ceph/src/isa-l/erasure_code/gf_2vect_mad_avx.asm

   1 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
   2 ;  Copyright(c) 2011-2015 Intel Corporation All rights reserved.
   3 ;
   4 ;  Redistribution and use in source and binary forms, with or without
   5 ;  modification, are permitted provided that the following conditions
   6 ;  are met:
   7 ;    * Redistributions of source code must retain the above copyright
   8 ;      notice, this list of conditions and the following disclaimer.
   9 ;    * Redistributions in binary form must reproduce the above copyright
  10 ;      notice, this list of conditions and the following disclaimer in
  11 ;      the documentation and/or other materials provided with the
  12 ;      distribution.
  13 ;    * Neither the name of Intel Corporation nor the names of its
  14 ;      contributors may be used to endorse or promote products derived
  15 ;      from this software without specific prior written permission.
  16 ;
  17 ;  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  18 ;  "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  19 ;  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  20 ;  A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  21 ;  OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  22 ;  SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  23 ;  LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  24 ;  DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  25 ;  THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  26 ;  (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  27 ;  OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  28 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
  29
  30 ;;;
  31 ;;; gf_2vect_mad_avx(len, vec, vec_i, mul_array, src, dest);
  32 ;;;
  33
  34 %include "reg_sizes.asm"
  35
  36 %define PS 8
  37
  38 %ifidn __OUTPUT_FORMAT__, win64
  39  %define arg0  rcx
  40  %define arg0.w ecx
  41  %define arg1  rdx
  42  %define arg2  r8
  43  %define arg3  r9
  44  %define arg4  r12
  45  %define arg5  r15
  46  %define tmp   r11
  47  %define tmp2   r10
  48  %define return rax
  49  %define return.w eax
  50  %define stack_size 16*9 + 3*8
  51  %define arg(x)      [rsp + stack_size + PS + PS*x]
  52  %define func(x) proc_frame x
  53
  54 %macro FUNC_SAVE 0
  55         sub     rsp, stack_size
  56         movdqa  [rsp+16*0],xmm6
  57         movdqa  [rsp+16*1],xmm7
  58         movdqa  [rsp+16*2],xmm8
  59         movdqa  [rsp+16*3],xmm9
  60         movdqa  [rsp+16*4],xmm10
  61         movdqa  [rsp+16*5],xmm11
  62         movdqa  [rsp+16*6],xmm12
  63         movdqa  [rsp+16*7],xmm13
  64         movdqa  [rsp+16*8],xmm14
  65         save_reg        r12,  9*16 + 0*8
  66         save_reg        r15,  9*16 + 1*8
  67         end_prolog
  68         mov     arg4, arg(4)
  69         mov     arg5, arg(5)
  70 %endmacro
  71
  72 %macro FUNC_RESTORE 0
  73         movdqa  xmm6, [rsp+16*0]
  74         movdqa  xmm7, [rsp+16*1]
  75         movdqa  xmm8, [rsp+16*2]
  76         movdqa  xmm9, [rsp+16*3]
  77         movdqa  xmm10, [rsp+16*4]
  78         movdqa  xmm11, [rsp+16*5]
  79         movdqa  xmm12, [rsp+16*6]
  80         movdqa  xmm13, [rsp+16*7]
  81         movdqa  xmm14, [rsp+16*8]
  82         mov     r12,  [rsp + 9*16 + 0*8]
  83         mov     r15,  [rsp + 9*16 + 1*8]
  84         add     rsp, stack_size
  85 %endmacro
  86
  87 %elifidn __OUTPUT_FORMAT__, elf64
  88  %define arg0  rdi
  89  %define arg0.w edi
  90  %define arg1  rsi
  91  %define arg2  rdx
  92  %define arg3  rcx
  93  %define arg4  r8
  94  %define arg5  r9
  95  %define tmp   r11
  96  %define tmp2   r10
  97  %define return rax
  98  %define return.w eax
  99
 100  %define func(x) x:
 101  %define FUNC_SAVE
 102  %define FUNC_RESTORE
 103 %endif
 104
 105 ;;; gf_2vect_mad_avx(len, vec, vec_i, mul_array, src, dest)
 106 %define len   arg0
 107 %define len.w arg0.w
 108 %define vec    arg1
 109 %define vec_i    arg2
 110 %define mul_array arg3
 111 %define src   arg4
 112 %define dest1 arg5
 113 %define pos   return
 114 %define pos.w return.w
 115
 116 %define dest2 tmp2
 117
 118 %ifndef EC_ALIGNED_ADDR
 119 ;;; Use Un-aligned load/store
 120  %define XLDR vmovdqu
 121  %define XSTR vmovdqu
 122 %else
 123 ;;; Use Non-temporal load/stor
 124  %ifdef NO_NT_LDST
 125   %define XLDR vmovdqa
 126   %define XSTR vmovdqa
 127  %else
 128   %define XLDR vmovntdqa
 129   %define XSTR vmovntdq
 130  %endif
 131 %endif
 132
 133
 134 default rel
 135
 136 [bits 64]
 137 section .text
 138
 139 %define xmask0f  xmm14
 140 %define xgft1_lo  xmm13
 141 %define xgft1_hi  xmm12
 142 %define xgft2_lo  xmm11
 143 %define xgft2_hi  xmm10
 144
 145 %define x0      xmm0
 146 %define xtmpa   xmm1
 147 %define xtmph1  xmm2
 148 %define xtmpl1  xmm3
 149 %define xtmph2  xmm4
 150 %define xtmpl2  xmm5
 151 %define xd1     xmm6
 152 %define xd2     xmm7
 153 %define xtmpd1  xmm8
 154 %define xtmpd2  xmm9
 155
 156
 157 align 16
 158 global gf_2vect_mad_avx:function
 159
 160 func(gf_2vect_mad_avx)
 161         FUNC_SAVE
 162         sub     len, 16
 163         jl      .return_fail
 164
 165         xor     pos, pos
 166         vmovdqa xmask0f, [mask0f]       ;Load mask of lower nibble in each byte
 167         sal     vec_i, 5                ;Multiply by 32
 168         sal     vec, 5
 169         lea     tmp, [mul_array + vec_i]
 170         vmovdqu xgft1_lo, [tmp]         ;Load array Ax{00}, Ax{01}, Ax{02}, ...
 171         vmovdqu xgft1_hi, [tmp+16]      ; " Ax{00}, Ax{10}, Ax{20}, ... , Ax{f0}
 172         vmovdqu xgft2_lo, [tmp+vec]     ;Load array Bx{00}, Bx{01}, Bx{02}, ...
 173         vmovdqu xgft2_hi, [tmp+vec+16]  ; " Bx{00}, Bx{10}, Bx{20}, ... , Bx{f0}
 174
 175         mov     dest2, [dest1+PS]
 176         mov     dest1, [dest1]
 177
 178         XLDR    xtmpd1, [dest1+len]     ;backup the last 16 bytes in dest
 179         XLDR    xtmpd2, [dest2+len]     ;backup the last 16 bytes in dest
 180
 181 .loop16:
 182         XLDR    xd1, [dest1+pos]                ;Get next dest vector
 183         XLDR    xd2, [dest2+pos]                ;Get next dest vector
 184 .loop16_overlap:
 185         XLDR    x0, [src+pos]           ;Get next source vector
 186
 187         vpand   xtmpa, x0, xmask0f      ;Mask low src nibble in bits 4-0
 188         vpsraw  x0, x0, 4               ;Shift to put high nibble into bits 4-0
 189         vpand   x0, x0, xmask0f         ;Mask high src nibble in bits 4-0
 190
 191         vpshufb xtmph1, xgft1_hi, x0    ;Lookup mul table of high nibble
 192         vpshufb xtmpl1, xgft1_lo, xtmpa ;Lookup mul table of low nibble
 193         vpxor   xtmph1, xtmph1, xtmpl1 ;GF add high and low partials
 194         vpxor   xd1, xd1, xtmph1        ;xd1 += partial
 195
 196         vpshufb xtmph2, xgft2_hi, x0    ;Lookup mul table of high nibble
 197         vpshufb xtmpl2, xgft2_lo, xtmpa ;Lookup mul table of low nibble
 198         vpxor   xtmph2, xtmph2, xtmpl2 ;GF add high and low partials
 199         vpxor   xd2, xd2, xtmph2        ;xd2 += partial
 200
 201         XSTR    [dest1+pos], xd1
 202         XSTR    [dest2+pos], xd2
 203
 204         add     pos, 16                 ;Loop on 16 bytes at a time
 205         cmp     pos, len
 206         jle     .loop16
 207
 208         lea     tmp, [len + 16]
 209         cmp     pos, tmp
 210         je      .return_pass
 211
 212         ;; Tail len
 213         mov     pos, len        ;Overlapped offset length-16
 214         vmovdqa xd1, xtmpd1     ;Restore xd1
 215         vmovdqa xd2, xtmpd2     ;Restore xd2
 216         jmp     .loop16_overlap ;Do one more overlap pass
 217
 218 .return_pass:
 219         mov     return, 0
 220         FUNC_RESTORE
 221         ret
 222
 223 .return_fail:
 224         mov     return, 1
 225         FUNC_RESTORE
 226         ret
 227
 228 endproc_frame
 229
 230 section .data
 231
 232 align 16
 233 mask0f: dq 0x0f0f0f0f0f0f0f0f, 0x0f0f0f0f0f0f0f0f
 234
 235 ;;;       func             core, ver, snum
 236 slversion gf_2vect_mad_avx, 02,  01,  0204