& now looking back at the patch-in-progress, I see it allows setting the OptBisector/OptPassGate as suggested in (2). If that becomes the /only/ option (ie: LLVMContext has no default OptPassGate) then the virtual interface could be kept down in IR (though it's still a bit questionable to have those Analysis types (Loop, Region, CallGraphSCC) even declared in IR). Then the implementation of OptBisector could be moved into Analysis - freely able to depend on the concrete Analysis types. - Dave On Thu, Mar 29, 2018 at 2:01 PM David Blaikie <dblaikie at gmail.com> wrote:> So... looking at OptBisect, I have a few thoughts: > > 1) what's the purpose of the virtual interface/OptPassGate? I'm guessing > maybe that worked around the circular referencing in these APIs? hmm, no, I > suppose that wouldn't work/be relevant here. > > 2) Why is OptBisector a ManagedStatic? That seems pretty antithetical to > the role of LLVMContext. When/why would a user be bisecting over multiple > LLVMContexts? & even then, maybe it'd be more suitable for that grouping > (the scope for the bisection) to be API driven - passing the bisector into > the LLVMContext ctor to define the set of contexts that share a bisector? > > On Wed, Mar 21, 2018 at 10:20 PM Yevgeny Rouban via llvm-dev < > llvm-dev at lists.llvm.org> wrote: > >> Andrew, >> >> I would not make the caller pass the description of the IR unit. That is >> because it would result in the description generated every time even if >> OptBisect is disabled. Description generation is not very chip. >> Thinking on the OptBisect extension, I believe passing the units are the >> right choice because OptPassGates may use them to make pass skipping >> decisions. >> >> -Yevgeny Rouban >> ----------------------------------------------------------- >> >> *From:* llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org >> <llvm-dev-bounces at lists.llvm.org>] *On Behalf Of *Kaylor, Andrew via llvm >> -dev >> *Sent:* Thursday, March 22, 2018 3:52 AM >> *To:* David Blaikie <dblaikie at gmail.com>; llvm-dev <llvm-dev at lists.llvm >> .org>; Friedman, Eli <efriedma at codeaurora.org> >> *Subject:* Re: [llvm-dev] Opt Bisect layering >> >> >> >> There is a patch under review right now from someone who wants to provide >> a mechanism to replace OptBisect as the pass gate keeping mechanism. >> >> >> >> https://reviews.llvm.org/D44464 >> >> >> >> Possibly we could take this opportunity to move OptBisect to a different >> layer, though I don’t know where else it would belong. >> >> >> >> The other possibility is that we could have the caller pass in a >> description instead of a pointer to the pass and the IR unit. OptBisect >> isn’t doing anything with them other than building a string for output. >> >> >> >> -Andy >> >> _______________________________________________ >> LLVM Developers mailing list >> llvm-dev at lists.llvm.org >> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev >> >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20180329/0c8fd140/attachment.html>
On 03/30/2018 12:05 AM, David Blaikie via llvm-dev wrote: > & now looking back at the patch-in-progress, I see it allows setting the OptBisector/OptPassGate as suggested in (2). Well, the patch currently discussed does not attempt to solve the passgate object management issue. It is left for the discretion of passgate object provider. > > If that becomes the /only/ option (ie: LLVMContext has no default OptPassGate) then the virtual interface could be kept down in IR (though it's still a bit questionable to have those Analysis types (Loop, Region, CallGraphSCC) even declared in IR). Then the implementation of OptBisector could be moved into Analysis - freely able to depend on the concrete Analysis types. To me this is a "Pass Manager catch" - entity that attempts to control all the passes needs to be part of (or tightly cooperate with) pass manager. Pass manager is currently in IR, and perhaps rightfully so. Yet passes that it controls work on "IR units" which are either IR or Analysis, thus Analysis leaks into the interfaces inevitably. Kinda logical conflict it is... regards, Fedor. > > - Dave > > On Thu, Mar 29, 2018 at 2:01 PM David Blaikie <dblaikie at gmail.com> wrote: > > So... looking at OptBisect, I have a few thoughts: > > 1) what's the purpose of the virtual interface/OptPassGate? I'm guessing maybe that worked around the circular referencing in these APIs? hmm, no, I suppose that wouldn't work/be relevant here. > > 2) Why is OptBisector a ManagedStatic? That seems pretty antithetical to the role of LLVMContext. When/why would a user be bisecting over multiple LLVMContexts? & even then, maybe it'd be more suitable for that grouping (the scope for the bisection) to be API driven - passing the bisector into the LLVMContext ctor to define the set of contexts that share a bisector? > > On Wed, Mar 21, 2018 at 10:20 PM Yevgeny Rouban via llvm-dev <llvm-dev at lists.llvm.org> wrote: > > Andrew, > > I would not make the caller pass the description of the IR unit. That is because it would result in the description generated every time even if OptBisect is disabled. Description generation is not very chip. > Thinking on the OptBisect extension, I believe passing the units are the right choice because OptPassGates may use them to make pass skipping decisions. > > -Yevgeny Rouban > ----------------------------------------------------------- > > From: llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org] On Behalf Of Kaylor, Andrew via llvm-dev > Sent: Thursday, March 22, 2018 3:52 AM > To: David Blaikie <dblaikie at gmail.com>; llvm-dev <llvm-dev at lists.llvm.org>; Friedman, Eli <efriedma at codeaurora.org> > Subject: Re: [llvm-dev] Opt Bisect layering > > > > There is a patch under review right now from someone who wants to provide a mechanism to replace OptBisect as the pass gate keeping mechanism. > > > > https://reviews.llvm.org/D44464 > > > > Possibly we could take this opportunity to move OptBisect to a different layer, though I don’t know where else it would belong. > > > > The other possibility is that we could have the caller pass in a description instead of a pointer to the pass and the IR unit. OptBisect isn’t doing anything with them other than building a string for output. > > > > -Andy > > _______________________________________________ > LLVM Developers mailing list > llvm-dev at lists.llvm.org > http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev > > > > _______________________________________________ > LLVM Developers mailing list > llvm-dev at lists.llvm.org > http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
On Mon, Apr 2, 2018 at 11:32 PM Fedor Sergeev via llvm-dev < llvm-dev at lists.llvm.org> wrote:> On 03/30/2018 12:05 AM, David Blaikie via llvm-dev wrote: > > & now looking back at the patch-in-progress, I see it allows setting > the OptBisector/OptPassGate as suggested in (2). > Well, the patch currently discussed does not attempt to solve the > passgate object management issue. > It is left for the discretion of passgate object provider. > > > > > If that becomes the /only/ option (ie: LLVMContext has no default > OptPassGate) then the virtual interface could be kept down in IR (though > it's still a bit questionable to have those Analysis types (Loop, > Region, CallGraphSCC) even declared in IR). Then the implementation of > OptBisector could be moved into Analysis - freely able to depend on the > concrete Analysis types. > > To me this is a "Pass Manager catch" - entity that attempts to control > all the passes needs to be part of (or tightly cooperate with) pass > manager. > Pass manager is currently in IR, and perhaps rightfully so. > Yet passes that it controls work on "IR units" which are either IR or > Analysis, thus Analysis leaks into the interfaces inevitably. > Kinda logical conflict it is... >This is in response to my "it's still a bit questionable" comment? That's not too important - I'm not pushing to change that if we can get the mechanical layering functional regardless, by only having forward declarations of those different Analysis entities in llvm/IR, but not need their definitions except in the implementation of this virtual interface which could live in llvm/Analysis. But to discuss it anyway: It seems a bit different that the "Pass Manager catch" depends on the concrete types but the Pass Manager (PassManager.h) itself, does not - it's only templates, none of it depends on Region, Loop, etc. If the catch could be implemented similarly to the manager itself, then it'd have the same layering requirements & no problem. But I haven't looked closely enough at the APIs to figure out if/how that might be done - the current implementation/mechanisms are at odds because of the incompatibility of templates and virtual dispatch (can't have a virtual function template - it'd have an unbounded/unknowable number of vtable entries, etc). Some sort of visitor-y thing might be needed/useful, I'm not sure. But again, not sure this is necessary to address/fix for the issues I'm seeing/pushing to deal with - but I'm happy to discuss/help design this area as well if you'd like :) - Dave> > regards, > Fedor. > > > > > - Dave > > > > On Thu, Mar 29, 2018 at 2:01 PM David Blaikie <dblaikie at gmail.com> > wrote: > > > > So... looking at OptBisect, I have a few thoughts: > > > > 1) what's the purpose of the virtual interface/OptPassGate? I'm > guessing maybe that worked around the circular referencing in these > APIs? hmm, no, I suppose that wouldn't work/be relevant here. > > > > 2) Why is OptBisector a ManagedStatic? That seems pretty > antithetical to the role of LLVMContext. When/why would a user be > bisecting over multiple LLVMContexts? & even then, maybe it'd be more > suitable for that grouping (the scope for the bisection) to be API > driven - passing the bisector into the LLVMContext ctor to define the > set of contexts that share a bisector? > > > > On Wed, Mar 21, 2018 at 10:20 PM Yevgeny Rouban via llvm-dev > <llvm-dev at lists.llvm.org> wrote: > > > > Andrew, > > > > I would not make the caller pass the description of the IR > unit. That is because it would result in the description generated every > time even if OptBisect is disabled. Description generation is not very > chip. > > Thinking on the OptBisect extension, I believe passing the > units are the right choice because OptPassGates may use them to make > pass skipping decisions. > > > > -Yevgeny Rouban > > ----------------------------------------------------------- > > > > From: llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org] On > Behalf Of Kaylor, Andrew via llvm-dev > > Sent: Thursday, March 22, 2018 3:52 AM > > To: David Blaikie <dblaikie at gmail.com>; llvm-dev > <llvm-dev at lists.llvm.org>; Friedman, Eli <efriedma at codeaurora.org> > > Subject: Re: [llvm-dev] Opt Bisect layering > > > > > > > > There is a patch under review right now from someone who > wants to provide a mechanism to replace OptBisect as the pass gate > keeping mechanism. > > > > > > > > https://reviews.llvm.org/D44464 > > > > > > > > Possibly we could take this opportunity to move OptBisect to > a different layer, though I don’t know where else it would belong. > > > > > > > > The other possibility is that we could have the caller pass > in a description instead of a pointer to the pass and the IR unit. > OptBisect isn’t doing anything with them other than building a string > for output. > > > > > > > > -Andy > > > > _______________________________________________ > > LLVM Developers mailing list > > llvm-dev at lists.llvm.org > > http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev > > > > > > > > _______________________________________________ > > LLVM Developers mailing list > > llvm-dev at lists.llvm.org > > http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev > > > _______________________________________________ > LLVM Developers mailing list > llvm-dev at lists.llvm.org > http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20180403/deffd189/attachment.html>