chunseoklee

On https://github.com/Samsung/ONE/pull/8481/commits/fd645eaf8c5b15769ca1bad7c3882c0c294c9d1b Need to reduce build time. it takes 840s till the following : ``` [ 840s] [100%] Generating ArgMin_U8_003.gen.tflite [ 840s] [100%] Generating AveragePool2D_000.gen.tflite [ 840s] [100%] Generating AveragePool2D_U8_000.gen.tflite...

on 5ec7b6a, fail at 1923s in ``` [ 1923s] /home/abuild/rpmbuild/BUILD/nnfw-1.20.0/externals/TENSORFLOW-2.6.0/tensorflow/lite/kernels/internal/optimized/neon_tensor_utils.h:30: undefined reference to `tflite::tensor_utils::NeonMatrixBatchVe ctorMultiplyAccumulate(float const*, int, int, float const*, int, float*)' [ 1923s] /usr/lib/gcc/armv7l-tizen-linux-gnueabi/9.2.0/../../../../armv7l-tizen-linux-gnueabi/bin/ld: ../kernels/libluci_interpreter_linux_pal.a(tensor_utils.cc.o): in function `tflite::tensor_utils::Matrix BatchVectorMultiplyAccumulate(signed...

On fe0f477, ``` 1725s] /home/abuild/rpmbuild/BUILD/nnfw-1.20.0/externals/TENSORFLOW-2.6.0-RUY/ruy/pack_arm.h:601: undefined reference to `ruy::Pack8bitRowMajorForNeon(unsigned char const*, int, int, int, int, int, int, s igned char*, int, int, int*, int, int)' [ 1725s] /usr/lib/gcc/armv7l-tizen-linux-gnueabi/9.2.0/../../../../armv7l-tizen-linux-gnueabi/bin/ld: libluci_interpreter_linux_pal.a(neon_tensor_utils.cc.o): in...

> Need to reduce build time. it takes 840s till the following : Unfortunately, tizen build system for gcc already accelerated !!! ``` ➜ grep accel /home/twoflower/GBS-ROOT-3.0TM1_llvm38/local/BUILD-ROOTS/scratch.armv7hl.0/tizen.conf Preinstall: qemu-accel-%{build_hostarch}-armv7l Runscripts:...

- cross arm32 compile_commands.json note that no string with neon in the file. ``` json { "directory": "/home/twoflower/dev/ONE/build/arm32.debug/compiler/luci-interpreter/src", "command": "/home/twoflower/dev/gcc-linaro-7.2.1-2017.11-x86_64_arm-linux-gnueabihf/bin/arm-linux-gnueabihf-g++ --sysroot=/home/twoflower/dev/rootfs/arm-ubuntu/arm -Dluci_interpreter_EXPORTS -I/home/twoflower/dev/ONE/compiler/luci-interpreter/include -I/home/twoflower/dev/ONE/compiler/luci-interpreter/src -I/home/twoflower/dev/ONE/compiler/luci/lang/include -I/home/twoflower/dev/ONE/compiler/loco/include -I/home/twoflower/dev/ONE/compiler/angkor/include -I/home/twoflower/dev/ONE/compiler/oops/include -I/home/twoflower/dev/ONE/compiler/pepper-str/include -I/home/twoflower/dev/ONE/compiler/luci-interpreter/pal/linux...

now, build is fine ob https://github.com/Samsung/ONE/pull/8481/commits/0be8a492a40c5ae92ac3d2a0f3edf708b67163c7. 1 test failed : ``` [ 1497s] [ OK ] QuantizedModelVerifierTest.LocalCreateConst (0 ms) [ 1497s] [ RUN ] QuantizedModelVerifierTest.InstanceNorm [ 1497s] QuantizeWithMinMaxPass Start [...

Now, all the tests passed on https://github.com/Samsung/ONE/pull/8481/commits/4b12f67c73e3ebd1633da4102da99c73469aa294. But, I am not sure that it is resolved since the commit avoids failed asserts by release build.

@binarman @SlavikMIPT Is it possible to investigate similar ( https://github.com/Samsung/ONE/issues/5080#issuecomment-743022914) performance/memory comparison with the latest tf-micro as of now ?

> "toy" example means sine generator model on tf-micro site, right ?

@BalyshevArtem Do you have benchmark result with luci-interpreter? (The result above is the one with tf-micro?)