Recherche avancée

Sur d’autres sites (7705)

• Building C library (FFmpeg) with Android NDK r17 : undefined reference to '__mulodi4'

  14 mai 2018, par fpsulli3

  My problem happens to be with FFmpeg but I suspect that this would happen with almost any C library.

  Problem Description

  My app uses FFmpeg that is compiled with NDK r10e. I am trying to update everything to NDK r17, while also switching to clang, since Google prefers us to use that over gcc.

  My first step is to just build FFmpeg.

  To that end, I have used the make_standalone_toolchain.py script to create a stand-alone toolchain for the x86 architecture, like so :

  make_standalone_toolchain.py --arch x86 --api 21 --install-dir ~/Development/ndk-toolchains/x86

  Then I configure the FFmpeg build as follows :

  TOOLCHAIN_DIR=~/Development/ndk-toolchains/x86
  
  
  
  ./configure \
  
  --prefix=$(pwd)/android/x86 \
  
  --cross-prefix=$TOOLCHAIN_DIR/bin/i686-linux-android- \
  
  --target-os=android \
  
  --arch=x86 \
  
  --enable-cross-compile \
  
  --disable-asm \
  
  --toolchain=clang-usan \
  
  --disable-stripping \
  
  --extra-cflags="-m32" \
  
  --sysroot=$TOOLCHAIN_DIR/sysroot/

  And then I build it as follows :

  make clean
  
  make -j4
  
  make install

  Everything seems to compile fine, but I get several linker errors that all say the same thing :

  undefined reference to ’__mulodi4’

  Solutions I’ve tried

  1. Linking against libclang_rt.builtins*

  I found a few places around the Web which suggested that this is caused by the fact that libgcc doesn’t provide __mulodi4. A github user named sitsofe was nice enough to post a work-around here. However, I am sure where to find this libclang_rt.builtins-i686.a library. Here is what I was able to find in my standalone toolchain directory :

  ./lib64/clang/6.0.2/lib/linux/libclang_rt.builtins-x86_64.a
  ./lib64/clang/6.0.2/lib/linux/libclang_rt.builtins-i386.a
  ./lib64/clang/6.0.2/lib/linux/libclang_rt.builtins-aarch64-android.a
  ./lib64/clang/6.0.2/lib/linux/libclang_rt.builtins-mips64-android.a
  ./lib64/clang/6.0.2/lib/linux/libclang_rt.builtins-x86_64-android.a
  ./lib64/clang/6.0.2/lib/linux/libclang_rt.builtins-i686-android.a
  ./lib64/clang/6.0.2/lib/linux/libclang_rt.builtins-arm-android.a
  ./lib64/clang/6.0.2/lib/linux/libclang_rt.builtins-mips-android.a

  The libclang_rt.builtins-i686-android.a library looks close but (I think) no cigar. When I try to link to it, I get the same error :

  undefined reference to ’__mulodi4’

  Here is my new FFmpeg build config command :

  ./configure \
  
  --prefix=$(pwd)/android/x86 \
  
  --cross-prefix=$TOOLCHAIN_DIR/bin/i686-linux-android- \
  
  --target-os=android \
  
  --arch=x86 \
  
  --enable-cross-compile \
  
  --disable-asm \
  
  --toolchain=clang-usan \
  
  --disable-stripping \
  
  --extra-cflags="-m32" \
  
  --extra-ldflags="-L${TOOLCHAIN_DIR}/lib64/clang/6.0.2/lib/linux/libclang_rt.builtins-i686-android.a" \
  
  --sysroot=$TOOLCHAIN_DIR/sysroot/

  I checked with -v to make sure that this line was added to the linker flags, and it was. However, I have no idea if this library should even be expected to work, let alone whether I’m adding it to the linker flags correctly. In any case, what I’m doing here doesn’t work.

  2. Switching to a different sanitizer

  Instead of using the undefined sanitizer, I tried switching to the address sanitizer. This is (frankly) a total stab in the dark, based on a vague mention of asan being available in r17 at Google I/O this week.

  In this case, FFmpeg builds just fine !

  However, when I try to pull FFmpeg into my test project (a simple AAR w/ C++ support, that just has one jni method that calls av_gettime(), I get a ton of linker errors :

  Error:error : undefined reference to ’__asan_option_detect_stack_use_after_return’
  Error:error : undefined reference to ’__asan_stack_malloc_0’
  Error:error : undefined reference to ’__asan_report_load4’
  Error:error : undefined reference to ’__asan_report_load4’
  Error:error : undefined reference to ’__asan_shadow_memory_dynamic_address’
  Error:error : undefined reference to ’__asan_option_detect_stack_use_after_return’
  Error:error : undefined reference to ’__asan_stack_malloc_0’
  Error:error : undefined reference to ’__asan_report_load4’
  Error:error : undefined reference to ’__asan_report_load4’
  Error:error : undefined reference to ’__asan_shadow_memory_dynamic_address’
  Error:error : undefined reference to ’__asan_option_detect_stack_use_after_return’
  Error:error : undefined reference to ’__asan_stack_malloc_0’
  Error:error : undefined reference to ’__asan_report_store4’
  Error:error : undefined reference to ’__asan_report_store4’
  Error:error : undefined reference to ’__asan_init’
  Error:error : undefined reference to ’__asan_version_mismatch_check_v9’

  So it seems to find the FFmpeg library just fine, indicating that that part of my CMake file is correct, but it can’t locate any of these asan references.

  This seems to be a common problem that people are running into, but I can’t see to find a work-around that actually works for me.

• AVFrame : How to get/replace plane data buffer(s) and size ?

  19 juillet 2018, par user10099431

  I’m working on gstreamer1.0-libav (1.6.3), trying to port custom FPGA based H264 video acceleration from gstreamer 0.10.

  The data planes (YUV) used to be allocated by a simple malloc back in gstreamer 0.10, so we simply replaced the AVFrame.data[i] pointers by pointers to memory in our video acceleration core. It seems to be MUCH more complicated in gstreamer 1.12.

  For starters, I tried copying the YUV planes from AVFrame.data[i] to a separate buffer - which worked fine ! Since I haven’t seen an immediate way to obtain the size of AVFrame.data[i] and I recognized that data[0], data[1], data[2] seem to be in a single continuous buffer, I simply used (data[1] - data [0]) for the size of the Y plane and (data[2] - data[1]) for the sizes of the U/V planes respectively. This works fine, expect for one scenario :

  • Input H264 stream with resolution of 800x600 or greater
  • The camera is covered (jacket, hand, ...)

  This causes a SEGFAULT in the memcpy of the V plane (data[2]) using the sizes determined as described above. Before covering the camera, the stream is displayed completely fine ... so for some reason the dark screen changes the plane sizes ?

  My ultimate goal is replacing the data[i] pointers allocated by gstreamer by my custom memory allocation (for futher processing) ... where exactly are these buffers assigned, can I change them and how can I obtain the size of each plane (data[0], data[1], data[2]) ?

• Trying to tonemap 14-bit grayscale video

  12 janvier 2018, par Trevor

  I’m trying to generate h.264 video from raw 2-byte gray video (14-bit range encoded in 16-bit values). I can do something like :

  ffmpeg -f rawvideo -pix_fmt gray16le -s:v 1280x720 -r 60 -i input.raw -c:v libx264 output.mp4

  And I get video but it’s pretty dark, not sure if it’s clipping, doing a linear remap, or storing the 16-bit data and VLC is doing the remap. ffprobe is reporting Video: h264 (High 4:4:4 Predictive) (avc1 / 0x31637661), yuvj444p(pc), 1280x720, 108 kb/s, 60 fps, 60 tbr, 15360 tbn, 120 tbc

  I was figuring I’d use the tonemap filter to make a better mapping. I added a filter before the output file with -vf.

  • tonemap=hable errors Impossible to convert between the formats supported by the filter 'graph 0 input from stream 0:0' and the filter 'auto_scaler_0'
  • zscale=transfer=linear,tonemap=hable errors Impossible to convert between the formats supported by the filter 'Parsed_tonemap_1' and the filter 'auto_scaler_1'
  • zscale=transfer=linear,tonemap=hable,zscale=transfer=bt709,format=yuvj444p errors code 3074: no path between colorspaces

  I’m not sure where to proceed from here...