RE: Broken SH2a patches

[Andrew STUBBS <andrew dot stubbs at st dot com>]

Hi Nick,

>    Features of this patch include:
> 
>      * I did not try to update the ASCII art in sh-opc.h with the new 
> relationships.  In my opinion it just makes the diagram too 
> cluttered to 
> be useful anymore.

Hmmm. That may be true, but I don't like it.

Here's my bash at it (note that the inheritance isn't quite as you had it -
see below):

                SH1
                 |
                SH2
   .------------'|`--------------------------------.
  /              |                                  \
SH-DSP          SH3-nommu/SH2A-nofpu               SH2E
 |               |          |`--------------------. |
 |               |          |                      \|
 |              SH3-nommu  SH4-nm-nf/SH2A-nofpu    SH3E/SH2A
 |               |\         |          |            |     |
 |               | `------. |        SH2A-nofpu     |     |
 |               |         \|            \          |     |
 |              SH3     SH4-nommu-nofpu   `---------+---. |
 |              /|\         |                       |    \|
 | .-----------' | `--------+---------------------. |    SH4/SH2A
 |/              |          /                      \|     |
 |               | .-------'                        |     |
 |               |/                                 |     |
SH3-dsp         SH4-nofpu                          SH3E  /|
 |               |`-------------------------------. | .-' |
 |               |                                 \|/    |
 |              SH4A-nofpu                         SH4   SH2A
 | .------------' `-------------------------------. |
 |/                                                \|
SH4AL-dsp                                          SH4A

Probably there ought to be a comment associated with the diagram explaining
that SHx/SHy refers to the intersection between the two architecures.
Alternatively the intersections could be made less confusing/dominant by
replacing their names with letters and then adding a legend (perhaps also
with details of shifts and prefetch differences).

>      * I have tried to use more meaningful names for the new 
> machine values.
> 
>     * I have added tests for the new machine values to the SH 
> specific 
> parts of the GAS and LD testsuites.
> 
> What do you think ?

These are all good things, and now there is a problem highlighted in the
test results:

One example:
+ sh2a-nofpu-or-sh3-nommu.s -isa=sh2e                 sh2a-or-sh3e

When -isa=sh2e it shouldn't be able to output any other isa, but it clearly
is. This is repeated many times. I think this is down to the fact that you
haven't created any new base architectures, but instead ored them together:

! #define arch_sh2a_or_sh3e                  (arch_sh2a_base|arch_sh3_base)

I haven't proven this though.

In addition to this, the inheritance tree still isn't quite right. I'm
beginning to think that if the head maintainer can't get this right then it
must be too complicated. I can't think of a better way though. Whatever
technique we use it needs to be able to select an architecture, based on the
instructions in the file, that specifies what architecture the file will run
on, without excluding any other similar, but different, architectures that
it may also run on (hence the fake architectures). Any ideas?

Specifically the problem is here

#define arch_sh3e_up       (arch_sh3e            | arch_sh2a_or_sh4_up)

and here

#define arch_sh3_nommu_up  (arch_sh3_nommu       \
		| arch_sh3_up | arch_sh2a_nofpu_or_sh4_nommu_nofpu_up)

Both these inheritances cause SH2A to allow SH3 instructions. E.g. 'ldc Rn,
SSR', 'ldc Rn, Rn_BANK' etc.  According to the document I have from the
Renesas web site the SH2A does not have such instructions. Joern and I
therefore think that the SH2A may descend from nothing but the SH2 and SH2e,
and instead it conceptually introduces some of the instructions used by the
SH3 and SH4. The diagram above shows this, but we have not checked this
instruction-by-instruction.

The code would then read:

/* Central branches.  */
#define arch_sh1_up                            (arch_sh1 \
		| arch_sh2_up)
#define arch_sh2_up                            (arch_sh2 \
		| arch_sh2e_up \
		| arch_sh2a_nofpu_or_sh3_nommu_up \
		| arch_sh_dsp_up)
#define arch_sh2a_nofpu_or_sh3_nommu_up        (arch_sh2a_nofpu_or_sh3_nommu
\
		| arch_sh2a_nofpu_or_sh4_nommu_nofpu_up \
		| arch_sh2a_or_sh3e_up \
		| arch_sh3_nommu_up)
#define arch_sh2a_nofpu_or_sh4_nommu_nofpu_up
(arch_sh2a_nofpu_or_sh4_nommu_nofpu \
		| arch_sh2a_nofpu_up \
		| arch_sh4_nommu_nofpu_up)
#define arch_sh2a_nofpu_up                     (arch_sh2a_nofpu \
		| arch_sh2a_or_sh4_up)
#define arch_sh3_nommu_up                      (arch_sh3_nommu \
		| arch_sh3_up \
		| arch_sh4_nommu_nofpu_up)
#define arch_sh3_up                            (arch_sh3 \
		| arch_sh3e_up \
		| arch_sh3_dsp_up \
		| arch_sh4_nofp_up)
#define arch_sh4_nommu_nofpu_up                (arch_sh4_nommu_nofpu \
		| arch_sh4_nofp_up)
#define arch_sh4_nofp_up                       (arch_sh4_nofpu \
		| arch_sh4_up \
		| arch_sh4a_nofp_up)
#define arch_sh4a_nofp_up                      (arch_sh4a_nofpu \
		| arch_sh4a_up \
		| arch_sh4al_dsp_up)

/* Right branches.  */
#define arch_sh2e_up                           (arch_sh2e \
		| arch_sh2a_or_sh3e_up)
#define arch_sh2a_or_sh3e_up                   (arch_sh2a_or_sh3e \
		| arch_sh2a_or_sh4_up \
		| arch_sh3e_up)
#define arch_sh2a_or_sh4_up                    (arch_sh2a_or_sh4 \
		| arch_sh2a_up \
		| arch_sh4_up)
#define arch_sh2a_up                           (arch_sh2a)
#define arch_sh3e_up                           (arch_sh3e \
		| arch_sh4_up)
#define arch_sh4_up                            (arch_sh4 \
		| arch_sh4a_up)
#define arch_sh4a_up                           (arch_sh4a)

/* Left branch.  */
#define arch_sh_dsp_up                         (arch_sh_dsp  \
		| arch_sh3_dsp_up)
#define arch_sh3_dsp_up                        (arch_sh3_dsp \
		| arch_sh4al_dsp_up)
#define arch_sh4al_dsp_up                      (arch_sh4al_dsp)

Please check this over yourself, but it looks right to me. I'm pretty sure
it matches the diagram above, but I'm not certain that both match the SH2A
architecture. There's plenty of room for screw ups here.

We're thinking about ways to automate/simplify this stuff, but nothings
going to happen this year I wouldn't think.

-- 
Andrew Stubbs
andrew.stubbs@st.com
(aka. andrew.stubbs@superh.com)