intel/dump_gpu: add an option to capture a single frame

We use the driver identifier buffer to get the information about the
current frame.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2201>

diff --git a/src/intel/tools/intel_dump_gpu.c b/src/intel/tools/intel_dump_gpu.c
index c3330e6cc2eba03d5ce5856727d018c1c00758e4..197aebd5b328d3153d3b8844849ead0efa8c232a 100644 (file)
--- a/src/intel/tools/intel_dump_gpu.c
+++ b/src/intel/tools/intel_dump_gpu.c
@@ -43,6 +43,7 @@
 #include "intel_aub.h"
 #include "aub_write.h"
 
+#include "dev/gen_debug.h"
 #include "dev/gen_device_info.h"
 #include "util/macros.h"
 
@@ -60,6 +61,8 @@ static FILE *output_file = NULL;
 static int verbose = 0;
 static bool device_override = false;
 static bool capture_only = false;
+static int64_t frame_id = -1;
+static bool capture_finished = false;
 
 #define MAX_FD_COUNT 64
 #define MAX_BO_COUNT 64 * 1024
@@ -229,6 +232,9 @@ dump_execbuffer2(int fd, struct drm_i915_gem_execbuffer2 *execbuffer2)
 
    ensure_device_info(fd);
 
+   if (capture_finished)
+      return;
+
    if (!aub_file.file) {
       aub_file_init(&aub_file, output_file,
                     verbose == 2 ? stdout : NULL,
@@ -245,9 +251,6 @@ dump_execbuffer2(int fd, struct drm_i915_gem_execbuffer2 *execbuffer2)
    else
       offset = aub_gtt_size(&aub_file);
 
-   if (verbose)
-      printf("Dumping execbuffer2:\n");
-
    for (uint32_t i = 0; i < execbuffer2->buffer_count; i++) {
       obj = &exec_objects[i];
       bo = get_bo(fd, obj->handle);
@@ -263,22 +266,63 @@ dump_execbuffer2(int fd, struct drm_i915_gem_execbuffer2 *execbuffer2)
 
       if (obj->flags & EXEC_OBJECT_PINNED) {
          bo->offset = obj->offset;
-         if (verbose)
-            printf("BO #%d (%dB) pinned @ 0x%" PRIx64 "\n",
-                   obj->handle, bo->size, bo->offset);
       } else {
          if (obj->alignment != 0)
             offset = align_u32(offset, obj->alignment);
          bo->offset = offset;
-         if (verbose)
-            printf("BO #%d (%dB) @ 0x%" PRIx64 "\n", obj->handle,
-                   bo->size, bo->offset);
          offset = align_u32(offset + bo->size + 4095, 4096);
       }
 
       if (bo->map == NULL && bo->size > 0)
          bo->map = gem_mmap(fd, obj->handle, 0, bo->size);
       fail_if(bo->map == MAP_FAILED, "bo mmap failed\n");
+   }
+
+   uint64_t current_frame_id = 0;
+   if (frame_id >= 0) {
+      for (uint32_t i = 0; i < execbuffer2->buffer_count; i++) {
+         obj = &exec_objects[i];
+         bo = get_bo(fd, obj->handle);
+
+         /* Check against frame_id requirements. */
+         if (memcmp(bo->map, intel_debug_identifier(),
+                    intel_debug_identifier_size()) == 0) {
+            const struct gen_debug_block_frame *frame_desc =
+               intel_debug_get_identifier_block(bo->map, bo->size,
+                                                GEN_DEBUG_BLOCK_TYPE_FRAME);
+
+            current_frame_id = frame_desc ? frame_desc->frame_id : 0;
+            break;
+         }
+      }
+   }
+
+   if (verbose)
+      printf("Dumping execbuffer2 (frame_id=%"PRIu64", buffers=%u):\n",
+             current_frame_id, execbuffer2->buffer_count);
+
+   /* Check whether we can stop right now. */
+   if (frame_id >= 0) {
+      if (current_frame_id < frame_id)
+         return;
+
+      if (current_frame_id > frame_id) {
+         aub_file_finish(&aub_file);
+         capture_finished = true;
+         return;
+      }
+   }
+
+
+   /* Map buffers into the PPGTT. */
+   for (uint32_t i = 0; i < execbuffer2->buffer_count; i++) {
+      obj = &exec_objects[i];
+      bo = get_bo(fd, obj->handle);
+
+      if (verbose) {
+         printf("BO #%d (%dB) @ 0x%" PRIx64 "\n",
+                obj->handle, bo->size, bo->offset);
+      }
 
       if (aub_use_execlists(&aub_file) && !bo->gtt_mapped) {
          aub_map_ppgtt(&aub_file, bo->offset, bo->size);
@@ -286,6 +330,7 @@ dump_execbuffer2(int fd, struct drm_i915_gem_execbuffer2 *execbuffer2)
       }
    }
 
+   /* Write the buffer content into the Aub. */
    batch_index = (execbuffer2->flags & I915_EXEC_BATCH_FIRST) ? 0 :
       execbuffer2->buffer_count - 1;
    batch_bo = get_bo(fd, exec_objects[batch_index].handle);
@@ -428,6 +473,8 @@ maybe_init(int fd)
                  output_filename);
       } else if (!strcmp(key, "capture_only")) {
          capture_only = atoi(value);
+      } else if (!strcmp(key, "frame")) {
+         frame_id = atol(value);
       } else {
          fprintf(stderr, "unknown option '%s'\n", key);
       }
@@ -728,7 +775,8 @@ fini(void)
 {
    if (devinfo.gen != 0) {
       free(output_filename);
-      aub_file_finish(&aub_file);
+      if (!capture_finished)
+         aub_file_finish(&aub_file);
       free(bos);
    }
 }

diff --git a/src/intel/tools/intel_dump_gpu.in b/src/intel/tools/intel_dump_gpu.in
index 0395af0a8eb50a31dc7fd513be843d2f4ff1a8b1..f486f34d3880a694cd04b87eb127917875647f3d 100755 (executable)
--- a/src/intel/tools/intel_dump_gpu.in
+++ b/src/intel/tools/intel_dump_gpu.in
@@ -20,6 +20,8 @@ contents and execution of the GEM application.
                       the output aub file. This helps reducing output file
                       size greatly but won't produce a file replayable
 
+  -f, --frame=ID      Only dump objects for frame ID
+
   -v                  Enable verbose output
 
   -vv                 Enable extra verbosity - dumps gtt mappings
@@ -36,6 +38,7 @@ args=""
 file=""
 gdb=""
 capture_only=""
+frame=""
 
 function add_arg() {
     arg=$1
@@ -86,6 +89,21 @@ while true; do
             add_arg "platform=${platform}"
             shift
             ;;
+        -f)
+            frame=$2
+            add_arg "frame=${frame}"
+            shift 2
+            ;;
+        -f*)
+            frame=${1##-f}
+            add_arg "frame=${frame}"
+            shift
+            ;;
+        --frame=*)
+            frame=${1##--frame=}
+            add_arg "frame=${frame}"
+            shift
+            ;;
         --gdb)
             gdb=1
             shift

diff --git a/src/intel/tools/meson.build b/src/intel/tools/meson.build
index d83812bab08bd9cb3a75e06a77826eb6b8a2e880..ac892677f9a4623f09d7d65d26e753bb9bb6ed73 100644 (file)
--- a/src/intel/tools/meson.build
+++ b/src/intel/tools/meson.build
@@ -116,7 +116,7 @@ libintel_dump_gpu = shared_library(
     'gen_context.h', 'gen8_context.h', 'gen10_context.h', 'intel_aub.h',
     'aub_write.h', 'aub_write.c', 'intel_dump_gpu.c',
   ),
-  dependencies : dep_dl,
+  dependencies : [dep_dl, idep_mesautil],
   include_directories : [inc_include, inc_src, inc_intel],
   link_with : libintel_dev,
   c_args : [no_override_init_args],