incorrect cmpxchg signature

[db7]

Here is a long standing bug we have fixed internally. I can create a PR if desired.

Description

The cmpxchg return value has I1 type, but the package defines it as I8 (see here ). That causes clang to fail to compile generated files that contain cmpxchg.

How to reproduce

Save the following code as input.ll :

; ModuleID = 'tests/cmpxchg.c'
source_filename = "tests/cmpxchg.c"
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-pc-linux-gnu"

@shared = common dso_local global i32 0, align 4
@expected = common dso_local global i32 0, align 4

; Function Attrs: noinline nounwind optnone uwtable
define dso_local i32 @main() #0 {
  %1 = alloca i32, align 4
  %2 = alloca i32, align 4
  %3 = alloca i8, align 1
  store i32 0, i32* %1, align 4
  store i32 1, i32* %2, align 4
  %4 = load i32, i32* @expected, align 4
  %5 = load i32, i32* %2, align 4
  %6 = cmpxchg i32* @shared, i32 %4, i32 %5 seq_cst seq_cst
  %7 = extractvalue { i32, i1 } %6, 0
  %8 = extractvalue { i32, i1 } %6, 1
  br i1 %8, label %10, label %9

9:                                                ; preds = %0
  store i32 %7, i32* @expected, align 4
  br label %10

10:                                               ; preds = %9, %0
  %11 = zext i1 %8 to i8
  store i8 %11, i8* %3, align 1
  %12 = load i8, i8* %3, align 1
  %13 = trunc i8 %12 to i1
  ret i32 0
}

attributes #0 = { noinline nounwind optnone uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }

!llvm.module.flags = !{!0}
!llvm.ident = !{!1}

!0 = !{i32 1, !"wchar_size", i32 4}
!1 = !{!"clang version 10.0.0-4ubuntu1 "}

Now use this small program with any version of package:

package main

import (
        "fmt"
        "os"

        "github.com/llir/llvm/asm"
)

func main() {
        mod, err := asm.ParseFile("input.ll")
        if err != nil {
                panic(err)
        }

        fp, err := os.Create("output.ll")
        if err != nil {
                panic(err)
        }
        defer fp.Close()
        fmt.Fprint(fp, mod)
}

Finally, try to compile the generated file:

$ clang output.ll
output.ll:18:32: error: '%6' defined with type '{ i32, i1 }' but expected '{ i32, i8 }'
        %7 = extractvalue { i32, i8 } %6, 0
                                      ^
1 error generated.

Fix

--- a/vendor/github.com/llir/llvm/asm/inst_memory.go
+++ b/vendor/github.com/llir/llvm/asm/inst_memory.go
@@ -44,7 +44,7 @@ func (fgen *funcGen) newCmpXchgInst(ident ir.LocalIdent, old *ast.CmpXchgInst) (
        if err != nil {
                return nil, errors.WithStack(err)
        }
-       typ := types.NewStruct(oldType, types.I8)
+       typ := types.NewStruct(oldType, types.I1)
        return &ir.InstCmpXchg{LocalIdent: ident, Typ: typ}, nil
 }

[dannypsnl]

Sure, please send PR

[mewmew]

@db7 thanks for the detailed bug report and also submitting the one-line fix! You're indeed correct that the second field of the struct returned by the cmpxchg should be a boolean of type i1.

Of curiosity, in what kind of projects are you using the llir/llvm package? : )

Wish you all the best and happy coding.

Cheers,
Robin

[db7]

Of curiosity, in what kind of projects are you using the llir/llvm package? : )

https://github.com/open-s4c/vsyncer

[mewmew]

Of curiosity, in what kind of projects are you using the llir/llvm package? : )

https://github.com/open-s4c/vsyncer

Hi Diogo,

Really cool project! Thanks for releasing it as open source.

Cheerful regards,
Robin