Justin Ruggles <[email protected]> writes:
> This allows for testing floating-point audio encoders across different
> platforms where exact comparisons are unreliable due to float rounding
> differences.
> ---
> tests/Makefile | 2 +-
> tests/fate-run.sh | 22 +++++++++++++++++++++-
> 2 files changed, 22 insertions(+), 2 deletions(-)
>
> diff --git a/tests/Makefile b/tests/Makefile
> index 5c65237..79bbba5 100644
> --- a/tests/Makefile
> +++ b/tests/Makefile
> @@ -116,7 +116,7 @@ fate: $(FATE)
>
> $(FATE): avconv$(EXESUF) $(FATE_UTILS:%=tests/%$(HOSTEXESUF))
> @echo "TEST $(@:fate-%=%)"
> - $(Q)$(SRC_PATH)/tests/fate-run.sh $@ "$(SAMPLES)" "$(TARGET_EXEC)"
> "$(TARGET_PATH)" '$(CMD)' '$(CMP)' '$(REF)' '$(FUZZ)' '$(THREADS)'
> '$(THREAD_TYPE)' '$(CPUFLAGS)'
> + $(Q)$(SRC_PATH)/tests/fate-run.sh $@ "$(SAMPLES)" "$(TARGET_EXEC)"
> "$(TARGET_PATH)" '$(CMD)' '$(CMP)' '$(REF)' '$(FUZZ)' '$(THREADS)'
> '$(THREAD_TYPE)' '$(CPUFLAGS)' '$(PSNR_SHIFT)'
>
> fate-list:
> @printf '%s\n' $(sort $(FATE))
> diff --git a/tests/fate-run.sh b/tests/fate-run.sh
> index 1d1540e..b48be6d 100755
> --- a/tests/fate-run.sh
> +++ b/tests/fate-run.sh
> @@ -18,6 +18,7 @@ fuzz=${8:-1}
> threads=${9:-1}
> thread_type=${10:-frame+slice}
> cpuflags=${11:-all}
> +psnr_shift=${12:-0}
>
> outdir="tests/data/fate"
> outfile="${outdir}/${test}"
> @@ -26,7 +27,7 @@ cmpfile="${outdir}/${test}.diff"
> repfile="${outdir}/${test}.rep"
>
> do_tiny_psnr(){
> - psnr=$(tests/tiny_psnr "$1" "$2" 2 0 0)
> + psnr=$(tests/tiny_psnr "$1" "$2" 2 $psnr_shift 0)
> val=$(expr "$psnr" : ".*$3: *\([0-9.]*\)")
> size1=$(expr "$psnr" : '.*bytes: *\([0-9]*\)')
> size2=$(expr "$psnr" : '.*bytes:[ 0-9]*/ *\([0-9]*\)')
> @@ -46,6 +47,12 @@ stddev(){
> do_tiny_psnr "$1" "$2" stddev 0 1
> }
>
> +stddev2(){
> + psnr1=$(cat $psnr_ref)
> + ref_val=$(expr "$psnr1" : ".*stddev: *\([0-9.]*\)")
Why so complicated? Why not simply pass the reference value directly
instead of storing the full tiny_psnr output?
> + do_tiny_psnr "$1" "$2" stddev $ref_val 0
> +}
I also don't see the need for this being a separate function. The same
can be achieved using a few extra parameters to the existing one, with
default values giving the current behaviour.
> run(){
> test "${V:-0}" -gt 0 && echo "$target_exec" $target_path/"$@" >&3
> $target_exec $target_path/"$@"
> @@ -75,6 +82,18 @@ pcm(){
> avconv "$@" -vn -f s16le -
> }
>
> +enc_dec_pcm(){
> + psnr_ref="${base}/ref/fate/${test}"
> + out_fmt=$1
> + in_ch=$2
> + in_sr=$3
> + in_fmt=$4
> + shift 4
> + encfile="${outdir}/${test}.${out_fmt}"
$encfile should be added to $cleanfiles so it gets deleted after the
test is run.
> + avconv -channels $in_ch -sample_rate $in_sr -f $in_fmt -i $ref "$@" -f
> $out_fmt -y $encfile
> + avconv -i $encfile -vn -f $in_fmt -
> +}
Perhaps this should abort if the encoding fails.
> regtest(){
> t="${test#$2-}"
> ref=${base}/ref/$2/$t
> @@ -125,6 +144,7 @@ if test -e "$ref"; then
> diff) diff -u -w "$ref" "$outfile" >$cmpfile ;;
> oneoff) oneoff "$ref" "$outfile" >$cmpfile ;;
> stddev) stddev "$ref" "$outfile" >$cmpfile ;;
> + stddev2) stddev2 "$ref" "$outfile" >$cmpfile ;;
> null) cat "$outfile" >$cmpfile ;;
> esac
> cmperr=$?
> --
> 1.7.1
>
--
Måns Rullgård
[email protected]
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
