broadcom/vc5: Remove leftover vc4 MSAA lowering setup in the FS key.
[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 * Returns the the offset of the scratch buffer for register spilling.
254 */
255 QUNIFORM_SPILL_OFFSET,
256 QUNIFORM_SPILL_SIZE_PER_THREAD,
257 };
258
259 struct v3d_varying_slot {
260 uint8_t slot_and_component;
261 };
262
263 static inline struct v3d_varying_slot
264 v3d_slot_from_slot_and_component(uint8_t slot, uint8_t component)
265 {
266 assert(slot < 255 / 4);
267 return (struct v3d_varying_slot){ (slot << 2) + component };
268 }
269
270 static inline uint8_t v3d_slot_get_slot(struct v3d_varying_slot slot)
271 {
272 return slot.slot_and_component >> 2;
273 }
274
275 static inline uint8_t v3d_slot_get_component(struct v3d_varying_slot slot)
276 {
277 return slot.slot_and_component & 3;
278 }
279
280 struct v3d_ubo_range {
281 /**
282 * offset in bytes from the start of the ubo where this range is
283 * uploaded.
284 *
285 * Only set once used is set.
286 */
287 uint32_t dst_offset;
288
289 /**
290 * offset in bytes from the start of the gallium uniforms where the
291 * data comes from.
292 */
293 uint32_t src_offset;
294
295 /** size in bytes of this ubo range */
296 uint32_t size;
297 };
298
299 struct v3d_key {
300 void *shader_state;
301 struct {
302 uint8_t swizzle[4];
303 uint8_t return_size;
304 uint8_t return_channels;
305 unsigned compare_mode:1;
306 unsigned compare_func:3;
307 bool clamp_s:1;
308 bool clamp_t:1;
309 bool clamp_r:1;
310 } tex[V3D_MAX_TEXTURE_SAMPLERS];
311 uint8_t ucp_enables;
312 };
313
314 struct v3d_fs_key {
315 struct v3d_key base;
316 bool depth_enabled;
317 bool is_points;
318 bool is_lines;
319 bool alpha_test;
320 bool point_coord_upper_left;
321 bool light_twoside;
322 bool msaa;
323 bool sample_coverage;
324 bool sample_alpha_to_coverage;
325 bool sample_alpha_to_one;
326 bool clamp_color;
327 bool shade_model_flat;
328 uint8_t nr_cbufs;
329 uint8_t swap_color_rb;
330 /* Mask of which render targets need to be written as 32-bit floats */
331 uint8_t f32_color_rb;
332 /* Masks of which render targets need to be written as ints/uints.
333 * Used by gallium to work around lost information in TGSI.
334 */
335 uint8_t int_color_rb;
336 uint8_t uint_color_rb;
337 uint8_t alpha_test_func;
338 uint8_t logicop_func;
339 uint32_t point_sprite_mask;
340
341 struct pipe_rt_blend_state blend;
342 };
343
344 struct v3d_vs_key {
345 struct v3d_key base;
346
347 struct v3d_varying_slot fs_inputs[V3D_MAX_FS_INPUTS];
348 uint8_t num_fs_inputs;
349
350 bool is_coord;
351 bool per_vertex_point_size;
352 bool clamp_color;
353 };
354
355 /** A basic block of VIR intructions. */
356 struct qblock {
357 struct list_head link;
358
359 struct list_head instructions;
360
361 struct set *predecessors;
362 struct qblock *successors[2];
363
364 int index;
365
366 /* Instruction IPs for the first and last instruction of the block.
367 * Set by qpu_schedule.c.
368 */
369 uint32_t start_qpu_ip;
370 uint32_t end_qpu_ip;
371
372 /* Instruction IP for the branch instruction of the block. Set by
373 * qpu_schedule.c.
374 */
375 uint32_t branch_qpu_ip;
376
377 /** Offset within the uniform stream at the start of the block. */
378 uint32_t start_uniform;
379 /** Offset within the uniform stream of the branch instruction */
380 uint32_t branch_uniform;
381
382 /** @{ used by v3d_vir_live_variables.c */
383 BITSET_WORD *def;
384 BITSET_WORD *use;
385 BITSET_WORD *live_in;
386 BITSET_WORD *live_out;
387 int start_ip, end_ip;
388 /** @} */
389 };
390
391 /** Which util/list.h add mode we should use when inserting an instruction. */
392 enum vir_cursor_mode {
393 vir_cursor_add,
394 vir_cursor_addtail,
395 };
396
397 /**
398 * Tracking structure for where new instructions should be inserted. Create
399 * with one of the vir_after_inst()-style helper functions.
400 *
401 * This does not protect against removal of the block or instruction, so we
402 * have an assert in instruction removal to try to catch it.
403 */
404 struct vir_cursor {
405 enum vir_cursor_mode mode;
406 struct list_head *link;
407 };
408
409 static inline struct vir_cursor
410 vir_before_inst(struct qinst *inst)
411 {
412 return (struct vir_cursor){ vir_cursor_addtail, &inst->link };
413 }
414
415 static inline struct vir_cursor
416 vir_after_inst(struct qinst *inst)
417 {
418 return (struct vir_cursor){ vir_cursor_add, &inst->link };
419 }
420
421 static inline struct vir_cursor
422 vir_before_block(struct qblock *block)
423 {
424 return (struct vir_cursor){ vir_cursor_add, &block->instructions };
425 }
426
427 static inline struct vir_cursor
428 vir_after_block(struct qblock *block)
429 {
430 return (struct vir_cursor){ vir_cursor_addtail, &block->instructions };
431 }
432
433 /**
434 * Compiler state saved across compiler invocations, for any expensive global
435 * setup.
436 */
437 struct v3d_compiler {
438 const struct v3d_device_info *devinfo;
439 struct ra_regs *regs;
440 unsigned int reg_class_phys[3];
441 unsigned int reg_class_phys_or_acc[3];
442 };
443
444 struct v3d_compile {
445 const struct v3d_device_info *devinfo;
446 nir_shader *s;
447 nir_function_impl *impl;
448 struct exec_list *cf_node_list;
449 const struct v3d_compiler *compiler;
450
451 /**
452 * Mapping from nir_register * or nir_ssa_def * to array of struct
453 * qreg for the values.
454 */
455 struct hash_table *def_ht;
456
457 /* For each temp, the instruction generating its value. */
458 struct qinst **defs;
459 uint32_t defs_array_size;
460
461 /**
462 * Inputs to the shader, arranged by TGSI declaration order.
463 *
464 * Not all fragment shader QFILE_VARY reads are present in this array.
465 */
466 struct qreg *inputs;
467 struct qreg *outputs;
468 bool msaa_per_sample_output;
469 struct qreg color_reads[V3D_MAX_SAMPLES];
470 struct qreg sample_colors[V3D_MAX_SAMPLES];
471 uint32_t inputs_array_size;
472 uint32_t outputs_array_size;
473 uint32_t uniforms_array_size;
474
475 /* Booleans for whether the corresponding QFILE_VARY[i] is
476 * flat-shaded. This includes gl_FragColor flat-shading, which is
477 * customized based on the shademodel_flat shader key.
478 */
479 uint32_t flat_shade_flags[BITSET_WORDS(V3D_MAX_FS_INPUTS)];
480
481 struct v3d_ubo_range *ubo_ranges;
482 bool *ubo_range_used;
483 uint32_t ubo_ranges_array_size;
484 /** Number of uniform areas tracked in ubo_ranges. */
485 uint32_t num_ubo_ranges;
486 uint32_t next_ubo_dst_offset;
487
488 /* State for whether we're executing on each channel currently. 0 if
489 * yes, otherwise a block number + 1 that the channel jumped to.
490 */
491 struct qreg execute;
492
493 struct qreg line_x, point_x, point_y;
494
495 /**
496 * Instance ID, which comes in before the vertex attribute payload if
497 * the shader record requests it.
498 */
499 struct qreg iid;
500
501 /**
502 * Vertex ID, which comes in before the vertex attribute payload
503 * (after Instance ID) if the shader record requests it.
504 */
505 struct qreg vid;
506
507 /* Fragment shader payload regs. */
508 struct qreg payload_w, payload_w_centroid, payload_z;
509
510 uint8_t vattr_sizes[V3D_MAX_VS_INPUTS];
511 uint32_t num_vpm_writes;
512
513 /* Size in bytes of registers that have been spilled. This is how much
514 * space needs to be available in the spill BO per thread per QPU.
515 */
516 uint32_t spill_size;
517 /* Shader-db stats for register spilling. */
518 uint32_t spills, fills;
519 /**
520 * Register spilling's per-thread base address, shared between each
521 * spill/fill's addressing calculations.
522 */
523 struct qreg spill_base;
524 /* Bit vector of which temps may be spilled */
525 BITSET_WORD *spillable;
526
527 /**
528 * Array of the VARYING_SLOT_* of all FS QFILE_VARY reads.
529 *
530 * This includes those that aren't part of the VPM varyings, like
531 * point/line coordinates.
532 */
533 struct v3d_varying_slot input_slots[V3D_MAX_FS_INPUTS];
534
535 /**
536 * An entry per outputs[] in the VS indicating what the VARYING_SLOT_*
537 * of the output is. Used to emit from the VS in the order that the
538 * FS needs.
539 */
540 struct v3d_varying_slot *output_slots;
541
542 struct pipe_shader_state *shader_state;
543 struct v3d_key *key;
544 struct v3d_fs_key *fs_key;
545 struct v3d_vs_key *vs_key;
546
547 /* Live ranges of temps. */
548 int *temp_start, *temp_end;
549 bool live_intervals_valid;
550
551 uint32_t *uniform_data;
552 enum quniform_contents *uniform_contents;
553 uint32_t uniform_array_size;
554 uint32_t num_uniforms;
555 uint32_t num_outputs;
556 uint32_t output_position_index;
557 nir_variable *output_color_var[4];
558 uint32_t output_point_size_index;
559 uint32_t output_sample_mask_index;
560
561 struct qreg undef;
562 uint32_t num_temps;
563
564 struct vir_cursor cursor;
565 struct list_head blocks;
566 int next_block_index;
567 struct qblock *cur_block;
568 struct qblock *loop_cont_block;
569 struct qblock *loop_break_block;
570
571 uint64_t *qpu_insts;
572 uint32_t qpu_inst_count;
573 uint32_t qpu_inst_size;
574
575 /* For the FS, the number of varying inputs not counting the
576 * point/line varyings payload
577 */
578 uint32_t num_inputs;
579
580 /**
581 * Number of inputs from num_inputs remaining to be queued to the read
582 * FIFO in the VS/CS.
583 */
584 uint32_t num_inputs_remaining;
585
586 /* Number of inputs currently in the read FIFO for the VS/CS */
587 uint32_t num_inputs_in_fifo;
588
589 /** Next offset in the VPM to read from in the VS/CS */
590 uint32_t vpm_read_offset;
591
592 uint32_t program_id;
593 uint32_t variant_id;
594
595 /* Set to compile program in in 1x, 2x, or 4x threaded mode, where
596 * SIG_THREAD_SWITCH is used to hide texturing latency at the cost of
597 * limiting ourselves to the part of the physical reg space.
598 *
599 * On V3D 3.x, 2x or 4x divide the physical reg space by 2x or 4x. On
600 * V3D 4.x, all shaders are 2x threaded, and 4x only divides the
601 * physical reg space in half.
602 */
603 uint8_t threads;
604 struct qinst *last_thrsw;
605 bool last_thrsw_at_top_level;
606
607 bool failed;
608 };
609
610 struct v3d_uniform_list {
611 enum quniform_contents *contents;
612 uint32_t *data;
613 uint32_t count;
614 };
615
616 struct v3d_prog_data {
617 struct v3d_uniform_list uniforms;
618
619 struct v3d_ubo_range *ubo_ranges;
620 uint32_t num_ubo_ranges;
621 uint32_t ubo_size;
622 uint32_t spill_size;
623
624 uint8_t num_inputs;
625 uint8_t threads;
626
627 /* For threads > 1, whether the program should be dispatched in the
628 * after-final-THRSW state.
629 */
630 bool single_seg;
631 };
632
633 struct v3d_vs_prog_data {
634 struct v3d_prog_data base;
635
636 bool uses_iid, uses_vid;
637
638 /* Number of components read from each vertex attribute. */
639 uint8_t vattr_sizes[32];
640
641 /* Total number of components read, for the shader state record. */
642 uint32_t vpm_input_size;
643
644 /* Total number of components written, for the shader state record. */
645 uint32_t vpm_output_size;
646 };
647
648 struct v3d_fs_prog_data {
649 struct v3d_prog_data base;
650
651 struct v3d_varying_slot input_slots[V3D_MAX_FS_INPUTS];
652
653 /* Array of flat shade flags.
654 *
655 * Each entry is only 24 bits (high 8 bits 0), to match the hardware
656 * packet layout.
657 */
658 uint32_t flat_shade_flags[((V3D_MAX_FS_INPUTS - 1) / 24) + 1];
659
660 bool writes_z;
661 bool discard;
662 };
663
664 /* Special nir_load_input intrinsic index for loading the current TLB
665 * destination color.
666 */
667 #define V3D_NIR_TLB_COLOR_READ_INPUT 2000000000
668
669 #define V3D_NIR_MS_MASK_OUTPUT 2000000000
670
671 extern const nir_shader_compiler_options v3d_nir_options;
672
673 const struct v3d_compiler *v3d_compiler_init(const struct v3d_device_info *devinfo);
674 void v3d_compiler_free(const struct v3d_compiler *compiler);
675 void v3d_optimize_nir(struct nir_shader *s);
676
677 uint64_t *v3d_compile_vs(const struct v3d_compiler *compiler,
678 struct v3d_vs_key *key,
679 struct v3d_vs_prog_data *prog_data,
680 nir_shader *s,
681 int program_id, int variant_id,
682 uint32_t *final_assembly_size);
683
684 uint64_t *v3d_compile_fs(const struct v3d_compiler *compiler,
685 struct v3d_fs_key *key,
686 struct v3d_fs_prog_data *prog_data,
687 nir_shader *s,
688 int program_id, int variant_id,
689 uint32_t *final_assembly_size);
690
691 void v3d_nir_to_vir(struct v3d_compile *c);
692
693 void vir_compile_destroy(struct v3d_compile *c);
694 const char *vir_get_stage_name(struct v3d_compile *c);
695 struct qblock *vir_new_block(struct v3d_compile *c);
696 void vir_set_emit_block(struct v3d_compile *c, struct qblock *block);
697 void vir_link_blocks(struct qblock *predecessor, struct qblock *successor);
698 struct qblock *vir_entry_block(struct v3d_compile *c);
699 struct qblock *vir_exit_block(struct v3d_compile *c);
700 struct qinst *vir_add_inst(enum v3d_qpu_add_op op, struct qreg dst,
701 struct qreg src0, struct qreg src1);
702 struct qinst *vir_mul_inst(enum v3d_qpu_mul_op op, struct qreg dst,
703 struct qreg src0, struct qreg src1);
704 struct qinst *vir_branch_inst(enum v3d_qpu_branch_cond cond, struct qreg src0);
705 void vir_remove_instruction(struct v3d_compile *c, struct qinst *qinst);
706 struct qreg vir_uniform(struct v3d_compile *c,
707 enum quniform_contents contents,
708 uint32_t data);
709 void vir_schedule_instructions(struct v3d_compile *c);
710 struct v3d_qpu_instr v3d_qpu_nop(void);
711
712 struct qreg vir_emit_def(struct v3d_compile *c, struct qinst *inst);
713 struct qinst *vir_emit_nondef(struct v3d_compile *c, struct qinst *inst);
714 void vir_set_cond(struct qinst *inst, enum v3d_qpu_cond cond);
715 void vir_set_pf(struct qinst *inst, enum v3d_qpu_pf pf);
716 void vir_set_unpack(struct qinst *inst, int src,
717 enum v3d_qpu_input_unpack unpack);
718
719 struct qreg vir_get_temp(struct v3d_compile *c);
720 void vir_emit_last_thrsw(struct v3d_compile *c);
721 void vir_calculate_live_intervals(struct v3d_compile *c);
722 bool vir_has_implicit_uniform(struct qinst *inst);
723 int vir_get_implicit_uniform_src(struct qinst *inst);
724 int vir_get_non_sideband_nsrc(struct qinst *inst);
725 int vir_get_nsrc(struct qinst *inst);
726 bool vir_has_side_effects(struct v3d_compile *c, struct qinst *inst);
727 bool vir_get_add_op(struct qinst *inst, enum v3d_qpu_add_op *op);
728 bool vir_get_mul_op(struct qinst *inst, enum v3d_qpu_mul_op *op);
729 bool vir_is_raw_mov(struct qinst *inst);
730 bool vir_is_tex(struct qinst *inst);
731 bool vir_is_add(struct qinst *inst);
732 bool vir_is_mul(struct qinst *inst);
733 bool vir_is_float_input(struct qinst *inst);
734 bool vir_depends_on_flags(struct qinst *inst);
735 bool vir_writes_r3(const struct v3d_device_info *devinfo, struct qinst *inst);
736 bool vir_writes_r4(const struct v3d_device_info *devinfo, struct qinst *inst);
737 struct qreg vir_follow_movs(struct v3d_compile *c, struct qreg reg);
738 uint8_t vir_channels_written(struct qinst *inst);
739 struct qreg ntq_get_src(struct v3d_compile *c, nir_src src, int i);
740 void ntq_store_dest(struct v3d_compile *c, nir_dest *dest, int chan,
741 struct qreg result);
742 void vir_emit_thrsw(struct v3d_compile *c);
743
744 void vir_dump(struct v3d_compile *c);
745 void vir_dump_inst(struct v3d_compile *c, struct qinst *inst);
746
747 void vir_validate(struct v3d_compile *c);
748
749 void vir_optimize(struct v3d_compile *c);
750 bool vir_opt_algebraic(struct v3d_compile *c);
751 bool vir_opt_constant_folding(struct v3d_compile *c);
752 bool vir_opt_copy_propagate(struct v3d_compile *c);
753 bool vir_opt_dead_code(struct v3d_compile *c);
754 bool vir_opt_peephole_sf(struct v3d_compile *c);
755 bool vir_opt_small_immediates(struct v3d_compile *c);
756 bool vir_opt_vpm(struct v3d_compile *c);
757 void v3d_nir_lower_blend(nir_shader *s, struct v3d_compile *c);
758 void v3d_nir_lower_io(nir_shader *s, struct v3d_compile *c);
759 void v3d_nir_lower_txf_ms(nir_shader *s, struct v3d_compile *c);
760 void vir_lower_uniforms(struct v3d_compile *c);
761
762 void v3d33_vir_vpm_read_setup(struct v3d_compile *c, int num_components);
763 void v3d33_vir_vpm_write_setup(struct v3d_compile *c);
764 void v3d33_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr);
765 void v3d40_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr);
766
767 void v3d_vir_to_qpu(struct v3d_compile *c, struct qpu_reg *temp_registers);
768 uint32_t v3d_qpu_schedule_instructions(struct v3d_compile *c);
769 void qpu_validate(struct v3d_compile *c);
770 struct qpu_reg *v3d_register_allocate(struct v3d_compile *c, bool *spilled);
771 bool vir_init_reg_sets(struct v3d_compiler *compiler);
772
773 void vir_PF(struct v3d_compile *c, struct qreg src, enum v3d_qpu_pf pf);
774
775 static inline bool
776 quniform_contents_is_texture_p0(enum quniform_contents contents)
777 {
778 return (contents >= QUNIFORM_TEXTURE_CONFIG_P0_0 &&
779 contents < (QUNIFORM_TEXTURE_CONFIG_P0_0 +
780 V3D_MAX_TEXTURE_SAMPLERS));
781 }
782
783 static inline struct qreg
784 vir_uniform_ui(struct v3d_compile *c, uint32_t ui)
785 {
786 return vir_uniform(c, QUNIFORM_CONSTANT, ui);
787 }
788
789 static inline struct qreg
790 vir_uniform_f(struct v3d_compile *c, float f)
791 {
792 return vir_uniform(c, QUNIFORM_CONSTANT, fui(f));
793 }
794
795 #define VIR_ALU0(name, vir_inst, op) \
796 static inline struct qreg \
797 vir_##name(struct v3d_compile *c) \
798 { \
799 return vir_emit_def(c, vir_inst(op, c->undef, \
800 c->undef, c->undef)); \
801 } \
802 static inline struct qinst * \
803 vir_##name##_dest(struct v3d_compile *c, struct qreg dest) \
804 { \
805 return vir_emit_nondef(c, vir_inst(op, dest, \
806 c->undef, c->undef)); \
807 }
808
809 #define VIR_ALU1(name, vir_inst, op) \
810 static inline struct qreg \
811 vir_##name(struct v3d_compile *c, struct qreg a) \
812 { \
813 return vir_emit_def(c, vir_inst(op, c->undef, \
814 a, c->undef)); \
815 } \
816 static inline struct qinst * \
817 vir_##name##_dest(struct v3d_compile *c, struct qreg dest, \
818 struct qreg a) \
819 { \
820 return vir_emit_nondef(c, vir_inst(op, dest, a, \
821 c->undef)); \
822 }
823
824 #define VIR_ALU2(name, vir_inst, op) \
825 static inline struct qreg \
826 vir_##name(struct v3d_compile *c, struct qreg a, struct qreg b) \
827 { \
828 return vir_emit_def(c, vir_inst(op, c->undef, a, b)); \
829 } \
830 static inline struct qinst * \
831 vir_##name##_dest(struct v3d_compile *c, struct qreg dest, \
832 struct qreg a, struct qreg b) \
833 { \
834 return vir_emit_nondef(c, vir_inst(op, dest, a, b)); \
835 }
836
837 #define VIR_NODST_0(name, vir_inst, op) \
838 static inline struct qinst * \
839 vir_##name(struct v3d_compile *c) \
840 { \
841 return vir_emit_nondef(c, vir_inst(op, c->undef, \
842 c->undef, c->undef)); \
843 }
844
845 #define VIR_NODST_1(name, vir_inst, op) \
846 static inline struct qinst * \
847 vir_##name(struct v3d_compile *c, struct qreg a) \
848 { \
849 return vir_emit_nondef(c, vir_inst(op, c->undef, \
850 a, c->undef)); \
851 }
852
853 #define VIR_NODST_2(name, vir_inst, op) \
854 static inline struct qinst * \
855 vir_##name(struct v3d_compile *c, struct qreg a, struct qreg b) \
856 { \
857 return vir_emit_nondef(c, vir_inst(op, c->undef, \
858 a, b)); \
859 }
860
861 #define VIR_A_ALU2(name) VIR_ALU2(name, vir_add_inst, V3D_QPU_A_##name)
862 #define VIR_M_ALU2(name) VIR_ALU2(name, vir_mul_inst, V3D_QPU_M_##name)
863 #define VIR_A_ALU1(name) VIR_ALU1(name, vir_add_inst, V3D_QPU_A_##name)
864 #define VIR_M_ALU1(name) VIR_ALU1(name, vir_mul_inst, V3D_QPU_M_##name)
865 #define VIR_A_ALU0(name) VIR_ALU0(name, vir_add_inst, V3D_QPU_A_##name)
866 #define VIR_M_ALU0(name) VIR_ALU0(name, vir_mul_inst, V3D_QPU_M_##name)
867 #define VIR_A_NODST_2(name) VIR_NODST_2(name, vir_add_inst, V3D_QPU_A_##name)
868 #define VIR_M_NODST_2(name) VIR_NODST_2(name, vir_mul_inst, V3D_QPU_M_##name)
869 #define VIR_A_NODST_1(name) VIR_NODST_1(name, vir_add_inst, V3D_QPU_A_##name)
870 #define VIR_M_NODST_1(name) VIR_NODST_1(name, vir_mul_inst, V3D_QPU_M_##name)
871 #define VIR_A_NODST_0(name) VIR_NODST_0(name, vir_add_inst, V3D_QPU_A_##name)
872
873 VIR_A_ALU2(FADD)
874 VIR_A_ALU2(VFPACK)
875 VIR_A_ALU2(FSUB)
876 VIR_A_ALU2(FMIN)
877 VIR_A_ALU2(FMAX)
878
879 VIR_A_ALU2(ADD)
880 VIR_A_ALU2(SUB)
881 VIR_A_ALU2(SHL)
882 VIR_A_ALU2(SHR)
883 VIR_A_ALU2(ASR)
884 VIR_A_ALU2(ROR)
885 VIR_A_ALU2(MIN)
886 VIR_A_ALU2(MAX)
887 VIR_A_ALU2(UMIN)
888 VIR_A_ALU2(UMAX)
889 VIR_A_ALU2(AND)
890 VIR_A_ALU2(OR)
891 VIR_A_ALU2(XOR)
892 VIR_A_ALU2(VADD)
893 VIR_A_ALU2(VSUB)
894 VIR_A_ALU2(STVPMV)
895 VIR_A_ALU1(NOT)
896 VIR_A_ALU1(NEG)
897 VIR_A_ALU1(FLAPUSH)
898 VIR_A_ALU1(FLBPUSH)
899 VIR_A_ALU1(FLBPOP)
900 VIR_A_ALU1(SETMSF)
901 VIR_A_ALU1(SETREVF)
902 VIR_A_ALU0(TIDX)
903 VIR_A_ALU0(EIDX)
904 VIR_A_ALU1(LDVPMV_IN)
905 VIR_A_ALU1(LDVPMV_OUT)
906
907 VIR_A_ALU0(FXCD)
908 VIR_A_ALU0(XCD)
909 VIR_A_ALU0(FYCD)
910 VIR_A_ALU0(YCD)
911 VIR_A_ALU0(MSF)
912 VIR_A_ALU0(REVF)
913 VIR_A_NODST_1(VPMSETUP)
914 VIR_A_NODST_0(VPMWT)
915 VIR_A_ALU2(FCMP)
916 VIR_A_ALU2(VFMAX)
917
918 VIR_A_ALU1(FROUND)
919 VIR_A_ALU1(FTOIN)
920 VIR_A_ALU1(FTRUNC)
921 VIR_A_ALU1(FTOIZ)
922 VIR_A_ALU1(FFLOOR)
923 VIR_A_ALU1(FTOUZ)
924 VIR_A_ALU1(FCEIL)
925 VIR_A_ALU1(FTOC)
926
927 VIR_A_ALU1(FDX)
928 VIR_A_ALU1(FDY)
929
930 VIR_A_ALU1(ITOF)
931 VIR_A_ALU1(CLZ)
932 VIR_A_ALU1(UTOF)
933
934 VIR_M_ALU2(UMUL24)
935 VIR_M_ALU2(FMUL)
936 VIR_M_ALU2(SMUL24)
937 VIR_M_NODST_2(MULTOP)
938
939 VIR_M_ALU1(MOV)
940 VIR_M_ALU1(FMOV)
941
942 static inline struct qinst *
943 vir_MOV_cond(struct v3d_compile *c, enum v3d_qpu_cond cond,
944 struct qreg dest, struct qreg src)
945 {
946 struct qinst *mov = vir_MOV_dest(c, dest, src);
947 vir_set_cond(mov, cond);
948 return mov;
949 }
950
951 static inline struct qreg
952 vir_SEL(struct v3d_compile *c, enum v3d_qpu_cond cond,
953 struct qreg src0, struct qreg src1)
954 {
955 struct qreg t = vir_get_temp(c);
956 vir_MOV_dest(c, t, src1);
957 vir_MOV_cond(c, cond, t, src0);
958 return t;
959 }
960
961 static inline struct qinst *
962 vir_NOP(struct v3d_compile *c)
963 {
964 return vir_emit_nondef(c, vir_add_inst(V3D_QPU_A_NOP,
965 c->undef, c->undef, c->undef));
966 }
967
968 static inline struct qreg
969 vir_LDTMU(struct v3d_compile *c)
970 {
971 if (c->devinfo->ver >= 41) {
972 struct qinst *ldtmu = vir_add_inst(V3D_QPU_A_NOP, c->undef,
973 c->undef, c->undef);
974 ldtmu->qpu.sig.ldtmu = true;
975
976 return vir_emit_def(c, ldtmu);
977 } else {
978 vir_NOP(c)->qpu.sig.ldtmu = true;
979 return vir_MOV(c, vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_R4));
980 }
981 }
982
983 static inline struct qreg
984 vir_UMUL(struct v3d_compile *c, struct qreg src0, struct qreg src1)
985 {
986 vir_MULTOP(c, src0, src1);
987 return vir_UMUL24(c, src0, src1);
988 }
989
990 /*
991 static inline struct qreg
992 vir_LOAD_IMM(struct v3d_compile *c, uint32_t val)
993 {
994 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM, c->undef,
995 vir_reg(QFILE_LOAD_IMM, val), c->undef));
996 }
997
998 static inline struct qreg
999 vir_LOAD_IMM_U2(struct v3d_compile *c, uint32_t val)
1000 {
1001 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM_U2, c->undef,
1002 vir_reg(QFILE_LOAD_IMM, val),
1003 c->undef));
1004 }
1005 static inline struct qreg
1006 vir_LOAD_IMM_I2(struct v3d_compile *c, uint32_t val)
1007 {
1008 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM_I2, c->undef,
1009 vir_reg(QFILE_LOAD_IMM, val),
1010 c->undef));
1011 }
1012 */
1013
1014 static inline struct qinst *
1015 vir_BRANCH(struct v3d_compile *c, enum v3d_qpu_cond cond)
1016 {
1017 /* The actual uniform_data value will be set at scheduling time */
1018 return vir_emit_nondef(c, vir_branch_inst(cond, vir_uniform_ui(c, 0)));
1019 }
1020
1021 #define vir_for_each_block(block, c) \
1022 list_for_each_entry(struct qblock, block, &c->blocks, link)
1023
1024 #define vir_for_each_block_rev(block, c) \
1025 list_for_each_entry_rev(struct qblock, block, &c->blocks, link)
1026
1027 /* Loop over the non-NULL members of the successors array. */
1028 #define vir_for_each_successor(succ, block) \
1029 for (struct qblock *succ = block->successors[0]; \
1030 succ != NULL; \
1031 succ = (succ == block->successors[1] ? NULL : \
1032 block->successors[1]))
1033
1034 #define vir_for_each_inst(inst, block) \
1035 list_for_each_entry(struct qinst, inst, &block->instructions, link)
1036
1037 #define vir_for_each_inst_rev(inst, block) \
1038 list_for_each_entry_rev(struct qinst, inst, &block->instructions, link)
1039
1040 #define vir_for_each_inst_safe(inst, block) \
1041 list_for_each_entry_safe(struct qinst, inst, &block->instructions, link)
1042
1043 #define vir_for_each_inst_inorder(inst, c) \
1044 vir_for_each_block(_block, c) \
1045 vir_for_each_inst(inst, _block)
1046
1047 #endif /* V3D_COMPILER_H */