summaryrefslogtreecommitdiffstats
path: root/src/video_core/gpu_thread.cpp
blob: 56b9621b18149988ab4abd0d2a428b40414836f3 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
// 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/microprofile.h"
#include "common/thread.h"
#include "core/core.h"
#include "core/frontend/emu_window.h"
#include "core/settings.h"
#include "video_core/dma_pusher.h"
#include "video_core/gpu.h"
#include "video_core/gpu_thread.h"
#include "video_core/renderer_base.h"

namespace VideoCommon::GPUThread {

/// Runs the GPU thread
static void RunThread(Core::System& system, VideoCore::RendererBase& renderer,
                      Core::Frontend::GraphicsContext& context, Tegra::DmaPusher& dma_pusher,
                      SynchState& state, Tegra::CDmaPusher& cdma_pusher) {
    std::string name = "yuzu:GPU";
    MicroProfileOnThreadCreate(name.c_str());
    Common::SetCurrentThreadName(name.c_str());
    Common::SetCurrentThreadPriority(Common::ThreadPriority::High);
    system.RegisterHostThread();

    // Wait for first GPU command before acquiring the window context
    while (state.queue.Empty())
        ;

    // If emulation was stopped during disk shader loading, abort before trying to acquire context
    if (!state.is_running) {
        return;
    }

    auto current_context = context.Acquire();

    CommandDataContainer next;
    while (state.is_running) {
        next = state.queue.PopWait();
        if (auto* submit_list = std::get_if<SubmitListCommand>(&next.data)) {
            dma_pusher.Push(std::move(submit_list->entries));
            dma_pusher.DispatchCalls();
        } else if (auto* command_list = std::get_if<SubmitChCommandEntries>(&next.data)) {
            // NVDEC
            cdma_pusher.Push(std::move(command_list->entries));
            cdma_pusher.DispatchCalls();
        } else if (const auto* data = std::get_if<SwapBuffersCommand>(&next.data)) {
            renderer.SwapBuffers(data->framebuffer ? &*data->framebuffer : nullptr);
        } else if (std::holds_alternative<OnCommandListEndCommand>(next.data)) {
            renderer.Rasterizer().ReleaseFences();
        } else if (std::holds_alternative<GPUTickCommand>(next.data)) {
            system.GPU().TickWork();
        } else if (const auto* flush = std::get_if<FlushRegionCommand>(&next.data)) {
            renderer.Rasterizer().FlushRegion(flush->addr, flush->size);
        } else if (const auto* invalidate = std::get_if<InvalidateRegionCommand>(&next.data)) {
            renderer.Rasterizer().OnCPUWrite(invalidate->addr, invalidate->size);
        } else if (std::holds_alternative<EndProcessingCommand>(next.data)) {
            return;
        } else {
            UNREACHABLE();
        }
        state.signaled_fence.store(next.fence);
    }
}

ThreadManager::ThreadManager(Core::System& system_, bool is_async_)
    : system{system_}, is_async{is_async_} {}

ThreadManager::~ThreadManager() {
    if (!thread.joinable()) {
        return;
    }

    // Notify GPU thread that a shutdown is pending
    PushCommand(EndProcessingCommand());
    thread.join();
}

void ThreadManager::StartThread(VideoCore::RendererBase& renderer,
                                Core::Frontend::GraphicsContext& context,
                                Tegra::DmaPusher& dma_pusher, Tegra::CDmaPusher& cdma_pusher) {
    thread = std::thread(RunThread, std::ref(system), std::ref(renderer), std::ref(context),
                         std::ref(dma_pusher), std::ref(state), std::ref(cdma_pusher));
}

void ThreadManager::SubmitList(Tegra::CommandList&& entries) {
    PushCommand(SubmitListCommand(std::move(entries)));
}

void ThreadManager::SubmitCommandBuffer(Tegra::ChCommandHeaderList&& entries) {
    PushCommand(SubmitChCommandEntries(std::move(entries)));
}

void ThreadManager::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
    PushCommand(SwapBuffersCommand(framebuffer ? std::make_optional(*framebuffer) : std::nullopt));
}

void ThreadManager::FlushRegion(VAddr addr, u64 size) {
    if (!is_async) {
        // Always flush with synchronous GPU mode
        PushCommand(FlushRegionCommand(addr, size));
        return;
    }

    // Asynchronous GPU mode
    switch (Settings::values.gpu_accuracy.GetValue()) {
    case Settings::GPUAccuracy::Normal:
        PushCommand(FlushRegionCommand(addr, size));
        break;
    case Settings::GPUAccuracy::High:
        // TODO(bunnei): Is this right? Preserving existing behavior for now
        break;
    case Settings::GPUAccuracy::Extreme: {
        auto& gpu = system.GPU();
        u64 fence = gpu.RequestFlush(addr, size);
        PushCommand(GPUTickCommand());
        while (fence > gpu.CurrentFlushRequestFence()) {
        }
        break;
    }
    default:
        UNIMPLEMENTED_MSG("Unsupported gpu_accuracy {}", Settings::values.gpu_accuracy.GetValue());
    }
}

void ThreadManager::InvalidateRegion(VAddr addr, u64 size) {
    system.Renderer().Rasterizer().OnCPUWrite(addr, size);
}

void ThreadManager::FlushAndInvalidateRegion(VAddr addr, u64 size) {
    // Skip flush on asynch mode, as FlushAndInvalidateRegion is not used for anything too important
    system.Renderer().Rasterizer().OnCPUWrite(addr, size);
}

void ThreadManager::WaitIdle() const {
    while (state.last_fence > state.signaled_fence.load(std::memory_order_relaxed)) {
    }
}

void ThreadManager::OnCommandListEnd() {
    PushCommand(OnCommandListEndCommand());
}

u64 ThreadManager::PushCommand(CommandData&& command_data) {
    const u64 fence{++state.last_fence};
    state.queue.Push(CommandDataContainer(std::move(command_data), fence));

    if (!is_async) {
        // In synchronous GPU mode, block the caller until the command has executed
        WaitIdle();
    }

    return fence;
}

} // namespace VideoCommon::GPUThread