Re: AW: [PATCH] Fix unaligned access when predictive commoning (PR 71083)

[Eric Botcazou <ebotcazou@adacore.com>]

> I tried to translate the c-test case to an equivalent ada test case with
> my not-so-fluent Ada speak...
> 
> So here is a test case that *actually* hits the if ((boff % BITS_PER_UNIT)
> != 0) code path.

I see, well done then.

> Before my patch there was an unaligned SImode access, and with
> the patch we have 3 QImode accesses here.
> 
> cat pr71083_pkg.ads
> package Pr71083_pkg is
>   type Nibble is mod 2**4;
>   type Int24  is mod 2**24;
>   type StructA is record
>     a : Nibble;
>     b : Int24;
>   end record;
>   pragma Pack(StructA);
>   type StructB is record
>     a : Nibble;
>     b : StructA;
>   end record;
>   pragma Pack(StructB);
>   type ArrayOfStructB is array(0..100) of StructB;
>   procedure Foo (X : in out ArrayOfStructB);
> end Pr71083_pkg;
> 
> cat pr71083_pkg.adb
> -- { dg-do compile }
> -- { dg-options "-O3" }
> package body Pr71083_pkg is
>   procedure Foo (X : in out ArrayOfStructB) is
>   begin
>     for K in 0..99 loop
>       X (K+1).b.b := X (K).b.b;
>     end loop;
>   end Foo;
> end Pr71083_pkg;
> 
> cat pr71083.adb
> -- { dg-do run }
> -- { dg-options "-O3" }
> with Pr71083_pkg;
> use Pr71083_pkg;
> procedure Pr71083 is
>   Test : ArrayOfStructB;
> begin
>   Test (0).b.b := 9999;
>   Foo (Test);
>   if Test (100).b.b /= 9999 then
>     raise Program_Error;
>   end if;
> end;
> 
> 
> I was not sure which name to choose,
> so I used the same convention as in the c-torture.
> How do you like that?

Worst convention ever. :-)  Either opt57 or loop_optimization23 at your 
convenience (you can add a reference to the PR in a comment).

> I would like to add that to the gnat.dg because
> it seems that there is zero testing for the predictive
> commoning in the gnat.dg test suite.

Very likely so.  The renamed testcase is OK for mainline, but please replace 
"Eric's" with "the" in the patch or copy-and-paste the explanation.

-- 
Eric Botcazou