Re: Compiler and interpreter origins

On Thursday 05 Aug 2004 7:27 pm, you wrote:
> [The IBM 360 architecture specifically permits instruction modification
> without extra serialization. I gather it's still a pain for hardware
> implementers, although it's now much more common to do an EXecute
> instruction to run one instruction created at runtime. -John]


Self-modifying code is common in IBM assembler code in daily use today.
The most common form is a NOP modified to a branch instruction,
for example:


LABEL1 NOP LABEL2
OI LABEL1+1,X'FO'


The Or-Immediate changes the second byte of the NOP: converting
the instruction from a "branch never" to a "branch always".
You save one bit of memory and a whole compare instruction!


The EXecute instruction loads the target instruction into memory,
overwrites part of it with the contents of a register, and executes
the result. It is most often used to produce a variable length move.
But I have seen EXecute used as a sort of one-instruction
subroutine call: EX is only four bytes, so if you use EX to EXecute
a six byte instruction, then you are saving two bytes!


--
Martin


Martin.Ward@durham.ac.uk http://www.cse.dmu.ac.uk/~mward/ Erdos number: 4
[Eeeww. I didn't even do that back in 1970. -John]

Return to the comp.compilers page.
Search the comp.compilers archives again.