shader_recompiler: Replace buffer pulling with attribute divisor for instance step rates (#3238)

* shader_recompiler: Replace buffer pulling with attribute divisor for instance step rates

* flatten_extended_userdata: Remove special step rate buffer handling

* Review comments

* spirv_emit_context: Name all instance rate attribs properly

* spirv: Merge ReadConstBuffer again

template function only has 1 user now

* attribute: Add missing attributes

* translate: Reimplement step rate instance id

* Resolve validation warnings

* shader_recompiler: Separate vertex inputs from LS stage, cleanup tess

src/video_core/buffer_cache/buffer_cache.cpp

@@ -198,10 +198,13 @@ void BufferCache::DownloadBufferMemory(Buffer& buffer, VAddr device_addr, u64 si
 }
 
 void BufferCache::BindVertexBuffers(const Vulkan::GraphicsPipeline& pipeline) {
+    const auto& regs = liverpool->regs;
     Vulkan::VertexInputs<vk::VertexInputAttributeDescription2EXT> attributes;
     Vulkan::VertexInputs<vk::VertexInputBindingDescription2EXT> bindings;
+    Vulkan::VertexInputs<vk::VertexInputBindingDivisorDescriptionEXT> divisors;
     Vulkan::VertexInputs<AmdGpu::Buffer> guest_buffers;
-    pipeline.GetVertexInputs(attributes, bindings, guest_buffers);
+    pipeline.GetVertexInputs(attributes, bindings, divisors, guest_buffers,
+                             regs.vgt_instance_step_rate_0, regs.vgt_instance_step_rate_1);
 
     if (instance.IsVertexInputDynamicState()) {
         // Update current vertex inputs.

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -72,12 +72,21 @@ GraphicsPipeline::GraphicsPipeline(
 
     VertexInputs<vk::VertexInputAttributeDescription> vertex_attributes;
     VertexInputs<vk::VertexInputBindingDescription> vertex_bindings;
+    VertexInputs<vk::VertexInputBindingDivisorDescriptionEXT> divisors;
     VertexInputs<AmdGpu::Buffer> guest_buffers;
     if (!instance.IsVertexInputDynamicState()) {
-        GetVertexInputs(vertex_attributes, vertex_bindings, guest_buffers);
+        const auto& vs_info = runtime_infos[u32(Shader::LogicalStage::Vertex)].vs_info;
+        GetVertexInputs(vertex_attributes, vertex_bindings, divisors, guest_buffers,
+                        vs_info.step_rate_0, vs_info.step_rate_1);
     }
 
+    const vk::PipelineVertexInputDivisorStateCreateInfo divisor_state = {
+        .vertexBindingDivisorCount = static_cast<u32>(divisors.size()),
+        .pVertexBindingDivisors = divisors.data(),
+    };
+
     const vk::PipelineVertexInputStateCreateInfo vertex_input_info = {
+        .pNext = divisors.empty() ? nullptr : &divisor_state,
         .vertexBindingDescriptionCount = static_cast<u32>(vertex_bindings.size()),
         .pVertexBindingDescriptions = vertex_bindings.data(),
         .vertexAttributeDescriptionCount = static_cast<u32>(vertex_attributes.size()),
@@ -304,19 +313,17 @@ GraphicsPipeline::GraphicsPipeline(
 GraphicsPipeline::~GraphicsPipeline() = default;
 
 template <typename Attribute, typename Binding>
-void GraphicsPipeline::GetVertexInputs(VertexInputs<Attribute>& attributes,
-                                       VertexInputs<Binding>& bindings,
-                                       VertexInputs<AmdGpu::Buffer>& guest_buffers) const {
+void GraphicsPipeline::GetVertexInputs(
+    VertexInputs<Attribute>& attributes, VertexInputs<Binding>& bindings,
+    VertexInputs<vk::VertexInputBindingDivisorDescriptionEXT>& divisors,
+    VertexInputs<AmdGpu::Buffer>& guest_buffers, u32 step_rate_0, u32 step_rate_1) const {
+    using InstanceIdType = Shader::Gcn::VertexAttribute::InstanceIdType;
     if (!fetch_shader || fetch_shader->attributes.empty()) {
         return;
     }
     const auto& vs_info = GetStage(Shader::LogicalStage::Vertex);
     for (const auto& attrib : fetch_shader->attributes) {
-        if (attrib.UsesStepRates()) {
-            // Skip attribute binding as the data will be pulled by shader.
-            continue;
-        }
-
+        const auto step_rate = attrib.GetStepRate();
         const auto& buffer = attrib.GetSharp(vs_info);
         attributes.push_back(Attribute{
             .location = attrib.semantic,
@@ -327,12 +334,19 @@ void GraphicsPipeline::GetVertexInputs(VertexInputs<Attribute>& attributes,
         bindings.push_back(Binding{
             .binding = attrib.semantic,
             .stride = buffer.GetStride(),
-            .inputRate = attrib.GetStepRate() == Shader::Gcn::VertexAttribute::InstanceIdType::None
-                             ? vk::VertexInputRate::eVertex
-                             : vk::VertexInputRate::eInstance,
+            .inputRate = step_rate == InstanceIdType::None ? vk::VertexInputRate::eVertex
+                                                           : vk::VertexInputRate::eInstance,
         });
+        const u32 divisor = step_rate == InstanceIdType::OverStepRate0
+                                ? step_rate_0
+                                : (step_rate == InstanceIdType::OverStepRate1 ? step_rate_1 : 1);
         if constexpr (std::is_same_v<Binding, vk::VertexInputBindingDescription2EXT>) {
-            bindings.back().divisor = 1;
+            bindings.back().divisor = divisor;
+        } else if (step_rate != InstanceIdType::None) {
+            divisors.push_back(vk::VertexInputBindingDivisorDescriptionEXT{
+                .binding = attrib.semantic,
+                .divisor = divisor,
+            });
         }
         guest_buffers.emplace_back(buffer);
     }
@@ -342,11 +356,13 @@ void GraphicsPipeline::GetVertexInputs(VertexInputs<Attribute>& attributes,
 template void GraphicsPipeline::GetVertexInputs(
     VertexInputs<vk::VertexInputAttributeDescription>& attributes,
     VertexInputs<vk::VertexInputBindingDescription>& bindings,
-    VertexInputs<AmdGpu::Buffer>& guest_buffers) const;
+    VertexInputs<vk::VertexInputBindingDivisorDescriptionEXT>& divisors,
+    VertexInputs<AmdGpu::Buffer>& guest_buffers, u32 step_rate_0, u32 step_rate_1) const;
 template void GraphicsPipeline::GetVertexInputs(
     VertexInputs<vk::VertexInputAttributeDescription2EXT>& attributes,
     VertexInputs<vk::VertexInputBindingDescription2EXT>& bindings,
-    VertexInputs<AmdGpu::Buffer>& guest_buffers) const;
+    VertexInputs<vk::VertexInputBindingDivisorDescriptionEXT>& divisors,
+    VertexInputs<AmdGpu::Buffer>& guest_buffers, u32 step_rate_0, u32 step_rate_1) const;
 
 void GraphicsPipeline::BuildDescSetLayout() {
     boost::container::small_vector<vk::DescriptorSetLayoutBinding, 32> bindings;

src/video_core/renderer_vulkan/vk_graphics_pipeline.h

@@ -81,7 +81,9 @@ public:
     /// Gets the attributes and bindings for vertex inputs.
     template <typename Attribute, typename Binding>
     void GetVertexInputs(VertexInputs<Attribute>& attributes, VertexInputs<Binding>& bindings,
-                         VertexInputs<AmdGpu::Buffer>& guest_buffers) const;
+                         VertexInputs<vk::VertexInputBindingDivisorDescriptionEXT>& divisors,
+                         VertexInputs<AmdGpu::Buffer>& guest_buffers, u32 step_rate_0,
+                         u32 step_rate_1) const;
 
 private:
     void BuildDescSetLayout();

src/video_core/renderer_vulkan/vk_instance.cpp

@@ -248,6 +248,7 @@ bool Instance::CreateDevice() {
     // Required
     ASSERT(add_extension(VK_KHR_SWAPCHAIN_EXTENSION_NAME));
     ASSERT(add_extension(VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME));
+    ASSERT(add_extension(VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME));
 
     // Optional
     depth_range_unrestricted = add_extension(VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME);
@@ -436,6 +437,9 @@ bool Instance::CreateDevice() {
         vk::PhysicalDeviceLegacyVertexAttributesFeaturesEXT{
             .legacyVertexAttributes = true,
         },
+        vk::PhysicalDeviceVertexAttributeDivisorFeatures{
+            .vertexAttributeInstanceRateDivisor = true,
+        },
         vk::PhysicalDeviceShaderAtomicFloat2FeaturesEXT{
             .shaderBufferFloat32AtomicMinMax =
                 shader_atomic_float2_features.shaderBufferFloat32AtomicMinMax,

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -94,15 +94,10 @@ const Shader::RuntimeInfo& PipelineCache::BuildRuntimeInfo(Stage stage, LogicalS
     switch (stage) {
     case Stage::Local: {
         BuildCommon(regs.ls_program);
-        if (regs.stage_enable.IsStageEnabled(static_cast<u32>(Stage::Hull))) {
-            info.ls_info.links_with_tcs = true;
-            Shader::TessellationDataConstantBuffer tess_constants;
-            const auto* pgm = regs.ProgramForStage(static_cast<u32>(Stage::Hull));
-            const auto params = Liverpool::GetParams(*pgm);
-            const auto& hull_info = program_cache.at(params.hash)->info;
-            hull_info.ReadTessConstantBuffer(tess_constants);
-            info.ls_info.ls_stride = tess_constants.ls_stride;
-        }
+        Shader::TessellationDataConstantBuffer tess_constants;
+        const auto* hull_info = infos[u32(Shader::LogicalStage::TessellationControl)];
+        hull_info->ReadTessConstantBuffer(tess_constants);
+        info.ls_info.ls_stride = tess_constants.ls_stride;
         break;
     }
     case Stage::Hull: {
@@ -122,6 +117,8 @@ const Shader::RuntimeInfo& PipelineCache::BuildRuntimeInfo(Stage stage, LogicalS
     case Stage::Vertex: {
         BuildCommon(regs.vs_program);
         GatherVertexOutputs(info.vs_info, regs.vs_output_control);
+        info.vs_info.step_rate_0 = regs.vgt_instance_step_rate_0;
+        info.vs_info.step_rate_1 = regs.vgt_instance_step_rate_1;
         info.vs_info.emulate_depth_negative_one_to_one =
             !instance.IsDepthClipControlSupported() &&
             regs.clipper_control.clip_space == Liverpool::ClipSpace::MinusWToW;
@@ -460,10 +457,6 @@ bool PipelineCache::RefreshGraphicsKey() {
         // Stride will still be handled outside the pipeline using dynamic state.
         u32 vertex_binding = 0;
         for (const auto& attrib : fetch_shader->attributes) {
-            if (attrib.UsesStepRates()) {
-                // Skip attribute binding as the data will be pulled by shader.
-                continue;
-            }
             const auto& buffer = attrib.GetSharp(*vs_info);
             ASSERT(vertex_binding < MaxVertexBufferCount);
             key.vertex_buffer_formats[vertex_binding++] =

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -20,12 +20,9 @@
 namespace Vulkan {
 
 static Shader::PushData MakeUserData(const AmdGpu::Liverpool::Regs& regs) {
-    Shader::PushData push_data{};
-    push_data.step0 = regs.vgt_instance_step_rate_0;
-    push_data.step1 = regs.vgt_instance_step_rate_1;
-
     // TODO(roamic): Add support for multiple viewports and geometry shaders when ViewportIndex
     // is encountered and implemented in the recompiler.
+    Shader::PushData push_data{};
     push_data.xoffset = regs.viewport_control.xoffset_enable ? regs.viewports[0].xoffset : 0.f;
     push_data.xscale = regs.viewport_control.xscale_enable ? regs.viewports[0].xscale : 1.f;
     push_data.yoffset = regs.viewport_control.yoffset_enable ? regs.viewports[0].yoffset : 0.f;