pytorch

b7b481bd - [PyTorch] Enable template build at aten operator level (#53801)

[PyTorch] Enable template build at aten operator level (#53801) Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/53801 ## Summary Enable partial explicit Aten level sources list for lite interpreter. More aten level source list will be added. 1. Use `gen_selected_mobile_ops_header.py ` to generate `selected_mobile_ops.h`. Currently, it only includes selected operators, and dtypes is all. 2. Add a custom target includes only `seleteted_mobile_ops.h`, and add it to `torch_cpu` dependency, when `BUILD_LITE_INTERPRETER` is enabled. As a note, the current input yaml file is slightly different than the one use in internal. Will align these two yaml as next step. **Android** x86: `SELECTED_OP_LIST=/Users/chenlai/Documents/pytorch/experiemnt/deeplabv3_scripted.yaml BUILD_LITE_INTERPRETER=1 ./scripts/build_pytorch_android.sh x86` libpytorch_jni_lite.so -- 3.4 MB armeabi-v7a `SELECTED_OP_LIST=/Users/chenlai/Documents/pytorch/experiemnt/deeplabv3_scripted.yaml BUILD_LITE_INTERPRETER=1 ./scripts/build_pytorch_android.sh armeabi-v7a` libpytorch_jni_lite.so -- 2.5 MB **iOS:** ``` (base) chenlai@chenlai-mp install % du -sh * 15M include 57M lib 2.8M share ``` ``` (base) chenlai@chenlai-mp lib % ls -lh total 117296 -rw-r--r-- 1 chenlai staff 3.2M Mar 15 22:03 libXNNPACK.a -rw-r--r-- 1 chenlai staff 913K Mar 15 22:03 libc10.a -rw-r--r-- 1 chenlai staff 4.6K Mar 15 22:03 libclog.a -rw-r--r-- 1 chenlai staff 42K Mar 15 22:03 libcpuinfo.a -rw-r--r-- 1 chenlai staff 1.5M Mar 15 22:03 libeigen_blas.a -rw-r--r-- 1 chenlai staff 44K Mar 15 22:03 libpthreadpool.a -rw-r--r-- 1 chenlai staff 166K Mar 15 22:03 libpytorch_qnnpack.a -rw-r--r-- 1 chenlai staff 384B Mar 15 22:03 libtorch.a -rw-r--r-- 1 chenlai staff 51M Mar 15 22:03 libtorch_cpu.a ``` ### **Master (Baseline):** **Android** x86: `SELECTED_OP_LIST=/Users/chenlai/Documents/pytorch/experiemnt/deeplabv3_scripted.yaml BUILD_LITE_INTERPRETER=1 ./scripts/build_pytorch_android.sh x86` libpytorch_jni_lite.so -- 3.8 MB armeabi-v7a `SELECTED_OP_LIST=/Users/chenlai/Documents/pytorch/experiemnt/deeplabv3_scripted.yaml BUILD_LITE_INTERPRETER=1 ./scripts/build_pytorch_android.sh armeabi-v7a` libpytorch_jni_lite.so -- 2.8 MB **iOS:** ``` (base) chenlai@chenlai-mp install % du -sh * 15M include 58M lib 2.8M share ``` ``` (base) chenlai@chenlai-mp lib % ls -lh total 119600 -rw-r--r-- 1 chenlai staff 3.2M Mar 4 23:16 libXNNPACK.a -rw-r--r-- 1 chenlai staff 910K Mar 4 23:16 libc10.a -rw-r--r-- 1 chenlai staff 4.6K Mar 4 23:16 libclog.a -rw-r--r-- 1 chenlai staff 42K Mar 4 23:16 libcpuinfo.a -rw-r--r-- 1 chenlai staff 1.5M Mar 4 23:16 libeigen_blas.a -rw-r--r-- 1 chenlai staff 44K Mar 4 23:16 libpthreadpool.a -rw-r--r-- 1 chenlai staff 166K Mar 4 23:16 libpytorch_qnnpack.a -rw-r--r-- 1 chenlai staff 384B Mar 4 23:16 libtorch.a -rw-r--r-- 1 chenlai staff 52M Mar 4 23:16 libtorch_cpu.a ``` Test Plan: Imported from OSS Reviewed By: dhruvbird Differential Revision: D27074814 Pulled By: cccclai fbshipit-source-id: 762b5ad5b87b6a262444392fd089249c4837ba18