usable sync of ringbuffer

Source/Plugins/Plugin_VideoOGL/Src/StreamBuffer.cpp

@@ -26,12 +26,12 @@ namespace OGL
 static const u32 SYNC_POINTS = 16;
 
 StreamBuffer::StreamBuffer(u32 type, size_t size, StreamType uploadType)
-: m_uploadtype(uploadType), m_buffertype(type), m_size(size), m_iterator(0), m_last_iterator(0)
+: m_uploadtype(uploadType), m_buffertype(type), m_size(size), m_iterator(0), m_used_iterator(0), m_free_iterator(0)
 {
 	glGenBuffers(1, &m_buffer);
 	
 	if(m_uploadtype == STREAM_DETECT)
-		m_uploadtype = MAP_AND_ORPHAN;
+		m_uploadtype = MAP_AND_SYNC;
 	
 	Init();
 }
@@ -42,6 +42,8 @@ StreamBuffer::~StreamBuffer()
 	glDeleteBuffers(1, &m_buffer);
 }
 
+#define SLOT(x) (x)*SYNC_POINTS/m_size
+
 void StreamBuffer::Alloc ( size_t size, u32 stride )
 {
 	size_t m_iterator_aligned = m_iterator;
@@ -49,35 +51,57 @@ void StreamBuffer::Alloc ( size_t size, u32 stride )
 		m_iterator_aligned--;
 		m_iterator_aligned = m_iterator_aligned - (m_iterator_aligned % stride) + stride;
 	}
+	size_t iter_end = m_iterator_aligned + size;
 	
 	switch(m_uploadtype) {
 	case MAP_AND_ORPHAN:
-		if(m_iterator_aligned+size >= m_size) {
+		if(iter_end >= m_size) {
 			glBufferData(m_buffertype, m_size, NULL, GL_STREAM_DRAW);
 			m_iterator_aligned = 0;
 		}
 		break;
 	case MAP_AND_SYNC:
-		for(u32 i=m_iterator*SYNC_POINTS/m_size+1; i<(m_iterator_aligned+size)*SYNC_POINTS/m_size+1 && i < SYNC_POINTS; i++)
+		
+		// insert waiting slots for used memory
+		for(u32 i=SLOT(m_used_iterator); i<SLOT(m_iterator); i++)
+		{
+			fences[i] = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
+		}
+		m_used_iterator = m_iterator;
+		
+		// wait for new slots to end of buffer
+		for(u32 i=SLOT(m_free_iterator)+1; i<=SLOT(iter_end) && i < SYNC_POINTS; i++)
 		{
 			glClientWaitSync(fences[i], GL_SYNC_FLUSH_COMMANDS_BIT, GL_TIMEOUT_IGNORED);
 			glDeleteSync(fences[i]);
 		}
+		m_free_iterator = iter_end;
 		
-		if(m_iterator_aligned+size >= m_size) {
+		// if buffer is full
-			for(u32 i=m_last_iterator*SYNC_POINTS/m_size; i < SYNC_POINTS; i++)
+		if(iter_end >= m_size) {
+			
+			// insert waiting slots in unused space at the end of the buffer
+			for(u32 i=SLOT(m_used_iterator); i < SYNC_POINTS; i++)
 				fences[i] = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
-			m_iterator_aligned = 0;
+			
-			m_last_iterator = 0;
+			// move to the start
-			glClientWaitSync(fences[0], GL_SYNC_FLUSH_COMMANDS_BIT, GL_TIMEOUT_IGNORED);
+			m_used_iterator = m_iterator_aligned = m_iterator = 0; // offset 0 is always aligned
-			glDeleteSync(fences[0]);
+			iter_end = size;
+			
+			// wait for space at the start
+			for(u32 i=0; i<=SLOT(iter_end); i++)
+			{
+				glClientWaitSync(fences[i], GL_SYNC_FLUSH_COMMANDS_BIT, GL_TIMEOUT_IGNORED);
+				glDeleteSync(fences[i]);
+			}
+			m_free_iterator = iter_end;
 		}
+		
 		break;
 	case BUFFERSUBDATA:
 		m_iterator_aligned = 0;
 		break;
 	}
-	
 	m_iterator = m_iterator_aligned;
 }
 
@@ -85,8 +109,6 @@ size_t StreamBuffer::Upload ( u8* data, size_t size )
 {
 	switch(m_uploadtype) {
 	case MAP_AND_SYNC:
-		for(u32 i=m_last_iterator*SYNC_POINTS/m_size; i<m_iterator*SYNC_POINTS/m_size; i++)
-			fences[i] = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
 	case MAP_AND_ORPHAN:
 		pointer = (u8*)glMapBufferRange(m_buffertype, m_iterator, size, GL_MAP_WRITE_BIT | GL_MAP_UNSYNCHRONIZED_BIT);
 		if(pointer) {
@@ -100,9 +122,9 @@ size_t StreamBuffer::Upload ( u8* data, size_t size )
 		glBufferSubData(m_buffertype, m_iterator, size, data);
 		break;
 	}
-	m_last_iterator = m_iterator;
+	size_t ret = m_iterator;
 	m_iterator += size;
-	return m_last_iterator;
+	return ret;
 }
 
 void StreamBuffer::Init()

Source/Plugins/Plugin_VideoOGL/Src/StreamBuffer.h

@@ -53,7 +53,8 @@ private:
 	size_t m_size;
 	u8 *pointer;
 	size_t m_iterator;
-	size_t m_last_iterator;
+	size_t m_used_iterator;
+	size_t m_free_iterator;
 	GLsync *fences;
 };