[vulkan] fix vuid 02999

Normally,  language_entries contained plain UTF-8 names.
In BotW v1.9.0, that title block is no longer directly readable as old UTF-8, so we were parsing binary data as text and going nuts.
...
In patch_manager.cpp (ParseControlNCA), I added a validity check: If update title text is unreadable, we keep update metadata but copy only the base language_entries block (0x0000..0x2FFF) so the game name is valid again.

UPDATE:
managed to decode the new language entries is a raw headerless zlib deflate.
added support for proper detection and inflation.

Co-authored-by: crueter <crueter@eden-emu.dev>
Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/3585
Reviewed-by: crueter <crueter@eden-emu.dev>
Reviewed-by: MaranBr <maranbr@eden-emu.dev>
Reviewed-by: CamilleLaVey <camillelavey99@gmail.com>
Co-authored-by: xbzk <xbzk@eden-emu.dev>
Co-committed-by: xbzk <xbzk@eden-emu.dev>

externals/CMakeLists.txt

@@ -407,12 +407,10 @@ if (APPLE)
     # moltenvk
     if (NOT YUZU_USE_BUNDLED_MOLTENVK)
         find_library(MOLTENVK_LIBRARY MoltenVK)
-    else()
-        unset(MOLTENVK_LIBRARY)
     endif()
 
     # TODO: kosmickrisp?
-    if (NOT MOLTENVK_LIBRARY)
+    if (NOT MOLTENVK_LIBRARY OR YUZU_USE_BUNDLED_MOLTENVK)
         AddJsonPackage(moltenvk)
 
         set(MOLTENVK_LIBRARY "${moltenvk_SOURCE_DIR}/MoltenVK/dylib/macOS/libMoltenVK.dylib" CACHE STRING "" FORCE)

src/common/host_memory.cpp

@@ -219,11 +219,6 @@ public:
         }
     }
 
-    bool ClearBackingRegion(size_t physical_offset, size_t length) {
-        // TODO: This does not seem to be possible on Windows.
-        return false;
-    }
-
     void EnableDirectMappedAddress() {
         // TODO
         UNREACHABLE();
@@ -617,24 +612,6 @@ public:
         ASSERT_MSG(ret == 0, "mprotect failed: {}", strerror(errno));
     }
 
-    bool ClearBackingRegion(size_t physical_offset, size_t length) {
-#ifdef __linux__
-        // Only incur syscall cost IF memset would be slower (theshold = 16MiB)
-        // TODO(lizzie): Smarter way to dynamically get this threshold (broadwell != raptor lake) for example
-        if (length >= 2097152UL * 8) {
-            // Set MADV_REMOVE on backing map to destroy it instantly.
-            // This also deletes the area from the backing file.
-            int ret = madvise(backing_base + physical_offset, length, MADV_REMOVE);
-            ASSERT_MSG(ret == 0, "madvise failed: {}", strerror(errno));
-            return true;
-        } else {
-            return false;
-        }
-#else
-        return false;
-#endif
-    }
-
     void EnableDirectMappedAddress() {
         virtual_base = nullptr;
     }
@@ -748,9 +725,7 @@ void HostMemory::Protect(size_t virtual_offset, size_t length, MemoryPermission
 }
 
 void HostMemory::ClearBackingRegion(size_t physical_offset, size_t length, u32 fill_value) {
-    if (!impl || fill_value != 0 || !impl->ClearBackingRegion(physical_offset, length)) {
-        std::memset(backing_base + physical_offset, fill_value, length);
-    }
+    std::memset(backing_base + physical_offset, fill_value, length);
 }
 
 void HostMemory::EnableDirectMappedAddress() {

src/core/CMakeLists.txt

@@ -1219,6 +1219,7 @@ target_link_libraries(core PRIVATE
     fmt::fmt
     nlohmann_json::nlohmann_json
     RenderDoc::API
+    ZLIB::ZLIB
     MbedTLS::mbedcrypto${MBEDTLS_LIB_SUFFIX}
     MbedTLS::mbedtls${MBEDTLS_LIB_SUFFIX})
 

src/core/device_memory_manager.h

@@ -127,6 +127,11 @@ public:
     void UpdatePagesCachedBatch(std::span<const std::pair<DAddr, size_t>> ranges, s32 delta);
 
 private:
+    struct TranslationEntry {
+        DAddr guest_page{};
+        u8* host_ptr{};
+    };
+
     // Internal helper that performs the update assuming the caller already holds the necessary lock.
     void UpdatePagesCachedCountNoLock(DAddr addr, size_t size, s32 delta);
 
@@ -195,6 +200,8 @@ private:
     }
 
     Common::VirtualBuffer<VAddr> cpu_backing_address;
+    std::array<TranslationEntry, 4> t_slot{};
+    u32 cache_cursor = 0;
     using CounterType = u8;
     using CounterAtomicType = std::atomic_uint8_t;
     static constexpr size_t subentries = 8 / sizeof(CounterType);

src/core/device_memory_manager.inc

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
@@ -247,6 +247,7 @@ void DeviceMemoryManager<Traits>::Map(DAddr address, VAddr virtual_address, size
         }
         impl->multi_dev_address.Register(new_dev, start_id);
     }
+    t_slot = {};
     if (track) {
         TrackContinuityImpl(address, virtual_address, size, asid);
     }
@@ -278,6 +279,7 @@ void DeviceMemoryManager<Traits>::Unmap(DAddr address, size_t size) {
             compressed_device_addr[phys_addr - 1] = new_start | MULTI_FLAG;
         }
     }
+    t_slot = {};
 }
 template <typename Traits>
 void DeviceMemoryManager<Traits>::TrackContinuityImpl(DAddr address, VAddr virtual_address,
@@ -417,6 +419,26 @@ void DeviceMemoryManager<Traits>::WalkBlock(DAddr addr, std::size_t size, auto o
 template <typename Traits>
 void DeviceMemoryManager<Traits>::ReadBlock(DAddr address, void* dest_pointer, size_t size) {
     device_inter->FlushRegion(address, size);
+    const std::size_t page_offset = address & Memory::YUZU_PAGEMASK;
+    if (size <= Memory::YUZU_PAGESIZE - page_offset) {
+        const DAddr guest_page = address & ~static_cast<DAddr>(Memory::YUZU_PAGEMASK);
+        for (size_t i = 0; i < 4; ++i) {
+            if (t_slot[i].guest_page == guest_page && t_slot[i].host_ptr != nullptr) {
+                std::memcpy(dest_pointer, t_slot[i].host_ptr + page_offset, size);
+                return;
+            }
+        }
+
+        const std::size_t page_index = address >> Memory::YUZU_PAGEBITS;
+        const auto phys_addr = compressed_physical_ptr[page_index];
+        if (phys_addr != 0) {
+            auto* const mem_ptr = GetPointerFromRaw<u8>((PAddr(phys_addr - 1) << Memory::YUZU_PAGEBITS));
+            t_slot[cache_cursor % t_slot.size()] = TranslationEntry{.guest_page = guest_page, .host_ptr = mem_ptr};
+            cache_cursor = (cache_cursor + 1) & 3U;
+            std::memcpy(dest_pointer, mem_ptr + page_offset, size);
+            return;
+        }
+    }
     WalkBlock(
         address, size,
         [&](size_t copy_amount, DAddr current_vaddr) {
@@ -455,6 +477,26 @@ void DeviceMemoryManager<Traits>::WriteBlock(DAddr address, const void* src_poin
 
 template <typename Traits>
 void DeviceMemoryManager<Traits>::ReadBlockUnsafe(DAddr address, void* dest_pointer, size_t size) {
+    const std::size_t page_offset = address & Memory::YUZU_PAGEMASK;
+    if (size <= Memory::YUZU_PAGESIZE - page_offset) {
+        const DAddr guest_page = address & ~static_cast<DAddr>(Memory::YUZU_PAGEMASK);
+        for (size_t i = 0; i < 4; ++i) {
+            if (t_slot[i].guest_page == guest_page && t_slot[i].host_ptr != nullptr) {
+                std::memcpy(dest_pointer, t_slot[i].host_ptr + page_offset, size);
+                return;
+            }
+        }
+
+        const std::size_t page_index = address >> Memory::YUZU_PAGEBITS;
+        const auto phys_addr = compressed_physical_ptr[page_index];
+        if (phys_addr != 0) {
+            auto* const mem_ptr = GetPointerFromRaw<u8>((PAddr(phys_addr - 1) << Memory::YUZU_PAGEBITS));
+            t_slot[cache_cursor % t_slot.size()] = TranslationEntry{.guest_page = guest_page, .host_ptr = mem_ptr};
+            cache_cursor = (cache_cursor + 1) & 3U;
+            std::memcpy(dest_pointer, mem_ptr + page_offset, size);
+            return;
+        }
+    }
     WalkBlock(
         address, size,
         [&](size_t copy_amount, DAddr current_vaddr) {

src/core/file_sys/control_metadata.cpp

@@ -4,6 +4,13 @@
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <array>
+#include <cstddef>
+#include <cstring>
+#include <limits>
+#include <span>
+#include <zlib.h>
+
 #include "common/settings.h"
 #include "common/string_util.h"
 #include "common/swap.h"
@@ -31,6 +38,73 @@ const std::array<const char*, 16> LANGUAGE_NAMES{{
     "BrazilianPortuguese",
 }};
 
+namespace {
+constexpr std::size_t LEGACY_LANGUAGE_REGION_SIZE = sizeof(std::array<LanguageEntry, 16>);
+constexpr std::size_t PACKED_LANGUAGE_REGION_MAX_SIZE = sizeof(LanguageEntry) * 32;
+
+bool InflateRawDeflate(std::span<const u8> compressed, std::vector<u8>& out) {
+    if (compressed.empty() || compressed.size() > std::numeric_limits<uInt>::max()) {
+        return false;
+    }
+
+    z_stream stream{};
+    stream.next_in = const_cast<Bytef*>(reinterpret_cast<const Bytef*>(compressed.data()));
+    stream.avail_in = static_cast<uInt>(compressed.size());
+    if (inflateInit2(&stream, -MAX_WBITS) != Z_OK) {
+        return false;
+    }
+
+    std::array<u8, 0x1000> chunk{};
+    int ret = Z_OK;
+    while (ret == Z_OK) {
+        stream.next_out = reinterpret_cast<Bytef*>(chunk.data());
+        stream.avail_out = static_cast<uInt>(chunk.size());
+        ret = inflate(&stream, Z_NO_FLUSH);
+        if (ret != Z_OK && ret != Z_STREAM_END) {
+            inflateEnd(&stream);
+            return false;
+        }
+
+        const auto produced = chunk.size() - static_cast<std::size_t>(stream.avail_out);
+        if (produced != 0) {
+            if (out.size() + produced > PACKED_LANGUAGE_REGION_MAX_SIZE) {
+                inflateEnd(&stream);
+                return false;
+            }
+            out.insert(out.end(), chunk.begin(),
+                       chunk.begin() + static_cast<std::ptrdiff_t>(produced));
+        }
+    }
+
+    inflateEnd(&stream);
+    return ret == Z_STREAM_END;
+}
+
+void DecodePackedLanguageEntries(RawNACP& raw) {
+    auto* packed_region = reinterpret_cast<u8*>(raw.language_entries.data());
+    u16_le compressed_size_le{};
+    std::memcpy(&compressed_size_le, packed_region, sizeof(compressed_size_le));
+    const auto compressed_size = static_cast<std::size_t>(compressed_size_le);
+
+    if (compressed_size == 0 || compressed_size > LEGACY_LANGUAGE_REGION_SIZE - sizeof(u16_le)) {
+        return;
+    }
+
+    std::vector<u8> decompressed;
+    if (!InflateRawDeflate(
+            std::span<const u8>(packed_region + sizeof(u16_le), compressed_size), decompressed)) {
+        return;
+    }
+
+    if (decompressed.size() < LEGACY_LANGUAGE_REGION_SIZE ||
+        decompressed.size() % sizeof(LanguageEntry) != 0) {
+        return;
+    }
+
+    std::memcpy(raw.language_entries.data(), decompressed.data(), LEGACY_LANGUAGE_REGION_SIZE);
+}
+} // namespace
+
 std::string LanguageEntry::GetApplicationName() const {
     return Common::StringFromFixedZeroTerminatedBuffer(application_name.data(),
                                                        application_name.size());
@@ -66,6 +140,7 @@ NACP::NACP() = default;
 
 NACP::NACP(VirtualFile file) {
     file->ReadObject(&raw);
+    DecodePackedLanguageEntries(raw);
 }
 
 NACP::~NACP() = default;

src/dynarmic/src/dynarmic/CMakeLists.txt

@@ -105,9 +105,6 @@ add_library(dynarmic STATIC
     frontend/A32/decoder/thumb32.inc
     frontend/A32/decoder/vfp.h
     frontend/A32/decoder/vfp.inc
-    frontend/A32/disassembler/disassembler.h
-    frontend/A32/disassembler/disassembler_arm.cpp
-    frontend/A32/disassembler/disassembler_thumb.cpp
     frontend/A32/FPSCR.h
     frontend/A32/ITState.h
     frontend/A32/PSR.h
@@ -122,7 +119,6 @@ add_library(dynarmic STATIC
     interface/A32/config.h
     interface/A32/coprocessor.h
     interface/A32/coprocessor_util.h
-    interface/A32/disassembler.h
     # A64
     frontend/A64/a64_ir_emitter.cpp
     frontend/A64/a64_ir_emitter.h

src/dynarmic/src/dynarmic/backend/x64/a32_emit_x64_memory.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -14,7 +14,7 @@
 #include <fmt/format.h>
 #include <fmt/ostream.h>
 #include <mcl/type_traits/integer_of_size.hpp>
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 
 #include "dynarmic/backend/x64/a32_emit_x64.h"
 #include "dynarmic/backend/x64/abi.h"

src/dynarmic/src/dynarmic/backend/x64/a64_emit_x64_memory.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -14,7 +14,7 @@
 #include <fmt/format.h>
 #include <fmt/ostream.h>
 #include <mcl/type_traits/integer_of_size.hpp>
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 
 #include "dynarmic/backend/x64/a64_emit_x64.h"
 #include "dynarmic/backend/x64/abi.h"

src/dynarmic/src/dynarmic/backend/x64/abi.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -11,7 +11,7 @@
 #include <algorithm>
 
 #include "dynarmic/common/common_types.h"
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 
 #include "dynarmic/backend/x64/block_of_code.h"
 

src/dynarmic/src/dynarmic/backend/x64/block_of_code.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -26,7 +26,7 @@
 
 #include "dynarmic/common/assert.h"
 #include <mcl/bit/bit_field.hpp>
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 
 #include "dynarmic/backend/x64/a32_jitstate.h"
 #include "dynarmic/backend/x64/abi.h"

src/dynarmic/src/dynarmic/backend/x64/block_of_code.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -15,9 +15,7 @@
 
 #include <mcl/bit/bit_field.hpp>
 #include "dynarmic/common/common_types.h"
-#include <xbyak/xbyak.h>
-#include <xbyak/xbyak_util.h>
-
+#include "dynarmic/backend/x64/xbyak.h"
 #include "dynarmic/backend/x64/abi.h"
 #include "dynarmic/backend/x64/callback.h"
 #include "dynarmic/backend/x64/constant_pool.h"

src/dynarmic/src/dynarmic/backend/x64/callback.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -12,7 +12,7 @@
 #include <vector>
 
 #include "dynarmic/common/common_types.h"
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 
 namespace Dynarmic::Backend::X64 {
 

src/dynarmic/src/dynarmic/backend/x64/constant_pool.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -15,7 +15,7 @@
 
 #include "dynarmic/common/common_types.h"
 #include <ankerl/unordered_dense.h>
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 
 namespace Dynarmic::Backend::X64 {
 

src/dynarmic/src/dynarmic/backend/x64/emit_x64.h

@@ -18,8 +18,7 @@
 
 #include <mcl/bitsizeof.hpp>
 #include <ankerl/unordered_dense.h>
-#include <xbyak/xbyak.h>
-#include <xbyak/xbyak_util.h>
+#include "dynarmic/backend/x64/xbyak.h"
 #include <boost/container/small_vector.hpp>
 
 #include "dynarmic/backend/exception_handler.h"

src/dynarmic/src/dynarmic/backend/x64/emit_x64_floating_point.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -19,7 +19,7 @@
 #include <mcl/mp/typelist/lower_to_tuple.hpp>
 #include "dynarmic/common/common_types.h"
 #include <mcl/type_traits/integer_of_size.hpp>
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 
 #include "dynarmic/backend/x64/abi.h"
 #include "dynarmic/backend/x64/block_of_code.h"

src/dynarmic/src/dynarmic/backend/x64/emit_x64_memory.h

@@ -7,7 +7,7 @@
  */
 
 #include <bit>
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 
 #include "dynarmic/backend/x64/a32_emit_x64.h"
 #include "dynarmic/backend/x64/a64_emit_x64.h"

src/dynarmic/src/dynarmic/backend/x64/emit_x64_vector.cpp

@@ -18,7 +18,7 @@
 #include <mcl/bitsizeof.hpp>
 #include "dynarmic/common/common_types.h"
 #include <mcl/type_traits/function_info.hpp>
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 
 #include "dynarmic/backend/x64/abi.h"
 #include "dynarmic/backend/x64/block_of_code.h"

src/dynarmic/src/dynarmic/backend/x64/emit_x64_vector_floating_point.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -21,7 +21,7 @@
 #include <mcl/mp/typelist/lower_to_tuple.hpp>
 #include <mcl/type_traits/function_info.hpp>
 #include <mcl/type_traits/integer_of_size.hpp>
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 
 #include "dynarmic/backend/x64/abi.h"
 #include "dynarmic/backend/x64/block_of_code.h"

src/dynarmic/src/dynarmic/backend/x64/hostloc.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -9,7 +9,7 @@
 
 #include "dynarmic/common/assert.h"
 #include "dynarmic/common/common_types.h"
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 
 namespace Dynarmic::Backend::X64 {
 

src/dynarmic/src/dynarmic/backend/x64/oparg.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -9,7 +9,7 @@
 #pragma once
 
 #include "dynarmic/common/assert.h"
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 
 namespace Dynarmic::Backend::X64 {
 

src/dynarmic/src/dynarmic/backend/x64/reg_alloc.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -16,7 +16,7 @@
 #include <fmt/ostream.h>
 #include "dynarmic/common/assert.h"
 #include <bit>
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 
 #include "dynarmic/backend/x64/abi.h"
 #include "dynarmic/backend/x64/stack_layout.h"

src/dynarmic/src/dynarmic/backend/x64/reg_alloc.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -14,7 +14,7 @@
 
 #include "boost/container/small_vector.hpp"
 #include "dynarmic/common/common_types.h"
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 #include <boost/container/static_vector.hpp>
 #include <boost/container/flat_set.hpp>
 #include <boost/pool/pool_alloc.hpp>

src/dynarmic/src/dynarmic/backend/x64/xbyak.h

@@ -0,0 +1,15 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+#include <unordered_map>
+#include <unordered_set>
+
+// TODO: Defining this crashes e v e r y t h i n g
+// #define XBYAK_STD_UNORDERED_SET ankerl::unordered_dense::set
+// #define XBYAK_STD_UNORDERED_MAP ankerl::unordered_dense::map
+// #define XBYAK_STD_UNORDERED_MULTIMAP boost::unordered_multimap
+
+#include <xbyak/xbyak.h>
+#include <xbyak/xbyak_util.h>

src/dynarmic/src/dynarmic/common/spin_lock_x64.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -8,7 +8,7 @@
 
 #include <mutex>
 #include <optional>
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 
 #include "dynarmic/backend/x64/abi.h"
 #include "dynarmic/backend/x64/hostloc.h"

src/dynarmic/src/dynarmic/common/spin_lock_x64.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -8,7 +8,7 @@
 
 #pragma once
 
-#include <xbyak/xbyak.h>
+#include "dynarmic/backend/x64/xbyak.h"
 
 namespace Dynarmic {
 

src/dynarmic/src/dynarmic/frontend/A32/disassembler/disassembler.h

@@ -1,20 +0,0 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
-/* This file is part of the dynarmic project.
- * Copyright (c) 2016 MerryMage
- * SPDX-License-Identifier: 0BSD
- */
-
-#pragma once
-
-#include <string>
-
-#include "dynarmic/common/common_types.h"
-
-namespace Dynarmic::A32 {
-
-std::string DisassembleArm(u32 instruction);
-std::string DisassembleThumb16(u16 instruction);
-
-}  // namespace Dynarmic::A32

src/dynarmic/src/dynarmic/frontend/A32/disassembler/disassembler_thumb.cpp

@@ -1,397 +0,0 @@
-/* This file is part of the dynarmic project.
- * Copyright (c) 2016 MerryMage
- * SPDX-License-Identifier: 0BSD
- */
-
-#include <cstdlib>
-#include <string>
-
-#include <fmt/format.h>
-#include <fmt/ostream.h>
-#include <mcl/bit/bit_field.hpp>
-
-#include "dynarmic/common/string_util.h"
-#include "dynarmic/frontend/A32/a32_types.h"
-#include "dynarmic/frontend/A32/decoder/thumb16.h"
-#include "dynarmic/frontend/A32/disassembler/disassembler.h"
-#include "dynarmic/frontend/imm.h"
-
-namespace Dynarmic::A32 {
-
-class DisassemblerVisitor {
-public:
-    using instruction_return_type = std::string;
-
-    std::string thumb16_LSL_imm(Imm<5> imm5, Reg m, Reg d) {
-        return fmt::format("lsls {}, {}, #{}", d, m, imm5.ZeroExtend());
-    }
-
-    std::string thumb16_LSR_imm(Imm<5> imm5, Reg m, Reg d) {
-        const u32 shift = imm5 != 0 ? imm5.ZeroExtend() : 32U;
-        return fmt::format("lsrs {}, {}, #{}", d, m, shift);
-    }
-
-    std::string thumb16_ASR_imm(Imm<5> imm5, Reg m, Reg d) {
-        const u32 shift = imm5 != 0 ? imm5.ZeroExtend() : 32U;
-        return fmt::format("asrs {}, {}, #{}", d, m, shift);
-    }
-
-    std::string thumb16_ADD_reg_t1(Reg m, Reg n, Reg d) {
-        return fmt::format("adds {}, {}, {}", d, n, m);
-    }
-
-    std::string thumb16_SUB_reg(Reg m, Reg n, Reg d) {
-        return fmt::format("subs {}, {}, {}", d, n, m);
-    }
-
-    std::string thumb16_ADD_imm_t1(Imm<3> imm3, Reg n, Reg d) {
-        return fmt::format("adds {}, {}, #{}", d, n, imm3.ZeroExtend());
-    }
-
-    std::string thumb16_SUB_imm_t1(Imm<3> imm3, Reg n, Reg d) {
-        return fmt::format("subs {}, {}, #{}", d, n, imm3.ZeroExtend());
-    }
-
-    std::string thumb16_MOV_imm(Reg d, Imm<8> imm8) {
-        return fmt::format("movs {}, #{}", d, imm8.ZeroExtend());
-    }
-
-    std::string thumb16_CMP_imm(Reg n, Imm<8> imm8) {
-        return fmt::format("cmp {}, #{}", n, imm8.ZeroExtend());
-    }
-
-    std::string thumb16_ADD_imm_t2(Reg d_n, Imm<8> imm8) {
-        return fmt::format("adds {}, #{}", d_n, imm8.ZeroExtend());
-    }
-
-    std::string thumb16_SUB_imm_t2(Reg d_n, Imm<8> imm8) {
-        return fmt::format("subs {}, #{}", d_n, imm8.ZeroExtend());
-    }
-
-    std::string thumb16_AND_reg(Reg m, Reg d_n) {
-        return fmt::format("ands {}, {}", d_n, m);
-    }
-
-    std::string thumb16_EOR_reg(Reg m, Reg d_n) {
-        return fmt::format("eors {}, {}", d_n, m);
-    }
-
-    std::string thumb16_LSL_reg(Reg m, Reg d_n) {
-        return fmt::format("lsls {}, {}", d_n, m);
-    }
-
-    std::string thumb16_LSR_reg(Reg m, Reg d_n) {
-        return fmt::format("lsrs {}, {}", d_n, m);
-    }
-
-    std::string thumb16_ASR_reg(Reg m, Reg d_n) {
-        return fmt::format("asrs {}, {}", d_n, m);
-    }
-
-    std::string thumb16_ADC_reg(Reg m, Reg d_n) {
-        return fmt::format("adcs {}, {}", d_n, m);
-    }
-
-    std::string thumb16_SBC_reg(Reg m, Reg d_n) {
-        return fmt::format("sbcs {}, {}", d_n, m);
-    }
-
-    std::string thumb16_ROR_reg(Reg m, Reg d_n) {
-        return fmt::format("rors {}, {}", d_n, m);
-    }
-
-    std::string thumb16_TST_reg(Reg m, Reg n) {
-        return fmt::format("tst {}, {}", n, m);
-    }
-
-    std::string thumb16_RSB_imm(Reg n, Reg d) {
-        // Pre-UAL syntax: NEGS <Rd>, <Rn>
-        return fmt::format("rsbs {}, {}, #0", d, n);
-    }
-
-    std::string thumb16_CMP_reg_t1(Reg m, Reg n) {
-        return fmt::format("cmp {}, {}", n, m);
-    }
-
-    std::string thumb16_CMN_reg(Reg m, Reg n) {
-        return fmt::format("cmn {}, {}", n, m);
-    }
-
-    std::string thumb16_ORR_reg(Reg m, Reg d_n) {
-        return fmt::format("orrs {}, {}", d_n, m);
-    }
-
-    std::string thumb16_MUL_reg(Reg n, Reg d_m) {
-        return fmt::format("muls {}, {}, {}", d_m, n, d_m);
-    }
-
-    std::string thumb16_BIC_reg(Reg m, Reg d_n) {
-        return fmt::format("bics {}, {}", d_n, m);
-    }
-
-    std::string thumb16_MVN_reg(Reg m, Reg d) {
-        return fmt::format("mvns {}, {}", d, m);
-    }
-
-    std::string thumb16_ADD_reg_t2(bool d_n_hi, Reg m, Reg d_n_lo) {
-        const Reg d_n = d_n_hi ? (d_n_lo + 8) : d_n_lo;
-        return fmt::format("add {}, {}", d_n, m);
-    }
-
-    std::string thumb16_CMP_reg_t2(bool n_hi, Reg m, Reg n_lo) {
-        const Reg n = n_hi ? (n_lo + 8) : n_lo;
-        return fmt::format("cmp {}, {}", n, m);
-    }
-
-    std::string thumb16_MOV_reg(bool d_hi, Reg m, Reg d_lo) {
-        const Reg d = d_hi ? (d_lo + 8) : d_lo;
-        return fmt::format("mov {}, {}", d, m);
-    }
-
-    std::string thumb16_LDR_literal(Reg t, Imm<8> imm8) {
-        const u32 imm32 = imm8.ZeroExtend() << 2;
-        return fmt::format("ldr {}, [pc, #{}]", t, imm32);
-    }
-
-    std::string thumb16_STR_reg(Reg m, Reg n, Reg t) {
-        return fmt::format("str {}, [{}, {}]", t, n, m);
-    }
-
-    std::string thumb16_STRH_reg(Reg m, Reg n, Reg t) {
-        return fmt::format("strh {}, [{}, {}]", t, n, m);
-    }
-
-    std::string thumb16_STRB_reg(Reg m, Reg n, Reg t) {
-        return fmt::format("strb {}, [{}, {}]", t, n, m);
-    }
-
-    std::string thumb16_LDRSB_reg(Reg m, Reg n, Reg t) {
-        return fmt::format("ldrsb {}, [{}, {}]", t, n, m);
-    }
-
-    std::string thumb16_LDR_reg(Reg m, Reg n, Reg t) {
-        return fmt::format("ldr {}, [{}, {}]", t, n, m);
-    }
-
-    std::string thumb16_LDRH_reg(Reg m, Reg n, Reg t) {
-        return fmt::format("ldrh {}, [%s, %s]", t, n, m);
-    }
-
-    std::string thumb16_LDRB_reg(Reg m, Reg n, Reg t) {
-        return fmt::format("ldrb {}, [{}, {}]", t, n, m);
-    }
-
-    std::string thumb16_LDRSH_reg(Reg m, Reg n, Reg t) {
-        return fmt::format("ldrsh {}, [{}, {}]", t, n, m);
-    }
-
-    std::string thumb16_STR_imm_t1(Imm<5> imm5, Reg n, Reg t) {
-        const u32 imm32 = imm5.ZeroExtend() << 2;
-        return fmt::format("str {}, [{}, #{}]", t, n, imm32);
-    }
-
-    std::string thumb16_LDR_imm_t1(Imm<5> imm5, Reg n, Reg t) {
-        const u32 imm32 = imm5.ZeroExtend() << 2;
-        return fmt::format("ldr {}, [{}, #{}]", t, n, imm32);
-    }
-
-    std::string thumb16_STRB_imm(Imm<5> imm5, Reg n, Reg t) {
-        const u32 imm32 = imm5.ZeroExtend();
-        return fmt::format("strb {}, [{}, #{}]", t, n, imm32);
-    }
-
-    std::string thumb16_LDRB_imm(Imm<5> imm5, Reg n, Reg t) {
-        const u32 imm32 = imm5.ZeroExtend();
-        return fmt::format("ldrb {}, [{}, #{}]", t, n, imm32);
-    }
-
-    std::string thumb16_STRH_imm(Imm<5> imm5, Reg n, Reg t) {
-        const u32 imm32 = imm5.ZeroExtend() << 1;
-        return fmt::format("strh {}, [{}, #{}]", t, n, imm32);
-    }
-
-    std::string thumb16_LDRH_imm(Imm<5> imm5, Reg n, Reg t) {
-        const u32 imm32 = imm5.ZeroExtend() << 1;
-        return fmt::format("ldrh {}, [{}, #{}]", t, n, imm32);
-    }
-
-    std::string thumb16_STR_imm_t2(Reg t, Imm<8> imm8) {
-        const u32 imm32 = imm8.ZeroExtend() << 2;
-        return fmt::format("str {}, [sp, #{}]", t, imm32);
-    }
-
-    std::string thumb16_LDR_imm_t2(Reg t, Imm<8> imm8) {
-        const u32 imm32 = imm8.ZeroExtend() << 2;
-        return fmt::format("ldr {}, [sp, #{}]", t, imm32);
-    }
-
-    std::string thumb16_ADR(Reg d, Imm<8> imm8) {
-        const u32 imm32 = imm8.ZeroExtend() << 2;
-        return fmt::format("adr {}, +#{}", d, imm32);
-    }
-
-    std::string thumb16_ADD_sp_t1(Reg d, Imm<8> imm8) {
-        const u32 imm32 = imm8.ZeroExtend() << 2;
-        return fmt::format("add {}, sp, #{}", d, imm32);
-    }
-
-    std::string thumb16_ADD_sp_t2(Imm<7> imm7) {
-        const u32 imm32 = imm7.ZeroExtend() << 2;
-        return fmt::format("add sp, sp, #{}", imm32);
-    }
-
-    std::string thumb16_SUB_sp(Imm<7> imm7) {
-        const u32 imm32 = imm7.ZeroExtend() << 2;
-        return fmt::format("sub sp, sp, #{}", imm32);
-    }
-
-    std::string thumb16_NOP() {
-        return "nop";
-    }
-
-    std::string thumb16_SEV() {
-        return "sev";
-    }
-
-    std::string thumb16_SEVL() {
-        return "sevl";
-    }
-
-    std::string thumb16_WFE() {
-        return "wfe";
-    }
-
-    std::string thumb16_WFI() {
-        return "wfi";
-    }
-
-    std::string thumb16_YIELD() {
-        return "yield";
-    }
-
-    std::string thumb16_IT(Imm<8> imm8) {
-        const Cond firstcond = imm8.Bits<4, 7, Cond>();
-        const bool firstcond0 = imm8.Bit<4>();
-        const auto [x, y, z] = [&] {
-            if (imm8.Bits<0, 3>() == 0b1000) {
-                return std::make_tuple("", "", "");
-            }
-            if (imm8.Bits<0, 2>() == 0b100) {
-                return std::make_tuple(imm8.Bit<3>() == firstcond0 ? "t" : "e", "", "");
-            }
-            if (imm8.Bits<0, 1>() == 0b10) {
-                return std::make_tuple(imm8.Bit<3>() == firstcond0 ? "t" : "e", imm8.Bit<2>() == firstcond0 ? "t" : "e", "");
-            }
-            // Sanity note: Here imm8.Bit<0>() is guaranteed to be == 1. (imm8 can never be 0bxxxx0000)
-            return std::make_tuple(imm8.Bit<3>() == firstcond0 ? "t" : "e", imm8.Bit<2>() == firstcond0 ? "t" : "e", imm8.Bit<1>() == firstcond0 ? "t" : "e");
-        }();
-        return fmt::format("it{}{}{} {}", x, y, z, CondToString(firstcond));
-    }
-
-    std::string thumb16_SXTH(Reg m, Reg d) {
-        return fmt::format("sxth {}, {}", d, m);
-    }
-
-    std::string thumb16_SXTB(Reg m, Reg d) {
-        return fmt::format("sxtb {}, {}", d, m);
-    }
-
-    std::string thumb16_UXTH(Reg m, Reg d) {
-        return fmt::format("uxth {}, {}", d, m);
-    }
-
-    std::string thumb16_UXTB(Reg m, Reg d) {
-        return fmt::format("uxtb {}, {}", d, m);
-    }
-
-    std::string thumb16_PUSH(bool M, RegList reg_list) {
-        if (M)
-            reg_list |= 1 << 14;
-        return fmt::format("push {{{}}}", RegListToString(reg_list));
-    }
-
-    std::string thumb16_POP(bool P, RegList reg_list) {
-        if (P)
-            reg_list |= 1 << 15;
-        return fmt::format("pop {{{}}}", RegListToString(reg_list));
-    }
-
-    std::string thumb16_SETEND(bool E) {
-        return fmt::format("setend {}", E ? "BE" : "LE");
-    }
-
-    std::string thumb16_CPS(bool im, bool a, bool i, bool f) {
-        return fmt::format("cps{} {}{}{}", im ? "id" : "ie", a ? "a" : "", i ? "i" : "", f ? "f" : "");
-    }
-
-    std::string thumb16_REV(Reg m, Reg d) {
-        return fmt::format("rev {}, {}", d, m);
-    }
-
-    std::string thumb16_REV16(Reg m, Reg d) {
-        return fmt::format("rev16 {}, {}", d, m);
-    }
-
-    std::string thumb16_REVSH(Reg m, Reg d) {
-        return fmt::format("revsh {}, {}", d, m);
-    }
-
-    std::string thumb16_BKPT(Imm<8> imm8) {
-        return fmt::format("bkpt #{}", imm8.ZeroExtend());
-    }
-
-    std::string thumb16_STMIA(Reg n, RegList reg_list) {
-        return fmt::format("stm {}!, {{{}}}", n, RegListToString(reg_list));
-    }
-
-    std::string thumb16_LDMIA(Reg n, RegList reg_list) {
-        const bool write_back = !mcl::bit::get_bit(static_cast<size_t>(n), reg_list);
-        return fmt::format("ldm {}{}, {{{}}}", n, write_back ? "!" : "", RegListToString(reg_list));
-    }
-
-    std::string thumb16_BX(Reg m) {
-        return fmt::format("bx {}", m);
-    }
-
-    std::string thumb16_BLX_reg(Reg m) {
-        return fmt::format("blx {}", m);
-    }
-
-    std::string thumb16_CBZ_CBNZ(bool nonzero, Imm<1> i, Imm<5> imm5, Reg n) {
-        const char* const name = nonzero ? "cbnz" : "cbz";
-        const u32 imm = concatenate(i, imm5, Imm<1>{0}).ZeroExtend();
-
-        return fmt::format("{} {}, #{}", name, n, imm);
-    }
-
-    std::string thumb16_UDF() {
-        return fmt::format("udf");
-    }
-
-    std::string thumb16_SVC(Imm<8> imm8) {
-        return fmt::format("svc #{}", imm8.ZeroExtend());
-    }
-
-    std::string thumb16_B_t1(Cond cond, Imm<8> imm8) {
-        const s32 imm32 = static_cast<s32>((imm8.SignExtend<u32>() << 1) + 4);
-        return fmt::format("b{} {}#{}",
-                           CondToString(cond),
-                           Common::SignToChar(imm32),
-                           abs(imm32));
-    }
-
-    std::string thumb16_B_t2(Imm<11> imm11) {
-        const s32 imm32 = static_cast<s32>((imm11.SignExtend<u32>() << 1) + 4);
-        return fmt::format("b {}#{}",
-                           Common::SignToChar(imm32),
-                           abs(imm32));
-    }
-};
-
-std::string DisassembleThumb16(u16 instruction) {
-    DisassemblerVisitor visitor;
-    auto decoder = DecodeThumb16<DisassemblerVisitor>(instruction);
-    return !decoder ? fmt::format("UNKNOWN: {:x}", instruction) : decoder->get().call(visitor, instruction);
-}
-
-}  // namespace Dynarmic::A32

src/dynarmic/src/dynarmic/interface/A32/disassembler.h

@@ -1,18 +0,0 @@
-/* This file is part of the dynarmic project.
- * Copyright (c) 2016 MerryMage
- * SPDX-License-Identifier: 0BSD
- */
-
-#pragma once
-
-#include <cstdint>
-#include <string>
-
-namespace Dynarmic {
-namespace A32 {
-
-std::string DisassembleArm(std::uint32_t instruction);
-std::string DisassembleThumb16(std::uint16_t instruction);
-
-}  // namespace A32
-}  // namespace Dynarmic

src/dynarmic/tests/A32/fuzz_thumb.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -26,7 +26,6 @@
 #include "dynarmic/frontend/A32/FPSCR.h"
 #include "dynarmic/frontend/A32/PSR.h"
 #include "dynarmic/frontend/A32/a32_location_descriptor.h"
-#include "dynarmic/frontend/A32/disassembler/disassembler.h"
 #include "dynarmic/frontend/A32/translate/a32_translate.h"
 #include "dynarmic/interface/A32/a32.h"
 #include "dynarmic/ir/basic_block.h"

src/dynarmic/tests/A32/test_arm_disassembler.cpp

@@ -1,381 +0,0 @@
-/* This file is part of the dynarmic project.
- * Copyright (c) 2016 MerryMage
- * SPDX-License-Identifier: 0BSD
- */
-
-#include <catch2/catch_test_macros.hpp>
-
-#include "dynarmic/frontend/A32/disassembler/disassembler.h"
-
-using Dynarmic::A32::DisassembleArm;
-
-TEST_CASE("Disassemble branch instructions", "[arm][disassembler]") {
-    REQUIRE(DisassembleArm(0xEAFFFFFE) == "b +#0");
-    REQUIRE(DisassembleArm(0xEB000008) == "bl +#40");
-    REQUIRE(DisassembleArm(0xFBFFFFFE) == "blx +#2");
-    REQUIRE(DisassembleArm(0xFAFFFFFF) == "blx +#4");
-    REQUIRE(DisassembleArm(0xFBE1E7FE) == "blx -#7888894");
-    REQUIRE(DisassembleArm(0xE12FFF3D) == "blx sp");
-    REQUIRE(DisassembleArm(0x312FFF13) == "bxcc r3");
-    REQUIRE(DisassembleArm(0x012FFF29) == "bxjeq r9");
-}
-
-TEST_CASE("Disassemble data processing instructions", "[arm][disassembler]") {
-    REQUIRE(DisassembleArm(0xE2A21004) == "adc r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE0A21143) == "adc r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE0A21103) == "adc r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE0A21123) == "adc r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE0A21163) == "adc r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE0A21003) == "adc r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE0A21063) == "adc r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE0A21453) == "adc r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE0A21413) == "adc r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE0A21433) == "adc r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE0A21473) == "adc r1, r2, r3, ror r4");
-    REQUIRE(DisassembleArm(0xE2B21004) == "adcs r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE0B21143) == "adcs r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE0B21103) == "adcs r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE0B21123) == "adcs r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE0B21163) == "adcs r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE0B21003) == "adcs r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE0B21063) == "adcs r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE0B21453) == "adcs r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE0B21413) == "adcs r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE0B21433) == "adcs r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE0B21473) == "adcs r1, r2, r3, ror r4");
-
-    REQUIRE(DisassembleArm(0xE2853004) == "add r3, r5, #4");
-    REQUIRE(DisassembleArm(0xE0821143) == "add r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE0821103) == "add r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE0821123) == "add r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE0821163) == "add r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE0821003) == "add r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE0821453) == "add r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE0821413) == "add r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE0821433) == "add r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE0821473) == "add r1, r2, r3, ror r4");
-    REQUIRE(DisassembleArm(0xE0821063) == "add r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE2953004) == "adds r3, r5, #4");
-    REQUIRE(DisassembleArm(0xE0921143) == "adds r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE0921103) == "adds r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE0921123) == "adds r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE0921163) == "adds r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE0921003) == "adds r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE0921063) == "adds r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE0921453) == "adds r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE0921413) == "adds r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE0921433) == "adds r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE0921473) == "adds r1, r2, r3, ror r4");
-
-    REQUIRE(DisassembleArm(0xE2021004) == "and r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE0021143) == "and r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE0021103) == "and r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE0021123) == "and r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE0021163) == "and r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE0021003) == "and r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE0021453) == "and r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE0021413) == "and r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE0021433) == "and r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE0021473) == "and r1, r2, r3, ror r4");
-    REQUIRE(DisassembleArm(0xE0021063) == "and r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE2121004) == "ands r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE0121143) == "ands r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE0121103) == "ands r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE0121123) == "ands r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE0121163) == "ands r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE0121003) == "ands r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE0121063) == "ands r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE0121453) == "ands r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE0121413) == "ands r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE0121433) == "ands r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE0121473) == "ands r1, r2, r3, ror r4");
-
-    REQUIRE(DisassembleArm(0xE3C21004) == "bic r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE1C21143) == "bic r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE1C21103) == "bic r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE1C21123) == "bic r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE1C21163) == "bic r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE1C21003) == "bic r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE1C21453) == "bic r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE1C21413) == "bic r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE1C21433) == "bic r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE1C21473) == "bic r1, r2, r3, ror r4");
-    REQUIRE(DisassembleArm(0xE1C21063) == "bic r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE3D21004) == "bics r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE1D21143) == "bics r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE1D21103) == "bics r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE1D21123) == "bics r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE1D21163) == "bics r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE1D21003) == "bics r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE1D21063) == "bics r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE1D21453) == "bics r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE1D21413) == "bics r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE1D21433) == "bics r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE1D21473) == "bics r1, r2, r3, ror r4");
-
-    REQUIRE(DisassembleArm(0xE3710004) == "cmn r1, #4");
-    REQUIRE(DisassembleArm(0xE1710142) == "cmn r1, r2, asr #2");
-    REQUIRE(DisassembleArm(0xE1710102) == "cmn r1, r2, lsl #2");
-    REQUIRE(DisassembleArm(0xE1710122) == "cmn r1, r2, lsr #2");
-    REQUIRE(DisassembleArm(0xE1710162) == "cmn r1, r2, ror #2");
-    REQUIRE(DisassembleArm(0xE1710002) == "cmn r1, r2");
-    REQUIRE(DisassembleArm(0xE1710062) == "cmn r1, r2, rrx");
-    REQUIRE(DisassembleArm(0xE1710352) == "cmn r1, r2, asr r3");
-    REQUIRE(DisassembleArm(0xE1710312) == "cmn r1, r2, lsl r3");
-    REQUIRE(DisassembleArm(0xE1710332) == "cmn r1, r2, lsr r3");
-    REQUIRE(DisassembleArm(0xE1710372) == "cmn r1, r2, ror r3");
-
-    REQUIRE(DisassembleArm(0xE3510004) == "cmp r1, #4");
-    REQUIRE(DisassembleArm(0xE1510142) == "cmp r1, r2, asr #2");
-    REQUIRE(DisassembleArm(0xE1510102) == "cmp r1, r2, lsl #2");
-    REQUIRE(DisassembleArm(0xE1510122) == "cmp r1, r2, lsr #2");
-    REQUIRE(DisassembleArm(0xE1510162) == "cmp r1, r2, ror #2");
-    REQUIRE(DisassembleArm(0xE1510002) == "cmp r1, r2");
-    REQUIRE(DisassembleArm(0xE1510062) == "cmp r1, r2, rrx");
-    REQUIRE(DisassembleArm(0xE1510352) == "cmp r1, r2, asr r3");
-    REQUIRE(DisassembleArm(0xE1510312) == "cmp r1, r2, lsl r3");
-    REQUIRE(DisassembleArm(0xE1510332) == "cmp r1, r2, lsr r3");
-    REQUIRE(DisassembleArm(0xE1510372) == "cmp r1, r2, ror r3");
-
-    REQUIRE(DisassembleArm(0xE2221004) == "eor r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE0221243) == "eor r1, r2, r3, asr #4");
-    REQUIRE(DisassembleArm(0xE0221203) == "eor r1, r2, r3, lsl #4");
-    REQUIRE(DisassembleArm(0xE0221223) == "eor r1, r2, r3, lsr #4");
-    REQUIRE(DisassembleArm(0xE0221263) == "eor r1, r2, r3, ror #4");
-    REQUIRE(DisassembleArm(0xE0221003) == "eor r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE0221453) == "eor r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE0221413) == "eor r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE0221433) == "eor r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE0221473) == "eor r1, r2, r3, ror r4");
-    REQUIRE(DisassembleArm(0xE0221063) == "eor r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE2321004) == "eors r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE0321243) == "eors r1, r2, r3, asr #4");
-    REQUIRE(DisassembleArm(0xE0321203) == "eors r1, r2, r3, lsl #4");
-    REQUIRE(DisassembleArm(0xE0321223) == "eors r1, r2, r3, lsr #4");
-    REQUIRE(DisassembleArm(0xE0321263) == "eors r1, r2, r3, ror #4");
-    REQUIRE(DisassembleArm(0xE0321003) == "eors r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE0321453) == "eors r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE0321413) == "eors r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE0321433) == "eors r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE0321473) == "eors r1, r2, r3, ror r4");
-    REQUIRE(DisassembleArm(0xE0321063) == "eors r1, r2, r3, rrx");
-
-    REQUIRE(DisassembleArm(0xE3A010FF) == "mov r1, #255");
-    REQUIRE(DisassembleArm(0xE1A01142) == "mov r1, r2, asr #2");
-    REQUIRE(DisassembleArm(0xE1A01102) == "mov r1, r2, lsl #2");
-    REQUIRE(DisassembleArm(0xE1A01122) == "mov r1, r2, lsr #2");
-    REQUIRE(DisassembleArm(0xE1A01162) == "mov r1, r2, ror #2");
-    REQUIRE(DisassembleArm(0xE1A01062) == "mov r1, r2, rrx");
-    REQUIRE(DisassembleArm(0xE1A0E00F) == "mov lr, pc");
-    REQUIRE(DisassembleArm(0xE3B010FF) == "movs r1, #255");
-    REQUIRE(DisassembleArm(0xE1B0E00F) == "movs lr, pc");
-    REQUIRE(DisassembleArm(0xE1B01142) == "movs r1, r2, asr #2");
-    REQUIRE(DisassembleArm(0xE1B01102) == "movs r1, r2, lsl #2");
-    REQUIRE(DisassembleArm(0xE1B01122) == "movs r1, r2, lsr #2");
-    REQUIRE(DisassembleArm(0xE1B01162) == "movs r1, r2, ror #2");
-
-    REQUIRE(DisassembleArm(0xE3E01004) == "mvn r1, #4");
-    REQUIRE(DisassembleArm(0xE1E01142) == "mvn r1, r2, asr #2");
-    REQUIRE(DisassembleArm(0xE1E01102) == "mvn r1, r2, lsl #2");
-    REQUIRE(DisassembleArm(0xE1E01122) == "mvn r1, r2, lsr #2");
-    REQUIRE(DisassembleArm(0xE1E01162) == "mvn r1, r2, ror #2");
-    REQUIRE(DisassembleArm(0xE1E01062) == "mvn r1, r2, rrx");
-    REQUIRE(DisassembleArm(0xE1E01002) == "mvn r1, r2");
-    REQUIRE(DisassembleArm(0xE1E01352) == "mvn r1, r2, asr r3");
-    REQUIRE(DisassembleArm(0xE1E01312) == "mvn r1, r2, lsl r3");
-    REQUIRE(DisassembleArm(0xE1E01332) == "mvn r1, r2, lsr r3");
-    REQUIRE(DisassembleArm(0xE1E01372) == "mvn r1, r2, ror r3");
-    REQUIRE(DisassembleArm(0xE3F01004) == "mvns r1, #4");
-    REQUIRE(DisassembleArm(0xE1F01142) == "mvns r1, r2, asr #2");
-    REQUIRE(DisassembleArm(0xE1F01102) == "mvns r1, r2, lsl #2");
-    REQUIRE(DisassembleArm(0xE1F01122) == "mvns r1, r2, lsr #2");
-    REQUIRE(DisassembleArm(0xE1F01162) == "mvns r1, r2, ror #2");
-    REQUIRE(DisassembleArm(0xE1F01062) == "mvns r1, r2, rrx");
-    REQUIRE(DisassembleArm(0xE1F01002) == "mvns r1, r2");
-    REQUIRE(DisassembleArm(0xE1F01352) == "mvns r1, r2, asr r3");
-    REQUIRE(DisassembleArm(0xE1F01312) == "mvns r1, r2, lsl r3");
-    REQUIRE(DisassembleArm(0xE1F01332) == "mvns r1, r2, lsr r3");
-    REQUIRE(DisassembleArm(0xE1F01372) == "mvns r1, r2, ror r3");
-
-    REQUIRE(DisassembleArm(0xE3821004) == "orr r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE1821143) == "orr r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE1821103) == "orr r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE1821123) == "orr r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE1821163) == "orr r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE1821063) == "orr r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE1821003) == "orr r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE1821453) == "orr r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE1821413) == "orr r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE1821433) == "orr r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE1821473) == "orr r1, r2, r3, ror r4");
-    REQUIRE(DisassembleArm(0xE3921004) == "orrs r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE1921143) == "orrs r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE1921103) == "orrs r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE1921123) == "orrs r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE1921163) == "orrs r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE1921063) == "orrs r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE1921003) == "orrs r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE1921453) == "orrs r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE1921413) == "orrs r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE1921433) == "orrs r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE1921473) == "orrs r1, r2, r3, ror r4");
-
-    REQUIRE(DisassembleArm(0xE2621004) == "rsb r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE0621143) == "rsb r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE0621103) == "rsb r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE0621123) == "rsb r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE0621163) == "rsb r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE0621063) == "rsb r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE0621003) == "rsb r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE0621453) == "rsb r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE0621413) == "rsb r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE0621433) == "rsb r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE0621473) == "rsb r1, r2, r3, ror r4");
-    REQUIRE(DisassembleArm(0xE2721004) == "rsbs r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE0721143) == "rsbs r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE0721103) == "rsbs r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE0721123) == "rsbs r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE0721163) == "rsbs r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE0721063) == "rsbs r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE0721003) == "rsbs r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE0721453) == "rsbs r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE0721413) == "rsbs r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE0721433) == "rsbs r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE0721473) == "rsbs r1, r2, r3, ror r4");
-
-    REQUIRE(DisassembleArm(0xE2E21004) == "rsc r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE0E21143) == "rsc r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE0E21103) == "rsc r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE0E21123) == "rsc r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE0E21163) == "rsc r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE0E21063) == "rsc r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE0E21003) == "rsc r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE0E21453) == "rsc r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE0E21413) == "rsc r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE0E21433) == "rsc r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE0E21473) == "rsc r1, r2, r3, ror r4");
-    REQUIRE(DisassembleArm(0xE2F21004) == "rscs r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE0F21143) == "rscs r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE0F21103) == "rscs r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE0F21123) == "rscs r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE0F21163) == "rscs r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE0F21063) == "rscs r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE0F21003) == "rscs r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE0F21453) == "rscs r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE0F21413) == "rscs r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE0F21433) == "rscs r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE0F21473) == "rscs r1, r2, r3, ror r4");
-
-    REQUIRE(DisassembleArm(0xE2C21004) == "sbc r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE0C21143) == "sbc r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE0C21103) == "sbc r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE0C21123) == "sbc r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE0C21163) == "sbc r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE0C21063) == "sbc r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE0C21003) == "sbc r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE0C21453) == "sbc r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE0C21413) == "sbc r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE0C21433) == "sbc r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE0C21473) == "sbc r1, r2, r3, ror r4");
-    REQUIRE(DisassembleArm(0xE2D21004) == "sbcs r1, r2, #4");
-    REQUIRE(DisassembleArm(0xE0D21143) == "sbcs r1, r2, r3, asr #2");
-    REQUIRE(DisassembleArm(0xE0D21103) == "sbcs r1, r2, r3, lsl #2");
-    REQUIRE(DisassembleArm(0xE0D21123) == "sbcs r1, r2, r3, lsr #2");
-    REQUIRE(DisassembleArm(0xE0D21163) == "sbcs r1, r2, r3, ror #2");
-    REQUIRE(DisassembleArm(0xE0D21063) == "sbcs r1, r2, r3, rrx");
-    REQUIRE(DisassembleArm(0xE0D21003) == "sbcs r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE0D21453) == "sbcs r1, r2, r3, asr r4");
-    REQUIRE(DisassembleArm(0xE0D21413) == "sbcs r1, r2, r3, lsl r4");
-    REQUIRE(DisassembleArm(0xE0D21433) == "sbcs r1, r2, r3, lsr r4");
-    REQUIRE(DisassembleArm(0xE0D21473) == "sbcs r1, r2, r3, ror r4");
-
-    REQUIRE(DisassembleArm(0xE3310004) == "teq r1, #4");
-    REQUIRE(DisassembleArm(0xE1310142) == "teq r1, r2, asr #2");
-    REQUIRE(DisassembleArm(0xE1310102) == "teq r1, r2, lsl #2");
-    REQUIRE(DisassembleArm(0xE1310122) == "teq r1, r2, lsr #2");
-    REQUIRE(DisassembleArm(0xE1310162) == "teq r1, r2, ror #2");
-    REQUIRE(DisassembleArm(0xE1310002) == "teq r1, r2");
-    REQUIRE(DisassembleArm(0xE1310062) == "teq r1, r2, rrx");
-    REQUIRE(DisassembleArm(0xE1310352) == "teq r1, r2, asr r3");
-    REQUIRE(DisassembleArm(0xE1310312) == "teq r1, r2, lsl r3");
-    REQUIRE(DisassembleArm(0xE1310332) == "teq r1, r2, lsr r3");
-    REQUIRE(DisassembleArm(0xE1310372) == "teq r1, r2, ror r3");
-
-    REQUIRE(DisassembleArm(0xE3110004) == "tst r1, #4");
-    REQUIRE(DisassembleArm(0xE1110142) == "tst r1, r2, asr #2");
-    REQUIRE(DisassembleArm(0xE1110102) == "tst r1, r2, lsl #2");
-    REQUIRE(DisassembleArm(0xE1110122) == "tst r1, r2, lsr #2");
-    REQUIRE(DisassembleArm(0xE1110162) == "tst r1, r2, ror #2");
-    REQUIRE(DisassembleArm(0xE1110002) == "tst r1, r2");
-    REQUIRE(DisassembleArm(0xE1110062) == "tst r1, r2, rrx");
-    REQUIRE(DisassembleArm(0xE1110352) == "tst r1, r2, asr r3");
-    REQUIRE(DisassembleArm(0xE1110312) == "tst r1, r2, lsl r3");
-    REQUIRE(DisassembleArm(0xE1110332) == "tst r1, r2, lsr r3");
-    REQUIRE(DisassembleArm(0xE1110372) == "tst r1, r2, ror r3");
-}
-
-TEST_CASE("Disassemble half-word multiply and multiply accumulate instructions", "[arm][disassembler]") {
-    REQUIRE(DisassembleArm(0xE1003281) == "smlabb r0, r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE10032C1) == "smlabt r0, r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE10032A1) == "smlatb r0, r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE10032E1) == "smlatt r0, r1, r2, r3");
-
-    REQUIRE(DisassembleArm(0xE1203281) == "smlawb r0, r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE12032C1) == "smlawt r0, r1, r2, r3");
-
-    REQUIRE(DisassembleArm(0xE12002A1) == "smulwb r0, r1, r2");
-    REQUIRE(DisassembleArm(0xE12002E1) == "smulwt r0, r1, r2");
-
-    REQUIRE(DisassembleArm(0xE1410382) == "smlalbb r0, r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE14103C2) == "smlalbt r0, r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE14103A2) == "smlaltb r0, r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE14103E2) == "smlaltt r0, r1, r2, r3");
-
-    REQUIRE(DisassembleArm(0xE1600281) == "smulbb r0, r1, r2");
-    REQUIRE(DisassembleArm(0xE16002C1) == "smulbt r0, r1, r2");
-    REQUIRE(DisassembleArm(0xE16002A1) == "smultb r0, r1, r2");
-    REQUIRE(DisassembleArm(0xE16002E1) == "smultt r0, r1, r2");
-}
-
-TEST_CASE("Disassemble multiply and multiply accumulate instructions", "[arm][disassembler]") {
-    REQUIRE(DisassembleArm(0xE0214392) == "mla r1, r2, r3, r4");
-    REQUIRE(DisassembleArm(0xE0314392) == "mlas r1, r2, r3, r4");
-
-    REQUIRE(DisassembleArm(0xE0010392) == "mul r1, r2, r3");
-    REQUIRE(DisassembleArm(0xE0110392) == "muls r1, r2, r3");
-
-    // TODO: MLS should be here whenever it's supported.
-
-    REQUIRE(DisassembleArm(0xE0E21493) == "smlal r1, r2, r3, r4");
-    REQUIRE(DisassembleArm(0xE0F21493) == "smlals r1, r2, r3, r4");
-
-    REQUIRE(DisassembleArm(0xE0C21493) == "smull r1, r2, r3, r4");
-    REQUIRE(DisassembleArm(0xE0D21493) == "smulls r1, r2, r3, r4");
-
-    REQUIRE(DisassembleArm(0xE0421493) == "umaal r1, r2, r3, r4");
-
-    REQUIRE(DisassembleArm(0xE0A21493) == "umlal r1, r2, r3, r4");
-    REQUIRE(DisassembleArm(0xE0B21493) == "umlals r1, r2, r3, r4");
-
-    REQUIRE(DisassembleArm(0xE0821493) == "umull r1, r2, r3, r4");
-    REQUIRE(DisassembleArm(0xE0921493) == "umulls r1, r2, r3, r4");
-}
-
-TEST_CASE("Disassemble synchronization primitive instructions", "[arm][disassembler]") {
-    REQUIRE(DisassembleArm(0xE1921F9F) == "ldrex r1, [r2]");
-    REQUIRE(DisassembleArm(0xE1D21F9F) == "ldrexb r1, [r2]");
-    REQUIRE(DisassembleArm(0xE1B31F9F) == "ldrexd r1, r2, [r3]");
-    REQUIRE(DisassembleArm(0xE1F21F9F) == "ldrexh r1, [r2]");
-
-    REQUIRE(DisassembleArm(0xE1831F92) == "strex r1, r2, [r3]");
-    REQUIRE(DisassembleArm(0xE1C31F92) == "strexb r1, r2, [r3]");
-    REQUIRE(DisassembleArm(0xE1A41F92) == "strexd r1, r2, r3, [r4]");
-    REQUIRE(DisassembleArm(0xE1E31F92) == "strexh r1, r2, [r3]");
-
-    REQUIRE(DisassembleArm(0xE1031092) == "swp r1, r2, [r3]");
-    REQUIRE(DisassembleArm(0xE1431092) == "swpb r1, r2, [r3]");
-}
-
-TEST_CASE("Disassemble load / store multiple instructions", "[arm][disassembler]") {
-    REQUIRE(DisassembleArm(0xE92D500F) == "stmdb sp!, {r0, r1, r2, r3, r12, lr}");
-}

src/dynarmic/tests/CMakeLists.txt

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+# SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 # SPDX-License-Identifier: GPL-3.0-or-later
 include(TargetArchitectureSpecificSources)
 
@@ -9,7 +9,6 @@ add_executable(dynarmic_tests
     fp/unpacked_tests.cpp
     rand_int.h
     # A32
-    A32/test_arm_disassembler.cpp
     A32/test_arm_instructions.cpp
     A32/test_coprocessor.cpp
     A32/test_svc.cpp
@@ -25,7 +24,7 @@ add_executable(dynarmic_tests
     A64/real_world.cpp
     A64/testenv.h
 )
-target_link_libraries(dynarmic_tests PRIVATE merry::oaknut)
+target_link_libraries(dynarmic_tests PRIVATE merry::oaknut unordered_dense::unordered_dense)
 
 if (DYNARMIC_TESTS_USE_UNICORN)
     target_link_libraries(dynarmic_tests PRIVATE Unicorn::Unicorn)
@@ -73,7 +72,7 @@ add_executable(dynarmic_print_info
     print_info.cpp
 )
 create_target_directory_groups(dynarmic_print_info)
-target_link_libraries(dynarmic_print_info PRIVATE dynarmic fmt::fmt merry::mcl)
+target_link_libraries(dynarmic_print_info PRIVATE dynarmic fmt::fmt merry::mcl unordered_dense::unordered_dense)
 if (BOOST_NO_HEADERS)
     target_link_libraries(dynarmic_print_info PRIVATE Boost::variant Boost::icl Boost::pool)
 else()
@@ -94,7 +93,7 @@ add_executable(dynarmic_test_generator
 
 create_target_directory_groups(dynarmic_test_generator)
 
-target_link_libraries(dynarmic_test_generator PRIVATE dynarmic fmt::fmt merry::mcl)
+target_link_libraries(dynarmic_test_generator PRIVATE dynarmic fmt::fmt merry::mcl unordered_dense::unordered_dense)
 if (BOOST_NO_HEADERS)
     target_link_libraries(dynarmic_test_generator PRIVATE Boost::variant Boost::icl Boost::pool)
 else()
@@ -111,7 +110,7 @@ add_executable(dynarmic_test_reader
     test_reader.cpp
 )
 create_target_directory_groups(dynarmic_test_reader)
-target_link_libraries(dynarmic_test_reader PRIVATE dynarmic fmt::fmt merry::mcl)
+target_link_libraries(dynarmic_test_reader PRIVATE dynarmic fmt::fmt merry::mcl unordered_dense::unordered_dense)
 if (BOOST_NO_HEADERS)
     target_link_libraries(dynarmic_test_reader PRIVATE Boost::variant Boost::icl Boost::pool)
 else()
@@ -124,7 +123,7 @@ target_compile_definitions(dynarmic_test_reader PRIVATE FMT_USE_USER_DEFINED_LIT
 #
 create_target_directory_groups(dynarmic_tests)
 
-target_link_libraries(dynarmic_tests PRIVATE dynarmic Catch2::Catch2WithMain fmt::fmt merry::mcl)
+target_link_libraries(dynarmic_tests PRIVATE dynarmic Catch2::Catch2WithMain fmt::fmt merry::mcl unordered_dense::unordered_dense)
 if (BOOST_NO_HEADERS)
     target_link_libraries(dynarmic_tests PRIVATE Boost::variant Boost::icl Boost::pool)
 else()

src/dynarmic/tests/print_info.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -34,7 +34,6 @@
 #include "dynarmic/frontend/A64/translate/impl/impl.h"
 #include "dynarmic/interface/A32/a32.h"
 #include "dynarmic/interface/A32/config.h"
-#include "dynarmic/interface/A32/disassembler.h"
 #include "dynarmic/ir/basic_block.h"
 #include "dynarmic/ir/opt_passes.h"
 

src/video_core/buffer_cache/buffer_base.h

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
@@ -39,7 +42,8 @@ public:
     static constexpr u64 BASE_PAGE_SIZE = 1ULL << BASE_PAGE_BITS;
 
     explicit BufferBase(VAddr cpu_addr_, u64 size_bytes_)
-        : cpu_addr{cpu_addr_}, size_bytes{size_bytes_} {}
+        : cpu_addr_cached{static_cast<DAddr>(cpu_addr_)}, cpu_addr{cpu_addr_},
+          size_bytes{size_bytes_} {}
 
     explicit BufferBase(NullBufferParams) {}
 
@@ -97,6 +101,8 @@ public:
         return cpu_addr;
     }
 
+    DAddr cpu_addr_cached = 0;
+
     /// Returns the offset relative to the given CPU address
     /// @pre IsInBounds returns true
     [[nodiscard]] u32 Offset(VAddr other_cpu_addr) const noexcept {

src/video_core/buffer_cache/buffer_cache.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
@@ -382,6 +382,10 @@ void BufferCache<P>::BindHostComputeBuffers() {
     BindHostComputeUniformBuffers();
     BindHostComputeStorageBuffers();
     BindHostComputeTextureBuffers();
+    if (any_buffer_uploaded) {
+        runtime.PostCopyBarrier();
+        any_buffer_uploaded = false;
+    }
 }
 
 template <class P>
@@ -763,45 +767,85 @@ void BufferCache<P>::BindHostIndexBuffer() {
     }
 }
 
+template <class P>
+void BufferCache<P>::BindHostVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size,
+                                          u32 stride) {
+    if constexpr (IS_OPENGL) {
+        runtime.BindVertexBuffer(index, buffer, offset, size, stride);
+    } else {
+        runtime.BindVertexBuffer(index, buffer.Handle(), offset, size, stride);
+    }
+}
+
+template <class P>
+Binding& BufferCache<P>::VertexBufferSlot(u32 index) {
+    ASSERT(index < NUM_VERTEX_BUFFERS);
+    return v_buffer[index];
+}
+
+template <class P>
+const Binding& BufferCache<P>::VertexBufferSlot(u32 index) const {
+    ASSERT(index < NUM_VERTEX_BUFFERS);
+    return v_buffer[index];
+}
+
+template <class P>
+void BufferCache<P>::UpdateVertexBufferSlot(u32 index, const Binding& binding) {
+    Binding& slot = VertexBufferSlot(index);
+    if (slot.device_addr != binding.device_addr || slot.size != binding.size) {
+        ++vertex_buffers_serial;
+    }
+    slot = binding;
+    if (binding.buffer_id != NULL_BUFFER_ID && binding.size != 0) {
+        enabled_vertex_buffers_mask |= (1u << index);
+    } else {
+        enabled_vertex_buffers_mask &= ~(1u << index);
+    }
+}
+
 template <class P>
 void BufferCache<P>::BindHostVertexBuffers() {
-    HostBindings<typename P::Buffer> host_bindings;
-    bool any_valid{false};
     auto& flags = maxwell3d->dirty.flags;
-    for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) {
-        const Binding& binding = channel_state->vertex_buffers[index];
+    u32 enabled_mask = enabled_vertex_buffers_mask;
+    HostBindings<Buffer> bindings{};
+    u32 last_index = std::numeric_limits<u32>::max();
+    const auto flush_bindings = [&]() {
+        if (bindings.buffers.empty()) {
+            return;
+        }
+        bindings.max_index = bindings.min_index + static_cast<u32>(bindings.buffers.size());
+        runtime.BindVertexBuffers(bindings);
+        bindings = HostBindings<Buffer>{};
+        last_index = std::numeric_limits<u32>::max();
+    };
+    while (enabled_mask != 0) {
+        const u32 index = std::countr_zero(enabled_mask);
+        enabled_mask &= (enabled_mask - 1);
+        const Binding& binding = VertexBufferSlot(index);
         Buffer& buffer = slot_buffers[binding.buffer_id];
         TouchBuffer(buffer, binding.buffer_id);
         SynchronizeBuffer(buffer, binding.device_addr, binding.size);
         if (!flags[Dirty::VertexBuffer0 + index]) {
+            flush_bindings();
             continue;
         }
         flags[Dirty::VertexBuffer0 + index] = false;
-
-        host_bindings.min_index = (std::min)(host_bindings.min_index, index);
-        host_bindings.max_index = (std::max)(host_bindings.max_index, index);
-        any_valid = true;
-    }
-
-    if (any_valid) {
-        host_bindings.max_index++;
-        for (u32 index = host_bindings.min_index; index < host_bindings.max_index; index++) {
-            flags[Dirty::VertexBuffer0 + index] = false;
-
-            const Binding& binding = channel_state->vertex_buffers[index];
-            Buffer& buffer = slot_buffers[binding.buffer_id];
-
-            const u32 stride = maxwell3d->regs.vertex_streams[index].stride;
-            const u32 offset = buffer.Offset(binding.device_addr);
-            buffer.MarkUsage(offset, binding.size);
-
-            host_bindings.buffers.push_back(&buffer);
-            host_bindings.offsets.push_back(offset);
-            host_bindings.sizes.push_back(binding.size);
-            host_bindings.strides.push_back(stride);
+        const u32 stride = maxwell3d->regs.vertex_streams[index].stride;
+        const u32 offset = buffer.Offset(binding.device_addr);
+        buffer.MarkUsage(offset, binding.size);
+        if (!bindings.buffers.empty() && index != last_index + 1) {
+            flush_bindings();
         }
-        runtime.BindVertexBuffers(host_bindings);
+        if (bindings.buffers.empty()) {
+            bindings.min_index = index;
+        }
+        bindings.buffers.push_back(&buffer);
+        bindings.offsets.push_back(offset);
+        bindings.sizes.push_back(binding.size);
+        bindings.strides.push_back(stride);
+        last_index = index;
     }
+    flush_bindings();
 }
 
 template <class P>
@@ -1205,17 +1249,20 @@ void BufferCache<P>::UpdateVertexBuffer(u32 index) {
     u32 size = address_size; // TODO: Analyze stride and number of vertices
     if (array.enable == 0 || size == 0 || !device_addr) {
         channel_state->vertex_buffers[index] = NULL_BINDING;
+        UpdateVertexBufferSlot(index, NULL_BINDING);
         return;
     }
     if (!gpu_memory->IsWithinGPUAddressRange(gpu_addr_end) || size >= 64_MiB) {
         size = static_cast<u32>(gpu_memory->MaxContinuousRange(gpu_addr_begin, size));
     }
     const BufferId buffer_id = FindBuffer(*device_addr, size);
-    channel_state->vertex_buffers[index] = Binding{
+    const Binding binding{
         .device_addr = *device_addr,
         .size = size,
         .buffer_id = buffer_id,
     };
+    channel_state->vertex_buffers[index] = binding;
+    UpdateVertexBufferSlot(index, binding);
 }
 
 template <class P>
@@ -1528,12 +1575,12 @@ void BufferCache<P>::TouchBuffer(Buffer& buffer, BufferId buffer_id) noexcept {
 
 template <class P>
 bool BufferCache<P>::SynchronizeBuffer(Buffer& buffer, DAddr device_addr, u32 size) {
-    boost::container::small_vector<BufferCopy, 4> copies;
+    upload_copies.clear();
     u64 total_size_bytes = 0;
     u64 largest_copy = 0;
-    DAddr buffer_start = buffer.CpuAddr();
+    const DAddr buffer_start = buffer.cpu_addr_cached;
     memory_tracker.ForEachUploadRange(device_addr, size, [&](u64 device_addr_out, u64 range_size) {
-        copies.push_back(BufferCopy{
+        upload_copies.push_back(BufferCopy{
             .src_offset = total_size_bytes,
             .dst_offset = device_addr_out - buffer_start,
             .size = range_size,
@@ -1544,8 +1591,9 @@ bool BufferCache<P>::SynchronizeBuffer(Buffer& buffer, DAddr device_addr, u32 si
     if (total_size_bytes == 0) {
         return true;
     }
-    const std::span<BufferCopy> copies_span(copies.data(), copies.size());
+    const std::span<BufferCopy> copies_span(upload_copies.data(), upload_copies.size());
     UploadMemory(buffer, total_size_bytes, largest_copy, copies_span);
+    any_buffer_uploaded = true;
     return false;
 }
 
@@ -1735,6 +1783,7 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id, bool do_not_mark) {
         auto& binding = channel_state->vertex_buffers[index];
         if (binding.buffer_id == buffer_id) {
             binding.buffer_id = BufferId{};
+            UpdateVertexBufferSlot(index, binding);
             dirty_vertex_buffers.push_back(index);
         }
     }

src/video_core/buffer_cache/buffer_cache_base.h

@@ -320,6 +320,7 @@ public:
 
     std::recursive_mutex mutex;
     Runtime& runtime;
+    bool any_buffer_uploaded = false;
 
 private:
     template <typename Func>
@@ -372,6 +373,8 @@ private:
 
     void BindHostTransformFeedbackBuffers();
 
+    void BindHostVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size, u32 stride);
+
     void BindHostComputeUniformBuffers();
 
     void BindHostComputeStorageBuffers();
@@ -453,6 +456,12 @@ private:
 
     [[nodiscard]] bool HasFastUniformBufferBound(size_t stage, u32 binding_index) const noexcept;
 
+    [[nodiscard]] Binding& VertexBufferSlot(u32 index);
+
+    [[nodiscard]] const Binding& VertexBufferSlot(u32 index) const;
+
+    void UpdateVertexBufferSlot(u32 index, const Binding& binding);
+
     void ClearDownload(DAddr base_addr, u64 size);
 
     void InlineMemoryImplementation(DAddr dest_address, size_t copy_size,
@@ -472,6 +481,12 @@ private:
 
     u32 last_index_count = 0;
 
+    u32 enabled_vertex_buffers_mask = 0;
+    u64 vertex_buffers_serial = 0;
+    std::array<Binding, 32> v_buffer{};
+
+    boost::container::small_vector<BufferCopy, 4> upload_copies;
+
     MemoryTracker memory_tracker;
     Common::RangeSet<DAddr> uncommitted_gpu_modified_ranges;
     Common::RangeSet<DAddr> gpu_modified_ranges;

src/video_core/gpu_thread.h

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -89,7 +92,7 @@ struct CommandDataContainer {
 
 /// Struct used to synchronize the GPU thread
 struct SynchState final {
-    using CommandQueue = Common::MPSCQueue<CommandDataContainer>;
+    using CommandQueue = Common::SPSCQueue<CommandDataContainer>;
     std::mutex write_lock;
     CommandQueue queue;
     u64 last_fence{};

src/video_core/renderer_opengl/gl_compute_pipeline.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -189,6 +189,10 @@ void ComputePipeline::Configure() {
     buffer_cache.runtime.SetEnableStorageBuffers(use_storage_buffers);
     buffer_cache.runtime.SetImagePointers(textures.data(), images.data());
     buffer_cache.BindHostComputeBuffers();
+    if (buffer_cache.any_buffer_uploaded) {
+        buffer_cache.runtime.PostCopyBarrier();
+        buffer_cache.any_buffer_uploaded = false;
+    }
 
     const VideoCommon::ImageViewInOut* views_it{views.data() + num_texture_buffers +
                                                 num_image_buffers};

src/video_core/renderer_opengl/gl_graphics_pipeline.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -558,6 +558,10 @@ bool GraphicsPipeline::ConfigureImpl(bool is_indexed) {
     if (image_binding != 0) {
         glBindImageTextures(0, image_binding, images.data());
     }
+    if (buffer_cache.any_buffer_uploaded) {
+        buffer_cache.runtime.PostCopyBarrier();
+        buffer_cache.any_buffer_uploaded = false;
+    }
 
     return true;
 }

src/video_core/renderer_vulkan/blit_image.cpp

@@ -436,23 +436,7 @@ void TransitionImageLayout(vk::CommandBuffer& cmdbuf, VkImage image, VkImageLayo
             .layerCount = 1,
         },
     };
-    // Blit image barrier - transition can happen at any stage
-    // Using specific stages for better parallelism, but keeping broad since this is a general utility
-    const VkPipelineStageFlags src_stages_blit =
-        VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-        VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-        VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-        VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-        VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-        VK_PIPELINE_STAGE_TRANSFER_BIT;
-    const VkPipelineStageFlags dst_stages_blit =
-        VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-        VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-        VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-        VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-        VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-        VK_PIPELINE_STAGE_TRANSFER_BIT;
-    cmdbuf.PipelineBarrier(src_stages_blit, dst_stages_blit,
+    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                            0, barrier);
 }
 

src/video_core/renderer_vulkan/present/util.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2024 yuzu Emulator Project
@@ -68,16 +68,7 @@ void TransitionImageLayout(vk::CommandBuffer& cmdbuf, VkImage image, VkImageLayo
             .layerCount = 1,
         },
     };
-    // Present path - use specific stages for better parallelism on tile-based GPUs
-    const VkPipelineStageFlags src_stages =
-        VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-        VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-        VK_PIPELINE_STAGE_TRANSFER_BIT;
-    const VkPipelineStageFlags dst_stages =
-        VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-        VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-        VK_PIPELINE_STAGE_TRANSFER_BIT;
-    cmdbuf.PipelineBarrier(src_stages, dst_stages,
+    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                            0, barrier);
 }
 
@@ -160,12 +151,11 @@ void DownloadColorImage(vk::CommandBuffer& cmdbuf, VkImage image, VkBuffer buffe
             .layerCount = VK_REMAINING_ARRAY_LAYERS,
         },
     };
-    // Host will read the downloaded data
     static constexpr VkMemoryBarrier memory_write_barrier{
         .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
         .pNext = nullptr,
-        .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-        .dstAccessMask = VK_ACCESS_HOST_READ_BIT,
+        .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+        .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
     };
     const VkBufferImageCopy copy{
         .bufferOffset = 0,
@@ -180,20 +170,10 @@ void DownloadColorImage(vk::CommandBuffer& cmdbuf, VkImage image, VkBuffer buffe
         .imageOffset{.x = 0, .y = 0, .z = 0},
         .imageExtent{extent},
     };
-    const VkPipelineStageFlags src_stages_copy =
-        VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-        VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-        VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-        VK_PIPELINE_STAGE_TRANSFER_BIT;
-    cmdbuf.PipelineBarrier(src_stages_copy, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
+    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
                            read_barrier);
     cmdbuf.CopyImageToBuffer(image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, buffer, copy);
-    const VkPipelineStageFlags dst_stages_copy =
-        VK_PIPELINE_STAGE_HOST_BIT |
-        VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-        VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-        VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
-    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stages_copy, 0,
+    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0,
                            memory_write_barrier, nullptr, image_write_barrier);
 }
 

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -436,27 +436,13 @@ void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer,
     scheduler.RequestOutsideRenderPassOperationContext();
     scheduler.Record([src_buffer, dst_buffer, vk_copies, barrier](vk::CommandBuffer cmdbuf) {
         if (barrier) {
-            // Buffer writes can come from vertex input, shaders, or compute
-            const VkPipelineStageFlags src_stages =
-                VK_PIPELINE_STAGE_VERTEX_INPUT_BIT |
-                VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-                VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-                VK_PIPELINE_STAGE_TRANSFER_BIT;
-            cmdbuf.PipelineBarrier(src_stages,
+            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                                    VK_PIPELINE_STAGE_TRANSFER_BIT, 0, READ_BARRIER);
         }
         cmdbuf.CopyBuffer(src_buffer, dst_buffer, VideoCommon::FixSmallVectorADL(vk_copies));
         if (barrier) {
-            // Buffer reads can go to vertex input, shaders, or compute
-            const VkPipelineStageFlags dst_stages =
-                VK_PIPELINE_STAGE_VERTEX_INPUT_BIT |
-                VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-                VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-                VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
             cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
-                                   dst_stages, 0, WRITE_BARRIER);
+                                   VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, WRITE_BARRIER);
         }
     });
 }
@@ -470,36 +456,21 @@ void BufferCacheRuntime::PreCopyBarrier() {
     };
     scheduler.RequestOutsideRenderPassOperationContext();
     scheduler.Record([](vk::CommandBuffer cmdbuf) {
-        const VkPipelineStageFlags src_stages =
-            VK_PIPELINE_STAGE_VERTEX_INPUT_BIT |
-            VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_TRANSFER_BIT;
-        cmdbuf.PipelineBarrier(src_stages, VK_PIPELINE_STAGE_TRANSFER_BIT,
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
                                0, READ_BARRIER);
     });
 }
 
 void BufferCacheRuntime::PostCopyBarrier() {
-    // Specific access flags for buffer destinations: vertex input, uniforms, storage, index
     static constexpr VkMemoryBarrier WRITE_BARRIER{
         .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
         .pNext = nullptr,
         .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-        .dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_INDEX_READ_BIT |
-                         VK_ACCESS_UNIFORM_READ_BIT | VK_ACCESS_SHADER_READ_BIT |
-                         VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT,
+        .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
     };
     scheduler.RequestOutsideRenderPassOperationContext();
     scheduler.Record([](vk::CommandBuffer cmdbuf) {
-        const VkPipelineStageFlags dst_stages =
-            VK_PIPELINE_STAGE_VERTEX_INPUT_BIT |
-            VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stages,
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                                0, WRITE_BARRIER);
     });
 }
@@ -523,24 +494,10 @@ void BufferCacheRuntime::ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t si
 
     scheduler.RequestOutsideRenderPassOperationContext();
     scheduler.Record([dest_buffer, offset, size, value](vk::CommandBuffer cmdbuf) {
-        // Buffer writes can come from vertex input, shaders, or compute
-        const VkPipelineStageFlags src_stages_clear =
-            VK_PIPELINE_STAGE_VERTEX_INPUT_BIT |
-            VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_TRANSFER_BIT;
-        cmdbuf.PipelineBarrier(src_stages_clear, VK_PIPELINE_STAGE_TRANSFER_BIT,
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
                                0, READ_BARRIER);
         cmdbuf.FillBuffer(dest_buffer, offset, size, value);
-        // Buffer reads can go to vertex input, shaders, or compute
-        const VkPipelineStageFlags dst_stages_clear =
-            VK_PIPELINE_STAGE_VERTEX_INPUT_BIT |
-            VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stages_clear,
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                                0, WRITE_BARRIER);
     });
 }

src/video_core/renderer_vulkan/vk_compute_pass.cpp

@@ -446,26 +446,13 @@ void ConditionalRenderingResolvePass::Resolve(VkBuffer dst_buffer, VkBuffer src_
         const VkDescriptorSet set = descriptor_allocator.Commit();
         device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
 
-        // Memory writes can come from graphics or compute stages
-        const VkPipelineStageFlags src_stages_compute =
-            VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-            VK_PIPELINE_STAGE_TRANSFER_BIT;
-        cmdbuf.PipelineBarrier(src_stages_compute,
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                                VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, read_barrier);
         cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline);
         cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
         cmdbuf.Dispatch(1, 1, 1);
-        // Conditional rendering result used by draw indirect or conditional rendering
-        const VkPipelineStageFlags dst_stages_cond =
-            VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT |
-            VK_PIPELINE_STAGE_CONDITIONAL_RENDERING_BIT_EXT |
-            VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT;
         cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
-                               dst_stages_cond, 0, write_barrier);
+                               VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, write_barrier);
     });
 }
 
@@ -533,29 +520,14 @@ void QueriesPrefixScanPass::Run(VkBuffer accumulation_buffer, VkBuffer dst_buffe
             const VkDescriptorSet set = descriptor_allocator.Commit();
             device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
 
-            // Memory writes can come from graphics or compute stages
-            const VkPipelineStageFlags src_stages_scan =
-                VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-                VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-                VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                VK_PIPELINE_STAGE_TRANSFER_BIT;
-            cmdbuf.PipelineBarrier(src_stages_scan,
+            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                                    VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, read_barrier);
             cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline);
             cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
             cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, uniforms);
             cmdbuf.Dispatch(1, 1, 1);
-            // Query prefix scan results used by draw indirect, conditional rendering, or shaders
-            const VkPipelineStageFlags dst_stages_scan =
-                VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT |
-                VK_PIPELINE_STAGE_CONDITIONAL_RENDERING_BIT_EXT |
-                VK_PIPELINE_STAGE_VERTEX_INPUT_BIT |
-                VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-                VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
             cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
-                                   dst_stages_scan, 0, write_barrier);
+                                   VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, write_barrier);
         });
     }
 }
@@ -607,15 +579,8 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
                 .layerCount = VK_REMAINING_ARRAY_LAYERS,
             },
         };
-        // If initialized, memory writes can come from graphics or compute stages
-        const VkPipelineStageFlags src_stages_astc = is_initialized ?
-            (VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-             VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-             VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-             VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-             VK_PIPELINE_STAGE_TRANSFER_BIT) :
-            VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT;
-        cmdbuf.PipelineBarrier(src_stages_astc,
+        cmdbuf.PipelineBarrier(is_initialized ? VK_PIPELINE_STAGE_ALL_COMMANDS_BIT
+                                              : VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
                                VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier);
         cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline);
     });

src/video_core/renderer_vulkan/vk_compute_pipeline.cpp

@@ -203,6 +203,10 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
 
     buffer_cache.UpdateComputeBuffers();
     buffer_cache.BindHostComputeBuffers();
+    if (buffer_cache.any_buffer_uploaded) {
+        buffer_cache.runtime.PostCopyBarrier();
+        buffer_cache.any_buffer_uploaded = false;
+    }
 
     RescalingPushConstant rescaling;
     const VideoCommon::SamplerId* samplers_it{samplers.data()};

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -496,6 +496,10 @@ bool GraphicsPipeline::ConfigureImpl(bool is_indexed) {
     if constexpr (Spec::enabled_stages[4]) {
         prepare_stage(4);
     }
+    if (buffer_cache.any_buffer_uploaded) {
+        buffer_cache.runtime.PostCopyBarrier();
+        buffer_cache.any_buffer_uploaded = false;
+    }
     texture_cache.UpdateRenderTargets(false);
     texture_cache.CheckFeedbackLoop(views);
     ConfigureDraw(rescaling, render_area);

src/video_core/renderer_vulkan/vk_present_manager.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
@@ -441,13 +441,7 @@ void PresentManager::CopyToSwapchainImpl(Frame* frame) {
         },
     };
 
-    // Previous operations that might have written to the frame image
-    const VkPipelineStageFlags src_stages_present =
-        VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-        VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-        VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-        VK_PIPELINE_STAGE_TRANSFER_BIT;
-    cmdbuf.PipelineBarrier(src_stages_present, VK_PIPELINE_STAGE_TRANSFER_BIT, {},
+    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, {},
                            {}, {}, pre_barriers);
 
     if (blit_supported) {

src/video_core/renderer_vulkan/vk_query_cache.cpp

@@ -809,19 +809,12 @@ public:
             .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
             .pNext = nullptr,
             .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-            .dstAccessMask = VK_ACCESS_HOST_READ_BIT | VK_ACCESS_SHADER_READ_BIT,
+            .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
         };
         scheduler.RequestOutsideRenderPassOperationContext();
         scheduler.Record([](vk::CommandBuffer cmdbuf) {
-            // After transfer, results may be read by host or used in subsequent operations
-            const VkPipelineStageFlags dst_stages_query =
-                VK_PIPELINE_STAGE_HOST_BIT |
-                VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-                VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-                VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
             cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
-                                   dst_stages_query, 0, WRITE_BARRIER);
+                                   VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, WRITE_BARRIER);
         });
 
         std::scoped_lock lk(flush_guard);
@@ -949,7 +942,7 @@ private:
             .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
             .pNext = nullptr,
             .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-            .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
+            .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT,
         };
         scheduler.RequestOutsideRenderPassOperationContext();
         scheduler.Record([dst_buffer = current_bank->GetBuffer(),
@@ -1482,40 +1475,22 @@ void QueryCacheRuntime::Barriers(bool is_prebarrier) {
         .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
         .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
     };
-    // Query results may be read by host or used in shaders/indirect commands
     static constexpr VkMemoryBarrier WRITE_BARRIER{
         .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
         .pNext = nullptr,
         .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-        .dstAccessMask = VK_ACCESS_HOST_READ_BIT | VK_ACCESS_SHADER_READ_BIT |
-                         VK_ACCESS_INDIRECT_COMMAND_READ_BIT,
+        .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
     };
     impl->scheduler.RequestOutsideRenderPassOperationContext();
     if (is_prebarrier) {
         impl->scheduler.Record([](vk::CommandBuffer cmdbuf) {
-            // Query data can be written by graphics or compute stages
-            const VkPipelineStageFlags src_stages_query =
-                VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-                VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-                VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-                VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-                VK_PIPELINE_STAGE_TRANSFER_BIT;
-            cmdbuf.PipelineBarrier(src_stages_query,
+            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                                    VK_PIPELINE_STAGE_TRANSFER_BIT, 0, READ_BARRIER);
         });
     } else {
         impl->scheduler.Record([](vk::CommandBuffer cmdbuf) {
-            // Query results may be read by host or used in shaders/indirect commands
-            const VkPipelineStageFlags dst_stages_query =
-                VK_PIPELINE_STAGE_HOST_BIT |
-                VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT |
-                VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-                VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
             cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
-                                   dst_stages_query, 0, WRITE_BARRIER);
+                                   VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, WRITE_BARRIER);
         });
     }
 }

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -586,23 +586,14 @@ void RasterizerVulkan::DispatchCompute() {
     }
     const std::array<u32, 3> dim{qmd.grid_dim_x, qmd.grid_dim_y, qmd.grid_dim_z};
     scheduler.RequestOutsideRenderPassOperationContext();
-    // Compute shader reads from storage buffers, uniforms, and images
     static constexpr VkMemoryBarrier READ_BARRIER{
         .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
         .pNext = nullptr,
-        .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
-        .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_UNIFORM_READ_BIT,
+        .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+        .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT,
     };
-    scheduler.Record([](vk::CommandBuffer cmdbuf) {
-        // Memory writes can come from graphics or compute stages
-        const VkPipelineStageFlags src_stages =
-            VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
-        cmdbuf.PipelineBarrier(src_stages, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
-                               0, READ_BARRIER);
-    });
+    scheduler.Record([](vk::CommandBuffer cmdbuf) { cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+                               0, READ_BARRIER); });
     scheduler.Record([dim](vk::CommandBuffer cmdbuf) { cmdbuf.Dispatch(dim[0], dim[1], dim[2]); });
 
     // Log compute dispatch

src/video_core/renderer_vulkan/vk_render_pass_cache.cpp

@@ -44,19 +44,13 @@ using VideoCore::Surface::SurfaceType;
         }
 
         VkAttachmentDescription AttachmentDescription(const Device& device, PixelFormat format,
-                                                      VkSampleCountFlagBits samples,
-                                                      bool is_depth_stencil) {
+                                                      VkSampleCountFlagBits samples) {
             using MaxwellToVK::SurfaceFormat;
 
             const SurfaceType surface_type = GetSurfaceType(format);
             const bool has_stencil = surface_type == SurfaceType::DepthStencil ||
                                      surface_type == SurfaceType::Stencil;
 
-            // Use optimal layouts for attachments - this allows drivers to optimize tiling and access patterns
-            const VkImageLayout attachment_layout = is_depth_stencil
-                ? VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL
-                : VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
-
             return {
                 .flags = {},
                 .format = SurfaceFormat(device, FormatType::Optimal, true, format).format,
@@ -67,8 +61,8 @@ using VideoCore::Surface::SurfaceType;
                                                  : VK_ATTACHMENT_LOAD_OP_DONT_CARE,
                 .stencilStoreOp = has_stencil ? VK_ATTACHMENT_STORE_OP_STORE
                                                   : VK_ATTACHMENT_STORE_OP_DONT_CARE,
-                .initialLayout = attachment_layout,
-                .finalLayout = attachment_layout,
+                .initialLayout = VK_IMAGE_LAYOUT_GENERAL,
+                .finalLayout = VK_IMAGE_LAYOUT_GENERAL,
             };
         }
     } // Anonymous namespace
@@ -90,10 +84,10 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) {
         const bool is_valid{format != PixelFormat::Invalid};
         references[index] = VkAttachmentReference{
             .attachment = is_valid ? num_colors : VK_ATTACHMENT_UNUSED,
-            .layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
+            .layout = VK_IMAGE_LAYOUT_GENERAL,
         };
         if (is_valid) {
-            descriptions.push_back(AttachmentDescription(*device, format, key.samples, false));
+            descriptions.push_back(AttachmentDescription(*device, format, key.samples));
             num_attachments = static_cast<u32>(index + 1);
             ++num_colors;
         }
@@ -103,9 +97,9 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) {
     if (key.depth_format != PixelFormat::Invalid) {
         depth_reference = VkAttachmentReference{
             .attachment = num_colors,
-            .layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL,
+            .layout = VK_IMAGE_LAYOUT_GENERAL,
         };
-        descriptions.push_back(AttachmentDescription(*device, key.depth_format, key.samples, true));
+        descriptions.push_back(AttachmentDescription(*device, key.depth_format, key.samples));
     }
     const VkSubpassDescription subpass{
         .flags = 0,

src/video_core/renderer_vulkan/vk_scheduler.cpp

@@ -264,26 +264,14 @@ u64 Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_se
     const u64 signal_value = master_semaphore->NextTick();
     RecordWithUploadBuffer([signal_semaphore, wait_semaphore, signal_value,
                             this](vk::CommandBuffer cmdbuf, vk::CommandBuffer upload_cmdbuf) {
-        // Specific access flags for upload destinations: vertex input, uniforms, storage, index, textures
         static constexpr VkMemoryBarrier WRITE_BARRIER{
             .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
             .pNext = nullptr,
             .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-            .dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_INDEX_READ_BIT |
-                             VK_ACCESS_UNIFORM_READ_BIT | VK_ACCESS_SHADER_READ_BIT |
-                             VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
-                             VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT,
+            .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
         };
-        // Specify exact stages that need the transfer results instead of ALL_COMMANDS_BIT
         upload_cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
-                                      VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-                                      VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                                      VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-                                      VK_PIPELINE_STAGE_VERTEX_INPUT_BIT |
-                                      VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                                      VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-                                      VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,
-                                      0, WRITE_BARRIER);
+                                      VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, WRITE_BARRIER);
         upload_cmdbuf.End();
         cmdbuf.End();
 
@@ -344,11 +332,11 @@ void Scheduler::EndRenderPass()
 
         query_cache->CounterEnable(VideoCommon::QueryType::ZPassPixelCount64, false);
         query_cache->NotifySegment(false);
+
         Record([num_images = num_renderpass_images,
                        images = renderpass_images,
                        ranges = renderpass_image_ranges](vk::CommandBuffer cmdbuf) {
             std::array<VkImageMemoryBarrier, 9> barriers;
-            VkPipelineStageFlags src_stages = 0;
             for (size_t i = 0; i < num_images; ++i) {
                 const VkImageSubresourceRange& range = ranges[i];
                 const bool is_color = (range.aspectMask & VK_IMAGE_ASPECT_COLOR_BIT) != 0;
@@ -356,76 +344,38 @@ void Scheduler::EndRenderPass()
                                               & (VK_IMAGE_ASPECT_DEPTH_BIT
                                                  | VK_IMAGE_ASPECT_STENCIL_BIT)) !=0;
 
-                // Determine optimal destination layout based on image usage
-                // After render pass, images may be used as shader resources or as attachments again
-                // Use optimal layouts to allow driver optimizations
                 VkAccessFlags src_access = 0;
-                VkAccessFlags dst_access = 0;
-                VkPipelineStageFlags this_stage = 0;
-                VkImageLayout new_layout;
 
-                if (is_color) {
-                    // Color attachments can be read as textures or used as attachments again
-                    src_access = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
-                    this_stage = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
-                    new_layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
-                    dst_access = VK_ACCESS_SHADER_READ_BIT
-                                 | VK_ACCESS_COLOR_ATTACHMENT_READ_BIT
-                                 | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
-                } else if (is_depth_stencil) {
-                    // Depth attachments can be read as textures or used as attachments again
-                    src_access = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
-                    this_stage = VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT
-                                 | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
-                    new_layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
-                    dst_access = VK_ACCESS_SHADER_READ_BIT
-                                 | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT
-                                 | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
-                } else {
-                    // Fallback to GENERAL for unknown usage
-                    src_access = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT
-                                 | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
-                    this_stage = VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT
-                                 | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT
-                                 | VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
-                    new_layout = VK_IMAGE_LAYOUT_GENERAL;
-                    dst_access = VK_ACCESS_SHADER_READ_BIT
-                                 | VK_ACCESS_SHADER_WRITE_BIT
-                                 | VK_ACCESS_COLOR_ATTACHMENT_READ_BIT
-                                 | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT
-                                 | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT
-                                 | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
-                }
-                src_stages |= this_stage;
+                if (is_color)
+                    src_access |= VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
+                else if (is_depth_stencil)
+                    src_access |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
+                else
+                    src_access |= VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT
+                                  | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
+
                 barriers[i] = VkImageMemoryBarrier{
                         .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                         .pNext = nullptr,
                         .srcAccessMask = src_access,
-                        .dstAccessMask = dst_access,
-                        .oldLayout = is_color ? VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL
-                                             : (is_depth_stencil ? VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL
-                                                                 : VK_IMAGE_LAYOUT_GENERAL),
-                        .newLayout = new_layout,
+                        .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT
+                                         | VK_ACCESS_COLOR_ATTACHMENT_READ_BIT
+                                         | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT
+                                         | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT
+                                         | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
+                        .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
+                        .newLayout = VK_IMAGE_LAYOUT_GENERAL,
                         .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                         .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                         .image = images[i],
                         .subresourceRange = range,
                 };
             }
-
             cmdbuf.EndRenderPass();
-
-            // Use specific pipeline stages instead of ALL_COMMANDS_BIT for better parallelism
-            // The destination stages depend on how the images will be used next
-            const VkPipelineStageFlags dst_stages =
-                VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-                VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-                VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
-
-            cmdbuf.PipelineBarrier(src_stages,
-                                   dst_stages,
+            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
+                                   VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
+                                   VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
+                                   VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                                    0,
                                    nullptr,
                                    nullptr,

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -810,8 +810,7 @@ void BlitScale(Scheduler& scheduler, VkImage src_image, VkImage dst_image, const
                 .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                 .pNext = nullptr,
                 .srcAccessMask = 0,
-                .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT |
-                                 VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
+                .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT | VK_ACCESS_MEMORY_READ_BIT,
                 .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
                 .newLayout = VK_IMAGE_LAYOUT_GENERAL,
                 .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
@@ -823,8 +822,7 @@ void BlitScale(Scheduler& scheduler, VkImage src_image, VkImage dst_image, const
                 .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                 .pNext = nullptr,
                 .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-                .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT |
-                                 VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
+                .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT | VK_ACCESS_MEMORY_READ_BIT,
                 .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
                 .newLayout = VK_IMAGE_LAYOUT_GENERAL,
                 .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
@@ -833,26 +831,11 @@ void BlitScale(Scheduler& scheduler, VkImage src_image, VkImage dst_image, const
                 .subresourceRange = subresource_range,
             },
         };
-        // Use specific pipeline stages for better parallelism
-        const VkPipelineStageFlags src_stages =
-            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-            VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-            VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_TRANSFER_BIT;
-        cmdbuf.PipelineBarrier(src_stages, VK_PIPELINE_STAGE_TRANSFER_BIT,
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
                                0, nullptr, nullptr, read_barriers);
         cmdbuf.BlitImage(src_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_image,
                          VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, regions, vk_filter);
-        // After transfer, images may be used in graphics, compute, or as attachments
-        const VkPipelineStageFlags dst_stages =
-            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-            VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-            VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stages,
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                                0, nullptr, nullptr, write_barriers);
     });
 }
@@ -1040,12 +1023,11 @@ void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src,
             .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
             .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
         };
-        // After reinterpret, buffers used as transfer source, uniforms, or storage
         static constexpr VkMemoryBarrier WRITE_BARRIER{
             .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
             .pNext = nullptr,
             .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-            .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_SHADER_READ_BIT,
+            .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
         };
         const std::array pre_barriers{
             VkImageMemoryBarrier{
@@ -1112,39 +1094,18 @@ void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src,
                 .subresourceRange = dst_range.SubresourceRange(dst_aspect_mask),
             },
         };
-        const VkPipelineStageFlags src_stages_transfer =
-            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-            VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-            VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_TRANSFER_BIT;
-        cmdbuf.PipelineBarrier(src_stages_transfer, VK_PIPELINE_STAGE_TRANSFER_BIT,
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
                                0, {}, {}, pre_barriers);
 
         cmdbuf.CopyImageToBuffer(src_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, copy_buffer,
                                  vk_in_copies);
-        const VkPipelineStageFlags dst_stages_transfer =
-            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-            VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-            VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-            VK_PIPELINE_STAGE_TRANSFER_BIT;
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stages_transfer,
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                                0, WRITE_BARRIER, nullptr, middle_in_barrier);
 
-        cmdbuf.PipelineBarrier(src_stages_transfer, VK_PIPELINE_STAGE_TRANSFER_BIT,
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
                                0, READ_BARRIER, {}, middle_out_barrier);
         cmdbuf.CopyBufferToImage(copy_buffer, dst_image, VK_IMAGE_LAYOUT_GENERAL, vk_out_copies);
-        // After reinterpret copy, image may be used in shaders or as attachment
-        const VkPipelineStageFlags dst_stages_post =
-            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-            VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-            VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stages_post,
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                                0, {}, {}, post_barriers);
     });
 }
@@ -1271,12 +1232,7 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst
                 .layerCount = VK_REMAINING_ARRAY_LAYERS,
             },
         };
-        const VkPipelineStageFlags src_stages_resolve =
-            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_TRANSFER_BIT;
-        cmdbuf.PipelineBarrier(src_stages_resolve, VK_PIPELINE_STAGE_TRANSFER_BIT,
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
                                0, nullptr, nullptr, read_barriers);
         if (is_resolve) {
             cmdbuf.ResolveImage(src_image, VK_IMAGE_LAYOUT_GENERAL, dst_image,
@@ -1289,14 +1245,7 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst
                 src_image, VK_IMAGE_LAYOUT_GENERAL, dst_image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
                 MakeImageBlit(dst_region, src_region, dst_layers, src_layers), vk_filter);
         }
-        // After blit/resolve, image may be used in shaders or as attachment
-        const VkPipelineStageFlags dst_stages_blit =
-            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-            VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-            VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stages_blit,
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                                0, write_barrier);
     });
 }
@@ -1874,14 +1823,7 @@ void Image::DownloadMemory(std::span<VkBuffer> buffers_span, std::span<size_t> o
                         .layerCount = VK_REMAINING_ARRAY_LAYERS,
                     },
                 };
-                const VkPipelineStageFlags src_stages_download =
-                    VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                    VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-                    VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-                    VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                    VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-                    VK_PIPELINE_STAGE_TRANSFER_BIT;
-                cmdbuf.PipelineBarrier(src_stages_download, VK_PIPELINE_STAGE_TRANSFER_BIT,
+                cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
                                        0, read_barrier);
 
                 for (size_t index = 0; index < buffers.size(); index++) {
@@ -1889,20 +1831,17 @@ void Image::DownloadMemory(std::span<VkBuffer> buffers_span, std::span<size_t> o
                                              vk_copies[index]);
                 }
 
-                // Host will read the downloaded data
                 const VkMemoryBarrier memory_write_barrier{
                     .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
                     .pNext = nullptr,
-                    .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-                    .dstAccessMask = VK_ACCESS_HOST_READ_BIT,
+                    .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+                    .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
                 };
-                // Image will be used for shaders or as attachment
                 const VkImageMemoryBarrier image_write_barrier{
                     .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                     .pNext = nullptr,
                     .srcAccessMask = 0,
-                    .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT |
-                                     VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
+                    .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
                     .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
                     .newLayout = VK_IMAGE_LAYOUT_GENERAL,
                     .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
@@ -1916,13 +1855,7 @@ void Image::DownloadMemory(std::span<VkBuffer> buffers_span, std::span<size_t> o
                         .layerCount = VK_REMAINING_ARRAY_LAYERS,
                     },
                 };
-                const VkPipelineStageFlags dst_stages_download =
-                    VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                    VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-                    VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                    VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-                    VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
-                cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stages_download,
+                cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                                        0, memory_write_barrier, nullptr, image_write_barrier);
             });
             return;
@@ -1957,15 +1890,7 @@ void Image::DownloadMemory(std::span<VkBuffer> buffers_span, std::span<size_t> o
                     .layerCount = VK_REMAINING_ARRAY_LAYERS,
                 },
             };
-            // Image writes can come from graphics or compute stages
-            const VkPipelineStageFlags src_stages_img =
-                VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-                VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-                VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-                VK_PIPELINE_STAGE_TRANSFER_BIT;
-            cmdbuf.PipelineBarrier(src_stages_img, VK_PIPELINE_STAGE_TRANSFER_BIT,
+            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
                                    0, read_barrier);
 
             for (size_t index = 0; index < buffers.size(); index++) {
@@ -1997,14 +1922,7 @@ void Image::DownloadMemory(std::span<VkBuffer> buffers_span, std::span<size_t> o
                     .layerCount = VK_REMAINING_ARRAY_LAYERS,
                 },
             };
-            // After download, image may be used in shaders or as attachment
-            const VkPipelineStageFlags dst_stages_img =
-                VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-                VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-                VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
-            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stages_img,
+            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
                                    0, memory_write_barrier, nullptr, image_write_barrier);
         });
     }
@@ -2566,9 +2484,7 @@ void TextureCacheRuntime::TransitionImageLayout(Image& image) {
             .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
             .pNext = nullptr,
             .srcAccessMask = VK_ACCESS_NONE,
-            .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT |
-                             VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
-                             VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
             .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED,
             .newLayout = VK_IMAGE_LAYOUT_GENERAL,
             .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
@@ -2584,16 +2500,8 @@ void TextureCacheRuntime::TransitionImageLayout(Image& image) {
         };
         scheduler.RequestOutsideRenderPassOperationContext();
         scheduler.Record([barrier](vk::CommandBuffer cmdbuf) {
-            // After layout transition, image may be used in shaders or as attachment
-            const VkPipelineStageFlags dst_stages_layout =
-                VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-                VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-                VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-                VK_PIPELINE_STAGE_TRANSFER_BIT;
             cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
-                                   dst_stages_layout, 0, barrier);
+                                   VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, barrier);
         });
     }
 }

src/video_core/renderer_vulkan/vk_update_descriptor.h

@@ -64,11 +64,19 @@ public:
     }
 
     void AddBuffer(VkBuffer buffer, VkDeviceSize offset, VkDeviceSize size) {
-        *(payload_cursor++) = VkDescriptorBufferInfo{
-            .buffer = buffer,
-            .offset = offset,
-            .range = size,
-        };
+        if (buffer == VK_NULL_HANDLE) {
+            *(payload_cursor++) = VkDescriptorBufferInfo{
+                .buffer = VK_NULL_HANDLE,
+                .offset = 0,
+                .range = VK_WHOLE_SIZE,
+            };
+        } else {
+            *(payload_cursor++) = VkDescriptorBufferInfo{
+                .buffer = buffer,
+                .offset = offset,
+                .range = size,
+            };
+        }
     }
 
     void AddTexelBuffer(VkBufferView texel_buffer) {

src/video_core/texture_cache/texture_cache.h

@@ -291,55 +291,59 @@ void TextureCache<P>::CheckFeedbackLoop(std::span<const ImageViewInOut> views) {
         return;
     }
 
+    if (render_targets_serial == last_feedback_loop_serial &&
+        texture_bindings_serial == last_feedback_texture_serial) {
+        if (last_feedback_loop_result) {
+            runtime.BarrierFeedbackLoop();
+        }
+        return;
+    }
+
+    if (rt_active_mask == 0) {
+        last_feedback_loop_serial = render_targets_serial;
+        last_feedback_texture_serial = texture_bindings_serial;
+        last_feedback_loop_result = false;
+        return;
+    }
+    const u32 depth_bit = 1u << NUM_RT;
+    const bool depth_active = (rt_active_mask & depth_bit) != 0;
+
     const bool requires_barrier = [&] {
         for (const auto& view : views) {
             if (!view.id) {
                 continue;
             }
 
-            bool is_render_target = false;
-
-            for (const auto& ct_view_id : render_targets.color_buffer_ids) {
-                if (ct_view_id && ct_view_id == view.id) {
-                    is_render_target = true;
-                    break;
-                }
-            }
-
-            if (!is_render_target && render_targets.depth_buffer_id == view.id) {
-                is_render_target = true;
-            }
-
-            if (is_render_target) {
-                continue;
-            }
-
-            auto& image_view = slot_image_views[view.id];
-
-            for (const auto& ct_view_id : render_targets.color_buffer_ids) {
-                if (!ct_view_id) {
+            {
+                bool is_continue = false;
+                for (size_t i = 0; i < 8; ++i)
+                    is_continue |= (rt_active_mask & (1u << i)) && view.id == render_targets.color_buffer_ids[i];
+                if (is_continue)
                     continue;
-                }
-
-                auto& ct_view = slot_image_views[ct_view_id];
-
-                if (image_view.image_id == ct_view.image_id) {
-                    return true;
-                }
             }
 
-            if (render_targets.depth_buffer_id) {
-                auto& zt_view = slot_image_views[render_targets.depth_buffer_id];
+            if (depth_active && view.id == render_targets.depth_buffer_id)
+                continue;
 
-                if (image_view.image_id == zt_view.image_id) {
-                    return true;
-                }
+            const ImageId view_image_id = slot_image_views[view.id].image_id;
+            {
+                bool is_continue = false;
+                for (size_t i = 0; i < 8; ++i)
+                    is_continue |= (rt_active_mask & (1u << i)) && view_image_id == rt_image_id[i];
+                if (is_continue)
+                    continue;
+            }
+            if (depth_active && view_image_id == rt_depth_image_id) {
+                return true;
             }
         }
 
         return false;
     }();
 
+    last_feedback_loop_serial = render_targets_serial;
+    last_feedback_texture_serial = texture_bindings_serial;
+    last_feedback_loop_result = requires_barrier;
     if (requires_barrier) {
         runtime.BarrierFeedbackLoop();
     }
@@ -399,13 +403,19 @@ void TextureCache<P>::SynchronizeGraphicsDescriptors() {
     const bool linked_tsc = maxwell3d->regs.sampler_binding == SamplerBinding::ViaHeaderBinding;
     const u32 tic_limit = maxwell3d->regs.tex_header.limit;
     const u32 tsc_limit = linked_tsc ? tic_limit : maxwell3d->regs.tex_sampler.limit;
+    bool bindings_changed = false;
     if (channel_state->graphics_sampler_table.Synchronize(maxwell3d->regs.tex_sampler.Address(),
                                                           tsc_limit)) {
         channel_state->graphics_sampler_ids.resize(tsc_limit + 1, CORRUPT_ID);
+        bindings_changed = true;
     }
     if (channel_state->graphics_image_table.Synchronize(maxwell3d->regs.tex_header.Address(),
                                                         tic_limit)) {
         channel_state->graphics_image_view_ids.resize(tic_limit + 1, CORRUPT_ID);
+        bindings_changed = true;
+    }
+    if (bindings_changed) {
+        ++texture_bindings_serial;
     }
 }
 
@@ -415,12 +425,18 @@ void TextureCache<P>::SynchronizeComputeDescriptors() {
     const u32 tic_limit = kepler_compute->regs.tic.limit;
     const u32 tsc_limit = linked_tsc ? tic_limit : kepler_compute->regs.tsc.limit;
     const GPUVAddr tsc_gpu_addr = kepler_compute->regs.tsc.Address();
+    bool bindings_changed = false;
     if (channel_state->compute_sampler_table.Synchronize(tsc_gpu_addr, tsc_limit)) {
         channel_state->compute_sampler_ids.resize(tsc_limit + 1, CORRUPT_ID);
+        bindings_changed = true;
     }
     if (channel_state->compute_image_table.Synchronize(kepler_compute->regs.tic.Address(),
                                                        tic_limit)) {
         channel_state->compute_image_view_ids.resize(tic_limit + 1, CORRUPT_ID);
+        bindings_changed = true;
+    }
+    if (bindings_changed) {
+        ++texture_bindings_serial;
     }
 }
 
@@ -534,6 +550,7 @@ void TextureCache<P>::UpdateRenderTargets(bool is_clear) {
         return;
     }
 
+    const VideoCommon::RenderTargets previous_render_targets = render_targets;
     const bool rescaled = RescaleRenderTargets();
     if (is_rescaling != rescaled) {
         flags[Dirty::RescaleViewports] = true;
@@ -549,6 +566,21 @@ void TextureCache<P>::UpdateRenderTargets(bool is_clear) {
 
     PrepareImageView(depth_buffer_id, true, is_clear && IsFullClear(depth_buffer_id));
 
+    rt_active_mask = 0;
+    rt_image_id = {};
+    for (size_t i = 0; i < rt_image_id.size(); ++i) {
+        if (ImageViewId const view = render_targets.color_buffer_ids[i]; view) {
+            rt_active_mask |= 1u << i;
+            rt_image_id[i] = slot_image_views[view].image_id;
+        }
+    }
+    if (depth_buffer_id) {
+        rt_active_mask |= (1u << NUM_RT);
+        rt_depth_image_id = slot_image_views[depth_buffer_id].image_id;
+    } else {
+        rt_depth_image_id = ImageId{};
+    }
+
     for (size_t index = 0; index < NUM_RT; ++index) {
         render_targets.draw_buffers[index] = static_cast<u8>(maxwell3d->regs.rt_control.Map(index));
     }
@@ -564,12 +596,22 @@ void TextureCache<P>::UpdateRenderTargets(bool is_clear) {
     };
     render_targets.is_rescaled = is_rescaling;
 
+    if (render_targets != previous_render_targets) {
+        ++render_targets_serial;
+    }
+
     flags[Dirty::DepthBiasGlobal] = true;
 }
 
 template <class P>
 typename P::Framebuffer* TextureCache<P>::GetFramebuffer() {
-    return &slot_framebuffers[GetFramebufferId(render_targets)];
+    if (last_framebuffer_id && last_framebuffer_serial == render_targets_serial) {
+        return &slot_framebuffers[last_framebuffer_id];
+    }
+    const FramebufferId framebuffer_id = GetFramebufferId(render_targets);
+    last_framebuffer_id = framebuffer_id;
+    last_framebuffer_serial = render_targets_serial;
+    return &slot_framebuffers[framebuffer_id];
 }
 
 template <class P>
@@ -2610,6 +2652,10 @@ void TextureCache<P>::RemoveFramebuffers(std::span<const ImageViewId> removed_vi
         if (it->first.Contains(removed_views)) {
             auto framebuffer_id = it->second;
             ASSERT(framebuffer_id);
+            if (framebuffer_id == last_framebuffer_id) {
+                last_framebuffer_id = {};
+                last_framebuffer_serial = 0;
+            }
             sentenced_framebuffers.Push(std::move(slot_framebuffers[framebuffer_id]));
             it = framebuffers.erase(it);
         } else {

src/video_core/texture_cache/texture_cache_base.h

@@ -455,6 +455,16 @@ private:
     std::deque<TextureCacheGPUMap> gpu_page_table_storage;
 
     RenderTargets render_targets;
+    u64 render_targets_serial = 0;
+    u32 rt_active_mask = 0;
+    std::array<ImageId, 8> rt_image_id{};
+    ImageId rt_depth_image_id{};
+    u64 texture_bindings_serial = 0;
+    u64 last_feedback_loop_serial = 0;
+    u64 last_feedback_texture_serial = 0;
+    bool last_feedback_loop_result = false;
+    FramebufferId last_framebuffer_id{};
+    u64 last_framebuffer_serial = 0;
 
     ankerl::unordered_dense::map<RenderTargets, FramebufferId> framebuffers;
     ankerl::unordered_dense::map<u64, std::vector<ImageMapId>, Common::IdentityHash<u64>> page_table;

src/yuzu/main_window.cpp

@@ -2142,7 +2142,7 @@ void MainWindow::BootGame(const QString& filename, Service::AM::FrontendAppletPa
         game_list->hide();
         game_list_placeholder->hide();
     }
-    status_bar_update_timer.start(250);
+    status_bar_update_timer.start(500);
     renderer_status_button->setDisabled(true);
     refresh_button->setDisabled(true);