extendf

pub inline fn extendf(
    comptime dst_t: type,
    comptime src_t: type,
    a: @Int(.unsigned, @typeInfo(src_t).float.bits),
) dst_t

File

Code

pub inline fn extendf(
    comptime dst_t: type,
    comptime src_t: type,
    a: @Int(.unsigned, @typeInfo(src_t).float.bits),
) dst_t {
    const src_rep_t = @Int(.unsigned, @typeInfo(src_t).float.bits);
    const dst_rep_t = @Int(.unsigned, @typeInfo(dst_t).float.bits);
    const srcSigBits = std.math.floatMantissaBits(src_t);
    const dstSigBits = std.math.floatMantissaBits(dst_t);

    // Various constants whose values follow from the type parameters.
    // Any reasonable optimizer will fold and propagate all of these.
    const srcBits = @bitSizeOf(src_t);
    const srcExpBits = srcBits - srcSigBits - 1;
    const srcInfExp = (1 << srcExpBits) - 1;
    const srcExpBias = srcInfExp >> 1;

    const srcMinNormal = 1 << srcSigBits;
    const srcInfinity = srcInfExp << srcSigBits;
    const srcSignMask = 1 << (srcSigBits + srcExpBits);
    const srcAbsMask = srcSignMask - 1;
    const srcQNaN = 1 << (srcSigBits - 1);
    const srcNaNCode = srcQNaN - 1;

    const dstBits = @bitSizeOf(dst_t);
    const dstExpBits = dstBits - dstSigBits - 1;
    const dstInfExp = (1 << dstExpBits) - 1;
    const dstExpBias = dstInfExp >> 1;

    const dstMinNormal: dst_rep_t = @as(dst_rep_t, 1) << dstSigBits;

    // Break a into a sign and representation of the absolute value
    const aRep: src_rep_t = @bitCast(a);
    const aAbs: src_rep_t = aRep & srcAbsMask;
    const sign: src_rep_t = aRep & srcSignMask;
    var absResult: dst_rep_t = undefined;

    if (aAbs -% srcMinNormal < srcInfinity - srcMinNormal) {
        // a is a normal number.
        // Extend to the destination type by shifting the significand and
        // exponent into the proper position and rebiasing the exponent.
        absResult = @as(dst_rep_t, aAbs) << (dstSigBits - srcSigBits);
        absResult += (dstExpBias - srcExpBias) << dstSigBits;
    } else if (aAbs >= srcInfinity) {
        // a is NaN or infinity.
        // Conjure the result by beginning with infinity, then setting the qNaN
        // bit (if needed) and right-aligning the rest of the trailing NaN
        // payload field.
        absResult = dstInfExp << dstSigBits;
        absResult |= @as(dst_rep_t, aAbs & srcQNaN) << (dstSigBits - srcSigBits);
        absResult |= @as(dst_rep_t, aAbs & srcNaNCode) << (dstSigBits - srcSigBits);
    } else if (aAbs != 0) {
        // a is denormal.
        // renormalize the significand and clear the leading bit, then insert
        // the correct adjusted exponent in the destination type.
        const scale: u32 = @clz(aAbs) - @clz(@as(src_rep_t, srcMinNormal));
        absResult = @as(dst_rep_t, aAbs) << @intCast(dstSigBits - srcSigBits + scale);
        absResult ^= dstMinNormal;
        const resultExponent: u32 = dstExpBias - srcExpBias - scale + 1;
        absResult |= @as(dst_rep_t, @intCast(resultExponent)) << dstSigBits;
    } else {
        // a is zero.
        absResult = 0;
    }

    // Apply the signbit to (dst_t)abs(a).
    const result: dst_rep_t align(@alignOf(dst_t)) = absResult | @as(dst_rep_t, sign) << (dstBits - srcBits);
    return @bitCast(result);
}