提交 e3fb95ae 编写于 作者: Y Yuan Shuai 提交者: Xiaoyang LI

[ARM][CPU] Fix time counter of arm cpu profiler (#1925)

* Fix timer of arm cpu profiler. test=develop

* Fix un-added op in cmake.test=develop

* fix cmake error

* fix cmake error, test=develop

* Fix pass sequence. test=develop

* replace option with lite_option. test=develop

* disable profile mode by default. test=develop

* Fix error option name. test=develop
上级 118fe30b
...@@ -14,8 +14,9 @@ ...@@ -14,8 +14,9 @@
cmake_minimum_required(VERSION 3.0) cmake_minimum_required(VERSION 3.0)
set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_CURRENT_SOURCE_DIR}/cmake") set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_CURRENT_SOURCE_DIR}/cmake")
include(lite_utils)
option(WITH_PADDLE_MOBILE "Use the paddle-mobile legacy build" OFF) lite_option(WITH_PADDLE_MOBILE "Use the paddle-mobile legacy build" OFF)
if (WITH_PADDLE_MOBILE) if (WITH_PADDLE_MOBILE)
add_subdirectory(mobile) add_subdirectory(mobile)
return() return()
...@@ -44,13 +45,13 @@ find_package(Threads REQUIRED) ...@@ -44,13 +45,13 @@ find_package(Threads REQUIRED)
include(simd) include(simd)
################################ Exposed Configurations ####################################### ################################ Exposed Configurations #######################################
option(WITH_DSO "Compile PaddlePaddle with dynamic linked CUDA" ON) lite_option(WITH_DSO "Compile PaddlePaddle with dynamic linked CUDA" ON)
option(WITH_AVX "Compile PaddlePaddle with AVX intrinsics" ${AVX_FOUND}) lite_option(WITH_AVX "Compile PaddlePaddle with AVX intrinsics" ON IF ${AVX_FOUND})
option(WITH_PYTHON "Compile PaddlePaddle with python interpreter" ON) lite_option(WITH_PYTHON "Compile PaddlePaddle with python interpreter" ON)
option(WITH_TESTING "Compile PaddlePaddle with unit testing" OFF) lite_option(WITH_TESTING "Compile PaddlePaddle with unit testing" OFF)
option(WITH_MKL "Compile PaddlePaddle with MKL support." ${AVX_FOUND}) lite_option(WITH_MKL "Compile PaddlePaddle with MKL support." ON IF ${AVX_FOUND})
option(WITH_ARM_DOTPROD "Compile PaddlePaddle with ARM dot production" ON) lite_option(WITH_ARM_DOTPROD "Compile PaddlePaddle with ARM dot production" ON)
option(WITH_SYSTEM_BLAS "Use system blas library" OFF) lite_option(WITH_SYSTEM_BLAS "Use system blas library" OFF)
# TODO(Superjomn) Remove WITH_ANAKIN option if not needed latter. # TODO(Superjomn) Remove WITH_ANAKIN option if not needed latter.
if(ANDROID OR IOS OR ARMLINUX) if(ANDROID OR IOS OR ARMLINUX)
set(WITH_GPU OFF CACHE STRING set(WITH_GPU OFF CACHE STRING
...@@ -68,20 +69,21 @@ if(ANDROID OR IOS OR ARMLINUX) ...@@ -68,20 +69,21 @@ if(ANDROID OR IOS OR ARMLINUX)
endif() endif()
# for lite, both server and mobile framework. # for lite, both server and mobile framework.
option(LITE_WITH_JAVA "Enable Java JNI lib in lite mode" OFF) lite_option(LITE_WITH_JAVA "Enable Java JNI lib in lite mode" OFF)
option(LITE_WITH_CUDA "Enable CUDA in lite mode" OFF) lite_option(LITE_WITH_CUDA "Enable CUDA in lite mode" OFF)
option(LITE_WITH_X86 "Enable X86 in lite mode" ON) lite_option(LITE_WITH_X86 "Enable X86 in lite mode" ON)
option(LITE_WITH_ARM "Enable ARM in lite mode" OFF) lite_option(LITE_WITH_ARM "Enable ARM in lite mode" OFF)
option(LITE_WITH_NPU "Enable NPU in lite mode" OFF) lite_option(LITE_WITH_NPU "Enable NPU in lite mode" OFF)
option(LITE_WITH_OPENMP "Enable OpenMP in lite framework" ON) lite_option(LITE_WITH_OPENMP "Enable OpenMP in lite framework" ON)
option(LITE_WITH_OPENCL "Enable OpenCL support in lite" OFF) lite_option(LITE_WITH_OPENCL "Enable OpenCL support in lite" OFF)
option(LITE_WITH_FPGA "Enable FPGA support in lite" OFF) lite_option(LITE_WITH_FPGA "Enable FPGA support in lite" OFF)
option(LITE_WITH_LIGHT_WEIGHT_FRAMEWORK "Enable light-weight framework" OFF) lite_option(LITE_WITH_LIGHT_WEIGHT_FRAMEWORK "Enable light-weight framework" OFF)
option(LITE_WITH_PROFILE "Enable profile mode in lite framework" OFF) lite_option(LITE_WITH_PROFILE "Enable profile mode in lite framework" OFF)
option(LITE_SHUTDOWN_LOG "Shutdown log system or not." OFF) lite_option(LITE_WITH_PRECISION_PROFILE "Enable precision profile in profile mode ON in lite" OFF IF LITE_WITH_PROFILE)
option(LITE_ON_TINY_PUBLISH "Publish tiny predictor lib." OFF) lite_option(LITE_SHUTDOWN_LOG "Shutdown log system or not." OFF)
lite_option(LITE_ON_TINY_PUBLISH "Publish tiny predictor lib." OFF)
# publish options # publish options
option(LITE_BUILD_EXTRA "Enable extra algorithm support in Lite, both kernels and operators" OFF) lite_option(LITE_BUILD_EXTRA "Enable extra algorithm support in Lite, both kernels and operators" OFF)
set(THIRD_PARTY_PATH "${CMAKE_BINARY_DIR}/third_party" CACHE STRING set(THIRD_PARTY_PATH "${CMAKE_BINARY_DIR}/third_party" CACHE STRING
"A path setting third party libraries download & build directories.") "A path setting third party libraries download & build directories.")
......
...@@ -164,6 +164,9 @@ endif() ...@@ -164,6 +164,9 @@ endif()
if (LITE_WITH_PROFILE) if (LITE_WITH_PROFILE)
add_definitions("-DLITE_WITH_PROFILE") add_definitions("-DLITE_WITH_PROFILE")
if (LITE_WITH_PRECISION_PROFILE)
add_definitions("-DLITE_WITH_PRECISION_PROFILE")
endif()
endif() endif()
if (LITE_WITH_LIGHT_WEIGHT_FRAMEWORK) if (LITE_WITH_LIGHT_WEIGHT_FRAMEWORK)
......
# Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ----------------------------------------------------------------------------
# section: Provides an paddle lite config option macro
# usage: lite_option(var "help string to describe the var" [if or IF (condition)])
# ----------------------------------------------------------------------------
macro(lite_option variable description value)
set(__value ${value})
set(__condition "")
set(__varname "__value")
foreach(arg ${ARGN})
if(arg STREQUAL "IF" OR arg STREQUAL "if")
set(__varname "__condition")
else()
list(APPEND ${__varname} ${arg})
endif()
endforeach()
unset(__varname)
if(__condition STREQUAL "")
set(__condition 2 GREATER 1)
endif()
if(${__condition})
if(__value MATCHES ";")
if(${__value})
option(${variable} "${description}" ON)
else()
option(${variable} "${description}" OFF)
endif()
elseif(DEFINED ${__value})
if(${__value})
option(${variable} "${description}" ON)
else()
option(${variable} "${description}" OFF)
endif()
else()
option(${variable} "${description}" ${__value})
endif()
else()
unset(${variable} CACHE)
endif()
unset(__condition)
unset(__value)
endmacro()
...@@ -54,14 +54,15 @@ class Optimizer { ...@@ -54,14 +54,15 @@ class Optimizer {
if (passes.empty()) { if (passes.empty()) {
RunPasses(std::vector<std::string>{ RunPasses(std::vector<std::string>{
{"lite_quant_dequant_fuse_pass", // {"lite_quant_dequant_fuse_pass", //
"lite_conv_bn_fuse_pass", // "lite_conv_elementwise_fuse_pass", // conv-elemwise-bn
"lite_conv_bn_fuse_pass", //
"lite_conv_elementwise_fuse_pass", // conv-bn-elemwise
// This pass is disabled to force some opencl kernels selected for // This pass is disabled to force some opencl kernels selected for
// final running, otherwise, they will be fused to ARM fusion // final running, otherwise, they will be fused to ARM fusion
// kernels, and the OpenCL devices will be discarded. // kernels, and the OpenCL devices will be discarded.
// TODO(Superjomn) Refine the fusion related design to select fusion // TODO(Superjomn) Refine the fusion related design to select fusion
// kernels for devices automatically. // kernels for devices automatically.
"lite_conv_elementwise_fuse_pass", //
"lite_conv_activation_fuse_pass", // "lite_conv_activation_fuse_pass", //
"lite_fc_fuse_pass", // "lite_fc_fuse_pass", //
"lite_shuffle_channel_fuse_pass", // "lite_shuffle_channel_fuse_pass", //
......
...@@ -61,7 +61,7 @@ class BasicTimer : TimerBase<BasicTimer> { ...@@ -61,7 +61,7 @@ class BasicTimer : TimerBase<BasicTimer> {
uint32_t min_{std::numeric_limits<uint32_t>::max()}; uint32_t min_{std::numeric_limits<uint32_t>::max()};
int id_{-1}; int id_{-1};
std::string key_; std::string key_;
std::chrono::time_point<std::chrono::high_resolution_clock> timer_{}; uint64_t timer_{};
// TODO(Superjomn) make static // TODO(Superjomn) make static
static const int name_w; static const int name_w;
...@@ -73,11 +73,20 @@ class BasicTimer : TimerBase<BasicTimer> { ...@@ -73,11 +73,20 @@ class BasicTimer : TimerBase<BasicTimer> {
void SetId(int id) { id_ = id; } void SetId(int id) { id_ = id; }
void SetKey(const std::string &key) { key_ = key; } void SetKey(const std::string &key) { key_ = key; }
void Start() { timer_ = std::chrono::high_resolution_clock::now(); } void Start() {
timer_ = static_cast<uint64_t>(
std::chrono::duration_cast<std::chrono::microseconds>(
std::chrono::system_clock::now().time_since_epoch())
.count());
}
void Stop() { void Stop() {
auto duration = std::chrono::duration_cast<std::chrono::milliseconds>( auto duration = static_cast<
std::chrono::high_resolution_clock::now() - timer_); uint64_t>( // timer unit: microsecond, 1second = 1e6 microsecond
Log(duration.count()); std::chrono::duration_cast<std::chrono::microseconds>(
std::chrono::system_clock::now().time_since_epoch())
.count() -
timer_);
Log(duration);
} }
int count() const { return count_; } int count() const { return count_; }
......
...@@ -118,8 +118,10 @@ void RuntimeProgram::Run() { ...@@ -118,8 +118,10 @@ void RuntimeProgram::Run() {
inst.Run(); inst.Run();
#ifdef LITE_WITH_PROFILE #ifdef LITE_WITH_PROFILE
#ifdef LITE_WITH_PRECISION_PROFILE
LITE_PRECISION_PROFILE(inst) LITE_PRECISION_PROFILE(inst)
#endif #endif // LITE_WITH_PRECISION_PROFILE
#endif // LITE_WITH_PROFILE
} }
} }
......
...@@ -63,6 +63,8 @@ add_operator(reduce_mean_op basic SRCS reduce_mean_op.cc DEPS ${op_DEPS}) ...@@ -63,6 +63,8 @@ add_operator(reduce_mean_op basic SRCS reduce_mean_op.cc DEPS ${op_DEPS})
add_operator(stack_op basic SRCS stack_op.cc DEPS ${op_DEPS}) add_operator(stack_op basic SRCS stack_op.cc DEPS ${op_DEPS})
add_operator(cast_op_lite basic SRCS cast_op.cc DEPS ${op_DEPS}) add_operator(cast_op_lite basic SRCS cast_op.cc DEPS ${op_DEPS})
add_operator(assign_op basic SRCS assign_op.cc DEPS ${op_DEPS}) add_operator(assign_op basic SRCS assign_op.cc DEPS ${op_DEPS})
add_operator(flatten_op basic SRCS flatten_op.cc DEPS ${op_DEPS})
add_operator(fake_quantize_range_abs_max_op basic SRCS fake_quantize_range_abs_max.cc DEPS ${op_DEPS})
# for OCR specific # for OCR specific
add_operator(while_op extra SRCS while_op.cc DEPS ${op_DEPS}) add_operator(while_op extra SRCS while_op.cc DEPS ${op_DEPS})
......
Markdown is supported
0% .
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