Reverted getFragmentSizeInBits to return an integral type and not TypeSize, because scalable variables cannot be part of structs/arrays - hence they cannot be pointed by fragments.

NFC. Rename declare in test.

I reverted the return type of the lambda function to be integral type instead of TypeSize.

NFC: fix code formatting.

Rename test file. NFC.

Fix formatting in test. NFC

I have extracted D92020 to implement only the change of interface for AllocaInst::getAllocationSizeInBits.

Add comment to DbgVariableIntrinsic::getFragmentSizeInBits(). NFC

Rebase on top of D92020. NFC.

I have added assertions around before TypeSize comparisons where it made sense to do so, but not in the lambda used in the sort invocation.

Thank you for the review @sdesmalen

I have removed the C test, as the LL test is enough to test the changes I have done in Local.cpp.

This last change is trying to remove the failure at https://reviews.llvm.org/harbormaster/unit/view/196974/
I haven’t been able to reproduce such failure on my dev machine, so I just changed the RUN lines in the test (added -c, used 2>%t) to see if the bot is going to be happy with the new invocation.

FYI: I am trying to figure out this failure that does not appear on my machine: https://reviews.llvm.org/harbormaster/unit/view/196974/

I updated the RUN lines of the .ll test to prevent the failure at https://reviews.llvm.org/harbormaster/unit/view/196899/.

This LGTM. May I ask to extend the commit message to add a reference to the paragraph in section "3.7.3.3", item 2 on page 23 of the specs version 00bet6?

HI @joechrisellis - thank you. This LGTM.

Parking this for the moment, as we need to refine the heuristic to decide when to prevent the spurious zero extension.

Improve comment in test.

@rengolin - thank you for looking into this. The patch has changed quite a bit since you approved it (faulty codegen for one of the macros, which I have removed from this patch). You might want to re-look at it.

Thank you @joechrisellis - LGTM!

I have added more test coverage for the codegen of the examples mentioned in section 3.7.3.3 of the SVE ACLE. The tests are generic to work for -msve-vector-bits=128|...|2048.

Hi @joechrisellis - thank you for this patch!

In the last update I removed the tests for __ARM_FEATURE_SVE_PREDICATE_OPERATORSbut forgot to remove the code that generates and tests the macro.

I removed the enablement of __ARM_FEATURE_SVE_PREDICATE_OPERATORS as the codegen is incorrect.

LGTM, thanks for explaining!

LGTM - but can you explain in the commit why you are doing this? In particular, can you state that the test case still fails without the fix that removed the warning?

NFC: rebase on top of master.

NFC: update tests to use %clang_cc1 instead of %clang to prevent errors for missing includes: https://reviews.llvm.org/B77907

Fix typo in comment.

Thank you for your review @resistor. I simplified the logic in the
lambda function as suggested in your comment.

Thank you for the review @resistor. I have implememted the
optimization also looking through a sign_extend node. Tests to cover
this have been added, but only for extending i32 into i64 because
illegal values like i8 are always sign extended inreg.

Updating the patch after splitting it in three patches. I'll link the remeining here after publishing them.

s/Mask/SignBitPos/ in the code I changed. Revert the change in a place were Mask was accidentally renamed.

1. fixed typo
2. described tests

LGTM, thank you!

I have addressed the comment from @samparker about making the check in
UsesDifferInSignExtension independent on the order of the uses. The
code now uses llvm::any_of on all uses.

Address all review comments but one (still working on the version of UsesDifferInSignExtension that doesn't care about the order of the uses).

Hi @c-rhodes, Peter asked me to take a look at this. LGTM, I only have minor stuff.

I have removed the method SelectionDAG::isZeroExtendInReg in favour of using the machinery already available in DAGCombiner.

LGTM! Thanks.

LGTM from the perspective of making sure that this solution can be extended to any of the architectures that libmvec supports.

Thank you for modifying the implementation to facilitate the extension to other targets.

Fix typo.

@sdesmalen - I have moved the check outside the switch statement as discussed in our phone call.

Nit: there is a typo in the commit message: whre -> where.

I have added test coverage for st1[bhwd] for both .s and .d version of the alias for [<Za>.<T>, #0].