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 "ir_hierarchical_visitor.h"
49#include "ir.h"
50
51class lower_clip_distance_visitor : public ir_hierarchical_visitor {
52public:
53   lower_clip_distance_visitor()
54      : progress(false), old_clip_distance_var(NULL),
55        new_clip_distance_var(NULL)
56   {
57   }
58
59   virtual ir_visitor_status visit(ir_variable *);
60   void create_indices(ir_rvalue*, ir_rvalue *&, ir_rvalue *&);
61   virtual ir_visitor_status visit_leave(ir_dereference_array *);
62   virtual ir_visitor_status visit_leave(ir_assignment *);
63   void visit_new_assignment(ir_assignment *ir);
64   virtual ir_visitor_status visit_leave(ir_call *);
65
66   bool progress;
67
68   /**
69    * Pointer to the declaration of gl_ClipDistance, if found.
70    */
71   ir_variable *old_clip_distance_var;
72
73   /**
74    * Pointer to the newly-created gl_ClipDistanceMESA variable.
75    */
76   ir_variable *new_clip_distance_var;
77};
78
79
80/**
81 * Replace any declaration of gl_ClipDistance as an array of floats with a
82 * declaration of gl_ClipDistanceMESA as an array of vec4's.
83 */
84ir_visitor_status
85lower_clip_distance_visitor::visit(ir_variable *ir)
86{
87   /* No point in looking for the declaration of gl_ClipDistance if
88    * we've already found it.
89    */
90   if (this->old_clip_distance_var)
91      return visit_continue;
92
93   if (ir->name && strcmp(ir->name, "gl_ClipDistance") == 0) {
94      this->progress = true;
95      this->old_clip_distance_var = ir;
96      assert (ir->type->is_array());
97      assert (ir->type->element_type() == glsl_type::float_type);
98      unsigned new_size = (ir->type->array_size() + 3) / 4;
99
100      /* Clone the old var so that we inherit all of its properties */
101      this->new_clip_distance_var = ir->clone(ralloc_parent(ir), NULL);
102
103      /* And change the properties that we need to change */
104      this->new_clip_distance_var->name
105         = ralloc_strdup(this->new_clip_distance_var, "gl_ClipDistanceMESA");
106      this->new_clip_distance_var->type
107         = glsl_type::get_array_instance(glsl_type::vec4_type, new_size);
108      this->new_clip_distance_var->max_array_access = ir->max_array_access / 4;
109
110      ir->replace_with(this->new_clip_distance_var);
111   }
112   return visit_continue;
113}
114
115
116/**
117 * Create the necessary GLSL rvalues to index into gl_ClipDistanceMESA based
118 * on the rvalue previously used to index into gl_ClipDistance.
119 *
120 * \param array_index Selects one of the vec4's in gl_ClipDistanceMESA
121 * \param swizzle_index Selects a component within the vec4 selected by
122 *        array_index.
123 */
124void
125lower_clip_distance_visitor::create_indices(ir_rvalue *old_index,
126                                            ir_rvalue *&array_index,
127                                            ir_rvalue *&swizzle_index)
128{
129   void *ctx = ralloc_parent(old_index);
130
131   /* Make sure old_index is a signed int so that the bitwise "shift" and
132    * "and" operations below type check properly.
133    */
134   if (old_index->type != glsl_type::int_type) {
135      assert (old_index->type == glsl_type::uint_type);
136      old_index = new(ctx) ir_expression(ir_unop_u2i, old_index);
137   }
138
139   ir_constant *old_index_constant = old_index->constant_expression_value();
140   if (old_index_constant) {
141      /* gl_ClipDistance is being accessed via a constant index.  Don't bother
142       * creating expressions to calculate the lowered indices.  Just create
143       * constants.
144       */
145      int const_val = old_index_constant->get_int_component(0);
146      array_index = new(ctx) ir_constant(const_val / 4);
147      swizzle_index = new(ctx) ir_constant(const_val % 4);
148   } else {
149      /* Create a variable to hold the value of old_index (so that we
150       * don't compute it twice).
151       */
152      ir_variable *old_index_var = new(ctx) ir_variable(
153         glsl_type::int_type, "clip_distance_index", ir_var_temporary);
154      this->base_ir->insert_before(old_index_var);
155      this->base_ir->insert_before(new(ctx) ir_assignment(
156         new(ctx) ir_dereference_variable(old_index_var), old_index));
157
158      /* Create the expression clip_distance_index / 4.  Do this as a bit
159       * shift because that's likely to be more efficient.
160       */
161      array_index = new(ctx) ir_expression(
162         ir_binop_rshift, new(ctx) ir_dereference_variable(old_index_var),
163         new(ctx) ir_constant(2));
164
165      /* Create the expression clip_distance_index % 4.  Do this as a bitwise
166       * AND because that's likely to be more efficient.
167       */
168      swizzle_index = new(ctx) ir_expression(
169         ir_binop_bit_and, new(ctx) ir_dereference_variable(old_index_var),
170         new(ctx) ir_constant(3));
171   }
172}
173
174
175/**
176 * Replace any expression that indexes into the gl_ClipDistance array with an
177 * expression that indexes into one of the vec4's in gl_ClipDistanceMESA and
178 * accesses the appropriate component.
179 */
180ir_visitor_status
181lower_clip_distance_visitor::visit_leave(ir_dereference_array *ir)
182{
183   /* If the gl_ClipDistance var hasn't been declared yet, then
184    * there's no way this deref can refer to it.
185    */
186   if (!this->old_clip_distance_var)
187      return visit_continue;
188
189   ir_dereference_variable *old_var_ref = ir->array->as_dereference_variable();
190   if (old_var_ref && old_var_ref->var == this->old_clip_distance_var) {
191      this->progress = true;
192      ir_rvalue *array_index;
193      ir_rvalue *swizzle_index;
194      this->create_indices(ir->array_index, array_index, swizzle_index);
195      void *mem_ctx = ralloc_parent(ir);
196      ir->array = new(mem_ctx) ir_dereference_array(
197         this->new_clip_distance_var, array_index);
198      ir->array_index = swizzle_index;
199   }
200
201   return visit_continue;
202}
203
204
205/**
206 * Replace any assignment having gl_ClipDistance (undereferenced) as its LHS
207 * or RHS with a sequence of assignments, one for each component of the array.
208 * Each of these assignments is lowered to refer to gl_ClipDistanceMESA as
209 * appropriate.
210 */
211ir_visitor_status
212lower_clip_distance_visitor::visit_leave(ir_assignment *ir)
213{
214   ir_dereference_variable *lhs_var = ir->lhs->as_dereference_variable();
215   ir_dereference_variable *rhs_var = ir->rhs->as_dereference_variable();
216   if ((lhs_var && lhs_var->var == this->old_clip_distance_var)
217       || (rhs_var && rhs_var->var == this->old_clip_distance_var)) {
218      /* LHS or RHS of the assignment is the entire gl_ClipDistance array.
219       * Since we are reshaping gl_ClipDistance from an array of floats to an
220       * array of vec4's, this isn't going to work as a bulk assignment
221       * anymore, so unroll it to element-by-element assignments and lower
222       * each of them.
223       *
224       * Note: to unroll into element-by-element assignments, we need to make
225       * clones of the LHS and RHS.  This is only safe if the LHS and RHS are
226       * side-effect free.  Fortunately, we know that they are, because the
227       * only kind of rvalue that can have side effects is an ir_call, and
228       * ir_calls only appear (a) as a statement on their own, or (b) as the
229       * RHS of an assignment that stores the result of the call in a
230       * temporary variable.
231       */
232      void *ctx = ralloc_parent(ir);
233      int array_size = this->old_clip_distance_var->type->array_size();
234      for (int i = 0; i < array_size; ++i) {
235         ir_dereference_array *new_lhs = new(ctx) ir_dereference_array(
236            ir->lhs->clone(ctx, NULL), new(ctx) ir_constant(i));
237         new_lhs->accept(this);
238         ir_dereference_array *new_rhs = new(ctx) ir_dereference_array(
239            ir->rhs->clone(ctx, NULL), new(ctx) ir_constant(i));
240         new_rhs->accept(this);
241         this->base_ir->insert_before(
242            new(ctx) ir_assignment(new_lhs, new_rhs));
243      }
244      ir->remove();
245   }
246
247   return visit_continue;
248}
249
250
251/**
252 * Set up base_ir properly and call visit_leave() on a newly created
253 * ir_assignment node.  This is used in cases where we have to insert an
254 * ir_assignment in a place where we know the hierarchical visitor won't see
255 * it.
256 */
257void
258lower_clip_distance_visitor::visit_new_assignment(ir_assignment *ir)
259{
260   ir_instruction *old_base_ir = this->base_ir;
261   this->base_ir = ir;
262   ir->accept(this);
263   this->base_ir = old_base_ir;
264}
265
266
267/**
268 * If gl_ClipDistance appears as an argument in an ir_call expression, replace
269 * it with a temporary variable, and make sure the ir_call is preceded and/or
270 * followed by assignments that copy the contents of the temporary variable to
271 * and/or from gl_ClipDistance.  Each of these assignments is then lowered to
272 * refer to gl_ClipDistanceMESA.
273 */
274ir_visitor_status
275lower_clip_distance_visitor::visit_leave(ir_call *ir)
276{
277   void *ctx = ralloc_parent(ir);
278
279   const exec_node *formal_param_node = ir->callee->parameters.head;
280   const exec_node *actual_param_node = ir->actual_parameters.head;
281   while (!actual_param_node->is_tail_sentinel()) {
282      ir_variable *formal_param = (ir_variable *) formal_param_node;
283      ir_rvalue *actual_param = (ir_rvalue *) actual_param_node;
284
285      /* Advance formal_param_node and actual_param_node now so that we can
286       * safely replace actual_param with another node, if necessary, below.
287       */
288      formal_param_node = formal_param_node->next;
289      actual_param_node = actual_param_node->next;
290
291      ir_dereference_variable *deref = actual_param->as_dereference_variable();
292      if (deref && deref->var == this->old_clip_distance_var) {
293         /* User is trying to pass the whole gl_ClipDistance array to a
294          * function call.  Since we are reshaping gl_ClipDistance from an
295          * array of floats to an array of vec4's, this isn't going to work
296          * anymore, so use a temporary array instead.
297          */
298         ir_variable *temp_clip_distance = new(ctx) ir_variable(
299            actual_param->type, "temp_clip_distance", ir_var_temporary);
300         this->base_ir->insert_before(temp_clip_distance);
301         actual_param->replace_with(
302            new(ctx) ir_dereference_variable(temp_clip_distance));
303         if (formal_param->mode == ir_var_in
304             || formal_param->mode == ir_var_inout) {
305            /* Copy from gl_ClipDistance to the temporary before the call.
306             * Since we are going to insert this copy before the current
307             * instruction, we need to visit it afterwards to make sure it
308             * gets lowered.
309             */
310            ir_assignment *new_assignment = new(ctx) ir_assignment(
311               new(ctx) ir_dereference_variable(temp_clip_distance),
312               new(ctx) ir_dereference_variable(old_clip_distance_var));
313            this->base_ir->insert_before(new_assignment);
314            this->visit_new_assignment(new_assignment);
315         }
316         if (formal_param->mode == ir_var_out
317             || formal_param->mode == ir_var_inout) {
318            /* Copy from the temporary to gl_ClipDistance after the call.
319             * Since visit_list_elements() has already decided which
320             * instruction it's going to visit next, we need to visit
321             * afterwards to make sure it gets lowered.
322             */
323            ir_assignment *new_assignment = new(ctx) ir_assignment(
324               new(ctx) ir_dereference_variable(old_clip_distance_var),
325               new(ctx) ir_dereference_variable(temp_clip_distance));
326            this->base_ir->insert_after(new_assignment);
327            this->visit_new_assignment(new_assignment);
328         }
329      }
330   }
331
332   return visit_continue;
333}
334
335
336bool
337lower_clip_distance(exec_list *instructions)
338{
339   lower_clip_distance_visitor v;
340
341   visit_list_elements(&v, instructions);
342
343   return v.progress;
344}
345