Hal Finkel via llvm-dev
2017-Nov-04 00:45 UTC
[llvm-dev] FW: clarification needed for the constrained fp implementation.
On 11/03/2017 05:26 PM, 陳韋任 via llvm-dev wrote:> > > 2017-11-04 4:29 GMT+08:00 Kaylor, Andrew via llvm-dev > <llvm-dev at lists.llvm.org <mailto:llvm-dev at lists.llvm.org>>: > > Copying the list on a discussion of potentially general interest…. > > *From:* Kaylor, Andrew > *Sent:* Friday, November 03, 2017 1:11 PM > *To:* 'Ding, Wei' <Wei.Ding2 at amd.com <mailto:Wei.Ding2 at amd.com>>; > Sumner, Brian <Brian.Sumner at amd.com > <mailto:Brian.Sumner at amd.com>>; Arsenault, Matthew > <Matthew.Arsenault at amd.com <mailto:Matthew.Arsenault at amd.com>> > *Subject:* RE: clarification needed for the constrained fp > implementation. > > Hi Wei, > > I’ve been meaning to write something up for discussion on the LLVM > Dev list about this. I hope you don’t mind if I copy the list now > to accomplish that while also answering your questions. > Eventually I create a document describing this in more detail and > less formally than the language definition. > > Basically, the “constraints” in the constrained FP intrinisics are > constraints on the optimizer. They are a way of telling the > optimizer what it can and cannot assume about rounding mode and FP > exception behavior. By default, the optimizer assumes that the > rounding mode is round-to-nearest and that FP exceptions are being > ignored. If the user code is going to do anything that > invalidates these assumptions, then we need a way to make the > optimizer stop assuming that. That’s what the intrinisics do. > Because most passes don’t recognize the intrinisics, they can’t do > anything with the operations they represent and therefore can’t > make any assumption about them. > > The intrinsics are not intended to do anything to change the > rounding mode or FP exception handling state. I have an idea in > mind for some additional intrinsics that would provide a way to > control the FP environment. There are already some > target-specific mechanisms for doing that, but I’d like to have > something that’s target independent. I’ll say more about this in a > minute. > > I mentioned in my review comments that my work on this has been > motivated by the STDC pragmas, and I think if I explain that it > might make the semantics of the intrinsics seem a little more > natural. The primary pragma I have in mind here is the “STDC > FENV_ACCESS” pragma. I believe this is part of the C99 standard, > but compiler support for it is still mostly (if not entirely) > missing. For instance, if you try to use this pragma with clang > you will get a message telling you that the pragma isn’t supported > and it will have no other effect. We want to change that. > > Basically, what the “STDC FENV_ACCESS” pragma does is provide > programmers with a way to tell the compiler that the program might > change the FP environment. This pragma represents a setting that > has only two states -- on and off. The default setting of this > state is documented as being implementation defined. In clang the > default state will be off. The C99 standard states that accessing > the FP environment (testing FP status flags, changing FP control > modes, etc.) when FENV_ACCESS is off is undefined behavior. The > C99 standard provides library calls to access the environment > (fesetround, fegetround, fetestexcept, etc.) but you can only > safely use these if you have set FENV_ACCESS to the “on” state. A > typical usage might look like this: > > #include <fenv.h> > > double someFunc(double A, double B, bool ForceRoundUp) { > > #pragma STDC FENV_ACCESS ON > > double Result; > > if (ForceRoundUp) { > > int OldRM = fegetround(); > > fesetround(FE_UPWARD); > > Result = A/B; > > fesetround(OldRM); > > } else { > > Result = A/B; > > } > > return Result; > > } > > > ... > abridge > ... > > > What I’m thinking is that we need something like this: > > void llvm.set.roundingmode(i32 mode) > > i32 lllvm.get.roundingmode() > > These would then get translated during instruction selection to > target-specific instructions, which is equivalent to what > fesetround() and fegetround() do. But I think it would also be > useful to have something like this: > > void llvm.begin.local.roundingmode(i32 mode) > > void llvm.end.local.roundingmode() > > > A little bit of curiously. What if user doesn't restore the OldRM > like this > > double someFunc(double A, double B, bool ForceRoundUp) { > #pragma STDC FENV_ACCESS ON > double Result; > if (ForceRoundUp) { > int OldRM = fegetround(); > fesetround(FE_UPWARD); > Result = A/B; > // > fesetround(OldRM); > } else { > Result = A/B; > } > return Result; > > } > > > Are we still going to generate > llvm.begin.local.roundingmode/llvm.end.local.roundingmode?I think that, however we do this, we'll need a way of dealing with "live in" and "live out" FP-environment state. Moreover, any external call can change these as well (unless we prove/know that it doesn't). Translating these calls into intrinsics so that the optimizer can reason about them seems like a reasonable plan. -Hal> > Regards, > chenwj > > -- > Wei-Ren Chen (陳韋任) > Homepage: https://people.cs.nctu.edu.tw/~chenwj > <https://people.cs.nctu.edu.tw/%7Echenwj> > > > _______________________________________________ > LLVM Developers mailing list > llvm-dev at lists.llvm.org > http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev-- Hal Finkel Lead, Compiler Technology and Programming Languages Leadership Computing Facility Argonne National Laboratory -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20171103/381f336d/attachment.html>
Kaylor, Andrew via llvm-dev
2017-Nov-06 19:43 UTC
[llvm-dev] FW: clarification needed for the constrained fp implementation.
> A little bit of curiously. What if user doesn't restore the OldRM like this: > > double someFunc(double A, double B, bool ForceRoundUp) { > #pragma STDC FENV_ACCESS ON > double Result; > if (ForceRoundUp) { > int OldRM = fegetround(); > fesetround(FE_UPWARD); > Result = A/B; > > // fesetround(OldRM); > } else { > Result = A/B; > } > return Result; > } > > Are we still going to generate llvm.begin.local.roundingmode/llvm.end.local.roundingmode?I was actually thinking that fegetround and fesetround would always be translated into llvm.get.roundingmode and llvm.set.roundingmode at least initially, because it’s a bit of trouble to prove that they are being used to implement the local rounding mode idiom. Certainly recognizing that kind of usage is more work than a front end should be doing. I would propose the begin/end local rounding mode intrinsics as a convenience for cases where some user (probably never clang) specifically knew that this was the intended behavior. I believe Wei has a use case like that, and I’ve also spoken to someone working on the Julia language who would like to be able to do this. That does still leave the problem of what happens if there is a function call inside a pair of begin/end local rounding mode intrinsics. I’m not entirely sure the best way to handle this. Do we try to restrict calls that might change the rounding mode or do we leave it up to the front end and/or user to make sure things are safe. As for the C/FENV_ACCESS case, I seriously feel like the way that pragma is defined leaves users with a huge amount of room to shoot themselves in the foot. If I’m reading the specifications correctly, it says that changing from a scope with FENV_ACCESS on to a scope with FENV_ACCESS off or vice versa without the FP environment in its default state is undefined behavior. At some point it would probably be nice to have a sanitizer that checks this. -Andy From: Hal Finkel [mailto:hfinkel at anl.gov] Sent: Friday, November 03, 2017 5:46 PM To: 陳韋任 <chenwj.cs97g at g2.nctu.edu.tw>; Kaylor, Andrew <andrew.kaylor at intel.com> Cc: llvm-dev <llvm-dev at lists.llvm.org> Subject: Re: [llvm-dev] FW: clarification needed for the constrained fp implementation. On 11/03/2017 05:26 PM, 陳韋任 via llvm-dev wrote: 2017-11-04 4:29 GMT+08:00 Kaylor, Andrew via llvm-dev <llvm-dev at lists.llvm.org<mailto:llvm-dev at lists.llvm.org>>: Copying the list on a discussion of potentially general interest…. From: Kaylor, Andrew Sent: Friday, November 03, 2017 1:11 PM To: 'Ding, Wei' <Wei.Ding2 at amd.com<mailto:Wei.Ding2 at amd.com>>; Sumner, Brian <Brian.Sumner at amd.com<mailto:Brian.Sumner at amd.com>>; Arsenault, Matthew <Matthew.Arsenault at amd.com<mailto:Matthew.Arsenault at amd.com>> Subject: RE: clarification needed for the constrained fp implementation. Hi Wei, I’ve been meaning to write something up for discussion on the LLVM Dev list about this. I hope you don’t mind if I copy the list now to accomplish that while also answering your questions. Eventually I create a document describing this in more detail and less formally than the language definition. Basically, the “constraints” in the constrained FP intrinisics are constraints on the optimizer. They are a way of telling the optimizer what it can and cannot assume about rounding mode and FP exception behavior. By default, the optimizer assumes that the rounding mode is round-to-nearest and that FP exceptions are being ignored. If the user code is going to do anything that invalidates these assumptions, then we need a way to make the optimizer stop assuming that. That’s what the intrinisics do. Because most passes don’t recognize the intrinisics, they can’t do anything with the operations they represent and therefore can’t make any assumption about them. The intrinsics are not intended to do anything to change the rounding mode or FP exception handling state. I have an idea in mind for some additional intrinsics that would provide a way to control the FP environment. There are already some target-specific mechanisms for doing that, but I’d like to have something that’s target independent. I’ll say more about this in a minute. I mentioned in my review comments that my work on this has been motivated by the STDC pragmas, and I think if I explain that it might make the semantics of the intrinsics seem a little more natural. The primary pragma I have in mind here is the “STDC FENV_ACCESS” pragma. I believe this is part of the C99 standard, but compiler support for it is still mostly (if not entirely) missing. For instance, if you try to use this pragma with clang you will get a message telling you that the pragma isn’t supported and it will have no other effect. We want to change that. Basically, what the “STDC FENV_ACCESS” pragma does is provide programmers with a way to tell the compiler that the program might change the FP environment. This pragma represents a setting that has only two states -- on and off. The default setting of this state is documented as being implementation defined. In clang the default state will be off. The C99 standard states that accessing the FP environment (testing FP status flags, changing FP control modes, etc.) when FENV_ACCESS is off is undefined behavior. The C99 standard provides library calls to access the environment (fesetround, fegetround, fetestexcept, etc.) but you can only safely use these if you have set FENV_ACCESS to the “on” state. A typical usage might look like this: #include <fenv.h> double someFunc(double A, double B, bool ForceRoundUp) { #pragma STDC FENV_ACCESS ON double Result; if (ForceRoundUp) { int OldRM = fegetround(); fesetround(FE_UPWARD); Result = A/B; fesetround(OldRM); } else { Result = A/B; } return Result; } ... abridge ... What I’m thinking is that we need something like this: void llvm.set.roundingmode(i32 mode) i32 lllvm.get.roundingmode() These would then get translated during instruction selection to target-specific instructions, which is equivalent to what fesetround() and fegetround() do. But I think it would also be useful to have something like this: void llvm.begin.local.roundingmode(i32 mode) void llvm.end.local.roundingmode() A little bit of curiously. What if user doesn't restore the OldRM like this double someFunc(double A, double B, bool ForceRoundUp) { #pragma STDC FENV_ACCESS ON double Result; if (ForceRoundUp) { int OldRM = fegetround(); fesetround(FE_UPWARD); Result = A/B; // fesetround(OldRM); } else { Result = A/B; } return Result; } Are we still going to generate llvm.begin.local.roundingmode/llvm.end.local.roundingmode? I think that, however we do this, we'll need a way of dealing with "live in" and "live out" FP-environment state. Moreover, any external call can change these as well (unless we prove/know that it doesn't). Translating these calls into intrinsics so that the optimizer can reason about them seems like a reasonable plan. -Hal Regards, chenwj -- Wei-Ren Chen (陳韋任) Homepage: https://people.cs.nctu.edu.tw/~chenwj<https://people.cs.nctu.edu.tw/%7Echenwj> _______________________________________________ LLVM Developers mailing list llvm-dev at lists.llvm.org<mailto:llvm-dev at lists.llvm.org> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev -- Hal Finkel Lead, Compiler Technology and Programming Languages Leadership Computing Facility Argonne National Laboratory -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20171106/6e943c4d/attachment.html>
陳韋任 via llvm-dev
2017-Nov-07 15:45 UTC
[llvm-dev] FW: clarification needed for the constrained fp implementation.
> > I was actually thinking that fegetround and fesetround would always be > translated into llvm.get.roundingmode and llvm.set.roundingmode at least > initially, because it’s a bit of trouble to prove that they are being used > to implement the local rounding mode idiom. Certainly recognizing that kind > of usage is more work than a front end should be doing. > > > > I would propose the begin/end local rounding mode intrinsics as a > convenience for cases where some user (probably never clang) specifically > knew that this was the intended behavior. I believe Wei has a use case > like that, and I’ve also spoken to someone working on the Julia language > who would like to be able to do this. >Okay.> That does still leave the problem of what happens if there is a function > call inside a pair of begin/end local rounding mode intrinsics. I’m not > entirely sure the best way to handle this. Do we try to restrict calls > that might change the rounding mode or do we leave it up to the front end > and/or user to make sure things are safe. >From the 2011 discussion thread on Intel Forum [1], seems the pragma only affects the function containing it. That sounds reasonable to me. [1] https://software.intel.com/en-us/forums/intel-c-compiler/topic/284430> As for the C/FENV_ACCESS case, I seriously feel like the way that pragma > is defined leaves users with a huge amount of room to shoot themselves in > the foot. If I’m reading the specifications correctly, it says that > changing from a scope with FENV_ACCESS on to a scope with FENV_ACCESS off > or vice versa without the FP environment in its default state is undefined > behavior. At some point it would probably be nice to have a sanitizer that > checks this. >I think you can just implement the pragma without worrying too much about how user might mis-use them. Leave the front-end or sanitizer to take care of those issues. -- Wei-Ren Chen (陳韋任) Homepage: https://people.cs.nctu.edu.tw/~chenwj -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20171107/e49b3f54/attachment.html>