Packed array problem

Packed array problem

[Simon Wright]

Investigating the SVD2Ada program[1], for converting an ARM System View
Description[2] into bindings, I got this message when compiling against
my STM32F429I runtime:

   stm32f429x-rcc.ads:149:09: packing of 3-bit components not supported
   by configuration

which is the same as in this little example:

     1. package Checks is
     2.    type Uint3 is mod 8 with Size => 3;
     3.    type Arry is array (0 .. 1) of Uint3
                |
        >>> packing of 3-bit components not supported by configuration

     4.    with Component_Size => 3, Size => 6;
     5. end Checks;

This code compiles perfectly happily against AdaCore's ravenscar-sfp
runtime.

The error message comes from freeze.adb:2721 in the compiler:

            if Is_Constrained (Arr)
              and then Is_Bit_Packed_Array (Arr)
              and then Present (Packed_Array_Impl_Type (Arr))
              and then Is_Array_Type (Packed_Array_Impl_Type (Arr))
            then
               declare
                  CS : constant Uint  := Component_Size (Arr);
                  RE : constant RE_Id := Get_Id (UI_To_Int (CS));

               begin
                  if RE /= RE_Null
                    and then not RTE_Available (RE)
                  then
                     Error_Msg_CRT
                       ("packing of " & UI_Image (CS) & "-bit components",
                        First_Subtype (Etype (Arr)));

and it would seem after much delving that the compiler is checking (in
rtsfind) for System.Pack_03, which isn't present in my runtime. If I
copy the spec from GCC 5.2.0 (not 4.9 - the interface has changed) then
check.ads compiles OK (I should say I normally use GCC 5.2.0, but this
problem is the same with GNAT GPL 2015).

BUT there is no s-pack03.ad? in AdaCore's arm-eabi runtimes! And when
compiling against the AdaCore runtimes there's no reference in the
generated binary to s-pack03!

SO has anyone got any idea as to what the compiler could be looking for
and finding in AdaCore's RTS but not in mine?

[1] https://github.com/AdaCore/svd2ada
[2] http://www.arm.com/products/processors/cortex-m/cortex-microcontroller-software-interface-standard.php

Re: Packed array problem

[rrr.eee.27]

On Wednesday, March 2, 2016 at 6:26:29 PM UTC+1, Simon Wright wrote:
> SO has anyone got any idea as to what the compiler could be looking for
> and finding in AdaCore's RTS but not in mine?

AFAIR, it is either s-unstyp.ads or s-bitops.ad[sb] or both.

RE

Re: Packed array problem

[Simon Wright]

rrr.eee.27@gmail.com writes:

> On Wednesday, March 2, 2016 at 6:26:29 PM UTC+1, Simon Wright wrote:
>> SO has anyone got any idea as to what the compiler could be looking for
>> and finding in AdaCore's RTS but not in mine?
>
> AFAIR, it is either s-unstyp.ads or s-bitops.ad[sb] or both.

Thanks for the thought, but I already have s-unstyp.ads, and
s-bitops.ad[bs] isn't in AdaCore's RTS and made no difference to mine.

Re: Packed array problem

[Simon Wright]

Simon Wright <simon@pushface.org> writes:

> Investigating the SVD2Ada program[1], for converting an ARM System
> View Description[2] into bindings, I got this message when compiling
> against my STM32F429I runtime:
>
>    stm32f429x-rcc.ads:149:09: packing of 3-bit components not
>    supported by configuration
>
> which is the same as in this little example:
>
>      1. package Checks is
>      2.    type Uint3 is mod 8 with Size => 3;
>      3.    type Arry is array (0 .. 1) of Uint3
>                 |
>         >>> packing of 3-bit components not supported by configuration
>
>      4.    with Component_Size => 3, Size => 6;
>      5. end Checks;
>
> This code compiles perfectly happily against AdaCore's ravenscar-sfp
> runtime.

Solved. I had pragma Normalize_Scalars in package System; removed this,
now all OK.

I don't quite see why Normalize_Scalars should affect matters, since the
compiler is quite capable of generating access-3-bits code directly in
all normal cases!

Normalize_Scalars came from the Ada Bare Bones tutorial at
http://wiki.osdev.org/Ada_Bare_bones.

Re: Packed array problem

[rieachus]

> I don't quite see why Normalize_Scalars should affect matters, since the  
> compiler is quite capable of generating access-3-bits code directly in 
> all normal cases!

GNAT's implementation of Normalize_Scalars will try to initialize scalars to
an invalid value if one exists.  I suspect the bug here is that Normalize_Scalars
is trying to choose a 4-bit value 1111 and that can't be packed into three bits.

But...Normalize_Scalars should not apply to array components. So I have no idea where GNAT is trying to jam four bits into three. Glad the fix was that easy.