#include "ir_rvalue_visitor.h"
#include "ir_basic_block.h"
#include "ir_optimization.h"
-#include "glsl_types.h"
+#include "compiler/glsl_types.h"
static bool debug = false;
+namespace {
+
class acp_entry : public exec_node
{
public:
ir_copy_propagation_elements_visitor()
{
this->progress = false;
+ this->killed_all = false;
this->mem_ctx = ralloc_context(NULL);
this->shader_mem_ctx = NULL;
this->acp = new(mem_ctx) exec_list;
virtual ir_visitor_status visit_leave(class ir_assignment *);
virtual ir_visitor_status visit_enter(class ir_call *);
virtual ir_visitor_status visit_enter(class ir_if *);
+ virtual ir_visitor_status visit_leave(class ir_swizzle *);
void handle_rvalue(ir_rvalue **rvalue);
void *shader_mem_ctx;
};
+} /* unnamed namespace */
+
ir_visitor_status
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;
ir_copy_propagation_elements_visitor::visit_leave(ir_assignment *ir)
{
ir_dereference_variable *lhs = ir->lhs->as_dereference_variable();
+ ir_variable *var = ir->lhs->variable_referenced();
+
+ if (var->type->is_scalar() || var->type->is_vector()) {
+ kill_entry *k;
+
+ if (lhs)
+ k = new(this->kills) kill_entry(var, ir->write_mask);
+ else
+ k = new(this->kills) kill_entry(var, ~0);
- if (lhs && (lhs->type->is_scalar() || lhs->type->is_vector())) {
- kill_entry *k = new(mem_ctx) kill_entry(lhs->var, ir->write_mask);
kill(k);
}
return visit_continue;
}
+ir_visitor_status
+ir_copy_propagation_elements_visitor::visit_leave(ir_swizzle *)
+{
+ /* Don't visit the values of swizzles since they are handled while
+ * visiting the swizzle itself.
+ */
+ return visit_continue;
+}
+
/**
* Replaces dereferences of ACP RHS variables with ACP LHS variables.
*
int swizzle_chan[4];
ir_dereference_variable *deref_var;
ir_variable *source[4] = {NULL, NULL, NULL, NULL};
- int source_chan[4];
+ int source_chan[4] = {0, 0, 0, 0};
int chans;
+ bool noop_swizzle = true;
if (!*ir)
return;
/* Try to find ACP entries covering swizzle_chan[], hoping they're
* the same source variable.
*/
- foreach_iter(exec_list_iterator, iter, *this->acp) {
- acp_entry *entry = (acp_entry *)iter.get();
-
+ foreach_in_list(acp_entry, entry, this->acp) {
if (var == entry->lhs) {
for (int c = 0; c < chans; c++) {
if (entry->write_mask & (1 << swizzle_chan[c])) {
source[c] = entry->rhs;
source_chan[c] = entry->swizzle[swizzle_chan[c]];
+
+ if (source_chan[c] != swizzle_chan[c])
+ noop_swizzle = false;
}
}
}
if (!shader_mem_ctx)
shader_mem_ctx = ralloc_parent(deref_var);
+ /* Don't pointlessly replace the rvalue with itself (or a noop swizzle
+ * of itself, which would just be deleted by opt_noop_swizzle).
+ */
+ if (source[0] == var && noop_swizzle)
+ return;
+
if (debug) {
printf("Copy propagation from:\n");
(*ir)->print();
source_chan[2],
source_chan[3],
chans);
+ progress = true;
if (debug) {
printf("to:\n");
ir_copy_propagation_elements_visitor::visit_enter(ir_call *ir)
{
/* Do copy propagation on call parameters, but skip any out params */
- exec_list_iterator sig_param_iter = ir->get_callee()->parameters.iterator();
- foreach_iter(exec_list_iterator, iter, ir->actual_parameters) {
- ir_variable *sig_param = (ir_variable *)sig_param_iter.get();
- ir_instruction *ir = (ir_instruction *)iter.get();
- if (sig_param->mode != ir_var_out && sig_param->mode != ir_var_inout) {
+ foreach_two_lists(formal_node, &ir->callee->parameters,
+ actual_node, &ir->actual_parameters) {
+ ir_variable *sig_param = (ir_variable *) formal_node;
+ ir_rvalue *ir = (ir_rvalue *) actual_node;
+ if (sig_param->data.mode != ir_var_function_out
+ && sig_param->data.mode != ir_var_function_inout) {
ir->accept(this);
}
- sig_param_iter.next();
}
/* Since we're unlinked, we don't (necessarily) know the side effects of
this->killed_all = false;
/* Populate the initial acp with a copy of the original */
- foreach_iter(exec_list_iterator, iter, *orig_acp) {
- acp_entry *a = (acp_entry *)iter.get();
- this->acp->push_tail(new(this->mem_ctx) acp_entry(a));
+ foreach_in_list(acp_entry, a, orig_acp) {
+ this->acp->push_tail(new(this->acp) acp_entry(a));
}
visit_list_elements(this, 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;
/* Move the new kills into the parent block's list, removing them
* from the parent's ACP list in the process.
*/
- foreach_list_safe(node, new_kills) {
- kill_entry *k = (kill_entry *)node;
+ foreach_in_list_safe(kill_entry, k, new_kills) {
kill(k);
}
+
+ ralloc_free(new_kills);
}
ir_visitor_status
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_list_safe(node, new_kills) {
- kill_entry *k = (kill_entry *)node;
+ foreach_in_list_safe(kill_entry, k, new_kills) {
kill(k);
}
+ ralloc_free(new_kills);
+
/* already descended into the children. */
return visit_continue_with_parent;
}
void
ir_copy_propagation_elements_visitor::kill(kill_entry *k)
{
- foreach_list_safe(node, acp) {
- acp_entry *entry = (acp_entry *)node;
-
+ foreach_in_list_safe(acp_entry, entry, acp) {
if (entry->lhs == k->var) {
entry->write_mask = entry->write_mask & ~k->write_mask;
- if (entry->write_mask == 0)
+ if (entry->write_mask == 0) {
entry->remove();
+ continue;
+ }
}
if (entry->rhs == k->var) {
entry->remove();
if (k->next)
k->remove();
+ ralloc_steal(this->kills, k);
this->kills->push_tail(k);
}
swizzle[i] = orig_swizzle[j++];
}
- entry = new(this->mem_ctx) acp_entry(lhs->var, rhs->var, ir->write_mask,
+ int write_mask = ir->write_mask;
+ if (lhs->var == rhs->var) {
+ /* If this is a copy from the variable to itself, then we need
+ * to be sure not to include the updated channels from this
+ * instruction in the set of new source channels to be
+ * copy-propagated from.
+ */
+ for (int i = 0; i < 4; i++) {
+ if (ir->write_mask & (1 << orig_swizzle[i]))
+ write_mask &= ~(1 << i);
+ }
+ }
+
+ entry = new(this->mem_ctx) acp_entry(lhs->var, rhs->var, write_mask,
swizzle);
this->acp->push_tail(entry);
}
projects / mesa.git / blobdiff