LLVM Bugzilla is read-only and represents the historical archive of all LLVM issues filled before November 26, 2021. Use github to submit LLVM bugs

Bug 41749 - In function: __mulxc3 Cannot select: 0x55d58830cbf0: f80 = ConstantFP<APFloat(604462909807314587353088)>
Summary: In function: __mulxc3 Cannot select: 0x55d58830cbf0: f80 = ConstantFP<APFloat...
Status: NEW
Alias: None
Product: libraries
Classification: Unclassified
Component: Backend: X86 (show other bugs)
Version: 8.0
Hardware: PC Linux
: P enhancement
Assignee: Unassigned LLVM Bugs
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2019-05-05 05:09 PDT by Constantin Runge
Modified: 2019-05-05 11:59 PDT (History)
5 users (show)

See Also:
Fixed By Commit(s):

Attachments
bugpoint-reduced-simplified.bc (1.41 KB, application/octet-stream)
2019-05-05 05:09 PDT, Constantin Runge 		Details
Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Constantin Runge 2019-05-05 05:09:43 PDT 
Created attachment 21896 [details]
bugpoint-reduced-simplified.bc

Clang 7 as well as Clang 8 ran into errors while compiling the respective compiler-rt.

Tracing the bug led me to a crashing llc:


    > llc -asm-show-inst -asm-verbose bugpoint-reduced-simplified.bc
    LLVM ERROR: Cannot select: 0x55d58830cbf0: f80 = ConstantFP<APFloat(604462909807314587353088)>
    In function: __mulxc3

The .bc file is attached.
Comment 1 Simon Pilgrim 2019-05-05 05:13:28 PDT 
; ModuleID = 'bugpoint-reduced-simplified.bc'
source_filename = "mulxc3-1a6977-min.i"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-pc-linux-gnu"

define dso_local void @__mulxc3() #0 {
  %1 = call x86_fp80 @llvm.copysign.f80(x86_fp80 0xK00000000000000000000, x86_fp80 undef)
  store x86_fp80 %1, x86_fp80* undef, align 16
  ret void
}

; Function Attrs: nounwind readnone speculatable
declare x86_fp80 @llvm.copysign.f80(x86_fp80, x86_fp80) #1

attributes #0 = { "use-soft-float"="false" }
attributes #1 = { nounwind readnone speculatable }

!llvm.ident = !{!0}

!0 = !{!"clang version 8.0.0 (tags/RELEASE_800/final)"}
Comment 2 Craig Topper 2019-05-05 11:01:28 PDT 
I'm not able to reproduce this. I've tried trunk and building from the 8.0 source. 

This is the pattern in X86InstrFPStack.td that this should correspond to
def : Pat<(f80 fpimmneg0), (CHS_Fp80 (LD_Fp080))>;

This line hasn't been edited in years.

Simon, did you reproduce this or just disassemble the bugpoint bitcode?
Comment 3 Simon Pilgrim 2019-05-05 11:12:56 PDT 
(In reply to Craig Topper from comment #2)
> Simon, did you reproduce this or just disassemble the bugpoint bitcode?

I just dissembled it - I can't repro it either (from the bc or the ir) but I forgot to mention that when I pasted the IR.

Constantin - please can you give us some more context to your build?
Comment 4 Constantin Runge 2019-05-05 11:17:13 PDT 
> Constantin - please can you give us some more context to your build?

Hmm.. Not sure what you need/what helps you. It's a Gentoo Linux machine with a Haswell CPU.

I can see if playing around with some optimization flags during clang/llvm build changes the behavior.
Comment 5 Constantin Runge 2019-05-05 11:59:55 PDT 
Ok, compiling llvm and clang with

    CFLAGS="-pipe -march=native -O2"
    CXXFLAGS="${CFLAGS}"
    LDFLAGS="-Wl,-O1 -Wl,--as-needed"

works (ie is not reproducible). Whereas with

    CFLAGS="-pipe -march=native -O3 -fomit-frame-pointer -fstack-protector-strong -ftree-vectorize -fassociative-math -fno-signed-zeros -fno-trapping-math -freciprocal-math -floop-parallelize-all -fgraphite-identity -floop-nest-optimize -flto=8"
    CXXFLAGS="-pipe -march=native -O3 -fomit-frame-pointer -fstack-protector-strong -ftree-vectorize -fassociative-math -fno-signed-zeros -fno-trapping-math -freciprocal-math -floop-parallelize-all -fgraphite-identity -floop-nest-optimize -flto=8 -fvisibility-inlines-hidden"
    LDFLAGS="-pipe -march=native -O3 -fomit-frame-pointer -fstack-protector-strong -ftree-vectorize -fassociative-math -fno-signed-zeros -fno-trapping-math -freciprocal-math -floop-parallelize-all -fgraphite-identity -floop-nest-optimize -flto=8 -Wl,-O1,--as-needed,--hash-style=gnu,--sort-common,-z,relro,-z,now"

I get the bug.

Probably I can isolate the flag.