CHAPTER 5 SOME EXCLUSIVE FEATURES OF A86 The IF Statement As a "nudge" in the direction of structured programming, A86 offers the IF statement. Suppose you want to conditionally skip around just one instruction. Ordinarily, this would require, for example: JNZ >L1 ; skip the following move if NZ MOV AX,BX ; make this move only if Z L1: ; this label exists only for the above skip You may replace the above code with the single line: IF Z MOV AX,BX The above line generates exactly the same code as the previous 3 lines-- a conditional jump of the opposite condition, around the statement given in the tail of the IF statement. The statement can be a macro call, giving you the opportunity to skip something more complicated. You may use any condition that would follow the "J" in a conditional jump instruction, except CXZ, which does not have a reverse condition. The assembler interprets the condition by appending a "J" to the beginning of the condition; so that the symbols "C", "NC", "Z", "NZ", etc. are not reserved by the assembler, and can be defined in other contexts. Multiple operands to PUSH, POP, INC, DEC A86 will accept any number of register operands for the instructions PUSH, POP, INC, and DEC; it will generate the appropriate machine instruction for each operand. For example, the statement PUSH AX,BX is the same as the two statements PUSH AX and PUSH BX. A numeric operand appearing in an INC or DEC statement will cause the previous INC(s) or DEC(s) to be propagated that number of times. For example, the statement INC AX,4 will generate 4 INC AX instructions. The statement DEC AL,BX,2 will generate DEC AL, DEC BX, DEC AL, DEC BX. Sorry, numeric operands are not allowed if any of the operands affected was a forward reference or relocatable quantity; e.g., INC FOO,2 where FOO is undefined. In most such cases, you'll want to code the more efficient ADD FOO,2 anyway. 5-2 Conditional Return Instructions Programmers accustomed to the conditional return instructions of the 8080/Z80 will appreciate the following feature: A86 allows the operand to a conditional jump instruction to be one of the three RET instructions RET, RETF, or IRET. The assembler will find a nearby return instruction of the indicated flavor, and use that as the target for the conditional jump. For example, JZ RET is the replacement for the 8080's RZ return-if-zero instruction. In other 8086 assembly languages, you have to find the nearby instruction yourself, attach a label to it, and use that label. Note that it does not suffice to attach a label to a single RET instruction and use that label throughout the program: the range of conditional jumps is only 128 bytes in either direction. What happens if A86 does not find a nearby return instruction? In that case, A86 issues an error, "02 Jump > 128", for the next matching return instruction in the program. If there is no subsequent return instruction, the return mnemonic will appear as an undefined symbol at the end of the program. In either case, you correct the problem by inserting a free-standing return instruction at some nearby point in the program, where it will not affect the existing code (typically following an unconditional JMP instruction). If there is no good place to insert a return instruction, you can always replace the "Jcond RET" with an "IF cond RET". A86 extensions to the MOV and XCHG instructions There are a number of MOV and XCHG instructions available in A86 that are not a part of the machine instruction set. First, moves between segment registers, and of immediate constants into segment registers are allowed. For example, if you code MOV ES,DS , the assembler will generate a PUSH DS followed by a POP ES; which will effect the move that you intended. If you code MOV DS,0 , the assembler will generate PUSH AX; MOV AX,0; MOV DS,AX; POP AX. This is mainly a convenience for D86 users to load segment registers manually. Second, MOV allows 3 operands. A statement MOV x,y,z is equivalent to the two statements MOV y,z followed by MOV x,y. Sorry, but segment overrides are not allowed in conjunction with 3-operand MOVs. The override preceding the MOV is ambiguous in its meaning; and overrides within operands cannot be handled correctly by A86. You'll have to code two MOV instructions if you want either or both to have a segment override. Third, A86 accepts a MOV of a word-sized memory operand into another word-sized memory operand. A86 handles this the same way it handles a MOV of segment registers: it generates a PUSH of the source followed by a POP of the destination. 5-3 Finally, A86 allows the XCHG of a segment register (except CS) with any other word-sized quantity, as well as the XCHG of two word-sized memory quantities. If there is no machine instruction available for XCHG a,b, then A86 generates PUSH a followed by MOV a,b followed by POP b. Local Symbols If you examine most assembly language program symbol tables, you will find that the symbols can be partitioned into two levels of significance. About half the symbols are the names of procedures and variables having global significance. If the names of these symbols are chosen intelligently and carefully, the program's readability improves drastically. (They usually aren't chosen well, most often because the assembler restricts symbols to 6 letters, or because the programmer's habits are influenced by such assemblers.) The other half of the symbols in a program have a much lower, local significance. They are only place markers used to implement small loops and local branching (e.g., "skip the next 2 instructions if the Z-flag is set"). Assigning full-blown names to these symbols reduces the readability of your program in two ways: First, it is harder to recognize local jumps for what they are-- they are usually the assembly language equivalent of high level language constructs like IF statements and WHILE loops. Second, it is harder to follow the global, significant symbols because they are buried in a sea of the place marker symbols in the symbol table. A86 solves this problem with local symbols. If a symbol in your program consists of a single letter followed by one or more decimal digits (L3, X123, Y37, etc.), then the symbol is a local symbol. Local symbols do not appear in the A86 XREF cross-reference listing. They can also be redefined to something completely different later in the program. Local symbols can be of any type: labels, memory variables, etc. Because local symbols can be redefined, you must take care to specify which one you are referring to in your program. If your reference is a forward reference (the label occurs further down in the program from the reference), then the reference must be preceded by a ">". For example, L2: MOVSB INC BX LOOP L2 ; lack of ">" means L2 is above this statement . . JNZ >L2 ; ">" indicates L2 is below this statement . JMP >L2 ; JMP L2 is disallowed here: cannot overlap ranges . L2: 5-4 I recommend that you assign all your local labels the names L0 through L9. If your program is so complex that it needs more than 10 place holders in any one stretch of code, then that stretch needs to be rewritten. Operands to AAM and AAD Instructions Those of you who have examined 86 family opcodes with an eagle eye will have noticed a somewhat spurious "0A" opcode generated after every AAM or AAD instruction. The opcode is there to provide the constant divisor or multiplicand for the instruction. Believe it or not, there wasn't enough room in the microcode of the original 8086 to hold this constant! Although Intel has never announced the generality of AAM and AAD, it is there: you can substitute any other constant for 0A (decimal 10), and that constant will be used. A86 supports this by letting you give a constant byte-sized operand to AAM or AAD. Particularly useful are the instructions AAM 16, which unpacks AL into nibbles AH and AL; and AAD 16, which reverses the process, packing nibbles AH and AL into AL. WARNING: A couple of my users point out to me that the AAD instruction with a general operand won't work on the NEC V20 and V30 chips. The operand is assumed to be 10 no matter what it really is. Since a large number of PC "speed up" kits involve switching to NEC chips, this will be seen on many PC's. You should not use AAD with an operand if you want your program to run on everybody's machine. Too bad. AAM works fine, though. Single-Operand Forms of the TEST Instruction A86 allows the TEST instruction to have a single operand, to set the flags according to the value of the operand. If the operand is a register, A86 generates a TEST of the register with itself. If the operand is a memory quantity, A86 generates a TEST of the memory with the constant -1 (i.e., the quantity will be ANDed with an all 1's constant). For example, instead of TEST DL,DL, you can code simply TEST DL. Instead of TEST WVAR,0FFFF, you can code simply TEST WVAR. Optimized LEA Instruction Many assembly-language programmers are in the habit of using, for example, LEA SI,MEMLOC instead of the equivalent MOV SI,OFFSET MEMLOC to load an immediate value that represents the pointer to a memory location. However, the LEA instruction form generates one more byte of object code than the MOV form. A86 recognizes this situation and generates the more-efficient MOV instruction when it can. This also applies to register moves: MOV AX,BX instead of LEA AX,[BX]. 5-5 I've gotten a little flak from some users about this feature. They claim it violates my policy against "behind your back" actions. But I feel that this feature is completely equivalent to code optimizations in other situations: the short JMP form instead of the equivalent near JMP; a byte operand to ADD SI,4 instead of a word operand; the one-byte XCHG AX,BX instead of the general XCHG rw,ew form; etc, etc, etc. In situations where there is absolute functional equivalence between forms, A86 tries to generate the most efficient form. But for those who are not convinced, I offer the +L2 switch, described in Chapter 3. Some users have also gotten the mistaken impression, from reading Intel's confusing specs, that the longer LEA is sometimes faster than the shorter MOV. This is never the case-- those users are reading the clock counts for the memory-fetch forms of MOV, not the register-only or immediate-value forms. If you don't believe it, try timing 1000 consecutive LEA's in a loop that executes 50000 times, vs. a similar loop with the equivalent MOV.