QNN SDK 1.17.0
When the following command is executed, an error occurs:
I follow the following procedure, can someone help me find out what the problem is
--------------
Convert Models
--------------
Follow official documentation in docs/general/tutorial2.html "Model Conversion and Build" and
docs/general/tutorial2.html "Execution on Android" in "HTP Backend Execution" up to the point
where you create *.serialized.bin.
For qnn-model-lib-generator, use -t option as needed (eg. if NDK is not available).
--------------------------
Push Files onto QNX Target
--------------------------
The following echoes the instructions from the Android Execution example in docs/general/tutorial2.html
(section "Execution on Android" in "HTP Backend Execution") but for a QNX enviroment. If FileZilla is used
to transfer files from host machine to the target, please make sure the file transfer type is set to binary.
1. Telnet to the QNX terminal and create directories:
* mkdir /data/local/tmp/lib
* mkdir /data/local/tmp/bin
2. FTP the following binaries (from the QNN SDK) to target:
* target/aarch64-qnx/lib/libQnnHtpV68Stub.so to /data/local/tmp/lib
* target/aarch64-qnx/lib/libQnnHtp.so to /data/local/tmp/lib
* target/aarch64-qnx/bin/qnn-net-run to /data/local/tmp/bin
* target/aarch64-qnx/bin/qnn-profile-viewer to /data/local/tmp/bin
3. In the QNX Terminal, enable permissions for QNN binaries:
* chmod 777 /data/local/tmp/bin/qnn-net-run
* chmod 777 /data/local/tmp/bin/qnn-profile-viewer
4. FTP the following binary (from the QNN SDK) to target's /mnt/etc/images/cdsp0/
* target/hexagon-v68/lib/unsigned/libQnnHtpV68Skel.so
5. FTP non-FuSa CDSP images to /mnt/etc/images/
* <QNX_Build_Path>/qnx_ap/prebuilt_NHLOS/cdsp0_adas/non_fusa/asic_8540/cdsp0.mbn
* <QNX_Build_Path>/qnx_ap/prebuilt_NHLOS/cdsp0_adas/non_fusa/asic_8540/cdsp0.mdt
* <QNX_Build_Path>/qnx_ap/prebuilt_NHLOS/cdsp0_adas/non_fusa/asic_8540/shared_obj/fastrpc_shell_3
* <QNX_Build_Path>/qnx_ap/prebuilt_NHLOS/cdsp0_adas/non_fusa/asic_8540/shared_obj/fastrpc_shell_unsigned_3
6. FTP an input list and input images from your host machine to target's /data/local/tmp/bin
* Example images and list can be found in the example in docs/general/tutorial2.html
"Execution on Android" in "HTP Backend Execution".
7. FTP the *serialized.bin to target's /data/local/tmp/bin
* From the Convert Models step
8. Reset the target with the following command in the QNX terminal:
* reset
---------------------------------------
Execute Neural Network using HTP on QNX
---------------------------------------
After following steps from "Push Files onto QNX Target" above, now qnn-net-run and qnn-profile-viewer
can be executed using HTP and QNX.
1. Telnet into the QNX Terminal and setup the execution environment
* cd /data/local/tmp/bin
* export PATH=/data/local/tmp/bin:$PATH
* export VENDOR_LIB=/data/local/tmp/lib
* export LD_LIBRARY_PATH=$VENDOR_LIB:$LD_LIBRARY_PATH
* export ADSP_LIBRARY_PATH="/mnt/etc/images/cdsp0;$VENDOR_LIB"
2. Execute qnn-net-run
* ./qnn-net-run --backend libQnnHtp.so \
--input_list <filename_for_input_list> \
--retrieve_context <serialized_network_binary.bin>
3. Execute qnn-net-run with profiling
* ./qnn-net-run --backend libQnnHtp.so \
--input_list <filename_for_input_list> \
--retrieve_context <serialized_network_binary.bin> \
--profile_level detailed \
--log_level error
Please read docs/general/tools.html#qnn-net-run for more options for qnn-net-run.
# echo $VENDOR_LIB