fix: buffer allocator

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

@@ -3,15 +3,16 @@
 
 const size_t BUFFER_ALLOCATION_SIZE = 8 * 1024 * 1024;
 
-// TODO: uncomment everything
+MetalBufferAllocation MetalBufferAllocator::GetBufferAllocation(size_t size, size_t alignment)
-MetalBufferAllocation MetalBufferAllocator::GetBufferAllocation(size_t size)
 {
+    // Align the size
+    size = (size + alignment - 1) & ~(alignment - 1);
+
     // 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)
+        if (size <= range.size)
         {
             MetalBufferAllocation allocation;
             allocation.bufferIndex = range.bufferIndex;
@@ -29,10 +30,9 @@ MetalBufferAllocation MetalBufferAllocator::GetBufferAllocation(size_t size)
             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);
+    MTL::Buffer* buffer = m_mtlr->GetDevice()->newBuffer(std::max(size, BUFFER_ALLOCATION_SIZE), MTL::ResourceStorageModeShared);
 
     MetalBufferAllocation allocation;
     allocation.bufferIndex = m_buffers.size();
@@ -42,7 +42,6 @@ MetalBufferAllocation MetalBufferAllocator::GetBufferAllocation(size_t size)
     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;
@@ -52,7 +51,6 @@ MetalBufferAllocation MetalBufferAllocator::GetBufferAllocation(size_t size)
 
         m_freeBufferRanges.push_back(range);
     }
-    */
 
     return allocation;
 }

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

@@ -41,7 +41,7 @@ public:
         return m_buffers[bufferIndex];
     }
 
-    MetalBufferAllocation GetBufferAllocation(size_t size);
+    MetalBufferAllocation GetBufferAllocation(size_t size, size_t alignment);
 
 private:
     class MetalRenderer* m_mtlr;
@@ -68,9 +68,9 @@ public:
         return m_bufferAllocator/*s[bufferAllocatorIndex]*/.GetBuffer(bufferIndex);
     }
 
-    MetalBufferAllocation GetBufferAllocation(size_t size)
+    MetalBufferAllocation GetBufferAllocation(size_t size, size_t alignment)
     {
-        auto allocation = m_bufferAllocator/*s[m_bufferAllocatorIndex]*/.GetBufferAllocation(size);
+        auto allocation = m_bufferAllocator/*s[m_bufferAllocatorIndex]*/.GetBufferAllocation(size, alignment);
         //allocation.bufferIndex |= (m_bufferAllocatorIndex << bufferAllocatorIndexShift);
 
         return allocation;

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

@@ -129,6 +129,7 @@ MTL::RenderPipelineState* MetalPipelineCache::GetPipelineState(const LatteFetchS
 	{
 	    auto texture = static_cast<LatteTextureViewMtl*>(activeFBO->depthBuffer.texture);
         desc->setDepthAttachmentPixelFormat(texture->GetTexture()->pixelFormat());
+        // TODO: stencil pixel format
 	}
 
 	NS::Error* error = nullptr;

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

@@ -16,6 +16,7 @@
 #include "Cemu/Logging/CemuDebugLogging.h"
 #include "HW/Latte/Core/Latte.h"
 #include "gui/guiWrapper.h"
+#include <stdexcept>
 
 extern bool hasValidFramebufferAttached;
 
@@ -143,6 +144,9 @@ void MetalRenderer::SwapBuffers(bool swapTV, bool swapDRC)
     m_drawableAcquired = false;
 
     CommitCommandBuffer();
+
+    // Reset temporary buffers
+    m_memoryManager->ResetTemporaryBuffers();
 }
 
 void MetalRenderer::DrawBackbufferQuad(LatteTextureView* texView, RendererOutputShader* shader, bool useLinearTexFilter,
@@ -661,7 +665,7 @@ void MetalRenderer::draw_execute(uint32 baseVertex, uint32 baseInstance, uint32
 	{
 	    auto mtlIndexType = GetMtlIndexType(hostIndexType);
 		MTL::Buffer* indexBuffer = m_memoryManager->GetBuffer(indexBufferIndex);
-		renderCommandEncoder->drawIndexedPrimitives(mtlPrimitiveType, hostIndexCount, mtlIndexType, indexBuffer, 0, instanceCount, baseVertex, baseInstance);
+		renderCommandEncoder->drawIndexedPrimitives(mtlPrimitiveType, hostIndexCount, mtlIndexType, indexBuffer, indexBufferOffset, instanceCount, baseVertex, baseInstance);
 	} else
 	{
 		renderCommandEncoder->drawPrimitives(mtlPrimitiveType, baseVertex, count, instanceCount, baseInstance);
@@ -675,7 +679,7 @@ void MetalRenderer::draw_endSequence()
 
 void* MetalRenderer::indexData_reserveIndexMemory(uint32 size, uint32& offset, uint32& bufferIndex)
 {
-    auto allocation = m_memoryManager->GetBufferAllocation(size);
+    auto allocation = m_memoryManager->GetBufferAllocation(size, 4);
 	offset = allocation.bufferOffset;
 	bufferIndex = allocation.bufferIndex;
 
@@ -684,7 +688,7 @@ void* MetalRenderer::indexData_reserveIndexMemory(uint32 size, uint32& offset, u
 
 void MetalRenderer::indexData_uploadIndexMemory(uint32 offset, uint32 size)
 {
-    debug_printf("MetalRenderer::indexData_uploadIndexMemory not implemented\n");
+    // Do nothing, since the buffer has shared storage mode
 }
 
 void MetalRenderer::EnsureCommandBuffer()
@@ -816,9 +820,6 @@ void MetalRenderer::CommitCommandBuffer()
         m_commandBuffer->release();
         m_commandBuffer = nullptr;
 
-        // Reset temporary buffers
-        m_memoryManager->ResetTemporaryBuffers();
-
         // Debug
         m_commandQueue->insertDebugCaptureBoundary();
     }