src/intel/common/gen_batch_decoder.c

   1 /*
   2  * Copyright © 2017 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include "common/gen_decoder.h"
  25 #include "gen_disasm.h"
  26 #include "util/macros.h"
  27
  28 #include <string.h>
  29
  30 void
  31 gen_batch_decode_ctx_init(struct gen_batch_decode_ctx *ctx,
  32                           const struct gen_device_info *devinfo,
  33                           FILE *fp, enum gen_batch_decode_flags flags,
  34                           const char *xml_path,
  35                           struct gen_batch_decode_bo (*get_bo)(void *,
  36                                                                uint64_t),
  37                           unsigned (*get_state_size)(void *, uint32_t),
  38                           void *user_data)
  39 {
  40    memset(ctx, 0, sizeof(*ctx));
  41
  42    ctx->get_bo = get_bo;
  43    ctx->get_state_size = get_state_size;
  44    ctx->user_data = user_data;
  45    ctx->fp = fp;
  46    ctx->flags = flags;
  47    ctx->max_vbo_decoded_lines = -1; /* No limit! */
  48
  49    if (xml_path == NULL)
  50       ctx->spec = gen_spec_load(devinfo);
  51    else
  52       ctx->spec = gen_spec_load_from_path(devinfo, xml_path);
  53    ctx->disasm = gen_disasm_create(devinfo);
  54 }
  55
  56 void
  57 gen_batch_decode_ctx_finish(struct gen_batch_decode_ctx *ctx)
  58 {
  59    gen_spec_destroy(ctx->spec);
  60    gen_disasm_destroy(ctx->disasm);
  61 }
  62
  63 #define CSI "\e["
  64 #define RED_COLOR    CSI "31m"
  65 #define BLUE_HEADER  CSI "0;44m"
  66 #define GREEN_HEADER CSI "1;42m"
  67 #define NORMAL       CSI "0m"
  68
  69 static void
  70 ctx_print_group(struct gen_batch_decode_ctx *ctx,
  71                 struct gen_group *group,
  72                 uint64_t address, const void *map)
  73 {
  74    gen_print_group(ctx->fp, group, address, map, 0,
  75                    (ctx->flags & GEN_BATCH_DECODE_IN_COLOR) != 0);
  76 }
  77
  78 static struct gen_batch_decode_bo
  79 ctx_get_bo(struct gen_batch_decode_ctx *ctx, uint64_t addr)
  80 {
  81    if (gen_spec_get_gen(ctx->spec) >= gen_make_gen(8,0)) {
  82       /* On Broadwell and above, we have 48-bit addresses which consume two
  83        * dwords.  Some packets require that these get stored in a "canonical
  84        * form" which means that bit 47 is sign-extended through the upper
  85        * bits. In order to correctly handle those aub dumps, we need to mask
  86        * off the top 16 bits.
  87        */
  88       addr &= (~0ull >> 16);
  89    }
  90
  91    struct gen_batch_decode_bo bo = ctx->get_bo(ctx->user_data, addr);
  92
  93    if (gen_spec_get_gen(ctx->spec) >= gen_make_gen(8,0))
  94       bo.addr &= (~0ull >> 16);
  95
  96    /* We may actually have an offset into the bo */
  97    if (bo.map != NULL) {
  98       assert(bo.addr <= addr);
  99       uint64_t offset = addr - bo.addr;
 100       bo.map += offset;
 101       bo.addr += offset;
 102       bo.size -= offset;
 103    }
 104
 105    return bo;
 106 }
 107
 108 static int
 109 update_count(struct gen_batch_decode_ctx *ctx,
 110              uint32_t offset_from_dsba,
 111              unsigned element_dwords,
 112              unsigned guess)
 113 {
 114    unsigned size = 0;
 115
 116    if (ctx->get_state_size)
 117       size = ctx->get_state_size(ctx->user_data, offset_from_dsba);
 118
 119    if (size > 0)
 120       return size / (sizeof(uint32_t) * element_dwords);
 121
 122    /* In the absence of any information, just guess arbitrarily. */
 123    return guess;
 124 }
 125
 126 static void
 127 ctx_disassemble_program(struct gen_batch_decode_ctx *ctx,
 128                         uint32_t ksp, const char *type)
 129 {
 130    uint64_t addr = ctx->instruction_base + ksp;
 131    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, addr);
 132    if (!bo.map)
 133       return;
 134
 135    fprintf(ctx->fp, "\nReferenced %s:\n", type);
 136    gen_disasm_disassemble(ctx->disasm, bo.map, 0, ctx->fp);
 137 }
 138
 139 /* Heuristic to determine whether a uint32_t is probably actually a float
 140  * (http://stackoverflow.com/a/2953466)
 141  */
 142
 143 static bool
 144 probably_float(uint32_t bits)
 145 {
 146    int exp = ((bits & 0x7f800000U) >> 23) - 127;
 147    uint32_t mant = bits & 0x007fffff;
 148
 149    /* +- 0.0 */
 150    if (exp == -127 && mant == 0)
 151       return true;
 152
 153    /* +- 1 billionth to 1 billion */
 154    if (-30 <= exp && exp <= 30)
 155       return true;
 156
 157    /* some value with only a few binary digits */
 158    if ((mant & 0x0000ffff) == 0)
 159       return true;
 160
 161    return false;
 162 }
 163
 164 static void
 165 ctx_print_buffer(struct gen_batch_decode_ctx *ctx,
 166                  struct gen_batch_decode_bo bo,
 167                  uint32_t read_length,
 168                  uint32_t pitch,
 169                  int max_lines)
 170 {
 171    const uint32_t *dw_end = bo.map + MIN2(bo.size, read_length);
 172
 173    int column_count = 0, line_count = -1;
 174    for (const uint32_t *dw = bo.map; dw < dw_end; dw++) {
 175       if (column_count * 4 == pitch || column_count == 8) {
 176          fprintf(ctx->fp, "\n");
 177          column_count = 0;
 178          line_count++;
 179
 180          if (max_lines >= 0 && line_count >= max_lines)
 181             break;
 182       }
 183       fprintf(ctx->fp, column_count == 0 ? "  " : " ");
 184
 185       if ((ctx->flags & GEN_BATCH_DECODE_FLOATS) && probably_float(*dw))
 186          fprintf(ctx->fp, "  %8.2f", *(float *) dw);
 187       else
 188          fprintf(ctx->fp, "  0x%08x", *dw);
 189
 190       column_count++;
 191    }
 192    fprintf(ctx->fp, "\n");
 193 }
 194
 195 static void
 196 handle_state_base_address(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
 197 {
 198    struct gen_group *inst = gen_spec_find_instruction(ctx->spec, p);
 199
 200    struct gen_field_iterator iter;
 201    gen_field_iterator_init(&iter, inst, p, 0, false);
 202
 203    uint64_t surface_base = 0, dynamic_base = 0, instruction_base = 0;
 204    bool surface_modify = 0, dynamic_modify = 0, instruction_modify = 0;
 205
 206    while (gen_field_iterator_next(&iter)) {
 207       if (strcmp(iter.name, "Surface State Base Address") == 0) {
 208          surface_base = iter.raw_value;
 209       } else if (strcmp(iter.name, "Dynamic State Base Address") == 0) {
 210          dynamic_base = iter.raw_value;
 211       } else if (strcmp(iter.name, "Instruction Base Address") == 0) {
 212          instruction_base = iter.raw_value;
 213       } else if (strcmp(iter.name, "Surface State Base Address Modify Enable") == 0) {
 214          surface_modify = iter.raw_value;
 215       } else if (strcmp(iter.name, "Dynamic State Base Address Modify Enable") == 0) {
 216          dynamic_modify = iter.raw_value;
 217       } else if (strcmp(iter.name, "Insntruction Base Address Modify Enable") == 0) {
 218          instruction_modify = iter.raw_value;
 219       }
 220    }
 221
 222    if (dynamic_modify)
 223       ctx->dynamic_base = dynamic_base;
 224
 225    if (surface_modify)
 226       ctx->surface_base = surface_base;
 227
 228    if (instruction_modify)
 229       ctx->instruction_base = instruction_base;
 230 }
 231
 232 static void
 233 dump_binding_table(struct gen_batch_decode_ctx *ctx, uint32_t offset, int count)
 234 {
 235    struct gen_group *strct =
 236       gen_spec_find_struct(ctx->spec, "RENDER_SURFACE_STATE");
 237    if (strct == NULL) {
 238       fprintf(ctx->fp, "did not find RENDER_SURFACE_STATE info\n");
 239       return;
 240    }
 241
 242    if (count < 0)
 243       count = update_count(ctx, offset, 1, 8);
 244
 245    if (offset % 32 != 0 || offset >= UINT16_MAX) {
 246       fprintf(ctx->fp, "  invalid binding table pointer\n");
 247       return;
 248    }
 249
 250    struct gen_batch_decode_bo bind_bo =
 251       ctx_get_bo(ctx, ctx->surface_base + offset);
 252
 253    if (bind_bo.map == NULL) {
 254       fprintf(ctx->fp, "  binding table unavailable\n");
 255       return;
 256    }
 257
 258    const uint32_t *pointers = bind_bo.map;
 259    for (int i = 0; i < count; i++) {
 260       if (pointers[i] == 0)
 261          continue;
 262
 263       uint64_t addr = ctx->surface_base + pointers[i];
 264       struct gen_batch_decode_bo bo = ctx_get_bo(ctx, addr);
 265       uint32_t size = strct->dw_length * 4;
 266
 267       if (pointers[i] % 32 != 0 ||
 268           addr < bo.addr || addr + size >= bo.addr + bo.size) {
 269          fprintf(ctx->fp, "pointer %u: 0x%08x <not valid>\n", i, pointers[i]);
 270          continue;
 271       }
 272
 273       fprintf(ctx->fp, "pointer %u: 0x%08x\n", i, pointers[i]);
 274       ctx_print_group(ctx, strct, addr, bo.map + (addr - bo.addr));
 275    }
 276 }
 277
 278 static void
 279 dump_samplers(struct gen_batch_decode_ctx *ctx, uint32_t offset, int count)
 280 {
 281    struct gen_group *strct = gen_spec_find_struct(ctx->spec, "SAMPLER_STATE");
 282
 283    if (count < 0)
 284       count = update_count(ctx, offset, strct->dw_length, 4);
 285
 286    uint64_t state_addr = ctx->dynamic_base + offset;
 287    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, state_addr);
 288    const void *state_map = bo.map;
 289
 290    if (state_map == NULL) {
 291       fprintf(ctx->fp, "  samplers unavailable\n");
 292       return;
 293    }
 294
 295    if (offset % 32 != 0 || state_addr - bo.addr >= bo.size) {
 296       fprintf(ctx->fp, "  invalid sampler state pointer\n");
 297       return;
 298    }
 299
 300    for (int i = 0; i < count; i++) {
 301       fprintf(ctx->fp, "sampler state %d\n", i);
 302       ctx_print_group(ctx, strct, state_addr, state_map);
 303       state_addr += 16;
 304       state_map += 16;
 305    }
 306 }
 307
 308 static void
 309 handle_media_interface_descriptor_load(struct gen_batch_decode_ctx *ctx,
 310                                        const uint32_t *p)
 311 {
 312    struct gen_group *inst = gen_spec_find_instruction(ctx->spec, p);
 313    struct gen_group *desc =
 314       gen_spec_find_struct(ctx->spec, "INTERFACE_DESCRIPTOR_DATA");
 315
 316    struct gen_field_iterator iter;
 317    gen_field_iterator_init(&iter, inst, p, 0, false);
 318    uint32_t descriptor_offset = 0;
 319    int descriptor_count = 0;
 320    while (gen_field_iterator_next(&iter)) {
 321       if (strcmp(iter.name, "Interface Descriptor Data Start Address") == 0) {
 322          descriptor_offset = strtol(iter.value, NULL, 16);
 323       } else if (strcmp(iter.name, "Interface Descriptor Total Length") == 0) {
 324          descriptor_count =
 325             strtol(iter.value, NULL, 16) / (desc->dw_length * 4);
 326       }
 327    }
 328
 329    uint64_t desc_addr = ctx->dynamic_base + descriptor_offset;
 330    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, desc_addr);
 331    const void *desc_map = bo.map;
 332
 333    if (desc_map == NULL) {
 334       fprintf(ctx->fp, "  interface descriptors unavailable\n");
 335       return;
 336    }
 337
 338    for (int i = 0; i < descriptor_count; i++) {
 339       fprintf(ctx->fp, "descriptor %d: %08x\n", i, descriptor_offset);
 340
 341       ctx_print_group(ctx, desc, desc_addr, desc_map);
 342
 343       gen_field_iterator_init(&iter, desc, desc_map, 0, false);
 344       uint64_t ksp = 0;
 345       uint32_t sampler_offset = 0, sampler_count = 0;
 346       uint32_t binding_table_offset = 0, binding_entry_count = 0;
 347       while (gen_field_iterator_next(&iter)) {
 348          if (strcmp(iter.name, "Kernel Start Pointer") == 0) {
 349             ksp = strtoll(iter.value, NULL, 16);
 350          } else if (strcmp(iter.name, "Sampler State Pointer") == 0) {
 351             sampler_offset = strtol(iter.value, NULL, 16);
 352          } else if (strcmp(iter.name, "Sampler Count") == 0) {
 353             sampler_count = strtol(iter.value, NULL, 10);
 354          } else if (strcmp(iter.name, "Binding Table Pointer") == 0) {
 355             binding_table_offset = strtol(iter.value, NULL, 16);
 356          } else if (strcmp(iter.name, "Binding Table Entry Count") == 0) {
 357             binding_entry_count = strtol(iter.value, NULL, 10);
 358          }
 359       }
 360
 361       ctx_disassemble_program(ctx, ksp, "compute shader");
 362       printf("\n");
 363
 364       dump_samplers(ctx, sampler_offset, sampler_count);
 365       dump_binding_table(ctx, binding_table_offset, binding_entry_count);
 366
 367       desc_map += desc->dw_length;
 368       desc_addr += desc->dw_length * 4;
 369    }
 370 }
 371
 372 static void
 373 handle_3dstate_vertex_buffers(struct gen_batch_decode_ctx *ctx,
 374                               const uint32_t *p)
 375 {
 376    struct gen_group *inst = gen_spec_find_instruction(ctx->spec, p);
 377    struct gen_group *vbs = gen_spec_find_struct(ctx->spec, "VERTEX_BUFFER_STATE");
 378
 379    struct gen_batch_decode_bo vb = {};
 380    uint32_t vb_size = 0;
 381    int index = -1;
 382    int pitch = -1;
 383    bool ready = false;
 384
 385    struct gen_field_iterator iter;
 386    gen_field_iterator_init(&iter, inst, p, 0, false);
 387    while (gen_field_iterator_next(&iter)) {
 388       if (iter.struct_desc != vbs)
 389          continue;
 390
 391       struct gen_field_iterator vbs_iter;
 392       gen_field_iterator_init(&vbs_iter, vbs, &iter.p[iter.start_bit / 32], 0, false);
 393       while (gen_field_iterator_next(&vbs_iter)) {
 394          if (strcmp(vbs_iter.name, "Vertex Buffer Index") == 0) {
 395             index = vbs_iter.raw_value;
 396          } else if (strcmp(vbs_iter.name, "Buffer Pitch") == 0) {
 397             pitch = vbs_iter.raw_value;
 398          } else if (strcmp(vbs_iter.name, "Buffer Starting Address") == 0) {
 399             vb = ctx_get_bo(ctx, vbs_iter.raw_value);
 400          } else if (strcmp(vbs_iter.name, "Buffer Size") == 0) {
 401             vb_size = vbs_iter.raw_value;
 402             ready = true;
 403          } else if (strcmp(vbs_iter.name, "End Address") == 0) {
 404             if (vb.map && vbs_iter.raw_value >= vb.addr)
 405                vb_size = vbs_iter.raw_value - vb.addr;
 406             else
 407                vb_size = 0;
 408             ready = true;
 409          }
 410
 411          if (!ready)
 412             continue;
 413
 414          fprintf(ctx->fp, "vertex buffer %d, size %d\n", index, vb_size);
 415
 416          if (vb.map == NULL) {
 417             fprintf(ctx->fp, "  buffer contents unavailable\n");
 418             continue;
 419          }
 420
 421          if (vb.map == 0 || vb_size == 0)
 422             continue;
 423
 424          ctx_print_buffer(ctx, vb, vb_size, pitch, ctx->max_vbo_decoded_lines);
 425
 426          vb.map = NULL;
 427          vb_size = 0;
 428          index = -1;
 429          pitch = -1;
 430          ready = false;
 431       }
 432    }
 433 }
 434
 435 static void
 436 handle_3dstate_index_buffer(struct gen_batch_decode_ctx *ctx,
 437                             const uint32_t *p)
 438 {
 439    struct gen_group *inst = gen_spec_find_instruction(ctx->spec, p);
 440
 441    struct gen_batch_decode_bo ib = {};
 442    uint32_t ib_size = 0;
 443    uint32_t format = 0;
 444
 445    struct gen_field_iterator iter;
 446    gen_field_iterator_init(&iter, inst, p, 0, false);
 447    while (gen_field_iterator_next(&iter)) {
 448       if (strcmp(iter.name, "Index Format") == 0) {
 449          format = iter.raw_value;
 450       } else if (strcmp(iter.name, "Buffer Starting Address") == 0) {
 451          ib = ctx_get_bo(ctx, iter.raw_value);
 452       } else if (strcmp(iter.name, "Buffer Size") == 0) {
 453          ib_size = iter.raw_value;
 454       }
 455    }
 456
 457    if (ib.map == NULL) {
 458       fprintf(ctx->fp, "  buffer contents unavailable\n");
 459       return;
 460    }
 461
 462    const void *m = ib.map;
 463    const void *ib_end = ib.map + MIN2(ib.size, ib_size);
 464    for (int i = 0; m < ib_end && i < 10; i++) {
 465       switch (format) {
 466       case 0:
 467          fprintf(ctx->fp, "%3d ", *(uint8_t *)m);
 468          m += 1;
 469          break;
 470       case 1:
 471          fprintf(ctx->fp, "%3d ", *(uint16_t *)m);
 472          m += 2;
 473          break;
 474       case 2:
 475          fprintf(ctx->fp, "%3d ", *(uint32_t *)m);
 476          m += 4;
 477          break;
 478       }
 479    }
 480
 481    if (m < ib_end)
 482       fprintf(ctx->fp, "...");
 483    fprintf(ctx->fp, "\n");
 484 }
 485
 486 static void
 487 decode_single_ksp(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
 488 {
 489    struct gen_group *inst = gen_spec_find_instruction(ctx->spec, p);
 490
 491    uint64_t ksp = 0;
 492    bool is_simd8 = false; /* vertex shaders on Gen8+ only */
 493    bool is_enabled = true;
 494
 495    struct gen_field_iterator iter;
 496    gen_field_iterator_init(&iter, inst, p, 0, false);
 497    while (gen_field_iterator_next(&iter)) {
 498       if (strcmp(iter.name, "Kernel Start Pointer") == 0) {
 499          ksp = iter.raw_value;
 500       } else if (strcmp(iter.name, "SIMD8 Dispatch Enable") == 0) {
 501          is_simd8 = iter.raw_value;
 502       } else if (strcmp(iter.name, "Dispatch Mode") == 0) {
 503          is_simd8 = strcmp(iter.value, "SIMD8") == 0;
 504       } else if (strcmp(iter.name, "Dispatch Enable") == 0) {
 505          is_simd8 = strcmp(iter.value, "SIMD8") == 0;
 506       } else if (strcmp(iter.name, "Enable") == 0) {
 507          is_enabled = iter.raw_value;
 508       }
 509    }
 510
 511    const char *type =
 512       strcmp(inst->name,   "VS_STATE") == 0 ? "vertex shader" :
 513       strcmp(inst->name,   "GS_STATE") == 0 ? "geometry shader" :
 514       strcmp(inst->name,   "SF_STATE") == 0 ? "strips and fans shader" :
 515       strcmp(inst->name, "CLIP_STATE") == 0 ? "clip shader" :
 516       strcmp(inst->name, "3DSTATE_DS") == 0 ? "tessellation evaluation shader" :
 517       strcmp(inst->name, "3DSTATE_HS") == 0 ? "tessellation control shader" :
 518       strcmp(inst->name, "3DSTATE_VS") == 0 ? (is_simd8 ? "SIMD8 vertex shader" : "vec4 vertex shader") :
 519       strcmp(inst->name, "3DSTATE_GS") == 0 ? (is_simd8 ? "SIMD8 geometry shader" : "vec4 geometry shader") :
 520       NULL;
 521
 522    if (is_enabled) {
 523       ctx_disassemble_program(ctx, ksp, type);
 524       printf("\n");
 525    }
 526 }
 527
 528 static void
 529 decode_ps_kernels(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
 530 {
 531    struct gen_group *inst = gen_spec_find_instruction(ctx->spec, p);
 532
 533    uint64_t ksp[3] = {0, 0, 0};
 534    bool enabled[3] = {false, false, false};
 535
 536    struct gen_field_iterator iter;
 537    gen_field_iterator_init(&iter, inst, p, 0, false);
 538    while (gen_field_iterator_next(&iter)) {
 539       if (strncmp(iter.name, "Kernel Start Pointer ",
 540                   strlen("Kernel Start Pointer ")) == 0) {
 541          int idx = iter.name[strlen("Kernel Start Pointer ")] - '0';
 542          ksp[idx] = strtol(iter.value, NULL, 16);
 543       } else if (strcmp(iter.name, "8 Pixel Dispatch Enable") == 0) {
 544          enabled[0] = strcmp(iter.value, "true") == 0;
 545       } else if (strcmp(iter.name, "16 Pixel Dispatch Enable") == 0) {
 546          enabled[1] = strcmp(iter.value, "true") == 0;
 547       } else if (strcmp(iter.name, "32 Pixel Dispatch Enable") == 0) {
 548          enabled[2] = strcmp(iter.value, "true") == 0;
 549       }
 550    }
 551
 552    /* Reorder KSPs to be [8, 16, 32] instead of the hardware order. */
 553    if (enabled[0] + enabled[1] + enabled[2] == 1) {
 554       if (enabled[1]) {
 555          ksp[1] = ksp[0];
 556          ksp[0] = 0;
 557       } else if (enabled[2]) {
 558          ksp[2] = ksp[0];
 559          ksp[0] = 0;
 560       }
 561    } else {
 562       uint64_t tmp = ksp[1];
 563       ksp[1] = ksp[2];
 564       ksp[2] = tmp;
 565    }
 566
 567    if (enabled[0])
 568       ctx_disassemble_program(ctx, ksp[0], "SIMD8 fragment shader");
 569    if (enabled[1])
 570       ctx_disassemble_program(ctx, ksp[1], "SIMD16 fragment shader");
 571    if (enabled[2])
 572       ctx_disassemble_program(ctx, ksp[2], "SIMD32 fragment shader");
 573    fprintf(ctx->fp, "\n");
 574 }
 575
 576 static void
 577 decode_3dstate_constant(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
 578 {
 579    struct gen_group *inst = gen_spec_find_instruction(ctx->spec, p);
 580    struct gen_group *body =
 581       gen_spec_find_struct(ctx->spec, "3DSTATE_CONSTANT_BODY");
 582
 583    uint32_t read_length[4] = {0};
 584    uint64_t read_addr[4];
 585
 586    struct gen_field_iterator outer;
 587    gen_field_iterator_init(&outer, inst, p, 0, false);
 588    while (gen_field_iterator_next(&outer)) {
 589       if (outer.struct_desc != body)
 590          continue;
 591
 592       struct gen_field_iterator iter;
 593       gen_field_iterator_init(&iter, body, &outer.p[outer.start_bit / 32],
 594                               0, false);
 595
 596       while (gen_field_iterator_next(&iter)) {
 597          int idx;
 598          if (sscanf(iter.name, "Read Length[%d]", &idx) == 1) {
 599             read_length[idx] = iter.raw_value;
 600          } else if (sscanf(iter.name, "Buffer[%d]", &idx) == 1) {
 601             read_addr[idx] = iter.raw_value;
 602          }
 603       }
 604
 605       for (int i = 0; i < 4; i++) {
 606          if (read_length[i] == 0)
 607             continue;
 608
 609          struct gen_batch_decode_bo buffer = ctx_get_bo(ctx, read_addr[i]);
 610          if (!buffer.map) {
 611             fprintf(ctx->fp, "constant buffer %d unavailable\n", i);
 612             continue;
 613          }
 614
 615          unsigned size = read_length[i] * 32;
 616          fprintf(ctx->fp, "constant buffer %d, size %u\n", i, size);
 617
 618          ctx_print_buffer(ctx, buffer, size, 0, -1);
 619       }
 620    }
 621 }
 622
 623 static void
 624 decode_3dstate_binding_table_pointers(struct gen_batch_decode_ctx *ctx,
 625                                       const uint32_t *p)
 626 {
 627    dump_binding_table(ctx, p[1], -1);
 628 }
 629
 630 static void
 631 decode_3dstate_sampler_state_pointers(struct gen_batch_decode_ctx *ctx,
 632                                       const uint32_t *p)
 633 {
 634    dump_samplers(ctx, p[1], -1);
 635 }
 636
 637 static void
 638 decode_3dstate_sampler_state_pointers_gen6(struct gen_batch_decode_ctx *ctx,
 639                                            const uint32_t *p)
 640 {
 641    dump_samplers(ctx, p[1], -1);
 642    dump_samplers(ctx, p[2], -1);
 643    dump_samplers(ctx, p[3], -1);
 644 }
 645
 646 static bool
 647 str_ends_with(const char *str, const char *end)
 648 {
 649    int offset = strlen(str) - strlen(end);
 650    if (offset < 0)
 651       return false;
 652
 653    return strcmp(str + offset, end) == 0;
 654 }
 655
 656 static void
 657 decode_dynamic_state_pointers(struct gen_batch_decode_ctx *ctx,
 658                               const char *struct_type, const uint32_t *p,
 659                               int count)
 660 {
 661    struct gen_group *inst = gen_spec_find_instruction(ctx->spec, p);
 662
 663    uint32_t state_offset = 0;
 664
 665    struct gen_field_iterator iter;
 666    gen_field_iterator_init(&iter, inst, p, 0, false);
 667    while (gen_field_iterator_next(&iter)) {
 668       if (str_ends_with(iter.name, "Pointer")) {
 669          state_offset = iter.raw_value;
 670          break;
 671       }
 672    }
 673
 674    uint64_t state_addr = ctx->dynamic_base + state_offset;
 675    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, state_addr);
 676    const void *state_map = bo.map;
 677
 678    if (state_map == NULL) {
 679       fprintf(ctx->fp, "  dynamic %s state unavailable\n", struct_type);
 680       return;
 681    }
 682
 683    struct gen_group *state = gen_spec_find_struct(ctx->spec, struct_type);
 684    if (strcmp(struct_type, "BLEND_STATE") == 0) {
 685       /* Blend states are different from the others because they have a header
 686        * struct called BLEND_STATE which is followed by a variable number of
 687        * BLEND_STATE_ENTRY structs.
 688        */
 689       fprintf(ctx->fp, "%s\n", struct_type);
 690       ctx_print_group(ctx, state, state_addr, state_map);
 691
 692       state_addr += state->dw_length * 4;
 693       state_map += state->dw_length * 4;
 694
 695       struct_type = "BLEND_STATE_ENTRY";
 696       state = gen_spec_find_struct(ctx->spec, struct_type);
 697    }
 698
 699    for (int i = 0; i < count; i++) {
 700       fprintf(ctx->fp, "%s %d\n", struct_type, i);
 701       ctx_print_group(ctx, state, state_addr, state_map);
 702
 703       state_addr += state->dw_length * 4;
 704       state_map += state->dw_length * 4;
 705    }
 706 }
 707
 708 static void
 709 decode_3dstate_viewport_state_pointers_cc(struct gen_batch_decode_ctx *ctx,
 710                                           const uint32_t *p)
 711 {
 712    decode_dynamic_state_pointers(ctx, "CC_VIEWPORT", p, 4);
 713 }
 714
 715 static void
 716 decode_3dstate_viewport_state_pointers_sf_clip(struct gen_batch_decode_ctx *ctx,
 717                                                const uint32_t *p)
 718 {
 719    decode_dynamic_state_pointers(ctx, "SF_CLIP_VIEWPORT", p, 4);
 720 }
 721
 722 static void
 723 decode_3dstate_blend_state_pointers(struct gen_batch_decode_ctx *ctx,
 724                                     const uint32_t *p)
 725 {
 726    decode_dynamic_state_pointers(ctx, "BLEND_STATE", p, 1);
 727 }
 728
 729 static void
 730 decode_3dstate_cc_state_pointers(struct gen_batch_decode_ctx *ctx,
 731                                  const uint32_t *p)
 732 {
 733    decode_dynamic_state_pointers(ctx, "COLOR_CALC_STATE", p, 1);
 734 }
 735
 736 static void
 737 decode_3dstate_scissor_state_pointers(struct gen_batch_decode_ctx *ctx,
 738                                       const uint32_t *p)
 739 {
 740    decode_dynamic_state_pointers(ctx, "SCISSOR_RECT", p, 1);
 741 }
 742
 743 static void
 744 decode_load_register_imm(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
 745 {
 746    struct gen_group *reg = gen_spec_find_register(ctx->spec, p[1]);
 747
 748    if (reg != NULL) {
 749       fprintf(ctx->fp, "register %s (0x%x): 0x%x\n",
 750               reg->name, reg->register_offset, p[2]);
 751       ctx_print_group(ctx, reg, reg->register_offset, &p[2]);
 752    }
 753 }
 754
 755 struct custom_decoder {
 756    const char *cmd_name;
 757    void (*decode)(struct gen_batch_decode_ctx *ctx, const uint32_t *p);
 758 } custom_decoders[] = {
 759    { "STATE_BASE_ADDRESS", handle_state_base_address },
 760    { "MEDIA_INTERFACE_DESCRIPTOR_LOAD", handle_media_interface_descriptor_load },
 761    { "3DSTATE_VERTEX_BUFFERS", handle_3dstate_vertex_buffers },
 762    { "3DSTATE_INDEX_BUFFER", handle_3dstate_index_buffer },
 763    { "3DSTATE_VS", decode_single_ksp },
 764    { "3DSTATE_GS", decode_single_ksp },
 765    { "3DSTATE_DS", decode_single_ksp },
 766    { "3DSTATE_HS", decode_single_ksp },
 767    { "3DSTATE_PS", decode_ps_kernels },
 768    { "3DSTATE_CONSTANT_VS", decode_3dstate_constant },
 769    { "3DSTATE_CONSTANT_GS", decode_3dstate_constant },
 770    { "3DSTATE_CONSTANT_PS", decode_3dstate_constant },
 771    { "3DSTATE_CONSTANT_HS", decode_3dstate_constant },
 772    { "3DSTATE_CONSTANT_DS", decode_3dstate_constant },
 773
 774    { "3DSTATE_BINDING_TABLE_POINTERS_VS", decode_3dstate_binding_table_pointers },
 775    { "3DSTATE_BINDING_TABLE_POINTERS_HS", decode_3dstate_binding_table_pointers },
 776    { "3DSTATE_BINDING_TABLE_POINTERS_DS", decode_3dstate_binding_table_pointers },
 777    { "3DSTATE_BINDING_TABLE_POINTERS_GS", decode_3dstate_binding_table_pointers },
 778    { "3DSTATE_BINDING_TABLE_POINTERS_PS", decode_3dstate_binding_table_pointers },
 779
 780    { "3DSTATE_SAMPLER_STATE_POINTERS_VS", decode_3dstate_sampler_state_pointers },
 781    { "3DSTATE_SAMPLER_STATE_POINTERS_HS", decode_3dstate_sampler_state_pointers },
 782    { "3DSTATE_SAMPLER_STATE_POINTERS_DS", decode_3dstate_sampler_state_pointers },
 783    { "3DSTATE_SAMPLER_STATE_POINTERS_GS", decode_3dstate_sampler_state_pointers },
 784    { "3DSTATE_SAMPLER_STATE_POINTERS_PS", decode_3dstate_sampler_state_pointers },
 785    { "3DSTATE_SAMPLER_STATE_POINTERS", decode_3dstate_sampler_state_pointers_gen6 },
 786
 787    { "3DSTATE_VIEWPORT_STATE_POINTERS_CC", decode_3dstate_viewport_state_pointers_cc },
 788    { "3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP", decode_3dstate_viewport_state_pointers_sf_clip },
 789    { "3DSTATE_BLEND_STATE_POINTERS", decode_3dstate_blend_state_pointers },
 790    { "3DSTATE_CC_STATE_POINTERS", decode_3dstate_cc_state_pointers },
 791    { "3DSTATE_SCISSOR_STATE_POINTERS", decode_3dstate_scissor_state_pointers },
 792    { "MI_LOAD_REGISTER_IMM", decode_load_register_imm }
 793 };
 794
 795 void
 796 gen_print_batch(struct gen_batch_decode_ctx *ctx,
 797                 const uint32_t *batch, uint32_t batch_size,
 798                 uint64_t batch_addr)
 799 {
 800    const uint32_t *p, *end = batch + batch_size / sizeof(uint32_t);
 801    int length;
 802    struct gen_group *inst;
 803
 804    for (p = batch; p < end; p += length) {
 805       inst = gen_spec_find_instruction(ctx->spec, p);
 806       length = gen_group_get_length(inst, p);
 807       assert(inst == NULL || length > 0);
 808       length = MAX2(1, length);
 809
 810       const char *reset_color = ctx->flags & GEN_BATCH_DECODE_IN_COLOR ? NORMAL : "";
 811
 812       uint64_t offset;
 813       if (ctx->flags & GEN_BATCH_DECODE_OFFSETS)
 814          offset = batch_addr + ((char *)p - (char *)batch);
 815       else
 816          offset = 0;
 817
 818       if (inst == NULL) {
 819          fprintf(ctx->fp, "%s0x%08"PRIx64": unknown instruction %08x%s\n",
 820                  (ctx->flags & GEN_BATCH_DECODE_IN_COLOR) ? RED_COLOR : "",
 821                  offset, p[0], reset_color);
 822          continue;
 823       }
 824
 825       const char *color;
 826       const char *inst_name = gen_group_get_name(inst);
 827       if (ctx->flags & GEN_BATCH_DECODE_IN_COLOR) {
 828          reset_color = NORMAL;
 829          if (ctx->flags & GEN_BATCH_DECODE_FULL) {
 830             if (strcmp(inst_name, "MI_BATCH_BUFFER_START") == 0 ||
 831                 strcmp(inst_name, "MI_BATCH_BUFFER_END") == 0)
 832                color = GREEN_HEADER;
 833             else
 834                color = BLUE_HEADER;
 835          } else {
 836             color = NORMAL;
 837          }
 838       } else {
 839          color = "";
 840          reset_color = "";
 841       }
 842
 843       fprintf(ctx->fp, "%s0x%08"PRIx64":  0x%08x:  %-80s%s\n",
 844               color, offset, p[0], inst_name, reset_color);
 845
 846       if (ctx->flags & GEN_BATCH_DECODE_FULL) {
 847          ctx_print_group(ctx, inst, offset, p);
 848
 849          for (int i = 0; i < ARRAY_SIZE(custom_decoders); i++) {
 850             if (strcmp(inst_name, custom_decoders[i].cmd_name) == 0) {
 851                custom_decoders[i].decode(ctx, p);
 852                break;
 853             }
 854          }
 855       }
 856
 857       if (strcmp(inst_name, "MI_BATCH_BUFFER_START") == 0) {
 858          struct gen_batch_decode_bo next_batch = {};
 859          bool second_level;
 860          struct gen_field_iterator iter;
 861          gen_field_iterator_init(&iter, inst, p, 0, false);
 862          while (gen_field_iterator_next(&iter)) {
 863             if (strcmp(iter.name, "Batch Buffer Start Address") == 0) {
 864                next_batch = ctx_get_bo(ctx, iter.raw_value);
 865             } else if (strcmp(iter.name, "Second Level Batch Buffer") == 0) {
 866                second_level = iter.raw_value;
 867             }
 868          }
 869
 870          if (next_batch.map == NULL) {
 871             fprintf(ctx->fp, "Secondary batch at 0x%08"PRIx64" unavailable\n",
 872                     next_batch.addr);
 873          } else {
 874             gen_print_batch(ctx, next_batch.map, next_batch.size,
 875                             next_batch.addr);
 876          }
 877          if (second_level) {
 878             /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" set acts
 879              * like a subroutine call.  Commands that come afterwards get
 880              * processed once the 2nd level batch buffer returns with
 881              * MI_BATCH_BUFFER_END.
 882              */
 883             continue;
 884          } else {
 885             /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" unset acts
 886              * like a goto.  Nothing after it will ever get processed.  In
 887              * order to prevent the recursion from growing, we just reset the
 888              * loop and continue;
 889              */
 890             break;
 891          }
 892       } else if (strcmp(inst_name, "MI_BATCH_BUFFER_END") == 0) {
 893          break;
 894       }
 895    }
 896 }