glsl: Add an option to clamp block indices when lowering UBO/SSBOs
[mesa.git] / src / compiler / glsl / linker.cpp
index d9a681ccca1da7dd8a37ef862f3e2401f5e0dd69..3d95540256c49d92f4bf842f30aee917b2b8e8d0 100644 (file)
@@ -216,7 +216,7 @@ public:
        * array using an index too large for its actual size assigned at link
        * time.
        */
-      if (var->data.max_array_access >= this->num_vertices) {
+      if (var->data.max_array_access >= (int)this->num_vertices) {
          linker_error(this->prog, "geometry shader accesses element %i of "
                       "%s, but only %i input vertices\n",
                       var->data.max_array_access, var->name, this->num_vertices);
@@ -641,19 +641,25 @@ link_invalidate_variable_locations(exec_list *ir)
 
 
 /**
- * Set clip_distance_array_size based on the given shader.
+ * Set clip_distance_array_size based and cull_distance_array_size on the given
+ * shader.
  *
  * Also check for errors based on incorrect usage of gl_ClipVertex and
- * gl_ClipDistance.
+ * gl_ClipDistance and gl_CullDistance.
+ * Additionally test whether the arrays gl_ClipDistance and gl_CullDistance
+ * exceed the maximum size defined by gl_MaxCombinedClipAndCullDistances.
  *
  * Return false if an error was reported.
  */
 static void
-analyze_clip_usage(struct gl_shader_program *prog,
-                   struct gl_shader *shader,
-                   GLuint *clip_distance_array_size)
+analyze_clip_cull_usage(struct gl_shader_program *prog,
+                        struct gl_shader *shader,
+                        struct gl_context *ctx,
+                        GLuint *clip_distance_array_size,
+                        GLuint *cull_distance_array_size)
 {
    *clip_distance_array_size = 0;
+   *cull_distance_array_size = 0;
 
    if (!prog->IsES && prog->Version >= 130) {
       /* From section 7.1 (Vertex Shader Special Variables) of the
@@ -667,23 +673,62 @@ analyze_clip_usage(struct gl_shader_program *prog,
        */
       find_assignment_visitor clip_vertex("gl_ClipVertex");
       find_assignment_visitor clip_distance("gl_ClipDistance");
+      find_assignment_visitor cull_distance("gl_CullDistance");
 
       clip_vertex.run(shader->ir);
       clip_distance.run(shader->ir);
+      cull_distance.run(shader->ir);
+
+      /* From the ARB_cull_distance spec:
+       *
+       * It is a compile-time or link-time error for the set of shaders forming
+       * a program to statically read or write both gl_ClipVertex and either
+       * gl_ClipDistance or gl_CullDistance.
+       *
+       * This does not apply to GLSL ES shaders, since GLSL ES defines neither
+       * gl_ClipVertex, gl_ClipDistance or gl_CullDistance.
+       */
       if (clip_vertex.variable_found() && clip_distance.variable_found()) {
          linker_error(prog, "%s shader writes to both `gl_ClipVertex' "
                       "and `gl_ClipDistance'\n",
                       _mesa_shader_stage_to_string(shader->Stage));
          return;
       }
+      if (clip_vertex.variable_found() && cull_distance.variable_found()) {
+         linker_error(prog, "%s shader writes to both `gl_ClipVertex' "
+                      "and `gl_CullDistance'\n",
+                      _mesa_shader_stage_to_string(shader->Stage));
+         return;
+      }
 
       if (clip_distance.variable_found()) {
          ir_variable *clip_distance_var =
-               shader->symbols->get_variable("gl_ClipDistance");
-
+                shader->symbols->get_variable("gl_ClipDistance");
          assert(clip_distance_var);
          *clip_distance_array_size = clip_distance_var->type->length;
       }
+      if (cull_distance.variable_found()) {
+         ir_variable *cull_distance_var =
+                shader->symbols->get_variable("gl_CullDistance");
+         assert(cull_distance_var);
+         *cull_distance_array_size = cull_distance_var->type->length;
+      }
+      /* From the ARB_cull_distance spec:
+       *
+       * It is a compile-time or link-time error for the set of shaders forming
+       * a program to have the sum of the sizes of the gl_ClipDistance and
+       * gl_CullDistance arrays to be larger than
+       * gl_MaxCombinedClipAndCullDistances.
+       */
+      if ((*clip_distance_array_size + *cull_distance_array_size) >
+          ctx->Const.MaxClipPlanes) {
+          linker_error(prog, "%s shader: the combined size of "
+                       "'gl_ClipDistance' and 'gl_CullDistance' size cannot "
+                       "be larger than "
+                       "gl_MaxCombinedClipAndCullDistances (%u)",
+                       _mesa_shader_stage_to_string(shader->Stage),
+                       ctx->Const.MaxClipPlanes);
+      }
    }
 }
 
@@ -691,13 +736,15 @@ analyze_clip_usage(struct gl_shader_program *prog,
 /**
  * Verify that a vertex shader executable meets all semantic requirements.
  *
- * Also sets prog->Vert.ClipDistanceArraySize as a side effect.
+ * Also sets prog->Vert.ClipDistanceArraySize and
+ * prog->Vert.CullDistanceArraySize as a side effect.
  *
  * \param shader  Vertex shader executable to be verified
  */
 void
 validate_vertex_shader_executable(struct gl_shader_program *prog,
-                                 struct gl_shader *shader)
+                                  struct gl_shader *shader,
+                                  struct gl_context *ctx)
 {
    if (shader == NULL)
       return;
@@ -744,17 +791,22 @@ validate_vertex_shader_executable(struct gl_shader_program *prog,
       }
    }
 
-   analyze_clip_usage(prog, shader, &prog->Vert.ClipDistanceArraySize);
+   analyze_clip_cull_usage(prog, shader, ctx,
+                           &prog->Vert.ClipDistanceArraySize,
+                           &prog->Vert.CullDistanceArraySize);
 }
 
 void
 validate_tess_eval_shader_executable(struct gl_shader_program *prog,
-                                     struct gl_shader *shader)
+                                     struct gl_shader *shader,
+                                     struct gl_context *ctx)
 {
    if (shader == NULL)
       return;
 
-   analyze_clip_usage(prog, shader, &prog->TessEval.ClipDistanceArraySize);
+   analyze_clip_cull_usage(prog, shader, ctx,
+                           &prog->TessEval.ClipDistanceArraySize,
+                           &prog->TessEval.CullDistanceArraySize);
 }
 
 
@@ -765,7 +817,7 @@ validate_tess_eval_shader_executable(struct gl_shader_program *prog,
  */
 void
 validate_fragment_shader_executable(struct gl_shader_program *prog,
-                                   struct gl_shader *shader)
+                                    struct gl_shader *shader)
 {
    if (shader == NULL)
       return;
@@ -785,14 +837,15 @@ validate_fragment_shader_executable(struct gl_shader_program *prog,
 /**
  * Verify that a geometry shader executable meets all semantic requirements
  *
- * Also sets prog->Geom.VerticesIn, and prog->Geom.ClipDistanceArraySize as
- * a side effect.
+ * Also sets prog->Geom.VerticesIn, and prog->Geom.ClipDistanceArraySize and
+ * prog->Geom.CullDistanceArraySize as a side effect.
  *
  * \param shader Geometry shader executable to be verified
  */
 void
 validate_geometry_shader_executable(struct gl_shader_program *prog,
-                                   struct gl_shader *shader)
+                                    struct gl_shader *shader,
+                                    struct gl_context *ctx)
 {
    if (shader == NULL)
       return;
@@ -800,7 +853,9 @@ validate_geometry_shader_executable(struct gl_shader_program *prog,
    unsigned num_vertices = vertices_per_prim(prog->Geom.InputType);
    prog->Geom.VerticesIn = num_vertices;
 
-   analyze_clip_usage(prog, shader, &prog->Geom.ClipDistanceArraySize);
+   analyze_clip_cull_usage(prog, shader, ctx,
+                           &prog->Geom.ClipDistanceArraySize,
+                           &prog->Geom.CullDistanceArraySize);
 }
 
 /**
@@ -869,7 +924,7 @@ validate_intrastage_arrays(struct gl_shader_program *prog,
       if ((var->type->fields.array == existing->type->fields.array) &&
           ((var->type->length == 0)|| (existing->type->length == 0))) {
          if (var->type->length != 0) {
-            if (var->type->length <= existing->data.max_array_access) {
+            if ((int)var->type->length <= existing->data.max_array_access) {
                linker_error(prog, "%s `%s' declared as type "
                            "`%s' but outermost dimension has an index"
                            " of `%i'\n",
@@ -880,7 +935,7 @@ validate_intrastage_arrays(struct gl_shader_program *prog,
             existing->type = var->type;
             return true;
          } else if (existing->type->length != 0) {
-            if(existing->type->length <= var->data.max_array_access &&
+            if((int)existing->type->length <= var->data.max_array_access &&
                !existing->data.from_ssbo_unsized_array) {
                linker_error(prog, "%s `%s' declared as type "
                            "`%s' but outermost dimension has an index"
@@ -988,6 +1043,13 @@ cross_validate_globals(struct gl_shader_program *prog,
                     return;
               }
 
+              if (var->data.location_frac != existing->data.location_frac) {
+                    linker_error(prog, "explicit components for %s "
+                                 "`%s' have differing values\n",
+                                 mode_string(var), var->name);
+                    return;
+              }
+
               existing->data.location = var->data.location;
               existing->data.explicit_location = true;
            } else {
@@ -1086,21 +1148,11 @@ cross_validate_globals(struct gl_shader_program *prog,
                     return;
                  }
               } else {
-                 /* If the first-seen instance of a particular uniform did not
-                  * have an initializer but a later instance does, copy the
-                  * initializer to the version stored in the symbol table.
-                  */
-                 /* FINISHME: This is wrong.  The constant_value field should
-                  * FINISHME: not be modified!  Imagine a case where a shader
-                  * FINISHME: without an initializer is linked in two different
-                  * FINISHME: programs with shaders that have differing
-                  * FINISHME: initializers.  Linking with the first will
-                  * FINISHME: modify the shader, and linking with the second
-                  * FINISHME: will fail.
-                  */
-                 existing->constant_initializer =
-                    var->constant_initializer->clone(ralloc_parent(existing),
-                                                     NULL);
+                  /* If the first-seen instance of a particular uniform did
+                   * not have an initializer but a later instance does,
+                   * replace the former with the later.
+                   */
+                  variables.replace_variable(existing->name, var);
               }
            }
 
@@ -1114,13 +1166,6 @@ cross_validate_globals(struct gl_shader_program *prog,
                               var->name);
                  return;
               }
-
-              /* Some instance had an initializer, so keep track of that.  In
-               * this location, all sorts of initializers (constant or
-               * otherwise) will propagate the existence to the variable
-               * stored in the symbol table.
-               */
-              existing->data.has_initializer = true;
            }
 
            if (existing->data.invariant != var->data.invariant) {
@@ -1165,39 +1210,58 @@ cross_validate_uniforms(struct gl_shader_program *prog)
 }
 
 /**
- * Accumulates the array of prog->BufferInterfaceBlocks and checks that all
- * definitons of blocks agree on their contents.
+ * Accumulates the array of buffer blocks and checks that all definitions of
+ * blocks agree on their contents.
  */
 static bool
-interstage_cross_validate_uniform_blocks(struct gl_shader_program *prog)
+interstage_cross_validate_uniform_blocks(struct gl_shader_program *prog,
+                                         bool validate_ssbo)
 {
    int *InterfaceBlockStageIndex[MESA_SHADER_STAGES];
+   struct gl_uniform_block *blks = NULL;
+   unsigned *num_blks = validate_ssbo ? &prog->NumShaderStorageBlocks :
+      &prog->NumUniformBlocks;
 
-   unsigned max_num_uniform_blocks = 0;
+   unsigned max_num_buffer_blocks = 0;
    for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
-      if (prog->_LinkedShaders[i])
-        max_num_uniform_blocks += prog->_LinkedShaders[i]->NumBufferInterfaceBlocks;
+      if (prog->_LinkedShaders[i]) {
+         if (validate_ssbo) {
+            max_num_buffer_blocks +=
+               prog->_LinkedShaders[i]->NumShaderStorageBlocks;
+         } else {
+            max_num_buffer_blocks +=
+               prog->_LinkedShaders[i]->NumUniformBlocks;
+         }
+      }
    }
 
    for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
       struct gl_shader *sh = prog->_LinkedShaders[i];
 
-      InterfaceBlockStageIndex[i] = new int[max_num_uniform_blocks];
-      for (unsigned int j = 0; j < max_num_uniform_blocks; j++)
+      InterfaceBlockStageIndex[i] = new int[max_num_buffer_blocks];
+      for (unsigned int j = 0; j < max_num_buffer_blocks; j++)
          InterfaceBlockStageIndex[i][j] = -1;
 
       if (sh == NULL)
         continue;
 
-      for (unsigned int j = 0; j < sh->NumBufferInterfaceBlocks; j++) {
-        int index = link_cross_validate_uniform_block(prog,
-                                                      &prog->BufferInterfaceBlocks,
-                                                      &prog->NumBufferInterfaceBlocks,
-                                                      sh->BufferInterfaceBlocks[j]);
+      unsigned sh_num_blocks;
+      struct gl_uniform_block **sh_blks;
+      if (validate_ssbo) {
+         sh_num_blocks = prog->_LinkedShaders[i]->NumShaderStorageBlocks;
+         sh_blks = sh->ShaderStorageBlocks;
+      } else {
+         sh_num_blocks = prog->_LinkedShaders[i]->NumUniformBlocks;
+         sh_blks = sh->UniformBlocks;
+      }
+
+      for (unsigned int j = 0; j < sh_num_blocks; j++) {
+         int index = link_cross_validate_uniform_block(prog, &blks, num_blks,
+                                                       sh_blks[j]);
 
          if (index == -1) {
-            linker_error(prog, "uniform block `%s' has mismatching definitions\n",
-                         sh->BufferInterfaceBlocks[j]->Name);
+            linker_error(prog, "buffer block `%s' has mismatching "
+                         "definitions\n", sh_blks[j]->Name);
 
             for (unsigned k = 0; k <= i; k++) {
                delete[] InterfaceBlockStageIndex[k];
@@ -1213,16 +1277,18 @@ interstage_cross_validate_uniform_blocks(struct gl_shader_program *prog)
     * FIXME: We should be able to free the per stage blocks here.
     */
    for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
-      for (unsigned j = 0; j < prog->NumBufferInterfaceBlocks; j++) {
+      for (unsigned j = 0; j < *num_blks; j++) {
          int stage_index = InterfaceBlockStageIndex[i][j];
 
         if (stage_index != -1) {
            struct gl_shader *sh = prog->_LinkedShaders[i];
 
-            prog->BufferInterfaceBlocks[j].stageref |= (1 << i);
+            blks[j].stageref |= (1 << i);
+
+            struct gl_uniform_block **sh_blks = validate_ssbo ?
+               sh->ShaderStorageBlocks : sh->UniformBlocks;
 
-            sh->BufferInterfaceBlocks[stage_index] =
-               &prog->BufferInterfaceBlocks[j];
+            sh_blks[stage_index] = &blks[j];
         }
       }
    }
@@ -1231,6 +1297,11 @@ interstage_cross_validate_uniform_blocks(struct gl_shader_program *prog)
       delete[] InterfaceBlockStageIndex[i];
    }
 
+   if (validate_ssbo)
+      prog->ShaderStorageBlocks = blks;
+   else
+      prog->UniformBlocks = blks;
+
    return true;
 }
 
@@ -1522,7 +1593,7 @@ private:
     */
    static const glsl_type *
    resize_interface_members(const glsl_type *type,
-                            const unsigned *max_ifc_array_access,
+                            const int *max_ifc_array_access,
                             bool is_ssbo)
    {
       unsigned num_fields = type->length;
@@ -2074,7 +2145,10 @@ link_intrastage_shaders(void *mem_ctx,
                        struct gl_shader **shader_list,
                        unsigned num_shaders)
 {
-   struct gl_uniform_block *uniform_blocks = NULL;
+   struct gl_uniform_block *ubo_blocks = NULL;
+   struct gl_uniform_block *ssbo_blocks = NULL;
+   unsigned num_ubo_blocks = 0;
+   unsigned num_ssbo_blocks = 0;
 
    /* Check that global variables defined in multiple shaders are consistent.
     */
@@ -2090,9 +2164,10 @@ link_intrastage_shaders(void *mem_ctx,
       return NULL;
 
    /* Link up uniform blocks defined within this stage. */
-   const unsigned num_uniform_blocks =
-      link_uniform_blocks(mem_ctx, ctx, prog, shader_list, num_shaders,
-                          &uniform_blocks);
+   link_uniform_blocks(mem_ctx, ctx, prog, shader_list, num_shaders,
+                       &ubo_blocks, &num_ubo_blocks, &ssbo_blocks,
+                       &num_ssbo_blocks);
+
    if (!prog->LinkStatus)
       return NULL;
 
@@ -2159,15 +2234,23 @@ link_intrastage_shaders(void *mem_ctx,
    linked->ir = new(linked) exec_list;
    clone_ir_list(mem_ctx, linked->ir, main->ir);
 
-   linked->BufferInterfaceBlocks =
-      ralloc_array(linked, gl_uniform_block *, num_uniform_blocks);
-
-   ralloc_steal(linked, uniform_blocks);
-   for (unsigned i = 0; i < num_uniform_blocks; i++) {
-      linked->BufferInterfaceBlocks[i] = &uniform_blocks[i];
+   /* Copy ubo blocks to linked shader list */
+   linked->UniformBlocks =
+      ralloc_array(linked, gl_uniform_block *, num_ubo_blocks);
+   ralloc_steal(linked, ubo_blocks);
+   for (unsigned i = 0; i < num_ubo_blocks; i++) {
+      linked->UniformBlocks[i] = &ubo_blocks[i];
    }
+   linked->NumUniformBlocks = num_ubo_blocks;
 
-   linked->NumBufferInterfaceBlocks = num_uniform_blocks;
+   /* Copy ssbo blocks to linked shader list */
+   linked->ShaderStorageBlocks =
+      ralloc_array(linked, gl_uniform_block *, num_ssbo_blocks);
+   ralloc_steal(linked, ssbo_blocks);
+   for (unsigned i = 0; i < num_ssbo_blocks; i++) {
+      linked->ShaderStorageBlocks[i] = &ssbo_blocks[i];
+   }
+   linked->NumShaderStorageBlocks = num_ssbo_blocks;
 
    link_fs_input_layout_qualifiers(prog, linked, shader_list, num_shaders);
    link_tcs_out_layout_qualifiers(prog, linked, shader_list, num_shaders);
@@ -2316,10 +2399,10 @@ update_array_sizes(struct gl_shader_program *prog)
           * Subroutine uniforms are not removed.
          */
         if (var->is_in_buffer_block() || var->type->contains_atomic() ||
-            var->type->contains_subroutine())
+            var->type->contains_subroutine() || var->constant_initializer)
            continue;
 
-        unsigned int size = var->data.max_array_access;
+        int size = var->data.max_array_access;
         for (unsigned j = 0; j < MESA_SHADER_STAGES; j++) {
               if (prog->_LinkedShaders[j] == NULL)
                  continue;
@@ -2336,7 +2419,7 @@ update_array_sizes(struct gl_shader_program *prog)
            }
         }
 
-        if (size + 1 != var->type->length) {
+        if (size + 1 != (int)var->type->length) {
            /* If this is a built-in uniform (i.e., it's backed by some
             * fixed-function state), adjust the number of state slots to
             * match the new array size.  The number of slots per array entry
@@ -2580,7 +2663,7 @@ assign_attribute_or_color_locations(gl_shader_program *prog,
          return false;
       }
 
-      const unsigned slots = var->type->count_attribute_slots(target_index == MESA_SHADER_VERTEX ? true : false);
+      const unsigned slots = var->type->count_attribute_slots(target_index == MESA_SHADER_VERTEX);
 
       /* If the variable is not a built-in and has a location statically
        * assigned in the shader (presumably via a layout qualifier), make sure
@@ -2790,6 +2873,26 @@ assign_attribute_or_color_locations(gl_shader_program *prog,
       to_assign[i].var->data.location = generic_base + location;
       to_assign[i].var->data.is_unmatched_generic_inout = 0;
       used_locations |= (use_mask << location);
+
+      if (to_assign[i].var->type->without_array()->is_dual_slot_double())
+         double_storage_locations |= (use_mask << location);
+   }
+
+   /* Now that we have all the locations, from the GL 4.5 core spec, section
+    * 11.1.1 (Vertex Attributes), dvec3, dvec4, dmat2x3, dmat2x4, dmat3,
+    * dmat3x4, dmat4x3, and dmat4 count as consuming twice as many attributes
+    * as equivalent single-precision types.
+    */
+   if (target_index == MESA_SHADER_VERTEX) {
+      unsigned total_attribs_size =
+         _mesa_bitcount(used_locations & ((1 << max_index) - 1)) +
+         _mesa_bitcount(double_storage_locations);
+      if (total_attribs_size > max_index) {
+        linker_error(prog,
+                     "attempt to use %d vertex attribute slots only %d available ",
+                     total_attribs_size, max_index);
+        return false;
+      }
    }
 
    return true;
@@ -2800,12 +2903,12 @@ assign_attribute_or_color_locations(gl_shader_program *prog,
  * unmatch flag if found so we don't optimise them away.
  */
 static void
-match_explicit_outputs_to_inputs(struct gl_shader_program *prog,
-                                 gl_shader *producer,
+match_explicit_outputs_to_inputs(gl_shader *producer,
                                  gl_shader *consumer)
 {
    glsl_symbol_table parameters;
-   ir_variable *explicit_locations[MAX_VARYING] = { NULL };
+   ir_variable *explicit_locations[MAX_VARYINGS_INCL_PATCH][4] =
+      { {NULL, NULL} };
 
    /* Find all shader outputs in the "producer" stage.
     */
@@ -2818,8 +2921,8 @@ match_explicit_outputs_to_inputs(struct gl_shader_program *prog,
       if (var->data.explicit_location &&
           var->data.location >= VARYING_SLOT_VAR0) {
          const unsigned idx = var->data.location - VARYING_SLOT_VAR0;
-         if (explicit_locations[idx] == NULL)
-            explicit_locations[idx] = var;
+         if (explicit_locations[idx][var->data.location_frac] == NULL)
+            explicit_locations[idx][var->data.location_frac] = var;
       }
    }
 
@@ -2833,7 +2936,8 @@ match_explicit_outputs_to_inputs(struct gl_shader_program *prog,
       ir_variable *output = NULL;
       if (input->data.explicit_location
           && input->data.location >= VARYING_SLOT_VAR0) {
-         output = explicit_locations[input->data.location - VARYING_SLOT_VAR0];
+         output = explicit_locations[input->data.location - VARYING_SLOT_VAR0]
+            [input->data.location_frac];
 
          if (output != NULL){
             input->data.is_unmatched_generic_inout = 0;
@@ -2973,21 +3077,22 @@ check_resources(struct gl_context *ctx, struct gl_shader_program *prog)
                    ctx->Const.MaxCombinedShaderStorageBlocks);
    }
 
-   for (unsigned i = 0; i < prog->NumBufferInterfaceBlocks; i++) {
-      /* Don't check SSBOs for Uniform Block Size */
-      if (!prog->BufferInterfaceBlocks[i].IsShaderStorage &&
-          prog->BufferInterfaceBlocks[i].UniformBufferSize > ctx->Const.MaxUniformBlockSize) {
+   for (unsigned i = 0; i < prog->NumUniformBlocks; i++) {
+      if (prog->UniformBlocks[i].UniformBufferSize >
+          ctx->Const.MaxUniformBlockSize) {
          linker_error(prog, "Uniform block %s too big (%d/%d)\n",
-                      prog->BufferInterfaceBlocks[i].Name,
-                      prog->BufferInterfaceBlocks[i].UniformBufferSize,
+                      prog->UniformBlocks[i].Name,
+                      prog->UniformBlocks[i].UniformBufferSize,
                       ctx->Const.MaxUniformBlockSize);
       }
+   }
 
-      if (prog->BufferInterfaceBlocks[i].IsShaderStorage &&
-          prog->BufferInterfaceBlocks[i].UniformBufferSize > ctx->Const.MaxShaderStorageBlockSize) {
+   for (unsigned i = 0; i < prog->NumShaderStorageBlocks; i++) {
+      if (prog->ShaderStorageBlocks[i].UniformBufferSize >
+          ctx->Const.MaxShaderStorageBlockSize) {
          linker_error(prog, "Shader storage block %s too big (%d/%d)\n",
-                      prog->BufferInterfaceBlocks[i].Name,
-                      prog->BufferInterfaceBlocks[i].UniformBufferSize,
+                      prog->ShaderStorageBlocks[i].Name,
+                      prog->ShaderStorageBlocks[i].UniformBufferSize,
                       ctx->Const.MaxShaderStorageBlockSize);
       }
    }
@@ -3003,12 +3108,20 @@ link_calculate_subroutine_compat(struct gl_shader_program *prog)
          continue;
 
       for (unsigned j = 0; j < sh->NumSubroutineUniformRemapTable; j++) {
+         if (sh->SubroutineUniformRemapTable[j] == INACTIVE_UNIFORM_EXPLICIT_LOCATION)
+            continue;
+
          struct gl_uniform_storage *uni = sh->SubroutineUniformRemapTable[j];
 
          if (!uni)
             continue;
 
+         sh->NumSubroutineUniforms++;
          count = 0;
+         if (sh->NumSubroutineFunctions == 0) {
+            linker_error(prog, "subroutine uniform %s defined but no valid functions found\n", uni->type->name);
+            continue;
+         }
          for (unsigned f = 0; f < sh->NumSubroutineFunctions; f++) {
             struct gl_subroutine_function *fn = &sh->SubroutineFunctions[f];
             for (int k = 0; k < fn->num_compat_types; k++) {
@@ -3210,12 +3323,12 @@ reserve_subroutine_explicit_locations(struct gl_shader_program *prog,
  * any optimizations happen to handle also inactive uniforms and
  * inactive array elements that may get trimmed away.
  */
-static int
+static unsigned
 check_explicit_uniform_locations(struct gl_context *ctx,
                                  struct gl_shader_program *prog)
 {
    if (!ctx->Extensions.ARB_explicit_uniform_location)
-      return -1;
+      return 0;
 
    /* This map is used to detect if overlapping explicit locations
     * occur with the same uniform (from different stage) or a different one.
@@ -3224,7 +3337,7 @@ check_explicit_uniform_locations(struct gl_context *ctx,
 
    if (!uniform_map) {
       linker_error(prog, "Out of memory during linking.\n");
-      return -1;
+      return 0;
    }
 
    unsigned entries_total = 0;
@@ -3253,7 +3366,7 @@ check_explicit_uniform_locations(struct gl_context *ctx,
             }
             if (!ret) {
                delete uniform_map;
-               return -1;
+               return 0;
             }
          }
       }
@@ -3295,8 +3408,8 @@ should_add_buffer_variable(struct gl_shader_program *shProg,
    if (type != GL_BUFFER_VARIABLE)
       return true;
 
-   for (unsigned i = 0; i < shProg->NumBufferInterfaceBlocks; i++) {
-      const char *block_name = shProg->BufferInterfaceBlocks[i].Name;
+   for (unsigned i = 0; i < shProg->NumShaderStorageBlocks; i++) {
+      const char *block_name = shProg->ShaderStorageBlocks[i].Name;
       block_name_len = strlen(block_name);
 
       const char *block_square_bracket = strchr(block_name, '[');
@@ -3329,12 +3442,12 @@ should_add_buffer_variable(struct gl_shader_program *shProg,
    if (found_interface)
       name = name + block_name_len + 1;
 
-   /* From: ARB_program_interface_query extension:
+   /* The ARB_program_interface_query spec says:
     *
-    *  "For an active shader storage block member declared as an array, an
-    *   entry will be generated only for the first array element, regardless
-    *   of its type.  For arrays of aggregate types, the enumeration rules are
-    *   applied recursively for the single enumerated array element.
+    *     "For an active shader storage block member declared as an array, an
+    *     entry will be generated only for the first array element, regardless
+    *     of its type.  For arrays of aggregate types, the enumeration rules
+    *     are applied recursively for the single enumerated array element."
     */
    const char *struct_first_dot = strchr(name, '.');
    const char *first_square_bracket = strchr(name, '[');
@@ -3479,8 +3592,9 @@ build_stageref(struct gl_shader_program *shProg, const char *name,
  */
 static gl_shader_variable *
 create_shader_variable(struct gl_shader_program *shProg,
-                       const ir_variable *in, bool use_implicit_location,
-                       int location_bias)
+                       const ir_variable *in,
+                       const char *name, const glsl_type *type,
+                       bool use_implicit_location, int location)
 {
    gl_shader_variable *out = ralloc(shProg, struct gl_shader_variable);
    if (!out)
@@ -3493,35 +3607,37 @@ create_shader_variable(struct gl_shader_program *shProg,
        in->data.location == SYSTEM_VALUE_VERTEX_ID_ZERO_BASE) {
       out->name = ralloc_strdup(shProg, "gl_VertexID");
    } else {
-      out->name = ralloc_strdup(shProg, in->name);
+      out->name = ralloc_strdup(shProg, name);
    }
 
    if (!out->name)
       return NULL;
 
-   /* From the ARB_program_interface_query specification:
+   /* The ARB_program_interface_query spec says:
     *
-    * "Not all active variables are assigned valid locations; the
-    *  following variables will have an effective location of -1:
+    *     "Not all active variables are assigned valid locations; the
+    *     following variables will have an effective location of -1:
     *
-    *  * uniforms declared as atomic counters;
+    *      * uniforms declared as atomic counters;
     *
-    *  * members of a uniform block;
+    *      * members of a uniform block;
     *
-    *  * built-in inputs, outputs, and uniforms (starting with "gl_"); and
+    *      * built-in inputs, outputs, and uniforms (starting with "gl_"); and
     *
-    *  * inputs or outputs not declared with a "location" layout qualifier,
-    *    except for vertex shader inputs and fragment shader outputs."
+    *      * inputs or outputs not declared with a "location" layout
+    *        qualifier, except for vertex shader inputs and fragment shader
+    *        outputs."
     */
    if (in->type->base_type == GLSL_TYPE_ATOMIC_UINT ||
        is_gl_identifier(in->name) ||
        !(in->data.explicit_location || use_implicit_location)) {
       out->location = -1;
    } else {
-      out->location = in->data.location - location_bias;
+      out->location = location;
    }
 
-   out->type = in->type;
+   out->type = type;
+   out->component = in->data.location_frac;
    out->index = in->data.index;
    out->patch = in->data.patch;
    out->mode = in->data.mode;
@@ -3529,6 +3645,76 @@ create_shader_variable(struct gl_shader_program *shProg,
    return out;
 }
 
+static bool
+add_shader_variable(struct gl_shader_program *shProg, unsigned stage_mask,
+                    GLenum programInterface, ir_variable *var,
+                    const char *name, const glsl_type *type,
+                    bool use_implicit_location, int location)
+{
+   const bool is_vertex_input =
+      programInterface == GL_PROGRAM_INPUT &&
+      stage_mask == MESA_SHADER_VERTEX;
+
+   switch (type->base_type) {
+   case GLSL_TYPE_STRUCT: {
+      /* The ARB_program_interface_query spec says:
+       *
+       *     "For an active variable declared as a structure, a separate entry
+       *     will be generated for each active structure member.  The name of
+       *     each entry is formed by concatenating the name of the structure,
+       *     the "."  character, and the name of the structure member.  If a
+       *     structure member to enumerate is itself a structure or array,
+       *     these enumeration rules are applied recursively."
+       */
+      unsigned field_location = location;
+      for (unsigned i = 0; i < type->length; i++) {
+         const struct glsl_struct_field *field = &type->fields.structure[i];
+         char *field_name = ralloc_asprintf(shProg, "%s.%s", name, field->name);
+         if (!add_shader_variable(shProg, stage_mask, programInterface,
+                                  var, field_name, field->type,
+                                  use_implicit_location, field_location))
+            return false;
+
+         field_location +=
+            field->type->count_attribute_slots(is_vertex_input);
+      }
+      return true;
+   }
+
+   default: {
+      /* Issue #16 of the ARB_program_interface_query spec says:
+       *
+       * "* If a variable is a member of an interface block without an
+       *    instance name, it is enumerated using just the variable name.
+       *
+       *  * If a variable is a member of an interface block with an instance
+       *    name, it is enumerated as "BlockName.Member", where "BlockName" is
+       *    the name of the interface block (not the instance name) and
+       *    "Member" is the name of the variable."
+       */
+      const char *prefixed_name = (var->data.from_named_ifc_block &&
+                                   !is_gl_identifier(var->name))
+         ? ralloc_asprintf(shProg, "%s.%s", var->get_interface_type()->name,
+                           name)
+         : name;
+
+      /* The ARB_program_interface_query spec says:
+       *
+       *     "For an active variable declared as a single instance of a basic
+       *     type, a single entry will be generated, using the variable name
+       *     from the shader source."
+       */
+      gl_shader_variable *sha_v =
+         create_shader_variable(shProg, var, prefixed_name, type,
+                                use_implicit_location, location);
+      if (!sha_v)
+         return false;
+
+      return add_program_resource(shProg, programInterface, sha_v, stage_mask);
+   }
+   }
+}
+
 static bool
 add_interface_variables(struct gl_shader_program *shProg,
                         unsigned stage, GLenum programInterface)
@@ -3577,12 +3763,9 @@ add_interface_variables(struct gl_shader_program *shProg,
          (stage == MESA_SHADER_VERTEX && var->data.mode == ir_var_shader_in) ||
          (stage == MESA_SHADER_FRAGMENT && var->data.mode == ir_var_shader_out);
 
-      gl_shader_variable *sha_v =
-         create_shader_variable(shProg, var, vs_input_or_fs_output, loc_bias);
-      if (!sha_v)
-         return false;
-
-      if (!add_program_resource(shProg, programInterface, sha_v, 1 << stage))
+      if (!add_shader_variable(shProg, 1 << stage, programInterface,
+                               var, var->name, var->type, vs_input_or_fs_output,
+                               var->data.location - loc_bias))
          return false;
    }
    return true;
@@ -3612,13 +3795,11 @@ add_packed_varyings(struct gl_shader_program *shProg, int stage, GLenum type)
          }
 
          if (type == iface) {
-            gl_shader_variable *sha_v =
-               create_shader_variable(shProg, var, false, VARYING_SLOT_VAR0);
-            if (!sha_v)
-               return false;
-            if (!add_program_resource(shProg, iface, sha_v,
-                                      build_stageref(shProg, sha_v->name,
-                                                     sha_v->mode)))
+            const int stage_mask =
+               build_stageref(shProg, var->name, var->data.mode);
+            if (!add_shader_variable(shProg, stage_mask,
+                                     iface, var, var->name, var->type, false,
+                                     var->data.location - VARYING_SLOT_VAR0))
                return false;
          }
       }
@@ -3638,12 +3819,11 @@ add_fragdata_arrays(struct gl_shader_program *shProg)
       ir_variable *var = node->as_variable();
       if (var) {
          assert(var->data.mode == ir_var_shader_out);
-         gl_shader_variable *sha_v =
-            create_shader_variable(shProg, var, true, FRAG_RESULT_DATA0);
-         if (!sha_v)
-            return false;
-         if (!add_program_resource(shProg, GL_PROGRAM_OUTPUT, sha_v,
-                                   1 << MESA_SHADER_FRAGMENT))
+
+         if (!add_shader_variable(shProg,
+                                  1 << MESA_SHADER_FRAGMENT,
+                                  GL_PROGRAM_OUTPUT, var, var->name, var->type,
+                                  true, var->data.location - FRAG_RESULT_DATA0))
             return false;
       }
    }
@@ -3656,14 +3836,16 @@ get_top_level_name(const char *name)
    const char *first_dot = strchr(name, '.');
    const char *first_square_bracket = strchr(name, '[');
    int name_size = 0;
-   /* From ARB_program_interface_query spec:
+
+   /* The ARB_program_interface_query spec says:
     *
-    * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying the
-    *  number of active array elements of the top-level shader storage block
-    *  member containing to the active variable is written to <params>.  If the
-    *  top-level block member is not declared as an array, the value one is
-    *  written to <params>.  If the top-level block member is an array with no
-    *  declared size, the value zero is written to <params>.
+    *     "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
+    *     the number of active array elements of the top-level shader storage
+    *     block member containing to the active variable is written to
+    *     <params>.  If the top-level block member is not declared as an
+    *     array, the value one is written to <params>.  If the top-level block
+    *     member is an array with no declared size, the value zero is written
+    *     to <params>."
     */
 
    /* The buffer variable is on top level.*/
@@ -3733,15 +3915,15 @@ static int
 get_array_size(struct gl_uniform_storage *uni, const glsl_struct_field *field,
                char *interface_name, char *var_name)
 {
-   /* From GL_ARB_program_interface_query spec:
+   /* The ARB_program_interface_query spec says:
     *
-    * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer
-    * identifying the number of active array elements of the top-level
-    * shader storage block member containing to the active variable is
-    * written to <params>.  If the top-level block member is not
-    * declared as an array, the value one is written to <params>.  If
-    * the top-level block member is an array with no declared size,
-    * the value zero is written to <params>.
+    *     "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
+    *     the number of active array elements of the top-level shader storage
+    *     block member containing to the active variable is written to
+    *     <params>.  If the top-level block member is not declared as an
+    *     array, the value one is written to <params>.  If the top-level block
+    *     member is an array with no declared size, the value zero is written
+    *     to <params>."
     */
    if (is_top_level_shader_storage_block_member(uni->name,
                                                 interface_name,
@@ -3760,17 +3942,16 @@ get_array_stride(struct gl_uniform_storage *uni, const glsl_type *interface,
                  const glsl_struct_field *field, char *interface_name,
                  char *var_name)
 {
-   /* From GL_ARB_program_interface_query:
+   /* The ARB_program_interface_query spec says:
     *
-    * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
-    *  identifying the stride between array elements of the top-level
-    *  shader storage block member containing the active variable is
-    *  written to <params>.  For top-level block members declared as
-    *  arrays, the value written is the difference, in basic machine
-    *  units, between the offsets of the active variable for
-    *  consecutive elements in the top-level array.  For top-level
-    *  block members not declared as an array, zero is written to
-    *  <params>."
+    *     "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
+    *     identifying the stride between array elements of the top-level
+    *     shader storage block member containing the active variable is
+    *     written to <params>.  For top-level block members declared as
+    *     arrays, the value written is the difference, in basic machine units,
+    *     between the offsets of the active variable for consecutive elements
+    *     in the top-level array.  For top-level block members not declared as
+    *     an array, zero is written to <params>."
     */
    if (field->type->is_array()) {
       const enum glsl_matrix_layout matrix_layout =
@@ -3805,8 +3986,8 @@ calculate_array_size_and_stride(struct gl_shader_program *shProg,
    char *var_name = get_top_level_name(uni->name);
    char *interface_name =
       get_top_level_name(uni->is_shader_storage ?
-                         shProg->ShaderStorageBlocks[block_index]->Name :
-                         shProg->UniformBlocks[block_index]->Name);
+                         shProg->ShaderStorageBlocks[block_index].Name :
+                         shProg->UniformBlocks[block_index].Name);
 
    if (strcmp(var_name, interface_name) == 0) {
       /* Deal with instanced array of SSBOs */
@@ -3947,8 +4128,8 @@ build_program_resource_list(struct gl_context *ctx,
       int block_index = shProg->UniformStorage[i].block_index;
       if (block_index != -1) {
          stageref |= is_shader_storage ?
-            shProg->ShaderStorageBlocks[block_index]->stageref :
-            shProg->UniformBlocks[block_index]->stageref;
+            shProg->ShaderStorageBlocks[block_index].stageref :
+            shProg->UniformBlocks[block_index].stageref;
       }
 
       GLenum type = is_shader_storage ? GL_BUFFER_VARIABLE : GL_UNIFORM;
@@ -3965,12 +4146,17 @@ build_program_resource_list(struct gl_context *ctx,
          return;
    }
 
-   /* Add program uniform blocks and shader storage blocks. */
-   for (unsigned i = 0; i < shProg->NumBufferInterfaceBlocks; i++) {
-      bool is_shader_storage = shProg->BufferInterfaceBlocks[i].IsShaderStorage;
-      GLenum type = is_shader_storage ? GL_SHADER_STORAGE_BLOCK : GL_UNIFORM_BLOCK;
-      if (!add_program_resource(shProg, type,
-          &shProg->BufferInterfaceBlocks[i], 0))
+   /* Add program uniform blocks. */
+   for (unsigned i = 0; i < shProg->NumUniformBlocks; i++) {
+      if (!add_program_resource(shProg, GL_UNIFORM_BLOCK,
+          &shProg->UniformBlocks[i], 0))
+         return;
+   }
+
+   /* Add program shader storage blocks. */
+   for (unsigned i = 0; i < shProg->NumShaderStorageBlocks; i++) {
+      if (!add_program_resource(shProg, GL_SHADER_STORAGE_BLOCK,
+          &shProg->ShaderStorageBlocks[i], 0))
          return;
    }
 
@@ -4056,6 +4242,7 @@ link_assign_subroutine_types(struct gl_shader_program *prog)
       if (sh == NULL)
          continue;
 
+      sh->MaxSubroutineFunctionIndex = 0;
       foreach_in_list(ir_instruction, node, sh->ir) {
          ir_function *fn = node->as_function();
          if (!fn)
@@ -4067,6 +4254,12 @@ link_assign_subroutine_types(struct gl_shader_program *prog)
          if (!fn->num_subroutine_types)
             continue;
 
+        /* these should have been calculated earlier. */
+        assert(fn->subroutine_index != -1);
+         if (sh->NumSubroutineFunctions + 1 > MAX_SUBROUTINES) {
+            linker_error(prog, "Too many subroutine functions declared.\n");
+            return;
+         }
          sh->SubroutineFunctions = reralloc(sh, sh->SubroutineFunctions,
                                             struct gl_subroutine_function,
                                             sh->NumSubroutineFunctions + 1);
@@ -4094,68 +4287,14 @@ link_assign_subroutine_types(struct gl_shader_program *prog)
          sh->SubroutineFunctions[sh->NumSubroutineFunctions].index =
             fn->subroutine_index;
 
+         if (fn->subroutine_index > (int)sh->MaxSubroutineFunctionIndex)
+            sh->MaxSubroutineFunctionIndex = fn->subroutine_index;
+
          for (int j = 0; j < fn->num_subroutine_types; j++)
             sh->SubroutineFunctions[sh->NumSubroutineFunctions].types[j] = fn->subroutine_types[j];
          sh->NumSubroutineFunctions++;
       }
-
-      /* Assign index for subroutines without an explicit index*/
-      int index = 0;
-      for (unsigned j = 0; j < sh->NumSubroutineFunctions; j++) {
-         while (sh->SubroutineFunctions[j].index == -1) {
-            for (unsigned k = 0; k < sh->NumSubroutineFunctions; k++) {
-               if (sh->SubroutineFunctions[k].index == index)
-                  break;
-               else if (k == sh->NumSubroutineFunctions - 1)
-                  sh->SubroutineFunctions[j].index = index;
-            }
-            index++;
-         }
-      }
-   }
-}
-
-static void
-split_ubos_and_ssbos(void *mem_ctx,
-                     struct gl_uniform_block **s_blks,
-                     struct gl_uniform_block *p_blks,
-                     unsigned num_blocks,
-                     struct gl_uniform_block ***ubos,
-                     unsigned *num_ubos,
-                     struct gl_uniform_block ***ssbos,
-                     unsigned *num_ssbos)
-{
-   unsigned num_ubo_blocks = 0;
-   unsigned num_ssbo_blocks = 0;
-
-   /* Are we spliting the list of blocks for the shader or the program */
-   bool is_shader = p_blks == NULL;
-
-   for (unsigned i = 0; i < num_blocks; i++) {
-      if (is_shader ? s_blks[i]->IsShaderStorage : p_blks[i].IsShaderStorage)
-         num_ssbo_blocks++;
-      else
-         num_ubo_blocks++;
-   }
-
-   *ubos = ralloc_array(mem_ctx, gl_uniform_block *, num_ubo_blocks);
-   *num_ubos = 0;
-
-   *ssbos = ralloc_array(mem_ctx, gl_uniform_block *, num_ssbo_blocks);
-   *num_ssbos = 0;
-
-   for (unsigned i = 0; i < num_blocks; i++) {
-      struct gl_uniform_block *blk = is_shader ? s_blks[i] : &p_blks[i];
-      if (blk->IsShaderStorage) {
-         (*ssbos)[*num_ssbos] = blk;
-         (*num_ssbos)++;
-      } else {
-         (*ubos)[*num_ubos] = blk;
-         (*num_ubos)++;
-      }
    }
-
-   assert(*num_ubos + *num_ssbos == num_blocks);
 }
 
 static void
@@ -4385,16 +4524,16 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)
 
          switch (stage) {
          case MESA_SHADER_VERTEX:
-            validate_vertex_shader_executable(prog, sh);
+            validate_vertex_shader_executable(prog, sh, ctx);
             break;
          case MESA_SHADER_TESS_CTRL:
             /* nothing to be done */
             break;
          case MESA_SHADER_TESS_EVAL:
-            validate_tess_eval_shader_executable(prog, sh);
+            validate_tess_eval_shader_executable(prog, sh, ctx);
             break;
          case MESA_SHADER_GEOMETRY:
-            validate_geometry_shader_executable(prog, sh);
+            validate_geometry_shader_executable(prog, sh, ctx);
             break;
          case MESA_SHADER_FRAGMENT:
             validate_fragment_shader_executable(prog, sh);
@@ -4410,14 +4549,19 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)
       }
    }
 
-   if (num_shaders[MESA_SHADER_GEOMETRY] > 0)
+   if (num_shaders[MESA_SHADER_GEOMETRY] > 0) {
       prog->LastClipDistanceArraySize = prog->Geom.ClipDistanceArraySize;
-   else if (num_shaders[MESA_SHADER_TESS_EVAL] > 0)
+      prog->LastCullDistanceArraySize = prog->Geom.CullDistanceArraySize;
+   } else if (num_shaders[MESA_SHADER_TESS_EVAL] > 0) {
       prog->LastClipDistanceArraySize = prog->TessEval.ClipDistanceArraySize;
-   else if (num_shaders[MESA_SHADER_VERTEX] > 0)
+      prog->LastCullDistanceArraySize = prog->TessEval.CullDistanceArraySize;
+   } else if (num_shaders[MESA_SHADER_VERTEX] > 0) {
       prog->LastClipDistanceArraySize = prog->Vert.ClipDistanceArraySize;
-   else
+      prog->LastCullDistanceArraySize = prog->Vert.CullDistanceArraySize;
+   } else {
       prog->LastClipDistanceArraySize = 0; /* Not used */
+      prog->LastCullDistanceArraySize = 0; /* Not used */
+   }
 
    /* Here begins the inter-stage linking phase.  Some initial validation is
     * performed, then locations are assigned for uniforms, attributes, and
@@ -4498,7 +4642,12 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)
    if (prog->SeparateShader)
       disable_varying_optimizations_for_sso(prog);
 
-   if (!interstage_cross_validate_uniform_blocks(prog))
+   /* Process UBOs */
+   if (!interstage_cross_validate_uniform_blocks(prog, false))
+      goto done;
+
+   /* Process SSBOs */
+   if (!interstage_cross_validate_uniform_blocks(prog, true))
       goto done;
 
    /* Do common optimization before assigning storage for attributes,
@@ -4513,8 +4662,8 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)
       if (!prog->LinkStatus)
         goto done;
 
-      if (ctx->Const.ShaderCompilerOptions[i].LowerClipDistance) {
-         lower_clip_distance(prog->_LinkedShaders[i]);
+      if (ctx->Const.ShaderCompilerOptions[i].LowerCombinedClipCullDistance) {
+         lower_clip_cull_distance(prog, prog->_LinkedShaders[i]);
       }
 
       if (ctx->Const.LowerTessLevel) {
@@ -4554,7 +4703,7 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)
       if (prog->_LinkedShaders[i] == NULL)
          continue;
 
-      match_explicit_outputs_to_inputs(prog, prog->_LinkedShaders[prev],
+      match_explicit_outputs_to_inputs(prog->_LinkedShaders[prev],
                                        prog->_LinkedShaders[i]);
       prev = i;
    }
@@ -4617,9 +4766,9 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)
 
    /* If there is no fragment shader we need to set transform feedback.
     *
-    * For SSO we need also need to assign output locations, we assign them
-    * here because we need to do it for both single stage programs and multi
-    * stage programs.
+    * For SSO we also need to assign output locations.  We assign them here
+    * because we need to do it for both single stage programs and multi stage
+    * programs.
     */
    if (last < MESA_SHADER_FRAGMENT &&
        (num_tfeedback_decls != 0 || prog->SeparateShader)) {
@@ -4665,7 +4814,7 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)
           */
          int next = last;
          for (int i = next - 1; i >= 0; i--) {
-            if (prog->_LinkedShaders[i] == NULL)
+            if (prog->_LinkedShaders[i] == NULL && i != 0)
                continue;
 
             gl_shader *const sh_i = prog->_LinkedShaders[i];
@@ -4681,8 +4830,11 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)
                       tfeedback_decls);
 
             /* This must be done after all dead varyings are eliminated. */
-            if (!check_against_output_limit(ctx, prog, sh_i))
-               goto done;
+            if (sh_i != NULL) {
+               if (!check_against_output_limit(ctx, prog, sh_i)) {
+                  goto done;
+               }
+            }
             if (!check_against_input_limit(ctx, prog, sh_next))
                goto done;
 
@@ -4695,33 +4847,6 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)
                              has_xfb_qualifiers))
       goto done;
 
-   /* Split BufferInterfaceBlocks into UniformBlocks and ShaderStorageBlocks
-    * for gl_shader_program and gl_shader, so that drivers that need separate
-    * index spaces for each set can have that.
-    */
-   for (unsigned i = MESA_SHADER_VERTEX; i < MESA_SHADER_STAGES; i++) {
-      if (prog->_LinkedShaders[i] != NULL) {
-         gl_shader *sh = prog->_LinkedShaders[i];
-         split_ubos_and_ssbos(sh,
-                              sh->BufferInterfaceBlocks,
-                              NULL,
-                              sh->NumBufferInterfaceBlocks,
-                              &sh->UniformBlocks,
-                              &sh->NumUniformBlocks,
-                              &sh->ShaderStorageBlocks,
-                              &sh->NumShaderStorageBlocks);
-      }
-   }
-
-   split_ubos_and_ssbos(prog,
-                        NULL,
-                        prog->BufferInterfaceBlocks,
-                        prog->NumBufferInterfaceBlocks,
-                        &prog->UniformBlocks,
-                        &prog->NumUniformBlocks,
-                        &prog->ShaderStorageBlocks,
-                        &prog->NumShaderStorageBlocks);
-
    update_array_sizes(prog);
    link_assign_uniform_locations(prog, ctx->Const.UniformBooleanTrue,
                                  num_explicit_uniform_locs,
@@ -4777,14 +4902,19 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)
       if (prog->_LinkedShaders[i] == NULL)
         continue;
 
-      if (ctx->Const.ShaderCompilerOptions[i].LowerBufferInterfaceBlocks)
-         lower_ubo_reference(prog->_LinkedShaders[i]);
+      const struct gl_shader_compiler_options *options =
+         &ctx->Const.ShaderCompilerOptions[i];
+
+      if (options->LowerBufferInterfaceBlocks)
+         lower_ubo_reference(prog->_LinkedShaders[i],
+                             options->ClampBlockIndicesToArrayBounds);
 
-      if (ctx->Const.ShaderCompilerOptions[i].LowerShaderSharedVariables)
+      if (options->LowerShaderSharedVariables)
          lower_shared_reference(prog->_LinkedShaders[i],
                                 &prog->Comp.SharedSize);
 
       lower_vector_derefs(prog->_LinkedShaders[i]);
+      do_vec_index_to_swizzle(prog->_LinkedShaders[i]->ir);
    }
 
 done: