src/mesa/drivers/dri/i965/brw_eu.h

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics to
   4  develop this 3D driver.
   5
   6  Permission is hereby granted, free of charge, to any person obtaining
   7  a copy of this software and associated documentation files (the
   8  "Software"), to deal in the Software without restriction, including
   9  without limitation the rights to use, copy, modify, merge, publish,
  10  distribute, sublicense, and/or sell copies of the Software, and to
  11  permit persons to whom the Software is furnished to do so, subject to
  12  the following conditions:
  13
  14  The above copyright notice and this permission notice (including the
  15  next paragraph) shall be included in all copies or substantial
  16  portions of the Software.
  17
  18  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  19  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  21  IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  22  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  23  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  24  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25
  26  **********************************************************************/
  27  /*
  28   * Authors:
  29   *   Keith Whitwell <keithw@vmware.com>
  30   */
  31
  32
  33 #ifndef BRW_EU_H
  34 #define BRW_EU_H
  35
  36 #include <stdbool.h>
  37 #include "brw_inst.h"
  38 #include "brw_structs.h"
  39 #include "brw_defines.h"
  40 #include "brw_reg.h"
  41 #include "intel_asm_annotation.h"
  42 #include "program/prog_instruction.h"
  43
  44 #ifdef __cplusplus
  45 extern "C" {
  46 #endif
  47
  48 #define BRW_EU_MAX_INSN_STACK 5
  49
  50 /* A helper for accessing the last instruction emitted.  This makes it easy
  51  * to set various bits on an instruction without having to create temporary
  52  * variable and assign the emitted instruction to those.
  53  */
  54 #define brw_last_inst (&p->store[p->nr_insn - 1])
  55
  56 struct brw_compile {
  57    brw_inst *store;
  58    int store_size;
  59    unsigned nr_insn;
  60    unsigned int next_insn_offset;
  61
  62    void *mem_ctx;
  63
  64    /* Allow clients to push/pop instruction state:
  65     */
  66    brw_inst stack[BRW_EU_MAX_INSN_STACK];
  67    bool compressed_stack[BRW_EU_MAX_INSN_STACK];
  68    brw_inst *current;
  69
  70    bool single_program_flow;
  71    bool compressed;
  72    struct brw_context *brw;
  73    const struct brw_device_info *devinfo;
  74
  75    /* Control flow stacks:
  76     * - if_stack contains IF and ELSE instructions which must be patched
  77     *   (and popped) once the matching ENDIF instruction is encountered.
  78     *
  79     *   Just store the instruction pointer(an index).
  80     */
  81    int *if_stack;
  82    int if_stack_depth;
  83    int if_stack_array_size;
  84
  85    /**
  86     * loop_stack contains the instruction pointers of the starts of loops which
  87     * must be patched (and popped) once the matching WHILE instruction is
  88     * encountered.
  89     */
  90    int *loop_stack;
  91    /**
  92     * pre-gen6, the BREAK and CONT instructions had to tell how many IF/ENDIF
  93     * blocks they were popping out of, to fix up the mask stack.  This tracks
  94     * the IF/ENDIF nesting in each current nested loop level.
  95     */
  96    int *if_depth_in_loop;
  97    int loop_stack_depth;
  98    int loop_stack_array_size;
  99 };
 100
 101 void brw_pop_insn_state( struct brw_compile *p );
 102 void brw_push_insn_state( struct brw_compile *p );
 103 void brw_set_default_exec_size(struct brw_compile *p, unsigned value);
 104 void brw_set_default_mask_control( struct brw_compile *p, unsigned value );
 105 void brw_set_default_saturate( struct brw_compile *p, bool enable );
 106 void brw_set_default_access_mode( struct brw_compile *p, unsigned access_mode );
 107 void brw_set_default_compression_control(struct brw_compile *p, enum brw_compression c);
 108 void brw_set_default_predicate_control( struct brw_compile *p, unsigned pc );
 109 void brw_set_default_predicate_inverse(struct brw_compile *p, bool predicate_inverse);
 110 void brw_set_default_flag_reg(struct brw_compile *p, int reg, int subreg);
 111 void brw_set_default_acc_write_control(struct brw_compile *p, unsigned value);
 112
 113 void brw_init_compile(struct brw_context *, struct brw_compile *p,
 114                       void *mem_ctx);
 115 void brw_disassemble(struct brw_context *brw, void *assembly,
 116                      int start, int end, FILE *out);
 117 const unsigned *brw_get_program( struct brw_compile *p, unsigned *sz );
 118
 119 brw_inst *brw_next_insn(struct brw_compile *p, unsigned opcode);
 120 void brw_set_dest(struct brw_compile *p, brw_inst *insn, struct brw_reg dest);
 121 void brw_set_src0(struct brw_compile *p, brw_inst *insn, struct brw_reg reg);
 122
 123 void gen6_resolve_implied_move(struct brw_compile *p,
 124                                struct brw_reg *src,
 125                                unsigned msg_reg_nr);
 126
 127 /* Helpers for regular instructions:
 128  */
 129 #define ALU1(OP)                                \
 130 brw_inst *brw_##OP(struct brw_compile *p,       \
 131               struct brw_reg dest,              \
 132               struct brw_reg src0);
 133
 134 #define ALU2(OP)                                \
 135 brw_inst *brw_##OP(struct brw_compile *p,       \
 136               struct brw_reg dest,              \
 137               struct brw_reg src0,              \
 138               struct brw_reg src1);
 139
 140 #define ALU3(OP)                                \
 141 brw_inst *brw_##OP(struct brw_compile *p,       \
 142               struct brw_reg dest,              \
 143               struct brw_reg src0,              \
 144               struct brw_reg src1,              \
 145               struct brw_reg src2);
 146
 147 #define ROUND(OP) \
 148 void brw_##OP(struct brw_compile *p, struct brw_reg dest, struct brw_reg src0);
 149
 150 ALU1(MOV)
 151 ALU2(SEL)
 152 ALU1(NOT)
 153 ALU2(AND)
 154 ALU2(OR)
 155 ALU2(XOR)
 156 ALU2(SHR)
 157 ALU2(SHL)
 158 ALU2(ASR)
 159 ALU1(F32TO16)
 160 ALU1(F16TO32)
 161 ALU2(ADD)
 162 ALU2(AVG)
 163 ALU2(MUL)
 164 ALU1(FRC)
 165 ALU1(RNDD)
 166 ALU2(MAC)
 167 ALU2(MACH)
 168 ALU1(LZD)
 169 ALU2(DP4)
 170 ALU2(DPH)
 171 ALU2(DP3)
 172 ALU2(DP2)
 173 ALU2(LINE)
 174 ALU2(PLN)
 175 ALU3(MAD)
 176 ALU3(LRP)
 177 ALU1(BFREV)
 178 ALU3(BFE)
 179 ALU2(BFI1)
 180 ALU3(BFI2)
 181 ALU1(FBH)
 182 ALU1(FBL)
 183 ALU1(CBIT)
 184 ALU2(ADDC)
 185 ALU2(SUBB)
 186 ALU2(MAC)
 187
 188 ROUND(RNDZ)
 189 ROUND(RNDE)
 190
 191 #undef ALU1
 192 #undef ALU2
 193 #undef ALU3
 194 #undef ROUND
 195
 196
 197 /* Helpers for SEND instruction:
 198  */
 199 void brw_set_sampler_message(struct brw_compile *p,
 200                              brw_inst *insn,
 201                              unsigned binding_table_index,
 202                              unsigned sampler,
 203                              unsigned msg_type,
 204                              unsigned response_length,
 205                              unsigned msg_length,
 206                              unsigned header_present,
 207                              unsigned simd_mode,
 208                              unsigned return_format);
 209
 210 void brw_set_dp_read_message(struct brw_compile *p,
 211                              brw_inst *insn,
 212                              unsigned binding_table_index,
 213                              unsigned msg_control,
 214                              unsigned msg_type,
 215                              unsigned target_cache,
 216                              unsigned msg_length,
 217                              bool header_present,
 218                              unsigned response_length);
 219
 220 void brw_set_dp_write_message(struct brw_compile *p,
 221                               brw_inst *insn,
 222                               unsigned binding_table_index,
 223                               unsigned msg_control,
 224                               unsigned msg_type,
 225                               unsigned msg_length,
 226                               bool header_present,
 227                               unsigned last_render_target,
 228                               unsigned response_length,
 229                               unsigned end_of_thread,
 230                               unsigned send_commit_msg);
 231
 232 void brw_urb_WRITE(struct brw_compile *p,
 233                    struct brw_reg dest,
 234                    unsigned msg_reg_nr,
 235                    struct brw_reg src0,
 236                    enum brw_urb_write_flags flags,
 237                    unsigned msg_length,
 238                    unsigned response_length,
 239                    unsigned offset,
 240                    unsigned swizzle);
 241
 242 /**
 243  * Send message to shared unit \p sfid with a possibly indirect descriptor \p
 244  * desc.  If \p desc is not an immediate it will be transparently loaded to an
 245  * address register using an OR instruction.  The returned instruction can be
 246  * passed as argument to the usual brw_set_*_message() functions in order to
 247  * specify any additional descriptor bits -- If \p desc is an immediate this
 248  * will be the SEND instruction itself, otherwise it will be the OR
 249  * instruction.
 250  */
 251 struct brw_inst *
 252 brw_send_indirect_message(struct brw_compile *p,
 253                           unsigned sfid,
 254                           struct brw_reg dst,
 255                           struct brw_reg payload,
 256                           struct brw_reg desc);
 257
 258 void brw_ff_sync(struct brw_compile *p,
 259                    struct brw_reg dest,
 260                    unsigned msg_reg_nr,
 261                    struct brw_reg src0,
 262                    bool allocate,
 263                    unsigned response_length,
 264                    bool eot);
 265
 266 void brw_svb_write(struct brw_compile *p,
 267                    struct brw_reg dest,
 268                    unsigned msg_reg_nr,
 269                    struct brw_reg src0,
 270                    unsigned binding_table_index,
 271                    bool   send_commit_msg);
 272
 273 void brw_fb_WRITE(struct brw_compile *p,
 274                   int dispatch_width,
 275                    struct brw_reg payload,
 276                    struct brw_reg implied_header,
 277                    unsigned msg_control,
 278                    unsigned binding_table_index,
 279                    unsigned msg_length,
 280                    unsigned response_length,
 281                    bool eot,
 282                    bool last_render_target,
 283                    bool header_present);
 284
 285 void brw_SAMPLE(struct brw_compile *p,
 286                 struct brw_reg dest,
 287                 unsigned msg_reg_nr,
 288                 struct brw_reg src0,
 289                 unsigned binding_table_index,
 290                 unsigned sampler,
 291                 unsigned msg_type,
 292                 unsigned response_length,
 293                 unsigned msg_length,
 294                 unsigned header_present,
 295                 unsigned simd_mode,
 296                 unsigned return_format);
 297
 298 void brw_adjust_sampler_state_pointer(struct brw_compile *p,
 299                                       struct brw_reg header,
 300                                       struct brw_reg sampler_index);
 301
 302 void gen4_math(struct brw_compile *p,
 303                struct brw_reg dest,
 304                unsigned function,
 305                unsigned msg_reg_nr,
 306                struct brw_reg src,
 307                unsigned precision );
 308
 309 void gen6_math(struct brw_compile *p,
 310                struct brw_reg dest,
 311                unsigned function,
 312                struct brw_reg src0,
 313                struct brw_reg src1);
 314
 315 void brw_oword_block_read(struct brw_compile *p,
 316                           struct brw_reg dest,
 317                           struct brw_reg mrf,
 318                           uint32_t offset,
 319                           uint32_t bind_table_index);
 320
 321 void brw_oword_block_read_scratch(struct brw_compile *p,
 322                                   struct brw_reg dest,
 323                                   struct brw_reg mrf,
 324                                   int num_regs,
 325                                   unsigned offset);
 326
 327 void brw_oword_block_write_scratch(struct brw_compile *p,
 328                                    struct brw_reg mrf,
 329                                    int num_regs,
 330                                    unsigned offset);
 331
 332 void gen7_block_read_scratch(struct brw_compile *p,
 333                              struct brw_reg dest,
 334                              int num_regs,
 335                              unsigned offset);
 336
 337 void brw_shader_time_add(struct brw_compile *p,
 338                          struct brw_reg payload,
 339                          uint32_t surf_index);
 340
 341 /**
 342  * Return the generation-specific jump distance scaling factor.
 343  *
 344  * Given the number of instructions to jump, we need to scale by
 345  * some number to obtain the actual jump distance to program in an
 346  * instruction.
 347  */
 348 static inline unsigned
 349 brw_jump_scale(const struct brw_context *brw)
 350 {
 351    /* Broadwell measures jump targets in bytes. */
 352    if (brw->gen >= 8)
 353       return 16;
 354
 355    /* Ironlake and later measure jump targets in 64-bit data chunks (in order
 356     * (to support compaction), so each 128-bit instruction requires 2 chunks.
 357     */
 358    if (brw->gen >= 5)
 359       return 2;
 360
 361    /* Gen4 simply uses the number of 128-bit instructions. */
 362    return 1;
 363 }
 364
 365 /* If/else/endif.  Works by manipulating the execution flags on each
 366  * channel.
 367  */
 368 brw_inst *brw_IF(struct brw_compile *p, unsigned execute_size);
 369 brw_inst *gen6_IF(struct brw_compile *p, enum brw_conditional_mod conditional,
 370                   struct brw_reg src0, struct brw_reg src1);
 371
 372 void brw_ELSE(struct brw_compile *p);
 373 void brw_ENDIF(struct brw_compile *p);
 374
 375 /* DO/WHILE loops:
 376  */
 377 brw_inst *brw_DO(struct brw_compile *p, unsigned execute_size);
 378
 379 brw_inst *brw_WHILE(struct brw_compile *p);
 380
 381 brw_inst *brw_BREAK(struct brw_compile *p);
 382 brw_inst *brw_CONT(struct brw_compile *p);
 383 brw_inst *gen6_HALT(struct brw_compile *p);
 384
 385 /* Forward jumps:
 386  */
 387 void brw_land_fwd_jump(struct brw_compile *p, int jmp_insn_idx);
 388
 389 brw_inst *brw_JMPI(struct brw_compile *p, struct brw_reg index,
 390                    unsigned predicate_control);
 391
 392 void brw_NOP(struct brw_compile *p);
 393
 394 /* Special case: there is never a destination, execution size will be
 395  * taken from src0:
 396  */
 397 void brw_CMP(struct brw_compile *p,
 398              struct brw_reg dest,
 399              unsigned conditional,
 400              struct brw_reg src0,
 401              struct brw_reg src1);
 402
 403 void
 404 brw_untyped_atomic(struct brw_compile *p,
 405                    struct brw_reg dest,
 406                    struct brw_reg payload,
 407                    unsigned atomic_op,
 408                    unsigned bind_table_index,
 409                    unsigned msg_length,
 410                    bool response_expected);
 411
 412 void
 413 brw_untyped_surface_read(struct brw_compile *p,
 414                          struct brw_reg dest,
 415                          struct brw_reg mrf,
 416                          unsigned bind_table_index,
 417                          unsigned msg_length,
 418                          unsigned num_channels);
 419
 420 void
 421 brw_pixel_interpolator_query(struct brw_compile *p,
 422                              struct brw_reg dest,
 423                              struct brw_reg mrf,
 424                              bool noperspective,
 425                              unsigned mode,
 426                              unsigned data,
 427                              unsigned msg_length,
 428                              unsigned response_length);
 429
 430 /***********************************************************************
 431  * brw_eu_util.c:
 432  */
 433
 434 void brw_copy_indirect_to_indirect(struct brw_compile *p,
 435                                    struct brw_indirect dst_ptr,
 436                                    struct brw_indirect src_ptr,
 437                                    unsigned count);
 438
 439 void brw_copy_from_indirect(struct brw_compile *p,
 440                             struct brw_reg dst,
 441                             struct brw_indirect ptr,
 442                             unsigned count);
 443
 444 void brw_copy4(struct brw_compile *p,
 445                struct brw_reg dst,
 446                struct brw_reg src,
 447                unsigned count);
 448
 449 void brw_copy8(struct brw_compile *p,
 450                struct brw_reg dst,
 451                struct brw_reg src,
 452                unsigned count);
 453
 454 void brw_math_invert( struct brw_compile *p,
 455                       struct brw_reg dst,
 456                       struct brw_reg src);
 457
 458 void brw_set_src1(struct brw_compile *p, brw_inst *insn, struct brw_reg reg);
 459
 460 void brw_set_uip_jip(struct brw_compile *p);
 461
 462 enum brw_conditional_mod brw_negate_cmod(uint32_t cmod);
 463 enum brw_conditional_mod brw_swap_cmod(uint32_t cmod);
 464
 465 /* brw_eu_compact.c */
 466 void brw_init_compaction_tables(struct brw_context *brw);
 467 void brw_compact_instructions(struct brw_compile *p, int start_offset,
 468                               int num_annotations, struct annotation *annotation);
 469 void brw_uncompact_instruction(struct brw_context *brw, brw_inst *dst,
 470                                brw_compact_inst *src);
 471 bool brw_try_compact_instruction(struct brw_context *brw, brw_compact_inst *dst,
 472                                  brw_inst *src);
 473
 474 void brw_debug_compact_uncompact(struct brw_context *brw, brw_inst *orig,
 475                                  brw_inst *uncompacted);
 476
 477 static inline int
 478 next_offset(const struct brw_context *brw, void *store, int offset)
 479 {
 480    brw_inst *insn = (brw_inst *)((char *)store + offset);
 481
 482    if (brw_inst_cmpt_control(brw, insn))
 483       return offset + 8;
 484    else
 485       return offset + 16;
 486 }
 487
 488 #ifdef __cplusplus
 489 }
 490 #endif
 491
 492 #endif