RFP: New vector operations and APIs.

[Bhateja, Jatin]

FYI, x86 ISA has direct instructions to perform signed / unsigned saturating addition / subtraction only for byte and short vector.


> -----Original Message-----
> From: Bhateja, Jatin
> Sent: Monday, May 13, 2024 7:25 AM
> To: panama-dev at openjdk.org
> Cc: John Rose <john.r.rose at oracle.com>; Paul Sandoz
> <paul.sandoz at oracle.com>; Vladimir Ivanov < vladimir.x.ivanov at oracle.com>; Viswanathan, Sandhya
> <sandhya.viswanathan at intel.com>
> Subject: RFP: New vector operations and APIs.
> 
> Hi All,
> 
> Please find below the proposal for adding new vector operators and API
> based on the user's feedback.
> 
> A) New VectorAPI operations.
>      . VectorOperators.SATURATING_UNSIGNED_ADD
>      . VectorOperators.SATURATING_UNSIGNED_SUB
>      . VectorOperators.SATURATING_ADD
>      . VectorOperators.SATURATING_SUB
> 
> Applicable Types for new operators:
> Prosed vector operators are applicable to integral types since their values
> wraparound in over/underflowing scenarios after setting appropriate status
> flags.
> For floating point types, as per IEEE 754 specs there are multiple schemes to
> handler underflow, one of them is gradual underflow[2] which transitions
> the value to subnormal range. Similarly, overflow implicitly.
> saturates the floating-point value to a representable range or an Infinite
> value.
> As the name suggest, these are saturating operations, i.e. the result of the
> computation is strictly capped by lower and upper bounds of the result type
> and is not wrapped around in underflowing or overflowing scenarios.
> 
> B) New Vector API.
>     . Vector<E>.selectFrom(Vector<E> v1, Vector<E> v2)
> 
> Using index values stored in the lanes of this vector, assemble values stored
> in first (v1) and second (v2) vector arguments. Thus, first and second vector
> serves as a table, whose elements are selected by indexes in the current
> vector. API is applicable to all integral and floating-point types.  The result of
> this operation is semantically equivalent to expression
> Vector.blend(v1.rearrange(this.toShuffle(), v2,rearrange(this.toShuffle),
> this.toShuffle().laneIsValid()). This is a convenience API for an existing
> rearrange API[2], we already have such a convenience API
> "Vector.selectFrom(Vector)"[3] for single vector permute, proposed API
> extends it to two vector permute.
> 
> Target Architecture Support:-
> New operators should emit optimum instruction sequence for VectorAPI supported
> target platforms.
> 
> We plan to target these extensions for JDK-24.
> 
> Please let us know your comments.
> 
> Best Regards,
> Jatin
> 
> [1] https://docs.oracle.com/cd/E19957-01/806-
> 3568/ncg_math.html#:~:text=How%20Does%20IEEE%20Arithmetic%20Treat%
> 20Underflow%3F
> [2]
> https://docs.oracle.com/en/java/javase/22/docs/api/jdk.incubator.vector/jdk
> /incubator/vector/Vector.html#rearrange(jdk.incubator.vector.VectorShuffle,j
> dk.incubator.vector.Vector)
> [3]
> https://docs.oracle.com/en/java/javase/22/docs/api/jdk.incubator.vector/jdk
> /incubator/vector/Vector.html#selectFrom(jdk.incubator.vector.Vector)
> 
> 
> 
>