x86/crypto/crct10dif-pcl-asm_64.S

2 # Implement fast CRC-T10DIF computation with SSE and PCLMULQDQ instructions
50 #  /white-papers/fast-crc-computation-generic-polynomials-pclmulqdq-paper.pdf
136 	# While >= 128 data bytes remain (not counting xmm0-7), fold the 128
137 	# bytes xmm0-7 into them, storing the result back into xmm0-7.
147 	# Now fold the 112 bytes in xmm0-xmm6 into the 16 bytes in xmm7.
167 	add	$128-16, len
200 	movdqu	-16(buf, len), %xmm1
203 	# xmm2 = high order part of second chunk: xmm7 left-shifted by 'len' bytes.
209 	# xmm7 = first chunk: xmm7 right-shifted by '16-len' bytes.
213 	# xmm1 = second chunk: 'len' bytes from xmm1 (low-order bytes),
214 	# then '16-len' bytes from xmm2 (high-order bytes).
225 	# Reduce the 128-bit value M(x), stored in xmm7, to the final 16-bit CRC
227 	# Load 'x^48 * (x^48 mod G(x))' and 'x^48 * (x^80 mod G(x))'.
230 	# Fold the high 64 bits into the low 64 bits, while also multiplying by
231 	# x^64.  This produces a 128-bit value congruent to x^64 * M(x) and
232 	# whose low 48 bits are 0.
234 	pclmulqdq	$0x11, FOLD_CONSTS, %xmm7 # high bits * x^48 * (x^80 mod G(x))
236 	pxor	%xmm0, %xmm7			  # + low bits * x^64
238 	# Fold the high 32 bits into the low 96 bits.  This produces a 96-bit
239 	# value congruent to x^64 * M(x) and whose low 48 bits are 0.
243 	pclmulqdq	$0x00, FOLD_CONSTS, %xmm7 # high 32 bits * x^48 * (x^48 mod G(x))
244 	pxor	%xmm0, %xmm7			  # + low bits
246 	# Load G(x) and floor(x^48 / G(x)).
251 	pclmulqdq	$0x11, FOLD_CONSTS, %xmm7 # high 32 bits * floor(x^48 / G(x))
253 	pclmulqdq	$0x00, FOLD_CONSTS, %xmm7 # *= G(x)
255 	pxor	%xmm7, %xmm0		     # + low 16 nonzero bits
256 	# Final CRC value (x^16 * M(x)) mod G(x) is in low 16 bits of xmm0.
288 # G(x) = x^16 + x^15 + x^11 + x^9 + x^8 + x^7 + x^5 + x^4 + x^2 + x^1 + x^0
290 	.quad		0x0000000000006123	# x^(8*128)	mod G(x)
291 	.quad		0x0000000000002295	# x^(8*128+64)	mod G(x)
293 	.quad		0x0000000000001069	# x^(4*128)	mod G(x)
294 	.quad		0x000000000000dd31	# x^(4*128+64)	mod G(x)
296 	.quad		0x000000000000857d	# x^(2*128)	mod G(x)
297 	.quad		0x0000000000007acc	# x^(2*128+64)	mod G(x)
299 	.quad		0x000000000000a010	# x^(1*128)	mod G(x)
300 	.quad		0x0000000000001faa	# x^(1*128+64)	mod G(x)
302 	.quad		0x1368000000000000	# x^48 * (x^48 mod G(x))
303 	.quad		0x2d56000000000000	# x^48 * (x^80 mod G(x))
305 	.quad		0x0000000000018bb7	# G(x)
306 	.quad		0x00000001f65a57f8	# floor(x^48 / G(x))
325 # For 1 <= len <= 15, the 16-byte vector beginning at &byteshift_table[16 - len]
327 # 0x80} XOR the index vector to shift right by '16 - len' bytes.