With AES-CTR, it is trivial; it's actually one of the advantages of using the scheme. In AES-CTR, there is a counter which is incremented for each block, encrypted, and then XOR'ed to the ciphertext in order to obtain the plaintext. To parallelize the decryption of, e.g., 100 blocks, you simply encrypt the counter for the 1st-50th blocks in one thread and the counter of the 50th-100th blocks in another (simply add 50 to the initial counter value and start from there).
AES-GCM is more or less the same, since it also uses something like CTR internally for encryption. You will have to also check the authenticity, but this can also be done in parallel. (Some people are not confortable with starting decryption before checking the authenticity, but that's another discussion...)