An interesting project might be to figure out where the performance boost from PGO comes. If it's just a few places in llvm/clang, adding manual hinting or whatnot would allow us to skip the instrumentation step for PGO and benefit the whole community by making clang faster out of the box.

Someone had sent a patch for this a while ago, https://codereview.chromium.org/2793343002/

Just in case you haven't seen it yet, GCC developers optimized Firefox builds recently with LTO/PGO and the optimized build showed some significant improvements in binary size and performance:

1) Original post: http://hubicka.blogspot.com/2018/12/firefox-64-built-with-gcc-and-clang.html

2) Follow up: http://hubicka.blogspot.com/2018/12/even-more-fun-with-building-and.html

Hence it might be worthwile for Chromium, too?

I don't see how a blog post about building Firefox with GCC/LTO/PGO is related to compiling clang?

Let me explain then, he is currently looking for real world test cases to polish the GCC LTO/PGO implementation and chose Firefox as a test case where he compares GCC vs. Clang PGO/LTO builds. There were some critical perfomance bugs identified and fixed along that way and both projects profited from that work. This effort made at least the Fedora package maintainer reconsider their decision of dropping GCC for their Firefox build in favor of Clang and users now enjoy a better performing browser than before.

There is a collaboration opportunity here for Chromium as well, as he might have a look on Chromium soon (as mentioned by him in the Phoronix forums today [3]). Also he works for Suse and they build Chromium still with GCC. I guess there is some interest on your side as well to see how much improvement LTO + PGO from both compilers could bring to the table even though upstream Chromium is built with Clang nowadays?  

[3] = https://www.phoronix.com/forums/forum/phoronix/latest-phoronix-articles/1071721-fedora-s-firefox-to-stick-with-gcc-over-clang-beefed-up-by-lto-pgo-optimizations?p=1071996#post1071996

That seems unrelated to this bug still (which is about building clang, not chrome).

Anyone is free to build chrome with gcc and publish perf metrics :-) I'd find that interesting to read.

Ah, ok. I got it now. I've thought this bug was about building Chromium with Clang + LTO + PGO. My bad! But great to see that such work would be of interest for you and the Chromium community nevertheless. :)