src/intel/tools/aubinator_viewer_decoder.cpp

   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 <string.h>
  25
  26 #include "util/macros.h"
  27
  28 #include "aubinator_viewer.h"
  29
  30 void
  31 aub_viewer_decode_ctx_init(struct aub_viewer_decode_ctx *ctx,
  32                            struct aub_viewer_cfg *cfg,
  33                            struct aub_viewer_decode_cfg *decode_cfg,
  34                            struct gen_spec *spec,
  35                            struct gen_disasm *disasm,
  36                            struct gen_batch_decode_bo (*get_bo)(void *, 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->engine = I915_ENGINE_CLASS_RENDER;
  46
  47    ctx->cfg = cfg;
  48    ctx->decode_cfg = decode_cfg;
  49    ctx->spec = spec;
  50    ctx->disasm = disasm;
  51 }
  52
  53 static void
  54 aub_viewer_print_group(struct aub_viewer_decode_ctx *ctx,
  55                        struct gen_group *group,
  56                        uint64_t address, const void *map)
  57 {
  58    struct gen_field_iterator iter;
  59    int last_dword = -1;
  60    const uint32_t *p = (const uint32_t *) map;
  61
  62    gen_field_iterator_init(&iter, group, p, 0, false);
  63    while (gen_field_iterator_next(&iter)) {
  64       if (ctx->decode_cfg->show_dwords) {
  65          int iter_dword = iter.end_bit / 32;
  66          if (last_dword != iter_dword) {
  67             for (int i = last_dword + 1; i <= iter_dword; i++) {
  68                ImGui::TextColored(ctx->cfg->dwords_color,
  69                                   "0x%012" PRIx64 ":  0x%012x : Dword %d",
  70                                   address + 4 * i, iter.p[i], i);
  71             }
  72             last_dword = iter_dword;
  73          }
  74       }
  75       if (!gen_field_is_header(iter.field)) {
  76          if (ctx->decode_cfg->field_filter.PassFilter(iter.name)) {
  77             if (iter.field->type.kind == gen_type::GEN_TYPE_BOOL && iter.raw_value) {
  78                ImGui::Text("%s: ", iter.name); ImGui::SameLine();
  79                ImGui::TextColored(ctx->cfg->boolean_color, "true");
  80             } else {
  81                ImGui::Text("%s: %s", iter.name, iter.value);
  82             }
  83             if (iter.struct_desc) {
  84                int struct_dword = iter.start_bit / 32;
  85                uint64_t struct_address = address + 4 * struct_dword;
  86                aub_viewer_print_group(ctx, iter.struct_desc, struct_address,
  87                                       &p[struct_dword]);
  88             }
  89          }
  90       }
  91    }
  92 }
  93
  94 static struct gen_batch_decode_bo
  95 ctx_get_bo(struct aub_viewer_decode_ctx *ctx, uint64_t addr)
  96 {
  97    if (gen_spec_get_gen(ctx->spec) >= gen_make_gen(8,0)) {
  98       /* On Broadwell and above, we have 48-bit addresses which consume two
  99        * dwords.  Some packets require that these get stored in a "canonical
 100        * form" which means that bit 47 is sign-extended through the upper
 101        * bits. In order to correctly handle those aub dumps, we need to mask
 102        * off the top 16 bits.
 103        */
 104       addr &= (~0ull >> 16);
 105    }
 106
 107    struct gen_batch_decode_bo bo = ctx->get_bo(ctx->user_data, addr);
 108
 109    if (gen_spec_get_gen(ctx->spec) >= gen_make_gen(8,0))
 110       bo.addr &= (~0ull >> 16);
 111
 112    /* We may actually have an offset into the bo */
 113    if (bo.map != NULL) {
 114       assert(bo.addr <= addr);
 115       uint64_t offset = addr - bo.addr;
 116       bo.map = (const uint8_t *)bo.map + offset;
 117       bo.addr += offset;
 118       bo.size -= offset;
 119    }
 120
 121    return bo;
 122 }
 123
 124 static int
 125 update_count(struct aub_viewer_decode_ctx *ctx,
 126              uint32_t offset_from_dsba,
 127              unsigned element_dwords,
 128              unsigned guess)
 129 {
 130    unsigned size = 0;
 131
 132    if (ctx->get_state_size)
 133       size = ctx->get_state_size(ctx->user_data, offset_from_dsba);
 134
 135    if (size > 0)
 136       return size / (sizeof(uint32_t) * element_dwords);
 137
 138    /* In the absence of any information, just guess arbitrarily. */
 139    return guess;
 140 }
 141
 142 static void
 143 ctx_disassemble_program(struct aub_viewer_decode_ctx *ctx,
 144                         uint32_t ksp, const char *type)
 145 {
 146    uint64_t addr = ctx->instruction_base + ksp;
 147    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, addr);
 148    if (!bo.map) {
 149       ImGui::TextColored(ctx->cfg->missing_color,
 150                          "Shader unavailable addr=0x%012" PRIx64, addr);
 151       return;
 152    }
 153
 154    ImGui::PushID(addr);
 155    if (ImGui::Button(type) && ctx->display_shader)
 156       ctx->display_shader(ctx->user_data, type, addr);
 157    ImGui::PopID();
 158 }
 159
 160 static void
 161 handle_state_base_address(struct aub_viewer_decode_ctx *ctx,
 162                           struct gen_group *inst,
 163                           const uint32_t *p)
 164 {
 165    struct gen_field_iterator iter;
 166    gen_field_iterator_init(&iter, inst, p, 0, false);
 167
 168    uint64_t surface_base = 0, dynamic_base = 0, instruction_base = 0;
 169    bool surface_modify = 0, dynamic_modify = 0, instruction_modify = 0;
 170
 171    while (gen_field_iterator_next(&iter)) {
 172       if (strcmp(iter.name, "Surface State Base Address") == 0) {
 173          surface_base = iter.raw_value;
 174       } else if (strcmp(iter.name, "Dynamic State Base Address") == 0) {
 175          dynamic_base = iter.raw_value;
 176       } else if (strcmp(iter.name, "Instruction Base Address") == 0) {
 177          instruction_base = iter.raw_value;
 178       } else if (strcmp(iter.name, "Surface State Base Address Modify Enable") == 0) {
 179          surface_modify = iter.raw_value;
 180       } else if (strcmp(iter.name, "Dynamic State Base Address Modify Enable") == 0) {
 181          dynamic_modify = iter.raw_value;
 182       } else if (strcmp(iter.name, "Instruction Base Address Modify Enable") == 0) {
 183          instruction_modify = iter.raw_value;
 184       }
 185    }
 186
 187    if (dynamic_modify)
 188       ctx->dynamic_base = dynamic_base;
 189
 190    if (surface_modify)
 191       ctx->surface_base = surface_base;
 192
 193    if (instruction_modify)
 194       ctx->instruction_base = instruction_base;
 195 }
 196
 197 static void
 198 dump_binding_table(struct aub_viewer_decode_ctx *ctx, uint32_t offset, int count)
 199 {
 200    struct gen_group *strct =
 201       gen_spec_find_struct(ctx->spec, "RENDER_SURFACE_STATE");
 202    if (strct == NULL) {
 203       ImGui::TextColored(ctx->cfg->missing_color, "did not find RENDER_SURFACE_STATE info");
 204       return;
 205    }
 206
 207    if (count < 0)
 208       count = update_count(ctx, offset, 1, 8);
 209
 210    if (offset % 32 != 0 || offset >= UINT16_MAX) {
 211       ImGui::TextColored(ctx->cfg->missing_color, "invalid binding table pointer");
 212       return;
 213    }
 214
 215    struct gen_batch_decode_bo bind_bo =
 216       ctx_get_bo(ctx, ctx->surface_base + offset);
 217
 218    if (bind_bo.map == NULL) {
 219       ImGui::TextColored(ctx->cfg->missing_color,
 220                          "binding table unavailable addr=0x%012" PRIx64,
 221                          ctx->surface_base + offset);
 222       return;
 223    }
 224
 225    const uint32_t *pointers = (const uint32_t *) bind_bo.map;
 226    for (int i = 0; i < count; i++) {
 227       if (pointers[i] == 0)
 228          continue;
 229
 230       uint64_t addr = ctx->surface_base + pointers[i];
 231       struct gen_batch_decode_bo bo = ctx_get_bo(ctx, addr);
 232       uint32_t size = strct->dw_length * 4;
 233
 234       if (pointers[i] % 32 != 0 ||
 235           addr < bo.addr || addr + size >= bo.addr + bo.size) {
 236          ImGui::TextColored(ctx->cfg->missing_color,
 237                             "pointer %u: %012x <not valid>", i, pointers[i]);
 238          continue;
 239       }
 240
 241       const uint8_t *state = (const uint8_t *) bo.map + (addr - bo.addr);
 242       if (ImGui::TreeNodeEx(&pointers[i], ImGuiTreeNodeFlags_Framed,
 243                             "pointer %u: %012x", i, pointers[i])) {
 244          aub_viewer_print_group(ctx, strct, addr, state);
 245          ImGui::TreePop();
 246       }
 247    }
 248 }
 249
 250 static void
 251 dump_samplers(struct aub_viewer_decode_ctx *ctx, uint32_t offset, int count)
 252 {
 253    struct gen_group *strct = gen_spec_find_struct(ctx->spec, "SAMPLER_STATE");
 254
 255    if (count < 0)
 256       count = update_count(ctx, offset, strct->dw_length, 4);
 257
 258    uint64_t state_addr = ctx->dynamic_base + offset;
 259    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, state_addr);
 260    const uint8_t *state_map = (const uint8_t *) bo.map;
 261
 262    if (state_map == NULL) {
 263       ImGui::TextColored(ctx->cfg->missing_color,
 264                          "samplers unavailable addr=0x%012" PRIx64, state_addr);
 265       return;
 266    }
 267
 268    if (offset % 32 != 0 || state_addr - bo.addr >= bo.size) {
 269       ImGui::TextColored(ctx->cfg->missing_color, "invalid sampler state pointer");
 270       return;
 271    }
 272
 273    for (int i = 0; i < count; i++) {
 274       if (ImGui::TreeNodeEx(state_map, ImGuiTreeNodeFlags_Framed,
 275                             "sampler state %d", i)) {
 276          aub_viewer_print_group(ctx, strct, state_addr, state_map);
 277          ImGui::TreePop();
 278       }
 279       state_addr += 16;
 280       state_map += 16;
 281    }
 282 }
 283
 284 static void
 285 handle_media_interface_descriptor_load(struct aub_viewer_decode_ctx *ctx,
 286                                        struct gen_group *inst,
 287                                        const uint32_t *p)
 288 {
 289    struct gen_group *desc =
 290       gen_spec_find_struct(ctx->spec, "INTERFACE_DESCRIPTOR_DATA");
 291
 292    struct gen_field_iterator iter;
 293    gen_field_iterator_init(&iter, inst, p, 0, false);
 294    uint32_t descriptor_offset = 0;
 295    int descriptor_count = 0;
 296    while (gen_field_iterator_next(&iter)) {
 297       if (strcmp(iter.name, "Interface Descriptor Data Start Address") == 0) {
 298          descriptor_offset = strtol(iter.value, NULL, 16);
 299       } else if (strcmp(iter.name, "Interface Descriptor Total Length") == 0) {
 300          descriptor_count =
 301             strtol(iter.value, NULL, 16) / (desc->dw_length * 4);
 302       }
 303    }
 304
 305    uint64_t desc_addr = ctx->dynamic_base + descriptor_offset;
 306    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, desc_addr);
 307    const uint32_t *desc_map = (const uint32_t *) bo.map;
 308
 309    if (desc_map == NULL) {
 310       ImGui::TextColored(ctx->cfg->missing_color,
 311                          "interface descriptors unavailable addr=0x%012" PRIx64, desc_addr);
 312       return;
 313    }
 314
 315    for (int i = 0; i < descriptor_count; i++) {
 316       ImGui::Text("descriptor %d: %012x", i, descriptor_offset);
 317
 318       aub_viewer_print_group(ctx, desc, desc_addr, desc_map);
 319
 320       gen_field_iterator_init(&iter, desc, desc_map, 0, false);
 321       uint64_t ksp = 0;
 322       uint32_t sampler_offset = 0, sampler_count = 0;
 323       uint32_t binding_table_offset = 0, binding_entry_count = 0;
 324       while (gen_field_iterator_next(&iter)) {
 325          if (strcmp(iter.name, "Kernel Start Pointer") == 0) {
 326             ksp = strtoll(iter.value, NULL, 16);
 327          } else if (strcmp(iter.name, "Sampler State Pointer") == 0) {
 328             sampler_offset = strtol(iter.value, NULL, 16);
 329          } else if (strcmp(iter.name, "Sampler Count") == 0) {
 330             sampler_count = strtol(iter.value, NULL, 10);
 331          } else if (strcmp(iter.name, "Binding Table Pointer") == 0) {
 332             binding_table_offset = strtol(iter.value, NULL, 16);
 333          } else if (strcmp(iter.name, "Binding Table Entry Count") == 0) {
 334             binding_entry_count = strtol(iter.value, NULL, 10);
 335          }
 336       }
 337
 338       ctx_disassemble_program(ctx, ksp, "compute shader");
 339
 340       dump_samplers(ctx, sampler_offset, sampler_count);
 341       dump_binding_table(ctx, binding_table_offset, binding_entry_count);
 342
 343       desc_map += desc->dw_length;
 344       desc_addr += desc->dw_length * 4;
 345    }
 346 }
 347
 348 static void
 349 handle_3dstate_vertex_buffers(struct aub_viewer_decode_ctx *ctx,
 350                               struct gen_group *inst,
 351                               const uint32_t *p)
 352 {
 353    struct gen_group *vbs = gen_spec_find_struct(ctx->spec, "VERTEX_BUFFER_STATE");
 354
 355    struct gen_batch_decode_bo vb = {};
 356    uint32_t vb_size = 0;
 357    int index = -1;
 358    int pitch = -1;
 359    bool ready = false;
 360
 361    struct gen_field_iterator iter;
 362    gen_field_iterator_init(&iter, inst, p, 0, false);
 363    while (gen_field_iterator_next(&iter)) {
 364       if (iter.struct_desc != vbs)
 365          continue;
 366
 367       uint64_t buffer_addr;
 368
 369       struct gen_field_iterator vbs_iter;
 370       gen_field_iterator_init(&vbs_iter, vbs, &iter.p[iter.start_bit / 32], 0, false);
 371       while (gen_field_iterator_next(&vbs_iter)) {
 372          if (strcmp(vbs_iter.name, "Vertex Buffer Index") == 0) {
 373             index = vbs_iter.raw_value;
 374          } else if (strcmp(vbs_iter.name, "Buffer Pitch") == 0) {
 375             pitch = vbs_iter.raw_value;
 376          } else if (strcmp(vbs_iter.name, "Buffer Starting Address") == 0) {
 377             buffer_addr = vbs_iter.raw_value;
 378             vb = ctx_get_bo(ctx, buffer_addr);
 379          } else if (strcmp(vbs_iter.name, "Buffer Size") == 0) {
 380             vb_size = vbs_iter.raw_value;
 381             ready = true;
 382          } else if (strcmp(vbs_iter.name, "End Address") == 0) {
 383             if (vb.map && vbs_iter.raw_value >= vb.addr)
 384                vb_size = vbs_iter.raw_value - vb.addr;
 385             else
 386                vb_size = 0;
 387             ready = true;
 388          }
 389
 390          if (!ready)
 391             continue;
 392
 393          ImGui::Text("vertex buffer %d, size %d, pitch %d", index, vb_size, pitch);
 394
 395          if (vb.map == NULL) {
 396             ImGui::TextColored(ctx->cfg->missing_color,
 397                                "buffer contents unavailable addr=0x%012" PRIx64, buffer_addr);
 398             continue;
 399          }
 400
 401          if (vb.map == 0 || vb_size == 0)
 402             continue;
 403
 404          vb.map = NULL;
 405          vb_size = 0;
 406          index = -1;
 407          pitch = -1;
 408          ready = false;
 409       }
 410    }
 411 }
 412
 413 static void
 414 handle_3dstate_index_buffer(struct aub_viewer_decode_ctx *ctx,
 415                             struct gen_group *inst,
 416                             const uint32_t *p)
 417 {
 418    struct gen_batch_decode_bo ib = {};
 419    uint64_t buffer_addr = 0;
 420    uint32_t ib_size = 0;
 421    uint32_t format = 0;
 422
 423    struct gen_field_iterator iter;
 424    gen_field_iterator_init(&iter, inst, p, 0, false);
 425    while (gen_field_iterator_next(&iter)) {
 426       if (strcmp(iter.name, "Index Format") == 0) {
 427          format = iter.raw_value;
 428       } else if (strcmp(iter.name, "Buffer Starting Address") == 0) {
 429          buffer_addr = iter.raw_value;
 430          ib = ctx_get_bo(ctx, buffer_addr);
 431       } else if (strcmp(iter.name, "Buffer Size") == 0) {
 432          ib_size = iter.raw_value;
 433       }
 434    }
 435
 436    if (ib.map == NULL) {
 437       ImGui::TextColored(ctx->cfg->missing_color,
 438                          "buffer contents unavailable addr=0x%012" PRIx64,
 439                          buffer_addr);
 440       return;
 441    }
 442
 443    const uint8_t *m = (const uint8_t *) ib.map;
 444    const uint8_t *ib_end = m + MIN2(ib.size, ib_size);
 445    for (int i = 0; m < ib_end && i < 10; i++) {
 446       switch (format) {
 447       case 0:
 448          m += 1;
 449          break;
 450       case 1:
 451          m += 2;
 452          break;
 453       case 2:
 454          m += 4;
 455          break;
 456       }
 457    }
 458 }
 459
 460 static void
 461 decode_single_ksp(struct aub_viewer_decode_ctx *ctx,
 462                   struct gen_group *inst,
 463                   const uint32_t *p)
 464 {
 465    uint64_t ksp = 0;
 466    bool is_simd8 = false; /* vertex shaders on Gen8+ only */
 467    bool is_enabled = true;
 468
 469    struct gen_field_iterator iter;
 470    gen_field_iterator_init(&iter, inst, p, 0, false);
 471    while (gen_field_iterator_next(&iter)) {
 472       if (strcmp(iter.name, "Kernel Start Pointer") == 0) {
 473          ksp = iter.raw_value;
 474       } else if (strcmp(iter.name, "SIMD8 Dispatch Enable") == 0) {
 475          is_simd8 = iter.raw_value;
 476       } else if (strcmp(iter.name, "Dispatch Mode") == 0) {
 477          is_simd8 = strcmp(iter.value, "SIMD8") == 0;
 478       } else if (strcmp(iter.name, "Dispatch Enable") == 0) {
 479          is_simd8 = strcmp(iter.value, "SIMD8") == 0;
 480       } else if (strcmp(iter.name, "Enable") == 0) {
 481          is_enabled = iter.raw_value;
 482       }
 483    }
 484
 485    const char *type =
 486       strcmp(inst->name,   "VS_STATE") == 0 ? "vertex shader" :
 487       strcmp(inst->name,   "GS_STATE") == 0 ? "geometry shader" :
 488       strcmp(inst->name,   "SF_STATE") == 0 ? "strips and fans shader" :
 489       strcmp(inst->name, "CLIP_STATE") == 0 ? "clip shader" :
 490       strcmp(inst->name, "3DSTATE_DS") == 0 ? "tessellation evaluation shader" :
 491       strcmp(inst->name, "3DSTATE_HS") == 0 ? "tessellation control shader" :
 492       strcmp(inst->name, "3DSTATE_VS") == 0 ? (is_simd8 ? "SIMD8 vertex shader" : "vec4 vertex shader") :
 493       strcmp(inst->name, "3DSTATE_GS") == 0 ? (is_simd8 ? "SIMD8 geometry shader" : "vec4 geometry shader") :
 494       NULL;
 495
 496    if (is_enabled)
 497       ctx_disassemble_program(ctx, ksp, type);
 498 }
 499
 500 static void
 501 decode_ps_kernels(struct aub_viewer_decode_ctx *ctx,
 502                   struct gen_group *inst,
 503                   const uint32_t *p)
 504 {
 505    uint64_t ksp[3] = {0, 0, 0};
 506    bool enabled[3] = {false, false, false};
 507
 508    struct gen_field_iterator iter;
 509    gen_field_iterator_init(&iter, inst, p, 0, false);
 510    while (gen_field_iterator_next(&iter)) {
 511       if (strncmp(iter.name, "Kernel Start Pointer ",
 512                   strlen("Kernel Start Pointer ")) == 0) {
 513          int idx = iter.name[strlen("Kernel Start Pointer ")] - '0';
 514          ksp[idx] = strtol(iter.value, NULL, 16);
 515       } else if (strcmp(iter.name, "8 Pixel Dispatch Enable") == 0) {
 516          enabled[0] = strcmp(iter.value, "true") == 0;
 517       } else if (strcmp(iter.name, "16 Pixel Dispatch Enable") == 0) {
 518          enabled[1] = strcmp(iter.value, "true") == 0;
 519       } else if (strcmp(iter.name, "32 Pixel Dispatch Enable") == 0) {
 520          enabled[2] = strcmp(iter.value, "true") == 0;
 521       }
 522    }
 523
 524    /* Reorder KSPs to be [8, 16, 32] instead of the hardware order. */
 525    if (enabled[0] + enabled[1] + enabled[2] == 1) {
 526       if (enabled[1]) {
 527          ksp[1] = ksp[0];
 528          ksp[0] = 0;
 529       } else if (enabled[2]) {
 530          ksp[2] = ksp[0];
 531          ksp[0] = 0;
 532       }
 533    } else {
 534       uint64_t tmp = ksp[1];
 535       ksp[1] = ksp[2];
 536       ksp[2] = tmp;
 537    }
 538
 539    if (enabled[0])
 540       ctx_disassemble_program(ctx, ksp[0], "SIMD8 fragment shader");
 541    if (enabled[1])
 542       ctx_disassemble_program(ctx, ksp[1], "SIMD16 fragment shader");
 543    if (enabled[2])
 544       ctx_disassemble_program(ctx, ksp[2], "SIMD32 fragment shader");
 545 }
 546
 547 static void
 548 decode_3dstate_constant(struct aub_viewer_decode_ctx *ctx,
 549                         struct gen_group *inst,
 550                         const uint32_t *p)
 551 {
 552    struct gen_group *body =
 553       gen_spec_find_struct(ctx->spec, "3DSTATE_CONSTANT_BODY");
 554
 555    uint32_t read_length[4] = {0};
 556    uint64_t read_addr[4];
 557
 558    struct gen_field_iterator outer;
 559    gen_field_iterator_init(&outer, inst, p, 0, false);
 560    while (gen_field_iterator_next(&outer)) {
 561       if (outer.struct_desc != body)
 562          continue;
 563
 564       struct gen_field_iterator iter;
 565       gen_field_iterator_init(&iter, body, &outer.p[outer.start_bit / 32],
 566                               0, false);
 567
 568       while (gen_field_iterator_next(&iter)) {
 569          int idx;
 570          if (sscanf(iter.name, "Read Length[%d]", &idx) == 1) {
 571             read_length[idx] = iter.raw_value;
 572          } else if (sscanf(iter.name, "Buffer[%d]", &idx) == 1) {
 573             read_addr[idx] = iter.raw_value;
 574          }
 575       }
 576
 577       for (int i = 0; i < 4; i++) {
 578          if (read_length[i] == 0)
 579             continue;
 580
 581          struct gen_batch_decode_bo buffer = ctx_get_bo(ctx, read_addr[i]);
 582          if (!buffer.map) {
 583             ImGui::TextColored(ctx->cfg->missing_color,
 584                                "constant buffer %d unavailable addr=0x%012" PRIx64,
 585                                i, read_addr[i]);
 586             continue;
 587          }
 588
 589          unsigned size = read_length[i] * 32;
 590          ImGui::Text("constant buffer %d, size %u", i, size);
 591
 592          if (ctx->edit_address) {
 593             if (ImGui::Button("Show/Edit buffer"))
 594                ctx->edit_address(ctx->user_data, read_addr[i], size);
 595          }
 596       }
 597    }
 598 }
 599
 600 static void
 601 decode_3dstate_binding_table_pointers(struct aub_viewer_decode_ctx *ctx,
 602                                       struct gen_group *inst,
 603                                       const uint32_t *p)
 604 {
 605    dump_binding_table(ctx, p[1], -1);
 606 }
 607
 608 static void
 609 decode_3dstate_sampler_state_pointers(struct aub_viewer_decode_ctx *ctx,
 610                                       struct gen_group *inst,
 611                                       const uint32_t *p)
 612 {
 613    dump_samplers(ctx, p[1], -1);
 614 }
 615
 616 static void
 617 decode_3dstate_sampler_state_pointers_gen6(struct aub_viewer_decode_ctx *ctx,
 618                                            struct gen_group *inst,
 619                                            const uint32_t *p)
 620 {
 621    dump_samplers(ctx, p[1], -1);
 622    dump_samplers(ctx, p[2], -1);
 623    dump_samplers(ctx, p[3], -1);
 624 }
 625
 626 static bool
 627 str_ends_with(const char *str, const char *end)
 628 {
 629    int offset = strlen(str) - strlen(end);
 630    if (offset < 0)
 631       return false;
 632
 633    return strcmp(str + offset, end) == 0;
 634 }
 635
 636 static void
 637 decode_dynamic_state_pointers(struct aub_viewer_decode_ctx *ctx,
 638                               struct gen_group *inst, const uint32_t *p,
 639                               const char *struct_type,  int count)
 640 {
 641    uint32_t state_offset = 0;
 642
 643    struct gen_field_iterator iter;
 644    gen_field_iterator_init(&iter, inst, p, 0, false);
 645    while (gen_field_iterator_next(&iter)) {
 646       if (str_ends_with(iter.name, "Pointer")) {
 647          state_offset = iter.raw_value;
 648          break;
 649       }
 650    }
 651
 652    uint64_t state_addr = ctx->dynamic_base + state_offset;
 653    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, state_addr);
 654    const uint8_t *state_map = (const uint8_t *) bo.map;
 655
 656    if (state_map == NULL) {
 657       ImGui::TextColored(ctx->cfg->missing_color,
 658                          "dynamic %s state unavailable addr=0x%012" PRIx64,
 659                          struct_type, state_addr);
 660       return;
 661    }
 662
 663    struct gen_group *state = gen_spec_find_struct(ctx->spec, struct_type);
 664    if (strcmp(struct_type, "BLEND_STATE") == 0) {
 665       /* Blend states are different from the others because they have a header
 666        * struct called BLEND_STATE which is followed by a variable number of
 667        * BLEND_STATE_ENTRY structs.
 668        */
 669       ImGui::Text("%s", struct_type);
 670       aub_viewer_print_group(ctx, state, state_addr, state_map);
 671
 672       state_addr += state->dw_length * 4;
 673       state_map += state->dw_length * 4;
 674
 675       struct_type = "BLEND_STATE_ENTRY";
 676       state = gen_spec_find_struct(ctx->spec, struct_type);
 677    }
 678
 679    for (int i = 0; i < count; i++) {
 680       ImGui::Text("%s %d", struct_type, i);
 681       aub_viewer_print_group(ctx, state, state_addr, state_map);
 682
 683       state_addr += state->dw_length * 4;
 684       state_map += state->dw_length * 4;
 685    }
 686 }
 687
 688 static void
 689 decode_3dstate_viewport_state_pointers_cc(struct aub_viewer_decode_ctx *ctx,
 690                                           struct gen_group *inst,
 691                                           const uint32_t *p)
 692 {
 693    decode_dynamic_state_pointers(ctx, inst, p, "CC_VIEWPORT", 4);
 694 }
 695
 696 static void
 697 decode_3dstate_viewport_state_pointers_sf_clip(struct aub_viewer_decode_ctx *ctx,
 698                                                struct gen_group *inst,
 699                                                const uint32_t *p)
 700 {
 701    decode_dynamic_state_pointers(ctx, inst, p, "SF_CLIP_VIEWPORT", 4);
 702 }
 703
 704 static void
 705 decode_3dstate_blend_state_pointers(struct aub_viewer_decode_ctx *ctx,
 706                                     struct gen_group *inst,
 707                                     const uint32_t *p)
 708 {
 709    decode_dynamic_state_pointers(ctx, inst, p, "BLEND_STATE", 1);
 710 }
 711
 712 static void
 713 decode_3dstate_cc_state_pointers(struct aub_viewer_decode_ctx *ctx,
 714                                  struct gen_group *inst,
 715                                  const uint32_t *p)
 716 {
 717    decode_dynamic_state_pointers(ctx, inst, p, "COLOR_CALC_STATE", 1);
 718 }
 719
 720 static void
 721 decode_3dstate_scissor_state_pointers(struct aub_viewer_decode_ctx *ctx,
 722                                       struct gen_group *inst,
 723                                       const uint32_t *p)
 724 {
 725    decode_dynamic_state_pointers(ctx, inst, p, "SCISSOR_RECT", 1);
 726 }
 727
 728 static void
 729 decode_load_register_imm(struct aub_viewer_decode_ctx *ctx,
 730                          struct gen_group *inst,
 731                          const uint32_t *p)
 732 {
 733    struct gen_group *reg = gen_spec_find_register(ctx->spec, p[1]);
 734
 735    if (reg != NULL &&
 736        ImGui::TreeNodeEx(&p[1], ImGuiTreeNodeFlags_Framed,
 737                          "%s (0x%x) = 0x%x",
 738                          reg->name, reg->register_offset, p[2])) {
 739       aub_viewer_print_group(ctx, reg, reg->register_offset, &p[2]);
 740       ImGui::TreePop();
 741    }
 742 }
 743
 744 static void
 745 decode_3dprimitive(struct aub_viewer_decode_ctx *ctx,
 746                    struct gen_group *inst,
 747                    const uint32_t *p)
 748 {
 749    if (ctx->display_urb) {
 750       if (ImGui::Button("Show URB"))
 751          ctx->display_urb(ctx->user_data, ctx->urb_stages);
 752    }
 753 }
 754
 755 static void
 756 handle_urb(struct aub_viewer_decode_ctx *ctx,
 757            struct gen_group *inst,
 758            const uint32_t *p)
 759 {
 760    struct gen_field_iterator iter;
 761    gen_field_iterator_init(&iter, inst, p, 0, false);
 762    while (gen_field_iterator_next(&iter)) {
 763       if (strstr(iter.name, "URB Starting Address")) {
 764          ctx->urb_stages[ctx->stage].start = iter.raw_value * 8192;
 765       } else if (strstr(iter.name, "URB Entry Allocation Size")) {
 766          ctx->urb_stages[ctx->stage].size = (iter.raw_value + 1) * 64;
 767       } else if (strstr(iter.name, "Number of URB Entries")) {
 768          ctx->urb_stages[ctx->stage].n_entries = iter.raw_value;
 769       }
 770    }
 771
 772    ctx->end_urb_offset = MAX2(ctx->urb_stages[ctx->stage].start +
 773                               ctx->urb_stages[ctx->stage].n_entries *
 774                               ctx->urb_stages[ctx->stage].size,
 775                               ctx->end_urb_offset);
 776 }
 777
 778 static void
 779 handle_urb_read(struct aub_viewer_decode_ctx *ctx,
 780                 struct gen_group *inst,
 781                 const uint32_t *p)
 782 {
 783    struct gen_field_iterator iter;
 784    gen_field_iterator_init(&iter, inst, p, 0, false);
 785    while (gen_field_iterator_next(&iter)) {
 786       /* Workaround the "Force * URB Entry Read Length" fields */
 787       if (iter.end_bit - iter.start_bit < 2)
 788          continue;
 789
 790       if (strstr(iter.name, "URB Entry Read Offset")) {
 791          ctx->urb_stages[ctx->stage].rd_offset = iter.raw_value * 32;
 792       } else if (strstr(iter.name, "URB Entry Read Length")) {
 793          ctx->urb_stages[ctx->stage].rd_length = iter.raw_value * 32;
 794       } else if (strstr(iter.name, "URB Entry Output Read Offset")) {
 795          ctx->urb_stages[ctx->stage].wr_offset = iter.raw_value * 32;
 796       } else if (strstr(iter.name, "URB Entry Output Length")) {
 797          ctx->urb_stages[ctx->stage].wr_length = iter.raw_value * 32;
 798       }
 799    }
 800 }
 801
 802 static void
 803 handle_urb_constant(struct aub_viewer_decode_ctx *ctx,
 804                     struct gen_group *inst,
 805                     const uint32_t *p)
 806 {
 807    struct gen_group *body =
 808       gen_spec_find_struct(ctx->spec, "3DSTATE_CONSTANT_BODY");
 809
 810    struct gen_field_iterator outer;
 811    gen_field_iterator_init(&outer, inst, p, 0, false);
 812    while (gen_field_iterator_next(&outer)) {
 813       if (outer.struct_desc != body)
 814          continue;
 815
 816       struct gen_field_iterator iter;
 817       gen_field_iterator_init(&iter, body, &outer.p[outer.start_bit / 32],
 818                               0, false);
 819
 820       ctx->urb_stages[ctx->stage].const_rd_length = 0;
 821       while (gen_field_iterator_next(&iter)) {
 822          int idx;
 823          if (sscanf(iter.name, "Read Length[%d]", &idx) == 1) {
 824             ctx->urb_stages[ctx->stage].const_rd_length += iter.raw_value * 32;
 825          }
 826       }
 827    }
 828 }
 829
 830 struct custom_decoder {
 831    const char *cmd_name;
 832    void (*decode)(struct aub_viewer_decode_ctx *ctx,
 833                   struct gen_group *inst,
 834                   const uint32_t *p);
 835    enum aub_decode_stage stage;
 836 } display_decoders[] = {
 837    { "STATE_BASE_ADDRESS", handle_state_base_address },
 838    { "MEDIA_INTERFACE_DESCRIPTOR_LOAD", handle_media_interface_descriptor_load },
 839    { "3DSTATE_VERTEX_BUFFERS", handle_3dstate_vertex_buffers },
 840    { "3DSTATE_INDEX_BUFFER", handle_3dstate_index_buffer },
 841    { "3DSTATE_VS", decode_single_ksp, AUB_DECODE_STAGE_VS, },
 842    { "3DSTATE_GS", decode_single_ksp, AUB_DECODE_STAGE_GS, },
 843    { "3DSTATE_DS", decode_single_ksp, AUB_DECODE_STAGE_DS, },
 844    { "3DSTATE_HS", decode_single_ksp, AUB_DECODE_STAGE_HS, },
 845    { "3DSTATE_PS", decode_ps_kernels, AUB_DECODE_STAGE_PS, },
 846    { "3DSTATE_CONSTANT_VS", decode_3dstate_constant, AUB_DECODE_STAGE_VS, },
 847    { "3DSTATE_CONSTANT_GS", decode_3dstate_constant, AUB_DECODE_STAGE_GS, },
 848    { "3DSTATE_CONSTANT_DS", decode_3dstate_constant, AUB_DECODE_STAGE_DS, },
 849    { "3DSTATE_CONSTANT_HS", decode_3dstate_constant, AUB_DECODE_STAGE_HS, },
 850    { "3DSTATE_CONSTANT_PS", decode_3dstate_constant, AUB_DECODE_STAGE_PS, },
 851
 852    { "3DSTATE_BINDING_TABLE_POINTERS_VS", decode_3dstate_binding_table_pointers, AUB_DECODE_STAGE_VS, },
 853    { "3DSTATE_BINDING_TABLE_POINTERS_GS", decode_3dstate_binding_table_pointers, AUB_DECODE_STAGE_GS, },
 854    { "3DSTATE_BINDING_TABLE_POINTERS_HS", decode_3dstate_binding_table_pointers, AUB_DECODE_STAGE_HS, },
 855    { "3DSTATE_BINDING_TABLE_POINTERS_DS", decode_3dstate_binding_table_pointers, AUB_DECODE_STAGE_DS, },
 856    { "3DSTATE_BINDING_TABLE_POINTERS_PS", decode_3dstate_binding_table_pointers, AUB_DECODE_STAGE_PS, },
 857
 858    { "3DSTATE_SAMPLER_STATE_POINTERS_VS", decode_3dstate_sampler_state_pointers, AUB_DECODE_STAGE_VS, },
 859    { "3DSTATE_SAMPLER_STATE_POINTERS_GS", decode_3dstate_sampler_state_pointers, AUB_DECODE_STAGE_GS, },
 860    { "3DSTATE_SAMPLER_STATE_POINTERS_DS", decode_3dstate_sampler_state_pointers, AUB_DECODE_STAGE_DS, },
 861    { "3DSTATE_SAMPLER_STATE_POINTERS_HS", decode_3dstate_sampler_state_pointers, AUB_DECODE_STAGE_HS, },
 862    { "3DSTATE_SAMPLER_STATE_POINTERS_PS", decode_3dstate_sampler_state_pointers, AUB_DECODE_STAGE_PS, },
 863    { "3DSTATE_SAMPLER_STATE_POINTERS", decode_3dstate_sampler_state_pointers_gen6 },
 864
 865    { "3DSTATE_VIEWPORT_STATE_POINTERS_CC", decode_3dstate_viewport_state_pointers_cc },
 866    { "3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP", decode_3dstate_viewport_state_pointers_sf_clip },
 867    { "3DSTATE_BLEND_STATE_POINTERS", decode_3dstate_blend_state_pointers },
 868    { "3DSTATE_CC_STATE_POINTERS", decode_3dstate_cc_state_pointers },
 869    { "3DSTATE_SCISSOR_STATE_POINTERS", decode_3dstate_scissor_state_pointers },
 870    { "MI_LOAD_REGISTER_IMM", decode_load_register_imm },
 871    { "3DPRIMITIVE", decode_3dprimitive },
 872 };
 873
 874 struct custom_decoder info_decoders[] = {
 875    { "STATE_BASE_ADDRESS", handle_state_base_address },
 876    { "3DSTATE_URB_VS", handle_urb, AUB_DECODE_STAGE_VS, },
 877    { "3DSTATE_URB_GS", handle_urb, AUB_DECODE_STAGE_GS, },
 878    { "3DSTATE_URB_DS", handle_urb, AUB_DECODE_STAGE_DS, },
 879    { "3DSTATE_URB_HS", handle_urb, AUB_DECODE_STAGE_HS, },
 880    { "3DSTATE_VS", handle_urb_read, AUB_DECODE_STAGE_VS, },
 881    { "3DSTATE_GS", handle_urb_read, AUB_DECODE_STAGE_GS, },
 882    { "3DSTATE_DS", handle_urb_read, AUB_DECODE_STAGE_DS, },
 883    { "3DSTATE_HS", handle_urb_read, AUB_DECODE_STAGE_HS, },
 884    { "3DSTATE_PS", handle_urb_read, AUB_DECODE_STAGE_PS, },
 885    { "3DSTATE_CONSTANT_VS", handle_urb_constant, AUB_DECODE_STAGE_VS, },
 886    { "3DSTATE_CONSTANT_GS", handle_urb_constant, AUB_DECODE_STAGE_GS, },
 887    { "3DSTATE_CONSTANT_DS", handle_urb_constant, AUB_DECODE_STAGE_DS, },
 888    { "3DSTATE_CONSTANT_HS", handle_urb_constant, AUB_DECODE_STAGE_HS, },
 889    { "3DSTATE_CONSTANT_PS", handle_urb_constant, AUB_DECODE_STAGE_PS, },
 890 };
 891
 892 void
 893 aub_viewer_render_batch(struct aub_viewer_decode_ctx *ctx,
 894                         const void *_batch, uint32_t batch_size,
 895                         uint64_t batch_addr)
 896 {
 897    struct gen_group *inst;
 898    const uint32_t *p, *batch = (const uint32_t *) _batch, *end = batch + batch_size / sizeof(uint32_t);
 899    int length;
 900
 901    for (p = batch; p < end; p += length) {
 902       inst = gen_spec_find_instruction(ctx->spec, ctx->engine, p);
 903       length = gen_group_get_length(inst, p);
 904       assert(inst == NULL || length > 0);
 905       length = MAX2(1, length);
 906
 907       uint64_t offset = batch_addr + ((char *)p - (char *)batch);
 908
 909       if (inst == NULL) {
 910          ImGui::TextColored(ctx->cfg->error_color,
 911                             "0x%012" PRIx64 ": unknown instruction %012x",
 912                             offset, p[0]);
 913          continue;
 914       }
 915
 916       const char *inst_name = gen_group_get_name(inst);
 917
 918       for (unsigned i = 0; i < ARRAY_SIZE(info_decoders); i++) {
 919          if (strcmp(inst_name, info_decoders[i].cmd_name) == 0) {
 920             ctx->stage = info_decoders[i].stage;
 921             info_decoders[i].decode(ctx, inst, p);
 922             break;
 923          }
 924       }
 925
 926       if (ctx->decode_cfg->command_filter.PassFilter(inst->name) &&
 927           ImGui::TreeNodeEx(p,
 928                             ImGuiTreeNodeFlags_Framed,
 929                             "0x%012" PRIx64 ":  %s",
 930                             offset, inst->name)) {
 931          aub_viewer_print_group(ctx, inst, offset, p);
 932
 933          for (unsigned i = 0; i < ARRAY_SIZE(display_decoders); i++) {
 934             if (strcmp(inst_name, display_decoders[i].cmd_name) == 0) {
 935                ctx->stage = display_decoders[i].stage;
 936                display_decoders[i].decode(ctx, inst, p);
 937                break;
 938             }
 939          }
 940
 941          if (ctx->edit_address) {
 942             if (ImGui::Button("Edit instruction"))
 943                ctx->edit_address(ctx->user_data, offset, length * 4);
 944          }
 945
 946          ImGui::TreePop();
 947       }
 948
 949       if (strcmp(inst_name, "MI_BATCH_BUFFER_START") == 0) {
 950          struct gen_batch_decode_bo next_batch = {};
 951          bool second_level;
 952          struct gen_field_iterator iter;
 953          gen_field_iterator_init(&iter, inst, p, 0, false);
 954          while (gen_field_iterator_next(&iter)) {
 955             if (strcmp(iter.name, "Batch Buffer Start Address") == 0) {
 956                next_batch = ctx_get_bo(ctx, iter.raw_value);
 957             } else if (strcmp(iter.name, "Second Level Batch Buffer") == 0) {
 958                second_level = iter.raw_value;
 959             }
 960          }
 961
 962          if (next_batch.map == NULL) {
 963             ImGui::TextColored(ctx->cfg->missing_color,
 964                                "Secondary batch at 0x%012" PRIx64 " unavailable",
 965                                next_batch.addr);
 966          } else {
 967             aub_viewer_render_batch(ctx, next_batch.map, next_batch.size,
 968                                     next_batch.addr);
 969          }
 970          if (second_level) {
 971             /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" set acts
 972              * like a subroutine call.  Commands that come afterwards get
 973              * processed once the 2nd level batch buffer returns with
 974              * MI_BATCH_BUFFER_END.
 975              */
 976             continue;
 977          } else {
 978             /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" unset acts
 979              * like a goto.  Nothing after it will ever get processed.  In
 980              * order to prevent the recursion from growing, we just reset the
 981              * loop and continue;
 982              */
 983             break;
 984          }
 985       } else if (strcmp(inst_name, "MI_BATCH_BUFFER_END") == 0) {
 986          break;
 987       }
 988    }
 989 }