* 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.
*/
* \file cos_tiling_arch35.cpp
* \brief cos_tiling_arch35
*/
#include <graph/utils/type_utils.h>
#include "cos_tiling_arch35.h"
#include "register/op_impl_registry.h"
#include "math/cos/op_kernel/arch35/cos_dag.h"
#include "math/cos/op_kernel/arch35/cos_struct.h"
#include "tiling/platform/platform_ascendc.h"
#include "op_host/tiling_base_util.h"
namespace optiling
{
using namespace Ops::Base;
const int64_t ASCEND_WORKSPACE = 16777216;
const int64_t ASCEND_API_BUFFER = 122880;
const int64_t DCACHE_SIZE = 32768;
ge::graphStatus CosTiling::SetTilingData()
{
OP_LOGD(tilingContext->GetNodeName(), "CosTiling SetTilingData enter.");
size_t* currentWorkspace = tilingContext->GetWorkspaceSizes(1);
OP_CHECK_NULL_WITH_CONTEXT(tilingContext, currentWorkspace);
currentWorkspace[0] = static_cast<size_t>(ASCEND_WORKSPACE);
const uint64_t tilingKey = GET_TPL_TILING_KEY(tiling->baseTiling.scheMode, dType);
OP_LOGD(tilingContext->GetNodeName(), "[TilingData] : tilingKey=%lu", tilingKey);
tilingContext->SetTilingKey(tilingKey);
tilingContext->SetBlockDim(tiling->baseTiling.blockNum);
uint64_t ubSize = 0;
auto platformInfo = tilingContext->GetPlatformInfo();
if (platformInfo == nullptr) {
auto compileInfoPtr = tilingContext->GetCompileInfo<ElewiseCompileInfo>();
OP_CHECK_IF(compileInfoPtr == nullptr, OP_LOGE_FOR_INVALID_VALUE_WITH_REASON(tilingContext->GetNodeName(), "compile_info", "nullptr", "compile info is null"),
return ge::GRAPH_FAILED);
ubSize = compileInfoPtr->ubSize;
} else {
auto ascendcPlatform = platform_ascendc::PlatformAscendC(platformInfo);
uint64_t ubSizePlatForm = 0;
ascendcPlatform.GetCoreMemSize(platform_ascendc::CoreMemType::UB, ubSizePlatForm);
ubSize = ubSizePlatForm;
}
tilingContext->SetLocalMemorySize(static_cast<uint32_t>(ubSize - DCACHE_SIZE));
return ge::GRAPH_SUCCESS;
}
ge::graphStatus CosTiling::CalcInputDtype()
{
OP_LOGD(tilingContext->GetNodeName(), "CosTiling CalcInputDtype enter.");
auto inputDesc = tilingContext->GetInputDesc(0);
OP_CHECK_NULL_WITH_CONTEXT(tilingContext, inputDesc);
this->inputDtype = inputDesc->GetDataType();
OP_CHECK_IF(
this->inputDtype != ge::DT_FLOAT16 && this->inputDtype != ge::DT_BF16 && this->inputDtype != ge::DT_FLOAT,
OP_LOGE_FOR_INVALID_DTYPE_WITH_REASON(tilingContext->GetNodeName(), "x", ge::TypeUtils::DataTypeToSerialString(this->inputDtype), "dtype not in [DT_FLOAT16, DT_BF16, DT_FLOAT]"),
return ge::GRAPH_FAILED);
return ge::GRAPH_SUCCESS;
}
ge::graphStatus CosTiling::CheckShape()
{
OP_LOGD(tilingContext->GetNodeName(), "CosTiling CheckShape enter.");
auto inputStorageShape = tilingContext->GetInputShape(0);
OP_CHECK_NULL_WITH_CONTEXT(tilingContext, inputStorageShape);
const gert::Shape& inputYShape = Ops::Base::EnsureNotScalar(inputStorageShape->GetStorageShape());
auto outputStorageShape = tilingContext->GetOutputShape(0);
OP_CHECK_NULL_WITH_CONTEXT(tilingContext, outputStorageShape);
const gert::Shape& outputZShape = Ops::Base::EnsureNotScalar(outputStorageShape->GetStorageShape());
OP_CHECK_IF(inputYShape != outputZShape,
OP_LOGE_FOR_INVALID_SHAPES_WITH_REASON(tilingContext->GetNodeName(), "x, y", (Ops::Base::ToString(inputYShape) + ", " + Ops::Base::ToString(outputZShape)).c_str(), "input shape must equal output shape"),
return ge::GRAPH_FAILED);
return ge::GRAPH_SUCCESS;
}
ge::graphStatus CosTiling::CalcOutputDtype()
{
OP_LOGD(tilingContext->GetNodeName(), "CosTiling CalcOutputDtype enter.");
auto outputDesc = tilingContext->GetOutputDesc(0);
OP_CHECK_NULL_WITH_CONTEXT(tilingContext, outputDesc);
this->outputDtype = outputDesc->GetDataType();
OP_CHECK_IF(this->outputDtype != this->inputDtype,
OP_LOGE_FOR_INVALID_DTYPES_WITH_REASON(tilingContext->GetNodeName(), "outputDtype, inputDtype", std::string(ge::TypeUtils::DataTypeToSerialString(this->outputDtype)) + ", " + std::string(ge::TypeUtils::DataTypeToSerialString(this->inputDtype)), "output dtype must be same as input dtype"),
return ge::GRAPH_FAILED);
return ge::GRAPH_SUCCESS;
}
ge::graphStatus CosTiling::RunTiling()
{
OP_LOGD(tilingContext->GetNodeName(), "CosTiling RunTiling enter.");
Ops::Base::ElewiseBaseTiling elewiseBaseTiling(tilingContext);
OP_CHECK_IF(CalcInputDtype() == ge::GRAPH_FAILED,
OP_LOGE(tilingContext->GetNodeName(), "get input dtype failed"), return ge::GRAPH_FAILED);
OP_CHECK_IF(CalcOutputDtype() == ge::GRAPH_FAILED,
OP_LOGE(tilingContext->GetNodeName(), "get output dtype failed"), return ge::GRAPH_FAILED);
OP_CHECK_IF(CheckShape() == ge::GRAPH_FAILED, OP_LOGE(tilingContext->GetNodeName(), "check shape failed"),
return ge::GRAPH_FAILED);
tiling = tilingContext->GetTilingData<CosTilingData>();
OP_CHECK_NULL_WITH_CONTEXT(tilingContext, tiling);
ge::graphStatus baseTilingResult = ge::GRAPH_FAILED;
if (this->outputDtype == ge::DT_FLOAT16) {
dType = TPL_FP16;
baseTilingResult = elewiseBaseTiling.DoTiling<CosOp::CosDAG<Ops::Base::half>::OpDag>(tiling->baseTiling, ASCEND_API_BUFFER + DCACHE_SIZE);
} else if (this->outputDtype == ge::DT_BF16) {
dType = TPL_BF16;
baseTilingResult = elewiseBaseTiling.DoTiling<CosOp::CosDAG<Ops::Base::bfloat16_t>::OpDag>(tiling->baseTiling, ASCEND_API_BUFFER + DCACHE_SIZE);
} else if (this->outputDtype == ge::DT_FLOAT) {
dType = TPL_FP32;
baseTilingResult = elewiseBaseTiling.DoTiling<CosOp::CosDAG<float>::OpDag>(tiling->baseTiling, ASCEND_API_BUFFER + DCACHE_SIZE);
} else {
OP_LOGE_FOR_INVALID_DTYPE_WITH_REASON(tilingContext->GetNodeName(), "y", ge::TypeUtils::DataTypeToSerialString(this->outputDtype), "dtype not in [DT_FLOAT16, DT_BF16, DT_FLOAT]");
return ge::GRAPH_FAILED;
}
OP_CHECK_IF(baseTilingResult == ge::GRAPH_FAILED,
OP_LOGE(tilingContext->GetNodeName(), "elewiseBaseTiling failed"), return ge::GRAPH_FAILED);
return SetTilingData();
}
static ge::graphStatus TilingForCos(gert::TilingContext* tilingContextGen)
{
OP_LOGD(tilingContextGen->GetNodeName(), "TilingForCos rt2.0 is running.");
auto compileInfo = tilingContextGen->GetCompileInfo<ElewiseCompileInfo>();
OP_CHECK_NULL_WITH_CONTEXT(tilingContextGen, compileInfo);
CosTiling baseOpTiling(tilingContextGen);
return baseOpTiling.RunTiling();
}
ge::graphStatus TilingPrepareForCos(gert::TilingParseContext* context)
{
OP_LOGD(context->GetNodeName(), "TilingPrepareForCos is running.");
auto compileInfo = context->GetCompiledInfo<ElewiseCompileInfo>();
OP_CHECK_NULL_WITH_CONTEXT(context, compileInfo);
auto platformInfo = context->GetPlatformInfo();
OP_CHECK_NULL_WITH_CONTEXT(context, platformInfo);
auto ascendcPlatform = platform_ascendc::PlatformAscendC(platformInfo);
compileInfo->coreNum = ascendcPlatform.GetCoreNumAiv();
uint64_t ubSize;
ascendcPlatform.GetCoreMemSize(platform_ascendc::CoreMemType::UB, ubSize);
compileInfo->ubSize = static_cast<int64_t>(ubSize);
return ge::GRAPH_SUCCESS;
}
IMPL_OP_OPTILING(Cos).Tiling(TilingForCos).TilingParse<ElewiseCompileInfo>(TilingPrepareForCos);
}
