llvm dev - May 2016 - Code which should exit 1 is exiting 0

I have IR at https://ghostbin.com/paste/daxv5
<https://ghostbin.com/paste/daxv5> which is meant to exit 1, but it is
always exiting 0.

I'm using it as a template for checking if two functions @test1 and @test2
are equivalent by checking against the exhaustive possible i16 values. For this
particular example it should be enough to know that for certain i16, @test1 and
@test2 are *not* equal. When an inequality is found, the program should exit 1.
However, it seems like it's always exiting 0. Is there something I'm
doing wrong?

 - CL

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160504/16324809/attachment.html>

> On May 4, 2016, at 5:25 PM, Carlos Liam via llvm-dev <llvm-dev at
lists.llvm.org> wrote:
> 
> I have IR at https://ghostbin.com/paste/daxv5
<https://ghostbin.com/paste/daxv5> which is meant to exit 1, but it is
always exiting 0.
> 
> I'm using it as a template for checking if two functions @test1 and
@test2 are equivalent by checking against the exhaustive possible i16 values.
For this particular example it should be enough to know that for certain i16,
@test1 and @test2 are *not* equal. When an inequality is found, the program
should exit 1. However, it seems like it's always exiting 0. Is there
something I'm doing wrong?
> 
>  - CL
How are you compiling it? For which target?

—escha
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160504/fcadc2ed/attachment.html>

I'm using lli, on x86-64.

 - CL
> On May 4, 2016, at 8:32 PM, escha at apple.com wrote:
> 
> 
>> On May 4, 2016, at 5:25 PM, Carlos Liam via llvm-dev <llvm-dev at
lists.llvm.org> wrote:
>> 
>> I have IR at https://ghostbin.com/paste/daxv5 which is meant to exit 1,
but it is always exiting 0.
>> 
>> I'm using it as a template for checking if two functions @test1 and
@test2 are equivalent by checking against the exhaustive possible i16 values.
For this particular example it should be enough to know that for certain i16,
@test1 and @test2 are *not* equal. When an inequality is found, the program
should exit 1. However, it seems like it's always exiting 0. Is there
something I'm doing wrong?
>> 
>>  - CL
> 
> How are you compiling it? For which target?
> 
> —escha
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160504/4e7bd84f/attachment.html>

On 5/4/16 8:25 PM, Carlos Liam via llvm-dev wrote:
> I have IR at https://ghostbin.com/paste/daxv5 which is meant to exit 
> 1, but it is always exiting 0.
>
1) It would help if you pasted the relevant code into your message. 
Paranoid people like me don't following links in emails.

2) Have you ensured that your program doesn't exercise undefined 
behavior?  If your program has a signed overflow, an out-of-bounds 
error, or some other undefined behavior, the compiler may optimize it in 
surprising and unexpected way.

Regards,

John Criswell
> I'm using it as a template for checking if two functions @test1 and 
> @test2 are equivalent by checking against the exhaustive possible i16 
> values. For this particular example it should be enough to know that 
> for certain i16, @test1 and @test2 are *not* equal. When an inequality 
> is found, the program should exit 1. However, it seems like it's 
> always exiting 0. Is there something I'm doing wrong?
>
>  - CL
>
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

-- 
John Criswell
Assistant Professor
Department of Computer Science, University of Rochester
http://www.cs.rochester.edu/u/criswell

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160504/33d03a55/attachment.html>

I checked the debug log: this is what’s happening.

1. Because f16 isn’t legal, SINT_TO_FP and FP_TO_SINT get promoted to f32.
2. f32 has 24 mantissa bits, which is enough to losslessly represent i16s. So
the conversion gets eliminated in FoldIntToFPToInt.

Part 1) here looks fairly suspicious.

—escha
> On May 4, 2016, at 5:41 PM, John Criswell via llvm-dev <llvm-dev at
lists.llvm.org> wrote:
> 
> On 5/4/16 8:25 PM, Carlos Liam via llvm-dev wrote:
>> I have IR at https://ghostbin.com/paste/daxv5
<https://ghostbin.com/paste/daxv5> which is meant to exit 1, but it is
always exiting 0.
>> 
> 
> 1) It would help if you pasted the relevant code into your message. 
Paranoid people like me don't following links in emails.
> 
> 2) Have you ensured that your program doesn't exercise undefined
behavior?  If your program has a signed overflow, an out-of-bounds error, or
some other undefined behavior, the compiler may optimize it in surprising and
unexpected way.
> 
> Regards,
> 
> John Criswell
> 
>> I'm using it as a template for checking if two functions @test1 and
@test2 are equivalent by checking against the exhaustive possible i16 values.
For this particular example it should be enough to know that for certain i16,
@test1 and @test2 are *not* equal. When an inequality is found, the program
should exit 1. However, it seems like it's always exiting 0. Is there
something I'm doing wrong?
>> 
>>  - CL
>> 
>> 
>> 
>> _______________________________________________
>> 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
<http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev>
> 
> 
> -- 
> John Criswell
> Assistant Professor
> Department of Computer Science, University of Rochester
> http://www.cs.rochester.edu/u/criswell
<http://www.cs.rochester.edu/u/criswell>_______________________________________________
> 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/20160504/bf86b679/attachment.html>

DAG legalization still runs with -O0.

—escha
> On May 4, 2016, at 6:00 PM, Carlos Liam <carlos at aarzee.me> wrote:
> 
> Shouldn't that not happen when I run with -O0?
> 
>  - CL
> 
>> On May 4, 2016, at 8:56 PM, escha at apple.com <mailto:escha at
apple.com> wrote:
>> 
>> I checked the debug log: this is what’s happening.
>> 
>> 1. Because f16 isn’t legal, SINT_TO_FP and FP_TO_SINT get promoted to
f32.
>> 2. f32 has 24 mantissa bits, which is enough to losslessly represent
i16s. So the conversion gets eliminated in FoldIntToFPToInt.
>> 
>> Part 1) here looks fairly suspicious.
>> 
>> —escha
>> 
>>> On May 4, 2016, at 5:41 PM, John Criswell via llvm-dev <llvm-dev
at lists.llvm.org <mailto:llvm-dev at lists.llvm.org>> wrote:
>>> 
>>> On 5/4/16 8:25 PM, Carlos Liam via llvm-dev wrote:
>>>> I have IR at https://ghostbin.com/paste/daxv5
<https://ghostbin.com/paste/daxv5> which is meant to exit 1, but it is
always exiting 0.
>>>> 
>>> 
>>> 1) It would help if you pasted the relevant code into your message.
Paranoid people like me don't following links in emails.
>>> 
>>> 2) Have you ensured that your program doesn't exercise
undefined behavior?  If your program has a signed overflow, an out-of-bounds
error, or some other undefined behavior, the compiler may optimize it in
surprising and unexpected way.
>>> 
>>> Regards,
>>> 
>>> John Criswell
>>> 
>>>> I'm using it as a template for checking if two functions
@test1 and @test2 are equivalent by checking against the exhaustive possible i16
values. For this particular example it should be enough to know that for certain
i16, @test1 and @test2 are *not* equal. When an inequality is found, the program
should exit 1. However, it seems like it's always exiting 0. Is there
something I'm doing wrong?
>>>> 
>>>>  - CL
>>>> 
>>>> 
>>>> 
>>>> _______________________________________________
>>>> 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
<http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev>
>>> 
>>> 
>>> -- 
>>> John Criswell
>>> Assistant Professor
>>> Department of Computer Science, University of Rochester
>>> http://www.cs.rochester.edu/u/criswell
<http://www.cs.rochester.edu/u/criswell>_______________________________________________
>>> 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
<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/20160504/58f6aefa/attachment.html>

I was curious about this for integers.  Looks like we have at least one bug
which is of a similar vein.  Comment out the branch here and the test will pass
on x86, otherwise it fails.

Note, command line I used was: llc int.ll -o int.s --filetype=asm &&
clang++ int.s -o main && ./main; echo $?

@g = constant i8 255
define i32 @main() {
  %ptr = bitcast i8* @g to i7*
  %v7 = load i7, i7* %ptr
;  br label %cmp
; cmp:
  %add7 = add i7 %v7, 1
  %icmp = icmp eq i7 %add7, 0
  br i1 %icmp, label %eq, label %ne
eq:
  ret i32 0
ne:
  ret i32 1
}

Pete
> On May 4, 2016, at 6:04 PM, via llvm-dev <llvm-dev at lists.llvm.org>
wrote:
> 
> DAG legalization still runs with -O0.
> 
> —escha
> 
>> On May 4, 2016, at 6:00 PM, Carlos Liam <carlos at aarzee.me
<mailto:carlos at aarzee.me>> wrote:
>> 
>> Shouldn't that not happen when I run with -O0?
>> 
>>  - CL
>> 
>>> On May 4, 2016, at 8:56 PM, escha at apple.com <mailto:escha at
apple.com> wrote:
>>> 
>>> I checked the debug log: this is what’s happening.
>>> 
>>> 1. Because f16 isn’t legal, SINT_TO_FP and FP_TO_SINT get promoted
to f32.
>>> 2. f32 has 24 mantissa bits, which is enough to losslessly
represent i16s. So the conversion gets eliminated in FoldIntToFPToInt.
>>> 
>>> Part 1) here looks fairly suspicious.
>>> 
>>> —escha
>>> 
>>>> On May 4, 2016, at 5:41 PM, John Criswell via llvm-dev
<llvm-dev at lists.llvm.org <mailto:llvm-dev at lists.llvm.org>>
wrote:
>>>> 
>>>> On 5/4/16 8:25 PM, Carlos Liam via llvm-dev wrote:
>>>>> I have IR at https://ghostbin.com/paste/daxv5
<https://ghostbin.com/paste/daxv5> which is meant to exit 1, but it is
always exiting 0.
>>>>> 
>>>> 
>>>> 1) It would help if you pasted the relevant code into your
message.  Paranoid people like me don't following links in emails.
>>>> 
>>>> 2) Have you ensured that your program doesn't exercise
undefined behavior?  If your program has a signed overflow, an out-of-bounds
error, or some other undefined behavior, the compiler may optimize it in
surprising and unexpected way.
>>>> 
>>>> Regards,
>>>> 
>>>> John Criswell
>>>> 
>>>>> I'm using it as a template for checking if two
functions @test1 and @test2 are equivalent by checking against the exhaustive
possible i16 values. For this particular example it should be enough to know
that for certain i16, @test1 and @test2 are *not* equal. When an inequality is
found, the program should exit 1. However, it seems like it's always exiting
0. Is there something I'm doing wrong?
>>>>> 
>>>>>  - CL
>>>>> 
>>>>> 
>>>>> 
>>>>> _______________________________________________
>>>>> 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
<http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev>
>>>> 
>>>> 
>>>> -- 
>>>> John Criswell
>>>> Assistant Professor
>>>> Department of Computer Science, University of Rochester
>>>> http://www.cs.rochester.edu/u/criswell
<http://www.cs.rochester.edu/u/criswell>_______________________________________________
>>>> 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
<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/20160504/6adb969f/attachment.html>