Star1296
1410
代码
代码
Issues108
Pull Requests213
Actions
项目讨论
Wiki
分析
Star1296
1410
已合并
【bugfix】修复math仓cleancode #3925
majiajian_hw创建于 7月8日
【bugfix】修复math仓cleancode #3925
已合并
majiajian_hw创建于 7月8日
从nunnons2/ops-mathbugfix_cleancode0708_001合入到cann/ops-mathmaster
共 5 个文件变更+101-110
Mconversion/chunk_cat/op_host/chunk_cat_tiling.cpp+39-40
@@ -19,26 +19,29 @@
19 19 
20namespace optiling {20namespace optiling {
21 21 
22-constexpr uint32_t UB_BLOCK_SIZE = 32; // UB块大小22+constexpr uint32_t UB_BLOCK_SIZE = 32; // UB块大小
23-constexpr uint32_t TRANS_BLOCK = 16; // 转置行数23+constexpr uint32_t TRANS_BLOCK = 16; // 转置行数
24-constexpr uint32_t RESERVE_UB = 256; // 接口获取UB的预留空间24+constexpr uint32_t RESERVE_UB = 256; // 接口获取UB的预留空间
25-constexpr uint32_t HALF = 2; // 半对齐/UB对半切分25+constexpr uint32_t HALF = 2; // 半对齐/UB对半切分
26-constexpr uint32_t ONETHIRD = 3; // UB对三切分26+constexpr uint32_t ONETHIRD = 3; // UB对三切分
27constexpr uint32_t DEFAUL_TILING_KEY = 0; // 默认tiling key27constexpr uint32_t DEFAUL_TILING_KEY = 0; // 默认tiling key
28+constexpr uint32_t NUM_THIRTY_TWO = 32;
28 29 
29-static const std::set<ge::DataType> supportedDtype =30+static const std::set<ge::DataType> supportedDtype = {ge::DT_FLOAT, ge::DT_FLOAT16, ge::DT_BF16};
30- {ge::DT_FLOAT, ge::DT_FLOAT16, ge::DT_BF16};
31 31 
32-std::string ChunkCatTiling::TilingDataToString() const {32+std::string ChunkCatTiling::TilingDataToString() const
33+{
33 return "blockRowNum = " + std::to_string(blockRowNum_) + ", blockColNum = " + std::to_string(blockColNum_) +34 return "blockRowNum = " + std::to_string(blockRowNum_) + ", blockColNum = " + std::to_string(blockColNum_) +
34- ", dim = " + std::to_string(dim_) + ", numChunk = " + std::to_string(numChunk_) +35+ ", dim = " + std::to_string(dim_) + ", numChunk = " + std::to_string(numChunk_) +
35- ", outputRow = " + std::to_string(outputRow_) + ", outputCol = " + std::to_string(outputCol_) +36+ ", outputRow = " + std::to_string(outputRow_) + ", outputCol = " + std::to_string(outputCol_) +
36- ", blockRowFactor = " + std::to_string(blockRowFactor_) + ", blockColFactor = " + std::to_string(blockColFactor_) +37+ ", blockRowFactor = " + std::to_string(blockRowFactor_) +
37- ", tailBlockRowFactor = " + std::to_string(tailBlockRowFactor_) + ", tailBlockColFactor = " + std::to_string(tailBlockColFactor_) +38+ ", blockColFactor = " + std::to_string(blockColFactor_) +
38- ", ubRowFactor = " + std::to_string(ubRowFactor_) + ", ubColFactor = " + std::to_string(ubColFactor_) +39+ ", tailBlockRowFactor = " + std::to_string(tailBlockRowFactor_) +
39- ", inputNum = " + std::to_string(inputNum_) + ", inUbSize = " + std::to_string(inUbSize_) +40+ ", tailBlockColFactor = " + std::to_string(tailBlockColFactor_) +
40- ", outUbSize = " + std::to_string(outUbSize_) + ", isAllAlign = " + std::to_string(isAllAlign_) +41+ ", ubRowFactor = " + std::to_string(ubRowFactor_) + ", ubColFactor = " + std::to_string(ubColFactor_) +
41- ", isHalfAlign = " + std::to_string(isHalfAlign_) + ", isOneConcat = " + std::to_string(isOneConcat_);42+ ", inputNum = " + std::to_string(inputNum_) + ", inUbSize = " + std::to_string(inUbSize_) +
43+ ", outUbSize = " + std::to_string(outUbSize_) + ", isAllAlign = " + std::to_string(isAllAlign_) +
44+ ", isHalfAlign = " + std::to_string(isHalfAlign_) + ", isOneConcat = " + std::to_string(isOneConcat_);
42}45}
43 46 
44// 获取硬件信息47// 获取硬件信息
@@ -83,8 +86,10 @@ ge::graphStatus ChunkCatTiling::GetInputInfo()
83 auto outputDesc = context_->GetOutputDesc(0);86 auto outputDesc = context_->GetOutputDesc(0);
84 OP_CHECK_NULL_WITH_CONTEXT(context_, outputDesc);87 OP_CHECK_NULL_WITH_CONTEXT(context_, outputDesc);
85 auto outputDataType = outputDesc->GetDataType();88 auto outputDataType = outputDesc->GetDataType();
86- OP_CHECK_IF(supportedDtype.count(inputDataType) == 0, OP_LOGE(context_, "input dtype is invalid"), return ge::GRAPH_FAILED);89+ OP_CHECK_IF(supportedDtype.count(inputDataType) == 0, OP_LOGE(context_, "input dtype is invalid"),
87- OP_CHECK_IF(supportedDtype.count(outputDataType) == 0, OP_LOGE(context_, "output dtype is invalid"), return ge::GRAPH_FAILED);90+ return ge::GRAPH_FAILED);
91+ OP_CHECK_IF(supportedDtype.count(outputDataType) == 0, OP_LOGE(context_, "output dtype is invalid"),
92+ return ge::GRAPH_FAILED);
88 if (inputDataType == ge::DT_FLOAT && outputDataType != ge::DT_FLOAT) {93 if (inputDataType == ge::DT_FLOAT && outputDataType != ge::DT_FLOAT) {
89 OP_LOGE(context_, "output dtype must be float when input dtype is float");94 OP_LOGE(context_, "output dtype must be float when input dtype is float");
90 return ge::GRAPH_FAILED;95 return ge::GRAPH_FAILED;
@@ -144,30 +149,30 @@ void ChunkCatTiling::DoUbSplit()
144 if (isRegBase) {149 if (isRegBase) {
145 // 列切150 // 列切
146 uint32_t colLimit = inUbSize_ / srcDtypeSize_;151 uint32_t colLimit = inUbSize_ / srcDtypeSize_;
147- colLimit = colLimit - 32 * srcEleUbBlock_;152+ colLimit = colLimit - NUM_THIRTY_TWO * srcEleUbBlock_;
148- int64_t ubColLoop = (outputCol_ + colLimit - 1) / colLimit ;153+ int64_t ubColLoop = (outputCol_ + colLimit - 1) / colLimit;
149 ubColFactor_ = (outputCol_ + ubColLoop - 1) / ubColLoop;154 ubColFactor_ = (outputCol_ + ubColLoop - 1) / ubColLoop;
150 ubColFactor_ = (ubColFactor_ + srcEleUbBlock_ - 1) / srcEleUbBlock_ * srcEleUbBlock_;155 ubColFactor_ = (ubColFactor_ + srcEleUbBlock_ - 1) / srcEleUbBlock_ * srcEleUbBlock_;
151 ubColFactor_ = ubColFactor_ > colLimit ? colLimit : ubColFactor_;156 ubColFactor_ = ubColFactor_ > colLimit ? colLimit : ubColFactor_;
152 // 行切157 // 行切
153 uint32_t rowLimit = ubColLoop == 1 ? colLimit / ubColFactor_ : 1;158 uint32_t rowLimit = ubColLoop == 1 ? colLimit / ubColFactor_ : 1;
154- int64_t ubRowLoop = (outputRow_ + rowLimit - 1) / rowLimit ;159+ int64_t ubRowLoop = (outputRow_ + rowLimit - 1) / rowLimit;
155 ubRowFactor_ = (outputRow_ + ubRowLoop - 1) / ubRowLoop;160 ubRowFactor_ = (outputRow_ + ubRowLoop - 1) / ubRowLoop;
156 } else if (isAllAlign_ || isOneConcat_) {161 } else if (isAllAlign_ || isOneConcat_) {
157 // 列切162 // 列切
158 uint32_t colLimit = inUbSize_ / srcDtypeSize_;163 uint32_t colLimit = inUbSize_ / srcDtypeSize_;
159 colLimit = (isOneConcat_) ? colLimit - 32 * (srcEleUbBlock_ - 1) : colLimit;164 colLimit = (isOneConcat_) ? colLimit - 32 * (srcEleUbBlock_ - 1) : colLimit;
160- int64_t ubColLoop = (outputCol_ + colLimit - 1) / colLimit ;165+ int64_t ubColLoop = (outputCol_ + colLimit - 1) / colLimit;
161 ubColFactor_ = (outputCol_ + ubColLoop - 1) / ubColLoop;166 ubColFactor_ = (outputCol_ + ubColLoop - 1) / ubColLoop;
162 ubColFactor_ = (ubColFactor_ + srcEleUbBlock_ - 1) / srcEleUbBlock_ * srcEleUbBlock_;167 ubColFactor_ = (ubColFactor_ + srcEleUbBlock_ - 1) / srcEleUbBlock_ * srcEleUbBlock_;
163 ubColFactor_ = ubColFactor_ > colLimit ? colLimit : ubColFactor_;168 ubColFactor_ = ubColFactor_ > colLimit ? colLimit : ubColFactor_;
164 // 行切169 // 行切
165 uint32_t rowLimit = ubColLoop == 1 ? colLimit / ubColFactor_ : 1;170 uint32_t rowLimit = ubColLoop == 1 ? colLimit / ubColFactor_ : 1;
166- int64_t ubRowLoop = (outputRow_ + rowLimit - 1) / rowLimit ;171+ int64_t ubRowLoop = (outputRow_ + rowLimit - 1) / rowLimit;
167 ubRowFactor_ = (outputRow_ + ubRowLoop - 1) / ubRowLoop;172 ubRowFactor_ = (outputRow_ + ubRowLoop - 1) / ubRowLoop;
168 } else {173 } else {
169 // 行切174 // 行切
170- uint32_t rowLimit = isHalfAlign_ ? TRANS_BLOCK * HALF: TRANS_BLOCK * srcEleUbBlock_;175+ uint32_t rowLimit = isHalfAlign_ ? TRANS_BLOCK * HALF : TRANS_BLOCK * srcEleUbBlock_;
171 int64_t ubRowLoop = (outputRow_ + rowLimit - 1) / rowLimit;176 int64_t ubRowLoop = (outputRow_ + rowLimit - 1) / rowLimit;
172 ubRowFactor_ = (outputRow_ + ubRowLoop - 1) / ubRowLoop;177 ubRowFactor_ = (outputRow_ + ubRowLoop - 1) / ubRowLoop;
173 // 列切178 // 列切
@@ -228,23 +233,19 @@ static ge::graphStatus Tiling4ChunkCat(gert::TilingContext* context)
228{233{
229 OP_LOGD(context, "ChunkCatTiling");234 OP_LOGD(context, "ChunkCatTiling");
230 ChunkCatTiling tiling(context);235 ChunkCatTiling tiling(context);
231- OP_CHECK_IF(236+ OP_CHECK_IF(tiling.GetPlatformInfo() != ge::GRAPH_SUCCESS, OP_LOGE(context, "GetPlatformInfo error"),
232- tiling.GetPlatformInfo() != ge::GRAPH_SUCCESS, OP_LOGE(context, "GetPlatformInfo error"),237+ return ge::GRAPH_FAILED);
233- return ge::GRAPH_FAILED);238+ OP_CHECK_IF(tiling.GetInputInfo() != ge::GRAPH_SUCCESS, OP_LOGE(context, "GetInputInfo error"),
234- OP_CHECK_IF(239+ return ge::GRAPH_FAILED);
235- tiling.GetInputInfo() != ge::GRAPH_SUCCESS, OP_LOGE(context, "GetInputInfo error"),240+ OP_CHECK_IF(tiling.CalculateOutputInfo() != ge::GRAPH_SUCCESS, OP_LOGE(context, "CalculateOutputInfo error"),
236- return ge::GRAPH_FAILED);241+ return ge::GRAPH_FAILED);
237- OP_CHECK_IF(
238- tiling.CalculateOutputInfo() != ge::GRAPH_SUCCESS, OP_LOGE(context, "CalculateOutputInfo error"),
239- return ge::GRAPH_FAILED);
240 tiling.DoUbSplit();242 tiling.DoUbSplit();
241 tiling.DoBlockSplit();243 tiling.DoBlockSplit();
242- 244+ 
243 ChunkCatTilingData* tilingData = context->GetTilingData<ChunkCatTilingData>();245 ChunkCatTilingData* tilingData = context->GetTilingData<ChunkCatTilingData>();
244 OP_CHECK_NULL_WITH_CONTEXT(context, tilingData);246 OP_CHECK_NULL_WITH_CONTEXT(context, tilingData);
245- OP_CHECK_IF(247+ OP_CHECK_IF(memset_s(tilingData, sizeof(ChunkCatTilingData), 0, sizeof(ChunkCatTilingData)) != EOK,
246- memset_s(tilingData, sizeof(ChunkCatTilingData), 0, sizeof(ChunkCatTilingData)) != EOK,248+ OP_LOGE(context, "set tiling data error"), return ge::GRAPH_FAILED);
247- OP_LOGE(context, "set tiling data error"), return ge::GRAPH_FAILED);
248 tiling.SetTilingData(tilingData);249 tiling.SetTilingData(tilingData);
249 OP_LOGD(context, "tiling data: %s", tiling.TilingDataToString().c_str());250 OP_LOGD(context, "tiling data: %s", tiling.TilingDataToString().c_str());
250 return ge::GRAPH_SUCCESS;251 return ge::GRAPH_SUCCESS;
@@ -255,7 +256,5 @@ static ge::graphStatus TilingPrepare4ChunkCat([[maybe_unused]] gert::TilingParse
255 return ge::GRAPH_SUCCESS;256 return ge::GRAPH_SUCCESS;
256}257}
257 258 
258-IMPL_OP_OPTILING(ChunkCat)259+IMPL_OP_OPTILING(ChunkCat).Tiling(Tiling4ChunkCat).TilingParse<ChunkCatCompileInfo>(TilingPrepare4ChunkCat);
259- .Tiling(Tiling4ChunkCat)
260- .TilingParse<ChunkCatCompileInfo>(TilingPrepare4ChunkCat);
261} // namespace optiling260} // namespace optiling
Mconversion/chunk_cat/op_host/chunk_cat_tiling.h+1-1
@@ -40,7 +40,7 @@ public:
40 40 
41private:41private:
42 gert::TilingContext* context_;42 gert::TilingContext* context_;
43- bool isRegBase;43+ bool isRegBase{false};
44 int32_t coreNum_{0};44 int32_t coreNum_{0};
45 uint64_t ubSize_{0};45 uint64_t ubSize_{0};
46 uint64_t sysWorkspaceSize_{0};46 uint64_t sysWorkspaceSize_{0};
Mrandom/dsa_random_normal/op_host/op_api/aclnn_normal.cpp+2-2
@@ -186,7 +186,7 @@ static const aclTensor* normalDavidPath(const aclTensor* selfContiguous, int64_t
186 186 
187 // 调用normal_算子kernel function(AI Cpu算子)187 // 调用normal_算子kernel function(AI Cpu算子)
188 int64_t alg = 1;188 int64_t alg = 1;
189- auto algScalar = executor->AllocScalar((void*)&alg, DataType::DT_INT32);189+ auto algScalar = executor->AllocScalar(reinterpret_cast<void*>(&alg), DataType::DT_INT32);
190 const aclTensor* algTensor = executor->ConvertToTensor(algScalar, op::ToOpDataType(ACL_INT32));190 const aclTensor* algTensor = executor->ConvertToTensor(algScalar, op::ToOpDataType(ACL_INT32));
191 auto stateLessOut = l0op::StatelessRandomNormalV2(selfContiguous, keyArr, counterArr, algTensor, executor);191 auto stateLessOut = l0op::StatelessRandomNormalV2(selfContiguous, keyArr, counterArr, algTensor, executor);
192 return normalDoublePath(selfContiguous, stateLessOut, mean, std, executor);192 return normalDoublePath(selfContiguous, stateLessOut, mean, std, executor);
@@ -237,7 +237,7 @@ static const aclTensor* normalTensorDavidPath(const aclTensor* selfContiguous, c
237 CHECK_RET(resultAddOut != nullptr, nullptr);237 CHECK_RET(resultAddOut != nullptr, nullptr);
238 238 
239 int64_t alg = 1;239 int64_t alg = 1;
240- auto algScalar = executor->AllocScalar((void*)&alg, DataType::DT_INT32);240+ auto algScalar = executor->AllocScalar(reinterpret_cast<void*>(&alg), DataType::DT_INT32);
241 const aclTensor* algTensor = executor->ConvertToTensor(algScalar, op::ToOpDataType(ACL_INT32));241 const aclTensor* algTensor = executor->ConvertToTensor(algScalar, op::ToOpDataType(ACL_INT32));
242 auto stateLessOut = l0op::StatelessRandomNormalV2(selfContiguous, normalSeedU64, resultAddOut, algTensor,242 auto stateLessOut = l0op::StatelessRandomNormalV2(selfContiguous, normalSeedU64, resultAddOut, algTensor,
243 executor);243 executor);
Mrandom/sim_thread_exponential/op_host/arch32/sim_thread_exponential_tiling_arch32.cpp+42-50
@@ -131,18 +131,17 @@ ge::graphStatus SimThreadExponentialTiling::GetInputTensorInfo()
131{131{
132 auto nodeName = context->GetNodeName();132 auto nodeName = context->GetNodeName();
133 133 
134- OP_CHECK_IF(134+ OP_CHECK_IF(count < 0, OP_LOGE(nodeName, "Count %ld must be greater than or equal to 0.", count),
135- count < 0, OP_LOGE(nodeName, "Count %ld must be greater than or equal to 0.", count), return ge::GRAPH_FAILED);135+ return ge::GRAPH_FAILED);
136- OP_CHECK_IF(136+ OP_CHECK_IF(threadPerProcessor <= 0,
137- threadPerProcessor <= 0, OP_LOGE(nodeName, "ThreadPerProcessor %d must be greater than 0.", threadPerProcessor),137+ OP_LOGE(nodeName, "ThreadPerProcessor %d must be greater than 0.", threadPerProcessor),
138- return ge::GRAPH_FAILED);138+ return ge::GRAPH_FAILED);
139- OP_CHECK_IF(139+ OP_CHECK_IF(streamProcessorCount <= 0,
140- streamProcessorCount <= 0,140+ OP_LOGE(nodeName, "StreamProcessorCount %d must be greater than 0.", streamProcessorCount),
141- OP_LOGE(nodeName, "StreamProcessorCount %d must be greater than 0.", streamProcessorCount),141+ return ge::GRAPH_FAILED);
142- return ge::GRAPH_FAILED);142+ OP_CHECK_IF(start > end, // 如何获取start、end？
143- OP_CHECK_IF(143+ OP_LOGE(nodeName, "Start %f must be less than or equal to end %f.", start, end),
144- start > end, // 如何获取start、end？144+ return ge::GRAPH_FAILED);
145- OP_LOGE(nodeName, "Start %f must be less than or equal to end %f.", start, end), return ge::GRAPH_FAILED);
146 145 
147 // 获取第一个输入gradOut的信息146 // 获取第一个输入gradOut的信息
148 auto selfShapePtr = context->GetInputShape(INPUT_SELF_IDX);147 auto selfShapePtr = context->GetInputShape(INPUT_SELF_IDX);
@@ -152,10 +151,9 @@ ge::graphStatus SimThreadExponentialTiling::GetInputTensorInfo()
152 for (size_t i = 0; i < selfShape.GetDimNum(); ++i) {151 for (size_t i = 0; i < selfShape.GetDimNum(); ++i) {
153 usrWorkspaceSize *= selfShape[i];152 usrWorkspaceSize *= selfShape[i];
154 }153 }
155- OP_CHECK_IF(154+ OP_CHECK_IF(usrWorkspaceSize != count,
156- usrWorkspaceSize != count,155+ OP_LOGE(nodeName, "Count %ld must be equal to the product of the elements in selfShape.", count),
157- OP_LOGE(nodeName, "Count %ld must be equal to the product of the elements in selfShape.", count),156+ return ge::GRAPH_FAILED);
158- return ge::GRAPH_FAILED);
159 usrWorkspaceSize = Ceil(usrWorkspaceSize, BATCHNUMPERHANDLE * BLOCKSIZE);157 usrWorkspaceSize = Ceil(usrWorkspaceSize, BATCHNUMPERHANDLE * BLOCKSIZE);
160 usrWorkspaceSize *= FP32_TYPESIZE;158 usrWorkspaceSize *= FP32_TYPESIZE;
161 159 
@@ -163,9 +161,9 @@ ge::graphStatus SimThreadExponentialTiling::GetInputTensorInfo()
163 OP_CHECK_NULL_WITH_CONTEXT(context, selfDesc);161 OP_CHECK_NULL_WITH_CONTEXT(context, selfDesc);
164 selfDType = selfDesc->GetDataType();162 selfDType = selfDesc->GetDataType();
165 GetDataTypeKey(selfDType);163 GetDataTypeKey(selfDType);
166- OP_CHECK_IF(164+ OP_CHECK_IF(GetDataTypeKey(selfDType) == false,
167- GetDataTypeKey(selfDType) == false,165+ OP_LOGE(nodeName, "The dtype of input self must be in [float32, float16, bfloat16]."),
168- OP_LOGE(nodeName, "The dtype of input self must be in [float32, float16, bfloat16]."), return ge::GRAPH_FAILED);166+ return ge::GRAPH_FAILED);
169 167 
170 return ge::GRAPH_SUCCESS;168 return ge::GRAPH_SUCCESS;
171}169}
@@ -195,8 +193,10 @@ ge::graphStatus SimThreadExponentialTiling::Tiling4Block()
195 offset_t_high = static_cast<uint32_t>(offset_t >> SHIFT_LEFT_32);193 offset_t_high = static_cast<uint32_t>(offset_t >> SHIFT_LEFT_32);
196 194 
197 // 分核计算195 // 分核计算
198- useCoreNum =196+ useCoreNum = static_cast<int64_t>(
199- static_cast<int64_t>(Ops::Base::CeilDiv(batchNumTotal, Ops::Base::CeilDiv(batchNumTotal, totalCoreNum)));197+ Ops::Base::CeilDiv(batchNumTotal, Ops::Base::CeilDiv(batchNumTotal, totalCoreNum)));
198+ OP_CHECK_IF(useCoreNum == 0, OP_LOGE(nodeName, "useCoreNum %u must be not equal to 0.", useCoreNum),
199+ return ge::GRAPH_FAILED);
200 // useCoreNum = static_cast<int64_t>(CeilDiv(batchNumTotal, CeilDiv(batchNumTotal, totalCoreNum)));200 // useCoreNum = static_cast<int64_t>(CeilDiv(batchNumTotal, CeilDiv(batchNumTotal, totalCoreNum)));
201 batchNumPerCore = (batchNumTotal + useCoreNum - 1) / useCoreNum;201 batchNumPerCore = (batchNumTotal + useCoreNum - 1) / useCoreNum;
202 batchNumTailCore = batchNumTotal - (useCoreNum - 1) * batchNumPerCore;202 batchNumTailCore = batchNumTotal - (useCoreNum - 1) * batchNumPerCore;
@@ -204,9 +204,8 @@ ge::graphStatus SimThreadExponentialTiling::Tiling4Block()
204 handleNumLoop = batchNumPerCore / BATCHNUMPERHANDLE;204 handleNumLoop = batchNumPerCore / BATCHNUMPERHANDLE;
205 handleNumTail = batchNumPerCore - handleNumLoop * BATCHNUMPERHANDLE;205 handleNumTail = batchNumPerCore - handleNumLoop * BATCHNUMPERHANDLE;
206 206 
207- OP_CHECK_IF(207+ OP_CHECK_IF(batchNumPerCore <= 0, OP_LOGE(nodeName, "batchNumPerCore %u must be greater than 0.", batchNumPerCore),
208- batchNumPerCore <= 0, OP_LOGE(nodeName, "batchNumPerCore %u must be greater than 0.", batchNumPerCore),208+ return ge::GRAPH_FAILED);
209- return ge::GRAPH_FAILED);
210 209 
211 return ge::GRAPH_SUCCESS;210 return ge::GRAPH_SUCCESS;
212}211}
@@ -223,10 +222,9 @@ ge::graphStatus SimThreadExponentialTiling::SetAttrParams()
223 const float* lambdaPtr = attrs->GetAttrPointer<float>(ATTR_1);222 const float* lambdaPtr = attrs->GetAttrPointer<float>(ATTR_1);
224 OP_CHECK_NULL_WITH_CONTEXT(context, lambdaPtr);223 OP_CHECK_NULL_WITH_CONTEXT(context, lambdaPtr);
225 lambda = static_cast<float>(*lambdaPtr);224 lambda = static_cast<float>(*lambdaPtr);
226- OP_CHECK_IF(225+ OP_CHECK_IF(lambda == 0,
227- lambda == 0,226+ OP_LOGE(context->GetNodeName(), "lambda is the denominator and cannot be zero, but get %f.", lambda),
228- OP_LOGE(context->GetNodeName(), "lambda is the denominator and cannot be zero, but get %f.", lambda),227+ return ge::GRAPH_FAILED);
229- return ge::GRAPH_FAILED);
230 const int64_t* seedPtr = attrs->GetAttrPointer<int64_t>(ATTR_2);228 const int64_t* seedPtr = attrs->GetAttrPointer<int64_t>(ATTR_2);
231 OP_CHECK_NULL_WITH_CONTEXT(context, seedPtr);229 OP_CHECK_NULL_WITH_CONTEXT(context, seedPtr);
232 seed = static_cast<uint64_t>(*seedPtr);230 seed = static_cast<uint64_t>(*seedPtr);
@@ -240,29 +238,22 @@ ge::graphStatus SimThreadExponentialTiling::SetAttrParams()
240 return ge::GRAPH_SUCCESS;238 return ge::GRAPH_SUCCESS;
241}239}
242 240 
243-void SimThreadExponentialTiling::SetTilingKey()241+void SimThreadExponentialTiling::SetTilingKey() { tilingKey_ = dataSizeType; }
244-{
245- tilingKey_ = dataSizeType;
246-}
247 242 
248-uint64_t SimThreadExponentialTiling::GetTilingKey()243+uint64_t SimThreadExponentialTiling::GetTilingKey() { return tilingKey_; }
249-{
250- return tilingKey_;
251-}
252 244 
253ge::graphStatus SimThreadExponentialTiling::DoTiling()245ge::graphStatus SimThreadExponentialTiling::DoTiling()
254{246{
255 auto nodeName = context->GetNodeName();247 auto nodeName = context->GetNodeName();
256 248 
257- OP_CHECK_IF(249+ OP_CHECK_IF(SetAttrParams() != ge::GRAPH_SUCCESS, OP_LOGE(nodeName, "SetAttrParams failed."),
258- SetAttrParams() != ge::GRAPH_SUCCESS, OP_LOGE(nodeName, "SetAttrParams failed."), return ge::GRAPH_FAILED);250+ return ge::GRAPH_FAILED);
259- OP_CHECK_IF(251+ OP_CHECK_IF(GetInputTensorInfo() != ge::GRAPH_SUCCESS, OP_LOGE(nodeName, "GetInputTensorInfo failed."),
260- GetInputTensorInfo() != ge::GRAPH_SUCCESS, OP_LOGE(nodeName, "GetInputTensorInfo failed."),252+ return ge::GRAPH_FAILED);
261- return ge::GRAPH_FAILED);253+ OP_CHECK_IF(GetPlatformInfo() != ge::GRAPH_SUCCESS, OP_LOGE(nodeName, "GetPlatformInfo failed."),
262- OP_CHECK_IF(254+ return ge::GRAPH_FAILED);
263- GetPlatformInfo() != ge::GRAPH_SUCCESS, OP_LOGE(nodeName, "GetPlatformInfo failed."), return ge::GRAPH_FAILED);255+ OP_CHECK_IF(Tiling4Block() != ge::GRAPH_SUCCESS, OP_LOGE(nodeName, "Tiling4Block failed."),
264- OP_CHECK_IF(256+ return ge::GRAPH_FAILED);
265- Tiling4Block() != ge::GRAPH_SUCCESS, OP_LOGE(nodeName, "Tiling4Block failed."), return ge::GRAPH_FAILED);
266 SetTilingData();257 SetTilingData();
267 258 
268 SetTilingKey();259 SetTilingKey();
@@ -275,10 +266,11 @@ ge::graphStatus SimThreadExponentialTiling::DoTiling()
275 266 
276 auto ascendcPlatform = platform_ascendc::PlatformAscendC(context->GetPlatformInfo());267 auto ascendcPlatform = platform_ascendc::PlatformAscendC(context->GetPlatformInfo());
277 size_t* currentWorkspace = context->GetWorkspaceSizes(1);268 size_t* currentWorkspace = context->GetWorkspaceSizes(1);
278- currentWorkspace[0] =269+ currentWorkspace
279- usrWorkspaceSize +270+ [0] = usrWorkspaceSize +
280- ascendcPlatform.GetLibApiWorkSpaceSize(); // 设置总的workspace的数值大小，总的workspace空间由框架来申请并管理。271+ ascendcPlatform
281- // 该workspace作为中转空间用来放T2类型数据272+ .GetLibApiWorkSpaceSize(); // 设置总的workspace的数值大小，总的workspace空间由框架来申请并管理。
273+ // 该workspace作为中转空间用来放T2类型数据
282 274 
283 return ge::GRAPH_SUCCESS;275 return ge::GRAPH_SUCCESS;
284}276}
@@ -303,4 +295,4 @@ ge::graphStatus TilingPrepare4SimThreadExponential(gert::TilingParseContext* con
303IMPL_OP_OPTILING(SimThreadExponential)295IMPL_OP_OPTILING(SimThreadExponential)
304 .Tiling(Tiling4SimThreadExponential)296 .Tiling(Tiling4SimThreadExponential)
305 .TilingParse<Tiling4SimThreadExponentialCompileInfo>(TilingPrepare4SimThreadExponential);297 .TilingParse<Tiling4SimThreadExponentialCompileInfo>(TilingPrepare4SimThreadExponential);
306-} // namespace optiling298+} // namespace optiling
Mrandom/sim_thread_exponential/op_host/arch32/sim_thread_exponential_tiling_arch32.h+17-17
@@ -86,15 +86,15 @@ private:
86 uint32_t batchNumTotal = 864;86 uint32_t batchNumTotal = 864;
87 int64_t numel = 250000;87 int64_t numel = 250000;
88 88 
89- uint32_t stepNum;89+ uint32_t stepNum = 0;
90- uint32_t stepBlock;90+ uint32_t stepBlock = 0;
91- uint32_t roundedSizeNum;91+ uint32_t roundedSizeNum = 0;
92- uint32_t roundedSizeBlock;92+ uint32_t roundedSizeBlock = 0;
93- float range;93+ float range = 0;
94- uint32_t handleNumLoop;94+ uint32_t handleNumLoop = 0;
95- uint32_t handleNumTail;95+ uint32_t handleNumTail = 0;
96- uint32_t totalCoreNum;96+ uint32_t totalCoreNum = 0;
97- uint32_t ubSize;97+ uint32_t ubSize = 0;
98 98 
99 uint64_t state = 0;99 uint64_t state = 0;
100 100 
@@ -102,15 +102,15 @@ private:
102 float end = 1;102 float end = 1;
103 103 
104 uint32_t dataSizeType = 3;104 uint32_t dataSizeType = 3;
105- int64_t count;105+ int64_t count = 0;
106- float lambda;106+ float lambda = 0;
107- uint64_t seed;107+ uint64_t seed = 0;
108- uint64_t offset;108+ uint64_t offset = 0;
109- int threadPerProcessor;109+ int threadPerProcessor = 0;
110- int streamProcessorCount;110+ int streamProcessorCount = 0;
111 111 
112- ge::DataType selfDType;112+ ge::DataType selfDType = ge::DT_UNDEFINED;
113};113};
114} // namespace optiling114} // namespace optiling
115 115 
116-#endif // OPS_BUILD_IN_OP_TILING_RUNTIME_SIM_THREAD_EXPONENTIAL_ARCH32_H_116+#endif // OPS_BUILD_IN_OP_TILING_RUNTIME_SIM_THREAD_EXPONENTIAL_ARCH32_H_