optimize index buffers

src/Cafe/HW/Latte/Renderer/Metal/MetalMemoryManager.cpp

@@ -1,5 +1,57 @@
+#include "Cafe/HW/Latte/Renderer/Metal/MetalMemoryManager.h"
 #include "Cafe/HW/Latte/Renderer/Metal/MetalRenderer.h"
 
+const size_t BUFFER_ALLOCATION_SIZE = 8 * 1024 * 1024;
+
+MetalBufferAllocation MetalBufferAllocator::GetBufferAllocation(size_t size)
+{
+    // First, try to find a free range
+    for (uint32 i = 0; i < m_freeBufferRanges.size(); i++)
+    {
+        auto& range = m_freeBufferRanges[i];
+        if (range.size >= size)
+        {
+            MetalBufferAllocation allocation;
+            allocation.bufferIndex = range.bufferIndex;
+            allocation.bufferOffset = range.offset;
+            allocation.data = (uint8*)m_buffers[range.bufferIndex]->contents() + range.offset;
+
+            range.offset += size;
+            range.size -= size;
+
+            if (range.size == 0)
+            {
+                m_freeBufferRanges.erase(m_freeBufferRanges.begin() + i);
+            }
+
+            return allocation;
+        }
+    }
+
+    // If no free range was found, allocate a new buffer
+    MTL::Buffer* buffer = m_mtlr->GetDevice()->newBuffer(std::max(size, BUFFER_ALLOCATION_SIZE), MTL::ResourceStorageModeShared);
+
+    MetalBufferAllocation allocation;
+    allocation.bufferIndex = m_buffers.size();
+    allocation.bufferOffset = 0;
+    allocation.data = buffer->contents();
+
+    m_buffers.push_back(buffer);
+
+    // If the buffer is larger than the requested size, add the remaining space to the free buffer ranges
+    if (size < BUFFER_ALLOCATION_SIZE)
+    {
+        MetalBufferRange range;
+        range.bufferIndex = allocation.bufferIndex;
+        range.offset = size;
+        range.size = BUFFER_ALLOCATION_SIZE - size;
+
+        m_freeBufferRanges.push_back(range);
+    }
+
+    return allocation;
+}
+
 void* MetalMemoryManager::GetTextureUploadBuffer(size_t size)
 {
     if (m_textureUploadBuffer.size() < size)
@@ -10,21 +62,6 @@ void* MetalMemoryManager::GetTextureUploadBuffer(size_t size)
     return m_textureUploadBuffer.data();
 }
 
-// TODO: optimize this
-MetalBufferAllocation MetalMemoryManager::GetBufferAllocation(size_t size)
-{
-    MTL::Buffer* buffer = m_mtlr->GetDevice()->newBuffer(size, MTL::ResourceStorageModeShared);
-
-    MetalBufferAllocation allocation;
-    allocation.bufferIndex = m_buffers.size();
-    allocation.bufferOffset = 0;
-    allocation.data = buffer->contents();
-
-    m_buffers.push_back(buffer);
-
-    return allocation;
-}
-
 void MetalMemoryManager::InitBufferCache(size_t size)
 {
     if (m_bufferCache)

src/Cafe/HW/Latte/Renderer/Metal/MetalMemoryManager.h

@@ -4,6 +4,8 @@
 
 #include "Cafe/HW/Latte/ISA/LatteReg.h"
 
+//const uint32 bufferAllocatorIndexShift = 24;
+
 struct MetalBufferAllocation
 {
     void* data;
@@ -11,16 +13,69 @@ struct MetalBufferAllocation
     size_t bufferOffset;
 };
 
-class MetalMemoryManager
+struct MetalBufferRange
+{
+    uint32 bufferIndex;
+    size_t offset;
+    size_t size;
+};
+
+class MetalBufferAllocator
 {
 public:
-    MetalMemoryManager(class MetalRenderer* metalRenderer) : m_mtlr{metalRenderer} {}
+    MetalBufferAllocator(class MetalRenderer* metalRenderer) : m_mtlr{metalRenderer} {}
+
+    void ResetTemporaryBuffers()
+    {
+        m_freeBufferRanges.clear();
+
+        // Register the free ranges
+        for (uint32 i = 0; i < m_buffers.size(); i++)
+        {
+            m_freeBufferRanges.push_back({i, 0, m_buffers[i]->length()});
+        }
+    }
 
     MTL::Buffer* GetBuffer(uint32 bufferIndex)
     {
         return m_buffers[bufferIndex];
     }
 
+    MetalBufferAllocation GetBufferAllocation(size_t size);
+
+private:
+    class MetalRenderer* m_mtlr;
+
+    std::vector<MTL::Buffer*> m_buffers;
+    std::vector<MetalBufferRange> m_freeBufferRanges;
+};
+
+class MetalMemoryManager
+{
+public:
+    MetalMemoryManager(class MetalRenderer* metalRenderer) : m_mtlr{metalRenderer}, m_bufferAllocator(metalRenderer) {}
+
+    void ResetTemporaryBuffers()
+    {
+        m_bufferAllocator/*s[m_bufferAllocatorIndex]*/.ResetTemporaryBuffers();
+        //m_bufferAllocatorIndex = (m_bufferAllocatorIndex + 1) % 2;
+    }
+
+    MTL::Buffer* GetBuffer(uint32 bufferIndex)
+    {
+        //uint32 bufferAllocatorIndex = (bufferIndex >> bufferAllocatorIndexShift);
+
+        return m_bufferAllocator/*s[bufferAllocatorIndex]*/.GetBuffer(bufferIndex);
+    }
+
+    MetalBufferAllocation GetBufferAllocation(size_t size)
+    {
+        auto allocation = m_bufferAllocator/*s[m_bufferAllocatorIndex]*/.GetBufferAllocation(size);
+        //allocation.bufferIndex |= (m_bufferAllocatorIndex << bufferAllocatorIndexShift);
+
+        return allocation;
+    }
+
     MTL::Buffer* GetBufferCache()
     {
         return m_bufferCache;
@@ -28,8 +83,6 @@ public:
 
     void* GetTextureUploadBuffer(size_t size);
 
-    MetalBufferAllocation GetBufferAllocation(size_t size);
-
     // Buffer cache
     void InitBufferCache(size_t size);
     void UploadToBufferCache(const void* data, size_t offset, size_t size);
@@ -39,6 +92,9 @@ private:
     class MetalRenderer* m_mtlr;
 
     std::vector<uint8> m_textureUploadBuffer;
-    std::vector<MTL::Buffer*> m_buffers;
+
+    MetalBufferAllocator m_bufferAllocator;//s[2];
+    //uint8 m_bufferAllocatorIndex = 0;
+
     MTL::Buffer* m_bufferCache = nullptr;
 };

src/Cafe/HW/Latte/Renderer/Metal/MetalRenderer.cpp

@@ -456,7 +456,7 @@ void MetalRenderer::draw_beginSequence()
 	LatteSHRC_UpdateActiveShaders();
 	if (LatteGPUState.activeShaderHasError)
 	{
-		cemuLog_logDebugOnce(LogType::Force, "Skipping drawcalls due to shader error");
+		debug_printf("Skipping drawcalls due to shader error\n");
 		m_state.skipDrawSequence = true;
 		cemu_assert_debug(false);
 		return;
@@ -506,10 +506,8 @@ void MetalRenderer::draw_beginSequence()
 
 void MetalRenderer::draw_execute(uint32 baseVertex, uint32 baseInstance, uint32 instanceCount, uint32 count, MPTR indexDataMPTR, Latte::LATTE_VGT_DMA_INDEX_TYPE::E_INDEX_TYPE indexType, bool isFirst)
 {
-    // TODO: uncomment
     //if (m_state.skipDrawSequence)
 	//{
-    //  printf("skipping draw\n");
 	//	return;
 	//}
 

src/Cafe/HW/Latte/Renderer/Metal/MetalRenderer.h

@@ -18,7 +18,7 @@
 
 #define MAX_MTL_TEXTURES 31
 
-struct MetalBufferRange
+struct MetalBoundBuffer
 {
     bool needsRebind = false;
     sint32 offset = -1;
@@ -28,7 +28,7 @@ struct MetalState
 {
     bool skipDrawSequence = false;
     class CachedFBOMtl* activeFBO = nullptr;
-    MetalBufferRange vertexBuffers[MAX_MTL_BUFFERS] = {{}};
+    MetalBoundBuffer vertexBuffers[MAX_MTL_BUFFERS] = {{}};
     class LatteTextureViewMtl* textures[MAX_MTL_TEXTURES] = {nullptr};
     MTL::Texture* colorRenderTargets[8] = {nullptr};
     MTL::Texture* depthRenderTarget = nullptr;
@@ -333,6 +333,9 @@ private:
             m_commandBuffer->release();
             m_commandBuffer = nullptr;
 
+            // Reset temporary buffers
+            m_memoryManager->ResetTemporaryBuffers();
+
             // Debug
             m_commandQueue->insertDebugCaptureBoundary();
         }

src/Cafe/HW/Latte/Renderer/Metal/ShaderSourcePresent.h

@@ -1,5 +1,4 @@
-#include <cmath>
-const char* presentLibrarySource = \
+inline const char* presentLibrarySource = \
 "#include <metal_stdlib>\n" \
 "using namespace metal;\n" \
 "\n" \