danalbert marked 3 inline comments as done.
danalbert added inline comments.
Comment at: clang/lib/Driver/ToolChains/Arch/ARM.cpp:389
+std::string DefaultFPU = getDefaultFPUName(Triple);
+if (DefaultFPU != "") {
+ if (!llvm::ARM::getFPUFeatures(llvm::ARM::parseFPU(De
peter.smith added a comment.
Thanks for the update. To summarise I'd like to keep the integrated and
non-integrated assembler in synch for Linux like Android, even at the risk of
adding an fpu when it isn't needed. I think that given how difficult it is to
not use NEON, I think the mfloat-abi=s
danalbert added inline comments.
Comment at: clang/lib/Driver/ToolChains/Arch/ARM.cpp:277
+ }
+ return "";
+}
peter.smith wrote:
> I'm a bit worried that we've changed the default behaviour for gnueabi[hf]
> targets here.
> For example with:
> ```
> .text
> v
danalbert updated this revision to Diff 187670.
danalbert marked 6 inline comments as done.
danalbert added a comment.
Updated to address some review comments:
- Additional tests for gnueabi/gnueabihf
- Fixed behavior for `-mfpu=neon -mfloat-abi=soft`, added test.
Repository:
rG LLVM Github M
peter.smith added a comment.
My main concern is that this changes the default behaviour for
arm-linux-gnueabi and arm-linux-gnueabihf targets. I've put some suggestions on
what I think the behaviour should be.
Comment at: clang/lib/Driver/ToolChains/Arch/ARM.cpp:277
+ }
+ r
danalbert created this revision.
danalbert added reviewers: peter.smith, kristof.beyls, srhines, pirama.
Herald added a subscriber: javed.absar.
Herald added a project: clang.
The ARM gas driver previously enabled NEON for ARMv7 and up, and a
handful of other extensions for ARMv8 and up (although
