i965: Make can_do_source_mods() a member of the instruction classes.

[mesa.git] / src / mesa / drivers / dri / i965 / brw_vec4_copy_propagation.cpp

diff --git a/src/mesa/drivers/dri/i965/brw_vec4_copy_propagation.cpp b/src/mesa/drivers/dri/i965/brw_vec4_copy_propagation.cpp
index 06f0e999ecf97044254c270abad6e5edf8891318..11571ad3b0a4ec2c86f586a159107ad55395632d 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_vec4_copy_propagation.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_copy_propagation.cpp
@@ -57,6 +57,21 @@ is_dominated_by_previous_instruction(vec4_instruction *inst)
           inst->opcode != BRW_OPCODE_ENDIF);
 }
 
+static bool
+is_channel_updated(vec4_instruction *inst, src_reg *values[4], int ch)
+{
+   const src_reg *src = values[ch];
+
+   /* consider GRF only */
+   assert(inst->dst.file == GRF);
+   if (!src || src->file != GRF)
+      return false;
+
+   return (src->reg == inst->dst.reg &&
+          src->reg_offset == inst->dst.reg_offset &&
+          inst->dst.writemask & (1 << BRW_GET_SWZ(src->swizzle, ch)));
+}
+
 static bool
 try_constant_propagation(vec4_instruction *inst, int arg, src_reg *values[4])
 {
@@ -67,7 +82,7 @@ try_constant_propagation(vec4_instruction *inst, int arg, src_reg *values[4])
     */
    src_reg value = *values[0];
    for (int i = 1; i < 4; i++) {
-      if (!value.equals(values[i]))
+      if (!value.equals(*values[i]))
         return false;
    }
 
@@ -95,9 +110,14 @@ try_constant_propagation(vec4_instruction *inst, int arg, src_reg *values[4])
       inst->src[arg] = value;
       return true;
 
+   case BRW_OPCODE_DP2:
+   case BRW_OPCODE_DP3:
+   case BRW_OPCODE_DP4:
+   case BRW_OPCODE_DPH:
+   case BRW_OPCODE_BFI1:
+   case BRW_OPCODE_ASR:
    case BRW_OPCODE_SHL:
    case BRW_OPCODE_SHR:
-   case BRW_OPCODE_ADDC:
    case BRW_OPCODE_SUBB:
       if (arg == 1) {
          inst->src[arg] = value;
@@ -111,6 +131,7 @@ try_constant_propagation(vec4_instruction *inst, int arg, src_reg *values[4])
    case BRW_OPCODE_OR:
    case BRW_OPCODE_AND:
    case BRW_OPCODE_XOR:
+   case BRW_OPCODE_ADDC:
       if (arg == 1) {
         inst->src[arg] = value;
         return true;
@@ -174,6 +195,15 @@ try_constant_propagation(vec4_instruction *inst, int arg, src_reg *values[4])
    return false;
 }
 
+static bool
+is_logic_op(enum opcode opcode)
+{
+   return (opcode == BRW_OPCODE_AND ||
+           opcode == BRW_OPCODE_OR  ||
+           opcode == BRW_OPCODE_XOR ||
+           opcode == BRW_OPCODE_NOT);
+}
+
 bool
 vec4_visitor::try_copy_propagation(vec4_instruction *inst, int arg,
                                    src_reg *values[4])
@@ -212,6 +242,14 @@ vec4_visitor::try_copy_propagation(vec4_instruction *inst, int arg,
        value.file != ATTR)
       return false;
 
+   if (brw->gen >= 8) {
+      if (value.negate) {
+         if (is_logic_op(inst->opcode)) {
+            return false;
+         }
+      }
+   }
+
    if (inst->src[arg].abs) {
       value.negate = false;
       value.abs = true;
@@ -225,7 +263,7 @@ vec4_visitor::try_copy_propagation(vec4_instruction *inst, int arg,
     * instructions.
     */
    if ((has_source_modifiers || value.file == UNIFORM ||
-        value.swizzle != BRW_SWIZZLE_XYZW) && !can_do_source_mods(inst))
+        value.swizzle != BRW_SWIZZLE_XYZW) && !inst->can_do_source_mods(brw))
       return false;
 
    if (has_source_modifiers && value.type != inst->src[arg].type)
@@ -251,7 +289,7 @@ vec4_visitor::try_copy_propagation(vec4_instruction *inst, int arg,
       return false;
 
    /* Don't report progress if this is a noop. */
-   if (value.equals(&inst->src[arg]))
+   if (value.equals(inst->src[arg]))
       return false;
 
    value.type = inst->src[arg].type;
@@ -351,11 +389,7 @@ vec4_visitor::opt_copy_propagation()
         else {
            for (int i = 0; i < virtual_grf_reg_count; i++) {
               for (int j = 0; j < 4; j++) {
-                 if (inst->dst.writemask & (1 << j) &&
-                     cur_value[i][j] &&
-                     cur_value[i][j]->file == GRF &&
-                     cur_value[i][j]->reg == inst->dst.reg &&
-                     cur_value[i][j]->reg_offset == inst->dst.reg_offset) {
+                 if (is_channel_updated(inst, cur_value[i], j)){
                     cur_value[i][j] = NULL;
                  }
               }