Performance degradation when targetting v80 from 2012 IDE

Oct 30, 2013 at 3:49 PM
Edited Oct 30, 2013 at 3:51 PM
does anyone else see the runtime performance degradation when targetting using Daffodil?
I am working on a large analytical system, and am targeting v80 from 2012 IDE.
The 2005 project conversion was done in two stages: from v80 to v100 and then from v100 to v110.

When comparing the results against the original 2005 build, the performance has dropped by as much as 10% . I analysed the command lines passed by both toolchains, and they indeed differ both for CL and LINK, albeit subtelly. I started manually bringing them to be more in-line, but the performance drop will just not go away.

Has anyone else experienced (and hopefully solved) this?
All the best,

PS While 2012 platform is not officially supported, it seems to be working fine. I assume that the same performance drop would happen with 2010 toolchain, although I have not tried this.
Oct 30, 2013 at 4:57 PM
I have not seen any noticeable performance degradation, and I don't recall anyone else reporting any performance problems. I can imagine a very slight effect due to the derived build task classes, but it should not be anywhere near 10%. As you surmise, it could be due to subtle differences in build tool options, but with some work you should be able to eliminate any meaningful differences. I'm more inclined to blame background work that the IDE is doing independently while the build is going on, or just simply that the VS2012 IDE process uses more physical memory than VS 2008, thus forcing the build process to page memory to and from the hard drive more frequently. It would be interesting to compare the performance to VS 2010 on an identical system; on the other hand, it's completely plausible that both exhibit the same problem.