OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-06 12:20:52 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity

Tensor construction codemod(ResizeLike) - 7/7 (#15087)

Browse Source 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/15087

Codemod generated with clangr shard mode, 25 files per diff,
motivation: https://github.com/pytorch/pytorch/pull/12407

Reviewed By: ezyang

Differential Revision: D13419765

fbshipit-source-id: 34d695309a66723281429610a12544598c507d74
This commit is contained in:
Jerry Zhang 2018-12-20 15:28:12 -08:00 committed by Facebook Github Bot
parent d6cbcb43c5
commit ed5b584f65
10 changed files with 47 additions and 50 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
caffe2/operators/swish_op.cc
View File

@ -24,10 +24,10 @@ bool SwishGradientOp<CPUContext>::DoRunWithType() {
auto& Xin = Input(X);
auto& Yin = Input(Y);
auto& DYin = Input(DY);
auto* DXout = Output(DX);
CAFFE_ENFORCE_EQ(Xin.numel(), Yin.numel());
CAFFE_ENFORCE_EQ(DYin.numel(), Yin.numel());
DXout->ResizeLike(Yin);
auto* DXout = Output(DX, Yin.sizes(), at::dtype<float>());
const float* Xdata = Xin.template data<float>();
const float* Ydata = Yin.template data<float>();

8
caffe2/operators/thresholded_relu_op.cc
View File

@ -8,8 +8,8 @@ namespace caffe2 {
template <>
bool ThresholdedReluOp<float, CPUContext>::RunOnDevice() {
auto& X = Input(0);
auto* Y = Output(0);
Y->ResizeLike(X);
auto* Y = Output(0, X.sizes(), at::dtype<float>());
ConstEigenVectorArrayMap<float> Xvec(X.data<float>(), X.numel());
EigenVectorArrayMap<float> Yvec(
@ -30,9 +30,9 @@ template <>
bool ThresholdedReluGradientOp<float, CPUContext>::RunOnDevice() {
auto& Y = Input(0);
auto& dY = Input(1);
auto* dX = Output(0);
CAFFE_ENFORCE_EQ(dY.numel(), Y.numel());
dX->ResizeLike(Y);
auto* dX = Output(0, Y.sizes(), at::dtype<float>());
const float* Ydata = Y.data<float>();
const float* dYdata = dY.data<float>();

4
caffe2/operators/unique_ops.cc
View File

@ -30,8 +30,8 @@ bool UniqueOp<CPUContext>::DoRunWithType() {
int* remapping = nullptr;
if (REMAPPING < OutputSize()) {
auto* remappingTensor = Output(REMAPPING);
remappingTensor->ResizeLike(inputTensor);
auto* remappingTensor =
Output(REMAPPING, inputTensor.sizes(), at::dtype<int>());
remapping = remappingTensor->template mutable_data<int>();
}

33
caffe2/operators/utility_ops.h
View File

@ -255,12 +255,19 @@ class SumOp : public Operator<Context> {
template <typename T, typename M>
bool DoRunWithType() {
auto& input0 = Input(0);
auto* output = Output(0);
if (InputSize() == 1) {
output->CopyFrom(input0, true /*async*/);
// TODO: better TensorOptions argument passing(e.g. default argument)
OutputTensorCopyFrom(
0,
// I'll change the order of argument in another diff, so that we don't
// need to write this
at::dtype(input0.dtype()),
input0,
true /*async*/);
return true;
}
output->ResizeLike(input0);
auto* output = Output(0, input0.sizes(), at::dtype<T>());
T* output_data = output->template mutable_data<T>();
// Dimension checking
for (int i = 1; i < InputSize(); ++i) {
@ -331,10 +338,9 @@ class WeightedSumOp : public Operator<Context> {
CAFFE_ENFORCE_GT(X0.numel(), 0);
CAFFE_ENFORCE_EQ(weight0.numel(), 1);
const int size = X0.numel();
auto* Y = Output(0);
if (Y != &X0) {
Y->ResizeLike(X0);
}
// Note: removed Aliasing check, since Output already has
// caching capability
auto* Y = Output(0, X0.sizes(), at::dtype<T>());
T* Y_data = Y->template mutable_data<T>();
if (input_size == 2) {
math::Scale<float, T>(
@ -346,15 +352,14 @@ class WeightedSumOp : public Operator<Context> {
return true;
}
const auto& X1 = Input(2);
CAFFE_ENFORCE_NE(
&X1,
Y,
CAFFE_ENFORCE(
!IsInputOutputAlias(2, 0),
"Input #2 is the same as output. If you want to do in-place updates, "
"put the output as input #0.");
const auto& weight1 = Input(3);
CAFFE_ENFORCE_EQ(X1.numel(), size);
CAFFE_ENFORCE_EQ(weight1.numel(), 1);
if (Y != &X0) {
if (!IsInputOutputAlias(0, 0)) {
context_.template CopySameDevice<T>(size, X0.template data<T>(), Y_data);
}
math::Axpby<float, T, Context>(
@ -371,7 +376,7 @@ class WeightedSumOp : public Operator<Context> {
const std::string err_msg = "Input #" + to_string(i) +
" is the same as output. If you want to do in-place updates, "
"put the output as input #0.";
CAFFE_ENFORCE_NE(&Xi, Y, err_msg);
CAFFE_ENFORCE(!IsInputOutputAlias(i, 0), err_msg);
const auto& weighti = Input(i + 1);
CAFFE_ENFORCE_EQ(Xi.numel(), size);
CAFFE_ENFORCE_EQ(weighti.numel(), 1);
@ -410,8 +415,8 @@ class WeightedSumGradientOp : public Operator<Context> {
for (int i = 0; i < InputSize() / 2; i++) {
auto& cur_w = Input(2 * i + 2);
CAFFE_ENFORCE_EQ(cur_w.numel(), 1);
auto* cur_dX = Output(i);
cur_dX->ResizeLike(dY);
auto* cur_dX = Output(i, dY.sizes(), at::dtype<DstType>());
math::Scale<float, DstType, Context>(
size,

17
caffe2/operators/utility_ops_cudnn.cc
View File

@ -45,10 +45,10 @@ class CuDNNWeightedSumOp : public Operator<CUDAContext> {
CAFFE_ENFORCE_EQ(weight0.numel(), 1);
const int input_size = X0.numel();
SetTensorDescriptor(cudnnTypeWrapper<T>::type, input_size);
auto* Y = Output(0);
if (Y != &X0) {
Y->ResizeLike(X0);
}
// Note: removed Aliasing check, since Output already has
// caching capability
auto* Y = Output(0, X0.sizes(), at::dtype<T>());
T* Y_data = Y->template mutable_data<T>();
T alpha = convert::To<float, T>(0.0f);
T beta = convert::To<float, T>(0.0f);
@ -65,9 +65,8 @@ class CuDNNWeightedSumOp : public Operator<CUDAContext> {
return true;
}
const auto& X1 = Input(2);
CAFFE_ENFORCE_NE(
&X1,
Y,
CAFFE_ENFORCE(
!IsInputOutputAlias(2, 0),
"Input #2 is the same as output. If you want to do in-place updates, "
"put the output as input #0.");
const auto& weight1 = Input(3);
@ -75,7 +74,7 @@ class CuDNNWeightedSumOp : public Operator<CUDAContext> {
CAFFE_ENFORCE_EQ(weight1.numel(), 1);
CopyWeightToHost<T>(weight1.template data<float>(), &alpha);
CopyWeightToHost<T>(weight0.template data<float>(), &beta);
if (Y == &X0) {
if (IsInputOutputAlias(0, 0)) {
CUDNN_ENFORCE(cudnnAddTensor(
cudnn_wrapper_.inline_cudnn_handle(),
&alpha,
@ -105,7 +104,7 @@ class CuDNNWeightedSumOp : public Operator<CUDAContext> {
const std::string err_msg = "Input #" + to_string(i) +
" is the same as output. If you want to do in-place updates, "
"put the output as input #0.";
CAFFE_ENFORCE_NE(&Xi, Y, err_msg);
CAFFE_ENFORCE(!IsInputOutputAlias(i, 0), err_msg);
const auto& weighti = Input(i + 1);
CAFFE_ENFORCE_EQ(Xi.numel(), input_size);
CAFFE_ENFORCE_EQ(weighti.numel(), 1);

8
caffe2/quantization/server/conv_relu_op.cc
View File

@ -20,8 +20,8 @@ bool ConvReluOp<T, Context>::RunOnDeviceWithOrderNCHW() {
BlobGetMutableTensor(local_output_blobs_[0], Context::GetDeviceType());
const T* output_local_data = local_output->template data<T>();
Tensor* output = Operator<Context>::Output(0);
output->ResizeLike(*local_output);
Tensor* output =
Operator<Context>::Output(0, local_output->sizes(), at::dtype<T>());
T* output_data = output->template mutable_data<T>();
#ifdef _OPENMP
#pragma omp parallel for
@ -51,8 +51,8 @@ bool ConvReluOp<T, Context>::RunOnDeviceWithOrderNHWC() {
BlobGetMutableTensor(local_output_blobs_[0], Context::GetDeviceType());
const T* output_local_data = local_output->template data<T>();
Tensor* output = Operator<Context>::Output(0);
output->ResizeLike(*local_output);
Tensor* output =
Operator<Context>::Output(0, local_output->sizes(), at::dtype<T>());
T* output_data = output->template mutable_data<T>();
#ifdef _OPENMP
#pragma omp parallel for

11
caffe2/quantization/server/fully_connected_fake_lowp_op.cc
View File

@ -179,9 +179,7 @@ bool FullyConnectedGradientFakeLowpFPOp<Q, Context, Engine, TransposeWeight>::
CAFFE_ENFORCE(M * K == X.size());
CAFFE_ENFORCE(K * N == W.size());
auto* dW = Output(0);
dW->ResizeLike(W);
auto* dW = Output(0, W.sizes(), at::dtype<T_DW>());
auto* db = Output(1, {N}, at::dtype<T_DB>());
if (X.size() == 0) {
@ -198,9 +196,7 @@ bool FullyConnectedGradientFakeLowpFPOp<Q, Context, Engine, TransposeWeight>::
&context_);
if (OutputSize() == 3) {
auto* dX = Output(2);
dX->ResizeLike(X);
dX->template mutable_data<T_DX>();
Output(2, X.sizes(), at::dtype<T_DX>());
}
return true;
@ -270,8 +266,7 @@ bool FullyConnectedGradientFakeLowpFPOp<Q, Context, Engine, TransposeWeight>::
// Compute dX
if (OutputSize() == 3) {
auto* dX = Output(2);
dX->ResizeLike(X);
auto* dX = Output(2, X.sizes(), at::dtype<T_DX>());
math::Gemm<T_DX, Context, Engine>(
CblasNoTrans,
TransposeWeight ? CblasNoTrans : CblasTrans,

3
caffe2/sgd/clip_tensor_op.h
View File

@ -29,8 +29,7 @@ class ClipTensorByScalingOp final : public Operator<Context> {
const auto* input_tensor_data = input_tensor.template data<float>();
const auto* val_data = val.template data<float>();
auto* clipped = Output(0);
clipped->ResizeLike(input_tensor);
auto* clipped = Output(0, input_tensor.sizes(), at::dtype<float>());
float* clipped_tensor_data = clipped->template mutable_data<float>();
if (InputSize() > 2) {

6
caffe2/sgd/yellowfin_op.h
View File

@ -157,9 +157,9 @@ for (int i = 0; i < param_tensor.dim(); ++i) {
// Output data
#define CAFFE2_YF_READ_OUTPUT(OUTPUT_NAME, VAR_NAME) \
auto VAR_NAME##_out_tensor = Output(OUTPUT_##OUTPUT_NAME); \
VAR_NAME##_out_tensor->ResizeLike(VAR_NAME##_tensor); \
#define CAFFE2_YF_READ_OUTPUT(OUTPUT_NAME, VAR_NAME) \
auto VAR_NAME##_out_tensor = \
Output(OUTPUT_##OUTPUT_NAME, VAR_NAME##_tensor.sizes(), at::dtype<T>()); \
VAR_NAME##_out_ = VAR_NAME##_out_tensor->template mutable_data<T>();
CAFFE2_YF_READ_OUTPUT(PARAM, param)

3
modules/detectron/batch_permutation_op.cc
View File

@ -78,7 +78,6 @@ template <>
bool BatchPermutationOp<float, CPUContext>::RunOnDevice() {
const auto& X = Input(0);
const auto& indices = Input(1);
auto* Y = Output(0);
CAFFE_ENFORCE_EQ(indices.dim(), 1, "indices must be 1-d");
CAFFE_ENFORCE_EQ(
@ -90,7 +89,7 @@ bool BatchPermutationOp<float, CPUContext>::RunOnDevice() {
indices.dim32(0),
")");
Y->ResizeLike(X);
auto* Y = Output(0, X.sizes(), at::dtype<float>());
const int N = X.dim32(0);
const int C = X.dim32(1);