rubyx/lib/arm/instructions/move_instruction.rb

module Arm
  class MoveInstruction < Register::Instruction
    include Constants
    include Attributed

    def initialize to , from , options = {}
      super(nil)
      @attributes = options
      if( from.is_a?(Symbol) and Register::RegisterValue.look_like_reg(from) )
        from = Register::RegisterValue.new(from , :Integer)
      end
      @from = from
      @to = to
      raise "move must have from set #{inspect}" unless from
      @attributes[:update_status] = 1 if @attributes[:update_status] == nil
      @attributes[:condition_code] = :al if @attributes[:condition_code] == nil
      @attributes[:opcode] = attributes[:opcode]
      @operand = 0

      @immediate = 0
      @rn = :r0 # register zero = zero bit pattern
      @extra = nil
    end
    attr_accessor :to , :from

    # arm intructions are pretty sensible, and always 4 bytes (thumb not supported)
    # but not all constants fit into the part of the instruction that is left after the instruction
    # code, so large moves have to be split into two instructions.
    # we handle this "transparently", just this instruction looks longer
    # alas, full transparency is not achieved as we only know when to use 2 instruction once we
    # know where the other object is, and that position is only set after code positions have been
    # determined (in link) and so see below in assemble
    def byte_length
      @extra ? 8 : 4
    end

    def assemble(io)
      # don't overwrite instance variables, to make assembly repeatable
      rn = @rn
      operand = @operand
      immediate = @immediate
      right = @from
      if (right.is_a?(Numeric))
        if (right.fits_u8?)
          # no shifting needed
          operand = right
          immediate = 1
        elsif (op_with_rot = calculate_u8_with_rr(right))
          operand = op_with_rot
          immediate = 1
        else
            # unfortunately i was wrong in thinking the pi is armv7. The good news is the code
            # below implements the movw instruction (armv7 for moving a word) and works
            #armv7 raise "Too big #{right} " if (right >> 16) > 0
            #armv7 operand = (right & 0xFFF)
            #armv7 immediate = 1
            #armv7 rn = (right >> 12)
            # a little STRANGE, that the armv7 movw (move a 2 byte word) is an old test opcode,
            # but there it is
            #armv7 @attributes[:opcode] = :tst
          raise "No negatives implemented #{right} " if right < 0
          # and so it continues: when we notice that the const doesn't fit, first time we raise an
          # error,but set the extra flag, to say the instruction is now 8 bytes
          # then on subsequent assemblies we can assemble
          unless @extra
            @extra = 1
            #puts "RELINK M at #{self.position.to_s(16)}"
            raise ::Register::LinkException.new("cannot fit numeric literal argument in operand #{right.inspect}")
          end
          # now we can do the actual breaking of instruction, by splitting the operand
          first = right & 0xFFFFFF00
          operand = calculate_u8_with_rr( first )
          raise "no fit for #{right}" unless operand
          immediate = 1
          @extra = ArmMachine.add( to , to , (right & 0xFF) )
          #TODO: this is still a hack, as it does not encode all possible values.
          # The way it _should_ be done
          # is to check that the first part is doabe with u8_with_rr AND leaves a u8 remainder
        end
      elsif( right.is_a? Register::RegisterValue)
        operand = reg_code(right)
        immediate = 0                # ie not immediate is register
      else
        raise "invalid operand argument #{right.class} , #{self.class}"
      end
      op =  shift_handling
      instuction_class = 0b00 # OPC_DATA_PROCESSING
      val = shift(operand , 0)
      val |= shift(op , 0) # any barrel action, is already shifted
      val |= shift(reg_code(@to) ,            12)
      val |= shift(reg_code(rn) ,            12+4)
      val |= shift(@attributes[:update_status] , 12+4+4)#20
      val |= shift(op_bit_code ,        12+4+4  + 1)
      val |= shift(immediate ,          12+4+4  + 1+4)
      val |= shift(instuction_class ,   12+4+4  + 1+4+1)
      val |= shift(cond_bit_code ,      12+4+4  + 1+4+1+2)
      io.write_uint32 val
      # by now we have the extra add so assemble that
      if(@extra)
        @extra.assemble(io)
        #puts "Assemble extra at #{val.to_s(16)}"
      end
    end

  end
end