glsl: Reduce memory consumption of copy propagation passes.

opt_copy_propagation and opt_copy_propagation_elements create new ACP
and Kill sets each time they enter a new control flow block.  For if
blocks, they also copy the entire existing ACP set contents into the
new set.

When we exit the control flow block, we discard the new sets.  However,
we weren't freeing them - so they lived on until the pass finished.
This can waste a lot of memory (57MB on one pessimal shader).

This patch makes the pass allocate ACP entries using this->acp as the
memory context, and Kill entries out of this->kill.  It also steals
kill entries when moving them from the inner kill list to the parent.

It then frees the lists, including their contents.

v2: Move ralloc_free(this->acp) just before this->acp = orig_acp
    (suggested by Eric Anholt).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: "10.5 10.4" <mesa-stable@lists.freedesktop.org>
This commit is contained in:
Kenneth Graunke 2014-12-16 18:33:39 -08:00
parent eda3dd0076
commit 76960a55e6
2 changed files with 25 additions and 6 deletions

View File

@ -128,6 +128,9 @@ ir_copy_propagation_visitor::visit_enter(ir_function_signature *ir)
visit_list_elements(this, &ir->body);
ralloc_free(this->acp);
ralloc_free(this->kills);
this->kills = orig_kills;
this->acp = orig_acp;
this->killed_all = orig_killed_all;
@ -215,7 +218,7 @@ ir_copy_propagation_visitor::handle_if_block(exec_list *instructions)
/* Populate the initial acp with a copy of the original */
foreach_in_list(acp_entry, a, orig_acp) {
this->acp->push_tail(new(this->mem_ctx) acp_entry(a->lhs, a->rhs));
this->acp->push_tail(new(this->acp) acp_entry(a->lhs, a->rhs));
}
visit_list_elements(this, instructions);
@ -226,12 +229,15 @@ ir_copy_propagation_visitor::handle_if_block(exec_list *instructions)
exec_list *new_kills = this->kills;
this->kills = orig_kills;
ralloc_free(this->acp);
this->acp = orig_acp;
this->killed_all = this->killed_all || orig_killed_all;
foreach_in_list(kill_entry, k, new_kills) {
kill(k->var);
}
ralloc_free(new_kills);
}
ir_visitor_status
@ -269,6 +275,7 @@ ir_copy_propagation_visitor::visit_enter(ir_loop *ir)
exec_list *new_kills = this->kills;
this->kills = orig_kills;
ralloc_free(this->acp);
this->acp = orig_acp;
this->killed_all = this->killed_all || orig_killed_all;
@ -276,6 +283,8 @@ ir_copy_propagation_visitor::visit_enter(ir_loop *ir)
kill(k->var);
}
ralloc_free(new_kills);
/* already descended into the children. */
return visit_continue_with_parent;
}
@ -294,7 +303,7 @@ ir_copy_propagation_visitor::kill(ir_variable *var)
/* Add the LHS variable to the list of killed variables in this block.
*/
this->kills->push_tail(new(this->mem_ctx) kill_entry(var));
this->kills->push_tail(new(this->kills) kill_entry(var));
}
/**
@ -322,7 +331,7 @@ ir_copy_propagation_visitor::add_copy(ir_assignment *ir)
ir->condition = new(ralloc_parent(ir)) ir_constant(false);
this->progress = true;
} else {
entry = new(this->mem_ctx) acp_entry(lhs_var, rhs_var);
entry = new(this->acp) acp_entry(lhs_var, rhs_var);
this->acp->push_tail(entry);
}
}

View File

@ -156,6 +156,9 @@ ir_copy_propagation_elements_visitor::visit_enter(ir_function_signature *ir)
visit_list_elements(this, &ir->body);
ralloc_free(this->acp);
ralloc_free(this->kills);
this->kills = orig_kills;
this->acp = orig_acp;
this->killed_all = orig_killed_all;
@ -173,9 +176,9 @@ ir_copy_propagation_elements_visitor::visit_leave(ir_assignment *ir)
kill_entry *k;
if (lhs)
k = new(mem_ctx) kill_entry(var, ir->write_mask);
k = new(this->kills) kill_entry(var, ir->write_mask);
else
k = new(mem_ctx) kill_entry(var, ~0);
k = new(this->kills) kill_entry(var, ~0);
kill(k);
}
@ -334,7 +337,7 @@ ir_copy_propagation_elements_visitor::handle_if_block(exec_list *instructions)
/* Populate the initial acp with a copy of the original */
foreach_in_list(acp_entry, a, orig_acp) {
this->acp->push_tail(new(this->mem_ctx) acp_entry(a));
this->acp->push_tail(new(this->acp) acp_entry(a));
}
visit_list_elements(this, instructions);
@ -345,6 +348,7 @@ ir_copy_propagation_elements_visitor::handle_if_block(exec_list *instructions)
exec_list *new_kills = this->kills;
this->kills = orig_kills;
ralloc_free(this->acp);
this->acp = orig_acp;
this->killed_all = this->killed_all || orig_killed_all;
@ -354,6 +358,8 @@ ir_copy_propagation_elements_visitor::handle_if_block(exec_list *instructions)
foreach_in_list_safe(kill_entry, k, new_kills) {
kill(k);
}
ralloc_free(new_kills);
}
ir_visitor_status
@ -391,6 +397,7 @@ ir_copy_propagation_elements_visitor::visit_enter(ir_loop *ir)
exec_list *new_kills = this->kills;
this->kills = orig_kills;
ralloc_free(this->acp);
this->acp = orig_acp;
this->killed_all = this->killed_all || orig_killed_all;
@ -398,6 +405,8 @@ ir_copy_propagation_elements_visitor::visit_enter(ir_loop *ir)
kill(k);
}
ralloc_free(new_kills);
/* already descended into the children. */
return visit_continue_with_parent;
}
@ -423,6 +432,7 @@ ir_copy_propagation_elements_visitor::kill(kill_entry *k)
if (k->next)
k->remove();
ralloc_steal(this->kills, k);
this->kills->push_tail(k);
}