RFR: 8354242: VectorAPI: combine vector not operation with compare [v6]

[erifan]

On Wed, 28 May 2025 12:18:15 GMT, Emanuel Peter <epeter at openjdk.org> wrote:

>> erifan has updated the pull request with a new target base due to a merge or a rebase. The incremental webrev excludes the unrelated changes brought in by the merge/rebase. The pull request contains 10 additional commits since the last revision:
>> 
>>  - Refactor the JTReg tests for compare.xor(maskAll)
>>    
>>    Also made a bit change to support pattern `VectorMask.fromLong()`.
>>  - Merge branch 'master' into JDK-8354242
>>  - Refactor code
>>    
>>    Add a new function XorVNode::Ideal_XorV_VectorMaskCmp to do this
>>    optimization, making the code more modular.
>>  - Merge branch 'master' into JDK-8354242
>>  - Update the jtreg test
>>  - Merge branch 'master' into JDK-8354242
>>  - Addressed some review comments
>>    
>>    1. Call VectorNode::Ideal() only once in XorVNode::Ideal.
>>    2. Improve code comments.
>>  - Merge branch 'master' into JDK-8354242
>>  - Merge branch 'master' into JDK-8354242
>>  - 8354242: VectorAPI: combine vector not operation with compare
>>    
>>    This patch optimizes the following patterns:
>>    For integer types:
>>    ```
>>    (XorV (VectorMaskCmp src1 src2 cond) (Replicate -1))
>>        => (VectorMaskCmp src1 src2 ncond)
>>    (XorVMask (VectorMaskCmp src1 src2 cond) (MaskAll m1))
>>        => (VectorMaskCmp src1 src2 ncond)
>>    ```
>>    cond can be eq, ne, le, ge, lt, gt, ule, uge, ult and ugt, ncond is the
>>    negative comparison of cond.
>>    
>>    For float and double types:
>>    ```
>>    (XorV (VectorMaskCast (VectorMaskCmp src1 src2 cond)) (Replicate -1))
>>        => (VectorMaskCast (VectorMaskCmp src1 src2 ncond))
>>    (XorVMask (VectorMaskCast (VectorMaskCmp src1 src2 cond)) (MaskAll m1))
>>        => (VectorMaskCast (VectorMaskCmp src1 src2 ncond))
>>    ```
>>    cond can be eq or ne.
>>    
>>    Benchmarks on Nvidia Grace machine with 128-bit SVE2:
>>    With option `-XX:UseSVE=2`:
>>    ```
>>    Benchmark			Unit	Before		Score Error	After		Score Error	Uplift
>>    testCompareEQMaskNotByte	ops/s	7912127.225	2677.289518	10266136.26	8955.008548	1.29
>>    testCompareEQMaskNotDouble	ops/s	884737.6799	446.963779	1179760.772	448.031844	1.33
>>    testCompareEQMaskNotFloat	ops/s	1765045.787	682.332214	2359520.803	896.305743	1.33
>>    testCompareEQMaskNotInt		ops/s	1787221.411	977.743935	2353952.519	960.069976	1.31
>>    testCompareEQMaskNotLong	ops/s	895297.1974	673.44808	1178449.02	323.804205	1.31
>>    testCompareEQMaskNotShort	ops/s	3339987.002	3415.2226	4712761.965	2110.862053	1.41
>>    testCompareGEMaskNotByte	ops/s	7907615.16	4...
>
> src/hotspot/share/opto/vectornode.cpp line 2251:
> 
>> 2249:                                       predicate_node, vt);
>> 2250:   if (vmcast_vt != nullptr) {
>> 2251:     // We optimized out an VectorMaskCast, and in order to ensure type
> 
> Suggestion:
> 
>     // We optimized out a VectorMaskCast, and in order to ensure type

Done.

> src/hotspot/share/opto/vectornode.cpp line 2253:
> 
>> 2251:     // We optimized out an VectorMaskCast, and in order to ensure type
>> 2252:     // correctness, we need to regenerate one. VectorMaskCast will be encoded as
>> 2253:     // empty for types with the same size.
> 
> Suggestion:
> 
>     // a no-op (identity function) for types with the same size.
> 
> Or what do you mean by "empty"? `TOP`? All zeros?

I mean `no-op`. Done, thanks.

> test/hotspot/jtreg/compiler/vectorapi/VectorMaskCompareNotTest.java line 96:
> 
>> 94:         Generator<Long> lGen = RD.uniformLongs(Long.MIN_VALUE, Long.MAX_VALUE);
>> 95:         Generator<Float> fGen = RD.uniformFloats(Float.MIN_VALUE, Float.MAX_VALUE);
>> 96:         Generator<Double> dGen = RD.uniformDoubles(Double.MIN_VALUE, Double.MAX_VALUE);
> 
> Are you sure you only want to draw from the uniform distribution?
> If you don't super care about the distribution, please just take `RD.ints/longs/floats/doubles()`.
> That way, you get all sorts of distributions, and also some that include NaN values etc. I think that would be important for your float cmp cases, no?

For float and double, we have to use the uniform distribution, because we have to make sure `NAN` is not generated. I added some comments about the reasons.

For other types, changed to use `RD.ints/longs`.

We have covered the special cases like +/- Inf, NaN.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/24674#discussion_r2128376851
PR Review Comment: https://git.openjdk.org/jdk/pull/24674#discussion_r2128378888
PR Review Comment: https://git.openjdk.org/jdk/pull/24674#discussion_r2128375032