This is an archive of the discontinued LLVM Phabricator instance.

ARM: Fix ARMSubtarget for WOA
ClosedPublic

Authored by martell on Jun 28 2016, 1:14 PM.

Details

Diff Detail

Repository
rL LLVM

Event Timeline

martell updated this revision to Diff 62129.Jun 28 2016, 1:14 PM
martell retitled this revision from to ARM: Fix ARMSubtarget for WOA.
martell updated this object.
martell added a reviewer: compnerd.
martell set the repository for this revision to rL LLVM.
martell added a subscriber: llvm-commits.
martell updated this revision to Diff 62135.Jun 28 2016, 1:46 PM
martell removed rL LLVM as the repository for this revision.

Hi,

Have you run all the tests on Windows (check-all)?

cheers,
--renato

lib/Support/Windows/Signals.inc
535 ↗(On Diff #62135)

Shouldn't this be Fp?

martell updated this revision to Diff 62145.Jun 28 2016, 3:47 PM

Have you run all the tests on Windows (check-all)?

No but I will do.
More changes need to go into Signals.inc to properly print everything out as was intended on ARM.
I just submitted the patch initially to get feedback to see if the change was accurate

lib/Support/Windows/Signals.inc
535 ↗(On Diff #62135)

Yes it should

Right, @compnerd knows better about WoA, though.

Also, when you do a diff for Phab, please use "diff -U999" so that we can see the context around the changes. Don't worry, Phab won't show a huge file. :)

cheers,
--renato

compnerd added inline comments.Jun 28 2016, 7:05 PM
lib/Target/ARM/ARMSubtarget.cpp
216 ↗(On Diff #62145)

Windows ARM NT is a pure thumb-2 environment. We really should not be generating ARM code. I dont think that this should be changed.

It should be changed, as Thumb-2 code is generated after this. With NoARM=true, no JIT code is generated at all

if you didn't see my NoARM comment.

Thumb-2 code is generated anyway, or at least code that fully runs on both 8.0 and 8.1, Fibonnaci is stable even with high test values, which even means that __chkstk works fine.

lib/Target/ARM/ARMSubtarget.cpp
216 ↗(On Diff #62145)

It should be changed, as Thumb-2 code is generated after this. With NoARM=true, no JIT code is generated at all

Note that I'm defining the target as arm-pc-win32

martell updated this revision to Diff 99711.May 21 2017, 7:12 PM
martell edited the summary of this revision. (Show Details)
martell set the repository for this revision to rL LLVM.

updated to support woa64 and falback to unknown instead of x86 when unknown

leaving the arm subtarget code out until I actually get a device to test this on.
Should be good to merge.

martell added a reviewer: rnk.May 28 2017, 5:47 PM

@rnk can you review this if you get a chance during the week.
I think compnerd is tied down and I don't want this to rot again :)

rnk accepted this revision.Aug 3 2017, 1:47 PM

lgtm, sorry I declared email bankruptcy at some point.

This revision is now accepted and ready to land.Aug 3 2017, 1:47 PM

Thanks Reid :)

This revision was automatically updated to reflect the committed changes.