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,
 113         QOP_UNPACK_8B,
 114         QOP_UNPACK_8C,
 115         QOP_UNPACK_8D,
 116
 117         /** Texture x coordinate parameter write */
 118         QOP_TEX_S,
 119         /** Texture y coordinate parameter write */
 120         QOP_TEX_T,
 121         /** Texture border color parameter or cube map z coordinate write */
 122         QOP_TEX_R,
 123         /** Texture LOD bias parameter write */
 124         QOP_TEX_B,
 125         /**
 126          * Signal of texture read being necessary and then reading r4 into
 127          * the destination
 128          */
 129         QOP_TEX_RESULT,
 130         QOP_R4_UNPACK_A,
 131         QOP_R4_UNPACK_B,
 132         QOP_R4_UNPACK_C,
 133         QOP_R4_UNPACK_D
 134 };
 135
 136 struct simple_node {
 137         struct simple_node *next;
 138         struct simple_node *prev;
 139 };
 140
 141 struct qinst {
 142         struct simple_node link;
 143
 144         enum qop op;
 145         struct qreg dst;
 146         struct qreg *src;
 147 };
 148
 149 enum qstage {
 150         /**
 151          * Coordinate shader, runs during binning, before the VS, and just
 152          * outputs position.
 153          */
 154         QSTAGE_COORD,
 155         QSTAGE_VERT,
 156         QSTAGE_FRAG,
 157 };
 158
 159 enum quniform_contents {
 160         /**
 161          * Indicates that a constant 32-bit value is copied from the program's
 162          * uniform contents.
 163          */
 164         QUNIFORM_CONSTANT,
 165         /**
 166          * Indicates that the program's uniform contents are used as an index
 167          * into the GL uniform storage.
 168          */
 169         QUNIFORM_UNIFORM,
 170
 171         /** @{
 172          * Scaling factors from clip coordinates to relative to the viewport
 173          * center.
 174          *
 175          * This is used by the coordinate and vertex shaders to produce the
 176          * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed
 177          * point offsets from the viewport ccenter.
 178          */
 179         QUNIFORM_VIEWPORT_X_SCALE,
 180         QUNIFORM_VIEWPORT_Y_SCALE,
 181         /** @} */
 182
 183         QUNIFORM_VIEWPORT_Z_OFFSET,
 184         QUNIFORM_VIEWPORT_Z_SCALE,
 185
 186         QUNIFORM_USER_CLIP_PLANE,
 187
 188         /**
 189          * A reference to a texture config parameter 0 uniform.
 190          *
 191          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 192          * defines texture type, miplevels, and such.  It will be found as a
 193          * parameter to the first QOP_TEX_[STRB] instruction in a sequence.
 194          */
 195         QUNIFORM_TEXTURE_CONFIG_P0,
 196
 197         /**
 198          * A reference to a texture config parameter 1 uniform.
 199          *
 200          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 201          * defines texture width, height, filters, and wrap modes.  It will be
 202          * found as a parameter to the second QOP_TEX_[STRB] instruction in a
 203          * sequence.
 204          */
 205         QUNIFORM_TEXTURE_CONFIG_P1,
 206
 207         /** A reference to a texture config parameter 2 cubemap stride uniform */
 208         QUNIFORM_TEXTURE_CONFIG_P2,
 209
 210         QUNIFORM_TEXRECT_SCALE_X,
 211         QUNIFORM_TEXRECT_SCALE_Y,
 212
 213         QUNIFORM_TEXTURE_BORDER_COLOR,
 214
 215         QUNIFORM_BLEND_CONST_COLOR,
 216         QUNIFORM_STENCIL,
 217
 218         QUNIFORM_ALPHA_REF,
 219 };
 220
 221 struct vc4_varying_semantic {
 222         uint8_t semantic;
 223         uint8_t index;
 224         uint8_t swizzle;
 225 };
 226
 227 struct vc4_compile {
 228         struct vc4_context *vc4;
 229         struct tgsi_parse_context parser;
 230         struct qreg *temps;
 231         /**
 232          * Inputs to the shader, arranged by TGSI declaration order.
 233          *
 234          * Not all fragment shader QFILE_VARY reads are present in this array.
 235          */
 236         struct qreg *inputs;
 237         struct qreg *outputs;
 238         struct qreg *consts;
 239         uint32_t temps_array_size;
 240         uint32_t inputs_array_size;
 241         uint32_t outputs_array_size;
 242         uint32_t uniforms_array_size;
 243         uint32_t consts_array_size;
 244         uint32_t num_consts;
 245         struct qreg line_x, point_x, point_y;
 246         struct qreg discard;
 247
 248         /**
 249          * Array of the TGSI semantics of all FS QFILE_VARY reads.
 250          *
 251          * This includes those that aren't part of the VPM varyings, like
 252          * point/line coordinates.
 253          */
 254         struct vc4_varying_semantic *input_semantics;
 255         uint32_t num_input_semantics;
 256         uint32_t input_semantics_array_size;
 257
 258         /**
 259          * An entry per outputs[] in the VS indicating what the semantic of
 260          * the output is.  Used to emit from the VS in the order that the FS
 261          * needs.
 262          */
 263         struct vc4_varying_semantic *output_semantics;
 264
 265         struct pipe_shader_state *shader_state;
 266         struct vc4_key *key;
 267         struct vc4_fs_key *fs_key;
 268         struct vc4_vs_key *vs_key;
 269
 270         uint32_t *uniform_data;
 271         enum quniform_contents *uniform_contents;
 272         uint32_t uniform_array_size;
 273         uint32_t num_uniforms;
 274         uint32_t num_outputs;
 275         uint32_t num_texture_samples;
 276         uint32_t output_position_index;
 277         uint32_t output_clipvertex_index;
 278         uint32_t output_color_index;
 279         uint32_t output_point_size_index;
 280
 281         struct qreg undef;
 282         enum qstage stage;
 283         uint32_t num_temps;
 284         struct simple_node instructions;
 285         uint32_t immediates[1024];
 286
 287         struct simple_node qpu_inst_list;
 288         uint64_t *qpu_insts;
 289         uint32_t qpu_inst_count;
 290         uint32_t qpu_inst_size;
 291         uint32_t num_inputs;
 292
 293         uint32_t program_id;
 294         uint32_t variant_id;
 295 };
 296
 297 struct vc4_compile *qir_compile_init(void);
 298 void qir_compile_destroy(struct vc4_compile *c);
 299 struct qinst *qir_inst(enum qop op, struct qreg dst,
 300                        struct qreg src0, struct qreg src1);
 301 struct qinst *qir_inst4(enum qop op, struct qreg dst,
 302                         struct qreg a,
 303                         struct qreg b,
 304                         struct qreg c,
 305                         struct qreg d);
 306 void qir_remove_instruction(struct qinst *qinst);
 307 void qir_reorder_uniforms(struct vc4_compile *c);
 308 void qir_emit(struct vc4_compile *c, struct qinst *inst);
 309 struct qreg qir_get_temp(struct vc4_compile *c);
 310 int qir_get_op_nsrc(enum qop qop);
 311 bool qir_reg_equals(struct qreg a, struct qreg b);
 312 bool qir_has_side_effects(struct vc4_compile *c, struct qinst *inst);
 313 bool qir_depends_on_flags(struct qinst *inst);
 314 bool qir_writes_r4(struct qinst *inst);
 315 bool qir_reads_r4(struct qinst *inst);
 316
 317 void qir_dump(struct vc4_compile *c);
 318 void qir_dump_inst(struct vc4_compile *c, struct qinst *inst);
 319 const char *qir_get_stage_name(enum qstage stage);
 320
 321 void qir_optimize(struct vc4_compile *c);
 322 bool qir_opt_algebraic(struct vc4_compile *c);
 323 bool qir_opt_copy_propagation(struct vc4_compile *c);
 324 bool qir_opt_cse(struct vc4_compile *c);
 325 bool qir_opt_dead_code(struct vc4_compile *c);
 326
 327 #define QIR_ALU0(name)                                                   \
 328 static inline struct qreg                                                \
 329 qir_##name(struct vc4_compile *c)                                        \
 330 {                                                                        \
 331         struct qreg t = qir_get_temp(c);                                 \
 332         qir_emit(c, qir_inst(QOP_##name, t, c->undef, c->undef));        \
 333         return t;                                                        \
 334 }
 335
 336 #define QIR_ALU1(name)                                                   \
 337 static inline struct qreg                                                \
 338 qir_##name(struct vc4_compile *c, struct qreg a)                         \
 339 {                                                                        \
 340         struct qreg t = qir_get_temp(c);                                 \
 341         qir_emit(c, qir_inst(QOP_##name, t, a, c->undef));               \
 342         return t;                                                        \
 343 }
 344
 345 #define QIR_ALU2(name)                                                   \
 346 static inline struct qreg                                                \
 347 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b)          \
 348 {                                                                        \
 349         struct qreg t = qir_get_temp(c);                                 \
 350         qir_emit(c, qir_inst(QOP_##name, t, a, b));                      \
 351         return t;                                                        \
 352 }
 353
 354 #define QIR_NODST_1(name)                                               \
 355 static inline void                                                      \
 356 qir_##name(struct vc4_compile *c, struct qreg a)                        \
 357 {                                                                       \
 358         qir_emit(c, qir_inst(QOP_##name, c->undef, a, c->undef));       \
 359 }
 360
 361 #define QIR_NODST_2(name)                                               \
 362 static inline void                                                      \
 363 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b)         \
 364 {                                                                       \
 365         qir_emit(c, qir_inst(QOP_##name, c->undef, a, b));       \
 366 }
 367
 368 QIR_ALU1(MOV)
 369 QIR_ALU2(FADD)
 370 QIR_ALU2(FSUB)
 371 QIR_ALU2(FMUL)
 372 QIR_ALU2(MUL24)
 373 QIR_NODST_1(SF)
 374 QIR_ALU1(SEL_X_0_ZS)
 375 QIR_ALU1(SEL_X_0_ZC)
 376 QIR_ALU1(SEL_X_0_NS)
 377 QIR_ALU1(SEL_X_0_NC)
 378 QIR_ALU2(SEL_X_Y_ZS)
 379 QIR_ALU2(SEL_X_Y_ZC)
 380 QIR_ALU2(SEL_X_Y_NS)
 381 QIR_ALU2(SEL_X_Y_NC)
 382 QIR_ALU2(FMIN)
 383 QIR_ALU2(FMAX)
 384 QIR_ALU2(FMINABS)
 385 QIR_ALU2(FMAXABS)
 386 QIR_ALU1(FTOI)
 387 QIR_ALU1(ITOF)
 388
 389 QIR_ALU2(ADD)
 390 QIR_ALU2(SUB)
 391 QIR_ALU2(SHL)
 392 QIR_ALU2(SHR)
 393 QIR_ALU2(ASR)
 394 QIR_ALU2(MIN)
 395 QIR_ALU2(MAX)
 396 QIR_ALU2(AND)
 397 QIR_ALU2(OR)
 398 QIR_ALU2(XOR)
 399 QIR_ALU1(NOT)
 400
 401 QIR_ALU1(RCP)
 402 QIR_ALU1(RSQ)
 403 QIR_ALU1(EXP2)
 404 QIR_ALU1(LOG2)
 405 QIR_ALU2(PACK_SCALED)
 406 QIR_ALU1(VARY_ADD_C)
 407 QIR_NODST_1(VPM_WRITE)
 408 QIR_NODST_2(TEX_S)
 409 QIR_NODST_2(TEX_T)
 410 QIR_NODST_2(TEX_R)
 411 QIR_NODST_2(TEX_B)
 412 QIR_ALU0(FRAG_X)
 413 QIR_ALU0(FRAG_Y)
 414 QIR_ALU0(FRAG_Z)
 415 QIR_ALU0(FRAG_W)
 416 QIR_ALU0(FRAG_REV_FLAG)
 417 QIR_ALU0(TEX_RESULT)
 418 QIR_ALU0(TLB_COLOR_READ)
 419 QIR_NODST_1(TLB_Z_WRITE)
 420 QIR_NODST_1(TLB_DISCARD_SETUP)
 421 QIR_NODST_1(TLB_STENCIL_SETUP)
 422
 423 static inline struct qreg
 424 qir_R4_UNPACK(struct vc4_compile *c, struct qreg r4, int i)
 425 {
 426         struct qreg t = qir_get_temp(c);
 427         qir_emit(c, qir_inst(QOP_R4_UNPACK_A + i, t, r4, c->undef));
 428         return t;
 429 }
 430
 431 static inline struct qreg
 432 qir_SEL_X_0_COND(struct vc4_compile *c, int i)
 433 {
 434         struct qreg t = qir_get_temp(c);
 435         qir_emit(c, qir_inst(QOP_R4_UNPACK_A + i, t, c->undef, c->undef));
 436         return t;
 437 }
 438
 439 static inline struct qreg
 440 qir_UNPACK_8(struct vc4_compile *c, struct qreg src, int i)
 441 {
 442         struct qreg t = qir_get_temp(c);
 443         qir_emit(c, qir_inst(QOP_UNPACK_8A + i, t, src, c->undef));
 444         return t;
 445 }
 446
 447 static inline struct qreg
 448 qir_POW(struct vc4_compile *c, struct qreg x, struct qreg y)
 449 {
 450         return qir_EXP2(c, qir_FMUL(c,
 451                                     y,
 452                                     qir_LOG2(c, x)));
 453 }
 454
 455 #endif /* VC4_QIR_H */