Re: RTL obscurities

Re: RTL obscurities

[Nick Maclaren]

> If course, but you haven't told us what you want to know.  The
> question above reads like a request for a detailed RTL cookbook and
> manual, and you aren't going to get that in an email response.

ANY pointer to ANY information that will save me time reverse
engineering the code!  It is the same requirement that I had before,
and which you gave me some pointers on the second aspect.
Unfortunately, it is the first one that is critical :-(

I really do mean that I have drawn a complete blank looking for any
useful information, and have looked through several gcc internals
documents and searched the Web as far as I can!  Some time back,
I found the right place to modify (ix86_expand_prologue in i386.c),
and have inserted some RTL, but not useful RTL.

What I need to do is to insert the following:

    extern void *__limit;
    extern void __failure (void);
    if (%SP < __limit) __failure();


And the call doesn't need to be a standard one, either, though it
does need to provide __failure with a way of getting back.


Regards,
Nick Maclaren,
University of Cambridge Computing Service,
New Museums Site, Pembroke Street, Cambridge CB2 3QH, England.
Email:  nmm1@cam.ac.uk
Tel.:  +44 1223 334761    Fax:  +44 1223 334679

Re: RTL obscurities

[Andrew Haley]

Nick Maclaren writes:
 > > If course, but you haven't told us what you want to know.  The
 > > question above reads like a request for a detailed RTL cookbook and
 > > manual, and you aren't going to get that in an email response.
 > 
 > ANY pointer to ANY information that will save me time reverse
 > engineering the code!

Well, I don't know what you don't know.  Have you found the GCC
internals manual?

 > It is the same requirement that I had before, and which you gave me
 > some pointers on the second aspect.  Unfortunately, it is the first
 > one that is critical :-(
 > 
 > I really do mean that I have drawn a complete blank looking for any
 > useful information, and have looked through several gcc internals
 > documents

Okay, so you have found the GCC internals manual.

 > and searched the Web as far as I can!  Some time back, I found the
 > right place to modify (ix86_expand_prologue in i386.c), and have
 > inserted some RTL, but not useful RTL.
 > 
 > What I need to do is to insert the following:
 > 
 >     extern void *__limit;
 >     extern void __failure (void);
 >     if (%SP < __limit) __failure();
 > 
 > 
 > And the call doesn't need to be a standard one, either, though it
 > does need to provide __failure with a way of getting back.

Firstly, you'll need to generate decls for the externs.  Then you'll
need to generate RTL like this:

(insn 13 12 14 1 (set (reg:CC 17 flags)
        (compare:CC (reg:SP)
            (mem:DI (symbol_ref:DI ("__limit") <var_decl 0x2aaaae1958f0 __limit>) ))) -1 (nil)
    (nil))

(jump_insn 14 13 15 1 (set (pc)
        (if_then_else (geu (reg:CC 17 flags)
                (const_int 0 [0x0]))
            (label_ref 18)
            (pc))) -1 (nil)
    )

(call_insn 17 16 18 2 (call (mem:DI (symbol_ref:DI ("__failure") <function_decl 0x2aaaae196b00 __failure>))
        (const_int 0 [0x0])) -1 (nil)
    (nil)
    (nil))

(code_label 18 17 19 3 2 "" [1 uses])

This construct doesn't need any scratch registers.  This is good.

You've already seen how to generate RTL by perusing
ix86_expand_prologue, so you know that generating this RTL will be
done by things like

  tmp = ix86_expand_int_compare (GEU, stack_pointer_rtx, limit);
  tmp = gen_rtx_IF_THEN_ELSE (VOIDmode, tmp,
			      gen_rtx_LABEL_REF (VOIDmode, label),
		              pc_rtx);

  emit_jump_insn (tmp);
  emit_call_insn ...
  emit_label (label);

... etc.

Andrew.

Re: RTL obscurities

[Ian Lance Taylor]

Nick Maclaren <nmm1@cus.cam.ac.uk> writes:

> What I need to do is to insert the following:
> 
>     extern void *__limit;
>     extern void __failure (void);
>     if (%SP < __limit) __failure();

This is similar to what probe_stack_range in explow.c does, so that a
look at that function.

It's obviously easier to answer more detailed questions.  The general
flow of RTL is:

  * Define insn FOO in .md file
  * Generate it using emit_insn (gen_FOO (operands)).
  * Use standard names for, e.g., addition insns so that gcc knows
    which ones to use when generating code.

There are literally thousands of example of RTL generation in the gcc
code base.

Ian

Re: RTL obscurities

[Nick Maclaren]

Thank you VERY much!  That is all I need for now.  I will insert
things along those lines and see how much further I can get.


Regards,
Nick Maclaren,
University of Cambridge Computing Service,
New Museums Site, Pembroke Street, Cambridge CB2 3QH, England.
Email:  nmm1@cam.ac.uk
Tel.:  +44 1223 334761    Fax:  +44 1223 334679

Re: RTL obscurities

[Andrew Haley]

Nick Maclaren writes:

 > I am trying to reverse engineer gcc's generation of RTL and failing
 > dismally.  Alll the documentation seems to be in terms of the
 > string representation, and I can find nothing that links that to
 > the relevant calls.  I can find some calls, but nothing that will
 > do what I need to two.
 > 
 > And this is just to generate 3 instructions in the prologue!
 > 
 > Can anyone help at all with this?

If course, but you haven't told us what you want to know.  The
question above reads like a request for a detailed RTL cookbook and
manual, and you aren't going to get that in an email response.

Andrew.

RTL obscurities

[Nick Maclaren]

I am trying to reverse engineer gcc's generation of RTL and failing
dismally.  Alll the documentation seems to be in terms of the string
representation, and I can find nothing that links that to the
relevant calls.  I can find some calls, but nothing that will do
what I need to two.

And this is just to generate 3 instructions in the prologue!

Can anyone help at all with this?


Regards,
Nick Maclaren,
University of Cambridge Computing Service,
New Museums Site, Pembroke Street, Cambridge CB2 3QH, England.
Email:  nmm1@cam.ac.uk
Tel.:  +44 1223 334761    Fax:  +44 1223 334679