Use a 64-bit data type for moving doubles to memory

Summary: When we're moving a double to memory because we have run out of VecD registers, we may need to use a temporary register to move the immediate into. In that case, we need to use a 64-bit register and not a double register, because we are treating the double as a 64-bit unsigned integer. In this case, we will explicitly set the temporary data type.

Reviewed By: DinoV

Differential Revision: D93621746

fbshipit-source-id: 13d5ade9883d9afdd1434b96b95bf1ca0884f81c

Author: kddnewton

cinderx/Jit/lir/postalloc.cpp

@@ -51,13 +51,13 @@ void insertMoveToMemoryLocation(
       block->allocateInstrBefore(
           instr_iter,
           Instruction::kMove,
-          OutPhyReg{temp, data_type},
-          Imm{constant, data_type});
+          OutPhyReg{temp, DataType::k64bit},
+          Imm{constant, DataType::k64bit});
       block->allocateInstrBefore(
           instr_iter,
           Instruction::kMove,
-          OutInd{base, index, data_type},
-          PhyReg{temp, data_type});
+          OutInd{base, index, DataType::k64bit},
+          PhyReg{temp, DataType::k64bit});
     } else {
       block->allocateInstrBefore(
           instr_iter,