hdkv/yuzu
1
Fork 0
Code Issues Pull Requests Releases 1 Activity

early-access version 1959

Browse Source
This commit is contained in:
pineappleEA
2021-08-02 19:12:49 +02:00
parent c1ab8428b1
commit ec76444e96
15 changed files with 54 additions and 293 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp
View File

@@ -68,7 +68,6 @@ void nvhost_nvdec::OnOpen(DeviceFD fd) {}
void nvhost_nvdec::OnClose(DeviceFD fd) {
LOG_INFO(Service_NVDRV, "NVDEC video stream ended");
system.GPU().ClearCdmaInstance();
system.GPU().MemoryManager().InvalidateQueuedCaches();
}
} // namespace Service::Nvidia::Devices

64
src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp
View File

@@ -166,8 +166,6 @@ NvResult nvhost_nvdec_common::MapBuffer(const std::vector<u8>& input, std::vecto
LOG_ERROR(Service_NVDRV, "failed to map size={}", object->size);
} else {
cmd_buffer.map_address = object->dma_map_addr;
AddBufferMap(object->dma_map_addr, object->size, object->addr,
object->status == nvmap::Object::Status::Allocated);
}
}
std::memcpy(output.data(), &params, sizeof(IoctlMapBuffer));
@@ -178,35 +176,10 @@ NvResult nvhost_nvdec_common::MapBuffer(const std::vector<u8>& input, std::vecto
}
NvResult nvhost_nvdec_common::UnmapBuffer(const std::vector<u8>& input, std::vector<u8>& output) {
IoctlMapBuffer params{};
std::memcpy(&params, input.data(), sizeof(IoctlMapBuffer));
std::vector<MapBufferEntry> cmd_buffer_handles(params.num_entries);
SliceVectors(input, cmd_buffer_handles, params.num_entries, sizeof(IoctlMapBuffer));
auto& gpu = system.GPU();
for (auto& cmd_buffer : cmd_buffer_handles) {
const auto object{nvmap_dev->GetObject(cmd_buffer.map_handle)};
if (!object) {
LOG_ERROR(Service_NVDRV, "invalid cmd_buffer nvmap_handle={:X}", cmd_buffer.map_handle);
std::memcpy(output.data(), &params, output.size());
return NvResult::InvalidState;
}
if (const auto size{RemoveBufferMap(object->dma_map_addr)}; size) {
if (vic_device) {
// UnmapVicFrame defers texture_cache invalidation of the frame address until
// the stream is over
gpu.MemoryManager().UnmapVicFrame(object->dma_map_addr, *size);
} else {
gpu.MemoryManager().Unmap(object->dma_map_addr, *size);
}
} else {
// This occurs quite frequently, however does not seem to impact functionality
LOG_DEBUG(Service_NVDRV, "invalid offset=0x{:X} dma=0x{:X}", object->addr,
object->dma_map_addr);
}
object->dma_map_addr = 0;
}
// This is intentionally left stubbed.
// By unmapping GPU buffers here, we break the continuity of the VP9 reference frame addresses,
// and risk invalidating data before the async GPU thread is done with it
LOG_DEBUG(Service_NVDRV, "(STUBBED) called");
std::memset(output.data(), 0, output.size());
return NvResult::Success;
}
@@ -218,33 +191,4 @@ NvResult nvhost_nvdec_common::SetSubmitTimeout(const std::vector<u8>& input,
return NvResult::Success;
}
std::optional<nvhost_nvdec_common::BufferMap> nvhost_nvdec_common::FindBufferMap(
GPUVAddr gpu_addr) const {
const auto it = std::find_if(
buffer_mappings.begin(), buffer_mappings.upper_bound(gpu_addr), [&](const auto& entry) {
return (gpu_addr >= entry.second.StartAddr() && gpu_addr < entry.second.EndAddr());
});
ASSERT(it != buffer_mappings.end());
return it->second;
}
void nvhost_nvdec_common::AddBufferMap(GPUVAddr gpu_addr, std::size_t size, VAddr cpu_addr,
bool is_allocated) {
buffer_mappings.insert_or_assign(gpu_addr, BufferMap{gpu_addr, size, cpu_addr, is_allocated});
}
std::optional<std::size_t> nvhost_nvdec_common::RemoveBufferMap(GPUVAddr gpu_addr) {
const auto iter{buffer_mappings.find(gpu_addr)};
if (iter == buffer_mappings.end()) {
return std::nullopt;
}
std::size_t size = 0;
if (iter->second.IsAllocated()) {
size = iter->second.Size();
}
buffer_mappings.erase(iter);
return size;
}
} // namespace Service::Nvidia::Devices

46
src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h
View File

@@ -23,45 +23,6 @@ public:
~nvhost_nvdec_common() override;
protected:
class BufferMap final {
public:
constexpr BufferMap() = default;
constexpr BufferMap(GPUVAddr start_addr_, std::size_t size_)
: start_addr{start_addr_}, end_addr{start_addr_ + size_} {}
constexpr BufferMap(GPUVAddr start_addr_, std::size_t size_, VAddr cpu_addr_,
bool is_allocated_)
: start_addr{start_addr_}, end_addr{start_addr_ + size_}, cpu_addr{cpu_addr_},
is_allocated{is_allocated_} {}
constexpr VAddr StartAddr() const {
return start_addr;
}
constexpr VAddr EndAddr() const {
return end_addr;
}
constexpr std::size_t Size() const {
return end_addr - start_addr;
}
constexpr VAddr CpuAddr() const {
return cpu_addr;
}
constexpr bool IsAllocated() const {
return is_allocated;
}
private:
GPUVAddr start_addr{};
GPUVAddr end_addr{};
VAddr cpu_addr{};
bool is_allocated{};
};
struct IoctlSetNvmapFD {
s32_le nvmap_fd{};
};
@@ -154,18 +115,11 @@ protected:
NvResult UnmapBuffer(const std::vector<u8>& input, std::vector<u8>& output);
NvResult SetSubmitTimeout(const std::vector<u8>& input, std::vector<u8>& output);
std::optional<BufferMap> FindBufferMap(GPUVAddr gpu_addr) const;
void AddBufferMap(GPUVAddr gpu_addr, std::size_t size, VAddr cpu_addr, bool is_allocated);
std::optional<std::size_t> RemoveBufferMap(GPUVAddr gpu_addr);
s32_le nvmap_fd{};
u32_le submit_timeout{};
bool vic_device{};
std::shared_ptr<nvmap> nvmap_dev;
SyncpointManager& syncpoint_manager;
std::array<u32, MaxSyncPoints> device_syncpoints{};
// This is expected to be ordered, therefore we must use a map, not unordered_map
std::map<GPUVAddr, BufferMap> buffer_mappings;
};
}; // namespace Devices
} // namespace Service::Nvidia

5
src/core/hle/service/nvdrv/devices/nvhost_vic.cpp
View File

@@ -12,9 +12,8 @@
namespace Service::Nvidia::Devices {
nvhost_vic::nvhost_vic(Core::System& system_, std::shared_ptr<nvmap> nvmap_dev_,
SyncpointManager& syncpoint_manager_)
: nvhost_nvdec_common(system_, std::move(nvmap_dev_), syncpoint_manager_) {
vic_device = true;
}
: nvhost_nvdec_common{system_, std::move(nvmap_dev_), syncpoint_manager_} {}
nvhost_vic::~nvhost_vic() = default;
NvResult nvhost_vic::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input,