* This file is part of the OpenBOAT project at Harbin Institute of Technology (HIT)
* and is contributed to the CANN Open Software.
*
* Copyright (c) 2025 AISS Group, Harbin Institute of Technology (HIT).
* All Rights Reserved.
*
* Authors (accounts):
* - Tu Yuanhang <@TuYHAAAAAA>
* - Su Tonghua <@sutonghua>
*
* This program is free software: you can redistribute it and/or modify it.
* Licensed under the CANN Open Software License Agreement Version 2.0 (the "License").
* You may not use this file except in compliance with the License.
* See the LICENSE file at the root of the repository for the full text of the License.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
*/
* \file complex_v2_tiling.cpp
* \brief
*/
#include "log/log.h"
#include "util/math_util.h"
#include "util/platform_util.h"
#include "register/op_impl_registry.h"
#include <graph/utils/type_utils.h>
#include "tiling/platform/platform_ascendc.h"
#include "../op_kernel/complex_v2_tiling_data.h"
#include "../op_kernel/complex_v2_tiling_key.h"
namespace optiling {
const uint32_t BUFFER_NUM = 2;
namespace {
struct ComplexV2CompileInfo {};
struct PlatformInfo {
uint64_t ubSize{0};
int64_t coreNum{0};
uint64_t blockSize{0};
};
}
struct ComplexV2CompileInfo {};
static ge::graphStatus GetPlatformInfo(gert::TilingContext* context, PlatformInfo& platformInfo)
{
auto& ubSize = platformInfo.ubSize;
auto& coreNum = platformInfo.coreNum;
auto& blockSize = platformInfo.blockSize;
fe::PlatFormInfos* platformInfoPtr = context->GetPlatformInfo();
OP_CHECK_NULL_WITH_CONTEXT(context, platformInfoPtr);
auto ascendcPlatform = platform_ascendc::PlatformAscendC(platformInfoPtr);
coreNum = ascendcPlatform.GetCoreNumAiv();
OP_CHECK_IF(coreNum <= 0, OP_LOGE(context, "coreNum is 0"), return ge::GRAPH_FAILED);
ascendcPlatform.GetCoreMemSize(platform_ascendc::CoreMemType::UB, ubSize);
OP_CHECK_IF(ubSize <= 0, OP_LOGE(context, "ubSize is 0"), return ge::GRAPH_FAILED);
blockSize = Ops::Base::GetUbBlockSize(context);
OP_CHECK_IF(blockSize == 0, OP_LOGE(context, "blockSize is 0"), return ge::GRAPH_FAILED);
return ge::GRAPH_SUCCESS;
}
static ge::graphStatus GetShapeAttrsInfo(gert::TilingContext* context, int64_t& totalIdx, ge::DataType& dataType)
{
auto inputX = context->GetInputShape(0);
OP_CHECK_NULL_WITH_CONTEXT(context, inputX);
totalIdx = inputX->GetStorageShape().GetShapeSize();
const std::set<ge::DataType> supportedDtype = {ge::DT_FLOAT, ge::DT_FLOAT16};
auto inputDesc = context->GetInputDesc(0);
OP_CHECK_NULL_WITH_CONTEXT(context, inputDesc);
dataType = inputDesc->GetDataType();
if (supportedDtype.count(dataType) == 0) {
OP_LOGE(context, "invalid dtype");
return ge::GRAPH_FAILED;
}
return ge::GRAPH_SUCCESS;
}
ge::graphStatus GetWorkspaceSize(gert::TilingContext* context)
{
auto ascendcPlatform = platform_ascendc:: PlatformAscendC(context->GetPlatformInfo());
uint32_t sysWorkspaceSize = ascendcPlatform.GetLibApiWorkSpaceSize();
size_t* currentWorkspace = context->GetWorkspaceSizes(1);
OP_CHECK_NULL_WITH_CONTEXT(context, currentWorkspace);
currentWorkspace[0] = sysWorkspaceSize;
return ge::GRAPH_SUCCESS;
}
static ge::graphStatus CalculateCoreNum(gert::TilingContext* context, ComplexV2TilingData* tiling,
const PlatformInfo& platformInfo, int64_t totalIdx)
{
auto coreNum = platformInfo.coreNum;
auto blockSize = platformInfo.blockSize;
auto ubSize = platformInfo.ubSize;
uint32_t typeLength = 0;
ge::TypeUtils::GetDataTypeLength(context->GetInputDesc(0)->GetDataType(), typeLength);
if (typeLength == 0) {
OP_LOGE(context, "typeLength is 0");
return ge::GRAPH_FAILED;
}
uint64_t inputBytes = static_cast<uint64_t>(typeLength);
uint64_t inputLengthBytes = static_cast<uint64_t>(totalIdx) * inputBytes;
uint32_t ubDataNumber = (inputBytes == 1ULL) ? 5U : 3U;
uint64_t tmp = (ubSize / blockSize / BUFFER_NUM);
uint32_t tileBlockNum = 1U;
if (tmp > 0) {
uint64_t tb = tmp / ubDataNumber;
tileBlockNum = (tb == 0) ? 1U : static_cast<uint32_t>(tb);
}
uint32_t tileDataNum = static_cast<uint32_t>((static_cast<uint64_t>(tileBlockNum) * blockSize) / inputBytes);
if (tileDataNum == 0U) {
tileDataNum = 1U;
}
uint64_t blocksTotal = Ops::Base::CeilDiv(inputLengthBytes,blockSize);
uint64_t coreNum64 = static_cast<uint64_t>(coreNum);
if (coreNum64 > blocksTotal) {
coreNum64 = blocksTotal;
}
if (coreNum64 == 0ULL) {
coreNum64 = 1ULL;
}
uint32_t finalCoreNum = static_cast<uint32_t>(coreNum64);
uint64_t everyCoreInputBlockNum = blocksTotal / coreNum64;
uint32_t tailBlockNum = static_cast<uint32_t>(blocksTotal % coreNum64);
uint64_t smallCoreDataNum_u = everyCoreInputBlockNum * blockSize / inputBytes;
uint32_t smallCoreDataNum = static_cast<uint32_t>(smallCoreDataNum_u);
uint32_t smallTileNum = static_cast<uint32_t>(everyCoreInputBlockNum / static_cast<uint64_t>(tileBlockNum));
uint32_t finalSmallTileNum = ((everyCoreInputBlockNum % tileBlockNum) == 0) ? smallTileNum : (smallTileNum + 1);
int64_t smallTailDataNum_i = static_cast<int64_t>(smallCoreDataNum) - static_cast<int64_t>(tileDataNum) * static_cast<int64_t>(smallTileNum);
uint32_t smallTailDataNum = (smallTailDataNum_i <= 0) ? tileDataNum : static_cast<uint32_t>(smallTailDataNum_i);
uint64_t bigEveryCoreBlockNum = everyCoreInputBlockNum + 1ULL;
uint64_t bigCoreDataNum_u = bigEveryCoreBlockNum * blockSize / inputBytes;
uint32_t bigCoreDataNum = static_cast<uint32_t>(bigCoreDataNum_u);
uint32_t bigTileNum = static_cast<uint32_t>(bigEveryCoreBlockNum / static_cast<uint64_t>(tileBlockNum));
uint32_t finalBigTileNum = ((bigEveryCoreBlockNum % tileBlockNum) == 0) ? bigTileNum : (bigTileNum + 1);
int64_t bigTailDataNum_i = static_cast<int64_t>(bigCoreDataNum) - static_cast<int64_t>(tileDataNum) * static_cast<int64_t>(bigTileNum);
uint32_t bigTailDataNum = (bigTailDataNum_i <= 0) ? tileDataNum : static_cast<uint32_t>(bigTailDataNum_i);
tiling->smallCoreDataNum = static_cast<int64_t>(smallCoreDataNum);
tiling->bigCoreDataNum = static_cast<int64_t>(bigCoreDataNum);
tiling->tileDataNum = static_cast<int64_t>(tileDataNum);
tiling->smallTailDataNum = static_cast<int64_t>(smallTailDataNum);
tiling->bigTailDataNum = static_cast<int64_t>(bigTailDataNum);
tiling->finalSmallTileNum = static_cast<int64_t>(finalSmallTileNum);
tiling->finalBigTileNum = static_cast<int64_t>(finalBigTileNum);
tiling->tailBlockNum = static_cast<int64_t>(tailBlockNum);
context->SetBlockDim(finalCoreNum);
return ge::GRAPH_SUCCESS;
}
static ge::graphStatus ComplexV2TilingFunc(gert::TilingContext* context)
{
PlatformInfo platformInfo;
OP_CHECK_IF(GetPlatformInfo(context, platformInfo) != ge::GRAPH_SUCCESS,
OP_LOGE(context, "GetPlatformInfo error"), return ge::GRAPH_FAILED);
int64_t totalIdx = 0;
ge::DataType dataType = ge::DT_FLOAT;
OP_CHECK_IF(GetShapeAttrsInfo(context, totalIdx, dataType) != ge::GRAPH_SUCCESS,
OP_LOGE(context, "GetShapeAttrsInfo error"), return ge::GRAPH_FAILED);
if (totalIdx <= 0) {
ComplexV2TilingData* tiling = context->GetTilingData<ComplexV2TilingData>();
context->GetRawTilingData()->SetDataSize(sizeof(ComplexV2TilingData));
OP_CHECK_NULL_WITH_CONTEXT(context, tiling);
memset_s(tiling, sizeof(ComplexV2TilingData), 0, sizeof(ComplexV2TilingData));
context->SetBlockDim(1);
context->SetTilingKey(GET_TPL_TILING_KEY(ELEMENTWISE_TPL_SCH_MODE_0));
return ge::GRAPH_SUCCESS;
}
OP_CHECK_IF(GetWorkspaceSize(context) != ge::GRAPH_SUCCESS,
OP_LOGE(context, "GetWorkspaceSize error"), return ge::GRAPH_FAILED);
ComplexV2TilingData* tiling = context->GetTilingData<ComplexV2TilingData>();
OP_CHECK_NULL_WITH_CONTEXT(context, tiling);
OP_CHECK_IF(memset_s(tiling, sizeof(ComplexV2TilingData), 0, sizeof(ComplexV2TilingData)) != EOK,
OP_LOGE(context, "set tiling data error"), return ge::GRAPH_FAILED);
OP_CHECK_IF(CalculateCoreNum(context, tiling, platformInfo, totalIdx) != ge::GRAPH_SUCCESS,
OP_LOGE(context, "CalculateCoreNum error"), return ge::GRAPH_FAILED);
context->GetRawTilingData()->SetDataSize(sizeof(ComplexV2TilingData));
return ge::GRAPH_SUCCESS;
}
IMPL_OP_OPTILING(ComplexV2).Tiling(ComplexV2TilingFunc);
}