[compiler-rt] r248853 - [CMake] [macho_embedded] We need to set some extra flags when building the macho_embedded builtins.

[Hans Wennborg via llvm-commits]

Building with the just-built Clang seems to work. We're building
compiler-rt separately, so I can just point it to the new clang
manually.

Thanks,
Hans

On Mon, Oct 5, 2015 at 3:36 PM, Chris Bieneman <beanz at apple.com> wrote:
> I’m guessing you’re on an old clang. Apple Clang 602 had bugs in its aeabi
> support that were exposed by that change.
>
> I have patches out for review that enable using the just-built clang when
> building compiler-rt that should make this issue go away:
>
> http://reviews.llvm.org/D13399
>
> Alternatively the current Xcode GM doesn’t exhibit this problem either.
>
> -Chris
>
> On Oct 5, 2015, at 3:32 PM, Hans Wennborg <hans at chromium.org> wrote:
>
> On Tue, Sep 29, 2015 at 4:21 PM, Chris Bieneman via llvm-commits
> <llvm-commits at lists.llvm.org> wrote:
>
> Author: cbieneman
> Date: Tue Sep 29 18:21:09 2015
> New Revision: 248853
>
> URL: http://llvm.org/viewvc/llvm-project?rev=248853&view=rev
> Log:
> [CMake] [macho_embedded] We need to set some extra flags when building the
> macho_embedded builtins.
>
> If we don't specify the arm target the float ABI compiler flags get ignored.
>
>
> This is causing problems when building Clang & compiler-rt on Mac for
> Chromium. The 10.10 SDK compiler now crashes like this:
>
> fatal error: error in backend: Cannot select: 0x110008210: i32,i32 =
> udivrem 0x110001e70, 0x10c006928 [ORD=2] [ID=10]
>  0x110001e70: i32,ch = CopyFromReg 0x104008a20, 0x10c006508 [ORD=2] [ID=7]
>    0x10c006508: i32 = Register %vreg1 [ID=1]
>  0x10c006928: i32,ch = CopyFromReg 0x104008a20, 0x110008420 [ORD=2] [ID=8]
>    0x110008420: i32 = Register %vreg3 [ID=2]
> In function: __udivmoddi4
>
> while compiling lib/builtins/udivmoddi4.c.
>
> I suppose this patch made some flags change. Is some bad flag getting
> passed to the compiler, or is it just broken?
>
> I put some more details here: http://crbug.com/539600
>
> Thanks,
> Hans
>
>