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_PASSTHROUGH_Z_WRITE,
 101         QOP_TLB_COLOR_WRITE,
 102         QOP_TLB_COLOR_READ,
 103         QOP_VARY_ADD_C,
 104
 105         QOP_FRAG_X,
 106         QOP_FRAG_Y,
 107         QOP_FRAG_Z,
 108         QOP_FRAG_RCP_W,
 109
 110         /** Texture x coordinate parameter write */
 111         QOP_TEX_S,
 112         /** Texture y coordinate parameter write */
 113         QOP_TEX_T,
 114         /** Texture border color parameter or cube map z coordinate write */
 115         QOP_TEX_R,
 116         /** Texture LOD bias parameter write */
 117         QOP_TEX_B,
 118         /**
 119          * Signal of texture read being necessary and then reading r4 into
 120          * the destination
 121          */
 122         QOP_TEX_RESULT,
 123         QOP_R4_UNPACK_A,
 124         QOP_R4_UNPACK_B,
 125         QOP_R4_UNPACK_C,
 126         QOP_R4_UNPACK_D
 127 };
 128
 129 struct simple_node {
 130         struct simple_node *next;
 131         struct simple_node *prev;
 132 };
 133
 134 struct qinst {
 135         struct simple_node link;
 136
 137         enum qop op;
 138         struct qreg dst;
 139         struct qreg *src;
 140 };
 141
 142 enum qstage {
 143         /**
 144          * Coordinate shader, runs during binning, before the VS, and just
 145          * outputs position.
 146          */
 147         QSTAGE_COORD,
 148         QSTAGE_VERT,
 149         QSTAGE_FRAG,
 150 };
 151
 152 enum quniform_contents {
 153         /**
 154          * Indicates that a constant 32-bit value is copied from the program's
 155          * uniform contents.
 156          */
 157         QUNIFORM_CONSTANT,
 158         /**
 159          * Indicates that the program's uniform contents are used as an index
 160          * into the GL uniform storage.
 161          */
 162         QUNIFORM_UNIFORM,
 163
 164         /** @{
 165          * Scaling factors from clip coordinates to relative to the viewport
 166          * center.
 167          *
 168          * This is used by the coordinate and vertex shaders to produce the
 169          * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed
 170          * point offsets from the viewport ccenter.
 171          */
 172         QUNIFORM_VIEWPORT_X_SCALE,
 173         QUNIFORM_VIEWPORT_Y_SCALE,
 174         /** @} */
 175
 176         QUNIFORM_VIEWPORT_Z_OFFSET,
 177         QUNIFORM_VIEWPORT_Z_SCALE,
 178
 179         /**
 180          * A reference to a texture config parameter 0 uniform.
 181          *
 182          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 183          * defines texture type, miplevels, and such.  It will be found as a
 184          * parameter to the first QOP_TEX_[STRB] instruction in a sequence.
 185          */
 186         QUNIFORM_TEXTURE_CONFIG_P0,
 187
 188         /**
 189          * A reference to a texture config parameter 1 uniform.
 190          *
 191          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 192          * defines texture width, height, filters, and wrap modes.  It will be
 193          * found as a parameter to the second QOP_TEX_[STRB] instruction in a
 194          * sequence.
 195          */
 196         QUNIFORM_TEXTURE_CONFIG_P1,
 197
 198         QUNIFORM_TEXRECT_SCALE_X,
 199         QUNIFORM_TEXRECT_SCALE_Y,
 200
 201         QUNIFORM_BLEND_CONST_COLOR,
 202 };
 203
 204 struct vc4_compile {
 205         struct tgsi_parse_context parser;
 206         struct qreg *temps;
 207         struct qreg *inputs;
 208         struct qreg *outputs;
 209         struct qreg *uniforms;
 210         struct qreg *consts;
 211         uint32_t num_consts;
 212         struct qreg line_x, point_x, point_y;
 213         struct qreg discard;
 214
 215         struct pipe_shader_state *shader_state;
 216         struct vc4_key *key;
 217         struct vc4_fs_key *fs_key;
 218         struct vc4_vs_key *vs_key;
 219
 220         uint32_t *uniform_data;
 221         enum quniform_contents *uniform_contents;
 222         uint32_t num_uniforms;
 223         uint32_t num_outputs;
 224         uint32_t num_texture_samples;
 225
 226         struct qreg undef;
 227         enum qstage stage;
 228         uint32_t num_temps;
 229         struct simple_node instructions;
 230         uint32_t immediates[1024];
 231
 232         struct simple_node qpu_inst_list;
 233         uint64_t *qpu_insts;
 234         uint32_t qpu_inst_count;
 235         uint32_t qpu_inst_size;
 236         uint32_t num_inputs;
 237 };
 238
 239 struct vc4_compile *qir_compile_init(void);
 240 void qir_compile_destroy(struct vc4_compile *c);
 241 struct qinst *qir_inst(enum qop op, struct qreg dst,
 242                        struct qreg src0, struct qreg src1);
 243 struct qinst *qir_inst4(enum qop op, struct qreg dst,
 244                         struct qreg a,
 245                         struct qreg b,
 246                         struct qreg c,
 247                         struct qreg d);
 248 void qir_emit(struct vc4_compile *c, struct qinst *inst);
 249 struct qreg qir_get_temp(struct vc4_compile *c);
 250 int qir_get_op_nsrc(enum qop qop);
 251 bool qir_reg_equals(struct qreg a, struct qreg b);
 252 bool qir_has_side_effects(struct qinst *inst);
 253 bool qir_depends_on_flags(struct qinst *inst);
 254 bool qir_writes_r4(struct qinst *inst);
 255 bool qir_reads_r4(struct qinst *inst);
 256
 257 void qir_dump(struct vc4_compile *c);
 258 void qir_dump_inst(struct qinst *inst);
 259 const char *qir_get_stage_name(enum qstage stage);
 260
 261 void qir_optimize(struct vc4_compile *c);
 262 bool qir_opt_algebraic(struct vc4_compile *c);
 263 bool qir_opt_copy_propagation(struct vc4_compile *c);
 264 bool qir_opt_cse(struct vc4_compile *c);
 265 bool qir_opt_dead_code(struct vc4_compile *c);
 266
 267 #define QIR_ALU0(name)                                                   \
 268 static inline struct qreg                                                \
 269 qir_##name(struct vc4_compile *c)                                        \
 270 {                                                                        \
 271         struct qreg t = qir_get_temp(c);                                 \
 272         qir_emit(c, qir_inst(QOP_##name, t, c->undef, c->undef));        \
 273         return t;                                                        \
 274 }
 275
 276 #define QIR_ALU1(name)                                                   \
 277 static inline struct qreg                                                \
 278 qir_##name(struct vc4_compile *c, struct qreg a)                         \
 279 {                                                                        \
 280         struct qreg t = qir_get_temp(c);                                 \
 281         qir_emit(c, qir_inst(QOP_##name, t, a, c->undef));               \
 282         return t;                                                        \
 283 }
 284
 285 #define QIR_ALU2(name)                                                   \
 286 static inline struct qreg                                                \
 287 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b)          \
 288 {                                                                        \
 289         struct qreg t = qir_get_temp(c);                                 \
 290         qir_emit(c, qir_inst(QOP_##name, t, a, b));                      \
 291         return t;                                                        \
 292 }
 293
 294 #define QIR_NODST_1(name)                                               \
 295 static inline void                                                      \
 296 qir_##name(struct vc4_compile *c, struct qreg a)                        \
 297 {                                                                       \
 298         qir_emit(c, qir_inst(QOP_##name, c->undef, a, c->undef));       \
 299 }
 300
 301 #define QIR_NODST_2(name)                                               \
 302 static inline void                                                      \
 303 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b)         \
 304 {                                                                       \
 305         qir_emit(c, qir_inst(QOP_##name, c->undef, a, b));       \
 306 }
 307
 308 QIR_ALU1(MOV)
 309 QIR_ALU2(FADD)
 310 QIR_ALU2(FSUB)
 311 QIR_ALU2(FMUL)
 312 QIR_ALU2(MUL24)
 313 QIR_NODST_1(SF)
 314 QIR_ALU1(SEL_X_0_ZS)
 315 QIR_ALU1(SEL_X_0_ZC)
 316 QIR_ALU1(SEL_X_0_NS)
 317 QIR_ALU1(SEL_X_0_NC)
 318 QIR_ALU2(SEL_X_Y_ZS)
 319 QIR_ALU2(SEL_X_Y_ZC)
 320 QIR_ALU2(SEL_X_Y_NS)
 321 QIR_ALU2(SEL_X_Y_NC)
 322 QIR_ALU2(FMIN)
 323 QIR_ALU2(FMAX)
 324 QIR_ALU2(FMINABS)
 325 QIR_ALU2(FMAXABS)
 326 QIR_ALU1(FTOI)
 327 QIR_ALU1(ITOF)
 328
 329 QIR_ALU2(ADD)
 330 QIR_ALU2(SUB)
 331 QIR_ALU2(SHL)
 332 QIR_ALU2(SHR)
 333 QIR_ALU2(ASR)
 334 QIR_ALU2(MIN)
 335 QIR_ALU2(MAX)
 336 QIR_ALU2(AND)
 337 QIR_ALU2(OR)
 338 QIR_ALU2(XOR)
 339 QIR_ALU1(NOT)
 340
 341 QIR_ALU1(RCP)
 342 QIR_ALU1(RSQ)
 343 QIR_ALU1(EXP2)
 344 QIR_ALU1(LOG2)
 345 QIR_ALU2(PACK_SCALED)
 346 QIR_ALU1(VARY_ADD_C)
 347 QIR_NODST_1(VPM_WRITE)
 348 QIR_NODST_2(TEX_S)
 349 QIR_NODST_2(TEX_T)
 350 QIR_NODST_2(TEX_R)
 351 QIR_NODST_2(TEX_B)
 352 QIR_ALU0(FRAG_X)
 353 QIR_ALU0(FRAG_Y)
 354 QIR_ALU0(FRAG_Z)
 355 QIR_ALU0(FRAG_RCP_W)
 356 QIR_NODST_1(TLB_DISCARD_SETUP)
 357
 358 static inline struct qreg
 359 qir_R4_UNPACK(struct vc4_compile *c, int i)
 360 {
 361         struct qreg t = qir_get_temp(c);
 362         qir_emit(c, qir_inst(QOP_R4_UNPACK_A + i, t, c->undef, c->undef));
 363         return t;
 364 }
 365
 366 static inline struct qreg
 367 qir_SEL_X_0_COND(struct vc4_compile *c, int i)
 368 {
 369         struct qreg t = qir_get_temp(c);
 370         qir_emit(c, qir_inst(QOP_R4_UNPACK_A + i, t, c->undef, c->undef));
 371         return t;
 372 }
 373
 374 #endif /* VC4_QIR_H */