[llvm-commits] [llvm] r79484 - in /llvm/trunk: lib/Target/X86/AsmPrinter/X86ATTAsmPrinter.cpp test/CodeGen/X86/2009-08-19-inline-asm-a-modifier.ll

[Dale Johannesen]

On Aug 19, 2009, at 8:16 PMPDT, Chris Lattner wrote:

>
> On Aug 19, 2009, at 5:40 PM, Dale Johannesen wrote:
>
>>>
>>> If regalloc picks a register other than eax, it's a bug that would  
>>> be good to catch because that would be increasing number of used  
>>> registers.
>>
>> No, using ecx or edx would not make the code quality any worse.
>
> They would clobber an extra register.  If/when we start doing  
> interprocedural regalloc, that is bad.  Forcing the test to check  
> for eax does not make the test any more fragile.

Yes it does.  The choice of register has nothing to do with what the  
test is testing.

The test is run twice, and one run is appropriate for FileCheck.  It  
occurred to me a mix might work, and it seems to, so I've checked that  
in.

You may be looking at this the wrong way; using FileCheck shouldn't be  
a goal in itself.  If it's adequate to express what you're trying to  
test for, there are good reasons to use it, but here that's not the  
case.