called from find_content_hash in the ConcreteTensor class. hash_in includes hash of dtype and interface.

called from find_content_hash in the ConcreteTensor class. hash_in includes hash of dtype and interface.

this is useful for defining findhash<X> on other types in other headers, in terms of std::hash<X>, without needing to include this header first.

this is useful for defining findhash<X> on other types in other headers, in terms of std::hash<X>, without needing to include this header first.

* @brief low-cost frexpf (but only the exponent result); * Generates only a few instructions on hexagon. * * Input must not be inf,nan, zero, or de

* @brief low-cost frexpf (but only the exponent result); * Generates only a few instructions on hexagon. * * Input must not be inf,nan, zero, or de

* @brief low-cost ldexpf * Same as ldexpf(val, iexpo) without a function call; * * Constraint: iexpo must be in range -126..127 */

* @brief low-cost ldexpf * Same as ldexpf(val, iexpo) without a function call; * * Constraint: iexpo must be in range -126..127 */

* @brief low-cost 2.0*n for integer n. * Same as powf(2.0f, iexpo) without a function call; * * Constraint: iexpo must be in range -126..127 */

* @brief low-cost 2.0*n for integer n. * Same as powf(2.0f, iexpo) without a function call; * * Constraint: iexpo must be in range -126..127 */

We can put arithmetic functions in a separate library, but we want the namespace here. */ Functions that need things like the context to evaluate shou

We can put arithmetic functions in a separate library, but we want the namespace here. */ Functions that need things like the context to evaluate shou

pointer to 'offs ' within data area

pointer to 'offs ' within data area

* @brief given a scale in float and a recip shift amount * return a quantized scale multiplier and change recip shamt inplace * */

* @brief given a scale in float and a recip shift amount * return a quantized scale multiplier and change recip shamt inplace * */

this is to support Tensor::get_dims()

this is to support Tensor::get_dims()

{size_align} of interface object (may or may not be quantized)

{size_align} of interface object (may or may not be quantized)

is_power_of_two: check for some number of zeros, followed by 1, followed by some number of zeros. FIXME: maybe should use bitset?

is_power_of_two: check for some number of zeros, followed by 1, followed by some number of zeros. FIXME: maybe should use bitset?

make a copy with the same output and no inputs

make a copy with the same output and no inputs

move_from is only used when rehashing; the 'from' record is known to be non-empty. We move-construct the value, and delete the old one.

move_from is only used when rehashing; the 'from' record is known to be non-empty. We move-construct the value, and delete the old one.

Helper method to populate all input tensors during execution.