diff options
author | reduz <reduzio@gmail.com> | 2021-06-24 10:58:36 -0300 |
---|---|---|
committer | reduz <reduzio@gmail.com> | 2021-06-24 15:59:15 -0300 |
commit | bde6f5eed1cec3639e5c81d1e5ee34ecf3fe7486 (patch) | |
tree | 8f339026a1db2addb5bd728d605b144d5258a320 | |
parent | 8ccadfabf21fadc01fda157988c66c6cecf38555 (diff) |
Implement Framebuffer Subpass support
* Required for better optimizing mobile platforms
* Will be used by the Vulkan mobile renderer.
-rw-r--r-- | doc/classes/RDFramebufferPass.xml | 35 | ||||
-rw-r--r-- | doc/classes/RenderingDevice.xml | 50 | ||||
-rw-r--r-- | drivers/vulkan/rendering_device_vulkan.cpp | 781 | ||||
-rw-r--r-- | drivers/vulkan/rendering_device_vulkan.h | 109 | ||||
-rw-r--r-- | servers/register_server_types.cpp | 1 | ||||
-rw-r--r-- | servers/rendering/renderer_rd/pipeline_cache_rd.cpp | 7 | ||||
-rw-r--r-- | servers/rendering/renderer_rd/pipeline_cache_rd.h | 9 | ||||
-rw-r--r-- | servers/rendering/rendering_device.cpp | 72 | ||||
-rw-r--r-- | servers/rendering/rendering_device.h | 30 | ||||
-rw-r--r-- | servers/rendering/rendering_device_binds.h | 28 |
10 files changed, 868 insertions, 254 deletions
diff --git a/doc/classes/RDFramebufferPass.xml b/doc/classes/RDFramebufferPass.xml new file mode 100644 index 0000000000..c26c41f93f --- /dev/null +++ b/doc/classes/RDFramebufferPass.xml @@ -0,0 +1,35 @@ +<?xml version="1.0" encoding="UTF-8" ?> +<class name="RDFramebufferPass" inherits="RefCounted" version="4.0"> + <brief_description> + Framebuffer pass attachment description. + </brief_description> + <description> + This class contains the list of attachment descriptions for a framebuffer pass. Each points with an index to a previously supplied list of texture attachments. + Multipass framebuffers can optimize some configurations in mobile, on desktop they provide little to no advantage. + </description> + <tutorials> + </tutorials> + <methods> + </methods> + <members> + <member name="color_attachments" type="PackedInt32Array" setter="set_color_attachments" getter="get_color_attachments" default="PackedInt32Array()"> + Color attachments in order starting from 0. If this attachment is not used by the shader, pass ATTACHMENT_UNUSED to skip. + </member> + <member name="depth_attachment" type="int" setter="set_depth_attachment" getter="get_depth_attachment" default="-1"> + Depth attachment. ATTACHMENT_UNUSED should be used if no depth buffer is required for this pass. + </member> + <member name="input_attachments" type="PackedInt32Array" setter="set_input_attachments" getter="get_input_attachments" default="PackedInt32Array()"> + Used for multipass framebuffers (more than one render pass). Converts an attachment to an input. Make sure to also supply it properly in the [RDUniform] for the uniform set. + </member> + <member name="preserve_attachments" type="PackedInt32Array" setter="set_preserve_attachments" getter="get_preserve_attachments" default="PackedInt32Array()"> + Attachments to preserve in this pass (otherwise they are erased). + </member> + <member name="resolve_attachments" type="PackedInt32Array" setter="set_resolve_attachments" getter="get_resolve_attachments" default="PackedInt32Array()"> + If the color attachments are multisampled, non-multisampled resolve attachments can be provided. + </member> + </members> + <constants> + <constant name="ATTACHMENT_UNUSED" value="-1"> + </constant> + </constants> +</class> diff --git a/doc/classes/RenderingDevice.xml b/doc/classes/RenderingDevice.xml index dc56e6fd5d..b66b945025 100644 --- a/doc/classes/RenderingDevice.xml +++ b/doc/classes/RenderingDevice.xml @@ -341,13 +341,29 @@ <description> </description> </method> + <method name="draw_list_switch_to_next_pass"> + <return type="int"> + </return> + <description> + </description> + </method> + <method name="draw_list_switch_to_next_pass_split"> + <return type="PackedInt64Array"> + </return> + <argument index="0" name="splits" type="int"> + </argument> + <description> + </description> + </method> <method name="framebuffer_create"> <return type="RID"> </return> - <argument index="0" name="textures" type="Array"> + <argument index="0" name="textures" type="RID[]"> </argument> <argument index="1" name="validate_with_format" type="int" default="-1"> </argument> + <argument index="2" name="view_count" type="int" default="1"> + </argument> <description> </description> </method> @@ -363,11 +379,27 @@ <description> </description> </method> + <method name="framebuffer_create_multipass"> + <return type="RID"> + </return> + <argument index="0" name="textures" type="RID[]"> + </argument> + <argument index="1" name="passes" type="RDFramebufferPass[]"> + </argument> + <argument index="2" name="validate_with_format" type="int" default="-1"> + </argument> + <argument index="3" name="view_count" type="int" default="1"> + </argument> + <description> + </description> + </method> <method name="framebuffer_format_create"> <return type="int"> </return> <argument index="0" name="attachments" type="RDAttachmentFormat[]"> </argument> + <argument index="1" name="view_count" type="int" default="1"> + </argument> <description> </description> </method> @@ -379,11 +411,25 @@ <description> </description> </method> + <method name="framebuffer_format_create_multipass"> + <return type="int"> + </return> + <argument index="0" name="attachments" type="RDAttachmentFormat[]"> + </argument> + <argument index="1" name="passes" type="RDFramebufferPass[]"> + </argument> + <argument index="2" name="view_count" type="int" default="1"> + </argument> + <description> + </description> + </method> <method name="framebuffer_format_get_texture_samples"> <return type="int" enum="RenderingDevice.TextureSamples"> </return> <argument index="0" name="format" type="int"> </argument> + <argument index="1" name="render_pass" type="int" default="0"> + </argument> <description> </description> </method> @@ -524,6 +570,8 @@ </argument> <argument index="8" name="dynamic_state_flags" type="int" default="0"> </argument> + <argument index="9" name="for_render_pass" type="int" default="0"> + </argument> <description> </description> </method> diff --git a/drivers/vulkan/rendering_device_vulkan.cpp b/drivers/vulkan/rendering_device_vulkan.cpp index c69b516f37..1539962803 100644 --- a/drivers/vulkan/rendering_device_vulkan.cpp +++ b/drivers/vulkan/rendering_device_vulkan.cpp @@ -3244,12 +3244,7 @@ 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, uint32_t p_view_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 | @@ -3262,27 +3257,31 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF { 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)), + Vector<VkAttachmentDescription> attachments; + + for (int i = 0; i < p_attachments.size(); i++) { + 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_RESOLVE_ATTACHMENT_BIT)), VK_NULL_HANDLE, "Texture format for index (" + itos(i) + ") requires an attachment (depth, stencil or resolve) bit set."); VkAttachmentDescription description = {}; description.flags = 0; - description.format = vulkan_formats[p_format[i].format]; - description.samples = rasterization_sample_count[p_format[i].samples]; + 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 // 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; @@ -3291,11 +3290,11 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF 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 +3307,11 @@ 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 +3325,11 @@ 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 +3345,14 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF } } - switch (is_depth_stencil ? p_final_depth_action : p_final_color_action) { + switch (is_depth ? p_final_depth_action : p_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 +3365,11 @@ 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 +3380,11 @@ 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 +3400,191 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF } } + attachment_last_pass[i] = -1; + attachments.push_back(description); + } + + 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; - VkAttachmentReference reference; - reference.attachment = i; + 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()); - if (p_format[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { - reference.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; + 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; + 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_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 input 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."); + reference.attachment = 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) + "), it's marked as depth, but it's not usable as resolve 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."); + 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; + reference.layout = 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; + } + + 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) + ")."); + 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."); + + attachment_last_pass[attachment] = i; + + preserve_references.push_back(attachment); + } + + 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; + depth_stencil_reference.layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; + attachment_last_pass[attachment] = i; + + if (!is_multisample_first) { + ERR_FAIL_COND_V_MSG(texture_samples != p_attachments[attachment].samples, VK_NULL_HANDLE, "Invalid framebuffer depth format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), if an attachment is marked as multisample, all of them should be multisample and use the same number of samples including the depth."); + } + + } else { + depth_stencil_reference.attachment = VK_ATTACHMENT_UNUSED; + depth_stencil_reference.layout = VK_IMAGE_LAYOUT_UNDEFINED; + } + + 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 { - 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."); + 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 @@ -3433,41 +3596,28 @@ 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; + } const uint32_t view_mask = (1 << p_view_count) - 1; const uint32_t correlation_mask = (1 << p_view_count) - 1; @@ -3498,17 +3648,29 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF 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, 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); @@ -3517,8 +3679,8 @@ RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_c 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, p_view_count); //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; @@ -3528,9 +3690,8 @@ 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; @@ -3538,6 +3699,7 @@ RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_c 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); if (E) { @@ -3583,18 +3745,18 @@ 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) { +RenderingDevice::TextureSamples RenderingDeviceVulkan::framebuffer_format_get_texture_samples(FramebufferFormatID p_format, uint32_t p_pass) { Map<FramebufferFormatID, FramebufferFormat>::Element *E = framebuffer_formats.find(p_format); ERR_FAIL_COND_V(!E, TEXTURE_SAMPLES_1); + ERR_FAIL_COND_V(p_pass >= uint32_t(E->get().pass_samples.size()), TEXTURE_SAMPLES_1); - return E->get().samples; + return E->get().pass_samples[p_pass]; } /***********************/ @@ -3615,6 +3777,30 @@ RID RenderingDeviceVulkan::framebuffer_create_empty(const Size2i &p_size, Textur RID RenderingDeviceVulkan::framebuffer_create(const Vector<RID> &p_texture_attachments, FramebufferFormatID p_format_check, uint32_t p_view_count) { _THREAD_SAFE_METHOD_ + FramebufferPass pass; + + for (int i = 0; i < p_texture_attachments.size(); i++) { + Texture *texture = texture_owner.getornull(p_texture_attachments[i]); + ERR_FAIL_COND_V_MSG(!texture, RID(), "Texture index supplied for framebuffer (" + itos(i) + ") is not a valid texture."); + + ERR_FAIL_COND_V_MSG(texture->layers != p_view_count, RID(), "Layers of our texture doesn't match view count for this framebuffer"); + + if (texture->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_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; Size2i size; @@ -3639,7 +3825,7 @@ RID RenderingDeviceVulkan::framebuffer_create(const Vector<RID> &p_texture_attac attachments.push_back(af); } - FramebufferFormatID format_id = framebuffer_format_create(attachments, p_view_count); + FramebufferFormatID format_id = framebuffer_format_create_multipass(attachments, p_passes, p_view_count); if (format_id == INVALID_ID) { return RID(); } @@ -4401,8 +4587,9 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed obtaining output variables."); for (uint32_t j = 0; j < ov_count; j++) { - if (output_vars[j]) { - fragment_outputs = MAX(fragment_outputs, output_vars[j]->location + 1); + const SpvReflectInterfaceVariable *refvar = output_vars[j]; + if (refvar != nullptr && refvar->built_in != SpvBuiltInFragDepth) { + fragment_outputs |= 1 << refvar->location; } } } @@ -4453,7 +4640,7 @@ 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]; @@ -5194,6 +5381,49 @@ 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.ids.size() != 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.ids.size()) + ")."); + } else { + ERR_FAIL_V_MSG(RID(), "InputAttachment (binding: " + itos(uniform.binding) + ") should provide one ID referencing a texture (IDs provided: " + itos(uniform.ids.size()) + ")."); + } + } + + Vector<VkDescriptorImageInfo> image_info; + + for (int j = 0; j < uniform.ids.size(); j++) { + Texture *texture = texture_owner.getornull(uniform.ids[j]); + + ERR_FAIL_COND_V_MSG(!texture, RID(), + "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.getornull(texture->owner); + ERR_FAIL_COND_V(!texture, RID()); //bug, should never happen + } + + img_info.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; + + image_info.push_back(img_info); + } + + write.dstArrayElement = 0; + write.descriptorCount = uniform.ids.size(); + write.descriptorType = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT; + write.pImageInfo = image_infos.push_back(image_info)->get().ptr(); + write.pBufferInfo = nullptr; + write.pTexelBufferView = nullptr; + + type_size = uniform.ids.size(); } break; default: { } @@ -5404,7 +5634,7 @@ 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) { _THREAD_SAFE_METHOD_ //needs a shader @@ -5423,8 +5653,16 @@ 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; @@ -5609,44 +5847,53 @@ 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; + { + const FramebufferPass &pass = fb_format.E->key().passes[p_for_render_pass]; + + for (int i = 0; i < pass.color_attachments.size(); i++) { + if (pass.color_attachments[i] != FramebufferPass::ATTACHMENT_UNUSED) { + int idx = attachment_states.size(); + + ERR_FAIL_INDEX_V(idx, p_blend_state.attachments.size(), RID()); + VkPipelineColorBlendAttachmentState state; + state.blendEnable = p_blend_state.attachments[idx].enable_blend; + + ERR_FAIL_INDEX_V(p_blend_state.attachments[idx].src_color_blend_factor, BLEND_FACTOR_MAX, RID()); + state.srcColorBlendFactor = blend_factors[p_blend_state.attachments[idx].src_color_blend_factor]; + ERR_FAIL_INDEX_V(p_blend_state.attachments[idx].dst_color_blend_factor, BLEND_FACTOR_MAX, RID()); + state.dstColorBlendFactor = blend_factors[p_blend_state.attachments[idx].dst_color_blend_factor]; + ERR_FAIL_INDEX_V(p_blend_state.attachments[idx].color_blend_op, BLEND_OP_MAX, RID()); + state.colorBlendOp = blend_operations[p_blend_state.attachments[idx].color_blend_op]; + + ERR_FAIL_INDEX_V(p_blend_state.attachments[idx].src_alpha_blend_factor, BLEND_FACTOR_MAX, RID()); + state.srcAlphaBlendFactor = blend_factors[p_blend_state.attachments[idx].src_alpha_blend_factor]; + ERR_FAIL_INDEX_V(p_blend_state.attachments[idx].dst_alpha_blend_factor, BLEND_FACTOR_MAX, RID()); + state.dstAlphaBlendFactor = blend_factors[p_blend_state.attachments[idx].dst_alpha_blend_factor]; + ERR_FAIL_INDEX_V(p_blend_state.attachments[idx].alpha_blend_op, BLEND_OP_MAX, RID()); + state.alphaBlendOp = blend_operations[p_blend_state.attachments[idx].alpha_blend_op]; + + state.colorWriteMask = 0; + if (p_blend_state.attachments[idx].write_r) { + state.colorWriteMask |= VK_COLOR_COMPONENT_R_BIT; + } + if (p_blend_state.attachments[idx].write_g) { + state.colorWriteMask |= VK_COLOR_COMPONENT_G_BIT; + } + if (p_blend_state.attachments[idx].write_b) { + state.colorWriteMask |= VK_COLOR_COMPONENT_B_BIT; + } + if (p_blend_state.attachments[idx].write_a) { + state.colorWriteMask |= VK_COLOR_COMPONENT_A_BIT; + } - for (int i = 0; i < p_blend_state.attachments.size(); i++) { - VkPipelineColorBlendAttachmentState state; - state.blendEnable = p_blend_state.attachments[i].enable_blend; - - ERR_FAIL_INDEX_V(p_blend_state.attachments[i].src_color_blend_factor, BLEND_FACTOR_MAX, RID()); - state.srcColorBlendFactor = blend_factors[p_blend_state.attachments[i].src_color_blend_factor]; - ERR_FAIL_INDEX_V(p_blend_state.attachments[i].dst_color_blend_factor, BLEND_FACTOR_MAX, RID()); - state.dstColorBlendFactor = blend_factors[p_blend_state.attachments[i].dst_color_blend_factor]; - ERR_FAIL_INDEX_V(p_blend_state.attachments[i].color_blend_op, BLEND_OP_MAX, RID()); - state.colorBlendOp = blend_operations[p_blend_state.attachments[i].color_blend_op]; - - ERR_FAIL_INDEX_V(p_blend_state.attachments[i].src_alpha_blend_factor, BLEND_FACTOR_MAX, RID()); - state.srcAlphaBlendFactor = blend_factors[p_blend_state.attachments[i].src_alpha_blend_factor]; - ERR_FAIL_INDEX_V(p_blend_state.attachments[i].dst_alpha_blend_factor, BLEND_FACTOR_MAX, RID()); - state.dstAlphaBlendFactor = blend_factors[p_blend_state.attachments[i].dst_alpha_blend_factor]; - ERR_FAIL_INDEX_V(p_blend_state.attachments[i].alpha_blend_op, BLEND_OP_MAX, RID()); - state.alphaBlendOp = blend_operations[p_blend_state.attachments[i].alpha_blend_op]; - - state.colorWriteMask = 0; - if (p_blend_state.attachments[i].write_r) { - state.colorWriteMask |= VK_COLOR_COMPONENT_R_BIT; - } - if (p_blend_state.attachments[i].write_g) { - state.colorWriteMask |= VK_COLOR_COMPONENT_G_BIT; - } - if (p_blend_state.attachments[i].write_b) { - state.colorWriteMask |= VK_COLOR_COMPONENT_B_BIT; - } - if (p_blend_state.attachments[i].write_a) { - state.colorWriteMask |= VK_COLOR_COMPONENT_A_BIT; + attachment_states.push_back(state); + idx++; + }; } - attachment_states.push_back(state); - }; + ERR_FAIL_COND_V(attachment_states.size() != p_blend_state.attachments.size(), RID()); + } color_blend_state_create_info.attachmentCount = attachment_states.size(); color_blend_state_create_info.pAttachments = attachment_states.ptr(); @@ -5719,7 +5966,7 @@ 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; @@ -5736,6 +5983,7 @@ 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; @@ -5874,13 +6122,14 @@ 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; + + 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; @@ -5888,8 +6137,8 @@ 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; @@ -5929,7 +6178,7 @@ 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; @@ -5941,7 +6190,7 @@ Error RenderingDeviceVulkan::_draw_list_setup_framebuffer(Framebuffer *p_framebu //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, nullptr, p_framebuffer->view_count); + 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; @@ -5965,11 +6214,14 @@ 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; } @@ -6159,15 +6411,23 @@ 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.getornull(framebuffer->texture_ids[i]); + + if (texture->usage_flags & TEXTURE_USAGE_COLOR_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; @@ -6177,13 +6437,14 @@ 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); @@ -6207,7 +6468,6 @@ 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; } @@ -6249,47 +6509,23 @@ 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.getornull(framebuffer->texture_ids[i]); - split_draw_list_allocators.write[i].command_buffers.push_back(command_buffer); + if (texture->usage_flags & TEXTURE_USAGE_COLOR_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; @@ -6299,53 +6535,24 @@ 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; - - 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_current_subpass = 0; - 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; @@ -6354,7 +6561,7 @@ 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; @@ -6362,10 +6569,8 @@ 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; @@ -6410,7 +6615,7 @@ void RenderingDeviceVulkan::draw_list_bind_render_pipeline(DrawListID p_list, RI const RenderPipeline *pipeline = render_pipeline_owner.getornull(p_render_pipeline); ERR_FAIL_COND(!pipeline); #ifdef DEBUG_ENABLED - ERR_FAIL_COND(pipeline->validation.framebuffer_format != dl->validation.framebuffer_format); + 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) { @@ -6749,30 +6954,162 @@ 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_ +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); - ERR_FAIL_COND_MSG(!draw_list, "Immediate draw list is already inactive."); + 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) { + 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(); + 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; } +} + +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]); diff --git a/drivers/vulkan/rendering_device_vulkan.h b/drivers/vulkan/rendering_device_vulkan.h index f4fe9cf956..9f5830103d 100644 --- a/drivers/vulkan/rendering_device_vulkan.h +++ b/drivers/vulkan/rendering_device_vulkan.h @@ -234,12 +234,87 @@ class RenderingDeviceVulkan : public RenderingDevice { struct FramebufferFormatKey { Vector<AttachmentFormat> attachments; + Vector<FramebufferPass> passes; uint32_t view_count = 1; bool operator<(const FramebufferFormatKey &p_key) const { if (view_count != p_key.view_count) { return view_count < p_key.view_count; } + uint32_t pass_size = passes.size(); + uint32_t key_pass_size = p_key.passes.size(); + if (pass_size != key_pass_size) { + return pass_size < key_pass_size; + } + const FramebufferPass *pass_ptr = passes.ptr(); + const FramebufferPass *key_pass_ptr = p_key.passes.ptr(); + + for (uint32_t i = 0; i < pass_size; i++) { + { //compare color attachments + uint32_t attachment_size = pass_ptr[i].color_attachments.size(); + uint32_t key_attachment_size = key_pass_ptr[i].color_attachments.size(); + if (attachment_size != key_attachment_size) { + return attachment_size < key_attachment_size; + } + const int32_t *pass_attachment_ptr = pass_ptr[i].color_attachments.ptr(); + const int32_t *key_pass_attachment_ptr = key_pass_ptr[i].color_attachments.ptr(); + + for (uint32_t j = 0; j < attachment_size; j++) { + if (pass_attachment_ptr[j] != key_pass_attachment_ptr[j]) { + return pass_attachment_ptr[j] < key_pass_attachment_ptr[j]; + } + } + } + { //compare input attachments + uint32_t attachment_size = pass_ptr[i].input_attachments.size(); + uint32_t key_attachment_size = key_pass_ptr[i].input_attachments.size(); + if (attachment_size != key_attachment_size) { + return attachment_size < key_attachment_size; + } + const int32_t *pass_attachment_ptr = pass_ptr[i].input_attachments.ptr(); + const int32_t *key_pass_attachment_ptr = key_pass_ptr[i].input_attachments.ptr(); + + for (uint32_t j = 0; j < attachment_size; j++) { + if (pass_attachment_ptr[j] != key_pass_attachment_ptr[j]) { + return pass_attachment_ptr[j] < key_pass_attachment_ptr[j]; + } + } + } + { //compare resolve attachments + uint32_t attachment_size = pass_ptr[i].resolve_attachments.size(); + uint32_t key_attachment_size = key_pass_ptr[i].resolve_attachments.size(); + if (attachment_size != key_attachment_size) { + return attachment_size < key_attachment_size; + } + const int32_t *pass_attachment_ptr = pass_ptr[i].resolve_attachments.ptr(); + const int32_t *key_pass_attachment_ptr = key_pass_ptr[i].resolve_attachments.ptr(); + + for (uint32_t j = 0; j < attachment_size; j++) { + if (pass_attachment_ptr[j] != key_pass_attachment_ptr[j]) { + return pass_attachment_ptr[j] < key_pass_attachment_ptr[j]; + } + } + } + { //compare preserve attachments + uint32_t attachment_size = pass_ptr[i].preserve_attachments.size(); + uint32_t key_attachment_size = key_pass_ptr[i].preserve_attachments.size(); + if (attachment_size != key_attachment_size) { + return attachment_size < key_attachment_size; + } + const int32_t *pass_attachment_ptr = pass_ptr[i].preserve_attachments.ptr(); + const int32_t *key_pass_attachment_ptr = key_pass_ptr[i].preserve_attachments.ptr(); + + for (uint32_t j = 0; j < attachment_size; j++) { + if (pass_attachment_ptr[j] != key_pass_attachment_ptr[j]) { + return pass_attachment_ptr[j] < key_pass_attachment_ptr[j]; + } + } + } + if (pass_ptr[i].depth_attachment != key_pass_ptr[i].depth_attachment) { + return pass_ptr[i].depth_attachment < key_pass_ptr[i].depth_attachment; + } + } + int as = attachments.size(); int bs = p_key.attachments.size(); if (as != bs) { @@ -266,16 +341,14 @@ class RenderingDeviceVulkan : public RenderingDevice { } }; - VkRenderPass _render_pass_create(const Vector<AttachmentFormat> &p_format, InitialAction p_initial_action, FinalAction p_final_action, InitialAction p_initial_depth_action, FinalAction p_final_depthcolor_action, int *r_color_attachment_count = nullptr, uint32_t p_view_count = 1); - + VkRenderPass _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 = 1, Vector<TextureSamples> *r_samples = nullptr); // This is a cache and it's never freed, it ensures // IDs for a given format are always unique. Map<FramebufferFormatKey, FramebufferFormatID> framebuffer_format_cache; struct FramebufferFormat { const Map<FramebufferFormatKey, FramebufferFormatID>::Element *E; VkRenderPass render_pass = VK_NULL_HANDLE; //here for constructing shaders, never used, see section (7.2. Render Pass Compatibility from Vulkan spec) - int color_attachments = 0; //used for pipeline validation - TextureSamples samples; + Vector<TextureSamples> pass_samples; uint32_t view_count = 1; // number of views }; @@ -289,6 +362,7 @@ class RenderingDeviceVulkan : public RenderingDevice { InitialAction initial_depth_action; FinalAction final_depth_action; uint32_t view_count; + bool operator<(const VersionKey &p_key) const { if (initial_color_action == p_key.initial_color_action) { if (final_color_action == p_key.final_color_action) { @@ -316,6 +390,7 @@ class RenderingDeviceVulkan : public RenderingDevice { struct Version { VkFramebuffer framebuffer = VK_NULL_HANDLE; VkRenderPass render_pass = VK_NULL_HANDLE; //this one is owned + uint32_t subpass_count = 1; }; Map<VersionKey, Version> framebuffers; @@ -536,7 +611,7 @@ class RenderingDeviceVulkan : public RenderingDevice { }; uint32_t vertex_input_mask = 0; //inputs used, this is mostly for validation - int fragment_outputs = 0; + uint32_t fragment_output_mask = 0; struct PushConstant { uint32_t push_constant_size = 0; @@ -680,6 +755,7 @@ class RenderingDeviceVulkan : public RenderingDevice { #ifdef DEBUG_ENABLED struct Validation { FramebufferFormatID framebuffer_format = 0; + uint32_t render_pass = 0; uint32_t dynamic_state = 0; VertexFormatID vertex_format = 0; bool uses_restart_indices = false; @@ -735,6 +811,7 @@ class RenderingDeviceVulkan : public RenderingDevice { struct DrawList { VkCommandBuffer command_buffer = VK_NULL_HANDLE; // If persistent, this is owned, otherwise it's shared with the ringbuffer. Rect2i viewport; + bool viewport_set = false; struct SetState { uint32_t pipeline_expected_format = 0; @@ -758,7 +835,6 @@ class RenderingDeviceVulkan : public RenderingDevice { #ifdef DEBUG_ENABLED struct Validation { bool active = true; // Means command buffer was not closed, so you can keep adding things. - FramebufferFormatID framebuffer_format = INVALID_ID; // Actual render pass values. uint32_t dynamic_state = 0; VertexFormatID vertex_format = INVALID_ID; @@ -794,7 +870,15 @@ class RenderingDeviceVulkan : public RenderingDevice { }; DrawList *draw_list = nullptr; // One for regular draw lists, multiple for split. + uint32_t draw_list_subpass_count = 0; uint32_t draw_list_count = 0; + VkRenderPass draw_list_render_pass; + VkFramebuffer draw_list_vkframebuffer; +#ifdef DEBUG_ENABLED + FramebufferFormatID draw_list_framebuffer_format = INVALID_ID; +#endif + uint32_t draw_list_current_subpass = 0; + bool draw_list_split = false; Vector<RID> draw_list_bound_textures; Vector<RID> draw_list_storage_textures; @@ -802,10 +886,12 @@ class RenderingDeviceVulkan : public RenderingDevice { bool draw_list_unbind_depth_textures = false; void _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); - Error _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 _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); Error _draw_list_render_pass_begin(Framebuffer *framebuffer, InitialAction p_initial_color_action, FinalAction p_final_color_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, const Vector<Color> &p_clear_colors, float p_clear_depth, uint32_t p_clear_stencil, Point2i viewport_offset, Point2i viewport_size, VkFramebuffer vkframebuffer, VkRenderPass render_pass, VkCommandBuffer command_buffer, VkSubpassContents subpass_contents, const Vector<RID> &p_storage_textures); _FORCE_INLINE_ DrawList *_get_draw_list_ptr(DrawListID p_id); Buffer *_get_buffer_from_owner(RID p_buffer, VkPipelineStageFlags &dst_stage_mask, VkAccessFlags &dst_access, uint32_t p_post_barrier); + Error _draw_list_allocate(const Rect2i &p_viewport, uint32_t p_splits, uint32_t p_subpass); + void _draw_list_free(Rect2i *r_last_viewport = nullptr); /**********************/ /**** COMPUTE LIST ****/ @@ -951,10 +1037,12 @@ public: /*********************/ virtual FramebufferFormatID framebuffer_format_create(const Vector<AttachmentFormat> &p_format, uint32_t p_view_count = 1); + virtual FramebufferFormatID framebuffer_format_create_multipass(const Vector<AttachmentFormat> &p_attachments, Vector<FramebufferPass> &p_passes, uint32_t p_view_count = 1); virtual FramebufferFormatID framebuffer_format_create_empty(TextureSamples p_samples = TEXTURE_SAMPLES_1); - virtual TextureSamples framebuffer_format_get_texture_samples(FramebufferFormatID p_format); + virtual TextureSamples framebuffer_format_get_texture_samples(FramebufferFormatID p_format, uint32_t p_pass = 0); virtual RID framebuffer_create(const Vector<RID> &p_texture_attachments, FramebufferFormatID p_format_check = INVALID_ID, uint32_t p_view_count = 1); + virtual RID framebuffer_create_multipass(const Vector<RID> &p_texture_attachments, Vector<FramebufferPass> &p_passes, FramebufferFormatID p_format_check = INVALID_ID, uint32_t p_view_count = 1); virtual RID framebuffer_create_empty(const Size2i &p_size, TextureSamples p_samples = TEXTURE_SAMPLES_1, FramebufferFormatID p_format_check = INVALID_ID); virtual FramebufferFormatID framebuffer_get_format(RID p_framebuffer); @@ -1005,7 +1093,7 @@ public: /**** RENDER PIPELINE ****/ /*************************/ - virtual RID 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 = 0); + virtual RID 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 = 0, uint32_t p_for_render_pass = 0); virtual bool render_pipeline_is_valid(RID p_pipeline); /**************************/ @@ -1044,6 +1132,9 @@ public: virtual void draw_list_enable_scissor(DrawListID p_list, const Rect2 &p_rect); virtual void draw_list_disable_scissor(DrawListID p_list); + virtual DrawListID draw_list_switch_to_next_pass(); + virtual Error draw_list_switch_to_next_pass_split(uint32_t p_splits, DrawListID *r_split_ids); + virtual void draw_list_end(uint32_t p_post_barrier = BARRIER_MASK_ALL); /***********************/ diff --git a/servers/register_server_types.cpp b/servers/register_server_types.cpp index deb230c4fb..fd11186751 100644 --- a/servers/register_server_types.cpp +++ b/servers/register_server_types.cpp @@ -194,6 +194,7 @@ void register_server_types() { ClassDB::register_class<RDTextureFormat>(); ClassDB::register_class<RDTextureView>(); ClassDB::register_class<RDAttachmentFormat>(); + ClassDB::register_class<RDFramebufferPass>(); ClassDB::register_class<RDSamplerState>(); ClassDB::register_class<RDVertexAttribute>(); ClassDB::register_class<RDUniform>(); diff --git a/servers/rendering/renderer_rd/pipeline_cache_rd.cpp b/servers/rendering/renderer_rd/pipeline_cache_rd.cpp index b2b919c40e..22888ddbe5 100644 --- a/servers/rendering/renderer_rd/pipeline_cache_rd.cpp +++ b/servers/rendering/renderer_rd/pipeline_cache_rd.cpp @@ -31,20 +31,21 @@ #include "pipeline_cache_rd.h" #include "core/os/memory.h" -RID PipelineCacheRD::_generate_version(RD::VertexFormatID p_vertex_format_id, RD::FramebufferFormatID p_framebuffer_format_id, bool p_wireframe) { +RID PipelineCacheRD::_generate_version(RD::VertexFormatID p_vertex_format_id, RD::FramebufferFormatID p_framebuffer_format_id, bool p_wireframe, uint32_t p_render_pass) { RD::PipelineMultisampleState multisample_state_version = multisample_state; - multisample_state_version.sample_count = RD::get_singleton()->framebuffer_format_get_texture_samples(p_framebuffer_format_id); + multisample_state_version.sample_count = RD::get_singleton()->framebuffer_format_get_texture_samples(p_framebuffer_format_id, p_render_pass); RD::PipelineRasterizationState raster_state_version = rasterization_state; raster_state_version.wireframe = p_wireframe; - RID pipeline = RD::get_singleton()->render_pipeline_create(shader, p_framebuffer_format_id, p_vertex_format_id, render_primitive, raster_state_version, multisample_state_version, depth_stencil_state, blend_state, dynamic_state_flags); + RID pipeline = RD::get_singleton()->render_pipeline_create(shader, p_framebuffer_format_id, p_vertex_format_id, render_primitive, raster_state_version, multisample_state_version, depth_stencil_state, blend_state, dynamic_state_flags, p_render_pass); ERR_FAIL_COND_V(pipeline.is_null(), RID()); versions = (Version *)memrealloc(versions, sizeof(Version) * (version_count + 1)); versions[version_count].framebuffer_id = p_framebuffer_format_id; versions[version_count].vertex_id = p_vertex_format_id; versions[version_count].wireframe = p_wireframe; versions[version_count].pipeline = pipeline; + versions[version_count].render_pass = p_render_pass; version_count++; return pipeline; } diff --git a/servers/rendering/renderer_rd/pipeline_cache_rd.h b/servers/rendering/renderer_rd/pipeline_cache_rd.h index b1c8f21ecc..387a8a038f 100644 --- a/servers/rendering/renderer_rd/pipeline_cache_rd.h +++ b/servers/rendering/renderer_rd/pipeline_cache_rd.h @@ -50,6 +50,7 @@ class PipelineCacheRD { struct Version { RD::VertexFormatID vertex_id; RD::FramebufferFormatID framebuffer_id; + uint32_t render_pass; bool wireframe; RID pipeline; }; @@ -57,7 +58,7 @@ class PipelineCacheRD { Version *versions; uint32_t version_count; - RID _generate_version(RD::VertexFormatID p_vertex_format_id, RD::FramebufferFormatID p_framebuffer_format_id, bool p_wireframe); + RID _generate_version(RD::VertexFormatID p_vertex_format_id, RD::FramebufferFormatID p_framebuffer_format_id, bool p_wireframe, uint32_t p_render_pass); void _clear(); @@ -65,7 +66,7 @@ public: void setup(RID p_shader, RD::RenderPrimitive p_primitive, const RD::PipelineRasterizationState &p_rasterization_state, RD::PipelineMultisampleState p_multisample, const RD::PipelineDepthStencilState &p_depth_stencil_state, const RD::PipelineColorBlendState &p_blend_state, int p_dynamic_state_flags = 0); void update_shader(RID p_shader); - _FORCE_INLINE_ RID get_render_pipeline(RD::VertexFormatID p_vertex_format_id, RD::FramebufferFormatID p_framebuffer_format_id, bool p_wireframe = false) { + _FORCE_INLINE_ RID get_render_pipeline(RD::VertexFormatID p_vertex_format_id, RD::FramebufferFormatID p_framebuffer_format_id, bool p_wireframe = false, uint32_t p_render_pass = 0) { #ifdef DEBUG_ENABLED ERR_FAIL_COND_V_MSG(shader.is_null(), RID(), "Attempted to use an unused shader variant (shader is null),"); @@ -74,13 +75,13 @@ public: spin_lock.lock(); RID result; for (uint32_t i = 0; i < version_count; i++) { - if (versions[i].vertex_id == p_vertex_format_id && versions[i].framebuffer_id == p_framebuffer_format_id && versions[i].wireframe == p_wireframe) { + if (versions[i].vertex_id == p_vertex_format_id && versions[i].framebuffer_id == p_framebuffer_format_id && versions[i].wireframe == p_wireframe && versions[i].render_pass == p_render_pass) { result = versions[i].pipeline; spin_lock.unlock(); return result; } } - result = _generate_version(p_vertex_format_id, p_framebuffer_format_id, p_wireframe); + result = _generate_version(p_vertex_format_id, p_framebuffer_format_id, p_wireframe, p_render_pass); spin_lock.unlock(); return result; } diff --git a/servers/rendering/rendering_device.cpp b/servers/rendering/rendering_device.cpp index ddf2ee9fe3..c52d97a3de 100644 --- a/servers/rendering/rendering_device.cpp +++ b/servers/rendering/rendering_device.cpp @@ -98,7 +98,7 @@ RID RenderingDevice::_texture_create_shared_from_slice(const Ref<RDTextureView> return texture_create_shared_from_slice(p_view->base, p_with_texture, p_layer, p_mipmap, p_slice_type); } -RenderingDevice::FramebufferFormatID RenderingDevice::_framebuffer_format_create(const TypedArray<RDAttachmentFormat> &p_attachments) { +RenderingDevice::FramebufferFormatID RenderingDevice::_framebuffer_format_create(const TypedArray<RDAttachmentFormat> &p_attachments, uint32_t p_view_count) { Vector<AttachmentFormat> attachments; attachments.resize(p_attachments.size()); @@ -107,12 +107,43 @@ RenderingDevice::FramebufferFormatID RenderingDevice::_framebuffer_format_create ERR_FAIL_COND_V(af.is_null(), INVALID_FORMAT_ID); attachments.write[i] = af->base; } - return framebuffer_format_create(attachments); + return framebuffer_format_create(attachments, p_view_count); } -RID RenderingDevice::_framebuffer_create(const Array &p_textures, FramebufferFormatID p_format_check) { +RenderingDevice::FramebufferFormatID RenderingDevice::_framebuffer_format_create_multipass(const TypedArray<RDAttachmentFormat> &p_attachments, const TypedArray<RDFramebufferPass> &p_passes, uint32_t p_view_count) { + Vector<AttachmentFormat> attachments; + attachments.resize(p_attachments.size()); + + for (int i = 0; i < p_attachments.size(); i++) { + Ref<RDAttachmentFormat> af = p_attachments[i]; + ERR_FAIL_COND_V(af.is_null(), INVALID_FORMAT_ID); + attachments.write[i] = af->base; + } + + Vector<FramebufferPass> passes; + for (int i = 0; i < p_passes.size(); i++) { + Ref<RDFramebufferPass> pass = p_passes[i]; + ERR_CONTINUE(pass.is_null()); + passes.push_back(pass->base); + } + + return framebuffer_format_create_multipass(attachments, passes, p_view_count); +} + +RID RenderingDevice::_framebuffer_create(const TypedArray<RID> &p_textures, FramebufferFormatID p_format_check, uint32_t p_view_count) { Vector<RID> textures = Variant(p_textures); - return framebuffer_create(textures, p_format_check); + return framebuffer_create(textures, p_format_check, p_view_count); +} + +RID RenderingDevice::_framebuffer_create_multipass(const TypedArray<RID> &p_textures, const TypedArray<RDFramebufferPass> &p_passes, FramebufferFormatID p_format_check, uint32_t p_view_count) { + Vector<RID> textures = Variant(p_textures); + Vector<FramebufferPass> passes; + for (int i = 0; i < p_passes.size(); i++) { + Ref<RDFramebufferPass> pass = p_passes[i]; + ERR_CONTINUE(pass.is_null()); + passes.push_back(pass->base); + } + return framebuffer_create_multipass(textures, passes, p_format_check, p_view_count); } RID RenderingDevice::_sampler_create(const Ref<RDSamplerState> &p_state) { @@ -190,7 +221,7 @@ Error RenderingDevice::_buffer_update(RID p_buffer, uint32_t p_offset, uint32_t return buffer_update(p_buffer, p_offset, p_size, p_data.ptr(), p_post_barrier); } -RID RenderingDevice::_render_pipeline_create(RID p_shader, FramebufferFormatID p_framebuffer_format, VertexFormatID p_vertex_format, RenderPrimitive p_render_primitive, const Ref<RDPipelineRasterizationState> &p_rasterization_state, const Ref<RDPipelineMultisampleState> &p_multisample_state, const Ref<RDPipelineDepthStencilState> &p_depth_stencil_state, const Ref<RDPipelineColorBlendState> &p_blend_state, int p_dynamic_state_flags) { +RID RenderingDevice::_render_pipeline_create(RID p_shader, FramebufferFormatID p_framebuffer_format, VertexFormatID p_vertex_format, RenderPrimitive p_render_primitive, const Ref<RDPipelineRasterizationState> &p_rasterization_state, const Ref<RDPipelineMultisampleState> &p_multisample_state, const Ref<RDPipelineDepthStencilState> &p_depth_stencil_state, const Ref<RDPipelineColorBlendState> &p_blend_state, int p_dynamic_state_flags, uint32_t p_for_render_pass) { PipelineRasterizationState rasterization_state; if (p_rasterization_state.is_valid()) { rasterization_state = p_rasterization_state->base; @@ -221,7 +252,7 @@ RID RenderingDevice::_render_pipeline_create(RID p_shader, FramebufferFormatID p } } - return render_pipeline_create(p_shader, p_framebuffer_format, p_vertex_format, p_render_primitive, rasterization_state, multisample_state, depth_stencil_state, color_blend_state, p_dynamic_state_flags); + return render_pipeline_create(p_shader, p_framebuffer_format, p_vertex_format, p_render_primitive, rasterization_state, multisample_state, depth_stencil_state, color_blend_state, p_dynamic_state_flags, p_for_render_pass); } Vector<int64_t> RenderingDevice::_draw_list_begin_split(RID p_framebuffer, uint32_t p_splits, InitialAction p_initial_color_action, FinalAction p_final_color_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, const Vector<Color> &p_clear_color_values, float p_clear_depth, uint32_t p_clear_stencil, const Rect2 &p_region, const TypedArray<RID> &p_storage_textures) { @@ -242,6 +273,22 @@ Vector<int64_t> RenderingDevice::_draw_list_begin_split(RID p_framebuffer, uint3 return split_ids; } +Vector<int64_t> RenderingDevice::_draw_list_switch_to_next_pass_split(uint32_t p_splits) { + Vector<DrawListID> splits; + splits.resize(p_splits); + + Error err = draw_list_switch_to_next_pass_split(p_splits, splits.ptrw()); + ERR_FAIL_COND_V(err != OK, Vector<int64_t>()); + + Vector<int64_t> split_ids; + split_ids.resize(splits.size()); + for (int i = 0; i < splits.size(); i++) { + split_ids.write[i] = splits[i]; + } + + return split_ids; +} + void RenderingDevice::_draw_list_set_push_constant(DrawListID p_list, const Vector<uint8_t> &p_data, uint32_t p_data_size) { ERR_FAIL_COND((uint32_t)p_data.size() > p_data_size); draw_list_set_push_constant(p_list, p_data.ptr(), p_data_size); @@ -269,10 +316,12 @@ void RenderingDevice::_bind_methods() { ClassDB::bind_method(D_METHOD("texture_clear", "texture", "color", "base_mipmap", "mipmap_count", "base_layer", "layer_count", "post_barrier"), &RenderingDevice::texture_clear, DEFVAL(BARRIER_MASK_ALL)); ClassDB::bind_method(D_METHOD("texture_resolve_multisample", "from_texture", "to_texture", "post_barrier"), &RenderingDevice::texture_resolve_multisample, DEFVAL(BARRIER_MASK_ALL)); - ClassDB::bind_method(D_METHOD("framebuffer_format_create", "attachments"), &RenderingDevice::_framebuffer_format_create); + ClassDB::bind_method(D_METHOD("framebuffer_format_create", "attachments", "view_count"), &RenderingDevice::_framebuffer_format_create, DEFVAL(1)); + ClassDB::bind_method(D_METHOD("framebuffer_format_create_multipass", "attachments", "passes", "view_count"), &RenderingDevice::_framebuffer_format_create_multipass, DEFVAL(1)); ClassDB::bind_method(D_METHOD("framebuffer_format_create_empty", "samples"), &RenderingDevice::framebuffer_format_create_empty, DEFVAL(TEXTURE_SAMPLES_1)); - ClassDB::bind_method(D_METHOD("framebuffer_format_get_texture_samples", "format"), &RenderingDevice::framebuffer_format_get_texture_samples); - ClassDB::bind_method(D_METHOD("framebuffer_create", "textures", "validate_with_format"), &RenderingDevice::_framebuffer_create, DEFVAL(INVALID_FORMAT_ID)); + ClassDB::bind_method(D_METHOD("framebuffer_format_get_texture_samples", "format", "render_pass"), &RenderingDevice::framebuffer_format_get_texture_samples, DEFVAL(0)); + ClassDB::bind_method(D_METHOD("framebuffer_create", "textures", "validate_with_format", "view_count"), &RenderingDevice::_framebuffer_create, DEFVAL(INVALID_FORMAT_ID), DEFVAL(1)); + ClassDB::bind_method(D_METHOD("framebuffer_create_multipass", "textures", "passes", "validate_with_format", "view_count"), &RenderingDevice::_framebuffer_create_multipass, DEFVAL(INVALID_FORMAT_ID), DEFVAL(1)); ClassDB::bind_method(D_METHOD("framebuffer_create_empty", "size", "samples", "validate_with_format"), &RenderingDevice::framebuffer_create_empty, DEFVAL(TEXTURE_SAMPLES_1), DEFVAL(INVALID_FORMAT_ID)); ClassDB::bind_method(D_METHOD("framebuffer_get_format", "framebuffer"), &RenderingDevice::framebuffer_get_format); @@ -299,7 +348,7 @@ void RenderingDevice::_bind_methods() { ClassDB::bind_method(D_METHOD("buffer_clear", "buffer", "offset", "size_bytes", "post_barrier"), &RenderingDevice::buffer_clear, DEFVAL(BARRIER_MASK_ALL)); ClassDB::bind_method(D_METHOD("buffer_get_data", "buffer"), &RenderingDevice::buffer_get_data); - ClassDB::bind_method(D_METHOD("render_pipeline_create", "shader", "framebuffer_format", "vertex_format", "primitive", "rasterization_state", "multisample_state", "stencil_state", "color_blend_state", "dynamic_state_flags"), &RenderingDevice::_render_pipeline_create, DEFVAL(0)); + ClassDB::bind_method(D_METHOD("render_pipeline_create", "shader", "framebuffer_format", "vertex_format", "primitive", "rasterization_state", "multisample_state", "stencil_state", "color_blend_state", "dynamic_state_flags", "for_render_pass"), &RenderingDevice::_render_pipeline_create, DEFVAL(0), DEFVAL(0)); ClassDB::bind_method(D_METHOD("render_pipeline_is_valid", "render_pipeline"), &RenderingDevice::render_pipeline_is_valid); ClassDB::bind_method(D_METHOD("compute_pipeline_create", "shader"), &RenderingDevice::compute_pipeline_create); @@ -325,6 +374,9 @@ void RenderingDevice::_bind_methods() { ClassDB::bind_method(D_METHOD("draw_list_enable_scissor", "draw_list", "rect"), &RenderingDevice::draw_list_enable_scissor, DEFVAL(Rect2i())); ClassDB::bind_method(D_METHOD("draw_list_disable_scissor", "draw_list"), &RenderingDevice::draw_list_disable_scissor); + ClassDB::bind_method(D_METHOD("draw_list_switch_to_next_pass"), &RenderingDevice::draw_list_switch_to_next_pass); + ClassDB::bind_method(D_METHOD("draw_list_switch_to_next_pass_split", "splits"), &RenderingDevice::_draw_list_switch_to_next_pass_split); + ClassDB::bind_method(D_METHOD("draw_list_end", "post_barrier"), &RenderingDevice::draw_list_end, DEFVAL(BARRIER_MASK_ALL)); ClassDB::bind_method(D_METHOD("compute_list_begin", "allow_draw_overlap"), &RenderingDevice::compute_list_begin, DEFVAL(false)); diff --git a/servers/rendering/rendering_device.h b/servers/rendering/rendering_device.h index c13dc01dd7..be7e127491 100644 --- a/servers/rendering/rendering_device.h +++ b/servers/rendering/rendering_device.h @@ -47,6 +47,7 @@ class RDPipelineRasterizationState; class RDPipelineMultisampleState; class RDPipelineDepthStencilState; class RDPipelineColorBlendState; +class RDFramebufferPass; class RenderingDevice : public Object { GDCLASS(RenderingDevice, Object) @@ -517,10 +518,23 @@ public: // This ID is warranted to be unique for the same formats, does not need to be freed virtual FramebufferFormatID framebuffer_format_create(const Vector<AttachmentFormat> &p_format, uint32_t p_view_count = 1) = 0; + struct FramebufferPass { + enum { + ATTACHMENT_UNUSED = -1 + }; + Vector<int32_t> color_attachments; + Vector<int32_t> input_attachments; + Vector<int32_t> resolve_attachments; + Vector<int32_t> preserve_attachments; + int32_t depth_attachment = ATTACHMENT_UNUSED; + }; + + virtual FramebufferFormatID framebuffer_format_create_multipass(const Vector<AttachmentFormat> &p_attachments, Vector<FramebufferPass> &p_passes, uint32_t p_view_count = 1) = 0; virtual FramebufferFormatID framebuffer_format_create_empty(TextureSamples p_samples = TEXTURE_SAMPLES_1) = 0; - virtual TextureSamples framebuffer_format_get_texture_samples(FramebufferFormatID p_format) = 0; + virtual TextureSamples framebuffer_format_get_texture_samples(FramebufferFormatID p_format, uint32_t p_pass = 0) = 0; virtual RID framebuffer_create(const Vector<RID> &p_texture_attachments, FramebufferFormatID p_format_check = INVALID_ID, uint32_t p_view_count = 1) = 0; + virtual RID framebuffer_create_multipass(const Vector<RID> &p_texture_attachments, Vector<FramebufferPass> &p_passes, FramebufferFormatID p_format_check = INVALID_ID, uint32_t p_view_count = 1) = 0; virtual RID framebuffer_create_empty(const Size2i &p_size, TextureSamples p_samples = TEXTURE_SAMPLES_1, FramebufferFormatID p_format_check = INVALID_ID) = 0; virtual FramebufferFormatID framebuffer_get_format(RID p_framebuffer) = 0; @@ -962,7 +976,7 @@ public: }; virtual bool render_pipeline_is_valid(RID p_pipeline) = 0; - virtual RID 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 = 0) = 0; + virtual RID 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 = 0, uint32_t p_for_render_pass = 0) = 0; /**************************/ /**** COMPUTE PIPELINE ****/ @@ -1018,6 +1032,9 @@ public: virtual void draw_list_enable_scissor(DrawListID p_list, const Rect2 &p_rect) = 0; virtual void draw_list_disable_scissor(DrawListID p_list) = 0; + virtual DrawListID draw_list_switch_to_next_pass() = 0; + virtual Error draw_list_switch_to_next_pass_split(uint32_t p_splits, DrawListID *r_split_ids) = 0; + virtual void draw_list_end(uint32_t p_post_barrier = BARRIER_MASK_ALL) = 0; /***********************/ @@ -1134,8 +1151,10 @@ protected: RID _texture_create_shared(const Ref<RDTextureView> &p_view, RID p_with_texture); RID _texture_create_shared_from_slice(const Ref<RDTextureView> &p_view, RID p_with_texture, uint32_t p_layer, uint32_t p_mipmap, TextureSliceType p_slice_type = TEXTURE_SLICE_2D); - FramebufferFormatID _framebuffer_format_create(const TypedArray<RDAttachmentFormat> &p_attachments); - RID _framebuffer_create(const Array &p_textures, FramebufferFormatID p_format_check = INVALID_ID); + FramebufferFormatID _framebuffer_format_create(const TypedArray<RDAttachmentFormat> &p_attachments, uint32_t p_view_count); + FramebufferFormatID _framebuffer_format_create_multipass(const TypedArray<RDAttachmentFormat> &p_attachments, const TypedArray<RDFramebufferPass> &p_passes, uint32_t p_view_count); + RID _framebuffer_create(const TypedArray<RID> &p_textures, FramebufferFormatID p_format_check = INVALID_ID, uint32_t p_view_count = 1); + RID _framebuffer_create_multipass(const TypedArray<RID> &p_textures, const TypedArray<RDFramebufferPass> &p_passes, FramebufferFormatID p_format_check = INVALID_ID, uint32_t p_view_count = 1); RID _sampler_create(const Ref<RDSamplerState> &p_state); VertexFormatID _vertex_format_create(const TypedArray<RDVertexAttribute> &p_vertex_formats); RID _vertex_array_create(uint32_t p_vertex_count, VertexFormatID p_vertex_format, const TypedArray<RID> &p_src_buffers); @@ -1146,11 +1165,12 @@ protected: Error _buffer_update(RID p_buffer, uint32_t p_offset, uint32_t p_size, const Vector<uint8_t> &p_data, uint32_t p_post_barrier = BARRIER_MASK_ALL); - RID _render_pipeline_create(RID p_shader, FramebufferFormatID p_framebuffer_format, VertexFormatID p_vertex_format, RenderPrimitive p_render_primitive, const Ref<RDPipelineRasterizationState> &p_rasterization_state, const Ref<RDPipelineMultisampleState> &p_multisample_state, const Ref<RDPipelineDepthStencilState> &p_depth_stencil_state, const Ref<RDPipelineColorBlendState> &p_blend_state, int p_dynamic_state_flags = 0); + RID _render_pipeline_create(RID p_shader, FramebufferFormatID p_framebuffer_format, VertexFormatID p_vertex_format, RenderPrimitive p_render_primitive, const Ref<RDPipelineRasterizationState> &p_rasterization_state, const Ref<RDPipelineMultisampleState> &p_multisample_state, const Ref<RDPipelineDepthStencilState> &p_depth_stencil_state, const Ref<RDPipelineColorBlendState> &p_blend_state, int p_dynamic_state_flags = 0, uint32_t p_for_render_pass = 0); Vector<int64_t> _draw_list_begin_split(RID p_framebuffer, uint32_t p_splits, InitialAction p_initial_color_action, FinalAction p_final_color_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, const Vector<Color> &p_clear_color_values = Vector<Color>(), float p_clear_depth = 1.0, uint32_t p_clear_stencil = 0, const Rect2 &p_region = Rect2(), const TypedArray<RID> &p_storage_textures = TypedArray<RID>()); void _draw_list_set_push_constant(DrawListID p_list, const Vector<uint8_t> &p_data, uint32_t p_data_size); void _compute_list_set_push_constant(ComputeListID p_list, const Vector<uint8_t> &p_data, uint32_t p_data_size); + Vector<int64_t> _draw_list_switch_to_next_pass_split(uint32_t p_splits); }; VARIANT_ENUM_CAST(RenderingDevice::ShaderStage) diff --git a/servers/rendering/rendering_device_binds.h b/servers/rendering/rendering_device_binds.h index 912674e309..dc59568ce9 100644 --- a/servers/rendering/rendering_device_binds.h +++ b/servers/rendering/rendering_device_binds.h @@ -128,6 +128,34 @@ protected: } }; +class RDFramebufferPass : public RefCounted { + GDCLASS(RDFramebufferPass, RefCounted) + friend class RenderingDevice; + + RD::FramebufferPass base; + +public: + RD_SETGET(PackedInt32Array, color_attachments) + RD_SETGET(PackedInt32Array, input_attachments) + RD_SETGET(PackedInt32Array, resolve_attachments) + RD_SETGET(PackedInt32Array, preserve_attachments) + RD_SETGET(int32_t, depth_attachment) +protected: + enum { + ATTACHMENT_UNUSED = -1 + }; + + static void _bind_methods() { + RD_BIND(Variant::PACKED_INT32_ARRAY, RDFramebufferPass, color_attachments); + RD_BIND(Variant::PACKED_INT32_ARRAY, RDFramebufferPass, input_attachments); + RD_BIND(Variant::PACKED_INT32_ARRAY, RDFramebufferPass, resolve_attachments); + RD_BIND(Variant::PACKED_INT32_ARRAY, RDFramebufferPass, preserve_attachments); + RD_BIND(Variant::INT, RDFramebufferPass, depth_attachment); + + BIND_CONSTANT(ATTACHMENT_UNUSED); + } +}; + class RDSamplerState : public RefCounted { GDCLASS(RDSamplerState, RefCounted) friend class RenderingDevice; |