summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--src/common/CMakeLists.txt2
-rw-r--r--src/common/make_unique_for_overwrite.h25
-rw-r--r--src/common/scratch_buffer.h95
-rw-r--r--src/core/hle/kernel/k_shared_memory.cpp69
-rw-r--r--src/core/hle/kernel/k_shared_memory.h8
-rw-r--r--src/core/hle/kernel/kernel.cpp43
-rw-r--r--src/core/hle/service/time/clock_types.h1
-rw-r--r--src/core/hle/service/time/time_sharedmemory.cpp17
-rw-r--r--src/core/hle/service/time/time_sharedmemory.h87
-rw-r--r--src/tests/CMakeLists.txt1
-rw-r--r--src/tests/common/scratch_buffer.cpp199
-rw-r--r--src/video_core/buffer_cache/buffer_cache.h11
-rw-r--r--src/video_core/dma_pusher.cpp19
-rw-r--r--src/video_core/dma_pusher.h8
-rw-r--r--src/video_core/engines/engine_upload.cpp4
-rw-r--r--src/video_core/engines/engine_upload.h7
-rw-r--r--src/video_core/engines/maxwell_dma.cpp34
-rw-r--r--src/video_core/engines/maxwell_dma.h8
-rw-r--r--src/video_core/host1x/vic.cpp6
-rw-r--r--src/video_core/host1x/vic.h7
20 files changed, 497 insertions, 154 deletions
diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt
index 25b22a281..eb05e46a8 100644
--- a/src/common/CMakeLists.txt
+++ b/src/common/CMakeLists.txt
@@ -78,6 +78,7 @@ add_library(common STATIC
logging/types.h
lz4_compression.cpp
lz4_compression.h
+ make_unique_for_overwrite.h
math_util.h
memory_detect.cpp
memory_detect.h
@@ -101,6 +102,7 @@ add_library(common STATIC
${CMAKE_CURRENT_BINARY_DIR}/scm_rev.cpp
scm_rev.h
scope_exit.h
+ scratch_buffer.h
settings.cpp
settings.h
settings_input.cpp
diff --git a/src/common/make_unique_for_overwrite.h b/src/common/make_unique_for_overwrite.h
new file mode 100644
index 000000000..c7413cf51
--- /dev/null
+++ b/src/common/make_unique_for_overwrite.h
@@ -0,0 +1,25 @@
+// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <memory>
+#include <type_traits>
+
+namespace Common {
+
+template <class T>
+requires(!std::is_array_v<T>) std::unique_ptr<T> make_unique_for_overwrite() {
+ return std::unique_ptr<T>(new T);
+}
+
+template <class T>
+requires std::is_unbounded_array_v<T> std::unique_ptr<T> make_unique_for_overwrite(std::size_t n) {
+ return std::unique_ptr<T>(new std::remove_extent_t<T>[n]);
+}
+
+template <class T, class... Args>
+requires std::is_bounded_array_v<T>
+void make_unique_for_overwrite(Args&&...) = delete;
+
+} // namespace Common
diff --git a/src/common/scratch_buffer.h b/src/common/scratch_buffer.h
new file mode 100644
index 000000000..1245a5086
--- /dev/null
+++ b/src/common/scratch_buffer.h
@@ -0,0 +1,95 @@
+// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include "common/make_unique_for_overwrite.h"
+
+namespace Common {
+
+/**
+ * ScratchBuffer class
+ * This class creates a default initialized heap allocated buffer for cases such as intermediate
+ * buffers being copied into entirely, where value initializing members during allocation or resize
+ * is redundant.
+ */
+template <typename T>
+class ScratchBuffer {
+public:
+ ScratchBuffer() = default;
+
+ explicit ScratchBuffer(size_t initial_capacity)
+ : last_requested_size{initial_capacity}, buffer_capacity{initial_capacity},
+ buffer{Common::make_unique_for_overwrite<T[]>(initial_capacity)} {}
+
+ ~ScratchBuffer() = default;
+
+ /// This will only grow the buffer's capacity if size is greater than the current capacity.
+ /// The previously held data will remain intact.
+ void resize(size_t size) {
+ if (size > buffer_capacity) {
+ auto new_buffer = Common::make_unique_for_overwrite<T[]>(size);
+ std::move(buffer.get(), buffer.get() + buffer_capacity, new_buffer.get());
+ buffer = std::move(new_buffer);
+ buffer_capacity = size;
+ }
+ last_requested_size = size;
+ }
+
+ /// This will only grow the buffer's capacity if size is greater than the current capacity.
+ /// The previously held data will be destroyed if a reallocation occurs.
+ void resize_destructive(size_t size) {
+ if (size > buffer_capacity) {
+ buffer_capacity = size;
+ buffer = Common::make_unique_for_overwrite<T[]>(buffer_capacity);
+ }
+ last_requested_size = size;
+ }
+
+ [[nodiscard]] T* data() noexcept {
+ return buffer.get();
+ }
+
+ [[nodiscard]] const T* data() const noexcept {
+ return buffer.get();
+ }
+
+ [[nodiscard]] T* begin() noexcept {
+ return data();
+ }
+
+ [[nodiscard]] const T* begin() const noexcept {
+ return data();
+ }
+
+ [[nodiscard]] T* end() noexcept {
+ return data() + last_requested_size;
+ }
+
+ [[nodiscard]] const T* end() const noexcept {
+ return data() + last_requested_size;
+ }
+
+ [[nodiscard]] T& operator[](size_t i) {
+ return buffer[i];
+ }
+
+ [[nodiscard]] const T& operator[](size_t i) const {
+ return buffer[i];
+ }
+
+ [[nodiscard]] size_t size() const noexcept {
+ return last_requested_size;
+ }
+
+ [[nodiscard]] size_t capacity() const noexcept {
+ return buffer_capacity;
+ }
+
+private:
+ size_t last_requested_size{};
+ size_t buffer_capacity{};
+ std::unique_ptr<T[]> buffer{};
+};
+
+} // namespace Common
diff --git a/src/core/hle/kernel/k_shared_memory.cpp b/src/core/hle/kernel/k_shared_memory.cpp
index 10cd4c43d..0aa68103c 100644
--- a/src/core/hle/kernel/k_shared_memory.cpp
+++ b/src/core/hle/kernel/k_shared_memory.cpp
@@ -6,6 +6,7 @@
#include "core/hle/kernel/k_page_table.h"
#include "core/hle/kernel/k_scoped_resource_reservation.h"
#include "core/hle/kernel/k_shared_memory.h"
+#include "core/hle/kernel/k_system_resource.h"
#include "core/hle/kernel/kernel.h"
#include "core/hle/kernel/svc_results.h"
@@ -18,19 +19,19 @@ KSharedMemory::~KSharedMemory() {
}
Result KSharedMemory::Initialize(Core::DeviceMemory& device_memory_, KProcess* owner_process_,
- KPageGroup&& page_list_, Svc::MemoryPermission owner_permission_,
- Svc::MemoryPermission user_permission_, PAddr physical_address_,
- std::size_t size_, std::string name_) {
+ Svc::MemoryPermission owner_permission_,
+ Svc::MemoryPermission user_permission_, std::size_t size_,
+ std::string name_) {
// Set members.
owner_process = owner_process_;
device_memory = &device_memory_;
- page_list = std::move(page_list_);
owner_permission = owner_permission_;
user_permission = user_permission_;
- physical_address = physical_address_;
- size = size_;
+ size = Common::AlignUp(size_, PageSize);
name = std::move(name_);
+ const size_t num_pages = Common::DivideUp(size, PageSize);
+
// Get the resource limit.
KResourceLimit* reslimit = kernel.GetSystemResourceLimit();
@@ -39,6 +40,17 @@ Result KSharedMemory::Initialize(Core::DeviceMemory& device_memory_, KProcess* o
size_);
R_UNLESS(memory_reservation.Succeeded(), ResultLimitReached);
+ // Allocate the memory.
+
+ //! HACK: Open continuous mapping from sysmodule pool.
+ auto option = KMemoryManager::EncodeOption(KMemoryManager::Pool::Secure,
+ KMemoryManager::Direction::FromBack);
+ physical_address = kernel.MemoryManager().AllocateAndOpenContinuous(num_pages, 1, option);
+ R_UNLESS(physical_address != 0, ResultOutOfMemory);
+
+ //! Insert the result into our page group.
+ page_group.emplace(physical_address, num_pages);
+
// Commit our reservation.
memory_reservation.Commit();
@@ -50,12 +62,23 @@ Result KSharedMemory::Initialize(Core::DeviceMemory& device_memory_, KProcess* o
is_initialized = true;
// Clear all pages in the memory.
- std::memset(device_memory_.GetPointer<void>(physical_address_), 0, size_);
+ for (const auto& block : page_group->Nodes()) {
+ std::memset(device_memory_.GetPointer<void>(block.GetAddress()), 0, block.GetSize());
+ }
return ResultSuccess;
}
void KSharedMemory::Finalize() {
+ // Close and finalize the page group.
+ // page_group->Close();
+ // page_group->Finalize();
+
+ //! HACK: Manually close.
+ for (const auto& block : page_group->Nodes()) {
+ kernel.MemoryManager().Close(block.GetAddress(), block.GetNumPages());
+ }
+
// Release the memory reservation.
resource_limit->Release(LimitableResource::PhysicalMemoryMax, size);
resource_limit->Close();
@@ -65,32 +88,28 @@ void KSharedMemory::Finalize() {
}
Result KSharedMemory::Map(KProcess& target_process, VAddr address, std::size_t map_size,
- Svc::MemoryPermission permissions) {
- const u64 page_count{(map_size + PageSize - 1) / PageSize};
-
- if (page_list.GetNumPages() != page_count) {
- UNIMPLEMENTED_MSG("Page count does not match");
- }
+ Svc::MemoryPermission map_perm) {
+ // Validate the size.
+ R_UNLESS(size == map_size, ResultInvalidSize);
- const Svc::MemoryPermission expected =
+ // Validate the permission.
+ const Svc::MemoryPermission test_perm =
&target_process == owner_process ? owner_permission : user_permission;
-
- if (permissions != expected) {
- UNIMPLEMENTED_MSG("Permission does not match");
+ if (test_perm == Svc::MemoryPermission::DontCare) {
+ ASSERT(map_perm == Svc::MemoryPermission::Read || map_perm == Svc::MemoryPermission::Write);
+ } else {
+ R_UNLESS(map_perm == test_perm, ResultInvalidNewMemoryPermission);
}
- return target_process.PageTable().MapPages(address, page_list, KMemoryState::Shared,
- ConvertToKMemoryPermission(permissions));
+ return target_process.PageTable().MapPages(address, *page_group, KMemoryState::Shared,
+ ConvertToKMemoryPermission(map_perm));
}
Result KSharedMemory::Unmap(KProcess& target_process, VAddr address, std::size_t unmap_size) {
- const u64 page_count{(unmap_size + PageSize - 1) / PageSize};
-
- if (page_list.GetNumPages() != page_count) {
- UNIMPLEMENTED_MSG("Page count does not match");
- }
+ // Validate the size.
+ R_UNLESS(size == unmap_size, ResultInvalidSize);
- return target_process.PageTable().UnmapPages(address, page_list, KMemoryState::Shared);
+ return target_process.PageTable().UnmapPages(address, *page_group, KMemoryState::Shared);
}
} // namespace Kernel
diff --git a/src/core/hle/kernel/k_shared_memory.h b/src/core/hle/kernel/k_shared_memory.h
index a96c55a3e..8b29f0b4a 100644
--- a/src/core/hle/kernel/k_shared_memory.h
+++ b/src/core/hle/kernel/k_shared_memory.h
@@ -3,6 +3,7 @@
#pragma once
+#include <optional>
#include <string>
#include "common/common_types.h"
@@ -26,9 +27,8 @@ public:
~KSharedMemory() override;
Result Initialize(Core::DeviceMemory& device_memory_, KProcess* owner_process_,
- KPageGroup&& page_list_, Svc::MemoryPermission owner_permission_,
- Svc::MemoryPermission user_permission_, PAddr physical_address_,
- std::size_t size_, std::string name_);
+ Svc::MemoryPermission owner_permission_,
+ Svc::MemoryPermission user_permission_, std::size_t size_, std::string name_);
/**
* Maps a shared memory block to an address in the target process' address space
@@ -76,7 +76,7 @@ public:
private:
Core::DeviceMemory* device_memory{};
KProcess* owner_process{};
- KPageGroup page_list;
+ std::optional<KPageGroup> page_group{};
Svc::MemoryPermission owner_permission{};
Svc::MemoryPermission user_permission{};
PAddr physical_address{};
diff --git a/src/core/hle/kernel/kernel.cpp b/src/core/hle/kernel/kernel.cpp
index b75bac5df..1fb25f221 100644
--- a/src/core/hle/kernel/kernel.cpp
+++ b/src/core/hle/kernel/kernel.cpp
@@ -94,6 +94,7 @@ struct KernelCore::Impl {
pt_heap_region.GetSize());
}
+ InitializeHackSharedMemory();
RegisterHostThread(nullptr);
default_service_thread = &CreateServiceThread(kernel, "DefaultServiceThread");
@@ -726,14 +727,14 @@ struct KernelCore::Impl {
}
void InitializeMemoryLayout() {
- const auto system_pool = memory_layout->GetKernelSystemPoolRegionPhysicalExtents();
-
// Initialize the memory manager.
memory_manager = std::make_unique<KMemoryManager>(system);
const auto& management_region = memory_layout->GetPoolManagementRegion();
ASSERT(management_region.GetEndAddress() != 0);
memory_manager->Initialize(management_region.GetAddress(), management_region.GetSize());
+ }
+ void InitializeHackSharedMemory() {
// Setup memory regions for emulated processes
// TODO(bunnei): These should not be hardcoded regions initialized within the kernel
constexpr std::size_t hid_size{0x40000};
@@ -742,39 +743,23 @@ struct KernelCore::Impl {
constexpr std::size_t time_size{0x1000};
constexpr std::size_t hidbus_size{0x1000};
- const PAddr hid_phys_addr{system_pool.GetAddress()};
- const PAddr font_phys_addr{system_pool.GetAddress() + hid_size};
- const PAddr irs_phys_addr{system_pool.GetAddress() + hid_size + font_size};
- const PAddr time_phys_addr{system_pool.GetAddress() + hid_size + font_size + irs_size};
- const PAddr hidbus_phys_addr{system_pool.GetAddress() + hid_size + font_size + irs_size +
- time_size};
-
hid_shared_mem = KSharedMemory::Create(system.Kernel());
font_shared_mem = KSharedMemory::Create(system.Kernel());
irs_shared_mem = KSharedMemory::Create(system.Kernel());
time_shared_mem = KSharedMemory::Create(system.Kernel());
hidbus_shared_mem = KSharedMemory::Create(system.Kernel());
- hid_shared_mem->Initialize(system.DeviceMemory(), nullptr,
- {hid_phys_addr, hid_size / PageSize},
- Svc::MemoryPermission::None, Svc::MemoryPermission::Read,
- hid_phys_addr, hid_size, "HID:SharedMemory");
- font_shared_mem->Initialize(system.DeviceMemory(), nullptr,
- {font_phys_addr, font_size / PageSize},
- Svc::MemoryPermission::None, Svc::MemoryPermission::Read,
- font_phys_addr, font_size, "Font:SharedMemory");
- irs_shared_mem->Initialize(system.DeviceMemory(), nullptr,
- {irs_phys_addr, irs_size / PageSize},
- Svc::MemoryPermission::None, Svc::MemoryPermission::Read,
- irs_phys_addr, irs_size, "IRS:SharedMemory");
- time_shared_mem->Initialize(system.DeviceMemory(), nullptr,
- {time_phys_addr, time_size / PageSize},
- Svc::MemoryPermission::None, Svc::MemoryPermission::Read,
- time_phys_addr, time_size, "Time:SharedMemory");
- hidbus_shared_mem->Initialize(system.DeviceMemory(), nullptr,
- {hidbus_phys_addr, hidbus_size / PageSize},
- Svc::MemoryPermission::None, Svc::MemoryPermission::Read,
- hidbus_phys_addr, hidbus_size, "HidBus:SharedMemory");
+ hid_shared_mem->Initialize(system.DeviceMemory(), nullptr, Svc::MemoryPermission::None,
+ Svc::MemoryPermission::Read, hid_size, "HID:SharedMemory");
+ font_shared_mem->Initialize(system.DeviceMemory(), nullptr, Svc::MemoryPermission::None,
+ Svc::MemoryPermission::Read, font_size, "Font:SharedMemory");
+ irs_shared_mem->Initialize(system.DeviceMemory(), nullptr, Svc::MemoryPermission::None,
+ Svc::MemoryPermission::Read, irs_size, "IRS:SharedMemory");
+ time_shared_mem->Initialize(system.DeviceMemory(), nullptr, Svc::MemoryPermission::None,
+ Svc::MemoryPermission::Read, time_size, "Time:SharedMemory");
+ hidbus_shared_mem->Initialize(system.DeviceMemory(), nullptr, Svc::MemoryPermission::None,
+ Svc::MemoryPermission::Read, hidbus_size,
+ "HidBus:SharedMemory");
}
KClientPort* CreateNamedServicePort(std::string name) {
diff --git a/src/core/hle/service/time/clock_types.h b/src/core/hle/service/time/clock_types.h
index ef070f32f..ed1eb5b2d 100644
--- a/src/core/hle/service/time/clock_types.h
+++ b/src/core/hle/service/time/clock_types.h
@@ -49,6 +49,7 @@ struct SteadyClockContext {
static_assert(sizeof(SteadyClockContext) == 0x18, "SteadyClockContext is incorrect size");
static_assert(std::is_trivially_copyable_v<SteadyClockContext>,
"SteadyClockContext must be trivially copyable");
+using StandardSteadyClockTimePointType = SteadyClockContext;
struct SystemClockContext {
s64 offset;
diff --git a/src/core/hle/service/time/time_sharedmemory.cpp b/src/core/hle/service/time/time_sharedmemory.cpp
index a3aa0e77f..ff53a7d6f 100644
--- a/src/core/hle/service/time/time_sharedmemory.cpp
+++ b/src/core/hle/service/time/time_sharedmemory.cpp
@@ -26,23 +26,24 @@ void SharedMemory::SetupStandardSteadyClock(const Common::UUID& clock_source_id,
const Clock::SteadyClockContext context{
static_cast<u64>(current_time_point.nanoseconds - ticks_time_span.nanoseconds),
clock_source_id};
- shared_memory_format.standard_steady_clock_timepoint.StoreData(
- system.Kernel().GetTimeSharedMem().GetPointer(), context);
+ StoreToLockFreeAtomicType(&GetFormat()->standard_steady_clock_timepoint, context);
}
void SharedMemory::UpdateLocalSystemClockContext(const Clock::SystemClockContext& context) {
- shared_memory_format.standard_local_system_clock_context.StoreData(
- system.Kernel().GetTimeSharedMem().GetPointer(), context);
+ StoreToLockFreeAtomicType(&GetFormat()->standard_local_system_clock_context, context);
}
void SharedMemory::UpdateNetworkSystemClockContext(const Clock::SystemClockContext& context) {
- shared_memory_format.standard_network_system_clock_context.StoreData(
- system.Kernel().GetTimeSharedMem().GetPointer(), context);
+ StoreToLockFreeAtomicType(&GetFormat()->standard_network_system_clock_context, context);
}
void SharedMemory::SetAutomaticCorrectionEnabled(bool is_enabled) {
- shared_memory_format.standard_user_system_clock_automatic_correction.StoreData(
- system.Kernel().GetTimeSharedMem().GetPointer(), is_enabled);
+ StoreToLockFreeAtomicType(
+ &GetFormat()->is_standard_user_system_clock_automatic_correction_enabled, is_enabled);
+}
+
+SharedMemory::Format* SharedMemory::GetFormat() {
+ return reinterpret_cast<SharedMemory::Format*>(system.Kernel().GetTimeSharedMem().GetPointer());
}
} // namespace Service::Time
diff --git a/src/core/hle/service/time/time_sharedmemory.h b/src/core/hle/service/time/time_sharedmemory.h
index 561685acd..044a4d24e 100644
--- a/src/core/hle/service/time/time_sharedmemory.h
+++ b/src/core/hle/service/time/time_sharedmemory.h
@@ -10,45 +10,68 @@
namespace Service::Time {
+// Note: this type is not safe for concurrent writes.
+template <typename T>
+struct LockFreeAtomicType {
+ u32 counter_;
+ std::array<T, 2> value_;
+};
+
+template <typename T>
+static inline void StoreToLockFreeAtomicType(LockFreeAtomicType<T>* p, const T& value) {
+ // Get the current counter.
+ auto counter = p->counter_;
+
+ // Increment the counter.
+ ++counter;
+
+ // Store the updated value.
+ p->value_[counter % 2] = value;
+
+ // Fence memory.
+ std::atomic_thread_fence(std::memory_order_release);
+
+ // Set the updated counter.
+ p->counter_ = counter;
+}
+
+template <typename T>
+static inline T LoadFromLockFreeAtomicType(const LockFreeAtomicType<T>* p) {
+ while (true) {
+ // Get the counter.
+ auto counter = p->counter_;
+
+ // Get the value.
+ auto value = p->value_[counter % 2];
+
+ // Fence memory.
+ std::atomic_thread_fence(std::memory_order_acquire);
+
+ // Check that the counter matches.
+ if (counter == p->counter_) {
+ return value;
+ }
+ }
+}
+
class SharedMemory final {
public:
explicit SharedMemory(Core::System& system_);
~SharedMemory();
- // TODO(ogniK): We have to properly simulate memory barriers, how are we going to do this?
- template <typename T, std::size_t Offset>
- struct MemoryBarrier {
- static_assert(std::is_trivially_copyable_v<T>, "T must be trivially copyable");
- u32_le read_attempt{};
- std::array<T, 2> data{};
-
- // These are not actually memory barriers at the moment as we don't have multicore and all
- // HLE is mutexed. This will need to properly be implemented when we start updating the time
- // points on threads. As of right now, we'll be updated both values synchronously and just
- // incrementing the read_attempt to indicate that we waited.
- void StoreData(u8* shared_memory, T data_to_store) {
- std::memcpy(this, shared_memory + Offset, sizeof(*this));
- read_attempt++;
- data[read_attempt & 1] = data_to_store;
- std::memcpy(shared_memory + Offset, this, sizeof(*this));
- }
-
- // For reading we're just going to read the last stored value. If there was no value stored
- // it will just end up reading an empty value as intended.
- T ReadData(u8* shared_memory) {
- std::memcpy(this, shared_memory + Offset, sizeof(*this));
- return data[(read_attempt - 1) & 1];
- }
- };
-
// Shared memory format
struct Format {
- MemoryBarrier<Clock::SteadyClockContext, 0x0> standard_steady_clock_timepoint;
- MemoryBarrier<Clock::SystemClockContext, 0x38> standard_local_system_clock_context;
- MemoryBarrier<Clock::SystemClockContext, 0x80> standard_network_system_clock_context;
- MemoryBarrier<bool, 0xc8> standard_user_system_clock_automatic_correction;
- u32_le format_version;
+ LockFreeAtomicType<Clock::StandardSteadyClockTimePointType> standard_steady_clock_timepoint;
+ LockFreeAtomicType<Clock::SystemClockContext> standard_local_system_clock_context;
+ LockFreeAtomicType<Clock::SystemClockContext> standard_network_system_clock_context;
+ LockFreeAtomicType<bool> is_standard_user_system_clock_automatic_correction_enabled;
+ u32 format_version;
};
+ static_assert(offsetof(Format, standard_steady_clock_timepoint) == 0x0);
+ static_assert(offsetof(Format, standard_local_system_clock_context) == 0x38);
+ static_assert(offsetof(Format, standard_network_system_clock_context) == 0x80);
+ static_assert(offsetof(Format, is_standard_user_system_clock_automatic_correction_enabled) ==
+ 0xc8);
static_assert(sizeof(Format) == 0xd8, "Format is an invalid size");
void SetupStandardSteadyClock(const Common::UUID& clock_source_id,
@@ -56,10 +79,10 @@ public:
void UpdateLocalSystemClockContext(const Clock::SystemClockContext& context);
void UpdateNetworkSystemClockContext(const Clock::SystemClockContext& context);
void SetAutomaticCorrectionEnabled(bool is_enabled);
+ Format* GetFormat();
private:
Core::System& system;
- Format shared_memory_format{};
};
} // namespace Service::Time
diff --git a/src/tests/CMakeLists.txt b/src/tests/CMakeLists.txt
index 348d1edf4..6a4022e45 100644
--- a/src/tests/CMakeLists.txt
+++ b/src/tests/CMakeLists.txt
@@ -8,6 +8,7 @@ add_executable(tests
common/host_memory.cpp
common/param_package.cpp
common/ring_buffer.cpp
+ common/scratch_buffer.cpp
common/unique_function.cpp
core/core_timing.cpp
core/internal_network/network.cpp
diff --git a/src/tests/common/scratch_buffer.cpp b/src/tests/common/scratch_buffer.cpp
new file mode 100644
index 000000000..b602c8d0a
--- /dev/null
+++ b/src/tests/common/scratch_buffer.cpp
@@ -0,0 +1,199 @@
+// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <algorithm>
+#include <array>
+#include <span>
+#include <catch2/catch.hpp>
+#include "common/common_types.h"
+#include "common/scratch_buffer.h"
+
+namespace Common {
+
+TEST_CASE("ScratchBuffer: Basic Test", "[common]") {
+ ScratchBuffer<u8> buf;
+
+ REQUIRE(buf.size() == 0U);
+ REQUIRE(buf.capacity() == 0U);
+
+ std::array<u8, 10> payload;
+ payload.fill(66);
+
+ buf.resize(payload.size());
+ REQUIRE(buf.size() == payload.size());
+ REQUIRE(buf.capacity() == payload.size());
+
+ std::memcpy(buf.data(), payload.data(), payload.size());
+ for (size_t i = 0; i < payload.size(); ++i) {
+ REQUIRE(buf[i] == payload[i]);
+ }
+}
+
+TEST_CASE("ScratchBuffer: resize_destructive Grow", "[common]") {
+ std::array<u8, 10> payload;
+ payload.fill(66);
+
+ ScratchBuffer<u8> buf(payload.size());
+ REQUIRE(buf.size() == payload.size());
+ REQUIRE(buf.capacity() == payload.size());
+
+ // Increasing the size should reallocate the buffer
+ buf.resize_destructive(payload.size() * 2);
+ REQUIRE(buf.size() == payload.size() * 2);
+ REQUIRE(buf.capacity() == payload.size() * 2);
+
+ // Since the buffer is not value initialized, reading its data will be garbage
+}
+
+TEST_CASE("ScratchBuffer: resize_destructive Shrink", "[common]") {
+ std::array<u8, 10> payload;
+ payload.fill(66);
+
+ ScratchBuffer<u8> buf(payload.size());
+ REQUIRE(buf.size() == payload.size());
+ REQUIRE(buf.capacity() == payload.size());
+
+ std::memcpy(buf.data(), payload.data(), payload.size());
+ for (size_t i = 0; i < payload.size(); ++i) {
+ REQUIRE(buf[i] == payload[i]);
+ }
+
+ // Decreasing the size should not cause a buffer reallocation
+ // This can be tested by ensuring the buffer capacity and data has not changed,
+ buf.resize_destructive(1U);
+ REQUIRE(buf.size() == 1U);
+ REQUIRE(buf.capacity() == payload.size());
+
+ for (size_t i = 0; i < payload.size(); ++i) {
+ REQUIRE(buf[i] == payload[i]);
+ }
+}
+
+TEST_CASE("ScratchBuffer: resize Grow u8", "[common]") {
+ std::array<u8, 10> payload;
+ payload.fill(66);
+
+ ScratchBuffer<u8> buf(payload.size());
+ REQUIRE(buf.size() == payload.size());
+ REQUIRE(buf.capacity() == payload.size());
+
+ std::memcpy(buf.data(), payload.data(), payload.size());
+ for (size_t i = 0; i < payload.size(); ++i) {
+ REQUIRE(buf[i] == payload[i]);
+ }
+
+ // Increasing the size should reallocate the buffer
+ buf.resize(payload.size() * 2);
+ REQUIRE(buf.size() == payload.size() * 2);
+ REQUIRE(buf.capacity() == payload.size() * 2);
+
+ // resize() keeps the previous data intact
+ for (size_t i = 0; i < payload.size(); ++i) {
+ REQUIRE(buf[i] == payload[i]);
+ }
+}
+
+TEST_CASE("ScratchBuffer: resize Grow u64", "[common]") {
+ std::array<u64, 10> payload;
+ payload.fill(6666);
+
+ ScratchBuffer<u64> buf(payload.size());
+ REQUIRE(buf.size() == payload.size());
+ REQUIRE(buf.capacity() == payload.size());
+
+ std::memcpy(buf.data(), payload.data(), payload.size() * sizeof(u64));
+ for (size_t i = 0; i < payload.size(); ++i) {
+ REQUIRE(buf[i] == payload[i]);
+ }
+
+ // Increasing the size should reallocate the buffer
+ buf.resize(payload.size() * 2);
+ REQUIRE(buf.size() == payload.size() * 2);
+ REQUIRE(buf.capacity() == payload.size() * 2);
+
+ // resize() keeps the previous data intact
+ for (size_t i = 0; i < payload.size(); ++i) {
+ REQUIRE(buf[i] == payload[i]);
+ }
+}
+
+TEST_CASE("ScratchBuffer: resize Shrink", "[common]") {
+ std::array<u8, 10> payload;
+ payload.fill(66);
+
+ ScratchBuffer<u8> buf(payload.size());
+ REQUIRE(buf.size() == payload.size());
+ REQUIRE(buf.capacity() == payload.size());
+
+ std::memcpy(buf.data(), payload.data(), payload.size());
+ for (size_t i = 0; i < payload.size(); ++i) {
+ REQUIRE(buf[i] == payload[i]);
+ }
+
+ // Decreasing the size should not cause a buffer reallocation
+ // This can be tested by ensuring the buffer capacity and data has not changed,
+ buf.resize(1U);
+ REQUIRE(buf.size() == 1U);
+ REQUIRE(buf.capacity() == payload.size());
+
+ for (size_t i = 0; i < payload.size(); ++i) {
+ REQUIRE(buf[i] == payload[i]);
+ }
+}
+
+TEST_CASE("ScratchBuffer: Span Size", "[common]") {
+ std::array<u8, 10> payload;
+ payload.fill(66);
+
+ ScratchBuffer<u8> buf(payload.size());
+ REQUIRE(buf.size() == payload.size());
+ REQUIRE(buf.capacity() == payload.size());
+
+ std::memcpy(buf.data(), payload.data(), payload.size());
+ for (size_t i = 0; i < payload.size(); ++i) {
+ REQUIRE(buf[i] == payload[i]);
+ }
+
+ buf.resize(3U);
+ REQUIRE(buf.size() == 3U);
+ REQUIRE(buf.capacity() == payload.size());
+
+ const auto buf_span = std::span<u8>(buf);
+ // The span size is the last requested size of the buffer, not its capacity
+ REQUIRE(buf_span.size() == buf.size());
+
+ for (size_t i = 0; i < buf_span.size(); ++i) {
+ REQUIRE(buf_span[i] == buf[i]);
+ REQUIRE(buf_span[i] == payload[i]);
+ }
+}
+
+TEST_CASE("ScratchBuffer: Span Writes", "[common]") {
+ std::array<u8, 10> payload;
+ payload.fill(66);
+
+ ScratchBuffer<u8> buf(payload.size());
+ REQUIRE(buf.size() == payload.size());
+ REQUIRE(buf.capacity() == payload.size());
+
+ std::memcpy(buf.data(), payload.data(), payload.size());
+ for (size_t i = 0; i < payload.size(); ++i) {
+ REQUIRE(buf[i] == payload[i]);
+ }
+
+ buf.resize(3U);
+ REQUIRE(buf.size() == 3U);
+ REQUIRE(buf.capacity() == payload.size());
+
+ const auto buf_span = std::span<u8>(buf);
+ REQUIRE(buf_span.size() == buf.size());
+
+ for (size_t i = 0; i < buf_span.size(); ++i) {
+ const auto new_value = static_cast<u8>(i + 1U);
+ // Writes to a span of the scratch buffer will propogate to the buffer itself
+ buf_span[i] = new_value;
+ REQUIRE(buf[i] == new_value);
+ }
+}
+
+} // namespace Common
diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
index 502b4d90a..6c8d98946 100644
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -20,6 +20,7 @@
#include "common/lru_cache.h"
#include "common/microprofile.h"
#include "common/polyfill_ranges.h"
+#include "common/scratch_buffer.h"
#include "common/settings.h"
#include "core/memory.h"
#include "video_core/buffer_cache/buffer_base.h"
@@ -422,8 +423,7 @@ private:
IntervalSet common_ranges;
std::deque<IntervalSet> committed_ranges;
- size_t immediate_buffer_capacity = 0;
- std::unique_ptr<u8[]> immediate_buffer_alloc;
+ Common::ScratchBuffer<u8> immediate_buffer_alloc;
struct LRUItemParams {
using ObjectType = BufferId;
@@ -1926,11 +1926,8 @@ std::span<const u8> BufferCache<P>::ImmediateBufferWithData(VAddr cpu_addr, size
template <class P>
std::span<u8> BufferCache<P>::ImmediateBuffer(size_t wanted_capacity) {
- if (wanted_capacity > immediate_buffer_capacity) {
- immediate_buffer_capacity = wanted_capacity;
- immediate_buffer_alloc = std::make_unique<u8[]>(wanted_capacity);
- }
- return std::span<u8>(immediate_buffer_alloc.get(), wanted_capacity);
+ immediate_buffer_alloc.resize_destructive(wanted_capacity);
+ return std::span<u8>(immediate_buffer_alloc.data(), wanted_capacity);
}
template <class P>
diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp
index 9835e3ac1..322de2606 100644
--- a/src/video_core/dma_pusher.cpp
+++ b/src/video_core/dma_pusher.cpp
@@ -56,7 +56,7 @@ bool DmaPusher::Step() {
if (command_list.prefetch_command_list.size()) {
// Prefetched command list from nvdrv, used for things like synchronization
- command_headers = std::move(command_list.prefetch_command_list);
+ ProcessCommands(command_list.prefetch_command_list);
dma_pushbuffer.pop();
} else {
const CommandListHeader command_list_header{
@@ -74,7 +74,7 @@ bool DmaPusher::Step() {
}
// Push buffer non-empty, read a word
- command_headers.resize(command_list_header.size);
+ command_headers.resize_destructive(command_list_header.size);
if (Settings::IsGPULevelHigh()) {
memory_manager.ReadBlock(dma_get, command_headers.data(),
command_list_header.size * sizeof(u32));
@@ -82,16 +82,21 @@ bool DmaPusher::Step() {
memory_manager.ReadBlockUnsafe(dma_get, command_headers.data(),
command_list_header.size * sizeof(u32));
}
+ ProcessCommands(command_headers);
}
- for (std::size_t index = 0; index < command_headers.size();) {
- const CommandHeader& command_header = command_headers[index];
+
+ return true;
+}
+
+void DmaPusher::ProcessCommands(std::span<const CommandHeader> commands) {
+ for (std::size_t index = 0; index < commands.size();) {
+ const CommandHeader& command_header = commands[index];
if (dma_state.method_count) {
// Data word of methods command
if (dma_state.non_incrementing) {
const u32 max_write = static_cast<u32>(
- std::min<std::size_t>(index + dma_state.method_count, command_headers.size()) -
- index);
+ std::min<std::size_t>(index + dma_state.method_count, commands.size()) - index);
CallMultiMethod(&command_header.argument, max_write);
dma_state.method_count -= max_write;
dma_state.is_last_call = true;
@@ -142,8 +147,6 @@ bool DmaPusher::Step() {
}
index++;
}
-
- return true;
}
void DmaPusher::SetState(const CommandHeader& command_header) {
diff --git a/src/video_core/dma_pusher.h b/src/video_core/dma_pusher.h
index 938f0f11c..6f00de937 100644
--- a/src/video_core/dma_pusher.h
+++ b/src/video_core/dma_pusher.h
@@ -4,11 +4,13 @@
#pragma once
#include <array>
+#include <span>
#include <vector>
#include <queue>
#include "common/bit_field.h"
#include "common/common_types.h"
+#include "common/scratch_buffer.h"
#include "video_core/engines/engine_interface.h"
#include "video_core/engines/puller.h"
@@ -136,13 +138,15 @@ private:
static constexpr u32 non_puller_methods = 0x40;
static constexpr u32 max_subchannels = 8;
bool Step();
+ void ProcessCommands(std::span<const CommandHeader> commands);
void SetState(const CommandHeader& command_header);
void CallMethod(u32 argument) const;
void CallMultiMethod(const u32* base_start, u32 num_methods) const;
- std::vector<CommandHeader> command_headers; ///< Buffer for list of commands fetched at once
+ Common::ScratchBuffer<CommandHeader>
+ command_headers; ///< Buffer for list of commands fetched at once
std::queue<CommandList> dma_pushbuffer; ///< Queue of command lists to be processed
std::size_t dma_pushbuffer_subindex{}; ///< Index within a command list within the pushbuffer
@@ -159,7 +163,7 @@ private:
DmaState dma_state{};
bool dma_increment_once{};
- bool ib_enable{true}; ///< IB mode enabled
+ const bool ib_enable{true}; ///< IB mode enabled
std::array<Engines::EngineInterface*, max_subchannels> subchannels{};
diff --git a/src/video_core/engines/engine_upload.cpp b/src/video_core/engines/engine_upload.cpp
index e4f8331ab..cea1dd8b0 100644
--- a/src/video_core/engines/engine_upload.cpp
+++ b/src/video_core/engines/engine_upload.cpp
@@ -24,7 +24,7 @@ void State::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) {
void State::ProcessExec(const bool is_linear_) {
write_offset = 0;
copy_size = regs.line_length_in * regs.line_count;
- inner_buffer.resize(copy_size);
+ inner_buffer.resize_destructive(copy_size);
is_linear = is_linear_;
}
@@ -70,7 +70,7 @@ void State::ProcessData(std::span<const u8> read_buffer) {
const std::size_t dst_size = Tegra::Texture::CalculateSize(
true, bytes_per_pixel, width, regs.dest.height, regs.dest.depth,
regs.dest.BlockHeight(), regs.dest.BlockDepth());
- tmp_buffer.resize(dst_size);
+ tmp_buffer.resize_destructive(dst_size);
memory_manager.ReadBlock(address, tmp_buffer.data(), dst_size);
Tegra::Texture::SwizzleSubrect(tmp_buffer, read_buffer, bytes_per_pixel, width,
regs.dest.height, regs.dest.depth, x_offset, regs.dest.y,
diff --git a/src/video_core/engines/engine_upload.h b/src/video_core/engines/engine_upload.h
index 94fafd9dc..7242d2529 100644
--- a/src/video_core/engines/engine_upload.h
+++ b/src/video_core/engines/engine_upload.h
@@ -4,9 +4,10 @@
#pragma once
#include <span>
-#include <vector>
+
#include "common/bit_field.h"
#include "common/common_types.h"
+#include "common/scratch_buffer.h"
namespace Tegra {
class MemoryManager;
@@ -73,8 +74,8 @@ private:
u32 write_offset = 0;
u32 copy_size = 0;
- std::vector<u8> inner_buffer;
- std::vector<u8> tmp_buffer;
+ Common::ScratchBuffer<u8> inner_buffer;
+ Common::ScratchBuffer<u8> tmp_buffer;
bool is_linear = false;
Registers& regs;
MemoryManager& memory_manager;
diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp
index a189e60ae..f73d7bf0f 100644
--- a/src/video_core/engines/maxwell_dma.cpp
+++ b/src/video_core/engines/maxwell_dma.cpp
@@ -184,12 +184,8 @@ void MaxwellDMA::CopyBlockLinearToPitch() {
const size_t src_size =
CalculateSize(true, bytes_per_pixel, width, height, depth, block_height, block_depth);
- if (read_buffer.size() < src_size) {
- read_buffer.resize(src_size);
- }
- if (write_buffer.size() < dst_size) {
- write_buffer.resize(dst_size);
- }
+ read_buffer.resize_destructive(src_size);
+ write_buffer.resize_destructive(dst_size);
memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size);
memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size);
@@ -235,12 +231,8 @@ void MaxwellDMA::CopyPitchToBlockLinear() {
CalculateSize(true, bytes_per_pixel, width, height, depth, block_height, block_depth);
const size_t src_size = static_cast<size_t>(regs.pitch_in) * regs.line_count;
- if (read_buffer.size() < src_size) {
- read_buffer.resize(src_size);
- }
- if (write_buffer.size() < dst_size) {
- write_buffer.resize(dst_size);
- }
+ read_buffer.resize_destructive(src_size);
+ write_buffer.resize_destructive(dst_size);
memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size);
if (Settings::IsGPULevelExtreme()) {
@@ -269,12 +261,8 @@ void MaxwellDMA::FastCopyBlockLinearToPitch() {
pos_x = pos_x % x_in_gob;
pos_y = pos_y % 8;
- if (read_buffer.size() < src_size) {
- read_buffer.resize(src_size);
- }
- if (write_buffer.size() < dst_size) {
- write_buffer.resize(dst_size);
- }
+ read_buffer.resize_destructive(src_size);
+ write_buffer.resize_destructive(dst_size);
if (Settings::IsGPULevelExtreme()) {
memory_manager.ReadBlock(regs.offset_in + offset, read_buffer.data(), src_size);
@@ -333,14 +321,10 @@ void MaxwellDMA::CopyBlockLinearToBlockLinear() {
const u32 pitch = x_elements * bytes_per_pixel;
const size_t mid_buffer_size = pitch * regs.line_count;
- if (read_buffer.size() < src_size) {
- read_buffer.resize(src_size);
- }
- if (write_buffer.size() < dst_size) {
- write_buffer.resize(dst_size);
- }
+ read_buffer.resize_destructive(src_size);
+ write_buffer.resize_destructive(dst_size);
- intermediate_buffer.resize(mid_buffer_size);
+ intermediate_buffer.resize_destructive(mid_buffer_size);
memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size);
memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size);
diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h
index d40d3d302..c88191a61 100644
--- a/src/video_core/engines/maxwell_dma.h
+++ b/src/video_core/engines/maxwell_dma.h
@@ -6,8 +6,10 @@
#include <array>
#include <cstddef>
#include <vector>
+
#include "common/bit_field.h"
#include "common/common_types.h"
+#include "common/scratch_buffer.h"
#include "video_core/engines/engine_interface.h"
namespace Core {
@@ -234,9 +236,9 @@ private:
MemoryManager& memory_manager;
VideoCore::RasterizerInterface* rasterizer = nullptr;
- std::vector<u8> read_buffer;
- std::vector<u8> write_buffer;
- std::vector<u8> intermediate_buffer;
+ Common::ScratchBuffer<u8> read_buffer;
+ Common::ScratchBuffer<u8> write_buffer;
+ Common::ScratchBuffer<u8> intermediate_buffer;
static constexpr std::size_t NUM_REGS = 0x800;
struct Regs {
diff --git a/src/video_core/host1x/vic.cpp b/src/video_core/host1x/vic.cpp
index ac0b7d20e..36a04e4e0 100644
--- a/src/video_core/host1x/vic.cpp
+++ b/src/video_core/host1x/vic.cpp
@@ -155,7 +155,7 @@ void Vic::WriteRGBFrame(const AVFrame* frame, const VicConfig& config) {
// swizzle pitch linear to block linear
const u32 block_height = static_cast<u32>(config.block_linear_height_log2);
const auto size = Texture::CalculateSize(true, 4, width, height, 1, block_height, 0);
- luma_buffer.resize(size);
+ luma_buffer.resize_destructive(size);
std::span<const u8> frame_buff(converted_frame_buf_addr, 4 * width * height);
Texture::SwizzleSubrect(luma_buffer, frame_buff, 4, width, height, 1, 0, 0, width, height,
block_height, 0, width * 4);
@@ -181,8 +181,8 @@ void Vic::WriteYUVFrame(const AVFrame* frame, const VicConfig& config) {
const auto stride = static_cast<size_t>(frame->linesize[0]);
- luma_buffer.resize(aligned_width * surface_height);
- chroma_buffer.resize(aligned_width * surface_height / 2);
+ luma_buffer.resize_destructive(aligned_width * surface_height);
+ chroma_buffer.resize_destructive(aligned_width * surface_height / 2);
// Populate luma buffer
const u8* luma_src = frame->data[0];
diff --git a/src/video_core/host1x/vic.h b/src/video_core/host1x/vic.h
index 2b78786e8..3d9753047 100644
--- a/src/video_core/host1x/vic.h
+++ b/src/video_core/host1x/vic.h
@@ -4,8 +4,9 @@
#pragma once
#include <memory>
-#include <vector>
+
#include "common/common_types.h"
+#include "common/scratch_buffer.h"
struct SwsContext;
@@ -49,8 +50,8 @@ private:
/// size does not change during a stream
using AVMallocPtr = std::unique_ptr<u8, decltype(&av_free)>;
AVMallocPtr converted_frame_buffer;
- std::vector<u8> luma_buffer;
- std::vector<u8> chroma_buffer;
+ Common::ScratchBuffer<u8> luma_buffer;
+ Common::ScratchBuffer<u8> chroma_buffer;
GPUVAddr config_struct_address{};
GPUVAddr output_surface_luma_address{};
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-22 01:19:23 +0100