i965/fs: Make patch_discard_jumps_to_fb_writes return bool.
[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 void *mem_ctx,
254 const struct brw_wm_prog_key *key,
255 struct brw_wm_prog_data *prog_data,
256 struct gl_shader_program *shader_prog,
257 struct gl_fragment_program *fp,
258 unsigned dispatch_width);
259 ~fs_visitor();
260
261 fs_reg *variable_storage(ir_variable *var);
262 int virtual_grf_alloc(int size);
263 void import_uniforms(fs_visitor *v);
264
265 void visit(ir_variable *ir);
266 void visit(ir_assignment *ir);
267 void visit(ir_dereference_variable *ir);
268 void visit(ir_dereference_record *ir);
269 void visit(ir_dereference_array *ir);
270 void visit(ir_expression *ir);
271 void visit(ir_texture *ir);
272 void visit(ir_if *ir);
273 void visit(ir_constant *ir);
274 void visit(ir_swizzle *ir);
275 void visit(ir_return *ir);
276 void visit(ir_loop *ir);
277 void visit(ir_loop_jump *ir);
278 void visit(ir_discard *ir);
279 void visit(ir_call *ir);
280 void visit(ir_function *ir);
281 void visit(ir_function_signature *ir);
282 void visit(ir_emit_vertex *);
283 void visit(ir_end_primitive *);
284
285 uint32_t gather_channel(ir_texture *ir, int sampler);
286 void swizzle_result(ir_texture *ir, fs_reg orig_val, int sampler);
287
288 bool can_do_source_mods(fs_inst *inst);
289
290 fs_inst *emit(fs_inst *inst);
291 void emit(exec_list list);
292
293 fs_inst *emit(enum opcode opcode);
294 fs_inst *emit(enum opcode opcode, fs_reg dst);
295 fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0);
296 fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1);
297 fs_inst *emit(enum opcode opcode, fs_reg dst,
298 fs_reg src0, fs_reg src1, fs_reg src2);
299
300 fs_inst *MOV(fs_reg dst, fs_reg src);
301 fs_inst *NOT(fs_reg dst, fs_reg src);
302 fs_inst *RNDD(fs_reg dst, fs_reg src);
303 fs_inst *RNDE(fs_reg dst, fs_reg src);
304 fs_inst *RNDZ(fs_reg dst, fs_reg src);
305 fs_inst *FRC(fs_reg dst, fs_reg src);
306 fs_inst *ADD(fs_reg dst, fs_reg src0, fs_reg src1);
307 fs_inst *MUL(fs_reg dst, fs_reg src0, fs_reg src1);
308 fs_inst *MACH(fs_reg dst, fs_reg src0, fs_reg src1);
309 fs_inst *MAC(fs_reg dst, fs_reg src0, fs_reg src1);
310 fs_inst *SHL(fs_reg dst, fs_reg src0, fs_reg src1);
311 fs_inst *SHR(fs_reg dst, fs_reg src0, fs_reg src1);
312 fs_inst *ASR(fs_reg dst, fs_reg src0, fs_reg src1);
313 fs_inst *AND(fs_reg dst, fs_reg src0, fs_reg src1);
314 fs_inst *OR(fs_reg dst, fs_reg src0, fs_reg src1);
315 fs_inst *XOR(fs_reg dst, fs_reg src0, fs_reg src1);
316 fs_inst *IF(uint32_t predicate);
317 fs_inst *IF(fs_reg src0, fs_reg src1, uint32_t condition);
318 fs_inst *CMP(fs_reg dst, fs_reg src0, fs_reg src1,
319 uint32_t condition);
320 fs_inst *LRP(fs_reg dst, fs_reg a, fs_reg y, fs_reg x);
321 fs_inst *DEP_RESOLVE_MOV(int grf);
322 fs_inst *BFREV(fs_reg dst, fs_reg value);
323 fs_inst *BFE(fs_reg dst, fs_reg bits, fs_reg offset, fs_reg value);
324 fs_inst *BFI1(fs_reg dst, fs_reg bits, fs_reg offset);
325 fs_inst *BFI2(fs_reg dst, fs_reg bfi1_dst, fs_reg insert, fs_reg base);
326 fs_inst *FBH(fs_reg dst, fs_reg value);
327 fs_inst *FBL(fs_reg dst, fs_reg value);
328 fs_inst *CBIT(fs_reg dst, fs_reg value);
329 fs_inst *MAD(fs_reg dst, fs_reg c, fs_reg b, fs_reg a);
330 fs_inst *ADDC(fs_reg dst, fs_reg src0, fs_reg src1);
331 fs_inst *SUBB(fs_reg dst, fs_reg src0, fs_reg src1);
332 fs_inst *SEL(fs_reg dst, fs_reg src0, fs_reg src1);
333
334 int type_size(const struct glsl_type *type);
335 fs_inst *get_instruction_generating_reg(fs_inst *start,
336 fs_inst *end,
337 const fs_reg &reg);
338
339 exec_list VARYING_PULL_CONSTANT_LOAD(const fs_reg &dst,
340 const fs_reg &surf_index,
341 const fs_reg &varying_offset,
342 uint32_t const_offset);
343
344 bool run();
345 void assign_binding_table_offsets();
346 void setup_payload_gen4();
347 void setup_payload_gen6();
348 void assign_curb_setup();
349 void calculate_urb_setup();
350 void assign_urb_setup();
351 bool assign_regs(bool allow_spilling);
352 void assign_regs_trivial();
353 void get_used_mrfs(bool *mrf_used);
354 void setup_payload_interference(struct ra_graph *g, int payload_reg_count,
355 int first_payload_node);
356 void setup_mrf_hack_interference(struct ra_graph *g,
357 int first_mrf_hack_node);
358 int choose_spill_reg(struct ra_graph *g);
359 void spill_reg(int spill_reg);
360 void split_virtual_grfs();
361 void compact_virtual_grfs();
362 void move_uniform_array_access_to_pull_constants();
363 void assign_constant_locations();
364 void demote_pull_constants();
365 void invalidate_live_intervals();
366 void calculate_live_intervals();
367 void calculate_register_pressure();
368 bool opt_algebraic();
369 bool opt_cse();
370 bool opt_cse_local(bblock_t *block, exec_list *aeb);
371 bool opt_copy_propagate();
372 bool try_copy_propagate(fs_inst *inst, int arg, acp_entry *entry);
373 bool try_constant_propagate(fs_inst *inst, acp_entry *entry);
374 bool opt_copy_propagate_local(void *mem_ctx, bblock_t *block,
375 exec_list *acp);
376 void opt_drop_redundant_mov_to_flags();
377 bool register_coalesce();
378 bool compute_to_mrf();
379 bool dead_code_eliminate();
380 bool remove_duplicate_mrf_writes();
381 bool virtual_grf_interferes(int a, int b);
382 void schedule_instructions(instruction_scheduler_mode mode);
383 void insert_gen4_send_dependency_workarounds();
384 void insert_gen4_pre_send_dependency_workarounds(fs_inst *inst);
385 void insert_gen4_post_send_dependency_workarounds(fs_inst *inst);
386 void vfail(const char *msg, va_list args);
387 void fail(const char *msg, ...);
388 void no16(const char *msg, ...);
389 void lower_uniform_pull_constant_loads();
390
391 void push_force_uncompressed();
392 void pop_force_uncompressed();
393
394 void emit_dummy_fs();
395 fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
396 fs_inst *emit_linterp(const fs_reg &attr, const fs_reg &interp,
397 glsl_interp_qualifier interpolation_mode,
398 bool is_centroid, bool is_sample);
399 fs_reg *emit_frontfacing_interpolation(ir_variable *ir);
400 fs_reg *emit_samplepos_setup(ir_variable *ir);
401 fs_reg *emit_sampleid_setup(ir_variable *ir);
402 fs_reg *emit_general_interpolation(ir_variable *ir);
403 void emit_interpolation_setup_gen4();
404 void emit_interpolation_setup_gen6();
405 void compute_sample_position(fs_reg dst, fs_reg int_sample_pos);
406 fs_reg rescale_texcoord(ir_texture *ir, fs_reg coordinate,
407 bool is_rect, int sampler, int texunit);
408 fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate,
409 fs_reg shadow_comp, fs_reg lod, fs_reg lod2);
410 fs_inst *emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate,
411 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
412 fs_reg sample_index);
413 fs_inst *emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
414 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
415 fs_reg sample_index, fs_reg mcs, int sampler);
416 fs_reg emit_mcs_fetch(ir_texture *ir, fs_reg coordinate, int sampler);
417 void emit_gen6_gather_wa(uint8_t wa, fs_reg dst);
418 fs_reg fix_math_operand(fs_reg src);
419 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0);
420 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0, fs_reg src1);
421 void emit_lrp(const fs_reg &dst, const fs_reg &x, const fs_reg &y,
422 const fs_reg &a);
423 void emit_minmax(uint32_t conditionalmod, const fs_reg &dst,
424 const fs_reg &src0, const fs_reg &src1);
425 bool try_emit_saturate(ir_expression *ir);
426 bool try_emit_mad(ir_expression *ir);
427 void try_replace_with_sel();
428 bool opt_peephole_sel();
429 bool opt_peephole_predicated_break();
430 bool opt_saturate_propagation();
431 void emit_bool_to_cond_code(ir_rvalue *condition);
432 void emit_if_gen6(ir_if *ir);
433 void emit_unspill(fs_inst *inst, fs_reg reg, uint32_t spill_offset,
434 int count);
435
436 void emit_fragment_program_code();
437 void setup_fp_regs();
438 fs_reg get_fp_src_reg(const prog_src_register *src);
439 fs_reg get_fp_dst_reg(const prog_dst_register *dst);
440 void emit_fp_alu1(enum opcode opcode,
441 const struct prog_instruction *fpi,
442 fs_reg dst, fs_reg src);
443 void emit_fp_alu2(enum opcode opcode,
444 const struct prog_instruction *fpi,
445 fs_reg dst, fs_reg src0, fs_reg src1);
446 void emit_fp_scalar_write(const struct prog_instruction *fpi,
447 fs_reg dst, fs_reg src);
448 void emit_fp_scalar_math(enum opcode opcode,
449 const struct prog_instruction *fpi,
450 fs_reg dst, fs_reg src);
451
452 void emit_fp_minmax(const struct prog_instruction *fpi,
453 fs_reg dst, fs_reg src0, fs_reg src1);
454
455 void emit_fp_sop(uint32_t conditional_mod,
456 const struct prog_instruction *fpi,
457 fs_reg dst, fs_reg src0, fs_reg src1, fs_reg one);
458
459 void emit_color_write(int target, int index, int first_color_mrf);
460 void emit_alpha_test();
461 void emit_fb_writes();
462
463 void emit_shader_time_begin();
464 void emit_shader_time_end();
465 void emit_shader_time_write(enum shader_time_shader_type type,
466 fs_reg value);
467
468 void emit_untyped_atomic(unsigned atomic_op, unsigned surf_index,
469 fs_reg dst, fs_reg offset, fs_reg src0,
470 fs_reg src1);
471
472 void emit_untyped_surface_read(unsigned surf_index, fs_reg dst,
473 fs_reg offset);
474
475 bool try_rewrite_rhs_to_dst(ir_assignment *ir,
476 fs_reg dst,
477 fs_reg src,
478 fs_inst *pre_rhs_inst,
479 fs_inst *last_rhs_inst);
480 void emit_assignment_writes(fs_reg &l, fs_reg &r,
481 const glsl_type *type, bool predicated);
482 void resolve_ud_negate(fs_reg *reg);
483 void resolve_bool_comparison(ir_rvalue *rvalue, fs_reg *reg);
484
485 fs_reg get_timestamp();
486
487 struct brw_reg interp_reg(int location, int channel);
488 void setup_uniform_values(ir_variable *ir);
489 void setup_builtin_uniform_values(ir_variable *ir);
490 int implied_mrf_writes(fs_inst *inst);
491
492 virtual void dump_instructions();
493 void dump_instruction(backend_instruction *inst);
494
495 void visit_atomic_counter_intrinsic(ir_call *ir);
496
497 struct gl_fragment_program *fp;
498 const struct brw_wm_prog_key *const key;
499 struct brw_wm_prog_data *prog_data;
500 unsigned int sanity_param_count;
501
502 int *param_size;
503
504 int *virtual_grf_sizes;
505 int virtual_grf_count;
506 int virtual_grf_array_size;
507 int *virtual_grf_start;
508 int *virtual_grf_end;
509 brw::fs_live_variables *live_intervals;
510
511 int *regs_live_at_ip;
512
513 /** Number of uniform variable components visited. */
514 unsigned uniforms;
515
516 /** Byte-offset for the next available spot in the scratch space buffer. */
517 unsigned last_scratch;
518
519 /**
520 * Array mapping UNIFORM register numbers to the pull parameter index,
521 * or -1 if this uniform register isn't being uploaded as a pull constant.
522 */
523 int *pull_constant_loc;
524
525 /**
526 * Array mapping UNIFORM register numbers to the push parameter index,
527 * or -1 if this uniform register isn't being uploaded as a push constant.
528 */
529 int *push_constant_loc;
530
531 struct hash_table *variable_ht;
532 fs_reg frag_depth;
533 fs_reg sample_mask;
534 fs_reg outputs[BRW_MAX_DRAW_BUFFERS];
535 unsigned output_components[BRW_MAX_DRAW_BUFFERS];
536 fs_reg dual_src_output;
537 bool do_dual_src;
538 int first_non_payload_grf;
539 /** Either BRW_MAX_GRF or GEN7_MRF_HACK_START */
540 int max_grf;
541
542 fs_reg *fp_temp_regs;
543 fs_reg *fp_input_regs;
544
545 /** @{ debug annotation info */
546 const char *current_annotation;
547 const void *base_ir;
548 /** @} */
549
550 bool failed;
551 char *fail_msg;
552 bool simd16_unsupported;
553 char *no16_msg;
554
555 /* Result of last visit() method. */
556 fs_reg result;
557
558 /** Register numbers for thread payload fields. */
559 struct {
560 uint8_t source_depth_reg;
561 uint8_t source_w_reg;
562 uint8_t aa_dest_stencil_reg;
563 uint8_t dest_depth_reg;
564 uint8_t sample_pos_reg;
565 uint8_t sample_mask_in_reg;
566 uint8_t barycentric_coord_reg[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
567
568 /** The number of thread payload registers the hardware will supply. */
569 uint8_t num_regs;
570 } payload;
571
572 bool source_depth_to_render_target;
573 bool runtime_check_aads_emit;
574
575 fs_reg pixel_x;
576 fs_reg pixel_y;
577 fs_reg wpos_w;
578 fs_reg pixel_w;
579 fs_reg delta_x[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
580 fs_reg delta_y[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
581 fs_reg shader_start_time;
582
583 int grf_used;
584 bool spilled_any_registers;
585
586 const unsigned dispatch_width; /**< 8 or 16 */
587
588 int force_uncompressed_stack;
589 };
590
591 /**
592 * The fragment shader code generator.
593 *
594 * Translates FS IR to actual i965 assembly code.
595 */
596 class fs_generator
597 {
598 public:
599 fs_generator(struct brw_context *brw,
600 void *mem_ctx,
601 const struct brw_wm_prog_key *key,
602 struct brw_wm_prog_data *prog_data,
603 struct gl_shader_program *prog,
604 struct gl_fragment_program *fp,
605 bool dual_source_output,
606 bool debug_flag);
607 ~fs_generator();
608
609 const unsigned *generate_assembly(exec_list *simd8_instructions,
610 exec_list *simd16_instructions,
611 unsigned *assembly_size);
612
613 private:
614 void generate_code(exec_list *instructions);
615 void generate_fb_write(fs_inst *inst);
616 void generate_blorp_fb_write(fs_inst *inst);
617 void generate_pixel_xy(struct brw_reg dst, bool is_x);
618 void generate_linterp(fs_inst *inst, struct brw_reg dst,
619 struct brw_reg *src);
620 void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
621 void generate_math1_gen7(fs_inst *inst,
622 struct brw_reg dst,
623 struct brw_reg src);
624 void generate_math2_gen7(fs_inst *inst,
625 struct brw_reg dst,
626 struct brw_reg src0,
627 struct brw_reg src1);
628 void generate_math1_gen6(fs_inst *inst,
629 struct brw_reg dst,
630 struct brw_reg src);
631 void generate_math2_gen6(fs_inst *inst,
632 struct brw_reg dst,
633 struct brw_reg src0,
634 struct brw_reg src1);
635 void generate_math_gen4(fs_inst *inst,
636 struct brw_reg dst,
637 struct brw_reg src);
638 void generate_math_g45(fs_inst *inst,
639 struct brw_reg dst,
640 struct brw_reg src);
641 void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
642 void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
643 bool negate_value);
644 void generate_scratch_write(fs_inst *inst, struct brw_reg src);
645 void generate_scratch_read(fs_inst *inst, struct brw_reg dst);
646 void generate_scratch_read_gen7(fs_inst *inst, struct brw_reg dst);
647 void generate_uniform_pull_constant_load(fs_inst *inst, struct brw_reg dst,
648 struct brw_reg index,
649 struct brw_reg offset);
650 void generate_uniform_pull_constant_load_gen7(fs_inst *inst,
651 struct brw_reg dst,
652 struct brw_reg surf_index,
653 struct brw_reg offset);
654 void generate_varying_pull_constant_load(fs_inst *inst, struct brw_reg dst,
655 struct brw_reg index,
656 struct brw_reg offset);
657 void generate_varying_pull_constant_load_gen7(fs_inst *inst,
658 struct brw_reg dst,
659 struct brw_reg index,
660 struct brw_reg offset);
661 void generate_mov_dispatch_to_flags(fs_inst *inst);
662
663 void generate_set_omask(fs_inst *inst,
664 struct brw_reg dst,
665 struct brw_reg sample_mask);
666
667 void generate_set_sample_id(fs_inst *inst,
668 struct brw_reg dst,
669 struct brw_reg src0,
670 struct brw_reg src1);
671
672 void generate_set_simd4x2_offset(fs_inst *inst,
673 struct brw_reg dst,
674 struct brw_reg offset);
675 void generate_discard_jump(fs_inst *inst);
676
677 void generate_pack_half_2x16_split(fs_inst *inst,
678 struct brw_reg dst,
679 struct brw_reg x,
680 struct brw_reg y);
681 void generate_unpack_half_2x16_split(fs_inst *inst,
682 struct brw_reg dst,
683 struct brw_reg src);
684
685 void generate_shader_time_add(fs_inst *inst,
686 struct brw_reg payload,
687 struct brw_reg offset,
688 struct brw_reg value);
689
690 void generate_untyped_atomic(fs_inst *inst,
691 struct brw_reg dst,
692 struct brw_reg atomic_op,
693 struct brw_reg surf_index);
694
695 void generate_untyped_surface_read(fs_inst *inst,
696 struct brw_reg dst,
697 struct brw_reg surf_index);
698
699 bool patch_discard_jumps_to_fb_writes();
700
701 struct brw_context *brw;
702 struct gl_context *ctx;
703
704 struct brw_compile *p;
705 const struct brw_wm_prog_key *const key;
706 struct brw_wm_prog_data *prog_data;
707
708 struct gl_shader_program *prog;
709 const struct gl_fragment_program *fp;
710
711 unsigned dispatch_width; /**< 8 or 16 */
712
713 exec_list discard_halt_patches;
714 bool dual_source_output;
715 const bool debug_flag;
716 void *mem_ctx;
717 };
718
719 /**
720 * The fragment shader code generator.
721 *
722 * Translates FS IR to actual i965 assembly code.
723 */
724 class gen8_fs_generator : public gen8_generator
725 {
726 public:
727 gen8_fs_generator(struct brw_context *brw,
728 void *mem_ctx,
729 const struct brw_wm_prog_key *key,
730 struct brw_wm_prog_data *prog_data,
731 struct gl_shader_program *prog,
732 struct gl_fragment_program *fp,
733 bool dual_source_output);
734 ~gen8_fs_generator();
735
736 const unsigned *generate_assembly(exec_list *simd8_instructions,
737 exec_list *simd16_instructions,
738 unsigned *assembly_size);
739
740 private:
741 void generate_code(exec_list *instructions);
742 void generate_fb_write(fs_inst *inst);
743 void generate_linterp(fs_inst *inst, struct brw_reg dst,
744 struct brw_reg *src);
745 void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
746 void generate_math1(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
747 void generate_math2(fs_inst *inst, struct brw_reg dst,
748 struct brw_reg src0, struct brw_reg src1);
749 void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
750 void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
751 bool negate_value);
752 void generate_scratch_write(fs_inst *inst, struct brw_reg src);
753 void generate_scratch_read(fs_inst *inst, struct brw_reg dst);
754 void generate_scratch_read_gen7(fs_inst *inst, struct brw_reg dst);
755 void generate_uniform_pull_constant_load(fs_inst *inst,
756 struct brw_reg dst,
757 struct brw_reg index,
758 struct brw_reg offset);
759 void generate_varying_pull_constant_load(fs_inst *inst,
760 struct brw_reg dst,
761 struct brw_reg index,
762 struct brw_reg offset);
763 void generate_mov_dispatch_to_flags(fs_inst *ir);
764 void generate_set_omask(fs_inst *ir,
765 struct brw_reg dst,
766 struct brw_reg sample_mask);
767 void generate_set_sample_id(fs_inst *ir,
768 struct brw_reg dst,
769 struct brw_reg src0,
770 struct brw_reg src1);
771 void generate_set_simd4x2_offset(fs_inst *ir,
772 struct brw_reg dst,
773 struct brw_reg offset);
774 void generate_pack_half_2x16_split(fs_inst *inst,
775 struct brw_reg dst,
776 struct brw_reg x,
777 struct brw_reg y);
778 void generate_unpack_half_2x16_split(fs_inst *inst,
779 struct brw_reg dst,
780 struct brw_reg src);
781 void generate_untyped_atomic(fs_inst *inst,
782 struct brw_reg dst,
783 struct brw_reg atomic_op,
784 struct brw_reg surf_index);
785
786 void generate_untyped_surface_read(fs_inst *inst,
787 struct brw_reg dst,
788 struct brw_reg surf_index);
789 void generate_discard_jump(fs_inst *ir);
790
791 bool patch_discard_jumps_to_fb_writes();
792
793 const struct brw_wm_prog_key *const key;
794 struct brw_wm_prog_data *prog_data;
795 const struct gl_fragment_program *fp;
796
797 unsigned dispatch_width; /** 8 or 16 */
798
799 bool dual_source_output;
800
801 exec_list discard_halt_patches;
802 };
803
804 bool brw_do_channel_expressions(struct exec_list *instructions);
805 bool brw_do_vector_splitting(struct exec_list *instructions);
806 bool brw_fs_precompile(struct gl_context *ctx, struct gl_shader_program *prog);
807
808 struct brw_reg brw_reg_from_fs_reg(fs_reg *reg);