c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2024-11-17 18:11:00 +03:00
Code Issues Packages Projects Releases Wiki Activity
045aeabce6
mirror_zfs/module/zfs/vdev_raidz_math_avx512bw.c

414 lines
11 KiB
C
Raw Normal View History

Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
/*
* 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.
ABD: Adapt avx512bw raidz assembly Adapt avx512bw implementation for use with abd buffers. Mul2 implementation is rewritten to take advantage of the BW instruction set. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Romain Dolbeau <romain.dolbeau@atos.net> Signed-off-by: Gvozden Neskovic <neskovic@gmail.com> Closes #5477
2016-12-16 04:31:33 +03:00
* Copyright (C) 2016 Gvozden Nešković. All rights reserved.
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
*/
#include <sys/isa_defs.h>
ABD: Adapt avx512bw raidz assembly Adapt avx512bw implementation for use with abd buffers. Mul2 implementation is rewritten to take advantage of the BW instruction set. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Romain Dolbeau <romain.dolbeau@atos.net> Signed-off-by: Gvozden Neskovic <neskovic@gmail.com> Closes #5477
2016-12-16 04:31:33 +03:00
#if defined(__x86_64) && defined(HAVE_AVX512BW)
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
Refactor ccompile.h to not include system headers This is a step toward being able to vendor the OpenZFS code in FreeBSD. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Ryan Moeller <ryan@iXsystems.com> Signed-off-by: Matt Macy <mmacy@FreeBSD.org> Closes #10625
2020-07-26 06:09:50 +03:00
#include <sys/param.h>
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
#include <sys/types.h>
OpenZFS restructuring - move platform specific headers Move platform specific Linux headers under include/os/linux/. Update the build system accordingly to detect the platform. This lays some of the initial groundwork to supporting building for other platforms. As part of this change it was necessary to create both a user and kernel space sys/simd.h header which can be included in either context. No functional change, the source has been refactored and the relevant #include's updated. Reviewed-by: Jorgen Lundman <lundman@lundman.net> Reviewed-by: Igor Kozhukhov <igor@dilos.org> Signed-off-by: Matthew Macy <mmacy@FreeBSD.org> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #9198
2019-09-05 19:34:54 +03:00
#include <sys/simd.h>
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
Refactor ccompile.h to not include system headers This is a step toward being able to vendor the OpenZFS code in FreeBSD. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Ryan Moeller <ryan@iXsystems.com> Signed-off-by: Matt Macy <mmacy@FreeBSD.org> Closes #10625
2020-07-26 06:09:50 +03:00
#ifdef __linux__
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
#define __asm __asm__ __volatile__
Refactor ccompile.h to not include system headers This is a step toward being able to vendor the OpenZFS code in FreeBSD. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Ryan Moeller <ryan@iXsystems.com> Signed-off-by: Matt Macy <mmacy@FreeBSD.org> Closes #10625
2020-07-26 06:09:50 +03:00
#endif
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
#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, ...) "zmm"#REG
#define VR1_(_1, REG, ...) "zmm"#REG
#define VR2_(_1, _2, REG, ...) "zmm"#REG
#define VR3_(_1, _2, _3, REG, ...) "zmm"#REG
#define VR4_(_1, _2, _3, _4, REG, ...) "zmm"#REG
#define VR5_(_1, _2, _3, _4, _5, REG, ...) "zmm"#REG
#define VR6_(_1, _2, _3, _4, _5, _6, REG, ...) "zmm"#REG
#define VR7_(_1, _2, _3, _4, _5, _6, _7, REG, ...) "zmm"#REG
#define VR0(r...) VR0_(r)
#define VR1(r...) VR1_(r)
#define VR2(r...) VR2_(r, 1)
#define VR3(r...) VR3_(r, 1, 2)
#define VR4(r...) VR4_(r, 1, 2)
#define VR5(r...) VR5_(r, 1, 2, 3)
#define VR6(r...) VR6_(r, 1, 2, 3, 4)
#define VR7(r...) VR7_(r, 1, 2, 3, 4, 5)
#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)
Linux 5.0 compat: ASM_BUG macro The 5.0 kernel defines the macro ASM_BUG. In order to prevent a conflict and build failure rename ASM_BUG to ZFS_ASM_BUG. This is currently only an issue on aarch64 but all instances of ASM_BUG we're renamed to avoid any future conflict on x86_64. Reviewed-by: Tomohiro Kusumi <kusumi.tomohiro@gmail.com> Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Chris Dunlop <chris@onthe.net.au> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #8725 Issue #8545
2019-05-08 20:18:40 +03:00
#define ZFS_ASM_BUG() ASSERT(0)
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
extern const uint8_t gf_clmul_mod_lt[4*256][16];
#define ELEM_SIZE 64
typedef struct v {
uint8_t b[ELEM_SIZE] __attribute__((aligned(ELEM_SIZE)));
} v_t;
#define XOR_ACC(src, r...) \
{ \
switch (REG_CNT(r)) { \
case 4: \
__asm( \
"vpxorq 0x00(%[SRC]), %%" VR0(r)", %%" VR0(r) "\n" \
"vpxorq 0x40(%[SRC]), %%" VR1(r)", %%" VR1(r) "\n" \
"vpxorq 0x80(%[SRC]), %%" VR2(r)", %%" VR2(r) "\n" \
"vpxorq 0xc0(%[SRC]), %%" VR3(r)", %%" VR3(r) "\n" \
: : [SRC] "r" (src)); \
break; \
case 2: \
__asm( \
"vpxorq 0x00(%[SRC]), %%" VR0(r)", %%" VR0(r) "\n" \
"vpxorq 0x40(%[SRC]), %%" VR1(r)", %%" VR1(r) "\n" \
: : [SRC] "r" (src)); \
break; \
default: \
Linux 5.0 compat: ASM_BUG macro The 5.0 kernel defines the macro ASM_BUG. In order to prevent a conflict and build failure rename ASM_BUG to ZFS_ASM_BUG. This is currently only an issue on aarch64 but all instances of ASM_BUG we're renamed to avoid any future conflict on x86_64. Reviewed-by: Tomohiro Kusumi <kusumi.tomohiro@gmail.com> Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Chris Dunlop <chris@onthe.net.au> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #8725 Issue #8545
2019-05-08 20:18:40 +03:00
ZFS_ASM_BUG(); \
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
} \
}
#define XOR(r...) \
{ \
switch (REG_CNT(r)) { \
case 8: \
__asm( \
"vpxorq %" VR0(r) ", %" VR4(r)", %" VR4(r) "\n" \
"vpxorq %" VR1(r) ", %" VR5(r)", %" VR5(r) "\n" \
"vpxorq %" VR2(r) ", %" VR6(r)", %" VR6(r) "\n" \
"vpxorq %" VR3(r) ", %" VR7(r)", %" VR7(r)); \
break; \
case 4: \
__asm( \
"vpxorq %" VR0(r) ", %" VR2(r)", %" VR2(r) "\n" \
"vpxorq %" VR1(r) ", %" VR3(r)", %" VR3(r)); \
break; \
default: \
Linux 5.0 compat: ASM_BUG macro The 5.0 kernel defines the macro ASM_BUG. In order to prevent a conflict and build failure rename ASM_BUG to ZFS_ASM_BUG. This is currently only an issue on aarch64 but all instances of ASM_BUG we're renamed to avoid any future conflict on x86_64. Reviewed-by: Tomohiro Kusumi <kusumi.tomohiro@gmail.com> Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Chris Dunlop <chris@onthe.net.au> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #8725 Issue #8545
2019-05-08 20:18:40 +03:00
ZFS_ASM_BUG(); \
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
} \
}
ABD: Adapt avx512bw raidz assembly Adapt avx512bw implementation for use with abd buffers. Mul2 implementation is rewritten to take advantage of the BW instruction set. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Romain Dolbeau <romain.dolbeau@atos.net> Signed-off-by: Gvozden Neskovic <neskovic@gmail.com> Closes #5477
2016-12-16 04:31:33 +03:00
#define ZERO(r...) XOR(r, r)
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
Use cstyle -cpP in `make cstyle` check Enable picky cstyle checks and resolve the new warnings. The vast majority of the changes needed were to handle minor issues with whitespace formatting. This patch contains no functional changes. Non-whitespace changes are as follows: * 8 times ; to { } in for/while loop * fix missing ; in cmd/zed/agents/zfs_diagnosis.c * comment (confim -> confirm) * change endline , to ; in cmd/zpool/zpool_main.c * a number of /* BEGIN CSTYLED */ /* END CSTYLED */ blocks * /* CSTYLED */ markers * change == 0 to ! * ulong to unsigned long in module/zfs/dsl_scan.c * rearrangement of module_param lines in module/zfs/metaslab.c * add { } block around statement after for_each_online_node Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov> Reviewed-by: Håkan Johansson <f96hajo@chalmers.se> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #5465
2016-12-12 21:46:26 +03:00
#define COPY(r...) \
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
{ \
switch (REG_CNT(r)) { \
case 8: \
__asm( \
"vmovdqa64 %" VR0(r) ", %" VR4(r) "\n" \
"vmovdqa64 %" VR1(r) ", %" VR5(r) "\n" \
"vmovdqa64 %" VR2(r) ", %" VR6(r) "\n" \
"vmovdqa64 %" VR3(r) ", %" VR7(r)); \
break; \
case 4: \
__asm( \
"vmovdqa64 %" VR0(r) ", %" VR2(r) "\n" \
"vmovdqa64 %" VR1(r) ", %" VR3(r)); \
break; \
default: \
Linux 5.0 compat: ASM_BUG macro The 5.0 kernel defines the macro ASM_BUG. In order to prevent a conflict and build failure rename ASM_BUG to ZFS_ASM_BUG. This is currently only an issue on aarch64 but all instances of ASM_BUG we're renamed to avoid any future conflict on x86_64. Reviewed-by: Tomohiro Kusumi <kusumi.tomohiro@gmail.com> Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Chris Dunlop <chris@onthe.net.au> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #8725 Issue #8545
2019-05-08 20:18:40 +03:00
ZFS_ASM_BUG(); \
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
} \
}
Use cstyle -cpP in `make cstyle` check Enable picky cstyle checks and resolve the new warnings. The vast majority of the changes needed were to handle minor issues with whitespace formatting. This patch contains no functional changes. Non-whitespace changes are as follows: * 8 times ; to { } in for/while loop * fix missing ; in cmd/zed/agents/zfs_diagnosis.c * comment (confim -> confirm) * change endline , to ; in cmd/zpool/zpool_main.c * a number of /* BEGIN CSTYLED */ /* END CSTYLED */ blocks * /* CSTYLED */ markers * change == 0 to ! * ulong to unsigned long in module/zfs/dsl_scan.c * rearrangement of module_param lines in module/zfs/metaslab.c * add { } block around statement after for_each_online_node Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov> Reviewed-by: Håkan Johansson <f96hajo@chalmers.se> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #5465
2016-12-12 21:46:26 +03:00
#define LOAD(src, r...) \
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
{ \
switch (REG_CNT(r)) { \
case 4: \
__asm( \
"vmovdqa64 0x00(%[SRC]), %%" VR0(r) "\n" \
"vmovdqa64 0x40(%[SRC]), %%" VR1(r) "\n" \
"vmovdqa64 0x80(%[SRC]), %%" VR2(r) "\n" \
"vmovdqa64 0xc0(%[SRC]), %%" VR3(r) "\n" \
: : [SRC] "r" (src)); \
break; \
case 2: \
__asm( \
"vmovdqa64 0x00(%[SRC]), %%" VR0(r) "\n" \
"vmovdqa64 0x40(%[SRC]), %%" VR1(r) "\n" \
: : [SRC] "r" (src)); \
break; \
default: \
Linux 5.0 compat: ASM_BUG macro The 5.0 kernel defines the macro ASM_BUG. In order to prevent a conflict and build failure rename ASM_BUG to ZFS_ASM_BUG. This is currently only an issue on aarch64 but all instances of ASM_BUG we're renamed to avoid any future conflict on x86_64. Reviewed-by: Tomohiro Kusumi <kusumi.tomohiro@gmail.com> Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Chris Dunlop <chris@onthe.net.au> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #8725 Issue #8545
2019-05-08 20:18:40 +03:00
ZFS_ASM_BUG(); \
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
} \
}
Use cstyle -cpP in `make cstyle` check Enable picky cstyle checks and resolve the new warnings. The vast majority of the changes needed were to handle minor issues with whitespace formatting. This patch contains no functional changes. Non-whitespace changes are as follows: * 8 times ; to { } in for/while loop * fix missing ; in cmd/zed/agents/zfs_diagnosis.c * comment (confim -> confirm) * change endline , to ; in cmd/zpool/zpool_main.c * a number of /* BEGIN CSTYLED */ /* END CSTYLED */ blocks * /* CSTYLED */ markers * change == 0 to ! * ulong to unsigned long in module/zfs/dsl_scan.c * rearrangement of module_param lines in module/zfs/metaslab.c * add { } block around statement after for_each_online_node Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov> Reviewed-by: Håkan Johansson <f96hajo@chalmers.se> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #5465
2016-12-12 21:46:26 +03:00
#define STORE(dst, r...) \
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
{ \
switch (REG_CNT(r)) { \
case 4: \
__asm( \
"vmovdqa64 %%" VR0(r) ", 0x00(%[DST])\n" \
"vmovdqa64 %%" VR1(r) ", 0x40(%[DST])\n" \
"vmovdqa64 %%" VR2(r) ", 0x80(%[DST])\n" \
"vmovdqa64 %%" VR3(r) ", 0xc0(%[DST])\n" \
: : [DST] "r" (dst)); \
break; \
case 2: \
__asm( \
"vmovdqa64 %%" VR0(r) ", 0x00(%[DST])\n" \
"vmovdqa64 %%" VR1(r) ", 0x40(%[DST])\n" \
: : [DST] "r" (dst)); \
break; \
default: \
Linux 5.0 compat: ASM_BUG macro The 5.0 kernel defines the macro ASM_BUG. In order to prevent a conflict and build failure rename ASM_BUG to ZFS_ASM_BUG. This is currently only an issue on aarch64 but all instances of ASM_BUG we're renamed to avoid any future conflict on x86_64. Reviewed-by: Tomohiro Kusumi <kusumi.tomohiro@gmail.com> Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Chris Dunlop <chris@onthe.net.au> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #8725 Issue #8545
2019-05-08 20:18:40 +03:00
ZFS_ASM_BUG(); \
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
} \
}
ABD: Adapt avx512bw raidz assembly Adapt avx512bw implementation for use with abd buffers. Mul2 implementation is rewritten to take advantage of the BW instruction set. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Romain Dolbeau <romain.dolbeau@atos.net> Signed-off-by: Gvozden Neskovic <neskovic@gmail.com> Closes #5477
2016-12-16 04:31:33 +03:00
#define MUL2_SETUP() \
{ \
__asm("vmovq %0, %%xmm22" :: "r"(0x1d1d1d1d1d1d1d1d)); \
__asm("vpbroadcastq %xmm22, %zmm22"); \
__asm("vpxord %zmm23, %zmm23 ,%zmm23"); \
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
}
Use cstyle -cpP in `make cstyle` check Enable picky cstyle checks and resolve the new warnings. The vast majority of the changes needed were to handle minor issues with whitespace formatting. This patch contains no functional changes. Non-whitespace changes are as follows: * 8 times ; to { } in for/while loop * fix missing ; in cmd/zed/agents/zfs_diagnosis.c * comment (confim -> confirm) * change endline , to ; in cmd/zpool/zpool_main.c * a number of /* BEGIN CSTYLED */ /* END CSTYLED */ blocks * /* CSTYLED */ markers * change == 0 to ! * ulong to unsigned long in module/zfs/dsl_scan.c * rearrangement of module_param lines in module/zfs/metaslab.c * add { } block around statement after for_each_online_node Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov> Reviewed-by: Håkan Johansson <f96hajo@chalmers.se> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #5465
2016-12-12 21:46:26 +03:00
#define _MUL2(r...) \
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
{ \
switch (REG_CNT(r)) { \
case 2: \
__asm( \
ABD: Adapt avx512bw raidz assembly Adapt avx512bw implementation for use with abd buffers. Mul2 implementation is rewritten to take advantage of the BW instruction set. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Romain Dolbeau <romain.dolbeau@atos.net> Signed-off-by: Gvozden Neskovic <neskovic@gmail.com> Closes #5477
2016-12-16 04:31:33 +03:00
"vpcmpb $1, %zmm23, %" VR0(r)", %k1\n" \
"vpcmpb $1, %zmm23, %" VR1(r)", %k2\n" \
"vpaddb %" VR0(r)", %" VR0(r)", %" VR0(r) "\n" \
"vpaddb %" VR1(r)", %" VR1(r)", %" VR1(r) "\n" \
"vpxord %zmm22, %" VR0(r)", %zmm12\n" \
"vpxord %zmm22, %" VR1(r)", %zmm13\n" \
"vmovdqu8 %zmm12, %" VR0(r) "{%k1}\n" \
"vmovdqu8 %zmm13, %" VR1(r) "{%k2}"); \
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
break; \
default: \
Linux 5.0 compat: ASM_BUG macro The 5.0 kernel defines the macro ASM_BUG. In order to prevent a conflict and build failure rename ASM_BUG to ZFS_ASM_BUG. This is currently only an issue on aarch64 but all instances of ASM_BUG we're renamed to avoid any future conflict on x86_64. Reviewed-by: Tomohiro Kusumi <kusumi.tomohiro@gmail.com> Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Chris Dunlop <chris@onthe.net.au> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #8725 Issue #8545
2019-05-08 20:18:40 +03:00
ZFS_ASM_BUG(); \
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
} \
}
#define MUL2(r...) \
{ \
switch (REG_CNT(r)) { \
case 4: \
_MUL2(R_01(r)); \
_MUL2(R_23(r)); \
break; \
case 2: \
_MUL2(r); \
break; \
default: \
Linux 5.0 compat: ASM_BUG macro The 5.0 kernel defines the macro ASM_BUG. In order to prevent a conflict and build failure rename ASM_BUG to ZFS_ASM_BUG. This is currently only an issue on aarch64 but all instances of ASM_BUG we're renamed to avoid any future conflict on x86_64. Reviewed-by: Tomohiro Kusumi <kusumi.tomohiro@gmail.com> Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Chris Dunlop <chris@onthe.net.au> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #8725 Issue #8545
2019-05-08 20:18:40 +03:00
ZFS_ASM_BUG(); \
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
} \
}
#define MUL4(r...) \
{ \
MUL2(r); \
MUL2(r); \
}
#define _0f "zmm15"
#define _as "zmm14"
#define _bs "zmm13"
#define _ltmod "zmm12"
#define _ltmul "zmm11"
#define _ta "zmm10"
#define _tb "zmm15"
ABD: Adapt avx512bw raidz assembly Adapt avx512bw implementation for use with abd buffers. Mul2 implementation is rewritten to take advantage of the BW instruction set. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Romain Dolbeau <romain.dolbeau@atos.net> Signed-off-by: Gvozden Neskovic <neskovic@gmail.com> Closes #5477
2016-12-16 04:31:33 +03:00
static const uint8_t __attribute__((aligned(64))) _mul_mask = 0x0F;
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
#define _MULx2(c, r...) \
{ \
switch (REG_CNT(r)) { \
case 2: \
__asm( \
"vpbroadcastb (%[mask]), %%" _0f "\n" \
/* upper bits */ \
"vbroadcasti32x4 0x00(%[lt]), %%" _ltmod "\n" \
"vbroadcasti32x4 0x10(%[lt]), %%" _ltmul "\n" \
\
"vpsraw $0x4, %%" VR0(r) ", %%"_as "\n" \
"vpsraw $0x4, %%" VR1(r) ", %%"_bs "\n" \
"vpandq %%" _0f ", %%" VR0(r) ", %%" VR0(r) "\n" \
"vpandq %%" _0f ", %%" VR1(r) ", %%" VR1(r) "\n" \
"vpandq %%" _0f ", %%" _as ", %%" _as "\n" \
"vpandq %%" _0f ", %%" _bs ", %%" _bs "\n" \
\
"vpshufb %%" _as ", %%" _ltmod ", %%" _ta "\n" \
"vpshufb %%" _bs ", %%" _ltmod ", %%" _tb "\n" \
"vpshufb %%" _as ", %%" _ltmul ", %%" _as "\n" \
"vpshufb %%" _bs ", %%" _ltmul ", %%" _bs "\n" \
/* lower bits */ \
"vbroadcasti32x4 0x20(%[lt]), %%" _ltmod "\n" \
"vbroadcasti32x4 0x30(%[lt]), %%" _ltmul "\n" \
\
"vpxorq %%" _ta ", %%" _as ", %%" _as "\n" \
"vpxorq %%" _tb ", %%" _bs ", %%" _bs "\n" \
\
"vpshufb %%" VR0(r) ", %%" _ltmod ", %%" _ta "\n" \
"vpshufb %%" VR1(r) ", %%" _ltmod ", %%" _tb "\n" \
"vpshufb %%" VR0(r) ", %%" _ltmul ", %%" VR0(r) "\n"\
"vpshufb %%" VR1(r) ", %%" _ltmul ", %%" VR1(r) "\n"\
\
"vpxorq %%" _ta ", %%" VR0(r) ", %%" VR0(r) "\n" \
"vpxorq %%" _as ", %%" VR0(r) ", %%" VR0(r) "\n" \
"vpxorq %%" _tb ", %%" VR1(r) ", %%" VR1(r) "\n" \
"vpxorq %%" _bs ", %%" VR1(r) ", %%" VR1(r) "\n" \
: : [mask] "r" (&_mul_mask), \
[lt] "r" (gf_clmul_mod_lt[4*(c)])); \
break; \
default: \
Linux 5.0 compat: ASM_BUG macro The 5.0 kernel defines the macro ASM_BUG. In order to prevent a conflict and build failure rename ASM_BUG to ZFS_ASM_BUG. This is currently only an issue on aarch64 but all instances of ASM_BUG we're renamed to avoid any future conflict on x86_64. Reviewed-by: Tomohiro Kusumi <kusumi.tomohiro@gmail.com> Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Chris Dunlop <chris@onthe.net.au> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #8725 Issue #8545
2019-05-08 20:18:40 +03:00
ZFS_ASM_BUG(); \
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
} \
}
#define MUL(c, r...) \
{ \
switch (REG_CNT(r)) { \
case 4: \
_MULx2(c, R_01(r)); \
_MULx2(c, R_23(r)); \
break; \
case 2: \
_MULx2(c, R_01(r)); \
break; \
default: \
Linux 5.0 compat: ASM_BUG macro The 5.0 kernel defines the macro ASM_BUG. In order to prevent a conflict and build failure rename ASM_BUG to ZFS_ASM_BUG. This is currently only an issue on aarch64 but all instances of ASM_BUG we're renamed to avoid any future conflict on x86_64. Reviewed-by: Tomohiro Kusumi <kusumi.tomohiro@gmail.com> Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Chris Dunlop <chris@onthe.net.au> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #8725 Issue #8545
2019-05-08 20:18:40 +03:00
ZFS_ASM_BUG(); \
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
} \
}
#define raidz_math_begin() kfpu_begin()
ABD: Adapt avx512bw raidz assembly Adapt avx512bw implementation for use with abd buffers. Mul2 implementation is rewritten to take advantage of the BW instruction set. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Romain Dolbeau <romain.dolbeau@atos.net> Signed-off-by: Gvozden Neskovic <neskovic@gmail.com> Closes #5477
2016-12-16 04:31:33 +03:00
#define raidz_math_end() kfpu_end()
/*
* ZERO, COPY, and MUL operations are already 2x unrolled, which means that
* the stride of these operations for avx512 must not exceed 4. Otherwise, a
* single step would exceed 512B block size.
*/
#define SYN_STRIDE 4
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
ABD Vectorized raidz Enable vectorized raidz code on ABD buffers. The avx512f, avx512bw, neon and aarch64_neonx2 are disabled in this commit. With the exception of avx512bw these implementations are updated for ABD in the subsequent commits. Signed-off-by: Gvozden Neskovic <neskovic@gmail.com>
2016-08-24 16:51:33 +03:00
#define ZERO_STRIDE 4
#define ZERO_DEFINE() {}
#define ZERO_D 0, 1, 2, 3
#define COPY_STRIDE 4
#define COPY_DEFINE() {}
#define COPY_D 0, 1, 2, 3
#define ADD_STRIDE 4
#define ADD_DEFINE() {}
Use cstyle -cpP in `make cstyle` check Enable picky cstyle checks and resolve the new warnings. The vast majority of the changes needed were to handle minor issues with whitespace formatting. This patch contains no functional changes. Non-whitespace changes are as follows: * 8 times ; to { } in for/while loop * fix missing ; in cmd/zed/agents/zfs_diagnosis.c * comment (confim -> confirm) * change endline , to ; in cmd/zpool/zpool_main.c * a number of /* BEGIN CSTYLED */ /* END CSTYLED */ blocks * /* CSTYLED */ markers * change == 0 to ! * ulong to unsigned long in module/zfs/dsl_scan.c * rearrangement of module_param lines in module/zfs/metaslab.c * add { } block around statement after for_each_online_node Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov> Reviewed-by: Håkan Johansson <f96hajo@chalmers.se> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #5465
2016-12-12 21:46:26 +03:00
#define ADD_D 0, 1, 2, 3
ABD Vectorized raidz Enable vectorized raidz code on ABD buffers. The avx512f, avx512bw, neon and aarch64_neonx2 are disabled in this commit. With the exception of avx512bw these implementations are updated for ABD in the subsequent commits. Signed-off-by: Gvozden Neskovic <neskovic@gmail.com>
2016-08-24 16:51:33 +03:00
#define MUL_STRIDE 4
Use cstyle -cpP in `make cstyle` check Enable picky cstyle checks and resolve the new warnings. The vast majority of the changes needed were to handle minor issues with whitespace formatting. This patch contains no functional changes. Non-whitespace changes are as follows: * 8 times ; to { } in for/while loop * fix missing ; in cmd/zed/agents/zfs_diagnosis.c * comment (confim -> confirm) * change endline , to ; in cmd/zpool/zpool_main.c * a number of /* BEGIN CSTYLED */ /* END CSTYLED */ blocks * /* CSTYLED */ markers * change == 0 to ! * ulong to unsigned long in module/zfs/dsl_scan.c * rearrangement of module_param lines in module/zfs/metaslab.c * add { } block around statement after for_each_online_node Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov> Reviewed-by: Håkan Johansson <f96hajo@chalmers.se> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #5465
2016-12-12 21:46:26 +03:00
#define MUL_DEFINE() {}
ABD Vectorized raidz Enable vectorized raidz code on ABD buffers. The avx512f, avx512bw, neon and aarch64_neonx2 are disabled in this commit. With the exception of avx512bw these implementations are updated for ABD in the subsequent commits. Signed-off-by: Gvozden Neskovic <neskovic@gmail.com>
2016-08-24 16:51:33 +03:00
#define MUL_D 0, 1, 2, 3
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
#define GEN_P_STRIDE 4
ABD: Adapt avx512bw raidz assembly Adapt avx512bw implementation for use with abd buffers. Mul2 implementation is rewritten to take advantage of the BW instruction set. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Romain Dolbeau <romain.dolbeau@atos.net> Signed-off-by: Gvozden Neskovic <neskovic@gmail.com> Closes #5477
2016-12-16 04:31:33 +03:00
#define GEN_P_DEFINE() {}
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
#define GEN_P_P 0, 1, 2, 3
#define GEN_PQ_STRIDE 4
ABD: Adapt avx512bw raidz assembly Adapt avx512bw implementation for use with abd buffers. Mul2 implementation is rewritten to take advantage of the BW instruction set. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Romain Dolbeau <romain.dolbeau@atos.net> Signed-off-by: Gvozden Neskovic <neskovic@gmail.com> Closes #5477
2016-12-16 04:31:33 +03:00
#define GEN_PQ_DEFINE() {}
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
#define GEN_PQ_D 0, 1, 2, 3
ABD: Adapt avx512bw raidz assembly Adapt avx512bw implementation for use with abd buffers. Mul2 implementation is rewritten to take advantage of the BW instruction set. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Romain Dolbeau <romain.dolbeau@atos.net> Signed-off-by: Gvozden Neskovic <neskovic@gmail.com> Closes #5477
2016-12-16 04:31:33 +03:00
#define GEN_PQ_C 4, 5, 6, 7
#define GEN_PQR_STRIDE 4
#define GEN_PQR_DEFINE() {}
#define GEN_PQR_D 0, 1, 2, 3
#define GEN_PQR_C 4, 5, 6, 7
#define SYN_Q_DEFINE() {}
#define SYN_Q_D 0, 1, 2, 3
#define SYN_Q_X 4, 5, 6, 7
#define SYN_R_DEFINE() {}
#define SYN_R_D 0, 1, 2, 3
#define SYN_R_X 4, 5, 6, 7
#define SYN_PQ_DEFINE() {}
#define SYN_PQ_D 0, 1, 2, 3
#define SYN_PQ_X 4, 5, 6, 7
#define REC_PQ_STRIDE 2
#define REC_PQ_DEFINE() {}
#define REC_PQ_X 0, 1
#define REC_PQ_Y 2, 3
#define REC_PQ_T 4, 5
#define SYN_PR_DEFINE() {}
#define SYN_PR_D 0, 1, 2, 3
#define SYN_PR_X 4, 5, 6, 7
#define REC_PR_STRIDE 2
#define REC_PR_DEFINE() {}
#define REC_PR_X 0, 1
#define REC_PR_Y 2, 3
#define REC_PR_T 4, 5
#define SYN_QR_DEFINE() {}
#define SYN_QR_D 0, 1, 2, 3
#define SYN_QR_X 4, 5, 6, 7
#define REC_QR_STRIDE 2
#define REC_QR_DEFINE() {}
#define REC_QR_X 0, 1
#define REC_QR_Y 2, 3
#define REC_QR_T 4, 5
#define SYN_PQR_DEFINE() {}
#define SYN_PQR_D 0, 1, 2, 3
#define SYN_PQR_X 4, 5, 6, 7
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
#define REC_PQR_STRIDE 2
ABD: Adapt avx512bw raidz assembly Adapt avx512bw implementation for use with abd buffers. Mul2 implementation is rewritten to take advantage of the BW instruction set. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Romain Dolbeau <romain.dolbeau@atos.net> Signed-off-by: Gvozden Neskovic <neskovic@gmail.com> Closes #5477
2016-12-16 04:31:33 +03:00
#define REC_PQR_DEFINE() {}
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
#define REC_PQR_X 0, 1
#define REC_PQR_Y 2, 3
#define REC_PQR_Z 4, 5
#define REC_PQR_XS 6, 7
#define REC_PQR_YS 8, 9
#include <sys/vdev_raidz_impl.h>
#include "vdev_raidz_math_impl.h"
DEFINE_GEN_METHODS(avx512bw);
DEFINE_REC_METHODS(avx512bw);
static boolean_t
raidz_will_avx512bw_work(void)
{
Linux 5.0 compat: SIMD compatibility Restore the SIMD optimization for 4.19.38 LTS, 4.14.120 LTS, and 5.0 and newer kernels. This is accomplished by leveraging the fact that by definition dedicated kernel threads never need to concern themselves with saving and restoring the user FPU state. Therefore, they may use the FPU as long as we can guarantee user tasks always restore their FPU state before context switching back to user space. For the 5.0 and 5.1 kernels disabling preemption and local interrupts is sufficient to allow the FPU to be used. All non-kernel threads will restore the preserved user FPU state. For 5.2 and latter kernels the user FPU state restoration will be skipped if the kernel determines the registers have not changed. Therefore, for these kernels we need to perform the additional step of saving and restoring the FPU registers. Invalidating the per-cpu global tracking the FPU state would force a restore but that functionality is private to the core x86 FPU implementation and unavailable. In practice, restricting SIMD to kernel threads is not a major restriction for ZFS. The vast majority of SIMD operations are already performed by the IO pipeline. The remaining cases are relatively infrequent and can be handled by the generic code without significant impact. The two most noteworthy cases are: 1) Decrypting the wrapping key for an encrypted dataset, i.e. `zfs load-key`. All other encryption and decryption operations will use the SIMD optimized implementations. 2) Generating the payload checksums for a `zfs send` stream. In order to avoid making any changes to the higher layers of ZFS all of the `*_get_ops()` functions were updated to take in to consideration the calling context. This allows for the fastest implementation to be used as appropriate (see kfpu_allowed()). The only other notable instance of SIMD operations being used outside a kernel thread was at module load time. This code was moved in to a taskq in order to accommodate the new kernel thread restriction. Finally, a few other modifications were made in order to further harden this code and facilitate testing. They include updating each implementations operations structure to be declared as a constant. And allowing "cycle" to be set when selecting the preferred ops in the kernel as well as user space. Reviewed-by: Tony Hutter <hutter2@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #8754 Closes #8793 Closes #8965
2019-07-12 19:31:20 +03:00
return (kfpu_allowed() && zfs_avx_available() &&
zfs_avx512f_available() && zfs_avx512bw_available());
Add parity generation/rebuild using AVX-512 for x86-64 avx512f should work on all AVX512 hardware, since it only uses Foundation instructions. avx512bw should be faster on hardware supporting the AVW512BW extension. We can use full-width pshufb (instead of relying on the 256 bits AVX2 pshufb). As a side-effect, the code is also unrolled more. Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Gvozden Neskovic <neskovic@gmail.com> Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name> Closes #5219
2016-11-02 22:40:23 +03:00
}
const raidz_impl_ops_t vdev_raidz_avx512bw_impl = {
.init = NULL,
.fini = NULL,
.gen = RAIDZ_GEN_METHODS(avx512bw),
.rec = RAIDZ_REC_METHODS(avx512bw),
.is_supported = &raidz_will_avx512bw_work,
.name = "avx512bw"
};
#endif /* defined(__x86_64) && defined(HAVE_AVX512BW) */
Reference in New Issue Copy Permalink