search for: addusersimpl

...sal orders. It looks to me that there are two pieces of the implementation of IVUsers that are leading to the fragility (i.e. dependence on input ordering) that I am seeing. A) The inclusion of Processed.count(User) as a condition of the two if statements at approx lines 235 & 241 in IVUsers::AddUsersImpl() of lib/Analysis/IVUsers.cpp. B) It doesn’t terminate a DFT if it encounters a phi node in the loop-header that it hasn’t seen before. To help illustrate these, here are the relevant DFT traces from the sample IR that I provided in the original post. (1) is the IR as-is, and (2) is with the usel...

...of the implementation of >>> IVUsers that are leading to the fragility (i.e. dependence on input >>> ordering) that I am seeing. >>> A) The inclusion of Processed.count(User) as a condition of the two >>> if statements at approx lines 235 & 241 in IVUsers::AddUsersImpl() >>> of lib/Analysis/IVUsers.cpp. >>> B) It doesn’t terminate a DFT if it encounters a phi node in the >>> loop-header that it hasn’t seen before. >>> >>> To help illustrate these, here are the relevant DFT traces from the >>> sample IR tha...

...sal orders. It looks to me that there are two pieces of the implementation of IVUsers that are leading to the fragility (i.e. dependence on input ordering) that I am seeing. A) The inclusion of Processed.count(User) as a condition of the two if statements at approx lines 235 & 241 in IVUsers::AddUsersImpl() of lib/Analysis/IVUsers.cpp. B) It doesn’t terminate a DFT if it encounters a phi node in the loop-header that it hasn’t seen before. To help illustrate these, here are the relevant DFT traces from the sample IR that I provided in the original post. (1) is the IR as-is, and (2) is with the usel...

Hi all, I’ve most recently been grappling with a difficult to reproduce bug. I’ve traced the source of the difficulty in reproduction to the IVUsers analysis pass that is used by Loop Strength Reduction. Specifically, the IVUsers pass’s output is very sensitive to both the use list ordering of the instructions that it is looking at and the ordering of the Phi nodes in the header block of the loop

----- Original Message ----- > From: "Bjarke Roune" <broune at google.com> > To: "Jingyue Wu" <jingyue at google.com> > Cc: llvmdev at cs.uiuc.edu > Sent: Tuesday, June 30, 2015 8:16:13 PM > Subject: Re: [LLVMdev] Deriving undefined behavior from nsw/inbounds/poison for scalar evolution > > Hi Adam, > > Jingyue is right. We need to keep