summaryrefslogtreecommitdiffstats
path: root/src/video_core
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core')
-rw-r--r--src/video_core/engines/maxwell_3d.cpp5
-rw-r--r--src/video_core/gpu.cpp48
-rw-r--r--src/video_core/gpu.h34
-rw-r--r--src/video_core/gpu_asynch.cpp9
-rw-r--r--src/video_core/gpu_asynch.h3
-rw-r--r--src/video_core/gpu_synch.cpp2
-rw-r--r--src/video_core/gpu_synch.h4
-rw-r--r--src/video_core/gpu_thread.cpp27
-rw-r--r--src/video_core/gpu_thread.h32
9 files changed, 105 insertions, 59 deletions
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index 74c46ec04..125c53360 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -525,8 +525,9 @@ void Maxwell3D::ProcessSyncPoint() {
const u32 sync_point = regs.sync_info.sync_point.Value();
const u32 increment = regs.sync_info.increment.Value();
const u32 cache_flush = regs.sync_info.unknown.Value();
- LOG_DEBUG(HW_GPU, "Syncpoint set {}, increment: {}, unk: {}", sync_point, increment,
- cache_flush);
+ if (increment) {
+ system.GPU().IncrementSyncPoint(sync_point);
+ }
}
void Maxwell3D::DrawArrays() {
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index 21007d8b2..1622332a4 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -29,7 +29,8 @@ u32 FramebufferConfig::BytesPerPixel(PixelFormat format) {
UNREACHABLE();
}
-GPU::GPU(Core::System& system, VideoCore::RendererBase& renderer) : renderer{renderer} {
+GPU::GPU(Core::System& system, VideoCore::RendererBase& renderer, bool is_async)
+ : system{system}, renderer{renderer}, is_async{is_async} {
auto& rasterizer{renderer.Rasterizer()};
memory_manager = std::make_unique<Tegra::MemoryManager>(system, rasterizer);
dma_pusher = std::make_unique<Tegra::DmaPusher>(*this);
@@ -74,6 +75,51 @@ const DmaPusher& GPU::DmaPusher() const {
return *dma_pusher;
}
+void GPU::IncrementSyncPoint(const u32 syncpoint_id) {
+ syncpoints[syncpoint_id]++;
+ std::lock_guard lock{sync_mutex};
+ if (!syncpt_interrupts[syncpoint_id].empty()) {
+ u32 value = syncpoints[syncpoint_id].load();
+ auto it = syncpt_interrupts[syncpoint_id].begin();
+ while (it != syncpt_interrupts[syncpoint_id].end()) {
+ if (value >= *it) {
+ TriggerCpuInterrupt(syncpoint_id, *it);
+ it = syncpt_interrupts[syncpoint_id].erase(it);
+ continue;
+ }
+ it++;
+ }
+ }
+}
+
+u32 GPU::GetSyncpointValue(const u32 syncpoint_id) const {
+ return syncpoints[syncpoint_id].load();
+}
+
+void GPU::RegisterSyncptInterrupt(const u32 syncpoint_id, const u32 value) {
+ auto& interrupt = syncpt_interrupts[syncpoint_id];
+ bool contains = std::any_of(interrupt.begin(), interrupt.end(),
+ [value](u32 in_value) { return in_value == value; });
+ if (contains) {
+ return;
+ }
+ syncpt_interrupts[syncpoint_id].emplace_back(value);
+}
+
+bool GPU::CancelSyncptInterrupt(const u32 syncpoint_id, const u32 value) {
+ std::lock_guard lock{sync_mutex};
+ auto& interrupt = syncpt_interrupts[syncpoint_id];
+ const auto iter =
+ std::find_if(interrupt.begin(), interrupt.end(),
+ [value](u32 interrupt_value) { return value == interrupt_value; });
+
+ if (iter == interrupt.end()) {
+ return false;
+ }
+ interrupt.erase(iter);
+ return true;
+}
+
u32 RenderTargetBytesPerPixel(RenderTargetFormat format) {
ASSERT(format != RenderTargetFormat::NONE);
diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h
index 0055e5326..87c96f46b 100644
--- a/src/video_core/gpu.h
+++ b/src/video_core/gpu.h
@@ -5,8 +5,12 @@
#pragma once
#include <array>
+#include <atomic>
+#include <list>
#include <memory>
+#include <mutex>
#include "common/common_types.h"
+#include "core/hle/service/nvdrv/nvdata.h"
#include "core/hle/service/nvflinger/buffer_queue.h"
#include "video_core/dma_pusher.h"
@@ -127,7 +131,7 @@ class MemoryManager;
class GPU {
public:
- explicit GPU(Core::System& system, VideoCore::RendererBase& renderer);
+ explicit GPU(Core::System& system, VideoCore::RendererBase& renderer, bool is_async);
virtual ~GPU();
@@ -170,6 +174,22 @@ public:
/// Returns a reference to the GPU DMA pusher.
Tegra::DmaPusher& DmaPusher();
+ void IncrementSyncPoint(u32 syncpoint_id);
+
+ u32 GetSyncpointValue(u32 syncpoint_id) const;
+
+ void RegisterSyncptInterrupt(u32 syncpoint_id, u32 value);
+
+ bool CancelSyncptInterrupt(u32 syncpoint_id, u32 value);
+
+ std::unique_lock<std::mutex> LockSync() {
+ return std::unique_lock{sync_mutex};
+ }
+
+ bool IsAsync() const {
+ return is_async;
+ }
+
/// Returns a const reference to the GPU DMA pusher.
const Tegra::DmaPusher& DmaPusher() const;
@@ -239,6 +259,9 @@ public:
/// Notify rasterizer that any caches of the specified region should be flushed and invalidated
virtual void FlushAndInvalidateRegion(CacheAddr addr, u64 size) = 0;
+protected:
+ virtual void TriggerCpuInterrupt(u32 syncpoint_id, u32 value) const = 0;
+
private:
void ProcessBindMethod(const MethodCall& method_call);
void ProcessSemaphoreTriggerMethod();
@@ -257,6 +280,7 @@ private:
protected:
std::unique_ptr<Tegra::DmaPusher> dma_pusher;
VideoCore::RendererBase& renderer;
+ Core::System& system;
private:
std::unique_ptr<Tegra::MemoryManager> memory_manager;
@@ -273,6 +297,14 @@ private:
std::unique_ptr<Engines::MaxwellDMA> maxwell_dma;
/// Inline memory engine
std::unique_ptr<Engines::KeplerMemory> kepler_memory;
+
+ std::array<std::atomic<u32>, Service::Nvidia::MaxSyncPoints> syncpoints{};
+
+ std::array<std::list<u32>, Service::Nvidia::MaxSyncPoints> syncpt_interrupts;
+
+ std::mutex sync_mutex;
+
+ const bool is_async;
};
#define ASSERT_REG_POSITION(field_name, position) \
diff --git a/src/video_core/gpu_asynch.cpp b/src/video_core/gpu_asynch.cpp
index d4e2553a9..ea67be831 100644
--- a/src/video_core/gpu_asynch.cpp
+++ b/src/video_core/gpu_asynch.cpp
@@ -2,6 +2,8 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
+#include "core/core.h"
+#include "core/hardware_interrupt_manager.h"
#include "video_core/gpu_asynch.h"
#include "video_core/gpu_thread.h"
#include "video_core/renderer_base.h"
@@ -9,7 +11,7 @@
namespace VideoCommon {
GPUAsynch::GPUAsynch(Core::System& system, VideoCore::RendererBase& renderer)
- : GPU(system, renderer), gpu_thread{system} {}
+ : GPU(system, renderer, true), gpu_thread{system} {}
GPUAsynch::~GPUAsynch() = default;
@@ -38,4 +40,9 @@ void GPUAsynch::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
gpu_thread.FlushAndInvalidateRegion(addr, size);
}
+void GPUAsynch::TriggerCpuInterrupt(const u32 syncpoint_id, const u32 value) const {
+ auto& interrupt_manager = system.InterruptManager();
+ interrupt_manager.GPUInterruptSyncpt(syncpoint_id, value);
+}
+
} // namespace VideoCommon
diff --git a/src/video_core/gpu_asynch.h b/src/video_core/gpu_asynch.h
index 30be74cba..36377d677 100644
--- a/src/video_core/gpu_asynch.h
+++ b/src/video_core/gpu_asynch.h
@@ -27,6 +27,9 @@ public:
void InvalidateRegion(CacheAddr addr, u64 size) override;
void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
+protected:
+ void TriggerCpuInterrupt(u32 syncpoint_id, u32 value) const override;
+
private:
GPUThread::ThreadManager gpu_thread;
};
diff --git a/src/video_core/gpu_synch.cpp b/src/video_core/gpu_synch.cpp
index 45e43b1dc..d4ead9c47 100644
--- a/src/video_core/gpu_synch.cpp
+++ b/src/video_core/gpu_synch.cpp
@@ -8,7 +8,7 @@
namespace VideoCommon {
GPUSynch::GPUSynch(Core::System& system, VideoCore::RendererBase& renderer)
- : GPU(system, renderer) {}
+ : GPU(system, renderer, false) {}
GPUSynch::~GPUSynch() = default;
diff --git a/src/video_core/gpu_synch.h b/src/video_core/gpu_synch.h
index 3031fcf72..07bcc47f1 100644
--- a/src/video_core/gpu_synch.h
+++ b/src/video_core/gpu_synch.h
@@ -25,6 +25,10 @@ public:
void FlushRegion(CacheAddr addr, u64 size) override;
void InvalidateRegion(CacheAddr addr, u64 size) override;
void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
+
+protected:
+ void TriggerCpuInterrupt([[maybe_unused]] u32 syncpoint_id,
+ [[maybe_unused]] u32 value) const override {}
};
} // namespace VideoCommon
diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp
index 3f0939ec9..b441e92b0 100644
--- a/src/video_core/gpu_thread.cpp
+++ b/src/video_core/gpu_thread.cpp
@@ -21,7 +21,8 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p
MicroProfileOnThreadCreate("GpuThread");
// Wait for first GPU command before acquiring the window context
- state.WaitForCommands();
+ while (state.queue.Empty())
+ ;
// If emulation was stopped during disk shader loading, abort before trying to acquire context
if (!state.is_running) {
@@ -32,7 +33,6 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p
CommandDataContainer next;
while (state.is_running) {
- state.WaitForCommands();
while (!state.queue.Empty()) {
state.queue.Pop(next);
if (const auto submit_list = std::get_if<SubmitListCommand>(&next.data)) {
@@ -49,8 +49,7 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p
} else {
UNREACHABLE();
}
- state.signaled_fence = next.fence;
- state.TrySynchronize();
+ state.signaled_fence.store(next.fence);
}
}
}
@@ -89,12 +88,7 @@ void ThreadManager::FlushRegion(CacheAddr addr, u64 size) {
}
void ThreadManager::InvalidateRegion(CacheAddr addr, u64 size) {
- if (state.queue.Empty()) {
- // It's quicker to invalidate a single region on the CPU if the queue is already empty
- system.Renderer().Rasterizer().InvalidateRegion(addr, size);
- } else {
- PushCommand(InvalidateRegionCommand(addr, size));
- }
+ system.Renderer().Rasterizer().InvalidateRegion(addr, size);
}
void ThreadManager::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
@@ -105,22 +99,13 @@ void ThreadManager::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
u64 ThreadManager::PushCommand(CommandData&& command_data) {
const u64 fence{++state.last_fence};
state.queue.Push(CommandDataContainer(std::move(command_data), fence));
- state.SignalCommands();
return fence;
}
MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192));
void SynchState::WaitForSynchronization(u64 fence) {
- if (signaled_fence >= fence) {
- return;
- }
-
- // Wait for the GPU to be idle (all commands to be executed)
- {
- MICROPROFILE_SCOPE(GPU_wait);
- std::unique_lock lock{synchronization_mutex};
- synchronization_condition.wait(lock, [this, fence] { return signaled_fence >= fence; });
- }
+ while (signaled_fence.load() < fence)
+ ;
}
} // namespace VideoCommon::GPUThread
diff --git a/src/video_core/gpu_thread.h b/src/video_core/gpu_thread.h
index 05a168a72..1d9d0c39e 100644
--- a/src/video_core/gpu_thread.h
+++ b/src/video_core/gpu_thread.h
@@ -88,41 +88,9 @@ struct CommandDataContainer {
/// Struct used to synchronize the GPU thread
struct SynchState final {
std::atomic_bool is_running{true};
- std::atomic_int queued_frame_count{};
- std::mutex synchronization_mutex;
- std::mutex commands_mutex;
- std::condition_variable commands_condition;
- std::condition_variable synchronization_condition;
-
- /// Returns true if the gap in GPU commands is small enough that we can consider the CPU and GPU
- /// synchronized. This is entirely empirical.
- bool IsSynchronized() const {
- constexpr std::size_t max_queue_gap{5};
- return queue.Size() <= max_queue_gap;
- }
-
- void TrySynchronize() {
- if (IsSynchronized()) {
- std::lock_guard lock{synchronization_mutex};
- synchronization_condition.notify_one();
- }
- }
void WaitForSynchronization(u64 fence);
- void SignalCommands() {
- if (queue.Empty()) {
- return;
- }
-
- commands_condition.notify_one();
- }
-
- void WaitForCommands() {
- std::unique_lock lock{commands_mutex};
- commands_condition.wait(lock, [this] { return !queue.Empty(); });
- }
-
using CommandQueue = Common::SPSCQueue<CommandDataContainer>;
CommandQueue queue;
u64 last_fence{};
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-04 03:42:57 +0200