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