1 files changed, 163 insertions, 48 deletions

diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index 8acf2eda2..ebd149c3a 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -2,6 +2,8 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <chrono>
+
 #include "common/assert.h"
 #include "common/microprofile.h"
 #include "core/core.h"
@@ -9,6 +11,7 @@
 #include "core/core_timing_util.h"
 #include "core/frontend/emu_window.h"
 #include "core/memory.h"
+#include "core/settings.h"
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/engines/kepler_compute.h"
 #include "video_core/engines/kepler_memory.h"
@@ -17,26 +20,36 @@
 #include "video_core/gpu.h"
 #include "video_core/memory_manager.h"
 #include "video_core/renderer_base.h"
+#include "video_core/shader_notify.h"
 #include "video_core/video_core.h"
 
 namespace Tegra {
 
 MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192));
 
-GPU::GPU(Core::System& system, std::unique_ptr<VideoCore::RendererBase>&& renderer_, bool is_async)
-    : system{system}, renderer{std::move(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);
-    maxwell_3d = std::make_unique<Engines::Maxwell3D>(system, rasterizer, *memory_manager);
-    fermi_2d = std::make_unique<Engines::Fermi2D>(rasterizer);
-    kepler_compute = std::make_unique<Engines::KeplerCompute>(system, rasterizer, *memory_manager);
-    maxwell_dma = std::make_unique<Engines::MaxwellDMA>(system, *memory_manager);
-    kepler_memory = std::make_unique<Engines::KeplerMemory>(system, *memory_manager);
-}
+GPU::GPU(Core::System& system_, bool is_async_, bool use_nvdec_)
+    : system{system_}, memory_manager{std::make_unique<Tegra::MemoryManager>(system)},
+      dma_pusher{std::make_unique<Tegra::DmaPusher>(system, *this)},
+      cdma_pusher{std::make_unique<Tegra::CDmaPusher>(*this)}, use_nvdec{use_nvdec_},
+      maxwell_3d{std::make_unique<Engines::Maxwell3D>(system, *memory_manager)},
+      fermi_2d{std::make_unique<Engines::Fermi2D>()},
+      kepler_compute{std::make_unique<Engines::KeplerCompute>(system, *memory_manager)},
+      maxwell_dma{std::make_unique<Engines::MaxwellDMA>(system, *memory_manager)},
+      kepler_memory{std::make_unique<Engines::KeplerMemory>(system, *memory_manager)},
+      shader_notify{std::make_unique<VideoCore::ShaderNotify>()}, is_async{is_async_} {}
 
 GPU::~GPU() = default;
 
+void GPU::BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer_) {
+    renderer = std::move(renderer_);
+
+    VideoCore::RasterizerInterface& rasterizer = renderer->Rasterizer();
+    memory_manager->BindRasterizer(rasterizer);
+    maxwell_3d->BindRasterizer(rasterizer);
+    fermi_2d->BindRasterizer(rasterizer);
+    kepler_compute->BindRasterizer(rasterizer);
+}
+
 Engines::Maxwell3D& GPU::Maxwell3D() {
     return *maxwell_3d;
 }
@@ -65,10 +78,18 @@ DmaPusher& GPU::DmaPusher() {
     return *dma_pusher;
 }
 
+Tegra::CDmaPusher& GPU::CDmaPusher() {
+    return *cdma_pusher;
+}
+
 const DmaPusher& GPU::DmaPusher() const {
     return *dma_pusher;
 }
 
+const Tegra::CDmaPusher& GPU::CDmaPusher() const {
+    return *cdma_pusher;
+}
+
 void GPU::WaitFence(u32 syncpoint_id, u32 value) {
     // Synced GPU, is always in sync
     if (!is_async) {
@@ -76,7 +97,7 @@ void GPU::WaitFence(u32 syncpoint_id, u32 value) {
     }
     MICROPROFILE_SCOPE(GPU_wait);
     std::unique_lock lock{sync_mutex};
-    sync_cv.wait(lock, [=]() { return syncpoints[syncpoint_id].load() >= value; });
+    sync_cv.wait(lock, [=, this] { return syncpoints[syncpoint_id].load() >= value; });
 }
 
 void GPU::IncrementSyncPoint(const u32 syncpoint_id) {
@@ -125,14 +146,38 @@ bool GPU::CancelSyncptInterrupt(const u32 syncpoint_id, const u32 value) {
     return true;
 }
 
+u64 GPU::RequestFlush(VAddr addr, std::size_t size) {
+    std::unique_lock lck{flush_request_mutex};
+    const u64 fence = ++last_flush_fence;
+    flush_requests.emplace_back(fence, addr, size);
+    return fence;
+}
+
+void GPU::TickWork() {
+    std::unique_lock lck{flush_request_mutex};
+    while (!flush_requests.empty()) {
+        auto& request = flush_requests.front();
+        const u64 fence = request.fence;
+        const VAddr addr = request.addr;
+        const std::size_t size = request.size;
+        flush_requests.pop_front();
+        flush_request_mutex.unlock();
+        renderer->Rasterizer().FlushRegion(addr, size);
+        current_flush_fence.store(fence);
+        flush_request_mutex.lock();
+    }
+}
+
 u64 GPU::GetTicks() const {
     // This values were reversed engineered by fincs from NVN
     // The gpu clock is reported in units of 385/625 nanoseconds
     constexpr u64 gpu_ticks_num = 384;
     constexpr u64 gpu_ticks_den = 625;
 
-    const u64 cpu_ticks = system.CoreTiming().GetTicks();
-    const u64 nanoseconds = Core::Timing::CyclesToNs(cpu_ticks).count();
+    u64 nanoseconds = system.CoreTiming().GetGlobalTimeNs().count();
+    if (Settings::values.use_fast_gpu_time.GetValue()) {
+        nanoseconds /= 256;
+    }
     const u64 nanoseconds_num = nanoseconds / gpu_ticks_den;
     const u64 nanoseconds_rem = nanoseconds % gpu_ticks_den;
     return nanoseconds_num * gpu_ticks_num + (nanoseconds_rem * gpu_ticks_num) / gpu_ticks_den;
@@ -142,30 +187,13 @@ void GPU::FlushCommands() {
     renderer->Rasterizer().FlushCommands();
 }
 
-// Note that, traditionally, methods are treated as 4-byte addressable locations, and hence
-// their numbers are written down multiplied by 4 in Docs. Here we are not multiply by 4.
-// So the values you see in docs might be multiplied by 4.
-enum class BufferMethods {
-    BindObject = 0x0,
-    Nop = 0x2,
-    SemaphoreAddressHigh = 0x4,
-    SemaphoreAddressLow = 0x5,
-    SemaphoreSequence = 0x6,
-    SemaphoreTrigger = 0x7,
-    NotifyIntr = 0x8,
-    WrcacheFlush = 0x9,
-    Unk28 = 0xA,
-    UnkCacheFlush = 0xB,
-    RefCnt = 0x14,
-    SemaphoreAcquire = 0x1A,
-    SemaphoreRelease = 0x1B,
-    FenceValue = 0x1C,
-    FenceAction = 0x1D,
-    Unk78 = 0x1E,
-    Unk7c = 0x1F,
-    Yield = 0x20,
-    NonPullerMethods = 0x40,
-};
+void GPU::SyncGuestHost() {
+    renderer->Rasterizer().SyncGuestHost();
+}
+
+void GPU::OnCommandListEnd() {
+    renderer->Rasterizer().ReleaseFences();
+}
 
 enum class GpuSemaphoreOperation {
     AcquireEqual = 0x1,
@@ -180,16 +208,32 @@ void GPU::CallMethod(const MethodCall& method_call) {
 
     ASSERT(method_call.subchannel < bound_engines.size());
 
-    if (ExecuteMethodOnEngine(method_call)) {
+    if (ExecuteMethodOnEngine(method_call.method)) {
         CallEngineMethod(method_call);
     } else {
         CallPullerMethod(method_call);
     }
 }
 
-bool GPU::ExecuteMethodOnEngine(const MethodCall& method_call) {
-    const auto method = static_cast<BufferMethods>(method_call.method);
-    return method >= BufferMethods::NonPullerMethods;
+void GPU::CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount,
+                          u32 methods_pending) {
+    LOG_TRACE(HW_GPU, "Processing method {:08X} on subchannel {}", method, subchannel);
+
+    ASSERT(subchannel < bound_engines.size());
+
+    if (ExecuteMethodOnEngine(method)) {
+        CallEngineMultiMethod(method, subchannel, base_start, amount, methods_pending);
+    } else {
+        for (std::size_t i = 0; i < amount; i++) {
+            CallPullerMethod(
+                {method, base_start[i], subchannel, methods_pending - static_cast<u32>(i)});
+        }
+    }
+}
+
+bool GPU::ExecuteMethodOnEngine(u32 method) {
+    const auto buffer_method = static_cast<BufferMethods>(method);
+    return buffer_method >= BufferMethods::NonPullerMethods;
 }
 
 void GPU::CallPullerMethod(const MethodCall& method_call) {
@@ -209,7 +253,12 @@ void GPU::CallPullerMethod(const MethodCall& method_call) {
     case BufferMethods::UnkCacheFlush:
     case BufferMethods::WrcacheFlush:
     case BufferMethods::FenceValue:
+        break;
     case BufferMethods::FenceAction:
+        ProcessFenceActionMethod();
+        break;
+    case BufferMethods::WaitForInterrupt:
+        ProcessWaitForInterruptMethod();
         break;
     case BufferMethods::SemaphoreTrigger: {
         ProcessSemaphoreTriggerMethod();
@@ -250,19 +299,46 @@ void GPU::CallEngineMethod(const MethodCall& method_call) {
 
     switch (engine) {
     case EngineID::FERMI_TWOD_A:
-        fermi_2d->CallMethod(method_call);
+        fermi_2d->CallMethod(method_call.method, method_call.argument, method_call.IsLastCall());
+        break;
+    case EngineID::MAXWELL_B:
+        maxwell_3d->CallMethod(method_call.method, method_call.argument, method_call.IsLastCall());
+        break;
+    case EngineID::KEPLER_COMPUTE_B:
+        kepler_compute->CallMethod(method_call.method, method_call.argument,
+                                   method_call.IsLastCall());
+        break;
+    case EngineID::MAXWELL_DMA_COPY_A:
+        maxwell_dma->CallMethod(method_call.method, method_call.argument, method_call.IsLastCall());
+        break;
+    case EngineID::KEPLER_INLINE_TO_MEMORY_B:
+        kepler_memory->CallMethod(method_call.method, method_call.argument,
+                                  method_call.IsLastCall());
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented engine");
+    }
+}
+
+void GPU::CallEngineMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount,
+                                u32 methods_pending) {
+    const EngineID engine = bound_engines[subchannel];
+
+    switch (engine) {
+    case EngineID::FERMI_TWOD_A:
+        fermi_2d->CallMultiMethod(method, base_start, amount, methods_pending);
         break;
     case EngineID::MAXWELL_B:
-        maxwell_3d->CallMethod(method_call);
+        maxwell_3d->CallMultiMethod(method, base_start, amount, methods_pending);
         break;
     case EngineID::KEPLER_COMPUTE_B:
-        kepler_compute->CallMethod(method_call);
+        kepler_compute->CallMultiMethod(method, base_start, amount, methods_pending);
         break;
     case EngineID::MAXWELL_DMA_COPY_A:
-        maxwell_dma->CallMethod(method_call);
+        maxwell_dma->CallMultiMethod(method, base_start, amount, methods_pending);
         break;
     case EngineID::KEPLER_INLINE_TO_MEMORY_B:
-        kepler_memory->CallMethod(method_call);
+        kepler_memory->CallMultiMethod(method, base_start, amount, methods_pending);
         break;
     default:
         UNIMPLEMENTED_MSG("Unimplemented engine");
@@ -273,7 +349,46 @@ void GPU::ProcessBindMethod(const MethodCall& method_call) {
     // Bind the current subchannel to the desired engine id.
     LOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {}", method_call.subchannel,
               method_call.argument);
-    bound_engines[method_call.subchannel] = static_cast<EngineID>(method_call.argument);
+    const auto engine_id = static_cast<EngineID>(method_call.argument);
+    bound_engines[method_call.subchannel] = static_cast<EngineID>(engine_id);
+    switch (engine_id) {
+    case EngineID::FERMI_TWOD_A:
+        dma_pusher->BindSubchannel(fermi_2d.get(), method_call.subchannel);
+        break;
+    case EngineID::MAXWELL_B:
+        dma_pusher->BindSubchannel(maxwell_3d.get(), method_call.subchannel);
+        break;
+    case EngineID::KEPLER_COMPUTE_B:
+        dma_pusher->BindSubchannel(kepler_compute.get(), method_call.subchannel);
+        break;
+    case EngineID::MAXWELL_DMA_COPY_A:
+        dma_pusher->BindSubchannel(maxwell_dma.get(), method_call.subchannel);
+        break;
+    case EngineID::KEPLER_INLINE_TO_MEMORY_B:
+        dma_pusher->BindSubchannel(kepler_memory.get(), method_call.subchannel);
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented engine {:04X}", static_cast<u32>(engine_id));
+    }
+}
+
+void GPU::ProcessFenceActionMethod() {
+    switch (regs.fence_action.op) {
+    case FenceOperation::Acquire:
+        WaitFence(regs.fence_action.syncpoint_id, regs.fence_value);
+        break;
+    case FenceOperation::Increment:
+        IncrementSyncPoint(regs.fence_action.syncpoint_id);
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented operation {}",
+                          static_cast<u32>(regs.fence_action.op.Value()));
+    }
+}
+
+void GPU::ProcessWaitForInterruptMethod() {
+    // TODO(bunnei) ImplementMe
+    LOG_WARNING(HW_GPU, "(STUBBED) called");
 }
 
 void GPU::ProcessSemaphoreTriggerMethod() {