thr3ads.net - llvm dev - [LLVMdev] Convert fdiv

If this information is useful, please help other people find it:
Share via:

Chad Rosier

2013-Aug-08 16:25 UTC

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

I would like to transform X/Y -> X*1/Y.  Specifically, I would like to
convert:

define void @t1a(double %a, double %b, double %d) {
entry:
  %div = fdiv fast double %a, %d
  %div1 = fdiv fast double %b, %d
  %call = tail call i32 @foo(double %div, double %div1)
  ret void
}

to:

define void @t1b(double %a, double %b, double %d) {
entry:
  %div = fdiv fast double 1.000000e+00, %d
  %mul = fmul fast double %div, %a
  %mul1 = fmul fast double %div, %b
  %call = tail call i32 @foo(double %mul, double %mul1)
  ret void
}

Is such a transformation best done as a (target-specific) DAG combine?

A similar instcombine already exists for the X/C->X*1/C case (see the
CvtFDivConstToReciprocal function in InstCombineMlDivRem.cpp), but I don't
believe the above can be done as an instcombine as it creates a new
instruction (in addition to replacing the original).  Also, I only want to
perform the transformation if there are multiple uses of 1/Y (like in my
test case).  Otherwise, the transformation replaces a fdiv with a fdiv+fmul
pair, which I doubt would be profitable.

FWIW, I'm also pretty sure this combine requires -fast-math.

Can someone point me in the right direction?

 Thanks,
  Chad
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130808/6ba77a65/attachment.html>

Hal Finkel

2013-Aug-08 16:49 UTC

head link

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

----- Original Message -----> 
> I would like to transform X/Y -> X*1/Y. Specifically, I would like to
> convert:
> 
> define void @t1a(double %a, double %b, double %d) {
> entry:
> %div = fdiv fast double %a, %d
> %div1 = fdiv fast double %b, %d
> %call = tail call i32 @foo(double %div, double %div1)
> ret void
> }
> 
> to:
> 
> define void @t1b(double %a, double %b, double %d) {
> entry:
> %div = fdiv fast double 1.000000e+00, %d
> %mul = fmul fast double %div, %a
> %mul1 = fmul fast double %div, %b
> %call = tail call i32 @foo(double %mul, double %mul1)
> ret void
> }
> 
> Is such a transformation best done as a (target-specific) DAG
> combine?
Perhaps only partially relevant, but I have a fast-math-only target-specific DAG
combine in the PowerPC backend which replaces X/Y -> X*1/Y because we can use
the fast reciprocal estimate + newton iteration for 1/Y instead of a full
divide.

 -Hal
> 
> A similar instcombine already exists for the X/C->X*1/C case (see the
> CvtFDivConstToReciprocal function in InstCombineMlDivRem.cpp), but I
> don't believe the above can be done as an instcombine as it creates
> a new instruction (in addition to replacing the original). Also, I
> only want to perform the transformation if there are multiple uses
> of 1/Y (like in my test case). Otherwise, the transformation
> replaces a fdiv with a fdiv+fmul pair, which I doubt would be
> profitable.
> 
> FWIW, I'm also pretty sure this combine requires -fast-math.
> 
> Can someone point me in the right direction?
> 
> Thanks,
> Chad
> 
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
> 
-- 
Hal Finkel
Assistant Computational Scientist
Leadership Computing Facility
Argonne National Laboratory

Jim Grosbach

2013-Aug-08 16:56 UTC

head link

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

Hi Chad,

This is a great transform to do, but you’re right that it’s only safe under
fast-math. This is particularly interesting when the original divisor is a
constant so you can materialize the reciprocal at compile-time. You’re right
that in either case, this optimization should only kick in when there is more
than one divide instruction that will be changed to a mul.

I don’t have a strong preference for instcombine vs. dagcombine, though I lean
slightly towards later when we’ll have more target information available if we
want to apply a more complicated cost function for some targets.

-Jim


On Aug 8, 2013, at 9:25 AM, Chad Rosier <chad.rosier at gmail.com> wrote:
> I would like to transform X/Y -> X*1/Y.  Specifically, I would like to
convert:
> 
> define void @t1a(double %a, double %b, double %d) {
> entry:
>   %div = fdiv fast double %a, %d
>   %div1 = fdiv fast double %b, %d
>   %call = tail call i32 @foo(double %div, double %div1)
>   ret void
> }
> 
> to:
> 
> define void @t1b(double %a, double %b, double %d) {
> entry:
>   %div = fdiv fast double 1.000000e+00, %d
>   %mul = fmul fast double %div, %a
>   %mul1 = fmul fast double %div, %b
>   %call = tail call i32 @foo(double %mul, double %mul1)
>   ret void
> }
> 
> Is such a transformation best done as a (target-specific) DAG combine?
> 
> A similar instcombine already exists for the X/C->X*1/C case (see the
CvtFDivConstToReciprocal function in InstCombineMlDivRem.cpp), but I don't
believe the above can be done as an instcombine as it creates a new instruction
(in addition to replacing the original).  Also, I only want to perform the
transformation if there are multiple uses of 1/Y (like in my test case). 
Otherwise, the transformation replaces a fdiv with a fdiv+fmul pair, which I
doubt would be profitable.
> 
> FWIW, I'm also pretty sure this combine requires -fast-math.
> 
> Can someone point me in the right direction?
> 
>  Thanks,
>   Chad
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130808/4b7d2f8b/attachment.html>

Weiming Zhao

2013-Aug-08 16:57 UTC

head link

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

Hi Chad,

 

Instcombine can create new instructions (e.g. some code in combine-select
does that). 

We can do it in DAG as well. (it may be easier to implement in DAG. It will
create multiple 1/Y, but later pass should remove the redundancy)

Btw,  it seems the transformation is beneficial only when %d is used as a
common denominator by >2 nominators. 

 

Thanks,

Weiming

From: llvmdev-bounces at cs.uiuc.edu [mailto:llvmdev-bounces at cs.uiuc.edu] On
Behalf Of Chad Rosier
Sent: Thursday, August 08, 2013 9:25 AM
To: llvmdev
Subject: [LLVMdev] Convert fdiv - X/Y -> X*1/Y

 

I would like to transform X/Y -> X*1/Y.  Specifically, I would like to
convert:

define void @t1a(double %a, double %b, double %d) {
entry:
  %div = fdiv fast double %a, %d
  %div1 = fdiv fast double %b, %d
  %call = tail call i32 @foo(double %div, double %div1)
  ret void
}

to:

define void @t1b(double %a, double %b, double %d) {
entry:
  %div = fdiv fast double 1.000000e+00, %d
  %mul = fmul fast double %div, %a
  %mul1 = fmul fast double %div, %b
  %call = tail call i32 @foo(double %mul, double %mul1)
  ret void
}

Is such a transformation best done as a (target-specific) DAG combine?

A similar instcombine already exists for the X/C->X*1/C case (see the
CvtFDivConstToReciprocal function in InstCombineMlDivRem.cpp), but I don't
believe the above can be done as an instcombine as it creates a new
instruction (in addition to replacing the original).  Also, I only want to
perform the transformation if there are multiple uses of 1/Y (like in my
test case).  Otherwise, the transformation replaces a fdiv with a fdiv+fmul
pair, which I doubt would be profitable.

FWIW, I'm also pretty sure this combine requires -fast-math.

Can someone point me in the right direction?

 Thanks,
  Chad

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130808/1e0acc88/attachment.html>

Benjamin Kramer

2013-Aug-08 17:05 UTC

head link

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

On 08.08.2013, at 18:25, Chad Rosier <chad.rosier at gmail.com> wrote:
> I would like to transform X/Y -> X*1/Y.  Specifically, I would like to
convert:
> 
> define void @t1a(double %a, double %b, double %d) {
> entry:
>   %div = fdiv fast double %a, %d
>   %div1 = fdiv fast double %b, %d
>   %call = tail call i32 @foo(double %div, double %div1)
>   ret void
> }
> 
> to:
> 
> define void @t1b(double %a, double %b, double %d) {
> entry:
>   %div = fdiv fast double 1.000000e+00, %d
>   %mul = fmul fast double %div, %a
>   %mul1 = fmul fast double %div, %b
>   %call = tail call i32 @foo(double %mul, double %mul1)
>   ret void
> }
> 
> Is such a transformation best done as a (target-specific) DAG combine?
> 
> A similar instcombine already exists for the X/C->X*1/C case (see the
CvtFDivConstToReciprocal function in InstCombineMlDivRem.cpp), but I don't
believe the above can be done as an instcombine as it creates a new instruction
(in addition to replacing the original).  Also, I only want to perform the
transformation if there are multiple uses of 1/Y (like in my test case). 
Otherwise, the transformation replaces a fdiv with a fdiv+fmul pair, which I
doubt would be profitable.
> 
> FWIW, I'm also pretty sure this combine requires -fast-math.
> 
> Can someone point me in the right direction?
Isn't this http://llvm.org/bugs/show_bug.cgi?id=16218 ? I posted a patch
there but never had the time to finish it.

- Ben
> 
>  Thanks,
>   Chad
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev

Chad Rosier

2013-Aug-08 17:12 UTC

head link

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

It is the very same bug.

On Thu, Aug 8, 2013 at 1:05 PM, Benjamin Kramer <benny.kra at gmail.com>
wrote:
>
> On 08.08.2013, at 18:25, Chad Rosier <chad.rosier at gmail.com>
wrote:
>
> > I would like to transform X/Y -> X*1/Y.  Specifically, I would like
to
> convert:
> >
> > define void @t1a(double %a, double %b, double %d) {
> > entry:
> >   %div = fdiv fast double %a, %d
> >   %div1 = fdiv fast double %b, %d
> >   %call = tail call i32 @foo(double %div, double %div1)
> >   ret void
> > }
> >
> > to:
> >
> > define void @t1b(double %a, double %b, double %d) {
> > entry:
> >   %div = fdiv fast double 1.000000e+00, %d
> >   %mul = fmul fast double %div, %a
> >   %mul1 = fmul fast double %div, %b
> >   %call = tail call i32 @foo(double %mul, double %mul1)
> >   ret void
> > }
> >
> > Is such a transformation best done as a (target-specific) DAG combine?
> >
> > A similar instcombine already exists for the X/C->X*1/C case (see
the
> CvtFDivConstToReciprocal function in InstCombineMlDivRem.cpp), but I
don't
> believe the above can be done as an instcombine as it creates a new
> instruction (in addition to replacing the original).  Also, I only want to
> perform the transformation if there are multiple uses of 1/Y (like in my
> test case).  Otherwise, the transformation replaces a fdiv with a fdiv+fmul
> pair, which I doubt would be profitable.
> >
> > FWIW, I'm also pretty sure this combine requires -fast-math.
> >
> > Can someone point me in the right direction?
>
> Isn't this http://llvm.org/bugs/show_bug.cgi?id=16218 ? I posted a
patch
> there but never had the time to finish it.
>
> - Ben
>
> >
> >  Thanks,
> >   Chad
> > _______________________________________________
> > LLVM Developers mailing list
> > LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>
>-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130808/414f75ab/attachment.html>

Shuxin Yang

2013-Aug-08 17:13 UTC

head link

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

I did few transformation in Instruction *InstCombiner::visitFDiv() in an 
attempt to remove some divs.
I may miss this case.  If you need to implement this rule, it is better 
done in Instcombine than in DAG combine.
Doing such xform early expose the redundancy of 1/y, which have positive 
impact to neighboring code,
while DAG combine is bit blind.

You should be very careful, reciprocal is very very very imprecise 
transformation. Sometimes you will see big different
with and without this xform.

On 8/8/13 9:25 AM, Chad Rosier wrote:> I would like to transform X/Y -> X*1/Y. Specifically, I would like to 
> convert:
>
> define void @t1a(double %a, double %b, double %d) {
> entry:
>   %div = fdiv fast double %a, %d
>   %div1 = fdiv fast double %b, %d
>   %call = tail call i32 @foo(double %div, double %div1)
>   ret void
> }
>
> to:
>
> define void @t1b(double %a, double %b, double %d) {
> entry:
>   %div = fdiv fast double 1.000000e+00, %d
>   %mul = fmul fast double %div, %a
>   %mul1 = fmul fast double %div, %b
>   %call = tail call i32 @foo(double %mul, double %mul1)
>   ret void
> }
>
> Is such a transformation best done as a (target-specific) DAG combine?
>
> A similar instcombine already exists for the X/C->X*1/C case (see the 
> CvtFDivConstToReciprocal function in InstCombineMlDivRem.cpp), but I 
> don't believe the above can be done as an instcombine as it creates a 
> new instruction (in addition to replacing the original).  Also, I only 
> want to perform the transformation if there are multiple uses of 1/Y 
> (like in my test case).  Otherwise, the transformation replaces a fdiv 
> with a fdiv+fmul pair, which I doubt would be profitable.
>
> FWIW, I'm also pretty sure this combine requires -fast-math.
>
> Can someone point me in the right direction?
>
>  Thanks,
>   Chad
>
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130808/169bb9a6/attachment.html>

Shuxin Yang

2013-Aug-08 17:21 UTC

head link

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

I remember why I didn't implement this rule in Instcombine. It add one 
instruction. So,
this xform should be driven by a redundancy eliminator if you care code 
size.

On 8/8/13 10:13 AM, Shuxin Yang wrote:> I did few transformation in Instruction *InstCombiner::visitFDiv() in 
> an attempt to remove some divs.
> I may miss this case.  If you need to implement this rule, it is 
> better done in Instcombine than in DAG combine.
> Doing such xform early expose the redundancy of 1/y, which have 
> positive impact to neighboring code,
> while DAG combine is bit blind.
>
> You should be very careful, reciprocal is very very very imprecise 
> transformation. Sometimes you will see big different
> with and without this xform.
>
> On 8/8/13 9:25 AM, Chad Rosier wrote:
>> I would like to transform X/Y -> X*1/Y. Specifically, I would like
to
>> convert:
>>
>> define void @t1a(double %a, double %b, double %d) {
>> entry:
>>   %div = fdiv fast double %a, %d
>>   %div1 = fdiv fast double %b, %d
>>   %call = tail call i32 @foo(double %div, double %div1)
>>   ret void
>> }
>>
>> to:
>>
>> define void @t1b(double %a, double %b, double %d) {
>> entry:
>>   %div = fdiv fast double 1.000000e+00, %d
>>   %mul = fmul fast double %div, %a
>>   %mul1 = fmul fast double %div, %b
>>   %call = tail call i32 @foo(double %mul, double %mul1)
>>   ret void
>> }
>>
>> Is such a transformation best done as a (target-specific) DAG combine?
>>
>> A similar instcombine already exists for the X/C->X*1/C case (see
the
>> CvtFDivConstToReciprocal function in InstCombineMlDivRem.cpp), but I 
>> don't believe the above can be done as an instcombine as it creates
a
>> new instruction (in addition to replacing the original).  Also, I 
>> only want to perform the transformation if there are multiple uses of 
>> 1/Y (like in my test case).  Otherwise, the transformation replaces a 
>> fdiv with a fdiv+fmul pair, which I doubt would be profitable.
>>
>> FWIW, I'm also pretty sure this combine requires -fast-math.
>>
>> Can someone point me in the right direction?
>>
>>  Thanks,
>>   Chad
>>
>>
>> _______________________________________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu          http://llvm.cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130808/fb576dd0/attachment.html>

Chad Rosier

2013-Aug-08 17:24 UTC

head link

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

I believe the advice that Owen gave me was that if it's simple enough, we
should do it in both places. (Feel free to chime in if I'm misquoting you,
Owen.)

Performing the transformation early as an InstCombine can expose additional
opportunities to optimize code.  The purpose of the redundant DAGCombine
would be to catch the case where the optimization is exposed by another DAG
combine.  And as Jim also pointed out, the DAG combine has the benefit of
target-specific knowledge.

I agree, this will require quite a bit of testing.

 Chad

On Thu, Aug 8, 2013 at 1:13 PM, Shuxin Yang <shuxin.llvm at gmail.com>
wrote:
>  I did few transformation in Instruction *InstCombiner::visitFDiv() in an
> attempt to remove some divs.
> I may miss this case.  If you need to implement this rule, it is better
> done in Instcombine than in DAG combine.
> Doing such xform early expose the redundancy of 1/y, which have positive
> impact to neighboring code,
> while DAG combine is bit blind.
>
> You should be very careful, reciprocal is very very very imprecise
> transformation. Sometimes you will see big different
> with and without this xform.
>
> On 8/8/13 9:25 AM, Chad Rosier wrote:
>
> I would like to transform X/Y -> X*1/Y.  Specifically, I would like to
> convert:
>
> define void @t1a(double %a, double %b, double %d) {
> entry:
>   %div = fdiv fast double %a, %d
>   %div1 = fdiv fast double %b, %d
>   %call = tail call i32 @foo(double %div, double %div1)
>   ret void
> }
>
> to:
>
> define void @t1b(double %a, double %b, double %d) {
> entry:
>   %div = fdiv fast double 1.000000e+00, %d
>   %mul = fmul fast double %div, %a
>   %mul1 = fmul fast double %div, %b
>   %call = tail call i32 @foo(double %mul, double %mul1)
>   ret void
> }
>
> Is such a transformation best done as a (target-specific) DAG combine?
>
> A similar instcombine already exists for the X/C->X*1/C case (see the
> CvtFDivConstToReciprocal function in InstCombineMlDivRem.cpp), but I
don't
> believe the above can be done as an instcombine as it creates a new
> instruction (in addition to replacing the original).  Also, I only want to
> perform the transformation if there are multiple uses of 1/Y (like in my
> test case).  Otherwise, the transformation replaces a fdiv with a fdiv+fmul
> pair, which I doubt would be profitable.
>
> FWIW, I'm also pretty sure this combine requires -fast-math.
>
> Can someone point me in the right direction?
>
>  Thanks,
>   Chad
>
>
> _______________________________________________
> LLVM Developers mailing listLLVMdev at cs.uiuc.edu        
http://llvm.cs.uiuc.eduhttp://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>
>
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>
>-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130808/3a1254c2/attachment.html>

Chad Rosier

2013-Aug-08 17:31 UTC

head link

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

Thanks for the input, Jim.  I'm kinda torn because I have the same
preference as you (i.e., a DAG combine), but Ben seems to have already
proposed a good solution as an InstCombine with a little cleanup.  Perhaps,
I'll implement both and then do some analysis. :D

 Chad

On Thu, Aug 8, 2013 at 12:56 PM, Jim Grosbach <grosbach at apple.com>
wrote:
> Hi Chad,
>
> This is a great transform to do, but you’re right that it’s only safe
> under fast-math. This is particularly interesting when the original divisor
> is a constant so you can materialize the reciprocal at compile-time. You’re
> right that in either case, this optimization should only kick in when there
> is more than one divide instruction that will be changed to a mul.
>
> I don’t have a strong preference for instcombine vs. dagcombine, though I
> lean slightly towards later when we’ll have more target information
> available if we want to apply a more complicated cost function for some
> targets.
>
> -Jim
>
>
> On Aug 8, 2013, at 9:25 AM, Chad Rosier <chad.rosier at gmail.com>
wrote:
>
> I would like to transform X/Y -> X*1/Y.  Specifically, I would like to
> convert:
>
> define void @t1a(double %a, double %b, double %d) {
> entry:
>   %div = fdiv fast double %a, %d
>   %div1 = fdiv fast double %b, %d
>   %call = tail call i32 @foo(double %div, double %div1)
>   ret void
> }
>
> to:
>
> define void @t1b(double %a, double %b, double %d) {
> entry:
>   %div = fdiv fast double 1.000000e+00, %d
>   %mul = fmul fast double %div, %a
>   %mul1 = fmul fast double %div, %b
>   %call = tail call i32 @foo(double %mul, double %mul1)
>   ret void
> }
>
> Is such a transformation best done as a (target-specific) DAG combine?
>
> A similar instcombine already exists for the X/C->X*1/C case (see the
> CvtFDivConstToReciprocal function in InstCombineMlDivRem.cpp), but I
don't
> believe the above can be done as an instcombine as it creates a new
> instruction (in addition to replacing the original).  Also, I only want to
> perform the transformation if there are multiple uses of 1/Y (like in my
> test case).  Otherwise, the transformation replaces a fdiv with a fdiv+fmul
> pair, which I doubt would be profitable.
>
> FWIW, I'm also pretty sure this combine requires -fast-math.
>
> Can someone point me in the right direction?
>
>  Thanks,
>   Chad
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>
>
>-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130808/99751415/attachment.html>

Joerg Sonnenberger

2013-Aug-08 17:32 UTC

head link

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

On Thu, Aug 08, 2013 at 10:13:07AM -0700, Shuxin Yang
wrote:> You should be very careful, reciprocal is very very very imprecise
> transformation. Sometimes you will see big different
> with and without this xform.
What about constant %a, %b in this case? At least for power-of-two
arguments, wouldn't it still be precise?

Joerg

Mark Lacey

2013-Aug-08 17:56 UTC

head link

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

On Aug 8, 2013, at 9:56 AM, Jim Grosbach <grosbach at apple.com> wrote:
> Hi Chad,
> 
> This is a great transform to do, but you’re right that it’s only safe under
fast-math. This is particularly interesting when the original divisor is a
constant so you can materialize the reciprocal at compile-time. You’re right
that in either case, this optimization should only kick in when there is more
than one divide instruction that will be changed to a mul.
It can be worthwhile to do this even in the case where there is only a single
divide since 1/Y might be loop invariant, and could then be hoisted out later by
LICM. You just need to be able to fold it back together when there is only a
single use, and that use is not inside a more deeply nested loop.
> 
> I don’t have a strong preference for instcombine vs. dagcombine, though I
lean slightly towards later when we’ll have more target information available if
we want to apply a more complicated cost function for some targets.
> 
> -Jim
> 
> 
> On Aug 8, 2013, at 9:25 AM, Chad Rosier <chad.rosier at gmail.com>
wrote:
> 
>> I would like to transform X/Y -> X*1/Y.  Specifically, I would like
to convert:
>> 
>> define void @t1a(double %a, double %b, double %d) {
>> entry:
>>   %div = fdiv fast double %a, %d
>>   %div1 = fdiv fast double %b, %d
>>   %call = tail call i32 @foo(double %div, double %div1)
>>   ret void
>> }
>> 
>> to:
>> 
>> define void @t1b(double %a, double %b, double %d) {
>> entry:
>>   %div = fdiv fast double 1.000000e+00, %d
>>   %mul = fmul fast double %div, %a
>>   %mul1 = fmul fast double %div, %b
>>   %call = tail call i32 @foo(double %mul, double %mul1)
>>   ret void
>> }
>> 
>> Is such a transformation best done as a (target-specific) DAG combine?
>> 
>> A similar instcombine already exists for the X/C->X*1/C case (see
the CvtFDivConstToReciprocal function in InstCombineMlDivRem.cpp), but I
don't believe the above can be done as an instcombine as it creates a new
instruction (in addition to replacing the original).  Also, I only want to
perform the transformation if there are multiple uses of 1/Y (like in my test
case).  Otherwise, the transformation replaces a fdiv with a fdiv+fmul pair,
which I doubt would be profitable.
>> 
>> FWIW, I'm also pretty sure this combine requires -fast-math.
>> 
>> Can someone point me in the right direction?
>> 
>>  Thanks,
>>   Chad
>> _______________________________________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
> 
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130808/6c2c3834/attachment.html>

Reed Kotler

2013-Aug-08 21:35 UTC

head link

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

Seems incorrect but I forget the IEEE fp rules.

What if both x and y are infinity?

in that case x/y = NAN but you transformation will yield 0 as the result.

On 08/08/2013 09:25 AM, Chad Rosier wrote:> I would like to transform X/Y -> X*1/Y.  Specifically, I would like to
> convert:
>
> define void @t1a(double %a, double %b, double %d) {
> entry:
>    %div = fdiv fast double %a, %d
>    %div1 = fdiv fast double %b, %d
>    %call = tail call i32 @foo(double %div, double %div1)
>    ret void
> }
>
> to:
>
> define void @t1b(double %a, double %b, double %d) {
> entry:
>    %div = fdiv fast double 1.000000e+00, %d
>    %mul = fmul fast double %div, %a
>    %mul1 = fmul fast double %div, %b
>    %call = tail call i32 @foo(double %mul, double %mul1)
>    ret void
> }
>
> Is such a transformation best done as a (target-specific) DAG combine?
>
> A similar instcombine already exists for the X/C->X*1/C case (see the
> CvtFDivConstToReciprocal function in InstCombineMlDivRem.cpp), but I
> don't believe the above can be done as an instcombine as it creates a
> new instruction (in addition to replacing the original).  Also, I only
> want to perform the transformation if there are multiple uses of 1/Y
> (like in my test case).  Otherwise, the transformation replaces a fdiv
> with a fdiv+fmul pair, which I doubt would be profitable.
>
> FWIW, I'm also pretty sure this combine requires -fast-math.
>
> Can someone point me in the right direction?
>
>   Thanks,
>    Chad
>
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>

Chandler Carruth

2013-Aug-08 21:44 UTC

head link

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

On Thu, Aug 8, 2013 at 2:35 PM, Reed Kotler <rkotler at mips.com> wrote:
> Seems incorrect but I forget the IEEE fp rules.
>
> What if both x and y are infinity?
>
> in that case x/y = NAN but you transformation will yield 0 as the result.
>
This is being proposed for -ffast-math.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130808/a17d9892/attachment.html>

Maybe Matching Threads

Search for more seemingly similar threads

llvm dev - Aug 2013 - [LLVMdev] Convert fdiv - X/Y -> X*1/Y

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

[LLVMdev] Convert fdiv - X/Y -> X*1/Y

Maybe Matching Threads