Cgo1.7.1 arm: irregular execution time

[HuanjunKong]

My colleague gives me a face_recognization.so, face_recognization.so calls libcaffe.so.
I compiled it with cgo based on arm.

cgo code:

package algo

/*
#include "load_so.h"
#cgo CFLAGS: -std=gnu99
#cgo LDFLAGS: -ldl
#cgo LDFLAGS: -L./ -L/usr/lib -L/usr/local/lib -L/usr/lib -L/usr/local/cuda/lib -L/home/ubuntu/caffe_gx/caffe/.build_release/lib  -lcaffe  -lcudart -lcublas -lcurand -lglog -lgflags -lprotobuf -lboost_system -lboost_filesystem -lm -lhdf5_hl -lhdf5 -lleveldb -lsnappy -llmdb -lopencv_core -lopencv_highgui -lopencv_imgproc -lboost_thread -lstdc++ -lcudnn -lcblas -latlas -lopencv_ml -lopencv_objdetect -lopencv_imgproc -lboost_thread -lstdc++ -lcblas -latlas -lopencv_core -lopencv_highgui -lopencv_imgproc -lboost_thread -lstdc++ -lboost_python  -lboost_filesystem -lboost_system -lboost_thread -lboost_serialization -lopencv_ml  -L./  -lface_recognization
*/
import "C"
import "unsafe"

func ExtractFeature(bytes []byte) ([]float32, error) {
	...
	errCode := C.extractFeature(pInput, inSize, pOuput, (*C.int)(unsafe.Pointer(&outSize)), (C.int)(mode))
	...
}

In load_so.c, I use dl_open to load relative function.

void init(char* pDir){
	soHandle = dlopen("./libface_recognization.so", RTLD_LAZY);
	if (!soHandle) {
		fprintf(stderr, "open so fail: %s\n", dlerror());
		return;
	}
	// feature extraction
    ...
	extract = (fptr_model_extract)dlsym(soHandle, "Face_Feature_Extract");
	if (!extract){
		fprintf(stderr, "dlsym Face_Feature_Extract fail: %s\n", dlerror());
		return;
	}
    ...
}
    
int extractFeature(char* input, int inLen, float* output, int* outLen, int mode){
    (*extract)(featureHandle, (const char*)input, inLen, output, outLen, mode);;
}

Calling relationship:

func ExtractFeature in golang --> extractFeature in load_so.c --> Face_Feature_Extract in face_rec.so --> libcaffe.so

1. In his C++ demo, Face_Feature_Extract function always cost 0.1 second.

2. In my golang code, the extractFeature in load_so.c costs 1.8 or 0.1 second, sometimes the caffe function in libcaffe.so spent 1.8 second, sometimes it is normal.

3. If I loop extractFeature function 1000 times in load_so.c, average execution time is 0.1 second.

I doubt that cgo1.7 is not stable, anybody can help me?

go env:
GOARCH=“arm"
GOBIN=”“
GOEXE=”"
GOHOSTARCH=“arm"
GOHOSTOS=“linux"
GOOS=“linux"
GOPATH=”/home/ubuntu/GoProjects"
GORACE=”“
GOROOT=”/home/ubuntu/golang/go17armv6l"
GOTOOLDIR=”/home/ubuntu/golang/go17armv6l/pkg/tool/linux_arm"
CC=“gcc"
GOGCCFLAGS=”-fPIC -marm -pthread -fmessage-length=0 -fdebug-prefix-map=/tmp/go-build053733439=/tmp/go-build -gno-record-gcc-switches"
CXX="g++"
CGO_ENABLED=“1”

hardware: nvidia jetson tx1

os: Linux tegra-ubuntu 3.10.96-tegra #1 SMP PREEMPT Tue May 17 16:31:40 PDT 2016 aarch64 aarch64 aarch64 GNU/Linux