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
  35 enum qfile {
  36         QFILE_NULL,
  37         QFILE_TEMP,
  38         QFILE_VARY,
  39         QFILE_UNIF,
  40 };
  41
  42 struct qreg {
  43         enum qfile file;
  44         uint32_t index;
  45 };
  46
  47 enum qop {
  48         QOP_UNDEF,
  49         QOP_MOV,
  50         QOP_FADD,
  51         QOP_FSUB,
  52         QOP_FMUL,
  53         QOP_MUL24,
  54         QOP_FMIN,
  55         QOP_FMAX,
  56         QOP_FMINABS,
  57         QOP_FMAXABS,
  58         QOP_ADD,
  59         QOP_SUB,
  60         QOP_SHL,
  61         QOP_SHR,
  62         QOP_ASR,
  63         QOP_MIN,
  64         QOP_MAX,
  65         QOP_AND,
  66         QOP_OR,
  67         QOP_XOR,
  68         QOP_NOT,
  69
  70         /* Sets the flag register according to src. */
  71         QOP_SF,
  72
  73         /* Note: Orderings of these compares must be the same as in
  74          * qpu_defines.h.  Selects the src[0] if the ns flag bit is set,
  75          * otherwise 0. */
  76         QOP_SEL_X_0_ZS,
  77         QOP_SEL_X_0_ZC,
  78         QOP_SEL_X_0_NS,
  79         QOP_SEL_X_0_NC,
  80         /* Selects the src[0] if the ns flag bit is set, otherwise src[1]. */
  81         QOP_SEL_X_Y_ZS,
  82         QOP_SEL_X_Y_ZC,
  83         QOP_SEL_X_Y_NS,
  84         QOP_SEL_X_Y_NC,
  85
  86         QOP_FTOI,
  87         QOP_ITOF,
  88         QOP_RCP,
  89         QOP_RSQ,
  90         QOP_EXP2,
  91         QOP_LOG2,
  92         QOP_VW_SETUP,
  93         QOP_VR_SETUP,
  94         QOP_PACK_SCALED,
  95         QOP_PACK_COLORS,
  96         QOP_VPM_WRITE,
  97         QOP_VPM_READ,
  98         QOP_TLB_DISCARD_SETUP,
  99         QOP_TLB_PASSTHROUGH_Z_WRITE,
 100         QOP_TLB_COLOR_WRITE,
 101         QOP_TLB_COLOR_READ,
 102         QOP_VARY_ADD_C,
 103
 104         QOP_FRAG_X,
 105         QOP_FRAG_Y,
 106         QOP_FRAG_Z,
 107         QOP_FRAG_RCP_W,
 108
 109         /** Texture x coordinate parameter write */
 110         QOP_TEX_S,
 111         /** Texture y coordinate parameter write */
 112         QOP_TEX_T,
 113         /** Texture border color parameter or cube map z coordinate write */
 114         QOP_TEX_R,
 115         /** Texture LOD bias parameter write */
 116         QOP_TEX_B,
 117         /**
 118          * Signal of texture read being necessary and then reading r4 into
 119          * the destination
 120          */
 121         QOP_TEX_RESULT,
 122         QOP_R4_UNPACK_A,
 123         QOP_R4_UNPACK_B,
 124         QOP_R4_UNPACK_C,
 125         QOP_R4_UNPACK_D
 126 };
 127
 128 struct simple_node {
 129         struct simple_node *next;
 130         struct simple_node *prev;
 131 };
 132
 133 struct qinst {
 134         struct simple_node link;
 135
 136         enum qop op;
 137         struct qreg dst;
 138         struct qreg *src;
 139 };
 140
 141 enum qstage {
 142         /**
 143          * Coordinate shader, runs during binning, before the VS, and just
 144          * outputs position.
 145          */
 146         QSTAGE_COORD,
 147         QSTAGE_VERT,
 148         QSTAGE_FRAG,
 149 };
 150
 151 enum quniform_contents {
 152         /**
 153          * Indicates that a constant 32-bit value is copied from the program's
 154          * uniform contents.
 155          */
 156         QUNIFORM_CONSTANT,
 157         /**
 158          * Indicates that the program's uniform contents are used as an index
 159          * into the GL uniform storage.
 160          */
 161         QUNIFORM_UNIFORM,
 162
 163         /** @{
 164          * Scaling factors from clip coordinates to relative to the viewport
 165          * center.
 166          *
 167          * This is used by the coordinate and vertex shaders to produce the
 168          * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed
 169          * point offsets from the viewport ccenter.
 170          */
 171         QUNIFORM_VIEWPORT_X_SCALE,
 172         QUNIFORM_VIEWPORT_Y_SCALE,
 173         /** @} */
 174
 175         QUNIFORM_VIEWPORT_Z_OFFSET,
 176         QUNIFORM_VIEWPORT_Z_SCALE,
 177
 178         /**
 179          * A reference to a texture config parameter 0 uniform.
 180          *
 181          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 182          * defines texture type, miplevels, and such.  It will be found as a
 183          * parameter to the first QOP_TEX_[STRB] instruction in a sequence.
 184          */
 185         QUNIFORM_TEXTURE_CONFIG_P0,
 186
 187         /**
 188          * A reference to a texture config parameter 1 uniform.
 189          *
 190          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
 191          * defines texture width, height, filters, and wrap modes.  It will be
 192          * found as a parameter to the second QOP_TEX_[STRB] instruction in a
 193          * sequence.
 194          */
 195         QUNIFORM_TEXTURE_CONFIG_P1,
 196
 197         QUNIFORM_TEXRECT_SCALE_X,
 198         QUNIFORM_TEXRECT_SCALE_Y,
 199
 200         QUNIFORM_BLEND_CONST_COLOR,
 201 };
 202
 203 struct qcompile {
 204         struct qreg undef;
 205         enum qstage stage;
 206         uint32_t num_temps;
 207         struct simple_node instructions;
 208         uint32_t immediates[1024];
 209
 210         struct simple_node qpu_inst_list;
 211         uint64_t *qpu_insts;
 212         uint32_t qpu_inst_count;
 213         uint32_t qpu_inst_size;
 214         uint32_t num_inputs;
 215 };
 216
 217 struct qcompile *qir_compile_init(void);
 218 void qir_compile_destroy(struct qcompile *c);
 219 struct qinst *qir_inst(enum qop op, struct qreg dst,
 220                        struct qreg src0, struct qreg src1);
 221 struct qinst *qir_inst4(enum qop op, struct qreg dst,
 222                         struct qreg a,
 223                         struct qreg b,
 224                         struct qreg c,
 225                         struct qreg d);
 226 void qir_emit(struct qcompile *c, struct qinst *inst);
 227 struct qreg qir_get_temp(struct qcompile *c);
 228 int qir_get_op_nsrc(enum qop qop);
 229 bool qir_reg_equals(struct qreg a, struct qreg b);
 230 bool qir_has_side_effects(struct qinst *inst);
 231 bool qir_depends_on_flags(struct qinst *inst);
 232 bool qir_writes_r4(struct qinst *inst);
 233 bool qir_reads_r4(struct qinst *inst);
 234
 235 void qir_dump(struct qcompile *c);
 236 void qir_dump_inst(struct qinst *inst);
 237 const char *qir_get_stage_name(enum qstage stage);
 238
 239 void qir_optimize(struct qcompile *c);
 240 bool qir_opt_algebraic(struct qcompile *c);
 241 bool qir_opt_copy_propagation(struct qcompile *c);
 242 bool qir_opt_cse(struct qcompile *c);
 243 bool qir_opt_dead_code(struct qcompile *c);
 244
 245 #define QIR_ALU0(name)                                                   \
 246 static inline struct qreg                                                \
 247 qir_##name(struct qcompile *c)                                           \
 248 {                                                                        \
 249         struct qreg t = qir_get_temp(c);                                 \
 250         qir_emit(c, qir_inst(QOP_##name, t, c->undef, c->undef));        \
 251         return t;                                                        \
 252 }
 253
 254 #define QIR_ALU1(name)                                                   \
 255 static inline struct qreg                                                \
 256 qir_##name(struct qcompile *c, struct qreg a)                            \
 257 {                                                                        \
 258         struct qreg t = qir_get_temp(c);                                 \
 259         qir_emit(c, qir_inst(QOP_##name, t, a, c->undef));               \
 260         return t;                                                        \
 261 }
 262
 263 #define QIR_ALU2(name)                                                   \
 264 static inline struct qreg                                                \
 265 qir_##name(struct qcompile *c, struct qreg a, struct qreg b)             \
 266 {                                                                        \
 267         struct qreg t = qir_get_temp(c);                                 \
 268         qir_emit(c, qir_inst(QOP_##name, t, a, b));                      \
 269         return t;                                                        \
 270 }
 271
 272 #define QIR_NODST_1(name)                                               \
 273 static inline void                                                      \
 274 qir_##name(struct qcompile *c, struct qreg a)                           \
 275 {                                                                       \
 276         qir_emit(c, qir_inst(QOP_##name, c->undef, a, c->undef));       \
 277 }
 278
 279 #define QIR_NODST_2(name)                                               \
 280 static inline void                                                      \
 281 qir_##name(struct qcompile *c, struct qreg a, struct qreg b)            \
 282 {                                                                       \
 283         qir_emit(c, qir_inst(QOP_##name, c->undef, a, b));       \
 284 }
 285
 286 QIR_ALU1(MOV)
 287 QIR_ALU2(FADD)
 288 QIR_ALU2(FSUB)
 289 QIR_ALU2(FMUL)
 290 QIR_ALU2(MUL24)
 291 QIR_NODST_1(SF)
 292 QIR_ALU1(SEL_X_0_ZS)
 293 QIR_ALU1(SEL_X_0_ZC)
 294 QIR_ALU1(SEL_X_0_NS)
 295 QIR_ALU1(SEL_X_0_NC)
 296 QIR_ALU2(SEL_X_Y_ZS)
 297 QIR_ALU2(SEL_X_Y_ZC)
 298 QIR_ALU2(SEL_X_Y_NS)
 299 QIR_ALU2(SEL_X_Y_NC)
 300 QIR_ALU2(FMIN)
 301 QIR_ALU2(FMAX)
 302 QIR_ALU2(FMINABS)
 303 QIR_ALU2(FMAXABS)
 304 QIR_ALU1(FTOI)
 305 QIR_ALU1(ITOF)
 306
 307 QIR_ALU2(ADD)
 308 QIR_ALU2(SUB)
 309 QIR_ALU2(SHL)
 310 QIR_ALU2(SHR)
 311 QIR_ALU2(ASR)
 312 QIR_ALU2(MIN)
 313 QIR_ALU2(MAX)
 314 QIR_ALU2(AND)
 315 QIR_ALU2(OR)
 316 QIR_ALU2(XOR)
 317 QIR_ALU1(NOT)
 318
 319 QIR_ALU1(RCP)
 320 QIR_ALU1(RSQ)
 321 QIR_ALU1(EXP2)
 322 QIR_ALU1(LOG2)
 323 QIR_ALU2(PACK_SCALED)
 324 QIR_ALU1(VARY_ADD_C)
 325 QIR_NODST_1(VPM_WRITE)
 326 QIR_NODST_2(TEX_S)
 327 QIR_NODST_2(TEX_T)
 328 QIR_NODST_2(TEX_R)
 329 QIR_NODST_2(TEX_B)
 330 QIR_ALU0(FRAG_X)
 331 QIR_ALU0(FRAG_Y)
 332 QIR_ALU0(FRAG_Z)
 333 QIR_ALU0(FRAG_RCP_W)
 334 QIR_NODST_1(TLB_DISCARD_SETUP)
 335
 336 static inline struct qreg
 337 qir_R4_UNPACK(struct qcompile *c, int i)
 338 {
 339         struct qreg t = qir_get_temp(c);
 340         qir_emit(c, qir_inst(QOP_R4_UNPACK_A + i, t, c->undef, c->undef));
 341         return t;
 342 }
 343
 344 static inline struct qreg
 345 qir_SEL_X_0_COND(struct qcompile *c, int i)
 346 {
 347         struct qreg t = qir_get_temp(c);
 348         qir_emit(c, qir_inst(QOP_R4_UNPACK_A + i, t, c->undef, c->undef));
 349         return t;
 350 }
 351
 352 #endif /* VC4_QIR_H */