Skip to content

Neon Depthwise Convolution Transpose Function #3792

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
Sep 7, 2017
Merged

Neon Depthwise Convolution Transpose Function #3792

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
f7e75a0
Refine the neon depthwise convolution code(separate the Function and …
hedaoyuan Aug 31, 2017
4b6b725
Refine NeonDepthwiseConv.
hedaoyuan Aug 31, 2017
40d47fa
[Refine code]Move class Padding into the NeonDepthwiseConv.h.
hedaoyuan Aug 31, 2017
840104c
Add NeonDepthwiseConvTransposeFunction.
hedaoyuan Aug 31, 2017
90bf4f6
Add stride support 2 for NeonDepthwiseConvTranspose.
hedaoyuan Aug 31, 2017
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
91 changes: 0 additions & 91 deletions paddle/function/Im2Col.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -94,95 +94,4 @@ class Col2ImFunctor {
int paddingWidth);
};

template <class T>
struct Padding {
static void run(const T* src,
T* dest,
int channels,
int inputHeight,
int inputWidth,
int paddingHeight,
int paddingWidth) {
const int destWidth = inputWidth + 2 * paddingWidth;
for (int c = 0; c < channels; c++) {
if (paddingHeight > 0) {
memset(dest, 0, destWidth * paddingHeight * sizeof(T));
dest += destWidth * paddingHeight;
}

for (int i = 0; i < inputHeight; i++) {
// padding head
for (int j = 0; j < paddingWidth; j++) {
*dest++ = T(0);
}

memcpy(dest, src, inputWidth * sizeof(T));
dest += inputWidth;
src += inputWidth;

// padding tail
for (int j = 0; j < paddingWidth; j++) {
*dest++ = T(0);
}
}

if (paddingHeight > 0) {
memset(dest, 0, destWidth * paddingHeight * sizeof(T));
dest += destWidth * paddingHeight;
}
}
}
};

#if defined(__ARM_NEON__) || defined(__ARM_NEON)
template <>
struct Padding<float> {
static void run(const float* src,
float* dest,
int channels,
int inputHeight,
int inputWidth,
int paddingHeight,
int paddingWidth) {
const int destWidth = inputWidth + 2 * paddingWidth;
for (int c = 0; c < channels; c++) {
if (paddingHeight > 0) {
memset(dest, 0, destWidth * paddingHeight * sizeof(float));
dest += destWidth * paddingHeight;
}

for (int i = 0; i < inputHeight; i++) {
// padding head
for (int j = 0; j < paddingWidth; j++) {
*dest++ = float(0);
}

int step = inputWidth >> 2;
int remain = inputWidth & 3;
for (int s = 0; s < step; s++) {
float32x4_t s0 = vld1q_f32(src);
vst1q_f32(dest, s0);
src += 4;
dest += 4;
}
for (int r = 0; r < remain; r++) {
*dest++ = *src++;
}

// padding tail
for (int j = 0; j < paddingWidth; j++) {
*dest++ = float(0);
}
}

if (paddingHeight > 0) {
memset(dest, 0, destWidth * paddingHeight * sizeof(float));
dest += destWidth * paddingHeight;
}
}
}
};

#endif

} // namespace paddle
Loading