."); for (uint32_t j = 0; j < binding_count; j++) { const SpvReflectDescriptorBinding &binding = *bindings[j]; VkDescriptorSetLayoutBinding layout_binding; UniformInfo info; bool need_array_dimensions = false; bool need_block_size = false; switch (binding.descriptor_type) { case SPV_REFLECT_DESCRIPTOR_TYPE_SAMPLER: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLER; info.type = UNIFORM_TYPE_SAMPLER; need_array_dimensions = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; info.type = UNIFORM_TYPE_SAMPLER_WITH_TEXTURE; need_array_dimensions = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_SAMPLED_IMAGE: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE; info.type = UNIFORM_TYPE_TEXTURE; need_array_dimensions = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_STORAGE_IMAGE: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE; info.type = UNIFORM_TYPE_IMAGE; need_array_dimensions = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; info.type = UNIFORM_TYPE_TEXTURE_BUFFER; need_array_dimensions = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER; info.type = UNIFORM_TYPE_IMAGE_BUFFER; need_array_dimensions = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_UNIFORM_BUFFER: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; info.type = UNIFORM_TYPE_UNIFORM_BUFFER; need_block_size = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_STORAGE_BUFFER: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER; info.type = UNIFORM_TYPE_STORAGE_BUFFER; need_block_size = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC: { ERR_PRINT("Dynamic uniform buffer not supported."); continue; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: { ERR_PRINT("Dynamic storage buffer not supported."); continue; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_INPUT_ATTACHMENT: { layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT; info.type = UNIFORM_TYPE_INPUT_ATTACHMENT; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_ACCELERATION_STRUCTURE_KHR: { ERR_PRINT("Acceleration structure not supported."); continue; } break; } if (need_array_dimensions) { if (binding.array.dims_count == 0) { info.length = 1; } else { for (uint32_t k = 0; k < binding.array.dims_count; k++) { if (k == 0) { info.length = binding.array.dims[0]; } else { info.length *= binding.array.dims[k]; } } } layout_binding.descriptorCount = info.length; } else if (need_block_size) { info.length = binding.block.size; layout_binding.descriptorCount = 1; } else { info.length = 0; layout_binding.descriptorCount = 1; } info.binding = binding.binding; uint32_t set = binding.set; //print_line("Stage: " + String(shader_stage_names[stage]) + " set=" + itos(set) + " binding=" + itos(info.binding) + " type=" + shader_uniform_names[info.type] + " length=" + itos(info.length)); ERR_FAIL_COND_V_MSG(set >= MAX_UNIFORM_SETS, RID(), "On shader stage '" + String(shader_stage_names[stage]) + "', uniform '" + binding.name + "' uses a set (" + itos(set) + ") index larger than what is supported (" + itos(MAX_UNIFORM_SETS) + ")."); ERR_FAIL_COND_V_MSG(set >= limits.maxBoundDescriptorSets, RID(), "On shader stage '" + String(shader_stage_names[stage]) + "', uniform '" + binding.name + "' uses a set (" + itos(set) + ") index larger than what is supported by the hardware (" + itos(limits.maxBoundDescriptorSets) + ")."); if (set < (uint32_t)set_bindings.size()) { //check if this already exists bool exists = false; for (int k = 0; k < set_bindings[set].size(); k++) { if (set_bindings[set][k].binding == (uint32_t)info.binding) { //already exists, verify that it's the same type ERR_FAIL_COND_V_MSG(set_bindings[set][k].descriptorType != layout_binding.descriptorType, RID(), "On shader stage '" + String(shader_stage_names[stage]) + "', uniform '" + binding.name + "' trying to re-use location for set=" + itos(set) + ", binding=" + itos(info.binding) + " with different uniform type."); //also, verify that it's the same size ERR_FAIL_COND_V_MSG(set_bindings[set][k].descriptorCount != layout_binding.descriptorCount || uniform_info[set][k].length != info.length, RID(), "On shader stage '" + String(shader_stage_names[stage]) + "', uniform '" + binding.name + "' trying to re-use location for set=" + itos(set) + ", binding=" + itos(info.binding) + " with different uniform size."); //just append stage mask and return set_bindings.write[set].write[k].stageFlags |= shader_stage_masks[stage]; uniform_info.write[set].write[k].stages |= 1 << stage; exists = true; } } if (exists) { continue; //merged } } layout_binding.binding = info.binding; layout_binding.stageFlags = shader_stage_masks[stage]; layout_binding.pImmutableSamplers = nullptr; //no support for this yet info.stages = 1 << stage; info.binding = info.binding; if (set >= (uint32_t)set_bindings.size()) { set_bindings.resize(set + 1); uniform_info.resize(set + 1); } set_bindings.write[set].push_back(layout_binding); uniform_info.write[set].push_back(info); } } if (stage == SHADER_STAGE_VERTEX) { uint32_t iv_count = 0; result = spvReflectEnumerateInputVariables(&module, &iv_count, nullptr); ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed enumerating input variables."); if (iv_count) { Vector input_vars; input_vars.resize(iv_count); result = spvReflectEnumerateInputVariables(&module, &iv_count, input_vars.ptrw()); ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed obtaining input variables."); for (uint32_t j = 0; j < iv_count; j++) { if (input_vars[j] && input_vars[j]->decoration_flags == 0) { //regular input vertex_input_mask |= (1 << uint32_t(input_vars[j]->location)); } } } } if (stage == SHADER_STAGE_FRAGMENT) { uint32_t ov_count = 0; result = spvReflectEnumerateOutputVariables(&module, &ov_count, nullptr); ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed enumerating output variables."); if (ov_count) { Vector output_vars; output_vars.resize(ov_count); result = spvReflectEnumerateOutputVariables(&module, &ov_count, output_vars.ptrw()); ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed obtaining output variables."); for (uint32_t j = 0; j < ov_count; j++) { if (output_vars[j]) { fragment_outputs = MAX(fragment_outputs, output_vars[j]->location + 1); } } } } uint32_t pc_count = 0; result = spvReflectEnumeratePushConstantBlocks(&module, &pc_count, nullptr); ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed enumerating push constants."); if (pc_count) { ERR_FAIL_COND_V_MSG(pc_count > 1, RID(), "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "': Only one push constant is supported, which should be the same across shader stages."); Vector pconstants; pconstants.resize(pc_count); result = spvReflectEnumeratePushConstantBlocks(&module, &pc_count, pconstants.ptrw()); ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed obtaining push constants."); #if 0 if (pconstants[0] == nullptr) { FileAccess *f = FileAccess::open("res://popo.spv", FileAccess::WRITE); f->store_buffer((const uint8_t *)&SpirV[0], SpirV.size() * sizeof(uint32_t)); memdelete(f); } #endif ERR_FAIL_COND_V_MSG(push_constant.push_constant_size && push_constant.push_constant_size != pconstants[0]->size, RID(), "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "': Push constant block must be the same across shader stages."); push_constant.push_constant_size = pconstants[0]->size; push_constant.push_constants_vk_stage |= shader_stage_masks[stage]; //print_line("Stage: " + String(shader_stage_names[stage]) + " push constant of size=" + itos(push_constant.push_constant_size)); } // Destroy the reflection data when no longer required. spvReflectDestroyShaderModule(&module); } stages_processed |= (1 << p_stages[i].shader_stage); } //all good, let's create modules _THREAD_SAFE_METHOD_ Shader shader; shader.vertex_input_mask = vertex_input_mask; shader.fragment_outputs = fragment_outputs; 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]; String error_text; bool success = true; for (int i = 0; i < p_stages.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 = p_stages[i].spir_v.size(); const uint8_t *r = p_stages[i].spir_v.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[p_stages[i].shader_stage]); break; } const VkShaderStageFlagBits shader_stage_bits[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, }; 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_bits[p_stages[i].shader_stage]; 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; Map::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 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.push_constant_size) { push_constant_range.stageFlags = push_constant.push_constants_vk_stage; push_constant_range.offset = 0; push_constant_range.size = push_constant.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); } ERR_FAIL_V_MSG(RID(), error_text); } 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.getornull(p_shader); ERR_FAIL_COND_V(!shader, 0); return shader->vertex_input_mask; } /******************/ /**** UNIFORMS ****/ /******************/ RID RenderingDeviceVulkan::uniform_buffer_create(uint32_t p_size_bytes, const Vector &p_data) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V(p_data.size() && (uint32_t)p_data.size() != p_size_bytes, RID()); ERR_FAIL_COND_V_MSG(draw_list != nullptr && p_data.size(), RID(), "Creating buffers with data is forbidden during creation of a draw list"); ERR_FAIL_COND_V_MSG(compute_list != nullptr && p_data.size(), RID(), "Creating buffers with data is forbidden during creation of a draw list"); 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_GPU_ONLY); 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); } return uniform_buffer_owner.make_rid(buffer); } RID RenderingDeviceVulkan::storage_buffer_create(uint32_t p_size_bytes, const Vector &p_data, uint32_t p_usage) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG(draw_list != nullptr && p_data.size(), RID(), "Creating buffers with data is forbidden during creation of a draw list"); ERR_FAIL_COND_V_MSG(compute_list != nullptr && p_data.size(), RID(), "Creating buffers with data is forbidden during creation of a draw list"); ERR_FAIL_COND_V(p_data.size() && (uint32_t)p_data.size() != p_size_bytes, RID()); Buffer buffer; buffer.usage = p_usage; uint32_t flags = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT; if (p_usage & STORAGE_BUFFER_USAGE_DISPATCH_INDIRECT) { flags |= VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT; } Error err = _buffer_allocate(&buffer, p_size_bytes, flags, VMA_MEMORY_USAGE_GPU_ONLY); 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 &p_data) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG(draw_list != nullptr && p_data.size(), RID(), "Creating buffers with data is forbidden during creation of a draw list"); ERR_FAIL_COND_V_MSG(compute_list != nullptr && p_data.size(), RID(), "Creating buffers with data is forbidden during creation of a draw list"); 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_GPU_ONLY); 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 return texture_buffer_owner.make_rid(texture_buffer); } RenderingDeviceVulkan::DescriptorPool *RenderingDeviceVulkan::_descriptor_pool_allocate(const DescriptorPoolKey &p_key) { if (!descriptor_pools.has(p_key)) { descriptor_pools[p_key] = Set(); } DescriptorPool *pool = nullptr; for (Set::Element *E = descriptor_pools[p_key].front(); E; E = E->next()) { if (E->get()->usage < max_descriptors_per_pool) { pool = E->get(); 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 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 &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.getornull(p_shader); ERR_FAIL_COND_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 writes; DescriptorPoolKey pool_key; //to keep them alive until update call List> buffer_infos; List> buffer_views; List> image_infos; //used for verification to make sure a uniform set does not use a framebuffer bound texture LocalVector attachable_textures; Vector mutable_sampled_textures; Vector 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.ids.size() != 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.ids.size()) + ")."); } else { ERR_FAIL_V_MSG(RID(), "Sampler (binding: " + itos(uniform.binding) + ") should provide one ID referencing a sampler (IDs provided: " + itos(uniform.ids.size()) + ")."); } } Vector image_info; for (int j = 0; j < uniform.ids.size(); j++) { VkSampler *sampler = sampler_owner.getornull(uniform.ids[j]); ERR_FAIL_COND_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.ids.size(); 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.ids.size(); } break; case UNIFORM_TYPE_SAMPLER_WITH_TEXTURE: { if (uniform.ids.size() != 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.ids.size()) + ")."); } 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.ids.size()) + ")."); } } Vector image_info; for (int j = 0; j < uniform.ids.size(); j += 2) { VkSampler *sampler = sampler_owner.getornull(uniform.ids[j + 0]); ERR_FAIL_COND_V_MSG(!sampler, RID(), "SamplerBuffer (binding: " + itos(uniform.binding) + ", index " + itos(j + 1) + ") is not a valid sampler."); Texture *texture = texture_owner.getornull(uniform.ids[j + 1]); ERR_FAIL_COND_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_RESOLVE_ATTACHMENT_BIT)) { UniformSet::AttachableTexture attachable_texture; attachable_texture.bind = set_uniform.binding; attachable_texture.texture = texture->owner.is_valid() ? texture->owner : uniform.ids[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); } if (texture->owner.is_valid()) { texture = texture_owner.getornull(texture->owner); ERR_FAIL_COND_V(!texture, RID()); //bug, should never happen } img_info.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; image_info.push_back(img_info); } write.dstArrayElement = 0; write.descriptorCount = uniform.ids.size() / 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.ids.size() / 2; } break; case UNIFORM_TYPE_TEXTURE: { if (uniform.ids.size() != 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.ids.size()) + ")."); } else { ERR_FAIL_V_MSG(RID(), "Texture (binding: " + itos(uniform.binding) + ") should provide one ID referencing a texture (IDs provided: " + itos(uniform.ids.size()) + ")."); } } Vector image_info; for (int j = 0; j < uniform.ids.size(); j++) { Texture *texture = texture_owner.getornull(uniform.ids[j]); ERR_FAIL_COND_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_RESOLVE_ATTACHMENT_BIT)) { UniformSet::AttachableTexture attachable_texture; attachable_texture.bind = set_uniform.binding; attachable_texture.texture = texture->owner.is_valid() ? texture->owner : uniform.ids[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); } if (texture->owner.is_valid()) { texture = texture_owner.getornull(texture->owner); ERR_FAIL_COND_V(!texture, RID()); //bug, should never happen } img_info.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; image_info.push_back(img_info); } write.dstArrayElement = 0; write.descriptorCount = uniform.ids.size(); 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.ids.size(); } break; case UNIFORM_TYPE_IMAGE: { if (uniform.ids.size() != 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.ids.size()) + ")."); } else { ERR_FAIL_V_MSG(RID(), "Image (binding: " + itos(uniform.binding) + ") should provide one ID referencing a texture (IDs provided: " + itos(uniform.ids.size()) + ")."); } } Vector image_info; for (int j = 0; j < uniform.ids.size(); j++) { Texture *texture = texture_owner.getornull(uniform.ids[j]); ERR_FAIL_COND_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); } if (texture->owner.is_valid()) { texture = texture_owner.getornull(texture->owner); ERR_FAIL_COND_V(!texture, RID()); //bug, should never happen } img_info.imageLayout = VK_IMAGE_LAYOUT_GENERAL; image_info.push_back(img_info); } write.dstArrayElement = 0; write.descriptorCount = uniform.ids.size(); 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.ids.size(); } break; case UNIFORM_TYPE_TEXTURE_BUFFER: { if (uniform.ids.size() != 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.ids.size()) + ")."); } else { ERR_FAIL_V_MSG(RID(), "Buffer (binding: " + itos(uniform.binding) + ") should provide one ID referencing a texture buffer (IDs provided: " + itos(uniform.ids.size()) + ")."); } } Vector buffer_info; Vector buffer_view; for (int j = 0; j < uniform.ids.size(); j++) { TextureBuffer *buffer = texture_buffer_owner.getornull(uniform.ids[j]); ERR_FAIL_COND_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.ids.size(); 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.ids.size(); } break; case UNIFORM_TYPE_SAMPLER_WITH_TEXTURE_BUFFER: { if (uniform.ids.size() != 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.ids.size()) + ")."); } 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.ids.size()) + ")."); } } Vector image_info; Vector buffer_info; Vector buffer_view; for (int j = 0; j < uniform.ids.size(); j += 2) { VkSampler *sampler = sampler_owner.getornull(uniform.ids[j + 0]); ERR_FAIL_COND_V_MSG(!sampler, RID(), "SamplerBuffer (binding: " + itos(uniform.binding) + ", index " + itos(j + 1) + ") is not a valid sampler."); TextureBuffer *buffer = texture_buffer_owner.getornull(uniform.ids[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_COND_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.ids.size() / 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.ids.size() / 2; } break; case UNIFORM_TYPE_IMAGE_BUFFER: { //todo } break; case UNIFORM_TYPE_UNIFORM_BUFFER: { ERR_FAIL_COND_V_MSG(uniform.ids.size() != 1, RID(), "Uniform buffer supplied (binding: " + itos(uniform.binding) + ") must provide one ID (" + itos(uniform.ids.size()) + " provided)."); Buffer *buffer = uniform_buffer_owner.getornull(uniform.ids[0]); ERR_FAIL_COND_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.ids.size() != 1, RID(), "Storage buffer supplied (binding: " + itos(uniform.binding) + ") must provide one ID (" + itos(uniform.ids.size()) + " provided)."); Buffer *buffer = nullptr; if (storage_buffer_owner.owns(uniform.ids[0])) { buffer = storage_buffer_owner.getornull(uniform.ids[0]); } else if (vertex_buffer_owner.owns(uniform.ids[0])) { buffer = vertex_buffer_owner.getornull(uniform.ids[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_COND_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: { } 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_COND_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); //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.ids.size(); const RID *ids = uniform.ids.ptr(); for (int j = 0; j < id_count; j++) { _add_dependency(id, ids[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); } Error RenderingDeviceVulkan::buffer_update(RID p_buffer, uint32_t p_offset, uint32_t p_size, const void *p_data, uint32_t 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 & 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, uint32_t 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 & 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."); // should not be needed // _buffer_memory_barrier(buffer->buffer, p_offset, p_size, dst_stage_mask, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_access, VK_ACCESS_TRANSFER_WRITE_BIT, p_post_barrier); 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, dst_stage_mask); #endif return OK; } Vector RenderingDeviceVulkan::buffer_get_data(RID p_buffer) { _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); if (!buffer) { ERR_FAIL_V_MSG(Vector(), "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 "false" gets us to the same command buffer as below. _buffer_memory_barrier(buffer->buffer, 0, buffer->size, src_stage_mask, src_access_mask, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_ACCESS_TRANSFER_READ_BIT, false); VkCommandBuffer command_buffer = frames[frame].setup_command_buffer; Buffer tmp_buffer; _buffer_allocate(&tmp_buffer, buffer->size, VK_BUFFER_USAGE_TRANSFER_DST_BIT, VMA_MEMORY_USAGE_CPU_ONLY); VkBufferCopy region; region.srcOffset = 0; region.dstOffset = 0; region.size = buffer->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(), "vmaMapMemory failed with error " + itos(vkerr) + "."); Vector buffer_data; { buffer_data.resize(buffer->size); uint8_t *w = buffer_data.ptrw(); copymem(w, buffer_mem, buffer->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, int p_dynamic_state_flags) { _THREAD_SAFE_METHOD_ //needs a shader Shader *shader = shader_owner.getornull(p_shader); ERR_FAIL_COND_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()); } 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(shader->fragment_outputs != fb_format.color_attachments, RID(), "Mismatch fragment output bindings (" + itos(shader->fragment_outputs) + ") and framebuffer color buffers (" + itos(fb_format.color_attachments) + ") 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 & (1 << 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(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 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_enable; 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 = rasterization_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 sample_mask; if (p_multisample_state.sample_mask.size()) { //use sample mask 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]; ERR_FAIL_COND_V(fb_format.color_attachments != p_blend_state.attachments.size(), RID()); Vector attachment_states; for (int i = 0; i < p_blend_state.attachments.size(); i++) { VkPipelineColorBlendAttachmentState state; 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.push_back(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 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 & DYNAMIC_STATE_LINE_WIDTH) { dynamic_states.push_back(VK_DYNAMIC_STATE_LINE_WIDTH); } if (p_dynamic_state_flags & DYNAMIC_STATE_DEPTH_BIAS) { dynamic_states.push_back(VK_DYNAMIC_STATE_DEPTH_BIAS); } if (p_dynamic_state_flags & DYNAMIC_STATE_BLEND_CONSTANTS) { dynamic_states.push_back(VK_DYNAMIC_STATE_BLEND_CONSTANTS); } if (p_dynamic_state_flags & DYNAMIC_STATE_DEPTH_BOUNDS) { dynamic_states.push_back(VK_DYNAMIC_STATE_DEPTH_BOUNDS); } if (p_dynamic_state_flags & DYNAMIC_STATE_STENCIL_COMPARE_MASK) { dynamic_states.push_back(VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK); } if (p_dynamic_state_flags & DYNAMIC_STATE_STENCIL_WRITE_MASK) { dynamic_states.push_back(VK_DYNAMIC_STATE_STENCIL_WRITE_MASK); } if (p_dynamic_state_flags & 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(); //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 = nullptr; graphics_pipeline_create_info.flags = 0; graphics_pipeline_create_info.stageCount = shader->pipeline_stages.size(); graphics_pipeline_create_info.pStages = shader->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 = 0; graphics_pipeline_create_info.basePipelineHandle = VK_NULL_HANDLE; graphics_pipeline_create_info.basePipelineIndex = 0; RenderPipeline pipeline; VkResult err = vkCreateGraphicsPipelines(device, VK_NULL_HANDLE, 1, &graphics_pipeline_create_info, nullptr, &pipeline.pipeline); ERR_FAIL_COND_V_MSG(err, RID(), "vkCreateGraphicsPipelines failed with error " + itos(err) + "."); pipeline.set_formats = shader->set_formats; pipeline.push_constant_stages = shader->push_constant.push_constants_vk_stage; pipeline.pipeline_layout = shader->pipeline_layout; pipeline.shader = p_shader; pipeline.push_constant_size = shader->push_constant.push_constant_size; #ifdef DEBUG_ENABLED pipeline.validation.dynamic_state = p_dynamic_state_flags; pipeline.validation.framebuffer_format = p_framebuffer_format; 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); //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) { _THREAD_SAFE_METHOD_ //needs a shader Shader *shader = shader_owner.getornull(p_shader); ERR_FAIL_COND_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; ComputePipeline pipeline; VkResult err = vkCreateComputePipelines(device, VK_NULL_HANDLE, 1, &compute_pipeline_create_info, nullptr, &pipeline.pipeline); ERR_FAIL_COND_V_MSG(err, RID(), "vkCreateComputePipelines failed with error " + itos(err) + "."); pipeline.set_formats = shader->set_formats; pipeline.push_constant_stages = shader->push_constant.push_constants_vk_stage; pipeline.pipeline_layout = shader->pipeline_layout; pipeline.shader = p_shader; pipeline.push_constant_size = shader->push_constant.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); //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 screen_attachment; screen_attachment.push_back(attachment); return const_cast(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; draw_list = memnew(DrawList); draw_list->command_buffer = command_buffer; #ifdef DEBUG_ENABLED draw_list->validation.framebuffer_format = screen_get_framebuffer_format(); #endif draw_list_count = 0; draw_list_split = false; 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 = context->window_get_width(p_screen); render_pass_begin.renderArea.extent.height = context->window_get_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(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) { 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; 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, p_initial_color_action, p_final_color_action, p_initial_depth_action, p_final_depth_action); 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 attachments; for (int i = 0; i < p_framebuffer->texture_ids.size(); i++) { Texture *texture = texture_owner.getornull(p_framebuffer->texture_ids[i]); ERR_FAIL_COND_V(!texture, ERR_BUG); attachments.push_back(texture->view); 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) + "."); p_framebuffer->framebuffers.insert(vk, version); } const Framebuffer::Version &version = p_framebuffer->framebuffers[vk]; *r_framebuffer = version.framebuffer; *r_render_pass = version.render_pass; 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 &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 &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 clear_values; clear_values.resize(framebuffer->texture_ids.size()); { int color_index = 0; for (int i = 0; i < framebuffer->texture_ids.size(); i++) { Texture *texture = texture_owner.getornull(framebuffer->texture_ids[i]); VkClearValue clear_value; 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[i] = clear_value; } } render_pass_begin.clearValueCount = clear_values.size(); render_pass_begin.pClearValues = clear_values.ptr(); for (int i = 0; i < p_storage_textures.size(); i++) { Texture *texture = texture_owner.getornull(p_storage_textures[i]); 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.getornull(framebuffer->texture_ids[i]); texture->bound = true; draw_list_bound_textures.push_back(framebuffer->texture_ids[i]); } return OK; } void RenderingDeviceVulkan::_draw_list_insert_clear_region(DrawList *draw_list, Framebuffer *framebuffer, Point2i viewport_offset, Point2i viewport_size, bool p_clear_color, const Vector &p_clear_colors, bool p_clear_depth, float p_depth, uint32_t p_stencil) { Vector clear_attachments; int color_index = 0; for (int i = 0; i < framebuffer->texture_ids.size(); i++) { Texture *texture = texture_owner.getornull(framebuffer->texture_ids[i]); VkClearAttachment clear_at = {}; if (p_clear_color && texture->usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { ERR_FAIL_INDEX(color_index, p_clear_colors.size()); //a bug Color clear_color = p_clear_colors[color_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 = viewport_offset.x; cr.rect.offset.y = viewport_offset.y; cr.rect.extent.width = viewport_size.width; cr.rect.extent.height = viewport_size.height; vkCmdClearAttachments(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 &p_clear_color_values, float p_clear_depth, uint32_t p_clear_stencil, const Rect2 &p_region, const Vector &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.getornull(p_framebuffer); ERR_FAIL_COND_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) { //check clear values int color_attachments = framebuffer_formats[framebuffer->format_id].color_attachments; ERR_FAIL_COND_V_MSG(p_clear_color_values.size() != color_attachments, INVALID_ID, "Clear color values supplied (" + itos(p_clear_color_values.size()) + ") differ from the amount required for framebuffer (" + itos(color_attachments) + ")."); } 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); 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 = memnew(DrawList); draw_list->command_buffer = command_buffer; #ifdef DEBUG_ENABLED draw_list->validation.framebuffer_format = framebuffer->format_id; #endif draw_list_count = 0; draw_list_split = false; 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); draw_list->viewport = Rect2i(viewport_offset, viewport_size); 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 &p_clear_color_values, float p_clear_depth, uint32_t p_clear_stencil, const Rect2 &p_region, const Vector &p_storage_textures) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V(p_splits < 1, ERR_INVALID_DECLARATION); Framebuffer *framebuffer = framebuffer_owner.getornull(p_framebuffer); ERR_FAIL_COND_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) { //check clear values int color_attachments = framebuffer_formats[framebuffer->format_id].color_attachments; ERR_FAIL_COND_V_MSG(p_clear_color_values.size() != color_attachments, ERR_INVALID_PARAMETER, "Clear color values supplied (" + itos(p_clear_color_values.size()) + ") differ from the amount required for framebuffer (" + itos(color_attachments) + ")."); } 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(); 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); } } } 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); 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 = 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 = render_pass; inheritance_info.subpass = 0; inheritance_info.framebuffer = 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; #ifdef DEBUG_ENABLED draw_list[i].validation.framebuffer_format = framebuffer->format_id; #endif if (i == 0 && (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); r_split_ids[i] = (int64_t(ID_TYPE_SPLIT_DRAW_LIST) << ID_BASE_SHIFT) + i; draw_list[i].viewport = Rect2i(viewport_offset, viewport_size); } 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_bind_render_pipeline(DrawListID p_list, RID p_render_pipeline) { DrawList *dl = _get_draw_list_ptr(p_list); ERR_FAIL_COND(!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.getornull(p_render_pipeline); ERR_FAIL_COND(!pipeline); #ifdef DEBUG_ENABLED ERR_FAIL_COND(pipeline->validation.framebuffer_format != dl->validation.framebuffer_format); #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; #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_COND(!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.getornull(p_uniform_set); ERR_FAIL_COND(!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(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_COND(!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.getornull(p_vertex_array); ERR_FAIL_COND(!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_COND(!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.getornull(p_index_array); ERR_FAIL_COND(!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, index_array->offset, 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_COND(!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_COND(!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_COND(!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.getornull(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."); if (dl->validation.pipeline_vertex_format != INVALID_ID) { //uses vertices, do some vertex validations ERR_FAIL_COND_MSG(dl->validation.vertex_array_size < dl->validation.index_array_max_index, "Index array references (max index: " + itos(dl->validation.index_array_max_index) + ") indices beyond the vertex array size (" + itos(dl->validation.vertex_array_size) + ")."); } 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_COND(!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_COND(!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); } void RenderingDeviceVulkan::draw_list_end(uint32_t p_post_barrier) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_MSG(!draw_list, "Immediate draw list is already inactive."); 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; } vkCmdExecuteCommands(frames[frame].draw_command_buffer, draw_list_count, command_buffers); vkCmdEndRenderPass(frames[frame].draw_command_buffer); memdelete_arr(draw_list); draw_list = nullptr; } else { //just end the list vkCmdEndRenderPass(draw_list->command_buffer); memdelete(draw_list); draw_list = nullptr; } for (int i = 0; i < draw_list_bound_textures.size(); i++) { Texture *texture = texture_owner.getornull(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 & 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 & BARRIER_MASK_RASTER) { barrier_flags |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | 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_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT /*| VK_ACCESS_INDIRECT_COMMAND_READ_BIT*/; } if (p_post_barrier & 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_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.getornull(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) #ifdef FORCE_FULL_BARRIER _full_barrier(true); #else 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); } #endif } /***********************/ /**** COMPUTE LISTS ****/ /***********************/ RenderingDevice::ComputeListID RenderingDeviceVulkan::compute_list_begin(bool p_allow_draw_overlap) { 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."); 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) { ERR_FAIL_COND(p_list != ID_TYPE_COMPUTE_LIST); ERR_FAIL_COND(!compute_list); ComputeList *cl = compute_list; const ComputePipeline *pipeline = compute_pipeline_owner.getornull(p_compute_pipeline); ERR_FAIL_COND(!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; #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) { ERR_FAIL_COND(p_list != ID_TYPE_COMPUTE_LIST); ERR_FAIL_COND(!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.getornull(p_uniform_set); ERR_FAIL_COND(!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_compute = 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_compute = 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_COND(!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) { ERR_FAIL_COND(p_list != ID_TYPE_COMPUTE_LIST); ERR_FAIL_COND(!compute_list); ComputeList *cl = compute_list; #ifdef DEBUG_ENABLED 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_x_groups) + ") is larger than device limit (" + itos(limits.maxComputeWorkGroupCount[0]) + ")"); ERR_FAIL_COND_MSG(p_z_groups > limits.maxComputeWorkGroupCount[2], "Dispatch amount of Z compute groups (" + itos(p_x_groups) + ") is larger than device limit (" + itos(limits.maxComputeWorkGroupCount[0]) + ")"); 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.getornull(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_COND(!compute_list); 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_COND(!compute_list); ComputeList *cl = compute_list; Buffer *buffer = storage_buffer_owner.getornull(p_buffer); ERR_FAIL_COND(!buffer); ERR_FAIL_COND_MSG(!(buffer->usage & STORAGE_BUFFER_USAGE_DISPATCH_INDIRECT), "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.getornull(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) { #ifdef FORCE_FULL_BARRIER _full_barrier(true); #else _memory_barrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_ACCESS_SHADER_WRITE_BIT, VK_ACCESS_SHADER_READ_BIT, true); #endif } void RenderingDeviceVulkan::compute_list_end(uint32_t p_post_barrier) { ERR_FAIL_COND(!compute_list); uint32_t barrier_flags = 0; uint32_t access_flags = 0; if (p_post_barrier & 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 & BARRIER_MASK_RASTER) { barrier_flags |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | 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_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT; } if (p_post_barrier & 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; } 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); } uint32_t barrier_idx = 0; for (Set::Element *E = compute_list->state.textures_to_sampled_layout.front(); E; E = E->next()) { VkImageMemoryBarrier &image_memory_barrier = image_barriers[barrier_idx++]; 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 = access_flags; image_memory_barrier.oldLayout = E->get()->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->get()->image; image_memory_barrier.subresourceRange.aspectMask = E->get()->read_aspect_mask; image_memory_barrier.subresourceRange.baseMipLevel = E->get()->base_mipmap; image_memory_barrier.subresourceRange.levelCount = E->get()->mipmaps; image_memory_barrier.subresourceRange.baseArrayLayer = E->get()->base_layer; image_memory_barrier.subresourceRange.layerCount = E->get()->layers; E->get()->layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; if (E->get()->used_in_frame != frames_drawn) { E->get()->used_in_transfer = false; E->get()->used_in_raster = false; E->get()->used_in_compute = false; E->get()->used_in_frame = frames_drawn; } } #ifdef FORCE_FULL_BARRIER _full_barrier(true); #else 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 = access_flags; if (image_barrier_count > 0 || p_post_barrier != BARRIER_MASK_NO_BARRIER) { vkCmdPipelineBarrier(compute_list->command_buffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, barrier_flags, 0, 1, &mem_barrier, 0, nullptr, image_barrier_count, image_barriers); } #endif memdelete(compute_list); compute_list = nullptr; } void RenderingDeviceVulkan::barrier(uint32_t p_from, uint32_t p_to) { uint32_t src_barrier_flags = 0; uint32_t src_access_flags = 0; if (p_from & BARRIER_MASK_COMPUTE) { src_barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; src_access_flags |= VK_ACCESS_SHADER_WRITE_BIT; } if (p_from & BARRIER_MASK_RASTER) { src_barrier_flags |= VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_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 & BARRIER_MASK_TRANSFER) { src_barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; src_access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT; } if (p_from == 0) { src_barrier_flags = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT; } uint32_t dst_barrier_flags = 0; uint32_t dst_access_flags = 0; if (p_to & 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 & BARRIER_MASK_RASTER) { dst_barrier_flags |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | 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_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT; } if (p_to & BARRIER_MASK_TRANSFER) { dst_barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; dst_access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_TRANSFER_READ_BIT; } if (p_to == 0) { dst_barrier_flags = VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_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 &p_clear_colors) { VkCommandBuffer frame_cmdbuf = frames[frame].frame_buffer; ERR_FAIL_COND(!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) { //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.getornull(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.getornull(p_id); frames[frame].framebuffers_to_dispose_of.push_back(*framebuffer); framebuffer_owner.free(p_id); } else if (sampler_owner.owns(p_id)) { VkSampler *sampler = sampler_owner.getornull(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.getornull(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.getornull(p_id); Buffer b; b.allocation = index_buffer->allocation; b.buffer = index_buffer->buffer; b.size = index_buffer->size; 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.getornull(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.getornull(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.getornull(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.getornull(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.getornull(p_id); frames[frame].uniform_sets_to_dispose_of.push_back(*uniform_set); uniform_set_owner.free(p_id); } else if (render_pipeline_owner.owns(p_id)) { RenderPipeline *pipeline = render_pipeline_owner.getornull(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.getornull(p_id); frames[frame].compute_pipelines_to_dispose_of.push_back(*pipeline); compute_pipeline_owner.free(p_id); } else { ERR_PRINT("Attempted to free invalid ID: " + itos(p_id.get_id())); } } 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.getornull(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.getornull(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.getornull(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.getornull(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.getornull(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.getornull(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.getornull(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.getornull(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.getornull(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.getornull(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.getornull(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.getornull(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())); } } void RenderingDeviceVulkan::draw_command_begin_label(String p_label_name, const Color p_color) { 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) { context->command_insert_label(frames[frame].draw_command_buffer, p_label_name, p_color); } void RenderingDeviceVulkan::draw_command_end_label() { 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(); } 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, 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(); } 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() { 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() { 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; } 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 (Map::Element *E = framebuffer->framebuffers.front(); E; E = E->next()) { //first framebuffer, then render pass because it depends on it vkDestroyFramebuffer(device, E->get().framebuffer, nullptr); vkDestroyRenderPass(device, E->get().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 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() const { VmaStats stats; vmaCalculateStats(allocator, &stats); return stats.total.usedBytes; } 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) { 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; vmaCreateAllocator(&allocatorInfo, &allocator); } frames = memnew_arr(Frame, 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(); 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 = memnew_arr(String, max_timestamp_query_elements); frames[i].timestamp_cpu_values = memnew_arr(uint64_t, max_timestamp_query_elements); frames[i].timestamp_count = 0; frames[i].timestamp_result_names = memnew_arr(String, max_timestamp_query_elements); frames[i].timestamp_cpu_result_values = memnew_arr(uint64_t, max_timestamp_query_elements); frames[i].timestamp_result_values = memnew_arr(uint64_t, 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); } } staging_buffer_block_size = GLOBAL_DEF("rendering/vulkan/staging_buffer/block_size_kb", 256); staging_buffer_block_size = MAX(4, staging_buffer_block_size); staging_buffer_block_size *= 1024; //kb -> bytes staging_buffer_max_size = GLOBAL_DEF("rendering/vulkan/staging_buffer/max_size_mb", 128); staging_buffer_max_size = MAX(1, 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_DEF("rendering/vulkan/staging_buffer/texture_upload_region_size_px", 64); 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_DEF("rendering/vulkan/descriptor_pools/max_descriptors_per_pool", 64); //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; } template void RenderingDeviceVulkan::_free_rids(T &p_owner, const char *p_type) { List owned; p_owner.get_owned_list(&owned); if (owned.size()) { WARN_PRINT(itos(owned.size()) + " RIDs of type '" + p_type + "' were leaked."); for (List::Element *E = owned.front(); E; E = E->next()) { free(E->get()); } } } void RenderingDeviceVulkan::capture_timestamp(const String &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++; } 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]; } int RenderingDeviceVulkan::limit_get(Limit p_limit) { 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]; 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 owned; texture_owner.get_owned_list(&owned); if (owned.size()) { WARN_PRINT(itos(owned.size()) + " RIDs of type 'Texture' were leaked."); //free shared first for (List::Element *E = owned.front(); E;) { List::Element *N = E->next(); if (texture_is_shared(E->get())) { free(E->get()); owned.erase(E->get()); } E = N; } //free non shared second, this will avoid an error trying to free unexisting textures due to dependencies. for (List::Element *E = owned.front(); E; E = E->next()) { free(E->get()); } } } //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); memdelete_arr(frames[i].timestamp_names); memdelete_arr(frames[i].timestamp_cpu_values); memdelete_arr(frames[i].timestamp_result_names); memdelete_arr(frames[i].timestamp_result_values); memdelete_arr(frames[i].timestamp_cpu_result_values); } for (int i = 0; i < split_draw_list_allocators.size(); i++) { vkDestroyCommandPool(device, split_draw_list_allocators[i].command_pool, nullptr); } memdelete_arr(frames); for (int i = 0; i < staging_buffer_blocks.size(); i++) { vmaDestroyBuffer(allocator, staging_buffer_blocks[i].buffer, staging_buffer_blocks[i].allocation); } vmaDestroyAllocator(allocator); while (vertex_formats.size()) { Map::Element *temp = vertex_formats.front(); memdelete_arr(temp->get().bindings); memdelete_arr(temp->get().attributes); vertex_formats.erase(temp); } for (int i = 0; i < framebuffer_formats.size(); i++) { vkDestroyRenderPass(device, framebuffer_formats[i].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; } RenderingDeviceVulkan::RenderingDeviceVulkan() { } RenderingDeviceVulkan::~RenderingDeviceVulkan() { if (local_device.is_valid()) { finalize(); context->local_device_free(local_device); } }