Lines Matching +full:low +full:- +full:g
2 // Accelerated CRC-T10DIF using arm64 NEON and Crypto Extensions instructions
5 // Copyright (C) 2019-2024 Google LLC
17 // Implement fast CRC-T10DIF computation with SSE and PCLMULQDQ instructions
65 // /white-papers/fast-crc-computation-generic-polynomials-pclmulqdq-paper.pdf
72 .arch armv8-a+crypto
96 * Pairwise long polynomial multiplication of two 16-bit values
100 * by two 64-bit values
128 * and after performing 8x8->16 bit long polynomial multiplication of
130 * we obtain the following four vectors of 16-bit elements:
141 * 80-bit results.
252 // While >= 128 data bytes remain (not counting v0-v7), fold the 128
253 // bytes v0-v7 into them, storing the result back into v0-v7.
263 // Now fold the 112 bytes in v0-v6 into the 16 bytes in v7.
282 adds len, len, #(128-16)
312 ldr q0, [buf, #-16]
316 // v1 = high order part of second chunk: v7 left-shifted by 'len' bytes.
322 // v3 = first chunk: v7 right-shifted by '16-len' bytes.
327 // Convert to 8-bit masks: 'len' 0x00 bytes, then '16-len' 0xff bytes.
330 // v2 = second chunk: 'len' bytes from v0 (low-order bytes),
331 // then '16-len' bytes from v1 (high-order bytes).
355 // Load the fold-across-16-bytes constants.
402 // Reduce the 128-bit value M(x), stored in v7, to the final 16-bit CRC.
406 // Load 'x^48 * (x^48 mod G(x))' and 'x^48 * (x^80 mod G(x))'.
409 // Fold the high 64 bits into the low 64 bits, while also multiplying by
410 // x^64. This produces a 128-bit value congruent to x^64 * M(x) and
411 // whose low 48 bits are 0.
413 pmull2 v7.1q, v7.2d, fold_consts.2d // high bits * x^48 * (x^80 mod G(x))
414 eor v0.16b, v0.16b, v7.16b // + low bits * x^64
416 // Fold the high 32 bits into the low 96 bits. This produces a 96-bit
417 // value congruent to x^64 * M(x) and whose low 48 bits are 0.
420 pmull v1.1q, v1.1d, fold_consts.1d // high 32 bits * x^48 * (x^48 mod G(x))
421 eor v0.16b, v0.16b, v1.16b // + low bits
423 // Load G(x) and floor(x^48 / G(x)).
427 pmull2 v1.1q, v0.2d, fold_consts.2d // high 32 bits * floor(x^48 / G(x))
429 pmull v1.1q, v1.1d, fold_consts.1d // *= G(x)
431 eor v0.16b, v0.16b, v1.16b // + low 16 nonzero bits
432 // Final CRC value (x^16 * M(x)) mod G(x) is in low 16 bits of v0.
442 // G(x) = x^16 + x^15 + x^11 + x^9 + x^8 + x^7 + x^5 + x^4 + x^2 + x^1 + x^0
444 .quad 0x0000000000006123 // x^(8*128) mod G(x)
445 .quad 0x0000000000002295 // x^(8*128+64) mod G(x)
447 .quad 0x0000000000001069 // x^(4*128) mod G(x)
448 .quad 0x000000000000dd31 // x^(4*128+64) mod G(x)
450 .quad 0x000000000000857d // x^(2*128) mod G(x)
451 .quad 0x0000000000007acc // x^(2*128+64) mod G(x)
453 .quad 0x000000000000a010 // x^(1*128) mod G(x)
454 .quad 0x0000000000001faa // x^(1*128+64) mod G(x)
456 .quad 0x1368000000000000 // x^48 * (x^48 mod G(x))
457 .quad 0x2d56000000000000 // x^48 * (x^80 mod G(x))
459 .quad 0x0000000000018bb7 // G(x)
460 .quad 0x00000001f65a57f8 // floor(x^48 / G(x))
462 // For 1 <= len <= 15, the 16-byte vector beginning at &byteshift_table[16 -
464 // ..., 0x80} XOR the index vector to shift right by '16 - len' bytes.