Repository navigation

Vectorize BitIncrement and BitDecrement for float/double in TensorPrimitives - #123610

Merged

stephentoub merged 15 commits into

mainfrom

copilot/vectorize-bit-operations

Jan 27, 2026

Commits on Jan 26, 2026

Initial plan
Copilot
committed
e8ae0e1
Implement vectorization for BitIncrement and BitDecrement operations

Copilot
and
stephentoub
committed
93c545f
Fix sign bit detection and NaN masking in BitIncrement and BitDecrement

Copilot
and
stephentoub
committed
7b31447
Refactor: Consolidate variables and move declarations closer to usage per code review

Copilot
and
stephentoub
committed
97ccc34
Add Half type support via TryUnaryInvokeHalfAsInt16

Copilot
and
stephentoub
committed
40e6702
Change else if to if and add blank lines for readability

Copilot
and
stephentoub
committed
1d4810c
Use Vector.IsNegative() API instead of manual sign bit checking

Copilot
and
stephentoub
committed
72a7b3c
Remove Half type support due to performance regression

Copilot
and
stephentoub
committed
bbe89c3
Optimize special case handling with single ConditionalSelect per @tannergooding

Copilot
and
stephentoub
committed
0522459
Fix IsZero to operate on float/double vectors instead of uint/ulong

Copilot
and
stephentoub
committed
1c6bb24
Remove redundant infinity handlers - already handled by general case

Copilot
and
tannergooding
committed
a1f645e
Re-add Half type support via TryUnaryInvokeHalfAsInt16

Copilot
and
stephentoub
committed
3ec0153
Optimize IsZero check to use bitwise comparison for positive zero only

Copilot
and
stephentoub
committed
4b3b884
Merge branch 'main' into copilot/vectorize-bit-operations
stephentoub
authored
3df9c30
Remove Half type support due to performance regression

Copilot
and
stephentoub
committed
394d589

Navigation Menu

Repository navigation

Vectorize BitIncrement and BitDecrement for float/double in TensorPrimitives - #123610

Vectorize BitIncrement and BitDecrement for float/double in TensorPrimitives#123610
stephentoub merged 15 commits into
mainfrom
copilot/vectorize-bit-operations

Commits on Jan 26, 2026

Initial plan

Implement vectorization for BitIncrement and BitDecrement operations

Fix sign bit detection and NaN masking in BitIncrement and BitDecrement

Refactor: Consolidate variables and move declarations closer to usage per code review

Add Half type support via TryUnaryInvokeHalfAsInt16

Change else if to if and add blank lines for readability

Use Vector.IsNegative() API instead of manual sign bit checking

Remove Half type support due to performance regression

Optimize special case handling with single ConditionalSelect per @tannergooding

Fix IsZero to operate on float/double vectors instead of uint/ulong

Remove redundant infinity handlers - already handled by general case

Re-add Half type support via TryUnaryInvokeHalfAsInt16

Optimize IsZero check to use bitwise comparison for positive zero only

Merge branch 'main' into copilot/vectorize-bit-operations

Remove Half type support due to performance regression