[llvm] [IR] Implementation and lowering of bitinsert and bitextract (PR #200605)

[Pedro Vicente via llvm-commits]

pedroMVicente wrote:

@aengelke Added globalisel tests. Globalisel has still a few problems in its implementation that do not allow to test the same as the SDag implementation.

- AMDGPU: Has problems with the OR operation for irregular size values.
- ARM: has integer legalization issues and problems with rotates in general. Similar to x86.
- X86: Rotates and Shifts do not work properly. Also it has some problems in merging the final results for big values (>=b128). We can not test bitinsert and bitextract properly.
- PowerPC: has problems because of endiness. Can not be tested.
- RISCV: has problems with legalization of integers. Can not be tested.
- Aarch64: Is mostly fine. Could not test test_bitinsert_double_b231_var and test_bitinsert_double_b231_var and similiar.

Should we keep the tests and the globalisel implementation?

https://github.com/llvm/llvm-project/pull/200605