* 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.
*/
* \file inv_tiling_arch35.cpp
* \brief Inv 算子 Host Tiling 实现（arch35 / Ascend950）
*
* Tiling 策略：
* 1. 多核切分：blockFactor = CeilAlign(CeilDiv(total, coreNum), 32B/typeSize)
* 2. UB 切分：bytesPerElem 按 dtype 分支（含 BUFFER_NUM=2 的 double buffer）
* - FP32: 2 * BUFFER_NUM * 4 + 4 = 20 (input+output*DB + ones)
* - FP16/BF16: 2 * BUFFER_NUM * 2 + 4 + 4 = 16 (input+output*DB + xFloat + ones)
* - INT32: 2 * BUFFER_NUM * 4 = 16 (input+output*DB，无 fp32 工作字节)
* ubFactor = FloorAlign(ubSize / bytesPerElem, 32B/typeSize)
* 3. Buffer layout: inputQueue(BUFFER_NUM) + outputQueue(BUFFER_NUM) + [tmpBuf1(fp32, 仅FP16/BF16)] + [tmpBuf2(fp32 ones, 仅浮点)]
* int32 路径仅 input+output（纯整型 Compare/Select，无 fp32 中转，评审 MED-2）
*
* TilingKey 编码（按 dtype 维度 D_T_SELF 分发）：
* key 0 -> C_DT_FLOAT (float32 直接 Div)
* key 1 -> C_DT_FLOAT16 (float16: Cast→Div→Cast)
* key 2 -> C_DT_BF16 (bfloat16: Cast→Div→Cast)
* key 3 -> C_DT_INT32 (int32: 整型 Compare+Select 三值映射)
*/
#include "register/op_def_registry.h"
#include "op_common/log/log.h"
#include "op_common/op_host/util/math_util.h"
#include "op_common/op_host/util/platform_util.h"
#include "../../op_kernel/arch35/inv_tiling_data.h"
#include "../../op_kernel/arch35/inv_tiling_key.h"
namespace optiling {
using Ops::Base::CeilDiv;
using Ops::Base::FloorDiv;
using Ops::Base::FloorAlign;
constexpr uint32_t WS_SYS_SIZE = 0U;
static const gert::Shape g_vec_1_shape = {1};
static inline const gert::Shape EnsureNotScalar(const gert::Shape& in_shape)
{
if (in_shape.GetDimNum() == 0) {
return g_vec_1_shape;
}
return in_shape;
}
static ge::graphStatus GetPlatformInfo(gert::TilingContext* context, uint64_t& ubSize, int64_t& coreNum)
{
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, "Inv: coreNum is 0"), return ge::GRAPH_FAILED);
ascendcPlatform.GetCoreMemSize(platform_ascendc::CoreMemType::UB, ubSize);
OP_CHECK_IF(ubSize == 0, OP_LOGE(context, "Inv: ubSize is 0"), return ge::GRAPH_FAILED);
return ge::GRAPH_SUCCESS;
}
static ge::graphStatus GetShapeInfo(gert::TilingContext* context, int64_t& totalElements,
ge::DataType& dataType)
{
auto inputSelf = context->GetInputShape(0);
OP_CHECK_NULL_WITH_CONTEXT(context, inputSelf);
auto inputShape = EnsureNotScalar(inputSelf->GetStorageShape());
totalElements = inputShape.GetShapeSize();
auto inputDesc = context->GetInputDesc(0);
OP_CHECK_NULL_WITH_CONTEXT(context, inputDesc);
dataType = inputDesc->GetDataType();
const std::set<ge::DataType> supportedDtype = {
ge::DT_FLOAT, ge::DT_FLOAT16, ge::DT_BF16, ge::DT_INT32
};
OP_CHECK_IF(supportedDtype.count(dataType) == 0,
OP_LOGE(context, "Inv: unsupported dtype %d", static_cast<int>(dataType)),
return ge::GRAPH_FAILED);
return ge::GRAPH_SUCCESS;
}
static ge::graphStatus GetWorkspaceSize(gert::TilingContext* context)
{
size_t* currentWorkspace = context->GetWorkspaceSizes(1);
OP_CHECK_NULL_WITH_CONTEXT(context, currentWorkspace);
currentWorkspace[0] = WS_SYS_SIZE;
return ge::GRAPH_SUCCESS;
}
static int64_t GetTypeSize(ge::DataType dataType)
{
return (dataType == ge::DT_FLOAT16 || dataType == ge::DT_BF16) ? 2 : 4;
}
static int64_t CalcBytesPerElem(ge::DataType dataType, int64_t typeSize)
{
constexpr int64_t BUFFER_NUM = 2;
if (dataType == ge::DT_INT32) {
return 2 * BUFFER_NUM * typeSize;
}
int64_t fp32WorkBytes = (dataType == ge::DT_FLOAT)
? static_cast<int64_t>(sizeof(float))
: 2 * static_cast<int64_t>(sizeof(float));
return 2 * BUFFER_NUM * typeSize + fp32WorkBytes;
}
static ge::graphStatus CalcTilingParams(gert::TilingContext* context, InvTilingData* tiling,
int64_t totalElements, int64_t coreNum,
uint64_t ubSize, ge::DataType dataType)
{
int64_t typeSize = GetTypeSize(dataType);
OP_CHECK_IF(typeSize == 0, OP_LOGE(context, "Inv: typeSize is 0"), return ge::GRAPH_FAILED);
int64_t ubBlockSize = 32 / typeSize;
int64_t blockFactor = CeilDiv(totalElements, coreNum);
blockFactor = ((blockFactor + ubBlockSize - 1) / ubBlockSize) * ubBlockSize;
int64_t usedCoreNum = CeilDiv(totalElements, blockFactor);
int64_t bytesPerElem = CalcBytesPerElem(dataType, typeSize);
OP_CHECK_IF(bytesPerElem == 0, OP_LOGE(context, "Inv: bytesPerElem is 0"), return ge::GRAPH_FAILED);
int64_t ubFactor = FloorAlign(static_cast<int64_t>(ubSize) / bytesPerElem, ubBlockSize);
OP_CHECK_IF(ubFactor <= 0, OP_LOGE(context, "Inv: ubFactor=%ld, UB too small", ubFactor),
return ge::GRAPH_FAILED);
tiling->totalElements = totalElements;
tiling->blockFactor = blockFactor;
tiling->ubFactor = ubFactor;
context->SetBlockDim(usedCoreNum);
return ge::GRAPH_SUCCESS;
}
static ge::graphStatus InvTilingFunc(gert::TilingContext* context)
{
OP_LOGD(context->GetNodeName(), "Enter InvTilingFunc");
uint64_t ubSize = 0;
int64_t coreNum = 0;
OP_CHECK_IF(GetPlatformInfo(context, ubSize, coreNum) != ge::GRAPH_SUCCESS,
OP_LOGE(context, "Inv: GetPlatformInfo error"), return ge::GRAPH_FAILED);
int64_t totalElements = 0;
ge::DataType dataType = ge::DT_FLOAT;
OP_CHECK_IF(GetShapeInfo(context, totalElements, dataType) != ge::GRAPH_SUCCESS,
OP_LOGE(context, "Inv: GetShapeInfo error"), return ge::GRAPH_FAILED);
OP_CHECK_IF(GetWorkspaceSize(context) != ge::GRAPH_SUCCESS,
OP_LOGE(context, "Inv: GetWorkspaceSize error"), return ge::GRAPH_FAILED);
InvTilingData* tiling = context->GetTilingData<InvTilingData>();
OP_CHECK_NULL_WITH_CONTEXT(context, tiling);
OP_CHECK_IF(memset_s(tiling, sizeof(InvTilingData), 0, sizeof(InvTilingData)) != EOK,
OP_LOGE(context, "Inv: set tiling data error"), return ge::GRAPH_FAILED);
if (totalElements == 0) {
context->SetBlockDim(1);
} else {
OP_CHECK_IF(CalcTilingParams(context, tiling, totalElements, coreNum, ubSize, dataType)
!= ge::GRAPH_SUCCESS,
OP_LOGE(context, "Inv: CalcTilingParams error"), return ge::GRAPH_FAILED);
}
ASCENDC_TPL_SEL_PARAM(context, static_cast<uint32_t>(dataType));
return ge::GRAPH_SUCCESS;
}
static ge::graphStatus TilingParseForInv([[maybe_unused]] gert::TilingParseContext* context)
{
return ge::GRAPH_SUCCESS;
}
struct InvCompileInfo {};
IMPL_OP_OPTILING(Inv).Tiling(InvTilingFunc).TilingParse<InvCompileInfo>(TilingParseForInv);
}
