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    LLVMValueRef prim_id;
 159 };
 160
 161
 162 /**
 163  * Sampler code generation interface.
 164  *
 165  * Although texture sampling is a requirement for TGSI translation, it is
 166  * a very different problem with several different approaches to it. This
 167  * structure establishes an interface for texture sampling code generation, so
 168  * that we can easily use different texture sampling strategies.
 169  */
 170 struct lp_build_sampler_soa
 171 {
 172    void
 173    (*destroy)( struct lp_build_sampler_soa *sampler );
 174
 175    void
 176    (*emit_fetch_texel)( const struct lp_build_sampler_soa *sampler,
 177                         struct gallivm_state *gallivm,
 178                         struct lp_type type,
 179                         boolean is_fetch,
 180                         unsigned texture_index,
 181                         unsigned sampler_index,
 182                         const LLVMValueRef *coords,
 183                         const LLVMValueRef *offsets,
 184                         const struct lp_derivatives *derivs,
 185                         LLVMValueRef lod_bias, /* optional */
 186                         LLVMValueRef explicit_lod, /* optional */
 187                         LLVMValueRef *texel);
 188
 189    void
 190    (*emit_size_query)( const struct lp_build_sampler_soa *sampler,
 191                        struct gallivm_state *gallivm,
 192                        struct lp_type type,
 193                        unsigned unit,
 194                        boolean need_nr_mips,
 195                        LLVMValueRef explicit_lod, /* optional */
 196                        LLVMValueRef *sizes_out);
 197 };
 198
 199
 200 struct lp_build_sampler_aos
 201 {
 202    LLVMValueRef
 203    (*emit_fetch_texel)( struct lp_build_sampler_aos *sampler,
 204                         struct lp_build_context *bld,
 205                         unsigned target, /* TGSI_TEXTURE_* */
 206                         unsigned unit,
 207                         LLVMValueRef coords,
 208                         const struct lp_derivatives derivs,
 209                         enum lp_build_tex_modifier modifier);
 210 };
 211
 212
 213 void
 214 lp_build_tgsi_info(const struct tgsi_token *tokens,
 215                    struct lp_tgsi_info *info);
 216
 217
 218 void
 219 lp_build_tgsi_soa(struct gallivm_state *gallivm,
 220                   const struct tgsi_token *tokens,
 221                   struct lp_type type,
 222                   struct lp_build_mask_context *mask,
 223                   LLVMValueRef consts_ptr,
 224                   const struct lp_bld_tgsi_system_values *system_values,
 225                   const LLVMValueRef *pos,
 226                   const LLVMValueRef (*inputs)[4],
 227                   LLVMValueRef (*outputs)[4],
 228                   struct lp_build_sampler_soa *sampler,
 229                   const struct tgsi_shader_info *info,
 230                   const struct lp_build_tgsi_gs_iface *gs_iface);
 231
 232
 233 void
 234 lp_build_tgsi_aos(struct gallivm_state *gallivm,
 235                   const struct tgsi_token *tokens,
 236                   struct lp_type type,
 237                   const unsigned char swizzles[4],
 238                   LLVMValueRef consts_ptr,
 239                   const LLVMValueRef *inputs,
 240                   LLVMValueRef *outputs,
 241                   struct lp_build_sampler_aos *sampler,
 242                   const struct tgsi_shader_info *info);
 243
 244
 245 struct lp_exec_mask {
 246    struct lp_build_context *bld;
 247
 248    boolean has_mask;
 249    boolean ret_in_main;
 250
 251    LLVMTypeRef int_vec_type;
 252
 253    LLVMValueRef cond_stack[LP_MAX_TGSI_NESTING];
 254    int cond_stack_size;
 255    LLVMValueRef cond_mask;
 256
 257    LLVMBasicBlockRef loop_block;
 258    LLVMValueRef cont_mask;
 259    LLVMValueRef break_mask;
 260    LLVMValueRef break_var;
 261    struct {
 262       LLVMBasicBlockRef loop_block;
 263       LLVMValueRef cont_mask;
 264       LLVMValueRef break_mask;
 265       LLVMValueRef break_var;
 266    } loop_stack[LP_MAX_TGSI_NESTING];
 267    int loop_stack_size;
 268
 269    LLVMValueRef ret_mask;
 270    struct {
 271       int pc;
 272       LLVMValueRef ret_mask;
 273    } call_stack[LP_MAX_TGSI_NESTING];
 274    int call_stack_size;
 275
 276    LLVMValueRef exec_mask;
 277    LLVMValueRef loop_limiter;
 278 };
 279
 280 struct lp_build_tgsi_inst_list
 281 {
 282    struct tgsi_full_instruction *instructions;
 283    uint max_instructions;
 284    uint num_instructions;
 285 };
 286
 287 unsigned lp_bld_tgsi_list_init(struct lp_build_tgsi_context * bld_base);
 288
 289
 290 unsigned lp_bld_tgsi_add_instruction(
 291    struct lp_build_tgsi_context * bld_base,
 292    struct tgsi_full_instruction *inst_to_add);
 293
 294
 295 struct lp_build_tgsi_context;
 296
 297
 298 typedef LLVMValueRef (*lp_build_emit_fetch_fn)(struct lp_build_tgsi_context *,
 299                                         const struct tgsi_full_src_register *,
 300                                         enum tgsi_opcode_type,
 301                                         unsigned);
 302
 303 struct lp_build_tgsi_context
 304 {
 305    struct lp_build_context base;
 306
 307    struct lp_build_context uint_bld;
 308    struct lp_build_context int_bld;
 309
 310    /** This array stores functions that are used to transform TGSI opcodes to
 311      * LLVM instructions.
 312      */
 313    struct lp_build_tgsi_action op_actions[TGSI_OPCODE_LAST];
 314
 315    /* TGSI_OPCODE_RSQ is defined as 1 / sqrt( abs(src0.x) ), rsq_action
 316     * should compute 1 / sqrt (src0.x) */
 317    struct lp_build_tgsi_action rsq_action;
 318
 319    struct lp_build_tgsi_action sqrt_action;
 320
 321    const struct tgsi_shader_info *info;
 322
 323    lp_build_emit_fetch_fn emit_fetch_funcs[TGSI_FILE_COUNT];
 324
 325    LLVMValueRef (*emit_swizzle)(struct lp_build_tgsi_context *,
 326                          LLVMValueRef, unsigned, unsigned, unsigned, unsigned);
 327
 328    void (*emit_store)(struct lp_build_tgsi_context *,
 329                       const struct tgsi_full_instruction *,
 330                       const struct tgsi_opcode_info *,
 331                       LLVMValueRef dst[4]);
 332
 333    void (*emit_declaration)(struct lp_build_tgsi_context *,
 334                              const struct tgsi_full_declaration *decl);
 335
 336    void (*emit_immediate)(struct lp_build_tgsi_context *,
 337                           const struct tgsi_full_immediate *imm);
 338
 339
 340    /* Allow the user to store data in this structure rather than passing it
 341     * to every function. */
 342    void * userdata;
 343
 344    boolean soa;
 345
 346    int pc;
 347
 348    struct tgsi_full_instruction *instructions;
 349    uint max_instructions;
 350    uint num_instructions;
 351
 352    /** This function allows the user to insert some instructions at the
 353      * beginning of the program.  It is optional and does not need to be
 354      * implemented.
 355      */
 356    void (*emit_prologue)(struct lp_build_tgsi_context*);
 357
 358    /** This function allows the user to insert some instructions at the end of
 359      * the program.  This callback is intended to be used for emitting
 360      * instructions to handle the export for the output registers, but it can
 361      * be used for any purpose.  Implementing this function is optiona, but
 362      * recommended.
 363      */
 364    void (*emit_epilogue)(struct lp_build_tgsi_context*);
 365 };
 366
 367 struct lp_build_tgsi_gs_iface
 368 {
 369    LLVMValueRef (*fetch_input)(const struct lp_build_tgsi_gs_iface *gs_iface,
 370                                struct lp_build_tgsi_context * bld_base,
 371                                LLVMValueRef vertex_index,
 372                                LLVMValueRef attrib_index,
 373                                LLVMValueRef swizzle_index);
 374    void (*emit_vertex)(const struct lp_build_tgsi_gs_iface *gs_iface,
 375                        struct lp_build_tgsi_context * bld_base,
 376                        LLVMValueRef (*outputs)[4],
 377                        LLVMValueRef emitted_vertices_vec);
 378    void (*end_primitive)(const struct lp_build_tgsi_gs_iface *gs_iface,
 379                          struct lp_build_tgsi_context * bld_base,
 380                          LLVMValueRef verts_per_prim_vec,
 381                          LLVMValueRef emitted_prims_vec);
 382    void (*gs_epilogue)(const struct lp_build_tgsi_gs_iface *gs_iface,
 383                        struct lp_build_tgsi_context * bld_base,
 384                        LLVMValueRef total_emitted_vertices_vec,
 385                        LLVMValueRef emitted_prims_vec);
 386 };
 387
 388 struct lp_build_tgsi_soa_context
 389 {
 390    struct lp_build_tgsi_context bld_base;
 391
 392    /* Builder for scalar elements of shader's data type (float) */
 393    struct lp_build_context elem_bld;
 394
 395    const struct lp_build_tgsi_gs_iface *gs_iface;
 396    LLVMValueRef emitted_prims_vec_ptr;
 397    LLVMValueRef total_emitted_vertices_vec_ptr;
 398    LLVMValueRef emitted_vertices_vec_ptr;
 399    /* if a shader doesn't have ENDPRIM instruction but it has
 400     * a number of EMIT instructions it means the END instruction
 401     * implicitly invokes ENDPRIM. handle this via a flag here
 402     * in the future maybe we can enforce TGSI to always have
 403     * an explicit ENDPRIM */
 404    boolean pending_end_primitive;
 405
 406    LLVMValueRef consts_ptr;
 407    const LLVMValueRef *pos;
 408    const LLVMValueRef (*inputs)[TGSI_NUM_CHANNELS];
 409    LLVMValueRef (*outputs)[TGSI_NUM_CHANNELS];
 410
 411    const struct lp_build_sampler_soa *sampler;
 412
 413    struct tgsi_declaration_sampler_view sv[PIPE_MAX_SHADER_SAMPLER_VIEWS];
 414
 415    LLVMValueRef immediates[LP_MAX_TGSI_IMMEDIATES][TGSI_NUM_CHANNELS];
 416    LLVMValueRef temps[LP_MAX_TGSI_TEMPS][TGSI_NUM_CHANNELS];
 417    LLVMValueRef addr[LP_MAX_TGSI_ADDRS][TGSI_NUM_CHANNELS];
 418    LLVMValueRef preds[LP_MAX_TGSI_PREDS][TGSI_NUM_CHANNELS];
 419
 420    /* We allocate/use this array of temps if (1 << TGSI_FILE_TEMPORARY) is
 421     * set in the indirect_files field.
 422     * The temps[] array above is unused then.
 423     */
 424    LLVMValueRef temps_array;
 425
 426    /* We allocate/use this array of output if (1 << TGSI_FILE_OUTPUT) is
 427     * set in the indirect_files field.
 428     * The outputs[] array above is unused then.
 429     */
 430    LLVMValueRef outputs_array;
 431
 432    /* We allocate/use this array of inputs if (1 << TGSI_FILE_INPUT) is
 433     * set in the indirect_files field.
 434     * The inputs[] array above is unused then.
 435     */
 436    LLVMValueRef inputs_array;
 437
 438    struct lp_bld_tgsi_system_values system_values;
 439
 440    /** bitmask indicating which register files are accessed indirectly */
 441    unsigned indirect_files;
 442
 443    struct lp_build_mask_context *mask;
 444    struct lp_exec_mask exec_mask;
 445
 446    uint num_immediates;
 447
 448 };
 449
 450 void
 451 lp_emit_declaration_soa(
 452    struct lp_build_tgsi_context *bld,
 453    const struct tgsi_full_declaration *decl);
 454
 455 void lp_emit_immediate_soa(
 456    struct lp_build_tgsi_context *bld_base,
 457    const struct tgsi_full_immediate *imm);
 458
 459 boolean
 460 lp_emit_instruction_soa(
 461    struct lp_build_tgsi_soa_context *bld,
 462    const struct tgsi_full_instruction *inst,
 463    const struct tgsi_opcode_info *info);
 464
 465
 466 LLVMValueRef
 467 lp_get_temp_ptr_soa(
 468    struct lp_build_tgsi_soa_context *bld,
 469    unsigned index,
 470    unsigned chan);
 471
 472 LLVMValueRef
 473 lp_get_output_ptr(
 474    struct lp_build_tgsi_soa_context *bld,
 475    unsigned index,
 476    unsigned chan);
 477
 478 struct lp_build_tgsi_aos_context
 479 {
 480    struct lp_build_tgsi_context bld_base;
 481
 482    /* Builder for integer masks and indices */
 483    struct lp_build_context int_bld;
 484
 485    /*
 486     * AoS swizzle used:
 487     * - swizzles[0] = red index
 488     * - swizzles[1] = green index
 489     * - swizzles[2] = blue index
 490     * - swizzles[3] = alpha index
 491     */
 492    unsigned char swizzles[4];
 493    unsigned char inv_swizzles[4];
 494
 495    LLVMValueRef consts_ptr;
 496    const LLVMValueRef *inputs;
 497    LLVMValueRef *outputs;
 498
 499    struct lp_build_sampler_aos *sampler;
 500
 501    LLVMValueRef immediates[LP_MAX_TGSI_IMMEDIATES];
 502    LLVMValueRef temps[LP_MAX_TGSI_TEMPS];
 503    LLVMValueRef addr[LP_MAX_TGSI_ADDRS];
 504    LLVMValueRef preds[LP_MAX_TGSI_PREDS];
 505
 506    /* We allocate/use this array of temps if (1 << TGSI_FILE_TEMPORARY) is
 507     * set in the indirect_files field.
 508     * The temps[] array above is unused then.
 509     */
 510    LLVMValueRef temps_array;
 511
 512    /** bitmask indicating which register files are accessed indirectly */
 513    unsigned indirect_files;
 514
 515 };
 516
 517 static INLINE struct lp_build_tgsi_soa_context *
 518 lp_soa_context(struct lp_build_tgsi_context *bld_base)
 519 {
 520    return (struct lp_build_tgsi_soa_context *)bld_base;
 521 }
 522
 523 static INLINE struct lp_build_tgsi_aos_context *
 524 lp_aos_context(struct lp_build_tgsi_context *bld_base)
 525 {
 526    return (struct lp_build_tgsi_aos_context *)bld_base;
 527 }
 528
 529 void
 530 lp_emit_declaration_aos(
 531    struct lp_build_tgsi_aos_context *bld,
 532    const struct tgsi_full_declaration *decl);
 533
 534
 535 boolean
 536 lp_emit_instruction_aos(
 537    struct lp_build_tgsi_aos_context *bld,
 538    const struct tgsi_full_instruction *inst,
 539    const struct tgsi_opcode_info *info,
 540    int *pc);
 541
 542 void
 543 lp_emit_store_aos(
 544    struct lp_build_tgsi_aos_context *bld,
 545    const struct tgsi_full_instruction *inst,
 546    unsigned index,
 547    LLVMValueRef value);
 548
 549 void lp_build_fetch_args(
 550    struct lp_build_tgsi_context * bld_base,
 551    struct lp_build_emit_data * emit_data);
 552
 553 LLVMValueRef
 554 lp_build_tgsi_inst_llvm_aos(
 555    struct lp_build_tgsi_context * bld_base,
 556    const struct tgsi_full_instruction *inst);
 557
 558 void
 559 lp_build_tgsi_intrinsic(
 560  const struct lp_build_tgsi_action * action,
 561  struct lp_build_tgsi_context * bld_base,
 562  struct lp_build_emit_data * emit_data);
 563
 564 LLVMValueRef
 565 lp_build_emit_llvm(
 566    struct lp_build_tgsi_context *bld_base,
 567    unsigned tgsi_opcode,
 568    struct lp_build_emit_data * emit_data);
 569
 570 LLVMValueRef
 571 lp_build_emit_llvm_unary(
 572    struct lp_build_tgsi_context *bld_base,
 573    unsigned tgsi_opcode,
 574    LLVMValueRef arg0);
 575
 576 LLVMValueRef
 577 lp_build_emit_llvm_binary(
 578    struct lp_build_tgsi_context *bld_base,
 579    unsigned tgsi_opcode,
 580    LLVMValueRef arg0,
 581    LLVMValueRef arg1);
 582
 583 LLVMValueRef
 584 lp_build_emit_llvm_ternary(
 585    struct lp_build_tgsi_context *bld_base,
 586    unsigned tgsi_opcode,
 587    LLVMValueRef arg0,
 588    LLVMValueRef arg1,
 589    LLVMValueRef arg2);
 590
 591 boolean
 592 lp_build_tgsi_inst_llvm(
 593    struct lp_build_tgsi_context * bld_base,
 594    const struct tgsi_full_instruction *inst);
 595
 596 LLVMValueRef
 597 lp_build_emit_fetch(
 598    struct lp_build_tgsi_context *bld_base,
 599    const struct tgsi_full_instruction *inst,
 600    unsigned src_op,
 601    const unsigned chan_index);
 602
 603
 604 LLVMValueRef
 605 lp_build_emit_fetch_texoffset(
 606    struct lp_build_tgsi_context *bld_base,
 607    const struct tgsi_full_instruction *inst,
 608    unsigned tex_off_op,
 609    const unsigned chan_index);
 610
 611 boolean
 612 lp_build_tgsi_llvm(
 613    struct lp_build_tgsi_context * bld_base,
 614    const struct tgsi_token *tokens);
 615
 616 #endif /* LP_BLD_TGSI_H */