Closed chrisseaton closed 1 year ago
Did you want to keep KillingBegin
in when you hid Begin
?
It seems the node count (148 nodes, branches, calls
) doesn't consider hidden nodes from that output.
Did you want to keep
KillingBegin
in when you hidBegin
?
That no longer exists since oracle/graal@9149851f8e83d3a79974e95004d1733ec6305549, so I don't think it's worth handling it.
doesn't consider hidden nodes
Wasn't that what you wanted?
Bad phrasing, the node count is higher after this PR than before, that's unexpected, i.e., it should be < 138 nodes (and I think the sum of the histogram is < 138).
Ah the overall count - I'll fix that.
% bundle exec bin/seafoam examples/fib-ruby.bgv.gz:2 describe
47 nodes, branches, calls
ConstantNode: 7
FixedGuardNode: 4
BoxNode$AllocatingBoxNode: 3
BoxNode$TrustedBoxedValue: 3
UnboxNode: 3
*TruffleArgument: 2
*TruffleNew: 2
DeoptimizeNode: 2
ExceptionObjectNode: 2
IntegerSubExactNode: 2
IntegerSubExactOverflowNode: 2
InvokeWithExceptionNode: 2
KillingBeginNode: 2
MethodCallTargetNode: 2
IfNode: 1
IntegerAddExactNode: 1
IntegerAddExactOverflowNode: 1
IntegerLessThanNode: 1
MergeNode: 1
ObjectEqualsNode: 1
ReturnNode: 1
StartNode: 1
ValuePhiNode: 1
Looks good, and that's almost a third of the nodes, making it clear the passes are quite worth it to simplify the graph.
Fixes #65
Before:
After:
Note that synthetic nodes have a
*
in front of their name.CC @eregon