Fix incorrect id check and potential out of bounds lookup

input_common: rename PAGE_SIZE to avoid conflict

.ci/scripts/windows/upload.ps1

@@ -26,7 +26,11 @@ $env:BUILD_ZIP = $MSVC_BUILD_ZIP
 $env:BUILD_SYMBOLS = $MSVC_BUILD_PDB
 $env:BUILD_UPDATE = $MSVC_SEVENZIP
 
-$BUILD_DIR = ".\build\bin\Release"
+if (Test-Path -Path ".\build\bin\Release") {
+    $BUILD_DIR = ".\build\bin\Release"
+} else {
+    $BUILD_DIR = ".\build\bin\"
+}
 
 # Cleanup unneeded data in submodules
 git submodule foreach git clean -fxd

CMakeModules/CopyYuzuFFmpegDeps.cmake

@@ -3,7 +3,7 @@
 
 function(copy_yuzu_FFmpeg_deps target_dir)
     include(WindowsCopyFiles)
-    set(DLL_DEST "${CMAKE_BINARY_DIR}/bin/$<CONFIG>/")
+    set(DLL_DEST "$<TARGET_FILE_DIR:${target_dir}>/")
     file(READ "${FFmpeg_PATH}/requirements.txt" FFmpeg_REQUIRED_DLLS)
     string(STRIP "${FFmpeg_REQUIRED_DLLS}" FFmpeg_REQUIRED_DLLS)
     windows_copy_files(${target_dir} ${FFmpeg_DLL_DIR} ${DLL_DEST} ${FFmpeg_REQUIRED_DLLS})

CMakeModules/CopyYuzuQt5Deps.cmake

@@ -4,7 +4,7 @@
 function(copy_yuzu_Qt5_deps target_dir)
     include(WindowsCopyFiles)
     if (MSVC)
-        set(DLL_DEST "${CMAKE_BINARY_DIR}/bin/$<CONFIG>/")
+        set(DLL_DEST "$<TARGET_FILE_DIR:${target_dir}>/")
         set(Qt5_DLL_DIR "${Qt5_DIR}/../../../bin")
     else()
         set(DLL_DEST "${CMAKE_BINARY_DIR}/bin/")

CMakeModules/CopyYuzuSDLDeps.cmake

@@ -3,6 +3,6 @@
 
 function(copy_yuzu_SDL_deps target_dir)
     include(WindowsCopyFiles)
-    set(DLL_DEST "${CMAKE_BINARY_DIR}/bin/$<CONFIG>/")
+    set(DLL_DEST "$<TARGET_FILE_DIR:${target_dir}>/")
     windows_copy_files(${target_dir} ${SDL2_DLL_DIR} ${DLL_DEST} SDL2.dll)
 endfunction(copy_yuzu_SDL_deps)

externals/CMakeLists.txt

@@ -8,15 +8,21 @@ set(CMAKE_POLICY_DEFAULT_CMP0077 NEW)
 # Disable tests in all externals supporting the standard option name
 set(BUILD_TESTING OFF)
 
+# Build only static externals
+set(BUILD_SHARED_LIBS OFF)
+
+# Skip install rules for all externals
+set_directory_properties(PROPERTIES EXCLUDE_FROM_ALL ON)
+
 # xbyak
 if ((ARCHITECTURE_x86 OR ARCHITECTURE_x86_64) AND NOT TARGET xbyak::xbyak)
-    add_subdirectory(xbyak EXCLUDE_FROM_ALL)
+    add_subdirectory(xbyak)
 endif()
 
 # Dynarmic
 if ((ARCHITECTURE_x86_64 OR ARCHITECTURE_arm64) AND NOT TARGET dynarmic::dynarmic)
     set(DYNARMIC_IGNORE_ASSERTS ON)
-    add_subdirectory(dynarmic EXCLUDE_FROM_ALL)
+    add_subdirectory(dynarmic)
     add_library(dynarmic::dynarmic ALIAS dynarmic)
 endif()
 
@@ -34,7 +40,7 @@ if (NOT TARGET inih::INIReader)
 endif()
 
 # mbedtls
-add_subdirectory(mbedtls EXCLUDE_FROM_ALL)
+add_subdirectory(mbedtls)
 target_include_directories(mbedtls PUBLIC ./mbedtls/include)
 
 # MicroProfile
@@ -48,7 +54,7 @@ endif()
 
 # libusb
 if (ENABLE_LIBUSB AND NOT TARGET libusb::usb)
-    add_subdirectory(libusb EXCLUDE_FROM_ALL)
+    add_subdirectory(libusb)
 endif()
 
 # SDL2
@@ -67,18 +73,16 @@ if (YUZU_USE_EXTERNAL_SDL2)
 
         set(HIDAPI ON)
     endif()
-    set(SDL_STATIC ON)
-    set(SDL_SHARED OFF)
     if (APPLE)
         set(SDL_FILE ON)
     endif()
 
-    add_subdirectory(SDL EXCLUDE_FROM_ALL)
+    add_subdirectory(SDL)
 endif()
 
 # ENet
 if (NOT TARGET enet::enet)
-    add_subdirectory(enet EXCLUDE_FROM_ALL)
+    add_subdirectory(enet)
     target_include_directories(enet INTERFACE ./enet/include)
     add_library(enet::enet ALIAS enet)
 endif()
@@ -86,24 +90,26 @@ endif()
 # Cubeb
 if (ENABLE_CUBEB AND NOT TARGET cubeb::cubeb)
     set(BUILD_TESTS OFF)
-    add_subdirectory(cubeb EXCLUDE_FROM_ALL)
+    set(BUILD_TOOLS OFF)
+    add_subdirectory(cubeb)
     add_library(cubeb::cubeb ALIAS cubeb)
 endif()
 
 # DiscordRPC
 if (USE_DISCORD_PRESENCE AND NOT TARGET DiscordRPC::discord-rpc)
-    add_subdirectory(discord-rpc EXCLUDE_FROM_ALL)
+    set(BUILD_EXAMPLES OFF)
+    add_subdirectory(discord-rpc)
     target_include_directories(discord-rpc INTERFACE ./discord-rpc/include)
     add_library(DiscordRPC::discord-rpc ALIAS discord-rpc)
 endif()
 
 # Sirit
-add_subdirectory(sirit EXCLUDE_FROM_ALL)
+add_subdirectory(sirit)
 
 # httplib
 if (ENABLE_WEB_SERVICE AND NOT TARGET httplib::httplib)
     set(HTTPLIB_REQUIRE_OPENSSL ON)
-    add_subdirectory(cpp-httplib EXCLUDE_FROM_ALL)
+    add_subdirectory(cpp-httplib)
 endif()
 
 # cpp-jwt
@@ -111,12 +117,12 @@ if (ENABLE_WEB_SERVICE AND NOT TARGET cpp-jwt::cpp-jwt)
     set(CPP_JWT_BUILD_EXAMPLES OFF)
     set(CPP_JWT_BUILD_TESTS OFF)
     set(CPP_JWT_USE_VENDORED_NLOHMANN_JSON OFF)
-    add_subdirectory(cpp-jwt EXCLUDE_FROM_ALL)
+    add_subdirectory(cpp-jwt)
 endif()
 
 # Opus
 if (NOT TARGET Opus::opus)
-    add_subdirectory(opus EXCLUDE_FROM_ALL)
+    add_subdirectory(opus)
 endif()
 
 # FFMpeg
@@ -130,12 +136,14 @@ endif()
 
 # Vulkan-Headers
 if (YUZU_USE_EXTERNAL_VULKAN_HEADERS)
-    add_subdirectory(Vulkan-Headers EXCLUDE_FROM_ALL)
+    add_subdirectory(Vulkan-Headers)
 endif()
 
 if (NOT TARGET LLVM::Demangle)
-    add_library(demangle STATIC)
+    add_library(demangle demangle/ItaniumDemangle.cpp)
     target_include_directories(demangle PUBLIC ./demangle)
-    target_sources(demangle PRIVATE demangle/ItaniumDemangle.cpp)
     add_library(LLVM::Demangle ALIAS demangle)
 endif()
+
+add_library(stb stb/stb_dxt.cpp)
+target_include_directories(stb PUBLIC ./stb)

externals/glad/CMakeLists.txt

@@ -1,7 +1,7 @@
 # SPDX-FileCopyrightText: 2015 Yuri Kunde Schlesner <yuriks@yuriks.net>
 # SPDX-License-Identifier: GPL-2.0-or-later
 
-add_library(glad STATIC
+add_library(glad
     src/glad.c
     include/KHR/khrplatform.h
     include/glad/glad.h

externals/libusb/CMakeLists.txt

@@ -122,7 +122,7 @@ else() # MINGW OR (${CMAKE_SYSTEM_NAME} MATCHES "Linux")
         add_compile_options(/utf-8)
     endif()
 
-    add_library(usb STATIC EXCLUDE_FROM_ALL
+    add_library(usb
         libusb/libusb/core.c
         libusb/libusb/core.c
         libusb/libusb/descriptor.c

externals/microprofile/microprofile.h

@@ -1697,7 +1697,13 @@ void MicroProfileFlip()
                             {
                                 int nTimer = MicroProfileLogTimerIndex(LE);
                                 uint8_t nGroup = pTimerToGroup[nTimer];
-                                MP_ASSERT(nStackPos < MICROPROFILE_STACK_MAX);
+
+                                // To avoid crashing due to OOB memory accesses/asserts
+                                // simply skip this iteration
+                                // MP_ASSERT(nStackPos < MICROPROFILE_STACK_MAX);
+                                if (nStackPos >= MICROPROFILE_STACK_MAX) {
+                                    break;
+                                }
                                 MP_ASSERT(nGroup < MICROPROFILE_MAX_GROUPS);
                                 pGroupStackPos[nGroup]++;
                                 pStack[nStackPos++] = k;

externals/opus/CMakeLists.txt

@@ -23,7 +23,7 @@ else()
     endif()
 endif()
 
-add_library(opus STATIC
+add_library(opus
     # CELT sources
     opus/celt/bands.c
     opus/celt/celt.c

externals/stb/stb_dxt.cpp

@@ -0,0 +1,765 @@
+// SPDX-FileCopyrightText: fabian "ryg" giesen
+// SPDX-License-Identifier: MIT
+
+// stb_dxt.h - v1.12 - DXT1/DXT5 compressor
+
+#include <stb_dxt.h>
+
+#include <stdlib.h>
+#include <string.h>
+
+#if !defined(STBD_FABS)
+#include <math.h>
+#endif
+
+#ifndef STBD_FABS
+#define STBD_FABS(x) fabs(x)
+#endif
+
+static const unsigned char stb__OMatch5[256][2] = {
+    {0, 0},   {0, 0},   {0, 1},   {0, 1},   {1, 0},   {1, 0},   {1, 0},   {1, 1},   {1, 1},
+    {1, 1},   {1, 2},   {0, 4},   {2, 1},   {2, 1},   {2, 1},   {2, 2},   {2, 2},   {2, 2},
+    {2, 3},   {1, 5},   {3, 2},   {3, 2},   {4, 0},   {3, 3},   {3, 3},   {3, 3},   {3, 4},
+    {3, 4},   {3, 4},   {3, 5},   {4, 3},   {4, 3},   {5, 2},   {4, 4},   {4, 4},   {4, 5},
+    {4, 5},   {5, 4},   {5, 4},   {5, 4},   {6, 3},   {5, 5},   {5, 5},   {5, 6},   {4, 8},
+    {6, 5},   {6, 5},   {6, 5},   {6, 6},   {6, 6},   {6, 6},   {6, 7},   {5, 9},   {7, 6},
+    {7, 6},   {8, 4},   {7, 7},   {7, 7},   {7, 7},   {7, 8},   {7, 8},   {7, 8},   {7, 9},
+    {8, 7},   {8, 7},   {9, 6},   {8, 8},   {8, 8},   {8, 9},   {8, 9},   {9, 8},   {9, 8},
+    {9, 8},   {10, 7},  {9, 9},   {9, 9},   {9, 10},  {8, 12},  {10, 9},  {10, 9},  {10, 9},
+    {10, 10}, {10, 10}, {10, 10}, {10, 11}, {9, 13},  {11, 10}, {11, 10}, {12, 8},  {11, 11},
+    {11, 11}, {11, 11}, {11, 12}, {11, 12}, {11, 12}, {11, 13}, {12, 11}, {12, 11}, {13, 10},
+    {12, 12}, {12, 12}, {12, 13}, {12, 13}, {13, 12}, {13, 12}, {13, 12}, {14, 11}, {13, 13},
+    {13, 13}, {13, 14}, {12, 16}, {14, 13}, {14, 13}, {14, 13}, {14, 14}, {14, 14}, {14, 14},
+    {14, 15}, {13, 17}, {15, 14}, {15, 14}, {16, 12}, {15, 15}, {15, 15}, {15, 15}, {15, 16},
+    {15, 16}, {15, 16}, {15, 17}, {16, 15}, {16, 15}, {17, 14}, {16, 16}, {16, 16}, {16, 17},
+    {16, 17}, {17, 16}, {17, 16}, {17, 16}, {18, 15}, {17, 17}, {17, 17}, {17, 18}, {16, 20},
+    {18, 17}, {18, 17}, {18, 17}, {18, 18}, {18, 18}, {18, 18}, {18, 19}, {17, 21}, {19, 18},
+    {19, 18}, {20, 16}, {19, 19}, {19, 19}, {19, 19}, {19, 20}, {19, 20}, {19, 20}, {19, 21},
+    {20, 19}, {20, 19}, {21, 18}, {20, 20}, {20, 20}, {20, 21}, {20, 21}, {21, 20}, {21, 20},
+    {21, 20}, {22, 19}, {21, 21}, {21, 21}, {21, 22}, {20, 24}, {22, 21}, {22, 21}, {22, 21},
+    {22, 22}, {22, 22}, {22, 22}, {22, 23}, {21, 25}, {23, 22}, {23, 22}, {24, 20}, {23, 23},
+    {23, 23}, {23, 23}, {23, 24}, {23, 24}, {23, 24}, {23, 25}, {24, 23}, {24, 23}, {25, 22},
+    {24, 24}, {24, 24}, {24, 25}, {24, 25}, {25, 24}, {25, 24}, {25, 24}, {26, 23}, {25, 25},
+    {25, 25}, {25, 26}, {24, 28}, {26, 25}, {26, 25}, {26, 25}, {26, 26}, {26, 26}, {26, 26},
+    {26, 27}, {25, 29}, {27, 26}, {27, 26}, {28, 24}, {27, 27}, {27, 27}, {27, 27}, {27, 28},
+    {27, 28}, {27, 28}, {27, 29}, {28, 27}, {28, 27}, {29, 26}, {28, 28}, {28, 28}, {28, 29},
+    {28, 29}, {29, 28}, {29, 28}, {29, 28}, {30, 27}, {29, 29}, {29, 29}, {29, 30}, {29, 30},
+    {30, 29}, {30, 29}, {30, 29}, {30, 30}, {30, 30}, {30, 30}, {30, 31}, {30, 31}, {31, 30},
+    {31, 30}, {31, 30}, {31, 31}, {31, 31},
+};
+static const unsigned char stb__OMatch6[256][2] = {
+    {0, 0},   {0, 1},   {1, 0},   {1, 1},   {1, 1},   {1, 2},   {2, 1},   {2, 2},   {2, 2},
+    {2, 3},   {3, 2},   {3, 3},   {3, 3},   {3, 4},   {4, 3},   {4, 4},   {4, 4},   {4, 5},
+    {5, 4},   {5, 5},   {5, 5},   {5, 6},   {6, 5},   {6, 6},   {6, 6},   {6, 7},   {7, 6},
+    {7, 7},   {7, 7},   {7, 8},   {8, 7},   {8, 8},   {8, 8},   {8, 9},   {9, 8},   {9, 9},
+    {9, 9},   {9, 10},  {10, 9},  {10, 10}, {10, 10}, {10, 11}, {11, 10}, {8, 16},  {11, 11},
+    {11, 12}, {12, 11}, {9, 17},  {12, 12}, {12, 13}, {13, 12}, {11, 16}, {13, 13}, {13, 14},
+    {14, 13}, {12, 17}, {14, 14}, {14, 15}, {15, 14}, {14, 16}, {15, 15}, {15, 16}, {16, 14},
+    {16, 15}, {17, 14}, {16, 16}, {16, 17}, {17, 16}, {18, 15}, {17, 17}, {17, 18}, {18, 17},
+    {20, 14}, {18, 18}, {18, 19}, {19, 18}, {21, 15}, {19, 19}, {19, 20}, {20, 19}, {20, 20},
+    {20, 20}, {20, 21}, {21, 20}, {21, 21}, {21, 21}, {21, 22}, {22, 21}, {22, 22}, {22, 22},
+    {22, 23}, {23, 22}, {23, 23}, {23, 23}, {23, 24}, {24, 23}, {24, 24}, {24, 24}, {24, 25},
+    {25, 24}, {25, 25}, {25, 25}, {25, 26}, {26, 25}, {26, 26}, {26, 26}, {26, 27}, {27, 26},
+    {24, 32}, {27, 27}, {27, 28}, {28, 27}, {25, 33}, {28, 28}, {28, 29}, {29, 28}, {27, 32},
+    {29, 29}, {29, 30}, {30, 29}, {28, 33}, {30, 30}, {30, 31}, {31, 30}, {30, 32}, {31, 31},
+    {31, 32}, {32, 30}, {32, 31}, {33, 30}, {32, 32}, {32, 33}, {33, 32}, {34, 31}, {33, 33},
+    {33, 34}, {34, 33}, {36, 30}, {34, 34}, {34, 35}, {35, 34}, {37, 31}, {35, 35}, {35, 36},
+    {36, 35}, {36, 36}, {36, 36}, {36, 37}, {37, 36}, {37, 37}, {37, 37}, {37, 38}, {38, 37},
+    {38, 38}, {38, 38}, {38, 39}, {39, 38}, {39, 39}, {39, 39}, {39, 40}, {40, 39}, {40, 40},
+    {40, 40}, {40, 41}, {41, 40}, {41, 41}, {41, 41}, {41, 42}, {42, 41}, {42, 42}, {42, 42},
+    {42, 43}, {43, 42}, {40, 48}, {43, 43}, {43, 44}, {44, 43}, {41, 49}, {44, 44}, {44, 45},
+    {45, 44}, {43, 48}, {45, 45}, {45, 46}, {46, 45}, {44, 49}, {46, 46}, {46, 47}, {47, 46},
+    {46, 48}, {47, 47}, {47, 48}, {48, 46}, {48, 47}, {49, 46}, {48, 48}, {48, 49}, {49, 48},
+    {50, 47}, {49, 49}, {49, 50}, {50, 49}, {52, 46}, {50, 50}, {50, 51}, {51, 50}, {53, 47},
+    {51, 51}, {51, 52}, {52, 51}, {52, 52}, {52, 52}, {52, 53}, {53, 52}, {53, 53}, {53, 53},
+    {53, 54}, {54, 53}, {54, 54}, {54, 54}, {54, 55}, {55, 54}, {55, 55}, {55, 55}, {55, 56},
+    {56, 55}, {56, 56}, {56, 56}, {56, 57}, {57, 56}, {57, 57}, {57, 57}, {57, 58}, {58, 57},
+    {58, 58}, {58, 58}, {58, 59}, {59, 58}, {59, 59}, {59, 59}, {59, 60}, {60, 59}, {60, 60},
+    {60, 60}, {60, 61}, {61, 60}, {61, 61}, {61, 61}, {61, 62}, {62, 61}, {62, 62}, {62, 62},
+    {62, 63}, {63, 62}, {63, 63}, {63, 63},
+};
+
+static int stb__Mul8Bit(int a, int b) {
+    int t = a * b + 128;
+    return (t + (t >> 8)) >> 8;
+}
+
+static void stb__From16Bit(unsigned char* out, unsigned short v) {
+    int rv = (v & 0xf800) >> 11;
+    int gv = (v & 0x07e0) >> 5;
+    int bv = (v & 0x001f) >> 0;
+
+    // expand to 8 bits via bit replication
+    out[0] = static_cast<unsigned char>((rv * 33) >> 2);
+    out[1] = static_cast<unsigned char>((gv * 65) >> 4);
+    out[2] = static_cast<unsigned char>((bv * 33) >> 2);
+    out[3] = 0;
+}
+
+static unsigned short stb__As16Bit(int r, int g, int b) {
+    return (unsigned short)((stb__Mul8Bit(r, 31) << 11) + (stb__Mul8Bit(g, 63) << 5) +
+                            stb__Mul8Bit(b, 31));
+}
+
+// linear interpolation at 1/3 point between a and b, using desired rounding
+// type
+static int stb__Lerp13(int a, int b) {
+#ifdef STB_DXT_USE_ROUNDING_BIAS
+    // with rounding bias
+    return a + stb__Mul8Bit(b - a, 0x55);
+#else
+    // without rounding bias
+    // replace "/ 3" by "* 0xaaab) >> 17" if your compiler sucks or you really
+    // need every ounce of speed.
+    return (2 * a + b) / 3;
+#endif
+}
+
+// linear interpolation at 1/2 point between a and b
+static int stb__Lerp12(int a, int b) {
+    return (a + b) / 2;
+}
+
+// lerp RGB color
+static void stb__Lerp13RGB(unsigned char* out, unsigned char* p1, unsigned char* p2) {
+    out[0] = (unsigned char)stb__Lerp13(p1[0], p2[0]);
+    out[1] = (unsigned char)stb__Lerp13(p1[1], p2[1]);
+    out[2] = (unsigned char)stb__Lerp13(p1[2], p2[2]);
+}
+
+static void stb__Lerp12RGB(unsigned char* out, unsigned char* p1, unsigned char* p2) {
+    out[0] = (unsigned char)stb__Lerp12(p1[0], p2[0]);
+    out[1] = (unsigned char)stb__Lerp12(p1[1], p2[1]);
+    out[2] = (unsigned char)stb__Lerp12(p1[2], p2[2]);
+}
+
+/****************************************************************************/
+
+static void stb__Eval4Colors(unsigned char* color, unsigned short c0, unsigned short c1) {
+    stb__From16Bit(color + 0, c0);
+    stb__From16Bit(color + 4, c1);
+    stb__Lerp13RGB(color + 8, color + 0, color + 4);
+    stb__Lerp13RGB(color + 12, color + 4, color + 0);
+}
+
+static void stb__Eval3Colors(unsigned char* color, unsigned short c0, unsigned short c1) {
+    stb__From16Bit(color + 0, c0);
+    stb__From16Bit(color + 4, c1);
+    stb__Lerp12RGB(color + 8, color + 0, color + 4);
+}
+
+// The color matching function
+static unsigned int stb__MatchColorsBlock(unsigned char* block, unsigned char* color) {
+    unsigned int mask = 0;
+    int dirr = color[0 * 4 + 0] - color[1 * 4 + 0];
+    int dirg = color[0 * 4 + 1] - color[1 * 4 + 1];
+    int dirb = color[0 * 4 + 2] - color[1 * 4 + 2];
+    int dots[16];
+    int stops[4];
+    int i;
+    int c0Point, halfPoint, c3Point;
+
+    for (i = 0; i < 16; i++)
+        dots[i] = block[i * 4 + 0] * dirr + block[i * 4 + 1] * dirg + block[i * 4 + 2] * dirb;
+
+    for (i = 0; i < 4; i++)
+        stops[i] = color[i * 4 + 0] * dirr + color[i * 4 + 1] * dirg + color[i * 4 + 2] * dirb;
+
+    // think of the colors as arranged on a line; project point onto that line,
+    // then choose next color out of available ones. we compute the crossover
+    // points for "best color in top half"/"best in bottom half" and then the same
+    // inside that subinterval.
+    //
+    // relying on this 1d approximation isn't always optimal in terms of euclidean
+    // distance, but it's very close and a lot faster.
+    // http://cbloomrants.blogspot.com/2008/12/12-08-08-dxtc-summary.html
+
+    c0Point = (stops[1] + stops[3]);
+    halfPoint = (stops[3] + stops[2]);
+    c3Point = (stops[2] + stops[0]);
+
+    for (i = 15; i >= 0; i--) {
+        int dot = dots[i] * 2;
+        mask <<= 2;
+
+        if (dot < halfPoint)
+            mask |= (dot < c0Point) ? 1 : 3;
+        else
+            mask |= (dot < c3Point) ? 2 : 0;
+    }
+
+    return mask;
+}
+
+static unsigned int stb__MatchColorsAlphaBlock(unsigned char* block, unsigned char* color) {
+    unsigned int mask = 0;
+    int dirr = color[0 * 4 + 0] - color[1 * 4 + 0];
+    int dirg = color[0 * 4 + 1] - color[1 * 4 + 1];
+    int dirb = color[0 * 4 + 2] - color[1 * 4 + 2];
+    int dots[16];
+    int stops[3];
+    int i;
+    int c0Point, c2Point;
+
+    for (i = 0; i < 16; i++)
+        dots[i] = block[i * 4 + 0] * dirr + block[i * 4 + 1] * dirg + block[i * 4 + 2] * dirb;
+
+    for (i = 0; i < 3; i++)
+        stops[i] = color[i * 4 + 0] * dirr + color[i * 4 + 1] * dirg + color[i * 4 + 2] * dirb;
+
+    c0Point = (stops[1] + stops[2]);
+    c2Point = (stops[2] + stops[0]);
+
+    for (i = 15; i >= 0; i--) {
+        int dot = dots[i] * 2;
+        mask <<= 2;
+
+        if (block[i * 4 + 3] == 0)
+            mask |= 3;
+        else if (dot < c2Point)
+            mask |= (dot < c0Point) ? 0 : 2;
+        else
+            mask |= (dot < c0Point) ? 1 : 0;
+    }
+
+    return mask;
+}
+
+static void stb__ReorderColors(unsigned short* pmax16, unsigned short* pmin16) {
+    if (*pmin16 < *pmax16) {
+        unsigned short t = *pmin16;
+        *pmin16 = *pmax16;
+        *pmax16 = t;
+    }
+}
+
+static void stb__FinalizeColors(unsigned short* pmax16, unsigned short* pmin16,
+                                unsigned int* pmask) {
+    if (*pmax16 < *pmin16) {
+        unsigned short t = *pmin16;
+        *pmin16 = *pmax16;
+        *pmax16 = t;
+        *pmask ^= 0x55555555;
+    }
+}
+
+// The color optimization function. (Clever code, part 1)
+static void stb__OptimizeColorsBlock(unsigned char* block, unsigned short* pmax16,
+                                     unsigned short* pmin16) {
+    int mind, maxd;
+    unsigned char *minp, *maxp;
+    double magn;
+    int v_r, v_g, v_b;
+    static const int nIterPower = 4;
+    float covf[6], vfr, vfg, vfb;
+
+    // determine color distribution
+    int cov[6];
+    int mu[3], min[3], max[3];
+    int ch, i, iter;
+
+    for (ch = 0; ch < 3; ch++) {
+        const unsigned char* bp = ((const unsigned char*)block) + ch;
+        int muv, minv, maxv;
+
+        muv = minv = maxv = bp[0];
+        for (i = 4; i < 64; i += 4) {
+            muv += bp[i];
+            if (bp[i] < minv)
+                minv = bp[i];
+            else if (bp[i] > maxv)
+                maxv = bp[i];
+        }
+
+        mu[ch] = (muv + 8) >> 4;
+        min[ch] = minv;
+        max[ch] = maxv;
+    }
+
+    // determine covariance matrix
+    for (i = 0; i < 6; i++)
+        cov[i] = 0;
+
+    for (i = 0; i < 16; i++) {
+        int r = block[i * 4 + 0] - mu[0];
+        int g = block[i * 4 + 1] - mu[1];
+        int b = block[i * 4 + 2] - mu[2];
+
+        cov[0] += r * r;
+        cov[1] += r * g;
+        cov[2] += r * b;
+        cov[3] += g * g;
+        cov[4] += g * b;
+        cov[5] += b * b;
+    }
+
+    // convert covariance matrix to float, find principal axis via power iter
+    for (i = 0; i < 6; i++)
+        covf[i] = static_cast<float>(cov[i]) / 255.0f;
+
+    vfr = (float)(max[0] - min[0]);
+    vfg = (float)(max[1] - min[1]);
+    vfb = (float)(max[2] - min[2]);
+
+    for (iter = 0; iter < nIterPower; iter++) {
+        float r = vfr * covf[0] + vfg * covf[1] + vfb * covf[2];
+        float g = vfr * covf[1] + vfg * covf[3] + vfb * covf[4];
+        float b = vfr * covf[2] + vfg * covf[4] + vfb * covf[5];
+
+        vfr = r;
+        vfg = g;
+        vfb = b;
+    }
+
+    magn = STBD_FABS(vfr);
+    if (STBD_FABS(vfg) > magn)
+        magn = STBD_FABS(vfg);
+    if (STBD_FABS(vfb) > magn)
+        magn = STBD_FABS(vfb);
+
+    if (magn < 4.0f) { // too small, default to luminance
+        v_r = 299;     // JPEG YCbCr luma coefs, scaled by 1000.
+        v_g = 587;
+        v_b = 114;
+    } else {
+        magn = 512.0 / magn;
+        v_r = (int)(vfr * magn);
+        v_g = (int)(vfg * magn);
+        v_b = (int)(vfb * magn);
+    }
+
+    minp = maxp = block;
+    mind = maxd = block[0] * v_r + block[1] * v_g + block[2] * v_b;
+    // Pick colors at extreme points
+    for (i = 1; i < 16; i++) {
+        int dot = block[i * 4 + 0] * v_r + block[i * 4 + 1] * v_g + block[i * 4 + 2] * v_b;
+
+        if (dot < mind) {
+            mind = dot;
+            minp = block + i * 4;
+        }
+
+        if (dot > maxd) {
+            maxd = dot;
+            maxp = block + i * 4;
+        }
+    }
+
+    *pmax16 = stb__As16Bit(maxp[0], maxp[1], maxp[2]);
+    *pmin16 = stb__As16Bit(minp[0], minp[1], minp[2]);
+    stb__ReorderColors(pmax16, pmin16);
+}
+
+static void stb__OptimizeColorsAlphaBlock(unsigned char* block, unsigned short* pmax16,
+                                          unsigned short* pmin16) {
+    int mind, maxd;
+    unsigned char *minp, *maxp;
+    double magn;
+    int v_r, v_g, v_b;
+    static const int nIterPower = 4;
+    float covf[6], vfr, vfg, vfb;
+
+    // determine color distribution
+    int cov[6];
+    int mu[3], min[3], max[3];
+    int ch, i, iter;
+
+    for (ch = 0; ch < 3; ch++) {
+        const unsigned char* bp = ((const unsigned char*)block) + ch;
+        int muv = 0, minv = 256, maxv = -1;
+        int num = 0;
+
+        for (i = 0; i < 64; i += 4) {
+            if (bp[3 - ch] == 0) {
+                continue;
+            }
+
+            muv += bp[i];
+            if (bp[i] < minv)
+                minv = bp[i];
+            else if (bp[i] > maxv)
+                maxv = bp[i];
+
+            num++;
+        }
+
+        mu[ch] = num > 0 ? (muv + 8) / num : 0;
+        min[ch] = minv;
+        max[ch] = maxv;
+    }
+
+    // determine covariance matrix
+    for (i = 0; i < 6; i++)
+        cov[i] = 0;
+
+    for (i = 0; i < 16; i++) {
+        if (block[i * 4 + 3] == 0) {
+            continue;
+        }
+
+        int r = block[i * 4 + 0] - mu[0];
+        int g = block[i * 4 + 1] - mu[1];
+        int b = block[i * 4 + 2] - mu[2];
+
+        cov[0] += r * r;
+        cov[1] += r * g;
+        cov[2] += r * b;
+        cov[3] += g * g;
+        cov[4] += g * b;
+        cov[5] += b * b;
+    }
+
+    // convert covariance matrix to float, find principal axis via power iter
+    for (i = 0; i < 6; i++)
+        covf[i] = static_cast<float>(cov[i]) / 255.0f;
+
+    vfr = (float)(max[0] - min[0]);
+    vfg = (float)(max[1] - min[1]);
+    vfb = (float)(max[2] - min[2]);
+
+    for (iter = 0; iter < nIterPower; iter++) {
+        float r = vfr * covf[0] + vfg * covf[1] + vfb * covf[2];
+        float g = vfr * covf[1] + vfg * covf[3] + vfb * covf[4];
+        float b = vfr * covf[2] + vfg * covf[4] + vfb * covf[5];
+
+        vfr = r;
+        vfg = g;
+        vfb = b;
+    }
+
+    magn = STBD_FABS(vfr);
+    if (STBD_FABS(vfg) > magn)
+        magn = STBD_FABS(vfg);
+    if (STBD_FABS(vfb) > magn)
+        magn = STBD_FABS(vfb);
+
+    if (magn < 4.0f) { // too small, default to luminance
+        v_r = 299;     // JPEG YCbCr luma coefs, scaled by 1000.
+        v_g = 587;
+        v_b = 114;
+    } else {
+        magn = 512.0 / magn;
+        v_r = (int)(vfr * magn);
+        v_g = (int)(vfg * magn);
+        v_b = (int)(vfb * magn);
+    }
+
+    minp = maxp = NULL;
+    mind = 0x7fffffff;
+    maxd = -0x80000000;
+
+    // Pick colors at extreme points
+    for (i = 0; i < 16; i++) {
+        if (block[i * 4 + 3] == 0) {
+            continue;
+        }
+
+        int dot = block[i * 4 + 0] * v_r + block[i * 4 + 1] * v_g + block[i * 4 + 2] * v_b;
+
+        if (dot < mind) {
+            mind = dot;
+            minp = block + i * 4;
+        }
+
+        if (dot > maxd) {
+            maxd = dot;
+            maxp = block + i * 4;
+        }
+    }
+
+    if (!maxp) {
+        // all alpha, no color
+        *pmin16 = 0xffff;
+        *pmax16 = 0;
+    } else {
+        // endpoint colors found
+        *pmax16 = stb__As16Bit(maxp[0], maxp[1], maxp[2]);
+        *pmin16 = stb__As16Bit(minp[0], minp[1], minp[2]);
+
+        if (*pmax16 == *pmin16) {
+            // modify the endpoints to indicate presence of an alpha block
+            if (*pmax16 > 0) {
+                (*pmax16)--;
+            } else {
+                (*pmin16)++;
+            }
+        }
+
+        stb__ReorderColors(pmax16, pmin16);
+    }
+}
+
+static const float stb__midpoints5[32] = {
+    0.015686f, 0.047059f, 0.078431f, 0.111765f, 0.145098f, 0.176471f, 0.207843f, 0.241176f,
+    0.274510f, 0.305882f, 0.337255f, 0.370588f, 0.403922f, 0.435294f, 0.466667f, 0.5f,
+    0.533333f, 0.564706f, 0.596078f, 0.629412f, 0.662745f, 0.694118f, 0.725490f, 0.758824f,
+    0.792157f, 0.823529f, 0.854902f, 0.888235f, 0.921569f, 0.952941f, 0.984314f, 1.0f};
+
+static const float stb__midpoints6[64] = {
+    0.007843f, 0.023529f, 0.039216f, 0.054902f, 0.070588f, 0.086275f, 0.101961f, 0.117647f,
+    0.133333f, 0.149020f, 0.164706f, 0.180392f, 0.196078f, 0.211765f, 0.227451f, 0.245098f,
+    0.262745f, 0.278431f, 0.294118f, 0.309804f, 0.325490f, 0.341176f, 0.356863f, 0.372549f,
+    0.388235f, 0.403922f, 0.419608f, 0.435294f, 0.450980f, 0.466667f, 0.482353f, 0.500000f,
+    0.517647f, 0.533333f, 0.549020f, 0.564706f, 0.580392f, 0.596078f, 0.611765f, 0.627451f,
+    0.643137f, 0.658824f, 0.674510f, 0.690196f, 0.705882f, 0.721569f, 0.737255f, 0.754902f,
+    0.772549f, 0.788235f, 0.803922f, 0.819608f, 0.835294f, 0.850980f, 0.866667f, 0.882353f,
+    0.898039f, 0.913725f, 0.929412f, 0.945098f, 0.960784f, 0.976471f, 0.992157f, 1.0f};
+
+static unsigned short stb__Quantize5(float x) {
+    unsigned short q;
+    x = x < 0 ? 0 : x > 1 ? 1 : x; // saturate
+    q = (unsigned short)(x * 31);
+    q += (x > stb__midpoints5[q]);
+    return q;
+}
+
+static unsigned short stb__Quantize6(float x) {
+    unsigned short q;
+    x = x < 0 ? 0 : x > 1 ? 1 : x; // saturate
+    q = (unsigned short)(x * 63);
+    q += (x > stb__midpoints6[q]);
+    return q;
+}
+
+// The refinement function. (Clever code, part 2)
+// Tries to optimize colors to suit block contents better.
+// (By solving a least squares system via normal equations+Cramer's rule)
+static int stb__RefineBlock(unsigned char* block, unsigned short* pmax16, unsigned short* pmin16,
+                            unsigned int mask) {
+    static const int w1Tab[4] = {3, 0, 2, 1};
+    static const int prods[4] = {0x090000, 0x000900, 0x040102, 0x010402};
+    // ^some magic to save a lot of multiplies in the accumulating loop...
+    // (precomputed products of weights for least squares system, accumulated
+    // inside one 32-bit register)
+
+    float f;
+    unsigned short oldMin, oldMax, min16, max16;
+    int i, akku = 0, xx, xy, yy;
+    int At1_r, At1_g, At1_b;
+    int At2_r, At2_g, At2_b;
+    unsigned int cm = mask;
+
+    oldMin = *pmin16;
+    oldMax = *pmax16;
+
+    if ((mask ^ (mask << 2)) < 4) // all pixels have the same index?
+    {
+        // yes, linear system would be singular; solve using optimal
+        // single-color match on average color
+        int r = 8, g = 8, b = 8;
+        for (i = 0; i < 16; ++i) {
+            r += block[i * 4 + 0];
+            g += block[i * 4 + 1];
+            b += block[i * 4 + 2];
+        }
+
+        r >>= 4;
+        g >>= 4;
+        b >>= 4;
+
+        max16 = static_cast<unsigned short>((stb__OMatch5[r][0] << 11) | (stb__OMatch6[g][0] << 5) |
+                                            stb__OMatch5[b][0]);
+        min16 = static_cast<unsigned short>((stb__OMatch5[r][1] << 11) | (stb__OMatch6[g][1] << 5) |
+                                            stb__OMatch5[b][1]);
+    } else {
+        At1_r = At1_g = At1_b = 0;
+        At2_r = At2_g = At2_b = 0;
+        for (i = 0; i < 16; ++i, cm >>= 2) {
+            int step = cm & 3;
+            int w1 = w1Tab[step];
+            int r = block[i * 4 + 0];
+            int g = block[i * 4 + 1];
+            int b = block[i * 4 + 2];
+
+            akku += prods[step];
+            At1_r += w1 * r;
+            At1_g += w1 * g;
+            At1_b += w1 * b;
+            At2_r += r;
+            At2_g += g;
+            At2_b += b;
+        }
+
+        At2_r = 3 * At2_r - At1_r;
+        At2_g = 3 * At2_g - At1_g;
+        At2_b = 3 * At2_b - At1_b;
+
+        // extract solutions and decide solvability
+        xx = akku >> 16;
+        yy = (akku >> 8) & 0xff;
+        xy = (akku >> 0) & 0xff;
+
+        f = 3.0f / 255.0f / static_cast<float>(xx * yy - xy * xy);
+
+        max16 = static_cast<unsigned short>(
+            stb__Quantize5(static_cast<float>(At1_r * yy - At2_r * xy) * f) << 11);
+        max16 |= static_cast<unsigned short>(
+            stb__Quantize6(static_cast<float>(At1_g * yy - At2_g * xy) * f) << 5);
+        max16 |= static_cast<unsigned short>(
+            stb__Quantize5(static_cast<float>(At1_b * yy - At2_b * xy) * f) << 0);
+
+        min16 = static_cast<unsigned short>(
+            stb__Quantize5(static_cast<float>(At2_r * xx - At1_r * xy) * f) << 11);
+        min16 |= static_cast<unsigned short>(
+            stb__Quantize6(static_cast<float>(At2_g * xx - At1_g * xy) * f) << 5);
+        min16 |= static_cast<unsigned short>(
+            stb__Quantize5(static_cast<float>(At2_b * xx - At1_b * xy) * f) << 0);
+    }
+
+    *pmin16 = min16;
+    *pmax16 = max16;
+    stb__ReorderColors(pmax16, pmin16);
+
+    return oldMin != min16 || oldMax != max16;
+}
+
+// Color block compression
+static void stb__CompressColorBlock(unsigned char* dest, unsigned char* block, int alpha,
+                                    int mode) {
+    unsigned int mask;
+    int i;
+    int refinecount;
+    unsigned short max16, min16;
+    unsigned char color[4 * 4];
+
+    refinecount = (mode & STB_DXT_HIGHQUAL) ? 2 : 1;
+
+    // check if block is constant
+    for (i = 1; i < 16; i++)
+        if (((unsigned int*)block)[i] != ((unsigned int*)block)[0])
+            break;
+
+    if (i == 16 && block[3] == 0 && alpha) { // constant alpha
+        mask = 0xffffffff;
+        max16 = 0;
+        min16 = 0xffff;
+    } else if (i == 16) { // constant color
+        int r = block[0], g = block[1], b = block[2];
+        mask = 0xaaaaaaaa;
+        max16 = static_cast<unsigned short>((stb__OMatch5[r][0] << 11) | (stb__OMatch6[g][0] << 5) |
+                                            stb__OMatch5[b][0]);
+        min16 = static_cast<unsigned short>((stb__OMatch5[r][1] << 11) | (stb__OMatch6[g][1] << 5) |
+                                            stb__OMatch5[b][1]);
+    } else if (alpha) {
+        stb__OptimizeColorsAlphaBlock(block, &max16, &min16);
+        stb__Eval3Colors(color, max16, min16);
+        mask = stb__MatchColorsAlphaBlock(block, color);
+    } else {
+        // first step: PCA+map along principal axis
+        stb__OptimizeColorsBlock(block, &max16, &min16);
+        if (max16 != min16) {
+            stb__Eval4Colors(color, max16, min16);
+            mask = stb__MatchColorsBlock(block, color);
+        } else
+            mask = 0;
+
+        // third step: refine (multiple times if requested)
+        for (i = 0; i < refinecount; i++) {
+            unsigned int lastmask = mask;
+
+            if (stb__RefineBlock(block, &max16, &min16, mask)) {
+                if (max16 != min16) {
+                    stb__Eval4Colors(color, max16, min16);
+                    mask = stb__MatchColorsBlock(block, color);
+                } else {
+                    mask = 0;
+                    break;
+                }
+            }
+
+            if (mask == lastmask)
+                break;
+        }
+    }
+
+    // write the color block
+    if (!alpha)
+        stb__FinalizeColors(&max16, &min16, &mask);
+
+    dest[0] = (unsigned char)(max16);
+    dest[1] = (unsigned char)(max16 >> 8);
+    dest[2] = (unsigned char)(min16);
+    dest[3] = (unsigned char)(min16 >> 8);
+    dest[4] = (unsigned char)(mask);
+    dest[5] = (unsigned char)(mask >> 8);
+    dest[6] = (unsigned char)(mask >> 16);
+    dest[7] = (unsigned char)(mask >> 24);
+}
+
+// Alpha block compression (this is easy for a change)
+static void stb__CompressAlphaBlock(unsigned char* dest, unsigned char* src, int stride) {
+    int i, dist, bias, dist4, dist2, bits, mask;
+
+    // find min/max color
+    int mn, mx;
+    mn = mx = src[0];
+
+    for (i = 1; i < 16; i++) {
+        if (src[i * stride] < mn)
+            mn = src[i * stride];
+        else if (src[i * stride] > mx)
+            mx = src[i * stride];
+    }
+
+    // encode them
+    dest[0] = (unsigned char)mx;
+    dest[1] = (unsigned char)mn;
+    dest += 2;
+
+    // determine bias and emit color indices
+    // given the choice of mx/mn, these indices are optimal:
+    // http://fgiesen.wordpress.com/2009/12/15/dxt5-alpha-block-index-determination/
+    dist = mx - mn;
+    dist4 = dist * 4;
+    dist2 = dist * 2;
+    bias = (dist < 8) ? (dist - 1) : (dist / 2 + 2);
+    bias -= mn * 7;
+    bits = 0, mask = 0;
+
+    for (i = 0; i < 16; i++) {
+        int a = src[i * stride] * 7 + bias;
+        int ind, t;
+
+        // select index. this is a "linear scale" lerp factor between 0 (val=min)
+        // and 7 (val=max).
+        t = (a >= dist4) ? -1 : 0;
+        ind = t & 4;
+        a -= dist4 & t;
+        t = (a >= dist2) ? -1 : 0;
+        ind += t & 2;
+        a -= dist2 & t;
+        ind += (a >= dist);
+
+        // turn linear scale into DXT index (0/1 are extremal pts)
+        ind = -ind & 7;
+        ind ^= (2 > ind);
+
+        // write index
+        mask |= ind << bits;
+        if ((bits += 3) >= 8) {
+            *dest++ = (unsigned char)mask;
+            mask >>= 8;
+            bits -= 8;
+        }
+    }
+}
+
+void stb_compress_bc1_block(unsigned char* dest, const unsigned char* src, int alpha, int mode) {
+    stb__CompressColorBlock(dest, (unsigned char*)src, alpha, mode);
+}
+
+void stb_compress_bc3_block(unsigned char* dest, const unsigned char* src, int mode) {
+    unsigned char data[16][4];
+    int i;
+
+    stb__CompressAlphaBlock(dest, (unsigned char*)src + 3, 4);
+    dest += 8;
+    // make a new copy of the data in which alpha is opaque,
+    // because code uses a fast test for color constancy
+    memcpy(data, src, 4 * 16);
+    for (i = 0; i < 16; ++i)
+        data[i][3] = 255;
+    src = &data[0][0];
+
+    stb__CompressColorBlock(dest, (unsigned char*)src, 0, mode);
+}

externals/stb/stb_dxt.h

@@ -0,0 +1,36 @@
+// SPDX-FileCopyrightText: fabian "ryg" giesen
+// SPDX-License-Identifier: MIT
+
+// stb_dxt.h - v1.12 - DXT1/DXT5 compressor
+
+#ifndef STB_INCLUDE_STB_DXT_H
+#define STB_INCLUDE_STB_DXT_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef STB_DXT_STATIC
+#define STBDDEF static
+#else
+#define STBDDEF extern
+#endif
+
+// compression mode (bitflags)
+#define STB_DXT_NORMAL 0
+#define STB_DXT_DITHER 1 // use dithering. was always dubious, now deprecated. does nothing!
+#define STB_DXT_HIGHQUAL                                                                           \
+    2 // high quality mode, does two refinement steps instead of 1. ~30-40% slower.
+
+STBDDEF void stb_compress_bc1_block(unsigned char* dest,
+                                    const unsigned char* src_rgba_four_bytes_per_pixel, int alpha,
+                                    int mode);
+
+STBDDEF void stb_compress_bc3_block(unsigned char* dest, const unsigned char* src, int mode);
+
+#define STB_COMPRESS_DXT_BLOCK
+
+#ifdef __cplusplus
+}
+#endif
+#endif // STB_INCLUDE_STB_DXT_H

src/audio_core/renderer/adsp/audio_renderer.cpp

@@ -154,6 +154,11 @@ void AudioRenderer::ThreadFunc() {
             return;
 
         case RenderMessage::AudioRenderer_Render: {
+            if (system.IsShuttingDown()) [[unlikely]] {
+                std::this_thread::sleep_for(std::chrono::milliseconds(5));
+                mailbox->ADSPSendMessage(RenderMessage::AudioRenderer_RenderResponse);
+                continue;
+            }
             std::array<bool, MaxRendererSessions> buffers_reset{};
             std::array<u64, MaxRendererSessions> render_times_taken{};
             const auto start_time{system.CoreTiming().GetClockTicks()};

src/audio_core/renderer/system_manager.cpp

@@ -27,7 +27,7 @@ bool SystemManager::InitializeUnsafe() {
     if (!active) {
         if (adsp.Start()) {
             active = true;
-            thread = std::jthread([this](std::stop_token stop_token) { ThreadFunc(); });
+            thread = std::jthread([this](std::stop_token stop_token) { ThreadFunc(stop_token); });
         }
     }
 
@@ -39,8 +39,7 @@ void SystemManager::Stop() {
         return;
     }
     active = false;
-    update.store(true);
-    update.notify_all();
+    thread.request_stop();
     thread.join();
     adsp.Stop();
 }
@@ -85,12 +84,12 @@ bool SystemManager::Remove(System& system_) {
     return true;
 }
 
-void SystemManager::ThreadFunc() {
+void SystemManager::ThreadFunc(std::stop_token stop_token) {
     static constexpr char name[]{"AudioRenderSystemManager"};
     MicroProfileOnThreadCreate(name);
     Common::SetCurrentThreadName(name);
     Common::SetCurrentThreadPriority(Common::ThreadPriority::High);
-    while (active) {
+    while (active && !stop_token.stop_requested()) {
         {
             std::scoped_lock l{mutex1};
 

src/audio_core/renderer/system_manager.h

@@ -66,13 +66,7 @@ private:
     /**
      * Main thread responsible for command generation.
      */
-    void ThreadFunc();
-
-    enum class StreamState {
-        Filling,
-        Steady,
-        Draining,
-    };
+    void ThreadFunc(std::stop_token stop_token);
 
     /// Core system
     Core::System& core;
@@ -90,8 +84,6 @@ private:
     ADSP::ADSP& adsp;
     /// AudioRenderer mailbox for communication
     ADSP::AudioRenderer_Mailbox* mailbox{};
-    /// Atomic for main thread to wait on
-    std::atomic<bool> update{};
 };
 
 } // namespace AudioCore::AudioRenderer

src/audio_core/sink/sink_stream.cpp

@@ -271,8 +271,11 @@ u64 SinkStream::GetExpectedPlayedSampleCount() {
 
 void SinkStream::WaitFreeSpace() {
     std::unique_lock lk{release_mutex};
-    release_cv.wait(
-        lk, [this]() { return queued_buffers < max_queue_size || system.IsShuttingDown(); });
+    release_cv.wait_for(lk, std::chrono::milliseconds(5),
+                        [this]() { return queued_buffers < max_queue_size; });
+    if (queued_buffers > max_queue_size + 3) {
+        release_cv.wait(lk, [this]() { return queued_buffers < max_queue_size; });
+    }
 }
 
 } // namespace AudioCore::Sink

src/common/bit_field.h

@@ -188,3 +188,8 @@ private:
 
 template <std::size_t Position, std::size_t Bits, typename T>
 using BitFieldBE = BitField<Position, Bits, T, BETag>;
+
+template <std::size_t Position, std::size_t Bits, typename T, typename EndianTag = LETag>
+inline auto format_as(BitField<Position, Bits, T, EndianTag> bitfield) {
+    return bitfield.Value();
+}

src/common/settings.cpp

@@ -61,6 +61,7 @@ void LogSettings() {
     log_setting("Renderer_NvdecEmulation", values.nvdec_emulation.GetValue());
     log_setting("Renderer_AccelerateASTC", values.accelerate_astc.GetValue());
     log_setting("Renderer_AsyncASTC", values.async_astc.GetValue());
+    log_setting("Renderer_AstcRecompression", values.astc_recompression.GetValue());
     log_setting("Renderer_UseVsync", values.vsync_mode.GetValue());
     log_setting("Renderer_UseReactiveFlushing", values.use_reactive_flushing.GetValue());
     log_setting("Renderer_ShaderBackend", values.shader_backend.GetValue());
@@ -224,6 +225,7 @@ void RestoreGlobalState(bool is_powered_on) {
     values.nvdec_emulation.SetGlobal(true);
     values.accelerate_astc.SetGlobal(true);
     values.async_astc.SetGlobal(true);
+    values.astc_recompression.SetGlobal(true);
     values.use_reactive_flushing.SetGlobal(true);
     values.shader_backend.SetGlobal(true);
     values.use_asynchronous_shaders.SetGlobal(true);

src/common/settings.h

@@ -90,6 +90,12 @@ enum class AntiAliasing : u32 {
     LastAA = Smaa,
 };
 
+enum class AstcRecompression : u32 {
+    Uncompressed = 0,
+    Bc1 = 1,
+    Bc3 = 2,
+};
+
 struct ResolutionScalingInfo {
     u32 up_scale{1};
     u32 down_shift{0};
@@ -473,6 +479,9 @@ struct Values {
     SwitchableSetting<bool> use_vulkan_driver_pipeline_cache{true,
                                                              "use_vulkan_driver_pipeline_cache"};
     SwitchableSetting<bool> enable_compute_pipelines{false, "enable_compute_pipelines"};
+    SwitchableSetting<AstcRecompression, true> astc_recompression{
+        AstcRecompression::Uncompressed, AstcRecompression::Uncompressed, AstcRecompression::Bc3,
+        "astc_recompression"};
 
     SwitchableSetting<u8> bg_red{0, "bg_red"};
     SwitchableSetting<u8> bg_green{0, "bg_green"};

src/core/core.cpp

@@ -117,8 +117,7 @@ FileSys::VirtualFile GetGameFileFromPath(const FileSys::VirtualFilesystem& vfs,
             return nullptr;
         }
 
-        return FileSys::ConcatenatedVfsFile::MakeConcatenatedFile(std::move(concat),
-                                                                  dir->GetName());
+        return FileSys::ConcatenatedVfsFile::MakeConcatenatedFile(concat, dir->GetName());
     }
 
     if (Common::FS::IsDir(path)) {

src/core/file_sys/romfs.cpp

@@ -140,7 +140,8 @@ VirtualFile CreateRomFS(VirtualDir dir, VirtualDir ext) {
         return nullptr;
 
     RomFSBuildContext ctx{dir, ext};
-    return ConcatenatedVfsFile::MakeConcatenatedFile(0, ctx.Build(), dir->GetName());
+    auto file_map = ctx.Build();
+    return ConcatenatedVfsFile::MakeConcatenatedFile(0, file_map, dir->GetName());
 }
 
 } // namespace FileSys

src/core/file_sys/vfs_concat.cpp

@@ -10,84 +10,105 @@
 
 namespace FileSys {
 
-static bool VerifyConcatenationMapContinuity(const std::multimap<u64, VirtualFile>& map) {
-    const auto last_valid = --map.end();
-    for (auto iter = map.begin(); iter != last_valid;) {
-        const auto old = iter++;
-        if (old->first + old->second->GetSize() != iter->first) {
+ConcatenatedVfsFile::ConcatenatedVfsFile(ConcatenationMap&& concatenation_map_, std::string&& name_)
+    : concatenation_map(std::move(concatenation_map_)), name(std::move(name_)) {
+    DEBUG_ASSERT(this->VerifyContinuity());
+}
+
+bool ConcatenatedVfsFile::VerifyContinuity() const {
+    u64 last_offset = 0;
+    for (auto& entry : concatenation_map) {
+        if (entry.offset != last_offset) {
             return false;
         }
+
+        last_offset = entry.offset + entry.file->GetSize();
     }
 
-    return map.begin()->first == 0;
-}
-
-ConcatenatedVfsFile::ConcatenatedVfsFile(std::vector<VirtualFile> files_, std::string name_)
-    : name(std::move(name_)) {
-    std::size_t next_offset = 0;
-    for (const auto& file : files_) {
-        files.emplace(next_offset, file);
-        next_offset += file->GetSize();
-    }
-}
-
-ConcatenatedVfsFile::ConcatenatedVfsFile(std::multimap<u64, VirtualFile> files_, std::string name_)
-    : files(std::move(files_)), name(std::move(name_)) {
-    ASSERT(VerifyConcatenationMapContinuity(files));
+    return true;
 }
 
 ConcatenatedVfsFile::~ConcatenatedVfsFile() = default;
 
-VirtualFile ConcatenatedVfsFile::MakeConcatenatedFile(std::vector<VirtualFile> files,
-                                                      std::string name) {
-    if (files.empty())
+VirtualFile ConcatenatedVfsFile::MakeConcatenatedFile(const std::vector<VirtualFile>& files,
+                                                      std::string&& name) {
+    // Fold trivial cases.
+    if (files.empty()) {
         return nullptr;
-    if (files.size() == 1)
-        return files[0];
+    }
+    if (files.size() == 1) {
+        return files.front();
+    }
 
-    return VirtualFile(new ConcatenatedVfsFile(std::move(files), std::move(name)));
+    // Make the concatenation map from the input.
+    std::vector<ConcatenationEntry> concatenation_map;
+    concatenation_map.reserve(files.size());
+    u64 last_offset = 0;
+
+    for (auto& file : files) {
+        concatenation_map.emplace_back(ConcatenationEntry{
+            .offset = last_offset,
+            .file = file,
+        });
+
+        last_offset += file->GetSize();
+    }
+
+    return VirtualFile(new ConcatenatedVfsFile(std::move(concatenation_map), std::move(name)));
 }
 
 VirtualFile ConcatenatedVfsFile::MakeConcatenatedFile(u8 filler_byte,
-                                                      std::multimap<u64, VirtualFile> files,
-                                                      std::string name) {
-    if (files.empty())
+                                                      const std::multimap<u64, VirtualFile>& files,
+                                                      std::string&& name) {
+    // Fold trivial cases.
+    if (files.empty()) {
         return nullptr;
-    if (files.size() == 1)
+    }
+    if (files.size() == 1) {
         return files.begin()->second;
-
-    const auto last_valid = --files.end();
-    for (auto iter = files.begin(); iter != last_valid;) {
-        const auto old = iter++;
-        if (old->first + old->second->GetSize() != iter->first) {
-            files.emplace(old->first + old->second->GetSize(),
-                          std::make_shared<StaticVfsFile>(filler_byte, iter->first - old->first -
-                                                                           old->second->GetSize()));
-        }
     }
 
-    // Ensure the map starts at offset 0 (start of file), otherwise pad to fill.
-    if (files.begin()->first != 0)
-        files.emplace(0, std::make_shared<StaticVfsFile>(filler_byte, files.begin()->first));
+    // Make the concatenation map from the input.
+    std::vector<ConcatenationEntry> concatenation_map;
 
-    return VirtualFile(new ConcatenatedVfsFile(std::move(files), std::move(name)));
+    concatenation_map.reserve(files.size());
+    u64 last_offset = 0;
+
+    // Iteration of a multimap is ordered, so offset will be strictly non-decreasing.
+    for (auto& [offset, file] : files) {
+        if (offset > last_offset) {
+            concatenation_map.emplace_back(ConcatenationEntry{
+                .offset = last_offset,
+                .file = std::make_shared<StaticVfsFile>(filler_byte, offset - last_offset),
+            });
+        }
+
+        concatenation_map.emplace_back(ConcatenationEntry{
+            .offset = offset,
+            .file = file,
+        });
+
+        last_offset = offset + file->GetSize();
+    }
+
+    return VirtualFile(new ConcatenatedVfsFile(std::move(concatenation_map), std::move(name)));
 }
 
 std::string ConcatenatedVfsFile::GetName() const {
-    if (files.empty()) {
+    if (concatenation_map.empty()) {
         return "";
     }
     if (!name.empty()) {
         return name;
     }
-    return files.begin()->second->GetName();
+    return concatenation_map.front().file->GetName();
 }
 
 std::size_t ConcatenatedVfsFile::GetSize() const {
-    if (files.empty()) {
+    if (concatenation_map.empty()) {
         return 0;
     }
-    return files.rbegin()->first + files.rbegin()->second->GetSize();
+    return concatenation_map.back().offset + concatenation_map.back().file->GetSize();
 }
 
 bool ConcatenatedVfsFile::Resize(std::size_t new_size) {
@@ -95,10 +116,10 @@ bool ConcatenatedVfsFile::Resize(std::size_t new_size) {
 }
 
 VirtualDir ConcatenatedVfsFile::GetContainingDirectory() const {
-    if (files.empty()) {
+    if (concatenation_map.empty()) {
         return nullptr;
     }
-    return files.begin()->second->GetContainingDirectory();
+    return concatenation_map.front().file->GetContainingDirectory();
 }
 
 bool ConcatenatedVfsFile::IsWritable() const {
@@ -110,25 +131,45 @@ bool ConcatenatedVfsFile::IsReadable() const {
 }
 
 std::size_t ConcatenatedVfsFile::Read(u8* data, std::size_t length, std::size_t offset) const {
-    auto entry = --files.end();
-    for (auto iter = files.begin(); iter != files.end(); ++iter) {
-        if (iter->first > offset) {
-            entry = --iter;
+    const ConcatenationEntry key{
+        .offset = offset,
+        .file = nullptr,
+    };
+
+    // Read nothing if the map is empty.
+    if (concatenation_map.empty()) {
+        return 0;
+    }
+
+    // Binary search to find the iterator to the first position we can check.
+    // It must exist, since we are not empty and are comparing unsigned integers.
+    auto it = std::prev(std::upper_bound(concatenation_map.begin(), concatenation_map.end(), key));
+    u64 cur_length = length;
+    u64 cur_offset = offset;
+
+    while (cur_length > 0 && it != concatenation_map.end()) {
+        // Check if we can read the file at this position.
+        const auto& file = it->file;
+        const u64 file_offset = it->offset;
+        const u64 file_size = file->GetSize();
+
+        if (cur_offset >= file_offset + file_size) {
+            // Entirely out of bounds read.
             break;
         }
+
+        // Read the file at this position.
+        const u64 intended_read_size = std::min<u64>(cur_length, file_size);
+        const u64 actual_read_size =
+            file->Read(data + (cur_offset - offset), intended_read_size, cur_offset - file_offset);
+
+        // Update tracking.
+        cur_offset += actual_read_size;
+        cur_length -= actual_read_size;
+        it++;
     }
 
-    if (entry->first + entry->second->GetSize() <= offset)
-        return 0;
-
-    const auto read_in =
-        std::min<u64>(entry->first + entry->second->GetSize() - offset, entry->second->GetSize());
-    if (length > read_in) {
-        return entry->second->Read(data, read_in, offset - entry->first) +
-               Read(data + read_in, length - read_in, offset + read_in);
-    }
-
-    return entry->second->Read(data, std::min<u64>(read_in, length), offset - entry->first);
+    return cur_offset - offset;
 }
 
 std::size_t ConcatenatedVfsFile::Write(const u8* data, std::size_t length, std::size_t offset) {

src/core/file_sys/vfs_concat.h

@@ -3,6 +3,7 @@
 
 #pragma once
 
+#include <compare>
 #include <map>
 #include <memory>
 #include "core/file_sys/vfs.h"
@@ -12,19 +13,33 @@ namespace FileSys {
 // Class that wraps multiple vfs files and concatenates them, making reads seamless. Currently
 // read-only.
 class ConcatenatedVfsFile : public VfsFile {
-    explicit ConcatenatedVfsFile(std::vector<VirtualFile> files, std::string name_);
-    explicit ConcatenatedVfsFile(std::multimap<u64, VirtualFile> files, std::string name_);
+private:
+    struct ConcatenationEntry {
+        u64 offset;
+        VirtualFile file;
+
+        auto operator<=>(const ConcatenationEntry& other) const {
+            return this->offset <=> other.offset;
+        }
+    };
+    using ConcatenationMap = std::vector<ConcatenationEntry>;
+
+    explicit ConcatenatedVfsFile(std::vector<ConcatenationEntry>&& concatenation_map,
+                                 std::string&& name);
+    bool VerifyContinuity() const;
 
 public:
     ~ConcatenatedVfsFile() override;
 
     /// Wrapper function to allow for more efficient handling of files.size() == 0, 1 cases.
-    static VirtualFile MakeConcatenatedFile(std::vector<VirtualFile> files, std::string name);
+    static VirtualFile MakeConcatenatedFile(const std::vector<VirtualFile>& files,
+                                            std::string&& name);
 
     /// Convenience function that turns a map of offsets to files into a concatenated file, filling
     /// gaps with a given filler byte.
-    static VirtualFile MakeConcatenatedFile(u8 filler_byte, std::multimap<u64, VirtualFile> files,
-                                            std::string name);
+    static VirtualFile MakeConcatenatedFile(u8 filler_byte,
+                                            const std::multimap<u64, VirtualFile>& files,
+                                            std::string&& name);
 
     std::string GetName() const override;
     std::size_t GetSize() const override;
@@ -37,8 +52,7 @@ public:
     bool Rename(std::string_view new_name) override;
 
 private:
-    // Maps starting offset to file -- more efficient.
-    std::multimap<u64, VirtualFile> files;
+    ConcatenationMap concatenation_map;
     std::string name;
 };
 

src/core/hid/emulated_controller.cpp

@@ -1283,9 +1283,14 @@ bool EmulatedController::HasNfc() const {
 }
 
 bool EmulatedController::WriteNfc(const std::vector<u8>& data) {
-    auto& nfc_output_device = output_devices[3];
+    auto& nfc_output_device = output_devices[static_cast<std::size_t>(DeviceIndex::Right)];
+    auto& nfc_virtual_output_device = output_devices[3];
 
-    return nfc_output_device->WriteNfcData(data) == Common::Input::NfcState::Success;
+    if (nfc_output_device->SupportsNfc() != Common::Input::NfcState::NotSupported) {
+        return nfc_output_device->WriteNfcData(data) == Common::Input::NfcState::Success;
+    }
+
+    return nfc_virtual_output_device->WriteNfcData(data) == Common::Input::NfcState::Success;
 }
 
 void EmulatedController::SetLedPattern() {

src/core/hle/kernel/k_memory_block_manager.h

@@ -144,14 +144,10 @@ private:
 
 class KScopedMemoryBlockManagerAuditor {
 public:
-    explicit KScopedMemoryBlockManagerAuditor(KMemoryBlockManager* m) : m_manager(m) {
-        ASSERT(m_manager->CheckState());
-    }
+    explicit KScopedMemoryBlockManagerAuditor(KMemoryBlockManager* m) : m_manager(m) {}
     explicit KScopedMemoryBlockManagerAuditor(KMemoryBlockManager& m)
         : KScopedMemoryBlockManagerAuditor(std::addressof(m)) {}
-    ~KScopedMemoryBlockManagerAuditor() {
-        ASSERT(m_manager->CheckState());
-    }
+    ~KScopedMemoryBlockManagerAuditor() = default;
 
 private:
     KMemoryBlockManager* m_manager;

src/core/hle/service/nfc/common/amiibo_crypto.cpp

@@ -52,9 +52,6 @@ bool IsAmiiboValid(const EncryptedNTAG215File& ntag_file) {
     if (ntag_file.compability_container != 0xEEFF10F1U) {
         return false;
     }
-    if (amiibo_data.constant_value != 0xA5) {
-        return false;
-    }
     if (amiibo_data.model_info.tag_type != NFC::PackedTagType::Type2) {
         return false;
     }

src/core/hle/service/nfc/common/device.cpp

@@ -119,18 +119,31 @@ bool NfcDevice::LoadNfcTag(std::span<const u8> data) {
 
     memcpy(&tag_data, data.data(), sizeof(NFP::EncryptedNTAG215File));
     is_plain_amiibo = NFP::AmiiboCrypto::IsAmiiboValid(tag_data);
-
-    if (is_plain_amiibo) {
-        encrypted_tag_data = NFP::AmiiboCrypto::EncodedDataToNfcData(tag_data);
-        LOG_INFO(Service_NFP, "Using plain amiibo");
-    } else {
-        tag_data = {};
-        memcpy(&encrypted_tag_data, data.data(), sizeof(NFP::EncryptedNTAG215File));
-    }
+    is_write_protected = false;
 
     device_state = DeviceState::TagFound;
     deactivate_event->GetReadableEvent().Clear();
     activate_event->Signal();
+
+    // Fallback for plain amiibos
+    if (is_plain_amiibo) {
+        LOG_INFO(Service_NFP, "Using plain amiibo");
+        encrypted_tag_data = NFP::AmiiboCrypto::EncodedDataToNfcData(tag_data);
+        return true;
+    }
+
+    // Fallback for encrypted amiibos without keys
+    if (!NFP::AmiiboCrypto::IsKeyAvailable()) {
+        LOG_INFO(Service_NFC, "Loading amiibo without keys");
+        memcpy(&encrypted_tag_data, data.data(), sizeof(NFP::EncryptedNTAG215File));
+        BuildAmiiboWithoutKeys();
+        is_plain_amiibo = true;
+        is_write_protected = true;
+        return true;
+    }
+
+    tag_data = {};
+    memcpy(&encrypted_tag_data, data.data(), sizeof(NFP::EncryptedNTAG215File));
     return true;
 }
 
@@ -346,23 +359,15 @@ Result NfcDevice::Mount(NFP::ModelType model_type, NFP::MountTarget mount_target
         return ResultWrongDeviceState;
     }
 
-    // The loaded amiibo is not encrypted
-    if (is_plain_amiibo) {
-        device_state = DeviceState::TagMounted;
-        mount_target = mount_target_;
-        return ResultSuccess;
-    }
-
     if (!NFP::AmiiboCrypto::IsAmiiboValid(encrypted_tag_data)) {
         LOG_ERROR(Service_NFP, "Not an amiibo");
         return ResultNotAnAmiibo;
     }
 
-    // Mark amiibos as read only when keys are missing
-    if (!NFP::AmiiboCrypto::IsKeyAvailable()) {
-        LOG_ERROR(Service_NFP, "No keys detected");
+    // The loaded amiibo is not encrypted
+    if (is_plain_amiibo) {
         device_state = DeviceState::TagMounted;
-        mount_target = NFP::MountTarget::Rom;
+        mount_target = mount_target_;
         return ResultSuccess;
     }
 
@@ -421,11 +426,11 @@ Result NfcDevice::Flush() {
 
     tag_data.write_counter++;
 
-    FlushWithBreak(NFP::BreakType::Normal);
+    const auto result = FlushWithBreak(NFP::BreakType::Normal);
 
     is_data_moddified = false;
 
-    return ResultSuccess;
+    return result;
 }
 
 Result NfcDevice::FlushDebug() {
@@ -444,11 +449,11 @@ Result NfcDevice::FlushDebug() {
 
     tag_data.write_counter++;
 
-    FlushWithBreak(NFP::BreakType::Normal);
+    const auto result = FlushWithBreak(NFP::BreakType::Normal);
 
     is_data_moddified = false;
 
-    return ResultSuccess;
+    return result;
 }
 
 Result NfcDevice::FlushWithBreak(NFP::BreakType break_type) {
@@ -457,6 +462,11 @@ Result NfcDevice::FlushWithBreak(NFP::BreakType break_type) {
         return ResultWrongDeviceState;
     }
 
+    if (is_write_protected) {
+        LOG_ERROR(Service_NFP, "No keys available skipping write request");
+        return ResultSuccess;
+    }
+
     std::vector<u8> data(sizeof(NFP::EncryptedNTAG215File));
     if (is_plain_amiibo) {
         memcpy(data.data(), &tag_data, sizeof(tag_data));
@@ -1033,7 +1043,6 @@ Result NfcDevice::GetAll(NFP::NfpData& data) const {
     }
 
     NFP::CommonInfo common_info{};
-    Service::Mii::MiiManager manager;
     const u64 application_id = tag_data.application_id;
 
     GetCommonInfo(common_info);
@@ -1249,6 +1258,28 @@ void NfcDevice::UpdateRegisterInfoCrc() {
     tag_data.register_info_crc = crc.checksum();
 }
 
+void NfcDevice::BuildAmiiboWithoutKeys() {
+    Service::Mii::MiiManager manager;
+    auto& settings = tag_data.settings;
+
+    tag_data = NFP::AmiiboCrypto::NfcDataToEncodedData(encrypted_tag_data);
+
+    // Common info
+    tag_data.write_counter = 0;
+    tag_data.amiibo_version = 0;
+    settings.write_date = GetAmiiboDate(GetCurrentPosixTime());
+
+    // Register info
+    SetAmiiboName(settings, {'y', 'u', 'z', 'u', 'A', 'm', 'i', 'i', 'b', 'o'});
+    settings.settings.font_region.Assign(0);
+    settings.init_date = GetAmiiboDate(GetCurrentPosixTime());
+    tag_data.owner_mii = manager.BuildFromStoreData(manager.BuildDefault(0));
+
+    // Admin info
+    settings.settings.amiibo_initialized.Assign(1);
+    settings.settings.appdata_initialized.Assign(0);
+}
+
 u64 NfcDevice::GetHandle() const {
     // Generate a handle based of the npad id
     return static_cast<u64>(npad_id);

src/core/hle/service/nfc/common/device.h

@@ -110,6 +110,8 @@ private:
     void UpdateSettingsCrc();
     void UpdateRegisterInfoCrc();
 
+    void BuildAmiiboWithoutKeys();
+
     bool is_controller_set{};
     int callback_key;
     const Core::HID::NpadIdType npad_id;
@@ -128,6 +130,7 @@ private:
     bool is_data_moddified{};
     bool is_app_area_open{};
     bool is_plain_amiibo{};
+    bool is_write_protected{};
     NFP::MountTarget mount_target{NFP::MountTarget::None};
 
     NFP::NTAG215File tag_data{};

src/core/hle/service/nvdrv/core/syncpoint_manager.cpp

@@ -64,7 +64,7 @@ void SyncpointManager::FreeSyncpoint(u32 id) {
 }
 
 bool SyncpointManager::IsSyncpointAllocated(u32 id) const {
-    return (id <= SyncpointCount) && syncpoints[id].reserved;
+    return (id < SyncpointCount) && syncpoints[id].reserved;
 }
 
 bool SyncpointManager::HasSyncpointExpired(u32 id, u32 threshold) const {

src/input_common/drivers/joycon.cpp

@@ -291,9 +291,13 @@ Common::Input::NfcState Joycons::SupportsNfc(const PadIdentifier& identifier_) c
     return Common::Input::NfcState::Success;
 };
 
-Common::Input::NfcState Joycons::WriteNfcData(const PadIdentifier& identifier_,
+Common::Input::NfcState Joycons::WriteNfcData(const PadIdentifier& identifier,
                                               const std::vector<u8>& data) {
-    return Common::Input::NfcState::NotSupported;
+    auto handle = GetHandle(identifier);
+    if (handle->WriteNfcData(data) != Joycon::DriverResult::Success) {
+        return Common::Input::NfcState::WriteFailed;
+    }
+    return Common::Input::NfcState::Success;
 };
 
 Common::Input::DriverResult Joycons::SetPollingMode(const PadIdentifier& identifier,

src/input_common/helpers/joycon_driver.cpp

@@ -492,6 +492,26 @@ DriverResult JoyconDriver::SetRingConMode() {
     return result;
 }
 
+DriverResult JoyconDriver::WriteNfcData(std::span<const u8> data) {
+    std::scoped_lock lock{mutex};
+    disable_input_thread = true;
+
+    if (!supported_features.nfc) {
+        return DriverResult::NotSupported;
+    }
+    if (!nfc_protocol->IsEnabled()) {
+        return DriverResult::Disabled;
+    }
+    if (!amiibo_detected) {
+        return DriverResult::ErrorWritingData;
+    }
+
+    const auto result = nfc_protocol->WriteAmiibo(data);
+
+    disable_input_thread = false;
+    return result;
+}
+
 bool JoyconDriver::IsConnected() const {
     std::scoped_lock lock{mutex};
     return is_connected.load();

src/input_common/helpers/joycon_driver.h

@@ -49,6 +49,7 @@ public:
     DriverResult SetIrMode();
     DriverResult SetNfcMode();
     DriverResult SetRingConMode();
+    DriverResult WriteNfcData(std::span<const u8> data);
 
     void SetCallbacks(const JoyconCallbacks& callbacks);
 

src/input_common/helpers/joycon_protocol/joycon_types.h

@@ -23,6 +23,7 @@ constexpr std::array<u8, 8> DefaultVibrationBuffer{0x0, 0x1, 0x40, 0x40, 0x0, 0x
 
 using MacAddress = std::array<u8, 6>;
 using SerialNumber = std::array<u8, 15>;
+using TagUUID = std::array<u8, 7>;
 
 enum class ControllerType : u8 {
     None = 0x00,
@@ -276,12 +277,13 @@ enum class MCUPacketFlag : u8 {
     LastCommandPacket = 0x08,
 };
 
-enum class NFCReadCommand : u8 {
+enum class NFCCommand : u8 {
     CancelAll = 0x00,
     StartPolling = 0x01,
     StopPolling = 0x02,
     StartWaitingRecieve = 0x04,
-    Ntag = 0x06,
+    ReadNtag = 0x06,
+    WriteNtag = 0x08,
     Mifare = 0x0F,
 };
 
@@ -292,14 +294,19 @@ enum class NFCTagType : u8 {
 
 enum class NFCPages {
     Block0 = 0,
+    Block3 = 3,
     Block45 = 45,
     Block135 = 135,
     Block231 = 231,
 };
 
 enum class NFCStatus : u8 {
+    Ready = 0x00,
+    Polling = 0x01,
     LastPackage = 0x04,
+    WriteDone = 0x05,
     TagLost = 0x07,
+    WriteReady = 0x09,
 };
 
 enum class IrsMode : u8 {
@@ -559,13 +566,32 @@ static_assert(sizeof(NFCReadBlockCommand) == 0x9, "NFCReadBlockCommand is an inv
 struct NFCReadCommandData {
     u8 unknown;
     u8 uuid_length;
-    u8 unknown_2;
-    std::array<u8, 6> uid;
+    TagUUID uid;
     NFCTagType tag_type;
     NFCReadBlockCommand read_block;
 };
 static_assert(sizeof(NFCReadCommandData) == 0x13, "NFCReadCommandData is an invalid size");
 
+#pragma pack(push, 1)
+struct NFCWriteCommandData {
+    u8 unknown;
+    u8 uuid_length;
+    TagUUID uid;
+    NFCTagType tag_type;
+    u8 unknown2;
+    u8 unknown3;
+    u8 unknown4;
+    u8 unknown5;
+    u8 unknown6;
+    u8 unknown7;
+    u8 unknown8;
+    u8 magic;
+    u16_be write_count;
+    u8 amiibo_version;
+};
+static_assert(sizeof(NFCWriteCommandData) == 0x15, "NFCWriteCommandData is an invalid size");
+#pragma pack(pop)
+
 struct NFCPollingCommandData {
     u8 enable_mifare;
     u8 unknown_1;
@@ -576,8 +602,8 @@ struct NFCPollingCommandData {
 static_assert(sizeof(NFCPollingCommandData) == 0x05, "NFCPollingCommandData is an invalid size");
 
 struct NFCRequestState {
-    NFCReadCommand command_argument;
-    INSERT_PADDING_BYTES(0x1);
+    NFCCommand command_argument;
+    u8 block_id;
     u8 packet_id;
     MCUPacketFlag packet_flag;
     u8 data_length;
@@ -591,6 +617,18 @@ struct NFCRequestState {
 };
 static_assert(sizeof(NFCRequestState) == 0x26, "NFCRequestState is an invalid size");
 
+struct NFCDataChunk {
+    u8 nfc_page;
+    u8 data_size;
+    std::array<u8, 0xFF> data;
+};
+
+struct NFCWritePackage {
+    NFCWriteCommandData command_data;
+    u8 number_of_chunks;
+    std::array<NFCDataChunk, 4> data_chunks;
+};
+
 struct IrsConfigure {
     MCUCommand command;
     MCUSubCommand sub_command;

src/input_common/helpers/joycon_protocol/nfc.cpp

@@ -34,6 +34,12 @@ DriverResult NfcProtocol::EnableNfc() {
 
         result = ConfigureMCU(config);
     }
+    if (result == DriverResult::Success) {
+        result = WaitSetMCUMode(ReportMode::NFC_IR_MODE_60HZ, MCUMode::NFC);
+    }
+    if (result == DriverResult::Success) {
+        result = WaitUntilNfcIs(NFCStatus::Ready);
+    }
 
     return result;
 }
@@ -56,27 +62,20 @@ DriverResult NfcProtocol::StartNFCPollingMode() {
     LOG_DEBUG(Input, "Start NFC pooling Mode");
     ScopedSetBlocking sb(this);
     DriverResult result{DriverResult::Success};
-    TagFoundData tag_data{};
 
-    if (result == DriverResult::Success) {
-        result = WaitSetMCUMode(ReportMode::NFC_IR_MODE_60HZ, MCUMode::NFC);
-    }
-    if (result == DriverResult::Success) {
-        result = WaitUntilNfcIsReady();
-    }
     if (result == DriverResult::Success) {
         MCUCommandResponse output{};
         result = SendStopPollingRequest(output);
     }
     if (result == DriverResult::Success) {
-        result = WaitUntilNfcIsReady();
+        result = WaitUntilNfcIs(NFCStatus::Ready);
     }
     if (result == DriverResult::Success) {
         MCUCommandResponse output{};
         result = SendStartPollingRequest(output);
     }
     if (result == DriverResult::Success) {
-        result = WaitUntilNfcIsPolling();
+        result = WaitUntilNfcIs(NFCStatus::Polling);
     }
     if (result == DriverResult::Success) {
         is_enabled = true;
@@ -112,6 +111,49 @@ DriverResult NfcProtocol::ScanAmiibo(std::vector<u8>& data) {
     return result;
 }
 
+DriverResult NfcProtocol::WriteAmiibo(std::span<const u8> data) {
+    LOG_DEBUG(Input, "Write amiibo");
+    ScopedSetBlocking sb(this);
+    DriverResult result{DriverResult::Success};
+    TagUUID tag_uuid = GetTagUUID(data);
+    TagFoundData tag_data{};
+
+    if (result == DriverResult::Success) {
+        result = IsTagInRange(tag_data, 7);
+    }
+    if (result == DriverResult::Success) {
+        if (tag_data.uuid != tag_uuid) {
+            result = DriverResult::InvalidParameters;
+        }
+    }
+    if (result == DriverResult::Success) {
+        MCUCommandResponse output{};
+        result = SendStopPollingRequest(output);
+    }
+    if (result == DriverResult::Success) {
+        result = WaitUntilNfcIs(NFCStatus::Ready);
+    }
+    if (result == DriverResult::Success) {
+        MCUCommandResponse output{};
+        result = SendStartPollingRequest(output, true);
+    }
+    if (result == DriverResult::Success) {
+        result = WaitUntilNfcIs(NFCStatus::WriteReady);
+    }
+    if (result == DriverResult::Success) {
+        result = WriteAmiiboData(tag_uuid, data);
+    }
+    if (result == DriverResult::Success) {
+        result = WaitUntilNfcIs(NFCStatus::WriteDone);
+    }
+    if (result == DriverResult::Success) {
+        MCUCommandResponse output{};
+        result = SendStopPollingRequest(output);
+    }
+
+    return result;
+}
+
 bool NfcProtocol::HasAmiibo() {
     if (update_counter++ < AMIIBO_UPDATE_DELAY) {
         return true;
@@ -129,7 +171,7 @@ bool NfcProtocol::HasAmiibo() {
     return result == DriverResult::Success;
 }
 
-DriverResult NfcProtocol::WaitUntilNfcIsReady() {
+DriverResult NfcProtocol::WaitUntilNfcIs(NFCStatus status) {
     constexpr std::size_t timeout_limit = 10;
     MCUCommandResponse output{};
     std::size_t tries = 0;
@@ -145,28 +187,7 @@ DriverResult NfcProtocol::WaitUntilNfcIsReady() {
         }
     } while (output.mcu_report != MCUReport::NFCState ||
              (output.mcu_data[1] << 8) + output.mcu_data[0] != 0x0500 ||
-             output.mcu_data[5] != 0x31 || output.mcu_data[6] != 0x00);
-
-    return DriverResult::Success;
-}
-
-DriverResult NfcProtocol::WaitUntilNfcIsPolling() {
-    constexpr std::size_t timeout_limit = 10;
-    MCUCommandResponse output{};
-    std::size_t tries = 0;
-
-    do {
-        auto result = SendNextPackageRequest(output, {});
-
-        if (result != DriverResult::Success) {
-            return result;
-        }
-        if (tries++ > timeout_limit) {
-            return DriverResult::Timeout;
-        }
-    } while (output.mcu_report != MCUReport::NFCState ||
-             (output.mcu_data[1] << 8) + output.mcu_data[0] != 0x0500 ||
-             output.mcu_data[5] != 0x31 || output.mcu_data[6] != 0x01);
+             output.mcu_data[5] != 0x31 || output.mcu_data[6] != static_cast<u8>(status));
 
     return DriverResult::Success;
 }
@@ -188,7 +209,7 @@ DriverResult NfcProtocol::IsTagInRange(TagFoundData& data, std::size_t timeout_l
              (output.mcu_data[6] != 0x09 && output.mcu_data[6] != 0x04));
 
     data.type = output.mcu_data[12];
-    data.uuid.resize(output.mcu_data[14]);
+    data.uuid_size = std::min(output.mcu_data[14], static_cast<u8>(sizeof(TagUUID)));
     memcpy(data.uuid.data(), output.mcu_data.data() + 15, data.uuid.size());
 
     return DriverResult::Success;
@@ -245,17 +266,94 @@ DriverResult NfcProtocol::GetAmiiboData(std::vector<u8>& ntag_data) {
     return DriverResult::Timeout;
 }
 
-DriverResult NfcProtocol::SendStartPollingRequest(MCUCommandResponse& output) {
+DriverResult NfcProtocol::WriteAmiiboData(const TagUUID& tag_uuid, std::span<const u8> data) {
+    constexpr std::size_t timeout_limit = 60;
+    const auto nfc_data = MakeAmiiboWritePackage(tag_uuid, data);
+    const std::vector<u8> nfc_buffer_data = SerializeWritePackage(nfc_data);
+    std::span<const u8> buffer(nfc_buffer_data);
+    MCUCommandResponse output{};
+    u8 block_id = 1;
+    u8 package_index = 0;
+    std::size_t tries = 0;
+    std::size_t current_position = 0;
+
+    LOG_INFO(Input, "Writing amiibo data");
+
+    auto result = SendWriteAmiiboRequest(output, tag_uuid);
+
+    if (result != DriverResult::Success) {
+        return result;
+    }
+
+    // Read Tag data but ignore the actual sent data
+    while (tries++ < timeout_limit) {
+        result = SendNextPackageRequest(output, package_index);
+        const auto nfc_status = static_cast<NFCStatus>(output.mcu_data[6]);
+
+        if (result != DriverResult::Success) {
+            return result;
+        }
+
+        if ((output.mcu_report == MCUReport::NFCReadData ||
+             output.mcu_report == MCUReport::NFCState) &&
+            nfc_status == NFCStatus::TagLost) {
+            return DriverResult::ErrorReadingData;
+        }
+
+        if (output.mcu_report == MCUReport::NFCReadData && output.mcu_data[1] == 0x07) {
+            package_index++;
+            continue;
+        }
+
+        if (output.mcu_report == MCUReport::NFCState && nfc_status == NFCStatus::LastPackage) {
+            LOG_INFO(Input, "Finished reading amiibo");
+            break;
+        }
+    }
+
+    // Send Data. Nfc buffer size is 31, Send the data in smaller packages
+    while (current_position < buffer.size() && tries++ < timeout_limit) {
+        const std::size_t next_position =
+            std::min(current_position + sizeof(NFCRequestState::raw_data), buffer.size());
+        const std::size_t block_size = next_position - current_position;
+        const bool is_last_packet = block_size < sizeof(NFCRequestState::raw_data);
+
+        SendWriteDataAmiiboRequest(output, block_id, is_last_packet,
+                                   buffer.subspan(current_position, block_size));
+
+        const auto nfc_status = static_cast<NFCStatus>(output.mcu_data[6]);
+
+        if ((output.mcu_report == MCUReport::NFCReadData ||
+             output.mcu_report == MCUReport::NFCState) &&
+            nfc_status == NFCStatus::TagLost) {
+            return DriverResult::ErrorReadingData;
+        }
+
+        // Increase position when data is confirmed by the joycon
+        if (output.mcu_report == MCUReport::NFCState &&
+            (output.mcu_data[1] << 8) + output.mcu_data[0] == 0x0500 &&
+            output.mcu_data[3] == block_id) {
+            block_id++;
+            current_position = next_position;
+        }
+    }
+
+    return result;
+}
+
+DriverResult NfcProtocol::SendStartPollingRequest(MCUCommandResponse& output,
+                                                  bool is_second_attempt) {
     NFCRequestState request{
-        .command_argument = NFCReadCommand::StartPolling,
-        .packet_id = 0x0,
+        .command_argument = NFCCommand::StartPolling,
+        .block_id = {},
+        .packet_id = {},
         .packet_flag = MCUPacketFlag::LastCommandPacket,
         .data_length = sizeof(NFCPollingCommandData),
         .nfc_polling =
             {
-                .enable_mifare = 0x01,
-                .unknown_1 = 0x00,
-                .unknown_2 = 0x00,
+                .enable_mifare = 0x00,
+                .unknown_1 = static_cast<u8>(is_second_attempt ? 0xe8 : 0x00),
+                .unknown_2 = static_cast<u8>(is_second_attempt ? 0x03 : 0x00),
                 .unknown_3 = 0x2c,
                 .unknown_4 = 0x01,
             },
@@ -271,10 +369,11 @@ DriverResult NfcProtocol::SendStartPollingRequest(MCUCommandResponse& output) {
 
 DriverResult NfcProtocol::SendStopPollingRequest(MCUCommandResponse& output) {
     NFCRequestState request{
-        .command_argument = NFCReadCommand::StopPolling,
-        .packet_id = 0x0,
+        .command_argument = NFCCommand::StopPolling,
+        .block_id = {},
+        .packet_id = {},
         .packet_flag = MCUPacketFlag::LastCommandPacket,
-        .data_length = 0,
+        .data_length = {},
         .raw_data = {},
         .crc = {},
     };
@@ -288,10 +387,11 @@ DriverResult NfcProtocol::SendStopPollingRequest(MCUCommandResponse& output) {
 
 DriverResult NfcProtocol::SendNextPackageRequest(MCUCommandResponse& output, u8 packet_id) {
     NFCRequestState request{
-        .command_argument = NFCReadCommand::StartWaitingRecieve,
+        .command_argument = NFCCommand::StartWaitingRecieve,
+        .block_id = {},
         .packet_id = packet_id,
         .packet_flag = MCUPacketFlag::LastCommandPacket,
-        .data_length = 0,
+        .data_length = {},
         .raw_data = {},
         .crc = {},
     };
@@ -305,17 +405,17 @@ DriverResult NfcProtocol::SendNextPackageRequest(MCUCommandResponse& output, u8
 
 DriverResult NfcProtocol::SendReadAmiiboRequest(MCUCommandResponse& output, NFCPages ntag_pages) {
     NFCRequestState request{
-        .command_argument = NFCReadCommand::Ntag,
-        .packet_id = 0x0,
+        .command_argument = NFCCommand::ReadNtag,
+        .block_id = {},
+        .packet_id = {},
         .packet_flag = MCUPacketFlag::LastCommandPacket,
         .data_length = sizeof(NFCReadCommandData),
         .nfc_read =
             {
                 .unknown = 0xd0,
-                .uuid_length = 0x07,
-                .unknown_2 = 0x00,
+                .uuid_length = sizeof(NFCReadCommandData::uid),
                 .uid = {},
-                .tag_type = NFCTagType::AllTags,
+                .tag_type = NFCTagType::Ntag215,
                 .read_block = GetReadBlockCommand(ntag_pages),
             },
         .crc = {},
@@ -328,12 +428,135 @@ DriverResult NfcProtocol::SendReadAmiiboRequest(MCUCommandResponse& output, NFCP
                        output);
 }
 
+DriverResult NfcProtocol::SendWriteAmiiboRequest(MCUCommandResponse& output,
+                                                 const TagUUID& tag_uuid) {
+    NFCRequestState request{
+        .command_argument = NFCCommand::ReadNtag,
+        .block_id = {},
+        .packet_id = {},
+        .packet_flag = MCUPacketFlag::LastCommandPacket,
+        .data_length = sizeof(NFCReadCommandData),
+        .nfc_read =
+            {
+                .unknown = 0xd0,
+                .uuid_length = sizeof(NFCReadCommandData::uid),
+                .uid = tag_uuid,
+                .tag_type = NFCTagType::Ntag215,
+                .read_block = GetReadBlockCommand(NFCPages::Block3),
+            },
+        .crc = {},
+    };
+
+    std::array<u8, sizeof(NFCRequestState)> request_data{};
+    memcpy(request_data.data(), &request, sizeof(NFCRequestState));
+    request_data[36] = CalculateMCU_CRC8(request_data.data(), 36);
+    return SendMCUData(ReportMode::NFC_IR_MODE_60HZ, MCUSubCommand::ReadDeviceMode, request_data,
+                       output);
+}
+
+DriverResult NfcProtocol::SendWriteDataAmiiboRequest(MCUCommandResponse& output, u8 block_id,
+                                                     bool is_last_packet,
+                                                     std::span<const u8> data) {
+    const auto data_size = std::min(data.size(), sizeof(NFCRequestState::raw_data));
+    NFCRequestState request{
+        .command_argument = NFCCommand::WriteNtag,
+        .block_id = block_id,
+        .packet_id = {},
+        .packet_flag =
+            is_last_packet ? MCUPacketFlag::LastCommandPacket : MCUPacketFlag::MorePacketsRemaining,
+        .data_length = static_cast<u8>(data_size),
+        .raw_data = {},
+        .crc = {},
+    };
+    memcpy(request.raw_data.data(), data.data(), data_size);
+
+    std::array<u8, sizeof(NFCRequestState)> request_data{};
+    memcpy(request_data.data(), &request, sizeof(NFCRequestState));
+    request_data[36] = CalculateMCU_CRC8(request_data.data(), 36);
+    return SendMCUData(ReportMode::NFC_IR_MODE_60HZ, MCUSubCommand::ReadDeviceMode, request_data,
+                       output);
+}
+
+std::vector<u8> NfcProtocol::SerializeWritePackage(const NFCWritePackage& package) const {
+    const std::size_t header_size =
+        sizeof(NFCWriteCommandData) + sizeof(NFCWritePackage::number_of_chunks);
+    std::vector<u8> serialized_data(header_size);
+    std::size_t start_index = 0;
+
+    memcpy(serialized_data.data(), &package, header_size);
+    start_index += header_size;
+
+    for (const auto& data_chunk : package.data_chunks) {
+        const std::size_t chunk_size =
+            sizeof(NFCDataChunk::nfc_page) + sizeof(NFCDataChunk::data_size) + data_chunk.data_size;
+
+        serialized_data.resize(start_index + chunk_size);
+        memcpy(serialized_data.data() + start_index, &data_chunk, chunk_size);
+        start_index += chunk_size;
+    }
+
+    return serialized_data;
+}
+
+NFCWritePackage NfcProtocol::MakeAmiiboWritePackage(const TagUUID& tag_uuid,
+                                                    std::span<const u8> data) const {
+    return {
+        .command_data{
+            .unknown = 0xd0,
+            .uuid_length = sizeof(NFCReadCommandData::uid),
+            .uid = tag_uuid,
+            .tag_type = NFCTagType::Ntag215,
+            .unknown2 = 0x00,
+            .unknown3 = 0x01,
+            .unknown4 = 0x04,
+            .unknown5 = 0xff,
+            .unknown6 = 0xff,
+            .unknown7 = 0xff,
+            .unknown8 = 0xff,
+            .magic = data[16],
+            .write_count = static_cast<u16>((data[17] << 8) + data[18]),
+            .amiibo_version = data[19],
+        },
+        .number_of_chunks = 3,
+        .data_chunks =
+            {
+                MakeAmiiboChunk(0x05, 0x20, data),
+                MakeAmiiboChunk(0x20, 0xf0, data),
+                MakeAmiiboChunk(0x5c, 0x98, data),
+            },
+    };
+}
+
+NFCDataChunk NfcProtocol::MakeAmiiboChunk(u8 page, u8 size, std::span<const u8> data) const {
+    constexpr u8 NFC_PAGE_SIZE = 4;
+
+    if (static_cast<std::size_t>(page * NFC_PAGE_SIZE) + size >= data.size()) {
+        return {};
+    }
+
+    NFCDataChunk chunk{
+        .nfc_page = page,
+        .data_size = size,
+        .data = {},
+    };
+    std::memcpy(chunk.data.data(), data.data() + (page * NFC_PAGE_SIZE), size);
+    return chunk;
+}
+
 NFCReadBlockCommand NfcProtocol::GetReadBlockCommand(NFCPages pages) const {
     switch (pages) {
     case NFCPages::Block0:
         return {
             .block_count = 1,
         };
+    case NFCPages::Block3:
+        return {
+            .block_count = 1,
+            .blocks =
+                {
+                    NFCReadBlock{0x03, 0x03},
+                },
+        };
     case NFCPages::Block45:
         return {
             .block_count = 1,
@@ -368,6 +591,17 @@ NFCReadBlockCommand NfcProtocol::GetReadBlockCommand(NFCPages pages) const {
     };
 }
 
+TagUUID NfcProtocol::GetTagUUID(std::span<const u8> data) const {
+    if (data.size() < 10) {
+        return {};
+    }
+
+    // crc byte 3 is omitted in this operation
+    return {
+        data[0], data[1], data[2], data[4], data[5], data[6], data[7],
+    };
+}
+
 bool NfcProtocol::IsEnabled() const {
     return is_enabled;
 }

src/input_common/helpers/joycon_protocol/nfc.h

@@ -27,6 +27,8 @@ public:
 
     DriverResult ScanAmiibo(std::vector<u8>& data);
 
+    DriverResult WriteAmiibo(std::span<const u8> data);
+
     bool HasAmiibo();
 
     bool IsEnabled() const;
@@ -37,18 +39,20 @@ private:
 
     struct TagFoundData {
         u8 type;
-        std::vector<u8> uuid;
+        u8 uuid_size;
+        TagUUID uuid;
     };
 
-    DriverResult WaitUntilNfcIsReady();
-
-    DriverResult WaitUntilNfcIsPolling();
+    DriverResult WaitUntilNfcIs(NFCStatus status);
 
     DriverResult IsTagInRange(TagFoundData& data, std::size_t timeout_limit = 1);
 
     DriverResult GetAmiiboData(std::vector<u8>& data);
 
-    DriverResult SendStartPollingRequest(MCUCommandResponse& output);
+    DriverResult WriteAmiiboData(const TagUUID& tag_uuid, std::span<const u8> data);
+
+    DriverResult SendStartPollingRequest(MCUCommandResponse& output,
+                                         bool is_second_attempt = false);
 
     DriverResult SendStopPollingRequest(MCUCommandResponse& output);
 
@@ -56,8 +60,21 @@ private:
 
     DriverResult SendReadAmiiboRequest(MCUCommandResponse& output, NFCPages ntag_pages);
 
+    DriverResult SendWriteAmiiboRequest(MCUCommandResponse& output, const TagUUID& tag_uuid);
+
+    DriverResult SendWriteDataAmiiboRequest(MCUCommandResponse& output, u8 block_id,
+                                            bool is_last_packet, std::span<const u8> data);
+
+    std::vector<u8> SerializeWritePackage(const NFCWritePackage& package) const;
+
+    NFCWritePackage MakeAmiiboWritePackage(const TagUUID& tag_uuid, std::span<const u8> data) const;
+
+    NFCDataChunk MakeAmiiboChunk(u8 page, u8 size, std::span<const u8> data) const;
+
     NFCReadBlockCommand GetReadBlockCommand(NFCPages pages) const;
 
+    TagUUID GetTagUUID(std::span<const u8> data) const;
+
     bool is_enabled{};
     std::size_t update_counter{};
 };

src/input_common/input_engine.cpp

@@ -380,13 +380,16 @@ void InputEngine::TriggerOnMotionChange(const PadIdentifier& identifier, int mot
     if (!configuring || !mapping_callback.on_data) {
         return;
     }
+    const auto old_value = GetMotion(identifier, motion);
     bool is_active = false;
-    if (std::abs(value.accel_x) > 1.5f || std::abs(value.accel_y) > 1.5f ||
-        std::abs(value.accel_z) > 1.5f) {
+    if (std::abs(value.accel_x - old_value.accel_x) > 1.5f ||
+        std::abs(value.accel_y - old_value.accel_y) > 1.5f ||
+        std::abs(value.accel_z - old_value.accel_z) > 1.5f) {
         is_active = true;
     }
-    if (std::abs(value.gyro_x) > 0.6f || std::abs(value.gyro_y) > 0.6f ||
-        std::abs(value.gyro_z) > 0.6f) {
+    if (std::abs(value.gyro_x - old_value.gyro_x) > 0.6f ||
+        std::abs(value.gyro_y - old_value.gyro_y) > 0.6f ||
+        std::abs(value.gyro_z - old_value.gyro_z) > 0.6f) {
         is_active = true;
     }
     if (!is_active) {

src/shader_recompiler/frontend/maxwell/translate/impl/integer_funnel_shift.cpp

@@ -30,7 +30,7 @@ void SHF(TranslatorVisitor& v, u64 insn, const IR::U32& shift, const IR::U32& hi
     union {
         u64 insn;
         BitField<0, 8, IR::Reg> dest_reg;
-        BitField<0, 8, IR::Reg> lo_bits_reg;
+        BitField<8, 8, IR::Reg> lo_bits_reg;
         BitField<37, 2, MaxShift> max_shift;
         BitField<47, 1, u64> cc;
         BitField<48, 2, u64> x_mode;

src/video_core/CMakeLists.txt

@@ -246,10 +246,14 @@ add_library(video_core STATIC
     texture_cache/util.h
     textures/astc.h
     textures/astc.cpp
+    textures/bcn.cpp
+    textures/bcn.h
     textures/decoders.cpp
     textures/decoders.h
     textures/texture.cpp
     textures/texture.h
+    textures/workers.cpp
+    textures/workers.h
     transform_feedback.cpp
     transform_feedback.h
     video_core.cpp
@@ -275,7 +279,7 @@ add_library(video_core STATIC
 create_target_directory_groups(video_core)
 
 target_link_libraries(video_core PUBLIC common core)
-target_link_libraries(video_core PUBLIC glad shader_recompiler)
+target_link_libraries(video_core PUBLIC glad shader_recompiler stb)
 
 if (YUZU_USE_BUNDLED_FFMPEG AND NOT WIN32)
     add_dependencies(video_core ffmpeg-build)

src/video_core/buffer_cache/buffer_cache.cpp

@@ -2,6 +2,8 @@
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 #include "common/microprofile.h"
+#include "video_core/buffer_cache/buffer_cache_base.h"
+#include "video_core/control/channel_state_cache.inc"
 
 namespace VideoCommon {
 
@@ -9,4 +11,6 @@ MICROPROFILE_DEFINE(GPU_PrepareBuffers, "GPU", "Prepare buffers", MP_RGB(224, 12
 MICROPROFILE_DEFINE(GPU_BindUploadBuffers, "GPU", "Bind and upload buffers", MP_RGB(224, 128, 128));
 MICROPROFILE_DEFINE(GPU_DownloadMemory, "GPU", "Download buffers", MP_RGB(224, 128, 128));
 
+template class VideoCommon::ChannelSetupCaches<VideoCommon::BufferCacheChannelInfo>;
+
 } // namespace VideoCommon

src/video_core/buffer_cache/buffer_cache.h

@@ -30,8 +30,8 @@ BufferCache<P>::BufferCache(VideoCore::RasterizerInterface& rasterizer_,
     }
 
     const s64 device_memory = static_cast<s64>(runtime.GetDeviceLocalMemory());
-    const s64 min_spacing_expected = device_memory - 1_GiB - 512_MiB;
-    const s64 min_spacing_critical = device_memory - 1_GiB;
+    const s64 min_spacing_expected = device_memory - 1_GiB;
+    const s64 min_spacing_critical = device_memory - 512_MiB;
     const s64 mem_threshold = std::min(device_memory, TARGET_THRESHOLD);
     const s64 min_vacancy_expected = (6 * mem_threshold) / 10;
     const s64 min_vacancy_critical = (3 * mem_threshold) / 10;
@@ -64,17 +64,22 @@ void BufferCache<P>::RunGarbageCollector() {
 template <class P>
 void BufferCache<P>::TickFrame() {
     // Calculate hits and shots and move hit bits to the right
-    const u32 hits = std::reduce(uniform_cache_hits.begin(), uniform_cache_hits.end());
-    const u32 shots = std::reduce(uniform_cache_shots.begin(), uniform_cache_shots.end());
-    std::copy_n(uniform_cache_hits.begin(), uniform_cache_hits.size() - 1,
-                uniform_cache_hits.begin() + 1);
-    std::copy_n(uniform_cache_shots.begin(), uniform_cache_shots.size() - 1,
-                uniform_cache_shots.begin() + 1);
-    uniform_cache_hits[0] = 0;
-    uniform_cache_shots[0] = 0;
+
+    const u32 hits = std::reduce(channel_state->uniform_cache_hits.begin(),
+                                 channel_state->uniform_cache_hits.end());
+    const u32 shots = std::reduce(channel_state->uniform_cache_shots.begin(),
+                                  channel_state->uniform_cache_shots.end());
+    std::copy_n(channel_state->uniform_cache_hits.begin(),
+                channel_state->uniform_cache_hits.size() - 1,
+                channel_state->uniform_cache_hits.begin() + 1);
+    std::copy_n(channel_state->uniform_cache_shots.begin(),
+                channel_state->uniform_cache_shots.size() - 1,
+                channel_state->uniform_cache_shots.begin() + 1);
+    channel_state->uniform_cache_hits[0] = 0;
+    channel_state->uniform_cache_shots[0] = 0;
 
     const bool skip_preferred = hits * 256 < shots * 251;
-    uniform_buffer_skip_cache_size = skip_preferred ? DEFAULT_SKIP_CACHE_SIZE : 0;
+    channel_state->uniform_buffer_skip_cache_size = skip_preferred ? DEFAULT_SKIP_CACHE_SIZE : 0;
 
     // If we can obtain the memory info, use it instead of the estimate.
     if (runtime.CanReportMemoryUsage()) {
@@ -164,10 +169,10 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
     BufferId buffer_a;
     BufferId buffer_b;
     do {
-        has_deleted_buffers = false;
+        channel_state->has_deleted_buffers = false;
         buffer_a = FindBuffer(*cpu_src_address, static_cast<u32>(amount));
         buffer_b = FindBuffer(*cpu_dest_address, static_cast<u32>(amount));
-    } while (has_deleted_buffers);
+    } while (channel_state->has_deleted_buffers);
     auto& src_buffer = slot_buffers[buffer_a];
     auto& dest_buffer = slot_buffers[buffer_b];
     SynchronizeBuffer(src_buffer, *cpu_src_address, static_cast<u32>(amount));
@@ -272,30 +277,30 @@ void BufferCache<P>::BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr
         .size = size,
         .buffer_id = BufferId{},
     };
-    uniform_buffers[stage][index] = binding;
+    channel_state->uniform_buffers[stage][index] = binding;
 }
 
 template <class P>
 void BufferCache<P>::DisableGraphicsUniformBuffer(size_t stage, u32 index) {
-    uniform_buffers[stage][index] = NULL_BINDING;
+    channel_state->uniform_buffers[stage][index] = NULL_BINDING;
 }
 
 template <class P>
 void BufferCache<P>::UpdateGraphicsBuffers(bool is_indexed) {
     MICROPROFILE_SCOPE(GPU_PrepareBuffers);
     do {
-        has_deleted_buffers = false;
+        channel_state->has_deleted_buffers = false;
         DoUpdateGraphicsBuffers(is_indexed);
-    } while (has_deleted_buffers);
+    } while (channel_state->has_deleted_buffers);
 }
 
 template <class P>
 void BufferCache<P>::UpdateComputeBuffers() {
     MICROPROFILE_SCOPE(GPU_PrepareBuffers);
     do {
-        has_deleted_buffers = false;
+        channel_state->has_deleted_buffers = false;
         DoUpdateComputeBuffers();
-    } while (has_deleted_buffers);
+    } while (channel_state->has_deleted_buffers);
 }
 
 template <class P>
@@ -338,98 +343,102 @@ template <class P>
 void BufferCache<P>::SetUniformBuffersState(const std::array<u32, NUM_STAGES>& mask,
                                             const UniformBufferSizes* sizes) {
     if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
-        if (enabled_uniform_buffer_masks != mask) {
+        if (channel_state->enabled_uniform_buffer_masks != mask) {
             if constexpr (IS_OPENGL) {
-                fast_bound_uniform_buffers.fill(0);
+                channel_state->fast_bound_uniform_buffers.fill(0);
             }
-            dirty_uniform_buffers.fill(~u32{0});
-            uniform_buffer_binding_sizes.fill({});
+            channel_state->dirty_uniform_buffers.fill(~u32{0});
+            channel_state->uniform_buffer_binding_sizes.fill({});
         }
     }
-    enabled_uniform_buffer_masks = mask;
-    uniform_buffer_sizes = sizes;
+    channel_state->enabled_uniform_buffer_masks = mask;
+    channel_state->uniform_buffer_sizes = sizes;
 }
 
 template <class P>
 void BufferCache<P>::SetComputeUniformBufferState(u32 mask,
                                                   const ComputeUniformBufferSizes* sizes) {
-    enabled_compute_uniform_buffer_mask = mask;
-    compute_uniform_buffer_sizes = sizes;
+    channel_state->enabled_compute_uniform_buffer_mask = mask;
+    channel_state->compute_uniform_buffer_sizes = sizes;
 }
 
 template <class P>
 void BufferCache<P>::UnbindGraphicsStorageBuffers(size_t stage) {
-    enabled_storage_buffers[stage] = 0;
-    written_storage_buffers[stage] = 0;
+    channel_state->enabled_storage_buffers[stage] = 0;
+    channel_state->written_storage_buffers[stage] = 0;
 }
 
 template <class P>
 void BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index,
                                                u32 cbuf_offset, bool is_written) {
-    enabled_storage_buffers[stage] |= 1U << ssbo_index;
-    written_storage_buffers[stage] |= (is_written ? 1U : 0U) << ssbo_index;
+    channel_state->enabled_storage_buffers[stage] |= 1U << ssbo_index;
+    channel_state->written_storage_buffers[stage] |= (is_written ? 1U : 0U) << ssbo_index;
 
     const auto& cbufs = maxwell3d->state.shader_stages[stage];
     const GPUVAddr ssbo_addr = cbufs.const_buffers[cbuf_index].address + cbuf_offset;
-    storage_buffers[stage][ssbo_index] = StorageBufferBinding(ssbo_addr, cbuf_index, is_written);
+    channel_state->storage_buffers[stage][ssbo_index] =
+        StorageBufferBinding(ssbo_addr, cbuf_index, is_written);
 }
 
 template <class P>
 void BufferCache<P>::UnbindGraphicsTextureBuffers(size_t stage) {
-    enabled_texture_buffers[stage] = 0;
-    written_texture_buffers[stage] = 0;
-    image_texture_buffers[stage] = 0;
+    channel_state->enabled_texture_buffers[stage] = 0;
+    channel_state->written_texture_buffers[stage] = 0;
+    channel_state->image_texture_buffers[stage] = 0;
 }
 
 template <class P>
 void BufferCache<P>::BindGraphicsTextureBuffer(size_t stage, size_t tbo_index, GPUVAddr gpu_addr,
                                                u32 size, PixelFormat format, bool is_written,
                                                bool is_image) {
-    enabled_texture_buffers[stage] |= 1U << tbo_index;
-    written_texture_buffers[stage] |= (is_written ? 1U : 0U) << tbo_index;
+    channel_state->enabled_texture_buffers[stage] |= 1U << tbo_index;
+    channel_state->written_texture_buffers[stage] |= (is_written ? 1U : 0U) << tbo_index;
     if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) {
-        image_texture_buffers[stage] |= (is_image ? 1U : 0U) << tbo_index;
+        channel_state->image_texture_buffers[stage] |= (is_image ? 1U : 0U) << tbo_index;
     }
-    texture_buffers[stage][tbo_index] = GetTextureBufferBinding(gpu_addr, size, format);
+    channel_state->texture_buffers[stage][tbo_index] =
+        GetTextureBufferBinding(gpu_addr, size, format);
 }
 
 template <class P>
 void BufferCache<P>::UnbindComputeStorageBuffers() {
-    enabled_compute_storage_buffers = 0;
-    written_compute_storage_buffers = 0;
-    image_compute_texture_buffers = 0;
+    channel_state->enabled_compute_storage_buffers = 0;
+    channel_state->written_compute_storage_buffers = 0;
+    channel_state->image_compute_texture_buffers = 0;
 }
 
 template <class P>
 void BufferCache<P>::BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
                                               bool is_written) {
-    enabled_compute_storage_buffers |= 1U << ssbo_index;
-    written_compute_storage_buffers |= (is_written ? 1U : 0U) << ssbo_index;
+    channel_state->enabled_compute_storage_buffers |= 1U << ssbo_index;
+    channel_state->written_compute_storage_buffers |= (is_written ? 1U : 0U) << ssbo_index;
 
     const auto& launch_desc = kepler_compute->launch_description;
     ASSERT(((launch_desc.const_buffer_enable_mask >> cbuf_index) & 1) != 0);
 
     const auto& cbufs = launch_desc.const_buffer_config;
     const GPUVAddr ssbo_addr = cbufs[cbuf_index].Address() + cbuf_offset;
-    compute_storage_buffers[ssbo_index] = StorageBufferBinding(ssbo_addr, cbuf_index, is_written);
+    channel_state->compute_storage_buffers[ssbo_index] =
+        StorageBufferBinding(ssbo_addr, cbuf_index, is_written);
 }
 
 template <class P>
 void BufferCache<P>::UnbindComputeTextureBuffers() {
-    enabled_compute_texture_buffers = 0;
-    written_compute_texture_buffers = 0;
-    image_compute_texture_buffers = 0;
+    channel_state->enabled_compute_texture_buffers = 0;
+    channel_state->written_compute_texture_buffers = 0;
+    channel_state->image_compute_texture_buffers = 0;
 }
 
 template <class P>
 void BufferCache<P>::BindComputeTextureBuffer(size_t tbo_index, GPUVAddr gpu_addr, u32 size,
                                               PixelFormat format, bool is_written, bool is_image) {
-    enabled_compute_texture_buffers |= 1U << tbo_index;
-    written_compute_texture_buffers |= (is_written ? 1U : 0U) << tbo_index;
+    channel_state->enabled_compute_texture_buffers |= 1U << tbo_index;
+    channel_state->written_compute_texture_buffers |= (is_written ? 1U : 0U) << tbo_index;
     if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) {
-        image_compute_texture_buffers |= (is_image ? 1U : 0U) << tbo_index;
+        channel_state->image_compute_texture_buffers |= (is_image ? 1U : 0U) << tbo_index;
     }
-    compute_texture_buffers[tbo_index] = GetTextureBufferBinding(gpu_addr, size, format);
+    channel_state->compute_texture_buffers[tbo_index] =
+        GetTextureBufferBinding(gpu_addr, size, format);
 }
 
 template <class P>
@@ -672,10 +681,10 @@ bool BufferCache<P>::IsRegionCpuModified(VAddr addr, size_t size) {
 
 template <class P>
 void BufferCache<P>::BindHostIndexBuffer() {
-    Buffer& buffer = slot_buffers[index_buffer.buffer_id];
-    TouchBuffer(buffer, index_buffer.buffer_id);
-    const u32 offset = buffer.Offset(index_buffer.cpu_addr);
-    const u32 size = index_buffer.size;
+    Buffer& buffer = slot_buffers[channel_state->index_buffer.buffer_id];
+    TouchBuffer(buffer, channel_state->index_buffer.buffer_id);
+    const u32 offset = buffer.Offset(channel_state->index_buffer.cpu_addr);
+    const u32 size = channel_state->index_buffer.size;
     const auto& draw_state = maxwell3d->draw_manager->GetDrawState();
     if (!draw_state.inline_index_draw_indexes.empty()) [[unlikely]] {
         if constexpr (USE_MEMORY_MAPS) {
@@ -689,7 +698,7 @@ void BufferCache<P>::BindHostIndexBuffer() {
             buffer.ImmediateUpload(0, draw_state.inline_index_draw_indexes);
         }
     } else {
-        SynchronizeBuffer(buffer, index_buffer.cpu_addr, size);
+        SynchronizeBuffer(buffer, channel_state->index_buffer.cpu_addr, size);
     }
     if constexpr (HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) {
         const u32 new_offset =
@@ -706,7 +715,7 @@ template <class P>
 void BufferCache<P>::BindHostVertexBuffers() {
     auto& flags = maxwell3d->dirty.flags;
     for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) {
-        const Binding& binding = vertex_buffers[index];
+        const Binding& binding = channel_state->vertex_buffers[index];
         Buffer& buffer = slot_buffers[binding.buffer_id];
         TouchBuffer(buffer, binding.buffer_id);
         SynchronizeBuffer(buffer, binding.cpu_addr, binding.size);
@@ -729,19 +738,19 @@ void BufferCache<P>::BindHostDrawIndirectBuffers() {
         SynchronizeBuffer(buffer, binding.cpu_addr, binding.size);
     };
     if (current_draw_indirect->include_count) {
-        bind_buffer(count_buffer_binding);
+        bind_buffer(channel_state->count_buffer_binding);
     }
-    bind_buffer(indirect_buffer_binding);
+    bind_buffer(channel_state->indirect_buffer_binding);
 }
 
 template <class P>
 void BufferCache<P>::BindHostGraphicsUniformBuffers(size_t stage) {
     u32 dirty = ~0U;
     if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
-        dirty = std::exchange(dirty_uniform_buffers[stage], 0);
+        dirty = std::exchange(channel_state->dirty_uniform_buffers[stage], 0);
     }
     u32 binding_index = 0;
-    ForEachEnabledBit(enabled_uniform_buffer_masks[stage], [&](u32 index) {
+    ForEachEnabledBit(channel_state->enabled_uniform_buffer_masks[stage], [&](u32 index) {
         const bool needs_bind = ((dirty >> index) & 1) != 0;
         BindHostGraphicsUniformBuffer(stage, index, binding_index, needs_bind);
         if constexpr (NEEDS_BIND_UNIFORM_INDEX) {
@@ -753,13 +762,13 @@ void BufferCache<P>::BindHostGraphicsUniformBuffers(size_t stage) {
 template <class P>
 void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 binding_index,
                                                    bool needs_bind) {
-    const Binding& binding = uniform_buffers[stage][index];
+    const Binding& binding = channel_state->uniform_buffers[stage][index];
     const VAddr cpu_addr = binding.cpu_addr;
-    const u32 size = std::min(binding.size, (*uniform_buffer_sizes)[stage][index]);
+    const u32 size = std::min(binding.size, (*channel_state->uniform_buffer_sizes)[stage][index]);
     Buffer& buffer = slot_buffers[binding.buffer_id];
     TouchBuffer(buffer, binding.buffer_id);
     const bool use_fast_buffer = binding.buffer_id != NULL_BUFFER_ID &&
-                                 size <= uniform_buffer_skip_cache_size &&
+                                 size <= channel_state->uniform_buffer_skip_cache_size &&
                                  !memory_tracker.IsRegionGpuModified(cpu_addr, size);
     if (use_fast_buffer) {
         if constexpr (IS_OPENGL) {
@@ -767,11 +776,11 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
                 // Fast path for Nvidia
                 const bool should_fast_bind =
                     !HasFastUniformBufferBound(stage, binding_index) ||
-                    uniform_buffer_binding_sizes[stage][binding_index] != size;
+                    channel_state->uniform_buffer_binding_sizes[stage][binding_index] != size;
                 if (should_fast_bind) {
                     // We only have to bind when the currently bound buffer is not the fast version
-                    fast_bound_uniform_buffers[stage] |= 1U << binding_index;
-                    uniform_buffer_binding_sizes[stage][binding_index] = size;
+                    channel_state->fast_bound_uniform_buffers[stage] |= 1U << binding_index;
+                    channel_state->uniform_buffer_binding_sizes[stage][binding_index] = size;
                     runtime.BindFastUniformBuffer(stage, binding_index, size);
                 }
                 const auto span = ImmediateBufferWithData(cpu_addr, size);
@@ -780,8 +789,8 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
             }
         }
         if constexpr (IS_OPENGL) {
-            fast_bound_uniform_buffers[stage] |= 1U << binding_index;
-            uniform_buffer_binding_sizes[stage][binding_index] = size;
+            channel_state->fast_bound_uniform_buffers[stage] |= 1U << binding_index;
+            channel_state->uniform_buffer_binding_sizes[stage][binding_index] = size;
         }
         // Stream buffer path to avoid stalling on non-Nvidia drivers or Vulkan
         const std::span<u8> span = runtime.BindMappedUniformBuffer(stage, binding_index, size);
@@ -791,15 +800,15 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
     // Classic cached path
     const bool sync_cached = SynchronizeBuffer(buffer, cpu_addr, size);
     if (sync_cached) {
-        ++uniform_cache_hits[0];
+        ++channel_state->uniform_cache_hits[0];
     }
-    ++uniform_cache_shots[0];
+    ++channel_state->uniform_cache_shots[0];
 
     // Skip binding if it's not needed and if the bound buffer is not the fast version
     // This exists to avoid instances where the fast buffer is bound and a GPU write happens
     needs_bind |= HasFastUniformBufferBound(stage, binding_index);
     if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
-        needs_bind |= uniform_buffer_binding_sizes[stage][binding_index] != size;
+        needs_bind |= channel_state->uniform_buffer_binding_sizes[stage][binding_index] != size;
     }
     if (!needs_bind) {
         return;
@@ -807,14 +816,14 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
     const u32 offset = buffer.Offset(cpu_addr);
     if constexpr (IS_OPENGL) {
         // Fast buffer will be unbound
-        fast_bound_uniform_buffers[stage] &= ~(1U << binding_index);
+        channel_state->fast_bound_uniform_buffers[stage] &= ~(1U << binding_index);
 
         // Mark the index as dirty if offset doesn't match
         const bool is_copy_bind = offset != 0 && !runtime.SupportsNonZeroUniformOffset();
-        dirty_uniform_buffers[stage] |= (is_copy_bind ? 1U : 0U) << index;
+        channel_state->dirty_uniform_buffers[stage] |= (is_copy_bind ? 1U : 0U) << index;
     }
     if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
-        uniform_buffer_binding_sizes[stage][binding_index] = size;
+        channel_state->uniform_buffer_binding_sizes[stage][binding_index] = size;
     }
     if constexpr (NEEDS_BIND_UNIFORM_INDEX) {
         runtime.BindUniformBuffer(stage, binding_index, buffer, offset, size);
@@ -826,15 +835,15 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
 template <class P>
 void BufferCache<P>::BindHostGraphicsStorageBuffers(size_t stage) {
     u32 binding_index = 0;
-    ForEachEnabledBit(enabled_storage_buffers[stage], [&](u32 index) {
-        const Binding& binding = storage_buffers[stage][index];
+    ForEachEnabledBit(channel_state->enabled_storage_buffers[stage], [&](u32 index) {
+        const Binding& binding = channel_state->storage_buffers[stage][index];
         Buffer& buffer = slot_buffers[binding.buffer_id];
         TouchBuffer(buffer, binding.buffer_id);
         const u32 size = binding.size;
         SynchronizeBuffer(buffer, binding.cpu_addr, size);
 
         const u32 offset = buffer.Offset(binding.cpu_addr);
-        const bool is_written = ((written_storage_buffers[stage] >> index) & 1) != 0;
+        const bool is_written = ((channel_state->written_storage_buffers[stage] >> index) & 1) != 0;
         if constexpr (NEEDS_BIND_STORAGE_INDEX) {
             runtime.BindStorageBuffer(stage, binding_index, buffer, offset, size, is_written);
             ++binding_index;
@@ -846,8 +855,8 @@ void BufferCache<P>::BindHostGraphicsStorageBuffers(size_t stage) {
 
 template <class P>
 void BufferCache<P>::BindHostGraphicsTextureBuffers(size_t stage) {
-    ForEachEnabledBit(enabled_texture_buffers[stage], [&](u32 index) {
-        const TextureBufferBinding& binding = texture_buffers[stage][index];
+    ForEachEnabledBit(channel_state->enabled_texture_buffers[stage], [&](u32 index) {
+        const TextureBufferBinding& binding = channel_state->texture_buffers[stage][index];
         Buffer& buffer = slot_buffers[binding.buffer_id];
         const u32 size = binding.size;
         SynchronizeBuffer(buffer, binding.cpu_addr, size);
@@ -855,7 +864,7 @@ void BufferCache<P>::BindHostGraphicsTextureBuffers(size_t stage) {
         const u32 offset = buffer.Offset(binding.cpu_addr);
         const PixelFormat format = binding.format;
         if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) {
-            if (((image_texture_buffers[stage] >> index) & 1) != 0) {
+            if (((channel_state->image_texture_buffers[stage] >> index) & 1) != 0) {
                 runtime.BindImageBuffer(buffer, offset, size, format);
             } else {
                 runtime.BindTextureBuffer(buffer, offset, size, format);
@@ -872,7 +881,7 @@ void BufferCache<P>::BindHostTransformFeedbackBuffers() {
         return;
     }
     for (u32 index = 0; index < NUM_TRANSFORM_FEEDBACK_BUFFERS; ++index) {
-        const Binding& binding = transform_feedback_buffers[index];
+        const Binding& binding = channel_state->transform_feedback_buffers[index];
         Buffer& buffer = slot_buffers[binding.buffer_id];
         TouchBuffer(buffer, binding.buffer_id);
         const u32 size = binding.size;
@@ -887,15 +896,16 @@ template <class P>
 void BufferCache<P>::BindHostComputeUniformBuffers() {
     if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
         // Mark all uniform buffers as dirty
-        dirty_uniform_buffers.fill(~u32{0});
-        fast_bound_uniform_buffers.fill(0);
+        channel_state->dirty_uniform_buffers.fill(~u32{0});
+        channel_state->fast_bound_uniform_buffers.fill(0);
     }
     u32 binding_index = 0;
-    ForEachEnabledBit(enabled_compute_uniform_buffer_mask, [&](u32 index) {
-        const Binding& binding = compute_uniform_buffers[index];
+    ForEachEnabledBit(channel_state->enabled_compute_uniform_buffer_mask, [&](u32 index) {
+        const Binding& binding = channel_state->compute_uniform_buffers[index];
         Buffer& buffer = slot_buffers[binding.buffer_id];
         TouchBuffer(buffer, binding.buffer_id);
-        const u32 size = std::min(binding.size, (*compute_uniform_buffer_sizes)[index]);
+        const u32 size =
+            std::min(binding.size, (*channel_state->compute_uniform_buffer_sizes)[index]);
         SynchronizeBuffer(buffer, binding.cpu_addr, size);
 
         const u32 offset = buffer.Offset(binding.cpu_addr);
@@ -911,15 +921,16 @@ void BufferCache<P>::BindHostComputeUniformBuffers() {
 template <class P>
 void BufferCache<P>::BindHostComputeStorageBuffers() {
     u32 binding_index = 0;
-    ForEachEnabledBit(enabled_compute_storage_buffers, [&](u32 index) {
-        const Binding& binding = compute_storage_buffers[index];
+    ForEachEnabledBit(channel_state->enabled_compute_storage_buffers, [&](u32 index) {
+        const Binding& binding = channel_state->compute_storage_buffers[index];
         Buffer& buffer = slot_buffers[binding.buffer_id];
         TouchBuffer(buffer, binding.buffer_id);
         const u32 size = binding.size;
         SynchronizeBuffer(buffer, binding.cpu_addr, size);
 
         const u32 offset = buffer.Offset(binding.cpu_addr);
-        const bool is_written = ((written_compute_storage_buffers >> index) & 1) != 0;
+        const bool is_written =
+            ((channel_state->written_compute_storage_buffers >> index) & 1) != 0;
         if constexpr (NEEDS_BIND_STORAGE_INDEX) {
             runtime.BindComputeStorageBuffer(binding_index, buffer, offset, size, is_written);
             ++binding_index;
@@ -931,8 +942,8 @@ void BufferCache<P>::BindHostComputeStorageBuffers() {
 
 template <class P>
 void BufferCache<P>::BindHostComputeTextureBuffers() {
-    ForEachEnabledBit(enabled_compute_texture_buffers, [&](u32 index) {
-        const TextureBufferBinding& binding = compute_texture_buffers[index];
+    ForEachEnabledBit(channel_state->enabled_compute_texture_buffers, [&](u32 index) {
+        const TextureBufferBinding& binding = channel_state->compute_texture_buffers[index];
         Buffer& buffer = slot_buffers[binding.buffer_id];
         const u32 size = binding.size;
         SynchronizeBuffer(buffer, binding.cpu_addr, size);
@@ -940,7 +951,7 @@ void BufferCache<P>::BindHostComputeTextureBuffers() {
         const u32 offset = buffer.Offset(binding.cpu_addr);
         const PixelFormat format = binding.format;
         if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) {
-            if (((image_compute_texture_buffers >> index) & 1) != 0) {
+            if (((channel_state->image_compute_texture_buffers >> index) & 1) != 0) {
                 runtime.BindImageBuffer(buffer, offset, size, format);
             } else {
                 runtime.BindTextureBuffer(buffer, offset, size, format);
@@ -954,7 +965,7 @@ void BufferCache<P>::BindHostComputeTextureBuffers() {
 template <class P>
 void BufferCache<P>::DoUpdateGraphicsBuffers(bool is_indexed) {
     do {
-        has_deleted_buffers = false;
+        channel_state->has_deleted_buffers = false;
         if (is_indexed) {
             UpdateIndexBuffer();
         }
@@ -968,7 +979,7 @@ void BufferCache<P>::DoUpdateGraphicsBuffers(bool is_indexed) {
         if (current_draw_indirect) {
             UpdateDrawIndirect();
         }
-    } while (has_deleted_buffers);
+    } while (channel_state->has_deleted_buffers);
 }
 
 template <class P>
@@ -999,7 +1010,7 @@ void BufferCache<P>::UpdateIndexBuffer() {
             slot_buffers.erase(inline_buffer_id);
             inline_buffer_id = CreateBuffer(0, buffer_size);
         }
-        index_buffer = Binding{
+        channel_state->index_buffer = Binding{
             .cpu_addr = 0,
             .size = inline_index_size,
             .buffer_id = inline_buffer_id,
@@ -1015,10 +1026,10 @@ void BufferCache<P>::UpdateIndexBuffer() {
         (index_buffer_ref.count + index_buffer_ref.first) * index_buffer_ref.FormatSizeInBytes();
     const u32 size = std::min(address_size, draw_size);
     if (size == 0 || !cpu_addr) {
-        index_buffer = NULL_BINDING;
+        channel_state->index_buffer = NULL_BINDING;
         return;
     }
-    index_buffer = Binding{
+    channel_state->index_buffer = Binding{
         .cpu_addr = *cpu_addr,
         .size = size,
         .buffer_id = FindBuffer(*cpu_addr, size),
@@ -1051,13 +1062,13 @@ void BufferCache<P>::UpdateVertexBuffer(u32 index) {
     const u32 address_size = static_cast<u32>(gpu_addr_end - gpu_addr_begin);
     u32 size = address_size; // TODO: Analyze stride and number of vertices
     if (array.enable == 0 || size == 0 || !cpu_addr) {
-        vertex_buffers[index] = NULL_BINDING;
+        channel_state->vertex_buffers[index] = NULL_BINDING;
         return;
     }
     if (!gpu_memory->IsWithinGPUAddressRange(gpu_addr_end)) {
         size = static_cast<u32>(gpu_memory->MaxContinuousRange(gpu_addr_begin, size));
     }
-    vertex_buffers[index] = Binding{
+    channel_state->vertex_buffers[index] = Binding{
         .cpu_addr = *cpu_addr,
         .size = size,
         .buffer_id = FindBuffer(*cpu_addr, size),
@@ -1079,23 +1090,24 @@ void BufferCache<P>::UpdateDrawIndirect() {
         };
     };
     if (current_draw_indirect->include_count) {
-        update(current_draw_indirect->count_start_address, sizeof(u32), count_buffer_binding);
+        update(current_draw_indirect->count_start_address, sizeof(u32),
+               channel_state->count_buffer_binding);
     }
     update(current_draw_indirect->indirect_start_address, current_draw_indirect->buffer_size,
-           indirect_buffer_binding);
+           channel_state->indirect_buffer_binding);
 }
 
 template <class P>
 void BufferCache<P>::UpdateUniformBuffers(size_t stage) {
-    ForEachEnabledBit(enabled_uniform_buffer_masks[stage], [&](u32 index) {
-        Binding& binding = uniform_buffers[stage][index];
+    ForEachEnabledBit(channel_state->enabled_uniform_buffer_masks[stage], [&](u32 index) {
+        Binding& binding = channel_state->uniform_buffers[stage][index];
         if (binding.buffer_id) {
             // Already updated
             return;
         }
         // Mark as dirty
         if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
-            dirty_uniform_buffers[stage] |= 1U << index;
+            channel_state->dirty_uniform_buffers[stage] |= 1U << index;
         }
         // Resolve buffer
         binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size);
@@ -1104,10 +1116,10 @@ void BufferCache<P>::UpdateUniformBuffers(size_t stage) {
 
 template <class P>
 void BufferCache<P>::UpdateStorageBuffers(size_t stage) {
-    const u32 written_mask = written_storage_buffers[stage];
-    ForEachEnabledBit(enabled_storage_buffers[stage], [&](u32 index) {
+    const u32 written_mask = channel_state->written_storage_buffers[stage];
+    ForEachEnabledBit(channel_state->enabled_storage_buffers[stage], [&](u32 index) {
         // Resolve buffer
-        Binding& binding = storage_buffers[stage][index];
+        Binding& binding = channel_state->storage_buffers[stage][index];
         const BufferId buffer_id = FindBuffer(binding.cpu_addr, binding.size);
         binding.buffer_id = buffer_id;
         // Mark buffer as written if needed
@@ -1119,11 +1131,11 @@ void BufferCache<P>::UpdateStorageBuffers(size_t stage) {
 
 template <class P>
 void BufferCache<P>::UpdateTextureBuffers(size_t stage) {
-    ForEachEnabledBit(enabled_texture_buffers[stage], [&](u32 index) {
-        Binding& binding = texture_buffers[stage][index];
+    ForEachEnabledBit(channel_state->enabled_texture_buffers[stage], [&](u32 index) {
+        Binding& binding = channel_state->texture_buffers[stage][index];
         binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size);
         // Mark buffer as written if needed
-        if (((written_texture_buffers[stage] >> index) & 1) != 0) {
+        if (((channel_state->written_texture_buffers[stage] >> index) & 1) != 0) {
             MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size);
         }
     });
@@ -1146,11 +1158,11 @@ void BufferCache<P>::UpdateTransformFeedbackBuffer(u32 index) {
     const u32 size = binding.size;
     const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
     if (binding.enable == 0 || size == 0 || !cpu_addr) {
-        transform_feedback_buffers[index] = NULL_BINDING;
+        channel_state->transform_feedback_buffers[index] = NULL_BINDING;
         return;
     }
     const BufferId buffer_id = FindBuffer(*cpu_addr, size);
-    transform_feedback_buffers[index] = Binding{
+    channel_state->transform_feedback_buffers[index] = Binding{
         .cpu_addr = *cpu_addr,
         .size = size,
         .buffer_id = buffer_id,
@@ -1160,8 +1172,8 @@ void BufferCache<P>::UpdateTransformFeedbackBuffer(u32 index) {
 
 template <class P>
 void BufferCache<P>::UpdateComputeUniformBuffers() {
-    ForEachEnabledBit(enabled_compute_uniform_buffer_mask, [&](u32 index) {
-        Binding& binding = compute_uniform_buffers[index];
+    ForEachEnabledBit(channel_state->enabled_compute_uniform_buffer_mask, [&](u32 index) {
+        Binding& binding = channel_state->compute_uniform_buffers[index];
         binding = NULL_BINDING;
         const auto& launch_desc = kepler_compute->launch_description;
         if (((launch_desc.const_buffer_enable_mask >> index) & 1) != 0) {
@@ -1178,12 +1190,12 @@ void BufferCache<P>::UpdateComputeUniformBuffers() {
 
 template <class P>
 void BufferCache<P>::UpdateComputeStorageBuffers() {
-    ForEachEnabledBit(enabled_compute_storage_buffers, [&](u32 index) {
+    ForEachEnabledBit(channel_state->enabled_compute_storage_buffers, [&](u32 index) {
         // Resolve buffer
-        Binding& binding = compute_storage_buffers[index];
+        Binding& binding = channel_state->compute_storage_buffers[index];
         binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size);
         // Mark as written if needed
-        if (((written_compute_storage_buffers >> index) & 1) != 0) {
+        if (((channel_state->written_compute_storage_buffers >> index) & 1) != 0) {
             MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size);
         }
     });
@@ -1191,11 +1203,11 @@ void BufferCache<P>::UpdateComputeStorageBuffers() {
 
 template <class P>
 void BufferCache<P>::UpdateComputeTextureBuffers() {
-    ForEachEnabledBit(enabled_compute_texture_buffers, [&](u32 index) {
-        Binding& binding = compute_texture_buffers[index];
+    ForEachEnabledBit(channel_state->enabled_compute_texture_buffers, [&](u32 index) {
+        Binding& binding = channel_state->compute_texture_buffers[index];
         binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size);
         // Mark as written if needed
-        if (((written_compute_texture_buffers >> index) & 1) != 0) {
+        if (((channel_state->written_compute_texture_buffers >> index) & 1) != 0) {
             MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size);
         }
     });
@@ -1610,13 +1622,13 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id, bool do_not_mark) {
     const auto replace = [scalar_replace](std::span<Binding> bindings) {
         std::ranges::for_each(bindings, scalar_replace);
     };
-    scalar_replace(index_buffer);
-    replace(vertex_buffers);
-    std::ranges::for_each(uniform_buffers, replace);
-    std::ranges::for_each(storage_buffers, replace);
-    replace(transform_feedback_buffers);
-    replace(compute_uniform_buffers);
-    replace(compute_storage_buffers);
+    scalar_replace(channel_state->index_buffer);
+    replace(channel_state->vertex_buffers);
+    std::ranges::for_each(channel_state->uniform_buffers, replace);
+    std::ranges::for_each(channel_state->storage_buffers, replace);
+    replace(channel_state->transform_feedback_buffers);
+    replace(channel_state->compute_uniform_buffers);
+    replace(channel_state->compute_storage_buffers);
 
     // Mark the whole buffer as CPU written to stop tracking CPU writes
     if (!do_not_mark) {
@@ -1634,8 +1646,8 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id, bool do_not_mark) {
 template <class P>
 void BufferCache<P>::NotifyBufferDeletion() {
     if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
-        dirty_uniform_buffers.fill(~u32{0});
-        uniform_buffer_binding_sizes.fill({});
+        channel_state->dirty_uniform_buffers.fill(~u32{0});
+        channel_state->uniform_buffer_binding_sizes.fill({});
     }
     auto& flags = maxwell3d->dirty.flags;
     flags[Dirty::IndexBuffer] = true;
@@ -1643,13 +1655,12 @@ void BufferCache<P>::NotifyBufferDeletion() {
     for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) {
         flags[Dirty::VertexBuffer0 + index] = true;
     }
-    has_deleted_buffers = true;
+    channel_state->has_deleted_buffers = true;
 }
 
 template <class P>
-typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr,
-                                                                      u32 cbuf_index,
-                                                                      bool is_written) const {
+Binding BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index,
+                                             bool is_written) const {
     const GPUVAddr gpu_addr = gpu_memory->Read<u64>(ssbo_addr);
     const auto size = [&]() {
         const bool is_nvn_cbuf = cbuf_index == 0;
@@ -1664,7 +1675,7 @@ typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr s
         // cbufs, which do not store the sizes adjacent to the addresses, so use the fully
         // mapped buffer size for now.
         const u32 memory_layout_size = static_cast<u32>(gpu_memory->GetMemoryLayoutSize(gpu_addr));
-        return memory_layout_size;
+        return std::min(memory_layout_size, static_cast<u32>(8_MiB));
     }();
     const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
     if (!cpu_addr || size == 0) {
@@ -1681,8 +1692,8 @@ typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr s
 }
 
 template <class P>
-typename BufferCache<P>::TextureBufferBinding BufferCache<P>::GetTextureBufferBinding(
-    GPUVAddr gpu_addr, u32 size, PixelFormat format) {
+TextureBufferBinding BufferCache<P>::GetTextureBufferBinding(GPUVAddr gpu_addr, u32 size,
+                                                             PixelFormat format) {
     const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
     TextureBufferBinding binding;
     if (!cpu_addr || size == 0) {
@@ -1721,7 +1732,7 @@ std::span<u8> BufferCache<P>::ImmediateBuffer(size_t wanted_capacity) {
 template <class P>
 bool BufferCache<P>::HasFastUniformBufferBound(size_t stage, u32 binding_index) const noexcept {
     if constexpr (IS_OPENGL) {
-        return ((fast_bound_uniform_buffers[stage] >> binding_index) & 1) != 0;
+        return ((channel_state->fast_bound_uniform_buffers[stage] >> binding_index) & 1) != 0;
     } else {
         // Only OpenGL has fast uniform buffers
         return false;
@@ -1730,14 +1741,14 @@ bool BufferCache<P>::HasFastUniformBufferBound(size_t stage, u32 binding_index)
 
 template <class P>
 std::pair<typename BufferCache<P>::Buffer*, u32> BufferCache<P>::GetDrawIndirectCount() {
-    auto& buffer = slot_buffers[count_buffer_binding.buffer_id];
-    return std::make_pair(&buffer, buffer.Offset(count_buffer_binding.cpu_addr));
+    auto& buffer = slot_buffers[channel_state->count_buffer_binding.buffer_id];
+    return std::make_pair(&buffer, buffer.Offset(channel_state->count_buffer_binding.cpu_addr));
 }
 
 template <class P>
 std::pair<typename BufferCache<P>::Buffer*, u32> BufferCache<P>::GetDrawIndirectBuffer() {
-    auto& buffer = slot_buffers[indirect_buffer_binding.buffer_id];
-    return std::make_pair(&buffer, buffer.Offset(indirect_buffer_binding.cpu_addr));
+    auto& buffer = slot_buffers[channel_state->indirect_buffer_binding.buffer_id];
+    return std::make_pair(&buffer, buffer.Offset(channel_state->indirect_buffer_binding.cpu_addr));
 }
 
 } // namespace VideoCommon

src/video_core/buffer_cache/buffer_cache_base.h

@@ -86,8 +86,78 @@ enum class ObtainBufferOperation : u32 {
     MarkQuery = 3,
 };
 
-template <typename P>
-class BufferCache : public VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> {
+static constexpr BufferId NULL_BUFFER_ID{0};
+static constexpr u32 DEFAULT_SKIP_CACHE_SIZE = static_cast<u32>(4_KiB);
+
+struct Binding {
+    VAddr cpu_addr{};
+    u32 size{};
+    BufferId buffer_id;
+};
+
+struct TextureBufferBinding : Binding {
+    PixelFormat format;
+};
+
+static constexpr Binding NULL_BINDING{
+    .cpu_addr = 0,
+    .size = 0,
+    .buffer_id = NULL_BUFFER_ID,
+};
+
+class BufferCacheChannelInfo : public ChannelInfo {
+public:
+    BufferCacheChannelInfo() = delete;
+    BufferCacheChannelInfo(Tegra::Control::ChannelState& state) noexcept : ChannelInfo(state) {}
+    BufferCacheChannelInfo(const BufferCacheChannelInfo& state) = delete;
+    BufferCacheChannelInfo& operator=(const BufferCacheChannelInfo&) = delete;
+
+    Binding index_buffer;
+    std::array<Binding, NUM_VERTEX_BUFFERS> vertex_buffers;
+    std::array<std::array<Binding, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES> uniform_buffers;
+    std::array<std::array<Binding, NUM_STORAGE_BUFFERS>, NUM_STAGES> storage_buffers;
+    std::array<std::array<TextureBufferBinding, NUM_TEXTURE_BUFFERS>, NUM_STAGES> texture_buffers;
+    std::array<Binding, NUM_TRANSFORM_FEEDBACK_BUFFERS> transform_feedback_buffers;
+    Binding count_buffer_binding;
+    Binding indirect_buffer_binding;
+
+    std::array<Binding, NUM_COMPUTE_UNIFORM_BUFFERS> compute_uniform_buffers;
+    std::array<Binding, NUM_STORAGE_BUFFERS> compute_storage_buffers;
+    std::array<TextureBufferBinding, NUM_TEXTURE_BUFFERS> compute_texture_buffers;
+
+    std::array<u32, NUM_STAGES> enabled_uniform_buffer_masks{};
+    u32 enabled_compute_uniform_buffer_mask = 0;
+
+    const UniformBufferSizes* uniform_buffer_sizes{};
+    const ComputeUniformBufferSizes* compute_uniform_buffer_sizes{};
+
+    std::array<u32, NUM_STAGES> enabled_storage_buffers{};
+    std::array<u32, NUM_STAGES> written_storage_buffers{};
+    u32 enabled_compute_storage_buffers = 0;
+    u32 written_compute_storage_buffers = 0;
+
+    std::array<u32, NUM_STAGES> enabled_texture_buffers{};
+    std::array<u32, NUM_STAGES> written_texture_buffers{};
+    std::array<u32, NUM_STAGES> image_texture_buffers{};
+    u32 enabled_compute_texture_buffers = 0;
+    u32 written_compute_texture_buffers = 0;
+    u32 image_compute_texture_buffers = 0;
+
+    std::array<u32, 16> uniform_cache_hits{};
+    std::array<u32, 16> uniform_cache_shots{};
+
+    u32 uniform_buffer_skip_cache_size = DEFAULT_SKIP_CACHE_SIZE;
+
+    bool has_deleted_buffers = false;
+
+    std::array<u32, NUM_STAGES> dirty_uniform_buffers{};
+    std::array<u32, NUM_STAGES> fast_bound_uniform_buffers{};
+    std::array<std::array<u32, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES>
+        uniform_buffer_binding_sizes{};
+};
+
+template <class P>
+class BufferCache : public VideoCommon::ChannelSetupCaches<BufferCacheChannelInfo> {
     // Page size for caching purposes.
     // This is unrelated to the CPU page size and it can be changed as it seems optimal.
     static constexpr u32 CACHING_PAGEBITS = 16;
@@ -104,8 +174,6 @@ class BufferCache : public VideoCommon::ChannelSetupCaches<VideoCommon::ChannelI
     static constexpr bool SEPARATE_IMAGE_BUFFERS_BINDINGS = P::SEPARATE_IMAGE_BUFFER_BINDINGS;
     static constexpr bool IMPLEMENTS_ASYNC_DOWNLOADS = P::IMPLEMENTS_ASYNC_DOWNLOADS;
 
-    static constexpr BufferId NULL_BUFFER_ID{0};
-
     static constexpr s64 DEFAULT_EXPECTED_MEMORY = 512_MiB;
     static constexpr s64 DEFAULT_CRITICAL_MEMORY = 1_GiB;
     static constexpr s64 TARGET_THRESHOLD = 4_GiB;
@@ -149,8 +217,6 @@ class BufferCache : public VideoCommon::ChannelSetupCaches<VideoCommon::ChannelI
     using OverlapSection = boost::icl::inter_section<int>;
     using OverlapCounter = boost::icl::split_interval_map<VAddr, int>;
 
-    struct Empty {};
-
     struct OverlapResult {
         std::vector<BufferId> ids;
         VAddr begin;
@@ -158,25 +224,7 @@ class BufferCache : public VideoCommon::ChannelSetupCaches<VideoCommon::ChannelI
         bool has_stream_leap = false;
     };
 
-    struct Binding {
-        VAddr cpu_addr{};
-        u32 size{};
-        BufferId buffer_id;
-    };
-
-    struct TextureBufferBinding : Binding {
-        PixelFormat format;
-    };
-
-    static constexpr Binding NULL_BINDING{
-        .cpu_addr = 0,
-        .size = 0,
-        .buffer_id = NULL_BUFFER_ID,
-    };
-
 public:
-    static constexpr u32 DEFAULT_SKIP_CACHE_SIZE = static_cast<u32>(4_KiB);
-
     explicit BufferCache(VideoCore::RasterizerInterface& rasterizer_,
                          Core::Memory::Memory& cpu_memory_, Runtime& runtime_);
 
@@ -496,51 +544,6 @@ private:
 
     u32 last_index_count = 0;
 
-    Binding index_buffer;
-    std::array<Binding, NUM_VERTEX_BUFFERS> vertex_buffers;
-    std::array<std::array<Binding, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES> uniform_buffers;
-    std::array<std::array<Binding, NUM_STORAGE_BUFFERS>, NUM_STAGES> storage_buffers;
-    std::array<std::array<TextureBufferBinding, NUM_TEXTURE_BUFFERS>, NUM_STAGES> texture_buffers;
-    std::array<Binding, NUM_TRANSFORM_FEEDBACK_BUFFERS> transform_feedback_buffers;
-    Binding count_buffer_binding;
-    Binding indirect_buffer_binding;
-
-    std::array<Binding, NUM_COMPUTE_UNIFORM_BUFFERS> compute_uniform_buffers;
-    std::array<Binding, NUM_STORAGE_BUFFERS> compute_storage_buffers;
-    std::array<TextureBufferBinding, NUM_TEXTURE_BUFFERS> compute_texture_buffers;
-
-    std::array<u32, NUM_STAGES> enabled_uniform_buffer_masks{};
-    u32 enabled_compute_uniform_buffer_mask = 0;
-
-    const UniformBufferSizes* uniform_buffer_sizes{};
-    const ComputeUniformBufferSizes* compute_uniform_buffer_sizes{};
-
-    std::array<u32, NUM_STAGES> enabled_storage_buffers{};
-    std::array<u32, NUM_STAGES> written_storage_buffers{};
-    u32 enabled_compute_storage_buffers = 0;
-    u32 written_compute_storage_buffers = 0;
-
-    std::array<u32, NUM_STAGES> enabled_texture_buffers{};
-    std::array<u32, NUM_STAGES> written_texture_buffers{};
-    std::array<u32, NUM_STAGES> image_texture_buffers{};
-    u32 enabled_compute_texture_buffers = 0;
-    u32 written_compute_texture_buffers = 0;
-    u32 image_compute_texture_buffers = 0;
-
-    std::array<u32, 16> uniform_cache_hits{};
-    std::array<u32, 16> uniform_cache_shots{};
-
-    u32 uniform_buffer_skip_cache_size = DEFAULT_SKIP_CACHE_SIZE;
-
-    bool has_deleted_buffers = false;
-
-    std::conditional_t<HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS, std::array<u32, NUM_STAGES>, Empty>
-        dirty_uniform_buffers{};
-    std::conditional_t<IS_OPENGL, std::array<u32, NUM_STAGES>, Empty> fast_bound_uniform_buffers{};
-    std::conditional_t<HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS,
-                       std::array<std::array<u32, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES>, Empty>
-        uniform_buffer_binding_sizes{};
-
     MemoryTracker memory_tracker;
     IntervalSet uncommitted_ranges;
     IntervalSet common_ranges;

src/video_core/renderer_opengl/gl_buffer_cache.cpp

@@ -117,7 +117,7 @@ BufferCacheRuntime::BufferCacheRuntime(const Device& device_)
     for (auto& stage_uniforms : fast_uniforms) {
         for (OGLBuffer& buffer : stage_uniforms) {
             buffer.Create();
-            glNamedBufferData(buffer.handle, BufferCache::DEFAULT_SKIP_CACHE_SIZE, nullptr,
+            glNamedBufferData(buffer.handle, VideoCommon::DEFAULT_SKIP_CACHE_SIZE, nullptr,
                               GL_STREAM_DRAW);
         }
     }

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -233,6 +233,8 @@ void ApplySwizzle(GLuint handle, PixelFormat format, std::array<SwizzleSource, 4
                                     const VideoCommon::ImageInfo& info) {
     if (IsPixelFormatASTC(info.format) && info.size.depth == 1 && !runtime.HasNativeASTC()) {
         return Settings::values.accelerate_astc.GetValue() &&
+               Settings::values.astc_recompression.GetValue() ==
+                   Settings::AstcRecompression::Uncompressed &&
                !Settings::values.async_astc.GetValue();
     }
     // Disable other accelerated uploads for now as they don't implement swizzled uploads
@@ -437,6 +439,24 @@ OGLTexture MakeImage(const VideoCommon::ImageInfo& info, GLenum gl_internal_form
     return GL_R32UI;
 }
 
+[[nodiscard]] bool IsAstcRecompressionEnabled() {
+    return Settings::values.astc_recompression.GetValue() !=
+           Settings::AstcRecompression::Uncompressed;
+}
+
+[[nodiscard]] GLenum SelectAstcFormat(PixelFormat format, bool is_srgb) {
+    switch (Settings::values.astc_recompression.GetValue()) {
+    case Settings::AstcRecompression::Bc1:
+        return is_srgb ? GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT : GL_COMPRESSED_RGBA_S3TC_DXT1_EXT;
+        break;
+    case Settings::AstcRecompression::Bc3:
+        return is_srgb ? GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT : GL_COMPRESSED_RGBA_S3TC_DXT5_EXT;
+        break;
+    default:
+        return is_srgb ? GL_SRGB8_ALPHA8 : GL_RGBA8;
+    }
+}
+
 } // Anonymous namespace
 
 ImageBufferMap::~ImageBufferMap() {
@@ -739,9 +759,16 @@ Image::Image(TextureCacheRuntime& runtime_, const VideoCommon::ImageInfo& info_,
     if (IsConverted(runtime->device, info.format, info.type)) {
         flags |= ImageFlagBits::Converted;
         flags |= ImageFlagBits::CostlyLoad;
-        gl_internal_format = IsPixelFormatSRGB(info.format) ? GL_SRGB8_ALPHA8 : GL_RGBA8;
+
+        const bool is_srgb = IsPixelFormatSRGB(info.format);
+        gl_internal_format = is_srgb ? GL_SRGB8_ALPHA8 : GL_RGBA8;
         gl_format = GL_RGBA;
         gl_type = GL_UNSIGNED_INT_8_8_8_8_REV;
+
+        if (IsPixelFormatASTC(info.format) && IsAstcRecompressionEnabled()) {
+            gl_internal_format = SelectAstcFormat(info.format, is_srgb);
+            gl_format = GL_NONE;
+        }
     } else {
         const auto& tuple = MaxwellToGL::GetFormatTuple(info.format);
         gl_internal_format = tuple.internal_format;
@@ -1130,7 +1157,12 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
       views{runtime.null_image_views} {
     const Device& device = runtime.device;
     if (True(image.flags & ImageFlagBits::Converted)) {
-        internal_format = IsPixelFormatSRGB(info.format) ? GL_SRGB8_ALPHA8 : GL_RGBA8;
+        const bool is_srgb = IsPixelFormatSRGB(info.format);
+        internal_format = is_srgb ? GL_SRGB8_ALPHA8 : GL_RGBA8;
+
+        if (IsPixelFormatASTC(info.format) && IsAstcRecompressionEnabled()) {
+            internal_format = SelectAstcFormat(info.format, is_srgb);
+        }
     } else {
         internal_format = MaxwellToGL::GetFormatTuple(format).internal_format;
     }

src/video_core/renderer_opengl/gl_texture_cache.h

@@ -144,6 +144,10 @@ public:
         return state_tracker;
     }
 
+    void BarrierFeedbackLoop() const noexcept {
+        // OpenGL does not require a barrier for attachment feedback loops.
+    }
+
 private:
     struct StagingBuffers {
         explicit StagingBuffers(GLenum storage_flags_, GLenum map_flags_);

src/video_core/renderer_vulkan/maxwell_to_vk.cpp

@@ -6,6 +6,7 @@
 #include "common/assert.h"
 #include "common/common_types.h"
 #include "common/logging/log.h"
+#include "common/settings.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_vulkan/maxwell_to_vk.h"
 #include "video_core/surface.h"
@@ -237,14 +238,25 @@ FormatInfo SurfaceFormat(const Device& device, FormatType format_type, bool with
                          PixelFormat pixel_format) {
     ASSERT(static_cast<size_t>(pixel_format) < std::size(tex_format_tuples));
     FormatTuple tuple = tex_format_tuples[static_cast<size_t>(pixel_format)];
-    // Use A8B8G8R8_UNORM on hardware that doesn't support ASTC natively
+    // Transcode on hardware that doesn't support ASTC natively
     if (!device.IsOptimalAstcSupported() && VideoCore::Surface::IsPixelFormatASTC(pixel_format)) {
         const bool is_srgb = with_srgb && VideoCore::Surface::IsPixelFormatSRGB(pixel_format);
-        if (is_srgb) {
-            tuple.format = VK_FORMAT_A8B8G8R8_SRGB_PACK32;
-        } else {
-            tuple.format = VK_FORMAT_A8B8G8R8_UNORM_PACK32;
-            tuple.usage |= Storage;
+
+        switch (Settings::values.astc_recompression.GetValue()) {
+        case Settings::AstcRecompression::Uncompressed:
+            if (is_srgb) {
+                tuple.format = VK_FORMAT_A8B8G8R8_SRGB_PACK32;
+            } else {
+                tuple.format = VK_FORMAT_A8B8G8R8_UNORM_PACK32;
+                tuple.usage |= Storage;
+            }
+            break;
+        case Settings::AstcRecompression::Bc1:
+            tuple.format = is_srgb ? VK_FORMAT_BC1_RGBA_SRGB_BLOCK : VK_FORMAT_BC1_RGBA_UNORM_BLOCK;
+            break;
+        case Settings::AstcRecompression::Bc3:
+            tuple.format = is_srgb ? VK_FORMAT_BC3_SRGB_BLOCK : VK_FORMAT_BC3_UNORM_BLOCK;
+            break;
         }
     }
     const bool attachable = (tuple.usage & Attachable) != 0;

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -481,12 +481,13 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) {
     if constexpr (Spec::enabled_stages[4]) {
         prepare_stage(4);
     }
+    texture_cache.UpdateRenderTargets(false);
+    texture_cache.CheckFeedbackLoop(views);
     ConfigureDraw(rescaling, render_area);
 }
 
 void GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling,
                                      const RenderAreaPushConstant& render_area) {
-    texture_cache.UpdateRenderTargets(false);
     scheduler.RequestRenderpass(texture_cache.GetFramebuffer());
 
     if (!is_built.load(std::memory_order::relaxed)) {

src/video_core/renderer_vulkan/vk_master_semaphore.cpp

@@ -10,11 +10,16 @@
 
 namespace Vulkan {
 
+constexpr u64 FENCE_RESERVE_SIZE = 8;
+
 MasterSemaphore::MasterSemaphore(const Device& device_) : device(device_) {
     if (!device.HasTimelineSemaphore()) {
         static constexpr VkFenceCreateInfo fence_ci{
             .sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO, .pNext = nullptr, .flags = 0};
-        fence = device.GetLogical().CreateFence(fence_ci);
+        free_queue.resize(FENCE_RESERVE_SIZE);
+        std::ranges::generate(free_queue,
+                              [&] { return device.GetLogical().CreateFence(fence_ci); });
+        wait_thread = std::jthread([this](std::stop_token token) { WaitThread(token); });
         return;
     }
 
@@ -167,16 +172,53 @@ VkResult MasterSemaphore::SubmitQueueFence(vk::CommandBuffer& cmdbuf, VkSemaphor
         .pSignalSemaphores = &signal_semaphore,
     };
 
+    auto fence = GetFreeFence();
     auto result = device.GetGraphicsQueue().Submit(submit_info, *fence);
 
     if (result == VK_SUCCESS) {
-        fence.Wait();
-        fence.Reset();
-        gpu_tick.store(host_tick);
-        gpu_tick.notify_all();
+        std::scoped_lock lock{wait_mutex};
+        wait_queue.emplace(host_tick, std::move(fence));
+        wait_cv.notify_one();
     }
 
     return result;
 }
 
+void MasterSemaphore::WaitThread(std::stop_token token) {
+    while (!token.stop_requested()) {
+        u64 host_tick;
+        vk::Fence fence;
+        {
+            std::unique_lock lock{wait_mutex};
+            Common::CondvarWait(wait_cv, lock, token, [this] { return !wait_queue.empty(); });
+            if (token.stop_requested()) {
+                return;
+            }
+            std::tie(host_tick, fence) = std::move(wait_queue.front());
+            wait_queue.pop();
+        }
+
+        fence.Wait();
+        fence.Reset();
+        gpu_tick.store(host_tick);
+        gpu_tick.notify_all();
+
+        std::scoped_lock lock{free_mutex};
+        free_queue.push_front(std::move(fence));
+    }
+}
+
+vk::Fence MasterSemaphore::GetFreeFence() {
+    std::scoped_lock lock{free_mutex};
+    if (free_queue.empty()) {
+        static constexpr VkFenceCreateInfo fence_ci{
+            .sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO, .pNext = nullptr, .flags = 0};
+        return device.GetLogical().CreateFence(fence_ci);
+    }
+
+    auto fence = std::move(free_queue.back());
+    free_queue.pop_back();
+    return fence;
+}
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_master_semaphore.h

@@ -5,8 +5,10 @@
 
 #include <atomic>
 #include <condition_variable>
+#include <deque>
 #include <mutex>
 #include <thread>
+#include <queue>
 
 #include "common/common_types.h"
 #include "common/polyfill_thread.h"
@@ -17,6 +19,8 @@ namespace Vulkan {
 class Device;
 
 class MasterSemaphore {
+    using Waitable = std::pair<u64, vk::Fence>;
+
 public:
     explicit MasterSemaphore(const Device& device);
     ~MasterSemaphore();
@@ -57,13 +61,22 @@ private:
     VkResult SubmitQueueFence(vk::CommandBuffer& cmdbuf, VkSemaphore signal_semaphore,
                               VkSemaphore wait_semaphore, u64 host_tick);
 
+    void WaitThread(std::stop_token token);
+
+    vk::Fence GetFreeFence();
+
 private:
     const Device& device;             ///< Device.
-    vk::Fence fence;                  ///< Fence.
     vk::Semaphore semaphore;          ///< Timeline semaphore.
     std::atomic<u64> gpu_tick{0};     ///< Current known GPU tick.
     std::atomic<u64> current_tick{1}; ///< Current logical tick.
+    std::mutex wait_mutex;
+    std::mutex free_mutex;
+    std::condition_variable_any wait_cv;
+    std::queue<Waitable> wait_queue;  ///< Queue for the fences to be waited on by the wait thread.
+    std::deque<vk::Fence> free_queue; ///< Holds available fences for submission.
     std::jthread debug_thread;        ///< Debug thread to workaround validation layer bugs.
+    std::jthread wait_thread;         ///< Helper thread that waits for submitted fences.
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -861,6 +861,10 @@ VkBuffer TextureCacheRuntime::GetTemporaryBuffer(size_t needed_size) {
     return *buffers[level];
 }
 
+void TextureCacheRuntime::BarrierFeedbackLoop() {
+    scheduler.RequestOutsideRenderPassOperationContext();
+}
+
 void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src,
                                            std::span<const VideoCommon::ImageCopy> copies) {
     std::vector<VkBufferImageCopy> vk_in_copies(copies.size());
@@ -1268,7 +1272,9 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu
     if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported()) {
         if (Settings::values.async_astc.GetValue()) {
             flags |= VideoCommon::ImageFlagBits::AsynchronousDecode;
-        } else if (Settings::values.accelerate_astc.GetValue() && info.size.depth == 1) {
+        } else if (Settings::values.astc_recompression.GetValue() ==
+                       Settings::AstcRecompression::Uncompressed &&
+                   Settings::values.accelerate_astc.GetValue() && info.size.depth == 1) {
             flags |= VideoCommon::ImageFlagBits::AcceleratedUpload;
         }
         flags |= VideoCommon::ImageFlagBits::Converted;
@@ -1283,7 +1289,9 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu
         .usage = VK_IMAGE_USAGE_STORAGE_BIT,
     };
     current_image = *original_image;
-    if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported()) {
+    if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported() &&
+        Settings::values.astc_recompression.GetValue() ==
+            Settings::AstcRecompression::Uncompressed) {
         const auto& device = runtime->device.GetLogical();
         storage_image_views.reserve(info.resources.levels);
         for (s32 level = 0; level < info.resources.levels; ++level) {

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -103,6 +103,8 @@ public:
 
     [[nodiscard]] VkBuffer GetTemporaryBuffer(size_t needed_size);
 
+    void BarrierFeedbackLoop();
+
     const Device& device;
     Scheduler& scheduler;
     MemoryAllocator& memory_allocator;

src/video_core/texture_cache/image_base.cpp

@@ -155,7 +155,7 @@ void ImageBase::CheckAliasState() {
     flags &= ~ImageFlagBits::Alias;
 }
 
-void AddImageAlias(ImageBase& lhs, ImageBase& rhs, ImageId lhs_id, ImageId rhs_id) {
+bool AddImageAlias(ImageBase& lhs, ImageBase& rhs, ImageId lhs_id, ImageId rhs_id) {
     static constexpr auto OPTIONS = RelaxedOptions::Size | RelaxedOptions::Format;
     ASSERT(lhs.info.type == rhs.info.type);
     std::optional<SubresourceBase> base;
@@ -169,7 +169,7 @@ void AddImageAlias(ImageBase& lhs, ImageBase& rhs, ImageId lhs_id, ImageId rhs_i
     }
     if (!base) {
         LOG_ERROR(HW_GPU, "Image alias should have been flipped");
-        return;
+        return false;
     }
     const PixelFormat lhs_format = lhs.info.format;
     const PixelFormat rhs_format = rhs.info.format;
@@ -248,12 +248,13 @@ void AddImageAlias(ImageBase& lhs, ImageBase& rhs, ImageId lhs_id, ImageId rhs_i
     }
     ASSERT(lhs_alias.copies.empty() == rhs_alias.copies.empty());
     if (lhs_alias.copies.empty()) {
-        return;
+        return false;
     }
     lhs.aliased_images.push_back(std::move(lhs_alias));
     rhs.aliased_images.push_back(std::move(rhs_alias));
     lhs.flags &= ~ImageFlagBits::IsRescalable;
     rhs.flags &= ~ImageFlagBits::IsRescalable;
+    return true;
 }
 
 } // namespace VideoCommon

src/video_core/texture_cache/image_base.h

@@ -142,6 +142,6 @@ struct ImageAllocBase {
     std::vector<ImageId> images;
 };
 
-void AddImageAlias(ImageBase& lhs, ImageBase& rhs, ImageId lhs_id, ImageId rhs_id);
+bool AddImageAlias(ImageBase& lhs, ImageBase& rhs, ImageId lhs_id, ImageId rhs_id);
 
 } // namespace VideoCommon

src/video_core/texture_cache/texture_cache.h

@@ -49,8 +49,8 @@ TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface&
 
     if constexpr (HAS_DEVICE_MEMORY_INFO) {
         const s64 device_memory = static_cast<s64>(runtime.GetDeviceLocalMemory());
-        const s64 min_spacing_expected = device_memory - 1_GiB - 512_MiB;
-        const s64 min_spacing_critical = device_memory - 1_GiB;
+        const s64 min_spacing_expected = device_memory - 1_GiB;
+        const s64 min_spacing_critical = device_memory - 512_MiB;
         const s64 mem_threshold = std::min(device_memory, TARGET_THRESHOLD);
         const s64 min_vacancy_expected = (6 * mem_threshold) / 10;
         const s64 min_vacancy_critical = (3 * mem_threshold) / 10;
@@ -86,10 +86,12 @@ void TextureCache<P>::RunGarbageCollector() {
             // used by the async decoder thread.
             return false;
         }
+        if (!aggressive_mode && True(image.flags & ImageFlagBits::CostlyLoad)) {
+            return false;
+        }
         const bool must_download =
             image.IsSafeDownload() && False(image.flags & ImageFlagBits::BadOverlap);
-        if (!high_priority_mode &&
-            (must_download || True(image.flags & ImageFlagBits::CostlyLoad))) {
+        if (!high_priority_mode && must_download) {
             return false;
         }
         if (must_download) {
@@ -137,7 +139,6 @@ void TextureCache<P>::TickFrame() {
     TickAsyncDecode();
 
     runtime.TickFrame();
-    critical_gc = 0;
     ++frame_tick;
 
     if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) {
@@ -183,6 +184,42 @@ void TextureCache<P>::FillComputeImageViews(std::span<ImageViewInOut> views) {
                          views);
 }
 
+template <class P>
+void TextureCache<P>::CheckFeedbackLoop(std::span<const ImageViewInOut> views) {
+    const bool requires_barrier = [&] {
+        for (const auto& view : views) {
+            if (!view.id) {
+                continue;
+            }
+            auto& image_view = slot_image_views[view.id];
+
+            // Check color targets
+            for (const auto& ct_view_id : render_targets.color_buffer_ids) {
+                if (ct_view_id) {
+                    auto& ct_view = slot_image_views[ct_view_id];
+                    if (image_view.image_id == ct_view.image_id) {
+                        return true;
+                    }
+                }
+            }
+
+            // Check zeta target
+            if (render_targets.depth_buffer_id) {
+                auto& zt_view = slot_image_views[render_targets.depth_buffer_id];
+                if (image_view.image_id == zt_view.image_id) {
+                    return true;
+                }
+            }
+        }
+
+        return false;
+    }();
+
+    if (requires_barrier) {
+        runtime.BarrierFeedbackLoop();
+    }
+}
+
 template <class P>
 typename P::Sampler* TextureCache<P>::GetGraphicsSampler(u32 index) {
     if (index > channel_state->graphics_sampler_table.Limit()) {
@@ -1274,17 +1311,18 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
     const size_t size_bytes = CalculateGuestSizeInBytes(new_info);
     const bool broken_views = runtime.HasBrokenTextureViewFormats();
     const bool native_bgr = runtime.HasNativeBgr();
-    boost::container::small_vector<ImageId, 4> overlap_ids;
-    std::unordered_set<ImageId> overlaps_found;
-    boost::container::small_vector<ImageId, 4> left_aliased_ids;
-    boost::container::small_vector<ImageId, 4> right_aliased_ids;
-    std::unordered_set<ImageId> ignore_textures;
-    boost::container::small_vector<ImageId, 4> bad_overlap_ids;
-    boost::container::small_vector<ImageId, 4> all_siblings;
+    join_overlap_ids.clear();
+    join_overlaps_found.clear();
+    join_left_aliased_ids.clear();
+    join_right_aliased_ids.clear();
+    join_ignore_textures.clear();
+    join_bad_overlap_ids.clear();
+    join_copies_to_do.clear();
+    join_alias_indices.clear();
     const bool this_is_linear = info.type == ImageType::Linear;
     const auto region_check = [&](ImageId overlap_id, ImageBase& overlap) {
         if (True(overlap.flags & ImageFlagBits::Remapped)) {
-            ignore_textures.insert(overlap_id);
+            join_ignore_textures.insert(overlap_id);
             return;
         }
         const bool overlap_is_linear = overlap.info.type == ImageType::Linear;
@@ -1294,11 +1332,11 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
         if (this_is_linear && overlap_is_linear) {
             if (info.pitch == overlap.info.pitch && gpu_addr == overlap.gpu_addr) {
                 // Alias linear images with the same pitch
-                left_aliased_ids.push_back(overlap_id);
+                join_left_aliased_ids.push_back(overlap_id);
             }
             return;
         }
-        overlaps_found.insert(overlap_id);
+        join_overlaps_found.insert(overlap_id);
         static constexpr bool strict_size = true;
         const std::optional<OverlapResult> solution = ResolveOverlap(
             new_info, gpu_addr, cpu_addr, overlap, strict_size, broken_views, native_bgr);
@@ -1306,33 +1344,33 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
             gpu_addr = solution->gpu_addr;
             cpu_addr = solution->cpu_addr;
             new_info.resources = solution->resources;
-            overlap_ids.push_back(overlap_id);
-            all_siblings.push_back(overlap_id);
+            join_overlap_ids.push_back(overlap_id);
+            join_copies_to_do.emplace_back(JoinCopy{false, overlap_id});
             return;
         }
         static constexpr auto options = RelaxedOptions::Size | RelaxedOptions::Format;
         const ImageBase new_image_base(new_info, gpu_addr, cpu_addr);
         if (IsSubresource(new_info, overlap, gpu_addr, options, broken_views, native_bgr)) {
-            left_aliased_ids.push_back(overlap_id);
+            join_left_aliased_ids.push_back(overlap_id);
             overlap.flags |= ImageFlagBits::Alias;
-            all_siblings.push_back(overlap_id);
+            join_copies_to_do.emplace_back(JoinCopy{true, overlap_id});
         } else if (IsSubresource(overlap.info, new_image_base, overlap.gpu_addr, options,
                                  broken_views, native_bgr)) {
-            right_aliased_ids.push_back(overlap_id);
+            join_right_aliased_ids.push_back(overlap_id);
             overlap.flags |= ImageFlagBits::Alias;
-            all_siblings.push_back(overlap_id);
+            join_copies_to_do.emplace_back(JoinCopy{true, overlap_id});
         } else {
-            bad_overlap_ids.push_back(overlap_id);
+            join_bad_overlap_ids.push_back(overlap_id);
         }
     };
     ForEachImageInRegion(cpu_addr, size_bytes, region_check);
     const auto region_check_gpu = [&](ImageId overlap_id, ImageBase& overlap) {
-        if (!overlaps_found.contains(overlap_id)) {
+        if (!join_overlaps_found.contains(overlap_id)) {
             if (True(overlap.flags & ImageFlagBits::Remapped)) {
-                ignore_textures.insert(overlap_id);
+                join_ignore_textures.insert(overlap_id);
             }
             if (overlap.gpu_addr == gpu_addr && overlap.guest_size_bytes == size_bytes) {
-                ignore_textures.insert(overlap_id);
+                join_ignore_textures.insert(overlap_id);
             }
         }
     };
@@ -1340,11 +1378,11 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
 
     bool can_rescale = info.rescaleable;
     bool any_rescaled = false;
-    for (const ImageId sibling_id : all_siblings) {
+    for (const auto& copy : join_copies_to_do) {
         if (!can_rescale) {
             break;
         }
-        Image& sibling = slot_images[sibling_id];
+        Image& sibling = slot_images[copy.id];
         can_rescale &= ImageCanRescale(sibling);
         any_rescaled |= True(sibling.flags & ImageFlagBits::Rescaled);
     }
@@ -1352,13 +1390,13 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
     can_rescale &= any_rescaled;
 
     if (can_rescale) {
-        for (const ImageId sibling_id : all_siblings) {
-            Image& sibling = slot_images[sibling_id];
+        for (const auto& copy : join_copies_to_do) {
+            Image& sibling = slot_images[copy.id];
             ScaleUp(sibling);
         }
     } else {
-        for (const ImageId sibling_id : all_siblings) {
-            Image& sibling = slot_images[sibling_id];
+        for (const auto& copy : join_copies_to_do) {
+            Image& sibling = slot_images[copy.id];
             ScaleDown(sibling);
         }
     }
@@ -1370,7 +1408,7 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
         new_image.flags |= ImageFlagBits::Sparse;
     }
 
-    for (const ImageId overlap_id : ignore_textures) {
+    for (const ImageId overlap_id : join_ignore_textures) {
         Image& overlap = slot_images[overlap_id];
         if (True(overlap.flags & ImageFlagBits::GpuModified)) {
             UNIMPLEMENTED();
@@ -1391,14 +1429,60 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
         ScaleDown(new_image);
     }
 
-    std::ranges::sort(overlap_ids, [this](const ImageId lhs, const ImageId rhs) {
-        const ImageBase& lhs_image = slot_images[lhs];
-        const ImageBase& rhs_image = slot_images[rhs];
+    std::ranges::sort(join_copies_to_do, [this](const JoinCopy& lhs, const JoinCopy& rhs) {
+        const ImageBase& lhs_image = slot_images[lhs.id];
+        const ImageBase& rhs_image = slot_images[rhs.id];
         return lhs_image.modification_tick < rhs_image.modification_tick;
     });
 
-    for (const ImageId overlap_id : overlap_ids) {
-        Image& overlap = slot_images[overlap_id];
+    ImageBase& new_image_base = new_image;
+    for (const ImageId aliased_id : join_right_aliased_ids) {
+        ImageBase& aliased = slot_images[aliased_id];
+        size_t alias_index = new_image_base.aliased_images.size();
+        if (!AddImageAlias(new_image_base, aliased, new_image_id, aliased_id)) {
+            continue;
+        }
+        join_alias_indices.emplace(aliased_id, alias_index);
+        new_image.flags |= ImageFlagBits::Alias;
+    }
+    for (const ImageId aliased_id : join_left_aliased_ids) {
+        ImageBase& aliased = slot_images[aliased_id];
+        size_t alias_index = new_image_base.aliased_images.size();
+        if (!AddImageAlias(aliased, new_image_base, aliased_id, new_image_id)) {
+            continue;
+        }
+        join_alias_indices.emplace(aliased_id, alias_index);
+        new_image.flags |= ImageFlagBits::Alias;
+    }
+    for (const ImageId aliased_id : join_bad_overlap_ids) {
+        ImageBase& aliased = slot_images[aliased_id];
+        aliased.overlapping_images.push_back(new_image_id);
+        new_image.overlapping_images.push_back(aliased_id);
+        if (aliased.info.resources.levels == 1 && aliased.info.block.depth == 0 &&
+            aliased.overlapping_images.size() > 1) {
+            aliased.flags |= ImageFlagBits::BadOverlap;
+        }
+        if (new_image.info.resources.levels == 1 && new_image.info.block.depth == 0 &&
+            new_image.overlapping_images.size() > 1) {
+            new_image.flags |= ImageFlagBits::BadOverlap;
+        }
+    }
+
+    for (const auto& copy_object : join_copies_to_do) {
+        Image& overlap = slot_images[copy_object.id];
+        if (copy_object.is_alias) {
+            if (!overlap.IsSafeDownload()) {
+                continue;
+            }
+            const auto alias_pointer = join_alias_indices.find(copy_object.id);
+            if (alias_pointer == join_alias_indices.end()) {
+                continue;
+            }
+            const AliasedImage& aliased = new_image.aliased_images[alias_pointer->second];
+            CopyImage(new_image_id, aliased.id, aliased.copies);
+            new_image.modification_tick = overlap.modification_tick;
+            continue;
+        }
         if (True(overlap.flags & ImageFlagBits::GpuModified)) {
             new_image.flags |= ImageFlagBits::GpuModified;
             const auto& resolution = Settings::values.resolution_info;
@@ -1411,35 +1495,15 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
             } else {
                 runtime.CopyImage(new_image, overlap, std::move(copies));
             }
+            new_image.modification_tick = overlap.modification_tick;
         }
         if (True(overlap.flags & ImageFlagBits::Tracked)) {
-            UntrackImage(overlap, overlap_id);
-        }
-        UnregisterImage(overlap_id);
-        DeleteImage(overlap_id);
-    }
-    ImageBase& new_image_base = new_image;
-    for (const ImageId aliased_id : right_aliased_ids) {
-        ImageBase& aliased = slot_images[aliased_id];
-        AddImageAlias(new_image_base, aliased, new_image_id, aliased_id);
-        new_image.flags |= ImageFlagBits::Alias;
-    }
-    for (const ImageId aliased_id : left_aliased_ids) {
-        ImageBase& aliased = slot_images[aliased_id];
-        AddImageAlias(aliased, new_image_base, aliased_id, new_image_id);
-        new_image.flags |= ImageFlagBits::Alias;
-    }
-    for (const ImageId aliased_id : bad_overlap_ids) {
-        ImageBase& aliased = slot_images[aliased_id];
-        aliased.overlapping_images.push_back(new_image_id);
-        new_image.overlapping_images.push_back(aliased_id);
-        if (aliased.info.resources.levels == 1 && aliased.overlapping_images.size() > 1) {
-            aliased.flags |= ImageFlagBits::BadOverlap;
-        }
-        if (new_image.info.resources.levels == 1 && new_image.overlapping_images.size() > 1) {
-            new_image.flags |= ImageFlagBits::BadOverlap;
+            UntrackImage(overlap, copy_object.id);
         }
+        UnregisterImage(copy_object.id);
+        DeleteImage(copy_object.id);
     }
+
     RegisterImage(new_image_id);
     return new_image_id;
 }
@@ -1469,7 +1533,7 @@ std::optional<typename TextureCache<P>::BlitImages> TextureCache<P>::GetBlitImag
         if (!copy.must_accelerate) {
             do {
                 if (!src_id && !dst_id) {
-                    break;
+                    return std::nullopt;
                 }
                 if (src_id && True(slot_images[src_id].flags & ImageFlagBits::GpuModified)) {
                     break;
@@ -1847,10 +1911,6 @@ void TextureCache<P>::RegisterImage(ImageId image_id) {
         tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format);
     }
     total_used_memory += Common::AlignUp(tentative_size, 1024);
-    if (total_used_memory > critical_memory && critical_gc < GC_EMERGENCY_COUNTS) {
-        RunGarbageCollector();
-        critical_gc++;
-    }
     image.lru_index = lru_cache.Insert(image_id, frame_tick);
 
     ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, image_id](u64 page) {

src/video_core/texture_cache/texture_cache_base.h

@@ -10,7 +10,9 @@
 #include <span>
 #include <type_traits>
 #include <unordered_map>
+#include <unordered_set>
 #include <vector>
+#include <boost/container/small_vector.hpp>
 #include <queue>
 
 #include "common/common_types.h"
@@ -148,6 +150,9 @@ public:
     /// Fill image_view_ids with the compute images in indices
     void FillComputeImageViews(std::span<ImageViewInOut> views);
 
+    /// Handle feedback loops during draws.
+    void CheckFeedbackLoop(std::span<const ImageViewInOut> views);
+
     /// Get the sampler from the graphics descriptor table in the specified index
     Sampler* GetGraphicsSampler(u32 index);
 
@@ -424,7 +429,6 @@ private:
     u64 minimum_memory;
     u64 expected_memory;
     u64 critical_memory;
-    size_t critical_gc;
 
     struct BufferDownload {
         GPUVAddr address;
@@ -474,6 +478,20 @@ private:
 
     Common::ThreadWorker texture_decode_worker{1, "TextureDecoder"};
     std::vector<std::unique_ptr<AsyncDecodeContext>> async_decodes;
+
+    // Join caching
+    boost::container::small_vector<ImageId, 4> join_overlap_ids;
+    std::unordered_set<ImageId> join_overlaps_found;
+    boost::container::small_vector<ImageId, 4> join_left_aliased_ids;
+    boost::container::small_vector<ImageId, 4> join_right_aliased_ids;
+    std::unordered_set<ImageId> join_ignore_textures;
+    boost::container::small_vector<ImageId, 4> join_bad_overlap_ids;
+    struct JoinCopy {
+        bool is_alias;
+        ImageId id;
+    };
+    boost::container::small_vector<JoinCopy, 4> join_copies_to_do;
+    std::unordered_map<ImageId, size_t> join_alias_indices;
 };
 
 } // namespace VideoCommon

src/video_core/texture_cache/util.cpp

@@ -18,6 +18,8 @@
 #include "common/bit_util.h"
 #include "common/common_types.h"
 #include "common/div_ceil.h"
+#include "common/scratch_buffer.h"
+#include "common/settings.h"
 #include "video_core/compatible_formats.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/memory_manager.h"
@@ -28,6 +30,7 @@
 #include "video_core/texture_cache/samples_helper.h"
 #include "video_core/texture_cache/util.h"
 #include "video_core/textures/astc.h"
+#include "video_core/textures/bcn.h"
 #include "video_core/textures/decoders.h"
 
 namespace VideoCommon {
@@ -120,7 +123,9 @@ template <u32 GOB_EXTENT>
     return {
         .width = AdjustMipBlockSize<GOB_SIZE_X>(num_tiles.width, block_size.width, level),
         .height = AdjustMipBlockSize<GOB_SIZE_Y>(num_tiles.height, block_size.height, level),
-        .depth = AdjustMipBlockSize<GOB_SIZE_Z>(num_tiles.depth, block_size.depth, level),
+        .depth = level == 0
+                     ? block_size.depth
+                     : AdjustMipBlockSize<GOB_SIZE_Z>(num_tiles.depth, block_size.depth, level),
     };
 }
 
@@ -162,6 +167,13 @@ template <u32 GOB_EXTENT>
 }
 
 [[nodiscard]] constexpr Extent3D TileShift(const LevelInfo& info, u32 level) {
+    if (level == 0) {
+        return Extent3D{
+            .width = info.block.width,
+            .height = info.block.height,
+            .depth = info.block.depth,
+        };
+    }
     const Extent3D blocks = NumLevelBlocks(info, level);
     return Extent3D{
         .width = AdjustTileSize(info.block.width, GOB_SIZE_X, blocks.width),
@@ -585,6 +597,21 @@ u32 CalculateConvertedSizeBytes(const ImageInfo& info) noexcept {
         return info.size.width * BytesPerBlock(info.format);
     }
     static constexpr Extent2D TILE_SIZE{1, 1};
+    if (IsPixelFormatASTC(info.format) && Settings::values.astc_recompression.GetValue() !=
+                                              Settings::AstcRecompression::Uncompressed) {
+        const u32 bpp_div =
+            Settings::values.astc_recompression.GetValue() == Settings::AstcRecompression::Bc1 ? 2
+                                                                                               : 1;
+        // NumBlocksPerLayer doesn't account for this correctly, so we have to do it manually.
+        u32 output_size = 0;
+        for (s32 i = 0; i < info.resources.levels; i++) {
+            const auto mip_size = AdjustMipSize(info.size, i);
+            const u32 plane_dim =
+                Common::AlignUp(mip_size.width, 4U) * Common::AlignUp(mip_size.height, 4U);
+            output_size += (plane_dim * info.size.depth * info.resources.layers) / bpp_div;
+        }
+        return output_size;
+    }
     return NumBlocksPerLayer(info, TILE_SIZE) * info.resources.layers * CONVERTED_BYTES_PER_BLOCK;
 }
 
@@ -885,6 +912,7 @@ BufferCopy UploadBufferCopy(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr,
 void ConvertImage(std::span<const u8> input, const ImageInfo& info, std::span<u8> output,
                   std::span<BufferImageCopy> copies) {
     u32 output_offset = 0;
+    Common::ScratchBuffer<u8> decode_scratch;
 
     const Extent2D tile_size = DefaultBlockSize(info.format);
     for (BufferImageCopy& copy : copies) {
@@ -895,22 +923,58 @@ void ConvertImage(std::span<const u8> input, const ImageInfo& info, std::span<u8
         ASSERT(copy.image_extent == mip_size);
         ASSERT(copy.buffer_row_length == Common::AlignUp(mip_size.width, tile_size.width));
         ASSERT(copy.buffer_image_height == Common::AlignUp(mip_size.height, tile_size.height));
-        if (IsPixelFormatASTC(info.format)) {
-            Tegra::Texture::ASTC::Decompress(
-                input.subspan(copy.buffer_offset), copy.image_extent.width,
-                copy.image_extent.height,
-                copy.image_subresource.num_layers * copy.image_extent.depth, tile_size.width,
-                tile_size.height, output.subspan(output_offset));
-        } else {
-            DecompressBC4(input.subspan(copy.buffer_offset), copy.image_extent,
-                          output.subspan(output_offset));
-        }
+
+        const auto input_offset = input.subspan(copy.buffer_offset);
         copy.buffer_offset = output_offset;
         copy.buffer_row_length = mip_size.width;
         copy.buffer_image_height = mip_size.height;
 
-        output_offset += copy.image_extent.width * copy.image_extent.height *
-                         copy.image_subresource.num_layers * CONVERTED_BYTES_PER_BLOCK;
+        const auto recompression_setting = Settings::values.astc_recompression.GetValue();
+        const bool astc = IsPixelFormatASTC(info.format);
+
+        if (astc && recompression_setting == Settings::AstcRecompression::Uncompressed) {
+            Tegra::Texture::ASTC::Decompress(
+                input_offset, copy.image_extent.width, copy.image_extent.height,
+                copy.image_subresource.num_layers * copy.image_extent.depth, tile_size.width,
+                tile_size.height, output.subspan(output_offset));
+
+            output_offset += copy.image_extent.width * copy.image_extent.height *
+                             copy.image_subresource.num_layers * CONVERTED_BYTES_PER_BLOCK;
+        } else if (astc) {
+            // BC1 uses 0.5 bytes per texel
+            // BC3 uses 1 byte per texel
+            const auto compress = recompression_setting == Settings::AstcRecompression::Bc1
+                                      ? Tegra::Texture::BCN::CompressBC1
+                                      : Tegra::Texture::BCN::CompressBC3;
+            const auto bpp_div = recompression_setting == Settings::AstcRecompression::Bc1 ? 2 : 1;
+
+            const u32 plane_dim = copy.image_extent.width * copy.image_extent.height;
+            const u32 level_size = plane_dim * copy.image_extent.depth *
+                                   copy.image_subresource.num_layers * CONVERTED_BYTES_PER_BLOCK;
+            decode_scratch.resize_destructive(level_size);
+
+            Tegra::Texture::ASTC::Decompress(
+                input_offset, copy.image_extent.width, copy.image_extent.height,
+                copy.image_subresource.num_layers * copy.image_extent.depth, tile_size.width,
+                tile_size.height, decode_scratch);
+
+            compress(decode_scratch, copy.image_extent.width, copy.image_extent.height,
+                     copy.image_subresource.num_layers * copy.image_extent.depth,
+                     output.subspan(output_offset));
+
+            const u32 aligned_plane_dim = Common::AlignUp(copy.image_extent.width, 4) *
+                                          Common::AlignUp(copy.image_extent.height, 4);
+
+            copy.buffer_size =
+                (aligned_plane_dim * copy.image_extent.depth * copy.image_subresource.num_layers) /
+                bpp_div;
+            output_offset += static_cast<u32>(copy.buffer_size);
+        } else {
+            DecompressBC4(input_offset, copy.image_extent, output.subspan(output_offset));
+
+            output_offset += copy.image_extent.width * copy.image_extent.height *
+                             copy.image_subresource.num_layers * CONVERTED_BYTES_PER_BLOCK;
+        }
     }
 }
 
@@ -1233,7 +1297,9 @@ u32 MapSizeBytes(const ImageBase& image) {
 
 static_assert(CalculateLevelSize(LevelInfo{{1920, 1080, 1}, {0, 2, 0}, {1, 1}, 2, 0}, 0) ==
               0x7f8000);
-static_assert(CalculateLevelSize(LevelInfo{{32, 32, 1}, {0, 0, 4}, {1, 1}, 4, 0}, 0) == 0x4000);
+static_assert(CalculateLevelSize(LevelInfo{{32, 32, 1}, {0, 0, 4}, {1, 1}, 4, 0}, 0) == 0x40000);
+
+static_assert(CalculateLevelSize(LevelInfo{{128, 8, 1}, {0, 4, 0}, {1, 1}, 4, 0}, 0) == 0x40000);
 
 static_assert(CalculateLevelOffset(PixelFormat::R8_SINT, {1920, 1080, 1}, {0, 2, 0}, 0, 7) ==
               0x2afc00);

src/video_core/textures/astc.cpp

@@ -16,8 +16,8 @@
 #include "common/alignment.h"
 #include "common/common_types.h"
 #include "common/polyfill_ranges.h"
-#include "common/thread_worker.h"
 #include "video_core/textures/astc.h"
+#include "video_core/textures/workers.h"
 
 class InputBitStream {
 public:
@@ -1656,8 +1656,7 @@ void Decompress(std::span<const uint8_t> data, uint32_t width, uint32_t height,
     const u32 rows = Common::DivideUp(height, block_height);
     const u32 cols = Common::DivideUp(width, block_width);
 
-    static Common::ThreadWorker workers{std::max(std::thread::hardware_concurrency(), 2U) / 2,
-                                        "ASTCDecompress"};
+    Common::ThreadWorker& workers{GetThreadWorkers()};
 
     for (u32 z = 0; z < depth; ++z) {
         const u32 depth_offset = z * height * width * 4;

src/video_core/textures/bcn.cpp

@@ -0,0 +1,87 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <stb_dxt.h>
+#include <string.h>
+
+#include "common/alignment.h"
+#include "video_core/textures/bcn.h"
+#include "video_core/textures/workers.h"
+
+namespace Tegra::Texture::BCN {
+
+using BCNCompressor = void(u8* block_output, const u8* block_input, bool any_alpha);
+
+template <u32 BytesPerBlock, bool ThresholdAlpha = false>
+void CompressBCN(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,
+                 std::span<uint8_t> output, BCNCompressor f) {
+    constexpr u8 alpha_threshold = 128;
+    constexpr u32 bytes_per_px = 4;
+    const u32 plane_dim = width * height;
+
+    Common::ThreadWorker& workers{GetThreadWorkers()};
+
+    for (u32 z = 0; z < depth; z++) {
+        for (u32 y = 0; y < height; y += 4) {
+            auto compress_row = [z, y, width, height, plane_dim, f, data, output]() {
+                for (u32 x = 0; x < width; x += 4) {
+                    // Gather 4x4 block of RGBA texels
+                    u8 input_colors[4][4][4];
+                    bool any_alpha = false;
+
+                    for (u32 j = 0; j < 4; j++) {
+                        for (u32 i = 0; i < 4; i++) {
+                            const size_t coord =
+                                (z * plane_dim + (y + j) * width + (x + i)) * bytes_per_px;
+
+                            if ((x + i < width) && (y + j < height)) {
+                                if constexpr (ThresholdAlpha) {
+                                    if (data[coord + 3] >= alpha_threshold) {
+                                        input_colors[j][i][0] = data[coord + 0];
+                                        input_colors[j][i][1] = data[coord + 1];
+                                        input_colors[j][i][2] = data[coord + 2];
+                                        input_colors[j][i][3] = 255;
+                                    } else {
+                                        any_alpha = true;
+                                        memset(input_colors[j][i], 0, bytes_per_px);
+                                    }
+                                } else {
+                                    memcpy(input_colors[j][i], &data[coord], bytes_per_px);
+                                }
+                            } else {
+                                memset(input_colors[j][i], 0, bytes_per_px);
+                            }
+                        }
+                    }
+
+                    const u32 bytes_per_row = BytesPerBlock * Common::DivideUp(width, 4U);
+                    const u32 bytes_per_plane = bytes_per_row * Common::DivideUp(height, 4U);
+                    f(output.data() + z * bytes_per_plane + (y / 4) * bytes_per_row +
+                          (x / 4) * BytesPerBlock,
+                      reinterpret_cast<u8*>(input_colors), any_alpha);
+                }
+            };
+            workers.QueueWork(std::move(compress_row));
+        }
+        workers.WaitForRequests();
+    }
+}
+
+void CompressBC1(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,
+                 std::span<uint8_t> output) {
+    CompressBCN<8, true>(data, width, height, depth, output,
+                         [](u8* block_output, const u8* block_input, bool any_alpha) {
+                             stb_compress_bc1_block(block_output, block_input, any_alpha,
+                                                    STB_DXT_NORMAL);
+                         });
+}
+
+void CompressBC3(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,
+                 std::span<uint8_t> output) {
+    CompressBCN<16, false>(data, width, height, depth, output,
+                           [](u8* block_output, const u8* block_input, bool any_alpha) {
+                               stb_compress_bc3_block(block_output, block_input, STB_DXT_NORMAL);
+                           });
+}
+
+} // namespace Tegra::Texture::BCN

src/video_core/textures/bcn.h

@@ -0,0 +1,17 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <span>
+#include <stdint.h>
+
+namespace Tegra::Texture::BCN {
+
+void CompressBC1(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,
+                 std::span<uint8_t> output);
+
+void CompressBC3(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,
+                 std::span<uint8_t> output);
+
+} // namespace Tegra::Texture::BCN

src/video_core/textures/workers.cpp

@@ -0,0 +1,15 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "video_core/textures/workers.h"
+
+namespace Tegra::Texture {
+
+Common::ThreadWorker& GetThreadWorkers() {
+    static Common::ThreadWorker workers{std::max(std::thread::hardware_concurrency(), 2U) / 2,
+                                        "ImageTranscode"};
+
+    return workers;
+}
+
+} // namespace Tegra::Texture

src/video_core/textures/workers.h

@@ -0,0 +1,12 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include "common/thread_worker.h"
+
+namespace Tegra::Texture {
+
+Common::ThreadWorker& GetThreadWorkers();
+
+}

src/video_core/vulkan_common/vulkan_device.cpp

@@ -473,11 +473,12 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
     }
     if (extensions.push_descriptor && is_intel_anv) {
         const u32 version = (properties.properties.driverVersion << 3) >> 3;
-        if (version >= VK_MAKE_API_VERSION(0, 22, 3, 0)) {
+        if (version >= VK_MAKE_API_VERSION(0, 22, 3, 0) &&
+            version < VK_MAKE_API_VERSION(0, 23, 2, 0)) {
             // Disable VK_KHR_push_descriptor due to
             // mesa/mesa/-/commit/ff91c5ca42bc80aa411cb3fd8f550aa6fdd16bdc
             LOG_WARNING(Render_Vulkan,
-                        "ANV drivers 22.3.0 and later have broken VK_KHR_push_descriptor");
+                        "ANV drivers 22.3.0 to 23.1.0 have broken VK_KHR_push_descriptor");
             extensions.push_descriptor = false;
             loaded_extensions.erase(VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME);
         }
@@ -1001,6 +1002,11 @@ u64 Device::GetDeviceMemoryUsage() const {
 }
 
 void Device::CollectPhysicalMemoryInfo() {
+    // Account for resolution scaling in memory limits
+    const size_t normal_memory = 6_GiB;
+    const size_t scaler_memory = 1_GiB * Settings::values.resolution_info.ScaleUp(1);
+
+    // Calculate limits using memory budget
     VkPhysicalDeviceMemoryBudgetPropertiesEXT budget{};
     budget.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MEMORY_BUDGET_PROPERTIES_EXT;
     const auto mem_info =
@@ -1030,11 +1036,12 @@ void Device::CollectPhysicalMemoryInfo() {
     if (!is_integrated) {
         const u64 reserve_memory = std::min<u64>(device_access_memory / 8, 1_GiB);
         device_access_memory -= reserve_memory;
+        device_access_memory = std::min<u64>(device_access_memory, normal_memory + scaler_memory);
         return;
     }
     const s64 available_memory = static_cast<s64>(device_access_memory - device_initial_usage);
     device_access_memory = static_cast<u64>(std::max<s64>(
-        std::min<s64>(available_memory - 8_GiB, 4_GiB), static_cast<s64>(local_memory)));
+        std::min<s64>(available_memory - 8_GiB, 4_GiB), std::min<s64>(local_memory, 4_GiB)));
 }
 
 void Device::CollectToolingInfo() {

src/yuzu/CMakeLists.txt

@@ -378,11 +378,7 @@ if(UNIX AND NOT APPLE)
 endif()
 
 if (WIN32 AND QT_VERSION VERSION_GREATER_EQUAL 6)
-    if (MSVC AND NOT ${CMAKE_GENERATOR} STREQUAL "Ninja")
-        set(YUZU_EXE_DIR "${CMAKE_BINARY_DIR}/bin/$<CONFIG>")
-    else()
-        set(YUZU_EXE_DIR "${CMAKE_BINARY_DIR}/bin")
-    endif()
+    set(YUZU_EXE_DIR "$<TARGET_FILE_DIR:yuzu>")
     add_custom_command(TARGET yuzu POST_BUILD COMMAND ${WINDEPLOYQT_EXECUTABLE} "${YUZU_EXE_DIR}/yuzu.exe" --dir "${YUZU_EXE_DIR}" --libdir "${YUZU_EXE_DIR}" --plugindir "${YUZU_EXE_DIR}/plugins" --no-compiler-runtime --no-opengl-sw --no-system-d3d-compiler --no-translations --verbose 0)
 endif()
 

src/yuzu/configuration/config.cpp

@@ -711,6 +711,7 @@ void Config::ReadRendererValues() {
     ReadGlobalSetting(Settings::values.nvdec_emulation);
     ReadGlobalSetting(Settings::values.accelerate_astc);
     ReadGlobalSetting(Settings::values.async_astc);
+    ReadGlobalSetting(Settings::values.astc_recompression);
     ReadGlobalSetting(Settings::values.use_reactive_flushing);
     ReadGlobalSetting(Settings::values.shader_backend);
     ReadGlobalSetting(Settings::values.use_asynchronous_shaders);
@@ -1359,6 +1360,10 @@ void Config::SaveRendererValues() {
                  Settings::values.nvdec_emulation.UsingGlobal());
     WriteGlobalSetting(Settings::values.accelerate_astc);
     WriteGlobalSetting(Settings::values.async_astc);
+    WriteSetting(QString::fromStdString(Settings::values.astc_recompression.GetLabel()),
+                 static_cast<u32>(Settings::values.astc_recompression.GetValue(global)),
+                 static_cast<u32>(Settings::values.astc_recompression.GetDefault()),
+                 Settings::values.astc_recompression.UsingGlobal());
     WriteGlobalSetting(Settings::values.use_reactive_flushing);
     WriteSetting(QString::fromStdString(Settings::values.shader_backend.GetLabel()),
                  static_cast<u32>(Settings::values.shader_backend.GetValue(global)),

src/yuzu/configuration/config.h

@@ -208,3 +208,4 @@ Q_DECLARE_METATYPE(Settings::ScalingFilter);
 Q_DECLARE_METATYPE(Settings::AntiAliasing);
 Q_DECLARE_METATYPE(Settings::RendererBackend);
 Q_DECLARE_METATYPE(Settings::ShaderBackend);
+Q_DECLARE_METATYPE(Settings::AstcRecompression);

src/yuzu/configuration/configure_graphics_advanced.cpp

@@ -27,6 +27,7 @@ void ConfigureGraphicsAdvanced::SetConfiguration() {
     ui->async_present->setEnabled(runtime_lock);
     ui->renderer_force_max_clock->setEnabled(runtime_lock);
     ui->async_astc->setEnabled(runtime_lock);
+    ui->astc_recompression_combobox->setEnabled(runtime_lock);
     ui->use_asynchronous_shaders->setEnabled(runtime_lock);
     ui->anisotropic_filtering_combobox->setEnabled(runtime_lock);
     ui->enable_compute_pipelines_checkbox->setEnabled(runtime_lock);
@@ -47,14 +48,20 @@ void ConfigureGraphicsAdvanced::SetConfiguration() {
             static_cast<int>(Settings::values.gpu_accuracy.GetValue()));
         ui->anisotropic_filtering_combobox->setCurrentIndex(
             Settings::values.max_anisotropy.GetValue());
+        ui->astc_recompression_combobox->setCurrentIndex(
+            static_cast<int>(Settings::values.astc_recompression.GetValue()));
     } else {
         ConfigurationShared::SetPerGameSetting(ui->gpu_accuracy, &Settings::values.gpu_accuracy);
         ConfigurationShared::SetPerGameSetting(ui->anisotropic_filtering_combobox,
                                                &Settings::values.max_anisotropy);
+        ConfigurationShared::SetPerGameSetting(ui->astc_recompression_combobox,
+                                               &Settings::values.astc_recompression);
         ConfigurationShared::SetHighlight(ui->label_gpu_accuracy,
                                           !Settings::values.gpu_accuracy.UsingGlobal());
         ConfigurationShared::SetHighlight(ui->af_label,
                                           !Settings::values.max_anisotropy.UsingGlobal());
+        ConfigurationShared::SetHighlight(ui->label_astc_recompression,
+                                          !Settings::values.astc_recompression.UsingGlobal());
     }
 }
 
@@ -71,6 +78,8 @@ void ConfigureGraphicsAdvanced::ApplyConfiguration() {
                                              ui->use_reactive_flushing, use_reactive_flushing);
     ConfigurationShared::ApplyPerGameSetting(&Settings::values.async_astc, ui->async_astc,
                                              async_astc);
+    ConfigurationShared::ApplyPerGameSetting(&Settings::values.astc_recompression,
+                                             ui->astc_recompression_combobox);
     ConfigurationShared::ApplyPerGameSetting(&Settings::values.use_asynchronous_shaders,
                                              ui->use_asynchronous_shaders,
                                              use_asynchronous_shaders);
@@ -105,6 +114,8 @@ void ConfigureGraphicsAdvanced::SetupPerGameUI() {
             Settings::values.renderer_force_max_clock.UsingGlobal());
         ui->use_reactive_flushing->setEnabled(Settings::values.use_reactive_flushing.UsingGlobal());
         ui->async_astc->setEnabled(Settings::values.async_astc.UsingGlobal());
+        ui->astc_recompression_combobox->setEnabled(
+            Settings::values.astc_recompression.UsingGlobal());
         ui->use_asynchronous_shaders->setEnabled(
             Settings::values.use_asynchronous_shaders.UsingGlobal());
         ui->use_fast_gpu_time->setEnabled(Settings::values.use_fast_gpu_time.UsingGlobal());
@@ -144,6 +155,9 @@ void ConfigureGraphicsAdvanced::SetupPerGameUI() {
     ConfigurationShared::SetColoredComboBox(
         ui->anisotropic_filtering_combobox, ui->af_label,
         static_cast<int>(Settings::values.max_anisotropy.GetValue(true)));
+    ConfigurationShared::SetColoredComboBox(
+        ui->astc_recompression_combobox, ui->label_astc_recompression,
+        static_cast<int>(Settings::values.astc_recompression.GetValue(true)));
 }
 
 void ConfigureGraphicsAdvanced::ExposeComputeOption() {

src/yuzu/configuration/configure_graphics_advanced.ui

@@ -69,6 +69,50 @@
           </layout>
          </widget>
         </item>
+        <item>
+         <widget class="QWidget" name="astc_recompression_layout" native="true">
+          <layout class="QHBoxLayout" name="horizontalLayout_3">
+            <property name="leftMargin">
+              <number>0</number>
+            </property>
+            <property name="topMargin">
+              <number>0</number>
+            </property>
+            <property name="rightMargin">
+              <number>0</number>
+            </property>
+            <property name="bottomMargin">
+              <number>0</number>
+            </property>
+            <item>
+              <widget class="QLabel" name="label_astc_recompression">
+              <property name="text">
+                <string>ASTC recompression:</string>
+              </property>
+              </widget>
+            </item>
+            <item>
+              <widget class="QComboBox" name="astc_recompression_combobox">
+                <item>
+                <property name="text">
+                  <string>Uncompressed (Best quality)</string>
+                </property>
+                </item>
+                <item>
+                <property name="text">
+                  <string>BC1 (Low quality)</string>
+                </property>
+                </item>
+                <item>
+                <property name="text">
+                  <string>BC3 (Medium quality)</string>
+                </property>
+                </item>
+              </widget>
+            </item>
+          </layout>
+         </widget>
+        </item>
         <item>
          <widget class="QCheckBox" name="async_present">
           <property name="text">

src/yuzu/game_list.cpp

@@ -544,6 +544,7 @@ void GameList::AddGamePopup(QMenu& context_menu, u64 program_id, const std::stri
     QAction* remove_update = remove_menu->addAction(tr("Remove Installed Update"));
     QAction* remove_dlc = remove_menu->addAction(tr("Remove All Installed DLC"));
     QAction* remove_custom_config = remove_menu->addAction(tr("Remove Custom Configuration"));
+    QAction* remove_cache_storage = remove_menu->addAction(tr("Remove Cache Storage"));
     QAction* remove_gl_shader_cache = remove_menu->addAction(tr("Remove OpenGL Pipeline Cache"));
     QAction* remove_vk_shader_cache = remove_menu->addAction(tr("Remove Vulkan Pipeline Cache"));
     remove_menu->addSeparator();
@@ -614,6 +615,9 @@ void GameList::AddGamePopup(QMenu& context_menu, u64 program_id, const std::stri
     connect(remove_custom_config, &QAction::triggered, [this, program_id, path]() {
         emit RemoveFileRequested(program_id, GameListRemoveTarget::CustomConfiguration, path);
     });
+    connect(remove_cache_storage, &QAction::triggered, [this, program_id, path] {
+        emit RemoveFileRequested(program_id, GameListRemoveTarget::CacheStorage, path);
+    });
     connect(dump_romfs, &QAction::triggered, [this, program_id, path]() {
         emit DumpRomFSRequested(program_id, path, DumpRomFSTarget::Normal);
     });

src/yuzu/game_list.h

@@ -45,6 +45,7 @@ enum class GameListRemoveTarget {
     VkShaderCache,
     AllShaderCache,
     CustomConfiguration,
+    CacheStorage,
 };
 
 enum class DumpRomFSTarget {

src/yuzu/main.cpp

@@ -2323,6 +2323,8 @@ void GMainWindow::OnGameListRemoveFile(u64 program_id, GameListRemoveTarget targ
             return tr("Delete All Transferable Shader Caches?");
         case GameListRemoveTarget::CustomConfiguration:
             return tr("Remove Custom Game Configuration?");
+        case GameListRemoveTarget::CacheStorage:
+            return tr("Remove Cache Storage?");
         default:
             return QString{};
         }
@@ -2346,6 +2348,9 @@ void GMainWindow::OnGameListRemoveFile(u64 program_id, GameListRemoveTarget targ
     case GameListRemoveTarget::CustomConfiguration:
         RemoveCustomConfiguration(program_id, game_path);
         break;
+    case GameListRemoveTarget::CacheStorage:
+        RemoveCacheStorage(program_id);
+        break;
     }
 }
 
@@ -2435,6 +2440,21 @@ void GMainWindow::RemoveCustomConfiguration(u64 program_id, const std::string& g
     }
 }
 
+void GMainWindow::RemoveCacheStorage(u64 program_id) {
+    const auto nand_dir = Common::FS::GetYuzuPath(Common::FS::YuzuPath::NANDDir);
+    auto vfs_nand_dir =
+        vfs->OpenDirectory(Common::FS::PathToUTF8String(nand_dir), FileSys::Mode::Read);
+
+    const auto cache_storage_path = FileSys::SaveDataFactory::GetFullPath(
+        *system, vfs_nand_dir, FileSys::SaveDataSpaceId::NandUser,
+        FileSys::SaveDataType::CacheStorage, 0 /* program_id */, {}, 0);
+
+    const auto path = Common::FS::ConcatPathSafe(nand_dir, cache_storage_path);
+
+    // Not an error if it wasn't cleared.
+    Common::FS::RemoveDirRecursively(path);
+}
+
 void GMainWindow::OnGameListDumpRomFS(u64 program_id, const std::string& game_path,
                                       DumpRomFSTarget target) {
     const auto failed = [this] {

src/yuzu/main.h

@@ -370,6 +370,7 @@ private:
     void RemoveVulkanDriverPipelineCache(u64 program_id);
     void RemoveAllTransferableShaderCaches(u64 program_id);
     void RemoveCustomConfiguration(u64 program_id, const std::string& game_path);
+    void RemoveCacheStorage(u64 program_id);
     std::optional<u64> SelectRomFSDumpTarget(const FileSys::ContentProvider&, u64 program_id);
     InstallResult InstallNSPXCI(const QString& filename);
     InstallResult InstallNCA(const QString& filename);

src/yuzu_cmd/config.cpp

@@ -318,6 +318,7 @@ void Config::ReadValues() {
     ReadSetting("Renderer", Settings::values.nvdec_emulation);
     ReadSetting("Renderer", Settings::values.accelerate_astc);
     ReadSetting("Renderer", Settings::values.async_astc);
+    ReadSetting("Renderer", Settings::values.astc_recompression);
     ReadSetting("Renderer", Settings::values.use_fast_gpu_time);
     ReadSetting("Renderer", Settings::values.use_vulkan_driver_pipeline_cache);
 

src/yuzu_cmd/default_ini.h

@@ -360,6 +360,10 @@ accelerate_astc =
 # 0 (default): Off, 1: On
 async_astc =
 
+# Recompress ASTC textures to a different format.
+# 0 (default): Uncompressed, 1: BC1 (Low quality), 2: BC3: (Medium quality)
+async_astc =
+
 # Turns on the speed limiter, which will limit the emulation speed to the desired speed limit value
 # 0: Off, 1: On (default)
 use_speed_limit =

src/yuzu_cmd/yuzu.cpp

@@ -227,7 +227,7 @@ int main(int argc, char** argv) {
     };
 
     while (optind < argc) {
-        int arg = getopt_long(argc, argv, "g:fhvp::c:", long_options, &option_index);
+        int arg = getopt_long(argc, argv, "g:fhvp::c:u:", long_options, &option_index);
         if (arg != -1) {
             switch (static_cast<char>(arg)) {
             case 'c':
@@ -283,7 +283,7 @@ int main(int argc, char** argv) {
                 break;
             case 'u':
                 selected_user = atoi(optarg);
-                return 0;
+                break;
             case 'v':
                 PrintVersion();
                 return 0;

vcpkg.json

@@ -1,7 +1,7 @@
 {
     "$schema": "https://raw.githubusercontent.com/microsoft/vcpkg-tool/main/docs/vcpkg.schema.json",
     "name": "yuzu",
-    "builtin-baseline": "acc3bcf76b84ae5041c86ab55fe138ae7b8255c7",
+    "builtin-baseline": "656fcc6ab2b05c6d999b7eaca717027ac3738f71",
     "version": "1.0",
     "dependencies": [
         "boost-algorithm",
@@ -53,7 +53,7 @@
         },
         {
             "name": "fmt",
-            "version": "9.0.0"
+            "version": "10.0.0"
         }
     ]
 }