Adds support for Ion Schema 2.0 ieee754_float constraint #213
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
popematt
force-pushed
the
isl2-ieee754float
branch
from
9b6e58b
to
f83fb9f
Compare
jobarr-amzn
approved these changes
Oct 8, 2022
Comment on lines
53
to
69
| private val BINARY_16_PRECISION_RANGES = listOf( | ||
| PrecisionInfo(0.0..1.0 / 16_384, interval = 1.0 / 16_777_216), // Subnormal numbers | ||
| PrecisionInfo(1.0 / 16_384..1.0 / 8_192, interval = 1.0 / 16_777_216), | ||
| PrecisionInfo(1.0 / 8_192..1.0 / 4_096, interval = 1.0 / 8_388_608), | ||
| PrecisionInfo(1.0 / 4_096..1.0 / 2_048, interval = 1.0 / 4_194_304), | ||
| PrecisionInfo(1.0 / 2_048..1.0 / 1_024, interval = 1.0 / 2_097_152), | ||
| PrecisionInfo(1.0 / 1_024..1.0 / 512, interval = 1.0 / 1_048_576), | ||
| PrecisionInfo(1.0 / 512..1.0 / 256, interval = 1.0 / 524_288), | ||
| PrecisionInfo(1.0 / 256..1.0 / 128, interval = 1.0 / 262_144), | ||
| PrecisionInfo(1.0 / 128..1.0 / 64, interval = 1.0 / 131_072), | ||
| PrecisionInfo(1.0 / 64..1.0 / 32, interval = 1.0 / 65_536), | ||
| PrecisionInfo(1.0 / 32..1.0 / 16, interval = 1.0 / 32_768), | ||
| PrecisionInfo(1.0 / 16..1.0 / 8, interval = 1.0 / 16_384), | ||
| PrecisionInfo(1.0 / 8..1.0 / 4, interval = 1.0 / 8_192), | ||
| PrecisionInfo(1.0 / 4..1.0 / 2, interval = 1.0 / 4_096), | ||
| PrecisionInfo(1.0 / 2..1.0, interval = 1.0 / 2_048), | ||
| PrecisionInfo(1.0..2.0, interval = 1.0 / 1_024), | ||
| PrecisionInfo(2.0..4.0, interval = 1.0 / 512), | ||
| PrecisionInfo(4.0..8.0, interval = 1.0 / 256), | ||
| PrecisionInfo(8.0..16.0, interval = 1.0 / 128), | ||
| PrecisionInfo(16.0..32.0, interval = 1.0 / 64), | ||
| PrecisionInfo(32.0..64.0, interval = 1.0 / 32), | ||
| PrecisionInfo(64.0..128.0, interval = 1.0 / 16), | ||
| PrecisionInfo(128.0..256.0, interval = 1.0 / 8), | ||
| PrecisionInfo(256.0..512.0, interval = 1.0 / 4), | ||
| PrecisionInfo(512.0..1_024.0, interval = 1.0 / 2), | ||
| PrecisionInfo(1_024.0..2_048.0, interval = 1.0), | ||
| PrecisionInfo(2_048.0..4_096.0, interval = 2.0), | ||
| PrecisionInfo(4_096.0..8_192.0, interval = 4.0), | ||
| PrecisionInfo(8_192.0..16_384.0, interval = 8.0), | ||
| PrecisionInfo(16_384.0..32_768.0, interval = 16.0), | ||
| PrecisionInfo(32_768.0..65_504.0, interval = 32.0), | ||
| ) | ||
| } |
There was a problem hiding this comment.
I played with this a bit to make sure I was understanding it :)
A couple alternate phrasings here, fwiw:
val BINARY_16_PRECISION_RANGES = sequence {
yield(PrecisionInfo( 0.0..2.0.pow(-14), interval = 2.0.pow(-24))) // Subnormal numbers
yieldAll((-14..14).asSequence().map {
PrecisionInfo(2.0.pow(it)..2.0.pow(it+1), interval = 2.0.pow(it-10))
})
// 65_504 is the largest precisely representable number
yield(PrecisionInfo(2.0.pow( 15)..65_504.0, interval = 2.0.pow( 5)))
}.toList()or the equally verbose but also equivalent:
val BINARY_16_PRECISION_RANGES = listOf(
PrecisionInfo( 0.0..2.0.pow(-14), interval = 2.0.pow(-24)), // Subnormal numbers
PrecisionInfo(2.0.pow(-14)..2.0.pow(-13), interval = 2.0.pow(-24)),
PrecisionInfo(2.0.pow(-13)..2.0.pow(-12), interval = 2.0.pow(-23)),
PrecisionInfo(2.0.pow(-12)..2.0.pow(-11), interval = 2.0.pow(-22)),
PrecisionInfo(2.0.pow(-11)..2.0.pow(-10), interval = 2.0.pow(-21)),
PrecisionInfo(2.0.pow(-10)..2.0.pow( -9), interval = 2.0.pow(-20)),
PrecisionInfo(2.0.pow( -9)..2.0.pow( -8), interval = 2.0.pow(-19)),
PrecisionInfo(2.0.pow( -8)..2.0.pow( -7), interval = 2.0.pow(-18)),
PrecisionInfo(2.0.pow( -7)..2.0.pow( -6), interval = 2.0.pow(-17)),
PrecisionInfo(2.0.pow( -6)..2.0.pow( -5), interval = 2.0.pow(-16)),
PrecisionInfo(2.0.pow( -5)..2.0.pow( -4), interval = 2.0.pow(-15)),
PrecisionInfo(2.0.pow( -4)..2.0.pow( -3), interval = 2.0.pow(-14)),
PrecisionInfo(2.0.pow( -3)..2.0.pow( -2), interval = 2.0.pow(-13)),
PrecisionInfo(2.0.pow( -2)..2.0.pow( -1), interval = 2.0.pow(-12)),
PrecisionInfo(2.0.pow( -1)..2.0.pow( 0), interval = 2.0.pow(-11)),
PrecisionInfo(2.0.pow( 0)..2.0.pow( 1), interval = 2.0.pow(-10)),
PrecisionInfo(2.0.pow( 1)..2.0.pow( 2), interval = 2.0.pow( -9)),
PrecisionInfo(2.0.pow( 2)..2.0.pow( 3), interval = 2.0.pow( -8)),
PrecisionInfo(2.0.pow( 3)..2.0.pow( 4), interval = 2.0.pow( -7)),
PrecisionInfo(2.0.pow( 4)..2.0.pow( 5), interval = 2.0.pow( -6)),
PrecisionInfo(2.0.pow( 5)..2.0.pow( 6), interval = 2.0.pow( -5)),
PrecisionInfo(2.0.pow( 6)..2.0.pow( 7), interval = 2.0.pow( -4)),
PrecisionInfo(2.0.pow( 7)..2.0.pow( 8), interval = 2.0.pow( -3)),
PrecisionInfo(2.0.pow( 8)..2.0.pow( 9), interval = 2.0.pow( -2)),
PrecisionInfo(2.0.pow( 9)..2.0.pow( 10), interval = 2.0.pow( -1)),
PrecisionInfo(2.0.pow( 10)..2.0.pow( 11), interval = 2.0.pow( 0)),
PrecisionInfo(2.0.pow( 11)..2.0.pow( 12), interval = 2.0.pow( 1)),
PrecisionInfo(2.0.pow( 12)..2.0.pow( 13), interval = 2.0.pow( 2)),
PrecisionInfo(2.0.pow( 13)..2.0.pow( 14), interval = 2.0.pow( 3)),
PrecisionInfo(2.0.pow( 14)..2.0.pow( 15), interval = 2.0.pow( 4)),
PrecisionInfo(2.0.pow( 15)..65_504.0, interval = 2.0.pow( 5)),
)There was a problem hiding this comment.
I also suspect there's a fast generalizeable way to detect this from bits, using toRawBits() and bit twiddling.
I looked at Float Toy and the manipulations of this gist, but ran out of energy and enthusiasm for reasoning through the corner cases.
I suspect that detecting "out of bounds" values in the double precision bit field is likely easier than straight conversion.
In any case, this was a fun diversion and I'm not at all asking for any changes here just sharing my thoughts.
There was a problem hiding this comment.
I'll create an issue for coming up with a generalizable solution. For now, I'm going to switch to the sequence-based factoring you suggested and add a link to the "Precision Limitations" table in the relevant wikipedia page.
popematt
force-pushed
the
isl2-ieee754float
branch
from
f83fb9f
to
8f0cfa9
Compare
popematt
force-pushed
the
isl2-ieee754float
branch
from
8f0cfa9
to
bca80bd
Compare
jobarr-amzn
approved these changes
Oct 11, 2022
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Issue #, if available:
#207
Description of changes:
ieee754_floatconstraintIonSchemaTests_2_0suite to include theieee754_floattest casesRelated PRs in ion-schema, ion-schema-tests, ion-schema-schemas:
Test cases: amazon-ion/ion-schema-tests#29
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.