[PATCH] D52193: RFC: [clang] Multithreaded compilation support

Mon Sep 24 07:21:31 PDT 2018

thakis added a comment.

In https://reviews.llvm.org/D52193#1241067, @aganea wrote:

> @thakis > clang-cl isn't supposed to do (explicit) registry accesses when you hold it right (pass in -fms-compatibility-version etc). Have you seen registry access costs, or is that speculation?
>
> Please see this log: F7268226: clang-cl-log.zip <https://reviews.llvm.org/F7268226> - the child `clang-cl -cc1` takes about ~117ms until it gets into the global initializers. This is on my Haswell PC. On the Skylake, this takes "only" ~60ms.
>  This probably explains why Ninja is slower on the Skylake when using `clang-cl` as a compiler. There should be a shorter codepath maybe when only a single .cpp is being compiled, and avoid running the child process.

Huh, interesting! I had a local hack years ago where I had measured how much not spawning a subprocess for cc1 saves (it looked like https://reviews.llvm.org/D52411) and over here it didn't do anything. Can you check if patching that in helps you a lot? If so, we should reconsider doing something like that.

Repository:
  rC Clang

https://reviews.llvm.org/D52193