i965/gen7: Implement code generation for untyped atomic instructions.
[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 * For virtual registers, this is a hardware register offset from
92 * the start of the register block (for example, a constant index
93 * in an array access).
94 */
95 int reg_offset;
96 /** Register type. BRW_REGISTER_TYPE_* */
97 int type;
98 bool negate;
99 bool abs;
100 bool sechalf;
101 struct brw_reg fixed_hw_reg;
102 int smear; /* -1, or a channel of the reg to smear to all channels. */
103
104 /** Value for file == IMM */
105 union {
106 int32_t i;
107 uint32_t u;
108 float f;
109 } imm;
110
111 fs_reg *reladdr;
112 };
113
114 static const fs_reg reg_undef;
115 static const fs_reg reg_null_f(retype(brw_null_reg(), BRW_REGISTER_TYPE_F));
116 static const fs_reg reg_null_d(retype(brw_null_reg(), BRW_REGISTER_TYPE_D));
117 static const fs_reg reg_null_ud(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
118
119 class ip_record : public exec_node {
120 public:
121 DECLARE_RALLOC_CXX_OPERATORS(ip_record)
122
123 ip_record(int ip)
124 {
125 this->ip = ip;
126 }
127
128 int ip;
129 };
130
131 class fs_inst : public backend_instruction {
132 public:
133 DECLARE_RALLOC_CXX_OPERATORS(fs_inst)
134
135 void init();
136
137 fs_inst();
138 fs_inst(enum opcode opcode);
139 fs_inst(enum opcode opcode, fs_reg dst);
140 fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0);
141 fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1);
142 fs_inst(enum opcode opcode, fs_reg dst,
143 fs_reg src0, fs_reg src1,fs_reg src2);
144
145 bool equals(fs_inst *inst);
146 bool overwrites_reg(const fs_reg &reg);
147 bool is_send_from_grf();
148 bool is_partial_write();
149 int regs_read(fs_visitor *v, int arg);
150
151 fs_reg dst;
152 fs_reg src[3];
153 bool saturate;
154 int conditional_mod; /**< BRW_CONDITIONAL_* */
155
156 /* Chooses which flag subregister (f0.0 or f0.1) is used for conditional
157 * mod and predication.
158 */
159 uint8_t flag_subreg;
160
161 int mlen; /**< SEND message length */
162 int regs_written; /**< Number of vgrfs written by a SEND message, or 1 */
163 int base_mrf; /**< First MRF in the SEND message, if mlen is nonzero. */
164 uint32_t texture_offset; /**< Texture offset bitfield */
165 int sampler;
166 int target; /**< MRT target. */
167 bool eot;
168 bool header_present;
169 bool shadow_compare;
170 bool force_uncompressed;
171 bool force_sechalf;
172 bool force_writemask_all;
173 uint32_t offset; /* spill/unspill offset */
174
175 /** @{
176 * Annotation for the generated IR. One of the two can be set.
177 */
178 const void *ir;
179 const char *annotation;
180 /** @} */
181 };
182
183 /**
184 * The fragment shader front-end.
185 *
186 * Translates either GLSL IR or Mesa IR (for ARB_fragment_program) into FS IR.
187 */
188 class fs_visitor : public backend_visitor
189 {
190 public:
191
192 fs_visitor(struct brw_context *brw,
193 struct brw_wm_compile *c,
194 struct gl_shader_program *shader_prog,
195 struct gl_fragment_program *fp,
196 unsigned dispatch_width);
197 ~fs_visitor();
198
199 fs_reg *variable_storage(ir_variable *var);
200 int virtual_grf_alloc(int size);
201 void import_uniforms(fs_visitor *v);
202
203 void visit(ir_variable *ir);
204 void visit(ir_assignment *ir);
205 void visit(ir_dereference_variable *ir);
206 void visit(ir_dereference_record *ir);
207 void visit(ir_dereference_array *ir);
208 void visit(ir_expression *ir);
209 void visit(ir_texture *ir);
210 void visit(ir_if *ir);
211 void visit(ir_constant *ir);
212 void visit(ir_swizzle *ir);
213 void visit(ir_return *ir);
214 void visit(ir_loop *ir);
215 void visit(ir_loop_jump *ir);
216 void visit(ir_discard *ir);
217 void visit(ir_call *ir);
218 void visit(ir_function *ir);
219 void visit(ir_function_signature *ir);
220 void visit(ir_emit_vertex *);
221 void visit(ir_end_primitive *);
222
223 uint32_t gather_channel(ir_texture *ir, int sampler);
224 void swizzle_result(ir_texture *ir, fs_reg orig_val, int sampler);
225
226 bool can_do_source_mods(fs_inst *inst);
227
228 fs_inst *emit(fs_inst inst);
229 fs_inst *emit(fs_inst *inst);
230 void emit(exec_list list);
231
232 fs_inst *emit(enum opcode opcode);
233 fs_inst *emit(enum opcode opcode, fs_reg dst);
234 fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0);
235 fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1);
236 fs_inst *emit(enum opcode opcode, fs_reg dst,
237 fs_reg src0, fs_reg src1, fs_reg src2);
238
239 fs_inst *MOV(fs_reg dst, fs_reg src);
240 fs_inst *NOT(fs_reg dst, fs_reg src);
241 fs_inst *RNDD(fs_reg dst, fs_reg src);
242 fs_inst *RNDE(fs_reg dst, fs_reg src);
243 fs_inst *RNDZ(fs_reg dst, fs_reg src);
244 fs_inst *FRC(fs_reg dst, fs_reg src);
245 fs_inst *ADD(fs_reg dst, fs_reg src0, fs_reg src1);
246 fs_inst *MUL(fs_reg dst, fs_reg src0, fs_reg src1);
247 fs_inst *MACH(fs_reg dst, fs_reg src0, fs_reg src1);
248 fs_inst *MAC(fs_reg dst, fs_reg src0, fs_reg src1);
249 fs_inst *SHL(fs_reg dst, fs_reg src0, fs_reg src1);
250 fs_inst *SHR(fs_reg dst, fs_reg src0, fs_reg src1);
251 fs_inst *ASR(fs_reg dst, fs_reg src0, fs_reg src1);
252 fs_inst *AND(fs_reg dst, fs_reg src0, fs_reg src1);
253 fs_inst *OR(fs_reg dst, fs_reg src0, fs_reg src1);
254 fs_inst *XOR(fs_reg dst, fs_reg src0, fs_reg src1);
255 fs_inst *IF(uint32_t predicate);
256 fs_inst *IF(fs_reg src0, fs_reg src1, uint32_t condition);
257 fs_inst *CMP(fs_reg dst, fs_reg src0, fs_reg src1,
258 uint32_t condition);
259 fs_inst *LRP(fs_reg dst, fs_reg a, fs_reg y, fs_reg x);
260 fs_inst *DEP_RESOLVE_MOV(int grf);
261 fs_inst *BFREV(fs_reg dst, fs_reg value);
262 fs_inst *BFE(fs_reg dst, fs_reg bits, fs_reg offset, fs_reg value);
263 fs_inst *BFI1(fs_reg dst, fs_reg bits, fs_reg offset);
264 fs_inst *BFI2(fs_reg dst, fs_reg bfi1_dst, fs_reg insert, fs_reg base);
265 fs_inst *FBH(fs_reg dst, fs_reg value);
266 fs_inst *FBL(fs_reg dst, fs_reg value);
267 fs_inst *CBIT(fs_reg dst, fs_reg value);
268 fs_inst *MAD(fs_reg dst, fs_reg c, fs_reg b, fs_reg a);
269 fs_inst *ADDC(fs_reg dst, fs_reg src0, fs_reg src1);
270 fs_inst *SUBB(fs_reg dst, fs_reg src0, fs_reg src1);
271
272 int type_size(const struct glsl_type *type);
273 fs_inst *get_instruction_generating_reg(fs_inst *start,
274 fs_inst *end,
275 fs_reg reg);
276
277 exec_list VARYING_PULL_CONSTANT_LOAD(fs_reg dst, fs_reg surf_index,
278 fs_reg varying_offset,
279 uint32_t const_offset);
280
281 bool run();
282 void assign_binding_table_offsets();
283 void setup_payload_gen4();
284 void setup_payload_gen6();
285 void assign_curb_setup();
286 void calculate_urb_setup();
287 void assign_urb_setup();
288 bool assign_regs();
289 void assign_regs_trivial();
290 void setup_payload_interference(struct ra_graph *g, int payload_reg_count,
291 int first_payload_node);
292 void setup_mrf_hack_interference(struct ra_graph *g,
293 int first_mrf_hack_node);
294 int choose_spill_reg(struct ra_graph *g);
295 void spill_reg(int spill_reg);
296 void split_virtual_grfs();
297 void compact_virtual_grfs();
298 void move_uniform_array_access_to_pull_constants();
299 void setup_pull_constants();
300 void invalidate_live_intervals();
301 void calculate_live_intervals();
302 bool opt_algebraic();
303 bool opt_cse();
304 bool opt_cse_local(bblock_t *block, exec_list *aeb);
305 bool opt_copy_propagate();
306 bool try_copy_propagate(fs_inst *inst, int arg, acp_entry *entry);
307 bool try_constant_propagate(fs_inst *inst, acp_entry *entry);
308 bool opt_copy_propagate_local(void *mem_ctx, bblock_t *block,
309 exec_list *acp);
310 bool register_coalesce();
311 bool register_coalesce_2();
312 bool compute_to_mrf();
313 bool dead_code_eliminate();
314 bool dead_code_eliminate_local();
315 bool remove_dead_constants();
316 bool remove_duplicate_mrf_writes();
317 bool virtual_grf_interferes(int a, int b);
318 void schedule_instructions(bool post_reg_alloc);
319 void insert_gen4_send_dependency_workarounds();
320 void insert_gen4_pre_send_dependency_workarounds(fs_inst *inst);
321 void insert_gen4_post_send_dependency_workarounds(fs_inst *inst);
322 void fail(const char *msg, ...);
323 void lower_uniform_pull_constant_loads();
324
325 void push_force_uncompressed();
326 void pop_force_uncompressed();
327 void push_force_sechalf();
328 void pop_force_sechalf();
329
330 void emit_dummy_fs();
331 fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
332 fs_inst *emit_linterp(const fs_reg &attr, const fs_reg &interp,
333 glsl_interp_qualifier interpolation_mode,
334 bool is_centroid);
335 fs_reg *emit_frontfacing_interpolation(ir_variable *ir);
336 fs_reg *emit_general_interpolation(ir_variable *ir);
337 void emit_interpolation_setup_gen4();
338 void emit_interpolation_setup_gen6();
339 fs_reg rescale_texcoord(ir_texture *ir, fs_reg coordinate,
340 bool is_rect, int sampler, int texunit);
341 fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate,
342 fs_reg shadow_comp, fs_reg lod, fs_reg lod2);
343 fs_inst *emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate,
344 fs_reg shadow_comp, fs_reg lod, fs_reg lod2,
345 fs_reg sample_index);
346 fs_inst *emit_texture_gen7(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_reg fix_math_operand(fs_reg src);
350 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0);
351 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0, fs_reg src1);
352 void emit_lrp(fs_reg dst, fs_reg x, fs_reg y, fs_reg a);
353 void emit_minmax(uint32_t conditionalmod, fs_reg dst,
354 fs_reg src0, fs_reg src1);
355 bool try_emit_saturate(ir_expression *ir);
356 bool try_emit_mad(ir_expression *ir, int mul_arg);
357 void try_replace_with_sel();
358 void emit_bool_to_cond_code(ir_rvalue *condition);
359 void emit_if_gen6(ir_if *ir);
360 void emit_unspill(fs_inst *inst, fs_reg reg, uint32_t spill_offset,
361 int count);
362
363 void emit_fragment_program_code();
364 void setup_fp_regs();
365 fs_reg get_fp_src_reg(const prog_src_register *src);
366 fs_reg get_fp_dst_reg(const prog_dst_register *dst);
367 void emit_fp_alu1(enum opcode opcode,
368 const struct prog_instruction *fpi,
369 fs_reg dst, fs_reg src);
370 void emit_fp_alu2(enum opcode opcode,
371 const struct prog_instruction *fpi,
372 fs_reg dst, fs_reg src0, fs_reg src1);
373 void emit_fp_scalar_write(const struct prog_instruction *fpi,
374 fs_reg dst, fs_reg src);
375 void emit_fp_scalar_math(enum opcode opcode,
376 const struct prog_instruction *fpi,
377 fs_reg dst, fs_reg src);
378
379 void emit_fp_minmax(const struct prog_instruction *fpi,
380 fs_reg dst, fs_reg src0, fs_reg src1);
381
382 void emit_fp_sop(uint32_t conditional_mod,
383 const struct prog_instruction *fpi,
384 fs_reg dst, fs_reg src0, fs_reg src1, fs_reg one);
385
386 void emit_color_write(int target, int index, int first_color_mrf);
387 void emit_fb_writes();
388
389 void emit_shader_time_begin();
390 void emit_shader_time_end();
391 void emit_shader_time_write(enum shader_time_shader_type type,
392 fs_reg value);
393
394 bool try_rewrite_rhs_to_dst(ir_assignment *ir,
395 fs_reg dst,
396 fs_reg src,
397 fs_inst *pre_rhs_inst,
398 fs_inst *last_rhs_inst);
399 void emit_assignment_writes(fs_reg &l, fs_reg &r,
400 const glsl_type *type, bool predicated);
401 void resolve_ud_negate(fs_reg *reg);
402 void resolve_bool_comparison(ir_rvalue *rvalue, fs_reg *reg);
403
404 fs_reg get_timestamp();
405
406 struct brw_reg interp_reg(int location, int channel);
407 void setup_uniform_values(ir_variable *ir);
408 void setup_builtin_uniform_values(ir_variable *ir);
409 int implied_mrf_writes(fs_inst *inst);
410
411 void dump_instruction(backend_instruction *inst);
412
413 struct gl_fragment_program *fp;
414 struct brw_wm_compile *c;
415 unsigned int sanity_param_count;
416
417 int param_size[MAX_UNIFORMS * 4];
418
419 int *virtual_grf_sizes;
420 int virtual_grf_count;
421 int virtual_grf_array_size;
422 int *virtual_grf_start;
423 int *virtual_grf_end;
424 brw::fs_live_variables *live_intervals;
425
426 /* This is the map from UNIFORM hw_reg + reg_offset as generated by
427 * the visitor to the packed uniform number after
428 * remove_dead_constants() that represents the actual uploaded
429 * uniform index.
430 */
431 int *params_remap;
432 int nr_params_remap;
433
434 struct hash_table *variable_ht;
435 fs_reg frag_depth;
436 fs_reg outputs[BRW_MAX_DRAW_BUFFERS];
437 unsigned output_components[BRW_MAX_DRAW_BUFFERS];
438 fs_reg dual_src_output;
439 int first_non_payload_grf;
440 /** Either BRW_MAX_GRF or GEN7_MRF_HACK_START */
441 int max_grf;
442
443 fs_reg *fp_temp_regs;
444 fs_reg *fp_input_regs;
445
446 /** @{ debug annotation info */
447 const char *current_annotation;
448 const void *base_ir;
449 /** @} */
450
451 bool failed;
452 char *fail_msg;
453
454 /* Result of last visit() method. */
455 fs_reg result;
456
457 fs_reg pixel_x;
458 fs_reg pixel_y;
459 fs_reg wpos_w;
460 fs_reg pixel_w;
461 fs_reg delta_x[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
462 fs_reg delta_y[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
463 fs_reg shader_start_time;
464
465 int grf_used;
466
467 const unsigned dispatch_width; /**< 8 or 16 */
468
469 int force_uncompressed_stack;
470 int force_sechalf_stack;
471 };
472
473 /**
474 * The fragment shader code generator.
475 *
476 * Translates FS IR to actual i965 assembly code.
477 */
478 class fs_generator
479 {
480 public:
481 fs_generator(struct brw_context *brw,
482 struct brw_wm_compile *c,
483 struct gl_shader_program *prog,
484 struct gl_fragment_program *fp,
485 bool dual_source_output);
486 ~fs_generator();
487
488 const unsigned *generate_assembly(exec_list *simd8_instructions,
489 exec_list *simd16_instructions,
490 unsigned *assembly_size);
491
492 private:
493 void generate_code(exec_list *instructions);
494 void generate_fb_write(fs_inst *inst);
495 void generate_pixel_xy(struct brw_reg dst, bool is_x);
496 void generate_linterp(fs_inst *inst, struct brw_reg dst,
497 struct brw_reg *src);
498 void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
499 void generate_math1_gen7(fs_inst *inst,
500 struct brw_reg dst,
501 struct brw_reg src);
502 void generate_math2_gen7(fs_inst *inst,
503 struct brw_reg dst,
504 struct brw_reg src0,
505 struct brw_reg src1);
506 void generate_math1_gen6(fs_inst *inst,
507 struct brw_reg dst,
508 struct brw_reg src);
509 void generate_math2_gen6(fs_inst *inst,
510 struct brw_reg dst,
511 struct brw_reg src0,
512 struct brw_reg src1);
513 void generate_math_gen4(fs_inst *inst,
514 struct brw_reg dst,
515 struct brw_reg src);
516 void generate_math_g45(fs_inst *inst,
517 struct brw_reg dst,
518 struct brw_reg src);
519 void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
520 void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
521 bool negate_value);
522 void generate_spill(fs_inst *inst, struct brw_reg src);
523 void generate_unspill(fs_inst *inst, struct brw_reg dst);
524 void generate_uniform_pull_constant_load(fs_inst *inst, struct brw_reg dst,
525 struct brw_reg index,
526 struct brw_reg offset);
527 void generate_uniform_pull_constant_load_gen7(fs_inst *inst,
528 struct brw_reg dst,
529 struct brw_reg surf_index,
530 struct brw_reg offset);
531 void generate_varying_pull_constant_load(fs_inst *inst, struct brw_reg dst,
532 struct brw_reg index,
533 struct brw_reg offset);
534 void generate_varying_pull_constant_load_gen7(fs_inst *inst,
535 struct brw_reg dst,
536 struct brw_reg index,
537 struct brw_reg offset);
538 void generate_mov_dispatch_to_flags(fs_inst *inst);
539 void generate_set_simd4x2_offset(fs_inst *inst,
540 struct brw_reg dst,
541 struct brw_reg offset);
542 void generate_discard_jump(fs_inst *inst);
543
544 void generate_pack_half_2x16_split(fs_inst *inst,
545 struct brw_reg dst,
546 struct brw_reg x,
547 struct brw_reg y);
548 void generate_unpack_half_2x16_split(fs_inst *inst,
549 struct brw_reg dst,
550 struct brw_reg src);
551
552 void generate_shader_time_add(fs_inst *inst,
553 struct brw_reg payload,
554 struct brw_reg offset,
555 struct brw_reg value);
556
557 void generate_untyped_atomic(fs_inst *inst,
558 struct brw_reg dst,
559 struct brw_reg atomic_op,
560 struct brw_reg surf_index);
561
562 void mark_surface_used(unsigned surf_index);
563
564 void patch_discard_jumps_to_fb_writes();
565
566 struct brw_context *brw;
567 struct gl_context *ctx;
568
569 struct brw_compile *p;
570 struct brw_wm_compile *c;
571
572 struct gl_shader_program *prog;
573 struct gl_shader *shader;
574 const struct gl_fragment_program *fp;
575
576 unsigned dispatch_width; /**< 8 or 16 */
577
578 exec_list discard_halt_patches;
579 bool dual_source_output;
580 void *mem_ctx;
581 };
582
583 bool brw_do_channel_expressions(struct exec_list *instructions);
584 bool brw_do_vector_splitting(struct exec_list *instructions);
585 bool brw_fs_precompile(struct gl_context *ctx, struct gl_shader_program *prog);