[SOLVED] FFmpeg 4.4.1 compilation error with Lensfun and other issues in slackware64 15.0

lecramyajiv · 03-17-2022, 02:07 PM

I have installed Slackware 15.0 64 bit. I am trying to recompile ffmpeg 4.4.1 to support more options, the problematic options are opencv, lensfun and glslang.

When I try to compile with with --enable-opencv it gives the error: opencv not found using pkg-config.I've added the location of opencv4.pc to the $PATH but it still fails. I've only found some couple links but not any answer to this problem.

https://trac.ffmpeg.org/ticket/7608
https://www.linuxquestions.org/quest...cv-4175645720/

Coming to lensfun, I'm using lensfun verison 0.3.95. FFmpeg 4.4.1 detects it during configure but during compilation it gives the following error

Code:

In file included from libavfilter/vf_lensfun.c:40:
/usr/include/lensfun/lensfun.h:3102:41: note: expected ‘const lfLens *’ but argument is of type ‘int’
 3102 |     lfModifier *modifier, const lfLens* lens, float focal, lfLensType target_projection);
      |                           ~~~~~~~~~~~~~~^~~~
libavfilter/vf_lensfun.c:234:17: error: too few arguments to function ‘lf_modifier_enable_projection_transform’
  234 |                 lf_modifier_enable_projection_transform(lensfun->modifier, lensfun->target_geometry);
      |                 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from libavfilter/vf_lensfun.c:40:
/usr/include/lensfun/lensfun.h:3101:17: note: declared here
 3101 | LF_EXPORT cbool lf_modifier_enable_projection_transform (
      |                 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
libavfilter/vf_lensfun.c:238:17: error: too few arguments to function ‘lf_modifier_enable_tca_correction’
  238 |                 lf_modifier_enable_tca_correction(lensfun->modifier);
      |                 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from libavfilter/vf_lensfun.c:40:
/usr/include/lensfun/lensfun.h:3094:15: note: declared here
 3094 | LF_EXPORT int lf_modifier_enable_tca_correction (lfModifier *modifier, const lfLens* lens, float focal);
      |               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
make: *** [ffbuild/common.mak:67: libavfilter/vf_lensfun.o] Error 1

I found this link for lensfun again no solutions to it,
https://trac.ffmpeg.org/ticket/9112

when trying it with ffmpeg 5.0 it gives the following error

Code:

libavfilter/vf_lensfun.c:234:17: error: implicit declaration of function ‘lf_modifier_enable_projection_transform’ [-Werror=implicit-function-declaration]
  234 |                 lf_modifier_enable_projection_transform(lensfun->modifier, lensfun->target_geometry);
      |                 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
libavfilter/vf_lensfun.c:235:17: error: implicit declaration of function ‘lf_modifier_enable_scaling’ [-Werror=implicit-function-declaration]
  235 |                 lf_modifier_enable_scaling(lensfun->modifier, lensfun->scale);
      |                 ^~~~~~~~~~~~~~~~~~~~~~~~~~
libavfilter/vf_lensfun.c:238:17: error: implicit declaration of function ‘lf_modifier_enable_tca_correction’ [-Werror=implicit-function-declaration]
  238 |                 lf_modifier_enable_tca_correction(lensfun->modifier);
      |                 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
cc1: some warnings being treated as errors
make: *** [ffbuild/common.mak:67: libavfilter/vf_lensfun.o] Error 1

In Debian and Arch, there seems to be no problem with enabling lensfun.

when enabling libglslang it gives an error libglslang not found. I have installed glslang version 11.8.0 but curiously libglslang gets detected in ffmpeg 5.0 and compiles with no problem but not in ffmpeg 4.4.1.

can you guys please help me with this.

drumz · 03-17-2022, 02:21 PM

I also had a problem with lensfun and opencv (and I found a patch for glslang). My solution: Don't enable those options. Here is my slackrepo hint file for ffmpeg (My slackrepo points to a local tree with my additions on top of the 15.0 branch of SBo):

Code:

ADDREQUIRES=" \
	nv-codec-headers \
	x264             \
	celt             \
	libdc1394        \
	libavc1394       \
	gsm              \
	rtmpdump         \
	xvidcore         \
	libass           \
	libiec61883      \
	libilbc          \
	libmodplug       \
	twolame          \
	ladspa_sdk       \
	flite            \
	x265             \
	zvbi             \
	libgme           \
	snappy           \
	libbs2b          \
	rubberband       \
	jack             \
	aom              \
	lv2              \
        lilv             \
	dav1d            \
	codec2           \
	soxr             \
	srt              \
	zimg             \
	chromaprint      \
	opencore-amr     \
	libfdk-aac"
OPTIONS=" \
	CUDA=yes \
	CUDAINCDIR=/usr/local/cuda/include \
	CUDALDDIR=/usr/local/cuda/lib64 \
	LIBNPP=yes \
	X264=yes \
	CELT=yes \
	DC1394=yes \
	GSM=yes \
	RTMP=yes \
	XVID=yes \
	ASS=yes \
	IEC61883=yes \
	ILBC=yes \
	MODPLUG=yes \
	TWOLAME=yes \
	LADSPA=yes \
	FLITE=yes \
	X265=yes \
	ZVBI=yes \
	OPENCV=no \
	GME=yes \
	SNAPPY=yes \
	BS2B=yes \
	RUBBERBAND=yes \
	JACK=yes \
	AOM=yes \
	LV2=yes \
	DAV1D=yes \
	CODEC2=yes \
	SOXR=yes \
	SRT=yes \
	ZIMG=yes \
	CHROMAPRINT=yes \
	OPENCORE=yes \
	FDK_AAC=yes \
	VULKAN=yes \
	GLSLANG=yes \
	LENSFUN=no \
	AAC=yes"

I just used ffmpeg-4.4.1 (with 2 additional patches - 1 for glslang). I did not try 5.0.

Here is the diff for my ffmpeg.SlackBuild:

Code:

diff --git a/ffmpeg.SlackBuild b/ffmpeg.SlackBuild
old mode 100644
new mode 100755
index 6d4c4eb..7d7d891
--- a/ffmpeg.SlackBuild
+++ b/ffmpeg.SlackBuild
@@ -167,6 +167,12 @@ libdrm=""     ; [ "${LIBDRM:-yes}" != "no" ]      && libdrm="--enable-libdrm"
 # No default patent encumbered features:
 aac=""        ; [ "${AAC:-no}" = "no" ]           && aac="--disable-encoder=aac"
 
+# Options added by Erich
+cuda=""       ; [ "${CUDA:-no}" != "no" ]         && cuda="--enable-cuda-nvcc"
+libnpp=""     ; [ "${LIBNPP:-no}" != "no" ]       && libnpp="--enable-libnpp"
+cuda_cflags=""; [ "${CUDAINCDIR:-no}" != "no" ]   && cuda_cflags="--extra-cflags=-I${CUDAINCDIR}"
+cuda_ldflags="";[ "${CUDALDDIR:-no}" != "no" ]    && cuda_ldflags="--extra-ldflags=-L${CUDALDDIR}"
+
 rm -rf $PKG
 mkdir -p $TMP $PKG
 cd $TMP
@@ -181,6 +187,12 @@ find -L . \
  \( -perm 666 -o -perm 664 -o -perm 640 -o -perm 600 -o -perm 444 \
   -o -perm 440 -o -perm 400 \) -exec chmod 644 {} \+
 
+# Patch for newer CUDA toolkits
+patch -p1 < $CWD/detect-nvcc-without-sm_30-support.patch
+
+# Patch for glslang link flags
+patch -p1 < $CWD/glslang-link-flags.patch
+
 # Fix linking with flite:
 sed -i "s| -lflite\"| -lflite -lm -lasound\"|" \
   ./configure
@@ -203,6 +215,9 @@ CXXFLAGS="$SLKCFLAGS" \
   $non_free \
   $aac \
   $chromaprint \
+  $cuda \
+  $cuda_cflags \
+  $cuda_ldflags \
   $decklink \
   $fdk \
   $fontconfig \
@@ -224,6 +239,7 @@ CXXFLAGS="$SLKCFLAGS" \
   $libiec61883 \
   $libilbc \
   $libmodplug \
+  $libnpp \
   $libopenal \
   $libopencv \
   $libopus \

And my 2 patches:

Code:

$ cat slackbuilds/multimedia/ffmpeg/detect-nvcc-without-sm_30-support.patch 
--- a/configure
+++ b/configure
@@ -4388,8 +4388,16 @@ else
     NVCC_C=""
 fi
 
+set_default nvcc
+
+if enabled cuda_nvcc; then
+    if $nvcc $nvccflags_default 2>&1 | grep -qi unsupported; then
+        nvccflags_default="-gencode arch=compute_60,code=sm_60 -O2"
+    fi
+fi
+
 set_default arch cc cxx doxygen pkg_config ranlib strip sysinclude \
-    target_exec x86asmexe nvcc
+    target_exec x86asmexe
 enabled cross_compile || host_cc_default=$cc
 set_default host_cc

Code:

$ cat slackbuilds/multimedia/ffmpeg/glslang-link-flags.patch 
diff --git a/configure b/configure
index 013780d..1de3802 100755
--- a/configure
+++ b/configure
@@ -6375,7 +6375,7 @@ enabled fontconfig        && enable libfontconfig
 enabled libfontconfig     && require_pkg_config libfontconfig fontconfig "fontconfig/fontconfig.h" FcInit
 enabled libfreetype       && require_pkg_config libfreetype freetype2 "ft2build.h FT_FREETYPE_H" FT_Init_FreeType
 enabled libfribidi        && require_pkg_config libfribidi fribidi fribidi.h fribidi_version_info
-enabled libglslang        && require_cpp libglslang glslang/SPIRV/GlslangToSpv.h "glslang::TIntermediate*" -lglslang -lMachineIndependent -lOSDependent -lHLSL -lOGLCompiler -lGenericCodeGen -lSPVRemapper -lSPIRV -lSPIRV-Tools-opt -lSPIRV-Tools -lpthread -lstdc++
+enabled libglslang        && require_cpp libglslang glslang/SPIRV/GlslangToSpv.h "glslang::TIntermediate*" -lglslang -lOSDependent -lHLSL -lOGLCompiler -lSPVRemapper -lSPIRV -lSPIRV-Tools-opt -lSPIRV-Tools -lpthread -lstdc++
 enabled libgme            && { check_pkg_config libgme libgme gme/gme.h gme_new_emu ||
                                require libgme gme/gme.h gme_new_emu -lgme -lstdc++; }
 enabled libgsm            && { for gsm_hdr in "gsm.h" "gsm/gsm.h"; do

notzed · 03-17-2022, 09:35 PM

lensfun just looks like a version mismatch going by that WONTFIX bug report. But the ffmpeg source doesn't seem to specify which is the correct one, so you might have to experiment a bit.

Looks like opencv is also a version mismatch and the one in slackware is just too new for ffmpeg. It's looking for the pkg-config of 'opencv' but it's 'opencv4' in slackware. Patching the configure script doesn't work because opencv4 seems to require a c++ compiler now. Perhaps you could try an ffmpeg snapshot/git but who knows, ffmpeg developers work to their own schedule and sometimes the filters aren't a priority.

lecramyajiv · 03-18-2022, 01:56 AM

Thanks for the reply guys, today I was checking the Debian package build and apparently opencv can't be enabled and conflicts with the opencv4 package. same with lensfun but I swear when i was using Debian it had been enabled. Glslang can be enabled with newer versions so no problem there. The solution would be do not enable these options for ffmpeg 4.4.1.
I'm attaching the ffmpeg debian.readme and TODO file from ffmpeg_5.0-3.debian.tar.xz file were they talk about it.