fixes #972 Very slow pull/push performance compared to ZMQ

This refactors the pipeline protocol to use lightweight mq
instead of the more expensive message queue structure.  It
also provides nicer backpressure and buffering support.
The test suite was updated and converted to NUTS as well.

This won't completely close the gap, but it should help quite
a bit.

1 files changed, 2 insertions, 2 deletions

diff --git a/src/core/lmq.c b/src/core/lmq.c
index bbf42356..47058fc9 100644
--- a/src/core/lmq.c
+++ b/src/core/lmq.c
@@ -22,7 +22,7 @@ nni_lmq_init(nni_lmq *lmq, size_t cap)
 	// operations as a power of two, for efficiency.  It does possibly
 	// waste some space, but never more than 2x.  Consumers should try
 	// for powers of two if they are concerned about efficiency.
-	alloc = 1;
+	alloc = 2;
 	while (alloc < cap) {
 		alloc *= 2;
 	}
@@ -127,7 +127,7 @@ nni_lmq_resize(nni_lmq *lmq, size_t cap)
 	size_t    alloc;
 	size_t    len;
 
-	alloc = 1;
+	alloc = 2;
 	while (alloc < cap) {
 		alloc *= 2;
 	}