Details
- Reviewers
stefanp nemanjai steven.zhang jsji - Group Reviewers
Restricted Project - Commits
- rGfebbe4b5a0ab: [PowerPC] [Clang] Enable float128 feature on VSX targets
Diff Detail
Event Timeline
clang/lib/Basic/Targets/PPC.cpp | ||
---|---|---|
350 | I think, we can support the float128 as far as the altivec is enabled. Do you have any specific reason on why limit the cpu ? |
clang/lib/Basic/Targets/PPC.cpp | ||
---|---|---|
350 |
Actually no. I also think we can enable for earlier targets. (GCC documented it's enabled for all targets with VSX) Previously, I thought we did not tested targets <P8 well. |
clang/lib/Basic/Targets/PPC.cpp | ||
---|---|---|
350 | The backend implementation only depends on altivec. I think it is better to align with backend implementation. And check the cpu will also have problem if we have something like: -mcpu=pwr8 -mattr=-altivec |
I understand that the back end only needs Altivec to support the operations, but enabling this on non-vsx subtargets is different from what GCC does. I don't see a good reason for us to depart from that.
Do we know why GCC guard it under VSX instead of Altivec ? Maybe, it will help us get more clear on what we should align with (GCC's implementation or what it should be).
Align with GCC, enable float128 for VSX and later subtargets.
But besides, GCC can enable it by -mcpu=power6 -mvsx, this revision did not do it.
Looks like this is causing failures at https://lab.llvm.org/buildbot/#/builders/76/builds/2422
Please revert.
Yes. Already reverted in https://reviews.llvm.org/rGcbd93cee9bf014402a7405479ba21f6f3340a126.
I think, we can support the float128 as far as the altivec is enabled. Do you have any specific reason on why limit the cpu ?