src/gallium/drivers/vc4/vc4_qir.h

   1 /*
   2  * Copyright © 2014 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 VC4_QIR_H
  25 #define VC4_QIR_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 "glsl/nir/nir.h"
  36 #include "util/list.h"
  37 #include "util/u_math.h"
  38
  39 #include "vc4_screen.h"
  40 #include "pipe/p_state.h"
  41
  42 struct nir_builder;
  43
  44 enum qfile {
  45         QFILE_NULL,
  46         QFILE_TEMP,
  47         QFILE_VARY,
  48         QFILE_UNIF,
  49         QFILE_VPM,
  50
  51         /**
  52          * Stores an immediate value in the index field that can be turned
  53          * into a small immediate field by qpu_encode_small_immediate().
  54          */
  55         QFILE_SMALL_IMM,
  56 };
  57
  58 struct qreg {
  59         enum qfile file;
  60         uint32_t index;
  61         int pack;
  62 };
  63
  64 enum qop {
  65         QOP_UNDEF,
  66         QOP_MOV,
  67         QOP_FADD,
  68         QOP_FSUB,
  69         QOP_FMUL,
  70         QOP_V8MULD,
  71         QOP_V8MIN,
  72         QOP_V8MAX,
  73         QOP_V8ADDS,
  74         QOP_V8SUBS,
  75         QOP_MUL24,
  76         QOP_FMIN,
  77         QOP_FMAX,
  78         QOP_FMINABS,
  79         QOP_FMAXABS,
  80         QOP_ADD,
  81         QOP_SUB,
  82         QOP_SHL,
  83         QOP_SHR,
  84         QOP_ASR,
  85         QOP_MIN,
  86         QOP_MAX,
  87         QOP_AND,
  88         QOP_OR,
  89         QOP_XOR,
  90         QOP_NOT,
  91
  92         /* Note: Orderings of these compares must be the same as in
  93          * qpu_defines.h.  Selects the src[0] if the ns flag bit is set,
  94          * otherwise 0. */
  95         QOP_SEL_X_0_ZS,
  96         QOP_SEL_X_0_ZC,
  97         QOP_SEL_X_0_NS,
  98         QOP_SEL_X_0_NC,
  99         /* Selects the src[0] if the ns flag bit is set, otherwise src[1]. */
 100         QOP_SEL_X_Y_ZS,
 101         QOP_SEL_X_Y_ZC,
 102         QOP_SEL_X_Y_NS,
 103         QOP_SEL_X_Y_NC,
 104
 105         QOP_FTOI,
 106         QOP_ITOF,
 107         QOP_RCP,
 108         QOP_RSQ,
 109         QOP_EXP2,
 110         QOP_LOG2,
 111         QOP_VW_SETUP,
 112         QOP_VR_SETUP,
 113         QOP_PACK_8888_F,
 114         QOP_PACK_8A_F,
 115         QOP_PACK_8B_F,
 116         QOP_PACK_8C_F,
 117         QOP_PACK_8D_F,
 118         QOP_TLB_DISCARD_SETUP,
 119         QOP_TLB_STENCIL_SETUP,
 120         QOP_TLB_Z_WRITE,
 121         QOP_TLB_COLOR_WRITE,
 122         QOP_TLB_COLOR_READ,
 123         QOP_VARY_ADD_C,
 124
 125         QOP_FRAG_X,
 126         QOP_FRAG_Y,
 127         QOP_FRAG_Z,
 128         QOP_FRAG_W,
 129         QOP_FRAG_REV_FLAG,
 130
 131         QOP_UNPACK_8A_F,
 132         QOP_UNPACK_8B_F,
 133         QOP_UNPACK_8C_F,
 134         QOP_UNPACK_8D_F,
 135         QOP_UNPACK_16A_F,
 136         QOP_UNPACK_16B_F,
 137
 138         QOP_UNPACK_8A_I,
 139         QOP_UNPACK_8B_I,
 140         QOP_UNPACK_8C_I,
 141         QOP_UNPACK_8D_I,
 142         QOP_UNPACK_16A_I,
 143         QOP_UNPACK_16B_I,
 144
 145         /** Texture x coordinate parameter write */
 146         QOP_TEX_S,
 147         /** Texture y coordinate parameter write */
 148         QOP_TEX_T,
 149         /** Texture border color parameter or cube map z coordinate write */
 150         QOP_TEX_R,
 151         /** Texture LOD bias parameter write */
 152         QOP_TEX_B,
 153
 154         /**
 155          * Texture-unit 4-byte read with address provided direct in S
 156          * cooordinate.
 157          *
 158          * The first operand is the offset from the start of the UBO, and the
 159          * second is the uniform that has the UBO's base pointer.
 160          */
 161         QOP_TEX_DIRECT,
 162
 163         /**
 164          * Signal of texture read being necessary and then reading r4 into
 165          * the destination
 166          */
 167         QOP_TEX_RESULT,
 168 };
 169
 170 struct queued_qpu_inst {
 171         struct list_head link;
 172         uint64_t inst;
 173 };
 174
 175 struct qinst {
 176         struct list_head link;
 177
 178         enum qop op;
 179         struct qreg dst;
 180         struct qreg *src;
 181         bool sf;
 182 };
 183
 184 enum qstage {
 185         /**
 186          * Coordinate shader, runs during binning, before the VS, and just
 187          * outputs position.
 188          */
 189         QSTAGE_COORD,
 190         QSTAGE_VERT,
 191         QSTAGE_FRAG,
 192 };
 193
 194 enum quniform_contents {
 195         /**
 196          * Indicates that a constant 32-bit value is copied from the program's
 197          * uniform contents.
 198          */
 199         QUNIFORM_CONSTANT,
 200         /**
 201          * Indicates that the program's uniform contents are used as an index
 202          * into the GL uniform storage.
 203          */
 204         QUNIFORM_UNIFORM,
 205
 206         /** @{
 207          * Scaling factors from clip coordinates to relative to the viewport
 208          * center.
 209          *
 210          * This is used by the coordinate and vertex shaders to produce the
 211          * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed
 212          * point offsets from the viewport ccenter.
 213          */
 214         QUNIFORM_VIEWPORT_X_SCALE,
 215         QUNIFORM_VIEWPORT_Y_SCALE,
 216         /** @} */
 217
 218         QUNIFORM_VIEWPORT_Z_OFFSET,
 219         QUNIFORM_VIEWPORT_Z_SCALE,
 220
 221         QUNIFORM_USER_CLIP_PLANE,
 222
 223         /**
 224          * A reference to a texture config parameter 0 uniform.
 225          *
 226          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 227          * defines texture type, miplevels, and such.  It will be found as a
 228          * parameter to the first QOP_TEX_[STRB] instruction in a sequence.
 229          */
 230         QUNIFORM_TEXTURE_CONFIG_P0,
 231
 232         /**
 233          * A reference to a texture config parameter 1 uniform.
 234          *
 235          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 236          * defines texture width, height, filters, and wrap modes.  It will be
 237          * found as a parameter to the second QOP_TEX_[STRB] instruction in a
 238          * sequence.
 239          */
 240         QUNIFORM_TEXTURE_CONFIG_P1,
 241
 242         /** A reference to a texture config parameter 2 cubemap stride uniform */
 243         QUNIFORM_TEXTURE_CONFIG_P2,
 244
 245         QUNIFORM_UBO_ADDR,
 246
 247         QUNIFORM_TEXRECT_SCALE_X,
 248         QUNIFORM_TEXRECT_SCALE_Y,
 249
 250         QUNIFORM_TEXTURE_BORDER_COLOR,
 251
 252         QUNIFORM_BLEND_CONST_COLOR_X,
 253         QUNIFORM_BLEND_CONST_COLOR_Y,
 254         QUNIFORM_BLEND_CONST_COLOR_Z,
 255         QUNIFORM_BLEND_CONST_COLOR_W,
 256         QUNIFORM_BLEND_CONST_COLOR_RGBA,
 257         QUNIFORM_BLEND_CONST_COLOR_AAAA,
 258
 259         QUNIFORM_STENCIL,
 260
 261         QUNIFORM_ALPHA_REF,
 262 };
 263
 264 struct vc4_varying_slot {
 265         uint8_t slot;
 266         uint8_t swizzle;
 267 };
 268
 269 struct vc4_compiler_ubo_range {
 270         /**
 271          * offset in bytes from the start of the ubo where this range is
 272          * uploaded.
 273          *
 274          * Only set once used is set.
 275          */
 276         uint32_t dst_offset;
 277
 278         /**
 279          * offset in bytes from the start of the gallium uniforms where the
 280          * data comes from.
 281          */
 282         uint32_t src_offset;
 283
 284         /** size in bytes of this ubo range */
 285         uint32_t size;
 286
 287         /**
 288          * Set if this range is used by the shader for indirect uniforms
 289          * access.
 290          */
 291         bool used;
 292 };
 293
 294 struct vc4_key {
 295         struct vc4_uncompiled_shader *shader_state;
 296         struct {
 297                 enum pipe_format format;
 298                 unsigned compare_mode:1;
 299                 unsigned compare_func:3;
 300                 unsigned wrap_s:3;
 301                 unsigned wrap_t:3;
 302                 uint8_t swizzle[4];
 303         } tex[VC4_MAX_TEXTURE_SAMPLERS];
 304         uint8_t ucp_enables;
 305 };
 306
 307 struct vc4_fs_key {
 308         struct vc4_key base;
 309         enum pipe_format color_format;
 310         bool depth_enabled;
 311         bool stencil_enabled;
 312         bool stencil_twoside;
 313         bool stencil_full_writemasks;
 314         bool is_points;
 315         bool is_lines;
 316         bool alpha_test;
 317         bool point_coord_upper_left;
 318         bool light_twoside;
 319         uint8_t alpha_test_func;
 320         uint8_t logicop_func;
 321         uint32_t point_sprite_mask;
 322
 323         struct pipe_rt_blend_state blend;
 324 };
 325
 326 struct vc4_vs_key {
 327         struct vc4_key base;
 328
 329         /**
 330          * This is a proxy for the array of FS input semantics, which is
 331          * larger than we would want to put in the key.
 332          */
 333         uint64_t compiled_fs_id;
 334
 335         enum pipe_format attr_formats[8];
 336         bool is_coord;
 337         bool per_vertex_point_size;
 338 };
 339
 340 struct vc4_compile {
 341         struct vc4_context *vc4;
 342         nir_shader *s;
 343         nir_function_impl *impl;
 344         struct exec_list *cf_node_list;
 345
 346         /**
 347          * Mapping from nir_register * or nir_ssa_def * to array of struct
 348          * qreg for the values.
 349          */
 350         struct hash_table *def_ht;
 351
 352         /* For each temp, the instruction generating its value. */
 353         struct qinst **defs;
 354         uint32_t defs_array_size;
 355
 356         /**
 357          * Inputs to the shader, arranged by TGSI declaration order.
 358          *
 359          * Not all fragment shader QFILE_VARY reads are present in this array.
 360          */
 361         struct qreg *inputs;
 362         struct qreg *outputs;
 363         uint32_t inputs_array_size;
 364         uint32_t outputs_array_size;
 365         uint32_t uniforms_array_size;
 366
 367         struct vc4_compiler_ubo_range *ubo_ranges;
 368         uint32_t ubo_ranges_array_size;
 369         /** Number of uniform areas declared in ubo_ranges. */
 370         uint32_t num_uniform_ranges;
 371         /** Number of uniform areas used for indirect addressed loads. */
 372         uint32_t num_ubo_ranges;
 373         uint32_t next_ubo_dst_offset;
 374
 375         struct qreg line_x, point_x, point_y;
 376         struct qreg discard;
 377
 378         uint8_t vattr_sizes[8];
 379
 380         /**
 381          * Array of the VARYING_SLOT_* of all FS QFILE_VARY reads.
 382          *
 383          * This includes those that aren't part of the VPM varyings, like
 384          * point/line coordinates.
 385          */
 386         struct vc4_varying_slot *input_slots;
 387         uint32_t num_input_slots;
 388         uint32_t input_slots_array_size;
 389
 390         /**
 391          * An entry per outputs[] in the VS indicating what the VARYING_SLOT_*
 392          * of the output is.  Used to emit from the VS in the order that the
 393          * FS needs.
 394          */
 395         struct vc4_varying_slot *output_slots;
 396
 397         struct pipe_shader_state *shader_state;
 398         struct vc4_key *key;
 399         struct vc4_fs_key *fs_key;
 400         struct vc4_vs_key *vs_key;
 401
 402         uint32_t *uniform_data;
 403         enum quniform_contents *uniform_contents;
 404         uint32_t uniform_array_size;
 405         uint32_t num_uniforms;
 406         uint32_t num_outputs;
 407         uint32_t num_texture_samples;
 408         uint32_t output_position_index;
 409         uint32_t output_color_index;
 410         uint32_t output_point_size_index;
 411
 412         struct qreg undef;
 413         enum qstage stage;
 414         uint32_t num_temps;
 415         struct list_head instructions;
 416         uint32_t immediates[1024];
 417
 418         struct list_head qpu_inst_list;
 419         uint64_t *qpu_insts;
 420         uint32_t qpu_inst_count;
 421         uint32_t qpu_inst_size;
 422         uint32_t num_inputs;
 423
 424         uint32_t program_id;
 425         uint32_t variant_id;
 426 };
 427
 428 /* Special nir_load_input intrinsic index for loading the current TLB
 429  * destination color.
 430  */
 431 #define VC4_NIR_TLB_COLOR_READ_INPUT            2000000000
 432
 433 /* Special offset for nir_load_uniform values to get a QUNIFORM_*
 434  * state-dependent value.
 435  */
 436 #define VC4_NIR_STATE_UNIFORM_OFFSET            2000000000
 437
 438 struct vc4_compile *qir_compile_init(void);
 439 void qir_compile_destroy(struct vc4_compile *c);
 440 struct qinst *qir_inst(enum qop op, struct qreg dst,
 441                        struct qreg src0, struct qreg src1);
 442 struct qinst *qir_inst4(enum qop op, struct qreg dst,
 443                         struct qreg a,
 444                         struct qreg b,
 445                         struct qreg c,
 446                         struct qreg d);
 447 void qir_remove_instruction(struct vc4_compile *c, struct qinst *qinst);
 448 struct qreg qir_uniform(struct vc4_compile *c,
 449                         enum quniform_contents contents,
 450                         uint32_t data);
 451 void qir_reorder_uniforms(struct vc4_compile *c);
 452
 453 void qir_emit(struct vc4_compile *c, struct qinst *inst);
 454 static inline void qir_emit_nodef(struct vc4_compile *c, struct qinst *inst)
 455 {
 456         list_addtail(&inst->link, &c->instructions);
 457 }
 458
 459 struct qreg qir_get_temp(struct vc4_compile *c);
 460 int qir_get_op_nsrc(enum qop qop);
 461 bool qir_reg_equals(struct qreg a, struct qreg b);
 462 bool qir_has_side_effects(struct vc4_compile *c, struct qinst *inst);
 463 bool qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst);
 464 bool qir_is_multi_instruction(struct qinst *inst);
 465 bool qir_is_mul(struct qinst *inst);
 466 bool qir_is_tex(struct qinst *inst);
 467 bool qir_depends_on_flags(struct qinst *inst);
 468 bool qir_writes_r4(struct qinst *inst);
 469 bool qir_src_needs_a_file(struct qinst *inst);
 470 struct qreg qir_follow_movs(struct vc4_compile *c, struct qreg reg);
 471
 472 void qir_dump(struct vc4_compile *c);
 473 void qir_dump_inst(struct vc4_compile *c, struct qinst *inst);
 474 const char *qir_get_stage_name(enum qstage stage);
 475
 476 void qir_optimize(struct vc4_compile *c);
 477 bool qir_opt_algebraic(struct vc4_compile *c);
 478 bool qir_opt_constant_folding(struct vc4_compile *c);
 479 bool qir_opt_copy_propagation(struct vc4_compile *c);
 480 bool qir_opt_cse(struct vc4_compile *c);
 481 bool qir_opt_dead_code(struct vc4_compile *c);
 482 bool qir_opt_small_immediates(struct vc4_compile *c);
 483 bool qir_opt_vpm_writes(struct vc4_compile *c);
 484 void vc4_nir_lower_blend(struct vc4_compile *c);
 485 void vc4_nir_lower_io(struct vc4_compile *c);
 486 nir_ssa_def *vc4_nir_get_state_uniform(struct nir_builder *b,
 487                                        enum quniform_contents contents);
 488 nir_ssa_def *vc4_nir_get_swizzled_channel(struct nir_builder *b,
 489                                           nir_ssa_def **srcs, int swiz);
 490 void qir_lower_uniforms(struct vc4_compile *c);
 491
 492 void qpu_schedule_instructions(struct vc4_compile *c);
 493
 494 void qir_SF(struct vc4_compile *c, struct qreg src);
 495
 496 static inline struct qreg
 497 qir_uniform_ui(struct vc4_compile *c, uint32_t ui)
 498 {
 499         return qir_uniform(c, QUNIFORM_CONSTANT, ui);
 500 }
 501
 502 static inline struct qreg
 503 qir_uniform_f(struct vc4_compile *c, float f)
 504 {
 505         return qir_uniform(c, QUNIFORM_CONSTANT, fui(f));
 506 }
 507
 508 #define QIR_ALU0(name)                                                   \
 509 static inline struct qreg                                                \
 510 qir_##name(struct vc4_compile *c)                                        \
 511 {                                                                        \
 512         struct qreg t = qir_get_temp(c);                                 \
 513         qir_emit(c, qir_inst(QOP_##name, t, c->undef, c->undef));        \
 514         return t;                                                        \
 515 }
 516
 517 #define QIR_ALU1(name)                                                   \
 518 static inline struct qreg                                                \
 519 qir_##name(struct vc4_compile *c, struct qreg a)                         \
 520 {                                                                        \
 521         struct qreg t = qir_get_temp(c);                                 \
 522         qir_emit(c, qir_inst(QOP_##name, t, a, c->undef));               \
 523         return t;                                                        \
 524 }                                                                        \
 525 static inline void                                                       \
 526 qir_##name##_dest(struct vc4_compile *c, struct qreg dest,               \
 527                   struct qreg a)                                         \
 528 {                                                                        \
 529         qir_emit_nodef(c, qir_inst(QOP_##name, dest, a, c->undef));      \
 530 }
 531
 532 #define QIR_ALU2(name)                                                   \
 533 static inline struct qreg                                                \
 534 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b)          \
 535 {                                                                        \
 536         struct qreg t = qir_get_temp(c);                                 \
 537         qir_emit(c, qir_inst(QOP_##name, t, a, b));                      \
 538         return t;                                                        \
 539 }                                                                        \
 540 static inline void                                                       \
 541 qir_##name##_dest(struct vc4_compile *c, struct qreg dest,               \
 542                   struct qreg a, struct qreg b)                          \
 543 {                                                                        \
 544         qir_emit_nodef(c, qir_inst(QOP_##name, dest, a, b));             \
 545 }
 546
 547 #define QIR_NODST_1(name)                                               \
 548 static inline void                                                      \
 549 qir_##name(struct vc4_compile *c, struct qreg a)                        \
 550 {                                                                       \
 551         qir_emit(c, qir_inst(QOP_##name, c->undef, a, c->undef));       \
 552 }
 553
 554 #define QIR_NODST_2(name)                                               \
 555 static inline void                                                      \
 556 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b)         \
 557 {                                                                       \
 558         qir_emit(c, qir_inst(QOP_##name, c->undef, a, b));       \
 559 }
 560
 561 #define QIR_PACK(name)                                                   \
 562 static inline struct qreg                                                \
 563 qir_##name(struct vc4_compile *c, struct qreg dest, struct qreg a)       \
 564 {                                                                        \
 565         qir_emit_nodef(c, qir_inst(QOP_##name, dest, a, c->undef));      \
 566         return dest;                                                     \
 567 }
 568
 569 QIR_ALU1(MOV)
 570 QIR_ALU2(FADD)
 571 QIR_ALU2(FSUB)
 572 QIR_ALU2(FMUL)
 573 QIR_ALU2(V8MULD)
 574 QIR_ALU2(V8MIN)
 575 QIR_ALU2(V8MAX)
 576 QIR_ALU2(V8ADDS)
 577 QIR_ALU2(V8SUBS)
 578 QIR_ALU2(MUL24)
 579 QIR_ALU1(SEL_X_0_ZS)
 580 QIR_ALU1(SEL_X_0_ZC)
 581 QIR_ALU1(SEL_X_0_NS)
 582 QIR_ALU1(SEL_X_0_NC)
 583 QIR_ALU2(SEL_X_Y_ZS)
 584 QIR_ALU2(SEL_X_Y_ZC)
 585 QIR_ALU2(SEL_X_Y_NS)
 586 QIR_ALU2(SEL_X_Y_NC)
 587 QIR_ALU2(FMIN)
 588 QIR_ALU2(FMAX)
 589 QIR_ALU2(FMINABS)
 590 QIR_ALU2(FMAXABS)
 591 QIR_ALU1(FTOI)
 592 QIR_ALU1(ITOF)
 593
 594 QIR_ALU2(ADD)
 595 QIR_ALU2(SUB)
 596 QIR_ALU2(SHL)
 597 QIR_ALU2(SHR)
 598 QIR_ALU2(ASR)
 599 QIR_ALU2(MIN)
 600 QIR_ALU2(MAX)
 601 QIR_ALU2(AND)
 602 QIR_ALU2(OR)
 603 QIR_ALU2(XOR)
 604 QIR_ALU1(NOT)
 605
 606 QIR_ALU1(RCP)
 607 QIR_ALU1(RSQ)
 608 QIR_ALU1(EXP2)
 609 QIR_ALU1(LOG2)
 610 QIR_ALU1(PACK_8888_F)
 611 QIR_PACK(PACK_8A_F)
 612 QIR_PACK(PACK_8B_F)
 613 QIR_PACK(PACK_8C_F)
 614 QIR_PACK(PACK_8D_F)
 615 QIR_ALU1(VARY_ADD_C)
 616 QIR_NODST_2(TEX_S)
 617 QIR_NODST_2(TEX_T)
 618 QIR_NODST_2(TEX_R)
 619 QIR_NODST_2(TEX_B)
 620 QIR_NODST_2(TEX_DIRECT)
 621 QIR_ALU0(FRAG_X)
 622 QIR_ALU0(FRAG_Y)
 623 QIR_ALU0(FRAG_Z)
 624 QIR_ALU0(FRAG_W)
 625 QIR_ALU0(FRAG_REV_FLAG)
 626 QIR_ALU0(TEX_RESULT)
 627 QIR_ALU0(TLB_COLOR_READ)
 628 QIR_NODST_1(TLB_COLOR_WRITE)
 629 QIR_NODST_1(TLB_Z_WRITE)
 630 QIR_NODST_1(TLB_DISCARD_SETUP)
 631 QIR_NODST_1(TLB_STENCIL_SETUP)
 632
 633 static inline struct qreg
 634 qir_UNPACK_8_F(struct vc4_compile *c, struct qreg src, int i)
 635 {
 636         struct qreg t = qir_get_temp(c);
 637         qir_emit(c, qir_inst(QOP_UNPACK_8A_F + i, t, src, c->undef));
 638         return t;
 639 }
 640
 641 static inline struct qreg
 642 qir_UNPACK_8_I(struct vc4_compile *c, struct qreg src, int i)
 643 {
 644         struct qreg t = qir_get_temp(c);
 645         qir_emit(c, qir_inst(QOP_UNPACK_8A_I + i, t, src, c->undef));
 646         return t;
 647 }
 648
 649 static inline struct qreg
 650 qir_UNPACK_16_F(struct vc4_compile *c, struct qreg src, int i)
 651 {
 652         struct qreg t = qir_get_temp(c);
 653         qir_emit(c, qir_inst(QOP_UNPACK_16A_F + i, t, src, c->undef));
 654         return t;
 655 }
 656
 657 static inline struct qreg
 658 qir_UNPACK_16_I(struct vc4_compile *c, struct qreg src, int i)
 659 {
 660         struct qreg t = qir_get_temp(c);
 661         qir_emit(c, qir_inst(QOP_UNPACK_16A_I + i, t, src, c->undef));
 662         return t;
 663 }
 664
 665 static inline struct qreg
 666 qir_PACK_8_F(struct vc4_compile *c, struct qreg dest, struct qreg val, int chan)
 667 {
 668         qir_emit(c, qir_inst(QOP_PACK_8A_F + chan, dest, val, c->undef));
 669         if (dest.file == QFILE_TEMP)
 670                 c->defs[dest.index] = NULL;
 671         return dest;
 672 }
 673
 674 static inline struct qreg
 675 qir_POW(struct vc4_compile *c, struct qreg x, struct qreg y)
 676 {
 677         return qir_EXP2(c, qir_FMUL(c,
 678                                     y,
 679                                     qir_LOG2(c, x)));
 680 }
 681
 682 static inline void
 683 qir_VPM_WRITE(struct vc4_compile *c, struct qreg val)
 684 {
 685         static const struct qreg vpm = { QFILE_VPM, 0 };
 686         qir_emit(c, qir_inst(QOP_MOV, vpm, val, c->undef));
 687 }
 688
 689 #endif /* VC4_QIR_H */