i965: Perform basic optimizations on the FIND_LIVE_CHANNEL opcode.

v2: Save some CPU cycles by doing 'return progress' rather than
    'depth++' in the discard jump special case.

Reviewed-by: Matt Turner <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>

6 files changed, 93 insertions, 0 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp
index 182c79fc83c..1ca7ca6caed 100644
--- a/src/mesa/drivers/dri/i965/brw_fs.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
@@ -2980,6 +2980,53 @@ fs_visitor::compute_to_mrf()
 }
 
 /**
+ * Eliminate FIND_LIVE_CHANNEL instructions occurring outside any control
+ * flow.  We could probably do better here with some form of divergence
+ * analysis.
+ */
+bool
+fs_visitor::eliminate_find_live_channel()
+{
+   bool progress = false;
+   unsigned depth = 0;
+
+   foreach_block_and_inst_safe(block, fs_inst, inst, cfg) {
+      switch (inst->opcode) {
+      case BRW_OPCODE_IF:
+      case BRW_OPCODE_DO:
+         depth++;
+         break;
+
+      case BRW_OPCODE_ENDIF:
+      case BRW_OPCODE_WHILE:
+         depth--;
+         break;
+
+      case FS_OPCODE_DISCARD_JUMP:
+         /* This can potentially make control flow non-uniform until the end
+          * of the program.
+          */
+         return progress;
+
+      case SHADER_OPCODE_FIND_LIVE_CHANNEL:
+         if (depth == 0) {
+            inst->opcode = BRW_OPCODE_MOV;
+            inst->src[0] = fs_reg(0);
+            inst->sources = 1;
+            inst->force_writemask_all = true;
+            progress = true;
+         }
+         break;
+
+      default:
+         break;
+      }
+   }
+
+   return progress;
+}
+
+/**
  * Once we've generated code, try to convert normal FS_OPCODE_FB_WRITE
  * instructions to FS_OPCODE_REP_FB_WRITE.
  */
@@ -3920,6 +3967,7 @@ fs_visitor::optimize()
       OPT(opt_zero_samples);
       OPT(register_coalesce);
       OPT(compute_to_mrf);
+      OPT(eliminate_find_live_channel);
 
       OPT(compact_virtual_grfs);
    } while (progress);
diff --git a/src/mesa/drivers/dri/i965/brw_fs.h b/src/mesa/drivers/dri/i965/brw_fs.h
index 0783754cba7..5f4ff15ba4d 100644
--- a/src/mesa/drivers/dri/i965/brw_fs.h
+++ b/src/mesa/drivers/dri/i965/brw_fs.h
@@ -239,6 +239,7 @@ public:
    bool opt_register_renaming();
    bool register_coalesce();
    bool compute_to_mrf();
+   bool eliminate_find_live_channel();
    bool dead_code_eliminate();
    bool remove_duplicate_mrf_writes();
 
diff --git a/src/mesa/drivers/dri/i965/brw_fs_cse.cpp b/src/mesa/drivers/dri/i965/brw_fs_cse.cpp
index 8958e621802..43370cb5c2e 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_cse.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_cse.cpp
@@ -89,6 +89,7 @@ is_expression(const fs_inst *const inst)
    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD:
    case FS_OPCODE_CINTERP:
    case FS_OPCODE_LINTERP:
+   case SHADER_OPCODE_FIND_LIVE_CHANNEL:
    case SHADER_OPCODE_BROADCAST:
       return true;
    case SHADER_OPCODE_RCP:
diff --git a/src/mesa/drivers/dri/i965/brw_vec4.cpp b/src/mesa/drivers/dri/i965/brw_vec4.cpp
index 607129bc63f..a8d0e4acfc9 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4.cpp
@@ -1132,6 +1132,46 @@ vec4_visitor::opt_register_coalesce()
 }
 
 /**
+ * Eliminate FIND_LIVE_CHANNEL instructions occurring outside any control
+ * flow.  We could probably do better here with some form of divergence
+ * analysis.
+ */
+bool
+vec4_visitor::eliminate_find_live_channel()
+{
+   bool progress = false;
+   unsigned depth = 0;
+
+   foreach_block_and_inst_safe(block, vec4_instruction, inst, cfg) {
+      switch (inst->opcode) {
+      case BRW_OPCODE_IF:
+      case BRW_OPCODE_DO:
+         depth++;
+         break;
+
+      case BRW_OPCODE_ENDIF:
+      case BRW_OPCODE_WHILE:
+         depth--;
+         break;
+
+      case SHADER_OPCODE_FIND_LIVE_CHANNEL:
+         if (depth == 0) {
+            inst->opcode = BRW_OPCODE_MOV;
+            inst->src[0] = src_reg(0);
+            inst->force_writemask_all = true;
+            progress = true;
+         }
+         break;
+
+      default:
+         break;
+      }
+   }
+
+   return progress;
+}
+
+/**
  * Splits virtual GRFs requesting more than one contiguous physical register.
  *
  * We initially create large virtual GRFs for temporary structures, arrays,
@@ -1759,6 +1799,7 @@ vec4_visitor::run()
       OPT(opt_cse);
       OPT(opt_algebraic);
       OPT(opt_register_coalesce);
+      OPT(eliminate_find_live_channel);
    } while (progress);
 
    pass_num = 0;
diff --git a/src/mesa/drivers/dri/i965/brw_vec4.h b/src/mesa/drivers/dri/i965/brw_vec4.h
index 1f1f100788e..3f56c5cf03f 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4.h
+++ b/src/mesa/drivers/dri/i965/brw_vec4.h
@@ -209,6 +209,7 @@ public:
    bool opt_cse();
    bool opt_algebraic();
    bool opt_register_coalesce();
+   bool eliminate_find_live_channel();
    bool is_dep_ctrl_unsafe(const vec4_instruction *inst);
    void opt_set_dependency_control();
    void opt_schedule_instructions();
diff --git a/src/mesa/drivers/dri/i965/brw_vec4_cse.cpp b/src/mesa/drivers/dri/i965/brw_vec4_cse.cpp
index 66b531c2909..9147c3cbb79 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_cse.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_cse.cpp
@@ -72,6 +72,7 @@ is_expression(const vec4_instruction *const inst)
    case BRW_OPCODE_MAD:
    case BRW_OPCODE_LRP:
    case VEC4_OPCODE_UNPACK_UNIFORM:
+   case SHADER_OPCODE_FIND_LIVE_CHANNEL:
    case SHADER_OPCODE_BROADCAST:
       return true;
    case SHADER_OPCODE_RCP: