i965/fs: Split "find what MRFs were used" to a helper function.
[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 "glsl/glsl_types.h"
51 #include "glsl/ir.h"
52
53 class bblock_t;
54 namespace {
55 struct acp_entry;
56 }
57
58 namespace brw {
59 class fs_live_variables;
60 }
61
62 class fs_reg {
63 public:
64 DECLARE_RALLOC_CXX_OPERATORS(fs_reg)
65
66 void init();
67
68 fs_reg();
69 fs_reg(float f);
70 fs_reg(int32_t i);
71 fs_reg(uint32_t u);
72 fs_reg(struct brw_reg fixed_hw_reg);
73 fs_reg(enum register_file file, int reg);
74 fs_reg(enum register_file file, int reg, uint32_t type);
75 fs_reg(class fs_visitor *v, const struct glsl_type *type);
76
77 bool equals(const fs_reg &r) const;
78 bool is_zero() const;
79 bool is_one() const;
80 bool is_valid_3src() const;
81 fs_reg retype(uint32_t type);
82
83 /** Register file: GRF, MRF, IMM. */
84 enum register_file file;
85 /**
86 * Register number. For MRF, it's the hardware register. For
87 * GRF, it's a virtual register number until register allocation
88 */
89 int reg;
90 /**
91 * Offset from the start of the contiguous register block.
92 *
93 * For pre-register-allocation GRFs, this is in units of a float per pixel
94 * (1 hardware register for SIMD8 mode, or 2 registers for SIMD16 mode).
95 * For uniforms, this is in units of 1 float.
96 */
97 int reg_offset;
98 /** Register type. BRW_REGISTER_TYPE_* */
99 int type;
100 bool negate;
101 bool abs;
102 bool sechalf;
103 struct brw_reg fixed_hw_reg;
104 int smear; /* -1, or a channel of the reg to smear to all channels. */
105
106 /** Value for file == IMM */
107 union {
108 int32_t i;
109 uint32_t u;
110 float f;
111 } imm;
112
113 fs_reg *reladdr;
114 };
115
116 static const fs_reg reg_undef;
117 static const fs_reg reg_null_f(retype(brw_null_reg(), BRW_REGISTER_TYPE_F));
118 static const fs_reg reg_null_d(retype(brw_null_reg(), BRW_REGISTER_TYPE_D));
119 static const fs_reg reg_null_ud(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
120
121 class ip_record : public exec_node {
122 public:
123 DECLARE_RALLOC_CXX_OPERATORS(ip_record)
124
125 ip_record(int ip)
126 {
127 this->ip = ip;
128 }
129
130 int ip;
131 };
132
133 class fs_inst : public backend_instruction {
134 public:
135 DECLARE_RALLOC_CXX_OPERATORS(fs_inst)
136
137 void init();
138
139 fs_inst();
140 fs_inst(enum opcode opcode);
141 fs_inst(enum opcode opcode, fs_reg dst);
142 fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0);
143 fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1);
144 fs_inst(enum opcode opcode, fs_reg dst,
145 fs_reg src0, fs_reg src1,fs_reg src2);
146
147 bool equals(fs_inst *inst);
148 bool overwrites_reg(const fs_reg &reg);
149 bool is_send_from_grf();
150 bool is_partial_write();
151 int regs_read(fs_visitor *v, int arg);
152
153 fs_reg dst;
154 fs_reg src[3];
155 bool saturate;
156 int conditional_mod; /**< BRW_CONDITIONAL_* */
157
158 /* Chooses which flag subregister (f0.0 or f0.1) is used for conditional
159 * mod and predication.
160 */
161 uint8_t flag_subreg;
162
163 int mlen; /**< SEND message length */
164 int regs_written; /**< Number of vgrfs written by a SEND message, or 1 */
165 int base_mrf; /**< First MRF in the SEND message, if mlen is nonzero. */
166 uint32_t texture_offset; /**< Texture offset bitfield */
167 int sampler;
168 int target; /**< MRT target. */
169 bool eot;
170 bool header_present;
171 bool shadow_compare;
172 bool force_uncompressed;
173 bool force_sechalf;
174 bool force_writemask_all;
175 uint32_t offset; /* spill/unspill offset */
176
177 /** @{
178 * Annotation for the generated IR. One of the two can be set.
179 */
180 const void *ir;
181 const char *annotation;
182 /** @} */
183 };
184
185 /**
186 * The fragment shader front-end.
187 *
188 * Translates either GLSL IR or Mesa IR (for ARB_fragment_program) into FS IR.
189 */
190 class fs_visitor : public backend_visitor
191 {
192 public:
193
194 fs_visitor(struct brw_context *brw,
195 struct brw_wm_compile *c,
196 struct gl_shader_program *shader_prog,
197 struct gl_fragment_program *fp,
198 unsigned dispatch_width);
199 ~fs_visitor();
200
201 fs_reg *variable_storage(ir_variable *var);
202 int virtual_grf_alloc(int size);
203 void import_uniforms(fs_visitor *v);
204
205 void visit(ir_variable *ir);
206 void visit(ir_assignment *ir);
207 void visit(ir_dereference_variable *ir);
208 void visit(ir_dereference_record *ir);
209 void visit(ir_dereference_array *ir);
210 void visit(ir_expression *ir);
211 void visit(ir_texture *ir);
212 void visit(ir_if *ir);
213 void visit(ir_constant *ir);
214 void visit(ir_swizzle *ir);
215 void visit(ir_return *ir);
216 void visit(ir_loop *ir);
217 void visit(ir_loop_jump *ir);
218 void visit(ir_discard *ir);
219 void visit(ir_call *ir);
220 void visit(ir_function *ir);
221 void visit(ir_function_signature *ir);
222 void visit(ir_emit_vertex *);
223 void visit(ir_end_primitive *);
224
225 uint32_t gather_channel(ir_texture *ir, int sampler);
226 void swizzle_result(ir_texture *ir, fs_reg orig_val, int sampler);
227
228 bool can_do_source_mods(fs_inst *inst);
229
230 fs_inst *emit(fs_inst inst);
231 fs_inst *emit(fs_inst *inst);
232 void emit(exec_list list);
233
234 fs_inst *emit(enum opcode opcode);
235 fs_inst *emit(enum opcode opcode, fs_reg dst);
236 fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0);
237 fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1);
238 fs_inst *emit(enum opcode opcode, fs_reg dst,
239 fs_reg src0, fs_reg src1, fs_reg src2);
240
241 fs_inst *MOV(fs_reg dst, fs_reg src);
242 fs_inst *NOT(fs_reg dst, fs_reg src);
243 fs_inst *RNDD(fs_reg dst, fs_reg src);
244 fs_inst *RNDE(fs_reg dst, fs_reg src);
245 fs_inst *RNDZ(fs_reg dst, fs_reg src);
246 fs_inst *FRC(fs_reg dst, fs_reg src);
247 fs_inst *ADD(fs_reg dst, fs_reg src0, fs_reg src1);
248 fs_inst *MUL(fs_reg dst, fs_reg src0, fs_reg src1);
249 fs_inst *MACH(fs_reg dst, fs_reg src0, fs_reg src1);
250 fs_inst *MAC(fs_reg dst, fs_reg src0, fs_reg src1);
251 fs_inst *SHL(fs_reg dst, fs_reg src0, fs_reg src1);
252 fs_inst *SHR(fs_reg dst, fs_reg src0, fs_reg src1);
253 fs_inst *ASR(fs_reg dst, fs_reg src0, fs_reg src1);
254 fs_inst *AND(fs_reg dst, fs_reg src0, fs_reg src1);
255 fs_inst *OR(fs_reg dst, fs_reg src0, fs_reg src1);
256 fs_inst *XOR(fs_reg dst, fs_reg src0, fs_reg src1);
257 fs_inst *IF(uint32_t predicate);
258 fs_inst *IF(fs_reg src0, fs_reg src1, uint32_t condition);
259 fs_inst *CMP(fs_reg dst, fs_reg src0, fs_reg src1,
260 uint32_t condition);
261 fs_inst *LRP(fs_reg dst, fs_reg a, fs_reg y, fs_reg x);
262 fs_inst *DEP_RESOLVE_MOV(int grf);
263 fs_inst *BFREV(fs_reg dst, fs_reg value);
264 fs_inst *BFE(fs_reg dst, fs_reg bits, fs_reg offset, fs_reg value);
265 fs_inst *BFI1(fs_reg dst, fs_reg bits, fs_reg offset);
266 fs_inst *BFI2(fs_reg dst, fs_reg bfi1_dst, fs_reg insert, fs_reg base);
267 fs_inst *FBH(fs_reg dst, fs_reg value);
268 fs_inst *FBL(fs_reg dst, fs_reg value);
269 fs_inst *CBIT(fs_reg dst, fs_reg value);
270 fs_inst *MAD(fs_reg dst, fs_reg c, fs_reg b, fs_reg a);
271 fs_inst *ADDC(fs_reg dst, fs_reg src0, fs_reg src1);
272 fs_inst *SUBB(fs_reg dst, fs_reg src0, fs_reg src1);
273
274 int type_size(const struct glsl_type *type);
275 fs_inst *get_instruction_generating_reg(fs_inst *start,
276 fs_inst *end,
277 fs_reg reg);
278
279 exec_list VARYING_PULL_CONSTANT_LOAD(fs_reg dst, fs_reg surf_index,
280 fs_reg varying_offset,
281 uint32_t const_offset);
282
283 bool run();
284 void assign_binding_table_offsets();
285 void setup_payload_gen4();
286 void setup_payload_gen6();
287 void assign_curb_setup();
288 void calculate_urb_setup();
289 void assign_urb_setup();
290 bool assign_regs();
291 void assign_regs_trivial();
292 void get_used_mrfs(bool *mrf_used);
293 void setup_payload_interference(struct ra_graph *g, int payload_reg_count,
294 int first_payload_node);
295 void setup_mrf_hack_interference(struct ra_graph *g,
296 int first_mrf_hack_node);
297 int choose_spill_reg(struct ra_graph *g);
298 void spill_reg(int spill_reg);
299 void split_virtual_grfs();
300 void compact_virtual_grfs();
301 void move_uniform_array_access_to_pull_constants();
302 void setup_pull_constants();
303 void invalidate_live_intervals();
304 void calculate_live_intervals();
305 bool opt_algebraic();
306 bool opt_cse();
307 bool opt_cse_local(bblock_t *block, exec_list *aeb);
308 bool opt_copy_propagate();
309 bool try_copy_propagate(fs_inst *inst, int arg, acp_entry *entry);
310 bool try_constant_propagate(fs_inst *inst, acp_entry *entry);
311 bool opt_copy_propagate_local(void *mem_ctx, bblock_t *block,
312 exec_list *acp);
313 bool register_coalesce();
314 bool register_coalesce_2();
315 bool compute_to_mrf();
316 bool dead_code_eliminate();
317 bool dead_code_eliminate_local();
318 bool remove_dead_constants();
319 bool remove_duplicate_mrf_writes();
320 bool virtual_grf_interferes(int a, int b);
321 void schedule_instructions(bool post_reg_alloc);
322 void insert_gen4_send_dependency_workarounds();
323 void insert_gen4_pre_send_dependency_workarounds(fs_inst *inst);
324 void insert_gen4_post_send_dependency_workarounds(fs_inst *inst);
325 void fail(const char *msg, ...);
326 void lower_uniform_pull_constant_loads();
327
328 void push_force_uncompressed();
329 void pop_force_uncompressed();
330 void push_force_sechalf();
331 void pop_force_sechalf();
332
333 void emit_dummy_fs();
334 fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
335 fs_inst *emit_linterp(const fs_reg &attr, const fs_reg &interp,
336 glsl_interp_qualifier interpolation_mode,
337 bool is_centroid);
338 fs_reg *emit_frontfacing_interpolation(ir_variable *ir);
339 fs_reg *emit_general_interpolation(ir_variable *ir);
340 void emit_interpolation_setup_gen4();
341 void emit_interpolation_setup_gen6();
342 fs_reg rescale_texcoord(ir_texture *ir, fs_reg coordinate,
343 bool is_rect, int sampler, int texunit);
344 fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate,
345 fs_reg shadow_comp, fs_reg lod, fs_reg lod2);
346 fs_inst *emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate,
347 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
348 fs_reg sample_index);
349 fs_inst *emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
350 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
351 fs_reg sample_index);
352 fs_reg fix_math_operand(fs_reg src);
353 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0);
354 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0, fs_reg src1);
355 void emit_lrp(fs_reg dst, fs_reg x, fs_reg y, fs_reg a);
356 void emit_minmax(uint32_t conditionalmod, fs_reg dst,
357 fs_reg src0, fs_reg src1);
358 bool try_emit_saturate(ir_expression *ir);
359 bool try_emit_mad(ir_expression *ir, int mul_arg);
360 void try_replace_with_sel();
361 void emit_bool_to_cond_code(ir_rvalue *condition);
362 void emit_if_gen6(ir_if *ir);
363 void emit_unspill(fs_inst *inst, fs_reg reg, uint32_t spill_offset,
364 int count);
365
366 void emit_fragment_program_code();
367 void setup_fp_regs();
368 fs_reg get_fp_src_reg(const prog_src_register *src);
369 fs_reg get_fp_dst_reg(const prog_dst_register *dst);
370 void emit_fp_alu1(enum opcode opcode,
371 const struct prog_instruction *fpi,
372 fs_reg dst, fs_reg src);
373 void emit_fp_alu2(enum opcode opcode,
374 const struct prog_instruction *fpi,
375 fs_reg dst, fs_reg src0, fs_reg src1);
376 void emit_fp_scalar_write(const struct prog_instruction *fpi,
377 fs_reg dst, fs_reg src);
378 void emit_fp_scalar_math(enum opcode opcode,
379 const struct prog_instruction *fpi,
380 fs_reg dst, fs_reg src);
381
382 void emit_fp_minmax(const struct prog_instruction *fpi,
383 fs_reg dst, fs_reg src0, fs_reg src1);
384
385 void emit_fp_sop(uint32_t conditional_mod,
386 const struct prog_instruction *fpi,
387 fs_reg dst, fs_reg src0, fs_reg src1, fs_reg one);
388
389 void emit_color_write(int target, int index, int first_color_mrf);
390 void emit_fb_writes();
391
392 void emit_shader_time_begin();
393 void emit_shader_time_end();
394 void emit_shader_time_write(enum shader_time_shader_type type,
395 fs_reg value);
396
397 bool try_rewrite_rhs_to_dst(ir_assignment *ir,
398 fs_reg dst,
399 fs_reg src,
400 fs_inst *pre_rhs_inst,
401 fs_inst *last_rhs_inst);
402 void emit_assignment_writes(fs_reg &l, fs_reg &r,
403 const glsl_type *type, bool predicated);
404 void resolve_ud_negate(fs_reg *reg);
405 void resolve_bool_comparison(ir_rvalue *rvalue, fs_reg *reg);
406
407 fs_reg get_timestamp();
408
409 struct brw_reg interp_reg(int location, int channel);
410 void setup_uniform_values(ir_variable *ir);
411 void setup_builtin_uniform_values(ir_variable *ir);
412 int implied_mrf_writes(fs_inst *inst);
413
414 void dump_instruction(backend_instruction *inst);
415
416 struct gl_fragment_program *fp;
417 struct brw_wm_compile *c;
418 unsigned int sanity_param_count;
419
420 int param_size[MAX_UNIFORMS * 4];
421
422 int *virtual_grf_sizes;
423 int virtual_grf_count;
424 int virtual_grf_array_size;
425 int *virtual_grf_start;
426 int *virtual_grf_end;
427 brw::fs_live_variables *live_intervals;
428
429 /* This is the map from UNIFORM hw_reg + reg_offset as generated by
430 * the visitor to the packed uniform number after
431 * remove_dead_constants() that represents the actual uploaded
432 * uniform index.
433 */
434 int *params_remap;
435 int nr_params_remap;
436
437 struct hash_table *variable_ht;
438 fs_reg frag_depth;
439 fs_reg outputs[BRW_MAX_DRAW_BUFFERS];
440 unsigned output_components[BRW_MAX_DRAW_BUFFERS];
441 fs_reg dual_src_output;
442 int first_non_payload_grf;
443 /** Either BRW_MAX_GRF or GEN7_MRF_HACK_START */
444 int max_grf;
445
446 fs_reg *fp_temp_regs;
447 fs_reg *fp_input_regs;
448
449 /** @{ debug annotation info */
450 const char *current_annotation;
451 const void *base_ir;
452 /** @} */
453
454 bool failed;
455 char *fail_msg;
456
457 /* Result of last visit() method. */
458 fs_reg result;
459
460 fs_reg pixel_x;
461 fs_reg pixel_y;
462 fs_reg wpos_w;
463 fs_reg pixel_w;
464 fs_reg delta_x[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
465 fs_reg delta_y[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
466 fs_reg shader_start_time;
467
468 int grf_used;
469
470 const unsigned dispatch_width; /**< 8 or 16 */
471
472 int force_uncompressed_stack;
473 int force_sechalf_stack;
474 };
475
476 /**
477 * The fragment shader code generator.
478 *
479 * Translates FS IR to actual i965 assembly code.
480 */
481 class fs_generator
482 {
483 public:
484 fs_generator(struct brw_context *brw,
485 struct brw_wm_compile *c,
486 struct gl_shader_program *prog,
487 struct gl_fragment_program *fp,
488 bool dual_source_output);
489 ~fs_generator();
490
491 const unsigned *generate_assembly(exec_list *simd8_instructions,
492 exec_list *simd16_instructions,
493 unsigned *assembly_size);
494
495 private:
496 void generate_code(exec_list *instructions);
497 void generate_fb_write(fs_inst *inst);
498 void generate_pixel_xy(struct brw_reg dst, bool is_x);
499 void generate_linterp(fs_inst *inst, struct brw_reg dst,
500 struct brw_reg *src);
501 void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
502 void generate_math1_gen7(fs_inst *inst,
503 struct brw_reg dst,
504 struct brw_reg src);
505 void generate_math2_gen7(fs_inst *inst,
506 struct brw_reg dst,
507 struct brw_reg src0,
508 struct brw_reg src1);
509 void generate_math1_gen6(fs_inst *inst,
510 struct brw_reg dst,
511 struct brw_reg src);
512 void generate_math2_gen6(fs_inst *inst,
513 struct brw_reg dst,
514 struct brw_reg src0,
515 struct brw_reg src1);
516 void generate_math_gen4(fs_inst *inst,
517 struct brw_reg dst,
518 struct brw_reg src);
519 void generate_math_g45(fs_inst *inst,
520 struct brw_reg dst,
521 struct brw_reg src);
522 void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
523 void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
524 bool negate_value);
525 void generate_spill(fs_inst *inst, struct brw_reg src);
526 void generate_unspill(fs_inst *inst, struct brw_reg dst);
527 void generate_uniform_pull_constant_load(fs_inst *inst, struct brw_reg dst,
528 struct brw_reg index,
529 struct brw_reg offset);
530 void generate_uniform_pull_constant_load_gen7(fs_inst *inst,
531 struct brw_reg dst,
532 struct brw_reg surf_index,
533 struct brw_reg offset);
534 void generate_varying_pull_constant_load(fs_inst *inst, struct brw_reg dst,
535 struct brw_reg index,
536 struct brw_reg offset);
537 void generate_varying_pull_constant_load_gen7(fs_inst *inst,
538 struct brw_reg dst,
539 struct brw_reg index,
540 struct brw_reg offset);
541 void generate_mov_dispatch_to_flags(fs_inst *inst);
542 void generate_set_simd4x2_offset(fs_inst *inst,
543 struct brw_reg dst,
544 struct brw_reg offset);
545 void generate_discard_jump(fs_inst *inst);
546
547 void generate_pack_half_2x16_split(fs_inst *inst,
548 struct brw_reg dst,
549 struct brw_reg x,
550 struct brw_reg y);
551 void generate_unpack_half_2x16_split(fs_inst *inst,
552 struct brw_reg dst,
553 struct brw_reg src);
554
555 void generate_shader_time_add(fs_inst *inst,
556 struct brw_reg payload,
557 struct brw_reg offset,
558 struct brw_reg value);
559
560 void generate_untyped_atomic(fs_inst *inst,
561 struct brw_reg dst,
562 struct brw_reg atomic_op,
563 struct brw_reg surf_index);
564
565 void generate_untyped_surface_read(fs_inst *inst,
566 struct brw_reg dst,
567 struct brw_reg surf_index);
568
569 void mark_surface_used(unsigned surf_index);
570
571 void patch_discard_jumps_to_fb_writes();
572
573 struct brw_context *brw;
574 struct gl_context *ctx;
575
576 struct brw_compile *p;
577 struct brw_wm_compile *c;
578
579 struct gl_shader_program *prog;
580 struct gl_shader *shader;
581 const struct gl_fragment_program *fp;
582
583 unsigned dispatch_width; /**< 8 or 16 */
584
585 exec_list discard_halt_patches;
586 bool dual_source_output;
587 void *mem_ctx;
588 };
589
590 bool brw_do_channel_expressions(struct exec_list *instructions);
591 bool brw_do_vector_splitting(struct exec_list *instructions);
592 bool brw_fs_precompile(struct gl_context *ctx, struct gl_shader_program *prog);