AMDGPU: Set implicit kernarg size to be of 256 bytes for code object version 5

[cfang]

Summary

If implicitarg_ptr intrinsic is not used, set implicit kernarg size to 0, otherwise
set it to 256 bytes for code object version 5 (and beyond).

[arsenm]

This can be tested in the existing llvm/test/CodeGen/AMDGPU/llvm.amdgcn.implicitarg.ptr.ll

[cfang]

In D123262#3434565, @arsenm wrote:

This can be tested in the existing llvm/test/CodeGen/AMDGPU/llvm.amdgcn.implicitarg.ptr.ll

This will make things much complicated since version 5 will generates quite different code from version 2, also
the dumped variable names are different. Maybe we should only check the kernarg size for version 5.

[cfang]

update LIT test

[arsenm]

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.implicitarg.ptr.ll

5 ↗

(On Diff #421061)

This is broken. You need to add back the GCN check prefix

[cfang]

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.implicitarg.ptr.ll

5 ↗

(On Diff #421061)

Can you tell how this is broken? It passed the test for my check.
The problem is that, if we add back the GCN check prefix, the COV5 check won't be able to check the label (When I add prefix to COV5 cases, the test will fail GCN check because COV5 is quite different).

[cfang]

For the LIT test, add back the GCN check prefix. For the GCN checks that are common to code object version 2 and MESA but different from code object version 5, we split the checks to HSA (version 2) and MESA.