* 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.
*
* Generated By CANNBot
*/
#include <iostream>
#include <fstream>
#include <string.h>
#include <stdint.h>
#include <vector>
#include <string>
#include <map>
#include <memory>
#include <new>
#include "assert.h"
#include "graph.h"
#include "types.h"
#include "tensor.h"
#include "ge_error_codes.h"
#include "ge_api_types.h"
#include "ge_api.h"
#include "array_ops.h"
#include "ge_ir_build.h"
#include "nn_other.h"
#include "../op_graph/ndtri_proto.h"
#define FAILED -1
#define SUCCESS 0
using namespace ge;
using std::map;
using std::string;
using std::vector;
#define ADD_INPUT(intputIndex, intputName, intputDtype, inputShape) \
vector<int64_t> placeholder##intputIndex##_shape = inputShape; \
auto placeholder##intputIndex = op::Data("placeholder" + intputIndex).set_attr_index(0); \
TensorDesc placeholder##intputIndex##_desc = TensorDesc(ge::Shape(placeholder##intputIndex##_shape), FORMAT_ND, \
intputDtype); \
placeholder##intputIndex##_desc.SetPlacement(ge::kPlacementHost); \
placeholder##intputIndex##_desc.SetFormat(FORMAT_ND); \
Tensor tensor_placeholder##intputIndex; \
ret = GenOnesDataFloat32(placeholder##intputIndex##_shape, tensor_placeholder##intputIndex, \
placeholder##intputIndex##_desc, 0.5f); \
if (ret != SUCCESS) { \
printf("%s - ERROR - [XIR]: Generate input data failed\n", GetTime().c_str()); \
return FAILED; \
} \
placeholder##intputIndex.update_input_desc_x(placeholder##intputIndex##_desc); \
placeholder##intputIndex.update_output_desc_y(placeholder##intputIndex##_desc); \
input.push_back(tensor_placeholder##intputIndex); \
graph.AddOp(placeholder##intputIndex); \
ndtri_op.set_input_##intputName(placeholder##intputIndex); \
inputs.push_back(placeholder##intputIndex);
#define LOG_PRINT(message, ...) \
do { \
printf(message, ##__VA_ARGS__); \
} while (0)
string GetTime()
{
time_t timep;
time(&timep);
char tmp[64];
strftime(tmp, sizeof(tmp), "%Y-%m-%d %H:%M:%S,000", localtime(&timep));
return tmp;
}
uint32_t GetDataTypeSize(DataType dt)
{
static const std::map<DataType, uint32_t> kSize = {
{ge::DT_FLOAT, 4}, {ge::DT_FLOAT16, 2}, {ge::DT_BF16, 2}, {ge::DT_INT16, 2}, {ge::DT_UINT16, 2},
{ge::DT_INT32, 4}, {ge::DT_UINT32, 4}, {ge::DT_INT64, 8}, {ge::DT_UINT64, 8}, {ge::DT_INT8, 1},
};
auto it = kSize.find(dt);
return it != kSize.end() ? it->second : 1;
}
static uint16_t FloatToHalfBits(float value)
{
uint32_t bits;
memcpy(&bits, &value, sizeof(bits));
uint32_t sign = (bits >> 16) & 0x8000U;
uint32_t exponent = (bits >> 23) & 0xffU;
uint32_t mantissa = bits & 0x7fffffU;
if (exponent == 0xffU) {
return static_cast<uint16_t>(sign | 0x7c00U | (mantissa != 0 ? 0x0200U : 0U));
}
int32_t half_exponent = static_cast<int32_t>(exponent) - 127 + 15;
if (half_exponent >= 31) {
return static_cast<uint16_t>(sign | 0x7c00U);
}
if (half_exponent < -10) {
return static_cast<uint16_t>(sign);
}
uint32_t shift = 13;
if (half_exponent <= 0) {
mantissa |= 0x800000U;
shift = static_cast<uint32_t>(14 - half_exponent);
half_exponent = 0;
}
uint32_t half_bits = (static_cast<uint32_t>(half_exponent) << 10) + (mantissa >> shift);
uint32_t remainder = mantissa & ((1U << shift) - 1U);
uint32_t halfway = 1U << (shift - 1U);
if (remainder > halfway || (remainder == halfway && (half_bits & 1U) != 0)) {
++half_bits;
}
return static_cast<uint16_t>(sign | half_bits);
}
static float HalfBitsToFloat(uint16_t bits)
{
uint32_t sign = static_cast<uint32_t>(bits & 0x8000U) << 16;
uint32_t exponent = (bits >> 10) & 0x1fU;
uint32_t mantissa = bits & 0x3ffU;
if (exponent == 0) {
float value = static_cast<float>(mantissa) / (1U << 24);
return sign != 0 ? -value : value;
}
uint32_t float_bits = sign | ((exponent == 31 ? 0xffU : exponent + 112U) << 23) | (mantissa << 13);
float value;
memcpy(&value, &float_bits, sizeof(value));
return value;
}
int32_t GenOnesDataFloat32(vector<int64_t> shapes, Tensor& input_tensor, TensorDesc& input_tensor_desc, float value)
{
input_tensor_desc.SetRealDimCnt(shapes.size());
size_t size = 1;
for (uint32_t i = 0; i < shapes.size(); i++) {
size *= shapes[i];
}
DataType dtype = input_tensor_desc.GetDataType();
uint32_t item_size = GetDataTypeSize(dtype);
uint32_t data_len = size * item_size;
std::unique_ptr<uint8_t[]> pData(new (std::nothrow) uint8_t[data_len]);
if (pData == nullptr) {
printf("alloc failed\n");
return FAILED;
}
for (size_t i = 0; i < size; ++i) {
if (dtype == DT_FLOAT) {
memcpy(pData.get() + i * item_size, &value, item_size);
} else {
uint32_t float_bits;
memcpy(&float_bits, &value, sizeof(float_bits));
uint16_t bits = dtype == DT_FLOAT16 ? FloatToHalfBits(value) : static_cast<uint16_t>(float_bits >> 16);
memcpy(pData.get() + i * item_size, &bits, item_size);
}
}
input_tensor = Tensor(input_tensor_desc);
return input_tensor.SetData(pData.get(), data_len) == SUCCESS ? SUCCESS : FAILED;
}
int32_t WriteDataToFile(string bin_file, uint64_t data_size, uint8_t* inputData)
{
FILE* fp = fopen(bin_file.c_str(), "wb");
if (fp == nullptr) {
printf("fopen %s failed\n", bin_file.c_str());
return FAILED;
}
size_t written = fwrite(inputData, sizeof(uint8_t), data_size, fp);
int close_ret = fclose(fp);
if (written != data_size || close_ret != 0) {
printf("write %s failed\n", bin_file.c_str());
return FAILED;
}
return SUCCESS;
}
int CreateOppInGraph(DataType inDtype, std::vector<ge::Tensor>& input, std::vector<Operator>& inputs,
std::vector<Operator>& outputs, Graph& graph)
{
Status ret = SUCCESS;
auto ndtri_op = op::Ndtri("ndtri_op");
std::vector<int64_t> xShape = {4, 2};
ADD_INPUT(1, x, inDtype, xShape);
outputs.push_back(ndtri_op);
return SUCCESS;
}
int main(int argc, char* argv[])
{
DataType inDtype = DT_FLOAT;
if (argc >= 2) {
string dtype_arg = argv[1];
if (dtype_arg == "float32" || dtype_arg == "0") {
inDtype = DT_FLOAT;
} else if (dtype_arg == "float16" || dtype_arg == "1") {
inDtype = DT_FLOAT16;
} else if (dtype_arg == "bfloat16" || dtype_arg == "27") {
inDtype = DT_BF16;
} else {
printf("Unsupported dtype: %s\n", argv[1]);
return FAILED;
}
}
if (argc < 2) {
printf("INFO: %s invoked without dtype arg, defaulting to DT_FLOAT\n", argv[0] ? argv[0] : "test_geir_ndtri");
}
const char* graph_name = "tc_ge_irrun_test";
Graph graph(graph_name);
std::vector<ge::Tensor> input;
printf("%s - INFO - [XIR]: Start to initialize ge using ge global options\n", GetTime().c_str());
std::map<AscendString, AscendString> global_options = {{"ge.exec.deviceId", "0"}, {"ge.graphRunMode", "1"}};
Status ret = ge::GEInitialize(global_options);
if (ret != SUCCESS) {
printf("%s - INFO - [XIR]: Initialize ge using ge global options failed\n", GetTime().c_str());
return FAILED;
}
printf("%s - INFO - [XIR]: Initialize ge using ge global options success\n", GetTime().c_str());
std::vector<Operator> inputs{};
std::vector<Operator> outputs{};
if (argc >= 2) {
std::cout << argv[1] << std::endl;
}
std::cout << inDtype << std::endl;
ret = CreateOppInGraph(inDtype, input, inputs, outputs, graph);
if (ret != SUCCESS) {
printf("%s - ERROR - [XIR]: Create ir session using build options failed\n", GetTime().c_str());
GEFinalize();
return FAILED;
}
if (!inputs.empty() && !outputs.empty()) {
graph.SetInputs(inputs).SetOutputs(outputs);
}
std::map<AscendString, AscendString> build_options = {
};
printf("%s - INFO - [XIR]: Start to create ir session using build options\n", GetTime().c_str());
std::unique_ptr<ge::Session> session(new (std::nothrow) Session(build_options));
if (session == nullptr) {
printf("%s - ERROR - [XIR]: Create ir session using build options failed\n", GetTime().c_str());
GEFinalize();
return FAILED;
}
printf("%s - INFO - [XIR]: Create ir session using build options success\n", GetTime().c_str());
printf("%s - INFO - [XIR]: Start to add compute graph to ir session\n", GetTime().c_str());
std::map<AscendString, AscendString> graph_options = {
};
uint32_t graph_id = 0;
ret = session->AddGraph(graph_id, graph, graph_options);
if (ret != SUCCESS) {
printf("%s - ERROR - [XIR]: Session add graph failed\n", GetTime().c_str());
session.reset();
GEFinalize();
return FAILED;
}
printf("%s - INFO - [XIR]: Session add ir compute graph to ir session success\n", GetTime().c_str());
printf("%s - INFO - [XIR]: dump graph to txt\n", GetTime().c_str());
std::string file_path = "./dump";
aclgrphDumpGraph(graph, file_path.c_str(), file_path.length());
printf("%s - INFO - [XIR]: Start to run ir compute graph\n", GetTime().c_str());
std::vector<ge::Tensor> output;
ret = session->RunGraph(graph_id, input, output);
if (ret != SUCCESS) {
printf("%s - INFO - [XIR]: Run graph failed\n", GetTime().c_str());
session.reset();
GEFinalize();
return FAILED;
}
printf("%s - INFO - [XIR]: Session run ir compute graph success\n", GetTime().c_str());
int input_num = input.size();
for (int i = 0; i < input_num; i++) {
std::cout << "input " << i << " dtype : " << input[i].GetTensorDesc().GetDataType() << std::endl;
string input_file = "./tc_ge_irrun_test_0008_npu_input_" + std::to_string(i) + ".bin";
uint8_t* input_data_i = input[i].GetData();
int64_t input_shape = input[i].GetTensorDesc().GetShape().GetShapeSize();
std::cout << "this is " << i << "th input, input shape size =" << input_shape << std::endl;
uint32_t data_size = input_shape * GetDataTypeSize(input[i].GetTensorDesc().GetDataType());
if (WriteDataToFile(input_file, data_size, input_data_i) != SUCCESS) {
session.reset();
GEFinalize();
return FAILED;
}
}
int output_num = output.size();
for (int i = 0; i < output_num; i++) {
std::cout << "output " << i << " dtype : " << output[i].GetTensorDesc().GetDataType() << std::endl;
string output_file = "./tc_ge_irrun_test_0008_npu_output_" + std::to_string(i) + ".bin";
uint8_t* output_data_i = output[i].GetData();
int64_t output_shape = output[i].GetTensorDesc().GetShape().GetShapeSize();
std::cout << "this is " << i << "th output, output shape size =" << output_shape << std::endl;
uint32_t data_size = output_shape * GetDataTypeSize(output[i].GetTensorDesc().GetDataType());
if (WriteDataToFile(output_file, data_size, output_data_i) != SUCCESS) {
session.reset();
GEFinalize();
return FAILED;
}
for (int64_t j = 0; j < output_shape; j++) {
float result_value;
DataType out_dtype = output[i].GetTensorDesc().GetDataType();
if (out_dtype == DT_FLOAT) {
memcpy(&result_value, output_data_i + j * sizeof(float), sizeof(float));
} else {
uint16_t bits;
memcpy(&bits, output_data_i + j * sizeof(uint16_t), sizeof(bits));
if (out_dtype == DT_FLOAT16) {
result_value = HalfBitsToFloat(bits);
} else {
uint32_t float_bits = static_cast<uint32_t>(bits) << 16;
memcpy(&result_value, &float_bits, sizeof(float));
}
}
LOG_PRINT("result[%ld] is: %f\n", j, result_value);
}
}
ge::AscendString error_msg = ge::GEGetErrorMsgV2();
std::string error_str(error_msg.GetString());
std::cout << "Error message: " << error_str << std::endl;
ge::AscendString warning_msg = ge::GEGetWarningMsgV2();
std::string warning_str(warning_msg.GetString());
std::cout << "Warning message: " << warning_str << std::endl;
printf("%s - INFO - [XIR]: Start to finalize ir graph session\n", GetTime().c_str());
session.reset();
ret = ge::GEFinalize();
if (ret != SUCCESS) {
printf("%s - INFO - [XIR]: Finalize ir graph session failed\n", GetTime().c_str());
return FAILED;
}
printf("%s - INFO - [XIR]: Finalize ir graph session success\n", GetTime().c_str());
return SUCCESS;
}
