[PATCH] D13771: [AArch64]Add support for converting halfword loads into a 32-bit word load

[Chad Rosier via llvm-commits]

James,
Unfortunately, we're not setup to conduct performance testing using the llvm test-suite.  Currently, we can only perform correctness testing, but we hope to have this resolved in the coming weeks.

 Chad

-----Original Message-----
From: Jun Bum Lim [mailto:junbuml at codeaurora.org] 
Sent: Monday, October 19, 2015 4:09 PM
To: junbuml at codeaurora.org; mzolotukhin at apple.com; ahmed.bougacha at gmail.com; james.molloy at arm.com; mcrosier at codeaurora.org
Cc: gberry at codeaurora.org; amara.emerson at arm.com; renato.golin at linaro.org; llvm-commits at lists.llvm.org
Subject: Re: [PATCH] D13771: [AArch64]Add support for converting halfword loads into a 32-bit word load

junbuml added a comment.

I tested it on A57 for spec2000 and spec2006. This patch was applied pretty widely, but clear performance was only observed in spce2006/h264ref (about 2%), and no performance regression was found. I can add a flag to turn it on only for a specific architecture. Please let me know any suggestion.


http://reviews.llvm.org/D13771