forked from jiuyuan/InfiniTensor
49 lines
1.4 KiB
C++
49 lines
1.4 KiB
C++
#include "core/graph.h"
|
|
#include "core/kernel.h"
|
|
#include "core/runtime.h"
|
|
#include "cuda/cuda_runtime.h"
|
|
#include "cuda/cuda_utility.h"
|
|
#include "operators/unary.h"
|
|
|
|
#include "test.h"
|
|
|
|
namespace infini {
|
|
|
|
template <class T>
|
|
void testClip(const std::function<void(void *, size_t, DataType)> &generator,
|
|
const Shape &shape) {
|
|
// Runtime
|
|
Runtime cpuRuntime = NativeCpuRuntimeObj::getInstance();
|
|
auto cudaRuntime = make_ref<CudaRuntimeObj>();
|
|
|
|
// Build input data on CPU
|
|
Tensor inputCpu = make_ref<TensorObj>(shape, DataType::Float32, cpuRuntime);
|
|
inputCpu->dataMalloc();
|
|
inputCpu->setData(generator);
|
|
|
|
// GPU
|
|
Graph cudaGraph = make_ref<GraphObj>(cudaRuntime);
|
|
auto inputGpu = cudaGraph->cloneTensor(inputCpu);
|
|
float min = 2.0;
|
|
float max = 4.0;
|
|
auto gpuOp = cudaGraph->addOp<T>(inputGpu, nullptr, min, max);
|
|
cudaGraph->dataMalloc();
|
|
cudaRuntime->run(cudaGraph);
|
|
auto outputGpu = gpuOp->getOutput();
|
|
auto outputGpu2Cpu = outputGpu->clone(cpuRuntime);
|
|
// CPU
|
|
Graph cpuGraph = make_ref<GraphObj>(cpuRuntime);
|
|
auto cpuOp = cpuGraph->addOp<T>(inputCpu, nullptr, min, max);
|
|
cpuGraph->dataMalloc();
|
|
cpuRuntime->run(cpuGraph);
|
|
auto outputCpu = cpuOp->getOutput();
|
|
// Check
|
|
EXPECT_TRUE(outputCpu->equalData(outputGpu2Cpu));
|
|
}
|
|
|
|
TEST(cuDNN_Unary, run) {
|
|
testClip<ClipObj>(IncrementalGenerator(), Shape{1, 2, 2, 3});
|
|
}
|
|
|
|
} // namespace infini
|