Here's the RFC of the overflow builtins: http://clang-developers.42468.n3.nabble.com/RFC-Introduce-overflow-builtins-td3838320.html Along with the tracking issue: https://bugs.llvm.org/show_bug.cgi?id=12290 And the patch: https://github.com/llvm-mirror/clang/commit/98d1ec1e99625176626b0bcd44cef7df6e89b289
There's also another patch that was added on top of this one which adds more overflow builtins: https://github.com/llvm-mirror/clang/commit/c41c63fbf84cc904580e733d1123d3b03bb5584c It seems clear that this optimization could bring big performance improvements on hot functions. It could also reduce binary size substantially (we're talking about 14->5 instructions in their case). Tristan On 14/08/2017 15:58, Ben Kelly wrote: > On Mon, Aug 14, 2017 at 6:11 AM, Till Schneidereit < > t...@tillschneidereit.net> wrote: > >> On Mon, Aug 14, 2017 at 9:27 AM, Julian Seward <jsew...@acm.org> wrote: >> >>> On 13/08/17 03:40, Ehsan Akhgari wrote: >>>> As you may have heard by now, Chromium has started to switch their >>> Windows >>>> builds to use clang-cl instead of MSVC [1]. This has improved their >>>> Speedometer v2 benchmark score on x86 (but not on x86-64) by about 30% >>>> according to AWFY [2]. [..] >>> Do we have any insight into why the Clang version is so much faster? >>> 30% strikes me as a large difference for two supposedly mature optimizing >>> compilers. And stranger still that it applies only for the 32-bit case. >>> So I'm curious to know what's changed. >>> >> AFAICT, the real change is about 19%: shortly before the jump to ~103, >> their score regressed from 86 to 78. I think using 86 as the baseline makes >> much more sense. 19% is obviously still a substantial improvement from a >> compiler change. >> > Google ran a bisect on the regression and improvements. Here is some > explanation of the regression: > > https://bugs.chromium.org/p/chromium/issues/detail?id=749359#c4 > > The comment suggests that the change would see an improvement in clang > since the code would then use class "overflow builtins". > > Its hard to tell if all of the clang-on-windows improvement is due to this > same set of code or not. > > The bisects were run from this issue: > > https://bugs.chromium.org/p/chromium/issues/detail?id=750672#c12 > > Ben > _______________________________________________ > dev-platform mailing list > dev-platform@lists.mozilla.org > https://lists.mozilla.org/listinfo/dev-platform _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
