Optimize render encoder end/start

examples/rt64_render_interface.cpp

@@ -1075,7 +1075,7 @@ namespace RT64 {
             ctx.commandList->setGraphicsPushConstants(0, &pushConstant1);
             drawRasterTriangle(ctx);
 
-            // Draw second triangle with push constants for red color
+            // Draw third triangle with push constants for green color
             offsetViewport.y -= 100.0f;
             ctx.commandList->setViewports(offsetViewport);
             RasterPushConstant pushConstant2;
@@ -1416,7 +1416,7 @@ namespace RT64 {
     using TestSetupFunc = std::function<std::unique_ptr<TestBase>()>;
     static std::vector<TestSetupFunc> g_Tests;
     static std::unique_ptr<TestBase> g_CurrentTest;
-    static uint32_t g_CurrentTestIndex = 2;
+    static uint32_t g_CurrentTestIndex = 1;
 
     void RegisterTests() {
         g_Tests.push_back([]() { return std::make_unique<ClearTest>(); });

src/contrib/metal-cpp/Metal/MTLTypes.hpp

@@ -90,6 +90,14 @@ struct SamplePosition
 
     float                 x;
     float                 y;
+    
+    bool operator==(const SamplePosition& other) const {
+        return x == other.x && y == other.y;
+    }
+    
+    bool operator!=(const SamplePosition& other) const {
+        return !(*this == other);
+    }
 } _MTL_PACKED;
 
 struct ResourceID

src/metal/rt64_metal.cpp

@@ -536,7 +536,6 @@ namespace RT64 {
         state->winding = MTL::WindingClockwise;
         state->sampleCount = desc.multisampling.sampleCount;
         if (desc.multisampling.sampleCount > 1) {
-            state->samplePositions = new MTL::SamplePosition[desc.multisampling.sampleCount];
             for (uint32_t i = 0; i < desc.multisampling.sampleCount; i++) {
                 state->samplePositions[i] = { (float)desc.multisampling.sampleLocations[i].x, (float)desc.multisampling.sampleLocations[i].y };
             }
@@ -562,8 +561,6 @@ namespace RT64 {
     MetalGraphicsPipeline::~MetalGraphicsPipeline() {
         state->renderPipelineState->release();
         state->depthStencilState->release();
-        // TODO: better way to dispose of sample positions
-        delete[] state->samplePositions;
         delete state;
     }
 
@@ -1195,8 +1192,15 @@ namespace RT64 {
             case MetalPipeline::Type::Graphics: {
                 const auto *graphicsPipeline = static_cast<const MetalGraphicsPipeline *>(interfacePipeline);
                 if (activeRenderState) {
-                    if (activeRenderState->samplePositions != graphicsPipeline->state->samplePositions || activeRenderState->sampleCount != graphicsPipeline->state->sampleCount) {
+                    if (activeRenderState->sampleCount != graphicsPipeline->state->sampleCount) {
                         endActiveRenderEncoder();
+                    } else if (activeRenderState->sampleCount > 1) {
+                        for (uint32_t i = 0; i < activeRenderState->sampleCount; i++) {
+                            if (activeRenderState->samplePositions[i] != graphicsPipeline->state->samplePositions[i]) {
+                                endActiveRenderEncoder();
+                                break;
+                            }
+                        }
                     }
                 }
 
@@ -1400,17 +1404,23 @@ namespace RT64 {
     }
 
     void MetalCommandList::setFramebuffer(const RenderFramebuffer *framebuffer) {
-        // If we're changing framebuffers, we need to end current encoder and reset state
-        if (targetFramebuffer != framebuffer) {
-            endOtherEncoders(EncoderType::Render);
-            endActiveRenderEncoder();
-
-            targetFramebuffer = static_cast<const MetalFramebuffer*>(framebuffer);
-
-            // Mark all state as needing update with the new encoder
-            if (targetFramebuffer != nullptr) {
-                dirtyGraphicsState.setAll();
+        endOtherEncoders(EncoderType::Render);
+        
+        if (framebuffer != nullptr) {
+            auto incomingFramebuffer = static_cast<const MetalFramebuffer*>(framebuffer);
+            
+            // check if we need to end the current encoder
+            if (targetFramebuffer == nullptr || *targetFramebuffer != *incomingFramebuffer) {
+                endActiveRenderEncoder();
+                targetFramebuffer = incomingFramebuffer;
+                
+                // Mark all state as needing update with the new encoder
+                if (targetFramebuffer != nullptr) {
+                    dirtyGraphicsState.setAll();
+                }
             }
+        } else {
+            targetFramebuffer = nullptr;
         }
     }
 

src/metal/rt64_metal.h

@@ -18,7 +18,6 @@ namespace RT64 {
     struct MetalGraphicsPipeline;
     struct MetalPool;
     struct MetalDrawable;
-    struct ExtendedRenderTexture;
 
     enum class EncoderType {
         Render,
@@ -92,7 +91,7 @@ namespace RT64 {
         MTL::CullMode cullMode = MTL::CullModeNone;
         MTL::DepthClipMode depthClipMode = MTL::DepthClipModeClip;
         MTL::Winding winding = MTL::WindingClockwise;
-        MTL::SamplePosition *samplePositions = nullptr;
+        MTL::SamplePosition samplePositions[16] = {};
         uint32_t sampleCount = 0;
     };
 
@@ -104,6 +103,11 @@ namespace RT64 {
             : buffer(buffer), offset(offset) {}
     };
 
+    struct ExtendedRenderTexture: RenderTexture {
+        RenderTextureDesc desc;
+        virtual MTL::Texture* getTexture() const = 0;
+    };
+
     struct MetalDescriptorSet : RenderDescriptorSet {
         std::unordered_map<uint32_t, MTL::Texture *> indicesToTextures;
         std::unordered_map<uint32_t, MetalBufferBinding> indicesToBuffers;
@@ -164,6 +168,38 @@ namespace RT64 {
         ~MetalFramebuffer() override;
         uint32_t getWidth() const override;
         uint32_t getHeight() const override;
+        
+        // this comparison is tailored towards whether we'll require a new encoder
+        bool operator==(const MetalFramebuffer& other) const {
+            if (colorAttachments.size() != other.colorAttachments.size()) {
+                return false;
+            }
+            
+            for (size_t i = 0; i < colorAttachments.size(); i++) {
+                if (colorAttachments[i]->getTexture() != other.colorAttachments[i]->getTexture() ||
+                    colorAttachments[i]->desc.multisampling.sampleCount != other.colorAttachments[i]->desc.multisampling.sampleCount) {
+                    return false;
+                }
+                
+                // Compare individual sample locations if multisampling is enabled
+                if (colorAttachments[i]->desc.multisampling.sampleCount > 1) {
+                    for (uint32_t s = 0; s < colorAttachments[i]->desc.multisampling.sampleCount; s++) {
+                        const auto& loc1 = colorAttachments[i]->desc.multisampling.sampleLocations[s];
+                        const auto& loc2 = other.colorAttachments[i]->desc.multisampling.sampleLocations[s];
+                        if (loc1 != loc2) {
+                            return false;
+                        }
+                    }
+                }
+            }
+            
+            return depthAttachment == other.depthAttachment;
+        }
+        
+        bool operator!=(const MetalFramebuffer& other) const {
+            return !(*this == other);
+        }
+        
     };
 
     struct MetalCommandList : RenderCommandList {
@@ -358,11 +394,6 @@ namespace RT64 {
         ~MetalBufferFormattedView() override;
     };
 
-    struct ExtendedRenderTexture: RenderTexture {
-        RenderTextureDesc desc;
-        virtual MTL::Texture* getTexture() const = 0;
-    };
-
     struct MetalDrawable : ExtendedRenderTexture {
         CA::MetalDrawable *mtl = nullptr;
         

src/rhi/rt64_render_interface_types.h

@@ -696,6 +696,14 @@ namespace RT64 {
         // Valid range is [-8, 7].
         int8_t x = 0;
         int8_t y = 0;
+        
+        bool operator==(const RenderMultisamplingLocation& other) const {
+            return x == other.x && y == other.y;
+        }
+        
+        bool operator!=(const RenderMultisamplingLocation& other) const {
+            return !(*this == other);
+        }
     };
 
     struct RenderMultisampling {