master

cann-robot支持Ascend950 mte atomic

e9f6cb63创建于 2 天前历史提交

文件	最后提交记录	最后更新时间
shmem_device_mte.h	支持Ascend950 mte atomic Co-authored-by: vector5<caobingjie@huawei.com> Co-authored-by: zhangyunqi<zhangyunqi5@huawei.com> Co-authored-by: QK_25415<zhuzhiming1@huawei.com> # message auto-generated for no-merge-commit merge: !283 merge mteatomic into master 支持Ascend950 mte atomic Created-by: vector5 Commit-by: zhangyunqi;vector5;QK_25415 Merged-by: cann-robot Description: ## 描述 <!--在这里详细描述你的改动，包括改动的原因和所采取的方法。--> 支持Ascend950 mte atomic的以下接口和UT T aclshmemx_mte_atomic_fetch(gm T *src, int32_t pe); void aclshmemx_mte_atomic_set(gm T *dst, T value, int32_t pe); T aclshmemx_mte_atomic_compare_swap(gm T *dst, T cond, T value, int32_t pe); T aclshmemx_mte_atomic_swap(gm T *dst, T value, int32_t pe); void aclshmemx_mte_atomic_inc(gm T *dst, int32_t pe); void aclshmemx_mte_atomic_add(gm T *dst, T value, int32_t pe); T aclshmemx_mte_atomic_fetch_inc(gm T *dst, int32_t pe); T aclshmemx_mte_atomic_fetch_add(gm T *dst, T value, int32_t pe); ## 关联的Issue <!-- 如果这个PR是为了解决特定的Issue，请在这里提供Issue链接。例如：关联Issue #123--> <!-- 如果这个PR是为了解决特定的问题单，请在这里描述问题单单号。--> https://gitcode.com/cann/shmem/issues/220 ## 测试 <!--描述进行了哪些测试来验证你的改动。包括但不限于二级冒烟、算子泛化等。--> A2验证ShmemAtomic  MteAtomic  A5验证ShmemAtomic  MteAtomic  ## 类型标签 <!-- [x] 表示选中 --> - [ ] Bug修复 - [x] 新特性 - [ ] 性能优化 - [ ] 文档更新 - [ ] 其他，请描述： See merge request: cann/shmem!283	2 天前
shmem_device_rdma.h	feat: 新增 rdma atomic 语义的 logic 操作 Co-authored-by: simonkorl28<machuan6@huawei.com> # message auto-generated for no-merge-commit merge: !326 merge xscale-atomic-4 into master feat: 新增 rdma atomic 语义的 logic 操作 Created-by: simonkorl28 Commit-by: simonkorl28 Merged-by: cann-robot Description: ## 描述 <!--在这里详细描述你的改动，包括改动的原因和所采取的方法。--> 1. 新增 rdma atomic 语义的 logic 操作相关低阶接口（逻辑运算，如 and, or, xor） 2. 新增 rdma atomic 语义的 logic 操作相关低阶接口对应 UT ## 关联的Issue <!-- 如果这个PR是为了解决特定的Issue，请在这里提供Issue链接。例如：关联Issue #123--> <!-- 如果这个PR是为了解决特定的问题单，请在这里描述问题单单号。--> [#234](https://gitcode.com/cann/shmem/issues/234) ## 测试 <!--描述进行了哪些测试来验证你的改动。包括但不限于二级冒烟、算子泛化等。--> 新增的 UT 运行通过  原始 RDMA UT 运行通过  与云脉网卡相关的原子语义测试 UT 测试通过  ## 文档更新 <!--如果这个PR包含文档的更新，请在这里指出。例如：更新了README.md文件。--> ## 类型标签 <!-- [x] 表示选中 --> - [ ] Bug修复 - [x] 新特性 - [ ] 性能优化 - [ ] 文档更新 - [ ] 其他，请描述： See merge request: cann/shmem!326	16 天前
shmem_device_sdma.h	适配aicpu，取消40个AIV限制 Co-authored-by: jiang-xinyu3<jiangxinyu3@hisilicon.com> # message auto-generated for no-merge-commit merge: !147 merge query_size into master 适配aicpu，取消40个AIV限制 Created-by: jiangxinyu3 Commit-by: jiangxinyu3;jiang-xinyu3 Merged-by: cann-robot Description: ## 描述 <!--在这里详细描述你的改动，包括改动的原因和所采取的方法。--> ## 关联的Issue <!-- 如果这个PR是为了解决特定的Issue，请在这里提供Issue链接。例如：关联Issue #123--> <!-- 如果这个PR是为了解决特定的问题单，请在这里描述问题单单号。--> ## 测试 <!--描述进行了哪些测试来验证你的改动。包括但不限于二级冒烟、算子泛化等。--> /home/jxy/shmem_jxy-query_size/examples/sdma 3467128 background process recorded 3467129 background process recorded 3467131 background process recorded 3467132 background process recorded 3467133 background process recorded 3467134 background process recorded 3467135 background process recorded 3467136 background process recorded Pe 7Finised !! Result Correct !! Pe 5Finised !! Result Correct !! Pe 3Finised !! Result Correct !! Pe 2Finised !! Result Correct !! Pe 6Finised !! Result Correct !! Pe 1Finised !! Result Correct !! Pe 4Finised !! Result Correct !! Pe 0Finised !! Result Correct !! [SUCCESS] demo run success in pe 1 [SUCCESS] demo run success in pe 5 [SUCCESS] demo run success in pe 4 [SUCCESS] demo run success in pe 7 [SUCCESS] demo run success in pe 3 [SUCCESS] demo run success in pe 2 [SUCCESS] demo run success in pe 6 [SUCCESS] demo run success in pe 0 wait process 3467128 done wait process 3467129 done wait process 3467131 done wait process 3467132 done wait process 3467133 done wait process 3467134 done wait process 3467135 done wait process 3467136 done root@hisi-hangzhou-ip69:notifywait$ rm -rf logs/* && bash run.sh -pes 8 -type int 3544418 background process recorded 3544419 background process recorded 3544420 background process recorded 3544421 background process recorded 3544422 background process recorded 3544423 background process recorded 3544424 background process recorded 3544425 background process recorded Pe 3 AllGather result in ptr_A after notify_wait: Pe 2 AllGather result in ptr_A after notify_wait: Pe 0 AllGather result in ptr_A after notify_wait: Pe 5 AllGather result in ptr_A after notify_wait: Pe 7 AllGather result in ptr_A after notify_wait: Pe 6 AllGather result in ptr_A after notify_wait: Pe 1 AllGather result in ptr_A after notify_wait: Pe 4 AllGather result in ptr_A after notify_wait: Pe 2 has 0 unexpected values. Pe 3 has 0 unexpected values. Pe 0 has 0 unexpected values. Pe 5 has 0 unexpected values. Pe 7 has 0 unexpected values. Pe 6 has 0 unexpected values. Pe 1 has 0 unexpected values. Pe 4 has 0 unexpected values. [SUCCESS] demo run success in pe 7 [SUCCESS] demo run success in pe 4 [SUCCESS] demo run success in pe 5 [SUCCESS] demo run success in pe 3 [SUCCESS] demo run success in pe 2 [SUCCESS] demo run success in pe 1 [SUCCESS] demo run success in pe 6 [SUCCESS] demo run success in pe 0 wait process 3544418 done wait process 3544419 done wait process 3544420 done wait process 3544421 done wait process 3544422 done wait process 3544423 done wait process 3544424 done wait process 3544425 done t@hisi-hangzhou-ip69:cmo$ bash run.sh -pes 8 -type int 3549967 background process recorded 3549968 background process recorded 3549969 background process recorded 3549970 background process recorded 3549971 background process recorded 3549972 background process recorded 3549973 background process recorded 3549974 background process recorded PE 6 Finised ! PE 4 Finised ! PE 0 Finised ! PE 2 Finised ! PE 1 Finised ! PE 5 Finised ! PE 3 Finised ! PE 7 Finised ! [SUCCESS] demo run success in pe 4 [SUCCESS] demo run success in pe 7 [SUCCESS] demo run success in pe 6 [SUCCESS] demo run success in pe 3 [SUCCESS] demo run success in pe 2 [SUCCESS] demo run success in pe 1 [SUCCESS] demo run success in pe 0 [SUCCESS] demo run success in pe 5 wait process 3549967 done wait process 3549968 done wait process 3549969 done wait process 3549970 done wait process 3549971 done wait process 3549972 done wait process 3549973 done wait process 3549974 done ## 文档更新 <!--如果这个PR包含文档的更新，请在这里指出。例如：更新了README.md文件。--> ## 类型标签 <!-- [x] 表示选中 --> - [x] Bug修复 - [ ] 新特性 - [ ] 性能优化 - [ ] 文档更新 - [ ] 其他，请描述： See merge request: cann/shmem!147	2 个月前
shmem_device_udma.h	UDMA场景补齐AMO接口 Co-authored-by: YeZZzzz1<yezhenni1@huawei.com> # message auto-generated for no-merge-commit merge: !284 merge master into master UDMA场景补齐AMO接口 Created-by: YeZZzzz1 Commit-by: YeZZzzz1 Merged-by: cann-robot Description: ## 描述 UMDA场景补齐AMO接口，包括：fetch, set, swap, fetch inc, inc, fetch and, and, fetch or, or, fetch xor, xor ## 关联的Issue https://gitcode.com/cann/shmem/issues/199 https://gitcode.com/cann/shmem/issues/205 ## 测试 examples： put  put signal   FAA  CAS  fetch  set  swap  fetch inc  fetch and  fetch or  fetch xor  uttests：         ## 文档更新 <!--如果这个PR包含文档的更新，请在这里指出。例如：更新了README.md文件。--> ## 类型标签 <!-- [x] 表示选中 --> - [ ] Bug修复 - [x] 新特性 - [ ] 性能优化 - [ ] 文档更新 - [ ] 其他，请描述： See merge request: cann/shmem!284	1 个月前