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
  46    ctx->cfg = cfg;
  47    ctx->decode_cfg = decode_cfg;
  48    ctx->spec = spec;
  49    ctx->disasm = disasm;
  50 }
  51
  52 static void
  53 aub_viewer_print_group(struct aub_viewer_decode_ctx *ctx,
  54                        struct gen_group *group,
  55                        uint64_t address, const void *map)
  56 {
  57    struct gen_field_iterator iter;
  58    int last_dword = -1;
  59    const uint32_t *p = (const uint32_t *) map;
  60
  61    gen_field_iterator_init(&iter, group, p, 0, false);
  62    while (gen_field_iterator_next(&iter)) {
  63       if (ctx->decode_cfg->show_dwords) {
  64          int iter_dword = iter.end_bit / 32;
  65          if (last_dword != iter_dword) {
  66             for (int i = last_dword + 1; i <= iter_dword; i++) {
  67                ImGui::TextColored(ctx->cfg->dwords_color,
  68                                   "0x%08" PRIx64 ":  0x%08x : Dword %d",
  69                                   address + 4 * i, iter.p[i], i);
  70             }
  71             last_dword = iter_dword;
  72          }
  73       }
  74       if (!gen_field_is_header(iter.field)) {
  75          if (ctx->decode_cfg->field_filter.PassFilter(iter.name)) {
  76             ImGui::Text("%s: %s", iter.name, iter.value);
  77             if (iter.struct_desc) {
  78                int struct_dword = iter.start_bit / 32;
  79                uint64_t struct_address = address + 4 * struct_dword;
  80                aub_viewer_print_group(ctx, iter.struct_desc, struct_address,
  81                                       &p[struct_dword]);
  82             }
  83          }
  84       }
  85    }
  86 }
  87
  88 static struct gen_batch_decode_bo
  89 ctx_get_bo(struct aub_viewer_decode_ctx *ctx, uint64_t addr)
  90 {
  91    if (gen_spec_get_gen(ctx->spec) >= gen_make_gen(8,0)) {
  92       /* On Broadwell and above, we have 48-bit addresses which consume two
  93        * dwords.  Some packets require that these get stored in a "canonical
  94        * form" which means that bit 47 is sign-extended through the upper
  95        * bits. In order to correctly handle those aub dumps, we need to mask
  96        * off the top 16 bits.
  97        */
  98       addr &= (~0ull >> 16);
  99    }
 100
 101    struct gen_batch_decode_bo bo = ctx->get_bo(ctx->user_data, addr);
 102
 103    if (gen_spec_get_gen(ctx->spec) >= gen_make_gen(8,0))
 104       bo.addr &= (~0ull >> 16);
 105
 106    /* We may actually have an offset into the bo */
 107    if (bo.map != NULL) {
 108       assert(bo.addr <= addr);
 109       uint64_t offset = addr - bo.addr;
 110       bo.map = (const uint8_t *)bo.map + offset;
 111       bo.addr += offset;
 112       bo.size -= offset;
 113    }
 114
 115    return bo;
 116 }
 117
 118 static int
 119 update_count(struct aub_viewer_decode_ctx *ctx,
 120              uint32_t offset_from_dsba,
 121              unsigned element_dwords,
 122              unsigned guess)
 123 {
 124    unsigned size = 0;
 125
 126    if (ctx->get_state_size)
 127       size = ctx->get_state_size(ctx->user_data, offset_from_dsba);
 128
 129    if (size > 0)
 130       return size / (sizeof(uint32_t) * element_dwords);
 131
 132    /* In the absence of any information, just guess arbitrarily. */
 133    return guess;
 134 }
 135
 136 static void
 137 ctx_disassemble_program(struct aub_viewer_decode_ctx *ctx,
 138                         uint32_t ksp, const char *type)
 139 {
 140    uint64_t addr = ctx->instruction_base + ksp;
 141    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, addr);
 142    if (!bo.map) {
 143       ImGui::TextColored(ctx->cfg->missing_color, "Shader unavailable");
 144       return;
 145    }
 146
 147    ImGui::PushID(addr);
 148    if (ImGui::Button(type) && ctx->display_shader)
 149       ctx->display_shader(ctx->user_data, type, addr);
 150    ImGui::PopID();
 151 }
 152
 153 static void
 154 handle_state_base_address(struct aub_viewer_decode_ctx *ctx,
 155                           struct gen_group *inst,
 156                           const uint32_t *p)
 157 {
 158    struct gen_field_iterator iter;
 159    gen_field_iterator_init(&iter, inst, p, 0, false);
 160
 161    uint64_t surface_base = 0, dynamic_base = 0, instruction_base = 0;
 162    bool surface_modify = 0, dynamic_modify = 0, instruction_modify = 0;
 163
 164    while (gen_field_iterator_next(&iter)) {
 165       if (strcmp(iter.name, "Surface State Base Address") == 0) {
 166          surface_base = iter.raw_value;
 167       } else if (strcmp(iter.name, "Dynamic State Base Address") == 0) {
 168          dynamic_base = iter.raw_value;
 169       } else if (strcmp(iter.name, "Instruction Base Address") == 0) {
 170          instruction_base = iter.raw_value;
 171       } else if (strcmp(iter.name, "Surface State Base Address Modify Enable") == 0) {
 172          surface_modify = iter.raw_value;
 173       } else if (strcmp(iter.name, "Dynamic State Base Address Modify Enable") == 0) {
 174          dynamic_modify = iter.raw_value;
 175       } else if (strcmp(iter.name, "Instruction Base Address Modify Enable") == 0) {
 176          instruction_modify = iter.raw_value;
 177       }
 178    }
 179
 180    if (dynamic_modify)
 181       ctx->dynamic_base = dynamic_base;
 182
 183    if (surface_modify)
 184       ctx->surface_base = surface_base;
 185
 186    if (instruction_modify)
 187       ctx->instruction_base = instruction_base;
 188 }
 189
 190 static void
 191 dump_binding_table(struct aub_viewer_decode_ctx *ctx, uint32_t offset, int count)
 192 {
 193    struct gen_group *strct =
 194       gen_spec_find_struct(ctx->spec, "RENDER_SURFACE_STATE");
 195    if (strct == NULL) {
 196       ImGui::TextColored(ctx->cfg->missing_color, "did not find RENDER_SURFACE_STATE info");
 197       return;
 198    }
 199
 200    if (count < 0)
 201       count = update_count(ctx, offset, 1, 8);
 202
 203    if (offset % 32 != 0 || offset >= UINT16_MAX) {
 204       ImGui::TextColored(ctx->cfg->missing_color, "invalid binding table pointer");
 205       return;
 206    }
 207
 208    struct gen_batch_decode_bo bind_bo =
 209       ctx_get_bo(ctx, ctx->surface_base + offset);
 210
 211    if (bind_bo.map == NULL) {
 212       ImGui::TextColored(ctx->cfg->missing_color,
 213                          "binding table unavailable addr=0x%08" PRIx64,
 214                          ctx->surface_base + offset);
 215       return;
 216    }
 217
 218    const uint32_t *pointers = (const uint32_t *) bind_bo.map;
 219    for (int i = 0; i < count; i++) {
 220       if (pointers[i] == 0)
 221          continue;
 222
 223       uint64_t addr = ctx->surface_base + pointers[i];
 224       struct gen_batch_decode_bo bo = ctx_get_bo(ctx, addr);
 225       uint32_t size = strct->dw_length * 4;
 226
 227       if (pointers[i] % 32 != 0 ||
 228           addr < bo.addr || addr + size >= bo.addr + bo.size) {
 229          ImGui::TextColored(ctx->cfg->missing_color,
 230                             "pointer %u: %08x <not valid>", i, pointers[i]);
 231          continue;
 232       }
 233
 234       ImGui::Text("pointer %u: %08x", i, pointers[i]);
 235       aub_viewer_print_group(ctx, strct, addr, (const uint8_t *) bo.map + (addr - bo.addr));
 236    }
 237 }
 238
 239 static void
 240 dump_samplers(struct aub_viewer_decode_ctx *ctx, uint32_t offset, int count)
 241 {
 242    struct gen_group *strct = gen_spec_find_struct(ctx->spec, "SAMPLER_STATE");
 243
 244    if (count < 0)
 245       count = update_count(ctx, offset, strct->dw_length, 4);
 246
 247    uint64_t state_addr = ctx->dynamic_base + offset;
 248    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, state_addr);
 249    const uint8_t *state_map = (const uint8_t *) bo.map;
 250
 251    if (state_map == NULL) {
 252       ImGui::TextColored(ctx->cfg->missing_color,
 253                          "samplers unavailable addr=0x%08" PRIx64, state_addr);
 254       return;
 255    }
 256
 257    if (offset % 32 != 0 || state_addr - bo.addr >= bo.size) {
 258       ImGui::TextColored(ctx->cfg->missing_color, "invalid sampler state pointer");
 259       return;
 260    }
 261
 262    for (int i = 0; i < count; i++) {
 263       ImGui::Text("sampler state %d", i);
 264       aub_viewer_print_group(ctx, strct, state_addr, state_map);
 265       state_addr += 16;
 266       state_map += 16;
 267    }
 268 }
 269
 270 static void
 271 handle_media_interface_descriptor_load(struct aub_viewer_decode_ctx *ctx,
 272                                        struct gen_group *inst,
 273                                        const uint32_t *p)
 274 {
 275    struct gen_group *desc =
 276       gen_spec_find_struct(ctx->spec, "INTERFACE_DESCRIPTOR_DATA");
 277
 278    struct gen_field_iterator iter;
 279    gen_field_iterator_init(&iter, inst, p, 0, false);
 280    uint32_t descriptor_offset = 0;
 281    int descriptor_count = 0;
 282    while (gen_field_iterator_next(&iter)) {
 283       if (strcmp(iter.name, "Interface Descriptor Data Start Address") == 0) {
 284          descriptor_offset = strtol(iter.value, NULL, 16);
 285       } else if (strcmp(iter.name, "Interface Descriptor Total Length") == 0) {
 286          descriptor_count =
 287             strtol(iter.value, NULL, 16) / (desc->dw_length * 4);
 288       }
 289    }
 290
 291    uint64_t desc_addr = ctx->dynamic_base + descriptor_offset;
 292    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, desc_addr);
 293    const uint32_t *desc_map = (const uint32_t *) bo.map;
 294
 295    if (desc_map == NULL) {
 296       ImGui::TextColored(ctx->cfg->missing_color,
 297                          "interface descriptors unavailable addr=0x%08" PRIx64, desc_addr);
 298       return;
 299    }
 300
 301    for (int i = 0; i < descriptor_count; i++) {
 302       ImGui::Text("descriptor %d: %08x", i, descriptor_offset);
 303
 304       aub_viewer_print_group(ctx, desc, desc_addr, desc_map);
 305
 306       gen_field_iterator_init(&iter, desc, desc_map, 0, false);
 307       uint64_t ksp = 0;
 308       uint32_t sampler_offset = 0, sampler_count = 0;
 309       uint32_t binding_table_offset = 0, binding_entry_count = 0;
 310       while (gen_field_iterator_next(&iter)) {
 311          if (strcmp(iter.name, "Kernel Start Pointer") == 0) {
 312             ksp = strtoll(iter.value, NULL, 16);
 313          } else if (strcmp(iter.name, "Sampler State Pointer") == 0) {
 314             sampler_offset = strtol(iter.value, NULL, 16);
 315          } else if (strcmp(iter.name, "Sampler Count") == 0) {
 316             sampler_count = strtol(iter.value, NULL, 10);
 317          } else if (strcmp(iter.name, "Binding Table Pointer") == 0) {
 318             binding_table_offset = strtol(iter.value, NULL, 16);
 319          } else if (strcmp(iter.name, "Binding Table Entry Count") == 0) {
 320             binding_entry_count = strtol(iter.value, NULL, 10);
 321          }
 322       }
 323
 324       ctx_disassemble_program(ctx, ksp, "compute shader");
 325
 326       dump_samplers(ctx, sampler_offset, sampler_count);
 327       dump_binding_table(ctx, binding_table_offset, binding_entry_count);
 328
 329       desc_map += desc->dw_length;
 330       desc_addr += desc->dw_length * 4;
 331    }
 332 }
 333
 334 static void
 335 handle_3dstate_vertex_buffers(struct aub_viewer_decode_ctx *ctx,
 336                               struct gen_group *inst,
 337                               const uint32_t *p)
 338 {
 339    struct gen_group *vbs = gen_spec_find_struct(ctx->spec, "VERTEX_BUFFER_STATE");
 340
 341    struct gen_batch_decode_bo vb = {};
 342    uint32_t vb_size = 0;
 343    int index = -1;
 344    int pitch = -1;
 345    bool ready = false;
 346
 347    struct gen_field_iterator iter;
 348    gen_field_iterator_init(&iter, inst, p, 0, false);
 349    while (gen_field_iterator_next(&iter)) {
 350       if (iter.struct_desc != vbs)
 351          continue;
 352
 353       uint64_t buffer_addr;
 354
 355       struct gen_field_iterator vbs_iter;
 356       gen_field_iterator_init(&vbs_iter, vbs, &iter.p[iter.start_bit / 32], 0, false);
 357       while (gen_field_iterator_next(&vbs_iter)) {
 358          if (strcmp(vbs_iter.name, "Vertex Buffer Index") == 0) {
 359             index = vbs_iter.raw_value;
 360          } else if (strcmp(vbs_iter.name, "Buffer Pitch") == 0) {
 361             pitch = vbs_iter.raw_value;
 362          } else if (strcmp(vbs_iter.name, "Buffer Starting Address") == 0) {
 363             buffer_addr = vbs_iter.raw_value;
 364             vb = ctx_get_bo(ctx, buffer_addr);
 365          } else if (strcmp(vbs_iter.name, "Buffer Size") == 0) {
 366             vb_size = vbs_iter.raw_value;
 367             ready = true;
 368          } else if (strcmp(vbs_iter.name, "End Address") == 0) {
 369             if (vb.map && vbs_iter.raw_value >= vb.addr)
 370                vb_size = vbs_iter.raw_value - vb.addr;
 371             else
 372                vb_size = 0;
 373             ready = true;
 374          }
 375
 376          if (!ready)
 377             continue;
 378
 379          ImGui::Text("vertex buffer %d, size %d, pitch %d", index, vb_size, pitch);
 380
 381          if (vb.map == NULL) {
 382             ImGui::TextColored(ctx->cfg->missing_color,
 383                                "buffer contents unavailable addr=0x%08" PRIx64, buffer_addr);
 384             continue;
 385          }
 386
 387          if (vb.map == 0 || vb_size == 0)
 388             continue;
 389
 390          vb.map = NULL;
 391          vb_size = 0;
 392          index = -1;
 393          pitch = -1;
 394          ready = false;
 395       }
 396    }
 397 }
 398
 399 static void
 400 handle_3dstate_index_buffer(struct aub_viewer_decode_ctx *ctx,
 401                             struct gen_group *inst,
 402                             const uint32_t *p)
 403 {
 404    struct gen_batch_decode_bo ib = {};
 405    uint64_t buffer_addr = 0;
 406    uint32_t ib_size = 0;
 407    uint32_t format = 0;
 408
 409    struct gen_field_iterator iter;
 410    gen_field_iterator_init(&iter, inst, p, 0, false);
 411    while (gen_field_iterator_next(&iter)) {
 412       if (strcmp(iter.name, "Index Format") == 0) {
 413          format = iter.raw_value;
 414       } else if (strcmp(iter.name, "Buffer Starting Address") == 0) {
 415          buffer_addr = iter.raw_value;
 416          ib = ctx_get_bo(ctx, buffer_addr);
 417       } else if (strcmp(iter.name, "Buffer Size") == 0) {
 418          ib_size = iter.raw_value;
 419       }
 420    }
 421
 422    if (ib.map == NULL) {
 423       ImGui::TextColored(ctx->cfg->missing_color,
 424                          "buffer contents unavailable addr=0x%08" PRIx64,
 425                          buffer_addr);
 426       return;
 427    }
 428
 429    const uint8_t *m = (const uint8_t *) ib.map;
 430    const uint8_t *ib_end = m + MIN2(ib.size, ib_size);
 431    for (int i = 0; m < ib_end && i < 10; i++) {
 432       switch (format) {
 433       case 0:
 434          m += 1;
 435          break;
 436       case 1:
 437          m += 2;
 438          break;
 439       case 2:
 440          m += 4;
 441          break;
 442       }
 443    }
 444 }
 445
 446 static void
 447 decode_single_ksp(struct aub_viewer_decode_ctx *ctx,
 448                   struct gen_group *inst,
 449                   const uint32_t *p)
 450 {
 451    uint64_t ksp = 0;
 452    bool is_simd8 = false; /* vertex shaders on Gen8+ only */
 453    bool is_enabled = true;
 454
 455    struct gen_field_iterator iter;
 456    gen_field_iterator_init(&iter, inst, p, 0, false);
 457    while (gen_field_iterator_next(&iter)) {
 458       if (strcmp(iter.name, "Kernel Start Pointer") == 0) {
 459          ksp = iter.raw_value;
 460       } else if (strcmp(iter.name, "SIMD8 Dispatch Enable") == 0) {
 461          is_simd8 = iter.raw_value;
 462       } else if (strcmp(iter.name, "Dispatch Mode") == 0) {
 463          is_simd8 = strcmp(iter.value, "SIMD8") == 0;
 464       } else if (strcmp(iter.name, "Dispatch Enable") == 0) {
 465          is_simd8 = strcmp(iter.value, "SIMD8") == 0;
 466       } else if (strcmp(iter.name, "Enable") == 0) {
 467          is_enabled = iter.raw_value;
 468       }
 469    }
 470
 471    const char *type =
 472       strcmp(inst->name,   "VS_STATE") == 0 ? "vertex shader" :
 473       strcmp(inst->name,   "GS_STATE") == 0 ? "geometry shader" :
 474       strcmp(inst->name,   "SF_STATE") == 0 ? "strips and fans shader" :
 475       strcmp(inst->name, "CLIP_STATE") == 0 ? "clip shader" :
 476       strcmp(inst->name, "3DSTATE_DS") == 0 ? "tessellation evaluation shader" :
 477       strcmp(inst->name, "3DSTATE_HS") == 0 ? "tessellation control shader" :
 478       strcmp(inst->name, "3DSTATE_VS") == 0 ? (is_simd8 ? "SIMD8 vertex shader" : "vec4 vertex shader") :
 479       strcmp(inst->name, "3DSTATE_GS") == 0 ? (is_simd8 ? "SIMD8 geometry shader" : "vec4 geometry shader") :
 480       NULL;
 481
 482    if (is_enabled)
 483       ctx_disassemble_program(ctx, ksp, type);
 484 }
 485
 486 static void
 487 decode_ps_kernels(struct aub_viewer_decode_ctx *ctx,
 488                   struct gen_group *inst,
 489                   const uint32_t *p)
 490 {
 491    uint64_t ksp[3] = {0, 0, 0};
 492    bool enabled[3] = {false, false, false};
 493
 494    struct gen_field_iterator iter;
 495    gen_field_iterator_init(&iter, inst, p, 0, false);
 496    while (gen_field_iterator_next(&iter)) {
 497       if (strncmp(iter.name, "Kernel Start Pointer ",
 498                   strlen("Kernel Start Pointer ")) == 0) {
 499          int idx = iter.name[strlen("Kernel Start Pointer ")] - '0';
 500          ksp[idx] = strtol(iter.value, NULL, 16);
 501       } else if (strcmp(iter.name, "8 Pixel Dispatch Enable") == 0) {
 502          enabled[0] = strcmp(iter.value, "true") == 0;
 503       } else if (strcmp(iter.name, "16 Pixel Dispatch Enable") == 0) {
 504          enabled[1] = strcmp(iter.value, "true") == 0;
 505       } else if (strcmp(iter.name, "32 Pixel Dispatch Enable") == 0) {
 506          enabled[2] = strcmp(iter.value, "true") == 0;
 507       }
 508    }
 509
 510    /* Reorder KSPs to be [8, 16, 32] instead of the hardware order. */
 511    if (enabled[0] + enabled[1] + enabled[2] == 1) {
 512       if (enabled[1]) {
 513          ksp[1] = ksp[0];
 514          ksp[0] = 0;
 515       } else if (enabled[2]) {
 516          ksp[2] = ksp[0];
 517          ksp[0] = 0;
 518       }
 519    } else {
 520       uint64_t tmp = ksp[1];
 521       ksp[1] = ksp[2];
 522       ksp[2] = tmp;
 523    }
 524
 525    if (enabled[0])
 526       ctx_disassemble_program(ctx, ksp[0], "SIMD8 fragment shader");
 527    if (enabled[1])
 528       ctx_disassemble_program(ctx, ksp[1], "SIMD16 fragment shader");
 529    if (enabled[2])
 530       ctx_disassemble_program(ctx, ksp[2], "SIMD32 fragment shader");
 531 }
 532
 533 static void
 534 decode_3dstate_constant(struct aub_viewer_decode_ctx *ctx,
 535                         struct gen_group *inst,
 536                         const uint32_t *p)
 537 {
 538    struct gen_group *body =
 539       gen_spec_find_struct(ctx->spec, "3DSTATE_CONSTANT_BODY");
 540
 541    uint32_t read_length[4] = {0};
 542    uint64_t read_addr[4];
 543
 544    struct gen_field_iterator outer;
 545    gen_field_iterator_init(&outer, inst, p, 0, false);
 546    while (gen_field_iterator_next(&outer)) {
 547       if (outer.struct_desc != body)
 548          continue;
 549
 550       struct gen_field_iterator iter;
 551       gen_field_iterator_init(&iter, body, &outer.p[outer.start_bit / 32],
 552                               0, false);
 553
 554       while (gen_field_iterator_next(&iter)) {
 555          int idx;
 556          if (sscanf(iter.name, "Read Length[%d]", &idx) == 1) {
 557             read_length[idx] = iter.raw_value;
 558          } else if (sscanf(iter.name, "Buffer[%d]", &idx) == 1) {
 559             read_addr[idx] = iter.raw_value;
 560          }
 561       }
 562
 563       for (int i = 0; i < 4; i++) {
 564          if (read_length[i] == 0)
 565             continue;
 566
 567          struct gen_batch_decode_bo buffer = ctx_get_bo(ctx, read_addr[i]);
 568          if (!buffer.map) {
 569             ImGui::TextColored(ctx->cfg->missing_color,
 570                                "constant buffer %d unavailable addr=0x%08" PRIx64,
 571                                i, read_addr[i]);
 572             continue;
 573          }
 574
 575          unsigned size = read_length[i] * 32;
 576          ImGui::Text("constant buffer %d, size %u", i, size);
 577
 578          if (ctx->edit_address) {
 579             if (ImGui::Button("Show/Edit buffer"))
 580                ctx->edit_address(ctx->user_data, read_addr[i], size);
 581          }
 582       }
 583    }
 584 }
 585
 586 static void
 587 decode_3dstate_binding_table_pointers(struct aub_viewer_decode_ctx *ctx,
 588                                       struct gen_group *inst,
 589                                       const uint32_t *p)
 590 {
 591    dump_binding_table(ctx, p[1], -1);
 592 }
 593
 594 static void
 595 decode_3dstate_sampler_state_pointers(struct aub_viewer_decode_ctx *ctx,
 596                                       struct gen_group *inst,
 597                                       const uint32_t *p)
 598 {
 599    dump_samplers(ctx, p[1], -1);
 600 }
 601
 602 static void
 603 decode_3dstate_sampler_state_pointers_gen6(struct aub_viewer_decode_ctx *ctx,
 604                                            struct gen_group *inst,
 605                                            const uint32_t *p)
 606 {
 607    dump_samplers(ctx, p[1], -1);
 608    dump_samplers(ctx, p[2], -1);
 609    dump_samplers(ctx, p[3], -1);
 610 }
 611
 612 static bool
 613 str_ends_with(const char *str, const char *end)
 614 {
 615    int offset = strlen(str) - strlen(end);
 616    if (offset < 0)
 617       return false;
 618
 619    return strcmp(str + offset, end) == 0;
 620 }
 621
 622 static void
 623 decode_dynamic_state_pointers(struct aub_viewer_decode_ctx *ctx,
 624                               struct gen_group *inst, const uint32_t *p,
 625                               const char *struct_type,  int count)
 626 {
 627    struct gen_group *state = gen_spec_find_struct(ctx->spec, struct_type);
 628
 629    uint32_t state_offset = 0;
 630
 631    struct gen_field_iterator iter;
 632    gen_field_iterator_init(&iter, inst, p, 0, false);
 633    while (gen_field_iterator_next(&iter)) {
 634       if (str_ends_with(iter.name, "Pointer")) {
 635          state_offset = iter.raw_value;
 636          break;
 637       }
 638    }
 639
 640    uint64_t state_addr = ctx->dynamic_base + state_offset;
 641    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, state_addr);
 642    const uint8_t *state_map = (const uint8_t *) bo.map;
 643
 644    if (state_map == NULL) {
 645       ImGui::TextColored(ctx->cfg->missing_color,
 646                          "dynamic %s state unavailable addr=0x%08" PRIx64,
 647                          struct_type, state_addr);
 648       return;
 649    }
 650
 651    for (int i = 0; i < count; i++) {
 652       ImGui::Text("%s %d", struct_type, i);
 653       aub_viewer_print_group(ctx, state, state_offset, state_map);
 654
 655       state_addr += state->dw_length * 4;
 656       state_map += state->dw_length;
 657    }
 658 }
 659
 660 static void
 661 decode_3dstate_viewport_state_pointers_cc(struct aub_viewer_decode_ctx *ctx,
 662                                           struct gen_group *inst,
 663                                           const uint32_t *p)
 664 {
 665    decode_dynamic_state_pointers(ctx, inst, p, "CC_VIEWPORT", 4);
 666 }
 667
 668 static void
 669 decode_3dstate_viewport_state_pointers_sf_clip(struct aub_viewer_decode_ctx *ctx,
 670                                                struct gen_group *inst,
 671                                                const uint32_t *p)
 672 {
 673    decode_dynamic_state_pointers(ctx, inst, p, "SF_CLIP_VIEWPORT", 4);
 674 }
 675
 676 static void
 677 decode_3dstate_blend_state_pointers(struct aub_viewer_decode_ctx *ctx,
 678                                     struct gen_group *inst,
 679                                     const uint32_t *p)
 680 {
 681    decode_dynamic_state_pointers(ctx, inst, p, "BLEND_STATE", 1);
 682 }
 683
 684 static void
 685 decode_3dstate_cc_state_pointers(struct aub_viewer_decode_ctx *ctx,
 686                                  struct gen_group *inst,
 687                                  const uint32_t *p)
 688 {
 689    decode_dynamic_state_pointers(ctx, inst, p, "COLOR_CALC_STATE", 1);
 690 }
 691
 692 static void
 693 decode_3dstate_scissor_state_pointers(struct aub_viewer_decode_ctx *ctx,
 694                                       struct gen_group *inst,
 695                                       const uint32_t *p)
 696 {
 697    decode_dynamic_state_pointers(ctx, inst, p, "SCISSOR_RECT", 1);
 698 }
 699
 700 static void
 701 decode_load_register_imm(struct aub_viewer_decode_ctx *ctx,
 702                          struct gen_group *inst,
 703                          const uint32_t *p)
 704 {
 705    struct gen_group *reg = gen_spec_find_register(ctx->spec, p[1]);
 706
 707    if (reg != NULL &&
 708        ImGui::TreeNodeEx(&p[1], ImGuiTreeNodeFlags_Framed,
 709                          "%s (0x%x) = 0x%x",
 710                          reg->name, reg->register_offset, p[2])) {
 711       aub_viewer_print_group(ctx, reg, reg->register_offset, &p[2]);
 712       ImGui::TreePop();
 713    }
 714 }
 715
 716 static void
 717 decode_3dprimitive(struct aub_viewer_decode_ctx *ctx,
 718                    struct gen_group *inst,
 719                    const uint32_t *p)
 720 {
 721    if (ctx->display_urb) {
 722       if (ImGui::Button("Show URB"))
 723          ctx->display_urb(ctx->user_data, ctx->urb_stages);
 724    }
 725 }
 726
 727 static void
 728 handle_urb(struct aub_viewer_decode_ctx *ctx,
 729            struct gen_group *inst,
 730            const uint32_t *p)
 731 {
 732    struct gen_field_iterator iter;
 733    gen_field_iterator_init(&iter, inst, p, 0, false);
 734    while (gen_field_iterator_next(&iter)) {
 735       if (strstr(iter.name, "URB Starting Address")) {
 736          ctx->urb_stages[ctx->stage].start = iter.raw_value * 8192;
 737       } else if (strstr(iter.name, "URB Entry Allocation Size")) {
 738          ctx->urb_stages[ctx->stage].size = (iter.raw_value + 1) * 64;
 739       } else if (strstr(iter.name, "Number of URB Entries")) {
 740          ctx->urb_stages[ctx->stage].n_entries = iter.raw_value;
 741       }
 742    }
 743
 744    ctx->end_urb_offset = MAX2(ctx->urb_stages[ctx->stage].start +
 745                               ctx->urb_stages[ctx->stage].n_entries *
 746                               ctx->urb_stages[ctx->stage].size,
 747                               ctx->end_urb_offset);
 748 }
 749
 750 static void
 751 handle_urb_read(struct aub_viewer_decode_ctx *ctx,
 752                 struct gen_group *inst,
 753                 const uint32_t *p)
 754 {
 755    struct gen_field_iterator iter;
 756    gen_field_iterator_init(&iter, inst, p, 0, false);
 757    while (gen_field_iterator_next(&iter)) {
 758       /* Workaround the "Force * URB Entry Read Length" fields */
 759       if (iter.end_bit - iter.start_bit < 2)
 760          continue;
 761
 762       if (strstr(iter.name, "URB Entry Read Offset")) {
 763          ctx->urb_stages[ctx->stage].rd_offset = iter.raw_value * 32;
 764       } else if (strstr(iter.name, "URB Entry Read Length")) {
 765          ctx->urb_stages[ctx->stage].rd_length = iter.raw_value * 32;
 766       } else if (strstr(iter.name, "URB Entry Output Read Offset")) {
 767          ctx->urb_stages[ctx->stage].wr_offset = iter.raw_value * 32;
 768       } else if (strstr(iter.name, "URB Entry Output Length")) {
 769          ctx->urb_stages[ctx->stage].wr_length = iter.raw_value * 32;
 770       }
 771    }
 772 }
 773
 774 static void
 775 handle_urb_constant(struct aub_viewer_decode_ctx *ctx,
 776                     struct gen_group *inst,
 777                     const uint32_t *p)
 778 {
 779    struct gen_group *body =
 780       gen_spec_find_struct(ctx->spec, "3DSTATE_CONSTANT_BODY");
 781
 782    struct gen_field_iterator outer;
 783    gen_field_iterator_init(&outer, inst, p, 0, false);
 784    while (gen_field_iterator_next(&outer)) {
 785       if (outer.struct_desc != body)
 786          continue;
 787
 788       struct gen_field_iterator iter;
 789       gen_field_iterator_init(&iter, body, &outer.p[outer.start_bit / 32],
 790                               0, false);
 791
 792       ctx->urb_stages[ctx->stage].const_rd_length = 0;
 793       while (gen_field_iterator_next(&iter)) {
 794          int idx;
 795          if (sscanf(iter.name, "Read Length[%d]", &idx) == 1) {
 796             ctx->urb_stages[ctx->stage].const_rd_length += iter.raw_value * 32;
 797          }
 798       }
 799    }
 800 }
 801
 802 struct custom_decoder {
 803    const char *cmd_name;
 804    void (*decode)(struct aub_viewer_decode_ctx *ctx,
 805                   struct gen_group *inst,
 806                   const uint32_t *p);
 807    enum aub_decode_stage stage;
 808 } display_decoders[] = {
 809    { "STATE_BASE_ADDRESS", handle_state_base_address },
 810    { "MEDIA_INTERFACE_DESCRIPTOR_LOAD", handle_media_interface_descriptor_load },
 811    { "3DSTATE_VERTEX_BUFFERS", handle_3dstate_vertex_buffers },
 812    { "3DSTATE_INDEX_BUFFER", handle_3dstate_index_buffer },
 813    { "3DSTATE_VS", decode_single_ksp, AUB_DECODE_STAGE_VS, },
 814    { "3DSTATE_GS", decode_single_ksp, AUB_DECODE_STAGE_GS, },
 815    { "3DSTATE_DS", decode_single_ksp, AUB_DECODE_STAGE_DS, },
 816    { "3DSTATE_HS", decode_single_ksp, AUB_DECODE_STAGE_HS, },
 817    { "3DSTATE_PS", decode_ps_kernels, AUB_DECODE_STAGE_PS, },
 818    { "3DSTATE_CONSTANT_VS", decode_3dstate_constant, AUB_DECODE_STAGE_VS, },
 819    { "3DSTATE_CONSTANT_GS", decode_3dstate_constant, AUB_DECODE_STAGE_GS, },
 820    { "3DSTATE_CONSTANT_DS", decode_3dstate_constant, AUB_DECODE_STAGE_DS, },
 821    { "3DSTATE_CONSTANT_HS", decode_3dstate_constant, AUB_DECODE_STAGE_HS, },
 822    { "3DSTATE_CONSTANT_PS", decode_3dstate_constant, AUB_DECODE_STAGE_PS, },
 823
 824    { "3DSTATE_BINDING_TABLE_POINTERS_VS", decode_3dstate_binding_table_pointers, AUB_DECODE_STAGE_VS, },
 825    { "3DSTATE_BINDING_TABLE_POINTERS_GS", decode_3dstate_binding_table_pointers, AUB_DECODE_STAGE_GS, },
 826    { "3DSTATE_BINDING_TABLE_POINTERS_HS", decode_3dstate_binding_table_pointers, AUB_DECODE_STAGE_HS, },
 827    { "3DSTATE_BINDING_TABLE_POINTERS_DS", decode_3dstate_binding_table_pointers, AUB_DECODE_STAGE_DS, },
 828    { "3DSTATE_BINDING_TABLE_POINTERS_PS", decode_3dstate_binding_table_pointers, AUB_DECODE_STAGE_PS, },
 829
 830    { "3DSTATE_SAMPLER_STATE_POINTERS_VS", decode_3dstate_sampler_state_pointers, AUB_DECODE_STAGE_VS, },
 831    { "3DSTATE_SAMPLER_STATE_POINTERS_GS", decode_3dstate_sampler_state_pointers, AUB_DECODE_STAGE_GS, },
 832    { "3DSTATE_SAMPLER_STATE_POINTERS_DS", decode_3dstate_sampler_state_pointers, AUB_DECODE_STAGE_DS, },
 833    { "3DSTATE_SAMPLER_STATE_POINTERS_HS", decode_3dstate_sampler_state_pointers, AUB_DECODE_STAGE_HS, },
 834    { "3DSTATE_SAMPLER_STATE_POINTERS_PS", decode_3dstate_sampler_state_pointers, AUB_DECODE_STAGE_PS, },
 835    { "3DSTATE_SAMPLER_STATE_POINTERS", decode_3dstate_sampler_state_pointers_gen6 },
 836
 837    { "3DSTATE_VIEWPORT_STATE_POINTERS_CC", decode_3dstate_viewport_state_pointers_cc },
 838    { "3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP", decode_3dstate_viewport_state_pointers_sf_clip },
 839    { "3DSTATE_BLEND_STATE_POINTERS", decode_3dstate_blend_state_pointers },
 840    { "3DSTATE_CC_STATE_POINTERS", decode_3dstate_cc_state_pointers },
 841    { "3DSTATE_SCISSOR_STATE_POINTERS", decode_3dstate_scissor_state_pointers },
 842    { "MI_LOAD_REGISTER_IMM", decode_load_register_imm },
 843    { "3DPRIMITIVE", decode_3dprimitive },
 844 };
 845
 846 struct custom_decoder info_decoders[] = {
 847    { "STATE_BASE_ADDRESS", handle_state_base_address },
 848    { "3DSTATE_URB_VS", handle_urb, AUB_DECODE_STAGE_VS, },
 849    { "3DSTATE_URB_GS", handle_urb, AUB_DECODE_STAGE_GS, },
 850    { "3DSTATE_URB_DS", handle_urb, AUB_DECODE_STAGE_DS, },
 851    { "3DSTATE_URB_HS", handle_urb, AUB_DECODE_STAGE_HS, },
 852    { "3DSTATE_VS", handle_urb_read, AUB_DECODE_STAGE_VS, },
 853    { "3DSTATE_GS", handle_urb_read, AUB_DECODE_STAGE_GS, },
 854    { "3DSTATE_DS", handle_urb_read, AUB_DECODE_STAGE_DS, },
 855    { "3DSTATE_HS", handle_urb_read, AUB_DECODE_STAGE_HS, },
 856    { "3DSTATE_PS", handle_urb_read, AUB_DECODE_STAGE_PS, },
 857    { "3DSTATE_CONSTANT_VS", handle_urb_constant, AUB_DECODE_STAGE_VS, },
 858    { "3DSTATE_CONSTANT_GS", handle_urb_constant, AUB_DECODE_STAGE_GS, },
 859    { "3DSTATE_CONSTANT_DS", handle_urb_constant, AUB_DECODE_STAGE_DS, },
 860    { "3DSTATE_CONSTANT_HS", handle_urb_constant, AUB_DECODE_STAGE_HS, },
 861    { "3DSTATE_CONSTANT_PS", handle_urb_constant, AUB_DECODE_STAGE_PS, },
 862 };
 863
 864 void
 865 aub_viewer_render_batch(struct aub_viewer_decode_ctx *ctx,
 866                         const void *_batch, uint32_t batch_size,
 867                         uint64_t batch_addr)
 868 {
 869    struct gen_group *inst;
 870    const uint32_t *p, *batch = (const uint32_t *) _batch, *end = batch + batch_size / sizeof(uint32_t);
 871    int length;
 872
 873    for (p = batch; p < end; p += length) {
 874       inst = gen_spec_find_instruction(ctx->spec, p);
 875       length = gen_group_get_length(inst, p);
 876       assert(inst == NULL || length > 0);
 877       length = MAX2(1, length);
 878
 879       uint64_t offset = batch_addr + ((char *)p - (char *)batch);
 880
 881       if (inst == NULL) {
 882          ImGui::TextColored(ctx->cfg->error_color,
 883                             "x%08" PRIx64 ": unknown instruction %08x",
 884                             offset, p[0]);
 885          continue;
 886       }
 887
 888       const char *inst_name = gen_group_get_name(inst);
 889
 890       for (unsigned i = 0; i < ARRAY_SIZE(info_decoders); i++) {
 891          if (strcmp(inst_name, info_decoders[i].cmd_name) == 0) {
 892             ctx->stage = info_decoders[i].stage;
 893             info_decoders[i].decode(ctx, inst, p);
 894             break;
 895          }
 896       }
 897
 898       if (ctx->decode_cfg->command_filter.PassFilter(inst->name) &&
 899           ImGui::TreeNodeEx(p,
 900                             ImGuiTreeNodeFlags_Framed,
 901                             "0x%08" PRIx64 ":  %s",
 902                             offset, inst->name)) {
 903          aub_viewer_print_group(ctx, inst, offset, p);
 904
 905          for (unsigned i = 0; i < ARRAY_SIZE(display_decoders); i++) {
 906             if (strcmp(inst_name, display_decoders[i].cmd_name) == 0) {
 907                ctx->stage = display_decoders[i].stage;
 908                display_decoders[i].decode(ctx, inst, p);
 909                break;
 910             }
 911          }
 912
 913          if (ctx->edit_address) {
 914             if (ImGui::Button("Edit instruction"))
 915                ctx->edit_address(ctx->user_data, offset, length * 4);
 916          }
 917
 918          ImGui::TreePop();
 919       }
 920
 921       if (strcmp(inst_name, "MI_BATCH_BUFFER_START") == 0) {
 922          struct gen_batch_decode_bo next_batch = {};
 923          bool second_level;
 924          struct gen_field_iterator iter;
 925          gen_field_iterator_init(&iter, inst, p, 0, false);
 926          while (gen_field_iterator_next(&iter)) {
 927             if (strcmp(iter.name, "Batch Buffer Start Address") == 0) {
 928                next_batch = ctx_get_bo(ctx, iter.raw_value);
 929             } else if (strcmp(iter.name, "Second Level Batch Buffer") == 0) {
 930                second_level = iter.raw_value;
 931             }
 932          }
 933
 934          if (next_batch.map == NULL) {
 935             ImGui::TextColored(ctx->cfg->missing_color,
 936                                "Secondary batch at 0x%08" PRIx64 " unavailable",
 937                                next_batch.addr);
 938          } else {
 939             aub_viewer_render_batch(ctx, next_batch.map, next_batch.size,
 940                                     next_batch.addr);
 941          }
 942          if (second_level) {
 943             /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" set acts
 944              * like a subroutine call.  Commands that come afterwards get
 945              * processed once the 2nd level batch buffer returns with
 946              * MI_BATCH_BUFFER_END.
 947              */
 948             continue;
 949          } else {
 950             /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" unset acts
 951              * like a goto.  Nothing after it will ever get processed.  In
 952              * order to prevent the recursion from growing, we just reset the
 953              * loop and continue;
 954              */
 955             break;
 956          }
 957       } else if (strcmp(inst_name, "MI_BATCH_BUFFER_END") == 0) {
 958          break;
 959       }
 960    }
 961 }