if window&1 == 0 { // If the window value is even, preserve the carry and continue. // Why is the carry preserved? // If carry == 0 and window & 1 == 0, // then the next carry should be 0 // If carry == 1 and window & 1 == 0, // then bit_buf & 1 == 1 so the next carry should be 1 pos += 1 continue } if window < width/2 { carry = 0 naf[pos] = int8(window) } else { carry = 1 naf[pos] = int8(window) - int8(width) } pos += w } return naf } func (s *Scalar) signedRadix16() [64]int8 { b := s.Bytes() if...

if window&1 == 0 { // If the window value is even, preserve the carry and continue. // Why is the carry preserved? // If carry == 0 and window & 1 == 0, // then the next carry should be 0 // If carry == 1 and window & 1 == 0, // then bit_buf & 1 == 1 so the next carry should be 1 pos += 1 continue } if window < width/2 { carry = 0 naf[pos] = int8(window) } else { carry = 1 naf[pos] = int8(window) - int8(width) } pos += w } return naf } func (s *Scalar) signedRadix16() [64]int8 { b := s.Bytes() if...

    `CUDA_ILLEGAL_ADDRESS` failures.

    Google discovered in mid-December 2017 that the PTX-to-SASS compiler in CUDA
    9 and CUDA 9.1 sometimes does not properly compute the carry bit when
    decomposing 64-bit address calculations with large offsets (e.g. `load [x +
    large_constant]`) into 32-bit arithmetic in SASS.