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 enum qfile {
  43         QFILE_NULL,
  44         QFILE_TEMP,
  45         QFILE_VARY,
  46         QFILE_UNIF,
  47         QFILE_VPM,
  48
  49         /**
  50          * Stores an immediate value in the index field that can be turned
  51          * into a small immediate field by qpu_encode_small_immediate().
  52          */
  53         QFILE_SMALL_IMM,
  54 };
  55
  56 struct qreg {
  57         enum qfile file;
  58         uint32_t index;
  59 };
  60
  61 enum qop {
  62         QOP_UNDEF,
  63         QOP_MOV,
  64         QOP_FADD,
  65         QOP_FSUB,
  66         QOP_FMUL,
  67         QOP_MUL24,
  68         QOP_FMIN,
  69         QOP_FMAX,
  70         QOP_FMINABS,
  71         QOP_FMAXABS,
  72         QOP_ADD,
  73         QOP_SUB,
  74         QOP_SHL,
  75         QOP_SHR,
  76         QOP_ASR,
  77         QOP_MIN,
  78         QOP_MAX,
  79         QOP_AND,
  80         QOP_OR,
  81         QOP_XOR,
  82         QOP_NOT,
  83
  84         /* Note: Orderings of these compares must be the same as in
  85          * qpu_defines.h.  Selects the src[0] if the ns flag bit is set,
  86          * otherwise 0. */
  87         QOP_SEL_X_0_ZS,
  88         QOP_SEL_X_0_ZC,
  89         QOP_SEL_X_0_NS,
  90         QOP_SEL_X_0_NC,
  91         /* Selects the src[0] if the ns flag bit is set, otherwise src[1]. */
  92         QOP_SEL_X_Y_ZS,
  93         QOP_SEL_X_Y_ZC,
  94         QOP_SEL_X_Y_NS,
  95         QOP_SEL_X_Y_NC,
  96
  97         QOP_FTOI,
  98         QOP_ITOF,
  99         QOP_RCP,
 100         QOP_RSQ,
 101         QOP_EXP2,
 102         QOP_LOG2,
 103         QOP_VW_SETUP,
 104         QOP_VR_SETUP,
 105         QOP_PACK_SCALED,
 106         QOP_PACK_8888_F,
 107         QOP_PACK_8A_F,
 108         QOP_PACK_8B_F,
 109         QOP_PACK_8C_F,
 110         QOP_PACK_8D_F,
 111         QOP_TLB_DISCARD_SETUP,
 112         QOP_TLB_STENCIL_SETUP,
 113         QOP_TLB_Z_WRITE,
 114         QOP_TLB_COLOR_WRITE,
 115         QOP_TLB_COLOR_READ,
 116         QOP_VARY_ADD_C,
 117
 118         QOP_FRAG_X,
 119         QOP_FRAG_Y,
 120         QOP_FRAG_Z,
 121         QOP_FRAG_W,
 122         QOP_FRAG_REV_FLAG,
 123
 124         QOP_UNPACK_8A_F,
 125         QOP_UNPACK_8B_F,
 126         QOP_UNPACK_8C_F,
 127         QOP_UNPACK_8D_F,
 128         QOP_UNPACK_16A_F,
 129         QOP_UNPACK_16B_F,
 130
 131         QOP_UNPACK_8A_I,
 132         QOP_UNPACK_8B_I,
 133         QOP_UNPACK_8C_I,
 134         QOP_UNPACK_8D_I,
 135         QOP_UNPACK_16A_I,
 136         QOP_UNPACK_16B_I,
 137
 138         /** Texture x coordinate parameter write */
 139         QOP_TEX_S,
 140         /** Texture y coordinate parameter write */
 141         QOP_TEX_T,
 142         /** Texture border color parameter or cube map z coordinate write */
 143         QOP_TEX_R,
 144         /** Texture LOD bias parameter write */
 145         QOP_TEX_B,
 146
 147         /**
 148          * Texture-unit 4-byte read with address provided direct in S
 149          * cooordinate.
 150          *
 151          * The first operand is the offset from the start of the UBO, and the
 152          * second is the uniform that has the UBO's base pointer.
 153          */
 154         QOP_TEX_DIRECT,
 155
 156         /**
 157          * Signal of texture read being necessary and then reading r4 into
 158          * the destination
 159          */
 160         QOP_TEX_RESULT,
 161 };
 162
 163 struct queued_qpu_inst {
 164         struct list_head link;
 165         uint64_t inst;
 166 };
 167
 168 struct qinst {
 169         struct list_head link;
 170
 171         enum qop op;
 172         struct qreg dst;
 173         struct qreg *src;
 174         bool sf;
 175 };
 176
 177 enum qstage {
 178         /**
 179          * Coordinate shader, runs during binning, before the VS, and just
 180          * outputs position.
 181          */
 182         QSTAGE_COORD,
 183         QSTAGE_VERT,
 184         QSTAGE_FRAG,
 185 };
 186
 187 enum quniform_contents {
 188         /**
 189          * Indicates that a constant 32-bit value is copied from the program's
 190          * uniform contents.
 191          */
 192         QUNIFORM_CONSTANT,
 193         /**
 194          * Indicates that the program's uniform contents are used as an index
 195          * into the GL uniform storage.
 196          */
 197         QUNIFORM_UNIFORM,
 198
 199         /** @{
 200          * Scaling factors from clip coordinates to relative to the viewport
 201          * center.
 202          *
 203          * This is used by the coordinate and vertex shaders to produce the
 204          * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed
 205          * point offsets from the viewport ccenter.
 206          */
 207         QUNIFORM_VIEWPORT_X_SCALE,
 208         QUNIFORM_VIEWPORT_Y_SCALE,
 209         /** @} */
 210
 211         QUNIFORM_VIEWPORT_Z_OFFSET,
 212         QUNIFORM_VIEWPORT_Z_SCALE,
 213
 214         QUNIFORM_USER_CLIP_PLANE,
 215
 216         /**
 217          * A reference to a texture config parameter 0 uniform.
 218          *
 219          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 220          * defines texture type, miplevels, and such.  It will be found as a
 221          * parameter to the first QOP_TEX_[STRB] instruction in a sequence.
 222          */
 223         QUNIFORM_TEXTURE_CONFIG_P0,
 224
 225         /**
 226          * A reference to a texture config parameter 1 uniform.
 227          *
 228          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 229          * defines texture width, height, filters, and wrap modes.  It will be
 230          * found as a parameter to the second QOP_TEX_[STRB] instruction in a
 231          * sequence.
 232          */
 233         QUNIFORM_TEXTURE_CONFIG_P1,
 234
 235         /** A reference to a texture config parameter 2 cubemap stride uniform */
 236         QUNIFORM_TEXTURE_CONFIG_P2,
 237
 238         QUNIFORM_UBO_ADDR,
 239
 240         QUNIFORM_TEXRECT_SCALE_X,
 241         QUNIFORM_TEXRECT_SCALE_Y,
 242
 243         QUNIFORM_TEXTURE_BORDER_COLOR,
 244
 245         QUNIFORM_BLEND_CONST_COLOR,
 246         QUNIFORM_STENCIL,
 247
 248         QUNIFORM_ALPHA_REF,
 249 };
 250
 251 struct vc4_varying_semantic {
 252         uint8_t semantic;
 253         uint8_t index;
 254         uint8_t swizzle;
 255 };
 256
 257 struct vc4_compiler_ubo_range {
 258         /**
 259          * offset in bytes from the start of the ubo where this range is
 260          * uploaded.
 261          *
 262          * Only set once used is set.
 263          */
 264         uint32_t dst_offset;
 265
 266         /**
 267          * offset in bytes from the start of the gallium uniforms where the
 268          * data comes from.
 269          */
 270         uint32_t src_offset;
 271
 272         /** size in bytes of this ubo range */
 273         uint32_t size;
 274
 275         /**
 276          * Set if this range is used by the shader for indirect uniforms
 277          * access.
 278          */
 279         bool used;
 280 };
 281
 282 struct vc4_key {
 283         struct vc4_uncompiled_shader *shader_state;
 284         struct {
 285                 enum pipe_format format;
 286                 unsigned compare_mode:1;
 287                 unsigned compare_func:3;
 288                 unsigned wrap_s:3;
 289                 unsigned wrap_t:3;
 290                 uint8_t swizzle[4];
 291         } tex[VC4_MAX_TEXTURE_SAMPLERS];
 292         uint8_t ucp_enables;
 293 };
 294
 295 struct vc4_fs_key {
 296         struct vc4_key base;
 297         enum pipe_format color_format;
 298         bool depth_enabled;
 299         bool stencil_enabled;
 300         bool stencil_twoside;
 301         bool stencil_full_writemasks;
 302         bool is_points;
 303         bool is_lines;
 304         bool alpha_test;
 305         bool point_coord_upper_left;
 306         bool light_twoside;
 307         uint8_t alpha_test_func;
 308         uint8_t logicop_func;
 309         uint32_t point_sprite_mask;
 310
 311         struct pipe_rt_blend_state blend;
 312 };
 313
 314 struct vc4_vs_key {
 315         struct vc4_key base;
 316
 317         /**
 318          * This is a proxy for the array of FS input semantics, which is
 319          * larger than we would want to put in the key.
 320          */
 321         uint64_t compiled_fs_id;
 322
 323         enum pipe_format attr_formats[8];
 324         bool is_coord;
 325         bool per_vertex_point_size;
 326 };
 327
 328 struct vc4_compile {
 329         struct vc4_context *vc4;
 330         nir_shader *s;
 331         nir_function_impl *impl;
 332         struct exec_list *cf_node_list;
 333
 334         /**
 335          * Mapping from nir_register * or nir_ssa_def * to array of struct
 336          * qreg for the values.
 337          */
 338         struct hash_table *def_ht;
 339
 340         /* For each temp, the instruction generating its value. */
 341         struct qinst **defs;
 342         uint32_t defs_array_size;
 343
 344         /**
 345          * Inputs to the shader, arranged by TGSI declaration order.
 346          *
 347          * Not all fragment shader QFILE_VARY reads are present in this array.
 348          */
 349         struct qreg *inputs;
 350         struct qreg *outputs;
 351         uint32_t inputs_array_size;
 352         uint32_t outputs_array_size;
 353         uint32_t uniforms_array_size;
 354
 355         struct vc4_compiler_ubo_range *ubo_ranges;
 356         uint32_t ubo_ranges_array_size;
 357         /** Number of uniform areas declared in ubo_ranges. */
 358         uint32_t num_uniform_ranges;
 359         /** Number of uniform areas used for indirect addressed loads. */
 360         uint32_t num_ubo_ranges;
 361         uint32_t next_ubo_dst_offset;
 362
 363         struct qreg line_x, point_x, point_y;
 364         struct qreg discard;
 365
 366         uint8_t vattr_sizes[8];
 367
 368         /**
 369          * Array of the TGSI semantics of all FS QFILE_VARY reads.
 370          *
 371          * This includes those that aren't part of the VPM varyings, like
 372          * point/line coordinates.
 373          */
 374         struct vc4_varying_semantic *input_semantics;
 375         uint32_t num_input_semantics;
 376         uint32_t input_semantics_array_size;
 377
 378         /**
 379          * An entry per outputs[] in the VS indicating what the semantic of
 380          * the output is.  Used to emit from the VS in the order that the FS
 381          * needs.
 382          */
 383         struct vc4_varying_semantic *output_semantics;
 384
 385         struct pipe_shader_state *shader_state;
 386         struct vc4_key *key;
 387         struct vc4_fs_key *fs_key;
 388         struct vc4_vs_key *vs_key;
 389
 390         uint32_t *uniform_data;
 391         enum quniform_contents *uniform_contents;
 392         uint32_t uniform_array_size;
 393         uint32_t num_uniforms;
 394         uint32_t num_outputs;
 395         uint32_t num_texture_samples;
 396         uint32_t output_position_index;
 397         uint32_t output_clipvertex_index;
 398         uint32_t output_color_index;
 399         uint32_t output_point_size_index;
 400
 401         struct qreg undef;
 402         enum qstage stage;
 403         uint32_t num_temps;
 404         struct list_head instructions;
 405         uint32_t immediates[1024];
 406
 407         struct list_head qpu_inst_list;
 408         uint64_t *qpu_insts;
 409         uint32_t qpu_inst_count;
 410         uint32_t qpu_inst_size;
 411         uint32_t num_inputs;
 412
 413         uint32_t program_id;
 414         uint32_t variant_id;
 415 };
 416
 417 /* Special offset for nir_load_uniform values to get a QUNIFORM_*
 418  * state-dependent value.
 419  */
 420 #define VC4_NIR_STATE_UNIFORM_OFFSET            2000000000
 421
 422 struct vc4_compile *qir_compile_init(void);
 423 void qir_compile_destroy(struct vc4_compile *c);
 424 struct qinst *qir_inst(enum qop op, struct qreg dst,
 425                        struct qreg src0, struct qreg src1);
 426 struct qinst *qir_inst4(enum qop op, struct qreg dst,
 427                         struct qreg a,
 428                         struct qreg b,
 429                         struct qreg c,
 430                         struct qreg d);
 431 void qir_remove_instruction(struct vc4_compile *c, struct qinst *qinst);
 432 struct qreg qir_uniform(struct vc4_compile *c,
 433                         enum quniform_contents contents,
 434                         uint32_t data);
 435 void qir_reorder_uniforms(struct vc4_compile *c);
 436 void qir_emit(struct vc4_compile *c, struct qinst *inst);
 437 struct qreg qir_get_temp(struct vc4_compile *c);
 438 int qir_get_op_nsrc(enum qop qop);
 439 bool qir_reg_equals(struct qreg a, struct qreg b);
 440 bool qir_has_side_effects(struct vc4_compile *c, struct qinst *inst);
 441 bool qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst);
 442 bool qir_is_multi_instruction(struct qinst *inst);
 443 bool qir_is_tex(struct qinst *inst);
 444 bool qir_depends_on_flags(struct qinst *inst);
 445 bool qir_writes_r4(struct qinst *inst);
 446 bool qir_src_needs_a_file(struct qinst *inst);
 447 struct qreg qir_follow_movs(struct vc4_compile *c, struct qreg reg);
 448
 449 void qir_dump(struct vc4_compile *c);
 450 void qir_dump_inst(struct vc4_compile *c, struct qinst *inst);
 451 const char *qir_get_stage_name(enum qstage stage);
 452
 453 void qir_optimize(struct vc4_compile *c);
 454 bool qir_opt_algebraic(struct vc4_compile *c);
 455 bool qir_opt_constant_folding(struct vc4_compile *c);
 456 bool qir_opt_copy_propagation(struct vc4_compile *c);
 457 bool qir_opt_cse(struct vc4_compile *c);
 458 bool qir_opt_dead_code(struct vc4_compile *c);
 459 bool qir_opt_small_immediates(struct vc4_compile *c);
 460 bool qir_opt_vpm_writes(struct vc4_compile *c);
 461 void vc4_nir_lower_io(struct vc4_compile *c);
 462 nir_ssa_def *vc4_nir_get_state_uniform(struct nir_builder *b,
 463                                        enum quniform_contents contents);
 464 void qir_lower_uniforms(struct vc4_compile *c);
 465
 466 void qpu_schedule_instructions(struct vc4_compile *c);
 467
 468 void qir_SF(struct vc4_compile *c, struct qreg src);
 469
 470 static inline struct qreg
 471 qir_uniform_ui(struct vc4_compile *c, uint32_t ui)
 472 {
 473         return qir_uniform(c, QUNIFORM_CONSTANT, ui);
 474 }
 475
 476 static inline struct qreg
 477 qir_uniform_f(struct vc4_compile *c, float f)
 478 {
 479         return qir_uniform(c, QUNIFORM_CONSTANT, fui(f));
 480 }
 481
 482 #define QIR_ALU0(name)                                                   \
 483 static inline struct qreg                                                \
 484 qir_##name(struct vc4_compile *c)                                        \
 485 {                                                                        \
 486         struct qreg t = qir_get_temp(c);                                 \
 487         qir_emit(c, qir_inst(QOP_##name, t, c->undef, c->undef));        \
 488         return t;                                                        \
 489 }
 490
 491 #define QIR_ALU1(name)                                                   \
 492 static inline struct qreg                                                \
 493 qir_##name(struct vc4_compile *c, struct qreg a)                         \
 494 {                                                                        \
 495         struct qreg t = qir_get_temp(c);                                 \
 496         qir_emit(c, qir_inst(QOP_##name, t, a, c->undef));               \
 497         return t;                                                        \
 498 }
 499
 500 #define QIR_ALU2(name)                                                   \
 501 static inline struct qreg                                                \
 502 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b)          \
 503 {                                                                        \
 504         struct qreg t = qir_get_temp(c);                                 \
 505         qir_emit(c, qir_inst(QOP_##name, t, a, b));                      \
 506         return t;                                                        \
 507 }
 508
 509 #define QIR_NODST_1(name)                                               \
 510 static inline void                                                      \
 511 qir_##name(struct vc4_compile *c, struct qreg a)                        \
 512 {                                                                       \
 513         qir_emit(c, qir_inst(QOP_##name, c->undef, a, c->undef));       \
 514 }
 515
 516 #define QIR_NODST_2(name)                                               \
 517 static inline void                                                      \
 518 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b)         \
 519 {                                                                       \
 520         qir_emit(c, qir_inst(QOP_##name, c->undef, a, b));       \
 521 }
 522
 523 QIR_ALU1(MOV)
 524 QIR_ALU2(FADD)
 525 QIR_ALU2(FSUB)
 526 QIR_ALU2(FMUL)
 527 QIR_ALU2(MUL24)
 528 QIR_ALU1(SEL_X_0_ZS)
 529 QIR_ALU1(SEL_X_0_ZC)
 530 QIR_ALU1(SEL_X_0_NS)
 531 QIR_ALU1(SEL_X_0_NC)
 532 QIR_ALU2(SEL_X_Y_ZS)
 533 QIR_ALU2(SEL_X_Y_ZC)
 534 QIR_ALU2(SEL_X_Y_NS)
 535 QIR_ALU2(SEL_X_Y_NC)
 536 QIR_ALU2(FMIN)
 537 QIR_ALU2(FMAX)
 538 QIR_ALU2(FMINABS)
 539 QIR_ALU2(FMAXABS)
 540 QIR_ALU1(FTOI)
 541 QIR_ALU1(ITOF)
 542
 543 QIR_ALU2(ADD)
 544 QIR_ALU2(SUB)
 545 QIR_ALU2(SHL)
 546 QIR_ALU2(SHR)
 547 QIR_ALU2(ASR)
 548 QIR_ALU2(MIN)
 549 QIR_ALU2(MAX)
 550 QIR_ALU2(AND)
 551 QIR_ALU2(OR)
 552 QIR_ALU2(XOR)
 553 QIR_ALU1(NOT)
 554
 555 QIR_ALU1(RCP)
 556 QIR_ALU1(RSQ)
 557 QIR_ALU1(EXP2)
 558 QIR_ALU1(LOG2)
 559 QIR_ALU2(PACK_SCALED)
 560 QIR_ALU1(PACK_8888_F)
 561 QIR_ALU2(PACK_8A_F)
 562 QIR_ALU2(PACK_8B_F)
 563 QIR_ALU2(PACK_8C_F)
 564 QIR_ALU2(PACK_8D_F)
 565 QIR_ALU1(VARY_ADD_C)
 566 QIR_NODST_2(TEX_S)
 567 QIR_NODST_2(TEX_T)
 568 QIR_NODST_2(TEX_R)
 569 QIR_NODST_2(TEX_B)
 570 QIR_NODST_2(TEX_DIRECT)
 571 QIR_ALU0(FRAG_X)
 572 QIR_ALU0(FRAG_Y)
 573 QIR_ALU0(FRAG_Z)
 574 QIR_ALU0(FRAG_W)
 575 QIR_ALU0(FRAG_REV_FLAG)
 576 QIR_ALU0(TEX_RESULT)
 577 QIR_ALU0(TLB_COLOR_READ)
 578 QIR_NODST_1(TLB_COLOR_WRITE)
 579 QIR_NODST_1(TLB_Z_WRITE)
 580 QIR_NODST_1(TLB_DISCARD_SETUP)
 581 QIR_NODST_1(TLB_STENCIL_SETUP)
 582
 583 static inline struct qreg
 584 qir_UNPACK_8_F(struct vc4_compile *c, struct qreg src, int i)
 585 {
 586         struct qreg t = qir_get_temp(c);
 587         qir_emit(c, qir_inst(QOP_UNPACK_8A_F + i, t, src, c->undef));
 588         return t;
 589 }
 590
 591 static inline struct qreg
 592 qir_UNPACK_8_I(struct vc4_compile *c, struct qreg src, int i)
 593 {
 594         struct qreg t = qir_get_temp(c);
 595         qir_emit(c, qir_inst(QOP_UNPACK_8A_I + i, t, src, c->undef));
 596         return t;
 597 }
 598
 599 static inline struct qreg
 600 qir_UNPACK_16_F(struct vc4_compile *c, struct qreg src, int i)
 601 {
 602         struct qreg t = qir_get_temp(c);
 603         qir_emit(c, qir_inst(QOP_UNPACK_16A_F + i, t, src, c->undef));
 604         return t;
 605 }
 606
 607 static inline struct qreg
 608 qir_UNPACK_16_I(struct vc4_compile *c, struct qreg src, int i)
 609 {
 610         struct qreg t = qir_get_temp(c);
 611         qir_emit(c, qir_inst(QOP_UNPACK_16A_I + i, t, src, c->undef));
 612         return t;
 613 }
 614
 615 static inline struct qreg
 616 qir_PACK_8_F(struct vc4_compile *c, struct qreg rest, struct qreg val, int chan)
 617 {
 618         struct qreg t = qir_get_temp(c);
 619         qir_emit(c, qir_inst(QOP_PACK_8A_F + chan, t, rest, val));
 620         return t;
 621 }
 622
 623 static inline struct qreg
 624 qir_POW(struct vc4_compile *c, struct qreg x, struct qreg y)
 625 {
 626         return qir_EXP2(c, qir_FMUL(c,
 627                                     y,
 628                                     qir_LOG2(c, x)));
 629 }
 630
 631 static inline void
 632 qir_VPM_WRITE(struct vc4_compile *c, struct qreg val)
 633 {
 634         static const struct qreg vpm = { QFILE_VPM, 0 };
 635         qir_emit(c, qir_inst(QOP_MOV, vpm, val, c->undef));
 636 }
 637
 638 #endif /* VC4_QIR_H */