"Merge Tagged PR 6598"

.ci/templates/build-msvc.yml

@@ -8,7 +8,7 @@ steps:
   displayName: 'Install vulkan-sdk'
 - script: python -m pip install --upgrade pip conan
   displayName: 'Install conan'
-- script: refreshenv && mkdir build && cd build && cmake -G "Visual Studio 16 2019" -A x64 -DYUZU_USE_BUNDLED_QT=1 -DYUZU_USE_BUNDLED_SDL2=1 -DYUZU_USE_QT_WEB_ENGINE=ON -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DYUZU_ENABLE_COMPATIBILITY_REPORTING=${COMPAT} -DUSE_DISCORD_PRESENCE=ON -DENABLE_QT_TRANSLATION=ON -DDISPLAY_VERSION=${{ parameters['version'] }} -DCMAKE_BUILD_TYPE=Release .. && cd ..
+- script: refreshenv && mkdir build && cd build && cmake -E env CXXFLAGS="/Gw /GA /Gr /Ob2" cmake -G "Visual Studio 16 2019" -A x64 -DCMAKE_INTERPROCEDURAL_OPTIMIZATION=ON -DCMAKE_POLICY_DEFAULT_CMP0069=NEW -DYUZU_USE_BUNDLED_QT=1 -DYUZU_USE_BUNDLED_SDL2=1 -DYUZU_USE_QT_WEB_ENGINE=ON -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DYUZU_ENABLE_COMPATIBILITY_REPORTING=${COMPAT} -DUSE_DISCORD_PRESENCE=ON -DENABLE_QT_TRANSLATION=ON -DDISPLAY_VERSION=${{ parameters['version'] }} -DCMAKE_BUILD_TYPE=Release .. && cd ..
   displayName: 'Configure CMake'
 - task: MSBuild@1
   displayName: 'Build'

src/CMakeLists.txt

@@ -37,7 +37,8 @@ if (MSVC)
     add_compile_options(
         /MP
         /Zi
-        /Zm200
+        /Zm300
+        /Zf
         /Zo
         /permissive-
         /EHsc
@@ -81,7 +82,7 @@ if (MSVC)
     add_compile_options("$<$<CONFIG:Release>:/GS->")
 
     set(CMAKE_EXE_LINKER_FLAGS_DEBUG   "/DEBUG /MANIFEST:NO" CACHE STRING "" FORCE)
-    set(CMAKE_EXE_LINKER_FLAGS_RELEASE "/DEBUG /MANIFEST:NO /INCREMENTAL:NO /OPT:REF,ICF" CACHE STRING "" FORCE)
+    set(CMAKE_EXE_LINKER_FLAGS_RELEASE "${CMAKE_EXE_LINKER_FLAGS_RELEASE} /DEBUG /MANIFEST:NO /INCREMENTAL:NO /OPT:REF,ICF" CACHE STRING "" FORCE)
 else()
     add_compile_options(
         -Wall

src/audio_core/stream.cpp

@@ -87,6 +87,14 @@ static void VolumeAdjustSamples(std::vector<s16>& samples, float game_volume) {
 }
 
 void Stream::PlayNextBuffer(std::chrono::nanoseconds ns_late) {
+    auto now = std::chrono::steady_clock::now();
+    auto duration = now.time_since_epoch();
+    auto nanoseconds = std::chrono::duration_cast<std::chrono::nanoseconds>(duration);
+
+    if (nanoseconds > expected_cb_time) {
+        ns_late = nanoseconds - expected_cb_time;
+    }
+
     if (!IsPlaying()) {
         // Ensure we are in playing state before playing the next buffer
         sink_stream.Flush();
@@ -121,6 +129,7 @@ void Stream::PlayNextBuffer(std::chrono::nanoseconds ns_late) {
         ns_late = {};
     }
 
+    expected_cb_time = nanoseconds + (buffer_release_ns - ns_late);
     core_timing.ScheduleEvent(buffer_release_ns - ns_late, release_event, {});
 }
 

src/audio_core/stream.h

@@ -117,13 +117,14 @@ private:
     ReleaseCallback release_callback; ///< Buffer release callback for the stream
     State state{State::Stopped};      ///< Playback state of the stream
     std::shared_ptr<Core::Timing::EventType>
-        release_event;                      ///< Core timing release event for the stream
-    BufferPtr active_buffer;                ///< Actively playing buffer in the stream
-    std::queue<BufferPtr> queued_buffers;   ///< Buffers queued to be played in the stream
-    std::queue<BufferPtr> released_buffers; ///< Buffers recently released from the stream
-    SinkStream& sink_stream;                ///< Output sink for the stream
-    Core::Timing::CoreTiming& core_timing;  ///< Core timing instance.
-    std::string name;                       ///< Name of the stream, must be unique
+        release_event;                              ///< Core timing release event for the stream
+    BufferPtr active_buffer;                        ///< Actively playing buffer in the stream
+    std::queue<BufferPtr> queued_buffers;           ///< Buffers queued to be played in the stream
+    std::queue<BufferPtr> released_buffers;         ///< Buffers recently released from the stream
+    SinkStream& sink_stream;                        ///< Output sink for the stream
+    Core::Timing::CoreTiming& core_timing;          ///< Core timing instance.
+    std::string name;                               ///< Name of the stream, must be unique
+    std::chrono::nanoseconds expected_cb_time = {}; ///< Estimated time of next callback
 };
 
 using StreamPtr = std::shared_ptr<Stream>;

src/common/assert.h

@@ -32,14 +32,14 @@ assert_noinline_call(const Fn& fn) {
 
 #define ASSERT(_a_)                                                                                \
     do                                                                                             \
-        if (!(_a_)) {                                                                              \
+        if (!(_a_)) [[unlikely]] {                                                                 \
             assert_noinline_call([] { LOG_CRITICAL(Debug, "Assertion Failed!"); });                \
         }                                                                                          \
     while (0)
 
 #define ASSERT_MSG(_a_, ...)                                                                       \
     do                                                                                             \
-        if (!(_a_)) {                                                                              \
+        if (!(_a_)) [[unlikely]] {                                                                 \
             assert_noinline_call([&] { LOG_CRITICAL(Debug, "Assertion Failed!\n" __VA_ARGS__); }); \
         }                                                                                          \
     while (0)
@@ -70,7 +70,7 @@ assert_noinline_call(const Fn& fn) {
 #define ASSERT_OR_EXECUTE(_a_, _b_)                                                                \
     do {                                                                                           \
         ASSERT(_a_);                                                                               \
-        if (!(_a_)) {                                                                              \
+        if (!(_a_)) [[unlikely]] {                                                                 \
             _b_                                                                                    \
         }                                                                                          \
     } while (0)
@@ -79,7 +79,7 @@ assert_noinline_call(const Fn& fn) {
 #define ASSERT_OR_EXECUTE_MSG(_a_, _b_, ...)                                                       \
     do {                                                                                           \
         ASSERT_MSG(_a_, __VA_ARGS__);                                                              \
-        if (!(_a_)) {                                                                              \
+        if (!(_a_)) [[unlikely]] {                                                                 \
             _b_                                                                                    \
         }                                                                                          \
     } while (0)

src/core/hle/kernel/k_page_table.cpp

@@ -285,72 +285,142 @@ ResultCode KPageTable::MapProcessCode(VAddr addr, std::size_t num_pages, KMemory
     return ResultSuccess;
 }
 
-ResultCode KPageTable::MapCodeMemory(VAddr dst_addr, VAddr src_addr, std::size_t size) {
+ResultCode KPageTable::MapCodeMemory(VAddr dst_address, VAddr src_address, std::size_t size) {
+    // Validate the mapping request.
+    R_UNLESS(this->CanContain(dst_address, size, KMemoryState::AliasCode),
+             ResultInvalidMemoryRegion);
+
+    // Lock the table.
     KScopedLightLock lk(general_lock);
 
-    const std::size_t num_pages{size / PageSize};
+    // Verify that the source memory is normal heap.
+    KMemoryState src_state{};
+    KMemoryPermission src_perm{};
+    std::size_t num_src_allocator_blocks{};
+    R_TRY(this->CheckMemoryState(std::addressof(src_state), std::addressof(src_perm), nullptr,
+                                 std::addressof(num_src_allocator_blocks), src_address, size,
+                                 KMemoryState::All, KMemoryState::Normal, KMemoryPermission::All,
+                                 KMemoryPermission::UserReadWrite, KMemoryAttribute::All,
+                                 KMemoryAttribute::None));
 
-    KMemoryState state{};
-    KMemoryPermission perm{};
-    CASCADE_CODE(CheckMemoryState(&state, &perm, nullptr, nullptr, src_addr, size,
-                                  KMemoryState::All, KMemoryState::Normal, KMemoryPermission::All,
-                                  KMemoryPermission::UserReadWrite, KMemoryAttribute::Mask,
-                                  KMemoryAttribute::None, KMemoryAttribute::IpcAndDeviceMapped));
-
-    if (IsRegionMapped(dst_addr, size)) {
-        return ResultInvalidCurrentMemory;
-    }
-
-    KPageLinkedList page_linked_list;
-    AddRegionToPages(src_addr, num_pages, page_linked_list);
+    // Verify that the destination memory is unmapped.
+    std::size_t num_dst_allocator_blocks{};
+    R_TRY(this->CheckMemoryState(std::addressof(num_dst_allocator_blocks), dst_address, size,
+                                 KMemoryState::All, KMemoryState::Free, KMemoryPermission::None,
+                                 KMemoryPermission::None, KMemoryAttribute::None,
+                                 KMemoryAttribute::None));
 
+    // Map the code memory.
     {
-        auto block_guard = detail::ScopeExit(
-            [&] { Operate(src_addr, num_pages, perm, OperationType::ChangePermissions); });
+        // Determine the number of pages being operated on.
+        const std::size_t num_pages = size / PageSize;
 
-        CASCADE_CODE(Operate(src_addr, num_pages, KMemoryPermission::None,
-                             OperationType::ChangePermissions));
-        CASCADE_CODE(MapPages(dst_addr, page_linked_list, KMemoryPermission::None));
+        // Create page groups for the memory being mapped.
+        KPageLinkedList pg;
+        AddRegionToPages(src_address, num_pages, pg);
 
-        block_guard.Cancel();
+        // Reprotect the source as kernel-read/not mapped.
+        const KMemoryPermission new_perm = static_cast<KMemoryPermission>(
+            KMemoryPermission::KernelRead | KMemoryPermission::NotMapped);
+        R_TRY(Operate(src_address, num_pages, new_perm, OperationType::ChangePermissions));
+
+        // Ensure that we unprotect the source pages on failure.
+        auto unprot_guard = SCOPE_GUARD({
+            ASSERT(this->Operate(src_address, num_pages, src_perm, OperationType::ChangePermissions)
+                       .IsSuccess());
+        });
+
+        // Map the alias pages.
+        R_TRY(MapPages(dst_address, pg, new_perm));
+
+        // We successfully mapped the alias pages, so we don't need to unprotect the src pages on
+        // failure.
+        unprot_guard.Cancel();
+
+        // Apply the memory block updates.
+        block_manager->Update(src_address, num_pages, src_state, new_perm,
+                              KMemoryAttribute::Locked);
+        block_manager->Update(dst_address, num_pages, KMemoryState::AliasCode, new_perm,
+                              KMemoryAttribute::None);
     }
 
-    block_manager->Update(src_addr, num_pages, state, KMemoryPermission::None,
-                          KMemoryAttribute::Locked);
-    block_manager->Update(dst_addr, num_pages, KMemoryState::AliasCode);
-
     return ResultSuccess;
 }
 
-ResultCode KPageTable::UnmapCodeMemory(VAddr dst_addr, VAddr src_addr, std::size_t size) {
+ResultCode KPageTable::UnmapCodeMemory(VAddr dst_address, VAddr src_address, std::size_t size) {
+    // Validate the mapping request.
+    R_UNLESS(this->CanContain(dst_address, size, KMemoryState::AliasCode),
+             ResultInvalidMemoryRegion);
+
+    // Lock the table.
     KScopedLightLock lk(general_lock);
 
-    if (!size) {
-        return ResultSuccess;
+    // Verify that the source memory is locked normal heap.
+    std::size_t num_src_allocator_blocks{};
+    R_TRY(this->CheckMemoryState(std::addressof(num_src_allocator_blocks), src_address, size,
+                                 KMemoryState::All, KMemoryState::Normal, KMemoryPermission::None,
+                                 KMemoryPermission::None, KMemoryAttribute::All,
+                                 KMemoryAttribute::Locked));
+
+    // Verify that the destination memory is aliasable code.
+    std::size_t num_dst_allocator_blocks{};
+    R_TRY(this->CheckMemoryStateContiguous(
+        std::addressof(num_dst_allocator_blocks), dst_address, size, KMemoryState::FlagCanCodeAlias,
+        KMemoryState::FlagCanCodeAlias, KMemoryPermission::None, KMemoryPermission::None,
+        KMemoryAttribute::All, KMemoryAttribute::None));
+
+    // Determine whether any pages being unmapped are code.
+    bool any_code_pages = false;
+    {
+        KMemoryBlockManager::const_iterator it = block_manager->FindIterator(dst_address);
+        while (true) {
+            // Get the memory info.
+            const KMemoryInfo info = it->GetMemoryInfo();
+
+            // Check if the memory has code flag.
+            if ((info.GetState() & KMemoryState::FlagCode) != KMemoryState::None) {
+                any_code_pages = true;
+                break;
+            }
+
+            // Check if we're done.
+            if (dst_address + size - 1 <= info.GetLastAddress()) {
+                break;
+            }
+
+            // Advance.
+            ++it;
+        }
     }
 
-    const std::size_t num_pages{size / PageSize};
+    // Ensure that we maintain the instruction cache.
+    bool reprotected_pages = false;
+    SCOPE_EXIT({
+        if (reprotected_pages && any_code_pages) {
+            system.InvalidateCpuInstructionCacheRange(dst_address, size);
+        }
+    });
 
-    CASCADE_CODE(CheckMemoryState(nullptr, nullptr, nullptr, nullptr, src_addr, size,
-                                  KMemoryState::All, KMemoryState::Normal, KMemoryPermission::None,
-                                  KMemoryPermission::None, KMemoryAttribute::Mask,
-                                  KMemoryAttribute::Locked, KMemoryAttribute::IpcAndDeviceMapped));
+    // Unmap.
+    {
+        // Determine the number of pages being operated on.
+        const std::size_t num_pages = size / PageSize;
 
-    KMemoryState state{};
-    CASCADE_CODE(CheckMemoryState(
-        &state, nullptr, nullptr, nullptr, dst_addr, PageSize, KMemoryState::FlagCanCodeAlias,
-        KMemoryState::FlagCanCodeAlias, KMemoryPermission::None, KMemoryPermission::None,
-        KMemoryAttribute::Mask, KMemoryAttribute::None, KMemoryAttribute::IpcAndDeviceMapped));
-    CASCADE_CODE(CheckMemoryState(dst_addr, size, KMemoryState::All, state, KMemoryPermission::None,
-                                  KMemoryPermission::None, KMemoryAttribute::Mask,
-                                  KMemoryAttribute::None));
-    CASCADE_CODE(Operate(dst_addr, num_pages, KMemoryPermission::None, OperationType::Unmap));
+        // Unmap the aliased copy of the pages.
+        R_TRY(Operate(dst_address, num_pages, KMemoryPermission::None, OperationType::Unmap));
 
-    block_manager->Update(dst_addr, num_pages, KMemoryState::Free);
-    block_manager->Update(src_addr, num_pages, KMemoryState::Normal,
-                          KMemoryPermission::UserReadWrite);
+        // Try to set the permissions for the source pages back to what they should be.
+        R_TRY(Operate(src_address, num_pages, KMemoryPermission::UserReadWrite,
+                      OperationType::ChangePermissions));
 
-    system.InvalidateCpuInstructionCacheRange(dst_addr, size);
+        // Apply the memory block updates.
+        block_manager->Update(dst_address, num_pages, KMemoryState::None);
+        block_manager->Update(src_address, num_pages, KMemoryState::Normal,
+                              KMemoryPermission::UserReadWrite);
+
+        // Note that we reprotected pages.
+        reprotected_pages = true;
+    }
 
     return ResultSuccess;
 }

src/core/hle/kernel/k_page_table.h

@@ -36,8 +36,8 @@ public:
                                     KMemoryManager::Pool pool);
     ResultCode MapProcessCode(VAddr addr, std::size_t pages_count, KMemoryState state,
                               KMemoryPermission perm);
-    ResultCode MapCodeMemory(VAddr dst_addr, VAddr src_addr, std::size_t size);
-    ResultCode UnmapCodeMemory(VAddr dst_addr, VAddr src_addr, std::size_t size);
+    ResultCode MapCodeMemory(VAddr dst_address, VAddr src_address, std::size_t size);
+    ResultCode UnmapCodeMemory(VAddr dst_address, VAddr src_address, std::size_t size);
     ResultCode UnmapProcessMemory(VAddr dst_addr, std::size_t size, KPageTable& src_page_table,
                                   VAddr src_addr);
     ResultCode MapPhysicalMemory(VAddr addr, std::size_t size);
@@ -253,7 +253,9 @@ public:
     constexpr bool IsInsideASLRRegion(VAddr address, std::size_t size) const {
         return !IsOutsideASLRRegion(address, size);
     }
-
+    constexpr std::size_t GetNumGuardPages() const {
+        return IsKernel() ? 1 : 4;
+    }
     PAddr GetPhysicalAddr(VAddr addr) const {
         const auto backing_addr = page_table_impl.backing_addr[addr >> PageBits];
         ASSERT(backing_addr);
@@ -275,10 +277,6 @@ private:
         return is_aslr_enabled;
     }
 
-    constexpr std::size_t GetNumGuardPages() const {
-        return IsKernel() ? 1 : 4;
-    }
-
     constexpr bool ContainsPages(VAddr addr, std::size_t num_pages) const {
         return (address_space_start <= addr) &&
                (num_pages <= (address_space_end - address_space_start) / PageSize) &&

src/core/hle/service/ldr/ldr.cpp

@@ -288,7 +288,7 @@ public:
     }
 
     bool ValidateRegionForMap(Kernel::KPageTable& page_table, VAddr start, std::size_t size) const {
-        constexpr std::size_t padding_size{4 * Kernel::PageSize};
+        const std::size_t padding_size{page_table.GetNumGuardPages() * Kernel::PageSize};
         const auto start_info{page_table.QueryInfo(start - 1)};
 
         if (start_info.state != Kernel::KMemoryState::Free) {
@@ -308,31 +308,69 @@ public:
         return (start + size + padding_size) <= (end_info.GetAddress() + end_info.GetSize());
     }
 
-    VAddr GetRandomMapRegion(const Kernel::KPageTable& page_table, std::size_t size) const {
-        VAddr addr{};
-        const std::size_t end_pages{(page_table.GetAliasCodeRegionSize() - size) >>
-                                    Kernel::PageBits};
-        do {
-            addr = page_table.GetAliasCodeRegionStart() +
-                   (Kernel::KSystemControl::GenerateRandomRange(0, end_pages) << Kernel::PageBits);
-        } while (!page_table.IsInsideAddressSpace(addr, size) ||
-                 page_table.IsInsideHeapRegion(addr, size) ||
-                 page_table.IsInsideAliasRegion(addr, size));
-        return addr;
+    ResultCode GetAvailableMapRegion(Kernel::KPageTable& page_table, u64 size, VAddr& out_addr) {
+        size = Common::AlignUp(size, Kernel::PageSize);
+        size += page_table.GetNumGuardPages() * Kernel::PageSize * 4;
+
+        auto is_region_available = [&](VAddr addr) {
+            const auto end_addr = addr + size;
+            while (addr < end_addr) {
+                if (system.Memory().IsValidVirtualAddress(addr)) {
+                    return false;
+                }
+
+                if (!page_table.IsInsideAddressSpace(out_addr, size)) {
+                    return false;
+                }
+
+                if (page_table.IsInsideHeapRegion(out_addr, size)) {
+                    return false;
+                }
+
+                if (page_table.IsInsideAliasRegion(out_addr, size)) {
+                    return false;
+                }
+
+                addr += Kernel::PageSize;
+            }
+            return true;
+        };
+
+        bool succeeded = false;
+        const auto map_region_end =
+            page_table.GetAliasCodeRegionStart() + page_table.GetAliasCodeRegionSize();
+        while (current_map_addr < map_region_end) {
+            if (is_region_available(current_map_addr)) {
+                succeeded = true;
+                break;
+            }
+            current_map_addr += 0x100000;
+        }
+
+        if (!succeeded) {
+            UNREACHABLE_MSG("Out of address space!");
+            return Kernel::ResultOutOfMemory;
+        }
+
+        out_addr = current_map_addr;
+        current_map_addr += size;
+
+        return ResultSuccess;
     }
 
-    ResultVal<VAddr> MapProcessCodeMemory(Kernel::KProcess* process, VAddr baseAddress,
-                                          u64 size) const {
-        for (std::size_t retry = 0; retry < MAXIMUM_MAP_RETRIES; retry++) {
-            auto& page_table{process->PageTable()};
-            const VAddr addr{GetRandomMapRegion(page_table, size)};
-            const ResultCode result{page_table.MapCodeMemory(addr, baseAddress, size)};
+    ResultVal<VAddr> MapProcessCodeMemory(Kernel::KProcess* process, VAddr base_addr, u64 size) {
+        auto& page_table{process->PageTable()};
+        VAddr addr{};
 
+        for (std::size_t retry = 0; retry < MAXIMUM_MAP_RETRIES; retry++) {
+            R_TRY(GetAvailableMapRegion(page_table, size, addr));
+
+            const ResultCode result{page_table.MapCodeMemory(addr, base_addr, size)};
             if (result == Kernel::ResultInvalidCurrentMemory) {
                 continue;
             }
 
-            CASCADE_CODE(result);
+            R_TRY(result);
 
             if (ValidateRegionForMap(page_table, addr, size)) {
                 return addr;
@@ -343,7 +381,7 @@ public:
     }
 
     ResultVal<VAddr> MapNro(Kernel::KProcess* process, VAddr nro_addr, std::size_t nro_size,
-                            VAddr bss_addr, std::size_t bss_size, std::size_t size) const {
+                            VAddr bss_addr, std::size_t bss_size, std::size_t size) {
         for (std::size_t retry = 0; retry < MAXIMUM_MAP_RETRIES; retry++) {
             auto& page_table{process->PageTable()};
             VAddr addr{};
@@ -597,6 +635,7 @@ public:
         LOG_WARNING(Service_LDR, "(STUBBED) called");
 
         initialized = true;
+        current_map_addr = system.CurrentProcess()->PageTable().GetAliasCodeRegionStart();
 
         IPC::ResponseBuilder rb{ctx, 2};
         rb.Push(ResultSuccess);
@@ -607,6 +646,7 @@ private:
 
     std::map<VAddr, NROInfo> nro;
     std::map<VAddr, std::vector<SHA256Hash>> nrr;
+    VAddr current_map_addr{};
 
     bool IsValidNROHash(const SHA256Hash& hash) const {
         return std::any_of(nrr.begin(), nrr.end(), [&hash](const auto& p) {

src/input_common/input_engine.h

@@ -207,7 +207,7 @@ private:
                                 int index) const;
 
     mutable std::mutex mutex;
-    mutable std::mutex mutex_callback;
+    mutable std::recursive_mutex mutex_callback;
     bool configuring{false};
     const std::string input_engine;
     int last_callback_key = 0;

src/video_core/buffer_cache/buffer_cache.h

@@ -1469,19 +1469,29 @@ typename BufferCache<P>::OverlapResult BufferCache<P>::ResolveOverlaps(VAddr cpu
         overlap_ids.push_back(overlap_id);
         overlap.Pick();
         const VAddr overlap_cpu_addr = overlap.CpuAddr();
+        bool goes_left = false;
         if (overlap_cpu_addr < begin) {
+            goes_left = true;
             cpu_addr = begin = overlap_cpu_addr;
         }
-        end = std::max(end, overlap_cpu_addr + overlap.SizeBytes());
-
+        const VAddr overlap_end = overlap_cpu_addr + overlap.SizeBytes();
+        bool goes_right = false;
+        if (overlap_end > end) {
+            goes_right = true;
+            end = overlap_end;
+        }
         stream_score += overlap.StreamScore();
         if (stream_score > STREAM_LEAP_THRESHOLD && !has_stream_leap) {
             // When this memory region has been joined a bunch of times, we assume it's being used
             // as a stream buffer. Increase the size to skip constantly recreating buffers.
             has_stream_leap = true;
-            begin -= PAGE_SIZE * 256;
-            cpu_addr = begin;
-            end += PAGE_SIZE * 256;
+            if (goes_right) {
+                begin -= PAGE_SIZE * 256;
+                cpu_addr = begin;
+            }
+            if (goes_left) {
+                end += PAGE_SIZE * 256;
+            }
         }
     }
     return OverlapResult{

src/video_core/macro/macro_hle.cpp

@@ -4,6 +4,8 @@
 
 #include <array>
 #include <vector>
+#include "common/scope_exit.h"
+#include "video_core/dirty_flags.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/macro/macro.h"
 #include "video_core/macro/macro_hle.h"
@@ -59,6 +61,7 @@ void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
     maxwell3d.regs.index_array.first = parameters[3];
     maxwell3d.regs.reg_array[0x446] = element_base; // vertex id base?
     maxwell3d.regs.index_array.count = parameters[1];
+    maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
     maxwell3d.regs.vb_element_base = element_base;
     maxwell3d.regs.vb_base_instance = base_instance;
     maxwell3d.mme_draw.instance_count = instance_count;
@@ -81,10 +84,67 @@ void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
     maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
 }
 
-constexpr std::array<std::pair<u64, HLEFunction>, 3> hle_funcs{{
+// Multidraw Indirect
+void HLE_3f5e74b9c9a50164(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& parameters) {
+    SCOPE_EXIT({
+        // Clean everything.
+        maxwell3d.regs.reg_array[0x446] = 0x0; // vertex id base?
+        maxwell3d.regs.index_array.count = 0;
+        maxwell3d.regs.vb_element_base = 0x0;
+        maxwell3d.regs.vb_base_instance = 0x0;
+        maxwell3d.mme_draw.instance_count = 0;
+        maxwell3d.CallMethodFromMME(0x8e3, 0x640);
+        maxwell3d.CallMethodFromMME(0x8e4, 0x0);
+        maxwell3d.CallMethodFromMME(0x8e5, 0x0);
+        maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
+        maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
+    });
+    const u32 start_indirect = parameters[0];
+    const u32 end_indirect = parameters[1];
+    if (start_indirect >= end_indirect) {
+        // Nothing to do.
+        return;
+    }
+    const auto topology =
+        static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[2]);
+    maxwell3d.regs.draw.topology.Assign(topology);
+    const u32 padding = parameters[3];
+    const std::size_t max_draws = parameters[4];
+
+    const u32 indirect_words = 5 + padding;
+    const std::size_t first_draw = start_indirect;
+    const std::size_t effective_draws = end_indirect - start_indirect;
+    const std::size_t last_draw = start_indirect + std::min(effective_draws, max_draws);
+
+    for (std::size_t index = first_draw; index < last_draw; index++) {
+        const std::size_t base = index * indirect_words + 5;
+        const u32 num_vertices = parameters[base];
+        const u32 instance_count = parameters[base + 1];
+        const u32 first_index = parameters[base + 2];
+        const u32 base_vertex = parameters[base + 3];
+        const u32 base_instance = parameters[base + 4];
+        maxwell3d.regs.index_array.first = first_index;
+        maxwell3d.regs.reg_array[0x446] = base_vertex;
+        maxwell3d.regs.index_array.count = num_vertices;
+        maxwell3d.regs.vb_element_base = base_vertex;
+        maxwell3d.regs.vb_base_instance = base_instance;
+        maxwell3d.mme_draw.instance_count = instance_count;
+        maxwell3d.CallMethodFromMME(0x8e3, 0x640);
+        maxwell3d.CallMethodFromMME(0x8e4, base_vertex);
+        maxwell3d.CallMethodFromMME(0x8e5, base_instance);
+        maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
+        if (maxwell3d.ShouldExecute()) {
+            maxwell3d.Rasterizer().Draw(true, true);
+        }
+        maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
+    }
+}
+
+constexpr std::array<std::pair<u64, HLEFunction>, 4> hle_funcs{{
     {0x771BB18C62444DA0, &HLE_771BB18C62444DA0},
     {0x0D61FC9FAAC9FCAD, &HLE_0D61FC9FAAC9FCAD},
     {0x0217920100488FF7, &HLE_0217920100488FF7},
+    {0x3f5e74b9c9a50164, &HLE_3f5e74b9c9a50164},
 }};
 
 class HLEMacroImpl final : public CachedMacro {
@@ -100,6 +160,7 @@ private:
     Engines::Maxwell3D& maxwell3d;
     HLEFunction func;
 };
+
 } // Anonymous namespace
 
 HLEMacro::HLEMacro(Engines::Maxwell3D& maxwell3d_) : maxwell3d{maxwell3d_} {}