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
  74    /* Control flow stacks:
  75     * - if_stack contains IF and ELSE instructions which must be patched
  76     *   (and popped) once the matching ENDIF instruction is encountered.
  77     *
  78     *   Just store the instruction pointer(an index).
  79     */
  80    int *if_stack;
  81    int if_stack_depth;
  82    int if_stack_array_size;
  83
  84    /**
  85     * loop_stack contains the instruction pointers of the starts of loops which
  86     * must be patched (and popped) once the matching WHILE instruction is
  87     * encountered.
  88     */
  89    int *loop_stack;
  90    /**
  91     * pre-gen6, the BREAK and CONT instructions had to tell how many IF/ENDIF
  92     * blocks they were popping out of, to fix up the mask stack.  This tracks
  93     * the IF/ENDIF nesting in each current nested loop level.
  94     */
  95    int *if_depth_in_loop;
  96    int loop_stack_depth;
  97    int loop_stack_array_size;
  98 };
  99
 100 void brw_pop_insn_state( struct brw_compile *p );
 101 void brw_push_insn_state( struct brw_compile *p );
 102 void brw_set_default_mask_control( struct brw_compile *p, unsigned value );
 103 void brw_set_default_saturate( struct brw_compile *p, bool enable );
 104 void brw_set_default_access_mode( struct brw_compile *p, unsigned access_mode );
 105 void brw_set_default_compression_control(struct brw_compile *p, enum brw_compression c);
 106 void brw_set_default_predicate_control( struct brw_compile *p, unsigned pc );
 107 void brw_set_default_predicate_inverse(struct brw_compile *p, bool predicate_inverse);
 108 void brw_set_default_flag_reg(struct brw_compile *p, int reg, int subreg);
 109 void brw_set_default_acc_write_control(struct brw_compile *p, unsigned value);
 110
 111 void brw_init_compile(struct brw_context *, struct brw_compile *p,
 112                       void *mem_ctx);
 113 void brw_disassemble(struct brw_context *brw, void *assembly,
 114                      int start, int end, FILE *out);
 115 const unsigned *brw_get_program( struct brw_compile *p, unsigned *sz );
 116
 117 brw_inst *brw_next_insn(struct brw_compile *p, unsigned opcode);
 118 void brw_set_dest(struct brw_compile *p, brw_inst *insn, struct brw_reg dest);
 119 void brw_set_src0(struct brw_compile *p, brw_inst *insn, struct brw_reg reg);
 120
 121 void gen6_resolve_implied_move(struct brw_compile *p,
 122                                struct brw_reg *src,
 123                                unsigned msg_reg_nr);
 124
 125 /* Helpers for regular instructions:
 126  */
 127 #define ALU1(OP)                                \
 128 brw_inst *brw_##OP(struct brw_compile *p,       \
 129               struct brw_reg dest,              \
 130               struct brw_reg src0);
 131
 132 #define ALU2(OP)                                \
 133 brw_inst *brw_##OP(struct brw_compile *p,       \
 134               struct brw_reg dest,              \
 135               struct brw_reg src0,              \
 136               struct brw_reg src1);
 137
 138 #define ALU3(OP)                                \
 139 brw_inst *brw_##OP(struct brw_compile *p,       \
 140               struct brw_reg dest,              \
 141               struct brw_reg src0,              \
 142               struct brw_reg src1,              \
 143               struct brw_reg src2);
 144
 145 #define ROUND(OP) \
 146 void brw_##OP(struct brw_compile *p, struct brw_reg dest, struct brw_reg src0);
 147
 148 ALU1(MOV)
 149 ALU2(SEL)
 150 ALU1(NOT)
 151 ALU2(AND)
 152 ALU2(OR)
 153 ALU2(XOR)
 154 ALU2(SHR)
 155 ALU2(SHL)
 156 ALU2(ASR)
 157 ALU1(F32TO16)
 158 ALU1(F16TO32)
 159 ALU2(ADD)
 160 ALU2(AVG)
 161 ALU2(MUL)
 162 ALU1(FRC)
 163 ALU1(RNDD)
 164 ALU2(MAC)
 165 ALU2(MACH)
 166 ALU1(LZD)
 167 ALU2(DP4)
 168 ALU2(DPH)
 169 ALU2(DP3)
 170 ALU2(DP2)
 171 ALU2(LINE)
 172 ALU2(PLN)
 173 ALU3(MAD)
 174 ALU3(LRP)
 175 ALU1(BFREV)
 176 ALU3(BFE)
 177 ALU2(BFI1)
 178 ALU3(BFI2)
 179 ALU1(FBH)
 180 ALU1(FBL)
 181 ALU1(CBIT)
 182 ALU2(ADDC)
 183 ALU2(SUBB)
 184 ALU2(MAC)
 185
 186 ROUND(RNDZ)
 187 ROUND(RNDE)
 188
 189 #undef ALU1
 190 #undef ALU2
 191 #undef ALU3
 192 #undef ROUND
 193
 194
 195 /* Helpers for SEND instruction:
 196  */
 197 void brw_set_sampler_message(struct brw_compile *p,
 198                              brw_inst *insn,
 199                              unsigned binding_table_index,
 200                              unsigned sampler,
 201                              unsigned msg_type,
 202                              unsigned response_length,
 203                              unsigned msg_length,
 204                              unsigned header_present,
 205                              unsigned simd_mode,
 206                              unsigned return_format);
 207
 208 void brw_set_indirect_send_descriptor(struct brw_compile *p,
 209                                       brw_inst *insn,
 210                                       unsigned sfid,
 211                                       struct brw_reg descriptor);
 212
 213 void brw_set_dp_read_message(struct brw_compile *p,
 214                              brw_inst *insn,
 215                              unsigned binding_table_index,
 216                              unsigned msg_control,
 217                              unsigned msg_type,
 218                              unsigned target_cache,
 219                              unsigned msg_length,
 220                              bool header_present,
 221                              unsigned response_length);
 222
 223 void brw_set_dp_write_message(struct brw_compile *p,
 224                               brw_inst *insn,
 225                               unsigned binding_table_index,
 226                               unsigned msg_control,
 227                               unsigned msg_type,
 228                               unsigned msg_length,
 229                               bool header_present,
 230                               unsigned last_render_target,
 231                               unsigned response_length,
 232                               unsigned end_of_thread,
 233                               unsigned send_commit_msg);
 234
 235 void brw_urb_WRITE(struct brw_compile *p,
 236                    struct brw_reg dest,
 237                    unsigned msg_reg_nr,
 238                    struct brw_reg src0,
 239                    enum brw_urb_write_flags flags,
 240                    unsigned msg_length,
 241                    unsigned response_length,
 242                    unsigned offset,
 243                    unsigned swizzle);
 244
 245 void brw_ff_sync(struct brw_compile *p,
 246                    struct brw_reg dest,
 247                    unsigned msg_reg_nr,
 248                    struct brw_reg src0,
 249                    bool allocate,
 250                    unsigned response_length,
 251                    bool eot);
 252
 253 void brw_svb_write(struct brw_compile *p,
 254                    struct brw_reg dest,
 255                    unsigned msg_reg_nr,
 256                    struct brw_reg src0,
 257                    unsigned binding_table_index,
 258                    bool   send_commit_msg);
 259
 260 void brw_fb_WRITE(struct brw_compile *p,
 261                   int dispatch_width,
 262                    unsigned msg_reg_nr,
 263                    struct brw_reg src0,
 264                    unsigned msg_control,
 265                    unsigned binding_table_index,
 266                    unsigned msg_length,
 267                    unsigned response_length,
 268                    bool eot,
 269                    bool header_present);
 270
 271 void brw_SAMPLE(struct brw_compile *p,
 272                 struct brw_reg dest,
 273                 unsigned msg_reg_nr,
 274                 struct brw_reg src0,
 275                 unsigned binding_table_index,
 276                 unsigned sampler,
 277                 unsigned msg_type,
 278                 unsigned response_length,
 279                 unsigned msg_length,
 280                 unsigned header_present,
 281                 unsigned simd_mode,
 282                 unsigned return_format);
 283
 284 void brw_adjust_sampler_state_pointer(struct brw_compile *p,
 285                                       struct brw_reg header,
 286                                       struct brw_reg sampler_index,
 287                                       struct brw_reg scratch);
 288
 289 void gen4_math(struct brw_compile *p,
 290                struct brw_reg dest,
 291                unsigned function,
 292                unsigned msg_reg_nr,
 293                struct brw_reg src,
 294                unsigned data_type,
 295                unsigned precision );
 296
 297 void gen6_math(struct brw_compile *p,
 298                struct brw_reg dest,
 299                unsigned function,
 300                struct brw_reg src0,
 301                struct brw_reg src1);
 302
 303 void brw_oword_block_read(struct brw_compile *p,
 304                           struct brw_reg dest,
 305                           struct brw_reg mrf,
 306                           uint32_t offset,
 307                           uint32_t bind_table_index);
 308
 309 void brw_oword_block_read_scratch(struct brw_compile *p,
 310                                   struct brw_reg dest,
 311                                   struct brw_reg mrf,
 312                                   int num_regs,
 313                                   unsigned offset);
 314
 315 void brw_oword_block_write_scratch(struct brw_compile *p,
 316                                    struct brw_reg mrf,
 317                                    int num_regs,
 318                                    unsigned offset);
 319
 320 void gen7_block_read_scratch(struct brw_compile *p,
 321                              struct brw_reg dest,
 322                              int num_regs,
 323                              unsigned offset);
 324
 325 void brw_shader_time_add(struct brw_compile *p,
 326                          struct brw_reg payload,
 327                          uint32_t surf_index);
 328
 329 /**
 330  * Return the generation-specific jump distance scaling factor.
 331  *
 332  * Given the number of instructions to jump, we need to scale by
 333  * some number to obtain the actual jump distance to program in an
 334  * instruction.
 335  */
 336 static inline unsigned
 337 brw_jump_scale(const struct brw_context *brw)
 338 {
 339    /* Broadwell measures jump targets in bytes. */
 340    if (brw->gen >= 8)
 341       return 16;
 342
 343    /* Ironlake and later measure jump targets in 64-bit data chunks (in order
 344     * (to support compaction), so each 128-bit instruction requires 2 chunks.
 345     */
 346    if (brw->gen >= 5)
 347       return 2;
 348
 349    /* Gen4 simply uses the number of 128-bit instructions. */
 350    return 1;
 351 }
 352
 353 /* If/else/endif.  Works by manipulating the execution flags on each
 354  * channel.
 355  */
 356 brw_inst *brw_IF(struct brw_compile *p, unsigned execute_size);
 357 brw_inst *gen6_IF(struct brw_compile *p, enum brw_conditional_mod conditional,
 358                   struct brw_reg src0, struct brw_reg src1);
 359
 360 void brw_ELSE(struct brw_compile *p);
 361 void brw_ENDIF(struct brw_compile *p);
 362
 363 /* DO/WHILE loops:
 364  */
 365 brw_inst *brw_DO(struct brw_compile *p, unsigned execute_size);
 366
 367 brw_inst *brw_WHILE(struct brw_compile *p);
 368
 369 brw_inst *brw_BREAK(struct brw_compile *p);
 370 brw_inst *brw_CONT(struct brw_compile *p);
 371 brw_inst *gen6_HALT(struct brw_compile *p);
 372
 373 /* Forward jumps:
 374  */
 375 void brw_land_fwd_jump(struct brw_compile *p, int jmp_insn_idx);
 376
 377 brw_inst *brw_JMPI(struct brw_compile *p, struct brw_reg index,
 378                    unsigned predicate_control);
 379
 380 void brw_NOP(struct brw_compile *p);
 381
 382 /* Special case: there is never a destination, execution size will be
 383  * taken from src0:
 384  */
 385 void brw_CMP(struct brw_compile *p,
 386              struct brw_reg dest,
 387              unsigned conditional,
 388              struct brw_reg src0,
 389              struct brw_reg src1);
 390
 391 void
 392 brw_untyped_atomic(struct brw_compile *p,
 393                    struct brw_reg dest,
 394                    struct brw_reg mrf,
 395                    unsigned atomic_op,
 396                    unsigned bind_table_index,
 397                    unsigned msg_length,
 398                    unsigned response_length);
 399
 400 void
 401 brw_untyped_surface_read(struct brw_compile *p,
 402                          struct brw_reg dest,
 403                          struct brw_reg mrf,
 404                          unsigned bind_table_index,
 405                          unsigned msg_length,
 406                          unsigned response_length);
 407
 408 void
 409 brw_pixel_interpolator_query(struct brw_compile *p,
 410                              struct brw_reg dest,
 411                              struct brw_reg mrf,
 412                              bool noperspective,
 413                              unsigned mode,
 414                              unsigned data,
 415                              unsigned msg_length,
 416                              unsigned response_length);
 417
 418 /***********************************************************************
 419  * brw_eu_util.c:
 420  */
 421
 422 void brw_copy_indirect_to_indirect(struct brw_compile *p,
 423                                    struct brw_indirect dst_ptr,
 424                                    struct brw_indirect src_ptr,
 425                                    unsigned count);
 426
 427 void brw_copy_from_indirect(struct brw_compile *p,
 428                             struct brw_reg dst,
 429                             struct brw_indirect ptr,
 430                             unsigned count);
 431
 432 void brw_copy4(struct brw_compile *p,
 433                struct brw_reg dst,
 434                struct brw_reg src,
 435                unsigned count);
 436
 437 void brw_copy8(struct brw_compile *p,
 438                struct brw_reg dst,
 439                struct brw_reg src,
 440                unsigned count);
 441
 442 void brw_math_invert( struct brw_compile *p,
 443                       struct brw_reg dst,
 444                       struct brw_reg src);
 445
 446 void brw_set_src1(struct brw_compile *p, brw_inst *insn, struct brw_reg reg);
 447
 448 void brw_set_uip_jip(struct brw_compile *p);
 449
 450 enum brw_conditional_mod brw_swap_cmod(uint32_t cmod);
 451
 452 /* brw_eu_compact.c */
 453 void brw_init_compaction_tables(struct brw_context *brw);
 454 void brw_compact_instructions(struct brw_compile *p, int start_offset,
 455                               int num_annotations, struct annotation *annotation);
 456 void brw_uncompact_instruction(struct brw_context *brw, brw_inst *dst,
 457                                brw_compact_inst *src);
 458 bool brw_try_compact_instruction(struct brw_context *brw, brw_compact_inst *dst,
 459                                  brw_inst *src);
 460
 461 void brw_debug_compact_uncompact(struct brw_context *brw, brw_inst *orig,
 462                                  brw_inst *uncompacted);
 463
 464 static inline int
 465 next_offset(const struct brw_context *brw, void *store, int offset)
 466 {
 467    brw_inst *insn = (brw_inst *)((char *)store + offset);
 468
 469    if (brw_inst_cmpt_control(brw, insn))
 470       return offset + 8;
 471    else
 472       return offset + 16;
 473 }
 474
 475 #ifdef __cplusplus
 476 }
 477 #endif
 478
 479 #endif