Add Vulkan stride dynamic state and robustness support

Fixes the waterfall in SMO by specifying vertex buffer bounds.

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

@@ -233,7 +233,9 @@ namespace skyline::gpu {
             vk::PhysicalDevicePrimitiveTopologyListRestartFeaturesEXT,
             vk::PhysicalDeviceImagelessFramebufferFeatures,
             vk::PhysicalDeviceTransformFeedbackFeaturesEXT,
-            vk::PhysicalDeviceIndexTypeUint8FeaturesEXT>()};
+            vk::PhysicalDeviceIndexTypeUint8FeaturesEXT,
+            vk::PhysicalDeviceExtendedDynamicStateFeaturesEXT,
+            vk::PhysicalDeviceRobustness2FeaturesEXT>()};
         decltype(deviceFeatures2) enabledFeatures2{}; // We only want to enable features we required due to potential overhead from unused features
 
         #define FEAT_REQ(structName, feature)                                            \

app/src/main/cpp/skyline/gpu/interconnect/maxwell_3d/active_state.cpp

@@ -31,9 +31,9 @@ namespace skyline::gpu::interconnect::maxwell3d {
 
                 if (megaBufferBinding = view->TryMegaBuffer(ctx.executor.cycle, ctx.gpu.megaBufferAllocator, ctx.executor.executionNumber);
                     megaBufferBinding)
-                    builder.SetVertexBuffer(index, megaBufferBinding);
+                    builder.SetVertexBuffer(index, megaBufferBinding, ctx.gpu.traits.supportsExtendedDynamicState, engine->vertexStream.format.stride);
                 else
-                    builder.SetVertexBuffer(index, *view);
+                    builder.SetVertexBuffer(index, *view, ctx.gpu.traits.supportsExtendedDynamicState, engine->vertexStream.format.stride);
 
                 return;
             } else {
@@ -41,8 +41,10 @@ namespace skyline::gpu::interconnect::maxwell3d {
             }
         }
 
-        // TODO: null descriptor
         megaBufferBinding = {};
+        if (ctx.gpu.traits.supportsNullDescriptor)
+            builder.SetVertexBuffer(index, BufferBinding{});
+        else
             builder.SetVertexBuffer(index, {ctx.gpu.megaBufferAllocator.Allocate(ctx.executor.cycle, 0).buffer});
     }
 

app/src/main/cpp/skyline/gpu/interconnect/maxwell_3d/maxwell_3d.cpp

@@ -247,7 +247,7 @@ namespace skyline::gpu::interconnect::maxwell3d {
 
         if (oldPipeline != pipeline)
             // If the pipeline has changed, we need to update the pipeline state
-            builder.SetPipeline(pipeline->compiledPipeline.pipeline);
+            builder.SetPipeline(pipeline->compiledPipeline.pipeline, vk::PipelineBindPoint::eGraphics);
 
         if (descUpdateInfo) {
             if (ctx.gpu.traits.supportsPushDescriptors) {

app/src/main/cpp/skyline/gpu/interconnect/maxwell_3d/packed_pipeline_state.cpp

@@ -19,7 +19,8 @@ namespace skyline::gpu::interconnect::maxwell3d {
     }
 
     void PackedPipelineState::SetVertexBinding(u32 index, engine::VertexStream stream, engine::VertexStreamInstance instance) {
-        vertexBindings[index].stride = stream.format.stride;
+        if (!dynamicStateActive)
+            vertexStrides[index] = stream.format.stride;
         vertexBindings[index].inputRate = static_cast<u8>(instance.isInstanced ? vk::VertexInputRate::eInstance : vk::VertexInputRate::eVertex);
         vertexBindings[index].enable = stream.format.enable;
         vertexBindings[index].divisor = stream.frequency;

app/src/main/cpp/skyline/gpu/interconnect/maxwell_3d/packed_pipeline_state.h

@@ -58,6 +58,7 @@ namespace skyline::gpu::interconnect::maxwell3d {
             u8 alphaFunc : 3; //!< Use {Set,Get}AlphaFunc
             bool alphaTestEnable : 1;
             bool depthClampEnable : 1; // Use SetDepthClampEnable
+            bool dynamicStateActive : 1;
         };
 
         u32 patchSize;
@@ -69,10 +70,8 @@ namespace skyline::gpu::interconnect::maxwell3d {
         std::array<u32, 8> postVtgShaderAttributeSkipMask;
 
         struct VertexBinding {
-            u16 stride : 12;
             u8 inputRate : 1;
             bool enable : 1;
-            u8 _pad_ : 2;
             u32 divisor;
 
             vk::VertexInputRate GetInputRate() const {
@@ -95,6 +94,8 @@ namespace skyline::gpu::interconnect::maxwell3d {
 
         std::array<AttachmentBlendState, engine::ColorTargetCount> attachmentBlendStates;
 
+        std::array<u16, engine::VertexStreamCount> vertexStrides; //!< Use {Set, Get}VertexBinding
+
         struct TransformFeedbackVarying {
             u16 stride;
             u8 offsetWords;
@@ -149,6 +150,8 @@ namespace skyline::gpu::interconnect::maxwell3d {
             // Only hash transform feedback state if it's enabled
             if (other.transformFeedbackEnable && transformFeedbackEnable)
                 return std::memcmp(this, &other, sizeof(PackedPipelineState)) == 0;
+            else if (dynamicStateActive)
+                return std::memcmp(this, &other, offsetof(PackedPipelineState, vertexStrides)) == 0;
             else
                 return std::memcmp(this, &other, offsetof(PackedPipelineState, transformFeedbackVaryings)) == 0;
         }
@@ -159,6 +162,9 @@ namespace skyline::gpu::interconnect::maxwell3d {
             // Only hash transform feedback state if it's enabled
             if (state.transformFeedbackEnable)
                 return XXH64(&state, sizeof(PackedPipelineState), 0);
+            else if (state.dynamicStateActive)
+                return XXH64(&state, offsetof(PackedPipelineState, vertexStrides), 0);
+
 
             return XXH64(&state, offsetof(PackedPipelineState, transformFeedbackVaryings), 0);
         }

app/src/main/cpp/skyline/gpu/interconnect/maxwell_3d/pipeline_manager.cpp

@@ -442,7 +442,7 @@ namespace skyline::gpu::interconnect::maxwell3d {
             const auto &binding{packedState.vertexBindings[i]};
             bindingDescs.push_back({
                                        .binding = i,
-                                       .stride = binding.stride,
+                                       .stride = packedState.vertexStrides[i],
                                        .inputRate = binding.GetInputRate(),
                                    });
 
@@ -535,7 +535,7 @@ namespace skyline::gpu::interconnect::maxwell3d {
             .pAttachments = attachmentBlendStates.data()
         };
 
-        constexpr std::array<vk::DynamicState, 9> dynamicStates{
+        constexpr std::array<vk::DynamicState, 10> dynamicStates{
             vk::DynamicState::eViewport,
             vk::DynamicState::eScissor,
             vk::DynamicState::eLineWidth,
@@ -544,11 +544,16 @@ namespace skyline::gpu::interconnect::maxwell3d {
             vk::DynamicState::eDepthBounds,
             vk::DynamicState::eStencilCompareMask,
             vk::DynamicState::eStencilWriteMask,
-            vk::DynamicState::eStencilReference
+            vk::DynamicState::eStencilReference,
+            // VK_EXT_dynamic_state starts here
+            vk::DynamicState::eVertexInputBindingStrideEXT
         };
 
+        static constexpr u32 BaseDynamicStateCount{9};
+        static constexpr u32 ExtendedDynamicStateCount{BaseDynamicStateCount + 1};
+
         vk::PipelineDynamicStateCreateInfo dynamicState{
-            .dynamicStateCount = static_cast<u32>(dynamicStates.size()),
+            .dynamicStateCount = ctx.gpu.traits.supportsExtendedDynamicState ? ExtendedDynamicStateCount : BaseDynamicStateCount,
             .pDynamicStates = dynamicStates.data()
         };
 

app/src/main/cpp/skyline/gpu/interconnect/maxwell_3d/pipeline_state.cpp

@@ -488,6 +488,8 @@ namespace skyline::gpu::interconnect::maxwell3d {
           ctSelect{engine.ctSelect} {}
 
     void PipelineState::Flush(InterconnectContext &ctx, Textures &textures, ConstantBufferSet &constantBuffers, StateUpdateBuilder &builder) {
+        packedState.dynamicStateActive = ctx.gpu.traits.supportsExtendedDynamicState;
+
         std::array<ShaderBinary, engine::PipelineCount> shaderBinaries;
         for (size_t i{}; i < engine::PipelineCount; i++) {
             const auto &stage{pipelineStages[i].UpdateGet(ctx)};

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

@@ -6,7 +6,7 @@
 
 namespace skyline::gpu {
     TraitManager::TraitManager(const DeviceFeatures2 &deviceFeatures2, DeviceFeatures2 &enabledFeatures2, const std::vector<vk::ExtensionProperties> &deviceExtensions, std::vector<std::array<char, VK_MAX_EXTENSION_NAME_SIZE>> &enabledExtensions, const DeviceProperties2 &deviceProperties2, const vk::raii::PhysicalDevice &physicalDevice) : quirks(deviceProperties2.get<vk::PhysicalDeviceProperties2>().properties, deviceProperties2.get<vk::PhysicalDeviceDriverProperties>()) {
-        bool hasCustomBorderColorExt{}, hasShaderAtomicInt64Ext{}, hasShaderFloat16Int8Ext{}, hasShaderDemoteToHelperExt{}, hasVertexAttributeDivisorExt{}, hasProvokingVertexExt{}, hasPrimitiveTopologyListRestartExt{}, hasImagelessFramebuffersExt{}, hasTransformFeedbackExt{}, hasUint8IndicesExt{};
+        bool hasCustomBorderColorExt{}, hasShaderAtomicInt64Ext{}, hasShaderFloat16Int8Ext{}, hasShaderDemoteToHelperExt{}, hasVertexAttributeDivisorExt{}, hasProvokingVertexExt{}, hasPrimitiveTopologyListRestartExt{}, hasImagelessFramebuffersExt{}, hasTransformFeedbackExt{}, hasUint8IndicesExt{}, hasExtendedDynamicStateExt{}, hasRobustness2Ext{};
         bool supportsUniformBufferStandardLayout{}; // We require VK_KHR_uniform_buffer_standard_layout but assume it is implicitly supported even when not present
 
         for (auto &extension : deviceExtensions) {
@@ -57,6 +57,8 @@ namespace skyline::gpu {
                 EXT_SET("VK_KHR_uniform_buffer_standard_layout", supportsUniformBufferStandardLayout);
                 EXT_SET("VK_EXT_primitive_topology_list_restart", hasPrimitiveTopologyListRestartExt);
                 EXT_SET("VK_EXT_transform_feedback", hasTransformFeedbackExt);
+                EXT_SET("VK_EXT_extended_dynamic_state", hasExtendedDynamicStateExt);
+                EXT_SET("VK_EXT_robustness2", hasRobustness2Ext);
             }
 
             #undef EXT_SET
@@ -83,6 +85,20 @@ namespace skyline::gpu {
         else
             enabledFeatures2.unlink<vk::PhysicalDeviceIndexTypeUint8FeaturesEXT>();
 
+
+        if (hasExtendedDynamicStateExt)
+            FEAT_SET(vk::PhysicalDeviceExtendedDynamicStateFeaturesEXT, extendedDynamicState, supportsExtendedDynamicState)
+        else
+            enabledFeatures2.unlink<vk::PhysicalDeviceExtendedDynamicStateFeaturesEXT>();
+
+        if (hasRobustness2Ext) {
+            FEAT_SET(vk::PhysicalDeviceRobustness2FeaturesEXT, nullDescriptor, supportsNullDescriptor)
+            FEAT_SET(vk::PhysicalDeviceRobustness2FeaturesEXT, robustBufferAccess2, std::ignore)
+            FEAT_SET(vk::PhysicalDeviceRobustness2FeaturesEXT, robustImageAccess2, std::ignore)
+        } else {
+            enabledFeatures2.unlink<vk::PhysicalDeviceRobustness2FeaturesEXT>();
+        }
+
         if (hasCustomBorderColorExt) {
             bool hasCustomBorderColorFeature{};
             FEAT_SET(vk::PhysicalDeviceCustomBorderColorFeaturesEXT, customBorderColors, hasCustomBorderColorFeature)

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

@@ -46,6 +46,8 @@ namespace skyline::gpu {
         bool supportsSubgroupVote{}; //!< If subgroup votes are supported in shaders with SPV_KHR_subgroup_vote
         bool supportsWideLines{}; //!< If the device supports the 'wideLines' Vulkan feature
         bool supportsDepthClamp{}; //!< If the device supports the 'depthClamp' Vulkan feature
+        bool supportsExtendedDynamicState{}; //!< If the device supports the 'VK_EXT_extended_dynamic_state' Vulkan extension
+        bool supportsNullDescriptor{}; //!< If the device supports the null descriptor feature in the 'VK_EXT_robustness2' Vulkan extension
         u32 subgroupSize{}; //!< Size of a subgroup on the host GPU
 
         std::bitset<7> bcnSupport{}; //!< Bitmask of BCn texture formats supported, it is ordered as BC1, BC2, BC3, BC4, BC5, BC6H and BC7
@@ -98,7 +100,9 @@ namespace skyline::gpu {
             vk::PhysicalDevicePrimitiveTopologyListRestartFeaturesEXT,
             vk::PhysicalDeviceImagelessFramebufferFeatures,
             vk::PhysicalDeviceTransformFeedbackFeaturesEXT,
-            vk::PhysicalDeviceIndexTypeUint8FeaturesEXT>;
+            vk::PhysicalDeviceIndexTypeUint8FeaturesEXT,
+            vk::PhysicalDeviceExtendedDynamicStateFeaturesEXT,
+            vk::PhysicalDeviceRobustness2FeaturesEXT>;
 
         TraitManager(const DeviceFeatures2 &deviceFeatures2, DeviceFeatures2 &enabledFeatures2, const std::vector<vk::ExtensionProperties> &deviceExtensions, std::vector<std::array<char, VK_MAX_EXTENSION_NAME_SIZE>> &enabledExtensions, const DeviceProperties2 &deviceProperties2, const vk::raii::PhysicalDevice& physicalDevice);