broadcom/vc5: Add cursors to the compiler infrastructure, like NIR's.
[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 int index;
119 };
120
121 struct qinst {
122 /** Entry in qblock->instructions */
123 struct list_head link;
124
125 /**
126 * The instruction being wrapped. Its condition codes, pack flags,
127 * signals, etc. will all be used, with just the register references
128 * being replaced by the contents of qinst->dst and qinst->src[].
129 */
130 struct v3d_qpu_instr qpu;
131
132 /* Pre-register-allocation references to src/dst registers */
133 struct qreg dst;
134 struct qreg src[3];
135 bool cond_is_exec_mask;
136 bool has_implicit_uniform;
137 bool is_last_thrsw;
138
139 /* After vir_to_qpu.c: If instr reads a uniform, which uniform from
140 * the uncompiled stream it is.
141 */
142 int uniform;
143 };
144
145 enum quniform_contents {
146 /**
147 * Indicates that a constant 32-bit value is copied from the program's
148 * uniform contents.
149 */
150 QUNIFORM_CONSTANT,
151 /**
152 * Indicates that the program's uniform contents are used as an index
153 * into the GL uniform storage.
154 */
155 QUNIFORM_UNIFORM,
156
157 /** @{
158 * Scaling factors from clip coordinates to relative to the viewport
159 * center.
160 *
161 * This is used by the coordinate and vertex shaders to produce the
162 * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed
163 * point offsets from the viewport ccenter.
164 */
165 QUNIFORM_VIEWPORT_X_SCALE,
166 QUNIFORM_VIEWPORT_Y_SCALE,
167 /** @} */
168
169 QUNIFORM_VIEWPORT_Z_OFFSET,
170 QUNIFORM_VIEWPORT_Z_SCALE,
171
172 QUNIFORM_USER_CLIP_PLANE,
173
174 /**
175 * A reference to a V3D 3.x texture config parameter 0 uniform.
176 *
177 * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
178 * defines texture type, miplevels, and such. It will be found as a
179 * parameter to the first QOP_TEX_[STRB] instruction in a sequence.
180 */
181 QUNIFORM_TEXTURE_CONFIG_P0_0,
182 QUNIFORM_TEXTURE_CONFIG_P0_1,
183 QUNIFORM_TEXTURE_CONFIG_P0_2,
184 QUNIFORM_TEXTURE_CONFIG_P0_3,
185 QUNIFORM_TEXTURE_CONFIG_P0_4,
186 QUNIFORM_TEXTURE_CONFIG_P0_5,
187 QUNIFORM_TEXTURE_CONFIG_P0_6,
188 QUNIFORM_TEXTURE_CONFIG_P0_7,
189 QUNIFORM_TEXTURE_CONFIG_P0_8,
190 QUNIFORM_TEXTURE_CONFIG_P0_9,
191 QUNIFORM_TEXTURE_CONFIG_P0_10,
192 QUNIFORM_TEXTURE_CONFIG_P0_11,
193 QUNIFORM_TEXTURE_CONFIG_P0_12,
194 QUNIFORM_TEXTURE_CONFIG_P0_13,
195 QUNIFORM_TEXTURE_CONFIG_P0_14,
196 QUNIFORM_TEXTURE_CONFIG_P0_15,
197 QUNIFORM_TEXTURE_CONFIG_P0_16,
198 QUNIFORM_TEXTURE_CONFIG_P0_17,
199 QUNIFORM_TEXTURE_CONFIG_P0_18,
200 QUNIFORM_TEXTURE_CONFIG_P0_19,
201 QUNIFORM_TEXTURE_CONFIG_P0_20,
202 QUNIFORM_TEXTURE_CONFIG_P0_21,
203 QUNIFORM_TEXTURE_CONFIG_P0_22,
204 QUNIFORM_TEXTURE_CONFIG_P0_23,
205 QUNIFORM_TEXTURE_CONFIG_P0_24,
206 QUNIFORM_TEXTURE_CONFIG_P0_25,
207 QUNIFORM_TEXTURE_CONFIG_P0_26,
208 QUNIFORM_TEXTURE_CONFIG_P0_27,
209 QUNIFORM_TEXTURE_CONFIG_P0_28,
210 QUNIFORM_TEXTURE_CONFIG_P0_29,
211 QUNIFORM_TEXTURE_CONFIG_P0_30,
212 QUNIFORM_TEXTURE_CONFIG_P0_31,
213 QUNIFORM_TEXTURE_CONFIG_P0_32,
214
215 /**
216 * A reference to a V3D 3.x texture config parameter 1 uniform.
217 *
218 * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
219 * has the pointer to the indirect texture state. Our data[] field
220 * will have a packed p1 value, but the address field will be just
221 * which texture unit's texture should be referenced.
222 */
223 QUNIFORM_TEXTURE_CONFIG_P1,
224
225 /* A a V3D 4.x texture config parameter. The high 8 bits will be
226 * which texture or sampler is being sampled, and the driver must
227 * replace the address field with the appropriate address.
228 */
229 QUNIFORM_TMU_CONFIG_P0,
230 QUNIFORM_TMU_CONFIG_P1,
231
232 QUNIFORM_TEXTURE_FIRST_LEVEL,
233
234 QUNIFORM_TEXTURE_WIDTH,
235 QUNIFORM_TEXTURE_HEIGHT,
236 QUNIFORM_TEXTURE_DEPTH,
237 QUNIFORM_TEXTURE_ARRAY_SIZE,
238 QUNIFORM_TEXTURE_LEVELS,
239
240 QUNIFORM_UBO_ADDR,
241
242 QUNIFORM_TEXRECT_SCALE_X,
243 QUNIFORM_TEXRECT_SCALE_Y,
244
245 QUNIFORM_TEXTURE_BORDER_COLOR,
246
247 QUNIFORM_STENCIL,
248
249 QUNIFORM_ALPHA_REF,
250 QUNIFORM_SAMPLE_MASK,
251 };
252
253 struct v3d_varying_slot {
254 uint8_t slot_and_component;
255 };
256
257 static inline struct v3d_varying_slot
258 v3d_slot_from_slot_and_component(uint8_t slot, uint8_t component)
259 {
260 assert(slot < 255 / 4);
261 return (struct v3d_varying_slot){ (slot << 2) + component };
262 }
263
264 static inline uint8_t v3d_slot_get_slot(struct v3d_varying_slot slot)
265 {
266 return slot.slot_and_component >> 2;
267 }
268
269 static inline uint8_t v3d_slot_get_component(struct v3d_varying_slot slot)
270 {
271 return slot.slot_and_component & 3;
272 }
273
274 struct v3d_ubo_range {
275 /**
276 * offset in bytes from the start of the ubo where this range is
277 * uploaded.
278 *
279 * Only set once used is set.
280 */
281 uint32_t dst_offset;
282
283 /**
284 * offset in bytes from the start of the gallium uniforms where the
285 * data comes from.
286 */
287 uint32_t src_offset;
288
289 /** size in bytes of this ubo range */
290 uint32_t size;
291 };
292
293 struct v3d_key {
294 void *shader_state;
295 struct {
296 uint8_t swizzle[4];
297 uint8_t return_size;
298 uint8_t return_channels;
299 union {
300 struct {
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 };
307 struct {
308 uint16_t msaa_width, msaa_height;
309 };
310 };
311 } tex[V3D_MAX_TEXTURE_SAMPLERS];
312 uint8_t ucp_enables;
313 };
314
315 struct v3d_fs_key {
316 struct v3d_key base;
317 bool depth_enabled;
318 bool is_points;
319 bool is_lines;
320 bool alpha_test;
321 bool point_coord_upper_left;
322 bool light_twoside;
323 bool msaa;
324 bool sample_coverage;
325 bool sample_alpha_to_coverage;
326 bool sample_alpha_to_one;
327 bool clamp_color;
328 bool shade_model_flat;
329 uint8_t nr_cbufs;
330 uint8_t swap_color_rb;
331 /* Mask of which render targets need to be written as 32-bit floats */
332 uint8_t f32_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 struct v3d_ubo_range *ubo_ranges;
478 bool *ubo_range_used;
479 uint32_t ubo_ranges_array_size;
480 /** Number of uniform areas tracked in ubo_ranges. */
481 uint32_t num_ubo_ranges;
482 uint32_t next_ubo_dst_offset;
483
484 /* State for whether we're executing on each channel currently. 0 if
485 * yes, otherwise a block number + 1 that the channel jumped to.
486 */
487 struct qreg execute;
488
489 struct qreg line_x, point_x, point_y;
490
491 /**
492 * Instance ID, which comes in before the vertex attribute payload if
493 * the shader record requests it.
494 */
495 struct qreg iid;
496
497 /**
498 * Vertex ID, which comes in before the vertex attribute payload
499 * (after Instance ID) if the shader record requests it.
500 */
501 struct qreg vid;
502
503 /* Fragment shader payload regs. */
504 struct qreg payload_w, payload_w_centroid, payload_z;
505
506 uint8_t vattr_sizes[V3D_MAX_VS_INPUTS];
507 uint32_t num_vpm_writes;
508
509 /**
510 * Array of the VARYING_SLOT_* of all FS QFILE_VARY reads.
511 *
512 * This includes those that aren't part of the VPM varyings, like
513 * point/line coordinates.
514 */
515 struct v3d_varying_slot input_slots[V3D_MAX_FS_INPUTS];
516
517 /**
518 * An entry per outputs[] in the VS indicating what the VARYING_SLOT_*
519 * of the output is. Used to emit from the VS in the order that the
520 * FS needs.
521 */
522 struct v3d_varying_slot *output_slots;
523
524 struct pipe_shader_state *shader_state;
525 struct v3d_key *key;
526 struct v3d_fs_key *fs_key;
527 struct v3d_vs_key *vs_key;
528
529 /* Live ranges of temps. */
530 int *temp_start, *temp_end;
531
532 uint32_t *uniform_data;
533 enum quniform_contents *uniform_contents;
534 uint32_t uniform_array_size;
535 uint32_t num_uniforms;
536 uint32_t num_outputs;
537 uint32_t output_position_index;
538 nir_variable *output_color_var[4];
539 uint32_t output_point_size_index;
540 uint32_t output_sample_mask_index;
541
542 struct qreg undef;
543 uint32_t num_temps;
544
545 struct vir_cursor cursor;
546 struct list_head blocks;
547 int next_block_index;
548 struct qblock *cur_block;
549 struct qblock *loop_cont_block;
550 struct qblock *loop_break_block;
551
552 uint64_t *qpu_insts;
553 uint32_t qpu_inst_count;
554 uint32_t qpu_inst_size;
555
556 /* For the FS, the number of varying inputs not counting the
557 * point/line varyings payload
558 */
559 uint32_t num_inputs;
560
561 /**
562 * Number of inputs from num_inputs remaining to be queued to the read
563 * FIFO in the VS/CS.
564 */
565 uint32_t num_inputs_remaining;
566
567 /* Number of inputs currently in the read FIFO for the VS/CS */
568 uint32_t num_inputs_in_fifo;
569
570 /** Next offset in the VPM to read from in the VS/CS */
571 uint32_t vpm_read_offset;
572
573 uint32_t program_id;
574 uint32_t variant_id;
575
576 /* Set to compile program in in 1x, 2x, or 4x threaded mode, where
577 * SIG_THREAD_SWITCH is used to hide texturing latency at the cost of
578 * limiting ourselves to the part of the physical reg space.
579 *
580 * On V3D 3.x, 2x or 4x divide the physical reg space by 2x or 4x. On
581 * V3D 4.x, all shaders are 2x threaded, and 4x only divides the
582 * physical reg space in half.
583 */
584 uint8_t threads;
585 struct qinst *last_thrsw;
586 bool last_thrsw_at_top_level;
587
588 bool failed;
589 };
590
591 struct v3d_uniform_list {
592 enum quniform_contents *contents;
593 uint32_t *data;
594 uint32_t count;
595 };
596
597 struct v3d_prog_data {
598 struct v3d_uniform_list uniforms;
599
600 struct v3d_ubo_range *ubo_ranges;
601 uint32_t num_ubo_ranges;
602 uint32_t ubo_size;
603
604 uint8_t num_inputs;
605 uint8_t threads;
606
607 /* For threads > 1, whether the program should be dispatched in the
608 * after-final-THRSW state.
609 */
610 bool single_seg;
611 };
612
613 struct v3d_vs_prog_data {
614 struct v3d_prog_data base;
615
616 bool uses_iid, uses_vid;
617
618 /* Number of components read from each vertex attribute. */
619 uint8_t vattr_sizes[32];
620
621 /* Total number of components read, for the shader state record. */
622 uint32_t vpm_input_size;
623
624 /* Total number of components written, for the shader state record. */
625 uint32_t vpm_output_size;
626 };
627
628 struct v3d_fs_prog_data {
629 struct v3d_prog_data base;
630
631 struct v3d_varying_slot input_slots[V3D_MAX_FS_INPUTS];
632
633 /* Array of flat shade flags.
634 *
635 * Each entry is only 24 bits (high 8 bits 0), to match the hardware
636 * packet layout.
637 */
638 uint32_t flat_shade_flags[((V3D_MAX_FS_INPUTS - 1) / 24) + 1];
639
640 bool writes_z;
641 bool discard;
642 };
643
644 /* Special nir_load_input intrinsic index for loading the current TLB
645 * destination color.
646 */
647 #define V3D_NIR_TLB_COLOR_READ_INPUT 2000000000
648
649 #define V3D_NIR_MS_MASK_OUTPUT 2000000000
650
651 extern const nir_shader_compiler_options v3d_nir_options;
652
653 const struct v3d_compiler *v3d_compiler_init(const struct v3d_device_info *devinfo);
654 void v3d_compiler_free(const struct v3d_compiler *compiler);
655 void v3d_optimize_nir(struct nir_shader *s);
656
657 uint64_t *v3d_compile_vs(const struct v3d_compiler *compiler,
658 struct v3d_vs_key *key,
659 struct v3d_vs_prog_data *prog_data,
660 nir_shader *s,
661 int program_id, int variant_id,
662 uint32_t *final_assembly_size);
663
664 uint64_t *v3d_compile_fs(const struct v3d_compiler *compiler,
665 struct v3d_fs_key *key,
666 struct v3d_fs_prog_data *prog_data,
667 nir_shader *s,
668 int program_id, int variant_id,
669 uint32_t *final_assembly_size);
670
671 void v3d_nir_to_vir(struct v3d_compile *c);
672
673 void vir_compile_destroy(struct v3d_compile *c);
674 const char *vir_get_stage_name(struct v3d_compile *c);
675 struct qblock *vir_new_block(struct v3d_compile *c);
676 void vir_set_emit_block(struct v3d_compile *c, struct qblock *block);
677 void vir_link_blocks(struct qblock *predecessor, struct qblock *successor);
678 struct qblock *vir_entry_block(struct v3d_compile *c);
679 struct qblock *vir_exit_block(struct v3d_compile *c);
680 struct qinst *vir_add_inst(enum v3d_qpu_add_op op, struct qreg dst,
681 struct qreg src0, struct qreg src1);
682 struct qinst *vir_mul_inst(enum v3d_qpu_mul_op op, struct qreg dst,
683 struct qreg src0, struct qreg src1);
684 struct qinst *vir_branch_inst(enum v3d_qpu_branch_cond cond, struct qreg src0);
685 void vir_remove_instruction(struct v3d_compile *c, struct qinst *qinst);
686 struct qreg vir_uniform(struct v3d_compile *c,
687 enum quniform_contents contents,
688 uint32_t data);
689 void vir_schedule_instructions(struct v3d_compile *c);
690 struct v3d_qpu_instr v3d_qpu_nop(void);
691
692 struct qreg vir_emit_def(struct v3d_compile *c, struct qinst *inst);
693 struct qinst *vir_emit_nondef(struct v3d_compile *c, struct qinst *inst);
694 void vir_set_cond(struct qinst *inst, enum v3d_qpu_cond cond);
695 void vir_set_pf(struct qinst *inst, enum v3d_qpu_pf pf);
696 void vir_set_unpack(struct qinst *inst, int src,
697 enum v3d_qpu_input_unpack unpack);
698
699 struct qreg vir_get_temp(struct v3d_compile *c);
700 void vir_calculate_live_intervals(struct v3d_compile *c);
701 bool vir_has_implicit_uniform(struct qinst *inst);
702 int vir_get_implicit_uniform_src(struct qinst *inst);
703 int vir_get_non_sideband_nsrc(struct qinst *inst);
704 int vir_get_nsrc(struct qinst *inst);
705 bool vir_has_side_effects(struct v3d_compile *c, struct qinst *inst);
706 bool vir_get_add_op(struct qinst *inst, enum v3d_qpu_add_op *op);
707 bool vir_get_mul_op(struct qinst *inst, enum v3d_qpu_mul_op *op);
708 bool vir_is_raw_mov(struct qinst *inst);
709 bool vir_is_tex(struct qinst *inst);
710 bool vir_is_add(struct qinst *inst);
711 bool vir_is_mul(struct qinst *inst);
712 bool vir_is_float_input(struct qinst *inst);
713 bool vir_depends_on_flags(struct qinst *inst);
714 bool vir_writes_r3(const struct v3d_device_info *devinfo, struct qinst *inst);
715 bool vir_writes_r4(const struct v3d_device_info *devinfo, struct qinst *inst);
716 struct qreg vir_follow_movs(struct v3d_compile *c, struct qreg reg);
717 uint8_t vir_channels_written(struct qinst *inst);
718 struct qreg ntq_get_src(struct v3d_compile *c, nir_src src, int i);
719 void ntq_store_dest(struct v3d_compile *c, nir_dest *dest, int chan,
720 struct qreg result);
721 void vir_emit_thrsw(struct v3d_compile *c);
722
723 void vir_dump(struct v3d_compile *c);
724 void vir_dump_inst(struct v3d_compile *c, struct qinst *inst);
725
726 void vir_validate(struct v3d_compile *c);
727
728 void vir_optimize(struct v3d_compile *c);
729 bool vir_opt_algebraic(struct v3d_compile *c);
730 bool vir_opt_constant_folding(struct v3d_compile *c);
731 bool vir_opt_copy_propagate(struct v3d_compile *c);
732 bool vir_opt_dead_code(struct v3d_compile *c);
733 bool vir_opt_peephole_sf(struct v3d_compile *c);
734 bool vir_opt_small_immediates(struct v3d_compile *c);
735 bool vir_opt_vpm(struct v3d_compile *c);
736 void v3d_nir_lower_blend(nir_shader *s, struct v3d_compile *c);
737 void v3d_nir_lower_io(nir_shader *s, struct v3d_compile *c);
738 void v3d_nir_lower_txf_ms(nir_shader *s, struct v3d_compile *c);
739 void vir_lower_uniforms(struct v3d_compile *c);
740
741 void v3d33_vir_vpm_read_setup(struct v3d_compile *c, int num_components);
742 void v3d33_vir_vpm_write_setup(struct v3d_compile *c);
743 void v3d33_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr);
744 void v3d40_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr);
745
746 void v3d_vir_to_qpu(struct v3d_compile *c, struct qpu_reg *temp_registers);
747 uint32_t v3d_qpu_schedule_instructions(struct v3d_compile *c);
748 void qpu_validate(struct v3d_compile *c);
749 struct qpu_reg *v3d_register_allocate(struct v3d_compile *c);
750 bool vir_init_reg_sets(struct v3d_compiler *compiler);
751
752 void vir_PF(struct v3d_compile *c, struct qreg src, enum v3d_qpu_pf pf);
753
754 static inline bool
755 quniform_contents_is_texture_p0(enum quniform_contents contents)
756 {
757 return (contents >= QUNIFORM_TEXTURE_CONFIG_P0_0 &&
758 contents < (QUNIFORM_TEXTURE_CONFIG_P0_0 +
759 V3D_MAX_TEXTURE_SAMPLERS));
760 }
761
762 static inline struct qreg
763 vir_uniform_ui(struct v3d_compile *c, uint32_t ui)
764 {
765 return vir_uniform(c, QUNIFORM_CONSTANT, ui);
766 }
767
768 static inline struct qreg
769 vir_uniform_f(struct v3d_compile *c, float f)
770 {
771 return vir_uniform(c, QUNIFORM_CONSTANT, fui(f));
772 }
773
774 #define VIR_ALU0(name, vir_inst, op) \
775 static inline struct qreg \
776 vir_##name(struct v3d_compile *c) \
777 { \
778 return vir_emit_def(c, vir_inst(op, c->undef, \
779 c->undef, c->undef)); \
780 } \
781 static inline struct qinst * \
782 vir_##name##_dest(struct v3d_compile *c, struct qreg dest) \
783 { \
784 return vir_emit_nondef(c, vir_inst(op, dest, \
785 c->undef, c->undef)); \
786 }
787
788 #define VIR_ALU1(name, vir_inst, op) \
789 static inline struct qreg \
790 vir_##name(struct v3d_compile *c, struct qreg a) \
791 { \
792 return vir_emit_def(c, vir_inst(op, c->undef, \
793 a, c->undef)); \
794 } \
795 static inline struct qinst * \
796 vir_##name##_dest(struct v3d_compile *c, struct qreg dest, \
797 struct qreg a) \
798 { \
799 return vir_emit_nondef(c, vir_inst(op, dest, a, \
800 c->undef)); \
801 }
802
803 #define VIR_ALU2(name, vir_inst, op) \
804 static inline struct qreg \
805 vir_##name(struct v3d_compile *c, struct qreg a, struct qreg b) \
806 { \
807 return vir_emit_def(c, vir_inst(op, c->undef, a, b)); \
808 } \
809 static inline struct qinst * \
810 vir_##name##_dest(struct v3d_compile *c, struct qreg dest, \
811 struct qreg a, struct qreg b) \
812 { \
813 return vir_emit_nondef(c, vir_inst(op, dest, a, b)); \
814 }
815
816 #define VIR_NODST_0(name, vir_inst, op) \
817 static inline struct qinst * \
818 vir_##name(struct v3d_compile *c) \
819 { \
820 return vir_emit_nondef(c, vir_inst(op, c->undef, \
821 c->undef, c->undef)); \
822 }
823
824 #define VIR_NODST_1(name, vir_inst, op) \
825 static inline struct qinst * \
826 vir_##name(struct v3d_compile *c, struct qreg a) \
827 { \
828 return vir_emit_nondef(c, vir_inst(op, c->undef, \
829 a, c->undef)); \
830 }
831
832 #define VIR_NODST_2(name, vir_inst, op) \
833 static inline struct qinst * \
834 vir_##name(struct v3d_compile *c, struct qreg a, struct qreg b) \
835 { \
836 return vir_emit_nondef(c, vir_inst(op, c->undef, \
837 a, b)); \
838 }
839
840 #define VIR_A_ALU2(name) VIR_ALU2(name, vir_add_inst, V3D_QPU_A_##name)
841 #define VIR_M_ALU2(name) VIR_ALU2(name, vir_mul_inst, V3D_QPU_M_##name)
842 #define VIR_A_ALU1(name) VIR_ALU1(name, vir_add_inst, V3D_QPU_A_##name)
843 #define VIR_M_ALU1(name) VIR_ALU1(name, vir_mul_inst, V3D_QPU_M_##name)
844 #define VIR_A_ALU0(name) VIR_ALU0(name, vir_add_inst, V3D_QPU_A_##name)
845 #define VIR_M_ALU0(name) VIR_ALU0(name, vir_mul_inst, V3D_QPU_M_##name)
846 #define VIR_A_NODST_2(name) VIR_NODST_2(name, vir_add_inst, V3D_QPU_A_##name)
847 #define VIR_M_NODST_2(name) VIR_NODST_2(name, vir_mul_inst, V3D_QPU_M_##name)
848 #define VIR_A_NODST_1(name) VIR_NODST_1(name, vir_add_inst, V3D_QPU_A_##name)
849 #define VIR_M_NODST_1(name) VIR_NODST_1(name, vir_mul_inst, V3D_QPU_M_##name)
850 #define VIR_A_NODST_0(name) VIR_NODST_0(name, vir_add_inst, V3D_QPU_A_##name)
851
852 VIR_A_ALU2(FADD)
853 VIR_A_ALU2(VFPACK)
854 VIR_A_ALU2(FSUB)
855 VIR_A_ALU2(FMIN)
856 VIR_A_ALU2(FMAX)
857
858 VIR_A_ALU2(ADD)
859 VIR_A_ALU2(SUB)
860 VIR_A_ALU2(SHL)
861 VIR_A_ALU2(SHR)
862 VIR_A_ALU2(ASR)
863 VIR_A_ALU2(ROR)
864 VIR_A_ALU2(MIN)
865 VIR_A_ALU2(MAX)
866 VIR_A_ALU2(UMIN)
867 VIR_A_ALU2(UMAX)
868 VIR_A_ALU2(AND)
869 VIR_A_ALU2(OR)
870 VIR_A_ALU2(XOR)
871 VIR_A_ALU2(VADD)
872 VIR_A_ALU2(VSUB)
873 VIR_A_ALU2(STVPMV)
874 VIR_A_ALU1(NOT)
875 VIR_A_ALU1(NEG)
876 VIR_A_ALU1(FLAPUSH)
877 VIR_A_ALU1(FLBPUSH)
878 VIR_A_ALU1(FLBPOP)
879 VIR_A_ALU1(SETMSF)
880 VIR_A_ALU1(SETREVF)
881 VIR_A_ALU0(TIDX)
882 VIR_A_ALU0(EIDX)
883 VIR_A_ALU1(LDVPMV_IN)
884 VIR_A_ALU1(LDVPMV_OUT)
885
886 VIR_A_ALU0(FXCD)
887 VIR_A_ALU0(XCD)
888 VIR_A_ALU0(FYCD)
889 VIR_A_ALU0(YCD)
890 VIR_A_ALU0(MSF)
891 VIR_A_ALU0(REVF)
892 VIR_A_NODST_1(VPMSETUP)
893 VIR_A_NODST_0(VPMWT)
894 VIR_A_ALU2(FCMP)
895 VIR_A_ALU2(VFMAX)
896
897 VIR_A_ALU1(FROUND)
898 VIR_A_ALU1(FTOIN)
899 VIR_A_ALU1(FTRUNC)
900 VIR_A_ALU1(FTOIZ)
901 VIR_A_ALU1(FFLOOR)
902 VIR_A_ALU1(FTOUZ)
903 VIR_A_ALU1(FCEIL)
904 VIR_A_ALU1(FTOC)
905
906 VIR_A_ALU1(FDX)
907 VIR_A_ALU1(FDY)
908
909 VIR_A_ALU1(ITOF)
910 VIR_A_ALU1(CLZ)
911 VIR_A_ALU1(UTOF)
912
913 VIR_M_ALU2(UMUL24)
914 VIR_M_ALU2(FMUL)
915 VIR_M_ALU2(SMUL24)
916 VIR_M_NODST_2(MULTOP)
917
918 VIR_M_ALU1(MOV)
919 VIR_M_ALU1(FMOV)
920
921 static inline struct qinst *
922 vir_MOV_cond(struct v3d_compile *c, enum v3d_qpu_cond cond,
923 struct qreg dest, struct qreg src)
924 {
925 struct qinst *mov = vir_MOV_dest(c, dest, src);
926 vir_set_cond(mov, cond);
927 return mov;
928 }
929
930 static inline struct qreg
931 vir_SEL(struct v3d_compile *c, enum v3d_qpu_cond cond,
932 struct qreg src0, struct qreg src1)
933 {
934 struct qreg t = vir_get_temp(c);
935 vir_MOV_dest(c, t, src1);
936 vir_MOV_cond(c, cond, t, src0);
937 return t;
938 }
939
940 static inline struct qinst *
941 vir_NOP(struct v3d_compile *c)
942 {
943 return vir_emit_nondef(c, vir_add_inst(V3D_QPU_A_NOP,
944 c->undef, c->undef, c->undef));
945 }
946
947 static inline struct qreg
948 vir_LDTMU(struct v3d_compile *c)
949 {
950 if (c->devinfo->ver >= 41) {
951 struct qinst *ldtmu = vir_add_inst(V3D_QPU_A_NOP, c->undef,
952 c->undef, c->undef);
953 ldtmu->qpu.sig.ldtmu = true;
954
955 return vir_emit_def(c, ldtmu);
956 } else {
957 vir_NOP(c)->qpu.sig.ldtmu = true;
958 return vir_MOV(c, vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_R4));
959 }
960 }
961
962 static inline struct qreg
963 vir_UMUL(struct v3d_compile *c, struct qreg src0, struct qreg src1)
964 {
965 vir_MULTOP(c, src0, src1);
966 return vir_UMUL24(c, src0, src1);
967 }
968
969 /*
970 static inline struct qreg
971 vir_LOAD_IMM(struct v3d_compile *c, uint32_t val)
972 {
973 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM, c->undef,
974 vir_reg(QFILE_LOAD_IMM, val), c->undef));
975 }
976
977 static inline struct qreg
978 vir_LOAD_IMM_U2(struct v3d_compile *c, uint32_t val)
979 {
980 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM_U2, c->undef,
981 vir_reg(QFILE_LOAD_IMM, val),
982 c->undef));
983 }
984 static inline struct qreg
985 vir_LOAD_IMM_I2(struct v3d_compile *c, uint32_t val)
986 {
987 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM_I2, c->undef,
988 vir_reg(QFILE_LOAD_IMM, val),
989 c->undef));
990 }
991 */
992
993 static inline struct qinst *
994 vir_BRANCH(struct v3d_compile *c, enum v3d_qpu_cond cond)
995 {
996 /* The actual uniform_data value will be set at scheduling time */
997 return vir_emit_nondef(c, vir_branch_inst(cond, vir_uniform_ui(c, 0)));
998 }
999
1000 #define vir_for_each_block(block, c) \
1001 list_for_each_entry(struct qblock, block, &c->blocks, link)
1002
1003 #define vir_for_each_block_rev(block, c) \
1004 list_for_each_entry_rev(struct qblock, block, &c->blocks, link)
1005
1006 /* Loop over the non-NULL members of the successors array. */
1007 #define vir_for_each_successor(succ, block) \
1008 for (struct qblock *succ = block->successors[0]; \
1009 succ != NULL; \
1010 succ = (succ == block->successors[1] ? NULL : \
1011 block->successors[1]))
1012
1013 #define vir_for_each_inst(inst, block) \
1014 list_for_each_entry(struct qinst, inst, &block->instructions, link)
1015
1016 #define vir_for_each_inst_rev(inst, block) \
1017 list_for_each_entry_rev(struct qinst, inst, &block->instructions, link)
1018
1019 #define vir_for_each_inst_safe(inst, block) \
1020 list_for_each_entry_safe(struct qinst, inst, &block->instructions, link)
1021
1022 #define vir_for_each_inst_inorder(inst, c) \
1023 vir_for_each_block(_block, c) \
1024 vir_for_each_inst(inst, _block)
1025
1026 #endif /* V3D_COMPILER_H */