This is an archive of the discontinued LLVM Phabricator instance.

[X86] Remove FeatureRTM from Skylake processor list
ClosedPublic

Authored by thiagomacieira on Oct 9 2018, 2:00 PM.

Diff Detail

Repository
rL LLVM

Event Timeline

thiagomacieira created this revision.Oct 9 2018, 2:00 PM
craig.topper retitled this revision from Remove FeatureRTM from Skylake processor list to [X86] Remove FeatureRTM from Skylake processor list.Oct 9 2018, 2:00 PM
This revision is now accepted and ready to land.Oct 9 2018, 2:05 PM
lebedev.ri added inline comments.
lib/Basic/Targets/X86.cpp
169 ↗(On Diff #168868)

high-end desktop and server chips (I can't find any SKX without).

Are you sure you didn't want to move it into an else of this if() ?

craig.topper added inline comments.Oct 9 2018, 5:39 PM
lib/Basic/Targets/X86.cpp
169 ↗(On Diff #168868)

This commit leaves it disabled for all processors, but can be re-enabled for specific builds with -mrtm.

So I think this was intended.

thiagomacieira added inline comments.Oct 9 2018, 5:44 PM
lib/Basic/Targets/X86.cpp
169 ↗(On Diff #168868)

I'm sure I didn't want to :-) Do you want me to do that? And do you want me to do it for IcelakeServer too?

This matches GCC's behaviour:

$ gcc -march=skylake-avx512 -dM -E -xc /dev/null | grep -c RTM
0
craig.topper requested changes to this revision.Oct 9 2018, 5:45 PM

But there is definitely at least one test that needs to be updated. @thiagomacieira, can update them or would you like me to?

lib/Basic/Targets/X86.cpp
169 ↗(On Diff #168868)

It's also consistent with gcc based on grepping for _RTM_ in their predefined macro output. https://godbolt.org/z/dtfrQU They don't infer it from -mcpu=skylake-avx512, but do when adding -mrtm

This revision now requires changes to proceed.Oct 9 2018, 5:45 PM

Added test update

This revision is now accepted and ready to land.Oct 9 2018, 9:40 PM
lebedev.ri added inline comments.Oct 9 2018, 11:23 PM
lib/Basic/Targets/X86.cpp
169 ↗(On Diff #168868)

I'm sure I didn't want to :-) Do you want me to do that?

Not really, just looked weird.

This revision was automatically updated to reflect the committed changes.