[PATCH] Don't constant fold bitcast/ptrtoint/inttoptr combinations without datalayout

[Eli Friedman]

Err, scratch that, my mistake.

Not sure why you fixed bitcast+ptrtoint and not inttoptr+bitcast, though.

-Eli

On Sun, Jul 21, 2013 at 4:39 PM, Eli Friedman <eli.friedman at gmail.com> wrote:
> From earlier in the thread:
>
>> I was thinking that for a while, but the original code would only fold if the intermediate integer was at least as big as the pointer size, so this was trying to match that behavior. It could constant fold even if the middle integer size is unknown, but that would change the behavior if the middle integer was smaller (I'm not sure why you would ever want that though).
>
> That applies to folding inttoptr+ptrtoint, but I fail to see how it
> applies to folding bitcast+ptrtoint.
>
> -Eli
>
> On Sun, Jul 21, 2013 at 2:21 AM, Matt Arsenault
> <Matthew.Arsenault at amd.com> wrote:
>>   Allow folding bitcast -> ptrtoint when the address spaces are the same.
>>
>>   Fix inline testcase by making it use casts that can still be folded away.
>>
>> Hi eli.friedman,
>>
>> http://llvm-reviews.chandlerc.com/D1157
>>
>> CHANGE SINCE LAST DIFF
>>   http://llvm-reviews.chandlerc.com/D1157?vs=2846&id=2932#toc
>>
>> Files:
>>   lib/IR/ConstantFold.cpp
>>   lib/IR/Instructions.cpp
>>   test/Transforms/Inline/inline_constprop.ll
>>   unittests/IR/InstructionsTest.cpp