mirror_zfs/module/zcommon/zfs_fletcher_avx512.c

160 lines
4.5 KiB
C
Raw Normal View History

/*
* 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.
*/
#if defined(__x86_64) && defined(HAVE_AVX512F)
#include <linux/simd_x86.h>
#include <sys/byteorder.h>
#include <sys/spa_checksum.h>
#include <zfs_fletcher.h>
#define __asm __asm__ __volatile__
typedef struct {
uint64_t v[8] __attribute__((aligned(64)));
} zfs_avx512_t;
static void
fletcher_4_avx512f_init(zio_cksum_t *zcp)
{
kfpu_begin();
/* clear registers */
__asm("vpxorq %zmm0, %zmm0, %zmm0");
__asm("vpxorq %zmm1, %zmm1, %zmm1");
__asm("vpxorq %zmm2, %zmm2, %zmm2");
__asm("vpxorq %zmm3, %zmm3, %zmm3");
}
static void
Rework of fletcher_4 module - Benchmark memory block is increased to 128kiB to reflect real block sizes more accurately. Measurements include all three stages needed for checksum generation, i.e. `init()/compute()/fini()`. The inner loop is repeated multiple times to offset overhead of time function. - Fastest implementation selects native and byteswap methods independently in benchmark. To support this new function pointers `init_byteswap()/fini_byteswap()` are introduced. - Implementation mutex lock is replaced by atomic variable. - To save time, benchmark is not executed in userspace. Instead, highest supported implementation is used for fastest. Default userspace selector is still 'cycle'. - `fletcher_4_native/byteswap()` methods use incremental methods to finish calculation if data size is not multiple of vector stride (currently 64B). - Added `fletcher_4_native_varsize()` special purpose method for use when buffer size is not known in advance. The method does not enforce 4B alignment on buffer size, and will ignore last (size % 4) bytes of the data buffer. - Benchmark `kstat` is changed to match the one of vdev_raidz. It now shows throughput for all supported implementations (in B/s), native and byteswap, as well as the code [fastest] is running. Example of `fletcher_4_bench` running on `Intel(R) Xeon(R) CPU E5-2660 v3 @ 2.60GHz`: implementation native byteswap scalar 4768120823 3426105750 sse2 7947841777 4318964249 ssse3 7951922722 6112191941 avx2 13269714358 11043200912 fastest avx2 avx2 Example of `fletcher_4_bench` running on `Intel(R) Xeon Phi(TM) CPU 7210 @ 1.30GHz`: implementation native byteswap scalar 1291115967 1031555336 sse2 2539571138 1280970926 ssse3 2537778746 1080016762 avx2 4950749767 1078493449 avx512f 9581379998 4010029046 fastest avx512f avx512f Signed-off-by: Gvozden Neskovic <neskovic@gmail.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #4952
2016-07-12 18:50:54 +03:00
fletcher_4_avx512f_native(const void *buf, uint64_t size, zio_cksum_t *unused)
{
const uint32_t *ip = buf;
const uint32_t *ipend = (uint32_t *)((uint8_t *)ip + size);
for (; ip < ipend; ip += 8) {
__asm("vpmovzxdq %0, %%zmm4"::"m" (*ip));
__asm("vpaddq %zmm4, %zmm0, %zmm0");
__asm("vpaddq %zmm0, %zmm1, %zmm1");
__asm("vpaddq %zmm1, %zmm2, %zmm2");
__asm("vpaddq %zmm2, %zmm3, %zmm3");
}
}
static void
fletcher_4_avx512f_byteswap(const void *buf, uint64_t size, zio_cksum_t *unused)
{
static const uint64_t byteswap_mask = 0xFFULL;
const uint32_t *ip = buf;
const uint32_t *ipend = (uint32_t *)((uint8_t *)ip + size);
__asm("vpbroadcastq %0, %%zmm8" :: "r" (byteswap_mask));
__asm("vpsllq $8, %zmm8, %zmm9");
__asm("vpsllq $16, %zmm8, %zmm10");
__asm("vpsllq $24, %zmm8, %zmm11");
for (; ip < ipend; ip += 8) {
__asm("vpmovzxdq %0, %%zmm5"::"m" (*ip));
__asm("vpsrlq $24, %zmm5, %zmm6");
__asm("vpandd %zmm8, %zmm6, %zmm6");
__asm("vpsrlq $8, %zmm5, %zmm7");
__asm("vpandd %zmm9, %zmm7, %zmm7");
__asm("vpord %zmm6, %zmm7, %zmm4");
__asm("vpsllq $8, %zmm5, %zmm6");
__asm("vpandd %zmm10, %zmm6, %zmm6");
__asm("vpord %zmm6, %zmm4, %zmm4");
__asm("vpsllq $24, %zmm5, %zmm5");
__asm("vpandd %zmm11, %zmm5, %zmm5");
__asm("vpord %zmm5, %zmm4, %zmm4");
__asm("vpaddq %zmm4, %zmm0, %zmm0");
__asm("vpaddq %zmm0, %zmm1, %zmm1");
__asm("vpaddq %zmm1, %zmm2, %zmm2");
__asm("vpaddq %zmm2, %zmm3, %zmm3");
}
}
static void
fletcher_4_avx512f_fini(zio_cksum_t *zcp)
{
static const uint64_t
CcA[] = { 0, 0, 1, 3, 6, 10, 15, 21 },
CcB[] = { 28, 36, 44, 52, 60, 68, 76, 84 },
DcA[] = { 0, 0, 0, 1, 4, 10, 20, 35 },
DcB[] = { 56, 84, 120, 164, 216, 276, 344, 420 },
DcC[] = { 448, 512, 576, 640, 704, 768, 832, 896 };
zfs_avx512_t a, b, c, b8, c64, d512;
uint64_t A, B, C, D;
uint64_t i;
__asm("vmovdqu64 %%zmm0, %0":"=m" (a));
__asm("vmovdqu64 %%zmm1, %0":"=m" (b));
__asm("vmovdqu64 %%zmm2, %0":"=m" (c));
__asm("vpsllq $3, %zmm1, %zmm1");
__asm("vpsllq $6, %zmm2, %zmm2");
__asm("vpsllq $9, %zmm3, %zmm3");
__asm("vmovdqu64 %%zmm1, %0":"=m" (b8));
__asm("vmovdqu64 %%zmm2, %0":"=m" (c64));
__asm("vmovdqu64 %%zmm3, %0":"=m" (d512));
kfpu_end();
A = a.v[0];
B = b8.v[0];
C = c64.v[0] - CcB[0] * b.v[0];
D = d512.v[0] - DcC[0] * c.v[0] + DcB[0] * b.v[0];
for (i = 1; i < 8; i++) {
A += a.v[i];
B += b8.v[i] - i * a.v[i];
C += c64.v[i] - CcB[i] * b.v[i] + CcA[i] * a.v[i];
D += d512.v[i] - DcC[i] * c.v[i] + DcB[i] * b.v[i] -
DcA[i] * a.v[i];
}
ZIO_SET_CHECKSUM(zcp, A, B, C, D);
}
static boolean_t
fletcher_4_avx512f_valid(void)
{
return (zfs_avx512f_available());
}
const fletcher_4_ops_t fletcher_4_avx512f_ops = {
Rework of fletcher_4 module - Benchmark memory block is increased to 128kiB to reflect real block sizes more accurately. Measurements include all three stages needed for checksum generation, i.e. `init()/compute()/fini()`. The inner loop is repeated multiple times to offset overhead of time function. - Fastest implementation selects native and byteswap methods independently in benchmark. To support this new function pointers `init_byteswap()/fini_byteswap()` are introduced. - Implementation mutex lock is replaced by atomic variable. - To save time, benchmark is not executed in userspace. Instead, highest supported implementation is used for fastest. Default userspace selector is still 'cycle'. - `fletcher_4_native/byteswap()` methods use incremental methods to finish calculation if data size is not multiple of vector stride (currently 64B). - Added `fletcher_4_native_varsize()` special purpose method for use when buffer size is not known in advance. The method does not enforce 4B alignment on buffer size, and will ignore last (size % 4) bytes of the data buffer. - Benchmark `kstat` is changed to match the one of vdev_raidz. It now shows throughput for all supported implementations (in B/s), native and byteswap, as well as the code [fastest] is running. Example of `fletcher_4_bench` running on `Intel(R) Xeon(R) CPU E5-2660 v3 @ 2.60GHz`: implementation native byteswap scalar 4768120823 3426105750 sse2 7947841777 4318964249 ssse3 7951922722 6112191941 avx2 13269714358 11043200912 fastest avx2 avx2 Example of `fletcher_4_bench` running on `Intel(R) Xeon Phi(TM) CPU 7210 @ 1.30GHz`: implementation native byteswap scalar 1291115967 1031555336 sse2 2539571138 1280970926 ssse3 2537778746 1080016762 avx2 4950749767 1078493449 avx512f 9581379998 4010029046 fastest avx512f avx512f Signed-off-by: Gvozden Neskovic <neskovic@gmail.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #4952
2016-07-12 18:50:54 +03:00
.init_native = fletcher_4_avx512f_init,
.fini_native = fletcher_4_avx512f_fini,
.compute_native = fletcher_4_avx512f_native,
.init_byteswap = fletcher_4_avx512f_init,
.fini_byteswap = fletcher_4_avx512f_fini,
.compute_byteswap = fletcher_4_avx512f_byteswap,
.valid = fletcher_4_avx512f_valid,
.name = "avx512f"
};
#endif /* defined(__x86_64) && defined(HAVE_AVX512F) */