1 /*
2  * Copyright © 2010 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 opt_function_inlining.cpp
26  *
27  * Replaces calls to functions with the body of the function.
28  */
29 
30 #include "ir.h"
31 #include "ir_visitor.h"
32 #include "ir_function_inlining.h"
33 #include "ir_expression_flattening.h"
34 #include "glsl_types.h"
35 #include "program/hash_table.h"
36 
37 static void
38 do_sampler_replacement(exec_list *instructions,
39 		       ir_variable *sampler,
40 		       ir_dereference *deref);
41 
42 namespace {
43 
44 class ir_function_inlining_visitor : public ir_hierarchical_visitor {
45 public:
ir_function_inlining_visitor()46    ir_function_inlining_visitor()
47    {
48       progress = false;
49    }
50 
~ir_function_inlining_visitor()51    virtual ~ir_function_inlining_visitor()
52    {
53       /* empty */
54    }
55 
56    virtual ir_visitor_status visit_enter(ir_expression *);
57    virtual ir_visitor_status visit_enter(ir_call *);
58    virtual ir_visitor_status visit_enter(ir_return *);
59    virtual ir_visitor_status visit_enter(ir_texture *);
60    virtual ir_visitor_status visit_enter(ir_swizzle *);
61 
62    bool progress;
63 };
64 
65 } /* unnamed namespace */
66 
67 bool
do_function_inlining(exec_list * instructions)68 do_function_inlining(exec_list *instructions)
69 {
70    ir_function_inlining_visitor v;
71 
72    v.run(instructions);
73 
74    return v.progress;
75 }
76 
77 static void
replace_return_with_assignment(ir_instruction * ir,void * data)78 replace_return_with_assignment(ir_instruction *ir, void *data)
79 {
80    void *ctx = ralloc_parent(ir);
81    ir_dereference *orig_deref = (ir_dereference *) data;
82    ir_return *ret = ir->as_return();
83 
84    if (ret) {
85       if (ret->value) {
86 	 ir_rvalue *lhs = orig_deref->clone(ctx, NULL);
87 	 ret->replace_with(new(ctx) ir_assignment(lhs, ret->value, NULL));
88       } else {
89 	 /* un-valued return has to be the last return, or we shouldn't
90 	  * have reached here. (see can_inline()).
91 	  */
92 	 assert(ret->next->is_tail_sentinel());
93 	 ret->remove();
94       }
95    }
96 }
97 
98 void
generate_inline(ir_instruction * next_ir)99 ir_call::generate_inline(ir_instruction *next_ir)
100 {
101    void *ctx = ralloc_parent(this);
102    ir_variable **parameters;
103    int num_parameters;
104    int i;
105    struct hash_table *ht;
106 
107    ht = hash_table_ctor(0, hash_table_pointer_hash, hash_table_pointer_compare);
108 
109    num_parameters = 0;
110    foreach_iter(exec_list_iterator, iter_sig, this->callee->parameters)
111       num_parameters++;
112 
113    parameters = new ir_variable *[num_parameters];
114 
115    /* Generate the declarations for the parameters to our inlined code,
116     * and set up the mapping of real function body variables to ours.
117     */
118    i = 0;
119    exec_list_iterator sig_param_iter = this->callee->parameters.iterator();
120    exec_list_iterator param_iter = this->actual_parameters.iterator();
121    for (i = 0; i < num_parameters; i++) {
122       ir_variable *sig_param = (ir_variable *) sig_param_iter.get();
123       ir_rvalue *param = (ir_rvalue *) param_iter.get();
124 
125       /* Generate a new variable for the parameter. */
126       if (sig_param->type->base_type == GLSL_TYPE_SAMPLER) {
127 	 /* For samplers, we want the inlined sampler references
128 	  * referencing the passed in sampler variable, since that
129 	  * will have the location information, which an assignment of
130 	  * a sampler wouldn't.  Fix it up below.
131 	  */
132 	 parameters[i] = NULL;
133       } else {
134 	 parameters[i] = sig_param->clone(ctx, ht);
135 	 parameters[i]->mode = ir_var_auto;
136 
137 	 /* Remove the read-only decoration becuase we're going to write
138 	  * directly to this variable.  If the cloned variable is left
139 	  * read-only and the inlined function is inside a loop, the loop
140 	  * analysis code will get confused.
141 	  */
142 	 parameters[i]->read_only = false;
143 	 next_ir->insert_before(parameters[i]);
144       }
145 
146       /* Move the actual param into our param variable if it's an 'in' type. */
147       if (parameters[i] && (sig_param->mode == ir_var_in ||
148 			    sig_param->mode == ir_var_const_in ||
149 			    sig_param->mode == ir_var_inout)) {
150 	 ir_assignment *assign;
151 
152 	 assign = new(ctx) ir_assignment(new(ctx) ir_dereference_variable(parameters[i]),
153 					 param, NULL);
154 	 next_ir->insert_before(assign);
155       }
156 
157       sig_param_iter.next();
158       param_iter.next();
159    }
160 
161    exec_list new_instructions;
162 
163    /* Generate the inlined body of the function to a new list */
164    foreach_iter(exec_list_iterator, iter, callee->body) {
165       ir_instruction *ir = (ir_instruction *)iter.get();
166       ir_instruction *new_ir = ir->clone(ctx, ht);
167 
168       new_instructions.push_tail(new_ir);
169       visit_tree(new_ir, replace_return_with_assignment, this->return_deref);
170    }
171 
172    /* If any samplers were passed in, replace any deref of the sampler
173     * with a deref of the sampler argument.
174     */
175    param_iter = this->actual_parameters.iterator();
176    sig_param_iter = this->callee->parameters.iterator();
177    for (i = 0; i < num_parameters; i++) {
178       ir_instruction *const param = (ir_instruction *) param_iter.get();
179       ir_variable *sig_param = (ir_variable *) sig_param_iter.get();
180 
181       if (sig_param->type->base_type == GLSL_TYPE_SAMPLER) {
182 	 ir_dereference *deref = param->as_dereference();
183 
184 	 assert(deref);
185 	 do_sampler_replacement(&new_instructions, sig_param, deref);
186       }
187       param_iter.next();
188       sig_param_iter.next();
189    }
190 
191    /* Now push those new instructions in. */
192    next_ir->insert_before(&new_instructions);
193 
194    /* Copy back the value of any 'out' parameters from the function body
195     * variables to our own.
196     */
197    i = 0;
198    param_iter = this->actual_parameters.iterator();
199    sig_param_iter = this->callee->parameters.iterator();
200    for (i = 0; i < num_parameters; i++) {
201       ir_instruction *const param = (ir_instruction *) param_iter.get();
202       const ir_variable *const sig_param = (ir_variable *) sig_param_iter.get();
203 
204       /* Move our param variable into the actual param if it's an 'out' type. */
205       if (parameters[i] && (sig_param->mode == ir_var_out ||
206 			    sig_param->mode == ir_var_inout)) {
207 	 ir_assignment *assign;
208 
209 	 assign = new(ctx) ir_assignment(param->clone(ctx, NULL)->as_rvalue(),
210 					 new(ctx) ir_dereference_variable(parameters[i]),
211 					 NULL);
212 	 next_ir->insert_before(assign);
213       }
214 
215       param_iter.next();
216       sig_param_iter.next();
217    }
218 
219    delete [] parameters;
220 
221    hash_table_dtor(ht);
222 }
223 
224 
225 ir_visitor_status
visit_enter(ir_expression * ir)226 ir_function_inlining_visitor::visit_enter(ir_expression *ir)
227 {
228    (void) ir;
229    return visit_continue_with_parent;
230 }
231 
232 
233 ir_visitor_status
visit_enter(ir_return * ir)234 ir_function_inlining_visitor::visit_enter(ir_return *ir)
235 {
236    (void) ir;
237    return visit_continue_with_parent;
238 }
239 
240 
241 ir_visitor_status
visit_enter(ir_texture * ir)242 ir_function_inlining_visitor::visit_enter(ir_texture *ir)
243 {
244    (void) ir;
245    return visit_continue_with_parent;
246 }
247 
248 
249 ir_visitor_status
visit_enter(ir_swizzle * ir)250 ir_function_inlining_visitor::visit_enter(ir_swizzle *ir)
251 {
252    (void) ir;
253    return visit_continue_with_parent;
254 }
255 
256 
257 ir_visitor_status
visit_enter(ir_call * ir)258 ir_function_inlining_visitor::visit_enter(ir_call *ir)
259 {
260    if (can_inline(ir)) {
261       ir->generate_inline(ir);
262       ir->remove();
263       this->progress = true;
264    }
265 
266    return visit_continue;
267 }
268 
269 
270 /**
271  * Replaces references to the "sampler" variable with a clone of "deref."
272  *
273  * From the spec, samplers can appear in the tree as function
274  * (non-out) parameters and as the result of array indexing and
275  * structure field selection.  In our builtin implementation, they
276  * also appear in the sampler field of an ir_tex instruction.
277  */
278 
279 class ir_sampler_replacement_visitor : public ir_hierarchical_visitor {
280 public:
ir_sampler_replacement_visitor(ir_variable * sampler,ir_dereference * deref)281    ir_sampler_replacement_visitor(ir_variable *sampler, ir_dereference *deref)
282    {
283       this->sampler = sampler;
284       this->deref = deref;
285    }
286 
~ir_sampler_replacement_visitor()287    virtual ~ir_sampler_replacement_visitor()
288    {
289    }
290 
291    virtual ir_visitor_status visit_leave(ir_call *);
292    virtual ir_visitor_status visit_leave(ir_dereference_array *);
293    virtual ir_visitor_status visit_leave(ir_dereference_record *);
294    virtual ir_visitor_status visit_leave(ir_texture *);
295 
296    void replace_deref(ir_dereference **deref);
297    void replace_rvalue(ir_rvalue **rvalue);
298 
299    ir_variable *sampler;
300    ir_dereference *deref;
301 };
302 
303 void
replace_deref(ir_dereference ** deref)304 ir_sampler_replacement_visitor::replace_deref(ir_dereference **deref)
305 {
306    ir_dereference_variable *deref_var = (*deref)->as_dereference_variable();
307    if (deref_var && deref_var->var == this->sampler) {
308       *deref = this->deref->clone(ralloc_parent(*deref), NULL);
309    }
310 }
311 
312 void
replace_rvalue(ir_rvalue ** rvalue)313 ir_sampler_replacement_visitor::replace_rvalue(ir_rvalue **rvalue)
314 {
315    if (!*rvalue)
316       return;
317 
318    ir_dereference *deref = (*rvalue)->as_dereference();
319 
320    if (!deref)
321       return;
322 
323    replace_deref(&deref);
324    *rvalue = deref;
325 }
326 
327 ir_visitor_status
visit_leave(ir_texture * ir)328 ir_sampler_replacement_visitor::visit_leave(ir_texture *ir)
329 {
330    replace_deref(&ir->sampler);
331 
332    return visit_continue;
333 }
334 
335 ir_visitor_status
visit_leave(ir_dereference_array * ir)336 ir_sampler_replacement_visitor::visit_leave(ir_dereference_array *ir)
337 {
338    replace_rvalue(&ir->array);
339    return visit_continue;
340 }
341 
342 ir_visitor_status
visit_leave(ir_dereference_record * ir)343 ir_sampler_replacement_visitor::visit_leave(ir_dereference_record *ir)
344 {
345    replace_rvalue(&ir->record);
346    return visit_continue;
347 }
348 
349 ir_visitor_status
visit_leave(ir_call * ir)350 ir_sampler_replacement_visitor::visit_leave(ir_call *ir)
351 {
352    foreach_iter(exec_list_iterator, iter, *ir) {
353       ir_rvalue *param = (ir_rvalue *)iter.get();
354       ir_rvalue *new_param = param;
355       replace_rvalue(&new_param);
356 
357       if (new_param != param) {
358 	 param->replace_with(new_param);
359       }
360    }
361    return visit_continue;
362 }
363 
364 static void
do_sampler_replacement(exec_list * instructions,ir_variable * sampler,ir_dereference * deref)365 do_sampler_replacement(exec_list *instructions,
366 		       ir_variable *sampler,
367 		       ir_dereference *deref)
368 {
369    ir_sampler_replacement_visitor v(sampler, deref);
370 
371    visit_list_elements(&v, instructions);
372 }
373