projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
ir_to_mesa: Remove unused member array_indexed from struct statevar_element.
[mesa.git] / src / mesa / program / ir_to_mesa.cpp
1 /*
2 * Copyright (C) 2005-2007 Brian Paul All Rights Reserved.
3 * Copyright (C) 2008 VMware, Inc. All Rights Reserved.
4 * Copyright © 2010 Intel Corporation
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
15 * Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
22 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
23 * DEALINGS IN THE SOFTWARE.
24 */
25
26 /**
27 * \file ir_to_mesa.cpp
28 *
29 * Translates the IR to ARB_fragment_program text if possible,
30 * printing the result
31 */
32
33 #include <stdio.h>
34 #include "main/compiler.h"
35 #include "ir.h"
36 #include "ir_visitor.h"
37 #include "ir_print_visitor.h"
38 #include "ir_expression_flattening.h"
39 #include "glsl_types.h"
40 #include "glsl_parser_extras.h"
41 #include "../glsl/program.h"
42 #include "ir_optimization.h"
43 #include "ast.h"
44
45 extern "C" {
46 #include "main/mtypes.h"
47 #include "main/shaderapi.h"
48 #include "main/shaderobj.h"
49 #include "main/uniforms.h"
50 #include "program/hash_table.h"
51 #include "program/prog_instruction.h"
52 #include "program/prog_optimize.h"
53 #include "program/prog_print.h"
54 #include "program/program.h"
55 #include "program/prog_uniform.h"
56 #include "program/prog_parameter.h"
57 }
58
59 static int swizzle_for_size(int size);
60
61 /**
62 * This struct is a corresponding struct to Mesa prog_src_register, with
63 * wider fields.
64 */
65 typedef struct ir_to_mesa_src_reg {
66 ir_to_mesa_src_reg(int file, int index, const glsl_type *type)
67 {
68 this->file = file;
69 this->index = index;
70 if (type && (type->is_scalar() || type->is_vector() || type->is_matrix()))
71 this->swizzle = swizzle_for_size(type->vector_elements);
72 else
73 this->swizzle = SWIZZLE_XYZW;
74 this->negate = 0;
75 this->reladdr = NULL;
76 }
77
78 ir_to_mesa_src_reg()
79 {
80 this->file = PROGRAM_UNDEFINED;
81 this->index = 0;
82 this->swizzle = 0;
83 this->negate = 0;
84 this->reladdr = NULL;
85 }
86
87 int file; /**< PROGRAM_* from Mesa */
88 int index; /**< temporary index, VERT_ATTRIB_*, FRAG_ATTRIB_*, etc. */
89 GLuint swizzle; /**< SWIZZLE_XYZWONEZERO swizzles from Mesa. */
90 int negate; /**< NEGATE_XYZW mask from mesa */
91 /** Register index should be offset by the integer in this reg. */
92 ir_to_mesa_src_reg *reladdr;
93 } ir_to_mesa_src_reg;
94
95 typedef struct ir_to_mesa_dst_reg {
96 int file; /**< PROGRAM_* from Mesa */
97 int index; /**< temporary index, VERT_ATTRIB_*, FRAG_ATTRIB_*, etc. */
98 int writemask; /**< Bitfield of WRITEMASK_[XYZW] */
99 GLuint cond_mask:4;
100 /** Register index should be offset by the integer in this reg. */
101 ir_to_mesa_src_reg *reladdr;
102 } ir_to_mesa_dst_reg;
103
104 extern ir_to_mesa_src_reg ir_to_mesa_undef;
105
106 class ir_to_mesa_instruction : public exec_node {
107 public:
108 /* Callers of this talloc-based new need not call delete. It's
109 * easier to just talloc_free 'ctx' (or any of its ancestors). */
110 static void* operator new(size_t size, void *ctx)
111 {
112 void *node;
113
114 node = talloc_zero_size(ctx, size);
115 assert(node != NULL);
116
117 return node;
118 }
119
120 enum prog_opcode op;
121 ir_to_mesa_dst_reg dst_reg;
122 ir_to_mesa_src_reg src_reg[3];
123 /** Pointer to the ir source this tree came from for debugging */
124 ir_instruction *ir;
125 GLboolean cond_update;
126 int sampler; /**< sampler index */
127 int tex_target; /**< One of TEXTURE_*_INDEX */
128 GLboolean tex_shadow;
129
130 class function_entry *function; /* Set on OPCODE_CAL or OPCODE_BGNSUB */
131 };
132
133 class variable_storage : public exec_node {
134 public:
135 variable_storage(ir_variable *var, int file, int index)
136 : file(file), index(index), var(var)
137 {
138 /* empty */
139 }
140
141 int file;
142 int index;
143 ir_variable *var; /* variable that maps to this, if any */
144 };
145
146 class function_entry : public exec_node {
147 public:
148 ir_function_signature *sig;
149
150 /**
151 * identifier of this function signature used by the program.
152 *
153 * At the point that Mesa instructions for function calls are
154 * generated, we don't know the address of the first instruction of
155 * the function body. So we make the BranchTarget that is called a
156 * small integer and rewrite them during set_branchtargets().
157 */
158 int sig_id;
159
160 /**
161 * Pointer to first instruction of the function body.
162 *
163 * Set during function body emits after main() is processed.
164 */
165 ir_to_mesa_instruction *bgn_inst;
166
167 /**
168 * Index of the first instruction of the function body in actual
169 * Mesa IR.
170 *
171 * Set after convertion from ir_to_mesa_instruction to prog_instruction.
172 */
173 int inst;
174
175 /** Storage for the return value. */
176 ir_to_mesa_src_reg return_reg;
177 };
178
179 class ir_to_mesa_visitor : public ir_visitor {
180 public:
181 ir_to_mesa_visitor();
182 ~ir_to_mesa_visitor();
183
184 function_entry *current_function;
185
186 GLcontext *ctx;
187 struct gl_program *prog;
188 struct gl_shader_program *shader_program;
189 struct gl_shader_compiler_options *options;
190
191 int next_temp;
192
193 variable_storage *find_variable_storage(ir_variable *var);
194
195 function_entry *get_function_signature(ir_function_signature *sig);
196
197 ir_to_mesa_src_reg get_temp(const glsl_type *type);
198 void reladdr_to_temp(ir_instruction *ir,
199 ir_to_mesa_src_reg *reg, int *num_reladdr);
200
201 struct ir_to_mesa_src_reg src_reg_for_float(float val);
202
203 /**
204 * \name Visit methods
205 *
206 * As typical for the visitor pattern, there must be one \c visit method for
207 * each concrete subclass of \c ir_instruction. Virtual base classes within
208 * the hierarchy should not have \c visit methods.
209 */
210 /*@{*/
211 virtual void visit(ir_variable *);
212 virtual void visit(ir_loop *);
213 virtual void visit(ir_loop_jump *);
214 virtual void visit(ir_function_signature *);
215 virtual void visit(ir_function *);
216 virtual void visit(ir_expression *);
217 virtual void visit(ir_swizzle *);
218 virtual void visit(ir_dereference_variable *);
219 virtual void visit(ir_dereference_array *);
220 virtual void visit(ir_dereference_record *);
221 virtual void visit(ir_assignment *);
222 virtual void visit(ir_constant *);
223 virtual void visit(ir_call *);
224 virtual void visit(ir_return *);
225 virtual void visit(ir_discard *);
226 virtual void visit(ir_texture *);
227 virtual void visit(ir_if *);
228 /*@}*/
229
230 struct ir_to_mesa_src_reg result;
231
232 /** List of variable_storage */
233 exec_list variables;
234
235 /** List of function_entry */
236 exec_list function_signatures;
237 int next_signature_id;
238
239 /** List of ir_to_mesa_instruction */
240 exec_list instructions;
241
242 ir_to_mesa_instruction *ir_to_mesa_emit_op0(ir_instruction *ir,
243 enum prog_opcode op);
244
245 ir_to_mesa_instruction *ir_to_mesa_emit_op1(ir_instruction *ir,
246 enum prog_opcode op,
247 ir_to_mesa_dst_reg dst,
248 ir_to_mesa_src_reg src0);
249
250 ir_to_mesa_instruction *ir_to_mesa_emit_op2(ir_instruction *ir,
251 enum prog_opcode op,
252 ir_to_mesa_dst_reg dst,
253 ir_to_mesa_src_reg src0,
254 ir_to_mesa_src_reg src1);
255
256 ir_to_mesa_instruction *ir_to_mesa_emit_op3(ir_instruction *ir,
257 enum prog_opcode op,
258 ir_to_mesa_dst_reg dst,
259 ir_to_mesa_src_reg src0,
260 ir_to_mesa_src_reg src1,
261 ir_to_mesa_src_reg src2);
262
263 void ir_to_mesa_emit_scalar_op1(ir_instruction *ir,
264 enum prog_opcode op,
265 ir_to_mesa_dst_reg dst,
266 ir_to_mesa_src_reg src0);
267
268 void ir_to_mesa_emit_scalar_op2(ir_instruction *ir,
269 enum prog_opcode op,
270 ir_to_mesa_dst_reg dst,
271 ir_to_mesa_src_reg src0,
272 ir_to_mesa_src_reg src1);
273
274 GLboolean try_emit_mad(ir_expression *ir,
275 int mul_operand);
276
277 int get_sampler_uniform_value(ir_dereference *deref);
278
279 void *mem_ctx;
280 };
281
282 ir_to_mesa_src_reg ir_to_mesa_undef = ir_to_mesa_src_reg(PROGRAM_UNDEFINED, 0, NULL);
283
284 ir_to_mesa_dst_reg ir_to_mesa_undef_dst = {
285 PROGRAM_UNDEFINED, 0, SWIZZLE_NOOP, COND_TR, NULL,
286 };
287
288 ir_to_mesa_dst_reg ir_to_mesa_address_reg = {
289 PROGRAM_ADDRESS, 0, WRITEMASK_X, COND_TR, NULL
290 };
291
292 static void fail_link(struct gl_shader_program *prog, const char *fmt, ...) PRINTFLIKE(2, 3);
293
294 static void fail_link(struct gl_shader_program *prog, const char *fmt, ...)
295 {
296 va_list args;
297 va_start(args, fmt);
298 prog->InfoLog = talloc_vasprintf_append(prog->InfoLog, fmt, args);
299 va_end(args);
300
301 prog->LinkStatus = GL_FALSE;
302 }
303
304 static int swizzle_for_size(int size)
305 {
306 int size_swizzles[4] = {
307 MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
308 MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y),
309 MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_Z),
310 MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W),
311 };
312
313 return size_swizzles[size - 1];
314 }
315
316 ir_to_mesa_instruction *
317 ir_to_mesa_visitor::ir_to_mesa_emit_op3(ir_instruction *ir,
318 enum prog_opcode op,
319 ir_to_mesa_dst_reg dst,
320 ir_to_mesa_src_reg src0,
321 ir_to_mesa_src_reg src1,
322 ir_to_mesa_src_reg src2)
323 {
324 ir_to_mesa_instruction *inst = new(mem_ctx) ir_to_mesa_instruction();
325 int num_reladdr = 0;
326
327 /* If we have to do relative addressing, we want to load the ARL
328 * reg directly for one of the regs, and preload the other reladdr
329 * sources into temps.
330 */
331 num_reladdr += dst.reladdr != NULL;
332 num_reladdr += src0.reladdr != NULL;
333 num_reladdr += src1.reladdr != NULL;
334 num_reladdr += src2.reladdr != NULL;
335
336 reladdr_to_temp(ir, &src2, &num_reladdr);
337 reladdr_to_temp(ir, &src1, &num_reladdr);
338 reladdr_to_temp(ir, &src0, &num_reladdr);
339
340 if (dst.reladdr) {
341 ir_to_mesa_emit_op1(ir, OPCODE_ARL, ir_to_mesa_address_reg,
342 *dst.reladdr);
343
344 num_reladdr--;
345 }
346 assert(num_reladdr == 0);
347
348 inst->op = op;
349 inst->dst_reg = dst;
350 inst->src_reg[0] = src0;
351 inst->src_reg[1] = src1;
352 inst->src_reg[2] = src2;
353 inst->ir = ir;
354
355 inst->function = NULL;
356
357 this->instructions.push_tail(inst);
358
359 return inst;
360 }
361
362
363 ir_to_mesa_instruction *
364 ir_to_mesa_visitor::ir_to_mesa_emit_op2(ir_instruction *ir,
365 enum prog_opcode op,
366 ir_to_mesa_dst_reg dst,
367 ir_to_mesa_src_reg src0,
368 ir_to_mesa_src_reg src1)
369 {
370 return ir_to_mesa_emit_op3(ir, op, dst, src0, src1, ir_to_mesa_undef);
371 }
372
373 ir_to_mesa_instruction *
374 ir_to_mesa_visitor::ir_to_mesa_emit_op1(ir_instruction *ir,
375 enum prog_opcode op,
376 ir_to_mesa_dst_reg dst,
377 ir_to_mesa_src_reg src0)
378 {
379 assert(dst.writemask != 0);
380 return ir_to_mesa_emit_op3(ir, op, dst,
381 src0, ir_to_mesa_undef, ir_to_mesa_undef);
382 }
383
384 ir_to_mesa_instruction *
385 ir_to_mesa_visitor::ir_to_mesa_emit_op0(ir_instruction *ir,
386 enum prog_opcode op)
387 {
388 return ir_to_mesa_emit_op3(ir, op, ir_to_mesa_undef_dst,
389 ir_to_mesa_undef,
390 ir_to_mesa_undef,
391 ir_to_mesa_undef);
392 }
393
394 inline ir_to_mesa_dst_reg
395 ir_to_mesa_dst_reg_from_src(ir_to_mesa_src_reg reg)
396 {
397 ir_to_mesa_dst_reg dst_reg;
398
399 dst_reg.file = reg.file;
400 dst_reg.index = reg.index;
401 dst_reg.writemask = WRITEMASK_XYZW;
402 dst_reg.cond_mask = COND_TR;
403 dst_reg.reladdr = reg.reladdr;
404
405 return dst_reg;
406 }
407
408 inline ir_to_mesa_src_reg
409 ir_to_mesa_src_reg_from_dst(ir_to_mesa_dst_reg reg)
410 {
411 return ir_to_mesa_src_reg(reg.file, reg.index, NULL);
412 }
413
414 /**
415 * Emits Mesa scalar opcodes to produce unique answers across channels.
416 *
417 * Some Mesa opcodes are scalar-only, like ARB_fp/vp. The src X
418 * channel determines the result across all channels. So to do a vec4
419 * of this operation, we want to emit a scalar per source channel used
420 * to produce dest channels.
421 */
422 void
423 ir_to_mesa_visitor::ir_to_mesa_emit_scalar_op2(ir_instruction *ir,
424 enum prog_opcode op,
425 ir_to_mesa_dst_reg dst,
426 ir_to_mesa_src_reg orig_src0,
427 ir_to_mesa_src_reg orig_src1)
428 {
429 int i, j;
430 int done_mask = ~dst.writemask;
431
432 /* Mesa RCP is a scalar operation splatting results to all channels,
433 * like ARB_fp/vp. So emit as many RCPs as necessary to cover our
434 * dst channels.
435 */
436 for (i = 0; i < 4; i++) {
437 GLuint this_mask = (1 << i);
438 ir_to_mesa_instruction *inst;
439 ir_to_mesa_src_reg src0 = orig_src0;
440 ir_to_mesa_src_reg src1 = orig_src1;
441
442 if (done_mask & this_mask)
443 continue;
444
445 GLuint src0_swiz = GET_SWZ(src0.swizzle, i);
446 GLuint src1_swiz = GET_SWZ(src1.swizzle, i);
447 for (j = i + 1; j < 4; j++) {
448 if (!(done_mask & (1 << j)) &&
449 GET_SWZ(src0.swizzle, j) == src0_swiz &&
450 GET_SWZ(src1.swizzle, j) == src1_swiz) {
451 this_mask |= (1 << j);
452 }
453 }
454 src0.swizzle = MAKE_SWIZZLE4(src0_swiz, src0_swiz,
455 src0_swiz, src0_swiz);
456 src1.swizzle = MAKE_SWIZZLE4(src1_swiz, src1_swiz,
457 src1_swiz, src1_swiz);
458
459 inst = ir_to_mesa_emit_op2(ir, op,
460 dst,
461 src0,
462 src1);
463 inst->dst_reg.writemask = this_mask;
464 done_mask |= this_mask;
465 }
466 }
467
468 void
469 ir_to_mesa_visitor::ir_to_mesa_emit_scalar_op1(ir_instruction *ir,
470 enum prog_opcode op,
471 ir_to_mesa_dst_reg dst,
472 ir_to_mesa_src_reg src0)
473 {
474 ir_to_mesa_src_reg undef = ir_to_mesa_undef;
475
476 undef.swizzle = SWIZZLE_XXXX;
477
478 ir_to_mesa_emit_scalar_op2(ir, op, dst, src0, undef);
479 }
480
481 struct ir_to_mesa_src_reg
482 ir_to_mesa_visitor::src_reg_for_float(float val)
483 {
484 ir_to_mesa_src_reg src_reg(PROGRAM_CONSTANT, -1, NULL);
485
486 src_reg.index = _mesa_add_unnamed_constant(this->prog->Parameters,
487 &val, 1, &src_reg.swizzle);
488
489 return src_reg;
490 }
491
492 static int
493 type_size(const struct glsl_type *type)
494 {
495 unsigned int i;
496 int size;
497
498 switch (type->base_type) {
499 case GLSL_TYPE_UINT:
500 case GLSL_TYPE_INT:
501 case GLSL_TYPE_FLOAT:
502 case GLSL_TYPE_BOOL:
503 if (type->is_matrix()) {
504 return type->matrix_columns;
505 } else {
506 /* Regardless of size of vector, it gets a vec4. This is bad
507 * packing for things like floats, but otherwise arrays become a
508 * mess. Hopefully a later pass over the code can pack scalars
509 * down if appropriate.
510 */
511 return 1;
512 }
513 case GLSL_TYPE_ARRAY:
514 return type_size(type->fields.array) * type->length;
515 case GLSL_TYPE_STRUCT:
516 size = 0;
517 for (i = 0; i < type->length; i++) {
518 size += type_size(type->fields.structure[i].type);
519 }
520 return size;
521 case GLSL_TYPE_SAMPLER:
522 /* Samplers take up one slot in UNIFORMS[], but they're baked in
523 * at link time.
524 */
525 return 1;
526 default:
527 assert(0);
528 return 0;
529 }
530 }
531
532 /**
533 * In the initial pass of codegen, we assign temporary numbers to
534 * intermediate results. (not SSA -- variable assignments will reuse
535 * storage). Actual register allocation for the Mesa VM occurs in a
536 * pass over the Mesa IR later.
537 */
538 ir_to_mesa_src_reg
539 ir_to_mesa_visitor::get_temp(const glsl_type *type)
540 {
541 ir_to_mesa_src_reg src_reg;
542 int swizzle[4];
543 int i;
544
545 src_reg.file = PROGRAM_TEMPORARY;
546 src_reg.index = next_temp;
547 src_reg.reladdr = NULL;
548 next_temp += type_size(type);
549
550 if (type->is_array() || type->is_record()) {
551 src_reg.swizzle = SWIZZLE_NOOP;
552 } else {
553 for (i = 0; i < type->vector_elements; i++)
554 swizzle[i] = i;
555 for (; i < 4; i++)
556 swizzle[i] = type->vector_elements - 1;
557 src_reg.swizzle = MAKE_SWIZZLE4(swizzle[0], swizzle[1],
558 swizzle[2], swizzle[3]);
559 }
560 src_reg.negate = 0;
561
562 return src_reg;
563 }
564
565 variable_storage *
566 ir_to_mesa_visitor::find_variable_storage(ir_variable *var)
567 {
568
569 variable_storage *entry;
570
571 foreach_iter(exec_list_iterator, iter, this->variables) {
572 entry = (variable_storage *)iter.get();
573
574 if (entry->var == var)
575 return entry;
576 }
577
578 return NULL;
579 }
580
581 struct statevar_element {
582 const char *field;
583 int tokens[STATE_LENGTH];
584 int swizzle;
585 };
586
587 static struct statevar_element gl_DepthRange_elements[] = {
588 {"near", {STATE_DEPTH_RANGE, 0, 0}, SWIZZLE_XXXX},
589 {"far", {STATE_DEPTH_RANGE, 0, 0}, SWIZZLE_YYYY},
590 {"diff", {STATE_DEPTH_RANGE, 0, 0}, SWIZZLE_ZZZZ},
591 };
592
593 static struct statevar_element gl_ClipPlane_elements[] = {
594 {NULL, {STATE_CLIPPLANE, 0, 0}, SWIZZLE_XYZW}
595 };
596
597 static struct statevar_element gl_Point_elements[] = {
598 {"size", {STATE_POINT_SIZE}, SWIZZLE_XXXX},
599 {"sizeMin", {STATE_POINT_SIZE}, SWIZZLE_YYYY},
600 {"sizeMax", {STATE_POINT_SIZE}, SWIZZLE_ZZZZ},
601 {"fadeThresholdSize", {STATE_POINT_SIZE}, SWIZZLE_WWWW},
602 {"distanceConstantAttenuation", {STATE_POINT_ATTENUATION}, SWIZZLE_XXXX},
603 {"distanceLinearAttenuation", {STATE_POINT_ATTENUATION}, SWIZZLE_YYYY},
604 {"distanceQuadraticAttenuation", {STATE_POINT_ATTENUATION}, SWIZZLE_ZZZZ},
605 };
606
607 static struct statevar_element gl_FrontMaterial_elements[] = {
608 {"emission", {STATE_MATERIAL, 0, STATE_EMISSION}, SWIZZLE_XYZW},
609 {"ambient", {STATE_MATERIAL, 0, STATE_AMBIENT}, SWIZZLE_XYZW},
610 {"diffuse", {STATE_MATERIAL, 0, STATE_DIFFUSE}, SWIZZLE_XYZW},
611 {"specular", {STATE_MATERIAL, 0, STATE_SPECULAR}, SWIZZLE_XYZW},
612 {"shininess", {STATE_MATERIAL, 0, STATE_SHININESS}, SWIZZLE_XXXX},
613 };
614
615 static struct statevar_element gl_BackMaterial_elements[] = {
616 {"emission", {STATE_MATERIAL, 1, STATE_EMISSION}, SWIZZLE_XYZW},
617 {"ambient", {STATE_MATERIAL, 1, STATE_AMBIENT}, SWIZZLE_XYZW},
618 {"diffuse", {STATE_MATERIAL, 1, STATE_DIFFUSE}, SWIZZLE_XYZW},
619 {"specular", {STATE_MATERIAL, 1, STATE_SPECULAR}, SWIZZLE_XYZW},
620 {"shininess", {STATE_MATERIAL, 1, STATE_SHININESS}, SWIZZLE_XXXX},
621 };
622
623 static struct statevar_element gl_LightSource_elements[] = {
624 {"ambient", {STATE_LIGHT, 0, STATE_AMBIENT}, SWIZZLE_XYZW},
625 {"diffuse", {STATE_LIGHT, 0, STATE_DIFFUSE}, SWIZZLE_XYZW},
626 {"specular", {STATE_LIGHT, 0, STATE_SPECULAR}, SWIZZLE_XYZW},
627 {"position", {STATE_LIGHT, 0, STATE_POSITION}, SWIZZLE_XYZW},
628 {"halfVector", {STATE_LIGHT, 0, STATE_HALF_VECTOR}, SWIZZLE_XYZW},
629 {"spotDirection", {STATE_LIGHT, 0, STATE_SPOT_DIRECTION}, SWIZZLE_XYZW},
630 {"spotCosCutoff", {STATE_LIGHT, 0, STATE_SPOT_DIRECTION}, SWIZZLE_WWWW},
631 {"spotCutoff", {STATE_LIGHT, 0, STATE_SPOT_CUTOFF}, SWIZZLE_XXXX},
632 {"spotExponent", {STATE_LIGHT, 0, STATE_ATTENUATION}, SWIZZLE_WWWW},
633 {"constantAttenuation", {STATE_LIGHT, 0, STATE_ATTENUATION}, SWIZZLE_XXXX},
634 {"linearAttenuation", {STATE_LIGHT, 0, STATE_ATTENUATION}, SWIZZLE_YYYY},
635 {"quadraticAttenuation", {STATE_LIGHT, 0, STATE_ATTENUATION}, SWIZZLE_ZZZZ},
636 };
637
638 static struct statevar_element gl_LightModel_elements[] = {
639 {"ambient", {STATE_LIGHTMODEL_AMBIENT, 0}, SWIZZLE_XYZW},
640 };
641
642 static struct statevar_element gl_FrontLightModelProduct_elements[] = {
643 {"sceneColor", {STATE_LIGHTMODEL_SCENECOLOR, 0}, SWIZZLE_XYZW},
644 };
645
646 static struct statevar_element gl_BackLightModelProduct_elements[] = {
647 {"sceneColor", {STATE_LIGHTMODEL_SCENECOLOR, 1}, SWIZZLE_XYZW},
648 };
649
650 static struct statevar_element gl_FrontLightProduct_elements[] = {
651 {"ambient", {STATE_LIGHTPROD, 0, 0, STATE_AMBIENT}, SWIZZLE_XYZW},
652 {"diffuse", {STATE_LIGHTPROD, 0, 0, STATE_DIFFUSE}, SWIZZLE_XYZW},
653 {"specular", {STATE_LIGHTPROD, 0, 0, STATE_SPECULAR}, SWIZZLE_XYZW},
654 };
655
656 static struct statevar_element gl_BackLightProduct_elements[] = {
657 {"ambient", {STATE_LIGHTPROD, 0, 1, STATE_AMBIENT}, SWIZZLE_XYZW},
658 {"diffuse", {STATE_LIGHTPROD, 0, 1, STATE_DIFFUSE}, SWIZZLE_XYZW},
659 {"specular", {STATE_LIGHTPROD, 0, 1, STATE_SPECULAR}, SWIZZLE_XYZW},
660 };
661
662 static struct statevar_element gl_TextureEnvColor_elements[] = {
663 {NULL, {STATE_TEXENV_COLOR, 0}, SWIZZLE_XYZW},
664 };
665
666 static struct statevar_element gl_EyePlaneS_elements[] = {
667 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_EYE_S}, SWIZZLE_XYZW},
668 };
669
670 static struct statevar_element gl_EyePlaneT_elements[] = {
671 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_EYE_T}, SWIZZLE_XYZW},
672 };
673
674 static struct statevar_element gl_EyePlaneR_elements[] = {
675 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_EYE_R}, SWIZZLE_XYZW},
676 };
677
678 static struct statevar_element gl_EyePlaneQ_elements[] = {
679 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_EYE_Q}, SWIZZLE_XYZW},
680 };
681
682 static struct statevar_element gl_ObjectPlaneS_elements[] = {
683 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_OBJECT_S}, SWIZZLE_XYZW},
684 };
685
686 static struct statevar_element gl_ObjectPlaneT_elements[] = {
687 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_OBJECT_T}, SWIZZLE_XYZW},
688 };
689
690 static struct statevar_element gl_ObjectPlaneR_elements[] = {
691 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_OBJECT_R}, SWIZZLE_XYZW},
692 };
693
694 static struct statevar_element gl_ObjectPlaneQ_elements[] = {
695 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_OBJECT_Q}, SWIZZLE_XYZW},
696 };
697
698 static struct statevar_element gl_Fog_elements[] = {
699 {"color", {STATE_FOG_COLOR}, SWIZZLE_XYZW},
700 {"density", {STATE_FOG_PARAMS}, SWIZZLE_XXXX},
701 {"start", {STATE_FOG_PARAMS}, SWIZZLE_YYYY},
702 {"end", {STATE_FOG_PARAMS}, SWIZZLE_ZZZZ},
703 {"scale", {STATE_FOG_PARAMS}, SWIZZLE_WWWW},
704 };
705
706 static struct statevar_element gl_NormalScale_elements[] = {
707 {NULL, {STATE_NORMAL_SCALE}, SWIZZLE_XXXX},
708 };
709
710 #define MATRIX(name, statevar, modifier) \
711 static struct statevar_element name ## _elements[] = { \
712 { NULL, { statevar, 0, 0, 0, modifier}, SWIZZLE_XYZW }, \
713 { NULL, { statevar, 0, 1, 1, modifier}, SWIZZLE_XYZW }, \
714 { NULL, { statevar, 0, 2, 2, modifier}, SWIZZLE_XYZW }, \
715 { NULL, { statevar, 0, 3, 3, modifier}, SWIZZLE_XYZW }, \
716 }
717
718 MATRIX(gl_ModelViewMatrix,
719 STATE_MODELVIEW_MATRIX, STATE_MATRIX_TRANSPOSE);
720 MATRIX(gl_ModelViewMatrixInverse,
721 STATE_MODELVIEW_MATRIX, STATE_MATRIX_INVTRANS);
722 MATRIX(gl_ModelViewMatrixTranspose,
723 STATE_MODELVIEW_MATRIX, 0);
724 MATRIX(gl_ModelViewMatrixInverseTranspose,
725 STATE_MODELVIEW_MATRIX, STATE_MATRIX_INVERSE);
726
727 MATRIX(gl_ProjectionMatrix,
728 STATE_PROJECTION_MATRIX, STATE_MATRIX_TRANSPOSE);
729 MATRIX(gl_ProjectionMatrixInverse,
730 STATE_PROJECTION_MATRIX, STATE_MATRIX_INVTRANS);
731 MATRIX(gl_ProjectionMatrixTranspose,
732 STATE_PROJECTION_MATRIX, 0);
733 MATRIX(gl_ProjectionMatrixInverseTranspose,
734 STATE_PROJECTION_MATRIX, STATE_MATRIX_INVERSE);
735
736 MATRIX(gl_ModelViewProjectionMatrix,
737 STATE_MVP_MATRIX, STATE_MATRIX_TRANSPOSE);
738 MATRIX(gl_ModelViewProjectionMatrixInverse,
739 STATE_MVP_MATRIX, STATE_MATRIX_INVTRANS);
740 MATRIX(gl_ModelViewProjectionMatrixTranspose,
741 STATE_MVP_MATRIX, 0);
742 MATRIX(gl_ModelViewProjectionMatrixInverseTranspose,
743 STATE_MVP_MATRIX, STATE_MATRIX_INVERSE);
744
745 MATRIX(gl_TextureMatrix,
746 STATE_TEXTURE_MATRIX, STATE_MATRIX_TRANSPOSE);
747 MATRIX(gl_TextureMatrixInverse,
748 STATE_TEXTURE_MATRIX, STATE_MATRIX_INVTRANS);
749 MATRIX(gl_TextureMatrixTranspose,
750 STATE_TEXTURE_MATRIX, 0);
751 MATRIX(gl_TextureMatrixInverseTranspose,
752 STATE_TEXTURE_MATRIX, STATE_MATRIX_INVERSE);
753
754 static struct statevar_element gl_NormalMatrix_elements[] = {
755 { NULL, { STATE_MODELVIEW_MATRIX, 0, 0, 0, STATE_MATRIX_INVERSE},
756 SWIZZLE_XYZW },
757 { NULL, { STATE_MODELVIEW_MATRIX, 0, 1, 1, STATE_MATRIX_INVERSE},
758 SWIZZLE_XYZW },
759 { NULL, { STATE_MODELVIEW_MATRIX, 0, 2, 2, STATE_MATRIX_INVERSE},
760 SWIZZLE_XYZW },
761 };
762
763 #undef MATRIX
764
765 #define STATEVAR(name) {#name, name ## _elements, Elements(name ## _elements)}
766
767 static const struct statevar {
768 const char *name;
769 struct statevar_element *elements;
770 unsigned int num_elements;
771 } statevars[] = {
772 STATEVAR(gl_DepthRange),
773 STATEVAR(gl_ClipPlane),
774 STATEVAR(gl_Point),
775 STATEVAR(gl_FrontMaterial),
776 STATEVAR(gl_BackMaterial),
777 STATEVAR(gl_LightSource),
778 STATEVAR(gl_LightModel),
779 STATEVAR(gl_FrontLightModelProduct),
780 STATEVAR(gl_BackLightModelProduct),
781 STATEVAR(gl_FrontLightProduct),
782 STATEVAR(gl_BackLightProduct),
783 STATEVAR(gl_TextureEnvColor),
784 STATEVAR(gl_EyePlaneS),
785 STATEVAR(gl_EyePlaneT),
786 STATEVAR(gl_EyePlaneR),
787 STATEVAR(gl_EyePlaneQ),
788 STATEVAR(gl_ObjectPlaneS),
789 STATEVAR(gl_ObjectPlaneT),
790 STATEVAR(gl_ObjectPlaneR),
791 STATEVAR(gl_ObjectPlaneQ),
792 STATEVAR(gl_Fog),
793
794 STATEVAR(gl_ModelViewMatrix),
795 STATEVAR(gl_ModelViewMatrixInverse),
796 STATEVAR(gl_ModelViewMatrixTranspose),
797 STATEVAR(gl_ModelViewMatrixInverseTranspose),
798
799 STATEVAR(gl_ProjectionMatrix),
800 STATEVAR(gl_ProjectionMatrixInverse),
801 STATEVAR(gl_ProjectionMatrixTranspose),
802 STATEVAR(gl_ProjectionMatrixInverseTranspose),
803
804 STATEVAR(gl_ModelViewProjectionMatrix),
805 STATEVAR(gl_ModelViewProjectionMatrixInverse),
806 STATEVAR(gl_ModelViewProjectionMatrixTranspose),
807 STATEVAR(gl_ModelViewProjectionMatrixInverseTranspose),
808
809 STATEVAR(gl_TextureMatrix),
810 STATEVAR(gl_TextureMatrixInverse),
811 STATEVAR(gl_TextureMatrixTranspose),
812 STATEVAR(gl_TextureMatrixInverseTranspose),
813
814 STATEVAR(gl_NormalMatrix),
815 STATEVAR(gl_NormalScale),
816 };
817
818 void
819 ir_to_mesa_visitor::visit(ir_variable *ir)
820 {
821 if (strcmp(ir->name, "gl_FragCoord") == 0) {
822 struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog;
823
824 fp->OriginUpperLeft = ir->origin_upper_left;
825 fp->PixelCenterInteger = ir->pixel_center_integer;
826 }
827
828 if (ir->mode == ir_var_uniform && strncmp(ir->name, "gl_", 3) == 0) {
829 unsigned int i;
830
831 for (i = 0; i < Elements(statevars); i++) {
832 if (strcmp(ir->name, statevars[i].name) == 0)
833 break;
834 }
835
836 if (i == Elements(statevars)) {
837 fail_link(this->shader_program,
838 "Failed to find builtin uniform `%s'\n", ir->name);
839 return;
840 }
841
842 const struct statevar *statevar = &statevars[i];
843
844 int array_count;
845 if (ir->type->is_array()) {
846 array_count = ir->type->length;
847 } else {
848 array_count = 1;
849 }
850
851 /* Check if this statevar's setup in the STATE file exactly
852 * matches how we'll want to reference it as a
853 * struct/array/whatever. If not, then we need to move it into
854 * temporary storage and hope that it'll get copy-propagated
855 * out.
856 */
857 for (i = 0; i < statevar->num_elements; i++) {
858 if (statevar->elements[i].swizzle != SWIZZLE_XYZW) {
859 break;
860 }
861 }
862
863 struct variable_storage *storage;
864 ir_to_mesa_dst_reg dst;
865 if (i == statevar->num_elements) {
866 /* We'll set the index later. */
867 storage = new(mem_ctx) variable_storage(ir, PROGRAM_STATE_VAR, -1);
868 this->variables.push_tail(storage);
869
870 dst = ir_to_mesa_undef_dst;
871 } else {
872 storage = new(mem_ctx) variable_storage(ir, PROGRAM_TEMPORARY,
873 this->next_temp);
874 this->variables.push_tail(storage);
875 this->next_temp += type_size(ir->type);
876
877 dst = ir_to_mesa_dst_reg_from_src(ir_to_mesa_src_reg(PROGRAM_TEMPORARY,
878 storage->index,
879 NULL));
880 }
881
882
883 for (int a = 0; a < array_count; a++) {
884 for (unsigned int i = 0; i < statevar->num_elements; i++) {
885 struct statevar_element *element = &statevar->elements[i];
886 int tokens[STATE_LENGTH];
887
888 memcpy(tokens, element->tokens, sizeof(element->tokens));
889 if (ir->type->is_array()) {
890 tokens[1] = a;
891 }
892
893 int index = _mesa_add_state_reference(this->prog->Parameters,
894 (gl_state_index *)tokens);
895
896 if (storage->file == PROGRAM_STATE_VAR) {
897 if (storage->index == -1) {
898 storage->index = index;
899 } else {
900 assert(index ==
901 (int)(storage->index + a * statevar->num_elements + i));
902 }
903 } else {
904 ir_to_mesa_src_reg src(PROGRAM_STATE_VAR, index, NULL);
905 src.swizzle = element->swizzle;
906 ir_to_mesa_emit_op1(ir, OPCODE_MOV, dst, src);
907 /* even a float takes up a whole vec4 reg in a struct/array. */
908 dst.index++;
909 }
910 }
911 }
912 if (storage->file == PROGRAM_TEMPORARY &&
913 dst.index != storage->index + type_size(ir->type)) {
914 fail_link(this->shader_program,
915 "failed to load builtin uniform `%s' (%d/%d regs loaded)\n",
916 ir->name, dst.index - storage->index,
917 type_size(ir->type));
918 }
919 }
920 }
921
922 void
923 ir_to_mesa_visitor::visit(ir_loop *ir)
924 {
925 ir_dereference_variable *counter = NULL;
926
927 if (ir->counter != NULL)
928 counter = new(ir) ir_dereference_variable(ir->counter);
929
930 if (ir->from != NULL) {
931 assert(ir->counter != NULL);
932
933 ir_assignment *a = new(ir) ir_assignment(counter, ir->from, NULL);
934
935 a->accept(this);
936 delete a;
937 }
938
939 ir_to_mesa_emit_op0(NULL, OPCODE_BGNLOOP);
940
941 if (ir->to) {
942 ir_expression *e =
943 new(ir) ir_expression(ir->cmp, glsl_type::bool_type,
944 counter, ir->to);
945 ir_if *if_stmt = new(ir) ir_if(e);
946
947 ir_loop_jump *brk = new(ir) ir_loop_jump(ir_loop_jump::jump_break);
948
949 if_stmt->then_instructions.push_tail(brk);
950
951 if_stmt->accept(this);
952
953 delete if_stmt;
954 delete e;
955 delete brk;
956 }
957
958 visit_exec_list(&ir->body_instructions, this);
959
960 if (ir->increment) {
961 ir_expression *e =
962 new(ir) ir_expression(ir_binop_add, counter->type,
963 counter, ir->increment);
964
965 ir_assignment *a = new(ir) ir_assignment(counter, e, NULL);
966
967 a->accept(this);
968 delete a;
969 delete e;
970 }
971
972 ir_to_mesa_emit_op0(NULL, OPCODE_ENDLOOP);
973 }
974
975 void
976 ir_to_mesa_visitor::visit(ir_loop_jump *ir)
977 {
978 switch (ir->mode) {
979 case ir_loop_jump::jump_break:
980 ir_to_mesa_emit_op0(NULL, OPCODE_BRK);
981 break;
982 case ir_loop_jump::jump_continue:
983 ir_to_mesa_emit_op0(NULL, OPCODE_CONT);
984 break;
985 }
986 }
987
988
989 void
990 ir_to_mesa_visitor::visit(ir_function_signature *ir)
991 {
992 assert(0);
993 (void)ir;
994 }
995
996 void
997 ir_to_mesa_visitor::visit(ir_function *ir)
998 {
999 /* Ignore function bodies other than main() -- we shouldn't see calls to
1000 * them since they should all be inlined before we get to ir_to_mesa.
1001 */
1002 if (strcmp(ir->name, "main") == 0) {
1003 const ir_function_signature *sig;
1004 exec_list empty;
1005
1006 sig = ir->matching_signature(&empty);
1007
1008 assert(sig);
1009
1010 foreach_iter(exec_list_iterator, iter, sig->body) {
1011 ir_instruction *ir = (ir_instruction *)iter.get();
1012
1013 ir->accept(this);
1014 }
1015 }
1016 }
1017
1018 GLboolean
1019 ir_to_mesa_visitor::try_emit_mad(ir_expression *ir, int mul_operand)
1020 {
1021 int nonmul_operand = 1 - mul_operand;
1022 ir_to_mesa_src_reg a, b, c;
1023
1024 ir_expression *expr = ir->operands[mul_operand]->as_expression();
1025 if (!expr || expr->operation != ir_binop_mul)
1026 return false;
1027
1028 expr->operands[0]->accept(this);
1029 a = this->result;
1030 expr->operands[1]->accept(this);
1031 b = this->result;
1032 ir->operands[nonmul_operand]->accept(this);
1033 c = this->result;
1034
1035 this->result = get_temp(ir->type);
1036 ir_to_mesa_emit_op3(ir, OPCODE_MAD,
1037 ir_to_mesa_dst_reg_from_src(this->result), a, b, c);
1038
1039 return true;
1040 }
1041
1042 void
1043 ir_to_mesa_visitor::reladdr_to_temp(ir_instruction *ir,
1044 ir_to_mesa_src_reg *reg, int *num_reladdr)
1045 {
1046 if (!reg->reladdr)
1047 return;
1048
1049 ir_to_mesa_emit_op1(ir, OPCODE_ARL, ir_to_mesa_address_reg, *reg->reladdr);
1050
1051 if (*num_reladdr != 1) {
1052 ir_to_mesa_src_reg temp = get_temp(glsl_type::vec4_type);
1053
1054 ir_to_mesa_emit_op1(ir, OPCODE_MOV,
1055 ir_to_mesa_dst_reg_from_src(temp), *reg);
1056 *reg = temp;
1057 }
1058
1059 (*num_reladdr)--;
1060 }
1061
1062 void
1063 ir_to_mesa_visitor::visit(ir_expression *ir)
1064 {
1065 unsigned int operand;
1066 struct ir_to_mesa_src_reg op[2];
1067 struct ir_to_mesa_src_reg result_src;
1068 struct ir_to_mesa_dst_reg result_dst;
1069 const glsl_type *vec4_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, 4, 1);
1070 const glsl_type *vec3_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, 3, 1);
1071 const glsl_type *vec2_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, 2, 1);
1072
1073 /* Quick peephole: Emit OPCODE_MAD(a, b, c) instead of ADD(MUL(a, b), c)
1074 */
1075 if (ir->operation == ir_binop_add) {
1076 if (try_emit_mad(ir, 1))
1077 return;
1078 if (try_emit_mad(ir, 0))
1079 return;
1080 }
1081
1082 for (operand = 0; operand < ir->get_num_operands(); operand++) {
1083 this->result.file = PROGRAM_UNDEFINED;
1084 ir->operands[operand]->accept(this);
1085 if (this->result.file == PROGRAM_UNDEFINED) {
1086 ir_print_visitor v;
1087 printf("Failed to get tree for expression operand:\n");
1088 ir->operands[operand]->accept(&v);
1089 exit(1);
1090 }
1091 op[operand] = this->result;
1092
1093 /* Matrix expression operands should have been broken down to vector
1094 * operations already.
1095 */
1096 assert(!ir->operands[operand]->type->is_matrix());
1097 }
1098
1099 this->result.file = PROGRAM_UNDEFINED;
1100
1101 /* Storage for our result. Ideally for an assignment we'd be using
1102 * the actual storage for the result here, instead.
1103 */
1104 result_src = get_temp(ir->type);
1105 /* convenience for the emit functions below. */
1106 result_dst = ir_to_mesa_dst_reg_from_src(result_src);
1107 /* Limit writes to the channels that will be used by result_src later.
1108 * This does limit this temp's use as a temporary for multi-instruction
1109 * sequences.
1110 */
1111 result_dst.writemask = (1 << ir->type->vector_elements) - 1;
1112
1113 switch (ir->operation) {
1114 case ir_unop_logic_not:
1115 ir_to_mesa_emit_op2(ir, OPCODE_SEQ, result_dst,
1116 op[0], src_reg_for_float(0.0));
1117 break;
1118 case ir_unop_neg:
1119 op[0].negate = ~op[0].negate;
1120 result_src = op[0];
1121 break;
1122 case ir_unop_abs:
1123 ir_to_mesa_emit_op1(ir, OPCODE_ABS, result_dst, op[0]);
1124 break;
1125 case ir_unop_sign:
1126 ir_to_mesa_emit_op1(ir, OPCODE_SSG, result_dst, op[0]);
1127 break;
1128 case ir_unop_rcp:
1129 ir_to_mesa_emit_scalar_op1(ir, OPCODE_RCP, result_dst, op[0]);
1130 break;
1131
1132 case ir_unop_exp2:
1133 ir_to_mesa_emit_scalar_op1(ir, OPCODE_EX2, result_dst, op[0]);
1134 break;
1135 case ir_unop_exp:
1136 case ir_unop_log:
1137 assert(!"not reached: should be handled by ir_explog_to_explog2");
1138 break;
1139 case ir_unop_log2:
1140 ir_to_mesa_emit_scalar_op1(ir, OPCODE_LG2, result_dst, op[0]);
1141 break;
1142 case ir_unop_sin:
1143 ir_to_mesa_emit_scalar_op1(ir, OPCODE_SIN, result_dst, op[0]);
1144 break;
1145 case ir_unop_cos:
1146 ir_to_mesa_emit_scalar_op1(ir, OPCODE_COS, result_dst, op[0]);
1147 break;
1148
1149 case ir_unop_dFdx:
1150 ir_to_mesa_emit_op1(ir, OPCODE_DDX, result_dst, op[0]);
1151 break;
1152 case ir_unop_dFdy:
1153 ir_to_mesa_emit_op1(ir, OPCODE_DDY, result_dst, op[0]);
1154 break;
1155
1156 case ir_unop_noise: {
1157 const enum prog_opcode opcode =
1158 prog_opcode(OPCODE_NOISE1
1159 + (ir->operands[0]->type->vector_elements) - 1);
1160 assert((opcode >= OPCODE_NOISE1) && (opcode <= OPCODE_NOISE4));
1161
1162 ir_to_mesa_emit_op1(ir, opcode, result_dst, op[0]);
1163 break;
1164 }
1165
1166 case ir_binop_add:
1167 ir_to_mesa_emit_op2(ir, OPCODE_ADD, result_dst, op[0], op[1]);
1168 break;
1169 case ir_binop_sub:
1170 ir_to_mesa_emit_op2(ir, OPCODE_SUB, result_dst, op[0], op[1]);
1171 break;
1172
1173 case ir_binop_mul:
1174 ir_to_mesa_emit_op2(ir, OPCODE_MUL, result_dst, op[0], op[1]);
1175 break;
1176 case ir_binop_div:
1177 assert(!"not reached: should be handled by ir_div_to_mul_rcp");
1178 case ir_binop_mod:
1179 assert(!"ir_binop_mod should have been converted to b * fract(a/b)");
1180 break;
1181
1182 case ir_binop_less:
1183 ir_to_mesa_emit_op2(ir, OPCODE_SLT, result_dst, op[0], op[1]);
1184 break;
1185 case ir_binop_greater:
1186 ir_to_mesa_emit_op2(ir, OPCODE_SGT, result_dst, op[0], op[1]);
1187 break;
1188 case ir_binop_lequal:
1189 ir_to_mesa_emit_op2(ir, OPCODE_SLE, result_dst, op[0], op[1]);
1190 break;
1191 case ir_binop_gequal:
1192 ir_to_mesa_emit_op2(ir, OPCODE_SGE, result_dst, op[0], op[1]);
1193 break;
1194 case ir_binop_equal:
1195 ir_to_mesa_emit_op2(ir, OPCODE_SEQ, result_dst, op[0], op[1]);
1196 break;
1197 case ir_binop_nequal:
1198 ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst, op[0], op[1]);
1199 break;
1200 case ir_binop_all_equal:
1201 /* "==" operator producing a scalar boolean. */
1202 if (ir->operands[0]->type->is_vector() ||
1203 ir->operands[1]->type->is_vector()) {
1204 ir_to_mesa_src_reg temp = get_temp(glsl_type::vec4_type);
1205 ir_to_mesa_emit_op2(ir, OPCODE_SNE,
1206 ir_to_mesa_dst_reg_from_src(temp), op[0], op[1]);
1207 ir_to_mesa_emit_op2(ir, OPCODE_DP4, result_dst, temp, temp);
1208 ir_to_mesa_emit_op2(ir, OPCODE_SEQ,
1209 result_dst, result_src, src_reg_for_float(0.0));
1210 } else {
1211 ir_to_mesa_emit_op2(ir, OPCODE_SEQ, result_dst, op[0], op[1]);
1212 }
1213 break;
1214 case ir_binop_any_nequal:
1215 /* "!=" operator producing a scalar boolean. */
1216 if (ir->operands[0]->type->is_vector() ||
1217 ir->operands[1]->type->is_vector()) {
1218 ir_to_mesa_src_reg temp = get_temp(glsl_type::vec4_type);
1219 ir_to_mesa_emit_op2(ir, OPCODE_SNE,
1220 ir_to_mesa_dst_reg_from_src(temp), op[0], op[1]);
1221 ir_to_mesa_emit_op2(ir, OPCODE_DP4, result_dst, temp, temp);
1222 ir_to_mesa_emit_op2(ir, OPCODE_SNE,
1223 result_dst, result_src, src_reg_for_float(0.0));
1224 } else {
1225 ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst, op[0], op[1]);
1226 }
1227 break;
1228
1229 case ir_unop_any:
1230 switch (ir->operands[0]->type->vector_elements) {
1231 case 4:
1232 ir_to_mesa_emit_op2(ir, OPCODE_DP4, result_dst, op[0], op[0]);
1233 break;
1234 case 3:
1235 ir_to_mesa_emit_op2(ir, OPCODE_DP3, result_dst, op[0], op[0]);
1236 break;
1237 case 2:
1238 ir_to_mesa_emit_op2(ir, OPCODE_DP2, result_dst, op[0], op[0]);
1239 break;
1240 default:
1241 assert(!"unreached: ir_unop_any of non-bvec");
1242 break;
1243 }
1244 ir_to_mesa_emit_op2(ir, OPCODE_SNE,
1245 result_dst, result_src, src_reg_for_float(0.0));
1246 break;
1247
1248 case ir_binop_logic_xor:
1249 ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst, op[0], op[1]);
1250 break;
1251
1252 case ir_binop_logic_or:
1253 /* This could be a saturated add and skip the SNE. */
1254 ir_to_mesa_emit_op2(ir, OPCODE_ADD,
1255 result_dst,
1256 op[0], op[1]);
1257
1258 ir_to_mesa_emit_op2(ir, OPCODE_SNE,
1259 result_dst,
1260 result_src, src_reg_for_float(0.0));
1261 break;
1262
1263 case ir_binop_logic_and:
1264 /* the bool args are stored as float 0.0 or 1.0, so "mul" gives us "and". */
1265 ir_to_mesa_emit_op2(ir, OPCODE_MUL,
1266 result_dst,
1267 op[0], op[1]);
1268 break;
1269
1270 case ir_binop_dot:
1271 if (ir->operands[0]->type == vec4_type) {
1272 assert(ir->operands[1]->type == vec4_type);
1273 ir_to_mesa_emit_op2(ir, OPCODE_DP4,
1274 result_dst,
1275 op[0], op[1]);
1276 } else if (ir->operands[0]->type == vec3_type) {
1277 assert(ir->operands[1]->type == vec3_type);
1278 ir_to_mesa_emit_op2(ir, OPCODE_DP3,
1279 result_dst,
1280 op[0], op[1]);
1281 } else if (ir->operands[0]->type == vec2_type) {
1282 assert(ir->operands[1]->type == vec2_type);
1283 ir_to_mesa_emit_op2(ir, OPCODE_DP2,
1284 result_dst,
1285 op[0], op[1]);
1286 }
1287 break;
1288
1289 case ir_binop_cross:
1290 ir_to_mesa_emit_op2(ir, OPCODE_XPD, result_dst, op[0], op[1]);
1291 break;
1292
1293 case ir_unop_sqrt:
1294 /* sqrt(x) = x * rsq(x). */
1295 ir_to_mesa_emit_scalar_op1(ir, OPCODE_RSQ, result_dst, op[0]);
1296 ir_to_mesa_emit_op2(ir, OPCODE_MUL, result_dst, result_src, op[0]);
1297 /* For incoming channels <= 0, set the result to 0. */
1298 op[0].negate = ~op[0].negate;
1299 ir_to_mesa_emit_op3(ir, OPCODE_CMP, result_dst,
1300 op[0], result_src, src_reg_for_float(0.0));
1301 break;
1302 case ir_unop_rsq:
1303 ir_to_mesa_emit_scalar_op1(ir, OPCODE_RSQ, result_dst, op[0]);
1304 break;
1305 case ir_unop_i2f:
1306 case ir_unop_b2f:
1307 case ir_unop_b2i:
1308 /* Mesa IR lacks types, ints are stored as truncated floats. */
1309 result_src = op[0];
1310 break;
1311 case ir_unop_f2i:
1312 ir_to_mesa_emit_op1(ir, OPCODE_TRUNC, result_dst, op[0]);
1313 break;
1314 case ir_unop_f2b:
1315 case ir_unop_i2b:
1316 ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst,
1317 op[0], src_reg_for_float(0.0));
1318 break;
1319 case ir_unop_trunc:
1320 ir_to_mesa_emit_op1(ir, OPCODE_TRUNC, result_dst, op[0]);
1321 break;
1322 case ir_unop_ceil:
1323 op[0].negate = ~op[0].negate;
1324 ir_to_mesa_emit_op1(ir, OPCODE_FLR, result_dst, op[0]);
1325 result_src.negate = ~result_src.negate;
1326 break;
1327 case ir_unop_floor:
1328 ir_to_mesa_emit_op1(ir, OPCODE_FLR, result_dst, op[0]);
1329 break;
1330 case ir_unop_fract:
1331 ir_to_mesa_emit_op1(ir, OPCODE_FRC, result_dst, op[0]);
1332 break;
1333
1334 case ir_binop_min:
1335 ir_to_mesa_emit_op2(ir, OPCODE_MIN, result_dst, op[0], op[1]);
1336 break;
1337 case ir_binop_max:
1338 ir_to_mesa_emit_op2(ir, OPCODE_MAX, result_dst, op[0], op[1]);
1339 break;
1340 case ir_binop_pow:
1341 ir_to_mesa_emit_scalar_op2(ir, OPCODE_POW, result_dst, op[0], op[1]);
1342 break;
1343
1344 case ir_unop_bit_not:
1345 case ir_unop_u2f:
1346 case ir_binop_lshift:
1347 case ir_binop_rshift:
1348 case ir_binop_bit_and:
1349 case ir_binop_bit_xor:
1350 case ir_binop_bit_or:
1351 assert(!"GLSL 1.30 features unsupported");
1352 break;
1353 }
1354
1355 this->result = result_src;
1356 }
1357
1358
1359 void
1360 ir_to_mesa_visitor::visit(ir_swizzle *ir)
1361 {
1362 ir_to_mesa_src_reg src_reg;
1363 int i;
1364 int swizzle[4];
1365
1366 /* Note that this is only swizzles in expressions, not those on the left
1367 * hand side of an assignment, which do write masking. See ir_assignment
1368 * for that.
1369 */
1370
1371 ir->val->accept(this);
1372 src_reg = this->result;
1373 assert(src_reg.file != PROGRAM_UNDEFINED);
1374
1375 for (i = 0; i < 4; i++) {
1376 if (i < ir->type->vector_elements) {
1377 switch (i) {
1378 case 0:
1379 swizzle[i] = GET_SWZ(src_reg.swizzle, ir->mask.x);
1380 break;
1381 case 1:
1382 swizzle[i] = GET_SWZ(src_reg.swizzle, ir->mask.y);
1383 break;
1384 case 2:
1385 swizzle[i] = GET_SWZ(src_reg.swizzle, ir->mask.z);
1386 break;
1387 case 3:
1388 swizzle[i] = GET_SWZ(src_reg.swizzle, ir->mask.w);
1389 break;
1390 }
1391 } else {
1392 /* If the type is smaller than a vec4, replicate the last
1393 * channel out.
1394 */
1395 swizzle[i] = swizzle[ir->type->vector_elements - 1];
1396 }
1397 }
1398
1399 src_reg.swizzle = MAKE_SWIZZLE4(swizzle[0],
1400 swizzle[1],
1401 swizzle[2],
1402 swizzle[3]);
1403
1404 this->result = src_reg;
1405 }
1406
1407 void
1408 ir_to_mesa_visitor::visit(ir_dereference_variable *ir)
1409 {
1410 variable_storage *entry = find_variable_storage(ir->var);
1411
1412 if (!entry) {
1413 switch (ir->var->mode) {
1414 case ir_var_uniform:
1415 entry = new(mem_ctx) variable_storage(ir->var, PROGRAM_UNIFORM,
1416 ir->var->location);
1417 this->variables.push_tail(entry);
1418 break;
1419 case ir_var_in:
1420 case ir_var_out:
1421 case ir_var_inout:
1422 /* The linker assigns locations for varyings and attributes,
1423 * including deprecated builtins (like gl_Color), user-assign
1424 * generic attributes (glBindVertexLocation), and
1425 * user-defined varyings.
1426 *
1427 * FINISHME: We would hit this path for function arguments. Fix!
1428 */
1429 assert(ir->var->location != -1);
1430 if (ir->var->mode == ir_var_in ||
1431 ir->var->mode == ir_var_inout) {
1432 entry = new(mem_ctx) variable_storage(ir->var,
1433 PROGRAM_INPUT,
1434 ir->var->location);
1435
1436 if (this->prog->Target == GL_VERTEX_PROGRAM_ARB &&
1437 ir->var->location >= VERT_ATTRIB_GENERIC0) {
1438 _mesa_add_attribute(prog->Attributes,
1439 ir->var->name,
1440 _mesa_sizeof_glsl_type(ir->var->type->gl_type),
1441 ir->var->type->gl_type,
1442 ir->var->location - VERT_ATTRIB_GENERIC0);
1443 }
1444 } else {
1445 entry = new(mem_ctx) variable_storage(ir->var,
1446 PROGRAM_OUTPUT,
1447 ir->var->location);
1448 }
1449
1450 break;
1451 case ir_var_auto:
1452 case ir_var_temporary:
1453 entry = new(mem_ctx) variable_storage(ir->var, PROGRAM_TEMPORARY,
1454 this->next_temp);
1455 this->variables.push_tail(entry);
1456
1457 next_temp += type_size(ir->var->type);
1458 break;
1459 }
1460
1461 if (!entry) {
1462 printf("Failed to make storage for %s\n", ir->var->name);
1463 exit(1);
1464 }
1465 }
1466
1467 this->result = ir_to_mesa_src_reg(entry->file, entry->index, ir->var->type);
1468 }
1469
1470 void
1471 ir_to_mesa_visitor::visit(ir_dereference_array *ir)
1472 {
1473 ir_constant *index;
1474 ir_to_mesa_src_reg src_reg;
1475 int element_size = type_size(ir->type);
1476
1477 index = ir->array_index->constant_expression_value();
1478
1479 ir->array->accept(this);
1480 src_reg = this->result;
1481
1482 if (index) {
1483 src_reg.index += index->value.i[0] * element_size;
1484 } else {
1485 ir_to_mesa_src_reg array_base = this->result;
1486 /* Variable index array dereference. It eats the "vec4" of the
1487 * base of the array and an index that offsets the Mesa register
1488 * index.
1489 */
1490 ir->array_index->accept(this);
1491
1492 ir_to_mesa_src_reg index_reg;
1493
1494 if (element_size == 1) {
1495 index_reg = this->result;
1496 } else {
1497 index_reg = get_temp(glsl_type::float_type);
1498
1499 ir_to_mesa_emit_op2(ir, OPCODE_MUL,
1500 ir_to_mesa_dst_reg_from_src(index_reg),
1501 this->result, src_reg_for_float(element_size));
1502 }
1503
1504 src_reg.reladdr = talloc(mem_ctx, ir_to_mesa_src_reg);
1505 memcpy(src_reg.reladdr, &index_reg, sizeof(index_reg));
1506 }
1507
1508 /* If the type is smaller than a vec4, replicate the last channel out. */
1509 if (ir->type->is_scalar() || ir->type->is_vector())
1510 src_reg.swizzle = swizzle_for_size(ir->type->vector_elements);
1511 else
1512 src_reg.swizzle = SWIZZLE_NOOP;
1513
1514 this->result = src_reg;
1515 }
1516
1517 void
1518 ir_to_mesa_visitor::visit(ir_dereference_record *ir)
1519 {
1520 unsigned int i;
1521 const glsl_type *struct_type = ir->record->type;
1522 int offset = 0;
1523
1524 ir->record->accept(this);
1525
1526 for (i = 0; i < struct_type->length; i++) {
1527 if (strcmp(struct_type->fields.structure[i].name, ir->field) == 0)
1528 break;
1529 offset += type_size(struct_type->fields.structure[i].type);
1530 }
1531 this->result.swizzle = swizzle_for_size(ir->type->vector_elements);
1532 this->result.index += offset;
1533 }
1534
1535 /**
1536 * We want to be careful in assignment setup to hit the actual storage
1537 * instead of potentially using a temporary like we might with the
1538 * ir_dereference handler.
1539 */
1540 static struct ir_to_mesa_dst_reg
1541 get_assignment_lhs(ir_dereference *ir, ir_to_mesa_visitor *v)
1542 {
1543 /* The LHS must be a dereference. If the LHS is a variable indexed array
1544 * access of a vector, it must be separated into a series conditional moves
1545 * before reaching this point (see ir_vec_index_to_cond_assign).
1546 */
1547 assert(ir->as_dereference());
1548 ir_dereference_array *deref_array = ir->as_dereference_array();
1549 if (deref_array) {
1550 assert(!deref_array->array->type->is_vector());
1551 }
1552
1553 /* Use the rvalue deref handler for the most part. We'll ignore
1554 * swizzles in it and write swizzles using writemask, though.
1555 */
1556 ir->accept(v);
1557 return ir_to_mesa_dst_reg_from_src(v->result);
1558 }
1559
1560 void
1561 ir_to_mesa_visitor::visit(ir_assignment *ir)
1562 {
1563 struct ir_to_mesa_dst_reg l;
1564 struct ir_to_mesa_src_reg r;
1565 int i;
1566
1567 ir->rhs->accept(this);
1568 r = this->result;
1569
1570 l = get_assignment_lhs(ir->lhs, this);
1571
1572 /* FINISHME: This should really set to the correct maximal writemask for each
1573 * FINISHME: component written (in the loops below). This case can only
1574 * FINISHME: occur for matrices, arrays, and structures.
1575 */
1576 if (ir->write_mask == 0) {
1577 assert(!ir->lhs->type->is_scalar() && !ir->lhs->type->is_vector());
1578 l.writemask = WRITEMASK_XYZW;
1579 } else if (ir->lhs->type->is_scalar()) {
1580 /* FINISHME: This hack makes writing to gl_FragData, which lives in the
1581 * FINISHME: W component of fragment shader output zero, work correctly.
1582 */
1583 l.writemask = WRITEMASK_XYZW;
1584 } else {
1585 assert(ir->lhs->type->is_vector());
1586 l.writemask = ir->write_mask;
1587 }
1588
1589 assert(l.file != PROGRAM_UNDEFINED);
1590 assert(r.file != PROGRAM_UNDEFINED);
1591
1592 if (ir->condition) {
1593 ir_to_mesa_src_reg condition;
1594
1595 ir->condition->accept(this);
1596 condition = this->result;
1597
1598 /* We use the OPCODE_CMP (a < 0 ? b : c) for conditional moves,
1599 * and the condition we produced is 0.0 or 1.0. By flipping the
1600 * sign, we can choose which value OPCODE_CMP produces without
1601 * an extra computing the condition.
1602 */
1603 condition.negate = ~condition.negate;
1604 for (i = 0; i < type_size(ir->lhs->type); i++) {
1605 ir_to_mesa_emit_op3(ir, OPCODE_CMP, l,
1606 condition, r, ir_to_mesa_src_reg_from_dst(l));
1607 l.index++;
1608 r.index++;
1609 }
1610 } else {
1611 for (i = 0; i < type_size(ir->lhs->type); i++) {
1612 ir_to_mesa_emit_op1(ir, OPCODE_MOV, l, r);
1613 l.index++;
1614 r.index++;
1615 }
1616 }
1617 }
1618
1619
1620 void
1621 ir_to_mesa_visitor::visit(ir_constant *ir)
1622 {
1623 ir_to_mesa_src_reg src_reg;
1624 GLfloat stack_vals[4] = { 0 };
1625 GLfloat *values = stack_vals;
1626 unsigned int i;
1627
1628 /* Unfortunately, 4 floats is all we can get into
1629 * _mesa_add_unnamed_constant. So, make a temp to store an
1630 * aggregate constant and move each constant value into it. If we
1631 * get lucky, copy propagation will eliminate the extra moves.
1632 */
1633
1634 if (ir->type->base_type == GLSL_TYPE_STRUCT) {
1635 ir_to_mesa_src_reg temp_base = get_temp(ir->type);
1636 ir_to_mesa_dst_reg temp = ir_to_mesa_dst_reg_from_src(temp_base);
1637
1638 foreach_iter(exec_list_iterator, iter, ir->components) {
1639 ir_constant *field_value = (ir_constant *)iter.get();
1640 int size = type_size(field_value->type);
1641
1642 assert(size > 0);
1643
1644 field_value->accept(this);
1645 src_reg = this->result;
1646
1647 for (i = 0; i < (unsigned int)size; i++) {
1648 ir_to_mesa_emit_op1(ir, OPCODE_MOV, temp, src_reg);
1649
1650 src_reg.index++;
1651 temp.index++;
1652 }
1653 }
1654 this->result = temp_base;
1655 return;
1656 }
1657
1658 if (ir->type->is_array()) {
1659 ir_to_mesa_src_reg temp_base = get_temp(ir->type);
1660 ir_to_mesa_dst_reg temp = ir_to_mesa_dst_reg_from_src(temp_base);
1661 int size = type_size(ir->type->fields.array);
1662
1663 assert(size > 0);
1664
1665 for (i = 0; i < ir->type->length; i++) {
1666 ir->array_elements[i]->accept(this);
1667 src_reg = this->result;
1668 for (int j = 0; j < size; j++) {
1669 ir_to_mesa_emit_op1(ir, OPCODE_MOV, temp, src_reg);
1670
1671 src_reg.index++;
1672 temp.index++;
1673 }
1674 }
1675 this->result = temp_base;
1676 return;
1677 }
1678
1679 if (ir->type->is_matrix()) {
1680 ir_to_mesa_src_reg mat = get_temp(ir->type);
1681 ir_to_mesa_dst_reg mat_column = ir_to_mesa_dst_reg_from_src(mat);
1682
1683 for (i = 0; i < ir->type->matrix_columns; i++) {
1684 assert(ir->type->base_type == GLSL_TYPE_FLOAT);
1685 values = &ir->value.f[i * ir->type->vector_elements];
1686
1687 src_reg = ir_to_mesa_src_reg(PROGRAM_CONSTANT, -1, NULL);
1688 src_reg.index = _mesa_add_unnamed_constant(this->prog->Parameters,
1689 values,
1690 ir->type->vector_elements,
1691 &src_reg.swizzle);
1692 ir_to_mesa_emit_op1(ir, OPCODE_MOV, mat_column, src_reg);
1693
1694 mat_column.index++;
1695 }
1696
1697 this->result = mat;
1698 return;
1699 }
1700
1701 src_reg.file = PROGRAM_CONSTANT;
1702 switch (ir->type->base_type) {
1703 case GLSL_TYPE_FLOAT:
1704 values = &ir->value.f[0];
1705 break;
1706 case GLSL_TYPE_UINT:
1707 for (i = 0; i < ir->type->vector_elements; i++) {
1708 values[i] = ir->value.u[i];
1709 }
1710 break;
1711 case GLSL_TYPE_INT:
1712 for (i = 0; i < ir->type->vector_elements; i++) {
1713 values[i] = ir->value.i[i];
1714 }
1715 break;
1716 case GLSL_TYPE_BOOL:
1717 for (i = 0; i < ir->type->vector_elements; i++) {
1718 values[i] = ir->value.b[i];
1719 }
1720 break;
1721 default:
1722 assert(!"Non-float/uint/int/bool constant");
1723 }
1724
1725 this->result = ir_to_mesa_src_reg(PROGRAM_CONSTANT, -1, ir->type);
1726 this->result.index = _mesa_add_unnamed_constant(this->prog->Parameters,
1727 values,
1728 ir->type->vector_elements,
1729 &this->result.swizzle);
1730 }
1731
1732 function_entry *
1733 ir_to_mesa_visitor::get_function_signature(ir_function_signature *sig)
1734 {
1735 function_entry *entry;
1736
1737 foreach_iter(exec_list_iterator, iter, this->function_signatures) {
1738 entry = (function_entry *)iter.get();
1739
1740 if (entry->sig == sig)
1741 return entry;
1742 }
1743
1744 entry = talloc(mem_ctx, function_entry);
1745 entry->sig = sig;
1746 entry->sig_id = this->next_signature_id++;
1747 entry->bgn_inst = NULL;
1748
1749 /* Allocate storage for all the parameters. */
1750 foreach_iter(exec_list_iterator, iter, sig->parameters) {
1751 ir_variable *param = (ir_variable *)iter.get();
1752 variable_storage *storage;
1753
1754 storage = find_variable_storage(param);
1755 assert(!storage);
1756
1757 storage = new(mem_ctx) variable_storage(param, PROGRAM_TEMPORARY,
1758 this->next_temp);
1759 this->variables.push_tail(storage);
1760
1761 this->next_temp += type_size(param->type);
1762 }
1763
1764 if (!sig->return_type->is_void()) {
1765 entry->return_reg = get_temp(sig->return_type);
1766 } else {
1767 entry->return_reg = ir_to_mesa_undef;
1768 }
1769
1770 this->function_signatures.push_tail(entry);
1771 return entry;
1772 }
1773
1774 void
1775 ir_to_mesa_visitor::visit(ir_call *ir)
1776 {
1777 ir_to_mesa_instruction *call_inst;
1778 ir_function_signature *sig = ir->get_callee();
1779 function_entry *entry = get_function_signature(sig);
1780 int i;
1781
1782 /* Process in parameters. */
1783 exec_list_iterator sig_iter = sig->parameters.iterator();
1784 foreach_iter(exec_list_iterator, iter, *ir) {
1785 ir_rvalue *param_rval = (ir_rvalue *)iter.get();
1786 ir_variable *param = (ir_variable *)sig_iter.get();
1787
1788 if (param->mode == ir_var_in ||
1789 param->mode == ir_var_inout) {
1790 variable_storage *storage = find_variable_storage(param);
1791 assert(storage);
1792
1793 param_rval->accept(this);
1794 ir_to_mesa_src_reg r = this->result;
1795
1796 ir_to_mesa_dst_reg l;
1797 l.file = storage->file;
1798 l.index = storage->index;
1799 l.reladdr = NULL;
1800 l.writemask = WRITEMASK_XYZW;
1801 l.cond_mask = COND_TR;
1802
1803 for (i = 0; i < type_size(param->type); i++) {
1804 ir_to_mesa_emit_op1(ir, OPCODE_MOV, l, r);
1805 l.index++;
1806 r.index++;
1807 }
1808 }
1809
1810 sig_iter.next();
1811 }
1812 assert(!sig_iter.has_next());
1813
1814 /* Emit call instruction */
1815 call_inst = ir_to_mesa_emit_op1(ir, OPCODE_CAL,
1816 ir_to_mesa_undef_dst, ir_to_mesa_undef);
1817 call_inst->function = entry;
1818
1819 /* Process out parameters. */
1820 sig_iter = sig->parameters.iterator();
1821 foreach_iter(exec_list_iterator, iter, *ir) {
1822 ir_rvalue *param_rval = (ir_rvalue *)iter.get();
1823 ir_variable *param = (ir_variable *)sig_iter.get();
1824
1825 if (param->mode == ir_var_out ||
1826 param->mode == ir_var_inout) {
1827 variable_storage *storage = find_variable_storage(param);
1828 assert(storage);
1829
1830 ir_to_mesa_src_reg r;
1831 r.file = storage->file;
1832 r.index = storage->index;
1833 r.reladdr = NULL;
1834 r.swizzle = SWIZZLE_NOOP;
1835 r.negate = 0;
1836
1837 param_rval->accept(this);
1838 ir_to_mesa_dst_reg l = ir_to_mesa_dst_reg_from_src(this->result);
1839
1840 for (i = 0; i < type_size(param->type); i++) {
1841 ir_to_mesa_emit_op1(ir, OPCODE_MOV, l, r);
1842 l.index++;
1843 r.index++;
1844 }
1845 }
1846
1847 sig_iter.next();
1848 }
1849 assert(!sig_iter.has_next());
1850
1851 /* Process return value. */
1852 this->result = entry->return_reg;
1853 }
1854
1855 class get_sampler_name : public ir_hierarchical_visitor
1856 {
1857 public:
1858 get_sampler_name(ir_to_mesa_visitor *mesa, ir_dereference *last)
1859 {
1860 this->mem_ctx = mesa->mem_ctx;
1861 this->mesa = mesa;
1862 this->name = NULL;
1863 this->offset = 0;
1864 this->last = last;
1865 }
1866
1867 virtual ir_visitor_status visit(ir_dereference_variable *ir)
1868 {
1869 this->name = ir->var->name;
1870 return visit_continue;
1871 }
1872
1873 virtual ir_visitor_status visit_leave(ir_dereference_record *ir)
1874 {
1875 this->name = talloc_asprintf(mem_ctx, "%s.%s", name, ir->field);
1876 return visit_continue;
1877 }
1878
1879 virtual ir_visitor_status visit_leave(ir_dereference_array *ir)
1880 {
1881 ir_constant *index = ir->array_index->as_constant();
1882 int i;
1883
1884 if (index) {
1885 i = index->value.i[0];
1886 } else {
1887 /* GLSL 1.10 and 1.20 allowed variable sampler array indices,
1888 * while GLSL 1.30 requires that the array indices be
1889 * constant integer expressions. We don't expect any driver
1890 * to actually work with a really variable array index, so
1891 * all that would work would be an unrolled loop counter that ends
1892 * up being constant above.
1893 */
1894 mesa->shader_program->InfoLog =
1895 talloc_asprintf_append(mesa->shader_program->InfoLog,
1896 "warning: Variable sampler array index "
1897 "unsupported.\nThis feature of the language "
1898 "was removed in GLSL 1.20 and is unlikely "
1899 "to be supported for 1.10 in Mesa.\n");
1900 i = 0;
1901 }
1902 if (ir != last) {
1903 this->name = talloc_asprintf(mem_ctx, "%s[%d]", name, i);
1904 } else {
1905 offset = i;
1906 }
1907 return visit_continue;
1908 }
1909
1910 ir_to_mesa_visitor *mesa;
1911 const char *name;
1912 void *mem_ctx;
1913 int offset;
1914 ir_dereference *last;
1915 };
1916
1917 int
1918 ir_to_mesa_visitor::get_sampler_uniform_value(ir_dereference *sampler)
1919 {
1920 get_sampler_name getname(this, sampler);
1921
1922 sampler->accept(&getname);
1923
1924 GLint index = _mesa_lookup_parameter_index(prog->Parameters, -1,
1925 getname.name);
1926
1927 if (index < 0) {
1928 fail_link(this->shader_program,
1929 "failed to find sampler named %s.\n", getname.name);
1930 return 0;
1931 }
1932
1933 index += getname.offset;
1934
1935 return this->prog->Parameters->ParameterValues[index][0];
1936 }
1937
1938 void
1939 ir_to_mesa_visitor::visit(ir_texture *ir)
1940 {
1941 ir_to_mesa_src_reg result_src, coord, lod_info, projector;
1942 ir_to_mesa_dst_reg result_dst, coord_dst;
1943 ir_to_mesa_instruction *inst = NULL;
1944 prog_opcode opcode = OPCODE_NOP;
1945
1946 ir->coordinate->accept(this);
1947
1948 /* Put our coords in a temp. We'll need to modify them for shadow,
1949 * projection, or LOD, so the only case we'd use it as is is if
1950 * we're doing plain old texturing. Mesa IR optimization should
1951 * handle cleaning up our mess in that case.
1952 */
1953 coord = get_temp(glsl_type::vec4_type);
1954 coord_dst = ir_to_mesa_dst_reg_from_src(coord);
1955 ir_to_mesa_emit_op1(ir, OPCODE_MOV, coord_dst,
1956 this->result);
1957
1958 if (ir->projector) {
1959 ir->projector->accept(this);
1960 projector = this->result;
1961 }
1962
1963 /* Storage for our result. Ideally for an assignment we'd be using
1964 * the actual storage for the result here, instead.
1965 */
1966 result_src = get_temp(glsl_type::vec4_type);
1967 result_dst = ir_to_mesa_dst_reg_from_src(result_src);
1968
1969 switch (ir->op) {
1970 case ir_tex:
1971 opcode = OPCODE_TEX;
1972 break;
1973 case ir_txb:
1974 opcode = OPCODE_TXB;
1975 ir->lod_info.bias->accept(this);
1976 lod_info = this->result;
1977 break;
1978 case ir_txl:
1979 opcode = OPCODE_TXL;
1980 ir->lod_info.lod->accept(this);
1981 lod_info = this->result;
1982 break;
1983 case ir_txd:
1984 case ir_txf:
1985 assert(!"GLSL 1.30 features unsupported");
1986 break;
1987 }
1988
1989 if (ir->projector) {
1990 if (opcode == OPCODE_TEX) {
1991 /* Slot the projector in as the last component of the coord. */
1992 coord_dst.writemask = WRITEMASK_W;
1993 ir_to_mesa_emit_op1(ir, OPCODE_MOV, coord_dst, projector);
1994 coord_dst.writemask = WRITEMASK_XYZW;
1995 opcode = OPCODE_TXP;
1996 } else {
1997 ir_to_mesa_src_reg coord_w = coord;
1998 coord_w.swizzle = SWIZZLE_WWWW;
1999
2000 /* For the other TEX opcodes there's no projective version
2001 * since the last slot is taken up by lod info. Do the
2002 * projective divide now.
2003 */
2004 coord_dst.writemask = WRITEMASK_W;
2005 ir_to_mesa_emit_op1(ir, OPCODE_RCP, coord_dst, projector);
2006
2007 coord_dst.writemask = WRITEMASK_XYZ;
2008 ir_to_mesa_emit_op2(ir, OPCODE_MUL, coord_dst, coord, coord_w);
2009
2010 coord_dst.writemask = WRITEMASK_XYZW;
2011 coord.swizzle = SWIZZLE_XYZW;
2012 }
2013 }
2014
2015 if (ir->shadow_comparitor) {
2016 /* Slot the shadow value in as the second to last component of the
2017 * coord.
2018 */
2019 ir->shadow_comparitor->accept(this);
2020 coord_dst.writemask = WRITEMASK_Z;
2021 ir_to_mesa_emit_op1(ir, OPCODE_MOV, coord_dst, this->result);
2022 coord_dst.writemask = WRITEMASK_XYZW;
2023 }
2024
2025 if (opcode == OPCODE_TXL || opcode == OPCODE_TXB) {
2026 /* Mesa IR stores lod or lod bias in the last channel of the coords. */
2027 coord_dst.writemask = WRITEMASK_W;
2028 ir_to_mesa_emit_op1(ir, OPCODE_MOV, coord_dst, lod_info);
2029 coord_dst.writemask = WRITEMASK_XYZW;
2030 }
2031
2032 inst = ir_to_mesa_emit_op1(ir, opcode, result_dst, coord);
2033
2034 if (ir->shadow_comparitor)
2035 inst->tex_shadow = GL_TRUE;
2036
2037 inst->sampler = get_sampler_uniform_value(ir->sampler);
2038
2039 const glsl_type *sampler_type = ir->sampler->type;
2040
2041 switch (sampler_type->sampler_dimensionality) {
2042 case GLSL_SAMPLER_DIM_1D:
2043 inst->tex_target = (sampler_type->sampler_array)
2044 ? TEXTURE_1D_ARRAY_INDEX : TEXTURE_1D_INDEX;
2045 break;
2046 case GLSL_SAMPLER_DIM_2D:
2047 inst->tex_target = (sampler_type->sampler_array)
2048 ? TEXTURE_2D_ARRAY_INDEX : TEXTURE_2D_INDEX;
2049 break;
2050 case GLSL_SAMPLER_DIM_3D:
2051 inst->tex_target = TEXTURE_3D_INDEX;
2052 break;
2053 case GLSL_SAMPLER_DIM_CUBE:
2054 inst->tex_target = TEXTURE_CUBE_INDEX;
2055 break;
2056 case GLSL_SAMPLER_DIM_RECT:
2057 inst->tex_target = TEXTURE_RECT_INDEX;
2058 break;
2059 case GLSL_SAMPLER_DIM_BUF:
2060 assert(!"FINISHME: Implement ARB_texture_buffer_object");
2061 break;
2062 default:
2063 assert(!"Should not get here.");
2064 }
2065
2066 this->result = result_src;
2067 }
2068
2069 void
2070 ir_to_mesa_visitor::visit(ir_return *ir)
2071 {
2072 if (ir->get_value()) {
2073 ir_to_mesa_dst_reg l;
2074 int i;
2075
2076 assert(current_function);
2077
2078 ir->get_value()->accept(this);
2079 ir_to_mesa_src_reg r = this->result;
2080
2081 l = ir_to_mesa_dst_reg_from_src(current_function->return_reg);
2082
2083 for (i = 0; i < type_size(current_function->sig->return_type); i++) {
2084 ir_to_mesa_emit_op1(ir, OPCODE_MOV, l, r);
2085 l.index++;
2086 r.index++;
2087 }
2088 }
2089
2090 ir_to_mesa_emit_op0(ir, OPCODE_RET);
2091 }
2092
2093 void
2094 ir_to_mesa_visitor::visit(ir_discard *ir)
2095 {
2096 struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog;
2097
2098 assert(ir->condition == NULL); /* FINISHME */
2099
2100 ir_to_mesa_emit_op0(ir, OPCODE_KIL_NV);
2101 fp->UsesKill = GL_TRUE;
2102 }
2103
2104 void
2105 ir_to_mesa_visitor::visit(ir_if *ir)
2106 {
2107 ir_to_mesa_instruction *cond_inst, *if_inst, *else_inst = NULL;
2108 ir_to_mesa_instruction *prev_inst;
2109
2110 prev_inst = (ir_to_mesa_instruction *)this->instructions.get_tail();
2111
2112 ir->condition->accept(this);
2113 assert(this->result.file != PROGRAM_UNDEFINED);
2114
2115 if (this->options->EmitCondCodes) {
2116 cond_inst = (ir_to_mesa_instruction *)this->instructions.get_tail();
2117
2118 /* See if we actually generated any instruction for generating
2119 * the condition. If not, then cook up a move to a temp so we
2120 * have something to set cond_update on.
2121 */
2122 if (cond_inst == prev_inst) {
2123 ir_to_mesa_src_reg temp = get_temp(glsl_type::bool_type);
2124 cond_inst = ir_to_mesa_emit_op1(ir->condition, OPCODE_MOV,
2125 ir_to_mesa_dst_reg_from_src(temp),
2126 result);
2127 }
2128 cond_inst->cond_update = GL_TRUE;
2129
2130 if_inst = ir_to_mesa_emit_op0(ir->condition, OPCODE_IF);
2131 if_inst->dst_reg.cond_mask = COND_NE;
2132 } else {
2133 if_inst = ir_to_mesa_emit_op1(ir->condition,
2134 OPCODE_IF, ir_to_mesa_undef_dst,
2135 this->result);
2136 }
2137
2138 this->instructions.push_tail(if_inst);
2139
2140 visit_exec_list(&ir->then_instructions, this);
2141
2142 if (!ir->else_instructions.is_empty()) {
2143 else_inst = ir_to_mesa_emit_op0(ir->condition, OPCODE_ELSE);
2144 visit_exec_list(&ir->else_instructions, this);
2145 }
2146
2147 if_inst = ir_to_mesa_emit_op1(ir->condition, OPCODE_ENDIF,
2148 ir_to_mesa_undef_dst, ir_to_mesa_undef);
2149 }
2150
2151 ir_to_mesa_visitor::ir_to_mesa_visitor()
2152 {
2153 result.file = PROGRAM_UNDEFINED;
2154 next_temp = 1;
2155 next_signature_id = 1;
2156 current_function = NULL;
2157 mem_ctx = talloc_new(NULL);
2158 }
2159
2160 ir_to_mesa_visitor::~ir_to_mesa_visitor()
2161 {
2162 talloc_free(mem_ctx);
2163 }
2164
2165 static struct prog_src_register
2166 mesa_src_reg_from_ir_src_reg(ir_to_mesa_src_reg reg)
2167 {
2168 struct prog_src_register mesa_reg;
2169
2170 mesa_reg.File = reg.file;
2171 assert(reg.index < (1 << INST_INDEX_BITS) - 1);
2172 mesa_reg.Index = reg.index;
2173 mesa_reg.Swizzle = reg.swizzle;
2174 mesa_reg.RelAddr = reg.reladdr != NULL;
2175 mesa_reg.Negate = reg.negate;
2176 mesa_reg.Abs = 0;
2177 mesa_reg.HasIndex2 = GL_FALSE;
2178 mesa_reg.RelAddr2 = 0;
2179 mesa_reg.Index2 = 0;
2180
2181 return mesa_reg;
2182 }
2183
2184 static void
2185 set_branchtargets(ir_to_mesa_visitor *v,
2186 struct prog_instruction *mesa_instructions,
2187 int num_instructions)
2188 {
2189 int if_count = 0, loop_count = 0;
2190 int *if_stack, *loop_stack;
2191 int if_stack_pos = 0, loop_stack_pos = 0;
2192 int i, j;
2193
2194 for (i = 0; i < num_instructions; i++) {
2195 switch (mesa_instructions[i].Opcode) {
2196 case OPCODE_IF:
2197 if_count++;
2198 break;
2199 case OPCODE_BGNLOOP:
2200 loop_count++;
2201 break;
2202 case OPCODE_BRK:
2203 case OPCODE_CONT:
2204 mesa_instructions[i].BranchTarget = -1;
2205 break;
2206 default:
2207 break;
2208 }
2209 }
2210
2211 if_stack = talloc_zero_array(v->mem_ctx, int, if_count);
2212 loop_stack = talloc_zero_array(v->mem_ctx, int, loop_count);
2213
2214 for (i = 0; i < num_instructions; i++) {
2215 switch (mesa_instructions[i].Opcode) {
2216 case OPCODE_IF:
2217 if_stack[if_stack_pos] = i;
2218 if_stack_pos++;
2219 break;
2220 case OPCODE_ELSE:
2221 mesa_instructions[if_stack[if_stack_pos - 1]].BranchTarget = i;
2222 if_stack[if_stack_pos - 1] = i;
2223 break;
2224 case OPCODE_ENDIF:
2225 mesa_instructions[if_stack[if_stack_pos - 1]].BranchTarget = i;
2226 if_stack_pos--;
2227 break;
2228 case OPCODE_BGNLOOP:
2229 loop_stack[loop_stack_pos] = i;
2230 loop_stack_pos++;
2231 break;
2232 case OPCODE_ENDLOOP:
2233 loop_stack_pos--;
2234 /* Rewrite any breaks/conts at this nesting level (haven't
2235 * already had a BranchTarget assigned) to point to the end
2236 * of the loop.
2237 */
2238 for (j = loop_stack[loop_stack_pos]; j < i; j++) {
2239 if (mesa_instructions[j].Opcode == OPCODE_BRK ||
2240 mesa_instructions[j].Opcode == OPCODE_CONT) {
2241 if (mesa_instructions[j].BranchTarget == -1) {
2242 mesa_instructions[j].BranchTarget = i;
2243 }
2244 }
2245 }
2246 /* The loop ends point at each other. */
2247 mesa_instructions[i].BranchTarget = loop_stack[loop_stack_pos];
2248 mesa_instructions[loop_stack[loop_stack_pos]].BranchTarget = i;
2249 break;
2250 case OPCODE_CAL:
2251 foreach_iter(exec_list_iterator, iter, v->function_signatures) {
2252 function_entry *entry = (function_entry *)iter.get();
2253
2254 if (entry->sig_id == mesa_instructions[i].BranchTarget) {
2255 mesa_instructions[i].BranchTarget = entry->inst;
2256 break;
2257 }
2258 }
2259 break;
2260 default:
2261 break;
2262 }
2263 }
2264 }
2265
2266 static void
2267 print_program(struct prog_instruction *mesa_instructions,
2268 ir_instruction **mesa_instruction_annotation,
2269 int num_instructions)
2270 {
2271 ir_instruction *last_ir = NULL;
2272 int i;
2273 int indent = 0;
2274
2275 for (i = 0; i < num_instructions; i++) {
2276 struct prog_instruction *mesa_inst = mesa_instructions + i;
2277 ir_instruction *ir = mesa_instruction_annotation[i];
2278
2279 fprintf(stdout, "%3d: ", i);
2280
2281 if (last_ir != ir && ir) {
2282 int j;
2283
2284 for (j = 0; j < indent; j++) {
2285 fprintf(stdout, " ");
2286 }
2287 ir->print();
2288 printf("\n");
2289 last_ir = ir;
2290
2291 fprintf(stdout, " "); /* line number spacing. */
2292 }
2293
2294 indent = _mesa_fprint_instruction_opt(stdout, mesa_inst, indent,
2295 PROG_PRINT_DEBUG, NULL);
2296 }
2297 }
2298
2299 static void
2300 count_resources(struct gl_program *prog)
2301 {
2302 unsigned int i;
2303
2304 prog->SamplersUsed = 0;
2305
2306 for (i = 0; i < prog->NumInstructions; i++) {
2307 struct prog_instruction *inst = &prog->Instructions[i];
2308
2309 if (_mesa_is_tex_instruction(inst->Opcode)) {
2310 prog->SamplerTargets[inst->TexSrcUnit] =
2311 (gl_texture_index)inst->TexSrcTarget;
2312 prog->SamplersUsed |= 1 << inst->TexSrcUnit;
2313 if (inst->TexShadow) {
2314 prog->ShadowSamplers |= 1 << inst->TexSrcUnit;
2315 }
2316 }
2317 }
2318
2319 _mesa_update_shader_textures_used(prog);
2320 }
2321
2322 struct uniform_sort {
2323 struct gl_uniform *u;
2324 int pos;
2325 };
2326
2327 /* The shader_program->Uniforms list is almost sorted in increasing
2328 * uniform->{Frag,Vert}Pos locations, but not quite when there are
2329 * uniforms shared between targets. We need to add parameters in
2330 * increasing order for the targets.
2331 */
2332 static int
2333 sort_uniforms(const void *a, const void *b)
2334 {
2335 struct uniform_sort *u1 = (struct uniform_sort *)a;
2336 struct uniform_sort *u2 = (struct uniform_sort *)b;
2337
2338 return u1->pos - u2->pos;
2339 }
2340
2341 /* Add the uniforms to the parameters. The linker chose locations
2342 * in our parameters lists (which weren't created yet), which the
2343 * uniforms code will use to poke values into our parameters list
2344 * when uniforms are updated.
2345 */
2346 static void
2347 add_uniforms_to_parameters_list(struct gl_shader_program *shader_program,
2348 struct gl_shader *shader,
2349 struct gl_program *prog)
2350 {
2351 unsigned int i;
2352 unsigned int next_sampler = 0, num_uniforms = 0;
2353 struct uniform_sort *sorted_uniforms;
2354
2355 sorted_uniforms = talloc_array(NULL, struct uniform_sort,
2356 shader_program->Uniforms->NumUniforms);
2357
2358 for (i = 0; i < shader_program->Uniforms->NumUniforms; i++) {
2359 struct gl_uniform *uniform = shader_program->Uniforms->Uniforms + i;
2360 int parameter_index = -1;
2361
2362 switch (shader->Type) {
2363 case GL_VERTEX_SHADER:
2364 parameter_index = uniform->VertPos;
2365 break;
2366 case GL_FRAGMENT_SHADER:
2367 parameter_index = uniform->FragPos;
2368 break;
2369 case GL_GEOMETRY_SHADER:
2370 parameter_index = uniform->GeomPos;
2371 break;
2372 }
2373
2374 /* Only add uniforms used in our target. */
2375 if (parameter_index != -1) {
2376 sorted_uniforms[num_uniforms].pos = parameter_index;
2377 sorted_uniforms[num_uniforms].u = uniform;
2378 num_uniforms++;
2379 }
2380 }
2381
2382 qsort(sorted_uniforms, num_uniforms, sizeof(struct uniform_sort),
2383 sort_uniforms);
2384
2385 for (i = 0; i < num_uniforms; i++) {
2386 struct gl_uniform *uniform = sorted_uniforms[i].u;
2387 int parameter_index = sorted_uniforms[i].pos;
2388 const glsl_type *type = uniform->Type;
2389 unsigned int size;
2390
2391 if (type->is_vector() ||
2392 type->is_scalar()) {
2393 size = type->vector_elements;
2394 } else {
2395 size = type_size(type) * 4;
2396 }
2397
2398 gl_register_file file;
2399 if (type->is_sampler() ||
2400 (type->is_array() && type->fields.array->is_sampler())) {
2401 file = PROGRAM_SAMPLER;
2402 } else {
2403 file = PROGRAM_UNIFORM;
2404 }
2405
2406 GLint index = _mesa_lookup_parameter_index(prog->Parameters, -1,
2407 uniform->Name);
2408
2409 if (index < 0) {
2410 index = _mesa_add_parameter(prog->Parameters, file,
2411 uniform->Name, size, type->gl_type,
2412 NULL, NULL, 0x0);
2413
2414 /* Sampler uniform values are stored in prog->SamplerUnits,
2415 * and the entry in that array is selected by this index we
2416 * store in ParameterValues[].
2417 */
2418 if (file == PROGRAM_SAMPLER) {
2419 for (unsigned int j = 0; j < size / 4; j++)
2420 prog->Parameters->ParameterValues[index + j][0] = next_sampler++;
2421 }
2422
2423 /* The location chosen in the Parameters list here (returned
2424 * from _mesa_add_uniform) has to match what the linker chose.
2425 */
2426 if (index != parameter_index) {
2427 fail_link(shader_program, "Allocation of uniform `%s' to target "
2428 "failed (%d vs %d)\n",
2429 uniform->Name, index, parameter_index);
2430 }
2431 }
2432 }
2433
2434 talloc_free(sorted_uniforms);
2435 }
2436
2437 static void
2438 set_uniform_initializer(GLcontext *ctx, void *mem_ctx,
2439 struct gl_shader_program *shader_program,
2440 const char *name, const glsl_type *type,
2441 ir_constant *val)
2442 {
2443 if (type->is_record()) {
2444 ir_constant *field_constant;
2445
2446 field_constant = (ir_constant *)val->components.get_head();
2447
2448 for (unsigned int i = 0; i < type->length; i++) {
2449 const glsl_type *field_type = type->fields.structure[i].type;
2450 const char *field_name = talloc_asprintf(mem_ctx, "%s.%s", name,
2451 type->fields.structure[i].name);
2452 set_uniform_initializer(ctx, mem_ctx, shader_program, field_name,
2453 field_type, field_constant);
2454 field_constant = (ir_constant *)field_constant->next;
2455 }
2456 return;
2457 }
2458
2459 int loc = _mesa_get_uniform_location(ctx, shader_program, name);
2460
2461 if (loc == -1) {
2462 fail_link(shader_program,
2463 "Couldn't find uniform for initializer %s\n", name);
2464 return;
2465 }
2466
2467 for (unsigned int i = 0; i < (type->is_array() ? type->length : 1); i++) {
2468 ir_constant *element;
2469 const glsl_type *element_type;
2470 if (type->is_array()) {
2471 element = val->array_elements[i];
2472 element_type = type->fields.array;
2473 } else {
2474 element = val;
2475 element_type = type;
2476 }
2477
2478 void *values;
2479
2480 if (element_type->base_type == GLSL_TYPE_BOOL) {
2481 int *conv = talloc_array(mem_ctx, int, element_type->components());
2482 for (unsigned int j = 0; j < element_type->components(); j++) {
2483 conv[j] = element->value.b[j];
2484 }
2485 values = (void *)conv;
2486 element_type = glsl_type::get_instance(GLSL_TYPE_INT,
2487 element_type->vector_elements,
2488 1);
2489 } else {
2490 values = &element->value;
2491 }
2492
2493 if (element_type->is_matrix()) {
2494 _mesa_uniform_matrix(ctx, shader_program,
2495 element_type->matrix_columns,
2496 element_type->vector_elements,
2497 loc, 1, GL_FALSE, (GLfloat *)values);
2498 loc += element_type->matrix_columns;
2499 } else {
2500 _mesa_uniform(ctx, shader_program, loc, element_type->matrix_columns,
2501 values, element_type->gl_type);
2502 loc += type_size(element_type);
2503 }
2504 }
2505 }
2506
2507 static void
2508 set_uniform_initializers(GLcontext *ctx,
2509 struct gl_shader_program *shader_program)
2510 {
2511 void *mem_ctx = NULL;
2512
2513 for (unsigned int i = 0; i < shader_program->_NumLinkedShaders; i++) {
2514 struct gl_shader *shader = shader_program->_LinkedShaders[i];
2515 foreach_iter(exec_list_iterator, iter, *shader->ir) {
2516 ir_instruction *ir = (ir_instruction *)iter.get();
2517 ir_variable *var = ir->as_variable();
2518
2519 if (!var || var->mode != ir_var_uniform || !var->constant_value)
2520 continue;
2521
2522 if (!mem_ctx)
2523 mem_ctx = talloc_new(NULL);
2524
2525 set_uniform_initializer(ctx, mem_ctx, shader_program, var->name,
2526 var->type, var->constant_value);
2527 }
2528 }
2529
2530 talloc_free(mem_ctx);
2531 }
2532
2533 struct gl_program *
2534 get_mesa_program(GLcontext *ctx, struct gl_shader_program *shader_program,
2535 struct gl_shader *shader)
2536 {
2537 ir_to_mesa_visitor v;
2538 struct prog_instruction *mesa_instructions, *mesa_inst;
2539 ir_instruction **mesa_instruction_annotation;
2540 int i;
2541 struct gl_program *prog;
2542 GLenum target;
2543 const char *target_string;
2544 GLboolean progress;
2545 struct gl_shader_compiler_options *options =
2546 &ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(shader->Type)];
2547
2548 switch (shader->Type) {
2549 case GL_VERTEX_SHADER:
2550 target = GL_VERTEX_PROGRAM_ARB;
2551 target_string = "vertex";
2552 break;
2553 case GL_FRAGMENT_SHADER:
2554 target = GL_FRAGMENT_PROGRAM_ARB;
2555 target_string = "fragment";
2556 break;
2557 default:
2558 assert(!"should not be reached");
2559 return NULL;
2560 }
2561
2562 validate_ir_tree(shader->ir);
2563
2564 prog = ctx->Driver.NewProgram(ctx, target, shader_program->Name);
2565 if (!prog)
2566 return NULL;
2567 prog->Parameters = _mesa_new_parameter_list();
2568 prog->Varying = _mesa_new_parameter_list();
2569 prog->Attributes = _mesa_new_parameter_list();
2570 v.ctx = ctx;
2571 v.prog = prog;
2572 v.shader_program = shader_program;
2573 v.options = options;
2574
2575 add_uniforms_to_parameters_list(shader_program, shader, prog);
2576
2577 /* Emit Mesa IR for main(). */
2578 visit_exec_list(shader->ir, &v);
2579 v.ir_to_mesa_emit_op0(NULL, OPCODE_END);
2580
2581 /* Now emit bodies for any functions that were used. */
2582 do {
2583 progress = GL_FALSE;
2584
2585 foreach_iter(exec_list_iterator, iter, v.function_signatures) {
2586 function_entry *entry = (function_entry *)iter.get();
2587
2588 if (!entry->bgn_inst) {
2589 v.current_function = entry;
2590
2591 entry->bgn_inst = v.ir_to_mesa_emit_op0(NULL, OPCODE_BGNSUB);
2592 entry->bgn_inst->function = entry;
2593
2594 visit_exec_list(&entry->sig->body, &v);
2595
2596 ir_to_mesa_instruction *last;
2597 last = (ir_to_mesa_instruction *)v.instructions.get_tail();
2598 if (last->op != OPCODE_RET)
2599 v.ir_to_mesa_emit_op0(NULL, OPCODE_RET);
2600
2601 ir_to_mesa_instruction *end;
2602 end = v.ir_to_mesa_emit_op0(NULL, OPCODE_ENDSUB);
2603 end->function = entry;
2604
2605 progress = GL_TRUE;
2606 }
2607 }
2608 } while (progress);
2609
2610 prog->NumTemporaries = v.next_temp;
2611
2612 int num_instructions = 0;
2613 foreach_iter(exec_list_iterator, iter, v.instructions) {
2614 num_instructions++;
2615 }
2616
2617 mesa_instructions =
2618 (struct prog_instruction *)calloc(num_instructions,
2619 sizeof(*mesa_instructions));
2620 mesa_instruction_annotation = talloc_array(v.mem_ctx, ir_instruction *,
2621 num_instructions);
2622
2623 mesa_inst = mesa_instructions;
2624 i = 0;
2625 foreach_iter(exec_list_iterator, iter, v.instructions) {
2626 ir_to_mesa_instruction *inst = (ir_to_mesa_instruction *)iter.get();
2627
2628 mesa_inst->Opcode = inst->op;
2629 mesa_inst->CondUpdate = inst->cond_update;
2630 mesa_inst->DstReg.File = inst->dst_reg.file;
2631 mesa_inst->DstReg.Index = inst->dst_reg.index;
2632 mesa_inst->DstReg.CondMask = inst->dst_reg.cond_mask;
2633 mesa_inst->DstReg.WriteMask = inst->dst_reg.writemask;
2634 mesa_inst->DstReg.RelAddr = inst->dst_reg.reladdr != NULL;
2635 mesa_inst->SrcReg[0] = mesa_src_reg_from_ir_src_reg(inst->src_reg[0]);
2636 mesa_inst->SrcReg[1] = mesa_src_reg_from_ir_src_reg(inst->src_reg[1]);
2637 mesa_inst->SrcReg[2] = mesa_src_reg_from_ir_src_reg(inst->src_reg[2]);
2638 mesa_inst->TexSrcUnit = inst->sampler;
2639 mesa_inst->TexSrcTarget = inst->tex_target;
2640 mesa_inst->TexShadow = inst->tex_shadow;
2641 mesa_instruction_annotation[i] = inst->ir;
2642
2643 /* Set IndirectRegisterFiles. */
2644 if (mesa_inst->DstReg.RelAddr)
2645 prog->IndirectRegisterFiles |= 1 << mesa_inst->DstReg.File;
2646
2647 for (unsigned src = 0; src < 3; src++)
2648 if (mesa_inst->SrcReg[src].RelAddr)
2649 prog->IndirectRegisterFiles |= 1 << mesa_inst->SrcReg[src].File;
2650
2651 if (options->EmitNoIfs && mesa_inst->Opcode == OPCODE_IF) {
2652 fail_link(shader_program, "Couldn't flatten if statement\n");
2653 }
2654
2655 switch (mesa_inst->Opcode) {
2656 case OPCODE_BGNSUB:
2657 inst->function->inst = i;
2658 mesa_inst->Comment = strdup(inst->function->sig->function_name());
2659 break;
2660 case OPCODE_ENDSUB:
2661 mesa_inst->Comment = strdup(inst->function->sig->function_name());
2662 break;
2663 case OPCODE_CAL:
2664 mesa_inst->BranchTarget = inst->function->sig_id; /* rewritten later */
2665 break;
2666 case OPCODE_ARL:
2667 prog->NumAddressRegs = 1;
2668 break;
2669 default:
2670 break;
2671 }
2672
2673 mesa_inst++;
2674 i++;
2675 }
2676
2677 set_branchtargets(&v, mesa_instructions, num_instructions);
2678
2679 if (ctx->Shader.Flags & GLSL_DUMP) {
2680 printf("\n");
2681 printf("GLSL IR for linked %s program %d:\n", target_string,
2682 shader_program->Name);
2683 _mesa_print_ir(shader->ir, NULL);
2684 printf("\n");
2685 printf("\n");
2686 printf("Mesa IR for linked %s program %d:\n", target_string,
2687 shader_program->Name);
2688 print_program(mesa_instructions, mesa_instruction_annotation,
2689 num_instructions);
2690 }
2691
2692 prog->Instructions = mesa_instructions;
2693 prog->NumInstructions = num_instructions;
2694
2695 do_set_program_inouts(shader->ir, prog);
2696 count_resources(prog);
2697
2698 _mesa_reference_program(ctx, &shader->Program, prog);
2699
2700 if ((ctx->Shader.Flags & GLSL_NO_OPT) == 0) {
2701 _mesa_optimize_program(ctx, prog);
2702 }
2703
2704 return prog;
2705 }
2706
2707 extern "C" {
2708 GLboolean
2709 _mesa_ir_compile_shader(GLcontext *ctx, struct gl_shader *shader)
2710 {
2711 assert(shader->CompileStatus);
2712 (void) ctx;
2713
2714 return GL_TRUE;
2715 }
2716
2717 GLboolean
2718 _mesa_ir_link_shader(GLcontext *ctx, struct gl_shader_program *prog)
2719 {
2720 assert(prog->LinkStatus);
2721
2722 for (unsigned i = 0; i < prog->_NumLinkedShaders; i++) {
2723 bool progress;
2724 exec_list *ir = prog->_LinkedShaders[i]->ir;
2725 struct gl_shader_compiler_options *options =
2726 &ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(prog->_LinkedShaders[i]->Type)];
2727
2728 do {
2729 progress = false;
2730
2731 /* Lowering */
2732 do_mat_op_to_vec(ir);
2733 do_mod_to_fract(ir);
2734 do_div_to_mul_rcp(ir);
2735 do_explog_to_explog2(ir);
2736
2737 progress = do_lower_jumps(ir, true, true, options->EmitNoMainReturn, options->EmitNoCont, options->EmitNoLoops) || progress;
2738
2739 progress = do_common_optimization(ir, true, options->MaxUnrollIterations) || progress;
2740
2741 if (options->EmitNoIfs)
2742 progress = do_if_to_cond_assign(ir) || progress;
2743
2744 if (options->EmitNoNoise)
2745 progress = lower_noise(ir) || progress;
2746
2747 /* If there are forms of indirect addressing that the driver
2748 * cannot handle, perform the lowering pass.
2749 */
2750 if (options->EmitNoIndirectInput || options->EmitNoIndirectOutput
2751 || options->EmitNoIndirectTemp || options->EmitNoIndirectUniform)
2752 progress =
2753 lower_variable_index_to_cond_assign(ir,
2754 options->EmitNoIndirectInput,
2755 options->EmitNoIndirectOutput,
2756 options->EmitNoIndirectTemp,
2757 options->EmitNoIndirectUniform)
2758 || progress;
2759
2760 progress = do_vec_index_to_cond_assign(ir) || progress;
2761 } while (progress);
2762
2763 validate_ir_tree(ir);
2764 }
2765
2766 for (unsigned i = 0; i < prog->_NumLinkedShaders; i++) {
2767 struct gl_program *linked_prog;
2768 bool ok = true;
2769
2770 linked_prog = get_mesa_program(ctx, prog, prog->_LinkedShaders[i]);
2771
2772 switch (prog->_LinkedShaders[i]->Type) {
2773 case GL_VERTEX_SHADER:
2774 _mesa_reference_vertprog(ctx, &prog->VertexProgram,
2775 (struct gl_vertex_program *)linked_prog);
2776 ok = ctx->Driver.ProgramStringNotify(ctx, GL_VERTEX_PROGRAM_ARB,
2777 linked_prog);
2778 break;
2779 case GL_FRAGMENT_SHADER:
2780 _mesa_reference_fragprog(ctx, &prog->FragmentProgram,
2781 (struct gl_fragment_program *)linked_prog);
2782 ok = ctx->Driver.ProgramStringNotify(ctx, GL_FRAGMENT_PROGRAM_ARB,
2783 linked_prog);
2784 break;
2785 }
2786 if (!ok) {
2787 return GL_FALSE;
2788 }
2789 _mesa_reference_program(ctx, &linked_prog, NULL);
2790 }
2791
2792 return GL_TRUE;
2793 }
2794
2795 void
2796 _mesa_glsl_compile_shader(GLcontext *ctx, struct gl_shader *shader)
2797 {
2798 struct _mesa_glsl_parse_state *state =
2799 new(shader) _mesa_glsl_parse_state(ctx, shader->Type, shader);
2800
2801 const char *source = shader->Source;
2802 /* Check if the user called glCompileShader without first calling
2803 * glShaderSource. This should fail to compile, but not raise a GL_ERROR.
2804 */
2805 if (source == NULL) {
2806 shader->CompileStatus = GL_FALSE;
2807 return;
2808 }
2809
2810 state->error = preprocess(state, &source, &state->info_log,
2811 &ctx->Extensions, ctx->API);
2812
2813 if (ctx->Shader.Flags & GLSL_DUMP) {
2814 printf("GLSL source for shader %d:\n", shader->Name);
2815 printf("%s\n", shader->Source);
2816 }
2817
2818 if (!state->error) {
2819 _mesa_glsl_lexer_ctor(state, source);
2820 _mesa_glsl_parse(state);
2821 _mesa_glsl_lexer_dtor(state);
2822 }
2823
2824 talloc_free(shader->ir);
2825 shader->ir = new(shader) exec_list;
2826 if (!state->error && !state->translation_unit.is_empty())
2827 _mesa_ast_to_hir(shader->ir, state);
2828
2829 if (!state->error && !shader->ir->is_empty()) {
2830 validate_ir_tree(shader->ir);
2831
2832 /* Do some optimization at compile time to reduce shader IR size
2833 * and reduce later work if the same shader is linked multiple times
2834 */
2835 while (do_common_optimization(shader->ir, false, 32))
2836 ;
2837
2838 validate_ir_tree(shader->ir);
2839 }
2840
2841 shader->symbols = state->symbols;
2842
2843 shader->CompileStatus = !state->error;
2844 shader->InfoLog = state->info_log;
2845 shader->Version = state->language_version;
2846 memcpy(shader->builtins_to_link, state->builtins_to_link,
2847 sizeof(shader->builtins_to_link[0]) * state->num_builtins_to_link);
2848 shader->num_builtins_to_link = state->num_builtins_to_link;
2849
2850 if (ctx->Shader.Flags & GLSL_LOG) {
2851 _mesa_write_shader_to_file(shader);
2852 }
2853
2854 if (ctx->Shader.Flags & GLSL_DUMP) {
2855 if (shader->CompileStatus) {
2856 printf("GLSL IR for shader %d:\n", shader->Name);
2857 _mesa_print_ir(shader->ir, NULL);
2858 printf("\n\n");
2859 } else {
2860 printf("GLSL shader %d failed to compile.\n", shader->Name);
2861 }
2862 if (shader->InfoLog && shader->InfoLog[0] != 0) {
2863 printf("GLSL shader %d info log:\n", shader->Name);
2864 printf("%s\n", shader->InfoLog);
2865 }
2866 }
2867
2868 /* Retain any live IR, but trash the rest. */
2869 reparent_ir(shader->ir, shader->ir);
2870
2871 talloc_free(state);
2872
2873 if (shader->CompileStatus) {
2874 if (!ctx->Driver.CompileShader(ctx, shader))
2875 shader->CompileStatus = GL_FALSE;
2876 }
2877 }
2878
2879 void
2880 _mesa_glsl_link_shader(GLcontext *ctx, struct gl_shader_program *prog)
2881 {
2882 unsigned int i;
2883
2884 _mesa_clear_shader_program_data(ctx, prog);
2885
2886 prog->LinkStatus = GL_TRUE;
2887
2888 for (i = 0; i < prog->NumShaders; i++) {
2889 if (!prog->Shaders[i]->CompileStatus) {
2890 fail_link(prog, "linking with uncompiled shader");
2891 prog->LinkStatus = GL_FALSE;
2892 }
2893 }
2894
2895 prog->Varying = _mesa_new_parameter_list();
2896 _mesa_reference_vertprog(ctx, &prog->VertexProgram, NULL);
2897 _mesa_reference_fragprog(ctx, &prog->FragmentProgram, NULL);
2898
2899 if (prog->LinkStatus) {
2900 link_shaders(ctx, prog);
2901 }
2902
2903 if (prog->LinkStatus) {
2904 if (!ctx->Driver.LinkShader(ctx, prog)) {
2905 prog->LinkStatus = GL_FALSE;
2906 }
2907 }
2908
2909 set_uniform_initializers(ctx, prog);
2910
2911 if (ctx->Shader.Flags & GLSL_DUMP) {
2912 if (!prog->LinkStatus) {
2913 printf("GLSL shader program %d failed to link\n", prog->Name);
2914 }
2915
2916 if (prog->InfoLog && prog->InfoLog[0] != 0) {
2917 printf("GLSL shader program %d info log:\n", prog->Name);
2918 printf("%s\n", prog->InfoLog);
2919 }
2920 }
2921 }
2922
2923 } /* extern "C" */