forked from jiuyuan/InfiniTensor
45 lines
1.7 KiB
C++
45 lines
1.7 KiB
C++
#include "core/graph.h"
|
|
#include "core/run_enigne.h"
|
|
#include "test.h"
|
|
|
|
namespace infini {
|
|
|
|
TEST(Graph, build_and_run) {
|
|
Graph g = make_ref<GraphNode>();
|
|
Tensor i0 = g->addTensor({1, 2, 3}, DataType::Int32);
|
|
Tensor w0 = g->addTensor({1, 3, 4}, DataType::Int32);
|
|
Tensor o0 = g->addTensor({1, 2, 4}, DataType::Int32);
|
|
g->dataMalloc();
|
|
i0->copyData(vector<VType>{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12}.data());
|
|
w0->copyData(vector<VType>{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12}.data());
|
|
g->addOp(make_ref<MatmulNode>(i0, w0, o0));
|
|
RunEngine(Device::CPU).run(g);
|
|
// check answer
|
|
auto ans = make_ref<TensorNode>(Shape{1, 2, 4}, DataType::Int32);
|
|
ans->dataMalloc();
|
|
ans->copyData(vector<VType>{38, 44, 50, 56, 83, 98, 113, 128}.data());
|
|
EXPECT_TRUE(o0->equalData(ans));
|
|
}
|
|
|
|
TEST(Graph, perf_engine) {
|
|
Graph g = make_ref<GraphNode>();
|
|
Tensor i0 = g->addTensor({1, 2, 3}, DataType::Int32);
|
|
Tensor w0 = g->addTensor({1, 3, 4}, DataType::Int32);
|
|
Tensor o0 = g->addTensor({1, 2, 4}, DataType::Int32);
|
|
g->dataMalloc();
|
|
i0->copyData(vector<VType>{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12}.data());
|
|
w0->copyData(vector<VType>{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12}.data());
|
|
g->addOp(make_ref<MatmulNode>(i0, w0, o0));
|
|
RunEngine(Device::CPU).run(g, true, true);
|
|
double perfTime = RunEngine(Device::CPU).getPerfTime(g);
|
|
// The example matmul takes 0.0036ms with one core
|
|
EXPECT_GT(perfTime, 0);
|
|
EXPECT_LT(perfTime, 0.01);
|
|
// check answer
|
|
auto ans = make_ref<TensorNode>(Shape{1, 2, 4}, DataType::Int32);
|
|
ans->dataMalloc();
|
|
ans->copyData(vector<VType>{38, 44, 50, 56, 83, 98, 113, 128}.data());
|
|
EXPECT_TRUE(o0->equalData(ans));
|
|
}
|
|
|
|
} // namespace infini
|