i965/fs: Restrict optimization that would fail for gen7's SENDs from GRFs
[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 class 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
126 static const fs_reg reg_undef;
127 static const fs_reg reg_null_f(ARF, BRW_ARF_NULL, BRW_REGISTER_TYPE_F);
128 static const fs_reg reg_null_d(ARF, BRW_ARF_NULL, BRW_REGISTER_TYPE_D);
129
130 class fs_inst : public backend_instruction {
131 public:
132 /* Callers of this ralloc-based new need not call delete. It's
133 * easier to just ralloc_free 'ctx' (or any of its ancestors). */
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 void init();
145
146 fs_inst();
147 fs_inst(enum opcode opcode);
148 fs_inst(enum opcode opcode, fs_reg dst);
149 fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0);
150 fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1);
151 fs_inst(enum opcode opcode, fs_reg dst,
152 fs_reg src0, fs_reg src1,fs_reg src2);
153
154 bool equals(fs_inst *inst);
155 int regs_written();
156 bool overwrites_reg(const fs_reg &reg);
157 bool is_tex();
158 bool is_math();
159 bool is_send_from_grf();
160
161 fs_reg dst;
162 fs_reg src[3];
163 bool saturate;
164 int conditional_mod; /**< BRW_CONDITIONAL_* */
165
166 int mlen; /**< SEND message length */
167 int base_mrf; /**< First MRF in the SEND message, if mlen is nonzero. */
168 uint32_t texture_offset; /**< Texture offset bitfield */
169 int sampler;
170 int target; /**< MRT target. */
171 bool eot;
172 bool header_present;
173 bool shadow_compare;
174 bool force_uncompressed;
175 bool force_sechalf;
176 uint32_t offset; /* spill/unspill offset */
177
178 /** @{
179 * Annotation for the generated IR. One of the two can be set.
180 */
181 const void *ir;
182 const char *annotation;
183 /** @} */
184 };
185
186 /**
187 * The fragment shader front-end.
188 *
189 * Translates either GLSL IR or Mesa IR (for ARB_fragment_program) into FS IR.
190 */
191 class fs_visitor : public backend_visitor
192 {
193 public:
194
195 fs_visitor(struct brw_context *brw,
196 struct brw_wm_compile *c,
197 struct gl_shader_program *prog,
198 struct gl_fragment_program *fp,
199 unsigned dispatch_width);
200 ~fs_visitor();
201
202 fs_reg *variable_storage(ir_variable *var);
203 int virtual_grf_alloc(int size);
204 void import_uniforms(fs_visitor *v);
205
206 void visit(ir_variable *ir);
207 void visit(ir_assignment *ir);
208 void visit(ir_dereference_variable *ir);
209 void visit(ir_dereference_record *ir);
210 void visit(ir_dereference_array *ir);
211 void visit(ir_expression *ir);
212 void visit(ir_texture *ir);
213 void visit(ir_if *ir);
214 void visit(ir_constant *ir);
215 void visit(ir_swizzle *ir);
216 void visit(ir_return *ir);
217 void visit(ir_loop *ir);
218 void visit(ir_loop_jump *ir);
219 void visit(ir_discard *ir);
220 void visit(ir_call *ir);
221 void visit(ir_function *ir);
222 void visit(ir_function_signature *ir);
223
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
231 fs_inst *emit(enum opcode opcode);
232 fs_inst *emit(enum opcode opcode, fs_reg dst);
233 fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0);
234 fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1);
235 fs_inst *emit(enum opcode opcode, fs_reg dst,
236 fs_reg src0, fs_reg src1, fs_reg src2);
237
238 fs_inst *MOV(fs_reg dst, fs_reg src);
239 fs_inst *NOT(fs_reg dst, fs_reg src);
240 fs_inst *RNDD(fs_reg dst, fs_reg src);
241 fs_inst *RNDE(fs_reg dst, fs_reg src);
242 fs_inst *RNDZ(fs_reg dst, fs_reg src);
243 fs_inst *FRC(fs_reg dst, fs_reg src);
244 fs_inst *ADD(fs_reg dst, fs_reg src0, fs_reg src1);
245 fs_inst *MUL(fs_reg dst, fs_reg src0, fs_reg src1);
246 fs_inst *MACH(fs_reg dst, fs_reg src0, fs_reg src1);
247 fs_inst *MAC(fs_reg dst, fs_reg src0, fs_reg src1);
248 fs_inst *SHL(fs_reg dst, fs_reg src0, fs_reg src1);
249 fs_inst *SHR(fs_reg dst, fs_reg src0, fs_reg src1);
250 fs_inst *ASR(fs_reg dst, fs_reg src0, fs_reg src1);
251 fs_inst *AND(fs_reg dst, fs_reg src0, fs_reg src1);
252 fs_inst *OR(fs_reg dst, fs_reg src0, fs_reg src1);
253 fs_inst *XOR(fs_reg dst, fs_reg src0, fs_reg src1);
254 fs_inst *IF(uint32_t predicate);
255 fs_inst *IF(fs_reg src0, fs_reg src1, uint32_t condition);
256 fs_inst *CMP(fs_reg dst, fs_reg src0, fs_reg src1,
257 uint32_t condition);
258
259 int type_size(const struct glsl_type *type);
260 fs_inst *get_instruction_generating_reg(fs_inst *start,
261 fs_inst *end,
262 fs_reg reg);
263
264 bool run();
265 void setup_payload_gen4();
266 void setup_payload_gen6();
267 void setup_paramvalues_refs();
268 void assign_curb_setup();
269 void calculate_urb_setup();
270 void assign_urb_setup();
271 bool assign_regs();
272 void assign_regs_trivial();
273 void setup_payload_interference(struct ra_graph *g, int payload_reg_count,
274 int first_payload_node);
275 void setup_mrf_hack_interference(struct ra_graph *g,
276 int first_mrf_hack_node);
277 int choose_spill_reg(struct ra_graph *g);
278 void spill_reg(int spill_reg);
279 void split_virtual_grfs();
280 void compact_virtual_grfs();
281 void setup_pull_constants();
282 void calculate_live_intervals();
283 bool opt_algebraic();
284 bool opt_cse();
285 bool opt_cse_local(bblock_t *block, exec_list *aeb);
286 bool opt_copy_propagate();
287 bool try_copy_propagate(fs_inst *inst, int arg, acp_entry *entry);
288 bool try_constant_propagate(fs_inst *inst, acp_entry *entry);
289 bool opt_copy_propagate_local(void *mem_ctx, bblock_t *block,
290 exec_list *acp);
291 bool register_coalesce();
292 bool register_coalesce_2();
293 bool compute_to_mrf();
294 bool dead_code_eliminate();
295 bool remove_dead_constants();
296 bool remove_duplicate_mrf_writes();
297 bool virtual_grf_interferes(int a, int b);
298 void schedule_instructions();
299 void fail(const char *msg, ...);
300
301 void push_force_uncompressed();
302 void pop_force_uncompressed();
303 void push_force_sechalf();
304 void pop_force_sechalf();
305
306 void emit_dummy_fs();
307 fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
308 fs_inst *emit_linterp(const fs_reg &attr, const fs_reg &interp,
309 glsl_interp_qualifier interpolation_mode,
310 bool is_centroid);
311 fs_reg *emit_frontfacing_interpolation(ir_variable *ir);
312 fs_reg *emit_general_interpolation(ir_variable *ir);
313 void emit_interpolation_setup_gen4();
314 void emit_interpolation_setup_gen6();
315 fs_reg rescale_texcoord(ir_texture *ir, fs_reg coordinate,
316 bool is_rect, int sampler, int texunit);
317 fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate,
318 fs_reg shadow_comp, fs_reg lod, fs_reg lod2);
319 fs_inst *emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate,
320 fs_reg shadow_comp, fs_reg lod, fs_reg lod2);
321 fs_inst *emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
322 fs_reg shadow_comp, fs_reg lod, fs_reg lod2);
323 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0);
324 fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0, fs_reg src1);
325 void emit_minmax(uint32_t conditionalmod, fs_reg dst,
326 fs_reg src0, fs_reg src1);
327 bool try_emit_saturate(ir_expression *ir);
328 bool try_emit_mad(ir_expression *ir, int mul_arg);
329 void emit_bool_to_cond_code(ir_rvalue *condition);
330 void emit_if_gen6(ir_if *ir);
331 void emit_unspill(fs_inst *inst, fs_reg reg, uint32_t spill_offset);
332
333 void emit_fragment_program_code();
334 void setup_fp_regs();
335 fs_reg get_fp_src_reg(const prog_src_register *src);
336 fs_reg get_fp_dst_reg(const prog_dst_register *dst);
337 void emit_fp_alu1(enum opcode opcode,
338 const struct prog_instruction *fpi,
339 fs_reg dst, fs_reg src);
340 void emit_fp_alu2(enum opcode opcode,
341 const struct prog_instruction *fpi,
342 fs_reg dst, fs_reg src0, fs_reg src1);
343 void emit_fp_scalar_write(const struct prog_instruction *fpi,
344 fs_reg dst, fs_reg src);
345 void emit_fp_scalar_math(enum opcode opcode,
346 const struct prog_instruction *fpi,
347 fs_reg dst, fs_reg src);
348
349 void emit_fp_minmax(const struct prog_instruction *fpi,
350 fs_reg dst, fs_reg src0, fs_reg src1);
351
352 void emit_fp_sop(uint32_t conditional_mod,
353 const struct prog_instruction *fpi,
354 fs_reg dst, fs_reg src0, fs_reg src1, fs_reg one);
355
356 void emit_color_write(int target, int index, int first_color_mrf);
357 void emit_fb_writes();
358 bool try_rewrite_rhs_to_dst(ir_assignment *ir,
359 fs_reg dst,
360 fs_reg src,
361 fs_inst *pre_rhs_inst,
362 fs_inst *last_rhs_inst);
363 void emit_assignment_writes(fs_reg &l, fs_reg &r,
364 const glsl_type *type, bool predicated);
365 void resolve_ud_negate(fs_reg *reg);
366 void resolve_bool_comparison(ir_rvalue *rvalue, fs_reg *reg);
367
368 struct brw_reg interp_reg(int location, int channel);
369 int setup_uniform_values(int loc, const glsl_type *type);
370 void setup_builtin_uniform_values(ir_variable *ir);
371 int implied_mrf_writes(fs_inst *inst);
372
373 void dump_instructions();
374 void dump_instruction(fs_inst *inst);
375
376 const struct gl_fragment_program *fp;
377 struct brw_wm_compile *c;
378
379 /* Delayed setup of c->prog_data.params[] due to realloc of
380 * ParamValues[] during compile.
381 */
382 int param_index[MAX_UNIFORMS * 4];
383 int param_offset[MAX_UNIFORMS * 4];
384
385 int *virtual_grf_sizes;
386 int virtual_grf_count;
387 int virtual_grf_array_size;
388 int *virtual_grf_def;
389 int *virtual_grf_use;
390 bool live_intervals_valid;
391
392 /* This is the map from UNIFORM hw_reg + reg_offset as generated by
393 * the visitor to the packed uniform number after
394 * remove_dead_constants() that represents the actual uploaded
395 * uniform index.
396 */
397 int *params_remap;
398
399 struct hash_table *variable_ht;
400 fs_reg frag_depth;
401 fs_reg outputs[BRW_MAX_DRAW_BUFFERS];
402 unsigned output_components[BRW_MAX_DRAW_BUFFERS];
403 fs_reg dual_src_output;
404 int first_non_payload_grf;
405 /** Either BRW_MAX_GRF or GEN7_MRF_HACK_START */
406 int max_grf;
407 int urb_setup[FRAG_ATTRIB_MAX];
408
409 fs_reg *fp_temp_regs;
410 fs_reg *fp_input_regs;
411
412 /** @{ debug annotation info */
413 const char *current_annotation;
414 const void *base_ir;
415 /** @} */
416
417 bool failed;
418 char *fail_msg;
419
420 /* Result of last visit() method. */
421 fs_reg result;
422
423 fs_reg pixel_x;
424 fs_reg pixel_y;
425 fs_reg wpos_w;
426 fs_reg pixel_w;
427 fs_reg delta_x[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
428 fs_reg delta_y[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
429
430 int grf_used;
431
432 const unsigned dispatch_width; /**< 8 or 16 */
433
434 int force_uncompressed_stack;
435 int force_sechalf_stack;
436 };
437
438 /**
439 * The fragment shader code generator.
440 *
441 * Translates FS IR to actual i965 assembly code.
442 */
443 class fs_generator
444 {
445 public:
446 fs_generator(struct brw_context *brw,
447 struct brw_wm_compile *c,
448 struct gl_shader_program *prog,
449 struct gl_fragment_program *fp,
450 bool dual_source_output);
451 ~fs_generator();
452
453 const unsigned *generate_assembly(exec_list *simd8_instructions,
454 exec_list *simd16_instructions,
455 unsigned *assembly_size);
456
457 private:
458 void generate_code(exec_list *instructions);
459 void generate_fb_write(fs_inst *inst);
460 void generate_pixel_xy(struct brw_reg dst, bool is_x);
461 void generate_linterp(fs_inst *inst, struct brw_reg dst,
462 struct brw_reg *src);
463 void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
464 void generate_math1_gen7(fs_inst *inst,
465 struct brw_reg dst,
466 struct brw_reg src);
467 void generate_math2_gen7(fs_inst *inst,
468 struct brw_reg dst,
469 struct brw_reg src0,
470 struct brw_reg src1);
471 void generate_math1_gen6(fs_inst *inst,
472 struct brw_reg dst,
473 struct brw_reg src);
474 void generate_math2_gen6(fs_inst *inst,
475 struct brw_reg dst,
476 struct brw_reg src0,
477 struct brw_reg src1);
478 void generate_math_gen4(fs_inst *inst,
479 struct brw_reg dst,
480 struct brw_reg src);
481 void generate_discard(fs_inst *inst);
482 void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
483 void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
484 bool negate_value);
485 void generate_spill(fs_inst *inst, struct brw_reg src);
486 void generate_unspill(fs_inst *inst, struct brw_reg dst);
487 void generate_uniform_pull_constant_load(fs_inst *inst, struct brw_reg dst,
488 struct brw_reg index,
489 struct brw_reg offset);
490 void generate_varying_pull_constant_load(fs_inst *inst, struct brw_reg dst,
491 struct brw_reg index);
492 void generate_varying_pull_constant_load_gen7(fs_inst *inst,
493 struct brw_reg dst,
494 struct brw_reg index,
495 struct brw_reg offset);
496 void generate_mov_dispatch_to_flags();
497
498 struct brw_context *brw;
499 struct intel_context *intel;
500 struct gl_context *ctx;
501
502 struct brw_compile *p;
503 struct brw_wm_compile *c;
504
505 struct gl_shader_program *prog;
506 struct gl_shader *shader;
507 const struct gl_fragment_program *fp;
508
509 unsigned dispatch_width; /**< 8 or 16 */
510
511 bool dual_source_output;
512 void *mem_ctx;
513 };
514
515 bool brw_do_channel_expressions(struct exec_list *instructions);
516 bool brw_do_vector_splitting(struct exec_list *instructions);
517 bool brw_fs_precompile(struct gl_context *ctx, struct gl_shader_program *prog);