DLSS 5 Autopilot 2.0.9

Updating from 2.0.x with update now downloads this build, checks its
SHA-256 against the SHA256SUMS.txt published with the release, and offers
a one-click restart. Your settings, library, hidden games and profiles carry
over.

Added

• did it work? reads which graphics runtime the game loaded. While the
  tool is open it records which of Windows' graphics DLLs a started game
  holds. A game set up as DirectX 10/11/12 that ran with OpenGL or
  DirectX 9 loaded and none of DirectX 10/11/12's DLLs is named as that,
  with the fix: if the game was already drawing, set graphics api to
  that renderer and install again (#682), where the answer had been "try
  another proxy name", which none of the DirectX 10/11/12 names would have
  fixed. This is read on the routes that load through ReShade (native,
  neural-upstream, renodx-dlss, bridge, feeder, standalone-dlssnr); the
  optiscaler and remix routes and autopilot do not read it. The record
  is made while the tool is watching the game, or when did it work? is
  pressed while the game runs. Checked by replaying earlier reports with
  the recorded line added; it has not been seen on a running game on this
  tool's test machine.
• ReShade's Vulkan layer is also registered for games that run as
  administrator. The Vulkan loader gives a program started as
  administrator none of the layers registered per user, which is where the tool registers ReShade's
  for Vulkan games and for games that go through DXVK. When the
  tool itself runs as administrator it also puts a copy of the layer under
  Program Files and registers that for all users; what will happen?
  lists it before installing, and the install log says where it went.
  Where ReShade's own setup already registered its layer for all users,
  the tool adds nothing there. An existing install has to be installed
  again with the tool run as administrator; updating the tool alone
  changes nothing. Uninstalling the last such game as administrator
  removes it; without administrator rights the tool says the registration
  is still there, and starting the tool once as administrator, while no
  Vulkan or DXVK install is left, takes it off (#562). This was checked against a simulated
  registry; it has not been run with an elevated game on this tool's test
  machine - a report from anyone whose game runs as administrator is
  welcome.
• The window opens the way it was closed: the same size and place, and
  maximised only if it was maximised when closed (#641). The first start
  after the update opens maximised once.
• A GTX 16 card is told DLSS cannot run on it. A GTX 1650 or 1660 is
  the same generation as RTX 20 and has no tensor cores; the driver offers
  no DLSS on it. The game page says so, and the install is refused before
  anything is written (#770); what will happen? and --check say the
  same. The MX 450/550 and the T400-T2000 workstation cards (Turing without
  tensor cores) and the GTX 10 cards are refused the same way. A card the
  tool cannot name is left to the build check, and so is any PC where one
  NVIDIA card has a name it cannot read - Windows keeps the entry of a card
  that was taken out.
• A bug report too long for the link opens with its start filled in
  (as much as fits: the header, the diagnosis and the file list first),
  under the instruction to paste the whole of it, which is on the
  clipboard. The page had held one line asking for the paste, and two of
  the four reports sent that way arrived with that line only (#751, #770).

Fixed

• A Source engine game was set up in front of an SDK tool in bin\
  (elementviewer.exe for Half-Life 2: Episode Two) rather than the game's
  executable in the root folder (#669). A game already installed that way
  keeps the wrong executable: press uninstall, set target exe to
  the game's own (hl2.exe) in the game's settings, then install.
  Checked on a folder built to that layout; no Source game is installed on
  this tool's test machine.
• On the bridge route, a game with DLSS on and Ray Reconstruction on was
  told to "turn DLSS on". The verdict says what the bridge's log shows - no
  DLSS Super Resolution call - and to turn Ray Reconstruction off if the
  game has it (#637). From one report; not run on this tool's test machine.
• On the optiscaler route, a Present that failed after neural rendering had
  drawn and nothing drew after it (in #768 on a window change, in the
  game's frame generation) was told "the game refused OptiScaler's
  swapchain - try another name in 'loads as'". It is named as a later
  failure, with the steps for it; a lost graphics device gets the driver's
  steps instead (#768).
• On the standalone route, a game that left no ReShade.log was told it had
  not been started, although the add-on's own log held the session (#744).
  When no other folder has the standalone route installed, that log is
  read, and the fault Windows records as the game closes is named as one.
• A crash Windows recorded in the same session replaces "play once and
  check again" (#748).
• An add-on in the process and no log written is named as ReShade having
  run and written no log - most likely because it cannot write to the
  game's folder - not as ReShade not starting, and it is not counted
  against the route (#787).
  The standalone add-on's log in a report is marked when it was written
  before the install, because it is one file for every game.
• A game's own nvngx_dlss.dll in its own folder (Streamline's
  streamline\, or an Unreal plugin's) is not called a copy "loaded from
  another folder", also from a record an earlier version made,
  and a native Vulkan install with nothing of it in the game is named as
  ReShade's layer not reaching it, not as a proxy name to change (#740).
• A report made after an uninstall, a reinstall, update all or a swap
  on the dlss page carried the diagnosis from before it ("Working." over an
  empty folder, #761).
• A crash in a file the tool installs names the project that file comes
  from (DLSS5-Feeder, OptiScaler, ReShade...) and says the caller can be
  another file in the same process, instead of "this is ours to fix"
  (#511).

Code signing policy