GHC: Ticket #4962: Dead code fed to CorePrep because RULEs keep it alive spuriously

I've implemented and validated a patch that does this (attached). Pretty straightforward, and it seems to work. Nofib results are: Min +0.0% -4.1% +0.0% +0.5% +0.0% Max +0.1% +0.0% +2.3% +274.5% +0.0%...

View Article

comment added

February 18, 2011, 4:25 pm

Sounds like a good plan, I'm trying it out.

View Article

comment added

February 18, 2011, 2:08 am

OK, here's a refined plan In CorePrep.cloneBndr, drop all unfoldings or rules In CorePrep.deFloatTop, which is the bit right at the end of CorePrep, do occurAnalyseExpr on the RHS of each top-level...

View Article

comment added

February 16, 2011, 4:30 am

Roman: yes, sorry - let's say that go is used in the body of the let. Simon: right. However, I'm not so sure that CoreTidy? is actually preserving rules on local lets right now.. certainly nothing in...

View Article

comment added

February 16, 2011, 3:17 am

Ah, I thought the "drop local rules" bit happens after recording unfoldings for the interface file. So the rules and local bindings would appear in the unfolding but not in the actual code. Basically,...

View Article

comment added

February 16, 2011, 3:01 am

But in principle if this definition appears in the interface file, it may be inlined, and then the local RULES may spring into action, no? What you say about dropping local rules is true if the defn...

View Article

comment added

February 16, 2011, 2:32 am

Where is go used in your example? If it isn't, should it be dropped together with the rule? Simon, didn't we discuss a similar problem a while ago? I vaguely remember that the conclusion was to drop...

View Article