Reduced CUDA dropout memory usage #1005
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…n number." This reverts commit 9a29b52.
Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com>
FedyuninV
requested changes
Dec 18, 2023
| ForwardRate(1.f - rate), | ||
| IsSpatial(isSpatial), | ||
| IsBatchwise(isBatchwise), | ||
| seed(seed) {}; |
There was a problem hiding this comment.
minor codestyle
No need for ;. The empty body should be located on a new line.
But it's your choice whether to write it like
seed(seed)
{}
or
seed(seed)
{
}
| @@ -35,4 +35,19 @@ struct CMathEngineDropoutDesc : public CDropoutDesc { | |||
| CFloatHandleVar Mask; | |||
| }; | |||
|
|
|||
| // DropoutDesc without mask for usage on CUDA | |||
| struct CCudaMathEngineDropoutDesc : public CDropoutDesc { | |||
There was a problem hiding this comment.
This class is for CUDA only. It's better to move somewhere to NeoMathEngine/src/GPU/CUDA/*.
E.g. in front of its' ctor definition in CudaMathEngineDnnDropout.cu
| const unsigned int threshold = forwardRate * UINT_MAX; | ||
| int row; | ||
| int col; | ||
| if (GetCudaTaskIndex2D( firstHeight, ( firstWidth + 3 ) / 4, row, col ) ) { |
There was a problem hiding this comment.
minor codestyle
if and ( must be together.
Spaces inside brackets must be used in the similar approach on both side. But you have if (GetCuda.. on one side with col ) ) on the other.
| int obj; | ||
| int row; | ||
| int col; | ||
| if ( GetCudaTaskIndex3D(inputObjectCount, inputObjectSize / maskObjectSize, maskObjectSize, obj, row, col ) ) { |
There was a problem hiding this comment.
minor codestyle There should be no space between if and ( (same rule applies to anything for( while( etc.
| const int index = row * firstWidth + col; | ||
|
|
||
| CIntArray<4> generated = random.Next(); | ||
| for (int j = 0; j < 4 && col + j < firstWidth; ++j) { |
|
|
||
| CIntArray<4> generated = random.Next(); | ||
| for (int j = 0; j < 4 && col + j < firstWidth; ++j) { | ||
| res[index + j] = (generated[j] <= threshold) ? (first[index + j] / forwardRate ) : 0.f; |
There was a problem hiding this comment.
minor codestyle fix spacing around first[index + j] / forwardRate
Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com>
daniyalaliev
force-pushed
the
dropouttask
branch
2 times, most recently
from
1bb0f8c
to
fc56af2
Compare
daniyalaliev
force-pushed
the
dropouttask
branch
from
fc56af2
to
7e20f40
Compare
FedyuninV
approved these changes
Dec 20, 2023
daniyalaliev
added a commit
to daniyalaliev/neoml
that referenced
this pull request
Jan 3, 2024
* Revert "Build Framework (NeoML-master 2.0.204.0): Incrementing version number." This reverts commit 9a29b52. * added test to branch * added rca22824 * started * revert some changes * added some new * reducing dropout memory usage on cuda Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * design changes have been made Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com>
daniyalaliev
added a commit
to daniyalaliev/neoml
that referenced
this pull request
Jan 3, 2024
* Revert "Build Framework (NeoML-master 2.0.204.0): Incrementing version number." This reverts commit 9a29b52. * added test to branch * added rca22824 * started * revert some changes * added some new * reducing dropout memory usage on cuda Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * design changes have been made Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com>
daniyalaliev
added a commit
to daniyalaliev/neoml
that referenced
this pull request
Jan 3, 2024
* Revert "Build Framework (NeoML-master 2.0.204.0): Incrementing version number." This reverts commit 9a29b52. * added test to branch * added rca22824 * started * revert some changes * added some new * reducing dropout memory usage on cuda Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * design changes have been made Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com>
daniyalaliev
added a commit
to daniyalaliev/neoml
that referenced
this pull request
Jan 3, 2024
* Revert "Build Framework (NeoML-master 2.0.204.0): Incrementing version number." This reverts commit 9a29b52. * added test to branch * added rca22824 * started * revert some changes * added some new * reducing dropout memory usage on cuda Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * design changes have been made Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com>
daniyalaliev
added a commit
to daniyalaliev/neoml
that referenced
this pull request
Jan 3, 2024
* Revert "Build Framework (NeoML-master 2.0.204.0): Incrementing version number." This reverts commit 9a29b52. * added test to branch * added rca22824 * started * revert some changes * added some new * reducing dropout memory usage on cuda Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * design changes have been made Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com>
daniyalaliev
added a commit
to daniyalaliev/neoml
that referenced
this pull request
Jan 3, 2024
* Revert "Build Framework (NeoML-master 2.0.204.0): Incrementing version number." This reverts commit 9a29b52. * added test to branch * added rca22824 * started * revert some changes * added some new * reducing dropout memory usage on cuda Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * design changes have been made Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com>
daniyalaliev
added a commit
to daniyalaliev/neoml
that referenced
this pull request
Jan 4, 2024
* Revert "Build Framework (NeoML-master 2.0.204.0): Incrementing version number." This reverts commit 9a29b52. * added test to branch * added rca22824 * started * revert some changes * added some new * reducing dropout memory usage on cuda Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * design changes have been made Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com>
daniyalaliev
added a commit
to daniyalaliev/neoml
that referenced
this pull request
Jan 4, 2024
* Revert "Build Framework (NeoML-master 2.0.204.0): Incrementing version number." This reverts commit 9a29b52. * added test to branch * added rca22824 * started * revert some changes * added some new * reducing dropout memory usage on cuda Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * design changes have been made Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com>
daniyalaliev
added a commit
to daniyalaliev/neoml
that referenced
this pull request
Jan 4, 2024
* Revert "Build Framework (NeoML-master 2.0.204.0): Incrementing version number." This reverts commit 9a29b52. * added test to branch * added rca22824 * started * revert some changes * added some new * reducing dropout memory usage on cuda Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * design changes have been made Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com>
daniyalaliev
added a commit
to daniyalaliev/neoml
that referenced
this pull request
Jan 4, 2024
* Revert "Build Framework (NeoML-master 2.0.204.0): Incrementing version number." This reverts commit 9a29b52. * added test to branch * added rca22824 * started * revert some changes * added some new * reducing dropout memory usage on cuda Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * design changes have been made Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com>
favorart
added a commit
that referenced
this pull request
Feb 27, 2024
* added test to branch Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * added rca22824 Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Build Framework (NeoML-master 2.0.207.0): Incrementing version number. Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Counters linux (#1003) * Revert "Build Framework (NeoML-master 2.0.204.0): Incrementing version number." This reverts commit 9a29b52. * added test to branch * added rca22824 * linux couners modified * fixing linux counters * small change * another count change * added flag for linux counter perfomance Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * added flag for linux counter perfomance Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * added flag for linux counter perfomance Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * added flag for linux counter perfomance Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * added flag for linux counter perfomance Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * fixed adding flag to perfomancecounter Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * another counting fixing Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * added flag for counters linux Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Build Framework (NeoML-master 2.0.208.0): Incrementing version number. Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Reduced CUDA dropout memory usage (#1005) * Revert "Build Framework (NeoML-master 2.0.204.0): Incrementing version number." This reverts commit 9a29b52. * added test to branch * added rca22824 * started * revert some changes * added some new * reducing dropout memory usage on cuda Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * design changes have been made Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * [CudaMathEngine] Fix warp-size iterations in max reducing functions (#1010) Signed-off-by: Kirill Golikov <kirill.golikov@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * [NeoML] Add min-max gradient clipping (#1009) Signed-off-by: Kirill Golikov <kirill.golikov@abbyy.com> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * [CudaMathEngine] Fix restrict modifier for function arguments (#1011) Signed-off-by: Kirill Golikov <kirill.golikov@abbyy.com> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Fix CImageToPixelLayer::SetImageHeight/Width (#1012) Signed-off-by: Valerii Fediunin <valery.fedyunin@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Build Framework (NeoML-master 2.0.209.0): Incrementing version number. Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * [CudaMathEngine] Fix shared buffers initializations and expf calls (#1004) Signed-off-by: Kirill Golikov <kirill.golikov@abbyy.com> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * update fineobj to 15 47 (#1013) * update fineobj to 15 47 Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Upgrade NeoMLTest Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Call set_global_variables before include(FineInstall) Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Remove unused variable Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Fix protobuf compilation errors on Darwin Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Remove invalid unicode symbols from copyright (+update) Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Remove unused variable Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Update NeoMLTest version Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Switch to next NeoMLTest Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Update NeoMLTest Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> --------- Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Build Framework (NeoML-master 2.0.210.0): Incrementing version number. Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Fix iOS CMake toolchain for newer versions of CMake (#1016) Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * [NeoML] Transfer CDnnBlob data in threads pools (#1014) Signed-off-by: Kirill Golikov <kirill.golikov@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * boost problem fixed Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Signed-off-by: Kirill Golikov <kirill.golikov@abbyy.com> Signed-off-by: Valerii Fediunin <valery.fedyunin@abbyy.com> Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> Co-authored-by: buildtech <buildtech@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Co-authored-by: Kirill Golikov <kirill.golikov@abbyy.com> Co-authored-by: NeoML-maintainer <65914319+NeoML-maintainer@users.noreply.github.com>
daniyalaliev
added a commit
to daniyalaliev/neoml
that referenced
this pull request
Mar 1, 2024
* added test to branch Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * added rca22824 Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Build Framework (NeoML-master 2.0.207.0): Incrementing version number. Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Counters linux (neoml-lib#1003) * Revert "Build Framework (NeoML-master 2.0.204.0): Incrementing version number." This reverts commit 9a29b52. * added test to branch * added rca22824 * linux couners modified * fixing linux counters * small change * another count change * added flag for linux counter perfomance Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * added flag for linux counter perfomance Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * added flag for linux counter perfomance Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * added flag for linux counter perfomance Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * added flag for linux counter perfomance Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * fixed adding flag to perfomancecounter Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * another counting fixing Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * added flag for counters linux Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Build Framework (NeoML-master 2.0.208.0): Incrementing version number. Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Reduced CUDA dropout memory usage (neoml-lib#1005) * Revert "Build Framework (NeoML-master 2.0.204.0): Incrementing version number." This reverts commit 9a29b52. * added test to branch * added rca22824 * started * revert some changes * added some new * reducing dropout memory usage on cuda Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * design changes have been made Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * [CudaMathEngine] Fix warp-size iterations in max reducing functions (neoml-lib#1010) Signed-off-by: Kirill Golikov <kirill.golikov@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * [NeoML] Add min-max gradient clipping (neoml-lib#1009) Signed-off-by: Kirill Golikov <kirill.golikov@abbyy.com> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * [CudaMathEngine] Fix restrict modifier for function arguments (neoml-lib#1011) Signed-off-by: Kirill Golikov <kirill.golikov@abbyy.com> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Fix CImageToPixelLayer::SetImageHeight/Width (neoml-lib#1012) Signed-off-by: Valerii Fediunin <valery.fedyunin@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Build Framework (NeoML-master 2.0.209.0): Incrementing version number. Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * [CudaMathEngine] Fix shared buffers initializations and expf calls (neoml-lib#1004) Signed-off-by: Kirill Golikov <kirill.golikov@abbyy.com> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * update fineobj to 15 47 (neoml-lib#1013) * update fineobj to 15 47 Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Upgrade NeoMLTest Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Call set_global_variables before include(FineInstall) Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Remove unused variable Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Fix protobuf compilation errors on Darwin Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Remove invalid unicode symbols from copyright (+update) Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Remove unused variable Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Update NeoMLTest version Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Switch to next NeoMLTest Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> * Update NeoMLTest Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> --------- Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Build Framework (NeoML-master 2.0.210.0): Incrementing version number. Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * Fix iOS CMake toolchain for newer versions of CMake (neoml-lib#1016) Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * [NeoML] Transfer CDnnBlob data in threads pools (neoml-lib#1014) Signed-off-by: Kirill Golikov <kirill.golikov@abbyy.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> * boost problem fixed Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> --------- Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com> Signed-off-by: Kirill Golikov <kirill.golikov@abbyy.com> Signed-off-by: Valerii Fediunin <valery.fedyunin@abbyy.com> Signed-off-by: Valery Fedyunin <valery.fedyunin@abbyy.com> Co-authored-by: buildtech <buildtech@abbyy.com> Co-authored-by: daniyalaliev <daniial.aliev@abbyy.com> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Co-authored-by: Kirill Golikov <kirill.golikov@abbyy.com> Co-authored-by: NeoML-maintainer <65914319+NeoML-maintainer@users.noreply.github.com> Signed-off-by: daniyalaliev <daniial.aliev@abbyy.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Now CUDA accepts only seed to randomly drop matrix elements without spending memory on mask storage