continuation from a previous thread [Cin]Re: Adding CIN_HW_DEV=qsv for GPU HW decoding? https://lists.cinelerra-gg.org/archives/list/[email protected]/mess... The current Slowroll update on Alder Lake /DG2 verifies use of vainfo | grep version libva info: VA-API version 1.23.0 libva info: Trying to open /usr/lib64/dri/iHD_drv_video.so libva info: Found init function __vaDriverInit_1_23 libva info: va_openDriver() returns 0 vainfo: VA-API version: 1.23 (libva 2.23.0) vainfo: Driver version: Intel iHD driver for Intel(R) Gen Graphics - 26.1.1 () Before further testing I had to rebuild my Cingg already patched with 0001-QSV-hw-decoder-test.patch cd /home3/cinelerra/cinelerra-5.1 make clean Edited the build.sh as a "current, all-in-one" ./configure line with libsvtav1, onevpl og vulkan cat bld.sh #!/bin/bash # IMPORTANT comments below to change the configure line # For newer operating system versions, add --enable-libsvtav1 # For really old versions, such as ubuntu 16, add --enable-libaom=no # Faster compile if include --disable-x265_hidepth # To add OpenCV plugins, add --with-opencv=sta,tar=https://download.cinelerra-gg.org/download.php?file=opencv%2Fopencv-20200306... # # original ./configure --with-single-user --with-booby # # 12.03.2026 extended ./configure with Intel QSV api and Vulkan api and filters # --with-onevpl Intel QSV api for unix (no) # --with-vulkan Vulkan api (no) # --with-libplacebo libplacebo ffmpeg Vulkan filter (no) # --with-libzimg libzimg ffmpeg filter (no) # ( ./autogen.sh ./configure --with-single-user --with-booby --enable-libsvtav1 --with-onevpl --with-vulkan --with-libplacebo --with-libzimg make && make install ) 2>&1 | tee log mv Makefile Makefile.cfg cp Makefile.devel Makefile This required two additional dependency packages on Slowroll before the script ran through zypper in libplacebo-devel zimg-devel ./bld.sh -------------------------- cd /home3/cinelerra/cinelerra-5.1 bin/cin Cinelerra Infinity - built: Mar 12 2026 23:52:16 Settings|Preferences|Performance|Use HW Device: none ---- Testing QSV decoding in CPU vs GPU respectively, that is before and after adding remap decoder mpeg2_qsv, shows that decoding in GPU isn't different faster than on CPU in this case, but moving most workload to the GPU should offload the CPU: cat bin/ffmpeg/decode.opts # apply at init decode loglevel=fatal formatprobesize=5000000 scan_all_pmts=1 remap_video_decoder libaom-av1=libdav1d remap_video_decoder mpeg2video=mpeg2_qsv mpeg2video -> h264_qsv ====================== CPU: ** rendered 5972 frames in 24.463 secs, 244.124 fps audio0 pad 64 0 (64) intel_gpu_top Video: < = 29% ------------- GPU: ** rendered 5972 frames in 24.562 secs, 243.140 fps audio0 pad 64 0 (64) intel_gpu_top Video: < = 42% mpeg2video -> hevc_qsv ====================== CPU: ** rendered 5972 frames in 21.910 secs, 272.570 fps audio0 pad 64 0 (64) intel_gpu_top Video: < = 24% -------------- GPU: ** rendered 5972 frames in 22.666 secs, 263.478 fps audio0 pad 64 0 (64) intel_gpu_top Video: < = 38% mpeg2video -> av1_qsv ====================== CPU: ** rendered 5972 frames in 24.330 secs, 245.458 fps audio0 pad 64 0 (64) intel_gpu_top Video: < = 24% -------- GPU: ** rendered 5972 frames in 24.451 secs, 244.244 fps audio0 pad 64 0 (64) intel_gpu_top Video: < = 39%