memory usage.
*   Fixed bfloat16 integration of LIBXSMM sparse mat-mul.
*   Improved performance and reduce memory usage by allowing ops to forward
    input buffers to output buffers and perform computations in-place.
*   Improved the performance of CPU assignment for strings.
*   Speed up matrix * vector multiplication and matrix * matrix with unknown
    shapes.

#include "textflag.h" // func xorBytes(dst, a, b *byte, n int) TEXT ·xorBytes(SB), NOSPLIT|NOFRAME, $0 MOV dst+0(FP), X10 MOV a+8(FP), X11 MOV b+16(FP), X12 MOV n+24(FP), X13 MOV $32, X15 BLT X13, X15, loop4_check // Check alignment - if alignment differs we have to do one byte at a time. AND $7, X10, X5 AND $7, X11, X6 AND $7, X12, X7 BNE X5, X6, loop4_check BNE X5, X7, loop4_check BEQZ X5, loop64_check // Check one byte at a time until we reach 8 byte alignment. MOV $8, X8 SUB X5, X8 SUB X8, X13...

  3. if no IP is found, NodeIP defaults to 127.0.0.1 and the IP family to IPv4 ([#91725](https://github.com/kubernetes/kubernetes/pull/91725), [@aojea](https://github.com/aojea)) [SIG Network]
- Kube-proxy, in dual-stack mode, infers the service IP family from the ClusterIP instead of using the `Service.Spec.IPFamily` field ([#91357](https://github.com/kubernetes/kubernetes/pull/91357), [@aojea](https://github.com/aojea))

Section 5.7.1.2. return p.BytesX() } // isZero reports whether x is all zeroes in constant time. func isZero(x []byte) bool { var acc byte for _, b := range x { acc |= b } return acc == 0 } // isLess reports whether a < b, where a and b are big-endian buffers of the // same length and shorter than 72 bytes. func isLess(a, b []byte) bool { if len(a) != len(b) { panic("crypto/ecdh: internal error: mismatched isLess inputs") } // Copy the values into a fixed-size preallocated little-endian buffer. // 72 bytes...