Merge pull request #6525 from ameerj/nvdec-fixes
nvdec: Fix Submit Ioctl data source, vic frame dimension computations
This commit is contained in:
commit
da4ca4f2f9
|
@ -19,26 +19,29 @@
|
||||||
namespace Service::Nvidia::Devices {
|
namespace Service::Nvidia::Devices {
|
||||||
|
|
||||||
namespace {
|
namespace {
|
||||||
// Splice vectors will copy count amount of type T from the input vector into the dst vector.
|
// Copies count amount of type T from the input vector into the dst vector.
|
||||||
|
// Returns the number of bytes written into dst.
|
||||||
template <typename T>
|
template <typename T>
|
||||||
std::size_t SpliceVectors(const std::vector<u8>& input, std::vector<T>& dst, std::size_t count,
|
std::size_t SliceVectors(const std::vector<u8>& input, std::vector<T>& dst, std::size_t count,
|
||||||
std::size_t offset) {
|
std::size_t offset) {
|
||||||
if (!dst.empty()) {
|
if (dst.empty()) {
|
||||||
std::memcpy(dst.data(), input.data() + offset, count * sizeof(T));
|
|
||||||
}
|
|
||||||
return 0;
|
return 0;
|
||||||
}
|
}
|
||||||
|
const size_t bytes_copied = count * sizeof(T);
|
||||||
|
std::memcpy(dst.data(), input.data() + offset, bytes_copied);
|
||||||
|
return bytes_copied;
|
||||||
|
}
|
||||||
|
|
||||||
// Write vectors will write data to the output buffer
|
// Writes the data in src to an offset into the dst vector. The offset is specified in bytes
|
||||||
|
// Returns the number of bytes written into dst.
|
||||||
template <typename T>
|
template <typename T>
|
||||||
std::size_t WriteVectors(std::vector<u8>& dst, const std::vector<T>& src, std::size_t offset) {
|
std::size_t WriteVectors(std::vector<u8>& dst, const std::vector<T>& src, std::size_t offset) {
|
||||||
if (src.empty()) {
|
if (src.empty()) {
|
||||||
return 0;
|
return 0;
|
||||||
} else {
|
|
||||||
std::memcpy(dst.data() + offset, src.data(), src.size() * sizeof(T));
|
|
||||||
offset += src.size() * sizeof(T);
|
|
||||||
return offset;
|
|
||||||
}
|
}
|
||||||
|
const size_t bytes_copied = src.size() * sizeof(T);
|
||||||
|
std::memcpy(dst.data() + offset, src.data(), bytes_copied);
|
||||||
|
return bytes_copied;
|
||||||
}
|
}
|
||||||
} // Anonymous namespace
|
} // Anonymous namespace
|
||||||
|
|
||||||
|
@ -62,7 +65,6 @@ NvResult nvhost_nvdec_common::Submit(const std::vector<u8>& input, std::vector<u
|
||||||
LOG_DEBUG(Service_NVDRV, "called NVDEC Submit, cmd_buffer_count={}", params.cmd_buffer_count);
|
LOG_DEBUG(Service_NVDRV, "called NVDEC Submit, cmd_buffer_count={}", params.cmd_buffer_count);
|
||||||
|
|
||||||
// Instantiate param buffers
|
// Instantiate param buffers
|
||||||
std::size_t offset = sizeof(IoctlSubmit);
|
|
||||||
std::vector<CommandBuffer> command_buffers(params.cmd_buffer_count);
|
std::vector<CommandBuffer> command_buffers(params.cmd_buffer_count);
|
||||||
std::vector<Reloc> relocs(params.relocation_count);
|
std::vector<Reloc> relocs(params.relocation_count);
|
||||||
std::vector<u32> reloc_shifts(params.relocation_count);
|
std::vector<u32> reloc_shifts(params.relocation_count);
|
||||||
|
@ -70,13 +72,14 @@ NvResult nvhost_nvdec_common::Submit(const std::vector<u8>& input, std::vector<u
|
||||||
std::vector<SyncptIncr> wait_checks(params.syncpoint_count);
|
std::vector<SyncptIncr> wait_checks(params.syncpoint_count);
|
||||||
std::vector<Fence> fences(params.fence_count);
|
std::vector<Fence> fences(params.fence_count);
|
||||||
|
|
||||||
// Splice input into their respective buffers
|
// Slice input into their respective buffers
|
||||||
offset = SpliceVectors(input, command_buffers, params.cmd_buffer_count, offset);
|
std::size_t offset = sizeof(IoctlSubmit);
|
||||||
offset = SpliceVectors(input, relocs, params.relocation_count, offset);
|
offset += SliceVectors(input, command_buffers, params.cmd_buffer_count, offset);
|
||||||
offset = SpliceVectors(input, reloc_shifts, params.relocation_count, offset);
|
offset += SliceVectors(input, relocs, params.relocation_count, offset);
|
||||||
offset = SpliceVectors(input, syncpt_increments, params.syncpoint_count, offset);
|
offset += SliceVectors(input, reloc_shifts, params.relocation_count, offset);
|
||||||
offset = SpliceVectors(input, wait_checks, params.syncpoint_count, offset);
|
offset += SliceVectors(input, syncpt_increments, params.syncpoint_count, offset);
|
||||||
offset = SpliceVectors(input, fences, params.fence_count, offset);
|
offset += SliceVectors(input, wait_checks, params.syncpoint_count, offset);
|
||||||
|
offset += SliceVectors(input, fences, params.fence_count, offset);
|
||||||
|
|
||||||
auto& gpu = system.GPU();
|
auto& gpu = system.GPU();
|
||||||
if (gpu.UseNvdec()) {
|
if (gpu.UseNvdec()) {
|
||||||
|
@ -88,35 +91,27 @@ NvResult nvhost_nvdec_common::Submit(const std::vector<u8>& input, std::vector<u
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
for (const auto& cmd_buffer : command_buffers) {
|
for (const auto& cmd_buffer : command_buffers) {
|
||||||
auto object = nvmap_dev->GetObject(cmd_buffer.memory_id);
|
const auto object = nvmap_dev->GetObject(cmd_buffer.memory_id);
|
||||||
ASSERT_OR_EXECUTE(object, return NvResult::InvalidState;);
|
ASSERT_OR_EXECUTE(object, return NvResult::InvalidState;);
|
||||||
const auto map = FindBufferMap(object->dma_map_addr);
|
|
||||||
if (!map) {
|
|
||||||
LOG_ERROR(Service_NVDRV, "Tried to submit an invalid offset 0x{:X} dma 0x{:X}",
|
|
||||||
object->addr, object->dma_map_addr);
|
|
||||||
return NvResult::Success;
|
|
||||||
}
|
|
||||||
Tegra::ChCommandHeaderList cmdlist(cmd_buffer.word_count);
|
Tegra::ChCommandHeaderList cmdlist(cmd_buffer.word_count);
|
||||||
gpu.MemoryManager().ReadBlock(map->StartAddr() + cmd_buffer.offset, cmdlist.data(),
|
system.Memory().ReadBlock(object->addr + cmd_buffer.offset, cmdlist.data(),
|
||||||
cmdlist.size() * sizeof(u32));
|
cmdlist.size() * sizeof(u32));
|
||||||
gpu.PushCommandBuffer(cmdlist);
|
gpu.PushCommandBuffer(cmdlist);
|
||||||
}
|
}
|
||||||
if (gpu.UseNvdec()) {
|
if (gpu.UseNvdec()) {
|
||||||
|
|
||||||
fences[0].value = syncpoint_manager.IncreaseSyncpoint(fences[0].id, 1);
|
fences[0].value = syncpoint_manager.IncreaseSyncpoint(fences[0].id, 1);
|
||||||
|
|
||||||
Tegra::ChCommandHeaderList cmdlist{{(4 << 28) | fences[0].id}};
|
Tegra::ChCommandHeaderList cmdlist{{(4 << 28) | fences[0].id}};
|
||||||
gpu.PushCommandBuffer(cmdlist);
|
gpu.PushCommandBuffer(cmdlist);
|
||||||
}
|
}
|
||||||
std::memcpy(output.data(), ¶ms, sizeof(IoctlSubmit));
|
std::memcpy(output.data(), ¶ms, sizeof(IoctlSubmit));
|
||||||
// Some games expect command_buffers to be written back
|
// Some games expect command_buffers to be written back
|
||||||
offset = sizeof(IoctlSubmit);
|
offset = sizeof(IoctlSubmit);
|
||||||
offset = WriteVectors(output, command_buffers, offset);
|
offset += WriteVectors(output, command_buffers, offset);
|
||||||
offset = WriteVectors(output, relocs, offset);
|
offset += WriteVectors(output, relocs, offset);
|
||||||
offset = WriteVectors(output, reloc_shifts, offset);
|
offset += WriteVectors(output, reloc_shifts, offset);
|
||||||
offset = WriteVectors(output, syncpt_increments, offset);
|
offset += WriteVectors(output, syncpt_increments, offset);
|
||||||
offset = WriteVectors(output, wait_checks, offset);
|
offset += WriteVectors(output, wait_checks, offset);
|
||||||
offset = WriteVectors(output, fences, offset);
|
offset += WriteVectors(output, fences, offset);
|
||||||
|
|
||||||
return NvResult::Success;
|
return NvResult::Success;
|
||||||
}
|
}
|
||||||
|
@ -148,14 +143,14 @@ NvResult nvhost_nvdec_common::MapBuffer(const std::vector<u8>& input, std::vecto
|
||||||
std::memcpy(¶ms, input.data(), sizeof(IoctlMapBuffer));
|
std::memcpy(¶ms, input.data(), sizeof(IoctlMapBuffer));
|
||||||
std::vector<MapBufferEntry> cmd_buffer_handles(params.num_entries);
|
std::vector<MapBufferEntry> cmd_buffer_handles(params.num_entries);
|
||||||
|
|
||||||
SpliceVectors(input, cmd_buffer_handles, params.num_entries, sizeof(IoctlMapBuffer));
|
SliceVectors(input, cmd_buffer_handles, params.num_entries, sizeof(IoctlMapBuffer));
|
||||||
|
|
||||||
auto& gpu = system.GPU();
|
auto& gpu = system.GPU();
|
||||||
|
|
||||||
for (auto& cmf_buff : cmd_buffer_handles) {
|
for (auto& cmd_buffer : cmd_buffer_handles) {
|
||||||
auto object{nvmap_dev->GetObject(cmf_buff.map_handle)};
|
auto object{nvmap_dev->GetObject(cmd_buffer.map_handle)};
|
||||||
if (!object) {
|
if (!object) {
|
||||||
LOG_ERROR(Service_NVDRV, "invalid cmd_buffer nvmap_handle={:X}", cmf_buff.map_handle);
|
LOG_ERROR(Service_NVDRV, "invalid cmd_buffer nvmap_handle={:X}", cmd_buffer.map_handle);
|
||||||
std::memcpy(output.data(), ¶ms, output.size());
|
std::memcpy(output.data(), ¶ms, output.size());
|
||||||
return NvResult::InvalidState;
|
return NvResult::InvalidState;
|
||||||
}
|
}
|
||||||
|
@ -170,7 +165,7 @@ NvResult nvhost_nvdec_common::MapBuffer(const std::vector<u8>& input, std::vecto
|
||||||
if (!object->dma_map_addr) {
|
if (!object->dma_map_addr) {
|
||||||
LOG_ERROR(Service_NVDRV, "failed to map size={}", object->size);
|
LOG_ERROR(Service_NVDRV, "failed to map size={}", object->size);
|
||||||
} else {
|
} else {
|
||||||
cmf_buff.map_address = object->dma_map_addr;
|
cmd_buffer.map_address = object->dma_map_addr;
|
||||||
AddBufferMap(object->dma_map_addr, object->size, object->addr,
|
AddBufferMap(object->dma_map_addr, object->size, object->addr,
|
||||||
object->status == nvmap::Object::Status::Allocated);
|
object->status == nvmap::Object::Status::Allocated);
|
||||||
}
|
}
|
||||||
|
@ -186,14 +181,14 @@ NvResult nvhost_nvdec_common::UnmapBuffer(const std::vector<u8>& input, std::vec
|
||||||
IoctlMapBuffer params{};
|
IoctlMapBuffer params{};
|
||||||
std::memcpy(¶ms, input.data(), sizeof(IoctlMapBuffer));
|
std::memcpy(¶ms, input.data(), sizeof(IoctlMapBuffer));
|
||||||
std::vector<MapBufferEntry> cmd_buffer_handles(params.num_entries);
|
std::vector<MapBufferEntry> cmd_buffer_handles(params.num_entries);
|
||||||
SpliceVectors(input, cmd_buffer_handles, params.num_entries, sizeof(IoctlMapBuffer));
|
SliceVectors(input, cmd_buffer_handles, params.num_entries, sizeof(IoctlMapBuffer));
|
||||||
|
|
||||||
auto& gpu = system.GPU();
|
auto& gpu = system.GPU();
|
||||||
|
|
||||||
for (auto& cmf_buff : cmd_buffer_handles) {
|
for (auto& cmd_buffer : cmd_buffer_handles) {
|
||||||
const auto object{nvmap_dev->GetObject(cmf_buff.map_handle)};
|
const auto object{nvmap_dev->GetObject(cmd_buffer.map_handle)};
|
||||||
if (!object) {
|
if (!object) {
|
||||||
LOG_ERROR(Service_NVDRV, "invalid cmd_buffer nvmap_handle={:X}", cmf_buff.map_handle);
|
LOG_ERROR(Service_NVDRV, "invalid cmd_buffer nvmap_handle={:X}", cmd_buffer.map_handle);
|
||||||
std::memcpy(output.data(), ¶ms, output.size());
|
std::memcpy(output.data(), ¶ms, output.size());
|
||||||
return NvResult::InvalidState;
|
return NvResult::InvalidState;
|
||||||
}
|
}
|
||||||
|
|
|
@ -129,28 +129,27 @@ void Vic::Execute() {
|
||||||
|
|
||||||
const std::size_t surface_width = config.surface_width_minus1 + 1;
|
const std::size_t surface_width = config.surface_width_minus1 + 1;
|
||||||
const std::size_t surface_height = config.surface_height_minus1 + 1;
|
const std::size_t surface_height = config.surface_height_minus1 + 1;
|
||||||
const std::size_t half_width = surface_width / 2;
|
const auto frame_width = std::min(surface_width, static_cast<size_t>(frame->width));
|
||||||
const std::size_t half_height = config.surface_height_minus1 / 2;
|
const auto frame_height = std::min(surface_height, static_cast<size_t>(frame->height));
|
||||||
|
const std::size_t half_width = frame_width / 2;
|
||||||
|
const std::size_t half_height = frame_height / 2;
|
||||||
const std::size_t aligned_width = (surface_width + 0xff) & ~0xff;
|
const std::size_t aligned_width = (surface_width + 0xff) & ~0xff;
|
||||||
|
|
||||||
const auto* luma_ptr = frame->data[0];
|
const auto* luma_ptr = frame->data[0];
|
||||||
const auto* chroma_b_ptr = frame->data[1];
|
const auto* chroma_b_ptr = frame->data[1];
|
||||||
const auto* chroma_r_ptr = frame->data[2];
|
const auto* chroma_r_ptr = frame->data[2];
|
||||||
const auto stride = frame->linesize[0];
|
const auto stride = static_cast<size_t>(frame->linesize[0]);
|
||||||
const auto half_stride = frame->linesize[1];
|
const auto half_stride = static_cast<size_t>(frame->linesize[1]);
|
||||||
|
|
||||||
luma_buffer.resize(aligned_width * surface_height);
|
luma_buffer.resize(aligned_width * surface_height);
|
||||||
chroma_buffer.resize(aligned_width * half_height);
|
chroma_buffer.resize(aligned_width * surface_height / 2);
|
||||||
|
|
||||||
// Populate luma buffer
|
// Populate luma buffer
|
||||||
for (std::size_t y = 0; y < surface_height - 1; ++y) {
|
for (std::size_t y = 0; y < frame_height; ++y) {
|
||||||
const std::size_t src = y * stride;
|
const std::size_t src = y * stride;
|
||||||
const std::size_t dst = y * aligned_width;
|
const std::size_t dst = y * aligned_width;
|
||||||
|
for (std::size_t x = 0; x < frame_width; ++x) {
|
||||||
const std::size_t size = surface_width;
|
luma_buffer[dst + x] = luma_ptr[src + x];
|
||||||
|
|
||||||
for (std::size_t offset = 0; offset < size; ++offset) {
|
|
||||||
luma_buffer[dst + offset] = luma_ptr[src + offset];
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
gpu.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(),
|
gpu.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(),
|
||||||
|
|
Reference in New Issue