Skip to content

P
Paddle-Lite

PaddlePaddle / Paddle-Lite

通知 338
Star 4
Fork 1
P
Paddle-Lite
“0c851cab2294356dd292b9b4458379d1bde4eadd”上不存在“paddle/phi/kernels/impl/selu_grad_kernel_impl.h”
提交 035c72f2 编写于 作者: E eclipsess
浏览文件
操作

add fc in types

上级 ff0f7f7b 86da216d
隐藏空白更改
内联 并排
Showing with 1863 addition and 2438 deletion
CMakeLists.txt
浏览文件 @ 035c72f2
cmake_minimum_required(VERSION 3.0)
project(paddle-mobile)
add_definitions(-DPADDLE_MOBILE_DEBUG)
add_definitions(-DENABLE_EXCEPTION)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
set(CMAKE_BUILD_TYPE RelWithDebInfo)
option(DEBUGING "enable debug mode" OFF)
option(USE_OPENMP "openmp support" OFF)
option(USE_EXCEPTION "use std exception" OFF)
if (DEBUGING)
set(CMAKE_BUILD_TYPE Debug)
else()
set(CMAKE_BUILD_TYPE Release)
endif ()
if(DEBUGING)
message(STATUS "debuging")
add_definitions(-DPADDLE_MOBILE_DEBUG)
if(ANDROID)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -llog")
endif()
else()
message(STATUS "releasing")
add_definitions(-fvisibility=hidden -fvisibility-inlines-hidden)
endif()
if (USE_EXCEPTION)
add_definitions(-DENABLE_EXCEPTION)
add_definitions(-fexceptions)
else()
add_definitions(-fno-exceptions)
endif ()
if(IS_MAC)
add_definitions(-DX86)
elseif(IS_IOS)
add_definitions(-DIOS)
elseif(V7)
add_definitions(-DARMV7)
elseif(V8)
add_definitions(-DARMV8)
else ()
add_definitions(-DX86)
endif()
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14")
set(CMAKE_VERBOSE_MAKEFILE ON)
set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY build)
......@@ -14,26 +52,86 @@ set(CMAKE_RUNTIME_OUTPUT_DIRECTORY build)
file(GLOB_RECURSE PADDLE_MOBILE_CC src/*.cc src/*.cpp src/*.c)
file(GLOB_RECURSE PADDLE_MOBILE_H src/*.h)
# include headers
if (NOT ANDROID)
list(REMOVE_ITEM PADDLE_MOBILE_CC ${CMAKE_CURRENT_SOURCE_DIR}/src/jni/*.cpp)
list(REMOVE_ITEM PADDLE_MOBILE_CC ${CMAKE_CURRENT_SOURCE_DIR}/src/jni/*.h)
list(REMOVE_ITEM PADDLE_MOBILE_CC ${CMAKE_CURRENT_SOURCE_DIR}/src/operators/math/math_func_neon.h)
endif ()
include_directories(src/)
#include(ExternalProject)
#ExternalProject_Add(openblas_proj
# GIT_REPOSITORY "https://github.com/xianyi/OpenBLAS.git"
# GIT_TAG "v0.2.20"
# SOURCE_DIR "openblas/"
# BUILD_IN_SOURCE 1
# CONFIGURE_COMMAND ""
# BUILD_COMMAND "make" "ONLY_CBLAS=1"
# INSTALL_COMMAND "make" "PREFIX=${CMAKE_BINARY_DIR}/" "install"
# )
#set_target_properties(openblas_proj PROPERTIES EXCLUDE_FROM_ALL 1)
if(USE_OPENMP)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopenmp")
add_definitions(-DPADDLE_MOBILE_USE_OPENMP)
endif()
if (googlenet)
add_definitions(-DCONCAT_OP)
add_definitions(-DCONV_OP)
add_definitions(-DLRN_OP)
add_definitions(-DMUL_OP)
add_definitions(-DELEMENTWISEADD_OP)
add_definitions(-DFUSION_FC_OP)
add_definitions(-DPOOL_OP)
add_definitions(-DRELU_OP)
elseif (mobilenet)
add_definitions(-DCONV_OP)
add_definitions(-DELEMENTWISEADD_OP)
add_definitions(-DRELU_OP)
add_definitions(-DSOFTMAX_OP)
add_definitions(-DSOFTMAX_OP)
add_definitions(-DDEPTHWISECONV_OP)
add_definitions(-DBATCHNORM_OP)
add_definitions(-DPOOL_OP)
add_definitions(-DRESHAPE_OP)
elseif (yolo)
add_definitions(-DBATCHNORM_OP)
add_definitions(-DCONV_OP)
add_definitions(-DRELU_OP)
add_definitions(-DELEMENTWISEADD_OP)
elseif (squeezenet)
add_definitions(-DCONCAT_OP)
add_definitions(-DCONV_OP)
add_definitions(-DRELU_OP)
add_definitions(-DELEMENTWISEADD_OP)
add_definitions(-DPOOL_OP)
add_definitions(-DRESHAPE_OP)
add_definitions(-DSOFTMAX_OP)
elseif(resnet)
add_definitions(-DCONV_OP)
add_definitions(-DBATCHNORM_OP)
add_definitions(-DELEMENTWISEADD_OP)
add_definitions(-DSOFTMAX_OP)
add_definitions(-DMUL_OP)
add_definitions(-DPOOL_OP)
add_definitions(-DRELU_OP)
else ()
add_definitions(-DBATCHNORM_OP)
add_definitions(-DBOXCODER_OP)
add_definitions(-DCONCAT_OP)
add_definitions(-DCONV_OP)
add_definitions(-DDEPTHWISECONV_OP)
add_definitions(-DELEMENTWISEADD_OP)
add_definitions(-DFUSIONCONVADD_OP)
add_definitions(-DCONVADDRELU_OP)
add_definitions(-DFUSION_FC_OP)
add_definitions(-DLRN_OP)
add_definitions(-DMUL_OP)
add_definitions(-DMULTICLASSNMS_OP)
add_definitions(-DPOOL_OP)
add_definitions(-DPRIORBOX_OP)
add_definitions(-DRELU_OP)
add_definitions(-DRESHAPE_OP)
add_definitions(-DSIGMOID_OP)
add_definitions(-DSOFTMAX_OP)
add_definitions(-DTRANSPOSE_OP)
endif()
#add_dependencies(paddle-mobile openblas_proj)
add_library(paddle-mobile SHARED ${PADDLE_MOBILE_CC} ${PADDLE_MOBILE_H})
# gen static
ADD_LIBRARY(paddle-mobile SHARED ${PADDLE_MOBILE_CC} ${PADDLE_MOBILE_H})
if(DEBUGING)
add_subdirectory(test)
endif()
#add_dependencies(paddle-mobile openblas_proj)
add_subdirectory(test)
src/common/enforce.h
浏览文件 @ 035c72f2
......@@ -17,8 +17,6 @@ limitations under the License. */
#ifdef ENABLE_EXCEPTION
#include <stdio.h>
#include <exception>
#include <sstream>
#include <stdexcept>
#include <string>
#endif
......@@ -32,12 +30,11 @@ struct PaddleMobileException : public std::exception {
PaddleMobileException(const char *header, const char *detail,
const char *file, const int line) {
std::stringstream ss;
ss << exception_prefix << "| " << header << "\n";
ss << "| [in file] : " << file << " \n";
ss << "| [on line] : " << line << " \n";
ss << "| [detail] : " << detail;
message = ss.str();
char buffer[1500];
snprintf(buffer, sizeof(buffer),
"%s| %s \n| [in file] : %s\n| [on line] : %d\n| [detail] : %s\n",
exception_prefix.c_str(), header, file, line, detail);
message = std::string(buffer);
}
const char *what() const noexcept { return message.c_str(); }
};
......
src/common/log.h
浏览文件 @ 035c72f2
......@@ -20,11 +20,38 @@ limitations under the License. */
#include <sstream>
#include <string>
#endif
#ifdef ANDROID
#include <android/log.h>
#endif
namespace paddle_mobile {
#ifdef PADDLE_MOBILE_DEBUG
#ifdef ANDROID
extern const char *ANDROID_LOG_TAG;
#define ANDROIDLOGI(...) \
__android_log_print(ANDROID_LOG_INFO, ANDROID_LOG_TAG, __VA_ARGS__); \
printf(__VA_ARGS__)
#define ANDROIDLOGW(...) \
__android_log_print(ANDROID_LOG_WARNING, ANDROID_LOG_TAG, __VA_ARGS__); \
printf(__VA_ARGS__)
#define ANDROIDLOGD(...) \
__android_log_print(ANDROID_LOG_DEBUG, ANDROID_LOG_TAG, __VA_ARGS__); \
printf(__VA_ARGS__)
#define ANDROIDLOGE(...) \
__android_log_print(ANDROID_LOG_ERROR, ANDROID_LOG_TAG, __VA_ARGS__); \
printf(__VA_ARGS__)
#else
#define ANDROIDLOGI(...)
#define ANDROIDLOGW(...)
#define ANDROIDLOGD(...)
#define ANDROIDLOGE(...)
#endif
enum LogLevel {
kNO_LOG,
kLOG_ERROR,
......@@ -122,6 +149,11 @@ struct ToLog {
#else
#define ANDROIDLOGI(...)
#define ANDROIDLOGW(...)
#define ANDROIDLOGD(...)
#define ANDROIDLOGE(...)
enum LogLevel {
kNO_LOG,
kLOG_ERROR,
......
src/common/log.cpp → src/common/macros.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,6 @@ 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. */
#include "log.h"
#pragma once
namespace paddle_mobile {}
#define EXPORT __attribute__((visibility("default")))
src/platform/macros.h → src/common/openmp-fix.cpp
浏览文件 @ 035c72f2
......@@ -12,14 +12,16 @@ 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. */
#pragma once
// Disable the copy and assignment operator for a class.
#ifndef DISABLE_COPY_AND_ASSIGN
#define DISABLE_COPY_AND_ASSIGN(classname) \
private: \
classname(const classname &) = delete; \
classname(classname &&) = delete; \
classname &operator=(const classname &) = delete; \
classname &operator=(classname &&) = delete
#ifdef PADDLE_MOBILE_USE_OPENMP
/**
* android-ndk-r17 has a problem when linking with openmp.
* if paddle-mobile enables -fopenmp, but didn't use those omp_* functions,
* after linking another binary with libpaddle-mobile.so, the omp_get_thread_num
* will not work. see test/common/test_openmp.cc the detailed reason is still
* unclear, but this trick will work. a better solution is hacking the linker,
* try some flags to make it link omp_* functions, but I didn't find out how to
* make it work.
*/
#include <omp.h>
static int _ = omp_get_num_procs();
#endif
src/common/protobuf-c.c
浏览文件 @ 035c72f2
......@@ -711,47 +711,6 @@ static inline size_t uint32_pack(uint32_t value, uint8_t *out) {
return rv;
}
/**
* Pack a signed 32-bit integer and return the number of bytes written.
* Negative numbers are encoded as two's complement 64-bit integers.
*
* \param value
* Value to encode.
* \param[out] out
* Packed value.
* \return
* Number of bytes written to `out`.
*/
static inline size_t int32_pack(int32_t value, uint8_t *out) {
if (value < 0) {
out[0] = value | 0x80;
out[1] = (value >> 7) | 0x80;
out[2] = (value >> 14) | 0x80;
out[3] = (value >> 21) | 0x80;
out[4] = (value >> 28) | 0x80;
out[5] = out[6] = out[7] = out[8] = 0xff;
out[9] = 0x01;
return 10;
} else {
return uint32_pack(value, out);
}
}
/**
* Pack a signed 32-bit integer using ZigZag encoding and return the number of
* bytes written.
*
* \param value
* Value to encode.
* \param[out] out
* Packed value.
* \return
* Number of bytes written to `out`.
*/
static inline size_t sint32_pack(int32_t value, uint8_t *out) {
return uint32_pack(zigzag32(value), out);
}
/**
* Pack a 64-bit unsigned integer using base-128 varint encoding and return the
* number of bytes written.
......@@ -789,116 +748,6 @@ static size_t uint64_pack(uint64_t value, uint8_t *out) {
return rv;
}
/**
* Pack a 64-bit signed integer in ZigZag encoding and return the number of
* bytes written.
*
* \param value
* Value to encode.
* \param[out] out
* Packed value.
* \return
* Number of bytes written to `out`.
*/
static inline size_t sint64_pack(int64_t value, uint8_t *out) {
return uint64_pack(zigzag64(value), out);
}
/**
* Pack a 32-bit quantity in little-endian byte order. Used for protobuf wire
* types fixed32, sfixed32, float. Similar to "htole32".
*
* \param value
* Value to encode.
* \param[out] out
* Packed value.
* \return
* Number of bytes written to `out`.
*/
static inline size_t fixed32_pack(uint32_t value, void *out) {
#if !defined(WORDS_BIGENDIAN)
memcpy(out, &value, 4);
#else
uint8_t *buf = out;
buf[0] = value;
buf[1] = value >> 8;
buf[2] = value >> 16;
buf[3] = value >> 24;
#endif
return 4;
}
/**
* Pack a 64-bit quantity in little-endian byte order. Used for protobuf wire
* types fixed64, sfixed64, double. Similar to "htole64".
*
* \todo The big-endian impl is really only good for 32-bit machines, a 64-bit
* version would be appreciated, plus a way to decide to use 64-bit math where
* convenient.
*
* \param value
* Value to encode.
* \param[out] out
* Packed value.
* \return
* Number of bytes written to `out`.
*/
static inline size_t fixed64_pack(uint64_t value, void *out) {
#if !defined(WORDS_BIGENDIAN)
memcpy(out, &value, 8);
#else
fixed32_pack(value, out);
fixed32_pack(value >> 32, ((char *)out) + 4);
#endif
return 8;
}
/**
* Pack a boolean value as an integer and return the number of bytes written.
*
* \todo Perhaps on some platforms *out = !!value would be a better impl, b/c
* that is idiomatic C++ in some STL implementations.
*
* \param value
* Value to encode.
* \param[out] out
* Packed value.
* \return
* Number of bytes written to `out`.
*/
static inline size_t boolean_pack(protobuf_c_boolean value, uint8_t *out) {
*out = value ? TRUE : FALSE;
return 1;
}
/**
* Pack a NUL-terminated C string and return the number of bytes written. The
* output includes a length delimiter.
*
* The NULL pointer is treated as an empty string. This isn't really necessary,
* but it allows people to leave required strings blank. (See Issue #13 in the
* bug tracker for a little more explanation).
*
* \param str
* String to encode.
* \param[out] out
* Packed value.
* \return
* Number of bytes written to `out`.
*/
static inline size_t string_pack(const char *str, uint8_t *out) {
if (str == NULL) {
out[0] = 0;
return 1;
} else {
size_t len = strlen(str);
size_t rv = uint32_pack(len, out);
memcpy(out + rv, str, len);
return rv + len;
}
}
/**
* Pack a ProtobufCBinaryData and return the number of bytes written. The output
* includes a length delimiter.
......@@ -918,30 +767,6 @@ static inline size_t binary_data_pack(const ProtobufCBinaryData *bd,
return rv + len;
}
/**
* Pack a ProtobufCMessage and return the number of bytes written. The output
* includes a length delimiter.
*
* \param message
* ProtobufCMessage object to pack.
* \param[out] out
* Packed message.
* \return
* Number of bytes written to `out`.
*/
static inline size_t prefixed_message_pack(const ProtobufCMessage *message,
uint8_t *out) {
if (message == NULL) {
out[0] = 0;
return 1;
} else {
size_t rv = protobuf_c_message_pack(message, out + 1);
uint32_t rv_packed_size = uint32_size(rv);
if (rv_packed_size != 1) memmove(out + rv_packed_size, out + 1, rv);
return uint32_pack(rv, out) + rv;
}
}
/**
* Pack a field tag.
*
......@@ -963,143 +788,6 @@ static size_t tag_pack(uint32_t id, uint8_t *out) {
return uint64_pack(((uint64_t)id) << 3, out);
}
/**
* Pack a required field and return the number of bytes written.
*
* \param field
* Field descriptor.
* \param member
* The field member.
* \param[out] out
* Packed value.
* \return
* Number of bytes written to `out`.
*/
static size_t required_field_pack(const ProtobufCFieldDescriptor *field,
const void *member, uint8_t *out) {
size_t rv = tag_pack(field->id, out);
switch (field->type) {
case PROTOBUF_C_TYPE_SINT32:
out[0] |= PROTOBUF_C_WIRE_TYPE_VARINT;
return rv + sint32_pack(*(const int32_t *)member, out + rv);
case PROTOBUF_C_TYPE_ENUM:
case PROTOBUF_C_TYPE_INT32:
out[0] |= PROTOBUF_C_WIRE_TYPE_VARINT;
return rv + int32_pack(*(const int32_t *)member, out + rv);
case PROTOBUF_C_TYPE_UINT32:
out[0] |= PROTOBUF_C_WIRE_TYPE_VARINT;
return rv + uint32_pack(*(const uint32_t *)member, out + rv);
case PROTOBUF_C_TYPE_SINT64:
out[0] |= PROTOBUF_C_WIRE_TYPE_VARINT;
return rv + sint64_pack(*(const int64_t *)member, out + rv);
case PROTOBUF_C_TYPE_INT64:
case PROTOBUF_C_TYPE_UINT64:
out[0] |= PROTOBUF_C_WIRE_TYPE_VARINT;
return rv + uint64_pack(*(const uint64_t *)member, out + rv);
case PROTOBUF_C_TYPE_SFIXED32:
case PROTOBUF_C_TYPE_FIXED32:
case PROTOBUF_C_TYPE_FLOAT:
out[0] |= PROTOBUF_C_WIRE_TYPE_32BIT;
return rv + fixed32_pack(*(const uint32_t *)member, out + rv);
case PROTOBUF_C_TYPE_SFIXED64:
case PROTOBUF_C_TYPE_FIXED64:
case PROTOBUF_C_TYPE_DOUBLE:
out[0] |= PROTOBUF_C_WIRE_TYPE_64BIT;
return rv + fixed64_pack(*(const uint64_t *)member, out + rv);
case PROTOBUF_C_TYPE_BOOL:
out[0] |= PROTOBUF_C_WIRE_TYPE_VARINT;
return rv + boolean_pack(*(const protobuf_c_boolean *)member, out + rv);
case PROTOBUF_C_TYPE_STRING:
out[0] |= PROTOBUF_C_WIRE_TYPE_LENGTH_PREFIXED;
return rv + string_pack(*(char *const *)member, out + rv);
case PROTOBUF_C_TYPE_BYTES:
out[0] |= PROTOBUF_C_WIRE_TYPE_LENGTH_PREFIXED;
return rv +
binary_data_pack((const ProtobufCBinaryData *)member, out + rv);
case PROTOBUF_C_TYPE_MESSAGE:
out[0] |= PROTOBUF_C_WIRE_TYPE_LENGTH_PREFIXED;
return rv + prefixed_message_pack(*(ProtobufCMessage *const *)member,
out + rv);
}
PROTOBUF_C__ASSERT_NOT_REACHED();
return 0;
}
/**
* Pack a oneof field and return the number of bytes written. Only packs the
* field that is selected by the case enum.
*
* \param field
* Field descriptor.
* \param oneof_case
* Enum value that selects the field in the oneof.
* \param member
* The field member.
* \param[out] out
* Packed value.
* \return
* Number of bytes written to `out`.
*/
static size_t oneof_field_pack(const ProtobufCFieldDescriptor *field,
uint32_t oneof_case, const void *member,
uint8_t *out) {
if (oneof_case != field->id) {
return 0;
}
if (field->type == PROTOBUF_C_TYPE_MESSAGE ||
field->type == PROTOBUF_C_TYPE_STRING) {
const void *ptr = *(const void *const *)member;
if (ptr == NULL || ptr == field->default_value) return 0;
}
return required_field_pack(field, member, out);
}
/**
* Pack an optional field and return the number of bytes written.
*
* \param field
* Field descriptor.
* \param has
* Whether the field is set.
* \param member
* The field member.
* \param[out] out
* Packed value.
* \return
* Number of bytes written to `out`.
*/
static size_t optional_field_pack(const ProtobufCFieldDescriptor *field,
const protobuf_c_boolean has,
const void *member, uint8_t *out) {
if (field->type == PROTOBUF_C_TYPE_MESSAGE ||
field->type == PROTOBUF_C_TYPE_STRING) {
const void *ptr = *(const void *const *)member;
if (ptr == NULL || ptr == field->default_value) return 0;
} else {
if (!has) return 0;
}
return required_field_pack(field, member, out);
}
/**
* Pack an unlabeled field and return the number of bytes written.
*
* \param field
* Field descriptor.
* \param member
* The field member.
* \param[out] out
* Packed value.
* \return
* Number of bytes written to `out`.
*/
static size_t unlabeled_field_pack(const ProtobufCFieldDescriptor *field,
const void *member, uint8_t *out) {
if (field_is_zeroish(field, member)) return 0;
return required_field_pack(field, member, out);
}
/**
* Given a field type, return the in-memory size.
*
......@@ -1139,236 +827,6 @@ static inline size_t sizeof_elt_in_repeated_array(ProtobufCType type) {
return 0;
}
/**
* Pack an array of 32-bit quantities.
*
* \param[out] out
* Destination.
* \param[in] in
* Source.
* \param[in] n
* Number of elements in the source array.
*/
static void copy_to_little_endian_32(void *out, const void *in,
const unsigned n) {
#if !defined(WORDS_BIGENDIAN)
memcpy(out, in, n * 4);
#else
unsigned i;
const uint32_t *ini = in;
for (i = 0; i < n; i++) fixed32_pack(ini[i], (uint32_t *)out + i);
#endif
}
/**
* Pack an array of 64-bit quantities.
*
* \param[out] out
* Destination.
* \param[in] in
* Source.
* \param[in] n
* Number of elements in the source array.
*/
static void copy_to_little_endian_64(void *out, const void *in,
const unsigned n) {
#if !defined(WORDS_BIGENDIAN)
memcpy(out, in, n * 8);
#else
unsigned i;
const uint64_t *ini = in;
for (i = 0; i < n; i++) fixed64_pack(ini[i], (uint64_t *)out + i);
#endif
}
/**
* Get the minimum number of bytes required to pack a field value of a
* particular type.
*
* \param type
* Field type.
* \return
* Number of bytes.
*/
static unsigned get_type_min_size(ProtobufCType type) {
if (type == PROTOBUF_C_TYPE_SFIXED32 || type == PROTOBUF_C_TYPE_FIXED32 ||
type == PROTOBUF_C_TYPE_FLOAT) {
return 4;
}
if (type == PROTOBUF_C_TYPE_SFIXED64 || type == PROTOBUF_C_TYPE_FIXED64 ||
type == PROTOBUF_C_TYPE_DOUBLE) {
return 8;
}
return 1;
}
/**
* Get the packed size of an array of same field type.
*
* \param field
* Field descriptor.
* \param count
* Number of elements of this type.
* \param array
* The elements to get the size of.
* \return
* Number of bytes required.
*/
static size_t get_packed_payload_length(const ProtobufCFieldDescriptor *field,
unsigned count, const void *array) {
unsigned rv = 0;
unsigned i;
switch (field->type) {
case PROTOBUF_C_TYPE_SFIXED32:
case PROTOBUF_C_TYPE_FIXED32:
case PROTOBUF_C_TYPE_FLOAT:
return count * 4;
case PROTOBUF_C_TYPE_SFIXED64:
case PROTOBUF_C_TYPE_FIXED64:
case PROTOBUF_C_TYPE_DOUBLE:
return count * 8;
case PROTOBUF_C_TYPE_ENUM:
case PROTOBUF_C_TYPE_INT32: {
const int32_t *arr = (const int32_t *)array;
for (i = 0; i < count; i++) rv += int32_size(arr[i]);
break;
}
case PROTOBUF_C_TYPE_SINT32: {
const int32_t *arr = (const int32_t *)array;
for (i = 0; i < count; i++) rv += sint32_size(arr[i]);
break;
}
case PROTOBUF_C_TYPE_UINT32: {
const uint32_t *arr = (const uint32_t *)array;
for (i = 0; i < count; i++) rv += uint32_size(arr[i]);
break;
}
case PROTOBUF_C_TYPE_SINT64: {
const int64_t *arr = (const int64_t *)array;
for (i = 0; i < count; i++) rv += sint64_size(arr[i]);
break;
}
case PROTOBUF_C_TYPE_INT64:
case PROTOBUF_C_TYPE_UINT64: {
const uint64_t *arr = (const uint64_t *)array;
for (i = 0; i < count; i++) rv += uint64_size(arr[i]);
break;
}
case PROTOBUF_C_TYPE_BOOL:
return count;
default:
PROTOBUF_C__ASSERT_NOT_REACHED();
}
return rv;
}
/**
* Pack an array of same field type to a virtual buffer.
*
* \param field
* Field descriptor.
* \param count
* Number of elements of this type.
* \param array
* The elements to get the size of.
* \param[out] buffer
* Virtual buffer to append data to.
* \return
* Number of bytes packed.
*/
static size_t pack_buffer_packed_payload(const ProtobufCFieldDescriptor *field,
unsigned count, const void *array,
ProtobufCBuffer *buffer) {
uint8_t scratch[16];
size_t rv = 0;
unsigned i;
switch (field->type) {
case PROTOBUF_C_TYPE_SFIXED32:
case PROTOBUF_C_TYPE_FIXED32:
case PROTOBUF_C_TYPE_FLOAT:
#if !defined(WORDS_BIGENDIAN)
rv = count * 4;
goto no_packing_needed;
#else
for (i = 0; i < count; i++) {
unsigned len = fixed32_pack(((uint32_t *)array)[i], scratch);
buffer->append(buffer, len, scratch);
rv += len;
}
break;
#endif
case PROTOBUF_C_TYPE_SFIXED64:
case PROTOBUF_C_TYPE_FIXED64:
case PROTOBUF_C_TYPE_DOUBLE:
#if !defined(WORDS_BIGENDIAN)
rv = count * 8;
goto no_packing_needed;
#else
for (i = 0; i < count; i++) {
unsigned len = fixed64_pack(((uint64_t *)array)[i], scratch);
buffer->append(buffer, len, scratch);
rv += len;
}
break;
#endif
case PROTOBUF_C_TYPE_ENUM:
case PROTOBUF_C_TYPE_INT32:
for (i = 0; i < count; i++) {
unsigned len = int32_pack(((int32_t *)array)[i], scratch);
buffer->append(buffer, len, scratch);
rv += len;
}
break;
case PROTOBUF_C_TYPE_SINT32:
for (i = 0; i < count; i++) {
unsigned len = sint32_pack(((int32_t *)array)[i], scratch);
buffer->append(buffer, len, scratch);
rv += len;
}
break;
case PROTOBUF_C_TYPE_UINT32:
for (i = 0; i < count; i++) {
unsigned len = uint32_pack(((uint32_t *)array)[i], scratch);
buffer->append(buffer, len, scratch);
rv += len;
}
break;
case PROTOBUF_C_TYPE_SINT64:
for (i = 0; i < count; i++) {
unsigned len = sint64_pack(((int64_t *)array)[i], scratch);
buffer->append(buffer, len, scratch);
rv += len;
}
break;
case PROTOBUF_C_TYPE_INT64:
case PROTOBUF_C_TYPE_UINT64:
for (i = 0; i < count; i++) {
unsigned len = uint64_pack(((uint64_t *)array)[i], scratch);
buffer->append(buffer, len, scratch);
rv += len;
}
break;
case PROTOBUF_C_TYPE_BOOL:
for (i = 0; i < count; i++) {
unsigned len = boolean_pack(((protobuf_c_boolean *)array)[i], scratch);
buffer->append(buffer, len, scratch);
rv += len;
}
return count;
default:
PROTOBUF_C__ASSERT_NOT_REACHED();
}
return rv;
#if !defined(WORDS_BIGENDIAN)
no_packing_needed:
buffer->append(buffer, rv, array);
return rv;
#endif
}
static inline int int_range_lookup(unsigned n_ranges,
const ProtobufCIntRange *ranges, int value) {
unsigned n;
......@@ -2638,147 +2096,3 @@ protobuf_c_boolean protobuf_c_message_check(const ProtobufCMessage *message) {
typedef void (*GenericHandler)(void *service, const ProtobufCMessage *input,
ProtobufCClosure closure, void *closure_data);
void protobuf_c_service_invoke_internal(ProtobufCService *service,
unsigned method_index,
const ProtobufCMessage *input,
ProtobufCClosure closure,
void *closure_data) {
GenericHandler *handlers;
GenericHandler handler;
/*
* Verify that method_index is within range. If this fails, you are
* likely invoking a newly added method on an old service. (Although
* other memory corruption bugs can cause this assertion too.)
*/
assert(method_index < service->descriptor->n_methods);
/*
* Get the array of virtual methods (which are enumerated by the
* generated code).
*/
handlers = (GenericHandler *)(service + 1);
/*
* Get our method and invoke it.
* \todo Seems like handler == NULL is a situation that needs handling.
*/
handler = handlers[method_index];
(*handler)(service, input, closure, closure_data);
}
void protobuf_c_service_generated_init(
ProtobufCService *service, const ProtobufCServiceDescriptor *descriptor,
ProtobufCServiceDestroy destroy) {
ASSERT_IS_SERVICE_DESCRIPTOR(descriptor);
service->descriptor = descriptor;
service->destroy = destroy;
service->invoke = protobuf_c_service_invoke_internal;
memset(service + 1, 0, descriptor->n_methods * sizeof(GenericHandler));
}
void protobuf_c_service_destroy(ProtobufCService *service) {
service->destroy(service);
}
/* --- querying the descriptors --- */
const ProtobufCEnumValue *protobuf_c_enum_descriptor_get_value_by_name(
const ProtobufCEnumDescriptor *desc, const char *name) {
unsigned start = 0;
unsigned count;
if (desc == NULL || desc->values_by_name == NULL) return NULL;
count = desc->n_value_names;
while (count > 1) {
unsigned mid = start + count / 2;
int rv = strcmp(desc->values_by_name[mid].name, name);
if (rv == 0)
return desc->values + desc->values_by_name[mid].index;
else if (rv < 0) {
count = start + count - (mid + 1);
start = mid + 1;
} else
count = mid - start;
}
if (count == 0) return NULL;
if (strcmp(desc->values_by_name[start].name, name) == 0)
return desc->values + desc->values_by_name[start].index;
return NULL;
}
const ProtobufCEnumValue *protobuf_c_enum_descriptor_get_value(
const ProtobufCEnumDescriptor *desc, int value) {
int rv = int_range_lookup(desc->n_value_ranges, desc->value_ranges, value);
if (rv < 0) return NULL;
return desc->values + rv;
}
const ProtobufCFieldDescriptor *protobuf_c_message_descriptor_get_field_by_name(
const ProtobufCMessageDescriptor *desc, const char *name) {
unsigned start = 0;
unsigned count;
const ProtobufCFieldDescriptor *field;
if (desc == NULL || desc->fields_sorted_by_name == NULL) return NULL;
count = desc->n_fields;
while (count > 1) {
unsigned mid = start + count / 2;
int rv;
field = desc->fields + desc->fields_sorted_by_name[mid];
rv = strcmp(field->name, name);
if (rv == 0)
return field;
else if (rv < 0) {
count = start + count - (mid + 1);
start = mid + 1;
} else
count = mid - start;
}
if (count == 0) return NULL;
field = desc->fields + desc->fields_sorted_by_name[start];
if (strcmp(field->name, name) == 0) return field;
return NULL;
}
const ProtobufCFieldDescriptor *protobuf_c_message_descriptor_get_field(
const ProtobufCMessageDescriptor *desc, unsigned value) {
int rv = int_range_lookup(desc->n_field_ranges, desc->field_ranges, value);
if (rv < 0) return NULL;
return desc->fields + rv;
}
const ProtobufCMethodDescriptor *
protobuf_c_service_descriptor_get_method_by_name(
const ProtobufCServiceDescriptor *desc, const char *name) {
unsigned start = 0;
unsigned count;
if (desc == NULL || desc->method_indices_by_name == NULL) return NULL;
count = desc->n_methods;
while (count > 1) {
unsigned mid = start + count / 2;
unsigned mid_index = desc->method_indices_by_name[mid];
const char *mid_name = desc->methods[mid_index].name;
int rv = strcmp(mid_name, name);
if (rv == 0) return desc->methods + desc->method_indices_by_name[mid];
if (rv < 0) {
count = start + count - (mid + 1);
start = mid + 1;
} else {
count = mid - start;
}
}
if (count == 0) return NULL;
if (strcmp(desc->methods[desc->method_indices_by_name[start]].name, name) ==
0)
return desc->methods + desc->method_indices_by_name[start];
return NULL;
}
src/common/protobuf-c.h
浏览文件 @ 035c72f2
......@@ -798,76 +798,6 @@ uint32_t protobuf_c_version_number(void);
*/
#define PROTOBUF_C_MIN_COMPILER_VERSION 1000000
/**
* Look up a `ProtobufCEnumValue` from a `ProtobufCEnumDescriptor` by name.
*
* \param desc
* The `ProtobufCEnumDescriptor` object.
* \param name
* The `name` field from the corresponding `ProtobufCEnumValue` object to
* match.
* \return
* A `ProtobufCEnumValue` object.
* \retval NULL
* If not found or if the optimize_for = CODE_SIZE option was set.
*/
PROTOBUF_C__API
const ProtobufCEnumValue *protobuf_c_enum_descriptor_get_value_by_name(
const ProtobufCEnumDescriptor *desc, const char *name);
/**
* Look up a `ProtobufCEnumValue` from a `ProtobufCEnumDescriptor` by numeric
* value.
*
* \param desc
* The `ProtobufCEnumDescriptor` object.
* \param value
* The `value` field from the corresponding `ProtobufCEnumValue` object to
* match.
*
* \return
* A `ProtobufCEnumValue` object.
* \retval NULL
* If not found.
*/
PROTOBUF_C__API
const ProtobufCEnumValue *protobuf_c_enum_descriptor_get_value(
const ProtobufCEnumDescriptor *desc, int value);
/**
* Look up a `ProtobufCFieldDescriptor` from a `ProtobufCMessageDescriptor` by
* the name of the field.
*
* \param desc
* The `ProtobufCMessageDescriptor` object.
* \param name
* The name of the field.
* \return
* A `ProtobufCFieldDescriptor` object.
* \retval NULL
* If not found or if the optimize_for = CODE_SIZE option was set.
*/
PROTOBUF_C__API
const ProtobufCFieldDescriptor *protobuf_c_message_descriptor_get_field_by_name(
const ProtobufCMessageDescriptor *desc, const char *name);
/**
* Look up a `ProtobufCFieldDescriptor` from a `ProtobufCMessageDescriptor` by
* the tag value of the field.
*
* \param desc
* The `ProtobufCMessageDescriptor` object.
* \param value
* The tag value of the field.
* \return
* A `ProtobufCFieldDescriptor` object.
* \retval NULL
* If not found.
*/
PROTOBUF_C__API
const ProtobufCFieldDescriptor *protobuf_c_message_descriptor_get_field(
const ProtobufCMessageDescriptor *desc, unsigned value);
/**
* Determine the number of bytes required to store the serialised message.
*
......@@ -947,33 +877,6 @@ PROTOBUF_C__API
void protobuf_c_message_init(const ProtobufCMessageDescriptor *descriptor,
void *message);
/**
* Free a service.
*
* \param service
* The service object to free.
*/
PROTOBUF_C__API
void protobuf_c_service_destroy(ProtobufCService *service);
/**
* Look up a `ProtobufCMethodDescriptor` by name.
*
* \param desc
* Service descriptor.
* \param name
* Name of the method.
*
* \return
* A `ProtobufCMethodDescriptor` object.
* \retval NULL
* If not found or if the optimize_for = CODE_SIZE option was set.
*/
PROTOBUF_C__API
const ProtobufCMethodDescriptor *
protobuf_c_service_descriptor_get_method_by_name(
const ProtobufCServiceDescriptor *desc, const char *name);
/**
* Initialise a `ProtobufCBufferSimple` object.
*/
......@@ -1011,18 +914,6 @@ PROTOBUF_C__API
void protobuf_c_buffer_simple_append(ProtobufCBuffer *buffer, size_t len,
const unsigned char *data);
PROTOBUF_C__API
void protobuf_c_service_generated_init(
ProtobufCService *service, const ProtobufCServiceDescriptor *descriptor,
ProtobufCServiceDestroy destroy);
PROTOBUF_C__API
void protobuf_c_service_invoke_internal(ProtobufCService *service,
unsigned method_index,
const ProtobufCMessage *input,
ProtobufCClosure closure,
void *closure_data);
/**@}*/
PROTOBUF_C__END_DECLS
......
src/common/type_define.h
浏览文件 @ 035c72f2
......@@ -16,7 +16,6 @@ limitations under the License. */
#include <map>
#include <string>
#include <unordered_set>
#include <vector>
#include "framework/attribute.h"
#include "framework/scope.h"
......@@ -40,13 +39,6 @@ using OpCreator = std::function<framework::OperatorBase<Dtype> *(
const framework::AttributeMap & /*attrs*/,
std::shared_ptr<framework::Scope> /*scope*/)>;
using GradOpMakerFN =
std::function<std::vector<std::unique_ptr<framework::OpDesc>>(
const framework::OpDesc &,
const std::unordered_set<std::string> & /*no_grad_set*/,
std::unordered_map<std::string, std::string> * /*grad_to_var*/,
const std::vector<framework::BlockDesc *> &grad_block)>;
using InferVarTypeFN = std::function<void(const framework::OpDesc & /*op_desc*/,
framework::BlockDesc * /*block*/)>;
......
src/common/types.h
浏览文件 @ 035c72f2
......@@ -16,7 +16,6 @@ limitations under the License. */
#include <string>
#include <unordered_map>
#include <utility>
namespace paddle_mobile {
enum class Precision : int { FP32 = 0 };
......@@ -80,6 +79,7 @@ static const std::string G_OP_TYPE_ELEMENTWISE_ADD = "elementwise_add";
static const std::string G_OP_TYPE_FUSION_CONV_ADD_RELU =
"fusion_conv_add_relu";
static const std::string G_OP_TYPE_FC = "fc";
static const std::string G_OP_TYPE_CONV_ADD = "conv_add";
static const std::string G_OP_TYPE_LRN = "lrn";
static const std::string G_OP_TYPE_MUL = "mul";
static const std::string G_OP_TYPE_MULTICLASS_NMS = "multiclass_nms";
......@@ -116,5 +116,6 @@ static std::unordered_map<
{G_OP_TYPE_PRIOR_BOX, {{"Image", "Input"}, {"Boxes", "Variances"}}},
{G_OP_TYPE_MULTICLASS_NMS, {{"BBoxes", "Scores"}, {"Out"}}},
{G_OP_TYPE_FC, {{"X", "Y", "Z"}, {"Out"}}},
{G_OP_TYPE_RESHAPE, {{"X"}, {"Out"}}}};
{G_OP_TYPE_RESHAPE, {{"X"}, {"Out"}}},
{G_OP_TYPE_DEPTHWISE_CONV, {{"Input"}, {"Output"}}}};
} // namespace paddle_mobile
src/common/variant.cpp 已删除 100644 → 0
浏览文件 @ ff0f7f7b
/* Copyright (c) 2018 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. */
src/common/variant.h
浏览文件 @ 035c72f2
......@@ -12,8 +12,7 @@ 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. */
#include <iostream>
#include "common/enforce.h"
#include "common/log.h"
#pragma once
......@@ -57,15 +56,11 @@ class RawData {
char data[size];
RawData() {}
RawData(const RawData &raw_data) { strcpy(data, raw_data.data); }
// void operator=(const RawData &raw_data){
// strcpy(data, raw_data.data);
// }
};
template <typename... Ts>
struct Variant {
Variant(const Variant &variant) {
// std::cout << " 赋值构造函数 " << std::endl;
type_id = variant.type_id;
data = variant.data;
}
......@@ -87,8 +82,7 @@ struct Variant {
if (type_id == typeid(T).hash_code()) {
return *const_cast<T *>(reinterpret_cast<const T *>(&data));
} else {
// std::cout << " bad cast in variant " << std::endl;
throw std::bad_cast();
PADDLE_MOBILE_THROW_EXCEPTION(" bad cast in variant ");
}
}
......
src/framework/attribute.cpp
浏览文件 @ 035c72f2
......@@ -17,14 +17,8 @@ limitations under the License. */
namespace paddle_mobile {
namespace framework {
/*
* Variant<int, float, std::string, std::vector<int>, std::vector<float>,
std::vector<std::string>, bool, std::vector<bool>, BlockDesc *,
int64_t>
* */
struct PrintVistor : Vistor<Print &> {
PrintVistor(Print &printer) : printer_(printer) {}
explicit PrintVistor(Print &printer) : printer_(printer) {}
template <typename T>
Print &operator()(const T &value) {
printer_ << value;
......
src/framework/attribute.h
浏览文件 @ 035c72f2
......@@ -14,7 +14,10 @@ limitations under the License. */
#pragma once
#include <string>
#include <unordered_map>
#include <vector>
#include "common/enforce.h"
#include "common/log.h"
#include "common/variant.h"
......@@ -22,28 +25,15 @@ limitations under the License. */
namespace paddle_mobile {
namespace framework {
using std::string;
using std::vector;
class BlockDesc;
class Attribute {
public:
/*
* PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__INT = 0,
PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__FLOAT = 1,
PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__STRING = 2,
PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__INTS = 3,
PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__FLOATS = 4,
PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__STRINGS = 5,
PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__BOOLEAN = 6,
PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__BOOLEANS = 7,
PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__BLOCK = 8,
PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__LONG = 9
PROTOBUF_C__FORCE_ENUM_TO_BE_INT_SIZE(PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE)
*
* */
static Attribute GetAttrValue(
PaddleMobile__Framework__Proto__OpDesc__Attr *attr_desc) {
// std::cout << "begin get attr value" << std::endl;
Attribute attr;
switch (attr_desc->type) {
case PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__BOOLEAN: {
......@@ -63,35 +53,35 @@ class Attribute {
break;
}
case PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__BOOLEANS: {
std::vector<bool> val(attr_desc->n_bools);
vector<bool> val(attr_desc->n_bools);
for (int i = 0; i < attr_desc->n_bools; ++i) {
val[i] = attr_desc->bools[i];
}
attr.Set<std::vector<bool>>(val);
attr.Set<vector<bool>>(val);
break;
}
case PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__INTS: {
std::vector<int> val(attr_desc->n_ints);
vector<int> val(attr_desc->n_ints);
for (int i = 0; i < attr_desc->n_ints; ++i) {
val[i] = attr_desc->ints[i];
}
attr.Set<std::vector<int>>(val);
attr.Set<vector<int>>(val);
break;
}
case PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__FLOATS: {
std::vector<float> val(attr_desc->n_floats);
vector<float> val(attr_desc->n_floats);
for (int i = 0; i < attr_desc->n_floats; ++i) {
val[i] = attr_desc->floats[i];
}
attr.Set<std::vector<float>>(val);
attr.Set<vector<float>>(val);
break;
}
case PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__STRINGS: {
std::vector<std::string> val(attr_desc->n_strings);
vector<string> val(attr_desc->n_strings);
for (int i = 0; i < attr_desc->n_strings; ++i) {
val[i] = attr_desc->strings[i];
}
attr.Set<std::vector<std::string>>(val);
attr.Set<vector<string>>(val);
break;
}
case PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__LONG: {
......@@ -122,47 +112,41 @@ class Attribute {
return vistor(attr.variant_.Get<int>());
} else if (attr.variant_.TypeId() == typeid(float).hash_code()) {
return vistor(attr.variant_.Get<float>());
} else if (attr.variant_.TypeId() == typeid(std::string).hash_code()) {
return vistor(attr.variant_.Get<std::string>());
} else if (attr.variant_.TypeId() == typeid(std::vector<int>).hash_code()) {
return vistor(attr.variant_.Get<std::vector<int>>());
} else if (attr.variant_.TypeId() ==
typeid(std::vector<float>).hash_code()) {
return vistor(attr.variant_.Get<std::vector<float>>());
} else if (attr.variant_.TypeId() ==
typeid(std::vector<std::string>).hash_code()) {
return vistor(attr.variant_.Get<std::vector<std::string>>());
} else if (attr.variant_.TypeId() == typeid(string).hash_code()) {
return vistor(attr.variant_.Get<string>());
} else if (attr.variant_.TypeId() == typeid(vector<int>).hash_code()) {
return vistor(attr.variant_.Get<vector<int>>());
} else if (attr.variant_.TypeId() == typeid(vector<float>).hash_code()) {
return vistor(attr.variant_.Get<vector<float>>());
} else if (attr.variant_.TypeId() == typeid(vector<string>).hash_code()) {
return vistor(attr.variant_.Get<vector<string>>());
} else if (attr.variant_.TypeId() == typeid(bool).hash_code()) {
return vistor(attr.variant_.Get<bool>());
} else if (attr.variant_.TypeId() ==
typeid(std::vector<bool>).hash_code()) {
return vistor(attr.variant_.Get<std::vector<bool>>());
} else if (attr.variant_.TypeId() == typeid(vector<bool>).hash_code()) {
return vistor(attr.variant_.Get<vector<bool>>());
} else if (attr.variant_.TypeId() == typeid(int64_t).hash_code()) {
return vistor(attr.variant_.Get<int64_t>());
} else {
throw std::bad_exception();
PADDLE_MOBILE_THROW_EXCEPTION("type not support");
}
}
private:
Variant<int, float, std::string, std::vector<int>, std::vector<float>,
std::vector<std::string>, bool, std::vector<bool>, BlockDesc *,
int64_t>
Variant<int, float, string, vector<int>, vector<float>, vector<string>, bool,
vector<bool>, BlockDesc *, int64_t>
variant_;
};
using AttributeMap = std::unordered_map<std::string, Attribute>;
using AttributeMap = std::unordered_map<string, Attribute>;
class AttrReader {
public:
explicit AttrReader(const AttributeMap &attrs) : attrs_(attrs) {}
template <typename T>
inline T Get(const std::string &name) const {
// PADDLE_ENFORCE(attrs_.count(name) != 0, "%s should
// be in
// AttributeMap",
// name);
inline T Get(const string &name) const {
PADDLE_MOBILE_ENFORCE(attrs_.count(name) != 0,
"%s should be in AttributeMap", name);
return ((Attribute)attrs_.at(name)).Get<T>();
}
......
src/framework/data_layout.h
浏览文件 @ 035c72f2
......@@ -15,7 +15,6 @@ limitations under the License. */
#pragma once
#include <cctype>
#include <iostream>
#include <string>
namespace paddle_mobile {
......@@ -40,7 +39,7 @@ inline DataLayout StringToDataLayout(const std::string &str) {
} else if (s == "ANYLAYOUT") {
return DataLayout::kAnyLayout;
} else {
// std::cout << "Unknown storage order string: %s", s;
PADDLE_MOBILE_THROW_EXCEPTION("Unknown storage order string: %s", s.c_str())
}
}
......@@ -54,14 +53,8 @@ inline std::string DataLayoutToString(const DataLayout &data_layout) {
return "ANY_LAYOUT";
default:
break;
// std::cout << "unknown DataLayou %d", data_layout;
}
}
inline std::ostream &operator<<(std::ostream &out, const DataLayout &l) {
out << DataLayoutToString(l);
return out;
}
} // namespace framework
} // namespace paddle_mobile
src/framework/data_transform.cpp 已删除 100644 → 0
浏览文件 @ ff0f7f7b
/* Copyright (c) 2018 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. */
#pragma once
#include "framework/data_transform.h"
namespace paddle_mobile {
namespace framework {
static void PassTensorData(Tensor *from, Tensor *to) {
to->ShareDataWith(*from);
*from = Tensor();
}
void DataTransform(const OpKernelType &expected_kernel_type,
const OpKernelType &kernel_type_for_var,
const Tensor &input_tensor, Tensor *output_tensor) {
bool transformed = false;
Tensor in;
in.ShareDataWith(input_tensor);
Tensor out;
// // do layout transform
// if (NeedTransformLayout(expected_kernel_type.data_layout_,
// kernel_type_for_var.data_layout_)) {
// TransDataLayout(kernel_type_for_var, expected_kernel_type, in,
// &out);
// transformed = true;
// PassTensorData(&out, &in);
// }
//
// // do data type transform
// if (expected_kernel_type.data_type_ !=
// kernel_type_for_var.data_type_) {
// TransDataType(kernel_type_for_var, expected_kernel_type, in,
// &out);
// transformed = true;
// PassTensorData(&out, &in);
// }
//
// // do device transform
// if (!platform::is_same_place(kernel_type_for_var.place_,
// expected_kernel_type.place_)) {
// TransDataDevice(in, expected_kernel_type.place_, &out);
// transformed = true;
// PassTensorData(&out, &in);
// }
//
// PADDLE_ENFORCE(transformed, "No transform is applied, please
// check!");
// get output data
output_tensor->ShareDataWith(in);
}
void CopyVariableWithTensor(const Variable &in_var, const Tensor &tensor,
Variable *out_var) {
// if (in_var.IsType<LoDTensor>()) {
// auto& in_lod_tensor = in_var.Get<LoDTensor>();
// auto* tran_lod_tensor = out_var.GetMutable<LoDTensor>();
// tran_lod_tensor->set_lod(in_lod_tensor.lod());
// tran_lod_tensor->set_layout(in_lod_tensor.layout());
// tran_lod_tensor->ShareDataWith(tensor);
// } else if (in_var.IsType<SelectedRows>()) {
// auto& in_selected_rows = in_var.Get<SelectedRows>();
// auto* trans_selected_rows =
// out_var.GetMutable<SelectedRows>();
// trans_selected_rows->set_height(in_selected_rows.height());
// trans_selected_rows->set_rows(in_selected_rows.rows());
// trans_selected_rows->mutable_value()->ShareDataWith(tensor);
// } else {
// PADDLE_THROW("unknown var type");
// }
}
} // namespace framework
} // namespace paddle_mobile
src/framework/data_type.h 已删除 100644 → 0
浏览文件 @ ff0f7f7b
/* Copyright (c) 2018 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. */
#pragma once
namespace paddle_mobile {
namespace framework {
// inline proto::VarType::Type ToDataType(std::type_index type) {
// using namespace paddle_mobile::framework::proto;
// if (typeid(float).hash_code() == type.hash_code()) {
// return proto::VarType::FP32;
// } else if (typeid(double).hash_code() == type.hash_code()) {
// return proto::VarType::FP64;
// } else if (typeid(int).hash_code() == type.hash_code()) {
// return proto::VarType::INT32;
// } else if (typeid(int64_t).hash_code() == type.hash_code()) {
// return proto::VarType::INT64;
// } else if (typeid(bool).hash_code() == type.hash_code()) {
// return proto::VarType::BOOL;
// } else {
//// PADDLE_THROW("Not supported");
// }
// }
} // namespace framework
} // namespace paddle_mobile
src/framework/ddim.cpp
浏览文件 @ 035c72f2
......@@ -63,9 +63,6 @@ void make_ddim(DDim &ddim, const int64_t *dims, int n) {
ddim = make_dim<9>(dims);
break;
default:
// std::cout << "Dynamic dimensions must have between [1,
// 9]
// dimensions.";
break;
}
}
......@@ -133,9 +130,6 @@ int64_t DDim::operator[](int idx) const {
int DDim::size() const { return arity(*this); }
bool DDim::operator==(DDim d) const {
// if (var.which() != d.getVar().which()) {
// return false;
// } else {
std::vector<int64_t> v1 = vectorize(*this);
std::vector<int64_t> v2 = vectorize(d);
......@@ -157,7 +151,7 @@ DDim DDim::operator+(DDim d) const {
std::vector<int64_t> v3;
assert(v1.size() == v2.size());
PADDLE_MOBILE_ENFORCE(v1.size() == v2.size(), "v1.size() != v2.size()");
for (unsigned int i = 0; i < v1.size(); i++) {
v3.push_back(v1[i] + v2[i]);
......@@ -172,7 +166,7 @@ DDim DDim::operator*(DDim d) const {
std::vector<int64_t> v3;
assert(v1.size() == v2.size());
PADDLE_MOBILE_ENFORCE(v1.size() == v2.size(), "v1.size() == v2.size()");
for (unsigned int i = 0; i < v1.size(); i++) {
v3.push_back(v1[i] * v2[i]);
......@@ -183,7 +177,7 @@ DDim DDim::operator*(DDim d) const {
int64_t get(const DDim &ddim, int idx) { return ddim[idx]; }
void set(DDim &ddim, int idx, int value) { ddim[idx] = value; }
void set(DDim *ddim, int idx, int value) { (*ddim)[idx] = value; }
/// @cond HIDDEN
struct VectorizeVisitor : Vistor<void> {
......@@ -235,13 +229,10 @@ struct SliceVectorizeVisitor : Vistor<void> {
SliceVectorizeVisitor(std::vector<int64_t> &v, int b, int e)
: vector(v), begin(b), end(e) {
// PADDLE_ENFORCE(begin < end,
// "Begin index must be less than end index in
// ddim
// slice.");
// PADDLE_ENFORCE(begin >= 0,
// "Begin index can't be less than zero in
// ddim slice.");
PADDLE_MOBILE_ENFORCE(
begin < end, "Begin index must be less than end index in ddim slice.");
PADDLE_MOBILE_ENFORCE(begin >= 0,
"Begin index can't be less than zero in ddim slice.");
}
template <int S>
......@@ -267,9 +258,7 @@ DDim slice_ddim(const DDim &ddim, int begin, int end) {
std::vector<int64_t> vec;
vec.reserve(end - begin);
SliceVectorizeVisitor visitor(vec, begin, end);
// boost::apply_visitor(visitor, dim);
DDim::ApplyVistor(visitor, ddim);
// visitor(ddim.var.Get<Dim<4>>());
return make_ddim(vec);
}
......@@ -287,31 +276,19 @@ struct ArityVisitor : Vistor<int> {
int arity(const DDim &d) {
ArityVisitor arityVisitor = ArityVisitor();
return DDim::ApplyVistor(arityVisitor, d);
// return arityVisitor(d.var.Get<Dim<4>>());
// return boost::apply_visitor(ArityVisitor(), d); }
}
/// \cond HIDDEN
/// \endcond
struct OSVistor : Vistor<std::ostream &> {
OSVistor(std::ostream &os) : os_(os) {}
template <int D>
std::ostream &operator()(Dim<D> dim) const {
return os_ << dim;
#ifdef PADDLE_MOBILE_DEBUG
Print &operator<<(Print &printer, const DDim &ddim) {
for (int j = 0; j < ddim.size(); ++j) {
printer << ddim[j] << " ";
}
private:
std::ostream &os_;
};
std::ostream &operator<<(std::ostream &os, const DDim &ddim) {
auto vistor = OSVistor(os);
DDim::ApplyVistor(vistor, ddim);
return os;
return printer;
}
#endif
DDim::DDim(std::initializer_list<int64_t> init_list) {
*this = make_ddim(init_list);
}
......
src/framework/ddim.h
浏览文件 @ 035c72f2
......@@ -14,10 +14,9 @@ limitations under the License. */
#pragma once
#include <assert.h>
#include <initializer_list>
#include <stdexcept>
#include <vector>
#include "common/enforce.h"
#include "common/variant.h"
#include "dim.h"
......@@ -58,9 +57,7 @@ struct DDim {
} else if (d.var.TypeId() == typeid(Dim<9>).hash_code()) {
return vistor(d.var.Get<Dim<9>>());
} else {
printf(" dim not support \n");
throw std::bad_exception();
// return typename Vistor::type_t();
DLOG << " dim not support";
}
}
......@@ -83,17 +80,6 @@ struct DDim {
int64_t operator[](int idx) const;
// template <typename Visitor>
// typename Visitor::result_type apply_visitor(Visitor& visitor) {
// return var.apply_visitor(visitor);
// }
//
// template <typename Visitor>
// typename Visitor::result_type apply_visitor(Visitor& visitor)
// const {
// return var.apply_visitor(visitor);
// }
DDimVar getVar() { return var; }
bool operator==(DDim d) const;
......@@ -126,7 +112,7 @@ DDim make_ddim(std::initializer_list<int64_t> dims);
int64_t get(const DDim &dim, int idx);
void set(DDim &dim, int idx, int val);
void set(DDim *dim, int idx, int val);
std::vector<int64_t> vectorize(const DDim &ddim);
......@@ -151,8 +137,6 @@ DDim slice_ddim(const DDim &dim, int begin, int end);
int arity(const DDim &ddim);
std::ostream &operator<<(std::ostream &, const DDim &);
// Reshape a tensor to a matrix. The matrix's first dimension(column
// length)
// will be the product of tensor's first `num_col_dims` dimensions.
......@@ -163,5 +147,9 @@ DDim flatten_to_1d(const DDim &src);
DDim stride(const DDim &ddim);
DDim stride_numel(const DDim &ddim);
#ifdef PADDLE_MOBILE_DEBUG
Print &operator<<(Print &printer, const DDim &ddim);
#endif
} // namespace framework
} // namespace paddle_mobile
src/framework/dim.h
浏览文件 @ 035c72f2
......@@ -14,13 +14,7 @@ limitations under the License. */
#pragma once
#include <iostream>
#include <sstream>
#include <stdexcept>
#include <type_traits>
#include "platform/hostdevice.h"
#include "common/enforce.h"
namespace paddle_mobile {
namespace framework {
......@@ -30,42 +24,35 @@ struct Dim {
static constexpr int dimensions = i;
template <typename... Args>
HOSTDEVICE Dim(int64_t _head, Args... _tail) : head(_head), tail(_tail...) {
Dim(int64_t _head, Args... _tail) : head(_head), tail(_tail...) {
static_assert(sizeof...(_tail) == i - 1,
"Dim initialized with the wrong number of parameters");
}
HOSTDEVICE
Dim(int64_t _head, const Dim<i - 1> &_tail) : head(_head), tail(_tail) {}
HOSTDEVICE
Dim() : head(0), tail() {}
/** Construct a Dim from a linear index and size. Uses Fortran
* order
* indexing. */
HOSTDEVICE
Dim(int64_t idx, const Dim<i> &size)
: head(idx % size.head), tail(idx / size.head, size.tail) {}
/** Construct a Dim with each dimension set to the given index */
HOSTDEVICE
Dim(int64_t idx) : head(idx), tail(idx) {}
HOSTDEVICE
bool operator==(const Dim<i> &o) const {
return (head == o.head) && (tail == o.tail);
}
HOSTDEVICE
bool operator!=(const Dim<i> &o) const { return !(*this == o); }
HOSTDEVICE
int64_t &operator[](int idx);
HOSTDEVICE
int64_t operator[](int idx) const;
HOST std::string to_string() const;
std::string to_string() const;
int64_t head;
Dim<i - 1> tail;
......@@ -76,32 +63,22 @@ template <>
struct Dim<0> {
static constexpr int dimensions = 0;
HOSTDEVICE
Dim(int64_t _head) {}
HOSTDEVICE
Dim() {}
HOSTDEVICE
Dim(int idx, const Dim<0> &size) {
#ifndef __CUDA_ARCH__
if (idx > 0) {
throw std::invalid_argument("Index out of range.");
PADDLE_MOBILE_THROW_EXCEPTION("Index out of range.")
}
#else
PADDLE_ASSERT(idx == 0);
#endif
}
HOSTDEVICE
bool operator==(const Dim<0> &o) const { return true; }
HOSTDEVICE
bool operator!=(const Dim<0> &o) const { return false; }
HOSTDEVICE
int64_t &operator[](int idx);
HOSTDEVICE
int64_t operator[](int idx) const;
};
......@@ -112,12 +89,12 @@ template <int i>
struct DimGetter {
// Return a copy if Dim is const
template <typename D>
HOSTDEVICE static int64_t impl(const D &d) {
static int64_t impl(const D &d) {
return DimGetter<i - 1>::impl(d.tail);
}
// Return a reference if Dim is mutable
template <typename D>
HOSTDEVICE static int64_t &impl(D &d) {
static int64_t &impl(D &d) {
return DimGetter<i - 1>::impl(d.tail);
}
};
......@@ -127,25 +104,22 @@ template <>
struct DimGetter<0> {
// Return a copy if Dim is const
template <typename D>
HOSTDEVICE static int64_t impl(const D &d) {
static int64_t impl(const D &d) {
return d.head;
}
// Return a reference if Dim is mutable
template <typename D>
HOSTDEVICE static int64_t &impl(D &d) {
static int64_t &impl(D &d) {
return d.head;
}
};
template <int D>
HOSTDEVICE int64_t &indexer(Dim<D> &dim, int idx) {
#ifndef __CUDA_ARCH__
int64_t &indexer(Dim<D> &dim, int idx) {
if (idx < 0) {
throw std::invalid_argument("Tried to access a negative dimension");
PADDLE_MOBILE_THROW_EXCEPTION("Tried to access a negative dimension")
}
#else
PADDLE_ASSERT(idx >= 0);
#endif
if (idx == 0) {
return dim.head;
}
......@@ -153,31 +127,15 @@ HOSTDEVICE int64_t &indexer(Dim<D> &dim, int idx) {
}
template <>
HOSTDEVICE int64_t &indexer<0>(Dim<0> &dim, int idx) {
#ifndef __CUDA_ARCH__
throw std::invalid_argument("Invalid index");
#else
PADDLE_ASSERT(false);
#if CUDA_VERSION < 8000
// On CUDA versions previous to 8.0, only __shared__ variables
// could be declared as static in the device code.
int64_t head = 0;
#else
static int64_t head = 0;
#endif
return head;
#endif
int64_t &indexer<0>(Dim<0> &dim, int idx) {
PADDLE_MOBILE_THROW_EXCEPTION("Invalid index")
}
template <int D>
HOSTDEVICE int64_t indexer(const Dim<D> &dim, int idx) {
#ifndef __CUDA_ARCH__
int64_t indexer(const Dim<D> &dim, int idx) {
if (idx < 0) {
throw std::invalid_argument("Tried to access a negative dimension");
PADDLE_MOBILE_THROW_EXCEPTION("Tried to access a negative dimension")
}
#else
PADDLE_ASSERT(idx >= 0);
#endif
if (idx == 0) {
return dim.head;
}
......@@ -185,102 +143,84 @@ HOSTDEVICE int64_t indexer(const Dim<D> &dim, int idx) {
}
template <>
HOSTDEVICE int64_t indexer<0>(const Dim<0> &dim, int idx) {
#ifndef __CUDA_ARCH__
throw std::invalid_argument("Invalid index");
#else
PADDLE_ASSERT(false);
#if CUDA_VERSION < 8000
// On CUDA versions previous to 8.0, only __shared__ variables
// could be declared as static in the device code.
int64_t head = 0;
#else
static int64_t head = 0;
#endif
return head;
#endif
int64_t indexer<0>(const Dim<0> &dim, int idx) {
PADDLE_MOBILE_THROW_EXCEPTION("Invalid index")
}
} // namespace
// Static access to constant Dim
template <int i, int l>
HOSTDEVICE int64_t get(const Dim<l> &d) {
int64_t get(const Dim<l> &d) {
return DimGetter<i>::impl(d);
}
// Static access to mutable Dim
template <int i, int l>
HOSTDEVICE int64_t &get(Dim<l> &d) {
int64_t &get(Dim<l> &d) {
return DimGetter<i>::impl(d);
}
// Dynamic access to constant Dim
template <int l>
HOSTDEVICE int64_t Dim<l>::operator[](int i) const {
int64_t Dim<l>::operator[](int i) const {
// std::cout << "l: " << l << std::endl;
return indexer(*this, i);
}
// Dynamic access to mutable Dim
template <int l>
HOSTDEVICE int64_t &Dim<l>::operator[](int i) {
int64_t &Dim<l>::operator[](int i) {
return indexer(*this, i);
}
// Dynamic access to constant Dim
inline HOSTDEVICE int64_t Dim<0>::operator[](int i) const {
return indexer(*this, i);
}
inline int64_t Dim<0>::operator[](int i) const { return indexer(*this, i); }
// Dynamic access to mutable Dim
inline HOSTDEVICE int64_t &Dim<0>::operator[](int i) {
return indexer(*this, i);
}
inline int64_t &Dim<0>::operator[](int i) { return indexer(*this, i); }
// Dynamic access to constant Dim
// without std::enable_if will try to instantiate this on get<0>(d)
template <int l>
HOSTDEVICE typename std::enable_if<(l > 0), int64_t>::type get(const Dim<l> &d,
int i) {
typename std::enable_if<(l > 0), int64_t>::type get(const Dim<l> &d, int i) {
return d[i];
}
// Dynamic access to mutable Dim
template <int l>
HOSTDEVICE typename std::enable_if<(l > 0), int64_t &>::type get(Dim<l> &d,
int i) {
typename std::enable_if<(l > 0), int64_t &>::type get(Dim<l> &d, int i) {
return d[i];
}
// Dot product of two dims
template <int i>
HOSTDEVICE int64_t linearize(const Dim<i> &a, const Dim<i> &b) {
int64_t linearize(const Dim<i> &a, const Dim<i> &b) {
return a.head * b.head + linearize(a.tail, b.tail);
}
// Base case dot product of two Dims
// Notice it is inline because it is no longer a template
template <>
HOSTDEVICE inline int64_t linearize(const Dim<0> &a, const Dim<0> &b) {
inline int64_t linearize(const Dim<0> &a, const Dim<0> &b) {
return 0;
}
// Product of a Dim
template <int i>
HOSTDEVICE int64_t product(const Dim<i> &a, int prod = 1) {
int64_t product(const Dim<i> &a, int prod = 1) {
return prod * a.head * product(a.tail);
}
// Base case product of a Dim
// Notice it is inline because it is no longer a template
template <>
HOSTDEVICE inline int64_t product(const Dim<0> &a, int prod) {
inline int64_t product(const Dim<0> &a, int prod) {
return prod;
}
// Is 0 <= idx_i < size_i for all i?
template <int i>
HOSTDEVICE bool contained(const Dim<i> &idx, const Dim<i> &size) {
bool contained(const Dim<i> &idx, const Dim<i> &size) {
return ((0 <= idx.head) && (idx.head < size.head) &&
contained(idx.tail, size.tail));
}
......@@ -288,7 +228,7 @@ HOSTDEVICE bool contained(const Dim<i> &idx, const Dim<i> &size) {
// Base case of is 0 <= idx_i < size_i ?
// Notice it is inline because it is no longer a template
template <>
HOSTDEVICE inline bool contained(const Dim<0> &idx, const Dim<0> &size) {
inline bool contained(const Dim<0> &idx, const Dim<0> &size) {
return true;
}
......@@ -296,7 +236,7 @@ HOSTDEVICE inline bool contained(const Dim<0> &idx, const Dim<0> &size) {
* \brief Compute exclusive prefix-multiply of a Dim.
*/
template <int i>
HOSTDEVICE Dim<i> ex_prefix_mul(const Dim<i> &src, int mul = 1) {
Dim<i> ex_prefix_mul(const Dim<i> &src, int mul = 1) {
return Dim<i>(mul, ex_prefix_mul(src.tail, mul * src.head));
}
......@@ -304,7 +244,7 @@ HOSTDEVICE Dim<i> ex_prefix_mul(const Dim<i> &src, int mul = 1) {
// Base case of ex_prefix_mul
// 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) {
inline Dim<0> ex_prefix_mul(const Dim<0> &src, int mul) {
return Dim<0>();
}
///\endcond
......@@ -313,18 +253,18 @@ HOSTDEVICE inline Dim<0> ex_prefix_mul(const Dim<0> &src, int mul) {
* Add two dimensions together
*/
template <int i>
HOSTDEVICE Dim<i> dim_plus(const Dim<i> &a, const Dim<i> &b) {
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));
}
// Base case
template <>
HOSTDEVICE inline Dim<0> dim_plus(const Dim<0> &a, const Dim<0> &b) {
inline Dim<0> dim_plus(const Dim<0> &a, const Dim<0> &b) {
return Dim<0>();
}
template <int i>
HOSTDEVICE Dim<i> operator+(const Dim<i> &lhs, const Dim<i> &rhs) {
Dim<i> operator+(const Dim<i> &lhs, const Dim<i> &rhs) {
return dim_plus(lhs, rhs);
}
......@@ -332,18 +272,18 @@ HOSTDEVICE Dim<i> operator+(const Dim<i> &lhs, const Dim<i> &rhs) {
* Multiply two dimensions together
*/
template <int i>
HOSTDEVICE Dim<i> dim_mult(const Dim<i> &a, const Dim<i> &b) {
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));
}
// Base case
template <>
HOSTDEVICE inline Dim<0> dim_mult(const Dim<0> &a, const Dim<0> &b) {
inline Dim<0> dim_mult(const Dim<0> &a, const Dim<0> &b) {
return Dim<0>();
}
template <int i>
HOSTDEVICE Dim<i> operator*(const Dim<i> &lhs, const Dim<i> &rhs) {
Dim<i> operator*(const Dim<i> &lhs, const Dim<i> &rhs) {
return dim_mult(lhs, rhs);
}
......@@ -358,7 +298,7 @@ HOSTDEVICE Dim<i> operator*(const Dim<i> &lhs, const Dim<i> &rhs) {
*/
template <int i>
HOSTDEVICE Dim<i> normalize_strides(const Dim<i> &size, const Dim<i> &stride) {
Dim<i> normalize_strides(const Dim<i> &size, const Dim<i> &stride) {
int norm_stride = size.head == 1 ? 0 : stride.head;
return Dim<i>(norm_stride, normalize_strides(size.tail, stride.tail));
}
......@@ -366,8 +306,7 @@ HOSTDEVICE Dim<i> normalize_strides(const Dim<i> &size, const Dim<i> &stride) {
///\cond HIDDEN
template <>
HOSTDEVICE inline Dim<0> normalize_strides(const Dim<0> &size,
const Dim<0> &stride) {
inline Dim<0> normalize_strides(const Dim<0> &size, const Dim<0> &stride) {
return Dim<0>();
}
......@@ -382,54 +321,9 @@ HOSTDEVICE inline Dim<0> normalize_strides(const Dim<0> &size,
*/
template <typename... Args>
HOSTDEVICE Dim<sizeof...(Args)> make_dim(Args... idxes) {
Dim<sizeof...(Args)> make_dim(Args... idxes) {
return Dim<sizeof...(Args)>(idxes...);
}
// Allows us to output a Dim
// XXX For some reason, overloading fails to resolve this correctly
template <int i>
typename std::enable_if<(i > 1), std::ostream &>::type operator<<(
std::ostream &os, const Dim<i> &d) {
os << d.head << ", " << d.tail;
return os;
}
// Base case that allows us to output a Dim
// XXX I wish this could be an overload instead of a template
template <int i>
typename std::enable_if<(i == 1), std::ostream &>::type operator<<(
std::ostream &os, const Dim<i> &d) {
os << d.head;
return os;
}
inline std::ostream &operator<<(std::ostream &os, const Dim<0> &d) {
return os;
}
template <int i>
HOST std::string Dim<i>::to_string() const {
std::stringstream stream;
stream << *this;
return stream.str();
}
template <int D>
HOSTDEVICE Dim<D> linear_to_dimension(int linear_index, Dim<D> extents) {
Dim<D> result;
for (int i = 0; i < D - 1; ++i) {
result[i] = linear_index % extents[i];
linear_index /= extents[i];
}
result[D - 1] = linear_index;
return result;
}
} // namespace framework
} // namespace paddle_mobile
src/framework/lod_tensor.cpp
浏览文件 @ 035c72f2
......@@ -13,72 +13,55 @@ See the License for the specific language governing permissions and
limitations under the License. */
#include "lod_tensor.h"
#include <stdint.h>
#include <string.h>
#include <algorithm>
#include <iterator>
namespace paddle_mobile {
namespace framework {
std::ostream &operator<<(std::ostream &os, const LoD &lod) {
os << "{";
for (auto &v : lod) {
os << "{";
bool is_first = true;
for (auto &i : v) {
if (is_first) {
os << i;
is_first = false;
} else {
os << ", " << i;
}
}
os << "}";
}
os << "}";
return os;
}
std::ostream &operator<<(std::ostream &os, const LoDTensor &t) {
// PADDLE_ENFORCE(t.type().hash_code() ==
// typeid(float).hash_code());
// if (!platform::is_cpu_place(t.place())) {
// LoDTensor tt;
// framework::TensorCopy(t, platform::CPUPlace(), &tt);
// platform::DeviceContextPool &pool =
// platform::DeviceContextPool::Instance(); auto &dev_ctx =
// *pool.Get(t.place()); dev_ctx.Wait();
//
// os << tt;
// return os;
// }
os << "dim: " << t.dims() << "\n";
os << "lod: " << t.lod() << "\n";
// only print first ten elements
int64_t size = t.numel() < 10 ? t.numel() : 10;
for (int64_t i = 0; i < size; ++i) {
os << t.data<float>()[i] << " ";
}
return os;
}
std::string LoDToString(const LoD &lod) {
std::ostringstream stream;
stream << lod;
return stream.str();
}
// std::ostream &operator<<(std::ostream &os, const LoD &lod) {
// os << "{";
// for (auto &v : lod) {
// os << "{";
// bool is_first = true;
// for (auto &i : v) {
// if (is_first) {
// os << i;
// is_first = false;
// } else {
// os << ", " << i;
// }
// }
// os << "}";
// }
// os << "}";
//
// return os;
//}
//
// std::ostream &operator<<(std::ostream &os, const LoDTensor &t) {
// PADDLE_MOBILE_ENFORCE(t.type().hash_code() == typeid(float).hash_code(),
// "t.type() is not float");
// os << "dim: " << t.dims() << "\n";
// os << "lod: " << t.lod() << "\n";
// // only print first ten elements
// int64_t size = t.numel() < 10 ? t.numel() : 10;
// for (int64_t i = 0; i < size; ++i) {
// os << t.data<float>()[i] << " ";
// }
//
// return os;
//}
// std::string LoDToString(const LoD &lod) {
// std::ostringstream stream;
// stream << lod;
// return stream.str();
//}
LoD SliceInLevel(const LoD &in, size_t level, size_t elem_begin,
size_t elem_end) {
// PADDLE_ENFORCE_LT(level, in.size());
// PADDLE_ENFORCE_LT(elem_end, in[level].size());
PADDLE_MOBILE_ENFORCE(level < in.size(), "level should >= in.size()");
PADDLE_MOBILE_ENFORCE(elem_end < in[level].size(),
"elem_end >= in[level].size()");
LoD res;
res.resize(in.size() - level);
// copy the first level
......@@ -152,7 +135,7 @@ bool CheckLoD(const LoD &in, int tensor_height) {
if (a < b) return true;
return false;
})) {
std::cout << "ascending error";
PADDLE_MOBILE_THROW_EXCEPTION("ascending error")
return false;
}
}
......@@ -211,8 +194,9 @@ LoDAndOffset GetSubLoDAndAbsoluteOffset(const LoD &lod, size_t start_idx,
LoD sub_lod;
for (size_t level_idx = start_level; level_idx < lod.size(); ++level_idx) {
// PADDLE_ENFORCE_LE(start_idx, end_idx);
// PADDLE_ENFORCE_LT(end_idx, lod[level_idx].size());
PADDLE_MOBILE_ENFORCE(start_idx <= end_idx, "start_idx > end_idx");
PADDLE_MOBILE_ENFORCE(end_idx < lod[level_idx].size(),
"end_idx >= lod[level_idx].size()");
std::vector<size_t> level_lens;
for (size_t i = start_idx; i < end_idx; ++i) {
level_lens.push_back(lod[level_idx][i + 1] - lod[level_idx][i]);
......@@ -226,10 +210,9 @@ LoDAndOffset GetSubLoDAndAbsoluteOffset(const LoD &lod, size_t start_idx,
}
void AppendLoD(LoD *lod, const LoD &lod_length) {
// PADDLE_ENFORCE(
// lod->empty() || lod->size() == lod_length.size(),
// "The lod_length should has the same size with the appended
// lod.");
PADDLE_MOBILE_ENFORCE(
lod->empty() || lod->size() == lod_length.size(),
"The lod_length should has the same size with the appended lod.");
if (lod->empty()) {
for (size_t i = 0; i < lod_length.size(); ++i) {
lod->emplace_back(1, 0); // size = 1, value = 0;
......
src/framework/lod_tensor.h
浏览文件 @ 035c72f2
......@@ -16,7 +16,6 @@ limitations under the License. */
#include <memory>
#include <string>
#include <utility>
#include <vector>
#include "tensor.h"
#include "tensor_util.h"
......
src/framework/op_info.h
浏览文件 @ 035c72f2
......@@ -25,9 +25,8 @@ template <typename Dtype>
struct OpInfo {
OpCreator<Dtype> creator_;
const OpCreator<Dtype> &Creator() const {
// PADDLE_ENFORCE_NOT_NULL(creator_,
// "Operator Creator has not been
// registered");
PADDLE_MOBILE_ENFORCE(creator_ != nullptr,
"Operator Creator has not been registered");
return creator_;
}
};
......@@ -48,17 +47,15 @@ class OpInfoMap {
}
void Insert(const std::string &type, const OpInfo<Dtype> &info) {
// PADDLE_ENFORCE(!Has(type), "Operator %s has been
// registered", type);
PADDLE_MOBILE_ENFORCE(!Has(type), "Operator %s has been registered",
type.c_str());
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);
PADDLE_MOBILE_ENFORCE(op_info_ptr != nullptr,
"Operator %s has not been registered", type.c_str());
return *op_info_ptr;
}
......
src/framework/operator.h
浏览文件 @ 035c72f2
......@@ -16,7 +16,6 @@ limitations under the License. */
#include <map>
#include <string>
#include <utility>
#include <vector>
#include "common/enforce.h"
......@@ -27,7 +26,6 @@ limitations under the License. */
#include "framework/op_info.h"
#include "framework/op_kernel_type.h"
#include "framework/op_registry.h"
#include "framework/paddle_mobile_object.h"
#include "framework/program/block_desc.h"
#include "framework/program/program-optimize/node.h"
#include "framework/scope.h"
......@@ -52,7 +50,7 @@ static T *GetVarValue(const string &key, const VariableNameMap &var_map,
}
template <typename Dtype>
class OperatorBase : PaddleMobileObject {
class OperatorBase {
public:
/*
* @b op 基类的实例化方法, op 获取到了 输入、参数以及提前分配好的输出 tensor
......@@ -121,7 +119,7 @@ class OperatorWithKernel : public OperatorBase<Dtype> {
* @b 所有kernel的父类
* */
template <typename Dtype, typename P>
class OpKernelBase : PaddleMobileObject {
class OpKernelBase {
public:
/*
* @b 所有kernel 需实现 Compute 方法
......@@ -139,14 +137,16 @@ class OpKernelBase : PaddleMobileObject {
std::shared_ptr<::paddle_mobile::framework::Scope> scope) \
: parent_cls<Dtype, T>(type, inputs, outputs, attrs, scope) {}
class FusionOpMatcher : PaddleMobileObject {
class FusionOpMatcher {
public:
FusionOpMatcher() {}
virtual std::string Type() = 0;
virtual void FolderNodes(Node *node) {
node->Folder(node_.Depth(), Type(), {});
virtual void FolderNodes(
Node *node,
std::vector<std::shared_ptr<framework::Node>> *removed_nodes) {
node->Folder(node_.Depth(), Type(), {}, removed_nodes);
}
virtual Node &BeginNode() { return node_; }
......
src/framework/paddle_mobile_object.cpp 已删除 100644 → 0
浏览文件 @ ff0f7f7b
/* Copyright (c) 2018 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. */
#include "paddle_mobile_object.h"
src/framework/program/block_desc.cpp
浏览文件 @ 035c72f2
......@@ -17,13 +17,7 @@ limitations under the License. */
namespace paddle_mobile {
namespace framework {
std::vector<std::shared_ptr<VarDesc>> BlockDesc::Vars() const {
std::vector<std::shared_ptr<VarDesc>> res;
for (const auto &p : vars_) {
res.push_back(p.second);
}
return res;
}
std::vector<std::shared_ptr<VarDesc>> BlockDesc::Vars() const { return vars_; }
std::vector<std::shared_ptr<OpDesc>> BlockDesc::Ops() const { return ops_; }
......@@ -31,10 +25,14 @@ BlockDesc::BlockDesc(PaddleMobile__Framework__Proto__BlockDesc *desc)
: index_(desc->idx), parent_index_(desc->idx) {
for (int i = 0; i < desc->n_vars; ++i) {
PaddleMobile__Framework__Proto__VarDesc *var_desc = desc->vars[i];
vars_[std::string(var_desc->name)] =
std::shared_ptr<VarDesc>(new VarDesc(var_desc));
vars_.emplace_back(std::shared_ptr<VarDesc>(new VarDesc(var_desc)));
}
std::sort(vars_.begin(), vars_.end(),
[](std::shared_ptr<VarDesc> left, std::shared_ptr<VarDesc> right) {
return left->Name() < right->Name();
});
for (int j = 0; j < desc->n_ops; ++j) {
PaddleMobile__Framework__Proto__OpDesc *op_desc = desc->ops[j];
ops_.emplace_back(new framework::OpDesc(op_desc));
......
src/framework/program/block_desc.h
浏览文件 @ 035c72f2
......@@ -15,14 +15,13 @@ limitations under the License. */
#pragma once
#include "framework/framework.pb-c.h"
#include "framework/paddle_mobile_object.h"
#include "framework/program/op_desc.h"
#include "framework/program/var_desc.h"
namespace paddle_mobile {
namespace framework {
class BlockDesc : PaddleMobileObject {
class BlockDesc {
public:
friend class Node;
friend class ProgramOptimize;
......@@ -35,10 +34,9 @@ class BlockDesc : PaddleMobileObject {
ops_.push_back(copy_op_desc);
}
for (auto &var_desc : block_desc.vars_) {
std::shared_ptr<VarDesc> copy_var_desc =
std::make_shared<VarDesc>(*var_desc.second);
vars_[var_desc.first] = copy_var_desc;
for (int i = 0; i < block_desc.vars_.size(); ++i) {
auto &var_desc = block_desc.vars_[i];
vars_.emplace_back(std::make_shared<VarDesc>(*var_desc));
}
}
......@@ -64,7 +62,7 @@ class BlockDesc : PaddleMobileObject {
bool multi_thread_;
int parent_index_;
std::vector<std::shared_ptr<OpDesc>> ops_;
std::unordered_map<std::string, std::shared_ptr<VarDesc>> vars_;
std::vector<std::shared_ptr<VarDesc>> vars_;
};
} // namespace framework
......
src/framework/program/op_desc.h
浏览文件 @ 035c72f2
......@@ -20,12 +20,11 @@ limitations under the License. */
#include "common/log.h"
#include "common/type_define.h"
#include "framework/framework.pb-c.h"
#include "framework/paddle_mobile_object.h"
namespace paddle_mobile {
namespace framework {
class OpDesc : PaddleMobileObject {
class OpDesc {
public:
friend class ProgramOptimize;
friend class FusionOpMatcher;
......
src/framework/program/program-optimize/fusion_op_register.cpp 已删除 100644 → 0
浏览文件 @ ff0f7f7b
/* Copyright (c) 2018 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. */
#include "fusion_op_register.h"
src/framework/program/program-optimize/node.cpp
浏览文件 @ 035c72f2
......@@ -12,10 +12,8 @@ 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. */
#include <sstream>
#include "framework/operator.h"
#include "framework/program/program-optimize/node.h"
#include "framework/operator.h"
namespace paddle_mobile {
......@@ -45,54 +43,13 @@ bool Node::operator==(const Node &in) {
return true;
}
bool Node::CanSplit(std::unordered_set<std::string> complex_compute_set) {
bool split = false;
CanSplit(&split, false, 0, &complex_compute_set, this);
return split;
}
void Node::CanSplit(bool *split, bool spliting, int complex_count,
std::unordered_set<std::string> *complex_compute_set,
Node *pre_node) {
if (spliting) {
if (complex_compute_set->find(this->type_) != complex_compute_set->end()) {
complex_count++;
}
}
if (inputs_.size() > 1 && pre_node != inputs_.back()) {
return;
}
if (inputs_.size() > 1 && pre_node == inputs_.back()) {
if (complex_count > 1) {
*split = true;
return;
}
}
// multi output, to check
if (outputs_.size() > 1) {
spliting = true;
complex_compute_set = 0;
} else {
if (spliting == true && inputs_.size() > 0) {
spliting = false;
} else {
}
}
for (auto &output : outputs_) {
output->CanSplit(split, spliting, complex_count, complex_compute_set, this);
}
}
std::vector<std::shared_ptr<framework::OpDesc>> Node::OpDescs(uint size) {
std::vector<std::shared_ptr<framework::OpDesc>> Node::OpDescs(int size) {
std::vector<std::shared_ptr<framework::OpDesc>> op_descs;
OpDescs(size - 1, &op_descs);
return op_descs;
}
void Node::OpDescs(uint index,
void Node::OpDescs(int index,
std::vector<std::shared_ptr<framework::OpDesc>> *op_desc) {
if (index == 0) {
return;
......@@ -103,107 +60,6 @@ void Node::OpDescs(uint index,
}
}
void Node::OpDescs(std::vector<std::shared_ptr<framework::OpDesc>> *op_desc,
Node *node, bool adding_thread, int thread_num) {
if (outputs_.size() > 1) {
adding_thread = false;
}
bool can_add_split = false;
// 如果当前节点有多个输出 并且 只有当前节点对应的 op_desc_ 输出数为 1 时支持
if (outputs_.size() > 1 &&
op_input_output_key[op_desc_->type_].second.size() == 1) {
can_add_split = true;
// 遍历当前节点的 output 节点
for (const auto &output : outputs_) {
// 不支持 output 有多个 output 的情况
if (output->outputs_.size() > 0) {
can_add_split = false;
break;
}
//与节点关联的 OpDesc
std::shared_ptr<framework::OpDesc> &op_desc = output->op_desc_;
//获取这个 op 的 inputs key 和 outputs key
auto inputs_and_outputs = op_input_output_key[op_desc->type_];
//判断现在 是否存在这个 op
//判断这个 output 和 input key 的 size 等于 1
if (op_input_output_key.find(op_desc->type_) !=
op_input_output_key.end() &&
inputs_and_outputs.first.size() == 1 &&
inputs_and_outputs.second.size() == 1) {
auto inputs_of_output = op_desc->Input(inputs_and_outputs.first[0]);
auto outputs_of_output = op_desc->Output(inputs_and_outputs.second[0]);
// 判断一下, 如果输入和输出没有同名, 是支持的
for (int i = 0; i < inputs_of_output.size(); ++i) {
std::string input_of_output = inputs_of_output[i];
for (int j = 0; j < outputs_of_output.size(); ++j) {
std::string output_of_output = outputs_of_output[j];
if (input_of_output == output_of_output) {
DLOG << "output的 output 包含 input" << input_of_output;
can_add_split = false;
break;
}
}
}
} else { // 如果模型中包含没有的 op, 则不支持添加 split
DLOG << "找不到 这个 op 类型: " << output->op_desc_->type_;
can_add_split = false;
}
}
}
if (inputs_.size() > 1 && node != inputs_.back()) {
return;
} else if (inputs_.size() > 1 && node == inputs_.back()) {
adding_thread = false;
op_desc->push_back(this->op_desc_);
} else {
op_desc->push_back(this->op_desc_);
}
if (adding_thread) {
Attribute attr;
attr.Set<int>(thread_num);
this->op_desc_->attrs_["thread"] = attr;
}
if (can_add_split) {
adding_thread = true;
std::shared_ptr<OpDesc> split_op_desc = std::make_shared<OpDesc>();
split_op_desc->type_ = G_OP_TYPE_SPLIT;
auto outputs = this->op_desc_->Output(
op_input_output_key[this->op_desc_->Type()].second[0]);
split_op_desc->inputs_ = {
{op_input_output_key[G_OP_TYPE_SPLIT].first[0], outputs}};
auto &split_outputs =
split_op_desc->outputs_[op_input_output_key[G_OP_TYPE_SPLIT].second[0]];
for (const auto &output : outputs_) {
split_outputs.push_back(outputs[0]);
}
DLOG << "add split";
op_desc->push_back(split_op_desc);
}
for (int i = 0; i < outputs_.size(); ++i) {
auto &output = outputs_[i];
if (can_add_split) {
output->OpDescs(op_desc, this, adding_thread, i);
} else {
output->OpDescs(op_desc, this, adding_thread, thread_num);
}
}
}
std::vector<std::shared_ptr<framework::OpDesc>> Node::OpDescs() {
std::vector<std::shared_ptr<framework::OpDesc>> op_descs;
OpDescs(&op_descs, this, false, 0);
return op_descs;
}
std::shared_ptr<Node> Node::To(int size) {
std::shared_ptr<Node> node = std::make_shared<Node>();
this->To(size - 1, node);
......@@ -224,24 +80,25 @@ void Node::To(int index, std::shared_ptr<Node> node) {
}
}
uint Node::Depth(uint begin) {
uint depth = 0;
int Node::Depth(int begin) {
int depth = 0;
begin++;
for (int i = 0; i < outputs_.size(); ++i) {
uint output_depth = outputs_[i]->Depth(begin);
int output_depth = outputs_[i]->Depth(begin);
depth = output_depth > depth ? output_depth : depth;
}
return begin > depth ? begin : depth;
}
Node &Node::Folder(
uint size, std::string type,
std::map<std::string, std::pair<std::string, std::string>> change) {
int size, std::string type,
std::map<std::string, std::pair<std::string, std::string>> change,
std::vector<std::shared_ptr<Node>> *removed_nodes) {
std::shared_ptr<framework::OpDesc> op_desc =
std::make_shared<framework::OpDesc>();
op_desc->inputs_ = this->op_desc_->inputs_;
std::vector<std::shared_ptr<Node>> outputs;
this->Folder(op_desc, &outputs, size - 1, &change, this);
this->Folder(op_desc, &outputs, size - 1, &change, this, removed_nodes);
this->outputs_ = outputs;
this->type_ = type;
this->op_desc_ = op_desc;
......@@ -251,9 +108,9 @@ Node &Node::Folder(
void Node::Folder(
std::shared_ptr<framework::OpDesc> op_desc,
std::vector<std::shared_ptr<Node>> *outputs, uint index,
std::vector<std::shared_ptr<Node>> *outputs, int index,
std::map<std::string, std::pair<std::string, std::string>> *change,
Node *begin_node) {
Node *begin_node, std::vector<std::shared_ptr<Node>> *removed_nodes) {
if (change->find(this->type_) != change->end()) {
auto change_pair = (*change)[this->type_];
op_desc->GetInputs()[change_pair.second] =
......@@ -266,7 +123,9 @@ void Node::Folder(
if (index > 0) {
--index;
for (auto output : outputs_) {
output->Folder(op_desc, outputs, index, change, begin_node);
removed_nodes->push_back(output);
output->Folder(op_desc, outputs, index, change, begin_node,
removed_nodes);
}
} else {
for (auto &op_output : this->op_desc_->outputs_) {
......@@ -285,7 +144,7 @@ void Node::Folder(
}
}
}
#ifdef PADDLE_MOBILE_DEBUG
std::string Node::ToString(std::string blank, const Node *node) const {
std::stringstream ss;
ss << type_ << "-> \n";
......@@ -316,6 +175,7 @@ Print &operator<<(Print &printer, const Node &node) {
printer << node.ToString();
return printer;
}
#endif
} // namespace framework
} // namespace paddle_mobile
src/framework/program/program-optimize/node.h
浏览文件 @ 035c72f2
......@@ -14,20 +14,17 @@ limitations under the License. */
#pragma once
#include <cinttypes>
#include <map>
#include <string>
#include <unordered_set>
#include <utility>
#include <vector>
#include "common/log.h"
#include "framework/paddle_mobile_object.h"
#include "framework/program/op_desc.h"
namespace paddle_mobile {
namespace framework {
class Node : PaddleMobileObject {
class Node {
friend class ProgramOptimize;
public:
......@@ -37,35 +34,34 @@ class Node : PaddleMobileObject {
: op_desc_(op_desc), type_(op_desc->Type()) {}
Node &operator>(std::shared_ptr<Node> node);
bool operator==(const Node &in);
bool CanSplit(std::unordered_set<std::string> complex_compute_set);
#ifdef PADDLE_MOBILE_DEBUG
std::string ToString() const;
void Description();
#endif
std::shared_ptr<Node> To(int size);
uint Depth(uint begin = 0);
int Depth(int begin = 0);
Node &Folder(
uint size, std::string type,
std::map<std::string, std::pair<std::string, std::string>> change_map);
std::vector<std::shared_ptr<framework::OpDesc>> OpDescs(uint size);
std::vector<std::shared_ptr<framework::OpDesc>> OpDescs();
int size, std::string type,
std::map<std::string, std::pair<std::string, std::string>> change_map,
std::vector<std::shared_ptr<Node>> *removed_nodes);
std::vector<std::shared_ptr<framework::OpDesc>> OpDescs(int size);
std::shared_ptr<framework::OpDesc> OpDescOfNode() { return op_desc_; }
std::string Type() { return type_; }
void Description();
private:
void CanSplit(bool *split, bool spliting, int complex_count,
std::unordered_set<std::string> *complex_compute_set,
Node *pre_node);
void OpDescs(std::vector<std::shared_ptr<framework::OpDesc>> *op_desc,
Node *node, bool adding_thread, int thread_num);
void OpDescs(uint size,
void OpDescs(int size,
std::vector<std::shared_ptr<framework::OpDesc>> *op_desc);
void To(int index, std::shared_ptr<Node>);
void Folder(
std::shared_ptr<framework::OpDesc> op_desc,
std::vector<std::shared_ptr<Node>> *outputs, uint index,
std::vector<std::shared_ptr<Node>> *outputs, int index,
std::map<std::string, std::pair<std::string, std::string>> *change,
Node *begin_node);
Node *begin_node, std::vector<std::shared_ptr<Node>> *removed_nodes);
std::shared_ptr<framework::OpDesc> op_desc_;
#ifdef PADDLE_MOBILE_DEBUG
std::string ToString(std::string blank, const Node *node) const;
#endif
std::vector<std::shared_ptr<Node>> outputs_;
std::vector<Node *> inputs_;
std::string type_;
......
src/framework/program/program-optimize/program_optimize.cpp
浏览文件 @ 035c72f2
......@@ -31,6 +31,8 @@ std::shared_ptr<ProgramDesc> ProgramOptimize::FushionOptimize(
std::unordered_map<std::string, std::vector<std::shared_ptr<Node>>>
type_map;
std::vector<std::shared_ptr<Node>> nodes;
std::shared_ptr<Node> begin_node;
auto block = optimize_program->Block(i);
// DLOG << " ops size: " << block->Ops().size();
......@@ -38,11 +40,13 @@ std::shared_ptr<ProgramDesc> ProgramOptimize::FushionOptimize(
auto op = block->Ops()[j];
auto op_type = op->Type();
if (op_input_output_key.find(op->Type()) == op_input_output_key.end()) {
LOG(kLOG_ERROR) << "return null ";
LOG(kLOG_ERROR) << "has not support op return null "
<< " op type: " << op->Type();
return nullptr;
}
std::shared_ptr<Node> node = std::make_shared<Node>(op);
nodes.push_back(node);
//
type_map[op->Type()].push_back(node);
......@@ -87,21 +91,26 @@ std::shared_ptr<ProgramDesc> ProgramOptimize::FushionOptimize(
// DLOG << " match success " << " fusion node: \n" <<
// matcher->BeginNode() << "\nsub node: \n" << *sub_node;
// DLOG << "match node\n"<< *match_node;
matcher->FolderNodes(match_node.get());
// DLOG << " after match node\n"<< *match_node;
// match_node->Description();
// DLOG << "begin node: \n" << *begin_node;
std::vector<std::shared_ptr<Node>> removed_nodes;
matcher->FolderNodes(match_node.get(), &removed_nodes);
for (int j = 0; j < removed_nodes.size(); ++j) {
auto removed_node = removed_nodes[j];
auto removed_ite =
std::find(nodes.begin(), nodes.end(), removed_node);
nodes.erase(removed_ite);
}
}
}
}
// DLOG << "node: \n" << *begin_node;
std::vector<std::shared_ptr<framework::OpDesc>> op_descs;
// bool can_splite = begin_node->CanSplit({G_OP_TYPE_CONV,
// G_OP_TYPE_BATCHNORM, G_OP_TYPE_DEPTHWISE_CONV});
GenerateOps(&op_descs, begin_node.get());
for (int m = 0; m < nodes.size(); ++m) {
auto &node = nodes[m];
op_descs.push_back(node->op_desc_);
}
// GenerateOps(&op_descs, begin_node.get());
block->ops_ = op_descs;
}
......@@ -118,6 +127,14 @@ void ProgramOptimize::GenerateOps(
Node *current_node) {
if (current_node->inputs_.size() > 1 &&
input_node != current_node->inputs_.back()) {
DLOG << " current type " << current_node->type_;
DLOG << " inputs size of current node > 0 ";
for (int i = 0; i < current_node->inputs_.size(); ++i) {
DLOG << " input i: " << current_node->inputs_[i]->type_;
}
return;
} else if (current_node->inputs_.size() > 1 &&
input_node == current_node->inputs_.back()) {
......
src/framework/program/program.cpp 已删除 100644 → 0
浏览文件 @ ff0f7f7b
/* Copyright (c) 2018 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. */
namespace paddle_mobile {
namespace framework {}
} // namespace paddle_mobile
src/framework/program/program.h
浏览文件 @ 035c72f2
......@@ -15,7 +15,6 @@ limitations under the License. */
#pragma once
#include "common/types.h"
#include "framework/paddle_mobile_object.h"
#include "framework/program/program_desc.h"
#include "framework/scope.h"
......@@ -23,12 +22,14 @@ namespace paddle_mobile {
namespace framework {
template <typename Dtype, Precision P = Precision::FP32>
class Program : PaddleMobileObject {
class Program {
public:
std::shared_ptr<ProgramDesc> originProgram;
std::shared_ptr<ProgramDesc> optimizeProgram;
std::shared_ptr<Scope> scope;
std::string model_path;
std::string para_path;
bool is_commbine = false;
private:
};
......
src/framework/program/program_desc.h
浏览文件 @ 035c72f2
......@@ -18,13 +18,12 @@ limitations under the License. */
#include "common/types.h"
#include "framework/framework.pb-c.h"
#include "framework/paddle_mobile_object.h"
#include "framework/program/block_desc.h"
namespace paddle_mobile {
namespace framework {
class ProgramDesc : PaddleMobileObject {
class ProgramDesc {
public:
friend class Node;
friend class ProgramOptimize;
......
src/framework/program/var_desc.h
浏览文件 @ 035c72f2
......@@ -14,40 +14,14 @@ limitations under the License. */
#pragma once
#include <string>
#include "framework/framework.pb-c.h"
#include "framework/paddle_mobile_object.h"
#include "framework/program/tensor_desc.h"
namespace paddle_mobile {
namespace framework {
/*
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__BOOL = 0,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__INT16 = 1,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__INT32 = 2,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__INT64 = 3,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__FP16 = 4,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__FP32 = 5,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__FP64 = 6,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__LOD_TENSOR = 7,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__SELECTED_ROWS = 8,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__FEED_MINIBATCH = 9,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__FETCH_LIST = 10,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__STEP_SCOPES = 11,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__LOD_RANK_TABLE = 12,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__LOD_TENSOR_ARRAY = 13,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__PLACE_LIST = 14,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__READER = 15,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__CHANNEL = 16,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__RAW = 17,
PADDLE_MOBILE__FRAMEWORK__PROTO__VAR_TYPE__TYPE__TUPLE = 18
*/
class VarDesc {
public:
VarDesc(const VarDesc &var_desc) {
......@@ -56,14 +30,6 @@ class VarDesc {
this->persistable_ = var_desc.persistable_;
this->tensor_desc_ = var_desc.tensor_desc_;
this->type_ = var_desc.type_;
/*
*
* std::string name_;
bool persistable_;
TensorDesc tensor_desc_;
VarType_Type type_;
VarType_Type data_type_;
* */
}
VarDesc(PaddleMobile__Framework__Proto__VarDesc *desc) {
type_ = (VarType_Type)desc->type->type;
......@@ -102,39 +68,6 @@ class VarDesc {
const TensorDesc &Tensor_desc() const { return tensor_desc_; }
// const proto::VarType::ChannelDesc &channel_desc() const {
// switch (desc_.type().type()) {
// case proto::VarType::CHANNEL:
// return desc_.type().channel();
// default:
// break;
// }
// }
// proto::VarType::Type GetDataType() const {
// switch (desc_.type().type()) {
// case proto::VarType::CHANNEL:
// return channel_desc().data_type();
// break;
// default:
// return tensor_desc().data_type();
// }
// }
// template <typename T>
// std::vector<T> RepeatedToVector(
// const google::protobuf::RepeatedField<T> &repeated_field) const {
// std::vector<T> ret;
// ret.reserve(repeated_field.size());
// std::copy(repeated_field.begin(), repeated_field.end(),
// std::back_inserter(ret));
// return ret;
// }
// std::vector<int64_t> GetShape() const {
// return this->RepeatedToVector(tensor_desc().dims());
// }
private:
std::string name_;
bool persistable_;
......
src/framework/scope.cpp
浏览文件 @ 035c72f2
......@@ -22,7 +22,6 @@ namespace paddle_mobile {
namespace framework {
Scope &Scope::NewScope() const {
std::unique_lock<std::mutex> lock(mutex_);
kids_.push_back(new Scope(this));
return *kids_.back();
}
......@@ -72,11 +71,9 @@ std::vector<std::string> Scope::LocalVarNames() const {
}
void Scope::DeleteScope(Scope *scope) const {
std::unique_lock<std::mutex> lock(mutex_);
auto it = std::find(kids_.begin(), kids_.end(), scope);
kids_.erase(it);
delete scope;
// deferent
}
void Scope::EraseVars(const std::vector<std::string> &var_names) {
......@@ -104,14 +101,6 @@ void Scope::Rename(const std::string &origin_name,
vars_[new_name] = origin_it->second;
vars_.erase(origin_it);
}
//
// std::string Scope::Rename(const std::string& origin_name)
// const {
// auto var_name = string::Sprintf("%p.%d", this,
// vars_.size());
// Rename(origin_name, var_name);
// return var_name;
// }
Variable *Scope::FindVarLocally(const std::string &name) const {
auto it = vars_.find(name);
......
src/framework/scope.h
浏览文件 @ 035c72f2
......@@ -14,17 +14,16 @@ limitations under the License. */
#pragma once
#include <list> //std::list
#include <mutex> //std::mutex
#include <unordered_map> //std::unordered_map
#include <list>
#include <unordered_map>
#include "variable.h"
namespace paddle_mobile {
namespace framework {
class Scope {
public:
Scope() {}
~Scope() {}
Scope() = default;
~Scope() = default;
Scope &NewScope() const;
......@@ -70,8 +69,6 @@ class Scope {
mutable std::unordered_map<std::string, Variable *> vars_;
mutable std::list<Scope *> kids_;
Scope const *parent_{nullptr};
mutable std::mutex mutex_;
};
} // namespace framework
} // namespace paddle_mobile
src/framework/tensor.h
浏览文件 @ 035c72f2
......@@ -14,14 +14,15 @@ limitations under the License. */
#pragma once
#include <common/enforce.h>
#include <cstdint>
#include <cstring>
#include <memory>
#include <type_traits>
#include <typeindex>
#include <vector>
#include "common/enforce.h"
#include "common/enforce.h"
#include "framework/data_layout.h"
#include "framework/ddim.h"
#include "memory/t_malloc.h"
......
src/framework/tensor_util.cpp
浏览文件 @ 035c72f2
......@@ -13,137 +13,19 @@ See the License for the specific language governing permissions and
limitations under the License. */
#include "tensor_util.h"
#include <algorithm>
#include <limits>
#include <vector>
namespace paddle_mobile {
namespace framework {
void TensorCopy(const Tensor &src, Tensor *dst) {
// VLOG(3) << "TensorCopy " << src.dims() << " from " <<
// src.place() << " to
// "
// << dst_place;
src.check_memory_size();
dst->Resize(src.dims());
dst->set_layout(src.layout());
auto src_ptr = src.data<void>();
auto dst_ptr = dst->mutable_data(src.type());
auto size = src.numel() * SizeOfType(src.type());
memory::Copy(dst_ptr, src_ptr, size);
}
void TensorCopySync(const Tensor &src, Tensor *dst) {
// VLOG(3) << "TensorCopySync " << src.dims() << " from " <<
// src.place()
// << " to " << dst_place;
src.check_memory_size();
dst->Resize(src.dims());
dst->set_layout(src.layout());
auto src_ptr = src.data<void>();
auto dst_ptr = dst->mutable_data(src.type());
auto size = src.numel() * SizeOfType(src.type());
memory::Copy(dst_ptr, src_ptr, size);
}
template <typename Predicate>
struct AnyDTypeVisitor {
Predicate predicate_;
const Tensor &tensor_;
Tensor *out_;
AnyDTypeVisitor(Predicate predicate, const Tensor &tensor, Tensor *out)
: predicate_(predicate), tensor_(tensor), out_(out) {}
template <typename T>
void operator()() const {
// auto t = EigenVector<T>::Flatten(tensor_);
// auto o = EigenScalar<bool>::From(*out_);
// return any of predicate_(t) is true.
// o.device(*ctx_.eigen_device()) = predicate_(t).any();
}
};
template <typename Predicate>
inline void AnyImpl(Predicate predicate, const Tensor &tensor,
framework::Tensor *out) {
VisitDataType(ToDataType(tensor.type()),
AnyDTypeVisitor<Predicate>(predicate, tensor, out));
}
template <typename Predicate>
struct AnyVisitor {
const framework::Tensor &tensor_;
Predicate predicate_;
AnyVisitor(const framework::Tensor &tensor, Predicate predicate)
: tensor_(tensor), predicate_(std::move(predicate)) {}
bool operator()(void) const {
framework::Tensor out;
out.Resize({1});
out.mutable_data<bool>();
AnyImpl(predicate_, tensor_, &out);
return this->GetResult(out);
}
bool GetResult(const framework::Tensor &out) const {
return *out.data<bool>();
}
};
template <typename Predicate>
inline bool Any(const framework::Tensor &tensor, Predicate predicate) {
AnyVisitor<Predicate> visitor(tensor, predicate);
// return platform::VisitPlace(visitor);
return visitor();
}
struct ContainsNANPredicate {
template <typename T>
auto operator()(const T &eigen_vec) const
-> decltype(std::declval<T>().isnan()) {
// Cast eigen_vector to vector of bool. true if is inf.
return eigen_vec.isnan();
}
};
bool TensorContainsNAN(const framework::Tensor &tensor) {
ContainsNANPredicate predicate;
return Any(tensor, predicate);
}
struct ContainsInfPredicate {
template <typename T>
auto operator()(const T &eigen_vec) const
-> decltype(std::declval<T>().isinf()) {
// Cast eigen_vector to vector of bool. true if is inf.
return eigen_vec.isinf();
}
};
bool TensorContainsInf(const framework::Tensor &tensor) {
ContainsInfPredicate predicate;
return Any(tensor, predicate);
}
struct DeserializedDataFunctor {
DeserializedDataFunctor(void **buf, Tensor *tensor)
: buf_(buf), tensor_(tensor) {}
template <typename T>
void operator()() {
*buf_ = tensor_->mutable_data<T>();
}
void **buf_;
Tensor *tensor_;
};
} // namespace framework
} // namespace paddle_mobile
src/framework/tensor_util.h
浏览文件 @ 035c72f2
......@@ -15,51 +15,12 @@ limitations under the License. */
#pragma once
#include <vector>
#include "memory/t_malloc.h"
#include "platform/data_type.h"
#include "tensor.h"
namespace paddle_mobile {
namespace framework {
void TensorCopy(const Tensor &src, Tensor *dst);
void TensorCopySync(const Tensor &src, Tensor *dst);
template <typename T>
void TensorFromVector(const std::vector<T> &src, Tensor *dst);
template <typename T>
void TesnorToVector(const Tensor &src, std::vector<T> *dst);
bool TensorContainsNAN(const framework::Tensor &tensor);
bool TensorContainsInf(const framework::Tensor &tensor);
void TensorToStream(std::ostream &os, const Tensor &tensor);
void TensorFromStream(std::istream &is, Tensor *tensor);
//
// The implementation of template functions.
//
template <typename T>
void TensorFromVector(const std::vector<T> &src, Tensor *dst) {
auto src_ptr = static_cast<const void *>(src.data());
dst->Resize({static_cast<int64_t>(src.size())});
auto dst_ptr = static_cast<void *>(dst->mutable_data<T>());
auto size = src.size() * sizeof(T);
memory::Copy(dst_ptr, src_ptr, size);
}
template <typename T>
void TensorToVector(const Tensor &src, std::vector<T> *dst) {
auto src_ptr = static_cast<const void *>(src.data<T>());
auto size = src.numel() * sizeof(T);
dst->resize(src.numel());
auto dst_ptr = static_cast<void *>(dst->data());
memory::Copy(dst_ptr, src_ptr, size);
}
} // namespace framework
} // namespace paddle_mobile
src/framework/var_type.h 已删除 100644 → 0
浏览文件 @ ff0f7f7b
/* Copyright (c) 2018 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. */
#pragma once
#include "framework.pb.h"
#include "lod_tensor.h"
#include "selected_rows.h"
#include "variable.h"
namespace paddle_mobile {
namespace framework {
inline proto::VarType::Type ToVarType(std::type_index type) {
if (type.hash_code() == typeid(LoDTensor).hash_code()) {
return proto::VarType_Type_LOD_TENSOR;
} else if (type.hash_code() == typeid(SelectedRows).hash_code()) {
return proto::VarType_Type_SELECTED_ROWS;
} else {
// PADDLE_THROW("ToVarType:Unsupported type %s",
// type.name());
}
}
} // namespace framework
} // namespace paddle_mobile
src/framework/variable.h
浏览文件 @ 035c72f2
......@@ -14,19 +14,17 @@ limitations under the License. */
#pragma once
#include <iostream>
#include <memory>
#include <string>
#include <typeindex>
#include <typeinfo>
#include "../common/variant.h"
#include "paddle_mobile_object.h"
namespace paddle_mobile {
namespace framework {
using std::string;
class Variable : public PaddleMobileObject {
class Variable {
public:
template <typename T>
const T *Get() const {
......
src/io.cpp → src/io/io.cpp
浏览文件 @ 035c72f2
......@@ -13,11 +13,10 @@ See the License for the specific language governing permissions and
limitations under the License. */
#include "io.h"
#include <fstream>
#include <vector>
#include "common/log.h"
#include "common/enforce.h"
#include "common/log.h"
#include "framework/framework.pb-c.h"
#include "framework/lod_tensor.h"
#include "framework/operator.h"
......@@ -30,16 +29,20 @@ limitations under the License. */
namespace paddle_mobile {
using framework::Variable;
void ReadBinaryFile(const std::string &filename, std::string *contents) {
std::ifstream fin(filename, std::ios::in | std::ios::binary);
PADDLE_MOBILE_ENFORCE(fin.is_open(), "open file: %s failed",
char *Get_binary_data(std::string filename) {
FILE *file = fopen(filename.c_str(), "rb");
PADDLE_MOBILE_ENFORCE(file != nullptr, "can't open file: %s ",
filename.c_str());
fin.seekg(0, std::ios::end);
contents->clear();
contents->resize(fin.tellg());
fin.seekg(0, std::ios::beg);
fin.read(&(contents->at(0)), contents->size());
fin.close();
fseek(file, 0, SEEK_END);
long size = ftell(file);
PADDLE_MOBILE_ENFORCE(size > 0, "size is too small");
rewind(file);
char *data = new char[size];
size_t bytes_read = fread(data, 1, size, file);
PADDLE_MOBILE_ENFORCE(bytes_read == size,
"read binary file bytes do not match with fseek");
fclose(file);
return data;
}
static size_t ReadBuffer(const char *file_name, uint8_t **out) {
......@@ -66,110 +69,27 @@ static size_t ReadBuffer(const char *file_name, uint8_t **out) {
}
template <typename Dtype, Precision P>
void Loader<Dtype, P>::LoadVar(framework::Variable *variable,
const framework::VarDesc &var_desc,
const std::string &file_path) {
auto tensor = variable->GetMutable<framework::LoDTensor>();
std::ifstream is(file_path);
PADDLE_MOBILE_ENFORCE(is.is_open(), "open file: %s failed",
file_path.c_str());
std::fpos<mbstate_t> pos;
pos = is.tellg(); // save current position
is.seekg(0, std::ios::end);
is.seekg(pos); // restore saved position
// 1. version
uint32_t version;
is.read(reinterpret_cast<char *>(&version), sizeof(version));
// 2 Lod information
uint64_t lod_level;
is.read(reinterpret_cast<char *>(&lod_level), sizeof(lod_level));
auto &lod = *tensor->mutable_lod();
lod.resize(lod_level);
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 (auto j : tmp) {
LOG(kLOG_DEBUG1) << " lod - " << j;
}
lod[i] = tmp;
}
// 3. tensor version
uint32_t tensor_version;
is.read(reinterpret_cast<char *>(&tensor_version), sizeof(tensor_version));
// 4. tensor desc
int32_t size;
is.read(reinterpret_cast<char *>(&size), sizeof(size));
std::unique_ptr<char[]> buf(new char[size]);
is.read(reinterpret_cast<char *>(buf.get()), size);
const framework::TensorDesc &desc = var_desc.Tensor_desc();
PaddleMobile__Framework__Proto__VarType__TensorDesc *tensor_desc = NULL;
// void *v;
// PaddleMobile__Framework__Proto__VarType__TensorDesc_Closure()(tensor_desc,
// buf.get());
// DLOG << "PaddleMobile__Framework__Proto__VarType__TensorDesc_Closure- " <<
// tensor_desc;
// framework::TensorDesc &tensor_desc = variable->
// PaddleMobile__Framework__Proto__ProgramDesc *c_program;
// uint8_t *proto_buf = NULL;
// size_t read_size = ReadBuffer(file_path.c_str(), &proto_buf);
// c_program = paddle_mobile__framework__proto__program_desc__unpack(NULL,
// read_size, buf);
// paddle_mobile__framework__proto__var_type__tensor_desc__init()
int memory_size = 1;
for (auto l : desc.Dims()) {
memory_size *= l;
}
tensor->Resize(framework::make_ddim(desc.Dims()));
void *memory = tensor;
int type_size = 0;
switch (desc.DataType()) {
case framework::VARTYPE_TYPE_FP16:
type_size = 2;
break;
case framework::VARTYPE_TYPE_FP32:
type_size = 4;
memory = tensor->mutable_data<float>();
break;
case framework::VARTYPE_TYPE_FP64:
type_size = 8;
break;
case framework::VARTYPE_TYPE_INT32:
type_size = 4;
break;
case framework::VARTYPE_TYPE_INT64:
type_size = 8;
break;
case framework::VARTYPE_TYPE_BOOL:
type_size = 1;
break;
default:
break;
}
is.read(static_cast<char *>(memory), memory_size * type_size);
is.close();
const framework::Program<Dtype, P> Loader<Dtype, P>::Load(
const std::string &dirname, bool optimize) {
auto program = this->LoadProgram(dirname + "/__model__", optimize);
program.model_path = dirname;
return program;
}
template <typename Dtype, Precision P>
const framework::Program<Dtype, P> Loader<Dtype, P>::Load(
const std::string &dirname, bool optimize) {
std::string model_filename = dirname + "/__model__";
const std::string &model_path, const std::string &para_path,
bool optimize) {
auto program = this->LoadProgram(model_path, optimize);
program.para_path = para_path;
program.is_commbine = true;
return program;
}
template <typename Dtype, Precision P>
const framework::Program<Dtype, P> Loader<Dtype, P>::LoadProgram(
const std::string &model_path, bool optimize) {
std::string model_filename = model_path;
PaddleMobile__Framework__Proto__ProgramDesc *c_program;
uint8_t *buf = NULL;
size_t read_size = ReadBuffer(model_filename.c_str(), &buf);
......@@ -183,22 +103,16 @@ const framework::Program<Dtype, P> Loader<Dtype, P>::Load(
//
DLOG << "n_ops: " << (*c_program->blocks)->n_ops;
//
std::shared_ptr<framework::ProgramDesc> originProgramDesc =
std::make_shared<framework::ProgramDesc>(c_program);
auto originProgramDesc = std::make_shared<framework::ProgramDesc>(c_program);
framework::Program<Dtype, P> program;
program.model_path = dirname;
program.originProgram = originProgramDesc;
std::shared_ptr<framework::Scope> scope =
std::make_shared<framework::Scope>();
auto scope = std::make_shared<framework::Scope>();
program.scope = scope;
originProgramDesc->Block(0);
for (const auto &block : originProgramDesc->Blocks()) {
for (int i = 0; i < block->Vars().size(); ++i) {
std::shared_ptr<framework::VarDesc> var_desc = block->Vars()[i];
// DLOG << "var name-- " << var_desc->Name();
for (auto var_desc : block->Vars()) {
auto var = scope->Var(var_desc->Name());
if (var_desc->Type() == framework::VARTYPE_TYPE_LOD_TENSOR) {
......@@ -221,6 +135,8 @@ const framework::Program<Dtype, P> Loader<Dtype, P>::Load(
}
}
// originProgramDesc->Description("program: ");
if (optimize) {
framework::ProgramOptimize program_optimize;
program.optimizeProgram =
......@@ -267,36 +183,38 @@ Executor<Dtype, P>::Executor(const framework::Program<Dtype> p, int batch_size,
ops_of_block_[*block_desc.get()].push_back(op_base);
}
}
InitMemory();
if (program_.is_commbine) {
InitCombineMemory();
} else {
InitMemory();
}
}
template <typename Dtype, Precision P>
void Executor<Dtype, P>::LoadMemory(const framework::VarDesc var_desc,
framework::LoDTensor *tensor,
const std::string &file_path) {
std::ifstream is(file_path);
PADDLE_MOBILE_ENFORCE(is.is_open(), "open file: %s failed",
file_path.c_str());
std::fpos<mbstate_t> pos;
pos = is.tellg(); // save current position
is.seekg(0, std::ios::end);
is.seekg(pos); // restore saved position
framework::LoDTensor *tensor, char *&data) {
// 1. version
uint32_t version;
is.read(reinterpret_cast<char *>(&version), sizeof(version));
uint32_t version = *(uint32_t *)data;
data += sizeof(uint32_t);
// 2 Lod information
uint64_t lod_level;
is.read(reinterpret_cast<char *>(&lod_level), sizeof(lod_level));
uint64_t lod_level = *(uint64_t *)data;
data += sizeof(uint64_t);
auto &lod = *tensor->mutable_lod();
lod.resize(lod_level);
for (uint64_t i = 0; i < lod_level; ++i) {
uint64_t size;
is.read(reinterpret_cast<char *>(&size), sizeof(size));
uint64_t size = *(uint64_t *)data;
data += sizeof(uint64_t);
DLOG << "lod size: " << i << size;
std::vector<size_t> tmp(size / sizeof(size_t));
is.read(reinterpret_cast<char *>(tmp.data()),
static_cast<std::streamsize>(size));
for (int k = 0; k < tmp.size(); ++k) {
tmp[k] = *(size_t *)data;
DLOG << "tmp[k]: " << k << *(size_t *)data;
data += sizeof(size_t);
}
for (auto j : tmp) {
LOG(kLOG_DEBUG1) << " lod - " << j;
}
......@@ -304,17 +222,20 @@ void Executor<Dtype, P>::LoadMemory(const framework::VarDesc var_desc,
}
// 3. tensor version
uint32_t tensor_version;
is.read(reinterpret_cast<char *>(&tensor_version), sizeof(tensor_version));
uint32_t tensor_version = *(uint32_t *)data;
data += sizeof(uint32_t);
// 4. tensor desc
int32_t size;
is.read(reinterpret_cast<char *>(&size), sizeof(size));
int32_t size = *(int32_t *)data;
data += sizeof(int32_t);
std::unique_ptr<char[]> buf(new char[size]);
is.read(reinterpret_cast<char *>(buf.get()), size);
for (int m = 0; m < size; ++m) {
buf.get()[m] = data[m];
}
data += (sizeof(char) * size);
const framework::TensorDesc &desc = var_desc.Tensor_desc();
int memory_size = 1;
for (auto l : desc.Dims()) {
memory_size *= l;
......@@ -348,8 +269,10 @@ void Executor<Dtype, P>::LoadMemory(const framework::VarDesc var_desc,
break;
}
is.read(static_cast<char *>(memory), memory_size * type_size);
is.close();
for (int n = 0; n < memory_size * type_size; ++n) {
static_cast<char *>(memory)[n] = data[n];
}
data += (sizeof(char) * memory_size * type_size);
}
template <typename Dtype, Precision P>
......@@ -362,8 +285,12 @@ void Executor<Dtype, P>::InitMemory() {
if (var_desc->Name() == "feed" || var_desc->Name() == "fetch") {
continue;
}
LoadMemory(*var_desc, tensor,
program_.model_path + "/" + var_desc->Name());
char *origin_data =
Get_binary_data(program_.model_path + "/" + var_desc->Name());
char *data = origin_data;
LoadMemory(*var_desc, tensor, data);
delete origin_data;
} else {
if (var_desc->Type() == framework::VARTYPE_TYPE_LOD_TENSOR) {
auto tensor = var->template GetMutable<framework::LoDTensor>();
......@@ -375,6 +302,30 @@ void Executor<Dtype, P>::InitMemory() {
}
}
template <typename Dtype, Precision P>
void Executor<Dtype, P>::InitCombineMemory() {
char *origin_data = Get_binary_data(program_.para_path);
char *data = origin_data;
for (const auto &block : to_predict_program_->Blocks()) {
for (const auto &var_desc : block->Vars()) {
auto var = program_.scope->Var(var_desc->Name());
if (var_desc->Persistable()) {
auto tensor = var->template GetMutable<framework::LoDTensor>();
if (var_desc->Name() == "feed" || var_desc->Name() == "fetch") {
continue;
}
LoadMemory(*var_desc, tensor, data);
} else {
if (var_desc->Type() == framework::VARTYPE_TYPE_LOD_TENSOR) {
auto tensor = var->template GetMutable<framework::LoDTensor>();
tensor->template mutable_data<Ptype>();
}
}
}
}
delete origin_data;
}
template <typename Dtype, Precision P>
std::shared_ptr<framework::Tensor> Executor<Dtype, P>::Predict(
const framework::Tensor &t) {
......
src/io.h → src/io/io.h
浏览文件 @ 035c72f2
......@@ -14,51 +14,73 @@ limitations under the License. */
#pragma once
#include <memory.h>
#include <map>
#include <memory>
#include <string>
#include <vector>
#include "common/types.h"
#include "framework/lod_tensor.h"
#include "framework/operator.h"
#include "framework/paddle_mobile_object.h"
#include "framework/program/program.h"
#include "framework/tensor.h"
namespace paddle_mobile {
template <typename Dtype, Precision P = Precision::FP32>
class Loader : PaddleMobileObject {
template <typename Dtype = CPU, Precision P = Precision::FP32>
class Loader {
public:
/*
* @b load separate format fluid model
* @b 加载分开形式的 fluid 模型
* */
const framework::Program<Dtype, P> Load(const std::string &dirname,
bool optimize = true);
bool optimize = false);
/*
* @b load combine format fluid mode
* @b 加载结合在一起格式的模型
* */
const framework::Program<Dtype, P> Load(const std::string &model_path,
const std::string &para_path,
bool optimize = false);
private:
void LoadVar(framework::Variable *variable,
const framework::VarDesc &var_desc,
const std::string &file_path);
const framework::Program<Dtype, P> LoadProgram(const std::string &model_path,
bool optimize = false);
};
template <typename Dtype, Precision P = Precision::FP32>
template <typename Dtype = CPU, Precision P = Precision::FP32>
class Executor {
public:
typedef typename PrecisionTrait<P>::ptype Ptype;
/*
* @b init executor with program load by Loader class
* @b 用 loader load 的 program 实例化 executor
* */
Executor(const framework::Program<Dtype> p, int batch_size = 1,
bool use_optimize = true);
/*
* @b to predict
* */
std::shared_ptr<framework::Tensor> Predict(const framework::Tensor &t);
/*
* @b to predict with vector and dim
*
* @b 使用 输入 和 输入的维度信息 进行预测
* */
std::vector<Ptype> Predict(const std::vector<Ptype> &input,
const std::vector<int64_t> &dims);
protected:
Executor() = default;
void InitMemory();
void LoadMemory(const framework::VarDesc var_desc,
framework::LoDTensor *tensor, const std::string &file_path);
framework::LoDTensor *tensor, char *&data);
void InitCombineMemory();
framework::Program<Dtype> program_;
int batch_size_ = 1;
std::shared_ptr<framework::ProgramDesc> to_predict_program_;
......
src/jni/paddle_mobile_jni.cpp 0 → 100644
浏览文件 @ 035c72f2
/* Copyright (c) 2018 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. */
#ifdef ANDROID
#include "paddle_mobile_jni.h"
#ifdef __cplusplus
extern "C" {
#endif
namespace paddle_mobile {
namespace jni {
using framework::DDim;
using framework::Program;
using framework::Tensor;
using paddle_mobile::CPU;
using std::string;
extern const char *ANDROID_LOG_TAG =
"paddle_mobile LOG built on " __DATE__ " " __TIME__;
static Executor<CPU> *shared_executor_instance = nullptr;
// toDo mutex lock
// static std::mutex shared_mutex;
Executor<CPU> *getExecutorInstance(const Program<CPU> p, int batch_size,
bool use_optimize) {
if (nullptr == shared_executor_instance) {
shared_executor_instance = new Executor<CPU>(p, batch_size, use_optimize);
}
return shared_executor_instance;
}
string jstring2cppstring(JNIEnv *env, jstring jstr) {
const char *cstr = env->GetStringUTFChars(jstr, 0);
string cppstr(cstr);
env->ReleaseStringUTFChars(jstr, cstr);
return cppstr;
}
JNIEXPORT jboolean JNICALL Java_com_baidu_paddle_PML_load(JNIEnv *env,
jclass thiz,
jstring modelPath) {
paddle_mobile::Loader<paddle_mobile::CPU> loader;
bool optimize = true;
auto program = loader.Load(jstring2cppstring(env, modelPath), optimize);
shared_executor_instance = getExecutorInstance(program, 1, optimize);
return shared_executor_instance != nullptr ? JNI_TRUE : JNI_FALSE;
}
JNIEXPORT jfloatArray JNICALL Java_com_baidu_paddle_PML_predictImage(
JNIEnv *env, jclass thiz, jfloatArray buf) {
jfloatArray result = NULL;
int count = 0;
float *dataPointer = nullptr;
if (nullptr != buf) {
dataPointer = env->GetFloatArrayElements(buf, NULL);
}
framework::Tensor input;
framework::DDim ddim = framework::make_ddim({1, 3, 224, 224});
input.Resize(ddim);
auto input_ptr = input.mutable_data<float>();
for (int i = 0; i < framework::product(ddim); i++) {
input_ptr[i] = dataPointer[i];
}
auto output = shared_executor_instance->Predict(input);
count = output->numel();
result = env->NewFloatArray(count);
env->SetFloatArrayRegion(result, 0, count, output->data<float>());
return result;
}
JNIEXPORT void JNICALL Java_com_baidu_paddle_PML_clear(JNIEnv *env,
jclass thiz) {}
} // namespace jni
} // namespace paddle_mobile
#ifdef __cplusplus
}
#endif
#endif
src/framework/data_transform.h → src/jni/paddle_mobile_jni.h
浏览文件 @ 035c72f2
......@@ -13,25 +13,39 @@ See the License for the specific language governing permissions and
limitations under the License. */
#pragma once
#include <functional>
#include <utility>
#include <vector>
#include "framework/op_kernel_type.h"
#include "framework/selected_rows.h"
#ifdef ANDROID
#include <jni.h>
#include "common/log.h"
#include "framework/tensor.h"
#include "framework/variable.h"
#include "io/io.h"
#ifdef __cplusplus
extern "C" {
#endif
namespace paddle_mobile {
namespace framework {
void DataTransform(const OpKernelType &expected_kernel_type,
const OpKernelType &kernel_type_for_var,
const Tensor &input_tensor, Tensor *out);
void CopyVariableWithTensor(const Variable &in_var, const Tensor &tensor,
Variable *out_var);
} // namespace framework
namespace jni {
/**
* load model & params of the net for android
*/
JNIEXPORT jboolean JNICALL Java_com_baidu_paddle_PML_load(JNIEnv *env,
jclass thiz,
jstring modelPath);
/**
* object detection for anroid
*/
JNIEXPORT jfloatArray JNICALL Java_com_baidu_paddle_PML_predictImage(
JNIEnv *env, jclass thiz, jfloatArray buf);
/**
* clear data of the net when destroy for android
*/
JNIEXPORT void JNICALL Java_com_baidu_paddle_PMLL_clear(JNIEnv *env,
jclass thiz);
} // namespace jni
} // namespace paddle_mobile
#ifdef __cplusplus
}
#endif
#endif
src/operators/batchnorm_op.cpp
浏览文件 @ 035c72f2
......@@ -12,7 +12,11 @@ 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. */
#ifdef BATCHNORM_OP
#include "batchnorm_op.h"
#include "framework/op_proto_maker.h"
#include "framework/op_registry.h"
namespace paddle_mobile {
namespace operators {
......@@ -29,3 +33,5 @@ template class BatchNormOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(batch_norm);
REGISTER_OPERATOR(batch_norm, ops::BatchNormOp);
#endif
src/operators/batchnorm_op.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef BATCHNORM_OP
#pragma once
#include <string>
......@@ -27,7 +29,7 @@ class BatchNormOp : public framework::OperatorWithKernel<DeviceType> {
public:
BatchNormOp(const string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs,
const framework::AttributeMap attrs,
const framework::AttributeMap &attrs,
std::shared_ptr<framework::Scope> scope)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
scope),
......@@ -47,3 +49,5 @@ class BatchNormOp : public framework::OperatorWithKernel<DeviceType> {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/box_coder_op.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef BOXCODER_OP
#include "operators/box_coder_op.h"
#include <vector>
namespace paddle_mobile {
......@@ -52,3 +54,5 @@ template class BoxCoderOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(box_coder);
REGISTER_OPERATOR(box_coder, ops::BoxCoderOp);
#endif
src/operators/box_coder_op.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef BOXCODER_OP
#pragma once
#include <string>
......@@ -30,7 +32,7 @@ class BoxCoderOp : public framework::OperatorWithKernel<DeviceType> {
public:
BoxCoderOp(const std::string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs,
const framework::AttributeMap attrs,
const framework::AttributeMap &attrs,
std::shared_ptr<framework::Scope> scope)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
scope),
......@@ -50,3 +52,5 @@ class BoxCoderOp : public framework::OperatorWithKernel<DeviceType> {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/concat_op.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef CONCAT_OP
#include "concat_op.h"
namespace paddle_mobile {
......@@ -62,3 +64,5 @@ template class ConcatOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(concat);
REGISTER_OPERATOR(concat, ops::ConcatOp);
#endif
src/operators/concat_op.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef CONCAT_OP
#pragma once
#include <string>
......@@ -25,7 +27,7 @@ template <typename DeviceType, typename T>
class ConcatOp : public framework::OperatorWithKernel<DeviceType> {
public:
ConcatOp(const 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)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
scope),
......@@ -45,3 +47,5 @@ class ConcatOp : public framework::OperatorWithKernel<DeviceType> {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/conv_op.cpp
浏览文件 @ 035c72f2
......@@ -12,9 +12,10 @@ 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. */
#ifdef CONV_OP
#include "operators/conv_op.h"
#include <vector>
#include "framework/data_type.h"
#include "framework/op_proto_maker.h"
#include "framework/op_registry.h"
......@@ -23,30 +24,11 @@ namespace operators {
template <typename Dtype, typename T>
void ConvOp<Dtype, T>::InferShape() const {
// std::cout << " begin get dims: " << std::endl;
auto in_dims = param_.Input()->dims();
// std::cout << " end get in dims: " << std::endl;
// std::cout << " in_dims: " << in_dims << std::endl;
// std::cout << " begin get Filter " << std::endl;
auto filter_dims = param_.Filter()->dims();
// std::cout << " end get Filter " << std::endl;
// std::cout << " begin get Attrs " << std::endl;
const std::vector<int> &strides = param_.Strides();
// std::cout << " end get Attrs " << strides[0] << std::endl;
std::vector<int> paddings = param_.Paddings();
int groups = param_.Groups();
std::vector<int> dilations = param_.Dilations();
PADDLE_MOBILE_ENFORCE((in_dims.size() == filter_dims.size() &&
......@@ -73,3 +55,5 @@ template class ConvOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(conv2d);
REGISTER_OPERATOR(conv2d, ops::ConvOp);
#endif
src/operators/conv_op.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef CONV_OP
#pragma once
#include <string>
......@@ -53,3 +55,5 @@ inline int ConvOutputSize(int input_size, int filter_size, int dilation,
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/depthwise_conv_op.cpp
浏览文件 @ 035c72f2
......@@ -12,9 +12,10 @@ 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. */
#ifdef DEPTHWISECONV_OP
#include "operators/depthwise_conv_op.h"
#include <vector>
#include "framework/data_type.h"
#include "framework/op_proto_maker.h"
#include "framework/op_registry.h"
#include "operators/conv_op.h"
......@@ -55,3 +56,5 @@ template class DepthwiseConvOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(depthwise_conv2d);
REGISTER_OPERATOR(depthwise_conv2d, ops::DepthwiseConvOp);
#endif
src/operators/depthwise_conv_op.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef DEPTHWISECONV_OP
#pragma once
#include <string>
......@@ -47,3 +49,5 @@ class DepthwiseConvOp : public framework::OperatorWithKernel<DeviceType> {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/elementwise_add_op.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef ELEMENTWISEADD_OP
#include "elementwise_add_op.h"
namespace paddle_mobile {
......@@ -29,3 +31,5 @@ template class ElementwiseAddOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(elementwise_add);
REGISTER_OPERATOR(elementwise_add, ops::ElementwiseAddOp);
#endif
src/operators/elementwise_add_op.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef ELEMENTWISEADD_OP
#pragma once
#include <string>
......@@ -27,7 +29,7 @@ class ElementwiseAddOp : public framework::OperatorWithKernel<DeviceType> {
public:
ElementwiseAddOp(const string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs,
const framework::AttributeMap attrs,
const framework::AttributeMap &attrs,
std::shared_ptr<framework::Scope> scope)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
scope),
......@@ -46,3 +48,5 @@ class ElementwiseAddOp : public framework::OperatorWithKernel<DeviceType> {
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/framework/program/var_desc.cpp → src/operators/fusion_conv_add.cpp
浏览文件 @ 035c72f2
......@@ -12,9 +12,20 @@ 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. */
#include "var_desc.h"
#ifdef FUSIONCONVADD_OP
#include "operators/fusion_conv_add.h"
namespace paddle_mobile {
namespace operators {
namespace framework {} // namespace framework
template <typename Dtype, typename T>
void FushionConvAddOp<Dtype, T>::InferShape() const {}
template class FushionConvAddOp<CPU, float>;
} // namespace operators
} // namespace paddle_mobile
namespace ops = paddle_mobile::operators;
USE_OP(conv_add);
REGISTER_OPERATOR(conv_add, ops::FushionConvAddOp);
#endif
src/framework/selected_rows.h → src/operators/fusion_conv_add.h
浏览文件 @ 035c72f2
......@@ -12,66 +12,61 @@ 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. */
#ifdef FUSIONCONVADD_OP
#pragma once
#include <string>
#include <vector>
#include "lod_tensor.h"
#include "tensor.h"
#include "framework/operator.h"
#include "framework/program/program-optimize/fusion_op_register.h"
namespace paddle_mobile {
namespace framework {
class SelectedRows {
namespace operators {
using std::string;
using std::vector;
class FusionConvAddMatcher : public framework::FusionOpMatcher {
public:
SelectedRows(const std::vector<int64_t> &rows, const int64_t &height)
: rows_(rows), height_(height) {
value_.reset(new Tensor());
FusionConvAddMatcher() {
node_ = framework::Node(G_OP_TYPE_CONV);
node_ > std::make_shared<framework::Node>(G_OP_TYPE_ELEMENTWISE_ADD);
}
SelectedRows() {
height_ = 0;
value_.reset(new Tensor());
void FolderNodes(
framework::Node *node,
std::vector<std::shared_ptr<framework::Node>> *removed_nodes) {
vector<std::shared_ptr<framework::OpDesc>> origin_descs =
node->OpDescs(node_.Depth());
node->Folder(node_.Depth(), Type(),
{{G_OP_TYPE_ELEMENTWISE_ADD, {"Y", "Y"}}}, removed_nodes);
}
const Tensor &value() const { return *value_; }
Tensor *mutable_value() { return value_.get(); }
int64_t height() const { return height_; }
void set_height(int64_t height) { height_ = height; }
const std::vector<int64_t> &rows() const { return rows_; }
std::vector<int64_t> *mutable_rows() { return &rows_; }
std::string Type() { return G_OP_TYPE_CONV_ADD; }
};
void set_rows(const std::vector<int64_t> &rows) { rows_ = rows; }
template <typename DeviceType, typename T>
class FushionConvAddOp : public framework::OperatorWithKernel<DeviceType> {
public:
FushionConvAddOp(const string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs,
const framework::AttributeMap &attrs,
std::shared_ptr<framework::Scope> scope)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
scope) {}
/**
* get the index of id in rows
*/
int64_t index(int64_t id) const {
auto it = std::find(rows_.begin(), rows_.end(), id);
// PADDLE_ENFORCE(it != rows_.end(), "id should be in rows");
return static_cast<int64_t>(std::distance(rows_.begin(), it));
}
void RunImpl() const {}
DDim GetCompleteDims() const {
std::vector<int64_t> dims = vectorize(value_->dims());
dims[0] = height_;
return make_ddim(dims);
}
using framework::OperatorWithKernel<DeviceType>::OperatorWithKernel;
void InferShape() const override;
private:
// Notice: rows can be duplicate. We can have {0, 4, 7, 0, 5, 7, 9}
// here.
// SelectedRows are simply concated when adding together. Until a
// SelectedRows add a Tensor, will the duplicate rows be handled.
std::vector<int64_t> rows_;
std::unique_ptr<Tensor> value_{nullptr};
int64_t height_;
protected:
// FushionFcParam param_;
};
} // namespace framework
// static framework::FusionOpRegistrar fc_registrar(new FusionConvAddMatcher());
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/fusion_conv_add_relu_op.cpp
浏览文件 @ 035c72f2
......@@ -12,4 +12,8 @@ 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. */
#ifdef CONVADDRELU_OP
#include "fusion_conv_add_relu_op.h"
#endif
src/operators/fusion_conv_add_relu_op.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef CONVADDRELU_OP
#pragma once
#include "framework/operator.h"
......@@ -28,16 +30,18 @@ class FushionConvAddReluOpMatcher : public framework::FusionOpMatcher {
std::make_shared<framework::Node>(G_OP_TYPE_RELU);
}
void FolderNodes(framework::Node *node) {
void FolderNodes(
framework::Node *node,
std::vector<std::shared_ptr<framework::Node>> *removed_nodes) {
std::vector<std::shared_ptr<framework::OpDesc>> origin_descs =
node->OpDescs(node_.Depth());
node->Folder(node_.Depth(), Type(),
{{G_OP_TYPE_ELEMENTWISE_ADD, {"Y", "Z"}}});
{{G_OP_TYPE_ELEMENTWISE_ADD, {"Y", "Z"}}}, removed_nodes);
}
std::string Type() { return G_OP_TYPE_FUSION_CONV_ADD_RELU; }
};
class FusionFcOp {
class ConvAddReluOp {
public:
private:
};
......@@ -47,3 +51,5 @@ class FusionFcOp {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/fusion_fc_op.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef FUSION_FC_OP
#include "operators/fusion_fc_op.h"
namespace paddle_mobile {
namespace operators {
......@@ -54,3 +56,5 @@ template class FushionFcOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(fc);
REGISTER_OPERATOR(fc, ops::FushionFcOp);
#endif
src/operators/fusion_fc_op.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef FUSION_FC_OP
#pragma once
#include <string>
......@@ -32,11 +34,13 @@ class FusionFcMatcher : public framework::FusionOpMatcher {
node_ > std::make_shared<framework::Node>(G_OP_TYPE_ELEMENTWISE_ADD);
}
void FolderNodes(framework::Node *node) {
void FolderNodes(
framework::Node *node,
std::vector<std::shared_ptr<framework::Node>> *removed_nodes) {
vector<std::shared_ptr<framework::OpDesc>> origin_descs =
node->OpDescs(node_.Depth());
node->Folder(node_.Depth(), Type(),
{{G_OP_TYPE_ELEMENTWISE_ADD, {"Y", "Z"}}});
{{G_OP_TYPE_ELEMENTWISE_ADD, {"Y", "Z"}}}, removed_nodes);
}
std::string Type() { return G_OP_TYPE_FC; }
......@@ -47,7 +51,7 @@ class FushionFcOp : public framework::OperatorWithKernel<DeviceType> {
public:
FushionFcOp(const string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs,
const framework::AttributeMap attrs,
const framework::AttributeMap &attrs,
std::shared_ptr<framework::Scope> scope)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
scope),
......@@ -65,7 +69,9 @@ class FushionFcOp : public framework::OperatorWithKernel<DeviceType> {
FushionFcParam param_;
};
static framework::FusionOpRegistrar fc_registrar(new FusionFcMatcher());
// static framework::FusionOpRegistrar fc_registrar(new FusionFcMatcher());
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/batchnorm_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef BATCHNORM_OP
#pragma once
#include "operators/kernel/batchnorm_kernel.h"
......@@ -47,7 +49,7 @@ void BatchNormKernel<CPU, float>::Compute(const BatchNormParam &param) const {
Tensor inv_std;
auto inv_std_ptr = inv_std.mutable_data<float>(make_ddim({C}));
if (C != variance->numel()) {
std::cout << "C must equal to variance.numel()" << std::endl;
DLOG << "C must equal to variance.numel()";
}
assert(C == variance->numel());
......@@ -91,3 +93,5 @@ void BatchNormKernel<CPU, float>::Compute(const BatchNormParam &param) const {
}
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/box_coder_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,7 +12,7 @@ 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. */
#pragma once
#ifdef BOXCODER_OP
#include "operators/kernel/box_coder_kernel.h"
......@@ -135,3 +135,5 @@ void BoxCoderKernel<CPU, float>::Compute(const BoxCoderParam& param) const {
}
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/concat_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,7 +12,7 @@ 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. */
#pragma once
#ifdef CONCAT_OP
#include "operators/kernel/concat_kernel.h"
......@@ -85,3 +85,5 @@ void ConcatKernel<CPU, float>::Compute(const ConcatParam &param) const {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/conv_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef CONV_OP
#include "operators/kernel/conv_kernel.h"
namespace paddle_mobile {
......@@ -112,3 +114,5 @@ template class ConvKernel<CPU, float>;
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/depthwise_conv_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef DEPTHWISECONV_OP
#include "operators/kernel/depthwise_conv_kernel.h"
#include "operators/kernel/conv_kernel.h"
......@@ -124,3 +126,5 @@ template class DepthwiseConvKernel<CPU, float>;
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/elementwise_add_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef ELEMENTWISEADD_OP
#pragma once
#include "operators/kernel/elementwise_add_kernel.h"
......@@ -40,3 +42,5 @@ template class ElementwiseAddKernel<CPU, float>;
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/fushion_fc_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef FUSION_FC_OP
#pragma once
#include "operators/kernel/fushion_fc_kernel.h"
......@@ -65,3 +67,5 @@ void FushionFcKernel<CPU, float>::Compute(const FushionFcParam &param) const {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/lrn_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef LRN_OP
#pragma once
#include "operators/kernel/lrn_kernel.h"
......@@ -42,3 +44,5 @@ template class LrnKernel<CPU, float>;
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/mul_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef MUL_OP
#pragma once
#include "operators/kernel/mul_kernel.h"
......@@ -48,3 +50,5 @@ template class MulKernel<CPU, float>;
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/multiclass_nms_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef MULTICLASSNMS_OP
#pragma once
#include "operators/kernel/multiclass_nms_kernel.h"
......@@ -273,3 +275,5 @@ void MultiClassNMSKernel<CPU, float>::Compute(
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/pool_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef POOL_OP
#include <operators/kernel/pool_kernel.h>
#include "common/log.h"
......@@ -73,3 +75,5 @@ void PoolKernel<CPU, float>::Compute(const PoolParam &param) const {
}
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/prior_box_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef PRIORBOX_OP
#pragma once
#include "operators/kernel/prior_box_kernel.h"
......@@ -143,3 +145,5 @@ void PriorBoxKernel<CPU, float>::Compute(const PriorBoxParam &param) const {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/relu_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef RELU_OP
#pragma once
#include "operators/kernel/relu_kernel.h"
......@@ -45,3 +47,5 @@ void ReluKernel<CPU, float>::Compute(const ReluParam &param) const {
}
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/reshape_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef RESHAPE_OP
#pragma once
#include "operators/kernel/reshape_kernel.h"
......@@ -49,3 +51,5 @@ void ReshapeKernel<CPU, float>::Compute(const ReshapeParam &param) const {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/sigmoid_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef SIGMOID_OP
#include "../sigmoid_kernel.h"
#if __ARM_NEON
#include "../../math/math_func_neon.h"
......@@ -25,35 +27,23 @@ using framework::Tensor;
void sigmoid(const Tensor *X, Tensor *Y) {
#if __ARM_NEON
DLOG << "step1";
const float *input = X->data<float>();
DLOG << "step11";
float *output = Y->mutable_data<float>();
DLOG << "step2";
const DDim &dDim = X->dims();
DLOG << "step3";
int axis_index = 1;
if (dDim.size() < 4) {
axis_index = 0;
}
DLOG << "step4";
DDim outer_ddim =
paddle_mobile::framework::slice_ddim(dDim, 0, axis_index + 1);
DDim inner_ddim =
paddle_mobile::framework::slice_ddim(dDim, axis_index + 1, dDim.size());
DLOG << "step5";
int out_size = paddle_mobile::framework::product(outer_ddim);
int inner_size = paddle_mobile::framework::product(inner_ddim);
DLOG << "step6";
#pragma omp parallel for
DLOG << "outsize=" << out_size;
DLOG << "innersize=" << inner_size;
#pragma omp parallel for
for (int i = 0; i < out_size; ++i) {
const float *input_outer_ptr = input + i * inner_size;
float *output_outer_ptr = output + i * inner_size;
......@@ -93,3 +83,5 @@ void SigmoidKernel<CPU, float>::Compute(const SigmoidParam &param) const {
template class SigmoidKernel<CPU, float>;
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/softmax_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef SOFTMAX_OP
#include "../softmax_kernel.h"
#include "../../math/softmax.h"
namespace paddle_mobile {
......@@ -29,3 +31,5 @@ void SoftmaxKernel<CPU, float>::Compute(const SoftmaxParam &param) const {
template class SoftmaxKernel<CPU, float>;
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/arm/transpose_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,7 +12,7 @@ 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. */
#pragma once
#ifdef TRANSPOSE_OP
#include "operators/kernel/transpose_kernel.h"
......@@ -70,3 +70,5 @@ void TransposeKernel<CPU, float>::Compute(const TransposeParam& param) const {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/batchnorm_kernel.h
浏览文件 @ 035c72f2
......@@ -12,9 +12,12 @@ 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. */
#ifdef BATCHNORM_OP
#pragma once
#include "framework/operator.h"
#include "operators/op_param.h"
#pragma once;
namespace paddle_mobile {
namespace operators {
......@@ -30,3 +33,5 @@ class BatchNormKernel
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/box_coder_kernel.h
浏览文件 @ 035c72f2
......@@ -12,14 +12,16 @@ 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. */
#ifdef BOXCODER_OP
#pragma once
#include <vector>
#include "framework/operator.h"
#include "operators/math/transform.h"
#include "operators/op_param.h"
#pragma once;
namespace paddle_mobile {
namespace operators {
......@@ -31,3 +33,5 @@ class BoxCoderKernel
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/concat_kernel.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef CONCAT_OP
#pragma once
#include "framework/operator.h"
#include "operators/op_param.h"
......@@ -29,3 +31,5 @@ class ConcatKernel : public framework::OpKernelBase<DeviceType, ConcatParam> {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/conv_kernel.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,10 @@ 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. */
#ifdef CONV_OP
#pragma once
#include <vector>
#include "framework/operator.h"
#include "operators/math/im2col.h"
......@@ -19,8 +23,6 @@ limitations under the License. */
#include "operators/math/vol2col.h"
#include "operators/op_param.h"
#pragma once;
namespace paddle_mobile {
namespace operators {
......@@ -49,3 +51,5 @@ inline bool IsExpand(const std::vector<int64_t> &filter_dim,
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/depthwise_conv_kernel.h
浏览文件 @ 035c72f2
......@@ -12,14 +12,16 @@ 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. */
#ifdef DEPTHWISECONV_OP
#pragma once
#include "framework/operator.h"
#include "operators/math/im2col.h"
#include "operators/math/math_function.h"
#include "operators/math/vol2col.h"
#include "operators/op_param.h"
#pragma once;
namespace paddle_mobile {
namespace operators {
......@@ -32,3 +34,5 @@ class DepthwiseConvKernel : public OpKernelBase<DeviceType, ConvParam> {
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/elementwise_add_kernel.h
浏览文件 @ 035c72f2
......@@ -12,7 +12,9 @@ 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. */
#pragma once;
#ifdef ELEMENTWISEADD_OP
#pragma once
#include "framework/operator.h"
#include "operators/math/elementwise_op_function.h"
......@@ -31,3 +33,5 @@ class ElementwiseAddKernel
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/fpga/conv_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef CONV_OP
namespace paddle_mobile {
namespace operators {
......@@ -22,3 +24,5 @@ namespace operators {
// template class ConvKernel<FPGA, float>;
}
} // namespace paddle_mobile
#endif
src/operators/kernel/fushion_fc_kernel.h
浏览文件 @ 035c72f2
......@@ -12,12 +12,14 @@ 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. */
#ifdef FUSION_FC_OP
#pragma once
#include "framework/operator.h"
#include "operators/math/math_function.h"
#include "operators/op_param.h"
#pragma once;
namespace paddle_mobile {
namespace operators {
......@@ -29,3 +31,5 @@ class FushionFcKernel
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/lrn_kernel.h
浏览文件 @ 035c72f2
......@@ -12,9 +12,12 @@ 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. */
#ifdef LRN_OP
#pragma once
#include "framework/operator.h"
#include "operators/op_param.h"
#pragma once;
namespace paddle_mobile {
namespace operators {
......@@ -70,3 +73,5 @@ class LrnKernel : public framework::OpKernelBase<DeviceType, LrnParam> {
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/framework/paddle_mobile_object.h → src/operators/kernel/mali/conv_kernel.cpp
浏览文件 @ 035c72f2
......@@ -12,21 +12,18 @@ 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. */
#pragma once
#ifdef CONV_OP
#include <string>
#include "stdio.h"
#include "operators/kernel/conv_kernel.h"
namespace paddle_mobile {
namespace operators {
class PaddleMobileObject {
public:
virtual std::string ToString() {
char address[128] = {0};
sprintf(address, "%p", this);
return std::string(address);
}
template <>
void ConvKernel<GPU_MALI, float>::Compute(const ConvParam &param) const {}
private:
};
template class ConvKernel<GPU_MALI, float>;
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/mul_kernel.h
浏览文件 @ 035c72f2
......@@ -12,10 +12,13 @@ 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. */
#ifdef MUL_OP
#pragma once
#include "framework/operator.h"
#include "operators/math/math_function.h"
#include "operators/op_param.h"
#pragma once;
namespace paddle_mobile {
namespace operators {
......@@ -29,3 +32,5 @@ class MulKernel : public framework::OpKernelBase<DeviceType, MulParam> {
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/multiclass_nms_kernel.h
浏览文件 @ 035c72f2
......@@ -12,10 +12,13 @@ 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. */
#ifdef MULTICLASSNMS_OP
#pragma once
#include "framework/operator.h"
#include "operators/op_param.h"
#pragma once;
#include "operators/op_param.h"
namespace paddle_mobile {
namespace operators {
......@@ -28,3 +31,5 @@ class MultiClassNMSKernel
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/pool_kernel.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef POOL_OP
#pragma once
#include "framework/operator.h"
......@@ -29,3 +31,5 @@ class PoolKernel : public OpKernelBase<DeviceType, PoolParam> {
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/prior_box_kernel.h
浏览文件 @ 035c72f2
......@@ -12,14 +12,16 @@ 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. */
#ifdef PRIORBOX_OP
#pragma once
#include <vector>
#include "framework/operator.h"
#include "operators/math/transform.h"
#include "operators/op_param.h"
#pragma once;
namespace paddle_mobile {
namespace operators {
......@@ -55,3 +57,5 @@ class PriorBoxKernel
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/relu_kernel.h
浏览文件 @ 035c72f2
......@@ -12,10 +12,13 @@ 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. */
#ifdef RELU_OP
#pragma once
#include "framework/operator.h"
#include "operators/op_param.h"
#pragma once;
#include "operators/op_param.h"
namespace paddle_mobile {
namespace operators {
......@@ -27,3 +30,5 @@ class ReluKernel : public framework::OpKernelBase<DeviceType, ReluParam> {
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/reshape_kernel.h
浏览文件 @ 035c72f2
......@@ -12,12 +12,14 @@ 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. */
#include <vector>
#ifdef RESHAPE_OP
#pragma once
#include <vector>
#include "framework/operator.h"
#include "operators/op_param.h"
#pragma once;
#include "operators/op_param.h"
namespace paddle_mobile {
namespace operators {
......@@ -72,3 +74,5 @@ class ReshapeKernel : public framework::OpKernelBase<DeviceType, ReshapeParam> {
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/sigmoid_kernel.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef SIGMOID_OP
#pragma once
#include "framework/operator.h"
......@@ -27,3 +29,5 @@ class SigmoidKernel : public OpKernelBase<DeviceType, SigmoidParam> {
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/softmax_kernel.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef SOFTMAX_OP
#pragma once
#include "framework/operator.h"
......@@ -30,3 +32,5 @@ class SoftmaxKernel : public OpKernelBase<DeviceType, SoftmaxParam> {
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/kernel/transpose_kernel.h
浏览文件 @ 035c72f2
......@@ -12,13 +12,15 @@ 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. */
#ifdef TRANSPOSE_OP
#pragma once
#include <vector>
#include "framework/operator.h"
#include "operators/op_param.h"
#pragma once;
namespace paddle_mobile {
namespace operators {
......@@ -30,3 +32,5 @@ class TransposeKernel
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/lrn_op.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef LRN_OP
#include "lrn_op.h"
namespace paddle_mobile {
......@@ -29,3 +31,5 @@ template class LrnOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(lrn);
REGISTER_OPERATOR(lrn, ops::LrnOp);
#endif
src/operators/lrn_op.h
浏览文件 @ 035c72f2
......@@ -11,6 +11,9 @@ 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. */
#ifdef LRN_OP
#pragma once
#include <string>
......@@ -25,7 +28,7 @@ template <typename DeviceType, typename T>
class LrnOp : public framework::OperatorWithKernel<DeviceType> {
public:
LrnOp(const 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)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
scope),
......@@ -45,3 +48,5 @@ class LrnOp : public framework::OperatorWithKernel<DeviceType> {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/math/gemm.cpp
浏览文件 @ 035c72f2
......@@ -13,10 +13,14 @@ See the License for the specific language governing permissions and
limitations under the License. */
#include "operators/math/gemm.h"
#ifndef X86
#include <arm_neon.h>
#endif
namespace paddle_mobile {
namespace operators {
namespace math {
float ab[MR * NR];
// 将A矩阵分块复制到连续内存(ColMajor)
void PackMatrixA(int m, int k, int paddingM, const float *A, int lda,
float *buffer) {
......@@ -170,17 +174,197 @@ void InnerKernel(int m, int n, int k, float alpha, const float *A, int lda,
}
// 计算一个更小的 4 * 4 的 C 矩阵分块
#if defined(IOS)
void AddDot4x4(int k, float alpha, const float *a, int lda, const float *b,
int ldb, float beta, float *C, int ldc, int mc, int nc) {
// init C
float32x4_t cv0 = vdupq_n_f32(0.0);
float32x4_t cv1 = vdupq_n_f32(0.0);
float32x4_t cv2 = vdupq_n_f32(0.0);
float32x4_t cv3 = vdupq_n_f32(0.0);
float32x4_t av;
float32x4_t bv;
float32x2_t av01;
float32x2_t av23;
for (int p = 0; p < k; p += 1) {
av = vld1q_f32(a);
bv = vld1q_f32(b);
av01 = vget_low_f32(av);
cv0 = vmlaq_lane_f32(cv0, bv, av01, 0);
cv1 = vmlaq_lane_f32(cv1, bv, av01, 1);
av23 = vget_high_f32(av);
cv2 = vmlaq_lane_f32(cv2, bv, av23, 0);
cv3 = vmlaq_lane_f32(cv3, bv, av23, 1);
a += MR;
b += NR;
}
float32x4x4_t cv = {cv0, cv1, cv2, cv3};
int i, j;
for (i = 0; i < mc; ++i) {
for (j = 0; j < nc; ++j) {
if (beta == 0.0) {
C(i, j) = 0.0;
} else if (beta != 1.0) {
C(i, j) *= beta;
}
if (j == 0) {
C(i, j) += alpha * vgetq_lane_f32(cv.val[i], 0);
} else if (j == 1) {
C(i, j) += alpha * vgetq_lane_f32(cv.val[i], 1);
} else if (j == 2) {
C(i, j) += alpha * vgetq_lane_f32(cv.val[i], 2);
} else if (j == 3) {
C(i, j) += alpha * vgetq_lane_f32(cv.val[i], 3);
}
}
}
}
#elif defined(ARMV7)
void AddDot4x4(int k, float alpha, const float *a, int lda, const float *b,
int ldb, float beta, float *C, int ldc, int mc, int nc) {
int kc1 = k / 2, kc2 = k % 2;
int bytes_ldc = 4 * ldc;
int flag_alpha = (alpha == 1.0) ? 1 : 2;
int flag_beta;
if (beta == 0.0) {
flag_beta = 0;
} else if (beta == 1.0) {
flag_beta = 1;
} else {
flag_beta = 2;
}
asm volatile(
"vmov.f32 q10, #0.0 \n\t"
"vmov.f32 q11, #0.0 \n\t"
"vmov.f32 q12, #0.0 \n\t"
"vmov.f32 q13, #0.0 \n\t"
"subs %[kc1], %[kc1], #1 \n\t"
"blt end_kc1_%= \n\t"
"loop_kc1_%=: \n\t"
"vld1.32 {q0, q1}, [%[a]]! \n\t"
"vld1.32 {q2, q3}, [%[b]]! \n\t"
"vmla.f32 q10, q2, d0[0] \n\t"
"vmla.f32 q11, q2, d0[1] \n\t"
"vmla.f32 q12, q2, d1[0] \n\t"
"vmla.f32 q13, q2, d1[1] \n\t"
"vmla.f32 q10, q3, d2[0] \n\t"
"vmla.f32 q11, q3, d2[1] \n\t"
"vmla.f32 q12, q3, d3[0] \n\t"
"vmla.f32 q13, q3, d3[1] \n\t"
"subs %[kc1], %[kc1], #1 \n\t"
"bge loop_kc1_%= \n\t"
"end_kc1_%=: \n\t"
"subs %[kc2], %[kc2], #1 \n\t"
"blt end_kc2_%= \n\t"
"vld1.32 {q0}, [%[a]]! \n\t"
"vld1.32 {q1}, [%[b]]! \n\t"
"vmla.f32 q10, q1, d0[0] \n\t"
"vmla.f32 q11, q1, d0[1] \n\t"
"vmla.f32 q12, q1, d1[0] \n\t"
"vmla.f32 q13, q1, d1[1] \n\t"
"end_kc2_%=: \n\t"
"cmp %[mc], #4 \n\t"
"bne temp_%= \n\t"
"cmp %[nc], #4 \n\t"
"bne temp_%= \n\t"
"vmov.f32 d8[0], %[alpha] \n\t"
"vmov.f32 d8[1], %[beta] \n\t"
"cmp %[flag_alpha], #1 \n\t"
"bne alpha_%= \n\t"
"alpha_%=: \n\t"
"vmul.f32 q10, q10, d8[0] \n\t"
"vmul.f32 q11, q11, d8[0] \n\t"
"vmul.f32 q12, q12, d8[0] \n\t"
"vmul.f32 q13, q13, d8[0] \n\t"
"beta_%=: \n\t"
"cmp %[flag_beta], #0 \n\t"
"beq memory_%= \n\t"
"mov r4, %[C] \n\t"
"mov r6, %[bytes_ldc]\n\t"
"vld1.32 {q0}, [r4], r6 \n\t"
"vld1.32 {q1}, [r4], r6 \n\t"
"vld1.32 {q2}, [r4], r6 \n\t"
"vld1.32 {q3}, [r4] \n\t"
"cmp %[flag_beta], #1 \n\t"
"beq beta_eq1_%= \n\t"
"bne beta_ne1_%= \n\t"
"beta_eq1_%=: \n\t"
"vadd.f32 q10, q10, q0 \n\t"
"vadd.f32 q11, q11, q1 \n\t"
"vadd.f32 q12, q12, q2 \n\t"
"vadd.f32 q13, q13, q3 \n\t"
"b memory_%= \n\t"
"beta_ne1_%=: \n\t"
"vmla.f32 q10, q0, d8[1] \n\t"
"vmla.f32 q11, q1, d8[1] \n\t"
"vmla.f32 q12, q2, d8[1] \n\t"
"vmla.f32 q13, q3, d8[1] \n\t"
"memory_%=: \n\t"
"mov r5, %[C] \n\t"
"mov r6, %[bytes_ldc]\n\t"
"vst1.32 {q10}, [r5], r6 \n\t"
"vst1.32 {q11}, [r5], r6 \n\t"
"vst1.32 {q12}, [r5], r6 \n\t"
"vst1.32 {q13}, [r5] \n\t"
"b end_%= \n\t"
"temp_%=: \n\t"
"vst1.32 {q10, q11}, [%[ab]]!\n\t"
"vst1.32 {q12, q13}, [%[ab]] \n\t"
"end_%=: \n\t"
:
: [a] "r"(a), [b] "r"(b), [C] "r"(C), [ab] "r"(ab), [kc1] "r"(kc1),
[kc2] "r"(kc2), [mc] "r"(mc), [nc] "r"(nc), [alpha] "r"(alpha),
[beta] "r"(beta), [bytes_ldc] "r"(bytes_ldc),
[flag_alpha] "r"(flag_alpha), [flag_beta] "r"(flag_beta)
: "memory", "q0", "q1", "q2", "q3", "q4", "q10", "q11", "q12", "q13");
if (mc != MR || nc != NR) {
int i, j;
for (i = 0; i < mc; ++i) {
for (j = 0; j < nc; ++j) {
if (beta == 0.0) {
if (alpha != 1.0) {
C(i, j) = alpha * ab[i * MR + j];
} else {
C(i, j) = ab[i * MR + j];
}
} else {
if (beta != 1.0) {
C(i, j) *= beta;
}
if (alpha != 1.0) {
C(i, j) += alpha * ab[i * MR + j];
} else {
C(i, j) += ab[i * MR + j];
}
}
}
}
}
}
#else
void AddDot4x4(int k, float alpha, const float *a, int lda, const float *b,
int ldb, float beta, float *C, int ldc, int mc, int nc) {
float c[16] = {0};
float reg_a0, reg_a1, reg_a2, reg_a3, reg_b0, reg_b1, reg_b2, reg_b3;
// // init C
// float32x4_t cv0 = vdup_n_f32(0.0);
// float32x4_t cv1 = vdup_n_f32(0.0);
// float32x4_t cv2 = vdup_n_f32(0.0);
// float32x4_t cv3 = vdup_n_f32(0.0);
for (int p = 0; p < k; p += 1) {
reg_b0 = *b++;
reg_b1 = *b++;
......@@ -232,6 +416,7 @@ void AddDot4x4(int k, float alpha, const float *a, int lda, const float *b,
}
}
}
#endif
// 32位 float 矩阵乘法
void sgemm(int m, int n, int k, float alpha, const float *A, int lda,
......
src/operators/math/pool3x3.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef POOL_OP
#pragma once
#if __ARM_NEON
......@@ -25,3 +27,5 @@ static void Pool3x3Max() {
static void Pool3x3Avg() {
// todo impl with neon
}
#endif
src/operators/math/pool_2x2.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef POOL_OP
#pragma once
#if __ARM_NEON
......@@ -25,3 +27,5 @@ static void Pool2x2Max() {
static void Pool2x2Avg() {
// todo impl with neon
}
#endif
src/operators/math/pooling.cpp
浏览文件 @ 035c72f2
......@@ -12,8 +12,10 @@ 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. */
#ifdef POOL_OP
#include "pooling.h"
#include <common/types.h>
#include "common/types.h"
namespace paddle_mobile {
namespace operators {
......@@ -57,7 +59,7 @@ class PoolFunctor<CPU, PoolProcess, T> {
T *output_data = output->mutable_data<T>();
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 ph = 0; ph < output_height; ++ph) {
int hstart = ph * stride_height - padding_height;
......@@ -91,3 +93,5 @@ template class PoolFunctor<CPU, math::MaxPool<float>, float>;
} // namespace math
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/math/pooling.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef POOL_OP
#pragma once
#include "common/log.h"
......@@ -64,3 +66,5 @@ class PoolFunctor {
}
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/math/softmax.cpp
浏览文件 @ 035c72f2
......@@ -11,6 +11,9 @@ 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. */
#ifdef SOFTMAX_OP
#include "operators/math/softmax.h"
#include "common/types.h"
#if __ARM_NEON
......@@ -153,3 +156,4 @@ template class SoftmaxFuntor<CPU, float>;
} // namespace math
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/math/softmax.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,7 @@ 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. */
#ifdef SOFTMAX_OP
#pragma once
#include "framework/tensor.h"
namespace paddle_mobile {
......@@ -26,3 +27,4 @@ class SoftmaxFuntor {
} // namespace math
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/mul_op.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef MUL_OP
#include "mul_op.h"
namespace paddle_mobile {
......@@ -55,3 +57,5 @@ template class MulOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(mul);
REGISTER_OPERATOR(mul, ops::MulOp);
#endif
src/operators/mul_op.h
浏览文件 @ 035c72f2
......@@ -11,6 +11,9 @@ 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. */
#ifdef MUL_OP
#pragma once
#include <string>
......@@ -25,7 +28,7 @@ template <typename DeviceType, typename T>
class MulOp : public framework::OperatorWithKernel<DeviceType> {
public:
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)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
scope),
......@@ -45,3 +48,5 @@ class MulOp : public framework::OperatorWithKernel<DeviceType> {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/multiclass_nms_op.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef MULTICLASSNMS_OP
#include "operators/multiclass_nms_op.h"
namespace paddle_mobile {
namespace operators {
......@@ -39,3 +41,5 @@ template class MultiClassNMSOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(multiclass_nms);
REGISTER_OPERATOR(multiclass_nms, ops::MultiClassNMSOp);
#endif
src/operators/multiclass_nms_op.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef MULTICLASSNMS_OP
#pragma once
#include <string>
......@@ -30,7 +32,7 @@ class MultiClassNMSOp : public framework::OperatorWithKernel<DeviceType> {
public:
MultiClassNMSOp(const std::string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs,
const framework::AttributeMap attrs,
const framework::AttributeMap &attrs,
std::shared_ptr<framework::Scope> scope)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
scope),
......@@ -50,3 +52,5 @@ class MultiClassNMSOp : public framework::OperatorWithKernel<DeviceType> {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/op_param.cpp
浏览文件 @ 035c72f2
......@@ -13,9 +13,10 @@ See the License for the specific language governing permissions and
limitations under the License. */
#include "op_param.h"
namespace paddle_mobile {
namespace operators {
#ifdef CONV_OP
Print &operator<<(Print &printer, const ConvParam &conv_param) {
printer << "parameter of conv: "
<< "\n";
......@@ -36,5 +37,7 @@ Print &operator<<(Print &printer, const ConvParam &conv_param) {
printer << " output dims: " << conv_param.Output()->dims();
return printer;
}
#endif
} // namespace operators
} // namespace paddle_mobile
src/operators/op_param.h
浏览文件 @ 035c72f2
......@@ -34,7 +34,7 @@ using framework::Tensor;
using std::string;
using std::vector;
class OpParam : PaddleMobileObject {
class OpParam {
protected:
template <typename T>
static T *InputFrom(const VariableNameMap &inputs, const Scope &scope) {
......@@ -167,8 +167,6 @@ class OpParam : PaddleMobileObject {
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 {
......@@ -191,6 +189,7 @@ class OpParam : PaddleMobileObject {
}
};
#ifdef CONV_OP
class ConvParam : OpParam {
public:
ConvParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
......@@ -230,7 +229,9 @@ class ConvParam : OpParam {
};
Print &operator<<(Print &printer, const ConvParam &conv_param);
#endif
#ifdef ELEMENTWISEADD_OP
class ElementwiseAddParam : OpParam {
public:
ElementwiseAddParam(const VariableNameMap &inputs,
......@@ -258,6 +259,9 @@ class ElementwiseAddParam : OpParam {
int axis_;
};
#endif
#ifdef MUL_OP
class MulParam : OpParam {
public:
MulParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
......@@ -287,7 +291,9 @@ class MulParam : OpParam {
int x_num_col_dims_;
int y_num_col_dims_;
};
#endif
#ifdef CONCAT_OP
class ConcatParam : public OpParam {
public:
ConcatParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
......@@ -309,7 +315,9 @@ class ConcatParam : public OpParam {
Tensor *out_;
int axis_;
};
#endif
#ifdef LRN_OP
class LrnParam : public OpParam {
public:
LrnParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
......@@ -351,6 +359,9 @@ class LrnParam : public OpParam {
float k_;
string data_format_;
};
#endif
#ifdef BATCHNORM_OP
class BatchNormParam : OpParam {
public:
BatchNormParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
......@@ -399,6 +410,9 @@ class BatchNormParam : OpParam {
bool is_test_;
string data_format_;
};
#endif
#ifdef POOL_OP
class PoolParam : public OpParam {
public:
PoolParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
......@@ -442,6 +456,9 @@ class PoolParam : public OpParam {
bool gloabal_pooling_ = false;
};
#endif
#ifdef PRIORBOX_OP
class PriorBoxParam : public OpParam {
public:
PriorBoxParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
......@@ -503,7 +520,9 @@ class PriorBoxParam : public OpParam {
float step_h_;
float offset_;
};
#endif
#ifdef BOXCODER_OP
class BoxCoderParam : public OpParam {
public:
BoxCoderParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
......@@ -533,7 +552,9 @@ class BoxCoderParam : public OpParam {
Tensor *output_box_;
std::string code_type_;
};
#endif
#ifdef SOFTMAX_OP
class SoftmaxParam : public OpParam {
public:
SoftmaxParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
......@@ -549,7 +570,9 @@ class SoftmaxParam : public OpParam {
Tensor *input_x_;
Tensor *out_;
};
#endif
#ifdef SIGMOID_OP
class SigmoidParam : public OpParam {
public:
SigmoidParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
......@@ -565,6 +588,9 @@ class SigmoidParam : public OpParam {
Tensor *input_x_;
Tensor *out_;
};
#endif
#ifdef MULTICLASSNMS_OP
class MultiClassNMSParam : public OpParam {
public:
MultiClassNMSParam(const VariableNameMap &inputs,
......@@ -610,6 +636,7 @@ class MultiClassNMSParam : public OpParam {
float nms_eta_;
float score_threshold_;
};
#endif
class FeedParam : public OpParam {
public:
......@@ -646,6 +673,7 @@ class FetchParam : public OpParam {
Tensor *out_;
};
#ifdef TRANSPOSE_OP
class TransposeParam : public OpParam {
public:
TransposeParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
......@@ -666,7 +694,9 @@ class TransposeParam : public OpParam {
Tensor *out_;
vector<int> axis_;
};
#endif
#ifdef RESHAPE_OP
class ReshapeParam : public OpParam {
public:
ReshapeParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
......@@ -695,7 +725,9 @@ class ReshapeParam : public OpParam {
vector<int> shape_;
bool inplace_;
};
#endif
#ifdef RELU_OP
/*
* @b op 层实例化好这个 param 传递给 kernel 层使用
* */
......@@ -715,7 +747,9 @@ class ReluParam : public OpParam {
Tensor *input_x_;
Tensor *out_;
};
#endif
#ifdef FUSION_FC_OP
class FushionFcParam : public OpParam {
public:
FushionFcParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
......@@ -751,6 +785,7 @@ class FushionFcParam : public OpParam {
int y_num_col_dims_;
int axis_;
};
#endif
} // namespace operators
} // namespace paddle_mobile
src/operators/pool_op.cpp
浏览文件 @ 035c72f2
......@@ -12,7 +12,11 @@ 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. */
#ifdef POOL_OP
#include "pool_op.h"
#include "framework/op_proto_maker.h"
#include "framework/op_registry.h"
namespace paddle_mobile {
namespace operators {
......@@ -57,3 +61,5 @@ template class PoolOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(pool2d);
REGISTER_OPERATOR(pool2d, ops::PoolOp);
#endif
src/operators/pool_op.h
浏览文件 @ 035c72f2
......@@ -12,13 +12,16 @@ 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. */
#ifdef POOL_OP
#pragma once
#include <framework/operator.h>
#include <operators/kernel/pool_kernel.h>
#include <operators/op_param.h>
#include <string>
#include "framework/operator.h"
#include "operators/kernel/pool_kernel.h"
#include "operators/op_param.h"
namespace paddle_mobile {
namespace operators {
using framework::AttributeMap;
......@@ -47,3 +50,5 @@ class PoolOp : public OperatorWithKernel<DeviceType> {
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/prior_box_op.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef PRIORBOX_OP
#include "operators/prior_box_op.h"
#include <vector>
namespace paddle_mobile {
......@@ -49,3 +51,5 @@ template class PriorBoxOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(prior_box);
REGISTER_OPERATOR(prior_box, ops::PriorBoxOp);
#endif
src/operators/prior_box_op.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef PRIORBOX_OP
#pragma once
#include <string>
......@@ -30,7 +32,7 @@ class PriorBoxOp : public framework::OperatorWithKernel<DeviceType> {
public:
PriorBoxOp(const std::string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs,
const framework::AttributeMap attrs,
const framework::AttributeMap &attrs,
std::shared_ptr<framework::Scope> scope)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
scope),
......@@ -50,3 +52,5 @@ class PriorBoxOp : public framework::OperatorWithKernel<DeviceType> {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/relu_op.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef RELU_OP
#include "operators/relu_op.h"
namespace paddle_mobile {
namespace operators {
......@@ -33,3 +35,5 @@ template class ReluOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(relu);
REGISTER_OPERATOR(relu, ops::ReluOp);
#endif
src/operators/relu_op.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef RELU_OP
#pragma once
#include <string>
......@@ -32,7 +34,7 @@ class ReluOp : public framework::OperatorWithKernel<DeviceType> {
* @b op 的实例化方法, 需要调用父类的实例化方法, 以及实例化自己的参数结构体
* */
ReluOp(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)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
scope),
......@@ -59,3 +61,5 @@ class ReluOp : public framework::OperatorWithKernel<DeviceType> {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/reshape_op.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef RESHAPE_OP
#include "operators/reshape_op.h"
#include <vector>
namespace paddle_mobile {
......@@ -32,3 +34,5 @@ template class ReshapeOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(reshape);
REGISTER_OPERATOR(reshape, ops::ReshapeOp);
#endif
src/operators/reshape_op.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef RESHAPE_OP
#pragma once
#include <string>
......@@ -29,7 +31,8 @@ template <typename DeviceType, typename T>
class ReshapeOp : public framework::OperatorWithKernel<DeviceType> {
public:
ReshapeOp(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)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
scope),
......@@ -49,3 +52,5 @@ class ReshapeOp : public framework::OperatorWithKernel<DeviceType> {
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/sigmoid_op.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef SIGMOID_OP
#include "operators/sigmoid_op.h"
namespace paddle_mobile {
......@@ -27,3 +29,5 @@ template class SigmoidOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(sigmoid);
REGISTER_OPERATOR(sigmoid, ops::SigmoidOp);
#endif
src/operators/sigmoid_op.h
浏览文件 @ 035c72f2
......@@ -12,12 +12,15 @@ 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. */
#ifdef SIGMOID_OP
#pragma once
#include <framework/operator.h>
#include <operators/op_param.h>
#include <string>
#include "framework/operator.h"
#include "operators/kernel/sigmoid_kernel.h"
#include "operators/op_param.h"
namespace paddle_mobile {
namespace operators {
......@@ -47,3 +50,5 @@ class SigmoidOp : public framework::OperatorWithKernel<DeviceType> {
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/softmax_op.cpp
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef SOFTMAX_OP
#include "operators/softmax_op.h"
namespace paddle_mobile {
......@@ -27,3 +29,5 @@ template class SoftmaxOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(softmax);
REGISTER_OPERATOR(softmax, ops::SoftmaxOp);
#endif
src/operators/softmax_op.h
浏览文件 @ 035c72f2
......@@ -12,12 +12,15 @@ 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. */
#ifdef SOFTMAX_OP
#pragma once
#include <framework/operator.h>
#include <operators/op_param.h>
#include <string>
#include "framework/operator.h"
#include "operators/kernel/softmax_kernel.h"
#include "operators/op_param.h"
namespace paddle_mobile {
namespace operators {
......@@ -47,3 +50,5 @@ class SoftmaxOp : public framework::OperatorWithKernel<DeviceType> {
};
} // namespace operators
} // namespace paddle_mobile
#endif
src/operators/transpose_op.cpp
浏览文件 @ 035c72f2
......@@ -12,9 +12,12 @@ 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. */
#include "operators/transpose_op.h"
#include <common/enforce.h>
#ifdef TRANSPOSE_OP
#include <vector>
#include "common/enforce.h"
#include "operators/transpose_op.h"
namespace paddle_mobile {
namespace operators {
......@@ -51,3 +54,5 @@ template class TransposeOp<CPU, float>;
namespace ops = paddle_mobile::operators;
USE_OP(transpose);
REGISTER_OPERATOR(transpose, ops::TransposeOp);
#endif
src/operators/transpose_op.h
浏览文件 @ 035c72f2
......@@ -12,6 +12,8 @@ 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. */
#ifdef TRANSPOSE_OP
#pragma once
#include <string>
......@@ -30,7 +32,7 @@ class TransposeOp : public framework::OperatorWithKernel<DeviceType> {
public:
TransposeOp(const std::string &type, const VariableNameMap &inputs,
const VariableNameMap &outputs,
const framework::AttributeMap attrs,
const framework::AttributeMap &attrs,
std::shared_ptr<framework::Scope> scope)
: framework::OperatorWithKernel<DeviceType>(type, inputs, outputs, attrs,
scope),
......@@ -50,3 +52,5 @@ class TransposeOp : public framework::OperatorWithKernel<DeviceType> {
} // namespace operators
} // namespace paddle_mobile
#endif
src/platform/data_type.h 已删除 100644 → 0
浏览文件 @ ff0f7f7b
/* Copyright (c) 2018 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. */
#pragma once
#include <string>
#include <typeindex>
#include "framework/program/tensor_desc.h"
namespace paddle_mobile {
namespace framework {
inline VarType_Type ToDataType(std::type_index type) {
/*if (typeid(platform::float16).hash_code() == type.hash_code()) {
return proto::VarType::FP16;
} else */
if (typeid(const float).hash_code() == type.hash_code()) {
// CPPLint complains Using C-style cast. Use
// static_cast<float>() instead
// One fix to this is to replace float with const float because
// typeid(T) == typeid(const T)
// http://en.cppreference.com/w/cpp/language/typeid
return VARTYPE_TYPE_FP32;
} else if (typeid(const double).hash_code() == type.hash_code()) {
return VARTYPE_TYPE_FP64;
} else if (typeid(const int).hash_code() == type.hash_code()) {
return VARTYPE_TYPE_INT32;
} else if (typeid(const int64_t).hash_code() == type.hash_code()) {
return VARTYPE_TYPE_INT64;
} else if (typeid(const bool).hash_code() == type.hash_code()) {
return VARTYPE_TYPE_BOOL;
} else {
// PADDLE_THROW("Not supported");
// std::cout << "Not supported";
}
}
inline std::type_index ToTypeIndex(VarType_Type type) {
switch (type) {
// case proto::VarType::FP16:
// return typeid(platform::float16);
case VARTYPE_TYPE_FP32:
return typeid(float);
case VARTYPE_TYPE_FP64:
return typeid(double);
case VARTYPE_TYPE_INT32:
return typeid(int);
case VARTYPE_TYPE_INT64:
return typeid(int64_t);
case VARTYPE_TYPE_BOOL:
return typeid(bool);
default:
// PADDLE_THROW("Not support type %d", type);
printf("Not support type %d", type);
}
}
template <typename Visitor>
inline void VisitDataType(VarType_Type type, Visitor visitor) {
switch (type) {
// case proto::VarType::FP16:
// visitor.template operator()<platform::float16>();
// break;
case VARTYPE_TYPE_FP32:
visitor.template operator()<float>();
break;
case VARTYPE_TYPE_FP64:
visitor.template operator()<double>();
break;
case VARTYPE_TYPE_INT32:
visitor.template operator()<int>();
break;
case VARTYPE_TYPE_INT64:
visitor.template operator()<int64_t>();
break;
case VARTYPE_TYPE_BOOL:
visitor.template operator()<bool>();
break;
default:
// PADDLE_THROW("Not supported");
printf("Not supported");
}
}
inline std::string DataTypeToString(const VarType_Type type) {
switch (type) {
case VARTYPE_TYPE_FP16:
return "float16";
case VARTYPE_TYPE_FP32:
return "float32";
case VARTYPE_TYPE_FP64:
return "float64";
case VARTYPE_TYPE_INT16:
return "int16";
case VARTYPE_TYPE_INT32:
return "int32";
case VARTYPE_TYPE_INT64:
return "int64";
case VARTYPE_TYPE_BOOL:
return "bool";
default:
// PADDLE_THROW("Not support type %d", type);
printf("Not support type %d", type);
}
}
inline std::ostream &operator<<(std::ostream &out, const VarType_Type &type) {
out << DataTypeToString(type);
return out;
}
} // namespace framework
} // namespace paddle_mobile
test/CMakeLists.txt
浏览文件 @ 035c72f2
set(dir ${CMAKE_CURRENT_SOURCE_DIR})
set(CMAKE_RUNTIME_OUTPUT_DIRECTORY "${dir}/build")
# gen test
ADD_EXECUTABLE(test-conv-op operators/test_cov_op.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-conv-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-mul-op operators/test_mul_op.cpp test_helper.h test_include.h)
target_link_libraries(test-mul-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-elementwiseadd-op operators/test_elementwise_add_op.cpp test_helper.h test_include.h)
target_link_libraries(test-elementwiseadd-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-concat-op operators/test_concat_op.cpp test_helper.h test_include.h)
target_link_libraries(test-concat-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-lrn-op operators/test_lrn_op.cpp test_helper.h test_include.h)
target_link_libraries(test-lrn-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-batchnorm-op operators/test_batchnorm_op.cpp test_helper.h test_include.h)
target_link_libraries(test-batchnorm-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-priorbox-op operators/test_prior_box_op.cpp test_helper.h test_include.h)
target_link_libraries(test-priorbox-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-boxcoder-op operators/test_box_coder_op.cpp test_helper.h test_include.h)
target_link_libraries(test-boxcoder-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-transpose-op operators/test_transpose_op.cpp test_helper.h test_include.h)
target_link_libraries(test-transpose-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-multiclassnms-op operators/test_multiclass_nms_op.cpp test_helper.h test_include.h)
target_link_libraries(test-multiclassnms-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-reshape-op operators/test_reshape_op.cpp test_helper.h test_include.h)
target_link_libraries(test-reshape-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-relu-op operators/test_relu_op.cpp test_helper.h test_include.h)
target_link_libraries(test-relu-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-fc-op operators/test_fushion_fc_op.cpp test_helper.h test_include.h)
target_link_libraries(test-fc-op paddle-mobile)
# gen test log
ADD_EXECUTABLE(test-log common/test_log.cpp)
target_link_libraries(test-log paddle-mobile)
# gen test log
ADD_EXECUTABLE(test-load framework/test_load.cpp)
target_link_libraries(test-load paddle-mobile)
# gen test log
# gen test
ADD_EXECUTABLE(test-optimize framework/test_optimize.cpp)
target_link_libraries(test-optimize paddle-mobile)
#gen test
ADD_EXECUTABLE(test-pool operators/test_pool_op.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-pool paddle-mobile)
#gen test
ADD_EXECUTABLE(test-softmax operators/test_softmax_op.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-softmax paddle-mobile)
if (googlenet)
# gen test
ADD_EXECUTABLE(test-googlenet net/test_googlenet.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-googlenet paddle-mobile)
elseif (mobilenet)
# gen test
ADD_EXECUTABLE(test-mobilenet net/test_mobilenet.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-mobilenet paddle-mobile)
elseif (yolo)
# gen test
ADD_EXECUTABLE(test-yolo net/test_yolo.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-yolo paddle-mobile)
elseif (squeezenet)
# gen test
ADD_EXECUTABLE(test-squeezenet net/test_squeezenet.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-squeezenet paddle-mobile)
elseif(resnet)
# gen test
ADD_EXECUTABLE(test-resnet net/test_resnet.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-resnet paddle-mobile)
else ()
# gen test
ADD_EXECUTABLE(test-resnet net/test_resnet.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-resnet paddle-mobile)
# gen test
ADD_EXECUTABLE(test-squeezenet net/test_squeezenet.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-squeezenet paddle-mobile)
# gen test
ADD_EXECUTABLE(test-yolo net/test_yolo.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-yolo paddle-mobile)
# gen test
ADD_EXECUTABLE(test-googlenet net/test_googlenet.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-googlenet paddle-mobile)
# gen test
ADD_EXECUTABLE(test-conv-op operators/test_cov_op.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-conv-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-mul-op operators/test_mul_op.cpp test_helper.h test_include.h)
target_link_libraries(test-mul-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-elementwiseadd-op operators/test_elementwise_add_op.cpp test_helper.h test_include.h)
target_link_libraries(test-elementwiseadd-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-concat-op operators/test_concat_op.cpp test_helper.h test_include.h)
target_link_libraries(test-concat-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-lrn-op operators/test_lrn_op.cpp test_helper.h test_include.h)
target_link_libraries(test-lrn-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-batchnorm-op operators/test_batchnorm_op.cpp test_helper.h test_include.h)
target_link_libraries(test-batchnorm-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-priorbox-op operators/test_prior_box_op.cpp test_helper.h test_include.h)
target_link_libraries(test-priorbox-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-boxcoder-op operators/test_box_coder_op.cpp test_helper.h test_include.h)
target_link_libraries(test-boxcoder-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-transpose-op operators/test_transpose_op.cpp test_helper.h test_include.h)
target_link_libraries(test-transpose-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-multiclassnms-op operators/test_multiclass_nms_op.cpp test_helper.h test_include.h)
target_link_libraries(test-multiclassnms-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-reshape-op operators/test_reshape_op.cpp test_helper.h test_include.h)
target_link_libraries(test-reshape-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-relu-op operators/test_relu_op.cpp test_helper.h test_include.h)
target_link_libraries(test-relu-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-fc-op operators/test_fushion_fc_op.cpp test_helper.h test_include.h)
target_link_libraries(test-fc-op paddle-mobile)
# gen test log
ADD_EXECUTABLE(test-log common/test_log.cpp)
target_link_libraries(test-log paddle-mobile)
# gen test log
ADD_EXECUTABLE(test-load framework/test_load.cpp)
target_link_libraries(test-load paddle-mobile)
# gen test log
# gen test
ADD_EXECUTABLE(test-optimize framework/test_optimize.cpp)
target_link_libraries(test-optimize paddle-mobile)
#gen test
ADD_EXECUTABLE(test-pool operators/test_pool_op.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-pool paddle-mobile)
#gen test
ADD_EXECUTABLE(test-softmax operators/test_softmax_op.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-softmax paddle-mobile)
# gen test
ADD_EXECUTABLE(test-gemm common/test_gemm.cpp)
target_link_libraries(test-gemm paddle-mobile)
# gen test
ADD_EXECUTABLE(test-gemm common/test_gemm.cpp)
target_link_libraries(test-gemm paddle-mobile)
# gen test
ADD_EXECUTABLE(test-enforce common/test_enforce.cpp)
target_link_libraries(test-enforce paddle-mobile)
# gen test
ADD_EXECUTABLE(test-enforce common/test_enforce.cpp)
target_link_libraries(test-enforce paddle-mobile)
# gen test - test if openmp works
ADD_EXECUTABLE(test-openmp common/test_openmp.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-openmp paddle-mobile)
# gen test
ADD_EXECUTABLE(test-yolo net/test_yolo.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-yolo paddle-mobile)
# gen test
ADD_EXECUTABLE(test-mobilenetssd net/test_mobilenet+ssd.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-mobilenetssd paddle-mobile)
# gen test
ADD_EXECUTABLE(test-googlenet net/test_googlenet.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-googlenet paddle-mobile)
# gen test
ADD_EXECUTABLE(test-sigmoid operators/test_sigmoid_op.cpp test_include.h)
target_link_libraries(test-sigmoid paddle-mobile)
# gen test
ADD_EXECUTABLE(test-mobilenet net/test_mobilenet.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-mobilenet paddle-mobile)
# gen test
ADD_EXECUTABLE(test-depthwise-conv-op operators/test_depthwise_conv_op.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-depthwise-conv-op paddle-mobile)
# gen test
ADD_EXECUTABLE(test-resnet net/test_resnet.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-resnet paddle-mobile)
# gen test
ADD_EXECUTABLE(test-mobilenetssd net/test_mobilenet+ssd.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-mobilenetssd paddle-mobile)
# gen test
ADD_EXECUTABLE(test-squeezenet net/test_squeezenet.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-squeezenet paddle-mobile)
# gen test
ADD_EXECUTABLE(test-sigmoid operators/test_sigmoid_op.cpp test_include.h)
target_link_libraries(test-sigmoid paddle-mobile)
# gen test
ADD_EXECUTABLE(test-depthwise-conv-op operators/test_depthwise_conv_op.cpp test_helper.h test_include.h executor_for_test.h)
target_link_libraries(test-depthwise-conv-op paddle-mobile)
endif()
src/framework/program/tensor_desc.cpp → test/common/test_lib_size.cpp
浏览文件 @ 035c72f2
......@@ -13,7 +13,9 @@ See the License for the specific language governing permissions and
limitations under the License. */
//
// Created by liuRuiLong on 2018/5/26.
// Created by liuRuiLong on 2018/6/6.
//
#include "tensor_desc.h"
#include "test_lib_size.h"
static test_lib_size t;
test/common/test_lib_size.h 0 → 100644
浏览文件 @ 035c72f2
/* Copyright (c) 2018 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. */
//
// Created by liuRuiLong on 2018/6/6.
//
#ifndef PADDLE_MOBILE_TEST_LIB_SIZE_H
#define PADDLE_MOBILE_TEST_LIB_SIZE_H
#include <vector>
//#include <list>
//#include <tuple>
//#include <typeinfo>
//#include <mutex>
//#include <initializer_list>
//#include <map>
//#include <string>
//#include <unordered_map>
//#include <unordered_set>
//#include <algorithm>
//#include <iostream>
//#include <sstream>
#include <memory>
//#include <stdio.h>
//#include <cstring>
void foo() {
// char *str = "1234";
// char dst[10];
// strcpy(dst, str);
// std::cout << "12345" << std::endl;
std::vector<int> vec = {1, 2, 3, 4, 5};
// std::find(vec.begin(), vec.end(), 1);
// std::find(vec.begin(), vec.end(), 1);
// std::list<int> l;
// std::mutex mutex_;
// std::map<int, float> m;
// std::unordered_map<int, float> u_m;
// std::unordered_set<int> u_s;
// std::string ss = "12345";
// printf("%f", ss.c_str());
// std::initializer_list<int> init_list = {1, 2};
// std::tuple<int, int> t = {1, 2};
// std::tuple_element<I, std::tuple<ARGS...>>::type
// std::tuple<>
// int i;
// int j;
// if (typeid(i) == typeid(j)){
// int z = 10;
// }
std::shared_ptr<int> s1 = std::make_shared<int>();
// std::stringstream ss;
// ss << "12345";
}
class test_lib_size {
public:
test_lib_size() {}
// std::shared_ptr<int> Test(){
// std::vector<int> vec = {1, 2, 3};
// std::shared_ptr<int> si = std::make_shared<int>();
// return si;
// }
// void test(){
// int i = 9;
// }
};
#endif // PADDLE_MOBILE_TEST_LIB_SIZE_H
src/platform/hostdevice.h → test/common/test_openmp.cpp
浏览文件 @ 035c72f2
......@@ -12,14 +12,18 @@ 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. */
#pragma once
//#include <omp.h>
#include <iostream>
#ifdef __CUDACC__
#define HOSTDEVICE __host__ __device__
#define DEVICE __device__
#define HOST __host__
#else
#define HOSTDEVICE
#define DEVICE
#define HOST
int main(void) {
#ifdef PADDLE_MOBILE_USE_OPENMP
#pragma omp parallel num_threads(2)
{
// int thread_id = omp_get_thread_num();
// int nthreads = omp_get_num_threads();
// std::cout << "Hello, OMP " << thread_id << "/" << nthreads <<
// "\n";
}
#endif
return 0;
}
test/executor_for_test.h
浏览文件 @ 035c72f2
......@@ -19,7 +19,7 @@ limitations under the License. */
#include "common/log.h"
#include "framework/op_registry.h"
#include "io.h"
#include "io/io.h"
#include "operators/conv_op.h"
#include "operators/elementwise_add_op.h"
#include "operators/pool_op.h"
......
test/framework/test_load.cpp
浏览文件 @ 035c72f2
......@@ -13,13 +13,16 @@ See the License for the specific language governing permissions and
limitations under the License. */
#include "../test_helper.h"
#include "io.h"
#include "io/io.h"
int main() {
paddle_mobile::Loader<paddle_mobile::CPU> loader;
// ../../../test/models/googlenet
// ../../../test/models/mobilenet
auto program = loader.Load(g_googlenet);
program.optimizeProgram->Description("program desc: ");
auto program = loader.Load(g_resnet, true);
loader.Load(g_googlenet_combine + "/model", g_googlenet_combine + "/params",
true);
program.originProgram->Description("program desc: ");
return 0;
}
test/framework/test_optimize.cpp
浏览文件 @ 035c72f2
......@@ -15,17 +15,17 @@ limitations under the License. */
#include "../test_helper.h"
#include "framework/program/program-optimize/node.h"
#include "framework/program/program-optimize/program_optimize.h"
#include "io.h"
#include "io/io.h"
int main() {
paddle_mobile::Loader<paddle_mobile::CPU> loader;
// "../../../test/models/googlenet"
auto program = loader.Load(g_googlenet);
auto program = loader.Load(g_mobilenet_ssd, true);
paddle_mobile::framework::ProgramOptimize optimize;
// program.originProgram->Description("origin");
auto optimize_program = optimize.FushionOptimize(program.originProgram);
if (optimize_program != nullptr) {
optimize_program->Description("optimize");
// optimize_program->Description("optimize");
} else {
LOG(paddle_mobile::kLOG_ERROR) << "optimize_program is null";
}
......
test/net/test_googlenet.cpp
浏览文件 @ 035c72f2
......@@ -18,9 +18,11 @@ limitations under the License. */
int main() {
paddle_mobile::Loader<paddle_mobile::CPU> loader;
bool optimize = true;
bool optimize = false;
auto time1 = time();
auto program = loader.Load(g_googlenet, optimize);
// auto program = loader.Load(g_googlenet, optimize);
auto program = loader.Load(g_googlenet_combine + "/model",
g_googlenet_combine + "/params", optimize);
auto time2 = time();
DLOG << "load cost :" << time_diff(time1, time2) << "ms\n";
paddle_mobile::Executor<paddle_mobile::CPU> executor(program, 1, optimize);
......
test/net/test_mobilenet+ssd.cpp
浏览文件 @ 035c72f2
......@@ -19,10 +19,10 @@ limitations under the License. */
int main() {
paddle_mobile::Loader<paddle_mobile::CPU> loader;
auto time1 = time();
auto program = loader.Load(g_mobilenet_ssd, false);
auto program = loader.Load(g_mobilenet_ssd, true);
auto time2 = time();
DLOG << "load cost :" << time_diff(time1, time1) << "ms";
paddle_mobile::Executor<paddle_mobile::CPU> executor(program, 1, false);
paddle_mobile::Executor<paddle_mobile::CPU> executor(program, 1, true);
std::vector<int64_t> dims{1, 3, 300, 300};
Tensor input_tensor;
......
test/operators/test_concat_op.cpp
浏览文件 @ 035c72f2
......@@ -12,7 +12,6 @@ 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. */
#include "../executor_for_test.h"
#include "../test_include.h"
#include "operators/concat_op.h"
......
test/operators/test_cov_op.cpp
浏览文件 @ 035c72f2
......@@ -12,8 +12,8 @@ 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. */
#include "../executor_for_test.h"
#include "../test_include.h"
#include "operators/conv_op.h"
int main() {
paddle_mobile::Loader<paddle_mobile::CPU> loader;
......
test/operators/test_depthwise_conv_op.cpp
浏览文件 @ 035c72f2
......@@ -12,7 +12,6 @@ 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. */
#include "../executor_for_test.h"
#include "../test_include.h"
#include "operators/depthwise_conv_op.h"
......
test/operators/test_elementwise_add_op.cpp
浏览文件 @ 035c72f2
......@@ -12,7 +12,6 @@ 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. */
#include "../executor_for_test.h"
#include "../test_include.h"
int main() {
......
test/operators/test_lrn_op.cpp
浏览文件 @ 035c72f2
......@@ -12,7 +12,6 @@ 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. */
#include "../executor_for_test.h"
#include "../test_include.h"
#include "operators/lrn_op.h"
......
test/operators/test_mul_op.cpp
浏览文件 @ 035c72f2
......@@ -12,7 +12,6 @@ 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. */
#include "../executor_for_test.h"
#include "../test_include.h"
#include "operators/mul_op.h"
......
test/operators/test_pool_op.cpp
浏览文件 @ 035c72f2
......@@ -12,9 +12,8 @@ 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. */
#include "../executor_for_test.h"
#include "../test_helper.h"
#include "io.h"
#include "../test_include.h"
#include "operators/pool_op.h"
int main() {
paddle_mobile::Loader<paddle_mobile::CPU> loader;
......
test/operators/test_relu_op.cpp
浏览文件 @ 035c72f2
......@@ -12,7 +12,6 @@ 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. */
#include "../executor_for_test.h"
#include "../test_include.h"
#include "operators/relu_op.h"
......
test/operators/test_reshape_op.cpp
浏览文件 @ 035c72f2
......@@ -12,9 +12,8 @@ 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. */
#include "../executor_for_test.h"
#include "../test_helper.h"
#include "io.h"
#include "../test_include.h"
#include "operators/reshape_op.h"
int main() {
paddle_mobile::Loader<paddle_mobile::CPU> loader;
......
test/operators/test_sigmoid_op.cpp
浏览文件 @ 035c72f2
......@@ -14,7 +14,7 @@ limitations under the License. */
#include "../../src/operators/kernel/sigmoid_kernel.h"
#include "../test_helper.h"
#include "io.h"
#include "io/io.h"
int main() {
paddle_mobile::framework::Tensor input;
......
test/operators/test_softmax_op.cpp
浏览文件 @ 035c72f2
......@@ -12,9 +12,9 @@ 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. */
#include "../executor_for_test.h"
#include "../test_helper.h"
#include "io.h"
#include "../test_include.h"
#include "operators/softmax_op.h"
int main() {
paddle_mobile::Loader<paddle_mobile::CPU> loader;
......
test/operators/test_transpose_op.cpp
浏览文件 @ 035c72f2
......@@ -12,10 +12,9 @@ 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. */
#include "../executor_for_test.h"
#include "../test_helper.h"
#include "io.h"
#include "../test_include.h"
#include "operators/transpose_op.h"
int main() {
paddle_mobile::Loader<paddle_mobile::CPU> loader;
auto program = loader.Load(std::string(g_mobilenet_ssd));
......
test/test_helper.h
浏览文件 @ 035c72f2
......@@ -22,12 +22,13 @@ limitations under the License. */
#include "framework/ddim.h"
#include "framework/tensor.h"
static const std::string g_googlenet = "../models/googlenet";
static const std::string g_mobilenet = "../models/mobilenet";
static const std::string g_mobilenet_ssd = "../models/mobilenet+ssd";
static const std::string g_squeezenet = "../models/squeezenet";
static const std::string g_resnet =
"../models/image_classification_resnet.inference.model";
static const std::string g_googlenet = "../models/googlenet";
static const std::string g_mobilenet = "../models/mobilenet";
static const std::string g_resnet_50 = "../models/resnet_50";
static const std::string g_resnet = "../models/resnet";
static const std::string g_googlenet_combine = "../models/googlenet_combine";
static const std::string g_yolo = "../models/yolo";
static const std::string g_test_image_1x3x224x224 =
"../images/test_image_1x3x224x224_float";
......
test/test_include.h
浏览文件 @ 035c72f2
......@@ -21,6 +21,7 @@ limitations under the License. */
#include "./test_helper.h"
#include "common/enforce.h"
#include "common/log.h"
#include "executor_for_test.h"
#include "framework/lod_tensor.h"
#include "framework/operator.h"
#include "framework/program/block_desc.h"
......@@ -29,4 +30,4 @@ limitations under the License. */
#include "framework/scope.h"
#include "framework/tensor.h"
#include "framework/variable.h"
#include "io.h"
#include "io/io.h"
build.sh → tools/build.sh
浏览文件 @ 035c72f2
......@@ -16,13 +16,13 @@ build_for_mac() {
PLATFORM="x86"
MODE="Release"
CXX_FLAGS="-std=c++11 -O3 -s"
BUILD_DIR=build/release/"${PLATFORM}"
BUILD_DIR=../build/release/"${PLATFORM}"
mkdir -p ${BUILD_DIR}/build
mkdir -p ${BUILD_DIR}/test
cp -r test/models ${BUILD_DIR}/test/models
cp -r ../test/models ${BUILD_DIR}/test/models
cmake . \
cmake .. \
-B"${BUILD_DIR}" \
-DCMAKE_BUILD_TYPE="${MODE}" \
-DCMAKE_CXX_FLAGS="${CXX_FLAGS}" \
......@@ -33,18 +33,20 @@ build_for_mac() {
}
build_for_android() {
rm -rf "../build"
if [ -z "${ANDROID_NDK}" ]; then
echo "ANDROID_NDK not found!"
exit -1
fi
PLATFORM="arm-v7a"
# PLATFORM="arm-v8a"
if [ -z "$PLATFORM" ]; then
PLATFORM="arm-v7a" # Users could choose "arm-v8a" or other platforms from the command line.
fi
if [ "${PLATFORM}" = "arm-v7a" ]; then
ABI="armeabi-v7a with NEON"
ARM_PLATFORM="V7"
CXX_FLAGS="-O3 -std=c++11 -s -march=armv7-a -mfpu=neon -mfloat-abi=softfp -pie -fPIE -w -Wno-error=format-security -llog"
CXX_FLAGS="-O3 -std=c++11 -s -march=armv7-a -mfpu=neon -mfloat-abi=softfp -pie -fPIE -w -Wno-error=format-security"
elif [ "${PLATFORM}" = "arm-v8a" ]; then
ABI="arm64-v8a"
ARM_PLATFORM="V8"
......@@ -54,13 +56,15 @@ build_for_android() {
exit -1
fi
MODE="Release"
ANDROID_PLATFORM_VERSION="android-15"
TOOLCHAIN_FILE="./tools/android-cmake/android.toolchain.cmake"
ANDROID_ARM_MODE="arm"
cmake . \
-B"build/release/${PLATFORM}" \
if [ $# -eq 1 ]; then
NET=$1
cmake .. \
-B"../build/release/${PLATFORM}" \
-DANDROID_ABI="${ABI}" \
-DCMAKE_BUILD_TYPE="${MODE}" \
-DCMAKE_TOOLCHAIN_FILE="${TOOLCHAIN_FILE}" \
......@@ -68,30 +72,55 @@ build_for_android() {
-DCMAKE_CXX_FLAGS="${CXX_FLAGS}" \
-DANDROID_STL=c++_static \
-DANDROID=true \
-D"${NET}=true" \
-D"${ARM_PLATFORM}"=true
else
cd "./build/release/${PLATFORM}"
cmake .. \
-B"../build/release/${PLATFORM}" \
-DANDROID_ABI="${ABI}" \
-DCMAKE_BUILD_TYPE="${MODE}" \
-DCMAKE_TOOLCHAIN_FILE="${TOOLCHAIN_FILE}" \
-DANDROID_PLATFORM="${ANDROID_PLATFORM_VERSION}" \
-DCMAKE_CXX_FLAGS="${CXX_FLAGS}" \
-DANDROID_STL=c++_static \
-DANDROID=true \
-D"${ARM_PLATFORM}"=true
fi
cd "../build/release/${PLATFORM}"
make -j 8
}
build_for_ios() {
rm -rf "../build"
PLATFORM="ios"
MODE="Release"
BUILD_DIR=build/release/"${PLATFORM}"
BUILD_DIR=../build/release/"${PLATFORM}"
TOOLCHAIN_FILE="./tools/ios-cmake/ios.toolchain.cmake"
C_FLAGS="-fobjc-abi-version=2 -fobjc-arc -isysroot ${CMAKE_OSX_SYSROOT}"
CXX_FLAGS="-fobjc-abi-version=2 -fobjc-arc -std=gnu++11 -stdlib=libc++ -isysroot ${CMAKE_OSX_SYSROOT}"
mkdir -p "${BUILD_DIR}"
cmake . \
-B"${BUILD_DIR}" \
-DCMAKE_BUILD_TYPE="${MODE}" \
-DCMAKE_TOOLCHAIN_FILE="${TOOLCHAIN_FILE}" \
-DIOS_PLATFORM=OS \
-DCMAKE_C_FLAGS="${C_FLAGS}" \
-DCMAKE_CXX_FLAGS="${CXX_FLAGS}" \
-DIS_IOS="true" \
if [ $# -eq 1 ]; then
NET=$1
cmake .. \
-B"${BUILD_DIR}" \
-DCMAKE_BUILD_TYPE="${MODE}" \
-DCMAKE_TOOLCHAIN_FILE="${TOOLCHAIN_FILE}" \
-DIOS_PLATFORM=OS \
-DCMAKE_C_FLAGS="${C_FLAGS}" \
-DCMAKE_CXX_FLAGS="${CXX_FLAGS}" \
-D"${NET}"=true \
-DIS_IOS="true"
else
cmake .. \
-B"${BUILD_DIR}" \
-DCMAKE_BUILD_TYPE="${MODE}" \
-DCMAKE_TOOLCHAIN_FILE="${TOOLCHAIN_FILE}" \
-DIOS_PLATFORM=OS \
-DCMAKE_C_FLAGS="${C_FLAGS}" \
-DCMAKE_CXX_FLAGS="${CXX_FLAGS}" \
-DIS_IOS="true"
fi
cd "${BUILD_DIR}"
make -j 8
}
......@@ -105,15 +134,43 @@ if [ $# -lt 1 ]; then
echo "available targets: mac|linux|ios|android"
echo "sample usage: ./build.sh mac"
else
if [ $1 = "mac" ]; then
build_for_mac
elif [ $1 = "linux" ]; then
build_for_linux
elif [ $1 = "android" ]; then
build_for_android
elif [ $1 = "ios" ]; then
build_for_ios
else
build_error
if [ $# -eq 2 ]; then
if [ $2 != "googlenet" -a $2 != "mobilenet" -a $2 != "yolo" -a $2 != "squeezenet" -a $2 != "resnet" ]; then
if [ $1 = "mac" ]; then
build_for_mac
elif [ $1 = "linux" ]; then
build_for_linux
elif [ $1 = "android" ]; then
build_for_android
elif [ $1 = "ios" ]; then
build_for_ios
else
build_error
fi
else
if [ $1 = "mac" ]; then
build_for_mac $2
elif [ $1 = "linux" ]; then
build_for_linux $2
elif [ $1 = "android" ]; then
build_for_android $2
elif [ $1 = "ios" ]; then
build_for_ios $2
else
build_error
fi
fi
else
if [ $1 = "mac" ]; then
build_for_mac
elif [ $1 = "linux" ]; then
build_for_linux
elif [ $1 = "android" ]; then
build_for_android
elif [ $1 = "ios" ]; then
build_for_ios
else
build_error
fi
fi
fi
tools/pre-commit.hooks/clang-format.hook
浏览文件 @ 035c72f2
......@@ -12,4 +12,8 @@ if ! [[ $version == *"$VERSION"* ]]; then
exit -1
fi
clang-format $@
# https://medicineyeh.wordpress.com/2017/07/13/clang-format-with-pragma/
shift
perl -i -pe 's|#pragma\s+omp|// <TRICKY-CLANG-FORMAT-PRAGMA-FIX> #pragma omp|' "$@"
clang-format -i $@
perl -i -pe 's|// <TRICKY-CLANG-FORMAT-PRAGMA-FIX> ||' "$@"
scripts/push2android.sh → tools/scripts/push2android.sh
浏览文件 @ 035c72f2
#!/usr/bin/env sh
push_fn () {
MODELS_PATH="../test/models/*"
EXE_FILE="../test/build/*"
MODELS_PATH="../../test/models/*"
IMAGE_PATH="../../test/images/*"
EXE_FILE="../../test/build/*"
EXE_DIR="data/local/tmp/bin"
MODELS_DIR="data/local/tmp/models"
LIB_PATH="../build/release/arm-v7a/build/*"
IMAGES_DIR="data/local/tmp/images"
LIB_PATH="../../build/release/arm-v7a/build/*"
adb push ${EXE_FILE} ${EXE_DIR}
adb push ${LIB_PATH} ${EXE_DIR}
adb push ${IMAGE_PATH} ${IMAGES_DIR}
adb push ${MODELS_PATH} ${MODELS_DIR}
echo "test files sync completed"
}
......
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