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 count
  17     # r6: x ptr
  18     # r7: y ptr
  19     # fp1: a mul-scalar
  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 into fp32-63
  25     sv.lfdup/els *64,8(7)  # load from y into fp64-95
  26     sv.fmadd *64,*64,1,*32 # (*y) = (*y) * (*x) + fp1
  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 ```
  51
  52 # SVE Version
  53
  54 ```
  55     1  // x0 = &x[0], x1 = &y[0], x2 = &a, x3 = &n
  56     2  daxpy_:
  57     3  ldrswx3, [x3]                     // x3=*n
  58     4  movx4, #0                         // x4=i=0
  59     5  whilelt p0.d, x4, x3              // p0=while(i++<n)
  60     6  ld1rdz0.d, p0/z, [x2]             // p0:z0=bcast(*a)
  61     7  .loop:
  62     8  ld1d z1.d, p0/z, [x0, x4, lsl #3] // p0:z1=x[i]
  63     9  ld1d z2.d, p0/z, [x1, x4, lsl #3] // p0:z2=y[i]
  64     10 fmla z2.d, p0/m, z1.d, z0.d       // p0?z2+=x[i]*a
  65     11 st1d z2.d, p0, [x1, x4, lsl #3]   // p0?y[i]=z2
  66     12 incd x4                           // i+=(VL/64)
  67     13 .latch:
  68     14 whilelt p0.d, x4, x3             // p0=while(i++<n)
  69     15 b.first .loop                    // more to do?
  70     16 ret
  71 ```