search

ClangBuiltLinux / continuous-integration

Continuous integration of latest Linux kernel with daily build of Clang & LLVM tools
https://travis-ci.com/ClangBuiltLinux/continuous-integration
Apache License 2.0
44 stars 18 forks source link

Unbreak ppc32 #194

Closed nathanchance closed 5 years ago

nathanchance commented 5 years ago

Message for first commit:

Currently, we collapse arm32_v{5,6,7} and ppc32/powerpc64le into arm and powerpc respectively. We don't want to apply patches that are for only one specific sub-architecture to all so add a SUBARCH variable that can be used to differentiate everything.

Message for second commit:

It does not seem like this is going to be dealt with in a timely manner; rather than disabling the target outright, just revert the one spot that breaks everything.

Presubmit: https://travis-ci.com/nathanchance/continuous-integration/builds/121137965

nathanchance commented 5 years ago

It's probably worth making this a clang only thing like you did, I'll update that + the commit message.

nickdesaulniers commented 5 years ago

It's probably worth making this a clang only

Sorry, yes I just thought of that, too. Please run it through the CI to make sure just dcbz is the only problematic function.

nathanchance commented 5 years ago

Hmmm, thinking about this a little bit more, I think I like your patch in #195 a little bit more than mine because all of these functions have this code generation issue:

https://godbolt.org/z/z57VU7

While dcbz might be the only problematic one at the moment, one of the other ones might be at another point in time.

At the same time, GCC 4.8.5 and GCC 8 both generate the exact same code before and after the problematic patch as Michael Ellerman pointed out. It might be worth pushing on a full revert in lieu of these workaround patches.

nickdesaulniers commented 5 years ago

Hmmm, thinking about this a little bit more, I think I like your patch in #195 a little bit more than mine because all of these functions have this code generation issue:

Shall I restart the CI of: https://github.com/ClangBuiltLinux/continuous-integration/pull/195 then or do you want to pull that patch file into this PR?

At the same time, GCC 4.8.5 and GCC 8 both generate the exact same code before and after the problematic patch as Michael Ellerman pointed out. It might be worth pushing on a full revert in lieu of these workaround patches.

We can suggest that "below the fold" when sending our patch.

nathanchance commented 5 years ago

I'll push that patch here under the ppc32 directory and kick off a presubmit run.

nathanchance commented 5 years ago

https://travis-ci.com/nathanchance/continuous-integration/builds/121137965