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    while (gen_field_iterator_next(&iter)) {
 162       if (strcmp(iter.name, "Surface State Base Address") == 0) {
 163          ctx->surface_base = iter.raw_value;
 164       } else if (strcmp(iter.name, "Dynamic State Base Address") == 0) {
 165          ctx->dynamic_base = iter.raw_value;
 166       } else if (strcmp(iter.name, "Instruction Base Address") == 0) {
 167          ctx->instruction_base = iter.raw_value;
 168       }
 169    }
 170 }
 171
 172 static void
 173 dump_binding_table(struct aub_viewer_decode_ctx *ctx, uint32_t offset, int count)
 174 {
 175    struct gen_group *strct =
 176       gen_spec_find_struct(ctx->spec, "RENDER_SURFACE_STATE");
 177    if (strct == NULL) {
 178       ImGui::TextColored(ctx->cfg->missing_color, "did not find RENDER_SURFACE_STATE info");
 179       return;
 180    }
 181
 182    if (count < 0)
 183       count = update_count(ctx, offset, 1, 8);
 184
 185    if (offset % 32 != 0 || offset >= UINT16_MAX) {
 186       ImGui::TextColored(ctx->cfg->missing_color, "invalid binding table pointer");
 187       return;
 188    }
 189
 190    struct gen_batch_decode_bo bind_bo =
 191       ctx_get_bo(ctx, ctx->surface_base + offset);
 192
 193    if (bind_bo.map == NULL) {
 194       ImGui::TextColored(ctx->cfg->missing_color,
 195                          "binding table unavailable addr=0x%08" PRIx64,
 196                          ctx->surface_base + offset);
 197       return;
 198    }
 199
 200    const uint32_t *pointers = (const uint32_t *) bind_bo.map;
 201    for (int i = 0; i < count; i++) {
 202       if (pointers[i] == 0)
 203          continue;
 204
 205       uint64_t addr = ctx->surface_base + pointers[i];
 206       struct gen_batch_decode_bo bo = ctx_get_bo(ctx, addr);
 207       uint32_t size = strct->dw_length * 4;
 208
 209       if (pointers[i] % 32 != 0 ||
 210           addr < bo.addr || addr + size >= bo.addr + bo.size) {
 211          ImGui::TextColored(ctx->cfg->missing_color,
 212                             "pointer %u: %08x <not valid>", i, pointers[i]);
 213          continue;
 214       }
 215
 216       ImGui::Text("pointer %u: %08x", i, pointers[i]);
 217       aub_viewer_print_group(ctx, strct, addr, (const uint8_t *) bo.map + (addr - bo.addr));
 218    }
 219 }
 220
 221 static void
 222 dump_samplers(struct aub_viewer_decode_ctx *ctx, uint32_t offset, int count)
 223 {
 224    struct gen_group *strct = gen_spec_find_struct(ctx->spec, "SAMPLER_STATE");
 225
 226    if (count < 0)
 227       count = update_count(ctx, offset, strct->dw_length, 4);
 228
 229    uint64_t state_addr = ctx->dynamic_base + offset;
 230    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, state_addr);
 231    const uint8_t *state_map = (const uint8_t *) bo.map;
 232
 233    if (state_map == NULL) {
 234       ImGui::TextColored(ctx->cfg->missing_color,
 235                          "samplers unavailable addr=0x%08" PRIx64, state_addr);
 236       return;
 237    }
 238
 239    if (offset % 32 != 0 || state_addr - bo.addr >= bo.size) {
 240       ImGui::TextColored(ctx->cfg->missing_color, "invalid sampler state pointer");
 241       return;
 242    }
 243
 244    for (int i = 0; i < count; i++) {
 245       ImGui::Text("sampler state %d", i);
 246       aub_viewer_print_group(ctx, strct, state_addr, state_map);
 247       state_addr += 16;
 248       state_map += 16;
 249    }
 250 }
 251
 252 static void
 253 handle_media_interface_descriptor_load(struct aub_viewer_decode_ctx *ctx,
 254                                        struct gen_group *inst,
 255                                        const uint32_t *p)
 256 {
 257    struct gen_group *desc =
 258       gen_spec_find_struct(ctx->spec, "INTERFACE_DESCRIPTOR_DATA");
 259
 260    struct gen_field_iterator iter;
 261    gen_field_iterator_init(&iter, inst, p, 0, false);
 262    uint32_t descriptor_offset = 0;
 263    int descriptor_count = 0;
 264    while (gen_field_iterator_next(&iter)) {
 265       if (strcmp(iter.name, "Interface Descriptor Data Start Address") == 0) {
 266          descriptor_offset = strtol(iter.value, NULL, 16);
 267       } else if (strcmp(iter.name, "Interface Descriptor Total Length") == 0) {
 268          descriptor_count =
 269             strtol(iter.value, NULL, 16) / (desc->dw_length * 4);
 270       }
 271    }
 272
 273    uint64_t desc_addr = ctx->dynamic_base + descriptor_offset;
 274    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, desc_addr);
 275    const uint32_t *desc_map = (const uint32_t *) bo.map;
 276
 277    if (desc_map == NULL) {
 278       ImGui::TextColored(ctx->cfg->missing_color,
 279                          "interface descriptors unavailable addr=0x%08" PRIx64, desc_addr);
 280       return;
 281    }
 282
 283    for (int i = 0; i < descriptor_count; i++) {
 284       ImGui::Text("descriptor %d: %08x", i, descriptor_offset);
 285
 286       aub_viewer_print_group(ctx, desc, desc_addr, desc_map);
 287
 288       gen_field_iterator_init(&iter, desc, desc_map, 0, false);
 289       uint64_t ksp = 0;
 290       uint32_t sampler_offset = 0, sampler_count = 0;
 291       uint32_t binding_table_offset = 0, binding_entry_count = 0;
 292       while (gen_field_iterator_next(&iter)) {
 293          if (strcmp(iter.name, "Kernel Start Pointer") == 0) {
 294             ksp = strtoll(iter.value, NULL, 16);
 295          } else if (strcmp(iter.name, "Sampler State Pointer") == 0) {
 296             sampler_offset = strtol(iter.value, NULL, 16);
 297          } else if (strcmp(iter.name, "Sampler Count") == 0) {
 298             sampler_count = strtol(iter.value, NULL, 10);
 299          } else if (strcmp(iter.name, "Binding Table Pointer") == 0) {
 300             binding_table_offset = strtol(iter.value, NULL, 16);
 301          } else if (strcmp(iter.name, "Binding Table Entry Count") == 0) {
 302             binding_entry_count = strtol(iter.value, NULL, 10);
 303          }
 304       }
 305
 306       ctx_disassemble_program(ctx, ksp, "compute shader");
 307
 308       dump_samplers(ctx, sampler_offset, sampler_count);
 309       dump_binding_table(ctx, binding_table_offset, binding_entry_count);
 310
 311       desc_map += desc->dw_length;
 312       desc_addr += desc->dw_length * 4;
 313    }
 314 }
 315
 316 static void
 317 handle_3dstate_vertex_buffers(struct aub_viewer_decode_ctx *ctx,
 318                               struct gen_group *inst,
 319                               const uint32_t *p)
 320 {
 321    struct gen_group *vbs = gen_spec_find_struct(ctx->spec, "VERTEX_BUFFER_STATE");
 322
 323    struct gen_batch_decode_bo vb = {};
 324    uint32_t vb_size = 0;
 325    int index = -1;
 326    int pitch = -1;
 327    bool ready = false;
 328
 329    struct gen_field_iterator iter;
 330    gen_field_iterator_init(&iter, inst, p, 0, false);
 331    while (gen_field_iterator_next(&iter)) {
 332       if (iter.struct_desc != vbs)
 333          continue;
 334
 335       uint64_t buffer_addr;
 336
 337       struct gen_field_iterator vbs_iter;
 338       gen_field_iterator_init(&vbs_iter, vbs, &iter.p[iter.start_bit / 32], 0, false);
 339       while (gen_field_iterator_next(&vbs_iter)) {
 340          if (strcmp(vbs_iter.name, "Vertex Buffer Index") == 0) {
 341             index = vbs_iter.raw_value;
 342          } else if (strcmp(vbs_iter.name, "Buffer Pitch") == 0) {
 343             pitch = vbs_iter.raw_value;
 344          } else if (strcmp(vbs_iter.name, "Buffer Starting Address") == 0) {
 345             buffer_addr = vbs_iter.raw_value;
 346             vb = ctx_get_bo(ctx, buffer_addr);
 347          } else if (strcmp(vbs_iter.name, "Buffer Size") == 0) {
 348             vb_size = vbs_iter.raw_value;
 349             ready = true;
 350          } else if (strcmp(vbs_iter.name, "End Address") == 0) {
 351             if (vb.map && vbs_iter.raw_value >= vb.addr)
 352                vb_size = vbs_iter.raw_value - vb.addr;
 353             else
 354                vb_size = 0;
 355             ready = true;
 356          }
 357
 358          if (!ready)
 359             continue;
 360
 361          ImGui::Text("vertex buffer %d, size %d", index, vb_size);
 362
 363          if (vb.map == NULL) {
 364             ImGui::TextColored(ctx->cfg->missing_color,
 365                                "buffer contents unavailable addr=0x%08" PRIx64, buffer_addr);
 366             continue;
 367          }
 368
 369          if (vb.map == 0 || vb_size == 0)
 370             continue;
 371
 372          vb.map = NULL;
 373          vb_size = 0;
 374          index = -1;
 375          pitch = -1;
 376          ready = false;
 377       }
 378    }
 379 }
 380
 381 static void
 382 handle_3dstate_index_buffer(struct aub_viewer_decode_ctx *ctx,
 383                             struct gen_group *inst,
 384                             const uint32_t *p)
 385 {
 386    struct gen_batch_decode_bo ib = {};
 387    uint64_t buffer_addr = 0;
 388    uint32_t ib_size = 0;
 389    uint32_t format = 0;
 390
 391    struct gen_field_iterator iter;
 392    gen_field_iterator_init(&iter, inst, p, 0, false);
 393    while (gen_field_iterator_next(&iter)) {
 394       if (strcmp(iter.name, "Index Format") == 0) {
 395          format = iter.raw_value;
 396       } else if (strcmp(iter.name, "Buffer Starting Address") == 0) {
 397          buffer_addr = iter.raw_value;
 398          ib = ctx_get_bo(ctx, buffer_addr);
 399       } else if (strcmp(iter.name, "Buffer Size") == 0) {
 400          ib_size = iter.raw_value;
 401       }
 402    }
 403
 404    if (ib.map == NULL) {
 405       ImGui::TextColored(ctx->cfg->missing_color,
 406                          "buffer contents unavailable addr=0x%08" PRIx64,
 407                          buffer_addr);
 408       return;
 409    }
 410
 411    const uint8_t *m = (const uint8_t *) ib.map;
 412    const uint8_t *ib_end = m + MIN2(ib.size, ib_size);
 413    for (int i = 0; m < ib_end && i < 10; i++) {
 414       switch (format) {
 415       case 0:
 416          m += 1;
 417          break;
 418       case 1:
 419          m += 2;
 420          break;
 421       case 2:
 422          m += 4;
 423          break;
 424       }
 425    }
 426 }
 427
 428 static void
 429 decode_single_ksp(struct aub_viewer_decode_ctx *ctx,
 430                   struct gen_group *inst,
 431                   const uint32_t *p)
 432 {
 433    uint64_t ksp = 0;
 434    bool is_simd8 = false; /* vertex shaders on Gen8+ only */
 435    bool is_enabled = true;
 436
 437    struct gen_field_iterator iter;
 438    gen_field_iterator_init(&iter, inst, p, 0, false);
 439    while (gen_field_iterator_next(&iter)) {
 440       if (strcmp(iter.name, "Kernel Start Pointer") == 0) {
 441          ksp = iter.raw_value;
 442       } else if (strcmp(iter.name, "SIMD8 Dispatch Enable") == 0) {
 443          is_simd8 = iter.raw_value;
 444       } else if (strcmp(iter.name, "Dispatch Mode") == 0) {
 445          is_simd8 = strcmp(iter.value, "SIMD8") == 0;
 446       } else if (strcmp(iter.name, "Dispatch Enable") == 0) {
 447          is_simd8 = strcmp(iter.value, "SIMD8") == 0;
 448       } else if (strcmp(iter.name, "Enable") == 0) {
 449          is_enabled = iter.raw_value;
 450       }
 451    }
 452
 453    const char *type =
 454       strcmp(inst->name,   "VS_STATE") == 0 ? "vertex shader" :
 455       strcmp(inst->name,   "GS_STATE") == 0 ? "geometry shader" :
 456       strcmp(inst->name,   "SF_STATE") == 0 ? "strips and fans shader" :
 457       strcmp(inst->name, "CLIP_STATE") == 0 ? "clip shader" :
 458       strcmp(inst->name, "3DSTATE_DS") == 0 ? "tessellation evaluation shader" :
 459       strcmp(inst->name, "3DSTATE_HS") == 0 ? "tessellation control shader" :
 460       strcmp(inst->name, "3DSTATE_VS") == 0 ? (is_simd8 ? "SIMD8 vertex shader" : "vec4 vertex shader") :
 461       strcmp(inst->name, "3DSTATE_GS") == 0 ? (is_simd8 ? "SIMD8 geometry shader" : "vec4 geometry shader") :
 462       NULL;
 463
 464    if (is_enabled)
 465       ctx_disassemble_program(ctx, ksp, type);
 466 }
 467
 468 static void
 469 decode_ps_kernels(struct aub_viewer_decode_ctx *ctx,
 470                   struct gen_group *inst,
 471                   const uint32_t *p)
 472 {
 473    uint64_t ksp[3] = {0, 0, 0};
 474    bool enabled[3] = {false, false, false};
 475
 476    struct gen_field_iterator iter;
 477    gen_field_iterator_init(&iter, inst, p, 0, false);
 478    while (gen_field_iterator_next(&iter)) {
 479       if (strncmp(iter.name, "Kernel Start Pointer ",
 480                   strlen("Kernel Start Pointer ")) == 0) {
 481          int idx = iter.name[strlen("Kernel Start Pointer ")] - '0';
 482          ksp[idx] = strtol(iter.value, NULL, 16);
 483       } else if (strcmp(iter.name, "8 Pixel Dispatch Enable") == 0) {
 484          enabled[0] = strcmp(iter.value, "true") == 0;
 485       } else if (strcmp(iter.name, "16 Pixel Dispatch Enable") == 0) {
 486          enabled[1] = strcmp(iter.value, "true") == 0;
 487       } else if (strcmp(iter.name, "32 Pixel Dispatch Enable") == 0) {
 488          enabled[2] = strcmp(iter.value, "true") == 0;
 489       }
 490    }
 491
 492    /* Reorder KSPs to be [8, 16, 32] instead of the hardware order. */
 493    if (enabled[0] + enabled[1] + enabled[2] == 1) {
 494       if (enabled[1]) {
 495          ksp[1] = ksp[0];
 496          ksp[0] = 0;
 497       } else if (enabled[2]) {
 498          ksp[2] = ksp[0];
 499          ksp[0] = 0;
 500       }
 501    } else {
 502       uint64_t tmp = ksp[1];
 503       ksp[1] = ksp[2];
 504       ksp[2] = tmp;
 505    }
 506
 507    if (enabled[0])
 508       ctx_disassemble_program(ctx, ksp[0], "SIMD8 fragment shader");
 509    if (enabled[1])
 510       ctx_disassemble_program(ctx, ksp[1], "SIMD16 fragment shader");
 511    if (enabled[2])
 512       ctx_disassemble_program(ctx, ksp[2], "SIMD32 fragment shader");
 513 }
 514
 515 static void
 516 decode_3dstate_constant(struct aub_viewer_decode_ctx *ctx,
 517                         struct gen_group *inst,
 518                         const uint32_t *p)
 519 {
 520    struct gen_group *body =
 521       gen_spec_find_struct(ctx->spec, "3DSTATE_CONSTANT_BODY");
 522
 523    uint32_t read_length[4] = {0};
 524    uint64_t read_addr[4];
 525
 526    struct gen_field_iterator outer;
 527    gen_field_iterator_init(&outer, inst, p, 0, false);
 528    while (gen_field_iterator_next(&outer)) {
 529       if (outer.struct_desc != body)
 530          continue;
 531
 532       struct gen_field_iterator iter;
 533       gen_field_iterator_init(&iter, body, &outer.p[outer.start_bit / 32],
 534                               0, false);
 535
 536       while (gen_field_iterator_next(&iter)) {
 537          int idx;
 538          if (sscanf(iter.name, "Read Length[%d]", &idx) == 1) {
 539             read_length[idx] = iter.raw_value;
 540          } else if (sscanf(iter.name, "Buffer[%d]", &idx) == 1) {
 541             read_addr[idx] = iter.raw_value;
 542          }
 543       }
 544
 545       for (int i = 0; i < 4; i++) {
 546          if (read_length[i] == 0)
 547             continue;
 548
 549          struct gen_batch_decode_bo buffer = ctx_get_bo(ctx, read_addr[i]);
 550          if (!buffer.map) {
 551             ImGui::TextColored(ctx->cfg->missing_color,
 552                                "constant buffer %d unavailable addr=0x%08" PRIx64,
 553                                i, read_addr[i]);
 554             continue;
 555          }
 556
 557          unsigned size = read_length[i] * 32;
 558          ImGui::Text("constant buffer %d, size %u", i, size);
 559
 560          if (ctx->edit_address) {
 561             if (ImGui::Button("Show/Edit buffer"))
 562                ctx->edit_address(ctx->user_data, read_addr[i], size);
 563          }
 564       }
 565    }
 566 }
 567
 568 static void
 569 decode_3dstate_binding_table_pointers(struct aub_viewer_decode_ctx *ctx,
 570                                       struct gen_group *inst,
 571                                       const uint32_t *p)
 572 {
 573    dump_binding_table(ctx, p[1], -1);
 574 }
 575
 576 static void
 577 decode_3dstate_sampler_state_pointers(struct aub_viewer_decode_ctx *ctx,
 578                                       struct gen_group *inst,
 579                                       const uint32_t *p)
 580 {
 581    dump_samplers(ctx, p[1], -1);
 582 }
 583
 584 static void
 585 decode_3dstate_sampler_state_pointers_gen6(struct aub_viewer_decode_ctx *ctx,
 586                                            struct gen_group *inst,
 587                                            const uint32_t *p)
 588 {
 589    dump_samplers(ctx, p[1], -1);
 590    dump_samplers(ctx, p[2], -1);
 591    dump_samplers(ctx, p[3], -1);
 592 }
 593
 594 static bool
 595 str_ends_with(const char *str, const char *end)
 596 {
 597    int offset = strlen(str) - strlen(end);
 598    if (offset < 0)
 599       return false;
 600
 601    return strcmp(str + offset, end) == 0;
 602 }
 603
 604 static void
 605 decode_dynamic_state_pointers(struct aub_viewer_decode_ctx *ctx,
 606                               struct gen_group *inst, const uint32_t *p,
 607                               const char *struct_type,  int count)
 608 {
 609    struct gen_group *state = gen_spec_find_struct(ctx->spec, struct_type);
 610
 611    uint32_t state_offset = 0;
 612
 613    struct gen_field_iterator iter;
 614    gen_field_iterator_init(&iter, inst, p, 0, false);
 615    while (gen_field_iterator_next(&iter)) {
 616       if (str_ends_with(iter.name, "Pointer")) {
 617          state_offset = iter.raw_value;
 618          break;
 619       }
 620    }
 621
 622    uint64_t state_addr = ctx->dynamic_base + state_offset;
 623    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, state_addr);
 624    const uint8_t *state_map = (const uint8_t *) bo.map;
 625
 626    if (state_map == NULL) {
 627       ImGui::TextColored(ctx->cfg->missing_color,
 628                          "dynamic %s state unavailable addr=0x%08" PRIx64,
 629                          struct_type, state_addr);
 630       return;
 631    }
 632
 633    for (int i = 0; i < count; i++) {
 634       ImGui::Text("%s %d", struct_type, i);
 635       aub_viewer_print_group(ctx, state, state_offset, state_map);
 636
 637       state_addr += state->dw_length * 4;
 638       state_map += state->dw_length;
 639    }
 640 }
 641
 642 static void
 643 decode_3dstate_viewport_state_pointers_cc(struct aub_viewer_decode_ctx *ctx,
 644                                           struct gen_group *inst,
 645                                           const uint32_t *p)
 646 {
 647    decode_dynamic_state_pointers(ctx, inst, p, "CC_VIEWPORT", 4);
 648 }
 649
 650 static void
 651 decode_3dstate_viewport_state_pointers_sf_clip(struct aub_viewer_decode_ctx *ctx,
 652                                                struct gen_group *inst,
 653                                                const uint32_t *p)
 654 {
 655    decode_dynamic_state_pointers(ctx, inst, p, "SF_CLIP_VIEWPORT", 4);
 656 }
 657
 658 static void
 659 decode_3dstate_blend_state_pointers(struct aub_viewer_decode_ctx *ctx,
 660                                     struct gen_group *inst,
 661                                     const uint32_t *p)
 662 {
 663    decode_dynamic_state_pointers(ctx, inst, p, "BLEND_STATE", 1);
 664 }
 665
 666 static void
 667 decode_3dstate_cc_state_pointers(struct aub_viewer_decode_ctx *ctx,
 668                                  struct gen_group *inst,
 669                                  const uint32_t *p)
 670 {
 671    decode_dynamic_state_pointers(ctx, inst, p, "COLOR_CALC_STATE", 1);
 672 }
 673
 674 static void
 675 decode_3dstate_scissor_state_pointers(struct aub_viewer_decode_ctx *ctx,
 676                                       struct gen_group *inst,
 677                                       const uint32_t *p)
 678 {
 679    decode_dynamic_state_pointers(ctx, inst, p, "SCISSOR_RECT", 1);
 680 }
 681
 682 static void
 683 decode_load_register_imm(struct aub_viewer_decode_ctx *ctx,
 684                          struct gen_group *inst,
 685                          const uint32_t *p)
 686 {
 687    struct gen_group *reg = gen_spec_find_register(ctx->spec, p[1]);
 688
 689    if (reg != NULL &&
 690        ImGui::TreeNodeEx(&p[1], ImGuiTreeNodeFlags_Framed,
 691                          "%s (0x%x) = 0x%x",
 692                          reg->name, reg->register_offset, p[2])) {
 693       aub_viewer_print_group(ctx, reg, reg->register_offset, &p[2]);
 694       ImGui::TreePop();
 695    }
 696 }
 697
 698 static void
 699 decode_3dprimitive(struct aub_viewer_decode_ctx *ctx,
 700                    struct gen_group *inst,
 701                    const uint32_t *p)
 702 {
 703    if (ctx->display_urb) {
 704       if (ImGui::Button("Show URB"))
 705          ctx->display_urb(ctx->user_data, ctx->urb_stages);
 706    }
 707 }
 708
 709 static void
 710 handle_urb(struct aub_viewer_decode_ctx *ctx,
 711            struct gen_group *inst,
 712            const uint32_t *p)
 713 {
 714    struct gen_field_iterator iter;
 715    gen_field_iterator_init(&iter, inst, p, 0, false);
 716    while (gen_field_iterator_next(&iter)) {
 717       if (strstr(iter.name, "URB Starting Address")) {
 718          ctx->urb_stages[ctx->stage].start = iter.raw_value * 8192;
 719       } else if (strstr(iter.name, "URB Entry Allocation Size")) {
 720          ctx->urb_stages[ctx->stage].size = (iter.raw_value + 1) * 64;
 721       } else if (strstr(iter.name, "Number of URB Entries")) {
 722          ctx->urb_stages[ctx->stage].n_entries = iter.raw_value;
 723       }
 724    }
 725
 726    ctx->end_urb_offset = MAX2(ctx->urb_stages[ctx->stage].start +
 727                               ctx->urb_stages[ctx->stage].n_entries *
 728                               ctx->urb_stages[ctx->stage].size,
 729                               ctx->end_urb_offset);
 730 }
 731
 732 static void
 733 handle_urb_read(struct aub_viewer_decode_ctx *ctx,
 734                 struct gen_group *inst,
 735                 const uint32_t *p)
 736 {
 737    struct gen_field_iterator iter;
 738    gen_field_iterator_init(&iter, inst, p, 0, false);
 739    while (gen_field_iterator_next(&iter)) {
 740       /* Workaround the "Force * URB Entry Read Length" fields */
 741       if (iter.end_bit - iter.start_bit < 2)
 742          continue;
 743
 744       if (strstr(iter.name, "URB Entry Read Offset")) {
 745          ctx->urb_stages[ctx->stage].rd_offset = iter.raw_value * 32;
 746       } else if (strstr(iter.name, "URB Entry Read Length")) {
 747          ctx->urb_stages[ctx->stage].rd_length = iter.raw_value * 32;
 748       } else if (strstr(iter.name, "URB Entry Output Read Offset")) {
 749          ctx->urb_stages[ctx->stage].wr_offset = iter.raw_value * 32;
 750       } else if (strstr(iter.name, "URB Entry Output Length")) {
 751          ctx->urb_stages[ctx->stage].wr_length = iter.raw_value * 32;
 752       }
 753    }
 754 }
 755
 756 static void
 757 handle_urb_constant(struct aub_viewer_decode_ctx *ctx,
 758                     struct gen_group *inst,
 759                     const uint32_t *p)
 760 {
 761    struct gen_group *body =
 762       gen_spec_find_struct(ctx->spec, "3DSTATE_CONSTANT_BODY");
 763
 764    struct gen_field_iterator outer;
 765    gen_field_iterator_init(&outer, inst, p, 0, false);
 766    while (gen_field_iterator_next(&outer)) {
 767       if (outer.struct_desc != body)
 768          continue;
 769
 770       struct gen_field_iterator iter;
 771       gen_field_iterator_init(&iter, body, &outer.p[outer.start_bit / 32],
 772                               0, false);
 773
 774       ctx->urb_stages[ctx->stage].const_rd_length = 0;
 775       while (gen_field_iterator_next(&iter)) {
 776          int idx;
 777          if (sscanf(iter.name, "Read Length[%d]", &idx) == 1) {
 778             ctx->urb_stages[ctx->stage].const_rd_length += iter.raw_value * 32;
 779          }
 780       }
 781    }
 782 }
 783
 784 struct custom_decoder {
 785    const char *cmd_name;
 786    void (*decode)(struct aub_viewer_decode_ctx *ctx,
 787                   struct gen_group *inst,
 788                   const uint32_t *p);
 789    enum aub_decode_stage stage;
 790 } display_decoders[] = {
 791    { "STATE_BASE_ADDRESS", handle_state_base_address },
 792    { "MEDIA_INTERFACE_DESCRIPTOR_LOAD", handle_media_interface_descriptor_load },
 793    { "3DSTATE_VERTEX_BUFFERS", handle_3dstate_vertex_buffers },
 794    { "3DSTATE_INDEX_BUFFER", handle_3dstate_index_buffer },
 795    { "3DSTATE_VS", decode_single_ksp, AUB_DECODE_STAGE_VS, },
 796    { "3DSTATE_GS", decode_single_ksp, AUB_DECODE_STAGE_GS, },
 797    { "3DSTATE_DS", decode_single_ksp, AUB_DECODE_STAGE_DS, },
 798    { "3DSTATE_HS", decode_single_ksp, AUB_DECODE_STAGE_HS, },
 799    { "3DSTATE_PS", decode_ps_kernels, AUB_DECODE_STAGE_PS, },
 800    { "3DSTATE_CONSTANT_VS", decode_3dstate_constant, AUB_DECODE_STAGE_VS, },
 801    { "3DSTATE_CONSTANT_GS", decode_3dstate_constant, AUB_DECODE_STAGE_GS, },
 802    { "3DSTATE_CONSTANT_DS", decode_3dstate_constant, AUB_DECODE_STAGE_DS, },
 803    { "3DSTATE_CONSTANT_HS", decode_3dstate_constant, AUB_DECODE_STAGE_HS, },
 804    { "3DSTATE_CONSTANT_PS", decode_3dstate_constant, AUB_DECODE_STAGE_PS, },
 805
 806    { "3DSTATE_BINDING_TABLE_POINTERS_VS", decode_3dstate_binding_table_pointers, AUB_DECODE_STAGE_VS, },
 807    { "3DSTATE_BINDING_TABLE_POINTERS_GS", decode_3dstate_binding_table_pointers, AUB_DECODE_STAGE_GS, },
 808    { "3DSTATE_BINDING_TABLE_POINTERS_HS", decode_3dstate_binding_table_pointers, AUB_DECODE_STAGE_HS, },
 809    { "3DSTATE_BINDING_TABLE_POINTERS_DS", decode_3dstate_binding_table_pointers, AUB_DECODE_STAGE_DS, },
 810    { "3DSTATE_BINDING_TABLE_POINTERS_PS", decode_3dstate_binding_table_pointers, AUB_DECODE_STAGE_PS, },
 811
 812    { "3DSTATE_SAMPLER_STATE_POINTERS_VS", decode_3dstate_sampler_state_pointers, AUB_DECODE_STAGE_VS, },
 813    { "3DSTATE_SAMPLER_STATE_POINTERS_GS", decode_3dstate_sampler_state_pointers, AUB_DECODE_STAGE_GS, },
 814    { "3DSTATE_SAMPLER_STATE_POINTERS_DS", decode_3dstate_sampler_state_pointers, AUB_DECODE_STAGE_DS, },
 815    { "3DSTATE_SAMPLER_STATE_POINTERS_HS", decode_3dstate_sampler_state_pointers, AUB_DECODE_STAGE_HS, },
 816    { "3DSTATE_SAMPLER_STATE_POINTERS_PS", decode_3dstate_sampler_state_pointers, AUB_DECODE_STAGE_PS, },
 817    { "3DSTATE_SAMPLER_STATE_POINTERS", decode_3dstate_sampler_state_pointers_gen6 },
 818
 819    { "3DSTATE_VIEWPORT_STATE_POINTERS_CC", decode_3dstate_viewport_state_pointers_cc },
 820    { "3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP", decode_3dstate_viewport_state_pointers_sf_clip },
 821    { "3DSTATE_BLEND_STATE_POINTERS", decode_3dstate_blend_state_pointers },
 822    { "3DSTATE_CC_STATE_POINTERS", decode_3dstate_cc_state_pointers },
 823    { "3DSTATE_SCISSOR_STATE_POINTERS", decode_3dstate_scissor_state_pointers },
 824    { "MI_LOAD_REGISTER_IMM", decode_load_register_imm },
 825    { "3DPRIMITIVE", decode_3dprimitive },
 826 };
 827
 828 struct custom_decoder info_decoders[] = {
 829    { "STATE_BASE_ADDRESS", handle_state_base_address },
 830    { "3DSTATE_URB_VS", handle_urb, AUB_DECODE_STAGE_VS, },
 831    { "3DSTATE_URB_GS", handle_urb, AUB_DECODE_STAGE_GS, },
 832    { "3DSTATE_URB_DS", handle_urb, AUB_DECODE_STAGE_DS, },
 833    { "3DSTATE_URB_HS", handle_urb, AUB_DECODE_STAGE_HS, },
 834    { "3DSTATE_VS", handle_urb_read, AUB_DECODE_STAGE_VS, },
 835    { "3DSTATE_GS", handle_urb_read, AUB_DECODE_STAGE_GS, },
 836    { "3DSTATE_DS", handle_urb_read, AUB_DECODE_STAGE_DS, },
 837    { "3DSTATE_HS", handle_urb_read, AUB_DECODE_STAGE_HS, },
 838    { "3DSTATE_PS", handle_urb_read, AUB_DECODE_STAGE_PS, },
 839    { "3DSTATE_CONSTANT_VS", handle_urb_constant, AUB_DECODE_STAGE_VS, },
 840    { "3DSTATE_CONSTANT_GS", handle_urb_constant, AUB_DECODE_STAGE_GS, },
 841    { "3DSTATE_CONSTANT_DS", handle_urb_constant, AUB_DECODE_STAGE_DS, },
 842    { "3DSTATE_CONSTANT_HS", handle_urb_constant, AUB_DECODE_STAGE_HS, },
 843    { "3DSTATE_CONSTANT_PS", handle_urb_constant, AUB_DECODE_STAGE_PS, },
 844 };
 845
 846 static inline uint64_t
 847 get_address(struct gen_spec *spec, const uint32_t *p)
 848 {
 849    /* Addresses are always guaranteed to be page-aligned and sometimes
 850     * hardware packets have extra stuff stuffed in the bottom 12 bits.
 851     */
 852    uint64_t addr = p[0] & ~0xfffu;
 853
 854    if (gen_spec_get_gen(spec) >= gen_make_gen(8,0)) {
 855       /* On Broadwell and above, we have 48-bit addresses which consume two
 856        * dwords.  Some packets require that these get stored in a "canonical
 857        * form" which means that bit 47 is sign-extended through the upper
 858        * bits. In order to correctly handle those aub dumps, we need to mask
 859        * off the top 16 bits.
 860        */
 861       addr |= ((uint64_t)p[1] & 0xffff) << 32;
 862    }
 863
 864    return addr;
 865 }
 866
 867 void
 868 aub_viewer_render_batch(struct aub_viewer_decode_ctx *ctx,
 869                         const void *_batch, uint32_t batch_size,
 870                         uint64_t batch_addr)
 871 {
 872    struct gen_group *inst;
 873    const uint32_t *p, *batch = (const uint32_t *) _batch, *end = batch + batch_size;
 874    int length;
 875
 876    for (p = batch; p < end; p += length) {
 877       inst = gen_spec_find_instruction(ctx->spec, p);
 878       length = gen_group_get_length(inst, p);
 879       assert(inst == NULL || length > 0);
 880       length = MAX2(1, length);
 881
 882       uint64_t offset = batch_addr + ((char *)p - (char *)batch);
 883
 884       if (inst == NULL) {
 885          ImGui::TextColored(ctx->cfg->error_color,
 886                             "x%08" PRIx64 ": unknown instruction %08x",
 887                             offset, p[0]);
 888          continue;
 889       }
 890
 891       const char *inst_name = gen_group_get_name(inst);
 892
 893       for (unsigned i = 0; i < ARRAY_SIZE(info_decoders); i++) {
 894          if (strcmp(inst_name, info_decoders[i].cmd_name) == 0) {
 895             ctx->stage = info_decoders[i].stage;
 896             info_decoders[i].decode(ctx, inst, p);
 897             break;
 898          }
 899       }
 900
 901       if (ctx->decode_cfg->command_filter.PassFilter(inst->name) &&
 902           ImGui::TreeNodeEx(p,
 903                             ImGuiTreeNodeFlags_Framed,
 904                             "0x%08" PRIx64 ":  %s",
 905                             offset, inst->name)) {
 906          aub_viewer_print_group(ctx, inst, offset, p);
 907
 908          for (unsigned i = 0; i < ARRAY_SIZE(display_decoders); i++) {
 909             if (strcmp(inst_name, display_decoders[i].cmd_name) == 0) {
 910                ctx->stage = display_decoders[i].stage;
 911                display_decoders[i].decode(ctx, inst, p);
 912                break;
 913             }
 914          }
 915
 916          if (ctx->edit_address) {
 917             if (ImGui::Button("Edit instruction"))
 918                ctx->edit_address(ctx->user_data, offset, length * 4);
 919          }
 920
 921          ImGui::TreePop();
 922       }
 923
 924       if (strcmp(inst_name, "MI_BATCH_BUFFER_START") == 0) {
 925          struct gen_batch_decode_bo next_batch = {};
 926          bool second_level;
 927          struct gen_field_iterator iter;
 928          gen_field_iterator_init(&iter, inst, p, 0, false);
 929          while (gen_field_iterator_next(&iter)) {
 930             if (strcmp(iter.name, "Batch Buffer Start Address") == 0) {
 931                next_batch = ctx_get_bo(ctx, iter.raw_value);
 932             } else if (strcmp(iter.name, "Second Level Batch Buffer") == 0) {
 933                second_level = iter.raw_value;
 934             }
 935          }
 936
 937          if (next_batch.map == NULL) {
 938             ImGui::TextColored(ctx->cfg->missing_color,
 939                                "Secondary batch at 0x%08" PRIx64 " unavailable",
 940                                next_batch.addr);
 941          } else {
 942             aub_viewer_render_batch(ctx, next_batch.map, next_batch.size,
 943                                     next_batch.addr);
 944          }
 945          if (second_level) {
 946             /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" set acts
 947              * like a subroutine call.  Commands that come afterwards get
 948              * processed once the 2nd level batch buffer returns with
 949              * MI_BATCH_BUFFER_END.
 950              */
 951             continue;
 952          } else {
 953             /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" unset acts
 954              * like a goto.  Nothing after it will ever get processed.  In
 955              * order to prevent the recursion from growing, we just reset the
 956              * loop and continue;
 957              */
 958             break;
 959          }
 960       } else if (strcmp(inst_name, "MI_BATCH_BUFFER_END") == 0) {
 961          break;
 962       }
 963    }
 964 }