i965/fs: Add a saturation propagation optimization pass.
[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 "program/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 }
50 #include "gen8_generator.h"
51 #include "glsl/glsl_types.h"
52 #include "glsl/ir.h"
53
54 #define MAX_SAMPLER_MESSAGE_SIZE 11
55
56 class bblock_t;
57 namespace {
58 struct acp_entry;
59 }
60
61 namespace brw {
62 class fs_live_variables;
63 }
64
65 class fs_reg {
66 public:
67 DECLARE_RALLOC_CXX_OPERATORS(fs_reg)
68
69 void init();
70
71 fs_reg();
72 fs_reg(float f);
73 fs_reg(int32_t i);
74 fs_reg(uint32_t u);
75 fs_reg(struct brw_reg fixed_hw_reg);
76 fs_reg(enum register_file file, int reg);
77 fs_reg(enum register_file file, int reg, uint32_t type);
78 fs_reg(class fs_visitor *v, const struct glsl_type *type);
79
80 bool equals(const fs_reg &r) const;
81 bool is_zero() const;
82 bool is_one() const;
83 bool is_null() const;
84 bool is_valid_3src() const;
85 fs_reg retype(uint32_t type);
86
87 /** Register file: GRF, MRF, IMM. */
88 enum register_file file;
89 /**
90 * Register number. For MRF, it's the hardware register. For
91 * GRF, it's a virtual register number until register allocation
92 */
93 int reg;
94 /**
95 * Offset from the start of the contiguous register block.
96 *
97 * For pre-register-allocation GRFs, this is in units of a float per pixel
98 * (1 hardware register for SIMD8 mode, or 2 registers for SIMD16 mode).
99 * For uniforms, this is in units of 1 float.
100 */
101 int reg_offset;
102 /** Register type. BRW_REGISTER_TYPE_* */
103 int type;
104 bool negate;
105 bool abs;
106 bool sechalf;
107 struct brw_reg fixed_hw_reg;
108 int smear; /* -1, or a channel of the reg to smear to all channels. */
109
110 /** Value for file == IMM */
111 union {
112 int32_t i;
113 uint32_t u;
114 float f;
115 } imm;
116
117 fs_reg *reladdr;
118 };
119
120 static const fs_reg reg_undef;
121 static const fs_reg reg_null_f(retype(brw_null_reg(), BRW_REGISTER_TYPE_F));
122 static const fs_reg reg_null_d(retype(brw_null_reg(), BRW_REGISTER_TYPE_D));
123 static const fs_reg reg_null_ud(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
124
125 class ip_record : public exec_node {
126 public:
127 DECLARE_RALLOC_CXX_OPERATORS(ip_record)
128
129 ip_record(int ip)
130 {
131 this->ip = ip;
132 }
133
134 int ip;
135 };
136
137 class fs_inst : public backend_instruction {
138 public:
139 DECLARE_RALLOC_CXX_OPERATORS(fs_inst)
140
141 void init();
142
143 fs_inst();
144 fs_inst(enum opcode opcode);
145 fs_inst(enum opcode opcode, fs_reg dst);
146 fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0);
147 fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1);
148 fs_inst(enum opcode opcode, fs_reg dst,
149 fs_reg src0, fs_reg src1,fs_reg src2);
150
151 bool equals(fs_inst *inst);
152 bool overwrites_reg(const fs_reg &reg);
153 bool is_send_from_grf();
154 bool is_partial_write();
155 int regs_read(fs_visitor *v, int arg);
156
157 bool reads_flag();
158 bool writes_flag();
159
160 fs_reg dst;
161 fs_reg src[3];
162 bool saturate;
163 int conditional_mod; /**< BRW_CONDITIONAL_* */
164
165 /* Chooses which flag subregister (f0.0 or f0.1) is used for conditional
166 * mod and predication.
167 */
168 uint8_t flag_subreg;
169
170 int mlen; /**< SEND message length */
171 int regs_written; /**< Number of vgrfs written by a SEND message, or 1 */
172 int base_mrf; /**< First MRF in the SEND message, if mlen is nonzero. */
173 uint32_t texture_offset; /**< Texture offset bitfield */
174 int sampler;
175 int target; /**< MRT target. */
176 bool eot;
177 bool header_present;
178 bool shadow_compare;
179 bool force_uncompressed;
180 bool force_sechalf;
181 bool force_writemask_all;
182 uint32_t offset; /* spill/unspill offset */
183
184 /** @{
185 * Annotation for the generated IR. One of the two can be set.
186 */
187 const void *ir;
188 const char *annotation;
189 /** @} */
190 };
191
192 /**
193 * The fragment shader front-end.
194 *
195 * Translates either GLSL IR or Mesa IR (for ARB_fragment_program) into FS IR.
196 */
197 class fs_visitor : public backend_visitor
198 {
199 public:
200
201 fs_visitor(struct brw_context *brw,
202 struct brw_wm_compile *c,
203 struct gl_shader_program *shader_prog,
204 struct gl_fragment_program *fp,
205 unsigned dispatch_width);
206 ~fs_visitor();
207
208 fs_reg *variable_storage(ir_variable *var);
209 int virtual_grf_alloc(int size);
210 void import_uniforms(fs_visitor *v);
211
212 void visit(ir_variable *ir);
213 void visit(ir_assignment *ir);
214 void visit(ir_dereference_variable *ir);
215 void visit(ir_dereference_record *ir);
216 void visit(ir_dereference_array *ir);
217 void visit(ir_expression *ir);
218 void visit(ir_texture *ir);
219 void visit(ir_if *ir);
220 void visit(ir_constant *ir);
221 void visit(ir_swizzle *ir);
222 void visit(ir_return *ir);
223 void visit(ir_loop *ir);
224 void visit(ir_loop_jump *ir);
225 void visit(ir_discard *ir);
226 void visit(ir_call *ir);
227 void visit(ir_function *ir);
228 void visit(ir_function_signature *ir);
229 void visit(ir_emit_vertex *);
230 void visit(ir_end_primitive *);
231
232 uint32_t gather_channel(ir_texture *ir, int sampler);
233 void swizzle_result(ir_texture *ir, fs_reg orig_val, int sampler);
234
235 bool can_do_source_mods(fs_inst *inst);
236
237 fs_inst *emit(fs_inst inst);
238 fs_inst *emit(fs_inst *inst);
239 void emit(exec_list list);
240
241 fs_inst *emit(enum opcode opcode);
242 fs_inst *emit(enum opcode opcode, fs_reg dst);
243 fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0);
244 fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1);
245 fs_inst *emit(enum opcode opcode, fs_reg dst,
246 fs_reg src0, fs_reg src1, fs_reg src2);
247
248 fs_inst *MOV(fs_reg dst, fs_reg src);
249 fs_inst *NOT(fs_reg dst, fs_reg src);
250 fs_inst *RNDD(fs_reg dst, fs_reg src);
251 fs_inst *RNDE(fs_reg dst, fs_reg src);
252 fs_inst *RNDZ(fs_reg dst, fs_reg src);
253 fs_inst *FRC(fs_reg dst, fs_reg src);
254 fs_inst *ADD(fs_reg dst, fs_reg src0, fs_reg src1);
255 fs_inst *MUL(fs_reg dst, fs_reg src0, fs_reg src1);
256 fs_inst *MACH(fs_reg dst, fs_reg src0, fs_reg src1);
257 fs_inst *MAC(fs_reg dst, fs_reg src0, fs_reg src1);
258 fs_inst *SHL(fs_reg dst, fs_reg src0, fs_reg src1);
259 fs_inst *SHR(fs_reg dst, fs_reg src0, fs_reg src1);
260 fs_inst *ASR(fs_reg dst, fs_reg src0, fs_reg src1);
261 fs_inst *AND(fs_reg dst, fs_reg src0, fs_reg src1);
262 fs_inst *OR(fs_reg dst, fs_reg src0, fs_reg src1);
263 fs_inst *XOR(fs_reg dst, fs_reg src0, fs_reg src1);
264 fs_inst *IF(uint32_t predicate);
265 fs_inst *IF(fs_reg src0, fs_reg src1, uint32_t condition);
266 fs_inst *CMP(fs_reg dst, fs_reg src0, fs_reg src1,
267 uint32_t condition);
268 fs_inst *LRP(fs_reg dst, fs_reg a, fs_reg y, fs_reg x);
269 fs_inst *DEP_RESOLVE_MOV(int grf);
270 fs_inst *BFREV(fs_reg dst, fs_reg value);
271 fs_inst *BFE(fs_reg dst, fs_reg bits, fs_reg offset, fs_reg value);
272 fs_inst *BFI1(fs_reg dst, fs_reg bits, fs_reg offset);
273 fs_inst *BFI2(fs_reg dst, fs_reg bfi1_dst, fs_reg insert, fs_reg base);
274 fs_inst *FBH(fs_reg dst, fs_reg value);
275 fs_inst *FBL(fs_reg dst, fs_reg value);
276 fs_inst *CBIT(fs_reg dst, fs_reg value);
277 fs_inst *MAD(fs_reg dst, fs_reg c, fs_reg b, fs_reg a);
278 fs_inst *ADDC(fs_reg dst, fs_reg src0, fs_reg src1);
279 fs_inst *SUBB(fs_reg dst, fs_reg src0, fs_reg src1);
280 fs_inst *SEL(fs_reg dst, fs_reg src0, fs_reg src1);
281
282 int type_size(const struct glsl_type *type);
283 fs_inst *get_instruction_generating_reg(fs_inst *start,
284 fs_inst *end,
285 fs_reg reg);
286
287 exec_list VARYING_PULL_CONSTANT_LOAD(fs_reg dst, fs_reg surf_index,
288 fs_reg varying_offset,
289 uint32_t const_offset);
290
291 bool run();
292 void assign_binding_table_offsets();
293 void setup_payload_gen4();
294 void setup_payload_gen6();
295 void assign_curb_setup();
296 void calculate_urb_setup();
297 void assign_urb_setup();
298 bool assign_regs(bool allow_spilling);
299 void assign_regs_trivial();
300 void get_used_mrfs(bool *mrf_used);
301 void setup_payload_interference(struct ra_graph *g, int payload_reg_count,
302 int first_payload_node);
303 void setup_mrf_hack_interference(struct ra_graph *g,
304 int first_mrf_hack_node);
305 int choose_spill_reg(struct ra_graph *g);
306 void spill_reg(int spill_reg);
307 void split_virtual_grfs();
308 void compact_virtual_grfs();
309 void move_uniform_array_access_to_pull_constants();
310 void setup_pull_constants();
311 void invalidate_live_intervals();
312 void calculate_live_intervals();
313 void calculate_register_pressure();
314 bool opt_algebraic();
315 bool opt_cse();
316 bool opt_cse_local(bblock_t *block, exec_list *aeb);
317 bool opt_copy_propagate();
318 bool try_copy_propagate(fs_inst *inst, int arg, acp_entry *entry);
319 bool try_constant_propagate(fs_inst *inst, acp_entry *entry);
320 bool opt_copy_propagate_local(void *mem_ctx, bblock_t *block,
321 exec_list *acp);
322 bool register_coalesce();
323 bool compute_to_mrf();
324 bool dead_code_eliminate();
325 bool dead_code_eliminate_local();
326 bool remove_dead_constants();
327 bool remove_duplicate_mrf_writes();
328 bool virtual_grf_interferes(int a, int b);
329 void schedule_instructions(instruction_scheduler_mode mode);
330 void insert_gen4_send_dependency_workarounds();
331 void insert_gen4_pre_send_dependency_workarounds(fs_inst *inst);
332 void insert_gen4_post_send_dependency_workarounds(fs_inst *inst);
333 void fail(const char *msg, ...);
334 void lower_uniform_pull_constant_loads();
335
336 void push_force_uncompressed();
337 void pop_force_uncompressed();
338
339 void emit_dummy_fs();
340 fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
341 fs_inst *emit_linterp(const fs_reg &attr, const fs_reg &interp,
342 glsl_interp_qualifier interpolation_mode,
343 bool is_centroid, bool is_sample);
344 fs_reg *emit_frontfacing_interpolation(ir_variable *ir);
345 fs_reg *emit_samplepos_setup(ir_variable *ir);
346 fs_reg *emit_sampleid_setup(ir_variable *ir);
347 fs_reg *emit_samplemaskin_setup(ir_variable *ir);
348 fs_reg *emit_general_interpolation(ir_variable *ir);
349 void emit_interpolation_setup_gen4();
350 void emit_interpolation_setup_gen6();
351 void compute_sample_position(fs_reg dst, fs_reg int_sample_pos);
352 fs_reg rescale_texcoord(ir_texture *ir, fs_reg coordinate,
353 bool is_rect, int sampler, int texunit);
354 fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate,
355 fs_reg shadow_comp, fs_reg lod, fs_reg lod2);
356 fs_inst *emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate,
357 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
358 fs_reg sample_index);
359 fs_inst *emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
360 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
361 fs_reg sample_index, fs_reg mcs, int sampler);
362 fs_reg emit_mcs_fetch(ir_texture *ir, fs_reg coordinate, int sampler);
363 fs_reg fix_math_operand(fs_reg src);
364 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0);
365 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0, fs_reg src1);
366 void emit_lrp(fs_reg dst, fs_reg x, fs_reg y, fs_reg a);
367 void emit_minmax(uint32_t conditionalmod, fs_reg dst,
368 fs_reg src0, fs_reg src1);
369 bool try_emit_saturate(ir_expression *ir);
370 bool try_emit_mad(ir_expression *ir, int mul_arg);
371 void try_replace_with_sel();
372 bool opt_peephole_sel();
373 bool opt_peephole_predicated_break();
374 bool opt_saturate_propagation();
375 void emit_bool_to_cond_code(ir_rvalue *condition);
376 void emit_if_gen6(ir_if *ir);
377 void emit_unspill(fs_inst *inst, fs_reg reg, uint32_t spill_offset,
378 int count);
379
380 void emit_fragment_program_code();
381 void setup_fp_regs();
382 fs_reg get_fp_src_reg(const prog_src_register *src);
383 fs_reg get_fp_dst_reg(const prog_dst_register *dst);
384 void emit_fp_alu1(enum opcode opcode,
385 const struct prog_instruction *fpi,
386 fs_reg dst, fs_reg src);
387 void emit_fp_alu2(enum opcode opcode,
388 const struct prog_instruction *fpi,
389 fs_reg dst, fs_reg src0, fs_reg src1);
390 void emit_fp_scalar_write(const struct prog_instruction *fpi,
391 fs_reg dst, fs_reg src);
392 void emit_fp_scalar_math(enum opcode opcode,
393 const struct prog_instruction *fpi,
394 fs_reg dst, fs_reg src);
395
396 void emit_fp_minmax(const struct prog_instruction *fpi,
397 fs_reg dst, fs_reg src0, fs_reg src1);
398
399 void emit_fp_sop(uint32_t conditional_mod,
400 const struct prog_instruction *fpi,
401 fs_reg dst, fs_reg src0, fs_reg src1, fs_reg one);
402
403 void emit_color_write(int target, int index, int first_color_mrf);
404 void emit_alpha_test();
405 void emit_fb_writes();
406
407 void emit_shader_time_begin();
408 void emit_shader_time_end();
409 void emit_shader_time_write(enum shader_time_shader_type type,
410 fs_reg value);
411
412 void emit_untyped_atomic(unsigned atomic_op, unsigned surf_index,
413 fs_reg dst, fs_reg offset, fs_reg src0,
414 fs_reg src1);
415
416 void emit_untyped_surface_read(unsigned surf_index, fs_reg dst,
417 fs_reg offset);
418
419 bool try_rewrite_rhs_to_dst(ir_assignment *ir,
420 fs_reg dst,
421 fs_reg src,
422 fs_inst *pre_rhs_inst,
423 fs_inst *last_rhs_inst);
424 void emit_assignment_writes(fs_reg &l, fs_reg &r,
425 const glsl_type *type, bool predicated);
426 void resolve_ud_negate(fs_reg *reg);
427 void resolve_bool_comparison(ir_rvalue *rvalue, fs_reg *reg);
428
429 fs_reg get_timestamp();
430
431 struct brw_reg interp_reg(int location, int channel);
432 void setup_uniform_values(ir_variable *ir);
433 void setup_builtin_uniform_values(ir_variable *ir);
434 int implied_mrf_writes(fs_inst *inst);
435
436 virtual void dump_instructions();
437 void dump_instruction(backend_instruction *inst);
438
439 void visit_atomic_counter_intrinsic(ir_call *ir);
440
441 struct gl_fragment_program *fp;
442 struct brw_wm_compile *c;
443 unsigned int sanity_param_count;
444
445 int param_size[MAX_UNIFORMS * 4];
446
447 int *virtual_grf_sizes;
448 int virtual_grf_count;
449 int virtual_grf_array_size;
450 int *virtual_grf_start;
451 int *virtual_grf_end;
452 brw::fs_live_variables *live_intervals;
453
454 int *regs_live_at_ip;
455
456 /* This is the map from UNIFORM hw_reg + reg_offset as generated by
457 * the visitor to the packed uniform number after
458 * remove_dead_constants() that represents the actual uploaded
459 * uniform index.
460 */
461 int *params_remap;
462 int nr_params_remap;
463
464 struct hash_table *variable_ht;
465 fs_reg frag_depth;
466 fs_reg sample_mask;
467 fs_reg outputs[BRW_MAX_DRAW_BUFFERS];
468 unsigned output_components[BRW_MAX_DRAW_BUFFERS];
469 fs_reg dual_src_output;
470 int first_non_payload_grf;
471 /** Either BRW_MAX_GRF or GEN7_MRF_HACK_START */
472 int max_grf;
473
474 fs_reg *fp_temp_regs;
475 fs_reg *fp_input_regs;
476
477 /** @{ debug annotation info */
478 const char *current_annotation;
479 const void *base_ir;
480 /** @} */
481
482 bool failed;
483 char *fail_msg;
484
485 /* Result of last visit() method. */
486 fs_reg result;
487
488 fs_reg pixel_x;
489 fs_reg pixel_y;
490 fs_reg wpos_w;
491 fs_reg pixel_w;
492 fs_reg delta_x[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
493 fs_reg delta_y[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
494 fs_reg shader_start_time;
495
496 int grf_used;
497 bool spilled_any_registers;
498
499 const unsigned dispatch_width; /**< 8 or 16 */
500
501 int force_uncompressed_stack;
502 };
503
504 /**
505 * The fragment shader code generator.
506 *
507 * Translates FS IR to actual i965 assembly code.
508 */
509 class fs_generator
510 {
511 public:
512 fs_generator(struct brw_context *brw,
513 struct brw_wm_compile *c,
514 struct gl_shader_program *prog,
515 struct gl_fragment_program *fp,
516 bool dual_source_output);
517 ~fs_generator();
518
519 const unsigned *generate_assembly(exec_list *simd8_instructions,
520 exec_list *simd16_instructions,
521 unsigned *assembly_size,
522 FILE *dump_file = NULL);
523
524 private:
525 void generate_code(exec_list *instructions, FILE *dump_file);
526 void generate_fb_write(fs_inst *inst);
527 void generate_blorp_fb_write(fs_inst *inst);
528 void generate_pixel_xy(struct brw_reg dst, bool is_x);
529 void generate_linterp(fs_inst *inst, struct brw_reg dst,
530 struct brw_reg *src);
531 void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
532 void generate_math1_gen7(fs_inst *inst,
533 struct brw_reg dst,
534 struct brw_reg src);
535 void generate_math2_gen7(fs_inst *inst,
536 struct brw_reg dst,
537 struct brw_reg src0,
538 struct brw_reg src1);
539 void generate_math1_gen6(fs_inst *inst,
540 struct brw_reg dst,
541 struct brw_reg src);
542 void generate_math2_gen6(fs_inst *inst,
543 struct brw_reg dst,
544 struct brw_reg src0,
545 struct brw_reg src1);
546 void generate_math_gen4(fs_inst *inst,
547 struct brw_reg dst,
548 struct brw_reg src);
549 void generate_math_g45(fs_inst *inst,
550 struct brw_reg dst,
551 struct brw_reg src);
552 void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
553 void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
554 bool negate_value);
555 void generate_scratch_write(fs_inst *inst, struct brw_reg src);
556 void generate_scratch_read(fs_inst *inst, struct brw_reg dst);
557 void generate_scratch_read_gen7(fs_inst *inst, struct brw_reg dst);
558 void generate_uniform_pull_constant_load(fs_inst *inst, struct brw_reg dst,
559 struct brw_reg index,
560 struct brw_reg offset);
561 void generate_uniform_pull_constant_load_gen7(fs_inst *inst,
562 struct brw_reg dst,
563 struct brw_reg surf_index,
564 struct brw_reg offset);
565 void generate_varying_pull_constant_load(fs_inst *inst, struct brw_reg dst,
566 struct brw_reg index,
567 struct brw_reg offset);
568 void generate_varying_pull_constant_load_gen7(fs_inst *inst,
569 struct brw_reg dst,
570 struct brw_reg index,
571 struct brw_reg offset);
572 void generate_mov_dispatch_to_flags(fs_inst *inst);
573
574 void generate_set_omask(fs_inst *inst,
575 struct brw_reg dst,
576 struct brw_reg sample_mask);
577
578 void generate_set_sample_id(fs_inst *inst,
579 struct brw_reg dst,
580 struct brw_reg src0,
581 struct brw_reg src1);
582
583 void generate_set_simd4x2_offset(fs_inst *inst,
584 struct brw_reg dst,
585 struct brw_reg offset);
586 void generate_discard_jump(fs_inst *inst);
587
588 void generate_pack_half_2x16_split(fs_inst *inst,
589 struct brw_reg dst,
590 struct brw_reg x,
591 struct brw_reg y);
592 void generate_unpack_half_2x16_split(fs_inst *inst,
593 struct brw_reg dst,
594 struct brw_reg src);
595
596 void generate_shader_time_add(fs_inst *inst,
597 struct brw_reg payload,
598 struct brw_reg offset,
599 struct brw_reg value);
600
601 void generate_untyped_atomic(fs_inst *inst,
602 struct brw_reg dst,
603 struct brw_reg atomic_op,
604 struct brw_reg surf_index);
605
606 void generate_untyped_surface_read(fs_inst *inst,
607 struct brw_reg dst,
608 struct brw_reg surf_index);
609
610 void mark_surface_used(unsigned surf_index);
611
612 void patch_discard_jumps_to_fb_writes();
613
614 struct brw_context *brw;
615 struct gl_context *ctx;
616
617 struct brw_compile *p;
618 struct brw_wm_compile *c;
619
620 struct gl_shader_program *prog;
621 const struct gl_fragment_program *fp;
622
623 unsigned dispatch_width; /**< 8 or 16 */
624
625 exec_list discard_halt_patches;
626 bool dual_source_output;
627 void *mem_ctx;
628 };
629
630 /**
631 * The fragment shader code generator.
632 *
633 * Translates FS IR to actual i965 assembly code.
634 */
635 class gen8_fs_generator : public gen8_generator
636 {
637 public:
638 gen8_fs_generator(struct brw_context *brw,
639 struct brw_wm_compile *c,
640 struct gl_shader_program *prog,
641 struct gl_fragment_program *fp,
642 bool dual_source_output);
643 ~gen8_fs_generator();
644
645 const unsigned *generate_assembly(exec_list *simd8_instructions,
646 exec_list *simd16_instructions,
647 unsigned *assembly_size);
648
649 private:
650 void generate_code(exec_list *instructions);
651 void generate_fb_write(fs_inst *inst);
652 void generate_linterp(fs_inst *inst, struct brw_reg dst,
653 struct brw_reg *src);
654 void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
655 void generate_math1(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
656 void generate_math2(fs_inst *inst, struct brw_reg dst,
657 struct brw_reg src0, struct brw_reg src1);
658 void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
659 void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
660 bool negate_value);
661 void generate_scratch_write(fs_inst *inst, struct brw_reg src);
662 void generate_scratch_read(fs_inst *inst, struct brw_reg dst);
663 void generate_scratch_read_gen7(fs_inst *inst, struct brw_reg dst);
664 void generate_uniform_pull_constant_load(fs_inst *inst,
665 struct brw_reg dst,
666 struct brw_reg index,
667 struct brw_reg offset);
668 void generate_varying_pull_constant_load(fs_inst *inst,
669 struct brw_reg dst,
670 struct brw_reg index,
671 struct brw_reg offset);
672 void generate_mov_dispatch_to_flags(fs_inst *ir);
673 void generate_set_simd4x2_offset(fs_inst *ir,
674 struct brw_reg dst,
675 struct brw_reg offset);
676 void generate_discard_jump(fs_inst *ir);
677
678 void patch_discard_jumps_to_fb_writes();
679
680 void mark_surface_used(unsigned surf_index);
681
682 struct brw_wm_compile *c;
683 const struct gl_fragment_program *fp;
684
685 unsigned dispatch_width; /** 8 or 16 */
686
687 bool dual_source_output;
688
689 exec_list discard_halt_patches;
690 };
691
692 bool brw_do_channel_expressions(struct exec_list *instructions);
693 bool brw_do_vector_splitting(struct exec_list *instructions);
694 bool brw_fs_precompile(struct gl_context *ctx, struct gl_shader_program *prog);
695
696 struct brw_reg brw_reg_from_fs_reg(fs_reg *reg);