## Known Bugs

*   Using XLA:GPU with CUDA 9 and CUDA 9.1 results in garbage results and/or
    `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 +

32CA          ; mapped                 ; 0031 0031 6708 #1.1  IDEOGRAPHIC TELEGRAPH SYMBOL FOR NOVEMBER
32CB          ; mapped                 ; 0031 0032 6708 #1.1  IDEOGRAPHIC TELEGRAPH SYMBOL FOR DECEMBER
32CC          ; mapped                 ; 0068 0067     # 4.0  SQUARE HG
32CD          ; mapped                 ; 0065 0072 0067 #4.0  SQUARE ERG
32CE          ; mapped                 ; 0065 0076     # 4.0  SQUARE EV

is governed by a BSD-style // license that can be found in the LICENSE file. // This Go implementation is derived in part from the reference // ANSI C implementation, which carries the following notice: // // rijndael-alg-fst.c // // @version 3.0 (December 2000) // // Optimised ANSI C code for the Rijndael cipher (now AES) // // @author Vincent Rijmen <******@****.***> // @author Antoon Bosselaers <******@****.***> // @author Paulo Barreto <******@****.***>...