diff options
author | ameerj <aj662@drexel.edu> | 2020-11-23 19:25:01 +0100 |
---|---|---|
committer | ameerj <52414509+ameerj@users.noreply.github.com> | 2021-02-13 19:07:31 +0100 |
commit | ac265a72ce4176ceb3cd10a5548ab71519771640 (patch) | |
tree | 0acde029388d465a5801db9106dd8f4e026e57e8 /src/video_core/command_classes/vic.cpp | |
parent | Merge pull request #5919 from ReinUsesLisp/stream-buffer-tragic (diff) | |
download | yuzu-ac265a72ce4176ceb3cd10a5548ab71519771640.tar yuzu-ac265a72ce4176ceb3cd10a5548ab71519771640.tar.gz yuzu-ac265a72ce4176ceb3cd10a5548ab71519771640.tar.bz2 yuzu-ac265a72ce4176ceb3cd10a5548ab71519771640.tar.lz yuzu-ac265a72ce4176ceb3cd10a5548ab71519771640.tar.xz yuzu-ac265a72ce4176ceb3cd10a5548ab71519771640.tar.zst yuzu-ac265a72ce4176ceb3cd10a5548ab71519771640.zip |
Diffstat (limited to '')
-rw-r--r-- | src/video_core/command_classes/vic.cpp | 21 |
1 files changed, 13 insertions, 8 deletions
diff --git a/src/video_core/command_classes/vic.cpp b/src/video_core/command_classes/vic.cpp index 2b7569335..73680d057 100644 --- a/src/video_core/command_classes/vic.cpp +++ b/src/video_core/command_classes/vic.cpp @@ -18,7 +18,10 @@ extern "C" { namespace Tegra { Vic::Vic(GPU& gpu_, std::shared_ptr<Nvdec> nvdec_processor_) - : gpu(gpu_), nvdec_processor(std::move(nvdec_processor_)) {} + : gpu(gpu_), + nvdec_processor(std::move(nvdec_processor_)), converted_frame_buffer{nullptr, av_free} + +{} Vic::~Vic() = default; void Vic::VicStateWrite(u32 offset, u32 arguments) { @@ -89,8 +92,10 @@ void Vic::Execute() { // Get Converted frame const std::size_t linear_size = frame->width * frame->height * 4; - using AVMallocPtr = std::unique_ptr<u8, decltype(&av_free)>; - AVMallocPtr converted_frame_buffer{static_cast<u8*>(av_malloc(linear_size)), av_free}; + // Only allocate frame_buffer once per stream, as the size is not expected to change + if (!converted_frame_buffer) { + converted_frame_buffer = AVMallocPtr{static_cast<u8*>(av_malloc(linear_size)), av_free}; + } const int converted_stride{frame->width * 4}; u8* const converted_frame_buf_addr{converted_frame_buffer.get()}; @@ -104,12 +109,12 @@ void Vic::Execute() { const u32 block_height = static_cast<u32>(config.block_linear_height_log2); const auto size = Tegra::Texture::CalculateSize(true, 4, frame->width, frame->height, 1, block_height, 0); - std::vector<u8> swizzled_data(size); + luma_buffer.resize(size); Tegra::Texture::SwizzleSubrect(frame->width, frame->height, frame->width * 4, - frame->width, 4, swizzled_data.data(), + frame->width, 4, luma_buffer.data(), converted_frame_buffer.get(), block_height, 0, 0); - gpu.MemoryManager().WriteBlock(output_surface_luma_address, swizzled_data.data(), size); + gpu.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), size); } else { // send pitch linear frame gpu.MemoryManager().WriteBlock(output_surface_luma_address, converted_frame_buf_addr, @@ -132,8 +137,8 @@ void Vic::Execute() { const auto stride = frame->linesize[0]; const auto half_stride = frame->linesize[1]; - std::vector<u8> luma_buffer(aligned_width * surface_height); - std::vector<u8> chroma_buffer(aligned_width * half_height); + luma_buffer.resize(aligned_width * surface_height); + chroma_buffer.resize(aligned_width * half_height); // Populate luma buffer for (std::size_t y = 0; y < surface_height - 1; ++y) { |