@@ -1532,7 +1532,8 @@ Status ModelManager::EnableExceptionDump(const std::map<string, string> &options
return SUCCESS;
}
Status ModelManager::LaunchKernelCheckAicpuOp(std::vector<std::string> &aicpu_optype_list, std::vector<std::string> &aicpu_tf_optype_list) {
Status ModelManager::LaunchKernelCheckAicpuOp(std::vector<std::string> &aicpu_optype_list,
std::vector<std::string> &aicpu_tf_optype_list) {
std::string kernel_name = "checkOpType";
GELOGI("LaunchKernelCheckAicpuOpType in, kernel name %s", kernel_name.c_str());
std::lock_guard<std::mutex> lock(cust_aicpu_mutex_);
@@ -1556,7 +1557,7 @@ Status ModelManager::LaunchKernelCheckAicpuOp(std::vector<std::string> &aicpu_op
size_t aicpu_op_nums = aicpu_optype_list.size();
size_t tf_op_nums = aicpu_tf_optype_list.size();
size_t op_nums = aicpu_op_nums + tf_op_nums;
size_t op_nums = aicpu_op_nums + tf_op_nums;
// malloc sysOpInfoList in SysOpCheckInfo
status = rtMalloc(&d_req_op_list, op_nums * sizeof(SysOpInfo), RT_MEMORY_HBM);
if (status != RT_ERROR_NONE) {
@@ -1591,12 +1592,11 @@ Status ModelManager::LaunchKernelCheckAicpuOp(std::vector<std::string> &aicpu_op
return RT_ERROR_TO_GE_STATUS(status);
}
allocated_mem.push_back(d_op_type_name);
GE_CHK_RT(rtMemcpy(d_op_type_name, op_type.length(), op_type.c_str(), op_type.length(), RT_MEMCPY_HOST_TO_DEVICE));
GE_CHK_RT(rtMemcpy(d_op_type_name, op_type.length(), op_type.c_str(), op_type.length(), RT_MEMCPY_HOST_TO_DEVICE));
op_info.opType = reinterpret_cast<uint64_t>(reinterpret_cast<uintptr_t>(d_op_type_name));
op_info.opLen = op_type.length();
op_info.kernelsType = CPU_KERNEL;
req_aicpu_op_info_list.emplace_back(op_info);
//GE_CHK_RT(rtMemcpy((char *)d_req_op_list + i * sizeof(SysOpInfo), sizeof(SysOpInfo), &op_info, sizeof(SysOpInfo), RT_MEMCPY_HOST_TO_DEVICE));
}
for (const auto &op_type : aicpu_tf_optype_list) {
@@ -1609,15 +1609,15 @@ Status ModelManager::LaunchKernelCheckAicpuOp(std::vector<std::string> &aicpu_op
return RT_ERROR_TO_GE_STATUS(status);
}
allocated_mem.push_back(d_op_type_name);
GE_CHK_RT(rtMemcpy(d_op_type_name, op_type.size(), op_type.c_str(), op_type.size(), RT_MEMCPY_HOST_TO_DEVICE));
GE_CHK_RT(rtMemcpy(d_op_type_name, op_type.size(), op_type.c_str(), op_type.size(), RT_MEMCPY_HOST_TO_DEVICE));
op_info.opType = reinterpret_cast<uint64_t>(reinterpret_cast<uintptr_t>(d_op_type_name));
op_info.opLen = op_type.size();
op_info.kernelsType = TF_KERNEL;
req_aicpu_op_info_list.emplace_back(op_info);
//GE_CHK_RT(rtMemcpy((char *)d_req_op_list + i * sizeof(SysOpInfo), sizeof(SysOpInfo), &op_info, sizeof(SysOpInfo), RT_MEMCPY_HOST_TO_DEVICE));
}
GELOGI("Check aicpu op all attr size: %zu, real attr size: %zu.", op_nums, req_aicpu_op_info_list.size());
GE_CHK_RT(rtMemcpy(d_req_op_list, sizeof(SysOpInfo) * req_aicpu_op_info_list.size(), req_aicpu_op_info_list.data(), sizeof(SysOpInfo) * req_aicpu_op_info_list.size(), RT_MEMCPY_HOST_TO_DEVICE));
GE_CHK_RT(rtMemcpy(d_req_op_list, sizeof(SysOpInfo) * req_aicpu_op_info_list.size(), req_aicpu_op_info_list.data(),
sizeof(SysOpInfo) * req_aicpu_op_info_list.size(), RT_MEMCPY_HOST_TO_DEVICE));
SysOpCheckInfo op_check_info_req;
SysOpCheckResp op_check_info_res;
@@ -1636,8 +1636,10 @@ Status ModelManager::LaunchKernelCheckAicpuOp(std::vector<std::string> &aicpu_op
return RT_ERROR_TO_GE_STATUS(status);
}
allocated_mem.push_back(args);
GE_CHK_RT(rtMemcpy(args, sizeof(SysOpCheckInfo), &op_check_info_req, sizeof(SysOpCheckInfo), RT_MEMCPY_HOST_TO_DEVICE));
GE_CHK_RT(rtMemcpy(reinterpret_cast<void *>(reinterpret_cast<uint64_t>(args) + op_check_info_req.offSetLen), sizeof(SysOpCheckResp), &op_check_info_res, sizeof(SysOpCheckResp), RT_MEMCPY_HOST_TO_DEVICE));
GE_CHK_RT(
rtMemcpy(args, sizeof(SysOpCheckInfo), &op_check_info_req, sizeof(SysOpCheckInfo), RT_MEMCPY_HOST_TO_DEVICE));
GE_CHK_RT(rtMemcpy(reinterpret_cast<void *>(reinterpret_cast<uint64_t>(args) + op_check_info_req.offSetLen),
sizeof(SysOpCheckResp), &op_check_info_res, sizeof(SysOpCheckResp), RT_MEMCPY_HOST_TO_DEVICE));
GE_CHK_RT(rtStreamCreate(&stream, 0));
GE_CHK_RT(rtCpuKernelLaunch(nullptr, kernel_name.c_str(), 1, args, args_size, nullptr, stream));
@@ -1650,7 +1652,8 @@ Status ModelManager::LaunchKernelCheckAicpuOp(std::vector<std::string> &aicpu_op
// Check the response
void *d_op_check_info_res = reinterpret_cast<void *>(reinterpret_cast<uint64_t>(args) + op_check_info_req.offSetLen);
GE_CHK_RT(rtMemcpy(&op_check_info_res, sizeof(SysOpCheckResp), d_op_check_info_res, sizeof(SysOpCheckResp), RT_MEMCPY_DEVICE_TO_HOST));
GE_CHK_RT(rtMemcpy(&op_check_info_res, sizeof(SysOpCheckResp), d_op_check_info_res, sizeof(SysOpCheckResp),
RT_MEMCPY_DEVICE_TO_HOST));
std::function<void()> callback = [&]() {
for (auto mem : allocated_mem) {
GE_CHK_RT(rtFree(mem));
@@ -1670,8 +1673,12 @@ Status ModelManager::LaunchKernelCheckAicpuOp(std::vector<std::string> &aicpu_op
res_ret_code_list.resize(res_op_nums);
res_aicpu_op_info_list.clear();
res_aicpu_op_info_list.resize(res_op_nums);
GE_CHK_RT(rtMemcpy(res_ret_code_list.data(), sizeof(ReturnCode) * res_op_nums, reinterpret_cast<void *>(reinterpret_cast<uintptr_t>(op_check_info_res.returnCodeList)), sizeof(ReturnCode) * res_op_nums, RT_MEMCPY_DEVICE_TO_HOST));
GE_CHK_RT(rtMemcpy(res_aicpu_op_info_list.data(), sizeof(SysOpInfo) * res_op_nums, reinterpret_cast<void *>(reinterpret_cast<uintptr_t>(op_check_info_res.sysOpInfoList)), sizeof(SysOpInfo) * res_op_nums, RT_MEMCPY_DEVICE_TO_HOST));
GE_CHK_RT(rtMemcpy(res_ret_code_list.data(), sizeof(ReturnCode) * res_op_nums,
reinterpret_cast<void *>(reinterpret_cast<uintptr_t>(op_check_info_res.returnCodeList)),
sizeof(ReturnCode) * res_op_nums, RT_MEMCPY_DEVICE_TO_HOST));
GE_CHK_RT(rtMemcpy(res_aicpu_op_info_list.data(), sizeof(SysOpInfo) * res_op_nums,
reinterpret_cast<void *>(reinterpret_cast<uintptr_t>(op_check_info_res.sysOpInfoList)),
sizeof(SysOpInfo) * res_op_nums, RT_MEMCPY_DEVICE_TO_HOST));
if (res_ret_code_list.size() != res_aicpu_op_info_list.size() || res_ret_code_list.size() != res_op_nums) {
GELOGE(FAILED, "Number of retcode is not equal to number of op type.");
GE_MAKE_GUARD(release, callback);
@@ -1681,14 +1688,19 @@ Status ModelManager::LaunchKernelCheckAicpuOp(std::vector<std::string> &aicpu_op
for (uint32_t i = 0; i < res_op_nums; i++) {
ReturnCode ret_code = res_ret_code_list.at(i);
SysOpInfo aicpu_info = res_aicpu_op_info_list.at(i);
GELOGI("Not support aicpu op type: %lu, kernel_type:%d, opLen:%d, ret_code:%d", aicpu_info.opType, aicpu_info.kernelsType, aicpu_info.opLen, ret_code);
GELOGI("Not support aicpu op type: %lu, kernel_type:%d, opLen:%d, ret_code:%d", aicpu_info.opType,
aicpu_info.kernelsType, aicpu_info.opLen, ret_code);
std::vector<char> op_name;
op_name.clear();
op_name.resize(kOpNameMaxSize);
GE_CHK_RT(rtMemcpy(op_name.data(), aicpu_info.opLen, reinterpret_cast<void *>(aicpu_info.opType), aicpu_info.opLen, RT_MEMCPY_DEVICE_TO_HOST));
std::string kernel_type = (static_cast<OpKernelType>(aicpu_info.kernelsType) == TF_KERNEL) ? "TF_KERNEL" : "CPU_KERNEL";
GE_CHK_RT(rtMemcpy(op_name.data(), aicpu_info.opLen, reinterpret_cast<void *>(aicpu_info.opType),
aicpu_info.opLen, RT_MEMCPY_DEVICE_TO_HOST));
std::string kernel_type =
(static_cast<OpKernelType>(aicpu_info.kernelsType) == TF_KERNEL) ? "TF_KERNEL" : "CPU_KERNEL";
string op_name_str(op_name.data());
fail_reason += "op_type: " + op_name_str + " kernel_type: " + kernel_type + " ret code:" + std::to_string(static_cast<int>(ret_code)) + "<0: op_type, 1: format, 2: datatype> \n";
fail_reason += "op_type: " + op_name_str + " kernel_type: " + kernel_type +
" ret code:" + std::to_string(static_cast<int>(ret_code)) +
"<0: op_type, 1: format, 2: datatype> \n";
}
fail_reason += "not support.";
GELOGE(FAILED, "Check aicpu op_type failed. details: %s", fail_reason.c_str());
@@ -1706,11 +1718,12 @@ Status ModelManager::CheckAicpuOpList(GeModelPtr ge_model) {
std::vector<std::string> aicpu_tf_optype_list;
bool aicpu_need_check = ge::AttrUtils::GetListStr(ge_model, "needCheckCpu", aicpu_optype_list);
bool tf_need_check = ge::AttrUtils::GetListStr(ge_model, "needCheckTf", aicpu_tf_optype_list);
if (!aicpu_need_check && !tf_need_check) {
if (!aicpu_need_check && !tf_need_check) {
GELOGI("Graph:%s No need to check aicpu optype.", ge_model->GetGraph().GetName().c_str());
return SUCCESS;
}
GE_CHK_STATUS_RET(LaunchKernelCheckAicpuOp(aicpu_optype_list, aicpu_tf_optype_list), "Launch check aicpu op type failed.");
GE_CHK_STATUS_RET(LaunchKernelCheckAicpuOp(aicpu_optype_list, aicpu_tf_optype_list),
"Launch check aicpu op type failed.");
return SUCCESS;
}