Lines Matching +full:low +full:- +full:g
2 // Accelerated CRC-T10DIF using arm64 NEON and Crypto Extensions instructions
14 // Implement fast CRC-T10DIF computation with SSE and PCLMULQDQ instructions
62 // /white-papers/fast-crc-computation-generic-polynomials-pclmulqdq-paper.pdf
69 .arch armv8-a+crypto
143 pmull t7.8h, ad.8b, bd2.8b // G = A*B2
157 pmull2 t7.8h, ad.16b, bd2.16b // G = A*B2
163 eor t5.16b, t5.16b, t7.16b // M = G + H
305 // While >= 128 data bytes remain (not counting v0-v7), fold the 128
306 // bytes v0-v7 into them, storing the result back into v0-v7.
316 // Now fold the 112 bytes in v0-v6 into the 16 bytes in v7.
336 adds len, len, #(128-16)
367 ldr q0, [buf, #-16]
371 // v1 = high order part of second chunk: v7 left-shifted by 'len' bytes.
377 // v3 = first chunk: v7 right-shifted by '16-len' bytes.
382 // Convert to 8-bit masks: 'len' 0x00 bytes, then '16-len' 0xff bytes.
385 // v2 = second chunk: 'len' bytes from v0 (low-order bytes),
386 // then '16-len' bytes from v1 (high-order bytes).
396 // Reduce the 128-bit value M(x), stored in v7, to the final 16-bit CRC.
400 // Load 'x^48 * (x^48 mod G(x))' and 'x^48 * (x^80 mod G(x))'.
404 // Fold the high 64 bits into the low 64 bits, while also multiplying by
405 // x^64. This produces a 128-bit value congruent to x^64 * M(x) and
406 // whose low 48 bits are 0.
408 __pmull_\p v7, v7, fold_consts, 2 // high bits * x^48 * (x^80 mod G(x))
409 eor v0.16b, v0.16b, v7.16b // + low bits * x^64
411 // Fold the high 32 bits into the low 96 bits. This produces a 96-bit
412 // value congruent to x^64 * M(x) and whose low 48 bits are 0.
415 __pmull_\p v1, v1, fold_consts // high 32 bits * x^48 * (x^48 mod G(x))
416 eor v0.16b, v0.16b, v1.16b // + low bits
418 // Load G(x) and floor(x^48 / G(x)).
423 __pmull_\p v1, v0, fold_consts, 2 // high 32 bits * floor(x^48 / G(x))
425 __pmull_\p v1, v1, fold_consts // *= G(x)
427 eor v0.16b, v0.16b, v1.16b // + low 16 nonzero bits
428 // Final CRC value (x^16 * M(x)) mod G(x) is in low 16 bits of v0.
451 // Load the fold-across-16-bytes constants.
469 stp x29, x30, [sp, #-16]!
488 // G(x) = x^16 + x^15 + x^11 + x^9 + x^8 + x^7 + x^5 + x^4 + x^2 + x^1 + x^0
490 .quad 0x0000000000006123 // x^(8*128) mod G(x)
491 .quad 0x0000000000002295 // x^(8*128+64) mod G(x)
493 .quad 0x0000000000001069 // x^(4*128) mod G(x)
494 .quad 0x000000000000dd31 // x^(4*128+64) mod G(x)
496 .quad 0x000000000000857d // x^(2*128) mod G(x)
497 .quad 0x0000000000007acc // x^(2*128+64) mod G(x)
499 .quad 0x000000000000a010 // x^(1*128) mod G(x)
500 .quad 0x0000000000001faa // x^(1*128+64) mod G(x)
502 .quad 0x1368000000000000 // x^48 * (x^48 mod G(x))
503 .quad 0x2d56000000000000 // x^48 * (x^80 mod G(x))
505 .quad 0x0000000000018bb7 // G(x)
506 .quad 0x00000001f65a57f8 // floor(x^48 / G(x))
508 // For 1 <= len <= 15, the 16-byte vector beginning at &byteshift_table[16 -
510 // ..., 0x80} XOR the index vector to shift right by '16 - len' bytes.