A wrapper around fwrite that throws on error.

Returns a cached power of 10 `c_k = c_k.f * pow(2, c_k.e)` such that its (binary) exponent satisfies `min_exponent <= c_k.e <= min_exponent + 28`.

Given the divisor (normally a power of 10), the remainder = v % divisor for some number v and the error, returns whether v should be rounded up, down,

Check if finite number. \param arg number to check \retval true if neither infinity nor NaN \retval false else template<typename T> typename enable<bo

the implementation of parseOption()

This function is paraphrased from: tensorflow/lite/kernels/internal/quantization_util.cc QuantizeMultiplier()

Nearest integer using half's internal rounding mode. \param arg half expression to round \return nearest integer using default rounding mode template<

create the png data from post-deflated data

decode one 64-entry block--

get some unsigned bits

returns 1 if "a*b*c*d + add" has no negative terms/factors and doesn't overflow

set up the kernels

read 16bit value and convert to 24bit RGB

@OPTIMIZE: provide an option that always forces left-predict or paeth predict

Each scan line uses the same kernel values so we should calculate the kernel values once and then we can use them for every scan line.

What output pixels does this input pixel contribute to?

What input pixels contribute to this output pixel?

Dummy implementations of strerror_r and strerror_s called if corresponding system functions are not available.

Hyperbolic tangent. \param arg function argument \return hyperbolic tangent value of \a arg template<typename T> typename enable<expr,T>::type tanh(T

* Gets the array2d total size. */