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
  32 #include "util/u_simple_list.h"
  33
  34 enum qfile {
  35         QFILE_NULL,
  36         QFILE_TEMP,
  37         QFILE_VARY,
  38         QFILE_UNIF,
  39 };
  40
  41 struct qreg {
  42         enum qfile file;
  43         uint32_t index;
  44 };
  45
  46 enum qop {
  47         QOP_UNDEF,
  48         QOP_MOV,
  49         QOP_FADD,
  50         QOP_FSUB,
  51         QOP_FMUL,
  52         QOP_MUL24,
  53         QOP_FMIN,
  54         QOP_FMAX,
  55         QOP_FMINABS,
  56         QOP_FMAXABS,
  57         QOP_ADD,
  58         QOP_SUB,
  59         QOP_SHL,
  60         QOP_SHR,
  61         QOP_ASR,
  62         QOP_MIN,
  63         QOP_MAX,
  64         QOP_AND,
  65         QOP_OR,
  66         QOP_XOR,
  67         QOP_NOT,
  68
  69         /* Sets the flag register according to src. */
  70         QOP_SF,
  71
  72         /* Note: Orderings of these compares must be the same as in
  73          * qpu_defines.h.  Selects the src[0] if the ns flag bit is set,
  74          * otherwise 0. */
  75         QOP_SEL_X_0_ZS,
  76         QOP_SEL_X_0_ZC,
  77         QOP_SEL_X_0_NS,
  78         QOP_SEL_X_0_NC,
  79         /* Selects the src[0] if the ns flag bit is set, otherwise src[1]. */
  80         QOP_SEL_X_Y_ZS,
  81         QOP_SEL_X_Y_ZC,
  82         QOP_SEL_X_Y_NS,
  83         QOP_SEL_X_Y_NC,
  84
  85         QOP_FTOI,
  86         QOP_ITOF,
  87         QOP_RCP,
  88         QOP_RSQ,
  89         QOP_EXP2,
  90         QOP_LOG2,
  91         QOP_VW_SETUP,
  92         QOP_VR_SETUP,
  93         QOP_PACK_SCALED,
  94         QOP_PACK_COLORS,
  95         QOP_VPM_WRITE,
  96         QOP_VPM_READ,
  97         QOP_TLB_DISCARD_SETUP,
  98         QOP_TLB_PASSTHROUGH_Z_WRITE,
  99         QOP_TLB_COLOR_WRITE,
 100         QOP_TLB_COLOR_READ,
 101         QOP_VARY_ADD_C,
 102
 103         QOP_FRAG_X,
 104         QOP_FRAG_Y,
 105         QOP_FRAG_Z,
 106         QOP_FRAG_RCP_W,
 107
 108         /** Texture x coordinate parameter write */
 109         QOP_TEX_S,
 110         /** Texture y coordinate parameter write */
 111         QOP_TEX_T,
 112         /** Texture border color parameter or cube map z coordinate write */
 113         QOP_TEX_R,
 114         /** Texture LOD bias parameter write */
 115         QOP_TEX_B,
 116         /**
 117          * Signal of texture read being necessary and then reading r4 into
 118          * the destination
 119          */
 120         QOP_TEX_RESULT,
 121         QOP_R4_UNPACK_A,
 122         QOP_R4_UNPACK_B,
 123         QOP_R4_UNPACK_C,
 124         QOP_R4_UNPACK_D
 125 };
 126
 127 struct simple_node {
 128         struct simple_node *next;
 129         struct simple_node *prev;
 130 };
 131
 132 struct qinst {
 133         struct simple_node link;
 134
 135         enum qop op;
 136         struct qreg dst;
 137         struct qreg *src;
 138 };
 139
 140 enum qstage {
 141         /**
 142          * Coordinate shader, runs during binning, before the VS, and just
 143          * outputs position.
 144          */
 145         QSTAGE_COORD,
 146         QSTAGE_VERT,
 147         QSTAGE_FRAG,
 148 };
 149
 150 enum quniform_contents {
 151         /**
 152          * Indicates that a constant 32-bit value is copied from the program's
 153          * uniform contents.
 154          */
 155         QUNIFORM_CONSTANT,
 156         /**
 157          * Indicates that the program's uniform contents are used as an index
 158          * into the GL uniform storage.
 159          */
 160         QUNIFORM_UNIFORM,
 161
 162         /** @{
 163          * Scaling factors from clip coordinates to relative to the viewport
 164          * center.
 165          *
 166          * This is used by the coordinate and vertex shaders to produce the
 167          * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed
 168          * point offsets from the viewport ccenter.
 169          */
 170         QUNIFORM_VIEWPORT_X_SCALE,
 171         QUNIFORM_VIEWPORT_Y_SCALE,
 172         /** @} */
 173
 174         QUNIFORM_VIEWPORT_Z_OFFSET,
 175         QUNIFORM_VIEWPORT_Z_SCALE,
 176
 177         /**
 178          * A reference to a texture config parameter 0 uniform.
 179          *
 180          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 181          * defines texture type, miplevels, and such.  It will be found as a
 182          * parameter to the first QOP_TEX_[STRB] instruction in a sequence.
 183          */
 184         QUNIFORM_TEXTURE_CONFIG_P0,
 185
 186         /**
 187          * A reference to a texture config parameter 1 uniform.
 188          *
 189          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 190          * defines texture width, height, filters, and wrap modes.  It will be
 191          * found as a parameter to the second QOP_TEX_[STRB] instruction in a
 192          * sequence.
 193          */
 194         QUNIFORM_TEXTURE_CONFIG_P1,
 195
 196         QUNIFORM_TEXRECT_SCALE_X,
 197         QUNIFORM_TEXRECT_SCALE_Y,
 198
 199         QUNIFORM_BLEND_CONST_COLOR,
 200 };
 201
 202 struct qcompile {
 203         struct qreg undef;
 204         enum qstage stage;
 205         uint32_t num_temps;
 206         struct simple_node instructions;
 207         uint32_t immediates[1024];
 208
 209         struct simple_node qpu_inst_list;
 210         uint64_t *qpu_insts;
 211         uint32_t qpu_inst_count;
 212         uint32_t qpu_inst_size;
 213         uint32_t num_inputs;
 214 };
 215
 216 struct qcompile *qir_compile_init(void);
 217 void qir_compile_destroy(struct qcompile *c);
 218 struct qinst *qir_inst(enum qop op, struct qreg dst,
 219                        struct qreg src0, struct qreg src1);
 220 struct qinst *qir_inst4(enum qop op, struct qreg dst,
 221                         struct qreg a,
 222                         struct qreg b,
 223                         struct qreg c,
 224                         struct qreg d);
 225 void qir_emit(struct qcompile *c, struct qinst *inst);
 226 struct qreg qir_get_temp(struct qcompile *c);
 227 int qir_get_op_nsrc(enum qop qop);
 228 bool qir_reg_equals(struct qreg a, struct qreg b);
 229 bool qir_has_side_effects(struct qinst *inst);
 230
 231 void qir_dump(struct qcompile *c);
 232 void qir_dump_inst(struct qinst *inst);
 233 const char *qir_get_stage_name(enum qstage stage);
 234
 235 void qir_optimize(struct qcompile *c);
 236 bool qir_opt_algebraic(struct qcompile *c);
 237 bool qir_opt_copy_propagation(struct qcompile *c);
 238 bool qir_opt_dead_code(struct qcompile *c);
 239
 240 #define QIR_ALU0(name)                                                   \
 241 static inline struct qreg                                                \
 242 qir_##name(struct qcompile *c)                                           \
 243 {                                                                        \
 244         struct qreg t = qir_get_temp(c);                                 \
 245         qir_emit(c, qir_inst(QOP_##name, t, c->undef, c->undef));        \
 246         return t;                                                        \
 247 }
 248
 249 #define QIR_ALU1(name)                                                   \
 250 static inline struct qreg                                                \
 251 qir_##name(struct qcompile *c, struct qreg a)                            \
 252 {                                                                        \
 253         struct qreg t = qir_get_temp(c);                                 \
 254         qir_emit(c, qir_inst(QOP_##name, t, a, c->undef));               \
 255         return t;                                                        \
 256 }
 257
 258 #define QIR_ALU2(name)                                                   \
 259 static inline struct qreg                                                \
 260 qir_##name(struct qcompile *c, struct qreg a, struct qreg b)             \
 261 {                                                                        \
 262         struct qreg t = qir_get_temp(c);                                 \
 263         qir_emit(c, qir_inst(QOP_##name, t, a, b));                      \
 264         return t;                                                        \
 265 }
 266
 267 #define QIR_NODST_1(name)                                               \
 268 static inline void                                                      \
 269 qir_##name(struct qcompile *c, struct qreg a)                           \
 270 {                                                                       \
 271         qir_emit(c, qir_inst(QOP_##name, c->undef, a, c->undef));       \
 272 }
 273
 274 #define QIR_NODST_2(name)                                               \
 275 static inline void                                                      \
 276 qir_##name(struct qcompile *c, struct qreg a, struct qreg b)            \
 277 {                                                                       \
 278         qir_emit(c, qir_inst(QOP_##name, c->undef, a, b));       \
 279 }
 280
 281 QIR_ALU1(MOV)
 282 QIR_ALU2(FADD)
 283 QIR_ALU2(FSUB)
 284 QIR_ALU2(FMUL)
 285 QIR_ALU2(MUL24)
 286 QIR_NODST_1(SF)
 287 QIR_ALU1(SEL_X_0_ZS)
 288 QIR_ALU1(SEL_X_0_ZC)
 289 QIR_ALU1(SEL_X_0_NS)
 290 QIR_ALU1(SEL_X_0_NC)
 291 QIR_ALU2(SEL_X_Y_ZS)
 292 QIR_ALU2(SEL_X_Y_ZC)
 293 QIR_ALU2(SEL_X_Y_NS)
 294 QIR_ALU2(SEL_X_Y_NC)
 295 QIR_ALU2(FMIN)
 296 QIR_ALU2(FMAX)
 297 QIR_ALU2(FMINABS)
 298 QIR_ALU2(FMAXABS)
 299 QIR_ALU1(FTOI)
 300 QIR_ALU1(ITOF)
 301
 302 QIR_ALU2(ADD)
 303 QIR_ALU2(SUB)
 304 QIR_ALU2(SHL)
 305 QIR_ALU2(SHR)
 306 QIR_ALU2(ASR)
 307 QIR_ALU2(MIN)
 308 QIR_ALU2(MAX)
 309 QIR_ALU2(AND)
 310 QIR_ALU2(OR)
 311 QIR_ALU2(XOR)
 312 QIR_ALU1(NOT)
 313
 314 QIR_ALU1(RCP)
 315 QIR_ALU1(RSQ)
 316 QIR_ALU1(EXP2)
 317 QIR_ALU1(LOG2)
 318 QIR_ALU2(PACK_SCALED)
 319 QIR_ALU1(VARY_ADD_C)
 320 QIR_NODST_1(VPM_WRITE)
 321 QIR_NODST_2(TEX_S)
 322 QIR_NODST_2(TEX_T)
 323 QIR_NODST_2(TEX_R)
 324 QIR_NODST_2(TEX_B)
 325 QIR_ALU0(FRAG_X)
 326 QIR_ALU0(FRAG_Y)
 327 QIR_ALU0(FRAG_Z)
 328 QIR_ALU0(FRAG_RCP_W)
 329 QIR_NODST_1(TLB_DISCARD_SETUP)
 330
 331 static inline struct qreg
 332 qir_R4_UNPACK(struct qcompile *c, int i)
 333 {
 334         struct qreg t = qir_get_temp(c);
 335         qir_emit(c, qir_inst(QOP_R4_UNPACK_A + i, t, c->undef, c->undef));
 336         return t;
 337 }
 338
 339 static inline struct qreg
 340 qir_SEL_X_0_COND(struct qcompile *c, int i)
 341 {
 342         struct qreg t = qir_get_temp(c);
 343         qir_emit(c, qir_inst(QOP_R4_UNPACK_A + i, t, c->undef, c->undef));
 344         return t;
 345 }
 346
 347 #endif /* VC4_QIR_H */