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