apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 16:48:03 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[Diffusion] Add C++ dpm solver (#714)

Browse Source 
* Add BetaForAlphaBar, ConvertModelOutput, SetTimesteps, and constructor for DPMSolverMultistepScheduler

* tmp

* Add DPMSolverFirstOrderUpdate

* Add ScaleModelInput

* Add MultiStepDPMSolverSecondOrderUpdate

* add MultiStepDPMSolverThirdOrderUpdate

* Add Step

* Add FASTDEPLOY_DECL

* Add AddNoise

* Fix operator

* update

* Fix DPMSolverMultistepScheduler

* Upgrade Slice

* Fix DPMSolverFirstOrderUpdate

* remove FASTDEPLOY_DECL

* Add config for dpm solver
This commit is contained in:
Jack Zhou
2022-11-30 13:41:22 +08:00
committed by GitHub
parent 3f8ed9bfee
commit d95094cfe5
14 changed files with 675 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
fastdeploy/function/clip.cc
View File

@@ -39,14 +39,15 @@ void ClipKernel(const FDTensor& x, double min, double max, FDTensor* out) {
"max should be greater than or equal to min. But received min = %f, "
"max = %f",
static_cast<float>(min_), static_cast<float>(max_));
out->Allocate(x.Shape(), x.Dtype());
FDTensor tmp;
tmp.Allocate(x.Shape(), x.Dtype());
const T* x_data = reinterpret_cast<const T*>(x.Data());
int64_t numel = x.Numel();
T* out_data = reinterpret_cast<T*>(out->Data());
T* out_data = reinterpret_cast<T*>(tmp.Data());
std::transform(x_data, x_data + numel, out_data, ClipFunctor<T>(min_, max_));
*out = std::move(tmp);
}
void Clip(const FDTensor& x, double min, double max, FDTensor* out) {

21
fastdeploy/function/elementwise.cc
View File

@@ -86,4 +86,25 @@ FDTensor operator/(const FDTensor& x, const FDTensor& y) {
return out;
}
#define INSTANTIATE_OPERATOR(operation_type) \
template FDTensor operator operation_type(const FDTensor& x, bool y); \
template FDTensor operator operation_type(const FDTensor& x, uint8_t y); \
template FDTensor operator operation_type(const FDTensor& x, int16_t y); \
template FDTensor operator operation_type(const FDTensor& x, int y); \
template FDTensor operator operation_type(const FDTensor& x, int64_t y); \
template FDTensor operator operation_type(const FDTensor& x, float y); \
template FDTensor operator operation_type(const FDTensor& x, double y); \
template FDTensor operator operation_type(bool x, const FDTensor& y); \
template FDTensor operator operation_type(uint8_t x, const FDTensor& y); \
template FDTensor operator operation_type(int16_t x, const FDTensor& y); \
template FDTensor operator operation_type(int x, const FDTensor& y); \
template FDTensor operator operation_type(int64_t x, const FDTensor& y); \
template FDTensor operator operation_type(float x, const FDTensor& y); \
template FDTensor operator operation_type(double x, const FDTensor& y)
INSTANTIATE_OPERATOR(+);
INSTANTIATE_OPERATOR(-);
INSTANTIATE_OPERATOR(*);
INSTANTIATE_OPERATOR(/);
} // namespace fastdeploy

34
fastdeploy/function/elementwise.h
View File

@@ -14,9 +14,11 @@
#pragma once
#include "fastdeploy/core/fd_scalar.h"
#include "fastdeploy/core/fd_tensor.h"
namespace fastdeploy {
namespace function {
/** Excute the add operation for input FDTensors. *out = x + y.
@@ -62,10 +64,42 @@ FASTDEPLOY_DECL void Maximum(const FDTensor& x, const FDTensor& y,
FASTDEPLOY_DECL FDTensor operator+(const FDTensor& x, const FDTensor& y);
template <typename T> FDTensor operator+(const FDTensor& x, T y) {
return x + FDTensor(Scalar(y));
}
template <typename T> FDTensor operator+(T x, const FDTensor& y) {
return FDTensor(Scalar(x)) + y;
}
FASTDEPLOY_DECL FDTensor operator-(const FDTensor& x, const FDTensor& y);
template <typename T> FDTensor operator-(const FDTensor& x, T y) {
return x - FDTensor(Scalar(y));
}
template <typename T> FDTensor operator-(T x, const FDTensor& y) {
return FDTensor(Scalar(x)) - y;
}
FASTDEPLOY_DECL FDTensor operator*(const FDTensor& x, const FDTensor& y);
template <typename T> FDTensor operator*(const FDTensor& x, T y) {
return x * FDTensor(Scalar(y));
}
template <typename T> FDTensor operator*(T x, const FDTensor& y) {
return FDTensor(Scalar(x)) * y;
}
FASTDEPLOY_DECL FDTensor operator/(const FDTensor& x, const FDTensor& y);
template <typename T> FDTensor operator/(const FDTensor& x, T y) {
return x / FDTensor(Scalar(y));
}
template <typename T> FDTensor operator/(T x, const FDTensor& y) {
return FDTensor(Scalar(x)) / y;
}
} // namespace fastdeploy

8
fastdeploy/function/elementwise_base.h
View File

@@ -213,10 +213,12 @@ void CommonElementwiseBroadcastForward(const FDTensor& x, const FDTensor& y,
GetBroadcastDimsArrays(x_dims, y_dims, x_dims_array.data(),
y_dims_array.data(), out_dims_array.data(), max_dim,
axis);
z->Allocate(out_dims_array, TypeToDataType<OutType>::dtype);
FDTensor tmp;
tmp.Allocate(out_dims_array, TypeToDataType<OutType>::dtype);
CommonForwardBroadcastCPU<Functor, T, OutType>(
x, y, z, x_dims_array.data(), y_dims_array.data(), out_dims_array.data(),
max_dim, func, is_xsize_larger);
x, y, &tmp, x_dims_array.data(), y_dims_array.data(),
out_dims_array.data(), max_dim, func, is_xsize_larger);
*z = std::move(tmp);
}
template <typename Functor, typename T, typename OutType = T>

15
fastdeploy/function/slice.cc
View File

@@ -163,5 +163,20 @@ void Slice(const FDTensor& x, const std::vector<int64_t>& axes,
}));
}
void Slice(const FDTensor& x, const std::vector<int64_t>& axes,
const std::vector<int64_t>& index, FDTensor* out) {
std::vector<int64_t> ends = index;
for (int i = 0; i < ends.size(); ++i) {
ends[i] += 1;
}
Slice(x, axes, index, ends, out);
for (int i = 0; i < axes.size(); ++i) {
if (out->Shape().size() <= 1) {
break;
}
out->Squeeze(axes[i]);
}
}
} // namespace function
} // namespace fastdeploy

3
fastdeploy/function/slice.h
View File

@@ -37,5 +37,8 @@ FASTDEPLOY_DECL void Slice(const FDTensor& x, const std::vector<int64_t>& axes,
const std::vector<int64_t>& starts,
const std::vector<int64_t>& ends, FDTensor* out);
FASTDEPLOY_DECL void Slice(const FDTensor& x, const std::vector<int64_t>& axes,
const std::vector<int64_t>& index, FDTensor* out);
} // namespace function
} // namespace fastdeploy