Commit f7d332b4 authored by Simon Marlow's avatar Simon Marlow
Browse files

slight improvement to scavenging of update frames when a collision has occurred

parent 523ecf30
......@@ -1575,11 +1575,13 @@ scavenge_stack(StgPtr p, StgPtr stack_end)
// threadPaused(). We could traverse the whole stack again
// before GC, but that seems like overkill.
// Scavenging this update frame as normal would be disastrous;
// the updatee would end up pointing to the value. So we turn
// the indirection into an IND_PERM, so that evacuate will
// copy the indirection into the old generation instead of
// discarding it.
// So, if the frame points to an indirection, it will get
// shorted out when we evacuate. If this happens, we have no
// closure to update any more. In the past we solved this by
// replacing the IND with an IND_PERM, but a better solution
// is to replace the update frame with a frame that no longer
// does the update and just uses the value already computed by
// the other thread, so that is what we now do.
// Note [upd-black-hole]
// One slight hiccup is that the THUNK_SELECTOR machinery can
......@@ -1590,22 +1592,16 @@ scavenge_stack(StgPtr p, StgPtr stack_end)
// the updatee is never a THUNK_SELECTOR and we're ok.
// NB. this is a new invariant: blackholing is not optional.
nat type;
const StgInfoTable *i;
StgClosure *updatee;
updatee = ((StgUpdateFrame *)p)->updatee;
i = updatee->;
type = get_itbl(updatee)->type;
if (type == IND) {
updatee-> = &stg_IND_PERM_info;
} else if (type == IND_OLDGEN) {
updatee-> = &stg_IND_OLDGEN_PERM_info;
StgClosure *v;
StgUpdateFrame *frame = (StgUpdateFrame *)p;
v = frame->updatee;
if (GET_CLOSURE_TAG(v) != 0 ||
(get_itbl(v)->type != BLACKHOLE &&
get_itbl(v)->type != CAF_BLACKHOLE)) {
frame-> = (const StgInfoTable*)&stg_gc_unpt_r1_info;
evacuate(&((StgUpdateFrame *)p)->updatee);
ASSERT(GET_CLOSURE_TAG(((StgUpdateFrame *)p)->updatee) == 0);
p += sizeofW(StgUpdateFrame);
Supports Markdown
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment