projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
glsl2: Add EmitNoNoise flag, use it to remove noise opcodes
[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 bool array_indexed;
586 };
587
588 static struct statevar_element gl_DepthRange_elements[] = {
589 {"near", {STATE_DEPTH_RANGE, 0, 0}, SWIZZLE_XXXX},
590 {"far", {STATE_DEPTH_RANGE, 0, 0}, SWIZZLE_YYYY},
591 {"diff", {STATE_DEPTH_RANGE, 0, 0}, SWIZZLE_ZZZZ},
592 };
593
594 static struct statevar_element gl_ClipPlane_elements[] = {
595 {NULL, {STATE_CLIPPLANE, 0, 0}, SWIZZLE_XYZW}
596 };
597
598 static struct statevar_element gl_Point_elements[] = {
599 {"size", {STATE_POINT_SIZE}, SWIZZLE_XXXX},
600 {"sizeMin", {STATE_POINT_SIZE}, SWIZZLE_YYYY},
601 {"sizeMax", {STATE_POINT_SIZE}, SWIZZLE_ZZZZ},
602 {"fadeThresholdSize", {STATE_POINT_SIZE}, SWIZZLE_WWWW},
603 {"distanceConstantAttenuation", {STATE_POINT_ATTENUATION}, SWIZZLE_XXXX},
604 {"distanceLinearAttenuation", {STATE_POINT_ATTENUATION}, SWIZZLE_YYYY},
605 {"distanceQuadraticAttenuation", {STATE_POINT_ATTENUATION}, SWIZZLE_ZZZZ},
606 };
607
608 static struct statevar_element gl_FrontMaterial_elements[] = {
609 {"emission", {STATE_MATERIAL, 0, STATE_EMISSION}, SWIZZLE_XYZW},
610 {"ambient", {STATE_MATERIAL, 0, STATE_AMBIENT}, SWIZZLE_XYZW},
611 {"diffuse", {STATE_MATERIAL, 0, STATE_DIFFUSE}, SWIZZLE_XYZW},
612 {"specular", {STATE_MATERIAL, 0, STATE_SPECULAR}, SWIZZLE_XYZW},
613 {"shininess", {STATE_MATERIAL, 0, STATE_SHININESS}, SWIZZLE_XXXX},
614 };
615
616 static struct statevar_element gl_BackMaterial_elements[] = {
617 {"emission", {STATE_MATERIAL, 1, STATE_EMISSION}, SWIZZLE_XYZW},
618 {"ambient", {STATE_MATERIAL, 1, STATE_AMBIENT}, SWIZZLE_XYZW},
619 {"diffuse", {STATE_MATERIAL, 1, STATE_DIFFUSE}, SWIZZLE_XYZW},
620 {"specular", {STATE_MATERIAL, 1, STATE_SPECULAR}, SWIZZLE_XYZW},
621 {"shininess", {STATE_MATERIAL, 1, STATE_SHININESS}, SWIZZLE_XXXX},
622 };
623
624 static struct statevar_element gl_LightSource_elements[] = {
625 {"ambient", {STATE_LIGHT, 0, STATE_AMBIENT}, SWIZZLE_XYZW},
626 {"diffuse", {STATE_LIGHT, 0, STATE_DIFFUSE}, SWIZZLE_XYZW},
627 {"specular", {STATE_LIGHT, 0, STATE_SPECULAR}, SWIZZLE_XYZW},
628 {"position", {STATE_LIGHT, 0, STATE_POSITION}, SWIZZLE_XYZW},
629 {"halfVector", {STATE_LIGHT, 0, STATE_HALF_VECTOR}, SWIZZLE_XYZW},
630 {"spotDirection", {STATE_LIGHT, 0, STATE_SPOT_DIRECTION}, SWIZZLE_XYZW},
631 {"spotCosCutoff", {STATE_LIGHT, 0, STATE_SPOT_DIRECTION}, SWIZZLE_WWWW},
632 {"spotCutoff", {STATE_LIGHT, 0, STATE_SPOT_CUTOFF}, SWIZZLE_XXXX},
633 {"spotExponent", {STATE_LIGHT, 0, STATE_ATTENUATION}, SWIZZLE_WWWW},
634 {"constantAttenuation", {STATE_LIGHT, 0, STATE_ATTENUATION}, SWIZZLE_XXXX},
635 {"linearAttenuation", {STATE_LIGHT, 0, STATE_ATTENUATION}, SWIZZLE_YYYY},
636 {"quadraticAttenuation", {STATE_LIGHT, 0, STATE_ATTENUATION}, SWIZZLE_ZZZZ},
637 };
638
639 static struct statevar_element gl_LightModel_elements[] = {
640 {"ambient", {STATE_LIGHTMODEL_AMBIENT, 0}, SWIZZLE_XYZW},
641 };
642
643 static struct statevar_element gl_FrontLightModelProduct_elements[] = {
644 {"sceneColor", {STATE_LIGHTMODEL_SCENECOLOR, 0}, SWIZZLE_XYZW},
645 };
646
647 static struct statevar_element gl_BackLightModelProduct_elements[] = {
648 {"sceneColor", {STATE_LIGHTMODEL_SCENECOLOR, 1}, SWIZZLE_XYZW},
649 };
650
651 static struct statevar_element gl_FrontLightProduct_elements[] = {
652 {"ambient", {STATE_LIGHTPROD, 0, 0, STATE_AMBIENT}, SWIZZLE_XYZW},
653 {"diffuse", {STATE_LIGHTPROD, 0, 0, STATE_DIFFUSE}, SWIZZLE_XYZW},
654 {"specular", {STATE_LIGHTPROD, 0, 0, STATE_SPECULAR}, SWIZZLE_XYZW},
655 };
656
657 static struct statevar_element gl_BackLightProduct_elements[] = {
658 {"ambient", {STATE_LIGHTPROD, 0, 1, STATE_AMBIENT}, SWIZZLE_XYZW},
659 {"diffuse", {STATE_LIGHTPROD, 0, 1, STATE_DIFFUSE}, SWIZZLE_XYZW},
660 {"specular", {STATE_LIGHTPROD, 0, 1, STATE_SPECULAR}, SWIZZLE_XYZW},
661 };
662
663 static struct statevar_element gl_TextureEnvColor_elements[] = {
664 {NULL, {STATE_TEXENV_COLOR, 0}, SWIZZLE_XYZW},
665 };
666
667 static struct statevar_element gl_EyePlaneS_elements[] = {
668 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_EYE_S}, SWIZZLE_XYZW},
669 };
670
671 static struct statevar_element gl_EyePlaneT_elements[] = {
672 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_EYE_T}, SWIZZLE_XYZW},
673 };
674
675 static struct statevar_element gl_EyePlaneR_elements[] = {
676 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_EYE_R}, SWIZZLE_XYZW},
677 };
678
679 static struct statevar_element gl_EyePlaneQ_elements[] = {
680 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_EYE_Q}, SWIZZLE_XYZW},
681 };
682
683 static struct statevar_element gl_ObjectPlaneS_elements[] = {
684 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_OBJECT_S}, SWIZZLE_XYZW},
685 };
686
687 static struct statevar_element gl_ObjectPlaneT_elements[] = {
688 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_OBJECT_T}, SWIZZLE_XYZW},
689 };
690
691 static struct statevar_element gl_ObjectPlaneR_elements[] = {
692 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_OBJECT_R}, SWIZZLE_XYZW},
693 };
694
695 static struct statevar_element gl_ObjectPlaneQ_elements[] = {
696 {NULL, {STATE_TEXGEN, 0, STATE_TEXGEN_OBJECT_Q}, SWIZZLE_XYZW},
697 };
698
699 static struct statevar_element gl_Fog_elements[] = {
700 {"color", {STATE_FOG_COLOR}, SWIZZLE_XYZW},
701 {"density", {STATE_FOG_PARAMS}, SWIZZLE_XXXX},
702 {"start", {STATE_FOG_PARAMS}, SWIZZLE_YYYY},
703 {"end", {STATE_FOG_PARAMS}, SWIZZLE_ZZZZ},
704 {"scale", {STATE_FOG_PARAMS}, SWIZZLE_WWWW},
705 };
706
707 static struct statevar_element gl_NormalScale_elements[] = {
708 {NULL, {STATE_NORMAL_SCALE}, SWIZZLE_XXXX},
709 };
710
711 #define MATRIX(name, statevar, modifier) \
712 static struct statevar_element name ## _elements[] = { \
713 { NULL, { statevar, 0, 0, 0, modifier}, SWIZZLE_XYZW }, \
714 { NULL, { statevar, 0, 1, 1, modifier}, SWIZZLE_XYZW }, \
715 { NULL, { statevar, 0, 2, 2, modifier}, SWIZZLE_XYZW }, \
716 { NULL, { statevar, 0, 3, 3, modifier}, SWIZZLE_XYZW }, \
717 }
718
719 MATRIX(gl_ModelViewMatrix,
720 STATE_MODELVIEW_MATRIX, STATE_MATRIX_TRANSPOSE);
721 MATRIX(gl_ModelViewMatrixInverse,
722 STATE_MODELVIEW_MATRIX, STATE_MATRIX_INVTRANS);
723 MATRIX(gl_ModelViewMatrixTranspose,
724 STATE_MODELVIEW_MATRIX, 0);
725 MATRIX(gl_ModelViewMatrixInverseTranspose,
726 STATE_MODELVIEW_MATRIX, STATE_MATRIX_INVERSE);
727
728 MATRIX(gl_ProjectionMatrix,
729 STATE_PROJECTION_MATRIX, STATE_MATRIX_TRANSPOSE);
730 MATRIX(gl_ProjectionMatrixInverse,
731 STATE_PROJECTION_MATRIX, STATE_MATRIX_INVTRANS);
732 MATRIX(gl_ProjectionMatrixTranspose,
733 STATE_PROJECTION_MATRIX, 0);
734 MATRIX(gl_ProjectionMatrixInverseTranspose,
735 STATE_PROJECTION_MATRIX, STATE_MATRIX_INVERSE);
736
737 MATRIX(gl_ModelViewProjectionMatrix,
738 STATE_MVP_MATRIX, STATE_MATRIX_TRANSPOSE);
739 MATRIX(gl_ModelViewProjectionMatrixInverse,
740 STATE_MVP_MATRIX, STATE_MATRIX_INVTRANS);
741 MATRIX(gl_ModelViewProjectionMatrixTranspose,
742 STATE_MVP_MATRIX, 0);
743 MATRIX(gl_ModelViewProjectionMatrixInverseTranspose,
744 STATE_MVP_MATRIX, STATE_MATRIX_INVERSE);
745
746 MATRIX(gl_TextureMatrix,
747 STATE_TEXTURE_MATRIX, STATE_MATRIX_TRANSPOSE);
748 MATRIX(gl_TextureMatrixInverse,
749 STATE_TEXTURE_MATRIX, STATE_MATRIX_INVTRANS);
750 MATRIX(gl_TextureMatrixTranspose,
751 STATE_TEXTURE_MATRIX, 0);
752 MATRIX(gl_TextureMatrixInverseTranspose,
753 STATE_TEXTURE_MATRIX, STATE_MATRIX_INVERSE);
754
755 static struct statevar_element gl_NormalMatrix_elements[] = {
756 { NULL, { STATE_MODELVIEW_MATRIX, 0, 0, 0, STATE_MATRIX_INVERSE},
757 SWIZZLE_XYZW },
758 { NULL, { STATE_MODELVIEW_MATRIX, 0, 1, 1, STATE_MATRIX_INVERSE},
759 SWIZZLE_XYZW },
760 { NULL, { STATE_MODELVIEW_MATRIX, 0, 2, 2, STATE_MATRIX_INVERSE},
761 SWIZZLE_XYZW },
762 };
763
764 #undef MATRIX
765
766 #define STATEVAR(name) {#name, name ## _elements, Elements(name ## _elements)}
767
768 static const struct statevar {
769 const char *name;
770 struct statevar_element *elements;
771 unsigned int num_elements;
772 } statevars[] = {
773 STATEVAR(gl_DepthRange),
774 STATEVAR(gl_ClipPlane),
775 STATEVAR(gl_Point),
776 STATEVAR(gl_FrontMaterial),
777 STATEVAR(gl_BackMaterial),
778 STATEVAR(gl_LightSource),
779 STATEVAR(gl_LightModel),
780 STATEVAR(gl_FrontLightModelProduct),
781 STATEVAR(gl_BackLightModelProduct),
782 STATEVAR(gl_FrontLightProduct),
783 STATEVAR(gl_BackLightProduct),
784 STATEVAR(gl_TextureEnvColor),
785 STATEVAR(gl_EyePlaneS),
786 STATEVAR(gl_EyePlaneT),
787 STATEVAR(gl_EyePlaneR),
788 STATEVAR(gl_EyePlaneQ),
789 STATEVAR(gl_ObjectPlaneS),
790 STATEVAR(gl_ObjectPlaneT),
791 STATEVAR(gl_ObjectPlaneR),
792 STATEVAR(gl_ObjectPlaneQ),
793 STATEVAR(gl_Fog),
794
795 STATEVAR(gl_ModelViewMatrix),
796 STATEVAR(gl_ModelViewMatrixInverse),
797 STATEVAR(gl_ModelViewMatrixTranspose),
798 STATEVAR(gl_ModelViewMatrixInverseTranspose),
799
800 STATEVAR(gl_ProjectionMatrix),
801 STATEVAR(gl_ProjectionMatrixInverse),
802 STATEVAR(gl_ProjectionMatrixTranspose),
803 STATEVAR(gl_ProjectionMatrixInverseTranspose),
804
805 STATEVAR(gl_ModelViewProjectionMatrix),
806 STATEVAR(gl_ModelViewProjectionMatrixInverse),
807 STATEVAR(gl_ModelViewProjectionMatrixTranspose),
808 STATEVAR(gl_ModelViewProjectionMatrixInverseTranspose),
809
810 STATEVAR(gl_TextureMatrix),
811 STATEVAR(gl_TextureMatrixInverse),
812 STATEVAR(gl_TextureMatrixTranspose),
813 STATEVAR(gl_TextureMatrixInverseTranspose),
814
815 STATEVAR(gl_NormalMatrix),
816 STATEVAR(gl_NormalScale),
817 };
818
819 void
820 ir_to_mesa_visitor::visit(ir_variable *ir)
821 {
822 if (strcmp(ir->name, "gl_FragCoord") == 0) {
823 struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog;
824
825 fp->OriginUpperLeft = ir->origin_upper_left;
826 fp->PixelCenterInteger = ir->pixel_center_integer;
827 }
828
829 if (ir->mode == ir_var_uniform && strncmp(ir->name, "gl_", 3) == 0) {
830 unsigned int i;
831
832 for (i = 0; i < Elements(statevars); i++) {
833 if (strcmp(ir->name, statevars[i].name) == 0)
834 break;
835 }
836
837 if (i == Elements(statevars)) {
838 fail_link(this->shader_program,
839 "Failed to find builtin uniform `%s'\n", ir->name);
840 return;
841 }
842
843 const struct statevar *statevar = &statevars[i];
844
845 int array_count;
846 if (ir->type->is_array()) {
847 array_count = ir->type->length;
848 } else {
849 array_count = 1;
850 }
851
852 /* Check if this statevar's setup in the STATE file exactly
853 * matches how we'll want to reference it as a
854 * struct/array/whatever. If not, then we need to move it into
855 * temporary storage and hope that it'll get copy-propagated
856 * out.
857 */
858 for (i = 0; i < statevar->num_elements; i++) {
859 if (statevar->elements[i].swizzle != SWIZZLE_XYZW) {
860 break;
861 }
862 }
863
864 struct variable_storage *storage;
865 ir_to_mesa_dst_reg dst;
866 if (i == statevar->num_elements) {
867 /* We'll set the index later. */
868 storage = new(mem_ctx) variable_storage(ir, PROGRAM_STATE_VAR, -1);
869 this->variables.push_tail(storage);
870
871 dst = ir_to_mesa_undef_dst;
872 } else {
873 storage = new(mem_ctx) variable_storage(ir, PROGRAM_TEMPORARY,
874 this->next_temp);
875 this->variables.push_tail(storage);
876 this->next_temp += type_size(ir->type);
877
878 dst = ir_to_mesa_dst_reg_from_src(ir_to_mesa_src_reg(PROGRAM_TEMPORARY,
879 storage->index,
880 NULL));
881 }
882
883
884 for (int a = 0; a < array_count; a++) {
885 for (unsigned int i = 0; i < statevar->num_elements; i++) {
886 struct statevar_element *element = &statevar->elements[i];
887 int tokens[STATE_LENGTH];
888
889 memcpy(tokens, element->tokens, sizeof(element->tokens));
890 if (ir->type->is_array()) {
891 tokens[1] = a;
892 }
893
894 int index = _mesa_add_state_reference(this->prog->Parameters,
895 (gl_state_index *)tokens);
896
897 if (storage->file == PROGRAM_STATE_VAR) {
898 if (storage->index == -1) {
899 storage->index = index;
900 } else {
901 assert(index == ((int)storage->index +
902 a * statevar->num_elements + i));
903 }
904 } else {
905 ir_to_mesa_src_reg src(PROGRAM_STATE_VAR, index, NULL);
906 src.swizzle = element->swizzle;
907 ir_to_mesa_emit_op1(ir, OPCODE_MOV, dst, src);
908 /* even a float takes up a whole vec4 reg in a struct/array. */
909 dst.index++;
910 }
911 }
912 }
913 if (storage->file == PROGRAM_TEMPORARY &&
914 dst.index != storage->index + type_size(ir->type)) {
915 fail_link(this->shader_program,
916 "failed to load builtin uniform `%s' (%d/%d regs loaded)\n",
917 ir->name, dst.index - storage->index,
918 type_size(ir->type));
919 }
920 }
921 }
922
923 void
924 ir_to_mesa_visitor::visit(ir_loop *ir)
925 {
926 ir_dereference_variable *counter = NULL;
927
928 if (ir->counter != NULL)
929 counter = new(ir) ir_dereference_variable(ir->counter);
930
931 if (ir->from != NULL) {
932 assert(ir->counter != NULL);
933
934 ir_assignment *a = new(ir) ir_assignment(counter, ir->from, NULL);
935
936 a->accept(this);
937 delete a;
938 }
939
940 ir_to_mesa_emit_op0(NULL, OPCODE_BGNLOOP);
941
942 if (ir->to) {
943 ir_expression *e =
944 new(ir) ir_expression(ir->cmp, glsl_type::bool_type,
945 counter, ir->to);
946 ir_if *if_stmt = new(ir) ir_if(e);
947
948 ir_loop_jump *brk = new(ir) ir_loop_jump(ir_loop_jump::jump_break);
949
950 if_stmt->then_instructions.push_tail(brk);
951
952 if_stmt->accept(this);
953
954 delete if_stmt;
955 delete e;
956 delete brk;
957 }
958
959 visit_exec_list(&ir->body_instructions, this);
960
961 if (ir->increment) {
962 ir_expression *e =
963 new(ir) ir_expression(ir_binop_add, counter->type,
964 counter, ir->increment);
965
966 ir_assignment *a = new(ir) ir_assignment(counter, e, NULL);
967
968 a->accept(this);
969 delete a;
970 delete e;
971 }
972
973 ir_to_mesa_emit_op0(NULL, OPCODE_ENDLOOP);
974 }
975
976 void
977 ir_to_mesa_visitor::visit(ir_loop_jump *ir)
978 {
979 switch (ir->mode) {
980 case ir_loop_jump::jump_break:
981 ir_to_mesa_emit_op0(NULL, OPCODE_BRK);
982 break;
983 case ir_loop_jump::jump_continue:
984 ir_to_mesa_emit_op0(NULL, OPCODE_CONT);
985 break;
986 }
987 }
988
989
990 void
991 ir_to_mesa_visitor::visit(ir_function_signature *ir)
992 {
993 assert(0);
994 (void)ir;
995 }
996
997 void
998 ir_to_mesa_visitor::visit(ir_function *ir)
999 {
1000 /* Ignore function bodies other than main() -- we shouldn't see calls to
1001 * them since they should all be inlined before we get to ir_to_mesa.
1002 */
1003 if (strcmp(ir->name, "main") == 0) {
1004 const ir_function_signature *sig;
1005 exec_list empty;
1006
1007 sig = ir->matching_signature(&empty);
1008
1009 assert(sig);
1010
1011 foreach_iter(exec_list_iterator, iter, sig->body) {
1012 ir_instruction *ir = (ir_instruction *)iter.get();
1013
1014 ir->accept(this);
1015 }
1016 }
1017 }
1018
1019 GLboolean
1020 ir_to_mesa_visitor::try_emit_mad(ir_expression *ir, int mul_operand)
1021 {
1022 int nonmul_operand = 1 - mul_operand;
1023 ir_to_mesa_src_reg a, b, c;
1024
1025 ir_expression *expr = ir->operands[mul_operand]->as_expression();
1026 if (!expr || expr->operation != ir_binop_mul)
1027 return false;
1028
1029 expr->operands[0]->accept(this);
1030 a = this->result;
1031 expr->operands[1]->accept(this);
1032 b = this->result;
1033 ir->operands[nonmul_operand]->accept(this);
1034 c = this->result;
1035
1036 this->result = get_temp(ir->type);
1037 ir_to_mesa_emit_op3(ir, OPCODE_MAD,
1038 ir_to_mesa_dst_reg_from_src(this->result), a, b, c);
1039
1040 return true;
1041 }
1042
1043 void
1044 ir_to_mesa_visitor::reladdr_to_temp(ir_instruction *ir,
1045 ir_to_mesa_src_reg *reg, int *num_reladdr)
1046 {
1047 if (!reg->reladdr)
1048 return;
1049
1050 ir_to_mesa_emit_op1(ir, OPCODE_ARL, ir_to_mesa_address_reg, *reg->reladdr);
1051
1052 if (*num_reladdr != 1) {
1053 ir_to_mesa_src_reg temp = get_temp(glsl_type::vec4_type);
1054
1055 ir_to_mesa_emit_op1(ir, OPCODE_MOV,
1056 ir_to_mesa_dst_reg_from_src(temp), *reg);
1057 *reg = temp;
1058 }
1059
1060 (*num_reladdr)--;
1061 }
1062
1063 void
1064 ir_to_mesa_visitor::visit(ir_expression *ir)
1065 {
1066 unsigned int operand;
1067 struct ir_to_mesa_src_reg op[2];
1068 struct ir_to_mesa_src_reg result_src;
1069 struct ir_to_mesa_dst_reg result_dst;
1070 const glsl_type *vec4_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, 4, 1);
1071 const glsl_type *vec3_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, 3, 1);
1072 const glsl_type *vec2_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, 2, 1);
1073
1074 /* Quick peephole: Emit OPCODE_MAD(a, b, c) instead of ADD(MUL(a, b), c)
1075 */
1076 if (ir->operation == ir_binop_add) {
1077 if (try_emit_mad(ir, 1))
1078 return;
1079 if (try_emit_mad(ir, 0))
1080 return;
1081 }
1082
1083 for (operand = 0; operand < ir->get_num_operands(); operand++) {
1084 this->result.file = PROGRAM_UNDEFINED;
1085 ir->operands[operand]->accept(this);
1086 if (this->result.file == PROGRAM_UNDEFINED) {
1087 ir_print_visitor v;
1088 printf("Failed to get tree for expression operand:\n");
1089 ir->operands[operand]->accept(&v);
1090 exit(1);
1091 }
1092 op[operand] = this->result;
1093
1094 /* Matrix expression operands should have been broken down to vector
1095 * operations already.
1096 */
1097 assert(!ir->operands[operand]->type->is_matrix());
1098 }
1099
1100 this->result.file = PROGRAM_UNDEFINED;
1101
1102 /* Storage for our result. Ideally for an assignment we'd be using
1103 * the actual storage for the result here, instead.
1104 */
1105 result_src = get_temp(ir->type);
1106 /* convenience for the emit functions below. */
1107 result_dst = ir_to_mesa_dst_reg_from_src(result_src);
1108 /* Limit writes to the channels that will be used by result_src later.
1109 * This does limit this temp's use as a temporary for multi-instruction
1110 * sequences.
1111 */
1112 result_dst.writemask = (1 << ir->type->vector_elements) - 1;
1113
1114 switch (ir->operation) {
1115 case ir_unop_logic_not:
1116 ir_to_mesa_emit_op2(ir, OPCODE_SEQ, result_dst,
1117 op[0], src_reg_for_float(0.0));
1118 break;
1119 case ir_unop_neg:
1120 op[0].negate = ~op[0].negate;
1121 result_src = op[0];
1122 break;
1123 case ir_unop_abs:
1124 ir_to_mesa_emit_op1(ir, OPCODE_ABS, result_dst, op[0]);
1125 break;
1126 case ir_unop_sign:
1127 ir_to_mesa_emit_op1(ir, OPCODE_SSG, result_dst, op[0]);
1128 break;
1129 case ir_unop_rcp:
1130 ir_to_mesa_emit_scalar_op1(ir, OPCODE_RCP, result_dst, op[0]);
1131 break;
1132
1133 case ir_unop_exp2:
1134 ir_to_mesa_emit_scalar_op1(ir, OPCODE_EX2, result_dst, op[0]);
1135 break;
1136 case ir_unop_exp:
1137 case ir_unop_log:
1138 assert(!"not reached: should be handled by ir_explog_to_explog2");
1139 break;
1140 case ir_unop_log2:
1141 ir_to_mesa_emit_scalar_op1(ir, OPCODE_LG2, result_dst, op[0]);
1142 break;
1143 case ir_unop_sin:
1144 ir_to_mesa_emit_scalar_op1(ir, OPCODE_SIN, result_dst, op[0]);
1145 break;
1146 case ir_unop_cos:
1147 ir_to_mesa_emit_scalar_op1(ir, OPCODE_COS, result_dst, op[0]);
1148 break;
1149
1150 case ir_unop_dFdx:
1151 ir_to_mesa_emit_op1(ir, OPCODE_DDX, result_dst, op[0]);
1152 break;
1153 case ir_unop_dFdy:
1154 ir_to_mesa_emit_op1(ir, OPCODE_DDY, result_dst, op[0]);
1155 break;
1156
1157 case ir_unop_noise: {
1158 const enum prog_opcode opcode =
1159 prog_opcode(OPCODE_NOISE1
1160 + (ir->operands[0]->type->vector_elements) - 1);
1161 assert((opcode >= OPCODE_NOISE1) && (opcode <= OPCODE_NOISE4));
1162
1163 ir_to_mesa_emit_op1(ir, opcode, result_dst, op[0]);
1164 break;
1165 }
1166
1167 case ir_binop_add:
1168 ir_to_mesa_emit_op2(ir, OPCODE_ADD, result_dst, op[0], op[1]);
1169 break;
1170 case ir_binop_sub:
1171 ir_to_mesa_emit_op2(ir, OPCODE_SUB, result_dst, op[0], op[1]);
1172 break;
1173
1174 case ir_binop_mul:
1175 ir_to_mesa_emit_op2(ir, OPCODE_MUL, result_dst, op[0], op[1]);
1176 break;
1177 case ir_binop_div:
1178 assert(!"not reached: should be handled by ir_div_to_mul_rcp");
1179 case ir_binop_mod:
1180 assert(!"ir_binop_mod should have been converted to b * fract(a/b)");
1181 break;
1182
1183 case ir_binop_less:
1184 ir_to_mesa_emit_op2(ir, OPCODE_SLT, result_dst, op[0], op[1]);
1185 break;
1186 case ir_binop_greater:
1187 ir_to_mesa_emit_op2(ir, OPCODE_SGT, result_dst, op[0], op[1]);
1188 break;
1189 case ir_binop_lequal:
1190 ir_to_mesa_emit_op2(ir, OPCODE_SLE, result_dst, op[0], op[1]);
1191 break;
1192 case ir_binop_gequal:
1193 ir_to_mesa_emit_op2(ir, OPCODE_SGE, result_dst, op[0], op[1]);
1194 break;
1195 case ir_binop_equal:
1196 /* "==" operator producing a scalar boolean. */
1197 if (ir->operands[0]->type->is_vector() ||
1198 ir->operands[1]->type->is_vector()) {
1199 ir_to_mesa_src_reg temp = get_temp(glsl_type::vec4_type);
1200 ir_to_mesa_emit_op2(ir, OPCODE_SNE,
1201 ir_to_mesa_dst_reg_from_src(temp), op[0], op[1]);
1202 ir_to_mesa_emit_op2(ir, OPCODE_DP4, result_dst, temp, temp);
1203 ir_to_mesa_emit_op2(ir, OPCODE_SEQ,
1204 result_dst, result_src, src_reg_for_float(0.0));
1205 } else {
1206 ir_to_mesa_emit_op2(ir, OPCODE_SEQ, result_dst, op[0], op[1]);
1207 }
1208 break;
1209 case ir_binop_nequal:
1210 /* "!=" operator producing a scalar boolean. */
1211 if (ir->operands[0]->type->is_vector() ||
1212 ir->operands[1]->type->is_vector()) {
1213 ir_to_mesa_src_reg temp = get_temp(glsl_type::vec4_type);
1214 ir_to_mesa_emit_op2(ir, OPCODE_SNE,
1215 ir_to_mesa_dst_reg_from_src(temp), op[0], op[1]);
1216 ir_to_mesa_emit_op2(ir, OPCODE_DP4, result_dst, temp, temp);
1217 ir_to_mesa_emit_op2(ir, OPCODE_SNE,
1218 result_dst, result_src, src_reg_for_float(0.0));
1219 } else {
1220 ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst, op[0], op[1]);
1221 }
1222 break;
1223
1224 case ir_unop_any:
1225 switch (ir->operands[0]->type->vector_elements) {
1226 case 4:
1227 ir_to_mesa_emit_op2(ir, OPCODE_DP4, result_dst, op[0], op[0]);
1228 break;
1229 case 3:
1230 ir_to_mesa_emit_op2(ir, OPCODE_DP3, result_dst, op[0], op[0]);
1231 break;
1232 case 2:
1233 ir_to_mesa_emit_op2(ir, OPCODE_DP2, result_dst, op[0], op[0]);
1234 break;
1235 default:
1236 assert(!"unreached: ir_unop_any of non-bvec");
1237 break;
1238 }
1239 ir_to_mesa_emit_op2(ir, OPCODE_SNE,
1240 result_dst, result_src, src_reg_for_float(0.0));
1241 break;
1242
1243 case ir_binop_logic_xor:
1244 ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst, op[0], op[1]);
1245 break;
1246
1247 case ir_binop_logic_or:
1248 /* This could be a saturated add and skip the SNE. */
1249 ir_to_mesa_emit_op2(ir, OPCODE_ADD,
1250 result_dst,
1251 op[0], op[1]);
1252
1253 ir_to_mesa_emit_op2(ir, OPCODE_SNE,
1254 result_dst,
1255 result_src, src_reg_for_float(0.0));
1256 break;
1257
1258 case ir_binop_logic_and:
1259 /* the bool args are stored as float 0.0 or 1.0, so "mul" gives us "and". */
1260 ir_to_mesa_emit_op2(ir, OPCODE_MUL,
1261 result_dst,
1262 op[0], op[1]);
1263 break;
1264
1265 case ir_binop_dot:
1266 if (ir->operands[0]->type == vec4_type) {
1267 assert(ir->operands[1]->type == vec4_type);
1268 ir_to_mesa_emit_op2(ir, OPCODE_DP4,
1269 result_dst,
1270 op[0], op[1]);
1271 } else if (ir->operands[0]->type == vec3_type) {
1272 assert(ir->operands[1]->type == vec3_type);
1273 ir_to_mesa_emit_op2(ir, OPCODE_DP3,
1274 result_dst,
1275 op[0], op[1]);
1276 } else if (ir->operands[0]->type == vec2_type) {
1277 assert(ir->operands[1]->type == vec2_type);
1278 ir_to_mesa_emit_op2(ir, OPCODE_DP2,
1279 result_dst,
1280 op[0], op[1]);
1281 }
1282 break;
1283
1284 case ir_binop_cross:
1285 ir_to_mesa_emit_op2(ir, OPCODE_XPD, result_dst, op[0], op[1]);
1286 break;
1287
1288 case ir_unop_sqrt:
1289 /* sqrt(x) = x * rsq(x). */
1290 ir_to_mesa_emit_scalar_op1(ir, OPCODE_RSQ, result_dst, op[0]);
1291 ir_to_mesa_emit_op2(ir, OPCODE_MUL, result_dst, result_src, op[0]);
1292 /* For incoming channels <= 0, set the result to 0. */
1293 op[0].negate = ~op[0].negate;
1294 ir_to_mesa_emit_op3(ir, OPCODE_CMP, result_dst,
1295 op[0], result_src, src_reg_for_float(0.0));
1296 break;
1297 case ir_unop_rsq:
1298 ir_to_mesa_emit_scalar_op1(ir, OPCODE_RSQ, result_dst, op[0]);
1299 break;
1300 case ir_unop_i2f:
1301 case ir_unop_b2f:
1302 case ir_unop_b2i:
1303 /* Mesa IR lacks types, ints are stored as truncated floats. */
1304 result_src = op[0];
1305 break;
1306 case ir_unop_f2i:
1307 ir_to_mesa_emit_op1(ir, OPCODE_TRUNC, result_dst, op[0]);
1308 break;
1309 case ir_unop_f2b:
1310 case ir_unop_i2b:
1311 ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst,
1312 op[0], src_reg_for_float(0.0));
1313 break;
1314 case ir_unop_trunc:
1315 ir_to_mesa_emit_op1(ir, OPCODE_TRUNC, result_dst, op[0]);
1316 break;
1317 case ir_unop_ceil:
1318 op[0].negate = ~op[0].negate;
1319 ir_to_mesa_emit_op1(ir, OPCODE_FLR, result_dst, op[0]);
1320 result_src.negate = ~result_src.negate;
1321 break;
1322 case ir_unop_floor:
1323 ir_to_mesa_emit_op1(ir, OPCODE_FLR, result_dst, op[0]);
1324 break;
1325 case ir_unop_fract:
1326 ir_to_mesa_emit_op1(ir, OPCODE_FRC, result_dst, op[0]);
1327 break;
1328
1329 case ir_binop_min:
1330 ir_to_mesa_emit_op2(ir, OPCODE_MIN, result_dst, op[0], op[1]);
1331 break;
1332 case ir_binop_max:
1333 ir_to_mesa_emit_op2(ir, OPCODE_MAX, result_dst, op[0], op[1]);
1334 break;
1335 case ir_binop_pow:
1336 ir_to_mesa_emit_scalar_op2(ir, OPCODE_POW, result_dst, op[0], op[1]);
1337 break;
1338
1339 case ir_unop_bit_not:
1340 case ir_unop_u2f:
1341 case ir_binop_lshift:
1342 case ir_binop_rshift:
1343 case ir_binop_bit_and:
1344 case ir_binop_bit_xor:
1345 case ir_binop_bit_or:
1346 assert(!"GLSL 1.30 features unsupported");
1347 break;
1348 }
1349
1350 this->result = result_src;
1351 }
1352
1353
1354 void
1355 ir_to_mesa_visitor::visit(ir_swizzle *ir)
1356 {
1357 ir_to_mesa_src_reg src_reg;
1358 int i;
1359 int swizzle[4];
1360
1361 /* Note that this is only swizzles in expressions, not those on the left
1362 * hand side of an assignment, which do write masking. See ir_assignment
1363 * for that.
1364 */
1365
1366 ir->val->accept(this);
1367 src_reg = this->result;
1368 assert(src_reg.file != PROGRAM_UNDEFINED);
1369
1370 for (i = 0; i < 4; i++) {
1371 if (i < ir->type->vector_elements) {
1372 switch (i) {
1373 case 0:
1374 swizzle[i] = GET_SWZ(src_reg.swizzle, ir->mask.x);
1375 break;
1376 case 1:
1377 swizzle[i] = GET_SWZ(src_reg.swizzle, ir->mask.y);
1378 break;
1379 case 2:
1380 swizzle[i] = GET_SWZ(src_reg.swizzle, ir->mask.z);
1381 break;
1382 case 3:
1383 swizzle[i] = GET_SWZ(src_reg.swizzle, ir->mask.w);
1384 break;
1385 }
1386 } else {
1387 /* If the type is smaller than a vec4, replicate the last
1388 * channel out.
1389 */
1390 swizzle[i] = swizzle[ir->type->vector_elements - 1];
1391 }
1392 }
1393
1394 src_reg.swizzle = MAKE_SWIZZLE4(swizzle[0],
1395 swizzle[1],
1396 swizzle[2],
1397 swizzle[3]);
1398
1399 this->result = src_reg;
1400 }
1401
1402 void
1403 ir_to_mesa_visitor::visit(ir_dereference_variable *ir)
1404 {
1405 variable_storage *entry = find_variable_storage(ir->var);
1406
1407 if (!entry) {
1408 switch (ir->var->mode) {
1409 case ir_var_uniform:
1410 entry = new(mem_ctx) variable_storage(ir->var, PROGRAM_UNIFORM,
1411 ir->var->location);
1412 this->variables.push_tail(entry);
1413 break;
1414 case ir_var_in:
1415 case ir_var_out:
1416 case ir_var_inout:
1417 /* The linker assigns locations for varyings and attributes,
1418 * including deprecated builtins (like gl_Color), user-assign
1419 * generic attributes (glBindVertexLocation), and
1420 * user-defined varyings.
1421 *
1422 * FINISHME: We would hit this path for function arguments. Fix!
1423 */
1424 assert(ir->var->location != -1);
1425 if (ir->var->mode == ir_var_in ||
1426 ir->var->mode == ir_var_inout) {
1427 entry = new(mem_ctx) variable_storage(ir->var,
1428 PROGRAM_INPUT,
1429 ir->var->location);
1430
1431 if (this->prog->Target == GL_VERTEX_PROGRAM_ARB &&
1432 ir->var->location >= VERT_ATTRIB_GENERIC0) {
1433 _mesa_add_attribute(prog->Attributes,
1434 ir->var->name,
1435 _mesa_sizeof_glsl_type(ir->var->type->gl_type),
1436 ir->var->type->gl_type,
1437 ir->var->location - VERT_ATTRIB_GENERIC0);
1438 }
1439 } else {
1440 entry = new(mem_ctx) variable_storage(ir->var,
1441 PROGRAM_OUTPUT,
1442 ir->var->location);
1443 }
1444
1445 break;
1446 case ir_var_auto:
1447 case ir_var_temporary:
1448 entry = new(mem_ctx) variable_storage(ir->var, PROGRAM_TEMPORARY,
1449 this->next_temp);
1450 this->variables.push_tail(entry);
1451
1452 next_temp += type_size(ir->var->type);
1453 break;
1454 }
1455
1456 if (!entry) {
1457 printf("Failed to make storage for %s\n", ir->var->name);
1458 exit(1);
1459 }
1460 }
1461
1462 this->result = ir_to_mesa_src_reg(entry->file, entry->index, ir->var->type);
1463 }
1464
1465 void
1466 ir_to_mesa_visitor::visit(ir_dereference_array *ir)
1467 {
1468 ir_constant *index;
1469 ir_to_mesa_src_reg src_reg;
1470 int element_size = type_size(ir->type);
1471
1472 index = ir->array_index->constant_expression_value();
1473
1474 ir->array->accept(this);
1475 src_reg = this->result;
1476
1477 if (index) {
1478 src_reg.index += index->value.i[0] * element_size;
1479 } else {
1480 ir_to_mesa_src_reg array_base = this->result;
1481 /* Variable index array dereference. It eats the "vec4" of the
1482 * base of the array and an index that offsets the Mesa register
1483 * index.
1484 */
1485 ir->array_index->accept(this);
1486
1487 ir_to_mesa_src_reg index_reg;
1488
1489 if (element_size == 1) {
1490 index_reg = this->result;
1491 } else {
1492 index_reg = get_temp(glsl_type::float_type);
1493
1494 ir_to_mesa_emit_op2(ir, OPCODE_MUL,
1495 ir_to_mesa_dst_reg_from_src(index_reg),
1496 this->result, src_reg_for_float(element_size));
1497 }
1498
1499 src_reg.reladdr = talloc(mem_ctx, ir_to_mesa_src_reg);
1500 memcpy(src_reg.reladdr, &index_reg, sizeof(index_reg));
1501 }
1502
1503 /* If the type is smaller than a vec4, replicate the last channel out. */
1504 if (ir->type->is_scalar() || ir->type->is_vector())
1505 src_reg.swizzle = swizzle_for_size(ir->type->vector_elements);
1506 else
1507 src_reg.swizzle = SWIZZLE_NOOP;
1508
1509 this->result = src_reg;
1510 }
1511
1512 void
1513 ir_to_mesa_visitor::visit(ir_dereference_record *ir)
1514 {
1515 unsigned int i;
1516 const glsl_type *struct_type = ir->record->type;
1517 int offset = 0;
1518
1519 ir->record->accept(this);
1520
1521 for (i = 0; i < struct_type->length; i++) {
1522 if (strcmp(struct_type->fields.structure[i].name, ir->field) == 0)
1523 break;
1524 offset += type_size(struct_type->fields.structure[i].type);
1525 }
1526 this->result.swizzle = swizzle_for_size(ir->type->vector_elements);
1527 this->result.index += offset;
1528 }
1529
1530 /**
1531 * We want to be careful in assignment setup to hit the actual storage
1532 * instead of potentially using a temporary like we might with the
1533 * ir_dereference handler.
1534 */
1535 static struct ir_to_mesa_dst_reg
1536 get_assignment_lhs(ir_dereference *ir, ir_to_mesa_visitor *v)
1537 {
1538 /* The LHS must be a dereference. If the LHS is a variable indexed array
1539 * access of a vector, it must be separated into a series conditional moves
1540 * before reaching this point (see ir_vec_index_to_cond_assign).
1541 */
1542 assert(ir->as_dereference());
1543 ir_dereference_array *deref_array = ir->as_dereference_array();
1544 if (deref_array) {
1545 assert(!deref_array->array->type->is_vector());
1546 }
1547
1548 /* Use the rvalue deref handler for the most part. We'll ignore
1549 * swizzles in it and write swizzles using writemask, though.
1550 */
1551 ir->accept(v);
1552 return ir_to_mesa_dst_reg_from_src(v->result);
1553 }
1554
1555 void
1556 ir_to_mesa_visitor::visit(ir_assignment *ir)
1557 {
1558 struct ir_to_mesa_dst_reg l;
1559 struct ir_to_mesa_src_reg r;
1560 int i;
1561
1562 ir->rhs->accept(this);
1563 r = this->result;
1564
1565 l = get_assignment_lhs(ir->lhs, this);
1566
1567 /* FINISHME: This should really set to the correct maximal writemask for each
1568 * FINISHME: component written (in the loops below). This case can only
1569 * FINISHME: occur for matrices, arrays, and structures.
1570 */
1571 if (ir->write_mask == 0) {
1572 assert(!ir->lhs->type->is_scalar() && !ir->lhs->type->is_vector());
1573 l.writemask = WRITEMASK_XYZW;
1574 } else if (ir->lhs->type->is_scalar()) {
1575 /* FINISHME: This hack makes writing to gl_FragData, which lives in the
1576 * FINISHME: W component of fragment shader output zero, work correctly.
1577 */
1578 l.writemask = WRITEMASK_XYZW;
1579 } else {
1580 assert(ir->lhs->type->is_vector());
1581 l.writemask = ir->write_mask;
1582 }
1583
1584 assert(l.file != PROGRAM_UNDEFINED);
1585 assert(r.file != PROGRAM_UNDEFINED);
1586
1587 if (ir->condition) {
1588 ir_to_mesa_src_reg condition;
1589
1590 ir->condition->accept(this);
1591 condition = this->result;
1592
1593 /* We use the OPCODE_CMP (a < 0 ? b : c) for conditional moves,
1594 * and the condition we produced is 0.0 or 1.0. By flipping the
1595 * sign, we can choose which value OPCODE_CMP produces without
1596 * an extra computing the condition.
1597 */
1598 condition.negate = ~condition.negate;
1599 for (i = 0; i < type_size(ir->lhs->type); i++) {
1600 ir_to_mesa_emit_op3(ir, OPCODE_CMP, l,
1601 condition, r, ir_to_mesa_src_reg_from_dst(l));
1602 l.index++;
1603 r.index++;
1604 }
1605 } else {
1606 for (i = 0; i < type_size(ir->lhs->type); i++) {
1607 ir_to_mesa_emit_op1(ir, OPCODE_MOV, l, r);
1608 l.index++;
1609 r.index++;
1610 }
1611 }
1612 }
1613
1614
1615 void
1616 ir_to_mesa_visitor::visit(ir_constant *ir)
1617 {
1618 ir_to_mesa_src_reg src_reg;
1619 GLfloat stack_vals[4] = { 0 };
1620 GLfloat *values = stack_vals;
1621 unsigned int i;
1622
1623 /* Unfortunately, 4 floats is all we can get into
1624 * _mesa_add_unnamed_constant. So, make a temp to store an
1625 * aggregate constant and move each constant value into it. If we
1626 * get lucky, copy propagation will eliminate the extra moves.
1627 */
1628
1629 if (ir->type->base_type == GLSL_TYPE_STRUCT) {
1630 ir_to_mesa_src_reg temp_base = get_temp(ir->type);
1631 ir_to_mesa_dst_reg temp = ir_to_mesa_dst_reg_from_src(temp_base);
1632
1633 foreach_iter(exec_list_iterator, iter, ir->components) {
1634 ir_constant *field_value = (ir_constant *)iter.get();
1635 int size = type_size(field_value->type);
1636
1637 assert(size > 0);
1638
1639 field_value->accept(this);
1640 src_reg = this->result;
1641
1642 for (i = 0; i < (unsigned int)size; i++) {
1643 ir_to_mesa_emit_op1(ir, OPCODE_MOV, temp, src_reg);
1644
1645 src_reg.index++;
1646 temp.index++;
1647 }
1648 }
1649 this->result = temp_base;
1650 return;
1651 }
1652
1653 if (ir->type->is_array()) {
1654 ir_to_mesa_src_reg temp_base = get_temp(ir->type);
1655 ir_to_mesa_dst_reg temp = ir_to_mesa_dst_reg_from_src(temp_base);
1656 int size = type_size(ir->type->fields.array);
1657
1658 assert(size > 0);
1659
1660 for (i = 0; i < ir->type->length; i++) {
1661 ir->array_elements[i]->accept(this);
1662 src_reg = this->result;
1663 for (int j = 0; j < size; j++) {
1664 ir_to_mesa_emit_op1(ir, OPCODE_MOV, temp, src_reg);
1665
1666 src_reg.index++;
1667 temp.index++;
1668 }
1669 }
1670 this->result = temp_base;
1671 return;
1672 }
1673
1674 if (ir->type->is_matrix()) {
1675 ir_to_mesa_src_reg mat = get_temp(ir->type);
1676 ir_to_mesa_dst_reg mat_column = ir_to_mesa_dst_reg_from_src(mat);
1677
1678 for (i = 0; i < ir->type->matrix_columns; i++) {
1679 assert(ir->type->base_type == GLSL_TYPE_FLOAT);
1680 values = &ir->value.f[i * ir->type->vector_elements];
1681
1682 src_reg = ir_to_mesa_src_reg(PROGRAM_CONSTANT, -1, NULL);
1683 src_reg.index = _mesa_add_unnamed_constant(this->prog->Parameters,
1684 values,
1685 ir->type->vector_elements,
1686 &src_reg.swizzle);
1687 ir_to_mesa_emit_op1(ir, OPCODE_MOV, mat_column, src_reg);
1688
1689 mat_column.index++;
1690 }
1691
1692 this->result = mat;
1693 return;
1694 }
1695
1696 src_reg.file = PROGRAM_CONSTANT;
1697 switch (ir->type->base_type) {
1698 case GLSL_TYPE_FLOAT:
1699 values = &ir->value.f[0];
1700 break;
1701 case GLSL_TYPE_UINT:
1702 for (i = 0; i < ir->type->vector_elements; i++) {
1703 values[i] = ir->value.u[i];
1704 }
1705 break;
1706 case GLSL_TYPE_INT:
1707 for (i = 0; i < ir->type->vector_elements; i++) {
1708 values[i] = ir->value.i[i];
1709 }
1710 break;
1711 case GLSL_TYPE_BOOL:
1712 for (i = 0; i < ir->type->vector_elements; i++) {
1713 values[i] = ir->value.b[i];
1714 }
1715 break;
1716 default:
1717 assert(!"Non-float/uint/int/bool constant");
1718 }
1719
1720 this->result = ir_to_mesa_src_reg(PROGRAM_CONSTANT, -1, ir->type);
1721 this->result.index = _mesa_add_unnamed_constant(this->prog->Parameters,
1722 values,
1723 ir->type->vector_elements,
1724 &this->result.swizzle);
1725 }
1726
1727 function_entry *
1728 ir_to_mesa_visitor::get_function_signature(ir_function_signature *sig)
1729 {
1730 function_entry *entry;
1731
1732 foreach_iter(exec_list_iterator, iter, this->function_signatures) {
1733 entry = (function_entry *)iter.get();
1734
1735 if (entry->sig == sig)
1736 return entry;
1737 }
1738
1739 entry = talloc(mem_ctx, function_entry);
1740 entry->sig = sig;
1741 entry->sig_id = this->next_signature_id++;
1742 entry->bgn_inst = NULL;
1743
1744 /* Allocate storage for all the parameters. */
1745 foreach_iter(exec_list_iterator, iter, sig->parameters) {
1746 ir_variable *param = (ir_variable *)iter.get();
1747 variable_storage *storage;
1748
1749 storage = find_variable_storage(param);
1750 assert(!storage);
1751
1752 storage = new(mem_ctx) variable_storage(param, PROGRAM_TEMPORARY,
1753 this->next_temp);
1754 this->variables.push_tail(storage);
1755
1756 this->next_temp += type_size(param->type);
1757 }
1758
1759 if (!sig->return_type->is_void()) {
1760 entry->return_reg = get_temp(sig->return_type);
1761 } else {
1762 entry->return_reg = ir_to_mesa_undef;
1763 }
1764
1765 this->function_signatures.push_tail(entry);
1766 return entry;
1767 }
1768
1769 void
1770 ir_to_mesa_visitor::visit(ir_call *ir)
1771 {
1772 ir_to_mesa_instruction *call_inst;
1773 ir_function_signature *sig = ir->get_callee();
1774 function_entry *entry = get_function_signature(sig);
1775 int i;
1776
1777 /* Process in parameters. */
1778 exec_list_iterator sig_iter = sig->parameters.iterator();
1779 foreach_iter(exec_list_iterator, iter, *ir) {
1780 ir_rvalue *param_rval = (ir_rvalue *)iter.get();
1781 ir_variable *param = (ir_variable *)sig_iter.get();
1782
1783 if (param->mode == ir_var_in ||
1784 param->mode == ir_var_inout) {
1785 variable_storage *storage = find_variable_storage(param);
1786 assert(storage);
1787
1788 param_rval->accept(this);
1789 ir_to_mesa_src_reg r = this->result;
1790
1791 ir_to_mesa_dst_reg l;
1792 l.file = storage->file;
1793 l.index = storage->index;
1794 l.reladdr = NULL;
1795 l.writemask = WRITEMASK_XYZW;
1796 l.cond_mask = COND_TR;
1797
1798 for (i = 0; i < type_size(param->type); i++) {
1799 ir_to_mesa_emit_op1(ir, OPCODE_MOV, l, r);
1800 l.index++;
1801 r.index++;
1802 }
1803 }
1804
1805 sig_iter.next();
1806 }
1807 assert(!sig_iter.has_next());
1808
1809 /* Emit call instruction */
1810 call_inst = ir_to_mesa_emit_op1(ir, OPCODE_CAL,
1811 ir_to_mesa_undef_dst, ir_to_mesa_undef);
1812 call_inst->function = entry;
1813
1814 /* Process out parameters. */
1815 sig_iter = sig->parameters.iterator();
1816 foreach_iter(exec_list_iterator, iter, *ir) {
1817 ir_rvalue *param_rval = (ir_rvalue *)iter.get();
1818 ir_variable *param = (ir_variable *)sig_iter.get();
1819
1820 if (param->mode == ir_var_out ||
1821 param->mode == ir_var_inout) {
1822 variable_storage *storage = find_variable_storage(param);
1823 assert(storage);
1824
1825 ir_to_mesa_src_reg r;
1826 r.file = storage->file;
1827 r.index = storage->index;
1828 r.reladdr = NULL;
1829 r.swizzle = SWIZZLE_NOOP;
1830 r.negate = 0;
1831
1832 param_rval->accept(this);
1833 ir_to_mesa_dst_reg l = ir_to_mesa_dst_reg_from_src(this->result);
1834
1835 for (i = 0; i < type_size(param->type); i++) {
1836 ir_to_mesa_emit_op1(ir, OPCODE_MOV, l, r);
1837 l.index++;
1838 r.index++;
1839 }
1840 }
1841
1842 sig_iter.next();
1843 }
1844 assert(!sig_iter.has_next());
1845
1846 /* Process return value. */
1847 this->result = entry->return_reg;
1848 }
1849
1850 class get_sampler_name : public ir_hierarchical_visitor
1851 {
1852 public:
1853 get_sampler_name(ir_to_mesa_visitor *mesa, ir_dereference *last)
1854 {
1855 this->mem_ctx = mesa->mem_ctx;
1856 this->mesa = mesa;
1857 this->name = NULL;
1858 this->offset = 0;
1859 this->last = last;
1860 }
1861
1862 virtual ir_visitor_status visit(ir_dereference_variable *ir)
1863 {
1864 this->name = ir->var->name;
1865 return visit_continue;
1866 }
1867
1868 virtual ir_visitor_status visit_leave(ir_dereference_record *ir)
1869 {
1870 this->name = talloc_asprintf(mem_ctx, "%s.%s", name, ir->field);
1871 return visit_continue;
1872 }
1873
1874 virtual ir_visitor_status visit_leave(ir_dereference_array *ir)
1875 {
1876 ir_constant *index = ir->array_index->as_constant();
1877 int i;
1878
1879 if (index) {
1880 i = index->value.i[0];
1881 } else {
1882 /* GLSL 1.10 and 1.20 allowed variable sampler array indices,
1883 * while GLSL 1.30 requires that the array indices be
1884 * constant integer expressions. We don't expect any driver
1885 * to actually work with a really variable array index, so
1886 * all that would work would be an unrolled loop counter that ends
1887 * up being constant above.
1888 */
1889 mesa->shader_program->InfoLog =
1890 talloc_asprintf_append(mesa->shader_program->InfoLog,
1891 "warning: Variable sampler array index "
1892 "unsupported.\nThis feature of the language "
1893 "was removed in GLSL 1.20 and is unlikely "
1894 "to be supported for 1.10 in Mesa.\n");
1895 i = 0;
1896 }
1897 if (ir != last) {
1898 this->name = talloc_asprintf(mem_ctx, "%s[%d]", name, i);
1899 } else {
1900 offset = i;
1901 }
1902 return visit_continue;
1903 }
1904
1905 ir_to_mesa_visitor *mesa;
1906 const char *name;
1907 void *mem_ctx;
1908 int offset;
1909 ir_dereference *last;
1910 };
1911
1912 int
1913 ir_to_mesa_visitor::get_sampler_uniform_value(ir_dereference *sampler)
1914 {
1915 get_sampler_name getname(this, sampler);
1916
1917 sampler->accept(&getname);
1918
1919 GLint index = _mesa_lookup_parameter_index(prog->Parameters, -1,
1920 getname.name);
1921
1922 if (index < 0) {
1923 fail_link(this->shader_program,
1924 "failed to find sampler named %s.\n", getname.name);
1925 return 0;
1926 }
1927
1928 index += getname.offset;
1929
1930 return this->prog->Parameters->ParameterValues[index][0];
1931 }
1932
1933 void
1934 ir_to_mesa_visitor::visit(ir_texture *ir)
1935 {
1936 ir_to_mesa_src_reg result_src, coord, lod_info, projector;
1937 ir_to_mesa_dst_reg result_dst, coord_dst;
1938 ir_to_mesa_instruction *inst = NULL;
1939 prog_opcode opcode = OPCODE_NOP;
1940
1941 ir->coordinate->accept(this);
1942
1943 /* Put our coords in a temp. We'll need to modify them for shadow,
1944 * projection, or LOD, so the only case we'd use it as is is if
1945 * we're doing plain old texturing. Mesa IR optimization should
1946 * handle cleaning up our mess in that case.
1947 */
1948 coord = get_temp(glsl_type::vec4_type);
1949 coord_dst = ir_to_mesa_dst_reg_from_src(coord);
1950 ir_to_mesa_emit_op1(ir, OPCODE_MOV, coord_dst,
1951 this->result);
1952
1953 if (ir->projector) {
1954 ir->projector->accept(this);
1955 projector = this->result;
1956 }
1957
1958 /* Storage for our result. Ideally for an assignment we'd be using
1959 * the actual storage for the result here, instead.
1960 */
1961 result_src = get_temp(glsl_type::vec4_type);
1962 result_dst = ir_to_mesa_dst_reg_from_src(result_src);
1963
1964 switch (ir->op) {
1965 case ir_tex:
1966 opcode = OPCODE_TEX;
1967 break;
1968 case ir_txb:
1969 opcode = OPCODE_TXB;
1970 ir->lod_info.bias->accept(this);
1971 lod_info = this->result;
1972 break;
1973 case ir_txl:
1974 opcode = OPCODE_TXL;
1975 ir->lod_info.lod->accept(this);
1976 lod_info = this->result;
1977 break;
1978 case ir_txd:
1979 case ir_txf:
1980 assert(!"GLSL 1.30 features unsupported");
1981 break;
1982 }
1983
1984 if (ir->projector) {
1985 if (opcode == OPCODE_TEX) {
1986 /* Slot the projector in as the last component of the coord. */
1987 coord_dst.writemask = WRITEMASK_W;
1988 ir_to_mesa_emit_op1(ir, OPCODE_MOV, coord_dst, projector);
1989 coord_dst.writemask = WRITEMASK_XYZW;
1990 opcode = OPCODE_TXP;
1991 } else {
1992 ir_to_mesa_src_reg coord_w = coord;
1993 coord_w.swizzle = SWIZZLE_WWWW;
1994
1995 /* For the other TEX opcodes there's no projective version
1996 * since the last slot is taken up by lod info. Do the
1997 * projective divide now.
1998 */
1999 coord_dst.writemask = WRITEMASK_W;
2000 ir_to_mesa_emit_op1(ir, OPCODE_RCP, coord_dst, projector);
2001
2002 coord_dst.writemask = WRITEMASK_XYZ;
2003 ir_to_mesa_emit_op2(ir, OPCODE_MUL, coord_dst, coord, coord_w);
2004
2005 coord_dst.writemask = WRITEMASK_XYZW;
2006 coord.swizzle = SWIZZLE_XYZW;
2007 }
2008 }
2009
2010 if (ir->shadow_comparitor) {
2011 /* Slot the shadow value in as the second to last component of the
2012 * coord.
2013 */
2014 ir->shadow_comparitor->accept(this);
2015 coord_dst.writemask = WRITEMASK_Z;
2016 ir_to_mesa_emit_op1(ir, OPCODE_MOV, coord_dst, this->result);
2017 coord_dst.writemask = WRITEMASK_XYZW;
2018 }
2019
2020 if (opcode == OPCODE_TXL || opcode == OPCODE_TXB) {
2021 /* Mesa IR stores lod or lod bias in the last channel of the coords. */
2022 coord_dst.writemask = WRITEMASK_W;
2023 ir_to_mesa_emit_op1(ir, OPCODE_MOV, coord_dst, lod_info);
2024 coord_dst.writemask = WRITEMASK_XYZW;
2025 }
2026
2027 inst = ir_to_mesa_emit_op1(ir, opcode, result_dst, coord);
2028
2029 if (ir->shadow_comparitor)
2030 inst->tex_shadow = GL_TRUE;
2031
2032 inst->sampler = get_sampler_uniform_value(ir->sampler);
2033
2034 const glsl_type *sampler_type = ir->sampler->type;
2035
2036 switch (sampler_type->sampler_dimensionality) {
2037 case GLSL_SAMPLER_DIM_1D:
2038 inst->tex_target = (sampler_type->sampler_array)
2039 ? TEXTURE_1D_ARRAY_INDEX : TEXTURE_1D_INDEX;
2040 break;
2041 case GLSL_SAMPLER_DIM_2D:
2042 inst->tex_target = (sampler_type->sampler_array)
2043 ? TEXTURE_2D_ARRAY_INDEX : TEXTURE_2D_INDEX;
2044 break;
2045 case GLSL_SAMPLER_DIM_3D:
2046 inst->tex_target = TEXTURE_3D_INDEX;
2047 break;
2048 case GLSL_SAMPLER_DIM_CUBE:
2049 inst->tex_target = TEXTURE_CUBE_INDEX;
2050 break;
2051 case GLSL_SAMPLER_DIM_RECT:
2052 inst->tex_target = TEXTURE_RECT_INDEX;
2053 break;
2054 case GLSL_SAMPLER_DIM_BUF:
2055 assert(!"FINISHME: Implement ARB_texture_buffer_object");
2056 break;
2057 default:
2058 assert(!"Should not get here.");
2059 }
2060
2061 this->result = result_src;
2062 }
2063
2064 void
2065 ir_to_mesa_visitor::visit(ir_return *ir)
2066 {
2067 if (ir->get_value()) {
2068 ir_to_mesa_dst_reg l;
2069 int i;
2070
2071 assert(current_function);
2072
2073 ir->get_value()->accept(this);
2074 ir_to_mesa_src_reg r = this->result;
2075
2076 l = ir_to_mesa_dst_reg_from_src(current_function->return_reg);
2077
2078 for (i = 0; i < type_size(current_function->sig->return_type); i++) {
2079 ir_to_mesa_emit_op1(ir, OPCODE_MOV, l, r);
2080 l.index++;
2081 r.index++;
2082 }
2083 }
2084
2085 ir_to_mesa_emit_op0(ir, OPCODE_RET);
2086 }
2087
2088 void
2089 ir_to_mesa_visitor::visit(ir_discard *ir)
2090 {
2091 struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog;
2092
2093 assert(ir->condition == NULL); /* FINISHME */
2094
2095 ir_to_mesa_emit_op0(ir, OPCODE_KIL_NV);
2096 fp->UsesKill = GL_TRUE;
2097 }
2098
2099 void
2100 ir_to_mesa_visitor::visit(ir_if *ir)
2101 {
2102 ir_to_mesa_instruction *cond_inst, *if_inst, *else_inst = NULL;
2103 ir_to_mesa_instruction *prev_inst;
2104
2105 prev_inst = (ir_to_mesa_instruction *)this->instructions.get_tail();
2106
2107 ir->condition->accept(this);
2108 assert(this->result.file != PROGRAM_UNDEFINED);
2109
2110 if (this->options->EmitCondCodes) {
2111 cond_inst = (ir_to_mesa_instruction *)this->instructions.get_tail();
2112
2113 /* See if we actually generated any instruction for generating
2114 * the condition. If not, then cook up a move to a temp so we
2115 * have something to set cond_update on.
2116 */
2117 if (cond_inst == prev_inst) {
2118 ir_to_mesa_src_reg temp = get_temp(glsl_type::bool_type);
2119 cond_inst = ir_to_mesa_emit_op1(ir->condition, OPCODE_MOV,
2120 ir_to_mesa_dst_reg_from_src(temp),
2121 result);
2122 }
2123 cond_inst->cond_update = GL_TRUE;
2124
2125 if_inst = ir_to_mesa_emit_op0(ir->condition, OPCODE_IF);
2126 if_inst->dst_reg.cond_mask = COND_NE;
2127 } else {
2128 if_inst = ir_to_mesa_emit_op1(ir->condition,
2129 OPCODE_IF, ir_to_mesa_undef_dst,
2130 this->result);
2131 }
2132
2133 this->instructions.push_tail(if_inst);
2134
2135 visit_exec_list(&ir->then_instructions, this);
2136
2137 if (!ir->else_instructions.is_empty()) {
2138 else_inst = ir_to_mesa_emit_op0(ir->condition, OPCODE_ELSE);
2139 visit_exec_list(&ir->else_instructions, this);
2140 }
2141
2142 if_inst = ir_to_mesa_emit_op1(ir->condition, OPCODE_ENDIF,
2143 ir_to_mesa_undef_dst, ir_to_mesa_undef);
2144 }
2145
2146 ir_to_mesa_visitor::ir_to_mesa_visitor()
2147 {
2148 result.file = PROGRAM_UNDEFINED;
2149 next_temp = 1;
2150 next_signature_id = 1;
2151 current_function = NULL;
2152 mem_ctx = talloc_new(NULL);
2153 }
2154
2155 ir_to_mesa_visitor::~ir_to_mesa_visitor()
2156 {
2157 talloc_free(mem_ctx);
2158 }
2159
2160 static struct prog_src_register
2161 mesa_src_reg_from_ir_src_reg(ir_to_mesa_src_reg reg)
2162 {
2163 struct prog_src_register mesa_reg;
2164
2165 mesa_reg.File = reg.file;
2166 assert(reg.index < (1 << INST_INDEX_BITS) - 1);
2167 mesa_reg.Index = reg.index;
2168 mesa_reg.Swizzle = reg.swizzle;
2169 mesa_reg.RelAddr = reg.reladdr != NULL;
2170 mesa_reg.Negate = reg.negate;
2171 mesa_reg.Abs = 0;
2172 mesa_reg.HasIndex2 = GL_FALSE;
2173 mesa_reg.RelAddr2 = 0;
2174 mesa_reg.Index2 = 0;
2175
2176 return mesa_reg;
2177 }
2178
2179 static void
2180 set_branchtargets(ir_to_mesa_visitor *v,
2181 struct prog_instruction *mesa_instructions,
2182 int num_instructions)
2183 {
2184 int if_count = 0, loop_count = 0;
2185 int *if_stack, *loop_stack;
2186 int if_stack_pos = 0, loop_stack_pos = 0;
2187 int i, j;
2188
2189 for (i = 0; i < num_instructions; i++) {
2190 switch (mesa_instructions[i].Opcode) {
2191 case OPCODE_IF:
2192 if_count++;
2193 break;
2194 case OPCODE_BGNLOOP:
2195 loop_count++;
2196 break;
2197 case OPCODE_BRK:
2198 case OPCODE_CONT:
2199 mesa_instructions[i].BranchTarget = -1;
2200 break;
2201 default:
2202 break;
2203 }
2204 }
2205
2206 if_stack = talloc_zero_array(v->mem_ctx, int, if_count);
2207 loop_stack = talloc_zero_array(v->mem_ctx, int, loop_count);
2208
2209 for (i = 0; i < num_instructions; i++) {
2210 switch (mesa_instructions[i].Opcode) {
2211 case OPCODE_IF:
2212 if_stack[if_stack_pos] = i;
2213 if_stack_pos++;
2214 break;
2215 case OPCODE_ELSE:
2216 mesa_instructions[if_stack[if_stack_pos - 1]].BranchTarget = i;
2217 if_stack[if_stack_pos - 1] = i;
2218 break;
2219 case OPCODE_ENDIF:
2220 mesa_instructions[if_stack[if_stack_pos - 1]].BranchTarget = i;
2221 if_stack_pos--;
2222 break;
2223 case OPCODE_BGNLOOP:
2224 loop_stack[loop_stack_pos] = i;
2225 loop_stack_pos++;
2226 break;
2227 case OPCODE_ENDLOOP:
2228 loop_stack_pos--;
2229 /* Rewrite any breaks/conts at this nesting level (haven't
2230 * already had a BranchTarget assigned) to point to the end
2231 * of the loop.
2232 */
2233 for (j = loop_stack[loop_stack_pos]; j < i; j++) {
2234 if (mesa_instructions[j].Opcode == OPCODE_BRK ||
2235 mesa_instructions[j].Opcode == OPCODE_CONT) {
2236 if (mesa_instructions[j].BranchTarget == -1) {
2237 mesa_instructions[j].BranchTarget = i;
2238 }
2239 }
2240 }
2241 /* The loop ends point at each other. */
2242 mesa_instructions[i].BranchTarget = loop_stack[loop_stack_pos];
2243 mesa_instructions[loop_stack[loop_stack_pos]].BranchTarget = i;
2244 break;
2245 case OPCODE_CAL:
2246 foreach_iter(exec_list_iterator, iter, v->function_signatures) {
2247 function_entry *entry = (function_entry *)iter.get();
2248
2249 if (entry->sig_id == mesa_instructions[i].BranchTarget) {
2250 mesa_instructions[i].BranchTarget = entry->inst;
2251 break;
2252 }
2253 }
2254 break;
2255 default:
2256 break;
2257 }
2258 }
2259 }
2260
2261 static void
2262 print_program(struct prog_instruction *mesa_instructions,
2263 ir_instruction **mesa_instruction_annotation,
2264 int num_instructions)
2265 {
2266 ir_instruction *last_ir = NULL;
2267 int i;
2268 int indent = 0;
2269
2270 for (i = 0; i < num_instructions; i++) {
2271 struct prog_instruction *mesa_inst = mesa_instructions + i;
2272 ir_instruction *ir = mesa_instruction_annotation[i];
2273
2274 fprintf(stdout, "%3d: ", i);
2275
2276 if (last_ir != ir && ir) {
2277 int j;
2278
2279 for (j = 0; j < indent; j++) {
2280 fprintf(stdout, " ");
2281 }
2282 ir->print();
2283 printf("\n");
2284 last_ir = ir;
2285
2286 fprintf(stdout, " "); /* line number spacing. */
2287 }
2288
2289 indent = _mesa_fprint_instruction_opt(stdout, mesa_inst, indent,
2290 PROG_PRINT_DEBUG, NULL);
2291 }
2292 }
2293
2294 static void
2295 count_resources(struct gl_program *prog)
2296 {
2297 unsigned int i;
2298
2299 prog->SamplersUsed = 0;
2300
2301 for (i = 0; i < prog->NumInstructions; i++) {
2302 struct prog_instruction *inst = &prog->Instructions[i];
2303
2304 if (_mesa_is_tex_instruction(inst->Opcode)) {
2305 prog->SamplerTargets[inst->TexSrcUnit] =
2306 (gl_texture_index)inst->TexSrcTarget;
2307 prog->SamplersUsed |= 1 << inst->TexSrcUnit;
2308 if (inst->TexShadow) {
2309 prog->ShadowSamplers |= 1 << inst->TexSrcUnit;
2310 }
2311 }
2312 }
2313
2314 _mesa_update_shader_textures_used(prog);
2315 }
2316
2317 struct uniform_sort {
2318 struct gl_uniform *u;
2319 int pos;
2320 };
2321
2322 /* The shader_program->Uniforms list is almost sorted in increasing
2323 * uniform->{Frag,Vert}Pos locations, but not quite when there are
2324 * uniforms shared between targets. We need to add parameters in
2325 * increasing order for the targets.
2326 */
2327 static int
2328 sort_uniforms(const void *a, const void *b)
2329 {
2330 struct uniform_sort *u1 = (struct uniform_sort *)a;
2331 struct uniform_sort *u2 = (struct uniform_sort *)b;
2332
2333 return u1->pos - u2->pos;
2334 }
2335
2336 /* Add the uniforms to the parameters. The linker chose locations
2337 * in our parameters lists (which weren't created yet), which the
2338 * uniforms code will use to poke values into our parameters list
2339 * when uniforms are updated.
2340 */
2341 static void
2342 add_uniforms_to_parameters_list(struct gl_shader_program *shader_program,
2343 struct gl_shader *shader,
2344 struct gl_program *prog)
2345 {
2346 unsigned int i;
2347 unsigned int next_sampler = 0, num_uniforms = 0;
2348 struct uniform_sort *sorted_uniforms;
2349
2350 sorted_uniforms = talloc_array(NULL, struct uniform_sort,
2351 shader_program->Uniforms->NumUniforms);
2352
2353 for (i = 0; i < shader_program->Uniforms->NumUniforms; i++) {
2354 struct gl_uniform *uniform = shader_program->Uniforms->Uniforms + i;
2355 int parameter_index = -1;
2356
2357 switch (shader->Type) {
2358 case GL_VERTEX_SHADER:
2359 parameter_index = uniform->VertPos;
2360 break;
2361 case GL_FRAGMENT_SHADER:
2362 parameter_index = uniform->FragPos;
2363 break;
2364 case GL_GEOMETRY_SHADER:
2365 parameter_index = uniform->GeomPos;
2366 break;
2367 }
2368
2369 /* Only add uniforms used in our target. */
2370 if (parameter_index != -1) {
2371 sorted_uniforms[num_uniforms].pos = parameter_index;
2372 sorted_uniforms[num_uniforms].u = uniform;
2373 num_uniforms++;
2374 }
2375 }
2376
2377 qsort(sorted_uniforms, num_uniforms, sizeof(struct uniform_sort),
2378 sort_uniforms);
2379
2380 for (i = 0; i < num_uniforms; i++) {
2381 struct gl_uniform *uniform = sorted_uniforms[i].u;
2382 int parameter_index = sorted_uniforms[i].pos;
2383 const glsl_type *type = uniform->Type;
2384 unsigned int size;
2385
2386 if (type->is_vector() ||
2387 type->is_scalar()) {
2388 size = type->vector_elements;
2389 } else {
2390 size = type_size(type) * 4;
2391 }
2392
2393 gl_register_file file;
2394 if (type->is_sampler() ||
2395 (type->is_array() && type->fields.array->is_sampler())) {
2396 file = PROGRAM_SAMPLER;
2397 } else {
2398 file = PROGRAM_UNIFORM;
2399 }
2400
2401 GLint index = _mesa_lookup_parameter_index(prog->Parameters, -1,
2402 uniform->Name);
2403
2404 if (index < 0) {
2405 index = _mesa_add_parameter(prog->Parameters, file,
2406 uniform->Name, size, type->gl_type,
2407 NULL, NULL, 0x0);
2408
2409 /* Sampler uniform values are stored in prog->SamplerUnits,
2410 * and the entry in that array is selected by this index we
2411 * store in ParameterValues[].
2412 */
2413 if (file == PROGRAM_SAMPLER) {
2414 for (unsigned int j = 0; j < size / 4; j++)
2415 prog->Parameters->ParameterValues[index + j][0] = next_sampler++;
2416 }
2417
2418 /* The location chosen in the Parameters list here (returned
2419 * from _mesa_add_uniform) has to match what the linker chose.
2420 */
2421 if (index != parameter_index) {
2422 fail_link(shader_program, "Allocation of uniform `%s' to target "
2423 "failed (%d vs %d)\n",
2424 uniform->Name, index, parameter_index);
2425 }
2426 }
2427 }
2428
2429 talloc_free(sorted_uniforms);
2430 }
2431
2432 static void
2433 set_uniform_initializer(GLcontext *ctx, void *mem_ctx,
2434 struct gl_shader_program *shader_program,
2435 const char *name, const glsl_type *type,
2436 ir_constant *val)
2437 {
2438 if (type->is_record()) {
2439 ir_constant *field_constant;
2440
2441 field_constant = (ir_constant *)val->components.get_head();
2442
2443 for (unsigned int i = 0; i < type->length; i++) {
2444 const glsl_type *field_type = type->fields.structure[i].type;
2445 const char *field_name = talloc_asprintf(mem_ctx, "%s.%s", name,
2446 type->fields.structure[i].name);
2447 set_uniform_initializer(ctx, mem_ctx, shader_program, field_name,
2448 field_type, field_constant);
2449 field_constant = (ir_constant *)field_constant->next;
2450 }
2451 return;
2452 }
2453
2454 int loc = _mesa_get_uniform_location(ctx, shader_program, name);
2455
2456 if (loc == -1) {
2457 fail_link(shader_program,
2458 "Couldn't find uniform for initializer %s\n", name);
2459 return;
2460 }
2461
2462 for (unsigned int i = 0; i < (type->is_array() ? type->length : 1); i++) {
2463 ir_constant *element;
2464 const glsl_type *element_type;
2465 if (type->is_array()) {
2466 element = val->array_elements[i];
2467 element_type = type->fields.array;
2468 } else {
2469 element = val;
2470 element_type = type;
2471 }
2472
2473 void *values;
2474
2475 if (element_type->base_type == GLSL_TYPE_BOOL) {
2476 int *conv = talloc_array(mem_ctx, int, element_type->components());
2477 for (unsigned int j = 0; j < element_type->components(); j++) {
2478 conv[j] = element->value.b[j];
2479 }
2480 values = (void *)conv;
2481 element_type = glsl_type::get_instance(GLSL_TYPE_INT,
2482 element_type->vector_elements,
2483 1);
2484 } else {
2485 values = &element->value;
2486 }
2487
2488 if (element_type->is_matrix()) {
2489 _mesa_uniform_matrix(ctx, shader_program,
2490 element_type->matrix_columns,
2491 element_type->vector_elements,
2492 loc, 1, GL_FALSE, (GLfloat *)values);
2493 loc += element_type->matrix_columns;
2494 } else {
2495 _mesa_uniform(ctx, shader_program, loc, element_type->matrix_columns,
2496 values, element_type->gl_type);
2497 loc += type_size(element_type);
2498 }
2499 }
2500 }
2501
2502 static void
2503 set_uniform_initializers(GLcontext *ctx,
2504 struct gl_shader_program *shader_program)
2505 {
2506 void *mem_ctx = NULL;
2507
2508 for (unsigned int i = 0; i < shader_program->_NumLinkedShaders; i++) {
2509 struct gl_shader *shader = shader_program->_LinkedShaders[i];
2510 foreach_iter(exec_list_iterator, iter, *shader->ir) {
2511 ir_instruction *ir = (ir_instruction *)iter.get();
2512 ir_variable *var = ir->as_variable();
2513
2514 if (!var || var->mode != ir_var_uniform || !var->constant_value)
2515 continue;
2516
2517 if (!mem_ctx)
2518 mem_ctx = talloc_new(NULL);
2519
2520 set_uniform_initializer(ctx, mem_ctx, shader_program, var->name,
2521 var->type, var->constant_value);
2522 }
2523 }
2524
2525 talloc_free(mem_ctx);
2526 }
2527
2528 struct gl_program *
2529 get_mesa_program(GLcontext *ctx, struct gl_shader_program *shader_program,
2530 struct gl_shader *shader)
2531 {
2532 ir_to_mesa_visitor v;
2533 struct prog_instruction *mesa_instructions, *mesa_inst;
2534 ir_instruction **mesa_instruction_annotation;
2535 int i;
2536 struct gl_program *prog;
2537 GLenum target;
2538 const char *target_string;
2539 GLboolean progress;
2540 struct gl_shader_compiler_options *options =
2541 &ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(shader->Type)];
2542
2543 switch (shader->Type) {
2544 case GL_VERTEX_SHADER:
2545 target = GL_VERTEX_PROGRAM_ARB;
2546 target_string = "vertex";
2547 break;
2548 case GL_FRAGMENT_SHADER:
2549 target = GL_FRAGMENT_PROGRAM_ARB;
2550 target_string = "fragment";
2551 break;
2552 default:
2553 assert(!"should not be reached");
2554 return NULL;
2555 }
2556
2557 validate_ir_tree(shader->ir);
2558
2559 prog = ctx->Driver.NewProgram(ctx, target, shader_program->Name);
2560 if (!prog)
2561 return NULL;
2562 prog->Parameters = _mesa_new_parameter_list();
2563 prog->Varying = _mesa_new_parameter_list();
2564 prog->Attributes = _mesa_new_parameter_list();
2565 v.ctx = ctx;
2566 v.prog = prog;
2567 v.shader_program = shader_program;
2568 v.options = options;
2569
2570 add_uniforms_to_parameters_list(shader_program, shader, prog);
2571
2572 /* Emit Mesa IR for main(). */
2573 visit_exec_list(shader->ir, &v);
2574 v.ir_to_mesa_emit_op0(NULL, OPCODE_END);
2575
2576 /* Now emit bodies for any functions that were used. */
2577 do {
2578 progress = GL_FALSE;
2579
2580 foreach_iter(exec_list_iterator, iter, v.function_signatures) {
2581 function_entry *entry = (function_entry *)iter.get();
2582
2583 if (!entry->bgn_inst) {
2584 v.current_function = entry;
2585
2586 entry->bgn_inst = v.ir_to_mesa_emit_op0(NULL, OPCODE_BGNSUB);
2587 entry->bgn_inst->function = entry;
2588
2589 visit_exec_list(&entry->sig->body, &v);
2590
2591 ir_to_mesa_instruction *last;
2592 last = (ir_to_mesa_instruction *)v.instructions.get_tail();
2593 if (last->op != OPCODE_RET)
2594 v.ir_to_mesa_emit_op0(NULL, OPCODE_RET);
2595
2596 ir_to_mesa_instruction *end;
2597 end = v.ir_to_mesa_emit_op0(NULL, OPCODE_ENDSUB);
2598 end->function = entry;
2599
2600 progress = GL_TRUE;
2601 }
2602 }
2603 } while (progress);
2604
2605 prog->NumTemporaries = v.next_temp;
2606
2607 int num_instructions = 0;
2608 foreach_iter(exec_list_iterator, iter, v.instructions) {
2609 num_instructions++;
2610 }
2611
2612 mesa_instructions =
2613 (struct prog_instruction *)calloc(num_instructions,
2614 sizeof(*mesa_instructions));
2615 mesa_instruction_annotation = talloc_array(v.mem_ctx, ir_instruction *,
2616 num_instructions);
2617
2618 mesa_inst = mesa_instructions;
2619 i = 0;
2620 foreach_iter(exec_list_iterator, iter, v.instructions) {
2621 ir_to_mesa_instruction *inst = (ir_to_mesa_instruction *)iter.get();
2622
2623 mesa_inst->Opcode = inst->op;
2624 mesa_inst->CondUpdate = inst->cond_update;
2625 mesa_inst->DstReg.File = inst->dst_reg.file;
2626 mesa_inst->DstReg.Index = inst->dst_reg.index;
2627 mesa_inst->DstReg.CondMask = inst->dst_reg.cond_mask;
2628 mesa_inst->DstReg.WriteMask = inst->dst_reg.writemask;
2629 mesa_inst->DstReg.RelAddr = inst->dst_reg.reladdr != NULL;
2630 mesa_inst->SrcReg[0] = mesa_src_reg_from_ir_src_reg(inst->src_reg[0]);
2631 mesa_inst->SrcReg[1] = mesa_src_reg_from_ir_src_reg(inst->src_reg[1]);
2632 mesa_inst->SrcReg[2] = mesa_src_reg_from_ir_src_reg(inst->src_reg[2]);
2633 mesa_inst->TexSrcUnit = inst->sampler;
2634 mesa_inst->TexSrcTarget = inst->tex_target;
2635 mesa_inst->TexShadow = inst->tex_shadow;
2636 mesa_instruction_annotation[i] = inst->ir;
2637
2638 /* Set IndirectRegisterFiles. */
2639 if (mesa_inst->DstReg.RelAddr)
2640 prog->IndirectRegisterFiles |= 1 << mesa_inst->DstReg.File;
2641
2642 for (unsigned src = 0; src < 3; src++)
2643 if (mesa_inst->SrcReg[src].RelAddr)
2644 prog->IndirectRegisterFiles |= 1 << mesa_inst->SrcReg[src].File;
2645
2646 if (options->EmitNoIfs && mesa_inst->Opcode == OPCODE_IF) {
2647 fail_link(shader_program, "Couldn't flatten if statement\n");
2648 }
2649
2650 switch (mesa_inst->Opcode) {
2651 case OPCODE_BGNSUB:
2652 inst->function->inst = i;
2653 mesa_inst->Comment = strdup(inst->function->sig->function_name());
2654 break;
2655 case OPCODE_ENDSUB:
2656 mesa_inst->Comment = strdup(inst->function->sig->function_name());
2657 break;
2658 case OPCODE_CAL:
2659 mesa_inst->BranchTarget = inst->function->sig_id; /* rewritten later */
2660 break;
2661 case OPCODE_ARL:
2662 prog->NumAddressRegs = 1;
2663 break;
2664 default:
2665 break;
2666 }
2667
2668 mesa_inst++;
2669 i++;
2670 }
2671
2672 set_branchtargets(&v, mesa_instructions, num_instructions);
2673
2674 if (ctx->Shader.Flags & GLSL_DUMP) {
2675 printf("\n");
2676 printf("GLSL IR for linked %s program %d:\n", target_string,
2677 shader_program->Name);
2678 _mesa_print_ir(shader->ir, NULL);
2679 printf("\n");
2680 printf("\n");
2681 printf("Mesa IR for linked %s program %d:\n", target_string,
2682 shader_program->Name);
2683 print_program(mesa_instructions, mesa_instruction_annotation,
2684 num_instructions);
2685 }
2686
2687 prog->Instructions = mesa_instructions;
2688 prog->NumInstructions = num_instructions;
2689
2690 do_set_program_inouts(shader->ir, prog);
2691 count_resources(prog);
2692
2693 _mesa_reference_program(ctx, &shader->Program, prog);
2694
2695 if ((ctx->Shader.Flags & GLSL_NO_OPT) == 0) {
2696 _mesa_optimize_program(ctx, prog);
2697 }
2698
2699 return prog;
2700 }
2701
2702 extern "C" {
2703 GLboolean
2704 _mesa_ir_compile_shader(GLcontext *ctx, struct gl_shader *shader)
2705 {
2706 assert(shader->CompileStatus);
2707 (void) ctx;
2708
2709 return GL_TRUE;
2710 }
2711
2712 GLboolean
2713 _mesa_ir_link_shader(GLcontext *ctx, struct gl_shader_program *prog)
2714 {
2715 assert(prog->LinkStatus);
2716
2717 for (unsigned i = 0; i < prog->_NumLinkedShaders; i++) {
2718 bool progress;
2719 exec_list *ir = prog->_LinkedShaders[i]->ir;
2720 struct gl_shader_compiler_options *options =
2721 &ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(prog->_LinkedShaders[i]->Type)];
2722
2723 do {
2724 progress = false;
2725
2726 /* Lowering */
2727 do_mat_op_to_vec(ir);
2728 do_mod_to_fract(ir);
2729 do_div_to_mul_rcp(ir);
2730 do_explog_to_explog2(ir);
2731
2732 progress = do_common_optimization(ir, true, options->MaxUnrollIterations) || progress;
2733
2734 if (options->EmitNoIfs)
2735 progress = do_if_to_cond_assign(ir) || progress;
2736
2737 if (options->EmitNoNoise)
2738 progress = lower_noise(ir) || progress;
2739
2740 progress = do_vec_index_to_cond_assign(ir) || progress;
2741 } while (progress);
2742
2743 validate_ir_tree(ir);
2744 }
2745
2746 for (unsigned i = 0; i < prog->_NumLinkedShaders; i++) {
2747 struct gl_program *linked_prog;
2748 bool ok = true;
2749
2750 linked_prog = get_mesa_program(ctx, prog, prog->_LinkedShaders[i]);
2751
2752 switch (prog->_LinkedShaders[i]->Type) {
2753 case GL_VERTEX_SHADER:
2754 _mesa_reference_vertprog(ctx, &prog->VertexProgram,
2755 (struct gl_vertex_program *)linked_prog);
2756 ok = ctx->Driver.ProgramStringNotify(ctx, GL_VERTEX_PROGRAM_ARB,
2757 linked_prog);
2758 break;
2759 case GL_FRAGMENT_SHADER:
2760 _mesa_reference_fragprog(ctx, &prog->FragmentProgram,
2761 (struct gl_fragment_program *)linked_prog);
2762 ok = ctx->Driver.ProgramStringNotify(ctx, GL_FRAGMENT_PROGRAM_ARB,
2763 linked_prog);
2764 break;
2765 }
2766 if (!ok) {
2767 return GL_FALSE;
2768 }
2769 _mesa_reference_program(ctx, &linked_prog, NULL);
2770 }
2771
2772 return GL_TRUE;
2773 }
2774
2775 void
2776 _mesa_glsl_compile_shader(GLcontext *ctx, struct gl_shader *shader)
2777 {
2778 struct _mesa_glsl_parse_state *state =
2779 new(shader) _mesa_glsl_parse_state(ctx, shader->Type, shader);
2780
2781 const char *source = shader->Source;
2782 /* Check if the user called glCompileShader without first calling
2783 * glShaderSource. This should fail to compile, but not raise a GL_ERROR.
2784 */
2785 if (source == NULL) {
2786 shader->CompileStatus = GL_FALSE;
2787 return;
2788 }
2789
2790 state->error = preprocess(state, &source, &state->info_log,
2791 &ctx->Extensions, ctx->API);
2792
2793 if (ctx->Shader.Flags & GLSL_DUMP) {
2794 printf("GLSL source for shader %d:\n", shader->Name);
2795 printf("%s\n", shader->Source);
2796 }
2797
2798 if (!state->error) {
2799 _mesa_glsl_lexer_ctor(state, source);
2800 _mesa_glsl_parse(state);
2801 _mesa_glsl_lexer_dtor(state);
2802 }
2803
2804 talloc_free(shader->ir);
2805 shader->ir = new(shader) exec_list;
2806 if (!state->error && !state->translation_unit.is_empty())
2807 _mesa_ast_to_hir(shader->ir, state);
2808
2809 if (!state->error && !shader->ir->is_empty()) {
2810 validate_ir_tree(shader->ir);
2811
2812 /* Do some optimization at compile time to reduce shader IR size
2813 * and reduce later work if the same shader is linked multiple times
2814 */
2815 while (do_common_optimization(shader->ir, false, 32))
2816 ;
2817
2818 validate_ir_tree(shader->ir);
2819 }
2820
2821 shader->symbols = state->symbols;
2822
2823 shader->CompileStatus = !state->error;
2824 shader->InfoLog = state->info_log;
2825 shader->Version = state->language_version;
2826 memcpy(shader->builtins_to_link, state->builtins_to_link,
2827 sizeof(shader->builtins_to_link[0]) * state->num_builtins_to_link);
2828 shader->num_builtins_to_link = state->num_builtins_to_link;
2829
2830 if (ctx->Shader.Flags & GLSL_LOG) {
2831 _mesa_write_shader_to_file(shader);
2832 }
2833
2834 if (ctx->Shader.Flags & GLSL_DUMP) {
2835 if (shader->CompileStatus) {
2836 printf("GLSL IR for shader %d:\n", shader->Name);
2837 _mesa_print_ir(shader->ir, NULL);
2838 printf("\n\n");
2839 } else {
2840 printf("GLSL shader %d failed to compile.\n", shader->Name);
2841 }
2842 if (shader->InfoLog && shader->InfoLog[0] != 0) {
2843 printf("GLSL shader %d info log:\n", shader->Name);
2844 printf("%s\n", shader->InfoLog);
2845 }
2846 }
2847
2848 /* Retain any live IR, but trash the rest. */
2849 reparent_ir(shader->ir, shader->ir);
2850
2851 talloc_free(state);
2852
2853 if (shader->CompileStatus) {
2854 if (!ctx->Driver.CompileShader(ctx, shader))
2855 shader->CompileStatus = GL_FALSE;
2856 }
2857 }
2858
2859 void
2860 _mesa_glsl_link_shader(GLcontext *ctx, struct gl_shader_program *prog)
2861 {
2862 unsigned int i;
2863
2864 _mesa_clear_shader_program_data(ctx, prog);
2865
2866 prog->LinkStatus = GL_TRUE;
2867
2868 for (i = 0; i < prog->NumShaders; i++) {
2869 if (!prog->Shaders[i]->CompileStatus) {
2870 fail_link(prog, "linking with uncompiled shader");
2871 prog->LinkStatus = GL_FALSE;
2872 }
2873 }
2874
2875 prog->Varying = _mesa_new_parameter_list();
2876 _mesa_reference_vertprog(ctx, &prog->VertexProgram, NULL);
2877 _mesa_reference_fragprog(ctx, &prog->FragmentProgram, NULL);
2878
2879 if (prog->LinkStatus) {
2880 link_shaders(ctx, prog);
2881 }
2882
2883 if (prog->LinkStatus) {
2884 if (!ctx->Driver.LinkShader(ctx, prog)) {
2885 prog->LinkStatus = GL_FALSE;
2886 }
2887 }
2888
2889 set_uniform_initializers(ctx, prog);
2890
2891 if (ctx->Shader.Flags & GLSL_DUMP) {
2892 if (!prog->LinkStatus) {
2893 printf("GLSL shader program %d failed to link\n", prog->Name);
2894 }
2895
2896 if (prog->InfoLog && prog->InfoLog[0] != 0) {
2897 printf("GLSL shader program %d info log:\n", prog->Name);
2898 printf("%s\n", prog->InfoLog);
2899 }
2900 }
2901 }
2902
2903 } /* extern "C" */