scratch/foo.lisp

   1 (in-package :sb-vm)
   2
   3 (define-vop (my-vop)
   4   (:policy :fast-safe)
   5
   6   (:args (vector1 :scs (descriptor-reg))
   7          (vector2 :scs (descriptor-reg)))
   8   (:arg-types simple-array-single-float simple-array-single-float)
   9
  10   (:temporary (:sc unsigned-reg) index)
  11
  12 ;;  (:temporary (:sc unsigned-reg) temp1)
  13 ;;  (:temporary (:sc unsigned-reg) temp2)
  14
  15   (:temporary (:sc sse-reg) sse-temp1)
  16   (:temporary (:sc sse-reg) sse-temp2)
  17
  18   (:generator 10
  19
  20               (inst xor index index)
  21
  22
  23               (inst movups sse-temp1
  24                     (make-ea :dword :base vector1 :index index
  25                              :disp (- (* vector-data-offset n-word-bytes) other-pointer-lowtag)))
  26               (inst movups sse-temp2
  27                     (make-ea :dword :base vector2 :index index
  28                              :disp (- (* vector-data-offset n-word-bytes) other-pointer-lowtag)))
  29
  30               (inst addps sse-temp1 sse-temp2)
  31
  32 ;;            (inst add index 4)
  33
  34               (inst movups
  35                     (make-ea :dword :base vector1 :index index
  36                              :disp (- (* vector-data-offset n-word-bytes) other-pointer-lowtag))
  37                     sse-temp1)
  38 #|
  39               (inst add index 4)
  40               (inst mov
  41                     (make-ea :dword :base vector1 :index index
  42                              :disp (- (* vector-data-offset n-word-bytes) other-pointer-lowtag))
  43                     index)
  44
  45               (inst add index 4)
  46               (inst movups
  47                     (make-ea :dword :base vector1 :index index
  48                              :disp (- (* vector-data-offset n-word-bytes) other-pointer-lowtag))
  49                     sse-temp2)
  50 |#
  51               ))
  52
  53 #|
  54 00000000 <my_func>:
  55    0:   31 c0                   xor    %eax,%eax                                                                      2:   0f 10 04 c6             movups (%esi,%eax,8),%xmm0
  56    6:   0f 10 0c c7             movups (%edi,%eax,8),%xmm1
  57    a:   0f 58 c1                addps  %xmm1,%xmm0
  58    d:   0f 11 44 c5 00          movups %xmm0,0x0(%ebp,%eax,8)
  59
  60 --- v2:
  61    0:   31 c0                   xor    %eax,%eax
  62    2:   0f 10 44 03 01          movups 0x1(%ebx,%eax,1),%xmm0
  63    7:   0f 10 4c 01 01          movups 0x1(%ecx,%eax,1),%xmm1
  64    c:   0f 58 c1                addps  %xmm1,%xmm0
  65    f:   0f 11 44 01 01          movups %xmm0,0x1(%ecx,%eax,1)
  66
  67 --- v3:
  68    2:   0f 10 43 01             movups 0x1(%ebx),%xmm0
  69    6:   0f 10 49 01             movups 0x1(%ecx),%xmm1
  70    a:   0f 58 c1                addps  %xmm1,%xmm0
  71    d:   0f 11 41 01             movups %xmm0,0x1(%ecx)
  72
  73 --- v4:
  74    2:   0f 10 44 1a 01          movups 0x1(%edx,%ebx,1),%xmm0
  75    7:   0f 10 4c 0e 01          movups 0x1(%esi,%ecx,1),%xmm1
  76    c:   0f 58 c1                addps  %xmm1,%xmm0
  77    f:   0f 11 41 01             movups %xmm0,0x1(%ecx)
  78
  79 10h = MOVUPS Vps, Wps
  80 11h = MOVUPS Wps, Vps
  81
  82 V = 128bit xmm reg specified by the modrm reg field.
  83 W = 128bit xmm register or mem op specified by the modrm byte.
  84 ps = 128bit single-precision float operand
  85
  86 movups  xmm0,        [ebx       + 01]
  87 movups  md reg r/m  sc idx bse  disp8
  88 0f 10   01 000 100  00 000 011    01
  89        +d8 xm0 sib  *0 +0  ebx   +01
  90 |#
  91 #|
  92 ;      43E: L4:   31C0             XOR EAX, EAX
  93
  94         7 6  5 4 3  2 1 0
  95         m d  r e g  r / m
  96
  97 44h = b 0 1  0 0 0  1 0 0
  98 4Ch = b 0 1  0 0 1  1 0 0
  99 64h = b 0 1  1 0 0  1 0 0
 100 E0h = b 1 1  1 0 0  0 0 0
 101 C1h = b 1 1  0 0 0  0 0 1
 102 43h = b 0 1  0 0 0  0 1 1
 103 49h = b 0 1  0 0 1  0 0 1
 104
 105 r/m b100 => has sib byte
 106
 107 modrm md+r/m field:
 108    r/m= 000, 001, 010, 011, 100, 101,    110  , 111
 109 md
 110 00 =     ax,  cx,  dx,  bx, sib, rip+d32, si,   di
 111 01 = --||-- + disp8                  , bp+disp8, ..
 112 10 = --||-- + disp32
 113 11 = al/ax/eax/mmx0/xmm0, 1, 2, 3, 4, 5, 6, 7
 114
 115 modrm reg:
 116        000 001 010 011 100 101 110 111
 117 reg32  eax ecx edx ebx esp ebp esi edi
 118 xmm    xm0 xm1 xm2 xm3 xm4 xm5 xm6 xm7 ;; actually xmm0..xmm7
 119
 120
 121 44h = md 01, r/m 100, reg 000, => xmm0, [sib + disp8] => 44 03 01 : xmm0, [ebx + 01], 44 01 01 : xmm0, [ecx + 01]
 122 64h = md 01, r/m 100, reg 100, => xmm4, [sib + disp8] => xmm4, [ebx + 01]
 123 04 C6 = md 00, r/m 100, reg 000, => xmm0, [sib] => xmm0, [esi*8]
 124 4C 01 01 = md 01, reg 001, r/m 100 => xmm1, [sib + disp8] => [ecx + 01]
 125 43h = md 01, reg 0, r/m 011 => xmm0, [ebx + 01]
 126 49h = md 01, reg 1, r/m 001 => xmm1, [ecx + 01]
 127
 128
 129         7 6  5 4 3  2 1 0
 130         s c  i d x  b a s
 131
 132 03h = b 0 0  0 0 0  0 1 1  = eax + ebx*1
 133 01h = b 0 0  0 0 0  0 0 1  = eax + ecx*1
 134 C6h = b 1 1  0 0 0  1 1 0  = eax + esi*8
 135 1Ah = b 0 0  0 1 1  0 1 0  = ebx + edx*1
 136 0Eh = b 0 0  0 0 1  1 1 0  = ecx + esi*1
 137
 138
 139 ;; movups xmm0, ea 0F 10 44 03 01
 140 ;      440:        0F 10 44 01 01         movups xmm0, [eax + ecx + 01]
 141
 142 ;; movups xmm1, ea 0F 10 4C 01 01
 143 ;      445:        0F 10 64 03 01         movups xmm4, [eax + ebx + 01]
 144
 145 ;; addps xmm0, xmm1 0F 58 C1
 146 ;      44A:         0F 58 E0        addps xmm0, xmm4
 147
 148 ;; movups ea, xmm0 0f 11 44 01 01
 149                    0F 11 44 01 01     movups [eax + ecx + 01], xmm0
 150
 151 ;      452:       83C004           ADD EAX, 4
 152
 153
 154 --
 155
 156    c:   0f 58 c4                addps  %xmm4,%xmm0
 157    f:   0f 58 e0                addps  %xmm0,%xmm4
 158
 159 |#