Skip to content

[big tensor] upsample 支持大tensor #74242

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 5 commits into from
Jul 30, 2025
Merged

[big tensor] upsample 支持大tensor #74242

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
a970b3b
fix
cszdrg Jul 25, 2025
3058d64
fix
cszdrg Jul 28, 2025
946c30f
fix
cszdrg Jul 28, 2025
100a04e
fix
cszdrg Jul 29, 2025
303fbe6
fix
cszdrg Jul 29, 2025
File filter

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
54 changes: 27 additions & 27 deletions paddle/phi/kernels/cpu/interpolate_grad_kernel.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,9 +28,9 @@ template <typename T>
static void LinearInterpolationGrad(const DenseTensor& output_grad,
DenseTensor* input_grad,
const float ratio_w,
const int in_w,
const int n,
const int c,
const int64_t in_w,
const int64_t n,
const int64_t c,
const int out_w,
const bool align_corners,
const int align_mode,
Expand Down Expand Up @@ -75,10 +75,10 @@ static void BilinearInterpolationGrad(const DenseTensor& output_grad,
DenseTensor* input_grad,
const float ratio_h,
const float ratio_w,
const int in_h,
const int in_w,
const int n,
const int c,
const int64_t in_h,
const int64_t in_w,
const int64_t n,
const int64_t c,
const int out_h,
const int out_w,
const bool align_corners,
Expand Down Expand Up @@ -142,8 +142,8 @@ static void NearestNeighborInterpolateGrad(const DenseTensor& output_grad,
DenseTensor* input_grad,
const float ratio_h,
const float ratio_w,
const int n,
const int c,
const int64_t n,
const int64_t c,
const int out_h,
const int out_w,
const bool align_corners,
Expand Down Expand Up @@ -179,10 +179,10 @@ static void BicubicInterpolationGrad(const DenseTensor& output_grad,
DenseTensor* input_grad,
const float ratio_h,
const float ratio_w,
const int in_h,
const int in_w,
const int n,
const int c,
const int64_t in_h,
const int64_t in_w,
const int64_t n,
const int64_t c,
const int out_h,
const int out_w,
const bool align_corners,
Expand All @@ -194,13 +194,13 @@ static void BicubicInterpolationGrad(const DenseTensor& output_grad,
for (int k = 0; k < out_h; k++) { // loop for images
MT y_n = align_corners ? ratio_h * static_cast<float>(k)
: ratio_h * (static_cast<float>(k) + 0.5f) - 0.5f;
int input_y = floorf(y_n);
int64_t input_y = floorf(y_n);
MT y_t = y_n - input_y;

for (int l = 0; l < out_w; l++) {
MT x_n = align_corners ? ratio_w * static_cast<float>(l)
: ratio_w * (static_cast<float>(l) + 0.5f) - 0.5f;
int input_x = floorf(x_n);
int64_t input_x = floorf(x_n);
MT x_t = x_n - input_x;

std::array<MT, 4> x_coeffs;
Expand All @@ -215,9 +215,9 @@ static void BicubicInterpolationGrad(const DenseTensor& output_grad,
for (int ii = 0; ii < 4; ii++) {
for (int jj = 0; jj < 4; jj++) {
int access_x = std::max(std::min(input_x - 1 + ii, in_w - 1),
static_cast<int>(0));
static_cast<int64_t>(0));
int access_y = std::max(std::min(input_y - 1 + jj, in_h - 1),
static_cast<int>(0));
static_cast<int64_t>(0));
if (data_layout == DataLayout::kNCHW) {
MT grad = static_cast<MT>(output_grad_t(i, j, k, l));
input_grad_t(i, j, access_y, access_x) +=
Expand All @@ -241,11 +241,11 @@ static void TrilinearInterpolationGrad(const DenseTensor& output_grad,
const float ratio_d,
const float ratio_h,
const float ratio_w,
const int in_d,
const int in_h,
const int in_w,
const int n,
const int c,
const int64_t in_d,
const int64_t in_h,
const int64_t in_w,
const int64_t n,
const int64_t c,
const int out_d,
const int out_h,
const int out_w,
Expand Down Expand Up @@ -348,8 +348,8 @@ static void NearestNeighbor3DInterpolateGrad(const DenseTensor& output_grad,
const float ratio_d,
const float ratio_h,
const float ratio_w,
const int n,
const int c,
const int64_t n,
const int64_t c,
const int out_d,
const int out_h,
const int out_w,
Expand Down Expand Up @@ -407,7 +407,7 @@ static void Interpolate1DCPUBwd(
int align_mode,
DenseTensor* input_grad) {
const DataLayout data_layout = common::StringToDataLayout(data_layout_str);
int n = 0, c = 0, in_d = 0, in_h = 0, in_w = 0;
int64_t n = 0, c = 0, in_d = 0, in_h = 0, in_w = 0;
funcs::ExtractNCDWH(input.dims(), data_layout, &n, &c, &in_d, &in_h, &in_w);

float scale_w = -1.0;
Expand Down Expand Up @@ -508,7 +508,7 @@ static void Interpolate2DCPUBwd(
int align_mode,
DenseTensor* input_grad) {
const DataLayout data_layout = common::StringToDataLayout(data_layout_str);
int n = 0, c = 0, in_d = 0, in_h = 0, in_w = 0;
int64_t n = 0, c = 0, in_d = 0, in_h = 0, in_w = 0;
funcs::ExtractNCDWH(input.dims(), data_layout, &n, &c, &in_d, &in_h, &in_w);

float scale_h = -1;
Expand Down Expand Up @@ -674,7 +674,7 @@ static void Interpolate3DCPUBwd(
int align_mode,
DenseTensor* input_grad) {
const DataLayout data_layout = common::StringToDataLayout(data_layout_str);
int n = 0, c = 0, in_d = 0, in_h = 0, in_w = 0;
int64_t n = 0, c = 0, in_d = 0, in_h = 0, in_w = 0;
funcs::ExtractNCDWH(input.dims(), data_layout, &n, &c, &in_d, &in_h, &in_w);

float scale_d = -1;
Expand Down
6 changes: 3 additions & 3 deletions paddle/phi/kernels/cpu/interpolate_kernel.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -561,7 +561,7 @@ static void Interpolate1DCPUFwd(
int align_mode,
DenseTensor* output) {
const DataLayout data_layout = common::StringToDataLayout(data_layout_str);
int n = 0, c = 0, in_d = 0, in_h = 0, in_w = 0;
int64_t n = 0, c = 0, in_d = 0, in_h = 0, in_w = 0;
funcs::ExtractNCDWH(x.dims(), data_layout, &n, &c, &in_d, &in_h, &in_w);

float scale_w = -1.;
Expand Down Expand Up @@ -662,7 +662,7 @@ static void Interpolate2DCPUFwd(
int align_mode,
DenseTensor* output) {
const DataLayout data_layout = common::StringToDataLayout(data_layout_str);
int n = 0, c = 0, in_d = 0, in_h = 0, in_w = 0;
int64_t n = 0, c = 0, in_d = 0, in_h = 0, in_w = 0;
funcs::ExtractNCDWH(x.dims(), data_layout, &n, &c, &in_d, &in_h, &in_w);

float scale_h = -1;
Expand Down Expand Up @@ -833,7 +833,7 @@ static void Interpolate3DCPUFwd(
int align_mode,
DenseTensor* output) {
const DataLayout data_layout = common::StringToDataLayout(data_layout_str);
int n = 0, c = 0, in_d = 0, in_h = 0, in_w = 0;
int64_t n = 0, c = 0, in_d = 0, in_h = 0, in_w = 0;
funcs::ExtractNCDWH(x.dims(), data_layout, &n, &c, &in_d, &in_h, &in_w);

float scale_d = -1;
Expand Down
10 changes: 5 additions & 5 deletions paddle/phi/kernels/funcs/interpolate_function.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,11 +54,11 @@ HOSTDEVICE inline void get_cubic_upsample_coefficients(T coeffs[4], T t) {

inline void ExtractNCDWH(const DDim& dims,
const DataLayout& data_layout,
int* N,
int* C,
int* D,
int* H,
int* W) {
int64_t* N,
int64_t* C,
int64_t* D,
int64_t* H,
int64_t* W) {
*N = dims[0];

if (dims.size() == 3) {
Expand Down
6 changes: 3 additions & 3 deletions paddle/phi/kernels/gpu/interpolate_grad_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -847,7 +847,7 @@ static void Interpolate1DCUDABwd(
int align_mode,
DenseTensor* input_grad) {
const DataLayout data_layout = common::StringToDataLayout(data_layout_str);
int n, c, in_d, in_h, in_w;
int64_t n, c, in_d, in_h, in_w;
funcs::ExtractNCDWH(input.dims(), data_layout, &n, &c, &in_d, &in_h, &in_w);

float scale_w = -1;
Expand Down Expand Up @@ -961,7 +961,7 @@ static void Interpolate2DCUDABwd(
int align_mode,
DenseTensor* input_grad) {
const DataLayout data_layout = common::StringToDataLayout(data_layout_str);
int n, c, in_d, in_h, in_w;
int64_t n, c, in_d, in_h, in_w;
funcs::ExtractNCDWH(input.dims(), data_layout, &n, &c, &in_d, &in_h, &in_w);

float scale_h = -1;
Expand Down Expand Up @@ -1239,7 +1239,7 @@ static void Interpolate3DCUDABwd(
int align_mode,
DenseTensor* input_grad) {
const DataLayout data_layout = common::StringToDataLayout(data_layout_str);
int n, c, in_d, in_h, in_w;
int64_t n, c, in_d, in_h, in_w;
funcs::ExtractNCDWH(input.dims(), data_layout, &n, &c, &in_d, &in_h, &in_w);

float scale_d = -1;
Expand Down
6 changes: 3 additions & 3 deletions paddle/phi/kernels/gpu/interpolate_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -645,7 +645,7 @@ static void Interpolate1DCUDAFwd(
auto* input_data = input.data<T>();

const DataLayout data_layout = common::StringToDataLayout(data_layout_str);
int n, c, in_d, in_h, in_w;
int64_t n, c, in_d, in_h, in_w;
funcs::ExtractNCDWH(input.dims(), data_layout, &n, &c, &in_d, &in_h, &in_w);

float scale_w = -1;
Expand Down Expand Up @@ -759,7 +759,7 @@ static void Interpolate2DCUDAFwd(
auto* input_data = input.data<T>();

const DataLayout data_layout = common::StringToDataLayout(data_layout_str);
int n, c, in_d, in_h, in_w;
int64_t n, c, in_d, in_h, in_w;
funcs::ExtractNCDWH(input.dims(), data_layout, &n, &c, &in_d, &in_h, &in_w);

float scale_w = -1;
Expand Down Expand Up @@ -1011,7 +1011,7 @@ static void Interpolate3DCUDAFwd(
auto* input_data = input.data<T>();

const DataLayout data_layout = common::StringToDataLayout(data_layout_str);
int n, c, in_d, in_h, in_w;
int64_t n, c, in_d, in_h, in_w;
funcs::ExtractNCDWH(input.dims(), data_layout, &n, &c, &in_d, &in_h, &in_w);

float scale_w = -1;
Expand Down
2 changes: 1 addition & 1 deletion paddle/phi/kernels/xpu/interpolate_grad_kernel.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,7 @@ void InterpolateGradKernel(
return;
}
const DataLayout data_layout = common::StringToDataLayout(data_layout_str);
int n, c, in_d, in_h, in_w;
int64_t n, c, in_d, in_h, in_w;
funcs::ExtractNCDWH(x.dims(), data_layout, &n, &c, &in_d, &in_h, &in_w);

float scale_h = -1;
Expand Down
2 changes: 1 addition & 1 deletion paddle/phi/kernels/xpu/interpolate_kernel.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,7 @@ void InterpolateKernel(
}
using XPUType = typename XPUTypeTrait<T>::Type;
const DataLayout data_layout = common::StringToDataLayout(data_layout_str);
int n, c, in_d, in_h, in_w;
int64_t n, c, in_d, in_h, in_w;
phi::funcs::ExtractNCDWH(x.dims(), data_layout, &n, &c, &in_d, &in_h, &in_w);

float scale_h = -1;
Expand Down
Loading