#!/bin/bash
#
# tst_api_demod_check
#
# Setup input and reference data for one of several versions of this test.

# Find the scripts directory
SCRIPTS="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null && pwd )"

# Setup common variables
source $SCRIPTS/run_tests_common.sh

# RUN_DIR - Directory where test will be run
RUN_DIR="${UNITTEST_BASE}/test_run/${FULL_TEST_NAME}"

# Call common setup function to make the directory
setup_common "${RUN_DIR}"

# Change to test directory
cd "${RUN_DIR}"

# way of performing a rough comparison of two output speech files that are not exactly the same

function compare_energy() {
    energy_ref=$(python3 -c "import numpy as np; x=np.fromfile(\"ref_demod.raw\",dtype=\"int16\").astype(float); print(10*np.log10(np.dot(x,x)))")
    energy_target=$(python3 -c "import numpy as np; x=np.fromfile(\"stm_out.raw\",dtype=\"int16\").astype(float); print(10*np.log10(np.dot(x,x)))")
    printf "ref energy: %f target energy: %f\n" $energy_ref $energy_target
	
    python3 -c "import sys; sys.exit(1) if abs($energy_ref-$energy_target) < 1 else sys.exit(0)"
    if [[ $? -eq 1 ]]; then echo "energy compare OK"; 
    else echo "energy compare BAD"; 
     	 let Fails=($Fails + 1)
    fi
}

#####################################################################
## Test CHECK actions:

declare -i Fails=0

case "${TEST_OPT}" in

    700D_plain_test)
	echo "Check reference decode"
	p1=$(grep '^BER\.*: 0.000' ref_gen.log | wc -l)
	p2=$(grep '^Coded BER: 0.000' ref_gen.log | wc -l)
	if [[ $p1 -eq 1 && $p2 -eq 1 ]]; then echo "OK"; 
	else echo "BAD"; 
     	    let Fails=($Fails + 1)
     	fi
	#
	echo "Check target decode"
	p1=$(grep '^BER\.*: 0.000' stderr.log | wc -l)
	p2=$(grep '^Coded BER: 0.000' stderr.log | wc -l)
	if [[ $p1 -eq 1 && $p2 -eq 1 ]]; then echo "OK"; 
	else echo "BAD"; 
     	    let Fails=($Fails + 1)
     	fi
	;;

    700D_AWGN_test)
	echo "Check reference decode"
	uber_ref=$(cat ref_gen.log | sed -n "s/^BER.*: \([0-9..]*\).*Tbits.*/\1/p")
	cber_ref=$(cat ref_gen.log | sed -n "s/^Coded BER.*: \([0-9..]*\).*Tbits.*/\1/p")
	printf "REF   uncoded BER: %f coded BER: %f\n" $uber_ref $cber_ref

	# As per notes in tst_api_demod_setup, coded BER is unreliable
	# for such a short test, so we'll just sanity check the
	# reference uncoded BER here. Bash can't compare floats
	# .... so use return code of some python script
	python3 -c "import sys; sys.exit(1) if $uber_ref < 0.1 else sys.exit(0)"
	if [[ $? -eq 1 ]]; then echo "OK"; 
	else echo "BAD"; 
     	    let Fails=($Fails + 1)
     	fi

	echo "Check target decode"
	uber_target=$(cat stderr.log | sed -n "s/^BER.*: \([0-9..]*\).*Tbits.*/\1/p")
	cber_target=$(cat stderr.log | sed -n "s/^Coded BER.*: \([0-9..]*\).*Tbits.*/\1/p")
	printf "TARGET uncoded BER: %f coded BER: %f\n" $uber_target $cber_target
	python3 -c "import sys; sys.exit(1) if $uber_target < 0.1 and abs($cber_ref-$cber_target) < 0.01 else sys.exit(0)"
	if [[ $? -eq 1 ]]; then echo "OK"; 
	else echo "BAD"; 
     	    let Fails=($Fails + 1)
     	fi
	;;

    700D_AWGN_codec)
	# 1/ The two output files sound OK, and when plotted look very
	#    similar, but they don't match on a sample-sample basis.
	
	# 2/ Suspect some small state difference, or perhaps random
	#    number generator diverging, sampling codec2_rand() at the
	#    end of the x86 and stm32 test programs might show up any
	#    differences.

	# 3/ At this stage - we can't make sample by sample automatic
	#    tests work.  However there is value in running the test
	#    to ensure no asserts are hit and the code doesn't crash
	#    (e.g. due to an out of memory issues). A simple energy
	#    comparison is used on the output speech files, which
	#    will trap any large errors.

	# 4/ We can also manually evaluate the output decoded speech by
	#    listening to the output speech files.

	compare_energy;

	# make sure execution time stays within bounds
	execution_time=mktmp
	cat stdout.log | sed  -n "s/.*freedv_rx \([0-9..]*\) msecs/\1/p" > $execution_time
	python3 -c "import sys; import numpy as np;  x=np.loadtxt(\"$execution_time\"); print(\"execution time max:: %5.2f mean: %5.2f ms\" % (np.max(x), np.mean(x))); sys.exit(1) if np.max(x) < 80.0 else sys.exit(0)"
	if [[ $? -eq 1 ]]; then echo "execution time OK"; 
        else echo "BAD"; 
     	   let Fails=($Fails + 1)
        fi

	;;

    700E_plain_test)
	echo "Check reference decode"
	p1=$(grep '^BER\.*: 0.000' ref_gen.log | wc -l)
	p2=$(grep '^Coded BER: 0.000' ref_gen.log | wc -l)
	if [[ $p1 -eq 1 && $p2 -eq 1 ]]; then echo "OK"; 
	else echo "BAD"; 
     	    let Fails=($Fails + 1)
     	fi
	#
	echo "Check target decode"
	p1=$(grep '^BER\.*: 0.000' stderr.log | wc -l)
	p2=$(grep '^Coded BER: 0.000' stderr.log | wc -l)
	if [[ $p1 -eq 1 && $p2 -eq 1 ]]; then echo "OK"; 
	else echo "BAD"; 
     	    let Fails=($Fails + 1)
     	fi
	;;

    700E_AWGN_test)
	echo "Check reference decode"
	uber_ref=$(cat ref_gen.log | sed -n "s/^BER.*: \([0-9..]*\).*Tbits.*/\1/p")
	cber_ref=$(cat ref_gen.log | sed -n "s/^Coded BER.*: \([0-9..]*\).*Tbits.*/\1/p")
	printf "REF   uncoded BER: %f coded BER: %f\n" $uber_ref $cber_ref

	# As per notes in tst_api_demod_setup, coded BER is unreliable
	# for such a short test, so we'll just sanity check the
	# reference uncoded BER here. Bash can't compare floats
	# .... so use return code of some python script
	python3 -c "import sys; sys.exit(1) if $uber_ref < 0.1 else sys.exit(0)"
	if [[ $? -eq 1 ]]; then echo "OK"; 
	else echo "BAD"; 
     	    let Fails=($Fails + 1)
     	fi

	echo "Check target decode"
	uber_target=$(cat stderr.log | sed -n "s/^BER.*: \([0-9..]*\).*Tbits.*/\1/p")
	cber_target=$(cat stderr.log | sed -n "s/^Coded BER.*: \([0-9..]*\).*Tbits.*/\1/p")
	printf "TARGET uncoded BER: %f coded BER: %f\n" $uber_target $cber_target
	python3 -c "import sys; sys.exit(1) if $uber_target < 0.1 and abs($cber_ref-$cber_target) < 0.01 else sys.exit(0)"
	if [[ $? -eq 1 ]]; then echo "OK"; 
	else echo "BAD"; 
     	    let Fails=($Fails + 1)
     	fi
	;;

    700E_AWGN_codec)
	# 1/ The two output files sound OK, and when plotted look very
	#    similar, but they don't match on a sample-sample basis.
	
	# 2/ Suspect some small state difference, or perhaps random
	#    number generator diverging, sampling codec2_rand() at the
	#    end of the x86 and stm32 test programs might show up any
	#    differences.

	# 3/ At this stage - we can't make sample by sample automatic
	#    tests work.  However there is value in running the test
	#    to ensure no asserts are hit and the code doesn't crash
	#    (e.g. due to an out of memory issues). A simple energy
	#    comparison is used on the output speech files, which
	#    will trap any large errors.

	# 4/ We can also manually evaluate the output decoded speech by
	#    listening to the output speech files.

	compare_energy;

	# make sure execution time stays within bounds
	execution_time=mktmp
	cat stdout.log | sed  -n "s/.*freedv_rx \([0-9..]*\) msecs/\1/p" > $execution_time
	python3 -c "import sys; import numpy as np;  x=np.loadtxt(\"$execution_time\"); print(\"execution time max:: %5.2f mean: %5.2f ms\" % (np.max(x), np.mean(x))); sys.exit(1) if np.max(x) < 80.0 else sys.exit(0)"
	if [[ $? -eq 1 ]]; then echo "execution time OK"; 
        else echo "BAD"; 
     	   let Fails=($Fails + 1)
        fi

	;;

    1600_plain_codec)
	compare_energy;
	;;

    *) 
       	printf "ERROR: invalid test option. Valid options are:\n  700D_plain_test\n  700D_AWGN_test\n  700D_plain_codec\n  1600_plain_codec\n"
	exit 1

    esac

if (( $Fails == 0 )); then
    echo -e "\nTest PASSED"
else
    echo -e "\nTest FAILED!"
fi


exit $Fails