#include <iostream>
#include <vector>
#include <getopt.h>
#include <unistd.h>
#include <dirent.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <sys/wait.h>
#include <dlfcn.h>
#include <cstdio>
#include <string>
#include <iomanip>
#include <string.h>
#include <list>
#include <algorithm>
#include <stdarg.h>
#include "acl/acl.h"
#include "aclnn/aclnn_base.h"
#include "acl/acl_base.h"
#include "acl/acl_rt.h"
#include "sec_common.h"
#include "utils.h"
#include <complex>

#include "../fuzzlib/secodeFuzz.h"
#include "aclnnop/aclnn_cat.h"

using namespace std;

static std::string ShapeToStr(const std::vector<int64_t>& s) {
    std::string r = "[";
    for (size_t k = 0; k < s.size(); k++) {
        if (k) r += ",";
        r += std::to_string(s[k]);
    }
    return r + "]";
}

// ===== [FUZZ-FILL] 输入数据按 dtype 随机泛化 =====
static uint16_t FuzzFloatToHalf(float f) {
    uint32_t x; memcpy(&x, &f, 4);
    uint16_t sign = (uint16_t)((x >> 16) & 0x8000);
    int32_t exp = (int32_t)((x >> 23) & 0xFF) - 127 + 15;
    uint32_t mant = (x >> 13) & 0x3FF;
    if (exp <= 0) { return sign; }
    if (exp >= 31) { return (uint16_t)(sign | 0x7C00); }
    return (uint16_t)(sign | (uint32_t)(exp << 10) | mant);
}
static uint16_t FuzzFloatToBf16(float f) {
    uint32_t x; memcpy(&x, &f, 4);
    uint32_t lsb = (x >> 16) & 1;
    x += 0x7FFF + lsb;
    return (uint16_t)(x >> 16);
}
static void FuzzInitHostData(void* data, size_t count, aclDataType dtype) {
    if (data == nullptr || count == 0) { return; }
    for (size_t i = 0; i < count; i++) {
        float rf = ((float)rand() / (float)RAND_MAX) * 2000.0f - 1000.0f;
        double rd = ((double)rand() / (double)RAND_MAX) * 2000.0 - 1000.0;
        switch ((int)dtype) {
        case ACL_FLOAT:   ((float*)data)[i] = rf; break;
        case ACL_FLOAT16: ((int16_t*)data)[i] = (int16_t)FuzzFloatToHalf(rf); break;
        case ACL_BF16:    ((int16_t*)data)[i] = (int16_t)FuzzFloatToBf16(rf); break;
        case ACL_DOUBLE:  ((double*)data)[i] = rd; break;
        case ACL_INT8:    ((int8_t*)data)[i] = (int8_t)(rand() % 256 - 128); break;
        case ACL_INT16:   ((int16_t*)data)[i] = (int16_t)(rand() - 16384); break;
        case ACL_INT32:   ((int32_t*)data)[i] = (int32_t)(((uint32_t)rand() << 16) ^ (uint32_t)rand()); break;
        case ACL_INT64:   ((int64_t*)data)[i] = (int64_t)(((uint64_t)rand() << 48) ^ ((uint64_t)rand() << 32) ^ ((uint64_t)rand() << 16) ^ (uint64_t)rand()); break;
        case ACL_UINT8:   ((uint8_t*)data)[i] = (uint8_t)rand(); break;
        case ACL_UINT16:  ((uint16_t*)data)[i] = (uint16_t)rand(); break;
        case ACL_UINT32:  ((uint32_t*)data)[i] = (((uint32_t)rand() << 16) ^ (uint32_t)rand()); break;
        case ACL_UINT64:  ((uint64_t*)data)[i] = (((uint64_t)rand() << 48) ^ ((uint64_t)rand() << 32) ^ ((uint64_t)rand() << 16) ^ (uint64_t)rand()); break;
        case ACL_BOOL:    ((int8_t*)data)[i] = (int8_t)(rand() & 1); break;
        default:          ((uint8_t*)data)[i] = (uint8_t)rand(); break;
        }
    }
}
// ===== [FUZZ-FILL END] =====

int aclnnCat_Normal(int fuzz_num)
{
    auto ret = Init(deviceId, &context, &stream);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);
    WriteDTLog("aclnnCat", "aclnnCat op fuzz start \n");
    const long long kSingleInputSizeMax = 1073741824LL;
    const long long kTotalInputSizeMax = 2147483648LL;

    // dtype enums
    int dtype0Enum[] = { 0, 1, 3, 9, 6, 2, 4, 12, 27, 11, 16 };
    int dtype0Num = sizeof(dtype0Enum) / sizeof(dtype0Enum[0]);

    aclrtSetOpExecuteTimeOut(1091);
    DT_Set_TimeOut_Second(3000);
    DT_FUZZ_START(0, fuzz_num, "aclnnCat", 0)
    {
        int i = 0;
        int shapeDimMax = 16;
        int aclnnRet = -1;  // aclnn 实际调用返回值（成功判定依据）
        int tensors_num = *(int *)DTSetGetNumberRange(&g_Element[i++], 2, 2, 5, "i");
        aclTensorList* tensorsList = nullptr;
        std::vector<aclTensor*> tensorsVec(tensors_num, nullptr);
        std::vector<void*> tensorsAddrVec(tensors_num, nullptr);
        void* outDeviceAddr = nullptr;
        aclTensor* out = nullptr;
        uint64_t workspaceSize = 0;
        aclOpExecutor* executor;
        void* workspaceAddr = nullptr;

        LOG_PRINT("aclnnCat_Normal %d\n", fuzzi);
        printf("[FUZZ_LOG] plog_file: aclnnCat_fuzz_plog_case_%d.log\n", fuzzi);
        aclnnRet = -1;
        try
        {
        // DT mutation data
            std::srand((unsigned int)((fuzzi + 1) * 2654435761u));
            int sizeBucket = *(int *)DTSetGetNumberRange(&g_Element[i++], 1, 0, 3, "i");
            int dim_num0 = *(int *)DTSetGetNumberRange(&g_Element[i++], 2, 0, 8, "i");  // 维度约束 [0,8]
            {
                const long long sizeTargets[4] = {1000LL, 1048576LL, 16777216LL, 134217728LL};
                int nTensors = 2;
                long long target = sizeTargets[sizeBucket] / nTensors;
                if (target < 1) { target = 1; }
                int nd = (dim_num0 > 0) ? dim_num0 : 1;
                long long perDim = (long long)pow((double)target, 1.0 / nd);
                if (perDim < 1) { perDim = 1; }
                if (perDim > MAX_SHAPE_SIZE) { perDim = MAX_SHAPE_SIZE; }
                shapeDimMax = (int)perDim;
            }
            int shape1 = *(int *)DTSetGetNumberRange(&g_Element[i++], 2, 1, shapeDimMax, "i");
            int shape2 = *(int *)DTSetGetNumberRange(&g_Element[i++], 2, 1, shapeDimMax, "i");
            int shape3 = *(int *)DTSetGetNumberRange(&g_Element[i++], 2, 1, shapeDimMax, "i");
            int shape4 = *(int *)DTSetGetNumberRange(&g_Element[i++], 2, 1, shapeDimMax, "i");
            int shape5 = *(int *)DTSetGetNumberRange(&g_Element[i++], 2, 1, shapeDimMax, "i");
            int shape6 = *(int *)DTSetGetNumberRange(&g_Element[i++], 2, 1, shapeDimMax, "i");
            int shape7 = *(int *)DTSetGetNumberRange(&g_Element[i++], 2, 1, shapeDimMax, "i");
            int shape8 = *(int *)DTSetGetNumberRange(&g_Element[i++], 2, 1, shapeDimMax, "i");
            auto dtype0 = (aclDataType)*(int *)DT_SetGetNumberEnumV3(i++, aclDataType::ACL_FLOAT, dtype0Enum, dtype0Num);
            int64_t dim = 0;
            int dimchoice = *(int *)DTSetGetNumberRange(&g_Element[i++], 0, 0, 1, "i");
            if (dim_num0 <= 0) { dim = 0; }
            else if (dimchoice == 0) { dim = (rand() % (dim_num0 + 1)) * -1; }
            else { dim = rand() % dim_num0; }
            int dim_num1 = *(int *)DTSetGetNumberRange(&g_Element[i++], 2, 0, 8, "i");
            std::vector<int64_t> outShape;
            std::vector<std::vector<int64_t>> tensorsShapeList;
            for (int t = 0; t < tensors_num; t++) {
                int dim_num_t = *(int *)DTSetGetNumberRange(&g_Element[i++], 2, 1, 8, "i");
                std::vector<int64_t> shape_t;
                if (dim_num_t == 1) { shape_t = {*(int *)DTSetGetNumberRange(&g_Element[i++], 4, 1, MAX_SHAPE_SIZE, "i")}; }
                else if (dim_num_t == 2) { shape_t = {*(int *)DTSetGetNumberRange(&g_Element[i++], 4, 1, MAX_SHAPE_SIZE, "i"), *(int *)DTSetGetNumberRange(&g_Element[i++], 2, 1, MAX_SHAPE_SIZE, "i")}; }
                tensorsShapeList.push_back(shape_t);
            }
            outShape = tensorsShapeList[0];
            long long shape_size = GetShapeSize(tensorsShapeList[0]);
            if (shape_size > kSingleInputSizeMax) { printf("shape_size too big\n"); goto ERROR; }
            if (dtype0 == aclDataType::ACL_FLOAT) {
                std::vector<float> outHostData(GetShapeSize(outShape), 0);
                FuzzInitHostData(outHostData.data(), outHostData.size(), dtype0);
                ret = CreateAclTensor_New(outHostData, outShape, &outDeviceAddr, dtype0, aclFormat::ACL_FORMAT_ND, &out);
            CHECK_RET(ret == ACL_SUCCESS, goto ERROR);
            } else if (dtype0 == aclDataType::ACL_FLOAT16) {
                std::vector<int16_t> outHostData(GetShapeSize(outShape), 0);
                FuzzInitHostData(outHostData.data(), outHostData.size(), dtype0);
                ret = CreateAclTensor_New(outHostData, outShape, &outDeviceAddr, dtype0, aclFormat::ACL_FORMAT_ND, &out);
            CHECK_RET(ret == ACL_SUCCESS, goto ERROR);
            } else if (dtype0 == aclDataType::ACL_INT32) {
                std::vector<int32_t> outHostData(GetShapeSize(outShape), 0);
                FuzzInitHostData(outHostData.data(), outHostData.size(), dtype0);
                ret = CreateAclTensor_New(outHostData, outShape, &outDeviceAddr, dtype0, aclFormat::ACL_FORMAT_ND, &out);
            CHECK_RET(ret == ACL_SUCCESS, goto ERROR);
            } else if (dtype0 == aclDataType::ACL_INT64) {
                std::vector<int64_t> outHostData(GetShapeSize(outShape), 0);
                FuzzInitHostData(outHostData.data(), outHostData.size(), dtype0);
                ret = CreateAclTensor_New(outHostData, outShape, &outDeviceAddr, dtype0, aclFormat::ACL_FORMAT_ND, &out);
            CHECK_RET(ret == ACL_SUCCESS, goto ERROR);
            } else if (dtype0 == aclDataType::ACL_INT16) {
                std::vector<int16_t> outHostData(GetShapeSize(outShape), 0);
                FuzzInitHostData(outHostData.data(), outHostData.size(), dtype0);
                ret = CreateAclTensor_New(outHostData, outShape, &outDeviceAddr, dtype0, aclFormat::ACL_FORMAT_ND, &out);
            CHECK_RET(ret == ACL_SUCCESS, goto ERROR);
            } else if (dtype0 == aclDataType::ACL_INT8) {
                std::vector<int8_t> outHostData(GetShapeSize(outShape), 0);
                FuzzInitHostData(outHostData.data(), outHostData.size(), dtype0);
                ret = CreateAclTensor_New(outHostData, outShape, &outDeviceAddr, dtype0, aclFormat::ACL_FORMAT_ND, &out);
            CHECK_RET(ret == ACL_SUCCESS, goto ERROR);
            } else if (dtype0 == aclDataType::ACL_UINT8) {
                std::vector<uint8_t> outHostData(GetShapeSize(outShape), 0);
                FuzzInitHostData(outHostData.data(), outHostData.size(), dtype0);
                ret = CreateAclTensor_New(outHostData, outShape, &outDeviceAddr, dtype0, aclFormat::ACL_FORMAT_ND, &out);
            CHECK_RET(ret == ACL_SUCCESS, goto ERROR);
            } else if (dtype0 == aclDataType::ACL_BOOL) {
                std::vector<int8_t> outHostData(GetShapeSize(outShape), 0);
                FuzzInitHostData(outHostData.data(), outHostData.size(), dtype0);
                ret = CreateAclTensor_New(outHostData, outShape, &outDeviceAddr, dtype0, aclFormat::ACL_FORMAT_ND, &out);
            CHECK_RET(ret == ACL_SUCCESS, goto ERROR);
            } else if (dtype0 == aclDataType::ACL_BF16) {
                std::vector<int16_t> outHostData(GetShapeSize(outShape), 0);
                FuzzInitHostData(outHostData.data(), outHostData.size(), dtype0);
                ret = CreateAclTensor_New(outHostData, outShape, &outDeviceAddr, dtype0, aclFormat::ACL_FORMAT_ND, &out);
            CHECK_RET(ret == ACL_SUCCESS, goto ERROR);
            } else if (dtype0 == aclDataType::ACL_DOUBLE) {
                std::vector<double> outHostData(GetShapeSize(outShape), 0);
                FuzzInitHostData(outHostData.data(), outHostData.size(), dtype0);
                ret = CreateAclTensor_New(outHostData, outShape, &outDeviceAddr, dtype0, aclFormat::ACL_FORMAT_ND, &out);
            CHECK_RET(ret == ACL_SUCCESS, goto ERROR);
            } else if (dtype0 == aclDataType::ACL_COMPLEX64) {
                std::vector<std::complex<float>> outHostData(GetShapeSize(outShape), 0);
                FuzzInitHostData(outHostData.data(), outHostData.size(), dtype0);
                ret = CreateAclTensor_New(outHostData, outShape, &outDeviceAddr, dtype0, aclFormat::ACL_FORMAT_ND, &out);
            CHECK_RET(ret == ACL_SUCCESS, goto ERROR);
            }

            ret = aclnnCatGetWorkspaceSize(tensorsList, dim, out, &workspaceSize, &executor);
printf("[FUZZ_PARAM] tensors: dtype=%d, shape=%s; out: dtype=%d, shape=%s; dim=%lld\n", (int)dtype0, ShapeToStr(tensorsShapeList[0]).c_str(), (int)dtype0, ShapeToStr(outShape).c_str(), (long long)dim);
            CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnCatGetWorkspaceSize failed. ERROR: %d\n", ret); goto ERROR);
            if (workspaceSize > 0) {
                ret = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
                CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); goto ERROR);
            }
            ret = aclnnCat(workspaceAddr, workspaceSize, executor, stream);
            aclnnRet = (int)ret;
            if (ret == ACL_SUCCESS) { printf("aclnnCat execute success :%d\n", (int)ret); }
            else { printf("aclnnCat execute failed :%d\n", (int)ret); goto ERROR; }
            ret = aclrtSynchronizeStream(stream);
            CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); goto ERROR);
            LOG_PRINT("aclnnCat execute success : %d\n", ret);
        }
        catch (std::exception& e)
        {
            printf("catch exception %s\n", e.what());
        }

ERROR:
        ret = aclrtSynchronizeStream(stream);
        if (aclnnRet == -1) { printf("[FUZZ_RESULT] case_%d: skip, ret=-1\n", fuzzi); }
        else { printf("[FUZZ_RESULT] case_%d: %s, ret=%d\n", fuzzi, (aclnnRet == 0) ? "success" : "faild", aclnnRet); }
        printf("end \n");
        CheckAndLog(ret);
        pid_t pid = getpid();
        std::string cmd = "grep -rin 'aicore error' /root/ascend/log/debug/plog/ ";
        printf("system return value %d\n", CheckResult(pid));
        if (tensorsList) { aclDestroyTensorList(tensorsList); }
        for (size_t ti = 0; ti < tensorsAddrVec.size(); ti++) { if (tensorsAddrVec[ti]) { aclrtFree(tensorsAddrVec[ti]); } }
        if (out) { aclDestroyTensor(out); }
        if (outDeviceAddr) { aclrtFree(outDeviceAddr); }
        if (workspaceSize > 0 && workspaceAddr) { aclrtFree(workspaceAddr); }
    }
    DT_FUZZ_END()
    WriteDTLog("aclnnCat", "aclnnCat op fuzz total number:%u, para error number %u, execute error number %u \n", fuzz_num, aclnn_INVALID_Para, aclnn_Inner_Error + Runtime_Error);
    WriteDTLog("aclnnCat", "aclnnCat op fuzz end \n");
    aclrtDestroyStream(stream);
    aclrtDestroyContext(context);
    aclrtResetDevice(deviceId);
    aclFinalize();
    return 0;
}

extern "C" int aclnnCat_fuzz(int argc, char *argv[]);
int aclnnCat_fuzz(int argc, char *argv[])
{
    int fuzz_num = MAXFUZZNUMBER;
    if (argc < 1) {
        cout << "[ERROR] input arg num must be bigger than 1 \n" << endl;
        return -1;
    } else if (argc > 1) {
        fuzz_num = atoi(argv[2]);
    }
    if (argc == 4) { deviceId = atoi(argv[3]); }
    printf("fuzz_num %d \n", fuzz_num);
    printf("aclnnCat_Normal :%d \n", aclnnCat_Normal(fuzz_num));
    return 0;
}