i965/fs: Add a concept of a width to fs_reg
[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 offset(fs_reg reg, unsigned delta)
128 {
129 assert(delta == 0 || (reg.file != HW_REG && reg.file != IMM));
130 reg.reg_offset += delta;
131 return reg;
132 }
133
134 static inline fs_reg
135 byte_offset(fs_reg reg, unsigned delta)
136 {
137 assert(delta == 0 || (reg.file != HW_REG && reg.file != IMM));
138 reg.subreg_offset += delta;
139 return reg;
140 }
141
142 /**
143 * Get either of the 8-component halves of a 16-component register.
144 *
145 * Note: this also works if \c reg represents a SIMD16 pair of registers.
146 */
147 static inline fs_reg
148 half(const fs_reg &reg, unsigned idx)
149 {
150 assert(idx < 2);
151 assert(idx == 0 || (reg.file != HW_REG && reg.file != IMM));
152 return byte_offset(reg, 8 * idx * reg.stride * type_sz(reg.type));
153 }
154
155 static const fs_reg reg_undef;
156
157 class ip_record : public exec_node {
158 public:
159 DECLARE_RALLOC_CXX_OPERATORS(ip_record)
160
161 ip_record(int ip)
162 {
163 this->ip = ip;
164 }
165
166 int ip;
167 };
168
169 class fs_inst : public backend_instruction {
170 fs_inst &operator=(const fs_inst &);
171
172 public:
173 DECLARE_RALLOC_CXX_OPERATORS(fs_inst)
174
175 void init(enum opcode opcode, const fs_reg &dst, fs_reg *src, int sources);
176
177 fs_inst(enum opcode opcode = BRW_OPCODE_NOP, const fs_reg &dst = reg_undef);
178 fs_inst(enum opcode opcode, const fs_reg &dst, const fs_reg &src0);
179 fs_inst(enum opcode opcode, const fs_reg &dst, const fs_reg &src0,
180 const fs_reg &src1);
181 fs_inst(enum opcode opcode, const fs_reg &dst, const fs_reg &src0,
182 const fs_reg &src1, const fs_reg &src2);
183 fs_inst(enum opcode opcode, const fs_reg &dst, fs_reg src[], int sources);
184 fs_inst(const fs_inst &that);
185
186 void resize_sources(uint8_t num_sources);
187
188 bool equals(fs_inst *inst) const;
189 bool overwrites_reg(const fs_reg &reg) const;
190 bool is_send_from_grf() const;
191 bool is_partial_write() const;
192 int regs_read(fs_visitor *v, int arg) const;
193 bool can_do_source_mods(struct brw_context *brw);
194
195 bool reads_flag() const;
196 bool writes_flag() const;
197
198 fs_reg dst;
199 fs_reg *src;
200
201 uint8_t sources; /**< Number of fs_reg sources. */
202
203 /* Chooses which flag subregister (f0.0 or f0.1) is used for conditional
204 * mod and predication.
205 */
206 uint8_t flag_subreg;
207
208 uint8_t regs_written; /**< Number of vgrfs written by a SEND message, or 1 */
209 bool eot:1;
210 bool header_present:1;
211 bool shadow_compare:1;
212 bool force_uncompressed:1;
213 bool force_sechalf:1;
214 bool pi_noperspective:1; /**< Pixel interpolator noperspective flag */
215 };
216
217 /**
218 * The fragment shader front-end.
219 *
220 * Translates either GLSL IR or Mesa IR (for ARB_fragment_program) into FS IR.
221 */
222 class fs_visitor : public backend_visitor
223 {
224 public:
225 const fs_reg reg_null_f;
226 const fs_reg reg_null_d;
227 const fs_reg reg_null_ud;
228
229 fs_visitor(struct brw_context *brw,
230 void *mem_ctx,
231 const struct brw_wm_prog_key *key,
232 struct brw_wm_prog_data *prog_data,
233 struct gl_shader_program *shader_prog,
234 struct gl_fragment_program *fp,
235 unsigned dispatch_width);
236 ~fs_visitor();
237 void init();
238
239 fs_reg *variable_storage(ir_variable *var);
240 int virtual_grf_alloc(int size);
241 void import_uniforms(fs_visitor *v);
242
243 void visit(ir_variable *ir);
244 void visit(ir_assignment *ir);
245 void visit(ir_dereference_variable *ir);
246 void visit(ir_dereference_record *ir);
247 void visit(ir_dereference_array *ir);
248 void visit(ir_expression *ir);
249 void visit(ir_texture *ir);
250 void visit(ir_if *ir);
251 void visit(ir_constant *ir);
252 void visit(ir_swizzle *ir);
253 void visit(ir_return *ir);
254 void visit(ir_loop *ir);
255 void visit(ir_loop_jump *ir);
256 void visit(ir_discard *ir);
257 void visit(ir_call *ir);
258 void visit(ir_function *ir);
259 void visit(ir_function_signature *ir);
260 void visit(ir_emit_vertex *);
261 void visit(ir_end_primitive *);
262
263 uint32_t gather_channel(ir_texture *ir, uint32_t sampler);
264 void swizzle_result(ir_texture *ir, fs_reg orig_val, uint32_t sampler);
265
266 fs_inst *emit(fs_inst *inst);
267 void emit(exec_list list);
268
269 fs_inst *emit(enum opcode opcode);
270 fs_inst *emit(enum opcode opcode, const fs_reg &dst);
271 fs_inst *emit(enum opcode opcode, const fs_reg &dst, const fs_reg &src0);
272 fs_inst *emit(enum opcode opcode, const fs_reg &dst, const fs_reg &src0,
273 const fs_reg &src1);
274 fs_inst *emit(enum opcode opcode, const fs_reg &dst,
275 const fs_reg &src0, const fs_reg &src1, const fs_reg &src2);
276 fs_inst *emit(enum opcode opcode, const fs_reg &dst,
277 fs_reg src[], int sources);
278
279 fs_inst *MOV(const fs_reg &dst, const fs_reg &src);
280 fs_inst *NOT(const fs_reg &dst, const fs_reg &src);
281 fs_inst *RNDD(const fs_reg &dst, const fs_reg &src);
282 fs_inst *RNDE(const fs_reg &dst, const fs_reg &src);
283 fs_inst *RNDZ(const fs_reg &dst, const fs_reg &src);
284 fs_inst *FRC(const fs_reg &dst, const fs_reg &src);
285 fs_inst *ADD(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
286 fs_inst *MUL(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
287 fs_inst *MACH(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
288 fs_inst *MAC(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
289 fs_inst *SHL(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
290 fs_inst *SHR(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
291 fs_inst *ASR(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
292 fs_inst *AND(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
293 fs_inst *OR(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
294 fs_inst *XOR(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
295 fs_inst *IF(enum brw_predicate predicate);
296 fs_inst *IF(const fs_reg &src0, const fs_reg &src1,
297 enum brw_conditional_mod condition);
298 fs_inst *CMP(fs_reg dst, fs_reg src0, fs_reg src1,
299 enum brw_conditional_mod condition);
300 fs_inst *LRP(const fs_reg &dst, const fs_reg &a, const fs_reg &y,
301 const fs_reg &x);
302 fs_inst *DEP_RESOLVE_MOV(int grf);
303 fs_inst *BFREV(const fs_reg &dst, const fs_reg &value);
304 fs_inst *BFE(const fs_reg &dst, const fs_reg &bits, const fs_reg &offset,
305 const fs_reg &value);
306 fs_inst *BFI1(const fs_reg &dst, const fs_reg &bits, const fs_reg &offset);
307 fs_inst *BFI2(const fs_reg &dst, const fs_reg &bfi1_dst,
308 const fs_reg &insert, const fs_reg &base);
309 fs_inst *FBH(const fs_reg &dst, const fs_reg &value);
310 fs_inst *FBL(const fs_reg &dst, const fs_reg &value);
311 fs_inst *CBIT(const fs_reg &dst, const fs_reg &value);
312 fs_inst *MAD(const fs_reg &dst, const fs_reg &c, const fs_reg &b,
313 const fs_reg &a);
314 fs_inst *ADDC(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
315 fs_inst *SUBB(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
316 fs_inst *SEL(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1);
317
318 int type_size(const struct glsl_type *type);
319 fs_inst *get_instruction_generating_reg(fs_inst *start,
320 fs_inst *end,
321 const fs_reg &reg);
322
323 fs_inst *LOAD_PAYLOAD(const fs_reg &dst, fs_reg *src, int sources);
324
325 exec_list VARYING_PULL_CONSTANT_LOAD(const fs_reg &dst,
326 const fs_reg &surf_index,
327 const fs_reg &varying_offset,
328 uint32_t const_offset);
329
330 bool run();
331 void assign_binding_table_offsets();
332 void setup_payload_gen4();
333 void setup_payload_gen6();
334 void assign_curb_setup();
335 void calculate_urb_setup();
336 void assign_urb_setup();
337 bool assign_regs(bool allow_spilling);
338 void assign_regs_trivial();
339 void get_used_mrfs(bool *mrf_used);
340 void setup_payload_interference(struct ra_graph *g, int payload_reg_count,
341 int first_payload_node);
342 void setup_mrf_hack_interference(struct ra_graph *g,
343 int first_mrf_hack_node);
344 int choose_spill_reg(struct ra_graph *g);
345 void spill_reg(int spill_reg);
346 void split_virtual_grfs();
347 bool compact_virtual_grfs();
348 void move_uniform_array_access_to_pull_constants();
349 void assign_constant_locations();
350 void demote_pull_constants();
351 void invalidate_live_intervals();
352 void calculate_live_intervals();
353 void calculate_register_pressure();
354 bool opt_algebraic();
355 bool opt_cse();
356 bool opt_cse_local(bblock_t *block);
357 bool opt_copy_propagate();
358 bool try_copy_propagate(fs_inst *inst, int arg, acp_entry *entry);
359 bool try_constant_propagate(fs_inst *inst, acp_entry *entry);
360 bool opt_copy_propagate_local(void *mem_ctx, bblock_t *block,
361 exec_list *acp);
362 void opt_drop_redundant_mov_to_flags();
363 bool opt_register_renaming();
364 bool register_coalesce();
365 bool compute_to_mrf();
366 bool dead_code_eliminate();
367 bool remove_duplicate_mrf_writes();
368 bool virtual_grf_interferes(int a, int b);
369 void schedule_instructions(instruction_scheduler_mode mode);
370 void insert_gen4_send_dependency_workarounds();
371 void insert_gen4_pre_send_dependency_workarounds(bblock_t *block,
372 fs_inst *inst);
373 void insert_gen4_post_send_dependency_workarounds(bblock_t *block,
374 fs_inst *inst);
375 void vfail(const char *msg, va_list args);
376 void fail(const char *msg, ...);
377 void no16(const char *msg, ...);
378 void lower_uniform_pull_constant_loads();
379 bool lower_load_payload();
380
381 void push_force_uncompressed();
382 void pop_force_uncompressed();
383
384 void emit_dummy_fs();
385 void emit_repclear_shader();
386 fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
387 fs_inst *emit_linterp(const fs_reg &attr, const fs_reg &interp,
388 glsl_interp_qualifier interpolation_mode,
389 bool is_centroid, bool is_sample);
390 fs_reg *emit_frontfacing_interpolation();
391 fs_reg *emit_samplepos_setup();
392 fs_reg *emit_sampleid_setup(ir_variable *ir);
393 fs_reg *emit_general_interpolation(ir_variable *ir);
394 void emit_interpolation_setup_gen4();
395 void emit_interpolation_setup_gen6();
396 void compute_sample_position(fs_reg dst, fs_reg int_sample_pos);
397 fs_reg rescale_texcoord(ir_texture *ir, fs_reg coordinate,
398 bool is_rect, uint32_t sampler, int texunit);
399 fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate,
400 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
401 uint32_t sampler);
402 fs_inst *emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate,
403 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
404 fs_reg sample_index, uint32_t sampler);
405 fs_inst *emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
406 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
407 fs_reg sample_index, fs_reg mcs, fs_reg sampler);
408 fs_reg emit_mcs_fetch(ir_texture *ir, fs_reg coordinate, fs_reg sampler);
409 void emit_gen6_gather_wa(uint8_t wa, fs_reg dst);
410 fs_reg fix_math_operand(fs_reg src);
411 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0);
412 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0, fs_reg src1);
413 void emit_lrp(const fs_reg &dst, const fs_reg &x, const fs_reg &y,
414 const fs_reg &a);
415 void emit_minmax(enum brw_conditional_mod conditionalmod, const fs_reg &dst,
416 const fs_reg &src0, const fs_reg &src1);
417 bool try_emit_saturate(ir_expression *ir);
418 bool try_emit_mad(ir_expression *ir);
419 void try_replace_with_sel();
420 bool opt_peephole_sel();
421 bool opt_peephole_predicated_break();
422 bool opt_saturate_propagation();
423 void emit_bool_to_cond_code(ir_rvalue *condition);
424 void emit_if_gen6(ir_if *ir);
425 void emit_unspill(bblock_t *block, fs_inst *inst, fs_reg reg,
426 uint32_t spill_offset, int count);
427
428 void emit_fragment_program_code();
429 void setup_fp_regs();
430 fs_reg get_fp_src_reg(const prog_src_register *src);
431 fs_reg get_fp_dst_reg(const prog_dst_register *dst);
432 void emit_fp_alu1(enum opcode opcode,
433 const struct prog_instruction *fpi,
434 fs_reg dst, fs_reg src);
435 void emit_fp_alu2(enum opcode opcode,
436 const struct prog_instruction *fpi,
437 fs_reg dst, fs_reg src0, fs_reg src1);
438 void emit_fp_scalar_write(const struct prog_instruction *fpi,
439 fs_reg dst, fs_reg src);
440 void emit_fp_scalar_math(enum opcode opcode,
441 const struct prog_instruction *fpi,
442 fs_reg dst, fs_reg src);
443
444 void emit_fp_minmax(const struct prog_instruction *fpi,
445 fs_reg dst, fs_reg src0, fs_reg src1);
446
447 void emit_fp_sop(enum brw_conditional_mod conditional_mod,
448 const struct prog_instruction *fpi,
449 fs_reg dst, fs_reg src0, fs_reg src1, fs_reg one);
450
451 void emit_color_write(fs_reg color, int index, int first_color_mrf);
452 void emit_alpha_test();
453 fs_inst *emit_single_fb_write(fs_reg color1, fs_reg color2,
454 fs_reg src0_alpha, unsigned components);
455 void emit_fb_writes();
456
457 void emit_shader_time_begin();
458 void emit_shader_time_end();
459 void emit_shader_time_write(enum shader_time_shader_type type,
460 fs_reg value);
461
462 void emit_untyped_atomic(unsigned atomic_op, unsigned surf_index,
463 fs_reg dst, fs_reg offset, fs_reg src0,
464 fs_reg src1);
465
466 void emit_untyped_surface_read(unsigned surf_index, fs_reg dst,
467 fs_reg offset);
468
469 void emit_interpolate_expression(ir_expression *ir);
470
471 bool try_rewrite_rhs_to_dst(ir_assignment *ir,
472 fs_reg dst,
473 fs_reg src,
474 fs_inst *pre_rhs_inst,
475 fs_inst *last_rhs_inst);
476 void emit_assignment_writes(fs_reg &l, fs_reg &r,
477 const glsl_type *type, bool predicated);
478 void resolve_ud_negate(fs_reg *reg);
479 void resolve_bool_comparison(ir_rvalue *rvalue, fs_reg *reg);
480
481 fs_reg get_timestamp();
482
483 struct brw_reg interp_reg(int location, int channel);
484 void setup_uniform_values(ir_variable *ir);
485 void setup_builtin_uniform_values(ir_variable *ir);
486 int implied_mrf_writes(fs_inst *inst);
487
488 virtual void dump_instructions();
489 virtual void dump_instructions(const char *name);
490 void dump_instruction(backend_instruction *inst);
491 void dump_instruction(backend_instruction *inst, FILE *file);
492
493 void visit_atomic_counter_intrinsic(ir_call *ir);
494
495 const void *const key;
496 struct brw_stage_prog_data *prog_data;
497 unsigned int sanity_param_count;
498
499 int *param_size;
500
501 int *virtual_grf_sizes;
502 int virtual_grf_count;
503 int virtual_grf_array_size;
504 int *virtual_grf_start;
505 int *virtual_grf_end;
506 brw::fs_live_variables *live_intervals;
507
508 int *regs_live_at_ip;
509
510 /** Number of uniform variable components visited. */
511 unsigned uniforms;
512
513 /** Byte-offset for the next available spot in the scratch space buffer. */
514 unsigned last_scratch;
515
516 /**
517 * Array mapping UNIFORM register numbers to the pull parameter index,
518 * or -1 if this uniform register isn't being uploaded as a pull constant.
519 */
520 int *pull_constant_loc;
521
522 /**
523 * Array mapping UNIFORM register numbers to the push parameter index,
524 * or -1 if this uniform register isn't being uploaded as a push constant.
525 */
526 int *push_constant_loc;
527
528 struct hash_table *variable_ht;
529 fs_reg frag_depth;
530 fs_reg sample_mask;
531 fs_reg outputs[BRW_MAX_DRAW_BUFFERS];
532 unsigned output_components[BRW_MAX_DRAW_BUFFERS];
533 fs_reg dual_src_output;
534 bool do_dual_src;
535 int first_non_payload_grf;
536 /** Either BRW_MAX_GRF or GEN7_MRF_HACK_START */
537 int max_grf;
538
539 fs_reg *fp_temp_regs;
540 fs_reg *fp_input_regs;
541
542 /** @{ debug annotation info */
543 const char *current_annotation;
544 const void *base_ir;
545 /** @} */
546
547 bool failed;
548 char *fail_msg;
549 bool simd16_unsupported;
550 char *no16_msg;
551
552 /* Result of last visit() method. */
553 fs_reg result;
554
555 /** Register numbers for thread payload fields. */
556 struct {
557 uint8_t source_depth_reg;
558 uint8_t source_w_reg;
559 uint8_t aa_dest_stencil_reg;
560 uint8_t dest_depth_reg;
561 uint8_t sample_pos_reg;
562 uint8_t sample_mask_in_reg;
563 uint8_t barycentric_coord_reg[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
564
565 /** The number of thread payload registers the hardware will supply. */
566 uint8_t num_regs;
567 } payload;
568
569 bool source_depth_to_render_target;
570 bool runtime_check_aads_emit;
571
572 fs_reg pixel_x;
573 fs_reg pixel_y;
574 fs_reg wpos_w;
575 fs_reg pixel_w;
576 fs_reg delta_x[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
577 fs_reg delta_y[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
578 fs_reg shader_start_time;
579
580 int grf_used;
581 bool spilled_any_registers;
582
583 const unsigned dispatch_width; /**< 8 or 16 */
584
585 int force_uncompressed_stack;
586 };
587
588 /**
589 * The fragment shader code generator.
590 *
591 * Translates FS IR to actual i965 assembly code.
592 */
593 class fs_generator
594 {
595 public:
596 fs_generator(struct brw_context *brw,
597 void *mem_ctx,
598 const struct brw_wm_prog_key *key,
599 struct brw_wm_prog_data *prog_data,
600 struct gl_shader_program *shader_prog,
601 struct gl_fragment_program *fp,
602 bool runtime_check_aads_emit,
603 bool debug_flag);
604 ~fs_generator();
605
606 const unsigned *generate_assembly(const cfg_t *simd8_cfg,
607 const cfg_t *simd16_cfg,
608 unsigned *assembly_size);
609
610 private:
611 void generate_code(const cfg_t *cfg);
612 void fire_fb_write(fs_inst *inst,
613 GLuint base_reg,
614 struct brw_reg implied_header,
615 GLuint nr);
616 void generate_fb_write(fs_inst *inst);
617 void generate_blorp_fb_write(fs_inst *inst);
618 void generate_rep_fb_write(fs_inst *inst);
619 void generate_pixel_xy(struct brw_reg dst, bool is_x);
620 void generate_linterp(fs_inst *inst, struct brw_reg dst,
621 struct brw_reg *src);
622 void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
623 struct brw_reg sampler_index);
624 void generate_math_gen6(fs_inst *inst,
625 struct brw_reg dst,
626 struct brw_reg src0,
627 struct brw_reg src1);
628 void generate_math_gen4(fs_inst *inst,
629 struct brw_reg dst,
630 struct brw_reg src);
631 void generate_math_g45(fs_inst *inst,
632 struct brw_reg dst,
633 struct brw_reg src);
634 void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src, struct brw_reg quality);
635 void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
636 struct brw_reg quality, bool negate_value);
637 void generate_scratch_write(fs_inst *inst, struct brw_reg src);
638 void generate_scratch_read(fs_inst *inst, struct brw_reg dst);
639 void generate_scratch_read_gen7(fs_inst *inst, struct brw_reg dst);
640 void generate_uniform_pull_constant_load(fs_inst *inst, struct brw_reg dst,
641 struct brw_reg index,
642 struct brw_reg offset);
643 void generate_uniform_pull_constant_load_gen7(fs_inst *inst,
644 struct brw_reg dst,
645 struct brw_reg surf_index,
646 struct brw_reg offset);
647 void generate_varying_pull_constant_load(fs_inst *inst, struct brw_reg dst,
648 struct brw_reg index,
649 struct brw_reg offset);
650 void generate_varying_pull_constant_load_gen7(fs_inst *inst,
651 struct brw_reg dst,
652 struct brw_reg index,
653 struct brw_reg offset);
654 void generate_mov_dispatch_to_flags(fs_inst *inst);
655
656 void generate_pixel_interpolator_query(fs_inst *inst,
657 struct brw_reg dst,
658 struct brw_reg src,
659 struct brw_reg msg_data,
660 unsigned msg_type);
661
662 void generate_set_omask(fs_inst *inst,
663 struct brw_reg dst,
664 struct brw_reg sample_mask);
665
666 void generate_set_sample_id(fs_inst *inst,
667 struct brw_reg dst,
668 struct brw_reg src0,
669 struct brw_reg src1);
670
671 void generate_set_simd4x2_offset(fs_inst *inst,
672 struct brw_reg dst,
673 struct brw_reg offset);
674 void generate_discard_jump(fs_inst *inst);
675
676 void generate_pack_half_2x16_split(fs_inst *inst,
677 struct brw_reg dst,
678 struct brw_reg x,
679 struct brw_reg y);
680 void generate_unpack_half_2x16_split(fs_inst *inst,
681 struct brw_reg dst,
682 struct brw_reg src);
683
684 void generate_shader_time_add(fs_inst *inst,
685 struct brw_reg payload,
686 struct brw_reg offset,
687 struct brw_reg value);
688
689 void generate_untyped_atomic(fs_inst *inst,
690 struct brw_reg dst,
691 struct brw_reg atomic_op,
692 struct brw_reg surf_index);
693
694 void generate_untyped_surface_read(fs_inst *inst,
695 struct brw_reg dst,
696 struct brw_reg surf_index);
697
698 bool patch_discard_jumps_to_fb_writes();
699
700 struct brw_context *brw;
701 struct gl_context *ctx;
702
703 struct brw_compile *p;
704 gl_shader_stage stage;
705 const void * const key;
706 struct brw_stage_prog_data * const prog_data;
707
708 struct gl_shader_program * const shader_prog;
709 const struct gl_program *prog;
710
711 unsigned dispatch_width; /**< 8 or 16 */
712
713 exec_list discard_halt_patches;
714 bool runtime_check_aads_emit;
715 const bool debug_flag;
716 void *mem_ctx;
717 };
718
719 bool brw_do_channel_expressions(struct exec_list *instructions);
720 bool brw_do_vector_splitting(struct exec_list *instructions);
721 bool brw_fs_precompile(struct gl_context *ctx, struct gl_shader_program *prog);
722
723 struct brw_reg brw_reg_from_fs_reg(fs_reg *reg);