VC fix for large vector kernel arguments

Align vector kernel arguments to element size instead of vector size

(cherry picked from commit 9abc8e4)

Author: vmustya

IGC/VectorCompiler/lib/GenXOpts/CMTrans/CMKernelArgOffset.cpp

@@ -591,6 +591,10 @@ void CMKernelArgOffset::processKernelOnOCLRT(Function *F) {
           Alignment = IGCLLVM::getAlignmentValue(
               DL.getPointerABIAlignment(Ty->getPointerAddressSpace()));
         }
+      } else if (auto *VTy = dyn_cast<IGCLLVM::FixedVectorType>(Ty)) {
+        auto *ETy = VTy->getElementType();
+        Bytes = DL.getTypeSizeInBits(Ty) / 8;
+        Alignment = IGCLLVM::getAlignmentValue(DL.getABITypeAlignment(ETy));
       } else {
         Bytes = DL.getTypeSizeInBits(Ty) / 8;
         Alignment = IGCLLVM::getAlignmentValue(DL.getABITypeAlignment(Ty));