Add wait before present option

Force wait on completion of queue operations for all queues before calling Present.
This is needed for accurate acquisition of instrumentation data on some platforms.

Change-Id: Ie5c07df95ad4420ed516f6f59c719d91887fa73a

USAGE_android.md

@@ -704,6 +704,7 @@ usage: gfxrecon.py replay [-h] [--push-file LOCAL_FILE] [--version] [--pause-fra
                           [--screenshot-prefix PREFIX] [--sfa] [--opcd]
                           [--surface-index N] [--sync] [--remove-unsupported]
                           [-m MODE] [--use-captured-swapchain-indices]
+                          [--wait-before-present]
                           [file]
 
 Launch the replay tool.
@@ -773,6 +774,10 @@ optional arguments:
                         setup for replay. The default without this option is to use a Virtual Swapchain
                         of images which match the swapchain in effect at capture time and which are
                         copied to the underlying swapchain of the implementation being replayed on.
+  --wait-before-present
+                        Force wait on completion of queue operations for all queues 
+                        before calling Present. This is needed for accurate acquisition 
+                        of instrumentation data on some platforms.
 ```
 
 The command will force-stop an active replay process before starting the replay

USAGE_desktop_Vulkan.md

@@ -368,7 +368,7 @@ gfxrecon-replay         [-h | --help] [--version] [--gpu <index>]
                         [-m <mode> | --memory-translation <mode>]
                         [--use-captured-swapchain-indices]
                         [--log-level <level>] [--log-file <file>] [--log-debugview]
-                        [--api <api>] [--no-debug-popup] <file>
+                        [--api <api>] [--no-debug-popup] [--wait-before-present] <file>
 
 Required arguments:
   <file>                Path to the capture file to replay.
@@ -470,6 +470,10 @@ Optional arguments:
                         setup for replay. The default without this option is to use a Virtual Swapchain
                         of images which match the swapchain in effect at capture time and which are
                         copied to the underlying swapchain of the implementation being replayed on.
+  --wait-before-present
+                        Force wait on completion of queue operations for all queues 
+                        before calling Present. This is needed for accurate acquisition 
+                        of instrumentation data on some platforms.
 ```
 
 ### Key Controls

android/scripts/gfxrecon.py

@@ -81,6 +81,7 @@ def CreateReplayParser():
     parser.add_argument('--validate', action='store_true', default=False, help='Enables the Khronos Vulkan validation layer (forwarded to replay tool)')
     parser.add_argument('--onhb', '--omit-null-hardware-buffers', action='store_true', default=False, help='Omit Vulkan calls that would pass a NULL AHardwareBuffer* (forwarded to replay tool)')
     parser.add_argument('--use-captured-swapchain-indices', action='store_true', default=False, help='Use the swapchain indices stored in the capture directly on the swapchain setup for replay. The default without this option is to use a Virtual Swapchain of images which match the swapchain in effect at capture time and which are copied to the underlying swapchain of the implementation being replayed on.')
+    parser.add_argument('--wait-before-present', action='store_true', default=False, help='Force wait on completion of queue operations for all queues before calling Present. This is needed for accurate acquisition of instrumentation data on some platforms.')
 
     parser.add_argument('-m', '--memory-translation', metavar='MODE', choices=['none', 'remap', 'realign', 'rebind'], help='Enable memory translation for replay on GPUs with memory types that are not compatible with the capture GPU\'s memory types.  Available modes are: none, remap, realign, rebind (forwarded to replay tool)')
     parser.add_argument('file', nargs='?', help='File on device to play (forwarded to replay tool)')
@@ -146,6 +147,9 @@ def MakeExtrasString(args):
         arg_list.append('-m')
         arg_list.append('{}'.format(args.memory_translation))
 
+    if args.wait_before_present:
+        arg_list.append('--wait-before-present')
+
     if args.file:
         arg_list.append(args.file)
     elif not args.version:

framework/decode/vulkan_replay_consumer_base.cpp

@@ -5522,6 +5522,13 @@ VulkanReplayConsumerBase::OverrideQueuePresentKHR(PFN_vkQueuePresentKHR
         modified_present_info.pImageIndices = modified_image_indices_.data();
     }
 
+    if (options_.wait_before_present)
+    {
+        format::HandleId device_handle = queue_info->parent_id;
+        VkDevice         device        = MapHandle<DeviceInfo>(device_handle, &VulkanObjectInfoTable::GetDeviceInfo);
+        GetDeviceTable((const void*)device_handle)->DeviceWaitIdle(device);
+    }
+
     // Only attempt to find imported or shadow semaphores if we know at least one around.
     if ((!have_imported_semaphores_) && (shadow_semaphores_.empty()) && (modified_present_info.swapchainCount != 0))
     {

framework/decode/vulkan_replay_options.h

@@ -56,6 +56,7 @@ struct VulkanReplayOptions : public ReplayOptions
     std::string                  screenshot_dir;
     std::string                  screenshot_file_prefix{ kDefaultScreenshotFilePrefix };
     std::string                  replace_dir;
+    bool                         wait_before_present{ false };
 };
 
 GFXRECON_END_NAMESPACE(decode)

tools/replay/replay_settings.h

@@ -31,7 +31,7 @@ const char kOptions[] =
     "opcd|--omit-pipeline-cache-data,--remove-unsupported,--validate,--debug-device-lost,--create-dummy-allocations,--"
     "screenshot-all,--onhb|--omit-null-hardware-buffers,--qamr|--quit-after-measurement-"
     "range,--fmr|--flush-measurement-range,--use-captured-swapchain-indices,--dcp,--discard-cached-psos,"
-    "--dx12-override-object-names";
+    "--dx12-override-object-names,--wait-before-present";
 const char kArguments[] =
     "--log-level,--log-file,--gpu,--gpu-group,--pause-frame,--wsi,--surface-index,-m|--memory-translation,"
     "--replace-shaders,--screenshots,--denied-messages,--allowed-messages,--screenshot-format,--"
@@ -196,6 +196,10 @@ static void PrintUsage(const char* exe_name)
     GFXRECON_WRITE_CONSOLE("          \t\treturned by vkEnumeratePhysicalDeviceGroups.  Replay may fail");
     GFXRECON_WRITE_CONSOLE("          \t\tif the specified device group is not compatible with the");
     GFXRECON_WRITE_CONSOLE("          \t\toriginal capture device group.");
+    GFXRECON_WRITE_CONSOLE("  --wait-before-present");
+    GFXRECON_WRITE_CONSOLE("          \t\tForce wait on completion of queue operations for all queues");
+    GFXRECON_WRITE_CONSOLE("          \t\tbefore calling Present. This is needed for accurate acquisition");
+    GFXRECON_WRITE_CONSOLE("          \t\tof instrumentation data on some platforms.");
 
 #if defined(WIN32)
     GFXRECON_WRITE_CONSOLE("")

tools/tool_settings.h

@@ -104,6 +104,8 @@ const char kFormatArgument[]                     = "--format";
 const char kIncludeBinariesOption[]              = "--include-binaries";
 const char kExpandFlagsOption[]                  = "--expand-flags";
 const char kFilePerFrameOption[]                 = "--file-per-frame";
+const char kWaitBeforePresent[]                  = "--wait-before-present";
+
 #if defined(WIN32)
 const char kApiFamilyOption[]       = "--api";
 const char kDxTwoPassReplay[]       = "--dx12-two-pass-replay";
@@ -785,6 +787,11 @@ GetVulkanReplayOptions(const gfxrecon::util::ArgumentParser&           arg_parse
         replay_options.surface_index = std::stoi(surface_index);
     }
 
+    if (arg_parser.IsOptionSet(kWaitBeforePresent))
+    {
+        replay_options.wait_before_present = true;
+    }
+
     return replay_options;
 }