i965: Drop mark_surface_used from gen8 generators.
[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 bool is_contiguous() const;
86 bool is_accumulator() const;
87
88 fs_reg &apply_stride(unsigned stride);
89 /** Smear a channel of the reg to all channels. */
90 fs_reg &set_smear(unsigned subreg);
91
92 /** Register file: GRF, MRF, IMM. */
93 enum register_file file;
94 /** Register type. BRW_REGISTER_TYPE_* */
95 uint8_t type;
96 /**
97 * Register number. For MRF, it's the hardware register. For
98 * GRF, it's a virtual register number until register allocation
99 */
100 uint16_t reg;
101 /**
102 * Offset from the start of the contiguous register block.
103 *
104 * For pre-register-allocation GRFs, this is in units of a float per pixel
105 * (1 hardware register for SIMD8 mode, or 2 registers for SIMD16 mode).
106 * For uniforms, this is in units of 1 float.
107 */
108 int reg_offset;
109 /**
110 * Offset in bytes from the start of the register. Values up to a
111 * backend_reg::reg_offset unit are valid.
112 */
113 int subreg_offset;
114
115 /** Value for file == IMM */
116 union {
117 int32_t i;
118 uint32_t u;
119 float f;
120 } imm;
121
122 struct brw_reg fixed_hw_reg;
123
124 fs_reg *reladdr;
125
126 bool negate;
127 bool abs;
128
129 /** Register region horizontal stride */
130 uint8_t stride;
131 };
132
133 static inline fs_reg
134 retype(fs_reg reg, unsigned type)
135 {
136 reg.fixed_hw_reg.type = reg.type = type;
137 return reg;
138 }
139
140 static inline fs_reg
141 offset(fs_reg reg, unsigned delta)
142 {
143 assert(delta == 0 || (reg.file != HW_REG && reg.file != IMM));
144 reg.reg_offset += delta;
145 return reg;
146 }
147
148 static inline fs_reg
149 byte_offset(fs_reg reg, unsigned delta)
150 {
151 assert(delta == 0 || (reg.file != HW_REG && reg.file != IMM));
152 reg.subreg_offset += delta;
153 return reg;
154 }
155
156 /**
157 * Get either of the 8-component halves of a 16-component register.
158 *
159 * Note: this also works if \c reg represents a SIMD16 pair of registers.
160 */
161 static inline fs_reg
162 half(const fs_reg &reg, unsigned idx)
163 {
164 assert(idx < 2);
165 assert(idx == 0 || (reg.file != HW_REG && reg.file != IMM));
166 return byte_offset(reg, 8 * idx * reg.stride * type_sz(reg.type));
167 }
168
169 static const fs_reg reg_undef;
170 static const fs_reg reg_null_f(retype(brw_null_reg(), BRW_REGISTER_TYPE_F));
171 static const fs_reg reg_null_d(retype(brw_null_reg(), BRW_REGISTER_TYPE_D));
172 static const fs_reg reg_null_ud(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
173
174 class ip_record : public exec_node {
175 public:
176 DECLARE_RALLOC_CXX_OPERATORS(ip_record)
177
178 ip_record(int ip)
179 {
180 this->ip = ip;
181 }
182
183 int ip;
184 };
185
186 class fs_inst : public backend_instruction {
187 public:
188 DECLARE_RALLOC_CXX_OPERATORS(fs_inst)
189
190 void init();
191
192 fs_inst();
193 fs_inst(enum opcode opcode);
194 fs_inst(enum opcode opcode, fs_reg dst);
195 fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0);
196 fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1);
197 fs_inst(enum opcode opcode, fs_reg dst,
198 fs_reg src0, fs_reg src1,fs_reg src2);
199
200 bool equals(fs_inst *inst) const;
201 bool overwrites_reg(const fs_reg &reg) const;
202 bool is_send_from_grf() const;
203 bool is_partial_write() const;
204 int regs_read(fs_visitor *v, int arg) const;
205
206 bool reads_flag() const;
207 bool writes_flag() const;
208
209 fs_reg dst;
210 fs_reg src[3];
211
212 /** @{
213 * Annotation for the generated IR. One of the two can be set.
214 */
215 const void *ir;
216 const char *annotation;
217 /** @} */
218
219 uint32_t texture_offset; /**< Texture offset bitfield */
220 uint32_t offset; /* spill/unspill offset */
221
222 uint8_t conditional_mod; /**< BRW_CONDITIONAL_* */
223
224 /* Chooses which flag subregister (f0.0 or f0.1) is used for conditional
225 * mod and predication.
226 */
227 uint8_t flag_subreg;
228
229 uint8_t mlen; /**< SEND message length */
230 uint8_t regs_written; /**< Number of vgrfs written by a SEND message, or 1 */
231 int8_t base_mrf; /**< First MRF in the SEND message, if mlen is nonzero. */
232 uint8_t sampler;
233 uint8_t target; /**< MRT target. */
234 bool saturate:1;
235 bool eot:1;
236 bool header_present:1;
237 bool shadow_compare:1;
238 bool force_uncompressed:1;
239 bool force_sechalf:1;
240 bool force_writemask_all:1;
241 };
242
243 /**
244 * The fragment shader front-end.
245 *
246 * Translates either GLSL IR or Mesa IR (for ARB_fragment_program) into FS IR.
247 */
248 class fs_visitor : public backend_visitor
249 {
250 public:
251
252 fs_visitor(struct brw_context *brw,
253 struct brw_wm_compile *c,
254 struct gl_shader_program *shader_prog,
255 struct gl_fragment_program *fp,
256 unsigned dispatch_width);
257 ~fs_visitor();
258
259 fs_reg *variable_storage(ir_variable *var);
260 int virtual_grf_alloc(int size);
261 void import_uniforms(fs_visitor *v);
262
263 void visit(ir_variable *ir);
264 void visit(ir_assignment *ir);
265 void visit(ir_dereference_variable *ir);
266 void visit(ir_dereference_record *ir);
267 void visit(ir_dereference_array *ir);
268 void visit(ir_expression *ir);
269 void visit(ir_texture *ir);
270 void visit(ir_if *ir);
271 void visit(ir_constant *ir);
272 void visit(ir_swizzle *ir);
273 void visit(ir_return *ir);
274 void visit(ir_loop *ir);
275 void visit(ir_loop_jump *ir);
276 void visit(ir_discard *ir);
277 void visit(ir_call *ir);
278 void visit(ir_function *ir);
279 void visit(ir_function_signature *ir);
280 void visit(ir_emit_vertex *);
281 void visit(ir_end_primitive *);
282
283 uint32_t gather_channel(ir_texture *ir, int sampler);
284 void swizzle_result(ir_texture *ir, fs_reg orig_val, int sampler);
285
286 bool can_do_source_mods(fs_inst *inst);
287
288 fs_inst *emit(fs_inst *inst);
289 void emit(exec_list list);
290
291 fs_inst *emit(enum opcode opcode);
292 fs_inst *emit(enum opcode opcode, fs_reg dst);
293 fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0);
294 fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1);
295 fs_inst *emit(enum opcode opcode, fs_reg dst,
296 fs_reg src0, fs_reg src1, fs_reg src2);
297
298 fs_inst *MOV(fs_reg dst, fs_reg src);
299 fs_inst *NOT(fs_reg dst, fs_reg src);
300 fs_inst *RNDD(fs_reg dst, fs_reg src);
301 fs_inst *RNDE(fs_reg dst, fs_reg src);
302 fs_inst *RNDZ(fs_reg dst, fs_reg src);
303 fs_inst *FRC(fs_reg dst, fs_reg src);
304 fs_inst *ADD(fs_reg dst, fs_reg src0, fs_reg src1);
305 fs_inst *MUL(fs_reg dst, fs_reg src0, fs_reg src1);
306 fs_inst *MACH(fs_reg dst, fs_reg src0, fs_reg src1);
307 fs_inst *MAC(fs_reg dst, fs_reg src0, fs_reg src1);
308 fs_inst *SHL(fs_reg dst, fs_reg src0, fs_reg src1);
309 fs_inst *SHR(fs_reg dst, fs_reg src0, fs_reg src1);
310 fs_inst *ASR(fs_reg dst, fs_reg src0, fs_reg src1);
311 fs_inst *AND(fs_reg dst, fs_reg src0, fs_reg src1);
312 fs_inst *OR(fs_reg dst, fs_reg src0, fs_reg src1);
313 fs_inst *XOR(fs_reg dst, fs_reg src0, fs_reg src1);
314 fs_inst *IF(uint32_t predicate);
315 fs_inst *IF(fs_reg src0, fs_reg src1, uint32_t condition);
316 fs_inst *CMP(fs_reg dst, fs_reg src0, fs_reg src1,
317 uint32_t condition);
318 fs_inst *LRP(fs_reg dst, fs_reg a, fs_reg y, fs_reg x);
319 fs_inst *DEP_RESOLVE_MOV(int grf);
320 fs_inst *BFREV(fs_reg dst, fs_reg value);
321 fs_inst *BFE(fs_reg dst, fs_reg bits, fs_reg offset, fs_reg value);
322 fs_inst *BFI1(fs_reg dst, fs_reg bits, fs_reg offset);
323 fs_inst *BFI2(fs_reg dst, fs_reg bfi1_dst, fs_reg insert, fs_reg base);
324 fs_inst *FBH(fs_reg dst, fs_reg value);
325 fs_inst *FBL(fs_reg dst, fs_reg value);
326 fs_inst *CBIT(fs_reg dst, fs_reg value);
327 fs_inst *MAD(fs_reg dst, fs_reg c, fs_reg b, fs_reg a);
328 fs_inst *ADDC(fs_reg dst, fs_reg src0, fs_reg src1);
329 fs_inst *SUBB(fs_reg dst, fs_reg src0, fs_reg src1);
330 fs_inst *SEL(fs_reg dst, fs_reg src0, fs_reg src1);
331
332 int type_size(const struct glsl_type *type);
333 fs_inst *get_instruction_generating_reg(fs_inst *start,
334 fs_inst *end,
335 const fs_reg &reg);
336
337 exec_list VARYING_PULL_CONSTANT_LOAD(const fs_reg &dst,
338 const fs_reg &surf_index,
339 const fs_reg &varying_offset,
340 uint32_t const_offset);
341
342 bool run();
343 void assign_binding_table_offsets();
344 void setup_payload_gen4();
345 void setup_payload_gen6();
346 void assign_curb_setup();
347 void calculate_urb_setup();
348 void assign_urb_setup();
349 bool assign_regs(bool allow_spilling);
350 void assign_regs_trivial();
351 void get_used_mrfs(bool *mrf_used);
352 void setup_payload_interference(struct ra_graph *g, int payload_reg_count,
353 int first_payload_node);
354 void setup_mrf_hack_interference(struct ra_graph *g,
355 int first_mrf_hack_node);
356 int choose_spill_reg(struct ra_graph *g);
357 void spill_reg(int spill_reg);
358 void split_virtual_grfs();
359 void compact_virtual_grfs();
360 void move_uniform_array_access_to_pull_constants();
361 void assign_constant_locations();
362 void demote_pull_constants();
363 void invalidate_live_intervals();
364 void calculate_live_intervals();
365 void calculate_register_pressure();
366 bool opt_algebraic();
367 bool opt_cse();
368 bool opt_cse_local(bblock_t *block, exec_list *aeb);
369 bool opt_copy_propagate();
370 bool try_copy_propagate(fs_inst *inst, int arg, acp_entry *entry);
371 bool try_constant_propagate(fs_inst *inst, acp_entry *entry);
372 bool opt_copy_propagate_local(void *mem_ctx, bblock_t *block,
373 exec_list *acp);
374 void opt_drop_redundant_mov_to_flags();
375 bool register_coalesce();
376 bool compute_to_mrf();
377 bool dead_code_eliminate();
378 bool remove_duplicate_mrf_writes();
379 bool virtual_grf_interferes(int a, int b);
380 void schedule_instructions(instruction_scheduler_mode mode);
381 void insert_gen4_send_dependency_workarounds();
382 void insert_gen4_pre_send_dependency_workarounds(fs_inst *inst);
383 void insert_gen4_post_send_dependency_workarounds(fs_inst *inst);
384 void vfail(const char *msg, va_list args);
385 void fail(const char *msg, ...);
386 void no16(const char *msg, ...);
387 void lower_uniform_pull_constant_loads();
388
389 void push_force_uncompressed();
390 void pop_force_uncompressed();
391
392 void emit_dummy_fs();
393 fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
394 fs_inst *emit_linterp(const fs_reg &attr, const fs_reg &interp,
395 glsl_interp_qualifier interpolation_mode,
396 bool is_centroid, bool is_sample);
397 fs_reg *emit_frontfacing_interpolation(ir_variable *ir);
398 fs_reg *emit_samplepos_setup(ir_variable *ir);
399 fs_reg *emit_sampleid_setup(ir_variable *ir);
400 fs_reg *emit_samplemaskin_setup(ir_variable *ir);
401 fs_reg *emit_general_interpolation(ir_variable *ir);
402 void emit_interpolation_setup_gen4();
403 void emit_interpolation_setup_gen6();
404 void compute_sample_position(fs_reg dst, fs_reg int_sample_pos);
405 fs_reg rescale_texcoord(ir_texture *ir, fs_reg coordinate,
406 bool is_rect, int sampler, int texunit);
407 fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate,
408 fs_reg shadow_comp, fs_reg lod, fs_reg lod2);
409 fs_inst *emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate,
410 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
411 fs_reg sample_index);
412 fs_inst *emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
413 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
414 fs_reg sample_index, fs_reg mcs, int sampler);
415 fs_reg emit_mcs_fetch(ir_texture *ir, fs_reg coordinate, int sampler);
416 void emit_gen6_gather_wa(uint8_t wa, fs_reg dst);
417 fs_reg fix_math_operand(fs_reg src);
418 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0);
419 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0, fs_reg src1);
420 void emit_lrp(const fs_reg &dst, const fs_reg &x, const fs_reg &y,
421 const fs_reg &a);
422 void emit_minmax(uint32_t conditionalmod, const fs_reg &dst,
423 const fs_reg &src0, const fs_reg &src1);
424 bool try_emit_saturate(ir_expression *ir);
425 bool try_emit_mad(ir_expression *ir);
426 void try_replace_with_sel();
427 bool opt_peephole_sel();
428 bool opt_peephole_predicated_break();
429 bool opt_saturate_propagation();
430 void emit_bool_to_cond_code(ir_rvalue *condition);
431 void emit_if_gen6(ir_if *ir);
432 void emit_unspill(fs_inst *inst, fs_reg reg, uint32_t spill_offset,
433 int count);
434
435 void emit_fragment_program_code();
436 void setup_fp_regs();
437 fs_reg get_fp_src_reg(const prog_src_register *src);
438 fs_reg get_fp_dst_reg(const prog_dst_register *dst);
439 void emit_fp_alu1(enum opcode opcode,
440 const struct prog_instruction *fpi,
441 fs_reg dst, fs_reg src);
442 void emit_fp_alu2(enum opcode opcode,
443 const struct prog_instruction *fpi,
444 fs_reg dst, fs_reg src0, fs_reg src1);
445 void emit_fp_scalar_write(const struct prog_instruction *fpi,
446 fs_reg dst, fs_reg src);
447 void emit_fp_scalar_math(enum opcode opcode,
448 const struct prog_instruction *fpi,
449 fs_reg dst, fs_reg src);
450
451 void emit_fp_minmax(const struct prog_instruction *fpi,
452 fs_reg dst, fs_reg src0, fs_reg src1);
453
454 void emit_fp_sop(uint32_t conditional_mod,
455 const struct prog_instruction *fpi,
456 fs_reg dst, fs_reg src0, fs_reg src1, fs_reg one);
457
458 void emit_color_write(int target, int index, int first_color_mrf);
459 void emit_alpha_test();
460 void emit_fb_writes();
461
462 void emit_shader_time_begin();
463 void emit_shader_time_end();
464 void emit_shader_time_write(enum shader_time_shader_type type,
465 fs_reg value);
466
467 void emit_untyped_atomic(unsigned atomic_op, unsigned surf_index,
468 fs_reg dst, fs_reg offset, fs_reg src0,
469 fs_reg src1);
470
471 void emit_untyped_surface_read(unsigned surf_index, fs_reg dst,
472 fs_reg offset);
473
474 bool try_rewrite_rhs_to_dst(ir_assignment *ir,
475 fs_reg dst,
476 fs_reg src,
477 fs_inst *pre_rhs_inst,
478 fs_inst *last_rhs_inst);
479 void emit_assignment_writes(fs_reg &l, fs_reg &r,
480 const glsl_type *type, bool predicated);
481 void resolve_ud_negate(fs_reg *reg);
482 void resolve_bool_comparison(ir_rvalue *rvalue, fs_reg *reg);
483
484 fs_reg get_timestamp();
485
486 struct brw_reg interp_reg(int location, int channel);
487 void setup_uniform_values(ir_variable *ir);
488 void setup_builtin_uniform_values(ir_variable *ir);
489 int implied_mrf_writes(fs_inst *inst);
490
491 virtual void dump_instructions();
492 void dump_instruction(backend_instruction *inst);
493
494 void visit_atomic_counter_intrinsic(ir_call *ir);
495
496 struct gl_fragment_program *fp;
497 struct brw_wm_compile *c;
498 unsigned int sanity_param_count;
499
500 int *param_size;
501
502 int *virtual_grf_sizes;
503 int virtual_grf_count;
504 int virtual_grf_array_size;
505 int *virtual_grf_start;
506 int *virtual_grf_end;
507 brw::fs_live_variables *live_intervals;
508
509 int *regs_live_at_ip;
510
511 /** Number of uniform variable components visited. */
512 unsigned uniforms;
513
514 /**
515 * Array mapping UNIFORM register numbers to the pull parameter index,
516 * or -1 if this uniform register isn't being uploaded as a pull constant.
517 */
518 int *pull_constant_loc;
519
520 /**
521 * Array mapping UNIFORM register numbers to the push parameter index,
522 * or -1 if this uniform register isn't being uploaded as a push constant.
523 */
524 int *push_constant_loc;
525
526 struct hash_table *variable_ht;
527 fs_reg frag_depth;
528 fs_reg sample_mask;
529 fs_reg outputs[BRW_MAX_DRAW_BUFFERS];
530 unsigned output_components[BRW_MAX_DRAW_BUFFERS];
531 fs_reg dual_src_output;
532 bool do_dual_src;
533 int first_non_payload_grf;
534 /** Either BRW_MAX_GRF or GEN7_MRF_HACK_START */
535 int max_grf;
536
537 fs_reg *fp_temp_regs;
538 fs_reg *fp_input_regs;
539
540 /** @{ debug annotation info */
541 const char *current_annotation;
542 const void *base_ir;
543 /** @} */
544
545 bool failed;
546 char *fail_msg;
547 bool simd16_unsupported;
548 char *no16_msg;
549
550 /* Result of last visit() method. */
551 fs_reg result;
552
553 fs_reg pixel_x;
554 fs_reg pixel_y;
555 fs_reg wpos_w;
556 fs_reg pixel_w;
557 fs_reg delta_x[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
558 fs_reg delta_y[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
559 fs_reg shader_start_time;
560
561 int grf_used;
562 bool spilled_any_registers;
563
564 const unsigned dispatch_width; /**< 8 or 16 */
565
566 int force_uncompressed_stack;
567 };
568
569 /**
570 * The fragment shader code generator.
571 *
572 * Translates FS IR to actual i965 assembly code.
573 */
574 class fs_generator
575 {
576 public:
577 fs_generator(struct brw_context *brw,
578 struct brw_wm_compile *c,
579 struct gl_shader_program *prog,
580 struct gl_fragment_program *fp,
581 bool dual_source_output);
582 ~fs_generator();
583
584 const unsigned *generate_assembly(exec_list *simd8_instructions,
585 exec_list *simd16_instructions,
586 unsigned *assembly_size,
587 FILE *dump_file = NULL);
588
589 private:
590 void generate_code(exec_list *instructions, FILE *dump_file);
591 void generate_fb_write(fs_inst *inst);
592 void generate_blorp_fb_write(fs_inst *inst);
593 void generate_pixel_xy(struct brw_reg dst, bool is_x);
594 void generate_linterp(fs_inst *inst, struct brw_reg dst,
595 struct brw_reg *src);
596 void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
597 void generate_math1_gen7(fs_inst *inst,
598 struct brw_reg dst,
599 struct brw_reg src);
600 void generate_math2_gen7(fs_inst *inst,
601 struct brw_reg dst,
602 struct brw_reg src0,
603 struct brw_reg src1);
604 void generate_math1_gen6(fs_inst *inst,
605 struct brw_reg dst,
606 struct brw_reg src);
607 void generate_math2_gen6(fs_inst *inst,
608 struct brw_reg dst,
609 struct brw_reg src0,
610 struct brw_reg src1);
611 void generate_math_gen4(fs_inst *inst,
612 struct brw_reg dst,
613 struct brw_reg src);
614 void generate_math_g45(fs_inst *inst,
615 struct brw_reg dst,
616 struct brw_reg src);
617 void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
618 void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
619 bool negate_value);
620 void generate_scratch_write(fs_inst *inst, struct brw_reg src);
621 void generate_scratch_read(fs_inst *inst, struct brw_reg dst);
622 void generate_scratch_read_gen7(fs_inst *inst, struct brw_reg dst);
623 void generate_uniform_pull_constant_load(fs_inst *inst, struct brw_reg dst,
624 struct brw_reg index,
625 struct brw_reg offset);
626 void generate_uniform_pull_constant_load_gen7(fs_inst *inst,
627 struct brw_reg dst,
628 struct brw_reg surf_index,
629 struct brw_reg offset);
630 void generate_varying_pull_constant_load(fs_inst *inst, struct brw_reg dst,
631 struct brw_reg index,
632 struct brw_reg offset);
633 void generate_varying_pull_constant_load_gen7(fs_inst *inst,
634 struct brw_reg dst,
635 struct brw_reg index,
636 struct brw_reg offset);
637 void generate_mov_dispatch_to_flags(fs_inst *inst);
638
639 void generate_set_omask(fs_inst *inst,
640 struct brw_reg dst,
641 struct brw_reg sample_mask);
642
643 void generate_set_sample_id(fs_inst *inst,
644 struct brw_reg dst,
645 struct brw_reg src0,
646 struct brw_reg src1);
647
648 void generate_set_simd4x2_offset(fs_inst *inst,
649 struct brw_reg dst,
650 struct brw_reg offset);
651 void generate_discard_jump(fs_inst *inst);
652
653 void generate_pack_half_2x16_split(fs_inst *inst,
654 struct brw_reg dst,
655 struct brw_reg x,
656 struct brw_reg y);
657 void generate_unpack_half_2x16_split(fs_inst *inst,
658 struct brw_reg dst,
659 struct brw_reg src);
660
661 void generate_shader_time_add(fs_inst *inst,
662 struct brw_reg payload,
663 struct brw_reg offset,
664 struct brw_reg value);
665
666 void generate_untyped_atomic(fs_inst *inst,
667 struct brw_reg dst,
668 struct brw_reg atomic_op,
669 struct brw_reg surf_index);
670
671 void generate_untyped_surface_read(fs_inst *inst,
672 struct brw_reg dst,
673 struct brw_reg surf_index);
674
675 void patch_discard_jumps_to_fb_writes();
676
677 struct brw_context *brw;
678 struct gl_context *ctx;
679
680 struct brw_compile *p;
681 struct brw_wm_compile *c;
682
683 struct gl_shader_program *prog;
684 const struct gl_fragment_program *fp;
685
686 unsigned dispatch_width; /**< 8 or 16 */
687
688 exec_list discard_halt_patches;
689 bool dual_source_output;
690 void *mem_ctx;
691 };
692
693 /**
694 * The fragment shader code generator.
695 *
696 * Translates FS IR to actual i965 assembly code.
697 */
698 class gen8_fs_generator : public gen8_generator
699 {
700 public:
701 gen8_fs_generator(struct brw_context *brw,
702 struct brw_wm_compile *c,
703 struct gl_shader_program *prog,
704 struct gl_fragment_program *fp,
705 bool dual_source_output);
706 ~gen8_fs_generator();
707
708 const unsigned *generate_assembly(exec_list *simd8_instructions,
709 exec_list *simd16_instructions,
710 unsigned *assembly_size);
711
712 private:
713 void generate_code(exec_list *instructions);
714 void generate_fb_write(fs_inst *inst);
715 void generate_linterp(fs_inst *inst, struct brw_reg dst,
716 struct brw_reg *src);
717 void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
718 void generate_math1(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
719 void generate_math2(fs_inst *inst, struct brw_reg dst,
720 struct brw_reg src0, struct brw_reg src1);
721 void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
722 void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
723 bool negate_value);
724 void generate_scratch_write(fs_inst *inst, struct brw_reg src);
725 void generate_scratch_read(fs_inst *inst, struct brw_reg dst);
726 void generate_scratch_read_gen7(fs_inst *inst, struct brw_reg dst);
727 void generate_uniform_pull_constant_load(fs_inst *inst,
728 struct brw_reg dst,
729 struct brw_reg index,
730 struct brw_reg offset);
731 void generate_varying_pull_constant_load(fs_inst *inst,
732 struct brw_reg dst,
733 struct brw_reg index,
734 struct brw_reg offset);
735 void generate_mov_dispatch_to_flags(fs_inst *ir);
736 void generate_set_omask(fs_inst *ir,
737 struct brw_reg dst,
738 struct brw_reg sample_mask);
739 void generate_set_sample_id(fs_inst *ir,
740 struct brw_reg dst,
741 struct brw_reg src0,
742 struct brw_reg src1);
743 void generate_set_simd4x2_offset(fs_inst *ir,
744 struct brw_reg dst,
745 struct brw_reg offset);
746 void generate_pack_half_2x16_split(fs_inst *inst,
747 struct brw_reg dst,
748 struct brw_reg x,
749 struct brw_reg y);
750 void generate_unpack_half_2x16_split(fs_inst *inst,
751 struct brw_reg dst,
752 struct brw_reg src);
753 void generate_discard_jump(fs_inst *ir);
754
755 void patch_discard_jumps_to_fb_writes();
756
757 struct brw_wm_compile *c;
758 const struct gl_fragment_program *fp;
759
760 unsigned dispatch_width; /** 8 or 16 */
761
762 bool dual_source_output;
763
764 exec_list discard_halt_patches;
765 };
766
767 bool brw_do_channel_expressions(struct exec_list *instructions);
768 bool brw_do_vector_splitting(struct exec_list *instructions);
769 bool brw_fs_precompile(struct gl_context *ctx, struct gl_shader_program *prog);
770
771 struct brw_reg brw_reg_from_fs_reg(fs_reg *reg);