zishuo/ascend-opencv
1
0
Fork 0
mirror of https://github.com/Ascend/ascend-opencv.git synced 2025-10-07 09:11:39 +08:00
Code Issues Packages Projects Releases Wiki Activity

update

Browse Source
This commit is contained in:
luoliang
2022-10-10 20:30:17 +08:00
parent afe9141dab
commit e47e5f027b
7 changed files with 115 additions and 81 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
acl/include/opencv2/acl/init_core.hpp
View File

@@ -34,7 +34,7 @@ inline aclEnv::~aclEnv() {
}
/////////////////////////////////////////aclCxt////////////////////////////
inline aclCxt::aclCxt(){};
inline aclCxt::aclCxt() {};
inline aclCxt::aclCxt(int device_id) : _device_id(device_id) {
_context = static_cast<aclrtContext *>(fastMalloc(sizeof(*_context)));

32
acl/src/matrices.cpp
View File

@@ -5,7 +5,6 @@ using namespace cv;
using namespace cv::acl;
namespace cv {
namespace acl {
static int merge_type(int depth, int channels) {
switch (depth) {
case CV_8U:
@@ -36,28 +35,27 @@ void merge(const vector<aclMat> &mv, aclMat &dest, int stream_id) {
for (size_t i = 0; i < mv.size(); ++i) {
int cols = mv[i].step / mv[i].elemSize();
vector<int64_t> inputShape{1, mv[i].rows, cols, mv[i].channels()};
vector<int64_t> inputShape {1, mv[i].rows, cols, mv[i].channels()};
opDesc.AddInputTensorDesc(dataType, inputShape.size(), inputShape.data(),
ACL_FORMAT_NHWC);
}
int cols = mv[0].step / mv[0].elemSize();
int channels = mv.size();
vector<int64_t> outputShape{1, mv[0].rows, cols, channels};
vector<int64_t> outputShape {1, mv[0].rows, cols, channels};
opDesc.AddOutputTensorDesc(dataType, outputShape.size(), outputShape.data(),
ACL_FORMAT_NHWC);
ino64_t N = mv.size();
constexpr int index2 = 2;
constexpr int index3 = 3;
constexpr int index4 = 4;
constexpr int index0 = 0, index1 = 1;
constexpr int index2 = 2, index3 = 3, index4 = 4;
constexpr int merge_size3 = 3;
constexpr int merge_size4 = 4;
aclopSetAttrInt(opDesc.opAttr, "N", N);
aclSetTensorDescName(opDesc.inputDesc[0], "concat_dim");
aclSetTensorDescName(opDesc.inputDesc[index0], "concat_dim");
aclSetTensorDescName(opDesc.inputDesc[1], "x0");
aclSetTensorDescName(opDesc.inputDesc[index1], "x0");
aclSetTensorDescName(opDesc.inputDesc[index2], "x1");
if (mv.size() == merge_size3)
aclSetTensorDescName(opDesc.inputDesc[index3], "x2");
@@ -91,7 +89,7 @@ void merge(const vector<aclMat> &mv, aclMat &dest, int stream_id) {
for (size_t i = 0; i < outputBuffers_.size(); i++)
AclSafeCall(aclDestroyDataBuffer(outputBuffers_[i]));
aclrtFree(dev);
AclSafeCall(aclrtFree(dev));
}
/**
@@ -123,18 +121,18 @@ void transpose(const aclMat &src, aclMat &dest, int stream_id) {
void *dev;
void *perm;
constexpr int dim0_t = 0;
constexpr int dim1_t = 1;
constexpr int dim2_t = 2;
constexpr int dim3_t = 3;
constexpr int dim0_t = 0, dim1_t = 1;
constexpr int dim2_t = 2, dim3_t = 3;
constexpr int index0 = 0, index1 = 1;
constexpr int index2 = 2, index3 = 3;
size_t size = aclGetTensorDescSize(opDesc.inputDesc[1]);
aclrtMalloc(&dev, size, ACL_MEM_MALLOC_NORMAL_ONLY);
aclrtMallocHost(&perm, aclGetTensorDescSize(opDesc.inputDesc.data()[1]));
((int *)perm)[0] = dim0_t;
((int *)perm)[1] = dim2_t;
((int *)perm)[2] = dim1_t;
((int *)perm)[3] = dim3_t;
((int *)perm)[index0] = dim0_t;
((int *)perm)[index1] = dim2_t;
((int *)perm)[index2] = dim1_t;
((int *)perm)[index3] = dim3_t;
aclrtMemcpy(dev, size, perm, size, ACL_MEMCPY_HOST_TO_DEVICE);
inputBuffers_.emplace_back(aclCreateDataBuffer(dev, size));

18
acl/test/test_acl.cpp
View File

@@ -13,6 +13,7 @@ void PERF_TEST::Test_operator_add_perf(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int rand_data_range = 1;
constexpr int min_format_flag = 128;
@@ -20,7 +21,7 @@ void PERF_TEST::Test_operator_add_perf(aclCxt *acl_context) {
vector<int> type{CV_8UC1, CV_32FC1, CV_32SC1, CV_64FC1};
for (size_t i = 0; i < type.size(); ++i) {
test.PrintLog("Perf test : Function: operator+=()", type[i]);
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src(val, val, type[i]);
Mat mat_dest(val, val, type[i]);
@@ -65,6 +66,7 @@ void PERF_TEST::Test_operator_sub_perf(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int rand_data_range1 = 4;
constexpr int rand_data_range2 = 32;
@@ -73,7 +75,7 @@ void PERF_TEST::Test_operator_sub_perf(aclCxt *acl_context) {
vector<int> type{CV_8UC1, CV_32FC1, CV_32SC1, CV_64FC1};
for (size_t i = 0; i < type.size(); ++i) {
test.PrintLog("Perf test : Function: operator-=()", type[i]);
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src(val, val, type[i]);
Mat mat_dest(val, val, type[i]);
@@ -116,17 +118,16 @@ void PERF_TEST::Test_operator_div_perf(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int s_val1 = 1;
constexpr int s_val2 = 2;
constexpr int s_val4 = 4;
constexpr int s_val8 = 8;
constexpr int s_val1 = 1, s_val2 = 2;
constexpr int s_val4 = 4, s_val8 = 8;
constexpr int min_format_flag = 128;
vector<int> type{CV_8UC1, CV_32FC1, CV_32SC1, CV_64FC1};
for (size_t i = 0; i < type.size(); ++i) {
test.PrintLog("Perf test : Function: operator/=()", type[i]);
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src(val, val, type[i], Scalar(s_val1, s_val2, s_val4));
Mat mat_dest(val, val, type[i], Scalar(s_val2, s_val4, s_val8));
@@ -166,13 +167,14 @@ void PERF_TEST::Test_operator_mul_perf(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int rand_data_range = 1;
constexpr int min_format_flag = 128;
vector<int> type{CV_32FC1};
for (size_t i = 0; i < type.size(); ++i) {
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src(val, val, type[i]);
Mat mat_dest(val, val, type[i]);

12
acl/test/test_correctness.cpp
View File

@@ -628,6 +628,8 @@ void AclMat_Test::Test_operator_sub(aclCxt *acl_context) {
const int rowsMax = 1048;
const int colsMax = 1048;
constexpr int lval = 1024;
constexpr int s_val1 = 1, s_val2 = 2, s_val3 = 3;
constexpr int s_val4 = 4, s_val6 = 6, s_val8 = 8;
vector<int> type{CV_8UC1, CV_8UC3, CV_32FC1, CV_32FC3,
CV_32SC1, CV_32SC3, CV_64FC1};
@@ -635,8 +637,8 @@ void AclMat_Test::Test_operator_sub(aclCxt *acl_context) {
test.PrintLog("Correctness test: Functoin: operator-=()", type[i]);
for (rows = lval; rows < rowsMax; rows++) {
for (cols = lval; cols < colsMax; cols++) {
Mat mat_src(rows, cols, type[i], Scalar(1, 2, 3));
Mat mat_dest(rows, cols, type[i], Scalar(4, 6, 8));
Mat mat_src(rows, cols, type[i], Scalar(s_val1, s_val2, s_val3));
Mat mat_dest(rows, cols, type[i], Scalar(s_val4, s_val6, s_val8));
Mat mat_dest1(rows, cols, type[i]);
aclMat aclmat_src(rows, cols, type[i], mat_src.data, acl_context,
@@ -664,6 +666,8 @@ void AclMat_Test::Test_operator_div(aclCxt *acl_context) {
const int rowsMax = 1048;
const int colsMax = 1048;
constexpr int lval = 1024;
constexpr int s_val1 = 1, s_val2 = 2;
constexpr int s_val4 = 4, s_val6 = 6, s_val8 = 8;
vector<int> type{CV_8UC1, CV_8UC3, CV_32FC1, CV_32FC3,
CV_32SC1, CV_32SC3, CV_64FC1};
@@ -671,8 +675,8 @@ void AclMat_Test::Test_operator_div(aclCxt *acl_context) {
test.PrintLog("Correctness test: Functoin: operator/=()", type[i]);
for (rows = lval; rows < rowsMax; rows++) {
for (cols = lval; cols < colsMax; cols++) {
Mat mat_src(rows, cols, type[i], Scalar(1, 2, 4));
Mat mat_dest(rows, cols, type[i], Scalar(4, 6, 8));
Mat mat_src(rows, cols, type[i], Scalar(s_val1, s_val2, s_val4));
Mat mat_dest(rows, cols, type[i], Scalar(s_val4, s_val6, s_val8));
Mat mat_dest1(rows, cols, type[i]);
aclMat aclmat_src(rows, cols, type[i], mat_src.data, acl_context,

17
acl/test/test_gemm.cpp
View File

@@ -14,12 +14,13 @@ void PERF_TEST::Test_MatMul(aclCxt *acl_context) {
double begin, end, time, acltime;
Common_Test test;
vector<int> type{CV_32FC1};
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int rand_data_range = 32;
constexpr int min_format_flag = 128;
for (size_t i = 0; i < type.size(); ++i) {
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
Mat mat_src(val, val, type[i]);
Mat mat_src1(val, val, type[i]);
Mat mat_dest(val, val, type[i]);
@@ -68,17 +69,21 @@ void PERF_TEST::Test_Convolution(aclCxt *acl_context) {
double begin, end, time, acltime;
Common_Test test;
vector<int> type{CV_32FC1};
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int min_format_flag = 128;
constexpr int s_val1 = 1, s_val2 = 2;
constexpr int s_val4 = 4, s_val6 = 6;
constexpr int kernel_val = 3;
for (size_t i = 0; i < type.size(); ++i) {
for (val = start_val; val <= valmax; val *= 2) {
Mat mat_src(val, val, type[i], Scalar{1, 2});
Mat mat_kernel(3, 3, type[i], Scalar(1, 4));
Mat mat_dest(val, val, type[i], Scalar{6});
for (val = start_val; val <= valmax; val *= base) {
Mat mat_src(val, val, type[i], Scalar{s_val1, s_val2});
Mat mat_kernel(kernel_val, kernel_val, type[i], Scalar(s_val1, s_val4));
Mat mat_dest(val, val, type[i], Scalar{s_val6});
aclMat aclmat_src(val, val, type[i], mat_src.data, acl_context);
aclMat aclmat_kernel(3, 3, type[i], mat_kernel.data, acl_context);
aclMat aclmat_kernel(kernel_val, kernel_val, type[i], mat_kernel.data, acl_context);
aclMat aclmat_dest(val, val, type[i], mat_dest.data, acl_context);
n = cycle_index;

82
acl/test/test_mathfuncs.cpp
View File

@@ -13,17 +13,19 @@ void PERF_TEST::Test_Abs(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int s_val_2 = -2, s_val_4 = -4, s_val_6 = -6;
constexpr int start_val = 8;
constexpr int min_format_flag = 128;
vector<int> type{CV_32FC1, CV_32SC1};
for (size_t i = 0; i < type.size(); ++i) {
test.PrintLog("Perf test : Function: Abs()", type[i]);
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src(val, val, type[i], Scalar{-2});
Mat mat_dest(val, val, type[i], Scalar{-4});
Mat mat_dest1(val, val, type[i], Scalar{-6});
Mat mat_src(val, val, type[i], Scalar{s_val_2});
Mat mat_dest(val, val, type[i], Scalar{s_val_4});
Mat mat_dest1(val, val, type[i], Scalar{s_val_6});
aclMat aclmat_src(val, val, type[i], mat_src.data, acl_context);
aclMat aclmat_dest(val, val, type[i], mat_dest.data, acl_context);
@@ -61,20 +63,22 @@ void PERF_TEST::Test_Pow(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int rand_data_range = 32;
constexpr int min_format_flag = 128;
vector<int> type{CV_8UC1, CV_32FC1, CV_32SC1};
for (size_t i = 0; i < type.size(); ++i) {
test.PrintLog("Perf test : Function: Pow()", type[i]);
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
int power = test.RandDom_(6);
Mat mat_src(val, val, type[i]);
Mat mat_dest(val, val, type[i]);
Mat mat_dest1(val, val, type[i]);
test.SetDataRange(mat_src, 32);
test.SetDataRange(mat_src, rand_data_range);
aclMat aclmat_src(val, val, type[i], mat_src.data, acl_context);
aclMat aclmat_dest(val, val, type[i], mat_dest.data, acl_context);
@@ -110,19 +114,21 @@ void PERF_TEST::Test_Sqrt(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int min_format_flag = 128;
constexpr int rand_data_range = 32;
vector<int> type{CV_32FC1, CV_64FC1};
for (size_t i = 0; i < type.size(); ++i) {
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src(val, val, type[i]);
Mat mat_dest(val, val, type[i]);
Mat mat_dest1(val, val, type[i]);
test.SetDataRange(mat_src, 32);
test.SetDataRange(mat_dest, 32);
test.SetDataRange(mat_src, rand_data_range);
test.SetDataRange(mat_dest, rand_data_range);
aclMat aclmat_src(val, val, type[i], mat_src.data, acl_context);
aclMat aclmat_dest(val, val, type[i], mat_dest.data, acl_context);
@@ -158,21 +164,23 @@ void PERF_TEST::Test_Add(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int min_format_flag = 128;
constexpr int rand_data_range = 32;
vector<int> type{CV_8UC1, CV_32FC1, CV_32SC1, CV_64FC1};
for (size_t i = 0; i < type.size(); ++i) {
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src1(val, val, type[i]);
Mat mat_src2(val, val, type[i]);
Mat mat_dest(val, val, type[i]);
Mat mat_dest1(val, val, type[i]);
test.SetDataRange(mat_src1, 32);
test.SetDataRange(mat_src2, 32);
test.SetDataRange(mat_dest, 32);
test.SetDataRange(mat_src1, rand_data_range);
test.SetDataRange(mat_src2, rand_data_range);
test.SetDataRange(mat_dest, rand_data_range);
aclMat aclmat_src1(val, val, type[i], mat_src1.data, acl_context);
aclMat aclmat_src2(val, val, type[i], mat_src2.data, acl_context);
@@ -211,21 +219,24 @@ void PERF_TEST::Test_Divide(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int min_format_flag = 128;
constexpr int rand_data_range1 = 32;
constexpr int rand_data_range2 = 4;
vector<int> type{CV_8UC1, CV_32FC1, CV_32SC1};
for (size_t i = 0; i < type.size(); ++i) {
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src1(val, val, type[i]);
Mat mat_src2(val, val, type[i]);
Mat mat_dest(val, val, type[i]);
Mat mat_dest1(val, val, type[i]);
test.SetDataRange(mat_src1, 32);
test.SetDataRange(mat_src2, 4);
test.SetDataRange(mat_dest, 32);
test.SetDataRange(mat_src1, rand_data_range1);
test.SetDataRange(mat_src2, rand_data_range2);
test.SetDataRange(mat_dest, rand_data_range1);
aclMat aclmat_src1(val, val, type[i], mat_src1.data, acl_context);
aclMat aclmat_src2(val, val, type[i], mat_src2.data, acl_context);
@@ -262,19 +273,22 @@ void PERF_TEST::Test_Exp(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int min_format_flag = 128;
constexpr int rand_data_range1 = 32;
constexpr int rand_data_range2 = 2;
vector<int> type{CV_32FC1, CV_64FC1};
for (size_t i = 0; i < type.size(); ++i) {
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src(val, val, type[i]);
Mat mat_dest(val, val, type[i]);
Mat mat_dest1(val, val, type[i]);
test.SetDataRange(mat_src, 32);
test.SetDataRange(mat_dest, 2);
test.SetDataRange(mat_src, rand_data_range1);
test.SetDataRange(mat_dest, rand_data_range2);
aclMat aclmat_src(val, val, type[i], mat_src.data, acl_context);
aclMat aclmat_dest(val, val, type[i], mat_dest.data, acl_context);
@@ -310,19 +324,21 @@ void PERF_TEST::Test_Log(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int min_format_flag = 128;
constexpr int rand_data_range = 32;
vector<int> type{CV_32FC1, CV_64FC1};
for (size_t i = 0; i < type.size(); ++i) {
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src(val, val, type[i]);
Mat mat_dest(val, val, type[i]);
Mat mat_dest1(val, val, type[i]);
test.SetDataRange(mat_src, 32);
test.SetDataRange(mat_dest, 32);
test.SetDataRange(mat_src, rand_data_range);
test.SetDataRange(mat_dest, rand_data_range);
aclMat aclmat_src(val, val, type[i], mat_src.data, acl_context);
aclMat aclmat_dest(val, val, type[i], mat_dest.data, acl_context);
@@ -358,21 +374,23 @@ void PERF_TEST::Test_Max(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int min_format_flag = 128;
constexpr int rand_data_range = 32;
vector<int> type{CV_32FC2, CV_32SC2, CV_64FC2};
for (size_t i = 0; i < type.size(); ++i) {
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src1(val, val, type[i]);
Mat mat_src2(val, val, type[i]);
Mat mat_dest(val, val, type[i]);
Mat mat_dest1(val, val, type[i]);
test.SetDataRange(mat_src1, 32);
test.SetDataRange(mat_src2, 32);
test.SetDataRange(mat_dest, 32);
test.SetDataRange(mat_src1, rand_data_range);
test.SetDataRange(mat_src2, rand_data_range);
test.SetDataRange(mat_dest, rand_data_range);
aclMat aclmat_src1(val, val, type[i], mat_src2.data, acl_context);
aclMat aclmat_src2(val, val, type[i], mat_src1.data, acl_context);
@@ -411,21 +429,23 @@ void PERF_TEST::Test_Min(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int min_format_flag = 128;
constexpr int rand_data_range = 32;
vector<int> type{CV_32FC3, CV_32SC3, CV_64FC3};
for (size_t i = 0; i < type.size(); ++i) {
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src1(val, val, type[i]);
Mat mat_src2(val, val, type[i]);
Mat mat_dest(val, val, type[i]);
Mat mat_dest1(val, val, type[i]);
test.SetDataRange(mat_src1, 32);
test.SetDataRange(mat_src2, 32);
test.SetDataRange(mat_dest, 32);
test.SetDataRange(mat_src1, rand_data_range);
test.SetDataRange(mat_src2, rand_data_range);
test.SetDataRange(mat_dest, rand_data_range);
aclMat aclmat_src1(val, val, type[i], mat_src2.data, acl_context);
aclMat aclmat_src2(val, val, type[i], mat_src1.data, acl_context);

33
acl/test/test_matrices.cpp
View File

@@ -13,9 +13,10 @@ void PERF_TEST::Test_Merge(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int rand_data_range = 32;
constexpr int s_val = 1;
constexpr int s_val1 = 1, s_val2 = 2, s_val3 = 3;
constexpr int min_format_flag = 128;
vector<int> srcType{CV_8UC1, CV_32FC1, CV_32SC1};
@@ -23,11 +24,11 @@ void PERF_TEST::Test_Merge(aclCxt *acl_context) {
vector<int> destType{CV_8UC3, CV_32FC3, CV_32SC3};
for (size_t i = 0; i < srcType.size(); ++i) {
test.PrintLog("Perf test : Function: merge()", srcType[i]);
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src1(val, val, srcType[i], Scalar(s_val));
Mat mat_src2(val, val, srcType[i], Scalar(s_val+1));
Mat mat_src3(val, val, srcType[i], Scalar(s_val+2));
Mat mat_src1(val, val, srcType[i], Scalar(s_val1));
Mat mat_src2(val, val, srcType[i], Scalar(s_val2));
Mat mat_src3(val, val, srcType[i], Scalar(s_val3));
Mat mat_dest(val, val, destType[i]);
Mat mat_dest1(val, val, destType[i]);
@@ -82,6 +83,7 @@ void PERF_TEST::Test_Transpose(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int rand_data_range = 32;
constexpr int min_format_flag = 128;
@@ -89,7 +91,7 @@ void PERF_TEST::Test_Transpose(aclCxt *acl_context) {
vector<int> type{CV_32FC1, CV_32SC1};
for (size_t i = 0; i < type.size(); ++i) {
test.PrintLog("Perf test : Function: transpose()", type[i]);
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src(val, val, type[i]);
Mat mat_dest(val, val, type[i]);
@@ -133,15 +135,17 @@ void PERF_TEST::Test_Split(aclCxt *acl_context) {
int cycle_index = 10;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int rand_data_range = 32;
constexpr int min_format_flag = 128;
constexpr int index0 = 0, index1 = 1, index2 = 2;
vector<int> srcType{CV_8UC3, CV_32FC3, CV_32SC3};
vector<int> destType{CV_8UC1, CV_32FC1, CV_32SC1};
for (size_t i = 0; i < srcType.size(); ++i) {
test.PrintLog("Perf test : Function: split()", srcType[i]);
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src(val, val, srcType[i]);
Mat mat_dest1(val, val, destType[i]);
@@ -179,13 +183,13 @@ void PERF_TEST::Test_Split(aclCxt *acl_context) {
end = static_cast<double>(getTickCount());
acltime = (end - begin) / getTickFrequency() / (cycle_index - 1);
(acl_dest.data())[0].download(mat_dest1);
(acl_dest.data())[1].download(mat_dest2);
(acl_dest.data())[2].download(mat_dest3);
(acl_dest.data())[index0].download(mat_dest1);
(acl_dest.data())[index1].download(mat_dest2);
(acl_dest.data())[index2].download(mat_dest3);
bool ret = test.Test_Diff((dest.data())[0], mat_dest1);
ret &= test.Test_Diff((dest.data())[1], mat_dest2);
ret &= test.Test_Diff((dest.data())[2], mat_dest3);
bool ret = test.Test_Diff((dest.data())[index0], mat_dest1);
ret &= test.Test_Diff((dest.data())[index1], mat_dest2);
ret &= test.Test_Diff((dest.data())[index2], mat_dest3);
ASSERT_TRUE(ret);
if (val < min_format_flag)
cout << "Shape: " << val << " x " << val << "\t\t";
@@ -203,6 +207,7 @@ void PERF_TEST::Test_Flip(aclCxt *acl_context) {
int cycle_index = 100;
double begin, end, time, acltime;
Common_Test test;
constexpr int base = 2;
constexpr int start_val = 8;
constexpr int rand_data_range = 32;
constexpr int min_format_flag = 128;
@@ -210,7 +215,7 @@ void PERF_TEST::Test_Flip(aclCxt *acl_context) {
vector<int> type{CV_8UC1, CV_32FC1, CV_32SC1, CV_64FC1};
for (size_t i = 0; i < type.size(); ++i) {
test.PrintLog("Perf test : Function: flip()", type[i]);
for (val = start_val; val <= valmax; val *= 2) {
for (val = start_val; val <= valmax; val *= base) {
n = cycle_index;
Mat mat_src(val, val, type[i]);
Mat mat_dest(val, val, type[i]);