[fio.git] / crc / crc32c-intel.c

#include <inttypes.h>
#include "crc32c.h"

/*
 * Based on a posting to lkml by Austin Zhang <austin.zhang@intel.com>
 *
 * Using hardware provided CRC32 instruction to accelerate the CRC32 disposal.
 * CRC32C polynomial:0x1EDC6F41(BE)/0x82F63B78(LE)
 * CRC32 is a new instruction in Intel SSE4.2, the reference can be found at:
 * http://www.intel.com/products/processor/manuals/
 * Intel(R) 64 and IA-32 Architectures Software Developer's Manual
 * Volume 2A: Instruction Set Reference, A-M
 */

#ifdef ARCH_HAVE_SSE

#if BITS_PER_LONG == 64
#define REX_PRE "0x48, "
#define SCALE_F 8
#else
#define REX_PRE
#define SCALE_F 4
#endif

uint32_t crc32c_intel_le_hw_byte(uint32_t crc, unsigned char const *data,
				 unsigned long length)
{
	while (length--) {
		__asm__ __volatile__(
			".byte 0xf2, 0xf, 0x38, 0xf0, 0xf1"
			:"=S"(crc)
			:"0"(crc), "c"(*data)
		);
		data++;
	}

	return crc;
}

/*
 * Steps through buffer one byte at at time, calculates reflected 
 * crc using table.
 */
uint32_t crc32c_intel(unsigned char const *data, unsigned long length)
{
	unsigned int iquotient = length / SCALE_F;
	unsigned int iremainder = length % SCALE_F;
#if BITS_PER_LONG == 64
	uint64_t *ptmp = (uint64_t *) data;
#else
	uint32_t *ptmp = (uint32_t *) data;
#endif
	uint32_t crc = ~0;

	while (iquotient--) {
		__asm__ __volatile__(
			".byte 0xf2, " REX_PRE "0xf, 0x38, 0xf1, 0xf1;"
			:"=S"(crc)
			:"0"(crc), "c"(*ptmp)
		);
		ptmp++;
	}

	if (iremainder)
		crc = crc32c_intel_le_hw_byte(crc, (unsigned char *)ptmp,
				 iremainder);

	return crc;
}

#endif /* ARCH_HAVE_SSE */
Commit	Line	Data
3845591f	1	#include <inttypes.h>
419484b9	2	#include "crc32c.h"
3845591f JA	3
	4	/*
	5	* Based on a posting to lkml by Austin Zhang <austin.zhang@intel.com>
	6	*
	7	* Using hardware provided CRC32 instruction to accelerate the CRC32 disposal.
	8	* CRC32C polynomial:0x1EDC6F41(BE)/0x82F63B78(LE)
	9	* CRC32 is a new instruction in Intel SSE4.2, the reference can be found at:
	10	* http://www.intel.com/products/processor/manuals/
	11	* Intel(R) 64 and IA-32 Architectures Software Developer's Manual
	12	* Volume 2A: Instruction Set Reference, A-M
	13	*/
	14
419484b9 AC	15	#ifdef ARCH_HAVE_SSE
419484b9 AC	16
3845591f JA	17	#if BITS_PER_LONG == 64
	18	#define REX_PRE "0x48, "
	19	#define SCALE_F 8
	20	#else
	21	#define REX_PRE
	22	#define SCALE_F 4
	23	#endif
	24
	25	uint32_t crc32c_intel_le_hw_byte(uint32_t crc, unsigned char const *data,
	26	unsigned long length)
	27	{
	28	while (length--) {
	29	__asm__ __volatile__(
	30	".byte 0xf2, 0xf, 0x38, 0xf0, 0xf1"
	31	:"=S"(crc)
	32	:"0"(crc), "c"(*data)
	33	);
	34	data++;
	35	}
	36
	37	return crc;
	38	}
	39
	40	/*
	41	* Steps through buffer one byte at at time, calculates reflected
	42	* crc using table.
	43	*/
	44	uint32_t crc32c_intel(unsigned char const *data, unsigned long length)
	45	{
	46	unsigned int iquotient = length / SCALE_F;
	47	unsigned int iremainder = length % SCALE_F;
	48	#if BITS_PER_LONG == 64
	49	uint64_t ptmp = (uint64_t ) data;
	50	#else
	51	uint32_t ptmp = (uint32_t ) data;
	52	#endif
	53	uint32_t crc = ~0;
	54
	55	while (iquotient--) {
	56	__asm__ __volatile__(
	57	".byte 0xf2, " REX_PRE "0xf, 0x38, 0xf1, 0xf1;"
	58	:"=S"(crc)
	59	:"0"(crc), "c"(*ptmp)
	60	);
	61	ptmp++;
	62	}
	63
	64	if (iremainder)
	65	crc = crc32c_intel_le_hw_byte(crc, (unsigned char *)ptmp,
	66	iremainder);
	67
	68	return crc;
	69	}
419484b9 AC	70
	71	#endif /* ARCH_HAVE_SSE */
	72