Skip to content

P
Paddle-Lite

PaddlePaddle / Paddle-Lite

通知 338
Star 4
Fork 1
P
Paddle-Lite
提交 dff4a781 编写于 作者: water94's avatar water94 提交者： GitHub
浏览文件
操作

Merge pull request #213 from allonli/develop 

add some gitignore option & modify to 2 spaces indent & format code & rm build folder
上级 e35ef6fe ba186e1d
隐藏空白更改
内联 并排
Showing with 13546 addition and 14089 deletion
.clang-format
浏览文件 @ dff4a781
...@@ -2,5 +2,4 @@ ...@@ -2,5 +2,4 @@
Language: Cpp Language: Cpp
BasedOnStyle: LLVM BasedOnStyle: LLVM
Standard: Cpp11 Standard: Cpp11
IndentWidth: 4
... ...
.gitignore
浏览文件 @ dff4a781
...@@ -54,5 +54,6 @@ paddle-mobile.cbp ...@@ -54,5 +54,6 @@ paddle-mobile.cbp
compile_commands.json compile_commands.json
cmake-build-debug/ cmake-build-debug/
cmake-build-release/
test/models/ test/models/
\ No newline at end of file
cmake-build-release/compile_commands.json 已删除 100644 → 0
浏览文件 @ e35ef6fe
[
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/framework/ddim.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/ddim.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/ddim.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/framework/lod_tensor.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/lod_tensor.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/lod_tensor.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/framework/scope.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/scope.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/scope.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/framework/tensor_util.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/tensor_util.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/tensor_util.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/memory/t_malloc.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/memory/t_malloc.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/memory/t_malloc.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/operators/math/im2col.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/math/im2col.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/math/im2col.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/operators/math/math_function.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/math/math_function.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/math/math_function.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/operators/math/vol2col.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/math/vol2col.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/math/vol2col.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/common/variant.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/common/variant.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/common/variant.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/framework/attribute.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/attribute.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/attribute.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/framework/block_desc.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/block_desc.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/block_desc.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/framework/data_transform.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/data_transform.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/data_transform.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/framework/executor.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/executor.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/executor.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/framework/framework.pb.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/framework.pb.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/framework.pb.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/framework/op_desc.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/op_desc.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/op_desc.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/framework/operator.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/operator.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/operator.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/framework/paddle_mobile_object.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/paddle_mobile_object.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/paddle_mobile_object.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/framework/program.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/program.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/program.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/framework/program_desc.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/program_desc.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/program_desc.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/framework/var_desc.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/var_desc.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/var_desc.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/io.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/io.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/io.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/operators/conv_op.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/conv_op.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/conv_op.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/operators/elementwise_add_op.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/elementwise_add_op.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/elementwise_add_op.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/operators/kernel/arm/conv_kernel.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/arm/conv_kernel.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/arm/conv_kernel.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/operators/kernel/arm/elementwise_add_kernel.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/arm/elementwise_add_kernel.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/arm/elementwise_add_kernel.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/operators/kernel/arm/mul_kernel.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/arm/mul_kernel.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/arm/mul_kernel.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/operators/kernel/fpga/conv_kernel.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/fpga/conv_kernel.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/fpga/conv_kernel.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/operators/mul_op.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/mul_op.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/mul_op.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-static.dir/src/operators/op_param.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/op_param.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/op_param.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/framework/ddim.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/ddim.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/ddim.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/framework/lod_tensor.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/lod_tensor.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/lod_tensor.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/framework/scope.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/scope.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/scope.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/framework/tensor_util.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/tensor_util.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/tensor_util.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/memory/t_malloc.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/memory/t_malloc.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/memory/t_malloc.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/operators/math/im2col.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/math/im2col.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/math/im2col.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/operators/math/math_function.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/math/math_function.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/math/math_function.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/operators/math/vol2col.cc.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/math/vol2col.cc",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/math/vol2col.cc"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/common/variant.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/common/variant.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/common/variant.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/framework/attribute.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/attribute.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/attribute.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/framework/block_desc.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/block_desc.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/block_desc.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/framework/data_transform.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/data_transform.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/data_transform.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/framework/executor.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/executor.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/executor.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/framework/framework.pb.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/framework.pb.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/framework.pb.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/framework/op_desc.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/op_desc.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/op_desc.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/framework/operator.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/operator.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/operator.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/framework/paddle_mobile_object.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/paddle_mobile_object.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/paddle_mobile_object.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/framework/program.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/program.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/program.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/framework/program_desc.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/program_desc.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/program_desc.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/framework/var_desc.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/framework/var_desc.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/framework/var_desc.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/io.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/io.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/io.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/operators/conv_op.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/conv_op.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/conv_op.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/operators/elementwise_add_op.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/elementwise_add_op.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/elementwise_add_op.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/operators/kernel/arm/conv_kernel.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/arm/conv_kernel.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/arm/conv_kernel.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/operators/kernel/arm/elementwise_add_kernel.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/arm/elementwise_add_kernel.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/arm/elementwise_add_kernel.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/operators/kernel/arm/mul_kernel.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/arm/mul_kernel.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/arm/mul_kernel.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/operators/kernel/fpga/conv_kernel.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/fpga/conv_kernel.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/kernel/fpga/conv_kernel.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/operators/mul_op.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/mul_op.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/mul_op.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -Dpaddle_mobile_EXPORTS -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -fPIC -std=c++11 -o CMakeFiles/paddle-mobile.dir/src/operators/op_param.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/src/operators/op_param.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/src/operators/op_param.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release/test",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/test-log.dir/common/test_log.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/test/common/test_log.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/test/common/test_log.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release/test",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/test-conv-op.dir/operators/test_cov_op.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/test/operators/test_cov_op.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/test/operators/test_cov_op.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release/test",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/test-load.dir/framework/test_load.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/test/framework/test_load.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/test/framework/test_load.cpp"
},
{
"directory": "/Users/allonli/Documents/workspace/paddle-mobile/cmake-build-release/test",
"command": "/Library/Developer/CommandLineTools/usr/bin/c++ -DPADDLE_MOBILE_DEBUG=\\\"true\\\" -I/Users/allonli/Documents/workspace/paddle-mobile/src -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/openblas/include -I/Users/allonli/Documents/workspace/paddle-mobile/third-party/protobuf/include -O2 -g -DNDEBUG -std=c++11 -o CMakeFiles/paddle-mobile-test.dir/main.cpp.o -c /Users/allonli/Documents/workspace/paddle-mobile/test/main.cpp",
"file": "/Users/allonli/Documents/workspace/paddle-mobile/test/main.cpp"
}
]
\ No newline at end of file
src/common/log.h
浏览文件 @ dff4a781
...@@ -28,15 +28,15 @@ SOFTWARE. ...@@ -28,15 +28,15 @@ SOFTWARE.
namespace paddle_mobile { namespace paddle_mobile {
enum LogLevel { enum LogLevel {
kNO_LOG, kNO_LOG,
kLOG_ERROR, kLOG_ERROR,
kLOG_WARNING, kLOG_WARNING,
kLOG_INFO, kLOG_INFO,
kLOG_DEBUG, kLOG_DEBUG,
kLOG_DEBUG1, kLOG_DEBUG1,
kLOG_DEBUG2, kLOG_DEBUG2,
kLOG_DEBUG3, kLOG_DEBUG3,
kLOG_DEBUG4 kLOG_DEBUG4
}; };
// log level // log level
...@@ -49,119 +49,117 @@ struct ToLog; ...@@ -49,119 +49,117 @@ struct ToLog;
struct Print; struct Print;
struct Print { struct Print {
friend struct ToLog; friend struct ToLog;
template <typename T> Print &operator<<(T const &value) { template <typename T> Print &operator<<(T const &value) {
buffer_ << value; buffer_ << value;
return *this; return *this;
} }
private: private:
void print(LogLevel level) { void print(LogLevel level) {
buffer_ << std::endl; buffer_ << std::endl;
if (level == kLOG_ERROR) { if (level == kLOG_ERROR) {
std::cerr << buffer_.str(); std::cerr << buffer_.str();
} else { } else {
std::cout << buffer_.str(); std::cout << buffer_.str();
}
} }
std::ostringstream buffer_; }
std::ostringstream buffer_;
}; };
struct ToLog { struct ToLog {
ToLog(LogLevel level = kLOG_DEBUG, const std::string &info = "") ToLog(LogLevel level = kLOG_DEBUG, const std::string &info = "")
: level_(level) { : level_(level) {
unsigned blanks = unsigned blanks =
(unsigned)(level > kLOG_DEBUG ? (level - kLOG_DEBUG) * 4 : 1); (unsigned)(level > kLOG_DEBUG ? (level - kLOG_DEBUG) * 4 : 1);
printer_ << logs[level] << " " << info << ":" printer_ << logs[level] << " " << info << ":" << std::string(blanks, ' ');
<< std::string(blanks, ' '); }
}
template <typename T> ToLog &operator<<(T const &value) {
template <typename T> ToLog &operator<<(T const &value) { printer_ << value;
printer_ << value; return *this;
return *this; }
}
~ToLog() { printer_.print(level_); }
~ToLog() { printer_.print(level_); }
private:
private: LogLevel level_;
LogLevel level_; Print printer_;
Print printer_;
}; };
#define LOG(level) \ #define LOG(level) \
if (level > paddle_mobile::log_level) { \ if (level > paddle_mobile::log_level) { \
} else \ } else \
paddle_mobile::ToLog( \ paddle_mobile::ToLog( \
level, (std::stringstream() \ level, \
<< "[file: " \ (std::stringstream() \
<< (strrchr(__FILE__, '/') ? (strrchr(__FILE__, '/') + 1) \ << "[file: " \
: __FILE__) \ << (strrchr(__FILE__, '/') ? (strrchr(__FILE__, '/') + 1) : __FILE__) \
<< "] [line: " << __LINE__ << "] ") \ << "] [line: " << __LINE__ << "] ") \
.str()) .str())
#define DLOG \ #define DLOG \
if (paddle_mobile::kLOG_DEBUG > paddle_mobile::log_level) { \ if (paddle_mobile::kLOG_DEBUG > paddle_mobile::log_level) { \
} else \ } else \
paddle_mobile::ToLog( \ paddle_mobile::ToLog( \
paddle_mobile::kLOG_DEBUG, \ paddle_mobile::kLOG_DEBUG, \
(std::stringstream() \ (std::stringstream() \
<< "[file: " \ << "[file: " \
<< (strrchr(__FILE__, '/') ? (strrchr(__FILE__, '/') + 1) \ << (strrchr(__FILE__, '/') ? (strrchr(__FILE__, '/') + 1) : __FILE__) \
: __FILE__) \ << "] [line: " << __LINE__ << "] ") \
<< "] [line: " << __LINE__ << "] ") \ .str())
.str())
} // namespace paddle_mobile } // namespace paddle_mobile
#define LOGF(level, format, ...) \ #define LOGF(level, format, ...) \
if (level > paddle_mobile::log_level) { \ if (level > paddle_mobile::log_level) { \
} else \ } else \
printf(format, ##__VA_ARGS__) printf(format, ##__VA_ARGS__)
#define DLOGF(format, ...) \ #define DLOGF(format, ...) \
if (paddle_mobile::kLOG_DEBUG > paddle_mobile::log_level) { \ if (paddle_mobile::kLOG_DEBUG > paddle_mobile::log_level) { \
} else \ } else \
printf(format, ##__VA_ARGS__) printf(format, ##__VA_ARGS__)
#else #else
namespace paddle_mobile { namespace paddle_mobile {
enum LogLevel { enum LogLevel {
kNO_LOG, kNO_LOG,
kLOG_ERROR, kLOG_ERROR,
kLOG_WARNING, kLOG_WARNING,
kLOG_INFO, kLOG_INFO,
kLOG_DEBUG, kLOG_DEBUG,
kLOG_DEBUG1, kLOG_DEBUG1,
kLOG_DEBUG2, kLOG_DEBUG2,
kLOG_DEBUG3, kLOG_DEBUG3,
kLOG_DEBUG4 kLOG_DEBUG4
}; };
struct ToLog; struct ToLog;
struct Print { struct Print {
friend struct ToLog; friend struct ToLog;
template <typename T> Print &operator<<(T const &value) {} template <typename T> Print &operator<<(T const &value) {}
private: private:
}; };
struct ToLog { struct ToLog {
ToLog(LogLevel level) {} ToLog(LogLevel level) {}
template <typename T> ToLog &operator<<(T const &value) { return *this; } template <typename T> ToLog &operator<<(T const &value) { return *this; }
}; };
#define LOG(level) \ #define LOG(level) \
if (true) { \ if (true) { \
} else \ } else \
paddle_mobile::ToLog(level) paddle_mobile::ToLog(level)
#define DLOG \ #define DLOG \
if (true) { \ if (true) { \
} else \ } else \
paddle_mobile::ToLog(paddle_mobile::kLOG_DEBUG) paddle_mobile::ToLog(paddle_mobile::kLOG_DEBUG)
#define LOGF(level, format, ...) #define LOGF(level, format, ...)
......
src/common/types.h
浏览文件 @ dff4a781
...@@ -32,32 +32,32 @@ typedef DeviceType<kGPU_MALI> GPU_MALI; ...@@ -32,32 +32,32 @@ typedef DeviceType<kGPU_MALI> GPU_MALI;
//! data type //! data type
enum DataType { enum DataType {
PM_INVALID = -1, PM_INVALID = -1,
PM_HALF = 0, PM_HALF = 0,
PM_FLOAT = 1, PM_FLOAT = 1,
PM_DOUBLE = 2, PM_DOUBLE = 2,
PM_INT8 = 3, PM_INT8 = 3,
PM_INT16 = 4, PM_INT16 = 4,
PM_INT32 = 5, PM_INT32 = 5,
PM_INT64 = 6, PM_INT64 = 6,
PM_UINT8 = 7, PM_UINT8 = 7,
PM_UINT16 = 8, PM_UINT16 = 8,
PM_UINT32 = 9, PM_UINT32 = 9,
PM_STRING = 10, PM_STRING = 10,
PM_BOOL = 11, PM_BOOL = 11,
PM_SHAPE = 12, PM_SHAPE = 12,
PM_TENSOR = 13 PM_TENSOR = 13
}; };
//! //!
enum PMStatus { enum PMStatus {
PMSuccess = 0xFF, /*!< No errors */ PMSuccess = 0xFF, /*!< No errors */
PMNotInitialized = 0x01, /*!< Data not initialized. */ PMNotInitialized = 0x01, /*!< Data not initialized. */
PMInvalidValue = 0x02, /*!< Incorrect variable value. */ PMInvalidValue = 0x02, /*!< Incorrect variable value. */
PMMemAllocFailed = 0x03, /*!< Memory allocation error. */ PMMemAllocFailed = 0x03, /*!< Memory allocation error. */
PMUnKownError = 0x04, /*!< Unknown error. */ PMUnKownError = 0x04, /*!< Unknown error. */
PMOutOfAuthority = 0x05, /*!< Try to modified data not your own*/ PMOutOfAuthority = 0x05, /*!< Try to modified data not your own*/
PMOutOfMem = 0x06, /*!< OOM error*/ PMOutOfMem = 0x06, /*!< OOM error*/
PMUnImplError = 0x07, /*!< Unimplement error. */ PMUnImplError = 0x07, /*!< Unimplement error. */
PMWrongDevice = 0x08 /*!< un-correct device. */ PMWrongDevice = 0x08 /*!< un-correct device. */
}; };
} // namespace paddle_mobile } // namespace paddle_mobile
src/common/variant.h
浏览文件 @ dff4a781
...@@ -24,74 +24,74 @@ namespace paddle_mobile { ...@@ -24,74 +24,74 @@ namespace paddle_mobile {
template <int ID, typename Type> struct IDToType { typedef Type type_t; }; template <int ID, typename Type> struct IDToType { typedef Type type_t; };
template <typename F, typename... Ts> struct VariantHelper { template <typename F, typename... Ts> struct VariantHelper {
static const size_t size = sizeof(F) > VariantHelper<Ts...>::size static const size_t size = sizeof(F) > VariantHelper<Ts...>::size
? sizeof(F) ? sizeof(F)
: VariantHelper<Ts...>::size; : VariantHelper<Ts...>::size;
inline static void Destroy(size_t id, void *data) { inline static void Destroy(size_t id, void *data) {
if (id == typeid(F).hash_code()) { if (id == typeid(F).hash_code()) {
reinterpret_cast<F *>(data)->~F(); reinterpret_cast<F *>(data)->~F();
} else { } else {
VariantHelper<Ts...>::Destroy(id, data); VariantHelper<Ts...>::Destroy(id, data);
}
} }
}
}; };
template <typename F> struct VariantHelper<F> { template <typename F> struct VariantHelper<F> {
static const size_t size = sizeof(F); static const size_t size = sizeof(F);
inline static void Destroy(size_t id, void *data) { inline static void Destroy(size_t id, void *data) {
if (id == typeid(F).hash_code()) { if (id == typeid(F).hash_code()) {
// reinterpret_cast<F*>(data)->~F(); // reinterpret_cast<F*>(data)->~F();
} else { } else {
// std::cout << "未匹配到 " << std::endl; // std::cout << "未匹配到 " << std::endl;
}
} }
}
}; };
template <size_t size> class RawData { template <size_t size> class RawData {
public: public:
char data[size]; char data[size];
RawData() {} RawData() {}
RawData(const RawData &raw_data) { strcpy(data, raw_data.data); } RawData(const RawData &raw_data) { strcpy(data, raw_data.data); }
// void operator=(const RawData &raw_data){ // void operator=(const RawData &raw_data){
// strcpy(data, raw_data.data); // strcpy(data, raw_data.data);
// } // }
}; };
template <typename... Ts> struct Variant { template <typename... Ts> struct Variant {
Variant(const Variant &variant) { Variant(const Variant &variant) {
// std::cout << " 赋值构造函数 " << std::endl; // std::cout << " 赋值构造函数 " << std::endl;
type_id = variant.type_id; type_id = variant.type_id;
data = variant.data; data = variant.data;
} }
Variant() : type_id(invalid_type()) {} Variant() : type_id(invalid_type()) {}
~Variant() { ~Variant() {
// helper::Destroy(type_id, &data); // helper::Destroy(type_id, &data);
} }
template <typename T, typename... Args> void Set(Args &&... args) { template <typename T, typename... Args> void Set(Args &&... args) {
helper::Destroy(type_id, &data); helper::Destroy(type_id, &data);
new (&data) T(std::forward<Args>(args)...); new (&data) T(std::forward<Args>(args)...);
type_id = typeid(T).hash_code(); type_id = typeid(T).hash_code();
} }
template <typename T> T &Get() const { template <typename T> T &Get() const {
if (type_id == typeid(T).hash_code()) { if (type_id == typeid(T).hash_code()) {
return *const_cast<T *>(reinterpret_cast<const T *>(&data)); return *const_cast<T *>(reinterpret_cast<const T *>(&data));
} else { } else {
// std::cout << " bad cast in variant " << std::endl; // std::cout << " bad cast in variant " << std::endl;
throw std::bad_cast(); throw std::bad_cast();
}
} }
}
size_t TypeId() const { return type_id; } size_t TypeId() const { return type_id; }
private: private:
static inline size_t invalid_type() { return typeid(void).hash_code(); } static inline size_t invalid_type() { return typeid(void).hash_code(); }
typedef VariantHelper<Ts...> helper; typedef VariantHelper<Ts...> helper;
size_t type_id; size_t type_id;
RawData<helper::size> data; RawData<helper::size> data;
}; };
template <typename T> struct Vistor { typedef T type_t; }; template <typename T> struct Vistor { typedef T type_t; };
......
src/framework/attribute.h
浏览文件 @ dff4a781
...@@ -27,102 +27,102 @@ namespace framework { ...@@ -27,102 +27,102 @@ namespace framework {
class BlockDesc; class BlockDesc;
class Attribute { class Attribute {
public: public:
static Attribute GetAttrValue(const proto::OpDesc::Attr &attr_desc) { static Attribute GetAttrValue(const proto::OpDesc::Attr &attr_desc) {
// std::cout << "begin get attr value" << std::endl; // std::cout << "begin get attr value" << std::endl;
Attribute attr; Attribute attr;
switch (attr_desc.type()) { switch (attr_desc.type()) {
case proto::AttrType::BOOLEAN: { case proto::AttrType::BOOLEAN: {
attr.Set<bool>(attr_desc.b()); attr.Set<bool>(attr_desc.b());
break; break;
}
case proto::AttrType::INT: {
attr.Set<int>(attr_desc.i());
break;
}
case proto::AttrType::FLOAT: {
attr.Set<float>(attr_desc.f());
break;
}
case proto::AttrType::STRING: {
attr.Set<std::string>(attr_desc.s());
break;
}
case proto::AttrType::BOOLEANS: {
std::vector<bool> val(attr_desc.bools_size());
for (int i = 0; i < attr_desc.bools_size(); ++i) {
val[i] = attr_desc.bools(i);
}
attr.Set<std::vector<bool>>(val);
break;
}
case proto::AttrType::INTS: {
std::vector<int> val(attr_desc.ints_size());
for (int i = 0; i < attr_desc.ints_size(); ++i) {
val[i] = attr_desc.ints(i);
}
attr.Set<std::vector<int>>(val);
break;
}
case proto::AttrType::FLOATS: {
std::vector<float> val(attr_desc.floats_size());
for (int i = 0; i < attr_desc.floats_size(); ++i) {
val[i] = attr_desc.floats(i);
}
attr.Set<std::vector<float>>(val);
break;
}
case proto::AttrType::STRINGS: {
std::vector<std::string> val(attr_desc.strings_size());
for (int i = 0; i < attr_desc.strings_size(); ++i) {
val[i] = attr_desc.strings(i);
}
attr.Set<std::vector<std::string>>(val);
break;
}
case proto::AttrType::LONG: {
attr.Set<int64_t>(attr_desc.l());
break;
}
default:
// std::cout << " not support " << std::endl;
break;
}
// std::cout << "end get attr value" << std::endl;
return attr;
} }
case proto::AttrType::INT: {
Attribute() {} attr.Set<int>(attr_desc.i());
template <typename T, typename... Args> Attribute &Set(Args &&... args) { break;
variant_.Set<T>(args...); }
return *this; case proto::AttrType::FLOAT: {
attr.Set<float>(attr_desc.f());
break;
}
case proto::AttrType::STRING: {
attr.Set<std::string>(attr_desc.s());
break;
}
case proto::AttrType::BOOLEANS: {
std::vector<bool> val(attr_desc.bools_size());
for (int i = 0; i < attr_desc.bools_size(); ++i) {
val[i] = attr_desc.bools(i);
}
attr.Set<std::vector<bool>>(val);
break;
}
case proto::AttrType::INTS: {
std::vector<int> val(attr_desc.ints_size());
for (int i = 0; i < attr_desc.ints_size(); ++i) {
val[i] = attr_desc.ints(i);
}
attr.Set<std::vector<int>>(val);
break;
}
case proto::AttrType::FLOATS: {
std::vector<float> val(attr_desc.floats_size());
for (int i = 0; i < attr_desc.floats_size(); ++i) {
val[i] = attr_desc.floats(i);
}
attr.Set<std::vector<float>>(val);
break;
} }
case proto::AttrType::STRINGS: {
std::vector<std::string> val(attr_desc.strings_size());
for (int i = 0; i < attr_desc.strings_size(); ++i) {
val[i] = attr_desc.strings(i);
}
attr.Set<std::vector<std::string>>(val);
break;
}
case proto::AttrType::LONG: {
attr.Set<int64_t>(attr_desc.l());
break;
}
default:
// std::cout << " not support " << std::endl;
break;
}
// std::cout << "end get attr value" << std::endl;
return attr;
}
template <typename T> T &Get() const { return variant_.Get<T>(); } Attribute() {}
template <typename T, typename... Args> Attribute &Set(Args &&... args) {
variant_.Set<T>(args...);
return *this;
}
private: template <typename T> T &Get() const { return variant_.Get<T>(); }
Variant<int, float, std::string, std::vector<int>, std::vector<float>,
std::vector<std::string>, bool, std::vector<bool>, BlockDesc *, private:
int64_t> Variant<int, float, std::string, std::vector<int>, std::vector<float>,
variant_; std::vector<std::string>, bool, std::vector<bool>, BlockDesc *,
int64_t>
variant_;
}; };
using AttributeMap = std::unordered_map<std::string, Attribute>; using AttributeMap = std::unordered_map<std::string, Attribute>;
class AttrReader { class AttrReader {
public: public:
explicit AttrReader(const AttributeMap &attrs) : attrs_(attrs) {} explicit AttrReader(const AttributeMap &attrs) : attrs_(attrs) {}
template <typename T> inline T Get(const std::string &name) const { template <typename T> inline T Get(const std::string &name) const {
// PADDLE_ENFORCE(attrs_.count(name) != 0, "%s should // PADDLE_ENFORCE(attrs_.count(name) != 0, "%s should
// be in // be in
// AttributeMap", // AttributeMap",
// name); // name);
return ((Attribute)attrs_.at(name)).Get<T>(); return ((Attribute)attrs_.at(name)).Get<T>();
} }
private: private:
const AttributeMap &attrs_; const AttributeMap &attrs_;
}; };
} // namespace framework } // namespace framework
......
src/framework/block_desc.cpp
浏览文件 @ dff4a781
...@@ -22,28 +22,28 @@ namespace paddle_mobile { ...@@ -22,28 +22,28 @@ namespace paddle_mobile {
namespace framework { namespace framework {
std::vector<std::shared_ptr<VarDesc>> BlockDesc::Vars() const { std::vector<std::shared_ptr<VarDesc>> BlockDesc::Vars() const {
std::vector<std::shared_ptr<VarDesc>> res; std::vector<std::shared_ptr<VarDesc>> res;
for (const auto &p : vars_) { for (const auto &p : vars_) {
res.push_back(p.second); res.push_back(p.second);
} }
return res; return res;
} }
std::vector<std::shared_ptr<OpDesc>> BlockDesc::Ops() const { std::vector<std::shared_ptr<OpDesc>> BlockDesc::Ops() const {
std::vector<std::shared_ptr<OpDesc>> res; std::vector<std::shared_ptr<OpDesc>> res;
for (const auto &op : ops_) { for (const auto &op : ops_) {
res.push_back(op); res.push_back(op);
} }
return res; return res;
} }
BlockDesc::BlockDesc(const proto::BlockDesc &desc) : desc_(desc) { BlockDesc::BlockDesc(const proto::BlockDesc &desc) : desc_(desc) {
for (const proto::VarDesc &var_desc : desc_.vars()) { for (const proto::VarDesc &var_desc : desc_.vars()) {
vars_[var_desc.name()].reset(new VarDesc(var_desc)); vars_[var_desc.name()].reset(new VarDesc(var_desc));
} }
for (const proto::OpDesc &op_desc : desc_.ops()) { for (const proto::OpDesc &op_desc : desc_.ops()) {
ops_.emplace_back(new framework::OpDesc(op_desc)); ops_.emplace_back(new framework::OpDesc(op_desc));
} }
} }
} // namespace framework } // namespace framework
......
src/framework/block_desc.h
浏览文件 @ dff4a781
...@@ -27,29 +27,28 @@ namespace paddle_mobile { ...@@ -27,29 +27,28 @@ namespace paddle_mobile {
namespace framework { namespace framework {
class BlockDesc : PaddleMobileObject { class BlockDesc : PaddleMobileObject {
public: public:
BlockDesc(const proto::BlockDesc &desc); BlockDesc(const proto::BlockDesc &desc);
const int &ID() const { return desc_.idx(); } const int &ID() const { return desc_.idx(); }
const int &Parent() const { return desc_.parent_idx(); } const int &Parent() const { return desc_.parent_idx(); }
bool operator==(const paddle_mobile::framework::BlockDesc &in_block) const { bool operator==(const paddle_mobile::framework::BlockDesc &in_block) const {
return this->ID() == in_block.ID() && return this->ID() == in_block.ID() && this->Parent() == in_block.Parent();
this->Parent() == in_block.Parent(); }
}
bool operator<(const paddle_mobile::framework::BlockDesc &in_block) const { bool operator<(const paddle_mobile::framework::BlockDesc &in_block) const {
return this->ID() < in_block.ID() && this->Parent() < in_block.Parent(); return this->ID() < in_block.ID() && this->Parent() < in_block.Parent();
} }
std::vector<std::shared_ptr<VarDesc>> Vars() const; std::vector<std::shared_ptr<VarDesc>> Vars() const;
std::vector<std::shared_ptr<OpDesc>> Ops() const; std::vector<std::shared_ptr<OpDesc>> Ops() const;
private: private:
proto::BlockDesc desc_; proto::BlockDesc desc_;
std::vector<std::shared_ptr<OpDesc>> ops_; std::vector<std::shared_ptr<OpDesc>> ops_;
std::unordered_map<std::string, std::shared_ptr<VarDesc>> vars_; std::unordered_map<std::string, std::shared_ptr<VarDesc>> vars_;
}; };
} // namespace framework } // namespace framework
...@@ -58,13 +57,13 @@ class BlockDesc : PaddleMobileObject { ...@@ -58,13 +57,13 @@ class BlockDesc : PaddleMobileObject {
namespace std { namespace std {
template <> struct hash<paddle_mobile::framework::BlockDesc> { template <> struct hash<paddle_mobile::framework::BlockDesc> {
typedef paddle_mobile::framework::BlockDesc argument_type; typedef paddle_mobile::framework::BlockDesc argument_type;
typedef std::size_t result_type; typedef std::size_t result_type;
result_type operator()(argument_type const &s) const noexcept { result_type operator()(argument_type const &s) const noexcept {
result_type const h1(std::hash<int>{}(s.ID())); result_type const h1(std::hash<int>{}(s.ID()));
result_type const h2(std::hash<int>{}(s.ID())); result_type const h2(std::hash<int>{}(s.ID()));
return h1 ^ (h2 << 1); return h1 ^ (h2 << 1);
} }
}; };
} // namespace std } // namespace std
src/framework/data_layout.h
浏览文件 @ dff4a781
...@@ -22,45 +22,45 @@ namespace paddle_mobile { ...@@ -22,45 +22,45 @@ namespace paddle_mobile {
namespace framework { namespace framework {
enum class DataLayout { enum class DataLayout {
kNHWC = 0, kNHWC = 0,
kNCHW = 1, kNCHW = 1,
kAnyLayout = 2, kAnyLayout = 2,
}; };
inline DataLayout StringToDataLayout(const std::string &str) { inline DataLayout StringToDataLayout(const std::string &str) {
std::string s(str); std::string s(str);
for (size_t i = 0; i < s.size(); ++i) { for (size_t i = 0; i < s.size(); ++i) {
s[i] = toupper(s[i]); s[i] = toupper(s[i]);
} }
if (s == "NHWC") { if (s == "NHWC") {
return DataLayout::kNHWC; return DataLayout::kNHWC;
} else if (s == "NCHW") { } else if (s == "NCHW") {
return DataLayout::kNCHW; return DataLayout::kNCHW;
} else if (s == "ANYLAYOUT") { } else if (s == "ANYLAYOUT") {
return DataLayout::kAnyLayout; return DataLayout::kAnyLayout;
} else { } else {
// std::cout << "Unknown storage order string: %s", s; // std::cout << "Unknown storage order string: %s", s;
} }
} }
inline std::string DataLayoutToString(const DataLayout &data_layout) { inline std::string DataLayoutToString(const DataLayout &data_layout) {
switch (data_layout) { switch (data_layout) {
case DataLayout::kNHWC: case DataLayout::kNHWC:
return "NHWC"; return "NHWC";
case DataLayout::kNCHW: case DataLayout::kNCHW:
return "NCHW"; return "NCHW";
case DataLayout::kAnyLayout: case DataLayout::kAnyLayout:
return "ANY_LAYOUT"; return "ANY_LAYOUT";
default: default:
break; break;
// std::cout << "unknown DataLayou %d", data_layout; // std::cout << "unknown DataLayou %d", data_layout;
} }
} }
inline std::ostream &operator<<(std::ostream &out, const DataLayout &l) { inline std::ostream &operator<<(std::ostream &out, const DataLayout &l) {
out << DataLayoutToString(l); out << DataLayoutToString(l);
return out; return out;
} }
} // namespace framework } // namespace framework
......
src/framework/data_transform.cpp
浏览文件 @ dff4a781
...@@ -24,68 +24,68 @@ namespace paddle_mobile { ...@@ -24,68 +24,68 @@ namespace paddle_mobile {
namespace framework { namespace framework {
static void PassTensorData(Tensor *from, Tensor *to) { static void PassTensorData(Tensor *from, Tensor *to) {
to->ShareDataWith(*from); to->ShareDataWith(*from);
*from = Tensor(); *from = Tensor();
} }
void DataTransform(const OpKernelType &expected_kernel_type, void DataTransform(const OpKernelType &expected_kernel_type,
const OpKernelType &kernel_type_for_var, const OpKernelType &kernel_type_for_var,
const Tensor &input_tensor, Tensor *output_tensor) { const Tensor &input_tensor, Tensor *output_tensor) {
bool transformed = false; bool transformed = false;
Tensor in; Tensor in;
in.ShareDataWith(input_tensor); in.ShareDataWith(input_tensor);
Tensor out; Tensor out;
// // do layout transform // // do layout transform
// if (NeedTransformLayout(expected_kernel_type.data_layout_, // if (NeedTransformLayout(expected_kernel_type.data_layout_,
// kernel_type_for_var.data_layout_)) { // kernel_type_for_var.data_layout_)) {
// TransDataLayout(kernel_type_for_var, expected_kernel_type, in, // TransDataLayout(kernel_type_for_var, expected_kernel_type, in,
// &out); // &out);
// transformed = true; // transformed = true;
// PassTensorData(&out, &in); // PassTensorData(&out, &in);
// } // }
// //
// // do data type transform // // do data type transform
// if (expected_kernel_type.data_type_ != // if (expected_kernel_type.data_type_ !=
// kernel_type_for_var.data_type_) { // kernel_type_for_var.data_type_) {
// TransDataType(kernel_type_for_var, expected_kernel_type, in, // TransDataType(kernel_type_for_var, expected_kernel_type, in,
// &out); // &out);
// transformed = true; // transformed = true;
// PassTensorData(&out, &in); // PassTensorData(&out, &in);
// } // }
// //
// // do device transform // // do device transform
// if (!platform::is_same_place(kernel_type_for_var.place_, // if (!platform::is_same_place(kernel_type_for_var.place_,
// expected_kernel_type.place_)) { // expected_kernel_type.place_)) {
// TransDataDevice(in, expected_kernel_type.place_, &out); // TransDataDevice(in, expected_kernel_type.place_, &out);
// transformed = true; // transformed = true;
// PassTensorData(&out, &in); // PassTensorData(&out, &in);
// } // }
// //
// PADDLE_ENFORCE(transformed, "No transform is applied, please // PADDLE_ENFORCE(transformed, "No transform is applied, please
// check!"); // check!");
// get output data // get output data
output_tensor->ShareDataWith(in); output_tensor->ShareDataWith(in);
} }
void CopyVariableWithTensor(const Variable &in_var, const Tensor &tensor, void CopyVariableWithTensor(const Variable &in_var, const Tensor &tensor,
Variable &out_var) { Variable &out_var) {
// if (in_var.IsType<LoDTensor>()) { // if (in_var.IsType<LoDTensor>()) {
// auto& in_lod_tensor = in_var.Get<LoDTensor>(); // auto& in_lod_tensor = in_var.Get<LoDTensor>();
// auto* tran_lod_tensor = out_var.GetMutable<LoDTensor>(); // auto* tran_lod_tensor = out_var.GetMutable<LoDTensor>();
// tran_lod_tensor->set_lod(in_lod_tensor.lod()); // tran_lod_tensor->set_lod(in_lod_tensor.lod());
// tran_lod_tensor->set_layout(in_lod_tensor.layout()); // tran_lod_tensor->set_layout(in_lod_tensor.layout());
// tran_lod_tensor->ShareDataWith(tensor); // tran_lod_tensor->ShareDataWith(tensor);
// } else if (in_var.IsType<SelectedRows>()) { // } else if (in_var.IsType<SelectedRows>()) {
// auto& in_selected_rows = in_var.Get<SelectedRows>(); // auto& in_selected_rows = in_var.Get<SelectedRows>();
// auto* trans_selected_rows = // auto* trans_selected_rows =
// out_var.GetMutable<SelectedRows>(); // out_var.GetMutable<SelectedRows>();
// trans_selected_rows->set_height(in_selected_rows.height()); // trans_selected_rows->set_height(in_selected_rows.height());
// trans_selected_rows->set_rows(in_selected_rows.rows()); // trans_selected_rows->set_rows(in_selected_rows.rows());
// trans_selected_rows->mutable_value()->ShareDataWith(tensor); // trans_selected_rows->mutable_value()->ShareDataWith(tensor);
// } else { // } else {
// PADDLE_THROW("unknown var type"); // PADDLE_THROW("unknown var type");
// } // }
} }
} // namespace framework } // namespace framework
......
src/framework/ddim.cc
浏览文件 @ dff4a781
...@@ -20,158 +20,156 @@ namespace framework { ...@@ -20,158 +20,156 @@ namespace framework {
/// @cond HIDDEN /// @cond HIDDEN
template <int i> Dim<i> make_dim(const int64_t *d) { template <int i> Dim<i> make_dim(const int64_t *d) {
return Dim<i>(*d, make_dim<i - 1>(d + 1)); return Dim<i>(*d, make_dim<i - 1>(d + 1));
} }
template <> Dim<0> make_dim<0>(const int64_t *d) { return Dim<0>(*d); } template <> Dim<0> make_dim<0>(const int64_t *d) { return Dim<0>(*d); }
void make_ddim(DDim &ddim, const int64_t *dims, int n) { void make_ddim(DDim &ddim, const int64_t *dims, int n) {
switch (n) { switch (n) {
case 0: case 0:
ddim = make_dim<0>(dims); ddim = make_dim<0>(dims);
break; break;
case 1: case 1:
ddim = make_dim<1>(dims); ddim = make_dim<1>(dims);
break; break;
case 2: case 2:
ddim = make_dim<2>(dims); ddim = make_dim<2>(dims);
break; break;
case 3: case 3:
ddim = make_dim<3>(dims); ddim = make_dim<3>(dims);
break; break;
case 4: case 4:
ddim = make_dim<4>(dims); ddim = make_dim<4>(dims);
break; break;
case 5: case 5:
ddim = make_dim<5>(dims); ddim = make_dim<5>(dims);
break; break;
case 6: case 6:
ddim = make_dim<6>(dims); ddim = make_dim<6>(dims);
break; break;
case 7: case 7:
ddim = make_dim<7>(dims); ddim = make_dim<7>(dims);
break; break;
case 8: case 8:
ddim = make_dim<8>(dims); ddim = make_dim<8>(dims);
break; break;
case 9: case 9:
ddim = make_dim<9>(dims); ddim = make_dim<9>(dims);
break; break;
default: default:
// std::cout << "Dynamic dimensions must have between [1, // std::cout << "Dynamic dimensions must have between [1,
// 9] // 9]
// dimensions."; // dimensions.";
break; break;
} }
} }
/// @endcond /// @endcond
DDim make_ddim(std::initializer_list<int64_t> dims) { DDim make_ddim(std::initializer_list<int64_t> dims) {
DDim result(make_dim(0)); DDim result(make_dim(0));
make_ddim(result, dims.begin(), dims.size()); make_ddim(result, dims.begin(), dims.size());
return result; return result;
} }
DDim make_ddim(const std::vector<int64_t> &dims) { DDim make_ddim(const std::vector<int64_t> &dims) {
DDim result(make_dim(0)); DDim result(make_dim(0));
make_ddim(result, &dims[0], dims.size()); make_ddim(result, &dims[0], dims.size());
return result; return result;
} }
DDim make_ddim(const std::vector<int> &dims) { DDim make_ddim(const std::vector<int> &dims) {
std::vector<int64_t> res(dims.size()); std::vector<int64_t> res(dims.size());
std::transform(dims.begin(), dims.end(), res.begin(), std::transform(dims.begin(), dims.end(), res.begin(),
[](int d) { return static_cast<int64_t>(d); }); [](int d) { return static_cast<int64_t>(d); });
return make_ddim(res); return make_ddim(res);
} }
/// @cond HIDDEN /// @cond HIDDEN
// XXX For some reason, putting this in an anonymous namespace causes // XXX For some reason, putting this in an anonymous namespace causes
// errors // errors
struct DynamicMutableIndexer : Vistor<int64_t &> { struct DynamicMutableIndexer : Vistor<int64_t &> {
public: public:
explicit DynamicMutableIndexer(int idx) : idx_(idx) {} explicit DynamicMutableIndexer(int idx) : idx_(idx) {}
template <int D> int64_t &operator()(Dim<D> &dim) const { template <int D> int64_t &operator()(Dim<D> &dim) const { return dim[idx_]; }
return dim[idx_];
}
private: private:
int idx_; int idx_;
}; };
struct DynamicConstIndexer : public Vistor<int64_t> { struct DynamicConstIndexer : public Vistor<int64_t> {
public: public:
explicit DynamicConstIndexer(int idx) : idx_(idx) {} explicit DynamicConstIndexer(int idx) : idx_(idx) {}
template <int D> int64_t operator()(const Dim<D> &dim) const { template <int D> int64_t operator()(const Dim<D> &dim) const {
return dim[idx_]; return dim[idx_];
} }
private: private:
int idx_; int idx_;
}; };
/// @endcond /// @endcond
int64_t &DDim::operator[](int idx) { int64_t &DDim::operator[](int idx) {
return DDim::ApplyVistor(DynamicMutableIndexer(idx), *this); return DDim::ApplyVistor(DynamicMutableIndexer(idx), *this);
} }
int64_t DDim::operator[](int idx) const { int64_t DDim::operator[](int idx) const {
return DDim::ApplyVistor(DynamicConstIndexer(idx), *this); return DDim::ApplyVistor(DynamicConstIndexer(idx), *this);
} }
int DDim::size() const { return arity(*this); } int DDim::size() const { return arity(*this); }
bool DDim::operator==(DDim d) const { bool DDim::operator==(DDim d) const {
// if (var.which() != d.getVar().which()) { // if (var.which() != d.getVar().which()) {
// return false; // return false;
// } else { // } else {
std::vector<int64_t> v1 = vectorize(*this); std::vector<int64_t> v1 = vectorize(*this);
std::vector<int64_t> v2 = vectorize(d); std::vector<int64_t> v2 = vectorize(d);
for (unsigned int i = 0; i < v1.size(); i++) { for (unsigned int i = 0; i < v1.size(); i++) {
if (v1[i] != v2[i]) { if (v1[i] != v2[i]) {
return false; return false;
}
} }
}
return true; return true;
// } // }
} }
bool DDim::operator!=(DDim d) const { return !(*this == d); } bool DDim::operator!=(DDim d) const { return !(*this == d); }
DDim DDim::operator+(DDim d) const { DDim DDim::operator+(DDim d) const {
std::vector<int64_t> v1 = vectorize(*this); std::vector<int64_t> v1 = vectorize(*this);
std::vector<int64_t> v2 = vectorize(d); std::vector<int64_t> v2 = vectorize(d);
std::vector<int64_t> v3; std::vector<int64_t> v3;
assert(v1.size() == v2.size()); assert(v1.size() == v2.size());
for (unsigned int i = 0; i < v1.size(); i++) { for (unsigned int i = 0; i < v1.size(); i++) {
v3.push_back(v1[i] + v2[i]); v3.push_back(v1[i] + v2[i]);
} }
return make_ddim(v3); return make_ddim(v3);
} }
DDim DDim::operator*(DDim d) const { DDim DDim::operator*(DDim d) const {
std::vector<int64_t> v1 = vectorize(*this); std::vector<int64_t> v1 = vectorize(*this);
std::vector<int64_t> v2 = vectorize(d); std::vector<int64_t> v2 = vectorize(d);
std::vector<int64_t> v3; std::vector<int64_t> v3;
assert(v1.size() == v2.size()); assert(v1.size() == v2.size());
for (unsigned int i = 0; i < v1.size(); i++) { for (unsigned int i = 0; i < v1.size(); i++) {
v3.push_back(v1[i] * v2[i]); v3.push_back(v1[i] * v2[i]);
} }
return make_ddim(v3); return make_ddim(v3);
} }
int64_t get(const DDim &ddim, int idx) { return ddim[idx]; } int64_t get(const DDim &ddim, int idx) { return ddim[idx]; }
...@@ -180,152 +178,152 @@ void set(DDim &ddim, int idx, int value) { ddim[idx] = value; } ...@@ -180,152 +178,152 @@ void set(DDim &ddim, int idx, int value) { ddim[idx] = value; }
/// @cond HIDDEN /// @cond HIDDEN
struct VectorizeVisitor : Vistor<void> { struct VectorizeVisitor : Vistor<void> {
std::vector<int64_t> &vector; std::vector<int64_t> &vector;
explicit VectorizeVisitor(std::vector<int64_t> &v) : vector(v) {} explicit VectorizeVisitor(std::vector<int64_t> &v) : vector(v) {}
template <typename T> void operator()(const T &t) { template <typename T> void operator()(const T &t) {
vector.push_back(t.head); vector.push_back(t.head);
this->operator()(t.tail); this->operator()(t.tail);
} }
void operator()(const Dim<0> &t) {} void operator()(const Dim<0> &t) {}
}; };
/// @endcond /// @endcond
std::vector<int64_t> vectorize(const DDim &ddim) { std::vector<int64_t> vectorize(const DDim &ddim) {
std::vector<int64_t> result; std::vector<int64_t> result;
VectorizeVisitor visitor(result); VectorizeVisitor visitor(result);
DDim::ApplyVistor(visitor, ddim); DDim::ApplyVistor(visitor, ddim);
return result; return result;
} }
// NOTE: framework::vectorize converts to type int64_t // NOTE: framework::vectorize converts to type int64_t
// which does not fit cudnn inputs. // which does not fit cudnn inputs.
std::vector<int> vectorize2int(const DDim &ddim) { std::vector<int> vectorize2int(const DDim &ddim) {
std::vector<int64_t> temp = vectorize(ddim); std::vector<int64_t> temp = vectorize(ddim);
std::vector<int> result(temp.begin(), temp.end()); std::vector<int> result(temp.begin(), temp.end());
return result; return result;
} }
struct ProductVisitor : Vistor<int64_t> { struct ProductVisitor : Vistor<int64_t> {
template <int D> int64_t operator()(const Dim<D> &dim) { template <int D> int64_t operator()(const Dim<D> &dim) {
return product(dim); return product(dim);
} }
}; };
int64_t product(const DDim &ddim) { int64_t product(const DDim &ddim) {
ProductVisitor visitor; ProductVisitor visitor;
return DDim::ApplyVistor(visitor, ddim); return DDim::ApplyVistor(visitor, ddim);
} }
struct SliceVectorizeVisitor : Vistor<void> { struct SliceVectorizeVisitor : Vistor<void> {
std::vector<int64_t> &vector; std::vector<int64_t> &vector;
int begin; int begin;
int end; int end;
SliceVectorizeVisitor(std::vector<int64_t> &v, int b, int e) SliceVectorizeVisitor(std::vector<int64_t> &v, int b, int e)
: vector(v), begin(b), end(e) { : vector(v), begin(b), end(e) {
// PADDLE_ENFORCE(begin < end, // PADDLE_ENFORCE(begin < end,
// "Begin index must be less than end index in // "Begin index must be less than end index in
// ddim // ddim
// slice."); // slice.");
// PADDLE_ENFORCE(begin >= 0, // PADDLE_ENFORCE(begin >= 0,
// "Begin index can't be less than zero in // "Begin index can't be less than zero in
// ddim slice."); // ddim slice.");
}
template <int S> void operator()(const Dim<S> &dim) {
if (begin == 0) {
vector.push_back(dim.head);
} else {
--begin;
} }
--end;
template <int S> void operator()(const Dim<S> &dim) { if (end > 0) {
if (begin == 0) { this->operator()(dim.tail);
vector.push_back(dim.head);
} else {
--begin;
}
--end;
if (end > 0) {
this->operator()(dim.tail);
}
} }
}
void operator()(const Dim<0> &dim) { void operator()(const Dim<0> &dim) {
// PADDLE_ENFORCE(end == 0, "End index in ddim slice is out // PADDLE_ENFORCE(end == 0, "End index in ddim slice is out
// of bound."); // of bound.");
} }
}; };
DDim slice_ddim(const DDim &ddim, int begin, int end) { DDim slice_ddim(const DDim &ddim, int begin, int end) {
std::vector<int64_t> vec; std::vector<int64_t> vec;
vec.reserve(end - begin); vec.reserve(end - begin);
SliceVectorizeVisitor visitor(vec, begin, end); SliceVectorizeVisitor visitor(vec, begin, end);
// boost::apply_visitor(visitor, dim); // boost::apply_visitor(visitor, dim);
DDim::ApplyVistor(visitor, ddim); DDim::ApplyVistor(visitor, ddim);
// visitor(ddim.var.Get<Dim<4>>()); // visitor(ddim.var.Get<Dim<4>>());
return make_ddim(vec); return make_ddim(vec);
} }
/// \cond HIDDEN /// \cond HIDDEN
struct ArityVisitor : Vistor<int> { struct ArityVisitor : Vistor<int> {
template <int D> int operator()(Dim<D>) const { return D; } template <int D> int operator()(Dim<D>) const { return D; }
}; };
/// \endcond /// \endcond
int arity(const DDim &d) { int arity(const DDim &d) {
ArityVisitor arityVisitor = ArityVisitor(); ArityVisitor arityVisitor = ArityVisitor();
return DDim::ApplyVistor(arityVisitor, d); return DDim::ApplyVistor(arityVisitor, d);
// return arityVisitor(d.var.Get<Dim<4>>()); // return arityVisitor(d.var.Get<Dim<4>>());
// return boost::apply_visitor(ArityVisitor(), d); } // return boost::apply_visitor(ArityVisitor(), d); }
} }
/// \cond HIDDEN /// \cond HIDDEN
/// \endcond /// \endcond
struct OSVistor : Vistor<std::ostream &> { struct OSVistor : Vistor<std::ostream &> {
OSVistor(std::ostream &os) : os_(os) {} OSVistor(std::ostream &os) : os_(os) {}
template <int D> std::ostream &operator()(Dim<D> dim) const { template <int D> std::ostream &operator()(Dim<D> dim) const {
return os_ << dim; return os_ << dim;
} }
private: private:
std::ostream &os_; std::ostream &os_;
}; };
std::ostream &operator<<(std::ostream &os, const DDim &ddim) { std::ostream &operator<<(std::ostream &os, const DDim &ddim) {
auto vistor = OSVistor(os); auto vistor = OSVistor(os);
DDim::ApplyVistor(vistor, ddim); DDim::ApplyVistor(vistor, ddim);
return os; return os;
} }
DDim::DDim(std::initializer_list<int64_t> init_list) { DDim::DDim(std::initializer_list<int64_t> init_list) {
*this = make_ddim(init_list); *this = make_ddim(init_list);
} }
DDim flatten_to_2d(const DDim &src, int num_col_dims) { DDim flatten_to_2d(const DDim &src, int num_col_dims) {
int rank = src.size(); int rank = src.size();
return make_ddim({product(slice_ddim(src, 0, num_col_dims)), return make_ddim({product(slice_ddim(src, 0, num_col_dims)),
product(slice_ddim(src, num_col_dims, rank))}); product(slice_ddim(src, num_col_dims, rank))});
} }
DDim flatten_to_1d(const DDim &src) { return make_ddim({product(src)}); } DDim flatten_to_1d(const DDim &src) { return make_ddim({product(src)}); }
DDim stride(const DDim &ddim) { DDim stride(const DDim &ddim) {
std::vector<int64_t> strides(ddim.size()); std::vector<int64_t> strides(ddim.size());
strides[ddim.size() - 1] = 1; strides[ddim.size() - 1] = 1;
for (int i = ddim.size() - 2; i >= 0; --i) { for (int i = ddim.size() - 2; i >= 0; --i) {
strides[i] = strides[i + 1] * ddim[i + 1]; strides[i] = strides[i + 1] * ddim[i + 1];
} }
return framework::make_ddim(strides); return framework::make_ddim(strides);
} }
DDim stride_numel(const framework::DDim &ddim) { DDim stride_numel(const framework::DDim &ddim) {
std::vector<int64_t> strides(ddim.size()); std::vector<int64_t> strides(ddim.size());
strides[ddim.size() - 1] = ddim[ddim.size() - 1]; strides[ddim.size() - 1] = ddim[ddim.size() - 1];
for (int i = ddim.size() - 2; i >= 0; --i) { for (int i = ddim.size() - 2; i >= 0; --i) {
strides[i] = strides[i + 1] * ddim[i]; strides[i] = strides[i + 1] * ddim[i];
} }
return framework::make_ddim(strides); return framework::make_ddim(strides);
} }
} // namespace framework } // namespace framework
......
src/framework/ddim.h
浏览文件 @ dff4a781
...@@ -30,77 +30,77 @@ namespace framework { ...@@ -30,77 +30,77 @@ namespace framework {
* The number of dimensions must be between [1, 9]. * The number of dimensions must be between [1, 9].
*/ */
struct DDim { struct DDim {
typedef Variant<Dim<0>, Dim<1>, Dim<2>, Dim<3>, Dim<4>, Dim<5>, Dim<6>, typedef Variant<Dim<0>, Dim<1>, Dim<2>, Dim<3>, Dim<4>, Dim<5>, Dim<6>,
Dim<7>, Dim<8>, Dim<9>> Dim<7>, Dim<8>, Dim<9>>
DDimVar; DDimVar;
DDimVar var; DDimVar var;
template <typename Vistor> template <typename Vistor>
static typename Vistor::type_t ApplyVistor(Vistor vistor, const DDim &d) { static typename Vistor::type_t ApplyVistor(Vistor vistor, const DDim &d) {
if (d.var.TypeId() == typeid(Dim<0>).hash_code()) { if (d.var.TypeId() == typeid(Dim<0>).hash_code()) {
return vistor(d.var.Get<Dim<0>>()); return vistor(d.var.Get<Dim<0>>());
} else if (d.var.TypeId() == typeid(Dim<1>).hash_code()) { } else if (d.var.TypeId() == typeid(Dim<1>).hash_code()) {
return vistor(d.var.Get<Dim<1>>()); return vistor(d.var.Get<Dim<1>>());
} else if (d.var.TypeId() == typeid(Dim<2>).hash_code()) { } else if (d.var.TypeId() == typeid(Dim<2>).hash_code()) {
return vistor(d.var.Get<Dim<2>>()); return vistor(d.var.Get<Dim<2>>());
} else if (d.var.TypeId() == typeid(Dim<3>).hash_code()) { } else if (d.var.TypeId() == typeid(Dim<3>).hash_code()) {
return vistor(d.var.Get<Dim<3>>()); return vistor(d.var.Get<Dim<3>>());
} else if (d.var.TypeId() == typeid(Dim<4>).hash_code()) { } else if (d.var.TypeId() == typeid(Dim<4>).hash_code()) {
return vistor(d.var.Get<Dim<4>>()); return vistor(d.var.Get<Dim<4>>());
} else if (d.var.TypeId() == typeid(Dim<5>).hash_code()) { } else if (d.var.TypeId() == typeid(Dim<5>).hash_code()) {
return vistor(d.var.Get<Dim<5>>()); return vistor(d.var.Get<Dim<5>>());
} else if (d.var.TypeId() == typeid(Dim<6>).hash_code()) { } else if (d.var.TypeId() == typeid(Dim<6>).hash_code()) {
return vistor(d.var.Get<Dim<6>>()); return vistor(d.var.Get<Dim<6>>());
} else if (d.var.TypeId() == typeid(Dim<7>).hash_code()) { } else if (d.var.TypeId() == typeid(Dim<7>).hash_code()) {
return vistor(d.var.Get<Dim<7>>()); return vistor(d.var.Get<Dim<7>>());
} else if (d.var.TypeId() == typeid(Dim<8>).hash_code()) { } else if (d.var.TypeId() == typeid(Dim<8>).hash_code()) {
return vistor(d.var.Get<Dim<8>>()); return vistor(d.var.Get<Dim<8>>());
} else if (d.var.TypeId() == typeid(Dim<9>).hash_code()) { } else if (d.var.TypeId() == typeid(Dim<9>).hash_code()) {
return vistor(d.var.Get<Dim<9>>()); return vistor(d.var.Get<Dim<9>>());
} else { } else {
printf(" dim not support \n"); printf(" dim not support \n");
throw std::bad_exception(); throw std::bad_exception();
// return typename Vistor::type_t(); // return typename Vistor::type_t();
}
} }
}
DDim() { var.Set<Dim<1>>(Dim<1>()); } DDim() { var.Set<Dim<1>>(Dim<1>()); }
template <int D> explicit DDim(const Dim<D> &in) { var.Set<Dim<D>>(in); } template <int D> explicit DDim(const Dim<D> &in) { var.Set<Dim<D>>(in); }
/*implicit*/ DDim(std::initializer_list<int64_t> init_list); /*implicit*/ DDim(std::initializer_list<int64_t> init_list);
template <int D> DDim &operator=(const Dim<D> &in) { template <int D> DDim &operator=(const Dim<D> &in) {
var.Set<Dim<D>>(in); var.Set<Dim<D>>(in);
return *this; return *this;
} }
int64_t &operator[](int idx); int64_t &operator[](int idx);
int64_t operator[](int idx) const; int64_t operator[](int idx) const;
// template <typename Visitor> // template <typename Visitor>
// typename Visitor::result_type apply_visitor(Visitor& visitor) { // typename Visitor::result_type apply_visitor(Visitor& visitor) {
// return var.apply_visitor(visitor); // return var.apply_visitor(visitor);
// } // }
// //
// template <typename Visitor> // template <typename Visitor>
// typename Visitor::result_type apply_visitor(Visitor& visitor) // typename Visitor::result_type apply_visitor(Visitor& visitor)
// const { // const {
// return var.apply_visitor(visitor); // return var.apply_visitor(visitor);
// } // }
DDimVar getVar() { return var; } DDimVar getVar() { return var; }
bool operator==(DDim d) const; bool operator==(DDim d) const;
bool operator!=(DDim d) const; bool operator!=(DDim d) const;
DDim operator+(DDim d) const; DDim operator+(DDim d) const;
DDim operator*(DDim d) const; DDim operator*(DDim d) const;
int size() const; int size() const;
}; };
/** /**
......
src/framework/dim.h
浏览文件 @ dff4a781
...@@ -25,199 +25,197 @@ namespace framework { ...@@ -25,199 +25,197 @@ namespace framework {
// Statically sized, statically indexed dimension // Statically sized, statically indexed dimension
template <int i> struct Dim { template <int i> struct Dim {
static constexpr int dimensions = i; static constexpr int dimensions = i;
template <typename... Args> template <typename... Args>
HOSTDEVICE Dim(int64_t _head, Args... _tail) : head(_head), tail(_tail...) { HOSTDEVICE Dim(int64_t _head, Args... _tail) : head(_head), tail(_tail...) {
static_assert(sizeof...(_tail) == i - 1, static_assert(sizeof...(_tail) == i - 1,
"Dim initialized with the wrong number of parameters"); "Dim initialized with the wrong number of parameters");
} }
HOSTDEVICE HOSTDEVICE
Dim(int64_t _head, const Dim<i - 1> &_tail) : head(_head), tail(_tail) {} Dim(int64_t _head, const Dim<i - 1> &_tail) : head(_head), tail(_tail) {}
HOSTDEVICE HOSTDEVICE
Dim() : head(0), tail() {} Dim() : head(0), tail() {}
/** Construct a Dim from a linear index and size. Uses Fortran /** Construct a Dim from a linear index and size. Uses Fortran
* order * order
* indexing. */ * indexing. */
HOSTDEVICE HOSTDEVICE
Dim(int64_t idx, const Dim<i> &size) Dim(int64_t idx, const Dim<i> &size)
: head(idx % size.head), tail(idx / size.head, size.tail) {} : head(idx % size.head), tail(idx / size.head, size.tail) {}
/** Construct a Dim with each dimension set to the given index */ /** Construct a Dim with each dimension set to the given index */
HOSTDEVICE HOSTDEVICE
Dim(int64_t idx) : head(idx), tail(idx) {} Dim(int64_t idx) : head(idx), tail(idx) {}
HOSTDEVICE HOSTDEVICE
bool operator==(const Dim<i> &o) const { bool operator==(const Dim<i> &o) const {
return (head == o.head) && (tail == o.tail); return (head == o.head) && (tail == o.tail);
} }
HOSTDEVICE HOSTDEVICE
bool operator!=(const Dim<i> &o) const { return !(*this == o); } bool operator!=(const Dim<i> &o) const { return !(*this == o); }
HOSTDEVICE HOSTDEVICE
int64_t &operator[](int idx); int64_t &operator[](int idx);
HOSTDEVICE HOSTDEVICE
int64_t operator[](int idx) const; int64_t operator[](int idx) const;
HOST std::string to_string() const; HOST std::string to_string() const;
int64_t head; int64_t head;
Dim<i - 1> tail; Dim<i - 1> tail;
}; };
// Base case specialization // Base case specialization
template <> struct Dim<0> { template <> struct Dim<0> {
static constexpr int dimensions = 0; static constexpr int dimensions = 0;
HOSTDEVICE HOSTDEVICE
Dim(int64_t _head) {} Dim(int64_t _head) {}
HOSTDEVICE HOSTDEVICE
Dim() {} Dim() {}
HOSTDEVICE HOSTDEVICE
Dim(int idx, const Dim<0> &size) { Dim(int idx, const Dim<0> &size) {
#ifndef __CUDA_ARCH__ #ifndef __CUDA_ARCH__
if (idx > 0) { if (idx > 0) {
throw std::invalid_argument("Index out of range."); throw std::invalid_argument("Index out of range.");
} }
#else #else
PADDLE_ASSERT(idx == 0); PADDLE_ASSERT(idx == 0);
#endif #endif
} }
HOSTDEVICE HOSTDEVICE
bool operator==(const Dim<0> &o) const { return true; } bool operator==(const Dim<0> &o) const { return true; }
HOSTDEVICE HOSTDEVICE
bool operator!=(const Dim<0> &o) const { return false; } bool operator!=(const Dim<0> &o) const { return false; }
HOSTDEVICE HOSTDEVICE
int64_t &operator[](int idx); int64_t &operator[](int idx);
HOSTDEVICE HOSTDEVICE
int64_t operator[](int idx) const; int64_t operator[](int idx) const;
}; };
namespace { namespace {
// Helper for accessing Dim classes // Helper for accessing Dim classes
template <int i> struct DimGetter { template <int i> struct DimGetter {
// Return a copy if Dim is const // Return a copy if Dim is const
template <typename D> HOSTDEVICE static int64_t impl(const D &d) { template <typename D> HOSTDEVICE static int64_t impl(const D &d) {
return DimGetter<i - 1>::impl(d.tail); return DimGetter<i - 1>::impl(d.tail);
} }
// Return a reference if Dim is mutable // Return a reference if Dim is mutable
template <typename D> HOSTDEVICE static int64_t &impl(D &d) { template <typename D> HOSTDEVICE static int64_t &impl(D &d) {
return DimGetter<i - 1>::impl(d.tail); return DimGetter<i - 1>::impl(d.tail);
} }
}; };
// Eureka! We found the element! // Eureka! We found the element!
template <> struct DimGetter<0> { template <> struct DimGetter<0> {
// Return a copy if Dim is const // Return a copy if Dim is const
template <typename D> HOSTDEVICE static int64_t impl(const D &d) { template <typename D> HOSTDEVICE static int64_t impl(const D &d) {
return d.head; return d.head;
} }
// Return a reference if Dim is mutable // Return a reference if Dim is mutable
template <typename D> HOSTDEVICE static int64_t &impl(D &d) { template <typename D> HOSTDEVICE static int64_t &impl(D &d) { return d.head; }
return d.head;
}
}; };
template <int D> HOSTDEVICE int64_t &indexer(Dim<D> &dim, int idx) { template <int D> HOSTDEVICE int64_t &indexer(Dim<D> &dim, int idx) {
#ifndef __CUDA_ARCH__ #ifndef __CUDA_ARCH__
if (idx < 0) { if (idx < 0) {
throw std::invalid_argument("Tried to access a negative dimension"); throw std::invalid_argument("Tried to access a negative dimension");
} }
#else #else
PADDLE_ASSERT(idx >= 0); PADDLE_ASSERT(idx >= 0);
#endif #endif
if (idx == 0) { if (idx == 0) {
return dim.head; return dim.head;
} }
return indexer(dim.tail, idx - 1); return indexer(dim.tail, idx - 1);
} }
template <> HOSTDEVICE int64_t &indexer<0>(Dim<0> &dim, int idx) { template <> HOSTDEVICE int64_t &indexer<0>(Dim<0> &dim, int idx) {
#ifndef __CUDA_ARCH__ #ifndef __CUDA_ARCH__
throw std::invalid_argument("Invalid index"); throw std::invalid_argument("Invalid index");
#else #else
PADDLE_ASSERT(false); PADDLE_ASSERT(false);
#if CUDA_VERSION < 8000 #if CUDA_VERSION < 8000
// On CUDA versions previous to 8.0, only __shared__ variables // On CUDA versions previous to 8.0, only __shared__ variables
// could be declared as static in the device code. // could be declared as static in the device code.
int64_t head = 0; int64_t head = 0;
#else #else
static int64_t head = 0; static int64_t head = 0;
#endif #endif
return head; return head;
#endif #endif
} }
template <int D> HOSTDEVICE int64_t indexer(const Dim<D> &dim, int idx) { template <int D> HOSTDEVICE int64_t indexer(const Dim<D> &dim, int idx) {
#ifndef __CUDA_ARCH__ #ifndef __CUDA_ARCH__
if (idx < 0) { if (idx < 0) {
throw std::invalid_argument("Tried to access a negative dimension"); throw std::invalid_argument("Tried to access a negative dimension");
} }
#else #else
PADDLE_ASSERT(idx >= 0); PADDLE_ASSERT(idx >= 0);
#endif #endif
if (idx == 0) { if (idx == 0) {
return dim.head; return dim.head;
} }
return indexer(dim.tail, idx - 1); return indexer(dim.tail, idx - 1);
} }
template <> HOSTDEVICE int64_t indexer<0>(const Dim<0> &dim, int idx) { template <> HOSTDEVICE int64_t indexer<0>(const Dim<0> &dim, int idx) {
#ifndef __CUDA_ARCH__ #ifndef __CUDA_ARCH__
throw std::invalid_argument("Invalid index"); throw std::invalid_argument("Invalid index");
#else #else
PADDLE_ASSERT(false); PADDLE_ASSERT(false);
#if CUDA_VERSION < 8000 #if CUDA_VERSION < 8000
// On CUDA versions previous to 8.0, only __shared__ variables // On CUDA versions previous to 8.0, only __shared__ variables
// could be declared as static in the device code. // could be declared as static in the device code.
int64_t head = 0; int64_t head = 0;
#else #else
static int64_t head = 0; static int64_t head = 0;
#endif #endif
return head; return head;
#endif #endif
} }
} // namespace } // namespace
// Static access to constant Dim // Static access to constant Dim
template <int i, int l> HOSTDEVICE int64_t get(const Dim<l> &d) { template <int i, int l> HOSTDEVICE int64_t get(const Dim<l> &d) {
return DimGetter<i>::impl(d); return DimGetter<i>::impl(d);
} }
// Static access to mutable Dim // Static access to mutable Dim
template <int i, int l> HOSTDEVICE int64_t &get(Dim<l> &d) { template <int i, int l> HOSTDEVICE int64_t &get(Dim<l> &d) {
return DimGetter<i>::impl(d); return DimGetter<i>::impl(d);
} }
// Dynamic access to constant Dim // Dynamic access to constant Dim
template <int l> HOSTDEVICE int64_t Dim<l>::operator[](int i) const { template <int l> HOSTDEVICE int64_t Dim<l>::operator[](int i) const {
// std::cout << "l: " << l << std::endl; // std::cout << "l: " << l << std::endl;
return indexer(*this, i); return indexer(*this, i);
} }
// Dynamic access to mutable Dim // Dynamic access to mutable Dim
template <int l> HOSTDEVICE int64_t &Dim<l>::operator[](int i) { template <int l> HOSTDEVICE int64_t &Dim<l>::operator[](int i) {
return indexer(*this, i); return indexer(*this, i);
} }
// Dynamic access to constant Dim // Dynamic access to constant Dim
inline HOSTDEVICE int64_t Dim<0>::operator[](int i) const { inline HOSTDEVICE int64_t Dim<0>::operator[](int i) const {
return indexer(*this, i); return indexer(*this, i);
} }
// Dynamic access to mutable Dim // Dynamic access to mutable Dim
inline HOSTDEVICE int64_t &Dim<0>::operator[](int i) { inline HOSTDEVICE int64_t &Dim<0>::operator[](int i) {
return indexer(*this, i); return indexer(*this, i);
} }
// Dynamic access to constant Dim // Dynamic access to constant Dim
...@@ -225,52 +223,52 @@ inline HOSTDEVICE int64_t &Dim<0>::operator[](int i) { ...@@ -225,52 +223,52 @@ inline HOSTDEVICE int64_t &Dim<0>::operator[](int i) {
template <int l> template <int l>
HOSTDEVICE typename std::enable_if<(l > 0), int64_t>::type get(const Dim<l> &d, HOSTDEVICE typename std::enable_if<(l > 0), int64_t>::type get(const Dim<l> &d,
int i) { int i) {
return d[i]; return d[i];
} }
// Dynamic access to mutable Dim // Dynamic access to mutable Dim
template <int l> template <int l>
HOSTDEVICE typename std::enable_if<(l > 0), int64_t &>::type get(Dim<l> &d, HOSTDEVICE typename std::enable_if<(l > 0), int64_t &>::type get(Dim<l> &d,
int i) { int i) {
return d[i]; return d[i];
} }
// Dot product of two dims // Dot product of two dims
template <int i> template <int i>
HOSTDEVICE int64_t linearize(const Dim<i> &a, const Dim<i> &b) { HOSTDEVICE int64_t linearize(const Dim<i> &a, const Dim<i> &b) {
return a.head * b.head + linearize(a.tail, b.tail); return a.head * b.head + linearize(a.tail, b.tail);
} }
// Base case dot product of two Dims // Base case dot product of two Dims
// Notice it is inline because it is no longer a template // Notice it is inline because it is no longer a template
template <> template <>
HOSTDEVICE inline int64_t linearize(const Dim<0> &a, const Dim<0> &b) { HOSTDEVICE inline int64_t linearize(const Dim<0> &a, const Dim<0> &b) {
return 0; return 0;
} }
// Product of a Dim // Product of a Dim
template <int i> HOSTDEVICE int64_t product(const Dim<i> &a, int prod = 1) { template <int i> HOSTDEVICE int64_t product(const Dim<i> &a, int prod = 1) {
return prod * a.head * product(a.tail); return prod * a.head * product(a.tail);
} }
// Base case product of a Dim // Base case product of a Dim
// Notice it is inline because it is no longer a template // Notice it is inline because it is no longer a template
template <> HOSTDEVICE inline int64_t product(const Dim<0> &a, int prod) { template <> HOSTDEVICE inline int64_t product(const Dim<0> &a, int prod) {
return prod; return prod;
} }
// Is 0 <= idx_i < size_i for all i? // Is 0 <= idx_i < size_i for all i?
template <int i> template <int i>
HOSTDEVICE bool contained(const Dim<i> &idx, const Dim<i> &size) { HOSTDEVICE bool contained(const Dim<i> &idx, const Dim<i> &size) {
return ((0 <= idx.head) && (idx.head < size.head) && return ((0 <= idx.head) && (idx.head < size.head) &&
contained(idx.tail, size.tail)); contained(idx.tail, size.tail));
} }
// Base case of is 0 <= idx_i < size_i ? // Base case of is 0 <= idx_i < size_i ?
// Notice it is inline because it is no longer a template // Notice it is inline because it is no longer a template
template <> template <>
HOSTDEVICE inline bool contained(const Dim<0> &idx, const Dim<0> &size) { HOSTDEVICE inline bool contained(const Dim<0> &idx, const Dim<0> &size) {
return true; return true;
} }
/** /**
...@@ -278,14 +276,14 @@ HOSTDEVICE inline bool contained(const Dim<0> &idx, const Dim<0> &size) { ...@@ -278,14 +276,14 @@ HOSTDEVICE inline bool contained(const Dim<0> &idx, const Dim<0> &size) {
*/ */
template <int i> template <int i>
HOSTDEVICE Dim<i> ex_prefix_mul(const Dim<i> &src, int mul = 1) { HOSTDEVICE Dim<i> ex_prefix_mul(const Dim<i> &src, int mul = 1) {
return Dim<i>(mul, ex_prefix_mul(src.tail, mul * src.head)); return Dim<i>(mul, ex_prefix_mul(src.tail, mul * src.head));
} }
///\cond HIDDEN ///\cond HIDDEN
// Base case of ex_prefix_mul // Base case of ex_prefix_mul
// Notice it is inline because it is no longer a template // Notice it is inline because it is no longer a template
template <> HOSTDEVICE inline Dim<0> ex_prefix_mul(const Dim<0> &src, int mul) { template <> HOSTDEVICE inline Dim<0> ex_prefix_mul(const Dim<0> &src, int mul) {
return Dim<0>(); return Dim<0>();
} }
///\endcond ///\endcond
...@@ -293,36 +291,36 @@ template <> HOSTDEVICE inline Dim<0> ex_prefix_mul(const Dim<0> &src, int mul) { ...@@ -293,36 +291,36 @@ template <> HOSTDEVICE inline Dim<0> ex_prefix_mul(const Dim<0> &src, int mul) {
* Add two dimensions together * Add two dimensions together
*/ */
template <int i> HOSTDEVICE Dim<i> dim_plus(const Dim<i> &a, const Dim<i> &b) { template <int i> HOSTDEVICE Dim<i> dim_plus(const Dim<i> &a, const Dim<i> &b) {
return Dim<i>(a.head + b.head, dim_plus(a.tail, b.tail)); return Dim<i>(a.head + b.head, dim_plus(a.tail, b.tail));
} }
// Base case // Base case
template <> template <>
HOSTDEVICE inline Dim<0> dim_plus(const Dim<0> &a, const Dim<0> &b) { HOSTDEVICE inline Dim<0> dim_plus(const Dim<0> &a, const Dim<0> &b) {
return Dim<0>(); return Dim<0>();
} }
template <int i> template <int i>
HOSTDEVICE Dim<i> operator+(const Dim<i> &lhs, const Dim<i> &rhs) { HOSTDEVICE Dim<i> operator+(const Dim<i> &lhs, const Dim<i> &rhs) {
return dim_plus(lhs, rhs); return dim_plus(lhs, rhs);
} }
/** /**
* Multiply two dimensions together * Multiply two dimensions together
*/ */
template <int i> HOSTDEVICE Dim<i> dim_mult(const Dim<i> &a, const Dim<i> &b) { template <int i> HOSTDEVICE Dim<i> dim_mult(const Dim<i> &a, const Dim<i> &b) {
return Dim<i>(a.head * b.head, dim_mult(a.tail, b.tail)); return Dim<i>(a.head * b.head, dim_mult(a.tail, b.tail));
} }
// Base case // Base case
template <> template <>
HOSTDEVICE inline Dim<0> dim_mult(const Dim<0> &a, const Dim<0> &b) { HOSTDEVICE inline Dim<0> dim_mult(const Dim<0> &a, const Dim<0> &b) {
return Dim<0>(); return Dim<0>();
} }
template <int i> template <int i>
HOSTDEVICE Dim<i> operator*(const Dim<i> &lhs, const Dim<i> &rhs) { HOSTDEVICE Dim<i> operator*(const Dim<i> &lhs, const Dim<i> &rhs) {
return dim_mult(lhs, rhs); return dim_mult(lhs, rhs);
} }
/** /**
...@@ -337,8 +335,8 @@ HOSTDEVICE Dim<i> operator*(const Dim<i> &lhs, const Dim<i> &rhs) { ...@@ -337,8 +335,8 @@ HOSTDEVICE Dim<i> operator*(const Dim<i> &lhs, const Dim<i> &rhs) {
template <int i> template <int i>
HOSTDEVICE Dim<i> normalize_strides(const Dim<i> &size, const Dim<i> &stride) { HOSTDEVICE Dim<i> normalize_strides(const Dim<i> &size, const Dim<i> &stride) {
int norm_stride = size.head == 1 ? 0 : stride.head; int norm_stride = size.head == 1 ? 0 : stride.head;
return Dim<i>(norm_stride, normalize_strides(size.tail, stride.tail)); return Dim<i>(norm_stride, normalize_strides(size.tail, stride.tail));
} }
///\cond HIDDEN ///\cond HIDDEN
...@@ -346,7 +344,7 @@ HOSTDEVICE Dim<i> normalize_strides(const Dim<i> &size, const Dim<i> &stride) { ...@@ -346,7 +344,7 @@ HOSTDEVICE Dim<i> normalize_strides(const Dim<i> &size, const Dim<i> &stride) {
template <> template <>
HOSTDEVICE inline Dim<0> normalize_strides(const Dim<0> &size, HOSTDEVICE inline Dim<0> normalize_strides(const Dim<0> &size,
const Dim<0> &stride) { const Dim<0> &stride) {
return Dim<0>(); return Dim<0>();
} }
///\endcond ///\endcond
...@@ -361,7 +359,7 @@ HOSTDEVICE inline Dim<0> normalize_strides(const Dim<0> &size, ...@@ -361,7 +359,7 @@ HOSTDEVICE inline Dim<0> normalize_strides(const Dim<0> &size,
template <typename... Args> template <typename... Args>
HOSTDEVICE Dim<sizeof...(Args)> make_dim(Args... idxes) { HOSTDEVICE Dim<sizeof...(Args)> make_dim(Args... idxes) {
return Dim<sizeof...(Args)>(idxes...); return Dim<sizeof...(Args)>(idxes...);
} }
// Allows us to output a Dim // Allows us to output a Dim
...@@ -369,8 +367,8 @@ HOSTDEVICE Dim<sizeof...(Args)> make_dim(Args... idxes) { ...@@ -369,8 +367,8 @@ HOSTDEVICE Dim<sizeof...(Args)> make_dim(Args... idxes) {
template <int i> template <int i>
typename std::enable_if<(i > 1), std::ostream &>::type typename std::enable_if<(i > 1), std::ostream &>::type
operator<<(std::ostream &os, const Dim<i> &d) { operator<<(std::ostream &os, const Dim<i> &d) {
os << d.head << ", " << d.tail; os << d.head << ", " << d.tail;
return os; return os;
} }
// Base case that allows us to output a Dim // Base case that allows us to output a Dim
...@@ -378,34 +376,34 @@ operator<<(std::ostream &os, const Dim<i> &d) { ...@@ -378,34 +376,34 @@ operator<<(std::ostream &os, const Dim<i> &d) {
template <int i> template <int i>
typename std::enable_if<(i == 1), std::ostream &>::type typename std::enable_if<(i == 1), std::ostream &>::type
operator<<(std::ostream &os, const Dim<i> &d) { operator<<(std::ostream &os, const Dim<i> &d) {
os << d.head; os << d.head;
return os; return os;
} }
inline std::ostream &operator<<(std::ostream &os, const Dim<0> &d) { inline std::ostream &operator<<(std::ostream &os, const Dim<0> &d) {
return os; return os;
} }
template <int i> HOST std::string Dim<i>::to_string() const { template <int i> HOST std::string Dim<i>::to_string() const {
std::stringstream stream; std::stringstream stream;
stream << *this; stream << *this;
return stream.str(); return stream.str();
} }
template <int D> template <int D>
HOSTDEVICE Dim<D> linear_to_dimension(int linear_index, Dim<D> extents) { HOSTDEVICE Dim<D> linear_to_dimension(int linear_index, Dim<D> extents) {
Dim<D> result; Dim<D> result;
for (int i = 0; i < D - 1; ++i) { for (int i = 0; i < D - 1; ++i) {
result[i] = linear_index % extents[i]; result[i] = linear_index % extents[i];
linear_index /= extents[i]; linear_index /= extents[i];
} }
result[D - 1] = linear_index; result[D - 1] = linear_index;
return result; return result;
} }
} // namespace framework } // namespace framework
......
src/framework/executor.cpp
浏览文件 @ dff4a781
...@@ -26,66 +26,66 @@ namespace framework { ...@@ -26,66 +26,66 @@ namespace framework {
template <typename Dtype> template <typename Dtype>
Executor<Dtype>::Executor(const Program<Dtype> p) : program_(p) { Executor<Dtype>::Executor(const Program<Dtype> p) : program_(p) {
if (use_optimize_) { if (use_optimize_) {
to_predict_program_ = program_.optimizeProgram; to_predict_program_ = program_.optimizeProgram;
} else { } else {
to_predict_program_ = program_.originProgram; to_predict_program_ = program_.originProgram;
} }
// const std::vector<std::shared_ptr<BlockDesc>> blocks = // const std::vector<std::shared_ptr<BlockDesc>> blocks =
to_predict_program_->Blocks(); to_predict_program_->Blocks();
// for (int i = 0; i < blocks.size(); ++i) { // for (int i = 0; i < blocks.size(); ++i) {
// std::shared_ptr<BlockDesc> block_desc = blocks[i]; // std::shared_ptr<BlockDesc> block_desc = blocks[i];
// std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops(); // std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops();
// for (int j = 0; j < ops.size(); ++j) { // for (int j = 0; j < ops.size(); ++j) {
// std::shared_ptr<OpDesc> op = ops[j]; // std::shared_ptr<OpDesc> op = ops[j];
// if (op->Type() == "conv2d" && op->Input("Input")[0] == // if (op->Type() == "conv2d" && op->Input("Input")[0] ==
// "pixel") { // "pixel") {
// Attribute strides_attr = op->GetAttrMap().at("strides"); // Attribute strides_attr = op->GetAttrMap().at("strides");
// std::vector<int> stride = // std::vector<int> stride =
// strides_attr.Get<std::vector<int>>(); for (int k = 0; k < // strides_attr.Get<std::vector<int>>(); for (int k = 0; k <
// stride.size(); ++k) { // stride.size(); ++k) {
// } // }
// std::shared_ptr<operators::ConvOp<Dtype, float>> conv = // std::shared_ptr<operators::ConvOp<Dtype, float>> conv =
// std::make_shared<operators::ConvOp<Dtype, float>>( // std::make_shared<operators::ConvOp<Dtype, float>>(
// op->Type(), op->GetInputs(), op->GetOutputs(), // op->Type(), op->GetInputs(), op->GetOutputs(),
// op->GetAttrMap(), program_.scope); // op->GetAttrMap(), program_.scope);
// ops_of_block_[*block_desc.get()].push_back(conv); // ops_of_block_[*block_desc.get()].push_back(conv);
// } // }
// } // }
// } // }
} }
template <typename Dtype> template <typename Dtype>
std::shared_ptr<Tensor> Executor<Dtype>::predict(Tensor &t) { std::shared_ptr<Tensor> Executor<Dtype>::predict(Tensor &t) {
// feed // feed
auto scope = program_.scope; auto scope = program_.scope;
Variable *g_feed_value = scope->Var("pixel"); Variable *g_feed_value = scope->Var("pixel");
auto tensor = g_feed_value->GetMutable<Tensor>(); auto tensor = g_feed_value->GetMutable<Tensor>();
tensor->ShareDataWith(t); tensor->ShareDataWith(t);
Variable *con_output = scope->Var("conv2d_0.tmp_0"); Variable *con_output = scope->Var("conv2d_0.tmp_0");
Tensor *output_tensor = con_output->GetMutable<Tensor>(); Tensor *output_tensor = con_output->GetMutable<Tensor>();
output_tensor->mutable_data<float>({1, 16, 32, 32}); output_tensor->mutable_data<float>({1, 16, 32, 32});
// std::cout << typeid(output_tensor).name() << std::endl; // std::cout << typeid(output_tensor).name() << std::endl;
// std::cout << "output_tensor dims: " << output_tensor->dims() << // std::cout << "output_tensor dims: " << output_tensor->dims() <<
// std::endl; // std::endl;
std::shared_ptr<Tensor> out_tensor = std::make_shared<LoDTensor>(); std::shared_ptr<Tensor> out_tensor = std::make_shared<LoDTensor>();
out_tensor.reset(output_tensor); out_tensor.reset(output_tensor);
predict(t, 0); predict(t, 0);
return out_tensor; return out_tensor;
} }
template <typename Dtype> template <typename Dtype>
void Executor<Dtype>::predict(const Tensor &t, int block_id) { void Executor<Dtype>::predict(const Tensor &t, int block_id) {
std::shared_ptr<BlockDesc> to_predict_block = std::shared_ptr<BlockDesc> to_predict_block =
to_predict_program_->Block(block_id); to_predict_program_->Block(block_id);
for (int j = 0; j < ops_of_block_[*to_predict_block.get()].size(); ++j) { for (int j = 0; j < ops_of_block_[*to_predict_block.get()].size(); ++j) {
auto op = ops_of_block_[*to_predict_block.get()][j]; auto op = ops_of_block_[*to_predict_block.get()][j];
op->Run(); op->Run();
} }
} }
template class Executor<CPU>; template class Executor<CPU>;
......
src/framework/executor.h
浏览文件 @ dff4a781
...@@ -35,23 +35,23 @@ namespace paddle_mobile { ...@@ -35,23 +35,23 @@ namespace paddle_mobile {
namespace framework { namespace framework {
template <typename Dtype> class Executor { template <typename Dtype> class Executor {
public: public:
Executor(); Executor();
Executor(const Program<Dtype> p); Executor(const Program<Dtype> p);
std::shared_ptr<Tensor> predict(Tensor &t); std::shared_ptr<Tensor> predict(Tensor &t);
public: public:
const framework::Program<Dtype> program_; const framework::Program<Dtype> program_;
std::shared_ptr<ProgramDesc> to_predict_program_; std::shared_ptr<ProgramDesc> to_predict_program_;
void predict(const Tensor &t, int block_id); void predict(const Tensor &t, int block_id);
std::map<framework::BlockDesc, std::map<framework::BlockDesc,
std::vector<std::shared_ptr<OperatorBase<Dtype>>>> std::vector<std::shared_ptr<OperatorBase<Dtype>>>>
ops_of_block_; ops_of_block_;
bool use_optimize_ = false; bool use_optimize_ = false;
}; };
} // namespace framework } // namespace framework
......
src/framework/framework.pb.cpp
浏览文件 @ dff4a781
因为 它太大了无法显示 source diff 。你可以改为 查看blob。
src/framework/framework.pb.h
浏览文件 @ dff4a781
因为 它太大了无法显示 source diff 。你可以改为 查看blob。
src/framework/lod_tensor.cc
浏览文件 @ dff4a781
...@@ -22,291 +22,289 @@ namespace paddle_mobile { ...@@ -22,291 +22,289 @@ namespace paddle_mobile {
namespace framework { namespace framework {
std::ostream &operator<<(std::ostream &os, const LoD &lod) { std::ostream &operator<<(std::ostream &os, const LoD &lod) {
os << "{";
for (auto &v : lod) {
os << "{"; os << "{";
for (auto &v : lod) { bool is_first = true;
os << "{"; for (auto &i : v) {
bool is_first = true; if (is_first) {
for (auto &i : v) { os << i;
if (is_first) { is_first = false;
os << i; } else {
is_first = false; os << ", " << i;
} else { }
os << ", " << i;
}
}
os << "}";
} }
os << "}"; os << "}";
}
os << "}";
return os; return os;
} }
std::ostream &operator<<(std::ostream &os, const LoDTensor &t) { std::ostream &operator<<(std::ostream &os, const LoDTensor &t) {
// PADDLE_ENFORCE(t.type().hash_code() == // PADDLE_ENFORCE(t.type().hash_code() ==
// typeid(float).hash_code()); // typeid(float).hash_code());
// if (!platform::is_cpu_place(t.place())) { // if (!platform::is_cpu_place(t.place())) {
// LoDTensor tt; // LoDTensor tt;
// framework::TensorCopy(t, platform::CPUPlace(), &tt); // framework::TensorCopy(t, platform::CPUPlace(), &tt);
// platform::DeviceContextPool &pool = // platform::DeviceContextPool &pool =
// platform::DeviceContextPool::Instance(); auto &dev_ctx = // platform::DeviceContextPool::Instance(); auto &dev_ctx =
// *pool.Get(t.place()); dev_ctx.Wait(); // *pool.Get(t.place()); dev_ctx.Wait();
// //
// os << tt; // os << tt;
// return os; // return os;
// } // }
os << "dim: " << t.dims() << "\n"; os << "dim: " << t.dims() << "\n";
os << "lod: " << t.lod() << "\n"; os << "lod: " << t.lod() << "\n";
// only print first ten elements // only print first ten elements
int64_t size = t.numel() < 10 ? t.numel() : 10; int64_t size = t.numel() < 10 ? t.numel() : 10;
for (int64_t i = 0; i < size; ++i) { for (int64_t i = 0; i < size; ++i) {
os << t.data<float>()[i] << " "; os << t.data<float>()[i] << " ";
} }
return os; return os;
} }
std::string LoDToString(const LoD &lod) { std::string LoDToString(const LoD &lod) {
std::ostringstream stream; std::ostringstream stream;
stream << lod; stream << lod;
return stream.str(); return stream.str();
} }
LoD SliceInLevel(const LoD &in, size_t level, size_t elem_begin, LoD SliceInLevel(const LoD &in, size_t level, size_t elem_begin,
size_t elem_end) { size_t elem_end) {
// PADDLE_ENFORCE_LT(level, in.size()); // PADDLE_ENFORCE_LT(level, in.size());
// PADDLE_ENFORCE_LT(elem_end, in[level].size()); // PADDLE_ENFORCE_LT(elem_end, in[level].size());
LoD res; LoD res;
res.resize(in.size() - level); res.resize(in.size() - level);
// copy the first level // copy the first level
res[0].assign(in[level].begin() + elem_begin, res[0].assign(in[level].begin() + elem_begin,
in[level].begin() + elem_end + 1); in[level].begin() + elem_end + 1);
for (size_t lvl = 1; lvl < res.size(); lvl++) { for (size_t lvl = 1; lvl < res.size(); lvl++) {
const auto &in_level = in[level + lvl]; const auto &in_level = in[level + lvl];
const auto &above_level = res[lvl - 1]; const auto &above_level = res[lvl - 1];
auto &out_level = res[lvl]; auto &out_level = res[lvl];
out_level.assign(in_level.begin() + above_level.front(), out_level.assign(in_level.begin() + above_level.front(),
in_level.begin() + above_level.back() + 1); in_level.begin() + above_level.back() + 1);
} }
for (size_t lvl = 0; lvl < res.size(); lvl++) { for (size_t lvl = 0; lvl < res.size(); lvl++) {
// to make the first offset equals 0, all the elements minus the // to make the first offset equals 0, all the elements minus the
// first // first
// element // element
size_t front = res[lvl].front(); size_t front = res[lvl].front();
for (auto &ele : res[lvl]) { for (auto &ele : res[lvl]) {
ele -= front; ele -= front;
}
} }
return res; }
return res;
} }
LoD ToAbsOffset(const LoD &in) { LoD ToAbsOffset(const LoD &in) {
// the lowest level stores relative offsets // the lowest level stores relative offsets
if (in.empty() || in.size() == 1) if (in.empty() || in.size() == 1)
return in; return in;
LoD result = in; LoD result = in;
for (auto level = static_cast<int>(in.size() - 2); level >= 0; level--) { for (auto level = static_cast<int>(in.size() - 2); level >= 0; level--) {
for (size_t i = 0; i < in[level].size(); ++i) { for (size_t i = 0; i < in[level].size(); ++i) {
size_t index = in[level][i]; size_t index = in[level][i];
result[level][i] = result[level + 1][index]; result[level][i] = result[level + 1][index];
}
} }
return result; }
return result;
} }
bool operator==(const LoD &a, const LoD &b) { bool operator==(const LoD &a, const LoD &b) {
if (a.size() != b.size()) { if (a.size() != b.size()) {
return false; return false;
}
for (size_t i = 0; i < a.size(); i++) {
const auto &a_level = a[i];
const auto &b_level = b[i];
if (a_level.size() != b_level.size()) {
return false;
} }
for (size_t j = 0; j < a_level.size(); j++) {
for (size_t i = 0; i < a.size(); i++) { if (a_level[j] != b_level[j]) {
const auto &a_level = a[i]; return false;
const auto &b_level = b[i]; }
if (a_level.size() != b_level.size()) {
return false;
}
for (size_t j = 0; j < a_level.size(); j++) {
if (a_level[j] != b_level[j]) {
return false;
}
}
} }
return true; }
return true;
} }
bool CheckLoD(const LoD &in, int tensor_height) { bool CheckLoD(const LoD &in, int tensor_height) {
if (in.empty()) if (in.empty())
return true;
for (const auto &level : in) {
// check: there should be more than 2 offsets existing in each
// level.
if (level.size() < 2)
return false;
// check: the first offset(the begin offset) of each level
// should be 0.
if (level.front() != 0)
return false;
// check: all the offsets in a level should be ascending(no same
// items
// allows).
if (!std::is_sorted(level.begin(), level.begin(),
[](size_t a, size_t b) {
if (a < b)
return true;
return false;
})) {
std::cout << "ascending error";
return false;
}
}
// check: the lowest level's last offset should equals
// `tensor_height` if
// tensor_height>0.
if (tensor_height > 0 && (size_t)tensor_height != in.back().back())
return false;
// check: the higher level's last offset should equals the lower
// level's
// size-1.
// NOTE LoD store the levels from top to bottom, so the higher level
// goes
// first.
for (size_t level = 0; level < in.size() - 1; level++) {
if (in[level].back() != in[level + 1].size() - 1)
return false;
}
return true; return true;
for (const auto &level : in) {
// check: there should be more than 2 offsets existing in each
// level.
if (level.size() < 2)
return false;
// check: the first offset(the begin offset) of each level
// should be 0.
if (level.front() != 0)
return false;
// check: all the offsets in a level should be ascending(no same
// items
// allows).
if (!std::is_sorted(level.begin(), level.begin(), [](size_t a, size_t b) {
if (a < b)
return true;
return false;
})) {
std::cout << "ascending error";
return false;
}
}
// check: the lowest level's last offset should equals
// `tensor_height` if
// tensor_height>0.
if (tensor_height > 0 && (size_t)tensor_height != in.back().back())
return false;
// check: the higher level's last offset should equals the lower
// level's
// size-1.
// NOTE LoD store the levels from top to bottom, so the higher level
// goes
// first.
for (size_t level = 0; level < in.size() - 1; level++) {
if (in[level].back() != in[level + 1].size() - 1)
return false;
}
return true;
} }
bool CheckAbsLoD(const LoD &in, int tensor_height) { bool CheckAbsLoD(const LoD &in, int tensor_height) {
if (in.empty()) if (in.empty())
return true;
for (const auto &level : in) {
// check: all the offsets in a level should be ascending(no same
// items
// allows).
if (!std::is_sorted(level.begin(), level.begin(),
[](size_t a, size_t b) {
if (a < b)
return true;
return false;
})) {
return false;
}
// check: there should be more than 2 offsets existing in each
// level.
if (level.size() < 2)
return false;
// check: the first offset of each level should be 0, and the
// last should be
// the same(the height of underlying tensor).
if (level.front() != 0)
return false;
if (tensor_height < 0) {
tensor_height = level.back();
} else if ((size_t)tensor_height != level.back()) {
return false;
}
}
return true; return true;
for (const auto &level : in) {
// check: all the offsets in a level should be ascending(no same
// items
// allows).
if (!std::is_sorted(level.begin(), level.begin(), [](size_t a, size_t b) {
if (a < b)
return true;
return false;
})) {
return false;
}
// check: there should be more than 2 offsets existing in each
// level.
if (level.size() < 2)
return false;
// check: the first offset of each level should be 0, and the
// last should be
// the same(the height of underlying tensor).
if (level.front() != 0)
return false;
if (tensor_height < 0) {
tensor_height = level.back();
} else if ((size_t)tensor_height != level.back()) {
return false;
}
}
return true;
} }
using LoDAndOffset = std::pair<LoD, std::pair<size_t, size_t>>; using LoDAndOffset = std::pair<LoD, std::pair<size_t, size_t>>;
LoDAndOffset GetSubLoDAndAbsoluteOffset(const LoD &lod, size_t start_idx, LoDAndOffset GetSubLoDAndAbsoluteOffset(const LoD &lod, size_t start_idx,
size_t end_idx, size_t start_level) { size_t end_idx, size_t start_level) {
LoD sub_lod; LoD sub_lod;
for (size_t level_idx = start_level; level_idx < lod.size(); ++level_idx) { for (size_t level_idx = start_level; level_idx < lod.size(); ++level_idx) {
// PADDLE_ENFORCE_LE(start_idx, end_idx); // PADDLE_ENFORCE_LE(start_idx, end_idx);
// PADDLE_ENFORCE_LT(end_idx, lod[level_idx].size()); // PADDLE_ENFORCE_LT(end_idx, lod[level_idx].size());
std::vector<size_t> level_lens; std::vector<size_t> level_lens;
for (size_t i = start_idx; i < end_idx; ++i) { for (size_t i = start_idx; i < end_idx; ++i) {
level_lens.push_back(lod[level_idx][i + 1] - lod[level_idx][i]); level_lens.push_back(lod[level_idx][i + 1] - lod[level_idx][i]);
}
sub_lod.emplace_back(level_lens);
start_idx = lod[level_idx][start_idx];
end_idx = lod[level_idx][end_idx];
} }
sub_lod.emplace_back(level_lens);
start_idx = lod[level_idx][start_idx];
end_idx = lod[level_idx][end_idx];
}
return LoDAndOffset{sub_lod, {start_idx, end_idx}}; return LoDAndOffset{sub_lod, {start_idx, end_idx}};
} }
void AppendLoD(LoD *lod, const LoD &lod_length) { void AppendLoD(LoD *lod, const LoD &lod_length) {
// PADDLE_ENFORCE( // PADDLE_ENFORCE(
// lod->empty() || lod->size() == lod_length.size(), // lod->empty() || lod->size() == lod_length.size(),
// "The lod_length should has the same size with the appended // "The lod_length should has the same size with the appended
// lod."); // lod.");
if (lod->empty()) { if (lod->empty()) {
for (size_t i = 0; i < lod_length.size(); ++i) { for (size_t i = 0; i < lod_length.size(); ++i) {
lod->emplace_back(1, 0); // size = 1, value = 0; lod->emplace_back(1, 0); // size = 1, value = 0;
}
*lod = LoD(lod_length.size(), std::vector<size_t>({0}));
} }
for (size_t i = 0; i < lod->size(); ++i) { *lod = LoD(lod_length.size(), std::vector<size_t>({0}));
auto &level = (*lod)[i]; }
for (size_t len : lod_length[i]) { for (size_t i = 0; i < lod->size(); ++i) {
level.push_back(level.back() + len); auto &level = (*lod)[i];
} for (size_t len : lod_length[i]) {
level.push_back(level.back() + len);
} }
}
} }
void SerializeToStream(std::ostream &os, const LoDTensor &tensor) { void SerializeToStream(std::ostream &os, const LoDTensor &tensor) {
{ // the 1st field, uint32_t version for LoDTensor { // the 1st field, uint32_t version for LoDTensor
constexpr uint32_t version = 0; constexpr uint32_t version = 0;
os.write(reinterpret_cast<const char *>(&version), sizeof(version)); os.write(reinterpret_cast<const char *>(&version), sizeof(version));
}
{
// the 2st field, LoD information
// uint64_t lod_level
// uint64_t lod_level_1 size in byte.
// int* lod_level_1 data
// ...
auto lod = tensor.lod();
uint64_t size = lod.size();
os.write(reinterpret_cast<const char *>(&size), sizeof(size));
for (auto &each : lod) {
size = each.size() * sizeof(framework::LoD::value_type::value_type);
os.write(reinterpret_cast<const char *>(&size), sizeof(size));
os.write(reinterpret_cast<const char *>(each.data()),
static_cast<std::streamsize>(size));
} }
{ }
// the 2st field, LoD information // the 3st field, Tensor
// uint64_t lod_level TensorToStream(os, static_cast<Tensor>(tensor));
// uint64_t lod_level_1 size in byte.
// int* lod_level_1 data
// ...
auto lod = tensor.lod();
uint64_t size = lod.size();
os.write(reinterpret_cast<const char *>(&size), sizeof(size));
for (auto &each : lod) {
size = each.size() * sizeof(framework::LoD::value_type::value_type);
os.write(reinterpret_cast<const char *>(&size), sizeof(size));
os.write(reinterpret_cast<const char *>(each.data()),
static_cast<std::streamsize>(size));
}
}
// the 3st field, Tensor
TensorToStream(os, static_cast<Tensor>(tensor));
} }
void DeserializeFromStream(std::istream &is, LoDTensor *tensor) { void DeserializeFromStream(std::istream &is, LoDTensor *tensor) {
{ {
// the 1st field, unit32_t version for LoDTensor // the 1st field, unit32_t version for LoDTensor
uint32_t version; uint32_t version;
is.read(reinterpret_cast<char *>(&version), sizeof(version)); is.read(reinterpret_cast<char *>(&version), sizeof(version));
// PADDLE_ENFORCE_EQ(version, 0U, "Only version 0 is // PADDLE_ENFORCE_EQ(version, 0U, "Only version 0 is
// supported"); // supported");
} }
{ {
// the 2st field, LoD information // the 2st field, LoD information
uint64_t lod_level; uint64_t lod_level;
is.read(reinterpret_cast<char *>(&lod_level), sizeof(lod_level)); is.read(reinterpret_cast<char *>(&lod_level), sizeof(lod_level));
auto &lod = *tensor->mutable_lod(); auto &lod = *tensor->mutable_lod();
lod.resize(lod_level); lod.resize(lod_level);
for (uint64_t i = 0; i < lod_level; ++i) { for (uint64_t i = 0; i < lod_level; ++i) {
uint64_t size; uint64_t size;
is.read(reinterpret_cast<char *>(&size), sizeof(size)); is.read(reinterpret_cast<char *>(&size), sizeof(size));
std::vector<size_t> tmp(size / sizeof(size_t)); std::vector<size_t> tmp(size / sizeof(size_t));
is.read(reinterpret_cast<char *>(tmp.data()), is.read(reinterpret_cast<char *>(tmp.data()),
static_cast<std::streamsize>(size)); static_cast<std::streamsize>(size));
lod[i] = tmp; lod[i] = tmp;
}
} }
// the 3st filed, Tensor }
TensorFromStream(is, static_cast<Tensor *>(tensor)); // the 3st filed, Tensor
TensorFromStream(is, static_cast<Tensor *>(tensor));
} }
} // namespace framework } // namespace framework
......
src/framework/lod_tensor.h
浏览文件 @ dff4a781
...@@ -102,45 +102,45 @@ bool CheckAbsLoD(const LoD &in, int tensor_height = -1); ...@@ -102,45 +102,45 @@ bool CheckAbsLoD(const LoD &in, int tensor_height = -1);
* see https://en.wikipedia.org/wiki/Level_of_details for reference. * see https://en.wikipedia.org/wiki/Level_of_details for reference.
*/ */
class LoDTensor : public Tensor { class LoDTensor : public Tensor {
public: public:
LoDTensor() : Tensor() {} LoDTensor() : Tensor() {}
explicit LoDTensor(const LoD &lod) : lod_(lod) {} explicit LoDTensor(const LoD &lod) : lod_(lod) {}
void set_lod(const LoD &lod) { lod_ = lod; } void set_lod(const LoD &lod) { lod_ = lod; }
const LoD &lod() const { return lod_; } const LoD &lod() const { return lod_; }
LoD *mutable_lod() { return &lod_; } LoD *mutable_lod() { return &lod_; }
/* /*
* Get the start offset and end offset of an element from LoD. * Get the start offset and end offset of an element from LoD.
*/ */
std::pair<size_t, size_t> lod_element(size_t level, size_t elem) const { std::pair<size_t, size_t> lod_element(size_t level, size_t elem) const {
// PADDLE_ENFORCE_LT(level, NumLevels()); // PADDLE_ENFORCE_LT(level, NumLevels());
// PADDLE_ENFORCE_LT(elem, NumElements(level)); // PADDLE_ENFORCE_LT(elem, NumElements(level));
return std::make_pair((lod_)[level][elem], (lod_)[level][elem + 1]); return std::make_pair((lod_)[level][elem], (lod_)[level][elem + 1]);
} }
/* /*
* Number of LoDTensor's levels, each level has units of data, for * Number of LoDTensor's levels, each level has units of data, for
* example, * example,
* in the sentence's view, article, paragraph, sentence are 3 * in the sentence's view, article, paragraph, sentence are 3
* levels. * levels.
*/ */
size_t NumLevels() const { return lod_.size(); } size_t NumLevels() const { return lod_.size(); }
/* /*
* Number of elements in a level. * Number of elements in a level.
*/ */
size_t NumElements(size_t level = 0) const { size_t NumElements(size_t level = 0) const {
// PADDLE_ENFORCE_LT(level, NumLevels()); // PADDLE_ENFORCE_LT(level, NumLevels());
// the last offset is the end of last element // the last offset is the end of last element
return (lod_)[level].size() - 1; return (lod_)[level].size() - 1;
} }
private: private:
LoD lod_; LoD lod_;
}; };
/* /*
...@@ -155,26 +155,26 @@ class LoDTensor : public Tensor { ...@@ -155,26 +155,26 @@ class LoDTensor : public Tensor {
*/ */
template <typename T> template <typename T>
LoDTensor LodExpand(const LoDTensor &source, const LoD &lod, size_t level) { LoDTensor LodExpand(const LoDTensor &source, const LoD &lod, size_t level) {
LoD abs_lod = ToAbsOffset(lod); LoD abs_lod = ToAbsOffset(lod);
const auto &lod_level = lod[level]; const auto &lod_level = lod[level];
size_t num_instances = source.dims()[0]; size_t num_instances = source.dims()[0];
// new tensor // new tensor
LoDTensor tensor; LoDTensor tensor;
tensor.set_lod(lod); tensor.set_lod(lod);
auto dims = source.dims(); auto dims = source.dims();
dims[0] = lod_level.back(); dims[0] = lod_level.back();
tensor.Resize(dims); tensor.Resize(dims);
tensor.mutable_data<T>(); tensor.mutable_data<T>();
// PADDLE_ENFORCE_EQ(num_instances, lod_level.size() - 1); // PADDLE_ENFORCE_EQ(num_instances, lod_level.size() - 1);
for (size_t ins = 0; ins < num_instances; ins++) { for (size_t ins = 0; ins < num_instances; ins++) {
for (size_t elem = lod_level[ins]; elem < lod_level[ins + 1]; elem++) { for (size_t elem = lod_level[ins]; elem < lod_level[ins + 1]; elem++) {
auto slice = tensor.Slice(elem, elem + 1); auto slice = tensor.Slice(elem, elem + 1);
TensorCopy(source.Slice(ins, ins + 1), &slice); TensorCopy(source.Slice(ins, ins + 1), &slice);
}
} }
return tensor; }
return tensor;
} }
// Get the absolute offset of a lod[start_level][start_idx:end_idx] and // Get the absolute offset of a lod[start_level][start_idx:end_idx] and
......
src/framework/op_desc.cpp
浏览文件 @ dff4a781
...@@ -8,51 +8,51 @@ namespace paddle_mobile { ...@@ -8,51 +8,51 @@ namespace paddle_mobile {
namespace framework { namespace framework {
OpDesc::OpDesc(const proto::OpDesc &desc) : desc_(desc) { OpDesc::OpDesc(const proto::OpDesc &desc) : desc_(desc) {
for (int i = 0; i < desc_.inputs_size(); ++i) { for (int i = 0; i < desc_.inputs_size(); ++i) {
const proto::OpDesc::Var &var = desc_.inputs(i); const proto::OpDesc::Var &var = desc_.inputs(i);
std::vector<std::string> &args = inputs_[var.parameter()]; std::vector<std::string> &args = inputs_[var.parameter()];
int arg_size = var.arguments_size(); int arg_size = var.arguments_size();
for (int j = 0; j < arg_size; ++j) { for (int j = 0; j < arg_size; ++j) {
args.push_back(var.arguments(j)); args.push_back(var.arguments(j));
}
} }
}
for (int i = 0; i < desc_.outputs_size(); ++i) {
const proto::OpDesc::Var &var = desc_.outputs(i); for (int i = 0; i < desc_.outputs_size(); ++i) {
std::vector<std::string> &args = outputs_[var.parameter()]; const proto::OpDesc::Var &var = desc_.outputs(i);
int arg_size = var.arguments_size(); std::vector<std::string> &args = outputs_[var.parameter()];
for (int j = 0; j < arg_size; ++j) { int arg_size = var.arguments_size();
args.push_back(var.arguments(j)); for (int j = 0; j < arg_size; ++j) {
} args.push_back(var.arguments(j));
} }
}
for (const proto::OpDesc::Attr &attr : desc_.attrs()) {
std::string attr_name = attr.name(); for (const proto::OpDesc::Attr &attr : desc_.attrs()) {
if (attr.type() != proto::AttrType::BLOCK) { std::string attr_name = attr.name();
attrs_[attr_name] = Attribute::GetAttrValue(attr); if (attr.type() != proto::AttrType::BLOCK) {
// if (attr.type() == proto::AttrType::INT){ attrs_[attr_name] = Attribute::GetAttrValue(attr);
// std::cout << " attrName " << attr_name << " " << // if (attr.type() == proto::AttrType::INT){
// attrs_[attr_name].Get<int>() << std::endl; // std::cout << " attrName " << attr_name << " " <<
// } // attrs_[attr_name].Get<int>() << std::endl;
} // }
} }
}
} }
const std::vector<std::string> &OpDesc::Input(const std::string &name) const { const std::vector<std::string> &OpDesc::Input(const std::string &name) const {
return inputs_.find(name)->second; return inputs_.find(name)->second;
} }
const std::vector<std::string> &OpDesc::Output(const std::string &name) const { const std::vector<std::string> &OpDesc::Output(const std::string &name) const {
return outputs_.find(name)->second; return outputs_.find(name)->second;
} }
Attribute OpDesc::GetAttr(const std::string &name) const { Attribute OpDesc::GetAttr(const std::string &name) const {
auto it = attrs_.find(name); auto it = attrs_.find(name);
return it->second; return it->second;
} }
const std::unordered_map<std::string, Attribute> &OpDesc::GetAttrMap() const { const std::unordered_map<std::string, Attribute> &OpDesc::GetAttrMap() const {
return attrs_; return attrs_;
} }
} // namespace framework } // namespace framework
......
src/framework/op_desc.h
浏览文件 @ dff4a781
...@@ -26,25 +26,25 @@ namespace paddle_mobile { ...@@ -26,25 +26,25 @@ namespace paddle_mobile {
namespace framework { namespace framework {
class OpDesc : PaddleMobileObject { class OpDesc : PaddleMobileObject {
public: public:
OpDesc(const proto::OpDesc &desc); OpDesc(const proto::OpDesc &desc);
const std::vector<std::string> &Input(const std::string &name) const; const std::vector<std::string> &Input(const std::string &name) const;
const std::vector<std::string> &Output(const std::string &name) const; const std::vector<std::string> &Output(const std::string &name) const;
Attribute GetAttr(const std::string &name) const; Attribute GetAttr(const std::string &name) const;
const VariableNameMap &GetInputs() { return inputs_; } const VariableNameMap &GetInputs() { return inputs_; }
const VariableNameMap &GetOutputs() { return outputs_; } const VariableNameMap &GetOutputs() { return outputs_; }
const AttributeMap &GetAttrMap() const; const AttributeMap &GetAttrMap() const;
const std::string &Type() { return desc_.type(); }; const std::string &Type() { return desc_.type(); };
private: private:
proto::OpDesc desc_; proto::OpDesc desc_;
VariableNameMap inputs_; VariableNameMap inputs_;
VariableNameMap outputs_; VariableNameMap outputs_;
AttributeMap attrs_; AttributeMap attrs_;
}; };
} // namespace framework } // namespace framework
......
src/framework/op_info.h
浏览文件 @ dff4a781
...@@ -25,13 +25,13 @@ namespace paddle_mobile { ...@@ -25,13 +25,13 @@ namespace paddle_mobile {
namespace framework { namespace framework {
template <typename Dtype> struct OpInfo { template <typename Dtype> struct OpInfo {
OpCreator<Dtype> creator_; OpCreator<Dtype> creator_;
const OpCreator<Dtype> &Creator() const { const OpCreator<Dtype> &Creator() const {
// PADDLE_ENFORCE_NOT_NULL(creator_, // PADDLE_ENFORCE_NOT_NULL(creator_,
// "Operator Creator has not been // "Operator Creator has not been
// registered"); // registered");
return creator_; return creator_;
} }
}; };
template <typename Dtype> class OpInfoMap; template <typename Dtype> class OpInfoMap;
...@@ -39,55 +39,55 @@ template <typename Dtype> class OpInfoMap; ...@@ -39,55 +39,55 @@ template <typename Dtype> class OpInfoMap;
template <typename Dtype> static OpInfoMap<Dtype> *g_op_info_map = nullptr; template <typename Dtype> static OpInfoMap<Dtype> *g_op_info_map = nullptr;
template <typename Dtype> class OpInfoMap { template <typename Dtype> class OpInfoMap {
public: public:
static OpInfoMap &Instance() { static OpInfoMap &Instance() {
if (g_op_info_map<Dtype> == nullptr) { if (g_op_info_map<Dtype> == nullptr) {
g_op_info_map<Dtype> = new OpInfoMap(); g_op_info_map<Dtype> = new OpInfoMap();
}
return *g_op_info_map<Dtype>;
};
bool Has(const std::string &op_type) const {
return map_.find(op_type) != map_.end();
}
void Insert(const std::string &type, const OpInfo<Dtype> &info) {
// PADDLE_ENFORCE(!Has(type), "Operator %s has been
// registered", type);
map_.insert({type, info});
}
const OpInfo<Dtype> &Get(const std::string &type) const {
auto op_info_ptr = GetNullable(type);
// PADDLE_ENFORCE_NOT_NULL(op_info_ptr, "Operator %s has not
// been
// registered",
// type);
return *op_info_ptr;
} }
return *g_op_info_map<Dtype>;
const OpInfo<Dtype> *GetNullable(const std::string &type) const { };
auto it = map_.find(type);
if (it == map_.end()) { bool Has(const std::string &op_type) const {
return nullptr; return map_.find(op_type) != map_.end();
} else { }
return &it->second;
} void Insert(const std::string &type, const OpInfo<Dtype> &info) {
// PADDLE_ENFORCE(!Has(type), "Operator %s has been
// registered", type);
map_.insert({type, info});
}
const OpInfo<Dtype> &Get(const std::string &type) const {
auto op_info_ptr = GetNullable(type);
// PADDLE_ENFORCE_NOT_NULL(op_info_ptr, "Operator %s has not
// been
// registered",
// type);
return *op_info_ptr;
}
const OpInfo<Dtype> *GetNullable(const std::string &type) const {
auto it = map_.find(type);
if (it == map_.end()) {
return nullptr;
} else {
return &it->second;
} }
}
const std::unordered_map<std::string, OpInfo<Dtype>> &map() const { const std::unordered_map<std::string, OpInfo<Dtype>> &map() const {
return map_; return map_;
} }
std::unordered_map<std::string, OpInfo<Dtype>> *mutable_map() { std::unordered_map<std::string, OpInfo<Dtype>> *mutable_map() {
return &map_; return &map_;
} }
private: private:
OpInfoMap() = default; OpInfoMap() = default;
std::unordered_map<std::string, OpInfo<Dtype>> map_; std::unordered_map<std::string, OpInfo<Dtype>> map_;
// DISABLE_COPY_AND_ASSIGN(OpInfoMap); // DISABLE_COPY_AND_ASSIGN(OpInfoMap);
}; };
} // namespace framework } // namespace framework
......
src/framework/op_kernel_type.h
浏览文件 @ dff4a781
...@@ -24,41 +24,40 @@ SOFTWARE. ...@@ -24,41 +24,40 @@ SOFTWARE.
namespace paddle_mobile { namespace paddle_mobile {
namespace framework { namespace framework {
struct OpKernelType { struct OpKernelType {
struct Hash { struct Hash {
size_t operator()(const OpKernelType &key) const { size_t operator()(const OpKernelType &key) const {
int data_type = static_cast<int>(key.data_type_) << LEFT_SHIFT; int data_type = static_cast<int>(key.data_type_) << LEFT_SHIFT;
int data_layout = static_cast<int>(key.data_layout_) int data_layout = static_cast<int>(key.data_layout_) << (LEFT_SHIFT * 2);
<< (LEFT_SHIFT * 2);
std::hash<int> hasher; std::hash<int> hasher;
return hasher(data_type + data_layout); return hasher(data_type + data_layout);
} }
}; };
// place, data_type, library_type kinds less than 2^8 // place, data_type, library_type kinds less than 2^8
constexpr static int LEFT_SHIFT = 8; constexpr static int LEFT_SHIFT = 8;
proto::VarType::Type data_type_; proto::VarType::Type data_type_;
DataLayout data_layout_; DataLayout data_layout_;
OpKernelType(proto::VarType::Type data_type, OpKernelType(proto::VarType::Type data_type,
DataLayout data_layout = DataLayout::kAnyLayout) DataLayout data_layout = DataLayout::kAnyLayout)
: data_type_(data_type), data_layout_(data_layout) {} : data_type_(data_type), data_layout_(data_layout) {}
bool operator==(const OpKernelType &o) const { bool operator==(const OpKernelType &o) const {
return data_type_ == o.data_type_ && data_layout_ == o.data_layout_; return data_type_ == o.data_type_ && data_layout_ == o.data_layout_;
} }
bool operator!=(const OpKernelType &o) const { return !(*this == o); } bool operator!=(const OpKernelType &o) const { return !(*this == o); }
}; };
inline bool NeedTransformLayout(const DataLayout &l, const DataLayout &r) { inline bool NeedTransformLayout(const DataLayout &l, const DataLayout &r) {
return l != DataLayout::kAnyLayout && r != DataLayout::kAnyLayout && l != r; return l != DataLayout::kAnyLayout && r != DataLayout::kAnyLayout && l != r;
} }
inline bool TransFromNeeded(const OpKernelType &l, const OpKernelType &r) { inline bool TransFromNeeded(const OpKernelType &l, const OpKernelType &r) {
return (l.data_type_ != r.data_type_) || return (l.data_type_ != r.data_type_) ||
NeedTransformLayout(l.data_layout_, r.data_layout_); NeedTransformLayout(l.data_layout_, r.data_layout_);
} }
} // namespace framework } // namespace framework
......
src/framework/operator.cpp
浏览文件 @ dff4a781
...@@ -30,7 +30,7 @@ OperatorBase<Dtype>::OperatorBase(const std::string &type, ...@@ -30,7 +30,7 @@ OperatorBase<Dtype>::OperatorBase(const std::string &type,
std::shared_ptr<Scope> scope) std::shared_ptr<Scope> scope)
: type_(type), inputs_(inputs), outputs_(outputs), attrs_(attrs), : type_(type), inputs_(inputs), outputs_(outputs), attrs_(attrs),
scope_(scope) { scope_(scope) {
CheckAllInputOutputSet(); CheckAllInputOutputSet();
} }
template <typename Dtype> template <typename Dtype>
void OperatorBase<Dtype>::CheckAllInputOutputSet() const {} void OperatorBase<Dtype>::CheckAllInputOutputSet() const {}
......
src/framework/operator.h
浏览文件 @ dff4a781
...@@ -49,50 +49,50 @@ static std::unordered_map< ...@@ -49,50 +49,50 @@ static std::unordered_map<
{"fetch", {{"X"}, {"Out"}}}}; {"fetch", {{"X"}, {"Out"}}}};
template <typename Dtype> class OperatorBase : PaddleMobileObject { template <typename Dtype> class OperatorBase : PaddleMobileObject {
public: public:
OperatorBase(const std::string &type, const VariableNameMap &inputs, OperatorBase(const std::string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs, const AttributeMap &attrs, const VariableNameMap &outputs, const AttributeMap &attrs,
std::shared_ptr<Scope> scope); std::shared_ptr<Scope> scope);
virtual ~OperatorBase() {} virtual ~OperatorBase() {}
virtual void Run() const = 0; virtual void Run() const = 0;
const VariableNameMap &Inputs() const { return inputs_; } const VariableNameMap &Inputs() const { return inputs_; }
const VariableNameMap &Outputs() const { return outputs_; } const VariableNameMap &Outputs() const { return outputs_; }
const std::string &Type() const { return type_; } const std::string &Type() const { return type_; }
const AttributeMap &Attrs() const { return attrs_; } const AttributeMap &Attrs() const { return attrs_; }
void ClearVariables(const std::vector<std::string> &var_names) const { void ClearVariables(const std::vector<std::string> &var_names) const {
if (this->scope_) { if (this->scope_) {
this->scope_->EraseVars(var_names); this->scope_->EraseVars(var_names);
}
} }
}
protected: protected:
std::shared_ptr<Scope> scope_; std::shared_ptr<Scope> scope_;
std::string type_; std::string type_;
VariableNameMap inputs_; VariableNameMap inputs_;
VariableNameMap outputs_; VariableNameMap outputs_;
AttributeMap attrs_; AttributeMap attrs_;
private: private:
void CheckAllInputOutputSet() const; void CheckAllInputOutputSet() const;
}; };
template <typename Dtype> template <typename Dtype>
class OperatorWithKernel : public OperatorBase<Dtype> { class OperatorWithKernel : public OperatorBase<Dtype> {
public: public:
OperatorWithKernel(const std::string &type, const VariableNameMap &inputs, OperatorWithKernel(const std::string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs, const VariableNameMap &outputs, const AttributeMap &attrs,
const AttributeMap &attrs, std::shared_ptr<Scope> scope) std::shared_ptr<Scope> scope)
: OperatorBase<Dtype>(type, inputs, outputs, attrs, scope) {} : OperatorBase<Dtype>(type, inputs, outputs, attrs, scope) {}
virtual void InferShape() const = 0; virtual void InferShape() const = 0;
virtual void Run() const = 0; virtual void Run() const = 0;
}; };
template <typename Dtype, typename P> class OpKernelBase : PaddleMobileObject { template <typename Dtype, typename P> class OpKernelBase : PaddleMobileObject {
public: public:
virtual void Compute(const P &para) const = 0; virtual void Compute(const P &para) const = 0;
virtual ~OpKernelBase() = default; virtual ~OpKernelBase() = default;
}; };
} // namespace framework } // namespace framework
......
src/framework/paddle_mobile_object.h
浏览文件 @ dff4a781
...@@ -24,13 +24,13 @@ SOFTWARE. ...@@ -24,13 +24,13 @@ SOFTWARE.
namespace paddle_mobile { namespace paddle_mobile {
class PaddleMobileObject { class PaddleMobileObject {
public: public:
virtual std::string ToString() { virtual std::string ToString() {
char address[128] = {0}; char address[128] = {0};
sprintf(address, "%p", this); sprintf(address, "%p", this);
return std::string(address); return std::string(address);
} }
private: private:
}; };
} // namespace paddle_mobile } // namespace paddle_mobile
src/framework/program-optimize/node.cpp
浏览文件 @ dff4a781
...@@ -25,71 +25,71 @@ namespace paddle_mobile { ...@@ -25,71 +25,71 @@ namespace paddle_mobile {
namespace framework { namespace framework {
Node &Node::operator>(std::shared_ptr<Node> node) { Node &Node::operator>(std::shared_ptr<Node> node) {
outputs_.push_back(node); outputs_.push_back(node);
std::shared_ptr<Node> this_node; std::shared_ptr<Node> this_node;
node->inputs_.push_back(this); node->inputs_.push_back(this);
return *node; return *node;
} }
bool Node::operator==(const Node &in) { bool Node::operator==(const Node &in) {
if (in.type_ == this->type_) { if (in.type_ == this->type_) {
if (this->outputs_.size() == in.outputs_.size()) { if (this->outputs_.size() == in.outputs_.size()) {
for (int i = 0; i < outputs_.size(); ++i) { for (int i = 0; i < outputs_.size(); ++i) {
if (!(*outputs_[i] == *in.outputs_[i])) { if (!(*outputs_[i] == *in.outputs_[i])) {
return false; return false;
}
}
} else {
return false;
} }
}
} else { } else {
return false; return false;
} }
return true; } else {
return false;
}
return true;
} }
std::string Node::ToString(std::string blank, const Node *node) const { std::string Node::ToString(std::string blank, const Node *node) const {
std::stringstream ss; std::stringstream ss;
ss << type_ << "-> \n"; ss << type_ << "-> \n";
if (inputs_.size() > 1 && node != inputs_.back()) { if (inputs_.size() > 1 && node != inputs_.back()) {
return ss.str();
} else if (inputs_.size() > 1 && node == inputs_.back()) {
ss << "\n" << blank << type_ << "\n";
}
for (int i = 0; i < outputs_.size(); ++i) {
ss << blank << outputs_[i]->ToString(blank + " ", this) << "";
}
return ss.str(); return ss.str();
} else if (inputs_.size() > 1 && node == inputs_.back()) {
ss << "\n" << blank << type_ << "\n";
}
for (int i = 0; i < outputs_.size(); ++i) {
ss << blank << outputs_[i]->ToString(blank + " ", this) << "";
}
return ss.str();
} }
std::string Node::ToString() const { return this->ToString(" ", this); } std::string Node::ToString() const { return this->ToString(" ", this); }
Node &Node::To(int index) { Node &Node::To(int index) {
if (index == 0) { if (index == 0) {
this->outputs_.clear(); this->outputs_.clear();
} }
for (int j = 0; j < this->outputs_.size(); ++j) { for (int j = 0; j < this->outputs_.size(); ++j) {
outputs_[j]->To(index - 1); outputs_[j]->To(index - 1);
} }
return *this; return *this;
} }
uint Node::depth(uint begin) { uint Node::depth(uint begin) {
uint depth = 0; uint depth = 0;
begin++; begin++;
for (int i = 0; i < outputs_.size(); ++i) { for (int i = 0; i < outputs_.size(); ++i) {
uint output_depth = outputs_[i]->depth(begin); uint output_depth = outputs_[i]->depth(begin);
depth = output_depth > depth ? output_depth : depth; depth = output_depth > depth ? output_depth : depth;
} }
return begin > depth ? begin : depth; return begin > depth ? begin : depth;
} }
Print &operator<<(Print &printer, const Node &node) { Print &operator<<(Print &printer, const Node &node) {
printer << node.ToString(); printer << node.ToString();
return printer; return printer;
} }
} // namespace framework } // namespace framework
......
src/framework/program-optimize/node.h
浏览文件 @ dff4a781
...@@ -29,22 +29,22 @@ namespace paddle_mobile { ...@@ -29,22 +29,22 @@ namespace paddle_mobile {
namespace framework { namespace framework {
class Node : PaddleMobileObject { class Node : PaddleMobileObject {
public: public:
Node(const std::string &type) : type_(type) {} Node(const std::string &type) : type_(type) {}
Node(std::shared_ptr<OpDesc> op_desc) Node(std::shared_ptr<OpDesc> op_desc)
: op_desc_(op_desc), type_(op_desc->Type()){}; : op_desc_(op_desc), type_(op_desc->Type()){};
Node &operator>(std::shared_ptr<Node> node); Node &operator>(std::shared_ptr<Node> node);
bool operator==(const Node &in); bool operator==(const Node &in);
std::string ToString() const; std::string ToString() const;
Node &To(int index); Node &To(int index);
uint depth(uint begin = 0); uint depth(uint begin = 0);
private: private:
std::shared_ptr<OpDesc> op_desc_; std::shared_ptr<OpDesc> op_desc_;
std::string ToString(std::string blank, const Node *node) const; std::string ToString(std::string blank, const Node *node) const;
std::vector<std::shared_ptr<Node>> outputs_; std::vector<std::shared_ptr<Node>> outputs_;
std::vector<Node *> inputs_; std::vector<Node *> inputs_;
std::string type_; std::string type_;
}; };
Print &operator<<(Print &printer, const Node &node); Print &operator<<(Print &printer, const Node &node);
......
src/framework/program-optimize/program_optimize.cpp
浏览文件 @ dff4a781
...@@ -26,49 +26,48 @@ std::shared_ptr<ProgramDesc> ProgramOptimize::Optimize() {} ...@@ -26,49 +26,48 @@ std::shared_ptr<ProgramDesc> ProgramOptimize::Optimize() {}
std::shared_ptr<ProgramDesc> std::shared_ptr<ProgramDesc>
ProgramOptimize::FushionOptimize(std::shared_ptr<ProgramDesc> ori_des) { ProgramOptimize::FushionOptimize(std::shared_ptr<ProgramDesc> ori_des) {
for (int i = 0; i < ori_des->Blocks().size(); ++i) { for (int i = 0; i < ori_des->Blocks().size(); ++i) {
std::unordered_map<std::string, std::shared_ptr<Node>> output_nodes; std::unordered_map<std::string, std::shared_ptr<Node>> output_nodes;
std::shared_ptr<Node> begin_node; std::shared_ptr<Node> begin_node;
auto block = ori_des->Block(i); auto block = ori_des->Block(i);
// DLOG << " ops size: " << block->Ops().size(); // DLOG << " ops size: " << block->Ops().size();
for (int j = 0; j < block->Ops().size(); ++j) { for (int j = 0; j < block->Ops().size(); ++j) {
auto op = block->Ops()[j]; auto op = block->Ops()[j];
auto op_type = op->Type(); auto op_type = op->Type();
// DLOG << "op type: " << op_type << " index: " << j; // DLOG << "op type: " << op_type << " index: " << j;
if (op_input_output_key.find(op->Type()) == if (op_input_output_key.find(op->Type()) == op_input_output_key.end()) {
op_input_output_key.end()) { return NULL;
return NULL; }
}
std::shared_ptr<Node> node = std::make_shared<Node>(op); std::shared_ptr<Node> node = std::make_shared<Node>(op);
if (j == 0) { if (j == 0) {
begin_node = node; begin_node = node;
} }
auto input_keys = op_input_output_key.at(op->Type()).first; auto input_keys = op_input_output_key.at(op->Type()).first;
for (auto input_key : input_keys) { for (auto input_key : input_keys) {
auto op_inputs = op->Input(input_key); auto op_inputs = op->Input(input_key);
for (int l = 0; l < op_inputs.size(); ++l) { for (int l = 0; l < op_inputs.size(); ++l) {
std::string input_key = op_inputs[l]; std::string input_key = op_inputs[l];
if (output_nodes.find(input_key) != output_nodes.end()) { if (output_nodes.find(input_key) != output_nodes.end()) {
auto input_node = output_nodes[input_key]; auto input_node = output_nodes[input_key];
*input_node > node; *input_node > node;
} }
}
}
auto output_keys = op_input_output_key.at(op_type).second;
for (auto output_key : output_keys) {
auto op_outputs = op->Output(output_key);
for (int k = 0; k < op_outputs.size(); ++k) {
output_nodes[op_outputs[k]] = node;
}
}
} }
}
DLOG << "node: \n" << *begin_node; auto output_keys = op_input_output_key.at(op_type).second;
for (auto output_key : output_keys) {
auto op_outputs = op->Output(output_key);
for (int k = 0; k < op_outputs.size(); ++k) {
output_nodes[op_outputs[k]] = node;
}
}
} }
return ori_des;
DLOG << "node: \n" << *begin_node;
}
return ori_des;
} }
} // namespace framework } // namespace framework
} // namespace paddle_mobile } // namespace paddle_mobile
src/framework/program-optimize/program_optimize.h
浏览文件 @ dff4a781
...@@ -26,16 +26,16 @@ namespace paddle_mobile { ...@@ -26,16 +26,16 @@ namespace paddle_mobile {
namespace framework { namespace framework {
class ProgramOptimize { class ProgramOptimize {
public: public:
ProgramOptimize() {} ProgramOptimize() {}
std::shared_ptr<ProgramDesc> Optimize(); std::shared_ptr<ProgramDesc> Optimize();
std::shared_ptr<ProgramDesc> std::shared_ptr<ProgramDesc>
FushionOptimize(std::shared_ptr<ProgramDesc> ori_des); FushionOptimize(std::shared_ptr<ProgramDesc> ori_des);
private: private:
// std::shared_ptr<ProgramDesc> ori_desc_; // std::shared_ptr<ProgramDesc> ori_desc_;
std::vector<std::unordered_map<std::string, std::shared_ptr<Node>>> std::vector<std::unordered_map<std::string, std::shared_ptr<Node>>>
outputs_nodes_; outputs_nodes_;
}; };
} // namespace framework } // namespace framework
} // namespace paddle_mobile } // namespace paddle_mobile
src/framework/program.h
浏览文件 @ dff4a781
...@@ -28,12 +28,12 @@ namespace framework { ...@@ -28,12 +28,12 @@ namespace framework {
template <typename Dtype, Precision P = Precision::FP32> template <typename Dtype, Precision P = Precision::FP32>
class Program : PaddleMobileObject { class Program : PaddleMobileObject {
public: public:
std::shared_ptr<ProgramDesc> originProgram; std::shared_ptr<ProgramDesc> originProgram;
std::shared_ptr<ProgramDesc> optimizeProgram; std::shared_ptr<ProgramDesc> optimizeProgram;
std::shared_ptr<Scope> scope; std::shared_ptr<Scope> scope;
private: private:
}; };
} // namespace framework } // namespace framework
......
src/framework/program_desc.cpp
浏览文件 @ dff4a781
...@@ -8,14 +8,14 @@ namespace paddle_mobile { ...@@ -8,14 +8,14 @@ namespace paddle_mobile {
namespace framework { namespace framework {
ProgramDesc::ProgramDesc(const proto::ProgramDesc &desc) : desc_(desc) { ProgramDesc::ProgramDesc(const proto::ProgramDesc &desc) : desc_(desc) {
for (auto &block_desc : *desc_.mutable_blocks()) { for (auto &block_desc : *desc_.mutable_blocks()) {
// new framework::BlockDesc(block_desc) // new framework::BlockDesc(block_desc)
blocks_.emplace_back(std::make_shared<BlockDesc>(block_desc)); blocks_.emplace_back(std::make_shared<BlockDesc>(block_desc));
} }
} }
std::shared_ptr<BlockDesc> ProgramDesc::Block(size_t idx) { std::shared_ptr<BlockDesc> ProgramDesc::Block(size_t idx) {
return blocks_[idx]; return blocks_[idx];
} }
} // namespace framework } // namespace framework
......
src/framework/program_desc.h
浏览文件 @ dff4a781
...@@ -28,14 +28,14 @@ namespace paddle_mobile { ...@@ -28,14 +28,14 @@ namespace paddle_mobile {
namespace framework { namespace framework {
class ProgramDesc : PaddleMobileObject { class ProgramDesc : PaddleMobileObject {
public: public:
ProgramDesc(const proto::ProgramDesc &desc); ProgramDesc(const proto::ProgramDesc &desc);
std::shared_ptr<BlockDesc> Block(size_t idx); std::shared_ptr<BlockDesc> Block(size_t idx);
const std::vector<std::shared_ptr<BlockDesc>> &Blocks() { return blocks_; }; const std::vector<std::shared_ptr<BlockDesc>> &Blocks() { return blocks_; };
private: private:
std::vector<std::shared_ptr<BlockDesc>> blocks_; std::vector<std::shared_ptr<BlockDesc>> blocks_;
proto::ProgramDesc desc_; proto::ProgramDesc desc_;
}; };
} // namespace framework } // namespace framework
......
src/framework/scope.cc
浏览文件 @ dff4a781
...@@ -7,20 +7,20 @@ namespace paddle_mobile { ...@@ -7,20 +7,20 @@ namespace paddle_mobile {
namespace framework { namespace framework {
Scope &Scope::NewScope() const { Scope &Scope::NewScope() const {
std::unique_lock<std::mutex> lock(mutex_); std::unique_lock<std::mutex> lock(mutex_);
kids_.push_back(new Scope(this)); kids_.push_back(new Scope(this));
return *kids_.back(); return *kids_.back();
} }
Variable *Scope::Var(const std::string &name) { Variable *Scope::Var(const std::string &name) {
auto *pvar = FindVarLocally(name); auto *pvar = FindVarLocally(name);
if (pvar != nullptr) { if (pvar != nullptr) {
return pvar;
};
pvar = new Variable;
vars_[name] = pvar;
pvar->name_ = &(vars_.find(name)->first);
return pvar; return pvar;
};
pvar = new Variable;
vars_[name] = pvar;
pvar->name_ = &(vars_.find(name)->first);
return pvar;
} }
// Variable* Scope::Var(std::string* name) { // Variable* Scope::Var(std::string* name) {
...@@ -33,70 +33,70 @@ Variable *Scope::Var(const std::string &name) { ...@@ -33,70 +33,70 @@ Variable *Scope::Var(const std::string &name) {
// } // }
Variable *Scope::FindVar(const std::string &name) const { Variable *Scope::FindVar(const std::string &name) const {
auto *pvar = FindVarLocally(name); auto *pvar = FindVarLocally(name);
if (pvar != nullptr) { if (pvar != nullptr) {
return pvar; return pvar;
} }
return (parent_ == nullptr) ? nullptr : parent_->FindVar(name); return (parent_ == nullptr) ? nullptr : parent_->FindVar(name);
} }
const Scope *Scope::FindScope(const Variable *var) const { const Scope *Scope::FindScope(const Variable *var) const {
for (auto &name_var : vars_) { for (auto &name_var : vars_) {
if (name_var.second == var) { if (name_var.second == var) {
return this; return this;
}
} }
return (parent_ == nullptr) ? nullptr : parent_->FindScope(var); }
return (parent_ == nullptr) ? nullptr : parent_->FindScope(var);
} }
void Scope::DropKids() { void Scope::DropKids() {
for (Scope *s : kids_) { for (Scope *s : kids_) {
delete s; delete s;
} }
kids_.clear(); kids_.clear();
} }
std::vector<std::string> Scope::LocalVarNames() const { std::vector<std::string> Scope::LocalVarNames() const {
std::vector<std::string> known_vars; std::vector<std::string> known_vars;
known_vars.reserve(vars_.size()); known_vars.reserve(vars_.size());
for (auto &name_var : vars_) { for (auto &name_var : vars_) {
known_vars.emplace_back(name_var.first); known_vars.emplace_back(name_var.first);
} }
return known_vars; return known_vars;
} }
void Scope::DeleteScope(Scope *scope) const { void Scope::DeleteScope(Scope *scope) const {
std::unique_lock<std::mutex> lock(mutex_); std::unique_lock<std::mutex> lock(mutex_);
auto it = std::find(kids_.begin(), kids_.end(), scope); auto it = std::find(kids_.begin(), kids_.end(), scope);
kids_.erase(it); kids_.erase(it);
delete scope; delete scope;
// deferent // deferent
} }
void Scope::EraseVars(const std::vector<std::string> &var_names) { void Scope::EraseVars(const std::vector<std::string> &var_names) {
std::set<std::string> var_set(var_names.begin(), var_names.end()); std::set<std::string> var_set(var_names.begin(), var_names.end());
for (auto it = vars_.begin(); it != vars_.end();) { for (auto it = vars_.begin(); it != vars_.end();) {
if (var_set.find(it->first) != var_set.end()) { if (var_set.find(it->first) != var_set.end()) {
delete it->second; delete it->second;
it = vars_.erase(it); it = vars_.erase(it);
} else { } else {
++it; ++it;
}
} }
}
} }
void Scope::Rename(const std::string &origin_name, void Scope::Rename(const std::string &origin_name,
const std::string &new_name) const { const std::string &new_name) const {
auto origin_it = vars_.find(origin_name); auto origin_it = vars_.find(origin_name);
if (origin_it == vars_.end()) { if (origin_it == vars_.end()) {
return; return;
} }
auto new_it = vars_.find(new_name); auto new_it = vars_.find(new_name);
if (new_it != vars_.end()) { if (new_it != vars_.end()) {
return; return;
} }
vars_[new_name] = origin_it->second; vars_[new_name] = origin_it->second;
vars_.erase(origin_it); vars_.erase(origin_it);
} }
// //
// std::string Scope::Rename(const std::string& origin_name) // std::string Scope::Rename(const std::string& origin_name)
...@@ -108,11 +108,11 @@ void Scope::Rename(const std::string &origin_name, ...@@ -108,11 +108,11 @@ void Scope::Rename(const std::string &origin_name,
// } // }
Variable *Scope::FindVarLocally(const std::string &name) const { Variable *Scope::FindVarLocally(const std::string &name) const {
auto it = vars_.find(name); auto it = vars_.find(name);
if (it != vars_.end()) { if (it != vars_.end()) {
return it->second; return it->second;
} }
return nullptr; return nullptr;
} }
} // namespace framework } // namespace framework
......
src/framework/scope.h
浏览文件 @ dff4a781
...@@ -26,56 +26,56 @@ SOFTWARE. ...@@ -26,56 +26,56 @@ SOFTWARE.
namespace paddle_mobile { namespace paddle_mobile {
namespace framework { namespace framework {
class Scope { class Scope {
public: public:
Scope() {} Scope() {}
~Scope() {} ~Scope() {}
Scope &NewScope() const; Scope &NewScope() const;
/// Create a variable with given name if it doesn't exist. /// Create a variable with given name if it doesn't exist.
Variable *Var(const std::string &name); Variable *Var(const std::string &name);
/// Create a variable with a scope-unique name. /// Create a variable with a scope-unique name.
Variable *Var(std::string *name = nullptr); Variable *Var(std::string *name = nullptr);
void EraseVars(const std::vector<std::string> &var_names); void EraseVars(const std::vector<std::string> &var_names);
/// Find a variable in the scope or any of its ancestors. Returns /// Find a variable in the scope or any of its ancestors. Returns
/// nullptr if cannot find. /// nullptr if cannot find.
Variable *FindVar(const std::string &name) const; Variable *FindVar(const std::string &name) const;
const Scope *parent() const { return parent_; } const Scope *parent() const { return parent_; }
/// Find the scope or an ancestor scope that contains the given /// Find the scope or an ancestor scope that contains the given
/// variable. /// variable.
const Scope *FindScope(const Variable *var) const; const Scope *FindScope(const Variable *var) const;
void DeleteScope(Scope *scope) const; void DeleteScope(Scope *scope) const;
/// Drop all kids scopes belonged to this scope. /// Drop all kids scopes belonged to this scope.
void DropKids(); void DropKids();
// enumerate all the variables current contains. // enumerate all the variables current contains.
std::vector<std::string> LocalVarNames() const; std::vector<std::string> LocalVarNames() const;
// Rename variable to a new name // Rename variable to a new name
void Rename(const std::string &origin_name, void Rename(const std::string &origin_name,
const std::string &new_name) const; const std::string &new_name) const;
// Rename variable to a new name and return the new name // Rename variable to a new name and return the new name
std::string Rename(const std::string &origin_name) const; std::string Rename(const std::string &origin_name) const;
Variable *FindVarLocally(const std::string &name) const; Variable *FindVarLocally(const std::string &name) const;
private: private:
// Call Scope::NewScope for a sub-scope. // Call Scope::NewScope for a sub-scope.
explicit Scope(Scope const *parent) : parent_(parent) {} explicit Scope(Scope const *parent) : parent_(parent) {}
mutable std::unordered_map<std::string, Variable *> vars_; mutable std::unordered_map<std::string, Variable *> vars_;
mutable std::list<Scope *> kids_; mutable std::list<Scope *> kids_;
Scope const *parent_{nullptr}; Scope const *parent_{nullptr};
mutable std::mutex mutex_; mutable std::mutex mutex_;
}; };
} // namespace framework } // namespace framework
} // namespace paddle_mobile } // namespace paddle_mobile
src/framework/selected_rows.h
浏览文件 @ dff4a781
...@@ -27,54 +27,54 @@ namespace paddle_mobile { ...@@ -27,54 +27,54 @@ namespace paddle_mobile {
namespace framework { namespace framework {
class SelectedRows { class SelectedRows {
public: public:
SelectedRows(const std::vector<int64_t> &rows, const int64_t &height) SelectedRows(const std::vector<int64_t> &rows, const int64_t &height)
: rows_(rows), height_(height) { : rows_(rows), height_(height) {
value_.reset(new Tensor()); value_.reset(new Tensor());
} }
SelectedRows() { SelectedRows() {
height_ = 0; height_ = 0;
value_.reset(new Tensor()); value_.reset(new Tensor());
} }
const Tensor &value() const { return *value_; } const Tensor &value() const { return *value_; }
Tensor *mutable_value() { return value_.get(); } Tensor *mutable_value() { return value_.get(); }
int64_t height() const { return height_; } int64_t height() const { return height_; }
void set_height(int64_t height) { height_ = height; } void set_height(int64_t height) { height_ = height; }
const std::vector<int64_t> &rows() const { return rows_; } const std::vector<int64_t> &rows() const { return rows_; }
std::vector<int64_t> *mutable_rows() { return &rows_; } std::vector<int64_t> *mutable_rows() { return &rows_; }
void set_rows(const std::vector<int64_t> &rows) { rows_ = rows; } void set_rows(const std::vector<int64_t> &rows) { rows_ = rows; }
/** /**
* get the index of id in rows * get the index of id in rows
*/ */
int64_t index(int64_t id) const { int64_t index(int64_t id) const {
auto it = std::find(rows_.begin(), rows_.end(), id); auto it = std::find(rows_.begin(), rows_.end(), id);
// PADDLE_ENFORCE(it != rows_.end(), "id should be in rows"); // PADDLE_ENFORCE(it != rows_.end(), "id should be in rows");
return static_cast<int64_t>(std::distance(rows_.begin(), it)); return static_cast<int64_t>(std::distance(rows_.begin(), it));
} }
DDim GetCompleteDims() const { DDim GetCompleteDims() const {
std::vector<int64_t> dims = vectorize(value_->dims()); std::vector<int64_t> dims = vectorize(value_->dims());
dims[0] = height_; dims[0] = height_;
return make_ddim(dims); return make_ddim(dims);
} }
private: private:
// Notice: rows can be duplicate. We can have {0, 4, 7, 0, 5, 7, 9} // Notice: rows can be duplicate. We can have {0, 4, 7, 0, 5, 7, 9}
// here. // here.
// SelectedRows are simply concated when adding together. Until a // SelectedRows are simply concated when adding together. Until a
// SelectedRows add a Tensor, will the duplicate rows be handled. // SelectedRows add a Tensor, will the duplicate rows be handled.
std::vector<int64_t> rows_; std::vector<int64_t> rows_;
std::unique_ptr<Tensor> value_{nullptr}; std::unique_ptr<Tensor> value_{nullptr};
int64_t height_; int64_t height_;
}; };
} // namespace framework } // namespace framework
......
src/framework/tensor.h
浏览文件 @ dff4a781
...@@ -29,305 +29,304 @@ namespace framework { ...@@ -29,305 +29,304 @@ namespace framework {
template <typename... T> struct SizeOfTypeFunctor; template <typename... T> struct SizeOfTypeFunctor;
template <typename T> struct SizeOfTypeFunctor<T> { template <typename T> struct SizeOfTypeFunctor<T> {
size_t operator()(std::type_index type) const { size_t operator()(std::type_index type) const {
if (typeid(T).hash_code() == type.hash_code()) { if (typeid(T).hash_code() == type.hash_code()) {
return sizeof(T); return sizeof(T);
} else { } else {
return 0UL; return 0UL;
}
} }
}
}; };
template <> struct SizeOfTypeFunctor<> { template <> struct SizeOfTypeFunctor<> {
size_t operator()(std::type_index type) const { return 0UL; } size_t operator()(std::type_index type) const { return 0UL; }
}; };
template <typename HEAD, typename... TAIL> template <typename HEAD, typename... TAIL>
struct SizeOfTypeFunctor<HEAD, TAIL...> { struct SizeOfTypeFunctor<HEAD, TAIL...> {
size_t operator()(std::type_index type) const { size_t operator()(std::type_index type) const {
SizeOfTypeFunctor<HEAD> head; SizeOfTypeFunctor<HEAD> head;
size_t head_size = head(type); size_t head_size = head(type);
if (head_size != 0) { if (head_size != 0) {
return head_size; return head_size;
}
SizeOfTypeFunctor<TAIL...> tail;
return tail(type);
} }
SizeOfTypeFunctor<TAIL...> tail;
return tail(type);
}
}; };
static inline size_t SizeOfType(std::type_index type) { static inline size_t SizeOfType(std::type_index type) {
SizeOfTypeFunctor<int, float, double, int16_t, int64_t, bool, size_t> SizeOfTypeFunctor<int, float, double, int16_t, int64_t, bool, size_t> functor;
functor; size_t size = functor(type);
size_t size = functor(type); // PADDLE_ENFORCE(size != 0UL, "Cannot get size of type %s",
// PADDLE_ENFORCE(size != 0UL, "Cannot get size of type %s", // type.name());
// type.name()); return size;
return size;
} }
class LoDTensor; class LoDTensor;
class Tensor { class Tensor {
public: public:
Tensor() : offset_(0) {} Tensor() : offset_(0) {}
/*! Return a pointer to mutable memory block. */ /*! Return a pointer to mutable memory block. */
template <typename T> inline T *data() { template <typename T> inline T *data() {
check_memory_size(); check_memory_size();
// PADDLE_ENFORCE(std::is_same<T, void>::value || // PADDLE_ENFORCE(std::is_same<T, void>::value ||
// holder_->type().hash_code() == // holder_->type().hash_code() ==
// typeid(T).hash_code(), // typeid(T).hash_code(),
// "Tensor holds the wrong type, it holds %s", // "Tensor holds the wrong type, it holds %s",
// this->holder_->type().name()); // this->holder_->type().name());
return reinterpret_cast<T *>( return reinterpret_cast<T *>(reinterpret_cast<uintptr_t>(holder_->ptr()) +
reinterpret_cast<uintptr_t>(holder_->ptr()) + offset_); offset_);
}
/*! Return a pointer to constant memory block. */
template <typename T> inline const T *data() const {
check_memory_size();
// PADDLE_ENFORCE(std::is_same<T, void>::value ||
// holder_->type().hash_code() ==
// typeid(T).hash_code(),
// "Tensor holds the wrong type, it holds %s",
// this->holder_->type().name());
return reinterpret_cast<const T *>(
reinterpret_cast<uintptr_t>(holder_->ptr()) + offset_);
}
inline bool IsInitialized() const { return holder_ != nullptr; }
/**
* @brief Return a pointer to mutable memory block.
* @note If not exist, then allocation.
*/
template <typename T> inline T *mutable_data() {
static_assert(std::is_pod<T>::value, "T must be POD");
return reinterpret_cast<T *>(mutable_data(typeid(T)));
}
inline void *mutable_data(std::type_index type) {
if (holder_ != nullptr) {
holder_->set_type(type);
} }
// PADDLE_ENFORCE_GE(numel(), 0,
/*! Return a pointer to constant memory block. */ // "When calling this method, the Tensor's
template <typename T> inline const T *data() const { // numel must be
check_memory_size(); // " "equal or larger than zero. " "Please
// PADDLE_ENFORCE(std::is_same<T, void>::value || // check
// holder_->type().hash_code() == // Tensor::Resize has been called first.");
// typeid(T).hash_code(), int64_t size = numel() * SizeOfType(type);
// "Tensor holds the wrong type, it holds %s", /* some versions of boost::variant don't have operator!= */
// this->holder_->type().name()); if (holder_ == nullptr || holder_->size() < size + offset_) {
holder_.reset(new PlaceholderImpl(size, type));
return reinterpret_cast<const T *>(
reinterpret_cast<uintptr_t>(holder_->ptr()) + offset_); offset_ = 0;
} }
return reinterpret_cast<void *>(
inline bool IsInitialized() const { return holder_ != nullptr; } reinterpret_cast<uintptr_t>(holder_->ptr()) + offset_);
}
/**
* @brief Return a pointer to mutable memory block. inline void *mutable_data() {
* @note If not exist, then allocation. // PADDLE_ENFORCE(this->holder_ != nullptr,
*/ // "Cannot invoke mutable data if current hold
template <typename T> inline T *mutable_data() { // nothing.");
static_assert(std::is_pod<T>::value, "T must be POD"); return mutable_data(holder_->type());
return reinterpret_cast<T *>(mutable_data(typeid(T))); }
/**
* @brief Return a pointer to mutable memory block.
*
* @param[in] dims The dimensions of the memory block.
* @param[in] place The place of the memory block.
*
* @note If not exist, then allocation.
*/
template <typename T> inline T *mutable_data(DDim dims) {
static_assert(std::is_pod<T>::value, "T must be POD");
Resize(dims);
return mutable_data<T>();
}
/*! Return the dimensions of the memory block. */
inline const DDim &dims() const { return dims_; }
/*! Return the numel of the memory block. */
inline int64_t numel() const { return product(dims_); }
/*! Resize the dimensions of the memory block. */
inline Tensor &Resize(const DDim &dims) {
dims_ = dims;
return *this;
}
/*! The internal of two tensors share the same memory block. */
inline Tensor &ShareDataWith(const Tensor &src) {
src.check_memory_size();
*this = src;
return *this;
}
/**
* @brief Return a sub-tensor of the given tensor.
*
* @param[in] begin_idx The index of the start row(inclusive) to
* slice.
* The index number begins from 0.
* @param[in] end_idx The index of the end row(exclusive) to
* slice.
* The index number begins from 0.
*/
inline Tensor Slice(int begin_idx, int end_idx) const {
check_memory_size();
// PADDLE_ENFORCE_GE(begin_idx, 0,
// "The start row index must be greater than
// 0.");
// PADDLE_ENFORCE_LE(end_idx, dims_[0], "The end row index is
// out of
// bound."); PADDLE_ENFORCE_LT(
// begin_idx, end_idx,
// "The start row index must be lesser than the end row
// index.");
if (dims_[0] == 1) {
return *this;
} else {
size_t base = numel() / dims_[0];
Tensor dst;
dst.holder_ = holder_;
dst.set_layout(layout_);
DDim dst_dims = dims_;
dst_dims[0] = end_idx - begin_idx;
dst.Resize(dst_dims);
dst.offset_ = offset_ + begin_idx * base * SizeOfType(type());
return dst;
} }
}
inline void *mutable_data(std::type_index type) {
if (holder_ != nullptr) { std::type_index type() const {
holder_->set_type(type); // PADDLE_ENFORCE_NOT_NULL(
} // holder_, "Tensor not initialized yet
// PADDLE_ENFORCE_GE(numel(), 0, // when
// "When calling this method, the Tensor's // Tensor::type() is called.");
// numel must be return holder_->type();
// " "equal or larger than zero. " "Please }
// check
// Tensor::Resize has been called first."); // memory size returns the holding memory size in byte.
int64_t size = numel() * SizeOfType(type); size_t memory_size() const {
/* some versions of boost::variant don't have operator!= */ return holder_ == nullptr ? 0UL : holder_->size() - offset_;
if (holder_ == nullptr || holder_->size() < size + offset_) { }
holder_.reset(new PlaceholderImpl(size, type));
inline void check_memory_size() const {
offset_ = 0; // PADDLE_ENFORCE_NOT_NULL(
} // holder_, "Tensor holds no memory. Call
return reinterpret_cast<void *>( // Tensor::mutable_data
reinterpret_cast<uintptr_t>(holder_->ptr()) + offset_); // first.");
// PADDLE_ENFORCE_LE(
// numel() * SizeOfType(type()), memory_size(),
// "Tensor's dims_ is out of bound. Call
// Tensor::mutable_data "
// "first to re-allocate memory.\n"
// "or maybe the required data-type mismatches the data
// already
// stored.");
}
inline DataLayout layout() const { return layout_; }
inline void set_layout(const DataLayout layout) { layout_ = layout; }
private:
/**
* @note Placeholder hides type T, so it doesn't appear as a
* template
* parameter of Variable.
*/
struct Placeholder {
virtual ~Placeholder() = default;
virtual void *ptr() const = 0;
virtual size_t size() const = 0;
virtual std::type_index type() const = 0;
virtual void set_type(std::type_index type) = 0;
};
struct PlaceholderImpl : public Placeholder {
PlaceholderImpl(size_t size, std::type_index type)
: ptr_(static_cast<uint8_t *>(memory::Alloc(size)),
memory::PODDeleter<uint8_t>()),
size_(size), type_(type) {
// PADDLE_ENFORCE_NOT_NULL(ptr_,
// "Insufficient %s
// memory to allocation.",
// (is_cpu_place(place_)
// ?
// "CPU" :
// "GPU"));
} }
inline void *mutable_data() { virtual size_t size() const { return size_; }
// PADDLE_ENFORCE(this->holder_ != nullptr,
// "Cannot invoke mutable data if current hold
// nothing.");
return mutable_data(holder_->type());
}
/** virtual void *ptr() const { return static_cast<void *>(ptr_.get()); }
* @brief Return a pointer to mutable memory block.
*
* @param[in] dims The dimensions of the memory block.
* @param[in] place The place of the memory block.
*
* @note If not exist, then allocation.
*/
template <typename T> inline T *mutable_data(DDim dims) {
static_assert(std::is_pod<T>::value, "T must be POD");
Resize(dims);
return mutable_data<T>();
}
/*! Return the dimensions of the memory block. */ virtual std::type_index type() const { return type_; }
inline const DDim &dims() const { return dims_; }
/*! Return the numel of the memory block. */ virtual void set_type(std::type_index type) { type_ = type; }
inline int64_t numel() const { return product(dims_); }
/*! Resize the dimensions of the memory block. */ /*! the pointer of memory block. */
inline Tensor &Resize(const DDim &dims) { std::unique_ptr<uint8_t, memory::PODDeleter<uint8_t>> ptr_;
dims_ = dims;
return *this;
}
/*! The internal of two tensors share the same memory block. */ /*! the size of memory block. */
inline Tensor &ShareDataWith(const Tensor &src) { size_t size_;
src.check_memory_size();
*this = src;
return *this;
}
/** /* the current type of memory */
* @brief Return a sub-tensor of the given tensor. std::type_index type_;
* };
* @param[in] begin_idx The index of the start row(inclusive) to
* slice.
* The index number begins from 0.
* @param[in] end_idx The index of the end row(exclusive) to
* slice.
* The index number begins from 0.
*/
inline Tensor Slice(int begin_idx, int end_idx) const {
check_memory_size();
// PADDLE_ENFORCE_GE(begin_idx, 0,
// "The start row index must be greater than
// 0.");
// PADDLE_ENFORCE_LE(end_idx, dims_[0], "The end row index is
// out of
// bound."); PADDLE_ENFORCE_LT(
// begin_idx, end_idx,
// "The start row index must be lesser than the end row
// index.");
if (dims_[0] == 1) {
return *this;
} else {
size_t base = numel() / dims_[0];
Tensor dst;
dst.holder_ = holder_;
dst.set_layout(layout_);
DDim dst_dims = dims_;
dst_dims[0] = end_idx - begin_idx;
dst.Resize(dst_dims);
dst.offset_ = offset_ + begin_idx * base * SizeOfType(type());
return dst;
}
}
std::type_index type() const { /*! holds the memory block if allocated. */
// PADDLE_ENFORCE_NOT_NULL( std::shared_ptr<Placeholder> holder_;
// holder_, "Tensor not initialized yet
// when
// Tensor::type() is called.");
return holder_->type();
}
// memory size returns the holding memory size in byte. /**
size_t memory_size() const { * @brief points to elements dimensions.
return holder_ == nullptr ? 0UL : holder_->size() - offset_; *
} * @note dims_ do not indicate the memory block size.
*/
inline void check_memory_size() const { DDim dims_;
// PADDLE_ENFORCE_NOT_NULL(
// holder_, "Tensor holds no memory. Call /**
// Tensor::mutable_data * @brief the layout of memory block, default is NHWC.
// first."); *
// PADDLE_ENFORCE_LE( * @note the memory allocation order, describe how weight/data is
// numel() * SizeOfType(type()), memory_size(), * stored
// "Tensor's dims_ is out of bound. Call * For example, in 4-D Tensor(rank=4), there are three
// Tensor::mutable_data " * commonly
// "first to re-allocate memory.\n" * used layout. They are
// "or maybe the required data-type mismatches the data * NCHW, NHWC, CHWN.
// already * N,C,H,W for respectively the batch size, the number of
// stored."); * feature maps, the height, the width.
} */
DataLayout layout_ = DataLayout::kNHWC;
inline DataLayout layout() const { return layout_; } /**
* @brief A PlaceHolder may be shared by more than one tensor.
inline void set_layout(const DataLayout layout) { layout_ = layout; } *
* @note Some of them may be slices of the others. So the offset_
private: * is introduced here to indicate the byte offset between
/** * PlaceHolder::ptr_ and where the tensor data really
* @note Placeholder hides type T, so it doesn't appear as a * begins.
* template */
* parameter of Variable. size_t offset_;
*/
struct Placeholder {
virtual ~Placeholder() = default;
virtual void *ptr() const = 0;
virtual size_t size() const = 0;
virtual std::type_index type() const = 0;
virtual void set_type(std::type_index type) = 0;
};
struct PlaceholderImpl : public Placeholder {
PlaceholderImpl(size_t size, std::type_index type)
: ptr_(static_cast<uint8_t *>(memory::Alloc(size)),
memory::PODDeleter<uint8_t>()),
size_(size), type_(type) {
// PADDLE_ENFORCE_NOT_NULL(ptr_,
// "Insufficient %s
// memory to allocation.",
// (is_cpu_place(place_)
// ?
// "CPU" :
// "GPU"));
}
virtual size_t size() const { return size_; }
virtual void *ptr() const { return static_cast<void *>(ptr_.get()); }
virtual std::type_index type() const { return type_; }
virtual void set_type(std::type_index type) { type_ = type; }
/*! the pointer of memory block. */
std::unique_ptr<uint8_t, memory::PODDeleter<uint8_t>> ptr_;
/*! the size of memory block. */
size_t size_;
/* the current type of memory */
std::type_index type_;
};
/*! holds the memory block if allocated. */
std::shared_ptr<Placeholder> holder_;
/**
* @brief points to elements dimensions.
*
* @note dims_ do not indicate the memory block size.
*/
DDim dims_;
/**
* @brief the layout of memory block, default is NHWC.
*
* @note the memory allocation order, describe how weight/data is
* stored
* For example, in 4-D Tensor(rank=4), there are three
* commonly
* used layout. They are
* NCHW, NHWC, CHWN.
* N,C,H,W for respectively the batch size, the number of
* feature maps, the height, the width.
*/
DataLayout layout_ = DataLayout::kNHWC;
/**
* @brief A PlaceHolder may be shared by more than one tensor.
*
* @note Some of them may be slices of the others. So the offset_
* is introduced here to indicate the byte offset between
* PlaceHolder::ptr_ and where the tensor data really
* begins.
*/
size_t offset_;
}; };
inline Tensor ReshapeToMatrix(const Tensor &src, int num_col_dims) { inline Tensor ReshapeToMatrix(const Tensor &src, int num_col_dims) {
Tensor res; Tensor res;
res.ShareDataWith(src); res.ShareDataWith(src);
res.Resize(flatten_to_2d(src.dims(), num_col_dims)); res.Resize(flatten_to_2d(src.dims(), num_col_dims));
return res; return res;
} }
} // namespace framework } // namespace framework
......
src/framework/tensor_util.cc
浏览文件 @ dff4a781
...@@ -21,183 +21,182 @@ namespace paddle_mobile { ...@@ -21,183 +21,182 @@ namespace paddle_mobile {
namespace framework { namespace framework {
void TensorCopy(const Tensor &src, Tensor *dst) { void TensorCopy(const Tensor &src, Tensor *dst) {
// VLOG(3) << "TensorCopy " << src.dims() << " from " << // VLOG(3) << "TensorCopy " << src.dims() << " from " <<
// src.place() << " to // src.place() << " to
// " // "
// << dst_place; // << dst_place;
src.check_memory_size(); src.check_memory_size();
dst->Resize(src.dims()); dst->Resize(src.dims());
dst->set_layout(src.layout()); dst->set_layout(src.layout());
auto src_ptr = src.data<void>(); auto src_ptr = src.data<void>();
auto dst_ptr = dst->mutable_data(src.type()); auto dst_ptr = dst->mutable_data(src.type());
auto size = src.numel() * SizeOfType(src.type()); auto size = src.numel() * SizeOfType(src.type());
memory::Copy(dst_ptr, src_ptr, size); memory::Copy(dst_ptr, src_ptr, size);
} }
void TensorCopySync(const Tensor &src, Tensor *dst) { void TensorCopySync(const Tensor &src, Tensor *dst) {
// VLOG(3) << "TensorCopySync " << src.dims() << " from " << // VLOG(3) << "TensorCopySync " << src.dims() << " from " <<
// src.place() // src.place()
// << " to " << dst_place; // << " to " << dst_place;
src.check_memory_size(); src.check_memory_size();
dst->Resize(src.dims()); dst->Resize(src.dims());
dst->set_layout(src.layout()); dst->set_layout(src.layout());
auto src_ptr = src.data<void>(); auto src_ptr = src.data<void>();
auto dst_ptr = dst->mutable_data(src.type()); auto dst_ptr = dst->mutable_data(src.type());
auto size = src.numel() * SizeOfType(src.type()); auto size = src.numel() * SizeOfType(src.type());
memory::Copy(dst_ptr, src_ptr, size); memory::Copy(dst_ptr, src_ptr, size);
} }
template <typename Predicate> struct AnyDTypeVisitor { template <typename Predicate> struct AnyDTypeVisitor {
Predicate predicate_; Predicate predicate_;
const Tensor &tensor_; const Tensor &tensor_;
Tensor *out_; Tensor *out_;
AnyDTypeVisitor(Predicate predicate, const Tensor &tensor, Tensor *out) AnyDTypeVisitor(Predicate predicate, const Tensor &tensor, Tensor *out)
: predicate_(predicate), tensor_(tensor), out_(out) {} : predicate_(predicate), tensor_(tensor), out_(out) {}
template <typename T> void operator()() const { template <typename T> void operator()() const {
// auto t = EigenVector<T>::Flatten(tensor_); // auto t = EigenVector<T>::Flatten(tensor_);
// auto o = EigenScalar<bool>::From(*out_); // auto o = EigenScalar<bool>::From(*out_);
// return any of predicate_(t) is true. // return any of predicate_(t) is true.
// o.device(*ctx_.eigen_device()) = predicate_(t).any(); // o.device(*ctx_.eigen_device()) = predicate_(t).any();
} }
}; };
template <typename Predicate> template <typename Predicate>
inline void AnyImpl(Predicate predicate, const Tensor &tensor, inline void AnyImpl(Predicate predicate, const Tensor &tensor,
framework::Tensor *out) { framework::Tensor *out) {
VisitDataType(ToDataType(tensor.type()), VisitDataType(ToDataType(tensor.type()),
AnyDTypeVisitor<Predicate>(predicate, tensor, out)); AnyDTypeVisitor<Predicate>(predicate, tensor, out));
} }
template <typename Predicate> struct AnyVisitor { template <typename Predicate> struct AnyVisitor {
const framework::Tensor &tensor_; const framework::Tensor &tensor_;
Predicate predicate_; Predicate predicate_;
AnyVisitor(const framework::Tensor &tensor, Predicate predicate) AnyVisitor(const framework::Tensor &tensor, Predicate predicate)
: tensor_(tensor), predicate_(std::move(predicate)) {} : tensor_(tensor), predicate_(std::move(predicate)) {}
bool operator()(void) const { bool operator()(void) const {
framework::Tensor out; framework::Tensor out;
out.Resize({1}); out.Resize({1});
out.mutable_data<bool>(); out.mutable_data<bool>();
AnyImpl(predicate_, tensor_, &out); AnyImpl(predicate_, tensor_, &out);
return this->GetResult(out); return this->GetResult(out);
} }
bool GetResult(const framework::Tensor &out) const { bool GetResult(const framework::Tensor &out) const {
return *out.data<bool>(); return *out.data<bool>();
} }
}; };
template <typename Predicate> template <typename Predicate>
inline bool Any(const framework::Tensor &tensor, Predicate predicate) { inline bool Any(const framework::Tensor &tensor, Predicate predicate) {
AnyVisitor<Predicate> visitor(tensor, predicate); AnyVisitor<Predicate> visitor(tensor, predicate);
// return platform::VisitPlace(visitor); // return platform::VisitPlace(visitor);
return visitor(); return visitor();
} }
struct ContainsNANPredicate { struct ContainsNANPredicate {
template <typename T> template <typename T>
auto operator()(const T &eigen_vec) const auto operator()(const T &eigen_vec) const
-> decltype(std::declval<T>().isnan()) { -> decltype(std::declval<T>().isnan()) {
// Cast eigen_vector to vector of bool. true if is inf. // Cast eigen_vector to vector of bool. true if is inf.
return eigen_vec.isnan(); return eigen_vec.isnan();
} }
}; };
bool TensorContainsNAN(const framework::Tensor &tensor) { bool TensorContainsNAN(const framework::Tensor &tensor) {
ContainsNANPredicate predicate; ContainsNANPredicate predicate;
return Any(tensor, predicate); return Any(tensor, predicate);
} }
struct ContainsInfPredicate { struct ContainsInfPredicate {
template <typename T> template <typename T>
auto operator()(const T &eigen_vec) const auto operator()(const T &eigen_vec) const
-> decltype(std::declval<T>().isinf()) { -> decltype(std::declval<T>().isinf()) {
// Cast eigen_vector to vector of bool. true if is inf. // Cast eigen_vector to vector of bool. true if is inf.
return eigen_vec.isinf(); return eigen_vec.isinf();
} }
}; };
bool TensorContainsInf(const framework::Tensor &tensor) { bool TensorContainsInf(const framework::Tensor &tensor) {
ContainsInfPredicate predicate; ContainsInfPredicate predicate;
return Any(tensor, predicate); return Any(tensor, predicate);
} }
void TensorToStream(std::ostream &os, const Tensor &tensor) { void TensorToStream(std::ostream &os, const Tensor &tensor) {
{ // the 1st field, uint32_t version { // the 1st field, uint32_t version
constexpr uint32_t version = 0; constexpr uint32_t version = 0;
os.write(reinterpret_cast<const char *>(&version), sizeof(version)); os.write(reinterpret_cast<const char *>(&version), sizeof(version));
} }
{ // the 2nd field, tensor description { // the 2nd field, tensor description
// int32_t size // int32_t size
// void* protobuf message // void* protobuf message
proto::VarType::TensorDesc desc; proto::VarType::TensorDesc desc;
desc.set_data_type(framework::ToDataType(tensor.type())); desc.set_data_type(framework::ToDataType(tensor.type()));
auto dims = framework::vectorize(tensor.dims()); auto dims = framework::vectorize(tensor.dims());
auto *pb_dims = desc.mutable_dims(); auto *pb_dims = desc.mutable_dims();
pb_dims->Resize(static_cast<int>(dims.size()), 0); pb_dims->Resize(static_cast<int>(dims.size()), 0);
std::copy(dims.begin(), dims.end(), pb_dims->begin()); std::copy(dims.begin(), dims.end(), pb_dims->begin());
int32_t size = desc.ByteSize(); int32_t size = desc.ByteSize();
os.write(reinterpret_cast<const char *>(&size), sizeof(size)); os.write(reinterpret_cast<const char *>(&size), sizeof(size));
auto out = desc.SerializeAsString(); auto out = desc.SerializeAsString();
os.write(out.data(), size); os.write(out.data(), size);
} }
{ // the 3rd field, tensor data { // the 3rd field, tensor data
uint64_t size = tensor.memory_size(); uint64_t size = tensor.memory_size();
auto *data_ptr = tensor.data<void>(); auto *data_ptr = tensor.data<void>();
// PADDLE_ENFORCE(size < // PADDLE_ENFORCE(size <
// std::numeric_limits<std::streamsize>::max(), // std::numeric_limits<std::streamsize>::max(),
// "Index overflow when writing tensor"); // "Index overflow when writing tensor");
os.write(static_cast<const char *>(data_ptr), os.write(static_cast<const char *>(data_ptr),
static_cast<std::streamsize>(size)); static_cast<std::streamsize>(size));
} }
} }
struct DeserializedDataFunctor { struct DeserializedDataFunctor {
DeserializedDataFunctor(void **buf, Tensor *tensor) DeserializedDataFunctor(void **buf, Tensor *tensor)
: buf_(buf), tensor_(tensor) {} : buf_(buf), tensor_(tensor) {}
template <typename T> void operator()() { template <typename T> void operator()() {
*buf_ = tensor_->mutable_data<T>(); *buf_ = tensor_->mutable_data<T>();
} }
void **buf_; void **buf_;
Tensor *tensor_; Tensor *tensor_;
}; };
void TensorFromStream(std::istream &is, framework::Tensor *tensor) { void TensorFromStream(std::istream &is, framework::Tensor *tensor) {
uint32_t version; uint32_t version;
is.read(reinterpret_cast<char *>(&version), sizeof(version)); is.read(reinterpret_cast<char *>(&version), sizeof(version));
// PADDLE_ENFORCE_EQ(version, 0U, "Only version 0 is supported"); // PADDLE_ENFORCE_EQ(version, 0U, "Only version 0 is supported");
proto::VarType::TensorDesc desc; proto::VarType::TensorDesc desc;
{ // int32_t size { // int32_t size
// proto buffer // proto buffer
int32_t size; int32_t size;
is.read(reinterpret_cast<char *>(&size), sizeof(size)); is.read(reinterpret_cast<char *>(&size), sizeof(size));
std::unique_ptr<char[]> buf(new char[size]); std::unique_ptr<char[]> buf(new char[size]);
is.read(reinterpret_cast<char *>(buf.get()), size); is.read(reinterpret_cast<char *>(buf.get()), size);
// PADDLE_ENFORCE(desc.ParseFromArray(buf.get(), size), // PADDLE_ENFORCE(desc.ParseFromArray(buf.get(), size),
// "Cannot parse tensor desc"); // "Cannot parse tensor desc");
} }
{ // read tensor { // read tensor
std::vector<int64_t> dims; std::vector<int64_t> dims;
dims.reserve(static_cast<size_t>(desc.dims().size())); dims.reserve(static_cast<size_t>(desc.dims().size()));
std::copy(desc.dims().begin(), desc.dims().end(), std::copy(desc.dims().begin(), desc.dims().end(), std::back_inserter(dims));
std::back_inserter(dims)); tensor->Resize(framework::make_ddim(dims));
tensor->Resize(framework::make_ddim(dims)); void *buf;
void *buf;
framework::VisitDataType(desc.data_type(),
framework::VisitDataType(desc.data_type(), DeserializedDataFunctor(&buf, tensor));
DeserializedDataFunctor(&buf, tensor)); is.read(static_cast<char *>(buf), tensor->memory_size());
is.read(static_cast<char *>(buf), tensor->memory_size()); }
}
} }
} // namespace framework } // namespace framework
......
src/framework/tensor_util.h
浏览文件 @ dff4a781
...@@ -43,23 +43,23 @@ void TensorFromStream(std::istream &is, Tensor *tensor); ...@@ -43,23 +43,23 @@ void TensorFromStream(std::istream &is, Tensor *tensor);
template <typename T> template <typename T>
void TensorFromVector(const std::vector<T> &src, Tensor *dst) { void TensorFromVector(const std::vector<T> &src, Tensor *dst) {
auto src_ptr = static_cast<const void *>(src.data()); auto src_ptr = static_cast<const void *>(src.data());
dst->Resize({static_cast<int64_t>(src.size())}); dst->Resize({static_cast<int64_t>(src.size())});
auto dst_ptr = static_cast<void *>(dst->mutable_data<T>()); auto dst_ptr = static_cast<void *>(dst->mutable_data<T>());
auto size = src.size() * sizeof(T); auto size = src.size() * sizeof(T);
memory::Copy(dst_ptr, src_ptr, size); memory::Copy(dst_ptr, src_ptr, size);
} }
template <typename T> template <typename T>
void TensorToVector(const Tensor &src, std::vector<T> *dst) { void TensorToVector(const Tensor &src, std::vector<T> *dst) {
auto src_ptr = static_cast<const void *>(src.data<T>()); auto src_ptr = static_cast<const void *>(src.data<T>());
auto size = src.numel() * sizeof(T); auto size = src.numel() * sizeof(T);
dst->resize(src.numel()); dst->resize(src.numel());
auto dst_ptr = static_cast<void *>(dst->data()); auto dst_ptr = static_cast<void *>(dst->data());
memory::Copy(dst_ptr, src_ptr, size); memory::Copy(dst_ptr, src_ptr, size);
} }
} // namespace framework } // namespace framework
......
src/framework/var_desc.h
浏览文件 @ dff4a781
...@@ -25,63 +25,63 @@ namespace paddle_mobile { ...@@ -25,63 +25,63 @@ namespace paddle_mobile {
namespace framework { namespace framework {
class VarDesc { class VarDesc {
public: public:
VarDesc(const proto::VarDesc &desc); VarDesc(const proto::VarDesc &desc);
std::string Name() const { return desc_.name(); } std::string Name() const { return desc_.name(); }
proto::VarType::Type GetType() const { return desc_.type().type(); } proto::VarType::Type GetType() const { return desc_.type().type(); }
bool Persistable() const { return desc_.persistable(); } bool Persistable() const { return desc_.persistable(); }
const proto::VarType::ChannelDesc &channel_desc() const { const proto::VarType::ChannelDesc &channel_desc() const {
switch (desc_.type().type()) { switch (desc_.type().type()) {
case proto::VarType::CHANNEL: case proto::VarType::CHANNEL:
return desc_.type().channel(); return desc_.type().channel();
default: default:
break; break;
}
} }
}
const proto::VarType::TensorDesc &tensor_desc() const { const proto::VarType::TensorDesc &tensor_desc() const {
switch (desc_.type().type()) { switch (desc_.type().type()) {
case proto::VarType::SELECTED_ROWS: case proto::VarType::SELECTED_ROWS:
return desc_.type().selected_rows(); return desc_.type().selected_rows();
case proto::VarType::LOD_TENSOR: case proto::VarType::LOD_TENSOR:
return desc_.type().lod_tensor().tensor(); return desc_.type().lod_tensor().tensor();
case proto::VarType::LOD_TENSOR_ARRAY: case proto::VarType::LOD_TENSOR_ARRAY:
return desc_.type().tensor_array().tensor(); return desc_.type().tensor_array().tensor();
default: default:
break; break;
}
} }
}
proto::VarType::Type GetDataType() const { proto::VarType::Type GetDataType() const {
switch (desc_.type().type()) { switch (desc_.type().type()) {
case proto::VarType::CHANNEL: case proto::VarType::CHANNEL:
return channel_desc().data_type(); return channel_desc().data_type();
break; break;
default: default:
return tensor_desc().data_type(); return tensor_desc().data_type();
}
} }
}
template <typename T> template <typename T>
std::vector<T> RepeatedToVector( std::vector<T> RepeatedToVector(
const google::protobuf::RepeatedField<T> &repeated_field) const { const google::protobuf::RepeatedField<T> &repeated_field) const {
std::vector<T> ret; std::vector<T> ret;
ret.reserve(repeated_field.size()); ret.reserve(repeated_field.size());
std::copy(repeated_field.begin(), repeated_field.end(), std::copy(repeated_field.begin(), repeated_field.end(),
std::back_inserter(ret)); std::back_inserter(ret));
return ret; return ret;
} }
std::vector<int64_t> GetShape() const { std::vector<int64_t> GetShape() const {
return this->RepeatedToVector(tensor_desc().dims()); return this->RepeatedToVector(tensor_desc().dims());
} }
private: private:
proto::VarDesc desc_; proto::VarDesc desc_;
}; };
} // namespace framework } // namespace framework
......
src/framework/var_type.h
浏览文件 @ dff4a781
...@@ -25,14 +25,14 @@ SOFTWARE. ...@@ -25,14 +25,14 @@ SOFTWARE.
namespace paddle_mobile { namespace paddle_mobile {
namespace framework { namespace framework {
inline proto::VarType::Type ToVarType(std::type_index type) { inline proto::VarType::Type ToVarType(std::type_index type) {
if (type.hash_code() == typeid(LoDTensor).hash_code()) { if (type.hash_code() == typeid(LoDTensor).hash_code()) {
return proto::VarType_Type_LOD_TENSOR; return proto::VarType_Type_LOD_TENSOR;
} else if (type.hash_code() == typeid(SelectedRows).hash_code()) { } else if (type.hash_code() == typeid(SelectedRows).hash_code()) {
return proto::VarType_Type_SELECTED_ROWS; return proto::VarType_Type_SELECTED_ROWS;
} else { } else {
// PADDLE_THROW("ToVarType:Unsupported type %s", // PADDLE_THROW("ToVarType:Unsupported type %s",
// type.name()); // type.name());
} }
} }
} // namespace framework } // namespace framework
......
src/framework/variable.h
浏览文件 @ dff4a781
...@@ -28,69 +28,69 @@ SOFTWARE. ...@@ -28,69 +28,69 @@ SOFTWARE.
namespace paddle_mobile { namespace paddle_mobile {
namespace framework { namespace framework {
class Variable : public PaddleMobileObject { class Variable : public PaddleMobileObject {
public: public:
template <typename T> const T *Get() const { template <typename T> const T *Get() const {
return static_cast<const T *>(holder_->Ptr()); return static_cast<const T *>(holder_->Ptr());
}
bool IsInitialized() const { return holder_ != nullptr; }
const std::string *Name() { return name_; }
template <typename T> T *GetMutable() {
if (!IsType<T>()) {
if (*Name() == "pixel") {
// std::cout << " reset " << *Name() <<
// std::endl;
}
holder_.reset(new PlaceholderImp<T>(new T()));
} }
return static_cast<T *>(holder_->Ptr());
bool IsInitialized() const { return holder_ != nullptr; } }
const std::string *Name() { return name_; } template <typename T> bool IsType() const {
if (holder_) {
template <typename T> T *GetMutable() { // printf("not null \n");
if (!IsType<T>()) { printf(" holder type : %s, this type %s \n", holder_->Type().name(),
if (*Name() == "pixel") { typeid(T).name());
// std::cout << " reset " << *Name() <<
// std::endl;
}
holder_.reset(new PlaceholderImp<T>(new T()));
}
return static_cast<T *>(holder_->Ptr());
} }
template <typename T> bool IsType() const { // std::cout << " " << holder_->Type() << " " <<
if (holder_) { // typeid(T) <<
// printf("not null \n"); // std::endl;
printf(" holder type : %s, this type %s \n", holder_->Type().name(), return holder_ != nullptr && holder_->Type() == typeid(T);
typeid(T).name()); }
}
// std::cout << " " << holder_->Type() << " " <<
// typeid(T) <<
// std::endl;
return holder_ != nullptr && holder_->Type() == typeid(T);
}
void Clear() { holder_.reset(); } void Clear() { holder_.reset(); }
std::type_index Type() const { return holder_->Type(); } std::type_index Type() const { return holder_->Type(); }
void SetName(const std::string *name) { name_ = name; } void SetName(const std::string *name) { name_ = name; }
private: private:
struct Placeholder { struct Placeholder {
Placeholder() = default; Placeholder() = default;
virtual ~Placeholder() = default; virtual ~Placeholder() = default;
virtual const std::type_info &Type() const = 0; virtual const std::type_info &Type() const = 0;
virtual void *Ptr() const = 0; virtual void *Ptr() const = 0;
}; };
template <typename T> struct PlaceholderImp : public Placeholder { template <typename T> struct PlaceholderImp : public Placeholder {
explicit PlaceholderImp(T *ptr) : ptr_(ptr), type_(typeid(T)) {} explicit PlaceholderImp(T *ptr) : ptr_(ptr), type_(typeid(T)) {}
virtual const std::type_info &Type() const { return type_; } virtual const std::type_info &Type() const { return type_; }
virtual void *Ptr() const override { virtual void *Ptr() const override {
return static_cast<void *>(ptr_.get()); return static_cast<void *>(ptr_.get());
} }
std::unique_ptr<T> ptr_; std::unique_ptr<T> ptr_;
const std::type_info &type_; const std::type_info &type_;
}; };
std::unique_ptr<Placeholder> holder_; std::unique_ptr<Placeholder> holder_;
friend class Scope; friend class Scope;
const std::string *name_; const std::string *name_;
}; };
} // namespace framework } // namespace framework
} // namespace paddle_mobile } // namespace paddle_mobile
src/io.cpp
浏览文件 @ dff4a781
...@@ -29,361 +29,357 @@ SOFTWARE. ...@@ -29,361 +29,357 @@ SOFTWARE.
namespace paddle_mobile { namespace paddle_mobile {
void ReadBinaryFile(const std::string &filename, std::string *contents) { void ReadBinaryFile(const std::string &filename, std::string *contents) {
std::ifstream fin(filename, std::ios::in | std::ios::binary); std::ifstream fin(filename, std::ios::in | std::ios::binary);
fin.seekg(0, std::ios::end); fin.seekg(0, std::ios::end);
contents->clear(); contents->clear();
contents->resize(fin.tellg()); contents->resize(fin.tellg());
fin.seekg(0, std::ios::beg); fin.seekg(0, std::ios::beg);
fin.read(&(contents->at(0)), contents->size()); fin.read(&(contents->at(0)), contents->size());
fin.close(); fin.close();
} }
template <typename Dtype, Precision P> template <typename Dtype, Precision P>
void Loader<Dtype, P>::LoadVar(framework::LoDTensor *tensor, void Loader<Dtype, P>::LoadVar(framework::LoDTensor *tensor,
const std::string &file_path) { const std::string &file_path) {
// LOG(kLOG_DEBUG) << " to load " << file_path; // LOG(kLOG_DEBUG) << " to load " << file_path;
// Log(kLOG_DEBUG) << "123"; // Log(kLOG_DEBUG) << "123";
std::ifstream is(file_path); std::ifstream is(file_path);
std::streampos pos = is.tellg(); // save current position std::streampos pos = is.tellg(); // save current position
is.seekg(0, std::ios::end); is.seekg(0, std::ios::end);
// LOG(kLOG_DEBUG) << " file length = " << is.tellg(); // LOG(kLOG_DEBUG) << " file length = " << is.tellg();
is.seekg(pos); // restore saved position is.seekg(pos); // restore saved position
// 1. version // 1. version
uint32_t version; uint32_t version;
is.read(reinterpret_cast<char *>(&version), sizeof(version)); is.read(reinterpret_cast<char *>(&version), sizeof(version));
// LOG(kLOG_INFO) << " version: " << version; // LOG(kLOG_INFO) << " version: " << version;
// 2 Lod information // 2 Lod information
uint64_t lod_level; uint64_t lod_level;
is.read(reinterpret_cast<char *>(&lod_level), sizeof(lod_level)); is.read(reinterpret_cast<char *>(&lod_level), sizeof(lod_level));
// LOG(kLOG_DEBUG) << " load level: " << lod_level; // LOG(kLOG_DEBUG) << " load level: " << lod_level;
// LOG(kLOG_DEBUG) << " lod info: "; // LOG(kLOG_DEBUG) << " lod info: ";
auto &lod = *tensor->mutable_lod(); auto &lod = *tensor->mutable_lod();
lod.resize(lod_level); lod.resize(lod_level);
for (uint64_t i = 0; i < lod_level; ++i) { for (uint64_t i = 0; i < lod_level; ++i) {
uint64_t size; uint64_t size;
is.read(reinterpret_cast<char *>(&size), sizeof(size));
std::vector<size_t> tmp(size / sizeof(size_t));
is.read(reinterpret_cast<char *>(tmp.data()),
static_cast<std::streamsize>(size));
for (int j = 0; j < tmp.size(); ++j) {
LOG(kLOG_DEBUG1) << " lod - " << tmp[j];
}
lod[i] = tmp;
}
// 3. tensor version
uint32_t tensor_version;
is.read(reinterpret_cast<char *>(&tensor_version), sizeof(tensor_version));
// std::cout << " tensor_version: " << tensor_version << std::endl;
// 4. tensor desc
int32_t size;
is.read(reinterpret_cast<char *>(&size), sizeof(size)); is.read(reinterpret_cast<char *>(&size), sizeof(size));
// std::cout << " tensor desc size: " << size << std::endl; std::vector<size_t> tmp(size / sizeof(size_t));
std::unique_ptr<char[]> buf(new char[size]); is.read(reinterpret_cast<char *>(tmp.data()),
is.read(reinterpret_cast<char *>(buf.get()), size); static_cast<std::streamsize>(size));
for (int j = 0; j < tmp.size(); ++j) {
framework::proto::VarType::TensorDesc desc; LOG(kLOG_DEBUG1) << " lod - " << tmp[j];
desc.ParseFromArray(buf.get(), size);
// std::cout << " desc dims size " << desc.dims().size() <<
// std::endl;
int memory_size = 1;
for (int l = 0; l < desc.dims().size(); ++l) {
// std::cout << " dim " << l << " value: " << desc.dims()[l]
// <<
// std::endl;
memory_size *= desc.dims()[l];
}
std::vector<int64_t> dims;
dims.reserve(static_cast<size_t>(desc.dims().size()));
std::copy(desc.dims().begin(), desc.dims().end(), std::back_inserter(dims));
tensor->Resize(framework::make_ddim(dims));
void *memory;
int type_size = 0;
// std::cout << " desc pre type: ";
switch (desc.data_type()) {
case framework::proto::VarType::FP16:
// std::cout << "FP16" << std::endl;
type_size = 2;
break;
case framework::proto::VarType::FP32:
type_size = 4;
memory = tensor->mutable_data<float>();
// std::cout << "FP32" << std::endl;
break;
case framework::proto::VarType::FP64:
type_size = 8;
// std::cout << "FP64" << std::endl;
break;
case framework::proto::VarType::INT32:
type_size = 4;
// std::cout << "INT32" << std::endl;
break;
case framework::proto::VarType::INT64:
type_size = 8;
// std::cout << "INT64" << std::endl;
break;
case framework::proto::VarType::BOOL:
type_size = 1;
// std::cout << "BOOL" << std::endl;
break;
default:
break;
// std::cout << " not support" << std::endl;
} }
lod[i] = tmp;
// std::cout << " malloc size: " << memory_size * type_size << }
// std::endl;
is.read(static_cast<char *>(memory), memory_size * type_size); // 3. tensor version
// std::cout << " memory: " << memory << std::endl; uint32_t tensor_version;
is.close(); is.read(reinterpret_cast<char *>(&tensor_version), sizeof(tensor_version));
// std::cout << " tensor_version: " << tensor_version << std::endl;
// 4. tensor desc
int32_t size;
is.read(reinterpret_cast<char *>(&size), sizeof(size));
// std::cout << " tensor desc size: " << size << std::endl;
std::unique_ptr<char[]> buf(new char[size]);
is.read(reinterpret_cast<char *>(buf.get()), size);
framework::proto::VarType::TensorDesc desc;
desc.ParseFromArray(buf.get(), size);
// std::cout << " desc dims size " << desc.dims().size() <<
// std::endl;
int memory_size = 1;
for (int l = 0; l < desc.dims().size(); ++l) {
// std::cout << " dim " << l << " value: " << desc.dims()[l]
// <<
// std::endl;
memory_size *= desc.dims()[l];
}
std::vector<int64_t> dims;
dims.reserve(static_cast<size_t>(desc.dims().size()));
std::copy(desc.dims().begin(), desc.dims().end(), std::back_inserter(dims));
tensor->Resize(framework::make_ddim(dims));
void *memory;
int type_size = 0;
// std::cout << " desc pre type: ";
switch (desc.data_type()) {
case framework::proto::VarType::FP16:
// std::cout << "FP16" << std::endl;
type_size = 2;
break;
case framework::proto::VarType::FP32:
type_size = 4;
memory = tensor->mutable_data<float>();
// std::cout << "FP32" << std::endl;
break;
case framework::proto::VarType::FP64:
type_size = 8;
// std::cout << "FP64" << std::endl;
break;
case framework::proto::VarType::INT32:
type_size = 4;
// std::cout << "INT32" << std::endl;
break;
case framework::proto::VarType::INT64:
type_size = 8;
// std::cout << "INT64" << std::endl;
break;
case framework::proto::VarType::BOOL:
type_size = 1;
// std::cout << "BOOL" << std::endl;
break;
default:
break;
// std::cout << " not support" << std::endl;
}
// std::cout << " malloc size: " << memory_size * type_size <<
// std::endl;
is.read(static_cast<char *>(memory), memory_size * type_size);
// std::cout << " memory: " << memory << std::endl;
is.close();
}; };
template <typename Dtype, Precision P> template <typename Dtype, Precision P>
const framework::Program<Dtype, P> const framework::Program<Dtype, P>
Loader<Dtype, P>::Load(const std::string &dirname) { Loader<Dtype, P>::Load(const std::string &dirname) {
std::string model_filename = dirname + "/__model__"; std::string model_filename = dirname + "/__model__";
std::string program_desc_str; std::string program_desc_str;
ReadBinaryFile(model_filename, &program_desc_str); ReadBinaryFile(model_filename, &program_desc_str);
framework::proto::ProgramDesc program_desc_proto; framework::proto::ProgramDesc program_desc_proto;
program_desc_proto.ParseFromString(program_desc_str); program_desc_proto.ParseFromString(program_desc_str);
std::shared_ptr<framework::ProgramDesc> originProgramDesc = std::shared_ptr<framework::ProgramDesc> originProgramDesc =
std::make_shared<framework::ProgramDesc>(program_desc_proto); std::make_shared<framework::ProgramDesc>(program_desc_proto);
framework::Program<Dtype, P> program; framework::Program<Dtype, P> program;
program.originProgram = originProgramDesc; program.originProgram = originProgramDesc;
std::shared_ptr<framework::Scope> scope = std::shared_ptr<framework::Scope> scope =
std::make_shared<framework::Scope>(); std::make_shared<framework::Scope>();
program.scope = scope; program.scope = scope;
auto block = originProgramDesc->Block(0); auto block = originProgramDesc->Block(0);
for (auto block : originProgramDesc->Blocks()) { for (auto block : originProgramDesc->Blocks()) {
// std::cout << "for block" << std::endl; // std::cout << "for block" << std::endl;
for (int i = 0; i < block->Vars().size(); ++i) { for (int i = 0; i < block->Vars().size(); ++i) {
std::shared_ptr<framework::VarDesc> var_desc = block->Vars()[i]; std::shared_ptr<framework::VarDesc> var_desc = block->Vars()[i];
auto var = scope->Var(var_desc->Name()); auto var = scope->Var(var_desc->Name());
if (var_desc->GetType() == framework::proto::VarType::LOD_TENSOR) { if (var_desc->GetType() == framework::proto::VarType::LOD_TENSOR) {
if (var_desc->Persistable() && if (var_desc->Persistable() &&
var_desc->GetType() != var_desc->GetType() != framework::proto::VarType::FEED_MINIBATCH &&
framework::proto::VarType::FEED_MINIBATCH && var_desc->GetType() != framework::proto::VarType::FETCH_LIST) {
var_desc->GetType() != framework::LoDTensor *tensor =
framework::proto::VarType::FETCH_LIST) { var->GetMutable<framework::LoDTensor>();
framework::LoDTensor *tensor = // to load
var->GetMutable<framework::LoDTensor>(); LoadVar(tensor, dirname + "/" + var_desc->Name());
// to load
LoadVar(tensor, dirname + "/" + var_desc->Name());
}
} else {
// std::cout << "非 lod" << std::endl;
}
} }
} else {
// std::cout << "非 lod" << std::endl;
}
} }
}
#ifdef PADDLE_MOBILE_DEBUG #ifdef PADDLE_MOBILE_DEBUG
for (int i = 0; i < program_desc_proto.blocks().size(); ++i) { for (int i = 0; i < program_desc_proto.blocks().size(); ++i) {
framework::proto::BlockDesc block = program_desc_proto.blocks()[i]; framework::proto::BlockDesc block = program_desc_proto.blocks()[i];
LOG(kLOG_DEBUG) << "block: " << block.idx(); LOG(kLOG_DEBUG) << "block: " << block.idx();
for (int j = 0; j < block.ops().size(); ++j) { for (int j = 0; j < block.ops().size(); ++j) {
if (j == 2) { if (j == 2) {
break; break;
} }
framework::proto::OpDesc op = block.ops()[j]; framework::proto::OpDesc op = block.ops()[j];
LOG(kLOG_DEBUG1) << "op: " << op.type(); LOG(kLOG_DEBUG1) << "op: " << op.type();
for (int m = 0; m < op.inputs_size(); ++m) { for (int m = 0; m < op.inputs_size(); ++m) {
const framework::proto::OpDesc::Var &var = op.inputs(m); const framework::proto::OpDesc::Var &var = op.inputs(m);
LOG(kLOG_DEBUG2) << "input parameter: " << var.parameter(); LOG(kLOG_DEBUG2) << "input parameter: " << var.parameter();
for (int n = 0; n < var.arguments().size(); ++n) { for (int n = 0; n < var.arguments().size(); ++n) {
LOG(kLOG_DEBUG3) << "argument - " << var.arguments()[n]; LOG(kLOG_DEBUG3) << "argument - " << var.arguments()[n];
} }
} }
for (int y = 0; y < op.outputs_size(); ++y) { for (int y = 0; y < op.outputs_size(); ++y) {
const framework::proto::OpDesc::Var &var = op.outputs(y); const framework::proto::OpDesc::Var &var = op.outputs(y);
LOG(kLOG_DEBUG2) << "out parameter: " << var.parameter(); LOG(kLOG_DEBUG2) << "out parameter: " << var.parameter();
for (int z = 0; z < var.arguments().size(); ++z) { for (int z = 0; z < var.arguments().size(); ++z) {
LOG(kLOG_DEBUG3) << "argument - " << var.arguments()[z]; LOG(kLOG_DEBUG3) << "argument - " << var.arguments()[z];
} }
} }
for (int x = 0; x < op.attrs().size(); ++x) { for (int x = 0; x < op.attrs().size(); ++x) {
const framework::proto::OpDesc_Attr attr = op.attrs()[x]; const framework::proto::OpDesc_Attr attr = op.attrs()[x];
LOG(kLOG_DEBUG2) << "attr name: " << attr.name(); LOG(kLOG_DEBUG2) << "attr name: " << attr.name();
switch (attr.type()) { switch (attr.type()) {
case framework::proto::AttrType::BOOLEAN: case framework::proto::AttrType::BOOLEAN:
LOG(kLOG_DEBUG3) << "boolen: " << attr.b(); LOG(kLOG_DEBUG3) << "boolen: " << attr.b();
break; break;
case framework::proto::AttrType::INT: case framework::proto::AttrType::INT:
LOG(kLOG_DEBUG3) << "int: " << attr.i(); LOG(kLOG_DEBUG3) << "int: " << attr.i();
break; break;
case framework::proto::AttrType::FLOAT: case framework::proto::AttrType::FLOAT:
LOG(kLOG_DEBUG3) << "float: " << attr.f(); LOG(kLOG_DEBUG3) << "float: " << attr.f();
case framework::proto::AttrType::STRING: case framework::proto::AttrType::STRING:
LOG(kLOG_DEBUG3) << "string: " << attr.s(); LOG(kLOG_DEBUG3) << "string: " << attr.s();
case framework::proto::AttrType::BOOLEANS: case framework::proto::AttrType::BOOLEANS:
for (int y = 0; y < attr.bools_size(); ++y) { for (int y = 0; y < attr.bools_size(); ++y) {
LOG(kLOG_DEBUG3) << "bools: " << attr.bools(y); LOG(kLOG_DEBUG3) << "bools: " << attr.bools(y);
} }
case framework::proto::AttrType::LONG: case framework::proto::AttrType::LONG:
LOG(kLOG_DEBUG3) << "long: " << attr.l(); LOG(kLOG_DEBUG3) << "long: " << attr.l();
case framework::proto::AttrType::FLOATS: case framework::proto::AttrType::FLOATS:
for (int y = 0; y < attr.floats_size(); ++y) { for (int y = 0; y < attr.floats_size(); ++y) {
LOG(kLOG_DEBUG3) << "floats: " << attr.floats(y); LOG(kLOG_DEBUG3) << "floats: " << attr.floats(y);
} }
case framework::proto::AttrType::INTS: case framework::proto::AttrType::INTS:
for (int y = 0; y < attr.ints_size(); ++y) { for (int y = 0; y < attr.ints_size(); ++y) {
LOG(kLOG_DEBUG3) << "ints: " << attr.ints(y); LOG(kLOG_DEBUG3) << "ints: " << attr.ints(y);
} }
case framework::proto::AttrType::STRINGS: case framework::proto::AttrType::STRINGS:
for (int y = 0; y < attr.strings_size(); ++y) { for (int y = 0; y < attr.strings_size(); ++y) {
LOG(kLOG_DEBUG3) << "strings: " << attr.strings(y); LOG(kLOG_DEBUG3) << "strings: " << attr.strings(y);
} }
} }
} }
}
for (int k = 0; k < block.vars().size(); ++k) {
framework::proto::VarDesc var = block.vars()[k];
if (var.type().type() == framework::proto::VarType::LOD_TENSOR) {
LOG(kLOG_DEBUG1) << "var name: " << var.name();
const framework::proto::VarType::TensorDesc &tensor_desc =
var.type().lod_tensor().tensor();
LOG(kLOG_DEBUG2) << "in var tensor desc dims size: "
<< tensor_desc.dims().size();
int memory_size = 1;
for (int l = 0; l < tensor_desc.dims().size(); ++l) {
LOG(kLOG_DEBUG3) << "var tensor desc dim " << l
<< " value: " << tensor_desc.dims()[l];
}
}
if (var.persistable() &&
var.type().type() != framework::proto::VarType::FEED_MINIBATCH &&
var.type().type() != framework::proto::VarType::FETCH_LIST) {
// std::cout << " to load " << var.name() <<
// std::endl;
std::string file_path = dirname + "/" + var.name();
std::ifstream is(file_path);
std::streampos pos = is.tellg(); // save current position
is.seekg(0, std::ios::end);
// std::cout << " file length = " << is.tellg() <<
// std::endl;
is.seekg(pos); // restore saved position
// 1. version
uint32_t version;
is.read(reinterpret_cast<char *>(&version), sizeof(version));
// std::cout << " version: " << version <<
// std::endl;
// 2 Lod information
uint64_t lod_level;
is.read(reinterpret_cast<char *>(&lod_level), sizeof(lod_level));
// std::cout << " load level: " << lod_level <<
// std::endl;
// std::cout << " lod info: " << std::endl;
for (uint64_t i = 0; i < lod_level; ++i) {
uint64_t size;
is.read(reinterpret_cast<char *>(&size), sizeof(size));
std::vector<size_t> tmp(size / sizeof(size_t));
is.read(reinterpret_cast<char *>(tmp.data()),
static_cast<std::streamsize>(size));
for (int j = 0; j < tmp.size(); ++j) {
// std::cout << " lod - " << tmp[j] <<
// std::endl;
}
} }
for (int k = 0; k < block.vars().size(); ++k) { uint32_t tensor_version;
framework::proto::VarDesc var = block.vars()[k]; is.read(reinterpret_cast<char *>(&version), sizeof(version));
if (var.type().type() == framework::proto::VarType::LOD_TENSOR) { // std::cout << " tensor_version: " <<
LOG(kLOG_DEBUG1) << "var name: " << var.name(); // tensor_version <<
const framework::proto::VarType::TensorDesc &tensor_desc = // std::endl;
var.type().lod_tensor().tensor();
LOG(kLOG_DEBUG2) << "in var tensor desc dims size: " int32_t size;
<< tensor_desc.dims().size(); is.read(reinterpret_cast<char *>(&size), sizeof(size));
int memory_size = 1; // std::cout << " tensor desc size: " << size <<
for (int l = 0; l < tensor_desc.dims().size(); ++l) { // std::endl;
LOG(kLOG_DEBUG3) << "var tensor desc dim " << l std::unique_ptr<char[]> buf(new char[size]);
<< " value: " << tensor_desc.dims()[l]; is.read(reinterpret_cast<char *>(buf.get()), size);
}
} framework::proto::VarType::TensorDesc desc;
desc.ParseFromArray(buf.get(), size);
if (var.persistable() &&
var.type().type() != // std::cout << " desc dims size " <<
framework::proto::VarType::FEED_MINIBATCH && // desc.dims().size() <<
var.type().type() != framework::proto::VarType::FETCH_LIST) { // std::endl;
// std::cout << " to load " << var.name() << int memory_size = 1;
// std::endl; for (int l = 0; l < desc.dims().size(); ++l) {
std::string file_path = dirname + "/" + var.name(); // std::cout << " dim " << l << " value: "
std::ifstream is(file_path); // <<
std::streampos pos = is.tellg(); // save current position // desc.dims()[l]
is.seekg(0, std::ios::end); // << std::endl;
// std::cout << " file length = " << is.tellg() << memory_size *= desc.dims()[l];
// std::endl;
is.seekg(pos); // restore saved position
// 1. version
uint32_t version;
is.read(reinterpret_cast<char *>(&version), sizeof(version));
// std::cout << " version: " << version <<
// std::endl;
// 2 Lod information
uint64_t lod_level;
is.read(reinterpret_cast<char *>(&lod_level),
sizeof(lod_level));
// std::cout << " load level: " << lod_level <<
// std::endl;
// std::cout << " lod info: " << std::endl;
for (uint64_t i = 0; i < lod_level; ++i) {
uint64_t size;
is.read(reinterpret_cast<char *>(&size), sizeof(size));
std::vector<size_t> tmp(size / sizeof(size_t));
is.read(reinterpret_cast<char *>(tmp.data()),
static_cast<std::streamsize>(size));
for (int j = 0; j < tmp.size(); ++j) {
// std::cout << " lod - " << tmp[j] <<
// std::endl;
}
}
uint32_t tensor_version;
is.read(reinterpret_cast<char *>(&version), sizeof(version));
// std::cout << " tensor_version: " <<
// tensor_version <<
// std::endl;
int32_t size;
is.read(reinterpret_cast<char *>(&size), sizeof(size));
// std::cout << " tensor desc size: " << size <<
// std::endl;
std::unique_ptr<char[]> buf(new char[size]);
is.read(reinterpret_cast<char *>(buf.get()), size);
framework::proto::VarType::TensorDesc desc;
desc.ParseFromArray(buf.get(), size);
// std::cout << " desc dims size " <<
// desc.dims().size() <<
// std::endl;
int memory_size = 1;
for (int l = 0; l < desc.dims().size(); ++l) {
// std::cout << " dim " << l << " value: "
// <<
// desc.dims()[l]
// << std::endl;
memory_size *= desc.dims()[l];
}
int type_size = 0;
// std::cout << " desc pre type: ";
switch (desc.data_type()) {
case framework::proto::VarType::FP16:
// std::cout << "FP16" << std::endl;
type_size = 2;
break;
case framework::proto::VarType::FP32:
type_size = 4;
// std::cout << "FP32" << std::endl;
break;
case framework::proto::VarType::FP64:
type_size = 8;
// std::cout << "FP64" << std::endl;
break;
case framework::proto::VarType::INT32:
type_size = 4;
// std::cout << "INT32" << std::endl;
break;
case framework::proto::VarType::INT64:
type_size = 8;
// std::cout << "INT64" << std::endl;
break;
case framework::proto::VarType::BOOL:
type_size = 1;
// std::cout << "BOOL" << std::endl;
break;
default:
break;
// std::cout << " not support" <<
// std::endl;
}
// std::cout << " malloc size: " << memory_size *
// type_size
// << std::endl;
void *memory = malloc(memory_size * type_size);
is.read(static_cast<char *>(memory), memory_size * type_size);
// std::cout << " memory: " << memory <<
// std::endl;
is.close();
} else {
// std::cout << " *not load "
// << " var : " << var.name() << std::endl;
}
} }
int type_size = 0;
// std::cout << " desc pre type: ";
switch (desc.data_type()) {
case framework::proto::VarType::FP16:
// std::cout << "FP16" << std::endl;
type_size = 2;
break;
case framework::proto::VarType::FP32:
type_size = 4;
// std::cout << "FP32" << std::endl;
break;
case framework::proto::VarType::FP64:
type_size = 8;
// std::cout << "FP64" << std::endl;
break;
case framework::proto::VarType::INT32:
type_size = 4;
// std::cout << "INT32" << std::endl;
break;
case framework::proto::VarType::INT64:
type_size = 8;
// std::cout << "INT64" << std::endl;
break;
case framework::proto::VarType::BOOL:
type_size = 1;
// std::cout << "BOOL" << std::endl;
break;
default:
break;
// std::cout << " not support" <<
// std::endl;
}
// std::cout << " malloc size: " << memory_size *
// type_size
// << std::endl;
void *memory = malloc(memory_size * type_size);
is.read(static_cast<char *>(memory), memory_size * type_size);
// std::cout << " memory: " << memory <<
// std::endl;
is.close();
} else {
// std::cout << " *not load "
// << " var : " << var.name() << std::endl;
}
} }
}
#endif #endif
return program; return program;
} }
template class Loader<CPU, Precision::FP32>; template class Loader<CPU, Precision::FP32>;
......
src/io.h
浏览文件 @ dff4a781
...@@ -29,11 +29,11 @@ namespace paddle_mobile { ...@@ -29,11 +29,11 @@ namespace paddle_mobile {
template <typename Dtype, Precision P = Precision::FP32> template <typename Dtype, Precision P = Precision::FP32>
class Loader : PaddleMobileObject { class Loader : PaddleMobileObject {
public: public:
const framework::Program<Dtype, P> Load(const std::string &dirname); const framework::Program<Dtype, P> Load(const std::string &dirname);
private: private:
void LoadVar(framework::LoDTensor *tensor, const std::string &file_path); void LoadVar(framework::LoDTensor *tensor, const std::string &file_path);
}; };
} // namespace paddle_mobile } // namespace paddle_mobile
src/memory/t_malloc.cc
浏览文件 @ dff4a781
...@@ -26,25 +26,25 @@ namespace memory { ...@@ -26,25 +26,25 @@ namespace memory {
const int MALLOC_ALIGN = 16; const int MALLOC_ALIGN = 16;
void Copy(void *dst, const void *src, size_t num) { void Copy(void *dst, const void *src, size_t num) {
std::memcpy(dst, src, num); std::memcpy(dst, src, num);
}; };
void *Alloc(size_t size) { void *Alloc(size_t size) {
size_t offset = sizeof(void *) + MALLOC_ALIGN - 1; size_t offset = sizeof(void *) + MALLOC_ALIGN - 1;
char *p = static_cast<char *>(malloc(offset + size)); char *p = static_cast<char *>(malloc(offset + size));
if (!p) { if (!p) {
return nullptr; return nullptr;
} }
void *r = reinterpret_cast<void *>(reinterpret_cast<size_t>(p + offset) & void *r = reinterpret_cast<void *>(reinterpret_cast<size_t>(p + offset) &
(~(MALLOC_ALIGN - 1))); (~(MALLOC_ALIGN - 1)));
static_cast<void **>(r)[-1] = p; static_cast<void **>(r)[-1] = p;
return r; return r;
} }
void Free(void *ptr) { void Free(void *ptr) {
if (ptr) { if (ptr) {
free(static_cast<void **>(ptr)[-1]); free(static_cast<void **>(ptr)[-1]);
} }
} }
} // namespace memory } // namespace memory
......
src/memory/t_malloc.h
浏览文件 @ dff4a781
...@@ -38,12 +38,12 @@ void Free(void *ptr); ...@@ -38,12 +38,12 @@ void Free(void *ptr);
* static_cast * static_cast
*/ */
template <typename T> class PODDeleter { template <typename T> class PODDeleter {
static_assert(std::is_pod<T>::value, "T must be POD"); static_assert(std::is_pod<T>::value, "T must be POD");
public: public:
explicit PODDeleter(){}; explicit PODDeleter(){};
void operator()(T *ptr) { Free(static_cast<void *>(ptr)); } void operator()(T *ptr) { Free(static_cast<void *>(ptr)); }
}; };
/** /**
...@@ -55,10 +55,10 @@ template <typename T> class PODDeleter { ...@@ -55,10 +55,10 @@ template <typename T> class PODDeleter {
* reinterpret_cast * reinterpret_cast
*/ */
template <typename T> class PlainDeleter { template <typename T> class PlainDeleter {
public: public:
explicit PlainDeleter(){}; explicit PlainDeleter(){};
void operator()(T *ptr) { Free(reinterpret_cast<void *>(ptr)); } void operator()(T *ptr) { Free(reinterpret_cast<void *>(ptr)); }
}; };
} // namespace memory } // namespace memory
} // namespace paddle_mobile } // namespace paddle_mobile
src/operators/batchnorm_op.cpp
浏览文件 @ dff4a781
...@@ -23,8 +23,8 @@ namespace operators { ...@@ -23,8 +23,8 @@ namespace operators {
template <typename Dtype, typename T> template <typename Dtype, typename T>
void BatchNormOp<Dtype, T>::InferShape() const { void BatchNormOp<Dtype, T>::InferShape() const {
auto x_dims = param_.InputX()->dims(); auto x_dims = param_.InputX()->dims();
param_.OutputY()->Resize(x_dims); param_.OutputY()->Resize(x_dims);
} }
template class BatchNormOp<CPU, float>; template class BatchNormOp<CPU, float>;
} // namespace operators } // namespace operators
......
src/operators/batchnorm_op.h
浏览文件 @ dff4a781
...@@ -27,25 +27,25 @@ using namespace framework; ...@@ -27,25 +27,25 @@ using namespace framework;
template <typename DeviceType, typename T> template <typename DeviceType, typename T>
class BatchNormOp : public framework::OperatorWithKernel<DeviceType> { class BatchNormOp : public framework::OperatorWithKernel<DeviceType> {
public: public:
BatchNormOp(const std::string &type, const VariableNameMap &inputs, BatchNormOp(const std::string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs, const VariableNameMap &outputs,
const framework::AttributeMap attrs, const framework::AttributeMap attrs,
std::shared_ptr<framework::Scope> scope) std::shared_ptr<framework::Scope> scope)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, : framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
attrs, scope), scope),
param_(inputs, outputs, attrs, *scope) {} param_(inputs, outputs, attrs, *scope) {}
void Run() const { void Run() const {
operators::BatchNormKernel<DeviceType, T> kernel; operators::BatchNormKernel<DeviceType, T> kernel;
kernel.Compute(param_); kernel.Compute(param_);
} }
using framework::OperatorWithKernel<DeviceType>::OperatorWithKernel; using framework::OperatorWithKernel<DeviceType>::OperatorWithKernel;
void InferShape() const override; void InferShape() const override;
protected: protected:
BatchNormParam param_; BatchNormParam param_;
}; };
} // namespace operators } // namespace operators
......
src/operators/concat_op.cpp
浏览文件 @ dff4a781
...@@ -23,40 +23,40 @@ namespace operators { ...@@ -23,40 +23,40 @@ namespace operators {
template <typename Dtype, typename T> template <typename Dtype, typename T>
void ConcatOp<Dtype, T>::InferShape() const { void ConcatOp<Dtype, T>::InferShape() const {
auto inputs = param_.Inputs(); auto inputs = param_.Inputs();
const size_t n = inputs.size(); const size_t n = inputs.size();
std::vector<DDim> inputs_dims; std::vector<DDim> inputs_dims;
inputs_dims.reserve(n); inputs_dims.reserve(n);
for (int i = 0; i < n; i++) { for (int i = 0; i < n; i++) {
inputs_dims.push_back(inputs[i]->dims()); inputs_dims.push_back(inputs[i]->dims());
} }
auto axis = static_cast<size_t>(param_.Axis()); auto axis = static_cast<size_t>(param_.Axis());
if (n == 1) { if (n == 1) {
DLOG << "Warning: concat op have only one input, " DLOG << "Warning: concat op have only one input, "
"may waste memory"; "may waste memory";
} }
/// add all dim[axis] and check other dims if equal. /// add all dim[axis] and check other dims if equal.
auto out_dims = inputs_dims[0]; auto out_dims = inputs_dims[0];
int in_zero_dims_size = out_dims.size(); int in_zero_dims_size = out_dims.size();
for (size_t i = 1; i < n; i++) { for (size_t i = 1; i < n; i++) {
for (size_t j = 0; j < in_zero_dims_size; j++) { for (size_t j = 0; j < in_zero_dims_size; j++) {
if (j == axis) { if (j == axis) {
out_dims[axis] += inputs_dims[i][j]; out_dims[axis] += inputs_dims[i][j];
} else { } else {
assert(out_dims[j] == inputs_dims[i][j]); assert(out_dims[j] == inputs_dims[i][j]);
} }
}
} }
}
if (out_dims[axis] < 0) { if (out_dims[axis] < 0) {
out_dims[axis] = -1; out_dims[axis] = -1;
} }
param_.Out()->Resize(out_dims); param_.Out()->Resize(out_dims);
} }
template class ConcatOp<CPU, float>; template class ConcatOp<CPU, float>;
......
src/operators/concat_op.h
浏览文件 @ dff4a781
...@@ -26,25 +26,24 @@ using namespace framework; ...@@ -26,25 +26,24 @@ using namespace framework;
template <typename DeviceType, typename T> template <typename DeviceType, typename T>
class ConcatOp : public framework::OperatorWithKernel<DeviceType> { class ConcatOp : public framework::OperatorWithKernel<DeviceType> {
public: public:
ConcatOp(const std::string &type, const VariableNameMap &inputs, ConcatOp(const std::string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs, const VariableNameMap &outputs, const framework::AttributeMap attrs,
const framework::AttributeMap attrs, std::shared_ptr<framework::Scope> scope)
std::shared_ptr<framework::Scope> scope) : framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, scope),
attrs, scope), param_(inputs, outputs, attrs, *scope) {}
param_(inputs, outputs, attrs, *scope) {}
void Run() const { void Run() const {
operators::ConcatKernel<DeviceType, T> kernel; operators::ConcatKernel<DeviceType, T> kernel;
kernel.Compute(param_); kernel.Compute(param_);
} }
using framework::OperatorWithKernel<DeviceType>::OperatorWithKernel; using framework::OperatorWithKernel<DeviceType>::OperatorWithKernel;
void InferShape() const override; void InferShape() const override;
protected: protected:
ConcatParam param_; ConcatParam param_;
}; };
} // namespace operators } // namespace operators
......
src/operators/conv_op.cpp
浏览文件 @ dff4a781
...@@ -25,48 +25,48 @@ namespace operators { ...@@ -25,48 +25,48 @@ namespace operators {
int ConvOutputSize(int input_size, int filter_size, int dilation, int padding, int ConvOutputSize(int input_size, int filter_size, int dilation, int padding,
int stride) { int stride) {
const int dkernel = dilation * (filter_size - 1) + 1; const int dkernel = dilation * (filter_size - 1) + 1;
int output_size = (input_size + 2 * padding - dkernel) / stride + 1; int output_size = (input_size + 2 * padding - dkernel) / stride + 1;
return output_size; return output_size;
} }
template <typename Dtype, typename T> template <typename Dtype, typename T>
void ConvOp<Dtype, T>::InferShape() const { void ConvOp<Dtype, T>::InferShape() const {
// std::cout << " begin get dims: " << std::endl; // std::cout << " begin get dims: " << std::endl;
auto in_dims = param_.Input()->dims(); auto in_dims = param_.Input()->dims();
// std::cout << " end get in dims: " << std::endl; // std::cout << " end get in dims: " << std::endl;
// std::cout << " in_dims: " << in_dims << std::endl; // std::cout << " in_dims: " << in_dims << std::endl;
// std::cout << " begin get Filter " << std::endl; // std::cout << " begin get Filter " << std::endl;
auto filter_dims = param_.Filter()->dims(); auto filter_dims = param_.Filter()->dims();
// std::cout << " end get Filter " << std::endl; // std::cout << " end get Filter " << std::endl;
// std::cout << " begin get Attrs " << std::endl; // std::cout << " begin get Attrs " << std::endl;
const std::vector<int> &strides = param_.Strides(); const std::vector<int> &strides = param_.Strides();
// std::cout << " end get Attrs " << strides[0] << std::endl; // std::cout << " end get Attrs " << strides[0] << std::endl;
std::vector<int> paddings = param_.Paddings(); std::vector<int> paddings = param_.Paddings();
int groups = param_.Groups(); int groups = param_.Groups();
std::vector<int> dilations = param_.Dilations(); std::vector<int> dilations = param_.Dilations();
std::vector<int64_t> output_shape({in_dims[0], filter_dims[0]}); std::vector<int64_t> output_shape({in_dims[0], filter_dims[0]});
for (size_t i = 0; i < strides.size(); ++i) { for (size_t i = 0; i < strides.size(); ++i) {
output_shape.push_back(ConvOutputSize(in_dims[i + 2], output_shape.push_back(ConvOutputSize(in_dims[i + 2], filter_dims[i + 2],
filter_dims[i + 2], dilations[i], dilations[i], paddings[i],
paddings[i], strides[i])); strides[i]));
} }
framework::DDim ddim = framework::make_ddim(output_shape); framework::DDim ddim = framework::make_ddim(output_shape);
param_.Output()->Resize(ddim); param_.Output()->Resize(ddim);
} }
template class ConvOp<CPU, float>; template class ConvOp<CPU, float>;
......
src/operators/conv_op.h
浏览文件 @ dff4a781
...@@ -28,25 +28,25 @@ using namespace framework; ...@@ -28,25 +28,25 @@ using namespace framework;
template <typename DeviceType, typename T> template <typename DeviceType, typename T>
class ConvOp : public framework::OperatorWithKernel<DeviceType> { class ConvOp : public framework::OperatorWithKernel<DeviceType> {
public: public:
ConvOp(const std::string &type, const VariableNameMap &inputs, ConvOp(const std::string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs, const framework::AttributeMap &attrs, const VariableNameMap &outputs, const framework::AttributeMap &attrs,
std::shared_ptr<framework::Scope> scope) std::shared_ptr<framework::Scope> scope)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, : framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
attrs, scope), scope),
param_(inputs, outputs, attrs, *scope) {} param_(inputs, outputs, attrs, *scope) {}
using framework::OperatorWithKernel<DeviceType>::OperatorWithKernel; using framework::OperatorWithKernel<DeviceType>::OperatorWithKernel;
void InferShape() const override; void InferShape() const override;
void Run() const { void Run() const {
operators::ConvKernel<DeviceType, T> kernel; operators::ConvKernel<DeviceType, T> kernel;
kernel.Compute(param_); kernel.Compute(param_);
this->ClearVariables({"Filter", "Input"}); this->ClearVariables({"Filter", "Input"});
} }
private: private:
ConvParam param_; ConvParam param_;
}; };
} // namespace operators } // namespace operators
......
src/operators/elementwise_add_op.cpp
浏览文件 @ dff4a781
...@@ -23,8 +23,8 @@ namespace operators { ...@@ -23,8 +23,8 @@ namespace operators {
template <typename Dtype, typename T> template <typename Dtype, typename T>
void ElementwiseAddOp<Dtype, T>::InferShape() const { void ElementwiseAddOp<Dtype, T>::InferShape() const {
auto x_dim = param_.InputX()->dims(); auto x_dim = param_.InputX()->dims();
param_.Out()->Resize(x_dim); param_.Out()->Resize(x_dim);
} }
template class ElementwiseAddOp<CPU, float>; template class ElementwiseAddOp<CPU, float>;
} // namespace operators } // namespace operators
......
src/operators/elementwise_add_op.h
浏览文件 @ dff4a781
...@@ -27,25 +27,25 @@ using namespace framework; ...@@ -27,25 +27,25 @@ using namespace framework;
template <typename DeviceType, typename T> template <typename DeviceType, typename T>
class ElementwiseAddOp : public framework::OperatorWithKernel<DeviceType> { class ElementwiseAddOp : public framework::OperatorWithKernel<DeviceType> {
public: public:
ElementwiseAddOp(const std::string &type, const VariableNameMap &inputs, ElementwiseAddOp(const std::string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs, const VariableNameMap &outputs,
const framework::AttributeMap attrs, const framework::AttributeMap attrs,
std::shared_ptr<framework::Scope> scope) std::shared_ptr<framework::Scope> scope)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, : framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
attrs, scope), scope),
param_(inputs, outputs, attrs, *scope) {} param_(inputs, outputs, attrs, *scope) {}
void Run() const { void Run() const {
operators::ElementwiseAddKernel<DeviceType, T> kernel; operators::ElementwiseAddKernel<DeviceType, T> kernel;
kernel.Compute(param_); kernel.Compute(param_);
} }
using framework::OperatorWithKernel<DeviceType>::OperatorWithKernel; using framework::OperatorWithKernel<DeviceType>::OperatorWithKernel;
void InferShape() const override; void InferShape() const override;
protected: protected:
ElementwiseAddParam param_; ElementwiseAddParam param_;
}; };
} // namespace operators } // namespace operators
} // namespace paddle_mobile } // namespace paddle_mobile
src/operators/kernel/arm/batchnorm_kernel.cpp
浏览文件 @ dff4a781
...@@ -21,73 +21,72 @@ namespace operators { ...@@ -21,73 +21,72 @@ namespace operators {
template <> template <>
void BatchNormKernel<CPU, float>::Compute(const BatchNormParam &param) const { void BatchNormKernel<CPU, float>::Compute(const BatchNormParam &param) const {
/// todo: test. /// todo: test.
const Tensor *input_x = param.InputX(); const Tensor *input_x = param.InputX();
auto input_x_ptr = input_x->data<float>(); auto input_x_ptr = input_x->data<float>();
const auto &x_dims = input_x->dims(); const auto &x_dims = input_x->dims();
const int N = x_dims[0]; const int N = x_dims[0];
const int C = x_dims[1]; const int C = x_dims[1];
const int H = x_dims[2]; const int H = x_dims[2];
const int W = x_dims[3]; const int W = x_dims[3];
const int stride0 = C * H * W; const int stride0 = C * H * W;
const int stride1 = H * W; const int stride1 = H * W;
const int stride2 = W; const int stride2 = W;
Tensor *out = param.OutputY(); Tensor *out = param.OutputY();
auto out_ptr = out->mutable_data<float>(); auto out_ptr = out->mutable_data<float>();
const float epsilon = param.Epsilon(); const float epsilon = param.Epsilon();
const Tensor *mean = param.InputMean(); const Tensor *mean = param.InputMean();
const Tensor *variance = param.InputVariance(); const Tensor *variance = param.InputVariance();
const Tensor *scale = param.InputScale(); const Tensor *scale = param.InputScale();
const Tensor *bias = param.InputBias(); const Tensor *bias = param.InputBias();
auto mean_ptr = mean->data<float>(); auto mean_ptr = mean->data<float>();
auto variance_ptr = variance->data<float>(); auto variance_ptr = variance->data<float>();
auto scale_ptr = scale->data<float>(); auto scale_ptr = scale->data<float>();
auto bias_ptr = bias->data<float>(); auto bias_ptr = bias->data<float>();
Tensor inv_std; Tensor inv_std;
auto inv_std_ptr = inv_std.mutable_data<float>(make_ddim({C})); auto inv_std_ptr = inv_std.mutable_data<float>(make_ddim({C}));
if (C != variance->numel()) { if (C != variance->numel()) {
std::cout << "C must equal to variance.numel()" << std::endl; std::cout << "C must equal to variance.numel()" << std::endl;
} }
assert(C == variance->numel()); assert(C == variance->numel());
/// std = (var + epsilon).sqrt(); /// std = (var + epsilon).sqrt();
/// inv_std = 1 / std; /// inv_std = 1 / std;
for (int i = 0; i < C; i++) { for (int i = 0; i < C; i++) {
inv_std_ptr[i] = inv_std_ptr[i] =
1 / static_cast<float>(pow((variance_ptr[i] + epsilon), 0.5)); 1 / static_cast<float>(pow((variance_ptr[i] + epsilon), 0.5));
} }
Tensor new_scale; Tensor new_scale;
auto new_scale_ptr = new_scale.mutable_data<float>(make_ddim({C})); auto new_scale_ptr = new_scale.mutable_data<float>(make_ddim({C}));
Tensor new_bias; Tensor new_bias;
auto new_bias_ptr = new_bias.mutable_data<float>(make_ddim({C})); auto new_bias_ptr = new_bias.mutable_data<float>(make_ddim({C}));
/// ((x - est_mean) * (inv_var) * scale + bias equal to /// ((x - est_mean) * (inv_var) * scale + bias equal to
/// (x * inv_var * scale) + (bias - est_mean * inv_var * scale) /// (x * inv_var * scale) + (bias - est_mean * inv_var * scale)
for (int i = 0; i < C; i++) { for (int i = 0; i < C; i++) {
new_scale_ptr[i] = inv_std_ptr[i] * scale_ptr[i]; new_scale_ptr[i] = inv_std_ptr[i] * scale_ptr[i];
new_bias_ptr[i] = new_bias_ptr[i] = bias_ptr[i] - mean_ptr[i] * inv_std_ptr[i] * scale_ptr[i];
bias_ptr[i] - mean_ptr[i] * inv_std_ptr[i] * scale_ptr[i]; {
{ for (int n = 0; n < N; n++) {
for (int n = 0; n < N; n++) { for (int h = 0; h < H; h++) {
for (int h = 0; h < H; h++) { for (int w = 0; w < W; w++) {
for (int w = 0; w < W; w++) { int index = n * stride0 + i * stride1 + h * stride2 + w;
int index = n * stride0 + i * stride1 + h * stride2 + w; out_ptr[index] =
out_ptr[index] = input_x_ptr[index] * new_scale_ptr[i] + input_x_ptr[index] * new_scale_ptr[i] + new_bias_ptr[i];
new_bias_ptr[i]; }
}
}
}
} }
}
} }
DLOG << "input[2,5,1,0](input[102]) ,channel 5 :"; }
DLOG << "input_x_ptr : " << input_x_ptr[102]; DLOG << "input[2,5,1,0](input[102]) ,channel 5 :";
DLOG << "variance : " << variance_ptr[5]; DLOG << "input_x_ptr : " << input_x_ptr[102];
DLOG << "inv_std_ptr : " << inv_std_ptr[5]; DLOG << "variance : " << variance_ptr[5];
DLOG << "new_scale_ptr : " << new_scale_ptr[5]; DLOG << "inv_std_ptr : " << inv_std_ptr[5];
DLOG << "new_bias_ptr : " << new_bias_ptr[5]; DLOG << "new_scale_ptr : " << new_scale_ptr[5];
DLOG << "out_ptr : " << out_ptr[102]; DLOG << "new_bias_ptr : " << new_bias_ptr[5];
DLOG << "out_ptr : " << out_ptr[102];
} }
} // namespace operators } // namespace operators
} // namespace paddle_mobile } // namespace paddle_mobile
src/operators/kernel/arm/concat_kernel.cpp
浏览文件 @ dff4a781
...@@ -19,36 +19,36 @@ limitations under the License. */ ...@@ -19,36 +19,36 @@ limitations under the License. */
namespace paddle_mobile { namespace paddle_mobile {
namespace operators { namespace operators {
template <typename T> class ConcatFunctor { template <typename T> class ConcatFunctor {
public: public:
void operator()(const std::vector<framework::Tensor> &input, const int axis, void operator()(const std::vector<framework::Tensor> &input, const int axis,
framework::Tensor *output) { framework::Tensor *output) {
size_t num = input.size(); size_t num = input.size();
int rows = 1; int rows = 1;
auto dim_0 = input[0].dims(); auto dim_0 = input[0].dims();
for (int i = 0; i < axis; ++i) { for (int i = 0; i < axis; ++i) {
rows *= dim_0[i]; rows *= dim_0[i];
} }
int out_rows = rows, out_cols = 0; int out_rows = rows, out_cols = 0;
std::vector<int64_t> input_cols(input.size()); std::vector<int64_t> input_cols(input.size());
for (int i = 0; i < num; ++i) { for (int i = 0; i < num; ++i) {
int t_cols = input[i].numel() / rows; int t_cols = input[i].numel() / rows;
out_cols += t_cols; out_cols += t_cols;
input_cols[i] = t_cols; input_cols[i] = t_cols;
} }
// computation // computation
for (int k = 0; k < out_rows; ++k) { for (int k = 0; k < out_rows; ++k) {
T *dst_ptr = output->data<T>() + k * out_cols; T *dst_ptr = output->data<T>() + k * out_cols;
int col_idx = 0; int col_idx = 0;
for (int j = 0; j < num; ++j) { for (int j = 0; j < num; ++j) {
int col_len = input_cols[j]; int col_len = input_cols[j];
const T *src_prt = input[j].data<T>() + k * col_len; const T *src_prt = input[j].data<T>() + k * col_len;
memory::Copy(dst_ptr + col_idx, src_prt, sizeof(T) * col_len); memory::Copy(dst_ptr + col_idx, src_prt, sizeof(T) * col_len);
col_idx += col_len; col_idx += col_len;
} }
}
} }
}
}; };
template <typename T> template <typename T>
void StridedNumelCopyWithAxis(int64_t axis, T *dst, void StridedNumelCopyWithAxis(int64_t axis, T *dst,
...@@ -56,61 +56,60 @@ void StridedNumelCopyWithAxis(int64_t axis, T *dst, ...@@ -56,61 +56,60 @@ void StridedNumelCopyWithAxis(int64_t axis, T *dst,
const T *src, const T *src,
const framework::DDim &src_stride_numel, const framework::DDim &src_stride_numel,
int64_t size) { int64_t size) {
int64_t before = dst_stride_numel[0] / dst_stride_numel[axis]; int64_t before = dst_stride_numel[0] / dst_stride_numel[axis];
int64_t src_after = src_stride_numel[axis]; int64_t src_after = src_stride_numel[axis];
int64_t dst_after = dst_stride_numel[axis]; int64_t dst_after = dst_stride_numel[axis];
///"src and dst tensor should have the same dims size." ///"src and dst tensor should have the same dims size."
assert(src_stride_numel.size() == dst_stride_numel.size()); assert(src_stride_numel.size() == dst_stride_numel.size());
for (int64_t i = 0; i < axis; ++i) { for (int64_t i = 0; i < axis; ++i) {
if (i < axis) { if (i < axis) {
/// src and dst should have the same elements /// src and dst should have the same elements
/// except the specified axis. /// except the specified axis.
assert(src_stride_numel[i] / src_stride_numel[axis] == assert(src_stride_numel[i] / src_stride_numel[axis] ==
dst_stride_numel[i] / dst_stride_numel[axis]); dst_stride_numel[i] / dst_stride_numel[axis]);
} else if (i == axis) { } else if (i == axis) {
continue; continue;
} else { } else {
/// "src and dst should have the same elements " /// "src and dst should have the same elements "
/// "except the specified axis." /// "except the specified axis."
assert(src_stride_numel[i] == dst_stride_numel[i]); assert(src_stride_numel[i] == dst_stride_numel[i]);
}
} }
}
for (int64_t i = 0; i < before; ++i) { for (int64_t i = 0; i < before; ++i) {
memory::Copy(dst + i * dst_after, src + i * src_after, memory::Copy(dst + i * dst_after, src + i * src_after, sizeof(T) * size);
sizeof(T) * size); }
}
} }
template <> template <>
void ConcatKernel<CPU, float>::Compute(const ConcatParam &param) const { void ConcatKernel<CPU, float>::Compute(const ConcatParam &param) const {
auto inputs = param.Inputs(); auto inputs = param.Inputs();
auto *out = param.Out(); auto *out = param.Out();
int64_t axis = param.Axis(); int64_t axis = param.Axis();
out->mutable_data<float>(); out->mutable_data<float>();
/// Sometimes direct copies will be faster, this maybe need deeply analysis. /// Sometimes direct copies will be faster, this maybe need deeply analysis.
if (axis == 0 && inputs.size() < 10) { if (axis == 0 && inputs.size() < 10) {
size_t output_offset = 0; size_t output_offset = 0;
for (auto *in : inputs) { for (auto *in : inputs) {
auto in_stride = framework::stride_numel(in->dims()); auto in_stride = framework::stride_numel(in->dims());
auto out_stride = framework::stride_numel(out->dims()); auto out_stride = framework::stride_numel(out->dims());
StridedNumelCopyWithAxis<float>( StridedNumelCopyWithAxis<float>(axis, out->data<float>() + output_offset,
axis, out->data<float>() + output_offset, out_stride, out_stride, in->data<float>(), in_stride,
in->data<float>(), in_stride, in_stride[axis]); in_stride[axis]);
output_offset += in_stride[axis]; output_offset += in_stride[axis];
} }
} else { } else {
std::vector<framework::Tensor> inputs_concat(inputs.size()); std::vector<framework::Tensor> inputs_concat(inputs.size());
for (int j = 0; j < inputs.size(); ++j) { for (int j = 0; j < inputs.size(); ++j) {
inputs_concat[j] = *inputs[j]; inputs_concat[j] = *inputs[j];
}
ConcatFunctor<float> concat_functor;
concat_functor(inputs_concat, static_cast<int>(axis), out);
} }
ConcatFunctor<float> concat_functor;
concat_functor(inputs_concat, static_cast<int>(axis), out);
}
} }
} // namespace operators } // namespace operators
......
src/operators/kernel/arm/conv_kernel.cpp
浏览文件 @ dff4a781
...@@ -24,128 +24,126 @@ namespace operators { ...@@ -24,128 +24,126 @@ namespace operators {
bool IsExpand(const std::vector<int64_t> &filter_dim, bool IsExpand(const std::vector<int64_t> &filter_dim,
const std::vector<int> &strides, const std::vector<int> &paddings, const std::vector<int> &strides, const std::vector<int> &paddings,
const std::vector<int> &dilations) { const std::vector<int> &dilations) {
bool filter_1 = true, strides_1 = true, padding_0 = true, dilation_1 = true; bool filter_1 = true, strides_1 = true, padding_0 = true, dilation_1 = true;
for (size_t j = 0; j < strides.size(); ++j) { for (size_t j = 0; j < strides.size(); ++j) {
filter_1 = filter_1 && (static_cast<int>(filter_dim[j + 2]) == 1); filter_1 = filter_1 && (static_cast<int>(filter_dim[j + 2]) == 1);
strides_1 = strides_1 && (strides[j] == 1); strides_1 = strides_1 && (strides[j] == 1);
padding_0 = padding_0 && (paddings[j] == 0); padding_0 = padding_0 && (paddings[j] == 0);
dilation_1 = dilation_1 && (dilations[j] == 1); dilation_1 = dilation_1 && (dilations[j] == 1);
} }
return !(filter_1 && strides_1 && padding_0 && dilation_1); return !(filter_1 && strides_1 && padding_0 && dilation_1);
} }
template <> void ConvKernel<CPU, float>::Compute(const ConvParam &param) const { template <> void ConvKernel<CPU, float>::Compute(const ConvParam &param) const {
LOG(kLOG_DEBUG) << param; LOG(kLOG_DEBUG) << param;
const Tensor *input = param.Input(); const Tensor *input = param.Input();
// The filter will be reshaped in the calculations, // The filter will be reshaped in the calculations,
// so here use an assignment operation, // so here use an assignment operation,
// that avoids modifying the variable in the Scope. // that avoids modifying the variable in the Scope.
Tensor filter = *param.Filter(); Tensor filter = *param.Filter();
Tensor *output = param.Output(); Tensor *output = param.Output();
// output->mutable_data<T>(context.GetPlace()); // output->mutable_data<T>(context.GetPlace());
int groups = param.Groups(); int groups = param.Groups();
std::vector<int> strides = param.Strides(); std::vector<int> strides = param.Strides();
std::vector<int> paddings = param.Paddings(); std::vector<int> paddings = param.Paddings();
std::vector<int> dilations = param.Dilations(); std::vector<int> dilations = param.Dilations();
DLOG << " compute end get Attrs " << strides[0]; DLOG << " compute end get Attrs " << strides[0];
const int batch_size = static_cast<int>(input->dims()[0]); const int batch_size = static_cast<int>(input->dims()[0]);
// filter_shape_vec: {k_o, k_i, k_h, k_w} or {k_o, k_i, k_d, k_h, // filter_shape_vec: {k_o, k_i, k_h, k_w} or {k_o, k_i, k_d, k_h,
// k_w} // k_w}
std::vector<int64_t> filter_shape_vec(framework::vectorize(filter.dims())); std::vector<int64_t> filter_shape_vec(framework::vectorize(filter.dims()));
// output_shape_vec: {o_n, o_c, o_h, o_w} or {o_n, o_c, o_d, o_h, // output_shape_vec: {o_n, o_c, o_h, o_w} or {o_n, o_c, o_d, o_h,
// o_w} // o_w}
std::vector<int64_t> output_shape_vec(framework::vectorize(output->dims())); std::vector<int64_t> output_shape_vec(framework::vectorize(output->dims()));
// use col_shape in the im2col calculation // use col_shape in the im2col calculation
// col_shape_vec: {i_c/g, k_h, k_w, o_h, o_w} or {i_c/g, k_d, k_h, // col_shape_vec: {i_c/g, k_h, k_w, o_h, o_w} or {i_c/g, k_d, k_h,
// k_w, o_d, // k_w, o_d,
// o_h, o_w} // o_h, o_w}
size_t data_dim = filter_shape_vec.size() - 2; size_t data_dim = filter_shape_vec.size() - 2;
std::vector<int64_t> col_shape_vec(1 + 2 * data_dim); std::vector<int64_t> col_shape_vec(1 + 2 * data_dim);
col_shape_vec[0] = input->dims()[1] / groups; col_shape_vec[0] = input->dims()[1] / groups;
for (size_t j = 0; j < data_dim; ++j) { for (size_t j = 0; j < data_dim; ++j) {
col_shape_vec[j + 1] = filter_shape_vec[j + 2]; col_shape_vec[j + 1] = filter_shape_vec[j + 2];
col_shape_vec[j + 1 + data_dim] = output_shape_vec[j + 2]; col_shape_vec[j + 1 + data_dim] = output_shape_vec[j + 2];
} }
framework::DDim col_shape(framework::make_ddim(col_shape_vec)); framework::DDim col_shape(framework::make_ddim(col_shape_vec));
// use col_matrix_shape in the gemm calculation // use col_matrix_shape in the gemm calculation
// size: (i_c/g * k_h * k_w, o_h * o_w) or (i_c/g * k_d * k_h * k_w, // size: (i_c/g * k_h * k_w, o_h * o_w) or (i_c/g * k_d * k_h * k_w,
// o_d * // o_d *
// o_h * o_w) // o_h * o_w)
framework::DDim col_matrix_shape = framework::DDim col_matrix_shape =
framework::flatten_to_2d(col_shape, data_dim + 1); framework::flatten_to_2d(col_shape, data_dim + 1);
bool is_expand = IsExpand(filter_shape_vec, strides, paddings, dilations); bool is_expand = IsExpand(filter_shape_vec, strides, paddings, dilations);
Tensor col; Tensor col;
// col_matrix shares the same piece of data with col, // col_matrix shares the same piece of data with col,
// but will be reshaped into a two-dimensional matrix shape // but will be reshaped into a two-dimensional matrix shape
// to call the matrix multiplication interface. // to call the matrix multiplication interface.
Tensor col_matrix; Tensor col_matrix;
if (is_expand) { if (is_expand) {
col.mutable_data<float>(col_shape); col.mutable_data<float>(col_shape);
col_matrix.ShareDataWith(col);
col_matrix.Resize(col_matrix_shape);
}
framework::DDim input_shape = framework::slice_ddim(
input->dims(), 1, static_cast<int>(input->dims().size()));
framework::DDim filter_matrix_shape = {filter.dims()[0],
filter.numel() / filter.dims()[0]};
filter.Resize(filter_matrix_shape);
framework::DDim output_matrix_shape = {
output->dims()[1],
output->numel() / (output->dims()[0] * output->dims()[1])};
// convolution operator: im2col(or vol2col) + gemm
int in_step = static_cast<int>(input->dims()[1]) / groups;
int out_step = static_cast<int>(output->dims()[1]) / groups;
math::Vol2ColFunctor<CPU, float> vol2col;
math::Im2ColFunctor<math::ColFormat::kCFO, CPU, float> im2col;
// auto& dev_ctx = context.template
// device_context<DeviceContext>();
for (int i = 0; i < batch_size; i++) {
Tensor in_batch = input->Slice(i, i + 1).Resize(input_shape);
Tensor out_batch = output->Slice(i, i + 1).Resize(output_matrix_shape);
for (int g = 0; g < groups; g++) {
Tensor in_slice = in_batch.Slice(g * in_step, (g + 1) * in_step);
if (!is_expand) {
col.ShareDataWith(in_slice);
col_matrix.ShareDataWith(col); col_matrix.ShareDataWith(col);
col_matrix.Resize(col_matrix_shape); col_matrix.Resize(col_matrix_shape);
} else if (data_dim == 2U) {
// im2col
im2col(in_slice, dilations, strides,
std::vector<int>{paddings[0], paddings[1], paddings[0],
paddings[1]},
&col);
} else if (data_dim == 3U) {
// vol2col
vol2col(in_slice, dilations, strides, paddings, &col);
}
// gemm
Tensor out_slice = out_batch.Slice(g * out_step, (g + 1) * out_step);
Tensor filter_slice = filter.Slice(g * out_step, (g + 1) * out_step);
math::matmul<float>(filter_slice, false, col_matrix, false, float(1.0),
&out_slice, float(0.0));
} }
}
framework::DDim input_shape = framework::slice_ddim(
input->dims(), 1, static_cast<int>(input->dims().size()));
framework::DDim filter_matrix_shape = {filter.dims()[0],
filter.numel() / filter.dims()[0]};
filter.Resize(filter_matrix_shape);
framework::DDim output_matrix_shape = {
output->dims()[1],
output->numel() / (output->dims()[0] * output->dims()[1])};
// convolution operator: im2col(or vol2col) + gemm
int in_step = static_cast<int>(input->dims()[1]) / groups;
int out_step = static_cast<int>(output->dims()[1]) / groups;
math::Vol2ColFunctor<CPU, float> vol2col;
math::Im2ColFunctor<math::ColFormat::kCFO, CPU, float> im2col;
// auto& dev_ctx = context.template
// device_context<DeviceContext>();
for (int i = 0; i < batch_size; i++) {
Tensor in_batch = input->Slice(i, i + 1).Resize(input_shape);
Tensor out_batch = output->Slice(i, i + 1).Resize(output_matrix_shape);
for (int g = 0; g < groups; g++) {
Tensor in_slice = in_batch.Slice(g * in_step, (g + 1) * in_step);
if (!is_expand) {
col.ShareDataWith(in_slice);
col_matrix.ShareDataWith(col);
col_matrix.Resize(col_matrix_shape);
} else if (data_dim == 2U) {
// im2col
im2col(in_slice, dilations, strides,
std::vector<int>{paddings[0], paddings[1], paddings[0],
paddings[1]},
&col);
} else if (data_dim == 3U) {
// vol2col
vol2col(in_slice, dilations, strides, paddings, &col);
}
// gemm
Tensor out_slice =
out_batch.Slice(g * out_step, (g + 1) * out_step);
Tensor filter_slice =
filter.Slice(g * out_step, (g + 1) * out_step);
math::matmul<float>(filter_slice, false, col_matrix, false,
float(1.0), &out_slice, float(0.0));
}
}
} }
template class ConvKernel<CPU, float>; template class ConvKernel<CPU, float>;
......
src/operators/kernel/arm/elementwise_add_kernel.cpp
浏览文件 @ dff4a781
...@@ -20,19 +20,19 @@ namespace paddle_mobile { ...@@ -20,19 +20,19 @@ namespace paddle_mobile {
namespace operators { namespace operators {
template <typename T> struct AddFunctor { template <typename T> struct AddFunctor {
inline T operator()(T a, T b) const { return a + b; } inline T operator()(T a, T b) const { return a + b; }
}; };
template <> template <>
void ElementwiseAddKernel<CPU, float>::Compute( void ElementwiseAddKernel<CPU, float>::Compute(
const ElementwiseAddParam &param) const { const ElementwiseAddParam &param) const {
const Tensor *input_x = param.InputX(); const Tensor *input_x = param.InputX();
const Tensor *input_y = param.InputY(); const Tensor *input_y = param.InputY();
Tensor *Out = param.Out(); Tensor *Out = param.Out();
Out->mutable_data<float>(); Out->mutable_data<float>();
const int axis = param.Axis(); const int axis = param.Axis();
ElementwiseComputeEx<AddFunctor<float>, float>(input_x, input_y, axis, ElementwiseComputeEx<AddFunctor<float>, float>(input_x, input_y, axis,
AddFunctor<float>(), Out); AddFunctor<float>(), Out);
} }
template class ElementwiseAddKernel<CPU, float>; template class ElementwiseAddKernel<CPU, float>;
......
src/operators/kernel/arm/lrn_kernel.cpp
浏览文件 @ dff4a781
...@@ -24,21 +24,21 @@ namespace paddle_mobile { ...@@ -24,21 +24,21 @@ namespace paddle_mobile {
namespace operators { namespace operators {
template <> void LrnKernel<CPU, float>::Compute(const LrnParam &param) const { template <> void LrnKernel<CPU, float>::Compute(const LrnParam &param) const {
const Tensor *input_x = param.InputX(); const Tensor *input_x = param.InputX();
auto x_dims = input_x->dims(); auto x_dims = input_x->dims();
/// data_format = NCHW /// data_format = NCHW
const int N = x_dims[0]; const int N = x_dims[0];
const int C = x_dims[1]; const int C = x_dims[1];
const int H = x_dims[2]; const int H = x_dims[2];
const int W = x_dims[3]; const int W = x_dims[3];
Tensor *out = param.Out(); Tensor *out = param.Out();
out->mutable_data<float>(); out->mutable_data<float>();
const int n = param.N(); const int n = param.N();
const float alpha = param.Alpha(); const float alpha = param.Alpha();
const float beta = param.Beta(); const float beta = param.Beta();
const float k = param.K(); const float k = param.K();
LRNFunctor<float> lrnFunctor; LRNFunctor<float> lrnFunctor;
lrnFunctor(*input_x, out, N, C, H, W, n, k, alpha, beta); lrnFunctor(*input_x, out, N, C, H, W, n, k, alpha, beta);
} }
template class LrnKernel<CPU, float>; template class LrnKernel<CPU, float>;
......
src/operators/kernel/arm/mul_kernel.cpp
浏览文件 @ dff4a781
...@@ -24,27 +24,27 @@ namespace paddle_mobile { ...@@ -24,27 +24,27 @@ namespace paddle_mobile {
namespace operators { namespace operators {
template <> void MulKernel<CPU, float>::Compute(const MulParam &param) const { template <> void MulKernel<CPU, float>::Compute(const MulParam &param) const {
const Tensor *input_x = param.InputX(); const Tensor *input_x = param.InputX();
const Tensor *input_y = param.InputY(); const Tensor *input_y = param.InputY();
Tensor *out = param.Out(); Tensor *out = param.Out();
out->mutable_data<float>(); out->mutable_data<float>();
const Tensor x_matrix = const Tensor x_matrix =
input_x->dims().size() > 2 input_x->dims().size() > 2
? framework::ReshapeToMatrix(*input_x, param.XNumColDims()) ? framework::ReshapeToMatrix(*input_x, param.XNumColDims())
: *input_x; : *input_x;
const Tensor y_matrix = const Tensor y_matrix =
input_y->dims().size() > 2 input_y->dims().size() > 2
? framework::ReshapeToMatrix(*input_y, param.YNumColDims()) ? framework::ReshapeToMatrix(*input_y, param.YNumColDims())
: *input_y; : *input_y;
auto out_dim = out->dims(); auto out_dim = out->dims();
if (out_dim.size() != 2) { if (out_dim.size() != 2) {
out->Resize({x_matrix.dims()[0], y_matrix.dims()[1]}); out->Resize({x_matrix.dims()[0], y_matrix.dims()[1]});
} }
math::matmul<float>(x_matrix, false, y_matrix, false, static_cast<float>(1), math::matmul<float>(x_matrix, false, y_matrix, false, static_cast<float>(1),
out, static_cast<float>(0)); out, static_cast<float>(0));
if (out_dim.size() != 2) { if (out_dim.size() != 2) {
out->Resize(out_dim); out->Resize(out_dim);
} }
} }
template class MulKernel<CPU, float>; template class MulKernel<CPU, float>;
......
src/operators/kernel/arm/pool_kernel.cpp
浏览文件 @ dff4a781
...@@ -24,54 +24,54 @@ namespace operators { ...@@ -24,54 +24,54 @@ namespace operators {
inline void PoolBasic(std::string pooling_type, std::vector<int> ksize, inline void PoolBasic(std::string pooling_type, std::vector<int> ksize,
std::vector<int> strides, std::vector<int> paddings, std::vector<int> strides, std::vector<int> paddings,
const Tensor *in_x, Tensor *out) { const Tensor *in_x, Tensor *out) {
if (pooling_type == "max") { if (pooling_type == "max") {
math::PoolFunctor<CPU, math::MaxPool<float>, float> pool2d_forward; math::PoolFunctor<CPU, math::MaxPool<float>, float> pool2d_forward;
math::MaxPool<float> pool_process; math::MaxPool<float> pool_process;
pool2d_forward(*in_x, ksize, strides, paddings, pool_process, out); pool2d_forward(*in_x, ksize, strides, paddings, pool_process, out);
} else if (pooling_type == "avg") { } else if (pooling_type == "avg") {
math::PoolFunctor<CPU, math::AvgPool<float>, float> pool2d_forward; math::PoolFunctor<CPU, math::AvgPool<float>, float> pool2d_forward;
math::AvgPool<float> pool_process; math::AvgPool<float> pool_process;
pool2d_forward(*in_x, ksize, strides, paddings, pool_process, out); pool2d_forward(*in_x, ksize, strides, paddings, pool_process, out);
} }
} }
template <> void PoolKernel<CPU, float>::Compute(const PoolParam &param) const { template <> void PoolKernel<CPU, float>::Compute(const PoolParam &param) const {
const Tensor *in_x = param.Input(); const Tensor *in_x = param.Input();
Tensor *out = param.Output(); Tensor *out = param.Output();
std::string pooling_type = param.PoolingType(); std::string pooling_type = param.PoolingType();
std::vector<int> ksize = param.Ksize(); std::vector<int> ksize = param.Ksize();
std::vector<int> strides = param.Strides(); std::vector<int> strides = param.Strides();
std::vector<int> paddings = param.Paddings(); std::vector<int> paddings = param.Paddings();
if (ksize.size() != 2) { if (ksize.size() != 2) {
LOG(paddle_mobile::LogLevel::kLOG_ERROR) LOG(paddle_mobile::LogLevel::kLOG_ERROR)
<< "Pool op only supports 2D and 3D input."; << "Pool op only supports 2D and 3D input.";
} }
if (param.isGlobalPooling()) { if (param.isGlobalPooling()) {
for (size_t i = 0; i < ksize.size(); ++i) { for (size_t i = 0; i < ksize.size(); ++i) {
paddings[i] = 0; paddings[i] = 0;
ksize[i] = static_cast<int>(in_x->dims()[i + 2]); ksize[i] = static_cast<int>(in_x->dims()[i + 2]);
}
} }
}
PoolBasic(pooling_type, ksize, strides, paddings, in_x, out); PoolBasic(pooling_type, ksize, strides, paddings, in_x, out);
// if (param.isGlobalPooling() || ksize[0] != ksize[1] || // if (param.isGlobalPooling() || ksize[0] != ksize[1] ||
// strides[0] != strides[1] || strides[1] != 2 || // strides[0] != strides[1] || strides[1] != 2 ||
// paddings[0] != paddings[1] || paddings[1] > 1) { // paddings[0] != paddings[1] || paddings[1] > 1) {
// PoolBasic(pooling_type, ksize, strides, paddings, in_x, out); // PoolBasic(pooling_type, ksize, strides, paddings, in_x, out);
// //
// } else if (ksize[0] == 2) { // } else if (ksize[0] == 2) {
// //
// } else if (ksize[0] == 3) { // } else if (ksize[0] == 3) {
// //
// } else { // } else {
// PoolBasic(pooling_type, ksize, strides, paddings, in_x, out); // PoolBasic(pooling_type, ksize, strides, paddings, in_x, out);
// } // }
} }
} // namespace operators } // namespace operators
} // namespace paddle_mobile } // namespace paddle_mobile
src/operators/kernel/batchnorm_kernel.h
浏览文件 @ dff4a781
...@@ -28,8 +28,8 @@ using namespace framework; ...@@ -28,8 +28,8 @@ using namespace framework;
template <typename DeviceType, typename T> template <typename DeviceType, typename T>
class BatchNormKernel class BatchNormKernel
: public framework::OpKernelBase<DeviceType, BatchNormParam> { : public framework::OpKernelBase<DeviceType, BatchNormParam> {
public: public:
void Compute(const BatchNormParam &param) const; void Compute(const BatchNormParam &param) const;
}; };
} // namespace operators } // namespace operators
......
src/operators/kernel/concat_kernel.h
浏览文件 @ dff4a781
...@@ -26,8 +26,8 @@ using namespace framework; ...@@ -26,8 +26,8 @@ using namespace framework;
template <typename DeviceType, typename T> template <typename DeviceType, typename T>
class ConcatKernel : public framework::OpKernelBase<DeviceType, ConcatParam> { class ConcatKernel : public framework::OpKernelBase<DeviceType, ConcatParam> {
public: public:
void Compute(const ConcatParam &param) const; void Compute(const ConcatParam &param) const;
}; };
} // namespace operators } // namespace operators
......
src/operators/kernel/conv_kernel.h
浏览文件 @ dff4a781
...@@ -31,8 +31,8 @@ using namespace framework; ...@@ -31,8 +31,8 @@ using namespace framework;
template <typename DeviceType, typename T> template <typename DeviceType, typename T>
class ConvKernel : public framework::OpKernelBase<DeviceType, ConvParam> { class ConvKernel : public framework::OpKernelBase<DeviceType, ConvParam> {
public: public:
void Compute(const ConvParam &param) const; void Compute(const ConvParam &param) const;
}; };
} // namespace operators } // namespace operators
} // namespace paddle_mobile } // namespace paddle_mobile
src/operators/kernel/elementwise_add_kernel.h
浏览文件 @ dff4a781
...@@ -29,8 +29,8 @@ using namespace framework; ...@@ -29,8 +29,8 @@ using namespace framework;
template <typename DeviceType, typename T> template <typename DeviceType, typename T>
class ElementwiseAddKernel class ElementwiseAddKernel
: public framework::OpKernelBase<DeviceType, ElementwiseAddParam> { : public framework::OpKernelBase<DeviceType, ElementwiseAddParam> {
public: public:
void Compute(const ElementwiseAddParam &param) const; void Compute(const ElementwiseAddParam &param) const;
}; };
} // namespace operators } // namespace operators
} // namespace paddle_mobile } // namespace paddle_mobile
src/operators/kernel/lrn_kernel.h
浏览文件 @ dff4a781
...@@ -26,52 +26,49 @@ namespace operators { ...@@ -26,52 +26,49 @@ namespace operators {
using namespace framework; using namespace framework;
template <typename T> struct LRNFunctor { template <typename T> struct LRNFunctor {
void operator()(const framework::Tensor &input, framework::Tensor *out, void operator()(const framework::Tensor &input, framework::Tensor *out, int N,
int N, int C, int H, int W, int n, T k, T alpha, T beta) { int C, int H, int W, int n, T k, T alpha, T beta) {
auto input_ptr = input.data<T>(); auto input_ptr = input.data<T>();
const int start = -(n - 1) / 2; const int start = -(n - 1) / 2;
const int end = start + n; const int end = start + n;
const int stride0 = C * H * W; const int stride0 = C * H * W;
const int stride1 = H * W; const int stride1 = H * W;
const int stride2 = W; const int stride2 = W;
const int stride3 = 1; const int stride3 = 1;
framework::Tensor sqr_buffer; framework::Tensor sqr_buffer;
auto sqr_buffer_ptr = sqr_buffer.mutable_data<T>(input.dims()); auto sqr_buffer_ptr = sqr_buffer.mutable_data<T>(input.dims());
std::fill(sqr_buffer_ptr, sqr_buffer_ptr + sqr_buffer.numel(), k); std::fill(sqr_buffer_ptr, sqr_buffer_ptr + sqr_buffer.numel(), k);
for (int a = 0; a < N; a++) { for (int a = 0; a < N; a++) {
for (int b = 0; b < C; b++) { for (int b = 0; b < C; b++) {
for (int index = start; index < end; index++) { for (int index = start; index < end; index++) {
int channel = b + index; int channel = b + index;
if (channel >= 0 && channel < C) { if (channel >= 0 && channel < C) {
for (int c = 0; c < H; c++) { for (int c = 0; c < H; c++) {
for (int d = 0; d < W; d++) { for (int d = 0; d < W; d++) {
int u = int u = a * stride0 + b * stride1 + c * stride2 + d;
a * stride0 + b * stride1 + c * stride2 + d;
int i = a * stride0 + channel * stride1 + int i = a * stride0 + channel * stride1 + c * stride2 + d;
c * stride2 + d;
sqr_buffer_ptr[u] += sqr_buffer_ptr[u] += alpha * input_ptr[i] * input_ptr[i];
alpha * input_ptr[i] * input_ptr[i]; }
}
}
}
}
} }
}
} }
auto out_ptr = out->data<T>(); }
for (int i = 0; i < input.numel(); i++) { }
out_ptr[i] = input_ptr[i] / pow(sqr_buffer_ptr[i], beta); auto out_ptr = out->data<T>();
} for (int i = 0; i < input.numel(); i++) {
out_ptr[i] = input_ptr[i] / pow(sqr_buffer_ptr[i], beta);
} }
}
}; };
template <typename DeviceType, typename T> template <typename DeviceType, typename T>
class LrnKernel : public framework::OpKernelBase<DeviceType, LrnParam> { class LrnKernel : public framework::OpKernelBase<DeviceType, LrnParam> {
public: public:
void Compute(const LrnParam &param) const; void Compute(const LrnParam &param) const;
}; };
} // namespace operators } // namespace operators
} // namespace paddle_mobile } // namespace paddle_mobile
src/operators/kernel/mul_kernel.h
浏览文件 @ dff4a781
...@@ -28,8 +28,8 @@ using namespace framework; ...@@ -28,8 +28,8 @@ using namespace framework;
template <typename DeviceType, typename T> template <typename DeviceType, typename T>
class MulKernel : public framework::OpKernelBase<DeviceType, MulParam> { class MulKernel : public framework::OpKernelBase<DeviceType, MulParam> {
public: public:
void Compute(const MulParam &param) const; void Compute(const MulParam &param) const;
}; };
} // namespace operators } // namespace operators
} // namespace paddle_mobile } // namespace paddle_mobile
src/operators/kernel/pool_kernel.h
浏览文件 @ dff4a781
...@@ -28,8 +28,8 @@ using namespace framework; ...@@ -28,8 +28,8 @@ using namespace framework;
template <typename DeviceType, typename T> template <typename DeviceType, typename T>
class PoolKernel : public framework::OpKernelBase<DeviceType, PoolParam> { class PoolKernel : public framework::OpKernelBase<DeviceType, PoolParam> {
public: public:
void Compute(const PoolParam &param) const; void Compute(const PoolParam &param) const;
}; };
} // namespace operators } // namespace operators
} // namespace paddle_mobile } // namespace paddle_mobile
src/operators/lrn_op.cpp
浏览文件 @ dff4a781
...@@ -22,8 +22,8 @@ namespace paddle_mobile { ...@@ -22,8 +22,8 @@ namespace paddle_mobile {
namespace operators { namespace operators {
template <typename Dtype, typename T> void LrnOp<Dtype, T>::InferShape() const { template <typename Dtype, typename T> void LrnOp<Dtype, T>::InferShape() const {
auto x_dims = param_.InputX()->dims(); auto x_dims = param_.InputX()->dims();
param_.Out()->Resize(x_dims); param_.Out()->Resize(x_dims);
} }
template class LrnOp<CPU, float>; template class LrnOp<CPU, float>;
} // namespace operators } // namespace operators
......
src/operators/lrn_op.h
浏览文件 @ dff4a781
...@@ -27,24 +27,24 @@ using namespace framework; ...@@ -27,24 +27,24 @@ using namespace framework;
template <typename DeviceType, typename T> template <typename DeviceType, typename T>
class LrnOp : public framework::OperatorWithKernel<DeviceType> { class LrnOp : public framework::OperatorWithKernel<DeviceType> {
public: public:
LrnOp(const std::string &type, const VariableNameMap &inputs, LrnOp(const std::string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs, const framework::AttributeMap attrs, const VariableNameMap &outputs, const framework::AttributeMap attrs,
std::shared_ptr<framework::Scope> scope) std::shared_ptr<framework::Scope> scope)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, : framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
attrs, scope), scope),
param_(inputs, outputs, attrs, *scope) {} param_(inputs, outputs, attrs, *scope) {}
void Run() const { void Run() const {
operators::LrnKernel<DeviceType, T> kernel; operators::LrnKernel<DeviceType, T> kernel;
kernel.Compute(param_); kernel.Compute(param_);
} }
using framework::OperatorWithKernel<DeviceType>::OperatorWithKernel; using framework::OperatorWithKernel<DeviceType>::OperatorWithKernel;
void InferShape() const override; void InferShape() const override;
protected: protected:
LrnParam param_; LrnParam param_;
}; };
} // namespace operators } // namespace operators
......
src/operators/math/elementwise_op_function.h
浏览文件 @ dff4a781
...@@ -34,66 +34,66 @@ namespace operators { ...@@ -34,66 +34,66 @@ namespace operators {
inline void get_mid_dims(const framework::DDim &x_dims, inline void get_mid_dims(const framework::DDim &x_dims,
const framework::DDim &y_dims, const int axis, const framework::DDim &y_dims, const int axis,
int *pre, int *n, int *post) { int *pre, int *n, int *post) {
*pre = 1; *pre = 1;
*n = 1; *n = 1;
*post = 1; *post = 1;
// compute pre // compute pre
for (int i = 0; i < axis; ++i) { for (int i = 0; i < axis; ++i) {
(*pre) *= x_dims[i]; (*pre) *= x_dims[i];
} }
for (int i = 0; i < y_dims.size(); ++i) { for (int i = 0; i < y_dims.size(); ++i) {
assert(x_dims[i + axis] == y_dims[i]); assert(x_dims[i + axis] == y_dims[i]);
/// "Broadcast dimension mismatch."); /// "Broadcast dimension mismatch.");
(*n) *= y_dims[i]; (*n) *= y_dims[i];
} }
for (int i = axis + y_dims.size(); i < x_dims.size(); ++i) { for (int i = axis + y_dims.size(); i < x_dims.size(); ++i) {
(*post) *= x_dims[i]; (*post) *= x_dims[i];
} }
} }
/// remove dims tail 1. (4,20,1,1) -> (4,20) /// remove dims tail 1. (4,20,1,1) -> (4,20)
inline void trim_trailing_singular_dims(framework::DDim *dims) { inline void trim_trailing_singular_dims(framework::DDim *dims) {
// Remove trailing dimensions of size 1 for y // Remove trailing dimensions of size 1 for y
auto actual_dims_size = dims->size(); auto actual_dims_size = dims->size();
for (; actual_dims_size != 0; --actual_dims_size) { for (; actual_dims_size != 0; --actual_dims_size) {
if ((*dims)[actual_dims_size - 1] != 1) if ((*dims)[actual_dims_size - 1] != 1)
break; break;
} }
if (actual_dims_size != dims->size()) { if (actual_dims_size != dims->size()) {
auto actual_dims = framework::vectorize(*dims); auto actual_dims = framework::vectorize(*dims);
actual_dims.resize(actual_dims_size); actual_dims.resize(actual_dims_size);
*dims = framework::make_ddim(actual_dims); *dims = framework::make_ddim(actual_dims);
} }
} }
template <typename T> class RowwiseTransformIterator { template <typename T> class RowwiseTransformIterator {
public: public:
RowwiseTransformIterator(const T *ptr, int n) : ptr_(ptr), i_(0), n_(n) {} RowwiseTransformIterator(const T *ptr, int n) : ptr_(ptr), i_(0), n_(n) {}
RowwiseTransformIterator<T> &operator++() {
++i_;
if (UNLIKELY(i_ == n_)) {
i_ = 0;
}
return *this;
}
bool operator==(const RowwiseTransformIterator<T> &rhs) const { RowwiseTransformIterator<T> &operator++() {
return (ptr_ + i_) == &(*rhs); ++i_;
if (UNLIKELY(i_ == n_)) {
i_ = 0;
} }
return *this;
}
bool operator!=(const RowwiseTransformIterator<T> &rhs) const { bool operator==(const RowwiseTransformIterator<T> &rhs) const {
return (ptr_ + i_) != &(*rhs); return (ptr_ + i_) == &(*rhs);
} }
const T &operator*() { return ptr_[i_]; } bool operator!=(const RowwiseTransformIterator<T> &rhs) const {
return (ptr_ + i_) != &(*rhs);
}
private: const T &operator*() { return ptr_[i_]; }
const T *ptr_;
int i_; private:
int64_t n_; const T *ptr_;
int i_;
int64_t n_;
}; };
/// (4,20,2)+(20,): (20,) just as (20,1), when move 2 strides in last /// (4,20,2)+(20,): (20,) just as (20,1), when move 2 strides in last
...@@ -101,106 +101,105 @@ template <typename T> class RowwiseTransformIterator { ...@@ -101,106 +101,105 @@ template <typename T> class RowwiseTransformIterator {
/// in (4,20,2) is 2 , /// in (4,20,2) is 2 ,
/// (20,1) move 1 stride , to fill(add) 2 element with the same number. /// (20,1) move 1 stride , to fill(add) 2 element with the same number.
template <typename T> class MidWiseTransformIterator { template <typename T> class MidWiseTransformIterator {
public: public:
MidWiseTransformIterator(const T *ptr, int n, int post) MidWiseTransformIterator(const T *ptr, int n, int post)
: ptr_(ptr), i_(0), j_(0), n_(n), post_(post) {} : ptr_(ptr), i_(0), j_(0), n_(n), post_(post) {}
MidWiseTransformIterator<T> &operator++() { MidWiseTransformIterator<T> &operator++() {
++j_; ++j_;
if (UNLIKELY(j_ == post_)) { if (UNLIKELY(j_ == post_)) {
++i_; ++i_;
j_ = 0; j_ = 0;
if (UNLIKELY(i_ == n_)) { if (UNLIKELY(i_ == n_)) {
i_ = 0; i_ = 0;
} }
} }
return *this; return *this;
} }
bool operator==(const MidWiseTransformIterator<T> &rhs) const { bool operator==(const MidWiseTransformIterator<T> &rhs) const {
return (ptr_ + i_) == &(*rhs); return (ptr_ + i_) == &(*rhs);
} }
bool operator!=(const MidWiseTransformIterator<T> &rhs) const { bool operator!=(const MidWiseTransformIterator<T> &rhs) const {
return (ptr_ + i_) != &(*rhs); return (ptr_ + i_) != &(*rhs);
} }
const T &operator*() { return ptr_[i_]; } const T &operator*() { return ptr_[i_]; }
private: private:
const T *ptr_; const T *ptr_;
int64_t i_; int64_t i_;
int64_t j_; int64_t j_;
int64_t n_; int64_t n_;
int64_t post_; int64_t post_;
}; };
template <typename Functor, typename T, typename OutType = T> template <typename Functor, typename T, typename OutType = T>
class TransformFunctor { class TransformFunctor {
public: public:
TransformFunctor(const framework::Tensor *x, const framework::Tensor *y, TransformFunctor(const framework::Tensor *x, const framework::Tensor *y,
framework::Tensor *z, Functor func) framework::Tensor *z, Functor func)
: x_(x->data<T>()), y_(y->data<T>()), z_(z->mutable_data<OutType>()), : x_(x->data<T>()), y_(y->data<T>()), z_(z->mutable_data<OutType>()),
nx_(x->numel()), func_(func) {} nx_(x->numel()), func_(func) {}
inline void Run() const { inline void Run() const {
math::Transform trans; math::Transform trans;
// 同时执行func(x_, y_)传入z_。 // 同时执行func(x_, y_)传入z_。
trans(x_, x_ + nx_, y_, z_, func_); trans(x_, x_ + nx_, y_, z_, func_);
} }
inline void RunRowWise(int n, int pre) const { inline void RunRowWise(int n, int pre) const {
math::Transform trans; math::Transform trans;
trans(x_, x_ + nx_, RowwiseTransformIterator<T>(y_, n), z_, func_); trans(x_, x_ + nx_, RowwiseTransformIterator<T>(y_, n), z_, func_);
} }
inline void RunMidWise(int n, int pre, int post) const { inline void RunMidWise(int n, int pre, int post) const {
math::Transform trans; math::Transform trans;
trans(x_, x_ + nx_, MidWiseTransformIterator<T>(y_, n, post), z_, trans(x_, x_ + nx_, MidWiseTransformIterator<T>(y_, n, post), z_, func_);
func_); }
}
private:
private: const T *x_;
const T *x_; const T *y_;
const T *y_; OutType *z_;
OutType *z_; int64_t nx_;
int64_t nx_; Functor func_;
Functor func_;
}; };
template <typename Functor, typename T, typename OutType = T> template <typename Functor, typename T, typename OutType = T>
void ElementwiseComputeEx(const framework::Tensor *x, void ElementwiseComputeEx(const framework::Tensor *x,
const framework::Tensor *y, int axis, Functor func, const framework::Tensor *y, int axis, Functor func,
framework::Tensor *z) { framework::Tensor *z) {
TransformFunctor<Functor, T, OutType> functor(x, y, z, func); TransformFunctor<Functor, T, OutType> functor(x, y, z, func);
auto x_dims = x->dims(); auto x_dims = x->dims();
auto y_dims = y->dims(); auto y_dims = y->dims();
// PADDLE_ENFORCE_GE(x_dims.size(), y_dims.size(), // PADDLE_ENFORCE_GE(x_dims.size(), y_dims.size(),
// "Rank of first input must >= rank of second // "Rank of first input must >= rank of second
// input."); // input.");
if (x_dims == y_dims) { if (x_dims == y_dims) {
functor.Run(); functor.Run();
return; return;
} }
/// axis = -1 represent the last dimension. /// axis = -1 represent the last dimension.
axis = (axis == -1 ? x_dims.size() - y_dims.size() : axis); axis = (axis == -1 ? x_dims.size() - y_dims.size() : axis);
// PADDLE_ENFORCE(axis >= 0 && axis < x_dims.size(), // PADDLE_ENFORCE(axis >= 0 && axis < x_dims.size(),
// "Axis should be in range [0, x_dims)"); // "Axis should be in range [0, x_dims)");
trim_trailing_singular_dims(&y_dims); trim_trailing_singular_dims(&y_dims);
axis = (y_dims.size() == 0) ? x_dims.size() : axis; axis = (y_dims.size() == 0) ? x_dims.size() : axis;
int pre, n, post; int pre, n, post;
get_mid_dims(x_dims, y_dims, axis, &pre, &n, &post); get_mid_dims(x_dims, y_dims, axis, &pre, &n, &post);
if (post == 1) { if (post == 1) {
functor.RunRowWise(n, pre); functor.RunRowWise(n, pre);
return; return;
} else { } else {
functor.RunMidWise(n, pre, post); functor.RunMidWise(n, pre, post);
return; return;
} }
} }
} // namespace operators } // namespace operators
......
src/operators/math/im2col.cc
浏览文件 @ dff4a781
...@@ -26,70 +26,65 @@ namespace math { ...@@ -26,70 +26,65 @@ namespace math {
* output_width] * output_width]
*/ */
template <class T> class Im2ColFunctor<ColFormat::kCFO, CPU, T> { template <class T> class Im2ColFunctor<ColFormat::kCFO, CPU, T> {
public: public:
void operator()(const framework::Tensor &im, void operator()(const framework::Tensor &im, const std::vector<int> &dilation,
const std::vector<int> &dilation, const std::vector<int> &stride,
const std::vector<int> &stride, const std::vector<int> &padding, framework::Tensor *col) {
const std::vector<int> &padding, framework::Tensor *col) { // PADDLE_ENFORCE(im.dims().size() == 3);
// PADDLE_ENFORCE(im.dims().size() == 3); // PADDLE_ENFORCE(col->dims().size() == 5);
// PADDLE_ENFORCE(col->dims().size() == 5);
int im_channels = im.dims()[0]; int im_channels = im.dims()[0];
int im_height = im.dims()[1]; int im_height = im.dims()[1];
int im_width = im.dims()[2]; int im_width = im.dims()[2];
int filter_height = col->dims()[1]; int filter_height = col->dims()[1];
int filter_width = col->dims()[2]; int filter_width = col->dims()[2];
int col_height = col->dims()[3]; int col_height = col->dims()[3];
int col_width = col->dims()[4]; int col_width = col->dims()[4];
// PADDLE_ENFORCE_EQ((im_height + padding[0] + padding[2] // PADDLE_ENFORCE_EQ((im_height + padding[0] + padding[2]
// - // -
// ((dilation[0] * (filter_height - 1) // ((dilation[0] * (filter_height - 1)
// + 1))) / // + 1))) /
// stride[0] + // stride[0] +
// 1, // 1,
// col_height, // col_height,
// "Output_height and // "Output_height and
// padding(padding_up, padding_down) // padding(padding_up, padding_down)
// are " "inconsistent."); // are " "inconsistent.");
// PADDLE_ENFORCE_EQ((im_width + padding[1] + padding[3] // PADDLE_ENFORCE_EQ((im_width + padding[1] + padding[3]
// - // -
// ((dilation[1] * (filter_width - 1) // ((dilation[1] * (filter_width - 1)
// + 1))) / // + 1))) /
// stride[1] + // stride[1] +
// 1, // 1,
// col_width, // col_width,
// "Output_height and // "Output_height and
// padding(padding_up, padding_down) // padding(padding_up, padding_down)
// are " "inconsistent."); // are " "inconsistent.");
int channels_col = im_channels * filter_height * filter_width; int channels_col = im_channels * filter_height * filter_width;
const T *im_data = im.data<T>(); const T *im_data = im.data<T>();
T *col_data = col->data<T>(); T *col_data = col->data<T>();
for (int c = 0; c < channels_col; ++c) { for (int c = 0; c < channels_col; ++c) {
int w_offset = c % filter_width; int w_offset = c % filter_width;
int h_offset = (c / filter_width) % filter_height; int h_offset = (c / filter_width) % filter_height;
int c_im = c / (filter_width * filter_height); int c_im = c / (filter_width * filter_height);
for (int h = 0; h < col_height; ++h) { for (int h = 0; h < col_height; ++h) {
int im_row_idx = int im_row_idx = h * stride[0] - padding[0] + h_offset * dilation[0];
h * stride[0] - padding[0] + h_offset * dilation[0]; for (int w = 0; w < col_width; ++w) {
for (int w = 0; w < col_width; ++w) { int im_col_idx = w * stride[1] - padding[1] + w_offset * dilation[1];
int im_col_idx = int col_idx = (c * col_height + h) * col_width + w;
w * stride[1] - padding[1] + w_offset * dilation[1]; int im_idx = (im_row_idx + c_im * im_height) * im_width + im_col_idx;
int col_idx = (c * col_height + h) * col_width + w;
int im_idx =
(im_row_idx + c_im * im_height) * im_width + im_col_idx;
col_data[col_idx] = col_data[col_idx] = (im_row_idx < 0 || im_row_idx >= im_height ||
(im_row_idx < 0 || im_row_idx >= im_height || im_col_idx < 0 || im_col_idx >= im_width)
im_col_idx < 0 || im_col_idx >= im_width) ? static_cast<T>(0)
? static_cast<T>(0) : im_data[im_idx];
: im_data[im_idx];
}
}
} }
}
} }
}
}; };
/* /*
...@@ -99,67 +94,64 @@ template <class T> class Im2ColFunctor<ColFormat::kCFO, CPU, T> { ...@@ -99,67 +94,64 @@ template <class T> class Im2ColFunctor<ColFormat::kCFO, CPU, T> {
* output_width] * output_width]
*/ */
template <class T> class Col2ImFunctor<ColFormat::kCFO, CPU, T> { template <class T> class Col2ImFunctor<ColFormat::kCFO, CPU, T> {
public: public:
void operator()(const framework::Tensor &col, void operator()(const framework::Tensor &col,
const std::vector<int> &dilation, const std::vector<int> &dilation,
const std::vector<int> &stride, const std::vector<int> &stride,
const std::vector<int> &padding, framework::Tensor *im) { const std::vector<int> &padding, framework::Tensor *im) {
// PADDLE_ENFORCE(im->dims().size() == 3); // PADDLE_ENFORCE(im->dims().size() == 3);
// PADDLE_ENFORCE(col.dims().size() == 5); // PADDLE_ENFORCE(col.dims().size() == 5);
int im_channels = im->dims()[0]; int im_channels = im->dims()[0];
int im_height = im->dims()[1]; int im_height = im->dims()[1];
int im_width = im->dims()[2]; int im_width = im->dims()[2];
int filter_height = col.dims()[1]; int filter_height = col.dims()[1];
int filter_width = col.dims()[2]; int filter_width = col.dims()[2];
int col_height = col.dims()[3]; int col_height = col.dims()[3];
int col_width = col.dims()[4]; int col_width = col.dims()[4];
// PADDLE_ENFORCE_EQ((im_height + padding[0] + padding[2] // PADDLE_ENFORCE_EQ((im_height + padding[0] + padding[2]
// - // -
// ((dilation[0] * (filter_height - 1) // ((dilation[0] * (filter_height - 1)
// + 1))) / // + 1))) /
// stride[0] + // stride[0] +
// 1, // 1,
// col_height, // col_height,
// "Output_height and // "Output_height and
// padding(padding_up, padding_down) // padding(padding_up, padding_down)
// are " "inconsistent."); // are " "inconsistent.");
// PADDLE_ENFORCE_EQ((im_width + padding[1] + padding[3] // PADDLE_ENFORCE_EQ((im_width + padding[1] + padding[3]
// - // -
// ((dilation[1] * (filter_width - 1) // ((dilation[1] * (filter_width - 1)
// + 1))) / // + 1))) /
// stride[1] + // stride[1] +
// 1, // 1,
// col_width, // col_width,
// "Output_height and // "Output_height and
// padding(padding_up, padding_down) // padding(padding_up, padding_down)
// are " "inconsistent."); // are " "inconsistent.");
int channels_col = im_channels * filter_height * filter_width; int channels_col = im_channels * filter_height * filter_width;
T *im_data = im->data<T>(); T *im_data = im->data<T>();
const T *col_data = col.data<T>(); const T *col_data = col.data<T>();
for (int c = 0; c < channels_col; ++c) { for (int c = 0; c < channels_col; ++c) {
int w_offset = c % filter_width; int w_offset = c % filter_width;
int h_offset = (c / filter_width) % filter_height; int h_offset = (c / filter_width) % filter_height;
int c_im = c / (filter_width * filter_height); int c_im = c / (filter_width * filter_height);
for (int h = 0; h < col_height; ++h) { for (int h = 0; h < col_height; ++h) {
int im_row_idx = int im_row_idx = h * stride[0] - padding[0] + h_offset * dilation[0];
h * stride[0] - padding[0] + h_offset * dilation[0]; for (int w = 0; w < col_width; ++w) {
for (int w = 0; w < col_width; ++w) { int im_col_idx = w * stride[1] - padding[1] + w_offset * dilation[1];
int im_col_idx = if ((im_row_idx) >= 0 && (im_row_idx) < im_height &&
w * stride[1] - padding[1] + w_offset * dilation[1]; (im_col_idx) >= 0 && (im_col_idx) < im_width) {
if ((im_row_idx) >= 0 && (im_row_idx) < im_height && im_data[(im_row_idx + c_im * im_height) * im_width + im_col_idx] +=
(im_col_idx) >= 0 && (im_col_idx) < im_width) { col_data[(c * col_height + h) * col_width + w];
im_data[(im_row_idx + c_im * im_height) * im_width + }
im_col_idx] +=
col_data[(c * col_height + h) * col_width + w];
}
}
}
} }
}
} }
}
}; };
template class Im2ColFunctor<ColFormat::kCFO, CPU, float>; template class Im2ColFunctor<ColFormat::kCFO, CPU, float>;
...@@ -174,74 +166,69 @@ template class Col2ImFunctor<ColFormat::kCFO, CPU, double>; ...@@ -174,74 +166,69 @@ template class Col2ImFunctor<ColFormat::kCFO, CPU, double>;
* filter_width] * filter_width]
*/ */
template <class T> class Im2ColFunctor<ColFormat::kOCF, CPU, T> { template <class T> class Im2ColFunctor<ColFormat::kOCF, CPU, T> {
public: public:
void operator()(const framework::Tensor &im, void operator()(const framework::Tensor &im, const std::vector<int> &dilation,
const std::vector<int> &dilation, const std::vector<int> &stride,
const std::vector<int> &stride, const std::vector<int> &padding, framework::Tensor *col) {
const std::vector<int> &padding, framework::Tensor *col) { // PADDLE_ENFORCE(im.dims().size() == 3);
// PADDLE_ENFORCE(im.dims().size() == 3); // PADDLE_ENFORCE(col->dims().size() == 5);
// PADDLE_ENFORCE(col->dims().size() == 5); int im_channels = im.dims()[0];
int im_channels = im.dims()[0]; int im_height = im.dims()[1];
int im_height = im.dims()[1]; int im_width = im.dims()[2];
int im_width = im.dims()[2]; int filter_height = col->dims()[3];
int filter_height = col->dims()[3]; int filter_width = col->dims()[4];
int filter_width = col->dims()[4]; int col_height = col->dims()[0];
int col_height = col->dims()[0]; int col_width = col->dims()[1];
int col_width = col->dims()[1];
// PADDLE_ENFORCE_EQ( // PADDLE_ENFORCE_EQ(
// (im_height + padding[0] + padding[2] - // (im_height + padding[0] + padding[2] -
// filter_height) / stride[0] // filter_height) / stride[0]
// + 1, col_height, "Output_height and // + 1, col_height, "Output_height and
// padding(padding_up, // padding(padding_up,
// padding_down) are " "inconsistent."); // padding_down) are " "inconsistent.");
// PADDLE_ENFORCE_EQ( // PADDLE_ENFORCE_EQ(
// (im_width + padding[1] + padding[3] - // (im_width + padding[1] + padding[3] -
// filter_width) / stride[1] + // filter_width) / stride[1] +
// 1, col_width, "col_width and padding(padding_left, // 1, col_width, "col_width and padding(padding_left,
// padding_right) // padding_right)
// are " "inconsistent."); // are " "inconsistent.");
const T *im_data = im.data<T>(); const T *im_data = im.data<T>();
T *col_data = col->data<T>(); T *col_data = col->data<T>();
for (int col_row_idx = 0; col_row_idx < col_height; ++col_row_idx) { for (int col_row_idx = 0; col_row_idx < col_height; ++col_row_idx) {
for (int col_col_idx = 0; col_col_idx < col_width; ++col_col_idx) { for (int col_col_idx = 0; col_col_idx < col_width; ++col_col_idx) {
for (int channel = 0; channel < im_channels; ++channel) { for (int channel = 0; channel < im_channels; ++channel) {
for (int filter_row_idx = 0; filter_row_idx < filter_height; for (int filter_row_idx = 0; filter_row_idx < filter_height;
++filter_row_idx) { ++filter_row_idx) {
int im_row_offset = col_row_idx * stride[0] + int im_row_offset =
filter_row_idx - padding[0]; col_row_idx * stride[0] + filter_row_idx - padding[0];
for (int filter_col_idx = 0; for (int filter_col_idx = 0; filter_col_idx < filter_width;
filter_col_idx < filter_width; ++filter_col_idx) { ++filter_col_idx) {
int im_col_offset = col_col_idx * stride[1] + int im_col_offset =
filter_col_idx - padding[1]; col_col_idx * stride[1] + filter_col_idx - padding[1];
int col_offset = int col_offset =
((((col_row_idx)*col_width + col_col_idx) * ((((col_row_idx)*col_width + col_col_idx) * im_channels +
im_channels + channel) *
channel) * filter_height +
filter_height + filter_row_idx) *
filter_row_idx) * filter_width +
filter_width + filter_col_idx;
filter_col_idx;
int im_offset = int im_offset = (channel * im_height + im_row_offset) * im_width +
(channel * im_height + im_row_offset) * im_col_offset;
im_width + col_data[col_offset] =
im_col_offset; (im_row_offset < 0 || im_row_offset >= im_height ||
col_data[col_offset] = im_col_offset < 0 || im_col_offset >= im_width)
(im_row_offset < 0 || ? static_cast<T>(0)
im_row_offset >= im_height || : im_data[im_offset];
im_col_offset < 0 || im_col_offset >= im_width)
? static_cast<T>(0)
: im_data[im_offset];
}
}
}
} }
}
} }
}
} }
}
}; };
/* /*
...@@ -251,74 +238,70 @@ template <class T> class Im2ColFunctor<ColFormat::kOCF, CPU, T> { ...@@ -251,74 +238,70 @@ template <class T> class Im2ColFunctor<ColFormat::kOCF, CPU, T> {
* filter_width] * filter_width]
*/ */
template <class T> class Col2ImFunctor<ColFormat::kOCF, CPU, T> { template <class T> class Col2ImFunctor<ColFormat::kOCF, CPU, T> {
public: public:
void operator()(const framework::Tensor &col, void operator()(const framework::Tensor &col,
const std::vector<int> &dilation, const std::vector<int> &dilation,
const std::vector<int> &stride, const std::vector<int> &stride,
const std::vector<int> &padding, framework::Tensor *im) { const std::vector<int> &padding, framework::Tensor *im) {
// PADDLE_ENFORCE(im->dims().size() == 3); // PADDLE_ENFORCE(im->dims().size() == 3);
// PADDLE_ENFORCE(col.dims().size() == 5); // PADDLE_ENFORCE(col.dims().size() == 5);
int im_channels = im->dims()[0]; int im_channels = im->dims()[0];
int im_height = im->dims()[1]; int im_height = im->dims()[1];
int im_width = im->dims()[2]; int im_width = im->dims()[2];
int filter_height = col.dims()[3]; int filter_height = col.dims()[3];
int filter_width = col.dims()[4]; int filter_width = col.dims()[4];
int col_height = col.dims()[0]; int col_height = col.dims()[0];
int col_width = col.dims()[1]; int col_width = col.dims()[1];
// PADDLE_ENFORCE_EQ( // PADDLE_ENFORCE_EQ(
// (im_height + padding[0] + padding[2] - // (im_height + padding[0] + padding[2] -
// filter_height) / stride[0] // filter_height) / stride[0]
// + 1, col_height, "Output_height and // + 1, col_height, "Output_height and
// padding(padding_up, // padding(padding_up,
// padding_down) are " "inconsistent."); // padding_down) are " "inconsistent.");
// PADDLE_ENFORCE_EQ( // PADDLE_ENFORCE_EQ(
// (im_width + padding[1] + padding[3] - // (im_width + padding[1] + padding[3] -
// filter_width) / stride[1] + // filter_width) / stride[1] +
// 1, col_width, "col_width and padding(padding_left, // 1, col_width, "col_width and padding(padding_left,
// padding_right) // padding_right)
// are " "inconsistent."); // are " "inconsistent.");
T *im_data = im->data<T>(); T *im_data = im->data<T>();
const T *col_data = col.data<T>(); const T *col_data = col.data<T>();
for (int col_row_idx = 0; col_row_idx < col_height; ++col_row_idx) { for (int col_row_idx = 0; col_row_idx < col_height; ++col_row_idx) {
for (int col_col_idx = 0; col_col_idx < col_width; ++col_col_idx) { for (int col_col_idx = 0; col_col_idx < col_width; ++col_col_idx) {
for (int channel = 0; channel < im_channels; ++channel) { for (int channel = 0; channel < im_channels; ++channel) {
for (int filter_row_idx = 0; filter_row_idx < filter_height; for (int filter_row_idx = 0; filter_row_idx < filter_height;
++filter_row_idx) { ++filter_row_idx) {
int im_row_offset = col_row_idx * stride[0] + int im_row_offset =
filter_row_idx - padding[0]; col_row_idx * stride[0] + filter_row_idx - padding[0];
for (int filter_col_idx = 0; for (int filter_col_idx = 0; filter_col_idx < filter_width;
filter_col_idx < filter_width; ++filter_col_idx) { ++filter_col_idx) {
int im_col_offset = col_col_idx * stride[1] + int im_col_offset =
filter_col_idx - padding[1]; col_col_idx * stride[1] + filter_col_idx - padding[1];
int col_offset = int col_offset =
(((col_row_idx * col_width + col_col_idx) * (((col_row_idx * col_width + col_col_idx) * im_channels +
im_channels + channel) *
channel) * filter_height +
filter_height + filter_row_idx) *
filter_row_idx) * filter_width +
filter_width + filter_col_idx;
filter_col_idx;
if (im_row_offset >= 0 && if (im_row_offset >= 0 && im_row_offset < im_height &&
im_row_offset < im_height && im_col_offset >= 0 && im_col_offset < im_width) {
im_col_offset >= 0 && int im_offset =
im_col_offset < im_width) { (channel * im_height + im_row_offset) * im_width +
int im_offset = im_col_offset;
(channel * im_height + im_row_offset) * im_data[im_offset] += col_data[col_offset];
im_width + }
im_col_offset;
im_data[im_offset] += col_data[col_offset];
}
}
}
}
} }
}
} }
}
} }
}
}; };
template class Im2ColFunctor<ColFormat::kOCF, CPU, float>; template class Im2ColFunctor<ColFormat::kOCF, CPU, float>;
......
src/operators/math/im2col.h
浏览文件 @ dff4a781
...@@ -89,20 +89,19 @@ enum class ColFormat { kCFO = 0, kOCF = 1 }; ...@@ -89,20 +89,19 @@ enum class ColFormat { kCFO = 0, kOCF = 1 };
*/ */
template <ColFormat Format, typename DeviceType, typename T> template <ColFormat Format, typename DeviceType, typename T>
class Im2ColFunctor { class Im2ColFunctor {
public: public:
void operator()(const framework::Tensor &im, void operator()(const framework::Tensor &im, const std::vector<int> &dilation,
const std::vector<int> &dilation, const std::vector<int> &stride,
const std::vector<int> &stride, const std::vector<int> &padding, framework::Tensor *col);
const std::vector<int> &padding, framework::Tensor *col);
}; };
template <ColFormat Format, typename DeviceType, typename T> template <ColFormat Format, typename DeviceType, typename T>
class Col2ImFunctor { class Col2ImFunctor {
public: public:
void operator()(const framework::Tensor &col, void operator()(const framework::Tensor &col,
const std::vector<int> &dilation, const std::vector<int> &dilation,
const std::vector<int> &stride, const std::vector<int> &stride,
const std::vector<int> &padding, framework::Tensor *im); const std::vector<int> &padding, framework::Tensor *im);
}; };
} // namespace math } // namespace math
......
src/operators/math/math_function.cc
浏览文件 @ dff4a781
...@@ -22,11 +22,11 @@ template <> ...@@ -22,11 +22,11 @@ template <>
void gemm<float>(const CBLAS_TRANSPOSE transA, const CBLAS_TRANSPOSE transB, void gemm<float>(const CBLAS_TRANSPOSE transA, const CBLAS_TRANSPOSE transB,
const int M, const int N, const int K, const float alpha, const int M, const int N, const int K, const float alpha,
const float *A, const float *B, const float beta, float *C) { const float *A, const float *B, const float beta, float *C) {
int lda = (transA == CblasNoTrans) ? K : M; int lda = (transA == CblasNoTrans) ? K : M;
int ldb = (transB == CblasNoTrans) ? N : K; int ldb = (transB == CblasNoTrans) ? N : K;
int ldc = N; int ldc = N;
cblas_sgemm(CblasRowMajor, transA, transB, M, N, K, alpha, A, lda, B, ldb, cblas_sgemm(CblasRowMajor, transA, transB, M, N, K, alpha, A, lda, B, ldb,
beta, C, ldc); beta, C, ldc);
} }
template <> template <>
...@@ -34,11 +34,11 @@ void gemm<double>(const CBLAS_TRANSPOSE transA, const CBLAS_TRANSPOSE transB, ...@@ -34,11 +34,11 @@ void gemm<double>(const CBLAS_TRANSPOSE transA, const CBLAS_TRANSPOSE transB,
const int M, const int N, const int K, const double alpha, const int M, const int N, const int K, const double alpha,
const double *A, const double *B, const double beta, const double *A, const double *B, const double beta,
double *C) { double *C) {
int lda = (transA == CblasNoTrans) ? K : M; int lda = (transA == CblasNoTrans) ? K : M;
int ldb = (transB == CblasNoTrans) ? N : K; int ldb = (transB == CblasNoTrans) ? N : K;
int ldc = N; int ldc = N;
cblas_dgemm(CblasRowMajor, transA, transB, M, N, K, alpha, A, lda, B, ldb, cblas_dgemm(CblasRowMajor, transA, transB, M, N, K, alpha, A, lda, B, ldb,
beta, C, ldc); beta, C, ldc);
} }
template <> template <>
...@@ -46,9 +46,9 @@ void gemm<float>(const bool transA, const bool transB, const int M, const int N, ...@@ -46,9 +46,9 @@ void gemm<float>(const bool transA, const bool transB, const int M, const int N,
const int K, const float alpha, const float *A, const int lda, const int K, const float alpha, const float *A, const int lda,
const float *B, const int ldb, const float beta, float *C, const float *B, const int ldb, const float beta, float *C,
const int ldc) { const int ldc) {
cblas_sgemm(CblasRowMajor, transA == false ? CblasNoTrans : CblasTrans, cblas_sgemm(CblasRowMajor, transA == false ? CblasNoTrans : CblasTrans,
transB == false ? CblasNoTrans : CblasTrans, M, N, K, alpha, A, transB == false ? CblasNoTrans : CblasTrans, M, N, K, alpha, A,
lda, B, ldb, beta, C, ldc); lda, B, ldb, beta, C, ldc);
} }
template <> template <>
...@@ -56,67 +56,67 @@ void gemm<double>(const bool transA, const bool transB, const int M, ...@@ -56,67 +56,67 @@ void gemm<double>(const bool transA, const bool transB, const int M,
const int N, const int K, const double alpha, const double *A, const int N, const int K, const double alpha, const double *A,
const int lda, const double *B, const int ldb, const int lda, const double *B, const int ldb,
const double beta, double *C, const int ldc) { const double beta, double *C, const int ldc) {
cblas_dgemm(CblasRowMajor, transA == false ? CblasNoTrans : CblasTrans, cblas_dgemm(CblasRowMajor, transA == false ? CblasNoTrans : CblasTrans,
transB == false ? CblasNoTrans : CblasTrans, M, N, K, alpha, A, transB == false ? CblasNoTrans : CblasTrans, M, N, K, alpha, A,
lda, B, ldb, beta, C, ldc); lda, B, ldb, beta, C, ldc);
} }
template <> template <>
void matmul<float>(const framework::Tensor &matrix_a, bool trans_a, void matmul<float>(const framework::Tensor &matrix_a, bool trans_a,
const framework::Tensor &matrix_b, bool trans_b, float alpha, const framework::Tensor &matrix_b, bool trans_b, float alpha,
framework::Tensor *matrix_out, float beta) { framework::Tensor *matrix_out, float beta) {
auto dim_a = matrix_a.dims(); auto dim_a = matrix_a.dims();
auto dim_b = matrix_b.dims(); auto dim_b = matrix_b.dims();
auto dim_out = matrix_out->dims(); auto dim_out = matrix_out->dims();
// PADDLE_ENFORCE(dim_a.size() == 2 && dim_b.size() == 2 && // PADDLE_ENFORCE(dim_a.size() == 2 && dim_b.size() == 2 &&
// dim_out.size() == // dim_out.size() ==
// 2, // 2,
// "The input and output of matmul be matrix"); // "The input and output of matmul be matrix");
// //
// PADDLE_ENFORCE(platform::is_cpu_place(matrix_a.place()) && // PADDLE_ENFORCE(platform::is_cpu_place(matrix_a.place()) &&
// platform::is_cpu_place(matrix_b.place()) // platform::is_cpu_place(matrix_b.place())
// && // &&
// platform::is_cpu_place(matrix_out->place()), // platform::is_cpu_place(matrix_out->place()),
// "Matrix must all be in CPUPlace"); // "Matrix must all be in CPUPlace");
int M = dim_out[0]; int M = dim_out[0];
int N = dim_out[1]; int N = dim_out[1];
int K = (trans_a == false) ? dim_a[1] : dim_a[0]; int K = (trans_a == false) ? dim_a[1] : dim_a[0];
CBLAS_TRANSPOSE transA = (trans_a == false) ? CblasNoTrans : CblasTrans; CBLAS_TRANSPOSE transA = (trans_a == false) ? CblasNoTrans : CblasTrans;
CBLAS_TRANSPOSE transB = (trans_b == false) ? CblasNoTrans : CblasTrans; CBLAS_TRANSPOSE transB = (trans_b == false) ? CblasNoTrans : CblasTrans;
gemm<float>(transA, transB, M, N, K, alpha, matrix_a.data<float>(), gemm<float>(transA, transB, M, N, K, alpha, matrix_a.data<float>(),
matrix_b.data<float>(), beta, matrix_out->data<float>()); matrix_b.data<float>(), beta, matrix_out->data<float>());
} }
template <> template <>
void matmul<double>(const framework::Tensor &matrix_a, bool trans_a, void matmul<double>(const framework::Tensor &matrix_a, bool trans_a,
const framework::Tensor &matrix_b, bool trans_b, const framework::Tensor &matrix_b, bool trans_b,
double alpha, framework::Tensor *matrix_out, double beta) { double alpha, framework::Tensor *matrix_out, double beta) {
auto dim_a = matrix_a.dims(); auto dim_a = matrix_a.dims();
auto dim_b = matrix_b.dims(); auto dim_b = matrix_b.dims();
auto dim_out = matrix_out->dims(); auto dim_out = matrix_out->dims();
// PADDLE_ENFORCE(dim_a.size() == 2 && dim_b.size() == 2 && // PADDLE_ENFORCE(dim_a.size() == 2 && dim_b.size() == 2 &&
// dim_out.size() == // dim_out.size() ==
// 2, // 2,
// "The input and output of matmul be matrix"); // "The input and output of matmul be matrix");
// //
// PADDLE_ENFORCE(platform::is_cpu_place(matrix_a.place()) && // PADDLE_ENFORCE(platform::is_cpu_place(matrix_a.place()) &&
// platform::is_cpu_place(matrix_b.place()) // platform::is_cpu_place(matrix_b.place())
// && // &&
// platform::is_cpu_place(matrix_out->place()), // platform::is_cpu_place(matrix_out->place()),
// "Matrix must all be in CPUPlace"); // "Matrix must all be in CPUPlace");
int M = dim_out[0]; int M = dim_out[0];
int N = dim_out[1]; int N = dim_out[1];
int K = (trans_a == false) ? dim_a[1] : dim_a[0]; int K = (trans_a == false) ? dim_a[1] : dim_a[0];
CBLAS_TRANSPOSE transA = (trans_a == false) ? CblasNoTrans : CblasTrans; CBLAS_TRANSPOSE transA = (trans_a == false) ? CblasNoTrans : CblasTrans;
CBLAS_TRANSPOSE transB = (trans_b == false) ? CblasNoTrans : CblasTrans; CBLAS_TRANSPOSE transB = (trans_b == false) ? CblasNoTrans : CblasTrans;
gemm<double>(transA, transB, M, N, K, alpha, matrix_a.data<double>(), gemm<double>(transA, transB, M, N, K, alpha, matrix_a.data<double>(),
matrix_b.data<double>(), beta, matrix_out->data<double>()); matrix_b.data<double>(), beta, matrix_out->data<double>());
} }
} // namespace math } // namespace math
......
src/operators/math/pool3x3.h
浏览文件 @ dff4a781
...@@ -22,9 +22,9 @@ SOFTWARE. ...@@ -22,9 +22,9 @@ SOFTWARE.
#endif // __ARM_NEON #endif // __ARM_NEON
static void Pool3x3Max() { static void Pool3x3Max() {
// todo impl with neon // todo impl with neon
} }
static void Pool3x3Avg() { static void Pool3x3Avg() {
// todo impl with neon // todo impl with neon
} }
src/operators/math/pool_2x2.h
浏览文件 @ dff4a781
...@@ -22,9 +22,9 @@ SOFTWARE. ...@@ -22,9 +22,9 @@ SOFTWARE.
#endif // __ARM_NEON #endif // __ARM_NEON
static void Pool2x2Max() { static void Pool2x2Max() {
// todo impl with neon // todo impl with neon
} }
static void Pool2x2Avg() { static void Pool2x2Avg() {
// todo impl with neon // todo impl with neon
} }
src/operators/math/pooling.cpp
浏览文件 @ dff4a781
...@@ -30,67 +30,65 @@ namespace math { ...@@ -30,67 +30,65 @@ namespace math {
*/ */
template <typename PoolProcess, typename T> template <typename PoolProcess, typename T>
class PoolFunctor<CPU, PoolProcess, T> { class PoolFunctor<CPU, PoolProcess, T> {
public: public:
void operator()(const framework::Tensor &input, void operator()(const framework::Tensor &input, const std::vector<int> &ksize,
const std::vector<int> &ksize, const std::vector<int> &strides,
const std::vector<int> &strides, const std::vector<int> &paddings, PoolProcess pool_process,
const std::vector<int> &paddings, PoolProcess pool_process, framework::Tensor *output) {
framework::Tensor *output) {
const int batch_size = input.dims()[0]; const int batch_size = input.dims()[0];
const int input_height = input.dims()[2]; const int input_height = input.dims()[2];
const int input_width = input.dims()[3]; const int input_width = input.dims()[3];
if (output == nullptr) { if (output == nullptr) {
DLOG << "output tensor is null"; DLOG << "output tensor is null";
} }
const int output_channels = output->dims()[1]; const int output_channels = output->dims()[1];
const int output_height = output->dims()[2]; const int output_height = output->dims()[2];
const int output_width = output->dims()[3]; const int output_width = output->dims()[3];
const int ksize_height = ksize[0]; const int ksize_height = ksize[0];
const int ksize_width = ksize[1]; const int ksize_width = ksize[1];
const int stride_height = strides[0]; const int stride_height = strides[0];
const int stride_width = strides[1]; const int stride_width = strides[1];
const int padding_height = paddings[0]; const int padding_height = paddings[0];
const int padding_width = paddings[1]; const int padding_width = paddings[1];
const int input_stride = input_height * input_width; const int input_stride = input_height * input_width;
const int output_stride = output_height * output_width; const int output_stride = output_height * output_width;
const T *input_data = input.data<T>(); const T *input_data = input.data<T>();
T *output_data = output->mutable_data<T>(); T *output_data = output->mutable_data<T>();
for (int i = 0; i < batch_size; i++) { for (int i = 0; i < batch_size; i++) {
#pragma omp parallel for #pragma omp parallel for
for (int c = 0; c < output_channels; ++c) { for (int c = 0; c < output_channels; ++c) {
for (int ph = 0; ph < output_height; ++ph) { for (int ph = 0; ph < output_height; ++ph) {
int hstart = ph * stride_height - padding_height; int hstart = ph * stride_height - padding_height;
int hend = std::min(hstart + ksize_height, input_height); int hend = std::min(hstart + ksize_height, input_height);
hstart = std::max(hstart, 0); hstart = std::max(hstart, 0);
for (int pw = 0; pw < output_width; ++pw) { for (int pw = 0; pw < output_width; ++pw) {
int wstart = pw * stride_width - padding_width; int wstart = pw * stride_width - padding_width;
int wend = std::min(wstart + ksize_width, input_width); int wend = std::min(wstart + ksize_width, input_width);
wstart = std::max(wstart, 0); wstart = std::max(wstart, 0);
T ele = pool_process.initial(); T ele = pool_process.initial();
for (int h = hstart; h < hend; ++h) { for (int h = hstart; h < hend; ++h) {
for (int w = wstart; w < wend; ++w) { for (int w = wstart; w < wend; ++w) {
pool_process.compute( pool_process.compute(input_data[h * input_width + w], &ele);
input_data[h * input_width + w], &ele); }
}
}
int pool_size = (hend - hstart) * (wend - wstart);
pool_process.finalize(static_cast<T>(pool_size), &ele);
output_data[ph * output_width + pw] = ele;
}
}
input_data += input_stride;
output_data += output_stride;
} }
int pool_size = (hend - hstart) * (wend - wstart);
pool_process.finalize(static_cast<T>(pool_size), &ele);
output_data[ph * output_width + pw] = ele;
}
} }
input_data += input_stride;
output_data += output_stride;
}
} }
}
}; };
template class PoolFunctor<CPU, math::AvgPool<float>, float>; template class PoolFunctor<CPU, math::AvgPool<float>, float>;
......
src/operators/math/pooling.h
浏览文件 @ dff4a781
...@@ -38,31 +38,30 @@ namespace math { ...@@ -38,31 +38,30 @@ namespace math {
* MaxPoolGrad and AvgPoolGrad are gradient operations respectively. * MaxPoolGrad and AvgPoolGrad are gradient operations respectively.
*/ */
template <class T> class MaxPool { template <class T> class MaxPool {
public: public:
inline T initial() { return static_cast<T>(-FLT_MAX); } inline T initial() { return static_cast<T>(-FLT_MAX); }
inline void compute(const T &x, T *y) { *y = *y > x ? *y : x; } inline void compute(const T &x, T *y) { *y = *y > x ? *y : x; }
inline void finalize(const T &pool_field, T *y) {} inline void finalize(const T &pool_field, T *y) {}
}; };
template <class T> class AvgPool { template <class T> class AvgPool {
public: public:
inline T initial() { return static_cast<T>(0); } inline T initial() { return static_cast<T>(0); }
inline void compute(const T &x, T *y) { *y += x; } inline void compute(const T &x, T *y) { *y += x; }
inline void finalize(const T &pool_field, T *y) { *y /= pool_field; } inline void finalize(const T &pool_field, T *y) { *y /= pool_field; }
}; };
template <typename DeviceType, typename PoolProcess, typename T> template <typename DeviceType, typename PoolProcess, typename T>
class PoolFunctor { class PoolFunctor {
public: public:
void operator()(const framework::Tensor &input, void operator()(const framework::Tensor &input, const std::vector<int> &ksize,
const std::vector<int> &ksize, const std::vector<int> &strides,
const std::vector<int> &strides, const std::vector<int> &paddings, PoolProcess pool_compute,
const std::vector<int> &paddings, PoolProcess pool_compute, framework::Tensor *output);
framework::Tensor *output);
}; };
} }
} // namespace operators } // namespace operators
......
src/operators/math/transform.h
浏览文件 @ dff4a781
...@@ -37,18 +37,18 @@ namespace math { ...@@ -37,18 +37,18 @@ namespace math {
// class, paddle::fluid::operators::RowwiseTRansformIterator. // class, paddle::fluid::operators::RowwiseTRansformIterator.
struct Transform { struct Transform {
template <typename InputIter, typename OutputIter, typename UnaryOperation> template <typename InputIter, typename OutputIter, typename UnaryOperation>
void operator()(InputIter first, InputIter last, OutputIter result, void operator()(InputIter first, InputIter last, OutputIter result,
UnaryOperation op) { UnaryOperation op) {
std::transform(first, last, result, op); std::transform(first, last, result, op);
} }
template <typename InputIter1, typename InputIter2, typename OutputIter, template <typename InputIter1, typename InputIter2, typename OutputIter,
typename BinaryOperation> typename BinaryOperation>
void operator()(InputIter1 first1, InputIter1 last1, InputIter2 first2, void operator()(InputIter1 first1, InputIter1 last1, InputIter2 first2,
OutputIter result, BinaryOperation op) { OutputIter result, BinaryOperation op) {
std::transform(first1, last1, first2, result, op); std::transform(first1, last1, first2, result, op);
} }
}; };
} // namespace math } // namespace math
} // namespace operators } // namespace operators
......
src/operators/math/vol2col.cc
浏览文件 @ dff4a781
...@@ -26,90 +26,83 @@ using Tensor = paddle_mobile::framework::Tensor; ...@@ -26,90 +26,83 @@ using Tensor = paddle_mobile::framework::Tensor;
* output_depth, output_height, output_width] * output_depth, output_height, output_width]
*/ */
template <typename T> class Vol2ColFunctor<CPU, T> { template <typename T> class Vol2ColFunctor<CPU, T> {
public: public:
void operator()(const Tensor &vol, const std::vector<int> &dilations, void operator()(const Tensor &vol, const std::vector<int> &dilations,
const std::vector<int> &strides, const std::vector<int> &strides,
const std::vector<int> &paddings, Tensor *col) const { const std::vector<int> &paddings, Tensor *col) const {
// PADDLE_ENFORCE(vol.dims().size() == 4); // PADDLE_ENFORCE(vol.dims().size() == 4);
// PADDLE_ENFORCE(col->dims().size() == 7); // PADDLE_ENFORCE(col->dims().size() == 7);
int input_channels = vol.dims()[0]; int input_channels = vol.dims()[0];
int input_depth = vol.dims()[1]; int input_depth = vol.dims()[1];
int input_height = vol.dims()[2]; int input_height = vol.dims()[2];
int input_width = vol.dims()[3]; int input_width = vol.dims()[3];
int filter_depth = col->dims()[1]; int filter_depth = col->dims()[1];
int filter_height = col->dims()[2]; int filter_height = col->dims()[2];
int filter_width = col->dims()[3]; int filter_width = col->dims()[3];
int output_depth = col->dims()[4]; int output_depth = col->dims()[4];
int output_height = col->dims()[5]; int output_height = col->dims()[5];
int output_width = col->dims()[6]; int output_width = col->dims()[6];
int channels_col = int channels_col =
input_channels * filter_depth * filter_height * filter_width; input_channels * filter_depth * filter_height * filter_width;
// PADDLE_ENFORCE_EQ((input_depth + 2 * paddings[0] - // PADDLE_ENFORCE_EQ((input_depth + 2 * paddings[0] -
// ((dilations[0] * (filter_depth - 1) // ((dilations[0] * (filter_depth - 1)
// + 1))) / // + 1))) /
// strides[0] + // strides[0] +
// 1, // 1,
// output_depth, // output_depth,
// "input_depth and output_depth are " // "input_depth and output_depth are "
// "mismatching."); // "mismatching.");
// PADDLE_ENFORCE_EQ((input_height + 2 * paddings[1] - // PADDLE_ENFORCE_EQ((input_height + 2 * paddings[1] -
// ((dilations[1] * (filter_height - // ((dilations[1] * (filter_height -
// 1) + 1))) / // 1) + 1))) /
// strides[1] + // strides[1] +
// 1, // 1,
// output_height, // output_height,
// "input_height and output_height are // "input_height and output_height are
// " // "
// "mismatching."); // "mismatching.");
// PADDLE_ENFORCE_EQ((input_width + 2 * paddings[2] - // PADDLE_ENFORCE_EQ((input_width + 2 * paddings[2] -
// ((dilations[2] * (filter_width - 1) // ((dilations[2] * (filter_width - 1)
// + 1))) / // + 1))) /
// strides[2] + // strides[2] +
// 1, // 1,
// output_width, // output_width,
// "input_width and output_width are " // "input_width and output_width are "
// "mismatching."); // "mismatching.");
const T *vol_data = vol.data<T>(); const T *vol_data = vol.data<T>();
T *col_data = col->data<T>(); T *col_data = col->data<T>();
for (int c = 0; c < channels_col; ++c) { for (int c = 0; c < channels_col; ++c) {
int w_offset = c % filter_width; int w_offset = c % filter_width;
int h_offset = (c / filter_width) % filter_height; int h_offset = (c / filter_width) % filter_height;
int d_offset = (c / filter_width / filter_height) % filter_depth; int d_offset = (c / filter_width / filter_height) % filter_depth;
int c_in = c / filter_width / filter_height / filter_depth; int c_in = c / filter_width / filter_height / filter_depth;
for (int d = 0; d < output_depth; ++d) { for (int d = 0; d < output_depth; ++d) {
int d_pad = int d_pad = d * strides[0] - paddings[0] + d_offset * dilations[0];
d * strides[0] - paddings[0] + d_offset * dilations[0]; for (int h = 0; h < output_height; ++h) {
for (int h = 0; h < output_height; ++h) { int h_pad = h * strides[1] - paddings[1] + h_offset * dilations[1];
int h_pad = for (int w = 0; w < output_width; ++w) {
h * strides[1] - paddings[1] + h_offset * dilations[1]; int w_pad = w * strides[2] - paddings[2] + w_offset * dilations[2];
for (int w = 0; w < output_width; ++w) {
int w_pad = w * strides[2] - paddings[2] + int col_idx =
w_offset * dilations[2]; ((c * output_depth + d) * output_height + h) * output_width + w;
int vol_idx =
int col_idx = ((c_in * input_depth + d_pad) * input_height + h_pad) *
((c * output_depth + d) * output_height + h) * input_width +
output_width + w_pad;
w; col_data[col_idx] =
int vol_idx = (h_pad < 0 || h_pad >= input_height || w_pad < 0 ||
((c_in * input_depth + d_pad) * input_height + w_pad >= input_width || d_pad < 0 || d_pad >= input_depth)
h_pad) * ? static_cast<T>(0)
input_width + : vol_data[vol_idx];
w_pad; }
col_data[col_idx] =
(h_pad < 0 || h_pad >= input_height || w_pad < 0 ||
w_pad >= input_width || d_pad < 0 ||
d_pad >= input_depth)
? static_cast<T>(0)
: vol_data[vol_idx];
}
}
}
} }
}
} }
}
}; };
/* /*
...@@ -119,89 +112,83 @@ template <typename T> class Vol2ColFunctor<CPU, T> { ...@@ -119,89 +112,83 @@ template <typename T> class Vol2ColFunctor<CPU, T> {
* output_depth, output_height, output_width] * output_depth, output_height, output_width]
*/ */
template <typename T> class Col2VolFunctor<CPU, T> { template <typename T> class Col2VolFunctor<CPU, T> {
public: public:
void operator()(const Tensor &col, const std::vector<int> &dilations, void operator()(const Tensor &col, const std::vector<int> &dilations,
const std::vector<int> &strides, const std::vector<int> &strides,
const std::vector<int> &paddings, Tensor *vol) const { const std::vector<int> &paddings, Tensor *vol) const {
// PADDLE_ENFORCE(vol->dims().size() == 4); // PADDLE_ENFORCE(vol->dims().size() == 4);
// PADDLE_ENFORCE(col.dims().size() == 7); // PADDLE_ENFORCE(col.dims().size() == 7);
int input_channels = vol->dims()[0]; int input_channels = vol->dims()[0];
int input_depth = vol->dims()[1]; int input_depth = vol->dims()[1];
int input_height = vol->dims()[2]; int input_height = vol->dims()[2];
int input_width = vol->dims()[3]; int input_width = vol->dims()[3];
int filter_depth = col.dims()[1]; int filter_depth = col.dims()[1];
int filter_height = col.dims()[2]; int filter_height = col.dims()[2];
int filter_width = col.dims()[3]; int filter_width = col.dims()[3];
int output_depth = col.dims()[4]; int output_depth = col.dims()[4];
int output_height = col.dims()[5]; int output_height = col.dims()[5];
int output_width = col.dims()[6]; int output_width = col.dims()[6];
int channels_col = int channels_col =
input_channels * filter_depth * filter_height * filter_width; input_channels * filter_depth * filter_height * filter_width;
// PADDLE_ENFORCE_EQ((input_depth + 2 * paddings[0] - // PADDLE_ENFORCE_EQ((input_depth + 2 * paddings[0] -
// ((dilations[0] * (filter_depth - 1) // ((dilations[0] * (filter_depth - 1)
// + 1))) / // + 1))) /
// strides[0] + // strides[0] +
// 1, // 1,
// output_depth, // output_depth,
// "input_depth and output_depth are " // "input_depth and output_depth are "
// "mismatching."); // "mismatching.");
// PADDLE_ENFORCE_EQ((input_height + 2 * paddings[1] - // PADDLE_ENFORCE_EQ((input_height + 2 * paddings[1] -
// ((dilations[1] * (filter_height - // ((dilations[1] * (filter_height -
// 1) + 1))) / // 1) + 1))) /
// strides[1] + // strides[1] +
// 1, // 1,
// output_height, // output_height,
// "input_height and output_height are // "input_height and output_height are
// " // "
// "mismatching."); // "mismatching.");
// PADDLE_ENFORCE_EQ((input_width + 2 * paddings[2] - // PADDLE_ENFORCE_EQ((input_width + 2 * paddings[2] -
// ((dilations[2] * (filter_width - 1) // ((dilations[2] * (filter_width - 1)
// + 1))) / // + 1))) /
// strides[2] + // strides[2] +
// 1, // 1,
// output_width, // output_width,
// "input_width and output_width are " // "input_width and output_width are "
// "mismatching."); // "mismatching.");
T *vol_data = vol->data<T>(); T *vol_data = vol->data<T>();
const T *col_data = col.data<T>(); const T *col_data = col.data<T>();
for (int c = 0; c < channels_col; ++c) { for (int c = 0; c < channels_col; ++c) {
int w_offset = c % filter_width; int w_offset = c % filter_width;
int h_offset = (c / filter_width) % filter_height; int h_offset = (c / filter_width) % filter_height;
int d_offset = (c / filter_width / filter_height) % filter_depth; int d_offset = (c / filter_width / filter_height) % filter_depth;
int cIm = c / filter_width / filter_height / filter_depth; int cIm = c / filter_width / filter_height / filter_depth;
for (int d = 0; d < output_depth; ++d) { for (int d = 0; d < output_depth; ++d) {
int d_pad = int d_pad = d * strides[0] - paddings[0] + d_offset * dilations[0];
d * strides[0] - paddings[0] + d_offset * dilations[0]; for (int h = 0; h < output_height; ++h) {
for (int h = 0; h < output_height; ++h) { int h_pad = h * strides[1] - paddings[1] + h_offset * dilations[1];
int h_pad = for (int w = 0; w < output_width; ++w) {
h * strides[1] - paddings[1] + h_offset * dilations[1]; int w_pad = w * strides[2] - paddings[2] + w_offset * dilations[2];
for (int w = 0; w < output_width; ++w) {
int w_pad = w * strides[2] - paddings[2] + if (h_pad >= 0 && h_pad < input_height && w_pad >= 0 &&
w_offset * dilations[2]; w_pad < input_width && d_pad >= 0 && d_pad < input_depth) {
int vol_idx =
if (h_pad >= 0 && h_pad < input_height && w_pad >= 0 && ((cIm * input_depth + d_pad) * input_height + h_pad) *
w_pad < input_width && d_pad >= 0 && input_width +
d_pad < input_depth) { w_pad;
int vol_idx =
((cIm * input_depth + d_pad) * input_height + int col_idx =
h_pad) * ((c * output_depth + d) * output_height + h) * output_width +
input_width + w;
w_pad; vol_data[vol_idx] += col_data[col_idx];
int col_idx =
((c * output_depth + d) * output_height + h) *
output_width +
w;
vol_data[vol_idx] += col_data[col_idx];
}
}
}
} }
}
} }
}
} }
}
}; };
template class Vol2ColFunctor<CPU, float>; template class Vol2ColFunctor<CPU, float>;
......
src/operators/math/vol2col.h
浏览文件 @ dff4a781
...@@ -73,17 +73,17 @@ namespace math { ...@@ -73,17 +73,17 @@ namespace math {
using Tensor = paddle_mobile::framework::Tensor; using Tensor = paddle_mobile::framework::Tensor;
template <typename DeviceType, typename T> class Vol2ColFunctor { template <typename DeviceType, typename T> class Vol2ColFunctor {
public: public:
void operator()(const Tensor &vol, const std::vector<int> &dilations, void operator()(const Tensor &vol, const std::vector<int> &dilations,
const std::vector<int> &strides, const std::vector<int> &strides,
const std::vector<int> &paddings, Tensor *col) const; const std::vector<int> &paddings, Tensor *col) const;
}; };
template <typename DeviceType, typename T> class Col2VolFunctor { template <typename DeviceType, typename T> class Col2VolFunctor {
public: public:
void operator()(const Tensor &col, const std::vector<int> &dilations, void operator()(const Tensor &col, const std::vector<int> &dilations,
const std::vector<int> &strides, const std::vector<int> &strides,
const std::vector<int> &paddings, Tensor *vol) const; const std::vector<int> &paddings, Tensor *vol) const;
}; };
} // namespace math } // namespace math
......
src/operators/mul_op.cpp
浏览文件 @ dff4a781
...@@ -22,34 +22,34 @@ namespace paddle_mobile { ...@@ -22,34 +22,34 @@ namespace paddle_mobile {
namespace operators { namespace operators {
template <typename Dtype, typename T> void MulOp<Dtype, T>::InferShape() const { template <typename Dtype, typename T> void MulOp<Dtype, T>::InferShape() const {
auto x_dims = param_.InputX()->dims(); auto x_dims = param_.InputX()->dims();
auto y_dims = param_.InputY()->dims(); auto y_dims = param_.InputY()->dims();
int x_num_col_dims = param_.XNumColDims(); int x_num_col_dims = param_.XNumColDims();
int y_num_col_dims = param_.YNumColDims(); int y_num_col_dims = param_.YNumColDims();
assert(x_dims.size() > x_num_col_dims); assert(x_dims.size() > x_num_col_dims);
assert(y_dims.size() > y_num_col_dims); assert(y_dims.size() > y_num_col_dims);
/// (1,2,3,4) , x_num_col_dims = 2 -> (2,12) /// (1,2,3,4) , x_num_col_dims = 2 -> (2,12)
auto x_mat_dims = framework::flatten_to_2d(x_dims, x_num_col_dims); auto x_mat_dims = framework::flatten_to_2d(x_dims, x_num_col_dims);
auto y_mat_dims = framework::flatten_to_2d(y_dims, y_num_col_dims); auto y_mat_dims = framework::flatten_to_2d(y_dims, y_num_col_dims);
assert(x_mat_dims[1] == y_mat_dims[0]); assert(x_mat_dims[1] == y_mat_dims[0]);
std::vector<int64_t> output_dims; std::vector<int64_t> output_dims;
output_dims.reserve( output_dims.reserve(
static_cast<size_t>(x_num_col_dims + y_dims.size() - y_num_col_dims)); static_cast<size_t>(x_num_col_dims + y_dims.size() - y_num_col_dims));
for (int i = 0; i < x_num_col_dims; ++i) { for (int i = 0; i < x_num_col_dims; ++i) {
output_dims.push_back(x_dims[i]); output_dims.push_back(x_dims[i]);
} }
for (int i = y_num_col_dims; i < y_dims.size(); ++i) { for (int i = y_num_col_dims; i < y_dims.size(); ++i) {
output_dims.push_back(y_dims[i]); output_dims.push_back(y_dims[i]);
} }
framework::DDim ddim = framework::make_ddim(output_dims); framework::DDim ddim = framework::make_ddim(output_dims);
param_.Out()->Resize(ddim); param_.Out()->Resize(ddim);
} }
template class MulOp<CPU, float>; template class MulOp<CPU, float>;
} // namespace operators } // namespace operators
......
src/operators/mul_op.h
浏览文件 @ dff4a781
...@@ -27,24 +27,24 @@ using namespace framework; ...@@ -27,24 +27,24 @@ using namespace framework;
template <typename DeviceType, typename T> template <typename DeviceType, typename T>
class MulOp : public framework::OperatorWithKernel<DeviceType> { class MulOp : public framework::OperatorWithKernel<DeviceType> {
public: public:
MulOp(const std::string &type, const VariableNameMap &inputs, MulOp(const std::string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs, const framework::AttributeMap attrs, const VariableNameMap &outputs, const framework::AttributeMap attrs,
std::shared_ptr<framework::Scope> scope) std::shared_ptr<framework::Scope> scope)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, : framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
attrs, scope), scope),
param_(inputs, outputs, attrs, *scope) {} param_(inputs, outputs, attrs, *scope) {}
void Run() const { void Run() const {
operators::MulKernel<DeviceType, T> kernel; operators::MulKernel<DeviceType, T> kernel;
kernel.Compute(param_); kernel.Compute(param_);
} }
using framework::OperatorWithKernel<DeviceType>::OperatorWithKernel; using framework::OperatorWithKernel<DeviceType>::OperatorWithKernel;
void InferShape() const override; void InferShape() const override;
protected: protected:
MulParam param_; MulParam param_;
}; };
} // namespace operators } // namespace operators
......
src/operators/op_param.cpp
浏览文件 @ dff4a781
...@@ -21,25 +21,24 @@ SOFTWARE. ...@@ -21,25 +21,24 @@ SOFTWARE.
namespace paddle_mobile { namespace paddle_mobile {
namespace operators { namespace operators {
Print &operator<<(Print &printer, const ConvParam &conv_param) { Print &operator<<(Print &printer, const ConvParam &conv_param) {
printer << "parameter of conv: " printer << "parameter of conv: "
<< "\n"; << "\n";
printer << " stride: " printer << " stride: "
<< " (" << conv_param.Strides()[0] << conv_param.Strides()[1] << " (" << conv_param.Strides()[0] << conv_param.Strides()[1] << ") "
<< ") " << "\n";
<< "\n"; printer << " paddings: "
printer << " paddings: " << " (" << conv_param.Paddings()[0] << conv_param.Paddings()[1]
<< " (" << conv_param.Paddings()[0] << conv_param.Paddings()[1] << ") "
<< ") " << "\n";
<< "\n"; printer << " dilations: "
printer << " dilations: " << " (" << conv_param.Dilations()[0] << conv_param.Dilations()[1]
<< " (" << conv_param.Dilations()[0] << conv_param.Dilations()[1] << ") "
<< ") " << "\n";
<< "\n"; printer << " groups: " << conv_param.Groups() << "\n";
printer << " groups: " << conv_param.Groups() << "\n"; printer << " input dims: " << conv_param.Input()->dims() << "\n";
printer << " input dims: " << conv_param.Input()->dims() << "\n"; printer << " filter dims: " << conv_param.Filter()->dims() << "\n";
printer << " filter dims: " << conv_param.Filter()->dims() << "\n"; printer << " output dims: " << conv_param.Output()->dims();
printer << " output dims: " << conv_param.Output()->dims(); return printer;
return printer;
} }
} // namespace operators } // namespace operators
} // namespace paddle_mobile } // namespace paddle_mobile
src/operators/op_param.h
浏览文件 @ dff4a781
...@@ -31,357 +31,355 @@ namespace operators { ...@@ -31,357 +31,355 @@ namespace operators {
using namespace framework; using namespace framework;
class OpParam : PaddleMobileObject { class OpParam : PaddleMobileObject {
public: public:
protected: protected:
template <typename T> template <typename T>
static T *InputFrom(const VariableNameMap &inputs, const Scope &scope) { static T *InputFrom(const VariableNameMap &inputs, const Scope &scope) {
return GetVarValue<T>("Input", inputs, scope); return GetVarValue<T>("Input", inputs, scope);
}
template <typename T>
static T *InputXFrom(const VariableNameMap &inputs, const Scope &scope) {
return GetVarValue<T>("X", inputs, scope);
}
template <typename T>
static T *InputYFrom(const VariableNameMap &inputs, const Scope &scope) {
return GetVarValue<T>("Y", inputs, scope);
}
template <typename T>
static T *InputBiasFrom(const VariableNameMap &inputs, const Scope &scope) {
return GetVarValue<T>("Bias", inputs, scope);
}
template <typename T>
static T *InputVarianceFrom(const VariableNameMap &inputs,
const Scope &scope) {
return GetVarValue<T>("Variance", inputs, scope);
}
template <typename T>
static T *InputMeanFrom(const VariableNameMap &inputs, const Scope &scope) {
return GetVarValue<T>("Mean", inputs, scope);
}
template <typename T>
static T *InputScaleFrom(const VariableNameMap &inputs, const Scope &scope) {
return GetVarValue<T>("Scale", inputs, scope);
}
template <typename T>
static std::vector<T *> InputMultiFrom(const VariableNameMap &inputs,
const Scope &scope) {
return GetMultiVarValue<T>("X", inputs, scope);
}
template <typename T>
static T *OutputFrom(const VariableNameMap &outputs, const Scope &scope) {
return GetVarValue<T>("Output", outputs, scope);
}
template <typename T>
static T *OutFrom(const VariableNameMap &outputs, const Scope &scope) {
return GetVarValue<T>("Out", outputs, scope);
}
template <typename T>
static T *OutputYFrom(const VariableNameMap &outputs, const Scope &scope) {
return GetVarValue<T>("Y", outputs, scope);
}
template <typename T>
static T *MidOutFrom(const VariableNameMap &outputs, const Scope &scope) {
return GetVarValue<T>("MidOut", outputs, scope);
}
template <typename T>
static T *FilterFrom(const VariableNameMap &inputs, const Scope &scope) {
return GetVarValue<T>("Filter", inputs, scope);
}
template <typename T>
static const T GetAttr(const std::string &key, const AttributeMap &map) {
return ((Attribute)map.at(key)).Get<T>();
}
template <typename T>
static T *GetVarValue(const std::string &key, const VariableNameMap &var_map,
const Scope &scope) {
auto var_vec = var_map.at(key);
if (!var_vec.empty()) {
// std::cout << " get var value -- " << var_vec[0] <<
// std::endl;
auto var = scope.FindVar(var_vec[0]);
return var->GetMutable<T>();
} else {
return nullptr;
} }
}
template <typename T> template <typename T>
static T *InputXFrom(const VariableNameMap &inputs, const Scope &scope) { static std::vector<T *> GetMultiVarValue(const std::string &key,
return GetVarValue<T>("X", inputs, scope); const VariableNameMap &var_map,
}
template <typename T>
static T *InputYFrom(const VariableNameMap &inputs, const Scope &scope) {
return GetVarValue<T>("Y", inputs, scope);
}
template <typename T>
static T *InputBiasFrom(const VariableNameMap &inputs, const Scope &scope) {
return GetVarValue<T>("Bias", inputs, scope);
}
template <typename T>
static T *InputVarianceFrom(const VariableNameMap &inputs,
const Scope &scope) {
return GetVarValue<T>("Variance", inputs, scope);
}
template <typename T>
static T *InputMeanFrom(const VariableNameMap &inputs, const Scope &scope) {
return GetVarValue<T>("Mean", inputs, scope);
}
template <typename T>
static T *InputScaleFrom(const VariableNameMap &inputs,
const Scope &scope) {
return GetVarValue<T>("Scale", inputs, scope);
}
template <typename T>
static std::vector<T *> InputMultiFrom(const VariableNameMap &inputs,
const Scope &scope) { const Scope &scope) {
return GetMultiVarValue<T>("X", inputs, scope); auto var_vecs = var_map.at(key);
} assert(var_vecs.size() > 1);
std::vector<T *> var_res;
template <typename T> for (auto &var_vec : var_vecs) {
static T *OutputFrom(const VariableNameMap &outputs, const Scope &scope) { auto var = scope.FindVar(var_vec);
return GetVarValue<T>("Output", outputs, scope); var_res.push_back(var->GetMutable<T>());
}
template <typename T>
static T *OutFrom(const VariableNameMap &outputs, const Scope &scope) {
return GetVarValue<T>("Out", outputs, scope);
}
template <typename T>
static T *OutputYFrom(const VariableNameMap &outputs, const Scope &scope) {
return GetVarValue<T>("Y", outputs, scope);
}
template <typename T>
static T *MidOutFrom(const VariableNameMap &outputs, const Scope &scope) {
return GetVarValue<T>("MidOut", outputs, scope);
}
template <typename T>
static T *FilterFrom(const VariableNameMap &inputs, const Scope &scope) {
return GetVarValue<T>("Filter", inputs, scope);
}
template <typename T>
static const T GetAttr(const std::string &key, const AttributeMap &map) {
return ((Attribute)map.at(key)).Get<T>();
}
template <typename T>
static T *GetVarValue(const std::string &key,
const VariableNameMap &var_map, const Scope &scope) {
auto var_vec = var_map.at(key);
if (!var_vec.empty()) {
// std::cout << " get var value -- " << var_vec[0] <<
// std::endl;
auto var = scope.FindVar(var_vec[0]);
return var->GetMutable<T>();
} else {
return nullptr;
}
}
template <typename T>
static std::vector<T *> GetMultiVarValue(const std::string &key,
const VariableNameMap &var_map,
const Scope &scope) {
auto var_vecs = var_map.at(key);
assert(var_vecs.size() > 1);
std::vector<T *> var_res;
for (auto &var_vec : var_vecs) {
auto var = scope.FindVar(var_vec);
var_res.push_back(var->GetMutable<T>());
}
return var_res;
} }
return var_res;
}
}; };
class ConvParam : OpParam { class ConvParam : OpParam {
public: public:
ConvParam(const VariableNameMap &inputs, const VariableNameMap &outputs, ConvParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
const framework::AttributeMap &attrs, const framework::AttributeMap &attrs,
const framework::Scope &scope) { const framework::Scope &scope) {
filter_ = FilterFrom<framework::LoDTensor>(inputs, scope); filter_ = FilterFrom<framework::LoDTensor>(inputs, scope);
input_ = InputFrom<framework::Tensor>(inputs, scope); input_ = InputFrom<framework::Tensor>(inputs, scope);
output_ = OutputFrom<framework::Tensor>(outputs, scope); output_ = OutputFrom<framework::Tensor>(outputs, scope);
strides_ = GetAttr<std::vector<int>>("strides", attrs); strides_ = GetAttr<std::vector<int>>("strides", attrs);
paddings_ = GetAttr<std::vector<int>>("paddings", attrs); paddings_ = GetAttr<std::vector<int>>("paddings", attrs);
dilations_ = GetAttr<std::vector<int>>("dilations", attrs); dilations_ = GetAttr<std::vector<int>>("dilations", attrs);
groups = GetAttr<int>("groups", attrs); groups = GetAttr<int>("groups", attrs);
} }
const Tensor *Input() const { return input_; } const Tensor *Input() const { return input_; }
const LoDTensor *Filter() const { return filter_; } const LoDTensor *Filter() const { return filter_; }
Tensor *Output() const { return output_; } Tensor *Output() const { return output_; }
const std::vector<int> &Strides() const { return strides_; } const std::vector<int> &Strides() const { return strides_; }
const std::vector<int> &Paddings() const { return paddings_; } const std::vector<int> &Paddings() const { return paddings_; }
const std::vector<int> &Dilations() const { return dilations_; } const std::vector<int> &Dilations() const { return dilations_; }
const int &Groups() const { return groups; } const int &Groups() const { return groups; }
private: private:
Tensor *input_; Tensor *input_;
Tensor *output_; Tensor *output_;
LoDTensor *filter_; LoDTensor *filter_;
std::vector<int> strides_; std::vector<int> strides_;
std::vector<int> paddings_; std::vector<int> paddings_;
std::vector<int> dilations_; std::vector<int> dilations_;
int groups; int groups;
}; };
Print &operator<<(Print &printer, const ConvParam &conv_param); Print &operator<<(Print &printer, const ConvParam &conv_param);
class ElementwiseAddParam : OpParam { class ElementwiseAddParam : OpParam {
public: public:
ElementwiseAddParam(const VariableNameMap &inputs, ElementwiseAddParam(const VariableNameMap &inputs,
const VariableNameMap &outputs, const VariableNameMap &outputs,
const framework::AttributeMap &attrs, const framework::AttributeMap &attrs,
const framework::Scope &scope) { const framework::Scope &scope) {
input_x_ = InputXFrom<framework::Tensor>(inputs, scope); input_x_ = InputXFrom<framework::Tensor>(inputs, scope);
input_y_ = InputYFrom<framework::Tensor>(inputs, scope); input_y_ = InputYFrom<framework::Tensor>(inputs, scope);
out_ = OutFrom<framework::Tensor>(outputs, scope); out_ = OutFrom<framework::Tensor>(outputs, scope);
axis_ = GetAttr<int>("axis", attrs); axis_ = GetAttr<int>("axis", attrs);
} }
const Tensor *InputX() const { return input_x_; } const Tensor *InputX() const { return input_x_; }
const Tensor *InputY() const { return input_y_; } const Tensor *InputY() const { return input_y_; }
Tensor *Out() const { return out_; } Tensor *Out() const { return out_; }
const int &Axis() const { return axis_; } const int &Axis() const { return axis_; }
private: private:
Tensor *input_x_; Tensor *input_x_;
Tensor *input_y_; Tensor *input_y_;
Tensor *out_; Tensor *out_;
int axis_; int axis_;
}; };
class MulParam : OpParam { class MulParam : OpParam {
public: public:
MulParam(const VariableNameMap &inputs, const VariableNameMap &outputs, MulParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
const framework::AttributeMap &attrs, const framework::AttributeMap &attrs,
const framework::Scope &scope) { const framework::Scope &scope) {
input_x_ = InputXFrom<framework::Tensor>(inputs, scope); input_x_ = InputXFrom<framework::Tensor>(inputs, scope);
input_y_ = InputYFrom<framework::Tensor>(inputs, scope); input_y_ = InputYFrom<framework::Tensor>(inputs, scope);
out_ = OutFrom<framework::Tensor>(outputs, scope); out_ = OutFrom<framework::Tensor>(outputs, scope);
x_num_col_dims_ = GetAttr<int>("x_num_col_dims", attrs); x_num_col_dims_ = GetAttr<int>("x_num_col_dims", attrs);
y_num_col_dims_ = GetAttr<int>("y_num_col_dims", attrs); y_num_col_dims_ = GetAttr<int>("y_num_col_dims", attrs);
} }
const Tensor *InputX() const { return input_x_; } const Tensor *InputX() const { return input_x_; }
const Tensor *InputY() const { return input_y_; } const Tensor *InputY() const { return input_y_; }
Tensor *Out() const { return out_; } Tensor *Out() const { return out_; }
const int &XNumColDims() const { return x_num_col_dims_; } const int &XNumColDims() const { return x_num_col_dims_; }
const int &YNumColDims() const { return y_num_col_dims_; } const int &YNumColDims() const { return y_num_col_dims_; }
private: private:
Tensor *input_x_; Tensor *input_x_;
Tensor *input_y_; Tensor *input_y_;
Tensor *out_; Tensor *out_;
int x_num_col_dims_; int x_num_col_dims_;
int y_num_col_dims_; int y_num_col_dims_;
}; };
class ConcatParam : public OpParam { class ConcatParam : public OpParam {
public: public:
ConcatParam(const VariableNameMap &inputs, const VariableNameMap &outputs, ConcatParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
const framework::AttributeMap &attrs, const framework::AttributeMap &attrs,
const framework::Scope &scope) { const framework::Scope &scope) {
inputs_ = InputMultiFrom<framework::Tensor>(inputs, scope); inputs_ = InputMultiFrom<framework::Tensor>(inputs, scope);
out_ = OutFrom<framework::Tensor>(outputs, scope); out_ = OutFrom<framework::Tensor>(outputs, scope);
axis_ = GetAttr<int>("axis", attrs); axis_ = GetAttr<int>("axis", attrs);
} }
std::vector<Tensor *> Inputs() const { return inputs_; } std::vector<Tensor *> Inputs() const { return inputs_; }
Tensor *Out() const { return out_; } Tensor *Out() const { return out_; }
const int &Axis() const { return axis_; } const int &Axis() const { return axis_; }
private: private:
std::vector<Tensor *> inputs_; std::vector<Tensor *> inputs_;
Tensor *out_; Tensor *out_;
int axis_; int axis_;
}; };
class LrnParam : public OpParam { class LrnParam : public OpParam {
public: public:
LrnParam(const VariableNameMap &inputs, const VariableNameMap &outputs, LrnParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
const framework::AttributeMap &attrs, const framework::AttributeMap &attrs,
const framework::Scope &scope) { const framework::Scope &scope) {
input_x_ = InputXFrom<framework::Tensor>(inputs, scope); input_x_ = InputXFrom<framework::Tensor>(inputs, scope);
out_ = OutFrom<framework::Tensor>(outputs, scope); out_ = OutFrom<framework::Tensor>(outputs, scope);
mid_out_ = MidOutFrom<framework::Tensor>(outputs, scope); mid_out_ = MidOutFrom<framework::Tensor>(outputs, scope);
n_ = GetAttr<int>("n", attrs); n_ = GetAttr<int>("n", attrs);
alpha_ = GetAttr<float>("alpha", attrs); alpha_ = GetAttr<float>("alpha", attrs);
beta_ = GetAttr<float>("beta", attrs); beta_ = GetAttr<float>("beta", attrs);
k_ = GetAttr<float>("k", attrs); k_ = GetAttr<float>("k", attrs);
data_format_ = GetAttr<std::string>("data_format", attrs); data_format_ = GetAttr<std::string>("data_format", attrs);
} }
const Tensor *InputX() const { return input_x_; } const Tensor *InputX() const { return input_x_; }
Tensor *Out() const { return out_; } Tensor *Out() const { return out_; }
Tensor *MidOut() const { return mid_out_; } Tensor *MidOut() const { return mid_out_; }
const int &N() const { return n_; } const int &N() const { return n_; }
const float &Alpha() const { return alpha_; } const float &Alpha() const { return alpha_; }
const float &Beta() const { return beta_; } const float &Beta() const { return beta_; }
const float &K() const { return k_; } const float &K() const { return k_; }
const std::string &DataFormat() const { return data_format_; } const std::string &DataFormat() const { return data_format_; }
private: private:
Tensor *input_x_; Tensor *input_x_;
Tensor *out_; Tensor *out_;
Tensor *mid_out_; Tensor *mid_out_;
int n_; int n_;
float alpha_; float alpha_;
float beta_; float beta_;
float k_; float k_;
std::string data_format_; std::string data_format_;
}; };
class BatchNormParam : OpParam { class BatchNormParam : OpParam {
public: public:
BatchNormParam(const VariableNameMap &inputs, BatchNormParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
const VariableNameMap &outputs, const framework::AttributeMap &attrs,
const framework::AttributeMap &attrs, const framework::Scope &scope) {
const framework::Scope &scope) { input_x_ = InputXFrom<framework::Tensor>(inputs, scope);
input_x_ = InputXFrom<framework::Tensor>(inputs, scope); output_y_ = OutputYFrom<framework::Tensor>(outputs, scope);
output_y_ = OutputYFrom<framework::Tensor>(outputs, scope); input_bias_ = InputBiasFrom<framework::Tensor>(inputs, scope);
input_bias_ = InputBiasFrom<framework::Tensor>(inputs, scope); input_mean_ = InputMeanFrom<framework::Tensor>(inputs, scope);
input_mean_ = InputMeanFrom<framework::Tensor>(inputs, scope); input_scale_ = InputScaleFrom<framework::Tensor>(inputs, scope);
input_scale_ = InputScaleFrom<framework::Tensor>(inputs, scope); input_variance_ = InputVarianceFrom<framework::Tensor>(inputs, scope);
input_variance_ = InputVarianceFrom<framework::Tensor>(inputs, scope); epsilon_ = GetAttr<float>("epsilon", attrs);
epsilon_ = GetAttr<float>("epsilon", attrs); momentum_ = GetAttr<float>("momentum", attrs);
momentum_ = GetAttr<float>("momentum", attrs); is_test_ = GetAttr<bool>("is_test", attrs);
is_test_ = GetAttr<bool>("is_test", attrs); }
}
const Tensor *InputX() const { return input_x_; } const Tensor *InputX() const { return input_x_; }
Tensor *OutputY() const { return output_y_; } Tensor *OutputY() const { return output_y_; }
const Tensor *InputBias() const { return input_bias_; } const Tensor *InputBias() const { return input_bias_; }
const Tensor *InputMean() const { return input_mean_; } const Tensor *InputMean() const { return input_mean_; }
const Tensor *InputScale() const { return input_scale_; } const Tensor *InputScale() const { return input_scale_; }
const Tensor *InputVariance() const { return input_variance_; } const Tensor *InputVariance() const { return input_variance_; }
const float &Epsilon() const { return epsilon_; } const float &Epsilon() const { return epsilon_; }
const float &Momentum() const { return momentum_; } const float &Momentum() const { return momentum_; }
const bool &IsTest() const { return is_test_; } const bool &IsTest() const { return is_test_; }
const std::string &DataFormat() const { return data_format_; } const std::string &DataFormat() const { return data_format_; }
private: private:
Tensor *input_x_; Tensor *input_x_;
Tensor *output_y_; Tensor *output_y_;
Tensor *input_bias_; Tensor *input_bias_;
Tensor *input_mean_; Tensor *input_mean_;
Tensor *input_scale_; Tensor *input_scale_;
Tensor *input_variance_; Tensor *input_variance_;
float epsilon_; float epsilon_;
float momentum_; float momentum_;
bool is_test_; bool is_test_;
std::string data_format_; std::string data_format_;
}; };
class PoolParam : public OpParam { class PoolParam : public OpParam {
public: public:
PoolParam(const VariableNameMap &inputs, const VariableNameMap &outputs, PoolParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
const framework::AttributeMap &attrs, const framework::AttributeMap &attrs,
const framework::Scope &scope) { const framework::Scope &scope) {
input_ = InputXFrom<framework::Tensor>(inputs, scope); input_ = InputXFrom<framework::Tensor>(inputs, scope);
output_ = OutFrom<framework::Tensor>(outputs, scope); output_ = OutFrom<framework::Tensor>(outputs, scope);
pooling_type_ = GetAttr<std::string>("pooling_type", attrs); pooling_type_ = GetAttr<std::string>("pooling_type", attrs);
ksize_ = GetAttr<std::vector<int>>("ksize", attrs); ksize_ = GetAttr<std::vector<int>>("ksize", attrs);
strides_ = GetAttr<std::vector<int>>("strides", attrs); strides_ = GetAttr<std::vector<int>>("strides", attrs);
paddings_ = GetAttr<std::vector<int>>("paddings", attrs); paddings_ = GetAttr<std::vector<int>>("paddings", attrs);
ceil_mode_ = GetAttr<bool>("ceil_mode", attrs); ceil_mode_ = GetAttr<bool>("ceil_mode", attrs);
gloabal_pooling_ = GetAttr<bool>("global_pooling", attrs); gloabal_pooling_ = GetAttr<bool>("global_pooling", attrs);
} }
const Tensor *Input() const { return input_; } const Tensor *Input() const { return input_; }
Tensor *Output() const { return output_; } Tensor *Output() const { return output_; }
const std::string &PoolingType() const { return pooling_type_; } const std::string &PoolingType() const { return pooling_type_; }
const std::vector<int> &Ksize() const { return ksize_; } const std::vector<int> &Ksize() const { return ksize_; }
const std::vector<int> &Strides() const { return strides_; } const std::vector<int> &Strides() const { return strides_; }
const std::vector<int> &Paddings() const { return paddings_; } const std::vector<int> &Paddings() const { return paddings_; }
bool isCeilMode() const { return ceil_mode_; } bool isCeilMode() const { return ceil_mode_; }
bool isGlobalPooling() const { return gloabal_pooling_; } bool isGlobalPooling() const { return gloabal_pooling_; }
private: private:
Tensor *input_; Tensor *input_;
Tensor *output_; Tensor *output_;
std::string pooling_type_; std::string pooling_type_;
std::vector<int> ksize_; std::vector<int> ksize_;
std::vector<int> strides_; std::vector<int> strides_;
std::vector<int> paddings_; std::vector<int> paddings_;
bool ceil_mode_; bool ceil_mode_;
bool gloabal_pooling_ = false; bool gloabal_pooling_ = false;
}; };
} // namespace operators } // namespace operators
......
src/operators/pool_op.cpp
浏览文件 @ dff4a781
...@@ -23,37 +23,37 @@ namespace operators { ...@@ -23,37 +23,37 @@ namespace operators {
int PoolOutputSize(int input_size, int filter_size, int padding, int stride, int PoolOutputSize(int input_size, int filter_size, int padding, int stride,
bool ceil_mode) { bool ceil_mode) {
int output_size; int output_size;
if (!ceil_mode) { if (!ceil_mode) {
output_size = (input_size - filter_size + 2 * padding) / stride + 1; output_size = (input_size - filter_size + 2 * padding) / stride + 1;
} else { } else {
output_size = output_size =
(input_size - filter_size + 2 * padding + stride - 1) / stride + 1; (input_size - filter_size + 2 * padding + stride - 1) / stride + 1;
} }
return output_size; return output_size;
} }
template <typename DeviceType, typename T> template <typename DeviceType, typename T>
void PoolOp<DeviceType, T>::InferShape() const { void PoolOp<DeviceType, T>::InferShape() const {
auto in_x_dims = param_.Input()->dims(); auto in_x_dims = param_.Input()->dims();
std::vector<int> ksize = param_.Ksize(); std::vector<int> ksize = param_.Ksize();
std::vector<int> paddings = param_.Paddings(); std::vector<int> paddings = param_.Paddings();
std::vector<int> strides = param_.Strides(); std::vector<int> strides = param_.Strides();
bool ceil_mode = param_.isCeilMode(); bool ceil_mode = param_.isCeilMode();
if (param_.isGlobalPooling()) { if (param_.isGlobalPooling()) {
ksize.resize(static_cast<size_t>(in_x_dims.size()) - 2); ksize.resize(static_cast<size_t>(in_x_dims.size()) - 2);
for (size_t i = 0; i < ksize.size(); ++i) {
paddings[i] = 0;
ksize[i] = static_cast<int>(in_x_dims[i + 2]);
}
}
std::vector<int64_t> output_shape({in_x_dims[0], in_x_dims[1]});
for (size_t i = 0; i < ksize.size(); ++i) { for (size_t i = 0; i < ksize.size(); ++i) {
output_shape.push_back(PoolOutputSize( paddings[i] = 0;
in_x_dims[i + 2], ksize[i], paddings[i], strides[i], ceil_mode)); ksize[i] = static_cast<int>(in_x_dims[i + 2]);
} }
param_.Output()->Resize(framework::make_ddim(output_shape)); }
DLOG << "infer shape out size =" << param_.Output()->numel(); std::vector<int64_t> output_shape({in_x_dims[0], in_x_dims[1]});
for (size_t i = 0; i < ksize.size(); ++i) {
output_shape.push_back(PoolOutputSize(in_x_dims[i + 2], ksize[i],
paddings[i], strides[i], ceil_mode));
}
param_.Output()->Resize(framework::make_ddim(output_shape));
DLOG << "infer shape out size =" << param_.Output()->numel();
} }
template class PoolOp<CPU, float>; template class PoolOp<CPU, float>;
} // namespace operators } // namespace operators
......
src/operators/pool_op.h
浏览文件 @ dff4a781
...@@ -28,25 +28,25 @@ using namespace framework; ...@@ -28,25 +28,25 @@ using namespace framework;
template <typename DeviceType, typename T> template <typename DeviceType, typename T>
class PoolOp : public framework::OperatorWithKernel<DeviceType> { class PoolOp : public framework::OperatorWithKernel<DeviceType> {
public: public:
PoolOp(const std::string &type, const VariableNameMap &inputs, PoolOp(const std::string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs, const framework::AttributeMap &attrs, const VariableNameMap &outputs, const framework::AttributeMap &attrs,
std::shared_ptr<framework::Scope> scope) std::shared_ptr<framework::Scope> scope)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, : framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
attrs, scope), scope),
param_(inputs, outputs, attrs, *scope) {} param_(inputs, outputs, attrs, *scope) {}
using framework::OperatorWithKernel<DeviceType>::OperatorWithKernel; using framework::OperatorWithKernel<DeviceType>::OperatorWithKernel;
void InferShape() const override; void InferShape() const override;
void Run() const { void Run() const {
// InferShape(); // InferShape();
operators::PoolKernel<DeviceType, T> kernel; operators::PoolKernel<DeviceType, T> kernel;
kernel.Compute(param_); kernel.Compute(param_);
this->ClearVariables({"X"}); this->ClearVariables({"X"});
} }
private: private:
PoolParam param_; PoolParam param_;
}; };
} // namespace operators } // namespace operators
} // namespace paddle_mobile } // namespace paddle_mobile
src/platform/data_type.h
浏览文件 @ dff4a781
...@@ -22,103 +22,103 @@ namespace paddle_mobile { ...@@ -22,103 +22,103 @@ namespace paddle_mobile {
namespace framework { namespace framework {
inline proto::VarType::Type ToDataType(std::type_index type) { inline proto::VarType::Type ToDataType(std::type_index type) {
/*if (typeid(platform::float16).hash_code() == type.hash_code()) { /*if (typeid(platform::float16).hash_code() == type.hash_code()) {
return proto::VarType::FP16; return proto::VarType::FP16;
} else */ } else */
if (typeid(const float).hash_code() == type.hash_code()) { if (typeid(const float).hash_code() == type.hash_code()) {
// CPPLint complains Using C-style cast. Use // CPPLint complains Using C-style cast. Use
// static_cast<float>() instead // static_cast<float>() instead
// One fix to this is to replace float with const float because // One fix to this is to replace float with const float because
// typeid(T) == typeid(const T) // typeid(T) == typeid(const T)
// http://en.cppreference.com/w/cpp/language/typeid // http://en.cppreference.com/w/cpp/language/typeid
return proto::VarType::FP32; return proto::VarType::FP32;
} else if (typeid(const double).hash_code() == type.hash_code()) { } else if (typeid(const double).hash_code() == type.hash_code()) {
return proto::VarType::FP64; return proto::VarType::FP64;
} else if (typeid(const int).hash_code() == type.hash_code()) { } else if (typeid(const int).hash_code() == type.hash_code()) {
return proto::VarType::INT32; return proto::VarType::INT32;
} else if (typeid(const int64_t).hash_code() == type.hash_code()) { } else if (typeid(const int64_t).hash_code() == type.hash_code()) {
return proto::VarType::INT64; return proto::VarType::INT64;
} else if (typeid(const bool).hash_code() == type.hash_code()) { } else if (typeid(const bool).hash_code() == type.hash_code()) {
return proto::VarType::BOOL; return proto::VarType::BOOL;
} else { } else {
// PADDLE_THROW("Not supported"); // PADDLE_THROW("Not supported");
// std::cout << "Not supported"; // std::cout << "Not supported";
} }
} }
inline std::type_index ToTypeIndex(proto::VarType::Type type) { inline std::type_index ToTypeIndex(proto::VarType::Type type) {
switch (type) { switch (type) {
// case proto::VarType::FP16: // case proto::VarType::FP16:
// return typeid(platform::float16); // return typeid(platform::float16);
case proto::VarType::FP32: case proto::VarType::FP32:
return typeid(float); return typeid(float);
case proto::VarType::FP64: case proto::VarType::FP64:
return typeid(double); return typeid(double);
case proto::VarType::INT32: case proto::VarType::INT32:
return typeid(int); return typeid(int);
case proto::VarType::INT64: case proto::VarType::INT64:
return typeid(int64_t); return typeid(int64_t);
case proto::VarType::BOOL: case proto::VarType::BOOL:
return typeid(bool); return typeid(bool);
default: default:
// PADDLE_THROW("Not support type %d", type); // PADDLE_THROW("Not support type %d", type);
printf("Not support type %d", type); printf("Not support type %d", type);
} }
} }
template <typename Visitor> template <typename Visitor>
inline void VisitDataType(proto::VarType::Type type, Visitor visitor) { inline void VisitDataType(proto::VarType::Type type, Visitor visitor) {
switch (type) { switch (type) {
// case proto::VarType::FP16: // case proto::VarType::FP16:
// visitor.template operator()<platform::float16>(); // visitor.template operator()<platform::float16>();
// break; // break;
case proto::VarType::FP32: case proto::VarType::FP32:
visitor.template operator()<float>(); visitor.template operator()<float>();
break; break;
case proto::VarType::FP64: case proto::VarType::FP64:
visitor.template operator()<double>(); visitor.template operator()<double>();
break; break;
case proto::VarType::INT32: case proto::VarType::INT32:
visitor.template operator()<int>(); visitor.template operator()<int>();
break; break;
case proto::VarType::INT64: case proto::VarType::INT64:
visitor.template operator()<int64_t>(); visitor.template operator()<int64_t>();
break; break;
case proto::VarType::BOOL: case proto::VarType::BOOL:
visitor.template operator()<bool>(); visitor.template operator()<bool>();
break; break;
default: default:
// PADDLE_THROW("Not supported"); // PADDLE_THROW("Not supported");
printf("Not supported"); printf("Not supported");
} }
} }
inline std::string DataTypeToString(const proto::VarType::Type type) { inline std::string DataTypeToString(const proto::VarType::Type type) {
switch (type) { switch (type) {
case proto::VarType::FP16: case proto::VarType::FP16:
return "float16"; return "float16";
case proto::VarType::FP32: case proto::VarType::FP32:
return "float32"; return "float32";
case proto::VarType::FP64: case proto::VarType::FP64:
return "float64"; return "float64";
case proto::VarType::INT16: case proto::VarType::INT16:
return "int16"; return "int16";
case proto::VarType::INT32: case proto::VarType::INT32:
return "int32"; return "int32";
case proto::VarType::INT64: case proto::VarType::INT64:
return "int64"; return "int64";
case proto::VarType::BOOL: case proto::VarType::BOOL:
return "bool"; return "bool";
default: default:
// PADDLE_THROW("Not support type %d", type); // PADDLE_THROW("Not support type %d", type);
printf("Not support type %d", type); printf("Not support type %d", type);
} }
} }
inline std::ostream &operator<<(std::ostream &out, inline std::ostream &operator<<(std::ostream &out,
const proto::VarType::Type &type) { const proto::VarType::Type &type) {
out << DataTypeToString(type); out << DataTypeToString(type);
return out; return out;
} }
} // namespace framework } // namespace framework
......
src/platform/macros.h
浏览文件 @ dff4a781
...@@ -17,9 +17,9 @@ limitations under the License. */ ...@@ -17,9 +17,9 @@ limitations under the License. */
// Disable the copy and assignment operator for a class. // Disable the copy and assignment operator for a class.
#ifndef DISABLE_COPY_AND_ASSIGN #ifndef DISABLE_COPY_AND_ASSIGN
#define DISABLE_COPY_AND_ASSIGN(classname) \ #define DISABLE_COPY_AND_ASSIGN(classname) \
private: \ private: \
classname(const classname &) = delete; \ classname(const classname &) = delete; \
classname(classname &&) = delete; \ classname(classname &&) = delete; \
classname &operator=(const classname &) = delete; \ classname &operator=(const classname &) = delete; \
classname &operator=(classname &&) = delete classname &operator=(classname &&) = delete
#endif #endif
test/common/test_log.cpp
浏览文件 @ dff4a781
...@@ -20,20 +20,20 @@ SOFTWARE. ...@@ -20,20 +20,20 @@ SOFTWARE.
int main() { int main() {
DLOGF("DASJFDAFJ%d -- %f", 12345, 344.234); DLOGF("DASJFDAFJ%d -- %f", 12345, 344.234);
LOGF(paddle_mobile::kLOG_DEBUG, "DASJFDAFJ%d -- %f", 12345, 344.234); LOGF(paddle_mobile::kLOG_DEBUG, "DASJFDAFJ%d -- %f", 12345, 344.234);
LOG(paddle_mobile::kLOG_DEBUG) << "test debug" LOG(paddle_mobile::kLOG_DEBUG) << "test debug"
<< " next log"; << " next log";
LOG(paddle_mobile::kLOG_DEBUG1) << "test debug1" LOG(paddle_mobile::kLOG_DEBUG1) << "test debug1"
<< " next log"; << " next log";
LOG(paddle_mobile::kLOG_DEBUG2) << "test debug2" LOG(paddle_mobile::kLOG_DEBUG2) << "test debug2"
<< " next log"; << " next log";
DLOG << "test DLOG"; DLOG << "test DLOG";
LOG(paddle_mobile::kLOG_ERROR) << " error occur !"; LOG(paddle_mobile::kLOG_ERROR) << " error occur !";
return 0; return 0;
} }
test/framework/executor_for_test.cpp
浏览文件 @ dff4a781
...@@ -23,29 +23,29 @@ Executor4Test<DeviceType, OpType>::Executor4Test(const Program<DeviceType> p, ...@@ -23,29 +23,29 @@ Executor4Test<DeviceType, OpType>::Executor4Test(const Program<DeviceType> p,
std::string op_type) std::string op_type)
: Executor<DeviceType>(p) { : Executor<DeviceType>(p) {
if (this->program_.originProgram == nullptr) { if (this->program_.originProgram == nullptr) {
LOG(paddle_mobile::LogLevel::kLOG_ERROR) LOG(paddle_mobile::LogLevel::kLOG_ERROR)
<< "to_predict_program_ == nullptr"; << "to_predict_program_ == nullptr";
} }
const std::vector<std::shared_ptr<BlockDesc>> blocks = const std::vector<std::shared_ptr<BlockDesc>> blocks =
this->to_predict_program_->Blocks(); this->to_predict_program_->Blocks();
for (int i = 0; i < blocks.size(); ++i) { for (int i = 0; i < blocks.size(); ++i) {
std::shared_ptr<BlockDesc> block_desc = blocks[i]; std::shared_ptr<BlockDesc> block_desc = blocks[i];
std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops(); std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops();
for (int j = 0; j < ops.size(); ++j) { for (int j = 0; j < ops.size(); ++j) {
std::shared_ptr<OpDesc> op = ops[j]; std::shared_ptr<OpDesc> op = ops[j];
if (op->Type() == op_type) { if (op->Type() == op_type) {
std::shared_ptr<OpType> op_ptr = std::make_shared<OpType>( std::shared_ptr<OpType> op_ptr = std::make_shared<OpType>(
op->Type(), op->GetInputs(), op->GetOutputs(), op->Type(), op->GetInputs(), op->GetOutputs(), op->GetAttrMap(),
op->GetAttrMap(), this->program_.scope); this->program_.scope);
this->ops_of_block_[*block_desc.get()].push_back(op_ptr); this->ops_of_block_[*block_desc.get()].push_back(op_ptr);
break; break;
} }
}
} }
}
} }
template <typename DeviceType, typename OpType> template <typename DeviceType, typename OpType>
...@@ -53,19 +53,19 @@ std::shared_ptr<Tensor> ...@@ -53,19 +53,19 @@ std::shared_ptr<Tensor>
Executor4Test<DeviceType, OpType>::predict(Tensor &t, std::string input, Executor4Test<DeviceType, OpType>::predict(Tensor &t, std::string input,
std::string output, DDim dDim) { std::string output, DDim dDim) {
auto scope = this->program_.scope; auto scope = this->program_.scope;
Variable *g_feed_value = scope->Var(input); Variable *g_feed_value = scope->Var(input);
auto tensor = g_feed_value->GetMutable<Tensor>(); auto tensor = g_feed_value->GetMutable<Tensor>();
tensor->ShareDataWith(t); tensor->ShareDataWith(t);
Variable *con_output = scope->Var(output); Variable *con_output = scope->Var(output);
Tensor *output_tensor = con_output->GetMutable<Tensor>(); Tensor *output_tensor = con_output->GetMutable<Tensor>();
output_tensor->mutable_data<float>(dDim); output_tensor->mutable_data<float>(dDim);
std::shared_ptr<Tensor> out_tensor = std::make_shared<LoDTensor>(); std::shared_ptr<Tensor> out_tensor = std::make_shared<LoDTensor>();
out_tensor.reset(output_tensor); out_tensor.reset(output_tensor);
Executor<DeviceType>::predict(t, 0); Executor<DeviceType>::predict(t, 0);
return out_tensor; return out_tensor;
} }
template class Executor4Test< template class Executor4Test<
......
test/framework/executor_for_test.h
浏览文件 @ dff4a781
...@@ -27,9 +27,9 @@ using namespace paddle_mobile::framework; ...@@ -27,9 +27,9 @@ using namespace paddle_mobile::framework;
template <typename DeviceType, typename OpType> template <typename DeviceType, typename OpType>
class Executor4Test : public Executor<DeviceType> { class Executor4Test : public Executor<DeviceType> {
public: public:
Executor4Test(const Program<DeviceType> p, std::string op_type); Executor4Test(const Program<DeviceType> p, std::string op_type);
std::shared_ptr<Tensor> predict(Tensor &t, std::string input, std::shared_ptr<Tensor> predict(Tensor &t, std::string input,
std::string output, DDim dDim); std::string output, DDim dDim);
}; };
test/framework/test_load.cpp
浏览文件 @ dff4a781
...@@ -19,10 +19,10 @@ SOFTWARE. ...@@ -19,10 +19,10 @@ SOFTWARE.
#include "io.h" #include "io.h"
int main() { int main() {
paddle_mobile::Loader<paddle_mobile::CPU> loader; paddle_mobile::Loader<paddle_mobile::CPU> loader;
//../../../test/models/googlenet //../../../test/models/googlenet
//../../../test/models/mobilenet //../../../test/models/mobilenet
auto program = loader.Load(std::string("../models/googlenet")); auto program = loader.Load(std::string("../models/googlenet"));
return 0; return 0;
} }
\ No newline at end of file
test/framework/test_optimize.cpp
浏览文件 @ dff4a781
...@@ -25,16 +25,16 @@ using namespace paddle_mobile::framework; ...@@ -25,16 +25,16 @@ using namespace paddle_mobile::framework;
int main() { int main() {
Loader<paddle_mobile::CPU> loader; Loader<paddle_mobile::CPU> loader;
// "../../../test/models/googlenet" // "../../../test/models/googlenet"
auto program = loader.Load("../models/googlenet"); auto program = loader.Load("../models/googlenet");
ProgramOptimize optimize; ProgramOptimize optimize;
auto optimize_program = optimize.FushionOptimize(program.originProgram); auto optimize_program = optimize.FushionOptimize(program.originProgram);
if (optimize_program) { if (optimize_program) {
} else { } else {
DLOG << "optimize_program is null"; DLOG << "optimize_program is null";
} }
} }
test/operators/test_batchnorm_op.cpp
浏览文件 @ dff4a781
...@@ -24,102 +24,100 @@ namespace paddle_mobile { ...@@ -24,102 +24,100 @@ namespace paddle_mobile {
namespace framework { namespace framework {
template <typename Dtype> class TestBatchNormOp { template <typename Dtype> class TestBatchNormOp {
public: public:
explicit TestBatchNormOp(const Program<Dtype> p) : program_(p) { explicit TestBatchNormOp(const Program<Dtype> p) : program_(p) {
if (use_optimize_) { if (use_optimize_) {
to_predict_program_ = program_.optimizeProgram; to_predict_program_ = program_.optimizeProgram;
} else { } else {
to_predict_program_ = program_.originProgram; to_predict_program_ = program_.originProgram;
}
const std::vector<std::shared_ptr<BlockDesc>> blocks =
to_predict_program_->Blocks();
// DLOG << " **block size " << blocks.size();
for (int i = 0; i < blocks.size(); ++i) {
std::shared_ptr<BlockDesc> block_desc = blocks[i];
std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops();
// DLOG << " ops " << ops.size();
for (int j = 0; j < ops.size(); ++j) {
std::shared_ptr<OpDesc> op = ops[j];
if (op->Type() == "batch_norm" &&
op->Input("X")[0] == "conv2d_0.tmp_0") {
DLOG << " mul attr size: " << op->GetAttrMap().size();
DLOG << " inputs size: " << op->GetInputs().size();
DLOG << " outputs size: " << op->GetOutputs().size();
DLOG << " Input X is : " << op->Input("X")[0];
DLOG << " Input Mean is : " << op->Input("Mean")[0];
DLOG << " Input Variance is : " << op->Input("Variance")[0];
DLOG << " Input Scale is : " << op->Input("Scale")[0];
DLOG << " Input Bias is : " << op->Input("Bias")[0];
DLOG << " Output Y is : " << op->Output("Y")[0];
DLOG << " epsilon : "
<< op->GetAttrMap().at("epsilon").Get<float>();
std::shared_ptr<operators::BatchNormOp<Dtype, float>> lrn =
std::make_shared<operators::BatchNormOp<Dtype, float>>(
op->Type(), op->GetInputs(), op->GetOutputs(),
op->GetAttrMap(), program_.scope);
ops_of_block_[*block_desc.get()].push_back(lrn);
}
}
}
} }
std::shared_ptr<Tensor> predict_bn(Tensor &t1, Tensor &t2, Tensor &t3, const std::vector<std::shared_ptr<BlockDesc>> blocks =
Tensor &t4, Tensor &t5) { to_predict_program_->Blocks();
// feed // DLOG << " **block size " << blocks.size();
auto scope = program_.scope; for (int i = 0; i < blocks.size(); ++i) {
Variable *x1_feed_value = scope->Var("conv2d_0.tmp_0"); std::shared_ptr<BlockDesc> block_desc = blocks[i];
auto tensor_x1 = x1_feed_value->GetMutable<Tensor>(); std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops();
tensor_x1->ShareDataWith(t1); // DLOG << " ops " << ops.size();
for (int j = 0; j < ops.size(); ++j) {
Variable *mean_feed_value = scope->Var("batch_norm_0.w_1"); std::shared_ptr<OpDesc> op = ops[j];
auto tensor_mean = mean_feed_value->GetMutable<Tensor>(); if (op->Type() == "batch_norm" &&
tensor_mean->ShareDataWith(t2); op->Input("X")[0] == "conv2d_0.tmp_0") {
DLOG << " mul attr size: " << op->GetAttrMap().size();
Variable *scale_feed_value = scope->Var("batch_norm_0.w_0"); DLOG << " inputs size: " << op->GetInputs().size();
auto tensor_scale = scale_feed_value->GetMutable<Tensor>(); DLOG << " outputs size: " << op->GetOutputs().size();
tensor_scale->ShareDataWith(t3); DLOG << " Input X is : " << op->Input("X")[0];
DLOG << " Input Mean is : " << op->Input("Mean")[0];
Variable *variance_feed_value = scope->Var("batch_norm_0.w_2"); DLOG << " Input Variance is : " << op->Input("Variance")[0];
auto tensor_variance = variance_feed_value->GetMutable<Tensor>(); DLOG << " Input Scale is : " << op->Input("Scale")[0];
tensor_variance->ShareDataWith(t4); DLOG << " Input Bias is : " << op->Input("Bias")[0];
DLOG << " Output Y is : " << op->Output("Y")[0];
Variable *bias_feed_value = scope->Var("batch_norm_0.b_0"); DLOG << " epsilon : " << op->GetAttrMap().at("epsilon").Get<float>();
auto tensor_bias = bias_feed_value->GetMutable<Tensor>(); std::shared_ptr<operators::BatchNormOp<Dtype, float>> lrn =
tensor_bias->ShareDataWith(t5); std::make_shared<operators::BatchNormOp<Dtype, float>>(
op->Type(), op->GetInputs(), op->GetOutputs(),
Variable *output = scope->Var("batch_norm_0.tmp_2"); op->GetAttrMap(), program_.scope);
auto *output_tensor = output->GetMutable<Tensor>(); ops_of_block_[*block_desc.get()].push_back(lrn);
output_tensor->mutable_data<float>({4, 10, 2, 2});
// DLOG << typeid(output_tensor).name();
// DLOG << "output_tensor dims: " << output_tensor->dims();
std::shared_ptr<Tensor> out_tensor = std::make_shared<LoDTensor>();
out_tensor.reset(output_tensor);
predict_bn(t1, t2, t3, t4, t5, 0);
return out_tensor;
}
private:
const framework::Program<Dtype> program_;
std::shared_ptr<ProgramDesc> to_predict_program_;
std::map<framework::BlockDesc,
std::vector<std::shared_ptr<OperatorBase<Dtype>>>>
ops_of_block_;
bool use_optimize_ = false;
void predict_bn(const Tensor &t1, const Tensor &t2, const Tensor &t3,
const Tensor &t4, const Tensor &t5, int block_id) {
std::shared_ptr<BlockDesc> to_predict_block =
to_predict_program_->Block(block_id);
for (int j = 0; j < ops_of_block_[*to_predict_block.get()].size();
++j) {
auto op = ops_of_block_[*to_predict_block.get()][j];
DLOG << "op -> run()";
op->Run();
} }
}
}
}
std::shared_ptr<Tensor> predict_bn(Tensor &t1, Tensor &t2, Tensor &t3,
Tensor &t4, Tensor &t5) {
// feed
auto scope = program_.scope;
Variable *x1_feed_value = scope->Var("conv2d_0.tmp_0");
auto tensor_x1 = x1_feed_value->GetMutable<Tensor>();
tensor_x1->ShareDataWith(t1);
Variable *mean_feed_value = scope->Var("batch_norm_0.w_1");
auto tensor_mean = mean_feed_value->GetMutable<Tensor>();
tensor_mean->ShareDataWith(t2);
Variable *scale_feed_value = scope->Var("batch_norm_0.w_0");
auto tensor_scale = scale_feed_value->GetMutable<Tensor>();
tensor_scale->ShareDataWith(t3);
Variable *variance_feed_value = scope->Var("batch_norm_0.w_2");
auto tensor_variance = variance_feed_value->GetMutable<Tensor>();
tensor_variance->ShareDataWith(t4);
Variable *bias_feed_value = scope->Var("batch_norm_0.b_0");
auto tensor_bias = bias_feed_value->GetMutable<Tensor>();
tensor_bias->ShareDataWith(t5);
Variable *output = scope->Var("batch_norm_0.tmp_2");
auto *output_tensor = output->GetMutable<Tensor>();
output_tensor->mutable_data<float>({4, 10, 2, 2});
// DLOG << typeid(output_tensor).name();
// DLOG << "output_tensor dims: " << output_tensor->dims();
std::shared_ptr<Tensor> out_tensor = std::make_shared<LoDTensor>();
out_tensor.reset(output_tensor);
predict_bn(t1, t2, t3, t4, t5, 0);
return out_tensor;
}
private:
const framework::Program<Dtype> program_;
std::shared_ptr<ProgramDesc> to_predict_program_;
std::map<framework::BlockDesc,
std::vector<std::shared_ptr<OperatorBase<Dtype>>>>
ops_of_block_;
bool use_optimize_ = false;
void predict_bn(const Tensor &t1, const Tensor &t2, const Tensor &t3,
const Tensor &t4, const Tensor &t5, int block_id) {
std::shared_ptr<BlockDesc> to_predict_block =
to_predict_program_->Block(block_id);
for (int j = 0; j < ops_of_block_[*to_predict_block.get()].size(); ++j) {
auto op = ops_of_block_[*to_predict_block.get()][j];
DLOG << "op -> run()";
op->Run();
} }
}
}; };
template class TestBatchNormOp<CPU>; template class TestBatchNormOp<CPU>;
...@@ -127,50 +125,48 @@ template class TestBatchNormOp<CPU>; ...@@ -127,50 +125,48 @@ template class TestBatchNormOp<CPU>;
} // namespace paddle_mobile } // namespace paddle_mobile
int main() { int main() {
DLOG << "----------**********----------"; DLOG << "----------**********----------";
DLOG << "begin to run BatchNormOp Test"; DLOG << "begin to run BatchNormOp Test";
paddle_mobile::Loader<paddle_mobile::CPU> loader; paddle_mobile::Loader<paddle_mobile::CPU> loader;
auto program = loader.Load(std::string( auto program = loader.Load(std::string(
"../../test/models/image_classification_resnet.inference.model")); "../../test/models/image_classification_resnet.inference.model"));
/// input x (4,10,2,2) /// input x (4,10,2,2)
paddle_mobile::framework::Tensor inputx1; paddle_mobile::framework::Tensor inputx1;
SetupTensor<float>(&inputx1, {4, 10, 2, 2}, static_cast<float>(0), SetupTensor<float>(&inputx1, {4, 10, 2, 2}, static_cast<float>(0),
static_cast<float>(1)); static_cast<float>(1));
auto *inputx1_ptr = inputx1.data<float>(); auto *inputx1_ptr = inputx1.data<float>();
paddle_mobile::framework::Tensor mean; paddle_mobile::framework::Tensor mean;
SetupTensor<float>(&mean, {10}, static_cast<float>(0), SetupTensor<float>(&mean, {10}, static_cast<float>(0), static_cast<float>(1));
static_cast<float>(1)); auto *mean_ptr = mean.data<float>();
auto *mean_ptr = mean.data<float>();
paddle_mobile::framework::Tensor scale;
paddle_mobile::framework::Tensor scale; SetupTensor<float>(&scale, {10}, static_cast<float>(0),
SetupTensor<float>(&scale, {10}, static_cast<float>(0), static_cast<float>(1));
static_cast<float>(1)); auto *scale_ptr = scale.data<float>();
auto *scale_ptr = scale.data<float>();
paddle_mobile::framework::Tensor variance;
paddle_mobile::framework::Tensor variance; SetupTensor<float>(&variance, {10}, static_cast<float>(0),
SetupTensor<float>(&variance, {10}, static_cast<float>(0), static_cast<float>(1));
static_cast<float>(1)); auto *variance_ptr = variance.data<float>();
auto *variance_ptr = variance.data<float>();
paddle_mobile::framework::Tensor bias;
paddle_mobile::framework::Tensor bias; SetupTensor<float>(&bias, {10}, static_cast<float>(0), static_cast<float>(1));
SetupTensor<float>(&bias, {10}, static_cast<float>(0), auto *bias_ptr = bias.data<float>();
static_cast<float>(1));
auto *bias_ptr = bias.data<float>(); paddle_mobile::framework::TestBatchNormOp<paddle_mobile::CPU> testBatchNormOp(
program);
paddle_mobile::framework::TestBatchNormOp<paddle_mobile::CPU>
testBatchNormOp(program); auto output_bn =
testBatchNormOp.predict_bn(inputx1, mean, scale, variance, bias);
auto output_bn = auto *output_bn_ptr = output_bn->data<float>();
testBatchNormOp.predict_bn(inputx1, mean, scale, variance, bias);
auto *output_bn_ptr = output_bn->data<float>(); /// [2, 5, 1, 0]
DLOG << " (" << inputx1_ptr[102] << " - " << mean_ptr[5] << ")/(("
/// [2, 5, 1, 0] << variance_ptr[5] << " + 0.00001"
DLOG << " (" << inputx1_ptr[102] << " - " << mean_ptr[5] << ")/((" << ")^0.5)* " << scale_ptr[5] << " + " << bias_ptr[5] << " = ";
<< variance_ptr[5] << " + 0.00001" DLOG << output_bn_ptr[102];
<< ")^0.5)* " << scale_ptr[5] << " + " << bias_ptr[5] << " = ";
DLOG << output_bn_ptr[102]; return 0;
return 0;
} }
test/operators/test_concat_op.cpp
浏览文件 @ dff4a781
...@@ -24,95 +24,92 @@ namespace paddle_mobile { ...@@ -24,95 +24,92 @@ namespace paddle_mobile {
namespace framework { namespace framework {
template <typename Dtype> class TestConcatOp { template <typename Dtype> class TestConcatOp {
public: public:
explicit TestConcatOp(const Program<Dtype> p) : program_(p) { explicit TestConcatOp(const Program<Dtype> p) : program_(p) {
if (use_optimize_) { if (use_optimize_) {
to_predict_program_ = program_.optimizeProgram; to_predict_program_ = program_.optimizeProgram;
} else { } else {
to_predict_program_ = program_.originProgram; to_predict_program_ = program_.originProgram;
}
const std::vector<std::shared_ptr<BlockDesc>> blocks =
to_predict_program_->Blocks();
// DLOG << " **block size " << blocks.size();
for (int i = 0; i < blocks.size(); ++i) {
std::shared_ptr<BlockDesc> block_desc = blocks[i];
std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops();
// DLOG << " ops " << ops.size();
for (int j = 0; j < ops.size(); ++j) {
std::shared_ptr<OpDesc> op = ops[j];
if (op->Type() == "concat" &&
op->Input("X")[0] == "conv2d_3.tmp_1") {
DLOG << " mul attr size: " << op->GetAttrMap().size();
DLOG << " inputs size: " << op->GetInputs().size();
DLOG << " outputs size: " << op->GetOutputs().size();
DLOG << " Input X is : " << op->Input("X")[0];
DLOG << " Output Out is : " << op->Output("Out")[0];
DLOG << " axis : "
<< op->GetAttrMap().at("axis").Get<int>();
std::shared_ptr<operators::ConcatOp<Dtype, float>> concat =
std::make_shared<operators::ConcatOp<Dtype, float>>(
op->Type(), op->GetInputs(), op->GetOutputs(),
op->GetAttrMap(), program_.scope);
ops_of_block_[*block_desc.get()].push_back(concat);
}
}
}
} }
std::shared_ptr<Tensor> predict_concat(Tensor &t1, Tensor &t2, Tensor &t3, const std::vector<std::shared_ptr<BlockDesc>> blocks =
Tensor &t4) { to_predict_program_->Blocks();
// feed // DLOG << " **block size " << blocks.size();
auto scope = program_.scope; for (int i = 0; i < blocks.size(); ++i) {
Variable *x1_feed_value = scope->Var("conv2d_3.tmp_1"); std::shared_ptr<BlockDesc> block_desc = blocks[i];
auto tensor_x1 = x1_feed_value->GetMutable<Tensor>(); std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops();
tensor_x1->ShareDataWith(t1); // DLOG << " ops " << ops.size();
for (int j = 0; j < ops.size(); ++j) {
Variable *x2_feed_value = scope->Var("conv2d_5.tmp_1"); std::shared_ptr<OpDesc> op = ops[j];
auto tensor_x2 = x2_feed_value->GetMutable<Tensor>(); if (op->Type() == "concat" && op->Input("X")[0] == "conv2d_3.tmp_1") {
tensor_x2->ShareDataWith(t2); DLOG << " mul attr size: " << op->GetAttrMap().size();
DLOG << " inputs size: " << op->GetInputs().size();
Variable *x3_feed_value = scope->Var("conv2d_7.tmp_1"); DLOG << " outputs size: " << op->GetOutputs().size();
auto tensor_x3 = x3_feed_value->GetMutable<Tensor>(); DLOG << " Input X is : " << op->Input("X")[0];
tensor_x3->ShareDataWith(t3); DLOG << " Output Out is : " << op->Output("Out")[0];
DLOG << " axis : " << op->GetAttrMap().at("axis").Get<int>();
Variable *x4_feed_value = scope->Var("conv2d_8.tmp_1");
auto tensor_x4 = x4_feed_value->GetMutable<Tensor>(); std::shared_ptr<operators::ConcatOp<Dtype, float>> concat =
tensor_x4->ShareDataWith(t4); std::make_shared<operators::ConcatOp<Dtype, float>>(
op->Type(), op->GetInputs(), op->GetOutputs(),
Variable *con_output = scope->Var("concat_0.tmp_0"); op->GetAttrMap(), program_.scope);
auto *output_tensor = con_output->GetMutable<Tensor>(); ops_of_block_[*block_desc.get()].push_back(concat);
output_tensor->mutable_data<float>({4, 100, 2, 2});
// DLOG << typeid(output_tensor).name();
// DLOG << "output_tensor dims: " << output_tensor->dims();
std::shared_ptr<Tensor> out_tensor = std::make_shared<LoDTensor>();
out_tensor.reset(output_tensor);
predict_concat(t1, t2, t3, t4, 0);
return out_tensor;
}
private:
const framework::Program<Dtype> program_;
std::shared_ptr<ProgramDesc> to_predict_program_;
std::map<framework::BlockDesc,
std::vector<std::shared_ptr<OperatorBase<Dtype>>>>
ops_of_block_;
bool use_optimize_ = false;
void predict_concat(const Tensor &t1, const Tensor &t2, const Tensor &t3,
const Tensor &t4, int block_id) {
std::shared_ptr<BlockDesc> to_predict_block =
to_predict_program_->Block(block_id);
for (int j = 0; j < ops_of_block_[*to_predict_block.get()].size();
++j) {
auto op = ops_of_block_[*to_predict_block.get()][j];
DLOG << "op -> run()";
op->Run();
} }
}
}
}
std::shared_ptr<Tensor> predict_concat(Tensor &t1, Tensor &t2, Tensor &t3,
Tensor &t4) {
// feed
auto scope = program_.scope;
Variable *x1_feed_value = scope->Var("conv2d_3.tmp_1");
auto tensor_x1 = x1_feed_value->GetMutable<Tensor>();
tensor_x1->ShareDataWith(t1);
Variable *x2_feed_value = scope->Var("conv2d_5.tmp_1");
auto tensor_x2 = x2_feed_value->GetMutable<Tensor>();
tensor_x2->ShareDataWith(t2);
Variable *x3_feed_value = scope->Var("conv2d_7.tmp_1");
auto tensor_x3 = x3_feed_value->GetMutable<Tensor>();
tensor_x3->ShareDataWith(t3);
Variable *x4_feed_value = scope->Var("conv2d_8.tmp_1");
auto tensor_x4 = x4_feed_value->GetMutable<Tensor>();
tensor_x4->ShareDataWith(t4);
Variable *con_output = scope->Var("concat_0.tmp_0");
auto *output_tensor = con_output->GetMutable<Tensor>();
output_tensor->mutable_data<float>({4, 100, 2, 2});
// DLOG << typeid(output_tensor).name();
// DLOG << "output_tensor dims: " << output_tensor->dims();
std::shared_ptr<Tensor> out_tensor = std::make_shared<LoDTensor>();
out_tensor.reset(output_tensor);
predict_concat(t1, t2, t3, t4, 0);
return out_tensor;
}
private:
const framework::Program<Dtype> program_;
std::shared_ptr<ProgramDesc> to_predict_program_;
std::map<framework::BlockDesc,
std::vector<std::shared_ptr<OperatorBase<Dtype>>>>
ops_of_block_;
bool use_optimize_ = false;
void predict_concat(const Tensor &t1, const Tensor &t2, const Tensor &t3,
const Tensor &t4, int block_id) {
std::shared_ptr<BlockDesc> to_predict_block =
to_predict_program_->Block(block_id);
for (int j = 0; j < ops_of_block_[*to_predict_block.get()].size(); ++j) {
auto op = ops_of_block_[*to_predict_block.get()][j];
DLOG << "op -> run()";
op->Run();
} }
}
}; };
template class TestConcatOp<CPU>; template class TestConcatOp<CPU>;
...@@ -120,60 +117,59 @@ template class TestConcatOp<CPU>; ...@@ -120,60 +117,59 @@ template class TestConcatOp<CPU>;
} // namespace paddle_mobile } // namespace paddle_mobile
int main() { int main() {
DLOG << "----------**********----------"; DLOG << "----------**********----------";
DLOG << "begin to run ConcatOp Test"; DLOG << "begin to run ConcatOp Test";
paddle_mobile::Loader<paddle_mobile::CPU> loader; paddle_mobile::Loader<paddle_mobile::CPU> loader;
auto program = loader.Load(std::string("../../test/models/googlenet")); auto program = loader.Load(std::string("../../test/models/googlenet"));
/// input x (4,10,2,2) /// input x (4,10,2,2)
paddle_mobile::framework::Tensor inputx1; paddle_mobile::framework::Tensor inputx1;
SetupTensor<float>(&inputx1, {4, 10, 2, 2}, static_cast<float>(0), SetupTensor<float>(&inputx1, {4, 10, 2, 2}, static_cast<float>(0),
static_cast<float>(1)); static_cast<float>(1));
auto *inputx1_ptr = inputx1.data<float>(); auto *inputx1_ptr = inputx1.data<float>();
/// input x (4,20,2,2) /// input x (4,20,2,2)
paddle_mobile::framework::Tensor inputx2; paddle_mobile::framework::Tensor inputx2;
SetupTensor<float>(&inputx2, {4, 20, 2, 2}, static_cast<float>(0), SetupTensor<float>(&inputx2, {4, 20, 2, 2}, static_cast<float>(0),
static_cast<float>(1)); static_cast<float>(1));
auto *inputx2_ptr = inputx2.data<float>(); auto *inputx2_ptr = inputx2.data<float>();
/// input x (4,30,2,2) /// input x (4,30,2,2)
paddle_mobile::framework::Tensor inputx3; paddle_mobile::framework::Tensor inputx3;
SetupTensor<float>(&inputx3, {4, 30, 2, 2}, static_cast<float>(0), SetupTensor<float>(&inputx3, {4, 30, 2, 2}, static_cast<float>(0),
static_cast<float>(1)); static_cast<float>(1));
auto *inputx3_ptr = inputx3.data<float>(); auto *inputx3_ptr = inputx3.data<float>();
/// input x (4,40,2,2) /// input x (4,40,2,2)
paddle_mobile::framework::Tensor inputx4; paddle_mobile::framework::Tensor inputx4;
SetupTensor<float>(&inputx4, {4, 40, 2, 2}, static_cast<float>(0), SetupTensor<float>(&inputx4, {4, 40, 2, 2}, static_cast<float>(0),
static_cast<float>(1)); static_cast<float>(1));
auto *inputx4_ptr = inputx4.data<float>(); auto *inputx4_ptr = inputx4.data<float>();
paddle_mobile::framework::TestConcatOp<paddle_mobile::CPU> testConcatOp( paddle_mobile::framework::TestConcatOp<paddle_mobile::CPU> testConcatOp(
program); program);
auto output_concat = auto output_concat =
testConcatOp.predict_concat(inputx1, inputx2, inputx3, inputx4); testConcatOp.predict_concat(inputx1, inputx2, inputx3, inputx4);
auto *output_concat_ptr = output_concat->data<float>(); auto *output_concat_ptr = output_concat->data<float>();
int input_n = 1; int input_n = 1;
int input_c = 2; int input_c = 2;
int input_h = 0; int input_h = 0;
int input_w = 1; int input_w = 1;
int stride0 = inputx3.numel() / inputx3.dims()[0]; int stride0 = inputx3.numel() / inputx3.dims()[0];
int stride1 = inputx3.numel() / inputx3.dims()[0] / inputx3.dims()[1]; int stride1 = inputx3.numel() / inputx3.dims()[0] / inputx3.dims()[1];
int stride2 = inputx3.dims()[3]; int stride2 = inputx3.dims()[3];
/// inputx1 (4,10,2,2), /// inputx1 (4,10,2,2),
/// inputx2 (4,20,2,2), /// inputx2 (4,20,2,2),
/// inputx3 (4,30,2,2), /// inputx3 (4,30,2,2),
/// inputx4 (4,40,2,2), /// inputx4 (4,40,2,2),
/// axis = 1 /// axis = 1
/// output (4,100,2,2) /// output (4,100,2,2)
int input_index = int input_index =
input_n * stride0 + input_c * stride1 + input_h * stride2 + input_w; input_n * stride0 + input_c * stride1 + input_h * stride2 + input_w;
int output_index = int output_index = input_n * 100 * 2 * 2 +
input_n * 100 * 2 * 2 + (input_c + inputx1.dims()[1] + inputx2.dims()[1]) * 2 * 2 +
(input_c + inputx1.dims()[1] + inputx2.dims()[1]) * 2 * 2 + input_h * 2 + input_w;
input_h * 2 + input_w;
DLOG << " inputx3[1,2,0,1] = " << inputx3_ptr[input_index];
DLOG << " inputx3[1,2,0,1] = " << inputx3_ptr[input_index]; DLOG << " output[1,12,0,1] = " << output_concat_ptr[output_index];
DLOG << " output[1,12,0,1] = " << output_concat_ptr[output_index]; return 0;
return 0;
} }
test/operators/test_cov_op.cpp
浏览文件 @ dff4a781
...@@ -21,26 +21,26 @@ SOFTWARE. ...@@ -21,26 +21,26 @@ SOFTWARE.
#include "io.h" #include "io.h"
int main() { int main() {
paddle_mobile::Loader<paddle_mobile::CPU> loader; paddle_mobile::Loader<paddle_mobile::CPU> loader;
auto program = loader.Load(std::string("../models/googlenet")); auto program = loader.Load(std::string("../models/googlenet"));
if (program.originProgram == nullptr) { if (program.originProgram == nullptr) {
DLOG << "program file read fail"; DLOG << "program file read fail";
} }
Executor4Test<paddle_mobile::CPU, Executor4Test<paddle_mobile::CPU,
paddle_mobile::operators::ConvOp<paddle_mobile::CPU, float>> paddle_mobile::operators::ConvOp<paddle_mobile::CPU, float>>
executor(program, "conv2d"); executor(program, "conv2d");
paddle_mobile::framework::Tensor input; paddle_mobile::framework::Tensor input;
SetupTensor<float>(&input, {1, 3, 32, 32}, static_cast<float>(0), SetupTensor<float>(&input, {1, 3, 32, 32}, static_cast<float>(0),
static_cast<float>(1)); static_cast<float>(1));
auto output = auto output =
executor.predict(input, "data", "conv2d_0.tmp_0", {1, 64, 56, 56}); executor.predict(input, "data", "conv2d_0.tmp_0", {1, 64, 56, 56});
float *output_ptr = output->data<float>(); float *output_ptr = output->data<float>();
for (int j = 0; j < output->numel(); ++j) { for (int j = 0; j < output->numel(); ++j) {
DLOG << " value of output: " << output_ptr[j]; DLOG << " value of output: " << output_ptr[j];
} }
return 0; return 0;
} }
test/operators/test_elementwise_add_op.cpp
浏览文件 @ dff4a781
...@@ -24,127 +24,124 @@ namespace paddle_mobile { ...@@ -24,127 +24,124 @@ namespace paddle_mobile {
namespace framework { namespace framework {
template <typename Dtype> class TestElementwiseAddOp { template <typename Dtype> class TestElementwiseAddOp {
public: public:
explicit TestElementwiseAddOp(const Program<Dtype> p) : program_(p) { explicit TestElementwiseAddOp(const Program<Dtype> p) : program_(p) {
if (use_optimize_) { if (use_optimize_) {
to_predict_program_ = program_.optimizeProgram; to_predict_program_ = program_.optimizeProgram;
} else { } else {
to_predict_program_ = program_.originProgram; to_predict_program_ = program_.originProgram;
}
const std::vector<std::shared_ptr<BlockDesc>> blocks =
to_predict_program_->Blocks();
// DLOG << " **block size " << blocks.size();
for (int i = 0; i < blocks.size(); ++i) {
std::shared_ptr<BlockDesc> block_desc = blocks[i];
std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops();
// DLOG << " ops " << ops.size();
for (int j = 0; j < ops.size(); ++j) {
std::shared_ptr<OpDesc> op = ops[j];
if (op->Type() == "elementwise_add" &&
op->Input("X")[0] == "batch_norm_2.tmp_2") {
DLOG << " elementwise_add attr size: "
<< op->GetAttrMap().size();
DLOG << " inputs size: " << op->GetInputs().size();
DLOG << " outputs size: " << op->GetOutputs().size();
DLOG << " Input X is : " << op->Input("X")[0];
DLOG << " Input Y is : " << op->Input("Y")[0];
DLOG << " Output Out is : " << op->Output("Out")[0];
Attribute axis_attr = op->GetAttrMap().at("axis");
int axis = axis_attr.Get<int>();
DLOG << " Attr axis is : " << axis;
std::shared_ptr<operators::ElementwiseAddOp<Dtype, float>>
add = std::make_shared<
operators::ElementwiseAddOp<Dtype, float>>(
op->Type(), op->GetInputs(), op->GetOutputs(),
op->GetAttrMap(), program_.scope);
ops_of_block_[*block_desc.get()].push_back(add);
}
}
}
}
std::shared_ptr<Tensor> predict_add(Tensor &t1, Tensor &t2) {
// feed
auto scope = program_.scope;
Variable *x_feed_value = scope->Var("batch_norm_2.tmp_2");
auto tensor_x = x_feed_value->GetMutable<Tensor>();
tensor_x->ShareDataWith(t1);
Variable *y_feed_value = scope->Var("batch_norm_0.tmp_3");
auto tensor_y = y_feed_value->GetMutable<Tensor>();
tensor_y->ShareDataWith(t2);
Variable *con_output = scope->Var("elementwise_add_0.tmp_0");
auto *output_tensor = con_output->GetMutable<Tensor>();
output_tensor->mutable_data<float>({1, 3, 224, 224});
// DLOG << typeid(output_tensor).name();
// DLOG << "output_tensor dims: " << output_tensor->dims();
std::shared_ptr<Tensor> out_tensor = std::make_shared<LoDTensor>();
out_tensor.reset(output_tensor);
predict_add(t1, t2, 0);
return out_tensor;
} }
private: const std::vector<std::shared_ptr<BlockDesc>> blocks =
const framework::Program<Dtype> program_; to_predict_program_->Blocks();
std::shared_ptr<ProgramDesc> to_predict_program_; // DLOG << " **block size " << blocks.size();
std::map<framework::BlockDesc, for (int i = 0; i < blocks.size(); ++i) {
std::vector<std::shared_ptr<OperatorBase<Dtype>>>> std::shared_ptr<BlockDesc> block_desc = blocks[i];
ops_of_block_; std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops();
bool use_optimize_ = false; // DLOG << " ops " << ops.size();
for (int j = 0; j < ops.size(); ++j) {
void predict_add(const Tensor &t1, const Tensor &t2, int block_id) { std::shared_ptr<OpDesc> op = ops[j];
std::shared_ptr<BlockDesc> to_predict_block = if (op->Type() == "elementwise_add" &&
to_predict_program_->Block(block_id); op->Input("X")[0] == "batch_norm_2.tmp_2") {
for (int j = 0; j < ops_of_block_[*to_predict_block.get()].size(); DLOG << " elementwise_add attr size: " << op->GetAttrMap().size();
++j) { DLOG << " inputs size: " << op->GetInputs().size();
auto op = ops_of_block_[*to_predict_block.get()][j]; DLOG << " outputs size: " << op->GetOutputs().size();
DLOG << "op -> run()"; DLOG << " Input X is : " << op->Input("X")[0];
op->Run(); DLOG << " Input Y is : " << op->Input("Y")[0];
DLOG << " Output Out is : " << op->Output("Out")[0];
Attribute axis_attr = op->GetAttrMap().at("axis");
int axis = axis_attr.Get<int>();
DLOG << " Attr axis is : " << axis;
std::shared_ptr<operators::ElementwiseAddOp<Dtype, float>> add =
std::make_shared<operators::ElementwiseAddOp<Dtype, float>>(
op->Type(), op->GetInputs(), op->GetOutputs(),
op->GetAttrMap(), program_.scope);
ops_of_block_[*block_desc.get()].push_back(add);
} }
}
}
}
std::shared_ptr<Tensor> predict_add(Tensor &t1, Tensor &t2) {
// feed
auto scope = program_.scope;
Variable *x_feed_value = scope->Var("batch_norm_2.tmp_2");
auto tensor_x = x_feed_value->GetMutable<Tensor>();
tensor_x->ShareDataWith(t1);
Variable *y_feed_value = scope->Var("batch_norm_0.tmp_3");
auto tensor_y = y_feed_value->GetMutable<Tensor>();
tensor_y->ShareDataWith(t2);
Variable *con_output = scope->Var("elementwise_add_0.tmp_0");
auto *output_tensor = con_output->GetMutable<Tensor>();
output_tensor->mutable_data<float>({1, 3, 224, 224});
// DLOG << typeid(output_tensor).name();
// DLOG << "output_tensor dims: " << output_tensor->dims();
std::shared_ptr<Tensor> out_tensor = std::make_shared<LoDTensor>();
out_tensor.reset(output_tensor);
predict_add(t1, t2, 0);
return out_tensor;
}
private:
const framework::Program<Dtype> program_;
std::shared_ptr<ProgramDesc> to_predict_program_;
std::map<framework::BlockDesc,
std::vector<std::shared_ptr<OperatorBase<Dtype>>>>
ops_of_block_;
bool use_optimize_ = false;
void predict_add(const Tensor &t1, const Tensor &t2, int block_id) {
std::shared_ptr<BlockDesc> to_predict_block =
to_predict_program_->Block(block_id);
for (int j = 0; j < ops_of_block_[*to_predict_block.get()].size(); ++j) {
auto op = ops_of_block_[*to_predict_block.get()][j];
DLOG << "op -> run()";
op->Run();
} }
}
}; };
template class TestElementwiseAddOp<CPU>; template class TestElementwiseAddOp<CPU>;
} // namespace framework } // namespace framework
} // namespace paddle_mobile } // namespace paddle_mobile
int main() { int main() {
DLOG << "----------**********----------"; DLOG << "----------**********----------";
DLOG << "begin to run ElementAddOp Test"; DLOG << "begin to run ElementAddOp Test";
paddle_mobile::Loader<paddle_mobile::CPU> loader; paddle_mobile::Loader<paddle_mobile::CPU> loader;
auto program = auto program =
loader.Load(std::string("../../test/models/" loader.Load(std::string("../../test/models/"
"image_classification_resnet.inference.model")); "image_classification_resnet.inference.model"));
/// input x (1,3,224,224) /// input x (1,3,224,224)
paddle_mobile::framework::Tensor inputx; paddle_mobile::framework::Tensor inputx;
SetupTensor<float>(&inputx, {1, 3, 224, 224}, static_cast<float>(0), SetupTensor<float>(&inputx, {1, 3, 224, 224}, static_cast<float>(0),
static_cast<float>(1)); static_cast<float>(1));
auto *inputx_ptr = inputx.data<float>(); auto *inputx_ptr = inputx.data<float>();
/// input y (224,) /// input y (224,)
paddle_mobile::framework::Tensor inputy; paddle_mobile::framework::Tensor inputy;
SetupTensor<float>(&inputy, {224}, static_cast<float>(0), SetupTensor<float>(&inputy, {224}, static_cast<float>(0),
static_cast<float>(1)); static_cast<float>(1));
auto *inputy_ptr = inputy.data<float>(); auto *inputy_ptr = inputy.data<float>();
paddle_mobile::framework::TestElementwiseAddOp<paddle_mobile::CPU> paddle_mobile::framework::TestElementwiseAddOp<paddle_mobile::CPU>
testElementwiseAddOp(program); testElementwiseAddOp(program);
auto output_add = testElementwiseAddOp.predict_add(inputx, inputy); auto output_add = testElementwiseAddOp.predict_add(inputx, inputy);
auto *output_add_ptr = output_add->data<float>(); auto *output_add_ptr = output_add->data<float>();
// for (int j = 0; j < output_add->numel(); ++j) { // for (int j = 0; j < output_add->numel(); ++j) {
// DLOG << "value of output: " << output_add_ptr[j]; // DLOG << "value of output: " << output_add_ptr[j];
// } // }
/// output (1,3,224,224) /// output (1,3,224,224)
DLOG << "output memory size : " << output_add->memory_size(); DLOG << "output memory size : " << output_add->memory_size();
DLOG << "output numel : " << output_add->numel(); DLOG << "output numel : " << output_add->numel();
DLOG << inputx_ptr[226] << " + " << inputy_ptr[2] << " = " DLOG << inputx_ptr[226] << " + " << inputy_ptr[2] << " = "
<< output_add_ptr[226]; << output_add_ptr[226];
return 0; return 0;
} }
test/operators/test_lrn_op.cpp
浏览文件 @ dff4a781
...@@ -24,84 +24,80 @@ namespace paddle_mobile { ...@@ -24,84 +24,80 @@ namespace paddle_mobile {
namespace framework { namespace framework {
template <typename Dtype> class TestLrnOp { template <typename Dtype> class TestLrnOp {
public: public:
explicit TestLrnOp(const Program<Dtype> p) : program_(p) { explicit TestLrnOp(const Program<Dtype> p) : program_(p) {
if (use_optimize_) { if (use_optimize_) {
to_predict_program_ = program_.optimizeProgram; to_predict_program_ = program_.optimizeProgram;
} else { } else {
to_predict_program_ = program_.originProgram; to_predict_program_ = program_.originProgram;
}
const std::vector<std::shared_ptr<BlockDesc>> blocks =
to_predict_program_->Blocks();
// DLOG << " **block size " << blocks.size();
for (int i = 0; i < blocks.size(); ++i) {
std::shared_ptr<BlockDesc> block_desc = blocks[i];
std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops();
// DLOG << " ops " << ops.size();
for (int j = 0; j < ops.size(); ++j) {
std::shared_ptr<OpDesc> op = ops[j];
if (op->Type() == "lrn" &&
op->Input("X")[0] == "pool2d_0.tmp_0") {
DLOG << " mul attr size: " << op->GetAttrMap().size();
DLOG << " inputs size: " << op->GetInputs().size();
DLOG << " outputs size: " << op->GetOutputs().size();
DLOG << " Input X is : " << op->Input("X")[0];
DLOG << " Output Out is : " << op->Output("Out")[0];
DLOG << " n : " << op->GetAttrMap().at("n").Get<int>();
DLOG << " alpha : "
<< op->GetAttrMap().at("alpha").Get<float>();
DLOG << " beta : "
<< op->GetAttrMap().at("beta").Get<float>();
DLOG << " k : " << op->GetAttrMap().at("k").Get<float>();
std::shared_ptr<operators::LrnOp<Dtype, float>> lrn =
std::make_shared<operators::LrnOp<Dtype, float>>(
op->Type(), op->GetInputs(), op->GetOutputs(),
op->GetAttrMap(), program_.scope);
ops_of_block_[*block_desc.get()].push_back(lrn);
}
}
}
} }
std::shared_ptr<Tensor> predict_lrn(Tensor &t1) { const std::vector<std::shared_ptr<BlockDesc>> blocks =
// feed to_predict_program_->Blocks();
auto scope = program_.scope; // DLOG << " **block size " << blocks.size();
Variable *x1_feed_value = scope->Var("pool2d_0.tmp_0"); for (int i = 0; i < blocks.size(); ++i) {
auto tensor_x1 = x1_feed_value->GetMutable<Tensor>(); std::shared_ptr<BlockDesc> block_desc = blocks[i];
tensor_x1->ShareDataWith(t1); std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops();
// DLOG << " ops " << ops.size();
Variable *con_output = scope->Var("pool1_norm1.tmp_1"); for (int j = 0; j < ops.size(); ++j) {
auto *output_tensor = con_output->GetMutable<Tensor>(); std::shared_ptr<OpDesc> op = ops[j];
output_tensor->mutable_data<float>({3, 4, 2, 2}); if (op->Type() == "lrn" && op->Input("X")[0] == "pool2d_0.tmp_0") {
// DLOG << typeid(output_tensor).name(); DLOG << " mul attr size: " << op->GetAttrMap().size();
// DLOG << "output_tensor dims: " << output_tensor->dims(); DLOG << " inputs size: " << op->GetInputs().size();
DLOG << " outputs size: " << op->GetOutputs().size();
std::shared_ptr<Tensor> out_tensor = std::make_shared<LoDTensor>(); DLOG << " Input X is : " << op->Input("X")[0];
out_tensor.reset(output_tensor); DLOG << " Output Out is : " << op->Output("Out")[0];
DLOG << " n : " << op->GetAttrMap().at("n").Get<int>();
predict_lrn(t1, 0); DLOG << " alpha : " << op->GetAttrMap().at("alpha").Get<float>();
return out_tensor; DLOG << " beta : " << op->GetAttrMap().at("beta").Get<float>();
} DLOG << " k : " << op->GetAttrMap().at("k").Get<float>();
std::shared_ptr<operators::LrnOp<Dtype, float>> lrn =
private: std::make_shared<operators::LrnOp<Dtype, float>>(
const framework::Program<Dtype> program_; op->Type(), op->GetInputs(), op->GetOutputs(),
std::shared_ptr<ProgramDesc> to_predict_program_; op->GetAttrMap(), program_.scope);
std::map<framework::BlockDesc, ops_of_block_[*block_desc.get()].push_back(lrn);
std::vector<std::shared_ptr<OperatorBase<Dtype>>>>
ops_of_block_;
bool use_optimize_ = false;
void predict_lrn(const Tensor &t1, int block_id) {
std::shared_ptr<BlockDesc> to_predict_block =
to_predict_program_->Block(block_id);
for (int j = 0; j < ops_of_block_[*to_predict_block.get()].size();
++j) {
auto op = ops_of_block_[*to_predict_block.get()][j];
DLOG << "op -> run()";
op->Run();
} }
}
}
}
std::shared_ptr<Tensor> predict_lrn(Tensor &t1) {
// feed
auto scope = program_.scope;
Variable *x1_feed_value = scope->Var("pool2d_0.tmp_0");
auto tensor_x1 = x1_feed_value->GetMutable<Tensor>();
tensor_x1->ShareDataWith(t1);
Variable *con_output = scope->Var("pool1_norm1.tmp_1");
auto *output_tensor = con_output->GetMutable<Tensor>();
output_tensor->mutable_data<float>({3, 4, 2, 2});
// DLOG << typeid(output_tensor).name();
// DLOG << "output_tensor dims: " << output_tensor->dims();
std::shared_ptr<Tensor> out_tensor = std::make_shared<LoDTensor>();
out_tensor.reset(output_tensor);
predict_lrn(t1, 0);
return out_tensor;
}
private:
const framework::Program<Dtype> program_;
std::shared_ptr<ProgramDesc> to_predict_program_;
std::map<framework::BlockDesc,
std::vector<std::shared_ptr<OperatorBase<Dtype>>>>
ops_of_block_;
bool use_optimize_ = false;
void predict_lrn(const Tensor &t1, int block_id) {
std::shared_ptr<BlockDesc> to_predict_block =
to_predict_program_->Block(block_id);
for (int j = 0; j < ops_of_block_[*to_predict_block.get()].size(); ++j) {
auto op = ops_of_block_[*to_predict_block.get()][j];
DLOG << "op -> run()";
op->Run();
} }
}
}; };
template class TestLrnOp<CPU>; template class TestLrnOp<CPU>;
...@@ -109,51 +105,50 @@ template class TestLrnOp<CPU>; ...@@ -109,51 +105,50 @@ template class TestLrnOp<CPU>;
} // namespace paddle_mobile } // namespace paddle_mobile
int main() { int main() {
DLOG << "----------**********----------"; DLOG << "----------**********----------";
DLOG << "begin to run LrnOp Test"; DLOG << "begin to run LrnOp Test";
paddle_mobile::Loader<paddle_mobile::CPU> loader; paddle_mobile::Loader<paddle_mobile::CPU> loader;
auto program = loader.Load(std::string("../../test/models/googlenet")); auto program = loader.Load(std::string("../../test/models/googlenet"));
/// input x (3,4,2,2) /// input x (3,4,2,2)
paddle_mobile::framework::Tensor inputx1; paddle_mobile::framework::Tensor inputx1;
SetupTensor<float>(&inputx1, {3, 4, 2, 2}, static_cast<float>(0), SetupTensor<float>(&inputx1, {3, 4, 2, 2}, static_cast<float>(0),
static_cast<float>(1)); static_cast<float>(1));
auto *inputx1_ptr = inputx1.data<float>(); auto *inputx1_ptr = inputx1.data<float>();
paddle_mobile::framework::TestLrnOp<paddle_mobile::CPU> testLrnOp(program); paddle_mobile::framework::TestLrnOp<paddle_mobile::CPU> testLrnOp(program);
auto output_lrn = testLrnOp.predict_lrn(inputx1); auto output_lrn = testLrnOp.predict_lrn(inputx1);
auto *output_lrn_ptr = output_lrn->data<float>(); auto *output_lrn_ptr = output_lrn->data<float>();
DLOG << " LrnOp input: "; DLOG << " LrnOp input: ";
for (int i = 0; i < 3; i++) { for (int i = 0; i < 3; i++) {
for (int j = 0; j < 4; j++) { for (int j = 0; j < 4; j++) {
for (int c = 0; c < 2; c++) { for (int c = 0; c < 2; c++) {
for (int d = 0; d < 2; d++) { for (int d = 0; d < 2; d++) {
DLOGF("%f ", inputx1_ptr[i * 16 + j * 4 + c * 2 + d]); DLOGF("%f ", inputx1_ptr[i * 16 + j * 4 + c * 2 + d]);
}
DLOGF("\n");
}
DLOGF("\n");
} }
DLOGF("\n"); DLOGF("\n");
}
DLOGF("\n");
} }
DLOG << " LrnOp output: "; DLOGF("\n");
for (int i = 0; i < 3; i++) { }
for (int j = 0; j < 4; j++) { DLOG << " LrnOp output: ";
for (int c = 0; c < 2; c++) { for (int i = 0; i < 3; i++) {
for (int d = 0; d < 2; d++) { for (int j = 0; j < 4; j++) {
DLOGF("%f ", output_lrn_ptr[i * 16 + j * 4 + c * 2 + d]); for (int c = 0; c < 2; c++) {
} for (int d = 0; d < 2; d++) {
DLOGF("\n"); DLOGF("%f ", output_lrn_ptr[i * 16 + j * 4 + c * 2 + d]);
}
DLOGF("\n");
} }
DLOGF("\n"); DLOGF("\n");
}
DLOGF("\n");
} }
DLOG << inputx1_ptr[0] << " / ((1 + 0.00002 * ( " << inputx1_ptr[0] DLOGF("\n");
<< "^2 + " << inputx1_ptr[4] << "^2 + " << inputx1_ptr[8] }
<< "^2 ))^0.75) = "; DLOG << inputx1_ptr[0] << " / ((1 + 0.00002 * ( " << inputx1_ptr[0] << "^2 + "
DLOG << output_lrn_ptr[0]; << inputx1_ptr[4] << "^2 + " << inputx1_ptr[8] << "^2 ))^0.75) = ";
return 0; DLOG << output_lrn_ptr[0];
return 0;
} }
test/operators/test_mul_op.cpp
浏览文件 @ dff4a781
...@@ -24,88 +24,86 @@ namespace paddle_mobile { ...@@ -24,88 +24,86 @@ namespace paddle_mobile {
namespace framework { namespace framework {
template <typename Dtype> class TestMulOp { template <typename Dtype> class TestMulOp {
public: public:
explicit TestMulOp(const Program<Dtype> p) : program_(p) { explicit TestMulOp(const Program<Dtype> p) : program_(p) {
if (use_optimize_) { if (use_optimize_) {
to_predict_program_ = program_.optimizeProgram; to_predict_program_ = program_.optimizeProgram;
} else { } else {
to_predict_program_ = program_.originProgram; to_predict_program_ = program_.originProgram;
}
const std::vector<std::shared_ptr<BlockDesc>> blocks =
to_predict_program_->Blocks();
// DLOG << " **block size " << blocks.size();
for (int i = 0; i < blocks.size(); ++i) {
std::shared_ptr<BlockDesc> block_desc = blocks[i];
std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops();
// DLOG << " ops " << ops.size();
for (int j = 0; j < ops.size(); ++j) {
std::shared_ptr<OpDesc> op = ops[j];
if (op->Type() == "mul" &&
op->Input("X")[0] == "pool2d_0.tmp_0") {
DLOG << " mul attr size: " << op->GetAttrMap().size();
DLOG << " inputs size: " << op->GetInputs().size();
DLOG << " outputs size: " << op->GetOutputs().size();
DLOG << " Input X is : " << op->Input("X")[0];
DLOG << " Input Y is : " << op->Input("Y")[0];
DLOG << " Output Out is : " << op->Output("Out")[0];
DLOG << "x_num_col_dims : "
<< op->GetAttrMap().at("x_num_col_dims").Get<int>();
DLOG << "y_num_col_dims : "
<< op->GetAttrMap().at("y_num_col_dims").Get<int>();
std::shared_ptr<operators::MulOp<Dtype, float>> mul =
std::make_shared<operators::MulOp<Dtype, float>>(
op->Type(), op->GetInputs(), op->GetOutputs(),
op->GetAttrMap(), program_.scope);
ops_of_block_[*block_desc.get()].push_back(mul);
}
}
}
} }
std::shared_ptr<Tensor> predict_mul(Tensor &t1, Tensor &t2) { const std::vector<std::shared_ptr<BlockDesc>> blocks =
// feed to_predict_program_->Blocks();
auto scope = program_.scope; // DLOG << " **block size " << blocks.size();
Variable *x_feed_value = scope->Var("pool2d_0.tmp_0"); for (int i = 0; i < blocks.size(); ++i) {
auto tensor_x = x_feed_value->GetMutable<Tensor>(); std::shared_ptr<BlockDesc> block_desc = blocks[i];
tensor_x->ShareDataWith(t1); std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops();
// DLOG << " ops " << ops.size();
Variable *y_feed_value = scope->Var("fc_0.w_0"); for (int j = 0; j < ops.size(); ++j) {
auto tensor_y = y_feed_value->GetMutable<Tensor>(); std::shared_ptr<OpDesc> op = ops[j];
tensor_y->ShareDataWith(t2); if (op->Type() == "mul" && op->Input("X")[0] == "pool2d_0.tmp_0") {
DLOG << " mul attr size: " << op->GetAttrMap().size();
Variable *con_output = scope->Var("fc_0.tmp_0"); DLOG << " inputs size: " << op->GetInputs().size();
auto *output_tensor = con_output->GetMutable<Tensor>(); DLOG << " outputs size: " << op->GetOutputs().size();
output_tensor->mutable_data<float>({3, 3}); DLOG << " Input X is : " << op->Input("X")[0];
// DLOG << typeid(output_tensor).name(); DLOG << " Input Y is : " << op->Input("Y")[0];
// DLOG << "output_tensor dims: " << output_tensor->dims(); DLOG << " Output Out is : " << op->Output("Out")[0];
DLOG << "x_num_col_dims : "
std::shared_ptr<Tensor> out_tensor = std::make_shared<LoDTensor>(); << op->GetAttrMap().at("x_num_col_dims").Get<int>();
out_tensor.reset(output_tensor); DLOG << "y_num_col_dims : "
<< op->GetAttrMap().at("y_num_col_dims").Get<int>();
predict_mul(t1, t2, 0);
return out_tensor; std::shared_ptr<operators::MulOp<Dtype, float>> mul =
} std::make_shared<operators::MulOp<Dtype, float>>(
op->Type(), op->GetInputs(), op->GetOutputs(),
private: op->GetAttrMap(), program_.scope);
const framework::Program<Dtype> program_; ops_of_block_[*block_desc.get()].push_back(mul);
std::shared_ptr<ProgramDesc> to_predict_program_;
std::map<framework::BlockDesc,
std::vector<std::shared_ptr<OperatorBase<Dtype>>>>
ops_of_block_;
bool use_optimize_ = false;
void predict_mul(const Tensor &t1, const Tensor &t2, int block_id) {
std::shared_ptr<BlockDesc> to_predict_block =
to_predict_program_->Block(block_id);
for (int j = 0; j < ops_of_block_[*to_predict_block.get()].size();
++j) {
auto op = ops_of_block_[*to_predict_block.get()][j];
DLOG << "op -> run()";
op->Run();
} }
}
}
}
std::shared_ptr<Tensor> predict_mul(Tensor &t1, Tensor &t2) {
// feed
auto scope = program_.scope;
Variable *x_feed_value = scope->Var("pool2d_0.tmp_0");
auto tensor_x = x_feed_value->GetMutable<Tensor>();
tensor_x->ShareDataWith(t1);
Variable *y_feed_value = scope->Var("fc_0.w_0");
auto tensor_y = y_feed_value->GetMutable<Tensor>();
tensor_y->ShareDataWith(t2);
Variable *con_output = scope->Var("fc_0.tmp_0");
auto *output_tensor = con_output->GetMutable<Tensor>();
output_tensor->mutable_data<float>({3, 3});
// DLOG << typeid(output_tensor).name();
// DLOG << "output_tensor dims: " << output_tensor->dims();
std::shared_ptr<Tensor> out_tensor = std::make_shared<LoDTensor>();
out_tensor.reset(output_tensor);
predict_mul(t1, t2, 0);
return out_tensor;
}
private:
const framework::Program<Dtype> program_;
std::shared_ptr<ProgramDesc> to_predict_program_;
std::map<framework::BlockDesc,
std::vector<std::shared_ptr<OperatorBase<Dtype>>>>
ops_of_block_;
bool use_optimize_ = false;
void predict_mul(const Tensor &t1, const Tensor &t2, int block_id) {
std::shared_ptr<BlockDesc> to_predict_block =
to_predict_program_->Block(block_id);
for (int j = 0; j < ops_of_block_[*to_predict_block.get()].size(); ++j) {
auto op = ops_of_block_[*to_predict_block.get()][j];
DLOG << "op -> run()";
op->Run();
} }
}
}; };
template class TestMulOp<CPU>; template class TestMulOp<CPU>;
...@@ -113,62 +111,62 @@ template class TestMulOp<CPU>; ...@@ -113,62 +111,62 @@ template class TestMulOp<CPU>;
} // namespace paddle_mobile } // namespace paddle_mobile
int main() { int main() {
DLOG << "----------**********----------"; DLOG << "----------**********----------";
DLOG << "begin to run MulOp Test"; DLOG << "begin to run MulOp Test";
paddle_mobile::Loader<paddle_mobile::CPU> loader; paddle_mobile::Loader<paddle_mobile::CPU> loader;
auto program = auto program =
loader.Load(std::string("../../test/models/" loader.Load(std::string("../../test/models/"
"image_classification_resnet.inference.model")); "image_classification_resnet.inference.model"));
/// input x (3,2,1,1) /// input x (3,2,1,1)
paddle_mobile::framework::Tensor inputx; paddle_mobile::framework::Tensor inputx;
SetupTensor<float>(&inputx, {3, 2, 1, 1}, static_cast<float>(0), SetupTensor<float>(&inputx, {3, 2, 1, 1}, static_cast<float>(0),
static_cast<float>(1)); static_cast<float>(1));
auto *inputx_ptr = inputx.data<float>(); auto *inputx_ptr = inputx.data<float>();
/// input y (2,3) /// input y (2,3)
paddle_mobile::framework::Tensor inputy; paddle_mobile::framework::Tensor inputy;
SetupTensor<float>(&inputy, {2, 3}, static_cast<float>(0), SetupTensor<float>(&inputy, {2, 3}, static_cast<float>(0),
static_cast<float>(1)); static_cast<float>(1));
auto *inputy_ptr = inputy.data<float>(); auto *inputy_ptr = inputy.data<float>();
paddle_mobile::framework::TestMulOp<paddle_mobile::CPU> testMulOp(program); paddle_mobile::framework::TestMulOp<paddle_mobile::CPU> testMulOp(program);
auto output_mul = testMulOp.predict_mul(inputx, inputy); auto output_mul = testMulOp.predict_mul(inputx, inputy);
auto *output_mul_ptr = output_mul->data<float>(); auto *output_mul_ptr = output_mul->data<float>();
auto dimx_1 = inputx.numel() / inputx.dims()[0]; auto dimx_1 = inputx.numel() / inputx.dims()[0];
DLOG << " inputx : "; DLOG << " inputx : ";
for (int i = 0; i < inputx.dims()[0]; ++i) { for (int i = 0; i < inputx.dims()[0]; ++i) {
for (int j = 0; j < dimx_1; ++j) { for (int j = 0; j < dimx_1; ++j) {
DLOGF("%f ", inputx_ptr[i * dimx_1 + j]); DLOGF("%f ", inputx_ptr[i * dimx_1 + j]);
}
DLOGF("\n");
} }
DLOGF("\n");
auto dimy_1 = inputy.numel() / inputy.dims()[0]; }
DLOG << " inputy : ";
for (int i = 0; i < inputy.dims()[0]; ++i) { auto dimy_1 = inputy.numel() / inputy.dims()[0];
for (int j = 0; j < dimy_1; ++j) { DLOG << " inputy : ";
DLOGF("%f ", inputy_ptr[i * dimx_1 + j]); for (int i = 0; i < inputy.dims()[0]; ++i) {
} for (int j = 0; j < dimy_1; ++j) {
DLOGF("\n"); DLOGF("%f ", inputy_ptr[i * dimx_1 + j]);
} }
DLOGF("\n");
auto dim_output_1 = output_mul->numel() / output_mul->dims()[0]; }
DLOG << " output : ";
for (int i = 0; i < output_mul->dims()[0]; ++i) { auto dim_output_1 = output_mul->numel() / output_mul->dims()[0];
for (int j = 0; j < dim_output_1; ++j) { DLOG << " output : ";
DLOGF("%f ", output_mul_ptr[i * dimy_1 + j]); for (int i = 0; i < output_mul->dims()[0]; ++i) {
} for (int j = 0; j < dim_output_1; ++j) {
DLOGF("\n"); DLOGF("%f ", output_mul_ptr[i * dimy_1 + j]);
} }
DLOGF("\n");
}
/// output (3,3) /// output (3,3)
DLOG << "output memory size : " << output_mul->memory_size(); DLOG << "output memory size : " << output_mul->memory_size();
DLOG << "output numel : " << output_mul->numel(); DLOG << "output numel : " << output_mul->numel();
DLOG << inputx_ptr[0] << " x " << inputy_ptr[0] << " + " << inputx_ptr[1] DLOG << inputx_ptr[0] << " x " << inputy_ptr[0] << " + " << inputx_ptr[1]
<< " x " << inputy_ptr[0 + 3] << " = " << output_mul_ptr[0]; << " x " << inputy_ptr[0 + 3] << " = " << output_mul_ptr[0];
return 0; return 0;
} }
test/operators/test_pool_op.cpp
浏览文件 @ dff4a781
...@@ -21,26 +21,26 @@ SOFTWARE. ...@@ -21,26 +21,26 @@ SOFTWARE.
#include "io.h" #include "io.h"
int main() { int main() {
paddle_mobile::Loader<paddle_mobile::CPU> loader; paddle_mobile::Loader<paddle_mobile::CPU> loader;
auto program = loader.Load(std::string("../models/googlenet")); auto program = loader.Load(std::string("../models/googlenet"));
if (program.originProgram == nullptr) { if (program.originProgram == nullptr) {
DLOG << "program read file"; DLOG << "program read file";
} }
Executor4Test<paddle_mobile::CPU, Executor4Test<paddle_mobile::CPU,
paddle_mobile::operators::PoolOp<paddle_mobile::CPU, float>> paddle_mobile::operators::PoolOp<paddle_mobile::CPU, float>>
executor(program, "pool2d"); executor(program, "pool2d");
paddle_mobile::framework::Tensor input; paddle_mobile::framework::Tensor input;
SetupTensor<float>(&input, {1, 64, 112, 112}, static_cast<float>(0), SetupTensor<float>(&input, {1, 64, 112, 112}, static_cast<float>(0),
static_cast<float>(1)); static_cast<float>(1));
auto output = executor.predict(input, "conv2d_0.tmp_1", "pool2d_0.tmp_0", auto output = executor.predict(input, "conv2d_0.tmp_1", "pool2d_0.tmp_0",
{1, 64, 56, 56}); {1, 64, 56, 56});
float *output_ptr = output->data<float>(); float *output_ptr = output->data<float>();
for (int j = 0; j < output->numel(); ++j) { for (int j = 0; j < output->numel(); ++j) {
DLOG << " value of output: " << output_ptr[j]; DLOG << " value of output: " << output_ptr[j];
} }
return 0; return 0;
} }
test/test_helper.h
浏览文件 @ dff4a781
...@@ -24,13 +24,12 @@ SOFTWARE. ...@@ -24,13 +24,12 @@ SOFTWARE.
template <typename T> template <typename T>
void SetupTensor(paddle_mobile::framework::Tensor *input, void SetupTensor(paddle_mobile::framework::Tensor *input,
paddle_mobile::framework::DDim dims, T lower, T upper) { paddle_mobile::framework::DDim dims, T lower, T upper) {
static unsigned int seed = 100; static unsigned int seed = 100;
std::mt19937 rng(seed++); std::mt19937 rng(seed++);
std::uniform_real_distribution<double> uniform_dist(0, 1); std::uniform_real_distribution<double> uniform_dist(0, 1);
T *input_ptr = input->mutable_data<T>(dims); T *input_ptr = input->mutable_data<T>(dims);
for (int i = 0; i < input->numel(); ++i) { for (int i = 0; i < input->numel(); ++i) {
input_ptr[i] = input_ptr[i] = static_cast<T>(uniform_dist(rng) * (upper - lower) + lower);
static_cast<T>(uniform_dist(rng) * (upper - lower) + lower); }
}
} }
tools/pre-commit.hooks/.clang-tidy.hook
浏览文件 @ dff4a781
...@@ -6,8 +6,7 @@ TOTAL_ERRORS=0 ...@@ -6,8 +6,7 @@ TOTAL_ERRORS=0
# The trick to remove deleted files: https://stackoverflow.com/a/2413151 # The trick to remove deleted files: https://stackoverflow.com/a/2413151
for file in $(git diff --cached --name-status | awk '$1 != "D" {print $2}' | grep -v "third-party/" | grep -v ".pb."); do for file in $(git diff --cached --name-status | awk '$1 != "D" {print $2}' | grep -v "third-party/" | grep -v ".pb."); do
python ./tools/pre-commit.hooks/run-clang-tidy.py $file; bash -c "python ./tools/pre-commit.hooks/run-clang-tidy.py $file -fix";
echo "python ./tools/pre-commit.hooks/run-clang-tidy.py $file"
TOTAL_ERRORS=$(expr $TOTAL_ERRORS + $?); TOTAL_ERRORS=$(expr $TOTAL_ERRORS + $?);
done done
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或