d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add parity generation/rebuild using 128-bits NEON for Aarch64

Browse Source 
This re-use the framework established for SSE2, SSSE3 and
AVX2. However, GCC is using FP registers on Aarch64, so
unlike SSE/AVX2 we can't rely on the registers being left alone
between ASM statements. So instead, the NEON code uses
C variables and GCC extended ASM syntax. Note that since
the kernel explicitly disable vector registers, they
have to be locally re-enabled explicitly.

As we use the variable's number to define the symbolic
name, and GCC won't allow duplicate symbolic names,
numbers have to be unique. Even when the code is not
going to be used (e.g. the case for 4 registers when
using the macro with only 2). Only the actually used
variables should be declared, otherwise the build
will fails in debug mode.

This requires the replacement of the XOR(X,X) syntax
by a new ZERO(X) macro, which does the same thing but
without repeating the argument. And perhaps someday
there will be a machine where there is a more efficient
way to zero a register than XOR with itself. This affects
scalar, SSE2, SSSE3 and AVX2 as they need the new macro.

It's possible to write faster implementations (different
scheduling, different unrolling, interleaving NEON and
scalar, ...) for various cores, but this one has the
advantage of fitting in the current state of the code,
and thus is likely easier to review/check/merge.

The only difference between aarch64-neon and aarch64-neonx2
is that aarch64-neonx2 unroll some functions some more.

Reviewed-by: Gvozden Neskovic <neskovic@gmail.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Romain Dolbeau <romain.dolbeau@atos.net>
Closes #4801
This commit is contained in:
Romain Dolbeau 2016-10-03 18:44:00 +02:00 committed by Brian Behlendorf
parent d1502e9ed0
commit 62a65a654e
16 changed files with 3216 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
cmd/raidz_test/raidz_test.h
View File

@ -34,6 +34,8 @@ static const char *raidz_impl_names[] = {
"sse2",
"ssse3",
"avx2",
"aarch64_neon",
"aarch64_neonx2",
NULL
};

1
include/linux/Makefile.am
View File

@ -8,6 +8,7 @@ KERNEL_H = \
$(top_srcdir)/include/linux/utsname_compat.h \
$(top_srcdir)/include/linux/kmap_compat.h \
$(top_srcdir)/include/linux/simd_x86.h \
$(top_srcdir)/include/linux/simd_aarch64.h \
$(top_srcdir)/include/linux/mod_compat.h
USER_H =

62
include/linux/simd_aarch64.h Normal file
View File

@ -0,0 +1,62 @@
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright (C) 2016 Romain Dolbeau <romain@dolbeau.org>.
*/
/*
* USER API:
*
* Kernel fpu methods:
* kfpu_begin()
* kfpu_end()
*/
#ifndef _SIMD_AARCH64_H
#define _SIMD_AARCH64_H
#include <sys/isa_defs.h>
#if defined(__aarch64__)
#include <sys/types.h>
#if defined(_KERNEL)
#include <asm/neon.h>
#define kfpu_begin() \
{ \
kernel_neon_begin(); \
}
#define kfpu_end() \
{ \
kernel_neon_end(); \
}
#else
/*
* fpu dummy methods for userspace
*/
#define kfpu_begin() do {} while (0)
#define kfpu_end() do {} while (0)
#endif /* defined(_KERNEL) */
#endif /* __aarch64__ */
#endif /* _SIMD_AARCH64_H */

4
include/sys/vdev_raidz_impl.h
View File

@ -141,6 +141,10 @@ extern const raidz_impl_ops_t vdev_raidz_ssse3_impl;
#if defined(__x86_64) && defined(HAVE_AVX2) /* only x86_64 for now */
extern const raidz_impl_ops_t vdev_raidz_avx2_impl;
#endif
#if defined(__aarch64__)
extern const raidz_impl_ops_t vdev_raidz_aarch64_neon_impl;
extern const raidz_impl_ops_t vdev_raidz_aarch64_neonx2_impl;
#endif
/*
* Commonly used raidz_map helpers

2
lib/libzpool/Makefile.am
View File

@ -99,6 +99,8 @@ KERNEL_C = \
vdev_raidz_math_sse2.c \
vdev_raidz_math_ssse3.c \
vdev_raidz_math_avx2.c \
vdev_raidz_math_aarch64_neon.c \
vdev_raidz_math_aarch64_neonx2.c \
vdev_root.c \
zap.c \
zap_leaf.c \

2
man/man5/zfs-module-parameters.5
View File

@ -1763,6 +1763,8 @@ Possible options are:
sse2 - implementation using SSE2 instruction set (64bit x86 only)
ssse3 - implementation using SSSE3 instruction set (64bit x86 only)
avx2 - implementation using AVX2 instruction set (64bit x86 only)
aarch64_neon - implementation using NEON (Aarch64/64 bit ARMv8 only)
aarch64_neonx2 - implementation using NEON with more unrolling (Aarch64/64 bit ARMv8 only)
.sp
Default value: \fBfastest\fR.
.RE

3
module/zfs/Makefile.in
View File

@ -115,3 +115,6 @@ $(MODULE)-objs += dsl_userhold.o
$(MODULE)-$(CONFIG_X86) += vdev_raidz_math_sse2.o
$(MODULE)-$(CONFIG_X86) += vdev_raidz_math_ssse3.o
$(MODULE)-$(CONFIG_X86) += vdev_raidz_math_avx2.o
$(MODULE)-$(CONFIG_ARM64) += vdev_raidz_math_aarch64_neon.o
$(MODULE)-$(CONFIG_ARM64) += vdev_raidz_math_aarch64_neonx2.o

18
module/zfs/vdev_raidz_math.c
View File

@ -55,7 +55,11 @@ const raidz_impl_ops_t *raidz_all_maths[] = {
&vdev_raidz_ssse3_impl,
#endif
#if defined(__x86_64) && defined(HAVE_AVX2) /* only x86_64 for now */
&vdev_raidz_avx2_impl
&vdev_raidz_avx2_impl,
#endif
#if defined(__aarch64__)
&vdev_raidz_aarch64_neon_impl,
&vdev_raidz_aarch64_neonx2_impl,
#endif
};
@ -275,11 +279,11 @@ raidz_math_kstat_headers(char *buf, size_t size)
off = snprintf(buf, size, "%-17s", "implementation");
for (i = 0; i < ARRAY_SIZE(raidz_gen_name); i++)
off += snprintf(buf + off, size - off, "%-12s",
off += snprintf(buf + off, size - off, "%-16s",
raidz_gen_name[i]);
for (i = 0; i < ARRAY_SIZE(raidz_rec_name); i++)
off += snprintf(buf + off, size - off, "%-12s",
off += snprintf(buf + off, size - off, "%-16s",
raidz_rec_name[i]);
(void) snprintf(buf + off, size - off, "\n");
@ -302,12 +306,12 @@ raidz_math_kstat_data(char *buf, size_t size, void *data)
for (i = 0; i < ARRAY_SIZE(raidz_gen_name); i++) {
int id = fstat->gen[i];
off += snprintf(buf + off, size - off, "%-12s",
off += snprintf(buf + off, size - off, "%-16s",
raidz_supp_impl[id]->name);
}
for (i = 0; i < ARRAY_SIZE(raidz_rec_name); i++) {
int id = fstat->rec[i];
off += snprintf(buf + off, size - off, "%-12s",
off += snprintf(buf + off, size - off, "%-16s",
raidz_supp_impl[id]->name);
}
} else {
@ -317,11 +321,11 @@ raidz_math_kstat_data(char *buf, size_t size, void *data)
raidz_supp_impl[id]->name);
for (i = 0; i < ARRAY_SIZE(raidz_gen_name); i++)
off += snprintf(buf + off, size - off, "%-12llu",
off += snprintf(buf + off, size - off, "%-16llu",
(u_longlong_t) cstat->gen[i]);
for (i = 0; i < ARRAY_SIZE(raidz_rec_name); i++)
off += snprintf(buf + off, size - off, "%-12llu",
off += snprintf(buf + off, size - off, "%-16llu",
(u_longlong_t) cstat->rec[i]);
}

2210
module/zfs/vdev_raidz_math_aarch64_neon.c Normal file
View File

File diff suppressed because it is too large Load Diff

685
module/zfs/vdev_raidz_math_aarch64_neon_common.h Normal file
View File

@ -0,0 +1,685 @@
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright (C) 2016 Romain Dolbeau. All rights reserved.
*/
#include <sys/types.h>
#include <linux/simd_aarch64.h>
#define __asm __asm__ __volatile__
#define _REG_CNT(_0, _1, _2, _3, _4, _5, _6, _7, N, ...) N
#define REG_CNT(r...) _REG_CNT(r, 8, 7, 6, 5, 4, 3, 2, 1)
#define VR0_(REG, ...) "%[w"#REG"]"
#define VR1_(_1, REG, ...) "%[w"#REG"]"
#define VR2_(_1, _2, REG, ...) "%[w"#REG"]"
#define VR3_(_1, _2, _3, REG, ...) "%[w"#REG"]"
#define VR4_(_1, _2, _3, _4, REG, ...) "%[w"#REG"]"
#define VR5_(_1, _2, _3, _4, _5, REG, ...) "%[w"#REG"]"
#define VR6_(_1, _2, _3, _4, _5, _6, REG, ...) "%[w"#REG"]"
#define VR7_(_1, _2, _3, _4, _5, _6, _7, REG, ...) "%[w"#REG"]"
/*
* Here we need registers not used otherwise.
* They will be used in unused ASM for the case
* with more registers than required... but GGC
* will still need to make sure the constraints
* are correct, and duplicate constraints are illegal
* ... and we use the "register" number as a name
*/
#define VR0(r...) VR0_(r)
#define VR1(r...) VR1_(r)
#define VR2(r...) VR2_(r, 36)
#define VR3(r...) VR3_(r, 36, 35)
#define VR4(r...) VR4_(r, 36, 35, 34, 33)
#define VR5(r...) VR5_(r, 36, 35, 34, 33, 32)
#define VR6(r...) VR6_(r, 36, 35, 34, 33, 32, 31)
#define VR7(r...) VR7_(r, 36, 35, 34, 33, 32, 31, 30)
#define VR(X) "%[w"#X"]"
#define RVR0_(REG, ...) [w##REG] "w" (w##REG)
#define RVR1_(_1, REG, ...) [w##REG] "w" (w##REG)
#define RVR2_(_1, _2, REG, ...) [w##REG] "w" (w##REG)
#define RVR3_(_1, _2, _3, REG, ...) [w##REG] "w" (w##REG)
#define RVR4_(_1, _2, _3, _4, REG, ...) [w##REG] "w" (w##REG)
#define RVR5_(_1, _2, _3, _4, _5, REG, ...) [w##REG] "w" (w##REG)
#define RVR6_(_1, _2, _3, _4, _5, _6, REG, ...) [w##REG] "w" (w##REG)
#define RVR7_(_1, _2, _3, _4, _5, _6, _7, REG, ...) [w##REG] "w" (w##REG)
#define RVR0(r...) RVR0_(r)
#define RVR1(r...) RVR1_(r)
#define RVR2(r...) RVR2_(r, 36)
#define RVR3(r...) RVR3_(r, 36, 35)
#define RVR4(r...) RVR4_(r, 36, 35, 34, 33)
#define RVR5(r...) RVR5_(r, 36, 35, 34, 33, 32)
#define RVR6(r...) RVR6_(r, 36, 35, 34, 33, 32, 31)
#define RVR7(r...) RVR7_(r, 36, 35, 34, 33, 32, 31, 30)
#define RVR(X) [w##X] "w" (w##X)
#define WVR0_(REG, ...) [w##REG] "=w" (w##REG)
#define WVR1_(_1, REG, ...) [w##REG] "=w" (w##REG)
#define WVR2_(_1, _2, REG, ...) [w##REG] "=w" (w##REG)
#define WVR3_(_1, _2, _3, REG, ...) [w##REG] "=w" (w##REG)
#define WVR4_(_1, _2, _3, _4, REG, ...) [w##REG] "=w" (w##REG)
#define WVR5_(_1, _2, _3, _4, _5, REG, ...) [w##REG] "=w" (w##REG)
#define WVR6_(_1, _2, _3, _4, _5, _6, REG, ...) [w##REG] "=w" (w##REG)
#define WVR7_(_1, _2, _3, _4, _5, _6, _7, REG, ...) [w##REG] "=w" (w##REG)
#define WVR0(r...) WVR0_(r)
#define WVR1(r...) WVR1_(r)
#define WVR2(r...) WVR2_(r, 36)
#define WVR3(r...) WVR3_(r, 36, 35)
#define WVR4(r...) WVR4_(r, 36, 35, 34, 33)
#define WVR5(r...) WVR5_(r, 36, 35, 34, 33, 32)
#define WVR6(r...) WVR6_(r, 36, 35, 34, 33, 32, 31)
#define WVR7(r...) WVR7_(r, 36, 35, 34, 33, 32, 31, 30)
#define WVR(X) [w##X] "=w" (w##X)
#define UVR0_(REG, ...) [w##REG] "+&w" (w##REG)
#define UVR1_(_1, REG, ...) [w##REG] "+&w" (w##REG)
#define UVR2_(_1, _2, REG, ...) [w##REG] "+&w" (w##REG)
#define UVR3_(_1, _2, _3, REG, ...) [w##REG] "+&w" (w##REG)
#define UVR4_(_1, _2, _3, _4, REG, ...) [w##REG] "+&w" (w##REG)
#define UVR5_(_1, _2, _3, _4, _5, REG, ...) [w##REG] "+&w" (w##REG)
#define UVR6_(_1, _2, _3, _4, _5, _6, REG, ...) [w##REG] "+&w" (w##REG)
#define UVR7_(_1, _2, _3, _4, _5, _6, _7, REG, ...) [w##REG] "+&w" (w##REG)
#define UVR0(r...) UVR0_(r)
#define UVR1(r...) UVR1_(r)
#define UVR2(r...) UVR2_(r, 36)
#define UVR3(r...) UVR3_(r, 36, 35)
#define UVR4(r...) UVR4_(r, 36, 35, 34, 33)
#define UVR5(r...) UVR5_(r, 36, 35, 34, 33, 32)
#define UVR6(r...) UVR6_(r, 36, 35, 34, 33, 32, 31)
#define UVR7(r...) UVR7_(r, 36, 35, 34, 33, 32, 31, 30)
#define UVR(X) [w##X] "+&w" (w##X)
#define R_01(REG1, REG2, ...) REG1, REG2
#define _R_23(_0, _1, REG2, REG3, ...) REG2, REG3
#define R_23(REG...) _R_23(REG, 1, 2, 3)
#define ASM_BUG() ASSERT(0)
#define OFFSET(ptr, val) (((unsigned char *)ptr)+val)
extern const uint8_t gf_clmul_mod_lt[4*256][16];
#define ELEM_SIZE 16
typedef struct v {
uint8_t b[ELEM_SIZE] __attribute__((aligned(ELEM_SIZE)));
} v_t;
#define PREFETCHNTA(ptr, offset) \
{ \
__asm( \
"prfm pstl1strm, %[MEM]\n" \
: : [MEM] "Q" (*(ptr + offset))); \
}
#define PREFETCH(ptr, offset) \
{ \
__asm( \
"prfm pldl1keep, %[MEM]\n" \
: : [MEM] "Q" (*(ptr + offset))); \
}
#define XOR_ACC(src, r...) \
{ \
switch (REG_CNT(r)) { \
case 8: \
__asm( \
"ld1 { v21.4s },%[SRC0]\n" \
"ld1 { v20.4s },%[SRC1]\n" \
"ld1 { v19.4s },%[SRC2]\n" \
"ld1 { v18.4s },%[SRC3]\n" \
"eor " VR0(r) ".16b," VR0(r) ".16b,v21.16b\n" \
"eor " VR1(r) ".16b," VR1(r) ".16b,v20.16b\n" \
"eor " VR2(r) ".16b," VR2(r) ".16b,v19.16b\n" \
"eor " VR3(r) ".16b," VR3(r) ".16b,v18.16b\n" \
"ld1 { v21.4s },%[SRC4]\n" \
"ld1 { v20.4s },%[SRC5]\n" \
"ld1 { v19.4s },%[SRC6]\n" \
"ld1 { v18.4s },%[SRC7]\n" \
"eor " VR4(r) ".16b," VR4(r) ".16b,v21.16b\n" \
"eor " VR5(r) ".16b," VR5(r) ".16b,v20.16b\n" \
"eor " VR6(r) ".16b," VR6(r) ".16b,v19.16b\n" \
"eor " VR7(r) ".16b," VR7(r) ".16b,v18.16b\n" \
: UVR0(r), UVR1(r), UVR2(r), UVR3(r), \
UVR4(r), UVR5(r), UVR6(r), UVR7(r) \
: [SRC0] "Q" (*(OFFSET(src, 0))), \
[SRC1] "Q" (*(OFFSET(src, 16))), \
[SRC2] "Q" (*(OFFSET(src, 32))), \
[SRC3] "Q" (*(OFFSET(src, 48))), \
[SRC4] "Q" (*(OFFSET(src, 64))), \
[SRC5] "Q" (*(OFFSET(src, 80))), \
[SRC6] "Q" (*(OFFSET(src, 96))), \
[SRC7] "Q" (*(OFFSET(src, 112))) \
: "v18", "v19", "v20", "v21"); \
break; \
case 4: \
__asm( \
"ld1 { v21.4s },%[SRC0]\n" \
"ld1 { v20.4s },%[SRC1]\n" \
"ld1 { v19.4s },%[SRC2]\n" \
"ld1 { v18.4s },%[SRC3]\n" \
"eor " VR0(r) ".16b," VR0(r) ".16b,v21.16b\n" \
"eor " VR1(r) ".16b," VR1(r) ".16b,v20.16b\n" \
"eor " VR2(r) ".16b," VR2(r) ".16b,v19.16b\n" \
"eor " VR3(r) ".16b," VR3(r) ".16b,v18.16b\n" \
: UVR0(r), UVR1(r), UVR2(r), UVR3(r) \
: [SRC0] "Q" (*(OFFSET(src, 0))), \
[SRC1] "Q" (*(OFFSET(src, 16))), \
[SRC2] "Q" (*(OFFSET(src, 32))), \
[SRC3] "Q" (*(OFFSET(src, 48))) \
: "v18", "v19", "v20", "v21"); \
break; \
case 2: \
__asm( \
"ld1 { v21.4s },%[SRC0]\n" \
"ld1 { v20.4s },%[SRC1]\n" \
"eor " VR0(r) ".16b," VR0(r) ".16b,v21.16b\n" \
"eor " VR1(r) ".16b," VR1(r) ".16b,v20.16b\n" \
: UVR0(r), UVR1(r) \
: [SRC0] "Q" (*(OFFSET(src, 0))), \
[SRC1] "Q" (*(OFFSET(src, 16))) \
: "v20", "v21"); \
break; \
default: \
ASM_BUG(); \
} \
}
#define XOR(r...) \
{ \
switch (REG_CNT(r)) { \
case 8: \
__asm( \
"eor " VR4(r) ".16b," VR4(r) ".16b," VR0(r) ".16b\n" \
"eor " VR5(r) ".16b," VR5(r) ".16b," VR1(r) ".16b\n" \
"eor " VR6(r) ".16b," VR6(r) ".16b," VR2(r) ".16b\n" \
"eor " VR7(r) ".16b," VR7(r) ".16b," VR3(r) ".16b\n" \
: UVR4(r), UVR5(r), UVR6(r), UVR7(r) \
: RVR0(r), RVR1(r), RVR2(r), RVR3(r)); \
break; \
case 4: \
__asm( \
"eor " VR2(r) ".16b," VR2(r) ".16b," VR0(r) ".16b\n" \
"eor " VR3(r) ".16b," VR3(r) ".16b," VR1(r) ".16b\n" \
: UVR2(r), UVR3(r) \
: RVR0(r), RVR1(r)); \
break; \
default: \
ASM_BUG(); \
} \
}
#define ZERO(r...) \
{ \
switch (REG_CNT(r)) { \
case 4: \
__asm( \
"eor " VR0(r) ".16b," VR0(r) ".16b," VR0(r) ".16b\n" \
"eor " VR1(r) ".16b," VR1(r) ".16b," VR1(r) ".16b\n" \
"eor " VR2(r) ".16b," VR2(r) ".16b," VR2(r) ".16b\n" \
"eor " VR3(r) ".16b," VR3(r) ".16b," VR3(r) ".16b\n" \
: WVR0(r), WVR1(r), WVR2(r), WVR3(r)); \
break; \
case 2: \
__asm( \
"eor " VR0(r) ".16b," VR0(r) ".16b," VR0(r) ".16b\n" \
"eor " VR1(r) ".16b," VR1(r) ".16b," VR1(r) ".16b\n" \
: WVR0(r), WVR1(r)); \
break; \
default: \
ASM_BUG(); \
} \
}
#define COPY(r...) \
{ \
switch (REG_CNT(r)) { \
case 8: \
__asm( \
"mov " VR4(r) ".16b," VR0(r) ".16b\n" \
"mov " VR5(r) ".16b," VR1(r) ".16b\n" \
"mov " VR6(r) ".16b," VR2(r) ".16b\n" \
"mov " VR7(r) ".16b," VR3(r) ".16b\n" \
: WVR4(r), WVR5(r), WVR6(r), WVR7(r) \
: RVR0(r), RVR1(r), RVR2(r), RVR3(r)); \
break; \
case 4: \
__asm( \
"mov " VR2(r) ".16b," VR0(r) ".16b\n" \
"mov " VR3(r) ".16b," VR1(r) ".16b\n" \
: WVR2(r), WVR3(r) \
: RVR0(r), RVR1(r)); \
break; \
default: \
ASM_BUG(); \
} \
}
#define LOAD(src, r...) \
{ \
switch (REG_CNT(r)) { \
case 8: \
__asm( \
"ld1 { " VR0(r) ".4s },%[SRC0]\n" \
"ld1 { " VR1(r) ".4s },%[SRC1]\n" \
"ld1 { " VR2(r) ".4s },%[SRC2]\n" \
"ld1 { " VR3(r) ".4s },%[SRC3]\n" \
"ld1 { " VR4(r) ".4s },%[SRC4]\n" \
"ld1 { " VR5(r) ".4s },%[SRC5]\n" \
"ld1 { " VR6(r) ".4s },%[SRC6]\n" \
"ld1 { " VR7(r) ".4s },%[SRC7]\n" \
: WVR0(r), WVR1(r), WVR2(r), WVR3(r), \
WVR4(r), WVR5(r), WVR6(r), WVR7(r) \
: [SRC0] "Q" (*(OFFSET(src, 0))), \
[SRC1] "Q" (*(OFFSET(src, 16))), \
[SRC2] "Q" (*(OFFSET(src, 32))), \
[SRC3] "Q" (*(OFFSET(src, 48))), \
[SRC4] "Q" (*(OFFSET(src, 64))), \
[SRC5] "Q" (*(OFFSET(src, 80))), \
[SRC6] "Q" (*(OFFSET(src, 96))), \
[SRC7] "Q" (*(OFFSET(src, 112)))); \
break; \
case 4: \
__asm( \
"ld1 { " VR0(r) ".4s },%[SRC0]\n" \
"ld1 { " VR1(r) ".4s },%[SRC1]\n" \
"ld1 { " VR2(r) ".4s },%[SRC2]\n" \
"ld1 { " VR3(r) ".4s },%[SRC3]\n" \
: WVR0(r), WVR1(r), WVR2(r), WVR3(r) \
: [SRC0] "Q" (*(OFFSET(src, 0))), \
[SRC1] "Q" (*(OFFSET(src, 16))), \
[SRC2] "Q" (*(OFFSET(src, 32))), \
[SRC3] "Q" (*(OFFSET(src, 48)))); \
break; \
case 2: \
__asm( \
"ld1 { " VR0(r) ".4s },%[SRC0]\n" \
"ld1 { " VR1(r) ".4s },%[SRC1]\n" \
: WVR0(r), WVR1(r) \
: [SRC0] "Q" (*(OFFSET(src, 0))), \
[SRC1] "Q" (*(OFFSET(src, 16)))); \
break; \
default: \
ASM_BUG(); \
} \
}
#define STORE(dst, r...) \
{ \
switch (REG_CNT(r)) { \
case 8: \
__asm( \
"st1 { " VR0(r) ".4s },%[DST0]\n" \
"st1 { " VR1(r) ".4s },%[DST1]\n" \
"st1 { " VR2(r) ".4s },%[DST2]\n" \
"st1 { " VR3(r) ".4s },%[DST3]\n" \
"st1 { " VR4(r) ".4s },%[DST4]\n" \
"st1 { " VR5(r) ".4s },%[DST5]\n" \
"st1 { " VR6(r) ".4s },%[DST6]\n" \
"st1 { " VR7(r) ".4s },%[DST7]\n" \
: [DST0] "=Q" (*(OFFSET(dst, 0))), \
[DST1] "=Q" (*(OFFSET(dst, 16))), \
[DST2] "=Q" (*(OFFSET(dst, 32))), \
[DST3] "=Q" (*(OFFSET(dst, 48))), \
[DST4] "=Q" (*(OFFSET(dst, 64))), \
[DST5] "=Q" (*(OFFSET(dst, 80))), \
[DST6] "=Q" (*(OFFSET(dst, 96))), \
[DST7] "=Q" (*(OFFSET(dst, 112))) \
: RVR0(r), RVR1(r), RVR2(r), RVR3(r), \
RVR4(r), RVR5(r), RVR6(r), RVR7(r)); \
break; \
case 4: \
__asm( \
"st1 { " VR0(r) ".4s },%[DST0]\n" \
"st1 { " VR1(r) ".4s },%[DST1]\n" \
"st1 { " VR2(r) ".4s },%[DST2]\n" \
"st1 { " VR3(r) ".4s },%[DST3]\n" \
: [DST0] "=Q" (*(OFFSET(dst, 0))), \
[DST1] "=Q" (*(OFFSET(dst, 16))), \
[DST2] "=Q" (*(OFFSET(dst, 32))), \
[DST3] "=Q" (*(OFFSET(dst, 48))) \
: RVR0(r), RVR1(r), RVR2(r), RVR3(r)); \
break; \
case 2: \
__asm( \
"st1 { " VR0(r) ".4s },%[DST0]\n" \
"st1 { " VR1(r) ".4s },%[DST1]\n" \
: [DST0] "=Q" (*(OFFSET(dst, 0))), \
[DST1] "=Q" (*(OFFSET(dst, 16))) \
: RVR0(r), RVR1(r)); \
break; \
default: \
ASM_BUG(); \
} \
}
/*
* Unfortunately cannot use the macro, because GCC
* will try to use the macro name and not value
* later on...
* Kept as a reference to what a numbered variable is
*/
#define _00 "v17"
#define _1d "v16"
#define _temp0 "v19"
#define _temp1 "v18"
#define MUL2_SETUP() \
{ \
__asm( \
"eor " VR(17) ".16b," VR(17) ".16b," VR(17) ".16b\n" \
"movi " VR(16) ".16b,#0x1d\n" \
: WVR(16), WVR(17)); \
}
#define MUL2(r...) \
{ \
switch (REG_CNT(r)) { \
case 4: \
__asm( \
"cmgt v19.16b," VR(17) ".16b," VR0(r) ".16b\n" \
"cmgt v18.16b," VR(17) ".16b," VR1(r) ".16b\n" \
"cmgt v21.16b," VR(17) ".16b," VR2(r) ".16b\n" \
"cmgt v20.16b," VR(17) ".16b," VR3(r) ".16b\n" \
"and v19.16b,v19.16b," VR(16) ".16b\n" \
"and v18.16b,v18.16b," VR(16) ".16b\n" \
"and v21.16b,v21.16b," VR(16) ".16b\n" \
"and v20.16b,v20.16b," VR(16) ".16b\n" \
"shl " VR0(r) ".16b," VR0(r) ".16b,#1\n" \
"shl " VR1(r) ".16b," VR1(r) ".16b,#1\n" \
"shl " VR2(r) ".16b," VR2(r) ".16b,#1\n" \
"shl " VR3(r) ".16b," VR3(r) ".16b,#1\n" \
"eor " VR0(r) ".16b,v19.16b," VR0(r) ".16b\n" \
"eor " VR1(r) ".16b,v18.16b," VR1(r) ".16b\n" \
"eor " VR2(r) ".16b,v21.16b," VR2(r) ".16b\n" \
"eor " VR3(r) ".16b,v20.16b," VR3(r) ".16b\n" \
: UVR0(r), UVR1(r), UVR2(r), UVR3(r) \
: RVR(17), RVR(16) \
: "v18", "v19", "v20", "v21"); \
break; \
case 2: \
__asm( \
"cmgt v19.16b," VR(17) ".16b," VR0(r) ".16b\n" \
"cmgt v18.16b," VR(17) ".16b," VR1(r) ".16b\n" \
"and v19.16b,v19.16b," VR(16) ".16b\n" \
"and v18.16b,v18.16b," VR(16) ".16b\n" \
"shl " VR0(r) ".16b," VR0(r) ".16b,#1\n" \
"shl " VR1(r) ".16b," VR1(r) ".16b,#1\n" \
"eor " VR0(r) ".16b,v19.16b," VR0(r) ".16b\n" \
"eor " VR1(r) ".16b,v18.16b," VR1(r) ".16b\n" \
: UVR0(r), UVR1(r) \
: RVR(17), RVR(16) \
: "v18", "v19"); \
break; \
default: \
ASM_BUG(); \
} \
}
#define MUL4(r...) \
{ \
MUL2(r); \
MUL2(r); \
}
/*
* Unfortunately cannot use the macro, because GCC
* will try to use the macro name and not value
* later on...
* Kept as a reference to what a register is
* (here we're using actual registers for the
* clobbered ones)
*/
#define _0f "v15"
#define _a_save "v14"
#define _b_save "v13"
#define _lt_mod_a "v12"
#define _lt_clmul_a "v11"
#define _lt_mod_b "v10"
#define _lt_clmul_b "v15"
#define _MULx2(c, r...) \
{ \
switch (REG_CNT(r)) { \
case 2: \
__asm( \
/* lts for upper part */ \
"movi v15.16b,#0x0f\n" \
"ld1 { v10.4s },%[lt0]\n" \
"ld1 { v11.4s },%[lt1]\n" \
/* upper part */ \
"and v14.16b," VR0(r) ".16b,v15.16b\n" \
"and v13.16b," VR1(r) ".16b,v15.16b\n" \
"sshr " VR0(r) ".8h," VR0(r) ".8h,#4\n" \
"sshr " VR1(r) ".8h," VR1(r) ".8h,#4\n" \
"and " VR0(r) ".16b," VR0(r) ".16b,v15.16b\n" \
"and " VR1(r) ".16b," VR1(r) ".16b,v15.16b\n" \
\
"tbl v12.16b,{v10.16b}," VR0(r) ".16b\n" \
"tbl v10.16b,{v10.16b}," VR1(r) ".16b\n" \
"tbl v15.16b,{v11.16b}," VR0(r) ".16b\n" \
"tbl v11.16b,{v11.16b}," VR1(r) ".16b\n" \
\
"eor " VR0(r) ".16b,v15.16b,v12.16b\n" \
"eor " VR1(r) ".16b,v11.16b,v10.16b\n" \
/* lts for lower part */ \
"ld1 { v10.4s },%[lt2]\n" \
"ld1 { v15.4s },%[lt3]\n" \
/* lower part */ \
"tbl v12.16b,{v10.16b},v14.16b\n" \
"tbl v10.16b,{v10.16b},v13.16b\n" \
"tbl v11.16b,{v15.16b},v14.16b\n" \
"tbl v15.16b,{v15.16b},v13.16b\n" \
\
"eor " VR0(r) ".16b," VR0(r) ".16b,v12.16b\n" \
"eor " VR1(r) ".16b," VR1(r) ".16b,v10.16b\n" \
"eor " VR0(r) ".16b," VR0(r) ".16b,v11.16b\n" \
"eor " VR1(r) ".16b," VR1(r) ".16b,v15.16b\n" \
: UVR0(r), UVR1(r) \
: [lt0] "Q" ((gf_clmul_mod_lt[4*(c)+0][0])), \
[lt1] "Q" ((gf_clmul_mod_lt[4*(c)+1][0])), \
[lt2] "Q" ((gf_clmul_mod_lt[4*(c)+2][0])), \
[lt3] "Q" ((gf_clmul_mod_lt[4*(c)+3][0])) \
: "v10", "v11", "v12", "v13", "v14", "v15"); \
break; \
default: \
ASM_BUG(); \
} \
}
#define MUL(c, r...) \
{ \
switch (REG_CNT(r)) { \
case 4: \
_MULx2(c, R_23(r)); \
_MULx2(c, R_01(r)); \
break; \
case 2: \
_MULx2(c, R_01(r)); \
break; \
default: \
ASM_BUG(); \
} \
}
#define raidz_math_begin() kfpu_begin()
#define raidz_math_end() kfpu_end()
/* Overkill... */
#if defined(_KERNEL)
#define GEN_X_DEFINE_0_3() \
register unsigned char w0 asm("v0") __attribute__((vector_size(16))); \
register unsigned char w1 asm("v1") __attribute__((vector_size(16))); \
register unsigned char w2 asm("v2") __attribute__((vector_size(16))); \
register unsigned char w3 asm("v3") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_4_5() \
register unsigned char w4 asm("v4") __attribute__((vector_size(16))); \
register unsigned char w5 asm("v5") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_6_7() \
register unsigned char w6 asm("v6") __attribute__((vector_size(16))); \
register unsigned char w7 asm("v7") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_8_9() \
register unsigned char w8 asm("v8") __attribute__((vector_size(16))); \
register unsigned char w9 asm("v9") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_10_11() \
register unsigned char w10 asm("v10") __attribute__((vector_size(16))); \
register unsigned char w11 asm("v11") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_12_15() \
register unsigned char w12 asm("v12") __attribute__((vector_size(16))); \
register unsigned char w13 asm("v13") __attribute__((vector_size(16))); \
register unsigned char w14 asm("v14") __attribute__((vector_size(16))); \
register unsigned char w15 asm("v15") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_16() \
register unsigned char w16 asm("v16") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_17() \
register unsigned char w17 asm("v17") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_18_21() \
register unsigned char w18 asm("v18") __attribute__((vector_size(16))); \
register unsigned char w19 asm("v19") __attribute__((vector_size(16))); \
register unsigned char w20 asm("v20") __attribute__((vector_size(16))); \
register unsigned char w21 asm("v21") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_22_23() \
register unsigned char w22 asm("v22") __attribute__((vector_size(16))); \
register unsigned char w23 asm("v23") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_24_27() \
register unsigned char w24 asm("v24") __attribute__((vector_size(16))); \
register unsigned char w25 asm("v25") __attribute__((vector_size(16))); \
register unsigned char w26 asm("v26") __attribute__((vector_size(16))); \
register unsigned char w27 asm("v27") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_28_30() \
register unsigned char w28 asm("v28") __attribute__((vector_size(16))); \
register unsigned char w29 asm("v29") __attribute__((vector_size(16))); \
register unsigned char w30 asm("v30") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_31() \
register unsigned char w31 asm("v31") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_32() \
register unsigned char w32 asm("v31") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_33_36() \
register unsigned char w33 asm("v31") __attribute__((vector_size(16))); \
register unsigned char w34 asm("v31") __attribute__((vector_size(16))); \
register unsigned char w35 asm("v31") __attribute__((vector_size(16))); \
register unsigned char w36 asm("v31") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_37_38() \
register unsigned char w37 asm("v31") __attribute__((vector_size(16))); \
register unsigned char w38 asm("v31") __attribute__((vector_size(16)));
#define GEN_X_DEFINE_ALL() \
GEN_X_DEFINE_0_3() \
GEN_X_DEFINE_4_5() \
GEN_X_DEFINE_6_7() \
GEN_X_DEFINE_8_9() \
GEN_X_DEFINE_10_11() \
GEN_X_DEFINE_12_15() \
GEN_X_DEFINE_16() \
GEN_X_DEFINE_17() \
GEN_X_DEFINE_18_21() \
GEN_X_DEFINE_22_23() \
GEN_X_DEFINE_24_27() \
GEN_X_DEFINE_28_30() \
GEN_X_DEFINE_31() \
GEN_X_DEFINE_32() \
GEN_X_DEFINE_33_36() \
GEN_X_DEFINE_37_38()
#else
#define GEN_X_DEFINE_0_3() \
unsigned char w0 __attribute__((vector_size(16))); \
unsigned char w1 __attribute__((vector_size(16))); \
unsigned char w2 __attribute__((vector_size(16))); \
unsigned char w3 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_4_5() \
unsigned char w4 __attribute__((vector_size(16))); \
unsigned char w5 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_6_7() \
unsigned char w6 __attribute__((vector_size(16))); \
unsigned char w7 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_8_9() \
unsigned char w8 __attribute__((vector_size(16))); \
unsigned char w9 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_10_11() \
unsigned char w10 __attribute__((vector_size(16))); \
unsigned char w11 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_12_15() \
unsigned char w12 __attribute__((vector_size(16))); \
unsigned char w13 __attribute__((vector_size(16))); \
unsigned char w14 __attribute__((vector_size(16))); \
unsigned char w15 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_16() \
unsigned char w16 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_17() \
unsigned char w17 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_18_21() \
unsigned char w18 __attribute__((vector_size(16))); \
unsigned char w19 __attribute__((vector_size(16))); \
unsigned char w20 __attribute__((vector_size(16))); \
unsigned char w21 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_22_23() \
unsigned char w22 __attribute__((vector_size(16))); \
unsigned char w23 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_24_27() \
unsigned char w24 __attribute__((vector_size(16))); \
unsigned char w25 __attribute__((vector_size(16))); \
unsigned char w26 __attribute__((vector_size(16))); \
unsigned char w27 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_28_30() \
unsigned char w28 __attribute__((vector_size(16))); \
unsigned char w29 __attribute__((vector_size(16))); \
unsigned char w30 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_31() \
unsigned char w31 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_32() \
unsigned char w32 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_33_36() \
unsigned char w33 __attribute__((vector_size(16))); \
unsigned char w34 __attribute__((vector_size(16))); \
unsigned char w35 __attribute__((vector_size(16))); \
unsigned char w36 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_37_38() \
unsigned char w37 __attribute__((vector_size(16))); \
unsigned char w38 __attribute__((vector_size(16)));
#define GEN_X_DEFINE_ALL() \
GEN_X_DEFINE_0_3() \
GEN_X_DEFINE_4_5() \
GEN_X_DEFINE_6_7() \
GEN_X_DEFINE_8_9() \
GEN_X_DEFINE_10_11() \
GEN_X_DEFINE_12_15() \
GEN_X_DEFINE_16() \
GEN_X_DEFINE_17() \
GEN_X_DEFINE_18_21() \
GEN_X_DEFINE_22_23() \
GEN_X_DEFINE_24_27() \
GEN_X_DEFINE_28_30() \
GEN_X_DEFINE_31() \
GEN_X_DEFINE_32() \
GEN_X_DEFINE_33_36() \
GEN_X_DEFINE_37_38()
#endif

164
module/zfs/vdev_raidz_math_aarch64_neonx2.c Normal file
View File

@ -0,0 +1,164 @@
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright (C) 2016 Romain Dolbeau. All rights reserved.
*/
#include <sys/isa_defs.h>
#if defined(__aarch64__)
#include "vdev_raidz_math_aarch64_neon_common.h"
#define GEN_P_DEFINE() \
GEN_X_DEFINE_0_3() \
GEN_X_DEFINE_4_5() \
GEN_X_DEFINE_6_7()
#define GEN_P_STRIDE 8
#define GEN_P_P 0, 1, 2, 3, 4, 5, 6, 7
#define GEN_PQ_DEFINE() \
GEN_X_DEFINE_0_3() \
GEN_X_DEFINE_4_5() \
GEN_X_DEFINE_6_7() \
GEN_X_DEFINE_8_9() \
GEN_X_DEFINE_10_11() \
GEN_X_DEFINE_16() \
GEN_X_DEFINE_17() \
GEN_X_DEFINE_33_36()
#define GEN_PQ_STRIDE 4
#define GEN_PQ_D 0, 1, 2, 3
#define GEN_PQ_P 4, 5, 6, 7
#define GEN_PQ_Q 8, 9, 10, 11
#define GEN_PQR_DEFINE() \
GEN_X_DEFINE_0_3() \
GEN_X_DEFINE_4_5() \
GEN_X_DEFINE_6_7() \
GEN_X_DEFINE_8_9() \
GEN_X_DEFINE_22_23() \
GEN_X_DEFINE_24_27() \
GEN_X_DEFINE_16() \
GEN_X_DEFINE_17() \
GEN_X_DEFINE_33_36()
#define GEN_PQR_STRIDE 4
#define GEN_PQR_D 0, 1, 2, 3
#define GEN_PQR_P 4, 5, 6, 7
#define GEN_PQR_Q 8, 9, 22, 23
#define GEN_PQR_R 24, 25, 26, 27
#define REC_P_DEFINE() \
GEN_X_DEFINE_0_3() \
GEN_X_DEFINE_33_36()
#define REC_P_STRIDE 4
#define REC_P_X 0, 1, 2, 3
#define REC_Q_DEFINE() \
GEN_X_DEFINE_0_3() \
GEN_X_DEFINE_16() \
GEN_X_DEFINE_17() \
GEN_X_DEFINE_33_36()
#define REC_Q_STRIDE 4
#define REC_Q_X 0, 1, 2, 3
#define REC_R_DEFINE() \
GEN_X_DEFINE_0_3() \
GEN_X_DEFINE_16() \
GEN_X_DEFINE_17() \
GEN_X_DEFINE_33_36()
#define REC_R_STRIDE 4
#define REC_R_X 0, 1, 2, 3
#define REC_PQ_DEFINE() \
GEN_X_DEFINE_0_3() \
GEN_X_DEFINE_4_5() \
GEN_X_DEFINE_6_7() \
GEN_X_DEFINE_8_9() \
GEN_X_DEFINE_16() \
GEN_X_DEFINE_17() \
GEN_X_DEFINE_22_23() \
GEN_X_DEFINE_33_36()
#define REC_PQ_STRIDE 4
#define REC_PQ_X 0, 1, 2, 3
#define REC_PQ_Y 4, 5, 6, 7
#define REC_PQ_D 8, 9, 22, 23
#define REC_PR_DEFINE() REC_PQ_DEFINE()
#define REC_PR_STRIDE 4
#define REC_PR_X 0, 1, 2, 3
#define REC_PR_Y 4, 5, 6, 7
#define REC_PR_D 8, 9, 22, 23
#define REC_QR_DEFINE() REC_PQ_DEFINE()
#define REC_QR_STRIDE 4
#define REC_QR_X 0, 1, 2, 3
#define REC_QR_Y 4, 5, 6, 7
#define REC_QR_D 8, 9, 22, 23
#define REC_PQR_DEFINE() \
GEN_X_DEFINE_0_3() \
GEN_X_DEFINE_4_5() \
GEN_X_DEFINE_6_7() \
GEN_X_DEFINE_8_9() \
GEN_X_DEFINE_16() \
GEN_X_DEFINE_17() \
GEN_X_DEFINE_22_23() \
GEN_X_DEFINE_24_27() \
GEN_X_DEFINE_28_30() \
GEN_X_DEFINE_31() \
GEN_X_DEFINE_33_36()
#define REC_PQR_STRIDE 4
#define REC_PQR_X 0, 1, 2, 3
#define REC_PQR_Y 4, 5, 6, 7
#define REC_PQR_Z 8, 9, 22, 23
#define REC_PQR_D 24, 25, 26, 27
#define REC_PQR_XS 24, 25, 26, 27
#define REC_PQR_YS 28, 29, 30, 31
#include <sys/vdev_raidz_impl.h>
#include "vdev_raidz_math_impl.h"
DEFINE_GEN_METHODS(aarch64_neonx2);
/*
* If compiled with -O0, gcc doesn't do any stack frame coalescing
* and -Wframe-larger-than=1024 is triggered in debug mode.
*/
#pragma GCC diagnostic ignored "-Wframe-larger-than="
DEFINE_REC_METHODS(aarch64_neonx2);
#pragma GCC diagnostic pop
static boolean_t
raidz_will_aarch64_neonx2_work(void)
{
return (B_TRUE); // __arch64__ requires NEON
}
const raidz_impl_ops_t vdev_raidz_aarch64_neonx2_impl = {
.init = NULL,
.fini = NULL,
.gen = RAIDZ_GEN_METHODS(aarch64_neonx2),
.rec = RAIDZ_REC_METHODS(aarch64_neonx2),
.is_supported = &raidz_will_aarch64_neonx2_work,
.name = "aarch64_neonx2"
};
#endif /* defined(__aarch64__) */

20
module/zfs/vdev_raidz_math_avx2.c
View File

@ -122,6 +122,26 @@ typedef struct v {
} \
}
#define ZERO(r...) \
{ \
switch (REG_CNT(r)) { \
case 4: \
__asm( \
"vpxor %" VR0(r) ", %" VR0(r)", %" VR0(r) "\n" \
"vpxor %" VR1(r) ", %" VR1(r)", %" VR1(r) "\n" \
"vpxor %" VR2(r) ", %" VR2(r)", %" VR2(r) "\n" \
"vpxor %" VR3(r) ", %" VR3(r)", %" VR3(r)); \
break; \
case 2: \
__asm( \
"vpxor %" VR0(r) ", %" VR0(r)", %" VR0(r) "\n" \
"vpxor %" VR1(r) ", %" VR1(r)", %" VR1(r)); \
break; \
default: \
ASM_BUG(); \
} \
}
#define COPY(r...) \
{ \
switch (REG_CNT(r)) { \

16
module/zfs/vdev_raidz_math_impl.h
View File

@ -490,7 +490,7 @@ REC_Q_BLOCK(raidz_map_t * const rm, const size_t off, const size_t end,
for (ioff = off; ioff < end; ioff += (REC_Q_STRIDE * sizeof (v_t))) {
MUL2_SETUP();
XOR(REC_Q_X, REC_Q_X);
ZERO(REC_Q_X);
if (ncols == nbigcols) {
for (c = firstdc; c < x; c++)
@ -587,7 +587,7 @@ REC_R_BLOCK(raidz_map_t * const rm, const size_t off, const size_t end,
for (ioff = off; ioff < end; ioff += (REC_R_STRIDE * sizeof (v_t))) {
MUL2_SETUP();
XOR(REC_R_X, REC_R_X);
ZERO(REC_R_X);
if (ncols == nbigcols) {
for (c = firstdc; c < x; c++)
@ -690,7 +690,7 @@ REC_PQ_BLOCK(raidz_map_t * const rm, const size_t off, const size_t end,
for (ioff = off; ioff < end; ioff += (REC_PQ_STRIDE * sizeof (v_t))) {
LOAD(COL_OFF(pcol, ioff), REC_PQ_X);
XOR(REC_PQ_Y, REC_PQ_Y);
ZERO(REC_PQ_Y);
MUL2_SETUP();
if (ncols == nbigcols) {
@ -816,7 +816,7 @@ REC_PR_BLOCK(raidz_map_t * const rm, const size_t off, const size_t end,
for (ioff = off; ioff < end; ioff += (REC_PR_STRIDE * sizeof (v_t))) {
LOAD(COL_OFF(pcol, ioff), REC_PR_X);
XOR(REC_PR_Y, REC_PR_Y);
ZERO(REC_PR_Y);
MUL2_SETUP();
if (ncols == nbigcols) {
@ -949,8 +949,8 @@ REC_QR_BLOCK(raidz_map_t * const rm, const size_t off, const size_t end,
for (ioff = off; ioff < end; ioff += (REC_QR_STRIDE * sizeof (v_t))) {
MUL2_SETUP();
XOR(REC_QR_X, REC_QR_X);
XOR(REC_QR_Y, REC_QR_Y);
ZERO(REC_QR_X);
ZERO(REC_QR_Y);
if (ncols == nbigcols) {
for (c = firstdc; c < x; c++)
@ -1090,8 +1090,8 @@ REC_PQR_BLOCK(raidz_map_t * const rm, const size_t off, const size_t end,
for (ioff = off; ioff < end; ioff += (REC_PQR_STRIDE * sizeof (v_t))) {
MUL2_SETUP();
LOAD(COL_OFF(pcol, ioff), REC_PQR_X);
XOR(REC_PQR_Y, REC_PQR_Y);
XOR(REC_PQR_Z, REC_PQR_Z);
ZERO(REC_PQR_Y);
ZERO(REC_PQR_Z);
if (ncols == nbigcols) {
for (c = firstdc; c < x; c++)

1
module/zfs/vdev_raidz_math_scalar.c
View File

@ -91,6 +91,7 @@ raidz_init_scalar(void)
#define XOR_ACC(src, acc) acc.e ^= ((v_t *)src)[0].e
#define XOR(src, acc) acc.e ^= src.e
#define ZERO(acc) acc.e = 0
#define COPY(src, dst) dst = src
#define LOAD(src, val) val = ((v_t *)src)[0]
#define STORE(dst, val) ((v_t *)dst)[0] = val

21
module/zfs/vdev_raidz_math_sse2.c
View File

@ -106,6 +106,27 @@ typedef struct v {
break; \
} \
}
#define ZERO(r...) \
{ \
switch (REG_CNT(r)) { \
case 4: \
__asm( \
"pxor %" VR0(r) ", %" VR0(r) "\n" \
"pxor %" VR1(r) ", %" VR1(r) "\n" \
"pxor %" VR2(r) ", %" VR2(r) "\n" \
"pxor %" VR3(r) ", %" VR3(r)); \
break; \
case 2: \
__asm( \
"pxor %" VR0(r) ", %" VR0(r) "\n" \
"pxor %" VR1(r) ", %" VR1(r)); \
break; \
case 1: \
__asm( \
"pxor %" VR0(r) ", %" VR0(r)); \
break; \
} \
}
#define COPY(r...) \
{ \

20
module/zfs/vdev_raidz_math_ssse3.c
View File

@ -122,6 +122,26 @@ typedef struct v {
} \
}
#define ZERO(r...) \
{ \
switch (REG_CNT(r)) { \
case 4: \
__asm( \
"pxor %" VR0(r) ", %" VR0(r) "\n" \
"pxor %" VR1(r) ", %" VR1(r) "\n" \
"pxor %" VR2(r) ", %" VR2(r) "\n" \
"pxor %" VR3(r) ", %" VR3(r)); \
break; \
case 2: \
__asm( \
"pxor %" VR0(r) ", %" VR0(r) "\n" \
"pxor %" VR1(r) ", %" VR1(r)); \
break; \
default: \
ASM_BUG(); \
} \
}
#define COPY(r...) \
{ \
switch (REG_CNT(r)) { \