Add quirk for relaxed render pass compatibility

As we require a relaxed version of the Vulkan render pass compatibility clause for caching multi-subpass render passes, we now utilize a quirk to determine if this is supported which it is on Nvidia/Adreno while AMD/Mali where it isn't supported we force single-subpass render passes.

app/src/main/cpp/skyline/gpu/interconnect/command_executor.cpp

@@ -51,7 +51,13 @@ namespace skyline::gpu::interconnect {
         cycle->AttachObject(dependency);
     }
 
-    void CommandExecutor::AddSubpass(std::function<void(vk::raii::CommandBuffer &, const std::shared_ptr<FenceCycle> &, GPU &, vk::RenderPass, u32)> &&function, vk::Rect2D renderArea, span<TextureView *> inputAttachments, span<TextureView *> colorAttachments, TextureView *depthStencilAttachment) {
+    void CommandExecutor::AddSubpass(std::function<void(vk::raii::CommandBuffer &, const std::shared_ptr<FenceCycle> &, GPU &, vk::RenderPass, u32)> &&function, vk::Rect2D renderArea, span<TextureView *> inputAttachments, span<TextureView *> colorAttachments, TextureView *depthStencilAttachment, bool exclusiveSubpass) {
+        if (exclusiveSubpass && renderPass) {
+            nodes.emplace_back(std::in_place_type_t<node::RenderPassEndNode>());
+            renderPass = nullptr;
+            subpassCount = 0;
+        }
+
         bool newRenderPass{CreateRenderPass(renderArea)};
         renderPass->AddSubpass(inputAttachments, colorAttachments, depthStencilAttachment ? &*depthStencilAttachment : nullptr);
         if (newRenderPass)

app/src/main/cpp/skyline/gpu/interconnect/command_executor.h

@@ -57,9 +57,10 @@ namespace skyline::gpu::interconnect {
 
         /**
          * @brief Adds a command that needs to be executed inside a subpass configured with certain attachments
+         * @param exclusiveSubpass If this subpass should be the only subpass in a render pass
          * @note Any supplied texture should be attached prior and not undergo any persistent layout transitions till execution
          */
-        void AddSubpass(std::function<void(vk::raii::CommandBuffer &, const std::shared_ptr<FenceCycle> &, GPU &, vk::RenderPass, u32)> &&function, vk::Rect2D renderArea, span<TextureView *> inputAttachments = {}, span<TextureView *> colorAttachments = {}, TextureView *depthStencilAttachment = {});
+        void AddSubpass(std::function<void(vk::raii::CommandBuffer &, const std::shared_ptr<FenceCycle> &, GPU &, vk::RenderPass, u32)> &&function, vk::Rect2D renderArea, span<TextureView *> inputAttachments = {}, span<TextureView *> colorAttachments = {}, TextureView *depthStencilAttachment = {}, bool exclusiveSubpass = false);
 
         /**
          * @brief Adds a subpass that clears the entirety of the specified attachment with a color value, it may utilize VK_ATTACHMENT_LOAD_OP_CLEAR for a more efficient clear when possible

app/src/main/cpp/skyline/gpu/interconnect/graphics_context.h

@@ -2897,7 +2897,7 @@ namespace skyline::gpu::interconnect {
                 cycle->AttachObject(drawStorage);
             }, vk::Rect2D{
                 .extent = activeColorRenderTargets.empty() ? depthRenderTarget.guest.dimensions : activeColorRenderTargets.front()->texture->dimensions,
-            }, {}, activeColorRenderTargets, depthRenderTargetView);
+            }, {}, activeColorRenderTargets, depthRenderTargetView, !gpu.traits.quirks.relaxedRenderPassCompatibility);
         }
 
         void DrawVertex(u32 vertexCount, u32 firstVertex) {

app/src/main/cpp/skyline/gpu/trait_manager.cpp

@@ -143,6 +143,7 @@ namespace skyline::gpu {
                 needsIndividualTextureBindingWrites = true;
                 vkImageMutableFormatCostly = true; // Disables UBWC
                 brokenDescriptorAliasing = true;
+                relaxedRenderPassCompatibility = true; // Adreno drivers support relaxed render pass compatibility rules
 
                 if (deviceProperties.driverVersion < VK_MAKE_VERSION(512, 600, 0))
                     maxSubpassCount = 64; // Driver will segfault while destroying the renderpass and associated objects if this is exceeded on all 5xx and below drivers
@@ -160,6 +161,11 @@ namespace skyline::gpu {
                 break;
             }
 
+            case vk::DriverId::eNvidiaProprietary: {
+                relaxedRenderPassCompatibility = true;
+                break;
+            }
+
             case vk::DriverId::eAmdProprietary: {
                 maxGlobalPriority = vk::QueueGlobalPriorityEXT::eHigh;
                 break;
@@ -172,8 +178,8 @@ namespace skyline::gpu {
 
     std::string TraitManager::QuirkManager::Summary() {
         return fmt::format(
-            "\n* Needs Individual Texture Binding Writes: {}\n* VkImage Mutable Format is costly: {}\n* Broken Descriptor Aliasing: {}\n* Max Subpass Count: {}\n* Max Global Queue Priority: {}",
+            "\n* Needs Individual Texture Binding Writes: {}\n* VkImage Mutable Format is costly: {}\n* Broken Descriptor Aliasing: {}\n* Relaxed Render Pass Compatibility: {}\n* Max Subpass Count: {}\n* Max Global Queue Priority: {}",
-            needsIndividualTextureBindingWrites, vkImageMutableFormatCostly, brokenDescriptorAliasing, maxSubpassCount, vk::to_string(maxGlobalPriority)
+            needsIndividualTextureBindingWrites, vkImageMutableFormatCostly, brokenDescriptorAliasing, relaxedRenderPassCompatibility, maxSubpassCount, vk::to_string(maxGlobalPriority)
         );
     }
 

app/src/main/cpp/skyline/gpu/trait_manager.h

@@ -46,9 +46,10 @@ namespace skyline::gpu {
             bool needsIndividualTextureBindingWrites{}; //!< [Adreno Proprietary] A bug that requires descriptor set writes for VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER to be done individually with descriptorCount = 1 rather than batched
             bool vkImageMutableFormatCostly{}; //!< [Adreno Proprietary/Freedreno] An indication that VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT is costly and should not be enabled unless absolutely necessary (Disables UBWC on Adreno GPUs)
             bool brokenDescriptorAliasing{}; //!< [Adreno Proprietary] A bug that causes alised descriptor sets to be incorrectly interpreted by the shader compiler leading to it buggering up LLVM function argument types and crashing
+            bool relaxedRenderPassCompatibility{}; //!< [Adreno Proprietary/Freedreno] A relaxed version of Vulkan specification's render pass compatibility clause which allows for caching pipeline objects for multi-subpass renderpasses, this is intentionally disabled by default as it requires testing prior to enabling
 
             u32 maxSubpassCount{std::numeric_limits<u32>::max()}; //!< The maximum amount of subpasses within a renderpass, this is limited to 64 on older Adreno proprietary drivers
-            vk::QueueGlobalPriorityEXT maximumGlobalPriority{vk::QueueGlobalPriorityEXT::eMedium}; //!< The highest allowed global priority of the queue, drivers will not allow higher priorities to be set on queues
+            vk::QueueGlobalPriorityEXT maxGlobalPriority{vk::QueueGlobalPriorityEXT::eMedium}; //!< The highest allowed global priority of the queue, drivers will not allow higher priorities to be set on queues
 
             QuirkManager() = default;