#!/bin/bash
# Copyright 2020-2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
#  * Redistributions of source code must retain the above copyright
#    notice, this list of conditions and the following disclaimer.
#  * Redistributions in binary form must reproduce the above copyright
#    notice, this list of conditions and the following disclaimer in the
#    documentation and/or other materials provided with the distribution.
#  * Neither the name of NVIDIA CORPORATION nor the names of its
#    contributors may be used to endorse or promote products derived
#    from this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
# PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT OWNER OR
# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

############################################################################
## This script generates the model repository needed for TensorRT testing
## on the Jetson device. Generating these models requires having TensorRT
## container.
############################################################################
#!/bin/bash -xe
# Make all generated files accessible outside of container
umask 0000
# Set the version of the models
TRITON_VERSION=${TRITON_VERSION:=26.07}
# Set the CUDA device to use
NVIDIA_VISIBLE_DEVICES=${NVIDIA_VISIBLE_DEVICES:=0}
# Set TensorRT image
TENSORRT_IMAGE=${TENSORRT_IMAGE:=nvcr.io/nvidia/tensorrt:$TRITON_VERSION-py3-igpu}
UBUNTU_IMAGE=${UBUNTU_IMAGE:=ubuntu:24.04}

# Set CI specific parameters
DOCKER_GPU_ARGS=${DOCKER_GPU_ARGS:-$([[ -v RUNNER_GPUS && $RUNNER_GPUS =~ ^[0-9] ]] && eval $NV_DOCKER_ARGS || echo "--runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=$NVIDIA_VISIBLE_DEVICES" )}

############################################################################
# Check if Docker volume exists
############################################################################
CI_JOB_ID=${CI_JOB_ID:=$(date +%Y%m%d_%H%M)}
DOCKER_VOLUME=${DOCKER_VOLUME:=volume_gen_qa_model_repositor_${CI_JOB_ID}}
if ! docker volume inspect $DOCKER_VOLUME > /dev/null 2>&1; then
    echo "Docker volume $DOCKER_VOLUME does not exist. Creating..."
    docker volume create $DOCKER_VOLUME
    docker volume inspect $DOCKER_VOLUME
fi

docker rm -f $DOCKER_VOLUME
docker run --rm -v $DOCKER_VOLUME:/mnt -w /mnt/$CI_JOB_ID $UBUNTU_IMAGE mkdir -p gen_srcdir ${TRITON_VERSION}
docker create --name $DOCKER_VOLUME -v $DOCKER_VOLUME:/mnt -w /mnt/$CI_JOB_ID $UBUNTU_IMAGE
docker cp . $DOCKER_VOLUME:/mnt/$CI_JOB_ID/gen_srcdir

# Set model output directories
VOLUME_BUILD_DIR=${VOLUME_BUILD_DIR:=/mnt/$CI_JOB_ID}
VOLUME_SRCDIR=${VOLUME_SRCDIR:=$VOLUME_BUILD_DIR/gen_srcdir}

VOLUME_DESTDIR=$VOLUME_BUILD_DIR/$TRITON_VERSION/qa_model_repository
VOLUME_DATADEPENDENTDIR=$VOLUME_BUILD_DIR/$TRITON_VERSION/qa_trt_data_dependent_model_repository
VOLUME_DYNASEQDESTDIR=$VOLUME_BUILD_DIR/$TRITON_VERSION/qa_dyna_sequence_model_repository
VOLUME_DYNASEQIMPLICITDESTDIR=$VOLUME_BUILD_DIR/$TRITON_VERSION/qa_dyna_sequence_implicit_model_repository
VOLUME_FORMATDESTDIR=$VOLUME_BUILD_DIR/$TRITON_VERSION/qa_trt_format_model_repository
VOLUME_IDENTITYBIGDESTDIR=$VOLUME_BUILD_DIR/$TRITON_VERSION/qa_identity_big_model_repository
VOLUME_IDENTITYDESTDIR=$VOLUME_BUILD_DIR/$TRITON_VERSION/qa_identity_model_repository
VOLUME_IMPLICITSEQDESTDIR=$VOLUME_BUILD_DIR/$TRITON_VERSION/qa_sequence_implicit_model_repository
VOLUME_RAGGEDDESTDIR=$VOLUME_BUILD_DIR/$TRITON_VERSION/qa_ragged_model_repository
VOLUME_RESHAPEDESTDIR=$VOLUME_BUILD_DIR/$TRITON_VERSION/qa_reshape_model_repository
VOLUME_SEQDESTDIR=$VOLUME_BUILD_DIR/$TRITON_VERSION/qa_sequence_model_repository
VOLUME_SHAPEDESTDIR=$VOLUME_BUILD_DIR/$TRITON_VERSION/qa_shapetensor_model_repository
VOLUME_VARDESTDIR=$VOLUME_BUILD_DIR/$TRITON_VERSION/qa_variable_model_repository
VOLUME_VARIMPLICITSEQDESTDIR=$VOLUME_BUILD_DIR/$TRITON_VERSION/qa_variable_sequence_implicit_model_repository
VOLUME_VARSEQDESTDIR=$VOLUME_BUILD_DIR/$TRITON_VERSION/qa_variable_sequence_model_repository

# Clean up host working directory
docker run --rm -v $DOCKER_VOLUME:/mnt -w /mnt/$CI_JOB_ID $UBUNTU_IMAGE \
mkdir -p \
$VOLUME_SRCDIR \
$VOLUME_DESTDIR \
$VOLUME_DATADEPENDENTDIR \
$VOLUME_DYNASEQDESTDIR \
$VOLUME_DYNASEQIMPLICITDESTDIR \
$VOLUME_FORMATDESTDIR \
$VOLUME_IDENTITYBIGDESTDIR \
$VOLUME_IDENTITYDESTDIR \
$VOLUME_IMPLICITSEQDESTDIR \
$VOLUME_RAGGEDDESTDIR \
$VOLUME_RESHAPEDESTDIR \
$VOLUME_SEQDESTDIR \
$VOLUME_SHAPEDESTDIR \
$VOLUME_VARDESTDIR \
$VOLUME_VARIMPLICITSEQDESTDIR \
$VOLUME_VARSEQDESTDIR

# Set TensorRT model generation script name
TRT_MODEL_SCRIPT=gen.TensorRT.gen_jetson_trt_models.cmds

# Set script to generate TensorRT models
cat > $TRT_MODEL_SCRIPT <<EOF
#!/bin/bash -xe
# Make all generated files accessible outside of container
umask 0000
nvidia-smi --query-gpu=compute_cap,compute_mode,driver_version,name,index --format=csv || true
export TRT_SUPPRESS_DEPRECATION_WARNINGS=1
ldconfig || true

cd $VOLUME_SRCDIR
# Models using shape tensor i/o
python3 $VOLUME_SRCDIR/gen_qa_identity_models.py --tensorrt-shape-io --models_dir=$VOLUME_SHAPEDESTDIR
python3 $VOLUME_SRCDIR/gen_qa_sequence_models.py --tensorrt-shape-io --models_dir=$VOLUME_SHAPEDESTDIR
python3 $VOLUME_SRCDIR/gen_qa_dyna_sequence_models.py --tensorrt-shape-io --models_dir=$VOLUME_SHAPEDESTDIR
python3 $VOLUME_SRCDIR/gen_qa_models.py --tensorrt --models_dir=$VOLUME_DESTDIR
python3 $VOLUME_SRCDIR/gen_qa_models.py --tensorrt --variable --models_dir=$VOLUME_VARDESTDIR
python3 $VOLUME_SRCDIR/gen_qa_identity_models.py --tensorrt --models_dir=$VOLUME_IDENTITYDESTDIR
python3 $VOLUME_SRCDIR/gen_qa_identity_models.py --tensorrt-big --models_dir=$VOLUME_IDENTITYBIGDESTDIR
python3 $VOLUME_SRCDIR/gen_qa_reshape_models.py --tensorrt --variable --models_dir=$VOLUME_RESHAPEDESTDIR
python3 $VOLUME_SRCDIR/gen_qa_sequence_models.py --tensorrt --models_dir=$VOLUME_SEQDESTDIR
python3 $VOLUME_SRCDIR/gen_qa_implicit_models.py --tensorrt --models_dir=$VOLUME_IMPLICITSEQDESTDIR
python3 $VOLUME_SRCDIR/gen_qa_implicit_models.py --tensorrt --variable --models_dir=$VOLUME_VARIMPLICITSEQDESTDIR
python3 $VOLUME_SRCDIR/gen_qa_dyna_sequence_models.py --tensorrt --models_dir=$VOLUME_DYNASEQDESTDIR
python3 $VOLUME_SRCDIR/gen_qa_sequence_models.py --tensorrt --variable --models_dir=$VOLUME_VARSEQDESTDIR
python3 $VOLUME_SRCDIR/gen_qa_dyna_sequence_implicit_models.py --tensorrt --models_dir=$VOLUME_DYNASEQIMPLICITDESTDIR
python3 $VOLUME_SRCDIR/gen_qa_ragged_models.py --tensorrt --models_dir=$VOLUME_RAGGEDDESTDIR
python3 $VOLUME_SRCDIR/gen_qa_trt_format_models.py --models_dir=$VOLUME_FORMATDESTDIR
python3 $VOLUME_SRCDIR/gen_qa_trt_data_dependent_shape.py --models_dir=$VOLUME_DATADEPENDENTDIR

EOF

chmod a+x $TRT_MODEL_SCRIPT

docker cp $TRT_MODEL_SCRIPT $DOCKER_VOLUME:$VOLUME_SRCDIR

docker pull $TENSORRT_IMAGE

docker run $DOCKER_GPU_ARGS \
   --rm -v $DOCKER_VOLUME:/mnt \
   -e TRT_VERBOSE \
  $TENSORRT_IMAGE bash -xe $VOLUME_SRCDIR/$TRT_MODEL_SCRIPT

# Copy generated models to /tmp/ if not running in CI
if [ -z $CI ] ; then
    echo "Copying generated models to /tmp/"
    docker cp $DOCKER_VOLUME:$VOLUME_BUILD_DIR/$TRITON_VERSION /tmp/
    echo "Removing Docker volume $DOCKER_VOLUME"
    docker rm -f $DOCKER_VOLUME
    docker volume rm $DOCKER_VOLUME
fi