broadcom/vc5: Add compiler support for V3D 4.x texturing.
[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_VARY,
77 QFILE_UNIF,
78 QFILE_TLB,
79 QFILE_TLBU,
80
81 /**
82 * VPM reads use this with an index value to say what part of the VPM
83 * is being read.
84 */
85 QFILE_VPM,
86
87 /**
88 * Stores an immediate value in the index field that will be used
89 * directly by qpu_load_imm().
90 */
91 QFILE_LOAD_IMM,
92
93 /**
94 * Stores an immediate value in the index field that can be turned
95 * into a small immediate field by qpu_encode_small_immediate().
96 */
97 QFILE_SMALL_IMM,
98 };
99
100 /**
101 * A reference to a QPU register or a virtual temp register.
102 */
103 struct qreg {
104 enum qfile file;
105 uint32_t index;
106 };
107
108 static inline struct qreg vir_reg(enum qfile file, uint32_t index)
109 {
110 return (struct qreg){file, index};
111 }
112
113 /**
114 * A reference to an actual register at the QPU level, for register
115 * allocation.
116 */
117 struct qpu_reg {
118 bool magic;
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 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_TEXTURE_BORDER_COLOR,
247
248 QUNIFORM_STENCIL,
249
250 QUNIFORM_ALPHA_REF,
251 QUNIFORM_SAMPLE_MASK,
252 };
253
254 struct v3d_varying_slot {
255 uint8_t slot_and_component;
256 };
257
258 static inline struct v3d_varying_slot
259 v3d_slot_from_slot_and_component(uint8_t slot, uint8_t component)
260 {
261 assert(slot < 255 / 4);
262 return (struct v3d_varying_slot){ (slot << 2) + component };
263 }
264
265 static inline uint8_t v3d_slot_get_slot(struct v3d_varying_slot slot)
266 {
267 return slot.slot_and_component >> 2;
268 }
269
270 static inline uint8_t v3d_slot_get_component(struct v3d_varying_slot slot)
271 {
272 return slot.slot_and_component & 3;
273 }
274
275 struct v3d_ubo_range {
276 /**
277 * offset in bytes from the start of the ubo where this range is
278 * uploaded.
279 *
280 * Only set once used is set.
281 */
282 uint32_t dst_offset;
283
284 /**
285 * offset in bytes from the start of the gallium uniforms where the
286 * data comes from.
287 */
288 uint32_t src_offset;
289
290 /** size in bytes of this ubo range */
291 uint32_t size;
292 };
293
294 struct v3d_key {
295 void *shader_state;
296 struct {
297 uint8_t swizzle[4];
298 uint8_t return_size;
299 uint8_t return_channels;
300 union {
301 struct {
302 unsigned compare_mode:1;
303 unsigned compare_func:3;
304 bool clamp_s:1;
305 bool clamp_t:1;
306 bool clamp_r:1;
307 };
308 struct {
309 uint16_t msaa_width, msaa_height;
310 };
311 };
312 } tex[V3D_MAX_TEXTURE_SAMPLERS];
313 uint8_t ucp_enables;
314 };
315
316 struct v3d_fs_key {
317 struct v3d_key base;
318 bool depth_enabled;
319 bool is_points;
320 bool is_lines;
321 bool alpha_test;
322 bool point_coord_upper_left;
323 bool light_twoside;
324 bool msaa;
325 bool sample_coverage;
326 bool sample_alpha_to_coverage;
327 bool sample_alpha_to_one;
328 bool clamp_color;
329 bool shade_model_flat;
330 uint8_t nr_cbufs;
331 uint8_t swap_color_rb;
332 /* Mask of which render targets need to be written as 32-bit floats */
333 uint8_t f32_color_rb;
334 uint8_t alpha_test_func;
335 uint8_t logicop_func;
336 uint32_t point_sprite_mask;
337
338 struct pipe_rt_blend_state blend;
339 };
340
341 struct v3d_vs_key {
342 struct v3d_key base;
343
344 struct v3d_varying_slot fs_inputs[V3D_MAX_FS_INPUTS];
345 uint8_t num_fs_inputs;
346
347 bool is_coord;
348 bool per_vertex_point_size;
349 bool clamp_color;
350 };
351
352 /** A basic block of VIR intructions. */
353 struct qblock {
354 struct list_head link;
355
356 struct list_head instructions;
357
358 struct set *predecessors;
359 struct qblock *successors[2];
360
361 int index;
362
363 /* Instruction IPs for the first and last instruction of the block.
364 * Set by qpu_schedule.c.
365 */
366 uint32_t start_qpu_ip;
367 uint32_t end_qpu_ip;
368
369 /* Instruction IP for the branch instruction of the block. Set by
370 * qpu_schedule.c.
371 */
372 uint32_t branch_qpu_ip;
373
374 /** Offset within the uniform stream at the start of the block. */
375 uint32_t start_uniform;
376 /** Offset within the uniform stream of the branch instruction */
377 uint32_t branch_uniform;
378
379 /** @{ used by v3d_vir_live_variables.c */
380 BITSET_WORD *def;
381 BITSET_WORD *use;
382 BITSET_WORD *live_in;
383 BITSET_WORD *live_out;
384 int start_ip, end_ip;
385 /** @} */
386 };
387
388 /**
389 * Compiler state saved across compiler invocations, for any expensive global
390 * setup.
391 */
392 struct v3d_compiler {
393 const struct v3d_device_info *devinfo;
394 struct ra_regs *regs;
395 unsigned int reg_class_phys[3];
396 unsigned int reg_class_phys_or_acc[3];
397 };
398
399 struct v3d_compile {
400 const struct v3d_device_info *devinfo;
401 nir_shader *s;
402 nir_function_impl *impl;
403 struct exec_list *cf_node_list;
404 const struct v3d_compiler *compiler;
405
406 /**
407 * Mapping from nir_register * or nir_ssa_def * to array of struct
408 * qreg for the values.
409 */
410 struct hash_table *def_ht;
411
412 /* For each temp, the instruction generating its value. */
413 struct qinst **defs;
414 uint32_t defs_array_size;
415
416 /**
417 * Inputs to the shader, arranged by TGSI declaration order.
418 *
419 * Not all fragment shader QFILE_VARY reads are present in this array.
420 */
421 struct qreg *inputs;
422 struct qreg *outputs;
423 bool msaa_per_sample_output;
424 struct qreg color_reads[V3D_MAX_SAMPLES];
425 struct qreg sample_colors[V3D_MAX_SAMPLES];
426 uint32_t inputs_array_size;
427 uint32_t outputs_array_size;
428 uint32_t uniforms_array_size;
429
430 /* Booleans for whether the corresponding QFILE_VARY[i] is
431 * flat-shaded. This includes gl_FragColor flat-shading, which is
432 * customized based on the shademodel_flat shader key.
433 */
434 uint32_t flat_shade_flags[BITSET_WORDS(V3D_MAX_FS_INPUTS)];
435
436 struct v3d_ubo_range *ubo_ranges;
437 bool *ubo_range_used;
438 uint32_t ubo_ranges_array_size;
439 /** Number of uniform areas tracked in ubo_ranges. */
440 uint32_t num_ubo_ranges;
441 uint32_t next_ubo_dst_offset;
442
443 /* State for whether we're executing on each channel currently. 0 if
444 * yes, otherwise a block number + 1 that the channel jumped to.
445 */
446 struct qreg execute;
447
448 struct qreg line_x, point_x, point_y;
449
450 /**
451 * Instance ID, which comes in before the vertex attribute payload if
452 * the shader record requests it.
453 */
454 struct qreg iid;
455
456 /**
457 * Vertex ID, which comes in before the vertex attribute payload
458 * (after Instance ID) if the shader record requests it.
459 */
460 struct qreg vid;
461
462 /* Fragment shader payload regs. */
463 struct qreg payload_w, payload_w_centroid, payload_z;
464
465 uint8_t vattr_sizes[V3D_MAX_VS_INPUTS];
466 uint32_t num_vpm_writes;
467
468 /**
469 * Array of the VARYING_SLOT_* of all FS QFILE_VARY reads.
470 *
471 * This includes those that aren't part of the VPM varyings, like
472 * point/line coordinates.
473 */
474 struct v3d_varying_slot input_slots[V3D_MAX_FS_INPUTS];
475
476 /**
477 * An entry per outputs[] in the VS indicating what the VARYING_SLOT_*
478 * of the output is. Used to emit from the VS in the order that the
479 * FS needs.
480 */
481 struct v3d_varying_slot *output_slots;
482
483 struct pipe_shader_state *shader_state;
484 struct v3d_key *key;
485 struct v3d_fs_key *fs_key;
486 struct v3d_vs_key *vs_key;
487
488 /* Live ranges of temps. */
489 int *temp_start, *temp_end;
490
491 uint32_t *uniform_data;
492 enum quniform_contents *uniform_contents;
493 uint32_t uniform_array_size;
494 uint32_t num_uniforms;
495 uint32_t num_outputs;
496 uint32_t output_position_index;
497 nir_variable *output_color_var[4];
498 uint32_t output_point_size_index;
499 uint32_t output_sample_mask_index;
500
501 struct qreg undef;
502 uint32_t num_temps;
503
504 struct list_head blocks;
505 int next_block_index;
506 struct qblock *cur_block;
507 struct qblock *loop_cont_block;
508 struct qblock *loop_break_block;
509
510 uint64_t *qpu_insts;
511 uint32_t qpu_inst_count;
512 uint32_t qpu_inst_size;
513
514 /* For the FS, the number of varying inputs not counting the
515 * point/line varyings payload
516 */
517 uint32_t num_inputs;
518
519 /**
520 * Number of inputs from num_inputs remaining to be queued to the read
521 * FIFO in the VS/CS.
522 */
523 uint32_t num_inputs_remaining;
524
525 /* Number of inputs currently in the read FIFO for the VS/CS */
526 uint32_t num_inputs_in_fifo;
527
528 /** Next offset in the VPM to read from in the VS/CS */
529 uint32_t vpm_read_offset;
530
531 uint32_t program_id;
532 uint32_t variant_id;
533
534 /* Set to compile program in in 1x, 2x, or 4x threaded mode, where
535 * SIG_THREAD_SWITCH is used to hide texturing latency at the cost of
536 * limiting ourselves to the part of the physical reg space.
537 *
538 * On V3D 3.x, 2x or 4x divide the physical reg space by 2x or 4x. On
539 * V3D 4.x, all shaders are 2x threaded, and 4x only divides the
540 * physical reg space in half.
541 */
542 uint8_t threads;
543 struct qinst *last_thrsw;
544 bool last_thrsw_at_top_level;
545
546 bool failed;
547 };
548
549 struct v3d_uniform_list {
550 enum quniform_contents *contents;
551 uint32_t *data;
552 uint32_t count;
553 };
554
555 struct v3d_prog_data {
556 struct v3d_uniform_list uniforms;
557
558 struct v3d_ubo_range *ubo_ranges;
559 uint32_t num_ubo_ranges;
560 uint32_t ubo_size;
561
562 uint8_t num_inputs;
563 uint8_t threads;
564
565 /* For threads > 1, whether the program should be dispatched in the
566 * after-final-THRSW state.
567 */
568 bool single_seg;
569 };
570
571 struct v3d_vs_prog_data {
572 struct v3d_prog_data base;
573
574 bool uses_iid, uses_vid;
575
576 /* Number of components read from each vertex attribute. */
577 uint8_t vattr_sizes[32];
578
579 /* Total number of components read, for the shader state record. */
580 uint32_t vpm_input_size;
581
582 /* Total number of components written, for the shader state record. */
583 uint32_t vpm_output_size;
584 };
585
586 struct v3d_fs_prog_data {
587 struct v3d_prog_data base;
588
589 struct v3d_varying_slot input_slots[V3D_MAX_FS_INPUTS];
590
591 /* Array of flat shade flags.
592 *
593 * Each entry is only 24 bits (high 8 bits 0), to match the hardware
594 * packet layout.
595 */
596 uint32_t flat_shade_flags[((V3D_MAX_FS_INPUTS - 1) / 24) + 1];
597
598 bool writes_z;
599 bool discard;
600 };
601
602 /* Special nir_load_input intrinsic index for loading the current TLB
603 * destination color.
604 */
605 #define V3D_NIR_TLB_COLOR_READ_INPUT 2000000000
606
607 #define V3D_NIR_MS_MASK_OUTPUT 2000000000
608
609 extern const nir_shader_compiler_options v3d_nir_options;
610
611 const struct v3d_compiler *v3d_compiler_init(const struct v3d_device_info *devinfo);
612 void v3d_compiler_free(const struct v3d_compiler *compiler);
613 void v3d_optimize_nir(struct nir_shader *s);
614
615 uint64_t *v3d_compile_vs(const struct v3d_compiler *compiler,
616 struct v3d_vs_key *key,
617 struct v3d_vs_prog_data *prog_data,
618 nir_shader *s,
619 int program_id, int variant_id,
620 uint32_t *final_assembly_size);
621
622 uint64_t *v3d_compile_fs(const struct v3d_compiler *compiler,
623 struct v3d_fs_key *key,
624 struct v3d_fs_prog_data *prog_data,
625 nir_shader *s,
626 int program_id, int variant_id,
627 uint32_t *final_assembly_size);
628
629 void v3d_nir_to_vir(struct v3d_compile *c);
630
631 void vir_compile_destroy(struct v3d_compile *c);
632 const char *vir_get_stage_name(struct v3d_compile *c);
633 struct qblock *vir_new_block(struct v3d_compile *c);
634 void vir_set_emit_block(struct v3d_compile *c, struct qblock *block);
635 void vir_link_blocks(struct qblock *predecessor, struct qblock *successor);
636 struct qblock *vir_entry_block(struct v3d_compile *c);
637 struct qblock *vir_exit_block(struct v3d_compile *c);
638 struct qinst *vir_add_inst(enum v3d_qpu_add_op op, struct qreg dst,
639 struct qreg src0, struct qreg src1);
640 struct qinst *vir_mul_inst(enum v3d_qpu_mul_op op, struct qreg dst,
641 struct qreg src0, struct qreg src1);
642 struct qinst *vir_branch_inst(enum v3d_qpu_branch_cond cond, struct qreg src0);
643 void vir_remove_instruction(struct v3d_compile *c, struct qinst *qinst);
644 struct qreg vir_uniform(struct v3d_compile *c,
645 enum quniform_contents contents,
646 uint32_t data);
647 void vir_schedule_instructions(struct v3d_compile *c);
648 struct v3d_qpu_instr v3d_qpu_nop(void);
649
650 struct qreg vir_emit_def(struct v3d_compile *c, struct qinst *inst);
651 struct qinst *vir_emit_nondef(struct v3d_compile *c, struct qinst *inst);
652 void vir_set_cond(struct qinst *inst, enum v3d_qpu_cond cond);
653 void vir_set_pf(struct qinst *inst, enum v3d_qpu_pf pf);
654 void vir_set_unpack(struct qinst *inst, int src,
655 enum v3d_qpu_input_unpack unpack);
656
657 struct qreg vir_get_temp(struct v3d_compile *c);
658 void vir_calculate_live_intervals(struct v3d_compile *c);
659 bool vir_has_implicit_uniform(struct qinst *inst);
660 int vir_get_implicit_uniform_src(struct qinst *inst);
661 int vir_get_non_sideband_nsrc(struct qinst *inst);
662 int vir_get_nsrc(struct qinst *inst);
663 bool vir_has_side_effects(struct v3d_compile *c, struct qinst *inst);
664 bool vir_get_add_op(struct qinst *inst, enum v3d_qpu_add_op *op);
665 bool vir_get_mul_op(struct qinst *inst, enum v3d_qpu_mul_op *op);
666 bool vir_is_raw_mov(struct qinst *inst);
667 bool vir_is_tex(struct qinst *inst);
668 bool vir_is_add(struct qinst *inst);
669 bool vir_is_mul(struct qinst *inst);
670 bool vir_is_float_input(struct qinst *inst);
671 bool vir_depends_on_flags(struct qinst *inst);
672 bool vir_writes_r3(const struct v3d_device_info *devinfo, struct qinst *inst);
673 bool vir_writes_r4(const struct v3d_device_info *devinfo, struct qinst *inst);
674 struct qreg vir_follow_movs(struct v3d_compile *c, struct qreg reg);
675 uint8_t vir_channels_written(struct qinst *inst);
676 struct qreg ntq_get_src(struct v3d_compile *c, nir_src src, int i);
677 void ntq_store_dest(struct v3d_compile *c, nir_dest *dest, int chan,
678 struct qreg result);
679 void vir_emit_thrsw(struct v3d_compile *c);
680
681 void vir_dump(struct v3d_compile *c);
682 void vir_dump_inst(struct v3d_compile *c, struct qinst *inst);
683
684 void vir_validate(struct v3d_compile *c);
685
686 void vir_optimize(struct v3d_compile *c);
687 bool vir_opt_algebraic(struct v3d_compile *c);
688 bool vir_opt_constant_folding(struct v3d_compile *c);
689 bool vir_opt_copy_propagate(struct v3d_compile *c);
690 bool vir_opt_dead_code(struct v3d_compile *c);
691 bool vir_opt_peephole_sf(struct v3d_compile *c);
692 bool vir_opt_small_immediates(struct v3d_compile *c);
693 bool vir_opt_vpm(struct v3d_compile *c);
694 void v3d_nir_lower_blend(nir_shader *s, struct v3d_compile *c);
695 void v3d_nir_lower_io(nir_shader *s, struct v3d_compile *c);
696 void v3d_nir_lower_txf_ms(nir_shader *s, struct v3d_compile *c);
697 void vir_lower_uniforms(struct v3d_compile *c);
698
699 void v3d33_vir_vpm_read_setup(struct v3d_compile *c, int num_components);
700 void v3d33_vir_vpm_write_setup(struct v3d_compile *c);
701 void v3d33_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr);
702 void v3d40_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr);
703
704 void v3d_vir_to_qpu(struct v3d_compile *c, struct qpu_reg *temp_registers);
705 uint32_t v3d_qpu_schedule_instructions(struct v3d_compile *c);
706 void qpu_validate(struct v3d_compile *c);
707 struct qpu_reg *v3d_register_allocate(struct v3d_compile *c);
708 bool vir_init_reg_sets(struct v3d_compiler *compiler);
709
710 void vir_PF(struct v3d_compile *c, struct qreg src, enum v3d_qpu_pf pf);
711
712 static inline bool
713 quniform_contents_is_texture_p0(enum quniform_contents contents)
714 {
715 return (contents >= QUNIFORM_TEXTURE_CONFIG_P0_0 &&
716 contents < (QUNIFORM_TEXTURE_CONFIG_P0_0 +
717 V3D_MAX_TEXTURE_SAMPLERS));
718 }
719
720 static inline struct qreg
721 vir_uniform_ui(struct v3d_compile *c, uint32_t ui)
722 {
723 return vir_uniform(c, QUNIFORM_CONSTANT, ui);
724 }
725
726 static inline struct qreg
727 vir_uniform_f(struct v3d_compile *c, float f)
728 {
729 return vir_uniform(c, QUNIFORM_CONSTANT, fui(f));
730 }
731
732 #define VIR_ALU0(name, vir_inst, op) \
733 static inline struct qreg \
734 vir_##name(struct v3d_compile *c) \
735 { \
736 return vir_emit_def(c, vir_inst(op, c->undef, \
737 c->undef, c->undef)); \
738 } \
739 static inline struct qinst * \
740 vir_##name##_dest(struct v3d_compile *c, struct qreg dest) \
741 { \
742 return vir_emit_nondef(c, vir_inst(op, dest, \
743 c->undef, c->undef)); \
744 }
745
746 #define VIR_ALU1(name, vir_inst, op) \
747 static inline struct qreg \
748 vir_##name(struct v3d_compile *c, struct qreg a) \
749 { \
750 return vir_emit_def(c, vir_inst(op, c->undef, \
751 a, c->undef)); \
752 } \
753 static inline struct qinst * \
754 vir_##name##_dest(struct v3d_compile *c, struct qreg dest, \
755 struct qreg a) \
756 { \
757 return vir_emit_nondef(c, vir_inst(op, dest, a, \
758 c->undef)); \
759 }
760
761 #define VIR_ALU2(name, vir_inst, op) \
762 static inline struct qreg \
763 vir_##name(struct v3d_compile *c, struct qreg a, struct qreg b) \
764 { \
765 return vir_emit_def(c, vir_inst(op, c->undef, a, b)); \
766 } \
767 static inline struct qinst * \
768 vir_##name##_dest(struct v3d_compile *c, struct qreg dest, \
769 struct qreg a, struct qreg b) \
770 { \
771 return vir_emit_nondef(c, vir_inst(op, dest, a, b)); \
772 }
773
774 #define VIR_NODST_0(name, vir_inst, op) \
775 static inline struct qinst * \
776 vir_##name(struct v3d_compile *c) \
777 { \
778 return vir_emit_nondef(c, vir_inst(op, c->undef, \
779 c->undef, c->undef)); \
780 }
781
782 #define VIR_NODST_1(name, vir_inst, op) \
783 static inline struct qinst * \
784 vir_##name(struct v3d_compile *c, struct qreg a) \
785 { \
786 return vir_emit_nondef(c, vir_inst(op, c->undef, \
787 a, c->undef)); \
788 }
789
790 #define VIR_NODST_2(name, vir_inst, op) \
791 static inline struct qinst * \
792 vir_##name(struct v3d_compile *c, struct qreg a, struct qreg b) \
793 { \
794 return vir_emit_nondef(c, vir_inst(op, c->undef, \
795 a, b)); \
796 }
797
798 #define VIR_A_ALU2(name) VIR_ALU2(name, vir_add_inst, V3D_QPU_A_##name)
799 #define VIR_M_ALU2(name) VIR_ALU2(name, vir_mul_inst, V3D_QPU_M_##name)
800 #define VIR_A_ALU1(name) VIR_ALU1(name, vir_add_inst, V3D_QPU_A_##name)
801 #define VIR_M_ALU1(name) VIR_ALU1(name, vir_mul_inst, V3D_QPU_M_##name)
802 #define VIR_A_ALU0(name) VIR_ALU0(name, vir_add_inst, V3D_QPU_A_##name)
803 #define VIR_M_ALU0(name) VIR_ALU0(name, vir_mul_inst, V3D_QPU_M_##name)
804 #define VIR_A_NODST_2(name) VIR_NODST_2(name, vir_add_inst, V3D_QPU_A_##name)
805 #define VIR_M_NODST_2(name) VIR_NODST_2(name, vir_mul_inst, V3D_QPU_M_##name)
806 #define VIR_A_NODST_1(name) VIR_NODST_1(name, vir_add_inst, V3D_QPU_A_##name)
807 #define VIR_M_NODST_1(name) VIR_NODST_1(name, vir_mul_inst, V3D_QPU_M_##name)
808 #define VIR_A_NODST_0(name) VIR_NODST_0(name, vir_add_inst, V3D_QPU_A_##name)
809
810 VIR_A_ALU2(FADD)
811 VIR_A_ALU2(VFPACK)
812 VIR_A_ALU2(FSUB)
813 VIR_A_ALU2(FMIN)
814 VIR_A_ALU2(FMAX)
815
816 VIR_A_ALU2(ADD)
817 VIR_A_ALU2(SUB)
818 VIR_A_ALU2(SHL)
819 VIR_A_ALU2(SHR)
820 VIR_A_ALU2(ASR)
821 VIR_A_ALU2(ROR)
822 VIR_A_ALU2(MIN)
823 VIR_A_ALU2(MAX)
824 VIR_A_ALU2(UMIN)
825 VIR_A_ALU2(UMAX)
826 VIR_A_ALU2(AND)
827 VIR_A_ALU2(OR)
828 VIR_A_ALU2(XOR)
829 VIR_A_ALU2(VADD)
830 VIR_A_ALU2(VSUB)
831 VIR_A_ALU2(STVPMV)
832 VIR_A_ALU1(NOT)
833 VIR_A_ALU1(NEG)
834 VIR_A_ALU1(FLAPUSH)
835 VIR_A_ALU1(FLBPUSH)
836 VIR_A_ALU1(FLBPOP)
837 VIR_A_ALU1(SETMSF)
838 VIR_A_ALU1(SETREVF)
839 VIR_A_ALU1(TIDX)
840 VIR_A_ALU1(EIDX)
841 VIR_A_ALU1(LDVPMV_IN)
842 VIR_A_ALU1(LDVPMV_OUT)
843
844 VIR_A_ALU0(FXCD)
845 VIR_A_ALU0(XCD)
846 VIR_A_ALU0(FYCD)
847 VIR_A_ALU0(YCD)
848 VIR_A_ALU0(MSF)
849 VIR_A_ALU0(REVF)
850 VIR_A_NODST_1(VPMSETUP)
851 VIR_A_NODST_0(VPMWT)
852 VIR_A_ALU2(FCMP)
853 VIR_A_ALU2(VFMAX)
854
855 VIR_A_ALU1(FROUND)
856 VIR_A_ALU1(FTOIN)
857 VIR_A_ALU1(FTRUNC)
858 VIR_A_ALU1(FTOIZ)
859 VIR_A_ALU1(FFLOOR)
860 VIR_A_ALU1(FTOUZ)
861 VIR_A_ALU1(FCEIL)
862 VIR_A_ALU1(FTOC)
863
864 VIR_A_ALU1(FDX)
865 VIR_A_ALU1(FDY)
866
867 VIR_A_ALU1(ITOF)
868 VIR_A_ALU1(CLZ)
869 VIR_A_ALU1(UTOF)
870
871 VIR_M_ALU2(UMUL24)
872 VIR_M_ALU2(FMUL)
873 VIR_M_ALU2(SMUL24)
874 VIR_M_NODST_2(MULTOP)
875
876 VIR_M_ALU1(MOV)
877 VIR_M_ALU1(FMOV)
878
879 static inline struct qinst *
880 vir_MOV_cond(struct v3d_compile *c, enum v3d_qpu_cond cond,
881 struct qreg dest, struct qreg src)
882 {
883 struct qinst *mov = vir_MOV_dest(c, dest, src);
884 vir_set_cond(mov, cond);
885 return mov;
886 }
887
888 static inline struct qreg
889 vir_SEL(struct v3d_compile *c, enum v3d_qpu_cond cond,
890 struct qreg src0, struct qreg src1)
891 {
892 struct qreg t = vir_get_temp(c);
893 vir_MOV_dest(c, t, src1);
894 vir_MOV_cond(c, cond, t, src0);
895 return t;
896 }
897
898 static inline struct qinst *
899 vir_NOP(struct v3d_compile *c)
900 {
901 return vir_emit_nondef(c, vir_add_inst(V3D_QPU_A_NOP,
902 c->undef, c->undef, c->undef));
903 }
904
905 static inline struct qreg
906 vir_LDTMU(struct v3d_compile *c)
907 {
908 if (c->devinfo->ver >= 41) {
909 struct qinst *ldtmu = vir_add_inst(V3D_QPU_A_NOP, c->undef,
910 c->undef, c->undef);
911 ldtmu->qpu.sig.ldtmu = true;
912
913 return vir_emit_def(c, ldtmu);
914 } else {
915 vir_NOP(c)->qpu.sig.ldtmu = true;
916 return vir_MOV(c, vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_R4));
917 }
918 }
919
920 /*
921 static inline struct qreg
922 vir_LOAD_IMM(struct v3d_compile *c, uint32_t val)
923 {
924 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM, c->undef,
925 vir_reg(QFILE_LOAD_IMM, val), c->undef));
926 }
927
928 static inline struct qreg
929 vir_LOAD_IMM_U2(struct v3d_compile *c, uint32_t val)
930 {
931 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM_U2, c->undef,
932 vir_reg(QFILE_LOAD_IMM, val),
933 c->undef));
934 }
935 static inline struct qreg
936 vir_LOAD_IMM_I2(struct v3d_compile *c, uint32_t val)
937 {
938 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM_I2, c->undef,
939 vir_reg(QFILE_LOAD_IMM, val),
940 c->undef));
941 }
942 */
943
944 static inline struct qinst *
945 vir_BRANCH(struct v3d_compile *c, enum v3d_qpu_cond cond)
946 {
947 /* The actual uniform_data value will be set at scheduling time */
948 return vir_emit_nondef(c, vir_branch_inst(cond, vir_uniform_ui(c, 0)));
949 }
950
951 #define vir_for_each_block(block, c) \
952 list_for_each_entry(struct qblock, block, &c->blocks, link)
953
954 #define vir_for_each_block_rev(block, c) \
955 list_for_each_entry_rev(struct qblock, block, &c->blocks, link)
956
957 /* Loop over the non-NULL members of the successors array. */
958 #define vir_for_each_successor(succ, block) \
959 for (struct qblock *succ = block->successors[0]; \
960 succ != NULL; \
961 succ = (succ == block->successors[1] ? NULL : \
962 block->successors[1]))
963
964 #define vir_for_each_inst(inst, block) \
965 list_for_each_entry(struct qinst, inst, &block->instructions, link)
966
967 #define vir_for_each_inst_rev(inst, block) \
968 list_for_each_entry_rev(struct qinst, inst, &block->instructions, link)
969
970 #define vir_for_each_inst_safe(inst, block) \
971 list_for_each_entry_safe(struct qinst, inst, &block->instructions, link)
972
973 #define vir_for_each_inst_inorder(inst, c) \
974 vir_for_each_block(_block, c) \
975 vir_for_each_inst(inst, _block)
976
977 #endif /* V3D_COMPILER_H */