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.

s, and r instead of ρ'. func sampleNTT(rho []byte, s, r byte) nttElement { G := sha3.NewShake128() G.Write(rho) G.Write([]byte{s, r}) var a nttElement var j int // index into a var buf [168]byte // buffered reads from B, matching the rate of SHAKE-128 off := len(buf) // index into buf, starts in a "buffer fully consumed" state for j < n { if off >= len(buf) { G.Read(buf[:]) off = 0 } v := uint32(buf[off]) | uint32(buf[off+1])<<8 | uint32(buf[off+2])<<16 off += 3 f, err := fieldToMontgomery(v & 0b0...

rightmost bits are the most significant // bits (dropped with a mask) and the leftmost bits are the least // significant bits (dropped with a right shift). var a nttElement var j int // index into a var buf [24]byte // buffered reads from B off := len(buf) // index into buf, starts in a "buffer fully consumed" state for { if off >= len(buf) { B.Read(buf[:]) off = 0 } d1 := byteorder.LEUint16(buf[off:]) & 0b1111_1111_1111 d2 := byteorder.LEUint16(buf[off+1:]) >> 4 off += 3 if d1 < q { a[j] = fieldElement(d1)...