[PATCH] NVTPX: Use MapMetadata() instead of incorrect, custom logic

Colin LeMahieu colinl at codeaurora.org
Wed Jan 14 09:36:17 PST 2015

It looks like this change is making a reference to a datastructure in
TransformUtils and the LLVMBuild doesn't have a reference which is causing
link issues at least on Windows mingw-64

The attached patch should fix it.

-----Original Message-----
From: llvm-commits-bounces at cs.uiuc.edu
[mailto:llvm-commits-bounces at cs.uiuc.edu] On Behalf Of Duncan P. N. Exon
Sent: Tuesday, January 13, 2015 11:18 PM
To: Justin Holewinski
Cc: llvm-commits
Subject: Re: [PATCH] NVTPX: Use MapMetadata() instead of incorrect, custom

r225943 and r225944.  Thanks for the review!

> On 2015 Jan 13, at 18:20, Justin Holewinski <justin.holewinski at gmail.com>
> This looks reasonable to me.  Thanks for updating this!
> On Tue, Jan 13, 2015 at 5:44 PM, Duncan P. N. Exon Smith
<dexonsmith at apple.com> wrote:
> NVTPX has its own `remapMDNode()` which (1) does strange things and
> (2) is bitrotting.
> The first patch removes the strange, invalid logic to replace a global 
> variable declaration (see the actual 
> `DIBuilder::createGlobalVariable()` for what the field *really* does 
> -- ignore the comment) with a ConstantInt representing the address 
> space.  (It'll mangle any other metadata schema that happens to 
> reference global variables, as well.)
> The second patch rewrites the logic to use `MapMetadata()`, so that 
> this code doesn't continue to bitrot.
> These aren't really NFC, but it doesn't cause any test failures.

llvm-commits mailing list
llvm-commits at cs.uiuc.edu
-------------- next part --------------
A non-text attachment was scrubbed...
Name: NVPTX.patch
Type: application/octet-stream
Size: 584 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150114/08c7948d/attachment.obj>

More information about the llvm-commits mailing list