* PATCH: Altivec support for RAID-6
@ 2004-11-18 6:52 H. Peter Anvin
2004-11-18 7:01 ` Andrew Morton
2004-11-18 20:23 ` Antonio Vargas
0 siblings, 2 replies; 6+ messages in thread
From: H. Peter Anvin @ 2004-11-18 6:52 UTC (permalink / raw)
To: Andrew Morton; +Cc: linux-kernel
[-- Attachment #1: Type: text/plain, Size: 316 bytes --]
This patch adds Altivec support for RAID-6, if appropriately configured
on the ppc or ppc64 architectures. Note that it changes the compile
flags for ppc64 in order to handle -maltivec correctly; this change was
vetted on the ppc64 mailing list and OK'd by paulus.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
[-- Attachment #2: raid6altivec.diff --]
[-- Type: text/x-patch, Size: 7285 bytes --]
Index: raid6/arch/ppc64/Makefile
===================================================================
RCS file: /home/hpa/kernel/bkcvs/linux-2.5/arch/ppc64/Makefile,v
retrieving revision 1.42
diff -u -r1.42 Makefile
--- raid6/arch/ppc64/Makefile 26 Oct 2004 14:53:34 -0000 1.42
+++ raid6/arch/ppc64/Makefile 18 Nov 2004 06:01:37 -0000
@@ -35,7 +35,11 @@
CFLAGS += -msoft-float -pipe -mminimal-toc -mtraceback=none
ifeq ($(CONFIG_POWER4_ONLY),y)
+ifeq ($(CONFIG_ALTIVEC),y)
+ CFLAGS += $(call cc-option,-mcpu=970)
+else
CFLAGS += $(call cc-option,-mcpu=power4)
+endif
else
CFLAGS += $(call cc-option,-mtune=power4)
endif
Index: raid6/drivers/md/Makefile
===================================================================
RCS file: /home/hpa/kernel/bkcvs/linux-2.5/drivers/md/Makefile,v
retrieving revision 1.23
diff -u -r1.23 Makefile
--- raid6/drivers/md/Makefile 11 Nov 2004 21:49:10 -0000 1.23
+++ raid6/drivers/md/Makefile 18 Nov 2004 06:01:38 -0000
@@ -9,6 +9,8 @@
raid6-objs := raid6main.o raid6algos.o raid6recov.o raid6tables.o \
raid6int1.o raid6int2.o raid6int4.o \
raid6int8.o raid6int16.o raid6int32.o \
+ raid6altivec1.o raid6altivec2.o raid6altivec4.o \
+ raid6altivec8.o \
raid6mmx.o raid6sse1.o raid6sse2.o
hostprogs-y := mktables
@@ -36,6 +38,10 @@
cmd_unroll = $(PERL) $(srctree)/$(src)/unroll.pl $(UNROLL) \
< $< > $@ || ( rm -f $@ && exit 1 )
+ifeq ($(CONFIG_ALTIVEC),y)
+altivec_flags := -maltivec -mabi=altivec
+endif
+
targets += raid6int1.c
$(obj)/raid6int1.c: UNROLL := 1
$(obj)/raid6int1.c: $(src)/raid6int.uc $(src)/unroll.pl FORCE
@@ -66,6 +72,30 @@
$(obj)/raid6int32.c: $(src)/raid6int.uc $(src)/unroll.pl FORCE
$(call if_changed,unroll)
+CFLAGS_raid6altivec1.o += $(altivec_flags)
+targets += raid6altivec1.c
+$(obj)/raid6altivec1.c: UNROLL := 1
+$(obj)/raid6altivec1.c: $(src)/raid6altivec.uc $(src)/unroll.pl FORCE
+ $(call if_changed,unroll)
+
+CFLAGS_raid6altivec2.o += $(altivec_flags)
+targets += raid6altivec2.c
+$(obj)/raid6altivec2.c: UNROLL := 2
+$(obj)/raid6altivec2.c: $(src)/raid6altivec.uc $(src)/unroll.pl FORCE
+ $(call if_changed,unroll)
+
+CFLAGS_raid6altivec4.o += $(altivec_flags)
+targets += raid6altivec4.c
+$(obj)/raid6altivec4.c: UNROLL := 4
+$(obj)/raid6altivec4.c: $(src)/raid6altivec.uc $(src)/unroll.pl FORCE
+ $(call if_changed,unroll)
+
+CFLAGS_raid6altivec8.o += $(altivec_flags)
+targets += raid6altivec8.c
+$(obj)/raid6altivec8.c: UNROLL := 8
+$(obj)/raid6altivec8.c: $(src)/raid6altivec.uc $(src)/unroll.pl FORCE
+ $(call if_changed,unroll)
+
quiet_cmd_mktable = TABLE $@
cmd_mktable = $(obj)/mktables > $@ || ( rm -f $@ && exit 1 )
Index: raid6/drivers/md/raid6algos.c
===================================================================
RCS file: /home/hpa/kernel/bkcvs/linux-2.5/drivers/md/raid6algos.c,v
retrieving revision 1.4
diff -u -r1.4 raid6algos.c
--- raid6/drivers/md/raid6algos.c 23 Jan 2004 16:04:36 -0000 1.4
+++ raid6/drivers/md/raid6algos.c 18 Nov 2004 06:01:38 -0000
@@ -37,6 +37,10 @@
extern const struct raid6_calls raid6_sse2x1;
extern const struct raid6_calls raid6_sse2x2;
extern const struct raid6_calls raid6_sse2x4;
+extern const struct raid6_calls raid6_altivec1;
+extern const struct raid6_calls raid6_altivec2;
+extern const struct raid6_calls raid6_altivec4;
+extern const struct raid6_calls raid6_altivec8;
const struct raid6_calls * const raid6_algos[] = {
&raid6_intx1,
@@ -60,6 +64,12 @@
&raid6_sse2x2,
&raid6_sse2x4,
#endif
+#ifdef CONFIG_ALTIVEC
+ &raid6_altivec1,
+ &raid6_altivec2,
+ &raid6_altivec4,
+ &raid6_altivec8,
+#endif
NULL
};
Index: raid6/drivers/md/raid6altivec.uc
===================================================================
RCS file: raid6/drivers/md/raid6altivec.uc
diff -N raid6/drivers/md/raid6altivec.uc
--- /dev/null 1 Jan 1970 00:00:00 -0000
+++ raid6/drivers/md/raid6altivec.uc 18 Nov 2004 06:01:38 -0000
@@ -0,0 +1,122 @@
+/* -*- linux-c -*- ------------------------------------------------------- *
+ *
+ * Copyright 2002-2004 H. Peter Anvin - All Rights Reserved
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, Inc., 53 Temple Place Ste 330,
+ * Bostom MA 02111-1307, USA; either version 2 of the License, or
+ * (at your option) any later version; incorporated herein by reference.
+ *
+ * ----------------------------------------------------------------------- */
+
+/*
+ * raid6altivec$#.c
+ *
+ * $#-way unrolled portable integer math RAID-6 instruction set
+ *
+ * This file is postprocessed using unroll.pl
+ *
+ * <benh> hpa: in process,
+ * you can just "steal" the vec unit with enable_kernel_altivec() (but
+ * bracked this with preempt_disable/enable or in a lock)
+ */
+
+#include "raid6.h"
+
+#ifdef CONFIG_ALTIVEC
+
+#include <altivec.h>
+#include <asm/system.h>
+#include <asm/cputable.h>
+
+/*
+ * This is the C data type to use
+ */
+
+typedef vector unsigned char unative_t;
+
+#define NBYTES(x) ((vector unsigned char) {x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x})
+#define NSIZE sizeof(unative_t)
+
+/*
+ * The SHLBYTE() operation shifts each byte left by 1, *not*
+ * rolling over into the next byte
+ */
+static inline __attribute_const__ unative_t SHLBYTE(unative_t v)
+{
+ return vec_add(v,v);
+}
+
+/*
+ * The MASK() operation returns 0xFF in any byte for which the high
+ * bit is 1, 0x00 for any byte for which the high bit is 0.
+ */
+static inline __attribute_const__ unative_t MASK(unative_t v)
+{
+ unative_t zv = NBYTES(0);
+
+ /* vec_cmpgt returns a vector bool char; thus the need for the cast */
+ return (unative_t)vec_cmpgt(zv, v);
+}
+
+
+/* This is noinline to make damned sure that gcc doesn't move any of the
+ Altivec code around the enable/disable code */
+static void noinline
+raid6_altivec$#_gen_syndrome_real(int disks, size_t bytes, void **ptrs)
+{
+ u8 **dptr = (u8 **)ptrs;
+ u8 *p, *q;
+ int d, z, z0;
+
+ unative_t wd$$, wq$$, wp$$, w1$$, w2$$;
+ unative_t x1d = NBYTES(0x1d);
+
+ z0 = disks - 3; /* Highest data disk */
+ p = dptr[z0+1]; /* XOR parity */
+ q = dptr[z0+2]; /* RS syndrome */
+
+ for ( d = 0 ; d < bytes ; d += NSIZE*$# ) {
+ wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE];
+ for ( z = z0-1 ; z >= 0 ; z-- ) {
+ wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE];
+ wp$$ = vec_xor(wp$$, wd$$);
+ w2$$ = MASK(wq$$);
+ w1$$ = SHLBYTE(wq$$);
+ w2$$ = vec_and(w2$$, x1d);
+ w1$$ = vec_xor(w1$$, w2$$);
+ wq$$ = vec_xor(w1$$, wd$$);
+ }
+ *(unative_t *)&p[d+NSIZE*$$] = wp$$;
+ *(unative_t *)&q[d+NSIZE*$$] = wq$$;
+ }
+}
+
+static void raid6_altivec$#_gen_syndrome(int disks, size_t bytes, void **ptrs)
+{
+ preempt_disable();
+ enable_kernel_altivec();
+
+ raid6_altivec$#_gen_syndrome_real(disks, bytes, ptrs);
+
+ preempt_enable();
+}
+
+int raid6_have_altivec(void);
+#if $# == 1
+int raid6_have_altivec(void)
+{
+ /* This assumes either all CPUs have Altivec or none does */
+ return cur_cpu_spec->cpu_features & CPU_FTR_ALTIVEC;
+}
+#endif
+
+const struct raid6_calls raid6_altivec$# = {
+ raid6_altivec$#_gen_syndrome,
+ raid6_have_altivec,
+ "altivecx$#",
+ 0
+};
+
+#endif /* CONFIG_ALTIVEC */
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: PATCH: Altivec support for RAID-6
2004-11-18 6:52 PATCH: Altivec support for RAID-6 H. Peter Anvin
@ 2004-11-18 7:01 ` Andrew Morton
2004-11-18 7:03 ` Andrew Morton
2004-11-18 20:23 ` Antonio Vargas
1 sibling, 1 reply; 6+ messages in thread
From: Andrew Morton @ 2004-11-18 7:01 UTC (permalink / raw)
To: H. Peter Anvin; +Cc: linux-kernel
"H. Peter Anvin" <hpa@zytor.com> wrote:
>
> $(obj)/raid6int1.c: $(src)/raid6int.uc $(src)/unroll.pl FORCE
So we require that raid6int.uc propagate through the system with the x bit
set.
I wonder if it would be safer to stick a $(SHELL) in there?
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: PATCH: Altivec support for RAID-6
2004-11-18 7:01 ` Andrew Morton
@ 2004-11-18 7:03 ` Andrew Morton
0 siblings, 0 replies; 6+ messages in thread
From: Andrew Morton @ 2004-11-18 7:03 UTC (permalink / raw)
To: hpa, linux-kernel
Andrew Morton <akpm@osdl.org> wrote:
>
> "H. Peter Anvin" <hpa@zytor.com> wrote:
> >
> > $(obj)/raid6int1.c: $(src)/raid6int.uc $(src)/unroll.pl FORCE
>
> So we require that raid6int.uc propagate through the system with the x bit
> set.
err, I think I misread that. Ignore.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: PATCH: Altivec support for RAID-6
2004-11-18 6:52 PATCH: Altivec support for RAID-6 H. Peter Anvin
2004-11-18 7:01 ` Andrew Morton
@ 2004-11-18 20:23 ` Antonio Vargas
2004-11-19 5:05 ` H. Peter Anvin
1 sibling, 1 reply; 6+ messages in thread
From: Antonio Vargas @ 2004-11-18 20:23 UTC (permalink / raw)
To: H. Peter Anvin; +Cc: Andrew Morton, linux-kernel
On Wed, 17 Nov 2004 22:52:55 -0800, H. Peter Anvin <hpa@zytor.com> wrote:
> This patch adds Altivec support for RAID-6, if appropriately configured
> on the ppc or ppc64 architectures. Note that it changes the compile
> flags for ppc64 in order to handle -maltivec correctly; this change was
> vetted on the ppc64 mailing list and OK'd by paulus.
>
> Signed-off-by: H. Peter Anvin <hpa@zytor.com>
>
hpa, are you aware of any other routines which should benefit from altivec?
--
Greetz, Antonio Vargas aka winden of network
Las cosas no son lo que parecen, excepto cuando parecen lo que si son.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: PATCH: Altivec support for RAID-6
2004-11-18 20:23 ` Antonio Vargas
@ 2004-11-19 5:05 ` H. Peter Anvin
[not found] ` <69304d1104111904046d02b5bd@mail.gmail.com>
0 siblings, 1 reply; 6+ messages in thread
From: H. Peter Anvin @ 2004-11-19 5:05 UTC (permalink / raw)
To: linux-kernel
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.
-hpa
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: PATCH: Altivec support for RAID-6
[not found] ` <69304d1104111904046d02b5bd@mail.gmail.com>
@ 2004-11-19 20:59 ` Antonio Vargas
0 siblings, 0 replies; 6+ messages in thread
From: Antonio Vargas @ 2004-11-19 20:59 UTC (permalink / raw)
To: H. Peter Anvin, lkml
[-- 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
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2004-11-19 21:02 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2004-11-18 6:52 PATCH: Altivec support for RAID-6 H. Peter Anvin
2004-11-18 7:01 ` Andrew Morton
2004-11-18 7:03 ` Andrew Morton
2004-11-18 20:23 ` Antonio Vargas
2004-11-19 5:05 ` H. Peter Anvin
[not found] ` <69304d1104111904046d02b5bd@mail.gmail.com>
2004-11-19 20:59 ` Antonio Vargas
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).