intel/tools: Convert aubinator over to the common framework

Reviewed-by: Lionel Landwerlin <[email protected]>

3 files changed, 33 insertions, 690 deletions

diff --git a/src/intel/Makefile.tools.am b/src/intel/Makefile.tools.am
index 9332ae3e7d3..944ee19805c 100644
--- a/src/intel/Makefile.tools.am
+++ b/src/intel/Makefile.tools.am
@@ -26,6 +26,7 @@ noinst_PROGRAMS += \
 tools_aubinator_SOURCES = \
 	tools/aubinator.c \
 	tools/disasm.c \
+	tools/gen_batch_decoder.c \
 	tools/gen_disasm.h \
 	tools/intel_aub.h
 
diff --git a/src/intel/tools/aubinator.c b/src/intel/tools/aubinator.c
index 4ae70ad4fbc..ed7446cf1ee 100644
--- a/src/intel/tools/aubinator.c
+++ b/src/intel/tools/aubinator.c
@@ -64,8 +64,8 @@ static enum { COLOR_AUTO, COLOR_ALWAYS, COLOR_NEVER } option_color;
 
 uint16_t pci_id = 0;
 char *input_file = NULL, *xml_path = NULL;
-struct gen_spec *spec;
-struct gen_disasm *disasm;
+struct gen_device_info devinfo;
+struct gen_batch_decode_ctx batch_ctx;
 
 uint64_t gtt_size, gtt_end;
 void *gtt;
@@ -95,683 +95,6 @@ valid_offset(uint32_t offset)
    return offset < gtt_end;
 }
 
-static void
-decode_group(struct gen_group *strct, const uint32_t *p, int starting_dword)
-{
-   uint64_t offset = option_print_offsets ? (void *) p - gtt : 0;
-
-   gen_print_group(outfile, strct, offset, p, 0, option_color == COLOR_ALWAYS);
-}
-
-static void
-dump_binding_table(struct gen_spec *spec, uint32_t offset)
-{
-   uint32_t *pointers, i;
-   uint64_t start;
-   struct gen_group *surface_state;
-
-   surface_state = gen_spec_find_struct(spec, "RENDER_SURFACE_STATE");
-   if (surface_state == NULL) {
-      fprintf(outfile, "did not find RENDER_SURFACE_STATE info\n");
-      return;
-   }
-
-   start = surface_state_base + offset;
-   pointers = gtt + start;
-   for (i = 0; i < 16; i++) {
-      if (pointers[i] == 0)
-         continue;
-      start = pointers[i] + surface_state_base;
-      if (!valid_offset(start)) {
-         fprintf(outfile, "pointer %u: %08x <not valid>\n",
-                 i, pointers[i]);
-         continue;
-      } else {
-         fprintf(outfile, "pointer %u: %08x\n", i, pointers[i]);
-      }
-
-      decode_group(surface_state, gtt + start, 0);
-   }
-}
-
-static void
-handle_3dstate_index_buffer(struct gen_spec *spec, uint32_t *p)
-{
-   void *start;
-   uint32_t length, i, type, size;
-
-   start = gtt + p[2];
-   type = (p[1] >> 8) & 3;
-   size = 1 << type;
-   length = p[4] / size;
-   if (length > 10)
-      length = 10;
-
-   fprintf(outfile, "\t");
-
-   for (i = 0; i < length; i++) {
-      switch (type) {
-      case 0:
-         fprintf(outfile, "%3d ", ((uint8_t *)start)[i]);
-         break;
-      case 1:
-         fprintf(outfile, "%3d ", ((uint16_t *)start)[i]);
-         break;
-      case 2:
-         fprintf(outfile, "%3d ", ((uint32_t *)start)[i]);
-         break;
-      }
-   }
-   if (length < p[4] / size)
-      fprintf(outfile, "...\n");
-   else
-      fprintf(outfile, "\n");
-}
-
-static inline uint64_t
-get_address(struct gen_spec *spec, uint32_t *p)
-{
-   /* Addresses are always guaranteed to be page-aligned and sometimes
-    * hardware packets have extra stuff stuffed in the bottom 12 bits.
-    */
-   uint64_t addr = p[0] & ~0xfffu;
-
-   if (gen_spec_get_gen(spec) >= gen_make_gen(8,0)) {
-      /* On Broadwell and above, we have 48-bit addresses which consume two
-       * dwords.  Some packets require that these get stored in a "canonical
-       * form" which means that bit 47 is sign-extended through the upper
-       * bits. In order to correctly handle those aub dumps, we need to mask
-       * off the top 16 bits.
-       */
-      addr |= ((uint64_t)p[1] & 0xffff) << 32;
-   }
-
-   return addr;
-}
-
-static inline uint64_t
-get_offset(uint32_t *p, uint32_t start, uint32_t end)
-{
-   assert(start <= end);
-   assert(end < 64);
-
-   uint64_t mask = (~0ull >> (64 - (end - start + 1))) << start;
-
-   uint64_t offset = p[0];
-   if (end >= 32)
-      offset |= (uint64_t) p[1] << 32;
-
-   return offset & mask;
-}
-
-static void
-handle_state_base_address(struct gen_spec *spec, uint32_t *p)
-{
-   if (gen_spec_get_gen(spec) >= gen_make_gen(8,0)) {
-      if (p[1] & 1)
-         general_state_base = get_address(spec, &p[1]);
-      if (p[4] & 1)
-         surface_state_base = get_address(spec, &p[4]);
-      if (p[6] & 1)
-         dynamic_state_base = get_address(spec, &p[6]);
-      if (p[10] & 1)
-         instruction_base = get_address(spec, &p[10]);
-      if (p[15] & 1)
-         instruction_bound = p[15] & 0xfff;
-   } else {
-      if (p[2] & 1)
-         surface_state_base = get_address(spec, &p[2]);
-      if (p[3] & 1)
-         dynamic_state_base = get_address(spec, &p[3]);
-      if (p[5] & 1)
-         instruction_base = get_address(spec, &p[5]);
-      if (p[9] & 1)
-         instruction_bound = get_address(spec, &p[9]);
-   }
-}
-
-static void
-dump_samplers(struct gen_spec *spec, uint32_t offset)
-{
-   uint32_t i;
-   uint64_t start;
-   struct gen_group *sampler_state;
-
-   sampler_state = gen_spec_find_struct(spec, "SAMPLER_STATE");
-
-   start = dynamic_state_base + offset;
-   for (i = 0; i < 4; i++) {
-      fprintf(outfile, "sampler state %d\n", i);
-      decode_group(sampler_state, gtt + start + i * 16, 0);
-   }
-}
-
-static void
-handle_media_interface_descriptor_load(struct gen_spec *spec, uint32_t *p)
-{
-   int i, length = p[2] / 32;
-   struct gen_group *descriptor_structure;
-   uint32_t *descriptors;
-   uint64_t start;
-   struct brw_instruction *insns;
-
-   descriptor_structure =
-      gen_spec_find_struct(spec, "INTERFACE_DESCRIPTOR_DATA");
-   if (descriptor_structure == NULL) {
-      fprintf(outfile, "did not find INTERFACE_DESCRIPTOR_DATA info\n");
-      return;
-   }
-
-   start = dynamic_state_base + p[3];
-   descriptors = gtt + start;
-   for (i = 0; i < length; i++, descriptors += 8) {
-      fprintf(outfile, "descriptor %u: %08x\n", i, *descriptors);
-      decode_group(descriptor_structure, descriptors, 0);
-
-      start = instruction_base + descriptors[0];
-      if (!valid_offset(start)) {
-         fprintf(outfile, "kernel: %08"PRIx64" <not valid>\n", start);
-         continue;
-      } else {
-         fprintf(outfile, "kernel: %08"PRIx64"\n", start);
-      }
-
-      insns = (struct brw_instruction *) (gtt + start);
-      gen_disasm_disassemble(disasm, insns, 0, stdout);
-
-      dump_samplers(spec, descriptors[3] & ~0x1f);
-      dump_binding_table(spec, descriptors[4] & ~0x1f);
-   }
-}
-
-/* Heuristic to determine whether a uint32_t is probably actually a float
- * (http://stackoverflow.com/a/2953466)
- */
-
-static bool
-probably_float(uint32_t bits)
-{
-   int exp = ((bits & 0x7f800000U) >> 23) - 127;
-   uint32_t mant = bits & 0x007fffff;
-
-   /* +- 0.0 */
-   if (exp == -127 && mant == 0)
-      return true;
-
-   /* +- 1 billionth to 1 billion */
-   if (-30 <= exp && exp <= 30)
-      return true;
-
-   /* some value with only a few binary digits */
-   if ((mant & 0x0000ffff) == 0)
-      return true;
-
-   return false;
-}
-
-static void
-handle_3dstate_vertex_buffers(struct gen_spec *spec, uint32_t *p)
-{
-   uint32_t *end, *s, *dw, *dwend;
-   uint64_t offset;
-   int n, i, count, stride;
-
-   end = (p[0] & 0xff) + p + 2;
-   for (s = &p[1], n = 0; s < end; s += 4, n++) {
-      if (gen_spec_get_gen(spec) >= gen_make_gen(8, 0)) {
-         offset = *(uint64_t *) &s[1];
-         dwend = gtt + offset + s[3];
-      } else {
-         offset = s[1];
-         dwend = gtt + s[2] + 1;
-      }
-
-      stride = field(s[0], 0, 11);
-      count = 0;
-      fprintf(outfile, "vertex buffer %d, size %d\n", n, s[3]);
-      for (dw = gtt + offset, i = 0; dw < dwend && i < 256; dw++) {
-         if (count == 0 && count % (8 * 4) == 0)
-            fprintf(outfile, "  ");
-
-         if (probably_float(*dw))
-            fprintf(outfile, "  %8.2f", *(float *) dw);
-         else
-            fprintf(outfile, "  0x%08x", *dw);
-
-         i++;
-         count += 4;
-
-         if (count == stride) {
-            fprintf(outfile, "\n");
-            count = 0;
-         } else if (count % (8 * 4) == 0) {
-            fprintf(outfile, "\n");
-         } else {
-            fprintf(outfile, " ");
-         }
-      }
-      if (count > 0 && count % (8 * 4) != 0)
-         fprintf(outfile, "\n");
-   }
-}
-
-static void
-handle_3dstate_vs(struct gen_spec *spec, uint32_t *p)
-{
-   uint64_t start;
-   struct brw_instruction *insns;
-   int vs_enable;
-
-   if (gen_spec_get_gen(spec) >= gen_make_gen(8, 0)) {
-      start = get_offset(&p[1], 6, 63);
-      vs_enable = p[7] & 1;
-   } else {
-      start = get_offset(&p[1], 6, 31);
-      vs_enable = p[5] & 1;
-   }
-
-   if (vs_enable) {
-      fprintf(outfile, "instruction_base %08"PRIx64", start %08"PRIx64"\n",
-              instruction_base, start);
-
-      insns = (struct brw_instruction *) (gtt + instruction_base + start);
-      gen_disasm_disassemble(disasm, insns, 0, stdout);
-   }
-}
-
-static void
-handle_3dstate_hs(struct gen_spec *spec, uint32_t *p)
-{
-   uint64_t start;
-   struct brw_instruction *insns;
-   int hs_enable;
-
-   if (gen_spec_get_gen(spec) >= gen_make_gen(8, 0)) {
-      start = get_offset(&p[3], 6, 63);
-   } else {
-      start = get_offset(&p[3], 6, 31);
-   }
-
-   hs_enable = p[2] & 0x80000000;
-
-   if (hs_enable) {
-      fprintf(outfile, "instruction_base %08"PRIx64", start %08"PRIx64"\n",
-              instruction_base, start);
-
-      insns = (struct brw_instruction *) (gtt + instruction_base + start);
-      gen_disasm_disassemble(disasm, insns, 0, stdout);
-   }
-}
-
-static void
-handle_3dstate_constant(struct gen_spec *spec, uint32_t *p)
-{
-   int i, j, length;
-   uint32_t *dw;
-   float *f;
-
-   for (i = 0; i < 4; i++) {
-      length = (p[1 + i / 2] >> (i & 1) * 16) & 0xffff;
-      f = (float *) (gtt + p[3 + i * 2] + dynamic_state_base);
-      dw = (uint32_t *) f;
-      for (j = 0; j < length * 8; j++) {
-         if (probably_float(dw[j]))
-            fprintf(outfile, "  %04.3f", f[j]);
-         else
-            fprintf(outfile, "  0x%08x", dw[j]);
-
-         if ((j & 7) == 7)
-            fprintf(outfile, "\n");
-      }
-   }
-}
-
-static void
-handle_3dstate_ps(struct gen_spec *spec, uint32_t *p)
-{
-   uint32_t mask = ~((1 << 6) - 1);
-   uint64_t start;
-   struct brw_instruction *insns;
-   static const char unused[] = "unused";
-   static const char *pixel_type[3] = {"8 pixel", "16 pixel", "32 pixel"};
-   const char *k0, *k1, *k2;
-   uint32_t k_mask, k1_offset, k2_offset;
-
-   if (gen_spec_get_gen(spec) >= gen_make_gen(8, 0)) {
-      k_mask = p[6] & 7;
-      k1_offset = 8;
-      k2_offset = 10;
-   } else {
-      k_mask = p[4] & 7;
-      k1_offset = 6;
-      k2_offset = 7;
-   }
-
-#define DISPATCH_8 1
-#define DISPATCH_16 2
-#define DISPATCH_32 4
-
-   switch (k_mask) {
-   case DISPATCH_8:
-      k0 = pixel_type[0];
-      k1 = unused;
-      k2 = unused;
-      break;
-   case DISPATCH_16:
-      k0 = pixel_type[1];
-      k1 = unused;
-      k2 = unused;
-      break;
-   case DISPATCH_8 | DISPATCH_16:
-      k0 = pixel_type[0];
-      k1 = unused;
-      k2 = pixel_type[1];
-      break;
-   case DISPATCH_32:
-      k0 = pixel_type[2];
-      k1 = unused;
-      k2 = unused;
-      break;
-   case DISPATCH_16 | DISPATCH_32:
-      k0 = unused;
-      k1 = pixel_type[2];
-      k2 = pixel_type[1];
-      break;
-   case DISPATCH_8 | DISPATCH_16 | DISPATCH_32:
-      k0 = pixel_type[0];
-      k1 = pixel_type[2];
-      k2 = pixel_type[1];
-      break;
-   default:
-      k0 = unused;
-      k1 = unused;
-      k2 = unused;
-      break;
-   }
-
-   start = instruction_base + (p[1] & mask);
-   fprintf(outfile, "  Kernel[0] %s\n", k0);
-   if (k0 != unused) {
-      insns = (struct brw_instruction *) (gtt + start);
-      gen_disasm_disassemble(disasm, insns, 0, stdout);
-   }
-
-   start = instruction_base + (p[k1_offset] & mask);
-   fprintf(outfile, "  Kernel[1] %s\n", k1);
-   if (k1 != unused) {
-      insns = (struct brw_instruction *) (gtt + start);
-      gen_disasm_disassemble(disasm, insns, 0, stdout);
-   }
-
-   start = instruction_base + (p[k2_offset] & mask);
-   fprintf(outfile, "  Kernel[2] %s\n", k2);
-   if (k2 != unused) {
-      insns = (struct brw_instruction *) (gtt + start);
-      gen_disasm_disassemble(disasm, insns, 0, stdout);
-   }
-}
-
-static void
-handle_3dstate_binding_table_pointers(struct gen_spec *spec, uint32_t *p)
-{
-   dump_binding_table(spec, p[1]);
-}
-
-static void
-handle_3dstate_sampler_state_pointers(struct gen_spec *spec, uint32_t *p)
-{
-   dump_samplers(spec, p[1]);
-}
-
-static void
-handle_3dstate_sampler_state_pointers_gen6(struct gen_spec *spec, uint32_t *p)
-{
-   dump_samplers(spec, p[1]);
-   dump_samplers(spec, p[2]);
-   dump_samplers(spec, p[3]);
-}
-
-static void
-handle_3dstate_viewport_state_pointers_cc(struct gen_spec *spec, uint32_t *p)
-{
-   uint64_t start;
-   struct gen_group *cc_viewport;
-
-   cc_viewport = gen_spec_find_struct(spec, "CC_VIEWPORT");
-
-   start = dynamic_state_base + (p[1] & ~0x1fu);
-   for (uint32_t i = 0; i < 4; i++) {
-      fprintf(outfile, "viewport %d\n", i);
-      decode_group(cc_viewport, gtt + start + i * 8, 0);
-   }
-}
-
-static void
-handle_3dstate_viewport_state_pointers_sf_clip(struct gen_spec *spec,
-                                               uint32_t *p)
-{
-   uint64_t start;
-   struct gen_group *sf_clip_viewport;
-
-   sf_clip_viewport = gen_spec_find_struct(spec, "SF_CLIP_VIEWPORT");
-
-   start = dynamic_state_base + (p[1] & ~0x3fu);
-   for (uint32_t i = 0; i < 4; i++) {
-      fprintf(outfile, "viewport %d\n", i);
-      decode_group(sf_clip_viewport, gtt + start + i * 64, 0);
-   }
-}
-
-static void
-handle_3dstate_blend_state_pointers(struct gen_spec *spec, uint32_t *p)
-{
-   uint64_t start;
-   struct gen_group *blend_state;
-
-   blend_state = gen_spec_find_struct(spec, "BLEND_STATE");
-
-   start = dynamic_state_base + (p[1] & ~0x3fu);
-   decode_group(blend_state, gtt + start, 0);
-}
-
-static void
-handle_3dstate_cc_state_pointers(struct gen_spec *spec, uint32_t *p)
-{
-   uint64_t start;
-   struct gen_group *cc_state;
-
-   cc_state = gen_spec_find_struct(spec, "COLOR_CALC_STATE");
-
-   start = dynamic_state_base + (p[1] & ~0x3fu);
-   decode_group(cc_state, gtt + start, 0);
-}
-
-static void
-handle_3dstate_scissor_state_pointers(struct gen_spec *spec, uint32_t *p)
-{
-   uint64_t start;
-   struct gen_group *scissor_rect;
-
-   scissor_rect = gen_spec_find_struct(spec, "SCISSOR_RECT");
-
-   start = dynamic_state_base + (p[1] & ~0x1fu);
-   decode_group(scissor_rect, gtt + start, 0);
-}
-
-static void
-handle_load_register_imm(struct gen_spec *spec, uint32_t *p)
-{
-   struct gen_group *reg = gen_spec_find_register(spec, p[1]);
-
-   if (reg != NULL) {
-      fprintf(outfile, "register %s (0x%x): 0x%x\n",
-              reg->name, reg->register_offset, p[2]);
-      decode_group(reg, &p[2], 0);
-   }
-}
-
-#define ARRAY_LENGTH(a) (sizeof (a) / sizeof (a)[0])
-
-#define STATE_BASE_ADDRESS                  0x61010000
-
-#define MEDIA_INTERFACE_DESCRIPTOR_LOAD     0x70020000
-
-#define _3DSTATE_INDEX_BUFFER               0x780a0000
-#define _3DSTATE_VERTEX_BUFFERS             0x78080000
-
-#define _3DSTATE_VS                         0x78100000
-#define _3DSTATE_GS                         0x78110000
-#define _3DSTATE_HS                         0x781b0000
-#define _3DSTATE_DS                         0x781d0000
-
-#define _3DSTATE_CONSTANT_VS                0x78150000
-#define _3DSTATE_CONSTANT_GS                0x78160000
-#define _3DSTATE_CONSTANT_PS                0x78170000
-#define _3DSTATE_CONSTANT_HS                0x78190000
-#define _3DSTATE_CONSTANT_DS                0x781A0000
-
-#define _3DSTATE_PS                         0x78200000
-
-#define _3DSTATE_BINDING_TABLE_POINTERS_VS  0x78260000
-#define _3DSTATE_BINDING_TABLE_POINTERS_HS  0x78270000
-#define _3DSTATE_BINDING_TABLE_POINTERS_DS  0x78280000
-#define _3DSTATE_BINDING_TABLE_POINTERS_GS  0x78290000
-#define _3DSTATE_BINDING_TABLE_POINTERS_PS  0x782a0000
-
-#define _3DSTATE_SAMPLER_STATE_POINTERS_VS  0x782b0000
-#define _3DSTATE_SAMPLER_STATE_POINTERS_GS  0x782e0000
-#define _3DSTATE_SAMPLER_STATE_POINTERS_PS  0x782f0000
-
-#define _3DSTATE_SAMPLER_STATE_POINTERS     0x78020000
-
-#define _3DSTATE_VIEWPORT_STATE_POINTERS_CC 0x78230000
-#define _3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP 0x78210000
-#define _3DSTATE_BLEND_STATE_POINTERS       0x78240000
-#define _3DSTATE_CC_STATE_POINTERS          0x780e0000
-#define _3DSTATE_SCISSOR_STATE_POINTERS     0x780f0000
-
-#define _MI_LOAD_REGISTER_IMM               0x11000000
-
-struct custom_handler {
-   uint32_t opcode;
-   void (*handle)(struct gen_spec *spec, uint32_t *p);
-} custom_handlers[] = {
-   { STATE_BASE_ADDRESS, handle_state_base_address },
-   { MEDIA_INTERFACE_DESCRIPTOR_LOAD, handle_media_interface_descriptor_load },
-   { _3DSTATE_VERTEX_BUFFERS, handle_3dstate_vertex_buffers },
-   { _3DSTATE_INDEX_BUFFER, handle_3dstate_index_buffer },
-   { _3DSTATE_VS, handle_3dstate_vs },
-   { _3DSTATE_GS, handle_3dstate_vs },
-   { _3DSTATE_DS, handle_3dstate_vs },
-   { _3DSTATE_HS, handle_3dstate_hs },
-   { _3DSTATE_CONSTANT_VS, handle_3dstate_constant },
-   { _3DSTATE_CONSTANT_GS, handle_3dstate_constant },
-   { _3DSTATE_CONSTANT_PS, handle_3dstate_constant },
-   { _3DSTATE_CONSTANT_HS, handle_3dstate_constant },
-   { _3DSTATE_CONSTANT_DS, handle_3dstate_constant },
-   { _3DSTATE_PS, handle_3dstate_ps },
-
-   { _3DSTATE_BINDING_TABLE_POINTERS_VS, handle_3dstate_binding_table_pointers },
-   { _3DSTATE_BINDING_TABLE_POINTERS_HS, handle_3dstate_binding_table_pointers },
-   { _3DSTATE_BINDING_TABLE_POINTERS_DS, handle_3dstate_binding_table_pointers },
-   { _3DSTATE_BINDING_TABLE_POINTERS_GS, handle_3dstate_binding_table_pointers },
-   { _3DSTATE_BINDING_TABLE_POINTERS_PS, handle_3dstate_binding_table_pointers },
-
-   { _3DSTATE_SAMPLER_STATE_POINTERS_VS, handle_3dstate_sampler_state_pointers },
-   { _3DSTATE_SAMPLER_STATE_POINTERS_GS, handle_3dstate_sampler_state_pointers },
-   { _3DSTATE_SAMPLER_STATE_POINTERS_PS, handle_3dstate_sampler_state_pointers },
-   { _3DSTATE_SAMPLER_STATE_POINTERS, handle_3dstate_sampler_state_pointers_gen6 },
-
-   { _3DSTATE_VIEWPORT_STATE_POINTERS_CC, handle_3dstate_viewport_state_pointers_cc },
-   { _3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP, handle_3dstate_viewport_state_pointers_sf_clip },
-   { _3DSTATE_BLEND_STATE_POINTERS, handle_3dstate_blend_state_pointers },
-   { _3DSTATE_CC_STATE_POINTERS, handle_3dstate_cc_state_pointers },
-   { _3DSTATE_SCISSOR_STATE_POINTERS, handle_3dstate_scissor_state_pointers },
-   { _MI_LOAD_REGISTER_IMM, handle_load_register_imm }
-};
-
-static void
-parse_commands(struct gen_spec *spec, uint32_t *cmds, int size, int engine)
-{
-   uint32_t *p, *end = cmds + size / 4;
-   int length, i;
-   struct gen_group *inst;
-
-   for (p = cmds; p < end; p += length) {
-      inst = gen_spec_find_instruction(spec, p);
-      length = gen_group_get_length(inst, p);
-      assert(inst == NULL || length > 0);
-      length = MAX2(1, length);
-      if (inst == NULL) {
-         fprintf(outfile, "unknown instruction %08x\n", p[0]);
-         continue;
-      }
-
-      const char *color, *reset_color = NORMAL;
-      uint64_t offset;
-
-      if (option_full_decode) {
-         if ((p[0] & 0xffff0000) == AUB_MI_BATCH_BUFFER_START ||
-             (p[0] & 0xffff0000) == AUB_MI_BATCH_BUFFER_END)
-            color = GREEN_HEADER;
-         else
-            color = BLUE_HEADER;
-      } else
-         color = NORMAL;
-
-      if (option_color == COLOR_NEVER) {
-         color = "";
-         reset_color = "";
-      }
-
-      if (option_print_offsets)
-         offset = (void *) p - gtt;
-      else
-         offset = 0;
-
-      fprintf(outfile, "%s0x%08"PRIx64":  0x%08x:  %-80s%s\n",
-              color, offset, p[0],
-              gen_group_get_name(inst), reset_color);
-
-      if (option_full_decode) {
-         decode_group(inst, p, 0);
-
-         for (i = 0; i < ARRAY_LENGTH(custom_handlers); i++) {
-            if (gen_group_get_opcode(inst) == custom_handlers[i].opcode) {
-               custom_handlers[i].handle(spec, p);
-               break;
-            }
-         }
-      }
-
-      if ((p[0] & 0xffff0000) == AUB_MI_BATCH_BUFFER_START) {
-         uint64_t start = get_address(spec, &p[1]);
-
-         if (p[0] & (1 << 22)) {
-            /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" set acts
-             * like a subroutine call.  Commands that come afterwards get
-             * processed once the 2nd level batch buffer returns with
-             * MI_BATCH_BUFFER_END.
-             */
-            parse_commands(spec, gtt + start, gtt_end - start, engine);
-         } else {
-            /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" unset acts
-             * like a goto.  Nothing after it will ever get processed.  In
-             * order to prevent the recursion from growing, we just reset the
-             * loop and continue;
-             */
-            p = gtt + start;
-            /* We don't know where secondaries end so use the GTT end */
-            end = gtt + gtt_end;
-            length = 0;
-            continue;
-         }
-      } else if ((p[0] & 0xffff0000) == AUB_MI_BATCH_BUFFER_END) {
-         break;
-      }
-   }
-}
-
 #define GEN_ENGINE_RENDER 1
 #define GEN_ENGINE_BLITTER 2
 
@@ -787,7 +110,7 @@ handle_trace_block(uint32_t *p)
    uint32_t *data = p + header_length + 2;
    int engine = GEN_ENGINE_RENDER;
 
-   if (gen_spec_get_gen(spec) >= gen_make_gen(8,0))
+   if (devinfo.gen >= 8)
       offset += (uint64_t) p[5] << 32;
 
    switch (operation) {
@@ -815,12 +138,28 @@ handle_trace_block(uint32_t *p)
          break;
       }
 
-      parse_commands(spec, data, size, engine);
+      (void)engine; /* TODO */
+      gen_print_batch(&batch_ctx, data, size, 0);
+
       gtt_end = 0;
       break;
    }
 }
 
+static struct gen_batch_decode_bo
+get_gen_batch_bo(void *user_data, uint64_t address)
+{
+   if (address > gtt_end)
+      return (struct gen_batch_decode_bo) { .map = NULL };
+
+   /* We really only have one giant address range */
+   return (struct gen_batch_decode_bo) {
+      .addr = 0,
+      .map = gtt,
+      .size = gtt_size
+   };
+}
+
 static void
 handle_trace_header(uint32_t *p)
 {
@@ -836,20 +175,22 @@ handle_trace_header(uint32_t *p)
    if (pci_id == 0)
       pci_id = aub_pci_id;
 
-   struct gen_device_info devinfo;
    if (!gen_get_device_info(pci_id, &devinfo)) {
       fprintf(stderr, "can't find device information: pci_id=0x%x\n", pci_id);
       exit(EXIT_FAILURE);
    }
 
-   if (xml_path == NULL)
-      spec = gen_spec_load(&devinfo);
-   else
-      spec = gen_spec_load_from_path(&devinfo, xml_path);
-   disasm = gen_disasm_create(&devinfo);
+   enum gen_batch_decode_flags batch_flags = 0;
+   if (option_color == COLOR_ALWAYS)
+      batch_flags |= GEN_BATCH_DECODE_IN_COLOR;
+   if (option_full_decode)
+      batch_flags |= GEN_BATCH_DECODE_FULL;
+   if (option_print_offsets)
+      batch_flags |= GEN_BATCH_DECODE_OFFSETS;
+   batch_flags |= GEN_BATCH_DECODE_FLOATS;
 
-   if (spec == NULL || disasm == NULL)
-      exit(EXIT_FAILURE);
+   gen_batch_decode_ctx_init(&batch_ctx, &devinfo, outfile, batch_flags,
+                             xml_path, get_gen_batch_bo, NULL);
 
    fprintf(outfile, "%sAubinator: Intel AUB file decoder.%-80s%s\n",
            GREEN_HEADER, "", NORMAL);
diff --git a/src/intel/tools/meson.build b/src/intel/tools/meson.build
index bff7e536630..7890eeca1dc 100644
--- a/src/intel/tools/meson.build
+++ b/src/intel/tools/meson.build
@@ -20,7 +20,8 @@
 
 aubinator = executable(
   'aubinator',
-  files('aubinator.c', 'disasm.c', 'gen_disasm.h', 'intel_aub.h'),
+  files('aubinator.c', 'disasm.c', 'gen_batch_decoder.c',
+        'gen_disasm.h', 'intel_aub.h'),
   dependencies : [dep_expat, dep_zlib, dep_dl, dep_thread, dep_m],
   include_directories : [inc_common, inc_intel],
   link_with : [libintel_common, libintel_compiler, libmesa_util],