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