src/gallium/auxiliary/gallivm/lp_bld_tgsi.h

   1 /**************************************************************************
   2  *
   3  * Copyright 2011-2012 Advanced Micro Devices, Inc.
   4  * Copyright 2009 VMware, Inc.
   5  * All Rights Reserved.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a
   8  * copy of this software and associated documentation files (the
   9  * "Software"), to deal in the Software without restriction, including
  10  * without limitation the rights to use, copy, modify, merge, publish,
  11  * distribute, sub license, and/or sell copies of the Software, and to
  12  * permit persons to whom the Software is furnished to do so, subject to
  13  * the following conditions:
  14  *
  15  * The above copyright notice and this permission notice (including the
  16  * next paragraph) shall be included in all copies or substantial portions
  17  * of the Software.
  18  *
  19  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  20  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  21  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  22  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  23  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  24  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  25  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  26  *
  27  **************************************************************************/
  28
  29 /**
  30  * @file
  31  * TGSI to LLVM IR translation.
  32  *
  33  * @author Jose Fonseca <jfonseca@vmware.com>
  34  * @author Tom Stellard <thomas.stellard@amd.com>
  35  */
  36
  37 #ifndef LP_BLD_TGSI_H
  38 #define LP_BLD_TGSI_H
  39
  40 #include "gallivm/lp_bld.h"
  41 #include "gallivm/lp_bld_tgsi_action.h"
  42 #include "gallivm/lp_bld_limits.h"
  43 #include "lp_bld_type.h"
  44 #include "pipe/p_compiler.h"
  45 #include "pipe/p_state.h"
  46 #include "tgsi/tgsi_exec.h"
  47 #include "tgsi/tgsi_scan.h"
  48 #include "tgsi/tgsi_info.h"
  49
  50 #define LP_CHAN_ALL ~0
  51
  52 #define LP_MAX_INSTRUCTIONS 256
  53
  54 struct tgsi_full_declaration;
  55 struct tgsi_full_immediate;
  56 struct tgsi_full_instruction;
  57 struct tgsi_full_src_register;
  58 struct tgsi_opcode_info;
  59 struct tgsi_token;
  60 struct tgsi_shader_info;
  61 struct lp_build_mask_context;
  62 struct gallivm_state;
  63 struct lp_derivatives;
  64 struct lp_build_tgsi_gs_iface;
  65
  66
  67 enum lp_build_tex_modifier {
  68    LP_BLD_TEX_MODIFIER_NONE = 0,
  69    LP_BLD_TEX_MODIFIER_PROJECTED,
  70    LP_BLD_TEX_MODIFIER_LOD_BIAS,
  71    LP_BLD_TEX_MODIFIER_EXPLICIT_LOD,
  72    LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV,
  73    LP_BLD_TEX_MODIFIER_LOD_ZERO
  74 };
  75
  76
  77 /**
  78  * Describe a channel of a register.
  79  *
  80  * The value can be a:
  81  * - immediate value (i.e. derived from a IMM register)
  82  * - CONST[n].x/y/z/w
  83  * - IN[n].x/y/z/w
  84  * - undetermined (when .file == TGSI_FILE_NULL)
  85  *
  86  * This is one of the analysis results, and is used to described
  87  * the output color in terms of inputs.
  88  */
  89 struct lp_tgsi_channel_info
  90 {
  91    unsigned file:4; /* TGSI_FILE_* */
  92    unsigned swizzle:3; /* PIPE_SWIZZLE_x */
  93    union {
  94       uint32_t index;
  95       float value; /* for TGSI_FILE_IMMEDIATE */
  96    } u;
  97 };
  98
  99
 100 /**
 101  * Describe a texture sampler interpolator.
 102  *
 103  * The interpolation is described in terms of regular inputs.
 104  */
 105 struct lp_tgsi_texture_info
 106 {
 107    struct lp_tgsi_channel_info coord[4];
 108    unsigned target:8; /* TGSI_TEXTURE_* */
 109    unsigned sampler_unit:8;  /* Sampler unit */
 110    unsigned texture_unit:8;  /* Texture unit */
 111    unsigned modifier:8; /* LP_BLD_TEX_MODIFIER_* */
 112 };
 113
 114
 115 struct lp_tgsi_info
 116 {
 117    struct tgsi_shader_info base;
 118
 119    /*
 120     * Whether any of the texture opcodes access a register file other than
 121     * TGSI_FILE_INPUT.
 122     *
 123     * We could also handle TGSI_FILE_CONST/IMMEDIATE here, but there is little
 124     * benefit.
 125     */
 126    unsigned indirect_textures:1;
 127
 128    /*
 129     * Whether any immediate values are outside the range of 0 and 1
 130     */
 131    unsigned unclamped_immediates:1;
 132
 133    /*
 134     * Texture opcode description. Aimed at detecting and described direct
 135     * texture opcodes.
 136     */
 137    unsigned num_texs;
 138    struct lp_tgsi_texture_info tex[PIPE_MAX_SAMPLERS];
 139
 140    /*
 141     * Output description. Aimed at detecting and describing simple blit
 142     * shaders.
 143     */
 144    struct lp_tgsi_channel_info output[PIPE_MAX_SHADER_OUTPUTS][4];
 145
 146    /*
 147     * Shortcut pointers into the above (for fragment shaders).
 148     */
 149    const struct lp_tgsi_channel_info *cbuf[PIPE_MAX_COLOR_BUFS];
 150 };
 151
 152 /**
 153  * Reference to system values.
 154  */
 155 struct lp_bld_tgsi_system_values {
 156    LLVMValueRef instance_id;
 157    LLVMValueRef vertex_id;
 158 };
 159
 160
 161 /**
 162  * Sampler code generation interface.
 163  *
 164  * Although texture sampling is a requirement for TGSI translation, it is
 165  * a very different problem with several different approaches to it. This
 166  * structure establishes an interface for texture sampling code generation, so
 167  * that we can easily use different texture sampling strategies.
 168  */
 169 struct lp_build_sampler_soa
 170 {
 171    void
 172    (*destroy)( struct lp_build_sampler_soa *sampler );
 173
 174    void
 175    (*emit_fetch_texel)( const struct lp_build_sampler_soa *sampler,
 176                         struct gallivm_state *gallivm,
 177                         struct lp_type type,
 178                         boolean is_fetch,
 179                         unsigned texture_index,
 180                         unsigned sampler_index,
 181                         const LLVMValueRef *coords,
 182                         const LLVMValueRef *offsets,
 183                         const struct lp_derivatives *derivs,
 184                         LLVMValueRef lod_bias, /* optional */
 185                         LLVMValueRef explicit_lod, /* optional */
 186                         LLVMValueRef *texel);
 187
 188    void
 189    (*emit_size_query)( const struct lp_build_sampler_soa *sampler,
 190                        struct gallivm_state *gallivm,
 191                        struct lp_type type,
 192                        unsigned unit,
 193                        boolean need_nr_mips,
 194                        LLVMValueRef explicit_lod, /* optional */
 195                        LLVMValueRef *sizes_out);
 196 };
 197
 198
 199 struct lp_build_sampler_aos
 200 {
 201    LLVMValueRef
 202    (*emit_fetch_texel)( struct lp_build_sampler_aos *sampler,
 203                         struct lp_build_context *bld,
 204                         unsigned target, /* TGSI_TEXTURE_* */
 205                         unsigned unit,
 206                         LLVMValueRef coords,
 207                         const struct lp_derivatives derivs,
 208                         enum lp_build_tex_modifier modifier);
 209 };
 210
 211
 212 void
 213 lp_build_tgsi_info(const struct tgsi_token *tokens,
 214                    struct lp_tgsi_info *info);
 215
 216
 217 void
 218 lp_build_tgsi_soa(struct gallivm_state *gallivm,
 219                   const struct tgsi_token *tokens,
 220                   struct lp_type type,
 221                   struct lp_build_mask_context *mask,
 222                   LLVMValueRef consts_ptr,
 223                   const struct lp_bld_tgsi_system_values *system_values,
 224                   const LLVMValueRef *pos,
 225                   const LLVMValueRef (*inputs)[4],
 226                   LLVMValueRef (*outputs)[4],
 227                   struct lp_build_sampler_soa *sampler,
 228                   const struct tgsi_shader_info *info,
 229                   const struct lp_build_tgsi_gs_iface *gs_iface);
 230
 231
 232 void
 233 lp_build_tgsi_aos(struct gallivm_state *gallivm,
 234                   const struct tgsi_token *tokens,
 235                   struct lp_type type,
 236                   const unsigned char swizzles[4],
 237                   LLVMValueRef consts_ptr,
 238                   const LLVMValueRef *inputs,
 239                   LLVMValueRef *outputs,
 240                   struct lp_build_sampler_aos *sampler,
 241                   const struct tgsi_shader_info *info);
 242
 243
 244 struct lp_exec_mask {
 245    struct lp_build_context *bld;
 246
 247    boolean has_mask;
 248    boolean ret_in_main;
 249
 250    LLVMTypeRef int_vec_type;
 251
 252    LLVMValueRef cond_stack[LP_MAX_TGSI_NESTING];
 253    int cond_stack_size;
 254    LLVMValueRef cond_mask;
 255
 256    LLVMBasicBlockRef loop_block;
 257    LLVMValueRef cont_mask;
 258    LLVMValueRef break_mask;
 259    LLVMValueRef break_var;
 260    struct {
 261       LLVMBasicBlockRef loop_block;
 262       LLVMValueRef cont_mask;
 263       LLVMValueRef break_mask;
 264       LLVMValueRef break_var;
 265    } loop_stack[LP_MAX_TGSI_NESTING];
 266    int loop_stack_size;
 267
 268    LLVMValueRef ret_mask;
 269    struct {
 270       int pc;
 271       LLVMValueRef ret_mask;
 272    } call_stack[LP_MAX_TGSI_NESTING];
 273    int call_stack_size;
 274
 275    LLVMValueRef exec_mask;
 276    LLVMValueRef loop_limiter;
 277 };
 278
 279 struct lp_build_tgsi_inst_list
 280 {
 281    struct tgsi_full_instruction *instructions;
 282    uint max_instructions;
 283    uint num_instructions;
 284 };
 285
 286 unsigned lp_bld_tgsi_list_init(struct lp_build_tgsi_context * bld_base);
 287
 288
 289 unsigned lp_bld_tgsi_add_instruction(
 290    struct lp_build_tgsi_context * bld_base,
 291    struct tgsi_full_instruction *inst_to_add);
 292
 293
 294 struct lp_build_tgsi_context;
 295
 296
 297 typedef LLVMValueRef (*lp_build_emit_fetch_fn)(struct lp_build_tgsi_context *,
 298                                         const struct tgsi_full_src_register *,
 299                                         enum tgsi_opcode_type,
 300                                         unsigned);
 301
 302 struct lp_build_tgsi_context
 303 {
 304    struct lp_build_context base;
 305
 306    struct lp_build_context uint_bld;
 307    struct lp_build_context int_bld;
 308
 309    /** This array stores functions that are used to transform TGSI opcodes to
 310      * LLVM instructions.
 311      */
 312    struct lp_build_tgsi_action op_actions[TGSI_OPCODE_LAST];
 313
 314    /* TGSI_OPCODE_RSQ is defined as 1 / sqrt( abs(src0.x) ), rsq_action
 315     * should compute 1 / sqrt (src0.x) */
 316    struct lp_build_tgsi_action rsq_action;
 317
 318    struct lp_build_tgsi_action sqrt_action;
 319
 320    const struct tgsi_shader_info *info;
 321
 322    lp_build_emit_fetch_fn emit_fetch_funcs[TGSI_FILE_COUNT];
 323
 324    LLVMValueRef (*emit_swizzle)(struct lp_build_tgsi_context *,
 325                          LLVMValueRef, unsigned, unsigned, unsigned, unsigned);
 326
 327    void (*emit_store)(struct lp_build_tgsi_context *,
 328                       const struct tgsi_full_instruction *,
 329                       const struct tgsi_opcode_info *,
 330                       LLVMValueRef dst[4]);
 331
 332    void (*emit_declaration)(struct lp_build_tgsi_context *,
 333                              const struct tgsi_full_declaration *decl);
 334
 335    void (*emit_immediate)(struct lp_build_tgsi_context *,
 336                           const struct tgsi_full_immediate *imm);
 337
 338
 339    /* Allow the user to store data in this structure rather than passing it
 340     * to every function. */
 341    void * userdata;
 342
 343    boolean soa;
 344
 345    int pc;
 346
 347    struct tgsi_full_instruction *instructions;
 348    uint max_instructions;
 349    uint num_instructions;
 350
 351    /** This function allows the user to insert some instructions at the
 352      * beginning of the program.  It is optional and does not need to be
 353      * implemented.
 354      */
 355    void (*emit_prologue)(struct lp_build_tgsi_context*);
 356
 357    /** This function allows the user to insert some instructions at the end of
 358      * the program.  This callback is intended to be used for emitting
 359      * instructions to handle the export for the output registers, but it can
 360      * be used for any purpose.  Implementing this function is optiona, but
 361      * recommended.
 362      */
 363    void (*emit_epilogue)(struct lp_build_tgsi_context*);
 364 };
 365
 366 struct lp_build_tgsi_gs_iface
 367 {
 368    LLVMValueRef input;
 369    void (*emit_vertex)(struct lp_build_tgsi_context * bld_base,
 370                        LLVMValueRef (*outputs)[4],
 371                        LLVMValueRef emitted_vertices_vec,
 372                        void *user_data);
 373    void (*end_primitive)(struct lp_build_tgsi_context * bld_base,
 374                          LLVMValueRef verts_per_prim_vec,
 375                          LLVMValueRef emitted_prims_vec,
 376                          void *user_data);
 377    void (*gs_epilogue)(struct lp_build_tgsi_context * bld_base,
 378                        LLVMValueRef total_emitted_vertices_vec,
 379                        LLVMValueRef emitted_prims_vec,
 380                        void *user_data);
 381    void *user_data;
 382 };
 383
 384 struct lp_build_tgsi_soa_context
 385 {
 386    struct lp_build_tgsi_context bld_base;
 387
 388    /* Builder for scalar elements of shader's data type (float) */
 389    struct lp_build_context elem_bld;
 390
 391    const struct lp_build_tgsi_gs_iface *gs_iface;
 392    LLVMValueRef emitted_prims_vec;
 393    LLVMValueRef total_emitted_vertices_vec;
 394    LLVMValueRef emitted_vertices_vec;
 395    /* if a shader doesn't have ENDPRIM instruction but it has
 396     * a number of EMIT instructions it means the END instruction
 397     * implicitly invokes ENDPRIM. handle this via a flag here
 398     * in the future maybe we can enforce TGSI to always have
 399     * an explicit ENDPRIM */
 400    boolean pending_end_primitive;
 401
 402    LLVMValueRef consts_ptr;
 403    const LLVMValueRef *pos;
 404    const LLVMValueRef (*inputs)[TGSI_NUM_CHANNELS];
 405    LLVMValueRef (*outputs)[TGSI_NUM_CHANNELS];
 406
 407    const struct lp_build_sampler_soa *sampler;
 408
 409    struct tgsi_declaration_sampler_view sv[PIPE_MAX_SHADER_SAMPLER_VIEWS];
 410
 411    LLVMValueRef immediates[LP_MAX_TGSI_IMMEDIATES][TGSI_NUM_CHANNELS];
 412    LLVMValueRef temps[LP_MAX_TGSI_TEMPS][TGSI_NUM_CHANNELS];
 413    LLVMValueRef addr[LP_MAX_TGSI_ADDRS][TGSI_NUM_CHANNELS];
 414    LLVMValueRef preds[LP_MAX_TGSI_PREDS][TGSI_NUM_CHANNELS];
 415
 416    /* We allocate/use this array of temps if (1 << TGSI_FILE_TEMPORARY) is
 417     * set in the indirect_files field.
 418     * The temps[] array above is unused then.
 419     */
 420    LLVMValueRef temps_array;
 421
 422    /* We allocate/use this array of output if (1 << TGSI_FILE_OUTPUT) is
 423     * set in the indirect_files field.
 424     * The outputs[] array above is unused then.
 425     */
 426    LLVMValueRef outputs_array;
 427
 428    /* We allocate/use this array of inputs if (1 << TGSI_FILE_INPUT) is
 429     * set in the indirect_files field.
 430     * The inputs[] array above is unused then.
 431     */
 432    LLVMValueRef inputs_array;
 433
 434    struct lp_bld_tgsi_system_values system_values;
 435
 436    /** bitmask indicating which register files are accessed indirectly */
 437    unsigned indirect_files;
 438
 439    struct lp_build_mask_context *mask;
 440    struct lp_exec_mask exec_mask;
 441
 442    uint num_immediates;
 443
 444 };
 445
 446 void
 447 lp_emit_declaration_soa(
 448    struct lp_build_tgsi_context *bld,
 449    const struct tgsi_full_declaration *decl);
 450
 451 void lp_emit_immediate_soa(
 452    struct lp_build_tgsi_context *bld_base,
 453    const struct tgsi_full_immediate *imm);
 454
 455 boolean
 456 lp_emit_instruction_soa(
 457    struct lp_build_tgsi_soa_context *bld,
 458    const struct tgsi_full_instruction *inst,
 459    const struct tgsi_opcode_info *info);
 460
 461
 462 LLVMValueRef
 463 lp_get_temp_ptr_soa(
 464    struct lp_build_tgsi_soa_context *bld,
 465    unsigned index,
 466    unsigned chan);
 467
 468 LLVMValueRef
 469 lp_get_output_ptr(
 470    struct lp_build_tgsi_soa_context *bld,
 471    unsigned index,
 472    unsigned chan);
 473
 474 struct lp_build_tgsi_aos_context
 475 {
 476    struct lp_build_tgsi_context bld_base;
 477
 478    /* Builder for integer masks and indices */
 479    struct lp_build_context int_bld;
 480
 481    /*
 482     * AoS swizzle used:
 483     * - swizzles[0] = red index
 484     * - swizzles[1] = green index
 485     * - swizzles[2] = blue index
 486     * - swizzles[3] = alpha index
 487     */
 488    unsigned char swizzles[4];
 489    unsigned char inv_swizzles[4];
 490
 491    LLVMValueRef consts_ptr;
 492    const LLVMValueRef *inputs;
 493    LLVMValueRef *outputs;
 494
 495    struct lp_build_sampler_aos *sampler;
 496
 497    LLVMValueRef immediates[LP_MAX_TGSI_IMMEDIATES];
 498    LLVMValueRef temps[LP_MAX_TGSI_TEMPS];
 499    LLVMValueRef addr[LP_MAX_TGSI_ADDRS];
 500    LLVMValueRef preds[LP_MAX_TGSI_PREDS];
 501
 502    /* We allocate/use this array of temps if (1 << TGSI_FILE_TEMPORARY) is
 503     * set in the indirect_files field.
 504     * The temps[] array above is unused then.
 505     */
 506    LLVMValueRef temps_array;
 507
 508    /** bitmask indicating which register files are accessed indirectly */
 509    unsigned indirect_files;
 510
 511 };
 512
 513 static INLINE struct lp_build_tgsi_soa_context *
 514 lp_soa_context(struct lp_build_tgsi_context *bld_base)
 515 {
 516    return (struct lp_build_tgsi_soa_context *)bld_base;
 517 }
 518
 519 static INLINE struct lp_build_tgsi_aos_context *
 520 lp_aos_context(struct lp_build_tgsi_context *bld_base)
 521 {
 522    return (struct lp_build_tgsi_aos_context *)bld_base;
 523 }
 524
 525 void
 526 lp_emit_declaration_aos(
 527    struct lp_build_tgsi_aos_context *bld,
 528    const struct tgsi_full_declaration *decl);
 529
 530
 531 boolean
 532 lp_emit_instruction_aos(
 533    struct lp_build_tgsi_aos_context *bld,
 534    const struct tgsi_full_instruction *inst,
 535    const struct tgsi_opcode_info *info,
 536    int *pc);
 537
 538 void
 539 lp_emit_store_aos(
 540    struct lp_build_tgsi_aos_context *bld,
 541    const struct tgsi_full_instruction *inst,
 542    unsigned index,
 543    LLVMValueRef value);
 544
 545 void lp_build_fetch_args(
 546    struct lp_build_tgsi_context * bld_base,
 547    struct lp_build_emit_data * emit_data);
 548
 549 LLVMValueRef
 550 lp_build_tgsi_inst_llvm_aos(
 551    struct lp_build_tgsi_context * bld_base,
 552    const struct tgsi_full_instruction *inst);
 553
 554 void
 555 lp_build_tgsi_intrinsic(
 556  const struct lp_build_tgsi_action * action,
 557  struct lp_build_tgsi_context * bld_base,
 558  struct lp_build_emit_data * emit_data);
 559
 560 LLVMValueRef
 561 lp_build_emit_llvm(
 562    struct lp_build_tgsi_context *bld_base,
 563    unsigned tgsi_opcode,
 564    struct lp_build_emit_data * emit_data);
 565
 566 LLVMValueRef
 567 lp_build_emit_llvm_unary(
 568    struct lp_build_tgsi_context *bld_base,
 569    unsigned tgsi_opcode,
 570    LLVMValueRef arg0);
 571
 572 LLVMValueRef
 573 lp_build_emit_llvm_binary(
 574    struct lp_build_tgsi_context *bld_base,
 575    unsigned tgsi_opcode,
 576    LLVMValueRef arg0,
 577    LLVMValueRef arg1);
 578
 579 LLVMValueRef
 580 lp_build_emit_llvm_ternary(
 581    struct lp_build_tgsi_context *bld_base,
 582    unsigned tgsi_opcode,
 583    LLVMValueRef arg0,
 584    LLVMValueRef arg1,
 585    LLVMValueRef arg2);
 586
 587 boolean
 588 lp_build_tgsi_inst_llvm(
 589    struct lp_build_tgsi_context * bld_base,
 590    const struct tgsi_full_instruction *inst);
 591
 592 LLVMValueRef
 593 lp_build_emit_fetch(
 594    struct lp_build_tgsi_context *bld_base,
 595    const struct tgsi_full_instruction *inst,
 596    unsigned src_op,
 597    const unsigned chan_index);
 598
 599
 600 LLVMValueRef
 601 lp_build_emit_fetch_texoffset(
 602    struct lp_build_tgsi_context *bld_base,
 603    const struct tgsi_full_instruction *inst,
 604    unsigned tex_off_op,
 605    const unsigned chan_index);
 606
 607 boolean
 608 lp_build_tgsi_llvm(
 609    struct lp_build_tgsi_context * bld_base,
 610    const struct tgsi_token *tokens);
 611
 612 #endif /* LP_BLD_TGSI_H */