Fix template compilation and warnings
Context
Compilation error when building with CUDA 12.6 and GCC 14.2
Modified files
include/aidge/backend/cuda/data/TensorImpl.hppsrc/data/TensorImpl.cusrc/operator/ILayerNormImpl_CUDA_kernels.cu
Detailed major modifications
- Moved enable_if template specialization from parameter to return type which helps the compiler to select the correct implementation.
- Changed factor type to avoid out of bound warning.
Merge request reports
Activity
added 126 commits
-
cd8c0d29...5d79b287 - 124 commits from branch
eclipse/aidge:dev - 81ee1832 - fix: TensorImpl: Enable if on return type instead of parameter
- 096db4c3 - edit: ILayerNormImpl: fixed type possible overflow warning
-
cd8c0d29...5d79b287 - 124 commits from branch
assigned to @silvanosky
requested review from @pineapple
added Enhancement ⭐ label
changed milestone to %aidge v0.5.0
added StatusReview Ready label
added 1 commit
- 1cadf046 - Update TensorImpl.hpp add include for type_traits
reset approvals from @pineapple by pushing to the branch
@silvanosky Could you rebase?
added 28 commits
-
07ab66a2...24446f6c - 24 commits from branch
eclipse/aidge:dev - 0b8c74f3 - fix: TensorImpl: Enable if on return type instead of parameter
- 718a916b - edit: ILayerNormImpl: fixed type possible overflow warning
- 505923e3 - Update TensorImpl.hpp add include for type_traits
- cc9be0bc - Update TensorImpl.cu
Toggle commit list-
07ab66a2...24446f6c - 24 commits from branch
99 101 return thrust::equal(thrustData, thrustData + mNbElts, thrustOtherData); 100 102 } 101 103 104 template void Aidge::thrust_copy<half_float::half, double>(half_float::half const*, double*, unsigned long); changed this line in version 9 of the diff
added 1 commit
- b08a6b1d - edit: TensorImpl: problem with half type conversion, comment for now
added 1 commit
- 3a82949e - edit: TensorImpl.{hpp,cu}: Fixed template in the general case
@silvanosky The pipeline is still failing, are you still on this? Can we help?
The pipeline keep failing using Clang, I don't know exactly why. It seems that the enable if condition don't have the same behavior as GCC hence symbol are undefined. CUDA seems to be compatible with GCC only so if you have a CLANG/CUDA expert to fix it.
I can't focus on this right now, I will have time in march to investigate deeper.
added 20 commits
-
e1452c59...ed7ea278 - 13 commits from branch
eclipse/aidge:dev - 1084f489 - fix: TensorImpl: Enable if on return type instead of parameter
- b2d4a2c1 - edit: ILayerNormImpl: fixed type possible overflow warning
- 741f3ea1 - Update TensorImpl.hpp add include for type_traits
- 8080e6fc - Update TensorImpl.cu
- 63926c14 - edit: TensorImpl: problem with half type conversion, comment for now
- 785d69c6 - edit: TensorImpl.{hpp,cu}: Fixed template in the general case
- 320db778 - edit: ILayerNorm: fixed test
Toggle commit list-
e1452c59...ed7ea278 - 13 commits from branch
added 19 commits
-
320db778...c4f53b0d - 13 commits from branch
eclipse/aidge:dev - 1ba7bbd3 - fix: TensorImpl: Enable if on return type instead of parameter
- f4721189 - edit: ILayerNormImpl: fixed type possible overflow warning
- a3c832f9 - Update TensorImpl.hpp add include for type_traits
- af3da233 - edit: TensorImpl.{hpp,cu}: Fixed template in the general case
- 2d392d87 - edit: ILayerNorm: fixed test
- 734cbe41 - upgrade: pybind: 2.13.6
Toggle commit list-
320db778...c4f53b0d - 13 commits from branch
reset approvals from @pineapple by pushing to the branch
added Stuck ❌ label
removed StatusReview Ready label
added 8 commits
-
734cbe41...ec169671 - 2 commits from branch
eclipse/aidge:dev - 43ee3a7a - fix: TensorImpl: Enable if on return type instead of parameter
- ad677378 - edit: ILayerNormImpl: fixed type possible overflow warning
- 8d8656b7 - Update TensorImpl.hpp add include for type_traits
- 1ce87625 - edit: TensorImpl.{hpp,cu}: Fixed template in the general case
- 2e74a4aa - edit: ILayerNorm: fixed test
- 8ebe2829 - upgrade: pybind: 2.13.6
Toggle commit list-
734cbe41...ec169671 - 2 commits from branch
added 1 commit
- f5ecf069 - edit: TensorImpl: gave up original approach to use class partial
added 1 commit
- bfb6e034 - edit: TensorImpl: gave up original approach to use class partial
changed milestone to %aidge v0.6.0
removed Stuck ❌ label
requested review from @pineapple
added 8 commits
-
00ffea41 - 1 commit from branch
eclipse/aidge:dev - 7d810245 - fix: TensorImpl: Enable if on return type instead of parameter
- dd3bd11d - edit: ILayerNormImpl: fixed type possible overflow warning
- a7164012 - Update TensorImpl.hpp add include for type_traits
- 27a9451c - edit: TensorImpl.{hpp,cu}: Fixed template in the general case
- 7b0be0af - edit: ILayerNorm: fixed test
- 931b62e2 - edit: TensorImpl: try to fix template ?
- 51450d3e - edit: TensorImpl: gave up original approach to use class partial
Toggle commit list-
00ffea41 - 1 commit from branch
added Stuck ❌ label
removed Stuck ❌ label
Wait @silvanosky is this review ready?
added StatusReview Ready label
added 11 commits
-
51450d3e...f5dbfb26 - 4 commits from branch
eclipse/aidge:dev - 1992ac67 - fix: TensorImpl: Enable if on return type instead of parameter
- 943081f8 - edit: ILayerNormImpl: fixed type possible overflow warning
- 6c3aa17b - Update TensorImpl.hpp add include for type_traits
- f80c3d59 - edit: TensorImpl.{hpp,cu}: Fixed template in the general case
- cf94ef7a - edit: ILayerNorm: fixed test
- 29181ce4 - edit: TensorImpl: try to fix template ?
- 1562648a - edit: TensorImpl: gave up original approach to use class partial
Toggle commit list-
51450d3e...f5dbfb26 - 4 commits from branch
