i965/urb: fixes division by zero
[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 #include "brw_ir_fs.h"
32 #include "brw_fs_builder.h"
33 #include "compiler/nir/nir.h"
34
35 struct bblock_t;
36 namespace {
37 struct acp_entry;
38 }
39
40 namespace brw {
41 class fs_live_variables;
42 }
43
44 struct brw_gs_compile;
45
46 static inline fs_reg
47 offset(fs_reg reg, const brw::fs_builder& bld, unsigned delta)
48 {
49 switch (reg.file) {
50 case BAD_FILE:
51 break;
52 case ARF:
53 case FIXED_GRF:
54 case MRF:
55 case VGRF:
56 case ATTR:
57 case UNIFORM:
58 return byte_offset(reg,
59 delta * reg.component_size(bld.dispatch_width()));
60 case IMM:
61 assert(delta == 0);
62 }
63 return reg;
64 }
65
66 /**
67 * The fragment shader front-end.
68 *
69 * Translates either GLSL IR or Mesa IR (for ARB_fragment_program) into FS IR.
70 */
71 class fs_visitor : public backend_shader
72 {
73 public:
74 fs_visitor(const struct brw_compiler *compiler, void *log_data,
75 void *mem_ctx,
76 const void *key,
77 struct brw_stage_prog_data *prog_data,
78 struct gl_program *prog,
79 const nir_shader *shader,
80 unsigned dispatch_width,
81 int shader_time_index,
82 const struct brw_vue_map *input_vue_map = NULL);
83 fs_visitor(const struct brw_compiler *compiler, void *log_data,
84 void *mem_ctx,
85 struct brw_gs_compile *gs_compile,
86 struct brw_gs_prog_data *prog_data,
87 const nir_shader *shader,
88 int shader_time_index);
89 void init();
90 ~fs_visitor();
91
92 fs_reg vgrf(const glsl_type *const type);
93 void import_uniforms(fs_visitor *v);
94 void setup_uniform_clipplane_values(gl_clip_plane *clip_planes);
95 void compute_clip_distance(gl_clip_plane *clip_planes);
96
97 fs_inst *get_instruction_generating_reg(fs_inst *start,
98 fs_inst *end,
99 const fs_reg &reg);
100
101 void VARYING_PULL_CONSTANT_LOAD(const brw::fs_builder &bld,
102 const fs_reg &dst,
103 const fs_reg &surf_index,
104 const fs_reg &varying_offset,
105 uint32_t const_offset);
106 void DEP_RESOLVE_MOV(const brw::fs_builder &bld, int grf);
107
108 bool run_fs(bool allow_spilling, bool do_rep_send);
109 bool run_vs(gl_clip_plane *clip_planes);
110 bool run_tcs_single_patch();
111 bool run_tes();
112 bool run_gs();
113 bool run_cs();
114 void optimize();
115 void allocate_registers(bool allow_spilling);
116 void setup_fs_payload_gen4();
117 void setup_fs_payload_gen6();
118 void setup_vs_payload();
119 void setup_gs_payload();
120 void setup_cs_payload();
121 void fixup_3src_null_dest();
122 void assign_curb_setup();
123 void calculate_urb_setup();
124 void assign_urb_setup();
125 void convert_attr_sources_to_hw_regs(fs_inst *inst);
126 void assign_vs_urb_setup();
127 void assign_tcs_single_patch_urb_setup();
128 void assign_tes_urb_setup();
129 void assign_gs_urb_setup();
130 bool assign_regs(bool allow_spilling, bool spill_all);
131 void assign_regs_trivial();
132 void calculate_payload_ranges(int payload_node_count,
133 int *payload_last_use_ip);
134 void setup_payload_interference(struct ra_graph *g, int payload_reg_count,
135 int first_payload_node);
136 int choose_spill_reg(struct ra_graph *g);
137 void spill_reg(int spill_reg);
138 void split_virtual_grfs();
139 bool compact_virtual_grfs();
140 void assign_constant_locations();
141 void lower_constant_loads();
142 void invalidate_live_intervals();
143 void calculate_live_intervals();
144 void calculate_register_pressure();
145 void validate();
146 bool opt_algebraic();
147 bool opt_redundant_discard_jumps();
148 bool opt_cse();
149 bool opt_cse_local(bblock_t *block);
150 bool opt_copy_propagate();
151 bool try_copy_propagate(fs_inst *inst, int arg, acp_entry *entry);
152 bool try_constant_propagate(fs_inst *inst, acp_entry *entry);
153 bool opt_copy_propagate_local(void *mem_ctx, bblock_t *block,
154 exec_list *acp);
155 bool opt_drop_redundant_mov_to_flags();
156 bool opt_register_renaming();
157 bool register_coalesce();
158 bool compute_to_mrf();
159 bool eliminate_find_live_channel();
160 bool dead_code_eliminate();
161 bool remove_duplicate_mrf_writes();
162
163 bool opt_sampler_eot();
164 bool virtual_grf_interferes(int a, int b);
165 void schedule_instructions(instruction_scheduler_mode mode);
166 void insert_gen4_send_dependency_workarounds();
167 void insert_gen4_pre_send_dependency_workarounds(bblock_t *block,
168 fs_inst *inst);
169 void insert_gen4_post_send_dependency_workarounds(bblock_t *block,
170 fs_inst *inst);
171 void vfail(const char *msg, va_list args);
172 void fail(const char *msg, ...);
173 void no16(const char *msg);
174 void lower_uniform_pull_constant_loads();
175 bool lower_load_payload();
176 bool lower_pack();
177 bool lower_d2x();
178 bool lower_logical_sends();
179 bool lower_integer_multiplication();
180 bool lower_minmax();
181 bool lower_simd_width();
182 bool opt_combine_constants();
183
184 void emit_dummy_fs();
185 void emit_repclear_shader();
186 fs_reg *emit_fragcoord_interpolation(bool pixel_center_integer,
187 bool origin_upper_left);
188 fs_inst *emit_linterp(const fs_reg &attr, const fs_reg &interp,
189 glsl_interp_qualifier interpolation_mode,
190 bool is_centroid, bool is_sample);
191 fs_reg *emit_frontfacing_interpolation();
192 fs_reg *emit_samplepos_setup();
193 fs_reg *emit_sampleid_setup();
194 fs_reg *emit_samplemaskin_setup();
195 void emit_general_interpolation(fs_reg *attr, const char *name,
196 const glsl_type *type,
197 glsl_interp_qualifier interpolation_mode,
198 int *location, bool mod_centroid,
199 bool mod_sample);
200 fs_reg *emit_vs_system_value(int location);
201 void emit_interpolation_setup_gen4();
202 void emit_interpolation_setup_gen6();
203 void compute_sample_position(fs_reg dst, fs_reg int_sample_pos);
204 fs_reg emit_mcs_fetch(const fs_reg &coordinate, unsigned components,
205 const fs_reg &sampler);
206 void emit_gen6_gather_wa(uint8_t wa, fs_reg dst);
207 fs_reg resolve_source_modifiers(const fs_reg &src);
208 void emit_discard_jump();
209 bool opt_peephole_sel();
210 bool opt_peephole_predicated_break();
211 bool opt_saturate_propagation();
212 bool opt_cmod_propagation();
213 bool opt_zero_samples();
214 void emit_unspill(bblock_t *block, fs_inst *inst, fs_reg reg,
215 uint32_t spill_offset, int count);
216 void emit_spill(bblock_t *block, fs_inst *inst, fs_reg reg,
217 uint32_t spill_offset, int count);
218
219 void emit_nir_code();
220 void nir_setup_inputs();
221 void nir_setup_single_output_varying(fs_reg *reg, const glsl_type *type,
222 unsigned *location);
223 void nir_setup_outputs();
224 void nir_setup_uniforms();
225 void nir_emit_system_values();
226 void nir_emit_impl(nir_function_impl *impl);
227 void nir_emit_cf_list(exec_list *list);
228 void nir_emit_if(nir_if *if_stmt);
229 void nir_emit_loop(nir_loop *loop);
230 void nir_emit_block(nir_block *block);
231 void nir_emit_instr(nir_instr *instr);
232 void nir_emit_alu(const brw::fs_builder &bld, nir_alu_instr *instr);
233 void nir_emit_load_const(const brw::fs_builder &bld,
234 nir_load_const_instr *instr);
235 void nir_emit_undef(const brw::fs_builder &bld,
236 nir_ssa_undef_instr *instr);
237 void nir_emit_vs_intrinsic(const brw::fs_builder &bld,
238 nir_intrinsic_instr *instr);
239 void nir_emit_tcs_intrinsic(const brw::fs_builder &bld,
240 nir_intrinsic_instr *instr);
241 void nir_emit_gs_intrinsic(const brw::fs_builder &bld,
242 nir_intrinsic_instr *instr);
243 void nir_emit_fs_intrinsic(const brw::fs_builder &bld,
244 nir_intrinsic_instr *instr);
245 void nir_emit_cs_intrinsic(const brw::fs_builder &bld,
246 nir_intrinsic_instr *instr);
247 void nir_emit_intrinsic(const brw::fs_builder &bld,
248 nir_intrinsic_instr *instr);
249 void nir_emit_tes_intrinsic(const brw::fs_builder &bld,
250 nir_intrinsic_instr *instr);
251 void nir_emit_ssbo_atomic(const brw::fs_builder &bld,
252 int op, nir_intrinsic_instr *instr);
253 void nir_emit_shared_atomic(const brw::fs_builder &bld,
254 int op, nir_intrinsic_instr *instr);
255 void nir_emit_texture(const brw::fs_builder &bld,
256 nir_tex_instr *instr);
257 void nir_emit_jump(const brw::fs_builder &bld,
258 nir_jump_instr *instr);
259 fs_reg get_nir_src(nir_src src);
260 fs_reg get_nir_dest(nir_dest dest);
261 fs_reg get_nir_image_deref(const nir_deref_var *deref);
262 fs_reg get_indirect_offset(nir_intrinsic_instr *instr);
263 void emit_percomp(const brw::fs_builder &bld, const fs_inst &inst,
264 unsigned wr_mask);
265
266 bool optimize_extract_to_float(nir_alu_instr *instr,
267 const fs_reg &result);
268 bool optimize_frontfacing_ternary(nir_alu_instr *instr,
269 const fs_reg &result);
270
271 void emit_alpha_test();
272 fs_inst *emit_single_fb_write(const brw::fs_builder &bld,
273 fs_reg color1, fs_reg color2,
274 fs_reg src0_alpha, unsigned components);
275 void emit_fb_writes();
276 void emit_urb_writes(const fs_reg &gs_vertex_count = fs_reg());
277 void set_gs_stream_control_data_bits(const fs_reg &vertex_count,
278 unsigned stream_id);
279 void emit_gs_control_data_bits(const fs_reg &vertex_count);
280 void emit_gs_end_primitive(const nir_src &vertex_count_nir_src);
281 void emit_gs_vertex(const nir_src &vertex_count_nir_src,
282 unsigned stream_id);
283 void emit_gs_thread_end();
284 void emit_gs_input_load(const fs_reg &dst, const nir_src &vertex_src,
285 unsigned base_offset, const nir_src &offset_src,
286 unsigned num_components);
287 void emit_cs_terminate();
288 fs_reg *emit_cs_local_invocation_id_setup();
289 fs_reg *emit_cs_work_group_id_setup();
290
291 void emit_barrier();
292
293 void emit_shader_time_begin();
294 void emit_shader_time_end();
295 void SHADER_TIME_ADD(const brw::fs_builder &bld,
296 int shader_time_subindex,
297 fs_reg value);
298
299 fs_reg get_timestamp(const brw::fs_builder &bld);
300
301 struct brw_reg interp_reg(int location, int channel);
302
303 int implied_mrf_writes(fs_inst *inst);
304
305 virtual void dump_instructions();
306 virtual void dump_instructions(const char *name);
307 void dump_instruction(backend_instruction *inst);
308 void dump_instruction(backend_instruction *inst, FILE *file);
309
310 const void *const key;
311 const struct brw_sampler_prog_key_data *key_tex;
312
313 struct brw_gs_compile *gs_compile;
314
315 struct brw_stage_prog_data *prog_data;
316 struct gl_program *prog;
317
318 const struct brw_vue_map *input_vue_map;
319
320 int *virtual_grf_start;
321 int *virtual_grf_end;
322 brw::fs_live_variables *live_intervals;
323
324 int *regs_live_at_ip;
325
326 /** Number of uniform variable components visited. */
327 unsigned uniforms;
328
329 /** Byte-offset for the next available spot in the scratch space buffer. */
330 unsigned last_scratch;
331
332 /**
333 * Array mapping UNIFORM register numbers to the pull parameter index,
334 * or -1 if this uniform register isn't being uploaded as a pull constant.
335 */
336 int *pull_constant_loc;
337
338 /**
339 * Array mapping UNIFORM register numbers to the push parameter index,
340 * or -1 if this uniform register isn't being uploaded as a push constant.
341 */
342 int *push_constant_loc;
343
344 fs_reg frag_depth;
345 fs_reg frag_stencil;
346 fs_reg sample_mask;
347 fs_reg outputs[VARYING_SLOT_MAX];
348 unsigned output_components[VARYING_SLOT_MAX];
349 fs_reg dual_src_output;
350 bool do_dual_src;
351 int first_non_payload_grf;
352 /** Either BRW_MAX_GRF or GEN7_MRF_HACK_START */
353 unsigned max_grf;
354
355 fs_reg *nir_locals;
356 fs_reg *nir_ssa_values;
357 fs_reg nir_inputs;
358 fs_reg nir_outputs;
359 fs_reg *nir_system_values;
360
361 bool failed;
362 char *fail_msg;
363 bool simd16_unsupported;
364 char *no16_msg;
365
366 /** Register numbers for thread payload fields. */
367 struct thread_payload {
368 uint8_t source_depth_reg;
369 uint8_t source_w_reg;
370 uint8_t aa_dest_stencil_reg;
371 uint8_t dest_depth_reg;
372 uint8_t sample_pos_reg;
373 uint8_t sample_mask_in_reg;
374 uint8_t barycentric_coord_reg[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
375 uint8_t local_invocation_id_reg;
376
377 /** The number of thread payload registers the hardware will supply. */
378 uint8_t num_regs;
379 } payload;
380
381 bool source_depth_to_render_target;
382 bool runtime_check_aads_emit;
383
384 fs_reg pixel_x;
385 fs_reg pixel_y;
386 fs_reg wpos_w;
387 fs_reg pixel_w;
388 fs_reg delta_xy[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
389 fs_reg shader_start_time;
390 fs_reg userplane[MAX_CLIP_PLANES];
391 fs_reg final_gs_vertex_count;
392 fs_reg control_data_bits;
393 fs_reg invocation_id;
394
395 unsigned grf_used;
396 bool spilled_any_registers;
397
398 const unsigned dispatch_width; /**< 8 or 16 */
399 unsigned min_dispatch_width;
400
401 int shader_time_index;
402
403 unsigned promoted_constants;
404 brw::fs_builder bld;
405 };
406
407 /**
408 * The fragment shader code generator.
409 *
410 * Translates FS IR to actual i965 assembly code.
411 */
412 class fs_generator
413 {
414 public:
415 fs_generator(const struct brw_compiler *compiler, void *log_data,
416 void *mem_ctx,
417 const void *key,
418 struct brw_stage_prog_data *prog_data,
419 unsigned promoted_constants,
420 bool runtime_check_aads_emit,
421 gl_shader_stage stage);
422 ~fs_generator();
423
424 void enable_debug(const char *shader_name);
425 int generate_code(const cfg_t *cfg, int dispatch_width);
426 const unsigned *get_assembly(unsigned int *assembly_size);
427
428 private:
429 void fire_fb_write(fs_inst *inst,
430 struct brw_reg payload,
431 struct brw_reg implied_header,
432 GLuint nr);
433 void generate_fb_write(fs_inst *inst, struct brw_reg payload);
434 void generate_urb_read(fs_inst *inst, struct brw_reg dst, struct brw_reg payload);
435 void generate_urb_write(fs_inst *inst, struct brw_reg payload);
436 void generate_cs_terminate(fs_inst *inst, struct brw_reg payload);
437 void generate_stencil_ref_packing(fs_inst *inst, struct brw_reg dst,
438 struct brw_reg src);
439 void generate_barrier(fs_inst *inst, struct brw_reg src);
440 void generate_linterp(fs_inst *inst, struct brw_reg dst,
441 struct brw_reg *src);
442 void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
443 struct brw_reg surface_index,
444 struct brw_reg sampler_index);
445 void generate_get_buffer_size(fs_inst *inst, struct brw_reg dst,
446 struct brw_reg src,
447 struct brw_reg surf_index);
448 void generate_math_gen6(fs_inst *inst,
449 struct brw_reg dst,
450 struct brw_reg src0,
451 struct brw_reg src1);
452 void generate_math_gen4(fs_inst *inst,
453 struct brw_reg dst,
454 struct brw_reg src);
455 void generate_math_g45(fs_inst *inst,
456 struct brw_reg dst,
457 struct brw_reg src);
458 void generate_ddx(enum opcode op, struct brw_reg dst, struct brw_reg src);
459 void generate_ddy(enum opcode op, struct brw_reg dst, struct brw_reg src,
460 bool negate_value);
461 void generate_scratch_write(fs_inst *inst, struct brw_reg src);
462 void generate_scratch_read(fs_inst *inst, struct brw_reg dst);
463 void generate_scratch_read_gen7(fs_inst *inst, struct brw_reg dst);
464 void generate_uniform_pull_constant_load(fs_inst *inst, struct brw_reg dst,
465 struct brw_reg index,
466 struct brw_reg offset);
467 void generate_uniform_pull_constant_load_gen7(fs_inst *inst,
468 struct brw_reg dst,
469 struct brw_reg surf_index,
470 struct brw_reg offset);
471 void generate_varying_pull_constant_load(fs_inst *inst, struct brw_reg dst,
472 struct brw_reg index,
473 struct brw_reg offset);
474 void generate_varying_pull_constant_load_gen7(fs_inst *inst,
475 struct brw_reg dst,
476 struct brw_reg index,
477 struct brw_reg offset);
478 void generate_mov_dispatch_to_flags(fs_inst *inst);
479
480 void generate_pixel_interpolator_query(fs_inst *inst,
481 struct brw_reg dst,
482 struct brw_reg src,
483 struct brw_reg msg_data,
484 unsigned msg_type);
485
486 void generate_set_sample_id(fs_inst *inst,
487 struct brw_reg dst,
488 struct brw_reg src0,
489 struct brw_reg src1);
490
491 void generate_set_simd4x2_offset(fs_inst *inst,
492 struct brw_reg dst,
493 struct brw_reg offset);
494 void generate_discard_jump(fs_inst *inst);
495
496 void generate_pack_half_2x16_split(fs_inst *inst,
497 struct brw_reg dst,
498 struct brw_reg x,
499 struct brw_reg y);
500 void generate_unpack_half_2x16_split(fs_inst *inst,
501 struct brw_reg dst,
502 struct brw_reg src);
503
504 void generate_shader_time_add(fs_inst *inst,
505 struct brw_reg payload,
506 struct brw_reg offset,
507 struct brw_reg value);
508
509 void generate_mov_indirect(fs_inst *inst,
510 struct brw_reg dst,
511 struct brw_reg reg,
512 struct brw_reg indirect_byte_offset);
513
514 bool patch_discard_jumps_to_fb_writes();
515
516 const struct brw_compiler *compiler;
517 void *log_data; /* Passed to compiler->*_log functions */
518
519 const struct brw_device_info *devinfo;
520
521 struct brw_codegen *p;
522 const void * const key;
523 struct brw_stage_prog_data * const prog_data;
524
525 unsigned dispatch_width; /**< 8 or 16 */
526
527 exec_list discard_halt_patches;
528 unsigned promoted_constants;
529 bool runtime_check_aads_emit;
530 bool debug_flag;
531 const char *shader_name;
532 gl_shader_stage stage;
533 void *mem_ctx;
534 };
535
536 bool brw_do_channel_expressions(struct exec_list *instructions);
537 bool brw_do_vector_splitting(struct exec_list *instructions);
538
539 void shuffle_32bit_load_result_to_64bit_data(const brw::fs_builder &bld,
540 const fs_reg &dst,
541 const fs_reg &src,
542 uint32_t components);
543
544 void shuffle_64bit_data_for_32bit_write(const brw::fs_builder &bld,
545 const fs_reg &dst,
546 const fs_reg &src,
547 uint32_t components);