sc: bone eater preliminary support (#895)

## Link to GitHub Issue or related Pull Request, if one exists
Fixes #114

## Description of change
Initial support for Silent Scope: Bone Eater, enough to be playable;
doesn't include QoS work needed to make this a smoother experience.

* Game detection
* Fix DX11 overlay to work with non-Unity games
* Implements sub monitor i/o (only really needed to get past I/O check)
* Try to make the game reasonably playable with a mouse
* Ignore the game's attempt to move windows, position them in a
reasonable place
* Fix language hook

## Testing

README.md

@@ -56,6 +56,10 @@ We explicitly do **NOT** have a Discord server for dicussions - we try to do eve
 
 Please see [CONTRIBUTING page](https://github.com/spice2x/spice2x.github.io/blob/main/CONTRIBUTING.md) for a full list of guidelines when submitting code.
 
+### AI-Use Policy
+
+Use AI tools at your discretion, but remember that a human will review every issue and pull request. Please ensure submissions are accurate, relevant, and worth the reviewer’s time.
+
 ## Additional information
 
 Please read [README.md](https://github.com/spice2x/spice2x.github.io/blob/main/src/spice2x/README.md) inside `src/spice2x`.

src/spice2x/CMakeLists.txt

@@ -531,6 +531,7 @@ set(SOURCE_FILES ${SOURCE_FILES}
         games/otoca/p4io.cpp
         games/silentscope/silentscope.cpp
         games/silentscope/io.cpp
+        games/silentscope/projector.cpp
         games/pcm/pcm.cpp
         games/pcm/io.cpp
         games/onpara/onpara.cpp

src/spice2x/README.md

@@ -144,6 +144,13 @@ restriction is that the ID has to be a valid 64-bit unsigned integer.
     `busy`
 
 #### Card
+- get_cards()
+  - returns the current card ID and source for each active card reader
+  - each entry contains `index`, `card_id`, and `source`
+  - `source` is `file` or `override`
+  - file entries also contain `file_name`, without the full configured path
+  - `-card0` and `-card1` overrides do not contain `file_name`
+  - this function only works when an API password is configured
 - insert(index: uint, card_id: hex)
   - inserts a card which gets read by the emulated card readers for the game
   - index has to be either 0 (for P1) or 1 (for P2)

src/spice2x/acio/bmpu/bmpu.cpp

@@ -1,5 +1,7 @@
 #include "bmpu.h"
 
+#include <algorithm>
+
 #include "acio/icca/icca.h"
 #include "avs/game.h"
 #include "cfg/api.h"
@@ -9,6 +11,7 @@
 #include "games/ftt/io.h"
 #include "games/museca/io.h"
 #include "games/silentscope/io.h"
+#include "hooks/graphics/graphics.h"
 #include "launcher/launcher.h"
 #include "misc/eamuse.h"
 
@@ -530,15 +533,38 @@ static bool __cdecl ac_io_bmpu_update_control_status_buffer() {
             STATUS_BUFFER[4] |= 0x20;
         }
 
-        // joy stick raw input
+        // gun position, sent big endian; the game keeps the top 12 bits of each pair
         auto &analogs = games::silentscope::get_analogs();
+        auto &gun_x = analogs.at(games::silentscope::Analogs::GUN_X);
+        auto &gun_y = analogs.at(games::silentscope::Analogs::GUN_Y);
         unsigned short joy_x = 0x7FFF;
         unsigned short joy_y = 0x7FFF;
-        if (analogs.at(games::silentscope::Analogs::GUN_X).isSet()) {
-            joy_x = (unsigned short) (Analogs::getState(RI_MGR, analogs.at(games::silentscope::Analogs::GUN_X)) * USHRT_MAX);
-        }
-        if (analogs.at(games::silentscope::Analogs::GUN_Y).isSet()) {
-            joy_y = (unsigned short) (Analogs::getState(RI_MGR, analogs.at(games::silentscope::Analogs::GUN_Y)) * USHRT_MAX);
+
+        if (gun_x.isSet() || gun_y.isSet()) {
+            if (gun_x.isSet()) {
+                joy_x = (unsigned short) (Analogs::getState(RI_MGR, gun_x) * USHRT_MAX);
+            }
+            if (gun_y.isSet()) {
+                joy_y = (unsigned short) (Analogs::getState(RI_MGR, gun_y) * USHRT_MAX);
+            }
+        } else {
+
+            // the gun reports where it is aimed, which a relative pointer cannot express - the
+            // calibration screen asks for the screen centre and two corners, so read the cursor
+            POINT cursor {};
+            RECT client {};
+            POINT origin {};
+            if (NDD_MAIN_WINDOW != nullptr &&
+                GetCursorPos(&cursor) &&
+                GetClientRect(NDD_MAIN_WINDOW, &client) &&
+                ClientToScreen(NDD_MAIN_WINDOW, &origin) &&
+                client.right > 1 && client.bottom > 1)
+            {
+                const LONG x = std::clamp(cursor.x - origin.x, 0L, client.right - 1);
+                const LONG y = std::clamp(cursor.y - origin.y, 0L, client.bottom - 1);
+                joy_x = (unsigned short) (x * USHRT_MAX / (client.right - 1));
+                joy_y = (unsigned short) (y * USHRT_MAX / (client.bottom - 1));
+            }
         }
 
         // invert X axis

src/spice2x/api/controller.cpp

@@ -337,9 +337,11 @@ bool Controller::process_request(ClientState *state, const char *in, size_t in_s
             if (module->name == request.module) {
                 module_found = true;
 
-                // check password force
-                if (module->password_force && this->password.empty() && request.function != "session_refresh") {
-                    Value err("Module requires the password to be set.");
+                // check password requirement
+                if (module->requires_password(request.function)
+                        && this->password.empty()
+                        && request.function != "session_refresh") {
+                    Value err("Function requires the password to be set.");
                     response.add_error(err);
                     break;
                 }

src/spice2x/api/h264_stream.cpp

@@ -118,7 +118,13 @@ namespace api {
                 param.i_height = height;
                 param.i_fps_num = this->fps;
                 param.i_fps_den = 1;
-                param.i_threads = 1;
+
+                // sliced threading, which zerolatency already selected, so a frame is split
+                // across workers rather than held back to be reordered. deliberately not the
+                // automatic count: this shares a machine with the game it is capturing, and
+                // taking every core to encode would win back frames at the game's expense
+                param.i_threads = 4;
+
                 param.b_annexb = 1;
                 // SPS/PPS ahead of every IDR, so a client can start decoding cold
                 param.b_repeat_headers = 1;

src/spice2x/api/module.cpp

@@ -16,6 +16,14 @@ namespace api {
         this->password_force = password_force;
     }
 
+    void Module::require_password(const std::string &function) {
+        this->password_force_functions.emplace(function);
+    }
+
+    bool Module::requires_password(const std::string &function) const {
+        return this->password_force || this->password_force_functions.contains(function);
+    }
+
     void Module::handle(Request &req, Response &res) {
 
         // log module access

src/spice2x/api/module.h

@@ -4,6 +4,7 @@
 #include <map>
 #include <string>
 #include <sstream>
+#include <unordered_set>
 #include <external/robin_hood.h>
 
 #include "response.h"
@@ -26,6 +27,13 @@ namespace api {
         // default constructor
         explicit Module(std::string name, bool password_force=false);
 
+        void require_password(const std::string &function);
+
+    private:
+
+        // functions which expose sensitive data or actions
+        std::unordered_set<std::string> password_force_functions;
+
     public:
 
         // virtual deconstructor
@@ -35,6 +43,8 @@ namespace api {
         std::string name;
         bool password_force;
 
+        bool requires_password(const std::string &function) const;
+
         // the magic
         void handle(Request &req, Response &res);
 

src/spice2x/api/modules/card.cpp

@@ -1,4 +1,5 @@
 #include "card.h"
+#include <fstream>
 #include <functional>
 #include "external/rapidjson/document.h"
 #include "util/logging.h"
@@ -11,8 +12,71 @@ using namespace rapidjson;
 
 namespace api::modules {
 
+    static bool normalize_card_id(const std::string &value, std::string &card_id) {
+        if (value.size() != 16) {
+            return false;
+        }
+
+        uint8_t card_bin[8] {};
+        if (!hex2bin(value.c_str(), card_bin)) {
+            return false;
+        }
+
+        card_id = bin2hex(card_bin, std::size(card_bin));
+        return true;
+    }
+
+    static bool read_card_id(const std::filesystem::path &path, std::string &card_id) {
+        std::ifstream file(path);
+        char buffer[16] {};
+        if (!file.read(buffer, std::size(buffer))) {
+            return false;
+        }
+
+        return normalize_card_id(std::string(buffer, std::size(buffer)), card_id);
+    }
+
     Card::Card() : Module("card") {
+        functions["get_cards"] = std::bind(&Card::get_cards, this, _1, _2);
         functions["insert"] = std::bind(&Card::insert, this, _1, _2);
+        require_password("get_cards");
+    }
+
+    /**
+     * get_cards()
+     */
+    void Card::get_cards(Request &req, Response &res) {
+        auto &alloc = res.doc()->GetAllocator();
+
+        for (int index = 0; index < eamuse_get_game_keypads(); index++) {
+            std::string card_id;
+            std::string filename;
+            const auto card_override = eamuse_get_card_override(index);
+            const bool has_override = !card_override.empty();
+
+            if (has_override) {
+                if (!normalize_card_id(card_override, card_id)) {
+                    continue;
+                }
+            } else {
+                const auto path = eamuse_get_card_path(index);
+                if (!read_card_id(path, card_id)) {
+                    continue;
+                }
+
+                const auto filename_u8 = path.filename().u8string();
+                filename.assign(filename_u8.begin(), filename_u8.end());
+            }
+
+            Value card(kObjectType);
+            card.AddMember("index", index, alloc);
+            card.AddMember("card_id", Value(card_id.c_str(), alloc), alloc);
+            card.AddMember("source", Value(has_override ? "override" : "file", alloc), alloc);
+            if (!has_override) {
+                card.AddMember("file_name", Value(filename.c_str(), alloc), alloc);
+            }
+            res.add_data(card);
+        }
     }
 
     /**

src/spice2x/api/modules/card.h

@@ -12,6 +12,7 @@ namespace api::modules {
     private:
 
         // function definitions
+        void get_cards(Request &req, Response &res);
         void insert(Request &req, Response &res);
     };
 }

src/spice2x/api/resources/dart/spiceapi-websocket/src/wrappers/card.dart

@@ -1,5 +1,32 @@
 part of spiceapi;
 
+class CardInfo {
+  final int index;
+  final String cardID;
+  final String source;
+  final String? fileName;
+
+  CardInfo(this.index, this.cardID, this.source, this.fileName);
+}
+
+Future<List<CardInfo>> cardGetCards(Connection con) {
+  var req = Request("card", "get_cards");
+  return con.request(req).then((res) {
+    List<CardInfo> cards = [];
+    for (var value in res.getData()) {
+      cards.add(
+        CardInfo(
+          value["index"],
+          value["card_id"],
+          value["source"],
+          value["file_name"],
+        ),
+      );
+    }
+    return cards;
+  });
+}
+
 Future<void> cardInsert(Connection con, int unit, String cardID) {
   var req = Request("card", "insert");
   req.addParam(unit);

src/spice2x/api/resources/dart/spiceapi/src/wrappers/card.dart

@@ -1,5 +1,32 @@
 part of spiceapi;
 
+class CardInfo {
+  final int index;
+  final String cardID;
+  final String source;
+  final String? fileName;
+
+  CardInfo(this.index, this.cardID, this.source, this.fileName);
+}
+
+Future<List<CardInfo>> cardGetCards(Connection con) {
+  var req = Request("card", "get_cards");
+  return con.request(req).then((res) {
+    List<CardInfo> cards = [];
+    for (var value in res.getData()) {
+      cards.add(
+        CardInfo(
+          value["index"],
+          value["card_id"],
+          value["source"],
+          value["file_name"],
+        ),
+      );
+    }
+    return cards;
+  });
+}
+
 Future<void> cardInsert(Connection con, int unit, String cardID) {
   var req = Request("card", "insert");
   req.addParam(unit);

src/spice2x/api/resources/python/spiceapi/card.py

@@ -2,6 +2,10 @@ from .connection import Connection
 from .request import Request
 
 
+def card_get_cards(con: Connection):
+    return con.request(Request("card", "get_cards")).get_data()
+
+
 def card_insert(con: Connection, unit: int, card_id: str):
     req = Request("card", "insert")
     req.add_param(unit)

src/spice2x/games/popn/bi3a_hook.cpp

@@ -2,6 +2,8 @@
 
 #if SPICE64 && !SPICE_XP
 
+#include <array>
+#include <mutex>
 #include <optional>
 #include <cstdint>
 #include "util/detour.h"
@@ -141,10 +143,100 @@ namespace games::popn {
     static AIO_SCI_COMM *aioSciComm;
     static AIO_IOB5_BI3A *aioIob5Bi3a;
 
+    static constexpr size_t PIKA_BUTTON_COUNT = 9;
+    static constexpr size_t PIKA_BUTTON_DATA_SIZE = PIKA_BUTTON_COUNT * 3;
+    static std::array<uint8_t, PIKA_BUTTON_DATA_SIZE> pika_button_staging{};
+    static bool pika_button_staging_dirty = false;
+    static std::mutex pika_button_staging_mutex;
+
     /*
      * implementations
      */
 
+    static void publish_pika_button_staging() {
+        std::array<uint8_t, PIKA_BUTTON_DATA_SIZE> data;
+        {
+            std::lock_guard<std::mutex> lock(pika_button_staging_mutex);
+            if (!pika_button_staging_dirty) {
+                return;
+            }
+            data = pika_button_staging;
+            pika_button_staging_dirty = false;
+        }
+
+        constexpr Lights::popn_lights_t legacy_button_lights[] = {
+            Lights::popn_lights_t::Button1,
+            Lights::popn_lights_t::Button2,
+            Lights::popn_lights_t::Button3,
+            Lights::popn_lights_t::Button4,
+            Lights::popn_lights_t::Button5,
+            Lights::popn_lights_t::Button6,
+            Lights::popn_lights_t::Button7,
+            Lights::popn_lights_t::Button8,
+            Lights::popn_lights_t::Button9
+        };
+        constexpr Lights::popn_lights_t button_lights[] = {
+            Lights::popn_lights_t::PikaButton1_R,
+            Lights::popn_lights_t::PikaButton1_G,
+            Lights::popn_lights_t::PikaButton1_B,
+            Lights::popn_lights_t::PikaButton2_R,
+            Lights::popn_lights_t::PikaButton2_G,
+            Lights::popn_lights_t::PikaButton2_B,
+            Lights::popn_lights_t::PikaButton3_R,
+            Lights::popn_lights_t::PikaButton3_G,
+            Lights::popn_lights_t::PikaButton3_B,
+            Lights::popn_lights_t::PikaButton4_R,
+            Lights::popn_lights_t::PikaButton4_G,
+            Lights::popn_lights_t::PikaButton4_B,
+            Lights::popn_lights_t::PikaButton5_R,
+            Lights::popn_lights_t::PikaButton5_G,
+            Lights::popn_lights_t::PikaButton5_B,
+            Lights::popn_lights_t::PikaButton6_R,
+            Lights::popn_lights_t::PikaButton6_G,
+            Lights::popn_lights_t::PikaButton6_B,
+            Lights::popn_lights_t::PikaButton7_R,
+            Lights::popn_lights_t::PikaButton7_G,
+            Lights::popn_lights_t::PikaButton7_B,
+            Lights::popn_lights_t::PikaButton8_R,
+            Lights::popn_lights_t::PikaButton8_G,
+            Lights::popn_lights_t::PikaButton8_B,
+            Lights::popn_lights_t::PikaButton9_R,
+            Lights::popn_lights_t::PikaButton9_G,
+            Lights::popn_lights_t::PikaButton9_B
+        };
+
+        static_assert(std::size(legacy_button_lights) == PIKA_BUTTON_COUNT);
+        static_assert(std::size(button_lights) == PIKA_BUTTON_DATA_SIZE);
+
+        auto &lights = get_lights();
+        for (size_t button = 0; button < PIKA_BUTTON_COUNT; button++) {
+            const auto data_offset = button * 3;
+            // special handling converting RGB data to non-RGB lights
+            // take the max(R, G, B) and use it to write the light value
+            const auto light_value = std::max({
+                data[data_offset],
+                data[data_offset + 1],
+                data[data_offset + 2]
+            });
+
+            // on the new cab, buttons are colorless plastic and rely on RGB to be lit
+            // at all times, even when "off". When translating to legacy on/off lights,
+            // treat only values above about 60% brightness as fully on to preserve the
+            // dim "off" state.
+            GameAPI::Lights::writeLight(
+                RI_MGR,
+                lights.at(legacy_button_lights[button]),
+                light_value > 150 ? 1.f : 0.f);
+        }
+
+        for (size_t light = 0; light < data.size(); light++) {
+            GameAPI::Lights::writeLight(
+                RI_MGR,
+                lights.at(button_lights[light]),
+                data[light] / 255.f);
+        }
+    }
+
     // libaio.dll
 
     static AIO_SCI_COMM_T *__fastcall aioSciComm_Open(AIO_SCI_COMM_T *unk) {
@@ -260,6 +352,11 @@ namespace games::popn {
         log_info("bi3a_hook", "aioIob5Bi3a_Create called with i_pNodeMgr={}, i={}, p={}", fmt::ptr(i_pNodeMgr), i, fmt::ptr(p));
 
         if (i_pNodeMgr == aioNmgrIob5) {
+            {
+                std::lock_guard<std::mutex> lock(pika_button_staging_mutex);
+                pika_button_staging.fill(0);
+                pika_button_staging_dirty = false;
+            }
             aioIob5Bi3a = new AIO_IOB5_BI3A;
             log_info("bi3a_hook", "aioIob5Bi3a_Create: returning custom AIO_IOB5_BI3A: {}", fmt::ptr(aioIob5Bi3a));
             return aioIob5Bi3a;
@@ -269,11 +366,15 @@ namespace games::popn {
     }
 
     static void __fastcall aioIob5Bi3a_GetDeviceStatus(AIO_IOB5_BI3A *i_pNodeCtl, AIO_IOB5_BI3A__DEVSTATUS *o_DevStatus) {
-        RI_MGR->devices_flush_output();
         if (i_pNodeCtl != aioIob5Bi3a) {
             return aioIob5Bi3a_GetDeviceStatus_orig(i_pNodeCtl, o_DevStatus);
         }
 
+        // snapshot the current state of lights and flush the output
+        // this closely mirrors what the game does with real I/O
+        publish_pika_button_staging();
+        RI_MGR->devices_flush_output();
+
         memset(o_DevStatus, 0, sizeof(*o_DevStatus));
         auto &buttons = get_buttons();
 
@@ -413,61 +514,15 @@ namespace games::popn {
                 i_CnPin, number_of_leds, (uint8_t)i_LedType, fmt::ptr(i_pData), data);
         }
 
-        auto &lights = get_lights();
         if (i_CnPin == 0 && number_of_leds == 9 * 3) {
-            // special handling converting to non-RGB lights
-            // take the max(R, G, B) and use it to write the light value
-            uint8_t light_value[9] = { 0 };
-            for (uint32_t i = 0; i < number_of_leds; i += 3) {
-                light_value[i / 3] =
-                    std::max({ ((uint8_t *)i_pData)[i], ((uint8_t *)i_pData)[i + 1], ((uint8_t *)i_pData)[i + 2] });
-            }
-
-            constexpr Lights::popn_lights_t legacy_button_lights[] = {
-                Lights::popn_lights_t::Button1,
-                Lights::popn_lights_t::Button2,
-                Lights::popn_lights_t::Button3,
-                Lights::popn_lights_t::Button4,
-                Lights::popn_lights_t::Button5,
-                Lights::popn_lights_t::Button6,
-                Lights::popn_lights_t::Button7,
-                Lights::popn_lights_t::Button8,
-                Lights::popn_lights_t::Button9
-            };
-
-            static_assert(std::size(legacy_button_lights) == 9);
-
-            for (size_t light = 0; light < 9; light++) {
-                // on the new cab, buttons are colorless plastic and they rely on RGB to be lit at all times, even when "off"
-                // when translating to legacy on/off lights, treat anything above ~60% brightness as fully on,
-                // otherwise off, to avoid dimly lit "off" state
-                GameAPI::Lights::writeLight(
-                    RI_MGR,
-                    lights.at(legacy_button_lights[light]),
-                    light_value[light] > 150 ? 1.f : 0.f);
-            }
-
-            // color buttons
-            constexpr Lights::popn_lights_t button_lights[] = {
-                Lights::popn_lights_t::PikaButton1_R, Lights::popn_lights_t::PikaButton1_G, Lights::popn_lights_t::PikaButton1_B,
-                Lights::popn_lights_t::PikaButton2_R, Lights::popn_lights_t::PikaButton2_G, Lights::popn_lights_t::PikaButton2_B,
-                Lights::popn_lights_t::PikaButton3_R, Lights::popn_lights_t::PikaButton3_G, Lights::popn_lights_t::PikaButton3_B,
-                Lights::popn_lights_t::PikaButton4_R, Lights::popn_lights_t::PikaButton4_G, Lights::popn_lights_t::PikaButton4_B,
-                Lights::popn_lights_t::PikaButton5_R, Lights::popn_lights_t::PikaButton5_G, Lights::popn_lights_t::PikaButton5_B,
-                Lights::popn_lights_t::PikaButton6_R, Lights::popn_lights_t::PikaButton6_G, Lights::popn_lights_t::PikaButton6_B,
-                Lights::popn_lights_t::PikaButton7_R, Lights::popn_lights_t::PikaButton7_G, Lights::popn_lights_t::PikaButton7_B,
-                Lights::popn_lights_t::PikaButton8_R, Lights::popn_lights_t::PikaButton8_G, Lights::popn_lights_t::PikaButton8_B,
-                Lights::popn_lights_t::PikaButton9_R, Lights::popn_lights_t::PikaButton9_G, Lights::popn_lights_t::PikaButton9_B
-            };
-
-            static_assert(std::size(button_lights) == 9 * 3);
-
-            for (size_t light = 0; light < number_of_leds; light++) {
-                GameAPI::Lights::writeLight(
-                    RI_MGR,
-                    lights.at(button_lights[light]),
-                    ((uint8_t *)i_pData)[light] / 255.f);
-            }
+            // SetTapeLedDataPart is meant to be a temporary staging buffer, so the game occasionally
+            // clears the buffer (set to 0 / lights off) between animations but they are not meant
+            // to be seen by the I/O
+            // therefore, instead of writing the lights out directly here, we stage them in a buffer
+            // and wait for GetDeviceStatus to be called
+            std::lock_guard<std::mutex> lock(pika_button_staging_mutex);
+            memcpy(pika_button_staging.data(), i_pData, pika_button_staging.size());
+            pika_button_staging_dirty = true;
         }
 
         if (tapeledutils::is_enabled() && i_CnPin > 0 && i_CnPin < std::size(TAPELED_MAPPING)) {

src/spice2x/games/silentscope/projector.cpp

@@ -0,0 +1,150 @@
+#include "projector.h"
+
+#include <algorithm>
+#include <cstring>
+#include <cwchar>
+
+#include "util/logging.h"
+
+namespace {
+
+    // request/response layout: identifier, command, two zero bytes, payload length,
+    // the payload itself and a checksum over everything before it
+    constexpr size_t HEADER_SIZE = 5;
+    constexpr uint8_t RESPONSE_ID = 0x23;
+
+    constexpr uint8_t COMMAND_COMMON_DATA = 0x8A;
+    constexpr uint8_t COMMAND_TEMPERATURE = 0x99;
+    constexpr uint8_t COMMAND_LAMP_CURRENT = 0x9B;
+
+    // shown in the test menu as PROJTIM, PROJHEATIN and PROJHEAT
+    constexpr uint32_t LAMP_SECONDS = 0;
+    constexpr uint32_t TEMPERATURE_INTAKE = 25;
+    constexpr uint32_t TEMPERATURE_EXHAUST = 35;
+
+    // anything below 1000 is treated as a lamp anomaly and raises PROJERROR unless the exact
+    // value was already recorded in /projecter/current on an earlier boot
+    constexpr uint32_t LAMP_CURRENT = 1000;
+
+    void put32(uint8_t *dest, uint32_t value) {
+        dest[0] = (uint8_t) (value & 0xFF);
+        dest[1] = (uint8_t) ((value >> 8) & 0xFF);
+        dest[2] = (uint8_t) ((value >> 16) & 0xFF);
+        dest[3] = (uint8_t) ((value >> 24) & 0xFF);
+    }
+}
+
+bool games::silentscope::ProjectorHandle::open(LPCWSTR lpFileName) {
+    if (wcscmp(lpFileName, L"COM2") != 0 && wcscmp(lpFileName, L"\\\\.\\COM2") != 0) {
+        return false;
+    }
+
+    log_info("silentscope", "Opened COM2 (projector)");
+
+    return true;
+}
+
+int games::silentscope::ProjectorHandle::read(LPVOID lpBuffer, DWORD nNumberOfBytesToRead) {
+    std::lock_guard<std::mutex> lock(this->mutex);
+
+    auto buffer = reinterpret_cast<uint8_t *>(lpBuffer);
+
+    DWORD bytes_read = 0;
+    while (bytes_read < nNumberOfBytesToRead && !this->response.empty()) {
+        buffer[bytes_read++] = this->response.front();
+        this->response.pop_front();
+    }
+
+    return (int) bytes_read;
+}
+
+int games::silentscope::ProjectorHandle::write(LPCVOID lpBuffer, DWORD nNumberOfBytesToWrite) {
+    std::lock_guard<std::mutex> lock(this->mutex);
+
+    auto buffer = reinterpret_cast<const uint8_t *>(lpBuffer);
+    this->request.insert(this->request.end(), buffer, buffer + nNumberOfBytesToWrite);
+
+    while (this->request.size() >= HEADER_SIZE) {
+        const size_t packet_size = HEADER_SIZE + this->request[4] + 1;
+        if (this->request.size() < packet_size) {
+            break;
+        }
+
+        this->process_request(this->request.data());
+        this->request.erase(this->request.begin(), this->request.begin() + packet_size);
+    }
+
+    return (int) nNumberOfBytesToWrite;
+}
+
+size_t games::silentscope::ProjectorHandle::bytes_available() {
+    std::lock_guard<std::mutex> lock(this->mutex);
+
+    return this->response.size();
+}
+
+bool games::silentscope::ProjectorHandle::close() {
+    std::lock_guard<std::mutex> lock(this->mutex);
+
+    this->request.clear();
+    this->response.clear();
+
+    log_info("silentscope", "Closed COM2 (projector)");
+
+    return true;
+}
+
+void games::silentscope::ProjectorHandle::process_request(const uint8_t *packet) {
+    const uint8_t command = packet[1];
+    const uint8_t length = packet[4];
+    const uint8_t *data = packet + HEADER_SIZE;
+
+    switch (command) {
+        case COMMAND_COMMON_DATA: {
+
+            // the game only looks at the lamp usage time near the end of the block
+            std::vector<uint8_t> payload(98, 0);
+            put32(&payload[94], LAMP_SECONDS);
+            this->reply(command, payload);
+            break;
+        }
+        case COMMAND_TEMPERATURE: {
+
+            // the requested sensor is echoed back along with its reading
+            const uint8_t sensor = length > 0 ? data[0] : 0;
+            std::vector<uint8_t> payload(5, 0);
+            payload[0] = sensor;
+            put32(&payload[1], sensor == 0 ? TEMPERATURE_INTAKE : TEMPERATURE_EXHAUST);
+            this->reply(command, payload);
+            break;
+        }
+        case COMMAND_LAMP_CURRENT: {
+
+            // the three byte item selector is echoed back along with the measurement
+            std::vector<uint8_t> payload(7, 0);
+            memcpy(payload.data(), data, std::min<size_t>(length, 3));
+            put32(&payload[3], LAMP_CURRENT);
+            this->reply(command, payload);
+            break;
+        }
+        default:
+            log_misc("silentscope", "unknown projector command {:#04x}", command);
+            this->reply(command, {});
+            break;
+    }
+}
+
+void games::silentscope::ProjectorHandle::reply(uint8_t command, const std::vector<uint8_t> &data) {
+    std::vector<uint8_t> packet {
+        RESPONSE_ID, command, 0x00, 0x00, (uint8_t) data.size()
+    };
+    packet.insert(packet.end(), data.begin(), data.end());
+
+    uint8_t checksum = 0;
+    for (auto byte : packet) {
+        checksum += byte;
+    }
+    packet.push_back(checksum);
+
+    this->response.insert(this->response.end(), packet.begin(), packet.end());
+}

src/spice2x/games/silentscope/projector.h

@@ -0,0 +1,30 @@
+#pragma once
+
+#include <cstdint>
+#include <deque>
+#include <mutex>
+#include <vector>
+
+#include "hooks/devicehook.h"
+
+namespace games::silentscope {
+
+    // The cabinet talks to its projector over COM2. Without an answer the game stops at
+    // I/O error 5-1560-0004 (IOCOM2, "the projector is not connected correctly").
+    class ProjectorHandle : public CustomHandle {
+    public:
+        bool open(LPCWSTR lpFileName) override;
+        int read(LPVOID lpBuffer, DWORD nNumberOfBytesToRead) override;
+        int write(LPCVOID lpBuffer, DWORD nNumberOfBytesToWrite) override;
+        size_t bytes_available() override;
+        bool close() override;
+
+    private:
+        std::mutex mutex;
+        std::vector<uint8_t> request;
+        std::deque<uint8_t> response;
+
+        void process_request(const uint8_t *packet);
+        void reply(uint8_t command, const std::vector<uint8_t> &data);
+    };
+}

src/spice2x/games/silentscope/silentscope.cpp

@@ -1,10 +1,11 @@
 #include "silentscope.h"
 
-#include "acioemu/handle.h"
 #include "cfg/configurator.h"
 #include "hooks/devicehook.h"
 #include "util/libutils.h"
 
+#include "projector.h"
+
 namespace games::silentscope {
 
     SilentScopeGame::SilentScopeGame() : Game("Silent Scope") {
@@ -16,9 +17,8 @@ namespace games::silentscope {
         // load the game DLL so hooks apply
         libutils::try_library("gamendd.dll");
 
-        // TODO(felix): implement stuff on this port
         devicehook_init();
-        devicehook_add(new acioemu::ACIOHandle(L"COM2"));
+        devicehook_add(new ProjectorHandle());
     }
 
     void SilentScopeGame::detach() {

src/spice2x/hooks/graphics/backends/d3d11/d3d11_backend.cpp

@@ -30,6 +30,7 @@ void graphics_d3d11_shutdown() {}
 #include <dxgi.h>
 #include <dxgi1_2.h>
 
+#include "avs/game.h"
 #include "d3d11_internal.h"
 #include "util/nt_loader.h"
 
@@ -237,10 +238,11 @@ bool d3dcompiler_available() {
 } // namespace
 
 void graphics_d3d11_init() {
-    // dx11 titles always run under execexe. skipping on pure-dx9 games keeps
-    // their startup path completely untouched (no exports patched, no poll
-    // thread, no LDR callback).
-    if (!GetModuleHandleW(L"execexe.dll")) {
+    // dx11 titles run under execexe, except Silent Scope: Bone Eater, whose Aska
+    // engine LoadLibrary's d3d11 itself. skipping on pure-dx9 games keeps their
+    // startup path completely untouched (no exports patched, no poll thread, no
+    // LDR callback).
+    if (!GetModuleHandleW(L"execexe.dll") && !avs::game::is_model("NDD")) {
         return;
     }
 

src/spice2x/hooks/graphics/backends/d3d9/d3d9_backend.cpp

@@ -944,9 +944,12 @@ HRESULT STDMETHODCALLTYPE WrappedIDirect3D9::CreateDevice(
     } else if (!D3D9_DEVICE_HOOK_DISABLE) {
         graphics_hook_window(hFocusWindow, pPresentationParameters);
 
-        *ppReturnedDeviceInterface = new WrappedIDirect3DDevice9(
+        auto *wrapped = new WrappedIDirect3DDevice9(
                 hFocusWindow,
                 *ppReturnedDeviceInterface);
+
+        wrapped->device_multithreaded = (BehaviorFlags & D3DCREATE_MULTITHREADED) != 0;
+        *ppReturnedDeviceInterface = wrapped;
     }
 
     // return result
@@ -1307,13 +1310,16 @@ HRESULT STDMETHODCALLTYPE WrappedIDirect3D9::CreateDeviceEx(
     } else if (!D3D9_DEVICE_HOOK_DISABLE) {
         graphics_hook_window(hFocusWindow, pPresentationParameters);
 
-        *ppReturnedDeviceInterface = new WrappedIDirect3DDevice9(
+        auto *wrapped = new WrappedIDirect3DDevice9(
                 hFocusWindow,
                 *ppReturnedDeviceInterface,
                 gfdm_parameters.logical_small_swapchain,
                 gfdm_two_head_exclusive() ? static_cast<IDirect3D9 *>(this) : nullptr,
                 gfdm_two_head_exclusive() ? pPresentationParameters : nullptr);
 
+        wrapped->device_multithreaded = (BehaviorFlags & D3DCREATE_MULTITHREADED) != 0;
+        *ppReturnedDeviceInterface = wrapped;
+
         // initialize sub screen if the game requested a multi-head context
         if (avs::game::is_model({"LDJ", "KFC", "M39", "M32"}) &&
             (orig_behavior_flags & D3DCREATE_ADAPTERGROUP_DEVICE)) {

src/spice2x/hooks/graphics/backends/d3d9/d3d9_device.cpp

@@ -634,6 +634,9 @@ HRESULT STDMETHODCALLTYPE WrappedIDirect3DDevice9::Reset(
         overlay::OVERLAY->reset_invalidate();
     }
 
+    // Reset refuses to run while any default pool resource is outstanding
+    d3d9_readback::discard_snapshot_targets(pReal);
+
     HRESULT res = pReal->Reset(pPresentationParameters);
 
     // recreate overlay
@@ -2321,6 +2324,9 @@ HRESULT STDMETHODCALLTYPE WrappedIDirect3DDevice9::ResetEx(
         overlay::OVERLAY->reset_invalidate();
     }
 
+    // ResetEx refuses to run while any default pool resource is outstanding
+    d3d9_readback::discard_snapshot_targets(pReal);
+
     HRESULT res = static_cast<IDirect3DDevice9Ex *>(pReal)->ResetEx(
             gfdm_parameters.presentation_parameters,
             gfdm_parameters.fullscreen_display_modes);

src/spice2x/hooks/graphics/backends/d3d9/d3d9_device.h

@@ -264,6 +264,10 @@ struct WrappedIDirect3DDevice9 : IDirect3DDevice9Ex {
     IDirect3DDevice9 *pReal;
     bool is_d3d9ex = false;
 
+    // set from the creation flags; without it the runtime does no locking of its own, so
+    // nothing but the present thread may touch the device
+    bool device_multithreaded = false;
+
     std::atomic_ulong refs = 1;
 
     WrappedIDirect3DSwapChain9 *main_swapchain = nullptr;

src/spice2x/hooks/graphics/backends/d3d9/d3d9_readback.cpp

@@ -1,5 +1,8 @@
 #include "d3d9_readback.h"
 
+#include <array>
+#include <chrono>
+#include <condition_variable>
 #include <mutex>
 #include <vector>
 
@@ -10,6 +13,13 @@ namespace d3d9_readback {
 
 namespace {
 
+// the snapshot path stamps frames with this to recognise one left behind by a break in the
+// request stream
+uint64_t now_us() {
+    return static_cast<uint64_t>(std::chrono::duration_cast<std::chrono::microseconds>(
+            std::chrono::steady_clock::now().time_since_epoch()).count());
+}
+
 SurfacePtr create_readback_surface(IDirect3DDevice9 *device, const D3DSURFACE_DESC &desc) {
     IDirect3DSurface9 *surface = nullptr;
     const HRESULT hr = device->CreateOffscreenPlainSurface(
@@ -160,21 +170,8 @@ ReadbackPool &pool() {
     return *instance;
 }
 
-} // namespace
-
-void release_device_resources(IDirect3DDevice9 *device) {
-    pool().clear_device(device);
-}
-
-BackbufferCopy::~BackbufferCopy() {
-    if (this->pooled && this->surface) {
-        pool().release(this->device, std::move(this->surface));
-    }
-}
-
-std::optional<BackbufferCopy> acquire_backbuffer_copy(
-        IDirect3DDevice9 *device, IDirect3DSwapChain9 *swap_chain, int screen, bool pooled) {
-
+// the back buffer plus the checks every caller has to make before copying out of it
+SurfacePtr open_backbuffer(IDirect3DSwapChain9 *swap_chain, int screen, D3DSURFACE_DESC &desc) {
     IDirect3DSurface9 *buffer = nullptr;
     HRESULT hr = swap_chain->GetBackBuffer(0, D3DBACKBUFFER_TYPE_MONO, &buffer);
     if (FAILED(hr) || buffer == nullptr) {
@@ -182,17 +179,17 @@ std::optional<BackbufferCopy> acquire_backbuffer_copy(
                 "failed to get back buffer for screen {}, hr={}",
                 screen,
                 FMT_HRESULT(hr));
-        return std::nullopt;
+        return nullptr;
     }
 
-    D3DSURFACE_DESC desc {};
-    hr = buffer->GetDesc(&desc);
+    SurfacePtr surface(buffer);
+
+    hr = surface->GetDesc(&desc);
     if (FAILED(hr)) {
         log_warning("graphics::d3d9",
                 "failed to acquire back buffer descriptor, hr={}",
                 FMT_HRESULT(hr));
-        buffer->Release();
-        return std::nullopt;
+        return nullptr;
     }
 
     // GetRenderTargetData rejects multisampled sources. no supported game has been
@@ -204,7 +201,399 @@ std::optional<BackbufferCopy> acquire_backbuffer_copy(
                     "back buffer is multisampled ({}), screenshots and capture are unsupported",
                     static_cast<uint32_t>(desc.MultiSampleType));
         });
-        buffer->Release();
+        return nullptr;
+    }
+
+    return surface;
+}
+
+SurfacePtr create_snapshot_target(IDirect3DDevice9 *device, const D3DSURFACE_DESC &desc) {
+    IDirect3DSurface9 *surface = nullptr;
+
+    // matching the back buffer keeps the blit a straight copy and leaves the pixels in the
+    // format the conversion step would have seen without the detour
+    const HRESULT hr = device->CreateRenderTarget(
+            desc.Width, desc.Height, desc.Format,
+            D3DMULTISAMPLE_NONE, 0, FALSE, &surface, nullptr);
+
+    if (FAILED(hr) || surface == nullptr) {
+        log_warning("graphics::d3d9",
+                "failed to create snapshot target, hr={}",
+                FMT_HRESULT(hr));
+        return nullptr;
+    }
+
+    return SurfacePtr(surface);
+}
+
+// two render targets per screen, reused across frames. two because one holds the frame waiting
+// out its deferral while the other takes the next one; a read only holds its target until the
+// pixels are in system memory, which is far shorter than the gap between requests.
+//
+// unlike the readback surfaces these live in the default pool, so they have to be gone before a
+// Reset and not merely before the device is released
+class SnapshotTargets {
+public:
+    SurfacePtr acquire(
+            IDirect3DDevice9 *device,
+            int screen,
+            const D3DSURFACE_DESC &desc,
+            uint64_t *out_generation) {
+
+        if (screen < 0 || screen >= static_cast<int>(GRAPHICS_CAPTURE_SCREEN_NO)) {
+            return nullptr;
+        }
+
+        std::lock_guard<std::mutex> lock(this->mutex);
+
+        if (this->device != device) {
+            this->drop();
+            this->device = device;
+            this->usable = true;
+        }
+
+        auto &entry = this->screens[screen];
+
+        Slot *free_slot = nullptr;
+        for (size_t i = 0; i < SLOTS_PER_SCREEN; i++) {
+            auto &candidate = entry.slots[(entry.next + i) % SLOTS_PER_SCREEN];
+            if (!candidate.busy) {
+                free_slot = &candidate;
+                entry.next = (entry.next + i + 1) % SLOTS_PER_SCREEN;
+                break;
+            }
+        }
+
+        if (free_slot == nullptr) {
+            return nullptr;
+        }
+
+        auto &slot = *free_slot;
+
+        if (slot.target
+                && (slot.width != desc.Width
+                        || slot.height != desc.Height
+                        || slot.format != desc.Format)) {
+            slot.target.reset();
+        }
+
+        if (!slot.target) {
+            slot.target = create_snapshot_target(device, desc);
+            if (!slot.target) {
+                // a back buffer format the device will not give us a render target for. the
+                // inline path can still read it, so stop trying rather than lose the stream
+                this->usable = false;
+                return nullptr;
+            }
+
+            slot.width = desc.Width;
+            slot.height = desc.Height;
+            slot.format = desc.Format;
+        }
+
+        slot.busy = true;
+        *out_generation = this->current_generation;
+
+        slot.target->AddRef();
+        return SurfacePtr(slot.target.get());
+    }
+
+    // holds the fresh snapshot back and returns the one from the previous request, which by now
+    // has had a full frame for its blit and transfer to land
+    std::optional<Snapshot> rotate(int screen, std::optional<Snapshot> fresh) {
+        if (screen < 0 || screen >= static_cast<int>(GRAPHICS_CAPTURE_SCREEN_NO)) {
+            return std::nullopt;
+        }
+
+        const uint64_t now = now_us();
+        std::optional<Snapshot> previous;
+
+        {
+            std::lock_guard<std::mutex> lock(this->mutex);
+            previous = std::move(this->pending[screen]);
+            this->pending[screen] = std::move(fresh);
+        }
+
+        // a frame waits here for the next request rather than for a deadline, so a gap in
+        // requests, a client reconnecting most obviously, would otherwise hand the new session
+        // a frame from before the gap. showing a stale frame is worse than showing none
+        if (previous.has_value() && now - previous->issued_us > MAX_DEFERRAL_US) {
+            return std::nullopt;
+        }
+
+        return previous;
+    }
+
+    // a reader whose target was already thrown away must not free the slot, or it would free
+    // the snapshot that replaced it while that one is still being read
+    void finish(int screen, uint64_t generation, IDirect3DSurface9 *target) {
+        if (screen < 0 || screen >= static_cast<int>(GRAPHICS_CAPTURE_SCREEN_NO)) {
+            return;
+        }
+
+        {
+            std::lock_guard<std::mutex> lock(this->mutex);
+            if (generation != this->current_generation) {
+                return;
+            }
+
+            this->free_slot(screen, target);
+        }
+
+        this->idle.notify_all();
+    }
+
+    bool still_current(uint64_t generation) {
+        std::lock_guard<std::mutex> lock(this->mutex);
+        return generation == this->current_generation;
+    }
+
+    bool is_usable() {
+        std::lock_guard<std::mutex> lock(this->mutex);
+        return this->usable;
+    }
+
+    // Reset fails outright while the device still owns default pool resources, so dropping our
+    // own references is not enough and any read in flight has to finish first. the wait is
+    // bounded because a failed Reset is something games retry and a stalled present thread is
+    // not something they survive
+    void discard(IDirect3DDevice9 *device) {
+        // destroyed after the lock is released, since dropping a snapshot calls back in here
+        std::array<std::optional<Snapshot>, GRAPHICS_CAPTURE_SCREEN_NO> stale;
+
+        {
+            std::unique_lock<std::mutex> lock(this->mutex);
+            if (this->device != nullptr && this->device != device) {
+                return;
+            }
+
+            // deferred frames are abandoned rather than waited for; only a read that is already
+            // running has to be allowed to finish
+            for (auto &held : this->pending) {
+                if (held.has_value() && held->surface) {
+                    this->free_slot(held->screen, held->surface.get());
+                }
+            }
+
+            stale = std::move(this->pending);
+            this->pending = {};
+
+            const bool drained = this->idle.wait_for(
+                    lock,
+                    std::chrono::milliseconds(100),
+                    [this] { return !this->any_busy(); });
+
+            if (!drained) {
+                log_warning("graphics::d3d9",
+                        "capture snapshot still in flight, discarding its target anyway");
+            }
+
+            this->drop();
+            this->device = nullptr;
+        }
+    }
+
+private:
+    static constexpr size_t SLOTS_PER_SCREEN = 2;
+
+    // generous next to the frame interval this is meant to bridge, so that ordinary jitter
+    // never trips it and only a real break in the request stream does
+    static constexpr uint64_t MAX_DEFERRAL_US = 250'000;
+
+    struct Slot {
+        SurfacePtr target;
+        UINT width = 0;
+        UINT height = 0;
+        D3DFORMAT format = D3DFMT_UNKNOWN;
+        bool busy = false;
+    };
+
+    struct Screen {
+        std::array<Slot, SLOTS_PER_SCREEN> slots;
+        size_t next = 0;
+    };
+
+    void free_slot(int screen, IDirect3DSurface9 *target) {
+        for (auto &slot : this->screens[screen].slots) {
+            if (slot.target.get() == target) {
+                slot.busy = false;
+                return;
+            }
+        }
+    }
+
+    bool any_busy() const {
+        for (const auto &entry : this->screens) {
+            for (const auto &slot : entry.slots) {
+                if (slot.busy) {
+                    return true;
+                }
+            }
+        }
+
+        return false;
+    }
+
+    void drop() {
+        for (auto &entry : this->screens) {
+            for (auto &slot : entry.slots) {
+                slot.target.reset();
+                slot.busy = false;
+            }
+
+            entry.next = 0;
+        }
+
+        this->current_generation++;
+    }
+
+    std::mutex mutex;
+    std::condition_variable idle;
+    std::array<Screen, GRAPHICS_CAPTURE_SCREEN_NO> screens;
+    std::array<std::optional<Snapshot>, GRAPHICS_CAPTURE_SCREEN_NO> pending;
+    IDirect3DDevice9 *device = nullptr;
+    uint64_t current_generation = 1;
+    bool usable = true;
+};
+
+// never destroyed, for the same reason the readback pool is not
+SnapshotTargets &targets() {
+    static SnapshotTargets *instance = new SnapshotTargets();
+    return *instance;
+}
+
+} // namespace
+
+void release_device_resources(IDirect3DDevice9 *device) {
+    targets().discard(device);
+    pool().clear_device(device);
+}
+
+void discard_snapshot_targets(IDirect3DDevice9 *device) {
+    targets().discard(device);
+}
+
+bool snapshots_supported() {
+    return targets().is_usable();
+}
+
+BackbufferCopy::~BackbufferCopy() {
+    if (this->pooled && this->surface) {
+        pool().release(this->device, std::move(this->surface));
+    }
+}
+
+Snapshot::~Snapshot() {
+    // still holding the target means the read never ran, and the slot would otherwise stay
+    // marked busy and take the screen out of capture permanently
+    if (this->surface) {
+        targets().finish(this->screen, this->generation, this->surface.get());
+    }
+}
+
+namespace {
+
+std::optional<Snapshot> take_snapshot(
+        IDirect3DDevice9 *device, IDirect3DSwapChain9 *swap_chain, int screen) {
+
+    const uint64_t started_us = now_us();
+
+    D3DSURFACE_DESC desc {};
+    auto buffer = open_backbuffer(swap_chain, screen, desc);
+    if (!buffer) {
+        return std::nullopt;
+    }
+
+    uint64_t generation = 0;
+    auto target = targets().acquire(device, screen, desc, &generation);
+    if (!target) {
+        return std::nullopt;
+    }
+
+    // built before the blit so that a failure below hands the slot back through the destructor
+    Snapshot snapshot;
+    snapshot.screen = screen;
+    snapshot.desc = desc;
+    snapshot.device = device;
+    snapshot.surface = std::move(target);
+    snapshot.generation = generation;
+    snapshot.issued_us = started_us;
+
+    // the point of the whole exercise: this is queued rather than waited on, so the game pays
+    // for issuing the copy and not for it completing. identical size and format, so there is
+    // no filtering to ask for
+    const HRESULT hr = device->StretchRect(
+            buffer.get(), nullptr, snapshot.surface.get(), nullptr, D3DTEXF_NONE);
+
+    if (FAILED(hr)) {
+        log_warning("graphics::d3d9",
+                "failed to snapshot back buffer for screen {}, hr={}",
+                screen,
+                FMT_HRESULT(hr));
+        return std::nullopt;
+    }
+
+    return snapshot;
+}
+
+} // namespace
+
+std::optional<Snapshot> snapshot_backbuffer(
+        IDirect3DDevice9 *device, IDirect3DSwapChain9 *swap_chain, int screen) {
+
+    return targets().rotate(screen, take_snapshot(device, swap_chain, screen));
+}
+
+std::optional<BackbufferCopy> read_snapshot(Snapshot snapshot) {
+    if (!snapshot.surface) {
+        return std::nullopt;
+    }
+
+    // a Reset between the blit and now means the target no longer holds the captured frame
+    if (!targets().still_current(snapshot.generation)) {
+        return std::nullopt;
+    }
+
+    auto destination = pool().acquire(snapshot.device, snapshot.desc);
+    if (!destination) {
+        return std::nullopt;
+    }
+
+    const HRESULT hr = snapshot.device->GetRenderTargetData(
+            snapshot.surface.get(), destination.get());
+
+    // the target is reusable as soon as the pixels are in system memory. dropping the
+    // reference before freeing the slot keeps the destructor from freeing it twice
+    const int screen = snapshot.screen;
+    const uint64_t generation = snapshot.generation;
+    IDirect3DSurface9 *target = snapshot.surface.get();
+    snapshot.surface.reset();
+    targets().finish(screen, generation, target);
+
+    if (FAILED(hr)) {
+        log_warning("graphics::d3d9",
+                "failed to read snapshot contents, hr={}",
+                FMT_HRESULT(hr));
+        pool().release(snapshot.device, std::move(destination));
+        return std::nullopt;
+    }
+
+    BackbufferCopy copy;
+    copy.screen = screen;
+    copy.desc = snapshot.desc;
+    copy.device = snapshot.device;
+    copy.surface = std::move(destination);
+    copy.pooled = true;
+
+    return copy;
+}
+
+std::optional<BackbufferCopy> acquire_backbuffer_copy(
+        IDirect3DDevice9 *device, IDirect3DSwapChain9 *swap_chain, int screen, bool pooled) {
+
+    D3DSURFACE_DESC desc {};
+    auto buffer = open_backbuffer(swap_chain, screen, desc);
+    if (!buffer) {
         return std::nullopt;
     }
 
@@ -212,12 +601,10 @@ std::optional<BackbufferCopy> acquire_backbuffer_copy(
             ? pool().acquire(device, desc)
             : create_readback_surface(device, desc);
     if (!destination) {
-        buffer->Release();
         return std::nullopt;
     }
 
-    hr = device->GetRenderTargetData(buffer, destination.get());
-    buffer->Release();
+    const HRESULT hr = device->GetRenderTargetData(buffer.get(), destination.get());
 
     if (FAILED(hr)) {
         log_warning("graphics::d3d9",

src/spice2x/hooks/graphics/backends/d3d9/d3d9_readback.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <cstdint>
 #include <memory>
 #include <optional>
 
@@ -39,6 +40,51 @@ namespace d3d9_readback {
             int screen,
             bool pooled);
 
+    // GPU side copy of a back buffer, taken while the contents are still the frame that was
+    // presented, so that reading them into system memory no longer has to happen before it
+    struct Snapshot {
+        int screen {};
+        D3DSURFACE_DESC desc {};
+        IDirect3DDevice9 *device = nullptr;
+        SurfacePtr surface;
+        uint64_t generation {};
+
+        // when the blit was issued, so a frame left behind by a break in the request stream can
+        // be recognised as stale rather than handed over
+        uint64_t issued_us {};
+
+        Snapshot() = default;
+        Snapshot(Snapshot &&) noexcept = default;
+        Snapshot &operator=(Snapshot &&) noexcept = default;
+        Snapshot(const Snapshot &) = delete;
+        Snapshot &operator=(const Snapshot &) = delete;
+        ~Snapshot();
+    };
+
+    // for the present thread, between the last EndScene and Present. blits the current frame,
+    // then returns the snapshot taken on the *previous* call: waiting a frame before reading
+    // means the blit and its system memory transfer have already happened, so the read does not
+    // stall on the GPU. costs the stream one frame of latency.
+    //
+    // returns nothing on the first call of a stream, and whenever the frame could not be taken,
+    // which is the caller's cue to skip rather than to wait
+    std::optional<Snapshot> snapshot_backbuffer(
+            IDirect3DDevice9 *device,
+            IDirect3DSwapChain9 *swap_chain,
+            int screen);
+
+    // the expensive half, for a thread that is not the present thread. only legal on a device
+    // created with D3DCREATE_MULTITHREADED
+    std::optional<BackbufferCopy> read_snapshot(Snapshot snapshot);
+
+    // false once a device has refused to give up a render target matching its back buffer,
+    // which leaves reading the back buffer directly as the only way to capture it
+    bool snapshots_supported();
+
+    // snapshot targets live in the default pool, so unlike the readback surfaces they have to
+    // be gone before a Reset and not merely before the device is released
+    void discard_snapshot_targets(IDirect3DDevice9 *device);
+
     // pooled surfaces hold references on the device; call this before releasing it
     void release_device_resources(IDirect3DDevice9 *device);
 }

src/spice2x/hooks/graphics/backends/d3d9/d3d9_screenshot.cpp

@@ -488,6 +488,63 @@ ThreadPool &capture_read_pool() {
     return *instance;
 }
 
+// Takes the frame on the present thread as a queued GPU blit and hands the readback to a pool
+// thread, so the game waits for neither. Only viable where the whole read can go off thread,
+// since the back buffer is overwritten right after Present and a snapshot the present thread
+// then had to read itself would cost more than reading the back buffer directly.
+//
+// Returns false when the frame could not be taken, including the ordinary case of the previous
+// snapshot of this screen still being read, which paces capture to what the reader sustains.
+static bool snapshot_capture(
+        IDirect3DDevice9 *device,
+        WrappedIDirect3DDevice9 *wrapped_device,
+        int screen) {
+
+    IDirect3DSwapChain9 *swap_chain = nullptr;
+    const HRESULT hr = wrapped_device->get_screenshot_swap_chain(screen, &swap_chain);
+    if (FAILED(hr) || swap_chain == nullptr) {
+        log_warning("graphics::d3d9",
+                "failed to get swap chain for screen {}, hr={}",
+                screen,
+                FMT_HRESULT(hr));
+        return false;
+    }
+
+    auto snapshot = d3d9_readback::snapshot_backbuffer(device, swap_chain, screen);
+    swap_chain->Release();
+
+    if (!snapshot.has_value()) {
+        return false;
+    }
+
+    try {
+        capture_read_pool().add([screen, snapshot = std::move(*snapshot)]() mutable {
+            // an escape from here would cross a thread boundary and terminate
+            try {
+                auto copy = d3d9_readback::read_snapshot(std::move(snapshot));
+                if (!copy.has_value()) {
+                    graphics_capture_skip(screen);
+                    return;
+                }
+
+                read_and_dispatch_capture(screen, std::move(*copy));
+            } catch (const std::exception &error) {
+                log_warning("graphics::d3d9", "capture read failed: {}", error.what());
+                graphics_capture_skip(screen);
+            } catch (...) {
+                log_warning("graphics::d3d9", "capture read failed");
+                graphics_capture_skip(screen);
+            }
+        });
+    } catch (const std::exception &) {
+        // the snapshot went into the lambda before the queue could fail, so it is already
+        // destroyed and its target handed back; the client just misses this frame
+        return false;
+    }
+
+    return true;
+}
+
 // by this point the pixels are plain memory, so none of this needs the device
 static void dispatch_screenshot_save(std::vector<PendingWrite> writes, size_t screen_count) {
     auto screenshot_process = [writes = std::move(writes), screen_count]() mutable {
@@ -632,6 +689,18 @@ static void process_image_request(
         const ImageRequest &request) {
     const bool screenshot = request.kind == ImageRequestKind::Screenshot;
 
+    if (!screenshot
+            && wrapped_device->device_multithreaded
+            && capture_read_off_thread(request.screen)
+            && d3d9_readback::snapshots_supported()) {
+
+        if (!snapshot_capture(device, wrapped_device, request.screen)) {
+            graphics_capture_skip(request.screen);
+        }
+
+        return;
+    }
+
     std::vector<int> screens { request.screen };
     if (screenshot && GRAPHICS_SCREENSHOT_SUBSCREENS) {
         screens.clear();

src/spice2x/hooks/graphics/graphics.cpp

@@ -52,6 +52,8 @@ static HWND GFDM_RIGHT_WINDOW = nullptr;
 static HMONITOR GFDM_TWO_HEAD_SMALL_MONITOR = nullptr;
 static HWND GFDM_TWO_HEAD_SMALL_WINDOW = nullptr;
 HWND POPN_SUBSCREEN_WINDOW = nullptr;
+static HWND NDD_SUBSCREEN_WINDOW = nullptr;
+HWND NDD_MAIN_WINDOW = nullptr;
 bool FAKE_SUBSCREEN_ADAPTER = false;
 
 // icon
@@ -634,6 +636,9 @@ static HWND WINAPI CreateWindowExA_hook(DWORD dwExStyle, LPCSTR lpClassName, LPC
     bool is_sdvx_sub_window = is_sdvx && window_name.ends_with(" Sub Screen");
     bool is_sdvx_main_window = is_sdvx && window_name.ends_with(" Main Screen");
     bool is_popn_sub_window = avs::game::is_model("M39") && window_name.ends_with("Sub Screen");
+    const bool is_ndd = avs::game::is_model("NDD");
+    bool is_ndd_sub_window = is_ndd && window_name.starts_with("Aska MultiDisplay");
+    bool is_ndd_main_window = is_ndd && window_name == "ASKA";
     const std::string gfdm_window_name = games::gitadora::is_arena_model()
         ? gitadora_canonical_window_name(effective_window_name)
         : "";
@@ -777,6 +782,14 @@ static HWND WINAPI CreateWindowExA_hook(DWORD dwExStyle, LPCSTR lpClassName, LPC
         }
     }
 
+    if (is_ndd_sub_window) {
+        NDD_SUBSCREEN_WINDOW = result;
+    }
+
+    if (is_ndd_main_window) {
+        NDD_MAIN_WINDOW = result;
+    }
+
     disable_touch_gestures(result);
     log_misc(
         "graphics",
@@ -902,6 +915,26 @@ static BOOL WINAPI EnumDisplayDevicesA_hook(LPCTSTR lpDevice, DWORD iDevNum,
     return value;
 }
 
+// the sub screen renders into a fixed 800x480 buffer, but the game's saved layout asks for rects
+// that do not match it, and dxgi stretches the buffer to fill whatever the client area ends up as
+static void ndd_subscreen_size(HWND hWnd, int &width, int &height) {
+    RECT rect {};
+    SetRect(&rect, 0, 0, 800, 480);
+    AdjustWindowRect(&rect, GetWindowLongA(hWnd, GWL_STYLE), 0);
+
+    width = rect.right - rect.left;
+    height = rect.bottom - rect.top;
+}
+
+// the saved layout drops the sub window wherever it sat on the machine that wrote the file
+static void ndd_subscreen_position(int &x, int &y) {
+    RECT main {};
+    if (NDD_MAIN_WINDOW != nullptr && GetWindowRect(NDD_MAIN_WINDOW, &main)) {
+        x = main.right;
+        y = main.top;
+    }
+}
+
 static BOOL WINAPI MoveWindow_hook(HWND hWnd, int X, int Y, int nWidth, int nHeight, BOOL bRepaint) {
     log_misc("graphics", "MoveWindow hook hit ({}, {}, {}, {}, {}, {})",
         fmt::ptr(hWnd),
@@ -931,6 +964,11 @@ static BOOL WINAPI MoveWindow_hook(HWND hWnd, int X, int Y, int nWidth, int nHei
         nHeight = rect.bottom - rect.top;
     }
 
+    if (GRAPHICS_WINDOWED && NDD_SUBSCREEN_WINDOW && hWnd == NDD_SUBSCREEN_WINDOW) {
+        ndd_subscreen_size(hWnd, nWidth, nHeight);
+        ndd_subscreen_position(X, Y);
+    }
+
     // iidx windowed TDJ mode
     if (GRAPHICS_WINDOWED && TDJ_SUBSCREEN_WINDOW && hWnd == TDJ_SUBSCREEN_WINDOW) {
        if (GRAPHICS_IIDX_WSUB) {
@@ -1073,6 +1111,15 @@ static LONG WINAPI SetWindowLongW_hook(HWND hWnd, int nIndex, LONG dwNewLong) {
 static BOOL WINAPI SetWindowPos_hook(HWND hWnd, HWND hWndInsertAfter,
         int X, int Y, int cx, int cy, UINT uFlags) {
 
+    if (GRAPHICS_WINDOWED && NDD_SUBSCREEN_WINDOW && hWnd == NDD_SUBSCREEN_WINDOW) {
+        if (!(uFlags & SWP_NOSIZE)) {
+            ndd_subscreen_size(hWnd, cx, cy);
+        }
+        if (!(uFlags & SWP_NOMOVE)) {
+            ndd_subscreen_position(X, Y);
+        }
+    }
+
     if (is_gfdm_two_head_small_window(hWnd) &&
             ((uFlags & SWP_HIDEWINDOW) ||
              (uFlags & (SWP_NOMOVE | SWP_NOSIZE)) != (SWP_NOMOVE | SWP_NOSIZE))) {

src/spice2x/hooks/graphics/graphics.h

@@ -103,6 +103,7 @@ extern HWND TDJ_SUBSCREEN_WINDOW;
 extern HWND SDVX_SUBSCREEN_WINDOW;
 extern HWND POPN_SUBSCREEN_WINDOW;
 extern HWND GFDM_SUBSCREEN_WINDOW;
+extern HWND NDD_MAIN_WINDOW;
 
 extern bool SUBSCREEN_FORCE_REDRAW;
 extern bool FAKE_SUBSCREEN_ADAPTER;

src/spice2x/hooks/lang.cpp

@@ -336,9 +336,11 @@ void hooks::lang::early_init() {
             &GetLocaleInfoA_orig);
     }
 
-    // for TDJ subscreen search keyboard and T44 narrow-string handling
+    // for TDJ subscreen search keyboard
+    // T44 narrow-string handling
+    // NDD text measuring
     if ((avs::game::is_model("LDJ") && games::iidx::TDJ_MODE) ||
-        avs::game::is_model("T44")) {
+        avs::game::is_model({ "T44", "NDD" })) {
         log_info("hooks::lang", "hooking IsDBCSLeadByte");
         detour::trampoline_try(
             "kernel32.dll",
@@ -358,8 +360,9 @@ void hooks::lang::early_init() {
 #endif
 
 #ifdef SPICE64
+    // NDD renders through GetTextExtentPoint32A, so its wide strings go back through CP_ACP first
     const auto hook_wide_char_to_multi_byte =
-        games::gitadora::is_arena_model() || avs::game::is_model("T44");
+        games::gitadora::is_arena_model() || avs::game::is_model({ "T44", "NDD" });
 #else
     // XG2 converts UTF-8 property strings through CP_ACP before rendering.
     const auto hook_wide_char_to_multi_byte = avs::game::is_model({ "K32", "K33" });

src/spice2x/launcher/options.cpp

@@ -1729,7 +1729,7 @@ static const std::vector<OptionDefinition> OPTION_DEFINITIONS = {
             "Value must be between 0 (poor quality) and 100 (best quality), inclusive. Default: 70.",
         .type = OptionType::Integer,
         .setting_name = "(0-100)",
-        .category = "Companion & API",
+        .category = "API Dev",
     },
     {
         // APIScreenMirrorDivide
@@ -1741,21 +1741,17 @@ static const std::vector<OptionDefinition> OPTION_DEFINITIONS = {
             "Value must be 1 or greater. Default: 1.",
         .type = OptionType::Integer,
         .setting_name = "1",
-        .category = "Companion & API",
+        .category = "API Dev",
     },
     {
         // APIStreamEnable
         .title = "API Video Stream Server Enable (EXPERIMENTAL)",
         .name = "apistream",
-        .desc = "Serves the mirrored screen as a video stream, on the API port plus two; "
-            "alternative to API screen capture. Requires -api.\n\n"
-            "http://host:apiport+2/stream.mjpg - MJPEG\n\n"
-            "http://host:apiport+2/stream.h264 - H.264\n\n"
-            "Parameters: screen (0-3), fps (1-60, default 30), q (1-100, default 70).\n\n"
-            "Example with -api 1337: http://host:1339/stream.h264?fps=30&q=70\n\n"
-            "VIEW ONLY - touch input still requires -api. "
-            "No password protection or encryption of any kind; video sent in the clear!\n\n"
-            "Streaming is known to cause older games to hang and crash.",
+        .desc = "Allows companion apps to receive compressed video streams over the API. "
+            "Companion app must support this feature to take advantage of it.\n\n"
+            "Video is served unencrypted over the network. "
+            "May cause older games to hang and crash.\n\n"
+            "Developers: see the wiki page for more details.",
         .type = OptionType::Bool,
         .category = "Companion & API",
     },

src/spice2x/misc/device.cpp

@@ -285,8 +285,10 @@ static int __cdecl device_read_secplug(int a1, int a2, int a3) {
     return 1;
 }
 
-static int __cdecl device_set_coinblocker_open(char number, char open) {
-    return 0;
+static void __cdecl device_set_coinblocker_open(char slot1_open, char slot2_open) {
+    // libdevice exposes one open flag per coin slot. Since spice has a shared
+    // software coin stock, consider it blocked only when both slots are closed.
+    eamuse_coin_set_block(!slot1_open && !slot2_open);
 }
 
 static void __cdecl device_set_coincounter_merge() {
@@ -617,7 +619,9 @@ void spicedevice_attach() {
     detour::inline_hook((void *) device_finalize, libutils::try_proc_list(
             DEVICE_INSTANCE, {"device_finalize", "?device_finalize@@YAXXZ"}));
     detour::inline_hook((void *) device_get_coinstock, libutils::try_proc_list(
-            DEVICE_INSTANCE, {"device_get_coinstock", "?device_get_coinstock@@YAXPEAG0@Z"}));
+            DEVICE_INSTANCE, {"device_get_coinstock",
+                              "?device_get_coinstock@@YAXPEAG0@Z",
+                              "?device_get_coinstock@@YAXPAG0@Z"}));
     detour::inline_hook((void *) device_get_coinstock_all, libutils::try_proc_list(
             DEVICE_INSTANCE, {"device_get_coinstock_all",
                               "?device_get_coinstock_all@@YAXPEAG0@Z",

src/spice2x/misc/eamuse.cpp

@@ -34,6 +34,7 @@ static uint16_t KEYPAD_STATE_OVERRIDES_READER[] = {0, 0};
 static uint16_t KEYPAD_STATE_OVERRIDES_OVERLAY[] = {0, 0};
 static std::string EAMUSE_GAME_NAME;
 static ConfigKeypadBindings KEYPAD_BINDINGS {};
+static std::mutex KEYPAD_BINDINGS_LOCK;
 
 // auto card
 bool AUTO_INSERT_CARD[2] = {false, false};
@@ -87,23 +88,23 @@ bool eamuse_get_card(int active_count, int unit_id, uint8_t *card) {
         return true;
     }
 
-    // get file path
-    std::filesystem::path path;
-    if (!KEYPAD_BINDINGS.card_paths[index].empty()) {
-        path = KEYPAD_BINDINGS.card_paths[index];
-    } else {
-        path = index > 0 ? "card1.txt" : "card0.txt";
-    }
-
     // call the next function
-    return eamuse_get_card(path, card, index);
+    return eamuse_get_card(eamuse_get_card_path(index), card, index);
+}
+
+std::filesystem::path eamuse_get_card_path(size_t index) {
+    std::lock_guard<std::mutex> lock(KEYPAD_BINDINGS_LOCK);
+    if (index >= std::size(KEYPAD_BINDINGS.card_paths)) {
+        return {};
+    }
+    if (!KEYPAD_BINDINGS.card_paths[index].empty()) {
+        return KEYPAD_BINDINGS.card_paths[index];
+    }
+    return index > 0 ? "card1.txt" : "card0.txt";
 }
 
 bool eamuse_get_card(const std::filesystem::path &path, uint8_t *card, int index) {
-    // do a quick copy under lock
-    std::unique_lock<std::mutex> lock(CARD_OVERRIDES_LOCK);
-    const auto card_override = CARD_OVERRIDES[index];
-    lock.unlock();
+    const auto card_override = eamuse_get_card_override(index);
 
     // Check if card overrides are present
     if (!card_override.empty()) {
@@ -149,6 +150,14 @@ bool eamuse_get_card(const std::filesystem::path &path, uint8_t *card, int index
     return eamuse_get_card_from_file(path, card, index);
 }
 
+std::string eamuse_get_card_override(size_t index) {
+    std::lock_guard<std::mutex> lock(CARD_OVERRIDES_LOCK);
+    if (index >= std::size(CARD_OVERRIDES)) {
+        return {};
+    }
+    return CARD_OVERRIDES[index];
+}
+
 bool eamuse_get_card_from_file(const std::filesystem::path &path, uint8_t *card, int index) {
 
     // open file
@@ -584,6 +593,7 @@ std::string eamuse_get_keypad_state_str(size_t unit) {
 }
 
 bool eamuse_keypad_state_naive() {
+    std::lock_guard<std::mutex> lock(KEYPAD_BINDINGS_LOCK);
     return KEYPAD_BINDINGS.keypads[0].empty() && KEYPAD_BINDINGS.keypads[1].empty();
 }
 
@@ -595,7 +605,9 @@ void eamuse_set_game(std::string game) {
 }
 
 void eamuse_update_keypad_bindings() {
-    KEYPAD_BINDINGS = Config::getInstance().getKeypadBindings(EAMUSE_GAME_NAME);
+    auto bindings = Config::getInstance().getKeypadBindings(EAMUSE_GAME_NAME);
+    std::lock_guard<std::mutex> lock(KEYPAD_BINDINGS_LOCK);
+    KEYPAD_BINDINGS = std::move(bindings);
 }
 
 const std::string &eamuse_get_game() {
@@ -718,6 +730,8 @@ void eamuse_autodetect_game() {
         eamuse_set_game("Mahjong Fight Girl");
     else if (avs::game::is_model("XIF"))
         eamuse_set_game("Polaris Chord");
+    else if (avs::game::is_model("NDD"))
+        eamuse_set_game("Silent Scope: Bone Eater");
     else {
         log_warning("eamuse", "unknown game model: {}", avs::game::MODEL);
         eamuse_set_game("unknown");

src/spice2x/misc/eamuse.h

@@ -43,6 +43,8 @@ extern std::string AUTO_PIN_MACRO_TRIGGER[2];
 bool eamuse_get_card(int active_count, int unit_id, uint8_t *card);
 bool eamuse_get_card(const std::filesystem::path &path, uint8_t *card, int unit_id);
 bool eamuse_get_card_from_file(const std::filesystem::path &path, uint8_t *card, int index);
+std::filesystem::path eamuse_get_card_path(size_t index);
+std::string eamuse_get_card_override(size_t index);
 
 void eamuse_card_insert(int unit);
 void eamuse_card_insert(int unit, const uint8_t *card);