Merge branch 'master' into quickstart-faq

.gitmodules

@@ -13,6 +13,9 @@
 [submodule "soundtouch"]
     path = externals/soundtouch
     url = https://github.com/citra-emu/ext-soundtouch.git
+[submodule "libressl"]
+    path = externals/libressl
+    url = https://github.com/citra-emu/ext-libressl-portable.git
 [submodule "discord-rpc"]
     path = externals/discord-rpc
     url = https://github.com/discordapp/discord-rpc.git

CMakeLists.txt

@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 3.11)
+cmake_minimum_required(VERSION 3.15)
 
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/CMakeModules")
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/externals/cmake-modules")
@@ -13,7 +13,7 @@ project(yuzu)
 option(ENABLE_SDL2 "Enable the SDL2 frontend" ON)
 
 option(ENABLE_QT "Enable the Qt frontend" ON)
-CMAKE_DEPENDENT_OPTION(YUZU_USE_BUNDLED_QT "Download bundled Qt binaries" OFF "ENABLE_QT;MSVC" OFF)
+CMAKE_DEPENDENT_OPTION(YUZU_USE_BUNDLED_QT "Download bundled Qt binaries" ON "ENABLE_QT;MSVC" OFF)
 
 option(ENABLE_WEB_SERVICE "Enable web services (telemetry, etc.)" ON)
 
@@ -152,7 +152,6 @@ macro(yuzu_find_packages)
         "Boost             1.71        boost/1.72.0"
         "Catch2            2.11        catch2/2.11.0"
         "fmt               6.2         fmt/6.2.0"
-        "OpenSSL           1.1         openssl/1.1.1f"
     # can't use until https://github.com/bincrafters/community/issues/1173
         #"libzip            1.5         libzip/1.5.2@bincrafters/stable"
         "lz4               1.8         lz4/1.9.2"
@@ -312,15 +311,6 @@ elseif (TARGET Boost::boost)
     add_library(boost ALIAS Boost::boost)
 endif()
 
-if (NOT TARGET OpenSSL::SSL)
-    set_target_properties(OpenSSL::OpenSSL PROPERTIES IMPORTED_GLOBAL TRUE)
-    add_library(OpenSSL::SSL ALIAS OpenSSL::OpenSSL)
-endif()
-if (NOT TARGET OpenSSL::Crypto)
-    set_target_properties(OpenSSL::OpenSSL PROPERTIES IMPORTED_GLOBAL TRUE)
-    add_library(OpenSSL::Crypto ALIAS OpenSSL::OpenSSL)
-endif()
-
 if (TARGET sdl2::sdl2)
     # imported from the conan generated sdl2Config.cmake
     set_target_properties(sdl2::sdl2 PROPERTIES IMPORTED_GLOBAL TRUE)

CMakeModules/GenerateSCMRev.cmake

@@ -51,6 +51,8 @@ endif()
 # The variable SRC_DIR must be passed into the script (since it uses the current build directory for all values of CMAKE_*_DIR)
 set(VIDEO_CORE "${SRC_DIR}/src/video_core")
 set(HASH_FILES
+    "${VIDEO_CORE}/renderer_opengl/gl_arb_decompiler.cpp"
+    "${VIDEO_CORE}/renderer_opengl/gl_arb_decompiler.h"
     "${VIDEO_CORE}/renderer_opengl/gl_shader_cache.cpp"
     "${VIDEO_CORE}/renderer_opengl/gl_shader_cache.h"
     "${VIDEO_CORE}/renderer_opengl/gl_shader_decompiler.cpp"

dist/qt_themes/qdarkstyle/style.qss

@@ -673,10 +673,6 @@ QTabWidget::pane {
     border-bottom-left-radius: 2px;
 }
 
-QTabWidget::tab-bar {
-    overflow: visible;
-}
-
 QTabBar {
     qproperty-drawBase: 0;
     border-radius: 3px;

dist/yuzu.manifest

@@ -1,24 +1,58 @@
 <?xml version="1.0" encoding="UTF-8" standalone="yes"?>
-<assembly xmlns="urn:schemas-microsoft-com:asm.v1" manifestVersion="1.0">
- <trustInfo xmlns="urn:schemas-microsoft-com:asm.v3">
-  <security>
-   <requestedPrivileges>
-    <requestedExecutionLevel level="asInvoker" uiAccess="false"/>
-   </requestedPrivileges>
-  </security>
- </trustInfo>
- <application xmlns="urn:schemas-microsoft-com:asm.v3">
-  <windowsSettings>
-   <dpiAware xmlns="http://schemas.microsoft.com/SMI/2005/WindowsSettings">True/PM</dpiAware>
-   <longPathAware xmlns="http://schemas.microsoft.com/SMI/2016/WindowsSettings">true</longPathAware>
-  </windowsSettings>
- </application>
- <compatibility xmlns="urn:schemas-microsoft-com:compatibility.v1">
-  <application>
-   <supportedOS Id="{35138b9a-5d96-4fbd-8e2d-a2440225f93a}"/>
-   <supportedOS Id="{4a2f28e3-53b9-4441-ba9c-d69d4a4a6e38}"/>
-   <supportedOS Id="{1f676c76-80e1-4239-95bb-83d0f6d0da78}"/>
-   <supportedOS Id="{8e0f7a12-bfb3-4fe8-b9a5-48fd50a15a9a}"/>
-  </application>
- </compatibility>
-</assembly>
+<assembly manifestVersion="1.0"
+    xmlns="urn:schemas-microsoft-com:asm.v1"
+    xmlns:asmv3="urn:schemas-microsoft-com:asm.v3">
+  <asmv3:application>
+    <asmv3:windowsSettings>
+      <!-- Windows 7/8/8.1/10 -->
+      <dpiAware
+        xmlns="http://schemas.microsoft.com/SMI/2005/WindowsSettings">
+        true/pm
+      </dpiAware>
+      <!-- Windows 10, version 1607 or later -->
+      <dpiAwareness
+        xmlns="http://schemas.microsoft.com/SMI/2016/WindowsSettings">
+        PerMonitorV2
+      </dpiAwareness>
+      <!-- Windows 10, version 1703 or later -->
+      <gdiScaling
+          xmlns="http://schemas.microsoft.com/SMI/2017/WindowsSettings">
+        true
+      </gdiScaling>
+      <ws2:longPathAware
+          xmlns:ws3="http://schemas.microsoft.com/SMI/2016/WindowsSettings">
+        true
+      </ws2:longPathAware>
+    </asmv3:windowsSettings>
+  </asmv3:application>
+  <compatibility
+      xmlns="urn:schemas-microsoft-com:compatibility.v1">
+    <application>
+      <!-- Windows 10 -->
+      <supportedOS Id="{8e0f7a12-bfb3-4fe8-b9a5-48fd50a15a9a}"/>
+      <!-- Windows 8.1 -->
+      <supportedOS Id="{1f676c76-80e1-4239-95bb-83d0f6d0da78}"/>
+      <!-- Windows 8 -->
+      <supportedOS Id="{4a2f28e3-53b9-4441-ba9c-d69d4a4a6e38}"/>
+      <!-- Windows 7 -->
+      <supportedOS Id="{35138b9a-5d96-4fbd-8e2d-a2440225f93a}"/>
+    </application>
+  </compatibility>
+  <trustInfo
+      xmlns="urn:schemas-microsoft-com:asm.v3">
+    <security>
+      <requestedPrivileges>
+        <!--
+          UAC settings:
+          - app should run at same integrity level as calling process
+          - app does not need to manipulate windows belonging to
+            higher-integrity-level processes
+          -->
+        <requestedExecutionLevel
+            level="asInvoker"
+            uiAccess="false"
+        />
+      </requestedPrivileges>
+    </security>
+  </trustInfo>
+</assembly>

externals/CMakeLists.txt

@@ -4,6 +4,13 @@ list(APPEND CMAKE_MODULE_PATH "${PROJECT_SOURCE_DIR}/CMakeModules")
 list(APPEND CMAKE_MODULE_PATH "${PROJECT_SOURCE_DIR}/externals/find-modules")
 include(DownloadExternals)
 
+# xbyak
+if (ARCHITECTURE_x86 OR ARCHITECTURE_x86_64)
+    add_library(xbyak INTERFACE)
+    target_include_directories(xbyak SYSTEM INTERFACE ./xbyak/xbyak)
+    target_compile_definitions(xbyak INTERFACE XBYAK_NO_OP_NAMES)
+endif()
+
 # Catch
 add_library(catch-single-include INTERFACE)
 target_include_directories(catch-single-include INTERFACE catch/single_include)
@@ -66,6 +73,15 @@ if (NOT LIBZIP_FOUND)
 endif()
 
 if (ENABLE_WEB_SERVICE)
+    # LibreSSL
+    set(LIBRESSL_SKIP_INSTALL ON CACHE BOOL "")
+    add_subdirectory(libressl EXCLUDE_FROM_ALL)
+    target_include_directories(ssl INTERFACE ./libressl/include)
+    target_compile_definitions(ssl PRIVATE -DHAVE_INET_NTOP)
+    get_directory_property(OPENSSL_LIBRARIES
+        DIRECTORY libressl
+        DEFINITION OPENSSL_LIBS)
+
     # lurlparser
     add_subdirectory(lurlparser EXCLUDE_FROM_ALL)
 
@@ -73,13 +89,5 @@ if (ENABLE_WEB_SERVICE)
     add_library(httplib INTERFACE)
     target_include_directories(httplib INTERFACE ./httplib)
     target_compile_definitions(httplib INTERFACE -DCPPHTTPLIB_OPENSSL_SUPPORT)
-    target_link_libraries(httplib INTERFACE OpenSSL::SSL OpenSSL::Crypto)
-endif()
-
-if (NOT TARGET xbyak)
-    if (ARCHITECTURE_x86 OR ARCHITECTURE_x86_64)
-        add_library(xbyak INTERFACE)
-        target_include_directories(xbyak SYSTEM INTERFACE ./xbyak/xbyak)
-        target_compile_definitions(xbyak INTERFACE XBYAK_NO_OP_NAMES)
-    endif()
+    target_link_libraries(httplib INTERFACE ${OPENSSL_LIBRARIES})
 endif()

externals/libressl

@@ -0,0 +1 @@
+Subproject commit 7d01cb01cb1a926ecb4c9c98b107ef3c26f59dfb

externals/sirit

@@ -1 +1 @@
-Subproject commit a62c5bbc100a5e5a31ea0ccc4a78d8fa6a4167ce
+Subproject commit eefca56afd49379bdebc97ded8b480839f930881

src/audio_core/audio_renderer.cpp

@@ -180,11 +180,12 @@ ResultVal<std::vector<u8>> AudioRenderer::UpdateAudioRenderer(const std::vector<
 
     // Copy output header
     UpdateDataHeader response_data{worker_params};
-    std::vector<u8> output_params(response_data.total_size);
     if (behavior_info.IsElapsedFrameCountSupported()) {
-        response_data.frame_count = 0x10;
-        response_data.total_size += 0x10;
+        response_data.render_info = sizeof(RendererInfo);
+        response_data.total_size += sizeof(RendererInfo);
     }
+
+    std::vector<u8> output_params(response_data.total_size);
     std::memcpy(output_params.data(), &response_data, sizeof(UpdateDataHeader));
 
     // Copy output memory pool entries
@@ -219,6 +220,17 @@ ResultVal<std::vector<u8>> AudioRenderer::UpdateAudioRenderer(const std::vector<
         return Audren::ERR_INVALID_PARAMETERS;
     }
 
+    if (behavior_info.IsElapsedFrameCountSupported()) {
+        const std::size_t renderer_info_offset{
+            sizeof(UpdateDataHeader) + response_data.memory_pools_size + response_data.voices_size +
+            response_data.effects_size + response_data.sinks_size +
+            response_data.performance_manager_size + response_data.behavior_size};
+        RendererInfo renderer_info{};
+        renderer_info.elasped_frame_count = elapsed_frame_count;
+        std::memcpy(output_params.data() + renderer_info_offset, &renderer_info,
+                    sizeof(RendererInfo));
+    }
+
     return MakeResult(output_params);
 }
 
@@ -447,6 +459,7 @@ void AudioRenderer::QueueMixedBuffer(Buffer::Tag tag) {
         }
     }
     audio_out->QueueBuffer(stream, tag, std::move(buffer));
+    elapsed_frame_count++;
 }
 
 void AudioRenderer::ReleaseAndQueueBuffers() {

src/audio_core/audio_renderer.h

@@ -196,6 +196,12 @@ struct EffectOutStatus {
 };
 static_assert(sizeof(EffectOutStatus) == 0x10, "EffectOutStatus is an invalid size");
 
+struct RendererInfo {
+    u64_le elasped_frame_count{};
+    INSERT_PADDING_WORDS(2);
+};
+static_assert(sizeof(RendererInfo) == 0x10, "RendererInfo is an invalid size");
+
 struct UpdateDataHeader {
     UpdateDataHeader() {}
 
@@ -209,7 +215,7 @@ struct UpdateDataHeader {
         mixes_size = 0x0;
         sinks_size = config.sink_count * 0x20;
         performance_manager_size = 0x10;
-        frame_count = 0;
+        render_info = 0;
         total_size = sizeof(UpdateDataHeader) + behavior_size + memory_pools_size + voices_size +
                      effects_size + sinks_size + performance_manager_size;
     }
@@ -223,8 +229,8 @@ struct UpdateDataHeader {
     u32_le mixes_size{};
     u32_le sinks_size{};
     u32_le performance_manager_size{};
-    INSERT_PADDING_WORDS(1);
-    u32_le frame_count{};
+    u32_le splitter_size{};
+    u32_le render_info{};
     INSERT_PADDING_WORDS(4);
     u32_le total_size{};
 };
@@ -258,6 +264,7 @@ private:
     std::unique_ptr<AudioOut> audio_out;
     StreamPtr stream;
     Core::Memory::Memory& memory;
+    std::size_t elapsed_frame_count{};
 };
 
 } // namespace AudioCore

src/common/CMakeLists.txt

@@ -32,6 +32,8 @@ add_custom_command(OUTPUT scm_rev.cpp
     DEPENDS
       # WARNING! It was too much work to try and make a common location for this list,
       # so if you need to change it, please update CMakeModules/GenerateSCMRev.cmake as well
+      "${VIDEO_CORE}/renderer_opengl/gl_arb_decompiler.cpp"
+      "${VIDEO_CORE}/renderer_opengl/gl_arb_decompiler.h"
       "${VIDEO_CORE}/renderer_opengl/gl_shader_cache.cpp"
       "${VIDEO_CORE}/renderer_opengl/gl_shader_cache.h"
       "${VIDEO_CORE}/renderer_opengl/gl_shader_decompiler.cpp"

src/common/telemetry.cpp

@@ -60,6 +60,7 @@ void AppendCPUInfo(FieldCollection& fc) {
     fc.AddField(FieldType::UserSystem, "CPU_Extension_x64_AES", Common::GetCPUCaps().aes);
     fc.AddField(FieldType::UserSystem, "CPU_Extension_x64_AVX", Common::GetCPUCaps().avx);
     fc.AddField(FieldType::UserSystem, "CPU_Extension_x64_AVX2", Common::GetCPUCaps().avx2);
+    fc.AddField(FieldType::UserSystem, "CPU_Extension_x64_AVX512", Common::GetCPUCaps().avx512);
     fc.AddField(FieldType::UserSystem, "CPU_Extension_x64_BMI1", Common::GetCPUCaps().bmi1);
     fc.AddField(FieldType::UserSystem, "CPU_Extension_x64_BMI2", Common::GetCPUCaps().bmi2);
     fc.AddField(FieldType::UserSystem, "CPU_Extension_x64_FMA", Common::GetCPUCaps().fma);

src/common/x64/cpu_detect.cpp

@@ -110,6 +110,11 @@ static CPUCaps Detect() {
                 caps.bmi1 = true;
             if ((cpu_id[1] >> 8) & 1)
                 caps.bmi2 = true;
+            // Checks for AVX512F, AVX512CD, AVX512VL, AVX512DQ, AVX512BW (Intel Skylake-X/SP)
+            if ((cpu_id[1] >> 16) & 1 && (cpu_id[1] >> 28) & 1 && (cpu_id[1] >> 31) & 1 &&
+                (cpu_id[1] >> 17) & 1 && (cpu_id[1] >> 30) & 1) {
+                caps.avx512 = caps.avx2;
+            }
         }
     }
 

src/common/x64/cpu_detect.h

@@ -19,6 +19,7 @@ struct CPUCaps {
     bool lzcnt;
     bool avx;
     bool avx2;
+    bool avx512;
     bool bmi1;
     bool bmi2;
     bool fma;

src/common/x64/xbyak_abi.h

@@ -11,7 +11,7 @@
 
 namespace Common::X64 {
 
-inline int RegToIndex(const Xbyak::Reg& reg) {
+inline std::size_t RegToIndex(const Xbyak::Reg& reg) {
     using Kind = Xbyak::Reg::Kind;
     ASSERT_MSG((reg.getKind() & (Kind::REG | Kind::XMM)) != 0,
                "RegSet only support GPRs and XMM registers.");
@@ -19,17 +19,17 @@ inline int RegToIndex(const Xbyak::Reg& reg) {
     return reg.getIdx() + (reg.getKind() == Kind::REG ? 0 : 16);
 }
 
-inline Xbyak::Reg64 IndexToReg64(int reg_index) {
+inline Xbyak::Reg64 IndexToReg64(std::size_t reg_index) {
     ASSERT(reg_index < 16);
-    return Xbyak::Reg64(reg_index);
+    return Xbyak::Reg64(static_cast<int>(reg_index));
 }
 
-inline Xbyak::Xmm IndexToXmm(int reg_index) {
+inline Xbyak::Xmm IndexToXmm(std::size_t reg_index) {
     ASSERT(reg_index >= 16 && reg_index < 32);
-    return Xbyak::Xmm(reg_index - 16);
+    return Xbyak::Xmm(static_cast<int>(reg_index - 16));
 }
 
-inline Xbyak::Reg IndexToReg(int reg_index) {
+inline Xbyak::Reg IndexToReg(std::size_t reg_index) {
     if (reg_index < 16) {
         return IndexToReg64(reg_index);
     } else {
@@ -151,9 +151,13 @@ constexpr size_t ABI_SHADOW_SPACE = 0;
 
 #endif
 
-inline void ABI_CalculateFrameSize(std::bitset<32> regs, size_t rsp_alignment,
-                                   size_t needed_frame_size, s32* out_subtraction,
-                                   s32* out_xmm_offset) {
+struct ABIFrameInfo {
+    s32 subtraction;
+    s32 xmm_offset;
+};
+
+inline ABIFrameInfo ABI_CalculateFrameSize(std::bitset<32> regs, size_t rsp_alignment,
+                                           size_t needed_frame_size) {
     const auto count = (regs & ABI_ALL_GPRS).count();
     rsp_alignment -= count * 8;
     size_t subtraction = 0;
@@ -170,33 +174,28 @@ inline void ABI_CalculateFrameSize(std::bitset<32> regs, size_t rsp_alignment,
     rsp_alignment -= subtraction;
     subtraction += rsp_alignment & 0xF;
 
-    *out_subtraction = (s32)subtraction;
-    *out_xmm_offset = (s32)(subtraction - xmm_base_subtraction);
+    return ABIFrameInfo{static_cast<s32>(subtraction),
+                        static_cast<s32>(subtraction - xmm_base_subtraction)};
 }
 
 inline size_t ABI_PushRegistersAndAdjustStack(Xbyak::CodeGenerator& code, std::bitset<32> regs,
                                               size_t rsp_alignment, size_t needed_frame_size = 0) {
-    s32 subtraction, xmm_offset;
-    ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset);
+    auto frame_info = ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size);
+
     for (std::size_t i = 0; i < regs.size(); ++i) {
         if (regs[i] && ABI_ALL_GPRS[i]) {
-            code.push(IndexToReg64(static_cast<int>(i)));
+            code.push(IndexToReg64(i));
         }
     }
-    if (subtraction != 0) {
-        code.sub(code.rsp, subtraction);
-    }
 
-    for (int i = 0; i < regs.count(); i++) {
-        if (regs.test(i) & ABI_ALL_GPRS.test(i)) {
-            code.push(IndexToReg64(i));
-        }
+    if (frame_info.subtraction != 0) {
+        code.sub(code.rsp, frame_info.subtraction);
     }
 
     for (std::size_t i = 0; i < regs.size(); ++i) {
         if (regs[i] && ABI_ALL_XMMS[i]) {
-            code.movaps(code.xword[code.rsp + xmm_offset], IndexToXmm(static_cast<int>(i)));
-            xmm_offset += 0x10;
+            code.movaps(code.xword[code.rsp + frame_info.xmm_offset], IndexToXmm(i));
+            frame_info.xmm_offset += 0x10;
         }
     }
 
@@ -205,59 +204,23 @@ inline size_t ABI_PushRegistersAndAdjustStack(Xbyak::CodeGenerator& code, std::b
 
 inline void ABI_PopRegistersAndAdjustStack(Xbyak::CodeGenerator& code, std::bitset<32> regs,
                                            size_t rsp_alignment, size_t needed_frame_size = 0) {
-    s32 subtraction, xmm_offset;
-    ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset);
+    auto frame_info = ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size);
 
     for (std::size_t i = 0; i < regs.size(); ++i) {
         if (regs[i] && ABI_ALL_XMMS[i]) {
-            code.movaps(IndexToXmm(static_cast<int>(i)), code.xword[code.rsp + xmm_offset]);
-            xmm_offset += 0x10;
+            code.movaps(IndexToXmm(i), code.xword[code.rsp + frame_info.xmm_offset]);
+            frame_info.xmm_offset += 0x10;
         }
     }
 
-    if (subtraction != 0) {
-        code.add(code.rsp, subtraction);
+    if (frame_info.subtraction != 0) {
+        code.add(code.rsp, frame_info.subtraction);
     }
 
     // GPRs need to be popped in reverse order
-    for (int i = 15; i >= 0; i--) {
-        if (regs[i]) {
-            code.pop(IndexToReg64(i));
-        }
-    }
-}
-
-inline size_t ABI_PushRegistersAndAdjustStackGPS(Xbyak::CodeGenerator& code, std::bitset<32> regs,
-                                                 size_t rsp_alignment,
-                                                 size_t needed_frame_size = 0) {
-    s32 subtraction, xmm_offset;
-    ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset);
-
-    for (std::size_t i = 0; i < regs.size(); ++i) {
+    for (std::size_t j = 0; j < regs.size(); ++j) {
+        const std::size_t i = regs.size() - j - 1;
         if (regs[i] && ABI_ALL_GPRS[i]) {
-            code.push(IndexToReg64(static_cast<int>(i)));
-        }
-    }
-
-    if (subtraction != 0) {
-        code.sub(code.rsp, subtraction);
-    }
-
-    return ABI_SHADOW_SPACE;
-}
-
-inline void ABI_PopRegistersAndAdjustStackGPS(Xbyak::CodeGenerator& code, std::bitset<32> regs,
-                                              size_t rsp_alignment, size_t needed_frame_size = 0) {
-    s32 subtraction, xmm_offset;
-    ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset);
-
-    if (subtraction != 0) {
-        code.add(code.rsp, subtraction);
-    }
-
-    // GPRs need to be popped in reverse order
-    for (int i = 15; i >= 0; i--) {
-        if (regs[i]) {
             code.pop(IndexToReg64(i));
         }
     }

src/core/CMakeLists.txt

@@ -606,11 +606,11 @@ endif()
 create_target_directory_groups(core)
 
 target_link_libraries(core PUBLIC common PRIVATE audio_core video_core)
-target_link_libraries(core PUBLIC Boost::boost PRIVATE fmt::fmt nlohmann_json::nlohmann_json mbedtls Opus::Opus unicorn)
+target_link_libraries(core PUBLIC Boost::boost PRIVATE fmt::fmt nlohmann_json::nlohmann_json mbedtls Opus::Opus unicorn zip)
 
 if (YUZU_ENABLE_BOXCAT)
     target_compile_definitions(core PRIVATE -DYUZU_ENABLE_BOXCAT)
-    target_link_libraries(core PRIVATE httplib nlohmann_json::nlohmann_json zip)
+    target_link_libraries(core PRIVATE httplib nlohmann_json::nlohmann_json)
 endif()
 
 if (ENABLE_WEB_SERVICE)

src/core/arm/dynarmic/arm_dynarmic_32.cpp

@@ -50,7 +50,8 @@ public:
     }
 
     void InterpreterFallback(u32 pc, std::size_t num_instructions) override {
-        UNIMPLEMENTED();
+        UNIMPLEMENTED_MSG("This should never happen, pc = {:08X}, code = {:08X}", pc,
+                          MemoryReadCode(pc));
     }
 
     void ExceptionRaised(u32 pc, Dynarmic::A32::Exception exception) override {
@@ -61,7 +62,7 @@ public:
         case Dynarmic::A32::Exception::Breakpoint:
             break;
         }
-        LOG_CRITICAL(HW_GPU, "ExceptionRaised(exception = {}, pc = {:08X}, code = {:08X})",
+        LOG_CRITICAL(Core_ARM, "ExceptionRaised(exception = {}, pc = {:08X}, code = {:08X})",
                      static_cast<std::size_t>(exception), pc, MemoryReadCode(pc));
         UNIMPLEMENTED();
     }
@@ -89,8 +90,6 @@ public:
 
     ARM_Dynarmic_32& parent;
     std::size_t num_interpreted_instructions{};
-    u64 tpidrro_el0{};
-    u64 tpidr_el0{};
 };
 
 std::shared_ptr<Dynarmic::A32::Jit> ARM_Dynarmic_32::MakeJit(Common::PageTable& page_table,
@@ -99,7 +98,7 @@ std::shared_ptr<Dynarmic::A32::Jit> ARM_Dynarmic_32::MakeJit(Common::PageTable&
     config.callbacks = cb.get();
     // TODO(bunnei): Implement page table for 32-bit
     // config.page_table = &page_table.pointers;
-    config.coprocessors[15] = std::make_shared<DynarmicCP15>((u32*)&CP15_regs[0]);
+    config.coprocessors[15] = cp15;
     config.define_unpredictable_behaviour = true;
     return std::make_unique<Dynarmic::A32::Jit>(config);
 }
@@ -112,13 +111,13 @@ void ARM_Dynarmic_32::Run() {
 }
 
 void ARM_Dynarmic_32::Step() {
-    cb->InterpreterFallback(jit->Regs()[15], 1);
+    jit->Step();
 }
 
 ARM_Dynarmic_32::ARM_Dynarmic_32(System& system, ExclusiveMonitor& exclusive_monitor,
                                  std::size_t core_index)
-    : ARM_Interface{system},
-      cb(std::make_unique<DynarmicCallbacks32>(*this)), core_index{core_index},
+    : ARM_Interface{system}, cb(std::make_unique<DynarmicCallbacks32>(*this)),
+      cp15(std::make_shared<DynarmicCP15>(*this)), core_index{core_index},
       exclusive_monitor{dynamic_cast<DynarmicExclusiveMonitor&>(exclusive_monitor)} {}
 
 ARM_Dynarmic_32::~ARM_Dynarmic_32() = default;
@@ -154,19 +153,19 @@ void ARM_Dynarmic_32::SetPSTATE(u32 cpsr) {
 }
 
 u64 ARM_Dynarmic_32::GetTlsAddress() const {
-    return CP15_regs[static_cast<std::size_t>(CP15Register::CP15_THREAD_URO)];
+    return cp15->uro;
 }
 
 void ARM_Dynarmic_32::SetTlsAddress(VAddr address) {
-    CP15_regs[static_cast<std::size_t>(CP15Register::CP15_THREAD_URO)] = static_cast<u32>(address);
+    cp15->uro = static_cast<u32>(address);
 }
 
 u64 ARM_Dynarmic_32::GetTPIDR_EL0() const {
-    return cb->tpidr_el0;
+    return cp15->uprw;
 }
 
 void ARM_Dynarmic_32::SetTPIDR_EL0(u64 value) {
-    cb->tpidr_el0 = value;
+    cp15->uprw = static_cast<u32>(value);
 }
 
 void ARM_Dynarmic_32::SaveContext(ThreadContext32& ctx) {

src/core/arm/dynarmic/arm_dynarmic_32.h

@@ -22,6 +22,7 @@ class Memory;
 namespace Core {
 
 class DynarmicCallbacks32;
+class DynarmicCP15;
 class DynarmicExclusiveMonitor;
 class System;
 
@@ -66,12 +67,14 @@ private:
         std::unordered_map<JitCacheKey, std::shared_ptr<Dynarmic::A32::Jit>, Common::PairHash>;
 
     friend class DynarmicCallbacks32;
+    friend class DynarmicCP15;
+
     std::unique_ptr<DynarmicCallbacks32> cb;
     JitCacheType jit_cache;
     std::shared_ptr<Dynarmic::A32::Jit> jit;
+    std::shared_ptr<DynarmicCP15> cp15;
     std::size_t core_index;
     DynarmicExclusiveMonitor& exclusive_monitor;
-    std::array<u32, 84> CP15_regs{};
 };
 
 } // namespace Core

src/core/arm/dynarmic/arm_dynarmic_64.cpp

@@ -98,8 +98,8 @@ public:
             }
             [[fallthrough]];
         default:
-            ASSERT_MSG(false, "ExceptionRaised(exception = {}, pc = {:X})",
-                       static_cast<std::size_t>(exception), pc);
+            ASSERT_MSG(false, "ExceptionRaised(exception = {}, pc = {:08X}, code = {:08X})",
+                       static_cast<std::size_t>(exception), pc, MemoryReadCode(pc));
         }
     }
 

src/core/arm/dynarmic/arm_dynarmic_cp15.cpp

@@ -2,79 +2,132 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <fmt/format.h>
+#include "common/logging/log.h"
+#include "core/arm/dynarmic/arm_dynarmic_32.h"
 #include "core/arm/dynarmic/arm_dynarmic_cp15.h"
+#include "core/core.h"
+#include "core/core_timing.h"
+#include "core/core_timing_util.h"
 
 using Callback = Dynarmic::A32::Coprocessor::Callback;
 using CallbackOrAccessOneWord = Dynarmic::A32::Coprocessor::CallbackOrAccessOneWord;
 using CallbackOrAccessTwoWords = Dynarmic::A32::Coprocessor::CallbackOrAccessTwoWords;
 
+template <>
+struct fmt::formatter<Dynarmic::A32::CoprocReg> {
+    constexpr auto parse(format_parse_context& ctx) {
+        return ctx.begin();
+    }
+    template <typename FormatContext>
+    auto format(const Dynarmic::A32::CoprocReg& reg, FormatContext& ctx) {
+        return format_to(ctx.out(), "cp{}", static_cast<size_t>(reg));
+    }
+};
+
+namespace Core {
+
+static u32 dummy_value;
+
 std::optional<Callback> DynarmicCP15::CompileInternalOperation(bool two, unsigned opc1,
                                                                CoprocReg CRd, CoprocReg CRn,
                                                                CoprocReg CRm, unsigned opc2) {
+    LOG_CRITICAL(Core_ARM, "CP15: cdp{} p15, {}, {}, {}, {}, {}", two ? "2" : "", opc1, CRd, CRn,
+                 CRm, opc2);
     return {};
 }
 
 CallbackOrAccessOneWord DynarmicCP15::CompileSendOneWord(bool two, unsigned opc1, CoprocReg CRn,
                                                          CoprocReg CRm, unsigned opc2) {
-    // TODO(merry): Privileged CP15 registers
-
     if (!two && CRn == CoprocReg::C7 && opc1 == 0 && CRm == CoprocReg::C5 && opc2 == 4) {
+        // CP15_FLUSH_PREFETCH_BUFFER
         // This is a dummy write, we ignore the value written here.
-        return &CP15[static_cast<std::size_t>(CP15Register::CP15_FLUSH_PREFETCH_BUFFER)];
+        return &dummy_value;
     }
 
     if (!two && CRn == CoprocReg::C7 && opc1 == 0 && CRm == CoprocReg::C10) {
         switch (opc2) {
         case 4:
+            // CP15_DATA_SYNC_BARRIER
             // This is a dummy write, we ignore the value written here.
-            return &CP15[static_cast<std::size_t>(CP15Register::CP15_DATA_SYNC_BARRIER)];
+            return &dummy_value;
         case 5:
+            // CP15_DATA_MEMORY_BARRIER
             // This is a dummy write, we ignore the value written here.
-            return &CP15[static_cast<std::size_t>(CP15Register::CP15_DATA_MEMORY_BARRIER)];
-        default:
-            return {};
+            return &dummy_value;
         }
     }
 
     if (!two && CRn == CoprocReg::C13 && opc1 == 0 && CRm == CoprocReg::C0 && opc2 == 2) {
-        return &CP15[static_cast<std::size_t>(CP15Register::CP15_THREAD_UPRW)];
+        // CP15_THREAD_UPRW
+        return &uprw;
     }
 
+    LOG_CRITICAL(Core_ARM, "CP15: mcr{} p15, {}, <Rt>, {}, {}, {}", two ? "2" : "", opc1, CRn, CRm,
+                 opc2);
     return {};
 }
 
 CallbackOrAccessTwoWords DynarmicCP15::CompileSendTwoWords(bool two, unsigned opc, CoprocReg CRm) {
+    LOG_CRITICAL(Core_ARM, "CP15: mcrr{} p15, {}, <Rt>, <Rt2>, {}", two ? "2" : "", opc, CRm);
     return {};
 }
 
 CallbackOrAccessOneWord DynarmicCP15::CompileGetOneWord(bool two, unsigned opc1, CoprocReg CRn,
                                                         CoprocReg CRm, unsigned opc2) {
-    // TODO(merry): Privileged CP15 registers
-
     if (!two && CRn == CoprocReg::C13 && opc1 == 0 && CRm == CoprocReg::C0) {
         switch (opc2) {
         case 2:
-            return &CP15[static_cast<std::size_t>(CP15Register::CP15_THREAD_UPRW)];
+            // CP15_THREAD_UPRW
+            return &uprw;
         case 3:
-            return &CP15[static_cast<std::size_t>(CP15Register::CP15_THREAD_URO)];
-        default:
-            return {};
+            // CP15_THREAD_URO
+            return &uro;
         }
     }
 
+    LOG_CRITICAL(Core_ARM, "CP15: mrc{} p15, {}, <Rt>, {}, {}, {}", two ? "2" : "", opc1, CRn, CRm,
+                 opc2);
     return {};
 }
 
 CallbackOrAccessTwoWords DynarmicCP15::CompileGetTwoWords(bool two, unsigned opc, CoprocReg CRm) {
+    if (!two && opc == 0 && CRm == CoprocReg::C14) {
+        // CNTPCT
+        const auto callback = static_cast<u64 (*)(Dynarmic::A32::Jit*, void*, u32, u32)>(
+            [](Dynarmic::A32::Jit*, void* arg, u32, u32) -> u64 {
+                ARM_Dynarmic_32& parent = *(ARM_Dynarmic_32*)arg;
+                return Timing::CpuCyclesToClockCycles(parent.system.CoreTiming().GetTicks());
+            });
+        return Dynarmic::A32::Coprocessor::Callback{callback, (void*)&parent};
+    }
+
+    LOG_CRITICAL(Core_ARM, "CP15: mrrc{} p15, {}, <Rt>, <Rt2>, {}", two ? "2" : "", opc, CRm);
     return {};
 }
 
 std::optional<Callback> DynarmicCP15::CompileLoadWords(bool two, bool long_transfer, CoprocReg CRd,
                                                        std::optional<u8> option) {
+    if (option) {
+        LOG_CRITICAL(Core_ARM, "CP15: mrrc{}{} p15, {}, [...], {}", two ? "2" : "",
+                     long_transfer ? "l" : "", CRd, *option);
+    } else {
+        LOG_CRITICAL(Core_ARM, "CP15: mrrc{}{} p15, {}, [...]", two ? "2" : "",
+                     long_transfer ? "l" : "", CRd);
+    }
     return {};
 }
 
 std::optional<Callback> DynarmicCP15::CompileStoreWords(bool two, bool long_transfer, CoprocReg CRd,
                                                         std::optional<u8> option) {
+    if (option) {
+        LOG_CRITICAL(Core_ARM, "CP15: mrrc{}{} p15, {}, [...], {}", two ? "2" : "",
+                     long_transfer ? "l" : "", CRd, *option);
+    } else {
+        LOG_CRITICAL(Core_ARM, "CP15: mrrc{}{} p15, {}, [...]", two ? "2" : "",
+                     long_transfer ? "l" : "", CRd);
+    }
     return {};
 }
+
+} // namespace Core

src/core/arm/dynarmic/arm_dynarmic_cp15.h

@@ -10,128 +10,15 @@
 #include <dynarmic/A32/coprocessor.h>
 #include "common/common_types.h"
 
-enum class CP15Register {
-    // c0 - Information registers
-    CP15_MAIN_ID,
-    CP15_CACHE_TYPE,
-    CP15_TCM_STATUS,
-    CP15_TLB_TYPE,
-    CP15_CPU_ID,
-    CP15_PROCESSOR_FEATURE_0,
-    CP15_PROCESSOR_FEATURE_1,
-    CP15_DEBUG_FEATURE_0,
-    CP15_AUXILIARY_FEATURE_0,
-    CP15_MEMORY_MODEL_FEATURE_0,
-    CP15_MEMORY_MODEL_FEATURE_1,
-    CP15_MEMORY_MODEL_FEATURE_2,
-    CP15_MEMORY_MODEL_FEATURE_3,
-    CP15_ISA_FEATURE_0,
-    CP15_ISA_FEATURE_1,
-    CP15_ISA_FEATURE_2,
-    CP15_ISA_FEATURE_3,
-    CP15_ISA_FEATURE_4,
+namespace Core {
 
-    // c1 - Control registers
-    CP15_CONTROL,
-    CP15_AUXILIARY_CONTROL,
-    CP15_COPROCESSOR_ACCESS_CONTROL,
-
-    // c2 - Translation table registers
-    CP15_TRANSLATION_BASE_TABLE_0,
-    CP15_TRANSLATION_BASE_TABLE_1,
-    CP15_TRANSLATION_BASE_CONTROL,
-    CP15_DOMAIN_ACCESS_CONTROL,
-    CP15_RESERVED,
-
-    // c5 - Fault status registers
-    CP15_FAULT_STATUS,
-    CP15_INSTR_FAULT_STATUS,
-    CP15_COMBINED_DATA_FSR = CP15_FAULT_STATUS,
-    CP15_INST_FSR,
-
-    // c6 - Fault Address registers
-    CP15_FAULT_ADDRESS,
-    CP15_COMBINED_DATA_FAR = CP15_FAULT_ADDRESS,
-    CP15_WFAR,
-    CP15_IFAR,
-
-    // c7 - Cache operation registers
-    CP15_WAIT_FOR_INTERRUPT,
-    CP15_PHYS_ADDRESS,
-    CP15_INVALIDATE_INSTR_CACHE,
-    CP15_INVALIDATE_INSTR_CACHE_USING_MVA,
-    CP15_INVALIDATE_INSTR_CACHE_USING_INDEX,
-    CP15_FLUSH_PREFETCH_BUFFER,
-    CP15_FLUSH_BRANCH_TARGET_CACHE,
-    CP15_FLUSH_BRANCH_TARGET_CACHE_ENTRY,
-    CP15_INVALIDATE_DATA_CACHE,
-    CP15_INVALIDATE_DATA_CACHE_LINE_USING_MVA,
-    CP15_INVALIDATE_DATA_CACHE_LINE_USING_INDEX,
-    CP15_INVALIDATE_DATA_AND_INSTR_CACHE,
-    CP15_CLEAN_DATA_CACHE,
-    CP15_CLEAN_DATA_CACHE_LINE_USING_MVA,
-    CP15_CLEAN_DATA_CACHE_LINE_USING_INDEX,
-    CP15_DATA_SYNC_BARRIER,
-    CP15_DATA_MEMORY_BARRIER,
-    CP15_CLEAN_AND_INVALIDATE_DATA_CACHE,
-    CP15_CLEAN_AND_INVALIDATE_DATA_CACHE_LINE_USING_MVA,
-    CP15_CLEAN_AND_INVALIDATE_DATA_CACHE_LINE_USING_INDEX,
-
-    // c8 - TLB operations
-    CP15_INVALIDATE_ITLB,
-    CP15_INVALIDATE_ITLB_SINGLE_ENTRY,
-    CP15_INVALIDATE_ITLB_ENTRY_ON_ASID_MATCH,
-    CP15_INVALIDATE_ITLB_ENTRY_ON_MVA,
-    CP15_INVALIDATE_DTLB,
-    CP15_INVALIDATE_DTLB_SINGLE_ENTRY,
-    CP15_INVALIDATE_DTLB_ENTRY_ON_ASID_MATCH,
-    CP15_INVALIDATE_DTLB_ENTRY_ON_MVA,
-    CP15_INVALIDATE_UTLB,
-    CP15_INVALIDATE_UTLB_SINGLE_ENTRY,
-    CP15_INVALIDATE_UTLB_ENTRY_ON_ASID_MATCH,
-    CP15_INVALIDATE_UTLB_ENTRY_ON_MVA,
-
-    // c9 - Data cache lockdown register
-    CP15_DATA_CACHE_LOCKDOWN,
-
-    // c10 - TLB/Memory map registers
-    CP15_TLB_LOCKDOWN,
-    CP15_PRIMARY_REGION_REMAP,
-    CP15_NORMAL_REGION_REMAP,
-
-    // c13 - Thread related registers
-    CP15_PID,
-    CP15_CONTEXT_ID,
-    CP15_THREAD_UPRW, // Thread ID register - User/Privileged Read/Write
-    CP15_THREAD_URO,  // Thread ID register - User Read Only (Privileged R/W)
-    CP15_THREAD_PRW,  // Thread ID register - Privileged R/W only.
-
-    // c15 - Performance and TLB lockdown registers
-    CP15_PERFORMANCE_MONITOR_CONTROL,
-    CP15_CYCLE_COUNTER,
-    CP15_COUNT_0,
-    CP15_COUNT_1,
-    CP15_READ_MAIN_TLB_LOCKDOWN_ENTRY,
-    CP15_WRITE_MAIN_TLB_LOCKDOWN_ENTRY,
-    CP15_MAIN_TLB_LOCKDOWN_VIRT_ADDRESS,
-    CP15_MAIN_TLB_LOCKDOWN_PHYS_ADDRESS,
-    CP15_MAIN_TLB_LOCKDOWN_ATTRIBUTE,
-    CP15_TLB_DEBUG_CONTROL,
-
-    // Skyeye defined
-    CP15_TLB_FAULT_ADDR,
-    CP15_TLB_FAULT_STATUS,
-
-    // Not an actual register.
-    // All registers should be defined above this.
-    CP15_REGISTER_COUNT,
-};
+class ARM_Dynarmic_32;
 
 class DynarmicCP15 final : public Dynarmic::A32::Coprocessor {
 public:
     using CoprocReg = Dynarmic::A32::CoprocReg;
 
-    explicit DynarmicCP15(u32* cp15) : CP15(cp15){};
+    explicit DynarmicCP15(ARM_Dynarmic_32& parent) : parent(parent) {}
 
     std::optional<Callback> CompileInternalOperation(bool two, unsigned opc1, CoprocReg CRd,
                                                      CoprocReg CRn, CoprocReg CRm,
@@ -147,6 +34,9 @@ public:
     std::optional<Callback> CompileStoreWords(bool two, bool long_transfer, CoprocReg CRd,
                                               std::optional<u8> option) override;
 
-private:
-    u32* CP15{};
+    ARM_Dynarmic_32& parent;
+    u32 uprw;
+    u32 uro;
 };
+
+} // namespace Core

src/core/file_sys/system_archive/mii_model.cpp

@@ -40,7 +40,7 @@ VirtualDir MiiModel() {
     out->AddFile(std::make_shared<ArrayVfsFile<MiiModelData::SHAPE_MID.size()>>(
         MiiModelData::SHAPE_MID, "ShapeMid.dat"));
 
-    return std::move(out);
+    return out;
 }
 
 } // namespace FileSys::SystemArchive

src/core/file_sys/system_archive/shared_font.cpp

@@ -23,7 +23,7 @@ VirtualFile PackBFTTF(const std::array<u8, Size>& data, const std::string& name)
 
     std::vector<u8> bfttf(Size + sizeof(u64));
 
-    u64 offset = 0;
+    size_t offset = 0;
     Service::NS::EncryptSharedFont(vec, bfttf, offset);
     return std::make_shared<VectorVfsFile>(std::move(bfttf), name);
 }

src/core/hle/kernel/memory/memory_manager.cpp

@@ -104,7 +104,7 @@ ResultCode MemoryManager::Allocate(PageLinkedList& page_list, std::size_t num_pa
     // Ensure that we don't leave anything un-freed
     auto group_guard = detail::ScopeExit([&] {
         for (const auto& it : page_list.Nodes()) {
-            const auto min_num_pages{std::min(
+            const auto min_num_pages{std::min<size_t>(
                 it.GetNumPages(), (chosen_manager.GetEndAddress() - it.GetAddress()) / PageSize)};
             chosen_manager.Free(it.GetAddress(), min_num_pages);
         }
@@ -165,7 +165,7 @@ ResultCode MemoryManager::Free(PageLinkedList& page_list, std::size_t num_pages,
 
     // Free all of the pages
     for (const auto& it : page_list.Nodes()) {
-        const auto min_num_pages{std::min(
+        const auto min_num_pages{std::min<size_t>(
             it.GetNumPages(), (chosen_manager.GetEndAddress() - it.GetAddress()) / PageSize)};
         chosen_manager.Free(it.GetAddress(), min_num_pages);
     }

src/core/hle/kernel/process.cpp

@@ -132,7 +132,8 @@ std::shared_ptr<ResourceLimit> Process::GetResourceLimit() const {
 
 u64 Process::GetTotalPhysicalMemoryAvailable() const {
     const u64 capacity{resource_limit->GetCurrentResourceValue(ResourceType::PhysicalMemory) +
-                       page_table->GetTotalHeapSize() + image_size + main_thread_stack_size};
+                       page_table->GetTotalHeapSize() + GetSystemResourceSize() + image_size +
+                       main_thread_stack_size};
 
     if (capacity < memory_usage_capacity) {
         return capacity;
@@ -146,7 +147,8 @@ u64 Process::GetTotalPhysicalMemoryAvailableWithoutSystemResource() const {
 }
 
 u64 Process::GetTotalPhysicalMemoryUsed() const {
-    return image_size + main_thread_stack_size + page_table->GetTotalHeapSize();
+    return image_size + main_thread_stack_size + page_table->GetTotalHeapSize() +
+           GetSystemResourceSize();
 }
 
 u64 Process::GetTotalPhysicalMemoryUsedWithoutSystemResource() const {

src/core/hle/kernel/readable_event.cpp

@@ -38,7 +38,7 @@ void ReadableEvent::Clear() {
 
 ResultCode ReadableEvent::Reset() {
     if (!is_signaled) {
-        LOG_ERROR(Kernel, "Handle is not signaled! object_id={}, object_type={}, object_name={}",
+        LOG_TRACE(Kernel, "Handle is not signaled! object_id={}, object_type={}, object_name={}",
                   GetObjectId(), GetTypeName(), GetName());
         return ERR_INVALID_STATE;
     }

src/core/hle/kernel/resource_limit.cpp

@@ -24,13 +24,9 @@ bool ResourceLimit::Reserve(ResourceType resource, s64 amount, u64 timeout) {
     const std::size_t index{ResourceTypeToIndex(resource)};
 
     s64 new_value = current[index] + amount;
-    while (new_value > limit[index] && available[index] + amount <= limit[index]) {
+    if (new_value > limit[index] && available[index] + amount <= limit[index]) {
         // TODO(bunnei): This is wrong for multicore, we should wait the calling thread for timeout
         new_value = current[index] + amount;
-
-        if (timeout >= 0) {
-            break;
-        }
     }
 
     if (new_value <= limit[index]) {

src/core/hle/service/acc/acc.cpp

@@ -44,6 +44,218 @@ static constexpr u32 SanitizeJPEGSize(std::size_t size) {
     return static_cast<u32>(std::min(size, max_jpeg_image_size));
 }
 
+class IManagerForSystemService final : public ServiceFramework<IManagerForSystemService> {
+public:
+    explicit IManagerForSystemService(Common::UUID user_id)
+        : ServiceFramework("IManagerForSystemService") {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {0, nullptr, "CheckAvailability"},
+            {1, nullptr, "GetAccountId"},
+            {2, nullptr, "EnsureIdTokenCacheAsync"},
+            {3, nullptr, "LoadIdTokenCache"},
+            {100, nullptr, "SetSystemProgramIdentification"},
+            {101, nullptr, "RefreshNotificationTokenAsync"}, // 7.0.0+
+            {110, nullptr, "GetServiceEntryRequirementCache"}, // 4.0.0+
+            {111, nullptr, "InvalidateServiceEntryRequirementCache"}, // 4.0.0+
+            {112, nullptr, "InvalidateTokenCache"}, // 4.0.0 - 6.2.0
+            {113, nullptr, "GetServiceEntryRequirementCacheForOnlinePlay"}, // 6.1.0+
+            {120, nullptr, "GetNintendoAccountId"},
+            {121, nullptr, "CalculateNintendoAccountAuthenticationFingerprint"}, // 9.0.0+
+            {130, nullptr, "GetNintendoAccountUserResourceCache"},
+            {131, nullptr, "RefreshNintendoAccountUserResourceCacheAsync"},
+            {132, nullptr, "RefreshNintendoAccountUserResourceCacheAsyncIfSecondsElapsed"},
+            {133, nullptr, "GetNintendoAccountVerificationUrlCache"}, // 9.0.0+
+            {134, nullptr, "RefreshNintendoAccountVerificationUrlCache"}, // 9.0.0+
+            {135, nullptr, "RefreshNintendoAccountVerificationUrlCacheAsyncIfSecondsElapsed"}, // 9.0.0+
+            {140, nullptr, "GetNetworkServiceLicenseCache"}, // 5.0.0+
+            {141, nullptr, "RefreshNetworkServiceLicenseCacheAsync"}, // 5.0.0+
+            {142, nullptr, "RefreshNetworkServiceLicenseCacheAsyncIfSecondsElapsed"}, // 5.0.0+
+            {150, nullptr, "CreateAuthorizationRequest"},
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+};
+
+// 3.0.0+
+class IFloatingRegistrationRequest final : public ServiceFramework<IFloatingRegistrationRequest> {
+public:
+    explicit IFloatingRegistrationRequest(Common::UUID user_id)
+        : ServiceFramework("IFloatingRegistrationRequest") {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {0, nullptr, "GetSessionId"},
+            {12, nullptr, "GetAccountId"},
+            {13, nullptr, "GetLinkedNintendoAccountId"},
+            {14, nullptr, "GetNickname"},
+            {15, nullptr, "GetProfileImage"},
+            {21, nullptr, "LoadIdTokenCache"},
+            {100, nullptr, "RegisterUser"}, // [1.0.0-3.0.2] RegisterAsync
+            {101, nullptr, "RegisterUserWithUid"}, // [1.0.0-3.0.2] RegisterWithUidAsync
+            {102, nullptr, "RegisterNetworkServiceAccountAsync"}, // 4.0.0+
+            {103, nullptr, "RegisterNetworkServiceAccountWithUidAsync"}, // 4.0.0+
+            {110, nullptr, "SetSystemProgramIdentification"},
+            {111, nullptr, "EnsureIdTokenCacheAsync"},
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+};
+
+class IAdministrator final : public ServiceFramework<IAdministrator> {
+public:
+    explicit IAdministrator(Common::UUID user_id) : ServiceFramework("IAdministrator") {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {0, nullptr, "CheckAvailability"},
+            {1, nullptr, "GetAccountId"},
+            {2, nullptr, "EnsureIdTokenCacheAsync"},
+            {3, nullptr, "LoadIdTokenCache"},
+            {100, nullptr, "SetSystemProgramIdentification"},
+            {101, nullptr, "RefreshNotificationTokenAsync"}, // 7.0.0+
+            {110, nullptr, "GetServiceEntryRequirementCache"}, // 4.0.0+
+            {111, nullptr, "InvalidateServiceEntryRequirementCache"}, // 4.0.0+
+            {112, nullptr, "InvalidateTokenCache"}, // 4.0.0 - 6.2.0
+            {113, nullptr, "GetServiceEntryRequirementCacheForOnlinePlay"}, // 6.1.0+
+            {120, nullptr, "GetNintendoAccountId"},
+            {121, nullptr, "CalculateNintendoAccountAuthenticationFingerprint"}, // 9.0.0+
+            {130, nullptr, "GetNintendoAccountUserResourceCache"},
+            {131, nullptr, "RefreshNintendoAccountUserResourceCacheAsync"},
+            {132, nullptr, "RefreshNintendoAccountUserResourceCacheAsyncIfSecondsElapsed"},
+            {133, nullptr, "GetNintendoAccountVerificationUrlCache"}, // 9.0.0+
+            {134, nullptr, "RefreshNintendoAccountVerificationUrlCacheAsync"}, // 9.0.0+
+            {135, nullptr, "RefreshNintendoAccountVerificationUrlCacheAsyncIfSecondsElapsed"}, // 9.0.0+
+            {140, nullptr, "GetNetworkServiceLicenseCache"}, // 5.0.0+
+            {141, nullptr, "RefreshNetworkServiceLicenseCacheAsync"}, // 5.0.0+
+            {142, nullptr, "RefreshNetworkServiceLicenseCacheAsyncIfSecondsElapsed"}, // 5.0.0+
+            {150, nullptr, "CreateAuthorizationRequest"},
+            {200, nullptr, "IsRegistered"},
+            {201, nullptr, "RegisterAsync"},
+            {202, nullptr, "UnregisterAsync"},
+            {203, nullptr, "DeleteRegistrationInfoLocally"},
+            {220, nullptr, "SynchronizeProfileAsync"},
+            {221, nullptr, "UploadProfileAsync"},
+            {222, nullptr, "SynchronizaProfileAsyncIfSecondsElapsed"},
+            {250, nullptr, "IsLinkedWithNintendoAccount"},
+            {251, nullptr, "CreateProcedureToLinkWithNintendoAccount"},
+            {252, nullptr, "ResumeProcedureToLinkWithNintendoAccount"},
+            {255, nullptr, "CreateProcedureToUpdateLinkageStateOfNintendoAccount"},
+            {256, nullptr, "ResumeProcedureToUpdateLinkageStateOfNintendoAccount"},
+            {260, nullptr, "CreateProcedureToLinkNnidWithNintendoAccount"}, // 3.0.0+
+            {261, nullptr, "ResumeProcedureToLinkNnidWithNintendoAccount"}, // 3.0.0+
+            {280, nullptr, "ProxyProcedureToAcquireApplicationAuthorizationForNintendoAccount"},
+            {290, nullptr, "GetRequestForNintendoAccountUserResourceView"}, // 8.0.0+
+            {300, nullptr, "TryRecoverNintendoAccountUserStateAsync"}, // 6.0.0+
+            {400, nullptr, "IsServiceEntryRequirementCacheRefreshRequiredForOnlinePlay"}, // 6.1.0+
+            {401, nullptr, "RefreshServiceEntryRequirementCacheForOnlinePlayAsync"}, // 6.1.0+
+            {900, nullptr, "GetAuthenticationInfoForWin"}, // 9.0.0+
+            {901, nullptr, "ImportAsyncForWin"}, // 9.0.0+
+            {997, nullptr, "DebugUnlinkNintendoAccountAsync"},
+            {998, nullptr, "DebugSetAvailabilityErrorDetail"},
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+};
+
+class IAuthorizationRequest final : public ServiceFramework<IAuthorizationRequest> {
+public:
+    explicit IAuthorizationRequest(Common::UUID user_id)
+        : ServiceFramework("IAuthorizationRequest") {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {0, nullptr, "GetSessionId"},
+            {10, nullptr, "InvokeWithoutInteractionAsync"},
+            {19, nullptr, "IsAuthorized"},
+            {20, nullptr, "GetAuthorizationCode"},
+            {21, nullptr, "GetIdToken"},
+            {22, nullptr, "GetState"},
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+};
+
+class IOAuthProcedure final : public ServiceFramework<IOAuthProcedure> {
+public:
+    explicit IOAuthProcedure(Common::UUID user_id) : ServiceFramework("IOAuthProcedure") {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {0, nullptr, "PrepareAsync"},
+            {1, nullptr, "GetRequest"},
+            {2, nullptr, "ApplyResponse"},
+            {3, nullptr, "ApplyResponseAsync"},
+            {10, nullptr, "Suspend"},
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+};
+
+// 3.0.0+
+class IOAuthProcedureForExternalNsa final : public ServiceFramework<IOAuthProcedureForExternalNsa> {
+public:
+    explicit IOAuthProcedureForExternalNsa(Common::UUID user_id)
+        : ServiceFramework("IOAuthProcedureForExternalNsa") {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {0, nullptr, "PrepareAsync"},
+            {1, nullptr, "GetRequest"},
+            {2, nullptr, "ApplyResponse"},
+            {3, nullptr, "ApplyResponseAsync"},
+            {10, nullptr, "Suspend"},
+            {100, nullptr, "GetAccountId"},
+            {101, nullptr, "GetLinkedNintendoAccountId"},
+            {102, nullptr, "GetNickname"},
+            {103, nullptr, "GetProfileImage"},
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+};
+
+class IOAuthProcedureForNintendoAccountLinkage final
+    : public ServiceFramework<IOAuthProcedureForNintendoAccountLinkage> {
+public:
+    explicit IOAuthProcedureForNintendoAccountLinkage(Common::UUID user_id)
+        : ServiceFramework("IOAuthProcedureForNintendoAccountLinkage") {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {0, nullptr, "PrepareAsync"},
+            {1, nullptr, "GetRequest"},
+            {2, nullptr, "ApplyResponse"},
+            {3, nullptr, "ApplyResponseAsync"},
+            {10, nullptr, "Suspend"},
+            {100, nullptr, "GetRequestWithTheme"},
+            {101, nullptr, "IsNetworkServiceAccountReplaced"},
+            {199, nullptr, "GetUrlForIntroductionOfExtraMembership"}, // 2.0.0 - 5.1.0
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+};
+
+class INotifier final : public ServiceFramework<INotifier> {
+public:
+    explicit INotifier(Common::UUID user_id) : ServiceFramework("INotifier") {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {0, nullptr, "GetSystemEvent"},
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+};
+
 class IProfileCommon : public ServiceFramework<IProfileCommon> {
 public:
     explicit IProfileCommon(const char* name, bool editor_commands, Common::UUID user_id,
@@ -226,6 +438,54 @@ public:
         : IProfileCommon("IProfileEditor", true, user_id, profile_manager) {}
 };
 
+class IAsyncContext final : public ServiceFramework<IAsyncContext> {
+public:
+    explicit IAsyncContext(Common::UUID user_id) : ServiceFramework("IAsyncContext") {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {0, nullptr, "GetSystemEvent"},
+            {1, nullptr, "Cancel"},
+            {2, nullptr, "HasDone"},
+            {3, nullptr, "GetResult"},
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+};
+
+class ISessionObject final : public ServiceFramework<ISessionObject> {
+public:
+    explicit ISessionObject(Common::UUID user_id) : ServiceFramework("ISessionObject") {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {999, nullptr, "Dummy"},
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+};
+
+class IGuestLoginRequest final : public ServiceFramework<IGuestLoginRequest> {
+public:
+    explicit IGuestLoginRequest(Common::UUID) : ServiceFramework("IGuestLoginRequest") {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {0, nullptr, "GetSessionId"},
+            {11, nullptr, "Unknown"}, // 1.0.0 - 2.3.0 (the name is blank on Switchbrew)
+            {12, nullptr, "GetAccountId"},
+            {13, nullptr, "GetLinkedNintendoAccountId"},
+            {14, nullptr, "GetNickname"},
+            {15, nullptr, "GetProfileImage"},
+            {21, nullptr, "LoadIdTokenCache"}, // 3.0.0+
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+};
+
 class IManagerForApplication final : public ServiceFramework<IManagerForApplication> {
 public:
     explicit IManagerForApplication(Common::UUID user_id)
@@ -265,6 +525,87 @@ private:
     Common::UUID user_id;
 };
 
+// 6.0.0+
+class IAsyncNetworkServiceLicenseKindContext final
+    : public ServiceFramework<IAsyncNetworkServiceLicenseKindContext> {
+public:
+    explicit IAsyncNetworkServiceLicenseKindContext(Common::UUID user_id)
+        : ServiceFramework("IAsyncNetworkServiceLicenseKindContext") {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {0, nullptr, "GetSystemEvent"},
+            {1, nullptr, "Cancel"},
+            {2, nullptr, "HasDone"},
+            {3, nullptr, "GetResult"},
+            {4, nullptr, "GetNetworkServiceLicenseKind"},
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+};
+
+// 8.0.0+
+class IOAuthProcedureForUserRegistration final
+    : public ServiceFramework<IOAuthProcedureForUserRegistration> {
+public:
+    explicit IOAuthProcedureForUserRegistration(Common::UUID user_id)
+        : ServiceFramework("IOAuthProcedureForUserRegistration") {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {0, nullptr, "PrepareAsync"},
+            {1, nullptr, "GetRequest"},
+            {2, nullptr, "ApplyResponse"},
+            {3, nullptr, "ApplyResponseAsync"},
+            {10, nullptr, "Suspend"},
+            {100, nullptr, "GetAccountId"},
+            {101, nullptr, "GetLinkedNintendoAccountId"},
+            {102, nullptr, "GetNickname"},
+            {103, nullptr, "GetProfileImage"},
+            {110, nullptr, "RegisterUserAsync"},
+            {111, nullptr, "GetUid"},
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+};
+
+class DAUTH_O final : public ServiceFramework<DAUTH_O> {
+public:
+    explicit DAUTH_O(Common::UUID) : ServiceFramework("dauth:o") {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {0, nullptr, "EnsureAuthenticationTokenCacheAsync"}, // [5.0.0-5.1.0] GeneratePostData
+            {1, nullptr, "LoadAuthenticationTokenCache"}, // 6.0.0+
+            {2, nullptr, "InvalidateAuthenticationTokenCache"}, // 6.0.0+
+            {10, nullptr, "EnsureEdgeTokenCacheAsync"}, // 6.0.0+
+            {11, nullptr, "LoadEdgeTokenCache"}, // 6.0.0+
+            {12, nullptr, "InvalidateEdgeTokenCache"}, // 6.0.0+
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+};
+
+// 6.0.0+
+class IAsyncResult final : public ServiceFramework<IAsyncResult> {
+public:
+    explicit IAsyncResult(Common::UUID user_id) : ServiceFramework("IAsyncResult") {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {0, nullptr, "GetResult"},
+            {1, nullptr, "Cancel"},
+            {2, nullptr, "IsAvailable"},
+            {3, nullptr, "GetSystemEvent"},
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+};
+
 void Module::Interface::GetUserCount(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service_ACC, "called");
     IPC::ResponseBuilder rb{ctx, 3};

src/core/hle/service/acc/acc_aa.cpp

@@ -13,8 +13,8 @@ ACC_AA::ACC_AA(std::shared_ptr<Module> module, std::shared_ptr<ProfileManager> p
         {0, nullptr, "EnsureCacheAsync"},
         {1, nullptr, "LoadCache"},
         {2, nullptr, "GetDeviceAccountId"},
-        {50, nullptr, "RegisterNotificationTokenAsync"},
-        {51, nullptr, "UnregisterNotificationTokenAsync"},
+        {50, nullptr, "RegisterNotificationTokenAsync"},   // 1.0.0 - 6.2.0
+        {51, nullptr, "UnregisterNotificationTokenAsync"}, // 1.0.0 - 6.2.0
     };
     RegisterHandlers(functions);
 }

src/core/hle/service/acc/acc_su.cpp

@@ -17,28 +17,28 @@ ACC_SU::ACC_SU(std::shared_ptr<Module> module, std::shared_ptr<ProfileManager> p
         {3, &ACC_SU::ListOpenUsers, "ListOpenUsers"},
         {4, &ACC_SU::GetLastOpenedUser, "GetLastOpenedUser"},
         {5, &ACC_SU::GetProfile, "GetProfile"},
-        {6, nullptr, "GetProfileDigest"},
+        {6, nullptr, "GetProfileDigest"}, // 3.0.0+
         {50, &ACC_SU::IsUserRegistrationRequestPermitted, "IsUserRegistrationRequestPermitted"},
         {51, &ACC_SU::TrySelectUserWithoutInteraction, "TrySelectUserWithoutInteraction"},
-        {60, nullptr, "ListOpenContextStoredUsers"},
-        {99, nullptr, "DebugActivateOpenContextRetention"},
+        {60, nullptr, "ListOpenContextStoredUsers"}, // 5.0.0 - 5.1.0
+        {99, nullptr, "DebugActivateOpenContextRetention"}, // 6.0.0+
         {100, nullptr, "GetUserRegistrationNotifier"},
         {101, nullptr, "GetUserStateChangeNotifier"},
         {102, nullptr, "GetBaasAccountManagerForSystemService"},
         {103, nullptr, "GetBaasUserAvailabilityChangeNotifier"},
         {104, nullptr, "GetProfileUpdateNotifier"},
-        {105, nullptr, "CheckNetworkServiceAvailabilityAsync"},
-        {106, nullptr, "GetProfileSyncNotifier"},
+        {105, nullptr, "CheckNetworkServiceAvailabilityAsync"}, // 4.0.0+
+        {106, nullptr, "GetProfileSyncNotifier"}, // 9.0.0+
         {110, nullptr, "StoreSaveDataThumbnail"},
         {111, nullptr, "ClearSaveDataThumbnail"},
         {112, nullptr, "LoadSaveDataThumbnail"},
-        {113, nullptr, "GetSaveDataThumbnailExistence"},
-        {120, nullptr, "ListOpenUsersInApplication"},
-        {130, nullptr, "ActivateOpenContextRetention"},
-        {140, &ACC_SU::ListQualifiedUsers, "ListQualifiedUsers"},
-        {150, nullptr, "AuthenticateApplicationAsync"},
-        {190, nullptr, "GetUserLastOpenedApplication"},
-        {191, nullptr, "ActivateOpenContextHolder"},
+        {113, nullptr, "GetSaveDataThumbnailExistence"}, // 5.0.0+
+        {120, nullptr, "ListOpenUsersInApplication"}, // 10.0.0+
+        {130, nullptr, "ActivateOpenContextRetention"}, // 6.0.0+
+        {140, &ACC_SU::ListQualifiedUsers, "ListQualifiedUsers"}, // 6.0.0+
+        {150, nullptr, "AuthenticateApplicationAsync"}, // 10.0.0+
+        {190, nullptr, "GetUserLastOpenedApplication"}, // 1.0.0 - 9.2.0
+        {191, nullptr, "ActivateOpenContextHolder"}, // 7.0.0+
         {200, nullptr, "BeginUserRegistration"},
         {201, nullptr, "CompleteUserRegistration"},
         {202, nullptr, "CancelUserRegistration"},
@@ -46,15 +46,15 @@ ACC_SU::ACC_SU(std::shared_ptr<Module> module, std::shared_ptr<ProfileManager> p
         {204, nullptr, "SetUserPosition"},
         {205, &ACC_SU::GetProfileEditor, "GetProfileEditor"},
         {206, nullptr, "CompleteUserRegistrationForcibly"},
-        {210, nullptr, "CreateFloatingRegistrationRequest"},
-        {211, nullptr, "CreateProcedureToRegisterUserWithNintendoAccount"},
-        {212, nullptr, "ResumeProcedureToRegisterUserWithNintendoAccount"},
+        {210, nullptr, "CreateFloatingRegistrationRequest"}, // 3.0.0+
+        {211, nullptr, "CreateProcedureToRegisterUserWithNintendoAccount"}, // 8.0.0+
+        {212, nullptr, "ResumeProcedureToRegisterUserWithNintendoAccount"}, // 8.0.0+
         {230, nullptr, "AuthenticateServiceAsync"},
         {250, nullptr, "GetBaasAccountAdministrator"},
         {290, nullptr, "ProxyProcedureForGuestLoginWithNintendoAccount"},
-        {291, nullptr, "ProxyProcedureForFloatingRegistrationWithNintendoAccount"},
+        {291, nullptr, "ProxyProcedureForFloatingRegistrationWithNintendoAccount"}, // 3.0.0+
         {299, nullptr, "SuspendBackgroundDaemon"},
-        {997, nullptr, "DebugInvalidateTokenCacheForUser"},
+        {997, nullptr, "DebugInvalidateTokenCacheForUser"}, // 3.0.0+
         {998, nullptr, "DebugSetUserStateClose"},
         {999, nullptr, "DebugSetUserStateOpen"},
     };

src/core/hle/service/acc/acc_u0.cpp

@@ -17,23 +17,23 @@ ACC_U0::ACC_U0(std::shared_ptr<Module> module, std::shared_ptr<ProfileManager> p
         {3, &ACC_U0::ListOpenUsers, "ListOpenUsers"},
         {4, &ACC_U0::GetLastOpenedUser, "GetLastOpenedUser"},
         {5, &ACC_U0::GetProfile, "GetProfile"},
-        {6, nullptr, "GetProfileDigest"},
+        {6, nullptr, "GetProfileDigest"}, // 3.0.0+
         {50, &ACC_U0::IsUserRegistrationRequestPermitted, "IsUserRegistrationRequestPermitted"},
         {51, &ACC_U0::TrySelectUserWithoutInteraction, "TrySelectUserWithoutInteraction"},
-        {60, nullptr, "ListOpenContextStoredUsers"},
-        {99, nullptr, "DebugActivateOpenContextRetention"},
+        {60, nullptr, "ListOpenContextStoredUsers"}, // 5.0.0 - 5.1.0
+        {99, nullptr, "DebugActivateOpenContextRetention"}, // 6.0.0+
         {100, &ACC_U0::InitializeApplicationInfo, "InitializeApplicationInfo"},
         {101, &ACC_U0::GetBaasAccountManagerForApplication, "GetBaasAccountManagerForApplication"},
         {102, nullptr, "AuthenticateApplicationAsync"},
-        {103, nullptr, "CheckNetworkServiceAvailabilityAsync"},
+        {103, nullptr, "CheckNetworkServiceAvailabilityAsync"}, // 4.0.0+
         {110, nullptr, "StoreSaveDataThumbnail"},
         {111, nullptr, "ClearSaveDataThumbnail"},
         {120, nullptr, "CreateGuestLoginRequest"},
-        {130, nullptr, "LoadOpenContext"},
-        {131, nullptr, "ListOpenContextStoredUsers"},
-        {140, &ACC_U0::InitializeApplicationInfoRestricted, "InitializeApplicationInfoRestricted"},
-        {141, &ACC_U0::ListQualifiedUsers, "ListQualifiedUsers"},
-        {150, &ACC_U0::IsUserAccountSwitchLocked, "IsUserAccountSwitchLocked"},
+        {130, nullptr, "LoadOpenContext"}, // 5.0.0+
+        {131, nullptr, "ListOpenContextStoredUsers"}, // 6.0.0+
+        {140, &ACC_U0::InitializeApplicationInfoRestricted, "InitializeApplicationInfoRestricted"}, // 6.0.0+
+        {141, &ACC_U0::ListQualifiedUsers, "ListQualifiedUsers"}, // 6.0.0+
+        {150, &ACC_U0::IsUserAccountSwitchLocked, "IsUserAccountSwitchLocked"}, // 6.0.0+
     };
     // clang-format on
 

src/core/hle/service/acc/acc_u1.cpp

@@ -17,28 +17,29 @@ ACC_U1::ACC_U1(std::shared_ptr<Module> module, std::shared_ptr<ProfileManager> p
         {3, &ACC_U1::ListOpenUsers, "ListOpenUsers"},
         {4, &ACC_U1::GetLastOpenedUser, "GetLastOpenedUser"},
         {5, &ACC_U1::GetProfile, "GetProfile"},
-        {6, nullptr, "GetProfileDigest"},
+        {6, nullptr, "GetProfileDigest"}, // 3.0.0+
         {50, &ACC_U1::IsUserRegistrationRequestPermitted, "IsUserRegistrationRequestPermitted"},
         {51, &ACC_U1::TrySelectUserWithoutInteraction, "TrySelectUserWithoutInteraction"},
-        {60, nullptr, "ListOpenContextStoredUsers"},
-        {99, nullptr, "DebugActivateOpenContextRetention"},
+        {60, nullptr, "ListOpenContextStoredUsers"}, // 5.0.0 - 5.1.0
+        {99, nullptr, "DebugActivateOpenContextRetention"}, // 6.0.0+
         {100, nullptr, "GetUserRegistrationNotifier"},
         {101, nullptr, "GetUserStateChangeNotifier"},
         {102, nullptr, "GetBaasAccountManagerForSystemService"},
-        {103, nullptr, "GetProfileUpdateNotifier"},
-        {104, nullptr, "CheckNetworkServiceAvailabilityAsync"},
-        {105, nullptr, "GetBaasUserAvailabilityChangeNotifier"},
-        {106, nullptr, "GetProfileSyncNotifier"},
+        {103, nullptr, "GetBaasUserAvailabilityChangeNotifier"},
+        {104, nullptr, "GetProfileUpdateNotifier"},
+        {105, nullptr, "CheckNetworkServiceAvailabilityAsync"}, // 4.0.0+
+        {106, nullptr, "GetProfileSyncNotifier"}, // 9.0.0+
         {110, nullptr, "StoreSaveDataThumbnail"},
         {111, nullptr, "ClearSaveDataThumbnail"},
         {112, nullptr, "LoadSaveDataThumbnail"},
-        {113, nullptr, "GetSaveDataThumbnailExistence"},
-        {130, nullptr, "ActivateOpenContextRetention"},
-        {140, &ACC_U1::ListQualifiedUsers, "ListQualifiedUsers"},
-        {150, nullptr, "AuthenticateApplicationAsync"},
-        {190, nullptr, "GetUserLastOpenedApplication"},
-        {191, nullptr, "ActivateOpenContextHolder"},
-        {997, nullptr, "DebugInvalidateTokenCacheForUser"},
+        {113, nullptr, "GetSaveDataThumbnailExistence"}, // 5.0.0+
+        {120, nullptr, "ListOpenUsersInApplication"}, // 10.0.0+
+        {130, nullptr, "ActivateOpenContextRetention"}, // 6.0.0+
+        {140, &ACC_U1::ListQualifiedUsers, "ListQualifiedUsers"}, // 6.0.0+
+        {150, nullptr, "AuthenticateApplicationAsync"}, // 10.0.0+
+        {190, nullptr, "GetUserLastOpenedApplication"}, // 1.0.0 - 9.2.0
+        {191, nullptr, "ActivateOpenContextHolder"}, // 7.0.0+
+        {997, nullptr, "DebugInvalidateTokenCacheForUser"}, // 3.0.0+
         {998, nullptr, "DebugSetUserStateClose"},
         {999, nullptr, "DebugSetUserStateOpen"},
     };

src/core/hle/service/am/applets/software_keyboard.cpp

@@ -30,7 +30,7 @@ static Core::Frontend::SoftwareKeyboardParameters ConvertToFrontendParameters(
                                                                        config.sub_text.size());
     params.guide_text = Common::UTF16StringFromFixedZeroTerminatedBuffer(config.guide_text.data(),
                                                                          config.guide_text.size());
-    params.initial_text = initial_text;
+    params.initial_text = std::move(initial_text);
     params.max_length = config.length_limit == 0 ? DEFAULT_MAX_LENGTH : config.length_limit;
     params.password = static_cast<bool>(config.is_password);
     params.cursor_at_beginning = static_cast<bool>(config.initial_cursor_position);
@@ -109,7 +109,7 @@ void SoftwareKeyboard::Execute() {
 
     const auto parameters = ConvertToFrontendParameters(config, initial_text);
 
-    frontend.RequestText([this](std::optional<std::u16string> text) { WriteText(text); },
+    frontend.RequestText([this](std::optional<std::u16string> text) { WriteText(std::move(text)); },
                          parameters);
 }
 

src/core/hle/service/hid/controllers/npad.cpp

@@ -566,6 +566,14 @@ void Controller_NPad::DisconnectNPad(u32 npad_id) {
     connected_controllers[NPadIdToIndex(npad_id)].is_connected = false;
 }
 
+void Controller_NPad::SetGyroscopeZeroDriftMode(GyroscopeZeroDriftMode drift_mode) {
+    gyroscope_zero_drift_mode = drift_mode;
+}
+
+Controller_NPad::GyroscopeZeroDriftMode Controller_NPad::GetGyroscopeZeroDriftMode() const {
+    return gyroscope_zero_drift_mode;
+}
+
 void Controller_NPad::StartLRAssignmentMode() {
     // Nothing internally is used for lr assignment mode. Since we have the ability to set the
     // controller types from boot, it doesn't really matter about showing a selection screen

src/core/hle/service/hid/controllers/npad.h

@@ -58,6 +58,12 @@ public:
     };
     static_assert(sizeof(Vibration) == 0x10, "Vibration is an invalid size");
 
+    enum class GyroscopeZeroDriftMode : u32 {
+        Loose = 0,
+        Standard = 1,
+        Tight = 2,
+    };
+
     enum class NpadHoldType : u64 {
         Vertical = 0,
         Horizontal = 1,
@@ -117,6 +123,8 @@ public:
 
     void ConnectNPad(u32 npad_id);
     void DisconnectNPad(u32 npad_id);
+    void SetGyroscopeZeroDriftMode(GyroscopeZeroDriftMode drift_mode);
+    GyroscopeZeroDriftMode GetGyroscopeZeroDriftMode() const;
     LedPattern GetLedPattern(u32 npad_id);
     void SetVibrationEnabled(bool can_vibrate);
     bool IsVibrationEnabled() const;
@@ -324,8 +332,8 @@ private:
     std::array<Kernel::EventPair, 10> styleset_changed_events;
     Vibration last_processed_vibration{};
     std::array<ControllerHolder, 10> connected_controllers{};
+    GyroscopeZeroDriftMode gyroscope_zero_drift_mode{GyroscopeZeroDriftMode::Standard};
     bool can_controllers_vibrate{true};
-
     std::array<ControllerPad, 10> npad_pad_states{};
     bool is_in_lr_assignment_mode{false};
     Core::System& system;

src/core/hle/service/hid/hid.cpp

@@ -161,7 +161,7 @@ Hid::Hid(Core::System& system) : ServiceFramework("hid"), system(system) {
         {40, nullptr, "AcquireXpadIdEventHandle"},
         {41, nullptr, "ReleaseXpadIdEventHandle"},
         {51, &Hid::ActivateXpad, "ActivateXpad"},
-        {55, nullptr, "GetXpadIds"},
+        {55, &Hid::GetXpadIDs, "GetXpadIds"},
         {56, nullptr, "ActivateJoyXpad"},
         {58, nullptr, "GetJoyXpadLifoHandle"},
         {59, nullptr, "GetJoyXpadIds"},
@@ -185,8 +185,8 @@ Hid::Hid(Core::System& system) : ServiceFramework("hid"), system(system) {
         {77, nullptr, "GetAccelerometerPlayMode"},
         {78, nullptr, "ResetAccelerometerPlayMode"},
         {79, &Hid::SetGyroscopeZeroDriftMode, "SetGyroscopeZeroDriftMode"},
-        {80, nullptr, "GetGyroscopeZeroDriftMode"},
-        {81, nullptr, "ResetGyroscopeZeroDriftMode"},
+        {80, &Hid::GetGyroscopeZeroDriftMode, "GetGyroscopeZeroDriftMode"},
+        {81, &Hid::ResetGyroscopeZeroDriftMode, "ResetGyroscopeZeroDriftMode"},
         {82, &Hid::IsSixAxisSensorAtRest, "IsSixAxisSensorAtRest"},
         {83, nullptr, "IsFirmwareUpdateAvailableForSixAxisSensor"},
         {91, &Hid::ActivateGesture, "ActivateGesture"},
@@ -230,15 +230,15 @@ Hid::Hid(Core::System& system) : ServiceFramework("hid"), system(system) {
         {211, nullptr, "IsVibrationDeviceMounted"},
         {300, &Hid::ActivateConsoleSixAxisSensor, "ActivateConsoleSixAxisSensor"},
         {301, &Hid::StartConsoleSixAxisSensor, "StartConsoleSixAxisSensor"},
-        {302, nullptr, "StopConsoleSixAxisSensor"},
-        {303, nullptr, "ActivateSevenSixAxisSensor"},
-        {304, nullptr, "StartSevenSixAxisSensor"},
+        {302, &Hid::StopConsoleSixAxisSensor, "StopConsoleSixAxisSensor"},
+        {303, &Hid::ActivateSevenSixAxisSensor, "ActivateSevenSixAxisSensor"},
+        {304, &Hid::StartSevenSixAxisSensor, "StartSevenSixAxisSensor"},
         {305, &Hid::StopSevenSixAxisSensor, "StopSevenSixAxisSensor"},
         {306, &Hid::InitializeSevenSixAxisSensor, "InitializeSevenSixAxisSensor"},
-        {307, nullptr, "FinalizeSevenSixAxisSensor"},
+        {307, &Hid::FinalizeSevenSixAxisSensor, "FinalizeSevenSixAxisSensor"},
         {308, nullptr, "SetSevenSixAxisSensorFusionStrength"},
         {309, nullptr, "GetSevenSixAxisSensorFusionStrength"},
-        {310, nullptr, "ResetSevenSixAxisSensorTimestamp"},
+        {310, &Hid::ResetSevenSixAxisSensorTimestamp, "ResetSevenSixAxisSensorTimestamp"},
         {400, nullptr, "IsUsbFullKeyControllerEnabled"},
         {401, nullptr, "EnableUsbFullKeyController"},
         {402, nullptr, "IsUsbFullKeyControllerConnected"},
@@ -319,6 +319,17 @@ void Hid::ActivateXpad(Kernel::HLERequestContext& ctx) {
     rb.Push(RESULT_SUCCESS);
 }
 
+void Hid::GetXpadIDs(Kernel::HLERequestContext& ctx) {
+    IPC::RequestParser rp{ctx};
+    const auto applet_resource_user_id{rp.Pop<u64>()};
+
+    LOG_DEBUG(Service_HID, "(STUBBED) called, applet_resource_user_id={}", applet_resource_user_id);
+
+    IPC::ResponseBuilder rb{ctx, 3};
+    rb.Push(RESULT_SUCCESS);
+    rb.Push(0);
+}
+
 void Hid::ActivateDebugPad(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp{ctx};
     const auto applet_resource_user_id{rp.Pop<u64>()};
@@ -363,6 +374,15 @@ void Hid::ActivateKeyboard(Kernel::HLERequestContext& ctx) {
     rb.Push(RESULT_SUCCESS);
 }
 
+void Hid::SendKeyboardLockKeyEvent(Kernel::HLERequestContext& ctx) {
+    IPC::RequestParser rp{ctx};
+    const auto flags{rp.Pop<u32>()};
+    LOG_WARNING(Service_HID, "(STUBBED) called. flags={}", flags);
+
+    IPC::ResponseBuilder rb{ctx, 2};
+    rb.Push(RESULT_SUCCESS);
+}
+
 void Hid::ActivateGesture(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp{ctx};
     const auto unknown{rp.Pop<u32>()};
@@ -402,15 +422,59 @@ void Hid::StartSixAxisSensor(Kernel::HLERequestContext& ctx) {
     rb.Push(RESULT_SUCCESS);
 }
 
+void Hid::StopSixAxisSensor(Kernel::HLERequestContext& ctx) {
+    IPC::RequestParser rp{ctx};
+    const auto handle{rp.Pop<u32>()};
+    const auto applet_resource_user_id{rp.Pop<u64>()};
+
+    LOG_WARNING(Service_HID, "(STUBBED) called, handle={}, applet_resource_user_id={}", handle,
+                applet_resource_user_id);
+
+    IPC::ResponseBuilder rb{ctx, 2};
+    rb.Push(RESULT_SUCCESS);
+}
+
 void Hid::SetGyroscopeZeroDriftMode(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp{ctx};
     const auto handle{rp.Pop<u32>()};
     const auto drift_mode{rp.Pop<u32>()};
     const auto applet_resource_user_id{rp.Pop<u64>()};
 
-    LOG_WARNING(Service_HID,
-                "(STUBBED) called, handle={}, drift_mode={}, applet_resource_user_id={}", handle,
-                drift_mode, applet_resource_user_id);
+    applet_resource->GetController<Controller_NPad>(HidController::NPad)
+        .SetGyroscopeZeroDriftMode(Controller_NPad::GyroscopeZeroDriftMode{drift_mode});
+
+    LOG_DEBUG(Service_HID, "called, handle={}, drift_mode={}, applet_resource_user_id={}", handle,
+              drift_mode, applet_resource_user_id);
+
+    IPC::ResponseBuilder rb{ctx, 2};
+    rb.Push(RESULT_SUCCESS);
+}
+
+void Hid::GetGyroscopeZeroDriftMode(Kernel::HLERequestContext& ctx) {
+    IPC::RequestParser rp{ctx};
+    const auto handle{rp.Pop<u32>()};
+    const auto applet_resource_user_id{rp.Pop<u64>()};
+
+    LOG_DEBUG(Service_HID, "called, handle={}, applet_resource_user_id={}", handle,
+              applet_resource_user_id);
+
+    IPC::ResponseBuilder rb{ctx, 3};
+    rb.Push(RESULT_SUCCESS);
+    rb.Push<u32>(
+        static_cast<u32>(applet_resource->GetController<Controller_NPad>(HidController::NPad)
+                             .GetGyroscopeZeroDriftMode()));
+}
+
+void Hid::ResetGyroscopeZeroDriftMode(Kernel::HLERequestContext& ctx) {
+    IPC::RequestParser rp{ctx};
+    const auto handle{rp.Pop<u32>()};
+    const auto applet_resource_user_id{rp.Pop<u64>()};
+
+    applet_resource->GetController<Controller_NPad>(HidController::NPad)
+        .SetGyroscopeZeroDriftMode(Controller_NPad::GyroscopeZeroDriftMode::Standard);
+
+    LOG_DEBUG(Service_HID, "called, handle={}, applet_resource_user_id={}", handle,
+              applet_resource_user_id);
 
     IPC::ResponseBuilder rb{ctx, 2};
     rb.Push(RESULT_SUCCESS);
@@ -821,33 +885,35 @@ void Hid::StartConsoleSixAxisSensor(Kernel::HLERequestContext& ctx) {
     rb.Push(RESULT_SUCCESS);
 }
 
-void Hid::StopSixAxisSensor(Kernel::HLERequestContext& ctx) {
+void Hid::StopConsoleSixAxisSensor(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp{ctx};
     const auto handle{rp.Pop<u32>()};
+    const auto applet_resource_user_id{rp.Pop<u64>()};
 
-    LOG_WARNING(Service_HID, "(STUBBED) called, handle={}", handle);
+    LOG_WARNING(Service_HID, "(STUBBED) called, handle={}, applet_resource_user_id={}", handle,
+                applet_resource_user_id);
 
     IPC::ResponseBuilder rb{ctx, 2};
     rb.Push(RESULT_SUCCESS);
 }
 
-void Hid::SetIsPalmaAllConnectable(Kernel::HLERequestContext& ctx) {
+void Hid::ActivateSevenSixAxisSensor(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp{ctx};
     const auto applet_resource_user_id{rp.Pop<u64>()};
-    const auto unknown{rp.Pop<u32>()};
 
-    LOG_WARNING(Service_HID, "(STUBBED) called, applet_resource_user_id={}, unknown={}",
-                applet_resource_user_id, unknown);
+    LOG_WARNING(Service_HID, "(STUBBED) called, applet_resource_user_id={}",
+                applet_resource_user_id);
 
     IPC::ResponseBuilder rb{ctx, 2};
     rb.Push(RESULT_SUCCESS);
 }
 
-void Hid::SetPalmaBoostMode(Kernel::HLERequestContext& ctx) {
+void Hid::StartSevenSixAxisSensor(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp{ctx};
-    const auto unknown{rp.Pop<u32>()};
+    const auto applet_resource_user_id{rp.Pop<u64>()};
 
-    LOG_WARNING(Service_HID, "(STUBBED) called, unknown={}", unknown);
+    LOG_WARNING(Service_HID, "(STUBBED) called, applet_resource_user_id={}",
+                applet_resource_user_id);
 
     IPC::ResponseBuilder rb{ctx, 2};
     rb.Push(RESULT_SUCCESS);
@@ -871,10 +937,46 @@ void Hid::InitializeSevenSixAxisSensor(Kernel::HLERequestContext& ctx) {
     rb.Push(RESULT_SUCCESS);
 }
 
-void Hid::SendKeyboardLockKeyEvent(Kernel::HLERequestContext& ctx) {
+void Hid::FinalizeSevenSixAxisSensor(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp{ctx};
-    const auto flags{rp.Pop<u32>()};
-    LOG_WARNING(Service_HID, "(STUBBED) called. flags={}", flags);
+    const auto applet_resource_user_id{rp.Pop<u64>()};
+
+    LOG_WARNING(Service_HID, "(STUBBED) called, applet_resource_user_id={}",
+                applet_resource_user_id);
+
+    IPC::ResponseBuilder rb{ctx, 2};
+    rb.Push(RESULT_SUCCESS);
+}
+
+void Hid::ResetSevenSixAxisSensorTimestamp(Kernel::HLERequestContext& ctx) {
+    IPC::RequestParser rp{ctx};
+    const auto applet_resource_user_id{rp.Pop<u64>()};
+
+    LOG_WARNING(Service_HID, "(STUBBED) called, applet_resource_user_id={}",
+                applet_resource_user_id);
+
+    IPC::ResponseBuilder rb{ctx, 2};
+    rb.Push(RESULT_SUCCESS);
+}
+
+void Hid::SetIsPalmaAllConnectable(Kernel::HLERequestContext& ctx) {
+    IPC::RequestParser rp{ctx};
+    const auto applet_resource_user_id{rp.Pop<u64>()};
+    const auto is_palma_all_connectable{rp.Pop<bool>()};
+
+    LOG_WARNING(Service_HID,
+                "(STUBBED) called, applet_resource_user_id={}, is_palma_all_connectable={}",
+                applet_resource_user_id, is_palma_all_connectable);
+
+    IPC::ResponseBuilder rb{ctx, 2};
+    rb.Push(RESULT_SUCCESS);
+}
+
+void Hid::SetPalmaBoostMode(Kernel::HLERequestContext& ctx) {
+    IPC::RequestParser rp{ctx};
+    const auto palma_boost_mode{rp.Pop<bool>()};
+
+    LOG_WARNING(Service_HID, "(STUBBED) called, palma_boost_mode={}", palma_boost_mode);
 
     IPC::ResponseBuilder rb{ctx, 2};
     rb.Push(RESULT_SUCCESS);

src/core/hle/service/hid/hid.h

@@ -86,14 +86,19 @@ public:
 private:
     void CreateAppletResource(Kernel::HLERequestContext& ctx);
     void ActivateXpad(Kernel::HLERequestContext& ctx);
+    void GetXpadIDs(Kernel::HLERequestContext& ctx);
     void ActivateDebugPad(Kernel::HLERequestContext& ctx);
     void ActivateTouchScreen(Kernel::HLERequestContext& ctx);
     void ActivateMouse(Kernel::HLERequestContext& ctx);
     void ActivateKeyboard(Kernel::HLERequestContext& ctx);
+    void SendKeyboardLockKeyEvent(Kernel::HLERequestContext& ctx);
     void ActivateGesture(Kernel::HLERequestContext& ctx);
     void ActivateNpadWithRevision(Kernel::HLERequestContext& ctx);
     void StartSixAxisSensor(Kernel::HLERequestContext& ctx);
+    void StopSixAxisSensor(Kernel::HLERequestContext& ctx);
     void SetGyroscopeZeroDriftMode(Kernel::HLERequestContext& ctx);
+    void GetGyroscopeZeroDriftMode(Kernel::HLERequestContext& ctx);
+    void ResetGyroscopeZeroDriftMode(Kernel::HLERequestContext& ctx);
     void IsSixAxisSensorAtRest(Kernel::HLERequestContext& ctx);
     void SetSupportedNpadStyleSet(Kernel::HLERequestContext& ctx);
     void GetSupportedNpadStyleSet(Kernel::HLERequestContext& ctx);
@@ -125,12 +130,15 @@ private:
     void IsVibrationPermitted(Kernel::HLERequestContext& ctx);
     void ActivateConsoleSixAxisSensor(Kernel::HLERequestContext& ctx);
     void StartConsoleSixAxisSensor(Kernel::HLERequestContext& ctx);
-    void StopSixAxisSensor(Kernel::HLERequestContext& ctx);
-    void SetIsPalmaAllConnectable(Kernel::HLERequestContext& ctx);
-    void SetPalmaBoostMode(Kernel::HLERequestContext& ctx);
+    void StopConsoleSixAxisSensor(Kernel::HLERequestContext& ctx);
+    void ActivateSevenSixAxisSensor(Kernel::HLERequestContext& ctx);
+    void StartSevenSixAxisSensor(Kernel::HLERequestContext& ctx);
     void StopSevenSixAxisSensor(Kernel::HLERequestContext& ctx);
     void InitializeSevenSixAxisSensor(Kernel::HLERequestContext& ctx);
-    void SendKeyboardLockKeyEvent(Kernel::HLERequestContext& ctx);
+    void FinalizeSevenSixAxisSensor(Kernel::HLERequestContext& ctx);
+    void ResetSevenSixAxisSensorTimestamp(Kernel::HLERequestContext& ctx);
+    void SetIsPalmaAllConnectable(Kernel::HLERequestContext& ctx);
+    void SetPalmaBoostMode(Kernel::HLERequestContext& ctx);
 
     std::shared_ptr<IAppletResource> applet_resource;
     Core::System& system;

src/core/hle/service/lm/manager.cpp

@@ -86,7 +86,8 @@ std::string FormatField(Field type, const std::vector<u8>& data) {
         return Common::StringFromFixedZeroTerminatedBuffer(
             reinterpret_cast<const char*>(data.data()), data.size());
     default:
-        UNIMPLEMENTED();
+        UNIMPLEMENTED_MSG("Unimplemented field type={}", type);
+        return "";
     }
 }
 

src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp

@@ -25,7 +25,7 @@ u32 nvhost_ctrl_gpu::ioctl(Ioctl command, const std::vector<u8>& input,
     case IoctlCommand::IocGetCharacteristicsCommand:
         return GetCharacteristics(input, output, output2, version);
     case IoctlCommand::IocGetTPCMasksCommand:
-        return GetTPCMasks(input, output);
+        return GetTPCMasks(input, output, output2, version);
     case IoctlCommand::IocGetActiveSlotMaskCommand:
         return GetActiveSlotMask(input, output);
     case IoctlCommand::IocZcullGetCtxSizeCommand:
@@ -98,17 +98,22 @@ u32 nvhost_ctrl_gpu::GetCharacteristics(const std::vector<u8>& input, std::vecto
     return 0;
 }
 
-u32 nvhost_ctrl_gpu::GetTPCMasks(const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_ctrl_gpu::GetTPCMasks(const std::vector<u8>& input, std::vector<u8>& output,
+                                 std::vector<u8>& output2, IoctlVersion version) {
     IoctlGpuGetTpcMasksArgs params{};
     std::memcpy(&params, input.data(), input.size());
-    LOG_INFO(Service_NVDRV, "called, mask=0x{:X}, mask_buf_addr=0x{:X}", params.mask_buf_size,
-             params.mask_buf_addr);
-    // TODO(ogniK): Confirm value on hardware
-    if (params.mask_buf_size)
-        params.tpc_mask_size = 4 * 1; // 4 * num_gpc
-    else
-        params.tpc_mask_size = 0;
-    std::memcpy(output.data(), &params, sizeof(params));
+    LOG_DEBUG(Service_NVDRV, "called, mask_buffer_size=0x{:X}", params.mask_buffer_size);
+    if (params.mask_buffer_size != 0) {
+        params.tcp_mask = 3;
+    }
+
+    if (version == IoctlVersion::Version3) {
+        std::memcpy(output.data(), input.data(), output.size());
+        std::memcpy(output2.data(), &params.tcp_mask, output2.size());
+    } else {
+        std::memcpy(output.data(), &params, output.size());
+    }
+
     return 0;
 }
 

src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h

@@ -92,16 +92,11 @@ private:
                   "IoctlCharacteristics is incorrect size");
 
     struct IoctlGpuGetTpcMasksArgs {
-        /// [in]  TPC mask buffer size reserved by userspace. Should be at least
-        /// sizeof(__u32) * fls(gpc_mask) to receive TPC mask for each GPC.
-        /// [out] full kernel buffer size
-        u32_le mask_buf_size;
-        u32_le reserved;
-
-        /// [in]  pointer to TPC mask buffer. It will receive one 32-bit TPC mask per GPC or 0 if
-        /// GPC is not enabled or not present. This parameter is ignored if mask_buf_size is 0.
-        u64_le mask_buf_addr;
-        u64_le tpc_mask_size; // Nintendo add this?
+        u32_le mask_buffer_size{};
+        INSERT_PADDING_WORDS(1);
+        u64_le mask_buffer_address{};
+        u32_le tcp_mask{};
+        INSERT_PADDING_WORDS(1);
     };
     static_assert(sizeof(IoctlGpuGetTpcMasksArgs) == 24,
                   "IoctlGpuGetTpcMasksArgs is incorrect size");
@@ -166,7 +161,8 @@ private:
 
     u32 GetCharacteristics(const std::vector<u8>& input, std::vector<u8>& output,
                            std::vector<u8>& output2, IoctlVersion version);
-    u32 GetTPCMasks(const std::vector<u8>& input, std::vector<u8>& output);
+    u32 GetTPCMasks(const std::vector<u8>& input, std::vector<u8>& output, std::vector<u8>& output2,
+                    IoctlVersion version);
     u32 GetActiveSlotMask(const std::vector<u8>& input, std::vector<u8>& output);
     u32 ZCullGetCtxSize(const std::vector<u8>& input, std::vector<u8>& output);
     u32 ZCullGetInfo(const std::vector<u8>& input, std::vector<u8>& output);

src/core/settings.h

@@ -437,7 +437,7 @@ struct Values {
     bool renderer_debug;
     int vulkan_device;
 
-    float resolution_factor;
+    u16 resolution_factor{1};
     int aspect_ratio;
     int max_anisotropy;
     bool use_frame_limit;
@@ -474,6 +474,7 @@ struct Values {
     bool reporting_services;
     bool quest_flag;
     bool disable_cpu_opt;
+    bool disable_macro_jit;
 
     // BCAT
     std::string bcat_backend;

src/input_common/keyboard.cpp

@@ -76,7 +76,7 @@ std::unique_ptr<Input::ButtonDevice> Keyboard::Create(const Common::ParamPackage
     int key_code = params.Get("code", 0);
     std::unique_ptr<KeyButton> button = std::make_unique<KeyButton>(key_button_list);
     key_button_list->AddKeyButton(key_code, button.get());
-    return std::move(button);
+    return button;
 }
 
 void Keyboard::PressKey(int key_code) {

src/input_common/motion_emu.cpp

@@ -145,7 +145,7 @@ std::unique_ptr<Input::MotionDevice> MotionEmu::Create(const Common::ParamPackag
     // Previously created device is disconnected here. Having two motion devices for 3DS is not
     // expected.
     current_device = device_wrapper->device;
-    return std::move(device_wrapper);
+    return device_wrapper;
 }
 
 void MotionEmu::BeginTilt(int x, int y) {

src/video_core/CMakeLists.txt

@@ -25,6 +25,14 @@ add_library(video_core STATIC
     engines/shader_bytecode.h
     engines/shader_header.h
     engines/shader_type.h
+    macro/macro.cpp
+    macro/macro.h
+    macro/macro_hle.cpp
+    macro/macro_hle.h
+    macro/macro_interpreter.cpp
+    macro/macro_interpreter.h
+    macro/macro_jit_x64.cpp
+    macro/macro_jit_x64.h
     fence_manager.h
     gpu.cpp
     gpu.h
@@ -36,8 +44,6 @@ add_library(video_core STATIC
     gpu_thread.h
     guest_driver.cpp
     guest_driver.h
-    macro_interpreter.cpp
-    macro_interpreter.h
     memory_manager.cpp
     memory_manager.h
     morton.cpp
@@ -45,11 +51,11 @@ add_library(video_core STATIC
     query_cache.h
     rasterizer_accelerated.cpp
     rasterizer_accelerated.h
-    rasterizer_cache.cpp
-    rasterizer_cache.h
     rasterizer_interface.h
     renderer_base.cpp
     renderer_base.h
+    renderer_opengl/gl_arb_decompiler.cpp
+    renderer_opengl/gl_arb_decompiler.h
     renderer_opengl/gl_buffer_cache.cpp
     renderer_opengl/gl_buffer_cache.h
     renderer_opengl/gl_device.cpp
@@ -89,6 +95,7 @@ add_library(video_core STATIC
     renderer_opengl/utils.h
     sampler_cache.cpp
     sampler_cache.h
+    shader_cache.h
     shader/decode/arithmetic.cpp
     shader/decode/arithmetic_immediate.cpp
     shader/decode/bfe.cpp

src/video_core/buffer_cache/buffer_block.h

@@ -15,48 +15,47 @@ namespace VideoCommon {
 
 class BufferBlock {
 public:
-    bool Overlaps(const VAddr start, const VAddr end) const {
+    bool Overlaps(VAddr start, VAddr end) const {
         return (cpu_addr < end) && (cpu_addr_end > start);
     }
 
-    bool IsInside(const VAddr other_start, const VAddr other_end) const {
+    bool IsInside(VAddr other_start, VAddr other_end) const {
         return cpu_addr <= other_start && other_end <= cpu_addr_end;
     }
 
-    std::size_t GetOffset(const VAddr in_addr) {
+    std::size_t Offset(VAddr in_addr) const {
         return static_cast<std::size_t>(in_addr - cpu_addr);
     }
 
-    VAddr GetCpuAddr() const {
+    VAddr CpuAddr() const {
         return cpu_addr;
     }
 
-    VAddr GetCpuAddrEnd() const {
+    VAddr CpuAddrEnd() const {
         return cpu_addr_end;
     }
 
-    void SetCpuAddr(const VAddr new_addr) {
+    void SetCpuAddr(VAddr new_addr) {
         cpu_addr = new_addr;
         cpu_addr_end = new_addr + size;
     }
 
-    std::size_t GetSize() const {
+    std::size_t Size() const {
         return size;
     }
 
-    void SetEpoch(u64 new_epoch) {
-        epoch = new_epoch;
+    u64 Epoch() const {
+        return epoch;
     }
 
-    u64 GetEpoch() {
-        return epoch;
+    void SetEpoch(u64 new_epoch) {
+        epoch = new_epoch;
     }
 
 protected:
-    explicit BufferBlock(VAddr cpu_addr, const std::size_t size) : size{size} {
-        SetCpuAddr(cpu_addr);
+    explicit BufferBlock(VAddr cpu_addr_, std::size_t size_) : size{size_} {
+        SetCpuAddr(cpu_addr_);
     }
-    ~BufferBlock() = default;
 
 private:
     VAddr cpu_addr{};

src/video_core/buffer_cache/buffer_cache.h

@@ -30,23 +30,31 @@
 
 namespace VideoCommon {
 
-template <typename OwnerBuffer, typename BufferType, typename StreamBuffer>
+template <typename Buffer, typename BufferType, typename StreamBuffer>
 class BufferCache {
     using IntervalSet = boost::icl::interval_set<VAddr>;
     using IntervalType = typename IntervalSet::interval_type;
     using VectorMapInterval = boost::container::small_vector<MapInterval*, 1>;
 
+    static constexpr u64 WRITE_PAGE_BIT = 11;
+    static constexpr u64 BLOCK_PAGE_BITS = 21;
+    static constexpr u64 BLOCK_PAGE_SIZE = 1ULL << BLOCK_PAGE_BITS;
+
 public:
-    using BufferInfo = std::pair<BufferType, u64>;
+    struct BufferInfo {
+        BufferType handle;
+        u64 offset;
+        u64 address;
+    };
 
     BufferInfo UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4,
                             bool is_written = false, bool use_fast_cbuf = false) {
         std::lock_guard lock{mutex};
 
-        const auto& memory_manager = system.GPU().MemoryManager();
+        auto& memory_manager = system.GPU().MemoryManager();
         const std::optional<VAddr> cpu_addr_opt = memory_manager.GpuToCpuAddress(gpu_addr);
         if (!cpu_addr_opt) {
-            return {GetEmptyBuffer(size), 0};
+            return GetEmptyBuffer(size);
         }
         const VAddr cpu_addr = *cpu_addr_opt;
 
@@ -55,33 +63,36 @@ public:
         constexpr std::size_t max_stream_size = 0x800;
         if (use_fast_cbuf || size < max_stream_size) {
             if (!is_written && !IsRegionWritten(cpu_addr, cpu_addr + size - 1)) {
-                auto& memory_manager = system.GPU().MemoryManager();
+                const bool is_granular = memory_manager.IsGranularRange(gpu_addr, size);
                 if (use_fast_cbuf) {
-                    if (memory_manager.IsGranularRange(gpu_addr, size)) {
-                        const auto host_ptr = memory_manager.GetPointer(gpu_addr);
-                        return ConstBufferUpload(host_ptr, size);
+                    u8* dest;
+                    if (is_granular) {
+                        dest = memory_manager.GetPointer(gpu_addr);
                     } else {
                         staging_buffer.resize(size);
-                        memory_manager.ReadBlockUnsafe(gpu_addr, staging_buffer.data(), size);
-                        return ConstBufferUpload(staging_buffer.data(), size);
+                        dest = staging_buffer.data();
+                        memory_manager.ReadBlockUnsafe(gpu_addr, dest, size);
                     }
+                    return ConstBufferUpload(dest, size);
+                }
+                if (is_granular) {
+                    u8* const host_ptr = memory_manager.GetPointer(gpu_addr);
+                    return StreamBufferUpload(size, alignment, [host_ptr, size](u8* dest) {
+                        std::memcpy(dest, host_ptr, size);
+                    });
                 } else {
-                    if (memory_manager.IsGranularRange(gpu_addr, size)) {
-                        const auto host_ptr = memory_manager.GetPointer(gpu_addr);
-                        return StreamBufferUpload(host_ptr, size, alignment);
-                    } else {
-                        staging_buffer.resize(size);
-                        memory_manager.ReadBlockUnsafe(gpu_addr, staging_buffer.data(), size);
-                        return StreamBufferUpload(staging_buffer.data(), size, alignment);
-                    }
+                    return StreamBufferUpload(
+                        size, alignment, [&memory_manager, gpu_addr, size](u8* dest) {
+                            memory_manager.ReadBlockUnsafe(gpu_addr, dest, size);
+                        });
                 }
             }
         }
 
-        OwnerBuffer block = GetBlock(cpu_addr, size);
+        Buffer* const block = GetBlock(cpu_addr, size);
         MapInterval* const map = MapAddress(block, gpu_addr, cpu_addr, size);
         if (!map) {
-            return {GetEmptyBuffer(size), 0};
+            return GetEmptyBuffer(size);
         }
         if (is_written) {
             map->MarkAsModified(true, GetModifiedTicks());
@@ -94,41 +105,49 @@ public:
             }
         }
 
-        return {ToHandle(block), static_cast<u64>(block->GetOffset(cpu_addr))};
+        return BufferInfo{block->Handle(), block->Offset(cpu_addr), block->Address()};
     }
 
     /// Uploads from a host memory. Returns the OpenGL buffer where it's located and its offset.
     BufferInfo UploadHostMemory(const void* raw_pointer, std::size_t size,
                                 std::size_t alignment = 4) {
         std::lock_guard lock{mutex};
-        return StreamBufferUpload(raw_pointer, size, alignment);
+        return StreamBufferUpload(size, alignment, [raw_pointer, size](u8* dest) {
+            std::memcpy(dest, raw_pointer, size);
+        });
     }
 
-    void Map(std::size_t max_size) {
+    /// Prepares the buffer cache for data uploading
+    /// @param max_size Maximum number of bytes that will be uploaded
+    /// @return True when a stream buffer invalidation was required, false otherwise
+    bool Map(std::size_t max_size) {
         std::lock_guard lock{mutex};
 
+        bool invalidated;
         std::tie(buffer_ptr, buffer_offset_base, invalidated) = stream_buffer->Map(max_size, 4);
         buffer_offset = buffer_offset_base;
+
+        return invalidated;
     }
 
-    /// Finishes the upload stream, returns true on bindings invalidation.
-    bool Unmap() {
+    /// Finishes the upload stream
+    void Unmap() {
         std::lock_guard lock{mutex};
-
         stream_buffer->Unmap(buffer_offset - buffer_offset_base);
-        return std::exchange(invalidated, false);
     }
 
+    /// Function called at the end of each frame, inteded for deferred operations
     void TickFrame() {
         ++epoch;
+
         while (!pending_destruction.empty()) {
             // Delay at least 4 frames before destruction.
             // This is due to triple buffering happening on some drivers.
             static constexpr u64 epochs_to_destroy = 5;
-            if (pending_destruction.front()->GetEpoch() + epochs_to_destroy > epoch) {
+            if (pending_destruction.front()->Epoch() + epochs_to_destroy > epoch) {
                 break;
             }
-            pending_destruction.pop_front();
+            pending_destruction.pop();
         }
     }
 
@@ -239,28 +258,16 @@ public:
         committed_flushes.pop_front();
     }
 
-    virtual BufferType GetEmptyBuffer(std::size_t size) = 0;
+    virtual BufferInfo GetEmptyBuffer(std::size_t size) = 0;
 
 protected:
     explicit BufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system,
                          std::unique_ptr<StreamBuffer> stream_buffer)
-        : rasterizer{rasterizer}, system{system}, stream_buffer{std::move(stream_buffer)},
-          stream_buffer_handle{this->stream_buffer->GetHandle()} {}
+        : rasterizer{rasterizer}, system{system}, stream_buffer{std::move(stream_buffer)} {}
 
     ~BufferCache() = default;
 
-    virtual BufferType ToHandle(const OwnerBuffer& storage) = 0;
-
-    virtual OwnerBuffer CreateBlock(VAddr cpu_addr, std::size_t size) = 0;
-
-    virtual void UploadBlockData(const OwnerBuffer& buffer, std::size_t offset, std::size_t size,
-                                 const u8* data) = 0;
-
-    virtual void DownloadBlockData(const OwnerBuffer& buffer, std::size_t offset, std::size_t size,
-                                   u8* data) = 0;
-
-    virtual void CopyBlock(const OwnerBuffer& src, const OwnerBuffer& dst, std::size_t src_offset,
-                           std::size_t dst_offset, std::size_t size) = 0;
+    virtual std::shared_ptr<Buffer> CreateBlock(VAddr cpu_addr, std::size_t size) = 0;
 
     virtual BufferInfo ConstBufferUpload(const void* raw_pointer, std::size_t size) {
         return {};
@@ -315,7 +322,7 @@ protected:
     }
 
 private:
-    MapInterval* MapAddress(const OwnerBuffer& block, GPUVAddr gpu_addr, VAddr cpu_addr,
+    MapInterval* MapAddress(const Buffer* block, GPUVAddr gpu_addr, VAddr cpu_addr,
                             std::size_t size) {
         const VectorMapInterval overlaps = GetMapsInRange(cpu_addr, size);
         if (overlaps.empty()) {
@@ -323,11 +330,11 @@ private:
             const VAddr cpu_addr_end = cpu_addr + size;
             if (memory_manager.IsGranularRange(gpu_addr, size)) {
                 u8* host_ptr = memory_manager.GetPointer(gpu_addr);
-                UploadBlockData(block, block->GetOffset(cpu_addr), size, host_ptr);
+                block->Upload(block->Offset(cpu_addr), size, host_ptr);
             } else {
                 staging_buffer.resize(size);
                 memory_manager.ReadBlockUnsafe(gpu_addr, staging_buffer.data(), size);
-                UploadBlockData(block, block->GetOffset(cpu_addr), size, staging_buffer.data());
+                block->Upload(block->Offset(cpu_addr), size, staging_buffer.data());
             }
             return Register(MapInterval(cpu_addr, cpu_addr_end, gpu_addr));
         }
@@ -370,7 +377,7 @@ private:
         return map;
     }
 
-    void UpdateBlock(const OwnerBuffer& block, VAddr start, VAddr end,
+    void UpdateBlock(const Buffer* block, VAddr start, VAddr end,
                      const VectorMapInterval& overlaps) {
         const IntervalType base_interval{start, end};
         IntervalSet interval_set{};
@@ -380,13 +387,13 @@ private:
             interval_set.subtract(subtract);
         }
         for (auto& interval : interval_set) {
-            std::size_t size = interval.upper() - interval.lower();
-            if (size > 0) {
-                staging_buffer.resize(size);
-                system.Memory().ReadBlockUnsafe(interval.lower(), staging_buffer.data(), size);
-                UploadBlockData(block, block->GetOffset(interval.lower()), size,
-                                staging_buffer.data());
+            const std::size_t size = interval.upper() - interval.lower();
+            if (size == 0) {
+                continue;
             }
+            staging_buffer.resize(size);
+            system.Memory().ReadBlockUnsafe(interval.lower(), staging_buffer.data(), size);
+            block->Upload(block->Offset(interval.lower()), size, staging_buffer.data());
         }
     }
 
@@ -416,23 +423,27 @@ private:
     }
 
     void FlushMap(MapInterval* map) {
+        const auto it = blocks.find(map->start >> BLOCK_PAGE_BITS);
+        ASSERT_OR_EXECUTE(it != blocks.end(), return;);
+
+        std::shared_ptr<Buffer> block = it->second;
+
         const std::size_t size = map->end - map->start;
-        OwnerBuffer block = blocks[map->start >> block_page_bits];
         staging_buffer.resize(size);
-        DownloadBlockData(block, block->GetOffset(map->start), size, staging_buffer.data());
+        block->Download(block->Offset(map->start), size, staging_buffer.data());
         system.Memory().WriteBlockUnsafe(map->start, staging_buffer.data(), size);
         map->MarkAsModified(false, 0);
     }
 
-    BufferInfo StreamBufferUpload(const void* raw_pointer, std::size_t size,
-                                  std::size_t alignment) {
+    template <typename Callable>
+    BufferInfo StreamBufferUpload(std::size_t size, std::size_t alignment, Callable&& callable) {
         AlignBuffer(alignment);
         const std::size_t uploaded_offset = buffer_offset;
-        std::memcpy(buffer_ptr, raw_pointer, size);
+        callable(buffer_ptr);
 
         buffer_ptr += size;
         buffer_offset += size;
-        return {stream_buffer_handle, uploaded_offset};
+        return BufferInfo{stream_buffer->Handle(), uploaded_offset, stream_buffer->Address()};
     }
 
     void AlignBuffer(std::size_t alignment) {
@@ -442,97 +453,89 @@ private:
         buffer_offset = offset_aligned;
     }
 
-    OwnerBuffer EnlargeBlock(OwnerBuffer buffer) {
-        const std::size_t old_size = buffer->GetSize();
-        const std::size_t new_size = old_size + block_page_size;
-        const VAddr cpu_addr = buffer->GetCpuAddr();
-        OwnerBuffer new_buffer = CreateBlock(cpu_addr, new_size);
-        CopyBlock(buffer, new_buffer, 0, 0, old_size);
-        buffer->SetEpoch(epoch);
-        pending_destruction.push_back(buffer);
+    std::shared_ptr<Buffer> EnlargeBlock(std::shared_ptr<Buffer> buffer) {
+        const std::size_t old_size = buffer->Size();
+        const std::size_t new_size = old_size + BLOCK_PAGE_SIZE;
+        const VAddr cpu_addr = buffer->CpuAddr();
+        std::shared_ptr<Buffer> new_buffer = CreateBlock(cpu_addr, new_size);
+        new_buffer->CopyFrom(*buffer, 0, 0, old_size);
+        QueueDestruction(std::move(buffer));
+
         const VAddr cpu_addr_end = cpu_addr + new_size - 1;
-        u64 page_start = cpu_addr >> block_page_bits;
-        const u64 page_end = cpu_addr_end >> block_page_bits;
-        while (page_start <= page_end) {
-            blocks[page_start] = new_buffer;
-            ++page_start;
+        const u64 page_end = cpu_addr_end >> BLOCK_PAGE_BITS;
+        for (u64 page_start = cpu_addr >> BLOCK_PAGE_BITS; page_start <= page_end; ++page_start) {
+            blocks.insert_or_assign(page_start, new_buffer);
         }
+
         return new_buffer;
     }
 
-    OwnerBuffer MergeBlocks(OwnerBuffer first, OwnerBuffer second) {
-        const std::size_t size_1 = first->GetSize();
-        const std::size_t size_2 = second->GetSize();
-        const VAddr first_addr = first->GetCpuAddr();
-        const VAddr second_addr = second->GetCpuAddr();
+    std::shared_ptr<Buffer> MergeBlocks(std::shared_ptr<Buffer> first,
+                                        std::shared_ptr<Buffer> second) {
+        const std::size_t size_1 = first->Size();
+        const std::size_t size_2 = second->Size();
+        const VAddr first_addr = first->CpuAddr();
+        const VAddr second_addr = second->CpuAddr();
         const VAddr new_addr = std::min(first_addr, second_addr);
         const std::size_t new_size = size_1 + size_2;
-        OwnerBuffer new_buffer = CreateBlock(new_addr, new_size);
-        CopyBlock(first, new_buffer, 0, new_buffer->GetOffset(first_addr), size_1);
-        CopyBlock(second, new_buffer, 0, new_buffer->GetOffset(second_addr), size_2);
-        first->SetEpoch(epoch);
-        second->SetEpoch(epoch);
-        pending_destruction.push_back(first);
-        pending_destruction.push_back(second);
+
+        std::shared_ptr<Buffer> new_buffer = CreateBlock(new_addr, new_size);
+        new_buffer->CopyFrom(*first, 0, new_buffer->Offset(first_addr), size_1);
+        new_buffer->CopyFrom(*second, 0, new_buffer->Offset(second_addr), size_2);
+        QueueDestruction(std::move(first));
+        QueueDestruction(std::move(second));
+
         const VAddr cpu_addr_end = new_addr + new_size - 1;
-        u64 page_start = new_addr >> block_page_bits;
-        const u64 page_end = cpu_addr_end >> block_page_bits;
-        while (page_start <= page_end) {
-            blocks[page_start] = new_buffer;
-            ++page_start;
+        const u64 page_end = cpu_addr_end >> BLOCK_PAGE_BITS;
+        for (u64 page_start = new_addr >> BLOCK_PAGE_BITS; page_start <= page_end; ++page_start) {
+            blocks.insert_or_assign(page_start, new_buffer);
         }
         return new_buffer;
     }
 
-    OwnerBuffer GetBlock(const VAddr cpu_addr, const std::size_t size) {
-        OwnerBuffer found;
+    Buffer* GetBlock(VAddr cpu_addr, std::size_t size) {
+        std::shared_ptr<Buffer> found;
+
         const VAddr cpu_addr_end = cpu_addr + size - 1;
-        u64 page_start = cpu_addr >> block_page_bits;
-        const u64 page_end = cpu_addr_end >> block_page_bits;
-        while (page_start <= page_end) {
+        const u64 page_end = cpu_addr_end >> BLOCK_PAGE_BITS;
+        for (u64 page_start = cpu_addr >> BLOCK_PAGE_BITS; page_start <= page_end; ++page_start) {
             auto it = blocks.find(page_start);
             if (it == blocks.end()) {
                 if (found) {
                     found = EnlargeBlock(found);
-                } else {
-                    const VAddr start_addr = (page_start << block_page_bits);
-                    found = CreateBlock(start_addr, block_page_size);
-                    blocks[page_start] = found;
-                }
-            } else {
-                if (found) {
-                    if (found == it->second) {
-                        ++page_start;
-                        continue;
-                    }
-                    found = MergeBlocks(found, it->second);
-                } else {
-                    found = it->second;
+                    continue;
                 }
+                const VAddr start_addr = page_start << BLOCK_PAGE_BITS;
+                found = CreateBlock(start_addr, BLOCK_PAGE_SIZE);
+                blocks.insert_or_assign(page_start, found);
+                continue;
+            }
+            if (!found) {
+                found = it->second;
+                continue;
+            }
+            if (found != it->second) {
+                found = MergeBlocks(std::move(found), it->second);
             }
-            ++page_start;
         }
-        return found;
+        return found.get();
     }
 
-    void MarkRegionAsWritten(const VAddr start, const VAddr end) {
-        u64 page_start = start >> write_page_bit;
-        const u64 page_end = end >> write_page_bit;
-        while (page_start <= page_end) {
+    void MarkRegionAsWritten(VAddr start, VAddr end) {
+        const u64 page_end = end >> WRITE_PAGE_BIT;
+        for (u64 page_start = start >> WRITE_PAGE_BIT; page_start <= page_end; ++page_start) {
             auto it = written_pages.find(page_start);
             if (it != written_pages.end()) {
                 it->second = it->second + 1;
             } else {
-                written_pages[page_start] = 1;
+                written_pages.insert_or_assign(page_start, 1);
             }
-            ++page_start;
         }
     }
 
-    void UnmarkRegionAsWritten(const VAddr start, const VAddr end) {
-        u64 page_start = start >> write_page_bit;
-        const u64 page_end = end >> write_page_bit;
-        while (page_start <= page_end) {
+    void UnmarkRegionAsWritten(VAddr start, VAddr end) {
+        const u64 page_end = end >> WRITE_PAGE_BIT;
+        for (u64 page_start = start >> WRITE_PAGE_BIT; page_start <= page_end; ++page_start) {
             auto it = written_pages.find(page_start);
             if (it != written_pages.end()) {
                 if (it->second > 1) {
@@ -541,22 +544,24 @@ private:
                     written_pages.erase(it);
                 }
             }
-            ++page_start;
         }
     }
 
-    bool IsRegionWritten(const VAddr start, const VAddr end) const {
-        u64 page_start = start >> write_page_bit;
-        const u64 page_end = end >> write_page_bit;
-        while (page_start <= page_end) {
+    bool IsRegionWritten(VAddr start, VAddr end) const {
+        const u64 page_end = end >> WRITE_PAGE_BIT;
+        for (u64 page_start = start >> WRITE_PAGE_BIT; page_start <= page_end; ++page_start) {
             if (written_pages.count(page_start) > 0) {
                 return true;
             }
-            ++page_start;
         }
         return false;
     }
 
+    void QueueDestruction(std::shared_ptr<Buffer> buffer) {
+        buffer->SetEpoch(epoch);
+        pending_destruction.push(std::move(buffer));
+    }
+
     void MarkForAsyncFlush(MapInterval* map) {
         if (!uncommitted_flushes) {
             uncommitted_flushes = std::make_shared<std::unordered_set<MapInterval*>>();
@@ -568,9 +573,7 @@ private:
     Core::System& system;
 
     std::unique_ptr<StreamBuffer> stream_buffer;
-    BufferType stream_buffer_handle{};
-
-    bool invalidated = false;
+    BufferType stream_buffer_handle;
 
     u8* buffer_ptr = nullptr;
     u64 buffer_offset = 0;
@@ -580,18 +583,15 @@ private:
     boost::intrusive::set<MapInterval, boost::intrusive::compare<MapIntervalCompare>>
         mapped_addresses;
 
-    static constexpr u64 write_page_bit = 11;
     std::unordered_map<u64, u32> written_pages;
+    std::unordered_map<u64, std::shared_ptr<Buffer>> blocks;
 
-    static constexpr u64 block_page_bits = 21;
-    static constexpr u64 block_page_size = 1ULL << block_page_bits;
-    std::unordered_map<u64, OwnerBuffer> blocks;
-
-    std::list<OwnerBuffer> pending_destruction;
+    std::queue<std::shared_ptr<Buffer>> pending_destruction;
     u64 epoch = 0;
     u64 modified_ticks = 0;
 
     std::vector<u8> staging_buffer;
+
     std::list<MapInterval*> marked_for_unregister;
 
     std::shared_ptr<std::unordered_set<MapInterval*>> uncommitted_flushes;

src/video_core/engines/const_buffer_engine_interface.h

@@ -93,6 +93,7 @@ public:
     virtual SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const = 0;
     virtual SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
                                                     u64 offset) const = 0;
+    virtual SamplerDescriptor AccessSampler(u32 handle) const = 0;
     virtual u32 GetBoundBuffer() const = 0;
 
     virtual VideoCore::GuestDriverProfile& AccessGuestDriverProfile() = 0;

src/video_core/engines/kepler_compute.cpp

@@ -92,8 +92,11 @@ SamplerDescriptor KeplerCompute::AccessBindlessSampler(ShaderType stage, u64 con
     ASSERT(stage == ShaderType::Compute);
     const auto& tex_info_buffer = launch_description.const_buffer_config[const_buffer];
     const GPUVAddr tex_info_address = tex_info_buffer.Address() + offset;
+    return AccessSampler(memory_manager.Read<u32>(tex_info_address));
+}
 
-    const Texture::TextureHandle tex_handle{memory_manager.Read<u32>(tex_info_address)};
+SamplerDescriptor KeplerCompute::AccessSampler(u32 handle) const {
+    const Texture::TextureHandle tex_handle{handle};
     const Texture::FullTextureInfo tex_info = GetTextureInfo(tex_handle);
     SamplerDescriptor result = SamplerDescriptor::FromTIC(tex_info.tic);
     result.is_shadow.Assign(tex_info.tsc.depth_compare_enabled.Value());

src/video_core/engines/kepler_compute.h

@@ -219,6 +219,8 @@ public:
     SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
                                             u64 offset) const override;
 
+    SamplerDescriptor AccessSampler(u32 handle) const override;
+
     u32 GetBoundBuffer() const override {
         return regs.tex_cb_index;
     }

src/video_core/engines/maxwell_3d.cpp

@@ -25,9 +25,8 @@ constexpr u32 MacroRegistersStart = 0xE00;
 Maxwell3D::Maxwell3D(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
                      MemoryManager& memory_manager)
     : system{system}, rasterizer{rasterizer}, memory_manager{memory_manager},
-      macro_interpreter{*this}, upload_state{memory_manager, regs.upload} {
+      macro_engine{GetMacroEngine(*this)}, upload_state{memory_manager, regs.upload} {
     dirty.flags.flip();
-
     InitializeRegisterDefaults();
 }
 
@@ -106,7 +105,11 @@ void Maxwell3D::InitializeRegisterDefaults() {
     regs.rasterize_enable = 1;
     regs.rt_separate_frag_data = 1;
     regs.framebuffer_srgb = 1;
+    regs.line_width_aliased = 1.0f;
+    regs.line_width_smooth = 1.0f;
     regs.front_face = Maxwell3D::Regs::FrontFace::ClockWise;
+    regs.polygon_mode_back = Maxwell3D::Regs::PolygonMode::Fill;
+    regs.polygon_mode_front = Maxwell3D::Regs::PolygonMode::Fill;
 
     shadow_state = regs;
 
@@ -116,7 +119,7 @@ void Maxwell3D::InitializeRegisterDefaults() {
     mme_inline[MAXWELL3D_REG_INDEX(index_array.count)] = true;
 }
 
-void Maxwell3D::CallMacroMethod(u32 method, std::size_t num_parameters, const u32* parameters) {
+void Maxwell3D::CallMacroMethod(u32 method, const std::vector<u32>& parameters) {
     // Reset the current macro.
     executing_macro = 0;
 
@@ -125,7 +128,7 @@ void Maxwell3D::CallMacroMethod(u32 method, std::size_t num_parameters, const u3
         ((method - MacroRegistersStart) >> 1) % static_cast<u32>(macro_positions.size());
 
     // Execute the current macro.
-    macro_interpreter.Execute(macro_positions[entry], num_parameters, parameters);
+    macro_engine->Execute(*this, macro_positions[entry], parameters);
     if (mme_draw.current_mode != MMEDrawMode::Undefined) {
         FlushMMEInlineDraw();
     }
@@ -161,7 +164,7 @@ void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
 
         // Call the macro when there are no more parameters in the command buffer
         if (is_last_call) {
-            CallMacroMethod(executing_macro, macro_params.size(), macro_params.data());
+            CallMacroMethod(executing_macro, macro_params);
             macro_params.clear();
         }
         return;
@@ -197,7 +200,7 @@ void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
         break;
     }
     case MAXWELL3D_REG_INDEX(macros.data): {
-        ProcessMacroUpload(arg);
+        macro_engine->AddCode(regs.macros.upload_address, arg);
         break;
     }
     case MAXWELL3D_REG_INDEX(macros.bind): {
@@ -306,7 +309,7 @@ void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
 
         // Call the macro when there are no more parameters in the command buffer
         if (amount == methods_pending) {
-            CallMacroMethod(executing_macro, macro_params.size(), macro_params.data());
+            CallMacroMethod(executing_macro, macro_params);
             macro_params.clear();
         }
         return;
@@ -420,9 +423,7 @@ void Maxwell3D::FlushMMEInlineDraw() {
 }
 
 void Maxwell3D::ProcessMacroUpload(u32 data) {
-    ASSERT_MSG(regs.macros.upload_address < macro_memory.size(),
-               "upload_address exceeded macro_memory size!");
-    macro_memory[regs.macros.upload_address++] = data;
+    macro_engine->AddCode(regs.macros.upload_address++, data);
 }
 
 void Maxwell3D::ProcessMacroBind(u32 data) {
@@ -739,8 +740,11 @@ SamplerDescriptor Maxwell3D::AccessBindlessSampler(ShaderType stage, u64 const_b
     const auto& shader = state.shader_stages[static_cast<std::size_t>(stage)];
     const auto& tex_info_buffer = shader.const_buffers[const_buffer];
     const GPUVAddr tex_info_address = tex_info_buffer.address + offset;
+    return AccessSampler(memory_manager.Read<u32>(tex_info_address));
+}
 
-    const Texture::TextureHandle tex_handle{memory_manager.Read<u32>(tex_info_address)};
+SamplerDescriptor Maxwell3D::AccessSampler(u32 handle) const {
+    const Texture::TextureHandle tex_handle{handle};
     const Texture::FullTextureInfo tex_info = GetTextureInfo(tex_handle);
     SamplerDescriptor result = SamplerDescriptor::FromTIC(tex_info.tic);
     result.is_shadow.Assign(tex_info.tsc.depth_compare_enabled.Value());

src/video_core/engines/maxwell_3d.h

@@ -23,7 +23,7 @@
 #include "video_core/engines/engine_upload.h"
 #include "video_core/engines/shader_type.h"
 #include "video_core/gpu.h"
-#include "video_core/macro_interpreter.h"
+#include "video_core/macro/macro.h"
 #include "video_core/textures/texture.h"
 
 namespace Core {
@@ -598,6 +598,7 @@ public:
                 BitField<4, 3, u32> block_height;
                 BitField<8, 3, u32> block_depth;
                 BitField<12, 1, InvMemoryLayout> type;
+                BitField<16, 1, u32> is_3d;
             } memory_layout;
             union {
                 BitField<0, 16, u32> layers;
@@ -1403,6 +1404,8 @@ public:
     SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
                                             u64 offset) const override;
 
+    SamplerDescriptor AccessSampler(u32 handle) const override;
+
     u32 GetBoundBuffer() const override {
         return regs.tex_cb_index;
     }
@@ -1411,17 +1414,16 @@ public:
 
     const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const override;
 
-    /// Memory for macro code - it's undetermined how big this is, however 1MB is much larger than
-    /// we've seen used.
-    using MacroMemory = std::array<u32, 0x40000>;
+    bool ShouldExecute() const {
+        return execute_on;
+    }
 
-    /// Gets a reference to macro memory.
-    const MacroMemory& GetMacroMemory() const {
-        return macro_memory;
+    VideoCore::RasterizerInterface& GetRasterizer() {
+        return rasterizer;
     }
 
-    bool ShouldExecute() const {
-        return execute_on;
+    const VideoCore::RasterizerInterface& GetRasterizer() const {
+        return rasterizer;
     }
 
     /// Notify a memory write has happened.
@@ -1468,16 +1470,13 @@ private:
 
     std::array<bool, Regs::NUM_REGS> mme_inline{};
 
-    /// Memory for macro code
-    MacroMemory macro_memory;
-
     /// Macro method that is currently being executed / being fed parameters.
     u32 executing_macro = 0;
     /// Parameters that have been submitted to the macro call so far.
     std::vector<u32> macro_params;
 
     /// Interpreter for the macro codes uploaded to the GPU.
-    MacroInterpreter macro_interpreter;
+    std::unique_ptr<MacroEngine> macro_engine;
 
     static constexpr u32 null_cb_data = 0xFFFFFFFF;
     struct {
@@ -1506,7 +1505,7 @@ private:
      * @param num_parameters Number of arguments
      * @param parameters Arguments to the method call
      */
-    void CallMacroMethod(u32 method, std::size_t num_parameters, const u32* parameters);
+    void CallMacroMethod(u32 method, const std::vector<u32>& parameters);
 
     /// Handles writes to the macro uploading register.
     void ProcessMacroUpload(u32 data);

src/video_core/macro/macro.cpp

@@ -0,0 +1,72 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <boost/container_hash/hash.hpp>
+#include "common/assert.h"
+#include "common/logging/log.h"
+#include "core/settings.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/macro/macro.h"
+#include "video_core/macro/macro_hle.h"
+#include "video_core/macro/macro_interpreter.h"
+#include "video_core/macro/macro_jit_x64.h"
+
+namespace Tegra {
+
+MacroEngine::MacroEngine(Engines::Maxwell3D& maxwell3d)
+    : hle_macros{std::make_unique<Tegra::HLEMacro>(maxwell3d)} {}
+
+MacroEngine::~MacroEngine() = default;
+
+void MacroEngine::AddCode(u32 method, u32 data) {
+    uploaded_macro_code[method].push_back(data);
+}
+
+void MacroEngine::Execute(Engines::Maxwell3D& maxwell3d, u32 method,
+                          const std::vector<u32>& parameters) {
+    auto compiled_macro = macro_cache.find(method);
+    if (compiled_macro != macro_cache.end()) {
+        const auto& cache_info = compiled_macro->second;
+        if (cache_info.has_hle_program) {
+            cache_info.hle_program->Execute(parameters, method);
+        } else {
+            cache_info.lle_program->Execute(parameters, method);
+        }
+    } else {
+        // Macro not compiled, check if it's uploaded and if so, compile it
+        auto macro_code = uploaded_macro_code.find(method);
+        if (macro_code == uploaded_macro_code.end()) {
+            UNREACHABLE_MSG("Macro 0x{0:x} was not uploaded", method);
+            return;
+        }
+        auto& cache_info = macro_cache[method];
+        cache_info.hash = boost::hash_value(macro_code->second);
+        cache_info.lle_program = Compile(macro_code->second);
+
+        auto hle_program = hle_macros->GetHLEProgram(cache_info.hash);
+        if (hle_program.has_value()) {
+            cache_info.has_hle_program = true;
+            cache_info.hle_program = std::move(hle_program.value());
+        }
+
+        if (cache_info.has_hle_program) {
+            cache_info.hle_program->Execute(parameters, method);
+        } else {
+            cache_info.lle_program->Execute(parameters, method);
+        }
+    }
+}
+
+std::unique_ptr<MacroEngine> GetMacroEngine(Engines::Maxwell3D& maxwell3d) {
+    if (Settings::values.disable_macro_jit) {
+        return std::make_unique<MacroInterpreter>(maxwell3d);
+    }
+#ifdef ARCHITECTURE_x86_64
+    return std::make_unique<MacroJITx64>(maxwell3d);
+#else
+    return std::make_unique<MacroInterpreter>(maxwell3d);
+#endif
+}
+
+} // namespace Tegra

src/video_core/macro/macro.h

@@ -0,0 +1,141 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+#include <unordered_map>
+#include <vector>
+#include "common/bit_field.h"
+#include "common/common_types.h"
+
+namespace Tegra {
+
+namespace Engines {
+class Maxwell3D;
+}
+
+namespace Macro {
+constexpr std::size_t NUM_MACRO_REGISTERS = 8;
+enum class Operation : u32 {
+    ALU = 0,
+    AddImmediate = 1,
+    ExtractInsert = 2,
+    ExtractShiftLeftImmediate = 3,
+    ExtractShiftLeftRegister = 4,
+    Read = 5,
+    Unused = 6, // This operation doesn't seem to be a valid encoding.
+    Branch = 7,
+};
+
+enum class ALUOperation : u32 {
+    Add = 0,
+    AddWithCarry = 1,
+    Subtract = 2,
+    SubtractWithBorrow = 3,
+    // Operations 4-7 don't seem to be valid encodings.
+    Xor = 8,
+    Or = 9,
+    And = 10,
+    AndNot = 11,
+    Nand = 12
+};
+
+enum class ResultOperation : u32 {
+    IgnoreAndFetch = 0,
+    Move = 1,
+    MoveAndSetMethod = 2,
+    FetchAndSend = 3,
+    MoveAndSend = 4,
+    FetchAndSetMethod = 5,
+    MoveAndSetMethodFetchAndSend = 6,
+    MoveAndSetMethodSend = 7
+};
+
+enum class BranchCondition : u32 {
+    Zero = 0,
+    NotZero = 1,
+};
+
+union Opcode {
+    u32 raw;
+    BitField<0, 3, Operation> operation;
+    BitField<4, 3, ResultOperation> result_operation;
+    BitField<4, 1, BranchCondition> branch_condition;
+    // If set on a branch, then the branch doesn't have a delay slot.
+    BitField<5, 1, u32> branch_annul;
+    BitField<7, 1, u32> is_exit;
+    BitField<8, 3, u32> dst;
+    BitField<11, 3, u32> src_a;
+    BitField<14, 3, u32> src_b;
+    // The signed immediate overlaps the second source operand and the alu operation.
+    BitField<14, 18, s32> immediate;
+
+    BitField<17, 5, ALUOperation> alu_operation;
+
+    // Bitfield instructions data
+    BitField<17, 5, u32> bf_src_bit;
+    BitField<22, 5, u32> bf_size;
+    BitField<27, 5, u32> bf_dst_bit;
+
+    u32 GetBitfieldMask() const {
+        return (1 << bf_size) - 1;
+    }
+
+    s32 GetBranchTarget() const {
+        return static_cast<s32>(immediate * sizeof(u32));
+    }
+};
+
+union MethodAddress {
+    u32 raw;
+    BitField<0, 12, u32> address;
+    BitField<12, 6, u32> increment;
+};
+
+} // namespace Macro
+
+class HLEMacro;
+
+class CachedMacro {
+public:
+    virtual ~CachedMacro() = default;
+    /**
+     * Executes the macro code with the specified input parameters.
+     * @param code The macro byte code to execute
+     * @param parameters The parameters of the macro
+     */
+    virtual void Execute(const std::vector<u32>& parameters, u32 method) = 0;
+};
+
+class MacroEngine {
+public:
+    explicit MacroEngine(Engines::Maxwell3D& maxwell3d);
+    virtual ~MacroEngine();
+
+    // Store the uploaded macro code to compile them when they're called.
+    void AddCode(u32 method, u32 data);
+
+    // Compiles the macro if its not in the cache, and executes the compiled macro
+    void Execute(Engines::Maxwell3D& maxwell3d, u32 method, const std::vector<u32>& parameters);
+
+protected:
+    virtual std::unique_ptr<CachedMacro> Compile(const std::vector<u32>& code) = 0;
+
+private:
+    struct CacheInfo {
+        std::unique_ptr<CachedMacro> lle_program{};
+        std::unique_ptr<CachedMacro> hle_program{};
+        u64 hash{};
+        bool has_hle_program{};
+    };
+
+    std::unordered_map<u32, CacheInfo> macro_cache;
+    std::unordered_map<u32, std::vector<u32>> uploaded_macro_code;
+    std::unique_ptr<HLEMacro> hle_macros;
+};
+
+std::unique_ptr<MacroEngine> GetMacroEngine(Engines::Maxwell3D& maxwell3d);
+
+} // namespace Tegra

src/video_core/macro/macro_hle.cpp

@@ -0,0 +1,113 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <array>
+#include <vector>
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/macro/macro_hle.h"
+#include "video_core/rasterizer_interface.h"
+
+namespace Tegra {
+
+namespace {
+// HLE'd functions
+static void HLE_771BB18C62444DA0(Engines::Maxwell3D& maxwell3d,
+                                 const std::vector<u32>& parameters) {
+    const u32 instance_count = parameters[2] & maxwell3d.GetRegisterValue(0xD1B);
+
+    maxwell3d.regs.draw.topology.Assign(
+        static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0] &
+                                                                        ~(0x3ffffff << 26)));
+    maxwell3d.regs.vb_base_instance = parameters[5];
+    maxwell3d.mme_draw.instance_count = instance_count;
+    maxwell3d.regs.vb_element_base = parameters[3];
+    maxwell3d.regs.index_array.count = parameters[1];
+    maxwell3d.regs.index_array.first = parameters[4];
+
+    if (maxwell3d.ShouldExecute()) {
+        maxwell3d.GetRasterizer().Draw(true, true);
+    }
+    maxwell3d.regs.index_array.count = 0;
+    maxwell3d.mme_draw.instance_count = 0;
+    maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
+}
+
+static void HLE_0D61FC9FAAC9FCAD(Engines::Maxwell3D& maxwell3d,
+                                 const std::vector<u32>& parameters) {
+    const u32 count = (maxwell3d.GetRegisterValue(0xD1B) & parameters[2]);
+
+    maxwell3d.regs.vertex_buffer.first = parameters[3];
+    maxwell3d.regs.vertex_buffer.count = parameters[1];
+    maxwell3d.regs.vb_base_instance = parameters[4];
+    maxwell3d.regs.draw.topology.Assign(
+        static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0]));
+    maxwell3d.mme_draw.instance_count = count;
+
+    if (maxwell3d.ShouldExecute()) {
+        maxwell3d.GetRasterizer().Draw(false, true);
+    }
+    maxwell3d.regs.vertex_buffer.count = 0;
+    maxwell3d.mme_draw.instance_count = 0;
+    maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
+}
+
+static void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d,
+                                 const std::vector<u32>& parameters) {
+    const u32 instance_count = (maxwell3d.GetRegisterValue(0xD1B) & parameters[2]);
+    const u32 element_base = parameters[4];
+    const u32 base_instance = parameters[5];
+    maxwell3d.regs.index_array.first = parameters[3];
+    maxwell3d.regs.reg_array[0x446] = element_base; // vertex id base?
+    maxwell3d.regs.index_array.count = parameters[1];
+    maxwell3d.regs.vb_element_base = element_base;
+    maxwell3d.regs.vb_base_instance = base_instance;
+    maxwell3d.mme_draw.instance_count = instance_count;
+    maxwell3d.CallMethodFromMME(0x8e3, 0x640);
+    maxwell3d.CallMethodFromMME(0x8e4, element_base);
+    maxwell3d.CallMethodFromMME(0x8e5, base_instance);
+    maxwell3d.regs.draw.topology.Assign(
+        static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0]));
+    if (maxwell3d.ShouldExecute()) {
+        maxwell3d.GetRasterizer().Draw(true, true);
+    }
+    maxwell3d.regs.reg_array[0x446] = 0x0; // vertex id base?
+    maxwell3d.regs.index_array.count = 0;
+    maxwell3d.regs.vb_element_base = 0x0;
+    maxwell3d.regs.vb_base_instance = 0x0;
+    maxwell3d.mme_draw.instance_count = 0;
+    maxwell3d.CallMethodFromMME(0x8e3, 0x640);
+    maxwell3d.CallMethodFromMME(0x8e4, 0x0);
+    maxwell3d.CallMethodFromMME(0x8e5, 0x0);
+    maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
+}
+} // namespace
+
+constexpr std::array<std::pair<u64, HLEFunction>, 3> hle_funcs{{
+    std::make_pair<u64, HLEFunction>(0x771BB18C62444DA0, &HLE_771BB18C62444DA0),
+    std::make_pair<u64, HLEFunction>(0x0D61FC9FAAC9FCAD, &HLE_0D61FC9FAAC9FCAD),
+    std::make_pair<u64, HLEFunction>(0x0217920100488FF7, &HLE_0217920100488FF7),
+}};
+
+HLEMacro::HLEMacro(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {}
+HLEMacro::~HLEMacro() = default;
+
+std::optional<std::unique_ptr<CachedMacro>> HLEMacro::GetHLEProgram(u64 hash) const {
+    const auto it = std::find_if(hle_funcs.cbegin(), hle_funcs.cend(),
+                                 [hash](const auto& pair) { return pair.first == hash; });
+    if (it == hle_funcs.end()) {
+        return std::nullopt;
+    }
+    return std::make_unique<HLEMacroImpl>(maxwell3d, it->second);
+}
+
+HLEMacroImpl::~HLEMacroImpl() = default;
+
+HLEMacroImpl::HLEMacroImpl(Engines::Maxwell3D& maxwell3d, HLEFunction func)
+    : maxwell3d(maxwell3d), func(func) {}
+
+void HLEMacroImpl::Execute(const std::vector<u32>& parameters, u32 method) {
+    func(maxwell3d, parameters);
+}
+
+} // namespace Tegra

src/video_core/macro/macro_hle.h

@@ -0,0 +1,44 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+#include <optional>
+#include <vector>
+#include "common/common_types.h"
+#include "video_core/macro/macro.h"
+
+namespace Tegra {
+
+namespace Engines {
+class Maxwell3D;
+}
+
+using HLEFunction = void (*)(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& parameters);
+
+class HLEMacro {
+public:
+    explicit HLEMacro(Engines::Maxwell3D& maxwell3d);
+    ~HLEMacro();
+
+    std::optional<std::unique_ptr<CachedMacro>> GetHLEProgram(u64 hash) const;
+
+private:
+    Engines::Maxwell3D& maxwell3d;
+};
+
+class HLEMacroImpl : public CachedMacro {
+public:
+    explicit HLEMacroImpl(Engines::Maxwell3D& maxwell3d, HLEFunction func);
+    ~HLEMacroImpl();
+
+    void Execute(const std::vector<u32>& parameters, u32 method) override;
+
+private:
+    Engines::Maxwell3D& maxwell3d;
+    HLEFunction func;
+};
+
+} // namespace Tegra

src/video_core/macro_interpreter.cpp → src/video_core/macro/macro_interpreter.cpp

@@ -1,4 +1,4 @@
-// Copyright 2018 yuzu Emulator Project
+// Copyright 2020 yuzu Emulator Project
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
@@ -6,109 +6,47 @@
 #include "common/logging/log.h"
 #include "common/microprofile.h"
 #include "video_core/engines/maxwell_3d.h"
-#include "video_core/macro_interpreter.h"
+#include "video_core/macro/macro_interpreter.h"
 
 MICROPROFILE_DEFINE(MacroInterp, "GPU", "Execute macro interpreter", MP_RGB(128, 128, 192));
 
 namespace Tegra {
-namespace {
-enum class Operation : u32 {
-    ALU = 0,
-    AddImmediate = 1,
-    ExtractInsert = 2,
-    ExtractShiftLeftImmediate = 3,
-    ExtractShiftLeftRegister = 4,
-    Read = 5,
-    Unused = 6, // This operation doesn't seem to be a valid encoding.
-    Branch = 7,
-};
-} // Anonymous namespace
-
-enum class MacroInterpreter::ALUOperation : u32 {
-    Add = 0,
-    AddWithCarry = 1,
-    Subtract = 2,
-    SubtractWithBorrow = 3,
-    // Operations 4-7 don't seem to be valid encodings.
-    Xor = 8,
-    Or = 9,
-    And = 10,
-    AndNot = 11,
-    Nand = 12
-};
-
-enum class MacroInterpreter::ResultOperation : u32 {
-    IgnoreAndFetch = 0,
-    Move = 1,
-    MoveAndSetMethod = 2,
-    FetchAndSend = 3,
-    MoveAndSend = 4,
-    FetchAndSetMethod = 5,
-    MoveAndSetMethodFetchAndSend = 6,
-    MoveAndSetMethodSend = 7
-};
-
-enum class MacroInterpreter::BranchCondition : u32 {
-    Zero = 0,
-    NotZero = 1,
-};
-
-union MacroInterpreter::Opcode {
-    u32 raw;
-    BitField<0, 3, Operation> operation;
-    BitField<4, 3, ResultOperation> result_operation;
-    BitField<4, 1, BranchCondition> branch_condition;
-    // If set on a branch, then the branch doesn't have a delay slot.
-    BitField<5, 1, u32> branch_annul;
-    BitField<7, 1, u32> is_exit;
-    BitField<8, 3, u32> dst;
-    BitField<11, 3, u32> src_a;
-    BitField<14, 3, u32> src_b;
-    // The signed immediate overlaps the second source operand and the alu operation.
-    BitField<14, 18, s32> immediate;
-
-    BitField<17, 5, ALUOperation> alu_operation;
-
-    // Bitfield instructions data
-    BitField<17, 5, u32> bf_src_bit;
-    BitField<22, 5, u32> bf_size;
-    BitField<27, 5, u32> bf_dst_bit;
-
-    u32 GetBitfieldMask() const {
-        return (1 << bf_size) - 1;
-    }
+MacroInterpreter::MacroInterpreter(Engines::Maxwell3D& maxwell3d)
+    : MacroEngine::MacroEngine(maxwell3d), maxwell3d(maxwell3d) {}
 
-    s32 GetBranchTarget() const {
-        return static_cast<s32>(immediate * sizeof(u32));
-    }
-};
+std::unique_ptr<CachedMacro> MacroInterpreter::Compile(const std::vector<u32>& code) {
+    return std::make_unique<MacroInterpreterImpl>(maxwell3d, code);
+}
 
-MacroInterpreter::MacroInterpreter(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {}
+MacroInterpreterImpl::MacroInterpreterImpl(Engines::Maxwell3D& maxwell3d,
+                                           const std::vector<u32>& code)
+    : maxwell3d(maxwell3d), code(code) {}
 
-void MacroInterpreter::Execute(u32 offset, std::size_t num_parameters, const u32* parameters) {
+void MacroInterpreterImpl::Execute(const std::vector<u32>& parameters, u32 method) {
     MICROPROFILE_SCOPE(MacroInterp);
     Reset();
 
     registers[1] = parameters[0];
+    num_parameters = parameters.size();
 
     if (num_parameters > parameters_capacity) {
         parameters_capacity = num_parameters;
         this->parameters = std::make_unique<u32[]>(num_parameters);
     }
-    std::memcpy(this->parameters.get(), parameters, num_parameters * sizeof(u32));
+    std::memcpy(this->parameters.get(), parameters.data(), num_parameters * sizeof(u32));
     this->num_parameters = num_parameters;
 
     // Execute the code until we hit an exit condition.
     bool keep_executing = true;
     while (keep_executing) {
-        keep_executing = Step(offset, false);
+        keep_executing = Step(false);
     }
 
     // Assert the the macro used all the input parameters
     ASSERT(next_parameter_index == num_parameters);
 }
 
-void MacroInterpreter::Reset() {
+void MacroInterpreterImpl::Reset() {
     registers = {};
     pc = 0;
     delayed_pc = {};
@@ -120,10 +58,10 @@ void MacroInterpreter::Reset() {
     carry_flag = false;
 }
 
-bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
+bool MacroInterpreterImpl::Step(bool is_delay_slot) {
     u32 base_address = pc;
 
-    Opcode opcode = GetOpcode(offset);
+    Macro::Opcode opcode = GetOpcode();
     pc += 4;
 
     // Update the program counter if we were delayed
@@ -134,18 +72,18 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
     }
 
     switch (opcode.operation) {
-    case Operation::ALU: {
+    case Macro::Operation::ALU: {
         u32 result = GetALUResult(opcode.alu_operation, GetRegister(opcode.src_a),
                                   GetRegister(opcode.src_b));
         ProcessResult(opcode.result_operation, opcode.dst, result);
         break;
     }
-    case Operation::AddImmediate: {
+    case Macro::Operation::AddImmediate: {
         ProcessResult(opcode.result_operation, opcode.dst,
                       GetRegister(opcode.src_a) + opcode.immediate);
         break;
     }
-    case Operation::ExtractInsert: {
+    case Macro::Operation::ExtractInsert: {
         u32 dst = GetRegister(opcode.src_a);
         u32 src = GetRegister(opcode.src_b);
 
@@ -155,7 +93,7 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
         ProcessResult(opcode.result_operation, opcode.dst, dst);
         break;
     }
-    case Operation::ExtractShiftLeftImmediate: {
+    case Macro::Operation::ExtractShiftLeftImmediate: {
         u32 dst = GetRegister(opcode.src_a);
         u32 src = GetRegister(opcode.src_b);
 
@@ -164,7 +102,7 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
         ProcessResult(opcode.result_operation, opcode.dst, result);
         break;
     }
-    case Operation::ExtractShiftLeftRegister: {
+    case Macro::Operation::ExtractShiftLeftRegister: {
         u32 dst = GetRegister(opcode.src_a);
         u32 src = GetRegister(opcode.src_b);
 
@@ -173,12 +111,12 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
         ProcessResult(opcode.result_operation, opcode.dst, result);
         break;
     }
-    case Operation::Read: {
+    case Macro::Operation::Read: {
         u32 result = Read(GetRegister(opcode.src_a) + opcode.immediate);
         ProcessResult(opcode.result_operation, opcode.dst, result);
         break;
     }
-    case Operation::Branch: {
+    case Macro::Operation::Branch: {
         ASSERT_MSG(!is_delay_slot, "Executing a branch in a delay slot is not valid");
         u32 value = GetRegister(opcode.src_a);
         bool taken = EvaluateBranchCondition(opcode.branch_condition, value);
@@ -191,7 +129,7 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
 
             delayed_pc = base_address + opcode.GetBranchTarget();
             // Execute one more instruction due to the delay slot.
-            return Step(offset, true);
+            return Step(true);
         }
         break;
     }
@@ -204,51 +142,44 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
     // cause an exit if it's executed inside a delay slot.
     if (opcode.is_exit && !is_delay_slot) {
         // Exit has a delay slot, execute the next instruction
-        Step(offset, true);
+        Step(true);
         return false;
     }
 
     return true;
 }
 
-MacroInterpreter::Opcode MacroInterpreter::GetOpcode(u32 offset) const {
-    const auto& macro_memory{maxwell3d.GetMacroMemory()};
-    ASSERT((pc % sizeof(u32)) == 0);
-    ASSERT((pc + offset) < macro_memory.size() * sizeof(u32));
-    return {macro_memory[offset + pc / sizeof(u32)]};
-}
-
-u32 MacroInterpreter::GetALUResult(ALUOperation operation, u32 src_a, u32 src_b) {
+u32 MacroInterpreterImpl::GetALUResult(Macro::ALUOperation operation, u32 src_a, u32 src_b) {
     switch (operation) {
-    case ALUOperation::Add: {
+    case Macro::ALUOperation::Add: {
         const u64 result{static_cast<u64>(src_a) + src_b};
         carry_flag = result > 0xffffffff;
         return static_cast<u32>(result);
     }
-    case ALUOperation::AddWithCarry: {
+    case Macro::ALUOperation::AddWithCarry: {
         const u64 result{static_cast<u64>(src_a) + src_b + (carry_flag ? 1ULL : 0ULL)};
         carry_flag = result > 0xffffffff;
         return static_cast<u32>(result);
     }
-    case ALUOperation::Subtract: {
+    case Macro::ALUOperation::Subtract: {
         const u64 result{static_cast<u64>(src_a) - src_b};
         carry_flag = result < 0x100000000;
         return static_cast<u32>(result);
     }
-    case ALUOperation::SubtractWithBorrow: {
+    case Macro::ALUOperation::SubtractWithBorrow: {
         const u64 result{static_cast<u64>(src_a) - src_b - (carry_flag ? 0ULL : 1ULL)};
         carry_flag = result < 0x100000000;
         return static_cast<u32>(result);
     }
-    case ALUOperation::Xor:
+    case Macro::ALUOperation::Xor:
         return src_a ^ src_b;
-    case ALUOperation::Or:
+    case Macro::ALUOperation::Or:
         return src_a | src_b;
-    case ALUOperation::And:
+    case Macro::ALUOperation::And:
         return src_a & src_b;
-    case ALUOperation::AndNot:
+    case Macro::ALUOperation::AndNot:
         return src_a & ~src_b;
-    case ALUOperation::Nand:
+    case Macro::ALUOperation::Nand:
         return ~(src_a & src_b);
 
     default:
@@ -257,43 +188,43 @@ u32 MacroInterpreter::GetALUResult(ALUOperation operation, u32 src_a, u32 src_b)
     }
 }
 
-void MacroInterpreter::ProcessResult(ResultOperation operation, u32 reg, u32 result) {
+void MacroInterpreterImpl::ProcessResult(Macro::ResultOperation operation, u32 reg, u32 result) {
     switch (operation) {
-    case ResultOperation::IgnoreAndFetch:
+    case Macro::ResultOperation::IgnoreAndFetch:
         // Fetch parameter and ignore result.
         SetRegister(reg, FetchParameter());
         break;
-    case ResultOperation::Move:
+    case Macro::ResultOperation::Move:
         // Move result.
         SetRegister(reg, result);
         break;
-    case ResultOperation::MoveAndSetMethod:
+    case Macro::ResultOperation::MoveAndSetMethod:
         // Move result and use as Method Address.
         SetRegister(reg, result);
         SetMethodAddress(result);
         break;
-    case ResultOperation::FetchAndSend:
+    case Macro::ResultOperation::FetchAndSend:
         // Fetch parameter and send result.
         SetRegister(reg, FetchParameter());
         Send(result);
         break;
-    case ResultOperation::MoveAndSend:
+    case Macro::ResultOperation::MoveAndSend:
         // Move and send result.
         SetRegister(reg, result);
         Send(result);
         break;
-    case ResultOperation::FetchAndSetMethod:
+    case Macro::ResultOperation::FetchAndSetMethod:
         // Fetch parameter and use result as Method Address.
         SetRegister(reg, FetchParameter());
         SetMethodAddress(result);
         break;
-    case ResultOperation::MoveAndSetMethodFetchAndSend:
+    case Macro::ResultOperation::MoveAndSetMethodFetchAndSend:
         // Move result and use as Method Address, then fetch and send parameter.
         SetRegister(reg, result);
         SetMethodAddress(result);
         Send(FetchParameter());
         break;
-    case ResultOperation::MoveAndSetMethodSend:
+    case Macro::ResultOperation::MoveAndSetMethodSend:
         // Move result and use as Method Address, then send bits 12:17 of result.
         SetRegister(reg, result);
         SetMethodAddress(result);
@@ -304,16 +235,28 @@ void MacroInterpreter::ProcessResult(ResultOperation operation, u32 reg, u32 res
     }
 }
 
-u32 MacroInterpreter::FetchParameter() {
-    ASSERT(next_parameter_index < num_parameters);
-    return parameters[next_parameter_index++];
+bool MacroInterpreterImpl::EvaluateBranchCondition(Macro::BranchCondition cond, u32 value) const {
+    switch (cond) {
+    case Macro::BranchCondition::Zero:
+        return value == 0;
+    case Macro::BranchCondition::NotZero:
+        return value != 0;
+    }
+    UNREACHABLE();
+    return true;
+}
+
+Macro::Opcode MacroInterpreterImpl::GetOpcode() const {
+    ASSERT((pc % sizeof(u32)) == 0);
+    ASSERT(pc < code.size() * sizeof(u32));
+    return {code[pc / sizeof(u32)]};
 }
 
-u32 MacroInterpreter::GetRegister(u32 register_id) const {
+u32 MacroInterpreterImpl::GetRegister(u32 register_id) const {
     return registers.at(register_id);
 }
 
-void MacroInterpreter::SetRegister(u32 register_id, u32 value) {
+void MacroInterpreterImpl::SetRegister(u32 register_id, u32 value) {
     // Register 0 is hardwired as the zero register.
     // Ensure no writes to it actually occur.
     if (register_id == 0) {
@@ -323,30 +266,24 @@ void MacroInterpreter::SetRegister(u32 register_id, u32 value) {
     registers.at(register_id) = value;
 }
 
-void MacroInterpreter::SetMethodAddress(u32 address) {
+void MacroInterpreterImpl::SetMethodAddress(u32 address) {
     method_address.raw = address;
 }
 
-void MacroInterpreter::Send(u32 value) {
+void MacroInterpreterImpl::Send(u32 value) {
     maxwell3d.CallMethodFromMME(method_address.address, value);
     // Increment the method address by the method increment.
     method_address.address.Assign(method_address.address.Value() +
                                   method_address.increment.Value());
 }
 
-u32 MacroInterpreter::Read(u32 method) const {
+u32 MacroInterpreterImpl::Read(u32 method) const {
     return maxwell3d.GetRegisterValue(method);
 }
 
-bool MacroInterpreter::EvaluateBranchCondition(BranchCondition cond, u32 value) const {
-    switch (cond) {
-    case BranchCondition::Zero:
-        return value == 0;
-    case BranchCondition::NotZero:
-        return value != 0;
-    }
-    UNREACHABLE();
-    return true;
+u32 MacroInterpreterImpl::FetchParameter() {
+    ASSERT(next_parameter_index < num_parameters);
+    return parameters[next_parameter_index++];
 }
 
 } // namespace Tegra

src/video_core/macro_interpreter.h → src/video_core/macro/macro_interpreter.h

@@ -1,44 +1,37 @@
-// Copyright 2018 yuzu Emulator Project
+// Copyright 2020 yuzu Emulator Project
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
 #pragma once
-
 #include <array>
 #include <optional>
-
+#include <vector>
 #include "common/bit_field.h"
 #include "common/common_types.h"
+#include "video_core/macro/macro.h"
 
 namespace Tegra {
 namespace Engines {
 class Maxwell3D;
 }
 
-class MacroInterpreter final {
+class MacroInterpreter final : public MacroEngine {
 public:
     explicit MacroInterpreter(Engines::Maxwell3D& maxwell3d);
 
-    /**
-     * Executes the macro code with the specified input parameters.
-     * @param offset Offset to start execution at.
-     * @param parameters The parameters of the macro.
-     */
-    void Execute(u32 offset, std::size_t num_parameters, const u32* parameters);
+protected:
+    std::unique_ptr<CachedMacro> Compile(const std::vector<u32>& code) override;
 
 private:
-    enum class ALUOperation : u32;
-    enum class BranchCondition : u32;
-    enum class ResultOperation : u32;
-
-    union Opcode;
+    Engines::Maxwell3D& maxwell3d;
+};
 
-    union MethodAddress {
-        u32 raw;
-        BitField<0, 12, u32> address;
-        BitField<12, 6, u32> increment;
-    };
+class MacroInterpreterImpl : public CachedMacro {
+public:
+    MacroInterpreterImpl(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& code);
+    void Execute(const std::vector<u32>& parameters, u32 method) override;
 
+private:
     /// Resets the execution engine state, zeroing registers, etc.
     void Reset();
 
@@ -49,20 +42,20 @@ private:
      * @param is_delay_slot Whether the current step is being executed due to a delay slot in a
      * previous instruction.
      */
-    bool Step(u32 offset, bool is_delay_slot);
+    bool Step(bool is_delay_slot);
 
     /// Calculates the result of an ALU operation. src_a OP src_b;
-    u32 GetALUResult(ALUOperation operation, u32 src_a, u32 src_b);
+    u32 GetALUResult(Macro::ALUOperation operation, u32 src_a, u32 src_b);
 
     /// Performs the result operation on the input result and stores it in the specified register
     /// (if necessary).
-    void ProcessResult(ResultOperation operation, u32 reg, u32 result);
+    void ProcessResult(Macro::ResultOperation operation, u32 reg, u32 result);
 
     /// Evaluates the branch condition and returns whether the branch should be taken or not.
-    bool EvaluateBranchCondition(BranchCondition cond, u32 value) const;
+    bool EvaluateBranchCondition(Macro::BranchCondition cond, u32 value) const;
 
     /// Reads an opcode at the current program counter location.
-    Opcode GetOpcode(u32 offset) const;
+    Macro::Opcode GetOpcode() const;
 
     /// Returns the specified register's value. Register 0 is hardcoded to always return 0.
     u32 GetRegister(u32 register_id) const;
@@ -89,13 +82,11 @@ private:
     /// Program counter to execute at after the delay slot is executed.
     std::optional<u32> delayed_pc;
 
-    static constexpr std::size_t NumMacroRegisters = 8;
-
     /// General purpose macro registers.
-    std::array<u32, NumMacroRegisters> registers = {};
+    std::array<u32, Macro::NUM_MACRO_REGISTERS> registers = {};
 
     /// Method address to use for the next Send instruction.
-    MethodAddress method_address = {};
+    Macro::MethodAddress method_address = {};
 
     /// Input parameters of the current macro.
     std::unique_ptr<u32[]> parameters;
@@ -105,5 +96,7 @@ private:
     u32 next_parameter_index = 0;
 
     bool carry_flag = false;
+    const std::vector<u32>& code;
 };
+
 } // namespace Tegra

src/video_core/macro/macro_jit_x64.cpp

@@ -0,0 +1,621 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/assert.h"
+#include "common/logging/log.h"
+#include "common/microprofile.h"
+#include "common/x64/xbyak_util.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/macro/macro_interpreter.h"
+#include "video_core/macro/macro_jit_x64.h"
+
+MICROPROFILE_DEFINE(MacroJitCompile, "GPU", "Compile macro JIT", MP_RGB(173, 255, 47));
+MICROPROFILE_DEFINE(MacroJitExecute, "GPU", "Execute macro JIT", MP_RGB(255, 255, 0));
+
+namespace Tegra {
+static const Xbyak::Reg64 STATE = Xbyak::util::rbx;
+static const Xbyak::Reg32 RESULT = Xbyak::util::ebp;
+static const Xbyak::Reg64 PARAMETERS = Xbyak::util::r12;
+static const Xbyak::Reg32 METHOD_ADDRESS = Xbyak::util::r14d;
+static const Xbyak::Reg64 BRANCH_HOLDER = Xbyak::util::r15;
+
+static const std::bitset<32> PERSISTENT_REGISTERS = Common::X64::BuildRegSet({
+    STATE,
+    RESULT,
+    PARAMETERS,
+    METHOD_ADDRESS,
+    BRANCH_HOLDER,
+});
+
+MacroJITx64::MacroJITx64(Engines::Maxwell3D& maxwell3d)
+    : MacroEngine::MacroEngine(maxwell3d), maxwell3d(maxwell3d) {}
+
+std::unique_ptr<CachedMacro> MacroJITx64::Compile(const std::vector<u32>& code) {
+    return std::make_unique<MacroJITx64Impl>(maxwell3d, code);
+}
+
+MacroJITx64Impl::MacroJITx64Impl(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& code)
+    : Xbyak::CodeGenerator(MAX_CODE_SIZE), code(code), maxwell3d(maxwell3d) {
+    Compile();
+}
+
+MacroJITx64Impl::~MacroJITx64Impl() = default;
+
+void MacroJITx64Impl::Execute(const std::vector<u32>& parameters, u32 method) {
+    MICROPROFILE_SCOPE(MacroJitExecute);
+    ASSERT_OR_EXECUTE(program != nullptr, { return; });
+    JITState state{};
+    state.maxwell3d = &maxwell3d;
+    state.registers = {};
+    program(&state, parameters.data());
+}
+
+void MacroJITx64Impl::Compile_ALU(Macro::Opcode opcode) {
+    const bool is_a_zero = opcode.src_a == 0;
+    const bool is_b_zero = opcode.src_b == 0;
+    const bool valid_operation = !is_a_zero && !is_b_zero;
+    [[maybe_unused]] const bool is_move_operation = !is_a_zero && is_b_zero;
+    const bool has_zero_register = is_a_zero || is_b_zero;
+    const bool no_zero_reg_skip = opcode.alu_operation == Macro::ALUOperation::AddWithCarry ||
+                                  opcode.alu_operation == Macro::ALUOperation::SubtractWithBorrow;
+
+    Xbyak::Reg32 src_a;
+    Xbyak::Reg32 src_b;
+
+    if (!optimizer.zero_reg_skip || no_zero_reg_skip) {
+        src_a = Compile_GetRegister(opcode.src_a, RESULT);
+        src_b = Compile_GetRegister(opcode.src_b, eax);
+    } else {
+        if (!is_a_zero) {
+            src_a = Compile_GetRegister(opcode.src_a, RESULT);
+        }
+        if (!is_b_zero) {
+            src_b = Compile_GetRegister(opcode.src_b, eax);
+        }
+    }
+
+    bool has_emitted = false;
+
+    switch (opcode.alu_operation) {
+    case Macro::ALUOperation::Add:
+        if (optimizer.zero_reg_skip) {
+            if (valid_operation) {
+                add(src_a, src_b);
+            }
+        } else {
+            add(src_a, src_b);
+        }
+
+        if (!optimizer.can_skip_carry) {
+            setc(byte[STATE + offsetof(JITState, carry_flag)]);
+        }
+        break;
+    case Macro::ALUOperation::AddWithCarry:
+        bt(dword[STATE + offsetof(JITState, carry_flag)], 0);
+        adc(src_a, src_b);
+        setc(byte[STATE + offsetof(JITState, carry_flag)]);
+        break;
+    case Macro::ALUOperation::Subtract:
+        if (optimizer.zero_reg_skip) {
+            if (valid_operation) {
+                sub(src_a, src_b);
+                has_emitted = true;
+            }
+        } else {
+            sub(src_a, src_b);
+            has_emitted = true;
+        }
+        if (!optimizer.can_skip_carry && has_emitted) {
+            setc(byte[STATE + offsetof(JITState, carry_flag)]);
+        }
+        break;
+    case Macro::ALUOperation::SubtractWithBorrow:
+        bt(dword[STATE + offsetof(JITState, carry_flag)], 0);
+        sbb(src_a, src_b);
+        setc(byte[STATE + offsetof(JITState, carry_flag)]);
+        break;
+    case Macro::ALUOperation::Xor:
+        if (optimizer.zero_reg_skip) {
+            if (valid_operation) {
+                xor_(src_a, src_b);
+            }
+        } else {
+            xor_(src_a, src_b);
+        }
+        break;
+    case Macro::ALUOperation::Or:
+        if (optimizer.zero_reg_skip) {
+            if (valid_operation) {
+                or_(src_a, src_b);
+            }
+        } else {
+            or_(src_a, src_b);
+        }
+        break;
+    case Macro::ALUOperation::And:
+        if (optimizer.zero_reg_skip) {
+            if (!has_zero_register) {
+                and_(src_a, src_b);
+            }
+        } else {
+            and_(src_a, src_b);
+        }
+        break;
+    case Macro::ALUOperation::AndNot:
+        if (optimizer.zero_reg_skip) {
+            if (!is_a_zero) {
+                not_(src_b);
+                and_(src_a, src_b);
+            }
+        } else {
+            not_(src_b);
+            and_(src_a, src_b);
+        }
+        break;
+    case Macro::ALUOperation::Nand:
+        if (optimizer.zero_reg_skip) {
+            if (!is_a_zero) {
+                and_(src_a, src_b);
+                not_(src_a);
+            }
+        } else {
+            and_(src_a, src_b);
+            not_(src_a);
+        }
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented ALU operation {}",
+                          static_cast<std::size_t>(opcode.alu_operation.Value()));
+        break;
+    }
+    Compile_ProcessResult(opcode.result_operation, opcode.dst);
+}
+
+void MacroJITx64Impl::Compile_AddImmediate(Macro::Opcode opcode) {
+    if (optimizer.skip_dummy_addimmediate) {
+        // Games tend to use this as an exit instruction placeholder. It's to encode an instruction
+        // without doing anything. In our case we can just not emit anything.
+        if (opcode.result_operation == Macro::ResultOperation::Move && opcode.dst == 0) {
+            return;
+        }
+    }
+    // Check for redundant moves
+    if (optimizer.optimize_for_method_move &&
+        opcode.result_operation == Macro::ResultOperation::MoveAndSetMethod) {
+        if (next_opcode.has_value()) {
+            const auto next = *next_opcode;
+            if (next.result_operation == Macro::ResultOperation::MoveAndSetMethod &&
+                opcode.dst == next.dst) {
+                return;
+            }
+        }
+    }
+    if (optimizer.zero_reg_skip && opcode.src_a == 0) {
+        if (opcode.immediate == 0) {
+            xor_(RESULT, RESULT);
+        } else {
+            mov(RESULT, opcode.immediate);
+        }
+    } else {
+        auto result = Compile_GetRegister(opcode.src_a, RESULT);
+        if (opcode.immediate > 2) {
+            add(result, opcode.immediate);
+        } else if (opcode.immediate == 1) {
+            inc(result);
+        } else if (opcode.immediate < 0) {
+            sub(result, opcode.immediate * -1);
+        }
+    }
+    Compile_ProcessResult(opcode.result_operation, opcode.dst);
+}
+
+void MacroJITx64Impl::Compile_ExtractInsert(Macro::Opcode opcode) {
+    auto dst = Compile_GetRegister(opcode.src_a, RESULT);
+    auto src = Compile_GetRegister(opcode.src_b, eax);
+
+    if (opcode.bf_src_bit != 0 && opcode.bf_src_bit != 31) {
+        shr(src, opcode.bf_src_bit);
+    } else if (opcode.bf_src_bit == 31) {
+        xor_(src, src);
+    }
+    // Don't bother masking the whole register since we're using a 32 bit register
+    if (opcode.bf_size != 31 && opcode.bf_size != 0) {
+        and_(src, opcode.GetBitfieldMask());
+    } else if (opcode.bf_size == 0) {
+        xor_(src, src);
+    }
+    if (opcode.bf_dst_bit != 31 && opcode.bf_dst_bit != 0) {
+        shl(src, opcode.bf_dst_bit);
+    } else if (opcode.bf_dst_bit == 31) {
+        xor_(src, src);
+    }
+
+    const u32 mask = ~(opcode.GetBitfieldMask() << opcode.bf_dst_bit);
+    if (mask != 0xffffffff) {
+        and_(dst, mask);
+    }
+    or_(dst, src);
+    Compile_ProcessResult(opcode.result_operation, opcode.dst);
+}
+
+void MacroJITx64Impl::Compile_ExtractShiftLeftImmediate(Macro::Opcode opcode) {
+    const auto dst = Compile_GetRegister(opcode.src_a, ecx);
+    const auto src = Compile_GetRegister(opcode.src_b, RESULT);
+
+    shr(src, dst.cvt8());
+    if (opcode.bf_size != 0 && opcode.bf_size != 31) {
+        and_(src, opcode.GetBitfieldMask());
+    } else if (opcode.bf_size == 0) {
+        xor_(src, src);
+    }
+
+    if (opcode.bf_dst_bit != 0 && opcode.bf_dst_bit != 31) {
+        shl(src, opcode.bf_dst_bit);
+    } else if (opcode.bf_dst_bit == 31) {
+        xor_(src, src);
+    }
+    Compile_ProcessResult(opcode.result_operation, opcode.dst);
+}
+
+void MacroJITx64Impl::Compile_ExtractShiftLeftRegister(Macro::Opcode opcode) {
+    const auto dst = Compile_GetRegister(opcode.src_a, ecx);
+    const auto src = Compile_GetRegister(opcode.src_b, RESULT);
+
+    if (opcode.bf_src_bit != 0) {
+        shr(src, opcode.bf_src_bit);
+    }
+
+    if (opcode.bf_size != 31) {
+        and_(src, opcode.GetBitfieldMask());
+    }
+    shl(src, dst.cvt8());
+
+    Compile_ProcessResult(opcode.result_operation, opcode.dst);
+}
+
+void MacroJITx64Impl::Compile_Read(Macro::Opcode opcode) {
+    if (optimizer.zero_reg_skip && opcode.src_a == 0) {
+        if (opcode.immediate == 0) {
+            xor_(RESULT, RESULT);
+        } else {
+            mov(RESULT, opcode.immediate);
+        }
+    } else {
+        auto result = Compile_GetRegister(opcode.src_a, RESULT);
+        if (opcode.immediate > 2) {
+            add(result, opcode.immediate);
+        } else if (opcode.immediate == 1) {
+            inc(result);
+        } else if (opcode.immediate < 0) {
+            sub(result, opcode.immediate * -1);
+        }
+    }
+
+    // Equivalent to Engines::Maxwell3D::GetRegisterValue:
+    if (optimizer.enable_asserts) {
+        Xbyak::Label pass_range_check;
+        cmp(RESULT, static_cast<u32>(Engines::Maxwell3D::Regs::NUM_REGS));
+        jb(pass_range_check);
+        int3();
+        L(pass_range_check);
+    }
+    mov(rax, qword[STATE]);
+    mov(RESULT,
+        dword[rax + offsetof(Engines::Maxwell3D, regs) +
+              offsetof(Engines::Maxwell3D::Regs, reg_array) + RESULT.cvt64() * sizeof(u32)]);
+
+    Compile_ProcessResult(opcode.result_operation, opcode.dst);
+}
+
+static void Send(Engines::Maxwell3D* maxwell3d, Macro::MethodAddress method_address, u32 value) {
+    maxwell3d->CallMethodFromMME(method_address.address, value);
+}
+
+void Tegra::MacroJITx64Impl::Compile_Send(Xbyak::Reg32 value) {
+    Common::X64::ABI_PushRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
+    mov(Common::X64::ABI_PARAM1, qword[STATE]);
+    mov(Common::X64::ABI_PARAM2, METHOD_ADDRESS);
+    mov(Common::X64::ABI_PARAM3, value);
+    Common::X64::CallFarFunction(*this, &Send);
+    Common::X64::ABI_PopRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
+
+    Xbyak::Label dont_process{};
+    // Get increment
+    test(METHOD_ADDRESS, 0x3f000);
+    // If zero, method address doesn't update
+    je(dont_process);
+
+    mov(ecx, METHOD_ADDRESS);
+    and_(METHOD_ADDRESS, 0xfff);
+    shr(ecx, 12);
+    and_(ecx, 0x3f);
+    lea(eax, ptr[rcx + METHOD_ADDRESS.cvt64()]);
+    sal(ecx, 12);
+    or_(eax, ecx);
+
+    mov(METHOD_ADDRESS, eax);
+
+    L(dont_process);
+}
+
+void Tegra::MacroJITx64Impl::Compile_Branch(Macro::Opcode opcode) {
+    ASSERT_MSG(!is_delay_slot, "Executing a branch in a delay slot is not valid");
+    const s32 jump_address =
+        static_cast<s32>(pc) + static_cast<s32>(opcode.GetBranchTarget() / sizeof(s32));
+
+    Xbyak::Label end;
+    auto value = Compile_GetRegister(opcode.src_a, eax);
+    test(value, value);
+    if (optimizer.has_delayed_pc) {
+        switch (opcode.branch_condition) {
+        case Macro::BranchCondition::Zero:
+            jne(end, T_NEAR);
+            break;
+        case Macro::BranchCondition::NotZero:
+            je(end, T_NEAR);
+            break;
+        }
+
+        if (opcode.branch_annul) {
+            xor_(BRANCH_HOLDER, BRANCH_HOLDER);
+            jmp(labels[jump_address], T_NEAR);
+        } else {
+            Xbyak::Label handle_post_exit{};
+            Xbyak::Label skip{};
+            jmp(skip, T_NEAR);
+            if (opcode.is_exit) {
+                L(handle_post_exit);
+                // Execute 1 instruction
+                mov(BRANCH_HOLDER, end_of_code);
+                // Jump to next instruction to skip delay slot check
+                jmp(labels[jump_address], T_NEAR);
+            } else {
+                L(handle_post_exit);
+                xor_(BRANCH_HOLDER, BRANCH_HOLDER);
+                jmp(labels[jump_address], T_NEAR);
+            }
+            L(skip);
+            mov(BRANCH_HOLDER, handle_post_exit);
+            jmp(delay_skip[pc], T_NEAR);
+        }
+    } else {
+        switch (opcode.branch_condition) {
+        case Macro::BranchCondition::Zero:
+            je(labels[jump_address], T_NEAR);
+            break;
+        case Macro::BranchCondition::NotZero:
+            jne(labels[jump_address], T_NEAR);
+            break;
+        }
+    }
+
+    L(end);
+}
+
+void Tegra::MacroJITx64Impl::Optimizer_ScanFlags() {
+    optimizer.can_skip_carry = true;
+    optimizer.has_delayed_pc = false;
+    for (auto raw_op : code) {
+        Macro::Opcode op{};
+        op.raw = raw_op;
+
+        if (op.operation == Macro::Operation::ALU) {
+            // Scan for any ALU operations which actually use the carry flag, if they don't exist in
+            // our current code we can skip emitting the carry flag handling operations
+            if (op.alu_operation == Macro::ALUOperation::AddWithCarry ||
+                op.alu_operation == Macro::ALUOperation::SubtractWithBorrow) {
+                optimizer.can_skip_carry = false;
+            }
+        }
+
+        if (op.operation == Macro::Operation::Branch) {
+            if (!op.branch_annul) {
+                optimizer.has_delayed_pc = true;
+            }
+        }
+    }
+}
+
+void MacroJITx64Impl::Compile() {
+    MICROPROFILE_SCOPE(MacroJitCompile);
+    bool keep_executing = true;
+    labels.fill(Xbyak::Label());
+
+    Common::X64::ABI_PushRegistersAndAdjustStack(*this, Common::X64::ABI_ALL_CALLEE_SAVED, 8);
+    // JIT state
+    mov(STATE, Common::X64::ABI_PARAM1);
+    mov(PARAMETERS, Common::X64::ABI_PARAM2);
+    xor_(RESULT, RESULT);
+    xor_(METHOD_ADDRESS, METHOD_ADDRESS);
+    xor_(BRANCH_HOLDER, BRANCH_HOLDER);
+
+    mov(dword[STATE + offsetof(JITState, registers) + 4], Compile_FetchParameter());
+
+    // Track get register for zero registers and mark it as no-op
+    optimizer.zero_reg_skip = true;
+
+    // AddImmediate tends to be used as a NOP instruction, if we detect this we can
+    // completely skip the entire code path and no emit anything
+    optimizer.skip_dummy_addimmediate = true;
+
+    // SMO tends to emit a lot of unnecessary method moves, we can mitigate this by only emitting
+    // one if our register isn't "dirty"
+    optimizer.optimize_for_method_move = true;
+
+    // Enable run-time assertions in JITted code
+    optimizer.enable_asserts = false;
+
+    // Check to see if we can skip emitting certain instructions
+    Optimizer_ScanFlags();
+
+    const u32 op_count = static_cast<u32>(code.size());
+    for (u32 i = 0; i < op_count; i++) {
+        if (i < op_count - 1) {
+            pc = i + 1;
+            next_opcode = GetOpCode();
+        } else {
+            next_opcode = {};
+        }
+        pc = i;
+        Compile_NextInstruction();
+    }
+
+    L(end_of_code);
+
+    Common::X64::ABI_PopRegistersAndAdjustStack(*this, Common::X64::ABI_ALL_CALLEE_SAVED, 8);
+    ret();
+    ready();
+    program = getCode<ProgramType>();
+}
+
+bool MacroJITx64Impl::Compile_NextInstruction() {
+    const auto opcode = GetOpCode();
+    if (labels[pc].getAddress()) {
+        return false;
+    }
+
+    L(labels[pc]);
+
+    switch (opcode.operation) {
+    case Macro::Operation::ALU:
+        Compile_ALU(opcode);
+        break;
+    case Macro::Operation::AddImmediate:
+        Compile_AddImmediate(opcode);
+        break;
+    case Macro::Operation::ExtractInsert:
+        Compile_ExtractInsert(opcode);
+        break;
+    case Macro::Operation::ExtractShiftLeftImmediate:
+        Compile_ExtractShiftLeftImmediate(opcode);
+        break;
+    case Macro::Operation::ExtractShiftLeftRegister:
+        Compile_ExtractShiftLeftRegister(opcode);
+        break;
+    case Macro::Operation::Read:
+        Compile_Read(opcode);
+        break;
+    case Macro::Operation::Branch:
+        Compile_Branch(opcode);
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented opcode {}", opcode.operation.Value());
+        break;
+    }
+
+    if (optimizer.has_delayed_pc) {
+        if (opcode.is_exit) {
+            mov(rax, end_of_code);
+            test(BRANCH_HOLDER, BRANCH_HOLDER);
+            cmove(BRANCH_HOLDER, rax);
+            // Jump to next instruction to skip delay slot check
+            je(labels[pc + 1], T_NEAR);
+        } else {
+            // TODO(ogniK): Optimize delay slot branching
+            Xbyak::Label no_delay_slot{};
+            test(BRANCH_HOLDER, BRANCH_HOLDER);
+            je(no_delay_slot, T_NEAR);
+            mov(rax, BRANCH_HOLDER);
+            xor_(BRANCH_HOLDER, BRANCH_HOLDER);
+            jmp(rax);
+            L(no_delay_slot);
+        }
+        L(delay_skip[pc]);
+        if (opcode.is_exit) {
+            return false;
+        }
+    } else {
+        test(BRANCH_HOLDER, BRANCH_HOLDER);
+        jne(end_of_code, T_NEAR);
+        if (opcode.is_exit) {
+            inc(BRANCH_HOLDER);
+            return false;
+        }
+    }
+    return true;
+}
+
+Xbyak::Reg32 Tegra::MacroJITx64Impl::Compile_FetchParameter() {
+    mov(eax, dword[PARAMETERS]);
+    add(PARAMETERS, sizeof(u32));
+    return eax;
+}
+
+Xbyak::Reg32 MacroJITx64Impl::Compile_GetRegister(u32 index, Xbyak::Reg32 dst) {
+    if (index == 0) {
+        // Register 0 is always zero
+        xor_(dst, dst);
+    } else {
+        mov(dst, dword[STATE + offsetof(JITState, registers) + index * sizeof(u32)]);
+    }
+
+    return dst;
+}
+
+void MacroJITx64Impl::Compile_ProcessResult(Macro::ResultOperation operation, u32 reg) {
+    const auto SetRegister = [this](u32 reg, const Xbyak::Reg32& result) {
+        // Register 0 is supposed to always return 0. NOP is implemented as a store to the zero
+        // register.
+        if (reg == 0) {
+            return;
+        }
+        mov(dword[STATE + offsetof(JITState, registers) + reg * sizeof(u32)], result);
+    };
+    const auto SetMethodAddress = [this](const Xbyak::Reg32& reg) { mov(METHOD_ADDRESS, reg); };
+
+    switch (operation) {
+    case Macro::ResultOperation::IgnoreAndFetch:
+        SetRegister(reg, Compile_FetchParameter());
+        break;
+    case Macro::ResultOperation::Move:
+        SetRegister(reg, RESULT);
+        break;
+    case Macro::ResultOperation::MoveAndSetMethod:
+        SetRegister(reg, RESULT);
+        SetMethodAddress(RESULT);
+        break;
+    case Macro::ResultOperation::FetchAndSend:
+        // Fetch parameter and send result.
+        SetRegister(reg, Compile_FetchParameter());
+        Compile_Send(RESULT);
+        break;
+    case Macro::ResultOperation::MoveAndSend:
+        // Move and send result.
+        SetRegister(reg, RESULT);
+        Compile_Send(RESULT);
+        break;
+    case Macro::ResultOperation::FetchAndSetMethod:
+        // Fetch parameter and use result as Method Address.
+        SetRegister(reg, Compile_FetchParameter());
+        SetMethodAddress(RESULT);
+        break;
+    case Macro::ResultOperation::MoveAndSetMethodFetchAndSend:
+        // Move result and use as Method Address, then fetch and send parameter.
+        SetRegister(reg, RESULT);
+        SetMethodAddress(RESULT);
+        Compile_Send(Compile_FetchParameter());
+        break;
+    case Macro::ResultOperation::MoveAndSetMethodSend:
+        // Move result and use as Method Address, then send bits 12:17 of result.
+        SetRegister(reg, RESULT);
+        SetMethodAddress(RESULT);
+        shr(RESULT, 12);
+        and_(RESULT, 0b111111);
+        Compile_Send(RESULT);
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented macro operation {}", static_cast<std::size_t>(operation));
+    }
+}
+
+Macro::Opcode MacroJITx64Impl::GetOpCode() const {
+    ASSERT(pc < code.size());
+    return {code[pc]};
+}
+
+std::bitset<32> MacroJITx64Impl::PersistentCallerSavedRegs() const {
+    return PERSISTENT_REGISTERS & Common::X64::ABI_ALL_CALLER_SAVED;
+}
+
+} // namespace Tegra

src/video_core/macro/macro_jit_x64.h

@@ -0,0 +1,98 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <bitset>
+#include <xbyak.h>
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "common/x64/xbyak_abi.h"
+#include "video_core/macro/macro.h"
+
+namespace Tegra {
+
+namespace Engines {
+class Maxwell3D;
+}
+
+/// MAX_CODE_SIZE is arbitrarily chosen based on current booting games
+constexpr size_t MAX_CODE_SIZE = 0x10000;
+
+class MacroJITx64 final : public MacroEngine {
+public:
+    explicit MacroJITx64(Engines::Maxwell3D& maxwell3d);
+
+protected:
+    std::unique_ptr<CachedMacro> Compile(const std::vector<u32>& code) override;
+
+private:
+    Engines::Maxwell3D& maxwell3d;
+};
+
+class MacroJITx64Impl : public Xbyak::CodeGenerator, public CachedMacro {
+public:
+    MacroJITx64Impl(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& code);
+    ~MacroJITx64Impl();
+
+    void Execute(const std::vector<u32>& parameters, u32 method) override;
+
+    void Compile_ALU(Macro::Opcode opcode);
+    void Compile_AddImmediate(Macro::Opcode opcode);
+    void Compile_ExtractInsert(Macro::Opcode opcode);
+    void Compile_ExtractShiftLeftImmediate(Macro::Opcode opcode);
+    void Compile_ExtractShiftLeftRegister(Macro::Opcode opcode);
+    void Compile_Read(Macro::Opcode opcode);
+    void Compile_Branch(Macro::Opcode opcode);
+
+private:
+    void Optimizer_ScanFlags();
+
+    void Compile();
+    bool Compile_NextInstruction();
+
+    Xbyak::Reg32 Compile_FetchParameter();
+    Xbyak::Reg32 Compile_GetRegister(u32 index, Xbyak::Reg32 dst);
+
+    void Compile_ProcessResult(Macro::ResultOperation operation, u32 reg);
+    void Compile_Send(Xbyak::Reg32 value);
+
+    Macro::Opcode GetOpCode() const;
+    std::bitset<32> PersistentCallerSavedRegs() const;
+
+    struct JITState {
+        Engines::Maxwell3D* maxwell3d{};
+        std::array<u32, Macro::NUM_MACRO_REGISTERS> registers{};
+        u32 carry_flag{};
+    };
+    static_assert(offsetof(JITState, maxwell3d) == 0, "Maxwell3D is not at 0x0");
+    using ProgramType = void (*)(JITState*, const u32*);
+
+    struct OptimizerState {
+        bool can_skip_carry{};
+        bool has_delayed_pc{};
+        bool zero_reg_skip{};
+        bool skip_dummy_addimmediate{};
+        bool optimize_for_method_move{};
+        bool enable_asserts{};
+    };
+    OptimizerState optimizer{};
+
+    std::optional<Macro::Opcode> next_opcode{};
+    ProgramType program{nullptr};
+
+    std::array<Xbyak::Label, MAX_CODE_SIZE> labels;
+    std::array<Xbyak::Label, MAX_CODE_SIZE> delay_skip;
+    Xbyak::Label end_of_code{};
+
+    bool is_delay_slot{};
+    u32 pc{};
+    std::optional<u32> delayed_pc;
+
+    const std::vector<u32>& code;
+    Engines::Maxwell3D& maxwell3d;
+};
+
+} // namespace Tegra

src/video_core/memory_manager.cpp

@@ -210,10 +210,11 @@ bool MemoryManager::IsBlockContinuous(const GPUVAddr start, const std::size_t si
     return range == inner_size;
 }
 
-void MemoryManager::ReadBlock(GPUVAddr src_addr, void* dest_buffer, const std::size_t size) const {
+void MemoryManager::ReadBlock(GPUVAddr gpu_src_addr, void* dest_buffer,
+                              const std::size_t size) const {
     std::size_t remaining_size{size};
-    std::size_t page_index{src_addr >> page_bits};
-    std::size_t page_offset{src_addr & page_mask};
+    std::size_t page_index{gpu_src_addr >> page_bits};
+    std::size_t page_offset{gpu_src_addr & page_mask};
 
     auto& memory = system.Memory();
 
@@ -234,11 +235,11 @@ void MemoryManager::ReadBlock(GPUVAddr src_addr, void* dest_buffer, const std::s
     }
 }
 
-void MemoryManager::ReadBlockUnsafe(GPUVAddr src_addr, void* dest_buffer,
+void MemoryManager::ReadBlockUnsafe(GPUVAddr gpu_src_addr, void* dest_buffer,
                                     const std::size_t size) const {
     std::size_t remaining_size{size};
-    std::size_t page_index{src_addr >> page_bits};
-    std::size_t page_offset{src_addr & page_mask};
+    std::size_t page_index{gpu_src_addr >> page_bits};
+    std::size_t page_offset{gpu_src_addr & page_mask};
 
     auto& memory = system.Memory();
 
@@ -259,10 +260,11 @@ void MemoryManager::ReadBlockUnsafe(GPUVAddr src_addr, void* dest_buffer,
     }
 }
 
-void MemoryManager::WriteBlock(GPUVAddr dest_addr, const void* src_buffer, const std::size_t size) {
+void MemoryManager::WriteBlock(GPUVAddr gpu_dest_addr, const void* src_buffer,
+                               const std::size_t size) {
     std::size_t remaining_size{size};
-    std::size_t page_index{dest_addr >> page_bits};
-    std::size_t page_offset{dest_addr & page_mask};
+    std::size_t page_index{gpu_dest_addr >> page_bits};
+    std::size_t page_offset{gpu_dest_addr & page_mask};
 
     auto& memory = system.Memory();
 
@@ -283,11 +285,11 @@ void MemoryManager::WriteBlock(GPUVAddr dest_addr, const void* src_buffer, const
     }
 }
 
-void MemoryManager::WriteBlockUnsafe(GPUVAddr dest_addr, const void* src_buffer,
+void MemoryManager::WriteBlockUnsafe(GPUVAddr gpu_dest_addr, const void* src_buffer,
                                      const std::size_t size) {
     std::size_t remaining_size{size};
-    std::size_t page_index{dest_addr >> page_bits};
-    std::size_t page_offset{dest_addr & page_mask};
+    std::size_t page_index{gpu_dest_addr >> page_bits};
+    std::size_t page_offset{gpu_dest_addr & page_mask};
 
     auto& memory = system.Memory();
 
@@ -306,16 +308,18 @@ void MemoryManager::WriteBlockUnsafe(GPUVAddr dest_addr, const void* src_buffer,
     }
 }
 
-void MemoryManager::CopyBlock(GPUVAddr dest_addr, GPUVAddr src_addr, const std::size_t size) {
+void MemoryManager::CopyBlock(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr,
+                              const std::size_t size) {
     std::vector<u8> tmp_buffer(size);
-    ReadBlock(src_addr, tmp_buffer.data(), size);
-    WriteBlock(dest_addr, tmp_buffer.data(), size);
+    ReadBlock(gpu_src_addr, tmp_buffer.data(), size);
+    WriteBlock(gpu_dest_addr, tmp_buffer.data(), size);
 }
 
-void MemoryManager::CopyBlockUnsafe(GPUVAddr dest_addr, GPUVAddr src_addr, const std::size_t size) {
+void MemoryManager::CopyBlockUnsafe(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr,
+                                    const std::size_t size) {
     std::vector<u8> tmp_buffer(size);
-    ReadBlockUnsafe(src_addr, tmp_buffer.data(), size);
-    WriteBlockUnsafe(dest_addr, tmp_buffer.data(), size);
+    ReadBlockUnsafe(gpu_src_addr, tmp_buffer.data(), size);
+    WriteBlockUnsafe(gpu_dest_addr, tmp_buffer.data(), size);
 }
 
 bool MemoryManager::IsGranularRange(GPUVAddr gpu_addr, std::size_t size) {

src/video_core/memory_manager.h

@@ -79,9 +79,9 @@ public:
      * in the Host Memory counterpart. Note: This functions cause Host GPU Memory
      * Flushes and Invalidations, respectively to each operation.
      */
-    void ReadBlock(GPUVAddr src_addr, void* dest_buffer, std::size_t size) const;
-    void WriteBlock(GPUVAddr dest_addr, const void* src_buffer, std::size_t size);
-    void CopyBlock(GPUVAddr dest_addr, GPUVAddr src_addr, std::size_t size);
+    void ReadBlock(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size) const;
+    void WriteBlock(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size);
+    void CopyBlock(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr, std::size_t size);
 
     /**
      * ReadBlockUnsafe and WriteBlockUnsafe are special versions of ReadBlock and
@@ -93,9 +93,9 @@ public:
      * WriteBlockUnsafe instead of WriteBlock since it shouldn't invalidate the texture
      * being flushed.
      */
-    void ReadBlockUnsafe(GPUVAddr src_addr, void* dest_buffer, std::size_t size) const;
-    void WriteBlockUnsafe(GPUVAddr dest_addr, const void* src_buffer, std::size_t size);
-    void CopyBlockUnsafe(GPUVAddr dest_addr, GPUVAddr src_addr, std::size_t size);
+    void ReadBlockUnsafe(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size) const;
+    void WriteBlockUnsafe(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size);
+    void CopyBlockUnsafe(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr, std::size_t size);
 
     /**
      * IsGranularRange checks if a gpu region can be simply read with a pointer

src/video_core/query_cache.h

@@ -220,8 +220,8 @@ private:
             return cache_begin < addr_end && addr_begin < cache_end;
         };
 
-        const u64 page_end = addr_end >> PAGE_SHIFT;
-        for (u64 page = addr_begin >> PAGE_SHIFT; page <= page_end; ++page) {
+        const u64 page_end = addr_end >> PAGE_BITS;
+        for (u64 page = addr_begin >> PAGE_BITS; page <= page_end; ++page) {
             const auto& it = cached_queries.find(page);
             if (it == std::end(cached_queries)) {
                 continue;
@@ -242,14 +242,14 @@ private:
     /// Registers the passed parameters as cached and returns a pointer to the stored cached query.
     CachedQuery* Register(VideoCore::QueryType type, VAddr cpu_addr, u8* host_ptr, bool timestamp) {
         rasterizer.UpdatePagesCachedCount(cpu_addr, CachedQuery::SizeInBytes(timestamp), 1);
-        const u64 page = static_cast<u64>(cpu_addr) >> PAGE_SHIFT;
+        const u64 page = static_cast<u64>(cpu_addr) >> PAGE_BITS;
         return &cached_queries[page].emplace_back(static_cast<QueryCache&>(*this), type, cpu_addr,
                                                   host_ptr);
     }
 
     /// Tries to a get a cached query. Returns nullptr on failure.
     CachedQuery* TryGet(VAddr addr) {
-        const u64 page = static_cast<u64>(addr) >> PAGE_SHIFT;
+        const u64 page = static_cast<u64>(addr) >> PAGE_BITS;
         const auto it = cached_queries.find(page);
         if (it == std::end(cached_queries)) {
             return nullptr;
@@ -268,7 +268,7 @@ private:
     }
 
     static constexpr std::uintptr_t PAGE_SIZE = 4096;
-    static constexpr unsigned PAGE_SHIFT = 12;
+    static constexpr unsigned PAGE_BITS = 12;
 
     Core::System& system;
     VideoCore::RasterizerInterface& rasterizer;

src/video_core/rasterizer_cache.cpp

@@ -1,7 +0,0 @@
-// Copyright 2018 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include "video_core/rasterizer_cache.h"
-
-RasterizerCacheObject::~RasterizerCacheObject() = default;

src/video_core/rasterizer_cache.h

@@ -1,253 +0,0 @@
-// Copyright 2018 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <mutex>
-#include <set>
-#include <unordered_map>
-
-#include <boost/icl/interval_map.hpp>
-#include <boost/range/iterator_range_core.hpp>
-
-#include "common/common_types.h"
-#include "core/settings.h"
-#include "video_core/gpu.h"
-#include "video_core/rasterizer_interface.h"
-
-class RasterizerCacheObject {
-public:
-    explicit RasterizerCacheObject(const VAddr cpu_addr) : cpu_addr{cpu_addr} {}
-
-    virtual ~RasterizerCacheObject();
-
-    VAddr GetCpuAddr() const {
-        return cpu_addr;
-    }
-
-    /// Gets the size of the shader in guest memory, required for cache management
-    virtual std::size_t GetSizeInBytes() const = 0;
-
-    /// Sets whether the cached object should be considered registered
-    void SetIsRegistered(bool registered) {
-        is_registered = registered;
-    }
-
-    /// Returns true if the cached object is registered
-    bool IsRegistered() const {
-        return is_registered;
-    }
-
-    /// Returns true if the cached object is dirty
-    bool IsDirty() const {
-        return is_dirty;
-    }
-
-    /// Returns ticks from when this cached object was last modified
-    u64 GetLastModifiedTicks() const {
-        return last_modified_ticks;
-    }
-
-    /// Marks an object as recently modified, used to specify whether it is clean or dirty
-    template <class T>
-    void MarkAsModified(bool dirty, T& cache) {
-        is_dirty = dirty;
-        last_modified_ticks = cache.GetModifiedTicks();
-    }
-
-    void SetMemoryMarked(bool is_memory_marked_) {
-        is_memory_marked = is_memory_marked_;
-    }
-
-    bool IsMemoryMarked() const {
-        return is_memory_marked;
-    }
-
-    void SetSyncPending(bool is_sync_pending_) {
-        is_sync_pending = is_sync_pending_;
-    }
-
-    bool IsSyncPending() const {
-        return is_sync_pending;
-    }
-
-private:
-    bool is_registered{};      ///< Whether the object is currently registered with the cache
-    bool is_dirty{};           ///< Whether the object is dirty (out of sync with guest memory)
-    bool is_memory_marked{};   ///< Whether the object is marking rasterizer memory.
-    bool is_sync_pending{};    ///< Whether the object is pending deletion.
-    u64 last_modified_ticks{}; ///< When the object was last modified, used for in-order flushing
-    VAddr cpu_addr{};          ///< Cpu address memory, unique from emulated virtual address space
-};
-
-template <class T>
-class RasterizerCache : NonCopyable {
-    friend class RasterizerCacheObject;
-
-public:
-    explicit RasterizerCache(VideoCore::RasterizerInterface& rasterizer) : rasterizer{rasterizer} {}
-
-    /// Write any cached resources overlapping the specified region back to memory
-    void FlushRegion(VAddr addr, std::size_t size) {
-        std::lock_guard lock{mutex};
-
-        const auto& objects{GetSortedObjectsFromRegion(addr, size)};
-        for (auto& object : objects) {
-            FlushObject(object);
-        }
-    }
-
-    /// Mark the specified region as being invalidated
-    void InvalidateRegion(VAddr addr, u64 size) {
-        std::lock_guard lock{mutex};
-
-        const auto& objects{GetSortedObjectsFromRegion(addr, size)};
-        for (auto& object : objects) {
-            if (!object->IsRegistered()) {
-                // Skip duplicates
-                continue;
-            }
-            Unregister(object);
-        }
-    }
-
-    void OnCPUWrite(VAddr addr, std::size_t size) {
-        std::lock_guard lock{mutex};
-
-        for (const auto& object : GetSortedObjectsFromRegion(addr, size)) {
-            if (object->IsRegistered()) {
-                UnmarkMemory(object);
-                object->SetSyncPending(true);
-                marked_for_unregister.emplace_back(object);
-            }
-        }
-    }
-
-    void SyncGuestHost() {
-        std::lock_guard lock{mutex};
-
-        for (const auto& object : marked_for_unregister) {
-            if (object->IsRegistered()) {
-                object->SetSyncPending(false);
-                Unregister(object);
-            }
-        }
-        marked_for_unregister.clear();
-    }
-
-    /// Invalidates everything in the cache
-    void InvalidateAll() {
-        std::lock_guard lock{mutex};
-
-        while (interval_cache.begin() != interval_cache.end()) {
-            Unregister(*interval_cache.begin()->second.begin());
-        }
-    }
-
-protected:
-    /// Tries to get an object from the cache with the specified cache address
-    T TryGet(VAddr addr) const {
-        const auto iter = map_cache.find(addr);
-        if (iter != map_cache.end())
-            return iter->second;
-        return nullptr;
-    }
-
-    /// Register an object into the cache
-    virtual void Register(const T& object) {
-        std::lock_guard lock{mutex};
-
-        object->SetIsRegistered(true);
-        interval_cache.add({GetInterval(object), ObjectSet{object}});
-        map_cache.insert({object->GetCpuAddr(), object});
-        rasterizer.UpdatePagesCachedCount(object->GetCpuAddr(), object->GetSizeInBytes(), 1);
-        object->SetMemoryMarked(true);
-    }
-
-    /// Unregisters an object from the cache
-    virtual void Unregister(const T& object) {
-        std::lock_guard lock{mutex};
-
-        UnmarkMemory(object);
-        object->SetIsRegistered(false);
-        if (object->IsSyncPending()) {
-            marked_for_unregister.remove(object);
-            object->SetSyncPending(false);
-        }
-        const VAddr addr = object->GetCpuAddr();
-        interval_cache.subtract({GetInterval(object), ObjectSet{object}});
-        map_cache.erase(addr);
-    }
-
-    void UnmarkMemory(const T& object) {
-        if (!object->IsMemoryMarked()) {
-            return;
-        }
-        rasterizer.UpdatePagesCachedCount(object->GetCpuAddr(), object->GetSizeInBytes(), -1);
-        object->SetMemoryMarked(false);
-    }
-
-    /// Returns a ticks counter used for tracking when cached objects were last modified
-    u64 GetModifiedTicks() {
-        std::lock_guard lock{mutex};
-
-        return ++modified_ticks;
-    }
-
-    virtual void FlushObjectInner(const T& object) = 0;
-
-    /// Flushes the specified object, updating appropriate cache state as needed
-    void FlushObject(const T& object) {
-        std::lock_guard lock{mutex};
-
-        if (!object->IsDirty()) {
-            return;
-        }
-        FlushObjectInner(object);
-        object->MarkAsModified(false, *this);
-    }
-
-    std::recursive_mutex mutex;
-
-private:
-    /// Returns a list of cached objects from the specified memory region, ordered by access time
-    std::vector<T> GetSortedObjectsFromRegion(VAddr addr, u64 size) {
-        if (size == 0) {
-            return {};
-        }
-
-        std::vector<T> objects;
-        const ObjectInterval interval{addr, addr + size};
-        for (auto& pair : boost::make_iterator_range(interval_cache.equal_range(interval))) {
-            for (auto& cached_object : pair.second) {
-                if (!cached_object) {
-                    continue;
-                }
-                objects.push_back(cached_object);
-            }
-        }
-
-        std::sort(objects.begin(), objects.end(), [](const T& a, const T& b) -> bool {
-            return a->GetLastModifiedTicks() < b->GetLastModifiedTicks();
-        });
-
-        return objects;
-    }
-
-    using ObjectSet = std::set<T>;
-    using ObjectCache = std::unordered_map<VAddr, T>;
-    using IntervalCache = boost::icl::interval_map<VAddr, ObjectSet>;
-    using ObjectInterval = typename IntervalCache::interval_type;
-
-    static auto GetInterval(const T& object) {
-        return ObjectInterval::right_open(object->GetCpuAddr(),
-                                          object->GetCpuAddr() + object->GetSizeInBytes());
-    }
-
-    ObjectCache map_cache;
-    IntervalCache interval_cache; ///< Cache of objects
-    u64 modified_ticks{};         ///< Counter of cache state ticks, used for in-order flushing
-    VideoCore::RasterizerInterface& rasterizer;
-    std::list<T> marked_for_unregister;
-};

src/video_core/renderer_opengl/gl_arb_decompiler.cpp

@@ -0,0 +1,2073 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <array>
+#include <cstddef>
+#include <string>
+#include <string_view>
+#include <utility>
+#include <variant>
+
+#include <fmt/format.h>
+
+#include "common/alignment.h"
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "video_core/renderer_opengl/gl_arb_decompiler.h"
+#include "video_core/renderer_opengl/gl_device.h"
+#include "video_core/shader/registry.h"
+#include "video_core/shader/shader_ir.h"
+
+// Predicates in the decompiled code follow the convention that -1 means true and 0 means false.
+// GLASM lacks booleans, so they have to be implemented as integers.
+// Using -1 for true is useful because both CMP.S and NOT.U can negate it, and CMP.S can be used to
+// select between two values, because -1 will be evaluated as true and 0 as false.
+
+namespace OpenGL {
+
+namespace {
+
+using Tegra::Engines::ShaderType;
+using Tegra::Shader::Attribute;
+using Tegra::Shader::PixelImap;
+using Tegra::Shader::Register;
+using namespace VideoCommon::Shader;
+using Operation = const OperationNode&;
+
+constexpr std::array INTERNAL_FLAG_NAMES = {"ZERO", "SIGN", "CARRY", "OVERFLOW"};
+
+char Swizzle(std::size_t component) {
+    ASSERT(component < 4);
+    return component["xyzw"];
+}
+
+constexpr bool IsGenericAttribute(Attribute::Index index) {
+    return index >= Attribute::Index::Attribute_0 && index <= Attribute::Index::Attribute_31;
+}
+
+u32 GetGenericAttributeIndex(Attribute::Index index) {
+    ASSERT(IsGenericAttribute(index));
+    return static_cast<u32>(index) - static_cast<u32>(Attribute::Index::Attribute_0);
+}
+
+std::string_view Modifiers(Operation operation) {
+    const auto meta = std::get_if<MetaArithmetic>(&operation.GetMeta());
+    if (meta && meta->precise) {
+        return ".PREC";
+    }
+    return "";
+}
+
+std::string_view GetInputFlags(PixelImap attribute) {
+    switch (attribute) {
+    case PixelImap::Perspective:
+        return "";
+    case PixelImap::Constant:
+        return "FLAT ";
+    case PixelImap::ScreenLinear:
+        return "NOPERSPECTIVE ";
+    case PixelImap::Unused:
+        break;
+    }
+    UNIMPLEMENTED_MSG("Unknown attribute usage index={}", static_cast<int>(attribute));
+    return {};
+}
+
+std::string_view ImageType(Tegra::Shader::ImageType image_type) {
+    switch (image_type) {
+    case Tegra::Shader::ImageType::Texture1D:
+        return "1D";
+    case Tegra::Shader::ImageType::TextureBuffer:
+        return "BUFFER";
+    case Tegra::Shader::ImageType::Texture1DArray:
+        return "ARRAY1D";
+    case Tegra::Shader::ImageType::Texture2D:
+        return "2D";
+    case Tegra::Shader::ImageType::Texture2DArray:
+        return "ARRAY2D";
+    case Tegra::Shader::ImageType::Texture3D:
+        return "3D";
+    }
+    UNREACHABLE();
+    return {};
+}
+
+std::string_view StackName(MetaStackClass stack) {
+    switch (stack) {
+    case MetaStackClass::Ssy:
+        return "SSY";
+    case MetaStackClass::Pbk:
+        return "PBK";
+    }
+    UNREACHABLE();
+    return "";
+};
+
+std::string_view PrimitiveDescription(Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology topology) {
+    switch (topology) {
+    case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::Points:
+        return "POINTS";
+    case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::Lines:
+    case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::LineStrip:
+        return "LINES";
+    case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::LinesAdjacency:
+    case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::LineStripAdjacency:
+        return "LINES_ADJACENCY";
+    case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::Triangles:
+    case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::TriangleStrip:
+    case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::TriangleFan:
+        return "TRIANGLES";
+    case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::TrianglesAdjacency:
+    case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::TriangleStripAdjacency:
+        return "TRIANGLES_ADJACENCY";
+    default:
+        UNIMPLEMENTED_MSG("topology={}", static_cast<int>(topology));
+        return "POINTS";
+    }
+}
+
+std::string_view TopologyName(Tegra::Shader::OutputTopology topology) {
+    switch (topology) {
+    case Tegra::Shader::OutputTopology::PointList:
+        return "POINTS";
+    case Tegra::Shader::OutputTopology::LineStrip:
+        return "LINE_STRIP";
+    case Tegra::Shader::OutputTopology::TriangleStrip:
+        return "TRIANGLE_STRIP";
+    default:
+        UNIMPLEMENTED_MSG("Unknown output topology: {}", static_cast<u32>(topology));
+        return "points";
+    }
+}
+
+std::string_view StageInputName(ShaderType stage) {
+    switch (stage) {
+    case ShaderType::Vertex:
+    case ShaderType::Geometry:
+        return "vertex";
+    case ShaderType::Fragment:
+        return "fragment";
+    case ShaderType::Compute:
+        return "invocation";
+    default:
+        UNREACHABLE();
+        return "";
+    }
+}
+
+std::string TextureType(const MetaTexture& meta) {
+    if (meta.sampler.is_buffer) {
+        return "BUFFER";
+    }
+    std::string type;
+    if (meta.sampler.is_shadow) {
+        type += "SHADOW";
+    }
+    if (meta.sampler.is_array) {
+        type += "ARRAY";
+    }
+    type += [&meta] {
+        switch (meta.sampler.type) {
+        case Tegra::Shader::TextureType::Texture1D:
+            return "1D";
+        case Tegra::Shader::TextureType::Texture2D:
+            return "2D";
+        case Tegra::Shader::TextureType::Texture3D:
+            return "3D";
+        case Tegra::Shader::TextureType::TextureCube:
+            return "CUBE";
+        }
+        UNREACHABLE();
+        return "2D";
+    }();
+    return type;
+}
+
+std::string GlobalMemoryName(const GlobalMemoryBase& base) {
+    return fmt::format("gmem{}_{}", base.cbuf_index, base.cbuf_offset);
+}
+
+class ARBDecompiler final {
+public:
+    explicit ARBDecompiler(const Device& device, const ShaderIR& ir, const Registry& registry,
+                           ShaderType stage, std::string_view identifier);
+
+    std::string Code() const {
+        return shader_source;
+    }
+
+private:
+    void DeclareHeader();
+    void DeclareVertex();
+    void DeclareGeometry();
+    void DeclareFragment();
+    void DeclareCompute();
+    void DeclareInputAttributes();
+    void DeclareOutputAttributes();
+    void DeclareLocalMemory();
+    void DeclareGlobalMemory();
+    void DeclareConstantBuffers();
+    void DeclareRegisters();
+    void DeclareTemporaries();
+    void DeclarePredicates();
+    void DeclareInternalFlags();
+
+    void InitializeVariables();
+
+    void DecompileAST();
+    void DecompileBranchMode();
+
+    void VisitAST(const ASTNode& node);
+    std::string VisitExpression(const Expr& node);
+
+    void VisitBlock(const NodeBlock& bb);
+
+    std::string Visit(const Node& node);
+
+    std::pair<std::string, std::size_t> BuildCoords(Operation);
+    std::string BuildAoffi(Operation);
+    void Exit();
+
+    std::string Assign(Operation);
+    std::string Select(Operation);
+    std::string FClamp(Operation);
+    std::string FCastHalf0(Operation);
+    std::string FCastHalf1(Operation);
+    std::string FSqrt(Operation);
+    std::string FSwizzleAdd(Operation);
+    std::string HAdd2(Operation);
+    std::string HMul2(Operation);
+    std::string HFma2(Operation);
+    std::string HAbsolute(Operation);
+    std::string HNegate(Operation);
+    std::string HClamp(Operation);
+    std::string HCastFloat(Operation);
+    std::string HUnpack(Operation);
+    std::string HMergeF32(Operation);
+    std::string HMergeH0(Operation);
+    std::string HMergeH1(Operation);
+    std::string HPack2(Operation);
+    std::string LogicalAssign(Operation);
+    std::string LogicalPick2(Operation);
+    std::string LogicalAnd2(Operation);
+    std::string FloatOrdered(Operation);
+    std::string FloatUnordered(Operation);
+    std::string LogicalAddCarry(Operation);
+    std::string Texture(Operation);
+    std::string TextureGather(Operation);
+    std::string TextureQueryDimensions(Operation);
+    std::string TextureQueryLod(Operation);
+    std::string TexelFetch(Operation);
+    std::string TextureGradient(Operation);
+    std::string ImageLoad(Operation);
+    std::string ImageStore(Operation);
+    std::string Branch(Operation);
+    std::string BranchIndirect(Operation);
+    std::string PushFlowStack(Operation);
+    std::string PopFlowStack(Operation);
+    std::string Exit(Operation);
+    std::string Discard(Operation);
+    std::string EmitVertex(Operation);
+    std::string EndPrimitive(Operation);
+    std::string InvocationId(Operation);
+    std::string YNegate(Operation);
+    std::string ThreadId(Operation);
+    std::string ShuffleIndexed(Operation);
+    std::string Barrier(Operation);
+    std::string MemoryBarrierGroup(Operation);
+    std::string MemoryBarrierGlobal(Operation);
+
+    template <const std::string_view& op>
+    std::string Unary(Operation operation) {
+        std::string temporary = AllocTemporary();
+        AddLine("{}{} {}, {};", op, Modifiers(operation), temporary, Visit(operation[0]));
+        return temporary;
+    }
+
+    template <const std::string_view& op>
+    std::string Binary(Operation operation) {
+        std::string temporary = AllocTemporary();
+        AddLine("{}{} {}, {}, {};", op, Modifiers(operation), temporary, Visit(operation[0]),
+                Visit(operation[1]));
+        return temporary;
+    }
+
+    template <const std::string_view& op>
+    std::string Trinary(Operation operation) {
+        std::string temporary = AllocTemporary();
+        AddLine("{}{} {}, {}, {}, {};", op, Modifiers(operation), temporary, Visit(operation[0]),
+                Visit(operation[1]), Visit(operation[2]));
+        return temporary;
+    }
+
+    template <const std::string_view& op, bool unordered>
+    std::string FloatComparison(Operation operation) {
+        std::string temporary = AllocTemporary();
+        AddLine("TRUNC.U.CC RC.x, {};", Binary<op>(operation));
+        AddLine("MOV.S {}, 0;", temporary);
+        AddLine("MOV.S {} (NE.x), -1;", temporary);
+
+        const std::string op_a = Visit(operation[0]);
+        const std::string op_b = Visit(operation[1]);
+        if constexpr (unordered) {
+            AddLine("SNE.F RC.x, {}, {};", op_a, op_a);
+            AddLine("TRUNC.U.CC RC.x, RC.x;");
+            AddLine("MOV.S {} (NE.x), -1;", temporary);
+            AddLine("SNE.F RC.x, {}, {};", op_b, op_b);
+            AddLine("TRUNC.U.CC RC.x, RC.x;");
+            AddLine("MOV.S {} (NE.x), -1;", temporary);
+        } else if (op == SNE_F) {
+            AddLine("SNE.F RC.x, {}, {};", op_a, op_a);
+            AddLine("TRUNC.U.CC RC.x, RC.x;");
+            AddLine("MOV.S {} (NE.x), 0;", temporary);
+            AddLine("SNE.F RC.x, {}, {};", op_b, op_b);
+            AddLine("TRUNC.U.CC RC.x, RC.x;");
+            AddLine("MOV.S {} (NE.x), 0;", temporary);
+        }
+        return temporary;
+    }
+
+    template <const std::string_view& op, bool is_nan>
+    std::string HalfComparison(Operation operation) {
+        std::string tmp1 = AllocVectorTemporary();
+        const std::string tmp2 = AllocVectorTemporary();
+        const std::string op_a = Visit(operation[0]);
+        const std::string op_b = Visit(operation[1]);
+        AddLine("UP2H.F {}, {};", tmp1, op_a);
+        AddLine("UP2H.F {}, {};", tmp2, op_b);
+        AddLine("{} {}, {}, {};", op, tmp1, tmp1, tmp2);
+        AddLine("TRUNC.U.CC RC.xy, {};", tmp1);
+        AddLine("MOV.S {}.xy, {{0, 0, 0, 0}};", tmp1);
+        AddLine("MOV.S {}.x (NE.x), -1;", tmp1);
+        AddLine("MOV.S {}.y (NE.y), -1;", tmp1);
+        if constexpr (is_nan) {
+            AddLine("MOVC.F RC.x, {};", op_a);
+            AddLine("MOV.S {}.x (NAN.x), -1;", tmp1);
+            AddLine("MOVC.F RC.x, {};", op_b);
+            AddLine("MOV.S {}.y (NAN.x), -1;", tmp1);
+        }
+        return tmp1;
+    }
+
+    template <const std::string_view& op, const std::string_view& type>
+    std::string AtomicImage(Operation operation) {
+        const auto& meta = std::get<MetaImage>(operation.GetMeta());
+        const u32 image_id = device.GetBaseBindings(stage).image + meta.image.index;
+        const std::size_t num_coords = operation.GetOperandsCount();
+        const std::size_t num_values = meta.values.size();
+
+        const std::string coord = AllocVectorTemporary();
+        const std::string value = AllocVectorTemporary();
+        for (std::size_t i = 0; i < num_coords; ++i) {
+            AddLine("MOV.S {}.{}, {};", coord, Swizzle(i), Visit(operation[i]));
+        }
+        for (std::size_t i = 0; i < num_values; ++i) {
+            AddLine("MOV.F {}.{}, {};", value, Swizzle(i), Visit(meta.values[i]));
+        }
+
+        AddLine("ATOMIM.{}.{} {}.x, {}, {}, image[{}], {};", op, type, coord, value, coord,
+                image_id, ImageType(meta.image.type));
+        return fmt::format("{}.x", coord);
+    }
+
+    template <const std::string_view& op, const std::string_view& type>
+    std::string Atomic(Operation operation) {
+        std::string temporary = AllocTemporary();
+        std::string address;
+        std::string_view opname;
+        if (const auto gmem = std::get_if<GmemNode>(&*operation[0])) {
+            AddLine("SUB.U {}, {}, {};", temporary, Visit(gmem->GetRealAddress()),
+                    Visit(gmem->GetBaseAddress()));
+            address = fmt::format("{}[{}]", GlobalMemoryName(gmem->GetDescriptor()), temporary);
+            opname = "ATOMB";
+        } else if (const auto smem = std::get_if<SmemNode>(&*operation[0])) {
+            address = fmt::format("shared_mem[{}]", Visit(smem->GetAddress()));
+            opname = "ATOMS";
+        } else {
+            UNREACHABLE();
+            return "{0, 0, 0, 0}";
+        }
+        AddLine("{}.{}.{} {}, {}, {};", opname, op, type, temporary, Visit(operation[1]), address);
+        return temporary;
+    }
+
+    template <char type>
+    std::string Negate(Operation operation) {
+        std::string temporary = AllocTemporary();
+        if constexpr (type == 'F') {
+            AddLine("MOV.F32 {}, -{};", temporary, Visit(operation[0]));
+        } else {
+            AddLine("MOV.{} {}, -{};", type, temporary, Visit(operation[0]));
+        }
+        return temporary;
+    }
+
+    template <char type>
+    std::string Absolute(Operation operation) {
+        std::string temporary = AllocTemporary();
+        AddLine("MOV.{} {}, |{}|;", type, temporary, Visit(operation[0]));
+        return temporary;
+    }
+
+    template <char type>
+    std::string BitfieldInsert(Operation operation) {
+        const std::string temporary = AllocVectorTemporary();
+        AddLine("MOV.{} {}.x, {};", type, temporary, Visit(operation[3]));
+        AddLine("MOV.{} {}.y, {};", type, temporary, Visit(operation[2]));
+        AddLine("BFI.{} {}.x, {}, {}, {};", type, temporary, temporary, Visit(operation[1]),
+                Visit(operation[0]));
+        return fmt::format("{}.x", temporary);
+    }
+
+    template <char type>
+    std::string BitfieldExtract(Operation operation) {
+        const std::string temporary = AllocVectorTemporary();
+        AddLine("MOV.{} {}.x, {};", type, temporary, Visit(operation[2]));
+        AddLine("MOV.{} {}.y, {};", type, temporary, Visit(operation[1]));
+        AddLine("BFE.{} {}.x, {}, {};", type, temporary, temporary, Visit(operation[0]));
+        return fmt::format("{}.x", temporary);
+    }
+
+    template <char swizzle>
+    std::string LocalInvocationId(Operation) {
+        return fmt::format("invocation.localid.{}", swizzle);
+    }
+
+    template <char swizzle>
+    std::string WorkGroupId(Operation) {
+        return fmt::format("invocation.groupid.{}", swizzle);
+    }
+
+    template <char c1, char c2>
+    std::string ThreadMask(Operation) {
+        return fmt::format("{}.thread{}{}mask", StageInputName(stage), c1, c2);
+    }
+
+    template <typename... Args>
+    void AddExpression(std::string_view text, Args&&... args) {
+        shader_source += fmt::format(text, std::forward<Args>(args)...);
+    }
+
+    template <typename... Args>
+    void AddLine(std::string_view text, Args&&... args) {
+        AddExpression(text, std::forward<Args>(args)...);
+        shader_source += '\n';
+    }
+
+    std::string AllocTemporary() {
+        max_temporaries = std::max(max_temporaries, num_temporaries + 1);
+        return fmt::format("T{}.x", num_temporaries++);
+    }
+
+    std::string AllocVectorTemporary() {
+        max_temporaries = std::max(max_temporaries, num_temporaries + 1);
+        return fmt::format("T{}", num_temporaries++);
+    }
+
+    void ResetTemporaries() noexcept {
+        num_temporaries = 0;
+    }
+
+    const Device& device;
+    const ShaderIR& ir;
+    const Registry& registry;
+    const ShaderType stage;
+
+    std::size_t num_temporaries = 0;
+    std::size_t max_temporaries = 0;
+
+    std::string shader_source;
+
+    static constexpr std::string_view ADD_F32 = "ADD.F32";
+    static constexpr std::string_view ADD_S = "ADD.S";
+    static constexpr std::string_view ADD_U = "ADD.U";
+    static constexpr std::string_view MUL_F32 = "MUL.F32";
+    static constexpr std::string_view MUL_S = "MUL.S";
+    static constexpr std::string_view MUL_U = "MUL.U";
+    static constexpr std::string_view DIV_F32 = "DIV.F32";
+    static constexpr std::string_view DIV_S = "DIV.S";
+    static constexpr std::string_view DIV_U = "DIV.U";
+    static constexpr std::string_view MAD_F32 = "MAD.F32";
+    static constexpr std::string_view RSQ_F32 = "RSQ.F32";
+    static constexpr std::string_view COS_F32 = "COS.F32";
+    static constexpr std::string_view SIN_F32 = "SIN.F32";
+    static constexpr std::string_view EX2_F32 = "EX2.F32";
+    static constexpr std::string_view LG2_F32 = "LG2.F32";
+    static constexpr std::string_view SLT_F = "SLT.F32";
+    static constexpr std::string_view SLT_S = "SLT.S";
+    static constexpr std::string_view SLT_U = "SLT.U";
+    static constexpr std::string_view SEQ_F = "SEQ.F32";
+    static constexpr std::string_view SEQ_S = "SEQ.S";
+    static constexpr std::string_view SEQ_U = "SEQ.U";
+    static constexpr std::string_view SLE_F = "SLE.F32";
+    static constexpr std::string_view SLE_S = "SLE.S";
+    static constexpr std::string_view SLE_U = "SLE.U";
+    static constexpr std::string_view SGT_F = "SGT.F32";
+    static constexpr std::string_view SGT_S = "SGT.S";
+    static constexpr std::string_view SGT_U = "SGT.U";
+    static constexpr std::string_view SNE_F = "SNE.F32";
+    static constexpr std::string_view SNE_S = "SNE.S";
+    static constexpr std::string_view SNE_U = "SNE.U";
+    static constexpr std::string_view SGE_F = "SGE.F32";
+    static constexpr std::string_view SGE_S = "SGE.S";
+    static constexpr std::string_view SGE_U = "SGE.U";
+    static constexpr std::string_view AND_S = "AND.S";
+    static constexpr std::string_view AND_U = "AND.U";
+    static constexpr std::string_view TRUNC_F = "TRUNC.F";
+    static constexpr std::string_view TRUNC_S = "TRUNC.S";
+    static constexpr std::string_view TRUNC_U = "TRUNC.U";
+    static constexpr std::string_view SHL_S = "SHL.S";
+    static constexpr std::string_view SHL_U = "SHL.U";
+    static constexpr std::string_view SHR_S = "SHR.S";
+    static constexpr std::string_view SHR_U = "SHR.U";
+    static constexpr std::string_view OR_S = "OR.S";
+    static constexpr std::string_view OR_U = "OR.U";
+    static constexpr std::string_view XOR_S = "XOR.S";
+    static constexpr std::string_view XOR_U = "XOR.U";
+    static constexpr std::string_view NOT_S = "NOT.S";
+    static constexpr std::string_view NOT_U = "NOT.U";
+    static constexpr std::string_view BTC_S = "BTC.S";
+    static constexpr std::string_view BTC_U = "BTC.U";
+    static constexpr std::string_view BTFM_S = "BTFM.S";
+    static constexpr std::string_view BTFM_U = "BTFM.U";
+    static constexpr std::string_view ROUND_F = "ROUND.F";
+    static constexpr std::string_view CEIL_F = "CEIL.F";
+    static constexpr std::string_view FLR_F = "FLR.F";
+    static constexpr std::string_view I2F_S = "I2F.S";
+    static constexpr std::string_view I2F_U = "I2F.U";
+    static constexpr std::string_view MIN_F = "MIN.F";
+    static constexpr std::string_view MIN_S = "MIN.S";
+    static constexpr std::string_view MIN_U = "MIN.U";
+    static constexpr std::string_view MAX_F = "MAX.F";
+    static constexpr std::string_view MAX_S = "MAX.S";
+    static constexpr std::string_view MAX_U = "MAX.U";
+    static constexpr std::string_view MOV_U = "MOV.U";
+    static constexpr std::string_view TGBALLOT_U = "TGBALLOT.U";
+    static constexpr std::string_view TGALL_U = "TGALL.U";
+    static constexpr std::string_view TGANY_U = "TGANY.U";
+    static constexpr std::string_view TGEQ_U = "TGEQ.U";
+    static constexpr std::string_view EXCH = "EXCH";
+    static constexpr std::string_view ADD = "ADD";
+    static constexpr std::string_view MIN = "MIN";
+    static constexpr std::string_view MAX = "MAX";
+    static constexpr std::string_view AND = "AND";
+    static constexpr std::string_view OR = "OR";
+    static constexpr std::string_view XOR = "XOR";
+    static constexpr std::string_view U32 = "U32";
+    static constexpr std::string_view S32 = "S32";
+
+    static constexpr std::size_t NUM_ENTRIES = static_cast<std::size_t>(OperationCode::Amount);
+    using DecompilerType = std::string (ARBDecompiler::*)(Operation);
+    static constexpr std::array<DecompilerType, NUM_ENTRIES> OPERATION_DECOMPILERS = {
+        &ARBDecompiler::Assign,
+
+        &ARBDecompiler::Select,
+
+        &ARBDecompiler::Binary<ADD_F32>,
+        &ARBDecompiler::Binary<MUL_F32>,
+        &ARBDecompiler::Binary<DIV_F32>,
+        &ARBDecompiler::Trinary<MAD_F32>,
+        &ARBDecompiler::Negate<'F'>,
+        &ARBDecompiler::Absolute<'F'>,
+        &ARBDecompiler::FClamp,
+        &ARBDecompiler::FCastHalf0,
+        &ARBDecompiler::FCastHalf1,
+        &ARBDecompiler::Binary<MIN_F>,
+        &ARBDecompiler::Binary<MAX_F>,
+        &ARBDecompiler::Unary<COS_F32>,
+        &ARBDecompiler::Unary<SIN_F32>,
+        &ARBDecompiler::Unary<EX2_F32>,
+        &ARBDecompiler::Unary<LG2_F32>,
+        &ARBDecompiler::Unary<RSQ_F32>,
+        &ARBDecompiler::FSqrt,
+        &ARBDecompiler::Unary<ROUND_F>,
+        &ARBDecompiler::Unary<FLR_F>,
+        &ARBDecompiler::Unary<CEIL_F>,
+        &ARBDecompiler::Unary<TRUNC_F>,
+        &ARBDecompiler::Unary<I2F_S>,
+        &ARBDecompiler::Unary<I2F_U>,
+        &ARBDecompiler::FSwizzleAdd,
+
+        &ARBDecompiler::Binary<ADD_S>,
+        &ARBDecompiler::Binary<MUL_S>,
+        &ARBDecompiler::Binary<DIV_S>,
+        &ARBDecompiler::Negate<'S'>,
+        &ARBDecompiler::Absolute<'S'>,
+        &ARBDecompiler::Binary<MIN_S>,
+        &ARBDecompiler::Binary<MAX_S>,
+
+        &ARBDecompiler::Unary<TRUNC_S>,
+        &ARBDecompiler::Unary<MOV_U>,
+        &ARBDecompiler::Binary<SHL_S>,
+        &ARBDecompiler::Binary<SHR_U>,
+        &ARBDecompiler::Binary<SHR_S>,
+        &ARBDecompiler::Binary<AND_S>,
+        &ARBDecompiler::Binary<OR_S>,
+        &ARBDecompiler::Binary<XOR_S>,
+        &ARBDecompiler::Unary<NOT_S>,
+        &ARBDecompiler::BitfieldInsert<'S'>,
+        &ARBDecompiler::BitfieldExtract<'S'>,
+        &ARBDecompiler::Unary<BTC_S>,
+        &ARBDecompiler::Unary<BTFM_S>,
+
+        &ARBDecompiler::Binary<ADD_U>,
+        &ARBDecompiler::Binary<MUL_U>,
+        &ARBDecompiler::Binary<DIV_U>,
+        &ARBDecompiler::Binary<MIN_U>,
+        &ARBDecompiler::Binary<MAX_U>,
+        &ARBDecompiler::Unary<TRUNC_U>,
+        &ARBDecompiler::Unary<MOV_U>,
+        &ARBDecompiler::Binary<SHL_U>,
+        &ARBDecompiler::Binary<SHR_U>,
+        &ARBDecompiler::Binary<SHR_U>,
+        &ARBDecompiler::Binary<AND_U>,
+        &ARBDecompiler::Binary<OR_U>,
+        &ARBDecompiler::Binary<XOR_U>,
+        &ARBDecompiler::Unary<NOT_U>,
+        &ARBDecompiler::BitfieldInsert<'U'>,
+        &ARBDecompiler::BitfieldExtract<'U'>,
+        &ARBDecompiler::Unary<BTC_U>,
+        &ARBDecompiler::Unary<BTFM_U>,
+
+        &ARBDecompiler::HAdd2,
+        &ARBDecompiler::HMul2,
+        &ARBDecompiler::HFma2,
+        &ARBDecompiler::HAbsolute,
+        &ARBDecompiler::HNegate,
+        &ARBDecompiler::HClamp,
+        &ARBDecompiler::HCastFloat,
+        &ARBDecompiler::HUnpack,
+        &ARBDecompiler::HMergeF32,
+        &ARBDecompiler::HMergeH0,
+        &ARBDecompiler::HMergeH1,
+        &ARBDecompiler::HPack2,
+
+        &ARBDecompiler::LogicalAssign,
+        &ARBDecompiler::Binary<AND_U>,
+        &ARBDecompiler::Binary<OR_U>,
+        &ARBDecompiler::Binary<XOR_U>,
+        &ARBDecompiler::Unary<NOT_U>,
+        &ARBDecompiler::LogicalPick2,
+        &ARBDecompiler::LogicalAnd2,
+
+        &ARBDecompiler::FloatComparison<SLT_F, false>,
+        &ARBDecompiler::FloatComparison<SEQ_F, false>,
+        &ARBDecompiler::FloatComparison<SLE_F, false>,
+        &ARBDecompiler::FloatComparison<SGT_F, false>,
+        &ARBDecompiler::FloatComparison<SNE_F, false>,
+        &ARBDecompiler::FloatComparison<SGE_F, false>,
+        &ARBDecompiler::FloatOrdered,
+        &ARBDecompiler::FloatUnordered,
+        &ARBDecompiler::FloatComparison<SLT_F, true>,
+        &ARBDecompiler::FloatComparison<SEQ_F, true>,
+        &ARBDecompiler::FloatComparison<SLE_F, true>,
+        &ARBDecompiler::FloatComparison<SGT_F, true>,
+        &ARBDecompiler::FloatComparison<SNE_F, true>,
+        &ARBDecompiler::FloatComparison<SGE_F, true>,
+
+        &ARBDecompiler::Binary<SLT_S>,
+        &ARBDecompiler::Binary<SEQ_S>,
+        &ARBDecompiler::Binary<SLE_S>,
+        &ARBDecompiler::Binary<SGT_S>,
+        &ARBDecompiler::Binary<SNE_S>,
+        &ARBDecompiler::Binary<SGE_S>,
+
+        &ARBDecompiler::Binary<SLT_U>,
+        &ARBDecompiler::Binary<SEQ_U>,
+        &ARBDecompiler::Binary<SLE_U>,
+        &ARBDecompiler::Binary<SGT_U>,
+        &ARBDecompiler::Binary<SNE_U>,
+        &ARBDecompiler::Binary<SGE_U>,
+
+        &ARBDecompiler::LogicalAddCarry,
+
+        &ARBDecompiler::HalfComparison<SLT_F, false>,
+        &ARBDecompiler::HalfComparison<SEQ_F, false>,
+        &ARBDecompiler::HalfComparison<SLE_F, false>,
+        &ARBDecompiler::HalfComparison<SGT_F, false>,
+        &ARBDecompiler::HalfComparison<SNE_F, false>,
+        &ARBDecompiler::HalfComparison<SGE_F, false>,
+        &ARBDecompiler::HalfComparison<SLT_F, true>,
+        &ARBDecompiler::HalfComparison<SEQ_F, true>,
+        &ARBDecompiler::HalfComparison<SLE_F, true>,
+        &ARBDecompiler::HalfComparison<SGT_F, true>,
+        &ARBDecompiler::HalfComparison<SNE_F, true>,
+        &ARBDecompiler::HalfComparison<SGE_F, true>,
+
+        &ARBDecompiler::Texture,
+        &ARBDecompiler::Texture,
+        &ARBDecompiler::TextureGather,
+        &ARBDecompiler::TextureQueryDimensions,
+        &ARBDecompiler::TextureQueryLod,
+        &ARBDecompiler::TexelFetch,
+        &ARBDecompiler::TextureGradient,
+
+        &ARBDecompiler::ImageLoad,
+        &ARBDecompiler::ImageStore,
+
+        &ARBDecompiler::AtomicImage<ADD, U32>,
+        &ARBDecompiler::AtomicImage<AND, U32>,
+        &ARBDecompiler::AtomicImage<OR, U32>,
+        &ARBDecompiler::AtomicImage<XOR, U32>,
+        &ARBDecompiler::AtomicImage<EXCH, U32>,
+
+        &ARBDecompiler::Atomic<EXCH, U32>,
+        &ARBDecompiler::Atomic<ADD, U32>,
+        &ARBDecompiler::Atomic<MIN, U32>,
+        &ARBDecompiler::Atomic<MAX, U32>,
+        &ARBDecompiler::Atomic<AND, U32>,
+        &ARBDecompiler::Atomic<OR, U32>,
+        &ARBDecompiler::Atomic<XOR, U32>,
+
+        &ARBDecompiler::Atomic<EXCH, S32>,
+        &ARBDecompiler::Atomic<ADD, S32>,
+        &ARBDecompiler::Atomic<MIN, S32>,
+        &ARBDecompiler::Atomic<MAX, S32>,
+        &ARBDecompiler::Atomic<AND, S32>,
+        &ARBDecompiler::Atomic<OR, S32>,
+        &ARBDecompiler::Atomic<XOR, S32>,
+
+        &ARBDecompiler::Atomic<ADD, U32>,
+        &ARBDecompiler::Atomic<MIN, U32>,
+        &ARBDecompiler::Atomic<MAX, U32>,
+        &ARBDecompiler::Atomic<AND, U32>,
+        &ARBDecompiler::Atomic<OR, U32>,
+        &ARBDecompiler::Atomic<XOR, U32>,
+
+        &ARBDecompiler::Atomic<ADD, S32>,
+        &ARBDecompiler::Atomic<MIN, S32>,
+        &ARBDecompiler::Atomic<MAX, S32>,
+        &ARBDecompiler::Atomic<AND, S32>,
+        &ARBDecompiler::Atomic<OR, S32>,
+        &ARBDecompiler::Atomic<XOR, S32>,
+
+        &ARBDecompiler::Branch,
+        &ARBDecompiler::BranchIndirect,
+        &ARBDecompiler::PushFlowStack,
+        &ARBDecompiler::PopFlowStack,
+        &ARBDecompiler::Exit,
+        &ARBDecompiler::Discard,
+
+        &ARBDecompiler::EmitVertex,
+        &ARBDecompiler::EndPrimitive,
+
+        &ARBDecompiler::InvocationId,
+        &ARBDecompiler::YNegate,
+        &ARBDecompiler::LocalInvocationId<'x'>,
+        &ARBDecompiler::LocalInvocationId<'y'>,
+        &ARBDecompiler::LocalInvocationId<'z'>,
+        &ARBDecompiler::WorkGroupId<'x'>,
+        &ARBDecompiler::WorkGroupId<'y'>,
+        &ARBDecompiler::WorkGroupId<'z'>,
+
+        &ARBDecompiler::Unary<TGBALLOT_U>,
+        &ARBDecompiler::Unary<TGALL_U>,
+        &ARBDecompiler::Unary<TGANY_U>,
+        &ARBDecompiler::Unary<TGEQ_U>,
+
+        &ARBDecompiler::ThreadId,
+        &ARBDecompiler::ThreadMask<'e', 'q'>,
+        &ARBDecompiler::ThreadMask<'g', 'e'>,
+        &ARBDecompiler::ThreadMask<'g', 't'>,
+        &ARBDecompiler::ThreadMask<'l', 'e'>,
+        &ARBDecompiler::ThreadMask<'l', 't'>,
+        &ARBDecompiler::ShuffleIndexed,
+
+        &ARBDecompiler::Barrier,
+        &ARBDecompiler::MemoryBarrierGroup,
+        &ARBDecompiler::MemoryBarrierGlobal,
+    };
+};
+
+ARBDecompiler::ARBDecompiler(const Device& device, const ShaderIR& ir, const Registry& registry,
+                             ShaderType stage, std::string_view identifier)
+    : device{device}, ir{ir}, registry{registry}, stage{stage} {
+    AddLine("TEMP RC;");
+    AddLine("TEMP FSWZA[4];");
+    AddLine("TEMP FSWZB[4];");
+    if (ir.IsDecompiled()) {
+        DecompileAST();
+    } else {
+        DecompileBranchMode();
+    }
+    AddLine("END");
+
+    const std::string code = std::move(shader_source);
+    DeclareHeader();
+    DeclareVertex();
+    DeclareGeometry();
+    DeclareFragment();
+    DeclareCompute();
+    DeclareInputAttributes();
+    DeclareOutputAttributes();
+    DeclareLocalMemory();
+    DeclareGlobalMemory();
+    DeclareConstantBuffers();
+    DeclareRegisters();
+    DeclareTemporaries();
+    DeclarePredicates();
+    DeclareInternalFlags();
+
+    shader_source += code;
+}
+
+std::string_view HeaderStageName(ShaderType stage) {
+    switch (stage) {
+    case ShaderType::Vertex:
+        return "vp";
+    case ShaderType::Geometry:
+        return "gp";
+    case ShaderType::Fragment:
+        return "fp";
+    case ShaderType::Compute:
+        return "cp";
+    default:
+        UNREACHABLE();
+        return "";
+    }
+}
+
+void ARBDecompiler::DeclareHeader() {
+    AddLine("!!NV{}5.0", HeaderStageName(stage));
+    // Enabling this allows us to cheat on some instructions like TXL with SHADOWARRAY2D
+    AddLine("OPTION NV_internal;");
+    AddLine("OPTION NV_gpu_program_fp64;");
+    AddLine("OPTION NV_shader_storage_buffer;");
+    AddLine("OPTION NV_shader_thread_group;");
+    if (ir.UsesWarps() && device.HasWarpIntrinsics()) {
+        AddLine("OPTION NV_shader_thread_shuffle;");
+    }
+    if (stage == ShaderType::Vertex) {
+        if (device.HasNvViewportArray2()) {
+            AddLine("OPTION NV_viewport_array2;");
+        }
+    }
+    if (stage == ShaderType::Fragment) {
+        AddLine("OPTION ARB_draw_buffers;");
+    }
+    if (device.HasImageLoadFormatted()) {
+        AddLine("OPTION EXT_shader_image_load_formatted;");
+    }
+}
+
+void ARBDecompiler::DeclareVertex() {
+    if (stage != ShaderType::Vertex) {
+        return;
+    }
+    AddLine("OUTPUT result_clip[] = {{ result.clip[0..7] }};");
+}
+
+void ARBDecompiler::DeclareGeometry() {
+    if (stage != ShaderType::Geometry) {
+        return;
+    }
+    const auto& info = registry.GetGraphicsInfo();
+    const auto& header = ir.GetHeader();
+    AddLine("PRIMITIVE_IN {};", PrimitiveDescription(info.primitive_topology));
+    AddLine("PRIMITIVE_OUT {};", TopologyName(header.common3.output_topology));
+    AddLine("VERTICES_OUT {};", header.common4.max_output_vertices.Value());
+    AddLine("ATTRIB vertex_position = vertex.position;");
+}
+
+void ARBDecompiler::DeclareFragment() {
+    if (stage != ShaderType::Fragment) {
+        return;
+    }
+    AddLine("OUTPUT result_color7 = result.color[7];");
+    AddLine("OUTPUT result_color6 = result.color[6];");
+    AddLine("OUTPUT result_color5 = result.color[5];");
+    AddLine("OUTPUT result_color4 = result.color[4];");
+    AddLine("OUTPUT result_color3 = result.color[3];");
+    AddLine("OUTPUT result_color2 = result.color[2];");
+    AddLine("OUTPUT result_color1 = result.color[1];");
+    AddLine("OUTPUT result_color0 = result.color;");
+}
+
+void ARBDecompiler::DeclareCompute() {
+    if (stage != ShaderType::Compute) {
+        return;
+    }
+    const ComputeInfo& info = registry.GetComputeInfo();
+    AddLine("GROUP_SIZE {} {} {};", info.workgroup_size[0], info.workgroup_size[1],
+            info.workgroup_size[2]);
+    if (info.shared_memory_size_in_words > 0) {
+        const u32 size_in_bytes = info.shared_memory_size_in_words * 4;
+        AddLine("SHARED_MEMORY {};", size_in_bytes);
+        AddLine("SHARED shared_mem[] = {{program.sharedmem}};");
+    }
+}
+
+void ARBDecompiler::DeclareInputAttributes() {
+    if (stage == ShaderType::Compute) {
+        return;
+    }
+    const std::string_view stage_name = StageInputName(stage);
+    for (const auto attribute : ir.GetInputAttributes()) {
+        if (!IsGenericAttribute(attribute)) {
+            continue;
+        }
+        const u32 index = GetGenericAttributeIndex(attribute);
+
+        std::string_view suffix;
+        if (stage == ShaderType::Fragment) {
+            const auto input_mode{ir.GetHeader().ps.GetPixelImap(index)};
+            if (input_mode == PixelImap::Unused) {
+                return;
+            }
+            suffix = GetInputFlags(input_mode);
+        }
+        AddLine("{}ATTRIB in_attr{}[] = {{ {}.attrib[{}..{}] }};", suffix, index, stage_name, index,
+                index);
+    }
+}
+
+void ARBDecompiler::DeclareOutputAttributes() {
+    if (stage == ShaderType::Compute) {
+        return;
+    }
+    for (const auto attribute : ir.GetOutputAttributes()) {
+        if (!IsGenericAttribute(attribute)) {
+            continue;
+        }
+        const u32 index = GetGenericAttributeIndex(attribute);
+        AddLine("OUTPUT out_attr{}[] = {{ result.attrib[{}..{}] }};", index, index, index);
+    }
+}
+
+void ARBDecompiler::DeclareLocalMemory() {
+    u64 size = 0;
+    if (stage == ShaderType::Compute) {
+        size = registry.GetComputeInfo().local_memory_size_in_words * 4ULL;
+    } else {
+        size = ir.GetHeader().GetLocalMemorySize();
+    }
+    if (size == 0) {
+        return;
+    }
+    const u64 element_count = Common::AlignUp(size, 4) / 4;
+    AddLine("TEMP lmem[{}];", element_count);
+}
+
+void ARBDecompiler::DeclareGlobalMemory() {
+    u32 binding = 0; // device.GetBaseBindings(stage).shader_storage_buffer;
+    for (const auto& pair : ir.GetGlobalMemory()) {
+        const auto& base = pair.first;
+        AddLine("STORAGE {}[] = {{ program.storage[{}] }};", GlobalMemoryName(base), binding);
+        ++binding;
+    }
+}
+
+void ARBDecompiler::DeclareConstantBuffers() {
+    u32 binding = 0;
+    for (const auto& cbuf : ir.GetConstantBuffers()) {
+        AddLine("CBUFFER cbuf{}[] = {{ program.buffer[{}] }};", cbuf.first, binding);
+        ++binding;
+    }
+}
+
+void ARBDecompiler::DeclareRegisters() {
+    for (const u32 gpr : ir.GetRegisters()) {
+        AddLine("TEMP R{};", gpr);
+    }
+}
+
+void ARBDecompiler::DeclareTemporaries() {
+    for (std::size_t i = 0; i < max_temporaries; ++i) {
+        AddLine("TEMP T{};", i);
+    }
+}
+
+void ARBDecompiler::DeclarePredicates() {
+    for (const Tegra::Shader::Pred pred : ir.GetPredicates()) {
+        AddLine("TEMP P{};", static_cast<u64>(pred));
+    }
+}
+
+void ARBDecompiler::DeclareInternalFlags() {
+    for (const char* name : INTERNAL_FLAG_NAMES) {
+        AddLine("TEMP {};", name);
+    }
+}
+
+void ARBDecompiler::InitializeVariables() {
+    AddLine("MOV.F32 FSWZA[0], -1;");
+    AddLine("MOV.F32 FSWZA[1], 1;");
+    AddLine("MOV.F32 FSWZA[2], -1;");
+    AddLine("MOV.F32 FSWZA[3], 0;");
+    AddLine("MOV.F32 FSWZB[0], -1;");
+    AddLine("MOV.F32 FSWZB[1], -1;");
+    AddLine("MOV.F32 FSWZB[2], 1;");
+    AddLine("MOV.F32 FSWZB[3], -1;");
+
+    if (stage == ShaderType::Vertex || stage == ShaderType::Geometry) {
+        AddLine("MOV.F result.position, {{0, 0, 0, 1}};");
+    }
+    for (const auto attribute : ir.GetOutputAttributes()) {
+        if (!IsGenericAttribute(attribute)) {
+            continue;
+        }
+        const u32 index = GetGenericAttributeIndex(attribute);
+        AddLine("MOV.F result.attrib[{}], {{0, 0, 0, 1}};", index);
+    }
+    for (const u32 gpr : ir.GetRegisters()) {
+        AddLine("MOV.F R{}, {{0, 0, 0, 0}};", gpr);
+    }
+    for (const Tegra::Shader::Pred pred : ir.GetPredicates()) {
+        AddLine("MOV.U P{}, {{0, 0, 0, 0}};", static_cast<u64>(pred));
+    }
+}
+
+void ARBDecompiler::DecompileAST() {
+    const u32 num_flow_variables = ir.GetASTNumVariables();
+    for (u32 i = 0; i < num_flow_variables; ++i) {
+        AddLine("TEMP F{};", i);
+    }
+    for (u32 i = 0; i < num_flow_variables; ++i) {
+        AddLine("MOV.U F{}, {{0, 0, 0, 0}};", i);
+    }
+
+    InitializeVariables();
+
+    VisitAST(ir.GetASTProgram());
+}
+
+void ARBDecompiler::DecompileBranchMode() {
+    static constexpr u32 FLOW_STACK_SIZE = 20;
+    if (!ir.IsFlowStackDisabled()) {
+        AddLine("TEMP SSY[{}];", FLOW_STACK_SIZE);
+        AddLine("TEMP PBK[{}];", FLOW_STACK_SIZE);
+        AddLine("TEMP SSY_TOP;");
+        AddLine("TEMP PBK_TOP;");
+    }
+
+    AddLine("TEMP PC;");
+
+    if (!ir.IsFlowStackDisabled()) {
+        AddLine("MOV.U SSY_TOP.x, 0;");
+        AddLine("MOV.U PBK_TOP.x, 0;");
+    }
+
+    InitializeVariables();
+
+    const auto basic_block_end = ir.GetBasicBlocks().end();
+    auto basic_block_it = ir.GetBasicBlocks().begin();
+    const u32 first_address = basic_block_it->first;
+    AddLine("MOV.U PC.x, {};", first_address);
+
+    AddLine("REP;");
+
+    std::size_t num_blocks = 0;
+    while (basic_block_it != basic_block_end) {
+        const auto& [address, bb] = *basic_block_it;
+        ++num_blocks;
+
+        AddLine("SEQ.S.CC RC.x, PC.x, {};", address);
+        AddLine("IF NE.x;");
+
+        VisitBlock(bb);
+
+        ++basic_block_it;
+
+        if (basic_block_it != basic_block_end) {
+            const auto op = std::get_if<OperationNode>(&*bb[bb.size() - 1]);
+            if (!op || op->GetCode() != OperationCode::Branch) {
+                const u32 next_address = basic_block_it->first;
+                AddLine("MOV.U PC.x, {};", next_address);
+                AddLine("CONT;");
+            }
+        }
+
+        AddLine("ELSE;");
+    }
+    AddLine("RET;");
+    while (num_blocks--) {
+        AddLine("ENDIF;");
+    }
+
+    AddLine("ENDREP;");
+}
+
+void ARBDecompiler::VisitAST(const ASTNode& node) {
+    if (const auto ast = std::get_if<ASTProgram>(&*node->GetInnerData())) {
+        for (ASTNode current = ast->nodes.GetFirst(); current; current = current->GetNext()) {
+            VisitAST(current);
+        }
+    } else if (const auto ast = std::get_if<ASTIfThen>(&*node->GetInnerData())) {
+        const std::string condition = VisitExpression(ast->condition);
+        ResetTemporaries();
+
+        AddLine("MOVC.U RC.x, {};", condition);
+        AddLine("IF NE.x;");
+        for (ASTNode current = ast->nodes.GetFirst(); current; current = current->GetNext()) {
+            VisitAST(current);
+        }
+        AddLine("ENDIF;");
+    } else if (const auto ast = std::get_if<ASTIfElse>(&*node->GetInnerData())) {
+        AddLine("ELSE;");
+        for (ASTNode current = ast->nodes.GetFirst(); current; current = current->GetNext()) {
+            VisitAST(current);
+        }
+    } else if (const auto ast = std::get_if<ASTBlockDecoded>(&*node->GetInnerData())) {
+        VisitBlock(ast->nodes);
+    } else if (const auto ast = std::get_if<ASTVarSet>(&*node->GetInnerData())) {
+        AddLine("MOV.U F{}, {};", ast->index, VisitExpression(ast->condition));
+        ResetTemporaries();
+    } else if (const auto ast = std::get_if<ASTDoWhile>(&*node->GetInnerData())) {
+        const std::string condition = VisitExpression(ast->condition);
+        ResetTemporaries();
+        AddLine("REP;");
+        for (ASTNode current = ast->nodes.GetFirst(); current; current = current->GetNext()) {
+            VisitAST(current);
+        }
+        AddLine("MOVC.U RC.x, {};", condition);
+        AddLine("BRK (NE.x);");
+        AddLine("ENDREP;");
+    } else if (const auto ast = std::get_if<ASTReturn>(&*node->GetInnerData())) {
+        const bool is_true = ExprIsTrue(ast->condition);
+        if (!is_true) {
+            AddLine("MOVC.U RC.x, {};", VisitExpression(ast->condition));
+            AddLine("IF NE.x;");
+            ResetTemporaries();
+        }
+        if (ast->kills) {
+            AddLine("KIL TR;");
+        } else {
+            Exit();
+        }
+        if (!is_true) {
+            AddLine("ENDIF;");
+        }
+    } else if (const auto ast = std::get_if<ASTBreak>(&*node->GetInnerData())) {
+        if (ExprIsTrue(ast->condition)) {
+            AddLine("BRK;");
+        } else {
+            AddLine("MOVC.U RC.x, {};", VisitExpression(ast->condition));
+            AddLine("BRK (NE.x);");
+            ResetTemporaries();
+        }
+    } else if (std::holds_alternative<ASTLabel>(*node->GetInnerData())) {
+        // Nothing to do
+    } else {
+        UNREACHABLE();
+    }
+}
+
+std::string ARBDecompiler::VisitExpression(const Expr& node) {
+    if (const auto expr = std::get_if<ExprAnd>(&*node)) {
+        std::string result = AllocTemporary();
+        AddLine("AND.U {}, {}, {};", result, VisitExpression(expr->operand1),
+                VisitExpression(expr->operand2));
+        return result;
+    }
+    if (const auto expr = std::get_if<ExprOr>(&*node)) {
+        std::string result = AllocTemporary();
+        AddLine("OR.U {}, {}, {};", result, VisitExpression(expr->operand1),
+                VisitExpression(expr->operand2));
+        return result;
+    }
+    if (const auto expr = std::get_if<ExprNot>(&*node)) {
+        std::string result = AllocTemporary();
+        AddLine("CMP.S {}, {}, 0, -1;", result, VisitExpression(expr->operand1));
+        return result;
+    }
+    if (const auto expr = std::get_if<ExprPredicate>(&*node)) {
+        return fmt::format("P{}.x", static_cast<u64>(expr->predicate));
+    }
+    if (const auto expr = std::get_if<ExprCondCode>(&*node)) {
+        return Visit(ir.GetConditionCode(expr->cc));
+    }
+    if (const auto expr = std::get_if<ExprVar>(&*node)) {
+        return fmt::format("F{}.x", expr->var_index);
+    }
+    if (const auto expr = std::get_if<ExprBoolean>(&*node)) {
+        return expr->value ? "0xffffffff" : "0";
+    }
+    if (const auto expr = std::get_if<ExprGprEqual>(&*node)) {
+        std::string result = AllocTemporary();
+        AddLine("SEQ.U {}, R{}.x, {};", result, expr->gpr, expr->value);
+        return result;
+    }
+    UNREACHABLE();
+    return "0";
+}
+
+void ARBDecompiler::VisitBlock(const NodeBlock& bb) {
+    for (const auto& node : bb) {
+        Visit(node);
+    }
+}
+
+std::string ARBDecompiler::Visit(const Node& node) {
+    if (const auto operation = std::get_if<OperationNode>(&*node)) {
+        if (const auto amend_index = operation->GetAmendIndex()) {
+            Visit(ir.GetAmendNode(*amend_index));
+        }
+        const std::size_t index = static_cast<std::size_t>(operation->GetCode());
+        if (index >= OPERATION_DECOMPILERS.size()) {
+            UNREACHABLE_MSG("Out of bounds operation: {}", index);
+            return {};
+        }
+        const auto decompiler = OPERATION_DECOMPILERS[index];
+        if (decompiler == nullptr) {
+            UNREACHABLE_MSG("Undefined operation: {}", index);
+            return {};
+        }
+        return (this->*decompiler)(*operation);
+    }
+
+    if (const auto gpr = std::get_if<GprNode>(&*node)) {
+        const u32 index = gpr->GetIndex();
+        if (index == Register::ZeroIndex) {
+            return "{0, 0, 0, 0}.x";
+        }
+        return fmt::format("R{}.x", index);
+    }
+
+    if (const auto cv = std::get_if<CustomVarNode>(&*node)) {
+        return fmt::format("CV{}.x", cv->GetIndex());
+    }
+
+    if (const auto immediate = std::get_if<ImmediateNode>(&*node)) {
+        std::string temporary = AllocTemporary();
+        AddLine("MOV.U {}, {};", temporary, immediate->GetValue());
+        return temporary;
+    }
+
+    if (const auto predicate = std::get_if<PredicateNode>(&*node)) {
+        std::string temporary = AllocTemporary();
+        switch (const auto index = predicate->GetIndex(); index) {
+        case Tegra::Shader::Pred::UnusedIndex:
+            AddLine("MOV.S {}, -1;", temporary);
+            break;
+        case Tegra::Shader::Pred::NeverExecute:
+            AddLine("MOV.S {}, 0;", temporary);
+            break;
+        default:
+            AddLine("MOV.S {}, P{}.x;", temporary, static_cast<u64>(index));
+            break;
+        }
+        if (predicate->IsNegated()) {
+            AddLine("CMP.S {}, {}, 0, -1;", temporary, temporary);
+        }
+        return temporary;
+    }
+
+    if (const auto abuf = std::get_if<AbufNode>(&*node)) {
+        if (abuf->IsPhysicalBuffer()) {
+            UNIMPLEMENTED_MSG("Physical buffers are not implemented");
+            return "{0, 0, 0, 0}.x";
+        }
+
+        const auto buffer_index = [this, &abuf]() -> std::string {
+            if (stage != ShaderType::Geometry) {
+                return "";
+            }
+            return fmt::format("[{}]", Visit(abuf->GetBuffer()));
+        };
+
+        const Attribute::Index index = abuf->GetIndex();
+        const u32 element = abuf->GetElement();
+        const char swizzle = Swizzle(element);
+        switch (index) {
+        case Attribute::Index::Position: {
+            if (stage == ShaderType::Geometry) {
+                return fmt::format("{}_position[{}].{}", StageInputName(stage),
+                                   Visit(abuf->GetBuffer()), swizzle);
+            } else {
+                return fmt::format("{}.position.{}", StageInputName(stage), swizzle);
+            }
+        }
+        case Attribute::Index::TessCoordInstanceIDVertexID:
+            ASSERT(stage == ShaderType::Vertex);
+            switch (element) {
+            case 2:
+                return "vertex.instance";
+            case 3:
+                return "vertex.id";
+            }
+            UNIMPLEMENTED_MSG("Unmanaged TessCoordInstanceIDVertexID element={}", element);
+            break;
+        case Attribute::Index::PointCoord:
+            switch (element) {
+            case 0:
+                return "fragment.pointcoord.x";
+            case 1:
+                return "fragment.pointcoord.y";
+            }
+            UNIMPLEMENTED();
+            break;
+        case Attribute::Index::FrontFacing: {
+            ASSERT(stage == ShaderType::Fragment);
+            ASSERT(element == 3);
+            const std::string temporary = AllocVectorTemporary();
+            AddLine("SGT.S RC.x, fragment.facing, {{0, 0, 0, 0}};");
+            AddLine("MOV.U.CC RC.x, -RC;");
+            AddLine("MOV.S {}.x, 0;", temporary);
+            AddLine("MOV.S {}.x (NE.x), -1;", temporary);
+            return fmt::format("{}.x", temporary);
+        }
+        default:
+            if (IsGenericAttribute(index)) {
+                if (stage == ShaderType::Geometry) {
+                    return fmt::format("in_attr{}[{}][0].{}", GetGenericAttributeIndex(index),
+                                       Visit(abuf->GetBuffer()), swizzle);
+                } else {
+                    return fmt::format("{}.attrib[{}].{}", StageInputName(stage),
+                                       GetGenericAttributeIndex(index), swizzle);
+                }
+            }
+            UNIMPLEMENTED_MSG("Unimplemented input attribute={}", static_cast<int>(index));
+            break;
+        }
+        return "{0, 0, 0, 0}.x";
+    }
+
+    if (const auto cbuf = std::get_if<CbufNode>(&*node)) {
+        std::string offset_string;
+        const auto& offset = cbuf->GetOffset();
+        if (const auto imm = std::get_if<ImmediateNode>(&*offset)) {
+            offset_string = std::to_string(imm->GetValue());
+        } else {
+            offset_string = Visit(offset);
+        }
+        std::string temporary = AllocTemporary();
+        AddLine("LDC.F32 {}, cbuf{}[{}];", temporary, cbuf->GetIndex(), offset_string);
+        return temporary;
+    }
+
+    if (const auto gmem = std::get_if<GmemNode>(&*node)) {
+        std::string temporary = AllocTemporary();
+        AddLine("SUB.U {}, {}, {};", temporary, Visit(gmem->GetRealAddress()),
+                Visit(gmem->GetBaseAddress()));
+        AddLine("LDB.U32 {}, {}[{}];", temporary, GlobalMemoryName(gmem->GetDescriptor()),
+                temporary);
+        return temporary;
+    }
+
+    if (const auto lmem = std::get_if<LmemNode>(&*node)) {
+        std::string temporary = Visit(lmem->GetAddress());
+        AddLine("SHR.U {}, {}, 2;", temporary, temporary);
+        AddLine("MOV.U {}, lmem[{}].x;", temporary, temporary);
+        return temporary;
+    }
+
+    if (const auto smem = std::get_if<SmemNode>(&*node)) {
+        std::string temporary = Visit(smem->GetAddress());
+        AddLine("LDS.U32 {}, shared_mem[{}];", temporary, temporary);
+        return temporary;
+    }
+
+    if (const auto internal_flag = std::get_if<InternalFlagNode>(&*node)) {
+        const std::size_t index = static_cast<std::size_t>(internal_flag->GetFlag());
+        return fmt::format("{}.x", INTERNAL_FLAG_NAMES[index]);
+    }
+
+    if (const auto conditional = std::get_if<ConditionalNode>(&*node)) {
+        if (const auto amend_index = conditional->GetAmendIndex()) {
+            Visit(ir.GetAmendNode(*amend_index));
+        }
+        AddLine("MOVC.U RC.x, {};", Visit(conditional->GetCondition()));
+        AddLine("IF NE.x;");
+        VisitBlock(conditional->GetCode());
+        AddLine("ENDIF;");
+        return {};
+    }
+
+    if (const auto cmt = std::get_if<CommentNode>(&*node)) {
+        // Uncommenting this will generate invalid code. GLASM lacks comments.
+        // AddLine("// {}", cmt->GetText());
+        return {};
+    }
+
+    UNIMPLEMENTED();
+    return {};
+}
+
+std::pair<std::string, std::size_t> ARBDecompiler::BuildCoords(Operation operation) {
+    const auto& meta = std::get<MetaTexture>(operation.GetMeta());
+    UNIMPLEMENTED_IF(meta.sampler.is_indexed);
+    UNIMPLEMENTED_IF(meta.sampler.is_shadow && meta.sampler.is_array &&
+                     meta.sampler.type == Tegra::Shader::TextureType::TextureCube);
+
+    const std::size_t count = operation.GetOperandsCount();
+    std::string temporary = AllocVectorTemporary();
+    std::size_t i = 0;
+    for (; i < count; ++i) {
+        AddLine("MOV.F {}.{}, {};", temporary, Swizzle(i), Visit(operation[i]));
+    }
+    if (meta.sampler.is_array) {
+        AddLine("I2F.S {}.{}, {};", temporary, Swizzle(i++), Visit(meta.array));
+    }
+    if (meta.sampler.is_shadow) {
+        AddLine("MOV.F {}.{}, {};", temporary, Swizzle(i++), Visit(meta.depth_compare));
+    }
+    return {std::move(temporary), i};
+}
+
+std::string ARBDecompiler::BuildAoffi(Operation operation) {
+    const auto& meta = std::get<MetaTexture>(operation.GetMeta());
+    if (meta.aoffi.empty()) {
+        return {};
+    }
+    const std::string temporary = AllocVectorTemporary();
+    std::size_t i = 0;
+    for (auto& node : meta.aoffi) {
+        AddLine("MOV.S {}.{}, {};", temporary, Swizzle(i++), Visit(node));
+    }
+    return fmt::format(", offset({})", temporary);
+}
+
+void ARBDecompiler::Exit() {
+    if (stage != ShaderType::Fragment) {
+        AddLine("RET;");
+        return;
+    }
+
+    const auto safe_get_register = [this](u32 reg) -> std::string {
+        // TODO(Rodrigo): Replace with contains once C++20 releases
+        const auto& used_registers = ir.GetRegisters();
+        if (used_registers.find(reg) != used_registers.end()) {
+            return fmt::format("R{}.x", reg);
+        }
+        return "{0, 0, 0, 0}.x";
+    };
+
+    const auto& header = ir.GetHeader();
+    u32 current_reg = 0;
+    for (u32 rt = 0; rt < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets; ++rt) {
+        for (u32 component = 0; component < 4; ++component) {
+            if (!header.ps.IsColorComponentOutputEnabled(rt, component)) {
+                continue;
+            }
+            AddLine("MOV.F result_color{}.{}, {};", rt, Swizzle(component),
+                    safe_get_register(current_reg));
+            ++current_reg;
+        }
+    }
+    if (header.ps.omap.depth) {
+        AddLine("MOV.F result.depth.z, {};", safe_get_register(current_reg + 1));
+    }
+
+    AddLine("RET;");
+}
+
+std::string ARBDecompiler::Assign(Operation operation) {
+    const Node& dest = operation[0];
+    const Node& src = operation[1];
+
+    std::string dest_name;
+    if (const auto gpr = std::get_if<GprNode>(&*dest)) {
+        if (gpr->GetIndex() == Register::ZeroIndex) {
+            // Writing to Register::ZeroIndex is a no op
+            return {};
+        }
+        dest_name = fmt::format("R{}.x", gpr->GetIndex());
+    } else if (const auto abuf = std::get_if<AbufNode>(&*dest)) {
+        const u32 element = abuf->GetElement();
+        const char swizzle = Swizzle(element);
+        switch (const Attribute::Index index = abuf->GetIndex()) {
+        case Attribute::Index::Position:
+            dest_name = fmt::format("result.position.{}", swizzle);
+            break;
+        case Attribute::Index::LayerViewportPointSize:
+            switch (element) {
+            case 0:
+                UNIMPLEMENTED();
+                return {};
+            case 1:
+            case 2:
+                if (!device.HasNvViewportArray2()) {
+                    LOG_ERROR(
+                        Render_OpenGL,
+                        "NV_viewport_array2 is missing. Maxwell gen 2 or better is required.");
+                    return {};
+                }
+                dest_name = element == 1 ? "result.layer.x" : "result.viewport.x";
+                break;
+            case 3:
+                dest_name = "result.pointsize.x";
+                break;
+            }
+            break;
+        case Attribute::Index::ClipDistances0123:
+            dest_name = fmt::format("result.clip[{}].x", element);
+            break;
+        case Attribute::Index::ClipDistances4567:
+            dest_name = fmt::format("result.clip[{}].x", element + 4);
+            break;
+        default:
+            if (!IsGenericAttribute(index)) {
+                UNREACHABLE();
+                return {};
+            }
+            dest_name =
+                fmt::format("result.attrib[{}].{}", GetGenericAttributeIndex(index), swizzle);
+            break;
+        }
+    } else if (const auto lmem = std::get_if<LmemNode>(&*dest)) {
+        const std::string address = Visit(lmem->GetAddress());
+        AddLine("SHR.U {}, {}, 2;", address, address);
+        dest_name = fmt::format("lmem[{}].x", address);
+    } else if (const auto smem = std::get_if<SmemNode>(&*dest)) {
+        AddLine("STS.U32 {}, shared_mem[{}];", Visit(src), Visit(smem->GetAddress()));
+        ResetTemporaries();
+        return {};
+    } else if (const auto gmem = std::get_if<GmemNode>(&*dest)) {
+        const std::string temporary = AllocTemporary();
+        AddLine("SUB.U {}, {}, {};", temporary, Visit(gmem->GetRealAddress()),
+                Visit(gmem->GetBaseAddress()));
+        AddLine("STB.U32 {}, {}[{}];", Visit(src), GlobalMemoryName(gmem->GetDescriptor()),
+                temporary);
+        ResetTemporaries();
+        return {};
+    } else {
+        UNREACHABLE();
+        ResetTemporaries();
+        return {};
+    }
+
+    AddLine("MOV.U {}, {};", dest_name, Visit(src));
+    ResetTemporaries();
+    return {};
+}
+
+std::string ARBDecompiler::Select(Operation operation) {
+    std::string temporary = AllocTemporary();
+    AddLine("CMP.S {}, {}, {}, {};", temporary, Visit(operation[0]), Visit(operation[1]),
+            Visit(operation[2]));
+    return temporary;
+}
+
+std::string ARBDecompiler::FClamp(Operation operation) {
+    // 1.0f in hex, replace with std::bit_cast on C++20
+    static constexpr u32 POSITIVE_ONE = 0x3f800000;
+
+    std::string temporary = AllocTemporary();
+    const Node& value = operation[0];
+    const Node& low = operation[1];
+    const Node& high = operation[2];
+    const auto* const imm_low = std::get_if<ImmediateNode>(&*low);
+    const auto* const imm_high = std::get_if<ImmediateNode>(&*high);
+    if (imm_low && imm_high && imm_low->GetValue() == 0 && imm_high->GetValue() == POSITIVE_ONE) {
+        AddLine("MOV.F32.SAT {}, {};", temporary, Visit(value));
+    } else {
+        AddLine("MIN.F {}, {}, {};", temporary, Visit(value), Visit(high));
+        AddLine("MAX.F {}, {}, {};", temporary, temporary, Visit(low));
+    }
+    return temporary;
+}
+
+std::string ARBDecompiler::FCastHalf0(Operation operation) {
+    const std::string temporary = AllocVectorTemporary();
+    AddLine("UP2H.F {}.x, {};", temporary, Visit(operation[0]));
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::FCastHalf1(Operation operation) {
+    const std::string temporary = AllocVectorTemporary();
+    AddLine("UP2H.F {}.y, {};", temporary, Visit(operation[0]));
+    AddLine("MOV {}.x, {}.y;", temporary, temporary);
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::FSqrt(Operation operation) {
+    std::string temporary = AllocTemporary();
+    AddLine("RSQ.F32 {}, {};", temporary, Visit(operation[0]));
+    AddLine("RCP.F32 {}, {};", temporary, temporary);
+    return temporary;
+}
+
+std::string ARBDecompiler::FSwizzleAdd(Operation operation) {
+    const std::string temporary = AllocVectorTemporary();
+    if (!device.HasWarpIntrinsics()) {
+        LOG_ERROR(Render_OpenGL,
+                  "NV_shader_thread_shuffle is missing. Kepler or better is required.");
+        AddLine("ADD.F {}.x, {}, {};", temporary, Visit(operation[0]), Visit(operation[1]));
+        return fmt::format("{}.x", temporary);
+    }
+
+    AddLine("AND.U {}.z, {}.threadid, 3;", temporary, StageInputName(stage));
+    AddLine("SHL.U {}.z, {}.z, 1;", temporary, temporary);
+    AddLine("SHR.U {}.z, {}, {}.z;", temporary, Visit(operation[2]), temporary);
+    AddLine("AND.U {}.z, {}.z, 3;", temporary, temporary);
+    AddLine("MUL.F32 {}.x, {}, FSWZA[{}.z];", temporary, Visit(operation[0]), temporary);
+    AddLine("MUL.F32 {}.y, {}, FSWZB[{}.z];", temporary, Visit(operation[1]), temporary);
+    AddLine("ADD.F32 {}.x, {}.x, {}.y;", temporary, temporary, temporary);
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::HAdd2(Operation operation) {
+    const std::string tmp1 = AllocVectorTemporary();
+    const std::string tmp2 = AllocVectorTemporary();
+    AddLine("UP2H.F {}.xy, {};", tmp1, Visit(operation[0]));
+    AddLine("UP2H.F {}.xy, {};", tmp2, Visit(operation[1]));
+    AddLine("ADD.F16 {}, {}, {};", tmp1, tmp1, tmp2);
+    AddLine("PK2H.F {}.x, {};", tmp1, tmp1);
+    return fmt::format("{}.x", tmp1);
+}
+
+std::string ARBDecompiler::HMul2(Operation operation) {
+    const std::string tmp1 = AllocVectorTemporary();
+    const std::string tmp2 = AllocVectorTemporary();
+    AddLine("UP2H.F {}.xy, {};", tmp1, Visit(operation[0]));
+    AddLine("UP2H.F {}.xy, {};", tmp2, Visit(operation[1]));
+    AddLine("MUL.F16 {}, {}, {};", tmp1, tmp1, tmp2);
+    AddLine("PK2H.F {}.x, {};", tmp1, tmp1);
+    return fmt::format("{}.x", tmp1);
+}
+
+std::string ARBDecompiler::HFma2(Operation operation) {
+    const std::string tmp1 = AllocVectorTemporary();
+    const std::string tmp2 = AllocVectorTemporary();
+    const std::string tmp3 = AllocVectorTemporary();
+    AddLine("UP2H.F {}.xy, {};", tmp1, Visit(operation[0]));
+    AddLine("UP2H.F {}.xy, {};", tmp2, Visit(operation[1]));
+    AddLine("UP2H.F {}.xy, {};", tmp3, Visit(operation[2]));
+    AddLine("MAD.F16 {}, {}, {}, {};", tmp1, tmp1, tmp2, tmp3);
+    AddLine("PK2H.F {}.x, {};", tmp1, tmp1);
+    return fmt::format("{}.x", tmp1);
+}
+
+std::string ARBDecompiler::HAbsolute(Operation operation) {
+    const std::string temporary = AllocVectorTemporary();
+    AddLine("UP2H.F {}.xy, {};", temporary, Visit(operation[0]));
+    AddLine("PK2H.F {}.x, |{}|;", temporary, temporary);
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::HNegate(Operation operation) {
+    const std::string temporary = AllocVectorTemporary();
+    AddLine("UP2H.F {}.xy, {};", temporary, Visit(operation[0]));
+    AddLine("MOVC.S RC.x, {};", Visit(operation[1]));
+    AddLine("MOV.F {}.x (NE.x), -{}.x;", temporary, temporary);
+    AddLine("MOVC.S RC.x, {};", Visit(operation[2]));
+    AddLine("MOV.F {}.y (NE.x), -{}.y;", temporary, temporary);
+    AddLine("PK2H.F {}.x, {};", temporary, temporary);
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::HClamp(Operation operation) {
+    const std::string tmp1 = AllocVectorTemporary();
+    const std::string tmp2 = AllocVectorTemporary();
+    AddLine("UP2H.F {}.xy, {};", tmp1, Visit(operation[0]));
+    AddLine("MOV.U {}.x, {};", tmp2, Visit(operation[1]));
+    AddLine("MOV.U {}.y, {}.x;", tmp2, tmp2);
+    AddLine("MAX.F {}, {}, {};", tmp1, tmp1, tmp2);
+    AddLine("MOV.U {}.x, {};", tmp2, Visit(operation[2]));
+    AddLine("MOV.U {}.y, {}.x;", tmp2, tmp2);
+    AddLine("MIN.F {}, {}, {};", tmp1, tmp1, tmp2);
+    AddLine("PK2H.F {}.x, {};", tmp1, tmp1);
+    return fmt::format("{}.x", tmp1);
+}
+
+std::string ARBDecompiler::HCastFloat(Operation operation) {
+    const std::string temporary = AllocVectorTemporary();
+    AddLine("MOV.F {}.y, {{0, 0, 0, 0}};", temporary);
+    AddLine("MOV.F {}.x, {};", temporary, Visit(operation[0]));
+    AddLine("PK2H.F {}.x, {};", temporary, temporary);
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::HUnpack(Operation operation) {
+    const std::string operand = Visit(operation[0]);
+    switch (std::get<Tegra::Shader::HalfType>(operation.GetMeta())) {
+    case Tegra::Shader::HalfType::H0_H1:
+        return operand;
+    case Tegra::Shader::HalfType::F32: {
+        const std::string temporary = AllocVectorTemporary();
+        AddLine("MOV.U {}.x, {};", temporary, operand);
+        AddLine("MOV.U {}.y, {}.x;", temporary, temporary);
+        AddLine("PK2H.F {}.x, {};", temporary, temporary);
+        return fmt::format("{}.x", temporary);
+    }
+    case Tegra::Shader::HalfType::H0_H0: {
+        const std::string temporary = AllocVectorTemporary();
+        AddLine("UP2H.F {}.xy, {};", temporary, operand);
+        AddLine("MOV.U {}.y, {}.x;", temporary, temporary);
+        AddLine("PK2H.F {}.x, {};", temporary, temporary);
+        return fmt::format("{}.x", temporary);
+    }
+    case Tegra::Shader::HalfType::H1_H1: {
+        const std::string temporary = AllocVectorTemporary();
+        AddLine("UP2H.F {}.xy, {};", temporary, operand);
+        AddLine("MOV.U {}.x, {}.y;", temporary, temporary);
+        AddLine("PK2H.F {}.x, {};", temporary, temporary);
+        return fmt::format("{}.x", temporary);
+    }
+    }
+    UNREACHABLE();
+    return "{0, 0, 0, 0}.x";
+}
+
+std::string ARBDecompiler::HMergeF32(Operation operation) {
+    const std::string temporary = AllocVectorTemporary();
+    AddLine("UP2H.F {}.xy, {};", temporary, Visit(operation[0]));
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::HMergeH0(Operation operation) {
+    const std::string temporary = AllocVectorTemporary();
+    AddLine("UP2H.F {}.xy, {};", temporary, Visit(operation[0]));
+    AddLine("UP2H.F {}.zw, {};", temporary, Visit(operation[1]));
+    AddLine("MOV.U {}.x, {}.z;", temporary, temporary);
+    AddLine("PK2H.F {}.x, {};", temporary, temporary);
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::HMergeH1(Operation operation) {
+    const std::string temporary = AllocVectorTemporary();
+    AddLine("UP2H.F {}.xy, {};", temporary, Visit(operation[0]));
+    AddLine("UP2H.F {}.zw, {};", temporary, Visit(operation[1]));
+    AddLine("MOV.U {}.y, {}.w;", temporary, temporary);
+    AddLine("PK2H.F {}.x, {};", temporary, temporary);
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::HPack2(Operation operation) {
+    const std::string temporary = AllocVectorTemporary();
+    AddLine("MOV.U {}.x, {};", temporary, Visit(operation[0]));
+    AddLine("MOV.U {}.y, {};", temporary, Visit(operation[1]));
+    AddLine("PK2H.F {}.x, {};", temporary, temporary);
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::LogicalAssign(Operation operation) {
+    const Node& dest = operation[0];
+    const Node& src = operation[1];
+
+    std::string target;
+
+    if (const auto pred = std::get_if<PredicateNode>(&*dest)) {
+        ASSERT_MSG(!pred->IsNegated(), "Negating logical assignment");
+
+        const Tegra::Shader::Pred index = pred->GetIndex();
+        switch (index) {
+        case Tegra::Shader::Pred::NeverExecute:
+        case Tegra::Shader::Pred::UnusedIndex:
+            // Writing to these predicates is a no-op
+            return {};
+        }
+        target = fmt::format("P{}.x", static_cast<u64>(index));
+    } else if (const auto internal_flag = std::get_if<InternalFlagNode>(&*dest)) {
+        const std::size_t index = static_cast<std::size_t>(internal_flag->GetFlag());
+        target = fmt::format("{}.x", INTERNAL_FLAG_NAMES[index]);
+    } else {
+        UNREACHABLE();
+        ResetTemporaries();
+        return {};
+    }
+
+    AddLine("MOV.U {}, {};", target, Visit(src));
+    ResetTemporaries();
+    return {};
+}
+
+std::string ARBDecompiler::LogicalPick2(Operation operation) {
+    std::string temporary = AllocTemporary();
+    const u32 index = std::get<ImmediateNode>(*operation[1]).GetValue();
+    AddLine("MOV.U {}, {}.{};", temporary, Visit(operation[0]), Swizzle(index));
+    return temporary;
+}
+
+std::string ARBDecompiler::LogicalAnd2(Operation operation) {
+    std::string temporary = AllocTemporary();
+    const std::string op = Visit(operation[0]);
+    AddLine("AND.U {}, {}.x, {}.y;", temporary, op, op);
+    return temporary;
+}
+
+std::string ARBDecompiler::FloatOrdered(Operation operation) {
+    std::string temporary = AllocTemporary();
+    AddLine("MOVC.F32 RC.x, {};", Visit(operation[0]));
+    AddLine("MOVC.F32 RC.y, {};", Visit(operation[1]));
+    AddLine("MOV.S {}, -1;", temporary);
+    AddLine("MOV.S {} (NAN.x), 0;", temporary);
+    AddLine("MOV.S {} (NAN.y), 0;", temporary);
+    return temporary;
+}
+
+std::string ARBDecompiler::FloatUnordered(Operation operation) {
+    std::string temporary = AllocTemporary();
+    AddLine("MOVC.F32 RC.x, {};", Visit(operation[0]));
+    AddLine("MOVC.F32 RC.y, {};", Visit(operation[1]));
+    AddLine("MOV.S {}, 0;", temporary);
+    AddLine("MOV.S {} (NAN.x), -1;", temporary);
+    AddLine("MOV.S {} (NAN.y), -1;", temporary);
+    return temporary;
+}
+
+std::string ARBDecompiler::LogicalAddCarry(Operation operation) {
+    std::string temporary = AllocTemporary();
+    AddLine("ADDC.U RC, {}, {};", Visit(operation[0]), Visit(operation[1]));
+    AddLine("MOV.S {}, 0;", temporary);
+    AddLine("IF CF.x;");
+    AddLine("MOV.S {}, -1;", temporary);
+    AddLine("ENDIF;");
+    return temporary;
+}
+
+std::string ARBDecompiler::Texture(Operation operation) {
+    const auto& meta = std::get<MetaTexture>(operation.GetMeta());
+    const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index;
+    const auto [temporary, swizzle] = BuildCoords(operation);
+
+    std::string_view opcode = "TEX";
+    std::string extra;
+    if (meta.bias) {
+        ASSERT(!meta.lod);
+        opcode = "TXB";
+
+        if (swizzle < 4) {
+            AddLine("MOV.F {}.w, {};", temporary, Visit(meta.bias));
+        } else {
+            const std::string bias = AllocTemporary();
+            AddLine("MOV.F {}, {};", bias, Visit(meta.bias));
+            extra = fmt::format(" {},", bias);
+        }
+    }
+    if (meta.lod) {
+        ASSERT(!meta.bias);
+        opcode = "TXL";
+
+        if (swizzle < 4) {
+            AddLine("MOV.F {}.w, {};", temporary, Visit(meta.lod));
+        } else {
+            const std::string lod = AllocTemporary();
+            AddLine("MOV.F {}, {};", lod, Visit(meta.lod));
+            extra = fmt::format(" {},", lod);
+        }
+    }
+
+    AddLine("{}.F {}, {},{} texture[{}], {}{};", opcode, temporary, temporary, extra, sampler_id,
+            TextureType(meta), BuildAoffi(operation));
+    AddLine("MOV.U {}.x, {}.{};", temporary, temporary, Swizzle(meta.element));
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::TextureGather(Operation operation) {
+    const auto& meta = std::get<MetaTexture>(operation.GetMeta());
+    const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index;
+    const auto [temporary, swizzle] = BuildCoords(operation);
+
+    std::string comp;
+    if (!meta.sampler.is_shadow) {
+        const auto& immediate = std::get<ImmediateNode>(*meta.component);
+        comp = fmt::format(".{}", Swizzle(immediate.GetValue()));
+    }
+
+    AddLine("TXG.F {}, {}, texture[{}]{}, {}{};", temporary, temporary, sampler_id, comp,
+            TextureType(meta), BuildAoffi(operation));
+    AddLine("MOV.U {}.x, {}.{};", temporary, temporary, Swizzle(meta.element));
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::TextureQueryDimensions(Operation operation) {
+    const auto& meta = std::get<MetaTexture>(operation.GetMeta());
+    const std::string temporary = AllocVectorTemporary();
+    const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index;
+
+    ASSERT(!meta.sampler.is_array);
+
+    const std::string lod = operation.GetOperandsCount() > 0 ? Visit(operation[0]) : "0";
+    AddLine("TXQ {}, {}, texture[{}], {};", temporary, lod, sampler_id, TextureType(meta));
+    AddLine("MOV.U {}.x, {}.{};", temporary, temporary, Swizzle(meta.element));
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::TextureQueryLod(Operation operation) {
+    const auto& meta = std::get<MetaTexture>(operation.GetMeta());
+    const std::string temporary = AllocVectorTemporary();
+    const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index;
+
+    ASSERT(!meta.sampler.is_array);
+
+    const std::size_t count = operation.GetOperandsCount();
+    for (std::size_t i = 0; i < count; ++i) {
+        AddLine("MOV.F {}.{}, {};", temporary, Swizzle(i), Visit(operation[i]));
+    }
+    AddLine("LOD.F {}, {}, texture[{}], {};", temporary, temporary, sampler_id, TextureType(meta));
+    AddLine("MUL.F32 {}, {}, {{256, 256, 0, 0}};", temporary, temporary);
+    AddLine("TRUNC.S {}, {};", temporary, temporary);
+    AddLine("MOV.U {}.x, {}.{};", temporary, temporary, Swizzle(meta.element));
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::TexelFetch(Operation operation) {
+    const auto& meta = std::get<MetaTexture>(operation.GetMeta());
+    const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index;
+    const auto [temporary, swizzle] = BuildCoords(operation);
+
+    if (!meta.sampler.is_buffer) {
+        ASSERT(swizzle < 4);
+        AddLine("MOV.F {}.w, {};", temporary, Visit(meta.lod));
+    }
+    AddLine("TXF.F {}, {}, texture[{}], {}{};", temporary, temporary, sampler_id, TextureType(meta),
+            BuildAoffi(operation));
+    AddLine("MOV.U {}.x, {}.{};", temporary, temporary, Swizzle(meta.element));
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::TextureGradient(Operation operation) {
+    const auto& meta = std::get<MetaTexture>(operation.GetMeta());
+    const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index;
+    const std::string ddx = AllocVectorTemporary();
+    const std::string ddy = AllocVectorTemporary();
+    const std::string coord = BuildCoords(operation).first;
+
+    const std::size_t num_components = meta.derivates.size() / 2;
+    for (std::size_t index = 0; index < num_components; ++index) {
+        const char swizzle = Swizzle(index);
+        AddLine("MOV.F {}.{}, {};", ddx, swizzle, Visit(meta.derivates[index * 2]));
+        AddLine("MOV.F {}.{}, {};", ddy, swizzle, Visit(meta.derivates[index * 2 + 1]));
+    }
+
+    const std::string_view result = coord;
+    AddLine("TXD.F {}, {}, {}, {}, texture[{}], {}{};", result, coord, ddx, ddy, sampler_id,
+            TextureType(meta), BuildAoffi(operation));
+    AddLine("MOV.F {}.x, {}.{};", result, result, Swizzle(meta.element));
+    return fmt::format("{}.x", result);
+}
+
+std::string ARBDecompiler::ImageLoad(Operation operation) {
+    const auto& meta = std::get<MetaImage>(operation.GetMeta());
+    const u32 image_id = device.GetBaseBindings(stage).image + meta.image.index;
+    const std::size_t count = operation.GetOperandsCount();
+    const std::string_view type = ImageType(meta.image.type);
+
+    const std::string temporary = AllocVectorTemporary();
+    for (std::size_t i = 0; i < count; ++i) {
+        AddLine("MOV.S {}.{}, {};", temporary, Swizzle(i), Visit(operation[i]));
+    }
+    AddLine("LOADIM.F {}, {}, image[{}], {};", temporary, temporary, image_id, type);
+    AddLine("MOV.F {}.x, {}.{};", temporary, temporary, Swizzle(meta.element));
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::ImageStore(Operation operation) {
+    const auto& meta = std::get<MetaImage>(operation.GetMeta());
+    const u32 image_id = device.GetBaseBindings(stage).image + meta.image.index;
+    const std::size_t num_coords = operation.GetOperandsCount();
+    const std::size_t num_values = meta.values.size();
+    const std::string_view type = ImageType(meta.image.type);
+
+    const std::string coord = AllocVectorTemporary();
+    const std::string value = AllocVectorTemporary();
+    for (std::size_t i = 0; i < num_coords; ++i) {
+        AddLine("MOV.S {}.{}, {};", coord, Swizzle(i), Visit(operation[i]));
+    }
+    for (std::size_t i = 0; i < num_values; ++i) {
+        AddLine("MOV.F {}.{}, {};", value, Swizzle(i), Visit(meta.values[i]));
+    }
+    AddLine("STOREIM.F image[{}], {}, {}, {};", image_id, value, coord, type);
+    return {};
+}
+
+std::string ARBDecompiler::Branch(Operation operation) {
+    const auto target = std::get<ImmediateNode>(*operation[0]);
+    AddLine("MOV.U PC.x, {};", target.GetValue());
+    AddLine("CONT;");
+    return {};
+}
+
+std::string ARBDecompiler::BranchIndirect(Operation operation) {
+    AddLine("MOV.U PC.x, {};", Visit(operation[0]));
+    AddLine("CONT;");
+    return {};
+}
+
+std::string ARBDecompiler::PushFlowStack(Operation operation) {
+    const auto stack = std::get<MetaStackClass>(operation.GetMeta());
+    const u32 target = std::get<ImmediateNode>(*operation[0]).GetValue();
+    const std::string_view stack_name = StackName(stack);
+    AddLine("MOV.U {}[{}_TOP.x].x, {};", stack_name, stack_name, target);
+    AddLine("ADD.S {}_TOP.x, {}_TOP.x, 1;", stack_name, stack_name);
+    return {};
+}
+
+std::string ARBDecompiler::PopFlowStack(Operation operation) {
+    const auto stack = std::get<MetaStackClass>(operation.GetMeta());
+    const std::string_view stack_name = StackName(stack);
+    AddLine("SUB.S {}_TOP.x, {}_TOP.x, 1;", stack_name, stack_name);
+    AddLine("MOV.U PC.x, {}[{}_TOP.x].x;", stack_name, stack_name);
+    AddLine("CONT;");
+    return {};
+}
+
+std::string ARBDecompiler::Exit(Operation) {
+    Exit();
+    return {};
+}
+
+std::string ARBDecompiler::Discard(Operation) {
+    AddLine("KIL TR;");
+    return {};
+}
+
+std::string ARBDecompiler::EmitVertex(Operation) {
+    AddLine("EMIT;");
+    return {};
+}
+
+std::string ARBDecompiler::EndPrimitive(Operation) {
+    AddLine("ENDPRIM;");
+    return {};
+}
+
+std::string ARBDecompiler::InvocationId(Operation) {
+    return "primitive.invocation";
+}
+
+std::string ARBDecompiler::YNegate(Operation) {
+    LOG_WARNING(Render_OpenGL, "(STUBBED)");
+    const std::string temporary = AllocTemporary();
+    AddLine("MOV.F {}, 1;", temporary);
+    return temporary;
+}
+
+std::string ARBDecompiler::ThreadId(Operation) {
+    return fmt::format("{}.threadid", StageInputName(stage));
+}
+
+std::string ARBDecompiler::ShuffleIndexed(Operation operation) {
+    if (!device.HasWarpIntrinsics()) {
+        LOG_ERROR(Render_OpenGL,
+                  "NV_shader_thread_shuffle is missing. Kepler or better is required.");
+        return Visit(operation[0]);
+    }
+    const std::string temporary = AllocVectorTemporary();
+    AddLine("SHFIDX.U {}, {}, {}, {{31, 0, 0, 0}};", temporary, Visit(operation[0]),
+            Visit(operation[1]));
+    AddLine("MOV.U {}.x, {}.y;", temporary, temporary);
+    return fmt::format("{}.x", temporary);
+}
+
+std::string ARBDecompiler::Barrier(Operation) {
+    if (!ir.IsDecompiled()) {
+        LOG_ERROR(Render_OpenGL, "BAR used but shader is not decompiled");
+        return {};
+    }
+    AddLine("BAR;");
+    return {};
+}
+
+std::string ARBDecompiler::MemoryBarrierGroup(Operation) {
+    AddLine("MEMBAR.CTA;");
+    return {};
+}
+
+std::string ARBDecompiler::MemoryBarrierGlobal(Operation) {
+    AddLine("MEMBAR;");
+    return {};
+}
+
+} // Anonymous namespace
+
+std::string DecompileAssemblyShader(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
+                                    const VideoCommon::Shader::Registry& registry,
+                                    Tegra::Engines::ShaderType stage, std::string_view identifier) {
+    return ARBDecompiler(device, ir, registry, stage, identifier).Code();
+}
+
+} // namespace OpenGL

src/video_core/renderer_opengl/gl_arb_decompiler.h

@@ -0,0 +1,29 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <string>
+#include <string_view>
+
+#include "common/common_types.h"
+
+namespace Tegra::Engines {
+enum class ShaderType : u32;
+}
+
+namespace VideoCommon::Shader {
+class ShaderIR;
+class Registry;
+} // namespace VideoCommon::Shader
+
+namespace OpenGL {
+
+class Device;
+
+std::string DecompileAssemblyShader(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
+                                    const VideoCommon::Shader::Registry& registry,
+                                    Tegra::Engines::ShaderType stage, std::string_view identifier);
+
+} // namespace OpenGL

src/video_core/renderer_opengl/gl_buffer_cache.cpp

@@ -22,22 +22,46 @@ using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 
 MICROPROFILE_DEFINE(OpenGL_Buffer_Download, "OpenGL", "Buffer Download", MP_RGB(192, 192, 128));
 
-CachedBufferBlock::CachedBufferBlock(VAddr cpu_addr, const std::size_t size)
+Buffer::Buffer(const Device& device, VAddr cpu_addr, std::size_t size)
     : VideoCommon::BufferBlock{cpu_addr, size} {
     gl_buffer.Create();
     glNamedBufferData(gl_buffer.handle, static_cast<GLsizeiptr>(size), nullptr, GL_DYNAMIC_DRAW);
+    if (device.HasVertexBufferUnifiedMemory()) {
+        glMakeNamedBufferResidentNV(gl_buffer.handle, GL_READ_WRITE);
+        glGetNamedBufferParameterui64vNV(gl_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV, &gpu_address);
+    }
 }
 
-CachedBufferBlock::~CachedBufferBlock() = default;
+Buffer::~Buffer() = default;
+
+void Buffer::Upload(std::size_t offset, std::size_t size, const u8* data) const {
+    glNamedBufferSubData(Handle(), static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size),
+                         data);
+}
+
+void Buffer::Download(std::size_t offset, std::size_t size, u8* data) const {
+    MICROPROFILE_SCOPE(OpenGL_Buffer_Download);
+    glMemoryBarrier(GL_BUFFER_UPDATE_BARRIER_BIT);
+    glGetNamedBufferSubData(Handle(), static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size),
+                            data);
+}
+
+void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset,
+                      std::size_t size) const {
+    glCopyNamedBufferSubData(src.Handle(), Handle(), static_cast<GLintptr>(src_offset),
+                             static_cast<GLintptr>(dst_offset), static_cast<GLsizeiptr>(size));
+}
 
 OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system,
-                               const Device& device, std::size_t stream_size)
-    : GenericBufferCache{rasterizer, system, std::make_unique<OGLStreamBuffer>(stream_size, true)} {
+                               const Device& device_, std::size_t stream_size)
+    : GenericBufferCache{rasterizer, system,
+                         std::make_unique<OGLStreamBuffer>(device_, stream_size, true)},
+      device{device_} {
     if (!device.HasFastBufferSubData()) {
         return;
     }
 
-    static constexpr auto size = static_cast<GLsizeiptr>(Maxwell::MaxConstBufferSize);
+    static constexpr GLsizeiptr size = static_cast<GLsizeiptr>(Maxwell::MaxConstBufferSize);
     glCreateBuffers(static_cast<GLsizei>(std::size(cbufs)), std::data(cbufs));
     for (const GLuint cbuf : cbufs) {
         glNamedBufferData(cbuf, size, nullptr, GL_STREAM_DRAW);
@@ -48,44 +72,21 @@ OGLBufferCache::~OGLBufferCache() {
     glDeleteBuffers(static_cast<GLsizei>(std::size(cbufs)), std::data(cbufs));
 }
 
-Buffer OGLBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) {
-    return std::make_shared<CachedBufferBlock>(cpu_addr, size);
-}
-
-GLuint OGLBufferCache::ToHandle(const Buffer& buffer) {
-    return buffer->GetHandle();
-}
-
-GLuint OGLBufferCache::GetEmptyBuffer(std::size_t) {
-    return 0;
+std::shared_ptr<Buffer> OGLBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) {
+    return std::make_shared<Buffer>(device, cpu_addr, size);
 }
 
-void OGLBufferCache::UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
-                                     const u8* data) {
-    glNamedBufferSubData(buffer->GetHandle(), static_cast<GLintptr>(offset),
-                         static_cast<GLsizeiptr>(size), data);
-}
-
-void OGLBufferCache::DownloadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
-                                       u8* data) {
-    MICROPROFILE_SCOPE(OpenGL_Buffer_Download);
-    glMemoryBarrier(GL_BUFFER_UPDATE_BARRIER_BIT);
-    glGetNamedBufferSubData(buffer->GetHandle(), static_cast<GLintptr>(offset),
-                            static_cast<GLsizeiptr>(size), data);
-}
-
-void OGLBufferCache::CopyBlock(const Buffer& src, const Buffer& dst, std::size_t src_offset,
-                               std::size_t dst_offset, std::size_t size) {
-    glCopyNamedBufferSubData(src->GetHandle(), dst->GetHandle(), static_cast<GLintptr>(src_offset),
-                             static_cast<GLintptr>(dst_offset), static_cast<GLsizeiptr>(size));
+OGLBufferCache::BufferInfo OGLBufferCache::GetEmptyBuffer(std::size_t) {
+    return {0, 0, 0};
 }
 
 OGLBufferCache::BufferInfo OGLBufferCache::ConstBufferUpload(const void* raw_pointer,
                                                              std::size_t size) {
     DEBUG_ASSERT(cbuf_cursor < std::size(cbufs));
-    const GLuint& cbuf = cbufs[cbuf_cursor++];
+    const GLuint cbuf = cbufs[cbuf_cursor++];
+
     glNamedBufferSubData(cbuf, 0, static_cast<GLsizeiptr>(size), raw_pointer);
-    return {cbuf, 0};
+    return {cbuf, 0, 0};
 }
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_buffer_cache.h

@@ -10,7 +10,6 @@
 #include "common/common_types.h"
 #include "video_core/buffer_cache/buffer_cache.h"
 #include "video_core/engines/maxwell_3d.h"
-#include "video_core/rasterizer_cache.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 #include "video_core/renderer_opengl/gl_stream_buffer.h"
 
@@ -24,57 +23,57 @@ class Device;
 class OGLStreamBuffer;
 class RasterizerOpenGL;
 
-class CachedBufferBlock;
+class Buffer : public VideoCommon::BufferBlock {
+public:
+    explicit Buffer(const Device& device, VAddr cpu_addr, std::size_t size);
+    ~Buffer();
 
-using Buffer = std::shared_ptr<CachedBufferBlock>;
-using GenericBufferCache = VideoCommon::BufferCache<Buffer, GLuint, OGLStreamBuffer>;
+    void Upload(std::size_t offset, std::size_t size, const u8* data) const;
 
-class CachedBufferBlock : public VideoCommon::BufferBlock {
-public:
-    explicit CachedBufferBlock(VAddr cpu_addr, const std::size_t size);
-    ~CachedBufferBlock();
+    void Download(std::size_t offset, std::size_t size, u8* data) const;
+
+    void CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset,
+                  std::size_t size) const;
 
-    GLuint GetHandle() const {
+    GLuint Handle() const noexcept {
         return gl_buffer.handle;
     }
 
+    u64 Address() const noexcept {
+        return gpu_address;
+    }
+
 private:
     OGLBuffer gl_buffer;
+    u64 gpu_address = 0;
 };
 
+using GenericBufferCache = VideoCommon::BufferCache<Buffer, GLuint, OGLStreamBuffer>;
 class OGLBufferCache final : public GenericBufferCache {
 public:
     explicit OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system,
                             const Device& device, std::size_t stream_size);
     ~OGLBufferCache();
 
-    GLuint GetEmptyBuffer(std::size_t) override;
+    BufferInfo GetEmptyBuffer(std::size_t) override;
 
     void Acquire() noexcept {
         cbuf_cursor = 0;
     }
 
 protected:
-    Buffer CreateBlock(VAddr cpu_addr, std::size_t size) override;
-
-    GLuint ToHandle(const Buffer& buffer) override;
-
-    void UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
-                         const u8* data) override;
-
-    void DownloadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
-                           u8* data) override;
-
-    void CopyBlock(const Buffer& src, const Buffer& dst, std::size_t src_offset,
-                   std::size_t dst_offset, std::size_t size) override;
+    std::shared_ptr<Buffer> CreateBlock(VAddr cpu_addr, std::size_t size) override;
 
     BufferInfo ConstBufferUpload(const void* raw_pointer, std::size_t size) override;
 
 private:
+    static constexpr std::size_t NUM_CBUFS = Tegra::Engines::Maxwell3D::Regs::MaxConstBuffers *
+                                             Tegra::Engines::Maxwell3D::Regs::MaxShaderProgram;
+
+    const Device& device;
+
     std::size_t cbuf_cursor = 0;
-    std::array<GLuint, Tegra::Engines::Maxwell3D::Regs::MaxConstBuffers *
-                           Tegra::Engines::Maxwell3D::Regs::MaxShaderProgram>
-        cbufs;
+    std::array<GLuint, NUM_CBUFS> cbufs{};
 };
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_device.cpp

@@ -6,6 +6,7 @@
 #include <array>
 #include <cstddef>
 #include <cstring>
+#include <limits>
 #include <optional>
 #include <vector>
 
@@ -26,24 +27,27 @@ constexpr u32 ReservedUniformBlocks = 1;
 
 constexpr u32 NumStages = 5;
 
-constexpr std::array LimitUBOs = {GL_MAX_VERTEX_UNIFORM_BLOCKS, GL_MAX_TESS_CONTROL_UNIFORM_BLOCKS,
-                                  GL_MAX_TESS_EVALUATION_UNIFORM_BLOCKS,
-                                  GL_MAX_GEOMETRY_UNIFORM_BLOCKS, GL_MAX_FRAGMENT_UNIFORM_BLOCKS};
+constexpr std::array LimitUBOs = {
+    GL_MAX_VERTEX_UNIFORM_BLOCKS,          GL_MAX_TESS_CONTROL_UNIFORM_BLOCKS,
+    GL_MAX_TESS_EVALUATION_UNIFORM_BLOCKS, GL_MAX_GEOMETRY_UNIFORM_BLOCKS,
+    GL_MAX_FRAGMENT_UNIFORM_BLOCKS,        GL_MAX_COMPUTE_UNIFORM_BLOCKS};
 
 constexpr std::array LimitSSBOs = {
-    GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS, GL_MAX_TESS_CONTROL_SHADER_STORAGE_BLOCKS,
+    GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS,          GL_MAX_TESS_CONTROL_SHADER_STORAGE_BLOCKS,
     GL_MAX_TESS_EVALUATION_SHADER_STORAGE_BLOCKS, GL_MAX_GEOMETRY_SHADER_STORAGE_BLOCKS,
-    GL_MAX_FRAGMENT_SHADER_STORAGE_BLOCKS};
+    GL_MAX_FRAGMENT_SHADER_STORAGE_BLOCKS,        GL_MAX_COMPUTE_SHADER_STORAGE_BLOCKS};
 
-constexpr std::array LimitSamplers = {
-    GL_MAX_VERTEX_TEXTURE_IMAGE_UNITS, GL_MAX_TESS_CONTROL_TEXTURE_IMAGE_UNITS,
-    GL_MAX_TESS_EVALUATION_TEXTURE_IMAGE_UNITS, GL_MAX_GEOMETRY_TEXTURE_IMAGE_UNITS,
-    GL_MAX_TEXTURE_IMAGE_UNITS};
+constexpr std::array LimitSamplers = {GL_MAX_VERTEX_TEXTURE_IMAGE_UNITS,
+                                      GL_MAX_TESS_CONTROL_TEXTURE_IMAGE_UNITS,
+                                      GL_MAX_TESS_EVALUATION_TEXTURE_IMAGE_UNITS,
+                                      GL_MAX_GEOMETRY_TEXTURE_IMAGE_UNITS,
+                                      GL_MAX_TEXTURE_IMAGE_UNITS,
+                                      GL_MAX_COMPUTE_TEXTURE_IMAGE_UNITS};
 
-constexpr std::array LimitImages = {GL_MAX_VERTEX_IMAGE_UNIFORMS,
-                                    GL_MAX_TESS_CONTROL_IMAGE_UNIFORMS,
-                                    GL_MAX_TESS_EVALUATION_IMAGE_UNIFORMS,
-                                    GL_MAX_GEOMETRY_IMAGE_UNIFORMS, GL_MAX_FRAGMENT_IMAGE_UNIFORMS};
+constexpr std::array LimitImages = {
+    GL_MAX_VERTEX_IMAGE_UNIFORMS,          GL_MAX_TESS_CONTROL_IMAGE_UNIFORMS,
+    GL_MAX_TESS_EVALUATION_IMAGE_UNIFORMS, GL_MAX_GEOMETRY_IMAGE_UNIFORMS,
+    GL_MAX_FRAGMENT_IMAGE_UNIFORMS,        GL_MAX_COMPUTE_IMAGE_UNIFORMS};
 
 template <typename T>
 T GetInteger(GLenum pname) {
@@ -85,6 +89,13 @@ u32 Extract(u32& base, u32& num, u32 amount, std::optional<GLenum> limit = {}) {
     return std::exchange(base, base + amount);
 }
 
+std::array<u32, Tegra::Engines::MaxShaderTypes> BuildMaxUniformBuffers() noexcept {
+    std::array<u32, Tegra::Engines::MaxShaderTypes> max;
+    std::transform(LimitUBOs.begin(), LimitUBOs.end(), max.begin(),
+                   [](GLenum pname) { return GetInteger<u32>(pname); });
+    return max;
+}
+
 std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> BuildBaseBindings() noexcept {
     std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> bindings;
 
@@ -112,16 +123,24 @@ std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> BuildBaseBindin
     u32 num_images = GetInteger<u32>(GL_MAX_IMAGE_UNITS);
     u32 base_images = 0;
 
-    // Reserve more image bindings on fragment and vertex stages.
+    // GL_MAX_IMAGE_UNITS is guaranteed by the spec to have a minimum value of 8.
+    // Due to the limitation of GL_MAX_IMAGE_UNITS, reserve at least 4 image bindings on the
+    // fragment stage, and at least 1 for the rest of the stages.
+    // So far games are observed to use 1 image binding on vertex and 4 on fragment stages.
+
+    // Reserve at least 4 image bindings on the fragment stage.
     bindings[4].image =
-        Extract(base_images, num_images, num_images / NumStages + 2, LimitImages[4]);
-    bindings[0].image =
-        Extract(base_images, num_images, num_images / NumStages + 1, LimitImages[0]);
+        Extract(base_images, num_images, std::max(4U, num_images / NumStages), LimitImages[4]);
+
+    // This is guaranteed to be at least 1.
+    const u32 total_extracted_images = num_images / (NumStages - 1);
 
     // Reserve the other image bindings.
-    const u32 total_extracted_images = num_images / (NumStages - 2);
-    for (std::size_t i = 2; i < NumStages; ++i) {
+    for (std::size_t i = 0; i < NumStages; ++i) {
         const std::size_t stage = stage_swizzle[i];
+        if (stage == 4) {
+            continue;
+        }
         bindings[stage].image =
             Extract(base_images, num_images, total_extracted_images, LimitImages[stage]);
     }
@@ -133,6 +152,7 @@ std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> BuildBaseBindin
 }
 
 bool IsASTCSupported() {
+    static constexpr std::array targets = {GL_TEXTURE_2D, GL_TEXTURE_2D_ARRAY};
     static constexpr std::array formats = {
         GL_COMPRESSED_RGBA_ASTC_4x4_KHR,           GL_COMPRESSED_RGBA_ASTC_5x4_KHR,
         GL_COMPRESSED_RGBA_ASTC_5x5_KHR,           GL_COMPRESSED_RGBA_ASTC_6x5_KHR,
@@ -149,25 +169,59 @@ bool IsASTCSupported() {
         GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x8_KHR,  GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x10_KHR,
         GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x10_KHR, GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x12_KHR,
     };
-    return std::find_if_not(formats.begin(), formats.end(), [](GLenum format) {
-               GLint supported;
-               glGetInternalformativ(GL_TEXTURE_2D, format, GL_INTERNALFORMAT_SUPPORTED, 1,
-                                     &supported);
-               return supported == GL_TRUE;
-           }) == formats.end();
+    static constexpr std::array required_support = {
+        GL_VERTEX_TEXTURE,   GL_TESS_CONTROL_TEXTURE, GL_TESS_EVALUATION_TEXTURE,
+        GL_GEOMETRY_TEXTURE, GL_FRAGMENT_TEXTURE,     GL_COMPUTE_TEXTURE,
+    };
+
+    for (const GLenum target : targets) {
+        for (const GLenum format : formats) {
+            for (const GLenum support : required_support) {
+                GLint value;
+                glGetInternalformativ(target, format, support, 1, &value);
+                if (value != GL_FULL_SUPPORT) {
+                    return false;
+                }
+            }
+        }
+    }
+    return true;
+}
+
+/// @brief Returns true when a GL_RENDERER is a Turing GPU
+/// @param renderer GL_RENDERER string
+bool IsTuring(std::string_view renderer) {
+    static constexpr std::array<std::string_view, 12> TURING_GPUS = {
+        "GTX 1650",        "GTX 1660",        "RTX 2060",        "RTX 2070",
+        "RTX 2080",        "TITAN RTX",       "Quadro RTX 3000", "Quadro RTX 4000",
+        "Quadro RTX 5000", "Quadro RTX 6000", "Quadro RTX 8000", "Tesla T4",
+    };
+    return std::any_of(TURING_GPUS.begin(), TURING_GPUS.end(),
+                       [renderer](std::string_view candidate) {
+                           return renderer.find(candidate) != std::string_view::npos;
+                       });
 }
 
 } // Anonymous namespace
 
-Device::Device() : base_bindings{BuildBaseBindings()} {
+Device::Device()
+    : max_uniform_buffers{BuildMaxUniformBuffers()}, base_bindings{BuildBaseBindings()} {
     const std::string_view vendor = reinterpret_cast<const char*>(glGetString(GL_VENDOR));
-    const auto renderer = reinterpret_cast<const char*>(glGetString(GL_RENDERER));
+    const std::string_view renderer = reinterpret_cast<const char*>(glGetString(GL_RENDERER));
+    const std::string_view version = reinterpret_cast<const char*>(glGetString(GL_VERSION));
     const std::vector extensions = GetExtensions();
 
     const bool is_nvidia = vendor == "NVIDIA Corporation";
     const bool is_amd = vendor == "ATI Technologies Inc.";
-    const bool is_intel = vendor == "Intel";
-    const bool is_intel_proprietary = is_intel && std::strstr(renderer, "Mesa") == nullptr;
+    const bool is_turing = is_nvidia && IsTuring(renderer);
+
+    bool disable_fast_buffer_sub_data = false;
+    if (is_nvidia && version == "4.6.0 NVIDIA 443.24") {
+        LOG_WARNING(
+            Render_OpenGL,
+            "Beta driver 443.24 is known to have issues. There might be performance issues.");
+        disable_fast_buffer_sub_data = true;
+    }
 
     uniform_buffer_alignment = GetInteger<std::size_t>(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT);
     shader_storage_alignment = GetInteger<std::size_t>(GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT);
@@ -178,14 +232,24 @@ Device::Device() : base_bindings{BuildBaseBindings()} {
     has_shader_ballot = GLAD_GL_ARB_shader_ballot;
     has_vertex_viewport_layer = GLAD_GL_ARB_shader_viewport_layer_array;
     has_image_load_formatted = HasExtension(extensions, "GL_EXT_shader_image_load_formatted");
+    has_texture_shadow_lod = HasExtension(extensions, "GL_EXT_texture_shadow_lod");
     has_astc = IsASTCSupported();
     has_variable_aoffi = TestVariableAoffi();
     has_component_indexing_bug = is_amd;
     has_precise_bug = TestPreciseBug();
-    has_broken_compute = is_intel_proprietary;
-    has_fast_buffer_sub_data = is_nvidia;
+    has_nv_viewport_array2 = GLAD_GL_NV_viewport_array2;
+
+    // At the moment of writing this, only Nvidia's driver optimizes BufferSubData on exclusive
+    // uniform buffers as "push constants"
+    has_fast_buffer_sub_data = is_nvidia && !disable_fast_buffer_sub_data;
+
+    // Nvidia's driver on Turing GPUs randomly crashes when the buffer is made resident, or on
+    // DeleteBuffers. Disable unified memory on these devices.
+    has_vertex_buffer_unified_memory = GLAD_GL_NV_vertex_buffer_unified_memory && !is_turing;
+
     use_assembly_shaders = Settings::values.use_assembly_shaders && GLAD_GL_NV_gpu_program5 &&
-                           GLAD_GL_NV_compute_program5;
+                           GLAD_GL_NV_compute_program5 && GLAD_GL_NV_transform_feedback &&
+                           GLAD_GL_NV_transform_feedback2;
 
     LOG_INFO(Render_OpenGL, "Renderer_VariableAOFFI: {}", has_variable_aoffi);
     LOG_INFO(Render_OpenGL, "Renderer_ComponentIndexingBug: {}", has_component_indexing_bug);
@@ -197,17 +261,17 @@ Device::Device() : base_bindings{BuildBaseBindings()} {
 }
 
 Device::Device(std::nullptr_t) {
-    uniform_buffer_alignment = 0;
+    max_uniform_buffers.fill(std::numeric_limits<u32>::max());
+    uniform_buffer_alignment = 4;
+    shader_storage_alignment = 4;
     max_vertex_attributes = 16;
     max_varyings = 15;
     has_warp_intrinsics = true;
     has_shader_ballot = true;
     has_vertex_viewport_layer = true;
     has_image_load_formatted = true;
+    has_texture_shadow_lod = true;
     has_variable_aoffi = true;
-    has_component_indexing_bug = false;
-    has_broken_compute = false;
-    has_precise_bug = false;
 }
 
 bool Device::TestVariableAoffi() {

src/video_core/renderer_opengl/gl_device.h

@@ -24,6 +24,10 @@ public:
     explicit Device();
     explicit Device(std::nullptr_t);
 
+    u32 GetMaxUniformBuffers(Tegra::Engines::ShaderType shader_type) const noexcept {
+        return max_uniform_buffers[static_cast<std::size_t>(shader_type)];
+    }
+
     const BaseBindings& GetBaseBindings(std::size_t stage_index) const noexcept {
         return base_bindings[stage_index];
     }
@@ -64,6 +68,14 @@ public:
         return has_image_load_formatted;
     }
 
+    bool HasTextureShadowLod() const {
+        return has_texture_shadow_lod;
+    }
+
+    bool HasVertexBufferUnifiedMemory() const {
+        return has_vertex_buffer_unified_memory;
+    }
+
     bool HasASTC() const {
         return has_astc;
     }
@@ -80,14 +92,14 @@ public:
         return has_precise_bug;
     }
 
-    bool HasBrokenCompute() const {
-        return has_broken_compute;
-    }
-
     bool HasFastBufferSubData() const {
         return has_fast_buffer_sub_data;
     }
 
+    bool HasNvViewportArray2() const {
+        return has_nv_viewport_array2;
+    }
+
     bool UseAssemblyShaders() const {
         return use_assembly_shaders;
     }
@@ -96,7 +108,8 @@ private:
     static bool TestVariableAoffi();
     static bool TestPreciseBug();
 
-    std::array<BaseBindings, Tegra::Engines::MaxShaderTypes> base_bindings;
+    std::array<u32, Tegra::Engines::MaxShaderTypes> max_uniform_buffers{};
+    std::array<BaseBindings, Tegra::Engines::MaxShaderTypes> base_bindings{};
     std::size_t uniform_buffer_alignment{};
     std::size_t shader_storage_alignment{};
     u32 max_vertex_attributes{};
@@ -105,12 +118,14 @@ private:
     bool has_shader_ballot{};
     bool has_vertex_viewport_layer{};
     bool has_image_load_formatted{};
+    bool has_texture_shadow_lod{};
+    bool has_vertex_buffer_unified_memory{};
     bool has_astc{};
     bool has_variable_aoffi{};
     bool has_component_indexing_bug{};
     bool has_precise_bug{};
-    bool has_broken_compute{};
     bool has_fast_buffer_sub_data{};
+    bool has_nv_viewport_array2{};
     bool use_assembly_shaders{};
 };
 

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -30,6 +30,7 @@
 #include "video_core/renderer_opengl/gl_shader_cache.h"
 #include "video_core/renderer_opengl/maxwell_to_gl.h"
 #include "video_core/renderer_opengl/renderer_opengl.h"
+#include "video_core/shader_cache.h"
 
 namespace OpenGL {
 
@@ -54,15 +55,34 @@ MICROPROFILE_DEFINE(OpenGL_PrimitiveAssembly, "OpenGL", "Prim Asmbl", MP_RGB(255
 
 namespace {
 
-constexpr std::size_t NumSupportedVertexAttributes = 16;
+constexpr std::size_t NUM_CONST_BUFFERS_PER_STAGE = 18;
+constexpr std::size_t NUM_CONST_BUFFERS_BYTES_PER_STAGE =
+    NUM_CONST_BUFFERS_PER_STAGE * Maxwell::MaxConstBufferSize;
+constexpr std::size_t TOTAL_CONST_BUFFER_BYTES =
+    NUM_CONST_BUFFERS_BYTES_PER_STAGE * Maxwell::MaxShaderStage;
+
+constexpr std::size_t NUM_SUPPORTED_VERTEX_ATTRIBUTES = 16;
+constexpr std::size_t NUM_SUPPORTED_VERTEX_BINDINGS = 16;
 
 template <typename Engine, typename Entry>
 Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry& entry,
                                                ShaderType shader_type, std::size_t index = 0) {
+    if constexpr (std::is_same_v<Entry, SamplerEntry>) {
+        if (entry.is_separated) {
+            const u32 buffer_1 = entry.buffer;
+            const u32 buffer_2 = entry.secondary_buffer;
+            const u32 offset_1 = entry.offset;
+            const u32 offset_2 = entry.secondary_offset;
+            const u32 handle_1 = engine.AccessConstBuffer32(shader_type, buffer_1, offset_1);
+            const u32 handle_2 = engine.AccessConstBuffer32(shader_type, buffer_2, offset_2);
+            return engine.GetTextureInfo(handle_1 | handle_2);
+        }
+    }
     if (entry.is_bindless) {
-        const auto tex_handle = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset);
-        return engine.GetTextureInfo(tex_handle);
+        const u32 handle = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset);
+        return engine.GetTextureInfo(handle);
     }
+
     const auto& gpu_profile = engine.AccessGuestDriverProfile();
     const u32 offset = entry.offset + static_cast<u32>(index * gpu_profile.GetTextureHandlerSize());
     if constexpr (std::is_same_v<Engine, Tegra::Engines::Maxwell3D>) {
@@ -87,6 +107,34 @@ std::size_t GetConstBufferSize(const Tegra::Engines::ConstBufferInfo& buffer,
     return buffer.size;
 }
 
+/// Translates hardware transform feedback indices
+/// @param location Hardware location
+/// @return Pair of ARB_transform_feedback3 token stream first and third arguments
+/// @note Read https://www.khronos.org/registry/OpenGL/extensions/ARB/ARB_transform_feedback3.txt
+std::pair<GLint, GLint> TransformFeedbackEnum(u8 location) {
+    const u8 index = location / 4;
+    if (index >= 8 && index <= 39) {
+        return {GL_GENERIC_ATTRIB_NV, index - 8};
+    }
+    if (index >= 48 && index <= 55) {
+        return {GL_TEXTURE_COORD_NV, index - 48};
+    }
+    switch (index) {
+    case 7:
+        return {GL_POSITION, 0};
+    case 40:
+        return {GL_PRIMARY_COLOR_NV, 0};
+    case 41:
+        return {GL_SECONDARY_COLOR_NV, 0};
+    case 42:
+        return {GL_BACK_PRIMARY_COLOR_NV, 0};
+    case 43:
+        return {GL_BACK_SECONDARY_COLOR_NV, 0};
+    }
+    UNIMPLEMENTED_MSG("index={}", static_cast<int>(index));
+    return {GL_POSITION, 0};
+}
+
 void oglEnable(GLenum cap, bool state) {
     (state ? glEnable : glDisable)(cap);
 }
@@ -104,6 +152,9 @@ RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWind
       screen_info{info}, program_manager{program_manager}, state_tracker{state_tracker} {
     CheckExtensions();
 
+    unified_uniform_buffer.Create();
+    glNamedBufferStorage(unified_uniform_buffer.handle, TOTAL_CONST_BUFFER_BYTES, nullptr, 0);
+
     if (device.UseAssemblyShaders()) {
         glCreateBuffers(static_cast<GLsizei>(staging_cbufs.size()), staging_cbufs.data());
         for (const GLuint cbuf : staging_cbufs) {
@@ -143,7 +194,7 @@ void RasterizerOpenGL::SetupVertexFormat() {
     // avoid OpenGL errors.
     // TODO(Subv): Analyze the shader to identify which attributes are actually used and don't
     // assume every shader uses them all.
-    for (std::size_t index = 0; index < NumSupportedVertexAttributes; ++index) {
+    for (std::size_t index = 0; index < NUM_SUPPORTED_VERTEX_ATTRIBUTES; ++index) {
         if (!flags[Dirty::VertexFormat0 + index]) {
             continue;
         }
@@ -181,9 +232,11 @@ void RasterizerOpenGL::SetupVertexBuffer() {
 
     MICROPROFILE_SCOPE(OpenGL_VB);
 
+    const bool use_unified_memory = device.HasVertexBufferUnifiedMemory();
+
     // Upload all guest vertex arrays sequentially to our buffer
     const auto& regs = gpu.regs;
-    for (std::size_t index = 0; index < Maxwell::NumVertexArrays; ++index) {
+    for (std::size_t index = 0; index < NUM_SUPPORTED_VERTEX_BINDINGS; ++index) {
         if (!flags[Dirty::VertexBuffer0 + index]) {
             continue;
         }
@@ -196,16 +249,25 @@ void RasterizerOpenGL::SetupVertexBuffer() {
 
         const GPUVAddr start = vertex_array.StartAddress();
         const GPUVAddr end = regs.vertex_array_limit[index].LimitAddress();
-
         ASSERT(end >= start);
+
+        const GLuint gl_index = static_cast<GLuint>(index);
         const u64 size = end - start;
         if (size == 0) {
-            glBindVertexBuffer(static_cast<GLuint>(index), 0, 0, vertex_array.stride);
+            glBindVertexBuffer(gl_index, 0, 0, vertex_array.stride);
+            if (use_unified_memory) {
+                glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, gl_index, 0, 0);
+            }
             continue;
         }
-        const auto [vertex_buffer, vertex_buffer_offset] = buffer_cache.UploadMemory(start, size);
-        glBindVertexBuffer(static_cast<GLuint>(index), vertex_buffer, vertex_buffer_offset,
-                           vertex_array.stride);
+        const auto info = buffer_cache.UploadMemory(start, size);
+        if (use_unified_memory) {
+            glBindVertexBuffer(gl_index, 0, 0, vertex_array.stride);
+            glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, gl_index,
+                                   info.address + info.offset, size);
+        } else {
+            glBindVertexBuffer(gl_index, info.handle, info.offset, vertex_array.stride);
+        }
     }
 }
 
@@ -218,7 +280,7 @@ void RasterizerOpenGL::SetupVertexInstances() {
     flags[Dirty::VertexInstances] = false;
 
     const auto& regs = gpu.regs;
-    for (std::size_t index = 0; index < NumSupportedVertexAttributes; ++index) {
+    for (std::size_t index = 0; index < NUM_SUPPORTED_VERTEX_ATTRIBUTES; ++index) {
         if (!flags[Dirty::VertexInstance0 + index]) {
             continue;
         }
@@ -235,9 +297,9 @@ GLintptr RasterizerOpenGL::SetupIndexBuffer() {
     MICROPROFILE_SCOPE(OpenGL_Index);
     const auto& regs = system.GPU().Maxwell3D().regs;
     const std::size_t size = CalculateIndexBufferSize();
-    const auto [buffer, offset] = buffer_cache.UploadMemory(regs.index_array.IndexStart(), size);
-    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, buffer);
-    return offset;
+    const auto info = buffer_cache.UploadMemory(regs.index_array.IndexStart(), size);
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, info.handle);
+    return info.offset;
 }
 
 void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
@@ -273,7 +335,7 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
             continue;
         }
 
-        Shader shader{shader_cache.GetStageProgram(program)};
+        Shader* const shader = shader_cache.GetStageProgram(program);
 
         if (device.UseAssemblyShaders()) {
             // Check for ARB limitation. We only have 16 SSBOs per context state. To workaround this
@@ -567,7 +629,16 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
                    (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment());
 
     // Prepare the vertex array.
-    buffer_cache.Map(buffer_size);
+    const bool invalidated = buffer_cache.Map(buffer_size);
+
+    if (invalidated) {
+        // When the stream buffer has been invalidated, we have to consider vertex buffers as dirty
+        auto& dirty = gpu.dirty.flags;
+        dirty[Dirty::VertexBuffers] = true;
+        for (int index = Dirty::VertexBuffer0; index <= Dirty::VertexBuffer31; ++index) {
+            dirty[index] = true;
+        }
+    }
 
     // Prepare vertex array format.
     SetupVertexFormat();
@@ -584,9 +655,9 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
     if (!device.UseAssemblyShaders()) {
         MaxwellUniformData ubo;
         ubo.SetFromRegs(gpu);
-        const auto [buffer, offset] =
+        const auto info =
             buffer_cache.UploadHostMemory(&ubo, sizeof(ubo), device.GetUniformBufferAlignment());
-        glBindBufferRange(GL_UNIFORM_BUFFER, EmulationUniformBlockBinding, buffer, offset,
+        glBindBufferRange(GL_UNIFORM_BUFFER, EmulationUniformBlockBinding, info.handle, info.offset,
                           static_cast<GLsizeiptr>(sizeof(ubo)));
     }
 
@@ -655,10 +726,6 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
 }
 
 void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
-    if (device.HasBrokenCompute()) {
-        return;
-    }
-
     buffer_cache.Acquire();
     current_cbuf = 0;
 
@@ -837,7 +904,7 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
     return true;
 }
 
-void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, const Shader& shader) {
+void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, Shader* shader) {
     static constexpr std::array PARAMETER_LUT = {
         GL_VERTEX_PROGRAM_PARAMETER_BUFFER_NV, GL_TESS_CONTROL_PROGRAM_PARAMETER_BUFFER_NV,
         GL_TESS_EVALUATION_PROGRAM_PARAMETER_BUFFER_NV, GL_GEOMETRY_PROGRAM_PARAMETER_BUFFER_NV,
@@ -846,41 +913,62 @@ void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, const Shad
     MICROPROFILE_SCOPE(OpenGL_UBO);
     const auto& stages = system.GPU().Maxwell3D().state.shader_stages;
     const auto& shader_stage = stages[stage_index];
+    const auto& entries = shader->GetEntries();
+    const bool use_unified = entries.use_unified_uniforms;
+    const std::size_t base_unified_offset = stage_index * NUM_CONST_BUFFERS_BYTES_PER_STAGE;
 
-    u32 binding =
-        device.UseAssemblyShaders() ? 0 : device.GetBaseBindings(stage_index).uniform_buffer;
-    for (const auto& entry : shader->GetEntries().const_buffers) {
-        const auto& buffer = shader_stage.const_buffers[entry.GetIndex()];
-        SetupConstBuffer(PARAMETER_LUT[stage_index], binding++, buffer, entry);
+    const auto base_bindings = device.GetBaseBindings(stage_index);
+    u32 binding = device.UseAssemblyShaders() ? 0 : base_bindings.uniform_buffer;
+    for (const auto& entry : entries.const_buffers) {
+        const u32 index = entry.GetIndex();
+        const auto& buffer = shader_stage.const_buffers[index];
+        SetupConstBuffer(PARAMETER_LUT[stage_index], binding, buffer, entry, use_unified,
+                         base_unified_offset + index * Maxwell::MaxConstBufferSize);
+        ++binding;
+    }
+    if (use_unified) {
+        const u32 index = static_cast<u32>(base_bindings.shader_storage_buffer +
+                                           entries.global_memory_entries.size());
+        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, index, unified_uniform_buffer.handle,
+                          base_unified_offset, NUM_CONST_BUFFERS_BYTES_PER_STAGE);
     }
 }
 
-void RasterizerOpenGL::SetupComputeConstBuffers(const Shader& kernel) {
+void RasterizerOpenGL::SetupComputeConstBuffers(Shader* kernel) {
     MICROPROFILE_SCOPE(OpenGL_UBO);
     const auto& launch_desc = system.GPU().KeplerCompute().launch_description;
+    const auto& entries = kernel->GetEntries();
+    const bool use_unified = entries.use_unified_uniforms;
 
     u32 binding = 0;
-    for (const auto& entry : kernel->GetEntries().const_buffers) {
+    for (const auto& entry : entries.const_buffers) {
         const auto& config = launch_desc.const_buffer_config[entry.GetIndex()];
         const std::bitset<8> mask = launch_desc.const_buffer_enable_mask.Value();
         Tegra::Engines::ConstBufferInfo buffer;
         buffer.address = config.Address();
         buffer.size = config.size;
         buffer.enabled = mask[entry.GetIndex()];
-        SetupConstBuffer(GL_COMPUTE_PROGRAM_PARAMETER_BUFFER_NV, binding++, buffer, entry);
+        SetupConstBuffer(GL_COMPUTE_PROGRAM_PARAMETER_BUFFER_NV, binding, buffer, entry,
+                         use_unified, entry.GetIndex() * Maxwell::MaxConstBufferSize);
+        ++binding;
+    }
+    if (use_unified) {
+        const GLuint index = static_cast<GLuint>(entries.global_memory_entries.size());
+        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, index, unified_uniform_buffer.handle, 0,
+                          NUM_CONST_BUFFERS_BYTES_PER_STAGE);
     }
 }
 
 void RasterizerOpenGL::SetupConstBuffer(GLenum stage, u32 binding,
                                         const Tegra::Engines::ConstBufferInfo& buffer,
-                                        const ConstBufferEntry& entry) {
+                                        const ConstBufferEntry& entry, bool use_unified,
+                                        std::size_t unified_offset) {
     if (!buffer.enabled) {
         // Set values to zero to unbind buffers
         if (device.UseAssemblyShaders()) {
             glBindBufferRangeNV(stage, entry.GetIndex(), 0, 0, 0);
         } else {
-            glBindBufferRange(GL_UNIFORM_BUFFER, binding,
-                              buffer_cache.GetEmptyBuffer(sizeof(float)), 0, sizeof(float));
+            glBindBufferRange(GL_UNIFORM_BUFFER, binding, 0, 0, sizeof(float));
         }
         return;
     }
@@ -889,23 +977,33 @@ void RasterizerOpenGL::SetupConstBuffer(GLenum stage, u32 binding,
     // UBO alignment requirements.
     const std::size_t size = Common::AlignUp(GetConstBufferSize(buffer, entry), sizeof(GLvec4));
 
-    const auto alignment = device.GetUniformBufferAlignment();
-    auto [cbuf, offset] = buffer_cache.UploadMemory(buffer.address, size, alignment, false,
-                                                    device.HasFastBufferSubData());
-    if (!device.UseAssemblyShaders()) {
-        glBindBufferRange(GL_UNIFORM_BUFFER, binding, cbuf, offset, size);
+    const bool fast_upload = !use_unified && device.HasFastBufferSubData();
+
+    const std::size_t alignment = use_unified ? 4 : device.GetUniformBufferAlignment();
+    const GPUVAddr gpu_addr = buffer.address;
+    auto info = buffer_cache.UploadMemory(gpu_addr, size, alignment, false, fast_upload);
+
+    if (device.UseAssemblyShaders()) {
+        UNIMPLEMENTED_IF(use_unified);
+        if (info.offset != 0) {
+            const GLuint staging_cbuf = staging_cbufs[current_cbuf++];
+            glCopyNamedBufferSubData(info.handle, staging_cbuf, info.offset, 0, size);
+            info.handle = staging_cbuf;
+            info.offset = 0;
+        }
+        glBindBufferRangeNV(stage, binding, info.handle, info.offset, size);
         return;
     }
-    if (offset != 0) {
-        const GLuint staging_cbuf = staging_cbufs[current_cbuf++];
-        glCopyNamedBufferSubData(cbuf, staging_cbuf, offset, 0, size);
-        cbuf = staging_cbuf;
-        offset = 0;
+
+    if (use_unified) {
+        glCopyNamedBufferSubData(info.handle, unified_uniform_buffer.handle, info.offset,
+                                 unified_offset, size);
+    } else {
+        glBindBufferRange(GL_UNIFORM_BUFFER, binding, info.handle, info.offset, size);
     }
-    glBindBufferRangeNV(stage, binding, cbuf, offset, size);
 }
 
-void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, const Shader& shader) {
+void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, Shader* shader) {
     auto& gpu{system.GPU()};
     auto& memory_manager{gpu.MemoryManager()};
     const auto cbufs{gpu.Maxwell3D().state.shader_stages[stage_index]};
@@ -920,7 +1018,7 @@ void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, const Shad
     }
 }
 
-void RasterizerOpenGL::SetupComputeGlobalMemory(const Shader& kernel) {
+void RasterizerOpenGL::SetupComputeGlobalMemory(Shader* kernel) {
     auto& gpu{system.GPU()};
     auto& memory_manager{gpu.MemoryManager()};
     const auto cbufs{gpu.KeplerCompute().launch_description.const_buffer_config};
@@ -937,13 +1035,12 @@ void RasterizerOpenGL::SetupComputeGlobalMemory(const Shader& kernel) {
 void RasterizerOpenGL::SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& entry,
                                          GPUVAddr gpu_addr, std::size_t size) {
     const auto alignment{device.GetShaderStorageBufferAlignment()};
-    const auto [ssbo, buffer_offset] =
-        buffer_cache.UploadMemory(gpu_addr, size, alignment, entry.is_written);
-    glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, ssbo, buffer_offset,
+    const auto info = buffer_cache.UploadMemory(gpu_addr, size, alignment, entry.is_written);
+    glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, info.handle, info.offset,
                       static_cast<GLsizeiptr>(size));
 }
 
-void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, const Shader& shader) {
+void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, Shader* shader) {
     MICROPROFILE_SCOPE(OpenGL_Texture);
     const auto& maxwell3d = system.GPU().Maxwell3D();
     u32 binding = device.GetBaseBindings(stage_index).sampler;
@@ -956,7 +1053,7 @@ void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, const Shader&
     }
 }
 
-void RasterizerOpenGL::SetupComputeTextures(const Shader& kernel) {
+void RasterizerOpenGL::SetupComputeTextures(Shader* kernel) {
     MICROPROFILE_SCOPE(OpenGL_Texture);
     const auto& compute = system.GPU().KeplerCompute();
     u32 binding = 0;
@@ -985,7 +1082,7 @@ void RasterizerOpenGL::SetupTexture(u32 binding, const Tegra::Texture::FullTextu
     }
 }
 
-void RasterizerOpenGL::SetupDrawImages(std::size_t stage_index, const Shader& shader) {
+void RasterizerOpenGL::SetupDrawImages(std::size_t stage_index, Shader* shader) {
     const auto& maxwell3d = system.GPU().Maxwell3D();
     u32 binding = device.GetBaseBindings(stage_index).image;
     for (const auto& entry : shader->GetEntries().images) {
@@ -995,7 +1092,7 @@ void RasterizerOpenGL::SetupDrawImages(std::size_t stage_index, const Shader& sh
     }
 }
 
-void RasterizerOpenGL::SetupComputeImages(const Shader& shader) {
+void RasterizerOpenGL::SetupComputeImages(Shader* shader) {
     const auto& compute = system.GPU().KeplerCompute();
     u32 binding = 0;
     for (const auto& entry : shader->GetEntries().images) {
@@ -1024,6 +1121,26 @@ void RasterizerOpenGL::SyncViewport() {
     const auto& regs = gpu.regs;
 
     const bool dirty_viewport = flags[Dirty::Viewports];
+    const bool dirty_clip_control = flags[Dirty::ClipControl];
+
+    if (dirty_clip_control || flags[Dirty::FrontFace]) {
+        flags[Dirty::FrontFace] = false;
+
+        GLenum mode = MaxwellToGL::FrontFace(regs.front_face);
+        if (regs.screen_y_control.triangle_rast_flip != 0 &&
+            regs.viewport_transform[0].scale_y < 0.0f) {
+            switch (mode) {
+            case GL_CW:
+                mode = GL_CCW;
+                break;
+            case GL_CCW:
+                mode = GL_CW;
+                break;
+            }
+        }
+        glFrontFace(mode);
+    }
+
     if (dirty_viewport || flags[Dirty::ClipControl]) {
         flags[Dirty::ClipControl] = false;
 
@@ -1121,11 +1238,6 @@ void RasterizerOpenGL::SyncCullMode() {
             glDisable(GL_CULL_FACE);
         }
     }
-
-    if (flags[Dirty::FrontFace]) {
-        flags[Dirty::FrontFace] = false;
-        glFrontFace(MaxwellToGL::FrontFace(regs.front_face));
-    }
 }
 
 void RasterizerOpenGL::SyncPrimitiveRestart() {
@@ -1496,12 +1608,70 @@ void RasterizerOpenGL::SyncFramebufferSRGB() {
     oglEnable(GL_FRAMEBUFFER_SRGB, gpu.regs.framebuffer_srgb);
 }
 
+void RasterizerOpenGL::SyncTransformFeedback() {
+    // TODO(Rodrigo): Inject SKIP_COMPONENTS*_NV when required. An unimplemented message will signal
+    // when this is required.
+    const auto& regs = system.GPU().Maxwell3D().regs;
+
+    static constexpr std::size_t STRIDE = 3;
+    std::array<GLint, 128 * STRIDE * Maxwell::NumTransformFeedbackBuffers> attribs;
+    std::array<GLint, Maxwell::NumTransformFeedbackBuffers> streams;
+
+    GLint* cursor = attribs.data();
+    GLint* current_stream = streams.data();
+
+    for (std::size_t feedback = 0; feedback < Maxwell::NumTransformFeedbackBuffers; ++feedback) {
+        const auto& layout = regs.tfb_layouts[feedback];
+        UNIMPLEMENTED_IF_MSG(layout.stride != layout.varying_count * 4, "Stride padding");
+        if (layout.varying_count == 0) {
+            continue;
+        }
+
+        *current_stream = static_cast<GLint>(feedback);
+        if (current_stream != streams.data()) {
+            // When stepping one stream, push the expected token
+            cursor[0] = GL_NEXT_BUFFER_NV;
+            cursor[1] = 0;
+            cursor[2] = 0;
+            cursor += STRIDE;
+        }
+        ++current_stream;
+
+        const auto& locations = regs.tfb_varying_locs[feedback];
+        std::optional<u8> current_index;
+        for (u32 offset = 0; offset < layout.varying_count; ++offset) {
+            const u8 location = locations[offset];
+            const u8 index = location / 4;
+
+            if (current_index == index) {
+                // Increase number of components of the previous attachment
+                ++cursor[-2];
+                continue;
+            }
+            current_index = index;
+
+            std::tie(cursor[0], cursor[2]) = TransformFeedbackEnum(location);
+            cursor[1] = 1;
+            cursor += STRIDE;
+        }
+    }
+
+    const GLsizei num_attribs = static_cast<GLsizei>((cursor - attribs.data()) / STRIDE);
+    const GLsizei num_strides = static_cast<GLsizei>(current_stream - streams.data());
+    glTransformFeedbackStreamAttribsNV(num_attribs, attribs.data(), num_strides, streams.data(),
+                                       GL_INTERLEAVED_ATTRIBS);
+}
+
 void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) {
     const auto& regs = system.GPU().Maxwell3D().regs;
     if (regs.tfb_enabled == 0) {
         return;
     }
 
+    if (device.UseAssemblyShaders()) {
+        SyncTransformFeedback();
+    }
+
     UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationControl) ||
                      regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationEval) ||
                      regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::Geometry));
@@ -1528,6 +1698,10 @@ void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) {
                           static_cast<GLsizeiptr>(size));
     }
 
+    // We may have to call BeginTransformFeedbackNV here since they seem to call different
+    // implementations on Nvidia's driver (the pointer is different) but we are using
+    // ARB_transform_feedback3 features with NV_transform_feedback interactions and the ARB
+    // extension doesn't define BeginTransformFeedback (without NV) interactions. It just works.
     glBeginTransformFeedback(GL_POINTS);
 }
 
@@ -1549,8 +1723,9 @@ void RasterizerOpenGL::EndTransformFeedback() {
         const GLuint handle = transform_feedback_buffers[index].handle;
         const GPUVAddr gpu_addr = binding.Address();
         const std::size_t size = binding.buffer_size;
-        const auto [dest_buffer, offset] = buffer_cache.UploadMemory(gpu_addr, size, 4, true);
-        glCopyNamedBufferSubData(handle, dest_buffer, 0, offset, static_cast<GLsizeiptr>(size));
+        const auto info = buffer_cache.UploadMemory(gpu_addr, size, 4, true);
+        glCopyNamedBufferSubData(handle, info.handle, 0, info.offset,
+                                 static_cast<GLsizeiptr>(size));
     }
 }
 

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -19,7 +19,6 @@
 #include "video_core/engines/const_buffer_info.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/rasterizer_accelerated.h"
-#include "video_core/rasterizer_cache.h"
 #include "video_core/rasterizer_interface.h"
 #include "video_core/renderer_opengl/gl_buffer_cache.h"
 #include "video_core/renderer_opengl/gl_device.h"
@@ -100,40 +99,41 @@ private:
     void ConfigureClearFramebuffer(bool using_color, bool using_depth_stencil);
 
     /// Configures the current constbuffers to use for the draw command.
-    void SetupDrawConstBuffers(std::size_t stage_index, const Shader& shader);
+    void SetupDrawConstBuffers(std::size_t stage_index, Shader* shader);
 
     /// Configures the current constbuffers to use for the kernel invocation.
-    void SetupComputeConstBuffers(const Shader& kernel);
+    void SetupComputeConstBuffers(Shader* kernel);
 
     /// Configures a constant buffer.
     void SetupConstBuffer(GLenum stage, u32 binding, const Tegra::Engines::ConstBufferInfo& buffer,
-                          const ConstBufferEntry& entry);
+                          const ConstBufferEntry& entry, bool use_unified,
+                          std::size_t unified_offset);
 
     /// Configures the current global memory entries to use for the draw command.
-    void SetupDrawGlobalMemory(std::size_t stage_index, const Shader& shader);
+    void SetupDrawGlobalMemory(std::size_t stage_index, Shader* shader);
 
     /// Configures the current global memory entries to use for the kernel invocation.
-    void SetupComputeGlobalMemory(const Shader& kernel);
+    void SetupComputeGlobalMemory(Shader* kernel);
 
     /// Configures a constant buffer.
     void SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& entry, GPUVAddr gpu_addr,
                            std::size_t size);
 
     /// Configures the current textures to use for the draw command.
-    void SetupDrawTextures(std::size_t stage_index, const Shader& shader);
+    void SetupDrawTextures(std::size_t stage_index, Shader* shader);
 
     /// Configures the textures used in a compute shader.
-    void SetupComputeTextures(const Shader& kernel);
+    void SetupComputeTextures(Shader* kernel);
 
     /// Configures a texture.
     void SetupTexture(u32 binding, const Tegra::Texture::FullTextureInfo& texture,
                       const SamplerEntry& entry);
 
     /// Configures images in a graphics shader.
-    void SetupDrawImages(std::size_t stage_index, const Shader& shader);
+    void SetupDrawImages(std::size_t stage_index, Shader* shader);
 
     /// Configures images in a compute shader.
-    void SetupComputeImages(const Shader& shader);
+    void SetupComputeImages(Shader* shader);
 
     /// Configures an image.
     void SetupImage(u32 binding, const Tegra::Texture::TICEntry& tic, const ImageEntry& entry);
@@ -201,6 +201,10 @@ private:
     /// Syncs the framebuffer sRGB state to match the guest state
     void SyncFramebufferSRGB();
 
+    /// Syncs transform feedback state to match guest state
+    /// @note Only valid on assembly shaders
+    void SyncTransformFeedback();
+
     /// Begin a transform feedback
     void BeginTransformFeedback(GLenum primitive_mode);
 
@@ -253,6 +257,7 @@ private:
         Tegra::Engines::Maxwell3D::Regs::MaxShaderProgram;
     std::array<GLuint, NUM_CONSTANT_BUFFERS> staging_cbufs{};
     std::size_t current_cbuf = 0;
+    OGLBuffer unified_uniform_buffer;
 
     /// Number of commands queued to the OpenGL driver. Reseted on flush.
     std::size_t num_queued_commands = 0;

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -20,6 +20,7 @@
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/engines/shader_type.h"
 #include "video_core/memory_manager.h"
+#include "video_core/renderer_opengl/gl_arb_decompiler.h"
 #include "video_core/renderer_opengl/gl_rasterizer.h"
 #include "video_core/renderer_opengl/gl_shader_cache.h"
 #include "video_core/renderer_opengl/gl_shader_decompiler.h"
@@ -29,6 +30,7 @@
 #include "video_core/shader/memory_util.h"
 #include "video_core/shader/registry.h"
 #include "video_core/shader/shader_ir.h"
+#include "video_core/shader_cache.h"
 
 namespace OpenGL {
 
@@ -147,7 +149,8 @@ ProgramSharedPtr BuildShader(const Device& device, ShaderType shader_type, u64 u
     auto program = std::make_shared<ProgramHandle>();
 
     if (device.UseAssemblyShaders()) {
-        const std::string arb = "Not implemented";
+        const std::string arb =
+            DecompileAssemblyShader(device, ir, registry, shader_type, shader_id);
 
         GLuint& arb_prog = program->assembly_program.handle;
 
@@ -194,12 +197,9 @@ std::unordered_set<GLenum> GetSupportedFormats() {
 
 } // Anonymous namespace
 
-CachedShader::CachedShader(VAddr cpu_addr, std::size_t size_in_bytes,
-                           std::shared_ptr<VideoCommon::Shader::Registry> registry,
-                           ShaderEntries entries, ProgramSharedPtr program_)
-    : RasterizerCacheObject{cpu_addr}, registry{std::move(registry)}, entries{std::move(entries)},
-      size_in_bytes{size_in_bytes}, program{std::move(program_)} {
-    // Assign either the assembly program or source program. We can't have both.
+Shader::Shader(std::shared_ptr<VideoCommon::Shader::Registry> registry_, ShaderEntries entries_,
+               ProgramSharedPtr program_)
+    : registry{std::move(registry_)}, entries{std::move(entries_)}, program{std::move(program_)} {
     handle = program->assembly_program.handle;
     if (handle == 0) {
         handle = program->source_program.handle;
@@ -207,16 +207,16 @@ CachedShader::CachedShader(VAddr cpu_addr, std::size_t size_in_bytes,
     ASSERT(handle != 0);
 }
 
-CachedShader::~CachedShader() = default;
+Shader::~Shader() = default;
 
-GLuint CachedShader::GetHandle() const {
+GLuint Shader::GetHandle() const {
     DEBUG_ASSERT(registry->IsConsistent());
     return handle;
 }
 
-Shader CachedShader::CreateStageFromMemory(const ShaderParameters& params,
-                                           Maxwell::ShaderProgram program_type, ProgramCode code,
-                                           ProgramCode code_b) {
+std::unique_ptr<Shader> Shader::CreateStageFromMemory(const ShaderParameters& params,
+                                                      Maxwell::ShaderProgram program_type,
+                                                      ProgramCode code, ProgramCode code_b) {
     const auto shader_type = GetShaderType(program_type);
     const std::size_t size_in_bytes = code.size() * sizeof(u64);
 
@@ -241,11 +241,12 @@ Shader CachedShader::CreateStageFromMemory(const ShaderParameters& params,
     entry.bindless_samplers = registry->GetBindlessSamplers();
     params.disk_cache.SaveEntry(std::move(entry));
 
-    return std::shared_ptr<CachedShader>(new CachedShader(
-        params.cpu_addr, size_in_bytes, std::move(registry), MakeEntries(ir), std::move(program)));
+    return std::unique_ptr<Shader>(new Shader(
+        std::move(registry), MakeEntries(params.device, ir, shader_type), std::move(program)));
 }
 
-Shader CachedShader::CreateKernelFromMemory(const ShaderParameters& params, ProgramCode code) {
+std::unique_ptr<Shader> Shader::CreateKernelFromMemory(const ShaderParameters& params,
+                                                       ProgramCode code) {
     const std::size_t size_in_bytes = code.size() * sizeof(u64);
 
     auto& engine = params.system.GPU().KeplerCompute();
@@ -265,22 +266,23 @@ Shader CachedShader::CreateKernelFromMemory(const ShaderParameters& params, Prog
     entry.bindless_samplers = registry->GetBindlessSamplers();
     params.disk_cache.SaveEntry(std::move(entry));
 
-    return std::shared_ptr<CachedShader>(new CachedShader(
-        params.cpu_addr, size_in_bytes, std::move(registry), MakeEntries(ir), std::move(program)));
+    return std::unique_ptr<Shader>(new Shader(std::move(registry),
+                                              MakeEntries(params.device, ir, ShaderType::Compute),
+                                              std::move(program)));
 }
 
-Shader CachedShader::CreateFromCache(const ShaderParameters& params,
-                                     const PrecompiledShader& precompiled_shader,
-                                     std::size_t size_in_bytes) {
-    return std::shared_ptr<CachedShader>(
-        new CachedShader(params.cpu_addr, size_in_bytes, precompiled_shader.registry,
-                         precompiled_shader.entries, precompiled_shader.program));
+std::unique_ptr<Shader> Shader::CreateFromCache(const ShaderParameters& params,
+                                                const PrecompiledShader& precompiled_shader) {
+    return std::unique_ptr<Shader>(new Shader(
+        precompiled_shader.registry, precompiled_shader.entries, precompiled_shader.program));
 }
 
 ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer, Core::System& system,
                                      Core::Frontend::EmuWindow& emu_window, const Device& device)
-    : RasterizerCache{rasterizer}, system{system}, emu_window{emu_window}, device{device},
-      disk_cache{system} {}
+    : VideoCommon::ShaderCache<Shader>{rasterizer}, system{system},
+      emu_window{emu_window}, device{device}, disk_cache{system} {}
+
+ShaderCacheOpenGL::~ShaderCacheOpenGL() = default;
 
 void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
                                       const VideoCore::DiskResourceLoadCallback& callback) {
@@ -348,7 +350,7 @@ void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
             PrecompiledShader shader;
             shader.program = std::move(program);
             shader.registry = std::move(registry);
-            shader.entries = MakeEntries(ir);
+            shader.entries = MakeEntries(device, ir, entry.type);
 
             std::scoped_lock lock{mutex};
             if (callback) {
@@ -434,7 +436,7 @@ ProgramSharedPtr ShaderCacheOpenGL::GeneratePrecompiledProgram(
     return program;
 }
 
-Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
+Shader* ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
     if (!system.GPU().Maxwell3D().dirty.flags[Dirty::Shaders]) {
         return last_shaders[static_cast<std::size_t>(program)];
     }
@@ -444,8 +446,7 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
 
     // Look up shader in the cache based on address
     const auto cpu_addr{memory_manager.GpuToCpuAddress(address)};
-    Shader shader{cpu_addr ? TryGet(*cpu_addr) : null_shader};
-    if (shader) {
+    if (Shader* const shader{cpu_addr ? TryGet(*cpu_addr) : null_shader.get()}) {
         return last_shaders[static_cast<std::size_t>(program)] = shader;
     }
 
@@ -466,30 +467,29 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
     const ShaderParameters params{system,    disk_cache, device,
                                   *cpu_addr, host_ptr,   unique_identifier};
 
+    std::unique_ptr<Shader> shader;
     const auto found = runtime_cache.find(unique_identifier);
     if (found == runtime_cache.end()) {
-        shader = CachedShader::CreateStageFromMemory(params, program, std::move(code),
-                                                     std::move(code_b));
+        shader = Shader::CreateStageFromMemory(params, program, std::move(code), std::move(code_b));
     } else {
-        const std::size_t size_in_bytes = code.size() * sizeof(u64);
-        shader = CachedShader::CreateFromCache(params, found->second, size_in_bytes);
+        shader = Shader::CreateFromCache(params, found->second);
     }
 
+    Shader* const result = shader.get();
     if (cpu_addr) {
-        Register(shader);
+        Register(std::move(shader), *cpu_addr, code.size() * sizeof(u64));
     } else {
-        null_shader = shader;
+        null_shader = std::move(shader);
     }
 
-    return last_shaders[static_cast<std::size_t>(program)] = shader;
+    return last_shaders[static_cast<std::size_t>(program)] = result;
 }
 
-Shader ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
+Shader* ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
     auto& memory_manager{system.GPU().MemoryManager()};
     const auto cpu_addr{memory_manager.GpuToCpuAddress(code_addr)};
 
-    auto kernel = cpu_addr ? TryGet(*cpu_addr) : null_kernel;
-    if (kernel) {
+    if (Shader* const kernel = cpu_addr ? TryGet(*cpu_addr) : null_kernel.get()) {
         return kernel;
     }
 
@@ -501,20 +501,21 @@ Shader ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
     const ShaderParameters params{system,    disk_cache, device,
                                   *cpu_addr, host_ptr,   unique_identifier};
 
+    std::unique_ptr<Shader> kernel;
     const auto found = runtime_cache.find(unique_identifier);
     if (found == runtime_cache.end()) {
-        kernel = CachedShader::CreateKernelFromMemory(params, std::move(code));
+        kernel = Shader::CreateKernelFromMemory(params, std::move(code));
     } else {
-        const std::size_t size_in_bytes = code.size() * sizeof(u64);
-        kernel = CachedShader::CreateFromCache(params, found->second, size_in_bytes);
+        kernel = Shader::CreateFromCache(params, found->second);
     }
 
+    Shader* const result = kernel.get();
     if (cpu_addr) {
-        Register(kernel);
+        Register(std::move(kernel), *cpu_addr, code.size() * sizeof(u64));
     } else {
-        null_kernel = kernel;
+        null_kernel = std::move(kernel);
     }
-    return kernel;
+    return result;
 }
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_shader_cache.h

@@ -18,12 +18,12 @@
 
 #include "common/common_types.h"
 #include "video_core/engines/shader_type.h"
-#include "video_core/rasterizer_cache.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 #include "video_core/renderer_opengl/gl_shader_decompiler.h"
 #include "video_core/renderer_opengl/gl_shader_disk_cache.h"
 #include "video_core/shader/registry.h"
 #include "video_core/shader/shader_ir.h"
+#include "video_core/shader_cache.h"
 
 namespace Core {
 class System;
@@ -35,12 +35,10 @@ class EmuWindow;
 
 namespace OpenGL {
 
-class CachedShader;
 class Device;
 class RasterizerOpenGL;
 struct UnspecializedShader;
 
-using Shader = std::shared_ptr<CachedShader>;
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 
 struct ProgramHandle {
@@ -64,62 +62,53 @@ struct ShaderParameters {
     u64 unique_identifier;
 };
 
-class CachedShader final : public RasterizerCacheObject {
+class Shader final {
 public:
-    ~CachedShader();
+    ~Shader();
 
     /// Gets the GL program handle for the shader
     GLuint GetHandle() const;
 
-    /// Returns the size in bytes of the shader
-    std::size_t GetSizeInBytes() const override {
-        return size_in_bytes;
-    }
-
     /// Gets the shader entries for the shader
     const ShaderEntries& GetEntries() const {
         return entries;
     }
 
-    static Shader CreateStageFromMemory(const ShaderParameters& params,
-                                        Maxwell::ShaderProgram program_type,
-                                        ProgramCode program_code, ProgramCode program_code_b);
-    static Shader CreateKernelFromMemory(const ShaderParameters& params, ProgramCode code);
+    static std::unique_ptr<Shader> CreateStageFromMemory(const ShaderParameters& params,
+                                                         Maxwell::ShaderProgram program_type,
+                                                         ProgramCode program_code,
+                                                         ProgramCode program_code_b);
+    static std::unique_ptr<Shader> CreateKernelFromMemory(const ShaderParameters& params,
+                                                          ProgramCode code);
 
-    static Shader CreateFromCache(const ShaderParameters& params,
-                                  const PrecompiledShader& precompiled_shader,
-                                  std::size_t size_in_bytes);
+    static std::unique_ptr<Shader> CreateFromCache(const ShaderParameters& params,
+                                                   const PrecompiledShader& precompiled_shader);
 
 private:
-    explicit CachedShader(VAddr cpu_addr, std::size_t size_in_bytes,
-                          std::shared_ptr<VideoCommon::Shader::Registry> registry,
-                          ShaderEntries entries, ProgramSharedPtr program);
+    explicit Shader(std::shared_ptr<VideoCommon::Shader::Registry> registry, ShaderEntries entries,
+                    ProgramSharedPtr program);
 
     std::shared_ptr<VideoCommon::Shader::Registry> registry;
     ShaderEntries entries;
-    std::size_t size_in_bytes = 0;
     ProgramSharedPtr program;
     GLuint handle = 0;
 };
 
-class ShaderCacheOpenGL final : public RasterizerCache<Shader> {
+class ShaderCacheOpenGL final : public VideoCommon::ShaderCache<Shader> {
 public:
     explicit ShaderCacheOpenGL(RasterizerOpenGL& rasterizer, Core::System& system,
                                Core::Frontend::EmuWindow& emu_window, const Device& device);
+    ~ShaderCacheOpenGL() override;
 
     /// Loads disk cache for the current game
     void LoadDiskCache(const std::atomic_bool& stop_loading,
                        const VideoCore::DiskResourceLoadCallback& callback);
 
     /// Gets the current specified shader stage program
-    Shader GetStageProgram(Maxwell::ShaderProgram program);
+    Shader* GetStageProgram(Maxwell::ShaderProgram program);
 
     /// Gets a compute kernel in the passed address
-    Shader GetComputeKernel(GPUVAddr code_addr);
-
-protected:
-    // We do not have to flush this cache as things in it are never modified by us.
-    void FlushObjectInner(const Shader& object) override {}
+    Shader* GetComputeKernel(GPUVAddr code_addr);
 
 private:
     ProgramSharedPtr GeneratePrecompiledProgram(
@@ -132,10 +121,10 @@ private:
     ShaderDiskCacheOpenGL disk_cache;
     std::unordered_map<u64, PrecompiledShader> runtime_cache;
 
-    Shader null_shader{};
-    Shader null_kernel{};
+    std::unique_ptr<Shader> null_shader;
+    std::unique_ptr<Shader> null_kernel;
 
-    std::array<Shader, Maxwell::MaxShaderProgram> last_shaders;
+    std::array<Shader*, Maxwell::MaxShaderProgram> last_shaders{};
 };
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -37,6 +37,7 @@ using Tegra::Shader::IpaMode;
 using Tegra::Shader::IpaSampleMode;
 using Tegra::Shader::PixelImap;
 using Tegra::Shader::Register;
+using Tegra::Shader::TextureType;
 using VideoCommon::Shader::BuildTransformFeedback;
 using VideoCommon::Shader::Registry;
 
@@ -61,8 +62,8 @@ struct TextureDerivates {};
 using TextureArgument = std::pair<Type, Node>;
 using TextureIR = std::variant<TextureOffset, TextureDerivates, TextureArgument>;
 
-constexpr u32 MAX_CONSTBUFFER_ELEMENTS =
-    static_cast<u32>(Maxwell::MaxConstBufferSize) / (4 * sizeof(float));
+constexpr u32 MAX_CONSTBUFFER_SCALARS = static_cast<u32>(Maxwell::MaxConstBufferSize) / sizeof(u32);
+constexpr u32 MAX_CONSTBUFFER_ELEMENTS = MAX_CONSTBUFFER_SCALARS / sizeof(u32);
 
 constexpr std::string_view CommonDeclarations = R"(#define ftoi floatBitsToInt
 #define ftou floatBitsToUint
@@ -402,6 +403,13 @@ std::string FlowStackTopName(MetaStackClass stack) {
     return fmt::format("{}_flow_stack_top", GetFlowStackPrefix(stack));
 }
 
+bool UseUnifiedUniforms(const Device& device, const ShaderIR& ir, ShaderType stage) {
+    const u32 num_ubos = static_cast<u32>(ir.GetConstantBuffers().size());
+    // We waste one UBO for emulation
+    const u32 num_available_ubos = device.GetMaxUniformBuffers(stage) - 1;
+    return num_ubos > num_available_ubos;
+}
+
 struct GenericVaryingDescription {
     std::string name;
     u8 first_element = 0;
@@ -412,8 +420,9 @@ class GLSLDecompiler final {
 public:
     explicit GLSLDecompiler(const Device& device, const ShaderIR& ir, const Registry& registry,
                             ShaderType stage, std::string_view identifier, std::string_view suffix)
-        : device{device}, ir{ir}, registry{registry}, stage{stage},
-          identifier{identifier}, suffix{suffix}, header{ir.GetHeader()} {
+        : device{device}, ir{ir}, registry{registry}, stage{stage}, identifier{identifier},
+          suffix{suffix}, header{ir.GetHeader()}, use_unified_uniforms{
+                                                      UseUnifiedUniforms(device, ir, stage)} {
         if (stage != ShaderType::Compute) {
             transform_feedback = BuildTransformFeedback(registry.GetGraphicsInfo());
         }
@@ -518,6 +527,9 @@ private:
         if (device.HasImageLoadFormatted()) {
             code.AddLine("#extension GL_EXT_shader_image_load_formatted : require");
         }
+        if (device.HasTextureShadowLod()) {
+            code.AddLine("#extension GL_EXT_texture_shadow_lod : require");
+        }
         if (device.HasWarpIntrinsics()) {
             code.AddLine("#extension GL_NV_gpu_shader5 : require");
             code.AddLine("#extension GL_NV_shader_thread_group : require");
@@ -618,7 +630,9 @@ private:
                 break;
             }
         }
-        if (stage != ShaderType::Vertex || device.HasVertexViewportLayer()) {
+
+        if (stage != ShaderType::Geometry &&
+            (stage != ShaderType::Vertex || device.HasVertexViewportLayer())) {
             if (ir.UsesLayer()) {
                 code.AddLine("int gl_Layer;");
             }
@@ -647,6 +661,16 @@ private:
         --code.scope;
         code.AddLine("}};");
         code.AddNewLine();
+
+        if (stage == ShaderType::Geometry) {
+            if (ir.UsesLayer()) {
+                code.AddLine("out int gl_Layer;");
+            }
+            if (ir.UsesViewportIndex()) {
+                code.AddLine("out int gl_ViewportIndex;");
+            }
+        }
+        code.AddNewLine();
     }
 
     void DeclareRegisters() {
@@ -834,12 +858,24 @@ private:
     }
 
     void DeclareConstantBuffers() {
+        if (use_unified_uniforms) {
+            const u32 binding = device.GetBaseBindings(stage).shader_storage_buffer +
+                                static_cast<u32>(ir.GetGlobalMemory().size());
+            code.AddLine("layout (std430, binding = {}) readonly buffer UnifiedUniforms {{",
+                         binding);
+            code.AddLine("    uint cbufs[];");
+            code.AddLine("}};");
+            code.AddNewLine();
+            return;
+        }
+
         u32 binding = device.GetBaseBindings(stage).uniform_buffer;
-        for (const auto& buffers : ir.GetConstantBuffers()) {
-            const auto index = buffers.first;
+        for (const auto [index, info] : ir.GetConstantBuffers()) {
+            const u32 num_elements = Common::AlignUp(info.GetSize(), 4) / 4;
+            const u32 size = info.IsIndirect() ? MAX_CONSTBUFFER_ELEMENTS : num_elements;
             code.AddLine("layout (std140, binding = {}) uniform {} {{", binding++,
                          GetConstBufferBlock(index));
-            code.AddLine("    uvec4 {}[{}];", GetConstBuffer(index), MAX_CONSTBUFFER_ELEMENTS);
+            code.AddLine("    uvec4 {}[{}];", GetConstBuffer(index), size);
             code.AddLine("}};");
             code.AddNewLine();
         }
@@ -877,13 +913,13 @@ private:
                     return "samplerBuffer";
                 }
                 switch (sampler.type) {
-                case Tegra::Shader::TextureType::Texture1D:
+                case TextureType::Texture1D:
                     return "sampler1D";
-                case Tegra::Shader::TextureType::Texture2D:
+                case TextureType::Texture2D:
                     return "sampler2D";
-                case Tegra::Shader::TextureType::Texture3D:
+                case TextureType::Texture3D:
                     return "sampler3D";
-                case Tegra::Shader::TextureType::TextureCube:
+                case TextureType::TextureCube:
                     return "samplerCube";
                 default:
                     UNREACHABLE();
@@ -1038,42 +1074,51 @@ private:
 
         if (const auto cbuf = std::get_if<CbufNode>(&*node)) {
             const Node offset = cbuf->GetOffset();
+            const u32 base_unified_offset = cbuf->GetIndex() * MAX_CONSTBUFFER_SCALARS;
+
             if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) {
                 // Direct access
                 const u32 offset_imm = immediate->GetValue();
                 ASSERT_MSG(offset_imm % 4 == 0, "Unaligned cbuf direct access");
-                return {fmt::format("{}[{}][{}]", GetConstBuffer(cbuf->GetIndex()),
-                                    offset_imm / (4 * 4), (offset_imm / 4) % 4),
-                        Type::Uint};
+                if (use_unified_uniforms) {
+                    return {fmt::format("cbufs[{}]", base_unified_offset + offset_imm / 4),
+                            Type::Uint};
+                } else {
+                    return {fmt::format("{}[{}][{}]", GetConstBuffer(cbuf->GetIndex()),
+                                        offset_imm / (4 * 4), (offset_imm / 4) % 4),
+                            Type::Uint};
+                }
             }
 
-            if (std::holds_alternative<OperationNode>(*offset)) {
-                // Indirect access
-                const std::string final_offset = code.GenerateTemporary();
-                code.AddLine("uint {} = {} >> 2;", final_offset, Visit(offset).AsUint());
+            // Indirect access
+            if (use_unified_uniforms) {
+                return {fmt::format("cbufs[{} + ({} >> 2)]", base_unified_offset,
+                                    Visit(offset).AsUint()),
+                        Type::Uint};
+            }
 
-                if (!device.HasComponentIndexingBug()) {
-                    return {fmt::format("{}[{} >> 2][{} & 3]", GetConstBuffer(cbuf->GetIndex()),
-                                        final_offset, final_offset),
-                            Type::Uint};
-                }
+            const std::string final_offset = code.GenerateTemporary();
+            code.AddLine("uint {} = {} >> 2;", final_offset, Visit(offset).AsUint());
 
-                // AMD's proprietary GLSL compiler emits ill code for variable component access.
-                // To bypass this driver bug generate 4 ifs, one per each component.
-                const std::string pack = code.GenerateTemporary();
-                code.AddLine("uvec4 {} = {}[{} >> 2];", pack, GetConstBuffer(cbuf->GetIndex()),
-                             final_offset);
-
-                const std::string result = code.GenerateTemporary();
-                code.AddLine("uint {};", result);
-                for (u32 swizzle = 0; swizzle < 4; ++swizzle) {
-                    code.AddLine("if (({} & 3) == {}) {} = {}{};", final_offset, swizzle, result,
-                                 pack, GetSwizzle(swizzle));
-                }
-                return {result, Type::Uint};
+            if (!device.HasComponentIndexingBug()) {
+                return {fmt::format("{}[{} >> 2][{} & 3]", GetConstBuffer(cbuf->GetIndex()),
+                                    final_offset, final_offset),
+                        Type::Uint};
             }
 
-            UNREACHABLE_MSG("Unmanaged offset node type");
+            // AMD's proprietary GLSL compiler emits ill code for variable component access.
+            // To bypass this driver bug generate 4 ifs, one per each component.
+            const std::string pack = code.GenerateTemporary();
+            code.AddLine("uvec4 {} = {}[{} >> 2];", pack, GetConstBuffer(cbuf->GetIndex()),
+                         final_offset);
+
+            const std::string result = code.GenerateTemporary();
+            code.AddLine("uint {};", result);
+            for (u32 swizzle = 0; swizzle < 4; ++swizzle) {
+                code.AddLine("if (({} & 3) == {}) {} = {}{};", final_offset, swizzle, result, pack,
+                             GetSwizzle(swizzle));
+            }
+            return {result, Type::Uint};
         }
 
         if (const auto gmem = std::get_if<GmemNode>(&*node)) {
@@ -1339,8 +1384,19 @@ private:
         const std::size_t count = operation.GetOperandsCount();
         const bool has_array = meta->sampler.is_array;
         const bool has_shadow = meta->sampler.is_shadow;
+        const bool workaround_lod_array_shadow_as_grad =
+            !device.HasTextureShadowLod() && function_suffix == "Lod" && meta->sampler.is_shadow &&
+            ((meta->sampler.type == TextureType::Texture2D && meta->sampler.is_array) ||
+             meta->sampler.type == TextureType::TextureCube);
+
+        std::string expr = "texture";
+
+        if (workaround_lod_array_shadow_as_grad) {
+            expr += "Grad";
+        } else {
+            expr += function_suffix;
+        }
 
-        std::string expr = "texture" + function_suffix;
         if (!meta->aoffi.empty()) {
             expr += "Offset";
         } else if (!meta->ptp.empty()) {
@@ -1374,6 +1430,16 @@ private:
             expr += ')';
         }
 
+        if (workaround_lod_array_shadow_as_grad) {
+            switch (meta->sampler.type) {
+            case TextureType::Texture2D:
+                return expr + ", vec2(0.0), vec2(0.0))";
+            case TextureType::TextureCube:
+                return expr + ", vec3(0.0), vec3(0.0))";
+            }
+            UNREACHABLE();
+        }
+
         for (const auto& variant : extras) {
             if (const auto argument = std::get_if<TextureArgument>(&variant)) {
                 expr += GenerateTextureArgument(*argument);
@@ -2000,8 +2066,19 @@ private:
         const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
         ASSERT(meta);
 
-        std::string expr = GenerateTexture(
-            operation, "Lod", {TextureArgument{Type::Float, meta->lod}, TextureOffset{}});
+        std::string expr{};
+
+        if (!device.HasTextureShadowLod() && meta->sampler.is_shadow &&
+            ((meta->sampler.type == TextureType::Texture2D && meta->sampler.is_array) ||
+             meta->sampler.type == TextureType::TextureCube)) {
+            LOG_ERROR(Render_OpenGL,
+                      "Device lacks GL_EXT_texture_shadow_lod, using textureGrad as a workaround");
+            expr = GenerateTexture(operation, "Lod", {});
+        } else {
+            expr = GenerateTexture(operation, "Lod",
+                                   {TextureArgument{Type::Float, meta->lod}, TextureOffset{}});
+        }
+
         if (meta->sampler.is_shadow) {
             expr = "vec4(" + expr + ')';
         }
@@ -2710,6 +2787,7 @@ private:
     const std::string_view identifier;
     const std::string_view suffix;
     const Header header;
+    const bool use_unified_uniforms;
     std::unordered_map<u8, VaryingTFB> transform_feedback;
 
     ShaderWriter code;
@@ -2905,7 +2983,7 @@ void GLSLDecompiler::DecompileAST() {
 
 } // Anonymous namespace
 
-ShaderEntries MakeEntries(const VideoCommon::Shader::ShaderIR& ir) {
+ShaderEntries MakeEntries(const Device& device, const ShaderIR& ir, ShaderType stage) {
     ShaderEntries entries;
     for (const auto& cbuf : ir.GetConstantBuffers()) {
         entries.const_buffers.emplace_back(cbuf.second.GetMaxOffset(), cbuf.second.IsIndirect(),
@@ -2926,6 +3004,7 @@ ShaderEntries MakeEntries(const VideoCommon::Shader::ShaderIR& ir) {
         entries.clip_distances = (clip_distances[i] ? 1U : 0U) << i;
     }
     entries.shader_length = ir.GetLength();
+    entries.use_unified_uniforms = UseUnifiedUniforms(device, ir, stage);
     return entries;
 }
 

src/video_core/renderer_opengl/gl_shader_decompiler.h

@@ -53,11 +53,13 @@ struct ShaderEntries {
     std::vector<GlobalMemoryEntry> global_memory_entries;
     std::vector<SamplerEntry> samplers;
     std::vector<ImageEntry> images;
-    u32 clip_distances{};
     std::size_t shader_length{};
+    u32 clip_distances{};
+    bool use_unified_uniforms{};
 };
 
-ShaderEntries MakeEntries(const VideoCommon::Shader::ShaderIR& ir);
+ShaderEntries MakeEntries(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
+                          Tegra::Engines::ShaderType stage);
 
 std::string DecompileShader(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
                             const VideoCommon::Shader::Registry& registry,

src/video_core/renderer_opengl/gl_shader_disk_cache.cpp

@@ -29,6 +29,8 @@ using VideoCommon::Shader::KeyMap;
 
 namespace {
 
+using VideoCommon::Shader::SeparateSamplerKey;
+
 using ShaderCacheVersionHash = std::array<u8, 64>;
 
 struct ConstBufferKey {
@@ -37,18 +39,26 @@ struct ConstBufferKey {
     u32 value = 0;
 };
 
-struct BoundSamplerKey {
+struct BoundSamplerEntry {
     u32 offset = 0;
     Tegra::Engines::SamplerDescriptor sampler;
 };
 
-struct BindlessSamplerKey {
+struct SeparateSamplerEntry {
+    u32 cbuf1 = 0;
+    u32 cbuf2 = 0;
+    u32 offset1 = 0;
+    u32 offset2 = 0;
+    Tegra::Engines::SamplerDescriptor sampler;
+};
+
+struct BindlessSamplerEntry {
     u32 cbuf = 0;
     u32 offset = 0;
     Tegra::Engines::SamplerDescriptor sampler;
 };
 
-constexpr u32 NativeVersion = 20;
+constexpr u32 NativeVersion = 21;
 
 ShaderCacheVersionHash GetShaderCacheVersionHash() {
     ShaderCacheVersionHash hash{};
@@ -87,12 +97,14 @@ bool ShaderDiskCacheEntry::Load(FileUtil::IOFile& file) {
     u32 texture_handler_size_value;
     u32 num_keys;
     u32 num_bound_samplers;
+    u32 num_separate_samplers;
     u32 num_bindless_samplers;
     if (file.ReadArray(&unique_identifier, 1) != 1 || file.ReadArray(&bound_buffer, 1) != 1 ||
         file.ReadArray(&is_texture_handler_size_known, 1) != 1 ||
         file.ReadArray(&texture_handler_size_value, 1) != 1 ||
         file.ReadArray(&graphics_info, 1) != 1 || file.ReadArray(&compute_info, 1) != 1 ||
         file.ReadArray(&num_keys, 1) != 1 || file.ReadArray(&num_bound_samplers, 1) != 1 ||
+        file.ReadArray(&num_separate_samplers, 1) != 1 ||
         file.ReadArray(&num_bindless_samplers, 1) != 1) {
         return false;
     }
@@ -101,23 +113,32 @@ bool ShaderDiskCacheEntry::Load(FileUtil::IOFile& file) {
     }
 
     std::vector<ConstBufferKey> flat_keys(num_keys);
-    std::vector<BoundSamplerKey> flat_bound_samplers(num_bound_samplers);
-    std::vector<BindlessSamplerKey> flat_bindless_samplers(num_bindless_samplers);
+    std::vector<BoundSamplerEntry> flat_bound_samplers(num_bound_samplers);
+    std::vector<SeparateSamplerEntry> flat_separate_samplers(num_separate_samplers);
+    std::vector<BindlessSamplerEntry> flat_bindless_samplers(num_bindless_samplers);
     if (file.ReadArray(flat_keys.data(), flat_keys.size()) != flat_keys.size() ||
         file.ReadArray(flat_bound_samplers.data(), flat_bound_samplers.size()) !=
             flat_bound_samplers.size() ||
+        file.ReadArray(flat_separate_samplers.data(), flat_separate_samplers.size()) !=
+            flat_separate_samplers.size() ||
         file.ReadArray(flat_bindless_samplers.data(), flat_bindless_samplers.size()) !=
             flat_bindless_samplers.size()) {
         return false;
     }
-    for (const auto& key : flat_keys) {
-        keys.insert({{key.cbuf, key.offset}, key.value});
+    for (const auto& entry : flat_keys) {
+        keys.insert({{entry.cbuf, entry.offset}, entry.value});
     }
-    for (const auto& key : flat_bound_samplers) {
-        bound_samplers.emplace(key.offset, key.sampler);
+    for (const auto& entry : flat_bound_samplers) {
+        bound_samplers.emplace(entry.offset, entry.sampler);
     }
-    for (const auto& key : flat_bindless_samplers) {
-        bindless_samplers.insert({{key.cbuf, key.offset}, key.sampler});
+    for (const auto& entry : flat_separate_samplers) {
+        SeparateSamplerKey key;
+        key.buffers = {entry.cbuf1, entry.cbuf2};
+        key.offsets = {entry.offset1, entry.offset2};
+        separate_samplers.emplace(key, entry.sampler);
+    }
+    for (const auto& entry : flat_bindless_samplers) {
+        bindless_samplers.insert({{entry.cbuf, entry.offset}, entry.sampler});
     }
 
     return true;
@@ -142,6 +163,7 @@ bool ShaderDiskCacheEntry::Save(FileUtil::IOFile& file) const {
         file.WriteObject(graphics_info) != 1 || file.WriteObject(compute_info) != 1 ||
         file.WriteObject(static_cast<u32>(keys.size())) != 1 ||
         file.WriteObject(static_cast<u32>(bound_samplers.size())) != 1 ||
+        file.WriteObject(static_cast<u32>(separate_samplers.size())) != 1 ||
         file.WriteObject(static_cast<u32>(bindless_samplers.size())) != 1) {
         return false;
     }
@@ -152,22 +174,34 @@ bool ShaderDiskCacheEntry::Save(FileUtil::IOFile& file) const {
         flat_keys.push_back(ConstBufferKey{address.first, address.second, value});
     }
 
-    std::vector<BoundSamplerKey> flat_bound_samplers;
+    std::vector<BoundSamplerEntry> flat_bound_samplers;
     flat_bound_samplers.reserve(bound_samplers.size());
     for (const auto& [address, sampler] : bound_samplers) {
-        flat_bound_samplers.push_back(BoundSamplerKey{address, sampler});
+        flat_bound_samplers.push_back(BoundSamplerEntry{address, sampler});
+    }
+
+    std::vector<SeparateSamplerEntry> flat_separate_samplers;
+    flat_separate_samplers.reserve(separate_samplers.size());
+    for (const auto& [key, sampler] : separate_samplers) {
+        SeparateSamplerEntry entry;
+        std::tie(entry.cbuf1, entry.cbuf2) = key.buffers;
+        std::tie(entry.offset1, entry.offset2) = key.offsets;
+        entry.sampler = sampler;
+        flat_separate_samplers.push_back(entry);
     }
 
-    std::vector<BindlessSamplerKey> flat_bindless_samplers;
+    std::vector<BindlessSamplerEntry> flat_bindless_samplers;
     flat_bindless_samplers.reserve(bindless_samplers.size());
     for (const auto& [address, sampler] : bindless_samplers) {
         flat_bindless_samplers.push_back(
-            BindlessSamplerKey{address.first, address.second, sampler});
+            BindlessSamplerEntry{address.first, address.second, sampler});
     }
 
     return file.WriteArray(flat_keys.data(), flat_keys.size()) == flat_keys.size() &&
            file.WriteArray(flat_bound_samplers.data(), flat_bound_samplers.size()) ==
                flat_bound_samplers.size() &&
+           file.WriteArray(flat_separate_samplers.data(), flat_separate_samplers.size()) ==
+               flat_separate_samplers.size() &&
            file.WriteArray(flat_bindless_samplers.data(), flat_bindless_samplers.size()) ==
                flat_bindless_samplers.size();
 }

src/video_core/renderer_opengl/gl_shader_disk_cache.h

@@ -57,6 +57,7 @@ struct ShaderDiskCacheEntry {
     VideoCommon::Shader::ComputeInfo compute_info;
     VideoCommon::Shader::KeyMap keys;
     VideoCommon::Shader::BoundSamplerMap bound_samplers;
+    VideoCommon::Shader::SeparateSamplerMap separate_samplers;
     VideoCommon::Shader::BindlessSamplerMap bindless_samplers;
 };
 

src/video_core/renderer_opengl/gl_stream_buffer.cpp

@@ -2,11 +2,13 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <deque>
+#include <tuple>
 #include <vector>
+
 #include "common/alignment.h"
 #include "common/assert.h"
 #include "common/microprofile.h"
+#include "video_core/renderer_opengl/gl_device.h"
 #include "video_core/renderer_opengl/gl_stream_buffer.h"
 
 MICROPROFILE_DEFINE(OpenGL_StreamBuffer, "OpenGL", "Stream Buffer Orphaning",
@@ -14,8 +16,7 @@ MICROPROFILE_DEFINE(OpenGL_StreamBuffer, "OpenGL", "Stream Buffer Orphaning",
 
 namespace OpenGL {
 
-OGLStreamBuffer::OGLStreamBuffer(GLsizeiptr size, bool vertex_data_usage, bool prefer_coherent,
-                                 bool use_persistent)
+OGLStreamBuffer::OGLStreamBuffer(const Device& device, GLsizeiptr size, bool vertex_data_usage)
     : buffer_size(size) {
     gl_buffer.Create();
 
@@ -29,34 +30,22 @@ OGLStreamBuffer::OGLStreamBuffer(GLsizeiptr size, bool vertex_data_usage, bool p
         allocate_size *= 2;
     }
 
-    if (use_persistent) {
-        persistent = true;
-        coherent = prefer_coherent;
-        const GLbitfield flags =
-            GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT | (coherent ? GL_MAP_COHERENT_BIT : 0);
-        glNamedBufferStorage(gl_buffer.handle, allocate_size, nullptr, flags);
-        mapped_ptr = static_cast<u8*>(glMapNamedBufferRange(
-            gl_buffer.handle, 0, buffer_size, flags | (coherent ? 0 : GL_MAP_FLUSH_EXPLICIT_BIT)));
-    } else {
-        glNamedBufferData(gl_buffer.handle, allocate_size, nullptr, GL_STREAM_DRAW);
+    static constexpr GLbitfield flags = GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT;
+    glNamedBufferStorage(gl_buffer.handle, allocate_size, nullptr, flags);
+    mapped_ptr = static_cast<u8*>(
+        glMapNamedBufferRange(gl_buffer.handle, 0, buffer_size, flags | GL_MAP_FLUSH_EXPLICIT_BIT));
+
+    if (device.HasVertexBufferUnifiedMemory()) {
+        glMakeNamedBufferResidentNV(gl_buffer.handle, GL_READ_ONLY);
+        glGetNamedBufferParameterui64vNV(gl_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV, &gpu_address);
     }
 }
 
 OGLStreamBuffer::~OGLStreamBuffer() {
-    if (persistent) {
-        glUnmapNamedBuffer(gl_buffer.handle);
-    }
+    glUnmapNamedBuffer(gl_buffer.handle);
     gl_buffer.Release();
 }
 
-GLuint OGLStreamBuffer::GetHandle() const {
-    return gl_buffer.handle;
-}
-
-GLsizeiptr OGLStreamBuffer::GetSize() const {
-    return buffer_size;
-}
-
 std::tuple<u8*, GLintptr, bool> OGLStreamBuffer::Map(GLsizeiptr size, GLintptr alignment) {
     ASSERT(size <= buffer_size);
     ASSERT(alignment <= buffer_size);
@@ -68,36 +57,21 @@ std::tuple<u8*, GLintptr, bool> OGLStreamBuffer::Map(GLsizeiptr size, GLintptr a
 
     bool invalidate = false;
     if (buffer_pos + size > buffer_size) {
+        MICROPROFILE_SCOPE(OpenGL_StreamBuffer);
+        glInvalidateBufferData(gl_buffer.handle);
+
         buffer_pos = 0;
         invalidate = true;
-
-        if (persistent) {
-            glUnmapNamedBuffer(gl_buffer.handle);
-        }
     }
 
-    if (invalidate || !persistent) {
-        MICROPROFILE_SCOPE(OpenGL_StreamBuffer);
-        GLbitfield flags = GL_MAP_WRITE_BIT | (persistent ? GL_MAP_PERSISTENT_BIT : 0) |
-                           (coherent ? GL_MAP_COHERENT_BIT : GL_MAP_FLUSH_EXPLICIT_BIT) |
-                           (invalidate ? GL_MAP_INVALIDATE_BUFFER_BIT : GL_MAP_UNSYNCHRONIZED_BIT);
-        mapped_ptr = static_cast<u8*>(
-            glMapNamedBufferRange(gl_buffer.handle, buffer_pos, buffer_size - buffer_pos, flags));
-        mapped_offset = buffer_pos;
-    }
-
-    return std::make_tuple(mapped_ptr + buffer_pos - mapped_offset, buffer_pos, invalidate);
+    return std::make_tuple(mapped_ptr + buffer_pos, buffer_pos, invalidate);
 }
 
 void OGLStreamBuffer::Unmap(GLsizeiptr size) {
     ASSERT(size <= mapped_size);
 
-    if (!coherent && size > 0) {
-        glFlushMappedNamedBufferRange(gl_buffer.handle, buffer_pos - mapped_offset, size);
-    }
-
-    if (!persistent) {
-        glUnmapNamedBuffer(gl_buffer.handle);
+    if (size > 0) {
+        glFlushMappedNamedBufferRange(gl_buffer.handle, buffer_pos, size);
     }
 
     buffer_pos += size;

src/video_core/renderer_opengl/gl_stream_buffer.h

@@ -11,15 +11,13 @@
 
 namespace OpenGL {
 
+class Device;
+
 class OGLStreamBuffer : private NonCopyable {
 public:
-    explicit OGLStreamBuffer(GLsizeiptr size, bool vertex_data_usage, bool prefer_coherent = false,
-                             bool use_persistent = true);
+    explicit OGLStreamBuffer(const Device& device, GLsizeiptr size, bool vertex_data_usage);
     ~OGLStreamBuffer();
 
-    GLuint GetHandle() const;
-    GLsizeiptr GetSize() const;
-
     /*
      * Allocates a linear chunk of memory in the GPU buffer with at least "size" bytes
      * and the optional alignment requirement.
@@ -32,15 +30,24 @@ public:
 
     void Unmap(GLsizeiptr size);
 
+    GLuint Handle() const {
+        return gl_buffer.handle;
+    }
+
+    u64 Address() const {
+        return gpu_address;
+    }
+
+    GLsizeiptr Size() const noexcept {
+        return buffer_size;
+    }
+
 private:
     OGLBuffer gl_buffer;
 
-    bool coherent = false;
-    bool persistent = false;
-
+    GLuint64EXT gpu_address = 0;
     GLintptr buffer_pos = 0;
     GLsizeiptr buffer_size = 0;
-    GLintptr mapped_offset = 0;
     GLsizeiptr mapped_size = 0;
     u8* mapped_ptr = nullptr;
 };

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -263,9 +263,14 @@ CachedSurface::CachedSurface(const GPUVAddr gpu_addr, const SurfaceParams& param
     target = GetTextureTarget(params.target);
     texture = CreateTexture(params, target, internal_format, texture_buffer);
     DecorateSurfaceName();
-    main_view = CreateViewInner(
-        ViewParams(params.target, 0, params.is_layered ? params.depth : 1, 0, params.num_levels),
-        true);
+
+    u32 num_layers = 1;
+    if (params.is_layered || params.target == SurfaceTarget::Texture3D) {
+        num_layers = params.depth;
+    }
+
+    main_view =
+        CreateViewInner(ViewParams(params.target, 0, num_layers, 0, params.num_levels), true);
 }
 
 CachedSurface::~CachedSurface() = default;
@@ -404,8 +409,7 @@ View CachedSurface::CreateViewInner(const ViewParams& view_key, const bool is_pr
 
 CachedSurfaceView::CachedSurfaceView(CachedSurface& surface, const ViewParams& params,
                                      bool is_proxy)
-    : VideoCommon::ViewBase(params), surface{surface},
-      format{GetFormatTuple(surface.GetSurfaceParams().pixel_format).internal_format},
+    : VideoCommon::ViewBase(params), surface{surface}, format{surface.internal_format},
       target{GetTextureTarget(params.target)}, is_proxy{is_proxy} {
     if (!is_proxy) {
         main_view = CreateTextureView();
@@ -414,20 +418,23 @@ CachedSurfaceView::CachedSurfaceView(CachedSurface& surface, const ViewParams& p
 
 CachedSurfaceView::~CachedSurfaceView() = default;
 
-void CachedSurfaceView::Attach(GLenum attachment, GLenum target) const {
+void CachedSurfaceView::Attach(GLenum attachment, GLenum fb_target) const {
     ASSERT(params.num_levels == 1);
 
+    if (params.target == SurfaceTarget::Texture3D) {
+        if (params.num_layers > 1) {
+            ASSERT(params.base_layer == 0);
+            glFramebufferTexture(fb_target, attachment, surface.texture.handle, params.base_level);
+        } else {
+            glFramebufferTexture3D(fb_target, attachment, target, surface.texture.handle,
+                                   params.base_level, params.base_layer);
+        }
+        return;
+    }
+
     if (params.num_layers > 1) {
-        // Layered framebuffer attachments
         UNIMPLEMENTED_IF(params.base_layer != 0);
-
-        switch (params.target) {
-        case SurfaceTarget::Texture2DArray:
-            glFramebufferTexture(target, attachment, GetTexture(), 0);
-            break;
-        default:
-            UNIMPLEMENTED();
-        }
+        glFramebufferTexture(fb_target, attachment, GetTexture(), 0);
         return;
     }
 
@@ -435,16 +442,16 @@ void CachedSurfaceView::Attach(GLenum attachment, GLenum target) const {
     const GLuint texture = surface.GetTexture();
     switch (surface.GetSurfaceParams().target) {
     case SurfaceTarget::Texture1D:
-        glFramebufferTexture1D(target, attachment, view_target, texture, params.base_level);
+        glFramebufferTexture1D(fb_target, attachment, view_target, texture, params.base_level);
         break;
     case SurfaceTarget::Texture2D:
-        glFramebufferTexture2D(target, attachment, view_target, texture, params.base_level);
+        glFramebufferTexture2D(fb_target, attachment, view_target, texture, params.base_level);
         break;
     case SurfaceTarget::Texture1DArray:
     case SurfaceTarget::Texture2DArray:
     case SurfaceTarget::TextureCubemap:
     case SurfaceTarget::TextureCubeArray:
-        glFramebufferTextureLayer(target, attachment, texture, params.base_level,
+        glFramebufferTextureLayer(fb_target, attachment, texture, params.base_level,
                                   params.base_layer);
         break;
     default:
@@ -501,8 +508,13 @@ OGLTextureView CachedSurfaceView::CreateTextureView() const {
     OGLTextureView texture_view;
     texture_view.Create();
 
-    glTextureView(texture_view.handle, target, surface.texture.handle, format, params.base_level,
-                  params.num_levels, params.base_layer, params.num_layers);
+    if (target == GL_TEXTURE_3D) {
+        glTextureView(texture_view.handle, target, surface.texture.handle, format,
+                      params.base_level, params.num_levels, 0, 1);
+    } else {
+        glTextureView(texture_view.handle, target, surface.texture.handle, format,
+                      params.base_level, params.num_levels, params.base_layer, params.num_layers);
+    }
     ApplyTextureDefaults(surface.GetSurfaceParams(), texture_view.handle);
 
     return texture_view;
@@ -545,8 +557,8 @@ void TextureCacheOpenGL::ImageBlit(View& src_view, View& dst_view,
                                    const Tegra::Engines::Fermi2D::Config& copy_config) {
     const auto& src_params{src_view->GetSurfaceParams()};
     const auto& dst_params{dst_view->GetSurfaceParams()};
-    UNIMPLEMENTED_IF(src_params.target == SurfaceTarget::Texture3D);
-    UNIMPLEMENTED_IF(dst_params.target == SurfaceTarget::Texture3D);
+    UNIMPLEMENTED_IF(src_params.depth != 1);
+    UNIMPLEMENTED_IF(dst_params.depth != 1);
 
     state_tracker.NotifyScissor0();
     state_tracker.NotifyFramebuffer();

src/video_core/renderer_opengl/gl_texture_cache.h

@@ -80,8 +80,10 @@ public:
     explicit CachedSurfaceView(CachedSurface& surface, const ViewParams& params, bool is_proxy);
     ~CachedSurfaceView();
 
-    /// Attaches this texture view to the current bound GL_DRAW_FRAMEBUFFER
-    void Attach(GLenum attachment, GLenum target) const;
+    /// @brief Attaches this texture view to the currently bound fb_target framebuffer
+    /// @param attachment   Attachment to bind textures to
+    /// @param fb_target    Framebuffer target to attach to (e.g. DRAW_FRAMEBUFFER)
+    void Attach(GLenum attachment, GLenum fb_target) const;
 
     GLuint GetTexture(Tegra::Texture::SwizzleSource x_source,
                       Tegra::Texture::SwizzleSource y_source,

src/video_core/renderer_opengl/maxwell_to_gl.h

@@ -46,10 +46,8 @@ inline GLenum VertexType(Maxwell::VertexAttribute attrib) {
             return GL_UNSIGNED_INT;
         case Maxwell::VertexAttribute::Size::Size_10_10_10_2:
             return GL_UNSIGNED_INT_2_10_10_10_REV;
-        default:
-            LOG_ERROR(Render_OpenGL, "Unimplemented vertex size={}", attrib.SizeString());
-            return {};
         }
+        break;
     case Maxwell::VertexAttribute::Type::SignedInt:
     case Maxwell::VertexAttribute::Type::SignedNorm:
         switch (attrib.size) {
@@ -70,10 +68,8 @@ inline GLenum VertexType(Maxwell::VertexAttribute attrib) {
             return GL_INT;
         case Maxwell::VertexAttribute::Size::Size_10_10_10_2:
             return GL_INT_2_10_10_10_REV;
-        default:
-            LOG_ERROR(Render_OpenGL, "Unimplemented vertex size={}", attrib.SizeString());
-            return {};
         }
+        break;
     case Maxwell::VertexAttribute::Type::Float:
         switch (attrib.size) {
         case Maxwell::VertexAttribute::Size::Size_16:
@@ -86,10 +82,8 @@ inline GLenum VertexType(Maxwell::VertexAttribute attrib) {
         case Maxwell::VertexAttribute::Size::Size_32_32_32:
         case Maxwell::VertexAttribute::Size::Size_32_32_32_32:
             return GL_FLOAT;
-        default:
-            LOG_ERROR(Render_OpenGL, "Unimplemented vertex size={}", attrib.SizeString());
-            return {};
         }
+        break;
     case Maxwell::VertexAttribute::Type::UnsignedScaled:
         switch (attrib.size) {
         case Maxwell::VertexAttribute::Size::Size_8:
@@ -102,10 +96,8 @@ inline GLenum VertexType(Maxwell::VertexAttribute attrib) {
         case Maxwell::VertexAttribute::Size::Size_16_16_16:
         case Maxwell::VertexAttribute::Size::Size_16_16_16_16:
             return GL_UNSIGNED_SHORT;
-        default:
-            LOG_ERROR(Render_OpenGL, "Unimplemented vertex size={}", attrib.SizeString());
-            return {};
         }
+        break;
     case Maxwell::VertexAttribute::Type::SignedScaled:
         switch (attrib.size) {
         case Maxwell::VertexAttribute::Size::Size_8:
@@ -118,14 +110,12 @@ inline GLenum VertexType(Maxwell::VertexAttribute attrib) {
         case Maxwell::VertexAttribute::Size::Size_16_16_16:
         case Maxwell::VertexAttribute::Size::Size_16_16_16_16:
             return GL_SHORT;
-        default:
-            LOG_ERROR(Render_OpenGL, "Unimplemented vertex size={}", attrib.SizeString());
-            return {};
         }
-    default:
-        LOG_ERROR(Render_OpenGL, "Unimplemented vertex type={}", attrib.TypeString());
-        return {};
+        break;
     }
+    UNIMPLEMENTED_MSG("Unimplemented vertex type={} and size={}", attrib.TypeString(),
+                      attrib.SizeString());
+    return {};
 }
 
 inline GLenum IndexFormat(Maxwell::IndexFormat index_format) {
@@ -137,8 +127,7 @@ inline GLenum IndexFormat(Maxwell::IndexFormat index_format) {
     case Maxwell::IndexFormat::UnsignedInt:
         return GL_UNSIGNED_INT;
     }
-    LOG_CRITICAL(Render_OpenGL, "Unimplemented index_format={}", static_cast<u32>(index_format));
-    UNREACHABLE();
+    UNREACHABLE_MSG("Invalid index_format={}", static_cast<u32>(index_format));
     return {};
 }
 
@@ -180,33 +169,32 @@ inline GLenum PrimitiveTopology(Maxwell::PrimitiveTopology topology) {
 }
 
 inline GLenum TextureFilterMode(Tegra::Texture::TextureFilter filter_mode,
-                                Tegra::Texture::TextureMipmapFilter mip_filter_mode) {
+                                Tegra::Texture::TextureMipmapFilter mipmap_filter_mode) {
     switch (filter_mode) {
-    case Tegra::Texture::TextureFilter::Linear: {
-        switch (mip_filter_mode) {
+    case Tegra::Texture::TextureFilter::Nearest:
+        switch (mipmap_filter_mode) {
         case Tegra::Texture::TextureMipmapFilter::None:
-            return GL_LINEAR;
+            return GL_NEAREST;
         case Tegra::Texture::TextureMipmapFilter::Nearest:
-            return GL_LINEAR_MIPMAP_NEAREST;
+            return GL_NEAREST_MIPMAP_NEAREST;
         case Tegra::Texture::TextureMipmapFilter::Linear:
-            return GL_LINEAR_MIPMAP_LINEAR;
+            return GL_NEAREST_MIPMAP_LINEAR;
         }
         break;
-    }
-    case Tegra::Texture::TextureFilter::Nearest: {
-        switch (mip_filter_mode) {
+    case Tegra::Texture::TextureFilter::Linear:
+        switch (mipmap_filter_mode) {
         case Tegra::Texture::TextureMipmapFilter::None:
-            return GL_NEAREST;
+            return GL_LINEAR;
         case Tegra::Texture::TextureMipmapFilter::Nearest:
-            return GL_NEAREST_MIPMAP_NEAREST;
+            return GL_LINEAR_MIPMAP_NEAREST;
         case Tegra::Texture::TextureMipmapFilter::Linear:
-            return GL_NEAREST_MIPMAP_LINEAR;
+            return GL_LINEAR_MIPMAP_LINEAR;
         }
         break;
     }
-    }
-    LOG_ERROR(Render_OpenGL, "Unimplemented texture filter mode={}", static_cast<u32>(filter_mode));
-    return GL_LINEAR;
+    UNREACHABLE_MSG("Invalid texture filter mode={} and mipmap filter mode={}",
+                    static_cast<u32>(filter_mode), static_cast<u32>(mipmap_filter_mode));
+    return GL_NEAREST;
 }
 
 inline GLenum WrapMode(Tegra::Texture::WrapMode wrap_mode) {
@@ -229,10 +217,9 @@ inline GLenum WrapMode(Tegra::Texture::WrapMode wrap_mode) {
         } else {
             return GL_MIRROR_CLAMP_TO_EDGE;
         }
-    default:
-        LOG_ERROR(Render_OpenGL, "Unimplemented texture wrap mode={}", static_cast<u32>(wrap_mode));
-        return GL_REPEAT;
     }
+    UNIMPLEMENTED_MSG("Unimplemented texture wrap mode={}", static_cast<u32>(wrap_mode));
+    return GL_REPEAT;
 }
 
 inline GLenum DepthCompareFunc(Tegra::Texture::DepthCompareFunc func) {
@@ -254,8 +241,7 @@ inline GLenum DepthCompareFunc(Tegra::Texture::DepthCompareFunc func) {
     case Tegra::Texture::DepthCompareFunc::Always:
         return GL_ALWAYS;
     }
-    LOG_ERROR(Render_OpenGL, "Unimplemented texture depth compare function ={}",
-              static_cast<u32>(func));
+    UNIMPLEMENTED_MSG("Unimplemented texture depth compare function={}", static_cast<u32>(func));
     return GL_GREATER;
 }
 
@@ -277,7 +263,7 @@ inline GLenum BlendEquation(Maxwell::Blend::Equation equation) {
     case Maxwell::Blend::Equation::MaxGL:
         return GL_MAX;
     }
-    LOG_ERROR(Render_OpenGL, "Unimplemented blend equation={}", static_cast<u32>(equation));
+    UNIMPLEMENTED_MSG("Unimplemented blend equation={}", static_cast<u32>(equation));
     return GL_FUNC_ADD;
 }
 
@@ -341,7 +327,7 @@ inline GLenum BlendFunc(Maxwell::Blend::Factor factor) {
     case Maxwell::Blend::Factor::OneMinusConstantAlphaGL:
         return GL_ONE_MINUS_CONSTANT_ALPHA;
     }
-    LOG_ERROR(Render_OpenGL, "Unimplemented blend factor={}", static_cast<u32>(factor));
+    UNIMPLEMENTED_MSG("Unimplemented blend factor={}", static_cast<u32>(factor));
     return GL_ZERO;
 }
 
@@ -361,7 +347,7 @@ inline GLenum SwizzleSource(Tegra::Texture::SwizzleSource source) {
     case Tegra::Texture::SwizzleSource::OneFloat:
         return GL_ONE;
     }
-    LOG_ERROR(Render_OpenGL, "Unimplemented swizzle source={}", static_cast<u32>(source));
+    UNIMPLEMENTED_MSG("Unimplemented swizzle source={}", static_cast<u32>(source));
     return GL_ZERO;
 }
 
@@ -392,7 +378,7 @@ inline GLenum ComparisonOp(Maxwell::ComparisonOp comparison) {
     case Maxwell::ComparisonOp::AlwaysOld:
         return GL_ALWAYS;
     }
-    LOG_ERROR(Render_OpenGL, "Unimplemented comparison op={}", static_cast<u32>(comparison));
+    UNIMPLEMENTED_MSG("Unimplemented comparison op={}", static_cast<u32>(comparison));
     return GL_ALWAYS;
 }
 
@@ -423,7 +409,7 @@ inline GLenum StencilOp(Maxwell::StencilOp stencil) {
     case Maxwell::StencilOp::DecrWrapOGL:
         return GL_DECR_WRAP;
     }
-    LOG_ERROR(Render_OpenGL, "Unimplemented stencil op={}", static_cast<u32>(stencil));
+    UNIMPLEMENTED_MSG("Unimplemented stencil op={}", static_cast<u32>(stencil));
     return GL_KEEP;
 }
 
@@ -434,7 +420,7 @@ inline GLenum FrontFace(Maxwell::FrontFace front_face) {
     case Maxwell::FrontFace::CounterClockWise:
         return GL_CCW;
     }
-    LOG_ERROR(Render_OpenGL, "Unimplemented front face cull={}", static_cast<u32>(front_face));
+    UNIMPLEMENTED_MSG("Unimplemented front face cull={}", static_cast<u32>(front_face));
     return GL_CCW;
 }
 
@@ -447,7 +433,7 @@ inline GLenum CullFace(Maxwell::CullFace cull_face) {
     case Maxwell::CullFace::FrontAndBack:
         return GL_FRONT_AND_BACK;
     }
-    LOG_ERROR(Render_OpenGL, "Unimplemented cull face={}", static_cast<u32>(cull_face));
+    UNIMPLEMENTED_MSG("Unimplemented cull face={}", static_cast<u32>(cull_face));
     return GL_BACK;
 }
 
@@ -486,7 +472,7 @@ inline GLenum LogicOp(Maxwell::LogicOperation operation) {
     case Maxwell::LogicOperation::Set:
         return GL_SET;
     }
-    LOG_ERROR(Render_OpenGL, "Unimplemented logic operation={}", static_cast<u32>(operation));
+    UNIMPLEMENTED_MSG("Unimplemented logic operation={}", static_cast<u32>(operation));
     return GL_COPY;
 }
 

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -488,6 +488,15 @@ void RendererOpenGL::InitOpenGLObjects() {
 
     // Clear screen to black
     LoadColorToActiveGLTexture(0, 0, 0, 0, screen_info.texture);
+
+    // Enable unified vertex attributes and query vertex buffer address when the driver supports it
+    if (device.HasVertexBufferUnifiedMemory()) {
+        glEnableClientState(GL_VERTEX_ATTRIB_ARRAY_UNIFIED_NV);
+
+        glMakeNamedBufferResidentNV(vertex_buffer.handle, GL_READ_ONLY);
+        glGetNamedBufferParameterui64vNV(vertex_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV,
+                                         &vertex_buffer_address);
+    }
 }
 
 void RendererOpenGL::AddTelemetryFields() {
@@ -656,7 +665,13 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) {
                          offsetof(ScreenRectVertex, tex_coord));
     glVertexAttribBinding(PositionLocation, 0);
     glVertexAttribBinding(TexCoordLocation, 0);
-    glBindVertexBuffer(0, vertex_buffer.handle, 0, sizeof(ScreenRectVertex));
+    if (device.HasVertexBufferUnifiedMemory()) {
+        glBindVertexBuffer(0, 0, 0, sizeof(ScreenRectVertex));
+        glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, 0, vertex_buffer_address,
+                               sizeof(vertices));
+    } else {
+        glBindVertexBuffer(0, vertex_buffer.handle, 0, sizeof(ScreenRectVertex));
+    }
 
     glBindTextureUnit(0, screen_info.display_texture);
     glBindSampler(0, 0);
@@ -751,8 +766,9 @@ void RendererOpenGL::RenderScreenshot() {
 }
 
 bool RendererOpenGL::Init() {
-    if (GLAD_GL_KHR_debug) {
+    if (Settings::values.renderer_debug && GLAD_GL_KHR_debug) {
         glEnable(GL_DEBUG_OUTPUT);
+        glEnable(GL_DEBUG_OUTPUT_SYNCHRONOUS);
         glDebugMessageCallback(DebugHandler, nullptr);
     }
 

src/video_core/renderer_opengl/renderer_opengl.h

@@ -107,6 +107,9 @@ private:
     OGLPipeline pipeline;
     OGLFramebuffer screenshot_framebuffer;
 
+    // GPU address of the vertex buffer
+    GLuint64EXT vertex_buffer_address = 0;
+
     /// Display information for Switch screen
     ScreenInfo screen_info;
 

src/video_core/renderer_vulkan/fixed_pipeline_state.cpp

@@ -71,8 +71,7 @@ void FixedPipelineState::Rasterizer::Fill(const Maxwell& regs) noexcept {
     const u32 topology_index = static_cast<u32>(regs.draw.topology.Value());
 
     u32 packed_front_face = PackFrontFace(regs.front_face);
-    if (regs.screen_y_control.triangle_rast_flip != 0 &&
-        regs.viewport_transform[0].scale_y > 0.0f) {
+    if (regs.screen_y_control.triangle_rast_flip != 0) {
         // Flip front face
         packed_front_face = 1 - packed_front_face;
     }

src/video_core/renderer_vulkan/maxwell_to_vk.cpp

@@ -21,29 +21,29 @@ namespace Sampler {
 
 VkFilter Filter(Tegra::Texture::TextureFilter filter) {
     switch (filter) {
-    case Tegra::Texture::TextureFilter::Linear:
-        return VK_FILTER_LINEAR;
     case Tegra::Texture::TextureFilter::Nearest:
         return VK_FILTER_NEAREST;
+    case Tegra::Texture::TextureFilter::Linear:
+        return VK_FILTER_LINEAR;
     }
-    UNIMPLEMENTED_MSG("Unimplemented sampler filter={}", static_cast<u32>(filter));
+    UNREACHABLE_MSG("Invalid sampler filter={}", static_cast<u32>(filter));
     return {};
 }
 
 VkSamplerMipmapMode MipmapMode(Tegra::Texture::TextureMipmapFilter mipmap_filter) {
     switch (mipmap_filter) {
     case Tegra::Texture::TextureMipmapFilter::None:
-        // TODO(Rodrigo): None seems to be mapped to OpenGL's mag and min filters without mipmapping
-        // (e.g. GL_NEAREST and GL_LINEAR). Vulkan doesn't have such a thing, find out if we have to
-        // use an image view with a single mipmap level to emulate this.
-        return VK_SAMPLER_MIPMAP_MODE_LINEAR;
-        ;
-    case Tegra::Texture::TextureMipmapFilter::Linear:
-        return VK_SAMPLER_MIPMAP_MODE_LINEAR;
+        // There are no Vulkan filter modes that directly correspond to OpenGL minification filters
+        // of GL_LINEAR or GL_NEAREST, but they can be emulated using
+        // VK_SAMPLER_MIPMAP_MODE_NEAREST, minLod = 0, and maxLod = 0.25, and using minFilter =
+        // VK_FILTER_LINEAR or minFilter = VK_FILTER_NEAREST, respectively.
+        return VK_SAMPLER_MIPMAP_MODE_NEAREST;
     case Tegra::Texture::TextureMipmapFilter::Nearest:
         return VK_SAMPLER_MIPMAP_MODE_NEAREST;
+    case Tegra::Texture::TextureMipmapFilter::Linear:
+        return VK_SAMPLER_MIPMAP_MODE_LINEAR;
     }
-    UNIMPLEMENTED_MSG("Unimplemented sampler mipmap mode={}", static_cast<u32>(mipmap_filter));
+    UNREACHABLE_MSG("Invalid sampler mipmap mode={}", static_cast<u32>(mipmap_filter));
     return {};
 }
 
@@ -78,10 +78,9 @@ VkSamplerAddressMode WrapMode(const VKDevice& device, Tegra::Texture::WrapMode w
     case Tegra::Texture::WrapMode::MirrorOnceBorder:
         UNIMPLEMENTED();
         return VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE;
-    default:
-        UNIMPLEMENTED_MSG("Unimplemented wrap mode={}", static_cast<u32>(wrap_mode));
-        return {};
     }
+    UNIMPLEMENTED_MSG("Unimplemented wrap mode={}", static_cast<u32>(wrap_mode));
+    return {};
 }
 
 VkCompareOp DepthCompareFunction(Tegra::Texture::DepthCompareFunc depth_compare_func) {
@@ -149,7 +148,7 @@ struct FormatTuple {
     {VK_FORMAT_R16_SFLOAT, Attachable | Storage},               // R16F
     {VK_FORMAT_R16_UNORM, Attachable | Storage},                // R16U
     {VK_FORMAT_UNDEFINED},                                      // R16S
-    {VK_FORMAT_UNDEFINED},                                      // R16UI
+    {VK_FORMAT_R16_UINT, Attachable | Storage},                 // R16UI
     {VK_FORMAT_UNDEFINED},                                      // R16I
     {VK_FORMAT_R16G16_UNORM, Attachable | Storage},             // RG16
     {VK_FORMAT_R16G16_SFLOAT, Attachable | Storage},            // RG16F
@@ -288,10 +287,9 @@ VkPrimitiveTopology PrimitiveTopology([[maybe_unused]] const VKDevice& device,
         return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
     case Maxwell::PrimitiveTopology::Patches:
         return VK_PRIMITIVE_TOPOLOGY_PATCH_LIST;
-    default:
-        UNIMPLEMENTED_MSG("Unimplemented topology={}", static_cast<u32>(topology));
-        return {};
     }
+    UNIMPLEMENTED_MSG("Unimplemented topology={}", static_cast<u32>(topology));
+    return {};
 }
 
 VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttribute::Size size) {

src/video_core/renderer_vulkan/renderer_vulkan.cpp

@@ -13,6 +13,7 @@
 #include <fmt/format.h>
 
 #include "common/dynamic_library.h"
+#include "common/file_util.h"
 #include "common/logging/log.h"
 #include "common/telemetry.h"
 #include "core/core.h"
@@ -76,7 +77,8 @@ Common::DynamicLibrary OpenVulkanLibrary() {
     char* libvulkan_env = getenv("LIBVULKAN_PATH");
     if (!libvulkan_env || !library.Open(libvulkan_env)) {
         // Use the libvulkan.dylib from the application bundle.
-        std::string filename = File::GetBundleDirectory() + "/Contents/Frameworks/libvulkan.dylib";
+        const std::string filename =
+            FileUtil::GetBundleDirectory() + "/Contents/Frameworks/libvulkan.dylib";
         library.Open(filename.c_str());
     }
 #else

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -37,9 +37,9 @@ std::unique_ptr<VKStreamBuffer> CreateStreamBuffer(const VKDevice& device, VKSch
 
 } // Anonymous namespace
 
-CachedBufferBlock::CachedBufferBlock(const VKDevice& device, VKMemoryManager& memory_manager,
-                                     VAddr cpu_addr, std::size_t size)
-    : VideoCommon::BufferBlock{cpu_addr, size} {
+Buffer::Buffer(const VKDevice& device, VKMemoryManager& memory_manager, VKScheduler& scheduler_,
+               VKStagingBufferPool& staging_pool_, VAddr cpu_addr, std::size_t size)
+    : VideoCommon::BufferBlock{cpu_addr, size}, scheduler{scheduler_}, staging_pool{staging_pool_} {
     VkBufferCreateInfo ci;
     ci.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
     ci.pNext = nullptr;
@@ -54,46 +54,17 @@ CachedBufferBlock::CachedBufferBlock(const VKDevice& device, VKMemoryManager& me
     buffer.commit = memory_manager.Commit(buffer.handle, false);
 }
 
-CachedBufferBlock::~CachedBufferBlock() = default;
+Buffer::~Buffer() = default;
 
-VKBufferCache::VKBufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system,
-                             const VKDevice& device, VKMemoryManager& memory_manager,
-                             VKScheduler& scheduler, VKStagingBufferPool& staging_pool)
-    : VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer>{rasterizer, system,
-                                                                 CreateStreamBuffer(device,
-                                                                                    scheduler)},
-      device{device}, memory_manager{memory_manager}, scheduler{scheduler}, staging_pool{
-                                                                                staging_pool} {}
-
-VKBufferCache::~VKBufferCache() = default;
-
-Buffer VKBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) {
-    return std::make_shared<CachedBufferBlock>(device, memory_manager, cpu_addr, size);
-}
-
-VkBuffer VKBufferCache::ToHandle(const Buffer& buffer) {
-    return buffer->GetHandle();
-}
-
-VkBuffer VKBufferCache::GetEmptyBuffer(std::size_t size) {
-    size = std::max(size, std::size_t(4));
-    const auto& empty = staging_pool.GetUnusedBuffer(size, false);
-    scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([size, buffer = *empty.handle](vk::CommandBuffer cmdbuf) {
-        cmdbuf.FillBuffer(buffer, 0, size, 0);
-    });
-    return *empty.handle;
-}
-
-void VKBufferCache::UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
-                                    const u8* data) {
+void Buffer::Upload(std::size_t offset, std::size_t size, const u8* data) const {
     const auto& staging = staging_pool.GetUnusedBuffer(size, true);
     std::memcpy(staging.commit->Map(size), data, size);
 
     scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([staging = *staging.handle, buffer = buffer->GetHandle(), offset,
-                      size](vk::CommandBuffer cmdbuf) {
-        cmdbuf.CopyBuffer(staging, buffer, VkBufferCopy{0, offset, size});
+
+    const VkBuffer handle = Handle();
+    scheduler.Record([staging = *staging.handle, handle, offset, size](vk::CommandBuffer cmdbuf) {
+        cmdbuf.CopyBuffer(staging, handle, VkBufferCopy{0, offset, size});
 
         VkBufferMemoryBarrier barrier;
         barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
@@ -102,7 +73,7 @@ void VKBufferCache::UploadBlockData(const Buffer& buffer, std::size_t offset, st
         barrier.dstAccessMask = UPLOAD_ACCESS_BARRIERS;
         barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
         barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
-        barrier.buffer = buffer;
+        barrier.buffer = handle;
         barrier.offset = offset;
         barrier.size = size;
         cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, UPLOAD_PIPELINE_STAGE, 0, {},
@@ -110,12 +81,12 @@ void VKBufferCache::UploadBlockData(const Buffer& buffer, std::size_t offset, st
     });
 }
 
-void VKBufferCache::DownloadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
-                                      u8* data) {
+void Buffer::Download(std::size_t offset, std::size_t size, u8* data) const {
     const auto& staging = staging_pool.GetUnusedBuffer(size, true);
     scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([staging = *staging.handle, buffer = buffer->GetHandle(), offset,
-                      size](vk::CommandBuffer cmdbuf) {
+
+    const VkBuffer handle = Handle();
+    scheduler.Record([staging = *staging.handle, handle, offset, size](vk::CommandBuffer cmdbuf) {
         VkBufferMemoryBarrier barrier;
         barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
         barrier.pNext = nullptr;
@@ -123,7 +94,7 @@ void VKBufferCache::DownloadBlockData(const Buffer& buffer, std::size_t offset,
         barrier.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
         barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
         barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
-        barrier.buffer = buffer;
+        barrier.buffer = handle;
         barrier.offset = offset;
         barrier.size = size;
 
@@ -131,18 +102,20 @@ void VKBufferCache::DownloadBlockData(const Buffer& buffer, std::size_t offset,
                                    VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
                                    VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                VK_PIPELINE_STAGE_TRANSFER_BIT, 0, {}, barrier, {});
-        cmdbuf.CopyBuffer(buffer, staging, VkBufferCopy{offset, 0, size});
+        cmdbuf.CopyBuffer(handle, staging, VkBufferCopy{offset, 0, size});
     });
     scheduler.Finish();
 
     std::memcpy(data, staging.commit->Map(size), size);
 }
 
-void VKBufferCache::CopyBlock(const Buffer& src, const Buffer& dst, std::size_t src_offset,
-                              std::size_t dst_offset, std::size_t size) {
+void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset,
+                      std::size_t size) const {
     scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([src_buffer = src->GetHandle(), dst_buffer = dst->GetHandle(), src_offset,
-                      dst_offset, size](vk::CommandBuffer cmdbuf) {
+
+    const VkBuffer dst_buffer = Handle();
+    scheduler.Record([src_buffer = src.Handle(), dst_buffer, src_offset, dst_offset,
+                      size](vk::CommandBuffer cmdbuf) {
         cmdbuf.CopyBuffer(src_buffer, dst_buffer, VkBufferCopy{src_offset, dst_offset, size});
 
         std::array<VkBufferMemoryBarrier, 2> barriers;
@@ -169,4 +142,30 @@ void VKBufferCache::CopyBlock(const Buffer& src, const Buffer& dst, std::size_t
     });
 }
 
+VKBufferCache::VKBufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system,
+                             const VKDevice& device, VKMemoryManager& memory_manager,
+                             VKScheduler& scheduler, VKStagingBufferPool& staging_pool)
+    : VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer>{rasterizer, system,
+                                                                 CreateStreamBuffer(device,
+                                                                                    scheduler)},
+      device{device}, memory_manager{memory_manager}, scheduler{scheduler}, staging_pool{
+                                                                                staging_pool} {}
+
+VKBufferCache::~VKBufferCache() = default;
+
+std::shared_ptr<Buffer> VKBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) {
+    return std::make_shared<Buffer>(device, memory_manager, scheduler, staging_pool, cpu_addr,
+                                    size);
+}
+
+VKBufferCache::BufferInfo VKBufferCache::GetEmptyBuffer(std::size_t size) {
+    size = std::max(size, std::size_t(4));
+    const auto& empty = staging_pool.GetUnusedBuffer(size, false);
+    scheduler.RequestOutsideRenderPassOperationContext();
+    scheduler.Record([size, buffer = *empty.handle](vk::CommandBuffer cmdbuf) {
+        cmdbuf.FillBuffer(buffer, 0, size, 0);
+    });
+    return {*empty.handle, 0, 0};
+}
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -8,7 +8,6 @@
 
 #include "common/common_types.h"
 #include "video_core/buffer_cache/buffer_cache.h"
-#include "video_core/rasterizer_cache.h"
 #include "video_core/renderer_vulkan/vk_memory_manager.h"
 #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
 #include "video_core/renderer_vulkan/vk_stream_buffer.h"
@@ -24,22 +23,34 @@ class VKDevice;
 class VKMemoryManager;
 class VKScheduler;
 
-class CachedBufferBlock final : public VideoCommon::BufferBlock {
+class Buffer final : public VideoCommon::BufferBlock {
 public:
-    explicit CachedBufferBlock(const VKDevice& device, VKMemoryManager& memory_manager,
-                               VAddr cpu_addr, std::size_t size);
-    ~CachedBufferBlock();
+    explicit Buffer(const VKDevice& device, VKMemoryManager& memory_manager, VKScheduler& scheduler,
+                    VKStagingBufferPool& staging_pool, VAddr cpu_addr, std::size_t size);
+    ~Buffer();
 
-    VkBuffer GetHandle() const {
+    void Upload(std::size_t offset, std::size_t size, const u8* data) const;
+
+    void Download(std::size_t offset, std::size_t size, u8* data) const;
+
+    void CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset,
+                  std::size_t size) const;
+
+    VkBuffer Handle() const {
         return *buffer.handle;
     }
 
+    u64 Address() const {
+        return 0;
+    }
+
 private:
+    VKScheduler& scheduler;
+    VKStagingBufferPool& staging_pool;
+
     VKBuffer buffer;
 };
 
-using Buffer = std::shared_ptr<CachedBufferBlock>;
-
 class VKBufferCache final : public VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer> {
 public:
     explicit VKBufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system,
@@ -47,21 +58,10 @@ public:
                            VKScheduler& scheduler, VKStagingBufferPool& staging_pool);
     ~VKBufferCache();
 
-    VkBuffer GetEmptyBuffer(std::size_t size) override;
+    BufferInfo GetEmptyBuffer(std::size_t size) override;
 
 protected:
-    VkBuffer ToHandle(const Buffer& buffer) override;
-
-    Buffer CreateBlock(VAddr cpu_addr, std::size_t size) override;
-
-    void UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
-                         const u8* data) override;
-
-    void DownloadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
-                           u8* data) override;
-
-    void CopyBlock(const Buffer& src, const Buffer& dst, std::size_t src_offset,
-                   std::size_t dst_offset, std::size_t size) override;
+    std::shared_ptr<Buffer> CreateBlock(VAddr cpu_addr, std::size_t size) override;
 
 private:
     const VKDevice& device;

src/video_core/renderer_vulkan/vk_compute_pipeline.cpp

@@ -53,8 +53,9 @@ vk::DescriptorSetLayout VKComputePipeline::CreateDescriptorSetLayout() const {
     };
     add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, entries.const_buffers.size());
     add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, entries.global_buffers.size());
-    add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, entries.texel_buffers.size());
+    add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, entries.uniform_texels.size());
     add_bindings(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, entries.samplers.size());
+    add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, entries.storage_texels.size());
     add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, entries.images.size());
 
     VkDescriptorSetLayoutCreateInfo ci;

src/video_core/renderer_vulkan/vk_descriptor_pool.cpp

@@ -42,6 +42,7 @@ vk::DescriptorPool* VKDescriptorPool::AllocateNewPool() {
         {VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, num_sets * 60},
         {VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, num_sets * 64},
         {VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, num_sets * 64},
+        {VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, num_sets * 64},
         {VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, num_sets * 40}};
 
     VkDescriptorPoolCreateInfo ci;

src/video_core/renderer_vulkan/vk_device.cpp

@@ -73,76 +73,79 @@ VkFormatFeatureFlags GetFormatFeatures(VkFormatProperties properties, FormatType
 
 std::unordered_map<VkFormat, VkFormatProperties> GetFormatProperties(
     vk::PhysicalDevice physical, const vk::InstanceDispatch& dld) {
-    static constexpr std::array formats{VK_FORMAT_A8B8G8R8_UNORM_PACK32,
-                                        VK_FORMAT_A8B8G8R8_UINT_PACK32,
-                                        VK_FORMAT_A8B8G8R8_SNORM_PACK32,
-                                        VK_FORMAT_A8B8G8R8_SRGB_PACK32,
-                                        VK_FORMAT_B5G6R5_UNORM_PACK16,
-                                        VK_FORMAT_A2B10G10R10_UNORM_PACK32,
-                                        VK_FORMAT_A1R5G5B5_UNORM_PACK16,
-                                        VK_FORMAT_R32G32B32A32_SFLOAT,
-                                        VK_FORMAT_R32G32B32A32_UINT,
-                                        VK_FORMAT_R32G32_SFLOAT,
-                                        VK_FORMAT_R32G32_UINT,
-                                        VK_FORMAT_R16G16B16A16_UINT,
-                                        VK_FORMAT_R16G16B16A16_SNORM,
-                                        VK_FORMAT_R16G16B16A16_UNORM,
-                                        VK_FORMAT_R16G16_UNORM,
-                                        VK_FORMAT_R16G16_SNORM,
-                                        VK_FORMAT_R16G16_SFLOAT,
-                                        VK_FORMAT_R16_UNORM,
-                                        VK_FORMAT_R8G8B8A8_SRGB,
-                                        VK_FORMAT_R8G8_UNORM,
-                                        VK_FORMAT_R8G8_SNORM,
-                                        VK_FORMAT_R8G8_UINT,
-                                        VK_FORMAT_R8_UNORM,
-                                        VK_FORMAT_R8_UINT,
-                                        VK_FORMAT_B10G11R11_UFLOAT_PACK32,
-                                        VK_FORMAT_R32_SFLOAT,
-                                        VK_FORMAT_R32_UINT,
-                                        VK_FORMAT_R32_SINT,
-                                        VK_FORMAT_R16_SFLOAT,
-                                        VK_FORMAT_R16G16B16A16_SFLOAT,
-                                        VK_FORMAT_B8G8R8A8_UNORM,
-                                        VK_FORMAT_B8G8R8A8_SRGB,
-                                        VK_FORMAT_R4G4B4A4_UNORM_PACK16,
-                                        VK_FORMAT_D32_SFLOAT,
-                                        VK_FORMAT_D16_UNORM,
-                                        VK_FORMAT_D16_UNORM_S8_UINT,
-                                        VK_FORMAT_D24_UNORM_S8_UINT,
-                                        VK_FORMAT_D32_SFLOAT_S8_UINT,
-                                        VK_FORMAT_BC1_RGBA_UNORM_BLOCK,
-                                        VK_FORMAT_BC2_UNORM_BLOCK,
-                                        VK_FORMAT_BC3_UNORM_BLOCK,
-                                        VK_FORMAT_BC4_UNORM_BLOCK,
-                                        VK_FORMAT_BC5_UNORM_BLOCK,
-                                        VK_FORMAT_BC5_SNORM_BLOCK,
-                                        VK_FORMAT_BC7_UNORM_BLOCK,
-                                        VK_FORMAT_BC6H_UFLOAT_BLOCK,
-                                        VK_FORMAT_BC6H_SFLOAT_BLOCK,
-                                        VK_FORMAT_BC1_RGBA_SRGB_BLOCK,
-                                        VK_FORMAT_BC2_SRGB_BLOCK,
-                                        VK_FORMAT_BC3_SRGB_BLOCK,
-                                        VK_FORMAT_BC7_SRGB_BLOCK,
-                                        VK_FORMAT_ASTC_4x4_SRGB_BLOCK,
-                                        VK_FORMAT_ASTC_8x8_SRGB_BLOCK,
-                                        VK_FORMAT_ASTC_8x5_SRGB_BLOCK,
-                                        VK_FORMAT_ASTC_5x4_SRGB_BLOCK,
-                                        VK_FORMAT_ASTC_5x5_UNORM_BLOCK,
-                                        VK_FORMAT_ASTC_5x5_SRGB_BLOCK,
-                                        VK_FORMAT_ASTC_10x8_UNORM_BLOCK,
-                                        VK_FORMAT_ASTC_10x8_SRGB_BLOCK,
-                                        VK_FORMAT_ASTC_6x6_UNORM_BLOCK,
-                                        VK_FORMAT_ASTC_6x6_SRGB_BLOCK,
-                                        VK_FORMAT_ASTC_10x10_UNORM_BLOCK,
-                                        VK_FORMAT_ASTC_10x10_SRGB_BLOCK,
-                                        VK_FORMAT_ASTC_12x12_UNORM_BLOCK,
-                                        VK_FORMAT_ASTC_12x12_SRGB_BLOCK,
-                                        VK_FORMAT_ASTC_8x6_UNORM_BLOCK,
-                                        VK_FORMAT_ASTC_8x6_SRGB_BLOCK,
-                                        VK_FORMAT_ASTC_6x5_UNORM_BLOCK,
-                                        VK_FORMAT_ASTC_6x5_SRGB_BLOCK,
-                                        VK_FORMAT_E5B9G9R9_UFLOAT_PACK32};
+    static constexpr std::array formats{
+        VK_FORMAT_A8B8G8R8_UNORM_PACK32,
+        VK_FORMAT_A8B8G8R8_UINT_PACK32,
+        VK_FORMAT_A8B8G8R8_SNORM_PACK32,
+        VK_FORMAT_A8B8G8R8_SRGB_PACK32,
+        VK_FORMAT_B5G6R5_UNORM_PACK16,
+        VK_FORMAT_A2B10G10R10_UNORM_PACK32,
+        VK_FORMAT_A1R5G5B5_UNORM_PACK16,
+        VK_FORMAT_R32G32B32A32_SFLOAT,
+        VK_FORMAT_R32G32B32A32_UINT,
+        VK_FORMAT_R32G32_SFLOAT,
+        VK_FORMAT_R32G32_UINT,
+        VK_FORMAT_R16G16B16A16_UINT,
+        VK_FORMAT_R16G16B16A16_SNORM,
+        VK_FORMAT_R16G16B16A16_UNORM,
+        VK_FORMAT_R16G16_UNORM,
+        VK_FORMAT_R16G16_SNORM,
+        VK_FORMAT_R16G16_SFLOAT,
+        VK_FORMAT_R16_UNORM,
+        VK_FORMAT_R16_UINT,
+        VK_FORMAT_R8G8B8A8_SRGB,
+        VK_FORMAT_R8G8_UNORM,
+        VK_FORMAT_R8G8_SNORM,
+        VK_FORMAT_R8G8_UINT,
+        VK_FORMAT_R8_UNORM,
+        VK_FORMAT_R8_UINT,
+        VK_FORMAT_B10G11R11_UFLOAT_PACK32,
+        VK_FORMAT_R32_SFLOAT,
+        VK_FORMAT_R32_UINT,
+        VK_FORMAT_R32_SINT,
+        VK_FORMAT_R16_SFLOAT,
+        VK_FORMAT_R16G16B16A16_SFLOAT,
+        VK_FORMAT_B8G8R8A8_UNORM,
+        VK_FORMAT_B8G8R8A8_SRGB,
+        VK_FORMAT_R4G4B4A4_UNORM_PACK16,
+        VK_FORMAT_D32_SFLOAT,
+        VK_FORMAT_D16_UNORM,
+        VK_FORMAT_D16_UNORM_S8_UINT,
+        VK_FORMAT_D24_UNORM_S8_UINT,
+        VK_FORMAT_D32_SFLOAT_S8_UINT,
+        VK_FORMAT_BC1_RGBA_UNORM_BLOCK,
+        VK_FORMAT_BC2_UNORM_BLOCK,
+        VK_FORMAT_BC3_UNORM_BLOCK,
+        VK_FORMAT_BC4_UNORM_BLOCK,
+        VK_FORMAT_BC5_UNORM_BLOCK,
+        VK_FORMAT_BC5_SNORM_BLOCK,
+        VK_FORMAT_BC7_UNORM_BLOCK,
+        VK_FORMAT_BC6H_UFLOAT_BLOCK,
+        VK_FORMAT_BC6H_SFLOAT_BLOCK,
+        VK_FORMAT_BC1_RGBA_SRGB_BLOCK,
+        VK_FORMAT_BC2_SRGB_BLOCK,
+        VK_FORMAT_BC3_SRGB_BLOCK,
+        VK_FORMAT_BC7_SRGB_BLOCK,
+        VK_FORMAT_ASTC_4x4_SRGB_BLOCK,
+        VK_FORMAT_ASTC_8x8_SRGB_BLOCK,
+        VK_FORMAT_ASTC_8x5_SRGB_BLOCK,
+        VK_FORMAT_ASTC_5x4_SRGB_BLOCK,
+        VK_FORMAT_ASTC_5x5_UNORM_BLOCK,
+        VK_FORMAT_ASTC_5x5_SRGB_BLOCK,
+        VK_FORMAT_ASTC_10x8_UNORM_BLOCK,
+        VK_FORMAT_ASTC_10x8_SRGB_BLOCK,
+        VK_FORMAT_ASTC_6x6_UNORM_BLOCK,
+        VK_FORMAT_ASTC_6x6_SRGB_BLOCK,
+        VK_FORMAT_ASTC_10x10_UNORM_BLOCK,
+        VK_FORMAT_ASTC_10x10_SRGB_BLOCK,
+        VK_FORMAT_ASTC_12x12_UNORM_BLOCK,
+        VK_FORMAT_ASTC_12x12_SRGB_BLOCK,
+        VK_FORMAT_ASTC_8x6_UNORM_BLOCK,
+        VK_FORMAT_ASTC_8x6_SRGB_BLOCK,
+        VK_FORMAT_ASTC_6x5_UNORM_BLOCK,
+        VK_FORMAT_ASTC_6x5_SRGB_BLOCK,
+        VK_FORMAT_E5B9G9R9_UFLOAT_PACK32,
+    };
     std::unordered_map<VkFormat, VkFormatProperties> format_properties;
     for (const auto format : formats) {
         format_properties.emplace(format, physical.GetFormatProperties(format));

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -27,6 +27,7 @@
 #include "video_core/renderer_vulkan/wrapper.h"
 #include "video_core/shader/compiler_settings.h"
 #include "video_core/shader/memory_util.h"
+#include "video_core/shader_cache.h"
 
 namespace Vulkan {
 
@@ -45,6 +46,7 @@ constexpr VkDescriptorType UNIFORM_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
 constexpr VkDescriptorType STORAGE_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
 constexpr VkDescriptorType UNIFORM_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER;
 constexpr VkDescriptorType COMBINED_IMAGE_SAMPLER = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
+constexpr VkDescriptorType STORAGE_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER;
 constexpr VkDescriptorType STORAGE_IMAGE = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE;
 
 constexpr VideoCommon::Shader::CompilerSettings compiler_settings{
@@ -104,8 +106,9 @@ u32 FillDescriptorLayout(const ShaderEntries& entries,
     u32 binding = base_binding;
     AddBindings<UNIFORM_BUFFER>(bindings, binding, flags, entries.const_buffers);
     AddBindings<STORAGE_BUFFER>(bindings, binding, flags, entries.global_buffers);
-    AddBindings<UNIFORM_TEXEL_BUFFER>(bindings, binding, flags, entries.texel_buffers);
+    AddBindings<UNIFORM_TEXEL_BUFFER>(bindings, binding, flags, entries.uniform_texels);
     AddBindings<COMBINED_IMAGE_SAMPLER>(bindings, binding, flags, entries.samplers);
+    AddBindings<STORAGE_TEXEL_BUFFER>(bindings, binding, flags, entries.storage_texels);
     AddBindings<STORAGE_IMAGE>(bindings, binding, flags, entries.images);
     return binding;
 }
@@ -130,19 +133,18 @@ bool ComputePipelineCacheKey::operator==(const ComputePipelineCacheKey& rhs) con
     return std::memcmp(&rhs, this, sizeof *this) == 0;
 }
 
-CachedShader::CachedShader(Core::System& system, Tegra::Engines::ShaderType stage,
-                           GPUVAddr gpu_addr, VAddr cpu_addr, ProgramCode program_code,
-                           u32 main_offset)
-    : RasterizerCacheObject{cpu_addr}, gpu_addr{gpu_addr}, program_code{std::move(program_code)},
+Shader::Shader(Core::System& system, Tegra::Engines::ShaderType stage, GPUVAddr gpu_addr,
+               VideoCommon::Shader::ProgramCode program_code, u32 main_offset)
+    : gpu_addr{gpu_addr}, program_code{std::move(program_code)},
       registry{stage, GetEngine(system, stage)}, shader_ir{this->program_code, main_offset,
                                                            compiler_settings, registry},
       entries{GenerateShaderEntries(shader_ir)} {}
 
-CachedShader::~CachedShader() = default;
+Shader::~Shader() = default;
 
-Tegra::Engines::ConstBufferEngineInterface& CachedShader::GetEngine(
-    Core::System& system, Tegra::Engines::ShaderType stage) {
-    if (stage == Tegra::Engines::ShaderType::Compute) {
+Tegra::Engines::ConstBufferEngineInterface& Shader::GetEngine(Core::System& system,
+                                                              Tegra::Engines::ShaderType stage) {
+    if (stage == ShaderType::Compute) {
         return system.GPU().KeplerCompute();
     } else {
         return system.GPU().Maxwell3D();
@@ -154,16 +156,16 @@ VKPipelineCache::VKPipelineCache(Core::System& system, RasterizerVulkan& rasteri
                                  VKDescriptorPool& descriptor_pool,
                                  VKUpdateDescriptorQueue& update_descriptor_queue,
                                  VKRenderPassCache& renderpass_cache)
-    : RasterizerCache{rasterizer}, system{system}, device{device}, scheduler{scheduler},
-      descriptor_pool{descriptor_pool}, update_descriptor_queue{update_descriptor_queue},
-      renderpass_cache{renderpass_cache} {}
+    : VideoCommon::ShaderCache<Shader>{rasterizer}, system{system}, device{device},
+      scheduler{scheduler}, descriptor_pool{descriptor_pool},
+      update_descriptor_queue{update_descriptor_queue}, renderpass_cache{renderpass_cache} {}
 
 VKPipelineCache::~VKPipelineCache() = default;
 
-std::array<Shader, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() {
+std::array<Shader*, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() {
     const auto& gpu = system.GPU().Maxwell3D();
 
-    std::array<Shader, Maxwell::MaxShaderProgram> shaders;
+    std::array<Shader*, Maxwell::MaxShaderProgram> shaders{};
     for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
         const auto program{static_cast<Maxwell::ShaderProgram>(index)};
 
@@ -176,24 +178,28 @@ std::array<Shader, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() {
         const GPUVAddr program_addr{GetShaderAddress(system, program)};
         const std::optional cpu_addr = memory_manager.GpuToCpuAddress(program_addr);
         ASSERT(cpu_addr);
-        auto shader = cpu_addr ? TryGet(*cpu_addr) : null_shader;
-        if (!shader) {
+
+        Shader* result = cpu_addr ? TryGet(*cpu_addr) : null_shader.get();
+        if (!result) {
             const auto host_ptr{memory_manager.GetPointer(program_addr)};
 
             // No shader found - create a new one
             constexpr u32 stage_offset = STAGE_MAIN_OFFSET;
-            const auto stage = static_cast<Tegra::Engines::ShaderType>(index == 0 ? 0 : index - 1);
+            const auto stage = static_cast<ShaderType>(index == 0 ? 0 : index - 1);
             ProgramCode code = GetShaderCode(memory_manager, program_addr, host_ptr, false);
+            const std::size_t size_in_bytes = code.size() * sizeof(u64);
+
+            auto shader = std::make_unique<Shader>(system, stage, program_addr, std::move(code),
+                                                   stage_offset);
+            result = shader.get();
 
-            shader = std::make_shared<CachedShader>(system, stage, program_addr, *cpu_addr,
-                                                    std::move(code), stage_offset);
             if (cpu_addr) {
-                Register(shader);
+                Register(std::move(shader), *cpu_addr, size_in_bytes);
             } else {
-                null_shader = shader;
+                null_shader = std::move(shader);
             }
         }
-        shaders[index] = std::move(shader);
+        shaders[index] = result;
     }
     return last_shaders = shaders;
 }
@@ -234,19 +240,22 @@ VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCach
     const auto cpu_addr = memory_manager.GpuToCpuAddress(program_addr);
     ASSERT(cpu_addr);
 
-    auto shader = cpu_addr ? TryGet(*cpu_addr) : null_kernel;
+    Shader* shader = cpu_addr ? TryGet(*cpu_addr) : null_kernel.get();
     if (!shader) {
         // No shader found - create a new one
         const auto host_ptr = memory_manager.GetPointer(program_addr);
 
         ProgramCode code = GetShaderCode(memory_manager, program_addr, host_ptr, true);
-        shader = std::make_shared<CachedShader>(system, Tegra::Engines::ShaderType::Compute,
-                                                program_addr, *cpu_addr, std::move(code),
-                                                KERNEL_MAIN_OFFSET);
+        const std::size_t size_in_bytes = code.size() * sizeof(u64);
+
+        auto shader_info = std::make_unique<Shader>(system, ShaderType::Compute, program_addr,
+                                                    std::move(code), KERNEL_MAIN_OFFSET);
+        shader = shader_info.get();
+
         if (cpu_addr) {
-            Register(shader);
+            Register(std::move(shader_info), *cpu_addr, size_in_bytes);
         } else {
-            null_kernel = shader;
+            null_kernel = std::move(shader_info);
         }
     }
 
@@ -262,7 +271,7 @@ VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCach
     return *entry;
 }
 
-void VKPipelineCache::Unregister(const Shader& shader) {
+void VKPipelineCache::OnShaderRemoval(Shader* shader) {
     bool finished = false;
     const auto Finish = [&] {
         // TODO(Rodrigo): Instead of finishing here, wait for the fences that use this pipeline and
@@ -294,8 +303,6 @@ void VKPipelineCache::Unregister(const Shader& shader) {
         Finish();
         it = compute_cache.erase(it);
     }
-
-    RasterizerCache::Unregister(shader);
 }
 
 std::pair<SPIRVProgram, std::vector<VkDescriptorSetLayoutBinding>>
@@ -312,7 +319,9 @@ VKPipelineCache::DecompileShaders(const GraphicsPipelineCacheKey& key) {
         ASSERT(point_size != 0.0f);
     }
     for (std::size_t i = 0; i < Maxwell::NumVertexAttributes; ++i) {
-        specialization.attribute_types[i] = fixed_state.vertex_input.attributes[i].Type();
+        const auto& attribute = fixed_state.vertex_input.attributes[i];
+        specialization.enabled_attributes[i] = attribute.enabled.Value() != 0;
+        specialization.attribute_types[i] = attribute.Type();
     }
     specialization.ndc_minus_one_to_one = fixed_state.rasterizer.ndc_minus_one_to_one;
 
@@ -328,12 +337,11 @@ VKPipelineCache::DecompileShaders(const GraphicsPipelineCacheKey& key) {
         }
 
         const GPUVAddr gpu_addr = GetShaderAddress(system, program_enum);
-        const auto cpu_addr = memory_manager.GpuToCpuAddress(gpu_addr);
-        const auto shader = cpu_addr ? TryGet(*cpu_addr) : null_shader;
-        ASSERT(shader);
+        const std::optional<VAddr> cpu_addr = memory_manager.GpuToCpuAddress(gpu_addr);
+        Shader* const shader = cpu_addr ? TryGet(*cpu_addr) : null_shader.get();
 
         const std::size_t stage = index == 0 ? 0 : index - 1; // Stage indices are 0 - 5
-        const auto program_type = GetShaderType(program_enum);
+        const ShaderType program_type = GetShaderType(program_enum);
         const auto& entries = shader->GetEntries();
         program[stage] = {
             Decompile(device, shader->GetIR(), program_type, shader->GetRegistry(), specialization),
@@ -375,16 +383,17 @@ void AddEntry(std::vector<VkDescriptorUpdateTemplateEntry>& template_entries, u3
         return;
     }
 
-    if constexpr (descriptor_type == UNIFORM_TEXEL_BUFFER) {
-        // Nvidia has a bug where updating multiple uniform texels at once causes the driver to
-        // crash.
+    if constexpr (descriptor_type == UNIFORM_TEXEL_BUFFER ||
+                  descriptor_type == STORAGE_TEXEL_BUFFER) {
+        // Nvidia has a bug where updating multiple texels at once causes the driver to crash.
+        // Note: Fixed in driver Windows 443.24, Linux 440.66.15
         for (u32 i = 0; i < count; ++i) {
             VkDescriptorUpdateTemplateEntry& entry = template_entries.emplace_back();
             entry.dstBinding = binding + i;
             entry.dstArrayElement = 0;
             entry.descriptorCount = 1;
             entry.descriptorType = descriptor_type;
-            entry.offset = offset + i * entry_size;
+            entry.offset = static_cast<std::size_t>(offset + i * entry_size);
             entry.stride = entry_size;
         }
     } else if (count > 0) {
@@ -405,8 +414,9 @@ void FillDescriptorUpdateTemplateEntries(
     std::vector<VkDescriptorUpdateTemplateEntryKHR>& template_entries) {
     AddEntry<UNIFORM_BUFFER>(template_entries, offset, binding, entries.const_buffers);
     AddEntry<STORAGE_BUFFER>(template_entries, offset, binding, entries.global_buffers);
-    AddEntry<UNIFORM_TEXEL_BUFFER>(template_entries, offset, binding, entries.texel_buffers);
+    AddEntry<UNIFORM_TEXEL_BUFFER>(template_entries, offset, binding, entries.uniform_texels);
     AddEntry<COMBINED_IMAGE_SAMPLER>(template_entries, offset, binding, entries.samplers);
+    AddEntry<STORAGE_TEXEL_BUFFER>(template_entries, offset, binding, entries.storage_texels);
     AddEntry<STORAGE_IMAGE>(template_entries, offset, binding, entries.images);
 }
 

src/video_core/renderer_vulkan/vk_pipeline_cache.h

@@ -17,7 +17,6 @@
 #include "common/common_types.h"
 #include "video_core/engines/const_buffer_engine_interface.h"
 #include "video_core/engines/maxwell_3d.h"
-#include "video_core/rasterizer_cache.h"
 #include "video_core/renderer_vulkan/fixed_pipeline_state.h"
 #include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
 #include "video_core/renderer_vulkan/vk_renderpass_cache.h"
@@ -26,6 +25,7 @@
 #include "video_core/shader/memory_util.h"
 #include "video_core/shader/registry.h"
 #include "video_core/shader/shader_ir.h"
+#include "video_core/shader_cache.h"
 
 namespace Core {
 class System;
@@ -41,8 +41,6 @@ class VKFence;
 class VKScheduler;
 class VKUpdateDescriptorQueue;
 
-class CachedShader;
-using Shader = std::shared_ptr<CachedShader>;
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 
 struct GraphicsPipelineCacheKey {
@@ -102,21 +100,16 @@ struct hash<Vulkan::ComputePipelineCacheKey> {
 
 namespace Vulkan {
 
-class CachedShader final : public RasterizerCacheObject {
+class Shader {
 public:
-    explicit CachedShader(Core::System& system, Tegra::Engines::ShaderType stage, GPUVAddr gpu_addr,
-                          VAddr cpu_addr, VideoCommon::Shader::ProgramCode program_code,
-                          u32 main_offset);
-    ~CachedShader();
+    explicit Shader(Core::System& system, Tegra::Engines::ShaderType stage, GPUVAddr gpu_addr,
+                    VideoCommon::Shader::ProgramCode program_code, u32 main_offset);
+    ~Shader();
 
     GPUVAddr GetGpuAddr() const {
         return gpu_addr;
     }
 
-    std::size_t GetSizeInBytes() const override {
-        return program_code.size() * sizeof(u64);
-    }
-
     VideoCommon::Shader::ShaderIR& GetIR() {
         return shader_ir;
     }
@@ -144,25 +137,23 @@ private:
     ShaderEntries entries;
 };
 
-class VKPipelineCache final : public RasterizerCache<Shader> {
+class VKPipelineCache final : public VideoCommon::ShaderCache<Shader> {
 public:
     explicit VKPipelineCache(Core::System& system, RasterizerVulkan& rasterizer,
                              const VKDevice& device, VKScheduler& scheduler,
                              VKDescriptorPool& descriptor_pool,
                              VKUpdateDescriptorQueue& update_descriptor_queue,
                              VKRenderPassCache& renderpass_cache);
-    ~VKPipelineCache();
+    ~VKPipelineCache() override;
 
-    std::array<Shader, Maxwell::MaxShaderProgram> GetShaders();
+    std::array<Shader*, Maxwell::MaxShaderProgram> GetShaders();
 
     VKGraphicsPipeline& GetGraphicsPipeline(const GraphicsPipelineCacheKey& key);
 
     VKComputePipeline& GetComputePipeline(const ComputePipelineCacheKey& key);
 
 protected:
-    void Unregister(const Shader& shader) override;
-
-    void FlushObjectInner(const Shader& object) override {}
+    void OnShaderRemoval(Shader* shader) final;
 
 private:
     std::pair<SPIRVProgram, std::vector<VkDescriptorSetLayoutBinding>> DecompileShaders(
@@ -175,10 +166,10 @@ private:
     VKUpdateDescriptorQueue& update_descriptor_queue;
     VKRenderPassCache& renderpass_cache;
 
-    Shader null_shader{};
-    Shader null_kernel{};
+    std::unique_ptr<Shader> null_shader;
+    std::unique_ptr<Shader> null_kernel;
 
-    std::array<Shader, Maxwell::MaxShaderProgram> last_shaders;
+    std::array<Shader*, Maxwell::MaxShaderProgram> last_shaders{};
 
     GraphicsPipelineCacheKey last_graphics_key;
     VKGraphicsPipeline* last_graphics_pipeline = nullptr;

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -38,6 +38,7 @@
 #include "video_core/renderer_vulkan/vk_texture_cache.h"
 #include "video_core/renderer_vulkan/vk_update_descriptor.h"
 #include "video_core/renderer_vulkan/wrapper.h"
+#include "video_core/shader_cache.h"
 
 namespace Vulkan {
 
@@ -98,7 +99,7 @@ VkRect2D GetScissorState(const Maxwell& regs, std::size_t index) {
 }
 
 std::array<GPUVAddr, Maxwell::MaxShaderProgram> GetShaderAddresses(
-    const std::array<Shader, Maxwell::MaxShaderProgram>& shaders) {
+    const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders) {
     std::array<GPUVAddr, Maxwell::MaxShaderProgram> addresses;
     for (std::size_t i = 0; i < std::size(addresses); ++i) {
         addresses[i] = shaders[i] ? shaders[i]->GetGpuAddr() : 0;
@@ -117,6 +118,17 @@ template <typename Engine, typename Entry>
 Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry& entry,
                                                std::size_t stage, std::size_t index = 0) {
     const auto stage_type = static_cast<Tegra::Engines::ShaderType>(stage);
+    if constexpr (std::is_same_v<Entry, SamplerEntry>) {
+        if (entry.is_separated) {
+            const u32 buffer_1 = entry.buffer;
+            const u32 buffer_2 = entry.secondary_buffer;
+            const u32 offset_1 = entry.offset;
+            const u32 offset_2 = entry.secondary_offset;
+            const u32 handle_1 = engine.AccessConstBuffer32(stage_type, buffer_1, offset_1);
+            const u32 handle_2 = engine.AccessConstBuffer32(stage_type, buffer_2, offset_2);
+            return engine.GetTextureInfo(handle_1 | handle_2);
+        }
+    }
     if (entry.is_bindless) {
         const auto tex_handle = engine.AccessConstBuffer32(stage_type, entry.buffer, entry.offset);
         return engine.GetTextureInfo(tex_handle);
@@ -468,8 +480,9 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) {
     const auto& entries = pipeline.GetEntries();
     SetupComputeConstBuffers(entries);
     SetupComputeGlobalBuffers(entries);
-    SetupComputeTexelBuffers(entries);
+    SetupComputeUniformTexels(entries);
     SetupComputeTextures(entries);
+    SetupComputeStorageTexels(entries);
     SetupComputeImages(entries);
 
     buffer_cache.Unmap();
@@ -715,7 +728,7 @@ std::tuple<VkFramebuffer, VkExtent2D> RasterizerVulkan::ConfigureFramebuffers(
         if (!view) {
             return false;
         }
-        key.views.push_back(view->GetHandle());
+        key.views.push_back(view->GetAttachment());
         key.width = std::min(key.width, view->GetWidth());
         key.height = std::min(key.height, view->GetHeight());
         key.layers = std::min(key.layers, view->GetNumLayers());
@@ -775,20 +788,21 @@ RasterizerVulkan::DrawParameters RasterizerVulkan::SetupGeometry(FixedPipelineSt
 }
 
 void RasterizerVulkan::SetupShaderDescriptors(
-    const std::array<Shader, Maxwell::MaxShaderProgram>& shaders) {
+    const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders) {
     texture_cache.GuardSamplers(true);
 
     for (std::size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
         // Skip VertexA stage
-        const auto& shader = shaders[stage + 1];
+        Shader* const shader = shaders[stage + 1];
         if (!shader) {
             continue;
         }
         const auto& entries = shader->GetEntries();
         SetupGraphicsConstBuffers(entries, stage);
         SetupGraphicsGlobalBuffers(entries, stage);
-        SetupGraphicsTexelBuffers(entries, stage);
+        SetupGraphicsUniformTexels(entries, stage);
         SetupGraphicsTextures(entries, stage);
+        SetupGraphicsStorageTexels(entries, stage);
         SetupGraphicsImages(entries, stage);
     }
     texture_cache.GuardSamplers(false);
@@ -838,6 +852,10 @@ void RasterizerVulkan::BeginTransformFeedback() {
     if (regs.tfb_enabled == 0) {
         return;
     }
+    if (!device.IsExtTransformFeedbackSupported()) {
+        LOG_ERROR(Render_Vulkan, "Transform feedbacks used but not supported");
+        return;
+    }
 
     UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationControl) ||
                      regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationEval) ||
@@ -852,10 +870,10 @@ void RasterizerVulkan::BeginTransformFeedback() {
     UNIMPLEMENTED_IF(binding.buffer_offset != 0);
 
     const GPUVAddr gpu_addr = binding.Address();
-    const std::size_t size = binding.buffer_size;
-    const auto [buffer, offset] = buffer_cache.UploadMemory(gpu_addr, size, 4, true);
+    const VkDeviceSize size = static_cast<VkDeviceSize>(binding.buffer_size);
+    const auto info = buffer_cache.UploadMemory(gpu_addr, size, 4, true);
 
-    scheduler.Record([buffer = buffer, offset = offset, size](vk::CommandBuffer cmdbuf) {
+    scheduler.Record([buffer = info.handle, offset = info.offset, size](vk::CommandBuffer cmdbuf) {
         cmdbuf.BindTransformFeedbackBuffersEXT(0, 1, &buffer, &offset, &size);
         cmdbuf.BeginTransformFeedbackEXT(0, 0, nullptr, nullptr);
     });
@@ -866,6 +884,9 @@ void RasterizerVulkan::EndTransformFeedback() {
     if (regs.tfb_enabled == 0) {
         return;
     }
+    if (!device.IsExtTransformFeedbackSupported()) {
+        return;
+    }
 
     scheduler.Record(
         [](vk::CommandBuffer cmdbuf) { cmdbuf.EndTransformFeedbackEXT(0, 0, nullptr, nullptr); });
@@ -877,14 +898,10 @@ void RasterizerVulkan::SetupVertexArrays(FixedPipelineState::VertexInput& vertex
 
     for (std::size_t index = 0; index < Maxwell::NumVertexAttributes; ++index) {
         const auto& attrib = regs.vertex_attrib_format[index];
-        if (!attrib.IsValid()) {
+        if (attrib.IsConstant()) {
             vertex_input.SetAttribute(index, false, 0, 0, {}, {});
             continue;
         }
-
-        [[maybe_unused]] const auto& buffer = regs.vertex_array[attrib.buffer];
-        ASSERT(buffer.IsEnabled());
-
         vertex_input.SetAttribute(index, true, attrib.buffer, attrib.offset, attrib.type.Value(),
                                   attrib.size.Value());
     }
@@ -908,8 +925,8 @@ void RasterizerVulkan::SetupVertexArrays(FixedPipelineState::VertexInput& vertex
             buffer_bindings.AddVertexBinding(DefaultBuffer(), 0);
             continue;
         }
-        const auto [buffer, offset] = buffer_cache.UploadMemory(start, size);
-        buffer_bindings.AddVertexBinding(buffer, offset);
+        const auto info = buffer_cache.UploadMemory(start, size);
+        buffer_bindings.AddVertexBinding(info.handle, info.offset);
     }
 }
 
@@ -931,7 +948,9 @@ void RasterizerVulkan::SetupIndexBuffer(BufferBindings& buffer_bindings, DrawPar
             break;
         }
         const GPUVAddr gpu_addr = regs.index_array.IndexStart();
-        auto [buffer, offset] = buffer_cache.UploadMemory(gpu_addr, CalculateIndexBufferSize());
+        const auto info = buffer_cache.UploadMemory(gpu_addr, CalculateIndexBufferSize());
+        VkBuffer buffer = info.handle;
+        u64 offset = info.offset;
         std::tie(buffer, offset) = quad_indexed_pass.Assemble(
             regs.index_array.format, params.num_vertices, params.base_vertex, buffer, offset);
 
@@ -945,7 +964,9 @@ void RasterizerVulkan::SetupIndexBuffer(BufferBindings& buffer_bindings, DrawPar
             break;
         }
         const GPUVAddr gpu_addr = regs.index_array.IndexStart();
-        auto [buffer, offset] = buffer_cache.UploadMemory(gpu_addr, CalculateIndexBufferSize());
+        const auto info = buffer_cache.UploadMemory(gpu_addr, CalculateIndexBufferSize());
+        VkBuffer buffer = info.handle;
+        u64 offset = info.offset;
 
         auto format = regs.index_array.format;
         const bool is_uint8 = format == Maxwell::IndexFormat::UnsignedByte;
@@ -980,12 +1001,12 @@ void RasterizerVulkan::SetupGraphicsGlobalBuffers(const ShaderEntries& entries,
     }
 }
 
-void RasterizerVulkan::SetupGraphicsTexelBuffers(const ShaderEntries& entries, std::size_t stage) {
+void RasterizerVulkan::SetupGraphicsUniformTexels(const ShaderEntries& entries, std::size_t stage) {
     MICROPROFILE_SCOPE(Vulkan_Textures);
     const auto& gpu = system.GPU().Maxwell3D();
-    for (const auto& entry : entries.texel_buffers) {
+    for (const auto& entry : entries.uniform_texels) {
         const auto image = GetTextureInfo(gpu, entry, stage).tic;
-        SetupTexelBuffer(image, entry);
+        SetupUniformTexels(image, entry);
     }
 }
 
@@ -1000,6 +1021,15 @@ void RasterizerVulkan::SetupGraphicsTextures(const ShaderEntries& entries, std::
     }
 }
 
+void RasterizerVulkan::SetupGraphicsStorageTexels(const ShaderEntries& entries, std::size_t stage) {
+    MICROPROFILE_SCOPE(Vulkan_Textures);
+    const auto& gpu = system.GPU().Maxwell3D();
+    for (const auto& entry : entries.storage_texels) {
+        const auto image = GetTextureInfo(gpu, entry, stage).tic;
+        SetupStorageTexel(image, entry);
+    }
+}
+
 void RasterizerVulkan::SetupGraphicsImages(const ShaderEntries& entries, std::size_t stage) {
     MICROPROFILE_SCOPE(Vulkan_Images);
     const auto& gpu = system.GPU().Maxwell3D();
@@ -1032,12 +1062,12 @@ void RasterizerVulkan::SetupComputeGlobalBuffers(const ShaderEntries& entries) {
     }
 }
 
-void RasterizerVulkan::SetupComputeTexelBuffers(const ShaderEntries& entries) {
+void RasterizerVulkan::SetupComputeUniformTexels(const ShaderEntries& entries) {
     MICROPROFILE_SCOPE(Vulkan_Textures);
     const auto& gpu = system.GPU().KeplerCompute();
-    for (const auto& entry : entries.texel_buffers) {
+    for (const auto& entry : entries.uniform_texels) {
         const auto image = GetTextureInfo(gpu, entry, ComputeShaderIndex).tic;
-        SetupTexelBuffer(image, entry);
+        SetupUniformTexels(image, entry);
     }
 }
 
@@ -1052,6 +1082,15 @@ void RasterizerVulkan::SetupComputeTextures(const ShaderEntries& entries) {
     }
 }
 
+void RasterizerVulkan::SetupComputeStorageTexels(const ShaderEntries& entries) {
+    MICROPROFILE_SCOPE(Vulkan_Textures);
+    const auto& gpu = system.GPU().KeplerCompute();
+    for (const auto& entry : entries.storage_texels) {
+        const auto image = GetTextureInfo(gpu, entry, ComputeShaderIndex).tic;
+        SetupStorageTexel(image, entry);
+    }
+}
+
 void RasterizerVulkan::SetupComputeImages(const ShaderEntries& entries) {
     MICROPROFILE_SCOPE(Vulkan_Images);
     const auto& gpu = system.GPU().KeplerCompute();
@@ -1074,10 +1113,9 @@ void RasterizerVulkan::SetupConstBuffer(const ConstBufferEntry& entry,
         Common::AlignUp(CalculateConstBufferSize(entry, buffer), 4 * sizeof(float));
     ASSERT(size <= MaxConstbufferSize);
 
-    const auto [buffer_handle, offset] =
+    const auto info =
         buffer_cache.UploadMemory(buffer.address, size, device.GetUniformBufferAlignment());
-
-    update_descriptor_queue.AddBuffer(buffer_handle, offset, size);
+    update_descriptor_queue.AddBuffer(info.handle, info.offset, size);
 }
 
 void RasterizerVulkan::SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAddr address) {
@@ -1091,18 +1129,18 @@ void RasterizerVulkan::SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAdd
         // Note: Do *not* use DefaultBuffer() here, storage buffers can be written breaking the
         // default buffer.
         static constexpr std::size_t dummy_size = 4;
-        const auto buffer = buffer_cache.GetEmptyBuffer(dummy_size);
-        update_descriptor_queue.AddBuffer(buffer, 0, dummy_size);
+        const auto info = buffer_cache.GetEmptyBuffer(dummy_size);
+        update_descriptor_queue.AddBuffer(info.handle, info.offset, dummy_size);
         return;
     }
 
-    const auto [buffer, offset] = buffer_cache.UploadMemory(
+    const auto info = buffer_cache.UploadMemory(
         actual_addr, size, device.GetStorageBufferAlignment(), entry.IsWritten());
-    update_descriptor_queue.AddBuffer(buffer, offset, size);
+    update_descriptor_queue.AddBuffer(info.handle, info.offset, size);
 }
 
-void RasterizerVulkan::SetupTexelBuffer(const Tegra::Texture::TICEntry& tic,
-                                        const TexelBufferEntry& entry) {
+void RasterizerVulkan::SetupUniformTexels(const Tegra::Texture::TICEntry& tic,
+                                          const UniformTexelEntry& entry) {
     const auto view = texture_cache.GetTextureSurface(tic, entry);
     ASSERT(view->IsBufferView());
 
@@ -1114,16 +1152,24 @@ void RasterizerVulkan::SetupTexture(const Tegra::Texture::FullTextureInfo& textu
     auto view = texture_cache.GetTextureSurface(texture.tic, entry);
     ASSERT(!view->IsBufferView());
 
-    const auto image_view = view->GetHandle(texture.tic.x_source, texture.tic.y_source,
-                                            texture.tic.z_source, texture.tic.w_source);
+    const VkImageView image_view = view->GetImageView(texture.tic.x_source, texture.tic.y_source,
+                                                      texture.tic.z_source, texture.tic.w_source);
     const auto sampler = sampler_cache.GetSampler(texture.tsc);
     update_descriptor_queue.AddSampledImage(sampler, image_view);
 
-    const auto image_layout = update_descriptor_queue.GetLastImageLayout();
+    VkImageLayout* const image_layout = update_descriptor_queue.LastImageLayout();
     *image_layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
     sampled_views.push_back(ImageView{std::move(view), image_layout});
 }
 
+void RasterizerVulkan::SetupStorageTexel(const Tegra::Texture::TICEntry& tic,
+                                         const StorageTexelEntry& entry) {
+    const auto view = texture_cache.GetImageSurface(tic, entry);
+    ASSERT(view->IsBufferView());
+
+    update_descriptor_queue.AddTexelBuffer(view->GetBufferView());
+}
+
 void RasterizerVulkan::SetupImage(const Tegra::Texture::TICEntry& tic, const ImageEntry& entry) {
     auto view = texture_cache.GetImageSurface(tic, entry);
 
@@ -1133,10 +1179,11 @@ void RasterizerVulkan::SetupImage(const Tegra::Texture::TICEntry& tic, const Ima
 
     UNIMPLEMENTED_IF(tic.IsBuffer());
 
-    const auto image_view = view->GetHandle(tic.x_source, tic.y_source, tic.z_source, tic.w_source);
+    const VkImageView image_view =
+        view->GetImageView(tic.x_source, tic.y_source, tic.z_source, tic.w_source);
     update_descriptor_queue.AddImage(image_view);
 
-    const auto image_layout = update_descriptor_queue.GetLastImageLayout();
+    VkImageLayout* const image_layout = update_descriptor_queue.LastImageLayout();
     *image_layout = VK_IMAGE_LAYOUT_GENERAL;
     image_views.push_back(ImageView{std::move(view), image_layout});
 }

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -168,7 +168,7 @@ private:
                                  bool is_indexed, bool is_instanced);
 
     /// Setup descriptors in the graphics pipeline.
-    void SetupShaderDescriptors(const std::array<Shader, Maxwell::MaxShaderProgram>& shaders);
+    void SetupShaderDescriptors(const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders);
 
     void SetupImageTransitions(Texceptions texceptions,
                                const std::array<View, Maxwell::NumRenderTargets>& color_attachments,
@@ -193,12 +193,15 @@ private:
     /// Setup global buffers in the graphics pipeline.
     void SetupGraphicsGlobalBuffers(const ShaderEntries& entries, std::size_t stage);
 
-    /// Setup texel buffers in the graphics pipeline.
-    void SetupGraphicsTexelBuffers(const ShaderEntries& entries, std::size_t stage);
+    /// Setup uniform texels in the graphics pipeline.
+    void SetupGraphicsUniformTexels(const ShaderEntries& entries, std::size_t stage);
 
     /// Setup textures in the graphics pipeline.
     void SetupGraphicsTextures(const ShaderEntries& entries, std::size_t stage);
 
+    /// Setup storage texels in the graphics pipeline.
+    void SetupGraphicsStorageTexels(const ShaderEntries& entries, std::size_t stage);
+
     /// Setup images in the graphics pipeline.
     void SetupGraphicsImages(const ShaderEntries& entries, std::size_t stage);
 
@@ -209,11 +212,14 @@ private:
     void SetupComputeGlobalBuffers(const ShaderEntries& entries);
 
     /// Setup texel buffers in the compute pipeline.
-    void SetupComputeTexelBuffers(const ShaderEntries& entries);
+    void SetupComputeUniformTexels(const ShaderEntries& entries);
 
     /// Setup textures in the compute pipeline.
     void SetupComputeTextures(const ShaderEntries& entries);
 
+    /// Setup storage texels in the compute pipeline.
+    void SetupComputeStorageTexels(const ShaderEntries& entries);
+
     /// Setup images in the compute pipeline.
     void SetupComputeImages(const ShaderEntries& entries);
 
@@ -222,10 +228,12 @@ private:
 
     void SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAddr address);
 
-    void SetupTexelBuffer(const Tegra::Texture::TICEntry& image, const TexelBufferEntry& entry);
+    void SetupUniformTexels(const Tegra::Texture::TICEntry& image, const UniformTexelEntry& entry);
 
     void SetupTexture(const Tegra::Texture::FullTextureInfo& texture, const SamplerEntry& entry);
 
+    void SetupStorageTexel(const Tegra::Texture::TICEntry& tic, const StorageTexelEntry& entry);
+
     void SetupImage(const Tegra::Texture::TICEntry& tic, const ImageEntry& entry);
 
     void UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs);

src/video_core/renderer_vulkan/vk_sampler_cache.cpp

@@ -9,6 +9,8 @@
 #include "video_core/renderer_vulkan/wrapper.h"
 #include "video_core/textures/texture.h"
 
+using Tegra::Texture::TextureMipmapFilter;
+
 namespace Vulkan {
 
 namespace {
@@ -63,8 +65,8 @@ vk::Sampler VKSamplerCache::CreateSampler(const Tegra::Texture::TSCEntry& tsc) c
     ci.maxAnisotropy = tsc.GetMaxAnisotropy();
     ci.compareEnable = tsc.depth_compare_enabled;
     ci.compareOp = MaxwellToVK::Sampler::DepthCompareFunction(tsc.depth_compare_func);
-    ci.minLod = tsc.GetMinLod();
-    ci.maxLod = tsc.GetMaxLod();
+    ci.minLod = tsc.mipmap_filter == TextureMipmapFilter::None ? 0.0f : tsc.GetMinLod();
+    ci.maxLod = tsc.mipmap_filter == TextureMipmapFilter::None ? 0.25f : tsc.GetMaxLod();
     ci.borderColor = arbitrary_borders ? VK_BORDER_COLOR_INT_CUSTOM_EXT : ConvertBorderColor(color);
     ci.unnormalizedCoordinates = VK_FALSE;
     return device.GetLogical().CreateSampler(ci);

src/video_core/renderer_vulkan/vk_shader_decompiler.cpp

@@ -400,8 +400,9 @@ private:
         u32 binding = specialization.base_binding;
         binding = DeclareConstantBuffers(binding);
         binding = DeclareGlobalBuffers(binding);
-        binding = DeclareTexelBuffers(binding);
+        binding = DeclareUniformTexels(binding);
         binding = DeclareSamplers(binding);
+        binding = DeclareStorageTexels(binding);
         binding = DeclareImages(binding);
 
         const Id main = OpFunction(t_void, {}, TypeFunction(t_void));
@@ -741,8 +742,10 @@ private:
             if (!IsGenericAttribute(index)) {
                 continue;
             }
-
             const u32 location = GetGenericAttributeLocation(index);
+            if (!IsAttributeEnabled(location)) {
+                continue;
+            }
             const auto type_descriptor = GetAttributeType(location);
             Id type;
             if (IsInputAttributeArray()) {
@@ -887,7 +890,7 @@ private:
         return binding;
     }
 
-    u32 DeclareTexelBuffers(u32 binding) {
+    u32 DeclareUniformTexels(u32 binding) {
         for (const auto& sampler : ir.GetSamplers()) {
             if (!sampler.is_buffer) {
                 continue;
@@ -908,7 +911,7 @@ private:
             Decorate(id, spv::Decoration::Binding, binding++);
             Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET);
 
-            texel_buffers.emplace(sampler.index, TexelBuffer{image_type, id});
+            uniform_texels.emplace(sampler.index, TexelBuffer{image_type, id});
         }
         return binding;
     }
@@ -943,31 +946,48 @@ private:
         return binding;
     }
 
-    u32 DeclareImages(u32 binding) {
+    u32 DeclareStorageTexels(u32 binding) {
         for (const auto& image : ir.GetImages()) {
-            const auto [dim, arrayed] = GetImageDim(image);
-            constexpr int depth = 0;
-            constexpr bool ms = false;
-            constexpr int sampled = 2; // This won't be accessed with a sampler
-            constexpr auto format = spv::ImageFormat::Unknown;
-            const Id image_type = TypeImage(t_uint, dim, depth, arrayed, ms, sampled, format, {});
-            const Id pointer_type = TypePointer(spv::StorageClass::UniformConstant, image_type);
-            const Id id = OpVariable(pointer_type, spv::StorageClass::UniformConstant);
-            AddGlobalVariable(Name(id, fmt::format("image_{}", image.index)));
-
-            Decorate(id, spv::Decoration::Binding, binding++);
-            Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET);
-            if (image.is_read && !image.is_written) {
-                Decorate(id, spv::Decoration::NonWritable);
-            } else if (image.is_written && !image.is_read) {
-                Decorate(id, spv::Decoration::NonReadable);
+            if (image.type != Tegra::Shader::ImageType::TextureBuffer) {
+                continue;
             }
+            DeclareImage(image, binding);
+        }
+        return binding;
+    }
 
-            images.emplace(image.index, StorageImage{image_type, id});
+    u32 DeclareImages(u32 binding) {
+        for (const auto& image : ir.GetImages()) {
+            if (image.type == Tegra::Shader::ImageType::TextureBuffer) {
+                continue;
+            }
+            DeclareImage(image, binding);
         }
         return binding;
     }
 
+    void DeclareImage(const Image& image, u32& binding) {
+        const auto [dim, arrayed] = GetImageDim(image);
+        constexpr int depth = 0;
+        constexpr bool ms = false;
+        constexpr int sampled = 2; // This won't be accessed with a sampler
+        const auto format = image.is_atomic ? spv::ImageFormat::R32ui : spv::ImageFormat::Unknown;
+        const Id image_type = TypeImage(t_uint, dim, depth, arrayed, ms, sampled, format, {});
+        const Id pointer_type = TypePointer(spv::StorageClass::UniformConstant, image_type);
+        const Id id = OpVariable(pointer_type, spv::StorageClass::UniformConstant);
+        AddGlobalVariable(Name(id, fmt::format("image_{}", image.index)));
+
+        Decorate(id, spv::Decoration::Binding, binding++);
+        Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET);
+        if (image.is_read && !image.is_written) {
+            Decorate(id, spv::Decoration::NonWritable);
+        } else if (image.is_written && !image.is_read) {
+            Decorate(id, spv::Decoration::NonReadable);
+        }
+
+        images.emplace(image.index, StorageImage{image_type, id});
+    }
+
     bool IsRenderTargetEnabled(u32 rt) const {
         for (u32 component = 0; component < 4; ++component) {
             if (header.ps.IsColorComponentOutputEnabled(rt, component)) {
@@ -986,6 +1006,10 @@ private:
         return stage == ShaderType::TesselationControl;
     }
 
+    bool IsAttributeEnabled(u32 location) const {
+        return stage != ShaderType::Vertex || specialization.enabled_attributes[location];
+    }
+
     u32 GetNumInputVertices() const {
         switch (stage) {
         case ShaderType::Geometry:
@@ -1201,16 +1225,20 @@ private:
                 UNIMPLEMENTED_MSG("Unmanaged FrontFacing element={}", element);
                 return {v_float_zero, Type::Float};
             default:
-                if (IsGenericAttribute(attribute)) {
-                    const u32 location = GetGenericAttributeLocation(attribute);
-                    const auto type_descriptor = GetAttributeType(location);
-                    const Type type = type_descriptor.type;
-                    const Id attribute_id = input_attributes.at(attribute);
-                    const std::vector elements = {element};
-                    const Id pointer = ArrayPass(type_descriptor.scalar, attribute_id, elements);
-                    return {OpLoad(GetTypeDefinition(type), pointer), type};
+                if (!IsGenericAttribute(attribute)) {
+                    break;
                 }
-                break;
+                const u32 location = GetGenericAttributeLocation(attribute);
+                if (!IsAttributeEnabled(location)) {
+                    // Disabled attributes (also known as constant attributes) always return zero.
+                    return {v_float_zero, Type::Float};
+                }
+                const auto type_descriptor = GetAttributeType(location);
+                const Type type = type_descriptor.type;
+                const Id attribute_id = input_attributes.at(attribute);
+                const std::vector elements = {element};
+                const Id pointer = ArrayPass(type_descriptor.scalar, attribute_id, elements);
+                return {OpLoad(GetTypeDefinition(type), pointer), type};
             }
             UNIMPLEMENTED_MSG("Unhandled input attribute: {}", static_cast<u32>(attribute));
             return {v_float_zero, Type::Float};
@@ -1246,7 +1274,7 @@ private:
                 } else {
                     UNREACHABLE_MSG("Unmanaged offset node type");
                 }
-                pointer = OpAccessChain(t_cbuf_float, buffer_id, Constant(t_uint, 0), buffer_index,
+                pointer = OpAccessChain(t_cbuf_float, buffer_id, v_uint_zero, buffer_index,
                                         buffer_element);
             }
             return {OpLoad(t_float, pointer), Type::Float};
@@ -1601,7 +1629,7 @@ private:
 
         const Id result = OpIAddCarry(TypeStruct({t_uint, t_uint}), op_a, op_b);
         const Id carry = OpCompositeExtract(t_uint, result, 1);
-        return {OpINotEqual(t_bool, carry, Constant(t_uint, 0)), Type::Bool};
+        return {OpINotEqual(t_bool, carry, v_uint_zero), Type::Bool};
     }
 
     Expression LogicalAssign(Operation operation) {
@@ -1664,7 +1692,7 @@ private:
         const auto& meta = std::get<MetaTexture>(operation.GetMeta());
         const u32 index = meta.sampler.index;
         if (meta.sampler.is_buffer) {
-            const auto& entry = texel_buffers.at(index);
+            const auto& entry = uniform_texels.at(index);
             return OpLoad(entry.image_type, entry.image);
         } else {
             const auto& entry = sampled_images.at(index);
@@ -1941,39 +1969,20 @@ private:
         return {};
     }
 
-    Expression AtomicImageAdd(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Expression AtomicImageMin(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Expression AtomicImageMax(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Expression AtomicImageAnd(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Expression AtomicImageOr(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
+    template <Id (Module::*func)(Id, Id, Id, Id, Id)>
+    Expression AtomicImage(Operation operation) {
+        const auto& meta{std::get<MetaImage>(operation.GetMeta())};
+        ASSERT(meta.values.size() == 1);
 
-    Expression AtomicImageXor(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
+        const Id coordinate = GetCoordinates(operation, Type::Int);
+        const Id image = images.at(meta.image.index).image;
+        const Id sample = v_uint_zero;
+        const Id pointer = OpImageTexelPointer(t_image_uint, image, coordinate, sample);
 
-    Expression AtomicImageExchange(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
+        const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device));
+        const Id semantics = v_uint_zero;
+        const Id value = AsUint(Visit(meta.values[0]));
+        return {(this->*func)(t_uint, pointer, scope, semantics, value), Type::Uint};
     }
 
     template <Id (Module::*func)(Id, Id, Id, Id, Id)>
@@ -1988,7 +1997,7 @@ private:
             return {v_float_zero, Type::Float};
         }
         const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device));
-        const Id semantics = Constant(t_uint, 0);
+        const Id semantics = v_uint_zero;
         const Id value = AsUint(Visit(operation[1]));
 
         return {(this->*func)(t_uint, pointer, scope, semantics, value), Type::Uint};
@@ -2612,11 +2621,11 @@ private:
 
         &SPIRVDecompiler::ImageLoad,
         &SPIRVDecompiler::ImageStore,
-        &SPIRVDecompiler::AtomicImageAdd,
-        &SPIRVDecompiler::AtomicImageAnd,
-        &SPIRVDecompiler::AtomicImageOr,
-        &SPIRVDecompiler::AtomicImageXor,
-        &SPIRVDecompiler::AtomicImageExchange,
+        &SPIRVDecompiler::AtomicImage<&Module::OpAtomicIAdd>,
+        &SPIRVDecompiler::AtomicImage<&Module::OpAtomicAnd>,
+        &SPIRVDecompiler::AtomicImage<&Module::OpAtomicOr>,
+        &SPIRVDecompiler::AtomicImage<&Module::OpAtomicXor>,
+        &SPIRVDecompiler::AtomicImage<&Module::OpAtomicExchange>,
 
         &SPIRVDecompiler::Atomic<&Module::OpAtomicExchange>,
         &SPIRVDecompiler::Atomic<&Module::OpAtomicIAdd>,
@@ -2758,8 +2767,11 @@ private:
         Decorate(TypeStruct(t_gmem_array), spv::Decoration::Block), 0, spv::Decoration::Offset, 0);
     const Id t_gmem_ssbo = TypePointer(spv::StorageClass::StorageBuffer, t_gmem_struct);
 
+    const Id t_image_uint = TypePointer(spv::StorageClass::Image, t_uint);
+
     const Id v_float_zero = Constant(t_float, 0.0f);
     const Id v_float_one = Constant(t_float, 1.0f);
+    const Id v_uint_zero = Constant(t_uint, 0);
 
     // Nvidia uses these defaults for varyings (e.g. position and generic attributes)
     const Id v_varying_default =
@@ -2784,15 +2796,16 @@ private:
     std::unordered_map<u8, GenericVaryingDescription> output_attributes;
     std::map<u32, Id> constant_buffers;
     std::map<GlobalMemoryBase, Id> global_buffers;
-    std::map<u32, TexelBuffer> texel_buffers;
+    std::map<u32, TexelBuffer> uniform_texels;
     std::map<u32, SampledImage> sampled_images;
+    std::map<u32, TexelBuffer> storage_texels;
     std::map<u32, StorageImage> images;
 
+    std::array<Id, Maxwell::NumRenderTargets> frag_colors{};
     Id instance_index{};
     Id vertex_index{};
     Id base_instance{};
     Id base_vertex{};
-    std::array<Id, Maxwell::NumRenderTargets> frag_colors{};
     Id frag_depth{};
     Id frag_coord{};
     Id front_facing{};
@@ -3048,13 +3061,17 @@ ShaderEntries GenerateShaderEntries(const VideoCommon::Shader::ShaderIR& ir) {
     }
     for (const auto& sampler : ir.GetSamplers()) {
         if (sampler.is_buffer) {
-            entries.texel_buffers.emplace_back(sampler);
+            entries.uniform_texels.emplace_back(sampler);
         } else {
             entries.samplers.emplace_back(sampler);
         }
     }
     for (const auto& image : ir.GetImages()) {
-        entries.images.emplace_back(image);
+        if (image.type == Tegra::Shader::ImageType::TextureBuffer) {
+            entries.storage_texels.emplace_back(image);
+        } else {
+            entries.images.emplace_back(image);
+        }
     }
     for (const auto& attribute : ir.GetInputAttributes()) {
         if (IsGenericAttribute(attribute)) {

src/video_core/renderer_vulkan/vk_shader_decompiler.h

@@ -21,8 +21,9 @@ class VKDevice;
 namespace Vulkan {
 
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
-using TexelBufferEntry = VideoCommon::Shader::Sampler;
+using UniformTexelEntry = VideoCommon::Shader::Sampler;
 using SamplerEntry = VideoCommon::Shader::Sampler;
+using StorageTexelEntry = VideoCommon::Shader::Image;
 using ImageEntry = VideoCommon::Shader::Image;
 
 constexpr u32 DESCRIPTOR_SET = 0;
@@ -66,13 +67,15 @@ private:
 struct ShaderEntries {
     u32 NumBindings() const {
         return static_cast<u32>(const_buffers.size() + global_buffers.size() +
-                                texel_buffers.size() + samplers.size() + images.size());
+                                uniform_texels.size() + samplers.size() + storage_texels.size() +
+                                images.size());
     }
 
     std::vector<ConstBufferEntry> const_buffers;
     std::vector<GlobalBufferEntry> global_buffers;
-    std::vector<TexelBufferEntry> texel_buffers;
+    std::vector<UniformTexelEntry> uniform_texels;
     std::vector<SamplerEntry> samplers;
+    std::vector<StorageTexelEntry> storage_texels;
     std::vector<ImageEntry> images;
     std::set<u32> attributes;
     std::array<bool, Maxwell::NumClipDistances> clip_distances{};
@@ -88,7 +91,8 @@ struct Specialization final {
     u32 shared_memory_size{};
 
     // Graphics specific
-    std::optional<float> point_size{};
+    std::optional<float> point_size;
+    std::bitset<Maxwell::NumVertexAttributes> enabled_attributes;
     std::array<Maxwell::VertexAttribute::Type, Maxwell::NumVertexAttributes> attribute_types{};
     bool ndc_minus_one_to_one{};
 };