RFR: 8370863: VectorAPI: Optimize the VectorMaskCast chain in specific patterns

[Eric Fang]

On Fri, 14 Nov 2025 01:17:50 GMT, Eric Fang <erfang at openjdk.org> wrote:

> `VectorMaskCastNode` is used to cast a vector mask from one type to another type. The cast may be generated by calling the vector API `cast` or generated by the compiler. For example, some vector mask operations like `trueCount` require the input mask to be integer types, so for floating point type masks, the compiler will cast the mask to the corresponding integer type mask automatically before doing the mask operation. This kind of cast is very common.
> 
> If the vector element size is not changed, the `VectorMaskCastNode` don't generate code, otherwise code will be generated to extend or narrow the mask. This IR node is not free no matter it generates code or not because it may block some optimizations. For example:
> 1. `(VectorStoremask (VectorMaskCast (VectorLoadMask x)))` The middle `VectorMaskCast` prevented the following optimization: `(VectorStoremask (VectorLoadMask x)) => (x)`
> 2. `(VectorMaskToLong (VectorMaskCast (VectorLongToMask x)))`, which blocks the optimization `(VectorMaskToLong (VectorLongToMask x)) => (x)`.
> 
> In these IR patterns, the value of the input `x` is not changed, so we can safely do the optimization. But if the input value is changed, we can't eliminate the cast.
> 
> The general idea of this PR is introducing an `uncast_mask` helper function, which can be used to uncast a chain of `VectorMaskCastNode`, like the existing `Node::uncast(bool)` function. The funtion returns the first non `VectorMaskCastNode`.
> 
> The intended use case is when the IR pattern to be optimized may contain one or more consecutive `VectorMaskCastNode` and this does not affect the correctness of the optimization. Then this function can be called to eliminate the `VectorMaskCastNode` chain.
> 
> Current optimizations related to `VectorMaskCastNode` include:
> 1. `(VectorMaskCast (VectorMaskCast x)) => (x)`, see JDK-8356760.
> 2. `(XorV (VectorMaskCast (VectorMaskCmp src1 src2 cond)) (Replicate -1)) => (VectorMaskCast (VectorMaskCmp src1 src2 ncond))`, see JDK-8354242.
> 
> This PR does the following optimizations:
> 1. Extends the optimization pattern `(VectorMaskCast (VectorMaskCast x)) => (x)` as `(VectorMaskCast (VectorMaskCast  ... (VectorMaskCast x))) => (x)`. Because as long as types of the head and tail `VectorMaskCastNode` are consistent, the optimization is correct.
> 2. Supports a new optimization pattern `(VectorStoreMask (VectorMaskCast ... (VectorLoadMask x))) => (x)`. Since the value before and after the pattern is a boolean vector, it remains unchanged as long as th...

Updated the JMH benchmarks and the new test results:

On a Nvidia Grace machine with 128-bit SVE2:

    Benchmark                       Unit    Before  Error   After   Error   Uplift
    microMaskLoadCastStoreByte64    ops/us  64.29   0.02    146.67  0.09    2.28
    microMaskLoadCastStoreDouble128 ops/us  10.05   0.00    38.10   0.01    3.79
    microMaskLoadCastStoreFloat128  ops/us  19.94   0.00    75.05   0.07    3.76
    microMaskLoadCastStoreInt128    ops/us  19.94   0.00    75.13   0.01    3.77
    microMaskLoadCastStoreLong128   ops/us  10.04   0.00    38.09   0.01    3.79
    microMaskLoadCastStoreShort64   ops/us  31.52   0.02    75.12   0.02    2.38

On a Nvidia Grace machine with 128-bit NEON:

    Benchmark                       Unit    Before  Error   After   Error   Uplift
    microMaskLoadCastStoreByte64    ops/us  73.33   0.01    147.01  0.06    2.00
    microMaskLoadCastStoreDouble128 ops/us  8.54    0.03    38.19   0.01    4.47
    microMaskLoadCastStoreFloat128  ops/us  23.75   0.01    75.27   0.10    3.17
    microMaskLoadCastStoreInt128    ops/us  23.73   0.01    75.25   0.07    3.17
    microMaskLoadCastStoreLong128   ops/us  8.56    0.03    38.19   0.01    4.46
    microMaskLoadCastStoreShort64   ops/us  24.32   0.00    75.35   0.07    3.10

On an AMD EPYC 9124 16-Core Processor with AVX3:

    Benchmark                       Unit    Before  Error   After   Error   Uplift
    microMaskLoadCastStoreByte64    ops/us  82.39   0.11    115.15  0.03    1.40
    microMaskLoadCastStoreDouble128 ops/us  0.32    0.00    0.32    0.00    0.99
    microMaskLoadCastStoreFloat128  ops/us  42.10   0.10    57.58   0.02    1.37
    microMaskLoadCastStoreInt128    ops/us  42.10   0.08    57.57   0.02    1.37
    microMaskLoadCastStoreLong128   ops/us  0.32    0.00    0.32    0.00    0.99
    microMaskLoadCastStoreShort64   ops/us  42.16   0.05    57.54   0.04    1.36


On an AMD EPYC 9124 16-Core Processor with AVX2:

    Benchmark                       Unit    Before  Error   After   Error   Uplift
    microMaskLoadCastStoreByte64    ops/us  73.59   0.27    115.14  0.04    1.56
    microMaskLoadCastStoreDouble128 ops/us  0.30    0.00    0.30    0.00    1.01
    microMaskLoadCastStoreFloat128  ops/us  30.68   0.09    57.57   0.02    1.88
    microMaskLoadCastStoreInt128    ops/us  30.75   0.09    57.58   0.01    1.87
    microMaskLoadCastStoreLong128   ops/us  0.30    0.00    0.30    0.00    1.00
    microMaskLoadCastStoreShort64   ops/us  24.95   0.01    57.59   0.01    2.31

On an AMD EPYC 9124 16-Core Processor with AVX1:

    Benchmark                       Unit    Before  Error   After   Error   Uplift
    microMaskLoadCastStoreByte64    ops/us  73.68   0.02    115.17  0.03    1.56
    microMaskLoadCastStoreDouble128 ops/us  0.30    0.00    0.30    0.00    1.01
    microMaskLoadCastStoreFloat128  ops/us  30.80   0.12    57.59   0.01    1.87
    microMaskLoadCastStoreInt128    ops/us  30.70   0.11    57.58   0.01    1.88
    microMaskLoadCastStoreLong128   ops/us  0.30    0.00    0.30    0.00    0.99
    microMaskLoadCastStoreShort64   ops/us  24.95   0.01    57.56   0.02    2.31

-------------

PR Comment: https://git.openjdk.org/jdk/pull/28313#issuecomment-3555660413