i965: Move annotation info into generate code.
[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 #include "intel_asm_printer.h"
50 }
51 #include "gen8_generator.h"
52 #include "glsl/glsl_types.h"
53 #include "glsl/ir.h"
54
55 #define MAX_SAMPLER_MESSAGE_SIZE 11
56
57 class bblock_t;
58 namespace {
59 struct acp_entry;
60 }
61
62 namespace brw {
63 class fs_live_variables;
64 }
65
66 class fs_reg {
67 public:
68 DECLARE_RALLOC_CXX_OPERATORS(fs_reg)
69
70 void init();
71
72 fs_reg();
73 fs_reg(float f);
74 fs_reg(int32_t i);
75 fs_reg(uint32_t u);
76 fs_reg(struct brw_reg fixed_hw_reg);
77 fs_reg(enum register_file file, int reg);
78 fs_reg(enum register_file file, int reg, uint32_t type);
79 fs_reg(class fs_visitor *v, const struct glsl_type *type);
80
81 bool equals(const fs_reg &r) const;
82 bool is_zero() const;
83 bool is_one() const;
84 bool is_null() const;
85 bool is_valid_3src() const;
86 bool is_contiguous() const;
87 bool is_accumulator() const;
88
89 fs_reg &apply_stride(unsigned stride);
90 /** Smear a channel of the reg to all channels. */
91 fs_reg &set_smear(unsigned subreg);
92
93 /** Register file: GRF, MRF, IMM. */
94 enum register_file file;
95 /** Register type. BRW_REGISTER_TYPE_* */
96 uint8_t type;
97 /**
98 * Register number. For MRF, it's the hardware register. For
99 * GRF, it's a virtual register number until register allocation
100 */
101 uint16_t reg;
102 /**
103 * Offset from the start of the contiguous register block.
104 *
105 * For pre-register-allocation GRFs, this is in units of a float per pixel
106 * (1 hardware register for SIMD8 mode, or 2 registers for SIMD16 mode).
107 * For uniforms, this is in units of 1 float.
108 */
109 int reg_offset;
110 /**
111 * Offset in bytes from the start of the register. Values up to a
112 * backend_reg::reg_offset unit are valid.
113 */
114 int subreg_offset;
115
116 /** Value for file == IMM */
117 union {
118 int32_t i;
119 uint32_t u;
120 float f;
121 } imm;
122
123 struct brw_reg fixed_hw_reg;
124
125 fs_reg *reladdr;
126
127 bool negate;
128 bool abs;
129
130 /** Register region horizontal stride */
131 uint8_t stride;
132 };
133
134 static inline fs_reg
135 retype(fs_reg reg, unsigned type)
136 {
137 reg.fixed_hw_reg.type = reg.type = type;
138 return reg;
139 }
140
141 static inline fs_reg
142 offset(fs_reg reg, unsigned delta)
143 {
144 assert(delta == 0 || (reg.file != HW_REG && reg.file != IMM));
145 reg.reg_offset += delta;
146 return reg;
147 }
148
149 static inline fs_reg
150 byte_offset(fs_reg reg, unsigned delta)
151 {
152 assert(delta == 0 || (reg.file != HW_REG && reg.file != IMM));
153 reg.subreg_offset += delta;
154 return reg;
155 }
156
157 /**
158 * Get either of the 8-component halves of a 16-component register.
159 *
160 * Note: this also works if \c reg represents a SIMD16 pair of registers.
161 */
162 static inline fs_reg
163 half(const fs_reg &reg, unsigned idx)
164 {
165 assert(idx < 2);
166 assert(idx == 0 || (reg.file != HW_REG && reg.file != IMM));
167 return byte_offset(reg, 8 * idx * reg.stride * type_sz(reg.type));
168 }
169
170 static const fs_reg reg_undef;
171 static const fs_reg reg_null_f(retype(brw_null_reg(), BRW_REGISTER_TYPE_F));
172 static const fs_reg reg_null_d(retype(brw_null_reg(), BRW_REGISTER_TYPE_D));
173 static const fs_reg reg_null_ud(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
174
175 class ip_record : public exec_node {
176 public:
177 DECLARE_RALLOC_CXX_OPERATORS(ip_record)
178
179 ip_record(int ip)
180 {
181 this->ip = ip;
182 }
183
184 int ip;
185 };
186
187 class fs_inst : public backend_instruction {
188 fs_inst &operator=(const fs_inst &);
189
190 public:
191 DECLARE_RALLOC_CXX_OPERATORS(fs_inst)
192
193 void init(enum opcode opcode, const fs_reg &dst, fs_reg *src, int sources);
194
195 fs_inst(enum opcode opcode = BRW_OPCODE_NOP, const fs_reg &dst = reg_undef);
196 fs_inst(enum opcode opcode, const fs_reg &dst, const fs_reg &src0);
197 fs_inst(enum opcode opcode, const fs_reg &dst, const fs_reg &src0,
198 const fs_reg &src1);
199 fs_inst(enum opcode opcode, const fs_reg &dst, const fs_reg &src0,
200 const fs_reg &src1, const fs_reg &src2);
201 fs_inst(enum opcode opcode, const fs_reg &dst, fs_reg src[], int sources);
202 fs_inst(const fs_inst &that);
203
204 void resize_sources(uint8_t num_sources);
205
206 bool equals(fs_inst *inst) const;
207 bool overwrites_reg(const fs_reg &reg) const;
208 bool is_send_from_grf() const;
209 bool is_partial_write() const;
210 int regs_read(fs_visitor *v, int arg) const;
211
212 bool reads_flag() const;
213 bool writes_flag() const;
214
215 fs_reg dst;
216 fs_reg *src;
217
218 uint32_t texture_offset; /**< Texture offset bitfield */
219 uint32_t offset; /* spill/unspill offset */
220
221 uint8_t sources; /**< Number of fs_reg sources. */
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 fs_inst *emit(enum opcode opcode, fs_reg dst,
300 fs_reg src[], int sources);
301
302 fs_inst *MOV(fs_reg dst, fs_reg src);
303 fs_inst *NOT(fs_reg dst, fs_reg src);
304 fs_inst *RNDD(fs_reg dst, fs_reg src);
305 fs_inst *RNDE(fs_reg dst, fs_reg src);
306 fs_inst *RNDZ(fs_reg dst, fs_reg src);
307 fs_inst *FRC(fs_reg dst, fs_reg src);
308 fs_inst *ADD(fs_reg dst, fs_reg src0, fs_reg src1);
309 fs_inst *MUL(fs_reg dst, fs_reg src0, fs_reg src1);
310 fs_inst *MACH(fs_reg dst, fs_reg src0, fs_reg src1);
311 fs_inst *MAC(fs_reg dst, fs_reg src0, fs_reg src1);
312 fs_inst *SHL(fs_reg dst, fs_reg src0, fs_reg src1);
313 fs_inst *SHR(fs_reg dst, fs_reg src0, fs_reg src1);
314 fs_inst *ASR(fs_reg dst, fs_reg src0, fs_reg src1);
315 fs_inst *AND(fs_reg dst, fs_reg src0, fs_reg src1);
316 fs_inst *OR(fs_reg dst, fs_reg src0, fs_reg src1);
317 fs_inst *XOR(fs_reg dst, fs_reg src0, fs_reg src1);
318 fs_inst *IF(uint32_t predicate);
319 fs_inst *IF(fs_reg src0, fs_reg src1, uint32_t condition);
320 fs_inst *CMP(fs_reg dst, fs_reg src0, fs_reg src1,
321 uint32_t condition);
322 fs_inst *LRP(fs_reg dst, fs_reg a, fs_reg y, fs_reg x);
323 fs_inst *DEP_RESOLVE_MOV(int grf);
324 fs_inst *BFREV(fs_reg dst, fs_reg value);
325 fs_inst *BFE(fs_reg dst, fs_reg bits, fs_reg offset, fs_reg value);
326 fs_inst *BFI1(fs_reg dst, fs_reg bits, fs_reg offset);
327 fs_inst *BFI2(fs_reg dst, fs_reg bfi1_dst, fs_reg insert, fs_reg base);
328 fs_inst *FBH(fs_reg dst, fs_reg value);
329 fs_inst *FBL(fs_reg dst, fs_reg value);
330 fs_inst *CBIT(fs_reg dst, fs_reg value);
331 fs_inst *MAD(fs_reg dst, fs_reg c, fs_reg b, fs_reg a);
332 fs_inst *ADDC(fs_reg dst, fs_reg src0, fs_reg src1);
333 fs_inst *SUBB(fs_reg dst, fs_reg src0, fs_reg src1);
334 fs_inst *SEL(fs_reg dst, fs_reg src0, fs_reg src1);
335
336 int type_size(const struct glsl_type *type);
337 fs_inst *get_instruction_generating_reg(fs_inst *start,
338 fs_inst *end,
339 const fs_reg &reg);
340
341 exec_list VARYING_PULL_CONSTANT_LOAD(const fs_reg &dst,
342 const fs_reg &surf_index,
343 const fs_reg &varying_offset,
344 uint32_t const_offset);
345
346 bool run();
347 void assign_binding_table_offsets();
348 void setup_payload_gen4();
349 void setup_payload_gen6();
350 void assign_curb_setup();
351 void calculate_urb_setup();
352 void assign_urb_setup();
353 bool assign_regs(bool allow_spilling);
354 void assign_regs_trivial();
355 void get_used_mrfs(bool *mrf_used);
356 void setup_payload_interference(struct ra_graph *g, int payload_reg_count,
357 int first_payload_node);
358 void setup_mrf_hack_interference(struct ra_graph *g,
359 int first_mrf_hack_node);
360 int choose_spill_reg(struct ra_graph *g);
361 void spill_reg(int spill_reg);
362 void split_virtual_grfs();
363 void compact_virtual_grfs();
364 void move_uniform_array_access_to_pull_constants();
365 void assign_constant_locations();
366 void demote_pull_constants();
367 void invalidate_live_intervals();
368 void calculate_live_intervals();
369 void calculate_register_pressure();
370 bool opt_algebraic();
371 bool opt_cse();
372 bool opt_cse_local(bblock_t *block, exec_list *aeb);
373 bool opt_copy_propagate();
374 bool try_copy_propagate(fs_inst *inst, int arg, acp_entry *entry);
375 bool try_constant_propagate(fs_inst *inst, acp_entry *entry);
376 bool opt_copy_propagate_local(void *mem_ctx, bblock_t *block,
377 exec_list *acp);
378 void opt_drop_redundant_mov_to_flags();
379 bool register_coalesce();
380 bool compute_to_mrf();
381 bool dead_code_eliminate();
382 bool remove_duplicate_mrf_writes();
383 bool virtual_grf_interferes(int a, int b);
384 void schedule_instructions(instruction_scheduler_mode mode);
385 void insert_gen4_send_dependency_workarounds();
386 void insert_gen4_pre_send_dependency_workarounds(fs_inst *inst);
387 void insert_gen4_post_send_dependency_workarounds(fs_inst *inst);
388 void vfail(const char *msg, va_list args);
389 void fail(const char *msg, ...);
390 void no16(const char *msg, ...);
391 void lower_uniform_pull_constant_loads();
392
393 void push_force_uncompressed();
394 void pop_force_uncompressed();
395
396 void emit_dummy_fs();
397 fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
398 fs_inst *emit_linterp(const fs_reg &attr, const fs_reg &interp,
399 glsl_interp_qualifier interpolation_mode,
400 bool is_centroid, bool is_sample);
401 fs_reg *emit_frontfacing_interpolation(ir_variable *ir);
402 fs_reg *emit_samplepos_setup(ir_variable *ir);
403 fs_reg *emit_sampleid_setup(ir_variable *ir);
404 fs_reg *emit_general_interpolation(ir_variable *ir);
405 void emit_interpolation_setup_gen4();
406 void emit_interpolation_setup_gen6();
407 void compute_sample_position(fs_reg dst, fs_reg int_sample_pos);
408 fs_reg rescale_texcoord(ir_texture *ir, fs_reg coordinate,
409 bool is_rect, int sampler, int texunit);
410 fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate,
411 fs_reg shadow_comp, fs_reg lod, fs_reg lod2);
412 fs_inst *emit_texture_gen5(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);
415 fs_inst *emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
416 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
417 fs_reg sample_index, fs_reg mcs, int sampler);
418 fs_reg emit_mcs_fetch(ir_texture *ir, fs_reg coordinate, int sampler);
419 void emit_gen6_gather_wa(uint8_t wa, fs_reg dst);
420 fs_reg fix_math_operand(fs_reg src);
421 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0);
422 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0, fs_reg src1);
423 void emit_lrp(const fs_reg &dst, const fs_reg &x, const fs_reg &y,
424 const fs_reg &a);
425 void emit_minmax(uint32_t conditionalmod, const fs_reg &dst,
426 const fs_reg &src0, const fs_reg &src1);
427 bool try_emit_saturate(ir_expression *ir);
428 bool try_emit_mad(ir_expression *ir);
429 void try_replace_with_sel();
430 bool opt_peephole_sel();
431 bool opt_peephole_predicated_break();
432 bool opt_saturate_propagation();
433 void emit_bool_to_cond_code(ir_rvalue *condition);
434 void emit_if_gen6(ir_if *ir);
435 void emit_unspill(fs_inst *inst, fs_reg reg, uint32_t spill_offset,
436 int count);
437
438 void emit_fragment_program_code();
439 void setup_fp_regs();
440 fs_reg get_fp_src_reg(const prog_src_register *src);
441 fs_reg get_fp_dst_reg(const prog_dst_register *dst);
442 void emit_fp_alu1(enum opcode opcode,
443 const struct prog_instruction *fpi,
444 fs_reg dst, fs_reg src);
445 void emit_fp_alu2(enum opcode opcode,
446 const struct prog_instruction *fpi,
447 fs_reg dst, fs_reg src0, fs_reg src1);
448 void emit_fp_scalar_write(const struct prog_instruction *fpi,
449 fs_reg dst, fs_reg src);
450 void emit_fp_scalar_math(enum opcode opcode,
451 const struct prog_instruction *fpi,
452 fs_reg dst, fs_reg src);
453
454 void emit_fp_minmax(const struct prog_instruction *fpi,
455 fs_reg dst, fs_reg src0, fs_reg src1);
456
457 void emit_fp_sop(uint32_t conditional_mod,
458 const struct prog_instruction *fpi,
459 fs_reg dst, fs_reg src0, fs_reg src1, fs_reg one);
460
461 void emit_color_write(int target, int index, int first_color_mrf);
462 void emit_alpha_test();
463 void emit_fb_writes();
464
465 void emit_shader_time_begin();
466 void emit_shader_time_end();
467 void emit_shader_time_write(enum shader_time_shader_type type,
468 fs_reg value);
469
470 void emit_untyped_atomic(unsigned atomic_op, unsigned surf_index,
471 fs_reg dst, fs_reg offset, fs_reg src0,
472 fs_reg src1);
473
474 void emit_untyped_surface_read(unsigned surf_index, fs_reg dst,
475 fs_reg offset);
476
477 bool try_rewrite_rhs_to_dst(ir_assignment *ir,
478 fs_reg dst,
479 fs_reg src,
480 fs_inst *pre_rhs_inst,
481 fs_inst *last_rhs_inst);
482 void emit_assignment_writes(fs_reg &l, fs_reg &r,
483 const glsl_type *type, bool predicated);
484 void resolve_ud_negate(fs_reg *reg);
485 void resolve_bool_comparison(ir_rvalue *rvalue, fs_reg *reg);
486
487 fs_reg get_timestamp();
488
489 struct brw_reg interp_reg(int location, int channel);
490 void setup_uniform_values(ir_variable *ir);
491 void setup_builtin_uniform_values(ir_variable *ir);
492 int implied_mrf_writes(fs_inst *inst);
493
494 virtual void dump_instructions();
495 virtual void dump_instructions(const char *name);
496 void dump_instruction(backend_instruction *inst);
497 void dump_instruction(backend_instruction *inst, FILE *file);
498
499 void visit_atomic_counter_intrinsic(ir_call *ir);
500
501 struct gl_fragment_program *fp;
502 const struct brw_wm_prog_key *const key;
503 struct brw_wm_prog_data *prog_data;
504 unsigned int sanity_param_count;
505
506 int *param_size;
507
508 int *virtual_grf_sizes;
509 int virtual_grf_count;
510 int virtual_grf_array_size;
511 int *virtual_grf_start;
512 int *virtual_grf_end;
513 brw::fs_live_variables *live_intervals;
514
515 int *regs_live_at_ip;
516
517 /** Number of uniform variable components visited. */
518 unsigned uniforms;
519
520 /** Byte-offset for the next available spot in the scratch space buffer. */
521 unsigned last_scratch;
522
523 /**
524 * Array mapping UNIFORM register numbers to the pull parameter index,
525 * or -1 if this uniform register isn't being uploaded as a pull constant.
526 */
527 int *pull_constant_loc;
528
529 /**
530 * Array mapping UNIFORM register numbers to the push parameter index,
531 * or -1 if this uniform register isn't being uploaded as a push constant.
532 */
533 int *push_constant_loc;
534
535 struct hash_table *variable_ht;
536 fs_reg frag_depth;
537 fs_reg sample_mask;
538 fs_reg outputs[BRW_MAX_DRAW_BUFFERS];
539 unsigned output_components[BRW_MAX_DRAW_BUFFERS];
540 fs_reg dual_src_output;
541 bool do_dual_src;
542 int first_non_payload_grf;
543 /** Either BRW_MAX_GRF or GEN7_MRF_HACK_START */
544 int max_grf;
545
546 fs_reg *fp_temp_regs;
547 fs_reg *fp_input_regs;
548
549 /** @{ debug annotation info */
550 const char *current_annotation;
551 const void *base_ir;
552 /** @} */
553
554 bool failed;
555 char *fail_msg;
556 bool simd16_unsupported;
557 char *no16_msg;
558
559 /* Result of last visit() method. */
560 fs_reg result;
561
562 /** Register numbers for thread payload fields. */
563 struct {
564 uint8_t source_depth_reg;
565 uint8_t source_w_reg;
566 uint8_t aa_dest_stencil_reg;
567 uint8_t dest_depth_reg;
568 uint8_t sample_pos_reg;
569 uint8_t sample_mask_in_reg;
570 uint8_t barycentric_coord_reg[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
571
572 /** The number of thread payload registers the hardware will supply. */
573 uint8_t num_regs;
574 } payload;
575
576 bool source_depth_to_render_target;
577 bool runtime_check_aads_emit;
578
579 fs_reg pixel_x;
580 fs_reg pixel_y;
581 fs_reg wpos_w;
582 fs_reg pixel_w;
583 fs_reg delta_x[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
584 fs_reg delta_y[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
585 fs_reg shader_start_time;
586
587 int grf_used;
588 bool spilled_any_registers;
589
590 const unsigned dispatch_width; /**< 8 or 16 */
591
592 int force_uncompressed_stack;
593 };
594
595 /**
596 * The fragment shader code generator.
597 *
598 * Translates FS IR to actual i965 assembly code.
599 */
600 class fs_generator
601 {
602 public:
603 fs_generator(struct brw_context *brw,
604 void *mem_ctx,
605 const struct brw_wm_prog_key *key,
606 struct brw_wm_prog_data *prog_data,
607 struct gl_shader_program *prog,
608 struct gl_fragment_program *fp,
609 bool dual_source_output,
610 bool debug_flag);
611 ~fs_generator();
612
613 const unsigned *generate_assembly(exec_list *simd8_instructions,
614 exec_list *simd16_instructions,
615 unsigned *assembly_size);
616
617 private:
618 void generate_code(exec_list *instructions);
619 void generate_fb_write(fs_inst *inst);
620 void generate_blorp_fb_write(fs_inst *inst);
621 void generate_pixel_xy(struct brw_reg dst, bool is_x);
622 void generate_linterp(fs_inst *inst, struct brw_reg dst,
623 struct brw_reg *src);
624 void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
625 void generate_math1_gen7(fs_inst *inst,
626 struct brw_reg dst,
627 struct brw_reg src);
628 void generate_math2_gen7(fs_inst *inst,
629 struct brw_reg dst,
630 struct brw_reg src0,
631 struct brw_reg src1);
632 void generate_math1_gen6(fs_inst *inst,
633 struct brw_reg dst,
634 struct brw_reg src);
635 void generate_math2_gen6(fs_inst *inst,
636 struct brw_reg dst,
637 struct brw_reg src0,
638 struct brw_reg src1);
639 void generate_math_gen4(fs_inst *inst,
640 struct brw_reg dst,
641 struct brw_reg src);
642 void generate_math_g45(fs_inst *inst,
643 struct brw_reg dst,
644 struct brw_reg src);
645 void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
646 void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
647 bool negate_value);
648 void generate_scratch_write(fs_inst *inst, struct brw_reg src);
649 void generate_scratch_read(fs_inst *inst, struct brw_reg dst);
650 void generate_scratch_read_gen7(fs_inst *inst, struct brw_reg dst);
651 void generate_uniform_pull_constant_load(fs_inst *inst, struct brw_reg dst,
652 struct brw_reg index,
653 struct brw_reg offset);
654 void generate_uniform_pull_constant_load_gen7(fs_inst *inst,
655 struct brw_reg dst,
656 struct brw_reg surf_index,
657 struct brw_reg offset);
658 void generate_varying_pull_constant_load(fs_inst *inst, struct brw_reg dst,
659 struct brw_reg index,
660 struct brw_reg offset);
661 void generate_varying_pull_constant_load_gen7(fs_inst *inst,
662 struct brw_reg dst,
663 struct brw_reg index,
664 struct brw_reg offset);
665 void generate_mov_dispatch_to_flags(fs_inst *inst);
666
667 void generate_set_omask(fs_inst *inst,
668 struct brw_reg dst,
669 struct brw_reg sample_mask);
670
671 void generate_set_sample_id(fs_inst *inst,
672 struct brw_reg dst,
673 struct brw_reg src0,
674 struct brw_reg src1);
675
676 void generate_set_simd4x2_offset(fs_inst *inst,
677 struct brw_reg dst,
678 struct brw_reg offset);
679 void generate_discard_jump(fs_inst *inst);
680
681 void generate_pack_half_2x16_split(fs_inst *inst,
682 struct brw_reg dst,
683 struct brw_reg x,
684 struct brw_reg y);
685 void generate_unpack_half_2x16_split(fs_inst *inst,
686 struct brw_reg dst,
687 struct brw_reg src);
688
689 void generate_shader_time_add(fs_inst *inst,
690 struct brw_reg payload,
691 struct brw_reg offset,
692 struct brw_reg value);
693
694 void generate_untyped_atomic(fs_inst *inst,
695 struct brw_reg dst,
696 struct brw_reg atomic_op,
697 struct brw_reg surf_index);
698
699 void generate_untyped_surface_read(fs_inst *inst,
700 struct brw_reg dst,
701 struct brw_reg surf_index);
702
703 bool patch_discard_jumps_to_fb_writes();
704
705 struct brw_context *brw;
706 struct gl_context *ctx;
707
708 struct brw_compile *p;
709 const struct brw_wm_prog_key *const key;
710 struct brw_wm_prog_data *prog_data;
711
712 struct gl_shader_program *prog;
713 const struct gl_fragment_program *fp;
714
715 unsigned dispatch_width; /**< 8 or 16 */
716
717 exec_list discard_halt_patches;
718 bool dual_source_output;
719 const bool debug_flag;
720 void *mem_ctx;
721 };
722
723 /**
724 * The fragment shader code generator.
725 *
726 * Translates FS IR to actual i965 assembly code.
727 */
728 class gen8_fs_generator : public gen8_generator
729 {
730 public:
731 gen8_fs_generator(struct brw_context *brw,
732 void *mem_ctx,
733 const struct brw_wm_prog_key *key,
734 struct brw_wm_prog_data *prog_data,
735 struct gl_shader_program *prog,
736 struct gl_fragment_program *fp,
737 bool dual_source_output);
738 ~gen8_fs_generator();
739
740 const unsigned *generate_assembly(exec_list *simd8_instructions,
741 exec_list *simd16_instructions,
742 unsigned *assembly_size);
743
744 private:
745 void generate_code(exec_list *instructions);
746 void generate_fb_write(fs_inst *inst);
747 void generate_linterp(fs_inst *inst, struct brw_reg dst,
748 struct brw_reg *src);
749 void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
750 void generate_math1(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
751 void generate_math2(fs_inst *inst, struct brw_reg dst,
752 struct brw_reg src0, struct brw_reg src1);
753 void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
754 void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
755 bool negate_value);
756 void generate_scratch_write(fs_inst *inst, struct brw_reg src);
757 void generate_scratch_read(fs_inst *inst, struct brw_reg dst);
758 void generate_scratch_read_gen7(fs_inst *inst, struct brw_reg dst);
759 void generate_uniform_pull_constant_load(fs_inst *inst,
760 struct brw_reg dst,
761 struct brw_reg index,
762 struct brw_reg offset);
763 void generate_varying_pull_constant_load(fs_inst *inst,
764 struct brw_reg dst,
765 struct brw_reg index,
766 struct brw_reg offset);
767 void generate_mov_dispatch_to_flags(fs_inst *ir);
768 void generate_set_omask(fs_inst *ir,
769 struct brw_reg dst,
770 struct brw_reg sample_mask);
771 void generate_set_sample_id(fs_inst *ir,
772 struct brw_reg dst,
773 struct brw_reg src0,
774 struct brw_reg src1);
775 void generate_set_simd4x2_offset(fs_inst *ir,
776 struct brw_reg dst,
777 struct brw_reg offset);
778 void generate_pack_half_2x16_split(fs_inst *inst,
779 struct brw_reg dst,
780 struct brw_reg x,
781 struct brw_reg y);
782 void generate_unpack_half_2x16_split(fs_inst *inst,
783 struct brw_reg dst,
784 struct brw_reg src);
785 void generate_untyped_atomic(fs_inst *inst,
786 struct brw_reg dst,
787 struct brw_reg atomic_op,
788 struct brw_reg surf_index);
789
790 void generate_untyped_surface_read(fs_inst *inst,
791 struct brw_reg dst,
792 struct brw_reg surf_index);
793 void generate_discard_jump(fs_inst *ir);
794
795 bool patch_discard_jumps_to_fb_writes();
796
797 const struct brw_wm_prog_key *const key;
798 struct brw_wm_prog_data *prog_data;
799 const struct gl_fragment_program *fp;
800
801 unsigned dispatch_width; /** 8 or 16 */
802
803 bool dual_source_output;
804
805 exec_list discard_halt_patches;
806 };
807
808 bool brw_do_channel_expressions(struct exec_list *instructions);
809 bool brw_do_vector_splitting(struct exec_list *instructions);
810 bool brw_fs_precompile(struct gl_context *ctx, struct gl_shader_program *prog);
811
812 struct brw_reg brw_reg_from_fs_reg(fs_reg *reg);