diff options
Diffstat (limited to 'drivers/vulkan/rendering_device_vulkan.cpp')
| -rw-r--r-- | drivers/vulkan/rendering_device_vulkan.cpp | 1536 |
1 files changed, 1199 insertions, 337 deletions
diff --git a/drivers/vulkan/rendering_device_vulkan.cpp b/drivers/vulkan/rendering_device_vulkan.cpp index dc92e04a43..b5bd6cef38 100644 --- a/drivers/vulkan/rendering_device_vulkan.cpp +++ b/drivers/vulkan/rendering_device_vulkan.cpp @@ -31,11 +31,14 @@ #include "rendering_device_vulkan.h" #include "core/config/project_settings.h" -#include "core/os/file_access.h" +#include "core/io/compression.h" +#include "core/io/file_access.h" +#include "core/io/marshalls.h" #include "core/os/os.h" #include "core/templates/hashfuncs.h" #include "drivers/vulkan/vulkan_context.h" +#include "thirdparty/misc/smolv.h" #include "thirdparty/spirv-reflect/spirv_reflect.h" //#define FORCE_FULL_BARRIER @@ -1398,12 +1401,15 @@ Error RenderingDeviceVulkan::_buffer_allocate(Buffer *p_buffer, uint32_t p_size, p_buffer->buffer_info.range = p_size; p_buffer->usage = p_usage; + buffer_memory += p_size; + return OK; } Error RenderingDeviceVulkan::_buffer_free(Buffer *p_buffer) { ERR_FAIL_COND_V(p_buffer->size == 0, ERR_INVALID_PARAMETER); + buffer_memory -= p_buffer->size; vmaDestroyBuffer(allocator, p_buffer->buffer, p_buffer->allocation); p_buffer->buffer = VK_NULL_HANDLE; p_buffer->allocation = nullptr; @@ -1798,6 +1804,10 @@ RID RenderingDeviceVulkan::texture_create(const TextureFormat &p_format, const T image_create_info.usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT; } + if (p_format.usage_bits & TEXTURE_USAGE_INPUT_ATTACHMENT_BIT) { + image_create_info.usage |= VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT; + } + if (p_format.usage_bits & TEXTURE_USAGE_CAN_UPDATE_BIT) { image_create_info.usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT; } @@ -1896,7 +1906,7 @@ RID RenderingDeviceVulkan::texture_create(const TextureFormat &p_format, const T VkResult err = vmaCreateImage(allocator, &image_create_info, &allocInfo, &texture.image, &texture.allocation, &texture.allocation_info); ERR_FAIL_COND_V_MSG(err, RID(), "vmaCreateImage failed with error " + itos(err) + "."); - + image_memory += texture.allocation_info.size; texture.type = p_format.texture_type; texture.format = p_format.format; texture.width = image_create_info.extent.width; @@ -2029,7 +2039,7 @@ RID RenderingDeviceVulkan::texture_create(const TextureFormat &p_format, const T if (p_data.size()) { for (uint32_t i = 0; i < image_create_info.arrayLayers; i++) { - texture_update(id, i, p_data[i]); + _texture_update(id, i, p_data[i], RD::BARRIER_MASK_ALL, true); } } return id; @@ -2128,6 +2138,10 @@ RID RenderingDeviceVulkan::texture_create_shared(const TextureView &p_view, RID } } + if (texture.usage_flags & TEXTURE_USAGE_INPUT_ATTACHMENT_BIT) { + usage_info.usage |= VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT; + } + if (texture.usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { usage_info.usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT; } @@ -2276,10 +2290,14 @@ RID RenderingDeviceVulkan::texture_create_shared_from_slice(const TextureView &p } Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, const Vector<uint8_t> &p_data, uint32_t p_post_barrier) { + return _texture_update(p_texture, p_layer, p_data, p_post_barrier, false); +} + +Error RenderingDeviceVulkan::_texture_update(RID p_texture, uint32_t p_layer, const Vector<uint8_t> &p_data, uint32_t p_post_barrier, bool p_use_setup_queue) { _THREAD_SAFE_METHOD_ - ERR_FAIL_COND_V_MSG(draw_list || compute_list, ERR_INVALID_PARAMETER, - "Updating textures in is forbidden during creation of a draw or compute list"); + ERR_FAIL_COND_V_MSG((draw_list || compute_list) && !p_use_setup_queue, ERR_INVALID_PARAMETER, + "Updating textures is forbidden during creation of a draw or compute list"); Texture *texture = texture_owner.getornull(p_texture); ERR_FAIL_COND_V(!texture, ERR_INVALID_PARAMETER); @@ -2320,7 +2338,7 @@ Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, con const uint8_t *r = p_data.ptr(); - VkCommandBuffer command_buffer = p_post_barrier ? frames[frame].draw_command_buffer : frames[frame].setup_command_buffer; + VkCommandBuffer command_buffer = p_use_setup_queue ? frames[frame].setup_command_buffer : frames[frame].draw_command_buffer; //barrier to transfer { @@ -2373,7 +2391,7 @@ Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, con to_allocate >>= get_compressed_image_format_pixel_rshift(texture->format); uint32_t alloc_offset, alloc_size; - Error err = _staging_buffer_allocate(to_allocate, required_align, alloc_offset, alloc_size, false, p_post_barrier); + Error err = _staging_buffer_allocate(to_allocate, required_align, alloc_offset, alloc_size, false, !p_use_setup_queue); ERR_FAIL_COND_V(err, ERR_CANT_CREATE); uint8_t *write_ptr; @@ -3244,12 +3262,7 @@ 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) { - 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 +3275,31 @@ 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)), - VK_NULL_HANDLE, "Texture format for index (" + itos(i) + ") requires an attachment (depth, stencil or resolve) bit set."); + 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_INPUT_ATTACHMENT_BIT)), + VK_NULL_HANDLE, "Texture format for index (" + itos(i) + ") requires an attachment (color, depth, input or stencil) bit set."); VkAttachmentDescription description = {}; description.flags = 0; - description.format = vulkan_formats[p_format[i].format]; - description.samples = rasterization_sample_count[p_format[i].samples]; + description.format = vulkan_formats[p_attachments[i].format]; + description.samples = rasterization_sample_count[p_attachments[i].samples]; - bool is_depth_stencil = p_format[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT; - bool is_sampled = p_format[i].usage_flags & TEXTURE_USAGE_SAMPLING_BIT; - bool is_storage = p_format[i].usage_flags & TEXTURE_USAGE_STORAGE_BIT; + bool is_sampled = p_attachments[i].usage_flags & TEXTURE_USAGE_SAMPLING_BIT; + bool is_storage = p_attachments[i].usage_flags & TEXTURE_USAGE_STORAGE_BIT; + bool is_depth = p_attachments[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT; // For each UNDEFINED, assume the prior use was a *read*, as we'd be discarding the output of a write // Also, each UNDEFINED will do an immediate layout transition (write), s.t. we must ensure execution synchronization vs. // the read. If this is a performance issue, one could track the actual last accessor of each resource, adding only that // 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 +3308,11 @@ 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 +3325,11 @@ 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 +3343,11 @@ 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 +3363,14 @@ 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 +3383,11 @@ 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 +3398,11 @@ 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 +3418,194 @@ 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; + + subpasses.resize(p_passes.size()); + color_reference_array.resize(p_passes.size()); + input_reference_array.resize(p_passes.size()); + resolve_reference_array.resize(p_passes.size()); + preserve_reference_array.resize(p_passes.size()); + depth_reference_array.resize(p_passes.size()); + + LocalVector<VkSubpassDependency> subpass_dependencies; + + for (int i = 0; i < p_passes.size(); i++) { + const FramebufferPass *pass = &p_passes[i]; + + LocalVector<VkAttachmentReference> &color_references = color_reference_array[i]; - VkAttachmentReference reference; - reference.attachment = i; + TextureSamples texture_samples = TEXTURE_SAMPLES_1; + bool is_multisample_first = true; - if (p_format[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { - reference.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; + 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_INPUT_ATTACHMENT_BIT), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it isn't marked as an input texture."); + ERR_FAIL_COND_V_MSG(attachment_last_pass[attachment] == i, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it already was used for something else before in this pass."); + reference.attachment = attachment; + reference.layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; + attachment_last_pass[attachment] = i; + } + input_references.push_back(reference); + } + + LocalVector<VkAttachmentReference> &resolve_references = resolve_reference_array[i]; + + if (pass->resolve_attachments.size() > 0) { + ERR_FAIL_COND_V_MSG(pass->resolve_attachments.size() != pass->color_attachments.size(), VK_NULL_HANDLE, "The amount of resolve attachments (" + itos(pass->resolve_attachments.size()) + ") must match the number of color attachments (" + itos(pass->color_attachments.size()) + ")."); + ERR_FAIL_COND_V_MSG(texture_samples == TEXTURE_SAMPLES_1, VK_NULL_HANDLE, "Resolve attachments specified, but color attachments are not multisample."); + } + for (int j = 0; j < pass->resolve_attachments.size(); j++) { + int32_t attachment = pass->resolve_attachments[j]; + VkAttachmentReference reference; + if (attachment == FramebufferPass::ATTACHMENT_UNUSED) { + reference.attachment = VK_ATTACHMENT_UNUSED; + reference.layout = VK_IMAGE_LAYOUT_UNDEFINED; + } else { + ERR_FAIL_INDEX_V_MSG(attachment, p_attachments.size(), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), resolve attachment (" + itos(j) + ")."); + ERR_FAIL_COND_V_MSG(pass->color_attachments[j] == FramebufferPass::ATTACHMENT_UNUSED, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), resolve attachment (" + itos(j) + "), the respective color attachment is marked as unused."); + ERR_FAIL_COND_V_MSG(!(p_attachments[attachment].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT), VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), resolve attachment, it isn't marked as a color texture."); + ERR_FAIL_COND_V_MSG(attachment_last_pass[attachment] == i, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), it already was used for something else before in this pass."); + bool multisample = p_attachments[attachment].samples > TEXTURE_SAMPLES_1; + ERR_FAIL_COND_V_MSG(multisample, VK_NULL_HANDLE, "Invalid framebuffer format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), resolve attachments can't be multisample."); + reference.attachment = attachment; + reference.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; // VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; + attachment_last_pass[attachment] = i; + } resolve_references.push_back(reference); - // if resolves are done, we need to ensure the copy is safe - dependency_to_external.dstStageMask |= VK_PIPELINE_STAGE_TRANSFER_BIT; - dependency_to_external.dstAccessMask |= VK_ACCESS_TRANSFER_READ_BIT; + } + + 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) { + texture_samples = p_attachments[attachment].samples; + is_multisample_first = false; + } else { + ERR_FAIL_COND_V_MSG(texture_samples != p_attachments[attachment].samples, VK_NULL_HANDLE, "Invalid framebuffer depth format attachment(" + itos(attachment) + "), in pass (" + itos(i) + "), if an attachment is marked as multisample, all of them should be multisample and use the same number of samples including the depth."); + } + } else { - ERR_FAIL_V_MSG(VK_NULL_HANDLE, "Texture index " + itos(i) + " is neither color, depth stencil or resolve so it can't be used as attachment."); + depth_stencil_reference.attachment = VK_ATTACHMENT_UNUSED; + depth_stencil_reference.layout = VK_IMAGE_LAYOUT_UNDEFINED; } + VkSubpassDescription &subpass = subpasses[i]; + subpass.flags = 0; + subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS; + subpass.inputAttachmentCount = input_references.size(); + if (input_references.size()) { + subpass.pInputAttachments = input_references.ptr(); + } else { + subpass.pInputAttachments = nullptr; + } + subpass.colorAttachmentCount = color_references.size(); + if (color_references.size()) { + subpass.pColorAttachments = color_references.ptr(); + } else { + subpass.pColorAttachments = nullptr; + } + if (depth_stencil_reference.attachment != VK_ATTACHMENT_UNUSED) { + subpass.pDepthStencilAttachment = &depth_stencil_reference; + } else { + subpass.pDepthStencilAttachment = nullptr; + } + + if (resolve_references.size()) { + subpass.pResolveAttachments = resolve_references.ptr(); + } else { + subpass.pResolveAttachments = nullptr; + } + + subpass.preserveAttachmentCount = preserve_references.size(); + if (preserve_references.size()) { + subpass.pPreserveAttachments = preserve_references.ptr(); + } else { + subpass.pPreserveAttachments = nullptr; + } + + if (r_samples) { + r_samples->push_back(texture_samples); + } + + if (i > 0) { + VkSubpassDependency dependency; + dependency.srcSubpass = i - 1; + dependency.dstSubpass = i; + dependency.srcStageMask = 0; + dependency.srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT; + dependency.dstStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; + + dependency.srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; + dependency.dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_INPUT_ATTACHMENT_READ_BIT; + dependency.dependencyFlags = VK_DEPENDENCY_BY_REGION_BIT; + subpass_dependencies.push_back(dependency); + } + /* // NOTE: Big Mallet Approach -- any layout transition causes a full barrier if (reference.layout != description.initialLayout) { // NOTE: this should be smarter based on the texture's knowledge of its previous role @@ -3433,57 +3617,89 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF dependency_to_external.dstStageMask |= VK_PIPELINE_STAGE_ALL_COMMANDS_BIT; dependency_to_external.dstAccessMask |= VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT; } + */ } - ERR_FAIL_COND_V_MSG(depth_stencil_references.size() > 1, VK_NULL_HANDLE, - "Formats can only have one depth/stencil attachment, supplied (" + itos(depth_stencil_references.size()) + ")."); - - ERR_FAIL_COND_V_MSG(resolve_references.size() > 1, VK_NULL_HANDLE, - "Formats can only have one resolve attachment, supplied (" + itos(resolve_references.size()) + ")."); - - VkSubpassDescription subpass; - subpass.flags = 0; - subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS; - subpass.inputAttachmentCount = 0; //unsupported for now - subpass.pInputAttachments = nullptr; - subpass.colorAttachmentCount = color_references.size(); - subpass.pColorAttachments = color_references.ptr(); - subpass.pDepthStencilAttachment = depth_stencil_references.ptr(); - subpass.pResolveAttachments = resolve_references.ptr(); - subpass.preserveAttachmentCount = 0; - subpass.pPreserveAttachments = nullptr; - VkRenderPassCreateInfo render_pass_create_info; render_pass_create_info.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO; render_pass_create_info.pNext = nullptr; render_pass_create_info.flags = 0; render_pass_create_info.attachmentCount = attachments.size(); render_pass_create_info.pAttachments = attachments.ptr(); - render_pass_create_info.subpassCount = 1; - render_pass_create_info.pSubpasses = &subpass; + render_pass_create_info.subpassCount = subpasses.size(); + render_pass_create_info.pSubpasses = subpasses.ptr(); // Commenting this because it seems it just avoids raster and compute to work at the same time. // Other barriers seem to be protecting the render pass fine. // render_pass_create_info.dependencyCount = 2; // render_pass_create_info.pDependencies = dependencies; - render_pass_create_info.dependencyCount = 0; - render_pass_create_info.pDependencies = nullptr; + render_pass_create_info.dependencyCount = subpass_dependencies.size(); + if (subpass_dependencies.size()) { + render_pass_create_info.pDependencies = subpass_dependencies.ptr(); + } else { + render_pass_create_info.pDependencies = nullptr; + } + + // These are only used if we use multiview but we need to define them in scope. + const uint32_t view_mask = (1 << p_view_count) - 1; + const uint32_t correlation_mask = (1 << p_view_count) - 1; + Vector<uint32_t> view_masks; + VkRenderPassMultiviewCreateInfo render_pass_multiview_create_info; + + if (p_view_count > 1) { + const VulkanContext::MultiviewCapabilities capabilities = context->get_multiview_capabilities(); + + // For now this only works with multiview! + ERR_FAIL_COND_V_MSG(!capabilities.is_supported, VK_NULL_HANDLE, "Multiview not supported"); + + // Make sure we limit this to the number of views we support. + ERR_FAIL_COND_V_MSG(p_view_count > capabilities.max_view_count, VK_NULL_HANDLE, "Hardware does not support requested number of views for Multiview render pass"); + + // Set view masks for each subpass + for (uint32_t i = 0; i < subpasses.size(); i++) { + view_masks.push_back(view_mask); + }; + + render_pass_multiview_create_info.sType = VK_STRUCTURE_TYPE_RENDER_PASS_MULTIVIEW_CREATE_INFO; + render_pass_multiview_create_info.pNext = nullptr; + render_pass_multiview_create_info.subpassCount = subpasses.size(); + render_pass_multiview_create_info.pViewMasks = view_masks.ptr(); + render_pass_multiview_create_info.dependencyCount = 0; + render_pass_multiview_create_info.pViewOffsets = nullptr; + render_pass_multiview_create_info.correlationMaskCount = 1; + render_pass_multiview_create_info.pCorrelationMasks = &correlation_mask; + + render_pass_create_info.pNext = &render_pass_multiview_create_info; + } VkRenderPass render_pass; VkResult res = vkCreateRenderPass(device, &render_pass_create_info, nullptr, &render_pass); ERR_FAIL_COND_V_MSG(res, VK_NULL_HANDLE, "vkCreateRenderPass failed with error " + itos(res) + "."); - if (r_color_attachment_count) { - *r_color_attachment_count = color_references.size(); - } return render_pass; } -RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_create(const Vector<AttachmentFormat> &p_format) { +RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_create(const Vector<AttachmentFormat> &p_format, uint32_t p_view_count) { + FramebufferPass pass; + for (int i = 0; i < p_format.size(); i++) { + if (p_format[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { + pass.depth_attachment = i; + } else { + pass.color_attachments.push_back(i); + } + } + + Vector<FramebufferPass> passes; + passes.push_back(pass); + return framebuffer_format_create_multipass(p_format, passes, p_view_count); +} +RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_create_multipass(const Vector<AttachmentFormat> &p_attachments, Vector<FramebufferPass> &p_passes, uint32_t p_view_count) { _THREAD_SAFE_METHOD_ FramebufferFormatKey key; - key.attachments = p_format; + key.attachments = p_attachments; + key.passes = p_passes; + key.view_count = p_view_count; const Map<FramebufferFormatKey, FramebufferFormatID>::Element *E = framebuffer_format_cache.find(key); if (E) { @@ -3491,8 +3707,8 @@ 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); //actions don't matter for this use case + Vector<TextureSamples> samples; + VkRenderPass render_pass = _render_pass_create(p_attachments, p_passes, INITIAL_ACTION_CLEAR, FINAL_ACTION_READ, INITIAL_ACTION_CLEAR, FINAL_ACTION_READ, p_view_count, &samples); //actions don't matter for this use case if (render_pass == VK_NULL_HANDLE) { //was likely invalid return INVALID_ID; @@ -3502,15 +3718,16 @@ RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_c E = framebuffer_format_cache.insert(key, id); FramebufferFormat fb_format; fb_format.E = E; - fb_format.color_attachments = color_references; fb_format.render_pass = render_pass; - fb_format.samples = p_format[0].samples; + fb_format.pass_samples = samples; + fb_format.view_count = p_view_count; framebuffer_formats[id] = fb_format; return id; } RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_create_empty(TextureSamples p_samples) { FramebufferFormatKey key; + key.passes.push_back(FramebufferPass()); const Map<FramebufferFormatKey, FramebufferFormatID>::Element *E = framebuffer_format_cache.find(key); if (E) { @@ -3556,18 +3773,18 @@ 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]; } /***********************/ @@ -3580,11 +3797,36 @@ RID RenderingDeviceVulkan::framebuffer_create_empty(const Size2i &p_size, Textur framebuffer.format_id = framebuffer_format_create_empty(p_samples); ERR_FAIL_COND_V(p_format_check != INVALID_FORMAT_ID && framebuffer.format_id != p_format_check, RID()); framebuffer.size = p_size; + framebuffer.view_count = 1; return framebuffer_owner.make_rid(framebuffer); } -RID RenderingDeviceVulkan::framebuffer_create(const Vector<RID> &p_texture_attachments, FramebufferFormatID p_format_check) { +RID RenderingDeviceVulkan::framebuffer_create(const Vector<RID> &p_texture_attachments, FramebufferFormatID p_format_check, uint32_t p_view_count) { + _THREAD_SAFE_METHOD_ + + 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; @@ -3594,6 +3836,8 @@ RID RenderingDeviceVulkan::framebuffer_create(const Vector<RID> &p_texture_attac 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 (i == 0) { size.width = texture->width; size.height = texture->height; @@ -3609,7 +3853,7 @@ RID RenderingDeviceVulkan::framebuffer_create(const Vector<RID> &p_texture_attac attachments.push_back(af); } - FramebufferFormatID format_id = framebuffer_format_create(attachments); + FramebufferFormatID format_id = framebuffer_format_create_multipass(attachments, p_passes, p_view_count); if (format_id == INVALID_ID) { return RID(); } @@ -3621,6 +3865,7 @@ RID RenderingDeviceVulkan::framebuffer_create(const Vector<RID> &p_texture_attac framebuffer.format_id = format_id; framebuffer.texture_ids = p_texture_attachments; framebuffer.size = size; + framebuffer.view_count = p_view_count; RID id = framebuffer_owner.make_rid(framebuffer); @@ -4133,51 +4378,87 @@ bool RenderingDeviceVulkan::_uniform_add_binding(Vector<Vector<VkDescriptorSetLa } #endif -RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages) { - //descriptor layouts - Vector<Vector<VkDescriptorSetLayoutBinding>> set_bindings; - Vector<Vector<UniformInfo>> uniform_info; - Shader::PushConstant push_constant; - push_constant.push_constant_size = 0; - push_constant.push_constants_vk_stage = 0; +#define SHADER_BINARY_VERSION 1 - uint32_t vertex_input_mask = 0; +String RenderingDeviceVulkan::shader_get_binary_cache_key() const { + return "Vulkan-SV" + itos(SHADER_BINARY_VERSION); +} - uint32_t fragment_outputs = 0; +struct RenderingDeviceVulkanShaderBinaryDataBinding { + uint32_t type; + uint32_t binding; + uint32_t stages; + uint32_t length; //size of arrays (in total elements), or ubos (in bytes * total elements) +}; - uint32_t stages_processed = 0; +struct RenderingDeviceVulkanShaderBinarySpecializationConstant { + uint32_t type; + uint32_t constant_id; + union { + uint32_t int_value; + float float_value; + bool bool_value; + }; + uint32_t stage_flags; +}; - bool is_compute = false; +struct RenderingDeviceVulkanShaderBinaryData { + uint32_t vertex_input_mask; + uint32_t fragment_outputs; + uint32_t specialization_constant_count; + uint32_t is_compute; + uint32_t compute_local_size[3]; + uint32_t set_count; + uint32_t push_constant_size; + uint32_t push_constants_vk_stage; + uint32_t stage_count; +}; + +Vector<uint8_t> RenderingDeviceVulkan::shader_compile_binary_from_spirv(const Vector<ShaderStageSPIRVData> &p_spirv) { + RenderingDeviceVulkanShaderBinaryData binary_data; + binary_data.vertex_input_mask = 0; + binary_data.fragment_outputs = 0; + binary_data.specialization_constant_count = 0; + binary_data.is_compute = 0; + binary_data.compute_local_size[0] = 0; + binary_data.compute_local_size[1] = 0; + binary_data.compute_local_size[2] = 0; + binary_data.set_count = 0; + binary_data.push_constant_size = 0; + binary_data.push_constants_vk_stage = 0; + + Vector<Vector<RenderingDeviceVulkanShaderBinaryDataBinding>> uniform_info; //set bindings + Vector<RenderingDeviceVulkanShaderBinarySpecializationConstant> specialization_constants; - uint32_t compute_local_size[3] = { 0, 0, 0 }; + uint32_t stages_processed = 0; - for (int i = 0; i < p_stages.size(); i++) { - if (p_stages[i].shader_stage == SHADER_STAGE_COMPUTE) { - is_compute = true; - ERR_FAIL_COND_V_MSG(p_stages.size() != 1, RID(), + for (int i = 0; i < p_spirv.size(); i++) { + if (p_spirv[i].shader_stage == SHADER_STAGE_COMPUTE) { + binary_data.is_compute = true; + ERR_FAIL_COND_V_MSG(p_spirv.size() != 1, Vector<uint8_t>(), "Compute shaders can only receive one stage, dedicated to compute."); } - ERR_FAIL_COND_V_MSG(stages_processed & (1 << p_stages[i].shader_stage), RID(), - "Stage " + String(shader_stage_names[p_stages[i].shader_stage]) + " submitted more than once."); + ERR_FAIL_COND_V_MSG(stages_processed & (1 << p_spirv[i].shader_stage), Vector<uint8_t>(), + "Stage " + String(shader_stage_names[p_spirv[i].shader_stage]) + " submitted more than once."); { SpvReflectShaderModule module; - const uint8_t *spirv = p_stages[i].spir_v.ptr(); - SpvReflectResult result = spvReflectCreateShaderModule(p_stages[i].spir_v.size(), spirv, &module); - ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), - "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed parsing shader."); - - if (is_compute) { - compute_local_size[0] = module.entry_points->local_size.x; - compute_local_size[1] = module.entry_points->local_size.y; - compute_local_size[2] = module.entry_points->local_size.z; + const uint8_t *spirv = p_spirv[i].spir_v.ptr(); + SpvReflectResult result = spvReflectCreateShaderModule(p_spirv[i].spir_v.size(), spirv, &module); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed parsing shader."); + + if (binary_data.is_compute) { + binary_data.compute_local_size[0] = module.entry_points->local_size.x; + binary_data.compute_local_size[1] = module.entry_points->local_size.y; + binary_data.compute_local_size[2] = module.entry_points->local_size.z; } uint32_t binding_count = 0; result = spvReflectEnumerateDescriptorBindings(&module, &binding_count, nullptr); - ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), - "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed enumerating descriptor bindings."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed enumerating descriptor bindings."); - uint32_t stage = p_stages[i].shader_stage; + uint32_t stage = p_spirv[i].shader_stage; if (binding_count > 0) { //Parse bindings @@ -4186,56 +4467,47 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages bindings.resize(binding_count); result = spvReflectEnumerateDescriptorBindings(&module, &binding_count, bindings.ptrw()); - ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), - "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed getting descriptor bindings."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed getting descriptor bindings."); for (uint32_t j = 0; j < binding_count; j++) { const SpvReflectDescriptorBinding &binding = *bindings[j]; - VkDescriptorSetLayoutBinding layout_binding; - UniformInfo info; + RenderingDeviceVulkanShaderBinaryDataBinding info; bool need_array_dimensions = false; bool need_block_size = false; switch (binding.descriptor_type) { case SPV_REFLECT_DESCRIPTOR_TYPE_SAMPLER: { - layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLER; info.type = UNIFORM_TYPE_SAMPLER; need_array_dimensions = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER: { - layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; info.type = UNIFORM_TYPE_SAMPLER_WITH_TEXTURE; need_array_dimensions = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_SAMPLED_IMAGE: { - layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE; info.type = UNIFORM_TYPE_TEXTURE; need_array_dimensions = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_STORAGE_IMAGE: { - layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE; info.type = UNIFORM_TYPE_IMAGE; need_array_dimensions = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER: { - layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; info.type = UNIFORM_TYPE_TEXTURE_BUFFER; need_array_dimensions = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER: { - layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER; info.type = UNIFORM_TYPE_IMAGE_BUFFER; need_array_dimensions = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_UNIFORM_BUFFER: { - layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; info.type = UNIFORM_TYPE_UNIFORM_BUFFER; need_block_size = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_STORAGE_BUFFER: { - layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER; info.type = UNIFORM_TYPE_STORAGE_BUFFER; need_block_size = true; } break; @@ -4248,8 +4520,8 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages continue; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_INPUT_ATTACHMENT: { - layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT; info.type = UNIFORM_TYPE_INPUT_ATTACHMENT; + need_array_dimensions = true; } break; case SPV_REFLECT_DESCRIPTOR_TYPE_ACCELERATION_STRUCTURE_KHR: { ERR_PRINT("Acceleration structure not supported."); @@ -4270,42 +4542,35 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages } } - layout_binding.descriptorCount = info.length; - } else if (need_block_size) { info.length = binding.block.size; - layout_binding.descriptorCount = 1; } else { info.length = 0; - layout_binding.descriptorCount = 1; } info.binding = binding.binding; uint32_t set = binding.set; - //print_line("Stage: " + String(shader_stage_names[stage]) + " set=" + itos(set) + " binding=" + itos(info.binding) + " type=" + shader_uniform_names[info.type] + " length=" + itos(info.length)); - - ERR_FAIL_COND_V_MSG(set >= MAX_UNIFORM_SETS, RID(), + ERR_FAIL_COND_V_MSG(set >= MAX_UNIFORM_SETS, Vector<uint8_t>(), "On shader stage '" + String(shader_stage_names[stage]) + "', uniform '" + binding.name + "' uses a set (" + itos(set) + ") index larger than what is supported (" + itos(MAX_UNIFORM_SETS) + ")."); - ERR_FAIL_COND_V_MSG(set >= limits.maxBoundDescriptorSets, RID(), + ERR_FAIL_COND_V_MSG(set >= limits.maxBoundDescriptorSets, Vector<uint8_t>(), "On shader stage '" + String(shader_stage_names[stage]) + "', uniform '" + binding.name + "' uses a set (" + itos(set) + ") index larger than what is supported by the hardware (" + itos(limits.maxBoundDescriptorSets) + ")."); - if (set < (uint32_t)set_bindings.size()) { + if (set < (uint32_t)uniform_info.size()) { //check if this already exists bool exists = false; - for (int k = 0; k < set_bindings[set].size(); k++) { - if (set_bindings[set][k].binding == (uint32_t)info.binding) { + for (int k = 0; k < uniform_info[set].size(); k++) { + if (uniform_info[set][k].binding == (uint32_t)info.binding) { //already exists, verify that it's the same type - ERR_FAIL_COND_V_MSG(set_bindings[set][k].descriptorType != layout_binding.descriptorType, RID(), + ERR_FAIL_COND_V_MSG(uniform_info[set][k].type != info.type, Vector<uint8_t>(), "On shader stage '" + String(shader_stage_names[stage]) + "', uniform '" + binding.name + "' trying to re-use location for set=" + itos(set) + ", binding=" + itos(info.binding) + " with different uniform type."); //also, verify that it's the same size - ERR_FAIL_COND_V_MSG(set_bindings[set][k].descriptorCount != layout_binding.descriptorCount || uniform_info[set][k].length != info.length, RID(), + ERR_FAIL_COND_V_MSG(uniform_info[set][k].length != info.length, Vector<uint8_t>(), "On shader stage '" + String(shader_stage_names[stage]) + "', uniform '" + binding.name + "' trying to re-use location for set=" + itos(set) + ", binding=" + itos(info.binding) + " with different uniform size."); //just append stage mask and return - set_bindings.write[set].write[k].stageFlags |= shader_stage_masks[stage]; uniform_info.write[set].write[k].stages |= 1 << stage; exists = true; } @@ -4316,40 +4581,87 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages } } - layout_binding.binding = info.binding; - layout_binding.stageFlags = shader_stage_masks[stage]; - layout_binding.pImmutableSamplers = nullptr; //no support for this yet - info.stages = 1 << stage; - info.binding = info.binding; - if (set >= (uint32_t)set_bindings.size()) { - set_bindings.resize(set + 1); + if (set >= (uint32_t)uniform_info.size()) { uniform_info.resize(set + 1); } - set_bindings.write[set].push_back(layout_binding); uniform_info.write[set].push_back(info); } } + { + //specialization constants + + uint32_t sc_count = 0; + result = spvReflectEnumerateSpecializationConstants(&module, &sc_count, nullptr); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed enumerating specialization constants."); + + if (sc_count) { + Vector<SpvReflectSpecializationConstant *> spec_constants; + spec_constants.resize(sc_count); + + result = spvReflectEnumerateSpecializationConstants(&module, &sc_count, spec_constants.ptrw()); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed obtaining specialization constants."); + + for (uint32_t j = 0; j < sc_count; j++) { + int32_t existing = -1; + RenderingDeviceVulkanShaderBinarySpecializationConstant sconst; + sconst.constant_id = spec_constants[j]->constant_id; + switch (spec_constants[j]->constant_type) { + case SPV_REFLECT_SPECIALIZATION_CONSTANT_BOOL: { + sconst.type = PIPELINE_SPECIALIZATION_CONSTANT_TYPE_BOOL; + sconst.bool_value = spec_constants[j]->default_value.int_bool_value != 0; + } break; + case SPV_REFLECT_SPECIALIZATION_CONSTANT_INT: { + sconst.type = PIPELINE_SPECIALIZATION_CONSTANT_TYPE_INT; + sconst.int_value = spec_constants[j]->default_value.int_bool_value; + } break; + case SPV_REFLECT_SPECIALIZATION_CONSTANT_FLOAT: { + sconst.type = PIPELINE_SPECIALIZATION_CONSTANT_TYPE_FLOAT; + sconst.float_value = spec_constants[j]->default_value.float_value; + } break; + } + sconst.stage_flags = 1 << p_spirv[i].shader_stage; + + for (int k = 0; k < specialization_constants.size(); k++) { + if (specialization_constants[k].constant_id == sconst.constant_id) { + ERR_FAIL_COND_V_MSG(specialization_constants[k].type != sconst.type, Vector<uint8_t>(), "More than one specialization constant used for id (" + itos(sconst.constant_id) + "), but their types differ."); + ERR_FAIL_COND_V_MSG(specialization_constants[k].int_value != sconst.int_value, Vector<uint8_t>(), "More than one specialization constant used for id (" + itos(sconst.constant_id) + "), but their default values differ."); + existing = k; + break; + } + } + + if (existing > 0) { + specialization_constants.write[existing].stage_flags |= sconst.stage_flags; + } else { + specialization_constants.push_back(sconst); + } + } + } + } + if (stage == SHADER_STAGE_VERTEX) { uint32_t iv_count = 0; result = spvReflectEnumerateInputVariables(&module, &iv_count, nullptr); - ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), - "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed enumerating input variables."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed enumerating input variables."); if (iv_count) { Vector<SpvReflectInterfaceVariable *> input_vars; input_vars.resize(iv_count); result = spvReflectEnumerateInputVariables(&module, &iv_count, input_vars.ptrw()); - ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), - "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed obtaining input variables."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed obtaining input variables."); for (uint32_t j = 0; j < iv_count; j++) { if (input_vars[j] && input_vars[j]->decoration_flags == 0) { //regular input - vertex_input_mask |= (1 << uint32_t(input_vars[j]->location)); + binary_data.vertex_input_mask |= (1 << uint32_t(input_vars[j]->location)); } } } @@ -4358,20 +4670,21 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages if (stage == SHADER_STAGE_FRAGMENT) { uint32_t ov_count = 0; result = spvReflectEnumerateOutputVariables(&module, &ov_count, nullptr); - ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), - "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed enumerating output variables."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed enumerating output variables."); if (ov_count) { Vector<SpvReflectInterfaceVariable *> output_vars; output_vars.resize(ov_count); result = spvReflectEnumerateOutputVariables(&module, &ov_count, output_vars.ptrw()); - ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), - "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed obtaining output variables."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed obtaining output variables."); for (uint32_t j = 0; j < ov_count; j++) { - if (output_vars[j]) { - fragment_outputs = MAX(fragment_outputs, output_vars[j]->location + 1); + const SpvReflectInterfaceVariable *refvar = output_vars[j]; + if (refvar != nullptr && refvar->built_in != SpvBuiltInFragDepth) { + binary_data.fragment_outputs |= 1 << refvar->location; } } } @@ -4379,18 +4692,18 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages uint32_t pc_count = 0; result = spvReflectEnumeratePushConstantBlocks(&module, &pc_count, nullptr); - ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), - "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed enumerating push constants."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed enumerating push constants."); if (pc_count) { - ERR_FAIL_COND_V_MSG(pc_count > 1, RID(), - "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "': Only one push constant is supported, which should be the same across shader stages."); + ERR_FAIL_COND_V_MSG(pc_count > 1, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "': Only one push constant is supported, which should be the same across shader stages."); Vector<SpvReflectBlockVariable *> pconstants; pconstants.resize(pc_count); result = spvReflectEnumeratePushConstantBlocks(&module, &pc_count, pconstants.ptrw()); - ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(), - "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed obtaining push constants."); + ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "' failed obtaining push constants."); #if 0 if (pconstants[0] == nullptr) { FileAccess *f = FileAccess::open("res://popo.spv", FileAccess::WRITE); @@ -4399,11 +4712,11 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages } #endif - ERR_FAIL_COND_V_MSG(push_constant.push_constant_size && push_constant.push_constant_size != pconstants[0]->size, RID(), - "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "': Push constant block must be the same across shader stages."); + ERR_FAIL_COND_V_MSG(binary_data.push_constant_size && binary_data.push_constant_size != pconstants[0]->size, Vector<uint8_t>(), + "Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_spirv[i].shader_stage]) + "': Push constant block must be the same across shader stages."); - push_constant.push_constant_size = pconstants[0]->size; - push_constant.push_constants_vk_stage |= shader_stage_masks[stage]; + binary_data.push_constant_size = pconstants[0]->size; + binary_data.push_constants_vk_stage |= shader_stage_masks[stage]; //print_line("Stage: " + String(shader_stage_names[stage]) + " push constant of size=" + itos(push_constant.push_constant_size)); } @@ -4412,9 +4725,291 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages spvReflectDestroyShaderModule(&module); } - stages_processed |= (1 << p_stages[i].shader_stage); + stages_processed |= (1 << p_spirv[i].shader_stage); + } + + Vector<Vector<uint8_t>> compressed_stages; + Vector<uint32_t> smolv_size; + Vector<uint32_t> zstd_size; //if 0, stdno t used + + uint32_t stages_binary_size = 0; + + bool strip_debug = false; + + for (int i = 0; i < p_spirv.size(); i++) { + smolv::ByteArray smolv; + if (!smolv::Encode(p_spirv[i].spir_v.ptr(), p_spirv[i].spir_v.size(), smolv, strip_debug ? smolv::kEncodeFlagStripDebugInfo : 0)) { + ERR_FAIL_V_MSG(Vector<uint8_t>(), "Error compressing shader stage :" + String(shader_stage_names[p_spirv[i].shader_stage])); + } else { + smolv_size.push_back(smolv.size()); + { //zstd + Vector<uint8_t> zstd; + zstd.resize(Compression::get_max_compressed_buffer_size(smolv.size(), Compression::MODE_ZSTD)); + int dst_size = Compression::compress(zstd.ptrw(), &smolv[0], smolv.size(), Compression::MODE_ZSTD); + + if (dst_size > 0 && (uint32_t)dst_size < smolv.size()) { + zstd_size.push_back(dst_size); + zstd.resize(dst_size); + compressed_stages.push_back(zstd); + } else { + Vector<uint8_t> smv; + smv.resize(smolv.size()); + memcpy(smv.ptrw(), &smolv[0], smolv.size()); + zstd_size.push_back(0); //not using zstd + compressed_stages.push_back(smv); + } + } + } + uint32_t s = compressed_stages[i].size(); + if (s % 4 != 0) { + s += 4 - (s % 4); + } + stages_binary_size += s; + } + + binary_data.specialization_constant_count = specialization_constants.size(); + binary_data.set_count = uniform_info.size(); + binary_data.stage_count = p_spirv.size(); + + uint32_t total_size = sizeof(uint32_t) * 3; //header + version + main datasize; + total_size += sizeof(RenderingDeviceVulkanShaderBinaryData); + + for (int i = 0; i < uniform_info.size(); i++) { + total_size += sizeof(uint32_t); + total_size += uniform_info[i].size() * sizeof(RenderingDeviceVulkanShaderBinaryDataBinding); + } + + total_size += sizeof(RenderingDeviceVulkanShaderBinarySpecializationConstant) * specialization_constants.size(); + + total_size += compressed_stages.size() * sizeof(uint32_t) * 3; //sizes + total_size += stages_binary_size; + + Vector<uint8_t> ret; + ret.resize(total_size); + uint32_t offset = 0; + { + uint8_t *binptr = ret.ptrw(); + binptr[0] = 'G'; + binptr[1] = 'V'; + binptr[2] = 'B'; + binptr[3] = 'D'; //godot vulkan binary data + offset += 4; + encode_uint32(SHADER_BINARY_VERSION, binptr + offset); + offset += sizeof(uint32_t); + encode_uint32(sizeof(RenderingDeviceVulkanShaderBinaryData), binptr + offset); + offset += sizeof(uint32_t); + memcpy(binptr + offset, &binary_data, sizeof(RenderingDeviceVulkanShaderBinaryData)); + offset += sizeof(RenderingDeviceVulkanShaderBinaryData); + + for (int i = 0; i < uniform_info.size(); i++) { + int count = uniform_info[i].size(); + encode_uint32(count, binptr + offset); + offset += sizeof(uint32_t); + if (count > 0) { + memcpy(binptr + offset, uniform_info[i].ptr(), sizeof(RenderingDeviceVulkanShaderBinaryDataBinding) * count); + offset += sizeof(RenderingDeviceVulkanShaderBinaryDataBinding) * count; + } + } + + if (specialization_constants.size()) { + memcpy(binptr + offset, specialization_constants.ptr(), sizeof(RenderingDeviceVulkanShaderBinarySpecializationConstant) * specialization_constants.size()); + offset += sizeof(RenderingDeviceVulkanShaderBinarySpecializationConstant) * specialization_constants.size(); + } + + for (int i = 0; i < compressed_stages.size(); i++) { + encode_uint32(p_spirv[i].shader_stage, binptr + offset); + offset += sizeof(uint32_t); + encode_uint32(smolv_size[i], binptr + offset); + offset += sizeof(uint32_t); + encode_uint32(zstd_size[i], binptr + offset); + offset += sizeof(uint32_t); + memcpy(binptr + offset, compressed_stages[i].ptr(), compressed_stages[i].size()); + + uint32_t s = compressed_stages[i].size(); + + if (s % 4 != 0) { + s += 4 - (s % 4); + } + + offset += s; + } + + ERR_FAIL_COND_V(offset != (uint32_t)ret.size(), Vector<uint8_t>()); + } + + return ret; +} + +RID RenderingDeviceVulkan::shader_create_from_bytecode(const Vector<uint8_t> &p_shader_binary) { + const uint8_t *binptr = p_shader_binary.ptr(); + uint32_t binsize = p_shader_binary.size(); + + uint32_t read_offset = 0; + //consistency check + ERR_FAIL_COND_V(binsize < sizeof(uint32_t) * 3 + sizeof(RenderingDeviceVulkanShaderBinaryData), RID()); + ERR_FAIL_COND_V(binptr[0] != 'G' || binptr[1] != 'V' || binptr[2] != 'B' || binptr[3] != 'D', RID()); + + uint32_t bin_version = decode_uint32(binptr + 4); + ERR_FAIL_COND_V(bin_version > SHADER_BINARY_VERSION, RID()); + + uint32_t bin_data_size = decode_uint32(binptr + 8); + + const RenderingDeviceVulkanShaderBinaryData &binary_data = *(const RenderingDeviceVulkanShaderBinaryData *)(binptr + 12); + + Shader::PushConstant push_constant; + push_constant.push_constant_size = binary_data.push_constant_size; + push_constant.push_constants_vk_stage = binary_data.push_constants_vk_stage; + + uint32_t vertex_input_mask = binary_data.vertex_input_mask; + + uint32_t fragment_outputs = binary_data.fragment_outputs; + + bool is_compute = binary_data.is_compute; + + uint32_t compute_local_size[3] = { binary_data.compute_local_size[0], binary_data.compute_local_size[1], binary_data.compute_local_size[2] }; + + read_offset += sizeof(uint32_t) * 3 + bin_data_size; + + Vector<Vector<VkDescriptorSetLayoutBinding>> set_bindings; + Vector<Vector<UniformInfo>> uniform_info; + + set_bindings.resize(binary_data.set_count); + uniform_info.resize(binary_data.set_count); + + for (uint32_t i = 0; i < binary_data.set_count; i++) { + ERR_FAIL_COND_V(read_offset + sizeof(uint32_t) >= binsize, RID()); + uint32_t set_count = decode_uint32(binptr + read_offset); + read_offset += sizeof(uint32_t); + const RenderingDeviceVulkanShaderBinaryDataBinding *set_ptr = (const RenderingDeviceVulkanShaderBinaryDataBinding *)(binptr + read_offset); + uint32_t set_size = set_count * sizeof(RenderingDeviceVulkanShaderBinaryDataBinding); + ERR_FAIL_COND_V(read_offset + set_size >= binsize, RID()); + + for (uint32_t j = 0; j < set_count; j++) { + UniformInfo info; + info.type = UniformType(set_ptr[j].type); + info.length = set_ptr[j].length; + info.binding = set_ptr[j].binding; + info.stages = set_ptr[j].stages; + + VkDescriptorSetLayoutBinding layout_binding; + layout_binding.pImmutableSamplers = nullptr; + layout_binding.binding = set_ptr[j].binding; + layout_binding.descriptorCount = 1; + layout_binding.stageFlags = 0; + for (uint32_t k = 0; k < SHADER_STAGE_MAX; k++) { + if (set_ptr[j].stages & (1 << k)) { + layout_binding.stageFlags |= shader_stage_masks[k]; + } + } + + switch (info.type) { + case UNIFORM_TYPE_SAMPLER: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLER; + layout_binding.descriptorCount = set_ptr[j].length; + } break; + case UNIFORM_TYPE_SAMPLER_WITH_TEXTURE: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; + layout_binding.descriptorCount = set_ptr[j].length; + } break; + case UNIFORM_TYPE_TEXTURE: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE; + layout_binding.descriptorCount = set_ptr[j].length; + } break; + case UNIFORM_TYPE_IMAGE: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE; + layout_binding.descriptorCount = set_ptr[j].length; + } break; + case UNIFORM_TYPE_TEXTURE_BUFFER: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; + layout_binding.descriptorCount = set_ptr[j].length; + } break; + case UNIFORM_TYPE_IMAGE_BUFFER: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER; + } break; + case UNIFORM_TYPE_UNIFORM_BUFFER: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; + } break; + case UNIFORM_TYPE_STORAGE_BUFFER: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER; + } break; + case UNIFORM_TYPE_INPUT_ATTACHMENT: { + layout_binding.descriptorType = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT; + } break; + default: { + ERR_FAIL_V(RID()); + } + } + + set_bindings.write[i].push_back(layout_binding); + uniform_info.write[i].push_back(info); + } + + read_offset += set_size; + } + + ERR_FAIL_COND_V(read_offset + binary_data.specialization_constant_count * sizeof(RenderingDeviceVulkanShaderBinarySpecializationConstant) >= binsize, RID()); + + Vector<Shader::SpecializationConstant> specialization_constants; + + for (uint32_t i = 0; i < binary_data.specialization_constant_count; i++) { + const RenderingDeviceVulkanShaderBinarySpecializationConstant &src_sc = *(const RenderingDeviceVulkanShaderBinarySpecializationConstant *)(binptr + read_offset); + Shader::SpecializationConstant sc; + sc.constant.int_value = src_sc.int_value; + sc.constant.type = PipelineSpecializationConstantType(src_sc.type); + sc.constant.constant_id = src_sc.constant_id; + sc.stage_flags = src_sc.stage_flags; + specialization_constants.push_back(sc); + + read_offset += sizeof(RenderingDeviceVulkanShaderBinarySpecializationConstant); + } + + Vector<Vector<uint8_t>> stage_spirv_data; + Vector<ShaderStage> stage_type; + + for (uint32_t i = 0; i < binary_data.stage_count; i++) { + ERR_FAIL_COND_V(read_offset + sizeof(uint32_t) * 3 >= binsize, RID()); + uint32_t stage = decode_uint32(binptr + read_offset); + read_offset += sizeof(uint32_t); + uint32_t smolv_size = decode_uint32(binptr + read_offset); + read_offset += sizeof(uint32_t); + uint32_t zstd_size = decode_uint32(binptr + read_offset); + read_offset += sizeof(uint32_t); + + uint32_t buf_size = (zstd_size > 0) ? zstd_size : smolv_size; + + Vector<uint8_t> smolv; + const uint8_t *src_smolv = nullptr; + + if (zstd_size > 0) { + //decompress to smolv + smolv.resize(smolv_size); + int dec_smolv_size = Compression::decompress(smolv.ptrw(), smolv.size(), binptr + read_offset, zstd_size, Compression::MODE_ZSTD); + ERR_FAIL_COND_V(dec_smolv_size != (int32_t)smolv_size, RID()); + src_smolv = smolv.ptr(); + } else { + src_smolv = binptr + read_offset; + } + + Vector<uint8_t> spirv; + uint32_t spirv_size = smolv::GetDecodedBufferSize(src_smolv, smolv_size); + spirv.resize(spirv_size); + if (!smolv::Decode(src_smolv, smolv_size, spirv.ptrw(), spirv_size)) { + ERR_FAIL_V_MSG(RID(), "Malformed smolv input uncompressing shader stage:" + String(shader_stage_names[stage])); + } + stage_spirv_data.push_back(spirv); + stage_type.push_back(ShaderStage(stage)); + + if (buf_size % 4 != 0) { + buf_size += 4 - (buf_size % 4); + } + + ERR_FAIL_COND_V(read_offset + buf_size > binsize, RID()); + + read_offset += buf_size; } + ERR_FAIL_COND_V(read_offset != binsize, RID()); + //all good, let's create modules _THREAD_SAFE_METHOD_ @@ -4422,23 +5017,24 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages Shader shader; shader.vertex_input_mask = vertex_input_mask; - shader.fragment_outputs = fragment_outputs; + shader.fragment_output_mask = fragment_outputs; shader.push_constant = push_constant; shader.is_compute = is_compute; shader.compute_local_size[0] = compute_local_size[0]; shader.compute_local_size[1] = compute_local_size[1]; shader.compute_local_size[2] = compute_local_size[2]; + shader.specialization_constants = specialization_constants; String error_text; bool success = true; - for (int i = 0; i < p_stages.size(); i++) { + for (int i = 0; i < stage_spirv_data.size(); i++) { VkShaderModuleCreateInfo shader_module_create_info; shader_module_create_info.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO; shader_module_create_info.pNext = nullptr; shader_module_create_info.flags = 0; - shader_module_create_info.codeSize = p_stages[i].spir_v.size(); - const uint8_t *r = p_stages[i].spir_v.ptr(); + shader_module_create_info.codeSize = stage_spirv_data[i].size(); + const uint8_t *r = stage_spirv_data[i].ptr(); shader_module_create_info.pCode = (const uint32_t *)r; @@ -4446,7 +5042,7 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages VkResult res = vkCreateShaderModule(device, &shader_module_create_info, nullptr, &module); if (res) { success = false; - error_text = "Error (" + itos(res) + ") creating shader module for stage: " + String(shader_stage_names[p_stages[i].shader_stage]); + error_text = "Error (" + itos(res) + ") creating shader module for stage: " + String(shader_stage_names[stage_type[i]]); break; } @@ -4462,7 +5058,7 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages shader_stage.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO; shader_stage.pNext = nullptr; shader_stage.flags = 0; - shader_stage.stage = shader_stage_bits[p_stages[i].shader_stage]; + shader_stage.stage = shader_stage_bits[stage_type[i]]; shader_stage.module = module; shader_stage.pName = "main"; shader_stage.pSpecializationInfo = nullptr; @@ -4910,7 +5506,7 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, img_info.sampler = *sampler; img_info.imageView = texture->view; - if (texture->usage_flags & (TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | TEXTURE_USAGE_RESOLVE_ATTACHMENT_BIT)) { + if (texture->usage_flags & (TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | TEXTURE_USAGE_INPUT_ATTACHMENT_BIT)) { UniformSet::AttachableTexture attachable_texture; attachable_texture.bind = set_uniform.binding; attachable_texture.texture = texture->owner.is_valid() ? texture->owner : uniform.ids[j + 1]; @@ -4963,7 +5559,7 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms, img_info.sampler = VK_NULL_HANDLE; img_info.imageView = texture->view; - if (texture->usage_flags & (TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | TEXTURE_USAGE_RESOLVE_ATTACHMENT_BIT)) { + if (texture->usage_flags & (TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | TEXTURE_USAGE_INPUT_ATTACHMENT_BIT)) { UniformSet::AttachableTexture attachable_texture; attachable_texture.bind = set_uniform.binding; attachable_texture.texture = texture->owner.is_valid() ? texture->owner : uniform.ids[j]; @@ -5163,6 +5759,49 @@ 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: { } @@ -5234,6 +5873,13 @@ bool RenderingDeviceVulkan::uniform_set_is_valid(RID p_uniform_set) { return uniform_set_owner.owns(p_uniform_set); } +void RenderingDeviceVulkan::uniform_set_set_invalidation_callback(RID p_uniform_set, UniformSetInvalidatedCallback p_callback, void *p_userdata) { + UniformSet *us = uniform_set_owner.getornull(p_uniform_set); + ERR_FAIL_COND(!us); + us->invalidated_callback = p_callback; + us->invalidated_callback_userdata = p_userdata; +} + Error RenderingDeviceVulkan::buffer_update(RID p_buffer, uint32_t p_offset, uint32_t p_size, const void *p_data, uint32_t p_post_barrier) { _THREAD_SAFE_METHOD_ @@ -5373,7 +6019,7 @@ 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, const Vector<PipelineSpecializationConstant> &p_specialization_constants) { _THREAD_SAFE_METHOD_ //needs a shader @@ -5392,8 +6038,16 @@ 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; @@ -5465,7 +6119,7 @@ RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferForma tessellation_create_info.sType = VK_STRUCTURE_TYPE_PIPELINE_TESSELLATION_STATE_CREATE_INFO; tessellation_create_info.pNext = nullptr; tessellation_create_info.flags = 0; - ERR_FAIL_COND_V(p_rasterization_state.patch_control_points < 1 || p_rasterization_state.patch_control_points > limits.maxTessellationPatchSize, RID()); + ERR_FAIL_COND_V(limits.maxTessellationPatchSize > 0 && (p_rasterization_state.patch_control_points < 1 || p_rasterization_state.patch_control_points > limits.maxTessellationPatchSize), RID()); tessellation_create_info.patchControlPoints = p_rasterization_state.patch_control_points; VkPipelineViewportStateCreateInfo viewport_state_create_info; @@ -5578,44 +6232,53 @@ 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(); @@ -5674,8 +6337,63 @@ RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferForma graphics_pipeline_create_info.pNext = nullptr; graphics_pipeline_create_info.flags = 0; - graphics_pipeline_create_info.stageCount = shader->pipeline_stages.size(); - graphics_pipeline_create_info.pStages = shader->pipeline_stages.ptr(); + Vector<VkPipelineShaderStageCreateInfo> pipeline_stages = shader->pipeline_stages; + Vector<VkSpecializationInfo> specialization_info; + Vector<Vector<VkSpecializationMapEntry>> specialization_map_entries; + Vector<uint32_t> specialization_constant_data; + + if (shader->specialization_constants.size()) { + specialization_constant_data.resize(shader->specialization_constants.size()); + uint32_t *data_ptr = specialization_constant_data.ptrw(); + specialization_info.resize(pipeline_stages.size()); + specialization_map_entries.resize(pipeline_stages.size()); + for (int i = 0; i < shader->specialization_constants.size(); i++) { + //see if overridden + const Shader::SpecializationConstant &sc = shader->specialization_constants[i]; + data_ptr[i] = sc.constant.int_value; //just copy the 32 bits + + for (int j = 0; j < p_specialization_constants.size(); j++) { + const PipelineSpecializationConstant &psc = p_specialization_constants[j]; + if (psc.constant_id == sc.constant.constant_id) { + ERR_FAIL_COND_V_MSG(psc.type != sc.constant.type, RID(), "Specialization constant provided for id (" + itos(sc.constant.constant_id) + ") is of the wrong type."); + data_ptr[i] = psc.int_value; + break; + } + } + + VkSpecializationMapEntry entry; + + entry.constantID = sc.constant.constant_id; + entry.offset = i * sizeof(uint32_t); + entry.size = sizeof(uint32_t); + + for (int j = 0; j < SHADER_STAGE_MAX; j++) { + if (sc.stage_flags & (1 << j)) { + VkShaderStageFlagBits stage = shader_stage_masks[j]; + for (int k = 0; k < pipeline_stages.size(); k++) { + if (pipeline_stages[k].stage == stage) { + specialization_map_entries.write[k].push_back(entry); + } + } + } + } + } + + for (int i = 0; i < pipeline_stages.size(); i++) { + if (specialization_map_entries[i].size()) { + specialization_info.write[i].dataSize = specialization_constant_data.size() * sizeof(uint32_t); + specialization_info.write[i].pData = data_ptr; + specialization_info.write[i].mapEntryCount = specialization_map_entries[i].size(); + specialization_info.write[i].pMapEntries = specialization_map_entries[i].ptr(); + + pipeline_stages.write[i].pSpecializationInfo = specialization_info.ptr() + i; + } + } + } + + graphics_pipeline_create_info.stageCount = pipeline_stages.size(); + graphics_pipeline_create_info.pStages = pipeline_stages.ptr(); + graphics_pipeline_create_info.pVertexInputState = &pipeline_vertex_input_state_create_info; graphics_pipeline_create_info.pInputAssemblyState = &input_assembly_create_info; graphics_pipeline_create_info.pTessellationState = &tessellation_create_info; @@ -5688,7 +6406,7 @@ 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; @@ -5705,6 +6423,7 @@ RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferForma #ifdef DEBUG_ENABLED pipeline.validation.dynamic_state = p_dynamic_state_flags; pipeline.validation.framebuffer_format = p_framebuffer_format; + pipeline.validation.render_pass = p_for_render_pass; pipeline.validation.vertex_format = p_vertex_format; pipeline.validation.uses_restart_indices = input_assembly_create_info.primitiveRestartEnable; @@ -5743,7 +6462,7 @@ bool RenderingDeviceVulkan::render_pipeline_is_valid(RID p_pipeline) { /**** COMPUTE PIPELINE ****/ /**************************/ -RID RenderingDeviceVulkan::compute_pipeline_create(RID p_shader) { +RID RenderingDeviceVulkan::compute_pipeline_create(RID p_shader, const Vector<PipelineSpecializationConstant> &p_specialization_constants) { _THREAD_SAFE_METHOD_ //needs a shader @@ -5765,6 +6484,44 @@ RID RenderingDeviceVulkan::compute_pipeline_create(RID p_shader) { compute_pipeline_create_info.basePipelineHandle = VK_NULL_HANDLE; compute_pipeline_create_info.basePipelineIndex = 0; + VkSpecializationInfo specialization_info; + Vector<VkSpecializationMapEntry> specialization_map_entries; + Vector<uint32_t> specialization_constant_data; + + if (shader->specialization_constants.size()) { + specialization_constant_data.resize(shader->specialization_constants.size()); + uint32_t *data_ptr = specialization_constant_data.ptrw(); + for (int i = 0; i < shader->specialization_constants.size(); i++) { + //see if overridden + const Shader::SpecializationConstant &sc = shader->specialization_constants[i]; + data_ptr[i] = sc.constant.int_value; //just copy the 32 bits + + for (int j = 0; j < p_specialization_constants.size(); j++) { + const PipelineSpecializationConstant &psc = p_specialization_constants[j]; + if (psc.constant_id == sc.constant.constant_id) { + ERR_FAIL_COND_V_MSG(psc.type != sc.constant.type, RID(), "Specialization constant provided for id (" + itos(sc.constant.constant_id) + ") is of the wrong type."); + data_ptr[i] = sc.constant.int_value; + break; + } + } + + VkSpecializationMapEntry entry; + + entry.constantID = sc.constant.constant_id; + entry.offset = i * sizeof(uint32_t); + entry.size = sizeof(uint32_t); + + specialization_map_entries.push_back(entry); + } + + specialization_info.dataSize = specialization_constant_data.size() * sizeof(uint32_t); + specialization_info.pData = data_ptr; + specialization_info.mapEntryCount = specialization_map_entries.size(); + specialization_info.pMapEntries = specialization_map_entries.ptr(); + + compute_pipeline_create_info.stage.pSpecializationInfo = &specialization_info; + } + ComputePipeline pipeline; VkResult err = vkCreateComputePipelines(device, VK_NULL_HANDLE, 1, &compute_pipeline_create_info, nullptr, &pipeline.pipeline); ERR_FAIL_COND_V_MSG(err, RID(), "vkCreateComputePipelines failed with error " + itos(err) + "."); @@ -5843,13 +6600,14 @@ 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; @@ -5857,8 +6615,8 @@ RenderingDevice::DrawListID RenderingDeviceVulkan::draw_list_begin_for_screen(Di render_pass_begin.renderPass = context->window_get_render_pass(p_screen); render_pass_begin.framebuffer = context->window_get_framebuffer(p_screen); - render_pass_begin.renderArea.extent.width = context->window_get_width(p_screen); - render_pass_begin.renderArea.extent.height = context->window_get_height(p_screen); + render_pass_begin.renderArea.extent.width = size.width; + render_pass_begin.renderArea.extent.height = size.height; render_pass_begin.renderArea.offset.x = 0; render_pass_begin.renderArea.offset.y = 0; @@ -5898,18 +6656,19 @@ RenderingDevice::DrawListID RenderingDeviceVulkan::draw_list_begin_for_screen(Di return int64_t(ID_TYPE_DRAW_LIST) << ID_BASE_SHIFT; } -Error RenderingDeviceVulkan::_draw_list_setup_framebuffer(Framebuffer *p_framebuffer, InitialAction p_initial_color_action, FinalAction p_final_color_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, VkFramebuffer *r_framebuffer, VkRenderPass *r_render_pass) { +Error RenderingDeviceVulkan::_draw_list_setup_framebuffer(Framebuffer *p_framebuffer, InitialAction p_initial_color_action, FinalAction p_final_color_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, VkFramebuffer *r_framebuffer, VkRenderPass *r_render_pass, uint32_t *r_subpass_count) { Framebuffer::VersionKey vk; vk.initial_color_action = p_initial_color_action; vk.final_color_action = p_final_color_action; vk.initial_depth_action = p_initial_depth_action; vk.final_depth_action = p_final_depth_action; + vk.view_count = p_framebuffer->view_count; if (!p_framebuffer->framebuffers.has(vk)) { //need to create this version Framebuffer::Version version; - version.render_pass = _render_pass_create(framebuffer_formats[p_framebuffer->format_id].E->key().attachments, p_initial_color_action, p_final_color_action, p_initial_depth_action, p_final_depth_action); + version.render_pass = _render_pass_create(framebuffer_formats[p_framebuffer->format_id].E->key().attachments, framebuffer_formats[p_framebuffer->format_id].E->key().passes, p_initial_color_action, p_final_color_action, p_initial_depth_action, p_final_depth_action, p_framebuffer->view_count); VkFramebufferCreateInfo framebuffer_create_info; framebuffer_create_info.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO; @@ -5933,11 +6692,14 @@ 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; } @@ -6127,15 +6889,23 @@ 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; @@ -6145,13 +6915,14 @@ RenderingDevice::DrawListID RenderingDeviceVulkan::draw_list_begin(RID p_framebu return INVALID_ID; } - draw_list = memnew(DrawList); - draw_list->command_buffer = command_buffer; + draw_list_render_pass = render_pass; + draw_list_vkframebuffer = vkframebuffer; + + _draw_list_allocate(Rect2i(viewport_offset, viewport_size), 0, 0); #ifdef DEBUG_ENABLED - draw_list->validation.framebuffer_format = framebuffer->format_id; + draw_list_framebuffer_format = framebuffer->format_id; #endif - draw_list_count = 0; - draw_list_split = false; + draw_list_current_subpass = 0; if (needs_clear_color || needs_clear_depth) { _draw_list_insert_clear_region(draw_list, framebuffer, viewport_offset, viewport_size, needs_clear_color, p_clear_color_values, needs_clear_depth, p_clear_depth, p_clear_stencil); @@ -6175,7 +6946,6 @@ 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; } @@ -6217,47 +6987,23 @@ 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; @@ -6267,53 +7013,24 @@ 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) + "."); - } + draw_list_current_subpass = 0; - res = vkBeginCommandBuffer(command_buffer, &cmdbuf_begin); - if (res) { - memdelete_arr(draw_list); - draw_list = nullptr; - ERR_FAIL_V_MSG(ERR_CANT_CREATE, "vkBeginCommandBuffer failed with error " + itos(res) + "."); - } - - draw_list[i].command_buffer = command_buffer; #ifdef DEBUG_ENABLED - draw_list[i].validation.framebuffer_format = framebuffer->format_id; + draw_list_framebuffer_format = framebuffer->format_id; #endif + draw_list_render_pass = render_pass; + draw_list_vkframebuffer = vkframebuffer; - if (i == 0 && (needs_clear_color || needs_clear_depth)) { - _draw_list_insert_clear_region(draw_list, framebuffer, viewport_offset, viewport_size, needs_clear_color, p_clear_color_values, needs_clear_depth, p_clear_depth, p_clear_stencil); - } + err = _draw_list_allocate(Rect2i(viewport_offset, viewport_size), p_splits, 0); + if (err != OK) { + return err; + } + + if (needs_clear_color || needs_clear_depth) { + _draw_list_insert_clear_region(&draw_list[0], framebuffer, viewport_offset, viewport_size, needs_clear_color, p_clear_color_values, needs_clear_depth, p_clear_depth, p_clear_stencil); + } + for (uint32_t i = 0; i < p_splits; i++) { VkViewport viewport; viewport.x = viewport_offset.x; viewport.y = viewport_offset.y; @@ -6322,7 +7039,7 @@ Error RenderingDeviceVulkan::draw_list_begin_split(RID p_framebuffer, uint32_t p viewport.minDepth = 0; viewport.maxDepth = 1.0; - vkCmdSetViewport(command_buffer, 0, 1, &viewport); + vkCmdSetViewport(draw_list[i].command_buffer, 0, 1, &viewport); VkRect2D scissor; scissor.offset.x = viewport_offset.x; @@ -6330,10 +7047,8 @@ 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; @@ -6378,7 +7093,7 @@ 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) { @@ -6717,30 +7432,166 @@ void RenderingDeviceVulkan::draw_list_disable_scissor(DrawListID p_list) { vkCmdSetScissor(dl->command_buffer, 0, 1, &scissor); } -void RenderingDeviceVulkan::draw_list_end(uint32_t p_post_barrier) { - _THREAD_SAFE_METHOD_ +uint32_t RenderingDeviceVulkan::draw_list_get_current_pass() { + return draw_list_current_subpass; +} - ERR_FAIL_COND_MSG(!draw_list, "Immediate draw list is already inactive."); +RenderingDevice::DrawListID RenderingDeviceVulkan::draw_list_switch_to_next_pass() { + ERR_FAIL_COND_V(draw_list == nullptr, INVALID_ID); + ERR_FAIL_COND_V(draw_list_current_subpass >= draw_list_subpass_count - 1, INVALID_FORMAT_ID); + + draw_list_current_subpass++; + + Rect2i viewport; + _draw_list_free(&viewport); + vkCmdNextSubpass(frames[frame].draw_command_buffer, VK_SUBPASS_CONTENTS_INLINE); + + _draw_list_allocate(viewport, 0, draw_list_current_subpass); + + return int64_t(ID_TYPE_DRAW_LIST) << ID_BASE_SHIFT; +} +Error RenderingDeviceVulkan::draw_list_switch_to_next_pass_split(uint32_t p_splits, DrawListID *r_split_ids) { + ERR_FAIL_COND_V(draw_list == nullptr, ERR_INVALID_PARAMETER); + ERR_FAIL_COND_V(draw_list_current_subpass >= draw_list_subpass_count - 1, ERR_INVALID_PARAMETER); + + draw_list_current_subpass++; + + Rect2i viewport; + _draw_list_free(&viewport); + + vkCmdNextSubpass(frames[frame].draw_command_buffer, VK_SUBPASS_CONTENTS_INLINE); + + _draw_list_allocate(viewport, p_splits, draw_list_current_subpass); + + for (uint32_t i = 0; i < p_splits; i++) { + r_split_ids[i] = (int64_t(ID_TYPE_SPLIT_DRAW_LIST) << ID_BASE_SHIFT) + i; + } + + return OK; +} + +Error RenderingDeviceVulkan::_draw_list_allocate(const Rect2i &p_viewport, uint32_t p_splits, uint32_t p_subpass) { + 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]); @@ -7465,6 +8316,10 @@ void RenderingDeviceVulkan::_free_internal(RID p_id) { } else if (uniform_set_owner.owns(p_id)) { UniformSet *uniform_set = uniform_set_owner.getornull(p_id); frames[frame].uniform_sets_to_dispose_of.push_back(*uniform_set); + if (uniform_set->invalidated_callback != nullptr) { + uniform_set->invalidated_callback(p_id, uniform_set->invalidated_callback_userdata); + } + uniform_set_owner.free(p_id); } else if (render_pipeline_owner.owns(p_id)) { RenderPipeline *pipeline = render_pipeline_owner.getornull(p_id); @@ -7749,6 +8604,7 @@ void RenderingDeviceVulkan::_free_pending_resources(int p_frame) { vkDestroyImageView(device, texture->view, nullptr); if (texture->owner.is_null()) { //actually owns the image and the allocation too + image_memory -= texture->allocation_info.size; vmaDestroyImage(allocator, texture->image, texture->allocation); } frames[p_frame].textures_to_dispose_of.pop_front(); @@ -7772,10 +8628,16 @@ uint32_t RenderingDeviceVulkan::get_frame_delay() const { return frame_count; } -uint64_t RenderingDeviceVulkan::get_memory_usage() const { - VmaStats stats; - vmaCalculateStats(allocator, &stats); - return stats.total.usedBytes; +uint64_t RenderingDeviceVulkan::get_memory_usage(MemoryType p_type) const { + if (p_type == MEMORY_BUFFERS) { + return buffer_memory; + } else if (p_type == MEMORY_TEXTURES) { + return image_memory; + } else { + VmaStats stats; + vmaCalculateStats(allocator, &stats); + return stats.total.usedBytes; + } } void RenderingDeviceVulkan::_flush(bool p_current_frame) { @@ -7995,8 +8857,8 @@ void RenderingDeviceVulkan::_free_rids(T &p_owner, const char *p_type) { } else { WARN_PRINT(vformat("%d RIDs of type \"%s\" were leaked.", owned.size(), p_type)); } - for (List<RID>::Element *E = owned.front(); E; E = E->next()) { - free(E->get()); + for (const RID &E : owned) { + free(E); } } } @@ -8217,13 +9079,13 @@ void RenderingDeviceVulkan::finalize() { List<RID>::Element *N = E->next(); if (texture_is_shared(E->get())) { free(E->get()); - owned.erase(E->get()); + owned.erase(E); } E = N; } //free non shared second, this will avoid an error trying to free unexisting textures due to dependencies. - for (List<RID>::Element *E = owned.front(); E; E = E->next()) { - free(E->get()); + for (const RID &E : owned) { + free(E); } } } |