已合并
修复Tril/Triu在corner case下的精度问题 #2704
Wangdongxu_mj007创建于 5月12日
修复Tril/Triu在corner case下的精度问题 #2704
已合并
共 1 个文件变更+1-1
| @@ -231,7 +231,7 @@ __aicore__ inline void SplitRowAndCol<T, IS_LOWER>::DoCopyVector(NormalBlockInfo | |||
| 231 | int64_t tailRows = info.tailRows > 0 ? info.tailRows : 0; | 231 | int64_t tailRows = info.tailRows > 0 ? info.tailRows : 0; |
| 232 | int64_t midRows = info.rowActual - tailRows - headRows; | 232 | int64_t midRows = info.rowActual - tailRows - headRows; |
| 233 | midRows = midRows > 0 ? midRows : 0; | 233 | midRows = midRows > 0 ? midRows : 0; |
| 234 | - uint16_t vfInnerLoops = CeilDiv(static_cast<uint32_t>(info.colActual), regCount_); | 234 | + uint16_t vfInnerLoops = CeilDiv(static_cast<uint32_t>(colInner_), regCount_); |
| 235 | uint16_t headLoops = headRows * vfInnerLoops; | 235 | uint16_t headLoops = headRows * vfInnerLoops; |
| 236 | uint16_t tailLoops = tailRows * vfInnerLoops; | 236 | uint16_t tailLoops = tailRows * vfInnerLoops; |
| 237 | uint16_t midRowsU16 = static_cast<uint16_t>(midRows); | 237 | uint16_t midRowsU16 = static_cast<uint16_t>(midRows); |