[polly] r258660 - BlockGenerators: Avoid redundant map lookup [NFC]
David Blaikie via llvm-commits
llvm-commits at lists.llvm.org
Mon Jan 25 13:05:14 PST 2016
On Mon, Jan 25, 2016 at 12:01 PM, Tobias Grosser <tobias at grosser.es> wrote:
> On 01/25/2016 08:52 PM, David Blaikie wrote:
>> On Sun, Jan 24, 2016 at 6:17 AM, Tobias Grosser via llvm-commits
>> <llvm-commits at lists.llvm.org <mailto:llvm-commits at lists.llvm.org>> wrote:
>> Author: grosser
>> Date: Sun Jan 24 08:16:59 2016
>> New Revision: 258660
>> URL: http://llvm.org/viewvc/llvm-project?rev=258660&view=rev
>> BlockGenerators: Avoid redundant map lookup [NFC]
>> Modified: polly/trunk/lib/CodeGen/BlockGenerators.cpp
>> --- polly/trunk/lib/CodeGen/BlockGenerators.cpp (original)
>> +++ polly/trunk/lib/CodeGen/BlockGenerators.cpp Sun Jan 24 08:16:59
>> @@ -338,8 +338,8 @@ Value *BlockGenerator::getOrCreateAlloca
>> auto Addr = Map[ScalarBase];
>> - if (GlobalMap.count(Addr))
>> - return GlobalMap[Addr];
>> + if (auto NewAddr = GlobalMap.lookup(Addr))
>> Probably best to include the '*' in auto here (if I'm understanding the
>> code correctly, auto is binding to Instruction* here) to make it clear
>> that it's a pointer (& that this if test is a null pointer test)?
> Thanks David, this is an AssertingVH, not a plain pointer.
Ah, right, I see now.
> Hence, it is
> is not possible to write it this way. However, the conversion to null
> still seems to work as expected. I have no idea how to write it better.
> Any suggestions?
Nothing great that comes to mind - could write out the type rather than
auto, but I doubt that'd add much. No worries
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the llvm-commits