Skip to content
Snippets Groups Projects

Vit operators

Merged Houssem ROUIS requested to merge (removed):vit_operators into dev
2 unresolved threads
  • [Add] OperatorTensors with Kernel implementaiton and unit-tests
    • Erf
    • Concat
    • Gather
    • ReduceMean
    • Reshape
    • Slice
    • Transpose
  • [Softmax kernel] update to support doing the operation on the attribute axis and not on a fixed one.
Edited by Maxence Naud

Merge request reports

Loading
Loading

Activity

Filter activity
  • Approvals
  • Assignees & reviewers
  • Comments (from bots)
  • Comments (from users)
  • Commits & branches
  • Edits
  • Labels
  • Lock status
  • Mentions
  • Merge request status
  • Tracking
6 * http://www.eclipse.org/legal/epl-2.0.
7 *
8 * SPDX-License-Identifier: EPL-2.0
9 *
10 ********************************************************************************/
11
12 #ifndef AIDGE_CPU_OPERATOR_SLICEIMPL_FORWARD_KERNEL_H_
13 #define AIDGE_CPU_OPERATOR_SLICEIMPL_FORWARD_KERNEL_H_
14
15 #include "aidge/utils/Registrar.hpp"
16
17 #include "aidge/backend/cpu/operator/SliceImpl.hpp"
18
19 namespace Aidge {
20 template <class I, class O>
21 void SliceImpl_cpu_forward_kernel(const std::vector<DimSize_t> inputDims,
  • Houssem ROUIS added 1 commit

    added 1 commit

    Compare with previous version

  • Houssem ROUIS added 2 commits

    added 2 commits

    Compare with previous version

  • Houssem ROUIS added 2 commits

    added 2 commits

    Compare with previous version

  • Houssem ROUIS added 4 commits

    added 4 commits

    Compare with previous version

  • Houssem ROUIS added 1 commit

    added 1 commit

    Compare with previous version

  • Houssem ROUIS mentioned in epic &2

    mentioned in epic &2

  • Houssem ROUIS added 4 commits

    added 4 commits

    • 3b6e6292 - fix python binding of concat by adding nb_in attr
    • 7717ad3c - minor code cleanings
    • fc2e005e - fix Slice outputDims and add test
    • 8dfe6ba6 - add more Registrar dims for Transpose

    Compare with previous version

  • Houssem ROUIS added 2 commits

    added 2 commits

    Compare with previous version

  • Houssem ROUIS marked this merge request as ready

    marked this merge request as ready

  • Cyril Moineau requested review from @pineapple

    requested review from @pineapple

  • Houssem ROUIS changed the description

    changed the description

  • Houssem ROUIS added 25 commits

    added 25 commits

    • bfdf7407...fafa52d8 - 24 commits from branch eclipse/aidge:master
    • 4260ba9f - Merge branch 'master' of gitlab.eclipse.org:hrouis/aidge_backend_cpu into vit_operators

    Compare with previous version

  • Houssem ROUIS added 1 commit

    added 1 commit

    Compare with previous version

  • assigned to @hrouis

  • Houssem ROUIS added 7 commits

    added 7 commits

    Compare with previous version

  • Houssem ROUIS added 4 commits

    added 4 commits

    Compare with previous version

  • Houssem ROUIS added 3 commits

    added 3 commits

    Compare with previous version

  • Houssem ROUIS added 2 commits

    added 2 commits

    • 42445cbd - switch shape input to attr for Reshape
    • b95fdae5 - change Slice attr to int64

    Compare with previous version

  • Houssem ROUIS added 1 commit

    added 1 commit

    Compare with previous version

  • Maxence Naud changed target branch from master to tiling

    changed target branch from master to tiling

    • I think Reshape Operator should not copy any memory! We could in forward() use setRawPtr() on output's Tensor to input's Tensor rawPtr.

    • I agree, there is no need to copy data here. However, such function doesn't need to be called each time in forward(). Setting the output Tensor's implementation pointer to the input's implementation pointer could be done in forwardDims() (hence in computeOutputDims()). At this stage, the implementation is already set as setBackend() has been previously called.

    • Or maybe we should not mix Operator's functions and OperatorImpl's functions?

      Edited by Maxence Naud
    • Author Developer

      I agree, but when I added Reshape, I saw that for all other operators Input ptr and Output ptr are different so I thought it would be dangerous to not copy (maybe there will be a double free memory or another problem)

    • This discussion has been transfered to the issue #4 (closed) as it more wildly talks about in-place operations.

    • Please register or sign in to reply
  • Maxence Naud changed target branch from tiling to master

    changed target branch from tiling to master

  • Houssem ROUIS added 38 commits

    added 38 commits

    Compare with previous version

  • Houssem ROUIS added 3 commits

    added 3 commits

    Compare with previous version

  • Houssem ROUIS added 1 commit

    added 1 commit

    Compare with previous version

  • Houssem ROUIS added 3 commits

    added 3 commits

    Compare with previous version

  • Houssem ROUIS added 1 commit

    added 1 commit

    Compare with previous version

  • Houssem ROUIS added 1 commit

    added 1 commit

    • 03a58652 - change attrs back to int64_t

    Compare with previous version

  • Maxence Naud added StatusReview Ready label and removed StatusWIP label

    added StatusReview Ready label and removed StatusWIP label

  • Maxence Naud added StatusWIP label and removed StatusReview Ready label

    added StatusWIP label and removed StatusReview Ready label

  • Maxence Naud changed the description

    changed the description

  • Maxence Naud changed target branch from master to dev

    changed target branch from master to dev

  • Maxence Naud added 15 commits

    added 15 commits

    Compare with previous version

  • Maxence Naud enabled an automatic merge when the pipeline for 7215ade7 succeeds

    enabled an automatic merge when the pipeline for 7215ade7 succeeds

  • Maxence Naud mentioned in issue #4 (closed)

    mentioned in issue #4 (closed)

  • Maxence Naud approved this merge request

    approved this merge request

  • Maxence Naud added 3 commits

    added 3 commits

    • 8d7f0c64 - [WIP][NF] Start fixing ReduceMean forward kernel memory leak
    • d87663c3 - Add tag to ReduceMean test
    • 0c6edb52 - [Fix] ReduceMean operator forward kernel with refactor

    Compare with previous version

  • removed IssueBug 🐛 label

  • Maxence Naud added 1 commit

    added 1 commit

    Compare with previous version

  • Maxence Naud added StatusReview Ready label and removed StatusWIP label

    added StatusReview Ready label and removed StatusWIP label

  • Maxence Naud approved this merge request

    approved this merge request

  • Maxence Naud added 1 commit

    added 1 commit

    Compare with previous version

  • Maxence Naud mentioned in commit 8521b218

    mentioned in commit 8521b218

  • merged

  • Maxence Naud mentioned in merge request !50 (merged)

    mentioned in merge request !50 (merged)

  • Please register or sign in to reply
    Loading