Some adjustments to vertex attributes

src/video_core/engines/draw_manager.cpp

@@ -14,6 +14,10 @@
 namespace Tegra::Engines {
 
 void Maxwell3D::DrawManager::ProcessMethodCall(Maxwell3D& maxwell3d, u32 method, u32 argument) {
+    if (draw_state.draw_mode == DrawMode::InstanceArray &&
+        method != MAXWELL3D_REG_INDEX(vertex_array_instance_subsequent)) {
+        DrawDeferred(maxwell3d);
+    }
     switch (method) {
     case MAXWELL3D_REG_INDEX(clear_surface):
         return Clear(maxwell3d, 1);
@@ -73,6 +77,15 @@ void Maxwell3D::DrawManager::Clear(Maxwell3D& maxwell3d, u32 layer_count) {
 }
 
 void Maxwell3D::DrawManager::DrawDeferred(Maxwell3D& maxwell3d) {
+    if (draw_state.draw_mode == DrawMode::InstanceArray) {
+        const u32 instance_count = draw_state.instance_count + 1;
+        draw_state.draw_mode = DrawMode::General;
+        draw_state.instance_count = 0;
+        if (maxwell3d.ShouldExecute()) {
+            maxwell3d.rasterizer->Draw(false, instance_count);
+        }
+        return;
+    }
     if (draw_state.draw_mode != DrawMode::Instance || draw_state.instance_count == 0) {
         return;
     }
@@ -89,16 +102,25 @@ void Maxwell3D::DrawManager::DrawArray(Maxwell3D& maxwell3d, Maxwell3D::Regs::Pr
 }
 
 void Maxwell3D::DrawManager::DrawArrayInstanced(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 vertex_first, u32 vertex_count, bool subsequent) {
+    if (subsequent && draw_state.draw_mode == DrawMode::InstanceArray &&
+        instance_topology == topology && draw_state.vertex_buffer.first == vertex_first &&
+        draw_state.vertex_buffer.count == vertex_count) {
+        ++draw_state.instance_count;
+        return;
+    }
+    u32 base_instance = 0;
+    if (subsequent) {
+        base_instance = draw_state.base_instance + draw_state.instance_count + 1;
+    }
+    DrawDeferred(maxwell3d);
+    instance_topology = topology;
     draw_state.topology = topology;
     draw_state.vertex_buffer.first = vertex_first;
     draw_state.vertex_buffer.count = vertex_count;
-    if (!subsequent) {
-        draw_state.instance_count = 1;
-    }
-    draw_state.base_instance = draw_state.instance_count - 1;
-    draw_state.draw_mode = DrawMode::Instance;
-    draw_state.instance_count++;
-    ProcessDraw(maxwell3d, false, 1);
+    draw_state.base_instance = base_instance;
+    draw_state.instance_count = 0;
+    draw_state.draw_mode = DrawMode::InstanceArray;
+    UpdateTopology(maxwell3d);
 }
 
 void Maxwell3D::DrawManager::DrawIndex(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 index_first, u32 index_count, u32 base_index, u32 base_instance, u32 num_instances) {

src/video_core/engines/maxwell_3d.h

@@ -3034,7 +3034,7 @@ public:
     // clang-format on
 
     struct DrawManager {
-        enum class DrawMode : u32 { General = 0, Instance, InlineIndex };
+        enum class DrawMode : u32 { General = 0, Instance, InlineIndex, InstanceArray };
         struct State {
             Maxwell3D::Regs::PrimitiveTopology topology{};
             DrawMode draw_mode{};
@@ -3086,6 +3086,7 @@ public:
         void ProcessDraw(Maxwell3D& maxwell3d, bool draw_indexed, u32 instance_count);
         void ProcessDrawIndirect(Maxwell3D& maxwell3d);
         State draw_state{};
+        Maxwell3D::Regs::PrimitiveTopology instance_topology{};
         DrawTextureState draw_texture_state{};
         IndirectParams indirect_state{};
     };

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -705,7 +705,8 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
             if (instanced) {
                 vertex_binding_divisors.push_back({
                     .binding = static_cast<u32>(index),
-                    .divisor = key.state.binding_divisors[index],
+                    .divisor = (std::min)(key.state.binding_divisors[index],
+                                          device.GetMaxVertexAttribDivisor()),
                 });
             }
         }

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -1918,13 +1918,17 @@ void RasterizerVulkan::UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs)
     for (u32 binding = 0; binding < max_bindings; ++binding) {
         const auto& input_binding{regs.vertex_streams[binding]};
         const bool is_instanced{regs.vertex_stream_instances.IsInstancingEnabled(binding)};
+        u32 divisor = 1;
+        if (is_instanced) {
+            divisor = (std::min)(input_binding.frequency, device.GetMaxVertexAttribDivisor());
+        }
         bindings.push_back({
             .sType = VK_STRUCTURE_TYPE_VERTEX_INPUT_BINDING_DESCRIPTION_2_EXT,
             .pNext = nullptr,
             .binding = binding,
             .stride = input_binding.stride,
             .inputRate = is_instanced ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX,
-            .divisor = is_instanced ? input_binding.frequency : 1,
+            .divisor = divisor,
         });
     }
 

src/video_core/vulkan_common/vulkan_device.cpp

@@ -1130,11 +1130,19 @@ bool Device::GetSuitability(bool requires_swapchain) {
         SetNext(next, properties.custom_border_color);
     }
 
+    if (extensions.vertex_attribute_divisor) {
+        properties.vertex_attribute_divisor.sType =
+            VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_PROPERTIES_EXT;
+        SetNext(next, properties.vertex_attribute_divisor);
+    }
+
     // Perform the property fetch.
     physical.GetProperties2(properties2);
 
     // Store base properties
     properties.properties = properties2.properties;
+    max_vertex_attrib_divisor =
+        (std::max)(1U, properties.vertex_attribute_divisor.maxVertexAttribDivisor);
 
     // Unload extensions if feature support is insufficient.
     RemoveUnsuitableExtensions();

src/video_core/vulkan_common/vulkan_device.h

@@ -984,6 +984,10 @@ FN_MAX_LIMIT_LIST
         return properties.properties.limits.maxVertexInputBindings;
     }
 
+    u32 GetMaxVertexAttribDivisor() const {
+        return max_vertex_attrib_divisor;
+    }
+
     u32 GetMaxViewports() const {
         return properties.properties.limits.maxViewports;
     }
@@ -1188,6 +1192,7 @@ private:
         VkPhysicalDeviceMaintenance5PropertiesKHR maintenance5{};
         VkPhysicalDeviceDepthStencilResolveProperties depth_stencil_resolve{};
         VkPhysicalDeviceCustomBorderColorPropertiesEXT custom_border_color{};
+        VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT vertex_attribute_divisor{};
 
         VkPhysicalDeviceProperties properties{};
     };
@@ -1195,6 +1200,7 @@ private:
     Extensions extensions{};
     Features features{};
     Properties properties{};
+    u32 max_vertex_attrib_divisor{1};
 
     VkPhysicalDeviceFeatures2 features2{};
     VkPhysicalDeviceProperties2 properties2{};