use of unconditional transfer of control instruction

Question

I did not understand why unconditional transfer of control instruction is used in cpu.So if we already know we have to jump to an instruction and skip some instruction irrespective of any condition then why do we not avoid writing those instruction as they will not be executed and replace them with instruction that will be executed after jump instruction

Answer 1

To implement if-then-else, when you get to the end of the "then" branch, you need to jump unconditionally past the "else" branch. The breaks in a C switch statement are also unconditional jumps.

You could probably also implement tail recursion by replacing "call" instructions with jumps.

Answer 2

As David Richerby's answer notes, the most common use is for implementing if-then-else, where the unconditional jump hops over the else code.

A related use is when the compiler knows conditional code is rarely executed and places it far from the more common executed code, so that an unconditional jump is used to return to the post-conditional code. This is a code transformation from:

 branch-on-cond-to-LABEL
 not-"cond"-code
 jump-to-JOIN
LABEL:
 "cond"-code
JOIN:


 branch-on-cond-to-LABEL
 not-"cond"-code

JOIN:

 ...

LABEL:
 "cond"-code
 jump-to-JOIN

This can enhance memory system performance by exploiting sequential prefetch and avoiding the fetching and storing of a partial cache block of code that is rarely used. (This is an extreme extension of placing the less frequently executed code of an if/else in the later position. This makes the common case a not taken branch which can avoid using a branch target buffer entry, can exploit simple static prediction (forward branches not taken), and can avoid the fetch redirection penalty in some processor implementations.)

It is common for instruction sets to provide a greater range for unconditional jumps than for conditional branches. This means that a conditional jump with a distant target might have to use a conditional jump (with an inverted condition) over an unconditional jump instruction.

An unconditional jump can also be used for hot patching. An unconditional jump to the next instruction (or other nop of the appropriate size) at the start of a block of code (typically a function) can be atomically changed to an unconditional jump to the new code replacing that block.

Another potential use for an unconditional jump instruction is to jump over inlined data. With a shared instruction and data cache, it might be useful to exploit block-granular retrieval from memory and sequential prefetching by placing data with code. This also allows the use of the program counter as a data address register, potentially avoiding the need to load a base address and slightly reducing register pressure.

In some early instruction sets there were no function return (or general indirect jump) instructions, so an unconditional jump instruction at the end of a function was (partially) overwritten to jump to the return address.