Merge remote branch 'origin/master' into pipe-video
[mesa.git] / src / glsl / ir.cpp
index 7cc55d40b78846c388154f4e790b59da58c15296..fc356ba52752dac3a6ee9e8e74b0362c1c3fd904 100644 (file)
@@ -31,6 +31,21 @@ ir_rvalue::ir_rvalue()
    this->type = glsl_type::error_type;
 }
 
+bool ir_rvalue::is_zero() const
+{
+   return false;
+}
+
+bool ir_rvalue::is_one() const
+{
+   return false;
+}
+
+bool ir_rvalue::is_negative_one() const
+{
+   return false;
+}
+
 /**
  * Modify the swizzle make to move one component to another
  *
@@ -55,6 +70,9 @@ update_rhs_swizzle(ir_swizzle_mask &m, unsigned from, unsigned to)
 void
 ir_assignment::set_lhs(ir_rvalue *lhs)
 {
+   void *mem_ctx = this;
+   bool swizzled = false;
+
    while (lhs != NULL) {
       ir_swizzle *swiz = lhs->as_swizzle();
 
@@ -82,7 +100,21 @@ ir_assignment::set_lhs(ir_rvalue *lhs)
       this->write_mask = write_mask;
       lhs = swiz->val;
 
-      this->rhs = new(this) ir_swizzle(this->rhs, rhs_swiz);
+      this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
+      swizzled = true;
+   }
+
+   if (swizzled) {
+      /* Now, RHS channels line up with the LHS writemask.  Collapse it
+       * to just the channels that will be written.
+       */
+      ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
+      int rhs_chan = 0;
+      for (int i = 0; i < 4; i++) {
+        if (write_mask & (1 << i))
+           update_rhs_swizzle(rhs_swiz, i, rhs_chan++);
+      }
+      this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
    }
 
    assert((lhs == NULL) || lhs->as_dereference());
@@ -122,6 +154,16 @@ ir_assignment::ir_assignment(ir_dereference *lhs, ir_rvalue *rhs,
    this->rhs = rhs;
    this->lhs = lhs;
    this->write_mask = write_mask;
+
+   if (lhs->type->is_scalar() || lhs->type->is_vector()) {
+      int lhs_components = 0;
+      for (int i = 0; i < 4; i++) {
+        if (write_mask & (1 << i))
+           lhs_components++;
+      }
+
+      assert(lhs_components == this->rhs->type->vector_elements);
+   }
 }
 
 ir_assignment::ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs,
@@ -149,91 +191,212 @@ ir_assignment::ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs,
 }
 
 
+ir_expression::ir_expression(int op, const struct glsl_type *type,
+                            ir_rvalue *op0)
+{
+   assert(get_num_operands(ir_expression_operation(op)) == 1);
+   this->ir_type = ir_type_expression;
+   this->type = type;
+   this->operation = ir_expression_operation(op);
+   this->operands[0] = op0;
+   this->operands[1] = NULL;
+   this->operands[2] = NULL;
+   this->operands[3] = NULL;
+}
+
 ir_expression::ir_expression(int op, const struct glsl_type *type,
                             ir_rvalue *op0, ir_rvalue *op1)
 {
+   assert(((op1 == NULL) && (get_num_operands(ir_expression_operation(op)) == 1))
+         || (get_num_operands(ir_expression_operation(op)) == 2));
    this->ir_type = ir_type_expression;
    this->type = type;
    this->operation = ir_expression_operation(op);
    this->operands[0] = op0;
    this->operands[1] = op1;
+   this->operands[2] = NULL;
+   this->operands[3] = NULL;
+}
+
+ir_expression::ir_expression(int op, const struct glsl_type *type,
+                            ir_rvalue *op0, ir_rvalue *op1,
+                            ir_rvalue *op2, ir_rvalue *op3)
+{
+   this->ir_type = ir_type_expression;
+   this->type = type;
+   this->operation = ir_expression_operation(op);
+   this->operands[0] = op0;
+   this->operands[1] = op1;
+   this->operands[2] = op2;
+   this->operands[3] = op3;
+}
+
+ir_expression::ir_expression(int op, ir_rvalue *op0)
+{
+   this->ir_type = ir_type_expression;
+
+   this->operation = ir_expression_operation(op);
+   this->operands[0] = op0;
+   this->operands[1] = NULL;
+   this->operands[2] = NULL;
+   this->operands[3] = NULL;
+
+   assert(op <= ir_last_unop);
+
+   switch (this->operation) {
+   case ir_unop_bit_not:
+   case ir_unop_logic_not:
+   case ir_unop_neg:
+   case ir_unop_abs:
+   case ir_unop_sign:
+   case ir_unop_rcp:
+   case ir_unop_rsq:
+   case ir_unop_sqrt:
+   case ir_unop_exp:
+   case ir_unop_log:
+   case ir_unop_exp2:
+   case ir_unop_log2:
+   case ir_unop_trunc:
+   case ir_unop_ceil:
+   case ir_unop_floor:
+   case ir_unop_fract:
+   case ir_unop_round_even:
+   case ir_unop_sin:
+   case ir_unop_cos:
+   case ir_unop_sin_reduced:
+   case ir_unop_cos_reduced:
+   case ir_unop_dFdx:
+   case ir_unop_dFdy:
+      this->type = op0->type;
+      break;
+
+   case ir_unop_f2i:
+   case ir_unop_b2i:
+      this->type = glsl_type::get_instance(GLSL_TYPE_INT,
+                                          op0->type->vector_elements, 1);
+      break;
+
+   case ir_unop_b2f:
+   case ir_unop_i2f:
+   case ir_unop_u2f:
+      this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT,
+                                          op0->type->vector_elements, 1);
+      break;
+
+   case ir_unop_f2b:
+   case ir_unop_i2b:
+      this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
+                                          op0->type->vector_elements, 1);
+      break;
+
+   case ir_unop_noise:
+      this->type = glsl_type::float_type;
+      break;
+
+   case ir_unop_any:
+      this->type = glsl_type::bool_type;
+      break;
+
+   default:
+      assert(!"not reached: missing automatic type setup for ir_expression");
+      this->type = op0->type;
+      break;
+   }
+}
+
+ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1)
+{
+   this->ir_type = ir_type_expression;
+
+   this->operation = ir_expression_operation(op);
+   this->operands[0] = op0;
+   this->operands[1] = op1;
+   this->operands[2] = NULL;
+   this->operands[3] = NULL;
+
+   assert(op > ir_last_unop);
+
+   switch (this->operation) {
+   case ir_binop_all_equal:
+   case ir_binop_any_nequal:
+      this->type = glsl_type::bool_type;
+      break;
+
+   case ir_binop_add:
+   case ir_binop_sub:
+   case ir_binop_min:
+   case ir_binop_max:
+   case ir_binop_pow:
+   case ir_binop_mul:
+   case ir_binop_div:
+   case ir_binop_mod:
+      if (op0->type->is_scalar()) {
+        this->type = op1->type;
+      } else if (op1->type->is_scalar()) {
+        this->type = op0->type;
+      } else {
+        /* FINISHME: matrix types */
+        assert(!op0->type->is_matrix() && !op1->type->is_matrix());
+        assert(op0->type == op1->type);
+        this->type = op0->type;
+      }
+      break;
+
+   case ir_binop_logic_and:
+   case ir_binop_logic_xor:
+   case ir_binop_logic_or:
+   case ir_binop_bit_and:
+   case ir_binop_bit_xor:
+   case ir_binop_bit_or:
+      if (op0->type->is_scalar()) {
+        this->type = op1->type;
+      } else if (op1->type->is_scalar()) {
+        this->type = op0->type;
+      }
+      break;
+
+   case ir_binop_equal:
+   case ir_binop_nequal:
+   case ir_binop_lequal:
+   case ir_binop_gequal:
+   case ir_binop_less:
+   case ir_binop_greater:
+      assert(op0->type == op1->type);
+      this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
+                                          op0->type->vector_elements, 1);
+      break;
+
+   case ir_binop_dot:
+      this->type = glsl_type::float_type;
+      break;
+
+   case ir_binop_lshift:
+   case ir_binop_rshift:
+      this->type = op0->type;
+      break;
+
+   default:
+      assert(!"not reached: missing automatic type setup for ir_expression");
+      this->type = glsl_type::float_type;
+   }
 }
 
 unsigned int
 ir_expression::get_num_operands(ir_expression_operation op)
 {
-/* Update ir_print_visitor.cpp when updating this list. */
-   const int num_operands[] = {
-      1, /* ir_unop_bit_not */
-      1, /* ir_unop_logic_not */
-      1, /* ir_unop_neg */
-      1, /* ir_unop_abs */
-      1, /* ir_unop_sign */
-      1, /* ir_unop_rcp */
-      1, /* ir_unop_rsq */
-      1, /* ir_unop_sqrt */
-      1, /* ir_unop_exp */
-      1, /* ir_unop_log */
-      1, /* ir_unop_exp2 */
-      1, /* ir_unop_log2 */
-      1, /* ir_unop_f2i */
-      1, /* ir_unop_i2f */
-      1, /* ir_unop_f2b */
-      1, /* ir_unop_b2f */
-      1, /* ir_unop_i2b */
-      1, /* ir_unop_b2i */
-      1, /* ir_unop_u2f */
-      1, /* ir_unop_any */
-
-      1, /* ir_unop_trunc */
-      1, /* ir_unop_ceil */
-      1, /* ir_unop_floor */
-      1, /* ir_unop_fract */
-
-      1, /* ir_unop_sin */
-      1, /* ir_unop_cos */
-
-      1, /* ir_unop_dFdx */
-      1, /* ir_unop_dFdy */
-
-      1, /* ir_unop_noise */
-
-      2, /* ir_binop_add */
-      2, /* ir_binop_sub */
-      2, /* ir_binop_mul */
-      2, /* ir_binop_div */
-      2, /* ir_binop_mod */
-
-      2, /* ir_binop_less */
-      2, /* ir_binop_greater */
-      2, /* ir_binop_lequal */
-      2, /* ir_binop_gequal */
-      2, /* ir_binop_equal */
-      2, /* ir_binop_nequal */
-      2, /* ir_binop_all_equal */
-      2, /* ir_binop_any_nequal */
-
-      2, /* ir_binop_lshift */
-      2, /* ir_binop_rshift */
-      2, /* ir_binop_bit_and */
-      2, /* ir_binop_bit_xor */
-      2, /* ir_binop_bit_or */
-
-      2, /* ir_binop_logic_and */
-      2, /* ir_binop_logic_xor */
-      2, /* ir_binop_logic_or */
-
-      2, /* ir_binop_dot */
-      2, /* ir_binop_cross */
-      2, /* ir_binop_min */
-      2, /* ir_binop_max */
-
-      2, /* ir_binop_pow */
-   };
+   assert(op <= ir_last_opcode);
+
+   if (op <= ir_last_unop)
+      return 1;
+
+   if (op <= ir_last_binop)
+      return 2;
 
-   assert(sizeof(num_operands) / sizeof(num_operands[0]) == ir_binop_pow + 1);
+   if (op == ir_quadop_vector)
+      return 4;
 
-   return num_operands[op];
+   assert(false);
+   return 0;
 }
 
 static const char *const operator_strs[] = {
@@ -261,8 +424,11 @@ static const char *const operator_strs[] = {
    "ceil",
    "floor",
    "fract",
+   "round_even",
    "sin",
    "cos",
+   "sin_reduced",
+   "cos_reduced",
    "dFdx",
    "dFdy",
    "noise",
@@ -288,16 +454,16 @@ static const char *const operator_strs[] = {
    "^^",
    "||",
    "dot",
-   "cross",
    "min",
    "max",
    "pow",
+   "vector",
 };
 
 const char *ir_expression::operator_string(ir_expression_operation op)
 {
    assert((unsigned int) op < Elements(operator_strs));
-   assert(Elements(operator_strs) == (ir_binop_pow + 1));
+   assert(Elements(operator_strs) == (ir_quadop_vector + 1));
    return operator_strs[op];
 }
 
@@ -306,6 +472,22 @@ const char *ir_expression::operator_string()
    return operator_string(this->operation);
 }
 
+const char*
+depth_layout_string(ir_depth_layout layout)
+{
+   switch(layout) {
+   case ir_depth_layout_none:      return "";
+   case ir_depth_layout_any:       return "depth_any";
+   case ir_depth_layout_greater:   return "depth_greater";
+   case ir_depth_layout_less:      return "depth_less";
+   case ir_depth_layout_unchanged: return "depth_unchanged";
+
+   default:
+      assert(0);
+      return "";
+   }
+}
+
 ir_expression_operation
 ir_expression::get_operator(const char *str)
 {
@@ -396,7 +578,7 @@ ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
          || type->is_record() || type->is_array());
 
    if (type->is_array()) {
-      this->array_elements = talloc_array(this, ir_constant *, type->length);
+      this->array_elements = ralloc_array(this, ir_constant *, type->length);
       unsigned i = 0;
       foreach_list(node, value_list) {
         ir_constant *value = (ir_constant *) node;
@@ -660,7 +842,7 @@ ir_constant::has_value(const ir_constant *c) const
 
    if (this->type->is_array()) {
       for (unsigned i = 0; i < this->type->length; i++) {
-        if (this->array_elements[i]->has_value(c->array_elements[i]))
+        if (!this->array_elements[i]->has_value(c->array_elements[i]))
            return false;
       }
       return true;
@@ -713,6 +895,115 @@ ir_constant::has_value(const ir_constant *c) const
    return true;
 }
 
+bool
+ir_constant::is_zero() const
+{
+   if (!this->type->is_scalar() && !this->type->is_vector())
+      return false;
+
+   for (unsigned c = 0; c < this->type->vector_elements; c++) {
+      switch (this->type->base_type) {
+      case GLSL_TYPE_FLOAT:
+        if (this->value.f[c] != 0.0)
+           return false;
+        break;
+      case GLSL_TYPE_INT:
+        if (this->value.i[c] != 0)
+           return false;
+        break;
+      case GLSL_TYPE_UINT:
+        if (this->value.u[c] != 0)
+           return false;
+        break;
+      case GLSL_TYPE_BOOL:
+        if (this->value.b[c] != false)
+           return false;
+        break;
+      default:
+        /* The only other base types are structures, arrays, and samplers.
+         * Samplers cannot be constants, and the others should have been
+         * filtered out above.
+         */
+        assert(!"Should not get here.");
+        return false;
+      }
+   }
+
+   return true;
+}
+
+bool
+ir_constant::is_one() const
+{
+   if (!this->type->is_scalar() && !this->type->is_vector())
+      return false;
+
+   for (unsigned c = 0; c < this->type->vector_elements; c++) {
+      switch (this->type->base_type) {
+      case GLSL_TYPE_FLOAT:
+        if (this->value.f[c] != 1.0)
+           return false;
+        break;
+      case GLSL_TYPE_INT:
+        if (this->value.i[c] != 1)
+           return false;
+        break;
+      case GLSL_TYPE_UINT:
+        if (this->value.u[c] != 1)
+           return false;
+        break;
+      case GLSL_TYPE_BOOL:
+        if (this->value.b[c] != true)
+           return false;
+        break;
+      default:
+        /* The only other base types are structures, arrays, and samplers.
+         * Samplers cannot be constants, and the others should have been
+         * filtered out above.
+         */
+        assert(!"Should not get here.");
+        return false;
+      }
+   }
+
+   return true;
+}
+
+bool
+ir_constant::is_negative_one() const
+{
+   if (!this->type->is_scalar() && !this->type->is_vector())
+      return false;
+
+   if (this->type->is_boolean())
+      return false;
+
+   for (unsigned c = 0; c < this->type->vector_elements; c++) {
+      switch (this->type->base_type) {
+      case GLSL_TYPE_FLOAT:
+        if (this->value.f[c] != -1.0)
+           return false;
+        break;
+      case GLSL_TYPE_INT:
+        if (this->value.i[c] != -1)
+           return false;
+        break;
+      case GLSL_TYPE_UINT:
+        if (int(this->value.u[c]) != -1)
+           return false;
+        break;
+      default:
+        /* The only other base types are structures, arrays, samplers, and
+         * booleans.  Samplers cannot be constants, and the others should
+         * have been filtered out above.
+         */
+        assert(!"Should not get here.");
+        return false;
+      }
+   }
+
+   return true;
+}
 
 ir_loop::ir_loop()
 {
@@ -745,7 +1036,7 @@ ir_dereference_array::ir_dereference_array(ir_rvalue *value,
 ir_dereference_array::ir_dereference_array(ir_variable *var,
                                           ir_rvalue *array_index)
 {
-   void *ctx = talloc_parent(var);
+   void *ctx = ralloc_parent(var);
 
    this->ir_type = ir_type_dereference_array;
    this->array_index = array_index;
@@ -778,7 +1069,7 @@ ir_dereference_record::ir_dereference_record(ir_rvalue *value,
 {
    this->ir_type = ir_type_dereference_record;
    this->record = value;
-   this->field = talloc_strdup(this, field);
+   this->field = ralloc_strdup(this, field);
    this->type = (this->record != NULL)
       ? this->record->type->field_type(field) : glsl_type::error_type;
 }
@@ -787,11 +1078,11 @@ ir_dereference_record::ir_dereference_record(ir_rvalue *value,
 ir_dereference_record::ir_dereference_record(ir_variable *var,
                                             const char *field)
 {
-   void *ctx = talloc_parent(var);
+   void *ctx = ralloc_parent(var);
 
    this->ir_type = ir_type_dereference_record;
    this->record = new(ctx) ir_dereference_variable(var);
-   this->field = talloc_strdup(this, field);
+   this->field = ralloc_strdup(this, field);
    this->type = (this->record != NULL)
       ? this->record->type->field_type(field) : glsl_type::error_type;
 }
@@ -954,7 +1245,7 @@ ir_swizzle::ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask)
 ir_swizzle *
 ir_swizzle::create(ir_rvalue *val, const char *str, unsigned vector_length)
 {
-   void *ctx = talloc_parent(val);
+   void *ctx = ralloc_parent(val);
 
    /* For each possible swizzle character, this table encodes the value in
     * \c idx_map that represents the 0th element of the vector.  For invalid
@@ -1043,12 +1334,15 @@ ir_variable::ir_variable(const struct glsl_type *type, const char *name,
 {
    this->ir_type = ir_type_variable;
    this->type = type;
-   this->name = talloc_strdup(this, name);
+   this->name = ralloc_strdup(this, name);
+   this->explicit_location = false;
    this->location = -1;
    this->warn_extension = NULL;
    this->constant_value = NULL;
    this->origin_upper_left = false;
    this->pixel_center_integer = false;
+   this->depth_layout = ir_depth_layout_none;
+   this->used = false;
 
    if (type && type->base_type == GLSL_TYPE_SAMPLER)
       this->read_only = true;
@@ -1085,6 +1379,21 @@ ir_function_signature::ir_function_signature(const glsl_type *return_type)
 }
 
 
+static bool
+modes_match(unsigned a, unsigned b)
+{
+   if (a == b)
+      return true;
+
+   /* Accept "in" vs. "const in" */
+   if ((a == ir_var_const_in && b == ir_var_in) ||
+       (b == ir_var_const_in && a == ir_var_in))
+      return true;
+
+   return false;
+}
+
+
 const char *
 ir_function_signature::qualifiers_match(exec_list *params)
 {
@@ -1097,7 +1406,7 @@ ir_function_signature::qualifiers_match(exec_list *params)
       ir_variable *b = (ir_variable *)iter_b.get();
 
       if (a->read_only != b->read_only ||
-         a->mode != b->mode ||
+         !modes_match(a->mode, b->mode) ||
          a->interpolation != b->interpolation ||
          a->centroid != b->centroid) {
 
@@ -1132,7 +1441,7 @@ ir_function_signature::replace_parameters(exec_list *new_params)
 ir_function::ir_function(const char *name)
 {
    this->ir_type = ir_type_function;
-   this->name = talloc_strdup(this, name);
+   this->name = ralloc_strdup(this, name);
 }
 
 
@@ -1198,7 +1507,7 @@ steal_memory(ir_instruction *ir, void *new_ctx)
       }
    }
 
-   talloc_steal(new_ctx, ir);
+   ralloc_steal(new_ctx, ir);
 }
 
 
@@ -1209,3 +1518,59 @@ reparent_ir(exec_list *list, void *mem_ctx)
       visit_tree((ir_instruction *) node, steal_memory, mem_ctx);
    }
 }
+
+
+static ir_rvalue *
+try_min_one(ir_rvalue *ir)
+{
+   ir_expression *expr = ir->as_expression();
+
+   if (!expr || expr->operation != ir_binop_min)
+      return NULL;
+
+   if (expr->operands[0]->is_one())
+      return expr->operands[1];
+
+   if (expr->operands[1]->is_one())
+      return expr->operands[0];
+
+   return NULL;
+}
+
+static ir_rvalue *
+try_max_zero(ir_rvalue *ir)
+{
+   ir_expression *expr = ir->as_expression();
+
+   if (!expr || expr->operation != ir_binop_max)
+      return NULL;
+
+   if (expr->operands[0]->is_zero())
+      return expr->operands[1];
+
+   if (expr->operands[1]->is_zero())
+      return expr->operands[0];
+
+   return NULL;
+}
+
+ir_rvalue *
+ir_rvalue::as_rvalue_to_saturate()
+{
+   ir_expression *expr = this->as_expression();
+
+   if (!expr)
+      return NULL;
+
+   ir_rvalue *max_zero = try_max_zero(expr);
+   if (max_zero) {
+      return try_min_one(max_zero);
+   } else {
+      ir_rvalue *min_one = try_min_one(expr);
+      if (min_one) {
+        return try_max_zero(min_one);
+      }
+   }
+
+   return NULL;
+}