@MaskRay I have rebased to D128433 which add LoongArch support to update_llc_test_checks. Do you have any futher comments? Thanks.

In D128430#3629155, @xen0n wrote:

LGTM apart from the stack ID assertion bit; @SixWeining do you think an assertion should be present nevertheless even if we're only using one type of stack currently?

Change 'rem' to 'mod' in comments.

@xen0n @MaskRay Thanks so much for your time reviewing the patches until late at night. Let us rebase the patch.

Hello everyone. @gonglingqin is another engineer from Loongson. Please help to review her patches. Thanks.

Address @MaskRay's comment. Rebase to D128433 since we need to call update_llc_test_checks.py.

disable trapping by default

In D128572#3609897, @xen0n wrote:

Trapping division/modulus operations are signatures of MIPS codegen, and indeed here the trapping-by-default behavior and the flag seem to come from MIPS. However, as division-by-zero in LLVM IR is undefined behavior, why can't we just omit the trapping behavior altogether (and match RISCV in this regard), or at least disable the trapping by default?

Use ";;" prefix for non-RUN and non-CHECK comment line.

In D128191#3596399, @xen0n wrote:

Nice overall improvements, thanks!

Only one minor suggestion. Do you think an extra test case that verifies something like (x & 0xfff0) >> 17 are properly optimized to 0? Or is such an optimization already covered by all the generic mechanisms so it's not needed here?

Does anyone have more comments on this patch? We still have a lot of patches waiting to be submitted after this one getting approved and landed. Thanks. ;-)

In D127203#3591233, @xen0n wrote:

Thinking about the discrepancy harder, you may want to do an experiment (movcf2gr into a register pre-filled with some non-trivial content) to see which version is correct. I don't know if the translators actually *fixed* the original manual during translation (because they have no permission to alter the original document AFAIK).

remove unnecessary andi 1 after movcf2gr

Remove redundant tests in xor.ll and add missing test or.ll.

Address @xen0n's comments that removing redundant tests. Thanks.

A polite ping. Hi @xen0n, do you have any comments? Thanks.

adjust the dependency

adjust the dependency

adjust the dependency

adjust the dependency

adjust the dependency

reverse the dependency

In D127205#3576929, @MaskRay wrote:

There are still many changes to sext-zext-trunc.ll

Use ;; to emphasize non-RUN non-CHECK lines.

rebase

rebase

rebase

rebase

Address @MaskRay's comment: use ;; to emphasize non-RUN non-CHECK lines.

Address @MaskRay's comment in D127200: use ;; to emphasize non-RUN non-CHECK lines.

Fix clang-format issue.

Address @MaskRay's comments:

1. Revise the title.
2. Remove necessary static keywords.

In D127205#3575174, @MaskRay wrote:

In D127205#3575164, @SixWeining wrote:

In D127205#3575143, @MaskRay wrote:

Depends on D127204

The dependency is inverted.

I think this should be placed before other patches.

The reason why I place it after other patches is that we can obviously see the improvements from the test (introduced by early patches) updates. If this is not good, I can revert the dependency. There is no problem. Thanks.

If you think it's worthwhile, you can mention this in D127204's summary. No need to add unnecessary intermediate steps to the tests to show the effect.

In D127205#3575143, @MaskRay wrote:

Depends on D127204

The dependency is inverted.

I think this should be placed before other patches.

Remove the patch number in the title.

1. Address @MaskRay's comment by moving imm0 testing to the top of the test file.
2. Remove the patch number in the title.

Remove the patch number in the title.

Remove the patch number in the title.

Remove the patch number in the title. (I already remove it in Phabricator yesterday.)

Remove the patch number in the title.

In D127206#3569149, @xen0n wrote:

It seems only the (x >> shamt) & mask case is handled, but not (x & in_place_mask) >> shamt; is this an intentional simplification? The current behavior is good though.

In D127203#3569152, @xen0n wrote:

This is big, and I hope to finish reviewing within this week. Just commenting to let you know ;-)

In D127200#3569120, @xen0n wrote:

This is just one fneg so you may change the title to just "[LoongArch] Add codegen support for fneg" instead.

Otherwise it looks good!

Address @myhsu's comment. Thanks.

rebase

Update the testcase since the parent commit is updated just now.

Add the missing update to shift-masked-shamt.ll