From 6a1f0aed360d7d5cd35f09f55602878a34552b71 Mon Sep 17 00:00:00 2001 From: Vincent Lejeune Date: Thu, 17 Mar 2016 21:31:34 +0100 Subject: [PATCH 1/2] vulkan: Precompute all possibles render pass --- rpcs3/Emu/RSX/VK/VKGSRender.cpp | 329 +++++++++++++++++++------------- rpcs3/Emu/RSX/VK/VKGSRender.h | 8 +- rpcs3/Emu/RSX/VK/VKHelpers.h | 1 + 3 files changed, 201 insertions(+), 137 deletions(-) diff --git a/rpcs3/Emu/RSX/VK/VKGSRender.cpp b/rpcs3/Emu/RSX/VK/VKGSRender.cpp index f8b9e151e4..65d6cb39cd 100644 --- a/rpcs3/Emu/RSX/VK/VKGSRender.cpp +++ b/rpcs3/Emu/RSX/VK/VKGSRender.cpp @@ -89,6 +89,63 @@ namespace vk } } + /** Maps color_format, depth_stencil_format and color count to an int as below : + * idx = color_count + 5 * depth_stencil_idx + 15 * color_format_idx + * This should perform a 1:1 mapping + */ + + size_t get_render_pass_location(VkFormat color_format, VkFormat depth_stencil_format, u8 color_count) + { + size_t color_format_idx = 0; + size_t depth_format_idx = 0; + + Expects(color_count < 5); + + switch (color_format) + { + case VK_FORMAT_R5G6B5_UNORM_PACK16: + color_format_idx = 0; + break; + case VK_FORMAT_B8G8R8A8_UNORM: + color_format_idx = 1; + break; + case VK_FORMAT_R16G16B16A16_SFLOAT: + color_format_idx = 2; + break; + case VK_FORMAT_R32G32B32A32_SFLOAT: + color_format_idx = 3; + break; + case VK_FORMAT_R8_UINT: + color_format_idx = 4; + break; + case VK_FORMAT_R8G8_UINT: + color_format_idx = 5; + break; + case VK_FORMAT_A1R5G5B5_UNORM_PACK16: + color_format_idx = 6; + break; + case VK_FORMAT_R32_SFLOAT: + color_format_idx = 7; + break; + } + + switch (depth_stencil_format) + { + case VK_FORMAT_D16_UNORM: + depth_format_idx = 0; + break; + case VK_FORMAT_D24_UNORM_S8_UINT: + case VK_FORMAT_D32_SFLOAT_S8_UINT: + depth_format_idx = 1; + break; + case VK_FORMAT_UNDEFINED: + depth_format_idx = 2; + break; + } + + return color_count + 5 * depth_format_idx + 5 * 3 * color_format_idx; + } + std::vector get_draw_buffers(rsx::surface_target fmt) { switch (fmt) @@ -149,6 +206,87 @@ namespace vk } } + +namespace +{ + VkRenderPass precompute_render_pass(VkDevice dev, VkFormat color_format, u8 number_of_color_surface, VkFormat depth_format) + { + /* Describe a render pass and framebuffer attachments */ + std::vector attachments = {}; + std::vector attachment_references; + + VkAttachmentDescription color_attachement_description = {}; + color_attachement_description.format = color_format; + color_attachement_description.samples = VK_SAMPLE_COUNT_1_BIT; + color_attachement_description.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD; + color_attachement_description.storeOp = VK_ATTACHMENT_STORE_OP_STORE; + color_attachement_description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; + color_attachement_description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; + color_attachement_description.initialLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; + color_attachement_description.finalLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; + for (u32 i = 0; i < number_of_color_surface; ++i) + { + attachments.push_back(color_attachement_description); + attachment_references.push_back({ i, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL }); + } + + if (depth_format != VK_FORMAT_UNDEFINED) + { + VkAttachmentDescription depth_attachement_description = {}; + depth_attachement_description.format = depth_format; + depth_attachement_description.samples = VK_SAMPLE_COUNT_1_BIT; + depth_attachement_description.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD; + depth_attachement_description.storeOp = VK_ATTACHMENT_STORE_OP_STORE; + depth_attachement_description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD; + depth_attachement_description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE; + depth_attachement_description.initialLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; + depth_attachement_description.finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; + attachments.push_back(depth_attachement_description); + + attachment_references.push_back({ number_of_color_surface, VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL }); + } + + VkSubpassDescription subpass = {}; + subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS; + subpass.colorAttachmentCount = number_of_color_surface; + subpass.pColorAttachments = number_of_color_surface > 0 ? attachment_references.data() : nullptr; + subpass.pDepthStencilAttachment = depth_format != VK_FORMAT_UNDEFINED ? &attachment_references.back() : nullptr; + + VkRenderPassCreateInfo rp_info = {}; + rp_info.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO; + rp_info.attachmentCount = attachments.size(); + rp_info.pAttachments = attachments.data(); + rp_info.subpassCount = 1; + rp_info.pSubpasses = &subpass; + + VkRenderPass result; + CHECK_RESULT(vkCreateRenderPass(dev, &rp_info, NULL, &result)); + return result; + } + + std::array get_precomputed_render_passes(VkDevice dev, const vk::gpu_formats_support &gpu_format_support) + { + std::array result = {}; + + const std::array depth_format_list = { VK_FORMAT_UNDEFINED, VK_FORMAT_D16_UNORM, gpu_format_support.d24_unorm_s8 ? VK_FORMAT_D24_UNORM_S8_UINT : VK_FORMAT_D32_SFLOAT_S8_UINT }; + const std::array color_format_list = { VK_FORMAT_R5G6B5_UNORM_PACK16, VK_FORMAT_B8G8R8A8_UNORM, VK_FORMAT_R16G16B16A16_SFLOAT, VK_FORMAT_R32G32B32A32_SFLOAT, VK_FORMAT_R8_UINT, VK_FORMAT_R8G8_UINT, VK_FORMAT_A1R5G5B5_UNORM_PACK16, VK_FORMAT_R32_SFLOAT }; + + + for (const VkFormat &color_format : color_format_list) + { + for (const VkFormat &depth_stencil_format : depth_format_list) + { + for (u8 number_of_draw_buffer = 0; number_of_draw_buffer <= 4; number_of_draw_buffer++) + { + size_t idx = vk::get_render_pass_location(color_format, depth_stencil_format, number_of_draw_buffer); + result[idx] = precompute_render_pass(dev, color_format, number_of_draw_buffer, depth_stencil_format); + } + } + } + return result; + } +} + VKGSRender::VKGSRender() : GSRender(frame_type::Vulkan) { shaders_cache.load(rsx::shader_language::glsl); @@ -213,6 +351,8 @@ VKGSRender::VKGSRender() : GSRender(frame_type::Vulkan) m_uniform_buffer.reset(new vk::buffer(*m_device, RING_BUFFER_SIZE, m_memory_type_mapping.host_visible_coherent, VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, 0)); m_index_buffer_ring_info.init(RING_BUFFER_SIZE); m_index_buffer.reset(new vk::buffer(*m_device, RING_BUFFER_SIZE, m_memory_type_mapping.host_visible_coherent, VK_BUFFER_USAGE_INDEX_BUFFER_BIT, 0)); + + m_render_passes = get_precomputed_render_passes(*m_device, m_optimal_tiling_supported_formats); } VKGSRender::~VKGSRender() @@ -235,8 +375,9 @@ VKGSRender::~VKGSRender() //TODO: Properly destroy shader modules instead of calling clear... m_prog_buffer.clear(); - if (m_render_pass) - destroy_render_pass(); + for (auto &render_pass : m_render_passes) + if (render_pass) + vkDestroyRenderPass(*m_device, render_pass, nullptr); m_command_buffer.destroy(); m_command_buffer_pool.destroy(); @@ -375,9 +516,15 @@ namespace void VKGSRender::end() { + size_t idx = vk::get_render_pass_location( + vk::get_compatible_surface_format(m_surface.color_format), + vk::get_compatible_depth_surface_format(m_optimal_tiling_supported_formats, m_surface.depth_format), + (u8)vk::get_draw_buffers(rsx::to_surface_target(rsx::method_registers[NV4097_SET_SURFACE_COLOR_TARGET])).size()); + VkRenderPass current_render_pass = m_render_passes[idx]; + VkRenderPassBeginInfo rp_begin = {}; rp_begin.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO; - rp_begin.renderPass = m_render_pass; + rp_begin.renderPass = current_render_pass; rp_begin.framebuffer = m_framebuffer; rp_begin.renderArea.offset.x = 0; rp_begin.renderArea.offset.y = 0; @@ -406,7 +553,7 @@ void VKGSRender::end() auto upload_info = upload_vertex_data(); m_program->set_primitive_topology(std::get<0>(upload_info)); - m_program->use(m_command_buffer, m_render_pass, 0); + m_program->use(m_command_buffer, current_render_pass, 0); if (!std::get<1>(upload_info)) vkCmdDraw(m_command_buffer, vertex_draw_count, 1, 0, 0); @@ -602,74 +749,6 @@ bool VKGSRender::do_method(u32 cmd, u32 arg) } } -void VKGSRender::init_render_pass(VkFormat surface_format, VkFormat depth_format, u8 num_draw_buffers, u8 *draw_buffers) -{ - //TODO: Create buffers as requested by the game. Render to swapchain for now.. - /* Describe a render pass and framebuffer attachments */ - std::array attachments; - - attachments[0].format = surface_format; - attachments[0].samples = VK_SAMPLE_COUNT_1_BIT; - attachments[0].loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; //Set to clear removes warnings about empty contents after flip; overwrites previous calls - attachments[0].storeOp = VK_ATTACHMENT_STORE_OP_STORE; - attachments[0].stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; - attachments[0].stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; - attachments[0].initialLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; //PRESENT_SRC_KHR?? - attachments[0].finalLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; - - attachments[1].format = depth_format; /* Depth buffer format. Should be more elegant than this */ - attachments[1].samples = VK_SAMPLE_COUNT_1_BIT; - attachments[1].loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; - attachments[1].storeOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; - attachments[1].stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; - attachments[1].stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; - attachments[1].initialLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; - attachments[1].finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; - - VkAttachmentReference template_color_reference = {}; - template_color_reference.attachment = VK_ATTACHMENT_UNUSED; - template_color_reference.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; - - VkAttachmentReference depth_reference = {}; - depth_reference.attachment = num_draw_buffers; - depth_reference.layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; - - //Fill in draw_buffers information... - std::array real_attachments; - std::array color_references; - - for (int i = 0; i < num_draw_buffers; ++i) - { - real_attachments[i] = attachments[0]; - - color_references[i] = template_color_reference; - color_references[i].attachment = (draw_buffers)? draw_buffers[i]: i; - } - - real_attachments[num_draw_buffers] = attachments[1]; - - VkSubpassDescription subpass = {}; - subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS; - subpass.colorAttachmentCount = num_draw_buffers; - subpass.pColorAttachments = num_draw_buffers? color_references.data() : nullptr; - subpass.pDepthStencilAttachment = &depth_reference; - - VkRenderPassCreateInfo rp_info = {}; - rp_info.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO; - rp_info.attachmentCount = num_draw_buffers + 1; - rp_info.pAttachments = real_attachments.data(); - rp_info.subpassCount = 1; - rp_info.pSubpasses = &subpass; - - CHECK_RESULT(vkCreateRenderPass((*m_device), &rp_info, NULL, &m_render_pass)); -} - -void VKGSRender::destroy_render_pass() -{ - vkDestroyRenderPass((*m_device), m_render_pass, nullptr); - m_render_pass = nullptr; -} - bool VKGSRender::load_program() { RSXVertexProgram vertex_program = get_current_vertex_program(); @@ -874,21 +953,10 @@ void VKGSRender::prepare_rtts() fbo_images.push_back(depth_image); } - if (reconfigure_render_pass) - { - //Create render pass with draw_buffers information - //Somewhat simliar to glDrawBuffers + size_t idx = vk::get_render_pass_location(vk::get_compatible_surface_format(m_surface.color_format), vk::get_compatible_depth_surface_format(m_optimal_tiling_supported_formats, m_surface.depth_format), (u8)draw_buffers.size()); + VkRenderPass current_render_pass = m_render_passes[idx]; - if (m_render_pass) - destroy_render_pass(); - - init_render_pass(vk::get_compatible_surface_format(m_surface.color_format), - vk::get_compatible_depth_surface_format(m_optimal_tiling_supported_formats, m_surface.depth_format), - (u8)draw_buffers.size(), - draw_buffers.data()); - } - - m_framebuffer.create((*m_device), m_render_pass, fbo_images.data(), fbo_images.size(), + m_framebuffer.create((*m_device), current_render_pass, fbo_images.data(), fbo_images.size(), clip_width, clip_height); m_draw_buffers_count = draw_buffers.size(); @@ -971,62 +1039,59 @@ void VKGSRender::flip(int buffer) present.pWaitSemaphores = &m_present_semaphore; present.waitSemaphoreCount = 1; - if (m_render_pass) + begin_command_buffer_recording(); + + if (m_present_semaphore) { - begin_command_buffer_recording(); + //Blit contents to screen.. + VkImage image_to_flip = nullptr; - if (m_present_semaphore) - { - //Blit contents to screen.. - VkImage image_to_flip = nullptr; - - if (std::get<1>(m_rtts.m_bound_render_targets[0]) != nullptr) - image_to_flip = (*std::get<1>(m_rtts.m_bound_render_targets[0])); - else - image_to_flip = (*std::get<1>(m_rtts.m_bound_render_targets[1])); - - VkImage target_image = m_swap_chain->get_swap_chain_image(m_current_present_image); - vk::copy_scaled_image(m_command_buffer, image_to_flip, target_image, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, VK_IMAGE_LAYOUT_PRESENT_SRC_KHR, - buffer_width, buffer_height, aspect_ratio.width, aspect_ratio.height, 1, VK_IMAGE_ASPECT_COLOR_BIT); - } + if (std::get<1>(m_rtts.m_bound_render_targets[0]) != nullptr) + image_to_flip = (*std::get<1>(m_rtts.m_bound_render_targets[0])); else - { - //No draw call was issued! - //TODO: Properly clear the background to rsx value - m_swap_chain->acquireNextImageKHR((*m_device), (*m_swap_chain), ~0ULL, VK_NULL_HANDLE, VK_NULL_HANDLE, &next_image_temp); + image_to_flip = (*std::get<1>(m_rtts.m_bound_render_targets[1])); - VkImageSubresourceRange range = vk::default_image_subresource_range(); - VkClearColorValue clear_black = { 0 }; - vkCmdClearColorImage(m_command_buffer, m_swap_chain->get_swap_chain_image(next_image_temp), VK_IMAGE_LAYOUT_PRESENT_SRC_KHR, &clear_black, 1, &range); - - present.pImageIndices = &next_image_temp; - present.waitSemaphoreCount = 0; - } + VkImage target_image = m_swap_chain->get_swap_chain_image(m_current_present_image); + vk::copy_scaled_image(m_command_buffer, image_to_flip, target_image, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, VK_IMAGE_LAYOUT_PRESENT_SRC_KHR, + buffer_width, buffer_height, aspect_ratio.width, aspect_ratio.height, 1, VK_IMAGE_ASPECT_COLOR_BIT); + } + else + { + //No draw call was issued! + //TODO: Properly clear the background to rsx value + m_swap_chain->acquireNextImageKHR((*m_device), (*m_swap_chain), ~0ULL, VK_NULL_HANDLE, VK_NULL_HANDLE, &next_image_temp); - end_command_buffer_recording(); - execute_command_buffer(false); + VkImageSubresourceRange range = vk::default_image_subresource_range(); + VkClearColorValue clear_black = { 0 }; + vkCmdClearColorImage(m_command_buffer, m_swap_chain->get_swap_chain_image(next_image_temp), VK_IMAGE_LAYOUT_PRESENT_SRC_KHR, &clear_black, 1, &range); - //Check if anything is waiting in queue and wait for it if possible.. - if (m_submit_fence) - { - CHECK_RESULT(vkWaitForFences((*m_device), 1, &m_submit_fence, VK_TRUE, ~0ULL)); + present.pImageIndices = &next_image_temp; + present.waitSemaphoreCount = 0; + } - vkDestroyFence((*m_device), m_submit_fence, nullptr); - m_submit_fence = nullptr; + end_command_buffer_recording(); + execute_command_buffer(false); - CHECK_RESULT(vkResetCommandBuffer(m_command_buffer, 0)); - } + //Check if anything is waiting in queue and wait for it if possible.. + if (m_submit_fence) + { + CHECK_RESULT(vkWaitForFences((*m_device), 1, &m_submit_fence, VK_TRUE, ~0ULL)); - CHECK_RESULT(m_swap_chain->queuePresentKHR(m_swap_chain->get_present_queue(), &present)); - CHECK_RESULT(vkQueueWaitIdle(m_swap_chain->get_present_queue())); + vkDestroyFence((*m_device), m_submit_fence, nullptr); + m_submit_fence = nullptr; - m_uniform_buffer_ring_info.m_get_pos = m_uniform_buffer_ring_info.get_current_put_pos_minus_one(); - m_index_buffer_ring_info.m_get_pos = m_index_buffer_ring_info.get_current_put_pos_minus_one(); - if (m_present_semaphore) - { - vkDestroySemaphore((*m_device), m_present_semaphore, nullptr); - m_present_semaphore = nullptr; - } + CHECK_RESULT(vkResetCommandBuffer(m_command_buffer, 0)); + } + + CHECK_RESULT(m_swap_chain->queuePresentKHR(m_swap_chain->get_present_queue(), &present)); + CHECK_RESULT(vkQueueWaitIdle(m_swap_chain->get_present_queue())); + + m_uniform_buffer_ring_info.m_get_pos = m_uniform_buffer_ring_info.get_current_put_pos_minus_one(); + m_index_buffer_ring_info.m_get_pos = m_index_buffer_ring_info.get_current_put_pos_minus_one(); + if (m_present_semaphore) + { + vkDestroySemaphore((*m_device), m_present_semaphore, nullptr); + m_present_semaphore = nullptr; } //Feed back damaged resources to the main texture cache for management... diff --git a/rpcs3/Emu/RSX/VK/VKGSRender.h b/rpcs3/Emu/RSX/VK/VKGSRender.h index e8636074ff..c5387994a7 100644 --- a/rpcs3/Emu/RSX/VK/VKGSRender.h +++ b/rpcs3/Emu/RSX/VK/VKGSRender.h @@ -139,9 +139,9 @@ private: bool recording = false; bool dirty_frame = true; - //Single render pass - VkRenderPass m_render_pass = nullptr; - + + std::array m_render_passes; + u32 m_draw_calls = 0; u8 m_draw_buffers_count = 0; @@ -153,8 +153,6 @@ public: private: void clear_surface(u32 mask); - void init_render_pass(VkFormat surface_format, VkFormat depth_format, u8 num_draw_buffers, u8 *draw_buffers); - void destroy_render_pass(); void execute_command_buffer(bool wait); void begin_command_buffer_recording(); void end_command_buffer_recording(); diff --git a/rpcs3/Emu/RSX/VK/VKHelpers.h b/rpcs3/Emu/RSX/VK/VKHelpers.h index f3137fa156..50a44f95b7 100644 --- a/rpcs3/Emu/RSX/VK/VKHelpers.h +++ b/rpcs3/Emu/RSX/VK/VKHelpers.h @@ -68,6 +68,7 @@ namespace vk VkFormat get_compatible_sampler_format(u32 format, VkComponentMapping& mapping, u8 swizzle_mask=0); VkFormat get_compatible_surface_format(rsx::surface_color_format color_format); + size_t get_render_pass_location(VkFormat color_surface_format, VkFormat depth_stencil_format, u8 color_surface_count); struct memory_type_mapping { From 24eb5440465495f8e9aa6ec6f429d4522e97a401 Mon Sep 17 00:00:00 2001 From: Vincent Lejeune Date: Thu, 17 Mar 2016 23:57:53 +0100 Subject: [PATCH 2/2] vulkan: Move descriptor sets and layout in VKGSRender class They're now shared between all programs. --- rpcs3/Emu/RSX/VK/VKGSRender.cpp | 99 ++++++++++++++++-- rpcs3/Emu/RSX/VK/VKGSRender.h | 4 + rpcs3/Emu/RSX/VK/VKHelpers.h | 16 +-- rpcs3/Emu/RSX/VK/VKProgramPipeline.cpp | 136 ++----------------------- rpcs3/Emu/RSX/VK/VKVertexBuffers.cpp | 6 +- 5 files changed, 112 insertions(+), 149 deletions(-) diff --git a/rpcs3/Emu/RSX/VK/VKGSRender.cpp b/rpcs3/Emu/RSX/VK/VKGSRender.cpp index 65d6cb39cd..823697dd55 100644 --- a/rpcs3/Emu/RSX/VK/VKGSRender.cpp +++ b/rpcs3/Emu/RSX/VK/VKGSRender.cpp @@ -285,6 +285,67 @@ namespace } return result; } + + std::tuple get_shared_pipeline_layout(VkDevice dev) + { + std::array bindings = {}; + + size_t idx = 0; + // Vertex buffer + for (int i = 0; i < 16; i++) + { + bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; + bindings[idx].descriptorCount = 1; + bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT; + bindings[idx].binding = VERTEX_BUFFERS_FIRST_BIND_SLOT + i; + idx++; + } + + bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; + bindings[idx].descriptorCount = 1; + bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT; + bindings[idx].binding = FRAGMENT_CONSTANT_BUFFERS_BIND_SLOT; + + idx++; + + bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; + bindings[idx].descriptorCount = 1; + bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT; + bindings[idx].binding = VERTEX_CONSTANT_BUFFERS_BIND_SLOT; + + idx++; + + for (int i = 0; i < 16; i++) + { + bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; + bindings[idx].descriptorCount = 1; + bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT; + bindings[idx].binding = TEXTURES_FIRST_BIND_SLOT + i; + idx++; + } + + bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; + bindings[idx].descriptorCount = 1; + bindings[idx].stageFlags = VK_SHADER_STAGE_ALL_GRAPHICS; + bindings[idx].binding = SCALE_OFFSET_BIND_SLOT; + + VkDescriptorSetLayoutCreateInfo infos = {}; + infos.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO; + infos.pBindings = bindings.data(); + infos.bindingCount = bindings.size(); + + VkDescriptorSetLayout set_layout; + CHECK_RESULT(vkCreateDescriptorSetLayout(dev, &infos, nullptr, &set_layout)); + + VkPipelineLayoutCreateInfo layout_info = {}; + layout_info.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO; + layout_info.setLayoutCount = 1; + layout_info.pSetLayouts = &set_layout; + + VkPipelineLayout result; + CHECK_RESULT(vkCreatePipelineLayout(dev, &layout_info, nullptr, &result)); + return std::make_tuple(result, set_layout); + } } VKGSRender::VKGSRender() : GSRender(frame_type::Vulkan) @@ -353,6 +414,24 @@ VKGSRender::VKGSRender() : GSRender(frame_type::Vulkan) m_index_buffer.reset(new vk::buffer(*m_device, RING_BUFFER_SIZE, m_memory_type_mapping.host_visible_coherent, VK_BUFFER_USAGE_INDEX_BUFFER_BIT, 0)); m_render_passes = get_precomputed_render_passes(*m_device, m_optimal_tiling_supported_formats); + + std::tie(pipeline_layout, descriptor_layouts) = get_shared_pipeline_layout(*m_device); + + VkDescriptorPoolSize uniform_buffer_pool = { VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER , 3 }; + VkDescriptorPoolSize uniform_texel_pool = { VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER , 16 }; + VkDescriptorPoolSize texture_pool = { VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER , 16 }; + + std::vector sizes{ uniform_buffer_pool, uniform_texel_pool, texture_pool }; + + descriptor_pool.create(*m_device, sizes.data(), sizes.size()); + + VkDescriptorSetAllocateInfo alloc_info = {}; + alloc_info.descriptorPool = descriptor_pool; + alloc_info.descriptorSetCount = 1; + alloc_info.pSetLayouts = &descriptor_layouts; + alloc_info.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO; + + CHECK_RESULT(vkAllocateDescriptorSets(*m_device, &alloc_info, &descriptor_sets)); } VKGSRender::~VKGSRender() @@ -379,6 +458,12 @@ VKGSRender::~VKGSRender() if (render_pass) vkDestroyRenderPass(*m_device, render_pass, nullptr); + vkFreeDescriptorSets(*m_device, descriptor_pool, 1, &descriptor_sets); + vkDestroyPipelineLayout(*m_device, pipeline_layout, nullptr); + vkDestroyDescriptorSetLayout(*m_device, descriptor_layouts, nullptr); + + descriptor_pool.destroy(); + m_command_buffer.destroy(); m_command_buffer_pool.destroy(); @@ -540,12 +625,12 @@ void VKGSRender::end() { if (!textures[i].enabled()) { - m_program->bind_uniform({ vk::null_sampler(), vk::null_image_view(), VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL}, "tex" + std::to_string(i)); + m_program->bind_uniform({ vk::null_sampler(), vk::null_image_view(), VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL}, "tex" + std::to_string(i), descriptor_sets); continue; } vk::texture &tex = (texture0)? (*texture0): m_texture_cache.upload_texture(m_command_buffer, textures[i], m_rtts); - m_program->bind_uniform({ tex, tex, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL }, "tex" + std::to_string(i)); + m_program->bind_uniform({ tex, tex, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL }, "tex" + std::to_string(i), descriptor_sets); texture0 = &tex; } } @@ -553,8 +638,8 @@ void VKGSRender::end() auto upload_info = upload_vertex_data(); m_program->set_primitive_topology(std::get<0>(upload_info)); - m_program->use(m_command_buffer, current_render_pass, 0); - + m_program->use(m_command_buffer, current_render_pass, pipeline_layout, descriptor_sets); + if (!std::get<1>(upload_info)) vkCmdDraw(m_command_buffer, vertex_draw_count, 1, 0, 0); else @@ -809,9 +894,9 @@ bool VKGSRender::load_program() m_prog_buffer.fill_fragment_constans_buffer({ reinterpret_cast(buf), gsl::narrow(fragment_constants_sz) }, fragment_program); m_uniform_buffer->unmap(); - m_program->bind_uniform({ m_uniform_buffer->value, scale_offset_offset, 256 }, SCALE_OFFSET_BIND_SLOT); - m_program->bind_uniform({ m_uniform_buffer->value, vertex_constants_offset, 512 * 4 * sizeof(float) }, VERTEX_CONSTANT_BUFFERS_BIND_SLOT); - m_program->bind_uniform({ m_uniform_buffer->value, fragment_constants_offset, fragment_constants_sz }, FRAGMENT_CONSTANT_BUFFERS_BIND_SLOT); + m_program->bind_uniform({ m_uniform_buffer->value, scale_offset_offset, 256 }, SCALE_OFFSET_BIND_SLOT, descriptor_sets); + m_program->bind_uniform({ m_uniform_buffer->value, vertex_constants_offset, 512 * 4 * sizeof(float) }, VERTEX_CONSTANT_BUFFERS_BIND_SLOT, descriptor_sets); + m_program->bind_uniform({ m_uniform_buffer->value, fragment_constants_offset, fragment_constants_sz }, FRAGMENT_CONSTANT_BUFFERS_BIND_SLOT, descriptor_sets); return true; } diff --git a/rpcs3/Emu/RSX/VK/VKGSRender.h b/rpcs3/Emu/RSX/VK/VKGSRender.h index c5387994a7..29f89743a8 100644 --- a/rpcs3/Emu/RSX/VK/VKGSRender.h +++ b/rpcs3/Emu/RSX/VK/VKGSRender.h @@ -141,6 +141,10 @@ private: std::array m_render_passes; + VkDescriptorSetLayout descriptor_layouts; + VkDescriptorSet descriptor_sets; + VkPipelineLayout pipeline_layout; + vk::descriptor_pool descriptor_pool; u32 m_draw_calls = 0; diff --git a/rpcs3/Emu/RSX/VK/VKHelpers.h b/rpcs3/Emu/RSX/VK/VKHelpers.h index 50a44f95b7..49572bd79f 100644 --- a/rpcs3/Emu/RSX/VK/VKHelpers.h +++ b/rpcs3/Emu/RSX/VK/VKHelpers.h @@ -1347,10 +1347,6 @@ namespace vk VkShaderModule vs, fs; VkPipeline pipeline_handle = nullptr; - VkDescriptorSetLayout descriptor_layouts; - VkDescriptorSet descriptor_sets; - VkPipelineLayout pipeline_layout; - int num_targets = 1; bool dirty; @@ -1362,7 +1358,6 @@ namespace vk vk::render_device *device = nullptr; std::vector uniforms; - vk::descriptor_pool descriptor_pool; void init_pipeline(); @@ -1394,15 +1389,12 @@ namespace vk void set_blend_op(int num_targets, u8* targets, VkBlendOp* color_ops, VkBlendOp* alpha_ops); void set_blend_op(int num_targets, u8 * targets, VkBlendOp color_op, VkBlendOp alpha_op); void set_primitive_restart(VkBool32 state); - - void init_descriptor_layout(); - void destroy_descriptors(); void set_draw_buffer_count(u8 draw_buffers); program& load_uniforms(program_domain domain, std::vector& inputs); - void use(vk::command_buffer& commands, VkRenderPass pass, u32 subpass); + void use(vk::command_buffer& commands, VkRenderPass pass, VkPipelineLayout pipeline_layout, VkDescriptorSet descriptor_set); bool has_uniform(std::string uniform_name); #define VERTEX_BUFFERS_FIRST_BIND_SLOT 3 @@ -1410,9 +1402,9 @@ namespace vk #define VERTEX_CONSTANT_BUFFERS_BIND_SLOT 1 #define TEXTURES_FIRST_BIND_SLOT 19 #define SCALE_OFFSET_BIND_SLOT 0 - void bind_uniform(VkDescriptorImageInfo image_descriptor, std::string uniform_name); - void bind_uniform(VkDescriptorBufferInfo buffer_descriptor, uint32_t binding_point); - void bind_uniform(const VkBufferView &buffer_view, const std::string &binding_name); + void bind_uniform(VkDescriptorImageInfo image_descriptor, std::string uniform_name, VkDescriptorSet &descriptor_set); + void bind_uniform(VkDescriptorBufferInfo buffer_descriptor, uint32_t binding_point, VkDescriptorSet &descriptor_set); + void bind_uniform(const VkBufferView &buffer_view, const std::string &binding_name, VkDescriptorSet &descriptor_set); program& operator = (const program&) = delete; program& operator = (program&& other); diff --git a/rpcs3/Emu/RSX/VK/VKProgramPipeline.cpp b/rpcs3/Emu/RSX/VK/VKProgramPipeline.cpp index db369fcff1..d72a932e01 100644 --- a/rpcs3/Emu/RSX/VK/VKProgramPipeline.cpp +++ b/rpcs3/Emu/RSX/VK/VKProgramPipeline.cpp @@ -31,10 +31,6 @@ namespace vk uniforms = other.uniforms; other.uniforms = tmp_uniforms; - vk::descriptor_pool tmp_pool; - descriptor_pool = other.descriptor_pool; - other.descriptor_pool = tmp_pool; - vk::render_device *tmp_dev = device; device = other.device; other.device = tmp_dev; @@ -55,10 +51,6 @@ namespace vk uniforms = other.uniforms; other.uniforms = tmp_uniforms; - vk::descriptor_pool tmp_pool; - descriptor_pool = other.descriptor_pool; - other.descriptor_pool = tmp_pool; - vk::render_device *tmp_dev = device; device = other.device; other.device = tmp_dev; @@ -382,108 +374,6 @@ namespace vk } } - namespace - { - std::tuple get_shared_pipeline_layout(VkDevice dev) - { - std::array bindings = {}; - - size_t idx = 0; - // Vertex buffer - for (int i = 0; i < 16; i++) - { - bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; - bindings[idx].descriptorCount = 1; - bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT; - bindings[idx].binding = VERTEX_BUFFERS_FIRST_BIND_SLOT + i; - idx++; - } - - bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; - bindings[idx].descriptorCount = 1; - bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT; - bindings[idx].binding = FRAGMENT_CONSTANT_BUFFERS_BIND_SLOT; - - idx++; - - bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; - bindings[idx].descriptorCount = 1; - bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT; - bindings[idx].binding = VERTEX_CONSTANT_BUFFERS_BIND_SLOT; - - idx++; - - for (int i = 0; i < 16; i++) - { - bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; - bindings[idx].descriptorCount = 1; - bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT; - bindings[idx].binding = TEXTURES_FIRST_BIND_SLOT + i; - idx++; - } - - bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; - bindings[idx].descriptorCount = 1; - bindings[idx].stageFlags = VK_SHADER_STAGE_ALL_GRAPHICS; - bindings[idx].binding = SCALE_OFFSET_BIND_SLOT; - - VkDescriptorSetLayoutCreateInfo infos = {}; - infos.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO; - infos.pBindings = bindings.data(); - infos.bindingCount = bindings.size(); - - VkDescriptorSetLayout set_layout; - CHECK_RESULT(vkCreateDescriptorSetLayout(dev, &infos, nullptr, &set_layout)); - - VkPipelineLayoutCreateInfo layout_info = {}; - layout_info.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO; - layout_info.setLayoutCount = 1; - layout_info.pSetLayouts = &set_layout; - - VkPipelineLayout result; - CHECK_RESULT(vkCreatePipelineLayout(dev, &layout_info, nullptr, &result)); - return std::make_tuple(result, set_layout); - } - } - - void program::init_descriptor_layout() - { - if (descriptor_pool.valid()) - descriptor_pool.destroy(); - - std::tie(pstate.pipeline_layout, pstate.descriptor_layouts) = get_shared_pipeline_layout(*device); - - VkDescriptorPoolSize uniform_buffer_pool = { VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER , 3}; - VkDescriptorPoolSize uniform_texel_pool = { VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER , 16}; - VkDescriptorPoolSize texture_pool = { VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER , 16 }; - - std::vector sizes{ uniform_buffer_pool, uniform_texel_pool, texture_pool }; - - descriptor_pool.create((*device), sizes.data(), sizes.size()); - - VkDescriptorSetAllocateInfo alloc_info = {}; - alloc_info.descriptorPool = descriptor_pool; - alloc_info.descriptorSetCount = 1; - alloc_info.pSetLayouts = &pstate.descriptor_layouts; - alloc_info.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO; - - CHECK_RESULT(vkAllocateDescriptorSets((*device), &alloc_info, &pstate.descriptor_sets)); - } - - void program::destroy_descriptors() - { - if (pstate.descriptor_sets) - vkFreeDescriptorSets((*device), descriptor_pool, 1, &pstate.descriptor_sets); - - if (pstate.pipeline_layout) - vkDestroyPipelineLayout((*device), pstate.pipeline_layout, nullptr); - - if (pstate.descriptor_layouts) - vkDestroyDescriptorSetLayout((*device), pstate.descriptor_layouts, nullptr); - - descriptor_pool.destroy(); - } - void program::set_draw_buffer_count(u8 draw_buffers) { if (pstate.num_targets != draw_buffers) @@ -509,7 +399,7 @@ namespace vk return *this; } - void program::use(vk::command_buffer& commands, VkRenderPass pass, u32 subpass) + void program::use(vk::command_buffer& commands, VkRenderPass pass, VkPipelineLayout pipeline_layout, VkDescriptorSet descriptor_set) { if (/*uniforms_changed*/true) { @@ -535,7 +425,7 @@ namespace vk pstate.pipeline.pDepthStencilState = &pstate.ds; pstate.pipeline.pStages = pstate.shader_stages; pstate.pipeline.pDynamicState = &pstate.dynamic_state; - pstate.pipeline.layout = pstate.pipeline_layout; + pstate.pipeline.layout = pipeline_layout; pstate.pipeline.basePipelineIndex = -1; pstate.pipeline.basePipelineHandle = VK_NULL_HANDLE; @@ -546,7 +436,7 @@ namespace vk } vkCmdBindPipeline(commands, VK_PIPELINE_BIND_POINT_GRAPHICS, pstate.pipeline_handle); - vkCmdBindDescriptorSets(commands, VK_PIPELINE_BIND_POINT_GRAPHICS, pstate.pipeline_layout, 0, 1, &pstate.descriptor_sets, 0, nullptr); + vkCmdBindDescriptorSets(commands, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline_layout, 0, 1, &descriptor_set, 0, nullptr); } bool program::has_uniform(std::string uniform_name) @@ -560,17 +450,15 @@ namespace vk return false; } - void program::bind_uniform(VkDescriptorImageInfo image_descriptor, std::string uniform_name) + void program::bind_uniform(VkDescriptorImageInfo image_descriptor, std::string uniform_name, VkDescriptorSet &descriptor_set) { - if (!pstate.descriptor_layouts) - init_descriptor_layout(); for (auto &uniform : uniforms) { if (uniform.name == uniform_name) { VkWriteDescriptorSet descriptor_writer = {}; descriptor_writer.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET; - descriptor_writer.dstSet = pstate.descriptor_sets; + descriptor_writer.dstSet = descriptor_set; descriptor_writer.descriptorCount = 1; descriptor_writer.pImageInfo = &image_descriptor; descriptor_writer.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; @@ -585,13 +473,11 @@ namespace vk throw EXCEPTION("texture not found"); } - void program::bind_uniform(VkDescriptorBufferInfo buffer_descriptor, uint32_t binding_point) + void program::bind_uniform(VkDescriptorBufferInfo buffer_descriptor, uint32_t binding_point, VkDescriptorSet &descriptor_set) { - if (!pstate.descriptor_layouts) - init_descriptor_layout(); VkWriteDescriptorSet descriptor_writer = {}; descriptor_writer.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET; - descriptor_writer.dstSet = pstate.descriptor_sets; + descriptor_writer.dstSet = descriptor_set; descriptor_writer.descriptorCount = 1; descriptor_writer.pBufferInfo = &buffer_descriptor; descriptor_writer.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; @@ -601,18 +487,15 @@ namespace vk vkUpdateDescriptorSets((*device), 1, &descriptor_writer, 0, nullptr); } - void program::bind_uniform(const VkBufferView &buffer_view, const std::string &binding_name) + void program::bind_uniform(const VkBufferView &buffer_view, const std::string &binding_name, VkDescriptorSet &descriptor_set) { - if (!pstate.descriptor_layouts) - init_descriptor_layout(); - for (auto &uniform : uniforms) { if (uniform.name == binding_name) { VkWriteDescriptorSet descriptor_writer = {}; descriptor_writer.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET; - descriptor_writer.dstSet = pstate.descriptor_sets; + descriptor_writer.dstSet = descriptor_set; descriptor_writer.descriptorCount = 1; descriptor_writer.pTexelBufferView = &buffer_view; descriptor_writer.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; @@ -630,7 +513,6 @@ namespace vk { if (device) { - destroy_descriptors(); uniforms.resize(0); if (pstate.pipeline_handle) diff --git a/rpcs3/Emu/RSX/VK/VKVertexBuffers.cpp b/rpcs3/Emu/RSX/VK/VKVertexBuffers.cpp index 47a3d9368f..28aca5f727 100644 --- a/rpcs3/Emu/RSX/VK/VKVertexBuffers.cpp +++ b/rpcs3/Emu/RSX/VK/VKVertexBuffers.cpp @@ -333,7 +333,7 @@ VKGSRender::upload_vertex_data() buffer.set_format(format); //Link texture to uniform location - m_program->bind_uniform(buffer, reg_table[index]); + m_program->bind_uniform(buffer, reg_table[index], descriptor_sets); } } @@ -421,7 +421,7 @@ VKGSRender::upload_vertex_data() buffer.sub_data(0, data_size, data_ptr); buffer.set_format(format); - m_program->bind_uniform(buffer, reg_table[index]); + m_program->bind_uniform(buffer, reg_table[index], descriptor_sets); } else if (register_vertex_info[index].size > 0) { @@ -460,7 +460,7 @@ VKGSRender::upload_vertex_data() buffer.sub_data(0, data_size, data_ptr); buffer.set_format(format); - m_program->bind_uniform(buffer, reg_table[index]); + m_program->bind_uniform(buffer, reg_table[index], descriptor_sets); break; } default: