X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fglsl%2Fir.cpp;h=d6594cd9a3e65063b4a72754aa8781d2100d3c7c;hb=440dfb45834e18ebd9acfc37aa31797d8803e51b;hp=d3f7302b54f47adf14dc2a66f2ce408f2f5b58e1;hpb=74e1802f5dd8921750851abc6128e4073602d405;p=mesa.git

diff --git a/src/glsl/ir.cpp b/src/glsl/ir.cpp
index d3f7302b54f..d6594cd9a3e 100644
--- a/src/glsl/ir.cpp
+++ b/src/glsl/ir.cpp
@@ -21,101 +21,388 @@
  * DEALINGS IN THE SOFTWARE.
  */
 #include <string.h>
-#include "main/imports.h"
+#include "main/core.h" /* for MAX2 */
 #include "ir.h"
 #include "ir_visitor.h"
 #include "glsl_types.h"
 
+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
+ *
+ * \param m    IR swizzle to be modified
+ * \param from Component in the RHS that is to be swizzled
+ * \param to   Desired swizzle location of \c from
+ */
+static void
+update_rhs_swizzle(ir_swizzle_mask &m, unsigned from, unsigned to)
+{
+   switch (to) {
+   case 0: m.x = from; break;
+   case 1: m.y = from; break;
+   case 2: m.z = from; break;
+   case 3: m.w = from; break;
+   default: assert(!"Should not get here.");
+   }
+
+   m.num_components = MAX2(m.num_components, (to + 1));
+}
+
+void
+ir_assignment::set_lhs(ir_rvalue *lhs)
+{
+   void *mem_ctx = this;
+   bool swizzled = false;
+
+   while (lhs != NULL) {
+      ir_swizzle *swiz = lhs->as_swizzle();
+
+      if (swiz == NULL)
+	 break;
+
+      unsigned write_mask = 0;
+      ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
+
+      for (unsigned i = 0; i < swiz->mask.num_components; i++) {
+	 unsigned c = 0;
+
+	 switch (i) {
+	 case 0: c = swiz->mask.x; break;
+	 case 1: c = swiz->mask.y; break;
+	 case 2: c = swiz->mask.z; break;
+	 case 3: c = swiz->mask.w; break;
+	 default: assert(!"Should not get here.");
+	 }
+
+	 write_mask |= (((this->write_mask >> i) & 1) << c);
+	 update_rhs_swizzle(rhs_swiz, i, c);
+      }
+
+      this->write_mask = write_mask;
+      lhs = swiz->val;
+
+      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());
+
+   this->lhs = (ir_dereference *) lhs;
+}
+
+ir_variable *
+ir_assignment::whole_variable_written()
+{
+   ir_variable *v = this->lhs->whole_variable_referenced();
+
+   if (v == NULL)
+      return NULL;
+
+   if (v->type->is_scalar())
+      return v;
+
+   if (v->type->is_vector()) {
+      const unsigned mask = (1U << v->type->vector_elements) - 1;
+
+      if (mask != this->write_mask)
+	 return NULL;
+   }
+
+   /* Either all the vector components are assigned or the variable is some
+    * composite type (and the whole thing is assigned.
+    */
+   return v;
+}
+
+ir_assignment::ir_assignment(ir_dereference *lhs, ir_rvalue *rhs,
+			     ir_rvalue *condition, unsigned write_mask)
+{
+   this->ir_type = ir_type_assignment;
+   this->condition = condition;
+   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,
 			     ir_rvalue *condition)
 {
    this->ir_type = ir_type_assignment;
-   this->lhs = lhs;
-   this->rhs = rhs;
    this->condition = condition;
+   this->rhs = rhs;
+
+   /* If the RHS is a vector type, assume that all components of the vector
+    * type are being written to the LHS.  The write mask comes from the RHS
+    * because we can have a case where the LHS is a vec4 and the RHS is a
+    * vec3.  In that case, the assignment is:
+    *
+    *     (assign (...) (xyz) (var_ref lhs) (var_ref rhs))
+    */
+   if (rhs->type->is_vector())
+      this->write_mask = (1U << rhs->type->vector_elements) - 1;
+   else if (rhs->type->is_scalar())
+      this->write_mask = 1;
+   else
+      this->write_mask = 0;
+
+   this->set_lhs(lhs);
 }
 
 
+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:
+   case ir_unop_u2i:
+      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_i2u:
+      this->type = glsl_type::get_instance(GLSL_TYPE_UINT,
+					   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_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 */
-
-      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_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;
 
-   assert(sizeof(num_operands) / sizeof(num_operands[0]) == ir_binop_pow + 1);
+   if (op <= ir_last_binop)
+      return 2;
 
-   return num_operands[op];
+   if (op == ir_quadop_vector)
+      return 4;
+
+   assert(false);
+   return 0;
 }
 
 static const char *const operator_strs[] = {
@@ -138,14 +425,21 @@ static const char *const operator_strs[] = {
    "i2b",
    "b2i",
    "u2f",
+   "i2u",
+   "u2i",
+   "any",
    "trunc",
    "ceil",
    "floor",
    "fract",
+   "round_even",
    "sin",
    "cos",
+   "sin_reduced",
+   "cos_reduced",
    "dFdx",
    "dFdy",
+   "noise",
    "+",
    "-",
    "*",
@@ -157,6 +451,8 @@ static const char *const operator_strs[] = {
    ">=",
    "==",
    "!=",
+   "all_equal",
+   "any_nequal",
    "<<",
    ">>",
    "&",
@@ -166,17 +462,38 @@ 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_quadop_vector + 1));
+   return operator_strs[op];
+}
+
 const char *ir_expression::operator_string()
 {
-   assert((unsigned int) operation <=
-	  sizeof(operator_strs) / sizeof(operator_strs[0]));
-   return operator_strs[operation];
+   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
@@ -211,6 +528,9 @@ ir_constant::ir_constant(float f)
    this->ir_type = ir_type_constant;
    this->type = glsl_type::float_type;
    this->value.f[0] = f;
+   for (int i = 1; i < 16; i++)  {
+      this->value.f[i] = 0;
+   }
 }
 
 ir_constant::ir_constant(unsigned int u)
@@ -218,6 +538,9 @@ ir_constant::ir_constant(unsigned int u)
    this->ir_type = ir_type_constant;
    this->type = glsl_type::uint_type;
    this->value.u[0] = u;
+   for (int i = 1; i < 16; i++) {
+      this->value.u[i] = 0;
+   }
 }
 
 ir_constant::ir_constant(int i)
@@ -225,6 +548,9 @@ ir_constant::ir_constant(int i)
    this->ir_type = ir_type_constant;
    this->type = glsl_type::int_type;
    this->value.i[0] = i;
+   for (int i = 1; i < 16; i++) {
+      this->value.i[i] = 0;
+   }
 }
 
 ir_constant::ir_constant(bool b)
@@ -232,6 +558,9 @@ ir_constant::ir_constant(bool b)
    this->ir_type = ir_type_constant;
    this->type = glsl_type::bool_type;
    this->value.b[0] = b;
+   for (int i = 1; i < 16; i++) {
+      this->value.b[i] = false;
+   }
 }
 
 ir_constant::ir_constant(const ir_constant *c, unsigned i)
@@ -257,7 +586,7 @@ 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;
@@ -282,15 +611,78 @@ ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
       return;
    }
 
+   for (unsigned i = 0; i < 16; i++) {
+      this->value.u[i] = 0;
+   }
 
    ir_constant *value = (ir_constant *) (value_list->head);
 
+   /* Constructors with exactly one scalar argument are special for vectors
+    * and matrices.  For vectors, the scalar value is replicated to fill all
+    * the components.  For matrices, the scalar fills the components of the
+    * diagonal while the rest is filled with 0.
+    */
+   if (value->type->is_scalar() && value->next->is_tail_sentinel()) {
+      if (type->is_matrix()) {
+	 /* Matrix - fill diagonal (rest is already set to 0) */
+	 assert(type->base_type == GLSL_TYPE_FLOAT);
+	 for (unsigned i = 0; i < type->matrix_columns; i++)
+	    this->value.f[i * type->vector_elements + i] = value->value.f[0];
+      } else {
+	 /* Vector or scalar - fill all components */
+	 switch (type->base_type) {
+	 case GLSL_TYPE_UINT:
+	 case GLSL_TYPE_INT:
+	    for (unsigned i = 0; i < type->components(); i++)
+	       this->value.u[i] = value->value.u[0];
+	    break;
+	 case GLSL_TYPE_FLOAT:
+	    for (unsigned i = 0; i < type->components(); i++)
+	       this->value.f[i] = value->value.f[0];
+	    break;
+	 case GLSL_TYPE_BOOL:
+	    for (unsigned i = 0; i < type->components(); i++)
+	       this->value.b[i] = value->value.b[0];
+	    break;
+	 default:
+	    assert(!"Should not get here.");
+	    break;
+	 }
+      }
+      return;
+   }
+
+   if (type->is_matrix() && value->type->is_matrix()) {
+      assert(value->next->is_tail_sentinel());
+
+      /* From section 5.4.2 of the GLSL 1.20 spec:
+       * "If a matrix is constructed from a matrix, then each component
+       *  (column i, row j) in the result that has a corresponding component
+       *  (column i, row j) in the argument will be initialized from there."
+       */
+      unsigned cols = MIN2(type->matrix_columns, value->type->matrix_columns);
+      unsigned rows = MIN2(type->vector_elements, value->type->vector_elements);
+      for (unsigned i = 0; i < cols; i++) {
+	 for (unsigned j = 0; j < rows; j++) {
+	    const unsigned src = i * value->type->vector_elements + j;
+	    const unsigned dst = i * type->vector_elements + j;
+	    this->value.f[dst] = value->value.f[src];
+	 }
+      }
+
+      /* "All other components will be initialized to the identity matrix." */
+      for (unsigned i = cols; i < type->matrix_columns; i++)
+	 this->value.f[i * type->vector_elements + i] = 1.0;
+
+      return;
+   }
+
    /* Use each component from each entry in the value_list to initialize one
     * component of the constant being constructed.
     */
    for (unsigned i = 0; i < type->components(); /* empty */) {
       assert(value->as_constant() != NULL);
-      assert(!value->is_tail_sentinal());
+      assert(!value->is_tail_sentinel());
 
       for (unsigned j = 0; j < value->type->components(); j++) {
 	 switch (type->base_type) {
@@ -321,6 +713,18 @@ ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
    }
 }
 
+ir_constant *
+ir_constant::zero(void *mem_ctx, const glsl_type *type)
+{
+   assert(type->is_numeric() || type->is_boolean());
+
+   ir_constant *c = new(mem_ctx) ir_constant;
+   c->type = type;
+   memset(&c->value, 0, sizeof(c->value));
+
+   return c;
+}
+
 bool
 ir_constant::get_bool_component(unsigned i) const
 {
@@ -393,7 +797,21 @@ ir_constant *
 ir_constant::get_array_element(unsigned i) const
 {
    assert(this->type->is_array());
-   assert(i < this->type->length);
+
+   /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
+    *
+    *     "Behavior is undefined if a shader subscripts an array with an index
+    *     less than 0 or greater than or equal to the size the array was
+    *     declared with."
+    *
+    * Most out-of-bounds accesses are removed before things could get this far.
+    * There are cases where non-constant array index values can get constant
+    * folded.
+    */
+   if (int(i) < 0)
+      i = 0;
+   else if (i >= this->type->length)
+      i = this->type->length - 1;
 
    return array_elements[i];
 }
@@ -416,7 +834,7 @@ ir_constant::get_record_field(const char *name)
       /* If the end of the list is encountered before the element matching the
        * requested field is found, return NULL.
        */
-      if (node->is_tail_sentinal())
+      if (node->is_tail_sentinel())
 	 return NULL;
    }
 
@@ -430,15 +848,20 @@ ir_constant::has_value(const ir_constant *c) const
    if (this->type != c->type)
       return false;
 
-   /* FINISHME: This will probably also handle constant arrays as soon as those
-    * FINISHME: are supported.
-    */
+   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]))
+	    return false;
+      }
+      return true;
+   }
+
    if (this->type->base_type == GLSL_TYPE_STRUCT) {
       const exec_node *a_node = this->components.head;
       const exec_node *b_node = c->components.head;
 
-      while (!a_node->is_tail_sentinal()) {
-	 assert(!b_node->is_tail_sentinal());
+      while (!a_node->is_tail_sentinel()) {
+	 assert(!b_node->is_tail_sentinel());
 
 	 const ir_constant *const a_field = (ir_constant *) a_node;
 	 const ir_constant *const b_field = (ir_constant *) b_node;
@@ -480,6 +903,127 @@ 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()
+{
+   this->ir_type = ir_type_loop;
+   this->cmp = ir_unop_neg;
+   this->from = NULL;
+   this->to = NULL;
+   this->increment = NULL;
+   this->counter = NULL;
+}
+
+
 ir_dereference_variable::ir_dereference_variable(ir_variable *var)
 {
    this->ir_type = ir_type_dereference_variable;
@@ -500,7 +1044,7 @@ 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;
@@ -533,7 +1077,7 @@ 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;
 }
@@ -542,18 +1086,17 @@ 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;
 }
 
-
 bool
-ir_dereference::is_lvalue()
+ir_dereference::is_lvalue() const
 {
    ir_variable *var = this->variable_referenced();
 
@@ -562,14 +1105,20 @@ ir_dereference::is_lvalue()
    if ((var == NULL) || var->read_only)
       return false;
 
-   if (this->type->is_array() && !var->array_lvalue)
+   /* From page 17 (page 23 of the PDF) of the GLSL 1.20 spec:
+    *
+    *    "Samplers cannot be treated as l-values; hence cannot be used
+    *     as out or inout function parameters, nor can they be
+    *     assigned into."
+    */
+   if (this->type->contains_sampler())
       return false;
 
    return true;
 }
 
 
-const char *tex_opcode_strs[] = { "tex", "txb", "txl", "txd", "txf" };
+const char *tex_opcode_strs[] = { "tex", "txb", "txl", "txd", "txf", "txs" };
 
 const char *ir_texture::opcode_string()
 {
@@ -591,21 +1140,21 @@ ir_texture::get_opcode(const char *str)
 
 
 void
-ir_texture::set_sampler(ir_dereference *sampler)
+ir_texture::set_sampler(ir_dereference *sampler, const glsl_type *type)
 {
    assert(sampler != NULL);
+   assert(type != NULL);
    this->sampler = sampler;
+   this->type = type;
 
-   switch (sampler->type->sampler_type) {
-   case GLSL_TYPE_FLOAT:
-      this->type = glsl_type::vec4_type;
-      break;
-   case GLSL_TYPE_INT:
-      this->type = glsl_type::ivec4_type;
-      break;
-   case GLSL_TYPE_UINT:
-      this->type = glsl_type::uvec4_type;
-      break;
+   if (this->op == ir_txs) {
+      assert(type->base_type == GLSL_TYPE_INT);
+   } else {
+      assert(sampler->type->sampler_type == (int) type->base_type);
+      if (sampler->type->sampler_shadow)
+	 assert(type->vector_elements == 4 || type->vector_elements == 1);
+      else
+	 assert(type->vector_elements == 4);
    }
 }
 
@@ -686,7 +1235,7 @@ 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
@@ -762,7 +1311,7 @@ ir_swizzle::create(ir_rvalue *val, const char *str, unsigned vector_length)
 #undef I
 
 ir_variable *
-ir_swizzle::variable_referenced()
+ir_swizzle::variable_referenced() const
 {
    return this->val->variable_referenced();
 }
@@ -771,15 +1320,19 @@ ir_swizzle::variable_referenced()
 ir_variable::ir_variable(const struct glsl_type *type, const char *name,
 			 ir_variable_mode mode)
    : max_array_access(0), read_only(false), centroid(false), invariant(false),
-     shader_in(false), shader_out(false),
-     mode(mode), interpolation(ir_var_smooth), array_lvalue(false)
+     mode(mode), interpolation(ir_var_smooth)
 {
    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;
@@ -789,9 +1342,6 @@ ir_variable::ir_variable(const struct glsl_type *type, const char *name,
 const char *
 ir_variable::interpolation_string() const
 {
-   if (!this->shader_in && !this->shader_out)
-      return "";
-
    switch (this->interpolation) {
    case ir_var_smooth:        return "smooth";
    case ir_var_flat:          return "flat";
@@ -815,6 +1365,22 @@ ir_function_signature::ir_function_signature(const glsl_type *return_type)
    : return_type(return_type), is_defined(false), _function(NULL)
 {
    this->ir_type = ir_type_function_signature;
+   this->is_builtin = false;
+}
+
+
+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;
 }
 
 
@@ -830,7 +1396,7 @@ 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) {
 
@@ -865,7 +1431,19 @@ 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);
+}
+
+
+bool
+ir_function::has_user_signature()
+{
+   foreach_list(n, &this->signatures) {
+      ir_function_signature *const sig = (ir_function_signature *) n;
+      if (!sig->is_builtin)
+	 return true;
+   }
+   return false;
 }
 
 
@@ -898,7 +1476,28 @@ visit_exec_list(exec_list *list, ir_visitor *visitor)
 static void
 steal_memory(ir_instruction *ir, void *new_ctx)
 {
-   talloc_steal(new_ctx, ir);
+   ir_variable *var = ir->as_variable();
+   ir_constant *constant = ir->as_constant();
+   if (var != NULL && var->constant_value != NULL)
+      steal_memory(var->constant_value, ir);
+
+   /* The components of aggregate constants are not visited by the normal
+    * visitor, so steal their values by hand.
+    */
+   if (constant != NULL) {
+      if (constant->type->is_record()) {
+	 foreach_iter(exec_list_iterator, iter, constant->components) {
+	    ir_constant *field = (ir_constant *)iter.get();
+	    steal_memory(field, ir);
+	 }
+      } else if (constant->type->is_array()) {
+	 for (unsigned int i = 0; i < constant->type->length; i++) {
+	    steal_memory(constant->array_elements[i], ir);
+	 }
+      }
+   }
+
+   ralloc_steal(new_ctx, ir);
 }
 
 
@@ -909,3 +1508,59 @@ 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;
+}