mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2025-06-08 01:53:19 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #25184 from dkurt:avoid_extra_memset

Browse Source 
Avoid extra memset
This commit is contained in:
Alexander Smorkalov 2024-03-25 13:07:49 +03:00 committed by GitHub
commit fc34554475
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 0 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
modules/dnn/src/layers/cpu_kernels/convolution.cpp
View File

@ -210,7 +210,6 @@ Ptr<FastConv> initFastConv(
{ {
conv->weightsBuf_FP16.resize(nweights + VEC_ALIGN); conv->weightsBuf_FP16.resize(nweights + VEC_ALIGN);
auto weightsPtr_FP16 = conv->getWeightsFP16(); auto weightsPtr_FP16 = conv->getWeightsFP16();
memset(reinterpret_cast<short*>(weightsPtr_FP16), 0, nweights * sizeof(weightsPtr_FP16[0]));
parallel_for_(Range(0, C), [&](const Range& r0){ parallel_for_(Range(0, C), [&](const Range& r0){
for(int c = r0.start; c < r0.end; c++) for(int c = r0.start; c < r0.end; c++)
@ -222,7 +221,6 @@ Ptr<FastConv> initFastConv(
{ {
conv->weightsBuf.resize(nweights + VEC_ALIGN); conv->weightsBuf.resize(nweights + VEC_ALIGN);
auto weightsPtr = conv->getWeights(); auto weightsPtr = conv->getWeights();
memset(weightsPtr, 0, nweights*sizeof(weightsPtr[0]));
parallel_for_(Range(0, C), [&](const Range& r0) { parallel_for_(Range(0, C), [&](const Range& r0) {
for(int c = r0.start; c < r0.end; c++) for(int c = r0.start; c < r0.end; c++)
@ -276,14 +274,12 @@ Ptr<FastConv> initFastConv(
{ {
conv->weightsWinoBuf_FP16.resize(nweights + VEC_ALIGN); conv->weightsWinoBuf_FP16.resize(nweights + VEC_ALIGN);
wptrWino_FP16 = conv->getWeightsWinoFP16(); wptrWino_FP16 = conv->getWeightsWinoFP16();
memset(reinterpret_cast<short*>(wptrWino_FP16), 0, nweights * sizeof(wptrWino_FP16[0]));
} }
else else
#endif #endif
{ {
conv->weightsWinoBuf.resize(nweights + VEC_ALIGN); conv->weightsWinoBuf.resize(nweights + VEC_ALIGN);
wptrWino = conv->getWeightsWino(); wptrWino = conv->getWeightsWino();
memset(wptrWino, 0, nweights * sizeof(wptrWino[0]));
} }
parallel_for_(Range(0, K), [&](const Range& r0){ parallel_for_(Range(0, K), [&](const Range& r0){
@ -377,14 +373,12 @@ Ptr<FastConv> initFastConv(
{ {
conv->weightsBuf_FP16.resize(nweights_FP16 + VEC_ALIGN); conv->weightsBuf_FP16.resize(nweights_FP16 + VEC_ALIGN);
weightsPtr_FP16 = conv->getWeightsFP16(); weightsPtr_FP16 = conv->getWeightsFP16();
memset(reinterpret_cast<short*>(weightsPtr_FP16), 0, nweights_FP16*sizeof(weightsPtr_FP16[0]));
} }
else else
#endif #endif
{ {
conv->weightsBuf.resize(nweights + VEC_ALIGN); conv->weightsBuf.resize(nweights + VEC_ALIGN);
weightsPtr = conv->getWeights(); weightsPtr = conv->getWeights();
memset(weightsPtr, 0, nweights*sizeof(weightsPtr[0]));
} }
// Pack the weight. // Pack the weight.
@ -651,7 +645,6 @@ static inline void packInputData(char* inpbuf_task, float* inp, const int* ofsta
for (int c = 0; c < Cg; c++, inptr += inp_planesize, inpbuf += CONV_NR_esz) for (int c = 0; c < Cg; c++, inptr += inp_planesize, inpbuf += CONV_NR_esz)
{ {
_cvt32f16f(inptr, (float16_t *)inpbuf, slice_len); _cvt32f16f(inptr, (float16_t *)inpbuf, slice_len);
memset(inpbuf + slice_len * esz, 0, (CONV_NR - slice_len) * esz);
} }
} }
else else
@ -659,7 +652,6 @@ static inline void packInputData(char* inpbuf_task, float* inp, const int* ofsta
for (int c = 0; c < Cg; c++, inptr += inp_planesize, inpbuf += CONV_NR_esz) for (int c = 0; c < Cg; c++, inptr += inp_planesize, inpbuf += CONV_NR_esz)
{ {
memcpy(inpbuf, inptr, slice_len * esz); memcpy(inpbuf, inptr, slice_len * esz);
memset(inpbuf + slice_len * esz, 0, (CONV_NR - slice_len) * esz);
} }
} }
} }