X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fstate_tracker%2Fst_glsl_to_tgsi.cpp;h=6cc655d70cf5f95b79f9ff36df30aef204e7b8e8;hb=20b0daf82de91fd57b7e8d825786789149f6358d;hp=abeb44a408380a66c891085b1e1802143e65c86a;hpb=8c50f18b29637470539d05ccc32b0cae0092aeac;p=mesa.git

diff --git a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
index abeb44a4083..6cc655d70cf 100644
--- a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
+++ b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
@@ -42,17 +42,17 @@
 #include "ir_optimization.h"
 #include "ast.h"
 
-extern "C" {
 #include "main/mtypes.h"
-#include "main/shaderapi.h"
 #include "main/shaderobj.h"
-#include "main/uniforms.h"
 #include "program/hash_table.h"
+
+extern "C" {
+#include "main/shaderapi.h"
+#include "main/uniforms.h"
 #include "program/prog_instruction.h"
 #include "program/prog_optimize.h"
 #include "program/prog_print.h"
 #include "program/program.h"
-#include "program/prog_uniform.h"
 #include "program/prog_parameter.h"
 #include "program/sampler.h"
 
@@ -70,6 +70,7 @@ extern "C" {
 #include "st_mesa_to_tgsi.h"
 }
 
+#define PROGRAM_IMMEDIATE PROGRAM_FILE_MAX
 #define PROGRAM_ANY_CONST ((1 << PROGRAM_LOCAL_PARAM) |  \
                            (1 << PROGRAM_ENV_PARAM) |    \
                            (1 << PROGRAM_STATE_VAR) |    \
@@ -77,6 +78,17 @@ extern "C" {
                            (1 << PROGRAM_CONSTANT) |     \
                            (1 << PROGRAM_UNIFORM))
 
+/**
+ * Maximum number of temporary registers.
+ *
+ * It is too big for stack allocated arrays -- it will cause stack overflow on
+ * Windows and likely Mac OS X.
+ */
+#define MAX_TEMPS         4096
+
+/* will be 4 for GLSL 4.00 */
+#define MAX_GLSL_TEXTURE_OFFSET 1
+
 class st_src_reg;
 class st_dst_reg;
 
@@ -171,7 +183,7 @@ st_src_reg::st_src_reg(st_dst_reg reg)
    this->index = reg.index;
    this->swizzle = SWIZZLE_XYZW;
    this->negate = 0;
-   this->reladdr = NULL;
+   this->reladdr = reg.reladdr;
 }
 
 st_dst_reg::st_dst_reg(st_src_reg reg)
@@ -208,6 +220,9 @@ public:
    int sampler; /**< sampler index */
    int tex_target; /**< One of TEXTURE_*_INDEX */
    GLboolean tex_shadow;
+   struct tgsi_texture_offset tex_offsets[MAX_GLSL_TEXTURE_OFFSET];
+   unsigned tex_offset_num_offset;
+   int dead_mask; /**< Used in dead code elimination */
 
    class function_entry *function; /* Set on TGSI_OPCODE_CAL or TGSI_OPCODE_BGNSUB */
 };
@@ -225,6 +240,20 @@ public:
    ir_variable *var; /* variable that maps to this, if any */
 };
 
+class immediate_storage : public exec_node {
+public:
+   immediate_storage(gl_constant_value *values, int size, int type)
+   {
+      memcpy(this->values, values, size * sizeof(gl_constant_value));
+      this->size = size;
+      this->type = type;
+   }
+   
+   gl_constant_value values[4];
+   int size; /**< Number of components (1-4) */
+   int type; /**< GL_FLOAT, GL_INT, GL_BOOL, or GL_UNSIGNED_INT */
+};
+
 class function_entry : public exec_node {
 public:
    ir_function_signature *sig;
@@ -232,7 +261,7 @@ public:
    /**
     * identifier of this function signature used by the program.
     *
-    * At the point that Mesa instructions for function calls are
+    * At the point that TGSI instructions for function calls are
     * generated, we don't know the address of the first instruction of
     * the function body.  So we make the BranchTarget that is called a
     * small integer and rewrite them during set_branchtargets().
@@ -247,10 +276,9 @@ public:
    glsl_to_tgsi_instruction *bgn_inst;
 
    /**
-    * Index of the first instruction of the function body in actual
-    * Mesa IR.
+    * Index of the first instruction of the function body in actual TGSI.
     *
-    * Set after convertion from glsl_to_tgsi_instruction to prog_instruction.
+    * Set after conversion from glsl_to_tgsi_instruction to TGSI.
     */
    int inst;
 
@@ -278,9 +306,13 @@ public:
    bool indirect_addr_consts;
    
    int glsl_version;
+   bool native_integers;
 
    variable_storage *find_variable_storage(ir_variable *var);
 
+   int add_constant(gl_register_file file, gl_constant_value values[4],
+                    int size, int datatype, GLuint *swizzle_out);
+
    function_entry *get_function_signature(ir_function_signature *sig);
 
    st_src_reg get_temp(const glsl_type *type);
@@ -322,6 +354,10 @@ public:
    /** List of variable_storage */
    exec_list variables;
 
+   /** List of immediate_storage */
+   exec_list immediates;
+   int num_immediates;
+
    /** List of function_entry */
    exec_list function_signatures;
    int next_signature_id;
@@ -348,11 +384,11 @@ public:
    /**
     * Emit the correct dot-product instruction for the type of arguments
     */
-   void emit_dp(ir_instruction *ir,
-                st_dst_reg dst,
-                st_src_reg src0,
-                st_src_reg src1,
-                unsigned elements);
+   glsl_to_tgsi_instruction *emit_dp(ir_instruction *ir,
+                                     st_dst_reg dst,
+                                     st_src_reg src0,
+                                     st_src_reg src1,
+                                     unsigned elements);
 
    void emit_scalar(ir_instruction *ir, unsigned op,
         	    st_dst_reg dst, st_src_reg src0);
@@ -360,14 +396,18 @@ public:
    void emit_scalar(ir_instruction *ir, unsigned op,
         	    st_dst_reg dst, st_src_reg src0, st_src_reg src1);
 
+   void try_emit_float_set(ir_instruction *ir, unsigned op, st_dst_reg dst);
+
    void emit_arl(ir_instruction *ir, st_dst_reg dst, st_src_reg src0);
 
    void emit_scs(ir_instruction *ir, unsigned op,
         	 st_dst_reg dst, const st_src_reg &src);
 
-   GLboolean try_emit_mad(ir_expression *ir,
-        		  int mul_operand);
-   GLboolean try_emit_sat(ir_expression *ir);
+   bool try_emit_mad(ir_expression *ir,
+              int mul_operand);
+   bool try_emit_mad_for_and_not(ir_expression *ir,
+              int mul_operand);
+   bool try_emit_sat(ir_expression *ir);
 
    void emit_swz(ir_expression *ir);
 
@@ -384,6 +424,7 @@ public:
 
    void copy_propagate(void);
    void eliminate_dead_code(void);
+   int eliminate_dead_code_advanced(void);
    void merge_registers(void);
    void renumber_registers(void);
 
@@ -480,10 +521,11 @@ glsl_to_tgsi_visitor::emit(ir_instruction *ir, unsigned op,
    inst->src[1] = src1;
    inst->src[2] = src2;
    inst->ir = ir;
+   inst->dead_mask = 0;
 
    inst->function = NULL;
    
-   if (op == TGSI_OPCODE_ARL)
+   if (op == TGSI_OPCODE_ARL || op == TGSI_OPCODE_UARL)
       this->num_address_regs = 1;
    
    /* Update indirect addressing status used by TGSI */
@@ -500,6 +542,9 @@ glsl_to_tgsi_visitor::emit(ir_instruction *ir, unsigned op,
       case PROGRAM_UNIFORM:
          this->indirect_addr_consts = true;
          break;
+      case PROGRAM_IMMEDIATE:
+         assert(!"immediates should not have indirect addressing");
+         break;
       default:
          break;
       }
@@ -519,6 +564,9 @@ glsl_to_tgsi_visitor::emit(ir_instruction *ir, unsigned op,
             case PROGRAM_UNIFORM:
                this->indirect_addr_consts = true;
                break;
+            case PROGRAM_IMMEDIATE:
+               assert(!"immediates should not have indirect addressing");
+               break;
             default:
                break;
             }
@@ -527,7 +575,10 @@ glsl_to_tgsi_visitor::emit(ir_instruction *ir, unsigned op,
    }
 
    this->instructions.push_tail(inst);
-   
+
+   if (native_integers)
+      try_emit_float_set(ir, op, dst);
+
    return inst;
 }
 
@@ -553,11 +604,28 @@ glsl_to_tgsi_visitor::emit(ir_instruction *ir, unsigned op)
    return emit(ir, op, undef_dst, undef_src, undef_src, undef_src);
 }
 
+ /**
+ * Emits the code to convert the result of float SET instructions to integers.
+ */
+void
+glsl_to_tgsi_visitor::try_emit_float_set(ir_instruction *ir, unsigned op,
+        		 st_dst_reg dst)
+{
+   if ((op == TGSI_OPCODE_SEQ ||
+        op == TGSI_OPCODE_SNE ||
+        op == TGSI_OPCODE_SGE ||
+        op == TGSI_OPCODE_SLT))
+   {
+      st_src_reg src = st_src_reg(dst);
+      src.negate = ~src.negate;
+      dst.type = GLSL_TYPE_FLOAT;
+      emit(ir, TGSI_OPCODE_F2I, dst, src);
+   }
+}
+
 /**
  * Determines whether to use an integer, unsigned integer, or float opcode 
  * based on the operands and input opcode, then emits the result.
- * 
- * TODO: type checking for remaining TGSI opcodes
  */
 unsigned
 glsl_to_tgsi_visitor::get_opcode(ir_instruction *ir, unsigned op,
@@ -568,8 +636,8 @@ glsl_to_tgsi_visitor::get_opcode(ir_instruction *ir, unsigned op,
    
    if (src0.type == GLSL_TYPE_FLOAT || src1.type == GLSL_TYPE_FLOAT)
       type = GLSL_TYPE_FLOAT;
-   else if (glsl_version >= 130)
-      type = src0.type;
+   else if (native_integers)
+      type = src0.type == GLSL_TYPE_BOOL ? GLSL_TYPE_INT : src0.type;
 
 #define case4(c, f, i, u) \
    case TGSI_OPCODE_##c: \
@@ -595,12 +663,7 @@ glsl_to_tgsi_visitor::get_opcode(ir_instruction *ir, unsigned op,
       case3(SGE, ISGE, USGE);
       case3(SLT, ISLT, USLT);
       
-      case2iu(SHL, SHL);
       case2iu(ISHR, USHR);
-      case2iu(NOT, NOT);
-      case2iu(AND, AND);
-      case2iu(OR, OR);
-      case2iu(XOR, XOR);
       
       default: break;
    }
@@ -609,7 +672,7 @@ glsl_to_tgsi_visitor::get_opcode(ir_instruction *ir, unsigned op,
    return op;
 }
 
-void
+glsl_to_tgsi_instruction *
 glsl_to_tgsi_visitor::emit_dp(ir_instruction *ir,
         		    st_dst_reg dst, st_src_reg src0, st_src_reg src1,
         		    unsigned elements)
@@ -618,7 +681,7 @@ glsl_to_tgsi_visitor::emit_dp(ir_instruction *ir,
       TGSI_OPCODE_DP2, TGSI_OPCODE_DP3, TGSI_OPCODE_DP4
    };
 
-   emit(ir, dot_opcodes[elements - 2], dst, src0, src1);
+   return emit(ir, dot_opcodes[elements - 2], dst, src0, src1);
 }
 
 /**
@@ -689,16 +752,12 @@ void
 glsl_to_tgsi_visitor::emit_arl(ir_instruction *ir,
         		        st_dst_reg dst, st_src_reg src0)
 {
-   st_src_reg tmp = get_temp(glsl_type::float_type);
+   int op = TGSI_OPCODE_ARL;
 
-   if (src0.type == GLSL_TYPE_INT)
-      emit(ir, TGSI_OPCODE_I2F, st_dst_reg(tmp), src0);
-   else if (src0.type == GLSL_TYPE_UINT)
-      emit(ir, TGSI_OPCODE_U2F, st_dst_reg(tmp), src0);
-   else
-      tmp = src0;
-   
-   emit(ir, TGSI_OPCODE_ARL, dst, tmp);
+   if (src0.type == GLSL_TYPE_INT || src0.type == GLSL_TYPE_UINT)
+      op = TGSI_OPCODE_UARL;
+
+   emit(NULL, op, dst, src0);
 }
 
 /**
@@ -796,38 +855,71 @@ glsl_to_tgsi_visitor::emit_scs(ir_instruction *ir, unsigned op,
    }
 }
 
-struct st_src_reg
+int
+glsl_to_tgsi_visitor::add_constant(gl_register_file file,
+        		     gl_constant_value values[4], int size, int datatype,
+        		     GLuint *swizzle_out)
+{
+   if (file == PROGRAM_CONSTANT) {
+      return _mesa_add_typed_unnamed_constant(this->prog->Parameters, values,
+                                              size, datatype, swizzle_out);
+   } else {
+      int index = 0;
+      immediate_storage *entry;
+      assert(file == PROGRAM_IMMEDIATE);
+
+      /* Search immediate storage to see if we already have an identical
+       * immediate that we can use instead of adding a duplicate entry.
+       */
+      foreach_iter(exec_list_iterator, iter, this->immediates) {
+         entry = (immediate_storage *)iter.get();
+         
+         if (entry->size == size &&
+             entry->type == datatype &&
+             !memcmp(entry->values, values, size * sizeof(gl_constant_value))) {
+             return index;
+         }
+         index++;
+      }
+      
+      /* Add this immediate to the list. */
+      entry = new(mem_ctx) immediate_storage(values, size, datatype);
+      this->immediates.push_tail(entry);
+      this->num_immediates++;
+      return index;
+   }
+}
+
+st_src_reg
 glsl_to_tgsi_visitor::st_src_reg_for_float(float val)
 {
-   st_src_reg src(PROGRAM_CONSTANT, -1, GLSL_TYPE_FLOAT);
+   st_src_reg src(PROGRAM_IMMEDIATE, -1, GLSL_TYPE_FLOAT);
    union gl_constant_value uval;
 
    uval.f = val;
-   src.index = _mesa_add_typed_unnamed_constant(this->prog->Parameters,
-        				  &uval, 1, GL_FLOAT, &src.swizzle);
+   src.index = add_constant(src.file, &uval, 1, GL_FLOAT, &src.swizzle);
 
    return src;
 }
 
-struct st_src_reg
+st_src_reg
 glsl_to_tgsi_visitor::st_src_reg_for_int(int val)
 {
-   st_src_reg src(PROGRAM_CONSTANT, -1, GLSL_TYPE_INT);
+   st_src_reg src(PROGRAM_IMMEDIATE, -1, GLSL_TYPE_INT);
    union gl_constant_value uval;
    
-   assert(glsl_version >= 130);
+   assert(native_integers);
 
    uval.i = val;
-   src.index = _mesa_add_typed_unnamed_constant(this->prog->Parameters,
-        				  &uval, 1, GL_INT, &src.swizzle);
+   src.index = add_constant(src.file, &uval, 1, GL_INT, &src.swizzle);
 
    return src;
 }
 
-struct st_src_reg
+st_src_reg
 glsl_to_tgsi_visitor::st_src_reg_for_type(int type, int val)
 {
-   if (glsl_version >= 130)
+   if (native_integers)
       return type == GLSL_TYPE_FLOAT ? st_src_reg_for_float(val) : 
                                        st_src_reg_for_int(val);
    else
@@ -884,10 +976,8 @@ st_src_reg
 glsl_to_tgsi_visitor::get_temp(const glsl_type *type)
 {
    st_src_reg src;
-   int swizzle[4];
-   int i;
 
-   src.type = glsl_version >= 130 ? type->base_type : GLSL_TYPE_FLOAT;
+   src.type = native_integers ? type->base_type : GLSL_TYPE_FLOAT;
    src.file = PROGRAM_TEMPORARY;
    src.index = next_temp;
    src.reladdr = NULL;
@@ -896,12 +986,7 @@ glsl_to_tgsi_visitor::get_temp(const glsl_type *type)
    if (type->is_array() || type->is_record()) {
       src.swizzle = SWIZZLE_NOOP;
    } else {
-      for (i = 0; i < type->vector_elements; i++)
-         swizzle[i] = i;
-      for (; i < 4; i++)
-         swizzle[i] = type->vector_elements - 1;
-      src.swizzle = MAKE_SWIZZLE4(swizzle[0], swizzle[1],
-        			  swizzle[2], swizzle[3]);
+      src.swizzle = swizzle_for_size(type->vector_elements);
    }
    src.negate = 0;
 
@@ -932,29 +1017,6 @@ glsl_to_tgsi_visitor::visit(ir_variable *ir)
 
       fp->OriginUpperLeft = ir->origin_upper_left;
       fp->PixelCenterInteger = ir->pixel_center_integer;
-
-   } else if (strcmp(ir->name, "gl_FragDepth") == 0) {
-      struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog;
-      switch (ir->depth_layout) {
-      case ir_depth_layout_none:
-         fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_NONE;
-         break;
-      case ir_depth_layout_any:
-         fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_ANY;
-         break;
-      case ir_depth_layout_greater:
-         fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_GREATER;
-         break;
-      case ir_depth_layout_less:
-         fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_LESS;
-         break;
-      case ir_depth_layout_unchanged:
-         fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_UNCHANGED;
-         break;
-      default:
-         assert(0);
-         break;
-      }
    }
 
    if (ir->mode == ir_var_uniform && strncmp(ir->name, "gl_", 3) == 0) {
@@ -974,7 +1036,7 @@ glsl_to_tgsi_visitor::visit(ir_variable *ir)
          }
       }
 
-      struct variable_storage *storage;
+      variable_storage *storage;
       st_dst_reg dst;
       if (i == ir->num_state_slots) {
          /* We'll set the index later. */
@@ -995,7 +1057,7 @@ glsl_to_tgsi_visitor::visit(ir_variable *ir)
          this->next_temp += type_size(ir->type);
 
          dst = st_dst_reg(st_src_reg(PROGRAM_TEMPORARY, storage->index,
-               glsl_version >= 130 ? ir->type->base_type : GLSL_TYPE_FLOAT));
+               native_integers ? ir->type->base_type : GLSL_TYPE_FLOAT));
       }
 
 
@@ -1011,7 +1073,7 @@ glsl_to_tgsi_visitor::visit(ir_variable *ir)
             }
          } else {
             st_src_reg src(PROGRAM_STATE_VAR, index,
-                  glsl_version >= 130 ? ir->type->base_type : GLSL_TYPE_FLOAT);
+                  native_integers ? ir->type->base_type : GLSL_TYPE_FLOAT);
             src.swizzle = slots[i].swizzle;
             emit(ir, TGSI_OPCODE_MOV, dst, src);
             /* even a float takes up a whole vec4 reg in a struct/array. */
@@ -1125,11 +1187,12 @@ glsl_to_tgsi_visitor::visit(ir_function *ir)
    }
 }
 
-GLboolean
+bool
 glsl_to_tgsi_visitor::try_emit_mad(ir_expression *ir, int mul_operand)
 {
    int nonmul_operand = 1 - mul_operand;
    st_src_reg a, b, c;
+   st_dst_reg result_dst;
 
    ir_expression *expr = ir->operands[mul_operand]->as_expression();
    if (!expr || expr->operation != ir_binop_mul)
@@ -1143,12 +1206,54 @@ glsl_to_tgsi_visitor::try_emit_mad(ir_expression *ir, int mul_operand)
    c = this->result;
 
    this->result = get_temp(ir->type);
-   emit(ir, TGSI_OPCODE_MAD, st_dst_reg(this->result), a, b, c);
+   result_dst = st_dst_reg(this->result);
+   result_dst.writemask = (1 << ir->type->vector_elements) - 1;
+   emit(ir, TGSI_OPCODE_MAD, result_dst, a, b, c);
 
    return true;
 }
 
-GLboolean
+/**
+ * Emit MAD(a, -b, a) instead of AND(a, NOT(b))
+ *
+ * The logic values are 1.0 for true and 0.0 for false.  Logical-and is
+ * implemented using multiplication, and logical-or is implemented using
+ * addition.  Logical-not can be implemented as (true - x), or (1.0 - x).
+ * As result, the logical expression (a & !b) can be rewritten as:
+ *
+ *     - a * !b
+ *     - a * (1 - b)
+ *     - (a * 1) - (a * b)
+ *     - a + -(a * b)
+ *     - a + (a * -b)
+ *
+ * This final expression can be implemented as a single MAD(a, -b, a)
+ * instruction.
+ */
+bool
+glsl_to_tgsi_visitor::try_emit_mad_for_and_not(ir_expression *ir, int try_operand)
+{
+   const int other_operand = 1 - try_operand;
+   st_src_reg a, b;
+
+   ir_expression *expr = ir->operands[try_operand]->as_expression();
+   if (!expr || expr->operation != ir_unop_logic_not)
+      return false;
+
+   ir->operands[other_operand]->accept(this);
+   a = this->result;
+   expr->operands[0]->accept(this);
+   b = this->result;
+
+   b.negate = ~b.negate;
+
+   this->result = get_temp(ir->type);
+   emit(ir, TGSI_OPCODE_MAD, st_dst_reg(this->result), a, b, a);
+
+   return true;
+}
+
+bool
 glsl_to_tgsi_visitor::try_emit_sat(ir_expression *ir)
 {
    /* Saturates were only introduced to vertex programs in
@@ -1164,10 +1269,32 @@ glsl_to_tgsi_visitor::try_emit_sat(ir_expression *ir)
    sat_src->accept(this);
    st_src_reg src = this->result;
 
-   this->result = get_temp(ir->type);
-   glsl_to_tgsi_instruction *inst;
-   inst = emit(ir, TGSI_OPCODE_MOV, st_dst_reg(this->result), src);
-   inst->saturate = true;
+   /* If we generated an expression instruction into a temporary in
+    * processing the saturate's operand, apply the saturate to that
+    * instruction.  Otherwise, generate a MOV to do the saturate.
+    *
+    * Note that we have to be careful to only do this optimization if
+    * the instruction in question was what generated src->result.  For
+    * example, ir_dereference_array might generate a MUL instruction
+    * to create the reladdr, and return us a src reg using that
+    * reladdr.  That MUL result is not the value we're trying to
+    * saturate.
+    */
+   ir_expression *sat_src_expr = sat_src->as_expression();
+   if (sat_src_expr && (sat_src_expr->operation == ir_binop_mul ||
+			sat_src_expr->operation == ir_binop_add ||
+			sat_src_expr->operation == ir_binop_dot)) {
+      glsl_to_tgsi_instruction *new_inst;
+      new_inst = (glsl_to_tgsi_instruction *)this->instructions.get_tail();
+      new_inst->saturate = true;
+   } else {
+      this->result = get_temp(ir->type);
+      st_dst_reg result_dst = st_dst_reg(this->result);
+      result_dst.writemask = (1 << ir->type->vector_elements) - 1;
+      glsl_to_tgsi_instruction *inst;
+      inst = emit(ir, TGSI_OPCODE_MOV, result_dst, src);
+      inst->saturate = true;
+   }
 
    return true;
 }
@@ -1207,6 +1334,16 @@ glsl_to_tgsi_visitor::visit(ir_expression *ir)
       if (try_emit_mad(ir, 0))
          return;
    }
+
+   /* Quick peephole: Emit OPCODE_MAD(-a, -b, a) instead of AND(a, NOT(b))
+    */
+   if (ir->operation == ir_binop_logic_and) {
+      if (try_emit_mad_for_and_not(ir, 1))
+	 return;
+      if (try_emit_mad_for_and_not(ir, 0))
+	 return;
+   }
+
    if (try_emit_sat(ir))
       return;
 
@@ -1252,7 +1389,17 @@ glsl_to_tgsi_visitor::visit(ir_expression *ir)
 
    switch (ir->operation) {
    case ir_unop_logic_not:
-      emit(ir, TGSI_OPCODE_SEQ, result_dst, op[0], st_src_reg_for_type(result_dst.type, 0));
+      if (result_dst.type != GLSL_TYPE_FLOAT)
+         emit(ir, TGSI_OPCODE_NOT, result_dst, op[0]);
+      else {
+         /* Previously 'SEQ dst, src, 0.0' was used for this.  However, many
+          * older GPUs implement SEQ using multiple instructions (i915 uses two
+          * SGE instructions and a MUL instruction).  Since our logic values are
+          * 0.0 and 1.0, 1-x also implements !x.
+          */
+         op[0].negate = ~op[0].negate;
+         emit(ir, TGSI_OPCODE_ADD, result_dst, op[0], st_src_reg_for_float(1.0));
+      }
       break;
    case ir_unop_neg:
       assert(result_dst.type == GLSL_TYPE_FLOAT || result_dst.type == GLSL_TYPE_INT);
@@ -1343,10 +1490,10 @@ glsl_to_tgsi_visitor::visit(ir_expression *ir)
       emit(ir, TGSI_OPCODE_SLT, result_dst, op[0], op[1]);
       break;
    case ir_binop_greater:
-      emit(ir, TGSI_OPCODE_SGT, result_dst, op[0], op[1]);
+      emit(ir, TGSI_OPCODE_SLT, result_dst, op[1], op[0]);
       break;
    case ir_binop_lequal:
-      emit(ir, TGSI_OPCODE_SLE, result_dst, op[0], op[1]);
+      emit(ir, TGSI_OPCODE_SGE, result_dst, op[1], op[0]);
       break;
    case ir_binop_gequal:
       emit(ir, TGSI_OPCODE_SGE, result_dst, op[0], op[1]);
@@ -1361,13 +1508,56 @@ glsl_to_tgsi_visitor::visit(ir_expression *ir)
       /* "==" operator producing a scalar boolean. */
       if (ir->operands[0]->type->is_vector() ||
           ir->operands[1]->type->is_vector()) {
-         st_src_reg temp = get_temp(glsl_version >= 130 ? 
+         st_src_reg temp = get_temp(native_integers ?
                glsl_type::get_instance(ir->operands[0]->type->base_type, 4, 1) :
                glsl_type::vec4_type);
-         assert(ir->operands[0]->type->base_type == GLSL_TYPE_FLOAT);
-         emit(ir, TGSI_OPCODE_SNE, st_dst_reg(temp), op[0], op[1]);
-         emit_dp(ir, result_dst, temp, temp, vector_elements);
-         emit(ir, TGSI_OPCODE_SEQ, result_dst, result_src, st_src_reg_for_float(0.0));
+         
+         if (native_integers) {
+            st_dst_reg temp_dst = st_dst_reg(temp);
+            st_src_reg temp1 = st_src_reg(temp), temp2 = st_src_reg(temp);
+            
+            emit(ir, TGSI_OPCODE_SEQ, st_dst_reg(temp), op[0], op[1]);
+            
+            /* Emit 1-3 AND operations to combine the SEQ results. */
+            switch (ir->operands[0]->type->vector_elements) {
+            case 2:
+               break;
+            case 3:
+               temp_dst.writemask = WRITEMASK_Y;
+               temp1.swizzle = SWIZZLE_YYYY;
+               temp2.swizzle = SWIZZLE_ZZZZ;
+               emit(ir, TGSI_OPCODE_AND, temp_dst, temp1, temp2);
+               break;
+            case 4:
+               temp_dst.writemask = WRITEMASK_X;
+               temp1.swizzle = SWIZZLE_XXXX;
+               temp2.swizzle = SWIZZLE_YYYY;
+               emit(ir, TGSI_OPCODE_AND, temp_dst, temp1, temp2);
+               temp_dst.writemask = WRITEMASK_Y;
+               temp1.swizzle = SWIZZLE_ZZZZ;
+               temp2.swizzle = SWIZZLE_WWWW;
+               emit(ir, TGSI_OPCODE_AND, temp_dst, temp1, temp2);
+            }
+            
+            temp1.swizzle = SWIZZLE_XXXX;
+            temp2.swizzle = SWIZZLE_YYYY;
+            emit(ir, TGSI_OPCODE_AND, result_dst, temp1, temp2);
+         } else {
+            emit(ir, TGSI_OPCODE_SNE, st_dst_reg(temp), op[0], op[1]);
+            
+            /* After the dot-product, the value will be an integer on the
+             * range [0,4].  Zero becomes 1.0, and positive values become zero.
+             */
+            emit_dp(ir, result_dst, temp, temp, vector_elements);
+
+            /* Negating the result of the dot-product gives values on the range
+             * [-4, 0].  Zero becomes 1.0, and negative values become zero.
+             * This is achieved using SGE.
+             */
+            st_src_reg sge_src = result_src;
+            sge_src.negate = ~sge_src.negate;
+            emit(ir, TGSI_OPCODE_SGE, result_dst, sge_src, st_src_reg_for_float(0.0));
+         }
       } else {
          emit(ir, TGSI_OPCODE_SEQ, result_dst, op[0], op[1]);
       }
@@ -1376,38 +1566,143 @@ glsl_to_tgsi_visitor::visit(ir_expression *ir)
       /* "!=" operator producing a scalar boolean. */
       if (ir->operands[0]->type->is_vector() ||
           ir->operands[1]->type->is_vector()) {
-         st_src_reg temp = get_temp(glsl_version >= 130 ? 
+         st_src_reg temp = get_temp(native_integers ?
                glsl_type::get_instance(ir->operands[0]->type->base_type, 4, 1) :
                glsl_type::vec4_type);
-         assert(ir->operands[0]->type->base_type == GLSL_TYPE_FLOAT);
          emit(ir, TGSI_OPCODE_SNE, st_dst_reg(temp), op[0], op[1]);
-         emit_dp(ir, result_dst, temp, temp, vector_elements);
-         emit(ir, TGSI_OPCODE_SNE, result_dst, result_src, st_src_reg_for_float(0.0));
+
+         if (native_integers) {
+            st_dst_reg temp_dst = st_dst_reg(temp);
+            st_src_reg temp1 = st_src_reg(temp), temp2 = st_src_reg(temp);
+            
+            /* Emit 1-3 OR operations to combine the SNE results. */
+            switch (ir->operands[0]->type->vector_elements) {
+            case 2:
+               break;
+            case 3:
+               temp_dst.writemask = WRITEMASK_Y;
+               temp1.swizzle = SWIZZLE_YYYY;
+               temp2.swizzle = SWIZZLE_ZZZZ;
+               emit(ir, TGSI_OPCODE_OR, temp_dst, temp1, temp2);
+               break;
+            case 4:
+               temp_dst.writemask = WRITEMASK_X;
+               temp1.swizzle = SWIZZLE_XXXX;
+               temp2.swizzle = SWIZZLE_YYYY;
+               emit(ir, TGSI_OPCODE_OR, temp_dst, temp1, temp2);
+               temp_dst.writemask = WRITEMASK_Y;
+               temp1.swizzle = SWIZZLE_ZZZZ;
+               temp2.swizzle = SWIZZLE_WWWW;
+               emit(ir, TGSI_OPCODE_OR, temp_dst, temp1, temp2);
+            }
+            
+            temp1.swizzle = SWIZZLE_XXXX;
+            temp2.swizzle = SWIZZLE_YYYY;
+            emit(ir, TGSI_OPCODE_OR, result_dst, temp1, temp2);
+         } else {
+            /* After the dot-product, the value will be an integer on the
+             * range [0,4].  Zero stays zero, and positive values become 1.0.
+             */
+            glsl_to_tgsi_instruction *const dp =
+                  emit_dp(ir, result_dst, temp, temp, vector_elements);
+            if (this->prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
+               /* The clamping to [0,1] can be done for free in the fragment
+                * shader with a saturate.
+                */
+               dp->saturate = true;
+            } else {
+               /* Negating the result of the dot-product gives values on the range
+                * [-4, 0].  Zero stays zero, and negative values become 1.0.  This
+                * achieved using SLT.
+                */
+               st_src_reg slt_src = result_src;
+               slt_src.negate = ~slt_src.negate;
+               emit(ir, TGSI_OPCODE_SLT, result_dst, slt_src, st_src_reg_for_float(0.0));
+            }
+         }
       } else {
          emit(ir, TGSI_OPCODE_SNE, result_dst, op[0], op[1]);
       }
       break;
 
-   case ir_unop_any:
+   case ir_unop_any: {
       assert(ir->operands[0]->type->is_vector());
-      emit_dp(ir, result_dst, op[0], op[0],
-              ir->operands[0]->type->vector_elements);
-      emit(ir, TGSI_OPCODE_SNE, result_dst, result_src, st_src_reg_for_float(0.0));
+
+      /* After the dot-product, the value will be an integer on the
+       * range [0,4].  Zero stays zero, and positive values become 1.0.
+       */
+      glsl_to_tgsi_instruction *const dp =
+         emit_dp(ir, result_dst, op[0], op[0],
+                 ir->operands[0]->type->vector_elements);
+      if (this->prog->Target == GL_FRAGMENT_PROGRAM_ARB &&
+          result_dst.type == GLSL_TYPE_FLOAT) {
+	      /* The clamping to [0,1] can be done for free in the fragment
+	       * shader with a saturate.
+	       */
+	      dp->saturate = true;
+      } else if (result_dst.type == GLSL_TYPE_FLOAT) {
+	      /* Negating the result of the dot-product gives values on the range
+	       * [-4, 0].  Zero stays zero, and negative values become 1.0.  This
+	       * is achieved using SLT.
+	       */
+	      st_src_reg slt_src = result_src;
+	      slt_src.negate = ~slt_src.negate;
+	      emit(ir, TGSI_OPCODE_SLT, result_dst, slt_src, st_src_reg_for_float(0.0));
+      }
+      else {
+         /* Use SNE 0 if integers are being used as boolean values. */
+         emit(ir, TGSI_OPCODE_SNE, result_dst, result_src, st_src_reg_for_int(0));
+      }
       break;
+   }
 
    case ir_binop_logic_xor:
-      emit(ir, TGSI_OPCODE_SNE, result_dst, op[0], op[1]);
+      if (native_integers)
+         emit(ir, TGSI_OPCODE_XOR, result_dst, op[0], op[1]);
+      else
+         emit(ir, TGSI_OPCODE_SNE, result_dst, op[0], op[1]);
       break;
 
-   case ir_binop_logic_or:
-      /* This could be a saturated add and skip the SNE. */
-      emit(ir, TGSI_OPCODE_ADD, result_dst, op[0], op[1]);
-      emit(ir, TGSI_OPCODE_SNE, result_dst, result_src, st_src_reg_for_float(0.0));
+   case ir_binop_logic_or: {
+      if (native_integers) {
+         /* If integers are used as booleans, we can use an actual "or" 
+          * instruction.
+          */
+         assert(native_integers);
+         emit(ir, TGSI_OPCODE_OR, result_dst, op[0], op[1]);
+      } else {
+         /* After the addition, the value will be an integer on the
+          * range [0,2].  Zero stays zero, and positive values become 1.0.
+          */
+         glsl_to_tgsi_instruction *add =
+            emit(ir, TGSI_OPCODE_ADD, result_dst, op[0], op[1]);
+         if (this->prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
+            /* The clamping to [0,1] can be done for free in the fragment
+             * shader with a saturate if floats are being used as boolean values.
+             */
+            add->saturate = true;
+         } else {
+            /* Negating the result of the addition gives values on the range
+             * [-2, 0].  Zero stays zero, and negative values become 1.0.  This
+             * is achieved using SLT.
+             */
+            st_src_reg slt_src = result_src;
+            slt_src.negate = ~slt_src.negate;
+            emit(ir, TGSI_OPCODE_SLT, result_dst, slt_src, st_src_reg_for_float(0.0));
+         }
+      }
       break;
+   }
 
    case ir_binop_logic_and:
-      /* the bool args are stored as float 0.0 or 1.0, so "mul" gives us "and". */
-      emit(ir, TGSI_OPCODE_MUL, result_dst, op[0], op[1]);
+      /* If native integers are disabled, the bool args are stored as float 0.0
+       * or 1.0, so "mul" gives us "and".  If they're enabled, just use the
+       * actual AND opcode.
+       */
+      if (native_integers)
+         emit(ir, TGSI_OPCODE_AND, result_dst, op[0], op[1]);
+      else
+         emit(ir, TGSI_OPCODE_MUL, result_dst, op[0], op[1]);
       break;
 
    case ir_binop_dot:
@@ -1430,25 +1725,50 @@ glsl_to_tgsi_visitor::visit(ir_expression *ir)
       emit_scalar(ir, TGSI_OPCODE_RSQ, result_dst, op[0]);
       break;
    case ir_unop_i2f:
-   case ir_unop_b2f:
-      if (glsl_version >= 130) {
+      if (native_integers) {
          emit(ir, TGSI_OPCODE_I2F, result_dst, op[0]);
          break;
       }
-   case ir_unop_b2i:
-      /* Booleans are stored as integers (or floats in GLSL 1.20 and lower). */
+      /* fallthrough to next case otherwise */
+   case ir_unop_b2f:
+      if (native_integers) {
+         emit(ir, TGSI_OPCODE_AND, result_dst, op[0], st_src_reg_for_float(1.0));
+         break;
+      }
+      /* fallthrough to next case otherwise */
+   case ir_unop_i2u:
+   case ir_unop_u2i:
+      /* Converting between signed and unsigned integers is a no-op. */
       result_src = op[0];
       break;
+   case ir_unop_b2i:
+      if (native_integers) {
+         /* Booleans are stored as integers using ~0 for true and 0 for false.
+          * GLSL requires that int(bool) return 1 for true and 0 for false.
+          * This conversion is done with AND, but it could be done with NEG.
+          */
+         emit(ir, TGSI_OPCODE_AND, result_dst, op[0], st_src_reg_for_int(1));
+      } else {
+         /* Booleans and integers are both stored as floats when native 
+          * integers are disabled.
+          */
+         result_src = op[0];
+      }
+      break;
    case ir_unop_f2i:
-      if (glsl_version >= 130)
+      if (native_integers)
          emit(ir, TGSI_OPCODE_F2I, result_dst, op[0]);
       else
          emit(ir, TGSI_OPCODE_TRUNC, result_dst, op[0]);
       break;
    case ir_unop_f2b:
+      emit(ir, TGSI_OPCODE_SNE, result_dst, op[0], st_src_reg_for_float(0.0));
+      break;
    case ir_unop_i2b:
-      emit(ir, TGSI_OPCODE_SNE, result_dst, op[0], 
-            st_src_reg_for_type(result_dst.type, 0));
+      if (native_integers)
+         emit(ir, TGSI_OPCODE_INEG, result_dst, op[0]);
+      else
+         emit(ir, TGSI_OPCODE_SNE, result_dst, op[0], st_src_reg_for_float(0.0));
       break;
    case ir_unop_trunc:
       emit(ir, TGSI_OPCODE_TRUNC, result_dst, op[0]);
@@ -1476,37 +1796,37 @@ glsl_to_tgsi_visitor::visit(ir_expression *ir)
       break;
 
    case ir_unop_bit_not:
-      if (glsl_version >= 130) {
+      if (native_integers) {
          emit(ir, TGSI_OPCODE_NOT, result_dst, op[0]);
          break;
       }
    case ir_unop_u2f:
-      if (glsl_version >= 130) {
+      if (native_integers) {
          emit(ir, TGSI_OPCODE_U2F, result_dst, op[0]);
          break;
       }
    case ir_binop_lshift:
-      if (glsl_version >= 130) {
+      if (native_integers) {
          emit(ir, TGSI_OPCODE_SHL, result_dst, op[0]);
          break;
       }
    case ir_binop_rshift:
-      if (glsl_version >= 130) {
+      if (native_integers) {
          emit(ir, TGSI_OPCODE_ISHR, result_dst, op[0]);
          break;
       }
    case ir_binop_bit_and:
-      if (glsl_version >= 130) {
+      if (native_integers) {
          emit(ir, TGSI_OPCODE_AND, result_dst, op[0]);
          break;
       }
    case ir_binop_bit_xor:
-      if (glsl_version >= 130) {
+      if (native_integers) {
          emit(ir, TGSI_OPCODE_XOR, result_dst, op[0]);
          break;
       }
    case ir_binop_bit_or:
-      if (glsl_version >= 130) {
+      if (native_integers) {
          emit(ir, TGSI_OPCODE_OR, result_dst, op[0]);
          break;
       }
@@ -1596,14 +1916,6 @@ glsl_to_tgsi_visitor::visit(ir_dereference_variable *ir)
          entry = new(mem_ctx) variable_storage(var,
                                                PROGRAM_INPUT,
                                                var->location);
-         if (this->prog->Target == GL_VERTEX_PROGRAM_ARB &&
-             var->location >= VERT_ATTRIB_GENERIC0) {
-            _mesa_add_attribute(this->prog->Attributes,
-                                var->name,
-                                _mesa_sizeof_glsl_type(var->type->gl_type),
-                                var->type->gl_type,
-                                var->location - VERT_ATTRIB_GENERIC0);
-         }
          break;
       case ir_var_out:
          assert(var->location != -1);
@@ -1633,7 +1945,7 @@ glsl_to_tgsi_visitor::visit(ir_dereference_variable *ir)
    }
 
    this->result = st_src_reg(entry->file, entry->index, var->type);
-   if (glsl_version <= 120)
+   if (!native_integers)
       this->result.type = GLSL_TYPE_FLOAT;
 }
 
@@ -1652,9 +1964,8 @@ glsl_to_tgsi_visitor::visit(ir_dereference_array *ir)
    if (index) {
       src.index += index->value.i[0] * element_size;
    } else {
-      st_src_reg array_base = this->result;
       /* Variable index array dereference.  It eats the "vec4" of the
-       * base of the array and an index that offsets the Mesa register
+       * base of the array and an index that offsets the TGSI register
        * index.
        */
       ir->array_index->accept(this);
@@ -1664,10 +1975,24 @@ glsl_to_tgsi_visitor::visit(ir_dereference_array *ir)
       if (element_size == 1) {
          index_reg = this->result;
       } else {
-         index_reg = get_temp(glsl_type::float_type);
+         index_reg = get_temp(native_integers ?
+                              glsl_type::int_type : glsl_type::float_type);
 
          emit(ir, TGSI_OPCODE_MUL, st_dst_reg(index_reg),
-              this->result, st_src_reg_for_float(element_size));
+              this->result, st_src_reg_for_type(index_reg.type, element_size));
+      }
+
+      /* If there was already a relative address register involved, add the
+       * new and the old together to get the new offset.
+       */
+      if (src.reladdr != NULL) {
+         st_src_reg accum_reg = get_temp(native_integers ?
+                                glsl_type::int_type : glsl_type::float_type);
+
+         emit(ir, TGSI_OPCODE_ADD, st_dst_reg(accum_reg),
+              index_reg, *src.reladdr);
+
+         index_reg = accum_reg;
       }
 
       src.reladdr = ralloc(mem_ctx, st_src_reg);
@@ -1838,7 +2163,8 @@ glsl_to_tgsi_visitor::visit(ir_assignment *ir)
    if (ir->write_mask == 0) {
       assert(!ir->lhs->type->is_scalar() && !ir->lhs->type->is_vector());
       l.writemask = WRITEMASK_XYZW;
-   } else if (ir->lhs->type->is_scalar()) {
+   } else if (ir->lhs->type->is_scalar() &&
+              ir->lhs->variable_referenced()->mode == ir_var_out) {
       /* FINISHME: This hack makes writing to gl_FragDepth, which lives in the
        * FINISHME: W component of fragment shader output zero, work correctly.
        */
@@ -1848,7 +2174,6 @@ glsl_to_tgsi_visitor::visit(ir_assignment *ir)
       int first_enabled_chan = 0;
       int rhs_chan = 0;
 
-      assert(ir->lhs->type->is_vector());
       l.writemask = ir->write_mask;
 
       for (int i = 0; i < 4; i++) {
@@ -1861,7 +2186,7 @@ glsl_to_tgsi_visitor::visit(ir_assignment *ir)
       /* Swizzle a small RHS vector into the channels being written.
        *
        * glsl ir treats write_mask as dictating how many channels are
-       * present on the RHS while Mesa IR treats write_mask as just
+       * present on the RHS while TGSI treats write_mask as just
        * showing which channels of the vec4 RHS get written.
        */
       for (int i = 0; i < 4; i++) {
@@ -1882,15 +2207,46 @@ glsl_to_tgsi_visitor::visit(ir_assignment *ir)
       st_src_reg condition = this->result;
 
       for (i = 0; i < type_size(ir->lhs->type); i++) {
+         st_src_reg l_src = st_src_reg(l);
+         st_src_reg condition_temp = condition;
+         l_src.swizzle = swizzle_for_size(ir->lhs->type->vector_elements);
+         
+         if (native_integers) {
+            /* This is necessary because TGSI's CMP instruction expects the
+             * condition to be a float, and we store booleans as integers.
+             * If TGSI had a UCMP instruction or similar, this extra
+             * instruction would not be necessary.
+             */
+            condition_temp = get_temp(glsl_type::vec4_type);
+            condition.negate = 0;
+            emit(ir, TGSI_OPCODE_I2F, st_dst_reg(condition_temp), condition);
+            condition_temp.swizzle = condition.swizzle;
+         }
+         
          if (switch_order) {
-            emit(ir, TGSI_OPCODE_CMP, l, condition, st_src_reg(l), r);
+            emit(ir, TGSI_OPCODE_CMP, l, condition_temp, l_src, r);
          } else {
-            emit(ir, TGSI_OPCODE_CMP, l, condition, r, st_src_reg(l));
+            emit(ir, TGSI_OPCODE_CMP, l, condition_temp, r, l_src);
          }
 
          l.index++;
          r.index++;
       }
+   } else if (ir->rhs->as_expression() &&
+              this->instructions.get_tail() &&
+              ir->rhs == ((glsl_to_tgsi_instruction *)this->instructions.get_tail())->ir &&
+              type_size(ir->lhs->type) == 1 &&
+              l.writemask == ((glsl_to_tgsi_instruction *)this->instructions.get_tail())->dst.writemask) {
+      /* To avoid emitting an extra MOV when assigning an expression to a 
+       * variable, emit the last instruction of the expression again, but
+       * replace the destination register with the target of the assignment.
+       * Dead code elimination will remove the original instruction.
+       */
+      glsl_to_tgsi_instruction *inst, *new_inst;
+      inst = (glsl_to_tgsi_instruction *)this->instructions.get_tail();
+      new_inst = emit(ir, inst->op, l, inst->src[0], inst->src[1], inst->src[2]);
+      new_inst->saturate = inst->saturate;
+      inst->dead_mask = inst->dst.writemask;
    } else {
       for (i = 0; i < type_size(ir->lhs->type); i++) {
          emit(ir, TGSI_OPCODE_MOV, l, r);
@@ -1909,9 +2265,11 @@ glsl_to_tgsi_visitor::visit(ir_constant *ir)
    gl_constant_value *values = (gl_constant_value *) stack_vals;
    GLenum gl_type = GL_NONE;
    unsigned int i;
+   static int in_array = 0;
+   gl_register_file file = in_array ? PROGRAM_CONSTANT : PROGRAM_IMMEDIATE;
 
    /* Unfortunately, 4 floats is all we can get into
-    * _mesa_add_unnamed_constant.  So, make a temp to store an
+    * _mesa_add_typed_unnamed_constant.  So, make a temp to store an
     * aggregate constant and move each constant value into it.  If we
     * get lucky, copy propagation will eliminate the extra moves.
     */
@@ -1945,6 +2303,7 @@ glsl_to_tgsi_visitor::visit(ir_constant *ir)
       int size = type_size(ir->type->fields.array);
 
       assert(size > 0);
+      in_array++;
 
       for (i = 0; i < ir->type->length; i++) {
          ir->array_elements[i]->accept(this);
@@ -1957,6 +2316,7 @@ glsl_to_tgsi_visitor::visit(ir_constant *ir)
          }
       }
       this->result = temp_base;
+      in_array--;
       return;
    }
 
@@ -1968,12 +2328,12 @@ glsl_to_tgsi_visitor::visit(ir_constant *ir)
          assert(ir->type->base_type == GLSL_TYPE_FLOAT);
          values = (gl_constant_value *) &ir->value.f[i * ir->type->vector_elements];
 
-         src = st_src_reg(PROGRAM_CONSTANT, -1, ir->type->base_type);
-         src.index = _mesa_add_typed_unnamed_constant(this->prog->Parameters,
-                                                      values,
-                                                      ir->type->vector_elements,
-                                                      GL_FLOAT,
-                                                      &src.swizzle);
+         src = st_src_reg(file, -1, ir->type->base_type);
+         src.index = add_constant(file,
+                                  values,
+                                  ir->type->vector_elements,
+                                  GL_FLOAT,
+                                  &src.swizzle);
          emit(ir, TGSI_OPCODE_MOV, mat_column, src);
 
          mat_column.index++;
@@ -1983,7 +2343,6 @@ glsl_to_tgsi_visitor::visit(ir_constant *ir)
       return;
    }
 
-   src.file = PROGRAM_CONSTANT;
    switch (ir->type->base_type) {
    case GLSL_TYPE_FLOAT:
       gl_type = GL_FLOAT;
@@ -1992,27 +2351,27 @@ glsl_to_tgsi_visitor::visit(ir_constant *ir)
       }
       break;
    case GLSL_TYPE_UINT:
-      gl_type = glsl_version >= 130 ? GL_UNSIGNED_INT : GL_FLOAT;
+      gl_type = native_integers ? GL_UNSIGNED_INT : GL_FLOAT;
       for (i = 0; i < ir->type->vector_elements; i++) {
-         if (glsl_version >= 130)
+         if (native_integers)
             values[i].u = ir->value.u[i];
          else
             values[i].f = ir->value.u[i];
       }
       break;
    case GLSL_TYPE_INT:
-      gl_type = glsl_version >= 130 ? GL_INT : GL_FLOAT;
+      gl_type = native_integers ? GL_INT : GL_FLOAT;
       for (i = 0; i < ir->type->vector_elements; i++) {
-         if (glsl_version >= 130)
+         if (native_integers)
             values[i].i = ir->value.i[i];
          else
             values[i].f = ir->value.i[i];
       }
       break;
    case GLSL_TYPE_BOOL:
-      gl_type = glsl_version >= 130 ? GL_BOOL : GL_FLOAT;
+      gl_type = native_integers ? GL_BOOL : GL_FLOAT;
       for (i = 0; i < ir->type->vector_elements; i++) {
-         if (glsl_version >= 130)
+         if (native_integers)
             values[i].b = ir->value.b[i];
          else
             values[i].f = ir->value.b[i];
@@ -2022,10 +2381,12 @@ glsl_to_tgsi_visitor::visit(ir_constant *ir)
       assert(!"Non-float/uint/int/bool constant");
    }
 
-   this->result = st_src_reg(PROGRAM_CONSTANT, -1, ir->type);
-   this->result.index = _mesa_add_typed_unnamed_constant(this->prog->Parameters,
-        					   values, ir->type->vector_elements, gl_type,
-        					   &this->result.swizzle);
+   this->result = st_src_reg(file, -1, ir->type);
+   this->result.index = add_constant(file,
+                                     values,
+                                     ir->type->vector_elements,
+                                     gl_type,
+                                     &this->result.swizzle);
 }
 
 function_entry *
@@ -2153,21 +2514,23 @@ glsl_to_tgsi_visitor::visit(ir_call *ir)
 void
 glsl_to_tgsi_visitor::visit(ir_texture *ir)
 {
-   st_src_reg result_src, coord, lod_info, projector, dx, dy;
+   st_src_reg result_src, coord, lod_info, projector, dx, dy, offset;
    st_dst_reg result_dst, coord_dst;
    glsl_to_tgsi_instruction *inst = NULL;
    unsigned opcode = TGSI_OPCODE_NOP;
 
-   ir->coordinate->accept(this);
+   if (ir->coordinate) {
+      ir->coordinate->accept(this);
 
-   /* Put our coords in a temp.  We'll need to modify them for shadow,
-    * projection, or LOD, so the only case we'd use it as is is if
-    * we're doing plain old texturing.  Mesa IR optimization should
-    * handle cleaning up our mess in that case.
-    */
-   coord = get_temp(glsl_type::vec4_type);
-   coord_dst = st_dst_reg(coord);
-   emit(ir, TGSI_OPCODE_MOV, coord_dst, this->result);
+      /* Put our coords in a temp.  We'll need to modify them for shadow,
+       * projection, or LOD, so the only case we'd use it as is is if
+       * we're doing plain old texturing.  The optimization passes on
+       * glsl_to_tgsi_visitor should handle cleaning up our mess in that case.
+       */
+      coord = get_temp(glsl_type::vec4_type);
+      coord_dst = st_dst_reg(coord);
+      emit(ir, TGSI_OPCODE_MOV, coord_dst, this->result);
+   }
 
    if (ir->projector) {
       ir->projector->accept(this);
@@ -2201,11 +2564,24 @@ glsl_to_tgsi_visitor::visit(ir_texture *ir)
       ir->lod_info.grad.dPdy->accept(this);
       dy = this->result;
       break;
-   case ir_txf: /* TODO: use TGSI_OPCODE_TXF here */
-      assert(!"GLSL 1.30 features unsupported");
+   case ir_txs:
+      opcode = TGSI_OPCODE_TXQ;
+      ir->lod_info.lod->accept(this);
+      lod_info = this->result;
+      break;
+   case ir_txf:
+      opcode = TGSI_OPCODE_TXF;
+      ir->lod_info.lod->accept(this);
+      lod_info = this->result;
+      if (ir->offset) {
+	 ir->offset->accept(this);
+	 offset = this->result;
+      }
       break;
    }
 
+   const glsl_type *sampler_type = ir->sampler->type;
+
    if (ir->projector) {
       if (opcode == TGSI_OPCODE_TEX) {
          /* Slot the projector in as the last component of the coord. */
@@ -2237,6 +2613,9 @@ glsl_to_tgsi_visitor::visit(ir_texture *ir)
             tmp_src = get_temp(glsl_type::vec4_type);
             st_dst_reg tmp_dst = st_dst_reg(tmp_src);
 
+	    /* Projective division not allowed for array samplers. */
+	    assert(!sampler_type->sampler_array);
+
             tmp_dst.writemask = WRITEMASK_Z;
             emit(ir, TGSI_OPCODE_MOV, tmp_dst, this->result);
 
@@ -2261,12 +2640,21 @@ glsl_to_tgsi_visitor::visit(ir_texture *ir)
        * coord.
        */
       ir->shadow_comparitor->accept(this);
-      coord_dst.writemask = WRITEMASK_Z;
+
+      /* XXX This will need to be updated for cubemap array samplers. */
+      if (sampler_type->sampler_dimensionality == GLSL_SAMPLER_DIM_2D &&
+          sampler_type->sampler_array) {
+         coord_dst.writemask = WRITEMASK_W;
+      } else {
+         coord_dst.writemask = WRITEMASK_Z;
+      }
+
       emit(ir, TGSI_OPCODE_MOV, coord_dst, this->result);
       coord_dst.writemask = WRITEMASK_XYZW;
    }
 
-   if (opcode == TGSI_OPCODE_TXL || opcode == TGSI_OPCODE_TXB) {
+   if (opcode == TGSI_OPCODE_TXL || opcode == TGSI_OPCODE_TXB ||
+       opcode == TGSI_OPCODE_TXF) {
       /* TGSI stores LOD or LOD bias in the last channel of the coords. */
       coord_dst.writemask = WRITEMASK_W;
       emit(ir, TGSI_OPCODE_MOV, coord_dst, lod_info);
@@ -2275,7 +2663,11 @@ glsl_to_tgsi_visitor::visit(ir_texture *ir)
 
    if (opcode == TGSI_OPCODE_TXD)
       inst = emit(ir, opcode, result_dst, coord, dx, dy);
-   else
+   else if (opcode == TGSI_OPCODE_TXQ)
+      inst = emit(ir, opcode, result_dst, lod_info);
+   else if (opcode == TGSI_OPCODE_TXF) {
+      inst = emit(ir, opcode, result_dst, coord);
+   } else
       inst = emit(ir, opcode, result_dst, coord);
 
    if (ir->shadow_comparitor)
@@ -2285,7 +2677,14 @@ glsl_to_tgsi_visitor::visit(ir_texture *ir)
         					   this->shader_program,
         					   this->prog);
 
-   const glsl_type *sampler_type = ir->sampler->type;
+   if (ir->offset) {
+       inst->tex_offset_num_offset = 1;
+       inst->tex_offsets[0].Index = offset.index;
+       inst->tex_offsets[0].File = offset.file;
+       inst->tex_offsets[0].SwizzleX = GET_SWZ(offset.swizzle, 0);
+       inst->tex_offsets[0].SwizzleY = GET_SWZ(offset.swizzle, 1);
+       inst->tex_offsets[0].SwizzleZ = GET_SWZ(offset.swizzle, 2);
+   }
 
    switch (sampler_type->sampler_dimensionality) {
    case GLSL_SAMPLER_DIM_1D:
@@ -2308,6 +2707,9 @@ glsl_to_tgsi_visitor::visit(ir_texture *ir)
    case GLSL_SAMPLER_DIM_BUF:
       assert(!"FINISHME: Implement ARB_texture_buffer_object");
       break;
+   case GLSL_SAMPLER_DIM_EXTERNAL:
+      inst->tex_target = TEXTURE_EXTERNAL_INDEX;
+      break;
    default:
       assert(!"Should not get here.");
    }
@@ -2358,7 +2760,7 @@ glsl_to_tgsi_visitor::visit(ir_discard *ir)
 void
 glsl_to_tgsi_visitor::visit(ir_if *ir)
 {
-   glsl_to_tgsi_instruction *cond_inst, *if_inst, *else_inst = NULL;
+   glsl_to_tgsi_instruction *cond_inst, *if_inst;
    glsl_to_tgsi_instruction *prev_inst;
 
    prev_inst = (glsl_to_tgsi_instruction *)this->instructions.get_tail();
@@ -2390,7 +2792,7 @@ glsl_to_tgsi_visitor::visit(ir_if *ir)
    visit_exec_list(&ir->then_instructions, this);
 
    if (!ir->else_instructions.is_empty()) {
-      else_inst = emit(ir->condition, TGSI_OPCODE_ELSE);
+      emit(ir->condition, TGSI_OPCODE_ELSE);
       visit_exec_list(&ir->else_instructions, this);
    }
 
@@ -2402,6 +2804,7 @@ glsl_to_tgsi_visitor::glsl_to_tgsi_visitor()
    result.file = PROGRAM_UNDEFINED;
    next_temp = 1;
    next_signature_id = 1;
+   num_immediates = 0;
    current_function = NULL;
    num_address_regs = 0;
    indirect_addr_temps = false;
@@ -2447,200 +2850,35 @@ count_resources(glsl_to_tgsi_visitor *v, gl_program *prog)
    _mesa_update_shader_textures_used(prog);
 }
 
-
-/**
- * Check if the given vertex/fragment/shader program is within the
- * resource limits of the context (number of texture units, etc).
- * If any of those checks fail, record a linker error.
- *
- * XXX more checks are needed...
- */
 static void
-check_resources(const struct gl_context *ctx,
-                struct gl_shader_program *shader_program,
-                glsl_to_tgsi_visitor *prog,
-                struct gl_program *proginfo)
+set_uniform_initializer(struct gl_context *ctx, void *mem_ctx,
+        		struct gl_shader_program *shader_program,
+        		const char *name, const glsl_type *type,
+        		ir_constant *val)
 {
-   switch (proginfo->Target) {
-   case GL_VERTEX_PROGRAM_ARB:
-      if (_mesa_bitcount(prog->samplers_used) >
-          ctx->Const.MaxVertexTextureImageUnits) {
-         fail_link(shader_program, "Too many vertex shader texture samplers");
-      }
-      if (proginfo->Parameters->NumParameters > MAX_UNIFORMS) {
-         fail_link(shader_program, "Too many vertex shader constants");
-      }
-      break;
-   case MESA_GEOMETRY_PROGRAM:
-      if (_mesa_bitcount(prog->samplers_used) >
-          ctx->Const.MaxGeometryTextureImageUnits) {
-         fail_link(shader_program, "Too many geometry shader texture samplers");
-      }
-      if (proginfo->Parameters->NumParameters >
-          MAX_GEOMETRY_UNIFORM_COMPONENTS / 4) {
-         fail_link(shader_program, "Too many geometry shader constants");
-      }
-      break;
-   case GL_FRAGMENT_PROGRAM_ARB:
-      if (_mesa_bitcount(prog->samplers_used) >
-          ctx->Const.MaxTextureImageUnits) {
-         fail_link(shader_program, "Too many fragment shader texture samplers");
-      }
-      if (proginfo->Parameters->NumParameters > MAX_UNIFORMS) {
-         fail_link(shader_program, "Too many fragment shader constants");
-      }
-      break;
-   default:
-      _mesa_problem(ctx, "unexpected program type in check_resources()");
-   }
-}
-
+   if (type->is_record()) {
+      ir_constant *field_constant;
 
+      field_constant = (ir_constant *)val->components.get_head();
 
-struct uniform_sort {
-   struct gl_uniform *u;
-   int pos;
-};
+      for (unsigned int i = 0; i < type->length; i++) {
+         const glsl_type *field_type = type->fields.structure[i].type;
+         const char *field_name = ralloc_asprintf(mem_ctx, "%s.%s", name,
+        				    type->fields.structure[i].name);
+         set_uniform_initializer(ctx, mem_ctx, shader_program, field_name,
+        			 field_type, field_constant);
+         field_constant = (ir_constant *)field_constant->next;
+      }
+      return;
+   }
 
-/* The shader_program->Uniforms list is almost sorted in increasing
- * uniform->{Frag,Vert}Pos locations, but not quite when there are
- * uniforms shared between targets.  We need to add parameters in
- * increasing order for the targets.
- */
-static int
-sort_uniforms(const void *a, const void *b)
-{
-   struct uniform_sort *u1 = (struct uniform_sort *)a;
-   struct uniform_sort *u2 = (struct uniform_sort *)b;
+   int loc = _mesa_get_uniform_location(ctx, shader_program, name);
 
-   return u1->pos - u2->pos;
-}
-
-/* Add the uniforms to the parameters.  The linker chose locations
- * in our parameters lists (which weren't created yet), which the
- * uniforms code will use to poke values into our parameters list
- * when uniforms are updated.
- */
-static void
-add_uniforms_to_parameters_list(struct gl_shader_program *shader_program,
-        			struct gl_shader *shader,
-        			struct gl_program *prog)
-{
-   unsigned int i;
-   unsigned int next_sampler = 0, num_uniforms = 0;
-   struct uniform_sort *sorted_uniforms;
-
-   sorted_uniforms = ralloc_array(NULL, struct uniform_sort,
-        			  shader_program->Uniforms->NumUniforms);
-
-   for (i = 0; i < shader_program->Uniforms->NumUniforms; i++) {
-      struct gl_uniform *uniform = shader_program->Uniforms->Uniforms + i;
-      int parameter_index = -1;
-
-      switch (shader->Type) {
-      case GL_VERTEX_SHADER:
-         parameter_index = uniform->VertPos;
-         break;
-      case GL_FRAGMENT_SHADER:
-         parameter_index = uniform->FragPos;
-         break;
-      case GL_GEOMETRY_SHADER:
-         parameter_index = uniform->GeomPos;
-         break;
-      }
-
-      /* Only add uniforms used in our target. */
-      if (parameter_index != -1) {
-         sorted_uniforms[num_uniforms].pos = parameter_index;
-         sorted_uniforms[num_uniforms].u = uniform;
-         num_uniforms++;
-      }
-   }
-
-   qsort(sorted_uniforms, num_uniforms, sizeof(struct uniform_sort),
-         sort_uniforms);
-
-   for (i = 0; i < num_uniforms; i++) {
-      struct gl_uniform *uniform = sorted_uniforms[i].u;
-      int parameter_index = sorted_uniforms[i].pos;
-      const glsl_type *type = uniform->Type;
-      unsigned int size;
-
-      if (type->is_vector() ||
-          type->is_scalar()) {
-         size = type->vector_elements;
-      } else {
-         size = type_size(type) * 4;
-      }
-
-      gl_register_file file;
-      if (type->is_sampler() ||
-          (type->is_array() && type->fields.array->is_sampler())) {
-         file = PROGRAM_SAMPLER;
-      } else {
-         file = PROGRAM_UNIFORM;
-      }
-
-      GLint index = _mesa_lookup_parameter_index(prog->Parameters, -1,
-        					 uniform->Name);
-
-      if (index < 0) {
-         index = _mesa_add_parameter(prog->Parameters, file,
-        			     uniform->Name, size, type->gl_type,
-        			     NULL, NULL, 0x0);
-
-         /* Sampler uniform values are stored in prog->SamplerUnits,
-          * and the entry in that array is selected by this index we
-          * store in ParameterValues[].
-          */
-         if (file == PROGRAM_SAMPLER) {
-            for (unsigned int j = 0; j < size / 4; j++)
-               prog->Parameters->ParameterValues[index + j][0].f = next_sampler++;
-         }
-
-         /* The location chosen in the Parameters list here (returned
-          * from _mesa_add_uniform) has to match what the linker chose.
-          */
-         if (index != parameter_index) {
-            fail_link(shader_program, "Allocation of uniform `%s' to target "
-        	      "failed (%d vs %d)\n",
-        	      uniform->Name, index, parameter_index);
-         }
-      }
-   }
-
-   ralloc_free(sorted_uniforms);
-}
-
-static void
-set_uniform_initializer(struct gl_context *ctx, void *mem_ctx,
-        		struct gl_shader_program *shader_program,
-        		const char *name, const glsl_type *type,
-        		ir_constant *val)
-{
-   if (type->is_record()) {
-      ir_constant *field_constant;
-
-      field_constant = (ir_constant *)val->components.get_head();
-
-      for (unsigned int i = 0; i < type->length; i++) {
-         const glsl_type *field_type = type->fields.structure[i].type;
-         const char *field_name = ralloc_asprintf(mem_ctx, "%s.%s", name,
-        				    type->fields.structure[i].name);
-         set_uniform_initializer(ctx, mem_ctx, shader_program, field_name,
-        			 field_type, field_constant);
-         field_constant = (ir_constant *)field_constant->next;
-      }
-      return;
-   }
-
-   int loc = _mesa_get_uniform_location(ctx, shader_program, name);
-
-   if (loc == -1) {
-      fail_link(shader_program,
-        	"Couldn't find uniform for initializer %s\n", name);
-      return;
-   }
+   if (loc == -1) {
+      fail_link(shader_program,
+        	"Couldn't find uniform for initializer %s\n", name);
+      return;
+   }
 
    for (unsigned int i = 0; i < (type->is_array() ? type->length : 1); i++) {
       ir_constant *element;
@@ -2673,43 +2911,13 @@ set_uniform_initializer(struct gl_context *ctx, void *mem_ctx,
         		      element_type->matrix_columns,
         		      element_type->vector_elements,
         		      loc, 1, GL_FALSE, (GLfloat *)values);
-         loc += element_type->matrix_columns;
       } else {
          _mesa_uniform(ctx, shader_program, loc, element_type->matrix_columns,
         	       values, element_type->gl_type);
-         loc += type_size(element_type);
       }
-   }
-}
-
-static void
-set_uniform_initializers(struct gl_context *ctx,
-        		 struct gl_shader_program *shader_program)
-{
-   void *mem_ctx = NULL;
-
-   for (unsigned int i = 0; i < MESA_SHADER_TYPES; i++) {
-      struct gl_shader *shader = shader_program->_LinkedShaders[i];
-
-      if (shader == NULL)
-         continue;
-
-      foreach_iter(exec_list_iterator, iter, *shader->ir) {
-         ir_instruction *ir = (ir_instruction *)iter.get();
-         ir_variable *var = ir->as_variable();
-
-         if (!var || var->mode != ir_var_uniform || !var->constant_value)
-            continue;
-
-         if (!mem_ctx)
-            mem_ctx = ralloc_context(NULL);
 
-         set_uniform_initializer(ctx, mem_ctx, shader_program, var->name,
-        			 var->type, var->constant_value);
-      }
+      loc++;
    }
-
-   ralloc_free(mem_ctx);
 }
 
 /*
@@ -2729,11 +2937,15 @@ glsl_to_tgsi_visitor::remove_output_reads(gl_register_file type)
    GLint outputMap[VERT_RESULT_MAX];
    GLint outputTypes[VERT_RESULT_MAX];
    GLuint numVaryingReads = 0;
-   GLboolean usedTemps[MAX_PROGRAM_TEMPS];
+   GLboolean *usedTemps;
    GLuint firstTemp = 0;
 
+   usedTemps = new GLboolean[MAX_TEMPS];
+   if (!usedTemps) {
+      return;
+   }
    _mesa_find_used_registers(prog, PROGRAM_TEMPORARY,
-                             usedTemps, MAX_PROGRAM_TEMPS);
+                             usedTemps, MAX_TEMPS);
 
    assert(type == PROGRAM_VARYING || type == PROGRAM_OUTPUT);
    assert(prog->Target == GL_VERTEX_PROGRAM_ARB || type != PROGRAM_VARYING);
@@ -2753,7 +2965,7 @@ glsl_to_tgsi_visitor::remove_output_reads(gl_register_file type)
             if (outputMap[var] == -1) {
                numVaryingReads++;
                outputMap[var] = _mesa_find_free_register(usedTemps,
-                                                         MAX_PROGRAM_TEMPS,
+                                                         MAX_TEMPS,
                                                          firstTemp);
                outputTypes[var] = inst->src[j].type;
                firstTemp = outputMap[var] + 1;
@@ -2764,6 +2976,8 @@ glsl_to_tgsi_visitor::remove_output_reads(gl_register_file type)
       }
    }
 
+   delete [] usedTemps;
+
    if (numVaryingReads == 0)
       return; /* nothing to be done */
 
@@ -2835,9 +3049,13 @@ get_src_arg_mask(st_dst_reg dst, st_src_reg src)
 void
 glsl_to_tgsi_visitor::simplify_cmp(void)
 {
-   unsigned tempWrites[MAX_PROGRAM_TEMPS];
+   unsigned *tempWrites;
    unsigned outputWrites[MAX_PROGRAM_OUTPUTS];
 
+   tempWrites = new unsigned[MAX_TEMPS];
+   if (!tempWrites) {
+      return;
+   }
    memset(tempWrites, 0, sizeof(tempWrites));
    memset(outputWrites, 0, sizeof(outputWrites));
 
@@ -2853,7 +3071,7 @@ glsl_to_tgsi_visitor::simplify_cmp(void)
           inst->op == TGSI_OPCODE_END ||
           inst->op == TGSI_OPCODE_ENDSUB ||
           inst->op == TGSI_OPCODE_RET) {
-         return;
+         break;
       }
 
       if (inst->dst.file == PROGRAM_OUTPUT) {
@@ -2861,7 +3079,7 @@ glsl_to_tgsi_visitor::simplify_cmp(void)
          prevWriteMask = outputWrites[inst->dst.index];
          outputWrites[inst->dst.index] |= inst->dst.writemask;
       } else if (inst->dst.file == PROGRAM_TEMPORARY) {
-         assert(inst->dst.index < MAX_PROGRAM_TEMPS);
+         assert(inst->dst.index < MAX_TEMPS);
          prevWriteMask = tempWrites[inst->dst.index];
          tempWrites[inst->dst.index] |= inst->dst.writemask;
       }
@@ -2878,6 +3096,8 @@ glsl_to_tgsi_visitor::simplify_cmp(void)
          inst->src[0] = inst->src[1];
       }
    }
+
+   delete [] tempWrites;
 }
 
 /* Replaces all references to a temporary register index with another index. */
@@ -3254,6 +3474,149 @@ glsl_to_tgsi_visitor::eliminate_dead_code(void)
    }
 }
 
+/*
+ * On a basic block basis, tracks available PROGRAM_TEMPORARY registers for dead
+ * code elimination.  This is less primitive than eliminate_dead_code(), as it
+ * is per-channel and can detect consecutive writes without a read between them
+ * as dead code.  However, there is some dead code that can be eliminated by 
+ * eliminate_dead_code() but not this function - for example, this function 
+ * cannot eliminate an instruction writing to a register that is never read and
+ * is the only instruction writing to that register.
+ *
+ * The glsl_to_tgsi_visitor lazily produces code assuming that this pass
+ * will occur.
+ */
+int
+glsl_to_tgsi_visitor::eliminate_dead_code_advanced(void)
+{
+   glsl_to_tgsi_instruction **writes = rzalloc_array(mem_ctx,
+                                                     glsl_to_tgsi_instruction *,
+                                                     this->next_temp * 4);
+   int *write_level = rzalloc_array(mem_ctx, int, this->next_temp * 4);
+   int level = 0;
+   int removed = 0;
+
+   foreach_iter(exec_list_iterator, iter, this->instructions) {
+      glsl_to_tgsi_instruction *inst = (glsl_to_tgsi_instruction *)iter.get();
+
+      assert(inst->dst.file != PROGRAM_TEMPORARY
+             || inst->dst.index < this->next_temp);
+      
+      switch (inst->op) {
+      case TGSI_OPCODE_BGNLOOP:
+      case TGSI_OPCODE_ENDLOOP:
+         /* End of a basic block, clear the write array entirely.
+          * FIXME: This keeps us from killing dead code when the writes are
+          * on either side of a loop, even when the register isn't touched
+          * inside the loop.
+          */
+         memset(writes, 0, sizeof(*writes) * this->next_temp * 4);
+         break;
+
+      case TGSI_OPCODE_ENDIF:
+      case TGSI_OPCODE_ELSE:
+         /* Promote the recorded level all channels written inside the preceding
+          * if or else block to the level above the if/else block.
+          */
+         for (int r = 0; r < this->next_temp; r++) {
+            for (int c = 0; c < 4; c++) {
+               if (!writes[4 * r + c])
+        	         continue;
+
+               if (write_level[4 * r + c] == level)
+        	         write_level[4 * r + c] = level-1;
+            }
+         }
+
+         if(inst->op == TGSI_OPCODE_ENDIF)
+            --level;
+         
+         break;
+
+      case TGSI_OPCODE_IF:
+         ++level;
+         /* fallthrough to default case to mark the condition as read */
+      
+      default:
+         /* Continuing the block, clear any channels from the write array that
+          * are read by this instruction.
+          */
+         for (unsigned i = 0; i < Elements(inst->src); i++) {
+            if (inst->src[i].file == PROGRAM_TEMPORARY && inst->src[i].reladdr){
+               /* Any temporary might be read, so no dead code elimination 
+                * across this instruction.
+                */
+               memset(writes, 0, sizeof(*writes) * this->next_temp * 4);
+            } else if (inst->src[i].file == PROGRAM_TEMPORARY) {
+               /* Clear where it's used as src. */
+               int src_chans = 1 << GET_SWZ(inst->src[i].swizzle, 0);
+               src_chans |= 1 << GET_SWZ(inst->src[i].swizzle, 1);
+               src_chans |= 1 << GET_SWZ(inst->src[i].swizzle, 2);
+               src_chans |= 1 << GET_SWZ(inst->src[i].swizzle, 3);
+               
+               for (int c = 0; c < 4; c++) {
+              	   if (src_chans & (1 << c)) {
+              	      writes[4 * inst->src[i].index + c] = NULL;
+              	   }
+               }
+            }
+         }
+         break;
+      }
+
+      /* If this instruction writes to a temporary, add it to the write array.
+       * If there is already an instruction in the write array for one or more
+       * of the channels, flag that channel write as dead.
+       */
+      if (inst->dst.file == PROGRAM_TEMPORARY &&
+          !inst->dst.reladdr &&
+          !inst->saturate) {
+         for (int c = 0; c < 4; c++) {
+            if (inst->dst.writemask & (1 << c)) {
+               if (writes[4 * inst->dst.index + c]) {
+                  if (write_level[4 * inst->dst.index + c] < level)
+                     continue;
+                  else
+                     writes[4 * inst->dst.index + c]->dead_mask |= (1 << c);
+               }
+               writes[4 * inst->dst.index + c] = inst;
+               write_level[4 * inst->dst.index + c] = level;
+            }
+         }
+      }
+   }
+
+   /* Anything still in the write array at this point is dead code. */
+   for (int r = 0; r < this->next_temp; r++) {
+      for (int c = 0; c < 4; c++) {
+         glsl_to_tgsi_instruction *inst = writes[4 * r + c];
+         if (inst)
+            inst->dead_mask |= (1 << c);
+      }
+   }
+
+   /* Now actually remove the instructions that are completely dead and update
+    * the writemask of other instructions with dead channels.
+    */
+   foreach_iter(exec_list_iterator, iter, this->instructions) {
+      glsl_to_tgsi_instruction *inst = (glsl_to_tgsi_instruction *)iter.get();
+      
+      if (!inst->dead_mask || !inst->dst.writemask)
+         continue;
+      else if ((inst->dst.writemask & ~inst->dead_mask) == 0) {
+         iter.remove();
+         delete inst;
+         removed++;
+      } else
+         inst->dst.writemask &= ~(inst->dead_mask);
+   }
+
+   ralloc_free(write_level);
+   ralloc_free(writes);
+   
+   return removed;
+}
+
 /* Merges temporary registers together where possible to reduce the number of 
  * registers needed to run a program.
  * 
@@ -3325,6 +3688,205 @@ glsl_to_tgsi_visitor::renumber_registers(void)
    this->next_temp = new_index;
 }
 
+/**
+ * Returns a fragment program which implements the current pixel transfer ops.
+ * Based on get_pixel_transfer_program in st_atom_pixeltransfer.c.
+ */
+extern "C" void
+get_pixel_transfer_visitor(struct st_fragment_program *fp,
+                           glsl_to_tgsi_visitor *original,
+                           int scale_and_bias, int pixel_maps)
+{
+   glsl_to_tgsi_visitor *v = new glsl_to_tgsi_visitor();
+   struct st_context *st = st_context(original->ctx);
+   struct gl_program *prog = &fp->Base.Base;
+   struct gl_program_parameter_list *params = _mesa_new_parameter_list();
+   st_src_reg coord, src0;
+   st_dst_reg dst0;
+   glsl_to_tgsi_instruction *inst;
+
+   /* Copy attributes of the glsl_to_tgsi_visitor in the original shader. */
+   v->ctx = original->ctx;
+   v->prog = prog;
+   v->glsl_version = original->glsl_version;
+   v->native_integers = original->native_integers;
+   v->options = original->options;
+   v->next_temp = original->next_temp;
+   v->num_address_regs = original->num_address_regs;
+   v->samplers_used = prog->SamplersUsed = original->samplers_used;
+   v->indirect_addr_temps = original->indirect_addr_temps;
+   v->indirect_addr_consts = original->indirect_addr_consts;
+   memcpy(&v->immediates, &original->immediates, sizeof(v->immediates));
+
+   /*
+    * Get initial pixel color from the texture.
+    * TEX colorTemp, fragment.texcoord[0], texture[0], 2D;
+    */
+   coord = st_src_reg(PROGRAM_INPUT, FRAG_ATTRIB_TEX0, glsl_type::vec2_type);
+   src0 = v->get_temp(glsl_type::vec4_type);
+   dst0 = st_dst_reg(src0);
+   inst = v->emit(NULL, TGSI_OPCODE_TEX, dst0, coord);
+   inst->sampler = 0;
+   inst->tex_target = TEXTURE_2D_INDEX;
+
+   prog->InputsRead |= FRAG_BIT_TEX0;
+   prog->SamplersUsed |= (1 << 0); /* mark sampler 0 as used */
+   v->samplers_used |= (1 << 0);
+
+   if (scale_and_bias) {
+      static const gl_state_index scale_state[STATE_LENGTH] =
+         { STATE_INTERNAL, STATE_PT_SCALE,
+           (gl_state_index) 0, (gl_state_index) 0, (gl_state_index) 0 };
+      static const gl_state_index bias_state[STATE_LENGTH] =
+         { STATE_INTERNAL, STATE_PT_BIAS,
+           (gl_state_index) 0, (gl_state_index) 0, (gl_state_index) 0 };
+      GLint scale_p, bias_p;
+      st_src_reg scale, bias;
+
+      scale_p = _mesa_add_state_reference(params, scale_state);
+      bias_p = _mesa_add_state_reference(params, bias_state);
+
+      /* MAD colorTemp, colorTemp, scale, bias; */
+      scale = st_src_reg(PROGRAM_STATE_VAR, scale_p, GLSL_TYPE_FLOAT);
+      bias = st_src_reg(PROGRAM_STATE_VAR, bias_p, GLSL_TYPE_FLOAT);
+      inst = v->emit(NULL, TGSI_OPCODE_MAD, dst0, src0, scale, bias);
+   }
+
+   if (pixel_maps) {
+      st_src_reg temp = v->get_temp(glsl_type::vec4_type);
+      st_dst_reg temp_dst = st_dst_reg(temp);
+
+      assert(st->pixel_xfer.pixelmap_texture);
+
+      /* With a little effort, we can do four pixel map look-ups with
+       * two TEX instructions:
+       */
+
+      /* TEX temp.rg, colorTemp.rgba, texture[1], 2D; */
+      temp_dst.writemask = WRITEMASK_XY; /* write R,G */
+      inst = v->emit(NULL, TGSI_OPCODE_TEX, temp_dst, src0);
+      inst->sampler = 1;
+      inst->tex_target = TEXTURE_2D_INDEX;
+
+      /* TEX temp.ba, colorTemp.baba, texture[1], 2D; */
+      src0.swizzle = MAKE_SWIZZLE4(SWIZZLE_Z, SWIZZLE_W, SWIZZLE_Z, SWIZZLE_W);
+      temp_dst.writemask = WRITEMASK_ZW; /* write B,A */
+      inst = v->emit(NULL, TGSI_OPCODE_TEX, temp_dst, src0);
+      inst->sampler = 1;
+      inst->tex_target = TEXTURE_2D_INDEX;
+
+      prog->SamplersUsed |= (1 << 1); /* mark sampler 1 as used */
+      v->samplers_used |= (1 << 1);
+
+      /* MOV colorTemp, temp; */
+      inst = v->emit(NULL, TGSI_OPCODE_MOV, dst0, temp);
+   }
+
+   /* Now copy the instructions from the original glsl_to_tgsi_visitor into the
+    * new visitor. */
+   foreach_iter(exec_list_iterator, iter, original->instructions) {
+      glsl_to_tgsi_instruction *inst = (glsl_to_tgsi_instruction *)iter.get();
+      st_src_reg src_regs[3];
+
+      if (inst->dst.file == PROGRAM_OUTPUT)
+         prog->OutputsWritten |= BITFIELD64_BIT(inst->dst.index);
+
+      for (int i=0; i<3; i++) {
+         src_regs[i] = inst->src[i];
+         if (src_regs[i].file == PROGRAM_INPUT &&
+             src_regs[i].index == FRAG_ATTRIB_COL0)
+         {
+            src_regs[i].file = PROGRAM_TEMPORARY;
+            src_regs[i].index = src0.index;
+         }
+         else if (src_regs[i].file == PROGRAM_INPUT)
+            prog->InputsRead |= BITFIELD64_BIT(src_regs[i].index);
+      }
+
+      v->emit(NULL, inst->op, inst->dst, src_regs[0], src_regs[1], src_regs[2]);
+   }
+
+   /* Make modifications to fragment program info. */
+   prog->Parameters = _mesa_combine_parameter_lists(params,
+                                                    original->prog->Parameters);
+   _mesa_free_parameter_list(params);
+   count_resources(v, prog);
+   fp->glsl_to_tgsi = v;
+}
+
+/**
+ * Make fragment program for glBitmap:
+ *   Sample the texture and kill the fragment if the bit is 0.
+ * This program will be combined with the user's fragment program.
+ *
+ * Based on make_bitmap_fragment_program in st_cb_bitmap.c.
+ */
+extern "C" void
+get_bitmap_visitor(struct st_fragment_program *fp,
+                   glsl_to_tgsi_visitor *original, int samplerIndex)
+{
+   glsl_to_tgsi_visitor *v = new glsl_to_tgsi_visitor();
+   struct st_context *st = st_context(original->ctx);
+   struct gl_program *prog = &fp->Base.Base;
+   st_src_reg coord, src0;
+   st_dst_reg dst0;
+   glsl_to_tgsi_instruction *inst;
+
+   /* Copy attributes of the glsl_to_tgsi_visitor in the original shader. */
+   v->ctx = original->ctx;
+   v->prog = prog;
+   v->glsl_version = original->glsl_version;
+   v->native_integers = original->native_integers;
+   v->options = original->options;
+   v->next_temp = original->next_temp;
+   v->num_address_regs = original->num_address_regs;
+   v->samplers_used = prog->SamplersUsed = original->samplers_used;
+   v->indirect_addr_temps = original->indirect_addr_temps;
+   v->indirect_addr_consts = original->indirect_addr_consts;
+   memcpy(&v->immediates, &original->immediates, sizeof(v->immediates));
+
+   /* TEX tmp0, fragment.texcoord[0], texture[0], 2D; */
+   coord = st_src_reg(PROGRAM_INPUT, FRAG_ATTRIB_TEX0, glsl_type::vec2_type);
+   src0 = v->get_temp(glsl_type::vec4_type);
+   dst0 = st_dst_reg(src0);
+   inst = v->emit(NULL, TGSI_OPCODE_TEX, dst0, coord);
+   inst->sampler = samplerIndex;
+   inst->tex_target = TEXTURE_2D_INDEX;
+
+   prog->InputsRead |= FRAG_BIT_TEX0;
+   prog->SamplersUsed |= (1 << samplerIndex); /* mark sampler as used */
+   v->samplers_used |= (1 << samplerIndex);
+
+   /* KIL if -tmp0 < 0 # texel=0 -> keep / texel=0 -> discard */
+   src0.negate = NEGATE_XYZW;
+   if (st->bitmap.tex_format == PIPE_FORMAT_L8_UNORM)
+      src0.swizzle = SWIZZLE_XXXX;
+   inst = v->emit(NULL, TGSI_OPCODE_KIL, undef_dst, src0);
+
+   /* Now copy the instructions from the original glsl_to_tgsi_visitor into the
+    * new visitor. */
+   foreach_iter(exec_list_iterator, iter, original->instructions) {
+      glsl_to_tgsi_instruction *inst = (glsl_to_tgsi_instruction *)iter.get();
+      st_src_reg src_regs[3];
+
+      if (inst->dst.file == PROGRAM_OUTPUT)
+         prog->OutputsWritten |= BITFIELD64_BIT(inst->dst.index);
+
+      for (int i=0; i<3; i++) {
+         src_regs[i] = inst->src[i];
+         if (src_regs[i].file == PROGRAM_INPUT)
+            prog->InputsRead |= BITFIELD64_BIT(src_regs[i].index);
+      }
+
+      v->emit(NULL, inst->op, inst->dst, src_regs[0], src_regs[1], src_regs[2]);
+   }
+
+   /* Make modifications to fragment program info. */
+   prog->Parameters = _mesa_clone_parameter_list(original->prog->Parameters);
+   count_resources(v, prog);
+   fp->glsl_to_tgsi = v;
+}
+
 /* ------------------------- TGSI conversion stuff -------------------------- */
 struct label {
    unsigned branch_target;
@@ -3337,8 +3899,9 @@ struct label {
 struct st_translate {
    struct ureg_program *ureg;
 
-   struct ureg_dst temps[MAX_PROGRAM_TEMPS];
+   struct ureg_dst temps[MAX_TEMPS];
    struct ureg_src *constants;
+   struct ureg_src *immediates;
    struct ureg_dst outputs[PIPE_MAX_SHADER_OUTPUTS];
    struct ureg_src inputs[PIPE_MAX_SHADER_INPUTS];
    struct ureg_dst address[1];
@@ -3378,6 +3941,7 @@ struct st_translate {
 /** Map Mesa's SYSTEM_VALUE_x to TGSI_SEMANTIC_x */
 static unsigned mesa_sysval_to_semantic[SYSTEM_VALUE_MAX] = {
    TGSI_SEMANTIC_FACE,
+   TGSI_SEMANTIC_VERTEXID,
    TGSI_SEMANTIC_INSTANCEID
 };
 
@@ -3387,15 +3951,14 @@ static unsigned mesa_sysval_to_semantic[SYSTEM_VALUE_MAX] = {
  * of labels built here and patch the TGSI code with the actual
  * location of each label.
  */
-static unsigned *get_label( struct st_translate *t,
-                            unsigned branch_target )
+static unsigned *get_label(struct st_translate *t, unsigned branch_target)
 {
    unsigned i;
 
    if (t->labels_count + 1 >= t->labels_size) {
       t->labels_size = 1 << (util_logbase2(t->labels_size) + 1);
       t->labels = (struct label *)realloc(t->labels, 
-                                          t->labels_size * sizeof t->labels[0]);
+                                          t->labels_size * sizeof(struct label));
       if (t->labels == NULL) {
          static unsigned dummy;
          t->error = TRUE;
@@ -3409,17 +3972,16 @@ static unsigned *get_label( struct st_translate *t,
 }
 
 /**
- * Called prior to emitting the TGSI code for each Mesa instruction.
+ * Called prior to emitting the TGSI code for each instruction.
  * Allocate additional space for instructions if needed.
- * Update the insn[] array so the next Mesa instruction points to
+ * Update the insn[] array so the next glsl_to_tgsi_instruction points to
  * the next TGSI instruction.
  */
-static void set_insn_start( struct st_translate *t,
-                            unsigned start )
+static void set_insn_start(struct st_translate *t, unsigned start)
 {
    if (t->insn_count + 1 >= t->insn_size) {
       t->insn_size = 1 << (util_logbase2(t->insn_size) + 1);
-      t->insn = (unsigned *)realloc(t->insn, t->insn_size * sizeof t->insn[0]);
+      t->insn = (unsigned *)realloc(t->insn, t->insn_size * sizeof(t->insn[0]));
       if (t->insn == NULL) {
          t->error = TRUE;
          return;
@@ -3430,20 +3992,45 @@ static void set_insn_start( struct st_translate *t,
 }
 
 /**
- * Map a Mesa dst register to a TGSI ureg_dst register.
+ * Map a glsl_to_tgsi constant/immediate to a TGSI immediate.
+ */
+static struct ureg_src
+emit_immediate(struct st_translate *t,
+               gl_constant_value values[4],
+               int type, int size)
+{
+   struct ureg_program *ureg = t->ureg;
+
+   switch(type)
+   {
+   case GL_FLOAT:
+      return ureg_DECL_immediate(ureg, &values[0].f, size);
+   case GL_INT:
+      return ureg_DECL_immediate_int(ureg, &values[0].i, size);
+   case GL_UNSIGNED_INT:
+   case GL_BOOL:
+      return ureg_DECL_immediate_uint(ureg, &values[0].u, size);
+   default:
+      assert(!"should not get here - type must be float, int, uint, or bool");
+      return ureg_src_undef();
+   }
+}
+
+/**
+ * Map a glsl_to_tgsi dst register to a TGSI ureg_dst register.
  */
 static struct ureg_dst
-dst_register( struct st_translate *t,
-              gl_register_file file,
-              GLuint index )
+dst_register(struct st_translate *t,
+             gl_register_file file,
+             GLuint index)
 {
-   switch( file ) {
+   switch(file) {
    case PROGRAM_UNDEFINED:
       return ureg_dst_undef();
 
    case PROGRAM_TEMPORARY:
       if (ureg_dst_is_undef(t->temps[index]))
-         t->temps[index] = ureg_DECL_temporary( t->ureg );
+         t->temps[index] = ureg_DECL_temporary(t->ureg);
 
       return t->temps[index];
 
@@ -3466,20 +4053,20 @@ dst_register( struct st_translate *t,
       return t->address[index];
 
    default:
-      debug_assert( 0 );
+      assert(!"unknown dst register file");
       return ureg_dst_undef();
    }
 }
 
 /**
- * Map a Mesa src register to a TGSI ureg_src register.
+ * Map a glsl_to_tgsi src register to a TGSI ureg_src register.
  */
 static struct ureg_src
-src_register( struct st_translate *t,
-              gl_register_file file,
-              GLuint index )
+src_register(struct st_translate *t,
+             gl_register_file file,
+             GLuint index)
 {
-   switch( file ) {
+   switch(file) {
    case PROGRAM_UNDEFINED:
       return ureg_src_undef();
 
@@ -3487,7 +4074,7 @@ src_register( struct st_translate *t,
       assert(index >= 0);
       assert(index < Elements(t->temps));
       if (ureg_dst_is_undef(t->temps[index]))
-         t->temps[index] = ureg_DECL_temporary( t->ureg );
+         t->temps[index] = ureg_DECL_temporary(t->ureg);
       return ureg_src(t->temps[index]);
 
    case PROGRAM_NAMED_PARAM:
@@ -3499,10 +4086,13 @@ src_register( struct st_translate *t,
    case PROGRAM_STATE_VAR:
    case PROGRAM_CONSTANT:       /* ie, immediate */
       if (index < 0)
-         return ureg_DECL_constant( t->ureg, 0 );
+         return ureg_DECL_constant(t->ureg, 0);
       else
          return t->constants[index];
 
+   case PROGRAM_IMMEDIATE:
+      return t->immediates[index];
+
    case PROGRAM_INPUT:
       assert(t->inputMapping[index] < Elements(t->inputs));
       return t->inputs[t->inputMapping[index]];
@@ -3519,7 +4109,7 @@ src_register( struct st_translate *t,
       return t->systemValues[index];
 
    default:
-      debug_assert( 0 );
+      assert(!"unknown src register file");
       return ureg_src_undef();
    }
 }
@@ -3528,22 +4118,21 @@ src_register( struct st_translate *t,
  * Create a TGSI ureg_dst register from an st_dst_reg.
  */
 static struct ureg_dst
-translate_dst( struct st_translate *t,
-               const st_dst_reg *dst_reg,
-               boolean saturate )
+translate_dst(struct st_translate *t,
+              const st_dst_reg *dst_reg,
+              bool saturate)
 {
-   struct ureg_dst dst = dst_register( t, 
-                                       dst_reg->file,
-                                       dst_reg->index );
+   struct ureg_dst dst = dst_register(t, 
+                                      dst_reg->file,
+                                      dst_reg->index);
 
-   dst = ureg_writemask( dst, 
-                         dst_reg->writemask );
+   dst = ureg_writemask(dst, dst_reg->writemask);
    
    if (saturate)
-      dst = ureg_saturate( dst );
+      dst = ureg_saturate(dst);
 
    if (dst_reg->reladdr != NULL)
-      dst = ureg_dst_indirect( dst, ureg_src(t->address[0]) );
+      dst = ureg_dst_indirect(dst, ureg_src(t->address[0]));
 
    return dst;
 }
@@ -3552,16 +4141,15 @@ translate_dst( struct st_translate *t,
  * Create a TGSI ureg_src register from an st_src_reg.
  */
 static struct ureg_src
-translate_src( struct st_translate *t,
-               const st_src_reg *src_reg )
+translate_src(struct st_translate *t, const st_src_reg *src_reg)
 {
-   struct ureg_src src = src_register( t, src_reg->file, src_reg->index );
+   struct ureg_src src = src_register(t, src_reg->file, src_reg->index);
 
-   src = ureg_swizzle( src,
-                       GET_SWZ( src_reg->swizzle, 0 ) & 0x3,
-                       GET_SWZ( src_reg->swizzle, 1 ) & 0x3,
-                       GET_SWZ( src_reg->swizzle, 2 ) & 0x3,
-                       GET_SWZ( src_reg->swizzle, 3 ) & 0x3);
+   src = ureg_swizzle(src,
+                      GET_SWZ(src_reg->swizzle, 0) & 0x3,
+                      GET_SWZ(src_reg->swizzle, 1) & 0x3,
+                      GET_SWZ(src_reg->swizzle, 2) & 0x3,
+                      GET_SWZ(src_reg->swizzle, 3) & 0x3);
 
    if ((src_reg->negate & 0xf) == NEGATE_XYZW)
       src = ureg_negate(src);
@@ -3592,40 +4180,59 @@ translate_src( struct st_translate *t,
    return src;
 }
 
+static struct tgsi_texture_offset
+translate_tex_offset(struct st_translate *t,
+                     const struct tgsi_texture_offset *in_offset)
+{
+   struct tgsi_texture_offset offset;
+
+   assert(in_offset->File == PROGRAM_IMMEDIATE);
+
+   offset.File = TGSI_FILE_IMMEDIATE;
+   offset.Index = in_offset->Index;
+   offset.SwizzleX = in_offset->SwizzleX;
+   offset.SwizzleY = in_offset->SwizzleY;
+   offset.SwizzleZ = in_offset->SwizzleZ;
+
+   return offset;
+}
+
 static void
-compile_tgsi_instruction(struct st_translate *t, 
-        			     const struct glsl_to_tgsi_instruction *inst)
+compile_tgsi_instruction(struct st_translate *t,
+                         const glsl_to_tgsi_instruction *inst)
 {
    struct ureg_program *ureg = t->ureg;
    GLuint i;
    struct ureg_dst dst[1];
    struct ureg_src src[4];
+   struct tgsi_texture_offset texoffsets[MAX_GLSL_TEXTURE_OFFSET];
+
    unsigned num_dst;
    unsigned num_src;
 
-   num_dst = num_inst_dst_regs( inst->op );
-   num_src = num_inst_src_regs( inst->op );
+   num_dst = num_inst_dst_regs(inst->op);
+   num_src = num_inst_src_regs(inst->op);
 
    if (num_dst) 
-      dst[0] = translate_dst( t, 
-                              &inst->dst,
-                              inst->saturate);
+      dst[0] = translate_dst(t, 
+                             &inst->dst,
+                             inst->saturate);
 
    for (i = 0; i < num_src; i++) 
-      src[i] = translate_src( t, &inst->src[i] );
+      src[i] = translate_src(t, &inst->src[i]);
 
-   switch( inst->op ) {
+   switch(inst->op) {
    case TGSI_OPCODE_BGNLOOP:
    case TGSI_OPCODE_CAL:
    case TGSI_OPCODE_ELSE:
    case TGSI_OPCODE_ENDLOOP:
    case TGSI_OPCODE_IF:
-      debug_assert(num_dst == 0);
-      ureg_label_insn( ureg,
-                       inst->op,
-                       src, num_src,
-                       get_label( t, 
-                                  inst->op == TGSI_OPCODE_CAL ? inst->function->sig_id : 0 ));
+      assert(num_dst == 0);
+      ureg_label_insn(ureg,
+                      inst->op,
+                      src, num_src,
+                      get_label(t, 
+                                inst->op == TGSI_OPCODE_CAL ? inst->function->sig_id : 0));
       return;
 
    case TGSI_OPCODE_TEX:
@@ -3633,72 +4240,44 @@ compile_tgsi_instruction(struct st_translate *t,
    case TGSI_OPCODE_TXD:
    case TGSI_OPCODE_TXL:
    case TGSI_OPCODE_TXP:
+   case TGSI_OPCODE_TXQ:
+   case TGSI_OPCODE_TXF:
       src[num_src++] = t->samplers[inst->sampler];
-      ureg_tex_insn( ureg,
-                     inst->op,
-                     dst, num_dst, 
-                     translate_texture_target( inst->tex_target,
-                                               inst->tex_shadow ),
-                     src, num_src );
+      for (i = 0; i < inst->tex_offset_num_offset; i++) {
+         texoffsets[i] = translate_tex_offset(t, &inst->tex_offsets[i]);
+      }
+      ureg_tex_insn(ureg,
+                    inst->op,
+                    dst, num_dst, 
+                    translate_texture_target(inst->tex_target, inst->tex_shadow),
+                    texoffsets, inst->tex_offset_num_offset,
+                    src, num_src);
       return;
 
    case TGSI_OPCODE_SCS:
-      dst[0] = ureg_writemask(dst[0], TGSI_WRITEMASK_XY );
-      ureg_insn( ureg, 
-                 inst->op, 
-                 dst, num_dst, 
-                 src, num_src );
-      break;
-
-   case TGSI_OPCODE_XPD:
-      dst[0] = ureg_writemask(dst[0], TGSI_WRITEMASK_XYZ );
-      ureg_insn( ureg, 
-                 inst->op, 
-                 dst, num_dst, 
-                 src, num_src );
+      dst[0] = ureg_writemask(dst[0], TGSI_WRITEMASK_XY);
+      ureg_insn(ureg, inst->op, dst, num_dst, src, num_src);
       break;
 
    default:
-      ureg_insn( ureg, 
-                 inst->op, 
-                 dst, num_dst, 
-                 src, num_src );
+      ureg_insn(ureg,
+                inst->op,
+                dst, num_dst,
+                src, num_src);
       break;
    }
 }
 
 /**
- * Emit the TGSI instructions to adjust the WPOS pixel center convention
- * Basically, add (adjX, adjY) to the fragment position.
- */
-static void
-emit_adjusted_wpos( struct st_translate *t,
-                    const struct gl_program *program,
-                    GLfloat adjX, GLfloat adjY)
-{
-   struct ureg_program *ureg = t->ureg;
-   struct ureg_dst wpos_temp = ureg_DECL_temporary(ureg);
-   struct ureg_src wpos_input = t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]];
-
-   /* Note that we bias X and Y and pass Z and W through unchanged.
-    * The shader might also use gl_FragCoord.w and .z.
-    */
-   ureg_ADD(ureg, wpos_temp, wpos_input,
-            ureg_imm4f(ureg, adjX, adjY, 0.0f, 0.0f));
-
-   t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]] = ureg_src(wpos_temp);
-}
-
-
-/**
- * Emit the TGSI instructions for inverting the WPOS y coordinate.
+ * Emit the TGSI instructions for inverting and adjusting WPOS.
  * This code is unavoidable because it also depends on whether
  * a FBO is bound (STATE_FB_WPOS_Y_TRANSFORM).
  */
 static void
-emit_wpos_inversion( struct st_translate *t,
-                     const struct gl_program *program,
-                     boolean invert)
+emit_wpos_adjustment( struct st_translate *t,
+                      const struct gl_program *program,
+                      boolean invert,
+                      GLfloat adjX, GLfloat adjY[2])
 {
    struct ureg_program *ureg = t->ureg;
 
@@ -3718,18 +4297,38 @@ emit_wpos_inversion( struct st_translate *t,
                                                        wposTransformState);
 
    struct ureg_src wpostrans = ureg_DECL_constant( ureg, wposTransConst );
-   struct ureg_dst wpos_temp;
+   struct ureg_dst wpos_temp = ureg_DECL_temporary( ureg );
    struct ureg_src wpos_input = t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]];
 
-   /* MOV wpos_temp, input[wpos]
-    */
-   if (wpos_input.File == TGSI_FILE_TEMPORARY)
-      wpos_temp = ureg_dst(wpos_input);
-   else {
-      wpos_temp = ureg_DECL_temporary( ureg );
+   /* First, apply the coordinate shift: */
+   if (adjX || adjY[0] || adjY[1]) {
+      if (adjY[0] != adjY[1]) {
+         /* Adjust the y coordinate by adjY[1] or adjY[0] respectively
+          * depending on whether inversion is actually going to be applied
+          * or not, which is determined by testing against the inversion
+          * state variable used below, which will be either +1 or -1.
+          */
+         struct ureg_dst adj_temp = ureg_DECL_temporary(ureg);
+
+         ureg_CMP(ureg, adj_temp,
+                  ureg_scalar(wpostrans, invert ? 2 : 0),
+                  ureg_imm4f(ureg, adjX, adjY[0], 0.0f, 0.0f),
+                  ureg_imm4f(ureg, adjX, adjY[1], 0.0f, 0.0f));
+         ureg_ADD(ureg, wpos_temp, wpos_input, ureg_src(adj_temp));
+      } else {
+         ureg_ADD(ureg, wpos_temp, wpos_input,
+                  ureg_imm4f(ureg, adjX, adjY[0], 0.0f, 0.0f));
+      }
+      wpos_input = ureg_src(wpos_temp);
+   } else {
+      /* MOV wpos_temp, input[wpos]
+       */
       ureg_MOV( ureg, wpos_temp, wpos_input );
    }
 
+   /* Now the conditional y flip: STATE_FB_WPOS_Y_TRANSFORM.xy/zw will be
+    * inversion/identity, or the other way around if we're drawing to an FBO.
+    */
    if (invert) {
       /* MAD wpos_temp.y, wpos_input, wpostrans.xxxx, wpostrans.yyyy
        */
@@ -3766,8 +4365,37 @@ emit_wpos(struct st_context *st,
    const struct gl_fragment_program *fp =
       (const struct gl_fragment_program *) program;
    struct pipe_screen *pscreen = st->pipe->screen;
+   GLfloat adjX = 0.0f;
+   GLfloat adjY[2] = { 0.0f, 0.0f };
    boolean invert = FALSE;
 
+   /* Query the pixel center conventions supported by the pipe driver and set
+    * adjX, adjY to help out if it cannot handle the requested one internally.
+    *
+    * The bias of the y-coordinate depends on whether y-inversion takes place
+    * (adjY[1]) or not (adjY[0]), which is in turn dependent on whether we are
+    * drawing to an FBO (causes additional inversion), and whether the the pipe
+    * driver origin and the requested origin differ (the latter condition is
+    * stored in the 'invert' variable).
+    *
+    * For height = 100 (i = integer, h = half-integer, l = lower, u = upper):
+    *
+    * center shift only:
+    * i -> h: +0.5
+    * h -> i: -0.5
+    *
+    * inversion only:
+    * l,i -> u,i: ( 0.0 + 1.0) * -1 + 100 = 99
+    * l,h -> u,h: ( 0.5 + 0.0) * -1 + 100 = 99.5
+    * u,i -> l,i: (99.0 + 1.0) * -1 + 100 = 0
+    * u,h -> l,h: (99.5 + 0.0) * -1 + 100 = 0.5
+    *
+    * inversion and center shift:
+    * l,i -> u,h: ( 0.0 + 0.5) * -1 + 100 = 99.5
+    * l,h -> u,i: ( 0.5 + 0.5) * -1 + 100 = 99
+    * u,i -> l,h: (99.0 + 0.5) * -1 + 100 = 0.5
+    * u,h -> l,i: (99.5 + 0.5) * -1 + 100 = 0
+    */
    if (fp->OriginUpperLeft) {
       /* Fragment shader wants origin in upper-left */
       if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT)) {
@@ -3795,12 +4423,17 @@ emit_wpos(struct st_context *st,
    
    if (fp->PixelCenterInteger) {
       /* Fragment shader wants pixel center integer */
-      if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER))
+      if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER)) {
          /* the driver supports pixel center integer */
+         adjY[1] = 1.0f;
          ureg_property_fs_coord_pixel_center(ureg, TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
-      else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER))
+      }
+      else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER)) {
          /* the driver supports pixel center half integer, need to bias X,Y */
-         emit_adjusted_wpos(t, program, 0.5f, invert ? 0.5f : -0.5f);
+         adjX = -0.5f;
+         adjY[0] = -0.5f;
+         adjY[1] = 0.5f;
+      }
       else
          assert(0);
    }
@@ -3811,8 +4444,8 @@ emit_wpos(struct st_context *st,
       }
       else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER)) {
          /* the driver supports pixel center integer, need to bias X,Y */
+         adjX = adjY[0] = adjY[1] = 0.5f;
          ureg_property_fs_coord_pixel_center(ureg, TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
-         emit_adjusted_wpos(t, program, 0.5f, invert ? -0.5f : 0.5f);
       }
       else
          assert(0);
@@ -3820,7 +4453,7 @@ emit_wpos(struct st_context *st,
 
    /* we invert after adjustment so that we avoid the MOV to temporary,
     * and reuse the adjustment ADD instead */
-   emit_wpos_inversion(t, program, invert);
+   emit_wpos_adjustment(t, program, invert, adjX, adjY);
 }
 
 /**
@@ -3889,16 +4522,21 @@ st_translate_program(
    const GLuint outputMapping[],
    const ubyte outputSemanticName[],
    const ubyte outputSemanticIndex[],
-   boolean passthrough_edgeflags )
+   boolean passthrough_edgeflags)
 {
-   struct st_translate translate, *t;
+   struct st_translate *t;
    unsigned i;
    enum pipe_error ret = PIPE_OK;
 
    assert(numInputs <= Elements(t->inputs));
    assert(numOutputs <= Elements(t->outputs));
 
-   t = &translate;
+   t = CALLOC_STRUCT(st_translate);
+   if (!t) {
+      ret = PIPE_ERROR_OUT_OF_MEMORY;
+      goto out;
+   }
+
    memset(t, 0, sizeof *t);
 
    t->procType = procType;
@@ -3935,28 +4573,26 @@ st_translate_program(
       for (i = 0; i < numOutputs; i++) {
          switch (outputSemanticName[i]) {
          case TGSI_SEMANTIC_POSITION:
-            t->outputs[i] = ureg_DECL_output( ureg,
-                                              TGSI_SEMANTIC_POSITION, /* Z / Depth */
-                                              outputSemanticIndex[i] );
-
-            t->outputs[i] = ureg_writemask( t->outputs[i],
-                                            TGSI_WRITEMASK_Z );
+            t->outputs[i] = ureg_DECL_output(ureg,
+                                             TGSI_SEMANTIC_POSITION, /* Z/Depth */
+                                             outputSemanticIndex[i]);
+            t->outputs[i] = ureg_writemask(t->outputs[i], TGSI_WRITEMASK_Z);
             break;
          case TGSI_SEMANTIC_STENCIL:
-            t->outputs[i] = ureg_DECL_output( ureg,
-                                              TGSI_SEMANTIC_STENCIL, /* Stencil */
-                                              outputSemanticIndex[i] );
-            t->outputs[i] = ureg_writemask( t->outputs[i],
-                                            TGSI_WRITEMASK_Y );
+            t->outputs[i] = ureg_DECL_output(ureg,
+                                             TGSI_SEMANTIC_STENCIL, /* Stencil */
+                                             outputSemanticIndex[i]);
+            t->outputs[i] = ureg_writemask(t->outputs[i], TGSI_WRITEMASK_Y);
             break;
          case TGSI_SEMANTIC_COLOR:
-            t->outputs[i] = ureg_DECL_output( ureg,
-                                              TGSI_SEMANTIC_COLOR,
-                                              outputSemanticIndex[i] );
+            t->outputs[i] = ureg_DECL_output(ureg,
+                                             TGSI_SEMANTIC_COLOR,
+                                             outputSemanticIndex[i]);
             break;
          default:
-            debug_assert(0);
-            return PIPE_ERROR_BAD_INPUT;
+            assert(!"fragment shader outputs must be POSITION/STENCIL/COLOR");
+            ret = PIPE_ERROR_BAD_INPUT;
+            goto out;
          }
       }
    }
@@ -3969,9 +4605,9 @@ st_translate_program(
       }
 
       for (i = 0; i < numOutputs; i++) {
-         t->outputs[i] = ureg_DECL_output( ureg,
-                                           outputSemanticName[i],
-                                           outputSemanticIndex[i] );
+         t->outputs[i] = ureg_DECL_output(ureg,
+                                          outputSemanticName[i],
+                                          outputSemanticIndex[i]);
       }
    }
    else {
@@ -3982,9 +4618,9 @@ st_translate_program(
       }
 
       for (i = 0; i < numOutputs; i++) {
-         t->outputs[i] = ureg_DECL_output( ureg,
-                                           outputSemanticName[i],
-                                           outputSemanticIndex[i] );
+         t->outputs[i] = ureg_DECL_output(ureg,
+                                          outputSemanticName[i],
+                                          outputSemanticIndex[i]);
          if ((outputSemanticName[i] == TGSI_SEMANTIC_PSIZE) && proginfo->Id) {
             /* Writing to the point size result register requires special
              * handling to implement clamping.
@@ -3998,8 +4634,8 @@ st_translate_program(
             unsigned pointSizeClampConst =
                _mesa_add_state_reference(proginfo->Parameters,
                                          pointSizeClampState);
-            struct ureg_dst psizregtemp = ureg_DECL_temporary( ureg );
-            t->pointSizeConst = ureg_DECL_constant( ureg, pointSizeClampConst );
+            struct ureg_dst psizregtemp = ureg_DECL_temporary(ureg);
+            t->pointSizeConst = ureg_DECL_constant(ureg, pointSizeClampConst);
             t->pointSizeResult = t->outputs[i];
             t->pointSizeOutIndex = i;
             t->outputs[i] = psizregtemp;
@@ -4012,8 +4648,8 @@ st_translate_program(
    /* Declare address register.
     */
    if (program->num_address_regs > 0) {
-      debug_assert( program->num_address_regs == 1 );
-      t->address[0] = ureg_DECL_address( ureg );
+      assert(program->num_address_regs == 1);
+      t->address[0] = ureg_DECL_address(ureg);
    }
 
    /* Declare misc input registers
@@ -4038,16 +4674,15 @@ st_translate_program(
        */
       for (i = 0; i < (unsigned)program->next_temp; i++) {
          /* XXX use TGSI_FILE_TEMPORARY_ARRAY when it's supported by ureg */
-         t->temps[i] = ureg_DECL_temporary( t->ureg );
+         t->temps[i] = ureg_DECL_temporary(t->ureg);
       }
    }
 
-   /* Emit constants and immediates.  Mesa uses a single index space
-    * for these, so we put all the translated regs in t->constants.
-    * XXX: this entire if block depends on proginfo->Parameters from Mesa IR
+   /* Emit constants and uniforms.  TGSI uses a single index space for these, 
+    * so we put all the translated regs in t->constants.
     */
    if (proginfo->Parameters) {
-      t->constants = (struct ureg_src *)CALLOC( proginfo->Parameters->NumParameters * sizeof t->constants[0] );
+      t->constants = (struct ureg_src *)CALLOC(proginfo->Parameters->NumParameters * sizeof(t->constants[0]));
       if (t->constants == NULL) {
          ret = PIPE_ERROR_OUT_OF_MEMORY;
          goto out;
@@ -4060,65 +4695,55 @@ st_translate_program(
          case PROGRAM_STATE_VAR:
          case PROGRAM_NAMED_PARAM:
          case PROGRAM_UNIFORM:
-            t->constants[i] = ureg_DECL_constant( ureg, i );
+            t->constants[i] = ureg_DECL_constant(ureg, i);
             break;
 
-            /* Emit immediates only when there's no indirect addressing of
-             * the const buffer.
-             * FIXME: Be smarter and recognize param arrays:
-             * indirect addressing is only valid within the referenced
-             * array.
-             */
+         /* Emit immediates for PROGRAM_CONSTANT only when there's no indirect
+          * addressing of the const buffer.
+          * FIXME: Be smarter and recognize param arrays:
+          * indirect addressing is only valid within the referenced
+          * array.
+          */
          case PROGRAM_CONSTANT:
             if (program->indirect_addr_consts)
-               t->constants[i] = ureg_DECL_constant( ureg, i );
+               t->constants[i] = ureg_DECL_constant(ureg, i);
             else
-               switch(proginfo->Parameters->Parameters[i].DataType)
-               {
-               case GL_FLOAT:
-               case GL_FLOAT_VEC2:
-               case GL_FLOAT_VEC3:
-               case GL_FLOAT_VEC4:
-                  t->constants[i] = ureg_DECL_immediate(ureg, (float *)proginfo->Parameters->ParameterValues[i], 4);
-                  break;
-               case GL_INT:
-               case GL_INT_VEC2:
-               case GL_INT_VEC3:
-               case GL_INT_VEC4:
-                  t->constants[i] = ureg_DECL_immediate_int(ureg, (int *)proginfo->Parameters->ParameterValues[i], 4);
-                  break;
-               case GL_UNSIGNED_INT:
-               case GL_UNSIGNED_INT_VEC2:
-               case GL_UNSIGNED_INT_VEC3:
-               case GL_UNSIGNED_INT_VEC4:
-               case GL_BOOL:
-               case GL_BOOL_VEC2:
-               case GL_BOOL_VEC3:
-               case GL_BOOL_VEC4:
-                  t->constants[i] = ureg_DECL_immediate_uint(ureg, (unsigned *)proginfo->Parameters->ParameterValues[i], 4);
-                  break;
-               default:
-                  assert(!"should not get here");
-               }
+               t->constants[i] = emit_immediate(t,
+                                                proginfo->Parameters->ParameterValues[i],
+                                                proginfo->Parameters->Parameters[i].DataType,
+                                                4);
             break;
          default:
             break;
          }
       }
    }
+   
+   /* Emit immediate values.
+    */
+   t->immediates = (struct ureg_src *)CALLOC(program->num_immediates * sizeof(struct ureg_src));
+   if (t->immediates == NULL) {
+      ret = PIPE_ERROR_OUT_OF_MEMORY;
+      goto out;
+   }
+   i = 0;
+   foreach_iter(exec_list_iterator, iter, program->immediates) {
+      immediate_storage *imm = (immediate_storage *)iter.get();
+      t->immediates[i++] = emit_immediate(t, imm->values, imm->type, imm->size);
+   }
 
    /* texture samplers */
    for (i = 0; i < ctx->Const.MaxTextureImageUnits; i++) {
       if (program->samplers_used & (1 << i)) {
-         t->samplers[i] = ureg_DECL_sampler( ureg, i );
+         t->samplers[i] = ureg_DECL_sampler(ureg, i);
       }
    }
 
    /* Emit each instruction in turn:
     */
    foreach_iter(exec_list_iterator, iter, program->instructions) {
-      set_insn_start( t, ureg_get_instruction_number( ureg ));
-      compile_tgsi_instruction( t, (glsl_to_tgsi_instruction *)iter.get() );
+      set_insn_start(t, ureg_get_instruction_number(ureg));
+      compile_tgsi_instruction(t, (glsl_to_tgsi_instruction *)iter.get());
 
       if (t->prevInstWrotePointSize && proginfo->Id) {
          /* The previous instruction wrote to the (fake) vertex point size
@@ -4128,14 +4753,14 @@ st_translate_program(
           * Note that we can't do this easily at the end of program due to
           * possible early return.
           */
-         set_insn_start( t, ureg_get_instruction_number( ureg ));
-         ureg_MAX( t->ureg,
-                   ureg_writemask(t->outputs[t->pointSizeOutIndex], WRITEMASK_X),
-                   ureg_src(t->outputs[t->pointSizeOutIndex]),
-                   ureg_swizzle(t->pointSizeConst, 1,1,1,1));
-         ureg_MIN( t->ureg, ureg_writemask(t->pointSizeResult, WRITEMASK_X),
-                   ureg_src(t->outputs[t->pointSizeOutIndex]),
-                   ureg_swizzle(t->pointSizeConst, 2,2,2,2));
+         set_insn_start(t, ureg_get_instruction_number(ureg));
+         ureg_MAX(t->ureg,
+                  ureg_writemask(t->outputs[t->pointSizeOutIndex], WRITEMASK_X),
+                  ureg_src(t->outputs[t->pointSizeOutIndex]),
+                  ureg_swizzle(t->pointSizeConst, 1,1,1,1));
+         ureg_MIN(t->ureg, ureg_writemask(t->pointSizeResult, WRITEMASK_X),
+                  ureg_src(t->outputs[t->pointSizeOutIndex]),
+                  ureg_swizzle(t->pointSizeConst, 2,2,2,2));
       }
       t->prevInstWrotePointSize = GL_FALSE;
    }
@@ -4143,18 +4768,22 @@ st_translate_program(
    /* Fix up all emitted labels:
     */
    for (i = 0; i < t->labels_count; i++) {
-      ureg_fixup_label( ureg,
-                        t->labels[i].token,
-                        t->insn[t->labels[i].branch_target] );
+      ureg_fixup_label(ureg, t->labels[i].token,
+                       t->insn[t->labels[i].branch_target]);
    }
 
 out:
-   FREE(t->insn);
-   FREE(t->labels);
-   FREE(t->constants);
+   if (t) {
+      FREE(t->insn);
+      FREE(t->labels);
+      FREE(t->constants);
+      FREE(t->immediates);
+
+      if (t->error) {
+         debug_printf("%s: translate error flag set\n", __FUNCTION__);
+      }
 
-   if (t->error) {
-      debug_printf("%s: translate error flag set\n", __FUNCTION__);
+      FREE(t);
    }
 
    return ret;
@@ -4172,9 +4801,11 @@ get_mesa_program(struct gl_context *ctx,
 {
    glsl_to_tgsi_visitor* v = new glsl_to_tgsi_visitor();
    struct gl_program *prog;
+   struct pipe_screen * screen = st_context(ctx)->pipe->screen;
+   unsigned pipe_shader_type;
    GLenum target;
    const char *target_string;
-   GLboolean progress;
+   bool progress;
    struct gl_shader_compiler_options *options =
          &ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(shader->Type)];
 
@@ -4182,14 +4813,17 @@ get_mesa_program(struct gl_context *ctx,
    case GL_VERTEX_SHADER:
       target = GL_VERTEX_PROGRAM_ARB;
       target_string = "vertex";
+      pipe_shader_type = PIPE_SHADER_VERTEX;
       break;
    case GL_FRAGMENT_SHADER:
       target = GL_FRAGMENT_PROGRAM_ARB;
       target_string = "fragment";
+      pipe_shader_type = PIPE_SHADER_FRAGMENT;
       break;
    case GL_GEOMETRY_SHADER:
       target = GL_GEOMETRY_PROGRAM_NV;
       target_string = "geometry";
+      pipe_shader_type = PIPE_SHADER_GEOMETRY;
       break;
    default:
       assert(!"should not be reached");
@@ -4202,15 +4836,15 @@ get_mesa_program(struct gl_context *ctx,
    if (!prog)
       return NULL;
    prog->Parameters = _mesa_new_parameter_list();
-   prog->Varying = _mesa_new_parameter_list();
-   prog->Attributes = _mesa_new_parameter_list();
    v->ctx = ctx;
    v->prog = prog;
    v->shader_program = shader_program;
    v->options = options;
    v->glsl_version = ctx->Const.GLSLVersion;
+   v->native_integers = ctx->Const.NativeIntegers;
 
-   add_uniforms_to_parameters_list(shader_program, shader, prog);
+   _mesa_generate_parameters_list_for_uniforms(shader_program, shader,
+					       prog->Parameters);
 
    /* Emit intermediate IR for main(). */
    visit_exec_list(shader->ir, v);
@@ -4258,22 +4892,25 @@ get_mesa_program(struct gl_context *ctx,
    }
 #endif
 
-   /* Remove reads to output registers, and to varyings in vertex shaders. */
-   v->remove_output_reads(PROGRAM_OUTPUT);
-   if (target == GL_VERTEX_PROGRAM_ARB)
-      v->remove_output_reads(PROGRAM_VARYING);
+   if (!screen->get_shader_param(screen, pipe_shader_type,
+                                 PIPE_SHADER_CAP_OUTPUT_READ)) {
+      /* Remove reads to output registers, and to varyings in vertex shaders. */
+      v->remove_output_reads(PROGRAM_OUTPUT);
+      if (target == GL_VERTEX_PROGRAM_ARB)
+         v->remove_output_reads(PROGRAM_VARYING);
+   }
    
-   /* Perform the simplify_cmp optimization, which is required by r300g. */
+   /* Perform optimizations on the instructions in the glsl_to_tgsi_visitor. */
    v->simplify_cmp();
+   v->copy_propagate();
+   while (v->eliminate_dead_code_advanced());
 
-   /* Perform optimizations on the instructions in the glsl_to_tgsi_visitor.
-    * FIXME: These passes to optimize temporary registers don't work when there
+   /* FIXME: These passes to optimize temporary registers don't work when there
     * is indirect addressing of the temporary register space.  We need proper 
     * array support so that we don't have to give up these passes in every 
     * shader that uses arrays.
     */
    if (!v->indirect_addr_temps) {
-      v->copy_propagate();
       v->eliminate_dead_code();
       v->merge_registers();
       v->renumber_registers();
@@ -4289,18 +4926,26 @@ get_mesa_program(struct gl_context *ctx,
       _mesa_print_ir(shader->ir, NULL);
       printf("\n");
       printf("\n");
+      fflush(stdout);
    }
 
    prog->Instructions = NULL;
    prog->NumInstructions = 0;
 
-   do_set_program_inouts(shader->ir, prog);
+   do_set_program_inouts(shader->ir, prog, shader->Type == GL_FRAGMENT_SHADER);
    count_resources(v, prog);
 
-   check_resources(ctx, shader_program, v, prog);
-
    _mesa_reference_program(ctx, &shader->Program, prog);
    
+   /* This has to be done last.  Any operation the can cause
+    * prog->ParameterValues to get reallocated (e.g., anything that adds a
+    * program constant) has to happen before creating this linkage.
+    */
+   _mesa_associate_uniform_storage(ctx, shader_program, prog->Parameters);
+   if (!shader_program->LinkStatus) {
+      return NULL;
+   }
+
    struct st_vertex_program *stvp;
    struct st_fragment_program *stfp;
    struct st_geometry_program *stgp;
@@ -4381,19 +5026,21 @@ st_link_shader(struct gl_context *ctx, struct gl_shader_program *prog)
          /* Lowering */
          do_mat_op_to_vec(ir);
          lower_instructions(ir, (MOD_TO_FRACT | DIV_TO_MUL_RCP | EXP_TO_EXP2
-        			 | LOG_TO_LOG2
+				 | LOG_TO_LOG2 | INT_DIV_TO_MUL_RCP
         			 | ((options->EmitNoPow) ? POW_TO_EXP2 : 0)));
 
          progress = do_lower_jumps(ir, true, true, options->EmitNoMainReturn, options->EmitNoCont, options->EmitNoLoops) || progress;
 
-         progress = do_common_optimization(ir, true, options->MaxUnrollIterations) || progress;
+         progress = do_common_optimization(ir, true, true,
+					   options->MaxUnrollIterations)
+	   || progress;
 
-         progress = lower_quadop_vector(ir, true) || progress;
+         progress = lower_quadop_vector(ir, false) || progress;
 
-         if (options->EmitNoIfs) {
+         if (options->MaxIfDepth == 0)
             progress = lower_discard(ir) || progress;
-            progress = lower_if_to_cond_assign(ir) || progress;
-         }
+
+         progress = lower_if_to_cond_assign(ir, options->MaxIfDepth) || progress;
 
          if (options->EmitNoNoise)
             progress = lower_noise(ir) || progress;
@@ -4426,29 +5073,18 @@ st_link_shader(struct gl_context *ctx, struct gl_shader_program *prog)
       linked_prog = get_mesa_program(ctx, prog, prog->_LinkedShaders[i]);
 
       if (linked_prog) {
-         bool ok = true;
-
-         switch (prog->_LinkedShaders[i]->Type) {
-         case GL_VERTEX_SHADER:
-            _mesa_reference_vertprog(ctx, &prog->VertexProgram,
-                                     (struct gl_vertex_program *)linked_prog);
-            ok = ctx->Driver.ProgramStringNotify(ctx, GL_VERTEX_PROGRAM_ARB,
-                                                 linked_prog);
-            break;
-         case GL_FRAGMENT_SHADER:
-            _mesa_reference_fragprog(ctx, &prog->FragmentProgram,
-                                     (struct gl_fragment_program *)linked_prog);
-            ok = ctx->Driver.ProgramStringNotify(ctx, GL_FRAGMENT_PROGRAM_ARB,
-                                                 linked_prog);
-            break;
-         case GL_GEOMETRY_SHADER:
-            _mesa_reference_geomprog(ctx, &prog->GeometryProgram,
-                                     (struct gl_geometry_program *)linked_prog);
-            ok = ctx->Driver.ProgramStringNotify(ctx, GL_GEOMETRY_PROGRAM_NV,
-                                                 linked_prog);
-            break;
-         }
-         if (!ok) {
+	 static const GLenum targets[] = {
+	    GL_VERTEX_PROGRAM_ARB,
+	    GL_FRAGMENT_PROGRAM_ARB,
+	    GL_GEOMETRY_PROGRAM_NV
+	 };
+
+	 _mesa_reference_program(ctx, &prog->_LinkedShaders[i]->Program,
+				 linked_prog);
+         if (!ctx->Driver.ProgramStringNotify(ctx, targets[i], linked_prog)) {
+	    _mesa_reference_program(ctx, &prog->_LinkedShaders[i]->Program,
+				    NULL);
+            _mesa_reference_program(ctx, &linked_prog, NULL);
             return GL_FALSE;
          }
       }
@@ -4459,53 +5095,4 @@ st_link_shader(struct gl_context *ctx, struct gl_shader_program *prog)
    return GL_TRUE;
 }
 
-
-/**
- * Link a GLSL shader program.  Called via glLinkProgram().
- */
-void
-st_glsl_link_shader(struct gl_context *ctx, struct gl_shader_program *prog)
-{
-   unsigned int i;
-
-   _mesa_clear_shader_program_data(ctx, prog);
-
-   prog->LinkStatus = GL_TRUE;
-
-   for (i = 0; i < prog->NumShaders; i++) {
-      if (!prog->Shaders[i]->CompileStatus) {
-         fail_link(prog, "linking with uncompiled shader");
-         prog->LinkStatus = GL_FALSE;
-      }
-   }
-
-   prog->Varying = _mesa_new_parameter_list();
-   _mesa_reference_vertprog(ctx, &prog->VertexProgram, NULL);
-   _mesa_reference_fragprog(ctx, &prog->FragmentProgram, NULL);
-   _mesa_reference_geomprog(ctx, &prog->GeometryProgram, NULL);
-
-   if (prog->LinkStatus) {
-      link_shaders(ctx, prog);
-   }
-
-   if (prog->LinkStatus) {
-      if (!ctx->Driver.LinkShader(ctx, prog)) {
-         prog->LinkStatus = GL_FALSE;
-      }
-   }
-
-   set_uniform_initializers(ctx, prog);
-
-   if (ctx->Shader.Flags & GLSL_DUMP) {
-      if (!prog->LinkStatus) {
-         printf("GLSL shader program %d failed to link\n", prog->Name);
-      }
-
-      if (prog->InfoLog && prog->InfoLog[0] != 0) {
-         printf("GLSL shader program %d info log:\n", prog->Name);
-         printf("%s\n", prog->InfoLog);
-      }
-   }
-}
-
 } /* extern "C" */