Recherche avancée

Sur d’autres sites (4553)

• avutil/hwcontext_cuda : add 4:2:2 pixel format support

  8 janvier, par Diego de Souza

  avutil/hwcontext_cuda : add 4:2:2 pixel format support
  This commit adds support for 4:2:2 pixel formats, namely NV16 and
  
  P216 for NVIDIA GPUs.
  Signed-off-by : Diego de Souza <ddesouza@nvidia.com>
  
  Signed-off-by : Timo Rothenpieler <timo@rothenpieler.org>
  

  • [DH] libavutil/hwcontext_cuda.c

• Configuration error when configuring the FFMPEG compilation with NVCC/CUDA [closed]

  3 février, par Lerenn

  I'm trying to compile the last FFMPEG version (49726a922fd2b358feb7753488d415180da5121c) on Fedora 41 with some libraries, including the CUDA libraries.

  


  Everything works well when compiling without CUDA, but I have 2 cryptic errors when trying to run the configure command with the NVCC :

  


    nvcc -gencode arch=compute_60,code=sm_60 -O2 -std=c&#x2B;&#x2B;11 -m64 -ptx -c -o /tmp/ffconf.q0uMcStN/test.o /tmp/ffconf.q0uMcStN/test.cu&#xA;  nvcc warning : Support for offline compilation for architectures prior to &#x27;<compute></compute>sm/lto>_75&#x27; will be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).&#xA;  /usr/include/c&#x2B;&#x2B;/14/type_traits(1610): error: "__is_nothrow_new_constructible" is not a function or static data member&#xA;        constexpr bool __is_nothrow_new_constructible&#xA;                    ^&#xA;&#xA;  /usr/include/c&#x2B;&#x2B;/14/type_traits(1610): error: "constexpr" is not valid here&#xA;        constexpr bool __is_nothrow_new_constructible&#xA;        ^&#xA;&#xA;  2 errors detected in the compilation of "/tmp/ffconf.q0uMcStN/test.cu".&#xA;  ERROR: failed checking for nvcc.&#xA;

  


  I tried compiling it with default gcc (gcc (GCC) 14.2.1 20250110 (Red Hat 14.2.1-7)) but my NVCC version (Build cuda_12.8.r12.8/compiler.35404655_0) seems to support only the GCC version 13.2 : https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html.

  


  I have compiled it and ran the configure command again but it seems that I have the same errors.

  


  Here is the configure command :

  


    PKG_CONFIG_PATH="/usr/local/lib/pkgconfig" \
        ./configure \
        --cc=gcc-13.2 \
        --extra-libs=-lpthread --prefix="/usr/local" \
        --extra-cflags="-I/usr/local/include -I/usr/local/cuda/include" \
        --extra-ldflags="-L/usr/local/lib -L/usr/local/cuda/lib64" \
        --pkg-config-flags="--static" --enable-gpl --enable-nonfree \
        --enable-libfdk-aac --enable-libmp3lame --enable-libopus \
        --enable-libvorbis --enable-libvpx --enable-libx264 --enable-libx265 \
        --enable-libvidstab --enable-libaom \
        --enable-cuda-nvcc --enable-libnpp \
        --disable-static --enable-shared


  


  If you have any idea or lead on the matter, I would be really grateful.

  


• Decoding HEVC with Alpha Channel via NVDEC : Monochrome (4:0:0) Workarounds ? [closed]

  3 mars, par Holy_diver

  I’m working on decoding HEVC streams that include an alpha (transparency) channel using NVIDIA’s NVDEC. The alpha channel is encoded in monochrome (YUV 4:0:0), but NVDEC’s HEVC decoder appears to lack support for monochrome formats. How can I work around this limitation ?

  


  Problem Details :

  


  HEVC Profile: Stream uses HEVC_Rext (Range Extensions) 
with a monochrome alpha layer (4:0:0 chroma subsampling).

NVDEC Limitations: The SDK documentation states support 
for 4:0:0 (8-bit) only for specific codecs like JPEG, not HEVC.
Attempting to decode returns 
cudaError_InvalidValue or NVCUDACB_STATUS_INVALID_PARAM.

Alpha Storage: The alpha is either a separate stream 
or a dual-layer HEVC bitstream (e.g., DUAL_LAYER_DEPTH_SEPARATE).