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 <stdio.h>
  28 #include <stdlib.h>
  29 #include <stdbool.h>
  30 #include <stdint.h>
  31 #include <string.h>
  32
  33 #include "util/u_simple_list.h"
  34 #include "tgsi/tgsi_parse.h"
  35
  36 enum qfile {
  37         QFILE_NULL,
  38         QFILE_TEMP,
  39         QFILE_VARY,
  40         QFILE_UNIF,
  41 };
  42
  43 struct qreg {
  44         enum qfile file;
  45         uint32_t index;
  46 };
  47
  48 enum qop {
  49         QOP_UNDEF,
  50         QOP_MOV,
  51         QOP_FADD,
  52         QOP_FSUB,
  53         QOP_FMUL,
  54         QOP_MUL24,
  55         QOP_FMIN,
  56         QOP_FMAX,
  57         QOP_FMINABS,
  58         QOP_FMAXABS,
  59         QOP_ADD,
  60         QOP_SUB,
  61         QOP_SHL,
  62         QOP_SHR,
  63         QOP_ASR,
  64         QOP_MIN,
  65         QOP_MAX,
  66         QOP_AND,
  67         QOP_OR,
  68         QOP_XOR,
  69         QOP_NOT,
  70
  71         /* Sets the flag register according to src. */
  72         QOP_SF,
  73
  74         /* Note: Orderings of these compares must be the same as in
  75          * qpu_defines.h.  Selects the src[0] if the ns flag bit is set,
  76          * otherwise 0. */
  77         QOP_SEL_X_0_ZS,
  78         QOP_SEL_X_0_ZC,
  79         QOP_SEL_X_0_NS,
  80         QOP_SEL_X_0_NC,
  81         /* Selects the src[0] if the ns flag bit is set, otherwise src[1]. */
  82         QOP_SEL_X_Y_ZS,
  83         QOP_SEL_X_Y_ZC,
  84         QOP_SEL_X_Y_NS,
  85         QOP_SEL_X_Y_NC,
  86
  87         QOP_FTOI,
  88         QOP_ITOF,
  89         QOP_RCP,
  90         QOP_RSQ,
  91         QOP_EXP2,
  92         QOP_LOG2,
  93         QOP_VW_SETUP,
  94         QOP_VR_SETUP,
  95         QOP_PACK_SCALED,
  96         QOP_PACK_COLORS,
  97         QOP_VPM_WRITE,
  98         QOP_VPM_READ,
  99         QOP_TLB_DISCARD_SETUP,
 100         QOP_TLB_STENCIL_SETUP,
 101         QOP_TLB_Z_WRITE,
 102         QOP_TLB_COLOR_WRITE,
 103         QOP_TLB_COLOR_READ,
 104         QOP_VARY_ADD_C,
 105
 106         QOP_FRAG_X,
 107         QOP_FRAG_Y,
 108         QOP_FRAG_Z,
 109         QOP_FRAG_W,
 110         QOP_FRAG_REV_FLAG,
 111
 112         QOP_UNPACK_8A_F,
 113         QOP_UNPACK_8B_F,
 114         QOP_UNPACK_8C_F,
 115         QOP_UNPACK_8D_F,
 116         QOP_UNPACK_16A_F,
 117         QOP_UNPACK_16B_F,
 118
 119         QOP_UNPACK_8A_I,
 120         QOP_UNPACK_8B_I,
 121         QOP_UNPACK_8C_I,
 122         QOP_UNPACK_8D_I,
 123         QOP_UNPACK_16A_I,
 124         QOP_UNPACK_16B_I,
 125
 126         /** Texture x coordinate parameter write */
 127         QOP_TEX_S,
 128         /** Texture y coordinate parameter write */
 129         QOP_TEX_T,
 130         /** Texture border color parameter or cube map z coordinate write */
 131         QOP_TEX_R,
 132         /** Texture LOD bias parameter write */
 133         QOP_TEX_B,
 134
 135         /**
 136          * Texture-unit 4-byte read with address provided direct in S
 137          * cooordinate.
 138          *
 139          * The first operand is the offset from the start of the UBO, and the
 140          * second is the uniform that has the UBO's base pointer.
 141          */
 142         QOP_TEX_DIRECT,
 143
 144         /**
 145          * Signal of texture read being necessary and then reading r4 into
 146          * the destination
 147          */
 148         QOP_TEX_RESULT,
 149         QOP_R4_UNPACK_A,
 150         QOP_R4_UNPACK_B,
 151         QOP_R4_UNPACK_C,
 152         QOP_R4_UNPACK_D
 153 };
 154
 155 struct simple_node {
 156         struct simple_node *next;
 157         struct simple_node *prev;
 158 };
 159
 160 struct queued_qpu_inst {
 161         struct simple_node link;
 162         uint64_t inst;
 163 };
 164
 165 struct qinst {
 166         struct simple_node link;
 167
 168         enum qop op;
 169         struct qreg dst;
 170         struct qreg *src;
 171 };
 172
 173 enum qstage {
 174         /**
 175          * Coordinate shader, runs during binning, before the VS, and just
 176          * outputs position.
 177          */
 178         QSTAGE_COORD,
 179         QSTAGE_VERT,
 180         QSTAGE_FRAG,
 181 };
 182
 183 enum quniform_contents {
 184         /**
 185          * Indicates that a constant 32-bit value is copied from the program's
 186          * uniform contents.
 187          */
 188         QUNIFORM_CONSTANT,
 189         /**
 190          * Indicates that the program's uniform contents are used as an index
 191          * into the GL uniform storage.
 192          */
 193         QUNIFORM_UNIFORM,
 194
 195         /** @{
 196          * Scaling factors from clip coordinates to relative to the viewport
 197          * center.
 198          *
 199          * This is used by the coordinate and vertex shaders to produce the
 200          * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed
 201          * point offsets from the viewport ccenter.
 202          */
 203         QUNIFORM_VIEWPORT_X_SCALE,
 204         QUNIFORM_VIEWPORT_Y_SCALE,
 205         /** @} */
 206
 207         QUNIFORM_VIEWPORT_Z_OFFSET,
 208         QUNIFORM_VIEWPORT_Z_SCALE,
 209
 210         QUNIFORM_USER_CLIP_PLANE,
 211
 212         /**
 213          * A reference to a texture config parameter 0 uniform.
 214          *
 215          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 216          * defines texture type, miplevels, and such.  It will be found as a
 217          * parameter to the first QOP_TEX_[STRB] instruction in a sequence.
 218          */
 219         QUNIFORM_TEXTURE_CONFIG_P0,
 220
 221         /**
 222          * A reference to a texture config parameter 1 uniform.
 223          *
 224          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 225          * defines texture width, height, filters, and wrap modes.  It will be
 226          * found as a parameter to the second QOP_TEX_[STRB] instruction in a
 227          * sequence.
 228          */
 229         QUNIFORM_TEXTURE_CONFIG_P1,
 230
 231         /** A reference to a texture config parameter 2 cubemap stride uniform */
 232         QUNIFORM_TEXTURE_CONFIG_P2,
 233
 234         QUNIFORM_UBO_ADDR,
 235
 236         QUNIFORM_TEXRECT_SCALE_X,
 237         QUNIFORM_TEXRECT_SCALE_Y,
 238
 239         QUNIFORM_TEXTURE_BORDER_COLOR,
 240
 241         QUNIFORM_BLEND_CONST_COLOR,
 242         QUNIFORM_STENCIL,
 243
 244         QUNIFORM_ALPHA_REF,
 245 };
 246
 247 struct vc4_varying_semantic {
 248         uint8_t semantic;
 249         uint8_t index;
 250         uint8_t swizzle;
 251 };
 252
 253 struct vc4_compiler_ubo_range {
 254         /**
 255          * offset in bytes from the start of the ubo where this range is
 256          * uploaded.
 257          *
 258          * Only set once used is set.
 259          */
 260         uint32_t dst_offset;
 261
 262         /**
 263          * offset in bytes from the start of the gallium uniforms where the
 264          * data comes from.
 265          */
 266         uint32_t src_offset;
 267
 268         /** size in bytes of this ubo range */
 269         uint32_t size;
 270
 271         /**
 272          * Set if this range is used by the shader for indirect uniforms
 273          * access.
 274          */
 275         bool used;
 276 };
 277
 278 struct vc4_compile {
 279         struct vc4_context *vc4;
 280         struct tgsi_parse_context parser;
 281         struct qreg *temps;
 282         /**
 283          * Inputs to the shader, arranged by TGSI declaration order.
 284          *
 285          * Not all fragment shader QFILE_VARY reads are present in this array.
 286          */
 287         struct qreg *inputs;
 288         struct qreg *outputs;
 289         struct qreg *consts;
 290         struct qreg addr[4]; /* TGSI ARL destination. */
 291         uint32_t temps_array_size;
 292         uint32_t inputs_array_size;
 293         uint32_t outputs_array_size;
 294         uint32_t uniforms_array_size;
 295         uint32_t consts_array_size;
 296         uint32_t num_consts;
 297
 298         struct vc4_compiler_ubo_range *ubo_ranges;
 299         uint32_t ubo_ranges_array_size;
 300         uint32_t num_ubo_ranges;
 301         uint32_t next_ubo_dst_offset;
 302
 303         struct qreg line_x, point_x, point_y;
 304         struct qreg discard;
 305
 306         /**
 307          * Array of the TGSI semantics of all FS QFILE_VARY reads.
 308          *
 309          * This includes those that aren't part of the VPM varyings, like
 310          * point/line coordinates.
 311          */
 312         struct vc4_varying_semantic *input_semantics;
 313         uint32_t num_input_semantics;
 314         uint32_t input_semantics_array_size;
 315
 316         /**
 317          * An entry per outputs[] in the VS indicating what the semantic of
 318          * the output is.  Used to emit from the VS in the order that the FS
 319          * needs.
 320          */
 321         struct vc4_varying_semantic *output_semantics;
 322
 323         struct pipe_shader_state *shader_state;
 324         struct vc4_key *key;
 325         struct vc4_fs_key *fs_key;
 326         struct vc4_vs_key *vs_key;
 327
 328         uint32_t *uniform_data;
 329         enum quniform_contents *uniform_contents;
 330         uint32_t uniform_array_size;
 331         uint32_t num_uniforms;
 332         uint32_t num_outputs;
 333         uint32_t num_texture_samples;
 334         uint32_t output_position_index;
 335         uint32_t output_clipvertex_index;
 336         uint32_t output_color_index;
 337         uint32_t output_point_size_index;
 338
 339         struct qreg undef;
 340         enum qstage stage;
 341         uint32_t num_temps;
 342         struct simple_node instructions;
 343         uint32_t immediates[1024];
 344
 345         struct simple_node qpu_inst_list;
 346         uint64_t *qpu_insts;
 347         uint32_t qpu_inst_count;
 348         uint32_t qpu_inst_size;
 349         uint32_t num_inputs;
 350
 351         uint32_t program_id;
 352         uint32_t variant_id;
 353 };
 354
 355 struct vc4_compile *qir_compile_init(void);
 356 void qir_compile_destroy(struct vc4_compile *c);
 357 struct qinst *qir_inst(enum qop op, struct qreg dst,
 358                        struct qreg src0, struct qreg src1);
 359 struct qinst *qir_inst4(enum qop op, struct qreg dst,
 360                         struct qreg a,
 361                         struct qreg b,
 362                         struct qreg c,
 363                         struct qreg d);
 364 void qir_remove_instruction(struct qinst *qinst);
 365 void qir_reorder_uniforms(struct vc4_compile *c);
 366 void qir_emit(struct vc4_compile *c, struct qinst *inst);
 367 struct qreg qir_get_temp(struct vc4_compile *c);
 368 int qir_get_op_nsrc(enum qop qop);
 369 bool qir_reg_equals(struct qreg a, struct qreg b);
 370 bool qir_has_side_effects(struct vc4_compile *c, struct qinst *inst);
 371 bool qir_depends_on_flags(struct qinst *inst);
 372 bool qir_writes_r4(struct qinst *inst);
 373 bool qir_reads_r4(struct qinst *inst);
 374
 375 void qir_dump(struct vc4_compile *c);
 376 void qir_dump_inst(struct vc4_compile *c, struct qinst *inst);
 377 const char *qir_get_stage_name(enum qstage stage);
 378
 379 void qir_optimize(struct vc4_compile *c);
 380 bool qir_opt_algebraic(struct vc4_compile *c);
 381 bool qir_opt_copy_propagation(struct vc4_compile *c);
 382 bool qir_opt_cse(struct vc4_compile *c);
 383 bool qir_opt_dead_code(struct vc4_compile *c);
 384
 385 void qpu_schedule_instructions(struct vc4_compile *c);
 386
 387 #define QIR_ALU0(name)                                                   \
 388 static inline struct qreg                                                \
 389 qir_##name(struct vc4_compile *c)                                        \
 390 {                                                                        \
 391         struct qreg t = qir_get_temp(c);                                 \
 392         qir_emit(c, qir_inst(QOP_##name, t, c->undef, c->undef));        \
 393         return t;                                                        \
 394 }
 395
 396 #define QIR_ALU1(name)                                                   \
 397 static inline struct qreg                                                \
 398 qir_##name(struct vc4_compile *c, struct qreg a)                         \
 399 {                                                                        \
 400         struct qreg t = qir_get_temp(c);                                 \
 401         qir_emit(c, qir_inst(QOP_##name, t, a, c->undef));               \
 402         return t;                                                        \
 403 }
 404
 405 #define QIR_ALU2(name)                                                   \
 406 static inline struct qreg                                                \
 407 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b)          \
 408 {                                                                        \
 409         struct qreg t = qir_get_temp(c);                                 \
 410         qir_emit(c, qir_inst(QOP_##name, t, a, b));                      \
 411         return t;                                                        \
 412 }
 413
 414 #define QIR_NODST_1(name)                                               \
 415 static inline void                                                      \
 416 qir_##name(struct vc4_compile *c, struct qreg a)                        \
 417 {                                                                       \
 418         qir_emit(c, qir_inst(QOP_##name, c->undef, a, c->undef));       \
 419 }
 420
 421 #define QIR_NODST_2(name)                                               \
 422 static inline void                                                      \
 423 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b)         \
 424 {                                                                       \
 425         qir_emit(c, qir_inst(QOP_##name, c->undef, a, b));       \
 426 }
 427
 428 QIR_ALU1(MOV)
 429 QIR_ALU2(FADD)
 430 QIR_ALU2(FSUB)
 431 QIR_ALU2(FMUL)
 432 QIR_ALU2(MUL24)
 433 QIR_NODST_1(SF)
 434 QIR_ALU1(SEL_X_0_ZS)
 435 QIR_ALU1(SEL_X_0_ZC)
 436 QIR_ALU1(SEL_X_0_NS)
 437 QIR_ALU1(SEL_X_0_NC)
 438 QIR_ALU2(SEL_X_Y_ZS)
 439 QIR_ALU2(SEL_X_Y_ZC)
 440 QIR_ALU2(SEL_X_Y_NS)
 441 QIR_ALU2(SEL_X_Y_NC)
 442 QIR_ALU2(FMIN)
 443 QIR_ALU2(FMAX)
 444 QIR_ALU2(FMINABS)
 445 QIR_ALU2(FMAXABS)
 446 QIR_ALU1(FTOI)
 447 QIR_ALU1(ITOF)
 448
 449 QIR_ALU2(ADD)
 450 QIR_ALU2(SUB)
 451 QIR_ALU2(SHL)
 452 QIR_ALU2(SHR)
 453 QIR_ALU2(ASR)
 454 QIR_ALU2(MIN)
 455 QIR_ALU2(MAX)
 456 QIR_ALU2(AND)
 457 QIR_ALU2(OR)
 458 QIR_ALU2(XOR)
 459 QIR_ALU1(NOT)
 460
 461 QIR_ALU1(RCP)
 462 QIR_ALU1(RSQ)
 463 QIR_ALU1(EXP2)
 464 QIR_ALU1(LOG2)
 465 QIR_ALU2(PACK_SCALED)
 466 QIR_ALU1(VARY_ADD_C)
 467 QIR_NODST_1(VPM_WRITE)
 468 QIR_NODST_2(TEX_S)
 469 QIR_NODST_2(TEX_T)
 470 QIR_NODST_2(TEX_R)
 471 QIR_NODST_2(TEX_B)
 472 QIR_NODST_2(TEX_DIRECT)
 473 QIR_ALU0(FRAG_X)
 474 QIR_ALU0(FRAG_Y)
 475 QIR_ALU0(FRAG_Z)
 476 QIR_ALU0(FRAG_W)
 477 QIR_ALU0(FRAG_REV_FLAG)
 478 QIR_ALU0(TEX_RESULT)
 479 QIR_ALU0(TLB_COLOR_READ)
 480 QIR_NODST_1(TLB_Z_WRITE)
 481 QIR_NODST_1(TLB_DISCARD_SETUP)
 482 QIR_NODST_1(TLB_STENCIL_SETUP)
 483
 484 static inline struct qreg
 485 qir_R4_UNPACK(struct vc4_compile *c, struct qreg r4, int i)
 486 {
 487         struct qreg t = qir_get_temp(c);
 488         qir_emit(c, qir_inst(QOP_R4_UNPACK_A + i, t, r4, c->undef));
 489         return t;
 490 }
 491
 492 static inline struct qreg
 493 qir_SEL_X_0_COND(struct vc4_compile *c, int i)
 494 {
 495         struct qreg t = qir_get_temp(c);
 496         qir_emit(c, qir_inst(QOP_R4_UNPACK_A + i, t, c->undef, c->undef));
 497         return t;
 498 }
 499
 500 static inline struct qreg
 501 qir_UNPACK_8_F(struct vc4_compile *c, struct qreg src, int i)
 502 {
 503         struct qreg t = qir_get_temp(c);
 504         qir_emit(c, qir_inst(QOP_UNPACK_8A_F + i, t, src, c->undef));
 505         return t;
 506 }
 507
 508 static inline struct qreg
 509 qir_UNPACK_8_I(struct vc4_compile *c, struct qreg src, int i)
 510 {
 511         struct qreg t = qir_get_temp(c);
 512         qir_emit(c, qir_inst(QOP_UNPACK_8A_I + i, t, src, c->undef));
 513         return t;
 514 }
 515
 516 static inline struct qreg
 517 qir_UNPACK_16_F(struct vc4_compile *c, struct qreg src, int i)
 518 {
 519         struct qreg t = qir_get_temp(c);
 520         qir_emit(c, qir_inst(QOP_UNPACK_16A_F + i, t, src, c->undef));
 521         return t;
 522 }
 523
 524 static inline struct qreg
 525 qir_UNPACK_16_I(struct vc4_compile *c, struct qreg src, int i)
 526 {
 527         struct qreg t = qir_get_temp(c);
 528         qir_emit(c, qir_inst(QOP_UNPACK_16A_I + i, t, src, c->undef));
 529         return t;
 530 }
 531
 532 static inline struct qreg
 533 qir_POW(struct vc4_compile *c, struct qreg x, struct qreg y)
 534 {
 535         return qir_EXP2(c, qir_FMUL(c,
 536                                     y,
 537                                     qir_LOG2(c, x)));
 538 }
 539
 540 #endif /* VC4_QIR_H */