simple_v_extension/daxpy_example.mdwn

   1 # c code
   2
   3 ```
   4     void daxpy(size_t n, double a, const double x[], double y[])
   5     {
   6      for (size_t i = 0; i < n; i++) {
   7        y[i] = a*x[i] + y[i];
   8      }
   9     }
  10 ```
  11
  12 # SVP64 Power ISA version
  13
  14 ```
  15
  16     # r5: n
  17     # r5: x
  18     # r6: y
  19     # fp1: a
  20     mtctr 5                 # move n to CTR
  21     addi r10,r6,0           # copy y-ptr into r10
  22 .L2
  23     setvl MAXVL=32,VL=CTR   # could do more
  24     sv.lfdup/els *32,8(6)   # load from x
  25     sv.lfdup/els *64,8(7)   # load from y
  26     sv.fmadd *64,*64,1,*32  # fmadd
  27     stfdup/els *64,8(10)    # store y-copy
  28     sv.bc/ctr .L2           # decrement VL by CTR
  29     blr                     # return
  30 ```
  31
  32 # RVV version
  33
  34 ```
  35     # a0 is n, a1 is pointer to x[0], a2 is pointer to y[0], fa0 is a
  36       li t0, 2<<25
  37       vsetdcfg t0             # enable 2 64b Fl.Pt. registers
  38     loop:
  39       setvl  t0, a0           # vl = t0 = min(mvl, n)
  40       vld    v0, a1           # load vector x
  41       c.slli   t1, t0, 3      # t1 = vl * 8 (in bytes)
  42       vld    v1, a2           # load vector y
  43       c.add    a1, a1, t1     # increment pointer to x by vl*8
  44       vfmadd v1, v0, fa0, v1  # v1 += v0 * fa0 (y = a * x + y)
  45       c.sub    a0, a0, t0     # n -= vl (t0)
  46       vst    v1, a2           # store Y
  47       c.add    a2, a2, t1     # increment pointer to y by vl*8
  48       c.bnez   a0, loop       # repeat if n != 0
  49       c.ret                   # return
  50 ```