[PATCH] propagate IR-level fast-math-flags to DAG nodes, disabled by default

[Nick Lewycky]

On 22 June 2015 at 14:09, Sanjay Patel <spatel at rotateright.com> wrote:

> Thanks, Nick.
>
> This is the same bug (convert repeated divisions by the same divisor to
> multiplication by the reciprocal) that should be solved by D9893 (
> http://reviews.llvm.org/D9893 ). After I apply that patch, the output asm
> looks correct to me.
>

OK! Let me know once that lands and I'll re-test it. There were a lot of
failures, so there may be something more left after this particular problem
is resolved.


>
> On Fri, Jun 19, 2015 at 5:53 PM, Nick Lewycky <nlewycky at google.com> wrote:
>
>> On 17 June 2015 at 13:57, Sanjay Patel <spatel at rotateright.com> wrote:
>>
>>>
>>>
>>> On Wed, Jun 17, 2015 at 12:59 AM, Nick Lewycky <nicholas at mxc.ca> wrote:
>>>
>>>> Sanjay Patel wrote:
>>>>
>>>>> Please let me know if I can still take you up on this offer:
>>>>> "Could you add FMF under a flag and then let me test it before turning
>>>>> it on? I can test for compiler crashes in 24 hours, or for miscompiles over
>>>>> a weekend."
>>>>>
>>>>
>>>> Right! Yes, I can do that.
>>>>
>>>
>>> Great! Let me know when you see bugs crawling out from under the rug. :)
>>>
>>
>> Just finished the first creduce.
>>
>> Testcase:
>>
>> int a, b, c;
>> void fn1() {
>>   b = c / (1.0 + c);
>>   a = 1 / (1.0 + c);
>> }
>>
>> Build with "clang -O2 -mllvm -enable-fmf-dag -ffast-math" to get
>>
>>  fatal error: error in backend: Cannot select: 0x2a8a760: f64 =
>>       ConstantFP<1.000000e+00> [ID=7]
>> In function: _Z3fn1v
>>
>> and here it is as .ll:
>>
>> target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
>> target triple = "x86_64-grtev4-linux-gnu"
>>
>> @a = global i32 0, align 4
>> @b = global i32 0, align 4
>> @c = global i32 0, align 4
>>
>> ; Function Attrs: nounwind uwtable
>> define void @_Z3fn1v() #0 {
>>   %1 = load i32, i32* @c, align 4, !tbaa !1
>>   %2 = sitofp i32 %1 to double
>>   %3 = fadd fast double %2, 1.000000e+00
>>   %4 = fdiv fast double %2, %3
>>   %5 = fptosi double %4 to i32
>>   store i32 %5, i32* @b, align 4, !tbaa !1
>>   %6 = fdiv fast double 1.000000e+00, %3
>>   %7 = fptosi double %6 to i32
>>   store i32 %7, i32* @a, align 4, !tbaa !1
>>   ret void
>> }
>>
>> attributes #0 = { nounwind uwtable "disable-tail-calls"="false"
>> "less-precise-fpmad"="false" "no-frame-pointer-elim"="false"
>> "no-infs-fp-math"="true" "no-nans-fp-math"="true"
>> "stack-protector-buffer-size"="8" "target-cpu"="x86-64"
>> "target-features"="+sse,+sse2" "unsafe-fp-math"="true"
>> "use-soft-float"="false" }
>>
>> !llvm.ident = !{!0}
>>
>> !0 = !{!"clang version google3-trunk (trunk r240125)"}
>> !1 = !{!2, !2, i64 0}
>> !2 = !{!"int", !3, i64 0}
>> !3 = !{!"omnipotent char", !4, i64 0}
>> !4 = !{!"Simple C/C++ TBAA"}
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150622/31b3528e/attachment.html>