Support multiple videos playing

src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp

@@ -21,7 +21,7 @@ NvResult nvhost_nvdec::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>&
     case 0x0:
         switch (command.cmd) {
         case 0x1:
-            return Submit(input, output);
+            return Submit(fd, input, output);
         case 0x2:
             return GetSyncpoint(input, output);
         case 0x3:
@@ -62,11 +62,16 @@ NvResult nvhost_nvdec::Ioctl3(DeviceFD fd, Ioctl command, const std::vector<u8>&
     return NvResult::NotImplemented;
 }
 
-void nvhost_nvdec::OnOpen(DeviceFD fd) {}
+void nvhost_nvdec::OnOpen(DeviceFD fd) {
+    static u32 next_id{};
+    fd_to_id[fd] = next_id++;
+}
 
 void nvhost_nvdec::OnClose(DeviceFD fd) {
     LOG_INFO(Service_NVDRV, "NVDEC video stream ended");
-    system.GPU().ClearCdmaInstance();
+    if (fd_to_id.find(fd) != fd_to_id.end()) {
+        system.GPU().ClearCdmaInstance(fd_to_id[fd]);
+    }
 }
 
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp

@@ -59,7 +59,8 @@ NvResult nvhost_nvdec_common::SetNVMAPfd(const std::vector<u8>& input) {
     return NvResult::Success;
 }
 
-NvResult nvhost_nvdec_common::Submit(const std::vector<u8>& input, std::vector<u8>& output) {
+NvResult nvhost_nvdec_common::Submit(DeviceFD fd, const std::vector<u8>& input,
+                                     std::vector<u8>& output) {
     IoctlSubmit params{};
     std::memcpy(&params, input.data(), sizeof(IoctlSubmit));
     LOG_DEBUG(Service_NVDRV, "called NVDEC Submit, cmd_buffer_count={}", params.cmd_buffer_count);
@@ -93,7 +94,7 @@ NvResult nvhost_nvdec_common::Submit(const std::vector<u8>& input, std::vector<u
         Tegra::ChCommandHeaderList cmdlist(cmd_buffer.word_count);
         system.Memory().ReadBlock(object->addr + cmd_buffer.offset, cmdlist.data(),
                                   cmdlist.size() * sizeof(u32));
-        gpu.PushCommandBuffer(cmdlist);
+        gpu.PushCommandBuffer(fd_to_id[fd], cmdlist);
     }
     std::memcpy(output.data(), &params, sizeof(IoctlSubmit));
     // Some games expect command_buffers to be written back

src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h

@@ -104,13 +104,14 @@ protected:
 
     /// Ioctl command implementations
     NvResult SetNVMAPfd(const std::vector<u8>& input);
-    NvResult Submit(const std::vector<u8>& input, std::vector<u8>& output);
+    NvResult Submit(DeviceFD fd, const std::vector<u8>& input, std::vector<u8>& output);
     NvResult GetSyncpoint(const std::vector<u8>& input, std::vector<u8>& output);
     NvResult GetWaitbase(const std::vector<u8>& input, std::vector<u8>& output);
     NvResult MapBuffer(const std::vector<u8>& input, std::vector<u8>& output);
     NvResult UnmapBuffer(const std::vector<u8>& input, std::vector<u8>& output);
     NvResult SetSubmitTimeout(const std::vector<u8>& input, std::vector<u8>& output);
 
+    std::unordered_map<DeviceFD, u32> fd_to_id{};
     s32_le nvmap_fd{};
     u32_le submit_timeout{};
     std::shared_ptr<nvmap> nvmap_dev;

src/core/hle/service/nvdrv/devices/nvhost_vic.cpp

@@ -21,7 +21,7 @@ NvResult nvhost_vic::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& i
     case 0x0:
         switch (command.cmd) {
         case 0x1:
-            return Submit(input, output);
+            return Submit(fd, input, output);
         case 0x2:
             return GetSyncpoint(input, output);
         case 0x3:
@@ -62,10 +62,15 @@ NvResult nvhost_vic::Ioctl3(DeviceFD fd, Ioctl command, const std::vector<u8>& i
     return NvResult::NotImplemented;
 }
 
-void nvhost_vic::OnOpen(DeviceFD fd) {}
+void nvhost_vic::OnOpen(DeviceFD fd) {
+    static u32 next_id{};
+    fd_to_id[fd] = next_id++;
+}
 
 void nvhost_vic::OnClose(DeviceFD fd) {
-    system.GPU().ClearCdmaInstance();
+    if (fd_to_id.find(fd) != fd_to_id.end()) {
+        system.GPU().ClearCdmaInstance(fd_to_id[fd]);
+    }
 }
 
 } // namespace Service::Nvidia::Devices

src/video_core/gpu.cpp

@@ -185,16 +185,6 @@ struct GPU::Impl {
         return *dma_pusher;
     }
 
-    /// Returns a reference to the GPU CDMA pusher.
-    [[nodiscard]] Tegra::CDmaPusher& CDmaPusher() {
-        return *cdma_pusher;
-    }
-
-    /// Returns a const reference to the GPU CDMA pusher.
-    [[nodiscard]] const Tegra::CDmaPusher& CDmaPusher() const {
-        return *cdma_pusher;
-    }
-
     /// Returns a reference to the underlying renderer.
     [[nodiscard]] VideoCore::RendererBase& Renderer() {
         return *renderer;
@@ -338,25 +328,26 @@ struct GPU::Impl {
     }
 
     /// Push GPU command buffer entries to be processed
-    void PushCommandBuffer(Tegra::ChCommandHeaderList& entries) {
+    void PushCommandBuffer(u32 id, Tegra::ChCommandHeaderList& entries) {
         if (!use_nvdec) {
             return;
         }
 
-        if (!cdma_pusher) {
-            cdma_pusher = std::make_unique<Tegra::CDmaPusher>(gpu);
+        if (cdma_pushers.find(id) == cdma_pushers.end()) {
+            cdma_pushers[id] = std::make_unique<Tegra::CDmaPusher>(gpu);
         }
 
         // SubmitCommandBuffer would make the nvdec operations async, this is not currently working
         // TODO(ameerj): RE proper async nvdec operation
         // gpu_thread.SubmitCommandBuffer(std::move(entries));
-
-        cdma_pusher->ProcessEntries(std::move(entries));
+        cdma_pushers[id]->ProcessEntries(std::move(entries));
     }
 
     /// Frees the CDMAPusher instance to free up resources
-    void ClearCdmaInstance() {
-        cdma_pusher.reset();
+    void ClearCdmaInstance(u32 id) {
+        if (cdma_pushers.find(id) != cdma_pushers.end()) {
+            cdma_pushers.erase(id);
+        }
     }
 
     /// Swap buffers (render frame)
@@ -659,7 +650,7 @@ struct GPU::Impl {
     Core::System& system;
     std::unique_ptr<Tegra::MemoryManager> memory_manager;
     std::unique_ptr<Tegra::DmaPusher> dma_pusher;
-    std::unique_ptr<Tegra::CDmaPusher> cdma_pusher;
+    std::map<u32, std::unique_ptr<Tegra::CDmaPusher>> cdma_pushers;
     std::unique_ptr<VideoCore::RendererBase> renderer;
     VideoCore::RasterizerInterface* rasterizer = nullptr;
     const bool use_nvdec;
@@ -811,14 +802,6 @@ const Tegra::DmaPusher& GPU::DmaPusher() const {
     return impl->DmaPusher();
 }
 
-Tegra::CDmaPusher& GPU::CDmaPusher() {
-    return impl->CDmaPusher();
-}
-
-const Tegra::CDmaPusher& GPU::CDmaPusher() const {
-    return impl->CDmaPusher();
-}
-
 VideoCore::RendererBase& GPU::Renderer() {
     return impl->Renderer();
 }
@@ -887,12 +870,12 @@ void GPU::PushGPUEntries(Tegra::CommandList&& entries) {
     impl->PushGPUEntries(std::move(entries));
 }
 
-void GPU::PushCommandBuffer(Tegra::ChCommandHeaderList& entries) {
-    impl->PushCommandBuffer(entries);
+void GPU::PushCommandBuffer(u32 id, Tegra::ChCommandHeaderList& entries) {
+    impl->PushCommandBuffer(id, entries);
 }
 
-void GPU::ClearCdmaInstance() {
-    impl->ClearCdmaInstance();
+void GPU::ClearCdmaInstance(u32 id) {
+    impl->ClearCdmaInstance(id);
 }
 
 void GPU::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {

src/video_core/gpu.h

@@ -242,10 +242,10 @@ public:
     void PushGPUEntries(Tegra::CommandList&& entries);
 
     /// Push GPU command buffer entries to be processed
-    void PushCommandBuffer(Tegra::ChCommandHeaderList& entries);
+    void PushCommandBuffer(u32 id, Tegra::ChCommandHeaderList& entries);
 
     /// Frees the CDMAPusher instance to free up resources
-    void ClearCdmaInstance();
+    void ClearCdmaInstance(u32 id);
 
     /// Swap buffers (render frame)
     void SwapBuffers(const Tegra::FramebufferConfig* framebuffer);