llvm dev - Dec 2015 - trouble hoisting GlobalValues

Hello LLVM,
To reduce the code-size cost of relocations, I'm trying to hoist
GlobalValues that are used many times.  A new pass hides each hoisted
GV behind a BITCAST in the dominating BB.  The pass then updates users
with the output of the BITCAST.  This pass works properly AFAICT.

The problems come in instruction selection.
SelectionDAGBuilder::visitBitCast() treats the BITCAST as a no-op and
eventually users of the GV fold the relocation as if hoisting never
happened.  Experiments turning the BITCAST into a copytoreg failed and
anyway I feel like I'm just shooting in the dark.

Can anyone can suggest a strategy for lowering a hoisted GV?  The end
result should be that the GV materializes as a simple move-immediate
to register.

Thanks,
-steve

On 11 December 2015 at 16:53, Steve King via llvm-dev
<llvm-dev at lists.llvm.org> wrote:
> Hello LLVM,
> To reduce the code-size cost of relocations, I'm trying to hoist
> GlobalValues that are used many times.  A new pass hides each hoisted
> GV behind a BITCAST in the dominating BB.  The pass then updates users
> with the output of the BITCAST.  This pass works properly AFAICT.
>
> The problems come in instruction selection.
> SelectionDAGBuilder::visitBitCast() treats the BITCAST as a no-op and
> eventually users of the GV fold the relocation as if hoisting never
> happened.  Experiments turning the BITCAST into a copytoreg failed and
> anyway I feel like I'm just shooting in the dark.
>
> Can anyone can suggest a strategy for lowering a hoisted GV?  The end
> result should be that the GV materializes as a simple move-immediate
> to register.
Nothing simple. One possibility would be to make constants explicitly
materialized. Something like:

https://github.com/apple/swift/blob/master/docs/SIL.rst#global-addr

Which would also be useful for architectures where some constants are
expensive to place in a register.

Cheers,
Rafael

----- Original Message -----
> From: "Rafael Espíndola via llvm-dev" <llvm-dev at
lists.llvm.org>
> To: "Steve King" <steve at metrokings.com>
> Cc: "llvm-dev" <llvm-dev at lists.llvm.org>
> Sent: Friday, December 11, 2015 4:28:33 PM
> Subject: Re: [llvm-dev] trouble hoisting GlobalValues
> 
> On 11 December 2015 at 16:53, Steve King via llvm-dev
> <llvm-dev at lists.llvm.org> wrote:
> > Hello LLVM,
> > To reduce the code-size cost of relocations, I'm trying to hoist
> > GlobalValues that are used many times.  A new pass hides each
> > hoisted
> > GV behind a BITCAST in the dominating BB.  The pass then updates
> > users
> > with the output of the BITCAST.  This pass works properly AFAICT.
> >
> > The problems come in instruction selection.
> > SelectionDAGBuilder::visitBitCast() treats the BITCAST as a no-op
> > and
> > eventually users of the GV fold the relocation as if hoisting never
> > happened.  Experiments turning the BITCAST into a copytoreg failed
> > and
> > anyway I feel like I'm just shooting in the dark.
> >
> > Can anyone can suggest a strategy for lowering a hoisted GV?  The
> > end
> > result should be that the GV materializes as a simple
> > move-immediate
> > to register.
Also, something similar-sounding to this is done in
lib/Transforms/Scalar/ConstantHoisting.cpp

 -Hal
> 
> Nothing simple. One possibility would be to make constants explicitly
> materialized. Something like:
> 
> https://github.com/apple/swift/blob/master/docs/SIL.rst#global-addr
> 
> Which would also be useful for architectures where some constants are
> expensive to place in a register.
> 
> Cheers,
> Rafael
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
> 
-- 
Hal Finkel
Assistant Computational Scientist
Leadership Computing Facility
Argonne National Laboratory

I thought Zia Ansari did something like what you want for constants to reduce
code size.  It is probably extendable for what you seem to be trying to
do. Back in something like mid-july or so.

Zia - ?

Kevin Smith
>-----Original Message-----
>From: llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org] On Behalf Of
Steve
>King via llvm-dev
>Sent: Friday, December 11, 2015 1:53 PM
>To: llvm-dev <llvm-dev at lists.llvm.org>
>Subject: [llvm-dev] trouble hoisting GlobalValues
>
>Hello LLVM,
>To reduce the code-size cost of relocations, I'm trying to hoist
>GlobalValues that are used many times.  A new pass hides each hoisted
>GV behind a BITCAST in the dominating BB.  The pass then updates users
>with the output of the BITCAST.  This pass works properly AFAICT.
>
>The problems come in instruction selection.
>SelectionDAGBuilder::visitBitCast() treats the BITCAST as a no-op and
>eventually users of the GV fold the relocation as if hoisting never
>happened.  Experiments turning the BITCAST into a copytoreg failed and
>anyway I feel like I'm just shooting in the dark.
>
>Can anyone can suggest a strategy for lowering a hoisted GV?  The end
>result should be that the GV materializes as a simple move-immediate
>to register.
>
>Thanks,
>-steve
>_______________________________________________
>LLVM Developers mailing list
>llvm-dev at lists.llvm.org
>http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

Hi,

Is it exactly the same question as what we discussed a few days ago in this
thread https://groups.google.com/forum/#!topic/llvm-dev/08EXrCDPboo ?

CC Preston who will be feel less lonely maybe.
— 
Mehdi
> On Dec 11, 2015, at 1:53 PM, Steve King via llvm-dev <llvm-dev at
lists.llvm.org> wrote:
> 
> Hello LLVM,
> To reduce the code-size cost of relocations, I'm trying to hoist
> GlobalValues that are used many times.  A new pass hides each hoisted
> GV behind a BITCAST in the dominating BB.  The pass then updates users
> with the output of the BITCAST.  This pass works properly AFAICT.
> 
> The problems come in instruction selection.
> SelectionDAGBuilder::visitBitCast() treats the BITCAST as a no-op and
> eventually users of the GV fold the relocation as if hoisting never
> happened.  Experiments turning the BITCAST into a copytoreg failed and
> anyway I feel like I'm just shooting in the dark.
> 
> Can anyone can suggest a strategy for lowering a hoisted GV?  The end
> result should be that the GV materializes as a simple move-immediate
> to register.
> 
> Thanks,
> -steve
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev