[llvm] [AMDGPU] Disable inline constants for pseudo scalar transcendentals (PR #104395)

[Shilei Tian via llvm-commits]

================
@@ -5776,6 +5776,18 @@ bool SIInstrInfo::isOperandLegal(const MachineInstr &MI, unsigned OpIdx,
           return false;
       }
     }
+  } else if (isVOP3(MI) && ST.hasNoF16PseudoScalarTransInlineConstants() &&
----------------
shiltian wrote:

For a "slightly" better performance, I'd put `ST.hasNoF16PseudoScalarTransInlineConstants()` before `isVOP3(MI)` because there are definitely more VOP3 instructions than the number of sub targets.

https://github.com/llvm/llvm-project/pull/104395