* @brief Normalize a non-zero length vector to unit length. */

* @brief Reset the tracker for a new processing batch. * * This must be called from single-threaded code before starting the multi-threaded proces

combined JPEG 'receive' and JPEG 'extend', since baseline always extends everything it receives.

returns 0..31 for the highest set bit

extract an arbitrarily-aligned N-bit value (N=bits) from v, and then make it 8-bits long and fractionally extend it to full full range.

returns STBI_rgb or whatever, 0 on error

* @brief Unsigned integer bit-interpreted as an SP float. * * @param v The value to bitcast. * * @return The converted value. */

* @brief Un-blue-contract a color. * * This function reverses any applied blue contraction. * * @param input The input color that has been blue-

* @brief Validate single channel compression swizzle. * * @param swizzle The swizzle to check. * * @return Return @c ASTCENC_SUCCESS if validate

* @brief Validate single channel decompression swizzle. * * @param swizzle The swizzle to check. * * @return Return @c ASTCENC_SUCCESS if valida

Parse wavelength from spectral channel name

Return sum of tile blocks. 0 = error

-1 = error

Add a custom error message

See header for documentation. */

* @brief Apply signed bit transfer. * * @param input0 The first encoded endpoint. * @param input1 The second encoded endpoint. */

* @brief Count the leading zeros for each lane in @c a. * * Valid for all data values of @c a; will return a per-lane value [0, 32]. */

* @brief Complete a task assignment. * * Mark @c count tasks as complete. This will notify all threads blocked on @c wait() if this * completes

* @brief The main function for the angular algorithm. * * @param weight_count The number of (decimated) weights. * @param de

See header for documentation. */

* @brief Compute the ideal endpoints and weights for 2 color components. * * @param blk The image block color data to compress. * @pa

See header for documentation. */

See header for documentation. */

* @brief Get the total number of blocks in an image. * * This function also validates that the total size of the compressed image, * in bytes, woul

* @brief Get the default alpha value for endpoints that don't store it. * * The default depends on whether the alpha endpoint is LDR or HDR. *

See header for documentation. */

* @brief Request a task assignment. * * Assign up to @c granule tasks to the caller for processing. * * @param granule Maximum number o

* @brief Unpack an HDR RGB direct encoding. * * @param input The packed endpoints (packed and modal). * @param[out] output0 The unpacked

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

* @brief Trigger the pipeline stage init step. * * This can be called from multi-threaded code. The first thread to hit this will process the *

* @brief Compute the integer linear interpolation of two color endpoints. * * @param u8_mask The mask for lanes using decode_unorm8 rather tha

* @brief sRGB gamma-encode a linear value. * * @return The gamma encoded value. */

* @brief Convert 16-bit LNS to float16. */

* @brief Compute an approximate log2(x) for each lane in the vector. * * Based on 5th degree minimax polynomials, ported from this blog * https://j

* @brief Return the min vector of a vector and a scalar. * * If either lane value is NaN, @c b will be returned for that lane. */

See header for documentation. */

See header for documentation. */

* @brief Compute an approximate pow(x, y) for each lane in the vector. * * Power function based on the exp2(log2(x) * y) transform. */

* @brief Quantize a HDR RGB color using direct RGB encoding. * * @param color0 The input unquantized color0 endpoint. * @param col

Read 1 byte

* @brief Read up to 16 bits from two bytes. * * This function reads a packed N-bit field from two bytes in memory. The stored value must exist * wi

* @brief Return lanes from @c b if @c cond is set, else @c a. */

See header for documentation */

* @brief Linearize an sRGB value. * * @return The linearized value. */

return 1 if the sum is valid, 0 on overflow. negative terms are considered invalid.

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

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

* @brief Return lanewise 2^a for each lane in @c a. * * Use of signed int means that this is only valid for values in range [0, 31]. */

* @brief Factory that returns a unit length 4 component vfloat4. */