[llvm] [SPIRV] Enable `bfloat16` arithmetic (PR #166031)

[Juan Manuel Martinez Caamaño via llvm-commits]

================
@@ -2688,7 +2688,7 @@ bool IRTranslator::translateKnownIntrinsic(const CallInst &CI, Intrinsic::ID ID,
 
 bool IRTranslator::translateInlineAsm(const CallBase &CB,
                                       MachineIRBuilder &MIRBuilder) {
-  if (containsBF16Type(CB))
+  if (!MF->getTarget().getTargetTriple().isSPIRV() && containsBF16Type(CB))
----------------
jmmartinez wrote:

NIT: Can we add a helper `canHandleBF16ForTarget()` that does the `MF->getTarget().getTargetTriple().isSPIRV()` check?

Adding a hook in `TargetMachine` would be ideal but it'd be an overkill.

https://github.com/llvm/llvm-project/pull/166031