Lines Matching +full:xor +full:- +full:v2
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:
138 * matching ranks. Then, the final XOR (*) can be pulled forward, and
141 * 80-bit results.
144 * to the 64x64 bit one above, but XOR'ing the outputs together will
225 CPU_LE( rev64 v2.16b, v2.16b )
233 CPU_LE( ext v2.16b, v2.16b, v2.16b, #8 )
240 // XOR the first 16 data *bits* with the initial CRC value.
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.
256 fold_32_bytes \p, v2, v3
263 // Now fold the 112 bytes in v0-v6 into the 16 bytes in v7.
270 fold_16_bytes \p, v2, v6
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.
319 ld1 {v2.16b}, [x4]
320 tbl v1.16b, {v7.16b}, v2.16b
322 // v3 = first chunk: v7 right-shifted by '16-len' bytes.
324 eor v2.16b, v2.16b, v3.16b
325 tbl v3.16b, {v7.16b}, v2.16b
327 // Convert to 8-bit masks: 'len' 0x00 bytes, then '16-len' 0xff bytes.
328 sshr v2.16b, v2.16b, #7
330 // v2 = second chunk: 'len' bytes from v0 (low-order bytes),
331 // then '16-len' bytes from v1 (high-order bytes).
332 bsl v2.16b, v1.16b, v0.16b
337 eor v7.16b, v7.16b, v2.16b
350 // XOR the first 16 data *bits* with the initial CRC value.
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.
404 movi v2.16b, #0 // init zero register
410 // x^64. This produces a 128-bit value congruent to x^64 * M(x) and
412 ext v0.16b, v2.16b, v7.16b, #8
416 // Fold the high 32 bits into the low 96 bits. This produces a 96-bit
418 ext v1.16b, v0.16b, v2.16b, #12 // extract high 32 bits
419 mov v0.s[3], v2.s[0] // zero high 32 bits
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.