metal: fix uniform data corruption

The wrong buffer binding offset was being used in 2/3rds of frames.

src/modules/graphics/StreamBuffer.h

@@ -57,6 +57,8 @@ public:
 	BufferUsage getMode() const { return mode; }
 	size_t getUsableSize() const { return bufferSize - frameGPUReadOffset; }
 
+	virtual size_t getGPUReadOffset() const = 0;
+
 	virtual MapInfo map(size_t minsize) = 0;
 	virtual size_t unmap(size_t usedsize) = 0;
 	virtual void markUsed(size_t usedsize) = 0;

src/modules/graphics/metal/Graphics.h

@@ -242,6 +242,7 @@ private:
 	StreamBuffer *uniformBuffer;
 	StreamBuffer::MapInfo uniformBufferData;
 	size_t uniformBufferOffset;
+	size_t uniformBufferGPUStart;
 
 	Buffer *defaultAttributesBuffer;
 

src/modules/graphics/metal/Graphics.mm

@@ -280,6 +280,7 @@ Graphics::Graphics()
 	, attachmentStoreActions()
 	, renderBindings()
 	, uniformBufferOffset(0)
+	, uniformBufferGPUStart(0)
 	, defaultAttributesBuffer(nullptr)
 	, families()
 { @autoreleasepool {
@@ -1035,14 +1036,19 @@ bool Graphics::applyShaderUniforms(id<MTLComputeCommandEncoder> encoder, love::g
 	if (uniformBuffer->getSize() < uniformBufferOffset + size)
 	{
 		size_t newsize = uniformBuffer->getSize() * 2;
+		if (uniformBufferOffset > 0)
+			uniformBuffer->nextFrame();
 		uniformBuffer->release();
-		uniformBuffer = CreateStreamBuffer(device, BUFFERUSAGE_VERTEX, newsize);
+		uniformBuffer = CreateStreamBuffer(device, BUFFERUSAGE_UNIFORM, newsize);
 		uniformBufferData = {};
 		uniformBufferOffset = 0;
 	}
 
 	if (uniformBufferData.data == nullptr)
+	{
 		uniformBufferData = uniformBuffer->map(uniformBuffer->getSize());
+		uniformBufferGPUStart = uniformBuffer->getGPUReadOffset();
+	}
 
 	memcpy(uniformBufferData.data + uniformBufferOffset, bufferdata, size);
 
@@ -1050,7 +1056,7 @@ bool Graphics::applyShaderUniforms(id<MTLComputeCommandEncoder> encoder, love::g
 	int uniformindex = Shader::getUniformBufferBinding();
 
 	auto &bindings = renderBindings;
-	setBuffer(encoder, bindings, uniformindex, buffer, uniformBufferOffset);
+	setBuffer(encoder, bindings, uniformindex, buffer, uniformBufferGPUStart + uniformBufferOffset);
 
 	uniformBufferOffset += alignUp(size, alignment);
 
@@ -1141,14 +1147,19 @@ void Graphics::applyShaderUniforms(id<MTLRenderCommandEncoder> renderEncoder, lo
 	if (uniformBuffer->getSize() < uniformBufferOffset + size)
 	{
 		size_t newsize = uniformBuffer->getSize() * 2;
+		if (uniformBufferOffset > 0)
+			uniformBuffer->nextFrame();
 		uniformBuffer->release();
-		uniformBuffer = CreateStreamBuffer(device, BUFFERUSAGE_VERTEX, newsize);
+		uniformBuffer = CreateStreamBuffer(device, BUFFERUSAGE_UNIFORM, newsize);
 		uniformBufferData = {};
 		uniformBufferOffset = 0;
 	}
 
 	if (uniformBufferData.data == nullptr)
+	{
 		uniformBufferData = uniformBuffer->map(uniformBuffer->getSize());
+		uniformBufferGPUStart = uniformBuffer->getGPUReadOffset();
+	}
 
 	memcpy(uniformBufferData.data + uniformBufferOffset, bufferdata, size);
 
@@ -1156,8 +1167,8 @@ void Graphics::applyShaderUniforms(id<MTLRenderCommandEncoder> renderEncoder, lo
 	int uniformindex = Shader::getUniformBufferBinding();
 
 	auto &bindings = renderBindings;
-	setBuffer(renderEncoder, bindings, SHADERSTAGE_VERTEX, uniformindex, buffer, uniformBufferOffset);
-	setBuffer(renderEncoder, bindings, SHADERSTAGE_PIXEL, uniformindex, buffer, uniformBufferOffset);
+	setBuffer(renderEncoder, bindings, SHADERSTAGE_VERTEX, uniformindex, buffer, uniformBufferGPUStart + uniformBufferOffset);
+	setBuffer(renderEncoder, bindings, SHADERSTAGE_PIXEL, uniformindex, buffer, uniformBufferGPUStart + uniformBufferOffset);
 
 	uniformBufferOffset += alignUp(size, alignment);
 

src/modules/graphics/metal/StreamBuffer.mm

@@ -67,6 +67,11 @@ public:
 		}
 	}}
 
+	size_t getGPUReadOffset() const override
+	{
+		return (frameIndex * bufferSize) + frameGPUReadOffset;
+	}
+
 	MapInfo map(size_t /*minsize*/) override
 	{
 		// Make sure this frame's section of the buffer is done being used.

src/modules/graphics/opengl/StreamBuffer.cpp

@@ -63,6 +63,11 @@ public:
 		delete[] data;
 	}
 
+	size_t getGPUReadOffset() const override
+	{
+		return (size_t) data;
+	}
+
 	MapInfo map(size_t /*minsize*/) override
 	{
 		return MapInfo(data, bufferSize);
@@ -111,6 +116,11 @@ public:
 		delete[] data;
 	}
 
+	size_t getGPUReadOffset() const override
+	{
+		return frameGPUReadOffset;
+	}
+
 	MapInfo map(size_t /*minsize*/) override
 	{
 		if (orphan)
@@ -192,6 +202,11 @@ public:
 
 	virtual ~StreamBufferSync() {}
 
+	size_t getGPUReadOffset() const override
+	{
+		return (frameIndex * bufferSize) + frameGPUReadOffset;
+	}
+
 	void nextFrame() override
 	{
 		// Insert a GPU fence for this frame's section of the data, we'll wait

src/modules/graphics/vulkan/StreamBuffer.cpp

@@ -96,6 +96,11 @@ ptrdiff_t StreamBuffer::getHandle() const
 	return (ptrdiff_t) buffer;
 }
 
+size_t getGPUReadOffset() const override
+{
+	return (frameIndex * bufferSize) + frameGPUReadOffset;
+}
+
 love::graphics::StreamBuffer::MapInfo StreamBuffer::map(size_t /*minsize*/)
 {
 	// TODO: do we also need to wait until a fence is complete, here?