[llvm] [AMDGPU][MC] Restrict op_sel in VOP3P dot instructions (PR #100485)

[Jun Wang via llvm-commits]

jwanggit86 wrote:

> Besides asm parser there is also codegen. Can we just select VOP3Features for dot4 and dot8 instructions to include VOP3_PACKED, but not VOP3_OPSEL? So that op_sel operands will not even exist on these. I guess it will need some encoding bits to be hardcoded in their Real definitions.

An earlier version of the patch did just that. Then op_sel is disallowed for v_dot4_* and v_dot8_*. The current version allows it, but it must be 0.

https://github.com/llvm/llvm-project/pull/100485