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_dp_read_message(struct brw_compile *p,
 209                              brw_inst *insn,
 210                              unsigned binding_table_index,
 211                              unsigned msg_control,
 212                              unsigned msg_type,
 213                              unsigned target_cache,
 214                              unsigned msg_length,
 215                              bool header_present,
 216                              unsigned response_length);
 217
 218 void brw_set_dp_write_message(struct brw_compile *p,
 219                               brw_inst *insn,
 220                               unsigned binding_table_index,
 221                               unsigned msg_control,
 222                               unsigned msg_type,
 223                               unsigned msg_length,
 224                               bool header_present,
 225                               unsigned last_render_target,
 226                               unsigned response_length,
 227                               unsigned end_of_thread,
 228                               unsigned send_commit_msg);
 229
 230 void brw_urb_WRITE(struct brw_compile *p,
 231                    struct brw_reg dest,
 232                    unsigned msg_reg_nr,
 233                    struct brw_reg src0,
 234                    enum brw_urb_write_flags flags,
 235                    unsigned msg_length,
 236                    unsigned response_length,
 237                    unsigned offset,
 238                    unsigned swizzle);
 239
 240 /**
 241  * Send message to shared unit \p sfid with a possibly indirect descriptor \p
 242  * desc.  If \p desc is not an immediate it will be transparently loaded to an
 243  * address register using an OR instruction.  The returned instruction can be
 244  * passed as argument to the usual brw_set_*_message() functions in order to
 245  * specify any additional descriptor bits -- If \p desc is an immediate this
 246  * will be the SEND instruction itself, otherwise it will be the OR
 247  * instruction.
 248  */
 249 struct brw_inst *
 250 brw_send_indirect_message(struct brw_compile *p,
 251                           unsigned sfid,
 252                           struct brw_reg dst,
 253                           struct brw_reg payload,
 254                           struct brw_reg desc);
 255
 256 void brw_ff_sync(struct brw_compile *p,
 257                    struct brw_reg dest,
 258                    unsigned msg_reg_nr,
 259                    struct brw_reg src0,
 260                    bool allocate,
 261                    unsigned response_length,
 262                    bool eot);
 263
 264 void brw_svb_write(struct brw_compile *p,
 265                    struct brw_reg dest,
 266                    unsigned msg_reg_nr,
 267                    struct brw_reg src0,
 268                    unsigned binding_table_index,
 269                    bool   send_commit_msg);
 270
 271 void brw_fb_WRITE(struct brw_compile *p,
 272                   int dispatch_width,
 273                    struct brw_reg payload,
 274                    struct brw_reg implied_header,
 275                    unsigned msg_control,
 276                    unsigned binding_table_index,
 277                    unsigned msg_length,
 278                    unsigned response_length,
 279                    bool eot,
 280                    bool last_render_target,
 281                    bool header_present);
 282
 283 void brw_SAMPLE(struct brw_compile *p,
 284                 struct brw_reg dest,
 285                 unsigned msg_reg_nr,
 286                 struct brw_reg src0,
 287                 unsigned binding_table_index,
 288                 unsigned sampler,
 289                 unsigned msg_type,
 290                 unsigned response_length,
 291                 unsigned msg_length,
 292                 unsigned header_present,
 293                 unsigned simd_mode,
 294                 unsigned return_format);
 295
 296 void brw_adjust_sampler_state_pointer(struct brw_compile *p,
 297                                       struct brw_reg header,
 298                                       struct brw_reg sampler_index);
 299
 300 void gen4_math(struct brw_compile *p,
 301                struct brw_reg dest,
 302                unsigned function,
 303                unsigned msg_reg_nr,
 304                struct brw_reg src,
 305                unsigned precision );
 306
 307 void gen6_math(struct brw_compile *p,
 308                struct brw_reg dest,
 309                unsigned function,
 310                struct brw_reg src0,
 311                struct brw_reg src1);
 312
 313 void brw_oword_block_read(struct brw_compile *p,
 314                           struct brw_reg dest,
 315                           struct brw_reg mrf,
 316                           uint32_t offset,
 317                           uint32_t bind_table_index);
 318
 319 void brw_oword_block_read_scratch(struct brw_compile *p,
 320                                   struct brw_reg dest,
 321                                   struct brw_reg mrf,
 322                                   int num_regs,
 323                                   unsigned offset);
 324
 325 void brw_oword_block_write_scratch(struct brw_compile *p,
 326                                    struct brw_reg mrf,
 327                                    int num_regs,
 328                                    unsigned offset);
 329
 330 void gen7_block_read_scratch(struct brw_compile *p,
 331                              struct brw_reg dest,
 332                              int num_regs,
 333                              unsigned offset);
 334
 335 void brw_shader_time_add(struct brw_compile *p,
 336                          struct brw_reg payload,
 337                          uint32_t surf_index);
 338
 339 /**
 340  * Return the generation-specific jump distance scaling factor.
 341  *
 342  * Given the number of instructions to jump, we need to scale by
 343  * some number to obtain the actual jump distance to program in an
 344  * instruction.
 345  */
 346 static inline unsigned
 347 brw_jump_scale(const struct brw_context *brw)
 348 {
 349    /* Broadwell measures jump targets in bytes. */
 350    if (brw->gen >= 8)
 351       return 16;
 352
 353    /* Ironlake and later measure jump targets in 64-bit data chunks (in order
 354     * (to support compaction), so each 128-bit instruction requires 2 chunks.
 355     */
 356    if (brw->gen >= 5)
 357       return 2;
 358
 359    /* Gen4 simply uses the number of 128-bit instructions. */
 360    return 1;
 361 }
 362
 363 /* If/else/endif.  Works by manipulating the execution flags on each
 364  * channel.
 365  */
 366 brw_inst *brw_IF(struct brw_compile *p, unsigned execute_size);
 367 brw_inst *gen6_IF(struct brw_compile *p, enum brw_conditional_mod conditional,
 368                   struct brw_reg src0, struct brw_reg src1);
 369
 370 void brw_ELSE(struct brw_compile *p);
 371 void brw_ENDIF(struct brw_compile *p);
 372
 373 /* DO/WHILE loops:
 374  */
 375 brw_inst *brw_DO(struct brw_compile *p, unsigned execute_size);
 376
 377 brw_inst *brw_WHILE(struct brw_compile *p);
 378
 379 brw_inst *brw_BREAK(struct brw_compile *p);
 380 brw_inst *brw_CONT(struct brw_compile *p);
 381 brw_inst *gen6_HALT(struct brw_compile *p);
 382
 383 /* Forward jumps:
 384  */
 385 void brw_land_fwd_jump(struct brw_compile *p, int jmp_insn_idx);
 386
 387 brw_inst *brw_JMPI(struct brw_compile *p, struct brw_reg index,
 388                    unsigned predicate_control);
 389
 390 void brw_NOP(struct brw_compile *p);
 391
 392 /* Special case: there is never a destination, execution size will be
 393  * taken from src0:
 394  */
 395 void brw_CMP(struct brw_compile *p,
 396              struct brw_reg dest,
 397              unsigned conditional,
 398              struct brw_reg src0,
 399              struct brw_reg src1);
 400
 401 void
 402 brw_untyped_atomic(struct brw_compile *p,
 403                    struct brw_reg dest,
 404                    struct brw_reg payload,
 405                    unsigned atomic_op,
 406                    unsigned bind_table_index,
 407                    unsigned msg_length,
 408                    bool response_expected);
 409
 410 void
 411 brw_untyped_surface_read(struct brw_compile *p,
 412                          struct brw_reg dest,
 413                          struct brw_reg mrf,
 414                          unsigned bind_table_index,
 415                          unsigned msg_length,
 416                          unsigned num_channels);
 417
 418 void
 419 brw_pixel_interpolator_query(struct brw_compile *p,
 420                              struct brw_reg dest,
 421                              struct brw_reg mrf,
 422                              bool noperspective,
 423                              unsigned mode,
 424                              unsigned data,
 425                              unsigned msg_length,
 426                              unsigned response_length);
 427
 428 /***********************************************************************
 429  * brw_eu_util.c:
 430  */
 431
 432 void brw_copy_indirect_to_indirect(struct brw_compile *p,
 433                                    struct brw_indirect dst_ptr,
 434                                    struct brw_indirect src_ptr,
 435                                    unsigned count);
 436
 437 void brw_copy_from_indirect(struct brw_compile *p,
 438                             struct brw_reg dst,
 439                             struct brw_indirect ptr,
 440                             unsigned count);
 441
 442 void brw_copy4(struct brw_compile *p,
 443                struct brw_reg dst,
 444                struct brw_reg src,
 445                unsigned count);
 446
 447 void brw_copy8(struct brw_compile *p,
 448                struct brw_reg dst,
 449                struct brw_reg src,
 450                unsigned count);
 451
 452 void brw_math_invert( struct brw_compile *p,
 453                       struct brw_reg dst,
 454                       struct brw_reg src);
 455
 456 void brw_set_src1(struct brw_compile *p, brw_inst *insn, struct brw_reg reg);
 457
 458 void brw_set_uip_jip(struct brw_compile *p);
 459
 460 enum brw_conditional_mod brw_negate_cmod(uint32_t cmod);
 461 enum brw_conditional_mod brw_swap_cmod(uint32_t cmod);
 462
 463 /* brw_eu_compact.c */
 464 void brw_init_compaction_tables(struct brw_context *brw);
 465 void brw_compact_instructions(struct brw_compile *p, int start_offset,
 466                               int num_annotations, struct annotation *annotation);
 467 void brw_uncompact_instruction(struct brw_context *brw, brw_inst *dst,
 468                                brw_compact_inst *src);
 469 bool brw_try_compact_instruction(struct brw_context *brw, brw_compact_inst *dst,
 470                                  brw_inst *src);
 471
 472 void brw_debug_compact_uncompact(struct brw_context *brw, brw_inst *orig,
 473                                  brw_inst *uncompacted);
 474
 475 static inline int
 476 next_offset(const struct brw_context *brw, void *store, int offset)
 477 {
 478    brw_inst *insn = (brw_inst *)((char *)store + offset);
 479
 480    if (brw_inst_cmpt_control(brw, insn))
 481       return offset + 8;
 482    else
 483       return offset + 16;
 484 }
 485
 486 #ifdef __cplusplus
 487 }
 488 #endif
 489
 490 #endif