#include <CZCopyRingBuffer.h>
class CZCopyRingBuffer
{
public:
CZCopyRingBuffer(CRingBuffer* pRing);
void* get(size_t nBytes); // Note blocks until nBytes are there.
void done();
};
The CRingBuffer class provides simple,
byte-stream oriented access to NSCLDAQ ring buffers. Retrieving data
from ring buffers using this class, however, results in a copy
being performed. If high performance is required, these
data copies can be a performance impediment.
CZCopyRingBuffer provides mostly copy free
access to data placed in a ring buffer. Copies will only take place
if the requested data spans the ring buffer wrap in which case
that request will be satisfied by creating a local copy
for the data that is contiguous in virtual memory.
CZCopyRingBuffer(CRingBuffer* pRing);
Creates a zero copy ring buffer access object given
a pointer to an underlying CRingBuffer
object. The ring buffer must have been attached as
a consumer (CRingBuffer::consumer).
void* get(size_t nBytes);
If necessar blocks until at least
nBytes of data are avalable
for consumption in the underlying ring. The return pointer
is then a pointer nBytes of
data. In most cases this pointer will point directly
the ring buffer and the data pointed to should be
treated as readonly. If the data wrapped, the pointer
will point at a copy of the data that is contiguous.
Once the callers is done using the data it should
invoke the done method.
void done();
Once the client is done using data returned by
get it should call
this method. The consumer pointer for the client
is then advanced past the data gotten by
get.
Failure to invoke done will
eventually result in the ring buffer being unable to
accept new data from the producer resulting in
a backpressure flow control condition.