[cmucl-imp] maybe-finish-mapping in map-allocated-objects must be inline
The local function MAYBE-FINISH-MAPPING in MAP-ALLOCATED-OBJECTS must be inlined otherwise the CURRENT pointer is boxed and moved into a value cell for every object. That boxing allocates something like 1MB of saps when (room t) is called with an almost empty heap and much more for if there is something on the heap. There's also a unused variable TESTED. It looks like some parts were written by a drunk :-) Helmut --- a/code/room.lisp +++ b/code/room.lisp @@ -368,7 +368,7 @@ (fixnum (sap+ current size)) (memory-size (sap+ current size))))) (setf current c)))) - (declare (inline next)) + (declare (inline next maybe-finish-mapping)) (loop (maybe-finish-mapping) (maybe-skip-page)
On 8/18/11 1:26 AM, Helmut Eller wrote:
The local function MAYBE-FINISH-MAPPING in MAP-ALLOCATED-OBJECTS must be inlined otherwise the CURRENT pointer is boxed and moved into a value cell for every object. That boxing allocates something like 1MB of saps when (room t) is called with an almost empty heap and much more for if there is something on the heap. Thanks for the patch. I had noticed this some time ago but never got around to investigating why this was happening. Such a simple fix too!
There's also a unused variable TESTED. It looks like some parts were written by a drunk :-) Could be me. I did port some of this from sbcl long ago.
Ray
On 8/18/11 1:26 AM, Helmut Eller wrote:
The local function MAYBE-FINISH-MAPPING in MAP-ALLOCATED-OBJECTS must be inlined otherwise the CURRENT pointer is boxed and moved into a value cell for every object. That boxing allocates something like 1MB of saps when (room t) is called with an almost empty heap and much more for if there is something on the heap.
Patch applied. Thanks! Ray
participants (2)
-
Helmut Eller -
Raymond Toy