Reduce synchronization overhead and improve performance in DMA operations

This function is a copy-cat implementation of Ryujinx's method

Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/3217
Reviewed-by: Caio Oliveira <caiooliveirafarias0@gmail.com>
Reviewed-by: crueter <crueter@eden-emu.dev>
Co-authored-by: Maufeat <sahyno1996@gmail.com>
Co-committed-by: Maufeat <sahyno1996@gmail.com>

src/audio_core/sink/sink_stream.cpp

@@ -32,7 +32,6 @@ void SinkStream::AppendBuffer(SinkBuffer& buffer, std::span<s16> samples) {
     auto yuzu_volume = Settings::Volume();
     if (yuzu_volume > 1.0f)
         yuzu_volume = 0.6f + 20.0f * std::log10(yuzu_volume);
-    yuzu_volume = std::max(yuzu_volume, 0.001f);
     auto const volume = system_volume * device_volume * yuzu_volume;
 
     if (system_channels > device_channels) {

src/core/hle/api_version.h

@@ -15,8 +15,8 @@ namespace HLE::ApiVersion {
 // Horizon OS version constants.
 
 constexpr u8 HOS_VERSION_MAJOR = 21;
-constexpr u8 HOS_VERSION_MINOR = 0;
-constexpr u8 HOS_VERSION_MICRO = 1;
+constexpr u8 HOS_VERSION_MINOR = 1;
+constexpr u8 HOS_VERSION_MICRO = 0;
 
 // NintendoSDK version constants.
 
@@ -24,9 +24,9 @@ constexpr u8 SDK_REVISION_MAJOR = 1;
 constexpr u8 SDK_REVISION_MINOR = 0;
 
 constexpr char PLATFORM_STRING[] = "NX";
-constexpr char VERSION_HASH[] = "066a75e6fab7316de34b88b60d229a0b2729e421";
-constexpr char DISPLAY_VERSION[] = "21.0.1";
-constexpr char DISPLAY_TITLE[] = "NintendoSDK Firmware for NX 21.0.1-1.0";
+constexpr char VERSION_HASH[] = "0d7a6334ddc3d637f69ec3976b17a260efd68fd4";
+constexpr char DISPLAY_VERSION[] = "21.1.0";
+constexpr char DISPLAY_TITLE[] = "NintendoSDK Firmware for NX 21.1.0-1.0";
 
 // Atmosphere version constants.
 

src/core/hle/kernel/svc/svc_info.cpp

@@ -6,6 +6,7 @@
 
 #include "core/core.h"
 #include "core/core_timing.h"
+#include "core/hle/kernel/k_memory_manager.h"
 #include "core/hle/kernel/k_process.h"
 #include "core/hle/kernel/k_resource_limit.h"
 #include "core/hle/kernel/svc.h"
@@ -271,8 +272,47 @@ Result GetInfo(Core::System& system, u64* result, InfoType info_id_type, Handle
 
 Result GetSystemInfo(Core::System& system, uint64_t* out, SystemInfoType info_type, Handle handle,
                      uint64_t info_subtype) {
-    UNIMPLEMENTED();
-    R_THROW(ResultNotImplemented);
+    const u32 info_id = static_cast<u32>(info_type);
+
+    R_UNLESS(info_id <= 2, ResultInvalidEnumValue);
+
+    R_UNLESS(handle == 0, ResultInvalidHandle);
+
+    if (info_id < 2) {
+        R_UNLESS(info_subtype <= 3, ResultInvalidCombination);
+
+        auto& memory_manager = system.Kernel().MemoryManager();
+        const auto pool = static_cast<KMemoryManager::Pool>(info_subtype);
+
+        switch (info_type) {
+        case SystemInfoType::TotalPhysicalMemorySize:
+            *out = memory_manager.GetSize(pool);
+            break;
+        case SystemInfoType::UsedPhysicalMemorySize:
+            *out = memory_manager.GetSize(pool) - memory_manager.GetFreeSize(pool);
+            break;
+        default:
+            R_THROW(ResultInvalidEnumValue);
+        }
+    } else {
+        R_UNLESS(info_subtype <= 1, ResultInvalidCombination);
+
+        constexpr u64 PrivilegedProcessLowestId = 1;
+        constexpr u64 PrivilegedProcessHighestId = 8;
+
+        switch (info_subtype) {
+        case 0:
+            *out = PrivilegedProcessLowestId;
+            break;
+        case 1:
+            *out = PrivilegedProcessHighestId;
+            break;
+        default:
+            R_THROW(ResultInvalidCombination);
+        }
+    }
+
+    R_SUCCEED();
 }
 
 Result GetInfo64(Core::System& system, uint64_t* out, InfoType info_type, Handle handle,

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

@@ -91,6 +91,8 @@ public:
         static const FunctionInfo functions[] = {
             {0, &ILogger::Log, "Log"},
             {1, &ILogger::SetDestination, "SetDestination"},
+            {2, nullptr, "TransmitHashedLog"}, //20.0.0+
+            {3, nullptr, "DevNotify"}, //20.0.0+
         };
         RegisterHandlers(functions);
     }

src/video_core/buffer_cache/buffer_cache.h

@@ -1705,21 +1705,26 @@ Binding BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index,
         return NULL_BINDING;
     }
 
+    // xbzk: New size logic. Fixes MCI.
+    // If ever the * comment below prove wrong, the 'if' block may be removed.
     const auto size = [&]() {
         const bool is_nvn_cbuf = cbuf_index == 0;
-        // The NVN driver buffer (index 0) is known to pack the SSBO address followed by its size.
         if (is_nvn_cbuf) {
-            const u32 ssbo_size = gpu_memory->Read<u32>(ssbo_addr + 8);
-            if (ssbo_size != 0) {
-                return ssbo_size;
+            // * The NVN driver buffer (index 0) is known to pack the SSBO address followed by its size.
+            const u64 next_qword = gpu_memory->Read<u64>(ssbo_addr + 8);
+            const u32 upper_32 = static_cast<u32>(next_qword >> 32);
+            // Hardware-based detection: GPU addresses have non-zero upper bits
+            if (upper_32 == 0) {
+                // This is a size field, not a GPU address
+                return static_cast<u32>(next_qword);  // Return lower_32
             }
         }
-        // Other titles (notably Doom Eternal) may use STG/LDG on buffer addresses in custom defined
-        // cbufs, which do not store the sizes adjacent to the addresses, so use the fully
-        // mapped buffer size for now.
+        // Fall through: either not NVN cbuf (Doom Eternal & +), or NVN but ssbo_addr+8 is a GPU address (MCI)
         const u32 memory_layout_size = static_cast<u32>(gpu_memory->GetMemoryLayoutSize(gpu_addr));
+        // Cap at 8MB to prevent allocator overflow from misinterpreted addresses
         return (std::min)(memory_layout_size, static_cast<u32>(8_MiB));
     }();
+
     // Alignment only applies to the offset of the buffer
     const u32 alignment = runtime.GetStorageBufferAlignment();
     const GPUVAddr aligned_gpu_addr = Common::AlignDown(gpu_addr, alignment);

src/video_core/dma_pusher.cpp

@@ -43,90 +43,66 @@ void DmaPusher::DispatchCalls() {
 
 bool DmaPusher::Step() {
     if (!ib_enable || dma_pushbuffer.empty()) {
-        // pushbuffer empty and IB empty or nonexistent - nothing to do
         return false;
     }
 
-    CommandList& command_list{dma_pushbuffer.front()};
+    CommandList& command_list = dma_pushbuffer.front();
 
-    ASSERT_OR_EXECUTE(
-        command_list.command_lists.size() || command_list.prefetch_command_list.size(), {
-            // Somehow the command_list is empty, in order to avoid a crash
-            // We ignore it and assume its size is 0.
-            dma_pushbuffer.pop();
-            dma_pushbuffer_subindex = 0;
-            return true;
-        });
+    const size_t prefetch_size = command_list.prefetch_command_list.size();
+    const size_t command_list_size = command_list.command_lists.size();
 
-    if (command_list.prefetch_command_list.size()) {
-        // Prefetched command list from nvdrv, used for things like synchronization
-        ProcessCommands(VideoCommon::FixSmallVectorADL(command_list.prefetch_command_list));
+    if (prefetch_size == 0 && command_list_size == 0) {
         dma_pushbuffer.pop();
-    } else {
-        const CommandListHeader command_list_header{
-            command_list.command_lists[dma_pushbuffer_subindex++]};
+        dma_pushbuffer_subindex = 0;
+        return true;
+    }
 
-        if (signal_sync) {
-            std::unique_lock lk(sync_mutex);
-            sync_cv.wait(lk, [this]() { return synced; });
-            signal_sync = false;
-            synced = false;
-        }
+    if (prefetch_size > 0) {
+        ProcessCommands(command_list.prefetch_command_list);
+        dma_pushbuffer.pop();
+        return true;
+    }
 
-        dma_state.dma_get = command_list_header.addr;
+    auto& current_command = command_list.command_lists[dma_pushbuffer_subindex];
+    const CommandListHeader& header = current_command;
+    dma_state.dma_get = header.addr;
 
-        if (command_list_header.size == 0) {
-            return true;
-        }
+    if (signal_sync && !synced) {
+        std::unique_lock lk(sync_mutex);
+        sync_cv.wait(lk, [this]() { return synced; });
+        signal_sync = false;
+        synced = false;
+    }
 
-        // Push buffer non-empty, read a word
-        if (dma_state.method >= MacroRegistersStart) {
-            if (subchannels[dma_state.subchannel]) {
-                subchannels[dma_state.subchannel]->current_dirty = memory_manager.IsMemoryDirty(
-                    dma_state.dma_get, command_list_header.size * sizeof(u32));
-            }
-        }
+    if (header.size > 0 && dma_state.method >= MacroRegistersStart && subchannels[dma_state.subchannel]) {
+        subchannels[dma_state.subchannel]->current_dirty = memory_manager.IsMemoryDirty(dma_state.dma_get, header.size * sizeof(u32));
+    }
 
-        const auto safe_process = [&] {
-            Tegra::Memory::GpuGuestMemory<Tegra::CommandHeader,
-                                          Tegra::Memory::GuestMemoryFlags::SafeRead>
-                headers(memory_manager, dma_state.dma_get, command_list_header.size,
-                        &command_headers);
+    if (header.size > 0) {
+        if (Settings::IsDMALevelDefault() ? (Settings::IsGPULevelMedium() || Settings::IsGPULevelHigh()) : Settings::IsDMALevelSafe()) {
+            Tegra::Memory::GpuGuestMemory<Tegra::CommandHeader, Tegra::Memory::GuestMemoryFlags::SafeRead>headers(memory_manager, dma_state.dma_get, header.size, &command_headers);
             ProcessCommands(headers);
-        };
-
-        const auto unsafe_process = [&] {
-            Tegra::Memory::GpuGuestMemory<Tegra::CommandHeader,
-                                          Tegra::Memory::GuestMemoryFlags::UnsafeRead>
-                headers(memory_manager, dma_state.dma_get, command_list_header.size,
-                        &command_headers);
-            ProcessCommands(headers);
-        };
-
-        const bool use_safe = Settings::IsDMALevelDefault() ? (Settings::IsGPULevelMedium() || Settings::IsGPULevelHigh()) : Settings::IsDMALevelSafe();
-
-        if (use_safe) {
-            safe_process();
         } else {
-            unsafe_process();
+            Tegra::Memory::GpuGuestMemory<Tegra::CommandHeader, Tegra::Memory::GuestMemoryFlags::UnsafeRead>headers(memory_manager, dma_state.dma_get, header.size, &command_headers);
+            ProcessCommands(headers);
         }
+    }
 
-        if (dma_pushbuffer_subindex >= command_list.command_lists.size()) {
-            // We've gone through the current list, remove it from the queue
-            dma_pushbuffer.pop();
-            dma_pushbuffer_subindex = 0;
-        } else if (command_list.command_lists[dma_pushbuffer_subindex].sync && Settings::values.sync_memory_operations.GetValue()) {
-            signal_sync = true;
-        }
+    if (++dma_pushbuffer_subindex >= command_list_size) {
+        dma_pushbuffer.pop();
+        dma_pushbuffer_subindex = 0;
+    } else {
+        signal_sync = command_list.command_lists[dma_pushbuffer_subindex].sync && Settings::values.sync_memory_operations.GetValue();
+    }
 
-        if (signal_sync) {
-            rasterizer->SignalFence([this]() {
+    if (signal_sync) {
+        rasterizer->SignalFence([this]() {
             std::scoped_lock lk(sync_mutex);
             synced = true;
             sync_cv.notify_all();
-            });
-        }
+        });
     }
+
     return true;
 }