RFR: 8283091: Support type conversion between different data sizes in SLP [v3]

[Fei Gao]

On Wed, 25 May 2022 01:13:36 GMT, Fei Gao <fgao at openjdk.org> wrote:

>> @fg1417  Thank you for suggesting this optimization. I see that it was not updated for some time. Do you still intend to work on it?
>> 
>> Please update to latest JDK (Loom was integrated) and run performance again. Also include % of changes.
>> 
>> I have the same concern as @DamonFool about regression when vectorizing some conversions. May be we should have additional `Matcher` property we could consult when trying to **auto-vectorize**. I understand that we need `vcvt2Dto2I` when VectorAPI specifically asking to generate it but we should not enforce auto-generation.
>
>> Please update to latest JDK (Loom was integrated) and run performance again. Also include % of changes.
>> 
>> I have the same concern as @DamonFool about regression when vectorizing some conversions. May be we should have additional `Matcher` property we could consult when trying to **auto-vectorize**. I understand that we need `vcvt2Dto2I` when VectorAPI specifically asking to generate it but we should not enforce auto-generation.
> 
> @vnkozlov thanks for your review and kind suggestion! I'll update the patch to resolve the potential performance regression.

> @fg1417 I don't see new update in this PR. Please also show performance numbers with new changes

Here is the perf uplift data (ns/op) on different machines for the latest patch.

  NEON          perf change (ns/op)
convertB2D      not supported
convertB2F      -45.55%
convertB2L      not supported
convertD2B      not supported
convertD2F      -42.32%
convertD2I      not supported (VectorAPI supported)
convertD2S      not supported
convertF2B      -42.95%
convertF2D      -45.28%
convertF2L      -5.78%
convertF2S      -51.30%
convertI2D      -27.82%
convertI2L      -44.54%
convertL2B      not supported
convertL2F      not supported (VectorAPI supported)
convertL2I      -28.58%
convertL2S      not supported
convertS2D      not supported
convertS2F      -53.37%
convertS2L      not supported

  SVE           perf change (ns/op)
convertB2D      -36.15%
convertB2F      -63.48%
convertB2L      -32.48%
convertD2B      0.02%
convertD2F      -47.85%
convertD2I      -46.42%
convertD2S      -32.08%
convertF2B      -59.54%
convertF2D      -60.81%
convertF2L      -61.81%
convertF2S      -67.67%
convertI2D      -60.63%
convertI2L      -57.23%
convertL2B      0.04%
convertL2F      -47.21%
convertL2I      -34.49%
convertL2S      -19.57%
convertS2D      -47.20%
convertS2F      -74.86%
convertS2L      -49.00%

  X86           perf change (ns/op)
convertB2D      -64.13%
convertB2F      -79.37%
convertB2L      -70.97%
convertD2B      not supported
convertD2F      -62.69%
convertD2I      not supported
convertD2S      not supported
convertF2B      not supported
convertF2D      -68.90%
convertF2L      not supported
convertF2S      not supported
convertI2D      -87.48%
convertI2L      -69.64%
convertL2B      -3.96%
convertL2F      -0.11%
convertL2I      -49.59%
convertL2S      -24.75%
convertS2D      -84.35%
convertS2F      -86.09%
convertS2L      -70.42%

  AVX512        perf change (ns/op)
convertB2D      -78.08%
convertB2F      -86.39%
convertB2L      -79.07%
convertD2B      not supported
convertD2F      -71.86%
convertD2I      not supported
convertD2S      not supported
convertF2B      not supported
convertF2D      -78.17%
convertF2L      not supported
convertF2S      not supported
convertI2D      -90.26%
convertI2L      -79.92%
convertL2B      -70.75%
convertL2F      -86.67%
convertL2I      -80.94%
convertL2S      -71.54%
convertS2D      -90.84%
convertS2F      -83.94%
convertS2L      -80.51%

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

PR: https://git.openjdk.java.net/jdk/pull/7806