[llvm-bugs] [Bug 25191] New: AArch64 does not implement FLT_ROUNDS

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=25191

            Bug ID: 25191
           Summary: AArch64 does not implement FLT_ROUNDS
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: Backend: AArch64
          Assignee: unassignedbugs at nondot.org
          Reporter: ed at 80386.nl
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

The C99 FLT_ROUNDS macro is supposed to return the current rounding mode of the
processor's floating point unit. Clang's __builtin_flt_rounds() can be used to
implement the FLT_ROUNDS macro.

Consider the following piece of code:

int foo(void) {
  return __builtin_flt_rounds();
}

For x86-64 we build the following piece of code:

0000000000000000 <foo>:
   0:   55                      push   %rbp
   1:   48 89 e5                mov    %rsp,%rbp
   4:   d9 7d f0                fnstcw -0x10(%rbp)
   7:   0f b7 45 f0             movzwl -0x10(%rbp),%eax
   b:   89 c1                   mov    %eax,%ecx
   d:   81 e1 00 04 00 00       and    $0x400,%ecx
  13:   c1 e9 09                shr    $0x9,%ecx
  16:   25 00 08 00 00          and    $0x800,%eax
  1b:   c1 e8 0b                shr    $0xb,%eax
  1e:   09 c8                   or     %ecx,%eax
  20:   ff c0                   inc    %eax
  22:   83 e0 03                and    $0x3,%eax
  25:   5d                      pop    %rbp
  26:   c3                      retq   

Which is correct. Now if we build exactly the same code with Clang targeted
against aarch64, we obtain:

0000000000000000 <foo>:
   0:   320003e0        orr     w0, wzr, #0x1
   4:   d65f03c0        ret

This is obviously not correct. It should return a value based on bits 22:23 of
FPCR.

I tried patching up Clang myself to support this, but I have to confess that I
don't know what I'm doing. I started out copying
ARMTargetLowering::LowerFLT_ROUNDS_() to
AArch64TargetLowering::LowerFLT_ROUNDS_(), but then I have no idea how to fetch
the FPCR register.

-- 
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/20151015/7628c572/attachment.html>