Re: How to generate object code

Nick Mudge wrote:
Do compilers normally turn code into an assembly language and then run
an assembler to to convert it into binary? Or does the compiler make
its own representation and itself turn the representation into object
code?

How do assemblers turn assembly into object code? Does the assembly
author have to know the binary numbers to translate the assembly into?
[Some compilers do it one way, some do it the other. Unix compilers have
traditionally compiled into assembler because early on there was a very
fast assembler available.

Re assemblers, of course they have to know the binary numbers in the
computer's instruction set. -John]

As John says there are examples either way. But I strongly assert
that the _only_ right way to do it is to produce assembler and
assemble this separately? Why?

A lng time ago I wrote Smalltalk VMs that did bytecocde
interpretation.
There were benefits if one wrote in C rather than assembler
(portablity, concision, debugging) but lots of speedups if one could
post-process the compiler-generated assembler (one could eliminate
redundant bounds checks on the bytecode dispatch switch, or even get
important variables into global registers - this before compilers like
gcc supported them directly). These tricks could yield cumulative
performance increases of 50%.

When I ported the VM from various Unix flavors (where the compilers
generate assembler) to Windows (which generated object code directly)
none of these techniques could be applied because although the
Microsoft compiler could be asked to spit out assembly code, the
assembly that it did produce was incorrect. If one assembled the
output it would only match the directly-compiled object code for
trivial cases. I presume that assembly was produced by decompiling
the object code. Because C->assembly->object was not the default path
its incorrectness was perhaps not noticed and certainly not corrected.

So if you're thinking of writing a compiler make it generate assembly
directly. It'll integrate better and be more extensible.
--
Eliot ,,,^..^,,, Smalltalk - scene not herd
[It certainly helps that Unix assemblers tend to be small and fast,
without macro languages or other addons that slow down assemblers
on other systems. If you really need macros, you can run your code
through m4 first. -John]

.

Relevant Pages

  • Re: Python for everything?
    ... > than you get in C - those are still done in assembler. ... the early days of DOS/Windows compilers, about which I know little, but ... device driver code for large Unix systems, ...
    (comp.lang.python)
  • Re: Calling asm from C
    ... char vendorbuf; ... *must* agree with the ABI that the assembler routine expects. ... I've been assuming that "Intel ABI" means something... ... If the C compilers are going to get all random on us, ...
    (alt.lang.asm)
  • Re: Why use relational tables in OO (please just take a look inside)
    ... B-) C is 30-50% slower than hand-tuned assembler. ... > library currently supports these two compilers. ... STL headers and libraries, I assumed I would send you the data files and ...
    (comp.object)
  • Re: Assembler?
    ... 6502 assembler I used was about 20K, and the disassembler was less than ... compilers optimized to them, they are optimised to the compilers and ... Register shmegister. ... What's cache RAM but a shitload of fast registers..especially for the stack?.. ...
    (comp.os.linux.misc)
  • Re: Future of programming
    ... but again, let's not forget in my opinion the main purpose of programming, ... the more abstract the programming language ... code-generators, or even efficient compilers, which is not the case ... > fastest solution to the problem, then assembler is still needed. ...
    (alt.lang.asm)