1 /**************************************************************************
2 *
3 * Copyright 2010 VMware, Inc.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
18 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
19 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
20 * USE OR OTHER DEALINGS IN THE SOFTWARE.
21 *
22 * The above copyright notice and this permission notice (including the
23 * next paragraph) shall be included in all copies or substantial portions
24 * of the Software.
25 *
26 **************************************************************************/
27
28
29 #include "util/compiler.h"
30 #include "util/u_memory.h"
31 #include "util/u_math.h"
32 #include "tgsi/tgsi_parse.h"
33 #include "tgsi/tgsi_util.h"
34 #include "tgsi/tgsi_dump.h"
35 #include "tgsi/tgsi_strings.h"
36 #include "lp_bld_debug.h"
37 #include "lp_bld_tgsi.h"
38
39
40 /**
41 * Analysis context.
42 *
43 * This is where we keep store the value of each channel of the IMM/TEMP/OUT
44 * register values, as we walk the shader.
45 */
46 struct analysis_context
47 {
48 struct lp_tgsi_info *info;
49
50 unsigned num_imms;
51 float imm[LP_MAX_TGSI_IMMEDIATES][4];
52 unsigned sample_target[PIPE_MAX_SHADER_SAMPLER_VIEWS];
53
54 struct lp_tgsi_channel_info temp[32][4];
55 };
56
57
58 /**
59 * Describe the specified channel of the src register.
60 */
61 static void
analyse_src(struct analysis_context * ctx,struct lp_tgsi_channel_info * chan_info,const struct tgsi_src_register * src,unsigned chan)62 analyse_src(struct analysis_context *ctx,
63 struct lp_tgsi_channel_info *chan_info,
64 const struct tgsi_src_register *src,
65 unsigned chan)
66 {
67 chan_info->file = TGSI_FILE_NULL;
68 if (!src->Indirect && !src->Absolute && !src->Negate) {
69 unsigned swizzle = tgsi_util_get_src_register_swizzle(src, chan);
70 if (src->File == TGSI_FILE_TEMPORARY) {
71 if (src->Index < ARRAY_SIZE(ctx->temp)) {
72 *chan_info = ctx->temp[src->Index][swizzle];
73 }
74 } else {
75 chan_info->file = src->File;
76 if (src->File == TGSI_FILE_IMMEDIATE) {
77 assert(src->Index < ARRAY_SIZE(ctx->imm));
78 if (src->Index < ARRAY_SIZE(ctx->imm)) {
79 chan_info->u.value = ctx->imm[src->Index][swizzle];
80 }
81 } else {
82 chan_info->u.index = src->Index;
83 chan_info->swizzle = swizzle;
84 }
85 }
86 }
87 }
88
89
90 /**
91 * Whether this register channel refers to a specific immediate value.
92 */
93 static boolean
is_immediate(const struct lp_tgsi_channel_info * chan_info,float value)94 is_immediate(const struct lp_tgsi_channel_info *chan_info, float value)
95 {
96 return chan_info->file == TGSI_FILE_IMMEDIATE &&
97 chan_info->u.value == value;
98 }
99
100
101 /**
102 * Analyse properties of tex instructions, in particular used
103 * to figure out if a texture is considered indirect.
104 * Not actually used by much except the tgsi dumping code.
105 */
106 static void
analyse_tex(struct analysis_context * ctx,const struct tgsi_full_instruction * inst,enum lp_build_tex_modifier modifier)107 analyse_tex(struct analysis_context *ctx,
108 const struct tgsi_full_instruction *inst,
109 enum lp_build_tex_modifier modifier)
110 {
111 struct lp_tgsi_info *info = ctx->info;
112 unsigned chan;
113
114 if (info->num_texs < ARRAY_SIZE(info->tex)) {
115 struct lp_tgsi_texture_info *tex_info = &info->tex[info->num_texs];
116 boolean indirect = FALSE;
117 unsigned readmask = 0;
118
119 tex_info->target = inst->Texture.Texture;
120 switch (inst->Texture.Texture) {
121 case TGSI_TEXTURE_1D:
122 readmask = TGSI_WRITEMASK_X;
123 break;
124 case TGSI_TEXTURE_1D_ARRAY:
125 case TGSI_TEXTURE_2D:
126 case TGSI_TEXTURE_RECT:
127 readmask = TGSI_WRITEMASK_XY;
128 break;
129 case TGSI_TEXTURE_SHADOW1D:
130 case TGSI_TEXTURE_SHADOW1D_ARRAY:
131 case TGSI_TEXTURE_SHADOW2D:
132 case TGSI_TEXTURE_SHADOWRECT:
133 case TGSI_TEXTURE_2D_ARRAY:
134 case TGSI_TEXTURE_2D_MSAA:
135 case TGSI_TEXTURE_3D:
136 case TGSI_TEXTURE_CUBE:
137 readmask = TGSI_WRITEMASK_XYZ;
138 break;
139 case TGSI_TEXTURE_SHADOW2D_ARRAY:
140 case TGSI_TEXTURE_SHADOWCUBE:
141 case TGSI_TEXTURE_2D_ARRAY_MSAA:
142 case TGSI_TEXTURE_CUBE_ARRAY:
143 readmask = TGSI_WRITEMASK_XYZW;
144 /* modifier would be in another not analyzed reg so just say indirect */
145 if (modifier != LP_BLD_TEX_MODIFIER_NONE) {
146 indirect = TRUE;
147 }
148 break;
149 case TGSI_TEXTURE_SHADOWCUBE_ARRAY:
150 readmask = TGSI_WRITEMASK_XYZW;
151 indirect = TRUE;
152 break;
153 default:
154 assert(0);
155 return;
156 }
157
158 if (modifier == LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV) {
159 /* We don't track explicit derivatives, although we could */
160 indirect = TRUE;
161 tex_info->sampler_unit = inst->Src[3].Register.Index;
162 tex_info->texture_unit = inst->Src[3].Register.Index;
163 } else {
164 if (modifier == LP_BLD_TEX_MODIFIER_PROJECTED ||
165 modifier == LP_BLD_TEX_MODIFIER_LOD_BIAS ||
166 modifier == LP_BLD_TEX_MODIFIER_EXPLICIT_LOD) {
167 readmask |= TGSI_WRITEMASK_W;
168 }
169 tex_info->sampler_unit = inst->Src[1].Register.Index;
170 tex_info->texture_unit = inst->Src[1].Register.Index;
171 }
172
173 for (chan = 0; chan < 4; ++chan) {
174 struct lp_tgsi_channel_info *chan_info = &tex_info->coord[chan];
175 if (readmask & (1 << chan)) {
176 analyse_src(ctx, chan_info, &inst->Src[0].Register, chan);
177 if (chan_info->file != TGSI_FILE_INPUT) {
178 indirect = TRUE;
179 }
180 } else {
181 memset(chan_info, 0, sizeof *chan_info);
182 }
183 }
184
185 if (indirect) {
186 info->indirect_textures = TRUE;
187 }
188
189 ++info->num_texs;
190 } else {
191 info->indirect_textures = TRUE;
192 }
193 }
194
195
196 /**
197 * Analyse properties of sample instructions, in particular used
198 * to figure out if a texture is considered indirect.
199 * Not actually used by much except the tgsi dumping code.
200 */
201 static void
analyse_sample(struct analysis_context * ctx,const struct tgsi_full_instruction * inst,enum lp_build_tex_modifier modifier,boolean shadow)202 analyse_sample(struct analysis_context *ctx,
203 const struct tgsi_full_instruction *inst,
204 enum lp_build_tex_modifier modifier,
205 boolean shadow)
206 {
207 struct lp_tgsi_info *info = ctx->info;
208 unsigned chan;
209
210 if (info->num_texs < ARRAY_SIZE(info->tex)) {
211 struct lp_tgsi_texture_info *tex_info = &info->tex[info->num_texs];
212 unsigned target = ctx->sample_target[inst->Src[1].Register.Index];
213 boolean indirect = FALSE;
214 boolean shadow = FALSE;
215 unsigned readmask;
216
217 switch (target) {
218 /* note no shadow targets here */
219 case TGSI_TEXTURE_BUFFER:
220 case TGSI_TEXTURE_1D:
221 readmask = TGSI_WRITEMASK_X;
222 break;
223 case TGSI_TEXTURE_1D_ARRAY:
224 case TGSI_TEXTURE_2D:
225 case TGSI_TEXTURE_RECT:
226 readmask = TGSI_WRITEMASK_XY;
227 break;
228 case TGSI_TEXTURE_2D_ARRAY:
229 case TGSI_TEXTURE_2D_MSAA:
230 case TGSI_TEXTURE_3D:
231 case TGSI_TEXTURE_CUBE:
232 readmask = TGSI_WRITEMASK_XYZ;
233 break;
234 case TGSI_TEXTURE_CUBE_ARRAY:
235 case TGSI_TEXTURE_2D_ARRAY_MSAA:
236 readmask = TGSI_WRITEMASK_XYZW;
237 break;
238 default:
239 assert(0);
240 return;
241 }
242
243 tex_info->target = target;
244 tex_info->texture_unit = inst->Src[1].Register.Index;
245 tex_info->sampler_unit = inst->Src[2].Register.Index;
246
247 if (tex_info->texture_unit != tex_info->sampler_unit) {
248 info->sampler_texture_units_different = TRUE;
249 }
250
251 if (modifier == LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV ||
252 modifier == LP_BLD_TEX_MODIFIER_EXPLICIT_LOD ||
253 modifier == LP_BLD_TEX_MODIFIER_LOD_BIAS || shadow) {
254 /* We don't track insts with additional regs, although we could */
255 indirect = TRUE;
256 }
257
258 for (chan = 0; chan < 4; ++chan) {
259 struct lp_tgsi_channel_info *chan_info = &tex_info->coord[chan];
260 if (readmask & (1 << chan)) {
261 analyse_src(ctx, chan_info, &inst->Src[0].Register, chan);
262 if (chan_info->file != TGSI_FILE_INPUT) {
263 indirect = TRUE;
264 }
265 } else {
266 memset(chan_info, 0, sizeof *chan_info);
267 }
268 }
269
270 if (indirect) {
271 info->indirect_textures = TRUE;
272 }
273
274 ++info->num_texs;
275 } else {
276 info->indirect_textures = TRUE;
277 }
278 }
279
280
281 /**
282 * Process an instruction, and update the register values accordingly.
283 */
284 static void
analyse_instruction(struct analysis_context * ctx,struct tgsi_full_instruction * inst)285 analyse_instruction(struct analysis_context *ctx,
286 struct tgsi_full_instruction *inst)
287 {
288 struct lp_tgsi_info *info = ctx->info;
289 struct lp_tgsi_channel_info (*regs)[4];
290 unsigned max_regs;
291 unsigned i;
292 unsigned index;
293 unsigned chan;
294
295 for (i = 0; i < inst->Instruction.NumDstRegs; ++i) {
296 const struct tgsi_dst_register *dst = &inst->Dst[i].Register;
297
298 /*
299 * Get the lp_tgsi_channel_info array corresponding to the destination
300 * register file.
301 */
302
303 if (dst->File == TGSI_FILE_TEMPORARY) {
304 regs = ctx->temp;
305 max_regs = ARRAY_SIZE(ctx->temp);
306 } else if (dst->File == TGSI_FILE_OUTPUT) {
307 regs = info->output;
308 max_regs = ARRAY_SIZE(info->output);
309 } else if (dst->File == TGSI_FILE_ADDRESS) {
310 continue;
311 } else if (dst->File == TGSI_FILE_BUFFER) {
312 continue;
313 } else if (dst->File == TGSI_FILE_IMAGE) {
314 continue;
315 } else if (dst->File == TGSI_FILE_MEMORY) {
316 continue;
317 } else {
318 assert(0);
319 continue;
320 }
321
322 /*
323 * Detect direct TEX instructions
324 */
325
326 switch (inst->Instruction.Opcode) {
327 case TGSI_OPCODE_TEX:
328 analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_NONE);
329 break;
330 case TGSI_OPCODE_TXD:
331 analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV);
332 break;
333 case TGSI_OPCODE_TXB:
334 analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_LOD_BIAS);
335 break;
336 case TGSI_OPCODE_TXL:
337 analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_EXPLICIT_LOD);
338 break;
339 case TGSI_OPCODE_TXP:
340 analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_PROJECTED);
341 break;
342 case TGSI_OPCODE_TEX2:
343 analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_NONE);
344 break;
345 case TGSI_OPCODE_TXB2:
346 analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_LOD_BIAS);
347 break;
348 case TGSI_OPCODE_TXL2:
349 analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_EXPLICIT_LOD);
350 break;
351 case TGSI_OPCODE_SAMPLE:
352 analyse_sample(ctx, inst, LP_BLD_TEX_MODIFIER_NONE, FALSE);
353 break;
354 case TGSI_OPCODE_SAMPLE_C:
355 analyse_sample(ctx, inst, LP_BLD_TEX_MODIFIER_NONE, TRUE);
356 break;
357 case TGSI_OPCODE_SAMPLE_C_LZ:
358 analyse_sample(ctx, inst, LP_BLD_TEX_MODIFIER_LOD_ZERO, TRUE);
359 break;
360 case TGSI_OPCODE_SAMPLE_D:
361 analyse_sample(ctx, inst, LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV, FALSE);
362 break;
363 case TGSI_OPCODE_SAMPLE_B:
364 analyse_sample(ctx, inst, LP_BLD_TEX_MODIFIER_LOD_BIAS, FALSE);
365 break;
366 case TGSI_OPCODE_SAMPLE_L:
367 analyse_sample(ctx, inst, LP_BLD_TEX_MODIFIER_EXPLICIT_LOD, FALSE);
368 break;
369 default:
370 break;
371 }
372
373 /*
374 * Keep track of assignments and writes
375 */
376
377 if (dst->Indirect) {
378 /*
379 * It could be any register index so clear all register indices.
380 */
381
382 for (chan = 0; chan < 4; ++chan) {
383 if (dst->WriteMask & (1 << chan)) {
384 for (index = 0; index < max_regs; ++index) {
385 regs[index][chan].file = TGSI_FILE_NULL;
386 }
387 }
388 }
389 } else if (dst->Index < max_regs) {
390 /*
391 * Update this destination register value.
392 */
393
394 struct lp_tgsi_channel_info res[4];
395
396 memset(res, 0, sizeof res);
397
398 if (!inst->Instruction.Saturate) {
399 for (chan = 0; chan < 4; ++chan) {
400 if (dst->WriteMask & (1 << chan)) {
401 if (inst->Instruction.Opcode == TGSI_OPCODE_MOV) {
402 analyse_src(ctx, &res[chan],
403 &inst->Src[0].Register, chan);
404 } else if (inst->Instruction.Opcode == TGSI_OPCODE_MUL) {
405 /*
406 * Propagate values across 1.0 and 0.0 multiplications.
407 */
408
409 struct lp_tgsi_channel_info src0;
410 struct lp_tgsi_channel_info src1;
411
412 analyse_src(ctx, &src0, &inst->Src[0].Register, chan);
413 analyse_src(ctx, &src1, &inst->Src[1].Register, chan);
414
415 if (is_immediate(&src0, 0.0f)) {
416 res[chan] = src0;
417 } else if (is_immediate(&src1, 0.0f)) {
418 res[chan] = src1;
419 } else if (is_immediate(&src0, 1.0f)) {
420 res[chan] = src1;
421 } else if (is_immediate(&src1, 1.0f)) {
422 res[chan] = src0;
423 }
424 }
425 }
426 }
427 }
428
429 for (chan = 0; chan < 4; ++chan) {
430 if (dst->WriteMask & (1 << chan)) {
431 regs[dst->Index][chan] = res[chan];
432 }
433 }
434 }
435 }
436
437 /*
438 * Clear all temporaries information in presence of a control flow opcode.
439 */
440
441 switch (inst->Instruction.Opcode) {
442 case TGSI_OPCODE_IF:
443 case TGSI_OPCODE_UIF:
444 case TGSI_OPCODE_ELSE:
445 case TGSI_OPCODE_ENDIF:
446 case TGSI_OPCODE_BGNLOOP:
447 case TGSI_OPCODE_BRK:
448 case TGSI_OPCODE_CONT:
449 case TGSI_OPCODE_ENDLOOP:
450 case TGSI_OPCODE_CAL:
451 case TGSI_OPCODE_BGNSUB:
452 case TGSI_OPCODE_ENDSUB:
453 case TGSI_OPCODE_SWITCH:
454 case TGSI_OPCODE_CASE:
455 case TGSI_OPCODE_DEFAULT:
456 case TGSI_OPCODE_ENDSWITCH:
457 case TGSI_OPCODE_RET:
458 case TGSI_OPCODE_END:
459 /* XXX: Are there more cases? */
460 memset(&ctx->temp, 0, sizeof ctx->temp);
461 memset(&info->output, 0, sizeof info->output);
462 FALLTHROUGH;
463 default:
464 break;
465 }
466 }
467
468
469 static inline void
dump_info(const struct tgsi_token * tokens,struct lp_tgsi_info * info)470 dump_info(const struct tgsi_token *tokens,
471 struct lp_tgsi_info *info)
472 {
473 unsigned index;
474 unsigned chan;
475
476 tgsi_dump(tokens, 0);
477
478 for (index = 0; index < info->num_texs; ++index) {
479 const struct lp_tgsi_texture_info *tex_info = &info->tex[index];
480 debug_printf("TEX[%u] =", index);
481 for (chan = 0; chan < 4; ++chan) {
482 const struct lp_tgsi_channel_info *chan_info =
483 &tex_info->coord[chan];
484 if (chan_info->file != TGSI_FILE_NULL) {
485 debug_printf(" %s[%u].%c",
486 tgsi_file_name(chan_info->file),
487 chan_info->u.index,
488 "xyzw01"[chan_info->swizzle]);
489 } else {
490 debug_printf(" _");
491 }
492 }
493 debug_printf(", RES[%u], SAMP[%u], %s\n",
494 tex_info->texture_unit,
495 tex_info->sampler_unit,
496 tgsi_texture_names[tex_info->target]);
497 }
498
499 for (index = 0; index < PIPE_MAX_SHADER_OUTPUTS; ++index) {
500 for (chan = 0; chan < 4; ++chan) {
501 const struct lp_tgsi_channel_info *chan_info =
502 &info->output[index][chan];
503 if (chan_info->file != TGSI_FILE_NULL) {
504 debug_printf("OUT[%u].%c = ", index, "xyzw"[chan]);
505 if (chan_info->file == TGSI_FILE_IMMEDIATE) {
506 debug_printf("%f", chan_info->u.value);
507 } else {
508 const char *file_name;
509 switch (chan_info->file) {
510 case TGSI_FILE_CONSTANT:
511 file_name = "CONST";
512 break;
513 case TGSI_FILE_INPUT:
514 file_name = "IN";
515 break;
516 default:
517 file_name = "???";
518 break;
519 }
520 debug_printf("%s[%u].%c",
521 file_name,
522 chan_info->u.index,
523 "xyzw01"[chan_info->swizzle]);
524 }
525 debug_printf("\n");
526 }
527 }
528 }
529 }
530
531
532 /**
533 * Detect any direct relationship between the output color
534 */
535 void
lp_build_tgsi_info(const struct tgsi_token * tokens,struct lp_tgsi_info * info)536 lp_build_tgsi_info(const struct tgsi_token *tokens,
537 struct lp_tgsi_info *info)
538 {
539 struct tgsi_parse_context parse;
540 struct analysis_context *ctx;
541 unsigned index;
542 unsigned chan;
543
544 memset(info, 0, sizeof *info);
545
546 tgsi_scan_shader(tokens, &info->base);
547
548 ctx = CALLOC(1, sizeof(struct analysis_context));
549 ctx->info = info;
550
551 tgsi_parse_init(&parse, tokens);
552
553 while (!tgsi_parse_end_of_tokens(&parse)) {
554 tgsi_parse_token(&parse);
555
556 switch (parse.FullToken.Token.Type) {
557 case TGSI_TOKEN_TYPE_DECLARATION: {
558 struct tgsi_full_declaration *decl = &parse.FullToken.FullDeclaration;
559 if (decl->Declaration.File == TGSI_FILE_SAMPLER_VIEW) {
560 for (index = decl->Range.First; index <= decl->Range.Last; index++) {
561 ctx->sample_target[index] = decl->SamplerView.Resource;
562 }
563 }
564 }
565 break;
566
567 case TGSI_TOKEN_TYPE_INSTRUCTION:
568 {
569 struct tgsi_full_instruction *inst =
570 &parse.FullToken.FullInstruction;
571
572 if (inst->Instruction.Opcode == TGSI_OPCODE_END ||
573 inst->Instruction.Opcode == TGSI_OPCODE_BGNSUB) {
574 /* We reached the end of main function body. */
575 goto finished;
576 }
577
578 analyse_instruction(ctx, inst);
579 }
580 break;
581
582 case TGSI_TOKEN_TYPE_IMMEDIATE:
583 {
584 const unsigned size =
585 parse.FullToken.FullImmediate.Immediate.NrTokens - 1;
586 assert(size <= 4);
587 if (ctx->num_imms < ARRAY_SIZE(ctx->imm)) {
588 for (chan = 0; chan < size; ++chan) {
589 float value = parse.FullToken.FullImmediate.u[chan].Float;
590 ctx->imm[ctx->num_imms][chan] = value;
591
592 if (value < 0.0f || value > 1.0f) {
593 info->unclamped_immediates = TRUE;
594 }
595 }
596 ++ctx->num_imms;
597 }
598 }
599 break;
600
601 case TGSI_TOKEN_TYPE_PROPERTY:
602 break;
603
604 default:
605 assert(0);
606 }
607 }
608 finished:
609
610 tgsi_parse_free(&parse);
611 FREE(ctx);
612
613
614 /*
615 * Link the output color values.
616 */
617
618 for (index = 0; index < PIPE_MAX_COLOR_BUFS; ++index) {
619 static const struct lp_tgsi_channel_info null_output[4];
620 info->cbuf[index] = null_output;
621 }
622
623 for (index = 0; index < info->base.num_outputs; ++index) {
624 unsigned semantic_name = info->base.output_semantic_name[index];
625 unsigned semantic_index = info->base.output_semantic_index[index];
626 if (semantic_name == TGSI_SEMANTIC_COLOR &&
627 semantic_index < PIPE_MAX_COLOR_BUFS) {
628 info->cbuf[semantic_index] = info->output[index];
629 }
630 }
631
632 if (gallivm_debug & GALLIVM_DEBUG_TGSI) {
633 dump_info(tokens, info);
634 }
635 }
636