i965/fs: Make resolve_source_modifiers consistent with the vec4 version

[mesa.git] / src / mesa / drivers / dri / i965 / brw_fs_peephole_predicated_break.cpp

diff --git a/src/mesa/drivers/dri/i965/brw_fs_peephole_predicated_break.cpp b/src/mesa/drivers/dri/i965/brw_fs_peephole_predicated_break.cpp
index 31b287aae364d0f7f511a35c3309f6d07e9b4119..b75f40ba5a1a222f5784e7804f37faba512f9018 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_fs_peephole_predicated_break.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_peephole_predicated_break.cpp
@@ -24,6 +24,8 @@
 #include "brw_fs.h"
 #include "brw_cfg.h"
 
+using namespace brw;
+
 /** @file brw_fs_peephole_predicated_break.cpp
  *
  * Loops are often structured as
@@ -38,6 +40,18 @@
  *
  * This peephole pass removes the IF and ENDIF instructions and predicates the
  * BREAK, dropping two instructions from the loop body.
+ *
+ * If the loop was a DO { ... } WHILE loop, it looks like
+ *
+ * loop:
+ *    ...
+ *    CMP.f0
+ *    (+f0) IF
+ *    BREAK
+ *    ENDIF
+ *    WHILE loop
+ *
+ * and we can remove the BREAK instruction and predicate the WHILE.
  */
 
 bool
@@ -72,10 +86,10 @@ fs_visitor::opt_peephole_predicated_break()
       /* For Sandybridge with IF with embedded comparison we need to emit an
        * instruction to set the flag register.
        */
-      if (brw->gen == 6 && if_inst->conditional_mod) {
-         fs_inst *cmp_inst = CMP(reg_null_d, if_inst->src[0], if_inst->src[1],
-                                 if_inst->conditional_mod);
-         if_inst->insert_before(if_block, cmp_inst);
+      if (devinfo->gen == 6 && if_inst->conditional_mod) {
+         const fs_builder ibld(this, if_block, if_inst);
+         ibld.CMP(ibld.null_reg_d(), if_inst->src[0], if_inst->src[1],
+                  if_inst->conditional_mod);
          jump_inst->predicate = BRW_PREDICATE_NORMAL;
       } else {
          jump_inst->predicate = if_inst->predicate;
@@ -95,10 +109,14 @@ fs_visitor::opt_peephole_predicated_break()
       }
       endif_inst->remove(endif_block);
 
-      earlier_block->children.make_empty();
-      later_block->parents.make_empty();
+      if (!earlier_block->ends_with_control_flow()) {
+         earlier_block->children.make_empty();
+         earlier_block->add_successor(cfg->mem_ctx, jump_block);
+      }
 
-      earlier_block->add_successor(cfg->mem_ctx, jump_block);
+      if (!later_block->starts_with_control_flow()) {
+         later_block->parents.make_empty();
+      }
       jump_block->add_successor(cfg->mem_ctx, later_block);
 
       if (earlier_block->can_combine_with(jump_block)) {
@@ -107,6 +125,30 @@ fs_visitor::opt_peephole_predicated_break()
          block = earlier_block;
       }
 
+      /* Now look at the first instruction of the block following the BREAK. If
+       * it's a WHILE, we can delete the break, predicate the WHILE, and join
+       * the two basic blocks.
+       */
+      bblock_t *while_block = earlier_block->next();
+      fs_inst *while_inst = (fs_inst *)while_block->start();
+
+      if (jump_inst->opcode == BRW_OPCODE_BREAK &&
+          while_inst->opcode == BRW_OPCODE_WHILE &&
+          while_inst->predicate == BRW_PREDICATE_NONE) {
+         jump_inst->remove(earlier_block);
+         while_inst->predicate = jump_inst->predicate;
+         while_inst->predicate_inverse = !jump_inst->predicate_inverse;
+
+         earlier_block->children.make_empty();
+         earlier_block->add_successor(cfg->mem_ctx, while_block);
+
+         assert(earlier_block->can_combine_with(while_block));
+         earlier_block->combine_with(while_block);
+
+         earlier_block->next()->parents.make_empty();
+         earlier_block->add_successor(cfg->mem_ctx, earlier_block->next());
+      }
+
       progress = true;
    }