[Nouveau] [PATCH 06/11] glsl: lower cull_distance into cull_distance_mesa

Tobias Klausmann tobias.johannes.klausmann at mni.thm.de
Sun May 24 10:58:04 PDT 2015


From: Dave Airlie <airlied at redhat.com>

Signed-off-by: Tobias Klausmann <tobias.johannes.klausmann at mni.thm.de>
---
 src/glsl/Makefile.sources        |   1 +
 src/glsl/ir_optimization.h       |   1 +
 src/glsl/link_varyings.cpp       |  15 +-
 src/glsl/link_varyings.h         |   3 +-
 src/glsl/linker.cpp              |   1 +
 src/glsl/lower_cull_distance.cpp | 549 +++++++++++++++++++++++++++++++++++++++
 6 files changed, 565 insertions(+), 5 deletions(-)
 create mode 100644 src/glsl/lower_cull_distance.cpp

diff --git a/src/glsl/Makefile.sources b/src/glsl/Makefile.sources
index d784a81..502b6ca 100644
--- a/src/glsl/Makefile.sources
+++ b/src/glsl/Makefile.sources
@@ -143,6 +143,7 @@ LIBGLSL_FILES = \
 	loop_unroll.cpp \
 	lower_clip_distance.cpp \
 	lower_const_arrays_to_uniforms.cpp \
+	lower_cull_distance.cpp \
 	lower_discard.cpp \
 	lower_discard_flow.cpp \
 	lower_if_to_cond_assign.cpp \
diff --git a/src/glsl/ir_optimization.h b/src/glsl/ir_optimization.h
index e6939f3..1220df6 100644
--- a/src/glsl/ir_optimization.h
+++ b/src/glsl/ir_optimization.h
@@ -119,6 +119,7 @@ bool lower_variable_index_to_cond_assign(exec_list *instructions,
 bool lower_quadop_vector(exec_list *instructions, bool dont_lower_swz);
 bool lower_const_arrays_to_uniforms(exec_list *instructions);
 bool lower_clip_distance(gl_shader *shader);
+bool lower_cull_distance(gl_shader *shader);
 void lower_output_reads(exec_list *instructions);
 bool lower_packing_builtins(exec_list *instructions, int op_mask);
 void lower_ubo_reference(struct gl_shader *shader, exec_list *instructions);
diff --git a/src/glsl/link_varyings.cpp b/src/glsl/link_varyings.cpp
index 7b2d4bd..46f84c6 100644
--- a/src/glsl/link_varyings.cpp
+++ b/src/glsl/link_varyings.cpp
@@ -301,6 +301,7 @@ tfeedback_decl::init(struct gl_context *ctx, const void *mem_ctx,
    this->location = -1;
    this->orig_name = input;
    this->is_clip_distance_mesa = false;
+   this->is_cull_distance_mesa = false;
    this->skip_components = 0;
    this->next_buffer_separator = false;
    this->matched_candidate = NULL;
@@ -351,6 +352,10 @@ tfeedback_decl::init(struct gl_context *ctx, const void *mem_ctx,
        strcmp(this->var_name, "gl_ClipDistance") == 0) {
       this->is_clip_distance_mesa = true;
    }
+   if (ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].LowerClipDistance &&
+       strcmp(this->var_name, "gl_CullDistance") == 0) {
+      this->is_cull_distance_mesa = true;
+   }
 }
 
 
@@ -397,7 +402,8 @@ tfeedback_decl::assign_location(struct gl_context *ctx,
          this->matched_candidate->type->fields.array->matrix_columns;
       const unsigned vector_elements =
          this->matched_candidate->type->fields.array->vector_elements;
-      unsigned actual_array_size = this->is_clip_distance_mesa ?
+      unsigned actual_array_size =
+         (this->is_clip_distance_mesa || this->is_cull_distance_mesa) ?
          prog->LastClipDistanceArraySize :
          this->matched_candidate->type->array_size();
 
@@ -410,7 +416,8 @@ tfeedback_decl::assign_location(struct gl_context *ctx,
                          actual_array_size);
             return false;
          }
-         unsigned array_elem_size = this->is_clip_distance_mesa ?
+         unsigned array_elem_size =
+            (this->is_clip_distance_mesa || this->is_cull_distance_mesa) ?
             1 : vector_elements * matrix_cols;
          fine_location += array_elem_size * this->array_subscript;
          this->size = 1;
@@ -419,7 +426,7 @@ tfeedback_decl::assign_location(struct gl_context *ctx,
       }
       this->vector_elements = vector_elements;
       this->matrix_columns = matrix_cols;
-      if (this->is_clip_distance_mesa)
+      if (this->is_clip_distance_mesa || this->is_cull_distance_mesa)
          this->type = GL_FLOAT;
       else
          this->type = this->matched_candidate->type->fields.array->gl_type;
@@ -542,7 +549,7 @@ const tfeedback_candidate *
 tfeedback_decl::find_candidate(gl_shader_program *prog,
                                hash_table *tfeedback_candidates)
 {
-   const char *name = this->is_clip_distance_mesa
+   const char *name = this->is_cull_distance_mesa ? "gl_CullDistanceMESA" : this->is_clip_distance_mesa
       ? "gl_ClipDistanceMESA" : this->var_name;
    this->matched_candidate = (const tfeedback_candidate *)
       hash_table_find(tfeedback_candidates, name);
diff --git a/src/glsl/link_varyings.h b/src/glsl/link_varyings.h
index afc16a8..842ab7c 100644
--- a/src/glsl/link_varyings.h
+++ b/src/glsl/link_varyings.h
@@ -128,7 +128,7 @@ public:
     */
    unsigned num_components() const
    {
-      if (this->is_clip_distance_mesa)
+      if (this->is_clip_distance_mesa || this->is_cull_distance_mesa)
          return this->size;
       else
          return this->vector_elements * this->matrix_columns * this->size;
@@ -165,6 +165,7 @@ private:
     * gl_ClipDistance to gl_ClipDistanceMESA.
     */
    bool is_clip_distance_mesa;
+   bool is_cull_distance_mesa;
 
    /**
     * The vertex shader output location that the linker assigned for this
diff --git a/src/glsl/linker.cpp b/src/glsl/linker.cpp
index 8eace14..e616520 100644
--- a/src/glsl/linker.cpp
+++ b/src/glsl/linker.cpp
@@ -2938,6 +2938,7 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)
 
       if (ctx->Const.ShaderCompilerOptions[i].LowerClipDistance) {
          lower_clip_distance(prog->_LinkedShaders[i]);
+         lower_cull_distance(prog->_LinkedShaders[i]);
       }
 
       while (do_common_optimization(prog->_LinkedShaders[i]->ir, true, false,
diff --git a/src/glsl/lower_cull_distance.cpp b/src/glsl/lower_cull_distance.cpp
new file mode 100644
index 0000000..d3b9592
--- /dev/null
+++ b/src/glsl/lower_cull_distance.cpp
@@ -0,0 +1,549 @@
+/*
+ * Copyright © 2011 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_cull_distance.cpp
+ *
+ * This pass accounts for the difference between the way
+ * gl_CullDistance is declared in standard GLSL (as an array of
+ * floats), and the way it is frequently implemented in hardware (as
+ * a pair of vec4s, with four cull distances packed into each).
+ *
+ * The declaration of gl_CullDistance is replaced with a declaration
+ * of gl_CullDistanceMESA, and any references to gl_CullDistance are
+ * translated to refer to gl_CullDistanceMESA with the appropriate
+ * swizzling of array indices.  For instance:
+ *
+ *   gl_CullDistance[i]
+ *
+ * is translated into:
+ *
+ *   gl_CullDistanceMESA[i>>2][i&3]
+ *
+ * Since some hardware may not internally represent gl_CullDistance as a pair
+ * of vec4's, this lowering pass is optional.  To enable it, set the
+ * LowerCullDistance flag in gl_shader_compiler_options to true.
+ */
+
+#include "glsl_symbol_table.h"
+#include "ir_rvalue_visitor.h"
+#include "ir.h"
+#include "program/prog_instruction.h" /* For WRITEMASK_* */
+
+namespace {
+
+class lower_cull_distance_visitor : public ir_rvalue_visitor {
+public:
+   explicit lower_cull_distance_visitor(gl_shader_stage shader_stage)
+      : progress(false), old_cull_distance_1d_var(NULL),
+        old_cull_distance_2d_var(NULL), new_cull_distance_1d_var(NULL),
+        new_cull_distance_2d_var(NULL), shader_stage(shader_stage)
+   {
+   }
+
+   virtual ir_visitor_status visit(ir_variable *);
+   void create_indices(ir_rvalue*, ir_rvalue *&, ir_rvalue *&);
+   bool is_cull_distance_vec8(ir_rvalue *ir);
+   ir_rvalue *lower_cull_distance_vec8(ir_rvalue *ir);
+   virtual ir_visitor_status visit_leave(ir_assignment *);
+   void visit_new_assignment(ir_assignment *ir);
+   virtual ir_visitor_status visit_leave(ir_call *);
+
+   virtual void handle_rvalue(ir_rvalue **rvalue);
+
+   void fix_lhs(ir_assignment *);
+
+   bool progress;
+
+   /**
+    * Pointer to the declaration of gl_CullDistance, if found.
+    *
+    * Note:
+    *
+    * - the 2d_var is for geometry shader input only.
+    *
+    * - since gl_CullDistance is available in geometry shaders as both an
+    *   input and an output, it's possible for both old_cull_distance_1d_var
+    *   and old_cull_distance_2d_var to be non-null.
+    */
+   ir_variable *old_cull_distance_1d_var;
+   ir_variable *old_cull_distance_2d_var;
+
+   /**
+    * Pointer to the newly-created gl_CullDistanceMESA variable.
+    */
+   ir_variable *new_cull_distance_1d_var;
+   ir_variable *new_cull_distance_2d_var;
+
+   /**
+    * Type of shader we are compiling (e.g. MESA_SHADER_VERTEX)
+    */
+   const gl_shader_stage shader_stage;
+};
+
+} /* anonymous namespace */
+
+/**
+ * Replace any declaration of gl_CullDistance as an array of floats with a
+ * declaration of gl_CullDistanceMESA as an array of vec4's.
+ */
+ir_visitor_status
+lower_cull_distance_visitor::visit(ir_variable *ir)
+{
+   if (!ir->name || strcmp(ir->name, "gl_CullDistance") != 0)
+      return visit_continue;
+   assert (ir->type->is_array());
+
+   if (!ir->type->fields.array->is_array()) {
+      /* 1D gl_CullDistance (used for vertex and geometry output, and fragment
+       * input).
+       */
+      if (this->old_cull_distance_1d_var)
+         return visit_continue;
+
+      this->progress = true;
+      this->old_cull_distance_1d_var = ir;
+      assert (ir->type->fields.array == glsl_type::float_type);
+      unsigned new_size = (ir->type->array_size() + 3) / 4;
+
+      /* Clone the old var so that we inherit all of its properties */
+      this->new_cull_distance_1d_var = ir->clone(ralloc_parent(ir), NULL);
+
+      /* And change the properties that we need to change */
+      this->new_cull_distance_1d_var->name
+         = ralloc_strdup(this->new_cull_distance_1d_var,
+                         "gl_CullDistanceMESA");
+      this->new_cull_distance_1d_var->type
+         = glsl_type::get_array_instance(glsl_type::vec4_type, new_size);
+      this->new_cull_distance_1d_var->data.max_array_access
+         = ir->data.max_array_access / 4;
+
+      ir->replace_with(this->new_cull_distance_1d_var);
+   } else {
+      /* 2D gl_CullDistance (used for geometry input). */
+      assert(ir->data.mode == ir_var_shader_in &&
+             this->shader_stage == MESA_SHADER_GEOMETRY);
+      if (this->old_cull_distance_2d_var)
+         return visit_continue;
+
+      this->progress = true;
+      this->old_cull_distance_2d_var = ir;
+      assert (ir->type->fields.array->fields.array == glsl_type::float_type);
+      unsigned new_size = (ir->type->fields.array->array_size() + 3) / 4;
+
+      /* Clone the old var so that we inherit all of its properties */
+      this->new_cull_distance_2d_var = ir->clone(ralloc_parent(ir), NULL);
+
+      /* And change the properties that we need to change */
+      this->new_cull_distance_2d_var->name
+         = ralloc_strdup(this->new_cull_distance_2d_var, "gl_CullDistanceMESA");
+      this->new_cull_distance_2d_var->type = glsl_type::get_array_instance(
+         glsl_type::get_array_instance(glsl_type::vec4_type,
+            new_size),
+         ir->type->array_size());
+      this->new_cull_distance_2d_var->data.max_array_access
+         = ir->data.max_array_access / 4;
+
+      ir->replace_with(this->new_cull_distance_2d_var);
+   }
+   return visit_continue;
+}
+
+
+/**
+ * Create the necessary GLSL rvalues to index into gl_CullDistanceMESA based
+ * on the rvalue previously used to index into gl_CullDistance.
+ *
+ * \param array_index Selects one of the vec4's in gl_CullDistanceMESA
+ * \param swizzle_index Selects a component within the vec4 selected by
+ *        array_index.
+ */
+void
+lower_cull_distance_visitor::create_indices(ir_rvalue *old_index,
+                                            ir_rvalue *&array_index,
+                                            ir_rvalue *&swizzle_index)
+{
+   void *ctx = ralloc_parent(old_index);
+
+   /* Make sure old_index is a signed int so that the bitwise "shift" and
+    * "and" operations below type check properly.
+    */
+   if (old_index->type != glsl_type::int_type) {
+      assert (old_index->type == glsl_type::uint_type);
+      old_index = new(ctx) ir_expression(ir_unop_u2i, old_index);
+   }
+
+   ir_constant *old_index_constant = old_index->constant_expression_value();
+   if (old_index_constant) {
+      /* gl_CullDistance is being accessed via a constant index.  Don't bother
+       * creating expressions to calculate the lowered indices.  Just create
+       * constants.
+       */
+      int const_val = old_index_constant->get_int_component(0);
+      array_index = new(ctx) ir_constant(const_val / 4);
+      swizzle_index = new(ctx) ir_constant(const_val % 4);
+   } else {
+      /* Create a variable to hold the value of old_index (so that we
+       * don't compute it twice).
+       */
+      ir_variable *old_index_var = new(ctx) ir_variable(
+         glsl_type::int_type, "cull_distance_index", ir_var_temporary);
+      this->base_ir->insert_before(old_index_var);
+      this->base_ir->insert_before(new(ctx) ir_assignment(
+         new(ctx) ir_dereference_variable(old_index_var), old_index));
+
+      /* Create the expression cull_distance_index / 4.  Do this as a bit
+       * shift because that's likely to be more efficient.
+       */
+      array_index = new(ctx) ir_expression(
+         ir_binop_rshift, new(ctx) ir_dereference_variable(old_index_var),
+         new(ctx) ir_constant(2));
+
+      /* Create the expression cull_distance_index % 4.  Do this as a bitwise
+       * AND because that's likely to be more efficient.
+       */
+      swizzle_index = new(ctx) ir_expression(
+         ir_binop_bit_and, new(ctx) ir_dereference_variable(old_index_var),
+         new(ctx) ir_constant(3));
+   }
+}
+
+
+/**
+ * Determine whether the given rvalue describes an array of 8 floats that
+ * needs to be lowered to an array of 2 vec4's; that is, determine whether it
+ * matches one of the following patterns:
+ *
+ * - gl_CullDistance (if gl_CullDistance is 1D)
+ * - gl_CullDistance[i] (if gl_CullDistance is 2D)
+ */
+bool
+lower_cull_distance_visitor::is_cull_distance_vec8(ir_rvalue *ir)
+{
+   /* Note that geometry shaders contain gl_CullDistance both as an input
+    * (which is a 2D array) and an output (which is a 1D array), so it's
+    * possible for both this->old_cull_distance_1d_var and
+    * this->old_cull_distance_2d_var to be non-NULL in the same shader.
+    */
+
+   if (this->old_cull_distance_1d_var) {
+      ir_dereference_variable *var_ref = ir->as_dereference_variable();
+      if (var_ref && var_ref->var == this->old_cull_distance_1d_var)
+         return true;
+   }
+   if (this->old_cull_distance_2d_var) {
+      /* 2D cull distance is only possible as a geometry input */
+      assert(this->shader_stage == MESA_SHADER_GEOMETRY);
+
+      ir_dereference_array *array_ref = ir->as_dereference_array();
+      if (array_ref) {
+         ir_dereference_variable *var_ref =
+            array_ref->array->as_dereference_variable();
+         if (var_ref && var_ref->var == this->old_cull_distance_2d_var)
+            return true;
+      }
+   }
+   return false;
+}
+
+
+/**
+ * If the given ir satisfies is_cull_distance_vec8(), return new ir
+ * representing its lowered equivalent.  That is, map:
+ *
+ * - gl_CullDistance    => gl_CullDistanceMESA    (if gl_CullDistance is 1D)
+ * - gl_CullDistance[i] => gl_CullDistanceMESA[i] (if gl_CullDistance is 2D)
+ *
+ * Otherwise return NULL.
+ */
+ir_rvalue *
+lower_cull_distance_visitor::lower_cull_distance_vec8(ir_rvalue *ir)
+{
+   if (this->old_cull_distance_1d_var) {
+      ir_dereference_variable *var_ref = ir->as_dereference_variable();
+      if (var_ref && var_ref->var == this->old_cull_distance_1d_var) {
+         return new(ralloc_parent(ir))
+            ir_dereference_variable(this->new_cull_distance_1d_var);
+      }
+   }
+   if (this->old_cull_distance_2d_var) {
+      /* 2D cull distance is only possible as a geometry input */
+      assert(this->shader_stage == MESA_SHADER_GEOMETRY);
+
+      ir_dereference_array *array_ref = ir->as_dereference_array();
+      if (array_ref) {
+         ir_dereference_variable *var_ref =
+            array_ref->array->as_dereference_variable();
+         if (var_ref && var_ref->var == this->old_cull_distance_2d_var) {
+            return new(ralloc_parent(ir))
+               ir_dereference_array(this->new_cull_distance_2d_var,
+                                    array_ref->array_index);
+         }
+      }
+   }
+   return NULL;
+}
+
+
+void
+lower_cull_distance_visitor::handle_rvalue(ir_rvalue **rv)
+{
+   if (*rv == NULL)
+      return;
+
+   ir_dereference_array *const array_deref = (*rv)->as_dereference_array();
+   if (array_deref == NULL)
+      return;
+
+   /* Replace any expression that indexes one of the floats in gl_CullDistance
+    * with an expression that indexes into one of the vec4's in
+    * gl_CullDistanceMESA and accesses the appropriate component.
+    */
+   ir_rvalue *lowered_vec8 =
+      this->lower_cull_distance_vec8(array_deref->array);
+   if (lowered_vec8 != NULL) {
+      this->progress = true;
+      ir_rvalue *array_index;
+      ir_rvalue *swizzle_index;
+      this->create_indices(array_deref->array_index, array_index, swizzle_index);
+      void *mem_ctx = ralloc_parent(array_deref);
+
+      ir_dereference_array *const new_array_deref =
+         new(mem_ctx) ir_dereference_array(lowered_vec8, array_index);
+
+      ir_expression *const expr =
+         new(mem_ctx) ir_expression(ir_binop_vector_extract,
+                                    new_array_deref,
+                                    swizzle_index);
+
+      *rv = expr;
+   }
+}
+
+void
+lower_cull_distance_visitor::fix_lhs(ir_assignment *ir)
+{
+   if (ir->lhs->ir_type == ir_type_expression) {
+      void *mem_ctx = ralloc_parent(ir);
+      ir_expression *const expr = (ir_expression *) ir->lhs;
+
+      /* The expression must be of the form:
+       *
+       *     (vector_extract gl_CullDistanceMESA[i], j).
+       */
+      assert(expr->operation == ir_binop_vector_extract);
+      assert(expr->operands[0]->ir_type == ir_type_dereference_array);
+      assert(expr->operands[0]->type == glsl_type::vec4_type);
+
+      ir_dereference *const new_lhs = (ir_dereference *) expr->operands[0];
+      ir->rhs = new(mem_ctx) ir_expression(ir_triop_vector_insert,
+					   glsl_type::vec4_type,
+					   new_lhs->clone(mem_ctx, NULL),
+					   ir->rhs,
+					   expr->operands[1]);
+      ir->set_lhs(new_lhs);
+      ir->write_mask = WRITEMASK_XYZW;
+   }
+}
+
+/**
+ * Replace any assignment having the 1D gl_CullDistance (undereferenced) as
+ * its LHS or RHS with a sequence of assignments, one for each component of
+ * the array.  Each of these assignments is lowered to refer to
+ * gl_CullDistanceMESA as appropriate.
+ *
+ * We need to do a similar replacement for 2D gl_CullDistance, however since
+ * it's an input, the only case we need to address is where a 1D slice of it
+ * is the entire RHS of an assignment, e.g.:
+ *
+ *     foo = gl_in[i].gl_CullDistance
+ */
+ir_visitor_status
+lower_cull_distance_visitor::visit_leave(ir_assignment *ir)
+{
+   /* First invoke the base class visitor.  This causes handle_rvalue() to be
+    * called on ir->rhs and ir->condition.
+    */
+   ir_rvalue_visitor::visit_leave(ir);
+
+   if (this->is_cull_distance_vec8(ir->lhs) ||
+       this->is_cull_distance_vec8(ir->rhs)) {
+      /* LHS or RHS of the assignment is the entire 1D gl_CullDistance array
+       * (or a 1D slice of a 2D gl_CullDistance input array).  Since we are
+       * reshaping gl_CullDistance from an array of floats to an array of
+       * vec4's, this isn't going to work as a bulk assignment anymore, so
+       * unroll it to element-by-element assignments and lower each of them.
+       *
+       * Note: to unroll into element-by-element assignments, we need to make
+       * clones of the LHS and RHS.  This is safe because expressions and
+       * l-values are side-effect free.
+       */
+      void *ctx = ralloc_parent(ir);
+      int array_size = ir->lhs->type->array_size();
+      for (int i = 0; i < array_size; ++i) {
+         ir_dereference_array *new_lhs = new(ctx) ir_dereference_array(
+            ir->lhs->clone(ctx, NULL), new(ctx) ir_constant(i));
+         ir_dereference_array *new_rhs = new(ctx) ir_dereference_array(
+            ir->rhs->clone(ctx, NULL), new(ctx) ir_constant(i));
+         this->handle_rvalue((ir_rvalue **) &new_rhs);
+
+         /* Handle the LHS after creating the new assignment.  This must
+          * happen in this order because handle_rvalue may replace the old LHS
+          * with an ir_expression of ir_binop_vector_extract.  Since this is
+          * not a valide l-value, this will cause an assertion in the
+          * ir_assignment constructor to fail.
+          *
+          * If this occurs, replace the mangled LHS with a dereference of the
+          * vector, and replace the RHS with an ir_triop_vector_insert.
+          */
+         ir_assignment *const assign = new(ctx) ir_assignment(new_lhs, new_rhs);
+         this->handle_rvalue((ir_rvalue **) &assign->lhs);
+         this->fix_lhs(assign);
+
+         this->base_ir->insert_before(assign);
+      }
+      ir->remove();
+
+      return visit_continue;
+   }
+
+   /* Handle the LHS as if it were an r-value.  Normally
+    * rvalue_visit(ir_assignment *) only visits the RHS, but we need to lower
+    * expressions in the LHS as well.
+    *
+    * This may cause the LHS to get replaced with an ir_expression of
+    * ir_binop_vector_extract.  If this occurs, replace it with a dereference
+    * of the vector, and replace the RHS with an ir_triop_vector_insert.
+    */
+   handle_rvalue((ir_rvalue **)&ir->lhs);
+   this->fix_lhs(ir);
+
+   return rvalue_visit(ir);
+}
+
+
+/**
+ * Set up base_ir properly and call visit_leave() on a newly created
+ * ir_assignment node.  This is used in cases where we have to insert an
+ * ir_assignment in a place where we know the hierarchical visitor won't see
+ * it.
+ */
+void
+lower_cull_distance_visitor::visit_new_assignment(ir_assignment *ir)
+{
+   ir_instruction *old_base_ir = this->base_ir;
+   this->base_ir = ir;
+   ir->accept(this);
+   this->base_ir = old_base_ir;
+}
+
+
+/**
+ * If a 1D gl_CullDistance variable appears as an argument in an ir_call
+ * expression, replace it with a temporary variable, and make sure the ir_call
+ * is preceded and/or followed by assignments that copy the contents of the
+ * temporary variable to and/or from gl_CullDistance.  Each of these
+ * assignments is then lowered to refer to gl_CullDistanceMESA.
+ *
+ * We need to do a similar replacement for 2D gl_CullDistance, however since
+ * it's an input, the only case we need to address is where a 1D slice of it
+ * is passed as an "in" parameter to an ir_call, e.g.:
+ *
+ *     foo(gl_in[i].gl_CullDistance)
+ */
+ir_visitor_status
+lower_cull_distance_visitor::visit_leave(ir_call *ir)
+{
+   void *ctx = ralloc_parent(ir);
+
+   const exec_node *formal_param_node = ir->callee->parameters.head;
+   const exec_node *actual_param_node = ir->actual_parameters.head;
+   while (!actual_param_node->is_tail_sentinel()) {
+      ir_variable *formal_param = (ir_variable *) formal_param_node;
+      ir_rvalue *actual_param = (ir_rvalue *) actual_param_node;
+
+      /* Advance formal_param_node and actual_param_node now so that we can
+       * safely replace actual_param with another node, if necessary, below.
+       */
+      formal_param_node = formal_param_node->next;
+      actual_param_node = actual_param_node->next;
+
+      if (this->is_cull_distance_vec8(actual_param)) {
+         /* User is trying to pass the whole 1D gl_CullDistance array (or a 1D
+          * slice of a 2D gl_CullDistance array) to a function call.  Since we
+          * are reshaping gl_CullDistance from an array of floats to an array
+          * of vec4's, this isn't going to work anymore, so use a temporary
+          * array instead.
+          */
+         ir_variable *temp_cull_distance = new(ctx) ir_variable(
+            actual_param->type, "temp_cull_distance", ir_var_temporary);
+         this->base_ir->insert_before(temp_cull_distance);
+         actual_param->replace_with(
+            new(ctx) ir_dereference_variable(temp_cull_distance));
+         if (formal_param->data.mode == ir_var_function_in
+             || formal_param->data.mode == ir_var_function_inout) {
+            /* Copy from gl_CullDistance to the temporary before the call.
+             * Since we are going to insert this copy before the current
+             * instruction, we need to visit it afterwards to make sure it
+             * gets lowered.
+             */
+            ir_assignment *new_assignment = new(ctx) ir_assignment(
+               new(ctx) ir_dereference_variable(temp_cull_distance),
+               actual_param->clone(ctx, NULL));
+            this->base_ir->insert_before(new_assignment);
+            this->visit_new_assignment(new_assignment);
+         }
+         if (formal_param->data.mode == ir_var_function_out
+             || formal_param->data.mode == ir_var_function_inout) {
+            /* Copy from the temporary to gl_CullDistance after the call.
+             * Since visit_list_elements() has already decided which
+             * instruction it's going to visit next, we need to visit
+             * afterwards to make sure it gets lowered.
+             */
+            ir_assignment *new_assignment = new(ctx) ir_assignment(
+               actual_param->clone(ctx, NULL),
+               new(ctx) ir_dereference_variable(temp_cull_distance));
+            this->base_ir->insert_after(new_assignment);
+            this->visit_new_assignment(new_assignment);
+         }
+      }
+   }
+
+   return rvalue_visit(ir);
+}
+
+
+bool
+lower_cull_distance(gl_shader *shader)
+{
+   lower_cull_distance_visitor v(shader->Stage);
+
+   visit_list_elements(&v, shader->ir);
+
+   if (v.new_cull_distance_1d_var)
+      shader->symbols->add_variable(v.new_cull_distance_1d_var);
+   if (v.new_cull_distance_2d_var)
+      shader->symbols->add_variable(v.new_cull_distance_2d_var);
+
+   return v.progress;
+}
-- 
2.4.1



More information about the Nouveau mailing list