提交 55d6950a 编写于 作者: S Sylwester Fraczek

rewrite conv_bn fuse pass to eigen

test=develop
上级 9c77b65c
...@@ -44,87 +44,16 @@ namespace ir { ...@@ -44,87 +44,16 @@ namespace ir {
GET_IR_NODE_FROM_SUBGRAPH(bn_saved_mean, bn_saved_mean, pattern_name); \ GET_IR_NODE_FROM_SUBGRAPH(bn_saved_mean, bn_saved_mean, pattern_name); \
GET_IR_NODE_FROM_SUBGRAPH(bn_saved_variance, bn_saved_variance, pattern_name) GET_IR_NODE_FROM_SUBGRAPH(bn_saved_variance, bn_saved_variance, pattern_name)
template <typename UnaryOperation>
LoDTensor tensor_apply(const LoDTensor& vec, UnaryOperation f) {
LoDTensor vec_y;
vec_y.Resize(vec.dims());
const float* x = vec.data<float>();
float* y = vec_y.mutable_data<float>(platform::CPUPlace());
for (int64_t i = 0; i < vec.numel(); i++) {
y[i] = f(x[i]);
}
return vec_y;
}
void tensor_apply_inplace(LoDTensor* vec, float (*f)(float)) {
float* data = vec->mutable_data<float>(platform::CPUPlace());
for (int64_t i = 0; i < vec->numel(); i++) {
data[i] = f(data[i]);
}
}
template <typename BinaryOperation>
LoDTensor tensor_apply_eltwise(const LoDTensor& vec_a, const LoDTensor& vec_b,
BinaryOperation f) {
PADDLE_ENFORCE_EQ(vec_a.dims(), vec_b.dims());
LoDTensor vec_y;
vec_y.Resize(vec_a.dims());
const float* a = vec_a.data<float>();
const float* b = vec_b.data<float>();
float* y = vec_y.mutable_data<float>(platform::CPUPlace());
for (int64_t i = 0; i < vec_a.numel(); i++) {
y[i] = f(a[i], b[i]);
}
return vec_y;
}
template <typename BinaryOperation>
LoDTensor tensor_apply_eltwise_broadcast(const LoDTensor& vec_a,
const LoDTensor& vec_b,
BinaryOperation f) {
PADDLE_ENFORCE_EQ(vec_a.dims().size(), 2);
PADDLE_ENFORCE_EQ(vec_b.dims().size(), 2);
PADDLE_ENFORCE_EQ(vec_a.dims()[0], vec_b.dims()[0]);
PADDLE_ENFORCE_EQ(vec_b.dims()[1], 1);
LoDTensor vec_y;
vec_y.Resize(vec_a.dims());
const float* a = vec_a.data<float>();
const float* b = vec_b.data<float>();
float* y = vec_y.mutable_data<float>(platform::CPUPlace());
size_t a_height = vec_a.dims()[0];
size_t a_width = vec_a.dims()[1];
for (size_t h = 0; h < a_height; h++) {
for (size_t w = 0; w < a_width; ++w) {
*(y++) = f(*(a++), b[h]);
}
}
return vec_y;
}
// reshape to two dimensions {A, B * C * ...} // reshape to two dimensions {A, B * C * ...}
void make_tensor_2d(LoDTensor* tensor_to_reshape) { DDim make_dims_2d(DDim dims) {
auto dims_count = tensor_to_reshape->dims().size(); auto dims_count = dims.size();
PADDLE_ENFORCE_GT(dims_count, 0); PADDLE_ENFORCE_GT(dims_count, 0);
int size2 = 1; int size2 = 1;
for (int i = 1; i < dims_count; i++) { for (int i = 1; i < dims_count; i++) {
size2 *= tensor_to_reshape->dims()[i]; size2 *= dims[i];
} }
tensor_to_reshape->Resize(make_ddim({tensor_to_reshape->dims()[0], size2})); return make_ddim({dims[0], size2});
}
void recompute_conv_weights(LoDTensor* weights, LoDTensor* tmp) {
// remember the weights tensor shape {A, B, C, ...}
auto weights_shape = weights->dims();
// reduce the weights to 2d {A, B * C * ...}
make_tensor_2d(weights);
// make tmp tensor 2d by adding 1 as second dim {A, 1}
make_tensor_2d(tmp);
*weights =
tensor_apply_eltwise_broadcast(*weights, *tmp, std::multiplies<float>());
// reshape weights to the original dims {A, B, C, ...}
weights->Resize(weights_shape);
} }
void recompute_bias_and_weights(const Scope* scope, void recompute_bias_and_weights(const Scope* scope,
...@@ -135,6 +64,13 @@ void recompute_bias_and_weights(const Scope* scope, ...@@ -135,6 +64,13 @@ void recompute_bias_and_weights(const Scope* scope,
const ir::Node& bn_variance, // const ir::Node& bn_variance, //
LoDTensor* eltwise_y_in_tensor, // LoDTensor* eltwise_y_in_tensor, //
float epsilon) { float epsilon) {
using EigenVectorArrayMap =
Eigen::Map<Eigen::Array<float, Eigen::Dynamic, 1>>;
using ConstEigenVectorArrayMap =
Eigen::Map<const Eigen::Array<float, Eigen::Dynamic, 1>>;
using EigenMatrixArrayMap = Eigen::Map<
Eigen::Array<float, Eigen::Dynamic, Eigen::Dynamic, Eigen::RowMajor>>;
// Re-compute bias of conv2d from BN // Re-compute bias of conv2d from BN
PADDLE_ENFORCE_EQ(eltwise_y_in_tensor->dims(), bn_bias_tensor.dims()); PADDLE_ENFORCE_EQ(eltwise_y_in_tensor->dims(), bn_bias_tensor.dims());
...@@ -143,31 +79,38 @@ void recompute_bias_and_weights(const Scope* scope, ...@@ -143,31 +79,38 @@ void recompute_bias_and_weights(const Scope* scope,
scope->FindVar(bn_variance.Name())->GetMutable<LoDTensor>(); scope->FindVar(bn_variance.Name())->GetMutable<LoDTensor>();
auto* mean_tensor = scope->FindVar(bn_mean.Name())->GetMutable<LoDTensor>(); auto* mean_tensor = scope->FindVar(bn_mean.Name())->GetMutable<LoDTensor>();
auto std_tensor = LoDTensor(); ConstEigenVectorArrayMap scale_array(scale_tensor->data<float>(),
std_tensor.Resize(bn_bias_tensor.dims()); scale_tensor->numel(), 1);
std_tensor = EigenVectorArrayMap variance_array(
tensor_apply(*variance_tensor, [&](float x) { return x + epsilon; }); variance_tensor->mutable_data<float>(platform::CPUPlace()),
variance_tensor->numel(), 1);
ConstEigenVectorArrayMap mean_array(mean_tensor->data<float>(),
mean_tensor->numel(), 1);
ConstEigenVectorArrayMap bn_bias_array(bn_bias_tensor.data<float>(),
bn_bias_tensor.numel(), 1);
using EigenVectorArrayMap = // variance will not be used anymore, so make it std_array and then tmp_array
Eigen::Map<Eigen::Array<float, Eigen::Dynamic, 1>>; variance_array += epsilon;
variance_array = variance_array.sqrt();
variance_array = scale_array / variance_array;
EigenVectorArrayMap eltwise_y_in_array(
eltwise_y_in_tensor->mutable_data<float>(platform::CPUPlace()),
eltwise_y_in_tensor->numel(), 1);
EigenVectorArrayMap std_vec( eltwise_y_in_array =
std_tensor.mutable_data<float>(platform::CPUPlace()), std_tensor.numel(), ((eltwise_y_in_array - mean_array) * variance_array) + bn_bias_array;
1);
std_vec = std_vec.sqrt();
auto tmp_tensor =
tensor_apply_eltwise(*scale_tensor, std_tensor, std::divides<float>());
auto tensor_minus = tensor_apply_eltwise(*eltwise_y_in_tensor, *mean_tensor,
std::minus<float>());
auto tensor_mul =
tensor_apply_eltwise(tensor_minus, tmp_tensor, std::multiplies<float>());
*eltwise_y_in_tensor =
tensor_apply_eltwise(tensor_mul, bn_bias_tensor, std::plus<float>());
// Re-compute weight of conv2d from BN // Re-compute weight of conv2d from BN
auto* current_param = auto* weights = scope->FindVar(conv_weight->Name())->GetMutable<LoDTensor>();
scope->FindVar(conv_weight->Name())->GetMutable<LoDTensor>(); auto weights_shape = weights->dims();
recompute_conv_weights(current_param, &tmp_tensor); auto weights_shape_2d = make_dims_2d(weights_shape);
EigenMatrixArrayMap weights_array_2d(
weights->mutable_data<float>(platform::CPUPlace()), weights_shape_2d[0],
weights_shape_2d[1]);
weights_array_2d.colwise() *= variance_array;
} }
std::unique_ptr<ir::Graph> ConvBNFusePass::ApplyImpl( std::unique_ptr<ir::Graph> ConvBNFusePass::ApplyImpl(
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册