mirror of
https://github.com/amadvance/snapraid.git
synced 2026-10-26 10:51:54 -04:00
1046 lines
36 KiB
C
1046 lines
36 KiB
C
// SPDX-License-Identifier: GPL-2.0-or-later
|
|
// Copyright (C) 2026 Andrea Mazzoleni
|
|
|
|
#include "internal.h"
|
|
#include "gf.h"
|
|
#include "cpu.h"
|
|
|
|
#ifdef CONFIG_X86_64
|
|
/*
|
|
* Generate one parity block (RAID5 with XOR) using AVX512BW implementation.
|
|
*
|
|
* Processes two disks per iteration using vpternlogq over 64-byte blocks.
|
|
*
|
|
* Note that in true AVX512F would suffice, but we don't want to add
|
|
* specific support for AVX512F because this would be the only function
|
|
* to benefit from that.
|
|
*/
|
|
static __always_inline void raid_gen1_avx512bw_gen(int nd, size_t size, void **vv, int streaming)
|
|
{
|
|
uint8_t **v = (uint8_t **)vv;
|
|
uint8_t *p;
|
|
int d, l;
|
|
size_t i;
|
|
|
|
l = nd - 1;
|
|
p = v[nd];
|
|
|
|
raid_avx_begin();
|
|
|
|
for (i = 0; i < size; i += 64) {
|
|
asm volatile ("vmovdqa64 %0,%%zmm0" : : "m" (v[0][i]));
|
|
|
|
for (d = 1; d <= l - 1; d += 2) {
|
|
asm volatile (
|
|
"vmovdqa64 %0,%%zmm1\n\t"
|
|
"vpternlogq $0x96,%1,%%zmm1,%%zmm0"
|
|
:
|
|
: "m" (v[d][i]), "m" (v[d + 1][i])
|
|
);
|
|
}
|
|
|
|
if (d == l) {
|
|
asm volatile ("vpxorq %0,%%zmm0,%%zmm0" : : "m" (v[l][i]));
|
|
}
|
|
|
|
if (streaming)
|
|
asm volatile ("vmovntdq %%zmm0,%0" : "=m" (p[i]));
|
|
else
|
|
asm volatile ("vmovdqa64 %%zmm0,%0" : "=m" (p[i]));
|
|
}
|
|
|
|
raid_avx_end(streaming);
|
|
}
|
|
|
|
/*
|
|
* Generate two parity blocks (RAID6 with Cauchy matrix) using AVX512BW implementation.
|
|
*
|
|
* Preloads 13 pairs of Q coefficient tables in ZMM registers (zmm4..zmm29)
|
|
* with an unrolled disk loop and early exits.
|
|
*/
|
|
static __always_inline void raid_gen2_avx512bw_gen(int nd, size_t size, void **vv, int streaming)
|
|
{
|
|
uint8_t **v = (uint8_t **)vv;
|
|
size_t i;
|
|
int d;
|
|
|
|
if (nd == 1) {
|
|
if (v[1] != v[0])
|
|
memcpy(v[1], v[0], size);
|
|
if (v[2] != v[0])
|
|
memcpy(v[2], v[0], size);
|
|
return;
|
|
}
|
|
|
|
raid_avx_begin();
|
|
|
|
asm volatile ("vpbroadcastb %0,%%zmm31" : : "m" (gfconst16.low4[0]));
|
|
|
|
/* preload as many Q coefficient tables as possible */
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm4" : : "m" (raid_gfcauchypshufb[1][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm5" : : "m" (raid_gfcauchypshufb[1][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm6" : : "m" (raid_gfcauchypshufb[2][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm7" : : "m" (raid_gfcauchypshufb[2][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm8" : : "m" (raid_gfcauchypshufb[3][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm9" : : "m" (raid_gfcauchypshufb[3][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm10" : : "m" (raid_gfcauchypshufb[4][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm11" : : "m" (raid_gfcauchypshufb[4][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm12" : : "m" (raid_gfcauchypshufb[5][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm13" : : "m" (raid_gfcauchypshufb[5][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm14" : : "m" (raid_gfcauchypshufb[6][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm15" : : "m" (raid_gfcauchypshufb[6][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm16" : : "m" (raid_gfcauchypshufb[7][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm17" : : "m" (raid_gfcauchypshufb[7][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm18" : : "m" (raid_gfcauchypshufb[8][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm19" : : "m" (raid_gfcauchypshufb[8][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm20" : : "m" (raid_gfcauchypshufb[9][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm21" : : "m" (raid_gfcauchypshufb[9][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm22" : : "m" (raid_gfcauchypshufb[10][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm23" : : "m" (raid_gfcauchypshufb[10][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm24" : : "m" (raid_gfcauchypshufb[11][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm25" : : "m" (raid_gfcauchypshufb[11][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm26" : : "m" (raid_gfcauchypshufb[12][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm27" : : "m" (raid_gfcauchypshufb[12][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm28" : : "m" (raid_gfcauchypshufb[13][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm29" : : "m" (raid_gfcauchypshufb[13][0][1][0]));
|
|
|
|
for (i = 0; i < size; i += 64) {
|
|
asm volatile ("vmovdqa64 %0,%%zmm0" : : "m" (v[0][i]));
|
|
asm volatile ("vmovdqa64 %zmm0,%zmm1");
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm2" : : "m" (v[1][i]));
|
|
asm volatile ("vpxorq %zmm2,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm2,%zmm3");
|
|
asm volatile ("vpandq %zmm31,%zmm2,%zmm2");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm3");
|
|
asm volatile ("vpshufb %zmm2,%zmm4,%zmm2");
|
|
asm volatile ("vpshufb %zmm3,%zmm5,%zmm3");
|
|
asm volatile ("vpternlogq $0x96,%zmm2,%zmm3,%zmm1");
|
|
if (nd == 2)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm2" : : "m" (v[2][i]));
|
|
asm volatile ("vpxorq %zmm2,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm2,%zmm3");
|
|
asm volatile ("vpandq %zmm31,%zmm2,%zmm2");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm3");
|
|
asm volatile ("vpshufb %zmm2,%zmm6,%zmm2");
|
|
asm volatile ("vpshufb %zmm3,%zmm7,%zmm3");
|
|
asm volatile ("vpternlogq $0x96,%zmm2,%zmm3,%zmm1");
|
|
if (nd == 3)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm2" : : "m" (v[3][i]));
|
|
asm volatile ("vpxorq %zmm2,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm2,%zmm3");
|
|
asm volatile ("vpandq %zmm31,%zmm2,%zmm2");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm3");
|
|
asm volatile ("vpshufb %zmm2,%zmm8,%zmm2");
|
|
asm volatile ("vpshufb %zmm3,%zmm9,%zmm3");
|
|
asm volatile ("vpternlogq $0x96,%zmm2,%zmm3,%zmm1");
|
|
if (nd == 4)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm2" : : "m" (v[4][i]));
|
|
asm volatile ("vpxorq %zmm2,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm2,%zmm3");
|
|
asm volatile ("vpandq %zmm31,%zmm2,%zmm2");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm3");
|
|
asm volatile ("vpshufb %zmm2,%zmm10,%zmm2");
|
|
asm volatile ("vpshufb %zmm3,%zmm11,%zmm3");
|
|
asm volatile ("vpternlogq $0x96,%zmm2,%zmm3,%zmm1");
|
|
if (nd == 5)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm2" : : "m" (v[5][i]));
|
|
asm volatile ("vpxorq %zmm2,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm2,%zmm3");
|
|
asm volatile ("vpandq %zmm31,%zmm2,%zmm2");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm3");
|
|
asm volatile ("vpshufb %zmm2,%zmm12,%zmm2");
|
|
asm volatile ("vpshufb %zmm3,%zmm13,%zmm3");
|
|
asm volatile ("vpternlogq $0x96,%zmm2,%zmm3,%zmm1");
|
|
if (nd == 6)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm2" : : "m" (v[6][i]));
|
|
asm volatile ("vpxorq %zmm2,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm2,%zmm3");
|
|
asm volatile ("vpandq %zmm31,%zmm2,%zmm2");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm3");
|
|
asm volatile ("vpshufb %zmm2,%zmm14,%zmm2");
|
|
asm volatile ("vpshufb %zmm3,%zmm15,%zmm3");
|
|
asm volatile ("vpternlogq $0x96,%zmm2,%zmm3,%zmm1");
|
|
if (nd == 7)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm2" : : "m" (v[7][i]));
|
|
asm volatile ("vpxorq %zmm2,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm2,%zmm3");
|
|
asm volatile ("vpandq %zmm31,%zmm2,%zmm2");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm3");
|
|
asm volatile ("vpshufb %zmm2,%zmm16,%zmm2");
|
|
asm volatile ("vpshufb %zmm3,%zmm17,%zmm3");
|
|
asm volatile ("vpternlogq $0x96,%zmm2,%zmm3,%zmm1");
|
|
if (nd == 8)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm2" : : "m" (v[8][i]));
|
|
asm volatile ("vpxorq %zmm2,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm2,%zmm3");
|
|
asm volatile ("vpandq %zmm31,%zmm2,%zmm2");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm3");
|
|
asm volatile ("vpshufb %zmm2,%zmm18,%zmm2");
|
|
asm volatile ("vpshufb %zmm3,%zmm19,%zmm3");
|
|
asm volatile ("vpternlogq $0x96,%zmm2,%zmm3,%zmm1");
|
|
if (nd == 9)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm2" : : "m" (v[9][i]));
|
|
asm volatile ("vpxorq %zmm2,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm2,%zmm3");
|
|
asm volatile ("vpandq %zmm31,%zmm2,%zmm2");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm3");
|
|
asm volatile ("vpshufb %zmm2,%zmm20,%zmm2");
|
|
asm volatile ("vpshufb %zmm3,%zmm21,%zmm3");
|
|
asm volatile ("vpternlogq $0x96,%zmm2,%zmm3,%zmm1");
|
|
if (nd == 10)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm2" : : "m" (v[10][i]));
|
|
asm volatile ("vpxorq %zmm2,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm2,%zmm3");
|
|
asm volatile ("vpandq %zmm31,%zmm2,%zmm2");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm3");
|
|
asm volatile ("vpshufb %zmm2,%zmm22,%zmm2");
|
|
asm volatile ("vpshufb %zmm3,%zmm23,%zmm3");
|
|
asm volatile ("vpternlogq $0x96,%zmm2,%zmm3,%zmm1");
|
|
if (nd == 11)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm2" : : "m" (v[11][i]));
|
|
asm volatile ("vpxorq %zmm2,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm2,%zmm3");
|
|
asm volatile ("vpandq %zmm31,%zmm2,%zmm2");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm3");
|
|
asm volatile ("vpshufb %zmm2,%zmm24,%zmm2");
|
|
asm volatile ("vpshufb %zmm3,%zmm25,%zmm3");
|
|
asm volatile ("vpternlogq $0x96,%zmm2,%zmm3,%zmm1");
|
|
if (nd == 12)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm2" : : "m" (v[12][i]));
|
|
asm volatile ("vpxorq %zmm2,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm2,%zmm3");
|
|
asm volatile ("vpandq %zmm31,%zmm2,%zmm2");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm3");
|
|
asm volatile ("vpshufb %zmm2,%zmm26,%zmm2");
|
|
asm volatile ("vpshufb %zmm3,%zmm27,%zmm3");
|
|
asm volatile ("vpternlogq $0x96,%zmm2,%zmm3,%zmm1");
|
|
if (nd == 13)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm2" : : "m" (v[13][i]));
|
|
asm volatile ("vpxorq %zmm2,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm2,%zmm3");
|
|
asm volatile ("vpandq %zmm31,%zmm2,%zmm2");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm3");
|
|
asm volatile ("vpshufb %zmm2,%zmm28,%zmm2");
|
|
asm volatile ("vpshufb %zmm3,%zmm29,%zmm3");
|
|
asm volatile ("vpternlogq $0x96,%zmm2,%zmm3,%zmm1");
|
|
if (nd == 14)
|
|
goto store;
|
|
|
|
/*
|
|
* No more registers are available for resident coefficient
|
|
* tables. Process D14 and later with the normal loop.
|
|
*/
|
|
for (d = 14; d < nd; ++d) {
|
|
asm volatile ("vmovdqa64 %0,%%zmm2" : : "m" (v[d][i]));
|
|
|
|
asm volatile ("vpxorq %zmm2,%zmm0,%zmm0");
|
|
|
|
asm volatile ("vpsrlw $4,%zmm2,%zmm3");
|
|
asm volatile ("vpandq %zmm31,%zmm2,%zmm2");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm3");
|
|
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm30" : : "m" (raid_gfcauchypshufb[d][0][0][0]));
|
|
asm volatile ("vpshufb %zmm2,%zmm30,%zmm2");
|
|
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm30" : : "m" (raid_gfcauchypshufb[d][0][1][0]));
|
|
asm volatile ("vpshufb %zmm3,%zmm30,%zmm3");
|
|
|
|
asm volatile ("vpternlogq $0x96,%zmm2,%zmm3,%zmm1");
|
|
}
|
|
|
|
store:
|
|
if (streaming) {
|
|
asm volatile ("vmovntdq %%zmm0,%0" : "=m" (v[nd][i]));
|
|
asm volatile ("vmovntdq %%zmm1,%0" : "=m" (v[nd + 1][i]));
|
|
} else {
|
|
asm volatile ("vmovdqa64 %%zmm0,%0" : "=m" (v[nd][i]));
|
|
asm volatile ("vmovdqa64 %%zmm1,%0" : "=m" (v[nd + 1][i]));
|
|
}
|
|
}
|
|
|
|
raid_avx_end(streaming);
|
|
}
|
|
|
|
/*
|
|
* Generate three parity blocks with Cauchy matrix using AVX512BW implementation.
|
|
*
|
|
* Preloads 6 sets of Q and R coefficient tables in ZMM registers (zmm7..zmm30)
|
|
* with an unrolled disk loop and early exits.
|
|
*/
|
|
static __always_inline void raid_gen3_avx512bw_gen(int nd, size_t size, void **vv, int streaming)
|
|
{
|
|
uint8_t **v = (uint8_t **)vv;
|
|
size_t i;
|
|
int d;
|
|
|
|
if (nd == 1) {
|
|
if (v[1] != v[0])
|
|
memcpy(v[1], v[0], size);
|
|
if (v[2] != v[0])
|
|
memcpy(v[2], v[0], size);
|
|
if (v[3] != v[0])
|
|
memcpy(v[3], v[0], size);
|
|
return;
|
|
}
|
|
|
|
raid_avx_begin();
|
|
|
|
asm volatile ("vpbroadcastb %0,%%zmm31" : : "m" (gfconst16.low4[0]));
|
|
|
|
/* preload as many Q and R coefficient tables as possible */
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm7" : : "m" (raid_gfcauchypshufb[1][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm8" : : "m" (raid_gfcauchypshufb[1][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm9" : : "m" (raid_gfcauchypshufb[1][1][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm10" : : "m" (raid_gfcauchypshufb[1][1][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm11" : : "m" (raid_gfcauchypshufb[2][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm12" : : "m" (raid_gfcauchypshufb[2][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm13" : : "m" (raid_gfcauchypshufb[2][1][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm14" : : "m" (raid_gfcauchypshufb[2][1][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm15" : : "m" (raid_gfcauchypshufb[3][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm16" : : "m" (raid_gfcauchypshufb[3][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm17" : : "m" (raid_gfcauchypshufb[3][1][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm18" : : "m" (raid_gfcauchypshufb[3][1][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm19" : : "m" (raid_gfcauchypshufb[4][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm20" : : "m" (raid_gfcauchypshufb[4][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm21" : : "m" (raid_gfcauchypshufb[4][1][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm22" : : "m" (raid_gfcauchypshufb[4][1][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm23" : : "m" (raid_gfcauchypshufb[5][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm24" : : "m" (raid_gfcauchypshufb[5][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm25" : : "m" (raid_gfcauchypshufb[5][1][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm26" : : "m" (raid_gfcauchypshufb[5][1][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm27" : : "m" (raid_gfcauchypshufb[6][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm28" : : "m" (raid_gfcauchypshufb[6][0][1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm29" : : "m" (raid_gfcauchypshufb[6][1][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm30" : : "m" (raid_gfcauchypshufb[6][1][1][0]));
|
|
|
|
for (i = 0; i < size; i += 64) {
|
|
asm volatile ("vmovdqa64 %0,%%zmm0" : : "m" (v[0][i]));
|
|
asm volatile ("vmovdqa64 %zmm0,%zmm1");
|
|
asm volatile ("vmovdqa64 %zmm0,%zmm2");
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm3" : : "m" (v[1][i]));
|
|
asm volatile ("vpxorq %zmm3,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm3,%zmm5");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm4");
|
|
asm volatile ("vpandq %zmm31,%zmm5,%zmm5");
|
|
asm volatile ("vpshufb %zmm4,%zmm7,%zmm3");
|
|
asm volatile ("vpshufb %zmm5,%zmm8,%zmm6");
|
|
asm volatile ("vpternlogq $0x96,%zmm3,%zmm6,%zmm1");
|
|
asm volatile ("vpshufb %zmm4,%zmm9,%zmm3");
|
|
asm volatile ("vpshufb %zmm5,%zmm10,%zmm6");
|
|
asm volatile ("vpternlogq $0x96,%zmm3,%zmm6,%zmm2");
|
|
if (nd == 2)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm3" : : "m" (v[2][i]));
|
|
asm volatile ("vpxorq %zmm3,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm3,%zmm5");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm4");
|
|
asm volatile ("vpandq %zmm31,%zmm5,%zmm5");
|
|
asm volatile ("vpshufb %zmm4,%zmm11,%zmm3");
|
|
asm volatile ("vpshufb %zmm5,%zmm12,%zmm6");
|
|
asm volatile ("vpternlogq $0x96,%zmm3,%zmm6,%zmm1");
|
|
asm volatile ("vpshufb %zmm4,%zmm13,%zmm3");
|
|
asm volatile ("vpshufb %zmm5,%zmm14,%zmm6");
|
|
asm volatile ("vpternlogq $0x96,%zmm3,%zmm6,%zmm2");
|
|
if (nd == 3)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm3" : : "m" (v[3][i]));
|
|
asm volatile ("vpxorq %zmm3,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm3,%zmm5");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm4");
|
|
asm volatile ("vpandq %zmm31,%zmm5,%zmm5");
|
|
asm volatile ("vpshufb %zmm4,%zmm15,%zmm3");
|
|
asm volatile ("vpshufb %zmm5,%zmm16,%zmm6");
|
|
asm volatile ("vpternlogq $0x96,%zmm3,%zmm6,%zmm1");
|
|
asm volatile ("vpshufb %zmm4,%zmm17,%zmm3");
|
|
asm volatile ("vpshufb %zmm5,%zmm18,%zmm6");
|
|
asm volatile ("vpternlogq $0x96,%zmm3,%zmm6,%zmm2");
|
|
if (nd == 4)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm3" : : "m" (v[4][i]));
|
|
asm volatile ("vpxorq %zmm3,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm3,%zmm5");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm4");
|
|
asm volatile ("vpandq %zmm31,%zmm5,%zmm5");
|
|
asm volatile ("vpshufb %zmm4,%zmm19,%zmm3");
|
|
asm volatile ("vpshufb %zmm5,%zmm20,%zmm6");
|
|
asm volatile ("vpternlogq $0x96,%zmm3,%zmm6,%zmm1");
|
|
asm volatile ("vpshufb %zmm4,%zmm21,%zmm3");
|
|
asm volatile ("vpshufb %zmm5,%zmm22,%zmm6");
|
|
asm volatile ("vpternlogq $0x96,%zmm3,%zmm6,%zmm2");
|
|
if (nd == 5)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm3" : : "m" (v[5][i]));
|
|
asm volatile ("vpxorq %zmm3,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm3,%zmm5");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm4");
|
|
asm volatile ("vpandq %zmm31,%zmm5,%zmm5");
|
|
asm volatile ("vpshufb %zmm4,%zmm23,%zmm3");
|
|
asm volatile ("vpshufb %zmm5,%zmm24,%zmm6");
|
|
asm volatile ("vpternlogq $0x96,%zmm3,%zmm6,%zmm1");
|
|
asm volatile ("vpshufb %zmm4,%zmm25,%zmm3");
|
|
asm volatile ("vpshufb %zmm5,%zmm26,%zmm6");
|
|
asm volatile ("vpternlogq $0x96,%zmm3,%zmm6,%zmm2");
|
|
if (nd == 6)
|
|
goto store;
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm3" : : "m" (v[6][i]));
|
|
asm volatile ("vpxorq %zmm3,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm3,%zmm5");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm4");
|
|
asm volatile ("vpandq %zmm31,%zmm5,%zmm5");
|
|
asm volatile ("vpshufb %zmm4,%zmm27,%zmm3");
|
|
asm volatile ("vpshufb %zmm5,%zmm28,%zmm6");
|
|
asm volatile ("vpternlogq $0x96,%zmm3,%zmm6,%zmm1");
|
|
asm volatile ("vpshufb %zmm4,%zmm29,%zmm3");
|
|
asm volatile ("vpshufb %zmm5,%zmm30,%zmm6");
|
|
asm volatile ("vpternlogq $0x96,%zmm3,%zmm6,%zmm2");
|
|
if (nd == 7)
|
|
goto store;
|
|
|
|
/*
|
|
* No more registers are available for resident coefficient
|
|
* tables. Process D7 and later with the normal loop.
|
|
*/
|
|
for (d = 7; d < nd; ++d) {
|
|
asm volatile ("vmovdqa64 %0,%%zmm3" : : "m" (v[d][i]));
|
|
asm volatile ("vpxorq %zmm3,%zmm0,%zmm0");
|
|
asm volatile ("vpsrlw $4,%zmm3,%zmm5");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm4");
|
|
asm volatile ("vpandq %zmm31,%zmm5,%zmm5");
|
|
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm6" : : "m" (raid_gfcauchypshufb[d][0][0][0]));
|
|
asm volatile ("vpshufb %zmm4,%zmm6,%zmm6");
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm3" : : "m" (raid_gfcauchypshufb[d][0][1][0]));
|
|
asm volatile ("vpshufb %zmm5,%zmm3,%zmm3");
|
|
asm volatile ("vpternlogq $0x96,%zmm3,%zmm6,%zmm1");
|
|
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm6" : : "m" (raid_gfcauchypshufb[d][1][0][0]));
|
|
asm volatile ("vpshufb %zmm4,%zmm6,%zmm6");
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm3" : : "m" (raid_gfcauchypshufb[d][1][1][0]));
|
|
asm volatile ("vpshufb %zmm5,%zmm3,%zmm3");
|
|
asm volatile ("vpternlogq $0x96,%zmm3,%zmm6,%zmm2");
|
|
}
|
|
|
|
store:
|
|
if (streaming) {
|
|
asm volatile ("vmovntdq %%zmm0,%0" : "=m" (v[nd][i]));
|
|
asm volatile ("vmovntdq %%zmm1,%0" : "=m" (v[nd + 1][i]));
|
|
asm volatile ("vmovntdq %%zmm2,%0" : "=m" (v[nd + 2][i]));
|
|
} else {
|
|
asm volatile ("vmovdqa64 %%zmm0,%0" : "=m" (v[nd][i]));
|
|
asm volatile ("vmovdqa64 %%zmm1,%0" : "=m" (v[nd + 1][i]));
|
|
asm volatile ("vmovdqa64 %%zmm2,%0" : "=m" (v[nd + 2][i]));
|
|
}
|
|
}
|
|
|
|
raid_avx_end(streaming);
|
|
}
|
|
|
|
/*
|
|
* Generate N parity blocks with Cauchy matrix using AVX512BW implementation.
|
|
*/
|
|
static __always_inline void raid_genX_avx512bw(int nd, size_t size, void **vv, int np, int streaming)
|
|
{
|
|
uint8_t **v = (uint8_t **)vv;
|
|
size_t i;
|
|
int d;
|
|
|
|
if (nd == 1) {
|
|
for (d = 0; d < np; ++d)
|
|
if (v[1 + d] != v[0])
|
|
memcpy(v[1 + d], v[0], size);
|
|
return;
|
|
}
|
|
|
|
raid_avx_begin();
|
|
|
|
asm volatile ("vpbroadcastb %0,%%zmm31" : : "m" (gfconst16.low4[0]));
|
|
|
|
for (i = 0; i < size; i += 64) {
|
|
asm volatile ("vmovdqa64 %0,%%zmm0" : : "m" (v[0][i]));
|
|
asm volatile ("vmovdqa64 %zmm0,%zmm1");
|
|
if (np >= 3)
|
|
asm volatile ("vmovdqa64 %zmm0,%zmm2");
|
|
if (np >= 4)
|
|
asm volatile ("vmovdqa64 %zmm0,%zmm3");
|
|
if (np >= 5)
|
|
asm volatile ("vmovdqa64 %zmm0,%zmm4");
|
|
if (np >= 6)
|
|
asm volatile ("vmovdqa64 %zmm0,%zmm5");
|
|
|
|
for (d = 1; d < nd; ++d) {
|
|
asm volatile ("vmovdqa64 %0,%%zmm10" : : "m" (v[d][i]));
|
|
|
|
asm volatile ("vpxorq %zmm10,%zmm0,%zmm0");
|
|
|
|
asm volatile ("vpsrlw $4,%zmm10,%zmm11");
|
|
asm volatile ("vpandq %zmm31,%zmm10,%zmm10");
|
|
asm volatile ("vpandq %zmm31,%zmm11,%zmm11");
|
|
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm12" : : "m" (raid_gfcauchypshufb[d][0][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm13" : : "m" (raid_gfcauchypshufb[d][0][1][0]));
|
|
if (np >= 3) {
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm14" : : "m" (raid_gfcauchypshufb[d][1][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm15" : : "m" (raid_gfcauchypshufb[d][1][1][0]));
|
|
}
|
|
if (np >= 4) {
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm16" : : "m" (raid_gfcauchypshufb[d][2][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm17" : : "m" (raid_gfcauchypshufb[d][2][1][0]));
|
|
}
|
|
if (np >= 5) {
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm18" : : "m" (raid_gfcauchypshufb[d][3][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm19" : : "m" (raid_gfcauchypshufb[d][3][1][0]));
|
|
}
|
|
if (np >= 6) {
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm20" : : "m" (raid_gfcauchypshufb[d][4][0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm21" : : "m" (raid_gfcauchypshufb[d][4][1][0]));
|
|
}
|
|
|
|
asm volatile ("vpshufb %zmm10,%zmm12,%zmm12");
|
|
asm volatile ("vpshufb %zmm11,%zmm13,%zmm13");
|
|
if (np >= 3) {
|
|
asm volatile ("vpshufb %zmm10,%zmm14,%zmm14");
|
|
asm volatile ("vpshufb %zmm11,%zmm15,%zmm15");
|
|
}
|
|
if (np >= 4) {
|
|
asm volatile ("vpshufb %zmm10,%zmm16,%zmm16");
|
|
asm volatile ("vpshufb %zmm11,%zmm17,%zmm17");
|
|
}
|
|
if (np >= 5) {
|
|
asm volatile ("vpshufb %zmm10,%zmm18,%zmm18");
|
|
asm volatile ("vpshufb %zmm11,%zmm19,%zmm19");
|
|
}
|
|
if (np >= 6) {
|
|
asm volatile ("vpshufb %zmm10,%zmm20,%zmm20");
|
|
asm volatile ("vpshufb %zmm11,%zmm21,%zmm21");
|
|
}
|
|
|
|
asm volatile ("vpternlogq $0x96,%zmm12,%zmm13,%zmm1");
|
|
if (np >= 3)
|
|
asm volatile ("vpternlogq $0x96,%zmm14,%zmm15,%zmm2");
|
|
if (np >= 4)
|
|
asm volatile ("vpternlogq $0x96,%zmm16,%zmm17,%zmm3");
|
|
if (np >= 5)
|
|
asm volatile ("vpternlogq $0x96,%zmm18,%zmm19,%zmm4");
|
|
if (np >= 6)
|
|
asm volatile ("vpternlogq $0x96,%zmm20,%zmm21,%zmm5");
|
|
}
|
|
|
|
if (streaming) {
|
|
asm volatile ("vmovntdq %%zmm0,%0" : "=m" (v[nd][i]));
|
|
asm volatile ("vmovntdq %%zmm1,%0" : "=m" (v[nd + 1][i]));
|
|
if (np >= 3)
|
|
asm volatile ("vmovntdq %%zmm2,%0" : "=m" (v[nd + 2][i]));
|
|
if (np >= 4)
|
|
asm volatile ("vmovntdq %%zmm3,%0" : "=m" (v[nd + 3][i]));
|
|
if (np >= 5)
|
|
asm volatile ("vmovntdq %%zmm4,%0" : "=m" (v[nd + 4][i]));
|
|
if (np >= 6)
|
|
asm volatile ("vmovntdq %%zmm5,%0" : "=m" (v[nd + 5][i]));
|
|
} else {
|
|
asm volatile ("vmovdqa64 %%zmm0,%0" : "=m" (v[nd][i]));
|
|
asm volatile ("vmovdqa64 %%zmm1,%0" : "=m" (v[nd + 1][i]));
|
|
if (np >= 3)
|
|
asm volatile ("vmovdqa64 %%zmm2,%0" : "=m" (v[nd + 2][i]));
|
|
if (np >= 4)
|
|
asm volatile ("vmovdqa64 %%zmm3,%0" : "=m" (v[nd + 3][i]));
|
|
if (np >= 5)
|
|
asm volatile ("vmovdqa64 %%zmm4,%0" : "=m" (v[nd + 4][i]));
|
|
if (np >= 6)
|
|
asm volatile ("vmovdqa64 %%zmm5,%0" : "=m" (v[nd + 5][i]));
|
|
}
|
|
}
|
|
|
|
raid_avx_end(streaming);
|
|
}
|
|
|
|
/*
|
|
* Recover multiple data failures using selected parity blocks with AVX512BW.
|
|
*
|
|
* Compute only the selected syndromes, keeping them in registers.
|
|
* This avoids temporary syndrome buffers, recomputation of
|
|
* parity blocks, and generation of unused parity rows.
|
|
*
|
|
* AVX512 provides enough registers to keep all six supported syndromes
|
|
* in registers in a single surviving-data scan.
|
|
*/
|
|
static __always_inline void raid_recX_avx512bw(int nr, int has_p, int *id, int *ip, int nd, size_t size, void **vv)
|
|
{
|
|
uint8_t **v = (uint8_t **)vv;
|
|
uint8_t *p[RAID_PARITY_MAX];
|
|
uint8_t *pa[RAID_PARITY_MAX];
|
|
uint8_t *src[RAID_DATA_MAX];
|
|
uint8_t G[RAID_PARITY_MAX * RAID_PARITY_MAX];
|
|
uint8_t V[RAID_PARITY_MAX * RAID_PARITY_MAX];
|
|
const uint8_t *S[RAID_DATA_MAX][RAID_PARITY_MAX];
|
|
const uint8_t *R[RAID_PARITY_MAX][RAID_PARITY_MAX];
|
|
size_t i;
|
|
int d, j, k, s;
|
|
int ns;
|
|
|
|
BUG_ON(nr < 1 || nr > RAID_PARITY_MAX);
|
|
|
|
/* setup the coefficients matrix */
|
|
for (j = 0; j < nr; ++j)
|
|
for (k = 0; k < nr; ++k)
|
|
G[j * nr + k] = A(ip[j], id[k]);
|
|
|
|
/* invert it to solve the system of linear equations */
|
|
raid_invert(G, V, nr);
|
|
|
|
/* setup selected parity and destination pointers */
|
|
for (j = 0; j < nr; ++j) {
|
|
p[j] = v[nd + ip[j]];
|
|
pa[j] = v[id[j]];
|
|
}
|
|
|
|
/*
|
|
* Build the compact list of surviving data blocks and precompute
|
|
* the multiplication-table pointers for each selected syndrome.
|
|
*/
|
|
ns = 0;
|
|
k = 0;
|
|
|
|
for (d = 0; d < nd; ++d) {
|
|
if (k < nr && d == id[k]) {
|
|
++k;
|
|
continue;
|
|
}
|
|
|
|
src[ns] = v[d];
|
|
|
|
for (j = 0; j < nr; ++j)
|
|
S[ns][j] = &raid_gfmulpshufb[A(ip[j], d)][0][0];
|
|
|
|
++ns;
|
|
}
|
|
|
|
BUG_ON(k != nr);
|
|
BUG_ON(ns != nd - nr);
|
|
|
|
/*
|
|
* If P is available, the last missing block is reconstructed
|
|
* directly from Pdelta, so the last inverse row isn't needed.
|
|
*/
|
|
for (j = 0; j < nr - has_p; ++j)
|
|
for (k = 0; k < nr; ++k)
|
|
R[j][k] = &raid_gfmulpshufb[V[j * nr + k]][0][0];
|
|
|
|
raid_avx_begin();
|
|
|
|
/*
|
|
* Register allocation during syndrome generation:
|
|
*
|
|
* zmm0 syndrome 0
|
|
* zmm2 syndrome 1
|
|
* zmm4 syndrome 2
|
|
* zmm6 syndrome 3
|
|
* zmm8 syndrome 4
|
|
* zmm10 syndrome 5
|
|
*
|
|
* zmm12 source / source low nibble
|
|
* zmm13 source high nibble
|
|
* zmm14 low multiplication table / result
|
|
* zmm15 high multiplication table / result
|
|
* zmm30 remaining P delta, if has_p
|
|
* zmm31 low-nibble mask
|
|
*
|
|
* After syndrome splitting:
|
|
*
|
|
* zmm0 /zmm1 syndrome 0 low/high
|
|
* zmm2 /zmm3 syndrome 1 low/high
|
|
* zmm4 /zmm5 syndrome 2 low/high
|
|
* zmm6 /zmm7 syndrome 3 low/high
|
|
* zmm8 /zmm9 syndrome 4 low/high
|
|
* zmm10/zmm11 syndrome 5 low/high
|
|
*/
|
|
|
|
asm volatile ("vpbroadcastb %0,%%zmm31" : : "m" (gfconst16.low4[0]));
|
|
|
|
for (i = 0; i < size; i += 64) {
|
|
/* start every selected syndrome from its stored parity block */
|
|
asm volatile ("vmovdqa64 %0,%%zmm0" : : "m" (p[0][i]));
|
|
|
|
if (nr >= 2)
|
|
asm volatile ("vmovdqa64 %0,%%zmm2" : : "m" (p[1][i]));
|
|
|
|
if (nr >= 3)
|
|
asm volatile ("vmovdqa64 %0,%%zmm4" : : "m" (p[2][i]));
|
|
|
|
if (nr >= 4)
|
|
asm volatile ("vmovdqa64 %0,%%zmm6" : : "m" (p[3][i]));
|
|
|
|
if (nr >= 5)
|
|
asm volatile ("vmovdqa64 %0,%%zmm8" : : "m" (p[4][i]));
|
|
|
|
if (nr >= 6)
|
|
asm volatile ("vmovdqa64 %0,%%zmm10" : : "m" (p[5][i]));
|
|
|
|
/* add all surviving data contributions in one source scan */
|
|
for (s = 0; s < ns; ++s) {
|
|
const uint8_t **t = S[s];
|
|
|
|
asm volatile ("vmovdqa64 %0,%%zmm12" : : "m" (src[s][i]));
|
|
|
|
/*
|
|
* P has coefficient 1 for every data disk.
|
|
* XOR the original source before splitting it.
|
|
*/
|
|
if (has_p) {
|
|
asm volatile ("vpxorq %zmm12,%zmm0,%zmm0");
|
|
|
|
asm volatile ("vpsrlw $4,%zmm12,%zmm13");
|
|
asm volatile ("vpandq %zmm31,%zmm12,%zmm12");
|
|
asm volatile ("vpandq %zmm31,%zmm13,%zmm13");
|
|
} else {
|
|
asm volatile ("vpsrlw $4,%zmm12,%zmm13");
|
|
asm volatile ("vpandq %zmm31,%zmm12,%zmm12");
|
|
asm volatile ("vpandq %zmm31,%zmm13,%zmm13");
|
|
|
|
/* syndrome 0 */
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm14" : : "m" (t[0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm15" : : "m" (t[0][16]));
|
|
asm volatile ("vpshufb %zmm12,%zmm14,%zmm14");
|
|
asm volatile ("vpshufb %zmm13,%zmm15,%zmm15");
|
|
asm volatile ("vpternlogq $0x96,%zmm14,%zmm15,%zmm0");
|
|
}
|
|
|
|
/* syndrome 1 */
|
|
if (nr >= 2) {
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm14" : : "m" (t[1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm15" : : "m" (t[1][16]));
|
|
asm volatile ("vpshufb %zmm12,%zmm14,%zmm14");
|
|
asm volatile ("vpshufb %zmm13,%zmm15,%zmm15");
|
|
asm volatile ("vpternlogq $0x96,%zmm14,%zmm15,%zmm2");
|
|
}
|
|
|
|
/* syndrome 2 */
|
|
if (nr >= 3) {
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm14" : : "m" (t[2][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm15" : : "m" (t[2][16]));
|
|
asm volatile ("vpshufb %zmm12,%zmm14,%zmm14");
|
|
asm volatile ("vpshufb %zmm13,%zmm15,%zmm15");
|
|
asm volatile ("vpternlogq $0x96,%zmm14,%zmm15,%zmm4");
|
|
}
|
|
|
|
/* syndrome 3 */
|
|
if (nr >= 4) {
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm14" : : "m" (t[3][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm15" : : "m" (t[3][16]));
|
|
asm volatile ("vpshufb %zmm12,%zmm14,%zmm14");
|
|
asm volatile ("vpshufb %zmm13,%zmm15,%zmm15");
|
|
asm volatile ("vpternlogq $0x96,%zmm14,%zmm15,%zmm6");
|
|
}
|
|
|
|
/* syndrome 4 */
|
|
if (nr >= 5) {
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm14" : : "m" (t[4][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm15" : : "m" (t[4][16]));
|
|
asm volatile ("vpshufb %zmm12,%zmm14,%zmm14");
|
|
asm volatile ("vpshufb %zmm13,%zmm15,%zmm15");
|
|
asm volatile ("vpternlogq $0x96,%zmm14,%zmm15,%zmm8");
|
|
}
|
|
|
|
/* syndrome 5 */
|
|
if (nr >= 6) {
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm14" : : "m" (t[5][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm15" : : "m" (t[5][16]));
|
|
asm volatile ("vpshufb %zmm12,%zmm14,%zmm14");
|
|
asm volatile ("vpshufb %zmm13,%zmm15,%zmm15");
|
|
asm volatile ("vpternlogq $0x96,%zmm14,%zmm15,%zmm10");
|
|
}
|
|
}
|
|
|
|
/*
|
|
* Preserve the complete P delta before syndrome 0 is
|
|
* destructively split into low/high nibbles.
|
|
*/
|
|
if (has_p)
|
|
asm volatile ("vmovdqa64 %zmm0,%zmm30");
|
|
|
|
/* split every completed syndrome once into low/high nibbles */
|
|
asm volatile ("vpsrlw $4,%zmm0,%zmm1");
|
|
asm volatile ("vpandq %zmm31,%zmm0,%zmm0");
|
|
asm volatile ("vpandq %zmm31,%zmm1,%zmm1");
|
|
|
|
if (nr >= 2) {
|
|
asm volatile ("vpsrlw $4,%zmm2,%zmm3");
|
|
asm volatile ("vpandq %zmm31,%zmm2,%zmm2");
|
|
asm volatile ("vpandq %zmm31,%zmm3,%zmm3");
|
|
}
|
|
|
|
if (nr >= 3) {
|
|
asm volatile ("vpsrlw $4,%zmm4,%zmm5");
|
|
asm volatile ("vpandq %zmm31,%zmm4,%zmm4");
|
|
asm volatile ("vpandq %zmm31,%zmm5,%zmm5");
|
|
}
|
|
|
|
if (nr >= 4) {
|
|
asm volatile ("vpsrlw $4,%zmm6,%zmm7");
|
|
asm volatile ("vpandq %zmm31,%zmm6,%zmm6");
|
|
asm volatile ("vpandq %zmm31,%zmm7,%zmm7");
|
|
}
|
|
|
|
if (nr >= 5) {
|
|
asm volatile ("vpsrlw $4,%zmm8,%zmm9");
|
|
asm volatile ("vpandq %zmm31,%zmm8,%zmm8");
|
|
asm volatile ("vpandq %zmm31,%zmm9,%zmm9");
|
|
}
|
|
|
|
if (nr >= 6) {
|
|
asm volatile ("vpsrlw $4,%zmm10,%zmm11");
|
|
asm volatile ("vpandq %zmm31,%zmm10,%zmm10");
|
|
asm volatile ("vpandq %zmm31,%zmm11,%zmm11");
|
|
}
|
|
|
|
/*
|
|
* Reconstruct the missing data blocks.
|
|
*
|
|
* If P is available, reconstruct only nr - 1 blocks through
|
|
* the inverse matrix. XOR each result out of zmm30, leaving
|
|
* the final missing block in zmm30.
|
|
*
|
|
* Keep independent low/high dependency chains:
|
|
*
|
|
* zmm12 low accumulator
|
|
* zmm13 high accumulator
|
|
* zmm14 low multiplication table/result
|
|
* zmm15 high multiplication table/result
|
|
* zmm30 remaining P delta, if has_p
|
|
*
|
|
* This is intentional. Do not serialize low/high products
|
|
* through a single temporary register.
|
|
*/
|
|
for (j = 0; j < nr - has_p; ++j) {
|
|
const uint8_t **t = R[j];
|
|
|
|
/* coefficient 0 initializes both accumulators */
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm12" : : "m" (t[0][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm13" : : "m" (t[0][16]));
|
|
asm volatile ("vpshufb %zmm0,%zmm12,%zmm12");
|
|
asm volatile ("vpshufb %zmm1,%zmm13,%zmm13");
|
|
|
|
if (nr >= 2) {
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm14" : : "m" (t[1][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm15" : : "m" (t[1][16]));
|
|
asm volatile ("vpshufb %zmm2,%zmm14,%zmm14");
|
|
asm volatile ("vpshufb %zmm3,%zmm15,%zmm15");
|
|
asm volatile ("vpxorq %zmm14,%zmm12,%zmm12");
|
|
asm volatile ("vpxorq %zmm15,%zmm13,%zmm13");
|
|
}
|
|
|
|
if (nr >= 3) {
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm14" : : "m" (t[2][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm15" : : "m" (t[2][16]));
|
|
asm volatile ("vpshufb %zmm4,%zmm14,%zmm14");
|
|
asm volatile ("vpshufb %zmm5,%zmm15,%zmm15");
|
|
asm volatile ("vpxorq %zmm14,%zmm12,%zmm12");
|
|
asm volatile ("vpxorq %zmm15,%zmm13,%zmm13");
|
|
}
|
|
|
|
if (nr >= 4) {
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm14" : : "m" (t[3][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm15" : : "m" (t[3][16]));
|
|
asm volatile ("vpshufb %zmm6,%zmm14,%zmm14");
|
|
asm volatile ("vpshufb %zmm7,%zmm15,%zmm15");
|
|
asm volatile ("vpxorq %zmm14,%zmm12,%zmm12");
|
|
asm volatile ("vpxorq %zmm15,%zmm13,%zmm13");
|
|
}
|
|
|
|
if (nr >= 5) {
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm14" : : "m" (t[4][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm15" : : "m" (t[4][16]));
|
|
asm volatile ("vpshufb %zmm8,%zmm14,%zmm14");
|
|
asm volatile ("vpshufb %zmm9,%zmm15,%zmm15");
|
|
asm volatile ("vpxorq %zmm14,%zmm12,%zmm12");
|
|
asm volatile ("vpxorq %zmm15,%zmm13,%zmm13");
|
|
}
|
|
|
|
if (nr >= 6) {
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm14" : : "m" (t[5][0]));
|
|
asm volatile ("vbroadcasti32x4 %0,%%zmm15" : : "m" (t[5][16]));
|
|
asm volatile ("vpshufb %zmm10,%zmm14,%zmm14");
|
|
asm volatile ("vpshufb %zmm11,%zmm15,%zmm15");
|
|
asm volatile ("vpxorq %zmm14,%zmm12,%zmm12");
|
|
asm volatile ("vpxorq %zmm15,%zmm13,%zmm13");
|
|
}
|
|
|
|
/* combine low/high products into the reconstructed block */
|
|
asm volatile ("vpxorq %zmm13,%zmm12,%zmm12");
|
|
|
|
/*
|
|
* Remove the reconstructed block from Pdelta.
|
|
* After nr - 1 iterations zmm30 contains the final
|
|
* missing data block.
|
|
*/
|
|
if (has_p)
|
|
asm volatile ("vpxorq %zmm12,%zmm30,%zmm30");
|
|
|
|
asm volatile ("vmovdqa64 %%zmm12,%0" : "=m" (pa[j][i]));
|
|
}
|
|
|
|
if (has_p)
|
|
asm volatile ("vmovdqa64 %%zmm30,%0" : "=m" (pa[nr - 1][i]));
|
|
}
|
|
|
|
raid_avx_end(0);
|
|
}
|
|
|
|
void raid_gen1_avx512bw(int nd, size_t size, void **vv, int streaming)
|
|
{
|
|
if (streaming)
|
|
raid_gen1_avx512bw_gen(nd, size, vv, 1);
|
|
else
|
|
raid_gen1_avx512bw_gen(nd, size, vv, 0);
|
|
}
|
|
|
|
void raid_gen2_avx512bw(int nd, size_t size, void **vv, int streaming)
|
|
{
|
|
if (streaming)
|
|
raid_gen2_avx512bw_gen(nd, size, vv, 1);
|
|
else
|
|
raid_gen2_avx512bw_gen(nd, size, vv, 0);
|
|
}
|
|
|
|
void raid_gen3_avx512bw(int nd, size_t size, void **vv, int streaming)
|
|
{
|
|
if (streaming)
|
|
raid_gen3_avx512bw_gen(nd, size, vv, 1);
|
|
else
|
|
raid_gen3_avx512bw_gen(nd, size, vv, 0);
|
|
}
|
|
|
|
void raid_gen4_avx512bw(int nd, size_t size, void **vv, int streaming)
|
|
{
|
|
if (streaming)
|
|
raid_genX_avx512bw(nd, size, vv, 4, 1);
|
|
else
|
|
raid_genX_avx512bw(nd, size, vv, 4, 0);
|
|
}
|
|
|
|
void raid_gen5_avx512bw(int nd, size_t size, void **vv, int streaming)
|
|
{
|
|
if (streaming)
|
|
raid_genX_avx512bw(nd, size, vv, 5, 1);
|
|
else
|
|
raid_genX_avx512bw(nd, size, vv, 5, 0);
|
|
}
|
|
|
|
void raid_gen6_avx512bw(int nd, size_t size, void **vv, int streaming)
|
|
{
|
|
if (streaming)
|
|
raid_genX_avx512bw(nd, size, vv, 6, 1);
|
|
else
|
|
raid_genX_avx512bw(nd, size, vv, 6, 0);
|
|
}
|
|
|
|
void raid_rec1_avx512bw(int nr, int *id, int *ip, int nd, size_t size, void **vv)
|
|
{
|
|
BUG_ON(nr != 1);
|
|
|
|
/* if recovering with P, use a custom XOR-only path with temporal stores */
|
|
if (ip[0] == 0) {
|
|
raid_rec1of1(id, nd, size, vv);
|
|
return;
|
|
}
|
|
|
|
raid_recX_avx512bw(1, 0, id, ip, nd, size, vv);
|
|
}
|
|
|
|
void raid_rec2_avx512bw(int nr, int *id, int *ip, int nd, size_t size, void **vv)
|
|
{
|
|
BUG_ON(nr != 2);
|
|
if (ip[0] == 0)
|
|
raid_recX_avx512bw(2, 1, id, ip, nd, size, vv);
|
|
else
|
|
raid_recX_avx512bw(2, 0, id, ip, nd, size, vv);
|
|
}
|
|
|
|
void raid_rec3_avx512bw(int nr, int *id, int *ip, int nd, size_t size, void **vv)
|
|
{
|
|
BUG_ON(nr != 3);
|
|
if (ip[0] == 0)
|
|
raid_recX_avx512bw(3, 1, id, ip, nd, size, vv);
|
|
else
|
|
raid_recX_avx512bw(3, 0, id, ip, nd, size, vv);
|
|
}
|
|
|
|
void raid_rec4_avx512bw(int nr, int *id, int *ip, int nd, size_t size, void **vv)
|
|
{
|
|
BUG_ON(nr != 4);
|
|
if (ip[0] == 0)
|
|
raid_recX_avx512bw(4, 1, id, ip, nd, size, vv);
|
|
else
|
|
raid_recX_avx512bw(4, 0, id, ip, nd, size, vv);
|
|
}
|
|
|
|
void raid_rec5_avx512bw(int nr, int *id, int *ip, int nd, size_t size, void **vv)
|
|
{
|
|
BUG_ON(nr != 5);
|
|
if (ip[0] == 0)
|
|
raid_recX_avx512bw(5, 1, id, ip, nd, size, vv);
|
|
else
|
|
raid_recX_avx512bw(5, 0, id, ip, nd, size, vv);
|
|
}
|
|
|
|
void raid_rec6_avx512bw(int nr, int *id, int *ip, int nd, size_t size, void **vv)
|
|
{
|
|
BUG_ON(nr != 6);
|
|
if (ip[0] == 0)
|
|
raid_recX_avx512bw(6, 1, id, ip, nd, size, vv);
|
|
else
|
|
raid_recX_avx512bw(6, 0, id, ip, nd, size, vv);
|
|
}
|
|
|
|
void raid_register_avx512(void)
|
|
{
|
|
if (raid_cpu_has_avx512bw()) {
|
|
if (!raid_cpu_has_slow_avx512()) {
|
|
raid_gen_register(RAID_ALGO_CAUCHY_PAR1, "avx512", raid_gen1_avx512bw, RAID_POLY_ANY);
|
|
raid_gen_register(RAID_ALGO_CAUCHY_PAR2, "avx512", raid_gen2_avx512bw, RAID_POLY_ANY);
|
|
raid_gen_register(RAID_ALGO_CAUCHY_PAR3, "avx512", raid_gen3_avx512bw, RAID_POLY_ANY);
|
|
raid_gen_register(RAID_ALGO_CAUCHY_PAR4, "avx512", raid_gen4_avx512bw, RAID_POLY_ANY);
|
|
raid_gen_register(RAID_ALGO_CAUCHY_PAR5, "avx512", raid_gen5_avx512bw, RAID_POLY_ANY);
|
|
raid_gen_register(RAID_ALGO_CAUCHY_PAR6, "avx512", raid_gen6_avx512bw, RAID_POLY_ANY);
|
|
|
|
raid_rec_register(RAID_ALGO_CAUCHY_PAR1, "avx512", raid_rec1_avx512bw, RAID_POLY_ANY);
|
|
raid_rec_register(RAID_ALGO_CAUCHY_PAR2, "avx512", raid_rec2_avx512bw, RAID_POLY_ANY);
|
|
raid_rec_register(RAID_ALGO_CAUCHY_PAR3, "avx512", raid_rec3_avx512bw, RAID_POLY_ANY);
|
|
raid_rec_register(RAID_ALGO_CAUCHY_PAR4, "avx512", raid_rec4_avx512bw, RAID_POLY_ANY);
|
|
raid_rec_register(RAID_ALGO_CAUCHY_PAR5, "avx512", raid_rec5_avx512bw, RAID_POLY_ANY);
|
|
raid_rec_register(RAID_ALGO_CAUCHY_PAR6, "avx512", raid_rec6_avx512bw, RAID_POLY_ANY);
|
|
}
|
|
}
|
|
}
|
|
|
|
#endif
|