Make clearNursery free

Summary:
clearNursery resets all the bd->free pointers of nursery blocks to
make the blocks empty.  In profiles we've seen clearNursery taking
significant amounts of time particularly with large -N and -A values.

This patch moves the work of clearNursery to the point at which we
actually need the new block, thereby introducing an invariant that
blocks to the right of the CurrentNursery pointer still need their
bd->free pointer reset.  This should make things faster overall,
because we don't need to clear blocks that we don't use.

Test Plan: validate

Reviewers: AndreasVoellmy, ezyang, austin

Subscribers: thomie, carter, ezyang, simonmar

Differential Revision: https://phabricator.haskell.org/D318

1 files changed, 21 insertions, 0 deletions

diff --git a/includes/rts/storage/GC.h b/includes/rts/storage/GC.h
index c171b67252..db71e951ea 100644
--- a/includes/rts/storage/GC.h
+++ b/includes/rts/storage/GC.h
@@ -66,6 +66,27 @@ typedef struct nursery_ {
     memcount       n_blocks;
 } nursery;
 
+// Nursery invariants:
+//
+//  - cap->r.rNursery points to the nursery for this capability
+//
+//  - cap->r.rCurrentNursery points to the block in the nursery that we are
+//    currently allocating into.  While in Haskell the current heap pointer is
+//    in Hp, outside Haskell it is stored in cap->r.rCurrentNursery->free.
+//
+//  - the blocks *after* cap->rCurrentNursery in the chain are empty
+//    (although their bd->free pointers have not been updated to
+//    reflect that)
+//
+//  - the blocks *before* cap->rCurrentNursery have been used.  Except
+//    for rCurrentAlloc.
+//
+//  - cap->r.rCurrentAlloc is either NULL, or it points to a block in
+//    the nursery *before* cap->r.rCurrentNursery.
+//
+// See also Note [allocation accounting] to understand how total
+// memory allocation is tracked.
+
 typedef struct generation_ {
     nat            no;                  // generation number