[llvm] [AMDGPU] Add half vector support for table-driven libcall optimzation (PR #178638)

[Matt Arsenault via llvm-commits]

================
@@ -743,16 +743,17 @@ bool AMDGPULibCalls::fold(CallInst *CI) {
   return false;
 }
 
-static Constant *
-_Z4coshdgetConstantFloatVectorForArgType(LLVMContext &Ctx,
-                                         AMDGPULibFunc::EType ArgType,
-                                         ArrayRef<double> Values, Type *Ty) {
+static Constant *getConstantFloatVectorForArgType(LLVMContext &Ctx,
+                                                  AMDGPULibFunc::EType ArgType,
+                                                  ArrayRef<double> Values,
----------------
arsenm wrote:

This probably should be APFloat. Though last I looked all of these constant aggregate APIs were kind of a mess, so I wouldn't be surprised if you're forced into using double 

https://github.com/llvm/llvm-project/pull/178638