svn merge ^/branches/released .

19 files changed, 3351 insertions, 3300 deletions

diff --git a/mesalib/src/glsl/ir_constant_expression.cpp b/mesalib/src/glsl/ir_constant_expression.cpp
index 2841fb350..1d39e31fa 100644
--- a/mesalib/src/glsl/ir_constant_expression.cpp
+++ b/mesalib/src/glsl/ir_constant_expression.cpp
@@ -1,1376 +1,1372 @@
-/*
- * Copyright © 2010 Intel Corporation
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
- * DEALINGS IN THE SOFTWARE.
- */
-
-/**
- * \file ir_constant_expression.cpp
- * Evaluate and process constant valued expressions
- *
- * In GLSL, constant valued expressions are used in several places.  These
- * must be processed and evaluated very early in the compilation process.
- *
- *    * Sizes of arrays
- *    * Initializers for uniforms
- *    * Initializers for \c const variables
- */
-
-#include <math.h>
-#include "main/core.h" /* for MAX2, MIN2, CLAMP */
-#include "ir.h"
-#include "ir_visitor.h"
-#include "glsl_types.h"
-
-static float
-dot(ir_constant *op0, ir_constant *op1)
-{
-   assert(op0->type->is_float() && op1->type->is_float());
-
-   float result = 0;
-   for (unsigned c = 0; c < op0->type->components(); c++)
-      result += op0->value.f[c] * op1->value.f[c];
-
-   return result;
-}
-
-ir_constant *
-ir_expression::constant_expression_value()
-{
-   if (this->type->is_error())
-      return NULL;
-
-   ir_constant *op[Elements(this->operands)] = { NULL, };
-   ir_constant_data data;
-
-   memset(&data, 0, sizeof(data));
-
-   for (unsigned operand = 0; operand < this->get_num_operands(); operand++) {
-      op[operand] = this->operands[operand]->constant_expression_value();
-      if (!op[operand])
-	 return NULL;
-   }
-
-   if (op[1] != NULL)
-      assert(op[0]->type->base_type == op[1]->type->base_type);
-
-   bool op0_scalar = op[0]->type->is_scalar();
-   bool op1_scalar = op[1] != NULL && op[1]->type->is_scalar();
-
-   /* When iterating over a vector or matrix's components, we want to increase
-    * the loop counter.  However, for scalars, we want to stay at 0.
-    */
-   unsigned c0_inc = op0_scalar ? 0 : 1;
-   unsigned c1_inc = op1_scalar ? 0 : 1;
-   unsigned components;
-   if (op1_scalar || !op[1]) {
-      components = op[0]->type->components();
-   } else {
-      components = op[1]->type->components();
-   }
-
-   void *ctx = ralloc_parent(this);
-
-   /* Handle array operations here, rather than below. */
-   if (op[0]->type->is_array()) {
-      assert(op[1] != NULL && op[1]->type->is_array());
-      switch (this->operation) {
-      case ir_binop_all_equal:
-	 return new(ctx) ir_constant(op[0]->has_value(op[1]));
-      case ir_binop_any_nequal:
-	 return new(ctx) ir_constant(!op[0]->has_value(op[1]));
-      default:
-	 break;
-      }
-      return NULL;
-   }
-
-   switch (this->operation) {
-   case ir_unop_bit_not:
-       switch (op[0]->type->base_type) {
-       case GLSL_TYPE_INT:
-           for (unsigned c = 0; c < components; c++)
-               data.i[c] = ~ op[0]->value.i[c];
-           break;
-       case GLSL_TYPE_UINT:
-           for (unsigned c = 0; c < components; c++)
-               data.u[c] = ~ op[0]->value.u[c];
-           break;
-       default:
-           assert(0);
-       }
-       break;
-
-   case ir_unop_logic_not:
-      assert(op[0]->type->base_type == GLSL_TYPE_BOOL);
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.b[c] = !op[0]->value.b[c];
-      break;
-
-   case ir_unop_f2i:
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.i[c] = (int) op[0]->value.f[c];
-      }
-      break;
-   case ir_unop_i2f:
-      assert(op[0]->type->base_type == GLSL_TYPE_INT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.f[c] = (float) op[0]->value.i[c];
-      }
-      break;
-   case ir_unop_u2f:
-      assert(op[0]->type->base_type == GLSL_TYPE_UINT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.f[c] = (float) op[0]->value.u[c];
-      }
-      break;
-   case ir_unop_b2f:
-      assert(op[0]->type->base_type == GLSL_TYPE_BOOL);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.f[c] = op[0]->value.b[c] ? 1.0F : 0.0F;
-      }
-      break;
-   case ir_unop_f2b:
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.b[c] = op[0]->value.f[c] != 0.0F ? true : false;
-      }
-      break;
-   case ir_unop_b2i:
-      assert(op[0]->type->base_type == GLSL_TYPE_BOOL);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.u[c] = op[0]->value.b[c] ? 1 : 0;
-      }
-      break;
-   case ir_unop_i2b:
-      assert(op[0]->type->is_integer());
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.b[c] = op[0]->value.u[c] ? true : false;
-      }
-      break;
-
-   case ir_unop_any:
-      assert(op[0]->type->is_boolean());
-      data.b[0] = false;
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 if (op[0]->value.b[c])
-	    data.b[0] = true;
-      }
-      break;
-
-   case ir_unop_trunc:
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.f[c] = truncf(op[0]->value.f[c]);
-      }
-      break;
-
-   case ir_unop_ceil:
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.f[c] = ceilf(op[0]->value.f[c]);
-      }
-      break;
-
-   case ir_unop_floor:
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.f[c] = floorf(op[0]->value.f[c]);
-      }
-      break;
-
-   case ir_unop_fract:
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 switch (this->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.u[c] = 0;
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.i[c] = 0;
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.f[c] = op[0]->value.f[c] - floor(op[0]->value.f[c]);
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-      break;
-
-   case ir_unop_sin:
-   case ir_unop_sin_reduced:
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.f[c] = sinf(op[0]->value.f[c]);
-      }
-      break;
-
-   case ir_unop_cos:
-   case ir_unop_cos_reduced:
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.f[c] = cosf(op[0]->value.f[c]);
-      }
-      break;
-
-   case ir_unop_neg:
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 switch (this->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.u[c] = -((int) op[0]->value.u[c]);
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.i[c] = -op[0]->value.i[c];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.f[c] = -op[0]->value.f[c];
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-      break;
-
-   case ir_unop_abs:
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 switch (this->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.u[c] = op[0]->value.u[c];
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.i[c] = op[0]->value.i[c];
-	    if (data.i[c] < 0)
-	       data.i[c] = -data.i[c];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.f[c] = fabs(op[0]->value.f[c]);
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-      break;
-
-   case ir_unop_sign:
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 switch (this->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.u[c] = op[0]->value.i[c] > 0;
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.i[c] = (op[0]->value.i[c] > 0) - (op[0]->value.i[c] < 0);
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.f[c] = float((op[0]->value.f[c] > 0)-(op[0]->value.f[c] < 0));
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-      break;
-
-   case ir_unop_rcp:
-      /* FINISHME: Emit warning when division-by-zero is detected. */
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 switch (this->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    if (op[0]->value.u[c] == 0.0)
-	       return NULL;
-	    data.u[c] = 1 / op[0]->value.u[c];
-	    break;
-	 case GLSL_TYPE_INT:
-	    if (op[0]->value.i[c] == 0.0)
-	       return NULL;
-	    data.i[c] = 1 / op[0]->value.i[c];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    if (op[0]->value.f[c] == 0.0)
-	       return NULL;
-	    data.f[c] = 1.0F / op[0]->value.f[c];
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-      break;
-
-   case ir_unop_rsq:
-      /* FINISHME: Emit warning when division-by-zero is detected. */
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 float s = sqrtf(op[0]->value.f[c]);
-	 if (s == 0)
-	    return NULL;
-	 data.f[c] = 1.0F / s;
-      }
-      break;
-
-   case ir_unop_sqrt:
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.f[c] = sqrtf(op[0]->value.f[c]);
-      }
-      break;
-
-   case ir_unop_exp:
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.f[c] = expf(op[0]->value.f[c]);
-      }
-      break;
-
-   case ir_unop_exp2:
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.f[c] = exp2f(op[0]->value.f[c]);
-      }
-      break;
-
-   case ir_unop_log:
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.f[c] = logf(op[0]->value.f[c]);
-      }
-      break;
-
-   case ir_unop_log2:
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.f[c] = log2f(op[0]->value.f[c]);
-      }
-      break;
-
-   case ir_unop_dFdx:
-   case ir_unop_dFdy:
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.f[c] = 0.0;
-      }
-      break;
-
-   case ir_binop_pow:
-      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 data.f[c] = powf(op[0]->value.f[c], op[1]->value.f[c]);
-      }
-      break;
-
-   case ir_binop_dot:
-      data.f[0] = dot(op[0], op[1]);
-      break;
-
-   case ir_binop_min:
-      assert(op[0]->type == op[1]->type || op0_scalar || op1_scalar);
-      for (unsigned c = 0, c0 = 0, c1 = 0;
-	   c < components;
-	   c0 += c0_inc, c1 += c1_inc, c++) {
-
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.u[c] = MIN2(op[0]->value.u[c0], op[1]->value.u[c1]);
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.i[c] = MIN2(op[0]->value.i[c0], op[1]->value.i[c1]);
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.f[c] = MIN2(op[0]->value.f[c0], op[1]->value.f[c1]);
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-
-      break;
-   case ir_binop_max:
-      assert(op[0]->type == op[1]->type || op0_scalar || op1_scalar);
-      for (unsigned c = 0, c0 = 0, c1 = 0;
-	   c < components;
-	   c0 += c0_inc, c1 += c1_inc, c++) {
-
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.u[c] = MAX2(op[0]->value.u[c0], op[1]->value.u[c1]);
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.i[c] = MAX2(op[0]->value.i[c0], op[1]->value.i[c1]);
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.f[c] = MAX2(op[0]->value.f[c0], op[1]->value.f[c1]);
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-      break;
-
-   case ir_binop_add:
-      assert(op[0]->type == op[1]->type || op0_scalar || op1_scalar);
-      for (unsigned c = 0, c0 = 0, c1 = 0;
-	   c < components;
-	   c0 += c0_inc, c1 += c1_inc, c++) {
-
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.u[c] = op[0]->value.u[c0] + op[1]->value.u[c1];
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.i[c] = op[0]->value.i[c0] + op[1]->value.i[c1];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.f[c] = op[0]->value.f[c0] + op[1]->value.f[c1];
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-
-      break;
-   case ir_binop_sub:
-      assert(op[0]->type == op[1]->type || op0_scalar || op1_scalar);
-      for (unsigned c = 0, c0 = 0, c1 = 0;
-	   c < components;
-	   c0 += c0_inc, c1 += c1_inc, c++) {
-
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.u[c] = op[0]->value.u[c0] - op[1]->value.u[c1];
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.i[c] = op[0]->value.i[c0] - op[1]->value.i[c1];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.f[c] = op[0]->value.f[c0] - op[1]->value.f[c1];
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-
-      break;
-   case ir_binop_mul:
-      /* Check for equal types, or unequal types involving scalars */
-      if ((op[0]->type == op[1]->type && !op[0]->type->is_matrix())
-	  || op0_scalar || op1_scalar) {
-	 for (unsigned c = 0, c0 = 0, c1 = 0;
-	      c < components;
-	      c0 += c0_inc, c1 += c1_inc, c++) {
-
-	    switch (op[0]->type->base_type) {
-	    case GLSL_TYPE_UINT:
-	       data.u[c] = op[0]->value.u[c0] * op[1]->value.u[c1];
-	       break;
-	    case GLSL_TYPE_INT:
-	       data.i[c] = op[0]->value.i[c0] * op[1]->value.i[c1];
-	       break;
-	    case GLSL_TYPE_FLOAT:
-	       data.f[c] = op[0]->value.f[c0] * op[1]->value.f[c1];
-	       break;
-	    default:
-	       assert(0);
-	    }
-	 }
-      } else {
-	 assert(op[0]->type->is_matrix() || op[1]->type->is_matrix());
-
-	 /* Multiply an N-by-M matrix with an M-by-P matrix.  Since either
-	  * matrix can be a GLSL vector, either N or P can be 1.
-	  *
-	  * For vec*mat, the vector is treated as a row vector.  This
-	  * means the vector is a 1-row x M-column matrix.
-	  *
-	  * For mat*vec, the vector is treated as a column vector.  Since
-	  * matrix_columns is 1 for vectors, this just works.
-	  */
-	 const unsigned n = op[0]->type->is_vector()
-	    ? 1 : op[0]->type->vector_elements;
-	 const unsigned m = op[1]->type->vector_elements;
-	 const unsigned p = op[1]->type->matrix_columns;
-	 for (unsigned j = 0; j < p; j++) {
-	    for (unsigned i = 0; i < n; i++) {
-	       for (unsigned k = 0; k < m; k++) {
-		  data.f[i+n*j] += op[0]->value.f[i+n*k]*op[1]->value.f[k+m*j];
-	       }
-	    }
-	 }
-      }
-
-      break;
-   case ir_binop_div:
-      /* FINISHME: Emit warning when division-by-zero is detected. */
-      assert(op[0]->type == op[1]->type || op0_scalar || op1_scalar);
-      for (unsigned c = 0, c0 = 0, c1 = 0;
-	   c < components;
-	   c0 += c0_inc, c1 += c1_inc, c++) {
-
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    if (op[1]->value.u[c1] == 0)
-	       return NULL;
-	    data.u[c] = op[0]->value.u[c0] / op[1]->value.u[c1];
-	    break;
-	 case GLSL_TYPE_INT:
-	    if (op[1]->value.i[c1] == 0)
-	       return NULL;
-	    data.i[c] = op[0]->value.i[c0] / op[1]->value.i[c1];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    if (op[1]->value.f[c1] == 0)
-	       return NULL;
-	    data.f[c] = op[0]->value.f[c0] / op[1]->value.f[c1];
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-
-      break;
-   case ir_binop_mod:
-      /* FINISHME: Emit warning when division-by-zero is detected. */
-      assert(op[0]->type == op[1]->type || op0_scalar || op1_scalar);
-      for (unsigned c = 0, c0 = 0, c1 = 0;
-	   c < components;
-	   c0 += c0_inc, c1 += c1_inc, c++) {
-
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    if (op[1]->value.u[c1] == 0)
-	       return NULL;
-	    data.u[c] = op[0]->value.u[c0] % op[1]->value.u[c1];
-	    break;
-	 case GLSL_TYPE_INT:
-	    if (op[1]->value.i[c1] == 0)
-	       return NULL;
-	    data.i[c] = op[0]->value.i[c0] % op[1]->value.i[c1];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    if (op[1]->value.f[c1] == 0)
-	       return NULL;
-	    /* We don't use fmod because it rounds toward zero; GLSL specifies
-	     * the use of floor.
-	     */
-	    data.f[c] = op[0]->value.f[c0] - op[1]->value.f[c1]
-	       * floorf(op[0]->value.f[c0] / op[1]->value.f[c1]);
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-
-      break;
-
-   case ir_binop_logic_and:
-      assert(op[0]->type->base_type == GLSL_TYPE_BOOL);
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.b[c] = op[0]->value.b[c] && op[1]->value.b[c];
-      break;
-   case ir_binop_logic_xor:
-      assert(op[0]->type->base_type == GLSL_TYPE_BOOL);
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.b[c] = op[0]->value.b[c] ^ op[1]->value.b[c];
-      break;
-   case ir_binop_logic_or:
-      assert(op[0]->type->base_type == GLSL_TYPE_BOOL);
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.b[c] = op[0]->value.b[c] || op[1]->value.b[c];
-      break;
-
-   case ir_binop_less:
-      assert(op[0]->type == op[1]->type);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.b[0] = op[0]->value.u[0] < op[1]->value.u[0];
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.b[0] = op[0]->value.i[0] < op[1]->value.i[0];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.b[0] = op[0]->value.f[0] < op[1]->value.f[0];
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-      break;
-   case ir_binop_greater:
-      assert(op[0]->type == op[1]->type);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.b[c] = op[0]->value.u[c] > op[1]->value.u[c];
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.b[c] = op[0]->value.i[c] > op[1]->value.i[c];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.b[c] = op[0]->value.f[c] > op[1]->value.f[c];
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-      break;
-   case ir_binop_lequal:
-      assert(op[0]->type == op[1]->type);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.b[0] = op[0]->value.u[0] <= op[1]->value.u[0];
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.b[0] = op[0]->value.i[0] <= op[1]->value.i[0];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.b[0] = op[0]->value.f[0] <= op[1]->value.f[0];
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-      break;
-   case ir_binop_gequal:
-      assert(op[0]->type == op[1]->type);
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.b[0] = op[0]->value.u[0] >= op[1]->value.u[0];
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.b[0] = op[0]->value.i[0] >= op[1]->value.i[0];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.b[0] = op[0]->value.f[0] >= op[1]->value.f[0];
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-      break;
-   case ir_binop_equal:
-      assert(op[0]->type == op[1]->type);
-      for (unsigned c = 0; c < components; c++) {
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.b[c] = op[0]->value.u[c] == op[1]->value.u[c];
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.b[c] = op[0]->value.i[c] == op[1]->value.i[c];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.b[c] = op[0]->value.f[c] == op[1]->value.f[c];
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-      break;
-   case ir_binop_nequal:
-      assert(op[0]->type != op[1]->type);
-      for (unsigned c = 0; c < components; c++) {
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.b[c] = op[0]->value.u[c] != op[1]->value.u[c];
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.b[c] = op[0]->value.i[c] != op[1]->value.i[c];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.b[c] = op[0]->value.f[c] != op[1]->value.f[c];
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-      break;
-   case ir_binop_all_equal:
-      data.b[0] = op[0]->has_value(op[1]);
-      break;
-   case ir_binop_any_nequal:
-      data.b[0] = !op[0]->has_value(op[1]);
-      break;
-
-   case ir_binop_lshift:
-      for (unsigned c = 0, c0 = 0, c1 = 0;
-           c < components;
-           c0 += c0_inc, c1 += c1_inc, c++) {
-
-          if (op[0]->type->base_type == GLSL_TYPE_INT &&
-              op[1]->type->base_type == GLSL_TYPE_INT) {
-              data.i[c] = op[0]->value.i[c0] << op[1]->value.i[c1];
-
-          } else if (op[0]->type->base_type == GLSL_TYPE_INT &&
-                     op[1]->type->base_type == GLSL_TYPE_UINT) {
-              data.i[c] = op[0]->value.i[c0] << op[1]->value.u[c1];
-
-          } else if (op[0]->type->base_type == GLSL_TYPE_UINT &&
-                     op[1]->type->base_type == GLSL_TYPE_INT) {
-              data.u[c] = op[0]->value.u[c0] << op[1]->value.i[c1];
-
-          } else if (op[0]->type->base_type == GLSL_TYPE_UINT &&
-                     op[1]->type->base_type == GLSL_TYPE_UINT) {
-              data.u[c] = op[0]->value.u[c0] << op[1]->value.u[c1];
-          }
-      }
-      break;
-
-   case ir_binop_rshift:
-       for (unsigned c = 0, c0 = 0, c1 = 0;
-            c < components;
-            c0 += c0_inc, c1 += c1_inc, c++) {
-
-           if (op[0]->type->base_type == GLSL_TYPE_INT &&
-               op[1]->type->base_type == GLSL_TYPE_INT) {
-               data.i[c] = op[0]->value.i[c0] >> op[1]->value.i[c1];
-
-           } else if (op[0]->type->base_type == GLSL_TYPE_INT &&
-                      op[1]->type->base_type == GLSL_TYPE_UINT) {
-               data.i[c] = op[0]->value.i[c0] >> op[1]->value.u[c1];
-
-           } else if (op[0]->type->base_type == GLSL_TYPE_UINT &&
-                      op[1]->type->base_type == GLSL_TYPE_INT) {
-               data.u[c] = op[0]->value.u[c0] >> op[1]->value.i[c1];
-
-           } else if (op[0]->type->base_type == GLSL_TYPE_UINT &&
-                      op[1]->type->base_type == GLSL_TYPE_UINT) {
-               data.u[c] = op[0]->value.u[c0] >> op[1]->value.u[c1];
-           }
-       }
-       break;
-
-   case ir_binop_bit_and:
-      for (unsigned c = 0, c0 = 0, c1 = 0;
-           c < components;
-           c0 += c0_inc, c1 += c1_inc, c++) {
-
-          switch (op[0]->type->base_type) {
-          case GLSL_TYPE_INT:
-              data.i[c] = op[0]->value.i[c0] & op[1]->value.i[c1];
-              break;
-          case GLSL_TYPE_UINT:
-              data.u[c] = op[0]->value.u[c0] & op[1]->value.u[c1];
-              break;
-          default:
-              assert(0);
-          }
-      }
-      break;
-
-   case ir_binop_bit_or:
-      for (unsigned c = 0, c0 = 0, c1 = 0;
-           c < components;
-           c0 += c0_inc, c1 += c1_inc, c++) {
-
-          switch (op[0]->type->base_type) {
-          case GLSL_TYPE_INT:
-              data.i[c] = op[0]->value.i[c0] | op[1]->value.i[c1];
-              break;
-          case GLSL_TYPE_UINT:
-              data.u[c] = op[0]->value.u[c0] | op[1]->value.u[c1];
-              break;
-          default:
-              assert(0);
-          }
-      }
-      break;
-
-   case ir_binop_bit_xor:
-      for (unsigned c = 0, c0 = 0, c1 = 0;
-           c < components;
-           c0 += c0_inc, c1 += c1_inc, c++) {
-
-          switch (op[0]->type->base_type) {
-          case GLSL_TYPE_INT:
-              data.i[c] = op[0]->value.i[c0] ^ op[1]->value.i[c1];
-              break;
-          case GLSL_TYPE_UINT:
-              data.u[c] = op[0]->value.u[c0] ^ op[1]->value.u[c1];
-              break;
-          default:
-              assert(0);
-          }
-      }
-      break;
-
-   case ir_quadop_vector:
-      for (unsigned c = 0; c < this->type->vector_elements; c++) {
-	 switch (this->type->base_type) {
-	 case GLSL_TYPE_INT:
-	    data.i[c] = op[c]->value.i[0];
-	    break;
-	 case GLSL_TYPE_UINT:
-	    data.u[c] = op[c]->value.u[0];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.f[c] = op[c]->value.f[0];
-	    break;
-	 default:
-	    assert(0);
-	 }
-      }
-      break;
-
-   default:
-      /* FINISHME: Should handle all expression types. */
-      return NULL;
-   }
-
-   return new(ctx) ir_constant(this->type, &data);
-}
-
-
-ir_constant *
-ir_texture::constant_expression_value()
-{
-   /* texture lookups aren't constant expressions */
-   return NULL;
-}
-
-
-ir_constant *
-ir_swizzle::constant_expression_value()
-{
-   ir_constant *v = this->val->constant_expression_value();
-
-   if (v != NULL) {
-      ir_constant_data data = { { 0 } };
-
-      const unsigned swiz_idx[4] = {
-	 this->mask.x, this->mask.y, this->mask.z, this->mask.w
-      };
-
-      for (unsigned i = 0; i < this->mask.num_components; i++) {
-	 switch (v->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	 case GLSL_TYPE_INT:   data.u[i] = v->value.u[swiz_idx[i]]; break;
-	 case GLSL_TYPE_FLOAT: data.f[i] = v->value.f[swiz_idx[i]]; break;
-	 case GLSL_TYPE_BOOL:  data.b[i] = v->value.b[swiz_idx[i]]; break;
-	 default:              assert(!"Should not get here."); break;
-	 }
-      }
-
-      void *ctx = ralloc_parent(this);
-      return new(ctx) ir_constant(this->type, &data);
-   }
-   return NULL;
-}
-
-
-ir_constant *
-ir_dereference_variable::constant_expression_value()
-{
-   /* This may occur during compile and var->type is glsl_type::error_type */
-   if (!var)
-      return NULL;
-
-   /* The constant_value of a uniform variable is its initializer,
-    * not the lifetime constant value of the uniform.
-    */
-   if (var->mode == ir_var_uniform)
-      return NULL;
-
-   if (!var->constant_value)
-      return NULL;
-
-   return var->constant_value->clone(ralloc_parent(var), NULL);
-}
-
-
-ir_constant *
-ir_dereference_array::constant_expression_value()
-{
-   ir_constant *array = this->array->constant_expression_value();
-   ir_constant *idx = this->array_index->constant_expression_value();
-
-   if ((array != NULL) && (idx != NULL)) {
-      void *ctx = ralloc_parent(this);
-      if (array->type->is_matrix()) {
-	 /* Array access of a matrix results in a vector.
-	  */
-	 const unsigned column = idx->value.u[0];
-
-	 const glsl_type *const column_type = array->type->column_type();
-
-	 /* Offset in the constant matrix to the first element of the column
-	  * to be extracted.
-	  */
-	 const unsigned mat_idx = column * column_type->vector_elements;
-
-	 ir_constant_data data = { { 0 } };
-
-	 switch (column_type->base_type) {
-	 case GLSL_TYPE_UINT:
-	 case GLSL_TYPE_INT:
-	    for (unsigned i = 0; i < column_type->vector_elements; i++)
-	       data.u[i] = array->value.u[mat_idx + i];
-
-	    break;
-
-	 case GLSL_TYPE_FLOAT:
-	    for (unsigned i = 0; i < column_type->vector_elements; i++)
-	       data.f[i] = array->value.f[mat_idx + i];
-
-	    break;
-
-	 default:
-	    assert(!"Should not get here.");
-	    break;
-	 }
-
-	 return new(ctx) ir_constant(column_type, &data);
-      } else if (array->type->is_vector()) {
-	 const unsigned component = idx->value.u[0];
-
-	 return new(ctx) ir_constant(array, component);
-      } else {
-	 const unsigned index = idx->value.u[0];
-	 return array->get_array_element(index)->clone(ctx, NULL);
-      }
-   }
-   return NULL;
-}
-
-
-ir_constant *
-ir_dereference_record::constant_expression_value()
-{
-   ir_constant *v = this->record->constant_expression_value();
-
-   return (v != NULL) ? v->get_record_field(this->field) : NULL;
-}
-
-
-ir_constant *
-ir_assignment::constant_expression_value()
-{
-   /* FINISHME: Handle CEs involving assignment (return RHS) */
-   return NULL;
-}
-
-
-ir_constant *
-ir_constant::constant_expression_value()
-{
-   return this;
-}
-
-
-ir_constant *
-ir_call::constant_expression_value()
-{
-   if (this->type == glsl_type::error_type)
-      return NULL;
-
-   /* From the GLSL 1.20 spec, page 23:
-    * "Function calls to user-defined functions (non-built-in functions)
-    *  cannot be used to form constant expressions."
-    */
-   if (!this->callee->is_builtin)
-      return NULL;
-
-   unsigned num_parameters = 0;
-
-   /* Check if all parameters are constant */
-   ir_constant *op[3];
-   foreach_list(n, &this->actual_parameters) {
-      ir_constant *constant = ((ir_rvalue *) n)->constant_expression_value();
-      if (constant == NULL)
-	 return NULL;
-
-      op[num_parameters] = constant;
-
-      assert(num_parameters < 3);
-      num_parameters++;
-   }
-
-   /* Individual cases below can either:
-    * - Assign "expr" a new ir_expression to evaluate (for basic opcodes)
-    * - Fill "data" with appopriate constant data
-    * - Return an ir_constant directly.
-    */
-   void *mem_ctx = ralloc_parent(this);
-   ir_expression *expr = NULL;
-
-   ir_constant_data data;
-   memset(&data, 0, sizeof(data));
-
-   const char *callee = this->callee_name();
-   if (strcmp(callee, "abs") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_unop_abs, type, op[0], NULL);
-   } else if (strcmp(callee, "all") == 0) {
-      assert(op[0]->type->is_boolean());
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 if (!op[0]->value.b[c])
-	    return new(mem_ctx) ir_constant(false);
-      }
-      return new(mem_ctx) ir_constant(true);
-   } else if (strcmp(callee, "any") == 0) {
-      assert(op[0]->type->is_boolean());
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 if (op[0]->value.b[c])
-	    return new(mem_ctx) ir_constant(true);
-      }
-      return new(mem_ctx) ir_constant(false);
-   } else if (strcmp(callee, "acos") == 0) {
-      assert(op[0]->type->is_float());
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.f[c] = acosf(op[0]->value.f[c]);
-   } else if (strcmp(callee, "acosh") == 0) {
-      assert(op[0]->type->is_float());
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.f[c] = acoshf(op[0]->value.f[c]);
-   } else if (strcmp(callee, "asin") == 0) {
-      assert(op[0]->type->is_float());
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.f[c] = asinf(op[0]->value.f[c]);
-   } else if (strcmp(callee, "asinh") == 0) {
-      assert(op[0]->type->is_float());
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.f[c] = asinhf(op[0]->value.f[c]);
-   } else if (strcmp(callee, "atan") == 0) {
-      assert(op[0]->type->is_float());
-      if (num_parameters == 2) {
-	 assert(op[1]->type->is_float());
-	 for (unsigned c = 0; c < op[0]->type->components(); c++)
-	    data.f[c] = atan2f(op[0]->value.f[c], op[1]->value.f[c]);
-      } else {
-	 for (unsigned c = 0; c < op[0]->type->components(); c++)
-	    data.f[c] = atanf(op[0]->value.f[c]);
-      }
-   } else if (strcmp(callee, "atanh") == 0) {
-      assert(op[0]->type->is_float());
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.f[c] = atanhf(op[0]->value.f[c]);
-   } else if (strcmp(callee, "dFdx") == 0 || strcmp(callee, "dFdy") == 0) {
-      return ir_constant::zero(mem_ctx, this->type);
-   } else if (strcmp(callee, "ceil") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_unop_ceil, type, op[0], NULL);
-   } else if (strcmp(callee, "clamp") == 0) {
-      assert(num_parameters == 3);
-      unsigned c1_inc = op[1]->type->is_scalar() ? 0 : 1;
-      unsigned c2_inc = op[2]->type->is_scalar() ? 0 : 1;
-      for (unsigned c = 0, c1 = 0, c2 = 0;
-	   c < op[0]->type->components();
-	   c1 += c1_inc, c2 += c2_inc, c++) {
-
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.u[c] = CLAMP(op[0]->value.u[c], op[1]->value.u[c1],
-			      op[2]->value.u[c2]);
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.i[c] = CLAMP(op[0]->value.i[c], op[1]->value.i[c1],
-			      op[2]->value.i[c2]);
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.f[c] = CLAMP(op[0]->value.f[c], op[1]->value.f[c1],
-			      op[2]->value.f[c2]);
-	    break;
-	 default:
-	    assert(!"Should not get here.");
-	 }
-      }
-   } else if (strcmp(callee, "cos") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_unop_cos, type, op[0], NULL);
-   } else if (strcmp(callee, "cosh") == 0) {
-      assert(op[0]->type->is_float());
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.f[c] = coshf(op[0]->value.f[c]);
-   } else if (strcmp(callee, "cross") == 0) {
-      assert(op[0]->type == glsl_type::vec3_type);
-      assert(op[1]->type == glsl_type::vec3_type);
-      data.f[0] = (op[0]->value.f[1] * op[1]->value.f[2] -
-		   op[1]->value.f[1] * op[0]->value.f[2]);
-      data.f[1] = (op[0]->value.f[2] * op[1]->value.f[0] -
-		   op[1]->value.f[2] * op[0]->value.f[0]);
-      data.f[2] = (op[0]->value.f[0] * op[1]->value.f[1] -
-		   op[1]->value.f[0] * op[0]->value.f[1]);
-   } else if (strcmp(callee, "degrees") == 0) {
-      assert(op[0]->type->is_float());
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.f[c] = 180.0F / M_PI * op[0]->value.f[c];
-   } else if (strcmp(callee, "distance") == 0) {
-      assert(op[0]->type->is_float() && op[1]->type->is_float());
-      float length_squared = 0.0;
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 float t = op[0]->value.f[c] - op[1]->value.f[c];
-	 length_squared += t * t;
-      }
-      return new(mem_ctx) ir_constant(sqrtf(length_squared));
-   } else if (strcmp(callee, "dot") == 0) {
-      return new(mem_ctx) ir_constant(dot(op[0], op[1]));
-   } else if (strcmp(callee, "equal") == 0) {
-      assert(op[0]->type->is_vector() && op[1] && op[1]->type->is_vector());
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.b[c] = op[0]->value.u[c] == op[1]->value.u[c];
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.b[c] = op[0]->value.i[c] == op[1]->value.i[c];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.b[c] = op[0]->value.f[c] == op[1]->value.f[c];
-	    break;
-	 case GLSL_TYPE_BOOL:
-	    data.b[c] = op[0]->value.b[c] == op[1]->value.b[c];
-	    break;
-	 default:
-	    assert(!"Should not get here.");
-	 }
-      }
-   } else if (strcmp(callee, "exp") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_unop_exp, type, op[0], NULL);
-   } else if (strcmp(callee, "exp2") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_unop_exp2, type, op[0], NULL);
-   } else if (strcmp(callee, "faceforward") == 0) {
-      if (dot(op[2], op[1]) < 0)
-	 return op[0];
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.f[c] = -op[0]->value.f[c];
-   } else if (strcmp(callee, "floor") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_unop_floor, type, op[0], NULL);
-   } else if (strcmp(callee, "fract") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_unop_fract, type, op[0], NULL);
-   } else if (strcmp(callee, "fwidth") == 0) {
-      return ir_constant::zero(mem_ctx, this->type);
-   } else if (strcmp(callee, "greaterThan") == 0) {
-      assert(op[0]->type->is_vector() && op[1] && op[1]->type->is_vector());
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.b[c] = op[0]->value.u[c] > op[1]->value.u[c];
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.b[c] = op[0]->value.i[c] > op[1]->value.i[c];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.b[c] = op[0]->value.f[c] > op[1]->value.f[c];
-	    break;
-	 default:
-	    assert(!"Should not get here.");
-	 }
-      }
-   } else if (strcmp(callee, "greaterThanEqual") == 0) {
-      assert(op[0]->type->is_vector() && op[1] && op[1]->type->is_vector());
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.b[c] = op[0]->value.u[c] >= op[1]->value.u[c];
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.b[c] = op[0]->value.i[c] >= op[1]->value.i[c];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.b[c] = op[0]->value.f[c] >= op[1]->value.f[c];
-	    break;
-	 default:
-	    assert(!"Should not get here.");
-	 }
-      }
-   } else if (strcmp(callee, "inversesqrt") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_unop_rsq, type, op[0], NULL);
-   } else if (strcmp(callee, "length") == 0) {
-      return new(mem_ctx) ir_constant(sqrtf(dot(op[0], op[0])));
-   } else if (strcmp(callee, "lessThan") == 0) {
-      assert(op[0]->type->is_vector() && op[1] && op[1]->type->is_vector());
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.b[c] = op[0]->value.u[c] < op[1]->value.u[c];
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.b[c] = op[0]->value.i[c] < op[1]->value.i[c];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.b[c] = op[0]->value.f[c] < op[1]->value.f[c];
-	    break;
-	 default:
-	    assert(!"Should not get here.");
-	 }
-      }
-   } else if (strcmp(callee, "lessThanEqual") == 0) {
-      assert(op[0]->type->is_vector() && op[1] && op[1]->type->is_vector());
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.b[c] = op[0]->value.u[c] <= op[1]->value.u[c];
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.b[c] = op[0]->value.i[c] <= op[1]->value.i[c];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.b[c] = op[0]->value.f[c] <= op[1]->value.f[c];
-	    break;
-	 default:
-	    assert(!"Should not get here.");
-	 }
-      }
-   } else if (strcmp(callee, "log") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_unop_log, type, op[0], NULL);
-   } else if (strcmp(callee, "log2") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_unop_log2, type, op[0], NULL);
-   } else if (strcmp(callee, "matrixCompMult") == 0) {
-      assert(op[0]->type->is_float() && op[1]->type->is_float());
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.f[c] = op[0]->value.f[c] * op[1]->value.f[c];
-   } else if (strcmp(callee, "max") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_binop_max, type, op[0], op[1]);
-   } else if (strcmp(callee, "min") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_binop_min, type, op[0], op[1]);
-   } else if (strcmp(callee, "mix") == 0) {
-      assert(op[0]->type->is_float() && op[1]->type->is_float());
-      if (op[2]->type->is_float()) {
-	 unsigned c2_inc = op[2]->type->is_scalar() ? 0 : 1;
-	 unsigned components = op[0]->type->components();
-	 for (unsigned c = 0, c2 = 0; c < components; c2 += c2_inc, c++) {
-	    data.f[c] = op[0]->value.f[c] * (1 - op[2]->value.f[c2]) +
-			op[1]->value.f[c] * op[2]->value.f[c2];
-	 }
-      } else {
-	 assert(op[2]->type->is_boolean());
-	 for (unsigned c = 0; c < op[0]->type->components(); c++)
-	    data.f[c] = op[op[2]->value.b[c] ? 1 : 0]->value.f[c];
-      }
-   } else if (strcmp(callee, "mod") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_binop_mod, type, op[0], op[1]);
-   } else if (strcmp(callee, "normalize") == 0) {
-      assert(op[0]->type->is_float());
-      float length = sqrtf(dot(op[0], op[0]));
-
-      if (length == 0)
-	 return ir_constant::zero(mem_ctx, this->type);
-
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.f[c] = op[0]->value.f[c] / length;
-   } else if (strcmp(callee, "not") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_unop_logic_not, type, op[0], NULL);
-   } else if (strcmp(callee, "notEqual") == 0) {
-      assert(op[0]->type->is_vector() && op[1] && op[1]->type->is_vector());
-      for (unsigned c = 0; c < op[0]->type->components(); c++) {
-	 switch (op[0]->type->base_type) {
-	 case GLSL_TYPE_UINT:
-	    data.b[c] = op[0]->value.u[c] != op[1]->value.u[c];
-	    break;
-	 case GLSL_TYPE_INT:
-	    data.b[c] = op[0]->value.i[c] != op[1]->value.i[c];
-	    break;
-	 case GLSL_TYPE_FLOAT:
-	    data.b[c] = op[0]->value.f[c] != op[1]->value.f[c];
-	    break;
-	 case GLSL_TYPE_BOOL:
-	    data.b[c] = op[0]->value.b[c] != op[1]->value.b[c];
-	    break;
-	 default:
-	    assert(!"Should not get here.");
-	 }
-      }
-   } else if (strcmp(callee, "outerProduct") == 0) {
-      assert(op[0]->type->is_vector() && op[1]->type->is_vector());
-      const unsigned m = op[0]->type->vector_elements;
-      const unsigned n = op[1]->type->vector_elements;
-      for (unsigned j = 0; j < n; j++) {
-	 for (unsigned i = 0; i < m; i++) {
-	    data.f[i+m*j] = op[0]->value.f[i] * op[1]->value.f[j];
-	 }
-      }
-   } else if (strcmp(callee, "pow") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_binop_pow, type, op[0], op[1]);
-   } else if (strcmp(callee, "radians") == 0) {
-      assert(op[0]->type->is_float());
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.f[c] = M_PI / 180.0F * op[0]->value.f[c];
-   } else if (strcmp(callee, "reflect") == 0) {
-      assert(op[0]->type->is_float());
-      float dot_NI = dot(op[1], op[0]);
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.f[c] = op[0]->value.f[c] - 2 * dot_NI * op[1]->value.f[c];
-   } else if (strcmp(callee, "refract") == 0) {
-      const float eta = op[2]->value.f[0];
-      const float dot_NI = dot(op[1], op[0]);
-      const float k = 1.0F - eta * eta * (1.0F - dot_NI * dot_NI);
-      if (k < 0.0) {
-	 return ir_constant::zero(mem_ctx, this->type);
-      } else {
-	 for (unsigned c = 0; c < type->components(); c++) {
-	    data.f[c] = eta * op[0]->value.f[c] - (eta * dot_NI + sqrtf(k))
-			    * op[1]->value.f[c];
-	 }
-      }
-   } else if (strcmp(callee, "sign") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_unop_sign, type, op[0], NULL);
-   } else if (strcmp(callee, "sin") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_unop_sin, type, op[0], NULL);
-   } else if (strcmp(callee, "sinh") == 0) {
-      assert(op[0]->type->is_float());
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.f[c] = sinhf(op[0]->value.f[c]);
-   } else if (strcmp(callee, "smoothstep") == 0) {
-      assert(num_parameters == 3);
-      assert(op[1]->type == op[0]->type);
-      unsigned edge_inc = op[0]->type->is_scalar() ? 0 : 1;
-      for (unsigned c = 0, e = 0; c < type->components(); e += edge_inc, c++) {
-	 const float edge0 = op[0]->value.f[e];
-	 const float edge1 = op[1]->value.f[e];
-	 if (edge0 == edge1) {
-	    data.f[c] = 0.0; /* Avoid a crash - results are undefined anyway */
-	 } else {
-	    const float numerator = op[2]->value.f[c] - edge0;
-	    const float denominator = edge1 - edge0;
-	    const float t = CLAMP(numerator/denominator, 0, 1);
-	    data.f[c] = t * t * (3 - 2 * t);
-	 }
-      }
-   } else if (strcmp(callee, "sqrt") == 0) {
-      expr = new(mem_ctx) ir_expression(ir_unop_sqrt, type, op[0], NULL);
-   } else if (strcmp(callee, "step") == 0) {
-      assert(op[0]->type->is_float() && op[1]->type->is_float());
-      /* op[0] (edge) may be either a scalar or a vector */
-      const unsigned c0_inc = op[0]->type->is_scalar() ? 0 : 1;
-      for (unsigned c = 0, c0 = 0; c < type->components(); c0 += c0_inc, c++)
-	 data.f[c] = (op[1]->value.f[c] < op[0]->value.f[c0]) ? 0.0F : 1.0F;
-   } else if (strcmp(callee, "tan") == 0) {
-      assert(op[0]->type->is_float());
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.f[c] = tanf(op[0]->value.f[c]);
-   } else if (strcmp(callee, "tanh") == 0) {
-      assert(op[0]->type->is_float());
-      for (unsigned c = 0; c < op[0]->type->components(); c++)
-	 data.f[c] = tanhf(op[0]->value.f[c]);
-   } else if (strcmp(callee, "transpose") == 0) {
-      assert(op[0]->type->is_matrix());
-      const unsigned n = op[0]->type->vector_elements;
-      const unsigned m = op[0]->type->matrix_columns;
-      for (unsigned j = 0; j < m; j++) {
-	 for (unsigned i = 0; i < n; i++) {
-	    data.f[m*i+j] += op[0]->value.f[i+n*j];
-	 }
-      }
-   } else {
-      /* Unsupported builtin - some are not allowed in constant expressions. */
-      return NULL;
-   }
-
-   if (expr != NULL)
-      return expr->constant_expression_value();
-
-   return new(mem_ctx) ir_constant(this->type, &data);
-}
+/*

+ * Copyright © 2010 Intel Corporation

+ *

+ * Permission is hereby granted, free of charge, to any person obtaining a

+ * copy of this software and associated documentation files (the "Software"),

+ * to deal in the Software without restriction, including without limitation

+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,

+ * and/or sell copies of the Software, and to permit persons to whom the

+ * Software is furnished to do so, subject to the following conditions:

+ *

+ * The above copyright notice and this permission notice (including the next

+ * paragraph) shall be included in all copies or substantial portions of the

+ * Software.

+ *

+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR

+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,

+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL

+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER

+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING

+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER

+ * DEALINGS IN THE SOFTWARE.

+ */

+

+/**

+ * \file ir_constant_expression.cpp

+ * Evaluate and process constant valued expressions

+ *

+ * In GLSL, constant valued expressions are used in several places.  These

+ * must be processed and evaluated very early in the compilation process.

+ *

+ *    * Sizes of arrays

+ *    * Initializers for uniforms

+ *    * Initializers for \c const variables

+ */

+

+#include <math.h>

+#include "main/core.h" /* for MAX2, MIN2, CLAMP */

+#include "ir.h"

+#include "ir_visitor.h"

+#include "glsl_types.h"

+

+static float

+dot(ir_constant *op0, ir_constant *op1)

+{

+   assert(op0->type->is_float() && op1->type->is_float());

+

+   float result = 0;

+   for (unsigned c = 0; c < op0->type->components(); c++)

+      result += op0->value.f[c] * op1->value.f[c];

+

+   return result;

+}

+

+ir_constant *

+ir_expression::constant_expression_value()

+{

+   if (this->type->is_error())

+      return NULL;

+

+   ir_constant *op[Elements(this->operands)] = { NULL, };

+   ir_constant_data data;

+

+   memset(&data, 0, sizeof(data));

+

+   for (unsigned operand = 0; operand < this->get_num_operands(); operand++) {

+      op[operand] = this->operands[operand]->constant_expression_value();

+      if (!op[operand])

+	 return NULL;

+   }

+

+   if (op[1] != NULL)

+      assert(op[0]->type->base_type == op[1]->type->base_type);

+

+   bool op0_scalar = op[0]->type->is_scalar();

+   bool op1_scalar = op[1] != NULL && op[1]->type->is_scalar();

+

+   /* When iterating over a vector or matrix's components, we want to increase

+    * the loop counter.  However, for scalars, we want to stay at 0.

+    */

+   unsigned c0_inc = op0_scalar ? 0 : 1;

+   unsigned c1_inc = op1_scalar ? 0 : 1;

+   unsigned components;

+   if (op1_scalar || !op[1]) {

+      components = op[0]->type->components();

+   } else {

+      components = op[1]->type->components();

+   }

+

+   void *ctx = ralloc_parent(this);

+

+   /* Handle array operations here, rather than below. */

+   if (op[0]->type->is_array()) {

+      assert(op[1] != NULL && op[1]->type->is_array());

+      switch (this->operation) {

+      case ir_binop_all_equal:

+	 return new(ctx) ir_constant(op[0]->has_value(op[1]));

+      case ir_binop_any_nequal:

+	 return new(ctx) ir_constant(!op[0]->has_value(op[1]));

+      default:

+	 break;

+      }

+      return NULL;

+   }

+

+   switch (this->operation) {

+   case ir_unop_bit_not:

+       switch (op[0]->type->base_type) {

+       case GLSL_TYPE_INT:

+           for (unsigned c = 0; c < components; c++)

+               data.i[c] = ~ op[0]->value.i[c];

+           break;

+       case GLSL_TYPE_UINT:

+           for (unsigned c = 0; c < components; c++)

+               data.u[c] = ~ op[0]->value.u[c];

+           break;

+       default:

+           assert(0);

+       }

+       break;

+

+   case ir_unop_logic_not:

+      assert(op[0]->type->base_type == GLSL_TYPE_BOOL);

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.b[c] = !op[0]->value.b[c];

+      break;

+

+   case ir_unop_f2i:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.i[c] = (int) op[0]->value.f[c];

+      }

+      break;

+   case ir_unop_i2f:

+      assert(op[0]->type->base_type == GLSL_TYPE_INT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = (float) op[0]->value.i[c];

+      }

+      break;

+   case ir_unop_u2f:

+      assert(op[0]->type->base_type == GLSL_TYPE_UINT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = (float) op[0]->value.u[c];

+      }

+      break;

+   case ir_unop_b2f:

+      assert(op[0]->type->base_type == GLSL_TYPE_BOOL);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = op[0]->value.b[c] ? 1.0F : 0.0F;

+      }

+      break;

+   case ir_unop_f2b:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.b[c] = op[0]->value.f[c] != 0.0F ? true : false;

+      }

+      break;

+   case ir_unop_b2i:

+      assert(op[0]->type->base_type == GLSL_TYPE_BOOL);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.u[c] = op[0]->value.b[c] ? 1 : 0;

+      }

+      break;

+   case ir_unop_i2b:

+      assert(op[0]->type->is_integer());

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.b[c] = op[0]->value.u[c] ? true : false;

+      }

+      break;

+

+   case ir_unop_any:

+      assert(op[0]->type->is_boolean());

+      data.b[0] = false;

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 if (op[0]->value.b[c])

+	    data.b[0] = true;

+      }

+      break;

+

+   case ir_unop_trunc:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = truncf(op[0]->value.f[c]);

+      }

+      break;

+

+   case ir_unop_ceil:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = ceilf(op[0]->value.f[c]);

+      }

+      break;

+

+   case ir_unop_floor:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = floorf(op[0]->value.f[c]);

+      }

+      break;

+

+   case ir_unop_fract:

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 switch (this->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.u[c] = 0;

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.i[c] = 0;

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.f[c] = op[0]->value.f[c] - floor(op[0]->value.f[c]);

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+      break;

+

+   case ir_unop_sin:

+   case ir_unop_sin_reduced:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = sinf(op[0]->value.f[c]);

+      }

+      break;

+

+   case ir_unop_cos:

+   case ir_unop_cos_reduced:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = cosf(op[0]->value.f[c]);

+      }

+      break;

+

+   case ir_unop_neg:

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 switch (this->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.u[c] = -((int) op[0]->value.u[c]);

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.i[c] = -op[0]->value.i[c];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.f[c] = -op[0]->value.f[c];

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+      break;

+

+   case ir_unop_abs:

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 switch (this->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.u[c] = op[0]->value.u[c];

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.i[c] = op[0]->value.i[c];

+	    if (data.i[c] < 0)

+	       data.i[c] = -data.i[c];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.f[c] = fabs(op[0]->value.f[c]);

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+      break;

+

+   case ir_unop_sign:

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 switch (this->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.u[c] = op[0]->value.i[c] > 0;

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.i[c] = (op[0]->value.i[c] > 0) - (op[0]->value.i[c] < 0);

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.f[c] = float((op[0]->value.f[c] > 0)-(op[0]->value.f[c] < 0));

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+      break;

+

+   case ir_unop_rcp:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 switch (this->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    if (op[0]->value.u[c] != 0.0)

+	       data.u[c] = 1 / op[0]->value.u[c];

+	    break;

+	 case GLSL_TYPE_INT:

+	    if (op[0]->value.i[c] != 0.0)

+	       data.i[c] = 1 / op[0]->value.i[c];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    if (op[0]->value.f[c] != 0.0)

+	       data.f[c] = 1.0F / op[0]->value.f[c];

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+      break;

+

+   case ir_unop_rsq:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = 1.0F / sqrtf(op[0]->value.f[c]);

+      }

+      break;

+

+   case ir_unop_sqrt:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = sqrtf(op[0]->value.f[c]);

+      }

+      break;

+

+   case ir_unop_exp:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = expf(op[0]->value.f[c]);

+      }

+      break;

+

+   case ir_unop_exp2:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = exp2f(op[0]->value.f[c]);

+      }

+      break;

+

+   case ir_unop_log:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = logf(op[0]->value.f[c]);

+      }

+      break;

+

+   case ir_unop_log2:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = log2f(op[0]->value.f[c]);

+      }

+      break;

+

+   case ir_unop_dFdx:

+   case ir_unop_dFdy:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = 0.0;

+      }

+      break;

+

+   case ir_binop_pow:

+      assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 data.f[c] = powf(op[0]->value.f[c], op[1]->value.f[c]);

+      }

+      break;

+

+   case ir_binop_dot:

+      data.f[0] = dot(op[0], op[1]);

+      break;

+

+   case ir_binop_min:

+      assert(op[0]->type == op[1]->type || op0_scalar || op1_scalar);

+      for (unsigned c = 0, c0 = 0, c1 = 0;

+	   c < components;

+	   c0 += c0_inc, c1 += c1_inc, c++) {

+

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.u[c] = MIN2(op[0]->value.u[c0], op[1]->value.u[c1]);

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.i[c] = MIN2(op[0]->value.i[c0], op[1]->value.i[c1]);

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.f[c] = MIN2(op[0]->value.f[c0], op[1]->value.f[c1]);

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+

+      break;

+   case ir_binop_max:

+      assert(op[0]->type == op[1]->type || op0_scalar || op1_scalar);

+      for (unsigned c = 0, c0 = 0, c1 = 0;

+	   c < components;

+	   c0 += c0_inc, c1 += c1_inc, c++) {

+

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.u[c] = MAX2(op[0]->value.u[c0], op[1]->value.u[c1]);

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.i[c] = MAX2(op[0]->value.i[c0], op[1]->value.i[c1]);

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.f[c] = MAX2(op[0]->value.f[c0], op[1]->value.f[c1]);

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+      break;

+

+   case ir_binop_add:

+      assert(op[0]->type == op[1]->type || op0_scalar || op1_scalar);

+      for (unsigned c = 0, c0 = 0, c1 = 0;

+	   c < components;

+	   c0 += c0_inc, c1 += c1_inc, c++) {

+

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.u[c] = op[0]->value.u[c0] + op[1]->value.u[c1];

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.i[c] = op[0]->value.i[c0] + op[1]->value.i[c1];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.f[c] = op[0]->value.f[c0] + op[1]->value.f[c1];

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+

+      break;

+   case ir_binop_sub:

+      assert(op[0]->type == op[1]->type || op0_scalar || op1_scalar);

+      for (unsigned c = 0, c0 = 0, c1 = 0;

+	   c < components;

+	   c0 += c0_inc, c1 += c1_inc, c++) {

+

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.u[c] = op[0]->value.u[c0] - op[1]->value.u[c1];

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.i[c] = op[0]->value.i[c0] - op[1]->value.i[c1];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.f[c] = op[0]->value.f[c0] - op[1]->value.f[c1];

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+

+      break;

+   case ir_binop_mul:

+      /* Check for equal types, or unequal types involving scalars */

+      if ((op[0]->type == op[1]->type && !op[0]->type->is_matrix())

+	  || op0_scalar || op1_scalar) {

+	 for (unsigned c = 0, c0 = 0, c1 = 0;

+	      c < components;

+	      c0 += c0_inc, c1 += c1_inc, c++) {

+

+	    switch (op[0]->type->base_type) {

+	    case GLSL_TYPE_UINT:

+	       data.u[c] = op[0]->value.u[c0] * op[1]->value.u[c1];

+	       break;

+	    case GLSL_TYPE_INT:

+	       data.i[c] = op[0]->value.i[c0] * op[1]->value.i[c1];

+	       break;

+	    case GLSL_TYPE_FLOAT:

+	       data.f[c] = op[0]->value.f[c0] * op[1]->value.f[c1];

+	       break;

+	    default:

+	       assert(0);

+	    }

+	 }

+      } else {

+	 assert(op[0]->type->is_matrix() || op[1]->type->is_matrix());

+

+	 /* Multiply an N-by-M matrix with an M-by-P matrix.  Since either

+	  * matrix can be a GLSL vector, either N or P can be 1.

+	  *

+	  * For vec*mat, the vector is treated as a row vector.  This

+	  * means the vector is a 1-row x M-column matrix.

+	  *

+	  * For mat*vec, the vector is treated as a column vector.  Since

+	  * matrix_columns is 1 for vectors, this just works.

+	  */

+	 const unsigned n = op[0]->type->is_vector()

+	    ? 1 : op[0]->type->vector_elements;

+	 const unsigned m = op[1]->type->vector_elements;

+	 const unsigned p = op[1]->type->matrix_columns;

+	 for (unsigned j = 0; j < p; j++) {

+	    for (unsigned i = 0; i < n; i++) {

+	       for (unsigned k = 0; k < m; k++) {

+		  data.f[i+n*j] += op[0]->value.f[i+n*k]*op[1]->value.f[k+m*j];

+	       }

+	    }

+	 }

+      }

+

+      break;

+   case ir_binop_div:

+      /* FINISHME: Emit warning when division-by-zero is detected. */

+      assert(op[0]->type == op[1]->type || op0_scalar || op1_scalar);

+      for (unsigned c = 0, c0 = 0, c1 = 0;

+	   c < components;

+	   c0 += c0_inc, c1 += c1_inc, c++) {

+

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    if (op[1]->value.u[c1] == 0) {

+	       data.u[c] = 0;

+	    } else {

+	       data.u[c] = op[0]->value.u[c0] / op[1]->value.u[c1];

+	    }

+	    break;

+	 case GLSL_TYPE_INT:

+	    if (op[1]->value.i[c1] == 0) {

+	       data.i[c] = 0;

+	    } else {

+	       data.i[c] = op[0]->value.i[c0] / op[1]->value.i[c1];

+	    }

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.f[c] = op[0]->value.f[c0] / op[1]->value.f[c1];

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+

+      break;

+   case ir_binop_mod:

+      /* FINISHME: Emit warning when division-by-zero is detected. */

+      assert(op[0]->type == op[1]->type || op0_scalar || op1_scalar);

+      for (unsigned c = 0, c0 = 0, c1 = 0;

+	   c < components;

+	   c0 += c0_inc, c1 += c1_inc, c++) {

+

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    if (op[1]->value.u[c1] == 0) {

+	       data.u[c] = 0;

+	    } else {

+	       data.u[c] = op[0]->value.u[c0] % op[1]->value.u[c1];

+	    }

+	    break;

+	 case GLSL_TYPE_INT:

+	    if (op[1]->value.i[c1] == 0) {

+	       data.i[c] = 0;

+	    } else {

+	       data.i[c] = op[0]->value.i[c0] % op[1]->value.i[c1];

+	    }

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    /* We don't use fmod because it rounds toward zero; GLSL specifies

+	     * the use of floor.

+	     */

+	    data.f[c] = op[0]->value.f[c0] - op[1]->value.f[c1]

+	       * floorf(op[0]->value.f[c0] / op[1]->value.f[c1]);

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+

+      break;

+

+   case ir_binop_logic_and:

+      assert(op[0]->type->base_type == GLSL_TYPE_BOOL);

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.b[c] = op[0]->value.b[c] && op[1]->value.b[c];

+      break;

+   case ir_binop_logic_xor:

+      assert(op[0]->type->base_type == GLSL_TYPE_BOOL);

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.b[c] = op[0]->value.b[c] ^ op[1]->value.b[c];

+      break;

+   case ir_binop_logic_or:

+      assert(op[0]->type->base_type == GLSL_TYPE_BOOL);

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.b[c] = op[0]->value.b[c] || op[1]->value.b[c];

+      break;

+

+   case ir_binop_less:

+      assert(op[0]->type == op[1]->type);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.b[0] = op[0]->value.u[0] < op[1]->value.u[0];

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.b[0] = op[0]->value.i[0] < op[1]->value.i[0];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.b[0] = op[0]->value.f[0] < op[1]->value.f[0];

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+      break;

+   case ir_binop_greater:

+      assert(op[0]->type == op[1]->type);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.b[c] = op[0]->value.u[c] > op[1]->value.u[c];

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.b[c] = op[0]->value.i[c] > op[1]->value.i[c];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.b[c] = op[0]->value.f[c] > op[1]->value.f[c];

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+      break;

+   case ir_binop_lequal:

+      assert(op[0]->type == op[1]->type);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.b[0] = op[0]->value.u[0] <= op[1]->value.u[0];

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.b[0] = op[0]->value.i[0] <= op[1]->value.i[0];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.b[0] = op[0]->value.f[0] <= op[1]->value.f[0];

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+      break;

+   case ir_binop_gequal:

+      assert(op[0]->type == op[1]->type);

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.b[0] = op[0]->value.u[0] >= op[1]->value.u[0];

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.b[0] = op[0]->value.i[0] >= op[1]->value.i[0];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.b[0] = op[0]->value.f[0] >= op[1]->value.f[0];

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+      break;

+   case ir_binop_equal:

+      assert(op[0]->type == op[1]->type);

+      for (unsigned c = 0; c < components; c++) {

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.b[c] = op[0]->value.u[c] == op[1]->value.u[c];

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.b[c] = op[0]->value.i[c] == op[1]->value.i[c];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.b[c] = op[0]->value.f[c] == op[1]->value.f[c];

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+      break;

+   case ir_binop_nequal:

+      assert(op[0]->type != op[1]->type);

+      for (unsigned c = 0; c < components; c++) {

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.b[c] = op[0]->value.u[c] != op[1]->value.u[c];

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.b[c] = op[0]->value.i[c] != op[1]->value.i[c];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.b[c] = op[0]->value.f[c] != op[1]->value.f[c];

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+      break;

+   case ir_binop_all_equal:

+      data.b[0] = op[0]->has_value(op[1]);

+      break;

+   case ir_binop_any_nequal:

+      data.b[0] = !op[0]->has_value(op[1]);

+      break;

+

+   case ir_binop_lshift:

+      for (unsigned c = 0, c0 = 0, c1 = 0;

+           c < components;

+           c0 += c0_inc, c1 += c1_inc, c++) {

+

+          if (op[0]->type->base_type == GLSL_TYPE_INT &&

+              op[1]->type->base_type == GLSL_TYPE_INT) {

+              data.i[c] = op[0]->value.i[c0] << op[1]->value.i[c1];

+

+          } else if (op[0]->type->base_type == GLSL_TYPE_INT &&

+                     op[1]->type->base_type == GLSL_TYPE_UINT) {

+              data.i[c] = op[0]->value.i[c0] << op[1]->value.u[c1];

+

+          } else if (op[0]->type->base_type == GLSL_TYPE_UINT &&

+                     op[1]->type->base_type == GLSL_TYPE_INT) {

+              data.u[c] = op[0]->value.u[c0] << op[1]->value.i[c1];

+

+          } else if (op[0]->type->base_type == GLSL_TYPE_UINT &&

+                     op[1]->type->base_type == GLSL_TYPE_UINT) {

+              data.u[c] = op[0]->value.u[c0] << op[1]->value.u[c1];

+          }

+      }

+      break;

+

+   case ir_binop_rshift:

+       for (unsigned c = 0, c0 = 0, c1 = 0;

+            c < components;

+            c0 += c0_inc, c1 += c1_inc, c++) {

+

+           if (op[0]->type->base_type == GLSL_TYPE_INT &&

+               op[1]->type->base_type == GLSL_TYPE_INT) {

+               data.i[c] = op[0]->value.i[c0] >> op[1]->value.i[c1];

+

+           } else if (op[0]->type->base_type == GLSL_TYPE_INT &&

+                      op[1]->type->base_type == GLSL_TYPE_UINT) {

+               data.i[c] = op[0]->value.i[c0] >> op[1]->value.u[c1];

+

+           } else if (op[0]->type->base_type == GLSL_TYPE_UINT &&

+                      op[1]->type->base_type == GLSL_TYPE_INT) {

+               data.u[c] = op[0]->value.u[c0] >> op[1]->value.i[c1];

+

+           } else if (op[0]->type->base_type == GLSL_TYPE_UINT &&

+                      op[1]->type->base_type == GLSL_TYPE_UINT) {

+               data.u[c] = op[0]->value.u[c0] >> op[1]->value.u[c1];

+           }

+       }

+       break;

+

+   case ir_binop_bit_and:

+      for (unsigned c = 0, c0 = 0, c1 = 0;

+           c < components;

+           c0 += c0_inc, c1 += c1_inc, c++) {

+

+          switch (op[0]->type->base_type) {

+          case GLSL_TYPE_INT:

+              data.i[c] = op[0]->value.i[c0] & op[1]->value.i[c1];

+              break;

+          case GLSL_TYPE_UINT:

+              data.u[c] = op[0]->value.u[c0] & op[1]->value.u[c1];

+              break;

+          default:

+              assert(0);

+          }

+      }

+      break;

+

+   case ir_binop_bit_or:

+      for (unsigned c = 0, c0 = 0, c1 = 0;

+           c < components;

+           c0 += c0_inc, c1 += c1_inc, c++) {

+

+          switch (op[0]->type->base_type) {

+          case GLSL_TYPE_INT:

+              data.i[c] = op[0]->value.i[c0] | op[1]->value.i[c1];

+              break;

+          case GLSL_TYPE_UINT:

+              data.u[c] = op[0]->value.u[c0] | op[1]->value.u[c1];

+              break;

+          default:

+              assert(0);

+          }

+      }

+      break;

+

+   case ir_binop_bit_xor:

+      for (unsigned c = 0, c0 = 0, c1 = 0;

+           c < components;

+           c0 += c0_inc, c1 += c1_inc, c++) {

+

+          switch (op[0]->type->base_type) {

+          case GLSL_TYPE_INT:

+              data.i[c] = op[0]->value.i[c0] ^ op[1]->value.i[c1];

+              break;

+          case GLSL_TYPE_UINT:

+              data.u[c] = op[0]->value.u[c0] ^ op[1]->value.u[c1];

+              break;

+          default:

+              assert(0);

+          }

+      }

+      break;

+

+   case ir_quadop_vector:

+      for (unsigned c = 0; c < this->type->vector_elements; c++) {

+	 switch (this->type->base_type) {

+	 case GLSL_TYPE_INT:

+	    data.i[c] = op[c]->value.i[0];

+	    break;

+	 case GLSL_TYPE_UINT:

+	    data.u[c] = op[c]->value.u[0];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.f[c] = op[c]->value.f[0];

+	    break;

+	 default:

+	    assert(0);

+	 }

+      }

+      break;

+

+   default:

+      /* FINISHME: Should handle all expression types. */

+      return NULL;

+   }

+

+   return new(ctx) ir_constant(this->type, &data);

+}

+

+

+ir_constant *

+ir_texture::constant_expression_value()

+{

+   /* texture lookups aren't constant expressions */

+   return NULL;

+}

+

+

+ir_constant *

+ir_swizzle::constant_expression_value()

+{

+   ir_constant *v = this->val->constant_expression_value();

+

+   if (v != NULL) {

+      ir_constant_data data = { { 0 } };

+

+      const unsigned swiz_idx[4] = {

+	 this->mask.x, this->mask.y, this->mask.z, this->mask.w

+      };

+

+      for (unsigned i = 0; i < this->mask.num_components; i++) {

+	 switch (v->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	 case GLSL_TYPE_INT:   data.u[i] = v->value.u[swiz_idx[i]]; break;

+	 case GLSL_TYPE_FLOAT: data.f[i] = v->value.f[swiz_idx[i]]; break;

+	 case GLSL_TYPE_BOOL:  data.b[i] = v->value.b[swiz_idx[i]]; break;

+	 default:              assert(!"Should not get here."); break;

+	 }

+      }

+

+      void *ctx = ralloc_parent(this);

+      return new(ctx) ir_constant(this->type, &data);

+   }

+   return NULL;

+}

+

+

+ir_constant *

+ir_dereference_variable::constant_expression_value()

+{

+   /* This may occur during compile and var->type is glsl_type::error_type */

+   if (!var)

+      return NULL;

+

+   /* The constant_value of a uniform variable is its initializer,

+    * not the lifetime constant value of the uniform.

+    */

+   if (var->mode == ir_var_uniform)

+      return NULL;

+

+   if (!var->constant_value)

+      return NULL;

+

+   return var->constant_value->clone(ralloc_parent(var), NULL);

+}

+

+

+ir_constant *

+ir_dereference_array::constant_expression_value()

+{

+   ir_constant *array = this->array->constant_expression_value();

+   ir_constant *idx = this->array_index->constant_expression_value();

+

+   if ((array != NULL) && (idx != NULL)) {

+      void *ctx = ralloc_parent(this);

+      if (array->type->is_matrix()) {

+	 /* Array access of a matrix results in a vector.

+	  */

+	 const unsigned column = idx->value.u[0];

+

+	 const glsl_type *const column_type = array->type->column_type();

+

+	 /* Offset in the constant matrix to the first element of the column

+	  * to be extracted.

+	  */

+	 const unsigned mat_idx = column * column_type->vector_elements;

+

+	 ir_constant_data data = { { 0 } };

+

+	 switch (column_type->base_type) {

+	 case GLSL_TYPE_UINT:

+	 case GLSL_TYPE_INT:

+	    for (unsigned i = 0; i < column_type->vector_elements; i++)

+	       data.u[i] = array->value.u[mat_idx + i];

+

+	    break;

+

+	 case GLSL_TYPE_FLOAT:

+	    for (unsigned i = 0; i < column_type->vector_elements; i++)

+	       data.f[i] = array->value.f[mat_idx + i];

+

+	    break;

+

+	 default:

+	    assert(!"Should not get here.");

+	    break;

+	 }

+

+	 return new(ctx) ir_constant(column_type, &data);

+      } else if (array->type->is_vector()) {

+	 const unsigned component = idx->value.u[0];

+

+	 return new(ctx) ir_constant(array, component);

+      } else {

+	 const unsigned index = idx->value.u[0];

+	 return array->get_array_element(index)->clone(ctx, NULL);

+      }

+   }

+   return NULL;

+}

+

+

+ir_constant *

+ir_dereference_record::constant_expression_value()

+{

+   ir_constant *v = this->record->constant_expression_value();

+

+   return (v != NULL) ? v->get_record_field(this->field) : NULL;

+}

+

+

+ir_constant *

+ir_assignment::constant_expression_value()

+{

+   /* FINISHME: Handle CEs involving assignment (return RHS) */

+   return NULL;

+}

+

+

+ir_constant *

+ir_constant::constant_expression_value()

+{

+   return this;

+}

+

+

+ir_constant *

+ir_call::constant_expression_value()

+{

+   if (this->type == glsl_type::error_type)

+      return NULL;

+

+   /* From the GLSL 1.20 spec, page 23:

+    * "Function calls to user-defined functions (non-built-in functions)

+    *  cannot be used to form constant expressions."

+    */

+   if (!this->callee->is_builtin)

+      return NULL;

+

+   unsigned num_parameters = 0;

+

+   /* Check if all parameters are constant */

+   ir_constant *op[3];

+   foreach_list(n, &this->actual_parameters) {

+      ir_constant *constant = ((ir_rvalue *) n)->constant_expression_value();

+      if (constant == NULL)

+	 return NULL;

+

+      op[num_parameters] = constant;

+

+      assert(num_parameters < 3);

+      num_parameters++;

+   }

+

+   /* Individual cases below can either:

+    * - Assign "expr" a new ir_expression to evaluate (for basic opcodes)

+    * - Fill "data" with appopriate constant data

+    * - Return an ir_constant directly.

+    */

+   void *mem_ctx = ralloc_parent(this);

+   ir_expression *expr = NULL;

+

+   ir_constant_data data;

+   memset(&data, 0, sizeof(data));

+

+   const char *callee = this->callee_name();

+   if (strcmp(callee, "abs") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_unop_abs, type, op[0], NULL);

+   } else if (strcmp(callee, "all") == 0) {

+      assert(op[0]->type->is_boolean());

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 if (!op[0]->value.b[c])

+	    return new(mem_ctx) ir_constant(false);

+      }

+      return new(mem_ctx) ir_constant(true);

+   } else if (strcmp(callee, "any") == 0) {

+      assert(op[0]->type->is_boolean());

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 if (op[0]->value.b[c])

+	    return new(mem_ctx) ir_constant(true);

+      }

+      return new(mem_ctx) ir_constant(false);

+   } else if (strcmp(callee, "acos") == 0) {

+      assert(op[0]->type->is_float());

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.f[c] = acosf(op[0]->value.f[c]);

+   } else if (strcmp(callee, "acosh") == 0) {

+      assert(op[0]->type->is_float());

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.f[c] = acoshf(op[0]->value.f[c]);

+   } else if (strcmp(callee, "asin") == 0) {

+      assert(op[0]->type->is_float());

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.f[c] = asinf(op[0]->value.f[c]);

+   } else if (strcmp(callee, "asinh") == 0) {

+      assert(op[0]->type->is_float());

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.f[c] = asinhf(op[0]->value.f[c]);

+   } else if (strcmp(callee, "atan") == 0) {

+      assert(op[0]->type->is_float());

+      if (num_parameters == 2) {

+	 assert(op[1]->type->is_float());

+	 for (unsigned c = 0; c < op[0]->type->components(); c++)

+	    data.f[c] = atan2f(op[0]->value.f[c], op[1]->value.f[c]);

+      } else {

+	 for (unsigned c = 0; c < op[0]->type->components(); c++)

+	    data.f[c] = atanf(op[0]->value.f[c]);

+      }

+   } else if (strcmp(callee, "atanh") == 0) {

+      assert(op[0]->type->is_float());

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.f[c] = atanhf(op[0]->value.f[c]);

+   } else if (strcmp(callee, "dFdx") == 0 || strcmp(callee, "dFdy") == 0) {

+      return ir_constant::zero(mem_ctx, this->type);

+   } else if (strcmp(callee, "ceil") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_unop_ceil, type, op[0], NULL);

+   } else if (strcmp(callee, "clamp") == 0) {

+      assert(num_parameters == 3);

+      unsigned c1_inc = op[1]->type->is_scalar() ? 0 : 1;

+      unsigned c2_inc = op[2]->type->is_scalar() ? 0 : 1;

+      for (unsigned c = 0, c1 = 0, c2 = 0;

+	   c < op[0]->type->components();

+	   c1 += c1_inc, c2 += c2_inc, c++) {

+

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.u[c] = CLAMP(op[0]->value.u[c], op[1]->value.u[c1],

+			      op[2]->value.u[c2]);

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.i[c] = CLAMP(op[0]->value.i[c], op[1]->value.i[c1],

+			      op[2]->value.i[c2]);

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.f[c] = CLAMP(op[0]->value.f[c], op[1]->value.f[c1],

+			      op[2]->value.f[c2]);

+	    break;

+	 default:

+	    assert(!"Should not get here.");

+	 }

+      }

+   } else if (strcmp(callee, "cos") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_unop_cos, type, op[0], NULL);

+   } else if (strcmp(callee, "cosh") == 0) {

+      assert(op[0]->type->is_float());

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.f[c] = coshf(op[0]->value.f[c]);

+   } else if (strcmp(callee, "cross") == 0) {

+      assert(op[0]->type == glsl_type::vec3_type);

+      assert(op[1]->type == glsl_type::vec3_type);

+      data.f[0] = (op[0]->value.f[1] * op[1]->value.f[2] -

+		   op[1]->value.f[1] * op[0]->value.f[2]);

+      data.f[1] = (op[0]->value.f[2] * op[1]->value.f[0] -

+		   op[1]->value.f[2] * op[0]->value.f[0]);

+      data.f[2] = (op[0]->value.f[0] * op[1]->value.f[1] -

+		   op[1]->value.f[0] * op[0]->value.f[1]);

+   } else if (strcmp(callee, "degrees") == 0) {

+      assert(op[0]->type->is_float());

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.f[c] = 180.0F / M_PI * op[0]->value.f[c];

+   } else if (strcmp(callee, "distance") == 0) {

+      assert(op[0]->type->is_float() && op[1]->type->is_float());

+      float length_squared = 0.0;

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 float t = op[0]->value.f[c] - op[1]->value.f[c];

+	 length_squared += t * t;

+      }

+      return new(mem_ctx) ir_constant(sqrtf(length_squared));

+   } else if (strcmp(callee, "dot") == 0) {

+      return new(mem_ctx) ir_constant(dot(op[0], op[1]));

+   } else if (strcmp(callee, "equal") == 0) {

+      assert(op[0]->type->is_vector() && op[1] && op[1]->type->is_vector());

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.b[c] = op[0]->value.u[c] == op[1]->value.u[c];

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.b[c] = op[0]->value.i[c] == op[1]->value.i[c];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.b[c] = op[0]->value.f[c] == op[1]->value.f[c];

+	    break;

+	 case GLSL_TYPE_BOOL:

+	    data.b[c] = op[0]->value.b[c] == op[1]->value.b[c];

+	    break;

+	 default:

+	    assert(!"Should not get here.");

+	 }

+      }

+   } else if (strcmp(callee, "exp") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_unop_exp, type, op[0], NULL);

+   } else if (strcmp(callee, "exp2") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_unop_exp2, type, op[0], NULL);

+   } else if (strcmp(callee, "faceforward") == 0) {

+      if (dot(op[2], op[1]) < 0)

+	 return op[0];

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.f[c] = -op[0]->value.f[c];

+   } else if (strcmp(callee, "floor") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_unop_floor, type, op[0], NULL);

+   } else if (strcmp(callee, "fract") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_unop_fract, type, op[0], NULL);

+   } else if (strcmp(callee, "fwidth") == 0) {

+      return ir_constant::zero(mem_ctx, this->type);

+   } else if (strcmp(callee, "greaterThan") == 0) {

+      assert(op[0]->type->is_vector() && op[1] && op[1]->type->is_vector());

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.b[c] = op[0]->value.u[c] > op[1]->value.u[c];

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.b[c] = op[0]->value.i[c] > op[1]->value.i[c];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.b[c] = op[0]->value.f[c] > op[1]->value.f[c];

+	    break;

+	 default:

+	    assert(!"Should not get here.");

+	 }

+      }

+   } else if (strcmp(callee, "greaterThanEqual") == 0) {

+      assert(op[0]->type->is_vector() && op[1] && op[1]->type->is_vector());

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.b[c] = op[0]->value.u[c] >= op[1]->value.u[c];

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.b[c] = op[0]->value.i[c] >= op[1]->value.i[c];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.b[c] = op[0]->value.f[c] >= op[1]->value.f[c];

+	    break;

+	 default:

+	    assert(!"Should not get here.");

+	 }

+      }

+   } else if (strcmp(callee, "inversesqrt") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_unop_rsq, type, op[0], NULL);

+   } else if (strcmp(callee, "length") == 0) {

+      return new(mem_ctx) ir_constant(sqrtf(dot(op[0], op[0])));

+   } else if (strcmp(callee, "lessThan") == 0) {

+      assert(op[0]->type->is_vector() && op[1] && op[1]->type->is_vector());

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.b[c] = op[0]->value.u[c] < op[1]->value.u[c];

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.b[c] = op[0]->value.i[c] < op[1]->value.i[c];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.b[c] = op[0]->value.f[c] < op[1]->value.f[c];

+	    break;

+	 default:

+	    assert(!"Should not get here.");

+	 }

+      }

+   } else if (strcmp(callee, "lessThanEqual") == 0) {

+      assert(op[0]->type->is_vector() && op[1] && op[1]->type->is_vector());

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.b[c] = op[0]->value.u[c] <= op[1]->value.u[c];

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.b[c] = op[0]->value.i[c] <= op[1]->value.i[c];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.b[c] = op[0]->value.f[c] <= op[1]->value.f[c];

+	    break;

+	 default:

+	    assert(!"Should not get here.");

+	 }

+      }

+   } else if (strcmp(callee, "log") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_unop_log, type, op[0], NULL);

+   } else if (strcmp(callee, "log2") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_unop_log2, type, op[0], NULL);

+   } else if (strcmp(callee, "matrixCompMult") == 0) {

+      assert(op[0]->type->is_float() && op[1]->type->is_float());

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.f[c] = op[0]->value.f[c] * op[1]->value.f[c];

+   } else if (strcmp(callee, "max") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_binop_max, type, op[0], op[1]);

+   } else if (strcmp(callee, "min") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_binop_min, type, op[0], op[1]);

+   } else if (strcmp(callee, "mix") == 0) {

+      assert(op[0]->type->is_float() && op[1]->type->is_float());

+      if (op[2]->type->is_float()) {

+	 unsigned c2_inc = op[2]->type->is_scalar() ? 0 : 1;

+	 unsigned components = op[0]->type->components();

+	 for (unsigned c = 0, c2 = 0; c < components; c2 += c2_inc, c++) {

+	    data.f[c] = op[0]->value.f[c] * (1 - op[2]->value.f[c2]) +

+			op[1]->value.f[c] * op[2]->value.f[c2];

+	 }

+      } else {

+	 assert(op[2]->type->is_boolean());

+	 for (unsigned c = 0; c < op[0]->type->components(); c++)

+	    data.f[c] = op[op[2]->value.b[c] ? 1 : 0]->value.f[c];

+      }

+   } else if (strcmp(callee, "mod") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_binop_mod, type, op[0], op[1]);

+   } else if (strcmp(callee, "normalize") == 0) {

+      assert(op[0]->type->is_float());

+      float length = sqrtf(dot(op[0], op[0]));

+

+      if (length == 0)

+	 return ir_constant::zero(mem_ctx, this->type);

+

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.f[c] = op[0]->value.f[c] / length;

+   } else if (strcmp(callee, "not") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_unop_logic_not, type, op[0], NULL);

+   } else if (strcmp(callee, "notEqual") == 0) {

+      assert(op[0]->type->is_vector() && op[1] && op[1]->type->is_vector());

+      for (unsigned c = 0; c < op[0]->type->components(); c++) {

+	 switch (op[0]->type->base_type) {

+	 case GLSL_TYPE_UINT:

+	    data.b[c] = op[0]->value.u[c] != op[1]->value.u[c];

+	    break;

+	 case GLSL_TYPE_INT:

+	    data.b[c] = op[0]->value.i[c] != op[1]->value.i[c];

+	    break;

+	 case GLSL_TYPE_FLOAT:

+	    data.b[c] = op[0]->value.f[c] != op[1]->value.f[c];

+	    break;

+	 case GLSL_TYPE_BOOL:

+	    data.b[c] = op[0]->value.b[c] != op[1]->value.b[c];

+	    break;

+	 default:

+	    assert(!"Should not get here.");

+	 }

+      }

+   } else if (strcmp(callee, "outerProduct") == 0) {

+      assert(op[0]->type->is_vector() && op[1]->type->is_vector());

+      const unsigned m = op[0]->type->vector_elements;

+      const unsigned n = op[1]->type->vector_elements;

+      for (unsigned j = 0; j < n; j++) {

+	 for (unsigned i = 0; i < m; i++) {

+	    data.f[i+m*j] = op[0]->value.f[i] * op[1]->value.f[j];

+	 }

+      }

+   } else if (strcmp(callee, "pow") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_binop_pow, type, op[0], op[1]);

+   } else if (strcmp(callee, "radians") == 0) {

+      assert(op[0]->type->is_float());

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.f[c] = M_PI / 180.0F * op[0]->value.f[c];

+   } else if (strcmp(callee, "reflect") == 0) {

+      assert(op[0]->type->is_float());

+      float dot_NI = dot(op[1], op[0]);

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.f[c] = op[0]->value.f[c] - 2 * dot_NI * op[1]->value.f[c];

+   } else if (strcmp(callee, "refract") == 0) {

+      const float eta = op[2]->value.f[0];

+      const float dot_NI = dot(op[1], op[0]);

+      const float k = 1.0F - eta * eta * (1.0F - dot_NI * dot_NI);

+      if (k < 0.0) {

+	 return ir_constant::zero(mem_ctx, this->type);

+      } else {

+	 for (unsigned c = 0; c < type->components(); c++) {

+	    data.f[c] = eta * op[0]->value.f[c] - (eta * dot_NI + sqrtf(k))

+			    * op[1]->value.f[c];

+	 }

+      }

+   } else if (strcmp(callee, "sign") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_unop_sign, type, op[0], NULL);

+   } else if (strcmp(callee, "sin") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_unop_sin, type, op[0], NULL);

+   } else if (strcmp(callee, "sinh") == 0) {

+      assert(op[0]->type->is_float());

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.f[c] = sinhf(op[0]->value.f[c]);

+   } else if (strcmp(callee, "smoothstep") == 0) {

+      assert(num_parameters == 3);

+      assert(op[1]->type == op[0]->type);

+      unsigned edge_inc = op[0]->type->is_scalar() ? 0 : 1;

+      for (unsigned c = 0, e = 0; c < type->components(); e += edge_inc, c++) {

+	 const float edge0 = op[0]->value.f[e];

+	 const float edge1 = op[1]->value.f[e];

+	 if (edge0 == edge1) {

+	    data.f[c] = 0.0; /* Avoid a crash - results are undefined anyway */

+	 } else {

+	    const float numerator = op[2]->value.f[c] - edge0;

+	    const float denominator = edge1 - edge0;

+	    const float t = CLAMP(numerator/denominator, 0, 1);

+	    data.f[c] = t * t * (3 - 2 * t);

+	 }

+      }

+   } else if (strcmp(callee, "sqrt") == 0) {

+      expr = new(mem_ctx) ir_expression(ir_unop_sqrt, type, op[0], NULL);

+   } else if (strcmp(callee, "step") == 0) {

+      assert(op[0]->type->is_float() && op[1]->type->is_float());

+      /* op[0] (edge) may be either a scalar or a vector */

+      const unsigned c0_inc = op[0]->type->is_scalar() ? 0 : 1;

+      for (unsigned c = 0, c0 = 0; c < type->components(); c0 += c0_inc, c++)

+	 data.f[c] = (op[1]->value.f[c] < op[0]->value.f[c0]) ? 0.0F : 1.0F;

+   } else if (strcmp(callee, "tan") == 0) {

+      assert(op[0]->type->is_float());

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.f[c] = tanf(op[0]->value.f[c]);

+   } else if (strcmp(callee, "tanh") == 0) {

+      assert(op[0]->type->is_float());

+      for (unsigned c = 0; c < op[0]->type->components(); c++)

+	 data.f[c] = tanhf(op[0]->value.f[c]);

+   } else if (strcmp(callee, "transpose") == 0) {

+      assert(op[0]->type->is_matrix());

+      const unsigned n = op[0]->type->vector_elements;

+      const unsigned m = op[0]->type->matrix_columns;

+      for (unsigned j = 0; j < m; j++) {

+	 for (unsigned i = 0; i < n; i++) {

+	    data.f[m*i+j] += op[0]->value.f[i+n*j];

+	 }

+      }

+   } else {

+      /* Unsupported builtin - some are not allowed in constant expressions. */

+      return NULL;

+   }

+

+   if (expr != NULL)

+      return expr->constant_expression_value();

+

+   return new(mem_ctx) ir_constant(this->type, &data);

+}

diff --git a/mesalib/src/glsl/linker.cpp b/mesalib/src/glsl/linker.cpp
index 46cd1950c..a9a2a5e20 100644
--- a/mesalib/src/glsl/linker.cpp
+++ b/mesalib/src/glsl/linker.cpp
@@ -1,1713 +1,1713 @@
-/*
- * Copyright © 2010 Intel Corporation
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
- * DEALINGS IN THE SOFTWARE.
- */
-
-/**
- * \file linker.cpp
- * GLSL linker implementation
- *
- * Given a set of shaders that are to be linked to generate a final program,
- * there are three distinct stages.
- *
- * In the first stage shaders are partitioned into groups based on the shader
- * type.  All shaders of a particular type (e.g., vertex shaders) are linked
- * together.
- *
- *   - Undefined references in each shader are resolve to definitions in
- *     another shader.
- *   - Types and qualifiers of uniforms, outputs, and global variables defined
- *     in multiple shaders with the same name are verified to be the same.
- *   - Initializers for uniforms and global variables defined
- *     in multiple shaders with the same name are verified to be the same.
- *
- * The result, in the terminology of the GLSL spec, is a set of shader
- * executables for each processing unit.
- *
- * After the first stage is complete, a series of semantic checks are performed
- * on each of the shader executables.
- *
- *   - Each shader executable must define a \c main function.
- *   - Each vertex shader executable must write to \c gl_Position.
- *   - Each fragment shader executable must write to either \c gl_FragData or
- *     \c gl_FragColor.
- *
- * In the final stage individual shader executables are linked to create a
- * complete exectuable.
- *
- *   - Types of uniforms defined in multiple shader stages with the same name
- *     are verified to be the same.
- *   - Initializers for uniforms defined in multiple shader stages with the
- *     same name are verified to be the same.
- *   - Types and qualifiers of outputs defined in one stage are verified to
- *     be the same as the types and qualifiers of inputs defined with the same
- *     name in a later stage.
- *
- * \author Ian Romanick <ian.d.romanick@intel.com>
- */
-#include <cstdlib>
-#include <cstdio>
-#include <cstdarg>
-#include <climits>
-
-#include "main/core.h"
-#include "glsl_symbol_table.h"
-#include "ir.h"
-#include "program.h"
-#include "program/hash_table.h"
-#include "linker.h"
-#include "ir_optimization.h"
-
-extern "C" {
-#include "main/shaderobj.h"
-}
-
-/**
- * Visitor that determines whether or not a variable is ever written.
- */
-class find_assignment_visitor : public ir_hierarchical_visitor {
-public:
-   find_assignment_visitor(const char *name)
-      : name(name), found(false)
-   {
-      /* empty */
-   }
-
-   virtual ir_visitor_status visit_enter(ir_assignment *ir)
-   {
-      ir_variable *const var = ir->lhs->variable_referenced();
-
-      if (strcmp(name, var->name) == 0) {
-	 found = true;
-	 return visit_stop;
-      }
-
-      return visit_continue_with_parent;
-   }
-
-   virtual ir_visitor_status visit_enter(ir_call *ir)
-   {
-      exec_list_iterator sig_iter = ir->get_callee()->parameters.iterator();
-      foreach_iter(exec_list_iterator, iter, *ir) {
-	 ir_rvalue *param_rval = (ir_rvalue *)iter.get();
-	 ir_variable *sig_param = (ir_variable *)sig_iter.get();
-
-	 if (sig_param->mode == ir_var_out ||
-	     sig_param->mode == ir_var_inout) {
-	    ir_variable *var = param_rval->variable_referenced();
-	    if (var && strcmp(name, var->name) == 0) {
-	       found = true;
-	       return visit_stop;
-	    }
-	 }
-	 sig_iter.next();
-      }
-
-      return visit_continue_with_parent;
-   }
-
-   bool variable_found()
-   {
-      return found;
-   }
-
-private:
-   const char *name;       /**< Find writes to a variable with this name. */
-   bool found;             /**< Was a write to the variable found? */
-};
-
-
-/**
- * Visitor that determines whether or not a variable is ever read.
- */
-class find_deref_visitor : public ir_hierarchical_visitor {
-public:
-   find_deref_visitor(const char *name)
-      : name(name), found(false)
-   {
-      /* empty */
-   }
-
-   virtual ir_visitor_status visit(ir_dereference_variable *ir)
-   {
-      if (strcmp(this->name, ir->var->name) == 0) {
-	 this->found = true;
-	 return visit_stop;
-      }
-
-      return visit_continue;
-   }
-
-   bool variable_found() const
-   {
-      return this->found;
-   }
-
-private:
-   const char *name;       /**< Find writes to a variable with this name. */
-   bool found;             /**< Was a write to the variable found? */
-};
-
-
-void
-linker_error_printf(gl_shader_program *prog, const char *fmt, ...)
-{
-   va_list ap;
-
-   ralloc_strcat(&prog->InfoLog, "error: ");
-   va_start(ap, fmt);
-   ralloc_vasprintf_append(&prog->InfoLog, fmt, ap);
-   va_end(ap);
-}
-
-
-void
-invalidate_variable_locations(gl_shader *sh, enum ir_variable_mode mode,
-			      int generic_base)
-{
-   foreach_list(node, sh->ir) {
-      ir_variable *const var = ((ir_instruction *) node)->as_variable();
-
-      if ((var == NULL) || (var->mode != (unsigned) mode))
-	 continue;
-
-      /* Only assign locations for generic attributes / varyings / etc.
-       */
-      if ((var->location >= generic_base) && !var->explicit_location)
-	  var->location = -1;
-   }
-}
-
-
-/**
- * Determine the number of attribute slots required for a particular type
- *
- * This code is here because it implements the language rules of a specific
- * GLSL version.  Since it's a property of the language and not a property of
- * types in general, it doesn't really belong in glsl_type.
- */
-unsigned
-count_attribute_slots(const glsl_type *t)
-{
-   /* From page 31 (page 37 of the PDF) of the GLSL 1.50 spec:
-    *
-    *     "A scalar input counts the same amount against this limit as a vec4,
-    *     so applications may want to consider packing groups of four
-    *     unrelated float inputs together into a vector to better utilize the
-    *     capabilities of the underlying hardware. A matrix input will use up
-    *     multiple locations.  The number of locations used will equal the
-    *     number of columns in the matrix."
-    *
-    * The spec does not explicitly say how arrays are counted.  However, it
-    * should be safe to assume the total number of slots consumed by an array
-    * is the number of entries in the array multiplied by the number of slots
-    * consumed by a single element of the array.
-    */
-
-   if (t->is_array())
-      return t->array_size() * count_attribute_slots(t->element_type());
-
-   if (t->is_matrix())
-      return t->matrix_columns;
-
-   return 1;
-}
-
-
-/**
- * Verify that a vertex shader executable meets all semantic requirements
- *
- * \param shader  Vertex shader executable to be verified
- */
-bool
-validate_vertex_shader_executable(struct gl_shader_program *prog,
-				  struct gl_shader *shader)
-{
-   if (shader == NULL)
-      return true;
-
-   find_assignment_visitor find("gl_Position");
-   find.run(shader->ir);
-   if (!find.variable_found()) {
-      linker_error_printf(prog,
-			  "vertex shader does not write to `gl_Position'\n");
-      return false;
-   }
-
-   return true;
-}
-
-
-/**
- * Verify that a fragment shader executable meets all semantic requirements
- *
- * \param shader  Fragment shader executable to be verified
- */
-bool
-validate_fragment_shader_executable(struct gl_shader_program *prog,
-				    struct gl_shader *shader)
-{
-   if (shader == NULL)
-      return true;
-
-   find_assignment_visitor frag_color("gl_FragColor");
-   find_assignment_visitor frag_data("gl_FragData");
-
-   frag_color.run(shader->ir);
-   frag_data.run(shader->ir);
-
-   if (frag_color.variable_found() && frag_data.variable_found()) {
-      linker_error_printf(prog,  "fragment shader writes to both "
-			  "`gl_FragColor' and `gl_FragData'\n");
-      return false;
-   }
-
-   return true;
-}
-
-
-/**
- * Generate a string describing the mode of a variable
- */
-static const char *
-mode_string(const ir_variable *var)
-{
-   switch (var->mode) {
-   case ir_var_auto:
-      return (var->read_only) ? "global constant" : "global variable";
-
-   case ir_var_uniform: return "uniform";
-   case ir_var_in:      return "shader input";
-   case ir_var_out:     return "shader output";
-   case ir_var_inout:   return "shader inout";
-
-   case ir_var_const_in:
-   case ir_var_temporary:
-   default:
-      assert(!"Should not get here.");
-      return "invalid variable";
-   }
-}
-
-
-/**
- * Perform validation of global variables used across multiple shaders
- */
-bool
-cross_validate_globals(struct gl_shader_program *prog,
-		       struct gl_shader **shader_list,
-		       unsigned num_shaders,
-		       bool uniforms_only)
-{
-   /* Examine all of the uniforms in all of the shaders and cross validate
-    * them.
-    */
-   glsl_symbol_table variables;
-   for (unsigned i = 0; i < num_shaders; i++) {
-      if (shader_list[i] == NULL)
-	 continue;
-
-      foreach_list(node, shader_list[i]->ir) {
-	 ir_variable *const var = ((ir_instruction *) node)->as_variable();
-
-	 if (var == NULL)
-	    continue;
-
-	 if (uniforms_only && (var->mode != ir_var_uniform))
-	    continue;
-
-	 /* Don't cross validate temporaries that are at global scope.  These
-	  * will eventually get pulled into the shaders 'main'.
-	  */
-	 if (var->mode == ir_var_temporary)
-	    continue;
-
-	 /* If a global with this name has already been seen, verify that the
-	  * new instance has the same type.  In addition, if the globals have
-	  * initializers, the values of the initializers must be the same.
-	  */
-	 ir_variable *const existing = variables.get_variable(var->name);
-	 if (existing != NULL) {
-	    if (var->type != existing->type) {
-	       /* Consider the types to be "the same" if both types are arrays
-		* of the same type and one of the arrays is implicitly sized.
-		* In addition, set the type of the linked variable to the
-		* explicitly sized array.
-		*/
-	       if (var->type->is_array()
-		   && existing->type->is_array()
-		   && (var->type->fields.array == existing->type->fields.array)
-		   && ((var->type->length == 0)
-		       || (existing->type->length == 0))) {
-		  if (var->type->length != 0) {
-		     existing->type = var->type;
-		  }
-	       } else {
-		  linker_error_printf(prog, "%s `%s' declared as type "
-				      "`%s' and type `%s'\n",
-				      mode_string(var),
-				      var->name, var->type->name,
-				      existing->type->name);
-		  return false;
-	       }
-	    }
-
-	    if (var->explicit_location) {
-	       if (existing->explicit_location
-		   && (var->location != existing->location)) {
-		     linker_error_printf(prog, "explicit locations for %s "
-					 "`%s' have differing values\n",
-					 mode_string(var), var->name);
-		     return false;
-	       }
-
-	       existing->location = var->location;
-	       existing->explicit_location = true;
-	    }
-
-        /* Validate layout qualifiers for gl_FragDepth.
-         *
-         * From the AMD_conservative_depth spec:
-         *    "If gl_FragDepth is redeclared in any fragment shader in
-         *    a program, it must be redeclared in all fragment shaders in that
-         *    program that have static assignments to gl_FragDepth. All
-         *    redeclarations of gl_FragDepth in all fragment shaders in
-         *    a single program must have the same set of qualifiers."
-         */
-        if (strcmp(var->name, "gl_FragDepth") == 0) {
-           bool layout_declared = var->depth_layout != ir_depth_layout_none;
-           bool layout_differs = var->depth_layout != existing->depth_layout;
-           if (layout_declared && layout_differs) {
-              linker_error_printf(prog,
-                 "All redeclarations of gl_FragDepth in all fragment shaders "
-                 "in a single program must have the same set of qualifiers.");
-           }
-           if (var->used && layout_differs) {
-              linker_error_printf(prog,
-                    "If gl_FragDepth is redeclared with a layout qualifier in"
-                    "any fragment shader, it must be redeclared with the same"
-                    "layout qualifier in all fragment shaders that have"
-                    "assignments to gl_FragDepth");
-           }
-        }
-
-	    /* FINISHME: Handle non-constant initializers.
-	     */
-	    if (var->constant_value != NULL) {
-	       if (existing->constant_value != NULL) {
-		  if (!var->constant_value->has_value(existing->constant_value)) {
-		     linker_error_printf(prog, "initializers for %s "
-					 "`%s' have differing values\n",
-					 mode_string(var), var->name);
-		     return false;
-		  }
-	       } else
-		  /* If the first-seen instance of a particular uniform did not
-		   * have an initializer but a later instance does, copy the
-		   * initializer to the version stored in the symbol table.
-		   */
-		  /* FINISHME: This is wrong.  The constant_value field should
-		   * FINISHME: not be modified!  Imagine a case where a shader
-		   * FINISHME: without an initializer is linked in two different
-		   * FINISHME: programs with shaders that have differing
-		   * FINISHME: initializers.  Linking with the first will
-		   * FINISHME: modify the shader, and linking with the second
-		   * FINISHME: will fail.
-		   */
-		  existing->constant_value =
-		     var->constant_value->clone(ralloc_parent(existing), NULL);
-	    }
-
-	    if (existing->invariant != var->invariant) {
-	       linker_error_printf(prog, "declarations for %s `%s' have "
-	                           "mismatching invariant qualifiers\n",
-	                           mode_string(var), var->name);
-	       return false;
-	    }
-            if (existing->centroid != var->centroid) {
-               linker_error_printf(prog, "declarations for %s `%s' have "
-                                   "mismatching centroid qualifiers\n",
-                                   mode_string(var), var->name);
-               return false;
-            }
-	 } else
-	    variables.add_variable(var);
-      }
-   }
-
-   return true;
-}
-
-
-/**
- * Perform validation of uniforms used across multiple shader stages
- */
-bool
-cross_validate_uniforms(struct gl_shader_program *prog)
-{
-   return cross_validate_globals(prog, prog->_LinkedShaders,
-				 MESA_SHADER_TYPES, true);
-}
-
-
-/**
- * Validate that outputs from one stage match inputs of another
- */
-bool
-cross_validate_outputs_to_inputs(struct gl_shader_program *prog,
-				 gl_shader *producer, gl_shader *consumer)
-{
-   glsl_symbol_table parameters;
-   /* FINISHME: Figure these out dynamically. */
-   const char *const producer_stage = "vertex";
-   const char *const consumer_stage = "fragment";
-
-   /* Find all shader outputs in the "producer" stage.
-    */
-   foreach_list(node, producer->ir) {
-      ir_variable *const var = ((ir_instruction *) node)->as_variable();
-
-      /* FINISHME: For geometry shaders, this should also look for inout
-       * FINISHME: variables.
-       */
-      if ((var == NULL) || (var->mode != ir_var_out))
-	 continue;
-
-      parameters.add_variable(var);
-   }
-
-
-   /* Find all shader inputs in the "consumer" stage.  Any variables that have
-    * matching outputs already in the symbol table must have the same type and
-    * qualifiers.
-    */
-   foreach_list(node, consumer->ir) {
-      ir_variable *const input = ((ir_instruction *) node)->as_variable();
-
-      /* FINISHME: For geometry shaders, this should also look for inout
-       * FINISHME: variables.
-       */
-      if ((input == NULL) || (input->mode != ir_var_in))
-	 continue;
-
-      ir_variable *const output = parameters.get_variable(input->name);
-      if (output != NULL) {
-	 /* Check that the types match between stages.
-	  */
-	 if (input->type != output->type) {
-	    /* There is a bit of a special case for gl_TexCoord.  This
-	     * built-in is unsized by default.  Appliations that variable
-	     * access it must redeclare it with a size.  There is some
-	     * language in the GLSL spec that implies the fragment shader
-	     * and vertex shader do not have to agree on this size.  Other
-	     * driver behave this way, and one or two applications seem to
-	     * rely on it.
-	     *
-	     * Neither declaration needs to be modified here because the array
-	     * sizes are fixed later when update_array_sizes is called.
-	     *
-	     * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec:
-	     *
-	     *     "Unlike user-defined varying variables, the built-in
-	     *     varying variables don't have a strict one-to-one
-	     *     correspondence between the vertex language and the
-	     *     fragment language."
-	     */
-	    if (!output->type->is_array()
-		|| (strncmp("gl_", output->name, 3) != 0)) {
-	       linker_error_printf(prog,
-				   "%s shader output `%s' declared as "
-				   "type `%s', but %s shader input declared "
-				   "as type `%s'\n",
-				   producer_stage, output->name,
-				   output->type->name,
-				   consumer_stage, input->type->name);
-	       return false;
-	    }
-	 }
-
-	 /* Check that all of the qualifiers match between stages.
-	  */
-	 if (input->centroid != output->centroid) {
-	    linker_error_printf(prog,
-				"%s shader output `%s' %s centroid qualifier, "
-				"but %s shader input %s centroid qualifier\n",
-				producer_stage,
-				output->name,
-				(output->centroid) ? "has" : "lacks",
-				consumer_stage,
-				(input->centroid) ? "has" : "lacks");
-	    return false;
-	 }
-
-	 if (input->invariant != output->invariant) {
-	    linker_error_printf(prog,
-				"%s shader output `%s' %s invariant qualifier, "
-				"but %s shader input %s invariant qualifier\n",
-				producer_stage,
-				output->name,
-				(output->invariant) ? "has" : "lacks",
-				consumer_stage,
-				(input->invariant) ? "has" : "lacks");
-	    return false;
-	 }
-
-	 if (input->interpolation != output->interpolation) {
-	    linker_error_printf(prog,
-				"%s shader output `%s' specifies %s "
-				"interpolation qualifier, "
-				"but %s shader input specifies %s "
-				"interpolation qualifier\n",
-				producer_stage,
-				output->name,
-				output->interpolation_string(),
-				consumer_stage,
-				input->interpolation_string());
-	    return false;
-	 }
-      }
-   }
-
-   return true;
-}
-
-
-/**
- * Populates a shaders symbol table with all global declarations
- */
-static void
-populate_symbol_table(gl_shader *sh)
-{
-   sh->symbols = new(sh) glsl_symbol_table;
-
-   foreach_list(node, sh->ir) {
-      ir_instruction *const inst = (ir_instruction *) node;
-      ir_variable *var;
-      ir_function *func;
-
-      if ((func = inst->as_function()) != NULL) {
-	 sh->symbols->add_function(func);
-      } else if ((var = inst->as_variable()) != NULL) {
-	 sh->symbols->add_variable(var);
-      }
-   }
-}
-
-
-/**
- * Remap variables referenced in an instruction tree
- *
- * This is used when instruction trees are cloned from one shader and placed in
- * another.  These trees will contain references to \c ir_variable nodes that
- * do not exist in the target shader.  This function finds these \c ir_variable
- * references and replaces the references with matching variables in the target
- * shader.
- *
- * If there is no matching variable in the target shader, a clone of the
- * \c ir_variable is made and added to the target shader.  The new variable is
- * added to \b both the instruction stream and the symbol table.
- *
- * \param inst         IR tree that is to be processed.
- * \param symbols      Symbol table containing global scope symbols in the
- *                     linked shader.
- * \param instructions Instruction stream where new variable declarations
- *                     should be added.
- */
-void
-remap_variables(ir_instruction *inst, struct gl_shader *target,
-		hash_table *temps)
-{
-   class remap_visitor : public ir_hierarchical_visitor {
-   public:
-	 remap_visitor(struct gl_shader *target,
-		    hash_table *temps)
-      {
-	 this->target = target;
-	 this->symbols = target->symbols;
-	 this->instructions = target->ir;
-	 this->temps = temps;
-      }
-
-      virtual ir_visitor_status visit(ir_dereference_variable *ir)
-      {
-	 if (ir->var->mode == ir_var_temporary) {
-	    ir_variable *var = (ir_variable *) hash_table_find(temps, ir->var);
-
-	    assert(var != NULL);
-	    ir->var = var;
-	    return visit_continue;
-	 }
-
-	 ir_variable *const existing =
-	    this->symbols->get_variable(ir->var->name);
-	 if (existing != NULL)
-	    ir->var = existing;
-	 else {
-	    ir_variable *copy = ir->var->clone(this->target, NULL);
-
-	    this->symbols->add_variable(copy);
-	    this->instructions->push_head(copy);
-	    ir->var = copy;
-	 }
-
-	 return visit_continue;
-      }
-
-   private:
-      struct gl_shader *target;
-      glsl_symbol_table *symbols;
-      exec_list *instructions;
-      hash_table *temps;
-   };
-
-   remap_visitor v(target, temps);
-
-   inst->accept(&v);
-}
-
-
-/**
- * Move non-declarations from one instruction stream to another
- *
- * The intended usage pattern of this function is to pass the pointer to the
- * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
- * pointer) for \c last and \c false for \c make_copies on the first
- * call.  Successive calls pass the return value of the previous call for
- * \c last and \c true for \c make_copies.
- *
- * \param instructions Source instruction stream
- * \param last         Instruction after which new instructions should be
- *                     inserted in the target instruction stream
- * \param make_copies  Flag selecting whether instructions in \c instructions
- *                     should be copied (via \c ir_instruction::clone) into the
- *                     target list or moved.
- *
- * \return
- * The new "last" instruction in the target instruction stream.  This pointer
- * is suitable for use as the \c last parameter of a later call to this
- * function.
- */
-exec_node *
-move_non_declarations(exec_list *instructions, exec_node *last,
-		      bool make_copies, gl_shader *target)
-{
-   hash_table *temps = NULL;
-
-   if (make_copies)
-      temps = hash_table_ctor(0, hash_table_pointer_hash,
-			      hash_table_pointer_compare);
-
-   foreach_list_safe(node, instructions) {
-      ir_instruction *inst = (ir_instruction *) node;
-
-      if (inst->as_function())
-	 continue;
-
-      ir_variable *var = inst->as_variable();
-      if ((var != NULL) && (var->mode != ir_var_temporary))
-	 continue;
-
-      assert(inst->as_assignment()
-	     || ((var != NULL) && (var->mode == ir_var_temporary)));
-
-      if (make_copies) {
-	 inst = inst->clone(target, NULL);
-
-	 if (var != NULL)
-	    hash_table_insert(temps, inst, var);
-	 else
-	    remap_variables(inst, target, temps);
-      } else {
-	 inst->remove();
-      }
-
-      last->insert_after(inst);
-      last = inst;
-   }
-
-   if (make_copies)
-      hash_table_dtor(temps);
-
-   return last;
-}
-
-/**
- * Get the function signature for main from a shader
- */
-static ir_function_signature *
-get_main_function_signature(gl_shader *sh)
-{
-   ir_function *const f = sh->symbols->get_function("main");
-   if (f != NULL) {
-      exec_list void_parameters;
-
-      /* Look for the 'void main()' signature and ensure that it's defined.
-       * This keeps the linker from accidentally pick a shader that just
-       * contains a prototype for main.
-       *
-       * We don't have to check for multiple definitions of main (in multiple
-       * shaders) because that would have already been caught above.
-       */
-      ir_function_signature *sig = f->matching_signature(&void_parameters);
-      if ((sig != NULL) && sig->is_defined) {
-	 return sig;
-      }
-   }
-
-   return NULL;
-}
-
-
-/**
- * Combine a group of shaders for a single stage to generate a linked shader
- *
- * \note
- * If this function is supplied a single shader, it is cloned, and the new
- * shader is returned.
- */
-static struct gl_shader *
-link_intrastage_shaders(void *mem_ctx,
-			struct gl_context *ctx,
-			struct gl_shader_program *prog,
-			struct gl_shader **shader_list,
-			unsigned num_shaders)
-{
-   /* Check that global variables defined in multiple shaders are consistent.
-    */
-   if (!cross_validate_globals(prog, shader_list, num_shaders, false))
-      return NULL;
-
-   /* Check that there is only a single definition of each function signature
-    * across all shaders.
-    */
-   for (unsigned i = 0; i < (num_shaders - 1); i++) {
-      foreach_list(node, shader_list[i]->ir) {
-	 ir_function *const f = ((ir_instruction *) node)->as_function();
-
-	 if (f == NULL)
-	    continue;
-
-	 for (unsigned j = i + 1; j < num_shaders; j++) {
-	    ir_function *const other =
-	       shader_list[j]->symbols->get_function(f->name);
-
-	    /* If the other shader has no function (and therefore no function
-	     * signatures) with the same name, skip to the next shader.
-	     */
-	    if (other == NULL)
-	       continue;
-
-	    foreach_iter (exec_list_iterator, iter, *f) {
-	       ir_function_signature *sig =
-		  (ir_function_signature *) iter.get();
-
-	       if (!sig->is_defined || sig->is_builtin)
-		  continue;
-
-	       ir_function_signature *other_sig =
-		  other->exact_matching_signature(& sig->parameters);
-
-	       if ((other_sig != NULL) && other_sig->is_defined
-		   && !other_sig->is_builtin) {
-		  linker_error_printf(prog,
-				      "function `%s' is multiply defined",
-				      f->name);
-		  return NULL;
-	       }
-	    }
-	 }
-      }
-   }
-
-   /* Find the shader that defines main, and make a clone of it.
-    *
-    * Starting with the clone, search for undefined references.  If one is
-    * found, find the shader that defines it.  Clone the reference and add
-    * it to the shader.  Repeat until there are no undefined references or
-    * until a reference cannot be resolved.
-    */
-   gl_shader *main = NULL;
-   for (unsigned i = 0; i < num_shaders; i++) {
-      if (get_main_function_signature(shader_list[i]) != NULL) {
-	 main = shader_list[i];
-	 break;
-      }
-   }
-
-   if (main == NULL) {
-      linker_error_printf(prog, "%s shader lacks `main'\n",
-			  (shader_list[0]->Type == GL_VERTEX_SHADER)
-			  ? "vertex" : "fragment");
-      return NULL;
-   }
-
-   gl_shader *linked = ctx->Driver.NewShader(NULL, 0, main->Type);
-   linked->ir = new(linked) exec_list;
-   clone_ir_list(mem_ctx, linked->ir, main->ir);
-
-   populate_symbol_table(linked);
-
-   /* The a pointer to the main function in the final linked shader (i.e., the
-    * copy of the original shader that contained the main function).
-    */
-   ir_function_signature *const main_sig = get_main_function_signature(linked);
-
-   /* Move any instructions other than variable declarations or function
-    * declarations into main.
-    */
-   exec_node *insertion_point =
-      move_non_declarations(linked->ir, (exec_node *) &main_sig->body, false,
-			    linked);
-
-   for (unsigned i = 0; i < num_shaders; i++) {
-      if (shader_list[i] == main)
-	 continue;
-
-      insertion_point = move_non_declarations(shader_list[i]->ir,
-					      insertion_point, true, linked);
-   }
-
-   /* Resolve initializers for global variables in the linked shader.
-    */
-   unsigned num_linking_shaders = num_shaders;
-   for (unsigned i = 0; i < num_shaders; i++)
-      num_linking_shaders += shader_list[i]->num_builtins_to_link;
-
-   gl_shader **linking_shaders =
-      (gl_shader **) calloc(num_linking_shaders, sizeof(gl_shader *));
-
-   memcpy(linking_shaders, shader_list,
-	  sizeof(linking_shaders[0]) * num_shaders);
-
-   unsigned idx = num_shaders;
-   for (unsigned i = 0; i < num_shaders; i++) {
-      memcpy(&linking_shaders[idx], shader_list[i]->builtins_to_link,
-	     sizeof(linking_shaders[0]) * shader_list[i]->num_builtins_to_link);
-      idx += shader_list[i]->num_builtins_to_link;
-   }
-
-   assert(idx == num_linking_shaders);
-
-   if (!link_function_calls(prog, linked, linking_shaders,
-			    num_linking_shaders)) {
-      ctx->Driver.DeleteShader(ctx, linked);
-      linked = NULL;
-   }
-
-   free(linking_shaders);
-
-   /* Make a pass over all variable declarations to ensure that arrays with
-    * unspecified sizes have a size specified.  The size is inferred from the
-    * max_array_access field.
-    */
-   if (linked != NULL) {
-      class array_sizing_visitor : public ir_hierarchical_visitor {
-      public:
-	 virtual ir_visitor_status visit(ir_variable *var)
-	 {
-	    if (var->type->is_array() && (var->type->length == 0)) {
-	       const glsl_type *type =
-		  glsl_type::get_array_instance(var->type->fields.array,
-						var->max_array_access);
-
-	       assert(type != NULL);
-	       var->type = type;
-	    }
-
-	    return visit_continue;
-	 }
-      } v;
-
-      v.run(linked->ir);
-   }
-
-   return linked;
-}
-
-
-struct uniform_node {
-   exec_node link;
-   struct gl_uniform *u;
-   unsigned slots;
-};
-
-/**
- * Update the sizes of linked shader uniform arrays to the maximum
- * array index used.
- *
- * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
- *
- *     If one or more elements of an array are active,
- *     GetActiveUniform will return the name of the array in name,
- *     subject to the restrictions listed above. The type of the array
- *     is returned in type. The size parameter contains the highest
- *     array element index used, plus one. The compiler or linker
- *     determines the highest index used.  There will be only one
- *     active uniform reported by the GL per uniform array.
-
- */
-static void
-update_array_sizes(struct gl_shader_program *prog)
-{
-   for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {
-	 if (prog->_LinkedShaders[i] == NULL)
-	    continue;
-
-      foreach_list(node, prog->_LinkedShaders[i]->ir) {
-	 ir_variable *const var = ((ir_instruction *) node)->as_variable();
-
-	 if ((var == NULL) || (var->mode != ir_var_uniform &&
-			       var->mode != ir_var_in &&
-			       var->mode != ir_var_out) ||
-	     !var->type->is_array())
-	    continue;
-
-	 unsigned int size = var->max_array_access;
-	 for (unsigned j = 0; j < MESA_SHADER_TYPES; j++) {
-	       if (prog->_LinkedShaders[j] == NULL)
-		  continue;
-
-	    foreach_list(node2, prog->_LinkedShaders[j]->ir) {
-	       ir_variable *other_var = ((ir_instruction *) node2)->as_variable();
-	       if (!other_var)
-		  continue;
-
-	       if (strcmp(var->name, other_var->name) == 0 &&
-		   other_var->max_array_access > size) {
-		  size = other_var->max_array_access;
-	       }
-	    }
-	 }
-
-	 if (size + 1 != var->type->fields.array->length) {
-	    var->type = glsl_type::get_array_instance(var->type->fields.array,
-						      size + 1);
-	    /* FINISHME: We should update the types of array
-	     * dereferences of this variable now.
-	     */
-	 }
-      }
-   }
-}
-
-static void
-add_uniform(void *mem_ctx, exec_list *uniforms, struct hash_table *ht,
-	    const char *name, const glsl_type *type, GLenum shader_type,
-	    unsigned *next_shader_pos, unsigned *total_uniforms)
-{
-   if (type->is_record()) {
-      for (unsigned int i = 0; i < type->length; i++) {
-	 const glsl_type *field_type = type->fields.structure[i].type;
-	 char *field_name = ralloc_asprintf(mem_ctx, "%s.%s", name,
-					    type->fields.structure[i].name);
-
-	 add_uniform(mem_ctx, uniforms, ht, field_name, field_type,
-		     shader_type, next_shader_pos, total_uniforms);
-      }
-   } else {
-      uniform_node *n = (uniform_node *) hash_table_find(ht, name);
-      unsigned int vec4_slots;
-      const glsl_type *array_elem_type = NULL;
-
-      if (type->is_array()) {
-	 array_elem_type = type->fields.array;
-	 /* Array of structures. */
-	 if (array_elem_type->is_record()) {
-	    for (unsigned int i = 0; i < type->length; i++) {
-	       char *elem_name = ralloc_asprintf(mem_ctx, "%s[%d]", name, i);
-	       add_uniform(mem_ctx, uniforms, ht, elem_name, array_elem_type,
-			   shader_type, next_shader_pos, total_uniforms);
-	    }
-	    return;
-	 }
-      }
-
-      /* Fix the storage size of samplers at 1 vec4 each. Be sure to pad out
-       * vectors to vec4 slots.
-       */
-      if (type->is_array()) {
-	 if (array_elem_type->is_sampler())
-	    vec4_slots = type->length;
-	 else
-	    vec4_slots = type->length * array_elem_type->matrix_columns;
-      } else if (type->is_sampler()) {
-	 vec4_slots = 1;
-      } else {
-	 vec4_slots = type->matrix_columns;
-      }
-
-      if (n == NULL) {
-	 n = (uniform_node *) calloc(1, sizeof(struct uniform_node));
-	 n->u = (gl_uniform *) calloc(1, sizeof(struct gl_uniform));
-	 n->slots = vec4_slots;
-
-	 n->u->Name = strdup(name);
-	 n->u->Type = type;
-	 n->u->VertPos = -1;
-	 n->u->FragPos = -1;
-	 n->u->GeomPos = -1;
-	 (*total_uniforms)++;
-
-	 hash_table_insert(ht, n, name);
-	 uniforms->push_tail(& n->link);
-      }
-
-      switch (shader_type) {
-      case GL_VERTEX_SHADER:
-	 n->u->VertPos = *next_shader_pos;
-	 break;
-      case GL_FRAGMENT_SHADER:
-	 n->u->FragPos = *next_shader_pos;
-	 break;
-      case GL_GEOMETRY_SHADER:
-	 n->u->GeomPos = *next_shader_pos;
-	 break;
-      }
-
-      (*next_shader_pos) += vec4_slots;
-   }
-}
-
-void
-assign_uniform_locations(struct gl_shader_program *prog)
-{
-   /* */
-   exec_list uniforms;
-   unsigned total_uniforms = 0;
-   hash_table *ht = hash_table_ctor(32, hash_table_string_hash,
-				    hash_table_string_compare);
-   void *mem_ctx = ralloc_context(NULL);
-
-   for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {
-      if (prog->_LinkedShaders[i] == NULL)
-	 continue;
-
-      unsigned next_position = 0;
-
-      foreach_list(node, prog->_LinkedShaders[i]->ir) {
-	 ir_variable *const var = ((ir_instruction *) node)->as_variable();
-
-	 if ((var == NULL) || (var->mode != ir_var_uniform))
-	    continue;
-
-	 if (strncmp(var->name, "gl_", 3) == 0) {
-	    /* At the moment, we don't allocate uniform locations for
-	     * builtin uniforms.  It's permitted by spec, and we'll
-	     * likely switch to doing that at some point, but not yet.
-	     */
-	    continue;
-	 }
-
-	 var->location = next_position;
-	 add_uniform(mem_ctx, &uniforms, ht, var->name, var->type,
-		     prog->_LinkedShaders[i]->Type,
-		     &next_position, &total_uniforms);
-      }
-   }
-
-   ralloc_free(mem_ctx);
-
-   gl_uniform_list *ul = (gl_uniform_list *)
-      calloc(1, sizeof(gl_uniform_list));
-
-   ul->Size = total_uniforms;
-   ul->NumUniforms = total_uniforms;
-   ul->Uniforms = (gl_uniform *) calloc(total_uniforms, sizeof(gl_uniform));
-
-   unsigned idx = 0;
-   uniform_node *next;
-   for (uniform_node *node = (uniform_node *) uniforms.head
-	   ; node->link.next != NULL
-	   ; node = next) {
-      next = (uniform_node *) node->link.next;
-
-      node->link.remove();
-      memcpy(&ul->Uniforms[idx], node->u, sizeof(gl_uniform));
-      idx++;
-
-      free(node->u);
-      free(node);
-   }
-
-   hash_table_dtor(ht);
-
-   prog->Uniforms = ul;
-}
-
-
-/**
- * Find a contiguous set of available bits in a bitmask
- *
- * \param used_mask     Bits representing used (1) and unused (0) locations
- * \param needed_count  Number of contiguous bits needed.
- *
- * \return
- * Base location of the available bits on success or -1 on failure.
- */
-int
-find_available_slots(unsigned used_mask, unsigned needed_count)
-{
-   unsigned needed_mask = (1 << needed_count) - 1;
-   const int max_bit_to_test = (8 * sizeof(used_mask)) - needed_count;
-
-   /* The comparison to 32 is redundant, but without it GCC emits "warning:
-    * cannot optimize possibly infinite loops" for the loop below.
-    */
-   if ((needed_count == 0) || (max_bit_to_test < 0) || (max_bit_to_test > 32))
-      return -1;
-
-   for (int i = 0; i <= max_bit_to_test; i++) {
-      if ((needed_mask & ~used_mask) == needed_mask)
-	 return i;
-
-      needed_mask <<= 1;
-   }
-
-   return -1;
-}
-
-
-bool
-assign_attribute_locations(gl_shader_program *prog, unsigned max_attribute_index)
-{
-   /* Mark invalid attribute locations as being used.
-    */
-   unsigned used_locations = (max_attribute_index >= 32)
-      ? ~0 : ~((1 << max_attribute_index) - 1);
-
-   gl_shader *const sh = prog->_LinkedShaders[0];
-   assert(sh->Type == GL_VERTEX_SHADER);
-
-   /* Operate in a total of four passes.
-    *
-    * 1. Invalidate the location assignments for all vertex shader inputs.
-    *
-    * 2. Assign locations for inputs that have user-defined (via
-    *    glBindVertexAttribLocation) locatoins.
-    *
-    * 3. Sort the attributes without assigned locations by number of slots
-    *    required in decreasing order.  Fragmentation caused by attribute
-    *    locations assigned by the application may prevent large attributes
-    *    from having enough contiguous space.
-    *
-    * 4. Assign locations to any inputs without assigned locations.
-    */
-
-   invalidate_variable_locations(sh, ir_var_in, VERT_ATTRIB_GENERIC0);
-
-   if (prog->Attributes != NULL) {
-      for (unsigned i = 0; i < prog->Attributes->NumParameters; i++) {
-	 ir_variable *const var =
-	    sh->symbols->get_variable(prog->Attributes->Parameters[i].Name);
-
-	 /* Note: attributes that occupy multiple slots, such as arrays or
-	  * matrices, may appear in the attrib array multiple times.
-	  */
-	 if ((var == NULL) || (var->location != -1))
-	    continue;
-
-	 /* From page 61 of the OpenGL 4.0 spec:
-	  *
-	  *     "LinkProgram will fail if the attribute bindings assigned by
-	  *     BindAttribLocation do not leave not enough space to assign a
-	  *     location for an active matrix attribute or an active attribute
-	  *     array, both of which require multiple contiguous generic
-	  *     attributes."
-	  *
-	  * Previous versions of the spec contain similar language but omit the
-	  * bit about attribute arrays.
-	  *
-	  * Page 61 of the OpenGL 4.0 spec also says:
-	  *
-	  *     "It is possible for an application to bind more than one
-	  *     attribute name to the same location. This is referred to as
-	  *     aliasing. This will only work if only one of the aliased
-	  *     attributes is active in the executable program, or if no path
-	  *     through the shader consumes more than one attribute of a set
-	  *     of attributes aliased to the same location. A link error can
-	  *     occur if the linker determines that every path through the
-	  *     shader consumes multiple aliased attributes, but
-	  *     implementations are not required to generate an error in this
-	  *     case."
-	  *
-	  * These two paragraphs are either somewhat contradictory, or I don't
-	  * fully understand one or both of them.
-	  */
-	 /* FINISHME: The code as currently written does not support attribute
-	  * FINISHME: location aliasing (see comment above).
-	  */
-	 const int attr = prog->Attributes->Parameters[i].StateIndexes[0];
-	 const unsigned slots = count_attribute_slots(var->type);
-
-	 /* Mask representing the contiguous slots that will be used by this
-	  * attribute.
-	  */
-	 const unsigned use_mask = (1 << slots) - 1;
-
-	 /* Generate a link error if the set of bits requested for this
-	  * attribute overlaps any previously allocated bits.
-	  */
-	 if ((~(use_mask << attr) & used_locations) != used_locations) {
-	    linker_error_printf(prog,
-				"insufficient contiguous attribute locations "
-				"available for vertex shader input `%s'",
-				var->name);
-	    return false;
-	 }
-
-	 var->location = VERT_ATTRIB_GENERIC0 + attr;
-	 used_locations |= (use_mask << attr);
-      }
-   }
-
-   /* Temporary storage for the set of attributes that need locations assigned.
-    */
-   struct temp_attr {
-      unsigned slots;
-      ir_variable *var;
-
-      /* Used below in the call to qsort. */
-      static int compare(const void *a, const void *b)
-      {
-	 const temp_attr *const l = (const temp_attr *) a;
-	 const temp_attr *const r = (const temp_attr *) b;
-
-	 /* Reversed because we want a descending order sort below. */
-	 return r->slots - l->slots;
-      }
-   } to_assign[16];
-
-   unsigned num_attr = 0;
-
-   foreach_list(node, sh->ir) {
-      ir_variable *const var = ((ir_instruction *) node)->as_variable();
-
-      if ((var == NULL) || (var->mode != ir_var_in))
-	 continue;
-
-      if (var->explicit_location) {
-	 const unsigned slots = count_attribute_slots(var->type);
-	 const unsigned use_mask = (1 << slots) - 1;
-	 const int attr = var->location - VERT_ATTRIB_GENERIC0;
-
-	 if ((var->location >= (int)(max_attribute_index + VERT_ATTRIB_GENERIC0))
-	     || (var->location < 0)) {
-	    linker_error_printf(prog,
-				"invalid explicit location %d specified for "
-				"`%s'\n",
-				(var->location < 0) ? var->location : attr,
-				var->name);
-	    return false;
-	 } else if (var->location >= VERT_ATTRIB_GENERIC0) {
-	    used_locations |= (use_mask << attr);
-	 }
-      }
-
-      /* The location was explicitly assigned, nothing to do here.
-       */
-      if (var->location != -1)
-	 continue;
-
-      to_assign[num_attr].slots = count_attribute_slots(var->type);
-      to_assign[num_attr].var = var;
-      num_attr++;
-   }
-
-   /* If all of the attributes were assigned locations by the application (or
-    * are built-in attributes with fixed locations), return early.  This should
-    * be the common case.
-    */
-   if (num_attr == 0)
-      return true;
-
-   qsort(to_assign, num_attr, sizeof(to_assign[0]), temp_attr::compare);
-
-   /* VERT_ATTRIB_GENERIC0 is a psdueo-alias for VERT_ATTRIB_POS.  It can only
-    * be explicitly assigned by via glBindAttribLocation.  Mark it as reserved
-    * to prevent it from being automatically allocated below.
-    */
-   find_deref_visitor find("gl_Vertex");
-   find.run(sh->ir);
-   if (find.variable_found())
-      used_locations |= (1 << 0);
-
-   for (unsigned i = 0; i < num_attr; i++) {
-      /* Mask representing the contiguous slots that will be used by this
-       * attribute.
-       */
-      const unsigned use_mask = (1 << to_assign[i].slots) - 1;
-
-      int location = find_available_slots(used_locations, to_assign[i].slots);
-
-      if (location < 0) {
-	 linker_error_printf(prog,
-			     "insufficient contiguous attribute locations "
-			     "available for vertex shader input `%s'",
-			     to_assign[i].var->name);
-	 return false;
-      }
-
-      to_assign[i].var->location = VERT_ATTRIB_GENERIC0 + location;
-      used_locations |= (use_mask << location);
-   }
-
-   return true;
-}
-
-
-/**
- * Demote shader inputs and outputs that are not used in other stages
- */
-void
-demote_shader_inputs_and_outputs(gl_shader *sh, enum ir_variable_mode mode)
-{
-   foreach_list(node, sh->ir) {
-      ir_variable *const var = ((ir_instruction *) node)->as_variable();
-
-      if ((var == NULL) || (var->mode != int(mode)))
-	 continue;
-
-      /* A shader 'in' or 'out' variable is only really an input or output if
-       * its value is used by other shader stages.  This will cause the variable
-       * to have a location assigned.
-       */
-      if (var->location == -1) {
-	 var->mode = ir_var_auto;
-      }
-   }
-}
-
-
-void
-assign_varying_locations(struct gl_shader_program *prog,
-			 gl_shader *producer, gl_shader *consumer)
-{
-   /* FINISHME: Set dynamically when geometry shader support is added. */
-   unsigned output_index = VERT_RESULT_VAR0;
-   unsigned input_index = FRAG_ATTRIB_VAR0;
-
-   /* Operate in a total of three passes.
-    *
-    * 1. Assign locations for any matching inputs and outputs.
-    *
-    * 2. Mark output variables in the producer that do not have locations as
-    *    not being outputs.  This lets the optimizer eliminate them.
-    *
-    * 3. Mark input variables in the consumer that do not have locations as
-    *    not being inputs.  This lets the optimizer eliminate them.
-    */
-
-   invalidate_variable_locations(producer, ir_var_out, VERT_RESULT_VAR0);
-   invalidate_variable_locations(consumer, ir_var_in, FRAG_ATTRIB_VAR0);
-
-   foreach_list(node, producer->ir) {
-      ir_variable *const output_var = ((ir_instruction *) node)->as_variable();
-
-      if ((output_var == NULL) || (output_var->mode != ir_var_out)
-	  || (output_var->location != -1))
-	 continue;
-
-      ir_variable *const input_var =
-	 consumer->symbols->get_variable(output_var->name);
-
-      if ((input_var == NULL) || (input_var->mode != ir_var_in))
-	 continue;
-
-      assert(input_var->location == -1);
-
-      output_var->location = output_index;
-      input_var->location = input_index;
-
-      /* FINISHME: Support for "varying" records in GLSL 1.50. */
-      assert(!output_var->type->is_record());
-
-      if (output_var->type->is_array()) {
-	 const unsigned slots = output_var->type->length
-	    * output_var->type->fields.array->matrix_columns;
-
-	 output_index += slots;
-	 input_index += slots;
-      } else {
-	 const unsigned slots = output_var->type->matrix_columns;
-
-	 output_index += slots;
-	 input_index += slots;
-      }
-   }
-
-   foreach_list(node, consumer->ir) {
-      ir_variable *const var = ((ir_instruction *) node)->as_variable();
-
-      if ((var == NULL) || (var->mode != ir_var_in))
-	 continue;
-
-      if (var->location == -1) {
-	 if (prog->Version <= 120) {
-	    /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec:
-	     *
-	     *     Only those varying variables used (i.e. read) in
-	     *     the fragment shader executable must be written to
-	     *     by the vertex shader executable; declaring
-	     *     superfluous varying variables in a vertex shader is
-	     *     permissible.
-	     *
-	     * We interpret this text as meaning that the VS must
-	     * write the variable for the FS to read it.  See
-	     * "glsl1-varying read but not written" in piglit.
-	     */
-
-	    linker_error_printf(prog, "fragment shader varying %s not written "
-				"by vertex shader\n.", var->name);
-	    prog->LinkStatus = false;
-	 }
-
-	 /* An 'in' variable is only really a shader input if its
-	  * value is written by the previous stage.
-	  */
-	 var->mode = ir_var_auto;
-      }
-   }
-}
-
-
-void
-link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)
-{
-   void *mem_ctx = ralloc_context(NULL); // temporary linker context
-
-   prog->LinkStatus = false;
-   prog->Validated = false;
-   prog->_Used = false;
-
-   if (prog->InfoLog != NULL)
-      ralloc_free(prog->InfoLog);
-
-   prog->InfoLog = ralloc_strdup(NULL, "");
-
-   /* Separate the shaders into groups based on their type.
-    */
-   struct gl_shader **vert_shader_list;
-   unsigned num_vert_shaders = 0;
-   struct gl_shader **frag_shader_list;
-   unsigned num_frag_shaders = 0;
-
-   vert_shader_list = (struct gl_shader **)
-      calloc(2 * prog->NumShaders, sizeof(struct gl_shader *));
-   frag_shader_list =  &vert_shader_list[prog->NumShaders];
-
-   unsigned min_version = UINT_MAX;
-   unsigned max_version = 0;
-   for (unsigned i = 0; i < prog->NumShaders; i++) {
-      min_version = MIN2(min_version, prog->Shaders[i]->Version);
-      max_version = MAX2(max_version, prog->Shaders[i]->Version);
-
-      switch (prog->Shaders[i]->Type) {
-      case GL_VERTEX_SHADER:
-	 vert_shader_list[num_vert_shaders] = prog->Shaders[i];
-	 num_vert_shaders++;
-	 break;
-      case GL_FRAGMENT_SHADER:
-	 frag_shader_list[num_frag_shaders] = prog->Shaders[i];
-	 num_frag_shaders++;
-	 break;
-      case GL_GEOMETRY_SHADER:
-	 /* FINISHME: Support geometry shaders. */
-	 assert(prog->Shaders[i]->Type != GL_GEOMETRY_SHADER);
-	 break;
-      }
-   }
-
-   /* Previous to GLSL version 1.30, different compilation units could mix and
-    * match shading language versions.  With GLSL 1.30 and later, the versions
-    * of all shaders must match.
-    */
-   assert(min_version >= 100);
-   assert(max_version <= 130);
-   if ((max_version >= 130 || min_version == 100)
-       && min_version != max_version) {
-      linker_error_printf(prog, "all shaders must use same shading "
-			  "language version\n");
-      goto done;
-   }
-
-   prog->Version = max_version;
-
-   for (unsigned int i = 0; i < MESA_SHADER_TYPES; i++) {
-      if (prog->_LinkedShaders[i] != NULL)
-	 ctx->Driver.DeleteShader(ctx, prog->_LinkedShaders[i]);
-
-      prog->_LinkedShaders[i] = NULL;
-   }
-
-   /* Link all shaders for a particular stage and validate the result.
-    */
-   if (num_vert_shaders > 0) {
-      gl_shader *const sh =
-	 link_intrastage_shaders(mem_ctx, ctx, prog, vert_shader_list,
-				 num_vert_shaders);
-
-      if (sh == NULL)
-	 goto done;
-
-      if (!validate_vertex_shader_executable(prog, sh))
-	 goto done;
-
-      _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_VERTEX],
-			     sh);
-   }
-
-   if (num_frag_shaders > 0) {
-      gl_shader *const sh =
-	 link_intrastage_shaders(mem_ctx, ctx, prog, frag_shader_list,
-				 num_frag_shaders);
-
-      if (sh == NULL)
-	 goto done;
-
-      if (!validate_fragment_shader_executable(prog, sh))
-	 goto done;
-
-      _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_FRAGMENT],
-			     sh);
-   }
-
-   /* Here begins the inter-stage linking phase.  Some initial validation is
-    * performed, then locations are assigned for uniforms, attributes, and
-    * varyings.
-    */
-   if (cross_validate_uniforms(prog)) {
-      unsigned prev;
-
-      for (prev = 0; prev < MESA_SHADER_TYPES; prev++) {
-	 if (prog->_LinkedShaders[prev] != NULL)
-	    break;
-      }
-
-      /* Validate the inputs of each stage with the output of the preceeding
-       * stage.
-       */
-      for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) {
-	 if (prog->_LinkedShaders[i] == NULL)
-	    continue;
-
-	 if (!cross_validate_outputs_to_inputs(prog,
-					       prog->_LinkedShaders[prev],
-					       prog->_LinkedShaders[i]))
-	    goto done;
-
-	 prev = i;
-      }
-
-      prog->LinkStatus = true;
-   }
-
-   /* Do common optimization before assigning storage for attributes,
-    * uniforms, and varyings.  Later optimization could possibly make
-    * some of that unused.
-    */
-   for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {
-      if (prog->_LinkedShaders[i] == NULL)
-	 continue;
-
-      while (do_common_optimization(prog->_LinkedShaders[i]->ir, true, 32))
-	 ;
-   }
-
-   update_array_sizes(prog);
-
-   assign_uniform_locations(prog);
-
-   if (prog->_LinkedShaders[MESA_SHADER_VERTEX] != NULL) {
-      /* FINISHME: The value of the max_attribute_index parameter is
-       * FINISHME: implementation dependent based on the value of
-       * FINISHME: GL_MAX_VERTEX_ATTRIBS.  GL_MAX_VERTEX_ATTRIBS must be
-       * FINISHME: at least 16, so hardcode 16 for now.
-       */
-      if (!assign_attribute_locations(prog, 16)) {
-	 prog->LinkStatus = false;
-	 goto done;
-      }
-   }
-
-   unsigned prev;
-   for (prev = 0; prev < MESA_SHADER_TYPES; prev++) {
-      if (prog->_LinkedShaders[prev] != NULL)
-	 break;
-   }
-
-   for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) {
-      if (prog->_LinkedShaders[i] == NULL)
-	 continue;
-
-      assign_varying_locations(prog,
-			       prog->_LinkedShaders[prev],
-			       prog->_LinkedShaders[i]);
-      prev = i;
-   }
-
-   if (prog->_LinkedShaders[MESA_SHADER_VERTEX] != NULL) {
-      demote_shader_inputs_and_outputs(prog->_LinkedShaders[MESA_SHADER_VERTEX],
-				       ir_var_out);
-   }
-
-   if (prog->_LinkedShaders[MESA_SHADER_GEOMETRY] != NULL) {
-      gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_GEOMETRY];
-
-      demote_shader_inputs_and_outputs(sh, ir_var_in);
-      demote_shader_inputs_and_outputs(sh, ir_var_inout);
-      demote_shader_inputs_and_outputs(sh, ir_var_out);
-   }
-
-   if (prog->_LinkedShaders[MESA_SHADER_FRAGMENT] != NULL) {
-      gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
-
-      demote_shader_inputs_and_outputs(sh, ir_var_in);
-   }
-
-   /* OpenGL ES requires that a vertex shader and a fragment shader both be
-    * present in a linked program.  By checking for use of shading language
-    * version 1.00, we also catch the GL_ARB_ES2_compatibility case.
-    */
-   if (ctx->API == API_OPENGLES2 || prog->Version == 100) {
-      if (prog->_LinkedShaders[MESA_SHADER_VERTEX] == NULL) {
-	 linker_error_printf(prog, "program lacks a vertex shader\n");
-	 prog->LinkStatus = false;
-      } else if (prog->_LinkedShaders[MESA_SHADER_FRAGMENT] == NULL) {
-	 linker_error_printf(prog, "program lacks a fragment shader\n");
-	 prog->LinkStatus = false;
-      }
-   }
-
-   /* FINISHME: Assign fragment shader output locations. */
-
-done:
-   free(vert_shader_list);
-
-   for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {
-      if (prog->_LinkedShaders[i] == NULL)
-	 continue;
-
-      /* Retain any live IR, but trash the rest. */
-      reparent_ir(prog->_LinkedShaders[i]->ir, prog->_LinkedShaders[i]->ir);
-   }
-
-   ralloc_free(mem_ctx);
-}
+/*

+ * Copyright © 2010 Intel Corporation

+ *

+ * Permission is hereby granted, free of charge, to any person obtaining a

+ * copy of this software and associated documentation files (the "Software"),

+ * to deal in the Software without restriction, including without limitation

+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,

+ * and/or sell copies of the Software, and to permit persons to whom the

+ * Software is furnished to do so, subject to the following conditions:

+ *

+ * The above copyright notice and this permission notice (including the next

+ * paragraph) shall be included in all copies or substantial portions of the

+ * Software.

+ *

+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR

+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,

+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL

+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER

+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING

+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER

+ * DEALINGS IN THE SOFTWARE.

+ */

+

+/**

+ * \file linker.cpp

+ * GLSL linker implementation

+ *

+ * Given a set of shaders that are to be linked to generate a final program,

+ * there are three distinct stages.

+ *

+ * In the first stage shaders are partitioned into groups based on the shader

+ * type.  All shaders of a particular type (e.g., vertex shaders) are linked

+ * together.

+ *

+ *   - Undefined references in each shader are resolve to definitions in

+ *     another shader.

+ *   - Types and qualifiers of uniforms, outputs, and global variables defined

+ *     in multiple shaders with the same name are verified to be the same.

+ *   - Initializers for uniforms and global variables defined

+ *     in multiple shaders with the same name are verified to be the same.

+ *

+ * The result, in the terminology of the GLSL spec, is a set of shader

+ * executables for each processing unit.

+ *

+ * After the first stage is complete, a series of semantic checks are performed

+ * on each of the shader executables.

+ *

+ *   - Each shader executable must define a \c main function.

+ *   - Each vertex shader executable must write to \c gl_Position.

+ *   - Each fragment shader executable must write to either \c gl_FragData or

+ *     \c gl_FragColor.

+ *

+ * In the final stage individual shader executables are linked to create a

+ * complete exectuable.

+ *

+ *   - Types of uniforms defined in multiple shader stages with the same name

+ *     are verified to be the same.

+ *   - Initializers for uniforms defined in multiple shader stages with the

+ *     same name are verified to be the same.

+ *   - Types and qualifiers of outputs defined in one stage are verified to

+ *     be the same as the types and qualifiers of inputs defined with the same

+ *     name in a later stage.

+ *

+ * \author Ian Romanick <ian.d.romanick@intel.com>

+ */

+#include <cstdlib>

+#include <cstdio>

+#include <cstdarg>

+#include <climits>

+

+#include "main/core.h"

+#include "glsl_symbol_table.h"

+#include "ir.h"

+#include "program.h"

+#include "program/hash_table.h"

+#include "linker.h"

+#include "ir_optimization.h"

+

+extern "C" {

+#include "main/shaderobj.h"

+}

+

+/**

+ * Visitor that determines whether or not a variable is ever written.

+ */

+class find_assignment_visitor : public ir_hierarchical_visitor {

+public:

+   find_assignment_visitor(const char *name)

+      : name(name), found(false)

+   {

+      /* empty */

+   }

+

+   virtual ir_visitor_status visit_enter(ir_assignment *ir)

+   {

+      ir_variable *const var = ir->lhs->variable_referenced();

+

+      if (strcmp(name, var->name) == 0) {

+	 found = true;

+	 return visit_stop;

+      }

+

+      return visit_continue_with_parent;

+   }

+

+   virtual ir_visitor_status visit_enter(ir_call *ir)

+   {

+      exec_list_iterator sig_iter = ir->get_callee()->parameters.iterator();

+      foreach_iter(exec_list_iterator, iter, *ir) {

+	 ir_rvalue *param_rval = (ir_rvalue *)iter.get();

+	 ir_variable *sig_param = (ir_variable *)sig_iter.get();

+

+	 if (sig_param->mode == ir_var_out ||

+	     sig_param->mode == ir_var_inout) {

+	    ir_variable *var = param_rval->variable_referenced();

+	    if (var && strcmp(name, var->name) == 0) {

+	       found = true;

+	       return visit_stop;

+	    }

+	 }

+	 sig_iter.next();

+      }

+

+      return visit_continue_with_parent;

+   }

+

+   bool variable_found()

+   {

+      return found;

+   }

+

+private:

+   const char *name;       /**< Find writes to a variable with this name. */

+   bool found;             /**< Was a write to the variable found? */

+};

+

+

+/**

+ * Visitor that determines whether or not a variable is ever read.

+ */

+class find_deref_visitor : public ir_hierarchical_visitor {

+public:

+   find_deref_visitor(const char *name)

+      : name(name), found(false)

+   {

+      /* empty */

+   }

+

+   virtual ir_visitor_status visit(ir_dereference_variable *ir)

+   {

+      if (strcmp(this->name, ir->var->name) == 0) {

+	 this->found = true;

+	 return visit_stop;

+      }

+

+      return visit_continue;

+   }

+

+   bool variable_found() const

+   {

+      return this->found;

+   }

+

+private:

+   const char *name;       /**< Find writes to a variable with this name. */

+   bool found;             /**< Was a write to the variable found? */

+};

+

+

+void

+linker_error_printf(gl_shader_program *prog, const char *fmt, ...)

+{

+   va_list ap;

+

+   ralloc_strcat(&prog->InfoLog, "error: ");

+   va_start(ap, fmt);

+   ralloc_vasprintf_append(&prog->InfoLog, fmt, ap);

+   va_end(ap);

+}

+

+

+void

+invalidate_variable_locations(gl_shader *sh, enum ir_variable_mode mode,

+			      int generic_base)

+{

+   foreach_list(node, sh->ir) {

+      ir_variable *const var = ((ir_instruction *) node)->as_variable();

+

+      if ((var == NULL) || (var->mode != (unsigned) mode))

+	 continue;

+

+      /* Only assign locations for generic attributes / varyings / etc.

+       */

+      if ((var->location >= generic_base) && !var->explicit_location)

+	  var->location = -1;

+   }

+}

+

+

+/**

+ * Determine the number of attribute slots required for a particular type

+ *

+ * This code is here because it implements the language rules of a specific

+ * GLSL version.  Since it's a property of the language and not a property of

+ * types in general, it doesn't really belong in glsl_type.

+ */

+unsigned

+count_attribute_slots(const glsl_type *t)

+{

+   /* From page 31 (page 37 of the PDF) of the GLSL 1.50 spec:

+    *

+    *     "A scalar input counts the same amount against this limit as a vec4,

+    *     so applications may want to consider packing groups of four

+    *     unrelated float inputs together into a vector to better utilize the

+    *     capabilities of the underlying hardware. A matrix input will use up

+    *     multiple locations.  The number of locations used will equal the

+    *     number of columns in the matrix."

+    *

+    * The spec does not explicitly say how arrays are counted.  However, it

+    * should be safe to assume the total number of slots consumed by an array

+    * is the number of entries in the array multiplied by the number of slots

+    * consumed by a single element of the array.

+    */

+

+   if (t->is_array())

+      return t->array_size() * count_attribute_slots(t->element_type());

+

+   if (t->is_matrix())

+      return t->matrix_columns;

+

+   return 1;

+}

+

+

+/**

+ * Verify that a vertex shader executable meets all semantic requirements

+ *

+ * \param shader  Vertex shader executable to be verified

+ */

+bool

+validate_vertex_shader_executable(struct gl_shader_program *prog,

+				  struct gl_shader *shader)

+{

+   if (shader == NULL)

+      return true;

+

+   find_assignment_visitor find("gl_Position");

+   find.run(shader->ir);

+   if (!find.variable_found()) {

+      linker_error_printf(prog,

+			  "vertex shader does not write to `gl_Position'\n");

+      return false;

+   }

+

+   return true;

+}

+

+

+/**

+ * Verify that a fragment shader executable meets all semantic requirements

+ *

+ * \param shader  Fragment shader executable to be verified

+ */

+bool

+validate_fragment_shader_executable(struct gl_shader_program *prog,

+				    struct gl_shader *shader)

+{

+   if (shader == NULL)

+      return true;

+

+   find_assignment_visitor frag_color("gl_FragColor");

+   find_assignment_visitor frag_data("gl_FragData");

+

+   frag_color.run(shader->ir);

+   frag_data.run(shader->ir);

+

+   if (frag_color.variable_found() && frag_data.variable_found()) {

+      linker_error_printf(prog,  "fragment shader writes to both "

+			  "`gl_FragColor' and `gl_FragData'\n");

+      return false;

+   }

+

+   return true;

+}

+

+

+/**

+ * Generate a string describing the mode of a variable

+ */

+static const char *

+mode_string(const ir_variable *var)

+{

+   switch (var->mode) {

+   case ir_var_auto:

+      return (var->read_only) ? "global constant" : "global variable";

+

+   case ir_var_uniform: return "uniform";

+   case ir_var_in:      return "shader input";

+   case ir_var_out:     return "shader output";

+   case ir_var_inout:   return "shader inout";

+

+   case ir_var_const_in:

+   case ir_var_temporary:

+   default:

+      assert(!"Should not get here.");

+      return "invalid variable";

+   }

+}

+

+

+/**

+ * Perform validation of global variables used across multiple shaders

+ */

+bool

+cross_validate_globals(struct gl_shader_program *prog,

+		       struct gl_shader **shader_list,

+		       unsigned num_shaders,

+		       bool uniforms_only)

+{

+   /* Examine all of the uniforms in all of the shaders and cross validate

+    * them.

+    */

+   glsl_symbol_table variables;

+   for (unsigned i = 0; i < num_shaders; i++) {

+      if (shader_list[i] == NULL)

+	 continue;

+

+      foreach_list(node, shader_list[i]->ir) {

+	 ir_variable *const var = ((ir_instruction *) node)->as_variable();

+

+	 if (var == NULL)

+	    continue;

+

+	 if (uniforms_only && (var->mode != ir_var_uniform))

+	    continue;

+

+	 /* Don't cross validate temporaries that are at global scope.  These

+	  * will eventually get pulled into the shaders 'main'.

+	  */

+	 if (var->mode == ir_var_temporary)

+	    continue;

+

+	 /* If a global with this name has already been seen, verify that the

+	  * new instance has the same type.  In addition, if the globals have

+	  * initializers, the values of the initializers must be the same.

+	  */

+	 ir_variable *const existing = variables.get_variable(var->name);

+	 if (existing != NULL) {

+	    if (var->type != existing->type) {

+	       /* Consider the types to be "the same" if both types are arrays

+		* of the same type and one of the arrays is implicitly sized.

+		* In addition, set the type of the linked variable to the

+		* explicitly sized array.

+		*/

+	       if (var->type->is_array()

+		   && existing->type->is_array()

+		   && (var->type->fields.array == existing->type->fields.array)

+		   && ((var->type->length == 0)

+		       || (existing->type->length == 0))) {

+		  if (var->type->length != 0) {

+		     existing->type = var->type;

+		  }

+	       } else {

+		  linker_error_printf(prog, "%s `%s' declared as type "

+				      "`%s' and type `%s'\n",

+				      mode_string(var),

+				      var->name, var->type->name,

+				      existing->type->name);

+		  return false;

+	       }

+	    }

+

+	    if (var->explicit_location) {

+	       if (existing->explicit_location

+		   && (var->location != existing->location)) {

+		     linker_error_printf(prog, "explicit locations for %s "

+					 "`%s' have differing values\n",

+					 mode_string(var), var->name);

+		     return false;

+	       }

+

+	       existing->location = var->location;

+	       existing->explicit_location = true;

+	    }

+

+        /* Validate layout qualifiers for gl_FragDepth.

+         *

+         * From the AMD_conservative_depth spec:

+         *    "If gl_FragDepth is redeclared in any fragment shader in

+         *    a program, it must be redeclared in all fragment shaders in that

+         *    program that have static assignments to gl_FragDepth. All

+         *    redeclarations of gl_FragDepth in all fragment shaders in

+         *    a single program must have the same set of qualifiers."

+         */

+        if (strcmp(var->name, "gl_FragDepth") == 0) {

+           bool layout_declared = var->depth_layout != ir_depth_layout_none;

+           bool layout_differs = var->depth_layout != existing->depth_layout;

+           if (layout_declared && layout_differs) {

+              linker_error_printf(prog,

+                 "All redeclarations of gl_FragDepth in all fragment shaders "

+                 "in a single program must have the same set of qualifiers.");

+           }

+           if (var->used && layout_differs) {

+              linker_error_printf(prog,

+                    "If gl_FragDepth is redeclared with a layout qualifier in"

+                    "any fragment shader, it must be redeclared with the same"

+                    "layout qualifier in all fragment shaders that have"

+                    "assignments to gl_FragDepth");

+           }

+        }

+

+	    /* FINISHME: Handle non-constant initializers.

+	     */

+	    if (var->constant_value != NULL) {

+	       if (existing->constant_value != NULL) {

+		  if (!var->constant_value->has_value(existing->constant_value)) {

+		     linker_error_printf(prog, "initializers for %s "

+					 "`%s' have differing values\n",

+					 mode_string(var), var->name);

+		     return false;

+		  }

+	       } else

+		  /* If the first-seen instance of a particular uniform did not

+		   * have an initializer but a later instance does, copy the

+		   * initializer to the version stored in the symbol table.

+		   */

+		  /* FINISHME: This is wrong.  The constant_value field should

+		   * FINISHME: not be modified!  Imagine a case where a shader

+		   * FINISHME: without an initializer is linked in two different

+		   * FINISHME: programs with shaders that have differing

+		   * FINISHME: initializers.  Linking with the first will

+		   * FINISHME: modify the shader, and linking with the second

+		   * FINISHME: will fail.

+		   */

+		  existing->constant_value =

+		     var->constant_value->clone(ralloc_parent(existing), NULL);

+	    }

+

+	    if (existing->invariant != var->invariant) {

+	       linker_error_printf(prog, "declarations for %s `%s' have "

+	                           "mismatching invariant qualifiers\n",

+	                           mode_string(var), var->name);

+	       return false;

+	    }

+            if (existing->centroid != var->centroid) {

+               linker_error_printf(prog, "declarations for %s `%s' have "

+                                   "mismatching centroid qualifiers\n",

+                                   mode_string(var), var->name);

+               return false;

+            }

+	 } else

+	    variables.add_variable(var);

+      }

+   }

+

+   return true;

+}

+

+

+/**

+ * Perform validation of uniforms used across multiple shader stages

+ */

+bool

+cross_validate_uniforms(struct gl_shader_program *prog)

+{

+   return cross_validate_globals(prog, prog->_LinkedShaders,

+				 MESA_SHADER_TYPES, true);

+}

+

+

+/**

+ * Validate that outputs from one stage match inputs of another

+ */

+bool

+cross_validate_outputs_to_inputs(struct gl_shader_program *prog,

+				 gl_shader *producer, gl_shader *consumer)

+{

+   glsl_symbol_table parameters;

+   /* FINISHME: Figure these out dynamically. */

+   const char *const producer_stage = "vertex";

+   const char *const consumer_stage = "fragment";

+

+   /* Find all shader outputs in the "producer" stage.

+    */

+   foreach_list(node, producer->ir) {

+      ir_variable *const var = ((ir_instruction *) node)->as_variable();

+

+      /* FINISHME: For geometry shaders, this should also look for inout

+       * FINISHME: variables.

+       */

+      if ((var == NULL) || (var->mode != ir_var_out))

+	 continue;

+

+      parameters.add_variable(var);

+   }

+

+

+   /* Find all shader inputs in the "consumer" stage.  Any variables that have

+    * matching outputs already in the symbol table must have the same type and

+    * qualifiers.

+    */

+   foreach_list(node, consumer->ir) {

+      ir_variable *const input = ((ir_instruction *) node)->as_variable();

+

+      /* FINISHME: For geometry shaders, this should also look for inout

+       * FINISHME: variables.

+       */

+      if ((input == NULL) || (input->mode != ir_var_in))

+	 continue;

+

+      ir_variable *const output = parameters.get_variable(input->name);

+      if (output != NULL) {

+	 /* Check that the types match between stages.

+	  */

+	 if (input->type != output->type) {

+	    /* There is a bit of a special case for gl_TexCoord.  This

+	     * built-in is unsized by default.  Appliations that variable

+	     * access it must redeclare it with a size.  There is some

+	     * language in the GLSL spec that implies the fragment shader

+	     * and vertex shader do not have to agree on this size.  Other

+	     * driver behave this way, and one or two applications seem to

+	     * rely on it.

+	     *

+	     * Neither declaration needs to be modified here because the array

+	     * sizes are fixed later when update_array_sizes is called.

+	     *

+	     * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec:

+	     *

+	     *     "Unlike user-defined varying variables, the built-in

+	     *     varying variables don't have a strict one-to-one

+	     *     correspondence between the vertex language and the

+	     *     fragment language."

+	     */

+	    if (!output->type->is_array()

+		|| (strncmp("gl_", output->name, 3) != 0)) {

+	       linker_error_printf(prog,

+				   "%s shader output `%s' declared as "

+				   "type `%s', but %s shader input declared "

+				   "as type `%s'\n",

+				   producer_stage, output->name,

+				   output->type->name,

+				   consumer_stage, input->type->name);

+	       return false;

+	    }

+	 }

+

+	 /* Check that all of the qualifiers match between stages.

+	  */

+	 if (input->centroid != output->centroid) {

+	    linker_error_printf(prog,

+				"%s shader output `%s' %s centroid qualifier, "

+				"but %s shader input %s centroid qualifier\n",

+				producer_stage,

+				output->name,

+				(output->centroid) ? "has" : "lacks",

+				consumer_stage,

+				(input->centroid) ? "has" : "lacks");

+	    return false;

+	 }

+

+	 if (input->invariant != output->invariant) {

+	    linker_error_printf(prog,

+				"%s shader output `%s' %s invariant qualifier, "

+				"but %s shader input %s invariant qualifier\n",

+				producer_stage,

+				output->name,

+				(output->invariant) ? "has" : "lacks",

+				consumer_stage,

+				(input->invariant) ? "has" : "lacks");

+	    return false;

+	 }

+

+	 if (input->interpolation != output->interpolation) {

+	    linker_error_printf(prog,

+				"%s shader output `%s' specifies %s "

+				"interpolation qualifier, "

+				"but %s shader input specifies %s "

+				"interpolation qualifier\n",

+				producer_stage,

+				output->name,

+				output->interpolation_string(),

+				consumer_stage,

+				input->interpolation_string());

+	    return false;

+	 }

+      }

+   }

+

+   return true;

+}

+

+

+/**

+ * Populates a shaders symbol table with all global declarations

+ */

+static void

+populate_symbol_table(gl_shader *sh)

+{

+   sh->symbols = new(sh) glsl_symbol_table;

+

+   foreach_list(node, sh->ir) {

+      ir_instruction *const inst = (ir_instruction *) node;

+      ir_variable *var;

+      ir_function *func;

+

+      if ((func = inst->as_function()) != NULL) {

+	 sh->symbols->add_function(func);

+      } else if ((var = inst->as_variable()) != NULL) {

+	 sh->symbols->add_variable(var);

+      }

+   }

+}

+

+

+/**

+ * Remap variables referenced in an instruction tree

+ *

+ * This is used when instruction trees are cloned from one shader and placed in

+ * another.  These trees will contain references to \c ir_variable nodes that

+ * do not exist in the target shader.  This function finds these \c ir_variable

+ * references and replaces the references with matching variables in the target

+ * shader.

+ *

+ * If there is no matching variable in the target shader, a clone of the

+ * \c ir_variable is made and added to the target shader.  The new variable is

+ * added to \b both the instruction stream and the symbol table.

+ *

+ * \param inst         IR tree that is to be processed.

+ * \param symbols      Symbol table containing global scope symbols in the

+ *                     linked shader.

+ * \param instructions Instruction stream where new variable declarations

+ *                     should be added.

+ */

+void

+remap_variables(ir_instruction *inst, struct gl_shader *target,

+		hash_table *temps)

+{

+   class remap_visitor : public ir_hierarchical_visitor {

+   public:

+	 remap_visitor(struct gl_shader *target,

+		    hash_table *temps)

+      {

+	 this->target = target;

+	 this->symbols = target->symbols;

+	 this->instructions = target->ir;

+	 this->temps = temps;

+      }

+

+      virtual ir_visitor_status visit(ir_dereference_variable *ir)

+      {

+	 if (ir->var->mode == ir_var_temporary) {

+	    ir_variable *var = (ir_variable *) hash_table_find(temps, ir->var);

+

+	    assert(var != NULL);

+	    ir->var = var;

+	    return visit_continue;

+	 }

+

+	 ir_variable *const existing =

+	    this->symbols->get_variable(ir->var->name);

+	 if (existing != NULL)

+	    ir->var = existing;

+	 else {

+	    ir_variable *copy = ir->var->clone(this->target, NULL);

+

+	    this->symbols->add_variable(copy);

+	    this->instructions->push_head(copy);

+	    ir->var = copy;

+	 }

+

+	 return visit_continue;

+      }

+

+   private:

+      struct gl_shader *target;

+      glsl_symbol_table *symbols;

+      exec_list *instructions;

+      hash_table *temps;

+   };

+

+   remap_visitor v(target, temps);

+

+   inst->accept(&v);

+}

+

+

+/**

+ * Move non-declarations from one instruction stream to another

+ *

+ * The intended usage pattern of this function is to pass the pointer to the

+ * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node

+ * pointer) for \c last and \c false for \c make_copies on the first

+ * call.  Successive calls pass the return value of the previous call for

+ * \c last and \c true for \c make_copies.

+ *

+ * \param instructions Source instruction stream

+ * \param last         Instruction after which new instructions should be

+ *                     inserted in the target instruction stream

+ * \param make_copies  Flag selecting whether instructions in \c instructions

+ *                     should be copied (via \c ir_instruction::clone) into the

+ *                     target list or moved.

+ *

+ * \return

+ * The new "last" instruction in the target instruction stream.  This pointer

+ * is suitable for use as the \c last parameter of a later call to this

+ * function.

+ */

+exec_node *

+move_non_declarations(exec_list *instructions, exec_node *last,

+		      bool make_copies, gl_shader *target)

+{

+   hash_table *temps = NULL;

+

+   if (make_copies)

+      temps = hash_table_ctor(0, hash_table_pointer_hash,

+			      hash_table_pointer_compare);

+

+   foreach_list_safe(node, instructions) {

+      ir_instruction *inst = (ir_instruction *) node;

+

+      if (inst->as_function())

+	 continue;

+

+      ir_variable *var = inst->as_variable();

+      if ((var != NULL) && (var->mode != ir_var_temporary))

+	 continue;

+

+      assert(inst->as_assignment()

+	     || ((var != NULL) && (var->mode == ir_var_temporary)));

+

+      if (make_copies) {

+	 inst = inst->clone(target, NULL);

+

+	 if (var != NULL)

+	    hash_table_insert(temps, inst, var);

+	 else

+	    remap_variables(inst, target, temps);

+      } else {

+	 inst->remove();

+      }

+

+      last->insert_after(inst);

+      last = inst;

+   }

+

+   if (make_copies)

+      hash_table_dtor(temps);

+

+   return last;

+}

+

+/**

+ * Get the function signature for main from a shader

+ */

+static ir_function_signature *

+get_main_function_signature(gl_shader *sh)

+{

+   ir_function *const f = sh->symbols->get_function("main");

+   if (f != NULL) {

+      exec_list void_parameters;

+

+      /* Look for the 'void main()' signature and ensure that it's defined.

+       * This keeps the linker from accidentally pick a shader that just

+       * contains a prototype for main.

+       *

+       * We don't have to check for multiple definitions of main (in multiple

+       * shaders) because that would have already been caught above.

+       */

+      ir_function_signature *sig = f->matching_signature(&void_parameters);

+      if ((sig != NULL) && sig->is_defined) {

+	 return sig;

+      }

+   }

+

+   return NULL;

+}

+

+

+/**

+ * Combine a group of shaders for a single stage to generate a linked shader

+ *

+ * \note

+ * If this function is supplied a single shader, it is cloned, and the new

+ * shader is returned.

+ */

+static struct gl_shader *

+link_intrastage_shaders(void *mem_ctx,

+			struct gl_context *ctx,

+			struct gl_shader_program *prog,

+			struct gl_shader **shader_list,

+			unsigned num_shaders)

+{

+   /* Check that global variables defined in multiple shaders are consistent.

+    */

+   if (!cross_validate_globals(prog, shader_list, num_shaders, false))

+      return NULL;

+

+   /* Check that there is only a single definition of each function signature

+    * across all shaders.

+    */

+   for (unsigned i = 0; i < (num_shaders - 1); i++) {

+      foreach_list(node, shader_list[i]->ir) {

+	 ir_function *const f = ((ir_instruction *) node)->as_function();

+

+	 if (f == NULL)

+	    continue;

+

+	 for (unsigned j = i + 1; j < num_shaders; j++) {

+	    ir_function *const other =

+	       shader_list[j]->symbols->get_function(f->name);

+

+	    /* If the other shader has no function (and therefore no function

+	     * signatures) with the same name, skip to the next shader.

+	     */

+	    if (other == NULL)

+	       continue;

+

+	    foreach_iter (exec_list_iterator, iter, *f) {

+	       ir_function_signature *sig =

+		  (ir_function_signature *) iter.get();

+

+	       if (!sig->is_defined || sig->is_builtin)

+		  continue;

+

+	       ir_function_signature *other_sig =

+		  other->exact_matching_signature(& sig->parameters);

+

+	       if ((other_sig != NULL) && other_sig->is_defined

+		   && !other_sig->is_builtin) {

+		  linker_error_printf(prog,

+				      "function `%s' is multiply defined",

+				      f->name);

+		  return NULL;

+	       }

+	    }

+	 }

+      }

+   }

+

+   /* Find the shader that defines main, and make a clone of it.

+    *

+    * Starting with the clone, search for undefined references.  If one is

+    * found, find the shader that defines it.  Clone the reference and add

+    * it to the shader.  Repeat until there are no undefined references or

+    * until a reference cannot be resolved.

+    */

+   gl_shader *main = NULL;

+   for (unsigned i = 0; i < num_shaders; i++) {

+      if (get_main_function_signature(shader_list[i]) != NULL) {

+	 main = shader_list[i];

+	 break;

+      }

+   }

+

+   if (main == NULL) {

+      linker_error_printf(prog, "%s shader lacks `main'\n",

+			  (shader_list[0]->Type == GL_VERTEX_SHADER)

+			  ? "vertex" : "fragment");

+      return NULL;

+   }

+

+   gl_shader *linked = ctx->Driver.NewShader(NULL, 0, main->Type);

+   linked->ir = new(linked) exec_list;

+   clone_ir_list(mem_ctx, linked->ir, main->ir);

+

+   populate_symbol_table(linked);

+

+   /* The a pointer to the main function in the final linked shader (i.e., the

+    * copy of the original shader that contained the main function).

+    */

+   ir_function_signature *const main_sig = get_main_function_signature(linked);

+

+   /* Move any instructions other than variable declarations or function

+    * declarations into main.

+    */

+   exec_node *insertion_point =

+      move_non_declarations(linked->ir, (exec_node *) &main_sig->body, false,

+			    linked);

+

+   for (unsigned i = 0; i < num_shaders; i++) {

+      if (shader_list[i] == main)

+	 continue;

+

+      insertion_point = move_non_declarations(shader_list[i]->ir,

+					      insertion_point, true, linked);

+   }

+

+   /* Resolve initializers for global variables in the linked shader.

+    */

+   unsigned num_linking_shaders = num_shaders;

+   for (unsigned i = 0; i < num_shaders; i++)

+      num_linking_shaders += shader_list[i]->num_builtins_to_link;

+

+   gl_shader **linking_shaders =

+      (gl_shader **) calloc(num_linking_shaders, sizeof(gl_shader *));

+

+   memcpy(linking_shaders, shader_list,

+	  sizeof(linking_shaders[0]) * num_shaders);

+

+   unsigned idx = num_shaders;

+   for (unsigned i = 0; i < num_shaders; i++) {

+      memcpy(&linking_shaders[idx], shader_list[i]->builtins_to_link,

+	     sizeof(linking_shaders[0]) * shader_list[i]->num_builtins_to_link);

+      idx += shader_list[i]->num_builtins_to_link;

+   }

+

+   assert(idx == num_linking_shaders);

+

+   if (!link_function_calls(prog, linked, linking_shaders,

+			    num_linking_shaders)) {

+      ctx->Driver.DeleteShader(ctx, linked);

+      linked = NULL;

+   }

+

+   free(linking_shaders);

+

+   /* Make a pass over all variable declarations to ensure that arrays with

+    * unspecified sizes have a size specified.  The size is inferred from the

+    * max_array_access field.

+    */

+   if (linked != NULL) {

+      class array_sizing_visitor : public ir_hierarchical_visitor {

+      public:

+	 virtual ir_visitor_status visit(ir_variable *var)

+	 {

+	    if (var->type->is_array() && (var->type->length == 0)) {

+	       const glsl_type *type =

+		  glsl_type::get_array_instance(var->type->fields.array,

+						var->max_array_access + 1);

+

+	       assert(type != NULL);

+	       var->type = type;

+	    }

+

+	    return visit_continue;

+	 }

+      } v;

+

+      v.run(linked->ir);

+   }

+

+   return linked;

+}

+

+

+struct uniform_node {

+   exec_node link;

+   struct gl_uniform *u;

+   unsigned slots;

+};

+

+/**

+ * Update the sizes of linked shader uniform arrays to the maximum

+ * array index used.

+ *

+ * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:

+ *

+ *     If one or more elements of an array are active,

+ *     GetActiveUniform will return the name of the array in name,

+ *     subject to the restrictions listed above. The type of the array

+ *     is returned in type. The size parameter contains the highest

+ *     array element index used, plus one. The compiler or linker

+ *     determines the highest index used.  There will be only one

+ *     active uniform reported by the GL per uniform array.

+

+ */

+static void

+update_array_sizes(struct gl_shader_program *prog)

+{

+   for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {

+	 if (prog->_LinkedShaders[i] == NULL)

+	    continue;

+

+      foreach_list(node, prog->_LinkedShaders[i]->ir) {

+	 ir_variable *const var = ((ir_instruction *) node)->as_variable();

+

+	 if ((var == NULL) || (var->mode != ir_var_uniform &&

+			       var->mode != ir_var_in &&

+			       var->mode != ir_var_out) ||

+	     !var->type->is_array())

+	    continue;

+

+	 unsigned int size = var->max_array_access;

+	 for (unsigned j = 0; j < MESA_SHADER_TYPES; j++) {

+	       if (prog->_LinkedShaders[j] == NULL)

+		  continue;

+

+	    foreach_list(node2, prog->_LinkedShaders[j]->ir) {

+	       ir_variable *other_var = ((ir_instruction *) node2)->as_variable();

+	       if (!other_var)

+		  continue;

+

+	       if (strcmp(var->name, other_var->name) == 0 &&

+		   other_var->max_array_access > size) {

+		  size = other_var->max_array_access;

+	       }

+	    }

+	 }

+

+	 if (size + 1 != var->type->fields.array->length) {

+	    var->type = glsl_type::get_array_instance(var->type->fields.array,

+						      size + 1);

+	    /* FINISHME: We should update the types of array

+	     * dereferences of this variable now.

+	     */

+	 }

+      }

+   }

+}

+

+static void

+add_uniform(void *mem_ctx, exec_list *uniforms, struct hash_table *ht,

+	    const char *name, const glsl_type *type, GLenum shader_type,

+	    unsigned *next_shader_pos, unsigned *total_uniforms)

+{

+   if (type->is_record()) {

+      for (unsigned int i = 0; i < type->length; i++) {

+	 const glsl_type *field_type = type->fields.structure[i].type;

+	 char *field_name = ralloc_asprintf(mem_ctx, "%s.%s", name,

+					    type->fields.structure[i].name);

+

+	 add_uniform(mem_ctx, uniforms, ht, field_name, field_type,

+		     shader_type, next_shader_pos, total_uniforms);

+      }

+   } else {

+      uniform_node *n = (uniform_node *) hash_table_find(ht, name);

+      unsigned int vec4_slots;

+      const glsl_type *array_elem_type = NULL;

+

+      if (type->is_array()) {

+	 array_elem_type = type->fields.array;

+	 /* Array of structures. */

+	 if (array_elem_type->is_record()) {

+	    for (unsigned int i = 0; i < type->length; i++) {

+	       char *elem_name = ralloc_asprintf(mem_ctx, "%s[%d]", name, i);

+	       add_uniform(mem_ctx, uniforms, ht, elem_name, array_elem_type,

+			   shader_type, next_shader_pos, total_uniforms);

+	    }

+	    return;

+	 }

+      }

+

+      /* Fix the storage size of samplers at 1 vec4 each. Be sure to pad out

+       * vectors to vec4 slots.

+       */

+      if (type->is_array()) {

+	 if (array_elem_type->is_sampler())

+	    vec4_slots = type->length;

+	 else

+	    vec4_slots = type->length * array_elem_type->matrix_columns;

+      } else if (type->is_sampler()) {

+	 vec4_slots = 1;

+      } else {

+	 vec4_slots = type->matrix_columns;

+      }

+

+      if (n == NULL) {

+	 n = (uniform_node *) calloc(1, sizeof(struct uniform_node));

+	 n->u = (gl_uniform *) calloc(1, sizeof(struct gl_uniform));

+	 n->slots = vec4_slots;

+

+	 n->u->Name = strdup(name);

+	 n->u->Type = type;

+	 n->u->VertPos = -1;

+	 n->u->FragPos = -1;

+	 n->u->GeomPos = -1;

+	 (*total_uniforms)++;

+

+	 hash_table_insert(ht, n, name);

+	 uniforms->push_tail(& n->link);

+      }

+

+      switch (shader_type) {

+      case GL_VERTEX_SHADER:

+	 n->u->VertPos = *next_shader_pos;

+	 break;

+      case GL_FRAGMENT_SHADER:

+	 n->u->FragPos = *next_shader_pos;

+	 break;

+      case GL_GEOMETRY_SHADER:

+	 n->u->GeomPos = *next_shader_pos;

+	 break;

+      }

+

+      (*next_shader_pos) += vec4_slots;

+   }

+}

+

+void

+assign_uniform_locations(struct gl_shader_program *prog)

+{

+   /* */

+   exec_list uniforms;

+   unsigned total_uniforms = 0;

+   hash_table *ht = hash_table_ctor(32, hash_table_string_hash,

+				    hash_table_string_compare);

+   void *mem_ctx = ralloc_context(NULL);

+

+   for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {

+      if (prog->_LinkedShaders[i] == NULL)

+	 continue;

+

+      unsigned next_position = 0;

+

+      foreach_list(node, prog->_LinkedShaders[i]->ir) {

+	 ir_variable *const var = ((ir_instruction *) node)->as_variable();

+

+	 if ((var == NULL) || (var->mode != ir_var_uniform))

+	    continue;

+

+	 if (strncmp(var->name, "gl_", 3) == 0) {

+	    /* At the moment, we don't allocate uniform locations for

+	     * builtin uniforms.  It's permitted by spec, and we'll

+	     * likely switch to doing that at some point, but not yet.

+	     */

+	    continue;

+	 }

+

+	 var->location = next_position;

+	 add_uniform(mem_ctx, &uniforms, ht, var->name, var->type,

+		     prog->_LinkedShaders[i]->Type,

+		     &next_position, &total_uniforms);

+      }

+   }

+

+   ralloc_free(mem_ctx);

+

+   gl_uniform_list *ul = (gl_uniform_list *)

+      calloc(1, sizeof(gl_uniform_list));

+

+   ul->Size = total_uniforms;

+   ul->NumUniforms = total_uniforms;

+   ul->Uniforms = (gl_uniform *) calloc(total_uniforms, sizeof(gl_uniform));

+

+   unsigned idx = 0;

+   uniform_node *next;

+   for (uniform_node *node = (uniform_node *) uniforms.head

+	   ; node->link.next != NULL

+	   ; node = next) {

+      next = (uniform_node *) node->link.next;

+

+      node->link.remove();

+      memcpy(&ul->Uniforms[idx], node->u, sizeof(gl_uniform));

+      idx++;

+

+      free(node->u);

+      free(node);

+   }

+

+   hash_table_dtor(ht);

+

+   prog->Uniforms = ul;

+}

+

+

+/**

+ * Find a contiguous set of available bits in a bitmask

+ *

+ * \param used_mask     Bits representing used (1) and unused (0) locations

+ * \param needed_count  Number of contiguous bits needed.

+ *

+ * \return

+ * Base location of the available bits on success or -1 on failure.

+ */

+int

+find_available_slots(unsigned used_mask, unsigned needed_count)

+{

+   unsigned needed_mask = (1 << needed_count) - 1;

+   const int max_bit_to_test = (8 * sizeof(used_mask)) - needed_count;

+

+   /* The comparison to 32 is redundant, but without it GCC emits "warning:

+    * cannot optimize possibly infinite loops" for the loop below.

+    */

+   if ((needed_count == 0) || (max_bit_to_test < 0) || (max_bit_to_test > 32))

+      return -1;

+

+   for (int i = 0; i <= max_bit_to_test; i++) {

+      if ((needed_mask & ~used_mask) == needed_mask)

+	 return i;

+

+      needed_mask <<= 1;

+   }

+

+   return -1;

+}

+

+

+bool

+assign_attribute_locations(gl_shader_program *prog, unsigned max_attribute_index)

+{

+   /* Mark invalid attribute locations as being used.

+    */

+   unsigned used_locations = (max_attribute_index >= 32)

+      ? ~0 : ~((1 << max_attribute_index) - 1);

+

+   gl_shader *const sh = prog->_LinkedShaders[0];

+   assert(sh->Type == GL_VERTEX_SHADER);

+

+   /* Operate in a total of four passes.

+    *

+    * 1. Invalidate the location assignments for all vertex shader inputs.

+    *

+    * 2. Assign locations for inputs that have user-defined (via

+    *    glBindVertexAttribLocation) locatoins.

+    *

+    * 3. Sort the attributes without assigned locations by number of slots

+    *    required in decreasing order.  Fragmentation caused by attribute

+    *    locations assigned by the application may prevent large attributes

+    *    from having enough contiguous space.

+    *

+    * 4. Assign locations to any inputs without assigned locations.

+    */

+

+   invalidate_variable_locations(sh, ir_var_in, VERT_ATTRIB_GENERIC0);

+

+   if (prog->Attributes != NULL) {

+      for (unsigned i = 0; i < prog->Attributes->NumParameters; i++) {

+	 ir_variable *const var =

+	    sh->symbols->get_variable(prog->Attributes->Parameters[i].Name);

+

+	 /* Note: attributes that occupy multiple slots, such as arrays or

+	  * matrices, may appear in the attrib array multiple times.

+	  */

+	 if ((var == NULL) || (var->location != -1))

+	    continue;

+

+	 /* From page 61 of the OpenGL 4.0 spec:

+	  *

+	  *     "LinkProgram will fail if the attribute bindings assigned by

+	  *     BindAttribLocation do not leave not enough space to assign a

+	  *     location for an active matrix attribute or an active attribute

+	  *     array, both of which require multiple contiguous generic

+	  *     attributes."

+	  *

+	  * Previous versions of the spec contain similar language but omit the

+	  * bit about attribute arrays.

+	  *

+	  * Page 61 of the OpenGL 4.0 spec also says:

+	  *

+	  *     "It is possible for an application to bind more than one

+	  *     attribute name to the same location. This is referred to as

+	  *     aliasing. This will only work if only one of the aliased

+	  *     attributes is active in the executable program, or if no path

+	  *     through the shader consumes more than one attribute of a set

+	  *     of attributes aliased to the same location. A link error can

+	  *     occur if the linker determines that every path through the

+	  *     shader consumes multiple aliased attributes, but

+	  *     implementations are not required to generate an error in this

+	  *     case."

+	  *

+	  * These two paragraphs are either somewhat contradictory, or I don't

+	  * fully understand one or both of them.

+	  */

+	 /* FINISHME: The code as currently written does not support attribute

+	  * FINISHME: location aliasing (see comment above).

+	  */

+	 const int attr = prog->Attributes->Parameters[i].StateIndexes[0];

+	 const unsigned slots = count_attribute_slots(var->type);

+

+	 /* Mask representing the contiguous slots that will be used by this

+	  * attribute.

+	  */

+	 const unsigned use_mask = (1 << slots) - 1;

+

+	 /* Generate a link error if the set of bits requested for this

+	  * attribute overlaps any previously allocated bits.

+	  */

+	 if ((~(use_mask << attr) & used_locations) != used_locations) {

+	    linker_error_printf(prog,

+				"insufficient contiguous attribute locations "

+				"available for vertex shader input `%s'",

+				var->name);

+	    return false;

+	 }

+

+	 var->location = VERT_ATTRIB_GENERIC0 + attr;

+	 used_locations |= (use_mask << attr);

+      }

+   }

+

+   /* Temporary storage for the set of attributes that need locations assigned.

+    */

+   struct temp_attr {

+      unsigned slots;

+      ir_variable *var;

+

+      /* Used below in the call to qsort. */

+      static int compare(const void *a, const void *b)

+      {

+	 const temp_attr *const l = (const temp_attr *) a;

+	 const temp_attr *const r = (const temp_attr *) b;

+

+	 /* Reversed because we want a descending order sort below. */

+	 return r->slots - l->slots;

+      }

+   } to_assign[16];

+

+   unsigned num_attr = 0;

+

+   foreach_list(node, sh->ir) {

+      ir_variable *const var = ((ir_instruction *) node)->as_variable();

+

+      if ((var == NULL) || (var->mode != ir_var_in))

+	 continue;

+

+      if (var->explicit_location) {

+	 const unsigned slots = count_attribute_slots(var->type);

+	 const unsigned use_mask = (1 << slots) - 1;

+	 const int attr = var->location - VERT_ATTRIB_GENERIC0;

+

+	 if ((var->location >= (int)(max_attribute_index + VERT_ATTRIB_GENERIC0))

+	     || (var->location < 0)) {

+	    linker_error_printf(prog,

+				"invalid explicit location %d specified for "

+				"`%s'\n",

+				(var->location < 0) ? var->location : attr,

+				var->name);

+	    return false;

+	 } else if (var->location >= VERT_ATTRIB_GENERIC0) {

+	    used_locations |= (use_mask << attr);

+	 }

+      }

+

+      /* The location was explicitly assigned, nothing to do here.

+       */

+      if (var->location != -1)

+	 continue;

+

+      to_assign[num_attr].slots = count_attribute_slots(var->type);

+      to_assign[num_attr].var = var;

+      num_attr++;

+   }

+

+   /* If all of the attributes were assigned locations by the application (or

+    * are built-in attributes with fixed locations), return early.  This should

+    * be the common case.

+    */

+   if (num_attr == 0)

+      return true;

+

+   qsort(to_assign, num_attr, sizeof(to_assign[0]), temp_attr::compare);

+

+   /* VERT_ATTRIB_GENERIC0 is a psdueo-alias for VERT_ATTRIB_POS.  It can only

+    * be explicitly assigned by via glBindAttribLocation.  Mark it as reserved

+    * to prevent it from being automatically allocated below.

+    */

+   find_deref_visitor find("gl_Vertex");

+   find.run(sh->ir);

+   if (find.variable_found())

+      used_locations |= (1 << 0);

+

+   for (unsigned i = 0; i < num_attr; i++) {

+      /* Mask representing the contiguous slots that will be used by this

+       * attribute.

+       */

+      const unsigned use_mask = (1 << to_assign[i].slots) - 1;

+

+      int location = find_available_slots(used_locations, to_assign[i].slots);

+

+      if (location < 0) {

+	 linker_error_printf(prog,

+			     "insufficient contiguous attribute locations "

+			     "available for vertex shader input `%s'",

+			     to_assign[i].var->name);

+	 return false;

+      }

+

+      to_assign[i].var->location = VERT_ATTRIB_GENERIC0 + location;

+      used_locations |= (use_mask << location);

+   }

+

+   return true;

+}

+

+

+/**

+ * Demote shader inputs and outputs that are not used in other stages

+ */

+void

+demote_shader_inputs_and_outputs(gl_shader *sh, enum ir_variable_mode mode)

+{

+   foreach_list(node, sh->ir) {

+      ir_variable *const var = ((ir_instruction *) node)->as_variable();

+

+      if ((var == NULL) || (var->mode != int(mode)))

+	 continue;

+

+      /* A shader 'in' or 'out' variable is only really an input or output if

+       * its value is used by other shader stages.  This will cause the variable

+       * to have a location assigned.

+       */

+      if (var->location == -1) {

+	 var->mode = ir_var_auto;

+      }

+   }

+}

+

+

+void

+assign_varying_locations(struct gl_shader_program *prog,

+			 gl_shader *producer, gl_shader *consumer)

+{

+   /* FINISHME: Set dynamically when geometry shader support is added. */

+   unsigned output_index = VERT_RESULT_VAR0;

+   unsigned input_index = FRAG_ATTRIB_VAR0;

+

+   /* Operate in a total of three passes.

+    *

+    * 1. Assign locations for any matching inputs and outputs.

+    *

+    * 2. Mark output variables in the producer that do not have locations as

+    *    not being outputs.  This lets the optimizer eliminate them.

+    *

+    * 3. Mark input variables in the consumer that do not have locations as

+    *    not being inputs.  This lets the optimizer eliminate them.

+    */

+

+   invalidate_variable_locations(producer, ir_var_out, VERT_RESULT_VAR0);

+   invalidate_variable_locations(consumer, ir_var_in, FRAG_ATTRIB_VAR0);

+

+   foreach_list(node, producer->ir) {

+      ir_variable *const output_var = ((ir_instruction *) node)->as_variable();

+

+      if ((output_var == NULL) || (output_var->mode != ir_var_out)

+	  || (output_var->location != -1))

+	 continue;

+

+      ir_variable *const input_var =

+	 consumer->symbols->get_variable(output_var->name);

+

+      if ((input_var == NULL) || (input_var->mode != ir_var_in))

+	 continue;

+

+      assert(input_var->location == -1);

+

+      output_var->location = output_index;

+      input_var->location = input_index;

+

+      /* FINISHME: Support for "varying" records in GLSL 1.50. */

+      assert(!output_var->type->is_record());

+

+      if (output_var->type->is_array()) {

+	 const unsigned slots = output_var->type->length

+	    * output_var->type->fields.array->matrix_columns;

+

+	 output_index += slots;

+	 input_index += slots;

+      } else {

+	 const unsigned slots = output_var->type->matrix_columns;

+

+	 output_index += slots;

+	 input_index += slots;

+      }

+   }

+

+   foreach_list(node, consumer->ir) {

+      ir_variable *const var = ((ir_instruction *) node)->as_variable();

+

+      if ((var == NULL) || (var->mode != ir_var_in))

+	 continue;

+

+      if (var->location == -1) {

+	 if (prog->Version <= 120) {

+	    /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec:

+	     *

+	     *     Only those varying variables used (i.e. read) in

+	     *     the fragment shader executable must be written to

+	     *     by the vertex shader executable; declaring

+	     *     superfluous varying variables in a vertex shader is

+	     *     permissible.

+	     *

+	     * We interpret this text as meaning that the VS must

+	     * write the variable for the FS to read it.  See

+	     * "glsl1-varying read but not written" in piglit.

+	     */

+

+	    linker_error_printf(prog, "fragment shader varying %s not written "

+				"by vertex shader\n.", var->name);

+	    prog->LinkStatus = false;

+	 }

+

+	 /* An 'in' variable is only really a shader input if its

+	  * value is written by the previous stage.

+	  */

+	 var->mode = ir_var_auto;

+      }

+   }

+}

+

+

+void

+link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)

+{

+   void *mem_ctx = ralloc_context(NULL); // temporary linker context

+

+   prog->LinkStatus = false;

+   prog->Validated = false;

+   prog->_Used = false;

+

+   if (prog->InfoLog != NULL)

+      ralloc_free(prog->InfoLog);

+

+   prog->InfoLog = ralloc_strdup(NULL, "");

+

+   /* Separate the shaders into groups based on their type.

+    */

+   struct gl_shader **vert_shader_list;

+   unsigned num_vert_shaders = 0;

+   struct gl_shader **frag_shader_list;

+   unsigned num_frag_shaders = 0;

+

+   vert_shader_list = (struct gl_shader **)

+      calloc(2 * prog->NumShaders, sizeof(struct gl_shader *));

+   frag_shader_list =  &vert_shader_list[prog->NumShaders];

+

+   unsigned min_version = UINT_MAX;

+   unsigned max_version = 0;

+   for (unsigned i = 0; i < prog->NumShaders; i++) {

+      min_version = MIN2(min_version, prog->Shaders[i]->Version);

+      max_version = MAX2(max_version, prog->Shaders[i]->Version);

+

+      switch (prog->Shaders[i]->Type) {

+      case GL_VERTEX_SHADER:

+	 vert_shader_list[num_vert_shaders] = prog->Shaders[i];

+	 num_vert_shaders++;

+	 break;

+      case GL_FRAGMENT_SHADER:

+	 frag_shader_list[num_frag_shaders] = prog->Shaders[i];

+	 num_frag_shaders++;

+	 break;

+      case GL_GEOMETRY_SHADER:

+	 /* FINISHME: Support geometry shaders. */

+	 assert(prog->Shaders[i]->Type != GL_GEOMETRY_SHADER);

+	 break;

+      }

+   }

+

+   /* Previous to GLSL version 1.30, different compilation units could mix and

+    * match shading language versions.  With GLSL 1.30 and later, the versions

+    * of all shaders must match.

+    */

+   assert(min_version >= 100);

+   assert(max_version <= 130);

+   if ((max_version >= 130 || min_version == 100)

+       && min_version != max_version) {

+      linker_error_printf(prog, "all shaders must use same shading "

+			  "language version\n");

+      goto done;

+   }

+

+   prog->Version = max_version;

+

+   for (unsigned int i = 0; i < MESA_SHADER_TYPES; i++) {

+      if (prog->_LinkedShaders[i] != NULL)

+	 ctx->Driver.DeleteShader(ctx, prog->_LinkedShaders[i]);

+

+      prog->_LinkedShaders[i] = NULL;

+   }

+

+   /* Link all shaders for a particular stage and validate the result.

+    */

+   if (num_vert_shaders > 0) {

+      gl_shader *const sh =

+	 link_intrastage_shaders(mem_ctx, ctx, prog, vert_shader_list,

+				 num_vert_shaders);

+

+      if (sh == NULL)

+	 goto done;

+

+      if (!validate_vertex_shader_executable(prog, sh))

+	 goto done;

+

+      _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_VERTEX],

+			     sh);

+   }

+

+   if (num_frag_shaders > 0) {

+      gl_shader *const sh =

+	 link_intrastage_shaders(mem_ctx, ctx, prog, frag_shader_list,

+				 num_frag_shaders);

+

+      if (sh == NULL)

+	 goto done;

+

+      if (!validate_fragment_shader_executable(prog, sh))

+	 goto done;

+

+      _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_FRAGMENT],

+			     sh);

+   }

+

+   /* Here begins the inter-stage linking phase.  Some initial validation is

+    * performed, then locations are assigned for uniforms, attributes, and

+    * varyings.

+    */

+   if (cross_validate_uniforms(prog)) {

+      unsigned prev;

+

+      for (prev = 0; prev < MESA_SHADER_TYPES; prev++) {

+	 if (prog->_LinkedShaders[prev] != NULL)

+	    break;

+      }

+

+      /* Validate the inputs of each stage with the output of the preceeding

+       * stage.

+       */

+      for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) {

+	 if (prog->_LinkedShaders[i] == NULL)

+	    continue;

+

+	 if (!cross_validate_outputs_to_inputs(prog,

+					       prog->_LinkedShaders[prev],

+					       prog->_LinkedShaders[i]))

+	    goto done;

+

+	 prev = i;

+      }

+

+      prog->LinkStatus = true;

+   }

+

+   /* Do common optimization before assigning storage for attributes,

+    * uniforms, and varyings.  Later optimization could possibly make

+    * some of that unused.

+    */

+   for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {

+      if (prog->_LinkedShaders[i] == NULL)

+	 continue;

+

+      while (do_common_optimization(prog->_LinkedShaders[i]->ir, true, 32))

+	 ;

+   }

+

+   update_array_sizes(prog);

+

+   assign_uniform_locations(prog);

+

+   if (prog->_LinkedShaders[MESA_SHADER_VERTEX] != NULL) {

+      /* FINISHME: The value of the max_attribute_index parameter is

+       * FINISHME: implementation dependent based on the value of

+       * FINISHME: GL_MAX_VERTEX_ATTRIBS.  GL_MAX_VERTEX_ATTRIBS must be

+       * FINISHME: at least 16, so hardcode 16 for now.

+       */

+      if (!assign_attribute_locations(prog, 16)) {

+	 prog->LinkStatus = false;

+	 goto done;

+      }

+   }

+

+   unsigned prev;

+   for (prev = 0; prev < MESA_SHADER_TYPES; prev++) {

+      if (prog->_LinkedShaders[prev] != NULL)

+	 break;

+   }

+

+   for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) {

+      if (prog->_LinkedShaders[i] == NULL)

+	 continue;

+

+      assign_varying_locations(prog,

+			       prog->_LinkedShaders[prev],

+			       prog->_LinkedShaders[i]);

+      prev = i;

+   }

+

+   if (prog->_LinkedShaders[MESA_SHADER_VERTEX] != NULL) {

+      demote_shader_inputs_and_outputs(prog->_LinkedShaders[MESA_SHADER_VERTEX],

+				       ir_var_out);

+   }

+

+   if (prog->_LinkedShaders[MESA_SHADER_GEOMETRY] != NULL) {

+      gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_GEOMETRY];

+

+      demote_shader_inputs_and_outputs(sh, ir_var_in);

+      demote_shader_inputs_and_outputs(sh, ir_var_inout);

+      demote_shader_inputs_and_outputs(sh, ir_var_out);

+   }

+

+   if (prog->_LinkedShaders[MESA_SHADER_FRAGMENT] != NULL) {

+      gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];

+

+      demote_shader_inputs_and_outputs(sh, ir_var_in);

+   }

+

+   /* OpenGL ES requires that a vertex shader and a fragment shader both be

+    * present in a linked program.  By checking for use of shading language

+    * version 1.00, we also catch the GL_ARB_ES2_compatibility case.

+    */

+   if (ctx->API == API_OPENGLES2 || prog->Version == 100) {

+      if (prog->_LinkedShaders[MESA_SHADER_VERTEX] == NULL) {

+	 linker_error_printf(prog, "program lacks a vertex shader\n");

+	 prog->LinkStatus = false;

+      } else if (prog->_LinkedShaders[MESA_SHADER_FRAGMENT] == NULL) {

+	 linker_error_printf(prog, "program lacks a fragment shader\n");

+	 prog->LinkStatus = false;

+      }

+   }

+

+   /* FINISHME: Assign fragment shader output locations. */

+

+done:

+   free(vert_shader_list);

+

+   for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {

+      if (prog->_LinkedShaders[i] == NULL)

+	 continue;

+

+      /* Retain any live IR, but trash the rest. */

+      reparent_ir(prog->_LinkedShaders[i]->ir, prog->_LinkedShaders[i]->ir);

+   }

+

+   ralloc_free(mem_ctx);

+}

diff --git a/mesalib/src/mesa/main/mtypes.h b/mesalib/src/mesa/main/mtypes.h
index 020595bd0..714d63afc 100644
--- a/mesalib/src/mesa/main/mtypes.h
+++ b/mesalib/src/mesa/main/mtypes.h
@@ -1240,7 +1240,7 @@ struct gl_texture_image
 				 *   GL_DEPTH_COMPONENT or GL_DEPTH_STENCIL_EXT

                                  *   only. Used for choosing TexEnv arithmetic.

 				 */

-   GLuint TexFormat;            /**< The actual format: MESA_FORMAT_x */

+   gl_format TexFormat;         /**< The actual texture memory format */

 

    GLuint Border;		/**< 0 or 1 */

    GLuint Width;		/**< = 2^WidthLog2 + 2*Border */

@@ -1643,6 +1643,7 @@ struct gl_array_attrib
    GLuint RestartIndex;

 

    GLbitfield NewState;		/**< mask of _NEW_ARRAY_* values */

+   GLboolean RebindArrays; /**< whether the VBO module should rebind arrays */

 

    /* GL_ARB_vertex_buffer_object */

    struct gl_buffer_object *ArrayBufferObj;

@@ -2379,7 +2380,7 @@ struct gl_renderbuffer
    GLenum InternalFormat; /**< The user-specified format */

    GLenum _BaseFormat;    /**< Either GL_RGB, GL_RGBA, GL_DEPTH_COMPONENT or

                                GL_STENCIL_INDEX. */

-   GLuint Format;         /**< The actual format: MESA_FORMAT_x */

+   gl_format Format;      /**< The actual renderbuffer memory format */

 

    GLubyte NumSamples;

 

diff --git a/mesalib/src/mesa/main/state.c b/mesalib/src/mesa/main/state.c
index c07e2c380..f11578faf 100644
--- a/mesalib/src/mesa/main/state.c
+++ b/mesalib/src/mesa/main/state.c
@@ -662,6 +662,8 @@ _mesa_update_state_locked( struct gl_context *ctx )
    ctx->NewState = 0;

    ctx->Driver.UpdateState(ctx, new_state);

    ctx->Array.NewState = 0;

+   if (!ctx->Array.RebindArrays)

+      ctx->Array.RebindArrays = (new_state & (_NEW_ARRAY | _NEW_PROGRAM)) != 0;

 }

 

 

diff --git a/mesalib/src/mesa/state_tracker/st_atom_blend.c b/mesalib/src/mesa/state_tracker/st_atom_blend.c
index 26bb3dab9..e3b3b9936 100644
--- a/mesalib/src/mesa/state_tracker/st_atom_blend.c
+++ b/mesalib/src/mesa/state_tracker/st_atom_blend.c
@@ -191,7 +191,7 @@ update_blend( struct st_context *st )
 {

    struct pipe_blend_state *blend = &st->state.blend;

    unsigned num_state = 1;

-   unsigned i;

+   unsigned i, j;

 

    memset(blend, 0, sizeof(*blend));

 

@@ -214,12 +214,15 @@ update_blend( struct st_context *st )
    }

    else if (st->ctx->Color.BlendEnabled) {

       /* blending enabled */

-      for (i = 0; i < num_state; i++) {

+      for (i = 0, j = 0; i < num_state; i++) {

 

          blend->rt[i].blend_enable = (st->ctx->Color.BlendEnabled >> i) & 0x1;

 

+         if (st->ctx->Extensions.ARB_draw_buffers_blend)

+            j = i;

+

          blend->rt[i].rgb_func =

-            translate_blend(st->ctx->Color.Blend[i].EquationRGB);

+            translate_blend(st->ctx->Color.Blend[j].EquationRGB);

 

          if (st->ctx->Color.Blend[i].EquationRGB == GL_MIN ||

              st->ctx->Color.Blend[i].EquationRGB == GL_MAX) {

@@ -229,13 +232,13 @@ update_blend( struct st_context *st )
          }

          else {

             blend->rt[i].rgb_src_factor =

-               translate_blend(st->ctx->Color.Blend[i].SrcRGB);

+               translate_blend(st->ctx->Color.Blend[j].SrcRGB);

             blend->rt[i].rgb_dst_factor =

-               translate_blend(st->ctx->Color.Blend[i].DstRGB);

+               translate_blend(st->ctx->Color.Blend[j].DstRGB);

          }

 

          blend->rt[i].alpha_func =

-            translate_blend(st->ctx->Color.Blend[i].EquationA);

+            translate_blend(st->ctx->Color.Blend[j].EquationA);

 

          if (st->ctx->Color.Blend[i].EquationA == GL_MIN ||

              st->ctx->Color.Blend[i].EquationA == GL_MAX) {

@@ -245,9 +248,9 @@ update_blend( struct st_context *st )
          }

          else {

             blend->rt[i].alpha_src_factor =

-               translate_blend(st->ctx->Color.Blend[i].SrcA);

+               translate_blend(st->ctx->Color.Blend[j].SrcA);

             blend->rt[i].alpha_dst_factor =

-               translate_blend(st->ctx->Color.Blend[i].DstA);

+               translate_blend(st->ctx->Color.Blend[j].DstA);

          }

       }

    }

diff --git a/mesalib/src/mesa/state_tracker/st_cb_bitmap.c b/mesalib/src/mesa/state_tracker/st_cb_bitmap.c
index ddd130a81..51623c179 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_bitmap.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_bitmap.c
@@ -349,6 +349,7 @@ setup_bitmap_vertex_data(struct st_context *st, bool normalized,
    if (!st->bitmap.vbuf) {

       st->bitmap.vbuf = pipe_buffer_create(pipe->screen, 

                                            PIPE_BIND_VERTEX_BUFFER,

+                                           PIPE_USAGE_STREAM,

                                            max_slots *

                                            sizeof(st->bitmap.vertices));

    }

@@ -456,6 +457,7 @@ draw_bitmap_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
    cso_save_fragment_shader(cso);

    cso_save_vertex_shader(cso);

    cso_save_vertex_elements(cso);

+   cso_save_vertex_buffers(cso);

 

    /* rasterizer state: just scissor */

    st->bitmap.rasterizer.scissor = ctx->Scissor.Enabled;

@@ -517,7 +519,7 @@ draw_bitmap_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
                                      sv->texture->target != PIPE_TEXTURE_RECT,

                                      x, y, width, height, z, color);

 

-   util_draw_vertex_buffer(pipe, st->bitmap.vbuf, offset,

+   util_draw_vertex_buffer(pipe, st->cso_context, st->bitmap.vbuf, offset,

                            PIPE_PRIM_TRIANGLE_FAN,

                            4,  /* verts */

                            3); /* attribs/vert */

@@ -531,6 +533,7 @@ draw_bitmap_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
    cso_restore_fragment_shader(cso);

    cso_restore_vertex_shader(cso);

    cso_restore_vertex_elements(cso);

+   cso_restore_vertex_buffers(cso);

 }

 

 

diff --git a/mesalib/src/mesa/state_tracker/st_cb_bufferobjects.c b/mesalib/src/mesa/state_tracker/st_cb_bufferobjects.c
index ba8a8cf89..9dd1f8a3e 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_bufferobjects.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_bufferobjects.c
@@ -171,7 +171,7 @@ st_bufferobj_data(struct gl_context *ctx,
    struct st_context *st = st_context(ctx);

    struct pipe_context *pipe = st->pipe;

    struct st_buffer_object *st_obj = st_buffer_object(obj);

-   unsigned buffer_usage;

+   unsigned bind, pipe_usage;

 

    st_obj->Base.Size = size;

    st_obj->Base.Usage = usage;

@@ -179,22 +179,43 @@ st_bufferobj_data(struct gl_context *ctx,
    switch(target) {

    case GL_PIXEL_PACK_BUFFER_ARB:

    case GL_PIXEL_UNPACK_BUFFER_ARB:

-      buffer_usage = PIPE_BIND_RENDER_TARGET;

+      bind = PIPE_BIND_RENDER_TARGET | PIPE_BIND_SAMPLER_VIEW;

       break;

    case GL_ARRAY_BUFFER_ARB:

-      buffer_usage = PIPE_BIND_VERTEX_BUFFER;

+      bind = PIPE_BIND_VERTEX_BUFFER;

       break;

    case GL_ELEMENT_ARRAY_BUFFER_ARB:

-      buffer_usage = PIPE_BIND_INDEX_BUFFER;

+      bind = PIPE_BIND_INDEX_BUFFER;

       break;

    default:

-      buffer_usage = 0;

+      bind = 0;

+   }

+

+   switch (usage) {

+   case GL_STATIC_DRAW:

+   case GL_STATIC_READ:

+   case GL_STATIC_COPY:

+      pipe_usage = PIPE_USAGE_STATIC;

+      break;

+   case GL_DYNAMIC_DRAW:

+   case GL_DYNAMIC_READ:

+   case GL_DYNAMIC_COPY:

+      pipe_usage = PIPE_USAGE_DYNAMIC;

+      break;

+   case GL_STREAM_DRAW:

+   case GL_STREAM_READ:

+   case GL_STREAM_COPY:

+      pipe_usage = PIPE_USAGE_STREAM;

+      break;

+   default:

+      pipe_usage = PIPE_USAGE_DEFAULT;

    }

 

    pipe_resource_reference( &st_obj->buffer, NULL );

 

    if (size != 0) {

-      st_obj->buffer = pipe_buffer_create(pipe->screen, buffer_usage, size);

+      st_obj->buffer = pipe_buffer_create(pipe->screen, bind,

+                                          pipe_usage, size);

 

       if (!st_obj->buffer) {

          return GL_FALSE;

diff --git a/mesalib/src/mesa/state_tracker/st_cb_clear.c b/mesalib/src/mesa/state_tracker/st_cb_clear.c
index 3e27be271..461f4e315 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_clear.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_clear.c
@@ -138,6 +138,7 @@ draw_quad(struct st_context *st,
    if (!st->clear.vbuf) {

       st->clear.vbuf = pipe_buffer_create(pipe->screen,

                                           PIPE_BIND_VERTEX_BUFFER,

+                                          PIPE_USAGE_STREAM,

                                           max_slots * sizeof(st->clear.vertices));

    }

 

@@ -172,7 +173,8 @@ draw_quad(struct st_context *st,
                                            st->clear.vertices);

 

    /* draw */

-   util_draw_vertex_buffer(pipe, 

+   util_draw_vertex_buffer(pipe,

+                           st->cso_context,

                            st->clear.vbuf, 

                            st->clear.vbuf_slot * sizeof(st->clear.vertices),

                            PIPE_PRIM_TRIANGLE_FAN,

@@ -221,6 +223,7 @@ clear_with_quad(struct gl_context *ctx,
    cso_save_fragment_shader(st->cso_context);

    cso_save_vertex_shader(st->cso_context);

    cso_save_vertex_elements(st->cso_context);

+   cso_save_vertex_buffers(st->cso_context);

 

    /* blend state: RGBA masking */

    {

@@ -309,6 +312,7 @@ clear_with_quad(struct gl_context *ctx,
    cso_restore_fragment_shader(st->cso_context);

    cso_restore_vertex_shader(st->cso_context);

    cso_restore_vertex_elements(st->cso_context);

+   cso_restore_vertex_buffers(st->cso_context);

 }

 

 

diff --git a/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c b/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c
index 56c7e8581..18ac6c8e0 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c
@@ -522,10 +522,11 @@ draw_quad(struct gl_context *ctx, GLfloat x0, GLfloat y0, GLfloat z,
       /* allocate/load buffer object with vertex data */

       buf = pipe_buffer_create(pipe->screen,

 			       PIPE_BIND_VERTEX_BUFFER,

+			       PIPE_USAGE_STATIC,

                                sizeof(verts));

       pipe_buffer_write(st->pipe, buf, 0, sizeof(verts), verts);

 

-      util_draw_vertex_buffer(pipe, buf, 0,

+      util_draw_vertex_buffer(pipe, st->cso_context, buf, 0,

                               PIPE_PRIM_QUADS,

                               4,  /* verts */

                               3); /* attribs/vert */

@@ -570,6 +571,7 @@ draw_textured_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
    cso_save_fragment_shader(cso);

    cso_save_vertex_shader(cso);

    cso_save_vertex_elements(cso);

+   cso_save_vertex_buffers(cso);

    if (write_stencil) {

       cso_save_depth_stencil_alpha(cso);

       cso_save_blend(cso);

@@ -686,6 +688,7 @@ draw_textured_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
    cso_restore_fragment_shader(cso);

    cso_restore_vertex_shader(cso);

    cso_restore_vertex_elements(cso);

+   cso_restore_vertex_buffers(cso);

    if (write_stencil) {

       cso_restore_depth_stencil_alpha(cso);

       cso_restore_blend(cso);

diff --git a/mesalib/src/mesa/state_tracker/st_cb_drawtex.c b/mesalib/src/mesa/state_tracker/st_cb_drawtex.c
index 5976f1048..db299eb8d 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_drawtex.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_drawtex.c
@@ -139,6 +139,7 @@ st_DrawTex(struct gl_context *ctx, GLfloat x, GLfloat y, GLfloat z,
 

    /* create the vertex buffer */

    vbuffer = pipe_buffer_create(pipe->screen, PIPE_BIND_VERTEX_BUFFER,

+                                PIPE_USAGE_STREAM,

                                 numAttribs * 4 * 4 * sizeof(GLfloat));

 

    /* load vertex buffer */

@@ -230,6 +231,7 @@ st_DrawTex(struct gl_context *ctx, GLfloat x, GLfloat y, GLfloat z,
    cso_save_viewport(cso);

    cso_save_vertex_shader(cso);

    cso_save_vertex_elements(cso);

+   cso_save_vertex_buffers(cso);

 

    {

       void *vs = lookup_shader(pipe, numAttribs,

@@ -264,7 +266,7 @@ st_DrawTex(struct gl_context *ctx, GLfloat x, GLfloat y, GLfloat z,
    }

 

 

-   util_draw_vertex_buffer(pipe, vbuffer,

+   util_draw_vertex_buffer(pipe, cso, vbuffer,

                            0,  /* offset */

                            PIPE_PRIM_TRIANGLE_FAN,

                            4,  /* verts */

@@ -277,6 +279,7 @@ st_DrawTex(struct gl_context *ctx, GLfloat x, GLfloat y, GLfloat z,
    cso_restore_viewport(cso);

    cso_restore_vertex_shader(cso);

    cso_restore_vertex_elements(cso);

+   cso_restore_vertex_buffers(cso);

 }

 

 

diff --git a/mesalib/src/mesa/state_tracker/st_context.c b/mesalib/src/mesa/state_tracker/st_context.c
index b16d965f8..8f3a7415d 100644
--- a/mesalib/src/mesa/state_tracker/st_context.c
+++ b/mesalib/src/mesa/state_tracker/st_context.c
@@ -203,6 +203,11 @@ static void st_destroy_context_priv( struct st_context *st )
    st_destroy_drawpix(st);

    st_destroy_drawtex(st);

 

+   /* Unreference any user vertex buffers. */

+   for (i = 0; i < st->num_user_vbs; i++) {

+      pipe_resource_reference(&st->user_vb[i], NULL);

+   }

+

    for (i = 0; i < Elements(st->state.sampler_views); i++) {

       pipe_sampler_view_reference(&st->state.sampler_views[i], NULL);

    }

diff --git a/mesalib/src/mesa/state_tracker/st_context.h b/mesalib/src/mesa/state_tracker/st_context.h
index 492ee600e..ef54fd7cd 100644
--- a/mesalib/src/mesa/state_tracker/st_context.h
+++ b/mesalib/src/mesa/state_tracker/st_context.h
@@ -185,6 +185,11 @@ struct st_context
 

    int force_msaa;

    void *winsys_drawable_handle;

+

+   /* User vertex buffers. */

+   struct pipe_resource *user_vb[PIPE_MAX_ATTRIBS];

+   unsigned user_vb_stride[PIPE_MAX_ATTRIBS];

+   unsigned num_user_vbs;

 };

 

 

diff --git a/mesalib/src/mesa/state_tracker/st_draw.c b/mesalib/src/mesa/state_tracker/st_draw.c
index 19466ea44..a31cbdc9e 100644
--- a/mesalib/src/mesa/state_tracker/st_draw.c
+++ b/mesalib/src/mesa/state_tracker/st_draw.c
@@ -243,13 +243,11 @@ st_pipe_vertex_format(GLenum type, GLuint size, GLenum format,
 static GLboolean

 is_interleaved_arrays(const struct st_vertex_program *vp,

                       const struct st_vp_variant *vpv,

-                      const struct gl_client_array **arrays,

-                      GLboolean *userSpace)

+                      const struct gl_client_array **arrays)

 {

    GLuint attr;

    const struct gl_buffer_object *firstBufObj = NULL;

    GLint firstStride = -1;

-   GLuint num_client_arrays = 0;

    const GLubyte *client_addr = NULL;

 

    for (attr = 0; attr < vpv->num_inputs; attr++) {

@@ -263,9 +261,8 @@ is_interleaved_arrays(const struct st_vertex_program *vp,
       else if (firstStride != stride) {

          return GL_FALSE;

       }

-         

+

       if (!bufObj || !bufObj->Name) {

-         num_client_arrays++;

          /* Try to detect if the client-space arrays are

           * "close" to each other.

           */

@@ -285,57 +282,11 @@ is_interleaved_arrays(const struct st_vertex_program *vp,
       }

    }

 

-   *userSpace = (num_client_arrays == vpv->num_inputs);

-   /* debug_printf("user space: %s (%d arrays, %d inputs)\n",

-      (int)*userSpace ? "Yes" : "No", num_client_arrays, vp->num_inputs); */

-

    return GL_TRUE;

 }

 

 

 /**

- * Compute the memory range occupied by the arrays.

- */

-static void

-get_arrays_bounds(const struct st_vertex_program *vp,

-                  const struct st_vp_variant *vpv,

-                  const struct gl_client_array **arrays,

-                  GLuint max_index,

-                  const GLubyte **low, const GLubyte **high)

-{

-   const GLubyte *low_addr = NULL;

-   const GLubyte *high_addr = NULL;

-   GLuint attr;

-

-   /* debug_printf("get_arrays_bounds: Handling %u attrs\n", vpv->num_inputs); */

-

-   for (attr = 0; attr < vpv->num_inputs; attr++) {

-      const GLuint mesaAttr = vp->index_to_input[attr];

-      const GLint stride = arrays[mesaAttr]->StrideB;

-      const GLubyte *start = arrays[mesaAttr]->Ptr;

-      const unsigned sz = (arrays[mesaAttr]->Size * 

-                           _mesa_sizeof_type(arrays[mesaAttr]->Type));

-      const GLubyte *end = start + (max_index * stride) + sz;

-

-      /* debug_printf("attr %u: stride %d size %u start %p end %p\n",

-         attr, stride, sz, start, end); */

-

-      if (attr == 0) {

-         low_addr = start;

-         high_addr = end;

-      }

-      else {

-         low_addr = MIN2(low_addr, start);

-         high_addr = MAX2(high_addr, end);

-      }

-   }

-

-   *low = low_addr;

-   *high = high_addr;

-}

-

-

-/**

  * Set up for drawing interleaved arrays that all live in one VBO

  * or all live in user space.

  * \param vbuffer  returns vertex buffer info

@@ -346,15 +297,21 @@ setup_interleaved_attribs(struct gl_context *ctx,
                           const struct st_vertex_program *vp,

                           const struct st_vp_variant *vpv,

                           const struct gl_client_array **arrays,

-                          GLuint max_index,

-                          GLboolean userSpace,

                           struct pipe_vertex_buffer *vbuffer,

-                          struct pipe_vertex_element velements[])

+                          struct pipe_vertex_element velements[],

+                          unsigned max_index)

 {

    struct st_context *st = st_context(ctx);

    struct pipe_context *pipe = st->pipe;

    GLuint attr;

-   const GLubyte *offset0 = NULL;

+   const GLubyte *low_addr = NULL;

+

+   /* Find the lowest address. */

+   for (attr = 0; attr < vpv->num_inputs; attr++) {

+      const GLubyte *start = arrays[vp->index_to_input[attr]]->Ptr;

+

+      low_addr = !low_addr ? start : MIN2(low_addr, start);

+   }

 

    for (attr = 0; attr < vpv->num_inputs; attr++) {

       const GLuint mesaAttr = vp->index_to_input[attr];

@@ -362,40 +319,28 @@ setup_interleaved_attribs(struct gl_context *ctx,
       struct st_buffer_object *stobj = st_buffer_object(bufobj);

       GLsizei stride = arrays[mesaAttr]->StrideB;

 

-      /*printf("stobj %u = %p\n", attr, (void*)stobj);*/

-

       if (attr == 0) {

-         const GLubyte *low, *high;

-

-         get_arrays_bounds(vp, vpv, arrays, max_index, &low, &high);

-         /* debug_printf("buffer range: %p %p range %d max index %u\n",

-            low, high, high - low, max_index); */

-

-         offset0 = low;

-         if (userSpace) {

+         if (bufobj && bufobj->Name) {

+            vbuffer->buffer = NULL;

+            pipe_resource_reference(&vbuffer->buffer, stobj->buffer);

+            vbuffer->buffer_offset = pointer_to_offset(low_addr);

+         } else {

             vbuffer->buffer =

-               pipe_user_buffer_create(pipe->screen, (void *) low, high - low,

+               pipe_user_buffer_create(pipe->screen, (void*)low_addr,

+                                       stride * (max_index + 1),

 				       PIPE_BIND_VERTEX_BUFFER);

             vbuffer->buffer_offset = 0;

-         }

-         else {

-            vbuffer->buffer = NULL;

-            pipe_resource_reference(&vbuffer->buffer, stobj->buffer);

-            vbuffer->buffer_offset = pointer_to_offset(low);

+

+            /* Track user vertex buffers. */

+            pipe_resource_reference(&st->user_vb[0], vbuffer->buffer);

+            st->user_vb_stride[0] = stride;

+            st->num_user_vbs = 1;

          }

          vbuffer->stride = stride; /* in bytes */

-         vbuffer->max_index = max_index;

       }

 

-      /*

-      if (arrays[mesaAttr]->InstanceDivisor)

-         vbuffer[attr].max_index = arrays[mesaAttr]->_MaxElement;

-      else

-         vbuffer[attr].max_index = max_index;

-      */

-

       velements[attr].src_offset =

-         (unsigned) (arrays[mesaAttr]->Ptr - offset0);

+         (unsigned) (arrays[mesaAttr]->Ptr - low_addr);

       velements[attr].instance_divisor = arrays[mesaAttr]->InstanceDivisor;

       velements[attr].vertex_buffer_index = 0;

       velements[attr].src_format =

@@ -419,10 +364,9 @@ setup_non_interleaved_attribs(struct gl_context *ctx,
                               const struct st_vertex_program *vp,

                               const struct st_vp_variant *vpv,

                               const struct gl_client_array **arrays,

-                              GLuint max_index,

-                              GLboolean *userSpace,

                               struct pipe_vertex_buffer vbuffer[],

-                              struct pipe_vertex_element velements[])

+                              struct pipe_vertex_element velements[],

+                              unsigned max_index)

 {

    struct st_context *st = st_context(ctx);

    struct pipe_context *pipe = st->pipe;

@@ -433,8 +377,6 @@ setup_non_interleaved_attribs(struct gl_context *ctx,
       struct gl_buffer_object *bufobj = arrays[mesaAttr]->BufferObj;

       GLsizei stride = arrays[mesaAttr]->StrideB;

 

-      *userSpace = GL_FALSE;

-

       if (bufobj && bufobj->Name) {

          /* Attribute data is in a VBO.

           * Recall that for VBOs, the gl_client_array->Ptr field is

@@ -442,37 +384,23 @@ setup_non_interleaved_attribs(struct gl_context *ctx,
           */

          struct st_buffer_object *stobj = st_buffer_object(bufobj);

          assert(stobj->buffer);

-         /*printf("stobj %u = %p\n", attr, (void*) stobj);*/

 

          vbuffer[attr].buffer = NULL;

          pipe_resource_reference(&vbuffer[attr].buffer, stobj->buffer);

          vbuffer[attr].buffer_offset = pointer_to_offset(arrays[mesaAttr]->Ptr);

       }

       else {

-         /* attribute data is in user-space memory, not a VBO */

-         uint bytes;

-         /*printf("user-space array %d stride %d\n", attr, stride);*/

-	

-         *userSpace = GL_TRUE;

-

          /* wrap user data */

          if (arrays[mesaAttr]->Ptr) {

-            /* user's vertex array */

-            if (arrays[mesaAttr]->StrideB) {

-               bytes = arrays[mesaAttr]->StrideB * (max_index + 1);

-            }

-            else {

-               bytes = arrays[mesaAttr]->Size

-                  * _mesa_sizeof_type(arrays[mesaAttr]->Type);

-            }

             vbuffer[attr].buffer = 

 	       pipe_user_buffer_create(pipe->screen,

-				       (void *) arrays[mesaAttr]->Ptr, bytes,

+				       (void *) arrays[mesaAttr]->Ptr,

+				       stride * (max_index + 1),

 				       PIPE_BIND_VERTEX_BUFFER);

          }

          else {

             /* no array, use ctx->Current.Attrib[] value */

-            bytes = sizeof(ctx->Current.Attrib[0]);

+            uint bytes = sizeof(ctx->Current.Attrib[0]);

             vbuffer[attr].buffer = 

 	       pipe_user_buffer_create(pipe->screen,

 				       (void *) ctx->Current.Attrib[mesaAttr],

@@ -482,16 +410,15 @@ setup_non_interleaved_attribs(struct gl_context *ctx,
          }

 

          vbuffer[attr].buffer_offset = 0;

-      }

 

-      assert(velements[attr].src_offset <= 2048); /* 11-bit field */

+         /* Track user vertex buffers. */

+         pipe_resource_reference(&st->user_vb[attr], vbuffer->buffer);

+         st->user_vb_stride[attr] = stride;

+         st->num_user_vbs = MAX2(st->num_user_vbs, attr+1);

+      }

 

       /* common-case setup */

       vbuffer[attr].stride = stride; /* in bytes */

-      if (arrays[mesaAttr]->InstanceDivisor)

-         vbuffer[attr].max_index = arrays[mesaAttr]->_MaxElement;

-      else

-         vbuffer[attr].max_index = max_index;

 

       velements[attr].src_offset = 0;

       velements[attr].instance_divisor = arrays[mesaAttr]->InstanceDivisor;

@@ -609,6 +536,62 @@ translate_prim(const struct gl_context *ctx, unsigned prim)
 }

 

 

+static void

+st_validate_varrays(struct gl_context *ctx,

+                    const struct gl_client_array **arrays,

+                    unsigned max_index)

+{

+   struct st_context *st = st_context(ctx);

+   const struct st_vertex_program *vp;

+   const struct st_vp_variant *vpv;

+   struct pipe_vertex_buffer vbuffer[PIPE_MAX_SHADER_INPUTS];

+   struct pipe_vertex_element velements[PIPE_MAX_ATTRIBS];

+   unsigned num_vbuffers, num_velements;

+   GLuint attr;

+   unsigned i;

+

+   /* must get these after state validation! */

+   vp = st->vp;

+   vpv = st->vp_variant;

+

+   memset(velements, 0, sizeof(struct pipe_vertex_element) * vpv->num_inputs);

+

+   /* Unreference any user vertex buffers. */

+   for (i = 0; i < st->num_user_vbs; i++) {

+      pipe_resource_reference(&st->user_vb[i], NULL);

+   }

+   st->num_user_vbs = 0;

+

+   /*

+    * Setup the vbuffer[] and velements[] arrays.

+    */

+   if (is_interleaved_arrays(vp, vpv, arrays)) {

+      setup_interleaved_attribs(ctx, vp, vpv, arrays, vbuffer, velements,

+                                max_index);

+      num_vbuffers = 1;

+      num_velements = vpv->num_inputs;

+      if (num_velements == 0)

+         num_vbuffers = 0;

+   }

+   else {

+      setup_non_interleaved_attribs(ctx, vp, vpv, arrays,

+                                    vbuffer, velements, max_index);

+      num_vbuffers = vpv->num_inputs;

+      num_velements = vpv->num_inputs;

+   }

+

+   cso_set_vertex_buffers(st->cso_context, num_vbuffers, vbuffer);

+   cso_set_vertex_elements(st->cso_context, num_velements, velements);

+

+   /* unreference buffers (frees wrapped user-space buffer objects)

+    * This is OK, because the pipe driver should reference buffers by itself

+    * in set_vertex_buffers. */

+   for (attr = 0; attr < num_vbuffers; attr++) {

+      pipe_resource_reference(&vbuffer[attr].buffer, NULL);

+      assert(!vbuffer[attr].buffer);

+   }

+}

+

 

 /**

  * This function gets plugged into the VBO module and is called when

@@ -627,90 +610,79 @@ st_draw_vbo(struct gl_context *ctx,
 {

    struct st_context *st = st_context(ctx);

    struct pipe_context *pipe = st->pipe;

-   const struct st_vertex_program *vp;

-   const struct st_vp_variant *vpv;

-   struct pipe_vertex_buffer vbuffer[PIPE_MAX_SHADER_INPUTS];

-   GLuint attr;

-   struct pipe_vertex_element velements[PIPE_MAX_ATTRIBS];

-   unsigned num_vbuffers, num_velements;

    struct pipe_index_buffer ibuffer;

-   GLboolean userSpace = GL_FALSE;

-   GLboolean vertDataEdgeFlags;

    struct pipe_draw_info info;

    unsigned i;

+   GLboolean new_array =

+         st->dirty.st && (st->dirty.mesa & (_NEW_ARRAY | _NEW_PROGRAM)) != 0;

 

    /* Mesa core state should have been validated already */

    assert(ctx->NewState == 0x0);

 

-   /* Gallium probably doesn't want this in some cases. */

-   if (!index_bounds_valid)

-      if (!vbo_all_varyings_in_vbos(arrays))

-	 vbo_get_minmax_index(ctx, prims, ib, &min_index, &max_index);

+   if (ib) {

+      /* Gallium probably doesn't want this in some cases. */

+      if (!index_bounds_valid)

+         if (!vbo_all_varyings_in_vbos(arrays))

+            vbo_get_minmax_index(ctx, prims, ib, &min_index, &max_index);

+   } else {

+      /* Get min/max index for non-indexed drawing. */

+      min_index = ~0;

+      max_index = 0;

+

+      for (i = 0; i < nr_prims; i++) {

+         min_index = MIN2(min_index, prims[i].start);

+         max_index = MAX2(max_index, prims[i].start + prims[i].count - 1);

+      }

+   }

 

-   /* sanity check for pointer arithmetic below */

-   assert(sizeof(arrays[0]->Ptr[0]) == 1);

+   /* Validate state. */

+   if (st->dirty.st) {

+      GLboolean vertDataEdgeFlags;

 

-   vertDataEdgeFlags = arrays[VERT_ATTRIB_EDGEFLAG]->BufferObj &&

-                       arrays[VERT_ATTRIB_EDGEFLAG]->BufferObj->Name;

-   if (vertDataEdgeFlags != st->vertdata_edgeflags) {

-      st->vertdata_edgeflags = vertDataEdgeFlags;

-      st->dirty.st |= ST_NEW_EDGEFLAGS_DATA;

-   }

+      /* sanity check for pointer arithmetic below */

+      assert(sizeof(arrays[0]->Ptr[0]) == 1);

 

-   st_validate_state(st);

+      vertDataEdgeFlags = arrays[VERT_ATTRIB_EDGEFLAG]->BufferObj &&

+                          arrays[VERT_ATTRIB_EDGEFLAG]->BufferObj->Name;

+      if (vertDataEdgeFlags != st->vertdata_edgeflags) {

+         st->vertdata_edgeflags = vertDataEdgeFlags;

+         st->dirty.st |= ST_NEW_EDGEFLAGS_DATA;

+      }

 

-   /* must get these after state validation! */

-   vp = st->vp;

-   vpv = st->vp_variant;

+      st_validate_state(st);

+

+      if (new_array) {

+         st_validate_varrays(ctx, arrays, max_index);

+      }

 

 #if 0

-   if (MESA_VERBOSE & VERBOSE_GLSL) {

-      check_uniforms(ctx);

-   }

+      if (MESA_VERBOSE & VERBOSE_GLSL) {

+         check_uniforms(ctx);

+      }

 #else

-   (void) check_uniforms;

+      (void) check_uniforms;

 #endif

-

-   memset(velements, 0, sizeof(struct pipe_vertex_element) * vpv->num_inputs);

-   /*

-    * Setup the vbuffer[] and velements[] arrays.

-    */

-   if (is_interleaved_arrays(vp, vpv, arrays, &userSpace)) {

-      /*printf("Draw interleaved\n");*/

-      setup_interleaved_attribs(ctx, vp, vpv, arrays, max_index, userSpace,

-                                vbuffer, velements);

-      num_vbuffers = 1;

-      num_velements = vpv->num_inputs;

-      if (num_velements == 0)

-         num_vbuffers = 0;

-   }

-   else {

-      /*printf("Draw non-interleaved\n");*/

-      setup_non_interleaved_attribs(ctx, vp, vpv, arrays, max_index,

-                                    &userSpace, vbuffer, velements);

-      num_vbuffers = vpv->num_inputs;

-      num_velements = vpv->num_inputs;

    }

 

-#if 0

-   {

-      GLuint i;

-      for (i = 0; i < num_vbuffers; i++) {

-         printf("buffers[%d].stride = %u\n", i, vbuffer[i].stride);

-         printf("buffers[%d].max_index = %u\n", i, vbuffer[i].max_index);

-         printf("buffers[%d].buffer_offset = %u\n", i, vbuffer[i].buffer_offset);

-         printf("buffers[%d].buffer = %p\n", i, (void*) vbuffer[i].buffer);

-      }

-      for (i = 0; i < num_velements; i++) {

-         printf("vlements[%d].vbuffer_index = %u\n", i, velements[i].vertex_buffer_index);

-         printf("vlements[%d].src_offset = %u\n", i, velements[i].src_offset);

-         printf("vlements[%d].format = %s\n", i, util_format_name(velements[i].src_format));

+   /* Notify the driver that the content of user buffers may have been

+    * changed. */

+   if (!new_array && st->num_user_vbs) {

+      for (i = 0; i < st->num_user_vbs; i++) {

+         if (st->user_vb[i]) {

+            unsigned stride = st->user_vb_stride[i];

+

+            if (stride) {

+               pipe->redefine_user_buffer(pipe, st->user_vb[i],

+                                          min_index * stride,

+                                          (max_index + 1 - min_index) * stride);

+            } else {

+               /* stride == 0 */

+               pipe->redefine_user_buffer(pipe, st->user_vb[i],

+                                          0, st->user_vb[i]->width0);

+            }

+         }

       }

    }

-#endif

-

-   pipe->set_vertex_buffers(pipe, num_vbuffers, vbuffer);

-   cso_set_vertex_elements(st->cso_context, num_velements, velements);

 

    setup_index_buffer(ctx, ib, &ibuffer);

    pipe->set_index_buffer(pipe, &ibuffer);

@@ -744,17 +716,6 @@ st_draw_vbo(struct gl_context *ctx,
    }

 

    pipe_resource_reference(&ibuffer.buffer, NULL);

-

-   /* unreference buffers (frees wrapped user-space buffer objects) */

-   for (attr = 0; attr < num_vbuffers; attr++) {

-      pipe_resource_reference(&vbuffer[attr].buffer, NULL);

-      assert(!vbuffer[attr].buffer);

-   }

-

-   if (userSpace) 

-   {

-      pipe->set_vertex_buffers(pipe, 0, NULL);

-   }

 }

 

 

diff --git a/mesalib/src/mesa/state_tracker/st_draw_feedback.c b/mesalib/src/mesa/state_tracker/st_draw_feedback.c
index 2ecb8e202..f0734eae0 100644
--- a/mesalib/src/mesa/state_tracker/st_draw_feedback.c
+++ b/mesalib/src/mesa/state_tracker/st_draw_feedback.c
@@ -179,7 +179,6 @@ st_feedback_draw_vbo(struct gl_context *ctx,
 

       /* common-case setup */

       vbuffers[attr].stride = arrays[mesaAttr]->StrideB; /* in bytes */

-      vbuffers[attr].max_index = max_index;

       velements[attr].instance_divisor = 0;

       velements[attr].vertex_buffer_index = attr;

       velements[attr].src_format = 

diff --git a/mesalib/src/mesa/state_tracker/st_gen_mipmap.c b/mesalib/src/mesa/state_tracker/st_gen_mipmap.c
index 0be66a2c2..5fcd806e0 100644
--- a/mesalib/src/mesa/state_tracker/st_gen_mipmap.c
+++ b/mesalib/src/mesa/state_tracker/st_gen_mipmap.c
@@ -105,13 +105,12 @@ st_render_mipmap(struct st_context *st,
 static void

 decompress_image(enum pipe_format format,

                  const uint8_t *src, uint8_t *dst,

-                 unsigned width, unsigned height)

+                 unsigned width, unsigned height, unsigned src_stride)

 {

    const struct util_format_description *desc = util_format_description(format);

    const uint bw = util_format_get_blockwidth(format);

    const uint bh = util_format_get_blockheight(format);

    const uint dst_stride = 4 * MAX2(width, bw);

-   const uint src_stride = util_format_get_stride(format, width);

 

    desc->unpack_rgba_8unorm(dst, dst_stride, src, src_stride, width, height);

 

@@ -144,10 +143,9 @@ decompress_image(enum pipe_format format,
 static void

 compress_image(enum pipe_format format,

                const uint8_t *src, uint8_t *dst,

-               unsigned width, unsigned height)

+               unsigned width, unsigned height, unsigned dst_stride)

 {

    const struct util_format_description *desc = util_format_description(format);

-   const uint dst_stride = util_format_get_stride(format, width);

    const uint src_stride = 4 * width;

 

    desc->pack_rgba_8unorm(dst, dst_stride, src, src_stride, width, height);

@@ -236,7 +234,7 @@ fallback_generate_mipmap(struct gl_context *ctx, GLenum target,
          dstTemp = malloc(dstWidth2 * dstHeight2 * comps + 000);

 

          /* decompress the src image: srcData -> srcTemp */

-         decompress_image(format, srcData, srcTemp, srcWidth, srcHeight);

+         decompress_image(format, srcData, srcTemp, srcWidth, srcHeight, srcTrans->stride);

 

          _mesa_generate_mipmap_level(target, datatype, comps,

                                      0 /*border*/,

@@ -248,7 +246,7 @@ fallback_generate_mipmap(struct gl_context *ctx, GLenum target,
                                      dstWidth2); /* stride in texels */

 

          /* compress the new image: dstTemp -> dstData */

-         compress_image(format, dstTemp, dstData, dstWidth, dstHeight);

+         compress_image(format, dstTemp, dstData, dstWidth, dstHeight, dstTrans->stride);

 

          free(srcTemp);

          free(dstTemp);

diff --git a/mesalib/src/mesa/tnl/t_draw.c b/mesalib/src/mesa/tnl/t_draw.c
index 741f0ed3f..a2ca6225b 100644
--- a/mesalib/src/mesa/tnl/t_draw.c
+++ b/mesalib/src/mesa/tnl/t_draw.c
@@ -125,6 +125,43 @@ convert_half_to_float(const struct gl_client_array *input,
    }

 }

 

+/**

+ * \brief Convert fixed-point to floating-point.

+ *

+ * In OpenGL, a fixed-point number is a "signed 2's complement 16.16 scaled

+ * integer" (Table 2.2 of the OpenGL ES 2.0 spec).

+ *

+ * If the buffer has the \c normalized flag set, the formula

+ *     \code normalize(x) := (2*x + 1) / (2^16 - 1) \endcode

+ * is used to map the fixed-point numbers into the range [-1, 1].

+ */

+static void

+convert_fixed_to_float(const struct gl_client_array *input,

+                       const GLubyte *ptr, GLfloat *fptr,

+                       GLuint count)

+{

+   GLuint i, j;

+   const GLint size = input->Size;

+

+   if (input->Normalized) {

+      for (i = 0; i < count; ++i) {

+         const GLfixed *in = (GLfixed *) ptr;

+         for (j = 0; j < size; ++j) {

+            *fptr++ = (GLfloat) (2 * in[j] + 1) / (GLfloat) ((1 << 16) - 1);

+         }

+         ptr += input->StrideB;

+      }

+   } else {

+      for (i = 0; i < count; ++i) {

+         const GLfixed *in = (GLfixed *) ptr;

+         for (j = 0; j < size; ++j) {

+            *fptr++ = in[j] / (GLfloat) (1 << 16);

+         }

+         ptr += input->StrideB;

+      }

+   }

+}

+

 /* Adjust pointer to point at first requested element, convert to

  * floating point, populate VB->AttribPtr[].

  */

@@ -174,6 +211,9 @@ static void _tnl_import_array( struct gl_context *ctx,
       case GL_HALF_FLOAT:

 	 convert_half_to_float(input, ptr, fptr, count, sz);

 	 break;

+      case GL_FIXED:

+         convert_fixed_to_float(input, ptr, fptr, count);

+         break;

       default:

 	 assert(0);

 	 break;

diff --git a/mesalib/src/mesa/vbo/vbo_exec_array.c b/mesalib/src/mesa/vbo/vbo_exec_array.c
index 13b54d59c..4304985aa 100644
--- a/mesalib/src/mesa/vbo/vbo_exec_array.c
+++ b/mesalib/src/mesa/vbo/vbo_exec_array.c
@@ -502,8 +502,13 @@ recalculate_input_bindings(struct gl_context *ctx)
 static void

 bind_arrays(struct gl_context *ctx)

 {

+   if (!ctx->Array.RebindArrays) {

+      return;

+   }

+

    bind_array_obj(ctx);

    recalculate_input_bindings(ctx);

+   ctx->Array.RebindArrays = GL_FALSE;

 }

 

 

diff --git a/mesalib/src/mesa/vbo/vbo_exec_draw.c b/mesalib/src/mesa/vbo/vbo_exec_draw.c
index 048f3d170..a2a098857 100644
--- a/mesalib/src/mesa/vbo/vbo_exec_draw.c
+++ b/mesalib/src/mesa/vbo/vbo_exec_draw.c
@@ -245,6 +245,7 @@ vbo_exec_bind_arrays( struct gl_context *ctx )
 	 arrays[attr]._MaxElement = count; /* ??? */

 

          varying_inputs |= 1 << attr;

+         ctx->NewState |= _NEW_ARRAY;

       }

    }

 

diff --git a/mesalib/src/mesa/vbo/vbo_save_draw.c b/mesalib/src/mesa/vbo/vbo_save_draw.c
index 6d8dbdb86..5a19b0d62 100644
--- a/mesalib/src/mesa/vbo/vbo_save_draw.c
+++ b/mesalib/src/mesa/vbo/vbo_save_draw.c
@@ -202,6 +202,7 @@ static void vbo_bind_vertex_list(struct gl_context *ctx,
 

 	 buffer_offset += node->attrsz[src] * sizeof(GLfloat);

          varying_inputs |= 1<<attr;

+         ctx->NewState |= _NEW_ARRAY;

       }

    }