diff options
Diffstat (limited to 'drivers/vulkan/rendering_device_vulkan.cpp')
| -rw-r--r-- | drivers/vulkan/rendering_device_vulkan.cpp | 2640 |
1 files changed, 1948 insertions, 692 deletions
diff --git a/drivers/vulkan/rendering_device_vulkan.cpp b/drivers/vulkan/rendering_device_vulkan.cpp index 469d80368d..77aab72d40 100644 --- a/drivers/vulkan/rendering_device_vulkan.cpp +++ b/drivers/vulkan/rendering_device_vulkan.cpp @@ -5,8 +5,8 @@ /* GODOT ENGINE */ /* https://godotengine.org */ /*************************************************************************/ -/* Copyright (c) 2007-2021 Juan Linietsky, Ariel Manzur. */ -/* Copyright (c) 2014-2021 Godot Engine contributors (cf. AUTHORS.md). */ +/* Copyright (c) 2007-2022 Juan Linietsky, Ariel Manzur. */ +/* Copyright (c) 2014-2022 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 */ @@ -31,20 +31,25 @@ #include "rendering_device_vulkan.h" #include "core/config/project_settings.h" -#include "core/os/file_access.h" +#include "core/io/compression.h" +#include "core/io/file_access.h" +#include "core/io/marshalls.h" #include "core/os/os.h" #include "core/templates/hashfuncs.h" #include "drivers/vulkan/vulkan_context.h" +#include "thirdparty/misc/smolv.h" #include "thirdparty/spirv-reflect/spirv_reflect.h" //#define FORCE_FULL_BARRIER +static const uint32_t SMALL_ALLOCATION_MAX_SIZE = 4096; + // Get the Vulkan object information and possible stage access types (bitwise OR'd with incoming values) RenderingDeviceVulkan::Buffer *RenderingDeviceVulkan::_get_buffer_from_owner(RID p_buffer, VkPipelineStageFlags &r_stage_mask, VkAccessFlags &r_access_mask, uint32_t p_post_barrier) { Buffer *buffer = nullptr; if (vertex_buffer_owner.owns(p_buffer)) { - buffer = vertex_buffer_owner.getornull(p_buffer); + buffer = vertex_buffer_owner.get_or_null(p_buffer); r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT; r_access_mask |= VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT; @@ -61,7 +66,7 @@ RenderingDeviceVulkan::Buffer *RenderingDeviceVulkan::_get_buffer_from_owner(RID } 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); + buffer = index_buffer_owner.get_or_null(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; @@ -70,7 +75,7 @@ RenderingDeviceVulkan::Buffer *RenderingDeviceVulkan::_get_buffer_from_owner(RID r_stage_mask |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; } r_access_mask |= VK_ACCESS_UNIFORM_READ_BIT; - buffer = uniform_buffer_owner.getornull(p_buffer); + buffer = uniform_buffer_owner.get_or_null(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; @@ -81,9 +86,9 @@ RenderingDeviceVulkan::Buffer *RenderingDeviceVulkan::_get_buffer_from_owner(RID r_access_mask |= VK_ACCESS_SHADER_READ_BIT; } - buffer = &texture_buffer_owner.getornull(p_buffer)->buffer; + buffer = &texture_buffer_owner.get_or_null(p_buffer)->buffer; } else if (storage_buffer_owner.owns(p_buffer)) { - buffer = storage_buffer_owner.getornull(p_buffer); + buffer = storage_buffer_owner.get_or_null(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; @@ -127,13 +132,13 @@ static void update_external_dependency_for_store(VkSubpassDependency &dependency void RenderingDeviceVulkan::_add_dependency(RID p_id, RID p_depends_on) { if (!dependency_map.has(p_depends_on)) { - dependency_map[p_depends_on] = Set<RID>(); + dependency_map[p_depends_on] = HashSet<RID>(); } dependency_map[p_depends_on].insert(p_id); if (!reverse_dependency_map.has(p_id)) { - reverse_dependency_map[p_id] = Set<RID>(); + reverse_dependency_map[p_id] = HashSet<RID>(); } reverse_dependency_map[p_id].insert(p_depends_on); @@ -142,26 +147,26 @@ void RenderingDeviceVulkan::_add_dependency(RID p_id, RID p_depends_on) { void RenderingDeviceVulkan::_free_dependencies(RID p_id) { //direct dependencies must be freed - Map<RID, Set<RID>>::Element *E = dependency_map.find(p_id); + HashMap<RID, HashSet<RID>>::Iterator E = dependency_map.find(p_id); if (E) { - while (E->get().size()) { - free(E->get().front()->get()); + while (E->value.size()) { + free(*E->value.begin()); } - dependency_map.erase(E); + dependency_map.remove(E); } //reverse dependencies must be unreferenced E = reverse_dependency_map.find(p_id); if (E) { - for (Set<RID>::Element *F = E->get().front(); F; F = F->next()) { - Map<RID, Set<RID>>::Element *G = dependency_map.find(F->get()); + for (const RID &F : E->value) { + HashMap<RID, HashSet<RID>>::Iterator G = dependency_map.find(F); ERR_CONTINUE(!G); - ERR_CONTINUE(!G->get().has(p_id)); - G->get().erase(p_id); + ERR_CONTINUE(!G->value.has(p_id)); + G->value.erase(p_id); } - reverse_dependency_map.erase(E); + reverse_dependency_map.remove(E); } } @@ -384,14 +389,6 @@ const VkFormat RenderingDeviceVulkan::vulkan_formats[RenderingDevice::DATA_FORMA 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] = { @@ -613,14 +610,6 @@ const char *RenderingDeviceVulkan::named_formats[RenderingDevice::DATA_FORMAT_MA "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) { @@ -967,15 +956,6 @@ uint32_t RenderingDeviceVulkan::get_image_format_pixel_size(DataFormat p_format) 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"); } @@ -1045,20 +1025,6 @@ void RenderingDeviceVulkan::get_compressed_image_format_block_dimensions(DataFor 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; @@ -1135,15 +1101,6 @@ uint32_t RenderingDeviceVulkan::get_compressed_image_format_block_byte_size(Data 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: { } } @@ -1164,16 +1121,7 @@ uint32_t RenderingDeviceVulkan::get_compressed_image_format_pixel_rshift(DataFor 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: { } } @@ -1228,7 +1176,7 @@ uint32_t RenderingDeviceVulkan::get_image_format_required_size(DataFormat p_form } w = MAX(blockw, w >> 1); h = MAX(blockh, h >> 1); - d = MAX(1, d >> 1); + d = MAX(1u, d >> 1); } return size; @@ -1236,23 +1184,23 @@ uint32_t RenderingDeviceVulkan::get_image_format_required_size(DataFormat p_form 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; + uint32_t w = p_width; + uint32_t h = p_height; + uint32_t d = p_depth; - int mipmaps = 1; + uint32_t 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); + w = MAX(1u, w >> 1); + h = MAX(1u, h >> 1); + d = MAX(1u, d >> 1); mipmaps++; - }; + } return mipmaps; } @@ -1370,7 +1318,7 @@ const VkImageType RenderingDeviceVulkan::vulkan_image_type[RenderingDevice::TEXT /**** BUFFER MANAGEMENT ****/ /***************************/ -Error RenderingDeviceVulkan::_buffer_allocate(Buffer *p_buffer, uint32_t p_size, uint32_t p_usage, VmaMemoryUsage p_mapping) { +Error RenderingDeviceVulkan::_buffer_allocate(Buffer *p_buffer, uint32_t p_size, uint32_t p_usage, VmaMemoryUsage p_mem_usage, VmaAllocationCreateFlags p_mem_flags) { VkBufferCreateInfo bufferInfo; bufferInfo.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO; bufferInfo.pNext = nullptr; @@ -1382,13 +1330,18 @@ Error RenderingDeviceVulkan::_buffer_allocate(Buffer *p_buffer, uint32_t p_size, bufferInfo.pQueueFamilyIndices = nullptr; VmaAllocationCreateInfo allocInfo; - allocInfo.flags = 0; - allocInfo.usage = p_mapping; + allocInfo.flags = p_mem_flags; + allocInfo.usage = p_mem_usage; allocInfo.requiredFlags = 0; allocInfo.preferredFlags = 0; allocInfo.memoryTypeBits = 0; allocInfo.pool = nullptr; allocInfo.pUserData = nullptr; + if (p_size <= SMALL_ALLOCATION_MAX_SIZE) { + uint32_t mem_type_index = 0; + vmaFindMemoryTypeIndexForBufferInfo(allocator, &bufferInfo, &allocInfo, &mem_type_index); + allocInfo.pool = _find_or_create_small_allocs_pool(mem_type_index); + } VkResult err = vmaCreateBuffer(allocator, &bufferInfo, &allocInfo, &p_buffer->buffer, &p_buffer->allocation, nullptr); ERR_FAIL_COND_V_MSG(err, ERR_CANT_CREATE, "Can't create buffer of size: " + itos(p_size) + ", error " + itos(err) + "."); @@ -1398,12 +1351,15 @@ Error RenderingDeviceVulkan::_buffer_allocate(Buffer *p_buffer, uint32_t p_size, p_buffer->buffer_info.range = p_size; p_buffer->usage = p_usage; + buffer_memory += p_size; + return OK; } Error RenderingDeviceVulkan::_buffer_free(Buffer *p_buffer) { ERR_FAIL_COND_V(p_buffer->size == 0, ERR_INVALID_PARAMETER); + buffer_memory -= p_buffer->size; vmaDestroyBuffer(allocator, p_buffer->buffer, p_buffer->allocation); p_buffer->buffer = VK_NULL_HANDLE; p_buffer->allocation = nullptr; @@ -1424,8 +1380,8 @@ Error RenderingDeviceVulkan::_insert_staging_block() { bufferInfo.pQueueFamilyIndices = nullptr; VmaAllocationCreateInfo allocInfo; - allocInfo.flags = 0; - allocInfo.usage = VMA_MEMORY_USAGE_CPU_ONLY; + allocInfo.flags = VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT; + allocInfo.usage = VMA_MEMORY_USAGE_AUTO_PREFER_HOST; allocInfo.requiredFlags = 0; allocInfo.preferredFlags = 0; allocInfo.memoryTypeBits = 0; @@ -1486,7 +1442,7 @@ Error RenderingDeviceVulkan::_staging_buffer_allocate(uint32_t p_amount, uint32_ // 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) { + if ((uint64_t)staging_buffer_blocks.size() * staging_buffer_block_size < staging_buffer_max_size) { //we can, so we are safe Error err = _insert_staging_block(); if (err) { @@ -1528,9 +1484,9 @@ Error RenderingDeviceVulkan::_staging_buffer_allocate(uint32_t p_amount, uint32_ //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) { + } else { //this block may still be in use, let's not touch it unless we have to, so.. can we create a new one? - if (staging_buffer_blocks.size() * staging_buffer_block_size < staging_buffer_max_size) { + if ((uint64_t)staging_buffer_blocks.size() * staging_buffer_block_size < staging_buffer_max_size) { //we are still allowed to create a new block, so let's do that and insert it for current pos Error err = _insert_staging_block(); if (err) { @@ -1600,7 +1556,7 @@ Error RenderingDeviceVulkan::_buffer_update(Buffer *p_buffer, size_t p_offset, c } //copy to staging buffer - copymem(((uint8_t *)data_ptr) + block_write_offset, p_data + submit_from, block_write_amount); + memcpy(((uint8_t *)data_ptr) + block_write_offset, p_data + submit_from, block_write_amount); //unmap vmaUnmapMemory(allocator, staging_buffer_blocks[staging_buffer_current].allocation); @@ -1798,6 +1754,10 @@ RID RenderingDeviceVulkan::texture_create(const TextureFormat &p_format, const T image_create_info.usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT; } + if (p_format.usage_bits & TEXTURE_USAGE_INPUT_ATTACHMENT_BIT) { + image_create_info.usage |= VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT; + } + if (p_format.usage_bits & TEXTURE_USAGE_CAN_UPDATE_BIT) { image_create_info.usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT; } @@ -1883,20 +1843,28 @@ RID RenderingDeviceVulkan::texture_create(const TextureFormat &p_format, const T //allocate memory + uint32_t width, height; + uint32_t image_size = get_image_format_required_size(p_format.format, p_format.width, p_format.height, p_format.depth, p_format.mipmaps, &width, &height); + VmaAllocationCreateInfo allocInfo; - allocInfo.flags = 0; - allocInfo.usage = p_format.usage_bits & TEXTURE_USAGE_CPU_READ_BIT ? VMA_MEMORY_USAGE_CPU_ONLY : VMA_MEMORY_USAGE_GPU_ONLY; + allocInfo.flags = (p_format.usage_bits & TEXTURE_USAGE_CPU_READ_BIT) ? VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT : 0; + allocInfo.pool = nullptr; + allocInfo.usage = VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE; allocInfo.requiredFlags = 0; allocInfo.preferredFlags = 0; allocInfo.memoryTypeBits = 0; - allocInfo.pool = nullptr; allocInfo.pUserData = nullptr; + if (image_size <= SMALL_ALLOCATION_MAX_SIZE) { + uint32_t mem_type_index = 0; + vmaFindMemoryTypeIndexForImageInfo(allocator, &image_create_info, &allocInfo, &mem_type_index); + allocInfo.pool = _find_or_create_small_allocs_pool(mem_type_index); + } Texture texture; VkResult err = vmaCreateImage(allocator, &image_create_info, &allocInfo, &texture.image, &texture.allocation, &texture.allocation_info); ERR_FAIL_COND_V_MSG(err, RID(), "vmaCreateImage failed with error " + itos(err) + "."); - + image_memory += texture.allocation_info.size; texture.type = p_format.texture_type; texture.format = p_format.format; texture.width = image_create_info.extent.width; @@ -2029,7 +1997,7 @@ RID RenderingDeviceVulkan::texture_create(const TextureFormat &p_format, const T if (p_data.size()) { for (uint32_t i = 0; i < image_create_info.arrayLayers; i++) { - texture_update(id, i, p_data[i]); + _texture_update(id, i, p_data[i], RD::BARRIER_MASK_ALL, true); } } return id; @@ -2038,12 +2006,12 @@ RID RenderingDeviceVulkan::texture_create(const TextureFormat &p_format, const T 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); + Texture *src_texture = texture_owner.get_or_null(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); + src_texture = texture_owner.get_or_null(src_texture->owner); ERR_FAIL_COND_V(!src_texture, RID()); //this is a bug } @@ -2128,6 +2096,10 @@ RID RenderingDeviceVulkan::texture_create_shared(const TextureView &p_view, RID } } + if (texture.usage_flags & TEXTURE_USAGE_INPUT_ATTACHMENT_BIT) { + usage_info.usage |= VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT; + } + if (texture.usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { usage_info.usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT; } @@ -2156,15 +2128,133 @@ RID RenderingDeviceVulkan::texture_create_shared(const TextureView &p_view, RID 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) { +RID RenderingDeviceVulkan::texture_create_from_extension(TextureType p_type, DataFormat p_format, TextureSamples p_samples, uint64_t p_flags, uint64_t p_image, uint64_t p_width, uint64_t p_height, uint64_t p_depth, uint64_t p_layers) { _THREAD_SAFE_METHOD_ + // This method creates a texture object using a VkImage created by an extension, module or other external source (OpenXR uses this). + VkImage image = (VkImage)p_image; + + Texture texture; + texture.image = image; + // if we leave texture.allocation as a nullptr, would that be enough to detect we don't "own" the image? + // also leave texture.allocation_info alone + // we'll set texture.view later on + texture.type = p_type; + texture.format = p_format; + texture.samples = p_samples; + texture.width = p_width; + texture.height = p_height; + texture.depth = p_depth; + texture.layers = p_layers; + texture.mipmaps = 0; // maybe make this settable too? + texture.usage_flags = p_flags; + texture.base_mipmap = 0; + texture.base_layer = 0; + texture.allowed_shared_formats.push_back(RD::DATA_FORMAT_R8G8B8A8_UNORM); + texture.allowed_shared_formats.push_back(RD::DATA_FORMAT_R8G8B8A8_SRGB); + + // Do we need to do something with texture.layout ? + + if (texture.usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { + texture.read_aspect_mask = VK_IMAGE_ASPECT_DEPTH_BIT; + texture.barrier_aspect_mask = VK_IMAGE_ASPECT_DEPTH_BIT; + + // if (format_has_stencil(p_format.format)) { + // texture.barrier_aspect_mask |= VK_IMAGE_ASPECT_STENCIL_BIT; + // } + } else { + texture.read_aspect_mask = VK_IMAGE_ASPECT_COLOR_BIT; + texture.barrier_aspect_mask = VK_IMAGE_ASPECT_COLOR_BIT; + } + + // Create a view for us to use + + VkImageViewCreateInfo image_view_create_info; + image_view_create_info.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO; + image_view_create_info.pNext = nullptr; + image_view_create_info.flags = 0; + image_view_create_info.image = texture.image; + + static const VkImageViewType view_types[TEXTURE_TYPE_MAX] = { + VK_IMAGE_VIEW_TYPE_1D, + VK_IMAGE_VIEW_TYPE_2D, + VK_IMAGE_VIEW_TYPE_3D, + VK_IMAGE_VIEW_TYPE_CUBE, + VK_IMAGE_VIEW_TYPE_1D_ARRAY, + VK_IMAGE_VIEW_TYPE_2D_ARRAY, + VK_IMAGE_VIEW_TYPE_CUBE_ARRAY, + }; + + image_view_create_info.viewType = view_types[texture.type]; + image_view_create_info.format = vulkan_formats[texture.format]; + + static const VkComponentSwizzle component_swizzles[TEXTURE_SWIZZLE_MAX] = { + VK_COMPONENT_SWIZZLE_IDENTITY, + VK_COMPONENT_SWIZZLE_ZERO, + VK_COMPONENT_SWIZZLE_ONE, + VK_COMPONENT_SWIZZLE_R, + VK_COMPONENT_SWIZZLE_G, + VK_COMPONENT_SWIZZLE_B, + VK_COMPONENT_SWIZZLE_A + }; + + // hardcode for now, maybe make this settable from outside.. + image_view_create_info.components.r = component_swizzles[TEXTURE_SWIZZLE_R]; + image_view_create_info.components.g = component_swizzles[TEXTURE_SWIZZLE_G]; + image_view_create_info.components.b = component_swizzles[TEXTURE_SWIZZLE_B]; + image_view_create_info.components.a = component_swizzles[TEXTURE_SWIZZLE_A]; + + image_view_create_info.subresourceRange.baseMipLevel = 0; + image_view_create_info.subresourceRange.levelCount = texture.mipmaps; + image_view_create_info.subresourceRange.baseArrayLayer = 0; + image_view_create_info.subresourceRange.layerCount = texture.layers; + if (texture.usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { + image_view_create_info.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT; + } else { + image_view_create_info.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; + } + + VkResult err = vkCreateImageView(device, &image_view_create_info, nullptr, &texture.view); - Texture *src_texture = texture_owner.getornull(p_with_texture); + if (err) { + // vmaDestroyImage(allocator, texture.image, texture.allocation); + ERR_FAIL_V_MSG(RID(), "vkCreateImageView failed with error " + itos(err) + "."); + } + + //barrier to set layout + { + VkImageMemoryBarrier image_memory_barrier; + image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; + image_memory_barrier.pNext = nullptr; + image_memory_barrier.srcAccessMask = 0; + image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT; + image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED; + image_memory_barrier.newLayout = texture.layout; + image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; + image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; + image_memory_barrier.image = texture.image; + image_memory_barrier.subresourceRange.aspectMask = texture.barrier_aspect_mask; + image_memory_barrier.subresourceRange.baseMipLevel = 0; + image_memory_barrier.subresourceRange.levelCount = texture.mipmaps; + image_memory_barrier.subresourceRange.baseArrayLayer = 0; + image_memory_barrier.subresourceRange.layerCount = texture.layers; + + vkCmdPipelineBarrier(frames[frame].setup_command_buffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); + } + + RID id = texture_owner.make_rid(texture); + + return id; +} + +RID RenderingDeviceVulkan::texture_create_shared_from_slice(const TextureView &p_view, RID p_with_texture, uint32_t p_layer, uint32_t p_mipmap, uint32_t p_mipmaps, TextureSliceType p_slice_type) { + _THREAD_SAFE_METHOD_ + + Texture *src_texture = texture_owner.get_or_null(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); + src_texture = texture_owner.get_or_null(src_texture->owner); ERR_FAIL_COND_V(!src_texture, RID()); //this is a bug } @@ -2180,6 +2270,7 @@ RID RenderingDeviceVulkan::texture_create_shared_from_slice(const TextureView &p //create view ERR_FAIL_UNSIGNED_INDEX_V(p_mipmap, src_texture->mipmaps, RID()); + ERR_FAIL_COND_V(p_mipmap + p_mipmaps > src_texture->mipmaps, RID()); ERR_FAIL_UNSIGNED_INDEX_V(p_layer, src_texture->layers, RID()); int slice_layers = 1; @@ -2192,7 +2283,7 @@ RID RenderingDeviceVulkan::texture_create_shared_from_slice(const TextureView &p 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.mipmaps = p_mipmaps; texture.layers = slice_layers; texture.base_mipmap = p_mipmap; texture.base_layer = p_layer; @@ -2255,7 +2346,7 @@ RID RenderingDeviceVulkan::texture_create_shared_from_slice(const TextureView &p "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.levelCount = p_mipmaps; image_view_create_info.subresourceRange.layerCount = slice_layers; image_view_create_info.subresourceRange.baseArrayLayer = p_layer; @@ -2276,17 +2367,21 @@ RID RenderingDeviceVulkan::texture_create_shared_from_slice(const TextureView &p } Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, const Vector<uint8_t> &p_data, uint32_t p_post_barrier) { + return _texture_update(p_texture, p_layer, p_data, p_post_barrier, false); +} + +Error RenderingDeviceVulkan::_texture_update(RID p_texture, uint32_t p_layer, const Vector<uint8_t> &p_data, uint32_t p_post_barrier, bool p_use_setup_queue) { _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"); + ERR_FAIL_COND_V_MSG((draw_list || compute_list) && !p_use_setup_queue, ERR_INVALID_PARAMETER, + "Updating textures is forbidden during creation of a draw or compute list"); - Texture *texture = texture_owner.getornull(p_texture); + Texture *texture = texture_owner.get_or_null(p_texture); ERR_FAIL_COND_V(!texture, ERR_INVALID_PARAMETER); if (texture->owner != RID()) { p_texture = texture->owner; - texture = texture_owner.getornull(texture->owner); + texture = texture_owner.get_or_null(texture->owner); ERR_FAIL_COND_V(!texture, ERR_BUG); //this is a bug } @@ -2320,7 +2415,7 @@ Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, con const uint8_t *r = p_data.ptr(); - VkCommandBuffer command_buffer = p_post_barrier ? frames[frame].draw_command_buffer : frames[frame].setup_command_buffer; + VkCommandBuffer command_buffer = p_use_setup_queue ? frames[frame].setup_command_buffer : frames[frame].draw_command_buffer; //barrier to transfer { @@ -2373,7 +2468,7 @@ Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, con 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); + Error err = _staging_buffer_allocate(to_allocate, required_align, alloc_offset, alloc_size, false, !p_use_setup_queue); ERR_FAIL_COND_V(err, ERR_CANT_CREATE); uint8_t *write_ptr; @@ -2461,8 +2556,8 @@ Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, con } mipmap_offset = image_total; - logic_width = MAX(1, logic_width >> 1); - logic_height = MAX(1, logic_height >> 1); + logic_width = MAX(1u, logic_width >> 1); + logic_height = MAX(1u, logic_height >> 1); } //barrier to restore layout @@ -2502,7 +2597,7 @@ Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, con 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); + vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, barrier_flags, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } if (texture->used_in_frame != frames_drawn) { @@ -2558,7 +2653,7 @@ Vector<uint8_t> RenderingDeviceVulkan::_texture_get_data_from_image(Texture *tex const uint8_t *rptr = slice_read_ptr + y * layout.rowPitch; uint8_t *wptr = write_ptr + y * line_width; - copymem(wptr, rptr, line_width); + memcpy(wptr, rptr, line_width); } } else { @@ -2566,7 +2661,7 @@ Vector<uint8_t> RenderingDeviceVulkan::_texture_get_data_from_image(Texture *tex 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); + memcpy(wptr, rptr, (uint64_t)pixel_size * width); } } } @@ -2583,7 +2678,7 @@ Vector<uint8_t> RenderingDeviceVulkan::_texture_get_data_from_image(Texture *tex Vector<uint8_t> RenderingDeviceVulkan::texture_get_data(RID p_texture, uint32_t p_layer) { _THREAD_SAFE_METHOD_ - Texture *tex = texture_owner.getornull(p_texture); + Texture *tex = texture_owner.get_or_null(p_texture); ERR_FAIL_COND_V(!tex, Vector<uint8_t>()); ERR_FAIL_COND_V_MSG(tex->bound, Vector<uint8_t>(), @@ -2608,7 +2703,7 @@ Vector<uint8_t> RenderingDeviceVulkan::texture_get_data(RID p_texture, uint32_t //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); + _buffer_allocate(&tmp_buffer, buffer_size, VK_BUFFER_USAGE_TRANSFER_DST_BIT, VMA_MEMORY_USAGE_AUTO_PREFER_HOST, VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT); { //Source image barrier VkImageMemoryBarrier image_memory_barrier; @@ -2660,9 +2755,9 @@ Vector<uint8_t> RenderingDeviceVulkan::texture_get_data(RID p_texture, uint32_t 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); + computed_w = MAX(1u, computed_w >> 1); + computed_h = MAX(1u, computed_h >> 1); + computed_d = MAX(1u, computed_d >> 1); offset += size; } @@ -2699,7 +2794,7 @@ Vector<uint8_t> RenderingDeviceVulkan::texture_get_data(RID p_texture, uint32_t { buffer_data.resize(buffer_size); uint8_t *w = buffer_data.ptrw(); - copymem(w, buffer_mem, buffer_size); + memcpy(w, buffer_mem, buffer_size); } vmaUnmapMemory(allocator, tmp_buffer.allocation); @@ -2713,7 +2808,7 @@ Vector<uint8_t> RenderingDeviceVulkan::texture_get_data(RID p_texture, uint32_t bool RenderingDeviceVulkan::texture_is_shared(RID p_texture) { _THREAD_SAFE_METHOD_ - Texture *tex = texture_owner.getornull(p_texture); + Texture *tex = texture_owner.get_or_null(p_texture); ERR_FAIL_COND_V(!tex, false); return tex->owner.is_valid(); } @@ -2722,10 +2817,18 @@ bool RenderingDeviceVulkan::texture_is_valid(RID p_texture) { return texture_owner.owns(p_texture); } +Size2i RenderingDeviceVulkan::texture_size(RID p_texture) { + _THREAD_SAFE_METHOD_ + + Texture *tex = texture_owner.get_or_null(p_texture); + ERR_FAIL_COND_V(!tex, Size2i()); + return Size2i(tex->width, tex->height); +} + 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); + Texture *src_tex = texture_owner.get_or_null(p_from_texture); ERR_FAIL_COND_V(!src_tex, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V_MSG(src_tex->bound, ERR_INVALID_PARAMETER, @@ -2746,7 +2849,7 @@ Error RenderingDeviceVulkan::texture_copy(RID p_from_texture, RID p_to_texture, 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); + Texture *dst_tex = texture_owner.get_or_null(p_to_texture); ERR_FAIL_COND_V(!dst_tex, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V_MSG(dst_tex->bound, ERR_INVALID_PARAMETER, @@ -2913,7 +3016,7 @@ Error RenderingDeviceVulkan::texture_copy(RID p_from_texture, RID p_to_texture, 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); + Texture *src_tex = texture_owner.get_or_null(p_from_texture); ERR_FAIL_COND_V(!src_tex, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V_MSG(src_tex->bound, ERR_INVALID_PARAMETER, @@ -2924,7 +3027,7 @@ Error RenderingDeviceVulkan::texture_resolve_multisample(RID p_from_texture, RID 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); + Texture *dst_tex = texture_owner.get_or_null(p_to_texture); ERR_FAIL_COND_V(!dst_tex, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V_MSG(dst_tex->bound, ERR_INVALID_PARAMETER, @@ -3084,7 +3187,7 @@ Error RenderingDeviceVulkan::texture_resolve_multisample(RID p_from_texture, RID 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); + Texture *src_tex = texture_owner.get_or_null(p_texture); ERR_FAIL_COND_V(!src_tex, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V_MSG(src_tex->bound, ERR_INVALID_PARAMETER, @@ -3244,58 +3347,75 @@ bool RenderingDeviceVulkan::texture_is_format_supported_for_usage(DataFormat p_f /**** ATTACHMENT ****/ /********************/ -VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentFormat> &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<VkAttachmentDescription> attachments; - Vector<VkAttachmentReference> color_references; - Vector<VkAttachmentReference> depth_stencil_references; - Vector<VkAttachmentReference> resolve_references; - +VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentFormat> &p_attachments, const Vector<FramebufferPass> &p_passes, InitialAction p_initial_action, FinalAction p_final_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, uint32_t p_view_count, Vector<TextureSamples> *r_samples) { // Set up dependencies from/to external equivalent to the default (implicit) one, and then amend them const VkPipelineStageFlags default_access_mask = VK_ACCESS_INPUT_ATTACHMENT_READ_BIT | - VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | - VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; // From Section 7.1 of Vulkan API Spec v1.1.148 + VK_ACCESS_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]; + LocalVector<int32_t> attachment_last_pass; + attachment_last_pass.resize(p_attachments.size()); - 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."); + Vector<VkAttachmentDescription> attachments; + Vector<int> attachment_remap; + + for (int i = 0; i < p_attachments.size(); i++) { + if (p_attachments[i].usage_flags == AttachmentFormat::UNUSED_ATTACHMENT) { + attachment_remap.push_back(VK_ATTACHMENT_UNUSED); + continue; + } + + ERR_FAIL_INDEX_V(p_attachments[i].format, DATA_FORMAT_MAX, VK_NULL_HANDLE); + ERR_FAIL_INDEX_V(p_attachments[i].samples, TEXTURE_SAMPLES_MAX, VK_NULL_HANDLE); + ERR_FAIL_COND_V_MSG(!(p_attachments[i].usage_flags & (TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | TEXTURE_USAGE_INPUT_ATTACHMENT_BIT)), + VK_NULL_HANDLE, "Texture format for index (" + itos(i) + ") requires an attachment (color, depth, input or stencil) bit set."); VkAttachmentDescription description = {}; description.flags = 0; - description.format = vulkan_formats[p_format[i].format]; - description.samples = rasterization_sample_count[p_format[i].samples]; + description.format = vulkan_formats[p_attachments[i].format]; + description.samples = rasterization_sample_count[p_attachments[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; + bool is_sampled = p_attachments[i].usage_flags & TEXTURE_USAGE_SAMPLING_BIT; + bool is_storage = p_attachments[i].usage_flags & TEXTURE_USAGE_STORAGE_BIT; + bool is_depth = p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT; // 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 + // 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) { + + switch (is_depth ? p_initial_depth_action : p_initial_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; + if (p_attachments[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { + description.loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR; + description.initialLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL); + description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; + } else if (p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { + description.loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR; + description.initialLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL); + description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_CLEAR; + dependency_from_external.srcStageMask |= reading_stages; + } else { + description.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; + description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; + description.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; //don't care what is there + dependency_from_external.srcStageMask |= reading_stages; + } } break; case INITIAL_ACTION_KEEP: { - if (p_format[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { + if (p_attachments[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { description.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD; description.initialLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL); description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; - } else if (p_format[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { + } else if (p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { description.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD; description.initialLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL); description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD; @@ -3308,11 +3428,11 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF } } break; case INITIAL_ACTION_DROP: { - if (p_format[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { + if (p_attachments[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { description.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; description.initialLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL); description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; - } else if (p_format[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { + } else if (p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { description.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; description.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; //don't care what is there description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; @@ -3326,11 +3446,11 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF } break; case INITIAL_ACTION_CLEAR_REGION_CONTINUE: case INITIAL_ACTION_CONTINUE: { - if (p_format[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { + if (p_attachments[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { description.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD; description.initialLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; - } else if (p_format[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { + } else if (p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { description.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD; description.initialLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD; @@ -3346,14 +3466,65 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF } } - switch (is_depth_stencil ? p_final_depth_action : p_final_color_action) { + bool used_last = false; + + { + int last_pass = p_passes.size() - 1; + + if (is_depth) { + //likely missing depth resolve? + if (p_passes[last_pass].depth_attachment == i) { + used_last = true; + } + } else { + if (p_passes[last_pass].resolve_attachments.size()) { + //if using resolve attachments, check resolve attachments + for (int j = 0; j < p_passes[last_pass].resolve_attachments.size(); j++) { + if (p_passes[last_pass].resolve_attachments[j] == i) { + used_last = true; + break; + } + } + } else { + for (int j = 0; j < p_passes[last_pass].color_attachments.size(); j++) { + if (p_passes[last_pass].color_attachments[j] == i) { + used_last = true; + break; + } + } + } + } + + if (!used_last) { + for (int j = 0; j < p_passes[last_pass].preserve_attachments.size(); j++) { + if (p_passes[last_pass].preserve_attachments[j] == i) { + used_last = true; + break; + } + } + } + } + + FinalAction final_action = p_final_action; + FinalAction final_depth_action = p_final_depth_action; + + if (!used_last) { + if (is_depth) { + final_depth_action = FINAL_ACTION_DISCARD; + + } else { + final_action = FINAL_ACTION_DISCARD; + } + } + + switch (is_depth ? final_depth_action : final_action) { case FINAL_ACTION_READ: { - if (p_format[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { + if (p_attachments[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { description.storeOp = VK_ATTACHMENT_STORE_OP_STORE; description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.finalLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL); update_external_dependency_for_store(dependency_to_external, is_sampled, is_storage, false); - } else if (p_format[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { + } else if (p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { description.storeOp = VK_ATTACHMENT_STORE_OP_STORE; description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE; description.finalLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL); @@ -3366,11 +3537,11 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF } } break; case FINAL_ACTION_DISCARD: { - if (p_format[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { + if (p_attachments[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { description.storeOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.finalLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL); - } else if (p_format[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { + } else if (p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { description.storeOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.finalLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL); @@ -3381,11 +3552,11 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF } } break; case FINAL_ACTION_CONTINUE: { - if (p_format[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { + if (p_attachments[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { description.storeOp = VK_ATTACHMENT_STORE_OP_STORE; description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; description.finalLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; - } else if (p_format[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { + } else if (p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { description.storeOp = VK_ATTACHMENT_STORE_OP_STORE; description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE; description.finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; @@ -3401,27 +3572,195 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF } } + attachment_last_pass[i] = -1; + attachment_remap.push_back(attachments.size()); attachments.push_back(description); + } - VkAttachmentReference reference; - reference.attachment = i; + LocalVector<VkSubpassDescription> subpasses; + LocalVector<LocalVector<VkAttachmentReference>> color_reference_array; + LocalVector<LocalVector<VkAttachmentReference>> input_reference_array; + LocalVector<LocalVector<VkAttachmentReference>> resolve_reference_array; + LocalVector<LocalVector<uint32_t>> preserve_reference_array; + LocalVector<VkAttachmentReference> depth_reference_array; - if (p_format[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { - reference.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; + subpasses.resize(p_passes.size()); + color_reference_array.resize(p_passes.size()); + input_reference_array.resize(p_passes.size()); + resolve_reference_array.resize(p_passes.size()); + preserve_reference_array.resize(p_passes.size()); + depth_reference_array.resize(p_passes.size()); + + LocalVector<VkSubpassDependency> subpass_dependencies; + + for (int i = 0; i < p_passes.size(); i++) { + const FramebufferPass *pass = &p_passes[i]; + + LocalVector<VkAttachmentReference> &color_references = color_reference_array[i]; + + TextureSamples texture_samples = TEXTURE_SAMPLES_1; + bool is_multisample_first = true; + + for (int j = 0; j < pass->color_attachments.size(); j++) { + int32_t attachment = pass->color_attachments[j]; + VkAttachmentReference reference; + if (attachment == FramebufferPass::ATTACHMENT_UNUSED) { + reference.attachment = VK_ATTACHMENT_UNUSED; + reference.layout = VK_IMAGE_LAYOUT_UNDEFINED; + } else { + ERR_FAIL_INDEX_V_MSG(attachment, p_attachments.size(), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), color attachment (" + itos(j) + ")."); + ERR_FAIL_COND_V_MSG(!(p_attachments[attachment].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it's marked as depth, but it's not usable as color attachment."); + ERR_FAIL_COND_V_MSG(attachment_last_pass[attachment] == i, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it already was used for something else before in this pass."); + + if (is_multisample_first) { + texture_samples = p_attachments[attachment].samples; + is_multisample_first = false; + } else { + ERR_FAIL_COND_V_MSG(texture_samples != p_attachments[attachment].samples, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), if an attachment is marked as multisample, all of them should be multisample and use the same number of samples."); + } + reference.attachment = attachment_remap[attachment]; + reference.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; + attachment_last_pass[attachment] = i; + } 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; + } + + LocalVector<VkAttachmentReference> &input_references = input_reference_array[i]; + + for (int j = 0; j < pass->input_attachments.size(); j++) { + int32_t attachment = pass->input_attachments[j]; + VkAttachmentReference reference; + if (attachment == FramebufferPass::ATTACHMENT_UNUSED) { + reference.attachment = VK_ATTACHMENT_UNUSED; + reference.layout = VK_IMAGE_LAYOUT_UNDEFINED; + } else { + ERR_FAIL_INDEX_V_MSG(attachment, p_attachments.size(), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), input attachment (" + itos(j) + ")."); + ERR_FAIL_COND_V_MSG(!(p_attachments[attachment].usage_flags & TEXTURE_USAGE_INPUT_ATTACHMENT_BIT), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it isn't marked as an input texture."); + ERR_FAIL_COND_V_MSG(attachment_last_pass[attachment] == i, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it already was used for something else before in this pass."); + reference.attachment = attachment_remap[attachment]; + reference.layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; + attachment_last_pass[attachment] = i; + } + input_references.push_back(reference); + } + + LocalVector<VkAttachmentReference> &resolve_references = resolve_reference_array[i]; + + if (pass->resolve_attachments.size() > 0) { + ERR_FAIL_COND_V_MSG(pass->resolve_attachments.size() != pass->color_attachments.size(), VK_NULL_HANDLE, "The amount of resolve attachments (" + itos(pass->resolve_attachments.size()) + ") must match the number of color attachments (" + itos(pass->color_attachments.size()) + ")."); + ERR_FAIL_COND_V_MSG(texture_samples == TEXTURE_SAMPLES_1, VK_NULL_HANDLE, "Resolve attachments specified, but color attachments are not multisample."); + } + for (int j = 0; j < pass->resolve_attachments.size(); j++) { + int32_t attachment = pass->resolve_attachments[j]; + VkAttachmentReference reference; + if (attachment == FramebufferPass::ATTACHMENT_UNUSED) { + reference.attachment = VK_ATTACHMENT_UNUSED; + reference.layout = VK_IMAGE_LAYOUT_UNDEFINED; + } else { + ERR_FAIL_INDEX_V_MSG(attachment, p_attachments.size(), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), resolve attachment (" + itos(j) + ")."); + ERR_FAIL_COND_V_MSG(pass->color_attachments[j] == FramebufferPass::ATTACHMENT_UNUSED, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), resolve attachment (" + itos(j) + "), the respective color attachment is marked as unused."); + ERR_FAIL_COND_V_MSG(!(p_attachments[attachment].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), resolve attachment, it isn't marked as a color texture."); + ERR_FAIL_COND_V_MSG(attachment_last_pass[attachment] == i, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it already was used for something else before in this pass."); + bool multisample = p_attachments[attachment].samples > TEXTURE_SAMPLES_1; + ERR_FAIL_COND_V_MSG(multisample, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), resolve attachments can't be multisample."); + reference.attachment = attachment_remap[attachment]; + reference.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; // VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; + attachment_last_pass[attachment] = i; + } 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; + } + + VkAttachmentReference &depth_stencil_reference = depth_reference_array[i]; + + if (pass->depth_attachment != FramebufferPass::ATTACHMENT_UNUSED) { + int32_t attachment = pass->depth_attachment; + ERR_FAIL_INDEX_V_MSG(attachment, p_attachments.size(), VK_NULL_HANDLE, "Invalid framebuffer depth format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), depth attachment."); + ERR_FAIL_COND_V_MSG(!(p_attachments[attachment].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT), VK_NULL_HANDLE, "Invalid framebuffer depth format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it's marked as depth, but it's not a depth attachment."); + ERR_FAIL_COND_V_MSG(attachment_last_pass[attachment] == i, VK_NULL_HANDLE, "Invalid framebuffer depth format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it already was used for something else before in this pass."); + depth_stencil_reference.attachment = attachment_remap[attachment]; + depth_stencil_reference.layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; + attachment_last_pass[attachment] = i; + + if (is_multisample_first) { + texture_samples = p_attachments[attachment].samples; + is_multisample_first = false; + } else { + ERR_FAIL_COND_V_MSG(texture_samples != p_attachments[attachment].samples, VK_NULL_HANDLE, "Invalid framebuffer depth format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), if an attachment is marked as multisample, all of them should be multisample and use the same number of samples including the depth."); + } + } else { - 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."); + depth_stencil_reference.attachment = VK_ATTACHMENT_UNUSED; + depth_stencil_reference.layout = VK_IMAGE_LAYOUT_UNDEFINED; + } + + LocalVector<uint32_t> &preserve_references = preserve_reference_array[i]; + + for (int j = 0; j < pass->preserve_attachments.size(); j++) { + int32_t attachment = pass->preserve_attachments[j]; + + ERR_FAIL_COND_V_MSG(attachment == FramebufferPass::ATTACHMENT_UNUSED, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), preserve attachment (" + itos(j) + "). Preserve attachments can't be unused."); + + ERR_FAIL_INDEX_V_MSG(attachment, p_attachments.size(), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), preserve attachment (" + itos(j) + ")."); + + if (attachment_last_pass[attachment] != i) { + //preserve can still be used to keep depth or color from being discarded after use + attachment_last_pass[attachment] = i; + preserve_references.push_back(attachment); + } } + VkSubpassDescription &subpass = subpasses[i]; + subpass.flags = 0; + subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS; + subpass.inputAttachmentCount = input_references.size(); + if (input_references.size()) { + subpass.pInputAttachments = input_references.ptr(); + } else { + subpass.pInputAttachments = nullptr; + } + subpass.colorAttachmentCount = color_references.size(); + if (color_references.size()) { + subpass.pColorAttachments = color_references.ptr(); + } else { + subpass.pColorAttachments = nullptr; + } + if (depth_stencil_reference.attachment != VK_ATTACHMENT_UNUSED) { + subpass.pDepthStencilAttachment = &depth_stencil_reference; + } else { + subpass.pDepthStencilAttachment = nullptr; + } + + if (resolve_references.size()) { + subpass.pResolveAttachments = resolve_references.ptr(); + } else { + subpass.pResolveAttachments = nullptr; + } + + subpass.preserveAttachmentCount = preserve_references.size(); + if (preserve_references.size()) { + subpass.pPreserveAttachments = preserve_references.ptr(); + } else { + subpass.pPreserveAttachments = nullptr; + } + + if (r_samples) { + r_samples->push_back(texture_samples); + } + + if (i > 0) { + VkSubpassDependency dependency; + dependency.srcSubpass = i - 1; + dependency.dstSubpass = i; + dependency.srcStageMask = 0; + dependency.srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT; + dependency.dstStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; + + dependency.srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; + dependency.dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_INPUT_ATTACHMENT_READ_BIT; + dependency.dependencyFlags = VK_DEPENDENCY_BY_REGION_BIT; + subpass_dependencies.push_back(dependency); + } + /* // NOTE: Big Mallet Approach -- any layout transition causes a full barrier if (reference.layout != description.initialLayout) { // NOTE: this should be smarter based on the texture's knowledge of its previous role @@ -3433,66 +3772,98 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF 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; + render_pass_create_info.subpassCount = subpasses.size(); + render_pass_create_info.pSubpasses = subpasses.ptr(); // Commenting this because it seems it just avoids raster and compute to work at the same time. // Other barriers seem to be protecting the render pass fine. // render_pass_create_info.dependencyCount = 2; // render_pass_create_info.pDependencies = dependencies; - render_pass_create_info.dependencyCount = 0; - render_pass_create_info.pDependencies = nullptr; + render_pass_create_info.dependencyCount = subpass_dependencies.size(); + if (subpass_dependencies.size()) { + render_pass_create_info.pDependencies = subpass_dependencies.ptr(); + } else { + render_pass_create_info.pDependencies = nullptr; + } + + // These are only used if we use multiview but we need to define them in scope. + const uint32_t view_mask = (1 << p_view_count) - 1; + const uint32_t correlation_mask = (1 << p_view_count) - 1; + Vector<uint32_t> view_masks; + VkRenderPassMultiviewCreateInfo render_pass_multiview_create_info; + + if (p_view_count > 1) { + const VulkanContext::MultiviewCapabilities capabilities = context->get_multiview_capabilities(); + + // For now this only works with multiview! + ERR_FAIL_COND_V_MSG(!capabilities.is_supported, VK_NULL_HANDLE, "Multiview not supported"); + + // Make sure we limit this to the number of views we support. + ERR_FAIL_COND_V_MSG(p_view_count > capabilities.max_view_count, VK_NULL_HANDLE, "Hardware does not support requested number of views for Multiview render pass"); + + // Set view masks for each subpass + for (uint32_t i = 0; i < subpasses.size(); i++) { + view_masks.push_back(view_mask); + } + + render_pass_multiview_create_info.sType = VK_STRUCTURE_TYPE_RENDER_PASS_MULTIVIEW_CREATE_INFO; + render_pass_multiview_create_info.pNext = nullptr; + render_pass_multiview_create_info.subpassCount = subpasses.size(); + render_pass_multiview_create_info.pViewMasks = view_masks.ptr(); + render_pass_multiview_create_info.dependencyCount = 0; + render_pass_multiview_create_info.pViewOffsets = nullptr; + render_pass_multiview_create_info.correlationMaskCount = 1; + render_pass_multiview_create_info.pCorrelationMasks = &correlation_mask; + + render_pass_create_info.pNext = &render_pass_multiview_create_info; + } VkRenderPass render_pass; VkResult res = 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<AttachmentFormat> &p_format) { +RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_create(const Vector<AttachmentFormat> &p_format, uint32_t p_view_count) { + FramebufferPass pass; + for (int i = 0; i < p_format.size(); i++) { + if (p_format[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { + pass.depth_attachment = i; + } else { + pass.color_attachments.push_back(i); + } + } + + Vector<FramebufferPass> passes; + passes.push_back(pass); + return framebuffer_format_create_multipass(p_format, passes, p_view_count); +} +RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_create_multipass(const Vector<AttachmentFormat> &p_attachments, Vector<FramebufferPass> &p_passes, uint32_t p_view_count) { _THREAD_SAFE_METHOD_ FramebufferFormatKey key; - key.attachments = p_format; + key.attachments = p_attachments; + key.passes = p_passes; + key.view_count = p_view_count; - const Map<FramebufferFormatKey, FramebufferFormatID>::Element *E = framebuffer_format_cache.find(key); + const RBMap<FramebufferFormatKey, FramebufferFormatID>::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 + Vector<TextureSamples> samples; + VkRenderPass render_pass = _render_pass_create(p_attachments, p_passes, INITIAL_ACTION_CLEAR, FINAL_ACTION_READ, INITIAL_ACTION_CLEAR, FINAL_ACTION_READ, p_view_count, &samples); //actions don't matter for this use case if (render_pass == VK_NULL_HANDLE) { //was likely invalid return INVALID_ID; @@ -3502,17 +3873,18 @@ RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_c 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; + fb_format.pass_samples = samples; + fb_format.view_count = p_view_count; framebuffer_formats[id] = fb_format; return id; } RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_create_empty(TextureSamples p_samples) { FramebufferFormatKey key; + key.passes.push_back(FramebufferPass()); - const Map<FramebufferFormatKey, FramebufferFormatID>::Element *E = framebuffer_format_cache.find(key); + const RBMap<FramebufferFormatKey, FramebufferFormatID>::Element *E = framebuffer_format_cache.find(key); if (E) { //exists, return return E->get(); @@ -3544,7 +3916,7 @@ RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_c 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) + "."); + ERR_FAIL_COND_V_MSG(res, 0, "vkCreateRenderPass for empty fb failed with error " + itos(res) + "."); if (render_pass == VK_NULL_HANDLE) { //was likely invalid return INVALID_ID; @@ -3556,18 +3928,18 @@ RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_c FramebufferFormat fb_format; fb_format.E = E; - fb_format.color_attachments = 0; fb_format.render_pass = render_pass; - fb_format.samples = p_samples; + fb_format.pass_samples.push_back(p_samples); framebuffer_formats[id] = fb_format; return id; } -RenderingDevice::TextureSamples RenderingDeviceVulkan::framebuffer_format_get_texture_samples(FramebufferFormatID p_format) { - Map<FramebufferFormatID, FramebufferFormat>::Element *E = framebuffer_formats.find(p_format); +RenderingDevice::TextureSamples RenderingDeviceVulkan::framebuffer_format_get_texture_samples(FramebufferFormatID p_format, uint32_t p_pass) { + HashMap<FramebufferFormatID, FramebufferFormat>::Iterator E = framebuffer_formats.find(p_format); ERR_FAIL_COND_V(!E, TEXTURE_SAMPLES_1); + ERR_FAIL_COND_V(p_pass >= uint32_t(E->value.pass_samples.size()), TEXTURE_SAMPLES_1); - return E->get().samples; + return E->value.pass_samples[p_pass]; } /***********************/ @@ -3580,36 +3952,68 @@ RID RenderingDeviceVulkan::framebuffer_create_empty(const Size2i &p_size, Textur framebuffer.format_id = framebuffer_format_create_empty(p_samples); ERR_FAIL_COND_V(p_format_check != INVALID_FORMAT_ID && framebuffer.format_id != p_format_check, RID()); framebuffer.size = p_size; + framebuffer.view_count = 1; return framebuffer_owner.make_rid(framebuffer); } -RID RenderingDeviceVulkan::framebuffer_create(const Vector<RID> &p_texture_attachments, FramebufferFormatID p_format_check) { +RID RenderingDeviceVulkan::framebuffer_create(const Vector<RID> &p_texture_attachments, FramebufferFormatID p_format_check, uint32_t p_view_count) { _THREAD_SAFE_METHOD_ - Vector<AttachmentFormat> attachments; - Size2i size; + FramebufferPass pass; 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."); + Texture *texture = texture_owner.get_or_null(p_texture_attachments[i]); + + ERR_FAIL_COND_V_MSG(texture && texture->layers != p_view_count, RID(), "Layers of our texture doesn't match view count for this framebuffer"); - if (i == 0) { - size.width = texture->width; - size.height = texture->height; + if (texture && texture->usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { + pass.depth_attachment = i; } 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."); + pass.color_attachments.push_back(texture ? i : FramebufferPass::ATTACHMENT_UNUSED); } + } + + Vector<FramebufferPass> passes; + passes.push_back(pass); + return framebuffer_create_multipass(p_texture_attachments, passes, p_format_check, p_view_count); +} + +RID RenderingDeviceVulkan::framebuffer_create_multipass(const Vector<RID> &p_texture_attachments, Vector<FramebufferPass> &p_passes, FramebufferFormatID p_format_check, uint32_t p_view_count) { + _THREAD_SAFE_METHOD_ + + Vector<AttachmentFormat> attachments; + attachments.resize(p_texture_attachments.size()); + Size2i size; + bool size_set = false; + for (int i = 0; i < p_texture_attachments.size(); i++) { AttachmentFormat af; - af.format = texture->format; - af.samples = texture->samples; - af.usage_flags = texture->usage_flags; - attachments.push_back(af); + Texture *texture = texture_owner.get_or_null(p_texture_attachments[i]); + if (!texture) { + af.usage_flags = AttachmentFormat::UNUSED_ATTACHMENT; + } else { + ERR_FAIL_COND_V_MSG(texture->layers != p_view_count, RID(), "Layers of our texture doesn't match view count for this framebuffer"); + + if (!size_set) { + size.width = texture->width; + size.height = texture->height; + size_set = true; + } else { + ERR_FAIL_COND_V_MSG((uint32_t)size.width != texture->width || (uint32_t)size.height != texture->height, RID(), + "All textures in a framebuffer should be the same size."); + } + + af.format = texture->format; + af.samples = texture->samples; + af.usage_flags = texture->usage_flags; + } + attachments.write[i] = af; } - FramebufferFormatID format_id = framebuffer_format_create(attachments); + ERR_FAIL_COND_V_MSG(!size_set, RID(), "All attachments unused."); + + FramebufferFormatID format_id = framebuffer_format_create_multipass(attachments, p_passes, p_view_count); if (format_id == INVALID_ID) { return RID(); } @@ -3621,11 +4025,14 @@ RID RenderingDeviceVulkan::framebuffer_create(const Vector<RID> &p_texture_attac framebuffer.format_id = format_id; framebuffer.texture_ids = p_texture_attachments; framebuffer.size = size; + framebuffer.view_count = p_view_count; RID id = framebuffer_owner.make_rid(framebuffer); for (int i = 0; i < p_texture_attachments.size(); i++) { - _add_dependency(id, p_texture_attachments[i]); + if (p_texture_attachments[i].is_valid()) { + _add_dependency(id, p_texture_attachments[i]); + } } return id; @@ -3634,7 +4041,7 @@ RID RenderingDeviceVulkan::framebuffer_create(const Vector<RID> &p_texture_attac RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_get_format(RID p_framebuffer) { _THREAD_SAFE_METHOD_ - Framebuffer *framebuffer = framebuffer_owner.getornull(p_framebuffer); + Framebuffer *framebuffer = framebuffer_owner.get_or_null(p_framebuffer); ERR_FAIL_COND_V(!framebuffer, INVALID_ID); return framebuffer->format_id; @@ -3703,7 +4110,7 @@ RID RenderingDeviceVulkan::vertex_buffer_create(uint32_t p_size_bytes, const Vec usage |= VK_BUFFER_USAGE_STORAGE_BUFFER_BIT; } Buffer buffer; - _buffer_allocate(&buffer, p_size_bytes, usage, VMA_MEMORY_USAGE_GPU_ONLY); + _buffer_allocate(&buffer, p_size_bytes, usage, VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE, 0); if (p_data.size()) { uint64_t data_size = p_data.size(); const uint8_t *r = p_data.ptr(); @@ -3731,7 +4138,7 @@ RenderingDevice::VertexFormatID RenderingDeviceVulkan::vertex_format_create(cons vdcache.bindings = memnew_arr(VkVertexInputBindingDescription, p_vertex_formats.size()); vdcache.attributes = memnew_arr(VkVertexInputAttributeDescription, p_vertex_formats.size()); - Set<int> used_locations; + HashSet<int> used_locations; for (int i = 0; i < p_vertex_formats.size(); i++) { ERR_CONTINUE(p_vertex_formats[i].format >= DATA_FORMAT_MAX); ERR_FAIL_COND_V(used_locations.has(p_vertex_formats[i].location), INVALID_ID); @@ -3784,7 +4191,7 @@ RID RenderingDeviceVulkan::vertex_array_create(uint32_t p_vertex_count, VertexFo 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]); + Buffer *buffer = vertex_buffer_owner.get_or_null(p_src_buffers[i]); //validate with buffer { @@ -3865,7 +4272,7 @@ RID RenderingDeviceVulkan::index_buffer_create(uint32_t p_index_count, IndexBuff #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); + _buffer_allocate(&index_buffer, size_bytes, VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT, VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE, 0); if (p_data.size()) { uint64_t data_size = p_data.size(); const uint8_t *r = p_data.ptr(); @@ -3880,7 +4287,7 @@ RID RenderingDeviceVulkan::index_array_create(RID p_index_buffer, uint32_t p_ind ERR_FAIL_COND_V(!index_buffer_owner.owns(p_index_buffer), RID()); - IndexBuffer *index_buffer = index_buffer_owner.getornull(p_index_buffer); + IndexBuffer *index_buffer = index_buffer_owner.get_or_null(p_index_buffer); ERR_FAIL_COND_V(p_index_count == 0, RID()); ERR_FAIL_COND_V(p_index_offset + p_index_count > index_buffer->index_count, RID()); @@ -3924,7 +4331,7 @@ static VkShaderStageFlagBits shader_stage_masks[RenderingDevice::SHADER_STAGE_MA String RenderingDeviceVulkan::_shader_uniform_debug(RID p_shader, int p_set) { String ret; - const Shader *shader = shader_owner.getornull(p_shader); + const Shader *shader = shader_owner.get_or_null(p_shader); ERR_FAIL_COND_V(!shader, String()); for (int i = 0; i < shader->sets.size(); i++) { if (p_set >= 0 && i != p_set) { @@ -3932,7 +4339,7 @@ String RenderingDeviceVulkan::_shader_uniform_debug(RID p_shader, int p_set) { } for (int j = 0; j < shader->sets[i].uniform_info.size(); j++) { const UniformInfo &ui = shader->sets[i].uniform_info[j]; - if (ret != String()) { + if (!ret.is_empty()) { ret += "\n"; } ret += "Set: " + itos(i) + " Binding: " + itos(ui.binding) + " Type: " + shader_uniform_names[ui.type] + " Length: " + itos(ui.length); @@ -4133,51 +4540,91 @@ bool RenderingDeviceVulkan::_uniform_add_binding(Vector<Vector<VkDescriptorSetLa } #endif -RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages) { - //descriptor layouts - Vector<Vector<VkDescriptorSetLayoutBinding>> set_bindings; - Vector<Vector<UniformInfo>> uniform_info; - Shader::PushConstant push_constant; - push_constant.push_constant_size = 0; - push_constant.push_constants_vk_stage = 0; +//version 1: initial +//version 2: Added shader name - uint32_t vertex_input_mask = 0; +#define SHADER_BINARY_VERSION 2 - uint32_t fragment_outputs = 0; +String RenderingDeviceVulkan::shader_get_binary_cache_key() const { + return "Vulkan-SV" + itos(SHADER_BINARY_VERSION); +} - uint32_t stages_processed = 0; +struct RenderingDeviceVulkanShaderBinaryDataBinding { + uint32_t type; + uint32_t binding; + uint32_t stages; + uint32_t length; //size of arrays (in total elements), or ubos (in bytes * total elements) +}; + +struct RenderingDeviceVulkanShaderBinarySpecializationConstant { + uint32_t type; + uint32_t constant_id; + union { + uint32_t int_value; + float float_value; + bool bool_value; + }; + uint32_t stage_flags; +}; - bool is_compute = false; +struct RenderingDeviceVulkanShaderBinaryData { + uint32_t vertex_input_mask; + uint32_t fragment_outputs; + uint32_t specialization_constant_count; + uint32_t is_compute; + uint32_t compute_local_size[3]; + uint32_t set_count; + uint32_t push_constant_size; + uint32_t push_constants_vk_stage; + uint32_t stage_count; + uint32_t shader_name_len; +}; - uint32_t compute_local_size[3] = { 0, 0, 0 }; +Vector<uint8_t> RenderingDeviceVulkan::shader_compile_binary_from_spirv(const Vector<ShaderStageSPIRVData> &p_spirv, const String &p_shader_name) { + RenderingDeviceVulkanShaderBinaryData binary_data; + binary_data.vertex_input_mask = 0; + binary_data.fragment_outputs = 0; + binary_data.specialization_constant_count = 0; + binary_data.is_compute = 0; + binary_data.compute_local_size[0] = 0; + binary_data.compute_local_size[1] = 0; + binary_data.compute_local_size[2] = 0; + binary_data.set_count = 0; + binary_data.push_constant_size = 0; + binary_data.push_constants_vk_stage = 0; + + Vector<Vector<RenderingDeviceVulkanShaderBinaryDataBinding>> uniform_info; //set bindings + Vector<RenderingDeviceVulkanShaderBinarySpecializationConstant> specialization_constants; - 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(), + uint32_t stages_processed = 0; + + for (int i = 0; i < p_spirv.size(); i++) { + if (p_spirv[i].shader_stage == SHADER_STAGE_COMPUTE) { + binary_data.is_compute = true; + ERR_FAIL_COND_V_MSG(p_spirv.size() != 1, Vector<uint8_t>(), "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."); + ERR_FAIL_COND_V_MSG(stages_processed & (1 << p_spirv[i].shader_stage), Vector<uint8_t>(), + "Stage " + String(shader_stage_names[p_spirv[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; + const uint8_t *spirv = p_spirv[i].spir_v.ptr(); + SpvReflectResult result = spvReflectCreateShaderModule(p_spirv[i].spir_v.size(), spirv, &module); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed parsing shader."); + + if (binary_data.is_compute) { + binary_data.compute_local_size[0] = module.entry_points->local_size.x; + binary_data.compute_local_size[1] = module.entry_points->local_size.y; + binary_data.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."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed enumerating descriptor bindings."); - uint32_t stage = p_stages[i].shader_stage; + uint32_t stage = p_spirv[i].shader_stage; if (binding_count > 0) { //Parse bindings @@ -4186,56 +4633,47 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages 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."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[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; + RenderingDeviceVulkanShaderBinaryDataBinding 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; @@ -4248,8 +4686,8 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages continue; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_INPUT_ATTACHMENT: { - layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT; info.type = UNIFORM_TYPE_INPUT_ATTACHMENT; + need_array_dimensions = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_ACCELERATION_STRUCTURE_KHR: { ERR_PRINT("Acceleration structure not supported."); @@ -4270,42 +4708,35 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages } } - 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(), + ERR_FAIL_COND_V_MSG(set >= MAX_UNIFORM_SETS, Vector<uint8_t>(), "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(), + ERR_FAIL_COND_V_MSG(set >= limits.maxBoundDescriptorSets, Vector<uint8_t>(), "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()) { + if (set < (uint32_t)uniform_info.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) { + for (int k = 0; k < uniform_info[set].size(); k++) { + if (uniform_info[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(), + ERR_FAIL_COND_V_MSG(uniform_info[set][k].type != info.type, Vector<uint8_t>(), "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(), + ERR_FAIL_COND_V_MSG(uniform_info[set][k].length != info.length, Vector<uint8_t>(), "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; } @@ -4316,40 +4747,90 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages } } - 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); + if (set >= (uint32_t)uniform_info.size()) { uniform_info.resize(set + 1); } - set_bindings.write[set].push_back(layout_binding); uniform_info.write[set].push_back(info); } } + { + //specialization constants + + uint32_t sc_count = 0; + result = spvReflectEnumerateSpecializationConstants(&module, &sc_count, nullptr); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed enumerating specialization constants."); + + if (sc_count) { + Vector<SpvReflectSpecializationConstant *> spec_constants; + spec_constants.resize(sc_count); + + result = spvReflectEnumerateSpecializationConstants(&module, &sc_count, spec_constants.ptrw()); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed obtaining specialization constants."); + + for (uint32_t j = 0; j < sc_count; j++) { + int32_t existing = -1; + RenderingDeviceVulkanShaderBinarySpecializationConstant sconst; + SpvReflectSpecializationConstant *spc = spec_constants[j]; + + sconst.constant_id = spc->constant_id; + sconst.int_value = 0.0; // clear previous value JIC + switch (spc->constant_type) { + case SPV_REFLECT_SPECIALIZATION_CONSTANT_BOOL: { + sconst.type = PIPELINE_SPECIALIZATION_CONSTANT_TYPE_BOOL; + sconst.bool_value = spc->default_value.int_bool_value != 0; + } break; + case SPV_REFLECT_SPECIALIZATION_CONSTANT_INT: { + sconst.type = PIPELINE_SPECIALIZATION_CONSTANT_TYPE_INT; + sconst.int_value = spc->default_value.int_bool_value; + } break; + case SPV_REFLECT_SPECIALIZATION_CONSTANT_FLOAT: { + sconst.type = PIPELINE_SPECIALIZATION_CONSTANT_TYPE_FLOAT; + sconst.float_value = spc->default_value.float_value; + } break; + } + sconst.stage_flags = 1 << p_spirv[i].shader_stage; + + for (int k = 0; k < specialization_constants.size(); k++) { + if (specialization_constants[k].constant_id == sconst.constant_id) { + ERR_FAIL_COND_V_MSG(specialization_constants[k].type != sconst.type, Vector<uint8_t>(), "More than one specialization constant used for id (" + itos(sconst.constant_id) + "), but their types differ."); + ERR_FAIL_COND_V_MSG(specialization_constants[k].int_value != sconst.int_value, Vector<uint8_t>(), "More than one specialization constant used for id (" + itos(sconst.constant_id) + "), but their default values differ."); + existing = k; + break; + } + } + + if (existing > 0) { + specialization_constants.write[existing].stage_flags |= sconst.stage_flags; + } else { + specialization_constants.push_back(sconst); + } + } + } + } + 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."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed enumerating input variables."); if (iv_count) { Vector<SpvReflectInterfaceVariable *> 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."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[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)); + binary_data.vertex_input_mask |= (1 << uint32_t(input_vars[j]->location)); } } } @@ -4358,20 +4839,21 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages 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."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed enumerating output variables."); if (ov_count) { Vector<SpvReflectInterfaceVariable *> 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."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[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); + const SpvReflectInterfaceVariable *refvar = output_vars[j]; + if (refvar != nullptr && refvar->built_in != SpvBuiltInFragDepth) { + binary_data.fragment_outputs |= 1 << refvar->location; } } } @@ -4379,31 +4861,30 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages 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."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[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."); + ERR_FAIL_COND_V_MSG(pc_count > 1, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "': Only one push constant is supported, which should be the same across shader stages."); Vector<SpvReflectBlockVariable *> 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."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed obtaining push constants."); #if 0 if (pconstants[0] == nullptr) { - FileAccess *f = FileAccess::open("res://popo.spv", FileAccess::WRITE); + Ref<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."); + ERR_FAIL_COND_V_MSG(binary_data.push_constant_size && binary_data.push_constant_size != pconstants[0]->size, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[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]; + binary_data.push_constant_size = pconstants[0]->size; + binary_data.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)); } @@ -4412,9 +4893,317 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages spvReflectDestroyShaderModule(&module); } - stages_processed |= (1 << p_stages[i].shader_stage); + stages_processed |= (1 << p_spirv[i].shader_stage); + } + + Vector<Vector<uint8_t>> compressed_stages; + Vector<uint32_t> smolv_size; + Vector<uint32_t> zstd_size; //if 0, stdno t used + + uint32_t stages_binary_size = 0; + + bool strip_debug = false; + + for (int i = 0; i < p_spirv.size(); i++) { + smolv::ByteArray smolv; + if (!smolv::Encode(p_spirv[i].spir_v.ptr(), p_spirv[i].spir_v.size(), smolv, strip_debug ? smolv::kEncodeFlagStripDebugInfo : 0)) { + ERR_FAIL_V_MSG(Vector<uint8_t>(), "Error compressing shader stage :" + String(shader_stage_names[p_spirv[i].shader_stage])); + } else { + smolv_size.push_back(smolv.size()); + { //zstd + Vector<uint8_t> zstd; + zstd.resize(Compression::get_max_compressed_buffer_size(smolv.size(), Compression::MODE_ZSTD)); + int dst_size = Compression::compress(zstd.ptrw(), &smolv[0], smolv.size(), Compression::MODE_ZSTD); + + if (dst_size > 0 && (uint32_t)dst_size < smolv.size()) { + zstd_size.push_back(dst_size); + zstd.resize(dst_size); + compressed_stages.push_back(zstd); + } else { + Vector<uint8_t> smv; + smv.resize(smolv.size()); + memcpy(smv.ptrw(), &smolv[0], smolv.size()); + zstd_size.push_back(0); //not using zstd + compressed_stages.push_back(smv); + } + } + } + uint32_t s = compressed_stages[i].size(); + if (s % 4 != 0) { + s += 4 - (s % 4); + } + stages_binary_size += s; + } + + binary_data.specialization_constant_count = specialization_constants.size(); + binary_data.set_count = uniform_info.size(); + binary_data.stage_count = p_spirv.size(); + + CharString shader_name_utf = p_shader_name.utf8(); + + binary_data.shader_name_len = shader_name_utf.length(); + + uint32_t total_size = sizeof(uint32_t) * 3; //header + version + main datasize; + total_size += sizeof(RenderingDeviceVulkanShaderBinaryData); + + total_size += binary_data.shader_name_len; + + if ((binary_data.shader_name_len % 4) != 0) { //alignment rules are really strange + total_size += 4 - (binary_data.shader_name_len % 4); + } + + for (int i = 0; i < uniform_info.size(); i++) { + total_size += sizeof(uint32_t); + total_size += uniform_info[i].size() * sizeof(RenderingDeviceVulkanShaderBinaryDataBinding); + } + + total_size += sizeof(RenderingDeviceVulkanShaderBinarySpecializationConstant) * specialization_constants.size(); + + total_size += compressed_stages.size() * sizeof(uint32_t) * 3; //sizes + total_size += stages_binary_size; + + Vector<uint8_t> ret; + ret.resize(total_size); + { + uint32_t offset = 0; + uint8_t *binptr = ret.ptrw(); + binptr[0] = 'G'; + binptr[1] = 'V'; + binptr[2] = 'B'; + binptr[3] = 'D'; //godot vulkan binary data + offset += 4; + encode_uint32(SHADER_BINARY_VERSION, binptr + offset); + offset += sizeof(uint32_t); + encode_uint32(sizeof(RenderingDeviceVulkanShaderBinaryData), binptr + offset); + offset += sizeof(uint32_t); + memcpy(binptr + offset, &binary_data, sizeof(RenderingDeviceVulkanShaderBinaryData)); + offset += sizeof(RenderingDeviceVulkanShaderBinaryData); + memcpy(binptr + offset, shader_name_utf.ptr(), binary_data.shader_name_len); + offset += binary_data.shader_name_len; + + if ((binary_data.shader_name_len % 4) != 0) { //alignment rules are really strange + offset += 4 - (binary_data.shader_name_len % 4); + } + + for (int i = 0; i < uniform_info.size(); i++) { + int count = uniform_info[i].size(); + encode_uint32(count, binptr + offset); + offset += sizeof(uint32_t); + if (count > 0) { + memcpy(binptr + offset, uniform_info[i].ptr(), sizeof(RenderingDeviceVulkanShaderBinaryDataBinding) * count); + offset += sizeof(RenderingDeviceVulkanShaderBinaryDataBinding) * count; + } + } + + if (specialization_constants.size()) { + memcpy(binptr + offset, specialization_constants.ptr(), sizeof(RenderingDeviceVulkanShaderBinarySpecializationConstant) * specialization_constants.size()); + offset += sizeof(RenderingDeviceVulkanShaderBinarySpecializationConstant) * specialization_constants.size(); + } + + for (int i = 0; i < compressed_stages.size(); i++) { + encode_uint32(p_spirv[i].shader_stage, binptr + offset); + offset += sizeof(uint32_t); + encode_uint32(smolv_size[i], binptr + offset); + offset += sizeof(uint32_t); + encode_uint32(zstd_size[i], binptr + offset); + offset += sizeof(uint32_t); + memcpy(binptr + offset, compressed_stages[i].ptr(), compressed_stages[i].size()); + + uint32_t s = compressed_stages[i].size(); + + if (s % 4 != 0) { + s += 4 - (s % 4); + } + + offset += s; + } + + ERR_FAIL_COND_V(offset != (uint32_t)ret.size(), Vector<uint8_t>()); + } + + return ret; +} + +RID RenderingDeviceVulkan::shader_create_from_bytecode(const Vector<uint8_t> &p_shader_binary) { + const uint8_t *binptr = p_shader_binary.ptr(); + uint32_t binsize = p_shader_binary.size(); + + uint32_t read_offset = 0; + //consistency check + ERR_FAIL_COND_V(binsize < sizeof(uint32_t) * 3 + sizeof(RenderingDeviceVulkanShaderBinaryData), RID()); + ERR_FAIL_COND_V(binptr[0] != 'G' || binptr[1] != 'V' || binptr[2] != 'B' || binptr[3] != 'D', RID()); + + uint32_t bin_version = decode_uint32(binptr + 4); + ERR_FAIL_COND_V(bin_version > SHADER_BINARY_VERSION, RID()); + + uint32_t bin_data_size = decode_uint32(binptr + 8); + + const RenderingDeviceVulkanShaderBinaryData &binary_data = *(reinterpret_cast<const RenderingDeviceVulkanShaderBinaryData *>(binptr + 12)); + + Shader::PushConstant push_constant; + push_constant.push_constant_size = binary_data.push_constant_size; + push_constant.push_constants_vk_stage = binary_data.push_constants_vk_stage; + + uint32_t vertex_input_mask = binary_data.vertex_input_mask; + + uint32_t fragment_outputs = binary_data.fragment_outputs; + + bool is_compute = binary_data.is_compute; + + const uint32_t compute_local_size[3] = { binary_data.compute_local_size[0], binary_data.compute_local_size[1], binary_data.compute_local_size[2] }; + + read_offset += sizeof(uint32_t) * 3 + bin_data_size; + + String name; + + if (binary_data.shader_name_len) { + name.parse_utf8((const char *)(binptr + read_offset), binary_data.shader_name_len); + read_offset += binary_data.shader_name_len; + if ((binary_data.shader_name_len % 4) != 0) { //alignment rules are really strange + read_offset += 4 - (binary_data.shader_name_len % 4); + } + } + + Vector<Vector<VkDescriptorSetLayoutBinding>> set_bindings; + Vector<Vector<UniformInfo>> uniform_info; + + set_bindings.resize(binary_data.set_count); + uniform_info.resize(binary_data.set_count); + + for (uint32_t i = 0; i < binary_data.set_count; i++) { + ERR_FAIL_COND_V(read_offset + sizeof(uint32_t) >= binsize, RID()); + uint32_t set_count = decode_uint32(binptr + read_offset); + read_offset += sizeof(uint32_t); + const RenderingDeviceVulkanShaderBinaryDataBinding *set_ptr = reinterpret_cast<const RenderingDeviceVulkanShaderBinaryDataBinding *>(binptr + read_offset); + uint32_t set_size = set_count * sizeof(RenderingDeviceVulkanShaderBinaryDataBinding); + ERR_FAIL_COND_V(read_offset + set_size >= binsize, RID()); + + for (uint32_t j = 0; j < set_count; j++) { + UniformInfo info; + info.type = UniformType(set_ptr[j].type); + info.length = set_ptr[j].length; + info.binding = set_ptr[j].binding; + info.stages = set_ptr[j].stages; + + VkDescriptorSetLayoutBinding layout_binding; + layout_binding.pImmutableSamplers = nullptr; + layout_binding.binding = set_ptr[j].binding; + layout_binding.descriptorCount = 1; + layout_binding.stageFlags = 0; + for (uint32_t k = 0; k < SHADER_STAGE_MAX; k++) { + if (set_ptr[j].stages & (1 << k)) { + layout_binding.stageFlags |= shader_stage_masks[k]; + } + } + + switch (info.type) { + case UNIFORM_TYPE_SAMPLER: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLER; + layout_binding.descriptorCount = set_ptr[j].length; + } break; + case UNIFORM_TYPE_SAMPLER_WITH_TEXTURE: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; + layout_binding.descriptorCount = set_ptr[j].length; + } break; + case UNIFORM_TYPE_TEXTURE: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE; + layout_binding.descriptorCount = set_ptr[j].length; + } break; + case UNIFORM_TYPE_IMAGE: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE; + layout_binding.descriptorCount = set_ptr[j].length; + } break; + case UNIFORM_TYPE_TEXTURE_BUFFER: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; + layout_binding.descriptorCount = set_ptr[j].length; + } break; + case UNIFORM_TYPE_IMAGE_BUFFER: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER; + } break; + case UNIFORM_TYPE_UNIFORM_BUFFER: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; + } break; + case UNIFORM_TYPE_STORAGE_BUFFER: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER; + } break; + case UNIFORM_TYPE_INPUT_ATTACHMENT: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT; + } break; + default: { + ERR_FAIL_V(RID()); + } + } + + set_bindings.write[i].push_back(layout_binding); + uniform_info.write[i].push_back(info); + } + + read_offset += set_size; + } + + ERR_FAIL_COND_V(read_offset + binary_data.specialization_constant_count * sizeof(RenderingDeviceVulkanShaderBinarySpecializationConstant) >= binsize, RID()); + + Vector<Shader::SpecializationConstant> specialization_constants; + + for (uint32_t i = 0; i < binary_data.specialization_constant_count; i++) { + const RenderingDeviceVulkanShaderBinarySpecializationConstant &src_sc = *(reinterpret_cast<const RenderingDeviceVulkanShaderBinarySpecializationConstant *>(binptr + read_offset)); + Shader::SpecializationConstant sc; + sc.constant.int_value = src_sc.int_value; + sc.constant.type = PipelineSpecializationConstantType(src_sc.type); + sc.constant.constant_id = src_sc.constant_id; + sc.stage_flags = src_sc.stage_flags; + specialization_constants.push_back(sc); + + read_offset += sizeof(RenderingDeviceVulkanShaderBinarySpecializationConstant); } + Vector<Vector<uint8_t>> stage_spirv_data; + Vector<ShaderStage> stage_type; + + for (uint32_t i = 0; i < binary_data.stage_count; i++) { + ERR_FAIL_COND_V(read_offset + sizeof(uint32_t) * 3 >= binsize, RID()); + uint32_t stage = decode_uint32(binptr + read_offset); + read_offset += sizeof(uint32_t); + uint32_t smolv_size = decode_uint32(binptr + read_offset); + read_offset += sizeof(uint32_t); + uint32_t zstd_size = decode_uint32(binptr + read_offset); + read_offset += sizeof(uint32_t); + + uint32_t buf_size = (zstd_size > 0) ? zstd_size : smolv_size; + + Vector<uint8_t> smolv; + const uint8_t *src_smolv = nullptr; + + if (zstd_size > 0) { + //decompress to smolv + smolv.resize(smolv_size); + int dec_smolv_size = Compression::decompress(smolv.ptrw(), smolv.size(), binptr + read_offset, zstd_size, Compression::MODE_ZSTD); + ERR_FAIL_COND_V(dec_smolv_size != (int32_t)smolv_size, RID()); + src_smolv = smolv.ptr(); + } else { + src_smolv = binptr + read_offset; + } + + Vector<uint8_t> spirv; + uint32_t spirv_size = smolv::GetDecodedBufferSize(src_smolv, smolv_size); + spirv.resize(spirv_size); + if (!smolv::Decode(src_smolv, smolv_size, spirv.ptrw(), spirv_size)) { + ERR_FAIL_V_MSG(RID(), "Malformed smolv input uncompressing shader stage:" + String(shader_stage_names[stage])); + } + stage_spirv_data.push_back(spirv); + stage_type.push_back(ShaderStage(stage)); + + if (buf_size % 4 != 0) { + buf_size += 4 - (buf_size % 4); + } + + ERR_FAIL_COND_V(read_offset + buf_size > binsize, RID()); + + read_offset += buf_size; + } + + ERR_FAIL_COND_V(read_offset != binsize, RID()); + //all good, let's create modules _THREAD_SAFE_METHOD_ @@ -4422,23 +5211,25 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages Shader shader; shader.vertex_input_mask = vertex_input_mask; - shader.fragment_outputs = fragment_outputs; + shader.fragment_output_mask = 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]; + shader.specialization_constants = specialization_constants; + shader.name = name; String error_text; bool success = true; - for (int i = 0; i < p_stages.size(); i++) { + for (int i = 0; i < stage_spirv_data.size(); i++) { VkShaderModuleCreateInfo shader_module_create_info; shader_module_create_info.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO; shader_module_create_info.pNext = nullptr; shader_module_create_info.flags = 0; - shader_module_create_info.codeSize = p_stages[i].spir_v.size(); - const uint8_t *r = p_stages[i].spir_v.ptr(); + shader_module_create_info.codeSize = stage_spirv_data[i].size(); + const uint8_t *r = stage_spirv_data[i].ptr(); shader_module_create_info.pCode = (const uint32_t *)r; @@ -4446,7 +5237,7 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages 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]); + error_text = "Error (" + itos(res) + ") creating shader module for stage: " + String(shader_stage_names[stage_type[i]]); break; } @@ -4462,7 +5253,7 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages 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.stage = shader_stage_bits[stage_type[i]]; shader_stage.module = module; shader_stage.pName = "main"; shader_stage.pSpecializationInfo = nullptr; @@ -4501,7 +5292,7 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages //has data, needs an actual format; UniformSetFormat usformat; usformat.uniform_info = set.uniform_info; - Map<UniformSetFormat, uint32_t>::Element *E = uniform_set_format_cache.find(usformat); + RBMap<UniformSetFormat, uint32_t>::Element *E = uniform_set_format_cache.find(usformat); if (E) { format = E->get(); } else { @@ -4573,7 +5364,7 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages uint32_t RenderingDeviceVulkan::shader_get_vertex_input_attribute_mask(RID p_shader) { _THREAD_SAFE_METHOD_ - const Shader *shader = shader_owner.getornull(p_shader); + const Shader *shader = shader_owner.get_or_null(p_shader); ERR_FAIL_COND_V(!shader, 0); return shader->vertex_input_mask; } @@ -4592,7 +5383,7 @@ RID RenderingDeviceVulkan::uniform_buffer_create(uint32_t p_size_bytes, const Ve "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); + Error err = _buffer_allocate(&buffer, p_size_bytes, VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE, 0); ERR_FAIL_COND_V(err != OK, RID()); if (p_data.size()) { uint64_t data_size = p_data.size(); @@ -4618,7 +5409,7 @@ RID RenderingDeviceVulkan::storage_buffer_create(uint32_t p_size_bytes, const Ve 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); + Error err = _buffer_allocate(&buffer, p_size_bytes, flags, VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE, 0); ERR_FAIL_COND_V(err != OK, RID()); if (p_data.size()) { @@ -4644,7 +5435,7 @@ RID RenderingDeviceVulkan::texture_buffer_create(uint32_t p_size_elements, DataF 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); + Error err = _buffer_allocate(&texture_buffer.buffer, size_bytes, VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT, VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE, 0); ERR_FAIL_COND_V(err != OK, RID()); if (p_data.size()) { @@ -4677,14 +5468,14 @@ RID RenderingDeviceVulkan::texture_buffer_create(uint32_t p_size_elements, DataF RenderingDeviceVulkan::DescriptorPool *RenderingDeviceVulkan::_descriptor_pool_allocate(const DescriptorPoolKey &p_key) { if (!descriptor_pools.has(p_key)) { - descriptor_pools[p_key] = Set<DescriptorPool *>(); + descriptor_pools[p_key] = HashSet<DescriptorPool *>(); } DescriptorPool *pool = nullptr; - for (Set<DescriptorPool *>::Element *E = descriptor_pools[p_key].front(); E; E = E->next()) { - if (E->get()->usage < max_descriptors_per_pool) { - pool = E->get(); + for (DescriptorPool *E : descriptor_pools[p_key]) { + if (E->usage < max_descriptors_per_pool) { + pool = E; break; } } @@ -4795,7 +5586,7 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, ERR_FAIL_COND_V(p_uniforms.size() == 0, RID()); - Shader *shader = shader_owner.getornull(p_shader); + Shader *shader = shader_owner.get_or_null(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(), @@ -4853,18 +5644,18 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, switch (uniform.uniform_type) { case UNIFORM_TYPE_SAMPLER: { - if (uniform.ids.size() != set_uniform.length) { + if (uniform.get_id_count() != (uint32_t)set_uniform.length) { if (set_uniform.length > 1) { - ERR_FAIL_V_MSG(RID(), "Sampler (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") sampler elements, so it should be provided equal number of sampler IDs to satisfy it (IDs provided: " + itos(uniform.ids.size()) + ")."); + ERR_FAIL_V_MSG(RID(), "Sampler (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") sampler elements, so it should be provided equal number of sampler IDs to satisfy it (IDs provided: " + itos(uniform.get_id_count()) + ")."); } else { - ERR_FAIL_V_MSG(RID(), "Sampler (binding: " + itos(uniform.binding) + ") should provide one ID referencing a sampler (IDs provided: " + itos(uniform.ids.size()) + ")."); + ERR_FAIL_V_MSG(RID(), "Sampler (binding: " + itos(uniform.binding) + ") should provide one ID referencing a sampler (IDs provided: " + itos(uniform.get_id_count()) + ")."); } } Vector<VkDescriptorImageInfo> image_info; - for (int j = 0; j < uniform.ids.size(); j++) { - VkSampler *sampler = sampler_owner.getornull(uniform.ids[j]); + for (uint32_t j = 0; j < uniform.get_id_count(); j++) { + VkSampler *sampler = sampler_owner.get_or_null(uniform.get_id(j)); ERR_FAIL_COND_V_MSG(!sampler, RID(), "Sampler (binding: " + itos(uniform.binding) + ", index " + itos(j) + ") is not a valid sampler."); VkDescriptorImageInfo img_info; @@ -4876,31 +5667,31 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, } write.dstArrayElement = 0; - write.descriptorCount = uniform.ids.size(); + write.descriptorCount = uniform.get_id_count(); write.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLER; write.pImageInfo = image_infos.push_back(image_info)->get().ptr(); write.pBufferInfo = nullptr; write.pTexelBufferView = nullptr; - type_size = uniform.ids.size(); + type_size = uniform.get_id_count(); } break; case UNIFORM_TYPE_SAMPLER_WITH_TEXTURE: { - if (uniform.ids.size() != set_uniform.length * 2) { + if (uniform.get_id_count() != (uint32_t)set_uniform.length * 2) { if (set_uniform.length > 1) { - ERR_FAIL_V_MSG(RID(), "SamplerTexture (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") sampler&texture elements, so it should provided twice the amount of IDs (sampler,texture pairs) to satisfy it (IDs provided: " + itos(uniform.ids.size()) + ")."); + ERR_FAIL_V_MSG(RID(), "SamplerTexture (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") sampler&texture elements, so it should provided twice the amount of IDs (sampler,texture pairs) to satisfy it (IDs provided: " + itos(uniform.get_id_count()) + ")."); } else { - ERR_FAIL_V_MSG(RID(), "SamplerTexture (binding: " + itos(uniform.binding) + ") should provide two IDs referencing a sampler and then a texture (IDs provided: " + itos(uniform.ids.size()) + ")."); + ERR_FAIL_V_MSG(RID(), "SamplerTexture (binding: " + itos(uniform.binding) + ") should provide two IDs referencing a sampler and then a texture (IDs provided: " + itos(uniform.get_id_count()) + ")."); } } Vector<VkDescriptorImageInfo> image_info; - for (int j = 0; j < uniform.ids.size(); j += 2) { - VkSampler *sampler = sampler_owner.getornull(uniform.ids[j + 0]); + for (uint32_t j = 0; j < uniform.get_id_count(); j += 2) { + VkSampler *sampler = sampler_owner.get_or_null(uniform.get_id(j + 0)); ERR_FAIL_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]); + Texture *texture = texture_owner.get_or_null(uniform.get_id(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(), @@ -4910,10 +5701,10 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, 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)) { + if (texture->usage_flags & (TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | TEXTURE_USAGE_INPUT_ATTACHMENT_BIT)) { UniformSet::AttachableTexture attachable_texture; attachable_texture.bind = set_uniform.binding; - attachable_texture.texture = texture->owner.is_valid() ? texture->owner : uniform.ids[j + 1]; + attachable_texture.texture = texture->owner.is_valid() ? texture->owner : uniform.get_id(j + 1); attachable_textures.push_back(attachable_texture); } @@ -4922,7 +5713,7 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, mutable_sampled_textures.push_back(texture); } if (texture->owner.is_valid()) { - texture = texture_owner.getornull(texture->owner); + texture = texture_owner.get_or_null(texture->owner); ERR_FAIL_COND_V(!texture, RID()); //bug, should never happen } @@ -4932,28 +5723,28 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, } write.dstArrayElement = 0; - write.descriptorCount = uniform.ids.size() / 2; + write.descriptorCount = uniform.get_id_count() / 2; write.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; write.pImageInfo = image_infos.push_back(image_info)->get().ptr(); write.pBufferInfo = nullptr; write.pTexelBufferView = nullptr; - type_size = uniform.ids.size() / 2; + type_size = uniform.get_id_count() / 2; } break; case UNIFORM_TYPE_TEXTURE: { - if (uniform.ids.size() != set_uniform.length) { + if (uniform.get_id_count() != (uint32_t)set_uniform.length) { if (set_uniform.length > 1) { - ERR_FAIL_V_MSG(RID(), "Texture (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") textures, so it should be provided equal number of texture IDs to satisfy it (IDs provided: " + itos(uniform.ids.size()) + ")."); + ERR_FAIL_V_MSG(RID(), "Texture (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") textures, so it should be provided equal number of texture IDs to satisfy it (IDs provided: " + itos(uniform.get_id_count()) + ")."); } else { - ERR_FAIL_V_MSG(RID(), "Texture (binding: " + itos(uniform.binding) + ") should provide one ID referencing a texture (IDs provided: " + itos(uniform.ids.size()) + ")."); + ERR_FAIL_V_MSG(RID(), "Texture (binding: " + itos(uniform.binding) + ") should provide one ID referencing a texture (IDs provided: " + itos(uniform.get_id_count()) + ")."); } } Vector<VkDescriptorImageInfo> image_info; - for (int j = 0; j < uniform.ids.size(); j++) { - Texture *texture = texture_owner.getornull(uniform.ids[j]); + for (uint32_t j = 0; j < uniform.get_id_count(); j++) { + Texture *texture = texture_owner.get_or_null(uniform.get_id(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(), @@ -4963,10 +5754,10 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, 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)) { + if (texture->usage_flags & (TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | TEXTURE_USAGE_INPUT_ATTACHMENT_BIT)) { UniformSet::AttachableTexture attachable_texture; attachable_texture.bind = set_uniform.binding; - attachable_texture.texture = texture->owner.is_valid() ? texture->owner : uniform.ids[j]; + attachable_texture.texture = texture->owner.is_valid() ? texture->owner : uniform.get_id(j); attachable_textures.push_back(attachable_texture); } @@ -4976,7 +5767,7 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, } if (texture->owner.is_valid()) { - texture = texture_owner.getornull(texture->owner); + texture = texture_owner.get_or_null(texture->owner); ERR_FAIL_COND_V(!texture, RID()); //bug, should never happen } @@ -4986,27 +5777,27 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, } write.dstArrayElement = 0; - write.descriptorCount = uniform.ids.size(); + write.descriptorCount = uniform.get_id_count(); write.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE; write.pImageInfo = image_infos.push_back(image_info)->get().ptr(); write.pBufferInfo = nullptr; write.pTexelBufferView = nullptr; - type_size = uniform.ids.size(); + type_size = uniform.get_id_count(); } break; case UNIFORM_TYPE_IMAGE: { - if (uniform.ids.size() != set_uniform.length) { + if (uniform.get_id_count() != (uint32_t)set_uniform.length) { if (set_uniform.length > 1) { - ERR_FAIL_V_MSG(RID(), "Image (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") textures, so it should be provided equal number of texture IDs to satisfy it (IDs provided: " + itos(uniform.ids.size()) + ")."); + ERR_FAIL_V_MSG(RID(), "Image (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") textures, so it should be provided equal number of texture IDs to satisfy it (IDs provided: " + itos(uniform.get_id_count()) + ")."); } else { - ERR_FAIL_V_MSG(RID(), "Image (binding: " + itos(uniform.binding) + ") should provide one ID referencing a texture (IDs provided: " + itos(uniform.ids.size()) + ")."); + ERR_FAIL_V_MSG(RID(), "Image (binding: " + itos(uniform.binding) + ") should provide one ID referencing a texture (IDs provided: " + itos(uniform.get_id_count()) + ")."); } } Vector<VkDescriptorImageInfo> image_info; - for (int j = 0; j < uniform.ids.size(); j++) { - Texture *texture = texture_owner.getornull(uniform.ids[j]); + for (uint32_t j = 0; j < uniform.get_id_count(); j++) { + Texture *texture = texture_owner.get_or_null(uniform.get_id(j)); ERR_FAIL_COND_V_MSG(!texture, RID(), "Image (binding: " + itos(uniform.binding) + ", index " + itos(j) + ") is not a valid texture."); @@ -5024,7 +5815,7 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, } if (texture->owner.is_valid()) { - texture = texture_owner.getornull(texture->owner); + texture = texture_owner.get_or_null(texture->owner); ERR_FAIL_COND_V(!texture, RID()); //bug, should never happen } @@ -5034,29 +5825,29 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, } write.dstArrayElement = 0; - write.descriptorCount = uniform.ids.size(); + write.descriptorCount = uniform.get_id_count(); write.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE; write.pImageInfo = image_infos.push_back(image_info)->get().ptr(); write.pBufferInfo = nullptr; write.pTexelBufferView = nullptr; - type_size = uniform.ids.size(); + type_size = uniform.get_id_count(); } break; case UNIFORM_TYPE_TEXTURE_BUFFER: { - if (uniform.ids.size() != set_uniform.length) { + if (uniform.get_id_count() != (uint32_t)set_uniform.length) { if (set_uniform.length > 1) { - ERR_FAIL_V_MSG(RID(), "Buffer (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") texture buffer elements, so it should be provided equal number of texture buffer IDs to satisfy it (IDs provided: " + itos(uniform.ids.size()) + ")."); + ERR_FAIL_V_MSG(RID(), "Buffer (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") texture buffer elements, so it should be provided equal number of texture buffer IDs to satisfy it (IDs provided: " + itos(uniform.get_id_count()) + ")."); } else { - ERR_FAIL_V_MSG(RID(), "Buffer (binding: " + itos(uniform.binding) + ") should provide one ID referencing a texture buffer (IDs provided: " + itos(uniform.ids.size()) + ")."); + ERR_FAIL_V_MSG(RID(), "Buffer (binding: " + itos(uniform.binding) + ") should provide one ID referencing a texture buffer (IDs provided: " + itos(uniform.get_id_count()) + ")."); } } Vector<VkDescriptorBufferInfo> buffer_info; Vector<VkBufferView> buffer_view; - for (int j = 0; j < uniform.ids.size(); j++) { - TextureBuffer *buffer = texture_buffer_owner.getornull(uniform.ids[j]); + for (uint32_t j = 0; j < uniform.get_id_count(); j++) { + TextureBuffer *buffer = texture_buffer_owner.get_or_null(uniform.get_id(j)); ERR_FAIL_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); @@ -5064,21 +5855,21 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, } write.dstArrayElement = 0; - write.descriptorCount = uniform.ids.size(); + write.descriptorCount = uniform.get_id_count(); write.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; write.pImageInfo = nullptr; write.pBufferInfo = buffer_infos.push_back(buffer_info)->get().ptr(); write.pTexelBufferView = buffer_views.push_back(buffer_view)->get().ptr(); - type_size = uniform.ids.size(); + type_size = uniform.get_id_count(); } break; case UNIFORM_TYPE_SAMPLER_WITH_TEXTURE_BUFFER: { - if (uniform.ids.size() != set_uniform.length * 2) { + if (uniform.get_id_count() != (uint32_t)set_uniform.length * 2) { if (set_uniform.length > 1) { - ERR_FAIL_V_MSG(RID(), "SamplerBuffer (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") sampler buffer elements, so it should provided twice the amount of IDs (sampler,buffer pairs) to satisfy it (IDs provided: " + itos(uniform.ids.size()) + ")."); + ERR_FAIL_V_MSG(RID(), "SamplerBuffer (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") sampler buffer elements, so it should provided twice the amount of IDs (sampler,buffer pairs) to satisfy it (IDs provided: " + itos(uniform.get_id_count()) + ")."); } else { - ERR_FAIL_V_MSG(RID(), "SamplerBuffer (binding: " + itos(uniform.binding) + ") should provide two IDs referencing a sampler and then a texture buffer (IDs provided: " + itos(uniform.ids.size()) + ")."); + ERR_FAIL_V_MSG(RID(), "SamplerBuffer (binding: " + itos(uniform.binding) + ") should provide two IDs referencing a sampler and then a texture buffer (IDs provided: " + itos(uniform.get_id_count()) + ")."); } } @@ -5086,11 +5877,11 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, Vector<VkDescriptorBufferInfo> buffer_info; Vector<VkBufferView> buffer_view; - for (int j = 0; j < uniform.ids.size(); j += 2) { - VkSampler *sampler = sampler_owner.getornull(uniform.ids[j + 0]); + for (uint32_t j = 0; j < uniform.get_id_count(); j += 2) { + VkSampler *sampler = sampler_owner.get_or_null(uniform.get_id(j + 0)); ERR_FAIL_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]); + TextureBuffer *buffer = texture_buffer_owner.get_or_null(uniform.get_id(j + 1)); VkDescriptorImageInfo img_info; img_info.sampler = *sampler; @@ -5106,23 +5897,23 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, } write.dstArrayElement = 0; - write.descriptorCount = uniform.ids.size() / 2; + write.descriptorCount = uniform.get_id_count() / 2; write.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; write.pImageInfo = image_infos.push_back(image_info)->get().ptr(); write.pBufferInfo = buffer_infos.push_back(buffer_info)->get().ptr(); write.pTexelBufferView = buffer_views.push_back(buffer_view)->get().ptr(); - type_size = uniform.ids.size() / 2; + type_size = uniform.get_id_count() / 2; } break; case UNIFORM_TYPE_IMAGE_BUFFER: { //todo } break; case UNIFORM_TYPE_UNIFORM_BUFFER: { - ERR_FAIL_COND_V_MSG(uniform.ids.size() != 1, RID(), - "Uniform buffer supplied (binding: " + itos(uniform.binding) + ") must provide one ID (" + itos(uniform.ids.size()) + " provided)."); + ERR_FAIL_COND_V_MSG(uniform.get_id_count() != 1, RID(), + "Uniform buffer supplied (binding: " + itos(uniform.binding) + ") must provide one ID (" + itos(uniform.get_id_count()) + " provided)."); - Buffer *buffer = uniform_buffer_owner.getornull(uniform.ids[0]); + Buffer *buffer = uniform_buffer_owner.get_or_null(uniform.get_id(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(), @@ -5137,15 +5928,15 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, } 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)."); + ERR_FAIL_COND_V_MSG(uniform.get_id_count() != 1, RID(), + "Storage buffer supplied (binding: " + itos(uniform.binding) + ") must provide one ID (" + itos(uniform.get_id_count()) + " provided)."); Buffer *buffer = nullptr; - if (storage_buffer_owner.owns(uniform.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]); + if (storage_buffer_owner.owns(uniform.get_id(0))) { + buffer = storage_buffer_owner.get_or_null(uniform.get_id(0)); + } else if (vertex_buffer_owner.owns(uniform.get_id(0))) { + buffer = vertex_buffer_owner.get_or_null(uniform.get_id(0)); ERR_FAIL_COND_V_MSG(!(buffer->usage & VK_BUFFER_USAGE_STORAGE_BUFFER_BIT), RID(), "Vertex buffer supplied (binding: " + itos(uniform.binding) + ") was not created with storage flag."); } @@ -5163,6 +5954,49 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, write.pTexelBufferView = nullptr; } break; case UNIFORM_TYPE_INPUT_ATTACHMENT: { + ERR_FAIL_COND_V_MSG(shader->is_compute, RID(), "InputAttachment (binding: " + itos(uniform.binding) + ") supplied for compute shader (this is not allowed)."); + + if (uniform.get_id_count() != (uint32_t)set_uniform.length) { + if (set_uniform.length > 1) { + ERR_FAIL_V_MSG(RID(), "InputAttachment (binding: " + itos(uniform.binding) + ") is an array of (" + itos(set_uniform.length) + ") textures, so it should be provided equal number of texture IDs to satisfy it (IDs provided: " + itos(uniform.get_id_count()) + ")."); + } else { + ERR_FAIL_V_MSG(RID(), "InputAttachment (binding: " + itos(uniform.binding) + ") should provide one ID referencing a texture (IDs provided: " + itos(uniform.get_id_count()) + ")."); + } + } + + Vector<VkDescriptorImageInfo> image_info; + + for (uint32_t j = 0; j < uniform.get_id_count(); j++) { + Texture *texture = texture_owner.get_or_null(uniform.get_id(j)); + + ERR_FAIL_COND_V_MSG(!texture, RID(), + "InputAttachment (binding: " + itos(uniform.binding) + ", index " + itos(j) + ") is not a valid texture."); + + ERR_FAIL_COND_V_MSG(!(texture->usage_flags & TEXTURE_USAGE_SAMPLING_BIT), RID(), + "InputAttachment (binding: " + itos(uniform.binding) + ", index " + itos(j) + ") needs the TEXTURE_USAGE_SAMPLING_BIT usage flag set in order to be used as uniform."); + + VkDescriptorImageInfo img_info; + img_info.sampler = VK_NULL_HANDLE; + img_info.imageView = texture->view; + + if (texture->owner.is_valid()) { + texture = texture_owner.get_or_null(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.get_id_count(); + write.descriptorType = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT; + write.pImageInfo = image_infos.push_back(image_info)->get().ptr(); + write.pBufferInfo = nullptr; + write.pTexelBufferView = nullptr; + + type_size = uniform.get_id_count(); } break; default: { } @@ -5212,10 +6046,9 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, _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(); + int id_count = uniform.get_id_count(); for (int j = 0; j < id_count; j++) { - _add_dependency(id, ids[j]); + _add_dependency(id, uniform.get_id(j)); } } @@ -5234,6 +6067,13 @@ bool RenderingDeviceVulkan::uniform_set_is_valid(RID p_uniform_set) { return uniform_set_owner.owns(p_uniform_set); } +void RenderingDeviceVulkan::uniform_set_set_invalidation_callback(RID p_uniform_set, UniformSetInvalidatedCallback p_callback, void *p_userdata) { + UniformSet *us = uniform_set_owner.get_or_null(p_uniform_set); + ERR_FAIL_COND(!us); + us->invalidated_callback = p_callback; + us->invalidated_callback_userdata = p_userdata; +} + 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_ @@ -5342,7 +6182,7 @@ Vector<uint8_t> RenderingDeviceVulkan::buffer_get_data(RID p_buffer) { 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); + _buffer_allocate(&tmp_buffer, buffer->size, VK_BUFFER_USAGE_TRANSFER_DST_BIT, VMA_MEMORY_USAGE_AUTO_PREFER_HOST, VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT); VkBufferCopy region; region.srcOffset = 0; region.dstOffset = 0; @@ -5359,7 +6199,7 @@ Vector<uint8_t> RenderingDeviceVulkan::buffer_get_data(RID p_buffer) { { buffer_data.resize(buffer->size); uint8_t *w = buffer_data.ptrw(); - copymem(w, buffer_mem, buffer->size); + memcpy(w, buffer_mem, buffer->size); } vmaUnmapMemory(allocator, tmp_buffer.allocation); @@ -5373,11 +6213,11 @@ Vector<uint8_t> RenderingDeviceVulkan::buffer_get_data(RID p_buffer) { /**** 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) { +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, uint32_t p_for_render_pass, const Vector<PipelineSpecializationConstant> &p_specialization_constants) { _THREAD_SAFE_METHOD_ //needs a shader - Shader *shader = shader_owner.getornull(p_shader); + Shader *shader = shader_owner.get_or_null(p_shader); ERR_FAIL_COND_V(!shader, RID()); ERR_FAIL_COND_V_MSG(shader->is_compute, RID(), @@ -5392,8 +6232,16 @@ RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferForma { //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."); + ERR_FAIL_COND_V_MSG(p_for_render_pass >= uint32_t(fb_format.E->key().passes.size()), RID(), "Render pass requested for pipeline creation (" + itos(p_for_render_pass) + ") is out of bounds"); + const FramebufferPass &pass = fb_format.E->key().passes[p_for_render_pass]; + uint32_t output_mask = 0; + for (int i = 0; i < pass.color_attachments.size(); i++) { + if (pass.color_attachments[i] != FramebufferPass::ATTACHMENT_UNUSED) { + output_mask |= 1 << i; + } + } + ERR_FAIL_COND_V_MSG(shader->fragment_output_mask != output_mask, RID(), + "Mismatch fragment shader output mask (" + itos(shader->fragment_output_mask) + ") and framebuffer color output mask (" + itos(output_mask) + ") when binding both in render pipeline."); } //vertex VkPipelineVertexInputStateCreateInfo pipeline_vertex_input_state_create_info; @@ -5407,7 +6255,7 @@ RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferForma //validate with inputs for (uint32_t i = 0; i < 32; i++) { - if (!(shader->vertex_input_mask & (1 << i))) { + if (!(shader->vertex_input_mask & (1UL << i))) { continue; } bool found = false; @@ -5465,7 +6313,7 @@ RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferForma 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()); + ERR_FAIL_COND_V(limits.maxTessellationPatchSize > 0 && (p_rasterization_state.patch_control_points < 1 || p_rasterization_state.patch_control_points > limits.maxTessellationPatchSize), RID()); tessellation_create_info.patchControlPoints = p_rasterization_state.patch_control_points; VkPipelineViewportStateCreateInfo viewport_state_create_info; @@ -5485,7 +6333,7 @@ RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferForma 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] = { + static const VkCullModeFlags cull_mode[3] = { VK_CULL_MODE_NONE, VK_CULL_MODE_FRONT_BIT, VK_CULL_MODE_BACK_BIT @@ -5512,7 +6360,7 @@ RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferForma Vector<VkSampleMask> sample_mask; if (p_multisample_state.sample_mask.size()) { //use sample mask - int rasterization_sample_mask_expected_size[TEXTURE_SAMPLES_MAX] = { + const int rasterization_sample_mask_expected_size[TEXTURE_SAMPLES_MAX] = { 1, 2, 4, 8, 16, 32, 64 }; ERR_FAIL_COND_V(rasterization_sample_mask_expected_size[p_multisample_state.sample_count] != p_multisample_state.sample_mask.size(), RID()); @@ -5578,44 +6426,59 @@ RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferForma 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<VkPipelineColorBlendAttachmentState> 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; + { + const FramebufferPass &pass = fb_format.E->key().passes[p_for_render_pass]; + attachment_states.resize(pass.color_attachments.size()); + ERR_FAIL_COND_V(p_blend_state.attachments.size() < pass.color_attachments.size(), RID()); + for (int i = 0; i < pass.color_attachments.size(); i++) { + VkPipelineColorBlendAttachmentState state; + if (pass.color_attachments[i] == FramebufferPass::ATTACHMENT_UNUSED) { + state.blendEnable = false; + + state.srcColorBlendFactor = VK_BLEND_FACTOR_ZERO; + state.dstColorBlendFactor = VK_BLEND_FACTOR_ZERO; + state.colorBlendOp = VK_BLEND_OP_ADD; + + state.srcAlphaBlendFactor = VK_BLEND_FACTOR_ZERO; + state.dstAlphaBlendFactor = VK_BLEND_FACTOR_ZERO; + state.alphaBlendOp = VK_BLEND_OP_ADD; + + state.colorWriteMask = 0; + } else { + state.blendEnable = p_blend_state.attachments[i].enable_blend; + + ERR_FAIL_INDEX_V(p_blend_state.attachments[i].src_color_blend_factor, BLEND_FACTOR_MAX, RID()); + state.srcColorBlendFactor = blend_factors[p_blend_state.attachments[i].src_color_blend_factor]; + ERR_FAIL_INDEX_V(p_blend_state.attachments[i].dst_color_blend_factor, BLEND_FACTOR_MAX, RID()); + state.dstColorBlendFactor = blend_factors[p_blend_state.attachments[i].dst_color_blend_factor]; + ERR_FAIL_INDEX_V(p_blend_state.attachments[i].color_blend_op, BLEND_OP_MAX, RID()); + state.colorBlendOp = blend_operations[p_blend_state.attachments[i].color_blend_op]; + + ERR_FAIL_INDEX_V(p_blend_state.attachments[i].src_alpha_blend_factor, BLEND_FACTOR_MAX, RID()); + state.srcAlphaBlendFactor = blend_factors[p_blend_state.attachments[i].src_alpha_blend_factor]; + ERR_FAIL_INDEX_V(p_blend_state.attachments[i].dst_alpha_blend_factor, BLEND_FACTOR_MAX, RID()); + state.dstAlphaBlendFactor = blend_factors[p_blend_state.attachments[i].dst_alpha_blend_factor]; + ERR_FAIL_INDEX_V(p_blend_state.attachments[i].alpha_blend_op, BLEND_OP_MAX, RID()); + state.alphaBlendOp = blend_operations[p_blend_state.attachments[i].alpha_blend_op]; + + state.colorWriteMask = 0; + if (p_blend_state.attachments[i].write_r) { + state.colorWriteMask |= VK_COLOR_COMPONENT_R_BIT; + } + if (p_blend_state.attachments[i].write_g) { + state.colorWriteMask |= VK_COLOR_COMPONENT_G_BIT; + } + if (p_blend_state.attachments[i].write_b) { + state.colorWriteMask |= VK_COLOR_COMPONENT_B_BIT; + } + if (p_blend_state.attachments[i].write_a) { + state.colorWriteMask |= VK_COLOR_COMPONENT_A_BIT; + } + } + attachment_states.write[i] = state; } - - attachment_states.push_back(state); - }; + } color_blend_state_create_info.attachmentCount = attachment_states.size(); color_blend_state_create_info.pAttachments = attachment_states.ptr(); @@ -5674,8 +6537,62 @@ RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferForma 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(); + Vector<VkPipelineShaderStageCreateInfo> pipeline_stages = shader->pipeline_stages; + Vector<VkSpecializationInfo> specialization_info; + Vector<Vector<VkSpecializationMapEntry>> specialization_map_entries; + Vector<uint32_t> specialization_constant_data; + + if (shader->specialization_constants.size()) { + specialization_constant_data.resize(shader->specialization_constants.size()); + uint32_t *data_ptr = specialization_constant_data.ptrw(); + specialization_info.resize(pipeline_stages.size()); + specialization_map_entries.resize(pipeline_stages.size()); + for (int i = 0; i < shader->specialization_constants.size(); i++) { + //see if overridden + const Shader::SpecializationConstant &sc = shader->specialization_constants[i]; + data_ptr[i] = sc.constant.int_value; //just copy the 32 bits + + for (int j = 0; j < p_specialization_constants.size(); j++) { + const PipelineSpecializationConstant &psc = p_specialization_constants[j]; + if (psc.constant_id == sc.constant.constant_id) { + ERR_FAIL_COND_V_MSG(psc.type != sc.constant.type, RID(), "Specialization constant provided for id (" + itos(sc.constant.constant_id) + ") is of the wrong type."); + data_ptr[i] = psc.int_value; + break; + } + } + + VkSpecializationMapEntry entry; + + entry.constantID = sc.constant.constant_id; + entry.offset = i * sizeof(uint32_t); + entry.size = sizeof(uint32_t); + + for (int j = 0; j < SHADER_STAGE_MAX; j++) { + if (sc.stage_flags & (1 << j)) { + VkShaderStageFlagBits stage = shader_stage_masks[j]; + for (int k = 0; k < pipeline_stages.size(); k++) { + if (pipeline_stages[k].stage == stage) { + specialization_map_entries.write[k].push_back(entry); + } + } + } + } + } + + for (int i = 0; i < pipeline_stages.size(); i++) { + if (specialization_map_entries[i].size()) { + specialization_info.write[i].dataSize = specialization_constant_data.size() * sizeof(uint32_t); + specialization_info.write[i].pData = data_ptr; + specialization_info.write[i].mapEntryCount = specialization_map_entries[i].size(); + specialization_info.write[i].pMapEntries = specialization_map_entries[i].ptr(); + pipeline_stages.write[i].pSpecializationInfo = specialization_info.ptr() + i; + } + } + } + + graphics_pipeline_create_info.stageCount = pipeline_stages.size(); + graphics_pipeline_create_info.pStages = pipeline_stages.ptr(); + graphics_pipeline_create_info.pVertexInputState = &pipeline_vertex_input_state_create_info; graphics_pipeline_create_info.pInputAssemblyState = &input_assembly_create_info; graphics_pipeline_create_info.pTessellationState = &tessellation_create_info; @@ -5688,13 +6605,13 @@ RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferForma 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.subpass = p_for_render_pass; graphics_pipeline_create_info.basePipelineHandle = VK_NULL_HANDLE; graphics_pipeline_create_info.basePipelineIndex = 0; RenderPipeline pipeline; VkResult err = vkCreateGraphicsPipelines(device, VK_NULL_HANDLE, 1, &graphics_pipeline_create_info, nullptr, &pipeline.pipeline); - ERR_FAIL_COND_V_MSG(err, RID(), "vkCreateGraphicsPipelines failed with error " + itos(err) + "."); + ERR_FAIL_COND_V_MSG(err, RID(), "vkCreateGraphicsPipelines failed with error " + itos(err) + " for shader '" + shader->name + "'."); pipeline.set_formats = shader->set_formats; pipeline.push_constant_stages = shader->push_constant.push_constants_vk_stage; @@ -5705,6 +6622,7 @@ RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferForma #ifdef DEBUG_ENABLED pipeline.validation.dynamic_state = p_dynamic_state_flags; pipeline.validation.framebuffer_format = p_framebuffer_format; + pipeline.validation.render_pass = p_for_render_pass; pipeline.validation.vertex_format = p_vertex_format; pipeline.validation.uses_restart_indices = input_assembly_create_info.primitiveRestartEnable; @@ -5743,11 +6661,11 @@ bool RenderingDeviceVulkan::render_pipeline_is_valid(RID p_pipeline) { /**** COMPUTE PIPELINE ****/ /**************************/ -RID RenderingDeviceVulkan::compute_pipeline_create(RID p_shader) { +RID RenderingDeviceVulkan::compute_pipeline_create(RID p_shader, const Vector<PipelineSpecializationConstant> &p_specialization_constants) { _THREAD_SAFE_METHOD_ //needs a shader - Shader *shader = shader_owner.getornull(p_shader); + Shader *shader = shader_owner.get_or_null(p_shader); ERR_FAIL_COND_V(!shader, RID()); ERR_FAIL_COND_V_MSG(!shader->is_compute, RID(), @@ -5765,6 +6683,44 @@ RID RenderingDeviceVulkan::compute_pipeline_create(RID p_shader) { compute_pipeline_create_info.basePipelineHandle = VK_NULL_HANDLE; compute_pipeline_create_info.basePipelineIndex = 0; + VkSpecializationInfo specialization_info; + Vector<VkSpecializationMapEntry> specialization_map_entries; + Vector<uint32_t> specialization_constant_data; + + if (shader->specialization_constants.size()) { + specialization_constant_data.resize(shader->specialization_constants.size()); + uint32_t *data_ptr = specialization_constant_data.ptrw(); + for (int i = 0; i < shader->specialization_constants.size(); i++) { + //see if overridden + const Shader::SpecializationConstant &sc = shader->specialization_constants[i]; + data_ptr[i] = sc.constant.int_value; //just copy the 32 bits + + for (int j = 0; j < p_specialization_constants.size(); j++) { + const PipelineSpecializationConstant &psc = p_specialization_constants[j]; + if (psc.constant_id == sc.constant.constant_id) { + ERR_FAIL_COND_V_MSG(psc.type != sc.constant.type, RID(), "Specialization constant provided for id (" + itos(sc.constant.constant_id) + ") is of the wrong type."); + data_ptr[i] = sc.constant.int_value; + break; + } + } + + VkSpecializationMapEntry entry; + + entry.constantID = sc.constant.constant_id; + entry.offset = i * sizeof(uint32_t); + entry.size = sizeof(uint32_t); + + specialization_map_entries.push_back(entry); + } + + specialization_info.dataSize = specialization_constant_data.size() * sizeof(uint32_t); + specialization_info.pData = data_ptr; + specialization_info.mapEntryCount = specialization_map_entries.size(); + specialization_info.pMapEntries = specialization_map_entries.ptr(); + + compute_pipeline_create_info.stage.pSpecializationInfo = &specialization_info; + } + ComputePipeline pipeline; VkResult err = vkCreateComputePipelines(device, VK_NULL_HANDLE, 1, &compute_pipeline_create_info, nullptr, &pipeline.pipeline); ERR_FAIL_COND_V_MSG(err, RID(), "vkCreateComputePipelines failed with error " + itos(err) + "."); @@ -5843,13 +6799,18 @@ RenderingDevice::DrawListID RenderingDeviceVulkan::draw_list_begin_for_screen(Di 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; + + if (!context->window_is_valid_swapchain(p_screen)) { + return INVALID_ID; + } + + Size2i size = Size2i(context->window_get_width(p_screen), context->window_get_height(p_screen)); + + _draw_list_allocate(Rect2i(Vector2i(), size), 0, 0); #ifdef DEBUG_ENABLED - draw_list->validation.framebuffer_format = screen_get_framebuffer_format(); + draw_list_framebuffer_format = screen_get_framebuffer_format(); #endif - draw_list_count = 0; - draw_list_split = false; + draw_list_subpass_count = 1; VkRenderPassBeginInfo render_pass_begin; render_pass_begin.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO; @@ -5857,8 +6818,8 @@ RenderingDevice::DrawListID RenderingDeviceVulkan::draw_list_begin_for_screen(Di 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.extent.width = size.width; + render_pass_begin.renderArea.extent.height = size.height; render_pass_begin.renderArea.offset.x = 0; render_pass_begin.renderArea.offset.y = 0; @@ -5898,18 +6859,19 @@ RenderingDevice::DrawListID RenderingDeviceVulkan::draw_list_begin_for_screen(Di 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) { +Error RenderingDeviceVulkan::_draw_list_setup_framebuffer(Framebuffer *p_framebuffer, InitialAction p_initial_color_action, FinalAction p_final_color_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, VkFramebuffer *r_framebuffer, VkRenderPass *r_render_pass, uint32_t *r_subpass_count) { Framebuffer::VersionKey vk; vk.initial_color_action = p_initial_color_action; vk.final_color_action = p_final_color_action; vk.initial_depth_action = p_initial_depth_action; vk.final_depth_action = p_final_depth_action; + vk.view_count = p_framebuffer->view_count; if (!p_framebuffer->framebuffers.has(vk)) { //need to create this version Framebuffer::Version version; - version.render_pass = _render_pass_create(framebuffer_formats[p_framebuffer->format_id].E->key().attachments, p_initial_color_action, p_final_color_action, p_initial_depth_action, p_final_depth_action); + version.render_pass = _render_pass_create(framebuffer_formats[p_framebuffer->format_id].E->key().attachments, framebuffer_formats[p_framebuffer->format_id].E->key().passes, p_initial_color_action, p_final_color_action, p_initial_depth_action, p_final_depth_action, p_framebuffer->view_count); VkFramebufferCreateInfo framebuffer_create_info; framebuffer_create_info.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO; @@ -5918,11 +6880,12 @@ Error RenderingDeviceVulkan::_draw_list_setup_framebuffer(Framebuffer *p_framebu framebuffer_create_info.renderPass = version.render_pass; Vector<VkImageView> attachments; for (int i = 0; i < p_framebuffer->texture_ids.size(); i++) { - Texture *texture = texture_owner.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); + Texture *texture = texture_owner.get_or_null(p_framebuffer->texture_ids[i]); + if (texture) { + 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(); @@ -5933,11 +6896,14 @@ Error RenderingDeviceVulkan::_draw_list_setup_framebuffer(Framebuffer *p_framebu VkResult err = vkCreateFramebuffer(device, &framebuffer_create_info, nullptr, &version.framebuffer); ERR_FAIL_COND_V_MSG(err, ERR_CANT_CREATE, "vkCreateFramebuffer failed with error " + itos(err) + "."); + version.subpass_count = framebuffer_formats[p_framebuffer->format_id].E->key().passes.size(); + p_framebuffer->framebuffers.insert(vk, version); } const Framebuffer::Version &version = p_framebuffer->framebuffers[vk]; *r_framebuffer = version.framebuffer; *r_render_pass = version.render_pass; + *r_subpass_count = version.subpass_count; return OK; } @@ -5963,13 +6929,18 @@ Error RenderingDeviceVulkan::_draw_list_render_pass_begin(Framebuffer *framebuff Vector<VkClearValue> clear_values; clear_values.resize(framebuffer->texture_ids.size()); - + int clear_values_count = 0; { int color_index = 0; for (int i = 0; i < framebuffer->texture_ids.size(); i++) { - Texture *texture = texture_owner.getornull(framebuffer->texture_ids[i]); VkClearValue clear_value; + Texture *texture = texture_owner.get_or_null(framebuffer->texture_ids[i]); + if (!texture) { + color_index++; + continue; + } + if (color_index < p_clear_colors.size() && texture->usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { ERR_FAIL_INDEX_V(color_index, p_clear_colors.size(), ERR_BUG); //a bug Color clear_color = p_clear_colors[color_index]; @@ -5987,15 +6958,18 @@ Error RenderingDeviceVulkan::_draw_list_render_pass_begin(Framebuffer *framebuff clear_value.color.float32[2] = 0; clear_value.color.float32[3] = 0; } - clear_values.write[i] = clear_value; + clear_values.write[clear_values_count++] = clear_value; } } - render_pass_begin.clearValueCount = clear_values.size(); + render_pass_begin.clearValueCount = clear_values_count; render_pass_begin.pClearValues = clear_values.ptr(); for (int i = 0; i < p_storage_textures.size(); i++) { - Texture *texture = texture_owner.getornull(p_storage_textures[i]); + Texture *texture = texture_owner.get_or_null(p_storage_textures[i]); + if (!texture) { + continue; + } ERR_CONTINUE_MSG(!(texture->usage_flags & TEXTURE_USAGE_STORAGE_BIT), "Supplied storage texture " + itos(i) + " for draw list is not set to be used for storage."); if (texture->usage_flags & TEXTURE_USAGE_SAMPLING_BIT) { @@ -6033,7 +7007,10 @@ Error RenderingDeviceVulkan::_draw_list_render_pass_begin(Framebuffer *framebuff 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 *texture = texture_owner.get_or_null(framebuffer->texture_ids[i]); + if (!texture) { + continue; + } texture->bound = true; draw_list_bound_textures.push_back(framebuffer->texture_ids[i]); } @@ -6042,15 +7019,21 @@ Error RenderingDeviceVulkan::_draw_list_render_pass_begin(Framebuffer *framebuff } void RenderingDeviceVulkan::_draw_list_insert_clear_region(DrawList *draw_list, Framebuffer *framebuffer, Point2i viewport_offset, Point2i viewport_size, bool p_clear_color, const Vector<Color> &p_clear_colors, bool p_clear_depth, float p_depth, uint32_t p_stencil) { + ERR_FAIL_COND_MSG(p_clear_color && p_clear_colors.size() != framebuffer->texture_ids.size(), "Clear color values supplied (" + itos(p_clear_colors.size()) + ") differ from the amount required for framebuffer color attachments (" + itos(framebuffer->texture_ids.size()) + ")."); Vector<VkClearAttachment> clear_attachments; int color_index = 0; + int texture_index = 0; for (int i = 0; i < framebuffer->texture_ids.size(); i++) { - Texture *texture = texture_owner.getornull(framebuffer->texture_ids[i]); - VkClearAttachment clear_at = {}; + Texture *texture = texture_owner.get_or_null(framebuffer->texture_ids[i]); + if (!texture) { + texture_index++; + continue; + } + + 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]; + Color clear_color = p_clear_colors[texture_index++]; clear_at.clearValue.color.float32[0] = clear_color.r; clear_at.clearValue.color.float32[1] = clear_color.g; clear_at.clearValue.color.float32[2] = clear_color.b; @@ -6088,7 +7071,7 @@ RenderingDevice::DrawListID RenderingDeviceVulkan::draw_list_begin(RID p_framebu 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); + Framebuffer *framebuffer = framebuffer_owner.get_or_null(p_framebuffer); ERR_FAIL_COND_V(!framebuffer, INVALID_ID); Point2i viewport_offset; @@ -6126,16 +7109,20 @@ RenderingDevice::DrawListID RenderingDeviceVulkan::draw_list_begin(RID p_framebu } 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) + ")."); + int color_count = 0; + for (int i = 0; i < framebuffer->texture_ids.size(); i++) { + Texture *texture = texture_owner.get_or_null(framebuffer->texture_ids[i]); + if (!texture || !(texture->usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT)) { + color_count++; + } + } + ERR_FAIL_COND_V_MSG(p_clear_color_values.size() != color_count, INVALID_ID, "Clear color values supplied (" + itos(p_clear_color_values.size()) + ") differ from the amount required for framebuffer color attachments (" + itos(color_count) + ")."); } VkFramebuffer vkframebuffer; VkRenderPass render_pass; - Error err = _draw_list_setup_framebuffer(framebuffer, p_initial_color_action, p_final_color_action, p_initial_depth_action, p_final_depth_action, &vkframebuffer, &render_pass); + Error err = _draw_list_setup_framebuffer(framebuffer, p_initial_color_action, p_final_color_action, p_initial_depth_action, p_final_depth_action, &vkframebuffer, &render_pass, &draw_list_subpass_count); ERR_FAIL_COND_V(err != OK, INVALID_ID); VkCommandBuffer command_buffer = frames[frame].draw_command_buffer; @@ -6145,13 +7132,14 @@ RenderingDevice::DrawListID RenderingDeviceVulkan::draw_list_begin(RID p_framebu return INVALID_ID; } - draw_list = memnew(DrawList); - draw_list->command_buffer = command_buffer; + draw_list_render_pass = render_pass; + draw_list_vkframebuffer = vkframebuffer; + + _draw_list_allocate(Rect2i(viewport_offset, viewport_size), 0, 0); #ifdef DEBUG_ENABLED - draw_list->validation.framebuffer_format = framebuffer->format_id; + draw_list_framebuffer_format = framebuffer->format_id; #endif - draw_list_count = 0; - draw_list_split = false; + draw_list_current_subpass = 0; if (needs_clear_color || needs_clear_depth) { _draw_list_insert_clear_region(draw_list, framebuffer, viewport_offset, viewport_size, needs_clear_color, p_clear_color_values, needs_clear_depth, p_clear_depth, p_clear_stencil); @@ -6175,7 +7163,6 @@ RenderingDevice::DrawListID RenderingDeviceVulkan::draw_list_begin(RID p_framebu vkCmdSetScissor(command_buffer, 0, 1, &scissor); - draw_list->viewport = Rect2i(viewport_offset, viewport_size); return int64_t(ID_TYPE_DRAW_LIST) << ID_BASE_SHIFT; } @@ -6184,7 +7171,7 @@ Error RenderingDeviceVulkan::draw_list_begin_split(RID p_framebuffer, uint32_t p ERR_FAIL_COND_V(p_splits < 1, ERR_INVALID_DECLARATION); - Framebuffer *framebuffer = framebuffer_owner.getornull(p_framebuffer); + Framebuffer *framebuffer = framebuffer_owner.get_or_null(p_framebuffer); ERR_FAIL_COND_V(!framebuffer, ERR_INVALID_DECLARATION); Point2i viewport_offset; @@ -6217,47 +7204,23 @@ Error RenderingDeviceVulkan::draw_list_begin_split(RID p_framebuffer, uint32_t p 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) + "."); + int color_count = 0; + for (int i = 0; i < framebuffer->texture_ids.size(); i++) { + Texture *texture = texture_owner.get_or_null(framebuffer->texture_ids[i]); - split_draw_list_allocators.write[i].command_buffers.push_back(command_buffer); + if (!texture || !(texture->usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT)) { + color_count++; } } + + ERR_FAIL_COND_V_MSG(p_clear_color_values.size() != color_count, ERR_INVALID_PARAMETER, + "Clear color values supplied (" + itos(p_clear_color_values.size()) + ") differ from the amount required for framebuffer (" + itos(color_count) + ")."); } VkFramebuffer vkframebuffer; VkRenderPass render_pass; - Error err = _draw_list_setup_framebuffer(framebuffer, p_initial_color_action, p_final_color_action, p_initial_depth_action, p_final_depth_action, &vkframebuffer, &render_pass); + Error err = _draw_list_setup_framebuffer(framebuffer, p_initial_color_action, p_final_color_action, p_initial_depth_action, p_final_depth_action, &vkframebuffer, &render_pass, &draw_list_subpass_count); ERR_FAIL_COND_V(err != OK, ERR_CANT_CREATE); VkCommandBuffer frame_command_buffer = frames[frame].draw_command_buffer; @@ -6267,53 +7230,24 @@ Error RenderingDeviceVulkan::draw_list_begin_split(RID p_framebuffer, uint32_t p return ERR_CANT_CREATE; } - draw_list = memnew_arr(DrawList, p_splits); - draw_list_count = p_splits; - draw_list_split = true; + draw_list_current_subpass = 0; - 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; + draw_list_framebuffer_format = framebuffer->format_id; #endif + draw_list_render_pass = render_pass; + draw_list_vkframebuffer = vkframebuffer; - 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); - } + err = _draw_list_allocate(Rect2i(viewport_offset, viewport_size), p_splits, 0); + if (err != OK) { + return err; + } + + if (needs_clear_color || needs_clear_depth) { + _draw_list_insert_clear_region(&draw_list[0], framebuffer, viewport_offset, viewport_size, needs_clear_color, p_clear_color_values, needs_clear_depth, p_clear_depth, p_clear_stencil); + } + for (uint32_t i = 0; i < p_splits; i++) { VkViewport viewport; viewport.x = viewport_offset.x; viewport.y = viewport_offset.y; @@ -6322,7 +7256,7 @@ Error RenderingDeviceVulkan::draw_list_begin_split(RID p_framebuffer, uint32_t p viewport.minDepth = 0; viewport.maxDepth = 1.0; - vkCmdSetViewport(command_buffer, 0, 1, &viewport); + vkCmdSetViewport(draw_list[i].command_buffer, 0, 1, &viewport); VkRect2D scissor; scissor.offset.x = viewport_offset.x; @@ -6330,10 +7264,8 @@ Error RenderingDeviceVulkan::draw_list_begin_split(RID p_framebuffer, uint32_t p scissor.extent.width = viewport_size.width; scissor.extent.height = viewport_size.height; - vkCmdSetScissor(command_buffer, 0, 1, &scissor); + vkCmdSetScissor(draw_list[i].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; @@ -6375,10 +7307,10 @@ void RenderingDeviceVulkan::draw_list_bind_render_pipeline(DrawListID p_list, RI 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); + const RenderPipeline *pipeline = render_pipeline_owner.get_or_null(p_render_pipeline); ERR_FAIL_COND(!pipeline); #ifdef DEBUG_ENABLED - ERR_FAIL_COND(pipeline->validation.framebuffer_format != dl->validation.framebuffer_format); + ERR_FAIL_COND(pipeline->validation.framebuffer_format != draw_list_framebuffer_format && pipeline->validation.render_pass != draw_list_current_subpass); #endif if (p_render_pipeline == dl->state.pipeline) { @@ -6450,7 +7382,7 @@ void RenderingDeviceVulkan::draw_list_bind_uniform_set(DrawListID p_list, RID p_ 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); + const UniformSet *uniform_set = uniform_set_owner.get_or_null(p_uniform_set); ERR_FAIL_COND(!uniform_set); if (p_index > dl->state.set_count) { @@ -6498,7 +7430,7 @@ void RenderingDeviceVulkan::draw_list_bind_vertex_array(DrawListID p_list, RID p 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); + const VertexArray *vertex_array = vertex_array_owner.get_or_null(p_vertex_array); ERR_FAIL_COND(!vertex_array); if (dl->state.vertex_array == p_vertex_array) { @@ -6522,7 +7454,7 @@ void RenderingDeviceVulkan::draw_list_bind_index_array(DrawListID p_list, RID p_ 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); + const IndexArray *index_array = index_array_owner.get_or_null(p_index_array); ERR_FAIL_COND(!index_array); if (dl->state.index_array == p_index_array) { @@ -6608,7 +7540,7 @@ void RenderingDeviceVulkan::draw_list_draw(DrawListID p_list, bool p_use_indices 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); + UniformSet *us = uniform_set_owner.get_or_null(dl->state.sets[i].uniform_set); ERR_FAIL_MSG("Uniforms supplied for set (" + itos(i) + "):\n" + _shader_uniform_debug(us->shader_id, us->shader_set) + "\nare not the same format as required by the pipeline shader. Pipeline shader requires the following bindings:\n" + _shader_uniform_debug(dl->state.pipeline_shader)); } else { ERR_FAIL_MSG("Uniforms supplied for set (" + itos(i) + ", which was was just freed) are not the same format as required by the pipeline shader. Pipeline shader requires the following bindings:\n" + _shader_uniform_debug(dl->state.pipeline_shader)); @@ -6717,33 +7649,175 @@ void RenderingDeviceVulkan::draw_list_disable_scissor(DrawListID p_list) { vkCmdSetScissor(dl->command_buffer, 0, 1, &scissor); } -void RenderingDeviceVulkan::draw_list_end(uint32_t p_post_barrier) { - _THREAD_SAFE_METHOD_ +uint32_t RenderingDeviceVulkan::draw_list_get_current_pass() { + return draw_list_current_subpass; +} - ERR_FAIL_COND_MSG(!draw_list, "Immediate draw list is already inactive."); +RenderingDevice::DrawListID RenderingDeviceVulkan::draw_list_switch_to_next_pass() { + ERR_FAIL_COND_V(draw_list == nullptr, INVALID_ID); + ERR_FAIL_COND_V(draw_list_current_subpass >= draw_list_subpass_count - 1, INVALID_FORMAT_ID); + + draw_list_current_subpass++; + + Rect2i viewport; + _draw_list_free(&viewport); + + vkCmdNextSubpass(frames[frame].draw_command_buffer, VK_SUBPASS_CONTENTS_INLINE); + + _draw_list_allocate(viewport, 0, draw_list_current_subpass); + + return int64_t(ID_TYPE_DRAW_LIST) << ID_BASE_SHIFT; +} +Error RenderingDeviceVulkan::draw_list_switch_to_next_pass_split(uint32_t p_splits, DrawListID *r_split_ids) { + ERR_FAIL_COND_V(draw_list == nullptr, ERR_INVALID_PARAMETER); + ERR_FAIL_COND_V(draw_list_current_subpass >= draw_list_subpass_count - 1, ERR_INVALID_PARAMETER); + + draw_list_current_subpass++; + + Rect2i viewport; + _draw_list_free(&viewport); + + vkCmdNextSubpass(frames[frame].draw_command_buffer, VK_SUBPASS_CONTENTS_INLINE); + + _draw_list_allocate(viewport, p_splits, draw_list_current_subpass); + + for (uint32_t i = 0; i < p_splits; i++) { + r_split_ids[i] = (int64_t(ID_TYPE_SPLIT_DRAW_LIST) << ID_BASE_SHIFT) + i; + } + + return OK; +} + +Error RenderingDeviceVulkan::_draw_list_allocate(const Rect2i &p_viewport, uint32_t p_splits, uint32_t p_subpass) { + // Lock while draw_list is active + _THREAD_SAFE_LOCK_ + + if (p_splits == 0) { + draw_list = memnew(DrawList); + draw_list->command_buffer = frames[frame].draw_command_buffer; + draw_list->viewport = p_viewport; + draw_list_count = 0; + draw_list_split = false; + } else { + if (p_splits > (uint32_t)split_draw_list_allocators.size()) { + uint32_t from = split_draw_list_allocators.size(); + split_draw_list_allocators.resize(p_splits); + for (uint32_t i = from; i < p_splits; i++) { + VkCommandPoolCreateInfo cmd_pool_info; + cmd_pool_info.sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO; + cmd_pool_info.pNext = nullptr; + cmd_pool_info.queueFamilyIndex = context->get_graphics_queue_family_index(); + cmd_pool_info.flags = VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT; + + VkResult res = vkCreateCommandPool(device, &cmd_pool_info, nullptr, &split_draw_list_allocators.write[i].command_pool); + ERR_FAIL_COND_V_MSG(res, ERR_CANT_CREATE, "vkCreateCommandPool failed with error " + itos(res) + "."); + + for (int j = 0; j < frame_count; j++) { + VkCommandBuffer command_buffer; + + VkCommandBufferAllocateInfo cmdbuf; + //no command buffer exists, create it. + cmdbuf.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO; + cmdbuf.pNext = nullptr; + cmdbuf.commandPool = split_draw_list_allocators[i].command_pool; + cmdbuf.level = VK_COMMAND_BUFFER_LEVEL_SECONDARY; + cmdbuf.commandBufferCount = 1; + + VkResult err = vkAllocateCommandBuffers(device, &cmdbuf, &command_buffer); + ERR_FAIL_COND_V_MSG(err, ERR_CANT_CREATE, "vkAllocateCommandBuffers failed with error " + itos(err) + "."); + + split_draw_list_allocators.write[i].command_buffers.push_back(command_buffer); + } + } + } + draw_list = memnew_arr(DrawList, p_splits); + draw_list_count = p_splits; + draw_list_split = true; + + for (uint32_t i = 0; i < p_splits; i++) { + //take a command buffer and initialize it + VkCommandBuffer command_buffer = split_draw_list_allocators[i].command_buffers[frame]; + + VkCommandBufferInheritanceInfo inheritance_info; + inheritance_info.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_INHERITANCE_INFO; + inheritance_info.pNext = nullptr; + inheritance_info.renderPass = draw_list_render_pass; + inheritance_info.subpass = p_subpass; + inheritance_info.framebuffer = draw_list_vkframebuffer; + inheritance_info.occlusionQueryEnable = false; + inheritance_info.queryFlags = 0; //? + inheritance_info.pipelineStatistics = 0; + + VkCommandBufferBeginInfo cmdbuf_begin; + cmdbuf_begin.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO; + cmdbuf_begin.pNext = nullptr; + cmdbuf_begin.flags = VK_COMMAND_BUFFER_USAGE_SIMULTANEOUS_USE_BIT | VK_COMMAND_BUFFER_USAGE_RENDER_PASS_CONTINUE_BIT; + cmdbuf_begin.pInheritanceInfo = &inheritance_info; + + VkResult res = vkResetCommandBuffer(command_buffer, 0); + if (res) { + memdelete_arr(draw_list); + draw_list = nullptr; + ERR_FAIL_V_MSG(ERR_CANT_CREATE, "vkResetCommandBuffer failed with error " + itos(res) + "."); + } + + res = vkBeginCommandBuffer(command_buffer, &cmdbuf_begin); + if (res) { + memdelete_arr(draw_list); + draw_list = nullptr; + ERR_FAIL_V_MSG(ERR_CANT_CREATE, "vkBeginCommandBuffer failed with error " + itos(res) + "."); + } + + draw_list[i].command_buffer = command_buffer; + draw_list[i].viewport = p_viewport; + } + } + + return OK; +} +void RenderingDeviceVulkan::_draw_list_free(Rect2i *r_last_viewport) { if (draw_list_split) { //send all command buffers VkCommandBuffer *command_buffers = (VkCommandBuffer *)alloca(sizeof(VkCommandBuffer) * draw_list_count); for (uint32_t i = 0; i < draw_list_count; i++) { vkEndCommandBuffer(draw_list[i].command_buffer); command_buffers[i] = draw_list[i].command_buffer; + if (r_last_viewport) { + if (i == 0 || draw_list[i].viewport_set) { + *r_last_viewport = draw_list[i].viewport; + } + } } vkCmdExecuteCommands(frames[frame].draw_command_buffer, draw_list_count, command_buffers); - vkCmdEndRenderPass(frames[frame].draw_command_buffer); memdelete_arr(draw_list); draw_list = nullptr; } else { + if (r_last_viewport) { + *r_last_viewport = draw_list->viewport; + } //just end the list - vkCmdEndRenderPass(draw_list->command_buffer); memdelete(draw_list); draw_list = nullptr; } + // draw_list is no longer active + _THREAD_SAFE_UNLOCK_ +} + +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."); + + _draw_list_free(); + + vkCmdEndRenderPass(frames[frame].draw_command_buffer); + for (int i = 0; i < draw_list_bound_textures.size(); i++) { - Texture *texture = texture_owner.getornull(draw_list_bound_textures[i]); + Texture *texture = texture_owner.get_or_null(draw_list_bound_textures[i]); ERR_CONTINUE(!texture); //wtf if (draw_list_unbind_color_textures && (texture->usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT)) { texture->bound = false; @@ -6791,7 +7865,7 @@ void RenderingDeviceVulkan::draw_list_end(uint32_t p_post_barrier) { } for (uint32_t i = 0; i < image_barrier_count; i++) { - Texture *texture = texture_owner.getornull(draw_list_storage_textures[i]); + Texture *texture = texture_owner.get_or_null(draw_list_storage_textures[i]); VkImageMemoryBarrier &image_memory_barrier = image_barriers[i]; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; @@ -6844,6 +7918,9 @@ RenderingDevice::ComputeListID RenderingDeviceVulkan::compute_list_begin(bool p_ ERR_FAIL_COND_V_MSG(!p_allow_draw_overlap && draw_list != nullptr, INVALID_ID, "Only one draw list can be active at the same time."); ERR_FAIL_COND_V_MSG(compute_list != nullptr, INVALID_ID, "Only one draw/compute list can be active at the same time."); + // Lock while compute_list is active + _THREAD_SAFE_LOCK_ + compute_list = memnew(ComputeList); compute_list->command_buffer = frames[frame].draw_command_buffer; compute_list->state.allow_draw_overlap = p_allow_draw_overlap; @@ -6857,7 +7934,7 @@ void RenderingDeviceVulkan::compute_list_bind_compute_pipeline(ComputeListID p_l ComputeList *cl = compute_list; - const ComputePipeline *pipeline = compute_pipeline_owner.getornull(p_compute_pipeline); + const ComputePipeline *pipeline = compute_pipeline_owner.get_or_null(p_compute_pipeline); ERR_FAIL_COND(!pipeline); if (p_compute_pipeline == cl->state.pipeline) { @@ -6930,7 +8007,7 @@ void RenderingDeviceVulkan::compute_list_bind_uniform_set(ComputeListID p_list, 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); + UniformSet *uniform_set = uniform_set_owner.get_or_null(p_uniform_set); ERR_FAIL_COND(!uniform_set); if (p_index > cl->state.set_count) { @@ -7012,13 +8089,13 @@ void RenderingDeviceVulkan::compute_list_bind_uniform_set(ComputeListID p_list, 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_transfer = false; } else { src_access_flags = 0; textures_to_storage[i]->used_in_compute = false; textures_to_storage[i]->used_in_raster = false; - textures_to_storage[i]->used_in_compute = false; + textures_to_storage[i]->used_in_transfer = false; textures_to_storage[i]->used_in_frame = frames_drawn; } @@ -7096,12 +8173,15 @@ void RenderingDeviceVulkan::compute_list_dispatch(ComputeListID p_list, uint32_t ComputeList *cl = compute_list; #ifdef DEBUG_ENABLED + ERR_FAIL_COND_MSG(p_x_groups == 0, "Dispatch amount of X compute groups (" + itos(p_x_groups) + ") is zero."); + ERR_FAIL_COND_MSG(p_z_groups == 0, "Dispatch amount of Z compute groups (" + itos(p_z_groups) + ") is zero."); + ERR_FAIL_COND_MSG(p_y_groups == 0, "Dispatch amount of Y compute groups (" + itos(p_y_groups) + ") is zero."); ERR_FAIL_COND_MSG(p_x_groups > limits.maxComputeWorkGroupCount[0], "Dispatch amount of X compute groups (" + itos(p_x_groups) + ") is larger than device limit (" + itos(limits.maxComputeWorkGroupCount[0]) + ")"); ERR_FAIL_COND_MSG(p_y_groups > limits.maxComputeWorkGroupCount[1], - "Dispatch amount of Y compute groups (" + itos(p_x_groups) + ") is larger than device limit (" + itos(limits.maxComputeWorkGroupCount[0]) + ")"); + "Dispatch amount of Y compute groups (" + itos(p_y_groups) + ") is larger than device limit (" + itos(limits.maxComputeWorkGroupCount[1]) + ")"); ERR_FAIL_COND_MSG(p_z_groups > limits.maxComputeWorkGroupCount[2], - "Dispatch amount of Z compute groups (" + itos(p_x_groups) + ") is larger than device limit (" + itos(limits.maxComputeWorkGroupCount[0]) + ")"); + "Dispatch amount of Z compute groups (" + itos(p_z_groups) + ") is larger than device limit (" + itos(limits.maxComputeWorkGroupCount[2]) + ")"); ERR_FAIL_COND_MSG(!cl->validation.active, "Submitted Compute Lists can no longer be modified."); #endif @@ -7129,7 +8209,7 @@ void RenderingDeviceVulkan::compute_list_dispatch(ComputeListID p_list, uint32_t 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); + UniformSet *us = uniform_set_owner.get_or_null(cl->state.sets[i].uniform_set); ERR_FAIL_MSG("Uniforms supplied for set (" + itos(i) + "):\n" + _shader_uniform_debug(us->shader_id, us->shader_set) + "\nare not the same format as required by the pipeline shader. Pipeline shader requires the following bindings:\n" + _shader_uniform_debug(cl->state.pipeline_shader)); } else { ERR_FAIL_MSG("Uniforms supplied for set (" + itos(i) + ", which was was just freed) are not the same format as required by the pipeline shader. Pipeline shader requires the following bindings:\n" + _shader_uniform_debug(cl->state.pipeline_shader)); @@ -7150,6 +8230,12 @@ void RenderingDeviceVulkan::compute_list_dispatch_threads(ComputeListID p_list, ERR_FAIL_COND(p_list != ID_TYPE_COMPUTE_LIST); ERR_FAIL_COND(!compute_list); +#ifdef DEBUG_ENABLED + ERR_FAIL_COND_MSG(p_x_threads == 0, "Dispatch amount of X compute threads (" + itos(p_x_threads) + ") is zero."); + ERR_FAIL_COND_MSG(p_y_threads == 0, "Dispatch amount of Y compute threads (" + itos(p_y_threads) + ") is zero."); + ERR_FAIL_COND_MSG(p_z_threads == 0, "Dispatch amount of Z compute threads (" + itos(p_z_threads) + ") is zero."); +#endif + ComputeList *cl = compute_list; #ifdef DEBUG_ENABLED @@ -7172,7 +8258,7 @@ void RenderingDeviceVulkan::compute_list_dispatch_indirect(ComputeListID p_list, ERR_FAIL_COND(!compute_list); ComputeList *cl = compute_list; - Buffer *buffer = storage_buffer_owner.getornull(p_buffer); + Buffer *buffer = storage_buffer_owner.get_or_null(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."); @@ -7206,7 +8292,7 @@ void RenderingDeviceVulkan::compute_list_dispatch_indirect(ComputeListID p_list, 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); + UniformSet *us = uniform_set_owner.get_or_null(cl->state.sets[i].uniform_set); ERR_FAIL_MSG("Uniforms supplied for set (" + itos(i) + "):\n" + _shader_uniform_debug(us->shader_id, us->shader_set) + "\nare not the same format as required by the pipeline shader. Pipeline shader requires the following bindings:\n" + _shader_uniform_debug(cl->state.pipeline_shader)); } else { ERR_FAIL_MSG("Uniforms supplied for set (" + itos(i) + ", which was was just freed) are not the same format as required by the pipeline shader. Pipeline shader requires the following bindings:\n" + _shader_uniform_debug(cl->state.pipeline_shader)); @@ -7263,31 +8349,31 @@ void RenderingDeviceVulkan::compute_list_end(uint32_t p_post_barrier) { uint32_t barrier_idx = 0; - for (Set<Texture *>::Element *E = compute_list->state.textures_to_sampled_layout.front(); E; E = E->next()) { + for (Texture *E : compute_list->state.textures_to_sampled_layout) { 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.oldLayout = E->layout; image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; - image_memory_barrier.image = E->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; + image_memory_barrier.image = E->image; + image_memory_barrier.subresourceRange.aspectMask = E->read_aspect_mask; + image_memory_barrier.subresourceRange.baseMipLevel = E->base_mipmap; + image_memory_barrier.subresourceRange.levelCount = E->mipmaps; + image_memory_barrier.subresourceRange.baseArrayLayer = E->base_layer; + image_memory_barrier.subresourceRange.layerCount = E->layers; - E->get()->layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; + E->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; + if (E->used_in_frame != frames_drawn) { + E->used_in_transfer = false; + E->used_in_raster = false; + E->used_in_compute = false; + E->used_in_frame = frames_drawn; } } @@ -7308,6 +8394,9 @@ void RenderingDeviceVulkan::compute_list_end(uint32_t p_post_barrier) { memdelete(compute_list); compute_list = nullptr; + + // compute_list is no longer active + _THREAD_SAFE_UNLOCK_ } void RenderingDeviceVulkan::barrier(uint32_t p_from, uint32_t p_to) { @@ -7418,60 +8507,65 @@ void RenderingDeviceVulkan::draw_list_render_secondary_to_framebuffer(ID p_frame 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); + Texture *texture = texture_owner.get_or_null(p_id); frames[frame].textures_to_dispose_of.push_back(*texture); texture_owner.free(p_id); } else if (framebuffer_owner.owns(p_id)) { - Framebuffer *framebuffer = framebuffer_owner.getornull(p_id); + Framebuffer *framebuffer = framebuffer_owner.get_or_null(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); + VkSampler *sampler = sampler_owner.get_or_null(p_id); frames[frame].samplers_to_dispose_of.push_back(*sampler); sampler_owner.free(p_id); } else if (vertex_buffer_owner.owns(p_id)) { - Buffer *vertex_buffer = vertex_buffer_owner.getornull(p_id); + Buffer *vertex_buffer = vertex_buffer_owner.get_or_null(p_id); frames[frame].buffers_to_dispose_of.push_back(*vertex_buffer); vertex_buffer_owner.free(p_id); } else if (vertex_array_owner.owns(p_id)) { vertex_array_owner.free(p_id); } else if (index_buffer_owner.owns(p_id)) { - IndexBuffer *index_buffer = index_buffer_owner.getornull(p_id); + IndexBuffer *index_buffer = index_buffer_owner.get_or_null(p_id); Buffer b; b.allocation = index_buffer->allocation; b.buffer = index_buffer->buffer; b.size = index_buffer->size; + b.buffer_info = {}; frames[frame].buffers_to_dispose_of.push_back(b); index_buffer_owner.free(p_id); } else if (index_array_owner.owns(p_id)) { index_array_owner.free(p_id); } else if (shader_owner.owns(p_id)) { - Shader *shader = shader_owner.getornull(p_id); + Shader *shader = shader_owner.get_or_null(p_id); frames[frame].shaders_to_dispose_of.push_back(*shader); shader_owner.free(p_id); } else if (uniform_buffer_owner.owns(p_id)) { - Buffer *uniform_buffer = uniform_buffer_owner.getornull(p_id); + Buffer *uniform_buffer = uniform_buffer_owner.get_or_null(p_id); frames[frame].buffers_to_dispose_of.push_back(*uniform_buffer); uniform_buffer_owner.free(p_id); } else if (texture_buffer_owner.owns(p_id)) { - TextureBuffer *texture_buffer = texture_buffer_owner.getornull(p_id); + TextureBuffer *texture_buffer = texture_buffer_owner.get_or_null(p_id); frames[frame].buffers_to_dispose_of.push_back(texture_buffer->buffer); frames[frame].buffer_views_to_dispose_of.push_back(texture_buffer->view); texture_buffer_owner.free(p_id); } else if (storage_buffer_owner.owns(p_id)) { - Buffer *storage_buffer = storage_buffer_owner.getornull(p_id); + Buffer *storage_buffer = storage_buffer_owner.get_or_null(p_id); frames[frame].buffers_to_dispose_of.push_back(*storage_buffer); storage_buffer_owner.free(p_id); } else if (uniform_set_owner.owns(p_id)) { - UniformSet *uniform_set = uniform_set_owner.getornull(p_id); + UniformSet *uniform_set = uniform_set_owner.get_or_null(p_id); frames[frame].uniform_sets_to_dispose_of.push_back(*uniform_set); uniform_set_owner.free(p_id); + + if (uniform_set->invalidated_callback != nullptr) { + uniform_set->invalidated_callback(uniform_set->invalidated_callback_userdata); + } } else if (render_pipeline_owner.owns(p_id)) { - RenderPipeline *pipeline = render_pipeline_owner.getornull(p_id); + RenderPipeline *pipeline = render_pipeline_owner.get_or_null(p_id); frames[frame].render_pipelines_to_dispose_of.push_back(*pipeline); render_pipeline_owner.free(p_id); } else if (compute_pipeline_owner.owns(p_id)) { - ComputePipeline *pipeline = compute_pipeline_owner.getornull(p_id); + ComputePipeline *pipeline = compute_pipeline_owner.get_or_null(p_id); frames[frame].compute_pipelines_to_dispose_of.push_back(*pipeline); compute_pipeline_owner.free(p_id); } else { @@ -7490,49 +8584,49 @@ void RenderingDeviceVulkan::free(RID p_id) { // 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); + Texture *texture = texture_owner.get_or_null(p_id); if (texture->owner.is_null()) { // Don't set the source texture's name when calling on a texture view context->set_object_name(VK_OBJECT_TYPE_IMAGE, uint64_t(texture->image), p_name); } context->set_object_name(VK_OBJECT_TYPE_IMAGE_VIEW, uint64_t(texture->view), p_name + " View"); } else if (framebuffer_owner.owns(p_id)) { - //Framebuffer *framebuffer = framebuffer_owner.getornull(p_id); + //Framebuffer *framebuffer = framebuffer_owner.get_or_null(p_id); // Not implemented for now as the relationship between Framebuffer and RenderPass is very complex } else if (sampler_owner.owns(p_id)) { - VkSampler *sampler = sampler_owner.getornull(p_id); + VkSampler *sampler = sampler_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_SAMPLER, uint64_t(*sampler), p_name); } else if (vertex_buffer_owner.owns(p_id)) { - Buffer *vertex_buffer = vertex_buffer_owner.getornull(p_id); + Buffer *vertex_buffer = vertex_buffer_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_BUFFER, uint64_t(vertex_buffer->buffer), p_name); } else if (index_buffer_owner.owns(p_id)) { - IndexBuffer *index_buffer = index_buffer_owner.getornull(p_id); + IndexBuffer *index_buffer = index_buffer_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_BUFFER, uint64_t(index_buffer->buffer), p_name); } else if (shader_owner.owns(p_id)) { - Shader *shader = shader_owner.getornull(p_id); + Shader *shader = shader_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_PIPELINE_LAYOUT, uint64_t(shader->pipeline_layout), p_name + " Pipeline Layout"); for (int i = 0; i < shader->sets.size(); i++) { context->set_object_name(VK_OBJECT_TYPE_DESCRIPTOR_SET_LAYOUT, uint64_t(shader->sets[i].descriptor_set_layout), p_name); } } else if (uniform_buffer_owner.owns(p_id)) { - Buffer *uniform_buffer = uniform_buffer_owner.getornull(p_id); + Buffer *uniform_buffer = uniform_buffer_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_BUFFER, uint64_t(uniform_buffer->buffer), p_name); } else if (texture_buffer_owner.owns(p_id)) { - TextureBuffer *texture_buffer = texture_buffer_owner.getornull(p_id); + TextureBuffer *texture_buffer = texture_buffer_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_BUFFER, uint64_t(texture_buffer->buffer.buffer), p_name); context->set_object_name(VK_OBJECT_TYPE_BUFFER_VIEW, uint64_t(texture_buffer->view), p_name + " View"); } else if (storage_buffer_owner.owns(p_id)) { - Buffer *storage_buffer = storage_buffer_owner.getornull(p_id); + Buffer *storage_buffer = storage_buffer_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_BUFFER, uint64_t(storage_buffer->buffer), p_name); } else if (uniform_set_owner.owns(p_id)) { - UniformSet *uniform_set = uniform_set_owner.getornull(p_id); + UniformSet *uniform_set = uniform_set_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_DESCRIPTOR_SET, uint64_t(uniform_set->descriptor_set), p_name); } else if (render_pipeline_owner.owns(p_id)) { - RenderPipeline *pipeline = render_pipeline_owner.getornull(p_id); + RenderPipeline *pipeline = render_pipeline_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_PIPELINE, uint64_t(pipeline->pipeline), p_name); context->set_object_name(VK_OBJECT_TYPE_PIPELINE_LAYOUT, uint64_t(pipeline->pipeline_layout), p_name + " Layout"); } else if (compute_pipeline_owner.owns(p_id)) { - ComputePipeline *pipeline = compute_pipeline_owner.getornull(p_id); + ComputePipeline *pipeline = compute_pipeline_owner.get_or_null(p_id); context->set_object_name(VK_OBJECT_TYPE_PIPELINE, uint64_t(pipeline->pipeline), p_name); context->set_object_name(VK_OBJECT_TYPE_PIPELINE_LAYOUT, uint64_t(pipeline->pipeline_layout), p_name + " Layout"); } else { @@ -7555,9 +8649,19 @@ void RenderingDeviceVulkan::draw_command_end_label() { String RenderingDeviceVulkan::get_device_vendor_name() const { return context->get_device_vendor_name(); } + String RenderingDeviceVulkan::get_device_name() const { return context->get_device_name(); } + +RenderingDevice::DeviceType RenderingDeviceVulkan::get_device_type() const { + return context->get_device_type(); +} + +String RenderingDeviceVulkan::get_device_api_version() const { + return context->get_device_api_version(); +} + String RenderingDeviceVulkan::get_device_pipeline_cache_uuid() const { return context->get_device_pipeline_cache_uuid(); } @@ -7659,6 +8763,30 @@ void RenderingDeviceVulkan::sync() { local_device_processing = false; } +VmaPool RenderingDeviceVulkan::_find_or_create_small_allocs_pool(uint32_t p_mem_type_index) { + if (small_allocs_pools.has(p_mem_type_index)) { + return small_allocs_pools[p_mem_type_index]; + } + + print_verbose("Creating VMA small objects pool for memory type index " + itos(p_mem_type_index)); + + VmaPoolCreateInfo pci; + pci.memoryTypeIndex = p_mem_type_index; + pci.flags = 0; + pci.blockSize = 0; + pci.minBlockCount = 0; + pci.maxBlockCount = SIZE_MAX; + pci.priority = 0.5f; + pci.minAllocationAlignment = 0; + pci.pMemoryAllocateNext = nullptr; + VmaPool pool = VK_NULL_HANDLE; + VkResult res = vmaCreatePool(allocator, &pci, &pool); + small_allocs_pools[p_mem_type_index] = pool; // Don't try to create it again if failed the first time + ERR_FAIL_COND_V_MSG(res, pool, "vmaCreatePool failed with error " + itos(res) + "."); + + return pool; +} + void RenderingDeviceVulkan::_free_pending_resources(int p_frame) { //free in dependency usage order, so nothing weird happens //pipelines @@ -7730,10 +8858,10 @@ void RenderingDeviceVulkan::_free_pending_resources(int p_frame) { while (frames[p_frame].framebuffers_to_dispose_of.front()) { Framebuffer *framebuffer = &frames[p_frame].framebuffers_to_dispose_of.front()->get(); - for (Map<Framebuffer::VersionKey, Framebuffer::Version>::Element *E = framebuffer->framebuffers.front(); E; E = E->next()) { + for (const KeyValue<Framebuffer::VersionKey, Framebuffer::Version> &E : framebuffer->framebuffers) { //first framebuffer, then render pass because it depends on it - vkDestroyFramebuffer(device, E->get().framebuffer, nullptr); - vkDestroyRenderPass(device, E->get().render_pass, nullptr); + vkDestroyFramebuffer(device, E.value.framebuffer, nullptr); + vkDestroyRenderPass(device, E.value.render_pass, nullptr); } frames[p_frame].framebuffers_to_dispose_of.pop_front(); @@ -7749,6 +8877,7 @@ void RenderingDeviceVulkan::_free_pending_resources(int p_frame) { vkDestroyImageView(device, texture->view, nullptr); if (texture->owner.is_null()) { //actually owns the image and the allocation too + image_memory -= texture->allocation_info.size; vmaDestroyImage(allocator, texture->image, texture->allocation); } frames[p_frame].textures_to_dispose_of.pop_front(); @@ -7772,10 +8901,16 @@ 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; +uint64_t RenderingDeviceVulkan::get_memory_usage(MemoryType p_type) const { + if (p_type == MEMORY_BUFFERS) { + return buffer_memory; + } else if (p_type == MEMORY_TEXTURES) { + return image_memory; + } else { + VmaTotalStatistics stats; + vmaCalculateStatistics(allocator, &stats); + return stats.total.statistics.allocationBytes; + } } void RenderingDeviceVulkan::_flush(bool p_current_frame) { @@ -7834,6 +8969,23 @@ void RenderingDeviceVulkan::_flush(bool p_current_frame) { } void RenderingDeviceVulkan::initialize(VulkanContext *p_context, bool p_local_device) { + // get our device capabilities + { + device_capabilities.version_major = p_context->get_vulkan_major(); + device_capabilities.version_minor = p_context->get_vulkan_minor(); + + // get info about subgroups + VulkanContext::SubgroupCapabilities subgroup_capabilities = p_context->get_subgroup_capabilities(); + device_capabilities.subgroup_size = subgroup_capabilities.size; + device_capabilities.subgroup_in_shaders = subgroup_capabilities.supported_stages_flags_rd(); + device_capabilities.subgroup_operations = subgroup_capabilities.supported_operations_flags_rd(); + + // get info about further features + VulkanContext::MultiviewCapabilities multiview_capabilies = p_context->get_multiview_capabilities(); + device_capabilities.supports_multiview = multiview_capabilies.is_supported && multiview_capabilies.max_view_count > 1; + device_capabilities.supports_fsr_half_float = p_context->get_shader_capabilities().shader_float16_is_supported && p_context->get_storage_buffer_capabilities().storage_buffer_16_bit_access_is_supported; + } + context = p_context; device = p_context->get_device(); if (p_local_device) { @@ -7852,6 +9004,7 @@ void RenderingDeviceVulkan::initialize(VulkanContext *p_context, bool p_local_de memset(&allocatorInfo, 0, sizeof(VmaAllocatorCreateInfo)); allocatorInfo.physicalDevice = p_context->get_physical_device(); allocatorInfo.device = device; + allocatorInfo.instance = p_context->get_instance(); vmaCreateAllocator(&allocatorInfo, &allocator); } @@ -7865,7 +9018,7 @@ void RenderingDeviceVulkan::initialize(VulkanContext *p_context, bool p_local_de 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.queueFamilyIndex = p_context->get_graphics_queue_family_index(); cmd_pool_info.flags = VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT; VkResult res = vkCreateCommandPool(device, &cmd_pool_info, nullptr, &frames[i].command_pool); @@ -7931,11 +9084,13 @@ void RenderingDeviceVulkan::initialize(VulkanContext *p_context, bool p_local_de } } + // Note: If adding new project settings here, also duplicate their definition in + // rendering_server.cpp for headless doctool. 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 = MAX(4u, 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 = MAX(1u, staging_buffer_max_size); staging_buffer_max_size *= 1024 * 1024; if (staging_buffer_max_size < staging_buffer_block_size * 4) { @@ -7974,14 +9129,19 @@ void RenderingDeviceVulkan::_free_rids(T &p_owner, const char *p_type) { List<RID> owned; p_owner.get_owned_list(&owned); if (owned.size()) { - WARN_PRINT(itos(owned.size()) + " RIDs of type '" + p_type + "' were leaked."); - for (List<RID>::Element *E = owned.front(); E; E = E->next()) { - free(E->get()); + if (owned.size() == 1) { + WARN_PRINT(vformat("1 RID of type \"%s\" was leaked.", p_type)); + } else { + WARN_PRINT(vformat("%d RIDs of type \"%s\" were leaked.", owned.size(), p_type)); + } + for (const RID &E : owned) { + free(E); } } } void RenderingDeviceVulkan::capture_timestamp(const String &p_name) { + ERR_FAIL_COND_MSG(draw_list != nullptr, "Capturing timestamps during draw list creation is not allowed. Offending timestamp was: " + p_name); ERR_FAIL_COND(frames[frame].timestamp_count >= max_timestamp_query_elements); //this should be optional for profiling, else it will slow things down @@ -7991,35 +9151,35 @@ void RenderingDeviceVulkan::capture_timestamp(const String &p_name) { 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; + 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; + 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); } @@ -8030,6 +9190,92 @@ void RenderingDeviceVulkan::capture_timestamp(const String &p_name) { frames[frame].timestamp_count++; } +uint64_t RenderingDeviceVulkan::get_driver_resource(DriverResource p_resource, RID p_rid, uint64_t p_index) { + _THREAD_SAFE_METHOD_ + + switch (p_resource) { + case DRIVER_RESOURCE_VULKAN_DEVICE: { + return (uint64_t)context->get_device(); + } break; + case DRIVER_RESOURCE_VULKAN_PHYSICAL_DEVICE: { + return (uint64_t)context->get_physical_device(); + } break; + case DRIVER_RESOURCE_VULKAN_INSTANCE: { + return (uint64_t)context->get_instance(); + } break; + case DRIVER_RESOURCE_VULKAN_QUEUE: { + return (uint64_t)context->get_graphics_queue(); + } break; + case DRIVER_RESOURCE_VULKAN_QUEUE_FAMILY_INDEX: { + return context->get_graphics_queue_family_index(); + } break; + case DRIVER_RESOURCE_VULKAN_IMAGE: { + Texture *tex = texture_owner.get_or_null(p_rid); + ERR_FAIL_NULL_V(tex, 0); + + return (uint64_t)tex->image; + } break; + case DRIVER_RESOURCE_VULKAN_IMAGE_VIEW: { + Texture *tex = texture_owner.get_or_null(p_rid); + ERR_FAIL_NULL_V(tex, 0); + + return (uint64_t)tex->view; + } break; + case DRIVER_RESOURCE_VULKAN_IMAGE_NATIVE_TEXTURE_FORMAT: { + Texture *tex = texture_owner.get_or_null(p_rid); + ERR_FAIL_NULL_V(tex, 0); + + return vulkan_formats[tex->format]; + } break; + case DRIVER_RESOURCE_VULKAN_SAMPLER: { + VkSampler *sampler = sampler_owner.get_or_null(p_rid); + ERR_FAIL_NULL_V(sampler, 0); + + return uint64_t(*sampler); + } break; + case DRIVER_RESOURCE_VULKAN_DESCRIPTOR_SET: { + UniformSet *uniform_set = uniform_set_owner.get_or_null(p_rid); + ERR_FAIL_NULL_V(uniform_set, 0); + + return uint64_t(uniform_set->descriptor_set); + } break; + case DRIVER_RESOURCE_VULKAN_BUFFER: { + Buffer *buffer = nullptr; + if (vertex_buffer_owner.owns(p_rid)) { + buffer = vertex_buffer_owner.get_or_null(p_rid); + } else if (index_buffer_owner.owns(p_rid)) { + buffer = index_buffer_owner.get_or_null(p_rid); + } else if (uniform_buffer_owner.owns(p_rid)) { + buffer = uniform_buffer_owner.get_or_null(p_rid); + } else if (texture_buffer_owner.owns(p_rid)) { + buffer = &texture_buffer_owner.get_or_null(p_rid)->buffer; + } else if (storage_buffer_owner.owns(p_rid)) { + buffer = storage_buffer_owner.get_or_null(p_rid); + } + + ERR_FAIL_NULL_V(buffer, 0); + + return uint64_t(buffer->buffer); + } break; + case DRIVER_RESOURCE_VULKAN_COMPUTE_PIPELINE: { + ComputePipeline *compute_pipeline = compute_pipeline_owner.get_or_null(p_rid); + ERR_FAIL_NULL_V(compute_pipeline, 0); + + return uint64_t(compute_pipeline->pipeline); + } break; + case DRIVER_RESOURCE_VULKAN_RENDER_PIPELINE: { + RenderPipeline *render_pipeline = render_pipeline_owner.get_or_null(p_rid); + ERR_FAIL_NULL_V(render_pipeline, 0); + + return uint64_t(render_pipeline->pipeline); + } break; + default: { + // not supported for this driver + return 0; + } break; + } +} + uint32_t RenderingDeviceVulkan::get_captured_timestamps_count() const { return frames[frame].timestamp_result_count; } @@ -8086,7 +9332,7 @@ String RenderingDeviceVulkan::get_captured_timestamp_name(uint32_t p_index) cons return frames[frame].timestamp_result_names[p_index]; } -int RenderingDeviceVulkan::limit_get(Limit p_limit) { +uint64_t RenderingDeviceVulkan::limit_get(Limit p_limit) { switch (p_limit) { case LIMIT_MAX_BOUND_UNIFORM_SETS: return limits.maxBoundDescriptorSets; @@ -8187,19 +9433,23 @@ void RenderingDeviceVulkan::finalize() { List<RID> owned; texture_owner.get_owned_list(&owned); if (owned.size()) { - WARN_PRINT(itos(owned.size()) + " RIDs of type 'Texture' were leaked."); + if (owned.size() == 1) { + WARN_PRINT("1 RID of type \"Texture\" was leaked."); + } else { + WARN_PRINT(vformat("%d RIDs of type \"Texture\" were leaked.", owned.size())); + } //free shared first for (List<RID>::Element *E = owned.front(); E;) { List<RID>::Element *N = E->next(); if (texture_is_shared(E->get())) { free(E->get()); - owned.erase(E->get()); + owned.erase(E); } E = N; } //free non shared second, this will avoid an error trying to free unexisting textures due to dependencies. - for (List<RID>::Element *E = owned.front(); E; E = E->next()) { - free(E->get()); + for (const RID &E : owned) { + free(E); } } } @@ -8226,17 +9476,22 @@ void RenderingDeviceVulkan::finalize() { for (int i = 0; i < staging_buffer_blocks.size(); i++) { vmaDestroyBuffer(allocator, staging_buffer_blocks[i].buffer, staging_buffer_blocks[i].allocation); } + while (small_allocs_pools.size()) { + HashMap<uint32_t, VmaPool>::Iterator E = small_allocs_pools.begin(); + vmaDestroyPool(allocator, E->value); + small_allocs_pools.remove(E); + } vmaDestroyAllocator(allocator); while (vertex_formats.size()) { - Map<VertexFormatID, VertexDescriptionCache>::Element *temp = vertex_formats.front(); - memdelete_arr(temp->get().bindings); - memdelete_arr(temp->get().attributes); - vertex_formats.erase(temp); + HashMap<VertexFormatID, VertexDescriptionCache>::Iterator temp = vertex_formats.begin(); + memdelete_arr(temp->value.bindings); + memdelete_arr(temp->value.attributes); + vertex_formats.remove(temp); } - for (int i = 0; i < framebuffer_formats.size(); i++) { - vkDestroyRenderPass(device, framebuffer_formats[i].render_pass, nullptr); + for (KeyValue<FramebufferFormatID, FramebufferFormat> &E : framebuffer_formats) { + vkDestroyRenderPass(device, E.value.render_pass, nullptr); } framebuffer_formats.clear(); @@ -8253,6 +9508,7 @@ RenderingDevice *RenderingDeviceVulkan::create_local_device() { } RenderingDeviceVulkan::RenderingDeviceVulkan() { + device_capabilities.device_family = DEVICE_VULKAN; } RenderingDeviceVulkan::~RenderingDeviceVulkan() { |