* Copyright (c) 2026 Huawei Technologies Co., Ltd.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
* \file test_pow_operation.cpp
* \brief
*/
#include "test_operation.h"
using namespace tile_fwk::test_operation;
namespace {
struct PowOpFuncArgs : public OpFuncArgs {
PowOpFuncArgs(const std::vector<int64_t>& viewShape, const std::vector<int64_t> tileShape)
: viewShape_(viewShape), tileShape_(tileShape)
{}
std::vector<int64_t> viewShape_;
std::vector<int64_t> tileShape_;
};
struct PowOpMetaData {
explicit PowOpMetaData(const OpFunc& opFunc, const nlohmann::json& test_data)
: opFunc_(opFunc), test_data_(test_data)
{}
OpFunc opFunc_;
nlohmann::json test_data_;
};
static inline void DoPow(
const std::vector<int64_t>& shape0, const std::vector<int64_t>& shape1,
const std::vector<SymbolicScalar>& validShape0, const std::vector<SymbolicScalar>& validShape1,
const std::vector<SymbolicScalar>& offset0, const std::vector<SymbolicScalar>& offset1,
const std::vector<SymbolicScalar>& offset2, const Tensor& input0, const Tensor& input1, Tensor& output)
{
Tensor tileTensor0 = View(input0, shape0, validShape0, offset0);
Tensor tileTensor1 = View(input1, shape1, validShape1, offset1);
auto res = Pow(tileTensor0, tileTensor1);
Assemble(res, offset2, output);
}
template <size_t shapeDim, size_t inputSize>
static inline void CalcWillBroadcast(bool willBroadcast[][shapeDim], const std::vector<Tensor>& inputs)
{
constexpr int64_t broadcast = 1;
for (size_t i = 0; i < shapeDim; ++i) {
for (size_t j = 0; j < inputSize; ++j) {
willBroadcast[j][i] = inputs[j].GetShape()[i] == broadcast && inputs[j ^ 1].GetShape()[i] != broadcast;
}
}
}
static void PowOperationExeFunc2Dims(
const std::vector<Tensor>& inputs, std::vector<Tensor>& outputs, const OpFuncArgs* opArgs)
{
FUNCTION("main", {inputs[0], inputs[1]}, {outputs[0]})
{
constexpr size_t shapeDim = 2;
constexpr size_t inputSize = 2;
bool willBroadcast[inputSize][shapeDim] = {};
CalcWillBroadcast<shapeDim, inputSize>(willBroadcast, inputs);
SymbolicScalar firstDim = std::max(inputs[0].GetShape()[0], inputs[1].GetShape()[0]);
SymbolicScalar secondDim = std::max(inputs[0].GetShape()[1], inputs[1].GetShape()[1]);
const std::vector<Tensor> tensors = {inputs[0], inputs[1], outputs[0]};
auto args = static_cast<const PowOpFuncArgs*>(opArgs);
const int64_t firstViewShape = args->viewShape_[0];
const int64_t secondViewShape = args->viewShape_[1];
const std::vector<int64_t> shape0 = {
willBroadcast[0][0] ? 1 : firstViewShape, willBroadcast[0][1] ? 1 : secondViewShape};
const std::vector<int64_t> shape1 = {
willBroadcast[1][0] ? 1 : firstViewShape, willBroadcast[1][1] ? 1 : secondViewShape};
const int bloop = CeilDiv(firstDim, firstViewShape);
const int sloop = CeilDiv(secondDim, secondViewShape);
LOOP("LOOP_L0_bIdx", FunctionType::DYNAMIC_LOOP, bIdx, LoopRange(0, bloop, 1))
{
LOOP("LOOP_L1_sIdx", FunctionType::DYNAMIC_LOOP, sIdx, LoopRange(0, sloop, 1))
{
std::vector<SymbolicScalar> offset0 = {
willBroadcast[0][0] ? SymbolicScalar(0) : bIdx * firstViewShape,
willBroadcast[0][1] ? SymbolicScalar(0) : sIdx * secondViewShape};
std::vector<SymbolicScalar> offset1 = {
willBroadcast[1][0] ? SymbolicScalar(0) : bIdx * firstViewShape,
willBroadcast[1][1] ? SymbolicScalar(0) : sIdx * secondViewShape};
std::vector<SymbolicScalar> offset2 = {bIdx * firstViewShape, sIdx * secondViewShape};
std::vector<SymbolicScalar> validShape0 = {
willBroadcast[0][0] ? SymbolicScalar(1) :
std::min(firstDim - bIdx * firstViewShape, firstViewShape),
willBroadcast[0][1] ? SymbolicScalar(1) :
std::min(secondDim - sIdx * secondViewShape, secondViewShape)};
std::vector<SymbolicScalar> validShape1 = {
willBroadcast[1][0] ? SymbolicScalar(1) :
std::min(firstDim - bIdx * firstViewShape, firstViewShape),
willBroadcast[1][1] ? SymbolicScalar(1) :
std::min(secondDim - sIdx * secondViewShape, secondViewShape)};
TileShape::Current().SetVecTile(args->tileShape_);
DoPow(
shape0, shape1, validShape0, validShape1, offset0, offset1, offset2, inputs[0], inputs[1],
outputs[0]);
}
}
}
}
static void PowOperationExeFunc3Dims(
const std::vector<Tensor>& inputs, std::vector<Tensor>& outputs, const OpFuncArgs* opArgs)
{
FUNCTION("main", {inputs[0], inputs[1]}, {outputs[0]})
{
constexpr size_t shapeDim = 3;
constexpr size_t inputSize = 2;
bool willBroadcast[inputSize][shapeDim] = {};
CalcWillBroadcast<shapeDim, inputSize>(willBroadcast, inputs);
SymbolicScalar firstDim = std::max(inputs[0].GetShape()[0], inputs[1].GetShape()[0]);
SymbolicScalar secondDim = std::max(inputs[0].GetShape()[1], inputs[1].GetShape()[1]);
SymbolicScalar thirdDim = std::max(inputs[0].GetShape()[2], inputs[1].GetShape()[2]);
const std::vector<Tensor> tensors = {inputs[0], inputs[1], outputs[0]};
auto args = static_cast<const PowOpFuncArgs*>(opArgs);
const int64_t firstViewShape = args->viewShape_[0];
const int64_t secondViewShape = args->viewShape_[1];
const int64_t thirdViewShape = args->viewShape_[2];
const std::vector<int64_t> shape0 = {
willBroadcast[0][0] ? 1 : firstViewShape, willBroadcast[0][1] ? 1 : secondViewShape,
willBroadcast[0][2] ? 1 : thirdViewShape};
const std::vector<int64_t> shape1 = {
willBroadcast[1][0] ? 1 : firstViewShape, willBroadcast[1][1] ? 1 : secondViewShape,
willBroadcast[1][2] ? 1 : thirdViewShape};
const int bloop = CeilDiv(firstDim, firstViewShape);
const int sloop = CeilDiv(secondDim, secondViewShape);
const int mloop = CeilDiv(thirdDim, thirdViewShape);
LOOP("LOOP_L0_bIdx", FunctionType::DYNAMIC_LOOP, bIdx, LoopRange(0, bloop, 1))
{
LOOP("LOOP_L1_sIdx", FunctionType::DYNAMIC_LOOP, sIdx, LoopRange(0, sloop, 1))
{
LOOP("LOOP_L2_mIdx", FunctionType::DYNAMIC_LOOP, mIdx, LoopRange(0, mloop, 1))
{
std::vector<SymbolicScalar> offset0 = {
willBroadcast[0][0] ? SymbolicScalar(0) : bIdx * firstViewShape,
willBroadcast[0][1] ? SymbolicScalar(0) : sIdx * secondViewShape,
willBroadcast[0][2] ? SymbolicScalar(0) : mIdx * thirdViewShape};
std::vector<SymbolicScalar> offset1 = {
willBroadcast[1][0] ? SymbolicScalar(0) : bIdx * firstViewShape,
willBroadcast[1][1] ? SymbolicScalar(0) : sIdx * secondViewShape,
willBroadcast[1][2] ? SymbolicScalar(0) : mIdx * thirdViewShape};
std::vector<SymbolicScalar> offset2 = {
bIdx * firstViewShape, sIdx * secondViewShape, mIdx * thirdViewShape};
std::vector<SymbolicScalar> validShape0 = {
willBroadcast[0][0] ? SymbolicScalar(1) :
std::min(firstDim - bIdx * firstViewShape, firstViewShape),
willBroadcast[0][1] ? SymbolicScalar(1) :
std::min(secondDim - sIdx * secondViewShape, secondViewShape),
willBroadcast[0][2] ? SymbolicScalar(1) :
std::min(thirdDim - mIdx * thirdViewShape, thirdViewShape)};
std::vector<SymbolicScalar> validShape1 = {
willBroadcast[1][0] ? SymbolicScalar(1) :
std::min(firstDim - bIdx * firstViewShape, firstViewShape),
willBroadcast[1][1] ? SymbolicScalar(1) :
std::min(secondDim - sIdx * secondViewShape, secondViewShape),
willBroadcast[1][2] ? SymbolicScalar(1) :
std::min(thirdDim - mIdx * thirdViewShape, thirdViewShape)};
TileShape::Current().SetVecTile(args->tileShape_);
DoPow(
shape0, shape1, validShape0, validShape1, offset0, offset1, offset2, inputs[0], inputs[1],
outputs[0]);
}
}
}
}
}
static void PowOperationExeFunc4Dims(
const std::vector<Tensor>& inputs, std::vector<Tensor>& outputs, const OpFuncArgs* opArgs)
{
FUNCTION("main", {inputs[0], inputs[1]}, {outputs[0]})
{
constexpr size_t shapeDim = 4;
constexpr size_t inputSize = 2;
bool willBroadcast[inputSize][shapeDim] = {};
CalcWillBroadcast<shapeDim, inputSize>(willBroadcast, inputs);
SymbolicScalar firstDim = std::max(inputs[0].GetShape()[0], inputs[1].GetShape()[0]);
SymbolicScalar secondDim = std::max(inputs[0].GetShape()[1], inputs[1].GetShape()[1]);
SymbolicScalar thirdDim = std::max(inputs[0].GetShape()[2], inputs[1].GetShape()[2]);
SymbolicScalar fourthDim = std::max(inputs[0].GetShape()[3], inputs[1].GetShape()[3]);
const std::vector<Tensor> tensors = {inputs[0], inputs[1], outputs[0]};
auto args = static_cast<const PowOpFuncArgs*>(opArgs);
const int64_t firstViewShape = args->viewShape_[0];
const int64_t secondViewShape = args->viewShape_[1];
const int64_t thirdViewShape = args->viewShape_[2];
const int64_t fourthViewShape = args->viewShape_[3];
const std::vector<int64_t> shape0 = {
willBroadcast[0][0] ? 1 : firstViewShape, willBroadcast[0][1] ? 1 : secondViewShape,
willBroadcast[0][2] ? 1 : thirdViewShape, willBroadcast[0][3] ? 1 : fourthViewShape};
const std::vector<int64_t> shape1 = {
willBroadcast[1][0] ? 1 : firstViewShape, willBroadcast[1][1] ? 1 : secondViewShape,
willBroadcast[1][2] ? 1 : thirdViewShape, willBroadcast[1][3] ? 1 : fourthViewShape};
const int bloop = CeilDiv(firstDim, firstViewShape);
const int sloop = CeilDiv(secondDim, secondViewShape);
const int mloop = CeilDiv(thirdDim, thirdViewShape);
const int nloop = CeilDiv(fourthDim, fourthViewShape);
LOOP("LOOP_L0_bIdx", FunctionType::DYNAMIC_LOOP, bIdx, LoopRange(0, bloop, 1))
{
LOOP("LOOP_L1_sIdx", FunctionType::DYNAMIC_LOOP, sIdx, LoopRange(0, sloop, 1))
{
LOOP("LOOP_L2_mIdx", FunctionType::DYNAMIC_LOOP, mIdx, LoopRange(0, mloop, 1))
{
LOOP("LOOP_L3_nIdx", FunctionType::DYNAMIC_LOOP, nIdx, LoopRange(0, nloop, 1))
{
std::vector<SymbolicScalar> offset0 = {
willBroadcast[0][0] ? SymbolicScalar(0) : bIdx * firstViewShape,
willBroadcast[0][1] ? SymbolicScalar(0) : sIdx * secondViewShape,
willBroadcast[0][2] ? SymbolicScalar(0) : mIdx * thirdViewShape,
willBroadcast[0][3] ? SymbolicScalar(0) : nIdx * fourthViewShape};
std::vector<SymbolicScalar> offset1 = {
willBroadcast[1][0] ? SymbolicScalar(0) : bIdx * firstViewShape,
willBroadcast[1][1] ? SymbolicScalar(0) : sIdx * secondViewShape,
willBroadcast[1][2] ? SymbolicScalar(0) : mIdx * thirdViewShape,
willBroadcast[1][3] ? SymbolicScalar(0) : nIdx * fourthViewShape};
std::vector<SymbolicScalar> offset2 = {
bIdx * firstViewShape, sIdx * secondViewShape, mIdx * thirdViewShape,
nIdx * fourthViewShape};
std::vector<SymbolicScalar> validShape0 = {
willBroadcast[0][0] ? SymbolicScalar(1) :
std::min(firstDim - bIdx * firstViewShape, firstViewShape),
willBroadcast[0][1] ? SymbolicScalar(1) :
std::min(secondDim - sIdx * secondViewShape, secondViewShape),
willBroadcast[0][2] ? SymbolicScalar(1) :
std::min(thirdDim - mIdx * thirdViewShape, thirdViewShape),
willBroadcast[0][3] ? SymbolicScalar(1) :
std::min(fourthDim - nIdx * fourthViewShape, fourthViewShape)};
std::vector<SymbolicScalar> validShape1 = {
willBroadcast[1][0] ? SymbolicScalar(1) :
std::min(firstDim - bIdx * firstViewShape, firstViewShape),
willBroadcast[1][1] ? SymbolicScalar(1) :
std::min(secondDim - sIdx * secondViewShape, secondViewShape),
willBroadcast[1][2] ? SymbolicScalar(1) :
std::min(thirdDim - mIdx * thirdViewShape, thirdViewShape),
willBroadcast[1][3] ? SymbolicScalar(1) :
std::min(fourthDim - nIdx * fourthViewShape, fourthViewShape)};
TileShape::Current().SetVecTile(args->tileShape_);
DoPow(
shape0, shape1, validShape0, validShape1, offset0, offset1, offset2, inputs[0], inputs[1],
outputs[0]);
}
}
}
}
}
}
class PowOperationTest : public npu::tile_fwk::stest::TestSuite_STest_Ops_Aihac_param<PowOpMetaData> {};
INSTANTIATE_TEST_SUITE_P(
TestPow, PowOperationTest,
::testing::ValuesIn(GetOpMetaData<PowOpMetaData>(
{PowOperationExeFunc2Dims, PowOperationExeFunc3Dims, PowOperationExeFunc4Dims}, "Pow")));
TEST_P(PowOperationTest, TestPow)
{
auto test_data = GetParam().test_data_;
auto args = PowOpFuncArgs(GetViewShape(test_data), GetTileShape(test_data));
auto testCase = CreateTestCaseDesc<PowOpMetaData>(GetParam(), &args);
TestExecutor::runTest(testCase);
}
}
