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

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics (http://www.tungstengraphics.com) 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 <keith@tungstengraphics.com>
  30   */
  31
  32
  33 #ifndef BRW_EU_H
  34 #define BRW_EU_H
  35
  36 #include <stdbool.h>
  37 #include "brw_structs.h"
  38 #include "brw_defines.h"
  39 #include "brw_reg.h"
  40 #include "program/prog_instruction.h"
  41
  42 #ifdef __cplusplus
  43 extern "C" {
  44 #endif
  45
  46 #define BRW_EU_MAX_INSN_STACK 5
  47
  48 struct brw_compile {
  49    struct brw_instruction *store;
  50    int store_size;
  51    GLuint nr_insn;
  52    unsigned int next_insn_offset;
  53
  54    void *mem_ctx;
  55
  56    /* Allow clients to push/pop instruction state:
  57     */
  58    struct brw_instruction stack[BRW_EU_MAX_INSN_STACK];
  59    bool compressed_stack[BRW_EU_MAX_INSN_STACK];
  60    struct brw_instruction *current;
  61
  62    GLuint flag_value;
  63    bool single_program_flow;
  64    bool compressed;
  65    struct brw_context *brw;
  66
  67    /* Control flow stacks:
  68     * - if_stack contains IF and ELSE instructions which must be patched
  69     *   (and popped) once the matching ENDIF instruction is encountered.
  70     *
  71     *   Just store the instruction pointer(an index).
  72     */
  73    int *if_stack;
  74    int if_stack_depth;
  75    int if_stack_array_size;
  76
  77    /**
  78     * loop_stack contains the instruction pointers of the starts of loops which
  79     * must be patched (and popped) once the matching WHILE instruction is
  80     * encountered.
  81     */
  82    int *loop_stack;
  83    /**
  84     * pre-gen6, the BREAK and CONT instructions had to tell how many IF/ENDIF
  85     * blocks they were popping out of, to fix up the mask stack.  This tracks
  86     * the IF/ENDIF nesting in each current nested loop level.
  87     */
  88    int *if_depth_in_loop;
  89    int loop_stack_depth;
  90    int loop_stack_array_size;
  91 };
  92
  93 static INLINE struct brw_instruction *current_insn( struct brw_compile *p)
  94 {
  95    return &p->store[p->nr_insn];
  96 }
  97
  98 void brw_pop_insn_state( struct brw_compile *p );
  99 void brw_push_insn_state( struct brw_compile *p );
 100 void brw_set_mask_control( struct brw_compile *p, GLuint value );
 101 void brw_set_saturate( struct brw_compile *p, bool enable );
 102 void brw_set_access_mode( struct brw_compile *p, GLuint access_mode );
 103 void brw_set_compression_control(struct brw_compile *p, enum brw_compression c);
 104 void brw_set_predicate_control_flag_value( struct brw_compile *p, GLuint value );
 105 void brw_set_predicate_control( struct brw_compile *p, GLuint pc );
 106 void brw_set_predicate_inverse(struct brw_compile *p, bool predicate_inverse);
 107 void brw_set_conditionalmod( struct brw_compile *p, GLuint conditional );
 108 void brw_set_flag_reg(struct brw_compile *p, int reg, int subreg);
 109 void brw_set_acc_write_control(struct brw_compile *p, GLuint value);
 110
 111 void brw_init_compile(struct brw_context *, struct brw_compile *p,
 112                       void *mem_ctx);
 113 void brw_dump_compile(struct brw_compile *p, FILE *out, int start, int end);
 114 const GLuint *brw_get_program( struct brw_compile *p, GLuint *sz );
 115
 116 struct brw_instruction *brw_next_insn(struct brw_compile *p, GLuint opcode);
 117 void brw_set_dest(struct brw_compile *p, struct brw_instruction *insn,
 118                   struct brw_reg dest);
 119 void brw_set_src0(struct brw_compile *p, struct brw_instruction *insn,
 120                   struct brw_reg reg);
 121
 122 void gen6_resolve_implied_move(struct brw_compile *p,
 123                                struct brw_reg *src,
 124                                GLuint msg_reg_nr);
 125
 126 /* Helpers for regular instructions:
 127  */
 128 #define ALU1(OP)                                        \
 129 struct brw_instruction *brw_##OP(struct brw_compile *p, \
 130               struct brw_reg dest,                      \
 131               struct brw_reg src0);
 132
 133 #define ALU2(OP)                                        \
 134 struct brw_instruction *brw_##OP(struct brw_compile *p, \
 135               struct brw_reg dest,                      \
 136               struct brw_reg src0,                      \
 137               struct brw_reg src1);
 138
 139 #define ALU3(OP)                                        \
 140 struct brw_instruction *brw_##OP(struct brw_compile *p, \
 141               struct brw_reg dest,                      \
 142               struct brw_reg src0,                      \
 143               struct brw_reg src1,                      \
 144               struct brw_reg src2);
 145
 146 #define ROUND(OP) \
 147 void brw_##OP(struct brw_compile *p, struct brw_reg dest, struct brw_reg src0);
 148
 149 ALU1(MOV)
 150 ALU2(SEL)
 151 ALU1(NOT)
 152 ALU2(AND)
 153 ALU2(OR)
 154 ALU2(XOR)
 155 ALU2(SHR)
 156 ALU2(SHL)
 157 ALU2(ASR)
 158 ALU1(F32TO16)
 159 ALU1(F16TO32)
 160 ALU2(JMPI)
 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
 185 ROUND(RNDZ)
 186 ROUND(RNDE)
 187
 188 #undef ALU1
 189 #undef ALU2
 190 #undef ALU3
 191 #undef ROUND
 192
 193
 194 /* Helpers for SEND instruction:
 195  */
 196 void brw_set_sampler_message(struct brw_compile *p,
 197                              struct brw_instruction *insn,
 198                              GLuint binding_table_index,
 199                              GLuint sampler,
 200                              GLuint msg_type,
 201                              GLuint response_length,
 202                              GLuint msg_length,
 203                              GLuint header_present,
 204                              GLuint simd_mode,
 205                              GLuint return_format);
 206
 207 void brw_set_dp_read_message(struct brw_compile *p,
 208                              struct brw_instruction *insn,
 209                              GLuint binding_table_index,
 210                              GLuint msg_control,
 211                              GLuint msg_type,
 212                              GLuint target_cache,
 213                              GLuint msg_length,
 214                              bool header_present,
 215                              GLuint response_length);
 216
 217 void brw_set_dp_write_message(struct brw_compile *p,
 218                               struct brw_instruction *insn,
 219                               GLuint binding_table_index,
 220                               GLuint msg_control,
 221                               GLuint msg_type,
 222                               GLuint msg_length,
 223                               bool header_present,
 224                               GLuint last_render_target,
 225                               GLuint response_length,
 226                               GLuint end_of_thread,
 227                               GLuint send_commit_msg);
 228
 229 enum brw_urb_write_flags {
 230    BRW_URB_WRITE_NO_FLAGS = 0,
 231
 232    /**
 233     * Causes a new URB entry to be allocated, and its address stored in the
 234     * destination register (gen < 7).
 235     */
 236    BRW_URB_WRITE_ALLOCATE = 0x1,
 237
 238    /**
 239     * Causes the current URB entry to be deallocated (gen < 7).
 240     */
 241    BRW_URB_WRITE_UNUSED = 0x2,
 242
 243    /**
 244     * Causes the thread to terminate.
 245     */
 246    BRW_URB_WRITE_EOT = 0x4,
 247
 248    /**
 249     * Indicates that the given URB entry is complete, and may be sent further
 250     * down the 3D pipeline (gen < 7).
 251     */
 252    BRW_URB_WRITE_COMPLETE = 0x8,
 253
 254    /**
 255     * Indicates that an additional offset (which may be different for the two
 256     * vec4 slots) is stored in the message header (gen == 7).
 257     */
 258    BRW_URB_WRITE_PER_SLOT_OFFSET = 0x10,
 259
 260    /**
 261     * Convenient combination of flags: end the thread while simultaneously
 262     * marking the given URB entry as complete.
 263     */
 264    BRW_URB_WRITE_EOT_COMPLETE = BRW_URB_WRITE_EOT | BRW_URB_WRITE_COMPLETE,
 265
 266    /**
 267     * Convenient combination of flags: mark the given URB entry as complete
 268     * and simultaneously allocate a new one.
 269     */
 270    BRW_URB_WRITE_ALLOCATE_COMPLETE =
 271       BRW_URB_WRITE_ALLOCATE | BRW_URB_WRITE_COMPLETE,
 272 };
 273
 274 #ifdef __cplusplus
 275 /**
 276  * Allow brw_urb_write_flags enums to be ORed together.
 277  */
 278 inline brw_urb_write_flags
 279 operator|(brw_urb_write_flags x, brw_urb_write_flags y)
 280 {
 281    return static_cast<brw_urb_write_flags>(static_cast<int>(x) |
 282                                            static_cast<int>(y));
 283 }
 284 #endif
 285
 286 void brw_urb_WRITE(struct brw_compile *p,
 287                    struct brw_reg dest,
 288                    GLuint msg_reg_nr,
 289                    struct brw_reg src0,
 290                    enum brw_urb_write_flags flags,
 291                    GLuint msg_length,
 292                    GLuint response_length,
 293                    GLuint offset,
 294                    GLuint swizzle);
 295
 296 void brw_ff_sync(struct brw_compile *p,
 297                    struct brw_reg dest,
 298                    GLuint msg_reg_nr,
 299                    struct brw_reg src0,
 300                    bool allocate,
 301                    GLuint response_length,
 302                    bool eot);
 303
 304 void brw_svb_write(struct brw_compile *p,
 305                    struct brw_reg dest,
 306                    GLuint msg_reg_nr,
 307                    struct brw_reg src0,
 308                    GLuint binding_table_index,
 309                    bool   send_commit_msg);
 310
 311 void brw_fb_WRITE(struct brw_compile *p,
 312                   int dispatch_width,
 313                    GLuint msg_reg_nr,
 314                    struct brw_reg src0,
 315                    GLuint msg_control,
 316                    GLuint binding_table_index,
 317                    GLuint msg_length,
 318                    GLuint response_length,
 319                    bool eot,
 320                    bool header_present);
 321
 322 void brw_SAMPLE(struct brw_compile *p,
 323                 struct brw_reg dest,
 324                 GLuint msg_reg_nr,
 325                 struct brw_reg src0,
 326                 GLuint binding_table_index,
 327                 GLuint sampler,
 328                 GLuint msg_type,
 329                 GLuint response_length,
 330                 GLuint msg_length,
 331                 GLuint header_present,
 332                 GLuint simd_mode,
 333                 GLuint return_format);
 334
 335 void brw_math( struct brw_compile *p,
 336                struct brw_reg dest,
 337                GLuint function,
 338                GLuint msg_reg_nr,
 339                struct brw_reg src,
 340                GLuint data_type,
 341                GLuint precision );
 342
 343 void brw_math2(struct brw_compile *p,
 344                struct brw_reg dest,
 345                GLuint function,
 346                struct brw_reg src0,
 347                struct brw_reg src1);
 348
 349 void brw_oword_block_read(struct brw_compile *p,
 350                           struct brw_reg dest,
 351                           struct brw_reg mrf,
 352                           uint32_t offset,
 353                           uint32_t bind_table_index);
 354
 355 void brw_oword_block_read_scratch(struct brw_compile *p,
 356                                   struct brw_reg dest,
 357                                   struct brw_reg mrf,
 358                                   int num_regs,
 359                                   GLuint offset);
 360
 361 void brw_oword_block_write_scratch(struct brw_compile *p,
 362                                    struct brw_reg mrf,
 363                                    int num_regs,
 364                                    GLuint offset);
 365
 366 void brw_shader_time_add(struct brw_compile *p,
 367                          struct brw_reg payload,
 368                          uint32_t surf_index);
 369
 370 /* If/else/endif.  Works by manipulating the execution flags on each
 371  * channel.
 372  */
 373 struct brw_instruction *brw_IF(struct brw_compile *p,
 374                                GLuint execute_size);
 375 struct brw_instruction *gen6_IF(struct brw_compile *p, uint32_t conditional,
 376                                 struct brw_reg src0, struct brw_reg src1);
 377
 378 void brw_ELSE(struct brw_compile *p);
 379 void brw_ENDIF(struct brw_compile *p);
 380
 381 /* DO/WHILE loops:
 382  */
 383 struct brw_instruction *brw_DO(struct brw_compile *p,
 384                                GLuint execute_size);
 385
 386 struct brw_instruction *brw_WHILE(struct brw_compile *p);
 387
 388 struct brw_instruction *brw_BREAK(struct brw_compile *p);
 389 struct brw_instruction *brw_CONT(struct brw_compile *p);
 390 struct brw_instruction *gen6_CONT(struct brw_compile *p);
 391 struct brw_instruction *gen6_HALT(struct brw_compile *p);
 392 /* Forward jumps:
 393  */
 394 void brw_land_fwd_jump(struct brw_compile *p, int jmp_insn_idx);
 395
 396
 397
 398 void brw_NOP(struct brw_compile *p);
 399
 400 void brw_WAIT(struct brw_compile *p);
 401
 402 /* Special case: there is never a destination, execution size will be
 403  * taken from src0:
 404  */
 405 void brw_CMP(struct brw_compile *p,
 406              struct brw_reg dest,
 407              GLuint conditional,
 408              struct brw_reg src0,
 409              struct brw_reg src1);
 410
 411 /***********************************************************************
 412  * brw_eu_util.c:
 413  */
 414
 415 void brw_copy_indirect_to_indirect(struct brw_compile *p,
 416                                    struct brw_indirect dst_ptr,
 417                                    struct brw_indirect src_ptr,
 418                                    GLuint count);
 419
 420 void brw_copy_from_indirect(struct brw_compile *p,
 421                             struct brw_reg dst,
 422                             struct brw_indirect ptr,
 423                             GLuint count);
 424
 425 void brw_copy4(struct brw_compile *p,
 426                struct brw_reg dst,
 427                struct brw_reg src,
 428                GLuint count);
 429
 430 void brw_copy8(struct brw_compile *p,
 431                struct brw_reg dst,
 432                struct brw_reg src,
 433                GLuint count);
 434
 435 void brw_math_invert( struct brw_compile *p,
 436                       struct brw_reg dst,
 437                       struct brw_reg src);
 438
 439 void brw_set_src1(struct brw_compile *p,
 440                   struct brw_instruction *insn,
 441                   struct brw_reg reg);
 442
 443 void brw_set_uip_jip(struct brw_compile *p);
 444
 445 uint32_t brw_swap_cmod(uint32_t cmod);
 446
 447 /* brw_eu_compact.c */
 448 void brw_init_compaction_tables(struct brw_context *brw);
 449 void brw_compact_instructions(struct brw_compile *p);
 450 void brw_uncompact_instruction(struct brw_context *brw,
 451                                struct brw_instruction *dst,
 452                                struct brw_compact_instruction *src);
 453 bool brw_try_compact_instruction(struct brw_compile *p,
 454                                  struct brw_compact_instruction *dst,
 455                                  struct brw_instruction *src);
 456
 457 void brw_debug_compact_uncompact(struct brw_context *brw,
 458                                  struct brw_instruction *orig,
 459                                  struct brw_instruction *uncompacted);
 460
 461 #ifdef __cplusplus
 462 }
 463 #endif
 464
 465 #endif