On Fri, 16 Oct 2026 16:34:12 GMT, Raffaello Giulietti <[email protected]> wrote:
>> The existing noDoubleRoundingToFloat16 check doesn't take into account the >> exponent / significand interaction of subnormals so only normal-range values >> are eligible to go through the existing fast-path. >> >> >> >> --------- >> - [x] I confirm that I make this contribution in accordance with the >> [OpenJDK Interim AI Policy](https://openjdk.org/legal/ai). > > src/jdk.incubator.vector/share/classes/jdk/incubator/vector/Float16.java line > 493: > >> 491: // handles >> infinities too >> 492: Double.isNaN(trialResult) || >> 493: (Math.abs(trialResult) > (0x1.0p-14) && // >> Float16.MIN_VALUE, avoid subnormal range > > Suggestion: > > (Math.abs(trialResult) > (0x1.0p-14) && // Float16.MIN_NORMAL, > avoid subnormal range > > > Any reason for using the static field in the comment but not in the code? Good catch; I'll fix that in the next push. Since the code needs the value as a double and it is a constant, I thought using the double literal was preferable to `Float16.MIN_NORMAL.doubleValue()` or similar expression. ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/33392#discussion_r4297046897
