[llvm-commits] [llvm] r134641 - in /llvm/trunk: lib/Target/X86/MCTargetDesc/X86MCTargetDesc.cpp lib/Target/X86/MCTargetDesc/X86MCTargetDesc.h lib/Target/X86/X86.td lib/Target/X86/X86Subtarget.cpp lib/Target/X86/X86Subtarget.h lib/Target/X86/X86Ta

[Evan Cheng]

On Jul 8, 2011, at 2:50 PM, Eli Friedman wrote:

> On Fri, Jul 8, 2011 at 2:47 PM, Chandler Carruth <chandlerc at google.com> wrote:
>> On Fri, Jul 8, 2011 at 2:35 PM, Eli Friedman <eli.friedman at gmail.com> wrote:
>>> 
>>> On Fri, Jul 8, 2011 at 2:15 PM, Evan Cheng <evan.cheng at apple.com> wrote:
>>>> r134741 should handle the non-x86 hosts. For the other cases, we should
>>>> either fix tests or add specific CPUs to X86.td.
>>> 
>>> r134741 does't help at all.  Consider the case where we're compiling
>>> for a triple like x86_64-apple-macosx10.7.0, no CPU is specified, and
>>> we're on ARM.  We end up in 64-bit mode, but with cmov and SSE
>>> disabled.  Stuff explodes.
>> 
>> I think I'm seeing this as well, 96 tests failing with: LLVM ERROR: SSE
>> register return with SSE disabled
>> FWIW, I'm running these tests on a recent Intel processor host (westmere I
>> think...)
> 
> Yes... it explodes there as well because we misdetect the host CPU
> type as "i686".
> 
> That said, we shouldn't be using sys::getHostCPUName() in
> X86Subtarget::X86Subtarget in the first place.

The code was broken before, it just wasn't triggered if no attributes are specified. It's no better if we default to "generic". I'm trying to figure a way to unbreak this.

Evan

> 
> -Eli