src/intel/common/gen_batch_decoder.c

   1 /*
   2  * Copyright © 2017 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include "common/gen_decoder.h"
  25 #include "gen_disasm.h"
  26 #include "util/macros.h"
  27
  28 #include <string.h>
  29
  30 void
  31 gen_batch_decode_ctx_init(struct gen_batch_decode_ctx *ctx,
  32                           const struct gen_device_info *devinfo,
  33                           FILE *fp, enum gen_batch_decode_flags flags,
  34                           const char *xml_path,
  35                           struct gen_batch_decode_bo (*get_bo)(void *,
  36                                                                uint64_t),
  37                           unsigned (*get_state_size)(void *, uint32_t),
  38                           void *user_data)
  39 {
  40    memset(ctx, 0, sizeof(*ctx));
  41
  42    ctx->get_bo = get_bo;
  43    ctx->get_state_size = get_state_size;
  44    ctx->user_data = user_data;
  45    ctx->fp = fp;
  46    ctx->flags = flags;
  47    ctx->max_vbo_decoded_lines = -1; /* No limit! */
  48    ctx->engine = I915_ENGINE_CLASS_RENDER;
  49
  50    if (xml_path == NULL)
  51       ctx->spec = gen_spec_load(devinfo);
  52    else
  53       ctx->spec = gen_spec_load_from_path(devinfo, xml_path);
  54    ctx->disasm = gen_disasm_create(devinfo);
  55 }
  56
  57 void
  58 gen_batch_decode_ctx_finish(struct gen_batch_decode_ctx *ctx)
  59 {
  60    gen_spec_destroy(ctx->spec);
  61    gen_disasm_destroy(ctx->disasm);
  62 }
  63
  64 #define CSI "\e["
  65 #define RED_COLOR    CSI "31m"
  66 #define BLUE_HEADER  CSI "0;44m"
  67 #define GREEN_HEADER CSI "1;42m"
  68 #define NORMAL       CSI "0m"
  69
  70 static void
  71 ctx_print_group(struct gen_batch_decode_ctx *ctx,
  72                 struct gen_group *group,
  73                 uint64_t address, const void *map)
  74 {
  75    gen_print_group(ctx->fp, group, address, map, 0,
  76                    (ctx->flags & GEN_BATCH_DECODE_IN_COLOR) != 0);
  77 }
  78
  79 static struct gen_batch_decode_bo
  80 ctx_get_bo(struct gen_batch_decode_ctx *ctx, uint64_t addr)
  81 {
  82    if (gen_spec_get_gen(ctx->spec) >= gen_make_gen(8,0)) {
  83       /* On Broadwell and above, we have 48-bit addresses which consume two
  84        * dwords.  Some packets require that these get stored in a "canonical
  85        * form" which means that bit 47 is sign-extended through the upper
  86        * bits. In order to correctly handle those aub dumps, we need to mask
  87        * off the top 16 bits.
  88        */
  89       addr &= (~0ull >> 16);
  90    }
  91
  92    struct gen_batch_decode_bo bo = ctx->get_bo(ctx->user_data, addr);
  93
  94    if (gen_spec_get_gen(ctx->spec) >= gen_make_gen(8,0))
  95       bo.addr &= (~0ull >> 16);
  96
  97    /* We may actually have an offset into the bo */
  98    if (bo.map != NULL) {
  99       assert(bo.addr <= addr);
 100       uint64_t offset = addr - bo.addr;
 101       bo.map += offset;
 102       bo.addr += offset;
 103       bo.size -= offset;
 104    }
 105
 106    return bo;
 107 }
 108
 109 static int
 110 update_count(struct gen_batch_decode_ctx *ctx,
 111              uint32_t offset_from_dsba,
 112              unsigned element_dwords,
 113              unsigned guess)
 114 {
 115    unsigned size = 0;
 116
 117    if (ctx->get_state_size)
 118       size = ctx->get_state_size(ctx->user_data, offset_from_dsba);
 119
 120    if (size > 0)
 121       return size / (sizeof(uint32_t) * element_dwords);
 122
 123    /* In the absence of any information, just guess arbitrarily. */
 124    return guess;
 125 }
 126
 127 static void
 128 ctx_disassemble_program(struct gen_batch_decode_ctx *ctx,
 129                         uint32_t ksp, const char *type)
 130 {
 131    uint64_t addr = ctx->instruction_base + ksp;
 132    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, addr);
 133    if (!bo.map)
 134       return;
 135
 136    fprintf(ctx->fp, "\nReferenced %s:\n", type);
 137    gen_disasm_disassemble(ctx->disasm, bo.map, 0, ctx->fp);
 138 }
 139
 140 /* Heuristic to determine whether a uint32_t is probably actually a float
 141  * (http://stackoverflow.com/a/2953466)
 142  */
 143
 144 static bool
 145 probably_float(uint32_t bits)
 146 {
 147    int exp = ((bits & 0x7f800000U) >> 23) - 127;
 148    uint32_t mant = bits & 0x007fffff;
 149
 150    /* +- 0.0 */
 151    if (exp == -127 && mant == 0)
 152       return true;
 153
 154    /* +- 1 billionth to 1 billion */
 155    if (-30 <= exp && exp <= 30)
 156       return true;
 157
 158    /* some value with only a few binary digits */
 159    if ((mant & 0x0000ffff) == 0)
 160       return true;
 161
 162    return false;
 163 }
 164
 165 static void
 166 ctx_print_buffer(struct gen_batch_decode_ctx *ctx,
 167                  struct gen_batch_decode_bo bo,
 168                  uint32_t read_length,
 169                  uint32_t pitch,
 170                  int max_lines)
 171 {
 172    const uint32_t *dw_end = bo.map + MIN2(bo.size, read_length);
 173
 174    int column_count = 0, line_count = -1;
 175    for (const uint32_t *dw = bo.map; dw < dw_end; dw++) {
 176       if (column_count * 4 == pitch || column_count == 8) {
 177          fprintf(ctx->fp, "\n");
 178          column_count = 0;
 179          line_count++;
 180
 181          if (max_lines >= 0 && line_count >= max_lines)
 182             break;
 183       }
 184       fprintf(ctx->fp, column_count == 0 ? "  " : " ");
 185
 186       if ((ctx->flags & GEN_BATCH_DECODE_FLOATS) && probably_float(*dw))
 187          fprintf(ctx->fp, "  %8.2f", *(float *) dw);
 188       else
 189          fprintf(ctx->fp, "  0x%08x", *dw);
 190
 191       column_count++;
 192    }
 193    fprintf(ctx->fp, "\n");
 194 }
 195
 196 static struct gen_group *
 197 gen_ctx_find_instruction(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
 198 {
 199    return gen_spec_find_instruction(ctx->spec, ctx->engine, p);
 200 }
 201
 202 static void
 203 handle_state_base_address(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
 204 {
 205    struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
 206
 207    struct gen_field_iterator iter;
 208    gen_field_iterator_init(&iter, inst, p, 0, false);
 209
 210    uint64_t surface_base = 0, dynamic_base = 0, instruction_base = 0;
 211    bool surface_modify = 0, dynamic_modify = 0, instruction_modify = 0;
 212
 213    while (gen_field_iterator_next(&iter)) {
 214       if (strcmp(iter.name, "Surface State Base Address") == 0) {
 215          surface_base = iter.raw_value;
 216       } else if (strcmp(iter.name, "Dynamic State Base Address") == 0) {
 217          dynamic_base = iter.raw_value;
 218       } else if (strcmp(iter.name, "Instruction Base Address") == 0) {
 219          instruction_base = iter.raw_value;
 220       } else if (strcmp(iter.name, "Surface State Base Address Modify Enable") == 0) {
 221          surface_modify = iter.raw_value;
 222       } else if (strcmp(iter.name, "Dynamic State Base Address Modify Enable") == 0) {
 223          dynamic_modify = iter.raw_value;
 224       } else if (strcmp(iter.name, "Instruction Base Address Modify Enable") == 0) {
 225          instruction_modify = iter.raw_value;
 226       }
 227    }
 228
 229    if (dynamic_modify)
 230       ctx->dynamic_base = dynamic_base;
 231
 232    if (surface_modify)
 233       ctx->surface_base = surface_base;
 234
 235    if (instruction_modify)
 236       ctx->instruction_base = instruction_base;
 237 }
 238
 239 static void
 240 dump_binding_table(struct gen_batch_decode_ctx *ctx, uint32_t offset, int count)
 241 {
 242    struct gen_group *strct =
 243       gen_spec_find_struct(ctx->spec, "RENDER_SURFACE_STATE");
 244    if (strct == NULL) {
 245       fprintf(ctx->fp, "did not find RENDER_SURFACE_STATE info\n");
 246       return;
 247    }
 248
 249    if (count < 0)
 250       count = update_count(ctx, offset, 1, 8);
 251
 252    if (offset % 32 != 0 || offset >= UINT16_MAX) {
 253       fprintf(ctx->fp, "  invalid binding table pointer\n");
 254       return;
 255    }
 256
 257    struct gen_batch_decode_bo bind_bo =
 258       ctx_get_bo(ctx, ctx->surface_base + offset);
 259
 260    if (bind_bo.map == NULL) {
 261       fprintf(ctx->fp, "  binding table unavailable\n");
 262       return;
 263    }
 264
 265    const uint32_t *pointers = bind_bo.map;
 266    for (int i = 0; i < count; i++) {
 267       if (pointers[i] == 0)
 268          continue;
 269
 270       uint64_t addr = ctx->surface_base + pointers[i];
 271       struct gen_batch_decode_bo bo = ctx_get_bo(ctx, addr);
 272       uint32_t size = strct->dw_length * 4;
 273
 274       if (pointers[i] % 32 != 0 ||
 275           addr < bo.addr || addr + size >= bo.addr + bo.size) {
 276          fprintf(ctx->fp, "pointer %u: 0x%08x <not valid>\n", i, pointers[i]);
 277          continue;
 278       }
 279
 280       fprintf(ctx->fp, "pointer %u: 0x%08x\n", i, pointers[i]);
 281       ctx_print_group(ctx, strct, addr, bo.map + (addr - bo.addr));
 282    }
 283 }
 284
 285 static void
 286 dump_samplers(struct gen_batch_decode_ctx *ctx, uint32_t offset, int count)
 287 {
 288    struct gen_group *strct = gen_spec_find_struct(ctx->spec, "SAMPLER_STATE");
 289
 290    if (count < 0)
 291       count = update_count(ctx, offset, strct->dw_length, 4);
 292
 293    uint64_t state_addr = ctx->dynamic_base + offset;
 294    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, state_addr);
 295    const void *state_map = bo.map;
 296
 297    if (state_map == NULL) {
 298       fprintf(ctx->fp, "  samplers unavailable\n");
 299       return;
 300    }
 301
 302    if (offset % 32 != 0 || state_addr - bo.addr >= bo.size) {
 303       fprintf(ctx->fp, "  invalid sampler state pointer\n");
 304       return;
 305    }
 306
 307    for (int i = 0; i < count; i++) {
 308       fprintf(ctx->fp, "sampler state %d\n", i);
 309       ctx_print_group(ctx, strct, state_addr, state_map);
 310       state_addr += 16;
 311       state_map += 16;
 312    }
 313 }
 314
 315 static void
 316 handle_media_interface_descriptor_load(struct gen_batch_decode_ctx *ctx,
 317                                        const uint32_t *p)
 318 {
 319    struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
 320    struct gen_group *desc =
 321       gen_spec_find_struct(ctx->spec, "INTERFACE_DESCRIPTOR_DATA");
 322
 323    struct gen_field_iterator iter;
 324    gen_field_iterator_init(&iter, inst, p, 0, false);
 325    uint32_t descriptor_offset = 0;
 326    int descriptor_count = 0;
 327    while (gen_field_iterator_next(&iter)) {
 328       if (strcmp(iter.name, "Interface Descriptor Data Start Address") == 0) {
 329          descriptor_offset = strtol(iter.value, NULL, 16);
 330       } else if (strcmp(iter.name, "Interface Descriptor Total Length") == 0) {
 331          descriptor_count =
 332             strtol(iter.value, NULL, 16) / (desc->dw_length * 4);
 333       }
 334    }
 335
 336    uint64_t desc_addr = ctx->dynamic_base + descriptor_offset;
 337    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, desc_addr);
 338    const void *desc_map = bo.map;
 339
 340    if (desc_map == NULL) {
 341       fprintf(ctx->fp, "  interface descriptors unavailable\n");
 342       return;
 343    }
 344
 345    for (int i = 0; i < descriptor_count; i++) {
 346       fprintf(ctx->fp, "descriptor %d: %08x\n", i, descriptor_offset);
 347
 348       ctx_print_group(ctx, desc, desc_addr, desc_map);
 349
 350       gen_field_iterator_init(&iter, desc, desc_map, 0, false);
 351       uint64_t ksp = 0;
 352       uint32_t sampler_offset = 0, sampler_count = 0;
 353       uint32_t binding_table_offset = 0, binding_entry_count = 0;
 354       while (gen_field_iterator_next(&iter)) {
 355          if (strcmp(iter.name, "Kernel Start Pointer") == 0) {
 356             ksp = strtoll(iter.value, NULL, 16);
 357          } else if (strcmp(iter.name, "Sampler State Pointer") == 0) {
 358             sampler_offset = strtol(iter.value, NULL, 16);
 359          } else if (strcmp(iter.name, "Sampler Count") == 0) {
 360             sampler_count = strtol(iter.value, NULL, 10);
 361          } else if (strcmp(iter.name, "Binding Table Pointer") == 0) {
 362             binding_table_offset = strtol(iter.value, NULL, 16);
 363          } else if (strcmp(iter.name, "Binding Table Entry Count") == 0) {
 364             binding_entry_count = strtol(iter.value, NULL, 10);
 365          }
 366       }
 367
 368       ctx_disassemble_program(ctx, ksp, "compute shader");
 369       printf("\n");
 370
 371       dump_samplers(ctx, sampler_offset, sampler_count);
 372       dump_binding_table(ctx, binding_table_offset, binding_entry_count);
 373
 374       desc_map += desc->dw_length;
 375       desc_addr += desc->dw_length * 4;
 376    }
 377 }
 378
 379 static void
 380 handle_3dstate_vertex_buffers(struct gen_batch_decode_ctx *ctx,
 381                               const uint32_t *p)
 382 {
 383    struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
 384    struct gen_group *vbs = gen_spec_find_struct(ctx->spec, "VERTEX_BUFFER_STATE");
 385
 386    struct gen_batch_decode_bo vb = {};
 387    uint32_t vb_size = 0;
 388    int index = -1;
 389    int pitch = -1;
 390    bool ready = false;
 391
 392    struct gen_field_iterator iter;
 393    gen_field_iterator_init(&iter, inst, p, 0, false);
 394    while (gen_field_iterator_next(&iter)) {
 395       if (iter.struct_desc != vbs)
 396          continue;
 397
 398       struct gen_field_iterator vbs_iter;
 399       gen_field_iterator_init(&vbs_iter, vbs, &iter.p[iter.start_bit / 32], 0, false);
 400       while (gen_field_iterator_next(&vbs_iter)) {
 401          if (strcmp(vbs_iter.name, "Vertex Buffer Index") == 0) {
 402             index = vbs_iter.raw_value;
 403          } else if (strcmp(vbs_iter.name, "Buffer Pitch") == 0) {
 404             pitch = vbs_iter.raw_value;
 405          } else if (strcmp(vbs_iter.name, "Buffer Starting Address") == 0) {
 406             vb = ctx_get_bo(ctx, vbs_iter.raw_value);
 407          } else if (strcmp(vbs_iter.name, "Buffer Size") == 0) {
 408             vb_size = vbs_iter.raw_value;
 409             ready = true;
 410          } else if (strcmp(vbs_iter.name, "End Address") == 0) {
 411             if (vb.map && vbs_iter.raw_value >= vb.addr)
 412                vb_size = (vbs_iter.raw_value + 1) - vb.addr;
 413             else
 414                vb_size = 0;
 415             ready = true;
 416          }
 417
 418          if (!ready)
 419             continue;
 420
 421          fprintf(ctx->fp, "vertex buffer %d, size %d\n", index, vb_size);
 422
 423          if (vb.map == NULL) {
 424             fprintf(ctx->fp, "  buffer contents unavailable\n");
 425             continue;
 426          }
 427
 428          if (vb.map == 0 || vb_size == 0)
 429             continue;
 430
 431          ctx_print_buffer(ctx, vb, vb_size, pitch, ctx->max_vbo_decoded_lines);
 432
 433          vb.map = NULL;
 434          vb_size = 0;
 435          index = -1;
 436          pitch = -1;
 437          ready = false;
 438       }
 439    }
 440 }
 441
 442 static void
 443 handle_3dstate_index_buffer(struct gen_batch_decode_ctx *ctx,
 444                             const uint32_t *p)
 445 {
 446    struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
 447
 448    struct gen_batch_decode_bo ib = {};
 449    uint32_t ib_size = 0;
 450    uint32_t format = 0;
 451
 452    struct gen_field_iterator iter;
 453    gen_field_iterator_init(&iter, inst, p, 0, false);
 454    while (gen_field_iterator_next(&iter)) {
 455       if (strcmp(iter.name, "Index Format") == 0) {
 456          format = iter.raw_value;
 457       } else if (strcmp(iter.name, "Buffer Starting Address") == 0) {
 458          ib = ctx_get_bo(ctx, iter.raw_value);
 459       } else if (strcmp(iter.name, "Buffer Size") == 0) {
 460          ib_size = iter.raw_value;
 461       }
 462    }
 463
 464    if (ib.map == NULL) {
 465       fprintf(ctx->fp, "  buffer contents unavailable\n");
 466       return;
 467    }
 468
 469    const void *m = ib.map;
 470    const void *ib_end = ib.map + MIN2(ib.size, ib_size);
 471    for (int i = 0; m < ib_end && i < 10; i++) {
 472       switch (format) {
 473       case 0:
 474          fprintf(ctx->fp, "%3d ", *(uint8_t *)m);
 475          m += 1;
 476          break;
 477       case 1:
 478          fprintf(ctx->fp, "%3d ", *(uint16_t *)m);
 479          m += 2;
 480          break;
 481       case 2:
 482          fprintf(ctx->fp, "%3d ", *(uint32_t *)m);
 483          m += 4;
 484          break;
 485       }
 486    }
 487
 488    if (m < ib_end)
 489       fprintf(ctx->fp, "...");
 490    fprintf(ctx->fp, "\n");
 491 }
 492
 493 static void
 494 decode_single_ksp(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
 495 {
 496    struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
 497
 498    uint64_t ksp = 0;
 499    bool is_simd8 = false; /* vertex shaders on Gen8+ only */
 500    bool is_enabled = true;
 501
 502    struct gen_field_iterator iter;
 503    gen_field_iterator_init(&iter, inst, p, 0, false);
 504    while (gen_field_iterator_next(&iter)) {
 505       if (strcmp(iter.name, "Kernel Start Pointer") == 0) {
 506          ksp = iter.raw_value;
 507       } else if (strcmp(iter.name, "SIMD8 Dispatch Enable") == 0) {
 508          is_simd8 = iter.raw_value;
 509       } else if (strcmp(iter.name, "Dispatch Mode") == 0) {
 510          is_simd8 = strcmp(iter.value, "SIMD8") == 0;
 511       } else if (strcmp(iter.name, "Dispatch Enable") == 0) {
 512          is_simd8 = strcmp(iter.value, "SIMD8") == 0;
 513       } else if (strcmp(iter.name, "Enable") == 0) {
 514          is_enabled = iter.raw_value;
 515       }
 516    }
 517
 518    const char *type =
 519       strcmp(inst->name,   "VS_STATE") == 0 ? "vertex shader" :
 520       strcmp(inst->name,   "GS_STATE") == 0 ? "geometry shader" :
 521       strcmp(inst->name,   "SF_STATE") == 0 ? "strips and fans shader" :
 522       strcmp(inst->name, "CLIP_STATE") == 0 ? "clip shader" :
 523       strcmp(inst->name, "3DSTATE_DS") == 0 ? "tessellation evaluation shader" :
 524       strcmp(inst->name, "3DSTATE_HS") == 0 ? "tessellation control shader" :
 525       strcmp(inst->name, "3DSTATE_VS") == 0 ? (is_simd8 ? "SIMD8 vertex shader" : "vec4 vertex shader") :
 526       strcmp(inst->name, "3DSTATE_GS") == 0 ? (is_simd8 ? "SIMD8 geometry shader" : "vec4 geometry shader") :
 527       NULL;
 528
 529    if (is_enabled) {
 530       ctx_disassemble_program(ctx, ksp, type);
 531       printf("\n");
 532    }
 533 }
 534
 535 static void
 536 decode_ps_kernels(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
 537 {
 538    struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
 539
 540    uint64_t ksp[3] = {0, 0, 0};
 541    bool enabled[3] = {false, false, false};
 542
 543    struct gen_field_iterator iter;
 544    gen_field_iterator_init(&iter, inst, p, 0, false);
 545    while (gen_field_iterator_next(&iter)) {
 546       if (strncmp(iter.name, "Kernel Start Pointer ",
 547                   strlen("Kernel Start Pointer ")) == 0) {
 548          int idx = iter.name[strlen("Kernel Start Pointer ")] - '0';
 549          ksp[idx] = strtol(iter.value, NULL, 16);
 550       } else if (strcmp(iter.name, "8 Pixel Dispatch Enable") == 0) {
 551          enabled[0] = strcmp(iter.value, "true") == 0;
 552       } else if (strcmp(iter.name, "16 Pixel Dispatch Enable") == 0) {
 553          enabled[1] = strcmp(iter.value, "true") == 0;
 554       } else if (strcmp(iter.name, "32 Pixel Dispatch Enable") == 0) {
 555          enabled[2] = strcmp(iter.value, "true") == 0;
 556       }
 557    }
 558
 559    /* Reorder KSPs to be [8, 16, 32] instead of the hardware order. */
 560    if (enabled[0] + enabled[1] + enabled[2] == 1) {
 561       if (enabled[1]) {
 562          ksp[1] = ksp[0];
 563          ksp[0] = 0;
 564       } else if (enabled[2]) {
 565          ksp[2] = ksp[0];
 566          ksp[0] = 0;
 567       }
 568    } else {
 569       uint64_t tmp = ksp[1];
 570       ksp[1] = ksp[2];
 571       ksp[2] = tmp;
 572    }
 573
 574    if (enabled[0])
 575       ctx_disassemble_program(ctx, ksp[0], "SIMD8 fragment shader");
 576    if (enabled[1])
 577       ctx_disassemble_program(ctx, ksp[1], "SIMD16 fragment shader");
 578    if (enabled[2])
 579       ctx_disassemble_program(ctx, ksp[2], "SIMD32 fragment shader");
 580    fprintf(ctx->fp, "\n");
 581 }
 582
 583 static void
 584 decode_3dstate_constant(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
 585 {
 586    struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
 587    struct gen_group *body =
 588       gen_spec_find_struct(ctx->spec, "3DSTATE_CONSTANT_BODY");
 589
 590    uint32_t read_length[4] = {0};
 591    uint64_t read_addr[4];
 592
 593    struct gen_field_iterator outer;
 594    gen_field_iterator_init(&outer, inst, p, 0, false);
 595    while (gen_field_iterator_next(&outer)) {
 596       if (outer.struct_desc != body)
 597          continue;
 598
 599       struct gen_field_iterator iter;
 600       gen_field_iterator_init(&iter, body, &outer.p[outer.start_bit / 32],
 601                               0, false);
 602
 603       while (gen_field_iterator_next(&iter)) {
 604          int idx;
 605          if (sscanf(iter.name, "Read Length[%d]", &idx) == 1) {
 606             read_length[idx] = iter.raw_value;
 607          } else if (sscanf(iter.name, "Buffer[%d]", &idx) == 1) {
 608             read_addr[idx] = iter.raw_value;
 609          }
 610       }
 611
 612       for (int i = 0; i < 4; i++) {
 613          if (read_length[i] == 0)
 614             continue;
 615
 616          struct gen_batch_decode_bo buffer = ctx_get_bo(ctx, read_addr[i]);
 617          if (!buffer.map) {
 618             fprintf(ctx->fp, "constant buffer %d unavailable\n", i);
 619             continue;
 620          }
 621
 622          unsigned size = read_length[i] * 32;
 623          fprintf(ctx->fp, "constant buffer %d, size %u\n", i, size);
 624
 625          ctx_print_buffer(ctx, buffer, size, 0, -1);
 626       }
 627    }
 628 }
 629
 630 static void
 631 decode_3dstate_binding_table_pointers(struct gen_batch_decode_ctx *ctx,
 632                                       const uint32_t *p)
 633 {
 634    dump_binding_table(ctx, p[1], -1);
 635 }
 636
 637 static void
 638 decode_3dstate_sampler_state_pointers(struct gen_batch_decode_ctx *ctx,
 639                                       const uint32_t *p)
 640 {
 641    dump_samplers(ctx, p[1], -1);
 642 }
 643
 644 static void
 645 decode_3dstate_sampler_state_pointers_gen6(struct gen_batch_decode_ctx *ctx,
 646                                            const uint32_t *p)
 647 {
 648    dump_samplers(ctx, p[1], -1);
 649    dump_samplers(ctx, p[2], -1);
 650    dump_samplers(ctx, p[3], -1);
 651 }
 652
 653 static bool
 654 str_ends_with(const char *str, const char *end)
 655 {
 656    int offset = strlen(str) - strlen(end);
 657    if (offset < 0)
 658       return false;
 659
 660    return strcmp(str + offset, end) == 0;
 661 }
 662
 663 static void
 664 decode_dynamic_state_pointers(struct gen_batch_decode_ctx *ctx,
 665                               const char *struct_type, const uint32_t *p,
 666                               int count)
 667 {
 668    struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
 669
 670    uint32_t state_offset = 0;
 671
 672    struct gen_field_iterator iter;
 673    gen_field_iterator_init(&iter, inst, p, 0, false);
 674    while (gen_field_iterator_next(&iter)) {
 675       if (str_ends_with(iter.name, "Pointer")) {
 676          state_offset = iter.raw_value;
 677          break;
 678       }
 679    }
 680
 681    uint64_t state_addr = ctx->dynamic_base + state_offset;
 682    struct gen_batch_decode_bo bo = ctx_get_bo(ctx, state_addr);
 683    const void *state_map = bo.map;
 684
 685    if (state_map == NULL) {
 686       fprintf(ctx->fp, "  dynamic %s state unavailable\n", struct_type);
 687       return;
 688    }
 689
 690    struct gen_group *state = gen_spec_find_struct(ctx->spec, struct_type);
 691    if (strcmp(struct_type, "BLEND_STATE") == 0) {
 692       /* Blend states are different from the others because they have a header
 693        * struct called BLEND_STATE which is followed by a variable number of
 694        * BLEND_STATE_ENTRY structs.
 695        */
 696       fprintf(ctx->fp, "%s\n", struct_type);
 697       ctx_print_group(ctx, state, state_addr, state_map);
 698
 699       state_addr += state->dw_length * 4;
 700       state_map += state->dw_length * 4;
 701
 702       struct_type = "BLEND_STATE_ENTRY";
 703       state = gen_spec_find_struct(ctx->spec, struct_type);
 704    }
 705
 706    for (int i = 0; i < count; i++) {
 707       fprintf(ctx->fp, "%s %d\n", struct_type, i);
 708       ctx_print_group(ctx, state, state_addr, state_map);
 709
 710       state_addr += state->dw_length * 4;
 711       state_map += state->dw_length * 4;
 712    }
 713 }
 714
 715 static void
 716 decode_3dstate_viewport_state_pointers_cc(struct gen_batch_decode_ctx *ctx,
 717                                           const uint32_t *p)
 718 {
 719    decode_dynamic_state_pointers(ctx, "CC_VIEWPORT", p, 4);
 720 }
 721
 722 static void
 723 decode_3dstate_viewport_state_pointers_sf_clip(struct gen_batch_decode_ctx *ctx,
 724                                                const uint32_t *p)
 725 {
 726    decode_dynamic_state_pointers(ctx, "SF_CLIP_VIEWPORT", p, 4);
 727 }
 728
 729 static void
 730 decode_3dstate_blend_state_pointers(struct gen_batch_decode_ctx *ctx,
 731                                     const uint32_t *p)
 732 {
 733    decode_dynamic_state_pointers(ctx, "BLEND_STATE", p, 1);
 734 }
 735
 736 static void
 737 decode_3dstate_cc_state_pointers(struct gen_batch_decode_ctx *ctx,
 738                                  const uint32_t *p)
 739 {
 740    decode_dynamic_state_pointers(ctx, "COLOR_CALC_STATE", p, 1);
 741 }
 742
 743 static void
 744 decode_3dstate_scissor_state_pointers(struct gen_batch_decode_ctx *ctx,
 745                                       const uint32_t *p)
 746 {
 747    decode_dynamic_state_pointers(ctx, "SCISSOR_RECT", p, 1);
 748 }
 749
 750 static void
 751 decode_load_register_imm(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
 752 {
 753    struct gen_group *reg = gen_spec_find_register(ctx->spec, p[1]);
 754
 755    if (reg != NULL) {
 756       fprintf(ctx->fp, "register %s (0x%x): 0x%x\n",
 757               reg->name, reg->register_offset, p[2]);
 758       ctx_print_group(ctx, reg, reg->register_offset, &p[2]);
 759    }
 760 }
 761
 762 struct custom_decoder {
 763    const char *cmd_name;
 764    void (*decode)(struct gen_batch_decode_ctx *ctx, const uint32_t *p);
 765 } custom_decoders[] = {
 766    { "STATE_BASE_ADDRESS", handle_state_base_address },
 767    { "MEDIA_INTERFACE_DESCRIPTOR_LOAD", handle_media_interface_descriptor_load },
 768    { "3DSTATE_VERTEX_BUFFERS", handle_3dstate_vertex_buffers },
 769    { "3DSTATE_INDEX_BUFFER", handle_3dstate_index_buffer },
 770    { "3DSTATE_VS", decode_single_ksp },
 771    { "3DSTATE_GS", decode_single_ksp },
 772    { "3DSTATE_DS", decode_single_ksp },
 773    { "3DSTATE_HS", decode_single_ksp },
 774    { "3DSTATE_PS", decode_ps_kernels },
 775    { "3DSTATE_CONSTANT_VS", decode_3dstate_constant },
 776    { "3DSTATE_CONSTANT_GS", decode_3dstate_constant },
 777    { "3DSTATE_CONSTANT_PS", decode_3dstate_constant },
 778    { "3DSTATE_CONSTANT_HS", decode_3dstate_constant },
 779    { "3DSTATE_CONSTANT_DS", decode_3dstate_constant },
 780
 781    { "3DSTATE_BINDING_TABLE_POINTERS_VS", decode_3dstate_binding_table_pointers },
 782    { "3DSTATE_BINDING_TABLE_POINTERS_HS", decode_3dstate_binding_table_pointers },
 783    { "3DSTATE_BINDING_TABLE_POINTERS_DS", decode_3dstate_binding_table_pointers },
 784    { "3DSTATE_BINDING_TABLE_POINTERS_GS", decode_3dstate_binding_table_pointers },
 785    { "3DSTATE_BINDING_TABLE_POINTERS_PS", decode_3dstate_binding_table_pointers },
 786
 787    { "3DSTATE_SAMPLER_STATE_POINTERS_VS", decode_3dstate_sampler_state_pointers },
 788    { "3DSTATE_SAMPLER_STATE_POINTERS_HS", decode_3dstate_sampler_state_pointers },
 789    { "3DSTATE_SAMPLER_STATE_POINTERS_DS", decode_3dstate_sampler_state_pointers },
 790    { "3DSTATE_SAMPLER_STATE_POINTERS_GS", decode_3dstate_sampler_state_pointers },
 791    { "3DSTATE_SAMPLER_STATE_POINTERS_PS", decode_3dstate_sampler_state_pointers },
 792    { "3DSTATE_SAMPLER_STATE_POINTERS", decode_3dstate_sampler_state_pointers_gen6 },
 793
 794    { "3DSTATE_VIEWPORT_STATE_POINTERS_CC", decode_3dstate_viewport_state_pointers_cc },
 795    { "3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP", decode_3dstate_viewport_state_pointers_sf_clip },
 796    { "3DSTATE_BLEND_STATE_POINTERS", decode_3dstate_blend_state_pointers },
 797    { "3DSTATE_CC_STATE_POINTERS", decode_3dstate_cc_state_pointers },
 798    { "3DSTATE_SCISSOR_STATE_POINTERS", decode_3dstate_scissor_state_pointers },
 799    { "MI_LOAD_REGISTER_IMM", decode_load_register_imm }
 800 };
 801
 802 void
 803 gen_print_batch(struct gen_batch_decode_ctx *ctx,
 804                 const uint32_t *batch, uint32_t batch_size,
 805                 uint64_t batch_addr)
 806 {
 807    const uint32_t *p, *end = batch + batch_size / sizeof(uint32_t);
 808    int length;
 809    struct gen_group *inst;
 810
 811    for (p = batch; p < end; p += length) {
 812       inst = gen_ctx_find_instruction(ctx, p);
 813       length = gen_group_get_length(inst, p);
 814       assert(inst == NULL || length > 0);
 815       length = MAX2(1, length);
 816
 817       const char *reset_color = ctx->flags & GEN_BATCH_DECODE_IN_COLOR ? NORMAL : "";
 818
 819       uint64_t offset;
 820       if (ctx->flags & GEN_BATCH_DECODE_OFFSETS)
 821          offset = batch_addr + ((char *)p - (char *)batch);
 822       else
 823          offset = 0;
 824
 825       if (inst == NULL) {
 826          fprintf(ctx->fp, "%s0x%08"PRIx64": unknown instruction %08x%s\n",
 827                  (ctx->flags & GEN_BATCH_DECODE_IN_COLOR) ? RED_COLOR : "",
 828                  offset, p[0], reset_color);
 829          continue;
 830       }
 831
 832       const char *color;
 833       const char *inst_name = gen_group_get_name(inst);
 834       if (ctx->flags & GEN_BATCH_DECODE_IN_COLOR) {
 835          reset_color = NORMAL;
 836          if (ctx->flags & GEN_BATCH_DECODE_FULL) {
 837             if (strcmp(inst_name, "MI_BATCH_BUFFER_START") == 0 ||
 838                 strcmp(inst_name, "MI_BATCH_BUFFER_END") == 0)
 839                color = GREEN_HEADER;
 840             else
 841                color = BLUE_HEADER;
 842          } else {
 843             color = NORMAL;
 844          }
 845       } else {
 846          color = "";
 847          reset_color = "";
 848       }
 849
 850       fprintf(ctx->fp, "%s0x%08"PRIx64":  0x%08x:  %-80s%s\n",
 851               color, offset, p[0], inst_name, reset_color);
 852
 853       if (ctx->flags & GEN_BATCH_DECODE_FULL) {
 854          ctx_print_group(ctx, inst, offset, p);
 855
 856          for (int i = 0; i < ARRAY_SIZE(custom_decoders); i++) {
 857             if (strcmp(inst_name, custom_decoders[i].cmd_name) == 0) {
 858                custom_decoders[i].decode(ctx, p);
 859                break;
 860             }
 861          }
 862       }
 863
 864       if (strcmp(inst_name, "MI_BATCH_BUFFER_START") == 0) {
 865          struct gen_batch_decode_bo next_batch = {};
 866          bool second_level;
 867          struct gen_field_iterator iter;
 868          gen_field_iterator_init(&iter, inst, p, 0, false);
 869          while (gen_field_iterator_next(&iter)) {
 870             if (strcmp(iter.name, "Batch Buffer Start Address") == 0) {
 871                next_batch = ctx_get_bo(ctx, iter.raw_value);
 872             } else if (strcmp(iter.name, "Second Level Batch Buffer") == 0) {
 873                second_level = iter.raw_value;
 874             }
 875          }
 876
 877          if (next_batch.map == NULL) {
 878             fprintf(ctx->fp, "Secondary batch at 0x%08"PRIx64" unavailable\n",
 879                     next_batch.addr);
 880          } else {
 881             gen_print_batch(ctx, next_batch.map, next_batch.size,
 882                             next_batch.addr);
 883          }
 884          if (second_level) {
 885             /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" set acts
 886              * like a subroutine call.  Commands that come afterwards get
 887              * processed once the 2nd level batch buffer returns with
 888              * MI_BATCH_BUFFER_END.
 889              */
 890             continue;
 891          } else {
 892             /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" unset acts
 893              * like a goto.  Nothing after it will ever get processed.  In
 894              * order to prevent the recursion from growing, we just reset the
 895              * loop and continue;
 896              */
 897             break;
 898          }
 899       } else if (strcmp(inst_name, "MI_BATCH_BUFFER_END") == 0) {
 900          break;
 901       }
 902    }
 903 }