[PowerPC] Implement Vector Multiply High/Divide Extended Builtins in LLVM/Clang

[amyk]

Summary

This patch implements the function prototypes vec_mulh and vec_dive in order to utilize
the vector multiply high (vmulh[s|u][w|d]) and vector divide extended (vdive[s|u][w|d])
instructions introduced in Power10.

Depends on D82576.

[anil9]

clang/test/CodeGen/builtins-ppc-p10vector.c

139

I may be wrong but where are the variables declared ? I do not see the variables delclared above in the file, i mean many of them.

llvm/test/CodeGen/PowerPC/p10-vector-divide.ll

52

nit : you put a comment right at this position in the multiply.ll file, for consistency you could add one here or remove the one there.

[amyk]

clang/test/CodeGen/builtins-ppc-p10vector.c

139

I accidentally uploaded the wrong diff, I will update this.

llvm/test/CodeGen/PowerPC/p10-vector-divide.ll

52

Will fix.

[amyk]

Addressed Anil's comments.

[amyk]

Rebase patch, remove MC tests from this patch.

[lei]

clang/test/CodeGen/builtins-ppc-p10vector.c

137

why does the ck stops matching at the first param? Shouldn't we check the remaining param type and number of param are correct as well?

llvm/lib/Target/PowerPC/PPCInstrPrefix.td

1212

nit: indent to match up with v4i32 on the previous line.

[amyk]

clang/test/CodeGen/builtins-ppc-p10vector.c

137

Yes, thanks for pointing that out. Will be fixing the CHECKs.

llvm/lib/Target/PowerPC/PPCInstrPrefix.td

1212

I will update with the proper indentation.

[amyk]

Addressed the following comments:

• updated CHECK lines to check for the full intrinsic call
• updated indentation

[amyk]

Ping.

[NeHuang]

llvm/test/CodeGen/PowerPC/p10-vector-divide.ll

59

nit: do we also need _intrinsic in the name as the test cases for the vector multiply high intrinsics.

[amyk]

llvm/test/CodeGen/PowerPC/p10-vector-divide.ll

59

I am thinking that it may be OK to leave the vector divide extended tests without the _intrinsic suffix.

The reason why I added _intrinsic to vector multiply high is because the vector multiply high instructions can be produced in two ways:

• by the mulhs node (which is present in the test, I have named them test_vmulh...)
• by the @llvm.ppc.altivec.vmulh[s|u][w|d] intrinsic (these tests have the _intrinsic suffix)

So far for the vector divide extended, we produce the instructions only if we have the intrinsic.

However I do not feel too strongly either way on this so if there is a preference to add _intrinsic to these tests, I can definitely do that.

[lei]

LGTM
Thx!