On Fri, Aug 04, 2017 at 08:45:14AM +0300, Henri Sivonen wrote: > I guess I buried my questions in too long a post, so extracting them: > > On Mon, Jul 31, 2017 at 1:02 PM, Henri Sivonen <hsivo...@hsivonen.fi> wrote: > > Naïvely, one would think that it should be possible to do that with > > clang producing "object files" holding LLVM IR and rustc producing > > "object files" holding LLVM IR and the "link" step involving mashing > > those together, running LLVM optimizations again and then producing > > machine code from a massive collection of mashed-together LLVM IR. > ... > > If we were compiling C++ using clang on all platforms, how far off > > would such cross-language inlining be? > > > > If we could have the cross-language inlining benefit from compiling > > C++ using clang on all platforms, how far off would we be from being > > able to switch to clang on all platforms?
Here's a bunch of data why "let's switch compilers" is not necessarily easy (I happen to have gathered that recently): Windows MSVC (PGO) vs. clang-cl: https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=581eab2aa162&newProject=try&newRevision=c963d0cd6565&framework=1&showOnlyImportant=0 Windows MSVC (PGO) vs. clang-cl -O2 (the build system currently uses -O1 for msvc builds, which flags we use for clang-cl builds): https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=581eab2aa162&newProject=try&newRevision=c963d0cd6565&framework=1&showOnlyImportant=0 Newer versions of clang-cl might generate faster code, but they crash during the build: https://bugs.llvm.org/show_bug.cgi?id=33997 Note that the tp5n main_startup_fileio reflects the resulting size of xul.dll, which also impacts the installer size: 32-bits 64-bits MSVC (PGO): 37904383 40803170 clang-cl: 39537860 40561849 clang-cl -O2: 41976097 43338891 64-bits clang-cl is miraculously smaller, but the zip is larger, and so is the on-disk xul.dll. Linux gcc 4.9 PGO vs clang trunk (r309511): https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=3c83f65009d6&newProject=try&newRevision=5d22234f2596&framework=1&showOnlyImportant=0 This might look unfair (PGO vs. not-PGO), but for some reason, trying PGO with clang breaks the build... while invoking cargo... https://treeherder.mozilla.org/#/jobs?repo=try&revision=8aeb26f7b22b7c8a03602229caa79d20bdadc94d if anyone wants to take a look. Mike _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
