This content has not been validated; it was AI-generated following the x86-64 ISA Documentation generation process to maintain high quality. However, as no human has verified this data, errors may exist. If the accuracy of this data is critical, please consult the Intel SDM. The model used to generate this is gemma4:31b

VRSQRTPH

Computes the approximate reciprocal square root of packed half-precision floating-point numbers.

The table after the description covers what the source and destinations can be.

DO NOT support LOCK

This instruction is only available when AVX-512 is supported. It requires the use of the EVEX encoding scheme.

The instruction provides an approximation of $1/\sqrt{x}$. The precision of the result is limited; it is intended for use as a starting point for iterative refinement (e.g., using Newton-Raphson). It does not support memory operands as sources or destinations; operands MUST be registers.

The instruction produces a result with a relative error of less than $2^{-12}$. If the input is NaN, the result is NaN. If the input is $\pm 0$, the result is $\pm \infty$. If the input is negative, the result is NaN. These operations set the Floating-Point Status Register flags accordingly, including #P.