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 };
  62
  63 enum qop {
  64         QOP_UNDEF,
  65         QOP_MOV,
  66         QOP_FADD,
  67         QOP_FSUB,
  68         QOP_FMUL,
  69         QOP_MUL24,
  70         QOP_FMIN,
  71         QOP_FMAX,
  72         QOP_FMINABS,
  73         QOP_FMAXABS,
  74         QOP_ADD,
  75         QOP_SUB,
  76         QOP_SHL,
  77         QOP_SHR,
  78         QOP_ASR,
  79         QOP_MIN,
  80         QOP_MAX,
  81         QOP_AND,
  82         QOP_OR,
  83         QOP_XOR,
  84         QOP_NOT,
  85
  86         /* Note: Orderings of these compares must be the same as in
  87          * qpu_defines.h.  Selects the src[0] if the ns flag bit is set,
  88          * otherwise 0. */
  89         QOP_SEL_X_0_ZS,
  90         QOP_SEL_X_0_ZC,
  91         QOP_SEL_X_0_NS,
  92         QOP_SEL_X_0_NC,
  93         /* Selects the src[0] if the ns flag bit is set, otherwise src[1]. */
  94         QOP_SEL_X_Y_ZS,
  95         QOP_SEL_X_Y_ZC,
  96         QOP_SEL_X_Y_NS,
  97         QOP_SEL_X_Y_NC,
  98
  99         QOP_FTOI,
 100         QOP_ITOF,
 101         QOP_RCP,
 102         QOP_RSQ,
 103         QOP_EXP2,
 104         QOP_LOG2,
 105         QOP_VW_SETUP,
 106         QOP_VR_SETUP,
 107         QOP_PACK_8888_F,
 108         QOP_PACK_8A_F,
 109         QOP_PACK_8B_F,
 110         QOP_PACK_8C_F,
 111         QOP_PACK_8D_F,
 112         QOP_PACK_16A_I,
 113         QOP_PACK_16B_I,
 114         QOP_TLB_DISCARD_SETUP,
 115         QOP_TLB_STENCIL_SETUP,
 116         QOP_TLB_Z_WRITE,
 117         QOP_TLB_COLOR_WRITE,
 118         QOP_TLB_COLOR_READ,
 119         QOP_VARY_ADD_C,
 120
 121         QOP_FRAG_X,
 122         QOP_FRAG_Y,
 123         QOP_FRAG_Z,
 124         QOP_FRAG_W,
 125         QOP_FRAG_REV_FLAG,
 126
 127         QOP_UNPACK_8A_F,
 128         QOP_UNPACK_8B_F,
 129         QOP_UNPACK_8C_F,
 130         QOP_UNPACK_8D_F,
 131         QOP_UNPACK_16A_F,
 132         QOP_UNPACK_16B_F,
 133
 134         QOP_UNPACK_8A_I,
 135         QOP_UNPACK_8B_I,
 136         QOP_UNPACK_8C_I,
 137         QOP_UNPACK_8D_I,
 138         QOP_UNPACK_16A_I,
 139         QOP_UNPACK_16B_I,
 140
 141         /** Texture x coordinate parameter write */
 142         QOP_TEX_S,
 143         /** Texture y coordinate parameter write */
 144         QOP_TEX_T,
 145         /** Texture border color parameter or cube map z coordinate write */
 146         QOP_TEX_R,
 147         /** Texture LOD bias parameter write */
 148         QOP_TEX_B,
 149
 150         /**
 151          * Texture-unit 4-byte read with address provided direct in S
 152          * cooordinate.
 153          *
 154          * The first operand is the offset from the start of the UBO, and the
 155          * second is the uniform that has the UBO's base pointer.
 156          */
 157         QOP_TEX_DIRECT,
 158
 159         /**
 160          * Signal of texture read being necessary and then reading r4 into
 161          * the destination
 162          */
 163         QOP_TEX_RESULT,
 164 };
 165
 166 struct queued_qpu_inst {
 167         struct list_head link;
 168         uint64_t inst;
 169 };
 170
 171 struct qinst {
 172         struct list_head link;
 173
 174         enum qop op;
 175         struct qreg dst;
 176         struct qreg *src;
 177         bool sf;
 178 };
 179
 180 enum qstage {
 181         /**
 182          * Coordinate shader, runs during binning, before the VS, and just
 183          * outputs position.
 184          */
 185         QSTAGE_COORD,
 186         QSTAGE_VERT,
 187         QSTAGE_FRAG,
 188 };
 189
 190 enum quniform_contents {
 191         /**
 192          * Indicates that a constant 32-bit value is copied from the program's
 193          * uniform contents.
 194          */
 195         QUNIFORM_CONSTANT,
 196         /**
 197          * Indicates that the program's uniform contents are used as an index
 198          * into the GL uniform storage.
 199          */
 200         QUNIFORM_UNIFORM,
 201
 202         /** @{
 203          * Scaling factors from clip coordinates to relative to the viewport
 204          * center.
 205          *
 206          * This is used by the coordinate and vertex shaders to produce the
 207          * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed
 208          * point offsets from the viewport ccenter.
 209          */
 210         QUNIFORM_VIEWPORT_X_SCALE,
 211         QUNIFORM_VIEWPORT_Y_SCALE,
 212         /** @} */
 213
 214         QUNIFORM_VIEWPORT_Z_OFFSET,
 215         QUNIFORM_VIEWPORT_Z_SCALE,
 216
 217         QUNIFORM_USER_CLIP_PLANE,
 218
 219         /**
 220          * A reference to a texture config parameter 0 uniform.
 221          *
 222          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 223          * defines texture type, miplevels, and such.  It will be found as a
 224          * parameter to the first QOP_TEX_[STRB] instruction in a sequence.
 225          */
 226         QUNIFORM_TEXTURE_CONFIG_P0,
 227
 228         /**
 229          * A reference to a texture config parameter 1 uniform.
 230          *
 231          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 232          * defines texture width, height, filters, and wrap modes.  It will be
 233          * found as a parameter to the second QOP_TEX_[STRB] instruction in a
 234          * sequence.
 235          */
 236         QUNIFORM_TEXTURE_CONFIG_P1,
 237
 238         /** A reference to a texture config parameter 2 cubemap stride uniform */
 239         QUNIFORM_TEXTURE_CONFIG_P2,
 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_BLEND_CONST_COLOR_X,
 249         QUNIFORM_BLEND_CONST_COLOR_Y,
 250         QUNIFORM_BLEND_CONST_COLOR_Z,
 251         QUNIFORM_BLEND_CONST_COLOR_W,
 252
 253         QUNIFORM_STENCIL,
 254
 255         QUNIFORM_ALPHA_REF,
 256 };
 257
 258 struct vc4_varying_semantic {
 259         uint8_t semantic;
 260         uint8_t index;
 261         uint8_t swizzle;
 262 };
 263
 264 struct vc4_compiler_ubo_range {
 265         /**
 266          * offset in bytes from the start of the ubo where this range is
 267          * uploaded.
 268          *
 269          * Only set once used is set.
 270          */
 271         uint32_t dst_offset;
 272
 273         /**
 274          * offset in bytes from the start of the gallium uniforms where the
 275          * data comes from.
 276          */
 277         uint32_t src_offset;
 278
 279         /** size in bytes of this ubo range */
 280         uint32_t size;
 281
 282         /**
 283          * Set if this range is used by the shader for indirect uniforms
 284          * access.
 285          */
 286         bool used;
 287 };
 288
 289 struct vc4_key {
 290         struct vc4_uncompiled_shader *shader_state;
 291         struct {
 292                 enum pipe_format format;
 293                 unsigned compare_mode:1;
 294                 unsigned compare_func:3;
 295                 unsigned wrap_s:3;
 296                 unsigned wrap_t:3;
 297                 uint8_t swizzle[4];
 298         } tex[VC4_MAX_TEXTURE_SAMPLERS];
 299         uint8_t ucp_enables;
 300 };
 301
 302 struct vc4_fs_key {
 303         struct vc4_key base;
 304         enum pipe_format color_format;
 305         bool depth_enabled;
 306         bool stencil_enabled;
 307         bool stencil_twoside;
 308         bool stencil_full_writemasks;
 309         bool is_points;
 310         bool is_lines;
 311         bool alpha_test;
 312         bool point_coord_upper_left;
 313         bool light_twoside;
 314         uint8_t alpha_test_func;
 315         uint8_t logicop_func;
 316         uint32_t point_sprite_mask;
 317
 318         struct pipe_rt_blend_state blend;
 319 };
 320
 321 struct vc4_vs_key {
 322         struct vc4_key base;
 323
 324         /**
 325          * This is a proxy for the array of FS input semantics, which is
 326          * larger than we would want to put in the key.
 327          */
 328         uint64_t compiled_fs_id;
 329
 330         enum pipe_format attr_formats[8];
 331         bool is_coord;
 332         bool per_vertex_point_size;
 333 };
 334
 335 struct vc4_compile {
 336         struct vc4_context *vc4;
 337         nir_shader *s;
 338         nir_function_impl *impl;
 339         struct exec_list *cf_node_list;
 340
 341         /**
 342          * Mapping from nir_register * or nir_ssa_def * to array of struct
 343          * qreg for the values.
 344          */
 345         struct hash_table *def_ht;
 346
 347         /* For each temp, the instruction generating its value. */
 348         struct qinst **defs;
 349         uint32_t defs_array_size;
 350
 351         /**
 352          * Inputs to the shader, arranged by TGSI declaration order.
 353          *
 354          * Not all fragment shader QFILE_VARY reads are present in this array.
 355          */
 356         struct qreg *inputs;
 357         struct qreg *outputs;
 358         uint32_t inputs_array_size;
 359         uint32_t outputs_array_size;
 360         uint32_t uniforms_array_size;
 361
 362         struct vc4_compiler_ubo_range *ubo_ranges;
 363         uint32_t ubo_ranges_array_size;
 364         /** Number of uniform areas declared in ubo_ranges. */
 365         uint32_t num_uniform_ranges;
 366         /** Number of uniform areas used for indirect addressed loads. */
 367         uint32_t num_ubo_ranges;
 368         uint32_t next_ubo_dst_offset;
 369
 370         struct qreg line_x, point_x, point_y;
 371         struct qreg discard;
 372
 373         uint8_t vattr_sizes[8];
 374
 375         /**
 376          * Array of the TGSI semantics of all FS QFILE_VARY reads.
 377          *
 378          * This includes those that aren't part of the VPM varyings, like
 379          * point/line coordinates.
 380          */
 381         struct vc4_varying_semantic *input_semantics;
 382         uint32_t num_input_semantics;
 383         uint32_t input_semantics_array_size;
 384
 385         /**
 386          * An entry per outputs[] in the VS indicating what the semantic of
 387          * the output is.  Used to emit from the VS in the order that the FS
 388          * needs.
 389          */
 390         struct vc4_varying_semantic *output_semantics;
 391
 392         struct pipe_shader_state *shader_state;
 393         struct vc4_key *key;
 394         struct vc4_fs_key *fs_key;
 395         struct vc4_vs_key *vs_key;
 396
 397         uint32_t *uniform_data;
 398         enum quniform_contents *uniform_contents;
 399         uint32_t uniform_array_size;
 400         uint32_t num_uniforms;
 401         uint32_t num_outputs;
 402         uint32_t num_texture_samples;
 403         uint32_t output_position_index;
 404         uint32_t output_clipvertex_index;
 405         uint32_t output_color_index;
 406         uint32_t output_point_size_index;
 407
 408         struct qreg undef;
 409         enum qstage stage;
 410         uint32_t num_temps;
 411         struct list_head instructions;
 412         uint32_t immediates[1024];
 413
 414         struct list_head qpu_inst_list;
 415         uint64_t *qpu_insts;
 416         uint32_t qpu_inst_count;
 417         uint32_t qpu_inst_size;
 418         uint32_t num_inputs;
 419
 420         uint32_t program_id;
 421         uint32_t variant_id;
 422 };
 423
 424 /* Special nir_load_input intrinsic index for loading the current TLB
 425  * destination color.
 426  */
 427 #define VC4_NIR_TLB_COLOR_READ_INPUT            2000000000
 428
 429 /* Special offset for nir_load_uniform values to get a QUNIFORM_*
 430  * state-dependent value.
 431  */
 432 #define VC4_NIR_STATE_UNIFORM_OFFSET            2000000000
 433
 434 struct vc4_compile *qir_compile_init(void);
 435 void qir_compile_destroy(struct vc4_compile *c);
 436 struct qinst *qir_inst(enum qop op, struct qreg dst,
 437                        struct qreg src0, struct qreg src1);
 438 struct qinst *qir_inst4(enum qop op, struct qreg dst,
 439                         struct qreg a,
 440                         struct qreg b,
 441                         struct qreg c,
 442                         struct qreg d);
 443 void qir_remove_instruction(struct vc4_compile *c, struct qinst *qinst);
 444 struct qreg qir_uniform(struct vc4_compile *c,
 445                         enum quniform_contents contents,
 446                         uint32_t data);
 447 void qir_reorder_uniforms(struct vc4_compile *c);
 448 void qir_emit(struct vc4_compile *c, struct qinst *inst);
 449 struct qreg qir_get_temp(struct vc4_compile *c);
 450 int qir_get_op_nsrc(enum qop qop);
 451 bool qir_reg_equals(struct qreg a, struct qreg b);
 452 bool qir_has_side_effects(struct vc4_compile *c, struct qinst *inst);
 453 bool qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst);
 454 bool qir_is_multi_instruction(struct qinst *inst);
 455 bool qir_is_tex(struct qinst *inst);
 456 bool qir_depends_on_flags(struct qinst *inst);
 457 bool qir_writes_r4(struct qinst *inst);
 458 bool qir_src_needs_a_file(struct qinst *inst);
 459 struct qreg qir_follow_movs(struct vc4_compile *c, struct qreg reg);
 460
 461 void qir_dump(struct vc4_compile *c);
 462 void qir_dump_inst(struct vc4_compile *c, struct qinst *inst);
 463 const char *qir_get_stage_name(enum qstage stage);
 464
 465 void qir_optimize(struct vc4_compile *c);
 466 bool qir_opt_algebraic(struct vc4_compile *c);
 467 bool qir_opt_constant_folding(struct vc4_compile *c);
 468 bool qir_opt_copy_propagation(struct vc4_compile *c);
 469 bool qir_opt_cse(struct vc4_compile *c);
 470 bool qir_opt_dead_code(struct vc4_compile *c);
 471 bool qir_opt_small_immediates(struct vc4_compile *c);
 472 bool qir_opt_vpm_writes(struct vc4_compile *c);
 473 void vc4_nir_lower_blend(struct vc4_compile *c);
 474 void vc4_nir_lower_io(struct vc4_compile *c);
 475 nir_ssa_def *vc4_nir_get_state_uniform(struct nir_builder *b,
 476                                        enum quniform_contents contents);
 477 nir_ssa_def *vc4_nir_get_swizzled_channel(struct nir_builder *b,
 478                                           nir_ssa_def **srcs, int swiz);
 479 void qir_lower_uniforms(struct vc4_compile *c);
 480
 481 void qpu_schedule_instructions(struct vc4_compile *c);
 482
 483 void qir_SF(struct vc4_compile *c, struct qreg src);
 484
 485 static inline struct qreg
 486 qir_uniform_ui(struct vc4_compile *c, uint32_t ui)
 487 {
 488         return qir_uniform(c, QUNIFORM_CONSTANT, ui);
 489 }
 490
 491 static inline struct qreg
 492 qir_uniform_f(struct vc4_compile *c, float f)
 493 {
 494         return qir_uniform(c, QUNIFORM_CONSTANT, fui(f));
 495 }
 496
 497 #define QIR_ALU0(name)                                                   \
 498 static inline struct qreg                                                \
 499 qir_##name(struct vc4_compile *c)                                        \
 500 {                                                                        \
 501         struct qreg t = qir_get_temp(c);                                 \
 502         qir_emit(c, qir_inst(QOP_##name, t, c->undef, c->undef));        \
 503         return t;                                                        \
 504 }
 505
 506 #define QIR_ALU1(name)                                                   \
 507 static inline struct qreg                                                \
 508 qir_##name(struct vc4_compile *c, struct qreg a)                         \
 509 {                                                                        \
 510         struct qreg t = qir_get_temp(c);                                 \
 511         qir_emit(c, qir_inst(QOP_##name, t, a, c->undef));               \
 512         return t;                                                        \
 513 }
 514
 515 #define QIR_ALU2(name)                                                   \
 516 static inline struct qreg                                                \
 517 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b)          \
 518 {                                                                        \
 519         struct qreg t = qir_get_temp(c);                                 \
 520         qir_emit(c, qir_inst(QOP_##name, t, a, b));                      \
 521         return t;                                                        \
 522 }
 523
 524 #define QIR_NODST_1(name)                                               \
 525 static inline void                                                      \
 526 qir_##name(struct vc4_compile *c, struct qreg a)                        \
 527 {                                                                       \
 528         qir_emit(c, qir_inst(QOP_##name, c->undef, a, c->undef));       \
 529 }
 530
 531 #define QIR_NODST_2(name)                                               \
 532 static inline void                                                      \
 533 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b)         \
 534 {                                                                       \
 535         qir_emit(c, qir_inst(QOP_##name, c->undef, a, b));       \
 536 }
 537
 538 #define QIR_PACK(name)                                                   \
 539 static inline struct qreg                                                \
 540 qir_##name(struct vc4_compile *c, struct qreg dest, struct qreg a)       \
 541 {                                                                        \
 542         qir_emit(c, qir_inst(QOP_##name, dest, a, c->undef));            \
 543         if (dest.file == QFILE_TEMP)                                     \
 544                 c->defs[dest.index] = NULL;                              \
 545         return dest;                                                     \
 546 }
 547
 548 QIR_ALU1(MOV)
 549 QIR_ALU2(FADD)
 550 QIR_ALU2(FSUB)
 551 QIR_ALU2(FMUL)
 552 QIR_ALU2(MUL24)
 553 QIR_ALU1(SEL_X_0_ZS)
 554 QIR_ALU1(SEL_X_0_ZC)
 555 QIR_ALU1(SEL_X_0_NS)
 556 QIR_ALU1(SEL_X_0_NC)
 557 QIR_ALU2(SEL_X_Y_ZS)
 558 QIR_ALU2(SEL_X_Y_ZC)
 559 QIR_ALU2(SEL_X_Y_NS)
 560 QIR_ALU2(SEL_X_Y_NC)
 561 QIR_ALU2(FMIN)
 562 QIR_ALU2(FMAX)
 563 QIR_ALU2(FMINABS)
 564 QIR_ALU2(FMAXABS)
 565 QIR_ALU1(FTOI)
 566 QIR_ALU1(ITOF)
 567
 568 QIR_ALU2(ADD)
 569 QIR_ALU2(SUB)
 570 QIR_ALU2(SHL)
 571 QIR_ALU2(SHR)
 572 QIR_ALU2(ASR)
 573 QIR_ALU2(MIN)
 574 QIR_ALU2(MAX)
 575 QIR_ALU2(AND)
 576 QIR_ALU2(OR)
 577 QIR_ALU2(XOR)
 578 QIR_ALU1(NOT)
 579
 580 QIR_ALU1(RCP)
 581 QIR_ALU1(RSQ)
 582 QIR_ALU1(EXP2)
 583 QIR_ALU1(LOG2)
 584 QIR_ALU1(PACK_8888_F)
 585 QIR_PACK(PACK_8A_F)
 586 QIR_PACK(PACK_8B_F)
 587 QIR_PACK(PACK_8C_F)
 588 QIR_PACK(PACK_8D_F)
 589 QIR_PACK(PACK_16A_I)
 590 QIR_PACK(PACK_16B_I)
 591 QIR_ALU1(VARY_ADD_C)
 592 QIR_NODST_2(TEX_S)
 593 QIR_NODST_2(TEX_T)
 594 QIR_NODST_2(TEX_R)
 595 QIR_NODST_2(TEX_B)
 596 QIR_NODST_2(TEX_DIRECT)
 597 QIR_ALU0(FRAG_X)
 598 QIR_ALU0(FRAG_Y)
 599 QIR_ALU0(FRAG_Z)
 600 QIR_ALU0(FRAG_W)
 601 QIR_ALU0(FRAG_REV_FLAG)
 602 QIR_ALU0(TEX_RESULT)
 603 QIR_ALU0(TLB_COLOR_READ)
 604 QIR_NODST_1(TLB_COLOR_WRITE)
 605 QIR_NODST_1(TLB_Z_WRITE)
 606 QIR_NODST_1(TLB_DISCARD_SETUP)
 607 QIR_NODST_1(TLB_STENCIL_SETUP)
 608
 609 static inline struct qreg
 610 qir_UNPACK_8_F(struct vc4_compile *c, struct qreg src, int i)
 611 {
 612         struct qreg t = qir_get_temp(c);
 613         qir_emit(c, qir_inst(QOP_UNPACK_8A_F + i, t, src, c->undef));
 614         return t;
 615 }
 616
 617 static inline struct qreg
 618 qir_UNPACK_8_I(struct vc4_compile *c, struct qreg src, int i)
 619 {
 620         struct qreg t = qir_get_temp(c);
 621         qir_emit(c, qir_inst(QOP_UNPACK_8A_I + i, t, src, c->undef));
 622         return t;
 623 }
 624
 625 static inline struct qreg
 626 qir_UNPACK_16_F(struct vc4_compile *c, struct qreg src, int i)
 627 {
 628         struct qreg t = qir_get_temp(c);
 629         qir_emit(c, qir_inst(QOP_UNPACK_16A_F + i, t, src, c->undef));
 630         return t;
 631 }
 632
 633 static inline struct qreg
 634 qir_UNPACK_16_I(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_16A_I + i, t, src, c->undef));
 638         return t;
 639 }
 640
 641 static inline struct qreg
 642 qir_PACK_8_F(struct vc4_compile *c, struct qreg dest, struct qreg val, int chan)
 643 {
 644         qir_emit(c, qir_inst(QOP_PACK_8A_F + chan, dest, val, c->undef));
 645         if (dest.file == QFILE_TEMP)
 646                 c->defs[dest.index] = NULL;
 647         return dest;
 648 }
 649
 650 static inline struct qreg
 651 qir_POW(struct vc4_compile *c, struct qreg x, struct qreg y)
 652 {
 653         return qir_EXP2(c, qir_FMUL(c,
 654                                     y,
 655                                     qir_LOG2(c, x)));
 656 }
 657
 658 static inline void
 659 qir_VPM_WRITE(struct vc4_compile *c, struct qreg val)
 660 {
 661         static const struct qreg vpm = { QFILE_VPM, 0 };
 662         qir_emit(c, qir_inst(QOP_MOV, vpm, val, c->undef));
 663 }
 664
 665 #endif /* VC4_QIR_H */