[llvm] [AMDGPU][True16] Support V_CEIL_F16. (PR #73108)

[Ivan Kosarev via llvm-commits]

================
@@ -2303,7 +2303,17 @@ bool AMDGPUOperand::isInlineValue() const {
 // AsmParser
 //===----------------------------------------------------------------------===//
 
-static int getRegClass(RegisterKind Is, unsigned RegWidth) {
+static int getRegClass(RegisterKind Is, unsigned RegWidth,
+                       RegisterSuffix Suffix) {
+  if (Suffix != RegisterSuffix::None) {
+    if (Is == IS_VGPR && RegWidth == 16) {
+      if (Suffix == RegisterSuffix::Lo)
+        return AMDGPU::VGPR_LO16RegClassID;
----------------
kosarev wrote:

Are we absolutely sure we want the register classes removed? If not yet, I would hesitate to complicate the code prematurely and suggest that we wait for the change removing the classes and then assess how much that looks a good idea.

How would the comment be useful? Once the classes are removed, there is no way this case could be missed.

https://github.com/llvm/llvm-project/pull/73108