zfs/module/zcommon/zfs_fletcher_avx512.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (C) 2016 Gvozden Nešković. All rights reserved.
  23  */
  24
  25 #if defined(__x86_64) && defined(HAVE_AVX512F)
  26
  27 #include <linux/simd_x86.h>
  28 #include <sys/byteorder.h>
  29 #include <sys/spa_checksum.h>
  30 #include <zfs_fletcher.h>
  31 #include <strings.h>
  32
  33 #define __asm __asm__ __volatile__
  34
  35 static void
  36 fletcher_4_avx512f_init(fletcher_4_ctx_t *ctx)
  37 {
  38         bzero(ctx->avx512, 4 * sizeof (zfs_fletcher_avx512_t));
  39 }
  40
  41 static void
  42 fletcher_4_avx512f_fini(fletcher_4_ctx_t *ctx, zio_cksum_t *zcp)
  43 {
  44         static const uint64_t
  45         CcA[] = {   0,   0,   1,   3,   6,  10,  15,  21 },
  46         CcB[] = {  28,  36,  44,  52,  60,  68,  76,  84 },
  47         DcA[] = {   0,   0,   0,   1,   4,  10,  20,  35 },
  48         DcB[] = {  56,  84, 120, 164, 216, 276, 344, 420 },
  49         DcC[] = { 448, 512, 576, 640, 704, 768, 832, 896 };
  50
  51         uint64_t A, B, C, D;
  52         uint64_t i;
  53
  54         A = ctx->avx512[0].v[0];
  55         B = 8 * ctx->avx512[1].v[0];
  56         C = 64 * ctx->avx512[2].v[0] - CcB[0] * ctx->avx512[1].v[0];
  57         D = 512 * ctx->avx512[3].v[0] - DcC[0] * ctx->avx512[2].v[0] +
  58             DcB[0] * ctx->avx512[1].v[0];
  59
  60         for (i = 1; i < 8; i++) {
  61                 A += ctx->avx512[0].v[i];
  62                 B += 8 * ctx->avx512[1].v[i] - i * ctx->avx512[0].v[i];
  63                 C += 64 * ctx->avx512[2].v[i] - CcB[i] * ctx->avx512[1].v[i] +
  64                     CcA[i] * ctx->avx512[0].v[i];
  65                 D += 512 * ctx->avx512[3].v[i] - DcC[i] * ctx->avx512[2].v[i] +
  66                     DcB[i] * ctx->avx512[1].v[i] - DcA[i] * ctx->avx512[0].v[i];
  67         }
  68
  69         ZIO_SET_CHECKSUM(zcp, A, B, C, D);
  70 }
  71
  72 #define FLETCHER_4_AVX512_RESTORE_CTX(ctx)                              \
  73 {                                                                       \
  74         __asm("vmovdqu64 %0, %%zmm0" :: "m" ((ctx)->avx512[0]));        \
  75         __asm("vmovdqu64 %0, %%zmm1" :: "m" ((ctx)->avx512[1]));        \
  76         __asm("vmovdqu64 %0, %%zmm2" :: "m" ((ctx)->avx512[2]));        \
  77         __asm("vmovdqu64 %0, %%zmm3" :: "m" ((ctx)->avx512[3]));        \
  78 }
  79
  80 #define FLETCHER_4_AVX512_SAVE_CTX(ctx)                                 \
  81 {                                                                       \
  82         __asm("vmovdqu64 %%zmm0, %0" : "=m" ((ctx)->avx512[0]));        \
  83         __asm("vmovdqu64 %%zmm1, %0" : "=m" ((ctx)->avx512[1]));        \
  84         __asm("vmovdqu64 %%zmm2, %0" : "=m" ((ctx)->avx512[2]));        \
  85         __asm("vmovdqu64 %%zmm3, %0" : "=m" ((ctx)->avx512[3]));        \
  86 }
  87
  88 static void
  89 fletcher_4_avx512f_native(fletcher_4_ctx_t *ctx, const void *buf, uint64_t size)
  90 {
  91         const uint32_t *ip = buf;
  92         const uint32_t *ipend = (uint32_t *)((uint8_t *)ip + size);
  93
  94         kfpu_begin();
  95
  96         FLETCHER_4_AVX512_RESTORE_CTX(ctx);
  97
  98         for (; ip < ipend; ip += 8) {
  99                 __asm("vpmovzxdq %0, %%zmm4"::"m" (*ip));
 100                 __asm("vpaddq %zmm4, %zmm0, %zmm0");
 101                 __asm("vpaddq %zmm0, %zmm1, %zmm1");
 102                 __asm("vpaddq %zmm1, %zmm2, %zmm2");
 103                 __asm("vpaddq %zmm2, %zmm3, %zmm3");
 104         }
 105
 106         FLETCHER_4_AVX512_SAVE_CTX(ctx);
 107
 108         kfpu_end();
 109 }
 110
 111 static void
 112 fletcher_4_avx512f_byteswap(fletcher_4_ctx_t *ctx, const void *buf,
 113     uint64_t size)
 114 {
 115         static const uint64_t byteswap_mask = 0xFFULL;
 116         const uint32_t *ip = buf;
 117         const uint32_t *ipend = (uint32_t *)((uint8_t *)ip + size);
 118
 119         kfpu_begin();
 120
 121         FLETCHER_4_AVX512_RESTORE_CTX(ctx);
 122
 123         __asm("vpbroadcastq %0, %%zmm8" :: "r" (byteswap_mask));
 124         __asm("vpsllq $8, %zmm8, %zmm9");
 125         __asm("vpsllq $16, %zmm8, %zmm10");
 126         __asm("vpsllq $24, %zmm8, %zmm11");
 127
 128         for (; ip < ipend; ip += 8) {
 129                 __asm("vpmovzxdq %0, %%zmm5"::"m" (*ip));
 130
 131                 __asm("vpsrlq $24, %zmm5, %zmm6");
 132                 __asm("vpandd %zmm8, %zmm6, %zmm6");
 133                 __asm("vpsrlq $8, %zmm5, %zmm7");
 134                 __asm("vpandd %zmm9, %zmm7, %zmm7");
 135                 __asm("vpord %zmm6, %zmm7, %zmm4");
 136                 __asm("vpsllq $8, %zmm5, %zmm6");
 137                 __asm("vpandd %zmm10, %zmm6, %zmm6");
 138                 __asm("vpord %zmm6, %zmm4, %zmm4");
 139                 __asm("vpsllq $24, %zmm5, %zmm5");
 140                 __asm("vpandd %zmm11, %zmm5, %zmm5");
 141                 __asm("vpord %zmm5, %zmm4, %zmm4");
 142
 143                 __asm("vpaddq %zmm4, %zmm0, %zmm0");
 144                 __asm("vpaddq %zmm0, %zmm1, %zmm1");
 145                 __asm("vpaddq %zmm1, %zmm2, %zmm2");
 146                 __asm("vpaddq %zmm2, %zmm3, %zmm3");
 147         }
 148
 149         FLETCHER_4_AVX512_SAVE_CTX(ctx)
 150
 151         kfpu_end();
 152 }
 153
 154 static boolean_t
 155 fletcher_4_avx512f_valid(void)
 156 {
 157         return (zfs_avx512f_available());
 158 }
 159
 160 const fletcher_4_ops_t fletcher_4_avx512f_ops = {
 161         .init_native = fletcher_4_avx512f_init,
 162         .fini_native = fletcher_4_avx512f_fini,
 163         .compute_native = fletcher_4_avx512f_native,
 164         .init_byteswap = fletcher_4_avx512f_init,
 165         .fini_byteswap = fletcher_4_avx512f_fini,
 166         .compute_byteswap = fletcher_4_avx512f_byteswap,
 167         .valid = fletcher_4_avx512f_valid,
 168         .name = "avx512f"
 169 };
 170
 171 #endif /* defined(__x86_64) && defined(HAVE_AVX512F) */