RFR: 8289552: Make intrinsic conversions between bit representations of half precision values and floats

[Smita Kamath]

On Fri, 5 Aug 2022 16:36:23 GMT, Smita Kamath <svkamath at openjdk.org> wrote:

> 8289552: Make intrinsic conversions between bit representations of half precision values and floats

JMH results
Before patch
Benchmark                                               (TESTSIZE)   Mode  Cnt    Score   Error   Units
Fp16ConversionBenchmark.float16ToFloat        2048  thrpt    3  359.680 ± 0.292  ops/ms
Fp16ConversionBenchmark.floatToFloat16        2048  thrpt    3  236.153 ± 8.164  ops/ms

After patch
Benchmark                                               (TESTSIZE)   Mode  Cnt     Score   Error   Units
Fp16ConversionBenchmark.float16ToFloat        2048  thrpt    3  1046.233 ± 0.019  ops/ms
Fp16ConversionBenchmark.floatToFloat16        2048  thrpt    3  1752.046 ± 0.058  ops/ms

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

PR: https://git.openjdk.org/jdk/pull/9781