6  * This software is provided 'as-is', without any express or implied warranty.
43 	 * clang (at least 3.9.[0-1]) pessimizes "rm" (y) and "m" (y)  in _mm_crc32_u8()
69 /* CRC-32C (iSCSI) polynomial in reversed bit order. */
73  * Block sizes for three-way parallel crc computation.  LONG and SHORT must
133 	uint32_t odd[32];       /* odd-power-of-two zeros operator */  in crc32c_zeros_op()
138 	odd[0] = POLY;              /* CRC-32C polynomial */  in crc32c_zeros_op()
153 	 * bits), in even -- next square puts operator for two zero bytes in  in crc32c_zeros_op()
165 	/* answer ended up in odd -- copy to even */  in crc32c_zeros_op()
172  * for that length, byte-by-byte on the operand.
214 /* Compute CRC-32C using the Intel hardware instruction. */
219 	const size_t align = 8;  in sse42_crc32c()  local
221 	const size_t align = 4;  in sse42_crc32c()
223 	const unsigned char *next, *end;  in sse42_crc32c()  local
234 	while (len && ((uintptr_t)next & (align - 1)) != 0) {  in sse42_crc32c()
237 		len--;  in sse42_crc32c()
243 	 * crc instructions, each on LONG bytes -- this is optimized for the  in sse42_crc32c()
252 		end = next + LONG;  in sse42_crc32c()
267 			next += align;  in sse42_crc32c()
268 		} while (next < end);  in sse42_crc32c()
269 		/*-  in sse42_crc32c()
294 		 * can run in 24 cycles, so the 3-way blocking is worse  in sse42_crc32c()
317 		len -= LONG * 3;  in sse42_crc32c()
330 		end = next + SHORT;  in sse42_crc32c()
345 			next += align;  in sse42_crc32c()
346 		} while (next < end);  in sse42_crc32c()
352 		len -= SHORT * 3;  in sse42_crc32c()
357 	end = next + (len - (len & (align - 1)));  in sse42_crc32c()
358 	while (next < end) {  in sse42_crc32c()
364 		next += align;  in sse42_crc32c()
366 	len &= (align - 1);  in sse42_crc32c()
372 		len--;  in sse42_crc32c()