simple_v_extension/daxpy_example.mdwn

   1     # c code
   2     void daxpy(size_t n, double a, const double x[], double y[])
   3     {
   4      for (size_t i = 0; i < n; i++) {
   5        y[i] = a*x[i] + y[i];
   6      }
   7     }
   8
   9     # SV Version
  10     # a0 is n, a1 is ptr to x[0], a2 is ptr to y[0], fa0 is a (scalar)
  11      VBLK.REG[0] = {type: F, isvec: 1, regkey: a3, regidx: a3, elwidth: dflt}
  12      VBLK.REG[1] = {type: F, isvec: 1, regkey: a7, regidx: a7, elwidth: dflt}
  13     loop:
  14      VBLK.SETVL  t0, a0, #4   # MVL=4, vl = t0 = min(a0, MVL))
  15      c.ld     a3, a1          # load 4 registers a3-6 from x
  16      c.slli   t1, t0, 3       # t1 = vl * 8 (in bytes)
  17      c.ld     a7, a2          # load 4 registers a7-10 from y
  18      c.add    a1, a1, t1      # increment pointer to x by vl*8
  19      fmadd  a7, a3, fa0, a7   # v1 += v0 * fa0 (y = a * x + y)
  20      c.sub    a0, a0, t0      # n -= vl (t0)
  21      c.st     a7, a2          # store 4 registers a7-10 to y
  22      c.add    a2, a2, t1      # increment pointer to y by vl*8
  23      c.bnez   a0, loop        # repeat if n != 0
  24
  25     # RVV version
  26     # a0 is n, a1 is pointer to x[0], a2 is pointer to y[0], fa0 is a
  27       0:  li t0, 2<<25
  28       4:  vsetdcfg t0             # enable 2 64b Fl.Pt. registers
  29     loop:
  30       8:  setvl  t0, a0           # vl = t0 = min(mvl, n)
  31       c:  vld    v0, a1           # load vector x
  32       10:  slli   t1, t0, 3        # t1 = vl * 8 (in bytes)
  33       14:  vld    v1, a2           # load vector y
  34       18:  add    a1, a1, t1       # increment pointer to x by vl*8
  35       1c:  vfmadd v1, v0, fa0, v1  # v1 += v0 * fa0 (y = a * x + y)
  36       20:  sub    a0, a0, t0       # n -= vl (t0)
  37       24:  vst    v1, a2           # store Y
  38       28:  add    a2, a2, t1       # increment pointer to y by vl*8
  39       2c:  bnez   a0, loop         # repeat if n != 0
  40       30:  ret                     # return