[{"name":"Bytes.Floating.Decode","comment":" Extra Floating-point binary representation for Elm\n\n@docs float16\n\n","unions":[],"aliases":[],"values":[{"name":"float16","comment":" Decode 2 bytes as an IEEE 754 half-precision (16-bit) floating-point\nnumber, in the given byte order, and widen it to Elm's native `Float` (64-bit).\n\nHalf-precision layout (16 bits):\n\n       exponent\n              |        mantissa\n    sign      |               |\n       |      |               |\n       |      |               |\n       | /---------\\/-------------------\\\n       *  * * * * *  * * * * * * * * * *  (16-bit)\n\nInterpretation by exponent value `e`:\n\n    e in [1..30] : normal    →  (-1)^s * 2^(e-15)   * (1 + m/1024)\n    e == 0, m/=0 : subnormal →  (-1)^s * 2^(-14)    * (0 + m/1024)\n    e == 0, m==0 : zero      →  +/- 0.0\n    e == 31, m==0: infinity  →  +/- Infinity\n    e == 31, m/=0: NaN\n\nThe conversion is exact: every float16 value is exactly representable\nin float64. The decoding uses pure arithmetic rather than bit\nreinterpretation through Bytes, making it ~35x faster than the\nfloat32-roundtrip approach.\n\n","type":"Bytes.Endianness -> Bytes.Decode.Decoder Basics.Float"}],"binops":[]},{"name":"Bytes.Floating.Encode","comment":" Extra Floating-point binary representation for Elm\n\n@docs float16\n\n","unions":[],"aliases":[],"values":[{"name":"float16","comment":" Encode a `Float` as an IEEE 754 half-precision (16-bit) floating-point\nnumber, in the given byte order.\n\nHalf-precision layout (16 bits):\n\n       exponent\n              |        mantissa\n    sign      |               |\n       |      |               |\n       |      |               |\n       | /---------\\/-------------------\\\n       *  * * * * *  * * * * * * * * * *  (16-bit)\n\nDepending on the exponent `e`:\n\n    e in [1..30] : normal    →  (-1)^s * 2^(e-15)   * (1 + m/1024)\n    e == 0, m/=0 : subnormal →  (-1)^s * 2^(-14)    * (0 + m/1024)\n    e == 0, m==0 : zero      →  +/- 0.0\n    e == 31, m==0: infinity  →  +/- Infinity\n    e == 31, m/=0: NaN\n\nRepresentable range: ~6.0e-8 (smallest subnormal) to 65504 (largest finite).\nValues outside this range are flushed to zero or infinity respectively.\n\nThe encoding uses pure arithmetic (frexp-style decomposition) rather than\nbit reinterpretation through Bytes, making it ~35x faster than the\nfloat32-roundtrip approach.\n\n","type":"Bytes.Endianness -> Basics.Float -> Bytes.Encode.Encoder"}],"binops":[]}]