Skip to content

P
Paddle

BaiXuePrincess / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
未验证 提交 be273ea9 编写于 作者: L Li-fAngyU 提交者： GitHub
浏览文件
操作

fix build warning: [Wsign-compare] on linux (#46644)

上级 ddf317ed
展开全部 隐藏空白更改
内联 并排
Showing with 244 addition and 232 deletion
paddle/fluid/distributed/ps/table/memory_dense_table.cc
浏览文件 @ be273ea9
......@@ -21,8 +21,8 @@ namespace distributed {
int FLAGS_pslib_table_save_max_retry_dense = 3;
void MemoryDenseTable::CreateInitializer(const std::string& attr,
const std::string& name) {
void MemoryDenseTable::CreateInitializer(const std::string &attr,
const std::string &name) {
auto slices = string::split_string<std::string>(attr, "&");
if (slices[0] == "gaussian_random") {
......@@ -60,14 +60,14 @@ int32_t MemoryDenseTable::InitializeValue() {
values_.resize(size);
total_dim_ = 0;
for (int x = 0; x < size; ++x) {
auto& varname = common.params()[x];
auto& dim = common.dims()[x];
auto &varname = common.params()[x];
auto &dim = common.dims()[x];
if (varname == "Param") {
param_dim_ = dim;
param_idx_ = x;
}
auto& initializer = common.initializers()[x];
auto &initializer = common.initializers()[x];
total_dim_ += dim;
CreateInitializer(initializer, varname);
......@@ -81,7 +81,7 @@ int32_t MemoryDenseTable::InitializeValue() {
fixed_len_params_dim_ = 0;
for (int x = 0; x < size; ++x) {
auto& dim = common.dims()[x];
auto &dim = common.dims()[x];
if (static_cast<int>(dim) != param_dim_) {
fixed_len_params_dim_ += dim;
} else {
......@@ -124,19 +124,19 @@ int32_t MemoryDenseTable::InitializeOptimizer() {
return 0;
}
int32_t MemoryDenseTable::SetGlobalLR(float* lr) {
int32_t MemoryDenseTable::SetGlobalLR(float *lr) {
_global_lr = lr;
optimizer_->SetGlobalLR(_global_lr);
return 0;
}
int32_t MemoryDenseTable::Pull(TableContext& context) {
int32_t MemoryDenseTable::Pull(TableContext &context) {
CHECK(context.value_type == Dense);
float* pull_values = context.pull_context.values;
float *pull_values = context.pull_context.values;
return PullDense(pull_values, context.num);
}
int32_t MemoryDenseTable::Push(TableContext& context) {
int32_t MemoryDenseTable::Push(TableContext &context) {
CHECK(context.value_type == Dense);
if (context.push_context.values != nullptr) {
if (!context.push_context.is_param) {
......@@ -148,13 +148,13 @@ int32_t MemoryDenseTable::Push(TableContext& context) {
return 0;
}
int32_t MemoryDenseTable::PullDense(float* pull_values, size_t num) {
int32_t MemoryDenseTable::PullDense(float *pull_values, size_t num) {
std::copy(
values_[param_idx_].begin(), values_[param_idx_].end(), pull_values);
return 0;
}
int32_t MemoryDenseTable::PushDenseParam(const float* values, size_t num) {
int32_t MemoryDenseTable::PushDenseParam(const float *values, size_t num) {
PADDLE_ENFORCE_GE(
num,
param_dim_,
......@@ -171,7 +171,7 @@ int32_t MemoryDenseTable::Pour() {
return 0;
}
int32_t MemoryDenseTable::PushDense(const float* values, size_t num) {
int32_t MemoryDenseTable::PushDense(const float *values, size_t num) {
if (sync) {
std::future<int> task =
_shards_task_pool[0]->enqueue([this, &values]() -> int {
......@@ -185,7 +185,7 @@ int32_t MemoryDenseTable::PushDense(const float* values, size_t num) {
return 0;
}
int32_t MemoryDenseTable::_PushDense(const float* values, size_t num) {
int32_t MemoryDenseTable::_PushDense(const float *values, size_t num) {
PADDLE_ENFORCE_GE(
num,
param_dim_,
......@@ -212,8 +212,8 @@ int32_t MemoryDenseTable::_PushDense(const float* values, size_t num) {
return 0;
}
int32_t MemoryDenseTable::Load(const std::string& path,
const std::string& param) {
int32_t MemoryDenseTable::Load(const std::string &path,
const std::string &param) {
if (param_dim_ <= 0) {
return 0;
}
......@@ -249,7 +249,7 @@ int32_t MemoryDenseTable::Load(const std::string& path,
try {
int dim_idx = 0;
float data_buffer[5];
float* data_buff_ptr = data_buffer;
float *data_buff_ptr = data_buffer;
std::string line_data;
auto common = _config.common();
......@@ -319,8 +319,8 @@ int32_t MemoryDenseTable::Load(const std::string& path,
return 0;
}
int32_t MemoryDenseTable::Save(const std::string& path,
const std::string& param) {
int32_t MemoryDenseTable::Save(const std::string &path,
const std::string &param) {
int save_param = atoi(param.c_str());
uint32_t feasign_size;
VLOG(0) << "MemoryDenseTable::save path " << path;
......@@ -353,7 +353,7 @@ int32_t MemoryDenseTable::Save(const std::string& path,
os.clear();
os.str("");
os << values_[param_col_ids_[0]][y] << " 0";
for (int x = 2; x < param_col_ids_.size(); ++x) {
for (size_t x = 2; x < param_col_ids_.size(); ++x) {
os << " ";
os << values_[param_col_ids_[x]][y];
}
......@@ -365,7 +365,7 @@ int32_t MemoryDenseTable::Save(const std::string& path,
os.clear();
os.str("");
os << values_[param_col_ids_[0]][y];
for (int x = 1; x < param_col_ids_.size(); ++x) {
for (size_t x = 1; x < param_col_ids_.size(); ++x) {
os << " ";
os << values_[param_col_ids_[x]][y];
}
......@@ -383,7 +383,7 @@ int32_t MemoryDenseTable::Save(const std::string& path,
auto write_channel =
_afs_client.open_w(channel_config, 1024 * 1024 * 40, &err_no);
for (auto& t : result_buffer_param) {
for (auto &t : result_buffer_param) {
if (0 != write_channel->write_line(t)) {
++retry_num;
is_write_failed = true;
......
paddle/fluid/distributed/ps/table/sparse_sgd_rule.cc
浏览文件 @ be273ea9
......@@ -23,7 +23,7 @@ DEFINE_bool(enable_show_scale_gradient, true, "enable show scale gradient");
namespace paddle {
namespace distributed {
void SparseNaiveSGDRule::LoadConfig(const SparseCommonSGDRuleParameter& param,
void SparseNaiveSGDRule::LoadConfig(const SparseCommonSGDRuleParameter &param,
size_t emb_dim) {
_embedding_dim = emb_dim;
auto naive_param = param.naive();
......@@ -41,9 +41,9 @@ void SparseNaiveSGDRule::LoadConfig(const SparseCommonSGDRuleParameter& param,
}
}
void SparseNaiveSGDRule::UpdateValueWork(float* w,
float* sgd,
const float* push_value,
void SparseNaiveSGDRule::UpdateValueWork(float *w,
float *sgd,
const float *push_value,
float scale) {
for (size_t i = 0; i < _embedding_dim; ++i) {
w[i] -= learning_rate_ * push_value[i];
......@@ -51,8 +51,8 @@ void SparseNaiveSGDRule::UpdateValueWork(float* w,
}
}
void SparseNaiveSGDRule::InitValueWork(float* value,
float* sgd,
void SparseNaiveSGDRule::InitValueWork(float *value,
float *sgd,
bool zero_init) {
if (zero_init) {
for (size_t i = 0; i < _embedding_dim; ++i) {
......@@ -68,7 +68,7 @@ void SparseNaiveSGDRule::InitValueWork(float* value,
}
}
}
void SparseAdaGradSGDRule::LoadConfig(const SparseCommonSGDRuleParameter& param,
void SparseAdaGradSGDRule::LoadConfig(const SparseCommonSGDRuleParameter &param,
size_t emb_dim) {
_embedding_dim = emb_dim;
auto adagrad_param = param.adagrad();
......@@ -88,11 +88,11 @@ void SparseAdaGradSGDRule::LoadConfig(const SparseCommonSGDRuleParameter& param,
}
}
void SparseAdaGradSGDRule::UpdateValueWork(float* w,
float* sgd,
const float* grad,
void SparseAdaGradSGDRule::UpdateValueWork(float *w,
float *sgd,
const float *grad,
float scale) {
float& g2sum = sgd[G2SumIndex()];
float &g2sum = sgd[G2SumIndex()];
double add_g2sum = 0;
for (size_t i = 0; i < _embedding_dim; i++) {
......@@ -106,8 +106,8 @@ void SparseAdaGradSGDRule::UpdateValueWork(float* w,
g2sum += add_g2sum / _embedding_dim;
}
void SparseAdaGradSGDRule::InitValueWork(float* value,
float* sgd,
void SparseAdaGradSGDRule::InitValueWork(float *value,
float *sgd,
bool zero_init) {
for (size_t i = 0; i < _embedding_dim; ++i) {
if (zero_init) {
......@@ -125,7 +125,7 @@ void SparseAdaGradSGDRule::InitValueWork(float* value,
sgd[G2SumIndex()] = 0;
}
void StdAdaGradSGDRule::LoadConfig(const SparseCommonSGDRuleParameter& param,
void StdAdaGradSGDRule::LoadConfig(const SparseCommonSGDRuleParameter &param,
size_t emb_dim) {
_embedding_dim = emb_dim;
auto adagrad_param = param.adagrad();
......@@ -145,12 +145,12 @@ void StdAdaGradSGDRule::LoadConfig(const SparseCommonSGDRuleParameter& param,
}
}
void StdAdaGradSGDRule::UpdateValueWork(float* w,
float* sgd,
const float* grad,
void StdAdaGradSGDRule::UpdateValueWork(float *w,
float *sgd,
const float *grad,
float scale) {
for (size_t i = 0; i < _embedding_dim; i++) {
float& g2sum = sgd[G2SumIndex() + i];
float &g2sum = sgd[G2SumIndex() + i];
double scaled_grad = grad[i] / scale;
w[i] -= learning_rate_ * scaled_grad *
sqrt(_initial_g2sum / (_initial_g2sum + g2sum));
......@@ -159,8 +159,8 @@ void StdAdaGradSGDRule::UpdateValueWork(float* w,
}
}
void StdAdaGradSGDRule::InitValueWork(float* value,
float* sgd,
void StdAdaGradSGDRule::InitValueWork(float *value,
float *sgd,
bool zero_init) {
for (size_t i = 0; i < _embedding_dim; ++i) {
if (zero_init) {
......@@ -178,7 +178,7 @@ void StdAdaGradSGDRule::InitValueWork(float* value,
}
}
void SparseAdamSGDRule::LoadConfig(const SparseCommonSGDRuleParameter& param,
void SparseAdamSGDRule::LoadConfig(const SparseCommonSGDRuleParameter &param,
size_t emb_dim) {
_embedding_dim = emb_dim;
auto adam_param = param.adam();
......@@ -199,15 +199,15 @@ void SparseAdamSGDRule::LoadConfig(const SparseCommonSGDRuleParameter& param,
}
}
void SparseAdamSGDRule::UpdateValueWork(float* w,
float* sgd,
const float* grad,
void SparseAdamSGDRule::UpdateValueWork(float *w,
float *sgd,
const float *grad,
float scale) {
float* gsum = sgd + GSumIndex();
float* g2sum = sgd + G2SumIndex();
float* beta1_pow = sgd + Beta1PowIndex();
float* beta2_pow = sgd + Beta2PowIndex();
const float* g = grad;
float *gsum = sgd + GSumIndex();
float *g2sum = sgd + G2SumIndex();
float *beta1_pow = sgd + Beta1PowIndex();
float *beta2_pow = sgd + Beta2PowIndex();
const float *g = grad;
float lr = learning_rate_;
float beta1_pow_ = *beta1_pow;
......@@ -227,8 +227,8 @@ void SparseAdamSGDRule::UpdateValueWork(float* w,
(*beta2_pow) *= _beta2_decay_rate;
}
void SparseAdamSGDRule::InitValueWork(float* value,
float* sgd,
void SparseAdamSGDRule::InitValueWork(float *value,
float *sgd,
bool zero_init) {
for (size_t i = 0; i < _embedding_dim; ++i) {
if (zero_init) {
......@@ -253,7 +253,7 @@ void SparseAdamSGDRule::InitValueWork(float* value,
}
void SparseSharedAdamSGDRule::LoadConfig(
const SparseCommonSGDRuleParameter& param, size_t emb_dim) {
const SparseCommonSGDRuleParameter &param, size_t emb_dim) {
_embedding_dim = emb_dim;
auto adam_param = param.adam();
learning_rate_ = adam_param.learning_rate();
......@@ -273,15 +273,15 @@ void SparseSharedAdamSGDRule::LoadConfig(
}
}
void SparseSharedAdamSGDRule::UpdateValueWork(float* w,
float* sgd,
const float* grad,
void SparseSharedAdamSGDRule::UpdateValueWork(float *w,
float *sgd,
const float *grad,
float scale) {
float* gsum = sgd + GSumIndex();
float* g2sum = sgd + G2SumIndex();
float* beta1_pow = sgd + Beta1PowIndex();
float* beta2_pow = sgd + Beta2PowIndex();
const float* g = grad;
float *gsum = sgd + GSumIndex();
float *g2sum = sgd + G2SumIndex();
float *beta1_pow = sgd + Beta1PowIndex();
float *beta2_pow = sgd + Beta2PowIndex();
const float *g = grad;
float lr = learning_rate_;
float beta1_pow_ = *beta1_pow;
......@@ -292,7 +292,7 @@ void SparseSharedAdamSGDRule::UpdateValueWork(float* w,
lr *= sqrt(1 - beta2_pow_) / (1 - beta1_pow_);
double sum_gsum = 0.0;
double sum_g2sum = 0.0;
for (int i = 0; i < _embedding_dim; i++) {
for (size_t i = 0; i < _embedding_dim; i++) {
// Calculation
double new_gsum =
_beta1_decay_rate * gsum_ + (1 - _beta1_decay_rate) * g[i];
......@@ -310,10 +310,10 @@ void SparseSharedAdamSGDRule::UpdateValueWork(float* w,
(*beta2_pow) *= _beta2_decay_rate;
}
void SparseSharedAdamSGDRule::InitValueWork(float* value,
float* sgd,
void SparseSharedAdamSGDRule::InitValueWork(float *value,
float *sgd,
bool zero_init) {
for (int i = 0; i < _embedding_dim; ++i) {
for (size_t i = 0; i < _embedding_dim; ++i) {
if (zero_init) {
value[i] = 0.0;
BoundValue(value[i]);
......@@ -327,7 +327,7 @@ void SparseSharedAdamSGDRule::InitValueWork(float* value,
}
}
// init rule gsum and g2sum
for (int i = GSumIndex(); i < Beta1PowIndex(); i++) {
for (size_t i = GSumIndex(); i < Beta1PowIndex(); i++) {
sgd[i] = 0.0;
}
// init beta1_pow and beta2_pow
......
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