AMDGPU/GlobalISel: Implement tail calls

[arsenm]

Summary

Or at least the sibling call cases which the DAG already handles.

[Flakebi]

It would be nice if we can put the common code with AArch64 into some generic place instead of duplicating it.
Other than that, looks good.

llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp

967–970

Checking the register mask seems pretty cheap, while checking argument assignment looks more expensive.
Does it make sense to do this check first? That should bail out faster if the calling conventions don’t match.

1224

Can this use getStackAlignment() instead of hardcoding 16?

1236

Same about getStackAlignment()

llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-sibling-call.ll

371–373

The parent does not use byval. Is this comment correct?

[arsenm]

Call common function, rebase on top of patch to fix split arguments

[arsenm]

ping

[Flakebi]

I left a nit inline. Apart from that, LGTM.

llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp

1172–1174

Can this and the assert below use getStackAlignment() instead of hardcoding 16?
Or does this have a different meaning?

[arsenm]

Don't hardcode stack alignment

[Flakebi]

llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp

1184

The assert should also use getStackAlignment().

[arsenm]

6a70874d27c73cf8b55a568449fd92f97b5bb7b3