freebsd-nq/module/zfs/vdev_raidz_math_avx2.c
Romain Dolbeau 62a65a654e Add parity generation/rebuild using 128-bits NEON for Aarch64
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
2016-10-03 09:44:00 -07:00

417 lines
11 KiB
C

/*
* 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 Gvozden Nešković. All rights reserved.
*/
#include <sys/isa_defs.h>
#if defined(__x86_64) && defined(HAVE_AVX2)
#include <sys/types.h>
#include <linux/simd_x86.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, ...) "ymm"#REG
#define VR1_(_1, REG, ...) "ymm"#REG
#define VR2_(_1, _2, REG, ...) "ymm"#REG
#define VR3_(_1, _2, _3, REG, ...) "ymm"#REG
#define VR4_(_1, _2, _3, _4, REG, ...) "ymm"#REG
#define VR5_(_1, _2, _3, _4, _5, REG, ...) "ymm"#REG
#define VR6_(_1, _2, _3, _4, _5, _6, REG, ...) "ymm"#REG
#define VR7_(_1, _2, _3, _4, _5, _6, _7, REG, ...) "ymm"#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)
#define ASM_BUG() ASSERT(0)
extern const uint8_t gf_clmul_mod_lt[4*256][16];
#define ELEM_SIZE 32
typedef struct v {
uint8_t b[ELEM_SIZE] __attribute__((aligned(ELEM_SIZE)));
} v_t;
#define PREFETCHNTA(ptr, offset) \
{ \
__asm( \
"prefetchnta " #offset "(%[MEM])\n" \
: : [MEM] "r" (ptr)); \
}
#define PREFETCH(ptr, offset) \
{ \
__asm( \
"prefetcht0 " #offset "(%[MEM])\n" \
: : [MEM] "r" (ptr)); \
}
#define XOR_ACC(src, r...) \
{ \
switch (REG_CNT(r)) { \
case 4: \
__asm( \
"vpxor 0x00(%[SRC]), %%" VR0(r)", %%" VR0(r) "\n" \
"vpxor 0x20(%[SRC]), %%" VR1(r)", %%" VR1(r) "\n" \
"vpxor 0x40(%[SRC]), %%" VR2(r)", %%" VR2(r) "\n" \
"vpxor 0x60(%[SRC]), %%" VR3(r)", %%" VR3(r) "\n" \
: : [SRC] "r" (src)); \
break; \
case 2: \
__asm( \
"vpxor 0x00(%[SRC]), %%" VR0(r)", %%" VR0(r) "\n" \
"vpxor 0x20(%[SRC]), %%" VR1(r)", %%" VR1(r) "\n" \
: : [SRC] "r" (src)); \
break; \
default: \
ASM_BUG(); \
} \
}
#define XOR(r...) \
{ \
switch (REG_CNT(r)) { \
case 8: \
__asm( \
"vpxor %" VR0(r) ", %" VR4(r)", %" VR4(r) "\n" \
"vpxor %" VR1(r) ", %" VR5(r)", %" VR5(r) "\n" \
"vpxor %" VR2(r) ", %" VR6(r)", %" VR6(r) "\n" \
"vpxor %" VR3(r) ", %" VR7(r)", %" VR7(r)); \
break; \
case 4: \
__asm( \
"vpxor %" VR0(r) ", %" VR2(r)", %" VR2(r) "\n" \
"vpxor %" VR1(r) ", %" VR3(r)", %" VR3(r)); \
break; \
default: \
ASM_BUG(); \
} \
}
#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)) { \
case 8: \
__asm( \
"vmovdqa %" VR0(r) ", %" VR4(r) "\n" \
"vmovdqa %" VR1(r) ", %" VR5(r) "\n" \
"vmovdqa %" VR2(r) ", %" VR6(r) "\n" \
"vmovdqa %" VR3(r) ", %" VR7(r)); \
break; \
case 4: \
__asm( \
"vmovdqa %" VR0(r) ", %" VR2(r) "\n" \
"vmovdqa %" VR1(r) ", %" VR3(r)); \
break; \
default: \
ASM_BUG(); \
} \
}
#define LOAD(src, r...) \
{ \
switch (REG_CNT(r)) { \
case 4: \
__asm( \
"vmovdqa 0x00(%[SRC]), %%" VR0(r) "\n" \
"vmovdqa 0x20(%[SRC]), %%" VR1(r) "\n" \
"vmovdqa 0x40(%[SRC]), %%" VR2(r) "\n" \
"vmovdqa 0x60(%[SRC]), %%" VR3(r) "\n" \
: : [SRC] "r" (src)); \
break; \
case 2: \
__asm( \
"vmovdqa 0x00(%[SRC]), %%" VR0(r) "\n" \
"vmovdqa 0x20(%[SRC]), %%" VR1(r) "\n" \
: : [SRC] "r" (src)); \
break; \
default: \
ASM_BUG(); \
} \
}
#define STORE(dst, r...) \
{ \
switch (REG_CNT(r)) { \
case 4: \
__asm( \
"vmovdqa %%" VR0(r) ", 0x00(%[DST])\n" \
"vmovdqa %%" VR1(r) ", 0x20(%[DST])\n" \
"vmovdqa %%" VR2(r) ", 0x40(%[DST])\n" \
"vmovdqa %%" VR3(r) ", 0x60(%[DST])\n" \
: : [DST] "r" (dst)); \
break; \
case 2: \
__asm( \
"vmovdqa %%" VR0(r) ", 0x00(%[DST])\n" \
"vmovdqa %%" VR1(r) ", 0x20(%[DST])\n" \
: : [DST] "r" (dst)); \
break; \
default: \
ASM_BUG(); \
} \
}
#define FLUSH() \
{ \
__asm("vzeroupper"); \
}
#define MUL2_SETUP() \
{ \
__asm("vmovq %0, %%xmm14" :: "r"(0x1d1d1d1d1d1d1d1d)); \
__asm("vpbroadcastq %xmm14, %ymm14"); \
__asm("vpxor %ymm15, %ymm15 ,%ymm15"); \
}
#define _MUL2(r...) \
{ \
switch (REG_CNT(r)) { \
case 2: \
__asm( \
"vpcmpgtb %" VR0(r)", %ymm15, %ymm12\n" \
"vpcmpgtb %" VR1(r)", %ymm15, %ymm13\n" \
"vpaddb %" VR0(r)", %" VR0(r)", %" VR0(r) "\n" \
"vpaddb %" VR1(r)", %" VR1(r)", %" VR1(r) "\n" \
"vpand %ymm14, %ymm12, %ymm12\n" \
"vpand %ymm14, %ymm13, %ymm13\n" \
"vpxor %ymm12, %" VR0(r)", %" VR0(r) "\n" \
"vpxor %ymm13, %" VR1(r)", %" VR1(r)); \
break; \
default: \
ASM_BUG(); \
} \
}
#define MUL2(r...) \
{ \
switch (REG_CNT(r)) { \
case 4: \
_MUL2(R_01(r)); \
_MUL2(R_23(r)); \
break; \
case 2: \
_MUL2(r); \
break; \
default: \
ASM_BUG(); \
} \
}
#define MUL4(r...) \
{ \
MUL2(r); \
MUL2(r); \
}
#define _0f "ymm15"
#define _as "ymm14"
#define _bs "ymm13"
#define _ltmod "ymm12"
#define _ltmul "ymm11"
#define _ta "ymm10"
#define _tb "ymm15"
static const uint8_t __attribute__((aligned(32))) _mul_mask = 0x0F;
#define _MULx2(c, r...) \
{ \
switch (REG_CNT(r)) { \
case 2: \
__asm( \
"vpbroadcastb (%[mask]), %%" _0f "\n" \
/* upper bits */ \
"vbroadcasti128 0x00(%[lt]), %%" _ltmod "\n" \
"vbroadcasti128 0x10(%[lt]), %%" _ltmul "\n" \
\
"vpsraw $0x4, %%" VR0(r) ", %%"_as "\n" \
"vpsraw $0x4, %%" VR1(r) ", %%"_bs "\n" \
"vpand %%" _0f ", %%" VR0(r) ", %%" VR0(r) "\n" \
"vpand %%" _0f ", %%" VR1(r) ", %%" VR1(r) "\n" \
"vpand %%" _0f ", %%" _as ", %%" _as "\n" \
"vpand %%" _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 */ \
"vbroadcasti128 0x20(%[lt]), %%" _ltmod "\n" \
"vbroadcasti128 0x30(%[lt]), %%" _ltmul "\n" \
\
"vpxor %%" _ta ", %%" _as ", %%" _as "\n" \
"vpxor %%" _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"\
\
"vpxor %%" _ta ", %%" VR0(r) ", %%" VR0(r) "\n" \
"vpxor %%" _as ", %%" VR0(r) ", %%" VR0(r) "\n" \
"vpxor %%" _tb ", %%" VR1(r) ", %%" VR1(r) "\n" \
"vpxor %%" _bs ", %%" VR1(r) ", %%" VR1(r) "\n" \
: : [mask] "r" (&_mul_mask), \
[lt] "r" (gf_clmul_mod_lt[4*(c)])); \
break; \
default: \
ASM_BUG(); \
} \
}
#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: \
ASM_BUG(); \
} \
}
#define raidz_math_begin() kfpu_begin()
#define raidz_math_end() \
{ \
FLUSH(); \
kfpu_end(); \
}
#define GEN_P_DEFINE() {}
#define GEN_P_STRIDE 4
#define GEN_P_P 0, 1, 2, 3
#define GEN_PQ_DEFINE() {}
#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() {}
#define GEN_PQR_STRIDE 2
#define GEN_PQR_D 0, 1
#define GEN_PQR_P 2, 3
#define GEN_PQR_Q 4, 5
#define GEN_PQR_R 6, 7
#define REC_P_DEFINE() {}
#define REC_P_STRIDE 4
#define REC_P_X 0, 1, 2, 3
#define REC_Q_DEFINE() {}
#define REC_Q_STRIDE 4
#define REC_Q_X 0, 1, 2, 3
#define REC_R_DEFINE() {}
#define REC_R_STRIDE 4
#define REC_R_X 0, 1, 2, 3
#define REC_PQ_DEFINE() {}
#define REC_PQ_STRIDE 2
#define REC_PQ_X 0, 1
#define REC_PQ_Y 2, 3
#define REC_PQ_D 4, 5
#define REC_PR_DEFINE() {}
#define REC_PR_STRIDE 2
#define REC_PR_X 0, 1
#define REC_PR_Y 2, 3
#define REC_PR_D 4, 5
#define REC_QR_DEFINE() {}
#define REC_QR_STRIDE 2
#define REC_QR_X 0, 1
#define REC_QR_Y 2, 3
#define REC_QR_D 4, 5
#define REC_PQR_DEFINE() {}
#define REC_PQR_STRIDE 2
#define REC_PQR_X 0, 1
#define REC_PQR_Y 2, 3
#define REC_PQR_Z 4, 5
#define REC_PQR_D 6, 7
#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(avx2);
DEFINE_REC_METHODS(avx2);
static boolean_t
raidz_will_avx2_work(void)
{
return (zfs_avx_available() && zfs_avx2_available());
}
const raidz_impl_ops_t vdev_raidz_avx2_impl = {
.init = NULL,
.fini = NULL,
.gen = RAIDZ_GEN_METHODS(avx2),
.rec = RAIDZ_REC_METHODS(avx2),
.is_supported = &raidz_will_avx2_work,
.name = "avx2"
};
#endif /* defined(__x86_64) && defined(HAVE_AVX2) */