JMP

JMP puts a new address into IP (and, for a far jump, into CS too) and execution carries on from there. It tests nothing, pushes nothing and changes no flag. It is the glue of assembly control flow: the jump over the "else" part, the jump back to the top of a loop, the way out of the middle of a block.

Forms

The direct near and short forms store a displacement, the distance from the end of the JMP to the target, so the code works wherever it is loaded. MASM chooses the shortest encoding that reaches; write SHORT to insist on two bytes, and MASM reports an error if the target is too far. The indirect forms read the destination at run time, which is how jump tables and dispatchers work:

MOV     BX, 2                ; case 1, times 2: the table holds words
JMP     WORD PTR [JTAB1+BX]  ; near, indirect: IP = the word at JTAB1+BX
PICK0:
MOV     AL, 0
JMP     SHOW1
PICK1:
MOV     AL, 1                ; this case runs
JMP     SHOW1
PICK2:
MOV     AL, 2
SHOW1:
ADD     AL, "0"              ; digit -> character
MOV     DL, AL
MOV     AH, 02h              ; DOS: print the character in DL
INT     21h
 
JTAB1   DW      PICK0, PICK1, PICK2

Pitfalls

A .COM file has no relocation table, so a far jump whose segment part names your own program cannot be linked into one; JMP FAR PTR to a label of your program belongs in .EXE programs. A far jump through a doubleword in memory (JMP DWORD PTR [OLDVEC1], the usual way to chain to a saved interrupt vector) is fine anywhere. JMP $ jumps to itself: an endless loop. Code right after a JMP never runs unless some label leads to it.

Later CPUs

In 16-bit code later processors behave the same. The 386 adds 32-bit displacements and offsets for 32-bit code, and in protected mode a far jump loads a selector instead of a segment.