Compare commits

...

35 Commits

Author SHA1 Message Date
tgsm
030814b1cb Remove GCC version checks
Citra can't be compiled using GCC <7 because of required C++17 support, so these version checks don't need to exist anymore.
2019-02-24 15:24:06 +01:00
bunnei
90c780e6f3 Merge pull request #2139 from degasus/dma_pusher
video_core/dma_pusher: The full list of headers at once.
2019-02-24 04:15:49 -05:00
bunnei
f7090bacc5 Merge pull request #2146 from ReinUsesLisp/vulkan-scheduler
vk_scheduler: Implement a scheduler
2019-02-23 23:32:43 -05:00
bunnei
d062991643 Merge pull request #2150 from ReinUsesLisp/fixup-layer-swizzle
gl_rasterizer_cache: Fixup parameter order in layered swizzle
2019-02-23 23:31:34 -05:00
bunnei
4ab978d670 Merge pull request #2151 from ReinUsesLisp/fixup-vk-memory-manager
vk_memory_manager: Fixup commit interval allocation
2019-02-23 23:29:53 -05:00
ReinUsesLisp
92050c4d86 vk_memory_manager: Fixup commit interval allocation
VKMemoryCommitImpl was using as the end of its interval "begin + end".
That ended up wasting memory.
2019-02-24 01:04:41 -03:00
ReinUsesLisp
abef11a540 gl_rasterizer_cache: Fixup parameter order in layered swizzle 2019-02-23 23:27:30 -03:00
ReinUsesLisp
f546fb35ed vk_scheduler: Implement a scheduler
The scheduler abstracts command buffer and fence management with an
interface that's able to do OpenGL-like operations on Vulkan command
buffers.

It returns by value a command buffer and fence that have to be used for
subsequent operations until Flush or Finish is executed, after that the
current execution context (the pair of command buffers and fences) gets
invalidated a new one must be fetched. Thankfully validation layers will
quickly detect if this is skipped throwing an error due to modifications
to a sent command buffer.
2019-02-22 01:33:32 -03:00
bunnei
94b27bb8a5 Merge pull request #2138 from ReinUsesLisp/vulkan-memory-manager
vk_memory_manager: Implement memory manager
2019-02-21 22:26:54 -05:00
bunnei
9539c4203b Merge pull request #2125 from ReinUsesLisp/fixup-glstate
gl_state: Synchronize gl_state even when state is disabled
2019-02-20 21:47:46 -05:00
bunnei
ae437320c8 Merge pull request #2130 from lioncash/system_engine
video_core: Remove usages of System::GetInstance() within the engines
2019-02-20 21:24:56 -05:00
Jungy
3273f93cd5 Fixes Unicode Key File Directories (#2120)
* Fixes Unicode Key File Directories

Adds code so that when loading a file it converts to UTF16 first, to
ensure the files can be opened. Code borrowed from FileUtil::Exists.

* Update src/core/crypto/key_manager.cpp

Co-Authored-By: Jungorend <Jungorend@users.noreply.github.com>

* Update src/core/crypto/key_manager.cpp

Co-Authored-By: Jungorend <Jungorend@users.noreply.github.com>

* Using FileUtil instead to be cleaner.

* Update src/core/crypto/key_manager.cpp

Co-Authored-By: Jungorend <Jungorend@users.noreply.github.com>
2019-02-20 21:24:25 -05:00
bunnei
ef559f5741 Merge pull request #2142 from lioncash/relocate
service/nvflinger: Relocate definitions of Layer and Display to the vi service
2019-02-20 21:21:55 -05:00
Lioncash
8d5d369b54 service/nvflinger: Relocate definitions of Layer and Display to the vi service
These are more closely related to the vi service as opposed to the
intermediary nvflinger.

This also places them in their relevant subfolder, as future changes to
these will likely result in subclassing to represent various displays
and services, as they're done within the service itself on hardware.

The reasoning for prefixing the display and layer source files is to
avoid potential clashing if two files with the same name are compiled
(e.g. if 'display.cpp/.h' or 'layer.cpp/.h' is added to another service
at any point), which MSVC will actually warn against. This prevents that
case from occurring.

This also presently coverts the std::array introduced within
f45c25aaba back to a std::vector to allow
the forward declaration of the Display type. Forward declaring a type
within a std::vector is allowed since the introduction of N4510
(http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2015/n4510.html) by
Zhihao Yuan.
2019-02-19 18:27:16 -05:00
Markus Wick
6dd40976d0 video_core/dma_pusher: Simplyfy Step() logic.
As fetching command list headers and and the list of command headers is a fixed 1:1 relation now, they can be implemented within a single call.
This cleans up the Step() logic quite a bit.
2019-02-19 10:28:42 +01:00
Markus Wick
717394c980 video_core/dma_pusher: The full list of headers at once.
Fetching every u32 from memory leads to a big overhead. So let's fetch all of them as a block if possible.
This reduces the Memory::* calls by the dma_pusher by a factor of 10.
2019-02-19 09:58:38 +01:00
ReinUsesLisp
b675c97cdd vk_memory_manager: Implement memory manager
A memory manager object handles the memory allocations for a device. It
allocates chunks of Vulkan memory objects and then suballocates.
2019-02-19 03:42:28 -03:00
bunnei
4bce08d497 Merge pull request #2122 from ReinUsesLisp/vulkan-resource-manager
vk_resource_manager: Implement fence and command buffer allocator
2019-02-18 21:05:28 -05:00
bunnei
2bb02a0b78 Merge pull request #2134 from lioncash/naming
audio_core/buffer: Make const and non-const getter for samples consistent
2019-02-17 11:26:33 -05:00
bunnei
e869c5ef1a Merge pull request #2133 from lioncash/arbiter
address_arbiter: Use nested namespaces where applicable
2019-02-16 15:37:21 -05:00
bunnei
4699fdca8f Merge pull request #2127 from FearlessTobi/fix-screenshot-srgb
renderer_opengl: respect the sRGB colorspace for the screenshot feature
2019-02-16 15:36:00 -05:00
bunnei
cd7e1183e2 Merge pull request #2128 from FearlessTobi/port-4197
Port citra-emu/citra#4197: "threadsafe_queue: Add PopWait and use it where possible "
2019-02-16 15:34:49 -05:00
Lioncash
b009bda67a audio_core/buffer: Make const and non-const getter for samples consistent
This way proper const/non-const selection can occur.
2019-02-16 15:21:35 -05:00
Lioncash
0113c36300 address_arbiter: Use nested namespaces where applicable
A fairly trivial change. Other sections of the codebase use nested
namespaces instead of separate namespaces here. This one must have just
been overlooked.
2019-02-16 12:41:30 -05:00
Lioncash
a8fa5019b5 video_core: Remove usages of System::GetInstance() within the engines
Avoids the use of the global accessor in favor of explicitly making the
system a dependency within the interface.
2019-02-15 22:06:23 -05:00
B3n30
2195f10d15 Adressed review comments 2019-02-15 22:14:54 +01:00
B3n30
4154936568 threadsafe_queue: Add WaitIfEmpty and use it in logging 2019-02-15 22:12:54 +01:00
fearlessTobi
9a56b99fa4 renderer_opengl: respect the sRGB colorspace for the screenshot feature
Previously, we were completely ignoring for screenshots whether the game uses RGB or sRGB.
This resulted in screenshot colors that looked off for some titles.
2019-02-15 21:27:29 +01:00
ReinUsesLisp
8dfc81239f gl_state: Synchronize gl_state even when state is disabled
There are some potential edge cases where gl_state may fail to track the
state if a related state changes while the toggle is disabled or it
didn't change. This addresses that.
2019-02-15 01:30:14 -03:00
ReinUsesLisp
ae6c052ed9 vk_resource_manager: Implement a command buffer pool with VKFencedPool 2019-02-14 18:44:26 -03:00
ReinUsesLisp
a2b6de7e9f vk_resource_manager: Add VKFencedPool interface
Handles a pool of resources protected by fences. Manages resource
overflow allocating more resources.

This class is intended to be used through inheritance.
2019-02-14 18:44:26 -03:00
ReinUsesLisp
0ffdd0a683 vk_resource_manager: Implement VKResourceManager and fence allocator
CommitFence iterates a pool of fences until one is found. If all fences
are being used at the same time, allocate more.
2019-02-14 18:44:26 -03:00
ReinUsesLisp
aa0b6babda vk_resource_manager: Implement VKFenceWatch
A fence watch is used to keep track of the usage of a fence and protect
a resource or set of resources without having to inherit from their
handlers.
2019-02-14 18:44:26 -03:00
ReinUsesLisp
25c2fe1c6b vk_resource_manager: Implement VKFence
Fences take ownership of objects, protecting them from GPU-side or
driver-side concurrent access. They must be commited from the resource
manager. Their usage flow is: commit the fence from the resource
manager, protect resources with it and use them, send the fence to an
execution queue and Wait for it if needed and then call Release. Used
resources will automatically be signaled when they are free to be
reused.
2019-02-14 18:44:26 -03:00
ReinUsesLisp
33a4cebc22 vk_resource_manager: Add VKResource interface
VKResource is an interface that gets signaled by a fence when it is free
to be reused.
2019-02-14 18:36:15 -03:00
38 changed files with 1247 additions and 220 deletions

View File

@@ -21,7 +21,7 @@ public:
Buffer(Tag tag, std::vector<s16>&& samples) : tag{tag}, samples{std::move(samples)} {}
/// Returns the raw audio data for the buffer
std::vector<s16>& Samples() {
std::vector<s16>& GetSamples() {
return samples;
}

View File

@@ -95,7 +95,7 @@ void Stream::PlayNextBuffer() {
active_buffer = queued_buffers.front();
queued_buffers.pop();
VolumeAdjustSamples(active_buffer->Samples());
VolumeAdjustSamples(active_buffer->GetSamples());
sink_stream.EnqueueSamples(GetNumChannels(), active_buffer->GetSamples());

View File

@@ -40,9 +40,7 @@ public:
const Impl& operator=(Impl const&) = delete;
void PushEntry(Entry e) {
std::lock_guard<std::mutex> lock(message_mutex);
message_queue.Push(std::move(e));
message_cv.notify_one();
}
void AddBackend(std::unique_ptr<Backend> backend) {
@@ -86,15 +84,13 @@ private:
}
};
while (true) {
{
std::unique_lock<std::mutex> lock(message_mutex);
message_cv.wait(lock, [&] { return !running || message_queue.Pop(entry); });
}
if (!running) {
entry = message_queue.PopWait();
if (entry.final_entry) {
break;
}
write_logs(entry);
}
// Drain the logging queue. Only writes out up to MAX_LOGS_TO_WRITE to prevent a case
// where a system is repeatedly spamming logs even on close.
const int MAX_LOGS_TO_WRITE = filter.IsDebug() ? INT_MAX : 100;
@@ -106,14 +102,13 @@ private:
}
~Impl() {
running = false;
message_cv.notify_one();
Entry entry;
entry.final_entry = true;
message_queue.Push(entry);
backend_thread.join();
}
std::atomic_bool running{true};
std::mutex message_mutex, writing_mutex;
std::condition_variable message_cv;
std::mutex writing_mutex;
std::thread backend_thread;
std::vector<std::unique_ptr<Backend>> backends;
Common::MPSCQueue<Log::Entry> message_queue;

View File

@@ -27,6 +27,7 @@ struct Entry {
unsigned int line_num;
std::string function;
std::string message;
bool final_entry = false;
Entry() = default;
Entry(Entry&& o) = default;

View File

@@ -28,8 +28,8 @@
#include <cstring>
#include "common/common_types.h"
// GCC 4.6+
#if __GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6)
// GCC
#ifdef __GNUC__
#if __BYTE_ORDER__ && (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) && !defined(COMMON_LITTLE_ENDIAN)
#define COMMON_LITTLE_ENDIAN 1
@@ -38,7 +38,7 @@
#endif
// LLVM/clang
#elif __clang__
#elif defined(__clang__)
#if __LITTLE_ENDIAN__ && !defined(COMMON_LITTLE_ENDIAN)
#define COMMON_LITTLE_ENDIAN 1

View File

@@ -8,6 +8,7 @@
// single reader, single writer queue
#include <atomic>
#include <condition_variable>
#include <cstddef>
#include <mutex>
#include <utility>
@@ -45,6 +46,7 @@ public:
ElementPtr* new_ptr = new ElementPtr();
write_ptr->next.store(new_ptr, std::memory_order_release);
write_ptr = new_ptr;
cv.notify_one();
++size;
}
@@ -74,6 +76,16 @@ public:
return true;
}
T PopWait() {
if (Empty()) {
std::unique_lock<std::mutex> lock(cv_mutex);
cv.wait(lock, [this]() { return !Empty(); });
}
T t;
Pop(t);
return t;
}
// not thread-safe
void Clear() {
size.store(0);
@@ -101,6 +113,8 @@ private:
ElementPtr* write_ptr;
ElementPtr* read_ptr;
std::atomic_size_t size{0};
std::mutex cv_mutex;
std::condition_variable cv;
};
// a simple thread-safe,
@@ -135,6 +149,10 @@ public:
return spsc_queue.Pop(t);
}
T PopWait() {
return spsc_queue.PopWait();
}
// not thread-safe
void Clear() {
spsc_queue.Clear();

View File

@@ -400,6 +400,10 @@ add_library(core STATIC
hle/service/time/time.h
hle/service/usb/usb.cpp
hle/service/usb/usb.h
hle/service/vi/display/vi_display.cpp
hle/service/vi/display/vi_display.h
hle/service/vi/layer/vi_layer.cpp
hle/service/vi/layer/vi_layer.h
hle/service/vi/vi.cpp
hle/service/vi/vi.h
hle/service/vi/vi_m.cpp

View File

@@ -128,7 +128,7 @@ struct System::Impl {
return ResultStatus::ErrorVideoCore;
}
gpu_core = std::make_unique<Tegra::GPU>(renderer->Rasterizer());
gpu_core = std::make_unique<Tegra::GPU>(system, renderer->Rasterizer());
cpu_core_manager.Initialize(system);
is_powered_on = true;

View File

@@ -398,7 +398,8 @@ static bool ValidCryptoRevisionString(std::string_view base, size_t begin, size_
}
void KeyManager::LoadFromFile(const std::string& filename, bool is_title_keys) {
std::ifstream file(filename);
std::ifstream file;
OpenFStream(file, filename, std::ios_base::in);
if (!file.is_open())
return;

View File

@@ -17,8 +17,7 @@
#include "core/hle/result.h"
#include "core/memory.h"
namespace Kernel {
namespace AddressArbiter {
namespace Kernel::AddressArbiter {
// Performs actual address waiting logic.
static ResultCode WaitForAddress(VAddr address, s64 timeout) {
@@ -176,5 +175,4 @@ ResultCode WaitForAddressIfEqual(VAddr address, s32 value, s64 timeout) {
return WaitForAddress(address, timeout);
}
} // namespace AddressArbiter
} // namespace Kernel
} // namespace Kernel::AddressArbiter

View File

@@ -8,9 +8,8 @@
union ResultCode;
namespace Kernel {
namespace Kernel::AddressArbiter {
namespace AddressArbiter {
enum class ArbitrationType {
WaitIfLessThan = 0,
DecrementAndWaitIfLessThan = 1,
@@ -29,6 +28,5 @@ ResultCode ModifyByWaitingCountAndSignalToAddressIfEqual(VAddr address, s32 valu
ResultCode WaitForAddressIfLessThan(VAddr address, s32 value, s64 timeout, bool should_decrement);
ResultCode WaitForAddressIfEqual(VAddr address, s32 value, s64 timeout);
} // namespace AddressArbiter
} // namespace Kernel
} // namespace Kernel::AddressArbiter

View File

@@ -14,11 +14,12 @@
#include "core/core_timing_util.h"
#include "core/hle/kernel/kernel.h"
#include "core/hle/kernel/readable_event.h"
#include "core/hle/kernel/writable_event.h"
#include "core/hle/service/nvdrv/devices/nvdisp_disp0.h"
#include "core/hle/service/nvdrv/nvdrv.h"
#include "core/hle/service/nvflinger/buffer_queue.h"
#include "core/hle/service/nvflinger/nvflinger.h"
#include "core/hle/service/vi/display/vi_display.h"
#include "core/hle/service/vi/layer/vi_layer.h"
#include "core/perf_stats.h"
#include "video_core/renderer_base.h"
@@ -27,7 +28,9 @@ namespace Service::NVFlinger {
constexpr std::size_t SCREEN_REFRESH_RATE = 60;
constexpr u64 frame_ticks = static_cast<u64>(Core::Timing::BASE_CLOCK_RATE / SCREEN_REFRESH_RATE);
NVFlinger::NVFlinger(Core::Timing::CoreTiming& core_timing) : core_timing{core_timing} {
NVFlinger::NVFlinger(Core::Timing::CoreTiming& core_timing)
: displays{{0, "Default"}, {1, "External"}, {2, "Edid"}, {3, "Internal"}, {4, "Null"}},
core_timing{core_timing} {
// Schedule the screen composition events
composition_event =
core_timing.RegisterEvent("ScreenComposition", [this](u64 userdata, int cycles_late) {
@@ -53,7 +56,7 @@ std::optional<u64> NVFlinger::OpenDisplay(std::string_view name) {
ASSERT(name == "Default");
const auto itr = std::find_if(displays.begin(), displays.end(),
[&](const Display& display) { return display.name == name; });
[&](const VI::Display& display) { return display.name == name; });
if (itr == displays.end()) {
return {};
}
@@ -106,9 +109,10 @@ std::shared_ptr<BufferQueue> NVFlinger::FindBufferQueue(u32 id) const {
return *itr;
}
Display* NVFlinger::FindDisplay(u64 display_id) {
const auto itr = std::find_if(displays.begin(), displays.end(),
[&](const Display& display) { return display.id == display_id; });
VI::Display* NVFlinger::FindDisplay(u64 display_id) {
const auto itr =
std::find_if(displays.begin(), displays.end(),
[&](const VI::Display& display) { return display.id == display_id; });
if (itr == displays.end()) {
return nullptr;
@@ -117,9 +121,10 @@ Display* NVFlinger::FindDisplay(u64 display_id) {
return &*itr;
}
const Display* NVFlinger::FindDisplay(u64 display_id) const {
const auto itr = std::find_if(displays.begin(), displays.end(),
[&](const Display& display) { return display.id == display_id; });
const VI::Display* NVFlinger::FindDisplay(u64 display_id) const {
const auto itr =
std::find_if(displays.begin(), displays.end(),
[&](const VI::Display& display) { return display.id == display_id; });
if (itr == displays.end()) {
return nullptr;
@@ -128,7 +133,7 @@ const Display* NVFlinger::FindDisplay(u64 display_id) const {
return &*itr;
}
Layer* NVFlinger::FindLayer(u64 display_id, u64 layer_id) {
VI::Layer* NVFlinger::FindLayer(u64 display_id, u64 layer_id) {
auto* const display = FindDisplay(display_id);
if (display == nullptr) {
@@ -136,7 +141,7 @@ Layer* NVFlinger::FindLayer(u64 display_id, u64 layer_id) {
}
const auto itr = std::find_if(display->layers.begin(), display->layers.end(),
[&](const Layer& layer) { return layer.id == layer_id; });
[&](const VI::Layer& layer) { return layer.id == layer_id; });
if (itr == display->layers.end()) {
return nullptr;
@@ -145,7 +150,7 @@ Layer* NVFlinger::FindLayer(u64 display_id, u64 layer_id) {
return &*itr;
}
const Layer* NVFlinger::FindLayer(u64 display_id, u64 layer_id) const {
const VI::Layer* NVFlinger::FindLayer(u64 display_id, u64 layer_id) const {
const auto* const display = FindDisplay(display_id);
if (display == nullptr) {
@@ -153,7 +158,7 @@ const Layer* NVFlinger::FindLayer(u64 display_id, u64 layer_id) const {
}
const auto itr = std::find_if(display->layers.begin(), display->layers.end(),
[&](const Layer& layer) { return layer.id == layer_id; });
[&](const VI::Layer& layer) { return layer.id == layer_id; });
if (itr == display->layers.end()) {
return nullptr;
@@ -174,7 +179,7 @@ void NVFlinger::Compose() {
// TODO(Subv): Support more than 1 layer.
ASSERT_MSG(display.layers.size() == 1, "Max 1 layer per display is supported");
Layer& layer = display.layers[0];
VI::Layer& layer = display.layers[0];
auto& buffer_queue = layer.buffer_queue;
// Search for a queued buffer and acquire it
@@ -207,15 +212,4 @@ void NVFlinger::Compose() {
}
}
Layer::Layer(u64 id, std::shared_ptr<BufferQueue> queue) : id(id), buffer_queue(std::move(queue)) {}
Layer::~Layer() = default;
Display::Display(u64 id, std::string name) : id(id), name(std::move(name)) {
auto& kernel = Core::System::GetInstance().Kernel();
vsync_event = Kernel::WritableEvent::CreateEventPair(kernel, Kernel::ResetType::Sticky,
fmt::format("Display VSync Event {}", id));
}
Display::~Display() = default;
} // namespace Service::NVFlinger

View File

@@ -4,7 +4,6 @@
#pragma once
#include <array>
#include <memory>
#include <optional>
#include <string>
@@ -26,31 +25,17 @@ class WritableEvent;
namespace Service::Nvidia {
class Module;
}
} // namespace Service::Nvidia
namespace Service::VI {
struct Display;
struct Layer;
} // namespace Service::VI
namespace Service::NVFlinger {
class BufferQueue;
struct Layer {
Layer(u64 id, std::shared_ptr<BufferQueue> queue);
~Layer();
u64 id;
std::shared_ptr<BufferQueue> buffer_queue;
};
struct Display {
Display(u64 id, std::string name);
~Display();
u64 id;
std::string name;
std::vector<Layer> layers;
Kernel::EventPair vsync_event;
};
class NVFlinger final {
public:
explicit NVFlinger(Core::Timing::CoreTiming& core_timing);
@@ -88,26 +73,20 @@ public:
private:
/// Finds the display identified by the specified ID.
Display* FindDisplay(u64 display_id);
VI::Display* FindDisplay(u64 display_id);
/// Finds the display identified by the specified ID.
const Display* FindDisplay(u64 display_id) const;
const VI::Display* FindDisplay(u64 display_id) const;
/// Finds the layer identified by the specified ID in the desired display.
Layer* FindLayer(u64 display_id, u64 layer_id);
VI::Layer* FindLayer(u64 display_id, u64 layer_id);
/// Finds the layer identified by the specified ID in the desired display.
const Layer* FindLayer(u64 display_id, u64 layer_id) const;
const VI::Layer* FindLayer(u64 display_id, u64 layer_id) const;
std::shared_ptr<Nvidia::Module> nvdrv;
std::array<Display, 5> displays{{
{0, "Default"},
{1, "External"},
{2, "Edid"},
{3, "Internal"},
{4, "Null"},
}};
std::vector<VI::Display> displays;
std::vector<std::shared_ptr<BufferQueue>> buffer_queues;
/// Id to use for the next layer that is created, this counter is shared among all displays.

View File

@@ -0,0 +1,22 @@
// Copyright 2019 yuzu emulator team
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <fmt/format.h>
#include "core/core.h"
#include "core/hle/kernel/readable_event.h"
#include "core/hle/service/vi/display/vi_display.h"
#include "core/hle/service/vi/layer/vi_layer.h"
namespace Service::VI {
Display::Display(u64 id, std::string name) : id{id}, name{std::move(name)} {
auto& kernel = Core::System::GetInstance().Kernel();
vsync_event = Kernel::WritableEvent::CreateEventPair(kernel, Kernel::ResetType::Sticky,
fmt::format("Display VSync Event {}", id));
}
Display::~Display() = default;
} // namespace Service::VI

View File

@@ -0,0 +1,28 @@
// Copyright 2019 yuzu emulator team
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <string>
#include <vector>
#include "common/common_types.h"
#include "core/hle/kernel/writable_event.h"
namespace Service::VI {
struct Layer;
struct Display {
Display(u64 id, std::string name);
~Display();
u64 id;
std::string name;
std::vector<Layer> layers;
Kernel::EventPair vsync_event;
};
} // namespace Service::VI

View File

@@ -0,0 +1,14 @@
// Copyright 2019 yuzu emulator team
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "core/hle/service/vi/layer/vi_layer.h"
namespace Service::VI {
Layer::Layer(u64 id, std::shared_ptr<NVFlinger::BufferQueue> queue)
: id{id}, buffer_queue{std::move(queue)} {}
Layer::~Layer() = default;
} // namespace Service::VI

View File

@@ -0,0 +1,25 @@
// Copyright 2019 yuzu emulator team
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <memory>
#include "common/common_types.h"
namespace Service::NVFlinger {
class BufferQueue;
}
namespace Service::VI {
struct Layer {
Layer(u64 id, std::shared_ptr<NVFlinger::BufferQueue> queue);
~Layer();
u64 id;
std::shared_ptr<NVFlinger::BufferQueue> buffer_queue;
};
} // namespace Service::VI

View File

@@ -105,7 +105,13 @@ if (ENABLE_VULKAN)
target_sources(video_core PRIVATE
renderer_vulkan/declarations.h
renderer_vulkan/vk_device.cpp
renderer_vulkan/vk_device.h)
renderer_vulkan/vk_device.h
renderer_vulkan/vk_memory_manager.cpp
renderer_vulkan/vk_memory_manager.h
renderer_vulkan/vk_resource_manager.cpp
renderer_vulkan/vk_resource_manager.h
renderer_vulkan/vk_scheduler.cpp
renderer_vulkan/vk_scheduler.h)
target_include_directories(video_core PRIVATE ../../externals/Vulkan-Headers/include)
target_compile_definitions(video_core PRIVATE HAS_VULKAN)

View File

@@ -33,18 +33,36 @@ void DmaPusher::DispatchCalls() {
}
bool DmaPusher::Step() {
if (dma_get != dma_put) {
// Push buffer non-empty, read a word
const auto address = gpu.MemoryManager().GpuToCpuAddress(dma_get);
ASSERT_MSG(address, "Invalid GPU address");
if (!ib_enable || dma_pushbuffer.empty()) {
// pushbuffer empty and IB empty or nonexistent - nothing to do
return false;
}
const CommandHeader command_header{Memory::Read32(*address)};
const CommandList& command_list{dma_pushbuffer.front()};
const CommandListHeader& command_list_header{command_list[dma_pushbuffer_subindex++]};
GPUVAddr dma_get = command_list_header.addr;
GPUVAddr dma_put = dma_get + command_list_header.size * sizeof(u32);
bool non_main = command_list_header.is_non_main;
dma_get += sizeof(u32);
if (dma_pushbuffer_subindex >= command_list.size()) {
// We've gone through the current list, remove it from the queue
dma_pushbuffer.pop();
dma_pushbuffer_subindex = 0;
}
if (!non_main) {
dma_mget = dma_get;
}
if (command_list_header.size == 0) {
return true;
}
// Push buffer non-empty, read a word
const auto address = gpu.MemoryManager().GpuToCpuAddress(dma_get);
ASSERT_MSG(address, "Invalid GPU address");
command_headers.resize(command_list_header.size);
Memory::ReadBlock(*address, command_headers.data(), command_list_header.size * sizeof(u32));
for (const CommandHeader& command_header : command_headers) {
// now, see if we're in the middle of a command
if (dma_state.length_pending) {
@@ -91,22 +109,11 @@ bool DmaPusher::Step() {
break;
}
}
} else if (ib_enable && !dma_pushbuffer.empty()) {
// Current pushbuffer empty, but we have more IB entries to read
const CommandList& command_list{dma_pushbuffer.front()};
const CommandListHeader& command_list_header{command_list[dma_pushbuffer_subindex++]};
dma_get = command_list_header.addr;
dma_put = dma_get + command_list_header.size * sizeof(u32);
non_main = command_list_header.is_non_main;
}
if (dma_pushbuffer_subindex >= command_list.size()) {
// We've gone through the current list, remove it from the queue
dma_pushbuffer.pop();
dma_pushbuffer_subindex = 0;
}
} else {
// Otherwise, pushbuffer empty and IB empty or nonexistent - nothing to do
return {};
if (!non_main) {
// TODO (degasus): This is dead code, as dma_mget is never read.
dma_mget = dma_put;
}
return true;

View File

@@ -75,6 +75,8 @@ private:
GPU& gpu;
std::vector<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
@@ -89,11 +91,8 @@ private:
DmaState dma_state{};
bool dma_increment_once{};
GPUVAddr dma_put{}; ///< pushbuffer current end address
GPUVAddr dma_get{}; ///< pushbuffer current read address
GPUVAddr dma_mget{}; ///< main pushbuffer last read address
bool ib_enable{true}; ///< IB mode enabled
bool non_main{}; ///< non-main pushbuffer active
};
} // namespace Tegra

View File

@@ -2,6 +2,7 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/logging/log.h"
#include "core/core.h"
#include "core/memory.h"
@@ -11,9 +12,9 @@
namespace Tegra::Engines {
KeplerMemory::KeplerMemory(VideoCore::RasterizerInterface& rasterizer,
KeplerMemory::KeplerMemory(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
MemoryManager& memory_manager)
: memory_manager(memory_manager), rasterizer{rasterizer} {}
: system{system}, memory_manager(memory_manager), rasterizer{rasterizer} {}
KeplerMemory::~KeplerMemory() = default;
@@ -50,7 +51,7 @@ void KeplerMemory::ProcessData(u32 data) {
rasterizer.InvalidateRegion(*dest_address, sizeof(u32));
Memory::Write32(*dest_address, data);
Core::System::GetInstance().GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
system.GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
state.write_offset++;
}

View File

@@ -5,13 +5,16 @@
#pragma once
#include <array>
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "video_core/gpu.h"
#include "video_core/memory_manager.h"
namespace Core {
class System;
}
namespace VideoCore {
class RasterizerInterface;
}
@@ -23,7 +26,8 @@ namespace Tegra::Engines {
class KeplerMemory final {
public:
KeplerMemory(VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager);
KeplerMemory(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
MemoryManager& memory_manager);
~KeplerMemory();
/// Write the value to the register identified by method.
@@ -76,6 +80,7 @@ public:
} state{};
private:
Core::System& system;
MemoryManager& memory_manager;
VideoCore::RasterizerInterface& rasterizer;

View File

@@ -19,8 +19,10 @@ namespace Tegra::Engines {
/// First register id that is actually a Macro call.
constexpr u32 MacroRegistersStart = 0xE00;
Maxwell3D::Maxwell3D(VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager)
: memory_manager(memory_manager), rasterizer{rasterizer}, macro_interpreter(*this) {
Maxwell3D::Maxwell3D(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
MemoryManager& memory_manager)
: memory_manager(memory_manager), system{system}, rasterizer{rasterizer},
macro_interpreter(*this) {
InitializeRegisterDefaults();
}
@@ -103,7 +105,7 @@ void Maxwell3D::CallMacroMethod(u32 method, std::vector<u32> parameters) {
}
void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
auto debug_context = Core::System::GetInstance().GetGPUDebugContext();
auto debug_context = system.GetGPUDebugContext();
// It is an error to write to a register other than the current macro's ARG register before it
// has finished execution.
@@ -317,7 +319,7 @@ void Maxwell3D::ProcessQueryGet() {
LongQueryResult query_result{};
query_result.value = result;
// TODO(Subv): Generate a real GPU timestamp and write it here instead of CoreTiming
query_result.timestamp = Core::System::GetInstance().CoreTiming().GetTicks();
query_result.timestamp = system.CoreTiming().GetTicks();
Memory::WriteBlock(*address, &query_result, sizeof(query_result));
}
dirty_flags.OnMemoryWrite();
@@ -334,7 +336,7 @@ void Maxwell3D::DrawArrays() {
regs.vertex_buffer.count);
ASSERT_MSG(!(regs.index_array.count && regs.vertex_buffer.count), "Both indexed and direct?");
auto debug_context = Core::System::GetInstance().GetGPUDebugContext();
auto debug_context = system.GetGPUDebugContext();
if (debug_context) {
debug_context->OnEvent(Tegra::DebugContext::Event::IncomingPrimitiveBatch, nullptr);

View File

@@ -17,6 +17,10 @@
#include "video_core/memory_manager.h"
#include "video_core/textures/texture.h"
namespace Core {
class System;
}
namespace VideoCore {
class RasterizerInterface;
}
@@ -28,7 +32,8 @@ namespace Tegra::Engines {
class Maxwell3D final {
public:
explicit Maxwell3D(VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager);
explicit Maxwell3D(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
MemoryManager& memory_manager);
~Maxwell3D() = default;
/// Register structure of the Maxwell3D engine.
@@ -1131,6 +1136,8 @@ public:
private:
void InitializeRegisterDefaults();
Core::System& system;
VideoCore::RasterizerInterface& rasterizer;
/// Start offsets of each macro in macro_memory

View File

@@ -2,6 +2,7 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "core/core.h"
#include "core/memory.h"
#include "video_core/engines/maxwell_3d.h"
@@ -11,8 +12,9 @@
namespace Tegra::Engines {
MaxwellDMA::MaxwellDMA(VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager)
: memory_manager(memory_manager), rasterizer{rasterizer} {}
MaxwellDMA::MaxwellDMA(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
MemoryManager& memory_manager)
: memory_manager(memory_manager), system{system}, rasterizer{rasterizer} {}
void MaxwellDMA::CallMethod(const GPU::MethodCall& method_call) {
ASSERT_MSG(method_call.method < Regs::NUM_REGS,
@@ -59,7 +61,7 @@ void MaxwellDMA::HandleCopy() {
}
// All copies here update the main memory, so mark all rasterizer states as invalid.
Core::System::GetInstance().GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
system.GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
if (regs.exec.is_dst_linear && regs.exec.is_src_linear) {
// When the enable_2d bit is disabled, the copy is performed as if we were copying a 1D

View File

@@ -5,13 +5,16 @@
#pragma once
#include <array>
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "video_core/gpu.h"
#include "video_core/memory_manager.h"
namespace Core {
class System;
}
namespace VideoCore {
class RasterizerInterface;
}
@@ -20,7 +23,8 @@ namespace Tegra::Engines {
class MaxwellDMA final {
public:
explicit MaxwellDMA(VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager);
explicit MaxwellDMA(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
MemoryManager& memory_manager);
~MaxwellDMA() = default;
/// Write the value to the register identified by method.
@@ -137,6 +141,8 @@ public:
MemoryManager& memory_manager;
private:
Core::System& system;
VideoCore::RasterizerInterface& rasterizer;
/// Performs the copy from the source buffer to the destination buffer as configured in the

View File

@@ -28,14 +28,14 @@ u32 FramebufferConfig::BytesPerPixel(PixelFormat format) {
UNREACHABLE();
}
GPU::GPU(VideoCore::RasterizerInterface& rasterizer) {
GPU::GPU(Core::System& system, VideoCore::RasterizerInterface& rasterizer) {
memory_manager = std::make_unique<Tegra::MemoryManager>();
dma_pusher = std::make_unique<Tegra::DmaPusher>(*this);
maxwell_3d = std::make_unique<Engines::Maxwell3D>(rasterizer, *memory_manager);
maxwell_3d = std::make_unique<Engines::Maxwell3D>(system, rasterizer, *memory_manager);
fermi_2d = std::make_unique<Engines::Fermi2D>(rasterizer, *memory_manager);
kepler_compute = std::make_unique<Engines::KeplerCompute>(*memory_manager);
maxwell_dma = std::make_unique<Engines::MaxwellDMA>(rasterizer, *memory_manager);
kepler_memory = std::make_unique<Engines::KeplerMemory>(rasterizer, *memory_manager);
maxwell_dma = std::make_unique<Engines::MaxwellDMA>(system, rasterizer, *memory_manager);
kepler_memory = std::make_unique<Engines::KeplerMemory>(system, rasterizer, *memory_manager);
}
GPU::~GPU() = default;

View File

@@ -6,12 +6,15 @@
#include <array>
#include <memory>
#include <vector>
#include "common/common_types.h"
#include "core/hle/service/nvflinger/buffer_queue.h"
#include "video_core/dma_pusher.h"
#include "video_core/memory_manager.h"
namespace Core {
class System;
}
namespace VideoCore {
class RasterizerInterface;
}
@@ -118,7 +121,7 @@ enum class EngineID {
class GPU final {
public:
explicit GPU(VideoCore::RasterizerInterface& rasterizer);
explicit GPU(Core::System& system, VideoCore::RasterizerInterface& rasterizer);
~GPU();
struct MethodCall {

View File

@@ -423,7 +423,7 @@ void SwizzleFunc(const MortonSwizzleMode& mode, const SurfaceParams& params,
for (u32 i = 0; i < params.depth; i++) {
MortonSwizzle(mode, params.pixel_format, params.MipWidth(mip_level),
params.MipBlockHeight(mip_level), params.MipHeight(mip_level),
params.MipBlockDepth(mip_level), params.tile_width_spacing, 1,
params.MipBlockDepth(mip_level), 1, params.tile_width_spacing,
gl_buffer.data() + offset_gl, gl_size, params.addr + offset);
offset += layer_size;
offset_gl += gl_size;

View File

@@ -11,7 +11,9 @@
namespace OpenGL {
OpenGLState OpenGLState::cur_state;
bool OpenGLState::s_rgb_used;
OpenGLState::OpenGLState() {
// These all match default OpenGL values
geometry_shaders.enabled = false;
@@ -112,7 +114,6 @@ void OpenGLState::ApplyDefaultState() {
}
void OpenGLState::ApplySRgb() const {
// sRGB
if (framebuffer_srgb.enabled != cur_state.framebuffer_srgb.enabled) {
if (framebuffer_srgb.enabled) {
// Track if sRGB is used
@@ -125,23 +126,20 @@ void OpenGLState::ApplySRgb() const {
}
void OpenGLState::ApplyCulling() const {
// Culling
const bool cull_changed = cull.enabled != cur_state.cull.enabled;
if (cull_changed) {
if (cull.enabled != cur_state.cull.enabled) {
if (cull.enabled) {
glEnable(GL_CULL_FACE);
} else {
glDisable(GL_CULL_FACE);
}
}
if (cull.enabled) {
if (cull_changed || cull.mode != cur_state.cull.mode) {
glCullFace(cull.mode);
}
if (cull_changed || cull.front_face != cur_state.cull.front_face) {
glFrontFace(cull.front_face);
}
if (cull.mode != cur_state.cull.mode) {
glCullFace(cull.mode);
}
if (cull.front_face != cur_state.cull.front_face) {
glFrontFace(cull.front_face);
}
}
@@ -172,72 +170,63 @@ void OpenGLState::ApplyColorMask() const {
}
void OpenGLState::ApplyDepth() const {
// Depth test
const bool depth_test_changed = depth.test_enabled != cur_state.depth.test_enabled;
if (depth_test_changed) {
if (depth.test_enabled != cur_state.depth.test_enabled) {
if (depth.test_enabled) {
glEnable(GL_DEPTH_TEST);
} else {
glDisable(GL_DEPTH_TEST);
}
}
if (depth.test_enabled &&
(depth_test_changed || depth.test_func != cur_state.depth.test_func)) {
if (depth.test_func != cur_state.depth.test_func) {
glDepthFunc(depth.test_func);
}
// Depth mask
if (depth.write_mask != cur_state.depth.write_mask) {
glDepthMask(depth.write_mask);
}
}
void OpenGLState::ApplyPrimitiveRestart() const {
const bool primitive_restart_changed =
primitive_restart.enabled != cur_state.primitive_restart.enabled;
if (primitive_restart_changed) {
if (primitive_restart.enabled != cur_state.primitive_restart.enabled) {
if (primitive_restart.enabled) {
glEnable(GL_PRIMITIVE_RESTART);
} else {
glDisable(GL_PRIMITIVE_RESTART);
}
}
if (primitive_restart_changed ||
(primitive_restart.enabled &&
primitive_restart.index != cur_state.primitive_restart.index)) {
if (primitive_restart.index != cur_state.primitive_restart.index) {
glPrimitiveRestartIndex(primitive_restart.index);
}
}
void OpenGLState::ApplyStencilTest() const {
const bool stencil_test_changed = stencil.test_enabled != cur_state.stencil.test_enabled;
if (stencil_test_changed) {
if (stencil.test_enabled != cur_state.stencil.test_enabled) {
if (stencil.test_enabled) {
glEnable(GL_STENCIL_TEST);
} else {
glDisable(GL_STENCIL_TEST);
}
}
if (stencil.test_enabled) {
auto config_stencil = [stencil_test_changed](GLenum face, const auto& config,
const auto& prev_config) {
if (stencil_test_changed || config.test_func != prev_config.test_func ||
config.test_ref != prev_config.test_ref ||
config.test_mask != prev_config.test_mask) {
glStencilFuncSeparate(face, config.test_func, config.test_ref, config.test_mask);
}
if (stencil_test_changed || config.action_depth_fail != prev_config.action_depth_fail ||
config.action_depth_pass != prev_config.action_depth_pass ||
config.action_stencil_fail != prev_config.action_stencil_fail) {
glStencilOpSeparate(face, config.action_stencil_fail, config.action_depth_fail,
config.action_depth_pass);
}
if (config.write_mask != prev_config.write_mask) {
glStencilMaskSeparate(face, config.write_mask);
}
};
config_stencil(GL_FRONT, stencil.front, cur_state.stencil.front);
config_stencil(GL_BACK, stencil.back, cur_state.stencil.back);
}
const auto ConfigStencil = [](GLenum face, const auto& config, const auto& prev_config) {
if (config.test_func != prev_config.test_func || config.test_ref != prev_config.test_ref ||
config.test_mask != prev_config.test_mask) {
glStencilFuncSeparate(face, config.test_func, config.test_ref, config.test_mask);
}
if (config.action_depth_fail != prev_config.action_depth_fail ||
config.action_depth_pass != prev_config.action_depth_pass ||
config.action_stencil_fail != prev_config.action_stencil_fail) {
glStencilOpSeparate(face, config.action_stencil_fail, config.action_depth_fail,
config.action_depth_pass);
}
if (config.write_mask != prev_config.write_mask) {
glStencilMaskSeparate(face, config.write_mask);
}
};
ConfigStencil(GL_FRONT, stencil.front, cur_state.stencil.front);
ConfigStencil(GL_BACK, stencil.back, cur_state.stencil.back);
}
// Viewport does not affects glClearBuffer so emulate viewport using scissor test
void OpenGLState::EmulateViewportWithScissor() {
@@ -278,19 +267,18 @@ void OpenGLState::ApplyViewport() const {
updated.depth_range_far != current.depth_range_far) {
glDepthRangeIndexed(i, updated.depth_range_near, updated.depth_range_far);
}
const bool scissor_changed = updated.scissor.enabled != current.scissor.enabled;
if (scissor_changed) {
if (updated.scissor.enabled != current.scissor.enabled) {
if (updated.scissor.enabled) {
glEnablei(GL_SCISSOR_TEST, i);
} else {
glDisablei(GL_SCISSOR_TEST, i);
}
}
if (updated.scissor.enabled &&
(scissor_changed || updated.scissor.x != current.scissor.x ||
updated.scissor.y != current.scissor.y ||
updated.scissor.width != current.scissor.width ||
updated.scissor.height != current.scissor.height)) {
if (updated.scissor.x != current.scissor.x || updated.scissor.y != current.scissor.y ||
updated.scissor.width != current.scissor.width ||
updated.scissor.height != current.scissor.height) {
glScissorIndexed(i, updated.scissor.x, updated.scissor.y, updated.scissor.width,
updated.scissor.height);
}
@@ -302,22 +290,23 @@ void OpenGLState::ApplyViewport() const {
updated.height != current.height) {
glViewport(updated.x, updated.y, updated.width, updated.height);
}
if (updated.depth_range_near != current.depth_range_near ||
updated.depth_range_far != current.depth_range_far) {
glDepthRange(updated.depth_range_near, updated.depth_range_far);
}
const bool scissor_changed = updated.scissor.enabled != current.scissor.enabled;
if (scissor_changed) {
if (updated.scissor.enabled != current.scissor.enabled) {
if (updated.scissor.enabled) {
glEnable(GL_SCISSOR_TEST);
} else {
glDisable(GL_SCISSOR_TEST);
}
}
if (updated.scissor.enabled && (scissor_changed || updated.scissor.x != current.scissor.x ||
updated.scissor.y != current.scissor.y ||
updated.scissor.width != current.scissor.width ||
updated.scissor.height != current.scissor.height)) {
if (updated.scissor.x != current.scissor.x || updated.scissor.y != current.scissor.y ||
updated.scissor.width != current.scissor.width ||
updated.scissor.height != current.scissor.height) {
glScissor(updated.scissor.x, updated.scissor.y, updated.scissor.width,
updated.scissor.height);
}
@@ -327,8 +316,7 @@ void OpenGLState::ApplyViewport() const {
void OpenGLState::ApplyGlobalBlending() const {
const Blend& current = cur_state.blend[0];
const Blend& updated = blend[0];
const bool blend_changed = updated.enabled != current.enabled;
if (blend_changed) {
if (updated.enabled != current.enabled) {
if (updated.enabled) {
glEnable(GL_BLEND);
} else {
@@ -338,15 +326,14 @@ void OpenGLState::ApplyGlobalBlending() const {
if (!updated.enabled) {
return;
}
if (blend_changed || updated.src_rgb_func != current.src_rgb_func ||
if (updated.src_rgb_func != current.src_rgb_func ||
updated.dst_rgb_func != current.dst_rgb_func || updated.src_a_func != current.src_a_func ||
updated.dst_a_func != current.dst_a_func) {
glBlendFuncSeparate(updated.src_rgb_func, updated.dst_rgb_func, updated.src_a_func,
updated.dst_a_func);
}
if (blend_changed || updated.rgb_equation != current.rgb_equation ||
updated.a_equation != current.a_equation) {
if (updated.rgb_equation != current.rgb_equation || updated.a_equation != current.a_equation) {
glBlendEquationSeparate(updated.rgb_equation, updated.a_equation);
}
}
@@ -354,26 +341,22 @@ void OpenGLState::ApplyGlobalBlending() const {
void OpenGLState::ApplyTargetBlending(std::size_t target, bool force) const {
const Blend& updated = blend[target];
const Blend& current = cur_state.blend[target];
const bool blend_changed = updated.enabled != current.enabled || force;
if (blend_changed) {
if (updated.enabled != current.enabled || force) {
if (updated.enabled) {
glEnablei(GL_BLEND, static_cast<GLuint>(target));
} else {
glDisablei(GL_BLEND, static_cast<GLuint>(target));
}
}
if (!updated.enabled) {
return;
}
if (blend_changed || updated.src_rgb_func != current.src_rgb_func ||
if (updated.src_rgb_func != current.src_rgb_func ||
updated.dst_rgb_func != current.dst_rgb_func || updated.src_a_func != current.src_a_func ||
updated.dst_a_func != current.dst_a_func) {
glBlendFuncSeparatei(static_cast<GLuint>(target), updated.src_rgb_func,
updated.dst_rgb_func, updated.src_a_func, updated.dst_a_func);
}
if (blend_changed || updated.rgb_equation != current.rgb_equation ||
updated.a_equation != current.a_equation) {
if (updated.rgb_equation != current.rgb_equation || updated.a_equation != current.a_equation) {
glBlendEquationSeparatei(static_cast<GLuint>(target), updated.rgb_equation,
updated.a_equation);
}
@@ -397,8 +380,7 @@ void OpenGLState::ApplyBlending() const {
}
void OpenGLState::ApplyLogicOp() const {
const bool logic_op_changed = logic_op.enabled != cur_state.logic_op.enabled;
if (logic_op_changed) {
if (logic_op.enabled != cur_state.logic_op.enabled) {
if (logic_op.enabled) {
glEnable(GL_COLOR_LOGIC_OP);
} else {
@@ -406,14 +388,12 @@ void OpenGLState::ApplyLogicOp() const {
}
}
if (logic_op.enabled &&
(logic_op_changed || logic_op.operation != cur_state.logic_op.operation)) {
if (logic_op.operation != cur_state.logic_op.operation) {
glLogicOp(logic_op.operation);
}
}
void OpenGLState::ApplyPolygonOffset() const {
const bool fill_enable_changed =
polygon_offset.fill_enable != cur_state.polygon_offset.fill_enable;
const bool line_enable_changed =
@@ -448,9 +428,7 @@ void OpenGLState::ApplyPolygonOffset() const {
}
}
if ((polygon_offset.fill_enable || polygon_offset.line_enable || polygon_offset.point_enable) &&
(factor_changed || units_changed || clamp_changed)) {
if (factor_changed || units_changed || clamp_changed) {
if (GLAD_GL_EXT_polygon_offset_clamp && polygon_offset.clamp != 0) {
glPolygonOffsetClamp(polygon_offset.factor, polygon_offset.units, polygon_offset.clamp);
} else {
@@ -528,9 +506,9 @@ void OpenGLState::ApplyDepthClamp() const {
depth_clamp.near_plane == cur_state.depth_clamp.near_plane) {
return;
}
if (depth_clamp.far_plane != depth_clamp.near_plane) {
UNIMPLEMENTED_MSG("Unimplemented Depth Clamp Separation!");
}
UNIMPLEMENTED_IF_MSG(depth_clamp.far_plane != depth_clamp.near_plane,
"Unimplemented Depth Clamp Separation!");
if (depth_clamp.far_plane || depth_clamp.near_plane) {
glEnable(GL_DEPTH_CLAMP);
} else {

View File

@@ -380,7 +380,8 @@ void RendererOpenGL::CaptureScreenshot() {
GLuint renderbuffer;
glGenRenderbuffers(1, &renderbuffer);
glBindRenderbuffer(GL_RENDERBUFFER, renderbuffer);
glRenderbufferStorage(GL_RENDERBUFFER, GL_RGB8, layout.width, layout.height);
glRenderbufferStorage(GL_RENDERBUFFER, state.GetsRGBUsed() ? GL_SRGB8 : GL_RGB8, layout.width,
layout.height);
glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER, renderbuffer);
DrawScreen(layout);

View File

@@ -0,0 +1,252 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <optional>
#include <tuple>
#include <vector>
#include "common/alignment.h"
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_memory_manager.h"
namespace Vulkan {
// TODO(Rodrigo): Fine tune this number
constexpr u64 ALLOC_CHUNK_SIZE = 64 * 1024 * 1024;
class VKMemoryAllocation final {
public:
explicit VKMemoryAllocation(const VKDevice& device, vk::DeviceMemory memory,
vk::MemoryPropertyFlags properties, u64 alloc_size, u32 type)
: device{device}, memory{memory}, properties{properties}, alloc_size{alloc_size},
shifted_type{ShiftType(type)}, is_mappable{properties &
vk::MemoryPropertyFlagBits::eHostVisible} {
if (is_mappable) {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
base_address = static_cast<u8*>(dev.mapMemory(memory, 0, alloc_size, {}, dld));
}
}
~VKMemoryAllocation() {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
if (is_mappable)
dev.unmapMemory(memory, dld);
dev.free(memory, nullptr, dld);
}
VKMemoryCommit Commit(vk::DeviceSize commit_size, vk::DeviceSize alignment) {
auto found = TryFindFreeSection(free_iterator, alloc_size, static_cast<u64>(commit_size),
static_cast<u64>(alignment));
if (!found) {
found = TryFindFreeSection(0, free_iterator, static_cast<u64>(commit_size),
static_cast<u64>(alignment));
if (!found) {
// Signal out of memory, it'll try to do more allocations.
return nullptr;
}
}
u8* address = is_mappable ? base_address + *found : nullptr;
auto commit = std::make_unique<VKMemoryCommitImpl>(this, memory, address, *found,
*found + commit_size);
commits.push_back(commit.get());
// Last commit's address is highly probable to be free.
free_iterator = *found + commit_size;
return commit;
}
void Free(const VKMemoryCommitImpl* commit) {
ASSERT(commit);
const auto it =
std::find_if(commits.begin(), commits.end(),
[&](const auto& stored_commit) { return stored_commit == commit; });
if (it == commits.end()) {
LOG_CRITICAL(Render_Vulkan, "Freeing unallocated commit!");
UNREACHABLE();
return;
}
commits.erase(it);
}
/// Returns whether this allocation is compatible with the arguments.
bool IsCompatible(vk::MemoryPropertyFlags wanted_properties, u32 type_mask) const {
return (wanted_properties & properties) != vk::MemoryPropertyFlagBits(0) &&
(type_mask & shifted_type) != 0;
}
private:
static constexpr u32 ShiftType(u32 type) {
return 1U << type;
}
/// A memory allocator, it may return a free region between "start" and "end" with the solicited
/// requeriments.
std::optional<u64> TryFindFreeSection(u64 start, u64 end, u64 size, u64 alignment) const {
u64 iterator = start;
while (iterator + size < end) {
const u64 try_left = Common::AlignUp(iterator, alignment);
const u64 try_right = try_left + size;
bool overlap = false;
for (const auto& commit : commits) {
const auto [commit_left, commit_right] = commit->interval;
if (try_left < commit_right && commit_left < try_right) {
// There's an overlap, continue the search where the overlapping commit ends.
iterator = commit_right;
overlap = true;
break;
}
}
if (!overlap) {
// A free address has been found.
return try_left;
}
}
// No free regions where found, return an empty optional.
return std::nullopt;
}
const VKDevice& device; ///< Vulkan device.
const vk::DeviceMemory memory; ///< Vulkan memory allocation handler.
const vk::MemoryPropertyFlags properties; ///< Vulkan properties.
const u64 alloc_size; ///< Size of this allocation.
const u32 shifted_type; ///< Stored Vulkan type of this allocation, shifted.
const bool is_mappable; ///< Whether the allocation is mappable.
/// Base address of the mapped pointer.
u8* base_address{};
/// Hints where the next free region is likely going to be.
u64 free_iterator{};
/// Stores all commits done from this allocation.
std::vector<const VKMemoryCommitImpl*> commits;
};
VKMemoryManager::VKMemoryManager(const VKDevice& device)
: device{device}, props{device.GetPhysical().getMemoryProperties(device.GetDispatchLoader())},
is_memory_unified{GetMemoryUnified(props)} {}
VKMemoryManager::~VKMemoryManager() = default;
VKMemoryCommit VKMemoryManager::Commit(const vk::MemoryRequirements& reqs, bool host_visible) {
ASSERT(reqs.size < ALLOC_CHUNK_SIZE);
// When a host visible commit is asked, search for host visible and coherent, otherwise search
// for a fast device local type.
const vk::MemoryPropertyFlags wanted_properties =
host_visible
? vk::MemoryPropertyFlagBits::eHostVisible | vk::MemoryPropertyFlagBits::eHostCoherent
: vk::MemoryPropertyFlagBits::eDeviceLocal;
const auto TryCommit = [&]() -> VKMemoryCommit {
for (auto& alloc : allocs) {
if (!alloc->IsCompatible(wanted_properties, reqs.memoryTypeBits))
continue;
if (auto commit = alloc->Commit(reqs.size, reqs.alignment); commit) {
return commit;
}
}
return {};
};
if (auto commit = TryCommit(); commit) {
return commit;
}
// Commit has failed, allocate more memory.
if (!AllocMemory(wanted_properties, reqs.memoryTypeBits, ALLOC_CHUNK_SIZE)) {
// TODO(Rodrigo): Try to use host memory.
LOG_CRITICAL(Render_Vulkan, "Ran out of memory!");
UNREACHABLE();
}
// Commit again, this time it won't fail since there's a fresh allocation above. If it does,
// there's a bug.
auto commit = TryCommit();
ASSERT(commit);
return commit;
}
VKMemoryCommit VKMemoryManager::Commit(vk::Buffer buffer, bool host_visible) {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
const auto requeriments = dev.getBufferMemoryRequirements(buffer, dld);
auto commit = Commit(requeriments, host_visible);
dev.bindBufferMemory(buffer, commit->GetMemory(), commit->GetOffset(), dld);
return commit;
}
VKMemoryCommit VKMemoryManager::Commit(vk::Image image, bool host_visible) {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
const auto requeriments = dev.getImageMemoryRequirements(image, dld);
auto commit = Commit(requeriments, host_visible);
dev.bindImageMemory(image, commit->GetMemory(), commit->GetOffset(), dld);
return commit;
}
bool VKMemoryManager::AllocMemory(vk::MemoryPropertyFlags wanted_properties, u32 type_mask,
u64 size) {
const u32 type = [&]() {
for (u32 type_index = 0; type_index < props.memoryTypeCount; ++type_index) {
const auto flags = props.memoryTypes[type_index].propertyFlags;
if ((type_mask & (1U << type_index)) && (flags & wanted_properties)) {
// The type matches in type and in the wanted properties.
return type_index;
}
}
LOG_CRITICAL(Render_Vulkan, "Couldn't find a compatible memory type!");
UNREACHABLE();
return 0u;
}();
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
// Try to allocate found type.
const vk::MemoryAllocateInfo memory_ai(size, type);
vk::DeviceMemory memory;
if (const vk::Result res = dev.allocateMemory(&memory_ai, nullptr, &memory, dld);
res != vk::Result::eSuccess) {
LOG_CRITICAL(Render_Vulkan, "Device allocation failed with code {}!", vk::to_string(res));
return false;
}
allocs.push_back(
std::make_unique<VKMemoryAllocation>(device, memory, wanted_properties, size, type));
return true;
}
/*static*/ bool VKMemoryManager::GetMemoryUnified(const vk::PhysicalDeviceMemoryProperties& props) {
for (u32 heap_index = 0; heap_index < props.memoryHeapCount; ++heap_index) {
if (!(props.memoryHeaps[heap_index].flags & vk::MemoryHeapFlagBits::eDeviceLocal)) {
// Memory is considered unified when heaps are device local only.
return false;
}
}
return true;
}
VKMemoryCommitImpl::VKMemoryCommitImpl(VKMemoryAllocation* allocation, vk::DeviceMemory memory,
u8* data, u64 begin, u64 end)
: allocation{allocation}, memory{memory}, data{data}, interval(std::make_pair(begin, end)) {}
VKMemoryCommitImpl::~VKMemoryCommitImpl() {
allocation->Free(this);
}
u8* VKMemoryCommitImpl::GetData() const {
ASSERT_MSG(data != nullptr, "Trying to access an unmapped commit.");
return data;
}
} // namespace Vulkan

View File

@@ -0,0 +1,87 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <memory>
#include <utility>
#include <vector>
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
namespace Vulkan {
class VKDevice;
class VKMemoryAllocation;
class VKMemoryCommitImpl;
using VKMemoryCommit = std::unique_ptr<VKMemoryCommitImpl>;
class VKMemoryManager final {
public:
explicit VKMemoryManager(const VKDevice& device);
~VKMemoryManager();
/**
* Commits a memory with the specified requeriments.
* @param reqs Requeriments returned from a Vulkan call.
* @param host_visible Signals the allocator that it *must* use host visible and coherent
* memory. When passing false, it will try to allocate device local memory.
* @returns A memory commit.
*/
VKMemoryCommit Commit(const vk::MemoryRequirements& reqs, bool host_visible);
/// Commits memory required by the buffer and binds it.
VKMemoryCommit Commit(vk::Buffer buffer, bool host_visible);
/// Commits memory required by the image and binds it.
VKMemoryCommit Commit(vk::Image image, bool host_visible);
/// Returns true if the memory allocations are done always in host visible and coherent memory.
bool IsMemoryUnified() const {
return is_memory_unified;
}
private:
/// Allocates a chunk of memory.
bool AllocMemory(vk::MemoryPropertyFlags wanted_properties, u32 type_mask, u64 size);
/// Returns true if the device uses an unified memory model.
static bool GetMemoryUnified(const vk::PhysicalDeviceMemoryProperties& props);
const VKDevice& device; ///< Device handler.
const vk::PhysicalDeviceMemoryProperties props; ///< Physical device properties.
const bool is_memory_unified; ///< True if memory model is unified.
std::vector<std::unique_ptr<VKMemoryAllocation>> allocs; ///< Current allocations.
};
class VKMemoryCommitImpl final {
friend VKMemoryAllocation;
public:
explicit VKMemoryCommitImpl(VKMemoryAllocation* allocation, vk::DeviceMemory memory, u8* data,
u64 begin, u64 end);
~VKMemoryCommitImpl();
/// Returns the writeable memory map. The commit has to be mappable.
u8* GetData() const;
/// Returns the Vulkan memory handler.
vk::DeviceMemory GetMemory() const {
return memory;
}
/// Returns the start position of the commit relative to the allocation.
vk::DeviceSize GetOffset() const {
return static_cast<vk::DeviceSize>(interval.first);
}
private:
std::pair<u64, u64> interval{}; ///< Interval where the commit exists.
vk::DeviceMemory memory; ///< Vulkan device memory handler.
VKMemoryAllocation* allocation{}; ///< Pointer to the large memory allocation.
u8* data{}; ///< Pointer to the host mapped memory, it has the commit offset included.
};
} // namespace Vulkan

View File

@@ -0,0 +1,285 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <optional>
#include "common/assert.h"
#include "common/logging/log.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
namespace Vulkan {
// TODO(Rodrigo): Fine tune these numbers.
constexpr std::size_t COMMAND_BUFFER_POOL_SIZE = 0x1000;
constexpr std::size_t FENCES_GROW_STEP = 0x40;
class CommandBufferPool final : public VKFencedPool {
public:
CommandBufferPool(const VKDevice& device)
: VKFencedPool(COMMAND_BUFFER_POOL_SIZE), device{device} {}
void Allocate(std::size_t begin, std::size_t end) {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
const u32 graphics_family = device.GetGraphicsFamily();
auto pool = std::make_unique<Pool>();
// Command buffers are going to be commited, recorded, executed every single usage cycle.
// They are also going to be reseted when commited.
const auto pool_flags = vk::CommandPoolCreateFlagBits::eTransient |
vk::CommandPoolCreateFlagBits::eResetCommandBuffer;
const vk::CommandPoolCreateInfo cmdbuf_pool_ci(pool_flags, graphics_family);
pool->handle = dev.createCommandPoolUnique(cmdbuf_pool_ci, nullptr, dld);
const vk::CommandBufferAllocateInfo cmdbuf_ai(*pool->handle,
vk::CommandBufferLevel::ePrimary,
static_cast<u32>(COMMAND_BUFFER_POOL_SIZE));
pool->cmdbufs =
dev.allocateCommandBuffersUnique<std::allocator<UniqueCommandBuffer>>(cmdbuf_ai, dld);
pools.push_back(std::move(pool));
}
vk::CommandBuffer Commit(VKFence& fence) {
const std::size_t index = CommitResource(fence);
const auto pool_index = index / COMMAND_BUFFER_POOL_SIZE;
const auto sub_index = index % COMMAND_BUFFER_POOL_SIZE;
return *pools[pool_index]->cmdbufs[sub_index];
}
private:
struct Pool {
UniqueCommandPool handle;
std::vector<UniqueCommandBuffer> cmdbufs;
};
const VKDevice& device;
std::vector<std::unique_ptr<Pool>> pools;
};
VKResource::VKResource() = default;
VKResource::~VKResource() = default;
VKFence::VKFence(const VKDevice& device, UniqueFence handle)
: device{device}, handle{std::move(handle)} {}
VKFence::~VKFence() = default;
void VKFence::Wait() {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
dev.waitForFences({*handle}, true, std::numeric_limits<u64>::max(), dld);
}
void VKFence::Release() {
is_owned = false;
}
void VKFence::Commit() {
is_owned = true;
is_used = true;
}
bool VKFence::Tick(bool gpu_wait, bool owner_wait) {
if (!is_used) {
// If a fence is not used it's always free.
return true;
}
if (is_owned && !owner_wait) {
// The fence is still being owned (Release has not been called) and ownership wait has
// not been asked.
return false;
}
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
if (gpu_wait) {
// Wait for the fence if it has been requested.
dev.waitForFences({*handle}, true, std::numeric_limits<u64>::max(), dld);
} else {
if (dev.getFenceStatus(*handle, dld) != vk::Result::eSuccess) {
// Vulkan fence is not ready, not much it can do here
return false;
}
}
// Broadcast resources their free state.
for (auto* resource : protected_resources) {
resource->OnFenceRemoval(this);
}
protected_resources.clear();
// Prepare fence for reusage.
dev.resetFences({*handle}, dld);
is_used = false;
return true;
}
void VKFence::Protect(VKResource* resource) {
protected_resources.push_back(resource);
}
void VKFence::Unprotect(const VKResource* resource) {
const auto it = std::find(protected_resources.begin(), protected_resources.end(), resource);
if (it != protected_resources.end()) {
protected_resources.erase(it);
}
}
VKFenceWatch::VKFenceWatch() = default;
VKFenceWatch::~VKFenceWatch() {
if (fence) {
fence->Unprotect(this);
}
}
void VKFenceWatch::Wait() {
if (!fence) {
return;
}
fence->Wait();
fence->Unprotect(this);
fence = nullptr;
}
void VKFenceWatch::Watch(VKFence& new_fence) {
Wait();
fence = &new_fence;
fence->Protect(this);
}
bool VKFenceWatch::TryWatch(VKFence& new_fence) {
if (fence) {
return false;
}
fence = &new_fence;
fence->Protect(this);
return true;
}
void VKFenceWatch::OnFenceRemoval(VKFence* signaling_fence) {
ASSERT_MSG(signaling_fence == fence, "Removing the wrong fence");
fence = nullptr;
}
VKFencedPool::VKFencedPool(std::size_t grow_step) : grow_step{grow_step} {}
VKFencedPool::~VKFencedPool() = default;
std::size_t VKFencedPool::CommitResource(VKFence& fence) {
const auto Search = [&](std::size_t begin, std::size_t end) -> std::optional<std::size_t> {
for (std::size_t iterator = begin; iterator < end; ++iterator) {
if (watches[iterator]->TryWatch(fence)) {
// The resource is now being watched, a free resource was successfully found.
return iterator;
}
}
return {};
};
// Try to find a free resource from the hinted position to the end.
auto found = Search(free_iterator, watches.size());
if (!found) {
// Search from beginning to the hinted position.
found = Search(0, free_iterator);
if (!found) {
// Both searches failed, the pool is full; handle it.
const std::size_t free_resource = ManageOverflow();
// Watch will wait for the resource to be free.
watches[free_resource]->Watch(fence);
found = free_resource;
}
}
// Free iterator is hinted to the resource after the one that's been commited.
free_iterator = (*found + 1) % watches.size();
return *found;
}
std::size_t VKFencedPool::ManageOverflow() {
const std::size_t old_capacity = watches.size();
Grow();
// The last entry is guaranted to be free, since it's the first element of the freshly
// allocated resources.
return old_capacity;
}
void VKFencedPool::Grow() {
const std::size_t old_capacity = watches.size();
watches.resize(old_capacity + grow_step);
std::generate(watches.begin() + old_capacity, watches.end(),
[]() { return std::make_unique<VKFenceWatch>(); });
Allocate(old_capacity, old_capacity + grow_step);
}
VKResourceManager::VKResourceManager(const VKDevice& device) : device{device} {
GrowFences(FENCES_GROW_STEP);
command_buffer_pool = std::make_unique<CommandBufferPool>(device);
}
VKResourceManager::~VKResourceManager() = default;
VKFence& VKResourceManager::CommitFence() {
const auto StepFences = [&](bool gpu_wait, bool owner_wait) -> VKFence* {
const auto Tick = [=](auto& fence) { return fence->Tick(gpu_wait, owner_wait); };
const auto hinted = fences.begin() + fences_iterator;
auto it = std::find_if(hinted, fences.end(), Tick);
if (it == fences.end()) {
it = std::find_if(fences.begin(), hinted, Tick);
if (it == hinted) {
return nullptr;
}
}
fences_iterator = std::distance(fences.begin(), it) + 1;
if (fences_iterator >= fences.size())
fences_iterator = 0;
auto& fence = *it;
fence->Commit();
return fence.get();
};
VKFence* found_fence = StepFences(false, false);
if (!found_fence) {
// Try again, this time waiting.
found_fence = StepFences(true, false);
if (!found_fence) {
// Allocate new fences and try again.
LOG_INFO(Render_Vulkan, "Allocating new fences {} -> {}", fences.size(),
fences.size() + FENCES_GROW_STEP);
GrowFences(FENCES_GROW_STEP);
found_fence = StepFences(true, false);
ASSERT(found_fence != nullptr);
}
}
return *found_fence;
}
vk::CommandBuffer VKResourceManager::CommitCommandBuffer(VKFence& fence) {
return command_buffer_pool->Commit(fence);
}
void VKResourceManager::GrowFences(std::size_t new_fences_count) {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
const vk::FenceCreateInfo fence_ci;
const std::size_t previous_size = fences.size();
fences.resize(previous_size + new_fences_count);
std::generate(fences.begin() + previous_size, fences.end(), [&]() {
return std::make_unique<VKFence>(device, dev.createFenceUnique(fence_ci, nullptr, dld));
});
}
} // namespace Vulkan

View File

@@ -0,0 +1,180 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <cstddef>
#include <memory>
#include <vector>
#include "video_core/renderer_vulkan/declarations.h"
namespace Vulkan {
class VKDevice;
class VKFence;
class VKResourceManager;
class CommandBufferPool;
/// Interface for a Vulkan resource
class VKResource {
public:
explicit VKResource();
virtual ~VKResource();
/**
* Signals the object that an owning fence has been signaled.
* @param signaling_fence Fence that signals its usage end.
*/
virtual void OnFenceRemoval(VKFence* signaling_fence) = 0;
};
/**
* Fences take ownership of objects, protecting them from GPU-side or driver-side concurrent access.
* They must be commited from the resource manager. Their usage flow is: commit the fence from the
* resource manager, protect resources with it and use them, send the fence to an execution queue
* and Wait for it if needed and then call Release. Used resources will automatically be signaled
* when they are free to be reused.
* @brief Protects resources for concurrent usage and signals its release.
*/
class VKFence {
friend class VKResourceManager;
public:
explicit VKFence(const VKDevice& device, UniqueFence handle);
~VKFence();
/**
* Waits for the fence to be signaled.
* @warning You must have ownership of the fence and it has to be previously sent to a queue to
* call this function.
*/
void Wait();
/**
* Releases ownership of the fence. Pass after it has been sent to an execution queue.
* Unmanaged usage of the fence after the call will result in undefined behavior because it may
* be being used for something else.
*/
void Release();
/// Protects a resource with this fence.
void Protect(VKResource* resource);
/// Removes protection for a resource.
void Unprotect(const VKResource* resource);
/// Retreives the fence.
operator vk::Fence() const {
return *handle;
}
private:
/// Take ownership of the fence.
void Commit();
/**
* Updates the fence status.
* @warning Waiting for the owner might soft lock the execution.
* @param gpu_wait Wait for the fence to be signaled by the driver.
* @param owner_wait Wait for the owner to signal its freedom.
* @returns True if the fence is free. Waiting for gpu and owner will always return true.
*/
bool Tick(bool gpu_wait, bool owner_wait);
const VKDevice& device; ///< Device handler
UniqueFence handle; ///< Vulkan fence
std::vector<VKResource*> protected_resources; ///< List of resources protected by this fence
bool is_owned = false; ///< The fence has been commited but not released yet.
bool is_used = false; ///< The fence has been commited but it has not been checked to be free.
};
/**
* A fence watch is used to keep track of the usage of a fence and protect a resource or set of
* resources without having to inherit VKResource from their handlers.
*/
class VKFenceWatch final : public VKResource {
public:
explicit VKFenceWatch();
~VKFenceWatch();
/// Waits for the fence to be released.
void Wait();
/**
* Waits for a previous fence and watches a new one.
* @param new_fence New fence to wait to.
*/
void Watch(VKFence& new_fence);
/**
* Checks if it's currently being watched and starts watching it if it's available.
* @returns True if a watch has started, false if it's being watched.
*/
bool TryWatch(VKFence& new_fence);
void OnFenceRemoval(VKFence* signaling_fence) override;
private:
VKFence* fence{}; ///< Fence watching this resource. nullptr when the watch is free.
};
/**
* Handles a pool of resources protected by fences. Manages resource overflow allocating more
* resources.
*/
class VKFencedPool {
public:
explicit VKFencedPool(std::size_t grow_step);
virtual ~VKFencedPool();
protected:
/**
* Commits a free resource and protects it with a fence. It may allocate new resources.
* @param fence Fence that protects the commited resource.
* @returns Index of the resource commited.
*/
std::size_t CommitResource(VKFence& fence);
/// Called when a chunk of resources have to be allocated.
virtual void Allocate(std::size_t begin, std::size_t end) = 0;
private:
/// Manages pool overflow allocating new resources.
std::size_t ManageOverflow();
/// Allocates a new page of resources.
void Grow();
std::size_t grow_step = 0; ///< Number of new resources created after an overflow
std::size_t free_iterator = 0; ///< Hint to where the next free resources is likely to be found
std::vector<std::unique_ptr<VKFenceWatch>> watches; ///< Set of watched resources
};
/**
* The resource manager handles all resources that can be protected with a fence avoiding
* driver-side or GPU-side concurrent usage. Usage is documented in VKFence.
*/
class VKResourceManager final {
public:
explicit VKResourceManager(const VKDevice& device);
~VKResourceManager();
/// Commits a fence. It has to be sent to a queue and released.
VKFence& CommitFence();
/// Commits an unused command buffer and protects it with a fence.
vk::CommandBuffer CommitCommandBuffer(VKFence& fence);
private:
/// Allocates new fences.
void GrowFences(std::size_t new_fences_count);
const VKDevice& device; ///< Device handler.
std::size_t fences_iterator = 0; ///< Index where a free fence is likely to be found.
std::vector<std::unique_ptr<VKFence>> fences; ///< Pool of fences.
std::unique_ptr<CommandBufferPool> command_buffer_pool; ///< Pool of command buffers.
};
} // namespace Vulkan

View File

@@ -0,0 +1,60 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/logging/log.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
namespace Vulkan {
VKScheduler::VKScheduler(const VKDevice& device, VKResourceManager& resource_manager)
: device{device}, resource_manager{resource_manager} {
next_fence = &resource_manager.CommitFence();
AllocateNewContext();
}
VKScheduler::~VKScheduler() = default;
VKExecutionContext VKScheduler::GetExecutionContext() const {
return VKExecutionContext(current_fence, current_cmdbuf);
}
VKExecutionContext VKScheduler::Flush(vk::Semaphore semaphore) {
SubmitExecution(semaphore);
current_fence->Release();
AllocateNewContext();
return GetExecutionContext();
}
VKExecutionContext VKScheduler::Finish(vk::Semaphore semaphore) {
SubmitExecution(semaphore);
current_fence->Wait();
current_fence->Release();
AllocateNewContext();
return GetExecutionContext();
}
void VKScheduler::SubmitExecution(vk::Semaphore semaphore) {
const auto& dld = device.GetDispatchLoader();
current_cmdbuf.end(dld);
const auto queue = device.GetGraphicsQueue();
const vk::SubmitInfo submit_info(0, nullptr, nullptr, 1, &current_cmdbuf, semaphore ? 1u : 0u,
&semaphore);
queue.submit({submit_info}, *current_fence, dld);
}
void VKScheduler::AllocateNewContext() {
current_fence = next_fence;
current_cmdbuf = resource_manager.CommitCommandBuffer(*current_fence);
next_fence = &resource_manager.CommitFence();
const auto& dld = device.GetDispatchLoader();
current_cmdbuf.begin({vk::CommandBufferUsageFlagBits::eOneTimeSubmit}, dld);
}
} // namespace Vulkan

View File

@@ -0,0 +1,69 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
namespace Vulkan {
class VKDevice;
class VKExecutionContext;
class VKFence;
class VKResourceManager;
/// The scheduler abstracts command buffer and fence management with an interface that's able to do
/// OpenGL-like operations on Vulkan command buffers.
class VKScheduler {
public:
explicit VKScheduler(const VKDevice& device, VKResourceManager& resource_manager);
~VKScheduler();
/// Gets the current execution context.
[[nodiscard]] VKExecutionContext GetExecutionContext() const;
/// Sends the current execution context to the GPU. It invalidates the current execution context
/// and returns a new one.
VKExecutionContext Flush(vk::Semaphore semaphore = nullptr);
/// Sends the current execution context to the GPU and waits for it to complete. It invalidates
/// the current execution context and returns a new one.
VKExecutionContext Finish(vk::Semaphore semaphore = nullptr);
private:
void SubmitExecution(vk::Semaphore semaphore);
void AllocateNewContext();
const VKDevice& device;
VKResourceManager& resource_manager;
vk::CommandBuffer current_cmdbuf;
VKFence* current_fence = nullptr;
VKFence* next_fence = nullptr;
};
class VKExecutionContext {
friend class VKScheduler;
public:
VKExecutionContext() = default;
VKFence& GetFence() const {
return *fence;
}
vk::CommandBuffer GetCommandBuffer() const {
return cmdbuf;
}
private:
explicit VKExecutionContext(VKFence* fence, vk::CommandBuffer cmdbuf)
: fence{fence}, cmdbuf{cmdbuf} {}
VKFence* fence{};
vk::CommandBuffer cmdbuf;
};
} // namespace Vulkan