On Thu, Jul 15, 2021 at 10:22 AM Erick Ochoa via Gcc <gcc@gcc.gnu.org> wrote:
>
> Hi,
>
> I was sorting SSA trees and DECL_P trees into different buckets. I
> used DECL_P as a proxy for it being a local/global variable/function
> (essentially a declaration). It seems that "probabilistically", I'm
> kinda right. Normally there is no tree that is both an SSA_VAR_P and a
> DECL_P at the same time, but there appears to be at least one in a
> couple of large benchmarks.
>
> What could lead to a tree being declared both SSA_VAR_P and DECL_P?
/* Nonzero if DECL represents an SSA name or a variable that can possibly
have an associated SSA name. */
#define SSA_VAR_P(DECL) \
(TREE_CODE (DECL) == VAR_DECL \
|| TREE_CODE (DECL) == PARM_DECL \
|| TREE_CODE (DECL) == RESULT_DECL \
|| TREE_CODE (DECL) == SSA_NAME)
there's your answer. SSA_VAR_P and DECL_P are not related in a way
you think.
> The first instance I've looked at where this happens the variable in
> the source is a static class member. But are there other cases? And
> why would it be the case that for smaller benchmarks there seems to be
> no instance of trees that are both SSA_VAR_P and DECL_P at the same
> time?
>
> I've been interpreting DECL_P as any variable that actually takes up
> memory. E.g., for every DECL_P variable there will be space allocated
> on the stack. But SSA_VAR_P does not need space allocated since they
> are just an abstraction to think about the flow of assignments. For
> example, we might have the following pseudo-gimple:
>
> PARM cond // variable cond is a boolean parameter (also DECL_P)
> DECL int i. // variable i is a local integer declaration
> // stack space is reserved for int i
> if (cond)
> {
> i_0 = 0 // i_0 is an SSA variable that corresponds to i
> }
> else
> {
> i_1 = 1 // same
> }
> i_2 = phi(i_0, i_1)
>
> In the end i_x will refer to the stack location reserved for i. I
> think that this is a somewhat naive view of the execution model but it
> is not until now that I found a counter example.
>
> Is the above not the case?
>
> As another note, I have not gone through the details on how static
> class member variables are implemented, so perhaps there are some
> details there where it would make sense to have a variable be both
> SSA_VAR_P and DECL_P.
>
> Can someone shed some light on this? Thanks!
>
> Thanks!
