raid6sse1.c 5.11 KB
Newer Older
Andrew Morton's avatar
Andrew Morton committed
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
/* -*- linux-c -*- ------------------------------------------------------- *
 *
 *   Copyright 2002 H. Peter Anvin - All Rights Reserved
 *
 *   This program is free software; you can redistribute it and/or modify
 *   it under the terms of the GNU General Public License as published by
 *   the Free Software Foundation, Inc., 53 Temple Place Ste 330,
 *   Bostom MA 02111-1307, USA; either version 2 of the License, or
 *   (at your option) any later version; incorporated herein by reference.
 *
 * ----------------------------------------------------------------------- */

/*
 * raid6sse1.c
 *
 * SSE-1/MMXEXT implementation of RAID-6 syndrome functions
 *
 * This is really an MMX implementation, but it requires SSE-1 or
 * AMD MMXEXT for prefetch support and a few other features.  The
 * support for nontemporal memory accesses is enough to make this
 * worthwhile as a separate implementation.
 */

Andrew Morton's avatar
Andrew Morton committed
24
#if defined(__i386__)
Andrew Morton's avatar
Andrew Morton committed
25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171

#include "raid6.h"
#include "raid6x86.h"

/* Defined in raid6mmx.c */
extern const struct raid6_mmx_constants {
	u64 x1d;
} raid6_mmx_constants;

static int raid6_have_sse1_or_mmxext(void)
{
#ifdef __KERNEL__
	/* Not really boot_cpu but "all_cpus" */
	return boot_cpu_has(X86_FEATURE_MMX) &&
		(boot_cpu_has(X86_FEATURE_XMM) ||
		 boot_cpu_has(X86_FEATURE_MMXEXT));
#else
	/* User space test code - this incorrectly breaks on some Athlons */
	u32 features = cpuid_features();
	return ( (features & (5<<23)) == (5<<23) );
#endif
}

/*
 * Plain SSE1 implementation
 */
static void raid6_sse11_gen_syndrome(int disks, size_t bytes, void **ptrs)
{
	u8 **dptr = (u8 **)ptrs;
	u8 *p, *q;
	int d, z, z0;
	raid6_mmx_save_t sa;

	z0 = disks - 3;		/* Highest data disk */
	p = dptr[z0+1];		/* XOR parity */
	q = dptr[z0+2];		/* RS syndrome */

	/* This is really MMX code, not SSE */
	raid6_before_mmx(&sa);

	asm volatile("movq %0,%%mm0" : : "m" (raid6_mmx_constants.x1d));
	asm volatile("pxor %mm5,%mm5");	/* Zero temp */

	for ( d = 0 ; d < bytes ; d += 8 ) {
		asm volatile("prefetchnta %0" : : "m" (dptr[z0][d]));
		asm volatile("movq %0,%%mm2" : : "m" (dptr[z0][d])); /* P[0] */
		asm volatile("prefetchnta %0" : : "m" (dptr[z0-1][d]));
		asm volatile("movq %mm2,%mm4");	/* Q[0] */
		asm volatile("movq %0,%%mm6" : : "m" (dptr[z0-1][d]));
		for ( z = z0-2 ; z >= 0 ; z-- ) {
			asm volatile("prefetchnta %0" : : "m" (dptr[z][d]));
			asm volatile("pcmpgtb %mm4,%mm5");
			asm volatile("paddb %mm4,%mm4");
			asm volatile("pand %mm0,%mm5");
			asm volatile("pxor %mm5,%mm4");
			asm volatile("pxor %mm5,%mm5");
			asm volatile("pxor %mm6,%mm2");
			asm volatile("pxor %mm6,%mm4");
			asm volatile("movq %0,%%mm6" : : "m" (dptr[z][d]));
		}
		asm volatile("pcmpgtb %mm4,%mm5");
		asm volatile("paddb %mm4,%mm4");
		asm volatile("pand %mm0,%mm5");
		asm volatile("pxor %mm5,%mm4");
		asm volatile("pxor %mm5,%mm5");
		asm volatile("pxor %mm6,%mm2");
		asm volatile("pxor %mm6,%mm4");

		asm volatile("movntq %%mm2,%0" : "=m" (p[d]));
		asm volatile("movntq %%mm4,%0" : "=m" (q[d]));
	}

	raid6_after_mmx(&sa);
	asm volatile("sfence" : : : "memory");
}

const struct raid6_calls raid6_sse1x1 = {
	raid6_sse11_gen_syndrome,
	raid6_have_sse1_or_mmxext,
	"sse1x1",
	1			/* Has cache hints */
};

/*
 * Unrolled-by-2 SSE1 implementation
 */
static void raid6_sse12_gen_syndrome(int disks, size_t bytes, void **ptrs)
{
	u8 **dptr = (u8 **)ptrs;
	u8 *p, *q;
	int d, z, z0;
	raid6_mmx_save_t sa;

	z0 = disks - 3;		/* Highest data disk */
	p = dptr[z0+1];		/* XOR parity */
	q = dptr[z0+2];		/* RS syndrome */

	raid6_before_mmx(&sa);

	asm volatile("movq %0,%%mm0" : : "m" (raid6_mmx_constants.x1d));
	asm volatile("pxor %mm5,%mm5");	/* Zero temp */
	asm volatile("pxor %mm7,%mm7"); /* Zero temp */

	/* We uniformly assume a single prefetch covers at least 16 bytes */
	for ( d = 0 ; d < bytes ; d += 16 ) {
		asm volatile("prefetchnta %0" : : "m" (dptr[z0][d]));
		asm volatile("movq %0,%%mm2" : : "m" (dptr[z0][d])); /* P[0] */
		asm volatile("movq %0,%%mm3" : : "m" (dptr[z0][d+8])); /* P[1] */
		asm volatile("movq %mm2,%mm4");	/* Q[0] */
		asm volatile("movq %mm3,%mm6"); /* Q[1] */
		for ( z = z0-1 ; z >= 0 ; z-- ) {
			asm volatile("prefetchnta %0" : : "m" (dptr[z][d]));
			asm volatile("pcmpgtb %mm4,%mm5");
			asm volatile("pcmpgtb %mm6,%mm7");
			asm volatile("paddb %mm4,%mm4");
			asm volatile("paddb %mm6,%mm6");
			asm volatile("pand %mm0,%mm5");
			asm volatile("pand %mm0,%mm7");
			asm volatile("pxor %mm5,%mm4");
			asm volatile("pxor %mm7,%mm6");
			asm volatile("movq %0,%%mm5" : : "m" (dptr[z][d]));
			asm volatile("movq %0,%%mm7" : : "m" (dptr[z][d+8]));
			asm volatile("pxor %mm5,%mm2");
			asm volatile("pxor %mm7,%mm3");
			asm volatile("pxor %mm5,%mm4");
			asm volatile("pxor %mm7,%mm6");
			asm volatile("pxor %mm5,%mm5");
			asm volatile("pxor %mm7,%mm7");
		}
		asm volatile("movntq %%mm2,%0" : "=m" (p[d]));
		asm volatile("movntq %%mm3,%0" : "=m" (p[d+8]));
		asm volatile("movntq %%mm4,%0" : "=m" (q[d]));
		asm volatile("movntq %%mm6,%0" : "=m" (q[d+8]));
	}

	raid6_after_mmx(&sa);
	asm volatile("sfence" : :: "memory");
}

const struct raid6_calls raid6_sse1x2 = {
	raid6_sse12_gen_syndrome,
	raid6_have_sse1_or_mmxext,
	"sse1x2",
	1			/* Has cache hints */
};

#endif