i965/fs: Add an exec_size field to fs_inst
[mesa.git] / src / mesa / drivers / dri / i965 / brw_fs.h
1 /*
2 * Copyright © 2010 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 *
23 * Authors:
24 * Eric Anholt <eric@anholt.net>
25 *
26 */
27
28 #pragma once
29
30 #include "brw_shader.h"
31
32 extern "C" {
33
34 #include <sys/types.h>
35
36 #include "main/macros.h"
37 #include "main/shaderobj.h"
38 #include "main/uniforms.h"
39 #include "program/prog_parameter.h"
40 #include "program/prog_print.h"
41 #include "program/prog_optimize.h"
42 #include "util/register_allocate.h"
43 #include "program/sampler.h"
44 #include "program/hash_table.h"
45 #include "brw_context.h"
46 #include "brw_eu.h"
47 #include "brw_wm.h"
48 #include "brw_shader.h"
49 #include "intel_asm_annotation.h"
50 }
51 #include "glsl/glsl_types.h"
52 #include "glsl/ir.h"
53
54 #define MAX_SAMPLER_MESSAGE_SIZE 11
55
56 struct bblock_t;
57 namespace {
58 struct acp_entry;
59 }
60
61 namespace brw {
62 class fs_live_variables;
63 }
64
65 class fs_visitor;
66
67 class fs_reg : public backend_reg {
68 public:
69 DECLARE_RALLOC_CXX_OPERATORS(fs_reg)
70
71 void init();
72
73 fs_reg();
74 explicit fs_reg(float f);
75 explicit fs_reg(int32_t i);
76 explicit fs_reg(uint32_t u);
77 fs_reg(struct brw_reg fixed_hw_reg);
78 fs_reg(enum register_file file, int reg);
79 fs_reg(enum register_file file, int reg, enum brw_reg_type type);
80 fs_reg(enum register_file file, int reg, enum brw_reg_type type, uint8_t width);
81 fs_reg(fs_visitor *v, const struct glsl_type *type);
82
83 bool equals(const fs_reg &r) const;
84 bool is_valid_3src() const;
85 bool is_contiguous() const;
86
87 fs_reg &apply_stride(unsigned stride);
88 /** Smear a channel of the reg to all channels. */
89 fs_reg &set_smear(unsigned subreg);
90
91 /**
92 * Offset in bytes from the start of the register. Values up to a
93 * backend_reg::reg_offset unit are valid.
94 */
95 int subreg_offset;
96
97 fs_reg *reladdr;
98
99 /**
100 * The register width. This indicates how many hardware values are
101 * represented by each virtual value. Valid values are 1, 8, or 16.
102 * For immediate values, this is 1. Most of the rest of the time, it
103 * will be equal to the dispatch width.
104 */
105 uint8_t width;
106
107 /**
108 * Returns the effective register width when used as a source in the
109 * given instruction. Registers such as uniforms and immediates
110 * effectively take on the width of the instruction in which they are
111 * used.
112 */
113 uint8_t effective_width(const fs_visitor *v) const;
114
115 /** Register region horizontal stride */
116 uint8_t stride;
117 };
118
119 static inline fs_reg
120 retype(fs_reg reg, enum brw_reg_type type)
121 {
122 reg.fixed_hw_reg.type = reg.type = type;
123 return reg;
124 }
125
126 static inline fs_reg
127 byte_offset(fs_reg reg, unsigned delta)
128 {
129 switch (reg.file) {
130 case BAD_FILE:
131 break;
132 case GRF:
133 reg.reg_offset += delta / 32;
134 break;
135 case MRF:
136 reg.reg += delta / 32;
137 break;
138 default:
139 assert(delta == 0);
140 }
141 reg.subreg_offset += delta % 32;
142 return reg;
143 }
144
145 static inline fs_reg
146 offset(fs_reg reg, unsigned delta)
147 {
148 assert(reg.stride > 0);
149 switch (reg.file) {
150 case BAD_FILE:
151 break;
152 case GRF:
153 case MRF:
154 return byte_offset(reg, delta * reg.width * reg.stride * type_sz(reg.type));
155 case UNIFORM:
156 reg.reg_offset += delta;
157 break;
158 default:
159 assert(delta == 0);
160 }
161 return reg;
162 }
163
164 /**
165 * Get either of the 8-component halves of a 16-component register.
166 *
167 * Note: this also works if \c reg represents a SIMD16 pair of registers.
168 */
169 static inline fs_reg
170 half(fs_reg reg, unsigned idx)
171 {
172 assert(idx < 2);
173 assert(idx == 0 || (reg.file != HW_REG && reg.file != IMM));
174 assert(reg.width == 16);
175 reg.width = 8;
176 return byte_offset(reg, 8 * idx * reg.stride * type_sz(reg.type));
177 }
178
179 static const fs_reg reg_undef;
180
181 class ip_record : public exec_node {
182 public:
183 DECLARE_RALLOC_CXX_OPERATORS(ip_record)
184
185 ip_record(int ip)
186 {
187 this->ip = ip;
188 }
189
190 int ip;
191 };
192
193 class fs_inst : public backend_instruction {
194 fs_inst &operator=(const fs_inst &);
195
196 void init(enum opcode opcode, uint8_t exec_width, const fs_reg &dst,
197 fs_reg *src, int sources);
198
199 public:
200 DECLARE_RALLOC_CXX_OPERATORS(fs_inst)
201
202 fs_inst();
203 fs_inst(enum opcode opcode, uint8_t exec_size);
204 fs_inst(enum opcode opcode, const fs_reg &dst);
205 fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst,
206 const fs_reg &src0);
207 fs_inst(enum opcode opcode, const fs_reg &dst, const fs_reg &src0);
208 fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst,
209 const fs_reg &src0, const fs_reg &src1);
210 fs_inst(enum opcode opcode, const fs_reg &dst, const fs_reg &src0,
211 const fs_reg &src1);
212 fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst,
213 const fs_reg &src0, const fs_reg &src1, const fs_reg &src2);
214 fs_inst(enum opcode opcode, const fs_reg &dst, const fs_reg &src0,
215 const fs_reg &src1, const fs_reg &src2);
216 fs_inst(enum opcode opcode, const fs_reg &dst, fs_reg src[], int sources);
217 fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst,
218 fs_reg src[], int sources);
219 fs_inst(const fs_inst &that);
220
221 void resize_sources(uint8_t num_sources);
222
223 bool equals(fs_inst *inst) const;
224 bool overwrites_reg(const fs_reg &reg) const;
225 bool is_send_from_grf() const;
226 bool is_partial_write() const;
227 int regs_read(fs_visitor *v, int arg) const;
228 bool can_do_source_mods(struct brw_context *brw);
229
230 bool reads_flag() const;
231 bool writes_flag() const;
232
233 fs_reg dst;
234 fs_reg *src;
235
236 uint8_t sources; /**< Number of fs_reg sources. */
237
238 /**
239 * Execution size of the instruction. This is used by the generator to
240 * generate the correct binary for the given fs_inst. Current valid
241 * values are 1, 8, 16.
242 */
243 uint8_t exec_size;
244
245 /* Chooses which flag subregister (f0.0 or f0.1) is used for conditional
246 * mod and predication.
247 */
248 uint8_t flag_subreg;
249
250 uint8_t regs_written; /**< Number of vgrfs written by a SEND message, or 1 */
251 bool eot:1;
252 bool header_present:1;
253 bool shadow_compare:1;
254 bool force_uncompressed:1;
255 bool force_sechalf:1;
256 bool pi_noperspective:1; /**< Pixel interpolator noperspective flag */
257 };
258
259 /**
260 * The fragment shader front-end.
261 *
262 * Translates either GLSL IR or Mesa IR (for ARB_fragment_program) into FS IR.
263 */
264 class fs_visitor : public backend_visitor
265 {
266 public:
267 const fs_reg reg_null_f;
268 const fs_reg reg_null_d;
269 const fs_reg reg_null_ud;
270
271 fs_visitor(struct brw_context *brw,
272 void *mem_ctx,
273 const struct brw_wm_prog_key *key,
274 struct brw_wm_prog_data *prog_data,
275 struct gl_shader_program *shader_prog,
276 struct gl_fragment_program *fp,
277 unsigned dispatch_width);
278 ~fs_visitor();
279 void init();
280
281 fs_reg *variable_storage(ir_variable *var);
282 int virtual_grf_alloc(int size);
283 void import_uniforms(fs_visitor *v);
284
285 void visit(ir_variable *ir);
286 void visit(ir_assignment *ir);
287 void visit(ir_dereference_variable *ir);
288 void visit(ir_dereference_record *ir);
289 void visit(ir_dereference_array *ir);
290 void visit(ir_expression *ir);
291 void visit(ir_texture *ir);
292 void visit(ir_if *ir);
293 void visit(ir_constant *ir);
294 void visit(ir_swizzle *ir);
295 void visit(ir_return *ir);
296 void visit(ir_loop *ir);
297 void visit(ir_loop_jump *ir);
298 void visit(ir_discard *ir);
299 void visit(ir_call *ir);
300 void visit(ir_function *ir);
301 void visit(ir_function_signature *ir);
302 void visit(ir_emit_vertex *);
303 void visit(ir_end_primitive *);
304
305 uint32_t gather_channel(ir_texture *ir, uint32_t sampler);
306 void swizzle_result(ir_texture *ir, fs_reg orig_val, uint32_t sampler);
307
308 fs_inst *emit(fs_inst *inst);
309 void emit(exec_list list);
310
311 fs_inst *emit(enum opcode opcode);
312 fs_inst *emit(enum opcode opcode, const fs_reg &dst);
313 fs_inst *emit(enum opcode opcode, const fs_reg &dst, const fs_reg &src0);
314 fs_inst *emit(enum opcode opcode, const fs_reg &dst, const fs_reg &src0,
315 const fs_reg &src1);
316 fs_inst *emit(enum opcode opcode, const fs_reg &dst,
317 const fs_reg &src0, const fs_reg &src1, const fs_reg &src2);
318 fs_inst *emit(enum opcode opcode, const fs_reg &dst,
319 fs_reg src[], int sources);
320
321 fs_inst *MOV(const fs_reg &dst, const fs_reg &src);
322 fs_inst *NOT(const fs_reg &dst, const fs_reg &src);
323 fs_inst *RNDD(const fs_reg &dst, const fs_reg &src);
324 fs_inst *RNDE(const fs_reg &dst, const fs_reg &src);
325 fs_inst *RNDZ(const fs_reg &dst, const fs_reg &src);
326 fs_inst *FRC(const fs_reg &dst, const fs_reg &src);
327 fs_inst *ADD(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
328 fs_inst *MUL(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
329 fs_inst *MACH(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
330 fs_inst *MAC(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
331 fs_inst *SHL(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
332 fs_inst *SHR(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
333 fs_inst *ASR(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
334 fs_inst *AND(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
335 fs_inst *OR(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
336 fs_inst *XOR(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
337 fs_inst *IF(enum brw_predicate predicate);
338 fs_inst *IF(const fs_reg &src0, const fs_reg &src1,
339 enum brw_conditional_mod condition);
340 fs_inst *CMP(fs_reg dst, fs_reg src0, fs_reg src1,
341 enum brw_conditional_mod condition);
342 fs_inst *LRP(const fs_reg &dst, const fs_reg &a, const fs_reg &y,
343 const fs_reg &x);
344 fs_inst *DEP_RESOLVE_MOV(int grf);
345 fs_inst *BFREV(const fs_reg &dst, const fs_reg &value);
346 fs_inst *BFE(const fs_reg &dst, const fs_reg &bits, const fs_reg &offset,
347 const fs_reg &value);
348 fs_inst *BFI1(const fs_reg &dst, const fs_reg &bits, const fs_reg &offset);
349 fs_inst *BFI2(const fs_reg &dst, const fs_reg &bfi1_dst,
350 const fs_reg &insert, const fs_reg &base);
351 fs_inst *FBH(const fs_reg &dst, const fs_reg &value);
352 fs_inst *FBL(const fs_reg &dst, const fs_reg &value);
353 fs_inst *CBIT(const fs_reg &dst, const fs_reg &value);
354 fs_inst *MAD(const fs_reg &dst, const fs_reg &c, const fs_reg &b,
355 const fs_reg &a);
356 fs_inst *ADDC(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
357 fs_inst *SUBB(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
358 fs_inst *SEL(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
359
360 int type_size(const struct glsl_type *type);
361 fs_inst *get_instruction_generating_reg(fs_inst *start,
362 fs_inst *end,
363 const fs_reg &reg);
364
365 fs_inst *LOAD_PAYLOAD(const fs_reg &dst, fs_reg *src, int sources);
366
367 exec_list VARYING_PULL_CONSTANT_LOAD(const fs_reg &dst,
368 const fs_reg &surf_index,
369 const fs_reg &varying_offset,
370 uint32_t const_offset);
371
372 bool run();
373 void assign_binding_table_offsets();
374 void setup_payload_gen4();
375 void setup_payload_gen6();
376 void assign_curb_setup();
377 void calculate_urb_setup();
378 void assign_urb_setup();
379 bool assign_regs(bool allow_spilling);
380 void assign_regs_trivial();
381 void get_used_mrfs(bool *mrf_used);
382 void setup_payload_interference(struct ra_graph *g, int payload_reg_count,
383 int first_payload_node);
384 void setup_mrf_hack_interference(struct ra_graph *g,
385 int first_mrf_hack_node);
386 int choose_spill_reg(struct ra_graph *g);
387 void spill_reg(int spill_reg);
388 void split_virtual_grfs();
389 bool compact_virtual_grfs();
390 void move_uniform_array_access_to_pull_constants();
391 void assign_constant_locations();
392 void demote_pull_constants();
393 void invalidate_live_intervals();
394 void calculate_live_intervals();
395 void calculate_register_pressure();
396 bool opt_algebraic();
397 bool opt_cse();
398 bool opt_cse_local(bblock_t *block);
399 bool opt_copy_propagate();
400 bool try_copy_propagate(fs_inst *inst, int arg, acp_entry *entry);
401 bool try_constant_propagate(fs_inst *inst, acp_entry *entry);
402 bool opt_copy_propagate_local(void *mem_ctx, bblock_t *block,
403 exec_list *acp);
404 void opt_drop_redundant_mov_to_flags();
405 bool opt_register_renaming();
406 bool register_coalesce();
407 bool compute_to_mrf();
408 bool dead_code_eliminate();
409 bool remove_duplicate_mrf_writes();
410 bool virtual_grf_interferes(int a, int b);
411 void schedule_instructions(instruction_scheduler_mode mode);
412 void insert_gen4_send_dependency_workarounds();
413 void insert_gen4_pre_send_dependency_workarounds(bblock_t *block,
414 fs_inst *inst);
415 void insert_gen4_post_send_dependency_workarounds(bblock_t *block,
416 fs_inst *inst);
417 void vfail(const char *msg, va_list args);
418 void fail(const char *msg, ...);
419 void no16(const char *msg, ...);
420 void lower_uniform_pull_constant_loads();
421 bool lower_load_payload();
422
423 void push_force_uncompressed();
424 void pop_force_uncompressed();
425
426 void emit_dummy_fs();
427 void emit_repclear_shader();
428 fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
429 fs_inst *emit_linterp(const fs_reg &attr, const fs_reg &interp,
430 glsl_interp_qualifier interpolation_mode,
431 bool is_centroid, bool is_sample);
432 fs_reg *emit_frontfacing_interpolation();
433 fs_reg *emit_samplepos_setup();
434 fs_reg *emit_sampleid_setup(ir_variable *ir);
435 fs_reg *emit_general_interpolation(ir_variable *ir);
436 void emit_interpolation_setup_gen4();
437 void emit_interpolation_setup_gen6();
438 void compute_sample_position(fs_reg dst, fs_reg int_sample_pos);
439 fs_reg rescale_texcoord(ir_texture *ir, fs_reg coordinate,
440 bool is_rect, uint32_t sampler, int texunit);
441 fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate,
442 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
443 uint32_t sampler);
444 fs_inst *emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate,
445 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
446 fs_reg sample_index, uint32_t sampler);
447 fs_inst *emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
448 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
449 fs_reg sample_index, fs_reg mcs, fs_reg sampler);
450 fs_reg emit_mcs_fetch(ir_texture *ir, fs_reg coordinate, fs_reg sampler);
451 void emit_gen6_gather_wa(uint8_t wa, fs_reg dst);
452 fs_reg fix_math_operand(fs_reg src);
453 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0);
454 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0, fs_reg src1);
455 void emit_lrp(const fs_reg &dst, const fs_reg &x, const fs_reg &y,
456 const fs_reg &a);
457 void emit_minmax(enum brw_conditional_mod conditionalmod, const fs_reg &dst,
458 const fs_reg &src0, const fs_reg &src1);
459 bool try_emit_saturate(ir_expression *ir);
460 bool try_emit_mad(ir_expression *ir);
461 void try_replace_with_sel();
462 bool opt_peephole_sel();
463 bool opt_peephole_predicated_break();
464 bool opt_saturate_propagation();
465 void emit_bool_to_cond_code(ir_rvalue *condition);
466 void emit_if_gen6(ir_if *ir);
467 void emit_unspill(bblock_t *block, fs_inst *inst, fs_reg reg,
468 uint32_t spill_offset, int count);
469 void emit_spill(bblock_t *block, fs_inst *inst, fs_reg reg,
470 uint32_t spill_offset, int count);
471
472 void emit_fragment_program_code();
473 void setup_fp_regs();
474 fs_reg get_fp_src_reg(const prog_src_register *src);
475 fs_reg get_fp_dst_reg(const prog_dst_register *dst);
476 void emit_fp_alu1(enum opcode opcode,
477 const struct prog_instruction *fpi,
478 fs_reg dst, fs_reg src);
479 void emit_fp_alu2(enum opcode opcode,
480 const struct prog_instruction *fpi,
481 fs_reg dst, fs_reg src0, fs_reg src1);
482 void emit_fp_scalar_write(const struct prog_instruction *fpi,
483 fs_reg dst, fs_reg src);
484 void emit_fp_scalar_math(enum opcode opcode,
485 const struct prog_instruction *fpi,
486 fs_reg dst, fs_reg src);
487
488 void emit_fp_minmax(const struct prog_instruction *fpi,
489 fs_reg dst, fs_reg src0, fs_reg src1);
490
491 void emit_fp_sop(enum brw_conditional_mod conditional_mod,
492 const struct prog_instruction *fpi,
493 fs_reg dst, fs_reg src0, fs_reg src1, fs_reg one);
494
495 void emit_color_write(fs_reg color, int index, int first_color_mrf);
496 void emit_alpha_test();
497 fs_inst *emit_single_fb_write(fs_reg color1, fs_reg color2,
498 fs_reg src0_alpha, unsigned components);
499 void emit_fb_writes();
500
501 void emit_shader_time_begin();
502 void emit_shader_time_end();
503 void emit_shader_time_write(enum shader_time_shader_type type,
504 fs_reg value);
505
506 void emit_untyped_atomic(unsigned atomic_op, unsigned surf_index,
507 fs_reg dst, fs_reg offset, fs_reg src0,
508 fs_reg src1);
509
510 void emit_untyped_surface_read(unsigned surf_index, fs_reg dst,
511 fs_reg offset);
512
513 void emit_interpolate_expression(ir_expression *ir);
514
515 bool try_rewrite_rhs_to_dst(ir_assignment *ir,
516 fs_reg dst,
517 fs_reg src,
518 fs_inst *pre_rhs_inst,
519 fs_inst *last_rhs_inst);
520 void emit_assignment_writes(fs_reg &l, fs_reg &r,
521 const glsl_type *type, bool predicated);
522 void resolve_ud_negate(fs_reg *reg);
523 void resolve_bool_comparison(ir_rvalue *rvalue, fs_reg *reg);
524
525 fs_reg get_timestamp();
526
527 struct brw_reg interp_reg(int location, int channel);
528 void setup_uniform_values(ir_variable *ir);
529 void setup_builtin_uniform_values(ir_variable *ir);
530 int implied_mrf_writes(fs_inst *inst);
531
532 virtual void dump_instructions();
533 virtual void dump_instructions(const char *name);
534 void dump_instruction(backend_instruction *inst);
535 void dump_instruction(backend_instruction *inst, FILE *file);
536
537 void visit_atomic_counter_intrinsic(ir_call *ir);
538
539 const void *const key;
540 struct brw_stage_prog_data *prog_data;
541 unsigned int sanity_param_count;
542
543 int *param_size;
544
545 int *virtual_grf_sizes;
546 int virtual_grf_count;
547 int virtual_grf_array_size;
548 int *virtual_grf_start;
549 int *virtual_grf_end;
550 brw::fs_live_variables *live_intervals;
551
552 int *regs_live_at_ip;
553
554 /** Number of uniform variable components visited. */
555 unsigned uniforms;
556
557 /** Byte-offset for the next available spot in the scratch space buffer. */
558 unsigned last_scratch;
559
560 /**
561 * Array mapping UNIFORM register numbers to the pull parameter index,
562 * or -1 if this uniform register isn't being uploaded as a pull constant.
563 */
564 int *pull_constant_loc;
565
566 /**
567 * Array mapping UNIFORM register numbers to the push parameter index,
568 * or -1 if this uniform register isn't being uploaded as a push constant.
569 */
570 int *push_constant_loc;
571
572 struct hash_table *variable_ht;
573 fs_reg frag_depth;
574 fs_reg sample_mask;
575 fs_reg outputs[BRW_MAX_DRAW_BUFFERS];
576 unsigned output_components[BRW_MAX_DRAW_BUFFERS];
577 fs_reg dual_src_output;
578 bool do_dual_src;
579 int first_non_payload_grf;
580 /** Either BRW_MAX_GRF or GEN7_MRF_HACK_START */
581 int max_grf;
582
583 fs_reg *fp_temp_regs;
584 fs_reg *fp_input_regs;
585
586 /** @{ debug annotation info */
587 const char *current_annotation;
588 const void *base_ir;
589 /** @} */
590
591 bool failed;
592 char *fail_msg;
593 bool simd16_unsupported;
594 char *no16_msg;
595
596 /* Result of last visit() method. */
597 fs_reg result;
598
599 /** Register numbers for thread payload fields. */
600 struct {
601 uint8_t source_depth_reg;
602 uint8_t source_w_reg;
603 uint8_t aa_dest_stencil_reg;
604 uint8_t dest_depth_reg;
605 uint8_t sample_pos_reg;
606 uint8_t sample_mask_in_reg;
607 uint8_t barycentric_coord_reg[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
608
609 /** The number of thread payload registers the hardware will supply. */
610 uint8_t num_regs;
611 } payload;
612
613 bool source_depth_to_render_target;
614 bool runtime_check_aads_emit;
615
616 fs_reg pixel_x;
617 fs_reg pixel_y;
618 fs_reg wpos_w;
619 fs_reg pixel_w;
620 fs_reg delta_x[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
621 fs_reg delta_y[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
622 fs_reg shader_start_time;
623
624 int grf_used;
625 bool spilled_any_registers;
626
627 const unsigned dispatch_width; /**< 8 or 16 */
628
629 int force_uncompressed_stack;
630 };
631
632 /**
633 * The fragment shader code generator.
634 *
635 * Translates FS IR to actual i965 assembly code.
636 */
637 class fs_generator
638 {
639 public:
640 fs_generator(struct brw_context *brw,
641 void *mem_ctx,
642 const struct brw_wm_prog_key *key,
643 struct brw_wm_prog_data *prog_data,
644 struct gl_shader_program *shader_prog,
645 struct gl_fragment_program *fp,
646 bool runtime_check_aads_emit,
647 bool debug_flag);
648 ~fs_generator();
649
650 const unsigned *generate_assembly(const cfg_t *simd8_cfg,
651 const cfg_t *simd16_cfg,
652 unsigned *assembly_size);
653
654 private:
655 void generate_code(const cfg_t *cfg);
656 void fire_fb_write(fs_inst *inst,
657 GLuint base_reg,
658 struct brw_reg implied_header,
659 GLuint nr);
660 void generate_fb_write(fs_inst *inst);
661 void generate_blorp_fb_write(fs_inst *inst);
662 void generate_rep_fb_write(fs_inst *inst);
663 void generate_pixel_xy(struct brw_reg dst, bool is_x);
664 void generate_linterp(fs_inst *inst, struct brw_reg dst,
665 struct brw_reg *src);
666 void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
667 struct brw_reg sampler_index);
668 void generate_math_gen6(fs_inst *inst,
669 struct brw_reg dst,
670 struct brw_reg src0,
671 struct brw_reg src1);
672 void generate_math_gen4(fs_inst *inst,
673 struct brw_reg dst,
674 struct brw_reg src);
675 void generate_math_g45(fs_inst *inst,
676 struct brw_reg dst,
677 struct brw_reg src);
678 void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src, struct brw_reg quality);
679 void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
680 struct brw_reg quality, bool negate_value);
681 void generate_scratch_write(fs_inst *inst, struct brw_reg src);
682 void generate_scratch_read(fs_inst *inst, struct brw_reg dst);
683 void generate_scratch_read_gen7(fs_inst *inst, struct brw_reg dst);
684 void generate_uniform_pull_constant_load(fs_inst *inst, struct brw_reg dst,
685 struct brw_reg index,
686 struct brw_reg offset);
687 void generate_uniform_pull_constant_load_gen7(fs_inst *inst,
688 struct brw_reg dst,
689 struct brw_reg surf_index,
690 struct brw_reg offset);
691 void generate_varying_pull_constant_load(fs_inst *inst, struct brw_reg dst,
692 struct brw_reg index,
693 struct brw_reg offset);
694 void generate_varying_pull_constant_load_gen7(fs_inst *inst,
695 struct brw_reg dst,
696 struct brw_reg index,
697 struct brw_reg offset);
698 void generate_mov_dispatch_to_flags(fs_inst *inst);
699
700 void generate_pixel_interpolator_query(fs_inst *inst,
701 struct brw_reg dst,
702 struct brw_reg src,
703 struct brw_reg msg_data,
704 unsigned msg_type);
705
706 void generate_set_omask(fs_inst *inst,
707 struct brw_reg dst,
708 struct brw_reg sample_mask);
709
710 void generate_set_sample_id(fs_inst *inst,
711 struct brw_reg dst,
712 struct brw_reg src0,
713 struct brw_reg src1);
714
715 void generate_set_simd4x2_offset(fs_inst *inst,
716 struct brw_reg dst,
717 struct brw_reg offset);
718 void generate_discard_jump(fs_inst *inst);
719
720 void generate_pack_half_2x16_split(fs_inst *inst,
721 struct brw_reg dst,
722 struct brw_reg x,
723 struct brw_reg y);
724 void generate_unpack_half_2x16_split(fs_inst *inst,
725 struct brw_reg dst,
726 struct brw_reg src);
727
728 void generate_shader_time_add(fs_inst *inst,
729 struct brw_reg payload,
730 struct brw_reg offset,
731 struct brw_reg value);
732
733 void generate_untyped_atomic(fs_inst *inst,
734 struct brw_reg dst,
735 struct brw_reg atomic_op,
736 struct brw_reg surf_index);
737
738 void generate_untyped_surface_read(fs_inst *inst,
739 struct brw_reg dst,
740 struct brw_reg surf_index);
741
742 bool patch_discard_jumps_to_fb_writes();
743
744 struct brw_context *brw;
745 struct gl_context *ctx;
746
747 struct brw_compile *p;
748 gl_shader_stage stage;
749 const void * const key;
750 struct brw_stage_prog_data * const prog_data;
751
752 struct gl_shader_program * const shader_prog;
753 const struct gl_program *prog;
754
755 unsigned dispatch_width; /**< 8 or 16 */
756
757 exec_list discard_halt_patches;
758 bool runtime_check_aads_emit;
759 const bool debug_flag;
760 void *mem_ctx;
761 };
762
763 bool brw_do_channel_expressions(struct exec_list *instructions);
764 bool brw_do_vector_splitting(struct exec_list *instructions);
765 bool brw_fs_precompile(struct gl_context *ctx, struct gl_shader_program *prog);
766
767 struct brw_reg brw_reg_from_fs_reg(fs_reg *reg);