Quelle  rvv.c   Sprache: C

// SPDX-License-Identifier: GPL-2.0-or-later
/*
 * RAID-6 syndrome calculation using RISC-V vector instructions
 *
 * Copyright 2024 Institute of Software, CAS.
 * Author: Chunyan Zhang <zhangchunyan@iscas.ac.cn>
 *
 * Based on neon.uc:
 * Copyright 2002-2004 H. Peter Anvin
 */

#include <asm/vector.h>
#include <linux/raid/pq.h>
#include "rvv.h"

#define NSIZE (riscv_v_vsize / 32) /* NSIZE = vlenb */

static int rvv_has_vector(void)
{
 return has_vector();
}

static void raid6_rvv1_gen_syndrome_real(int disks, unsigned long bytes, void **ptrs)
{
 u8 **dptr = (u8 **)ptrs;
 u8 *p, *q;
 unsigned long vl, d;
 int z, z0;

 z0 = disks - 3;  /* Highest data disk */
 p = dptr[z0 + 1];  /* XOR parity */
 q = dptr[z0 + 2];  /* RS syndrome */

 asm volatile (".option push\n"
        ".option arch,+v\n"
        "vsetvli %0, x0, e8, m1, ta, ma\n"
        ".option pop\n"
        : "=&r" (vl)
 );

  /* v0:wp0, v1:wq0, v2:wd0/w20, v3:w10 */
 for (d = 0; d < bytes; d += NSIZE * 1) {
  /* wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE]; */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vle8.v v0, (%[wp0])\n"
         "vle8.v v1, (%[wp0])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&dptr[z0][d + 0 * NSIZE])
  );

  for (z = z0 - 1 ; z >= 0 ; z--) {
   /*
 * w2$$ = MASK(wq$$);
 * w1$$ = SHLBYTE(wq$$);
 * w2$$ &= NBYTES(0x1d);
 * w1$$ ^= w2$$;
 * wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE];
 * wq$$ = w1$$ ^ wd$$;
 * wp$$ ^= wd$$;
 */
   asm volatile (".option push\n"
          ".option arch,+v\n"
          "vsra.vi v2, v1, 7\n"
          "vsll.vi v3, v1, 1\n"
          "vand.vx v2, v2, %[x1d]\n"
          "vxor.vv v3, v3, v2\n"
          "vle8.v v2, (%[wd0])\n"
          "vxor.vv v1, v3, v2\n"
          "vxor.vv v0, v0, v2\n"
          ".option pop\n"
          : :
          [wd0]"r"(&dptr[z][d + 0 * NSIZE]),
          [x1d]"r"(0x1d)
   );
  }

  /*
 * *(unative_t *)&p[d+NSIZE*$$] = wp$$;
 * *(unative_t *)&q[d+NSIZE*$$] = wq$$;
 */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vse8.v v0, (%[wp0])\n"
         "vse8.v v1, (%[wq0])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&p[d + NSIZE * 0]),
         [wq0]"r"(&q[d + NSIZE * 0])
  );
 }
}

static void raid6_rvv1_xor_syndrome_real(int disks, int start, int stop,
      unsigned long bytes, void **ptrs)
{
 u8 **dptr = (u8 **)ptrs;
 u8 *p, *q;
 unsigned long vl, d;
 int z, z0;

 z0 = stop;  /* P/Q right side optimization */
 p = dptr[disks - 2]; /* XOR parity */
 q = dptr[disks - 1]; /* RS syndrome */

 asm volatile (".option push\n"
        ".option arch,+v\n"
        "vsetvli %0, x0, e8, m1, ta, ma\n"
        ".option pop\n"
        : "=&r" (vl)
 );

 /* v0:wp0, v1:wq0, v2:wd0/w20, v3:w10 */
 for (d = 0 ; d < bytes ; d += NSIZE * 1) {
  /* wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE]; */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vle8.v v0, (%[wp0])\n"
         "vle8.v v1, (%[wp0])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&dptr[z0][d + 0 * NSIZE])
  );

  /* P/Q data pages */
  for (z = z0 - 1; z >= start; z--) {
   /*
 * w2$$ = MASK(wq$$);
 * w1$$ = SHLBYTE(wq$$);
 * w2$$ &= NBYTES(0x1d);
 * w1$$ ^= w2$$;
 * wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE];
 * wq$$ = w1$$ ^ wd$$;
 * wp$$ ^= wd$$;
 */
   asm volatile (".option push\n"
          ".option arch,+v\n"
          "vsra.vi v2, v1, 7\n"
          "vsll.vi v3, v1, 1\n"
          "vand.vx v2, v2, %[x1d]\n"
          "vxor.vv v3, v3, v2\n"
          "vle8.v v2, (%[wd0])\n"
          "vxor.vv v1, v3, v2\n"
          "vxor.vv v0, v0, v2\n"
          ".option pop\n"
          : :
          [wd0]"r"(&dptr[z][d + 0 * NSIZE]),
          [x1d]"r"(0x1d)
   );
  }

  /* P/Q left side optimization */
  for (z = start - 1; z >= 0; z--) {
   /*
 * w2$$ = MASK(wq$$);
 * w1$$ = SHLBYTE(wq$$);
 * w2$$ &= NBYTES(0x1d);
 * wq$$ = w1$$ ^ w2$$;
 */
   asm volatile (".option push\n"
          ".option arch,+v\n"
          "vsra.vi v2, v1, 7\n"
          "vsll.vi v3, v1, 1\n"
          "vand.vx v2, v2, %[x1d]\n"
          "vxor.vv v1, v3, v2\n"
          ".option pop\n"
          : :
          [x1d]"r"(0x1d)
   );
  }

  /*
 * *(unative_t *)&p[d+NSIZE*$$] ^= wp$$;
 * *(unative_t *)&q[d+NSIZE*$$] ^= wq$$;
 * v0:wp0, v1:wq0, v2:p0, v3:q0
 */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vle8.v v2, (%[wp0])\n"
         "vle8.v v3, (%[wq0])\n"
         "vxor.vv v2, v2, v0\n"
         "vxor.vv v3, v3, v1\n"
         "vse8.v v2, (%[wp0])\n"
         "vse8.v v3, (%[wq0])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&p[d + NSIZE * 0]),
         [wq0]"r"(&q[d + NSIZE * 0])
  );
 }
}

static void raid6_rvv2_gen_syndrome_real(int disks, unsigned long bytes, void **ptrs)
{
 u8 **dptr = (u8 **)ptrs;
 u8 *p, *q;
 unsigned long vl, d;
 int z, z0;

 z0 = disks - 3;  /* Highest data disk */
 p = dptr[z0 + 1];  /* XOR parity */
 q = dptr[z0 + 2];  /* RS syndrome */

 asm volatile (".option push\n"
        ".option arch,+v\n"
        "vsetvli %0, x0, e8, m1, ta, ma\n"
        ".option pop\n"
        : "=&r" (vl)
 );

 /*
 * v0:wp0, v1:wq0, v2:wd0/w20, v3:w10
 * v4:wp1, v5:wq1, v6:wd1/w21, v7:w11
 */
 for (d = 0; d < bytes; d += NSIZE * 2) {
  /* wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE]; */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vle8.v v0, (%[wp0])\n"
         "vle8.v v1, (%[wp0])\n"
         "vle8.v v4, (%[wp1])\n"
         "vle8.v v5, (%[wp1])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&dptr[z0][d + 0 * NSIZE]),
         [wp1]"r"(&dptr[z0][d + 1 * NSIZE])
  );

  for (z = z0 - 1; z >= 0; z--) {
   /*
 * w2$$ = MASK(wq$$);
 * w1$$ = SHLBYTE(wq$$);
 * w2$$ &= NBYTES(0x1d);
 * w1$$ ^= w2$$;
 * wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE];
 * wq$$ = w1$$ ^ wd$$;
 * wp$$ ^= wd$$;
 */
   asm volatile (".option push\n"
          ".option arch,+v\n"
          "vsra.vi v2, v1, 7\n"
          "vsll.vi v3, v1, 1\n"
          "vand.vx v2, v2, %[x1d]\n"
          "vxor.vv v3, v3, v2\n"
          "vle8.v v2, (%[wd0])\n"
          "vxor.vv v1, v3, v2\n"
          "vxor.vv v0, v0, v2\n"

          "vsra.vi v6, v5, 7\n"
          "vsll.vi v7, v5, 1\n"
          "vand.vx v6, v6, %[x1d]\n"
          "vxor.vv v7, v7, v6\n"
          "vle8.v v6, (%[wd1])\n"
          "vxor.vv v5, v7, v6\n"
          "vxor.vv v4, v4, v6\n"
          ".option pop\n"
          : :
          [wd0]"r"(&dptr[z][d + 0 * NSIZE]),
          [wd1]"r"(&dptr[z][d + 1 * NSIZE]),
          [x1d]"r"(0x1d)
   );
  }

  /*
 * *(unative_t *)&p[d+NSIZE*$$] = wp$$;
 * *(unative_t *)&q[d+NSIZE*$$] = wq$$;
 */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vse8.v v0, (%[wp0])\n"
         "vse8.v v1, (%[wq0])\n"
         "vse8.v v4, (%[wp1])\n"
         "vse8.v v5, (%[wq1])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&p[d + NSIZE * 0]),
         [wq0]"r"(&q[d + NSIZE * 0]),
         [wp1]"r"(&p[d + NSIZE * 1]),
         [wq1]"r"(&q[d + NSIZE * 1])
  );
 }
}

static void raid6_rvv2_xor_syndrome_real(int disks, int start, int stop,
      unsigned long bytes, void **ptrs)
{
 u8 **dptr = (u8 **)ptrs;
 u8 *p, *q;
 unsigned long vl, d;
 int z, z0;

 z0 = stop;  /* P/Q right side optimization */
 p = dptr[disks - 2]; /* XOR parity */
 q = dptr[disks - 1]; /* RS syndrome */

 asm volatile (".option push\n"
        ".option arch,+v\n"
        "vsetvli %0, x0, e8, m1, ta, ma\n"
        ".option pop\n"
        : "=&r" (vl)
 );

 /*
 * v0:wp0, v1:wq0, v2:wd0/w20, v3:w10
 * v4:wp1, v5:wq1, v6:wd1/w21, v7:w11
 */
 for (d = 0; d < bytes; d += NSIZE * 2) {
   /* wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE]; */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vle8.v v0, (%[wp0])\n"
         "vle8.v v1, (%[wp0])\n"
         "vle8.v v4, (%[wp1])\n"
         "vle8.v v5, (%[wp1])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&dptr[z0][d + 0 * NSIZE]),
         [wp1]"r"(&dptr[z0][d + 1 * NSIZE])
  );

  /* P/Q data pages */
  for (z = z0 - 1; z >= start; z--) {
   /*
 * w2$$ = MASK(wq$$);
 * w1$$ = SHLBYTE(wq$$);
 * w2$$ &= NBYTES(0x1d);
 * w1$$ ^= w2$$;
 * wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE];
 * wq$$ = w1$$ ^ wd$$;
 * wp$$ ^= wd$$;
 */
   asm volatile (".option push\n"
          ".option arch,+v\n"
          "vsra.vi v2, v1, 7\n"
          "vsll.vi v3, v1, 1\n"
          "vand.vx v2, v2, %[x1d]\n"
          "vxor.vv v3, v3, v2\n"
          "vle8.v v2, (%[wd0])\n"
          "vxor.vv v1, v3, v2\n"
          "vxor.vv v0, v0, v2\n"

          "vsra.vi v6, v5, 7\n"
          "vsll.vi v7, v5, 1\n"
          "vand.vx v6, v6, %[x1d]\n"
          "vxor.vv v7, v7, v6\n"
          "vle8.v v6, (%[wd1])\n"
          "vxor.vv v5, v7, v6\n"
          "vxor.vv v4, v4, v6\n"
          ".option pop\n"
          : :
          [wd0]"r"(&dptr[z][d + 0 * NSIZE]),
          [wd1]"r"(&dptr[z][d + 1 * NSIZE]),
          [x1d]"r"(0x1d)
   );
  }

  /* P/Q left side optimization */
  for (z = start - 1; z >= 0; z--) {
   /*
 * w2$$ = MASK(wq$$);
 * w1$$ = SHLBYTE(wq$$);
 * w2$$ &= NBYTES(0x1d);
 * wq$$ = w1$$ ^ w2$$;
 */
   asm volatile (".option push\n"
          ".option arch,+v\n"
          "vsra.vi v2, v1, 7\n"
          "vsll.vi v3, v1, 1\n"
          "vand.vx v2, v2, %[x1d]\n"
          "vxor.vv v1, v3, v2\n"

          "vsra.vi v6, v5, 7\n"
          "vsll.vi v7, v5, 1\n"
          "vand.vx v6, v6, %[x1d]\n"
          "vxor.vv v5, v7, v6\n"
          ".option pop\n"
          : :
          [x1d]"r"(0x1d)
   );
  }

  /*
 * *(unative_t *)&p[d+NSIZE*$$] ^= wp$$;
 * *(unative_t *)&q[d+NSIZE*$$] ^= wq$$;
 * v0:wp0, v1:wq0, v2:p0, v3:q0
 * v4:wp1, v5:wq1, v6:p1, v7:q1
 */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vle8.v v2, (%[wp0])\n"
         "vle8.v v3, (%[wq0])\n"
         "vxor.vv v2, v2, v0\n"
         "vxor.vv v3, v3, v1\n"
         "vse8.v v2, (%[wp0])\n"
         "vse8.v v3, (%[wq0])\n"

         "vle8.v v6, (%[wp1])\n"
         "vle8.v v7, (%[wq1])\n"
         "vxor.vv v6, v6, v4\n"
         "vxor.vv v7, v7, v5\n"
         "vse8.v v6, (%[wp1])\n"
         "vse8.v v7, (%[wq1])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&p[d + NSIZE * 0]),
         [wq0]"r"(&q[d + NSIZE * 0]),
         [wp1]"r"(&p[d + NSIZE * 1]),
         [wq1]"r"(&q[d + NSIZE * 1])
  );
 }
}

static void raid6_rvv4_gen_syndrome_real(int disks, unsigned long bytes, void **ptrs)
{
 u8 **dptr = (u8 **)ptrs;
 u8 *p, *q;
 unsigned long vl, d;
 int z, z0;

 z0 = disks - 3; /* Highest data disk */
 p = dptr[z0 + 1]; /* XOR parity */
 q = dptr[z0 + 2]; /* RS syndrome */

 asm volatile (".option push\n"
        ".option arch,+v\n"
        "vsetvli %0, x0, e8, m1, ta, ma\n"
        ".option pop\n"
        : "=&r" (vl)
 );

 /*
 * v0:wp0, v1:wq0, v2:wd0/w20, v3:w10
 * v4:wp1, v5:wq1, v6:wd1/w21, v7:w11
 * v8:wp2, v9:wq2, v10:wd2/w22, v11:w12
 * v12:wp3, v13:wq3, v14:wd3/w23, v15:w13
 */
 for (d = 0; d < bytes; d += NSIZE * 4) {
  /* wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE]; */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vle8.v v0, (%[wp0])\n"
         "vle8.v v1, (%[wp0])\n"
         "vle8.v v4, (%[wp1])\n"
         "vle8.v v5, (%[wp1])\n"
         "vle8.v v8, (%[wp2])\n"
         "vle8.v v9, (%[wp2])\n"
         "vle8.v v12, (%[wp3])\n"
         "vle8.v v13, (%[wp3])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&dptr[z0][d + 0 * NSIZE]),
         [wp1]"r"(&dptr[z0][d + 1 * NSIZE]),
         [wp2]"r"(&dptr[z0][d + 2 * NSIZE]),
         [wp3]"r"(&dptr[z0][d + 3 * NSIZE])
  );

  for (z = z0 - 1; z >= 0; z--) {
   /*
 * w2$$ = MASK(wq$$);
 * w1$$ = SHLBYTE(wq$$);
 * w2$$ &= NBYTES(0x1d);
 * w1$$ ^= w2$$;
 * wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE];
 * wq$$ = w1$$ ^ wd$$;
 * wp$$ ^= wd$$;
 */
   asm volatile (".option push\n"
          ".option arch,+v\n"
          "vsra.vi v2, v1, 7\n"
          "vsll.vi v3, v1, 1\n"
          "vand.vx v2, v2, %[x1d]\n"
          "vxor.vv v3, v3, v2\n"
          "vle8.v v2, (%[wd0])\n"
          "vxor.vv v1, v3, v2\n"
          "vxor.vv v0, v0, v2\n"

          "vsra.vi v6, v5, 7\n"
          "vsll.vi v7, v5, 1\n"
          "vand.vx v6, v6, %[x1d]\n"
          "vxor.vv v7, v7, v6\n"
          "vle8.v v6, (%[wd1])\n"
          "vxor.vv v5, v7, v6\n"
          "vxor.vv v4, v4, v6\n"

          "vsra.vi v10, v9, 7\n"
          "vsll.vi v11, v9, 1\n"
          "vand.vx v10, v10, %[x1d]\n"
          "vxor.vv v11, v11, v10\n"
          "vle8.v v10, (%[wd2])\n"
          "vxor.vv v9, v11, v10\n"
          "vxor.vv v8, v8, v10\n"

          "vsra.vi v14, v13, 7\n"
          "vsll.vi v15, v13, 1\n"
          "vand.vx v14, v14, %[x1d]\n"
          "vxor.vv v15, v15, v14\n"
          "vle8.v v14, (%[wd3])\n"
          "vxor.vv v13, v15, v14\n"
          "vxor.vv v12, v12, v14\n"
          ".option pop\n"
          : :
          [wd0]"r"(&dptr[z][d + 0 * NSIZE]),
          [wd1]"r"(&dptr[z][d + 1 * NSIZE]),
          [wd2]"r"(&dptr[z][d + 2 * NSIZE]),
          [wd3]"r"(&dptr[z][d + 3 * NSIZE]),
          [x1d]"r"(0x1d)
   );
  }

  /*
 * *(unative_t *)&p[d+NSIZE*$$] = wp$$;
 * *(unative_t *)&q[d+NSIZE*$$] = wq$$;
 */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vse8.v v0, (%[wp0])\n"
         "vse8.v v1, (%[wq0])\n"
         "vse8.v v4, (%[wp1])\n"
         "vse8.v v5, (%[wq1])\n"
         "vse8.v v8, (%[wp2])\n"
         "vse8.v v9, (%[wq2])\n"
         "vse8.v v12, (%[wp3])\n"
         "vse8.v v13, (%[wq3])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&p[d + NSIZE * 0]),
         [wq0]"r"(&q[d + NSIZE * 0]),
         [wp1]"r"(&p[d + NSIZE * 1]),
         [wq1]"r"(&q[d + NSIZE * 1]),
         [wp2]"r"(&p[d + NSIZE * 2]),
         [wq2]"r"(&q[d + NSIZE * 2]),
         [wp3]"r"(&p[d + NSIZE * 3]),
         [wq3]"r"(&q[d + NSIZE * 3])
  );
 }
}

static void raid6_rvv4_xor_syndrome_real(int disks, int start, int stop,
      unsigned long bytes, void **ptrs)
{
 u8 **dptr = (u8 **)ptrs;
 u8 *p, *q;
 unsigned long vl, d;
 int z, z0;

 z0 = stop;  /* P/Q right side optimization */
 p = dptr[disks - 2]; /* XOR parity */
 q = dptr[disks - 1]; /* RS syndrome */

 asm volatile (".option push\n"
        ".option arch,+v\n"
        "vsetvli %0, x0, e8, m1, ta, ma\n"
        ".option pop\n"
        : "=&r" (vl)
 );

 /*
 * v0:wp0, v1:wq0, v2:wd0/w20, v3:w10
 * v4:wp1, v5:wq1, v6:wd1/w21, v7:w11
 * v8:wp2, v9:wq2, v10:wd2/w22, v11:w12
 * v12:wp3, v13:wq3, v14:wd3/w23, v15:w13
 */
 for (d = 0; d < bytes; d += NSIZE * 4) {
   /* wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE]; */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vle8.v v0, (%[wp0])\n"
         "vle8.v v1, (%[wp0])\n"
         "vle8.v v4, (%[wp1])\n"
         "vle8.v v5, (%[wp1])\n"
         "vle8.v v8, (%[wp2])\n"
         "vle8.v v9, (%[wp2])\n"
         "vle8.v v12, (%[wp3])\n"
         "vle8.v v13, (%[wp3])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&dptr[z0][d + 0 * NSIZE]),
         [wp1]"r"(&dptr[z0][d + 1 * NSIZE]),
         [wp2]"r"(&dptr[z0][d + 2 * NSIZE]),
         [wp3]"r"(&dptr[z0][d + 3 * NSIZE])
  );

  /* P/Q data pages */
  for (z = z0 - 1; z >= start; z--) {
   /*
 * w2$$ = MASK(wq$$);
 * w1$$ = SHLBYTE(wq$$);
 * w2$$ &= NBYTES(0x1d);
 * w1$$ ^= w2$$;
 * wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE];
 * wq$$ = w1$$ ^ wd$$;
 * wp$$ ^= wd$$;
 */
   asm volatile (".option push\n"
          ".option arch,+v\n"
          "vsra.vi v2, v1, 7\n"
          "vsll.vi v3, v1, 1\n"
          "vand.vx v2, v2, %[x1d]\n"
          "vxor.vv v3, v3, v2\n"
          "vle8.v v2, (%[wd0])\n"
          "vxor.vv v1, v3, v2\n"
          "vxor.vv v0, v0, v2\n"

          "vsra.vi v6, v5, 7\n"
          "vsll.vi v7, v5, 1\n"
          "vand.vx v6, v6, %[x1d]\n"
          "vxor.vv v7, v7, v6\n"
          "vle8.v v6, (%[wd1])\n"
          "vxor.vv v5, v7, v6\n"
          "vxor.vv v4, v4, v6\n"

          "vsra.vi v10, v9, 7\n"
          "vsll.vi v11, v9, 1\n"
          "vand.vx v10, v10, %[x1d]\n"
          "vxor.vv v11, v11, v10\n"
          "vle8.v v10, (%[wd2])\n"
          "vxor.vv v9, v11, v10\n"
          "vxor.vv v8, v8, v10\n"

          "vsra.vi v14, v13, 7\n"
          "vsll.vi v15, v13, 1\n"
          "vand.vx v14, v14, %[x1d]\n"
          "vxor.vv v15, v15, v14\n"
          "vle8.v v14, (%[wd3])\n"
          "vxor.vv v13, v15, v14\n"
          "vxor.vv v12, v12, v14\n"
          ".option pop\n"
          : :
          [wd0]"r"(&dptr[z][d + 0 * NSIZE]),
          [wd1]"r"(&dptr[z][d + 1 * NSIZE]),
          [wd2]"r"(&dptr[z][d + 2 * NSIZE]),
          [wd3]"r"(&dptr[z][d + 3 * NSIZE]),
          [x1d]"r"(0x1d)
   );
  }

  /* P/Q left side optimization */
  for (z = start - 1; z >= 0; z--) {
   /*
 * w2$$ = MASK(wq$$);
 * w1$$ = SHLBYTE(wq$$);
 * w2$$ &= NBYTES(0x1d);
 * wq$$ = w1$$ ^ w2$$;
 */
   asm volatile (".option push\n"
          ".option arch,+v\n"
          "vsra.vi v2, v1, 7\n"
          "vsll.vi v3, v1, 1\n"
          "vand.vx v2, v2, %[x1d]\n"
          "vxor.vv v1, v3, v2\n"

          "vsra.vi v6, v5, 7\n"
          "vsll.vi v7, v5, 1\n"
          "vand.vx v6, v6, %[x1d]\n"
          "vxor.vv v5, v7, v6\n"

          "vsra.vi v10, v9, 7\n"
          "vsll.vi v11, v9, 1\n"
          "vand.vx v10, v10, %[x1d]\n"
          "vxor.vv v9, v11, v10\n"

          "vsra.vi v14, v13, 7\n"
          "vsll.vi v15, v13, 1\n"
          "vand.vx v14, v14, %[x1d]\n"
          "vxor.vv v13, v15, v14\n"
          ".option pop\n"
          : :
          [x1d]"r"(0x1d)
   );
  }

  /*
 * *(unative_t *)&p[d+NSIZE*$$] ^= wp$$;
 * *(unative_t *)&q[d+NSIZE*$$] ^= wq$$;
 * v0:wp0, v1:wq0, v2:p0, v3:q0
 * v4:wp1, v5:wq1, v6:p1, v7:q1
 * v8:wp2, v9:wq2, v10:p2, v11:q2
 * v12:wp3, v13:wq3, v14:p3, v15:q3
 */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vle8.v v2, (%[wp0])\n"
         "vle8.v v3, (%[wq0])\n"
         "vxor.vv v2, v2, v0\n"
         "vxor.vv v3, v3, v1\n"
         "vse8.v v2, (%[wp0])\n"
         "vse8.v v3, (%[wq0])\n"

         "vle8.v v6, (%[wp1])\n"
         "vle8.v v7, (%[wq1])\n"
         "vxor.vv v6, v6, v4\n"
         "vxor.vv v7, v7, v5\n"
         "vse8.v v6, (%[wp1])\n"
         "vse8.v v7, (%[wq1])\n"

         "vle8.v v10, (%[wp2])\n"
         "vle8.v v11, (%[wq2])\n"
         "vxor.vv v10, v10, v8\n"
         "vxor.vv v11, v11, v9\n"
         "vse8.v v10, (%[wp2])\n"
         "vse8.v v11, (%[wq2])\n"

         "vle8.v v14, (%[wp3])\n"
         "vle8.v v15, (%[wq3])\n"
         "vxor.vv v14, v14, v12\n"
         "vxor.vv v15, v15, v13\n"
         "vse8.v v14, (%[wp3])\n"
         "vse8.v v15, (%[wq3])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&p[d + NSIZE * 0]),
         [wq0]"r"(&q[d + NSIZE * 0]),
         [wp1]"r"(&p[d + NSIZE * 1]),
         [wq1]"r"(&q[d + NSIZE * 1]),
         [wp2]"r"(&p[d + NSIZE * 2]),
         [wq2]"r"(&q[d + NSIZE * 2]),
         [wp3]"r"(&p[d + NSIZE * 3]),
         [wq3]"r"(&q[d + NSIZE * 3])
  );
 }
}

static void raid6_rvv8_gen_syndrome_real(int disks, unsigned long bytes, void **ptrs)
{
 u8 **dptr = (u8 **)ptrs;
 u8 *p, *q;
 unsigned long vl, d;
 int z, z0;

 z0 = disks - 3; /* Highest data disk */
 p = dptr[z0 + 1]; /* XOR parity */
 q = dptr[z0 + 2]; /* RS syndrome */

 asm volatile (".option push\n"
        ".option arch,+v\n"
        "vsetvli %0, x0, e8, m1, ta, ma\n"
        ".option pop\n"
        : "=&r" (vl)
 );

 /*
 * v0:wp0,   v1:wq0,  v2:wd0/w20,  v3:w10
 * v4:wp1,   v5:wq1,  v6:wd1/w21,  v7:w11
 * v8:wp2,   v9:wq2, v10:wd2/w22, v11:w12
 * v12:wp3, v13:wq3, v14:wd3/w23, v15:w13
 * v16:wp4, v17:wq4, v18:wd4/w24, v19:w14
 * v20:wp5, v21:wq5, v22:wd5/w25, v23:w15
 * v24:wp6, v25:wq6, v26:wd6/w26, v27:w16
 * v28:wp7, v29:wq7, v30:wd7/w27, v31:w17
 */
 for (d = 0; d < bytes; d += NSIZE * 8) {
  /* wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE]; */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vle8.v v0, (%[wp0])\n"
         "vle8.v v1, (%[wp0])\n"
         "vle8.v v4, (%[wp1])\n"
         "vle8.v v5, (%[wp1])\n"
         "vle8.v v8, (%[wp2])\n"
         "vle8.v v9, (%[wp2])\n"
         "vle8.v v12, (%[wp3])\n"
         "vle8.v v13, (%[wp3])\n"
         "vle8.v v16, (%[wp4])\n"
         "vle8.v v17, (%[wp4])\n"
         "vle8.v v20, (%[wp5])\n"
         "vle8.v v21, (%[wp5])\n"
         "vle8.v v24, (%[wp6])\n"
         "vle8.v v25, (%[wp6])\n"
         "vle8.v v28, (%[wp7])\n"
         "vle8.v v29, (%[wp7])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&dptr[z0][d + 0 * NSIZE]),
         [wp1]"r"(&dptr[z0][d + 1 * NSIZE]),
         [wp2]"r"(&dptr[z0][d + 2 * NSIZE]),
         [wp3]"r"(&dptr[z0][d + 3 * NSIZE]),
         [wp4]"r"(&dptr[z0][d + 4 * NSIZE]),
         [wp5]"r"(&dptr[z0][d + 5 * NSIZE]),
         [wp6]"r"(&dptr[z0][d + 6 * NSIZE]),
         [wp7]"r"(&dptr[z0][d + 7 * NSIZE])
  );

  for (z = z0 - 1; z >= 0; z--) {
   /*
 * w2$$ = MASK(wq$$);
 * w1$$ = SHLBYTE(wq$$);
 * w2$$ &= NBYTES(0x1d);
 * w1$$ ^= w2$$;
 * wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE];
 * wq$$ = w1$$ ^ wd$$;
 * wp$$ ^= wd$$;
 */
   asm volatile (".option push\n"
          ".option arch,+v\n"
          "vsra.vi v2, v1, 7\n"
          "vsll.vi v3, v1, 1\n"
          "vand.vx v2, v2, %[x1d]\n"
          "vxor.vv v3, v3, v2\n"
          "vle8.v v2, (%[wd0])\n"
          "vxor.vv v1, v3, v2\n"
          "vxor.vv v0, v0, v2\n"

          "vsra.vi v6, v5, 7\n"
          "vsll.vi v7, v5, 1\n"
          "vand.vx v6, v6, %[x1d]\n"
          "vxor.vv v7, v7, v6\n"
          "vle8.v v6, (%[wd1])\n"
          "vxor.vv v5, v7, v6\n"
          "vxor.vv v4, v4, v6\n"

          "vsra.vi v10, v9, 7\n"
          "vsll.vi v11, v9, 1\n"
          "vand.vx v10, v10, %[x1d]\n"
          "vxor.vv v11, v11, v10\n"
          "vle8.v v10, (%[wd2])\n"
          "vxor.vv v9, v11, v10\n"
          "vxor.vv v8, v8, v10\n"

          "vsra.vi v14, v13, 7\n"
          "vsll.vi v15, v13, 1\n"
          "vand.vx v14, v14, %[x1d]\n"
          "vxor.vv v15, v15, v14\n"
          "vle8.v v14, (%[wd3])\n"
          "vxor.vv v13, v15, v14\n"
          "vxor.vv v12, v12, v14\n"

          "vsra.vi v18, v17, 7\n"
          "vsll.vi v19, v17, 1\n"
          "vand.vx v18, v18, %[x1d]\n"
          "vxor.vv v19, v19, v18\n"
          "vle8.v v18, (%[wd4])\n"
          "vxor.vv v17, v19, v18\n"
          "vxor.vv v16, v16, v18\n"

          "vsra.vi v22, v21, 7\n"
          "vsll.vi v23, v21, 1\n"
          "vand.vx v22, v22, %[x1d]\n"
          "vxor.vv v23, v23, v22\n"
          "vle8.v v22, (%[wd5])\n"
          "vxor.vv v21, v23, v22\n"
          "vxor.vv v20, v20, v22\n"

          "vsra.vi v26, v25, 7\n"
          "vsll.vi v27, v25, 1\n"
          "vand.vx v26, v26, %[x1d]\n"
          "vxor.vv v27, v27, v26\n"
          "vle8.v v26, (%[wd6])\n"
          "vxor.vv v25, v27, v26\n"
          "vxor.vv v24, v24, v26\n"

          "vsra.vi v30, v29, 7\n"
          "vsll.vi v31, v29, 1\n"
          "vand.vx v30, v30, %[x1d]\n"
          "vxor.vv v31, v31, v30\n"
          "vle8.v v30, (%[wd7])\n"
          "vxor.vv v29, v31, v30\n"
          "vxor.vv v28, v28, v30\n"
          ".option pop\n"
          : :
          [wd0]"r"(&dptr[z][d + 0 * NSIZE]),
          [wd1]"r"(&dptr[z][d + 1 * NSIZE]),
          [wd2]"r"(&dptr[z][d + 2 * NSIZE]),
          [wd3]"r"(&dptr[z][d + 3 * NSIZE]),
          [wd4]"r"(&dptr[z][d + 4 * NSIZE]),
          [wd5]"r"(&dptr[z][d + 5 * NSIZE]),
          [wd6]"r"(&dptr[z][d + 6 * NSIZE]),
          [wd7]"r"(&dptr[z][d + 7 * NSIZE]),
          [x1d]"r"(0x1d)
   );
  }

  /*
 * *(unative_t *)&p[d+NSIZE*$$] = wp$$;
 * *(unative_t *)&q[d+NSIZE*$$] = wq$$;
 */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vse8.v v0, (%[wp0])\n"
         "vse8.v v1, (%[wq0])\n"
         "vse8.v v4, (%[wp1])\n"
         "vse8.v v5, (%[wq1])\n"
         "vse8.v v8, (%[wp2])\n"
         "vse8.v v9, (%[wq2])\n"
         "vse8.v v12, (%[wp3])\n"
         "vse8.v v13, (%[wq3])\n"
         "vse8.v v16, (%[wp4])\n"
         "vse8.v v17, (%[wq4])\n"
         "vse8.v v20, (%[wp5])\n"
         "vse8.v v21, (%[wq5])\n"
         "vse8.v v24, (%[wp6])\n"
         "vse8.v v25, (%[wq6])\n"
         "vse8.v v28, (%[wp7])\n"
         "vse8.v v29, (%[wq7])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&p[d + NSIZE * 0]),
         [wq0]"r"(&q[d + NSIZE * 0]),
         [wp1]"r"(&p[d + NSIZE * 1]),
         [wq1]"r"(&q[d + NSIZE * 1]),
         [wp2]"r"(&p[d + NSIZE * 2]),
         [wq2]"r"(&q[d + NSIZE * 2]),
         [wp3]"r"(&p[d + NSIZE * 3]),
         [wq3]"r"(&q[d + NSIZE * 3]),
         [wp4]"r"(&p[d + NSIZE * 4]),
         [wq4]"r"(&q[d + NSIZE * 4]),
         [wp5]"r"(&p[d + NSIZE * 5]),
         [wq5]"r"(&q[d + NSIZE * 5]),
         [wp6]"r"(&p[d + NSIZE * 6]),
         [wq6]"r"(&q[d + NSIZE * 6]),
         [wp7]"r"(&p[d + NSIZE * 7]),
         [wq7]"r"(&q[d + NSIZE * 7])
  );
 }
}

static void raid6_rvv8_xor_syndrome_real(int disks, int start, int stop,
      unsigned long bytes, void **ptrs)
{
 u8 **dptr = (u8 **)ptrs;
 u8 *p, *q;
 unsigned long vl, d;
 int z, z0;

 z0 = stop;  /* P/Q right side optimization */
 p = dptr[disks - 2]; /* XOR parity */
 q = dptr[disks - 1]; /* RS syndrome */

 asm volatile (".option push\n"
        ".option arch,+v\n"
        "vsetvli %0, x0, e8, m1, ta, ma\n"
        ".option pop\n"
        : "=&r" (vl)
 );

 /*
 * v0:wp0, v1:wq0, v2:wd0/w20, v3:w10
 * v4:wp1, v5:wq1, v6:wd1/w21, v7:w11
 * v8:wp2, v9:wq2, v10:wd2/w22, v11:w12
 * v12:wp3, v13:wq3, v14:wd3/w23, v15:w13
 * v16:wp4, v17:wq4, v18:wd4/w24, v19:w14
 * v20:wp5, v21:wq5, v22:wd5/w25, v23:w15
 * v24:wp6, v25:wq6, v26:wd6/w26, v27:w16
 * v28:wp7, v29:wq7, v30:wd7/w27, v31:w17
 */
 for (d = 0; d < bytes; d += NSIZE * 8) {
   /* wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE]; */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vle8.v v0, (%[wp0])\n"
         "vle8.v v1, (%[wp0])\n"
         "vle8.v v4, (%[wp1])\n"
         "vle8.v v5, (%[wp1])\n"
         "vle8.v v8, (%[wp2])\n"
         "vle8.v v9, (%[wp2])\n"
         "vle8.v v12, (%[wp3])\n"
         "vle8.v v13, (%[wp3])\n"
         "vle8.v v16, (%[wp4])\n"
         "vle8.v v17, (%[wp4])\n"
         "vle8.v v20, (%[wp5])\n"
         "vle8.v v21, (%[wp5])\n"
         "vle8.v v24, (%[wp6])\n"
         "vle8.v v25, (%[wp6])\n"
         "vle8.v v28, (%[wp7])\n"
         "vle8.v v29, (%[wp7])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&dptr[z0][d + 0 * NSIZE]),
         [wp1]"r"(&dptr[z0][d + 1 * NSIZE]),
         [wp2]"r"(&dptr[z0][d + 2 * NSIZE]),
         [wp3]"r"(&dptr[z0][d + 3 * NSIZE]),
         [wp4]"r"(&dptr[z0][d + 4 * NSIZE]),
         [wp5]"r"(&dptr[z0][d + 5 * NSIZE]),
         [wp6]"r"(&dptr[z0][d + 6 * NSIZE]),
         [wp7]"r"(&dptr[z0][d + 7 * NSIZE])
  );

  /* P/Q data pages */
  for (z = z0 - 1; z >= start; z--) {
   /*
 * w2$$ = MASK(wq$$);
 * w1$$ = SHLBYTE(wq$$);
 * w2$$ &= NBYTES(0x1d);
 * w1$$ ^= w2$$;
 * wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE];
 * wq$$ = w1$$ ^ wd$$;
 * wp$$ ^= wd$$;
 */
   asm volatile (".option push\n"
          ".option arch,+v\n"
          "vsra.vi v2, v1, 7\n"
          "vsll.vi v3, v1, 1\n"
          "vand.vx v2, v2, %[x1d]\n"
          "vxor.vv v3, v3, v2\n"
          "vle8.v v2, (%[wd0])\n"
          "vxor.vv v1, v3, v2\n"
          "vxor.vv v0, v0, v2\n"

          "vsra.vi v6, v5, 7\n"
          "vsll.vi v7, v5, 1\n"
          "vand.vx v6, v6, %[x1d]\n"
          "vxor.vv v7, v7, v6\n"
          "vle8.v v6, (%[wd1])\n"
          "vxor.vv v5, v7, v6\n"
          "vxor.vv v4, v4, v6\n"

          "vsra.vi v10, v9, 7\n"
          "vsll.vi v11, v9, 1\n"
          "vand.vx v10, v10, %[x1d]\n"
          "vxor.vv v11, v11, v10\n"
          "vle8.v v10, (%[wd2])\n"
          "vxor.vv v9, v11, v10\n"
          "vxor.vv v8, v8, v10\n"

          "vsra.vi v14, v13, 7\n"
          "vsll.vi v15, v13, 1\n"
          "vand.vx v14, v14, %[x1d]\n"
          "vxor.vv v15, v15, v14\n"
          "vle8.v v14, (%[wd3])\n"
          "vxor.vv v13, v15, v14\n"
          "vxor.vv v12, v12, v14\n"

          "vsra.vi v18, v17, 7\n"
          "vsll.vi v19, v17, 1\n"
          "vand.vx v18, v18, %[x1d]\n"
          "vxor.vv v19, v19, v18\n"
          "vle8.v v18, (%[wd4])\n"
          "vxor.vv v17, v19, v18\n"
          "vxor.vv v16, v16, v18\n"

          "vsra.vi v22, v21, 7\n"
          "vsll.vi v23, v21, 1\n"
          "vand.vx v22, v22, %[x1d]\n"
          "vxor.vv v23, v23, v22\n"
          "vle8.v v22, (%[wd5])\n"
          "vxor.vv v21, v23, v22\n"
          "vxor.vv v20, v20, v22\n"

          "vsra.vi v26, v25, 7\n"
          "vsll.vi v27, v25, 1\n"
          "vand.vx v26, v26, %[x1d]\n"
          "vxor.vv v27, v27, v26\n"
          "vle8.v v26, (%[wd6])\n"
          "vxor.vv v25, v27, v26\n"
          "vxor.vv v24, v24, v26\n"

          "vsra.vi v30, v29, 7\n"
          "vsll.vi v31, v29, 1\n"
          "vand.vx v30, v30, %[x1d]\n"
          "vxor.vv v31, v31, v30\n"
          "vle8.v v30, (%[wd7])\n"
          "vxor.vv v29, v31, v30\n"
          "vxor.vv v28, v28, v30\n"
          ".option pop\n"
          : :
          [wd0]"r"(&dptr[z][d + 0 * NSIZE]),
          [wd1]"r"(&dptr[z][d + 1 * NSIZE]),
          [wd2]"r"(&dptr[z][d + 2 * NSIZE]),
          [wd3]"r"(&dptr[z][d + 3 * NSIZE]),
          [wd4]"r"(&dptr[z][d + 4 * NSIZE]),
          [wd5]"r"(&dptr[z][d + 5 * NSIZE]),
          [wd6]"r"(&dptr[z][d + 6 * NSIZE]),
          [wd7]"r"(&dptr[z][d + 7 * NSIZE]),
          [x1d]"r"(0x1d)
   );
  }

  /* P/Q left side optimization */
  for (z = start - 1; z >= 0; z--) {
   /*
 * w2$$ = MASK(wq$$);
 * w1$$ = SHLBYTE(wq$$);
 * w2$$ &= NBYTES(0x1d);
 * wq$$ = w1$$ ^ w2$$;
 */
   asm volatile (".option push\n"
          ".option arch,+v\n"
          "vsra.vi v2, v1, 7\n"
          "vsll.vi v3, v1, 1\n"
          "vand.vx v2, v2, %[x1d]\n"
          "vxor.vv v1, v3, v2\n"

          "vsra.vi v6, v5, 7\n"
          "vsll.vi v7, v5, 1\n"
          "vand.vx v6, v6, %[x1d]\n"
          "vxor.vv v5, v7, v6\n"

          "vsra.vi v10, v9, 7\n"
          "vsll.vi v11, v9, 1\n"
          "vand.vx v10, v10, %[x1d]\n"
          "vxor.vv v9, v11, v10\n"

          "vsra.vi v14, v13, 7\n"
          "vsll.vi v15, v13, 1\n"
          "vand.vx v14, v14, %[x1d]\n"
          "vxor.vv v13, v15, v14\n"

          "vsra.vi v18, v17, 7\n"
          "vsll.vi v19, v17, 1\n"
          "vand.vx v18, v18, %[x1d]\n"
          "vxor.vv v17, v19, v18\n"

          "vsra.vi v22, v21, 7\n"
          "vsll.vi v23, v21, 1\n"
          "vand.vx v22, v22, %[x1d]\n"
          "vxor.vv v21, v23, v22\n"

          "vsra.vi v26, v25, 7\n"
          "vsll.vi v27, v25, 1\n"
          "vand.vx v26, v26, %[x1d]\n"
          "vxor.vv v25, v27, v26\n"

          "vsra.vi v30, v29, 7\n"
          "vsll.vi v31, v29, 1\n"
          "vand.vx v30, v30, %[x1d]\n"
          "vxor.vv v29, v31, v30\n"
          ".option pop\n"
          : :
          [x1d]"r"(0x1d)
   );
  }

  /*
 * *(unative_t *)&p[d+NSIZE*$$] ^= wp$$;
 * *(unative_t *)&q[d+NSIZE*$$] ^= wq$$;
 * v0:wp0, v1:wq0, v2:p0, v3:q0
 * v4:wp1, v5:wq1, v6:p1, v7:q1
 * v8:wp2, v9:wq2, v10:p2, v11:q2
 * v12:wp3, v13:wq3, v14:p3, v15:q3
 * v16:wp4, v17:wq4, v18:p4, v19:q4
 * v20:wp5, v21:wq5, v22:p5, v23:q5
 * v24:wp6, v25:wq6, v26:p6, v27:q6
 * v28:wp7, v29:wq7, v30:p7, v31:q7
 */
  asm volatile (".option push\n"
         ".option arch,+v\n"
         "vle8.v v2, (%[wp0])\n"
         "vle8.v v3, (%[wq0])\n"
         "vxor.vv v2, v2, v0\n"
         "vxor.vv v3, v3, v1\n"
         "vse8.v v2, (%[wp0])\n"
         "vse8.v v3, (%[wq0])\n"

         "vle8.v v6, (%[wp1])\n"
         "vle8.v v7, (%[wq1])\n"
         "vxor.vv v6, v6, v4\n"
         "vxor.vv v7, v7, v5\n"
         "vse8.v v6, (%[wp1])\n"
         "vse8.v v7, (%[wq1])\n"

         "vle8.v v10, (%[wp2])\n"
         "vle8.v v11, (%[wq2])\n"
         "vxor.vv v10, v10, v8\n"
         "vxor.vv v11, v11, v9\n"
         "vse8.v v10, (%[wp2])\n"
         "vse8.v v11, (%[wq2])\n"

         "vle8.v v14, (%[wp3])\n"
         "vle8.v v15, (%[wq3])\n"
         "vxor.vv v14, v14, v12\n"
         "vxor.vv v15, v15, v13\n"
         "vse8.v v14, (%[wp3])\n"
         "vse8.v v15, (%[wq3])\n"

         "vle8.v v18, (%[wp4])\n"
         "vle8.v v19, (%[wq4])\n"
         "vxor.vv v18, v18, v16\n"
         "vxor.vv v19, v19, v17\n"
         "vse8.v v18, (%[wp4])\n"
         "vse8.v v19, (%[wq4])\n"

         "vle8.v v22, (%[wp5])\n"
         "vle8.v v23, (%[wq5])\n"
         "vxor.vv v22, v22, v20\n"
         "vxor.vv v23, v23, v21\n"
         "vse8.v v22, (%[wp5])\n"
         "vse8.v v23, (%[wq5])\n"

         "vle8.v v26, (%[wp6])\n"
         "vle8.v v27, (%[wq6])\n"
         "vxor.vv v26, v26, v24\n"
         "vxor.vv v27, v27, v25\n"
         "vse8.v v26, (%[wp6])\n"
         "vse8.v v27, (%[wq6])\n"

         "vle8.v v30, (%[wp7])\n"
         "vle8.v v31, (%[wq7])\n"
         "vxor.vv v30, v30, v28\n"
         "vxor.vv v31, v31, v29\n"
         "vse8.v v30, (%[wp7])\n"
         "vse8.v v31, (%[wq7])\n"
         ".option pop\n"
         : :
         [wp0]"r"(&p[d + NSIZE * 0]),
         [wq0]"r"(&q[d + NSIZE * 0]),
         [wp1]"r"(&p[d + NSIZE * 1]),
         [wq1]"r"(&q[d + NSIZE * 1]),
         [wp2]"r"(&p[d + NSIZE * 2]),
         [wq2]"r"(&q[d + NSIZE * 2]),
         [wp3]"r"(&p[d + NSIZE * 3]),
         [wq3]"r"(&q[d + NSIZE * 3]),
         [wp4]"r"(&p[d + NSIZE * 4]),
         [wq4]"r"(&q[d + NSIZE * 4]),
         [wp5]"r"(&p[d + NSIZE * 5]),
         [wq5]"r"(&q[d + NSIZE * 5]),
         [wp6]"r"(&p[d + NSIZE * 6]),
         [wq6]"r"(&q[d + NSIZE * 6]),
         [wp7]"r"(&p[d + NSIZE * 7]),
         [wq7]"r"(&q[d + NSIZE * 7])
  );
 }
}

RAID6_RVV_WRAPPER(1);
RAID6_RVV_WRAPPER(2);
RAID6_RVV_WRAPPER(4);
RAID6_RVV_WRAPPER(8);