[m-dev.] for review: add MR_virtual_hp, MR_virtual_sp, etc.

[Zoltan Somogyi]

> +#ifdef USE_CONSERVATIVE_GC
> +  #define incr_saved_hp(A, B) incr_hp(A, B)
> +  #define incr_saved_hp_atomic(A, B) incr_hp_atomic(A, B)

Make the macro bodies line up, and add a comment saying that the
reason why the "saved" part has no effect is that there is in fact
no hp registers.

> +#else
> +  /* same as tag_incr_hp except we use MR_virtual_hp instead of MR_hp */
> +  #define incr_saved_hp(dest, count)				\
> +	(							\
> +		(dest) = (Word) mkword(0, (Word) MR_virtual_hp),\
> +		debugincrhp(count, MR_virtual_hp),		\
> +		MR_virtual_hp += (count),			\
> +		saved_heap_overflow_check(),			\
> +		(void)0						\
> +	)
> +  #define incr_saved_hp_atomic(A, B) incr_saved_hp(A, B)
> +#endif

Add a comment that atomicity is meaningful only with the Boehm collector.

>  /*
>  ** The registers of the Mercury virtual machine are built up using
>  ** three levels of abstraction.
>  **
> -** The first level defines the first NUM_REAL_REGS register variables
> -** mr0, mr1, etc. as the physical machine registers, and defines an
> -** array fake_regs[n] of pseudo registers.
> +** The bottom level is the hardware description layer.
> +** This layer is defined seperately for each architecture, in machdeps/*.h.

s/seperate/separate/

I have seen you make this mistake often enough that I suggest that you
put it (and variations with endings such as "ly") into a vim abbreviation.

>  ** Since the set of most frequently used Mercury virtual machine
> -** registers can be different for each program, we want to make
> +** registers can be different for each program or grade, we want to make
>  ** this mapping as easy to change as possible. This is why the
>  ** map is in a minimal header file, mercury_regorder.h.

Making the mapping different for each program is an idea whose time
has passed, so you may want to delete that part of the comment.

There is a subtle problem with your diff. The purpose of these changes
is to fix the XXX problem in mercury_trace_util.c, but defining MR_virtual*
to refer to fake_reg is not what is wanted there. Instead that code wants to
refer to MR_saved_regs, because this has the values from the event, and
fake_reg will not, except when displaying the value of the first variable
(since fake_reg will be overwritten by the display routine itself).

Zoltan.