Go to most recent revision | Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
5564 | serge | 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 ir_set_program_inouts.cpp |
||
26 | * |
||
27 | * Sets the InputsRead and OutputsWritten of Mesa programs. |
||
28 | * |
||
29 | * Additionally, for fragment shaders, sets the InterpQualifier array, the |
||
30 | * IsCentroid and IsSample bitfields, and the UsesDFdy flag. |
||
31 | * |
||
32 | * Mesa programs (gl_program, not gl_shader_program) have a set of |
||
33 | * flags indicating which varyings are read and written. Computing |
||
34 | * which are actually read from some sort of backend code can be |
||
35 | * tricky when variable array indexing involved. So this pass |
||
36 | * provides support for setting InputsRead and OutputsWritten right |
||
37 | * from the GLSL IR. |
||
38 | */ |
||
39 | |||
40 | #include "main/core.h" /* for struct gl_program */ |
||
41 | #include "ir.h" |
||
42 | #include "ir_visitor.h" |
||
43 | #include "glsl_types.h" |
||
44 | |||
45 | namespace { |
||
46 | |||
47 | class ir_set_program_inouts_visitor : public ir_hierarchical_visitor { |
||
48 | public: |
||
49 | ir_set_program_inouts_visitor(struct gl_program *prog, |
||
50 | gl_shader_stage shader_stage) |
||
51 | { |
||
52 | this->prog = prog; |
||
53 | this->shader_stage = shader_stage; |
||
54 | } |
||
55 | ~ir_set_program_inouts_visitor() |
||
56 | { |
||
57 | } |
||
58 | |||
59 | virtual ir_visitor_status visit_enter(ir_dereference_array *); |
||
60 | virtual ir_visitor_status visit_enter(ir_function_signature *); |
||
61 | virtual ir_visitor_status visit_enter(ir_expression *); |
||
62 | virtual ir_visitor_status visit_enter(ir_discard *); |
||
63 | virtual ir_visitor_status visit_enter(ir_texture *); |
||
64 | virtual ir_visitor_status visit(ir_dereference_variable *); |
||
65 | |||
66 | private: |
||
67 | void mark_whole_variable(ir_variable *var); |
||
68 | bool try_mark_partial_variable(ir_variable *var, ir_rvalue *index); |
||
69 | |||
70 | struct gl_program *prog; |
||
71 | gl_shader_stage shader_stage; |
||
72 | }; |
||
73 | |||
74 | } /* anonymous namespace */ |
||
75 | |||
76 | static inline bool |
||
77 | is_shader_inout(ir_variable *var) |
||
78 | { |
||
79 | return var->data.mode == ir_var_shader_in || |
||
80 | var->data.mode == ir_var_shader_out || |
||
81 | var->data.mode == ir_var_system_value; |
||
82 | } |
||
83 | |||
84 | static inline bool |
||
85 | is_dual_slot(ir_variable *var) |
||
86 | { |
||
87 | const glsl_type *type = var->type->without_array(); |
||
88 | return type == glsl_type::dvec4_type || type == glsl_type::dvec3_type; |
||
89 | } |
||
90 | |||
91 | static void |
||
92 | mark(struct gl_program *prog, ir_variable *var, int offset, int len, |
||
93 | bool is_fragment_shader) |
||
94 | { |
||
95 | /* As of GLSL 1.20, varyings can only be floats, floating-point |
||
96 | * vectors or matrices, or arrays of them. For Mesa programs using |
||
97 | * InputsRead/OutputsWritten, everything but matrices uses one |
||
98 | * slot, while matrices use a slot per column. Presumably |
||
99 | * something doing a more clever packing would use something other |
||
100 | * than InputsRead/OutputsWritten. |
||
101 | */ |
||
102 | |||
103 | for (int i = 0; i < len; i++) { |
||
104 | bool dual_slot = is_dual_slot(var); |
||
105 | int idx = var->data.location + var->data.index + offset + i; |
||
106 | GLbitfield64 bitfield = BITFIELD64_BIT(idx); |
||
107 | |||
108 | if (var->data.mode == ir_var_shader_in) { |
||
109 | prog->InputsRead |= bitfield; |
||
110 | if (dual_slot) |
||
111 | prog->DoubleInputsRead |= bitfield; |
||
112 | if (is_fragment_shader) { |
||
113 | gl_fragment_program *fprog = (gl_fragment_program *) prog; |
||
114 | fprog->InterpQualifier[idx] = |
||
115 | (glsl_interp_qualifier) var->data.interpolation; |
||
116 | if (var->data.centroid) |
||
117 | fprog->IsCentroid |= bitfield; |
||
118 | if (var->data.sample) |
||
119 | fprog->IsSample |= bitfield; |
||
120 | } |
||
121 | } else if (var->data.mode == ir_var_system_value) { |
||
122 | prog->SystemValuesRead |= bitfield; |
||
123 | } else { |
||
124 | assert(var->data.mode == ir_var_shader_out); |
||
125 | prog->OutputsWritten |= bitfield; |
||
126 | } |
||
127 | } |
||
128 | } |
||
129 | |||
130 | /** |
||
131 | * Mark an entire variable as used. Caller must ensure that the variable |
||
132 | * represents a shader input or output. |
||
133 | */ |
||
134 | void |
||
135 | ir_set_program_inouts_visitor::mark_whole_variable(ir_variable *var) |
||
136 | { |
||
137 | const glsl_type *type = var->type; |
||
138 | if (this->shader_stage == MESA_SHADER_GEOMETRY && |
||
139 | var->data.mode == ir_var_shader_in && type->is_array()) { |
||
140 | type = type->fields.array; |
||
141 | } |
||
142 | |||
143 | mark(this->prog, var, 0, type->count_attribute_slots(), |
||
144 | this->shader_stage == MESA_SHADER_FRAGMENT); |
||
145 | } |
||
146 | |||
147 | /* Default handler: Mark all the locations in the variable as used. */ |
||
148 | ir_visitor_status |
||
149 | ir_set_program_inouts_visitor::visit(ir_dereference_variable *ir) |
||
150 | { |
||
151 | if (!is_shader_inout(ir->var)) |
||
152 | return visit_continue; |
||
153 | |||
154 | mark_whole_variable(ir->var); |
||
155 | |||
156 | return visit_continue; |
||
157 | } |
||
158 | |||
159 | /** |
||
160 | * Try to mark a portion of the given variable as used. Caller must ensure |
||
161 | * that the variable represents a shader input or output which can be indexed |
||
162 | * into in array fashion (an array or matrix). For the purpose of geometry |
||
163 | * shader inputs (which are always arrays*), this means that the array element |
||
164 | * must be something that can be indexed into in array fashion. |
||
165 | * |
||
166 | * *Except gl_PrimitiveIDIn, as noted below. |
||
167 | * |
||
168 | * If the index can't be interpreted as a constant, or some other problem |
||
169 | * occurs, then nothing will be marked and false will be returned. |
||
170 | */ |
||
171 | bool |
||
172 | ir_set_program_inouts_visitor::try_mark_partial_variable(ir_variable *var, |
||
173 | ir_rvalue *index) |
||
174 | { |
||
175 | const glsl_type *type = var->type; |
||
176 | |||
177 | if (this->shader_stage == MESA_SHADER_GEOMETRY && |
||
178 | var->data.mode == ir_var_shader_in) { |
||
179 | /* The only geometry shader input that is not an array is |
||
180 | * gl_PrimitiveIDIn, and in that case, this code will never be reached, |
||
181 | * because gl_PrimitiveIDIn can't be indexed into in array fashion. |
||
182 | */ |
||
183 | assert(type->is_array()); |
||
184 | type = type->fields.array; |
||
185 | } |
||
186 | |||
187 | /* The code below only handles: |
||
188 | * |
||
189 | * - Indexing into matrices |
||
190 | * - Indexing into arrays of (matrices, vectors, or scalars) |
||
191 | * |
||
192 | * All other possibilities are either prohibited by GLSL (vertex inputs and |
||
193 | * fragment outputs can't be structs) or should have been eliminated by |
||
194 | * lowering passes (do_vec_index_to_swizzle() gets rid of indexing into |
||
195 | * vectors, and lower_packed_varyings() gets rid of structs that occur in |
||
196 | * varyings). |
||
197 | */ |
||
198 | if (!(type->is_matrix() || |
||
199 | (type->is_array() && |
||
200 | (type->fields.array->is_numeric() || |
||
201 | type->fields.array->is_boolean())))) { |
||
202 | assert(!"Unexpected indexing in ir_set_program_inouts"); |
||
203 | |||
204 | /* For safety in release builds, in case we ever encounter unexpected |
||
205 | * indexing, give up and let the caller mark the whole variable as used. |
||
206 | */ |
||
207 | return false; |
||
208 | } |
||
209 | |||
210 | ir_constant *index_as_constant = index->as_constant(); |
||
211 | if (!index_as_constant) |
||
212 | return false; |
||
213 | |||
214 | unsigned elem_width; |
||
215 | unsigned num_elems; |
||
216 | if (type->is_array()) { |
||
217 | num_elems = type->length; |
||
218 | if (type->fields.array->is_matrix()) |
||
219 | elem_width = type->fields.array->matrix_columns; |
||
220 | else |
||
221 | elem_width = 1; |
||
222 | } else { |
||
223 | num_elems = type->matrix_columns; |
||
224 | elem_width = 1; |
||
225 | } |
||
226 | |||
227 | if (index_as_constant->value.u[0] >= num_elems) { |
||
228 | /* Constant index outside the bounds of the matrix/array. This could |
||
229 | * arise as a result of constant folding of a legal GLSL program. |
||
230 | * |
||
231 | * Even though the spec says that indexing outside the bounds of a |
||
232 | * matrix/array results in undefined behaviour, we don't want to pass |
||
233 | * out-of-range values to mark() (since this could result in slots that |
||
234 | * don't exist being marked as used), so just let the caller mark the |
||
235 | * whole variable as used. |
||
236 | */ |
||
237 | return false; |
||
238 | } |
||
239 | |||
240 | mark(this->prog, var, index_as_constant->value.u[0] * elem_width, |
||
241 | elem_width, this->shader_stage == MESA_SHADER_FRAGMENT); |
||
242 | return true; |
||
243 | } |
||
244 | |||
245 | ir_visitor_status |
||
246 | ir_set_program_inouts_visitor::visit_enter(ir_dereference_array *ir) |
||
247 | { |
||
248 | /* Note: for geometry shader inputs, lower_named_interface_blocks may |
||
249 | * create 2D arrays, so we need to be able to handle those. 2D arrays |
||
250 | * shouldn't be able to crop up for any other reason. |
||
251 | */ |
||
252 | if (ir_dereference_array * const inner_array = |
||
253 | ir->array->as_dereference_array()) { |
||
254 | /* ir => foo[i][j] |
||
255 | * inner_array => foo[i] |
||
256 | */ |
||
257 | if (ir_dereference_variable * const deref_var = |
||
258 | inner_array->array->as_dereference_variable()) { |
||
259 | if (this->shader_stage == MESA_SHADER_GEOMETRY && |
||
260 | deref_var->var->data.mode == ir_var_shader_in) { |
||
261 | /* foo is a geometry shader input, so i is the vertex, and j the |
||
262 | * part of the input we're accessing. |
||
263 | */ |
||
264 | if (try_mark_partial_variable(deref_var->var, ir->array_index)) |
||
265 | { |
||
266 | /* We've now taken care of foo and j, but i might contain a |
||
267 | * subexpression that accesses shader inputs. So manually |
||
268 | * visit i and then continue with the parent. |
||
269 | */ |
||
270 | inner_array->array_index->accept(this); |
||
271 | return visit_continue_with_parent; |
||
272 | } |
||
273 | } |
||
274 | } |
||
275 | } else if (ir_dereference_variable * const deref_var = |
||
276 | ir->array->as_dereference_variable()) { |
||
277 | /* ir => foo[i], where foo is a variable. */ |
||
278 | if (this->shader_stage == MESA_SHADER_GEOMETRY && |
||
279 | deref_var->var->data.mode == ir_var_shader_in) { |
||
280 | /* foo is a geometry shader input, so i is the vertex, and we're |
||
281 | * accessing the entire input. |
||
282 | */ |
||
283 | mark_whole_variable(deref_var->var); |
||
284 | /* We've now taken care of foo, but i might contain a subexpression |
||
285 | * that accesses shader inputs. So manually visit i and then |
||
286 | * continue with the parent. |
||
287 | */ |
||
288 | ir->array_index->accept(this); |
||
289 | return visit_continue_with_parent; |
||
290 | } else if (is_shader_inout(deref_var->var)) { |
||
291 | /* foo is a shader input/output, but not a geometry shader input, |
||
292 | * so i is the part of the input we're accessing. |
||
293 | */ |
||
294 | if (try_mark_partial_variable(deref_var->var, ir->array_index)) |
||
295 | return visit_continue_with_parent; |
||
296 | } |
||
297 | } |
||
298 | |||
299 | /* The expression is something we don't recognize. Just visit its |
||
300 | * subexpressions. |
||
301 | */ |
||
302 | return visit_continue; |
||
303 | } |
||
304 | |||
305 | ir_visitor_status |
||
306 | ir_set_program_inouts_visitor::visit_enter(ir_function_signature *ir) |
||
307 | { |
||
308 | /* We don't want to descend into the function parameters and |
||
309 | * consider them as shader inputs or outputs. |
||
310 | */ |
||
311 | visit_list_elements(this, &ir->body); |
||
312 | return visit_continue_with_parent; |
||
313 | } |
||
314 | |||
315 | ir_visitor_status |
||
316 | ir_set_program_inouts_visitor::visit_enter(ir_expression *ir) |
||
317 | { |
||
318 | if (this->shader_stage == MESA_SHADER_FRAGMENT && |
||
319 | (ir->operation == ir_unop_dFdy || |
||
320 | ir->operation == ir_unop_dFdy_coarse || |
||
321 | ir->operation == ir_unop_dFdy_fine)) { |
||
322 | gl_fragment_program *fprog = (gl_fragment_program *) prog; |
||
323 | fprog->UsesDFdy = true; |
||
324 | } |
||
325 | return visit_continue; |
||
326 | } |
||
327 | |||
328 | ir_visitor_status |
||
329 | ir_set_program_inouts_visitor::visit_enter(ir_discard *) |
||
330 | { |
||
331 | /* discards are only allowed in fragment shaders. */ |
||
332 | assert(this->shader_stage == MESA_SHADER_FRAGMENT); |
||
333 | |||
334 | gl_fragment_program *fprog = (gl_fragment_program *) prog; |
||
335 | fprog->UsesKill = true; |
||
336 | |||
337 | return visit_continue; |
||
338 | } |
||
339 | |||
340 | ir_visitor_status |
||
341 | ir_set_program_inouts_visitor::visit_enter(ir_texture *ir) |
||
342 | { |
||
343 | if (ir->op == ir_tg4) |
||
344 | prog->UsesGather = true; |
||
345 | return visit_continue; |
||
346 | } |
||
347 | |||
348 | void |
||
349 | do_set_program_inouts(exec_list *instructions, struct gl_program *prog, |
||
350 | gl_shader_stage shader_stage) |
||
351 | { |
||
352 | ir_set_program_inouts_visitor v(prog, shader_stage); |
||
353 | |||
354 | prog->InputsRead = 0; |
||
355 | prog->OutputsWritten = 0; |
||
356 | prog->SystemValuesRead = 0; |
||
357 | if (shader_stage == MESA_SHADER_FRAGMENT) { |
||
358 | gl_fragment_program *fprog = (gl_fragment_program *) prog; |
||
359 | memset(fprog->InterpQualifier, 0, sizeof(fprog->InterpQualifier)); |
||
360 | fprog->IsCentroid = 0; |
||
361 | fprog->IsSample = 0; |
||
362 | fprog->UsesDFdy = false; |
||
363 | fprog->UsesKill = false; |
||
364 | } |
||
365 | visit_list_elements(&v, instructions); |
||
366 | }> |