src/glsl/lower_clip_distance.cpp

   1 /*
   2  * Copyright © 2011 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  21  * DEALINGS IN THE SOFTWARE.
  22  */
  23
  24 /**
  25  * \file lower_clip_distance.cpp
  26  *
  27  * This pass accounts for the difference between the way
  28  * gl_ClipDistance is declared in standard GLSL (as an array of
  29  * floats), and the way it is frequently implemented in hardware (as
  30  * a pair of vec4s, with four clip distances packed into each).
  31  *
  32  * The declaration of gl_ClipDistance is replaced with a declaration
  33  * of gl_ClipDistanceMESA, and any references to gl_ClipDistance are
  34  * translated to refer to gl_ClipDistanceMESA with the appropriate
  35  * swizzling of array indices.  For instance:
  36  *
  37  *   gl_ClipDistance[i]
  38  *
  39  * is translated into:
  40  *
  41  *   gl_ClipDistanceMESA[i>>2][i&3]
  42  *
  43  * Since some hardware may not internally represent gl_ClipDistance as a pair
  44  * of vec4's, this lowering pass is optional.  To enable it, set the
  45  * LowerClipDistance flag in gl_shader_compiler_options to true.
  46  */
  47
  48 #include "glsl_symbol_table.h"
  49 #include "ir_hierarchical_visitor.h"
  50 #include "ir.h"
  51
  52 class lower_clip_distance_visitor : public ir_hierarchical_visitor {
  53 public:
  54    lower_clip_distance_visitor()
  55       : progress(false), old_clip_distance_var(NULL),
  56         new_clip_distance_var(NULL)
  57    {
  58    }
  59
  60    virtual ir_visitor_status visit(ir_variable *);
  61    void create_indices(ir_rvalue*, ir_rvalue *&, ir_rvalue *&);
  62    virtual ir_visitor_status visit_leave(ir_dereference_array *);
  63    virtual ir_visitor_status visit_leave(ir_assignment *);
  64    void visit_new_assignment(ir_assignment *ir);
  65    virtual ir_visitor_status visit_leave(ir_call *);
  66
  67    bool progress;
  68
  69    /**
  70     * Pointer to the declaration of gl_ClipDistance, if found.
  71     */
  72    ir_variable *old_clip_distance_var;
  73
  74    /**
  75     * Pointer to the newly-created gl_ClipDistanceMESA variable.
  76     */
  77    ir_variable *new_clip_distance_var;
  78 };
  79
  80
  81 /**
  82  * Replace any declaration of gl_ClipDistance as an array of floats with a
  83  * declaration of gl_ClipDistanceMESA as an array of vec4's.
  84  */
  85 ir_visitor_status
  86 lower_clip_distance_visitor::visit(ir_variable *ir)
  87 {
  88    /* No point in looking for the declaration of gl_ClipDistance if
  89     * we've already found it.
  90     */
  91    if (this->old_clip_distance_var)
  92       return visit_continue;
  93
  94    if (ir->name && strcmp(ir->name, "gl_ClipDistance") == 0) {
  95       this->progress = true;
  96       this->old_clip_distance_var = ir;
  97       assert (ir->type->is_array());
  98       assert (ir->type->element_type() == glsl_type::float_type);
  99       unsigned new_size = (ir->type->array_size() + 3) / 4;
 100
 101       /* Clone the old var so that we inherit all of its properties */
 102       this->new_clip_distance_var = ir->clone(ralloc_parent(ir), NULL);
 103
 104       /* And change the properties that we need to change */
 105       this->new_clip_distance_var->name
 106          = ralloc_strdup(this->new_clip_distance_var, "gl_ClipDistanceMESA");
 107       this->new_clip_distance_var->type
 108          = glsl_type::get_array_instance(glsl_type::vec4_type, new_size);
 109       this->new_clip_distance_var->max_array_access = ir->max_array_access / 4;
 110
 111       ir->replace_with(this->new_clip_distance_var);
 112    }
 113    return visit_continue;
 114 }
 115
 116
 117 /**
 118  * Create the necessary GLSL rvalues to index into gl_ClipDistanceMESA based
 119  * on the rvalue previously used to index into gl_ClipDistance.
 120  *
 121  * \param array_index Selects one of the vec4's in gl_ClipDistanceMESA
 122  * \param swizzle_index Selects a component within the vec4 selected by
 123  *        array_index.
 124  */
 125 void
 126 lower_clip_distance_visitor::create_indices(ir_rvalue *old_index,
 127                                             ir_rvalue *&array_index,
 128                                             ir_rvalue *&swizzle_index)
 129 {
 130    void *ctx = ralloc_parent(old_index);
 131
 132    /* Make sure old_index is a signed int so that the bitwise "shift" and
 133     * "and" operations below type check properly.
 134     */
 135    if (old_index->type != glsl_type::int_type) {
 136       assert (old_index->type == glsl_type::uint_type);
 137       old_index = new(ctx) ir_expression(ir_unop_u2i, old_index);
 138    }
 139
 140    ir_constant *old_index_constant = old_index->constant_expression_value();
 141    if (old_index_constant) {
 142       /* gl_ClipDistance is being accessed via a constant index.  Don't bother
 143        * creating expressions to calculate the lowered indices.  Just create
 144        * constants.
 145        */
 146       int const_val = old_index_constant->get_int_component(0);
 147       array_index = new(ctx) ir_constant(const_val / 4);
 148       swizzle_index = new(ctx) ir_constant(const_val % 4);
 149    } else {
 150       /* Create a variable to hold the value of old_index (so that we
 151        * don't compute it twice).
 152        */
 153       ir_variable *old_index_var = new(ctx) ir_variable(
 154          glsl_type::int_type, "clip_distance_index", ir_var_temporary);
 155       this->base_ir->insert_before(old_index_var);
 156       this->base_ir->insert_before(new(ctx) ir_assignment(
 157          new(ctx) ir_dereference_variable(old_index_var), old_index));
 158
 159       /* Create the expression clip_distance_index / 4.  Do this as a bit
 160        * shift because that's likely to be more efficient.
 161        */
 162       array_index = new(ctx) ir_expression(
 163          ir_binop_rshift, new(ctx) ir_dereference_variable(old_index_var),
 164          new(ctx) ir_constant(2));
 165
 166       /* Create the expression clip_distance_index % 4.  Do this as a bitwise
 167        * AND because that's likely to be more efficient.
 168        */
 169       swizzle_index = new(ctx) ir_expression(
 170          ir_binop_bit_and, new(ctx) ir_dereference_variable(old_index_var),
 171          new(ctx) ir_constant(3));
 172    }
 173 }
 174
 175
 176 /**
 177  * Replace any expression that indexes into the gl_ClipDistance array with an
 178  * expression that indexes into one of the vec4's in gl_ClipDistanceMESA and
 179  * accesses the appropriate component.
 180  */
 181 ir_visitor_status
 182 lower_clip_distance_visitor::visit_leave(ir_dereference_array *ir)
 183 {
 184    /* If the gl_ClipDistance var hasn't been declared yet, then
 185     * there's no way this deref can refer to it.
 186     */
 187    if (!this->old_clip_distance_var)
 188       return visit_continue;
 189
 190    ir_dereference_variable *old_var_ref = ir->array->as_dereference_variable();
 191    if (old_var_ref && old_var_ref->var == this->old_clip_distance_var) {
 192       this->progress = true;
 193       ir_rvalue *array_index;
 194       ir_rvalue *swizzle_index;
 195       this->create_indices(ir->array_index, array_index, swizzle_index);
 196       void *mem_ctx = ralloc_parent(ir);
 197       ir->array = new(mem_ctx) ir_dereference_array(
 198          this->new_clip_distance_var, array_index);
 199       ir->array_index = swizzle_index;
 200    }
 201
 202    return visit_continue;
 203 }
 204
 205
 206 /**
 207  * Replace any assignment having gl_ClipDistance (undereferenced) as its LHS
 208  * or RHS with a sequence of assignments, one for each component of the array.
 209  * Each of these assignments is lowered to refer to gl_ClipDistanceMESA as
 210  * appropriate.
 211  */
 212 ir_visitor_status
 213 lower_clip_distance_visitor::visit_leave(ir_assignment *ir)
 214 {
 215    ir_dereference_variable *lhs_var = ir->lhs->as_dereference_variable();
 216    ir_dereference_variable *rhs_var = ir->rhs->as_dereference_variable();
 217    if ((lhs_var && lhs_var->var == this->old_clip_distance_var)
 218        || (rhs_var && rhs_var->var == this->old_clip_distance_var)) {
 219       /* LHS or RHS of the assignment is the entire gl_ClipDistance array.
 220        * Since we are reshaping gl_ClipDistance from an array of floats to an
 221        * array of vec4's, this isn't going to work as a bulk assignment
 222        * anymore, so unroll it to element-by-element assignments and lower
 223        * each of them.
 224        *
 225        * Note: to unroll into element-by-element assignments, we need to make
 226        * clones of the LHS and RHS.  This is only safe if the LHS and RHS are
 227        * side-effect free.  Fortunately, we know that they are, because the
 228        * only kind of rvalue that can have side effects is an ir_call, and
 229        * ir_calls only appear (a) as a statement on their own, or (b) as the
 230        * RHS of an assignment that stores the result of the call in a
 231        * temporary variable.
 232        */
 233       void *ctx = ralloc_parent(ir);
 234       int array_size = this->old_clip_distance_var->type->array_size();
 235       for (int i = 0; i < array_size; ++i) {
 236          ir_dereference_array *new_lhs = new(ctx) ir_dereference_array(
 237             ir->lhs->clone(ctx, NULL), new(ctx) ir_constant(i));
 238          new_lhs->accept(this);
 239          ir_dereference_array *new_rhs = new(ctx) ir_dereference_array(
 240             ir->rhs->clone(ctx, NULL), new(ctx) ir_constant(i));
 241          new_rhs->accept(this);
 242          this->base_ir->insert_before(
 243             new(ctx) ir_assignment(new_lhs, new_rhs));
 244       }
 245       ir->remove();
 246    }
 247
 248    return visit_continue;
 249 }
 250
 251
 252 /**
 253  * Set up base_ir properly and call visit_leave() on a newly created
 254  * ir_assignment node.  This is used in cases where we have to insert an
 255  * ir_assignment in a place where we know the hierarchical visitor won't see
 256  * it.
 257  */
 258 void
 259 lower_clip_distance_visitor::visit_new_assignment(ir_assignment *ir)
 260 {
 261    ir_instruction *old_base_ir = this->base_ir;
 262    this->base_ir = ir;
 263    ir->accept(this);
 264    this->base_ir = old_base_ir;
 265 }
 266
 267
 268 /**
 269  * If gl_ClipDistance appears as an argument in an ir_call expression, replace
 270  * it with a temporary variable, and make sure the ir_call is preceded and/or
 271  * followed by assignments that copy the contents of the temporary variable to
 272  * and/or from gl_ClipDistance.  Each of these assignments is then lowered to
 273  * refer to gl_ClipDistanceMESA.
 274  */
 275 ir_visitor_status
 276 lower_clip_distance_visitor::visit_leave(ir_call *ir)
 277 {
 278    void *ctx = ralloc_parent(ir);
 279
 280    const exec_node *formal_param_node = ir->callee->parameters.head;
 281    const exec_node *actual_param_node = ir->actual_parameters.head;
 282    while (!actual_param_node->is_tail_sentinel()) {
 283       ir_variable *formal_param = (ir_variable *) formal_param_node;
 284       ir_rvalue *actual_param = (ir_rvalue *) actual_param_node;
 285
 286       /* Advance formal_param_node and actual_param_node now so that we can
 287        * safely replace actual_param with another node, if necessary, below.
 288        */
 289       formal_param_node = formal_param_node->next;
 290       actual_param_node = actual_param_node->next;
 291
 292       ir_dereference_variable *deref = actual_param->as_dereference_variable();
 293       if (deref && deref->var == this->old_clip_distance_var) {
 294          /* User is trying to pass the whole gl_ClipDistance array to a
 295           * function call.  Since we are reshaping gl_ClipDistance from an
 296           * array of floats to an array of vec4's, this isn't going to work
 297           * anymore, so use a temporary array instead.
 298           */
 299          ir_variable *temp_clip_distance = new(ctx) ir_variable(
 300             actual_param->type, "temp_clip_distance", ir_var_temporary);
 301          this->base_ir->insert_before(temp_clip_distance);
 302          actual_param->replace_with(
 303             new(ctx) ir_dereference_variable(temp_clip_distance));
 304          if (formal_param->mode == ir_var_function_in
 305              || formal_param->mode == ir_var_function_inout) {
 306             /* Copy from gl_ClipDistance to the temporary before the call.
 307              * Since we are going to insert this copy before the current
 308              * instruction, we need to visit it afterwards to make sure it
 309              * gets lowered.
 310              */
 311             ir_assignment *new_assignment = new(ctx) ir_assignment(
 312                new(ctx) ir_dereference_variable(temp_clip_distance),
 313                new(ctx) ir_dereference_variable(old_clip_distance_var));
 314             this->base_ir->insert_before(new_assignment);
 315             this->visit_new_assignment(new_assignment);
 316          }
 317          if (formal_param->mode == ir_var_function_out
 318              || formal_param->mode == ir_var_function_inout) {
 319             /* Copy from the temporary to gl_ClipDistance after the call.
 320              * Since visit_list_elements() has already decided which
 321              * instruction it's going to visit next, we need to visit
 322              * afterwards to make sure it gets lowered.
 323              */
 324             ir_assignment *new_assignment = new(ctx) ir_assignment(
 325                new(ctx) ir_dereference_variable(old_clip_distance_var),
 326                new(ctx) ir_dereference_variable(temp_clip_distance));
 327             this->base_ir->insert_after(new_assignment);
 328             this->visit_new_assignment(new_assignment);
 329          }
 330       }
 331    }
 332
 333    return visit_continue;
 334 }
 335
 336
 337 bool
 338 lower_clip_distance(gl_shader *shader)
 339 {
 340    lower_clip_distance_visitor v;
 341
 342    visit_list_elements(&v, shader->ir);
 343
 344    if (v.new_clip_distance_var)
 345       shader->symbols->add_variable(v.new_clip_distance_var);
 346
 347    return v.progress;
 348 }