v3d: Garbage collect unused uniforms code.
[mesa.git] / src / broadcom / compiler / v3d_compiler.h
1 /*
2 * Copyright © 2016 Broadcom
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
24 #ifndef V3D_COMPILER_H
25 #define V3D_COMPILER_H
26
27 #include <assert.h>
28 #include <stdio.h>
29 #include <stdlib.h>
30 #include <stdbool.h>
31 #include <stdint.h>
32 #include <string.h>
33
34 #include "util/macros.h"
35 #include "common/v3d_debug.h"
36 #include "common/v3d_device_info.h"
37 #include "compiler/nir/nir.h"
38 #include "util/list.h"
39 #include "util/u_math.h"
40
41 #include "qpu/qpu_instr.h"
42 #include "pipe/p_state.h"
43
44 #define V3D_MAX_TEXTURE_SAMPLERS 32
45 #define V3D_MAX_SAMPLES 4
46 #define V3D_MAX_FS_INPUTS 64
47 #define V3D_MAX_VS_INPUTS 64
48
49 struct nir_builder;
50
51 struct v3d_fs_inputs {
52 /**
53 * Array of the meanings of the VPM inputs this shader needs.
54 *
55 * It doesn't include those that aren't part of the VPM, like
56 * point/line coordinates.
57 */
58 struct v3d_varying_slot *input_slots;
59 uint32_t num_inputs;
60 };
61
62 enum qfile {
63 /** An unused source or destination register. */
64 QFILE_NULL,
65
66 /** A physical register, such as the W coordinate payload. */
67 QFILE_REG,
68 /** One of the regsiters for fixed function interactions. */
69 QFILE_MAGIC,
70
71 /**
72 * A virtual register, that will be allocated to actual accumulator
73 * or physical registers later.
74 */
75 QFILE_TEMP,
76 QFILE_UNIF,
77 QFILE_TLB,
78 QFILE_TLBU,
79
80 /**
81 * VPM reads use this with an index value to say what part of the VPM
82 * is being read.
83 */
84 QFILE_VPM,
85
86 /**
87 * Stores an immediate value in the index field that will be used
88 * directly by qpu_load_imm().
89 */
90 QFILE_LOAD_IMM,
91
92 /**
93 * Stores an immediate value in the index field that can be turned
94 * into a small immediate field by qpu_encode_small_immediate().
95 */
96 QFILE_SMALL_IMM,
97 };
98
99 /**
100 * A reference to a QPU register or a virtual temp register.
101 */
102 struct qreg {
103 enum qfile file;
104 uint32_t index;
105 };
106
107 static inline struct qreg vir_reg(enum qfile file, uint32_t index)
108 {
109 return (struct qreg){file, index};
110 }
111
112 /**
113 * A reference to an actual register at the QPU level, for register
114 * allocation.
115 */
116 struct qpu_reg {
117 bool magic;
118 bool smimm;
119 int index;
120 };
121
122 struct qinst {
123 /** Entry in qblock->instructions */
124 struct list_head link;
125
126 /**
127 * The instruction being wrapped. Its condition codes, pack flags,
128 * signals, etc. will all be used, with just the register references
129 * being replaced by the contents of qinst->dst and qinst->src[].
130 */
131 struct v3d_qpu_instr qpu;
132
133 /* Pre-register-allocation references to src/dst registers */
134 struct qreg dst;
135 struct qreg src[3];
136 bool cond_is_exec_mask;
137 bool has_implicit_uniform;
138 bool is_last_thrsw;
139
140 /* After vir_to_qpu.c: If instr reads a uniform, which uniform from
141 * the uncompiled stream it is.
142 */
143 int uniform;
144 };
145
146 enum quniform_contents {
147 /**
148 * Indicates that a constant 32-bit value is copied from the program's
149 * uniform contents.
150 */
151 QUNIFORM_CONSTANT,
152 /**
153 * Indicates that the program's uniform contents are used as an index
154 * into the GL uniform storage.
155 */
156 QUNIFORM_UNIFORM,
157
158 /** @{
159 * Scaling factors from clip coordinates to relative to the viewport
160 * center.
161 *
162 * This is used by the coordinate and vertex shaders to produce the
163 * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed
164 * point offsets from the viewport ccenter.
165 */
166 QUNIFORM_VIEWPORT_X_SCALE,
167 QUNIFORM_VIEWPORT_Y_SCALE,
168 /** @} */
169
170 QUNIFORM_VIEWPORT_Z_OFFSET,
171 QUNIFORM_VIEWPORT_Z_SCALE,
172
173 QUNIFORM_USER_CLIP_PLANE,
174
175 /**
176 * A reference to a V3D 3.x texture config parameter 0 uniform.
177 *
178 * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
179 * defines texture type, miplevels, and such. It will be found as a
180 * parameter to the first QOP_TEX_[STRB] instruction in a sequence.
181 */
182 QUNIFORM_TEXTURE_CONFIG_P0_0,
183 QUNIFORM_TEXTURE_CONFIG_P0_1,
184 QUNIFORM_TEXTURE_CONFIG_P0_2,
185 QUNIFORM_TEXTURE_CONFIG_P0_3,
186 QUNIFORM_TEXTURE_CONFIG_P0_4,
187 QUNIFORM_TEXTURE_CONFIG_P0_5,
188 QUNIFORM_TEXTURE_CONFIG_P0_6,
189 QUNIFORM_TEXTURE_CONFIG_P0_7,
190 QUNIFORM_TEXTURE_CONFIG_P0_8,
191 QUNIFORM_TEXTURE_CONFIG_P0_9,
192 QUNIFORM_TEXTURE_CONFIG_P0_10,
193 QUNIFORM_TEXTURE_CONFIG_P0_11,
194 QUNIFORM_TEXTURE_CONFIG_P0_12,
195 QUNIFORM_TEXTURE_CONFIG_P0_13,
196 QUNIFORM_TEXTURE_CONFIG_P0_14,
197 QUNIFORM_TEXTURE_CONFIG_P0_15,
198 QUNIFORM_TEXTURE_CONFIG_P0_16,
199 QUNIFORM_TEXTURE_CONFIG_P0_17,
200 QUNIFORM_TEXTURE_CONFIG_P0_18,
201 QUNIFORM_TEXTURE_CONFIG_P0_19,
202 QUNIFORM_TEXTURE_CONFIG_P0_20,
203 QUNIFORM_TEXTURE_CONFIG_P0_21,
204 QUNIFORM_TEXTURE_CONFIG_P0_22,
205 QUNIFORM_TEXTURE_CONFIG_P0_23,
206 QUNIFORM_TEXTURE_CONFIG_P0_24,
207 QUNIFORM_TEXTURE_CONFIG_P0_25,
208 QUNIFORM_TEXTURE_CONFIG_P0_26,
209 QUNIFORM_TEXTURE_CONFIG_P0_27,
210 QUNIFORM_TEXTURE_CONFIG_P0_28,
211 QUNIFORM_TEXTURE_CONFIG_P0_29,
212 QUNIFORM_TEXTURE_CONFIG_P0_30,
213 QUNIFORM_TEXTURE_CONFIG_P0_31,
214 QUNIFORM_TEXTURE_CONFIG_P0_32,
215
216 /**
217 * A reference to a V3D 3.x texture config parameter 1 uniform.
218 *
219 * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
220 * has the pointer to the indirect texture state. Our data[] field
221 * will have a packed p1 value, but the address field will be just
222 * which texture unit's texture should be referenced.
223 */
224 QUNIFORM_TEXTURE_CONFIG_P1,
225
226 /* A V3D 4.x texture config parameter. The high 8 bits will be
227 * which texture or sampler is being sampled, and the driver must
228 * replace the address field with the appropriate address.
229 */
230 QUNIFORM_TMU_CONFIG_P0,
231 QUNIFORM_TMU_CONFIG_P1,
232
233 QUNIFORM_TEXTURE_FIRST_LEVEL,
234
235 QUNIFORM_TEXTURE_WIDTH,
236 QUNIFORM_TEXTURE_HEIGHT,
237 QUNIFORM_TEXTURE_DEPTH,
238 QUNIFORM_TEXTURE_ARRAY_SIZE,
239 QUNIFORM_TEXTURE_LEVELS,
240
241 QUNIFORM_UBO_ADDR,
242
243 QUNIFORM_TEXRECT_SCALE_X,
244 QUNIFORM_TEXRECT_SCALE_Y,
245
246 QUNIFORM_ALPHA_REF,
247
248 /**
249 * Returns the the offset of the scratch buffer for register spilling.
250 */
251 QUNIFORM_SPILL_OFFSET,
252 QUNIFORM_SPILL_SIZE_PER_THREAD,
253 };
254
255 struct v3d_varying_slot {
256 uint8_t slot_and_component;
257 };
258
259 static inline struct v3d_varying_slot
260 v3d_slot_from_slot_and_component(uint8_t slot, uint8_t component)
261 {
262 assert(slot < 255 / 4);
263 return (struct v3d_varying_slot){ (slot << 2) + component };
264 }
265
266 static inline uint8_t v3d_slot_get_slot(struct v3d_varying_slot slot)
267 {
268 return slot.slot_and_component >> 2;
269 }
270
271 static inline uint8_t v3d_slot_get_component(struct v3d_varying_slot slot)
272 {
273 return slot.slot_and_component & 3;
274 }
275
276 struct v3d_ubo_range {
277 /**
278 * offset in bytes from the start of the ubo where this range is
279 * uploaded.
280 *
281 * Only set once used is set.
282 */
283 uint32_t dst_offset;
284
285 /**
286 * offset in bytes from the start of the gallium uniforms where the
287 * data comes from.
288 */
289 uint32_t src_offset;
290
291 /** size in bytes of this ubo range */
292 uint32_t size;
293 };
294
295 struct v3d_key {
296 void *shader_state;
297 struct {
298 uint8_t swizzle[4];
299 uint8_t return_size;
300 uint8_t return_channels;
301 unsigned compare_mode:1;
302 unsigned compare_func:3;
303 bool clamp_s:1;
304 bool clamp_t:1;
305 bool clamp_r:1;
306 } tex[V3D_MAX_TEXTURE_SAMPLERS];
307 uint8_t ucp_enables;
308 };
309
310 struct v3d_fs_key {
311 struct v3d_key base;
312 bool depth_enabled;
313 bool is_points;
314 bool is_lines;
315 bool alpha_test;
316 bool point_coord_upper_left;
317 bool light_twoside;
318 bool msaa;
319 bool sample_coverage;
320 bool sample_alpha_to_coverage;
321 bool sample_alpha_to_one;
322 bool clamp_color;
323 bool shade_model_flat;
324 uint8_t nr_cbufs;
325 uint8_t swap_color_rb;
326 /* Mask of which render targets need to be written as 32-bit floats */
327 uint8_t f32_color_rb;
328 /* Masks of which render targets need to be written as ints/uints.
329 * Used by gallium to work around lost information in TGSI.
330 */
331 uint8_t int_color_rb;
332 uint8_t uint_color_rb;
333 uint8_t alpha_test_func;
334 uint8_t logicop_func;
335 uint32_t point_sprite_mask;
336
337 struct pipe_rt_blend_state blend;
338 };
339
340 struct v3d_vs_key {
341 struct v3d_key base;
342
343 struct v3d_varying_slot fs_inputs[V3D_MAX_FS_INPUTS];
344 uint8_t num_fs_inputs;
345
346 bool is_coord;
347 bool per_vertex_point_size;
348 bool clamp_color;
349 };
350
351 /** A basic block of VIR intructions. */
352 struct qblock {
353 struct list_head link;
354
355 struct list_head instructions;
356
357 struct set *predecessors;
358 struct qblock *successors[2];
359
360 int index;
361
362 /* Instruction IPs for the first and last instruction of the block.
363 * Set by qpu_schedule.c.
364 */
365 uint32_t start_qpu_ip;
366 uint32_t end_qpu_ip;
367
368 /* Instruction IP for the branch instruction of the block. Set by
369 * qpu_schedule.c.
370 */
371 uint32_t branch_qpu_ip;
372
373 /** Offset within the uniform stream at the start of the block. */
374 uint32_t start_uniform;
375 /** Offset within the uniform stream of the branch instruction */
376 uint32_t branch_uniform;
377
378 /** @{ used by v3d_vir_live_variables.c */
379 BITSET_WORD *def;
380 BITSET_WORD *use;
381 BITSET_WORD *live_in;
382 BITSET_WORD *live_out;
383 int start_ip, end_ip;
384 /** @} */
385 };
386
387 /** Which util/list.h add mode we should use when inserting an instruction. */
388 enum vir_cursor_mode {
389 vir_cursor_add,
390 vir_cursor_addtail,
391 };
392
393 /**
394 * Tracking structure for where new instructions should be inserted. Create
395 * with one of the vir_after_inst()-style helper functions.
396 *
397 * This does not protect against removal of the block or instruction, so we
398 * have an assert in instruction removal to try to catch it.
399 */
400 struct vir_cursor {
401 enum vir_cursor_mode mode;
402 struct list_head *link;
403 };
404
405 static inline struct vir_cursor
406 vir_before_inst(struct qinst *inst)
407 {
408 return (struct vir_cursor){ vir_cursor_addtail, &inst->link };
409 }
410
411 static inline struct vir_cursor
412 vir_after_inst(struct qinst *inst)
413 {
414 return (struct vir_cursor){ vir_cursor_add, &inst->link };
415 }
416
417 static inline struct vir_cursor
418 vir_before_block(struct qblock *block)
419 {
420 return (struct vir_cursor){ vir_cursor_add, &block->instructions };
421 }
422
423 static inline struct vir_cursor
424 vir_after_block(struct qblock *block)
425 {
426 return (struct vir_cursor){ vir_cursor_addtail, &block->instructions };
427 }
428
429 /**
430 * Compiler state saved across compiler invocations, for any expensive global
431 * setup.
432 */
433 struct v3d_compiler {
434 const struct v3d_device_info *devinfo;
435 struct ra_regs *regs;
436 unsigned int reg_class_phys[3];
437 unsigned int reg_class_phys_or_acc[3];
438 };
439
440 struct v3d_compile {
441 const struct v3d_device_info *devinfo;
442 nir_shader *s;
443 nir_function_impl *impl;
444 struct exec_list *cf_node_list;
445 const struct v3d_compiler *compiler;
446
447 /**
448 * Mapping from nir_register * or nir_ssa_def * to array of struct
449 * qreg for the values.
450 */
451 struct hash_table *def_ht;
452
453 /* For each temp, the instruction generating its value. */
454 struct qinst **defs;
455 uint32_t defs_array_size;
456
457 /**
458 * Inputs to the shader, arranged by TGSI declaration order.
459 *
460 * Not all fragment shader QFILE_VARY reads are present in this array.
461 */
462 struct qreg *inputs;
463 struct qreg *outputs;
464 bool msaa_per_sample_output;
465 struct qreg color_reads[V3D_MAX_SAMPLES];
466 struct qreg sample_colors[V3D_MAX_SAMPLES];
467 uint32_t inputs_array_size;
468 uint32_t outputs_array_size;
469 uint32_t uniforms_array_size;
470
471 /* Booleans for whether the corresponding QFILE_VARY[i] is
472 * flat-shaded. This includes gl_FragColor flat-shading, which is
473 * customized based on the shademodel_flat shader key.
474 */
475 uint32_t flat_shade_flags[BITSET_WORDS(V3D_MAX_FS_INPUTS)];
476
477 uint32_t noperspective_flags[BITSET_WORDS(V3D_MAX_FS_INPUTS)];
478
479 uint32_t centroid_flags[BITSET_WORDS(V3D_MAX_FS_INPUTS)];
480
481 bool uses_center_w;
482
483 struct v3d_ubo_range *ubo_ranges;
484 bool *ubo_range_used;
485 uint32_t ubo_ranges_array_size;
486 /** Number of uniform areas tracked in ubo_ranges. */
487 uint32_t num_ubo_ranges;
488 uint32_t next_ubo_dst_offset;
489
490 /* State for whether we're executing on each channel currently. 0 if
491 * yes, otherwise a block number + 1 that the channel jumped to.
492 */
493 struct qreg execute;
494
495 struct qreg line_x, point_x, point_y;
496
497 /**
498 * Instance ID, which comes in before the vertex attribute payload if
499 * the shader record requests it.
500 */
501 struct qreg iid;
502
503 /**
504 * Vertex ID, which comes in before the vertex attribute payload
505 * (after Instance ID) if the shader record requests it.
506 */
507 struct qreg vid;
508
509 /* Fragment shader payload regs. */
510 struct qreg payload_w, payload_w_centroid, payload_z;
511
512 uint8_t vattr_sizes[V3D_MAX_VS_INPUTS];
513 uint32_t num_vpm_writes;
514
515 /* Size in bytes of registers that have been spilled. This is how much
516 * space needs to be available in the spill BO per thread per QPU.
517 */
518 uint32_t spill_size;
519 /* Shader-db stats for register spilling. */
520 uint32_t spills, fills;
521 /**
522 * Register spilling's per-thread base address, shared between each
523 * spill/fill's addressing calculations.
524 */
525 struct qreg spill_base;
526 /* Bit vector of which temps may be spilled */
527 BITSET_WORD *spillable;
528
529 /**
530 * Array of the VARYING_SLOT_* of all FS QFILE_VARY reads.
531 *
532 * This includes those that aren't part of the VPM varyings, like
533 * point/line coordinates.
534 */
535 struct v3d_varying_slot input_slots[V3D_MAX_FS_INPUTS];
536
537 /**
538 * An entry per outputs[] in the VS indicating what the VARYING_SLOT_*
539 * of the output is. Used to emit from the VS in the order that the
540 * FS needs.
541 */
542 struct v3d_varying_slot *output_slots;
543
544 struct pipe_shader_state *shader_state;
545 struct v3d_key *key;
546 struct v3d_fs_key *fs_key;
547 struct v3d_vs_key *vs_key;
548
549 /* Live ranges of temps. */
550 int *temp_start, *temp_end;
551 bool live_intervals_valid;
552
553 uint32_t *uniform_data;
554 enum quniform_contents *uniform_contents;
555 uint32_t uniform_array_size;
556 uint32_t num_uniforms;
557 uint32_t num_outputs;
558 uint32_t output_position_index;
559 nir_variable *output_color_var[4];
560 uint32_t output_point_size_index;
561 uint32_t output_sample_mask_index;
562
563 struct qreg undef;
564 uint32_t num_temps;
565
566 struct vir_cursor cursor;
567 struct list_head blocks;
568 int next_block_index;
569 struct qblock *cur_block;
570 struct qblock *loop_cont_block;
571 struct qblock *loop_break_block;
572
573 uint64_t *qpu_insts;
574 uint32_t qpu_inst_count;
575 uint32_t qpu_inst_size;
576
577 /* For the FS, the number of varying inputs not counting the
578 * point/line varyings payload
579 */
580 uint32_t num_inputs;
581
582 /**
583 * Number of inputs from num_inputs remaining to be queued to the read
584 * FIFO in the VS/CS.
585 */
586 uint32_t num_inputs_remaining;
587
588 /* Number of inputs currently in the read FIFO for the VS/CS */
589 uint32_t num_inputs_in_fifo;
590
591 /** Next offset in the VPM to read from in the VS/CS */
592 uint32_t vpm_read_offset;
593
594 uint32_t program_id;
595 uint32_t variant_id;
596
597 /* Set to compile program in in 1x, 2x, or 4x threaded mode, where
598 * SIG_THREAD_SWITCH is used to hide texturing latency at the cost of
599 * limiting ourselves to the part of the physical reg space.
600 *
601 * On V3D 3.x, 2x or 4x divide the physical reg space by 2x or 4x. On
602 * V3D 4.x, all shaders are 2x threaded, and 4x only divides the
603 * physical reg space in half.
604 */
605 uint8_t threads;
606 struct qinst *last_thrsw;
607 bool last_thrsw_at_top_level;
608
609 bool failed;
610 };
611
612 struct v3d_uniform_list {
613 enum quniform_contents *contents;
614 uint32_t *data;
615 uint32_t count;
616 };
617
618 struct v3d_prog_data {
619 struct v3d_uniform_list uniforms;
620
621 struct v3d_ubo_range *ubo_ranges;
622 uint32_t num_ubo_ranges;
623 uint32_t ubo_size;
624 uint32_t spill_size;
625
626 uint8_t num_inputs;
627 uint8_t threads;
628
629 /* For threads > 1, whether the program should be dispatched in the
630 * after-final-THRSW state.
631 */
632 bool single_seg;
633 };
634
635 struct v3d_vs_prog_data {
636 struct v3d_prog_data base;
637
638 bool uses_iid, uses_vid;
639
640 /* Number of components read from each vertex attribute. */
641 uint8_t vattr_sizes[32];
642
643 /* Total number of components read, for the shader state record. */
644 uint32_t vpm_input_size;
645
646 /* Total number of components written, for the shader state record. */
647 uint32_t vpm_output_size;
648
649 /* Set if there should be separate VPM segments for input and output.
650 * If unset, vpm_input_size will be 0.
651 */
652 bool separate_segments;
653
654 /* Value to be programmed in VCM_CACHE_SIZE. */
655 uint8_t vcm_cache_size;
656 };
657
658 struct v3d_fs_prog_data {
659 struct v3d_prog_data base;
660
661 struct v3d_varying_slot input_slots[V3D_MAX_FS_INPUTS];
662
663 /* Array of flat shade flags.
664 *
665 * Each entry is only 24 bits (high 8 bits 0), to match the hardware
666 * packet layout.
667 */
668 uint32_t flat_shade_flags[((V3D_MAX_FS_INPUTS - 1) / 24) + 1];
669
670 uint32_t noperspective_flags[((V3D_MAX_FS_INPUTS - 1) / 24) + 1];
671
672 uint32_t centroid_flags[((V3D_MAX_FS_INPUTS - 1) / 24) + 1];
673
674 bool writes_z;
675 bool discard;
676 bool uses_center_w;
677 };
678
679 /* Special nir_load_input intrinsic index for loading the current TLB
680 * destination color.
681 */
682 #define V3D_NIR_TLB_COLOR_READ_INPUT 2000000000
683
684 #define V3D_NIR_MS_MASK_OUTPUT 2000000000
685
686 extern const nir_shader_compiler_options v3d_nir_options;
687
688 const struct v3d_compiler *v3d_compiler_init(const struct v3d_device_info *devinfo);
689 void v3d_compiler_free(const struct v3d_compiler *compiler);
690 void v3d_optimize_nir(struct nir_shader *s);
691
692 uint64_t *v3d_compile_vs(const struct v3d_compiler *compiler,
693 struct v3d_vs_key *key,
694 struct v3d_vs_prog_data *prog_data,
695 nir_shader *s,
696 int program_id, int variant_id,
697 uint32_t *final_assembly_size);
698
699 uint64_t *v3d_compile_fs(const struct v3d_compiler *compiler,
700 struct v3d_fs_key *key,
701 struct v3d_fs_prog_data *prog_data,
702 nir_shader *s,
703 int program_id, int variant_id,
704 uint32_t *final_assembly_size);
705
706 void v3d_nir_to_vir(struct v3d_compile *c);
707
708 void vir_compile_destroy(struct v3d_compile *c);
709 const char *vir_get_stage_name(struct v3d_compile *c);
710 struct qblock *vir_new_block(struct v3d_compile *c);
711 void vir_set_emit_block(struct v3d_compile *c, struct qblock *block);
712 void vir_link_blocks(struct qblock *predecessor, struct qblock *successor);
713 struct qblock *vir_entry_block(struct v3d_compile *c);
714 struct qblock *vir_exit_block(struct v3d_compile *c);
715 struct qinst *vir_add_inst(enum v3d_qpu_add_op op, struct qreg dst,
716 struct qreg src0, struct qreg src1);
717 struct qinst *vir_mul_inst(enum v3d_qpu_mul_op op, struct qreg dst,
718 struct qreg src0, struct qreg src1);
719 struct qinst *vir_branch_inst(enum v3d_qpu_branch_cond cond, struct qreg src0);
720 void vir_remove_instruction(struct v3d_compile *c, struct qinst *qinst);
721 struct qreg vir_uniform(struct v3d_compile *c,
722 enum quniform_contents contents,
723 uint32_t data);
724 void vir_schedule_instructions(struct v3d_compile *c);
725 struct v3d_qpu_instr v3d_qpu_nop(void);
726
727 struct qreg vir_emit_def(struct v3d_compile *c, struct qinst *inst);
728 struct qinst *vir_emit_nondef(struct v3d_compile *c, struct qinst *inst);
729 void vir_set_cond(struct qinst *inst, enum v3d_qpu_cond cond);
730 void vir_set_pf(struct qinst *inst, enum v3d_qpu_pf pf);
731 void vir_set_unpack(struct qinst *inst, int src,
732 enum v3d_qpu_input_unpack unpack);
733
734 struct qreg vir_get_temp(struct v3d_compile *c);
735 void vir_emit_last_thrsw(struct v3d_compile *c);
736 void vir_calculate_live_intervals(struct v3d_compile *c);
737 bool vir_has_implicit_uniform(struct qinst *inst);
738 int vir_get_implicit_uniform_src(struct qinst *inst);
739 int vir_get_non_sideband_nsrc(struct qinst *inst);
740 int vir_get_nsrc(struct qinst *inst);
741 bool vir_has_side_effects(struct v3d_compile *c, struct qinst *inst);
742 bool vir_get_add_op(struct qinst *inst, enum v3d_qpu_add_op *op);
743 bool vir_get_mul_op(struct qinst *inst, enum v3d_qpu_mul_op *op);
744 bool vir_is_raw_mov(struct qinst *inst);
745 bool vir_is_tex(struct qinst *inst);
746 bool vir_is_add(struct qinst *inst);
747 bool vir_is_mul(struct qinst *inst);
748 bool vir_is_float_input(struct qinst *inst);
749 bool vir_depends_on_flags(struct qinst *inst);
750 bool vir_writes_r3(const struct v3d_device_info *devinfo, struct qinst *inst);
751 bool vir_writes_r4(const struct v3d_device_info *devinfo, struct qinst *inst);
752 struct qreg vir_follow_movs(struct v3d_compile *c, struct qreg reg);
753 uint8_t vir_channels_written(struct qinst *inst);
754 struct qreg ntq_get_src(struct v3d_compile *c, nir_src src, int i);
755 void ntq_store_dest(struct v3d_compile *c, nir_dest *dest, int chan,
756 struct qreg result);
757 void vir_emit_thrsw(struct v3d_compile *c);
758
759 void vir_dump(struct v3d_compile *c);
760 void vir_dump_inst(struct v3d_compile *c, struct qinst *inst);
761
762 void vir_validate(struct v3d_compile *c);
763
764 void vir_optimize(struct v3d_compile *c);
765 bool vir_opt_algebraic(struct v3d_compile *c);
766 bool vir_opt_constant_folding(struct v3d_compile *c);
767 bool vir_opt_copy_propagate(struct v3d_compile *c);
768 bool vir_opt_dead_code(struct v3d_compile *c);
769 bool vir_opt_peephole_sf(struct v3d_compile *c);
770 bool vir_opt_small_immediates(struct v3d_compile *c);
771 bool vir_opt_vpm(struct v3d_compile *c);
772 void v3d_nir_lower_blend(nir_shader *s, struct v3d_compile *c);
773 void v3d_nir_lower_io(nir_shader *s, struct v3d_compile *c);
774 void v3d_nir_lower_txf_ms(nir_shader *s, struct v3d_compile *c);
775 void vir_lower_uniforms(struct v3d_compile *c);
776
777 void v3d33_vir_vpm_read_setup(struct v3d_compile *c, int num_components);
778 void v3d33_vir_vpm_write_setup(struct v3d_compile *c);
779 void v3d33_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr);
780 void v3d40_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr);
781
782 void v3d_vir_to_qpu(struct v3d_compile *c, struct qpu_reg *temp_registers);
783 uint32_t v3d_qpu_schedule_instructions(struct v3d_compile *c);
784 void qpu_validate(struct v3d_compile *c);
785 struct qpu_reg *v3d_register_allocate(struct v3d_compile *c, bool *spilled);
786 bool vir_init_reg_sets(struct v3d_compiler *compiler);
787
788 void vir_PF(struct v3d_compile *c, struct qreg src, enum v3d_qpu_pf pf);
789
790 static inline bool
791 quniform_contents_is_texture_p0(enum quniform_contents contents)
792 {
793 return (contents >= QUNIFORM_TEXTURE_CONFIG_P0_0 &&
794 contents < (QUNIFORM_TEXTURE_CONFIG_P0_0 +
795 V3D_MAX_TEXTURE_SAMPLERS));
796 }
797
798 static inline struct qreg
799 vir_uniform_ui(struct v3d_compile *c, uint32_t ui)
800 {
801 return vir_uniform(c, QUNIFORM_CONSTANT, ui);
802 }
803
804 static inline struct qreg
805 vir_uniform_f(struct v3d_compile *c, float f)
806 {
807 return vir_uniform(c, QUNIFORM_CONSTANT, fui(f));
808 }
809
810 #define VIR_ALU0(name, vir_inst, op) \
811 static inline struct qreg \
812 vir_##name(struct v3d_compile *c) \
813 { \
814 return vir_emit_def(c, vir_inst(op, c->undef, \
815 c->undef, c->undef)); \
816 } \
817 static inline struct qinst * \
818 vir_##name##_dest(struct v3d_compile *c, struct qreg dest) \
819 { \
820 return vir_emit_nondef(c, vir_inst(op, dest, \
821 c->undef, c->undef)); \
822 }
823
824 #define VIR_ALU1(name, vir_inst, op) \
825 static inline struct qreg \
826 vir_##name(struct v3d_compile *c, struct qreg a) \
827 { \
828 return vir_emit_def(c, vir_inst(op, c->undef, \
829 a, c->undef)); \
830 } \
831 static inline struct qinst * \
832 vir_##name##_dest(struct v3d_compile *c, struct qreg dest, \
833 struct qreg a) \
834 { \
835 return vir_emit_nondef(c, vir_inst(op, dest, a, \
836 c->undef)); \
837 }
838
839 #define VIR_ALU2(name, vir_inst, op) \
840 static inline struct qreg \
841 vir_##name(struct v3d_compile *c, struct qreg a, struct qreg b) \
842 { \
843 return vir_emit_def(c, vir_inst(op, c->undef, a, b)); \
844 } \
845 static inline struct qinst * \
846 vir_##name##_dest(struct v3d_compile *c, struct qreg dest, \
847 struct qreg a, struct qreg b) \
848 { \
849 return vir_emit_nondef(c, vir_inst(op, dest, a, b)); \
850 }
851
852 #define VIR_NODST_0(name, vir_inst, op) \
853 static inline struct qinst * \
854 vir_##name(struct v3d_compile *c) \
855 { \
856 return vir_emit_nondef(c, vir_inst(op, c->undef, \
857 c->undef, c->undef)); \
858 }
859
860 #define VIR_NODST_1(name, vir_inst, op) \
861 static inline struct qinst * \
862 vir_##name(struct v3d_compile *c, struct qreg a) \
863 { \
864 return vir_emit_nondef(c, vir_inst(op, c->undef, \
865 a, c->undef)); \
866 }
867
868 #define VIR_NODST_2(name, vir_inst, op) \
869 static inline struct qinst * \
870 vir_##name(struct v3d_compile *c, struct qreg a, struct qreg b) \
871 { \
872 return vir_emit_nondef(c, vir_inst(op, c->undef, \
873 a, b)); \
874 }
875
876 #define VIR_SFU(name) \
877 static inline struct qreg \
878 vir_##name(struct v3d_compile *c, struct qreg a) \
879 { \
880 if (c->devinfo->ver >= 41) { \
881 return vir_emit_def(c, vir_add_inst(V3D_QPU_A_##name, \
882 c->undef, \
883 a, c->undef)); \
884 } else { \
885 vir_FMOV_dest(c, vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_##name), a); \
886 return vir_FMOV(c, vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_R4)); \
887 } \
888 } \
889 static inline struct qinst * \
890 vir_##name##_dest(struct v3d_compile *c, struct qreg dest, \
891 struct qreg a) \
892 { \
893 if (c->devinfo->ver >= 41) { \
894 return vir_emit_nondef(c, vir_add_inst(V3D_QPU_A_##name, \
895 dest, \
896 a, c->undef)); \
897 } else { \
898 vir_FMOV_dest(c, vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_##name), a); \
899 return vir_FMOV_dest(c, dest, vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_R4)); \
900 } \
901 }
902
903 #define VIR_A_ALU2(name) VIR_ALU2(name, vir_add_inst, V3D_QPU_A_##name)
904 #define VIR_M_ALU2(name) VIR_ALU2(name, vir_mul_inst, V3D_QPU_M_##name)
905 #define VIR_A_ALU1(name) VIR_ALU1(name, vir_add_inst, V3D_QPU_A_##name)
906 #define VIR_M_ALU1(name) VIR_ALU1(name, vir_mul_inst, V3D_QPU_M_##name)
907 #define VIR_A_ALU0(name) VIR_ALU0(name, vir_add_inst, V3D_QPU_A_##name)
908 #define VIR_M_ALU0(name) VIR_ALU0(name, vir_mul_inst, V3D_QPU_M_##name)
909 #define VIR_A_NODST_2(name) VIR_NODST_2(name, vir_add_inst, V3D_QPU_A_##name)
910 #define VIR_M_NODST_2(name) VIR_NODST_2(name, vir_mul_inst, V3D_QPU_M_##name)
911 #define VIR_A_NODST_1(name) VIR_NODST_1(name, vir_add_inst, V3D_QPU_A_##name)
912 #define VIR_M_NODST_1(name) VIR_NODST_1(name, vir_mul_inst, V3D_QPU_M_##name)
913 #define VIR_A_NODST_0(name) VIR_NODST_0(name, vir_add_inst, V3D_QPU_A_##name)
914
915 VIR_A_ALU2(FADD)
916 VIR_A_ALU2(VFPACK)
917 VIR_A_ALU2(FSUB)
918 VIR_A_ALU2(FMIN)
919 VIR_A_ALU2(FMAX)
920
921 VIR_A_ALU2(ADD)
922 VIR_A_ALU2(SUB)
923 VIR_A_ALU2(SHL)
924 VIR_A_ALU2(SHR)
925 VIR_A_ALU2(ASR)
926 VIR_A_ALU2(ROR)
927 VIR_A_ALU2(MIN)
928 VIR_A_ALU2(MAX)
929 VIR_A_ALU2(UMIN)
930 VIR_A_ALU2(UMAX)
931 VIR_A_ALU2(AND)
932 VIR_A_ALU2(OR)
933 VIR_A_ALU2(XOR)
934 VIR_A_ALU2(VADD)
935 VIR_A_ALU2(VSUB)
936 VIR_A_NODST_2(STVPMV)
937 VIR_A_ALU1(NOT)
938 VIR_A_ALU1(NEG)
939 VIR_A_ALU1(FLAPUSH)
940 VIR_A_ALU1(FLBPUSH)
941 VIR_A_ALU1(FLPOP)
942 VIR_A_ALU1(SETMSF)
943 VIR_A_ALU1(SETREVF)
944 VIR_A_ALU0(TIDX)
945 VIR_A_ALU0(EIDX)
946 VIR_A_ALU1(LDVPMV_IN)
947 VIR_A_ALU1(LDVPMV_OUT)
948 VIR_A_ALU0(TMUWT)
949
950 VIR_A_ALU0(FXCD)
951 VIR_A_ALU0(XCD)
952 VIR_A_ALU0(FYCD)
953 VIR_A_ALU0(YCD)
954 VIR_A_ALU0(MSF)
955 VIR_A_ALU0(REVF)
956 VIR_A_NODST_1(VPMSETUP)
957 VIR_A_NODST_0(VPMWT)
958 VIR_A_ALU2(FCMP)
959 VIR_A_ALU2(VFMAX)
960
961 VIR_A_ALU1(FROUND)
962 VIR_A_ALU1(FTOIN)
963 VIR_A_ALU1(FTRUNC)
964 VIR_A_ALU1(FTOIZ)
965 VIR_A_ALU1(FFLOOR)
966 VIR_A_ALU1(FTOUZ)
967 VIR_A_ALU1(FCEIL)
968 VIR_A_ALU1(FTOC)
969
970 VIR_A_ALU1(FDX)
971 VIR_A_ALU1(FDY)
972
973 VIR_A_ALU1(ITOF)
974 VIR_A_ALU1(CLZ)
975 VIR_A_ALU1(UTOF)
976
977 VIR_M_ALU2(UMUL24)
978 VIR_M_ALU2(FMUL)
979 VIR_M_ALU2(SMUL24)
980 VIR_M_NODST_2(MULTOP)
981
982 VIR_M_ALU1(MOV)
983 VIR_M_ALU1(FMOV)
984
985 VIR_SFU(RECIP)
986 VIR_SFU(RSQRT)
987 VIR_SFU(EXP)
988 VIR_SFU(LOG)
989 VIR_SFU(SIN)
990 VIR_SFU(RSQRT2)
991
992 static inline struct qinst *
993 vir_MOV_cond(struct v3d_compile *c, enum v3d_qpu_cond cond,
994 struct qreg dest, struct qreg src)
995 {
996 struct qinst *mov = vir_MOV_dest(c, dest, src);
997 vir_set_cond(mov, cond);
998 return mov;
999 }
1000
1001 static inline struct qreg
1002 vir_SEL(struct v3d_compile *c, enum v3d_qpu_cond cond,
1003 struct qreg src0, struct qreg src1)
1004 {
1005 struct qreg t = vir_get_temp(c);
1006 vir_MOV_dest(c, t, src1);
1007 vir_MOV_cond(c, cond, t, src0);
1008 return t;
1009 }
1010
1011 static inline struct qinst *
1012 vir_NOP(struct v3d_compile *c)
1013 {
1014 return vir_emit_nondef(c, vir_add_inst(V3D_QPU_A_NOP,
1015 c->undef, c->undef, c->undef));
1016 }
1017
1018 static inline struct qreg
1019 vir_LDTMU(struct v3d_compile *c)
1020 {
1021 if (c->devinfo->ver >= 41) {
1022 struct qinst *ldtmu = vir_add_inst(V3D_QPU_A_NOP, c->undef,
1023 c->undef, c->undef);
1024 ldtmu->qpu.sig.ldtmu = true;
1025
1026 return vir_emit_def(c, ldtmu);
1027 } else {
1028 vir_NOP(c)->qpu.sig.ldtmu = true;
1029 return vir_MOV(c, vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_R4));
1030 }
1031 }
1032
1033 static inline struct qreg
1034 vir_UMUL(struct v3d_compile *c, struct qreg src0, struct qreg src1)
1035 {
1036 vir_MULTOP(c, src0, src1);
1037 return vir_UMUL24(c, src0, src1);
1038 }
1039
1040 /*
1041 static inline struct qreg
1042 vir_LOAD_IMM(struct v3d_compile *c, uint32_t val)
1043 {
1044 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM, c->undef,
1045 vir_reg(QFILE_LOAD_IMM, val), c->undef));
1046 }
1047
1048 static inline struct qreg
1049 vir_LOAD_IMM_U2(struct v3d_compile *c, uint32_t val)
1050 {
1051 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM_U2, c->undef,
1052 vir_reg(QFILE_LOAD_IMM, val),
1053 c->undef));
1054 }
1055 static inline struct qreg
1056 vir_LOAD_IMM_I2(struct v3d_compile *c, uint32_t val)
1057 {
1058 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM_I2, c->undef,
1059 vir_reg(QFILE_LOAD_IMM, val),
1060 c->undef));
1061 }
1062 */
1063
1064 static inline struct qinst *
1065 vir_BRANCH(struct v3d_compile *c, enum v3d_qpu_cond cond)
1066 {
1067 /* The actual uniform_data value will be set at scheduling time */
1068 return vir_emit_nondef(c, vir_branch_inst(cond, vir_uniform_ui(c, 0)));
1069 }
1070
1071 #define vir_for_each_block(block, c) \
1072 list_for_each_entry(struct qblock, block, &c->blocks, link)
1073
1074 #define vir_for_each_block_rev(block, c) \
1075 list_for_each_entry_rev(struct qblock, block, &c->blocks, link)
1076
1077 /* Loop over the non-NULL members of the successors array. */
1078 #define vir_for_each_successor(succ, block) \
1079 for (struct qblock *succ = block->successors[0]; \
1080 succ != NULL; \
1081 succ = (succ == block->successors[1] ? NULL : \
1082 block->successors[1]))
1083
1084 #define vir_for_each_inst(inst, block) \
1085 list_for_each_entry(struct qinst, inst, &block->instructions, link)
1086
1087 #define vir_for_each_inst_rev(inst, block) \
1088 list_for_each_entry_rev(struct qinst, inst, &block->instructions, link)
1089
1090 #define vir_for_each_inst_safe(inst, block) \
1091 list_for_each_entry_safe(struct qinst, inst, &block->instructions, link)
1092
1093 #define vir_for_each_inst_inorder(inst, c) \
1094 vir_for_each_block(_block, c) \
1095 vir_for_each_inst(inst, _block)
1096
1097 #endif /* V3D_COMPILER_H */