llvmpipe: do constant buffer bounds checking in shaders
[mesa.git] / src / gallium / auxiliary / gallivm / lp_bld_tgsi.h
1 /**************************************************************************
2 *
3 * Copyright 2011-2012 Advanced Micro Devices, Inc.
4 * Copyright 2009 VMware, Inc.
5 * All Rights Reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
14 *
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
17 * of the Software.
18 *
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 *
27 **************************************************************************/
28
29 /**
30 * @file
31 * TGSI to LLVM IR translation.
32 *
33 * @author Jose Fonseca <jfonseca@vmware.com>
34 * @author Tom Stellard <thomas.stellard@amd.com>
35 */
36
37 #ifndef LP_BLD_TGSI_H
38 #define LP_BLD_TGSI_H
39
40 #include "gallivm/lp_bld.h"
41 #include "gallivm/lp_bld_tgsi_action.h"
42 #include "gallivm/lp_bld_limits.h"
43 #include "gallivm/lp_bld_sample.h"
44 #include "lp_bld_type.h"
45 #include "pipe/p_compiler.h"
46 #include "pipe/p_state.h"
47 #include "tgsi/tgsi_exec.h"
48 #include "tgsi/tgsi_scan.h"
49 #include "tgsi/tgsi_info.h"
50
51 #define LP_CHAN_ALL ~0
52
53 #define LP_MAX_INSTRUCTIONS 256
54
55 struct tgsi_full_declaration;
56 struct tgsi_full_immediate;
57 struct tgsi_full_instruction;
58 struct tgsi_full_src_register;
59 struct tgsi_opcode_info;
60 struct tgsi_token;
61 struct tgsi_shader_info;
62 struct lp_build_mask_context;
63 struct gallivm_state;
64 struct lp_derivatives;
65 struct lp_build_tgsi_gs_iface;
66
67
68 enum lp_build_tex_modifier {
69 LP_BLD_TEX_MODIFIER_NONE = 0,
70 LP_BLD_TEX_MODIFIER_PROJECTED,
71 LP_BLD_TEX_MODIFIER_LOD_BIAS,
72 LP_BLD_TEX_MODIFIER_EXPLICIT_LOD,
73 LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV,
74 LP_BLD_TEX_MODIFIER_LOD_ZERO
75 };
76
77
78 /**
79 * Describe a channel of a register.
80 *
81 * The value can be a:
82 * - immediate value (i.e. derived from a IMM register)
83 * - CONST[n].x/y/z/w
84 * - IN[n].x/y/z/w
85 * - undetermined (when .file == TGSI_FILE_NULL)
86 *
87 * This is one of the analysis results, and is used to described
88 * the output color in terms of inputs.
89 */
90 struct lp_tgsi_channel_info
91 {
92 unsigned file:4; /* TGSI_FILE_* */
93 unsigned swizzle:3; /* PIPE_SWIZZLE_x */
94 union {
95 uint32_t index;
96 float value; /* for TGSI_FILE_IMMEDIATE */
97 } u;
98 };
99
100
101 /**
102 * Describe a texture sampler interpolator.
103 *
104 * The interpolation is described in terms of regular inputs.
105 */
106 struct lp_tgsi_texture_info
107 {
108 struct lp_tgsi_channel_info coord[4];
109 unsigned target:8; /* TGSI_TEXTURE_* */
110 unsigned sampler_unit:8; /* Sampler unit */
111 unsigned texture_unit:8; /* Texture unit */
112 unsigned modifier:8; /* LP_BLD_TEX_MODIFIER_* */
113 };
114
115
116 struct lp_tgsi_info
117 {
118 struct tgsi_shader_info base;
119
120 /*
121 * Whether any of the texture opcodes access a register file other than
122 * TGSI_FILE_INPUT.
123 *
124 * We could also handle TGSI_FILE_CONST/IMMEDIATE here, but there is little
125 * benefit.
126 */
127 unsigned indirect_textures:1;
128
129 /*
130 * Whether any immediate values are outside the range of 0 and 1
131 */
132 unsigned unclamped_immediates:1;
133
134 /*
135 * Texture opcode description. Aimed at detecting and described direct
136 * texture opcodes.
137 */
138 unsigned num_texs;
139 struct lp_tgsi_texture_info tex[PIPE_MAX_SAMPLERS];
140
141 /*
142 * Output description. Aimed at detecting and describing simple blit
143 * shaders.
144 */
145 struct lp_tgsi_channel_info output[PIPE_MAX_SHADER_OUTPUTS][4];
146
147 /*
148 * Shortcut pointers into the above (for fragment shaders).
149 */
150 const struct lp_tgsi_channel_info *cbuf[PIPE_MAX_COLOR_BUFS];
151 };
152
153 /**
154 * Reference to system values.
155 */
156 struct lp_bld_tgsi_system_values {
157 LLVMValueRef instance_id;
158 LLVMValueRef vertex_id;
159 LLVMValueRef prim_id;
160 };
161
162
163 /**
164 * Sampler code generation interface.
165 *
166 * Although texture sampling is a requirement for TGSI translation, it is
167 * a very different problem with several different approaches to it. This
168 * structure establishes an interface for texture sampling code generation, so
169 * that we can easily use different texture sampling strategies.
170 */
171 struct lp_build_sampler_soa
172 {
173 void
174 (*destroy)( struct lp_build_sampler_soa *sampler );
175
176 void
177 (*emit_fetch_texel)( const struct lp_build_sampler_soa *sampler,
178 struct gallivm_state *gallivm,
179 struct lp_type type,
180 boolean is_fetch,
181 unsigned texture_index,
182 unsigned sampler_index,
183 const LLVMValueRef *coords,
184 const LLVMValueRef *offsets,
185 const struct lp_derivatives *derivs,
186 LLVMValueRef lod_bias, /* optional */
187 LLVMValueRef explicit_lod, /* optional */
188 enum lp_sampler_lod_property,
189 LLVMValueRef *texel);
190
191 void
192 (*emit_size_query)( const struct lp_build_sampler_soa *sampler,
193 struct gallivm_state *gallivm,
194 struct lp_type type,
195 unsigned unit,
196 unsigned target,
197 boolean need_nr_mips,
198 enum lp_sampler_lod_property,
199 LLVMValueRef explicit_lod, /* optional */
200 LLVMValueRef *sizes_out);
201 };
202
203
204 struct lp_build_sampler_aos
205 {
206 LLVMValueRef
207 (*emit_fetch_texel)( struct lp_build_sampler_aos *sampler,
208 struct lp_build_context *bld,
209 unsigned target, /* TGSI_TEXTURE_* */
210 unsigned unit,
211 LLVMValueRef coords,
212 const struct lp_derivatives derivs,
213 enum lp_build_tex_modifier modifier);
214 };
215
216
217 void
218 lp_build_tgsi_info(const struct tgsi_token *tokens,
219 struct lp_tgsi_info *info);
220
221
222 void
223 lp_build_tgsi_soa(struct gallivm_state *gallivm,
224 const struct tgsi_token *tokens,
225 struct lp_type type,
226 struct lp_build_mask_context *mask,
227 LLVMValueRef consts_ptr,
228 LLVMValueRef const_sizes_ptr,
229 const struct lp_bld_tgsi_system_values *system_values,
230 const LLVMValueRef (*inputs)[4],
231 LLVMValueRef (*outputs)[4],
232 struct lp_build_sampler_soa *sampler,
233 const struct tgsi_shader_info *info,
234 const struct lp_build_tgsi_gs_iface *gs_iface);
235
236
237 void
238 lp_build_tgsi_aos(struct gallivm_state *gallivm,
239 const struct tgsi_token *tokens,
240 struct lp_type type,
241 const unsigned char swizzles[4],
242 LLVMValueRef consts_ptr,
243 const LLVMValueRef *inputs,
244 LLVMValueRef *outputs,
245 struct lp_build_sampler_aos *sampler,
246 const struct tgsi_shader_info *info);
247
248
249 enum lp_exec_mask_break_type {
250 LP_EXEC_MASK_BREAK_TYPE_LOOP,
251 LP_EXEC_MASK_BREAK_TYPE_SWITCH
252 };
253
254
255 struct lp_exec_mask {
256 struct lp_build_context *bld;
257
258 boolean has_mask;
259 boolean ret_in_main;
260
261 LLVMTypeRef int_vec_type;
262
263 LLVMValueRef cond_stack[LP_MAX_TGSI_NESTING];
264 int cond_stack_size;
265 LLVMValueRef cond_mask;
266
267 /* keep track if break belongs to switch or loop */
268 enum lp_exec_mask_break_type break_type_stack[LP_MAX_TGSI_NESTING];
269 enum lp_exec_mask_break_type break_type;
270
271 struct {
272 LLVMValueRef switch_val;
273 LLVMValueRef switch_mask;
274 LLVMValueRef switch_mask_default;
275 boolean switch_in_default;
276 unsigned switch_pc;
277 } switch_stack[LP_MAX_TGSI_NESTING];
278 int switch_stack_size;
279 LLVMValueRef switch_val;
280 LLVMValueRef switch_mask; /* current switch exec mask */
281 LLVMValueRef switch_mask_default; /* reverse of switch mask used for default */
282 boolean switch_in_default; /* if switch exec is currently in default */
283 unsigned switch_pc; /* when used points to default or endswitch-1 */
284
285 LLVMBasicBlockRef loop_block;
286 LLVMValueRef cont_mask;
287 LLVMValueRef break_mask;
288 LLVMValueRef break_var;
289 struct {
290 LLVMBasicBlockRef loop_block;
291 LLVMValueRef cont_mask;
292 LLVMValueRef break_mask;
293 LLVMValueRef break_var;
294 } loop_stack[LP_MAX_TGSI_NESTING];
295 int loop_stack_size;
296
297 LLVMValueRef ret_mask;
298 struct {
299 int pc;
300 LLVMValueRef ret_mask;
301 } call_stack[LP_MAX_TGSI_NESTING];
302 int call_stack_size;
303
304 LLVMValueRef exec_mask;
305 LLVMValueRef loop_limiter;
306 };
307
308 struct lp_build_tgsi_inst_list
309 {
310 struct tgsi_full_instruction *instructions;
311 uint max_instructions;
312 uint num_instructions;
313 };
314
315 unsigned lp_bld_tgsi_list_init(struct lp_build_tgsi_context * bld_base);
316
317
318 unsigned lp_bld_tgsi_add_instruction(
319 struct lp_build_tgsi_context * bld_base,
320 struct tgsi_full_instruction *inst_to_add);
321
322
323 struct lp_build_tgsi_context;
324
325
326 typedef LLVMValueRef (*lp_build_emit_fetch_fn)(struct lp_build_tgsi_context *,
327 const struct tgsi_full_src_register *,
328 enum tgsi_opcode_type,
329 unsigned);
330
331 struct lp_build_tgsi_context
332 {
333 struct lp_build_context base;
334
335 struct lp_build_context uint_bld;
336 struct lp_build_context int_bld;
337
338 /** This array stores functions that are used to transform TGSI opcodes to
339 * LLVM instructions.
340 */
341 struct lp_build_tgsi_action op_actions[TGSI_OPCODE_LAST];
342
343 /* TGSI_OPCODE_RSQ is defined as 1 / sqrt( abs(src0.x) ), rsq_action
344 * should compute 1 / sqrt (src0.x) */
345 struct lp_build_tgsi_action rsq_action;
346
347 struct lp_build_tgsi_action sqrt_action;
348
349 const struct tgsi_shader_info *info;
350
351 lp_build_emit_fetch_fn emit_fetch_funcs[TGSI_FILE_COUNT];
352
353 LLVMValueRef (*emit_swizzle)(struct lp_build_tgsi_context *,
354 LLVMValueRef, unsigned, unsigned, unsigned, unsigned);
355
356
357 void (*emit_debug)(struct lp_build_tgsi_context *,
358 const struct tgsi_full_instruction *,
359 const struct tgsi_opcode_info *);
360
361 void (*emit_store)(struct lp_build_tgsi_context *,
362 const struct tgsi_full_instruction *,
363 const struct tgsi_opcode_info *,
364 LLVMValueRef dst[4]);
365
366 void (*emit_declaration)(struct lp_build_tgsi_context *,
367 const struct tgsi_full_declaration *decl);
368
369 void (*emit_immediate)(struct lp_build_tgsi_context *,
370 const struct tgsi_full_immediate *imm);
371
372
373 /* Allow the user to store data in this structure rather than passing it
374 * to every function. */
375 void * userdata;
376
377 boolean soa;
378
379 int pc;
380
381 struct tgsi_full_instruction *instructions;
382 uint max_instructions;
383 uint num_instructions;
384
385 /** This function allows the user to insert some instructions at the
386 * beginning of the program. It is optional and does not need to be
387 * implemented.
388 */
389 void (*emit_prologue)(struct lp_build_tgsi_context*);
390
391 /** This function allows the user to insert some instructions at the end of
392 * the program. This callback is intended to be used for emitting
393 * instructions to handle the export for the output registers, but it can
394 * be used for any purpose. Implementing this function is optiona, but
395 * recommended.
396 */
397 void (*emit_epilogue)(struct lp_build_tgsi_context*);
398 };
399
400 struct lp_build_tgsi_gs_iface
401 {
402 LLVMValueRef (*fetch_input)(const struct lp_build_tgsi_gs_iface *gs_iface,
403 struct lp_build_tgsi_context * bld_base,
404 boolean is_vindex_indirect,
405 LLVMValueRef vertex_index,
406 boolean is_aindex_indirect,
407 LLVMValueRef attrib_index,
408 LLVMValueRef swizzle_index);
409 void (*emit_vertex)(const struct lp_build_tgsi_gs_iface *gs_iface,
410 struct lp_build_tgsi_context * bld_base,
411 LLVMValueRef (*outputs)[4],
412 LLVMValueRef emitted_vertices_vec);
413 void (*end_primitive)(const struct lp_build_tgsi_gs_iface *gs_iface,
414 struct lp_build_tgsi_context * bld_base,
415 LLVMValueRef verts_per_prim_vec,
416 LLVMValueRef emitted_prims_vec);
417 void (*gs_epilogue)(const struct lp_build_tgsi_gs_iface *gs_iface,
418 struct lp_build_tgsi_context * bld_base,
419 LLVMValueRef total_emitted_vertices_vec,
420 LLVMValueRef emitted_prims_vec);
421 };
422
423 struct lp_build_tgsi_soa_context
424 {
425 struct lp_build_tgsi_context bld_base;
426
427 /* Builder for scalar elements of shader's data type (float) */
428 struct lp_build_context elem_bld;
429
430 const struct lp_build_tgsi_gs_iface *gs_iface;
431 LLVMValueRef emitted_prims_vec_ptr;
432 LLVMValueRef total_emitted_vertices_vec_ptr;
433 LLVMValueRef emitted_vertices_vec_ptr;
434 LLVMValueRef max_output_vertices_vec;
435
436 LLVMValueRef consts_ptr;
437 LLVMValueRef const_sizes_ptr;
438 const LLVMValueRef (*inputs)[TGSI_NUM_CHANNELS];
439 LLVMValueRef (*outputs)[TGSI_NUM_CHANNELS];
440
441 const struct lp_build_sampler_soa *sampler;
442
443 struct tgsi_declaration_sampler_view sv[PIPE_MAX_SHADER_SAMPLER_VIEWS];
444
445 LLVMValueRef immediates[LP_MAX_TGSI_IMMEDIATES][TGSI_NUM_CHANNELS];
446 LLVMValueRef temps[LP_MAX_TGSI_TEMPS][TGSI_NUM_CHANNELS];
447 LLVMValueRef addr[LP_MAX_TGSI_ADDRS][TGSI_NUM_CHANNELS];
448 LLVMValueRef preds[LP_MAX_TGSI_PREDS][TGSI_NUM_CHANNELS];
449
450 /* We allocate/use this array of temps if (1 << TGSI_FILE_TEMPORARY) is
451 * set in the indirect_files field.
452 * The temps[] array above is unused then.
453 */
454 LLVMValueRef temps_array;
455
456 /* We allocate/use this array of output if (1 << TGSI_FILE_OUTPUT) is
457 * set in the indirect_files field.
458 * The outputs[] array above is unused then.
459 */
460 LLVMValueRef outputs_array;
461
462 /* We allocate/use this array of inputs if (1 << TGSI_FILE_INPUT) is
463 * set in the indirect_files field.
464 * The inputs[] array above is unused then.
465 */
466 LLVMValueRef inputs_array;
467
468 /* We allocate/use this array of temps if (1 << TGSI_FILE_IMMEDIATE) is
469 * set in the indirect_files field.
470 */
471 LLVMValueRef imms_array;
472
473
474 struct lp_bld_tgsi_system_values system_values;
475
476 /** bitmask indicating which register files are accessed indirectly */
477 unsigned indirect_files;
478
479 struct lp_build_mask_context *mask;
480 struct lp_exec_mask exec_mask;
481
482 uint num_immediates;
483
484 };
485
486 void
487 lp_emit_declaration_soa(
488 struct lp_build_tgsi_context *bld,
489 const struct tgsi_full_declaration *decl);
490
491 void lp_emit_immediate_soa(
492 struct lp_build_tgsi_context *bld_base,
493 const struct tgsi_full_immediate *imm);
494
495 boolean
496 lp_emit_instruction_soa(
497 struct lp_build_tgsi_soa_context *bld,
498 const struct tgsi_full_instruction *inst,
499 const struct tgsi_opcode_info *info);
500
501
502 LLVMValueRef
503 lp_get_temp_ptr_soa(
504 struct lp_build_tgsi_soa_context *bld,
505 unsigned index,
506 unsigned chan);
507
508 LLVMValueRef
509 lp_get_output_ptr(
510 struct lp_build_tgsi_soa_context *bld,
511 unsigned index,
512 unsigned chan);
513
514 struct lp_build_tgsi_aos_context
515 {
516 struct lp_build_tgsi_context bld_base;
517
518 /* Builder for integer masks and indices */
519 struct lp_build_context int_bld;
520
521 /*
522 * AoS swizzle used:
523 * - swizzles[0] = red index
524 * - swizzles[1] = green index
525 * - swizzles[2] = blue index
526 * - swizzles[3] = alpha index
527 */
528 unsigned char swizzles[4];
529 unsigned char inv_swizzles[4];
530
531 LLVMValueRef consts_ptr;
532 const LLVMValueRef *inputs;
533 LLVMValueRef *outputs;
534
535 struct lp_build_sampler_aos *sampler;
536
537 LLVMValueRef immediates[LP_MAX_TGSI_IMMEDIATES];
538 LLVMValueRef temps[LP_MAX_TGSI_TEMPS];
539 LLVMValueRef addr[LP_MAX_TGSI_ADDRS];
540 LLVMValueRef preds[LP_MAX_TGSI_PREDS];
541
542 /* We allocate/use this array of temps if (1 << TGSI_FILE_TEMPORARY) is
543 * set in the indirect_files field.
544 * The temps[] array above is unused then.
545 */
546 LLVMValueRef temps_array;
547
548 /** bitmask indicating which register files are accessed indirectly */
549 unsigned indirect_files;
550
551 };
552
553 static INLINE struct lp_build_tgsi_soa_context *
554 lp_soa_context(struct lp_build_tgsi_context *bld_base)
555 {
556 return (struct lp_build_tgsi_soa_context *)bld_base;
557 }
558
559 static INLINE struct lp_build_tgsi_aos_context *
560 lp_aos_context(struct lp_build_tgsi_context *bld_base)
561 {
562 return (struct lp_build_tgsi_aos_context *)bld_base;
563 }
564
565 void
566 lp_emit_declaration_aos(
567 struct lp_build_tgsi_aos_context *bld,
568 const struct tgsi_full_declaration *decl);
569
570
571 boolean
572 lp_emit_instruction_aos(
573 struct lp_build_tgsi_aos_context *bld,
574 const struct tgsi_full_instruction *inst,
575 const struct tgsi_opcode_info *info,
576 int *pc);
577
578 void
579 lp_emit_store_aos(
580 struct lp_build_tgsi_aos_context *bld,
581 const struct tgsi_full_instruction *inst,
582 unsigned index,
583 LLVMValueRef value);
584
585 void lp_build_fetch_args(
586 struct lp_build_tgsi_context * bld_base,
587 struct lp_build_emit_data * emit_data);
588
589 LLVMValueRef
590 lp_build_tgsi_inst_llvm_aos(
591 struct lp_build_tgsi_context * bld_base,
592 const struct tgsi_full_instruction *inst);
593
594 void
595 lp_build_tgsi_intrinsic(
596 const struct lp_build_tgsi_action * action,
597 struct lp_build_tgsi_context * bld_base,
598 struct lp_build_emit_data * emit_data);
599
600 LLVMValueRef
601 lp_build_emit_llvm(
602 struct lp_build_tgsi_context *bld_base,
603 unsigned tgsi_opcode,
604 struct lp_build_emit_data * emit_data);
605
606 LLVMValueRef
607 lp_build_emit_llvm_unary(
608 struct lp_build_tgsi_context *bld_base,
609 unsigned tgsi_opcode,
610 LLVMValueRef arg0);
611
612 LLVMValueRef
613 lp_build_emit_llvm_binary(
614 struct lp_build_tgsi_context *bld_base,
615 unsigned tgsi_opcode,
616 LLVMValueRef arg0,
617 LLVMValueRef arg1);
618
619 LLVMValueRef
620 lp_build_emit_llvm_ternary(
621 struct lp_build_tgsi_context *bld_base,
622 unsigned tgsi_opcode,
623 LLVMValueRef arg0,
624 LLVMValueRef arg1,
625 LLVMValueRef arg2);
626
627 boolean
628 lp_build_tgsi_inst_llvm(
629 struct lp_build_tgsi_context * bld_base,
630 const struct tgsi_full_instruction *inst);
631
632 LLVMValueRef
633 lp_build_emit_fetch(
634 struct lp_build_tgsi_context *bld_base,
635 const struct tgsi_full_instruction *inst,
636 unsigned src_op,
637 const unsigned chan_index);
638
639
640 LLVMValueRef
641 lp_build_emit_fetch_texoffset(
642 struct lp_build_tgsi_context *bld_base,
643 const struct tgsi_full_instruction *inst,
644 unsigned tex_off_op,
645 const unsigned chan_index);
646
647 boolean
648 lp_build_tgsi_llvm(
649 struct lp_build_tgsi_context * bld_base,
650 const struct tgsi_token *tokens);
651
652 #endif /* LP_BLD_TGSI_H */