Clean up. To be applied on top of D100250.

Fix one more test case.

In D99754#2678502, @MaskRay wrote:

llvm/BinaryFormat/ELF.h is typically added along with tests to llvm-readobj/yaml2obj/obj2yaml. See D95728 for an example.

Use CHECK-NEXT in more places.

Ping (& fix one small style issue pointed out by clang-tidy)

Use suffix instead of an explicit cast.

In D99465#2654633, @craig.topper wrote:

We can't use preprocessor defines in the intrinsic header files. It prevents being able to use the intrinsics with attribute(target("avx512f,vaes")). What's preventing __m512i from being defined?

Run clang-format on the modified line.

Rebase to main.

Ping ?

Split this patch, the SelectionDAG part is in D98273

Sure. I'll have the first one introduce the SelectionDAG rules and the second one introduce the fixes for the MIPS codegen together with the various tests.

In D98189#2614070, @atanasyan wrote:

I have not checked the patch carefully yet. But it looks like some tests was removed. For example MIPS64R5EL-LABEL: i8_2... Is it intended?

Apply the TRUNCATE + BITCAST strategy everywhere, at least on MIPS the generated code is noticeably better.

Apply clang-format

Assert all the operand types are the same.

Properly handle cases where the true/false operand have different types than the comparison ones.

But, I'm rather curious about how the miscompilation happened from this optimization. A gep with such offset isn't common, unless a programmer writes a code that subtracts a pointer from null (which is already fishy)?
It would be great if I can see the input that causes miscompilation.

Don't mark the fmin/fmax ops as legal if fullfp16 is not available.
Update test cases.

Update some documentation comments.

Remove unused parameter.

Rebased after changes in master, feel free to land the changes right away as I have no commit access (yet?).

You can use "-stop-after=finalize-isel" and check the MIR output which should print the memory operands. I think maybe update_mir_test_checks.py can generate the check lines for MIR output from a .ll test.

• Add MIR test
• Replace dyn_cast<> with cast<>

Roll back to something simpler, let's only implement %lm and its encoding.

The test is not so solid (is it?), ideally we should check the MI to make sure the load ops have the correct offset but I have no idea of what kind of test does that.

As usual please merge if no more reviews are needed.

@MaskRay Is this ready to go now?

I am once again asking for your financial support.
@MaskRay would you mind landing this now that it got another approval mark?

sigh
Ping?

Ping?

It tries to fix a couple of things, so it would probably been better to divide it up into more than one patch and add some more tests, but this is fine. I compared the result with GAS for V8 and it matches.

In D89628#2358994, @lebedev.ri wrote:

In D89628#2342684, @LemonBoy wrote:

Looks good to me, thanks.

Great, can you commit this on my behalf? I have no commit access.

@hans Is it too early to request this patch to be included in 11.0.1?

Please specify the Author: name <e@ma.il> to be used for the commit.

Ping?

Looks good to me, thanks.

Cleaned up the true/false building parts.
Added a note about LLVM not detecting if signed multiplication overflow, not much of a problem and can be fixed later.

Looks good to me - thanks!

Doesn't the same issue apply to the isNeutralValue branch?

Fix the neutral value code path too.

Oh well, it turns out the results are evaluated according to all the elements so it's ok to splat it.

Use ScopeExit instead of a bare lambda.

Perhaps this'd be more robust with ScopeExit?

Ping with some more reviewers, hoping to land this in time for LLVM 99.

LGTM, I typo'ed the comparison operator when I removed the switch. Sorry about that.

@LemonBoy from your previous comment it looks like you don't intend to work on this further, is that correct? If so, are you okay with me taking over this patch, thus uploading this new patch for example?

I won't have time to update this diff in the near future so feel free to add the value range checks in a follow-up commit/patch.