[m-rev.] for post-commit review: requantify after inlining

[Julien Fischer]

Hi Zoltan,

On Fri, 21 Jun 2019, Zoltan Somogyi wrote:

> For review by anyone. The attached DIFF.diet.c shows the effect of the change
> on diet.c.
>
> By the way, I got the idea for this change when I saw messages like this from
> the C# compiler:
>
> Mercury/css/diet.cs(386,25): warning CS0219: The variable `TypeInfo' is assigned but its value is never used



> diff --git a/compiler/inlining.m b/compiler/inlining.m
> index 64d0c9eaf..7fc3bd8db 100644
> --- a/compiler/inlining.m
> +++ b/compiler/inlining.m

...

> @@ -725,15 +718,15 @@ inline_in_proc(Params, ShouldInlineProcs, ShouldInlineTailProcs, PredProcId,
>              InlinedParallel = no
>          ),
> 
> -        (
> -            Requantify = yes,
> -            requantify_proc_general(ordinary_nonlocals_no_lambda, !ProcInfo)
> -        ;
> -            Requantify = no
> -        ),
> -
>          (
>              DidInlining = yes,
> +            % We want to requantify the procedure body if we did any inlining.
> +            % If the body of the some inlined call did not use some of the

Delete the first "some".

> +            % call's input arg vars, and this was the only use of the
> +            % corresponding caller variables, this will tell the simplification
> +            % pass we invoke before code generation that the goal(s) that
> +            % generate those caller variables can be optimized away.
> +            requantify_proc_general(ordinary_nonlocals_no_lambda, !ProcInfo),
>              recompute_instmap_delta_proc(recompute_atomic_instmap_deltas,

Looks fine otherwise.

Julien.