renderer: added support for instance step rates

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -30,12 +30,19 @@ GraphicsPipeline::GraphicsPipeline(const Instance& instance_, Scheduler& schedul
         stages[i] = *infos[i];
     }
     BuildDescSetLayout();
+
+    const vk::PushConstantRange push_constants = {
+        .stageFlags = vk::ShaderStageFlagBits::eVertex,
+        .offset = 0,
+        .size = 2 * sizeof(u32),
+    };
+
     const vk::DescriptorSetLayout set_layout = *desc_layout;
     const vk::PipelineLayoutCreateInfo layout_info = {
         .setLayoutCount = 1U,
         .pSetLayouts = &set_layout,
-        .pushConstantRangeCount = 0,
-        .pPushConstantRanges = nullptr,
+        .pushConstantRangeCount = 1,
+        .pPushConstantRanges = &push_constants,
     };
     pipeline_layout = instance.GetDevice().createPipelineLayoutUnique(layout_info);
 
@@ -43,6 +50,12 @@ GraphicsPipeline::GraphicsPipeline(const Instance& instance_, Scheduler& schedul
     boost::container::static_vector<vk::VertexInputAttributeDescription, 32> attributes;
     const auto& vs_info = stages[0];
     for (const auto& input : vs_info.vs_inputs) {
+        if (input.instance_step_rate == Shader::Info::VsInput::InstanceIdType::OverStepRate0 ||
+            input.instance_step_rate == Shader::Info::VsInput::InstanceIdType::OverStepRate1) {
+            // Skip attribute binding as the data will be pulled by shader
+            continue;
+        }
+
         const auto buffer = vs_info.ReadUd<AmdGpu::Buffer>(input.sgpr_base, input.dword_offset);
         attributes.push_back({
             .location = input.binding,
@@ -420,6 +433,11 @@ void GraphicsPipeline::BindVertexBuffers(StreamBuffer& staging) const {
     // Calculate buffers memory overlaps
     boost::container::static_vector<BufferRange, MaxVertexBufferCount> ranges{};
     for (const auto& input : vs_info.vs_inputs) {
+        if (input.instance_step_rate == Shader::Info::VsInput::InstanceIdType::OverStepRate0 ||
+            input.instance_step_rate == Shader::Info::VsInput::InstanceIdType::OverStepRate1) {
+            continue;
+        }
+
         const auto& buffer = vs_info.ReadUd<AmdGpu::Buffer>(input.sgpr_base, input.dword_offset);
         if (buffer.GetSize() == 0) {
             continue;

src/video_core/renderer_vulkan/vk_graphics_pipeline.h

@@ -67,20 +67,24 @@ public:
     void BindResources(Core::MemoryManager* memory, StreamBuffer& staging,
                        VideoCore::TextureCache& texture_cache) const;
 
-    [[nodiscard]] vk::Pipeline Handle() const noexcept {
+    vk::Pipeline Handle() const noexcept {
         return *pipeline;
     }
 
-    [[nodiscard]] bool IsEmbeddedVs() const noexcept {
+    vk::PipelineLayout GetLayout() const {
+        return *pipeline_layout;
+    }
+
+    bool IsEmbeddedVs() const noexcept {
         static constexpr size_t EmbeddedVsHash = 0x9b2da5cf47f8c29f;
         return key.stage_hashes[0] == EmbeddedVsHash;
     }
 
-    [[nodiscard]] auto GetWriteMasks() const {
+    auto GetWriteMasks() const {
         return key.write_masks;
     }
 
-    [[nodiscard]] bool IsDepthEnabled() const {
+    bool IsDepthEnabled() const {
         return key.depth.depth_enable.Value();
     }
 

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -54,6 +54,13 @@ void Rasterizer::Draw(bool is_indexed, u32 index_offset) {
     UpdateDynamicState(*pipeline);
 
     cmdbuf.bindPipeline(vk::PipelineBindPoint::eGraphics, pipeline->Handle());
+
+    const u32 step_rates[] = {
+        regs.vgt_instance_step_rate_0,
+        regs.vgt_instance_step_rate_1,
+    };
+    cmdbuf.pushConstants(pipeline->GetLayout(), vk::ShaderStageFlagBits::eVertex, 0u,
+                         sizeof(step_rates), &step_rates);
     if (is_indexed) {
         cmdbuf.drawIndexed(num_indices, regs.num_instances.NumInstances(), 0, 0, 0);
     } else {