1. 20 1月, 2018 2 次提交
  2. 19 1月, 2018 5 次提交
  3. 17 1月, 2018 7 次提交
    • W
      Blob::col_id · 1f7765e0
      willzhang4a58 提交于
      1f7765e0
    • W
      add set_col_id for blob · 271de42e
      willzhang4a58 提交于
      271de42e
    • D
      change dataloader to move col_id and max_col_id to register (#520) · 2e47a8ee
      Daniel Sun 提交于
      * change dataloader to move col_id and max_col_id to register
      
      * some miner changes
      2e47a8ee
    • W
      set default col_id to make cnn runable · f27ace65
      willzhang4a58 提交于
      f27ace65
    • C
      refine fw(recurrent/normal) comp task node (#511) · 5970b3d3
      chengtbf 提交于
      * fw task node for recurrent out edge
      
      * recurrent task node 2 out regst
      
      * set min regst num of rec_ht_regst
      
      * todo is recurrent out edge
      
      * refine name
      
      * implement is recurrent out edge for comp task node
      
      * reduce code for last impl
      
      * fix for sub class fuc
      
      * change bind out regst in fw task node by SunPeiWen
      
      * fix for review
      
      * unexpected_run for virtual func
      
      * fix bug by zyeric fault
      
      * change interface: delete isrecurrentoutedge add get succ/pred chain node on edge
      
      * rec_ht
      
      * fix name
      
      * fix name
      
      * refine fw(recurrent/nomal) compute task node
      
      * fix bug of bind in regst
      
      * delete node
      
      * fix for review
      
      * abstract new func for succ/pred chain node on edge
      
      * fix max regst in model parallel
      
      * fix back
      5970b3d3
    • N
      feat: update accumulate actor (#509) · 2e100ea6
      Niu Chong 提交于
      * feat: update accumulate actor
      
      * fix: fix comment and add enum class ColIdOrder
      
      * refactor: rename ColIdOrder::UnSet to ColIdOrder::UnCertain
      
      * chore: rename IsFirstRegstInPieceOfThisOrder and move it out of Actor Class
      
      * fix: initialize col_id/max_col_id of blobheader as 0 instead of -1
      2e100ea6
    • W
      use read_id instead of read_done_id · 65258e6c
      willzhang4a58 提交于
      65258e6c
  4. 16 1月, 2018 6 次提交
    • W
      move col_id from blob to register · cad45673
      willzhang4a58 提交于
      cad45673
    • C
      refine recurrent op (#518) · 988f5d9b
      chengtbf 提交于
      * refine recurrent op
      
      * ht -> rec_ht , out-> ht
      
      * notes
      
      * refine recurretn op conf
      
      * sigmoid
      
      * for blob header and col num
      988f5d9b
    • W
      IsLastCol -> IsMaxCol · 2fcd8431
      willzhang4a58 提交于
      2fcd8431
    • W
      fix: has_header_field · cf20a9f8
      willzhang4a58 提交于
      cf20a9f8
    • W
      fix: error shape of packed_blob_desc · 29d76763
      willzhang4a58 提交于
      29d76763
    • D
      rewrite BasicDataLoader based on dev_nc_blob (#496) · d51a32e1
      Daniel Sun 提交于
      * change dataloader
      
      * make basic_data_loader_kernel.cpp readable
      
      * debug
      
      * debug dataloader
      
      * change something due to review
      
      * do some changes
      
      * refactor to reduce code
      
      * remove comment
      
      * refactor the dataloader kernel
      
      * debug
      
      * debug a runtime bug
      
      * remove fillblobwithzero method
      
      * changes based on reviews
      
      * debug and make it runnable
      
      * change the declarations due to zyeric's change request
      
      * changes due to willzhang's change request
      
      * make it able to compile
      
      * change the type of blob header and move the position of DataLoadStatus
      
      * change the declaration of function ReadColFromBufferToOutBlob()
      
      * change the wrong use of memcpy() and memset()
      
      * add 0 after the end of each line
      
      * refactor ReadOneDataContent() function
      
      * revert a wrong change
      
      * remove unnecessary CHECK()
      
      * rename and move some declaration
      
      * bug for compiles
      d51a32e1
  5. 15 1月, 2018 2 次提交
  6. 14 1月, 2018 2 次提交
    • Y
      fix bugs && make runnable (#514) · c6789e54
      Yi Zhu 提交于
      c6789e54
    • duduscript's avatar
      Bug: Gassian -> Normal (#512) · 8815ce16
      duduscript 提交于
      * add l1 l2 regularization
      
      * fix some problem
      
      * Fix bug
      
      * fix op API
      
      * Fix a bug in op
      
      * add l1l2
      
      * remove regularization_diff_blob
      
      * remove sign in kernel_util
      
      * fix naming problem and add l1l2 function
      
      * simplify code
      
      * remove else
      
      * update modelupdate API
      
      * Fix bug
      
      * ~
      
      * fix bug
      8815ce16
  7. 12 1月, 2018 5 次提交
  8. 10 1月, 2018 5 次提交
    • Y
      Add IBVerbsCommNet to System & Fix Bug of Initializing Model on GPU (#502) · a57877c4
      Yi Zhu 提交于
      * branch init
      
      * init rdma_comm_network interface
      
      * Rdma rebuild (#369)
      
      * rebuild rdma main class
      
      * remove linux dir
      
      * fix bugs
      
      * Rdma rebuild (#370)
      
      * rebuild rdma main class
      
      * remove linux dir
      
      * add connection rpc
      
      * Post API params change to const ptr
      
      * fix compile bugs
      
      * implement read recv send function
      
      * updt connection interface & implement poll queue
      
      * fix bugs & updat interface
      
      * implement RdmaCommNet functions
      
      * Rdma rebuild (#371)
      
      * rebuild rdma main class
      
      * remove linux dir
      
      * add connection rpc
      
      * Post API params change to const ptr
      
      * Add MemDesc rpc and registerdone
      
      * Add clear rpc
      
      * Useless file
      
      * Rdma rebuild (#372)
      
      * rebuild rdma main class
      
      * remove linux dir
      
      * add connection rpc
      
      * Post API params change to const ptr
      
      * Add MemDesc rpc and registerdone
      
      * Add clear rpc
      
      * remove connection_pool and change API for MenDesc
      
      * remove connection pool file
      
      * remove .file
      
      * rebuild connection to endpoint manager
      
      * remove GetConnection
      
      * fix constructor and destructor fir rdmacommnet
      
      * fix bugs
      
      * fix deconstructor
      
      * Rdma rebuild (#373)
      
      * rebuild rdma main class
      
      * remove linux dir
      
      * add connection rpc
      
      * Post API params change to const ptr
      
      * Add MemDesc rpc and registerdone
      
      * Add clear rpc
      
      * remove connection_pool and change API for MenDesc
      
      * remove connection pool file
      
      * remove .file
      
      * rebuild connection to endpoint manager
      
      * remove GetConnection
      
      * Add RDMA to JOBCONF
      
      * Rdma rebuild (#374)
      
      * rebuild rdma main class
      
      * remove linux dir
      
      * add connection rpc
      
      * Post API params change to const ptr
      
      * Add MemDesc rpc and registerdone
      
      * Add clear rpc
      
      * remove connection_pool and change API for MenDesc
      
      * remove connection pool file
      
      * remove .file
      
      * rebuild connection to endpoint manager
      
      * remove GetConnection
      
      * Add RDMA to JOBCONF
      
      * fix RdmaInit bug
      
      * add log
      
      * Rdma rebuild (#375)
      
      * rebuild rdma main class
      
      * remove linux dir
      
      * add connection rpc
      
      * Post API params change to const ptr
      
      * Add MemDesc rpc and registerdone
      
      * Add clear rpc
      
      * remove connection_pool and change API for MenDesc
      
      * remove connection pool file
      
      * remove .file
      
      * rebuild connection to endpoint manager
      
      * remove GetConnection
      
      * Add RDMA to JOBCONF
      
      * fix RdmaInit bug
      
      * add log
      
      * Add Log
      
      * fix bug
      
      * fix bugs
      
      * Rdma rebuild (#376)
      
      * rebuild rdma main class
      
      * remove linux dir
      
      * add connection rpc
      
      * Post API params change to const ptr
      
      * Add MemDesc rpc and registerdone
      
      * Add clear rpc
      
      * remove connection_pool and change API for MenDesc
      
      * remove connection pool file
      
      * remove .file
      
      * rebuild connection to endpoint manager
      
      * remove GetConnection
      
      * Add RDMA to JOBCONF
      
      * fix RdmaInit bug
      
      * add log
      
      * Add Log
      
      * commnet add establish network
      
      * add qpn
      
      * fix bug
      
      * create conn_info and qp for each conn
      
      * Add commit and destruct for endpointmanager
      
      * Add proto use_rdma
      
      * fix bug
      
      * Delete redundent file
      
      * Dev rdma yx (#377)
      
      * rebuild rdma main class
      
      * remove linux dir
      
      * add connection rpc
      
      * Post API params change to const ptr
      
      * Add MemDesc rpc and registerdone
      
      * Add clear rpc
      
      * remove connection_pool and change API for MenDesc
      
      * remove connection pool file
      
      * remove .file
      
      * rebuild connection to endpoint manager
      
      * remove GetConnection
      
      * Add RDMA to JOBCONF
      
      * fix RdmaInit bug
      
      * add log
      
      * Add Log
      
      * commnet add establish network
      
      * add qpn
      
      * fix bug
      
      * create conn_info and qp for each conn
      
      * Add commit and destruct for endpointmanager
      
      * Add proto use_rdma
      
      * fix rpc bug
      
      * fix some problem
      
      * int64 and uint64
      
      * Dev rdma yx (#378)
      
      * rebuild rdma main class
      
      * remove linux dir
      
      * add connection rpc
      
      * Post API params change to const ptr
      
      * Add MemDesc rpc and registerdone
      
      * Add clear rpc
      
      * remove connection_pool and change API for MenDesc
      
      * remove connection pool file
      
      * remove .file
      
      * rebuild connection to endpoint manager
      
      * remove GetConnection
      
      * Add RDMA to JOBCONF
      
      * fix RdmaInit bug
      
      * add log
      
      * Add Log
      
      * commnet add establish network
      
      * add qpn
      
      * fix bug
      
      * create conn_info and qp for each conn
      
      * Add commit and destruct for endpointmanager
      
      * Add proto use_rdma
      
      * fix rpc bug
      
      * fix some problem
      
      * int64 and uint64
      
      * Fix bug
      
      * remove useless head file
      
      * fix bug
      
      * fix bug
      
      * fix bug
      
      * updt interface
      
      * change conninfo rpc (#379)
      
      * change conninfo rpc
      
      * swap the place of var declare
      
      * fix zlib dependency
      
      * remove useless function (#381)
      
      * updt recv buffer size
      
      * updt proto & sh for multi-machines
      
      * remove useless LOG
      
      * updt cmake
      
      * remove soft link file
      
      * rearrage code
      
      * rdma proto & updt to ctrl
      
      * merge & fix bugs
      
      * updt to KV
      
      * add debug info
      
      * fix bug
      
      * remove redudant code
      
      * remove redundant code
      
      * recover blank line
      
      * updt interfaces & fix bug
      
      * updt proto
      
      * updt proto
      
      * updt proto
      
      * merge master
      
      * add macro guardian
      
      * fix include
      
      * fix rdma_memory
      
      * add RdmaMem
      
      * fix code
      
      * add endpoint_manager & rdma_comm_network
      
      * remove EstablishNetwork
      
      * updt resource port
      
      * refine SendMsg
      
      * refine connection interface
      
      * refine code
      
      * refine code
      
      * remove redundant changes
      
      * fix macro guardian bug
      
      * refine endpoint_manager && ibverbs_comm_net
      
      * merge master
      
      * refine code
      
      * replace ptr with normal var
      
      * fix bugs
      
      * remove useless log
      
      * rollback changes to job.prototxt
      
      * fix proto
      
      * refine code
      
      * updt to LOG(FATAL)
      a57877c4
    • N
      fix(register/blob.cpp): fix typo (#503) · 3facafa3
      Niu Chong 提交于
      3facafa3
    • Y
      IBVerbsCommNet and EndpointManager (#479) · 9072ef48
      Yi Zhu 提交于
      * add endpoint_manager & rdma_comm_network
      
      * refine code
      
      * refine code
      
      * fix macro guardian bug
      
      * refine endpoint_manager && ibverbs_comm_net
      
      * merge master
      
      * refine code
      
      * replace ptr with normal var
      9072ef48
    • Y
      Refine interface in CommNet (#499) · 39cc7788
      Yi Zhu 提交于
      * fix bug
      
      * updt interface
      
      * refine code
      39cc7788
    • N
      feat: add col_num(that is sequence length) in Blob and BlobDesc (#494) · 0719c5a8
      Niu Chong 提交于
      * feat: add seq_len in Blob and BlobDesc
      
      * fix: rename PieceStatus to BlobHeader and insert it to mem_ptr of Blob
      
      * fix: fix the comments
      
      * fix: fix typos and remove  vim .swp file
      
      * fix: fix the comment
      
      * fix: update max_col_num() to col_num() in Register
      
      * fix: fix due to the comment
      0719c5a8
  9. 09 1月, 2018 4 次提交
  10. 08 1月, 2018 1 次提交
  11. 05 1月, 2018 1 次提交