[llvm-commits] [llvm] r66902 - /llvm/trunk/lib/Target/X86/X86FastISel.cpp

[Chris Lattner]

On Mar 13, 2009, at 10:59 AM, Dan Gohman wrote:

>
> On Mar 13, 2009, at 9:36 AM, Chris Lattner wrote:
>
>> Author: lattner
>> Date: Fri Mar 13 11:36:42 2009
>> New Revision: 66902
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=66902&view=rev
>> Log:
>> generalize this code so that fast isel handles integer truncates to
>> i1, which
>> codegen to the same thing as integer truncates to i8 (the top bits  
>> are
>> just undefined).  This implements rdar://6667338
>
> Hi Chris,
>
> X86SelectZExt, at least, currently assumes that the top bits in
> the register holding an i1 value are zero.

How is that safe?  What if the i1 was produced by the non-fast isel?

-Chris