old API will be deprecated in a future release. See
        https://www.tensorflow.org/xla/custom_call for details.
    *   XLA:GPU reductions are deterministic by default (reductions within
        `jit_compile=True` are now deterministic).
    *   XLA:GPU works with Horovod (OSS contribution by Trent Lo from NVidia)

bits, a reseed interval of 2^48 requests, and a // maximum request size of 2^19 bits (2^16 bytes, 64 KiB). // // We support a narrow range of parameters that fit the needs of our RNG: // AES-256, no derivation function, no personalization string, no prediction // resistance, and 384-bit additional input. type Counter struct { // c is instantiated with K as the key and V as the counter. c aes.CTR reseedCounter uint64 } const ( keySize = 256 / 8 SeedSize = keySize + aes.BlockSize reseedInterval = 1 << 48...