Re: PATCH: Altivec support for RAID-6

[Antonio Vargas <windenntw@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 951 bytes --]

On Fri, 19 Nov 2004 13:04:11 +0100, Antonio Vargas <windenntw@gmail.com> wrote:
> On Fri, 19 Nov 2004 05:05:54 +0000 (UTC), H. Peter Anvin <hpa@zytor.com> wrote:
> > Followup to:  <69304d1104111812234656a606@mail.gmail.com>
> > By author:    Antonio Vargas <windenntw@gmail.com>
> > In newsgroup: linux.dev.kernel
> > >
> > > hpa, are you aware of any other routines which should benefit from altivec?
> > >
> >
> > Presumably the XOR code used by RAID-5, and quite possibly some of the
> > cryptography stuff.  Unlike most SIMD instruction sets, it should be
> > possible to write AES using Altivec.
> 
> I'll take a crack at the RAID-5 stuff first then.
> 

Here we go... this file is compile-tested on gcc 3.3 from userspace,
it can serve as an starting point for testing.

Signed-off-by: Antonio Vargas <windenntw@gmail.com>

-- 
Greetz, Antonio Vargas aka winden of network

Las cosas no son lo que parecen, excepto cuando parecen lo que si son.

[-- Attachment #2: xor.h --]
[-- Type: text/plain, Size: 3228 bytes --]

/*
 * include/asm-ppc/xor.h
 *
 * Altivec optimized RAID-5 checksumming functions.
 * Copyright 2004 Antonio Vargas, windenntw@gmail.com
 *
 * Based on include/asm-generic/xor.h and drivers/md/raid6altivec.uc
 *
 * 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, version 2.
 *
 * You should have received a copy of the GNU General Public License
 * (for example /usr/src/linux/COPYING); if not, write to the Free
 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */

#ifdef KERNEL
#include <asm-generic/xor.h>
#endif

#ifdef CONFIG_ALTIVEC

#ifdef KERNEL
#include <altivec.h>
#include <asm/cputable.h>
#include <asm/processor.h>
#include <asm/system.h>
#else
#define prefetch(x)
#define prefetchw(x)
void enable_kernel_altivec(){};
void preempt_enable(){};
void preempt_disable(){};
#endif

typedef vector unsigned char unative_t;

#define BODY\
	long lines = bytes / (4 * sizeof(unative_t)) - 1;\
	int i;\
	preempt_disable();\
	enable_kernel_altivec();\
	PREF(0)\
	do {\
		PREF(4);\
		CALC(0);\
		CALC(1);\
		CALC(2);\
		CALC(3);\
		INC;\
	} while (--lines > 0);\
	for(i = 0 ; i < 4 ; i++)\
		CALC(i);\
	preempt_enable();


#define PREF2(x) prefetchw(p1 + x); prefetch(p2  + x);
#define PREF3(x) PREF2(x) prefetch(p3  + x);
#define PREF4(x) PREF3(x) prefetch(p4  + x);
#define PREF5(x) PREF4(x) prefetch(p5  + x);

#define CALC2(x) p1[x] ^= p2[x]
#define CALC3(x) CALC2(x) ^ p3[x]
#define CALC4(x) CALC3(x) ^ p4[x]
#define CALC5(x) CALC4(x) ^ p5[x]

#define INC2 p1 += 4; p2 += 4;
#define INC3 INC2 p3 += 4;
#define INC4 INC3 p4 += 4;
#define INC5 INC4 p5 += 4;

static void
xor_altivec2(unsigned long bytes, unative_t *p1, unative_t *p2)
{
#undef PREF
#undef CALC
#undef INC
#define PREF(x) PREF2(x)
#define CALC(x) CALC2(x)
#define INC     INC2
	BODY
}

static void
xor_altivec3(unsigned long bytes, unative_t *p1, unative_t *p2, unative_t *p3)
{
#undef PREF
#undef CALC
#undef INC
#define PREF(x) PREF3(x)
#define CALC(x) CALC3(x)
#define INC     INC3
	BODY
}


static void
xor_altivec4(unsigned long bytes, unative_t *p1, unative_t *p2, unative_t *p3, unative_t *p4)
{
#undef PREF
#undef CALC
#undef INC
#define PREF(x) PREF4(x)
#define CALC(x) CALC4(x)
#define INC     INC4
	BODY
}

static void
xor_altivec5(unsigned long bytes, unative_t *p1, unative_t *p2, unative_t *p3, unative_t *p4, unative_t *p5)
{
#undef PREF
#undef CALC
#undef INC
#define PREF(x) PREF5(x)
#define CALC(x) CALC5(x)
#define INC     INC5
	BODY
}

#ifdef KERNEL
static struct xor_block_template xor_block_altivec = {
	.name = "altivec",
	.do_2 = xor_altivec2,
	.do_3 = xor_altivec3,
	.do_4 = xor_altivec4,
	.do_5 = xor_altivec5,
};

#undef  XOR_TRY_TEMPLATES
#define XOR_TRY_TEMPLATES                       \
	do {                                    \
		xor_speed(&xor_block_8regs);    \
		xor_speed(&xor_block_8regs_p);  \
		xor_speed(&xor_block_32regs);   \
		xor_speed(&xor_block_32regs_p); \
		xor_speed(&xor_block_altivec);  \
	} while (0)
#endif

#endif