forked from jiuyuan/InfiniTensor
44 lines
1.2 KiB
C++
44 lines
1.2 KiB
C++
#include "core/graph.h"
|
|
#include "core/kernel.h"
|
|
#include "core/runtime.h"
|
|
#include "kunlun/kunlun_runtime.h"
|
|
#include "operators/transpose.h"
|
|
|
|
#include "test.h"
|
|
|
|
namespace infini {
|
|
|
|
template <class T>
|
|
void testTranspose(
|
|
const std::function<void(void *, size_t, DataType)> &generator,
|
|
const Shape &shape) {
|
|
// Runtime
|
|
Runtime cpuRuntime = NativeCpuRuntimeObj::getInstance();
|
|
auto xpuRuntime = make_ref<KUNLUNRuntimeObj>();
|
|
|
|
// Build input data on CPU
|
|
Tensor inputCpu = make_ref<TensorObj>(shape, DataType::Float32, cpuRuntime);
|
|
inputCpu->dataMalloc();
|
|
inputCpu->setData(generator);
|
|
|
|
// GPU
|
|
Graph xpuGraph = make_ref<GraphObj>(xpuRuntime);
|
|
auto inputGpu = xpuGraph->cloneTensor(inputCpu);
|
|
vector<int> permute = {0, 1, 3, 2};
|
|
auto gpuOp = xpuGraph->addOp<T>(inputGpu, nullptr, permute);
|
|
xpuGraph->dataMalloc();
|
|
xpuRuntime->run(xpuGraph);
|
|
auto outputGpu = gpuOp->getOutput();
|
|
auto outputGpu2Cpu = outputGpu->clone(cpuRuntime);
|
|
// Check
|
|
inputCpu->printData();
|
|
outputGpu2Cpu->printData();
|
|
EXPECT_TRUE(1);
|
|
}
|
|
|
|
TEST(xpu_Transpose, run) {
|
|
testTranspose<TransposeObj>(IncrementalGenerator(), Shape{1, 1, 2, 3});
|
|
}
|
|
|
|
} // namespace infini
|