similar to: RFC: A change in InstCombine canonical form

On Wed, Mar 16, 2016 at 8:34 AM, Mehdi Amini via llvm-dev < llvm-dev at lists.llvm.org> wrote: > Hi, > > How do it interact with the "typeless pointers" work? > Right - the goal of the typeless pointer work is to fix all these bugs related to "didn't look through bitcasts" in optimizations. Sometimes that's going to mean more work (because the code

On Wed, Mar 16, 2016 at 11:00 AM, Ehsan Amiri <ehsanamiri at gmail.com> wrote: > David, > > Could you give us an update on the status of typeless pointer work? How > much work is left and when you think it might be ready? > It's a bit of an onion peel, really - since it will eventually involve generalizing/fixing every optimization that's currently leaning on typed

Back to the discussion on the RFC, I still see some advantage in following the proposed solution. I see two paths forward: 1- Change canonical form, possibly lower memcpy to non-integer load and store in InstCombine. Then teach the backends to convert that to integer load and store if that is more profitable. Notice that we are talking about loads that have no use other than store. So it is a

I don't know enough about the tradeoff for 1, but 2 seems like a bandaid for something that is not a correctness issue neither a regression. I'm not sure it justifies "bandaid patches" while there is a clear path forward, i.e. typeless pointers, unless there is an acknowledgement that typeless pointers won't be there before a couple of years. -- Mehdi > On Mar 22, 2016,

I don't really mind, but the intermediate stage will not be very nice: that a lot of code / tests that needs to be written with bitcast, and all of that while they are deemed to disappear. The added value isn't clear to me considering the added work. I'm not sure it wouldn't add more work for all the cleanup required by the "typeless pointer", but I'm not sure

I feel very strongly that blocking work on making optimization bitcast-ignorant on the typeless pointer work would be a major mistake. Unless we expected the typeless pointer work to be concluded within the near term (say 3-6 months maximum), we should not block any development which would be accepted in the typeless pointer work wasn't planned. In my view, this is one of the largest

I'd phrase this differently: being pointer-bitcast agnostic is a step towards support typeless pointers. :) We can either become bitcast agnostic all in one big change or incrementally. Personally, I'd prefer the later since it reduces the risk associated with enabling typeless pointers in the end. Philip On 03/22/2016 12:16 PM, Mehdi Amini via llvm-dev wrote: > I don't know

This is roughly what I wrote... > On Mar 22, 2016, at 1:31 PM, Philip Reames <listmail at philipreames.com> wrote: > > I feel very strongly that blocking work on making optimization bitcast-ignorant on the typeless pointer work would be a major mistake. Unless we expected the typeless pointer work to be concluded within the near term (say 3-6 months maximum), we should not block

But not what David was stating, unless I misread? I was specifically responding to David's wording: "If we're talking about making an optimization able to ignore the bitcast instructions - yes, that work is unnecessary & perhaps questionable given the typeless pointer work. Not outright off limits, but the same time might be better invested in moving typeless pointers

Ultimately everything is going to be made to not rely on the types of pointers - that's nearly equivalent to bitcast-ignorant (the difference being that the presence of an extra instruction (the bitcast) might trip up some optimizations - but the presence of the /type/ information implied by the bitcast should not trip up or be necessary for optimizations (two sides of the same coin)) If

----- Original Message ----- > From: "Mehdi Amini via llvm-dev" <llvm-dev at lists.llvm.org> > To: "Philip Reames" <listmail at philipreames.com>, "David Blaikie" > <dblaikie at gmail.com> > Cc: "llvm-dev" <llvm-dev at lists.llvm.org>, uweigand at de.ibm.com, "Tom > Stellard" <thomas.stellard at

On Tue, Mar 22, 2016 at 1:41 PM, Mehdi Amini <mehdi.amini at apple.com> wrote: > Sorry I should have been more clear (writing to many email in parallel) > > You're right. I was adding a +1 to you here. > > Especially I wrote "unless there is an acknowledgement that typeless > pointers won't be there before a couple of years" with the PassManager in >

Sorry I should have been more clear (writing to many email in parallel) You're right. I was adding a +1 to you here. Especially I wrote "unless there is an acknowledgement that typeless pointers won't be there before a couple of years" with the PassManager in mind, and I was expecting from David some good indication of a timeframe for the typeless pointers. If the typeless

Thanks. *Phillip, *As Hal said I do not think (1) is a very large item. Please let me know if I am mistaken. *David *I think (1) is more inline with typeless pointer work than (2). Contributing to typeless pointer work will be great, but given its unknown time frame we cannot stop fixing existing problems. Of course, we should follow an approach consistent with the long-term solution. On Tue,

On 03/22/2016 02:44 PM, Ehsan Amiri wrote: > Thanks. > > *Phillip, *As Hal said I do not think (1) is a very large item. Please > let me know if I am mistaken. I have no specific reason to believe it will be a large amount of work, but prior experience tells me changes to canonical form have a tendency of exposing unexpected issues. To be clear, I am supportive of you implementing

On 03/22/2016 06:50 PM, Ehsan Amiri wrote: > > > On Tue, Mar 22, 2016 at 6:42 PM, Philip Reames > <listmail at philipreames.com <mailto:listmail at philipreames.com>> wrote: > > > > On 03/22/2016 02:44 PM, Ehsan Amiri wrote: >> Thanks. >> >> *Phillip, *As Hal said I do not think (1) is a very large item. >> Please let me

So this change did indeed have an effect! :) I’m seeing regressions in a number of benchmarks mainly due to a host of extra bitcasts that get introduced. Here’s the problem I’m seeing in a nutshell: 1) There is a Phi with input type double 2) Polly demotes the phi into a load/store of type double 3) InstCombine canonicalizes the load/store to use i64 instead of double 4)

Thanks for the reply Pete. Unfortunately, I don’t think it is going to be as simple as ignoring those loads which only store. In findCommonType(), only one alloca is passed in at a time. So, while you could find those cases where that alloca was loaded from and stored elsewhere, you can’t find those places that store to that alloca from somewhere else (at least not easily that I can see). So

Hi all, I have a question regarding dynamic indexing into a vector with GEP. I see that in the ScalarReplAggregates pass in the LLVM 3.2 release the call SROA::isSafeGEP() will now allow alloca scalarization in the case where a GEP index into a vector isn’t a constant. My question is: what is the expected behavior when the index is out of bounds of the vector? Is it undefined? I have an

> On Jan 21, 2015, at 3:02 PM, Chandler Carruth <chandlerc at gmail.com> wrote: > > > On Wed, Jan 21, 2015 at 2:43 PM, Pete Cooper <peter_cooper at apple.com <mailto:peter_cooper at apple.com>> wrote: > The first thing that springs to mind is that I don’t trust the backend to get this right. I don’t think it will understand when an i32 load/store would have been