[llvm-bugs] [Bug 36061] New: miscompiles llvm.powi.f64 libcall on mips64

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=36061

            Bug ID: 36061
           Summary: miscompiles llvm.powi.f64 libcall on mips64
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: Backend: MIPS
          Assignee: unassignedbugs at nondot.org
          Reporter: jcowgill+llvm at jcowgill.uk
                CC: llvm-bugs at lists.llvm.org

This code:

declare double @llvm.powi.f64(double, i32)

define double @powi(double %value, i32 %power) {
        %1 = tail call double @llvm.powi.f64(double %value, i32 %power)
        ret double %1
}

Compiled with "llc -mtriple mips64el-unknown-linux-gnuabi64 -mcpu mips64r2
test.ll" gives this assembly exerpt:

# %bb.0:
        daddiu  $sp, $sp, -16
        .cfi_def_cfa_offset 16
        sd      $ra, 8($sp)             # 8-byte Folded Spill
        .cfi_offset 31, -8
        jal     __powidf2
        dext    $5, $5, 0, 32
        ld      $ra, 8($sp)             # 8-byte Folded Reload
        jr      $ra
        daddiu  $sp, $sp, 16

This is incorrect. The dext instruction causes $a1 to be _zero_ extended from
32 to 64 bits, however since __powidf2 is a C function it must be called using
the C ABI which requires 32-bit integers to be sign extended.

This causes all powi calls with negative non-constant powers to give bogus
results.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20180123/1151d740/attachment.html>