Multiply packed single-precision (32-bit) floating-point elements in a and b, and store the results in dst. https://www.intel.com/content/www/us/en/do

std::swap has warning on aligned data

* @brief Determine the unquantized value given a quantization level. * * @param quant_level The quantization level to use. * @param value

! ZSTD_isError() : * tells if a return value is an error code * symbol is required for external callers */

Again in str4ncmp we use a memcpy to avoid undefined behavior. The memcpy may appear expensive. Yet all decent optimizing compilers will compile memcp

Again in str4ncmp we use a memcpy to avoid undefined behavior. The memcpy may appear expensive. Yet all decent optimizing compilers will compile memcp

* @brief Clamp a float value between 0.0f and 255.0f. * * NaNs are turned into 0.0f. * * @param v The value to clamp. * * @return The clamped v

* @brief Unquantize a color. * * This function uses a lookup table as the quantization is encoded to make * hardware implementations easier, and is

Add packed single-precision (32-bit) floating-point elements in a and b, and store the results in dst. https://www.intel.com/content/www/us/en/docs/in

Check whether the current bytes are valid UTF-8.

Check whether the current bytes are valid UTF-8.

Returns a pointer to the output buffer content. No terminating null character is appended. */

dot

* @brief Population bit count. * * @param v The value to population count. * * @return The number of 1 bits. */

Zero constructor

Zero constructor

conversion from malloc headers to user pointers, and back

We call a fallback floating-point parser that might be slow. Note it will accept JSON numbers, but the JSON spec. is more restrictive so before you ca

We call a fallback floating-point parser that might be slow. Note it will accept JSON numbers, but the JSON spec. is more restrictive so before you ca

* @brief Store a vector to a 16B aligned memory address. */

Treat space at ptr +/- offset as a chunk

was using kram::format, but wanted to decouple this lib

Copied from GLSLGenerator

* @brief Return the horizontal sum of RGB vector lanes as a scalar. */

resize(0) sets the container to empty, but does not free the allocated block.

We sometimes call trailing_zero on inputs that are zero, but the algorithms do not end up using the returned value. Sadly, sanitizers are not smart en

We sometimes call trailing_zero on inputs that are zero, but the algorithms do not end up using the returned value. Sadly, sanitizers are not smart en

_umul128 for x86, arm this is a slow emulation routine for 32-bit

_umul128 for x86, arm this is a slow emulation routine for 32-bit

Attempts to compute i * 10^(power) exactly; and if "negative" is true, negate the result. This function will only work in some cases, when it does not

Attempts to compute i * 10^(power) exactly; and if "negative" is true, negate the result. This function will only work in some cases, when it does not