Relative Addresses (Relocatables) and the Minus Operator (80381)

SUMMARY

The Microsoft Macro Assembler does not support using the minus operator (-) to negate a relative address, as in the following statement:

   lea ax, [bx - offset Table]

Under MASM 5.1 and 5.1a, the error generated for this statement is:

A2040: operands must be in same segment, or one must be constant

Under MASM 6.0, the error generated is:

A2094: operand must be relocatable

MORE INFORMATION

Relocatable means not having an absolute address. Relocatables are often referred to as "relative addresses." The assembler does not know where the label, data, or code will be located in memory. The linker provides the address.

The assembler generates a fixup record (FIXUPP) for each address it cannot determine at assembly time, which contains the information the linker will need to determine the address. The FIXUPP record contains the LOCATION of the address to be fixed up, the TARGET address to which the fixup refers, and the FRAME relative to which the address computation is performed. These fields are explained in detail on pages 682-693 in "The MS-DOS Encyclopedia." The FIXUPP record follows the data record that link needs to update.

Note that an address that needs to be fixed up is indicated in the source listing produced by the assembler by an R following the instruction. This is a handy way to observe some of the common fixups.

A statement such as the following

   lea ax, [offset Table1 - offset Table]

(where Table1 and Table are in the same segment) includes a difference of relocatables and will be an immediate value. The assembler will perform the computation. This is one of the recommendations in the A2040 error shown in the summary.

A statement like such as the following

   lea ax, [bx + offset Table]

is equivalent to lea ax, Table[bx]. The offset of Table is known by the assembler, stored in a data record, fixed up, and added to bx at run time.

In the following statement introduced in the summary

   lea ax, [bx - offset Table]

the offset of Table is known, but the negation of an address is not defined for any language. The assembler could merely toggle the bits, but how would the linker interpret the new, negated address and perform the fixup? What would happen at run time? As it stands, the value produced would have no meaning. Since this fixup is not supported, an error is generated.