* Copyright (c) 2025 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.
*/
#include "mul_kernel.h"
#include <memory>
#include <set>
#include "framework/common/debug/log.h"
#include "common/math/math_util.h"
#include "framework/common/framework_types_internal.h"
#include "framework/common/util.h"
#include "framework/common/debug/ge_log.h"
#include "framework/common/ge_inner_error_codes.h"
#include "common/b_cast/b_cast.h"
#include "common/checker.h"
#include "graph/utils/type_utils.h"
#include "host_kernels/kernel_factory.h"
namespace ge {
namespace {
const std::set<DataType> kMulSupportedType = {DT_INT8, DT_INT16, DT_INT32, DT_INT64, DT_UINT8,
DT_UINT16, DT_UINT32, DT_UINT64, DT_FLOAT16, DT_FLOAT,
DT_DOUBLE, DT_COMPLEX32, DT_COMPLEX64, DT_COMPLEX128};
template <typename T>
Status OverflowCheckMul(T const &x, T const &y, DataType const &type) {
switch (type) {
case DT_INT8:
FMK_INT8_MULCHECK(x, y)
break;
case DT_INT16:
FMK_INT16_MULCHECK(x, y)
break;
case DT_INT32:
FMK_INT32_MULCHECK(x, y)
break;
case DT_INT64:
FMK_INT64_MULCHECK(x, y)
break;
case DT_UINT8:
FMK_UINT8_MULCHECK(x, y)
break;
case DT_UINT16:
FMK_UINT16_MULCHECK(x, y)
break;
case DT_UINT32:
FMK_UINT32_MULCHECK(x, y)
break;
case DT_UINT64:
FMK_UINT64_MULCHECK(x, y)
break;
case DT_FLOAT16:
case DT_COMPLEX32:
FMK_FP16_MULCHECK(x, y)
break;
case DT_FLOAT:
case DT_COMPLEX64:
FMK_FLOAT_MULCHECK(x, y)
break;
case DT_DOUBLE:
case DT_COMPLEX128:
FMK_DOUBLE_MULCHECK(x, y)
break;
default:
break;
}
return SUCCESS;
}
template <typename T>
Status OverflowCheckAdd(T const &x, T const &y, DataType const &type) {
switch (type) {
case DT_COMPLEX32:
FMK_FP16_ADDCHECK(x, y)
break;
case DT_COMPLEX64:
FMK_FLOAT_ADDCHECK(x, y)
break;
case DT_COMPLEX128:
FMK_DOUBLE_ADDCHECK(x, y)
break;
default:
break;
}
return SUCCESS;
}
template <typename T>
Status OverflowCheckSub(T const &x, T const &y, DataType const &type) {
switch (type) {
case DT_COMPLEX32:
FMK_FP16_SUBCHECK(x, y)
break;
case DT_COMPLEX64:
FMK_FLOAT_SUBCHECK(x, y)
break;
case DT_COMPLEX128:
FMK_DOUBLE_SUBCHECK(x, y)
break;
default:
break;
}
return SUCCESS;
}
#define DEFINE_FUNC_WITH_STATUS_BY_TYPE(TYPE) \
std::function<TYPE(TYPE const &, TYPE const &, DataType &, Status &)> func_##TYPE = \
[](TYPE const &a, TYPE const &b, DataType &type, Status &ret) -> TYPE { \
ret = OverflowCheckMul(a, b, type); \
if (ret != SUCCESS) { \
GELOGE(PARAM_INVALID, "Result of mul is overflow."); \
return static_cast<TYPE>(0); \
} \
return static_cast<TYPE>(a) * static_cast<TYPE>(b); \
}
#define SET_BCAST_COMPUTE_CASE(DTYPE, TYPE) \
case DTYPE: \
ret = bcast.BCastComputeCheck(input, y_data_##TYPE##_, func_##TYPE); \
break
#define SET_OUTPUT(DTYPE, TYPE) \
case DTYPE: \
(void)output_ptr->SetData(reinterpret_cast<uint8_t *>(y_data_##TYPE##_.data()), y_data_##TYPE##_.size() * length); \
break
DEFINE_FUNC_WITH_STATUS_BY_TYPE(int8_t);
DEFINE_FUNC_WITH_STATUS_BY_TYPE(int16_t);
DEFINE_FUNC_WITH_STATUS_BY_TYPE(int32_t);
DEFINE_FUNC_WITH_STATUS_BY_TYPE(int64_t);
DEFINE_FUNC_WITH_STATUS_BY_TYPE(uint8_t);
DEFINE_FUNC_WITH_STATUS_BY_TYPE(uint16_t);
DEFINE_FUNC_WITH_STATUS_BY_TYPE(uint32_t);
DEFINE_FUNC_WITH_STATUS_BY_TYPE(uint64_t);
DEFINE_FUNC_WITH_STATUS_BY_TYPE(fp16_t);
DEFINE_FUNC_WITH_STATUS_BY_TYPE(float);
DEFINE_FUNC_WITH_STATUS_BY_TYPE(double);
template <typename InT>
Status ComplexCompute(const OpDescPtr op_desc_ptr, const std::vector<ConstGeTensorPtr> &input,
std::vector<GeTensorPtr> &v_output) {
BCast bcast;
Status ret = bcast.GenerateBcastInfo(BCast::TransShapeToDimVec(input[0U]->GetTensorDesc()),
BCast::TransShapeToDimVec(input[1U]->GetTensorDesc()));
GE_ASSERT_SUCCESS(ret, "Greater broadcasting failed.");
std::vector<int64_t> x_indexes;
std::vector<int64_t> y_indexes;
bcast.BCastIndexes(x_indexes, y_indexes);
auto x1_data = reinterpret_cast<const InT *>(input[0U]->GetData().data());
auto x2_data = reinterpret_cast<const InT *>(input[1U]->GetData().data());
size_t data_num = x_indexes.size();
constexpr int64_t kComplexWidth = 2;
const size_t value_num = data_num * kComplexWidth;
std::unique_ptr<InT[]> buf = ge::MakeUnique<InT[]>(value_num);
GE_CHECK_NOTNULL(buf, "New sizeof(T) * data_num(%zu) memory failed", static_cast<size_t>(sizeof(InT) * value_num));
DataType data_type = input[0U]->GetTensorDesc().GetDataType();
GE_ASSERT_TRUE((value_num * sizeof(InT)) == input[0U]->GetData().size(),
"complex value size should be an integer multiple of 2.");
for (size_t i = 0U; i < data_num; ++i) {
auto x_index_real = x1_data[x_indexes[i] * kComplexWidth];
auto x_index_imaginary = x1_data[(x_indexes[i] * kComplexWidth) + 1U];
auto y_index_real = x2_data[y_indexes[i] * kComplexWidth];
auto y_index_imaginary = x2_data[(y_indexes[i] * kComplexWidth) + 1U];
if ((OverflowCheckMul<InT>(x_index_real, y_index_real, data_type) != SUCCESS) ||
(OverflowCheckMul<InT>(x_index_real, y_index_imaginary, data_type) != SUCCESS) ||
(OverflowCheckMul<InT>(x_index_imaginary, y_index_real, data_type) != SUCCESS) ||
(OverflowCheckMul<InT>(x_index_imaginary, y_index_imaginary, data_type) != SUCCESS)) {
GELOGE(PARAM_INVALID, "Result of mul is overflow.");
return PARAM_INVALID;
}
auto xr_yr = x_index_real * y_index_real;
auto xi_yi = x_index_imaginary * y_index_imaginary;
auto xr_yi = x_index_real * y_index_imaginary;
auto xi_yr = x_index_imaginary * y_index_real;
if ((OverflowCheckSub<InT>(xr_yr, xi_yi, data_type) != SUCCESS) ||
(OverflowCheckAdd<InT>(xr_yi, xi_yr, data_type) != SUCCESS)) {
GELOGE(PARAM_INVALID, "Result of mul is overflow.");
return PARAM_INVALID;
}
buf[i * kComplexWidth]= xr_yr - xi_yi;
buf[(i * kComplexWidth) + 1UL] = xr_yi + xi_yr;
}
GeTensorPtr output_ptr = MakeShared<GeTensor>(op_desc_ptr->GetOutputDesc(0));
GE_CHECK_NOTNULL(output_ptr, "Make shared failed");
output_ptr->SetData(reinterpret_cast<uint8_t *>(buf.get()), value_num * sizeof(InT));
output_ptr->MutableTensorDesc().SetDataType(data_type);
output_ptr->MutableTensorDesc().SetShape(GeShape(bcast.GetOutputShape()));
v_output.push_back(output_ptr);
GELOGD("MulKernel success");
return SUCCESS;
}
}
Status MulKernel::Compute(const OpDescPtr op_desc_ptr, const std::vector<ConstGeTensorPtr> &input,
std::vector<GeTensorPtr> &v_output) {
GELOGD("MulKernel in");
if (op_desc_ptr == nullptr) {
GELOGE(PARAM_INVALID, "Parameter's invalid, input opDescPtr is nullptr.");
return PARAM_INVALID;
}
Status ret = MulCheck(input);
if (ret != SUCCESS) {
return ret;
}
DataType data_type = input[0]->GetTensorDesc().GetDataType();
BCast bcast;
switch (data_type) {
case DT_COMPLEX32:
return ComplexCompute<fp16_t>(op_desc_ptr, input, v_output);
case DT_COMPLEX64:
return ComplexCompute<float>(op_desc_ptr, input, v_output);
case DT_COMPLEX128:
return ComplexCompute<double>(op_desc_ptr, input, v_output);
SET_BCAST_COMPUTE_CASE(DT_INT8, int8_t);
SET_BCAST_COMPUTE_CASE(DT_INT16, int16_t);
SET_BCAST_COMPUTE_CASE(DT_INT32, int32_t);
SET_BCAST_COMPUTE_CASE(DT_INT64, int64_t);
SET_BCAST_COMPUTE_CASE(DT_UINT8, uint8_t);
SET_BCAST_COMPUTE_CASE(DT_UINT16, uint16_t);
SET_BCAST_COMPUTE_CASE(DT_UINT32, uint32_t);
SET_BCAST_COMPUTE_CASE(DT_UINT64, uint64_t);
SET_BCAST_COMPUTE_CASE(DT_FLOAT16, fp16_t);
SET_BCAST_COMPUTE_CASE(DT_FLOAT, float);
SET_BCAST_COMPUTE_CASE(DT_DOUBLE, double);
default:
ret = NOT_CHANGED;
break;
}
if (ret != SUCCESS) {
GELOGW("BCastCompute fail, data_type: %s, ret: %s", TypeUtils::DataTypeToSerialString(data_type).c_str(),
GET_ERRORNO_STR(ret).c_str());
return NOT_CHANGED;
}
uint32_t length = 1;
if (!TypeUtils::GetDataTypeLength(data_type, length)) {
GELOGW("Can't GetDataTypeLength of data_type: %s", TypeUtils::DataTypeToSerialString(data_type).c_str());
return NOT_CHANGED;
}
GeTensorPtr output_ptr = MakeShared<GeTensor>(op_desc_ptr->GetOutputDesc(0));
if (output_ptr == nullptr) {
GELOGE(MEMALLOC_FAILED, "Make shared failed");
return MEMALLOC_FAILED;
}
output_ptr->MutableTensorDesc().SetShape(GeShape(bcast.GetOutputShape()));
switch (data_type) {
SET_OUTPUT(DT_INT8, int8_t);
SET_OUTPUT(DT_INT16, int16_t);
SET_OUTPUT(DT_INT32, int32_t);
SET_OUTPUT(DT_INT64, int64_t);
SET_OUTPUT(DT_UINT8, uint8_t);
SET_OUTPUT(DT_UINT16, uint16_t);
SET_OUTPUT(DT_UINT32, uint32_t);
SET_OUTPUT(DT_UINT64, uint64_t);
SET_OUTPUT(DT_FLOAT16, fp16_t);
SET_OUTPUT(DT_FLOAT, float);
SET_OUTPUT(DT_DOUBLE, double);
default:
break;
}
output_ptr->MutableTensorDesc().SetDataType(data_type);
v_output.push_back(output_ptr);
GELOGD("MulKernel success");
return SUCCESS;
}
Status MulKernel::MulCheck(const std::vector<ConstGeTensorPtr> &input) const {
if (input.size() != static_cast<size_t>(MUL_INPUT_NUM)) {
GELOGI("The number of input for Mul must be %u.", MUL_INPUT_NUM);
return NOT_CHANGED;
}
ConstGeTensorPtr input_x1 = input.at(0);
ConstGeTensorPtr input_x2 = input.at(1);
GE_CHECK_NOTNULL(input_x1);
GE_CHECK_NOTNULL(input_x2);
if (input_x1->GetData().size() == 0 || input_x2->GetData().size() == 0) {
GELOGI("Data size check skipped: empty input. x1: %zu, x2: %zu", input_x1->GetData().size(), input_x2->GetData().size());
return NOT_CHANGED;
}
DataType type = input_x1->GetTensorDesc().GetDataType();
if (type != input_x2->GetTensorDesc().GetDataType()) {
GELOGI("Data type of inputs for Mul not matched.");
return NOT_CHANGED;
}
if (kMulSupportedType.find(type) == kMulSupportedType.end()) {
GELOGI("Mul does not support this Data type: %s", TypeUtils::DataTypeToSerialString(type).c_str());
return NOT_CHANGED;
}
return SUCCESS;
}
REGISTER_COMPUTE_NODE_KERNEL(MUL, MulKernel);
}
