Merge remote-tracking branch 'origin/released'

13 files changed, 543 insertions, 26 deletions

diff --git a/mesalib/src/glsl/Makefile.sources b/mesalib/src/glsl/Makefile.sources
index f2743f750..765f06a27 100644
--- a/mesalib/src/glsl/Makefile.sources
+++ b/mesalib/src/glsl/Makefile.sources
@@ -66,6 +66,7 @@ LIBGLSL_CXX_FILES = \
 	$(GLSL_SRCDIR)/lower_vec_index_to_swizzle.cpp \
 	$(GLSL_SRCDIR)/lower_vector.cpp \
 	$(GLSL_SRCDIR)/lower_output_reads.cpp \
+	$(GLSL_SRCDIR)/lower_ubo_reference.cpp \
 	$(GLSL_SRCDIR)/opt_algebraic.cpp \
 	$(GLSL_SRCDIR)/opt_array_splitting.cpp \
 	$(GLSL_SRCDIR)/opt_constant_folding.cpp \
diff --git a/mesalib/src/glsl/README b/mesalib/src/glsl/README
index dd80a53d4..0a0afccdc 100644
--- a/mesalib/src/glsl/README
+++ b/mesalib/src/glsl/README
@@ -177,7 +177,6 @@ ir_unop_fract was added.  The following areas need updating to add a
 new expression type:
 
 ir.h (new enum)
-ir.cpp:get_num_operands() (used for ir_reader)
 ir.cpp:operator_strs (used for ir_reader)
 ir_constant_expression.cpp (you probably want to be able to constant fold)
 ir_validate.cpp (check users have the right types)
diff --git a/mesalib/src/glsl/ast_to_hir.cpp b/mesalib/src/glsl/ast_to_hir.cpp
index 1c54991cf..02fe66b60 100644
--- a/mesalib/src/glsl/ast_to_hir.cpp
+++ b/mesalib/src/glsl/ast_to_hir.cpp
@@ -4054,11 +4054,15 @@ ast_uniform_block::hir(exec_list *instructions,
 	 ubo_var->Type = var->type;
 	 ubo_var->Buffer = ubo - state->uniform_blocks;
 	 ubo_var->Offset = 0; /* Assigned at link time. */
-	 ubo_var->RowMajor = block_row_major;
-	 if (decl_list->type->qualifier.flags.q.row_major)
-	    ubo_var->RowMajor = true;
-	 else if (decl_list->type->qualifier.flags.q.column_major)
-	    ubo_var->RowMajor = false;
+
+	 if (var->type->is_matrix() ||
+	     (var->type->is_array() && var->type->fields.array->is_matrix())) {
+	    ubo_var->RowMajor = block_row_major;
+	    if (decl_list->type->qualifier.flags.q.row_major)
+	       ubo_var->RowMajor = true;
+	    else if (decl_list->type->qualifier.flags.q.column_major)
+	       ubo_var->RowMajor = false;
+	 }
 
 	 /* From the GL_ARB_uniform_buffer_object spec:
 	  *
diff --git a/mesalib/src/glsl/glsl_parser.yy b/mesalib/src/glsl/glsl_parser.yy
index 04c64f096..ee6a67288 100644
--- a/mesalib/src/glsl/glsl_parser.yy
+++ b/mesalib/src/glsl/glsl_parser.yy
@@ -1929,6 +1929,16 @@ uniform_block:
 	   void *ctx = state;
 	   $$ = new(ctx) ast_uniform_block(*state->default_uniform_qualifier,
 					   $2, $4);
+
+	   if (!state->ARB_uniform_buffer_object_enable) {
+	      _mesa_glsl_error(& @1, state,
+			       "#version 140 / GL_ARB_uniform_buffer_object "
+			       "required for defining uniform blocks\n");
+	   } else if (state->ARB_uniform_buffer_object_warn) {
+	      _mesa_glsl_warning(& @1, state,
+				 "#version 140 / GL_ARB_uniform_buffer_object "
+				 "required for defining uniform blocks\n");
+	   }
 	}
 	| layout_qualifier UNIFORM NEW_IDENTIFIER '{' member_list '}' ';'
 	{
@@ -1939,6 +1949,16 @@ uniform_block:
 	      YYERROR;
 	   }
 	   $$ = new(ctx) ast_uniform_block(qual, $3, $5);
+
+	   if (!state->ARB_uniform_buffer_object_enable) {
+	      _mesa_glsl_error(& @1, state,
+			       "#version 140 / GL_ARB_uniform_buffer_object "
+			       "required for defining uniform blocks\n");
+	   } else if (state->ARB_uniform_buffer_object_warn) {
+	      _mesa_glsl_warning(& @1, state,
+				 "#version 140 / GL_ARB_uniform_buffer_object "
+				 "required for defining uniform blocks\n");
+	   }
 	}
 	;
 
diff --git a/mesalib/src/glsl/glsl_types.cpp b/mesalib/src/glsl/glsl_types.cpp
index 3d7866058..2aa51f0b3 100644
--- a/mesalib/src/glsl/glsl_types.cpp
+++ b/mesalib/src/glsl/glsl_types.cpp
@@ -694,14 +694,19 @@ glsl_type::std140_base_alignment(bool row_major) const
     *     row vectors with <C> components each, according to rule (4).
     */
    if (this->is_matrix()) {
-      const struct glsl_type *vec_type;
+      const struct glsl_type *vec_type, *array_type;
+      int c = this->matrix_columns;
+      int r = this->vector_elements;
+
       if (row_major) {
-	 vec_type = get_instance(GLSL_TYPE_FLOAT, this->vector_elements, 1);
+	 vec_type = get_instance(GLSL_TYPE_FLOAT, c, 1);
+	 array_type = glsl_type::get_array_instance(vec_type, r);
       } else {
-	 vec_type = get_instance(GLSL_TYPE_FLOAT, this->matrix_columns, 1);
+	 vec_type = get_instance(GLSL_TYPE_FLOAT, r, 1);
+	 array_type = glsl_type::get_array_instance(vec_type, c);
       }
 
-      return vec_type->std140_base_alignment(false);
+      return array_type->std140_base_alignment(false);
    }
 
    /* (9) If the member is a structure, the base alignment of the
diff --git a/mesalib/src/glsl/ir.cpp b/mesalib/src/glsl/ir.cpp
index b0e38d820..f59cdd29a 100644
--- a/mesalib/src/glsl/ir.cpp
+++ b/mesalib/src/glsl/ir.cpp
@@ -480,6 +480,7 @@ static const char *const operator_strs[] = {
    "min",
    "max",
    "pow",
+   "ubo_load",
    "vector",
 };
 
diff --git a/mesalib/src/glsl/ir.h b/mesalib/src/glsl/ir.h
index f019837d5..89c516c87 100644
--- a/mesalib/src/glsl/ir.h
+++ b/mesalib/src/glsl/ir.h
@@ -1018,9 +1018,17 @@ enum ir_expression_operation {
    ir_binop_pow,
 
    /**
+    * Load a value the size of a given GLSL type from a uniform block.
+    *
+    * operand0 is the ir_constant uniform block index in the linked shader.
+    * operand1 is a byte offset within the uniform block.
+    */
+   ir_binop_ubo_load,
+
+   /**
     * A sentinel marking the last of the binary operations.
     */
-   ir_last_binop = ir_binop_pow,
+   ir_last_binop = ir_binop_ubo_load,
 
    ir_quadop_vector,
 
diff --git a/mesalib/src/glsl/ir_optimization.h b/mesalib/src/glsl/ir_optimization.h
index c435d7717..2220d511e 100644
--- a/mesalib/src/glsl/ir_optimization.h
+++ b/mesalib/src/glsl/ir_optimization.h
@@ -74,6 +74,7 @@ bool lower_variable_index_to_cond_assign(exec_list *instructions,
 bool lower_quadop_vector(exec_list *instructions, bool dont_lower_swz);
 bool lower_clip_distance(exec_list *instructions);
 void lower_output_reads(exec_list *instructions);
+void lower_ubo_reference(struct gl_shader *shader, exec_list *instructions);
 bool optimize_redundant_jumps(exec_list *instructions);
 bool optimize_split_arrays(exec_list *instructions, bool linked);
 
diff --git a/mesalib/src/glsl/ir_rvalue_visitor.cpp b/mesalib/src/glsl/ir_rvalue_visitor.cpp
index 193bcd2d7..b34a419e8 100644
--- a/mesalib/src/glsl/ir_rvalue_visitor.cpp
+++ b/mesalib/src/glsl/ir_rvalue_visitor.cpp
@@ -36,7 +36,7 @@
 #include "glsl_types.h"
 
 ir_visitor_status
-ir_rvalue_visitor::visit_leave(ir_expression *ir)
+ir_rvalue_base_visitor::rvalue_visit(ir_expression *ir)
 {
    unsigned int operand;
 
@@ -48,7 +48,7 @@ ir_rvalue_visitor::visit_leave(ir_expression *ir)
 }
 
 ir_visitor_status
-ir_rvalue_visitor::visit_leave(ir_texture *ir)
+ir_rvalue_base_visitor::rvalue_visit(ir_texture *ir)
 {
    handle_rvalue(&ir->coordinate);
    handle_rvalue(&ir->projector);
@@ -76,14 +76,14 @@ ir_rvalue_visitor::visit_leave(ir_texture *ir)
 }
 
 ir_visitor_status
-ir_rvalue_visitor::visit_leave(ir_swizzle *ir)
+ir_rvalue_base_visitor::rvalue_visit(ir_swizzle *ir)
 {
    handle_rvalue(&ir->val);
    return visit_continue;
 }
 
 ir_visitor_status
-ir_rvalue_visitor::visit_leave(ir_dereference_array *ir)
+ir_rvalue_base_visitor::rvalue_visit(ir_dereference_array *ir)
 {
    /* The array index is not the target of the assignment, so clear the
     * 'in_assignee' flag.  Restore it after returning from the array index.
@@ -98,14 +98,14 @@ ir_rvalue_visitor::visit_leave(ir_dereference_array *ir)
 }
 
 ir_visitor_status
-ir_rvalue_visitor::visit_leave(ir_dereference_record *ir)
+ir_rvalue_base_visitor::rvalue_visit(ir_dereference_record *ir)
 {
    handle_rvalue(&ir->record);
    return visit_continue;
 }
 
 ir_visitor_status
-ir_rvalue_visitor::visit_leave(ir_assignment *ir)
+ir_rvalue_base_visitor::rvalue_visit(ir_assignment *ir)
 {
    handle_rvalue(&ir->rhs);
    handle_rvalue(&ir->condition);
@@ -114,7 +114,7 @@ ir_rvalue_visitor::visit_leave(ir_assignment *ir)
 }
 
 ir_visitor_status
-ir_rvalue_visitor::visit_leave(ir_call *ir)
+ir_rvalue_base_visitor::rvalue_visit(ir_call *ir)
 {
    foreach_iter(exec_list_iterator, iter, *ir) {
       ir_rvalue *param = (ir_rvalue *)iter.get();
@@ -129,15 +129,124 @@ ir_rvalue_visitor::visit_leave(ir_call *ir)
 }
 
 ir_visitor_status
-ir_rvalue_visitor::visit_leave(ir_return *ir)
+ir_rvalue_base_visitor::rvalue_visit(ir_return *ir)
 {
    handle_rvalue(&ir->value);;
    return visit_continue;
 }
 
 ir_visitor_status
-ir_rvalue_visitor::visit_leave(ir_if *ir)
+ir_rvalue_base_visitor::rvalue_visit(ir_if *ir)
 {
    handle_rvalue(&ir->condition);
    return visit_continue;
 }
+
+
+ir_visitor_status
+ir_rvalue_visitor::visit_leave(ir_expression *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_visitor::visit_leave(ir_texture *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_visitor::visit_leave(ir_swizzle *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_visitor::visit_leave(ir_dereference_array *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_visitor::visit_leave(ir_dereference_record *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_visitor::visit_leave(ir_assignment *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_visitor::visit_leave(ir_call *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_visitor::visit_leave(ir_return *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_visitor::visit_leave(ir_if *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_enter_visitor::visit_enter(ir_expression *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_enter_visitor::visit_enter(ir_texture *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_enter_visitor::visit_enter(ir_swizzle *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_enter_visitor::visit_enter(ir_dereference_array *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_enter_visitor::visit_enter(ir_dereference_record *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_enter_visitor::visit_enter(ir_assignment *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_enter_visitor::visit_enter(ir_call *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_enter_visitor::visit_enter(ir_return *ir)
+{
+   return rvalue_visit(ir);
+}
+
+ir_visitor_status
+ir_rvalue_enter_visitor::visit_enter(ir_if *ir)
+{
+   return rvalue_visit(ir);
+}
diff --git a/mesalib/src/glsl/ir_rvalue_visitor.h b/mesalib/src/glsl/ir_rvalue_visitor.h
index 31a56beb9..2179fa5a8 100644
--- a/mesalib/src/glsl/ir_rvalue_visitor.h
+++ b/mesalib/src/glsl/ir_rvalue_visitor.h
@@ -30,7 +30,22 @@
  * a pointer to each rvalue in the tree.
  */
 
-class ir_rvalue_visitor : public ir_hierarchical_visitor {
+class ir_rvalue_base_visitor : public ir_hierarchical_visitor {
+public:
+   ir_visitor_status rvalue_visit(ir_assignment *);
+   ir_visitor_status rvalue_visit(ir_call *);
+   ir_visitor_status rvalue_visit(ir_dereference_array *);
+   ir_visitor_status rvalue_visit(ir_dereference_record *);
+   ir_visitor_status rvalue_visit(ir_expression *);
+   ir_visitor_status rvalue_visit(ir_if *);
+   ir_visitor_status rvalue_visit(ir_return *);
+   ir_visitor_status rvalue_visit(ir_swizzle *);
+   ir_visitor_status rvalue_visit(ir_texture *);
+
+   virtual void handle_rvalue(ir_rvalue **rvalue) = 0;
+};
+
+class ir_rvalue_visitor : public ir_rvalue_base_visitor {
 public:
 
    virtual ir_visitor_status visit_leave(ir_assignment *);
@@ -42,6 +57,18 @@ public:
    virtual ir_visitor_status visit_leave(ir_return *);
    virtual ir_visitor_status visit_leave(ir_swizzle *);
    virtual ir_visitor_status visit_leave(ir_texture *);
+};
 
-   virtual void handle_rvalue(ir_rvalue **rvalue) = 0;
+class ir_rvalue_enter_visitor : public ir_rvalue_base_visitor {
+public:
+
+   virtual ir_visitor_status visit_enter(ir_assignment *);
+   virtual ir_visitor_status visit_enter(ir_call *);
+   virtual ir_visitor_status visit_enter(ir_dereference_array *);
+   virtual ir_visitor_status visit_enter(ir_dereference_record *);
+   virtual ir_visitor_status visit_enter(ir_expression *);
+   virtual ir_visitor_status visit_enter(ir_if *);
+   virtual ir_visitor_status visit_enter(ir_return *);
+   virtual ir_visitor_status visit_enter(ir_swizzle *);
+   virtual ir_visitor_status visit_enter(ir_texture *);
 };
diff --git a/mesalib/src/glsl/ir_validate.cpp b/mesalib/src/glsl/ir_validate.cpp
index 191d39831..af0b5768a 100644
--- a/mesalib/src/glsl/ir_validate.cpp
+++ b/mesalib/src/glsl/ir_validate.cpp
@@ -423,6 +423,13 @@ ir_validate::visit_leave(ir_expression *ir)
       assert(ir->operands[0]->type == ir->operands[1]->type);
       break;
 
+   case ir_binop_ubo_load:
+      assert(ir->operands[0]->as_constant());
+      assert(ir->operands[0]->type == glsl_type::uint_type);
+
+      assert(ir->operands[1]->type == glsl_type::uint_type);
+      break;
+
    case ir_quadop_vector:
       /* The vector operator collects some number of scalars and generates a
        * vector from them.
diff --git a/mesalib/src/glsl/link_uniforms.cpp b/mesalib/src/glsl/link_uniforms.cpp
index 7c3d74b82..27694b3f3 100644
--- a/mesalib/src/glsl/link_uniforms.cpp
+++ b/mesalib/src/glsl/link_uniforms.cpp
@@ -224,14 +224,24 @@ public:
    }
 
    void set_and_process(struct gl_shader_program *prog,
+			struct gl_shader *shader,
 			ir_variable *var)
    {
       ubo_var = NULL;
       if (var->uniform_block != -1) {
 	 struct gl_uniform_block *block =
-	    &prog->UniformBlocks[var->uniform_block];
+	    &shader->UniformBlocks[var->uniform_block];
+
+	 ubo_block_index = -1;
+	 for (unsigned i = 0; i < prog->NumUniformBlocks; i++) {
+	    if (!strcmp(prog->UniformBlocks[i].Name,
+			shader->UniformBlocks[var->uniform_block].Name)) {
+	       ubo_block_index = i;
+	       break;
+	    }
+	 }
+	 assert(ubo_block_index != -1);
 
-	 ubo_block_index = var->uniform_block;
 	 ubo_var_index = var->location;
 	 ubo_var = &block->Uniforms[var->location];
 	 ubo_byte_offset = ubo_var->Offset;
@@ -490,7 +500,19 @@ link_assign_uniform_block_offsets(struct gl_shader *shader)
 	 ubo_var->Offset = offset;
 	 offset += size;
       }
-      block->UniformBufferSize = offset;
+
+      /* From the GL_ARB_uniform_buffer_object spec:
+       *
+       *     "For uniform blocks laid out according to [std140] rules,
+       *      the minimum buffer object size returned by the
+       *      UNIFORM_BLOCK_DATA_SIZE query is derived by taking the
+       *      offset of the last basic machine unit consumed by the
+       *      last uniform of the uniform block (including any
+       *      end-of-array or end-of-structure padding), adding one,
+       *      and rounding up to the next multiple of the base
+       *      alignment required for a vec4."
+       */
+      block->UniformBufferSize = align(offset, 16);
    }
 }
 
@@ -598,7 +620,7 @@ link_assign_uniform_locations(struct gl_shader_program *prog)
 	 if (strncmp("gl_", var->name, 3) == 0)
 	    continue;
 
-	 parcel.set_and_process(prog, var);
+	 parcel.set_and_process(prog, prog->_LinkedShaders[i], var);
       }
 
       prog->_LinkedShaders[i]->active_samplers = parcel.shader_samplers_used;
diff --git a/mesalib/src/glsl/lower_ubo_reference.cpp b/mesalib/src/glsl/lower_ubo_reference.cpp
new file mode 100644
index 000000000..e8d2c4742
--- /dev/null
+++ b/mesalib/src/glsl/lower_ubo_reference.cpp
@@ -0,0 +1,313 @@
+/*
+ * Copyright © 2012 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+/**
+ * \file lower_ubo_reference.cpp
+ *
+ * IR lower pass to replace dereferences of variables in a uniform
+ * buffer object with usage of ir_binop_ubo_load expressions, each of
+ * which can read data up to the size of a vec4.
+ *
+ * This relieves drivers of the responsibility to deal with tricky UBO
+ * layout issues like std140 structures and row_major matrices on
+ * their own.
+ */
+
+#include "ir.h"
+#include "ir_builder.h"
+#include "ir_rvalue_visitor.h"
+#include "main/macros.h"
+
+using namespace ir_builder;
+
+namespace {
+class lower_ubo_reference_visitor : public ir_rvalue_enter_visitor {
+public:
+   lower_ubo_reference_visitor(struct gl_shader *shader)
+   : shader(shader)
+   {
+   }
+
+   void handle_rvalue(ir_rvalue **rvalue);
+   void emit_ubo_loads(ir_dereference *deref, ir_variable *base_offset,
+		       unsigned int deref_offset);
+   ir_expression *ubo_load(const struct glsl_type *type,
+			   ir_rvalue *offset);
+
+   void *mem_ctx;
+   struct gl_shader *shader;
+   struct gl_uniform_buffer_variable *ubo_var;
+   unsigned uniform_block;
+   bool progress;
+};
+
+static inline unsigned int
+align(unsigned int a, unsigned int align)
+{
+   return (a + align - 1) / align * align;
+}
+
+void
+lower_ubo_reference_visitor::handle_rvalue(ir_rvalue **rvalue)
+{
+   if (!*rvalue)
+      return;
+
+   ir_dereference *deref = (*rvalue)->as_dereference();
+   if (!deref)
+      return;
+
+   ir_variable *var = deref->variable_referenced();
+   if (!var || var->uniform_block == -1)
+      return;
+
+   mem_ctx = ralloc_parent(*rvalue);
+   uniform_block = var->uniform_block;
+   struct gl_uniform_block *block = &shader->UniformBlocks[uniform_block];
+   this->ubo_var = &block->Uniforms[var->location];
+   ir_rvalue *offset = new(mem_ctx) ir_constant(0u);
+   unsigned const_offset = 0;
+   bool row_major = ubo_var->RowMajor;
+
+   /* Calculate the offset to the start of the region of the UBO
+    * dereferenced by *rvalue.  This may be a variable offset if an
+    * array dereference has a variable index.
+    */
+   while (deref) {
+      switch (deref->ir_type) {
+      case ir_type_dereference_variable: {
+	 const_offset += ubo_var->Offset;
+	 deref = NULL;
+	 break;
+      }
+
+      case ir_type_dereference_array: {
+	 ir_dereference_array *deref_array = (ir_dereference_array *)deref;
+	 unsigned array_stride;
+	 if (deref_array->array->type->is_matrix() && row_major) {
+	    /* When loading a vector out of a row major matrix, the
+	     * step between the columns (vectors) is the size of a
+	     * float, while the step between the rows (elements of a
+	     * vector) is handled below in emit_ubo_loads.
+	     */
+	    array_stride = 4;
+	 } else {
+	    array_stride = deref_array->type->std140_size(row_major);
+	    array_stride = align(array_stride, 16);
+	 }
+
+	 ir_constant *const_index = deref_array->array_index->as_constant();
+	 if (const_index) {
+	    const_offset += array_stride * const_index->value.i[0];
+	 } else {
+	    offset = add(offset,
+			 mul(deref_array->array_index,
+			     new(mem_ctx) ir_constant(array_stride)));
+	 }
+	 deref = deref_array->array->as_dereference();
+	 break;
+      }
+
+      case ir_type_dereference_record: {
+	 ir_dereference_record *deref_record = (ir_dereference_record *)deref;
+	 const glsl_type *struct_type = deref_record->record->type;
+	 unsigned intra_struct_offset = 0;
+
+	 unsigned max_field_align = 16;
+	 for (unsigned int i = 0; i < struct_type->length; i++) {
+	    const glsl_type *type = struct_type->fields.structure[i].type;
+	    unsigned field_align = type->std140_base_alignment(row_major);
+	    max_field_align = MAX2(field_align, max_field_align);
+	    intra_struct_offset = align(intra_struct_offset, field_align);
+
+	    if (strcmp(struct_type->fields.structure[i].name,
+		       deref_record->field) == 0)
+	       break;
+	    intra_struct_offset += type->std140_size(row_major);
+	 }
+
+	 const_offset = align(const_offset, max_field_align);
+	 const_offset += intra_struct_offset;
+
+	 deref = deref_record->record->as_dereference();
+	 break;
+      }
+      default:
+	 assert(!"not reached");
+	 deref = NULL;
+	 break;
+      }
+   }
+
+   /* Now that we've calculated the offset to the start of the
+    * dereference, walk over the type and emit loads into a temporary.
+    */
+   const glsl_type *type = (*rvalue)->type;
+   ir_variable *load_var = new(mem_ctx) ir_variable(type,
+						    "ubo_load_temp",
+						    ir_var_temporary);
+   base_ir->insert_before(load_var);
+
+   ir_variable *load_offset = new(mem_ctx) ir_variable(glsl_type::uint_type,
+						       "ubo_load_temp_offset",
+						       ir_var_temporary);
+   base_ir->insert_before(load_offset);
+   base_ir->insert_before(assign(load_offset, offset));
+
+   deref = new(mem_ctx) ir_dereference_variable(load_var);
+   emit_ubo_loads(deref, load_offset, const_offset);
+   *rvalue = deref;
+
+   progress = true;
+}
+
+ir_expression *
+lower_ubo_reference_visitor::ubo_load(const glsl_type *type,
+				      ir_rvalue *offset)
+{
+   return new(mem_ctx)
+      ir_expression(ir_binop_ubo_load,
+		    type,
+		    new(mem_ctx) ir_constant(this->uniform_block),
+		    offset);
+
+}
+
+/**
+ * Takes LHS and emits a series of assignments into its components
+ * from the UBO variable at variable_offset + deref_offset.
+ *
+ * Recursively calls itself to break the deref down to the point that
+ * the ir_binop_ubo_load expressions generated are contiguous scalars
+ * or vectors.
+ */
+void
+lower_ubo_reference_visitor::emit_ubo_loads(ir_dereference *deref,
+					    ir_variable *base_offset,
+					    unsigned int deref_offset)
+{
+   if (deref->type->is_record()) {
+      unsigned int field_offset = 0;
+
+      for (unsigned i = 0; i < deref->type->length; i++) {
+	 const struct glsl_struct_field *field =
+	    &deref->type->fields.structure[i];
+	 ir_dereference *field_deref =
+	    new(mem_ctx) ir_dereference_record(deref->clone(mem_ctx, NULL),
+					       field->name);
+
+	 field_offset =
+	    align(field_offset,
+		  field->type->std140_base_alignment(ubo_var->RowMajor));
+
+	 emit_ubo_loads(field_deref, base_offset, deref_offset + field_offset);
+
+	 field_offset += field->type->std140_size(ubo_var->RowMajor);
+      }
+      return;
+   }
+
+   if (deref->type->is_array()) {
+      unsigned array_stride =
+	 align(deref->type->fields.array->std140_size(ubo_var->RowMajor), 16);
+
+      for (unsigned i = 0; i < deref->type->length; i++) {
+	 ir_constant *element = new(mem_ctx) ir_constant(i);
+	 ir_dereference *element_deref =
+	    new(mem_ctx) ir_dereference_array(deref->clone(mem_ctx, NULL),
+					      element);
+	 emit_ubo_loads(element_deref, base_offset,
+			deref_offset + i * array_stride);
+      }
+      return;
+   }
+
+   if (deref->type->is_matrix()) {
+      for (unsigned i = 0; i < deref->type->matrix_columns; i++) {
+	 ir_constant *col = new(mem_ctx) ir_constant(i);
+	 ir_dereference *col_deref =
+	    new(mem_ctx) ir_dereference_array(deref->clone(mem_ctx, NULL),
+					      col);
+
+	 /* std140 always rounds the stride of arrays (and matrices)
+	  * to a vec4, so matrices are always 16 between columns/rows.
+	  */
+	 emit_ubo_loads(col_deref, base_offset, deref_offset + i * 16);
+      }
+      return;
+   }
+
+   assert(deref->type->is_scalar() ||
+	  deref->type->is_vector());
+
+   if (!ubo_var->RowMajor) {
+      ir_rvalue *offset = add(base_offset,
+			      new(mem_ctx) ir_constant(deref_offset));
+      base_ir->insert_before(assign(deref->clone(mem_ctx, NULL),
+				    ubo_load(deref->type, offset)));
+   } else {
+      /* We're dereffing a column out of a row-major matrix, so we
+       * gather the vector from each stored row.
+      */
+      assert(deref->type->base_type == GLSL_TYPE_FLOAT);
+      /* Matrices, row_major or not, are stored as if they were
+       * arrays of vectors of the appropriate size in std140.
+       * Arrays have their strides rounded up to a vec4, so the
+       * matrix stride is always 16.
+       */
+      unsigned matrix_stride = 16;
+
+      for (unsigned i = 0; i < deref->type->vector_elements; i++) {
+	 ir_rvalue *chan = new(mem_ctx) ir_constant((int)i);
+	 ir_dereference *deref_chan =
+	    new(mem_ctx) ir_dereference_array(deref->clone(mem_ctx, NULL),
+					      chan);
+
+	 ir_rvalue *chan_offset =
+	    add(base_offset,
+		new(mem_ctx) ir_constant(deref_offset + i * matrix_stride));
+
+	 base_ir->insert_before(assign(deref_chan,
+				       ubo_load(glsl_type::float_type,
+						chan_offset)));
+      }
+   }
+}
+
+} /* unnamed namespace */
+
+void
+lower_ubo_reference(struct gl_shader *shader, exec_list *instructions)
+{
+   lower_ubo_reference_visitor v(shader);
+
+   /* Loop over the instructions lowering references, because we take
+    * a deref of a UBO array using a UBO dereference as the index will
+    * produce a collection of instructions all of which have cloned
+    * UBO dereferences for that array index.
+    */
+   do {
+      v.progress = false;
+      visit_list_elements(&v, instructions);
+   } while (v.progress);
+}