ceph/src/isa-l/erasure_code/gf_3vect_mad_sse.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_3vect_mad_sse(len, vec, vec_i, mul_array, src, dest);
  32 ;;;
  33
  34 %include "reg_sizes.asm"
  35
  36 %define PS 8
  37 %ifidn __OUTPUT_FORMAT__, win64
  38  %define arg0  rcx
  39  %define arg0.w ecx
  40  %define arg1  rdx
  41  %define arg2  r8
  42  %define arg3  r9
  43  %define arg4  r12
  44  %define arg5  r15
  45  %define tmp   r11
  46  %define return rax
  47  %define return.w eax
  48  %define stack_size 16*10 + 3*8
  49  %define arg(x)      [rsp + stack_size + PS + PS*x]
  50  %define func(x) proc_frame x
  51
  52 %macro FUNC_SAVE 0
  53         sub     rsp, stack_size
  54         movdqa  [rsp+16*0],xmm6
  55         movdqa  [rsp+16*1],xmm7
  56         movdqa  [rsp+16*2],xmm8
  57         movdqa  [rsp+16*3],xmm9
  58         movdqa  [rsp+16*4],xmm10
  59         movdqa  [rsp+16*5],xmm11
  60         movdqa  [rsp+16*6],xmm12
  61         movdqa  [rsp+16*7],xmm13
  62         movdqa  [rsp+16*8],xmm14
  63         movdqa  [rsp+16*9],xmm15
  64         save_reg        r12,  10*16 + 0*8
  65         save_reg        r15,  10*16 + 1*8
  66         end_prolog
  67         mov     arg4, arg(4)
  68         mov     arg5, arg(5)
  69 %endmacro
  70
  71 %macro FUNC_RESTORE 0
  72         movdqa  xmm6, [rsp+16*0]
  73         movdqa  xmm7, [rsp+16*1]
  74         movdqa  xmm8, [rsp+16*2]
  75         movdqa  xmm9, [rsp+16*3]
  76         movdqa  xmm10, [rsp+16*4]
  77         movdqa  xmm11, [rsp+16*5]
  78         movdqa  xmm12, [rsp+16*6]
  79         movdqa  xmm13, [rsp+16*7]
  80         movdqa  xmm14, [rsp+16*8]
  81         movdqa  xmm15, [rsp+16*9]
  82         mov     r12,  [rsp + 10*16 + 0*8]
  83         mov     r15,  [rsp + 10*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 return rax
  97  %define return.w eax
  98
  99  %define func(x) x:
 100  %define FUNC_SAVE
 101  %define FUNC_RESTORE
 102 %endif
 103
 104 ;;; gf_3vect_mad_sse(len, vec, vec_i, mul_array, src, dest)
 105 %define len   arg0
 106 %define len.w arg0.w
 107 %define vec    arg1
 108 %define vec_i    arg2
 109 %define mul_array arg3
 110 %define src   arg4
 111 %define dest1  arg5
 112 %define pos   return
 113 %define pos.w return.w
 114
 115 %define dest2 mul_array
 116 %define dest3 vec_i
 117
 118 %ifndef EC_ALIGNED_ADDR
 119 ;;; Use Un-aligned load/store
 120  %define XLDR movdqu
 121  %define XSTR movdqu
 122 %else
 123 ;;; Use Non-temporal load/stor
 124  %ifdef NO_NT_LDST
 125   %define XLDR movdqa
 126   %define XSTR movdqa
 127  %else
 128   %define XLDR movntdqa
 129   %define XSTR movntdq
 130  %endif
 131 %endif
 132
 133 default rel
 134
 135 [bits 64]
 136 section .text
 137
 138 %define xmask0f  xmm15
 139 %define xgft1_lo  xmm14
 140 %define xgft1_hi  xmm13
 141 %define xgft2_lo  xmm12
 142 %define xgft2_hi  xmm11
 143 %define xgft3_lo  xmm10
 144 %define xgft3_hi  xmm9
 145
 146 %define x0      xmm0
 147 %define xtmpa   xmm1
 148 %define xtmph1  xmm2
 149 %define xtmpl1  xmm3
 150 %define xtmph2  xmm4
 151 %define xtmpl2  xmm5
 152 %define xtmph3  xmm6
 153 %define xtmpl3  xmm7
 154 %define xd1     xmm8
 155 %define xd2     xtmpl1
 156 %define xd3     xtmph1
 157
 158 align 16
 159 global gf_3vect_mad_sse:ISAL_SYM_TYPE_FUNCTION
 160 func(gf_3vect_mad_sse)
 161         FUNC_SAVE
 162         sub     len, 16
 163         jl      .return_fail
 164         xor     pos, pos
 165         movdqa  xmask0f, [mask0f]       ;Load mask of lower nibble in each byte
 166         sal     vec_i, 5                ;Multiply by 32
 167         sal     vec, 5
 168         lea     tmp, [mul_array + vec_i]
 169
 170         movdqu  xgft1_lo, [tmp]         ;Load array Ax{00}, Ax{01}, Ax{02}, ...
 171         movdqu  xgft1_hi, [tmp+16]      ; " Ax{00}, Ax{10}, Ax{20}, ... , Ax{f0}
 172         movdqu  xgft2_lo, [tmp+vec]     ;Load array Bx{00}, Bx{01}, Bx{02}, ...
 173         movdqu  xgft2_hi, [tmp+vec+16]  ; " Bx{00}, Bx{10}, Bx{20}, ... , Bx{f0}
 174         movdqu  xgft3_lo, [tmp+2*vec]   ;Load array Cx{00}, Cx{01}, Cx{02}, ...
 175         movdqu  xgft3_hi, [tmp+2*vec+16]        ; " Cx{00}, Cx{10}, Cx{20}, ... , Cx{f0}
 176         mov     dest2, [dest1+PS]       ; reuse mul_array
 177         mov     dest3, [dest1+2*PS]     ; reuse vec_i
 178         mov     dest1, [dest1]
 179
 180 .loop16:
 181         XLDR    x0, [src+pos]           ;Get next source vector
 182         movdqa  xtmph1, xgft1_hi        ;Reload const array registers
 183         movdqa  xtmpl1, xgft1_lo
 184         movdqa  xtmph2, xgft2_hi        ;Reload const array registers
 185         movdqa  xtmpl2, xgft2_lo
 186         movdqa  xtmph3, xgft3_hi        ;Reload const array registers
 187         movdqa  xtmpl3, xgft3_lo
 188
 189         XLDR    xd1, [dest1+pos]        ;Get next dest vector
 190
 191         movdqa  xtmpa, x0               ;Keep unshifted copy of src
 192         psraw   x0, 4                   ;Shift to put high nibble into bits 4-0
 193         pand    x0, xmask0f             ;Mask high src nibble in bits 4-0
 194         pand    xtmpa, xmask0f          ;Mask low src nibble in bits 4-0
 195
 196         ; dest1
 197         pshufb  xtmph1, x0              ;Lookup mul table of high nibble
 198         pshufb  xtmpl1, xtmpa           ;Lookup mul table of low nibble
 199         pxor    xtmph1, xtmpl1          ;GF add high and low partials
 200         pxor    xd1, xtmph1
 201
 202         XLDR    xd2, [dest2+pos]        ;reuse xtmpl1. Get next dest vector
 203         XLDR    xd3, [dest3+pos]        ;reuse xtmph1. Get next dest vector
 204
 205         ; dest2
 206         pshufb  xtmph2, x0              ;Lookup mul table of high nibble
 207         pshufb  xtmpl2, xtmpa           ;Lookup mul table of low nibble
 208         pxor    xtmph2, xtmpl2          ;GF add high and low partials
 209         pxor    xd2, xtmph2
 210
 211         ; dest3
 212         pshufb  xtmph3, x0              ;Lookup mul table of high nibble
 213         pshufb  xtmpl3, xtmpa           ;Lookup mul table of low nibble
 214         pxor    xtmph3, xtmpl3          ;GF add high and low partials
 215         pxor    xd3, xtmph3
 216
 217         XSTR    [dest1+pos], xd1        ;Store result
 218         XSTR    [dest2+pos], xd2        ;Store result
 219         XSTR    [dest3+pos], xd3        ;Store result
 220
 221         add     pos, 16                 ;Loop on 16 bytes at a time
 222         cmp     pos, len
 223         jle     .loop16
 224
 225         lea     tmp, [len + 16]
 226         cmp     pos, tmp
 227         je      .return_pass
 228
 229 .lessthan16:
 230         ;; Tail len
 231         ;; Do one more overlap pass
 232         mov     tmp, len                ;Overlapped offset length-16
 233
 234         XLDR    x0, [src+tmp]           ;Get next source vector
 235         XLDR    xd1, [dest1+tmp]        ;Get next dest vector
 236         XLDR    xd2, [dest2+tmp]        ;reuse xtmpl1. Get next dest vector
 237         XLDR    xd3, [dest3+tmp]        ;reuse xtmph1. Get next dest vector
 238
 239         sub     len, pos
 240
 241         movdqa  xtmph3, [constip16]     ;Load const of i + 16
 242         pinsrb  xtmpl3, len.w, 15
 243         pshufb  xtmpl3, xmask0f         ;Broadcast len to all bytes
 244         pcmpgtb xtmpl3, xtmph3
 245
 246         movdqa  xtmpa, x0               ;Keep unshifted copy of src
 247         psraw   x0, 4                   ;Shift to put high nibble into bits 4-0
 248         pand    x0, xmask0f             ;Mask high src nibble in bits 4-0
 249         pand    xtmpa, xmask0f          ;Mask low src nibble in bits 4-0
 250
 251         ; dest1
 252         pshufb  xgft1_hi, x0            ;Lookup mul table of high nibble
 253         pshufb  xgft1_lo, xtmpa         ;Lookup mul table of low nibble
 254         pxor    xgft1_hi, xgft1_lo      ;GF add high and low partials
 255         pand    xgft1_hi, xtmpl3
 256         pxor    xd1, xgft1_hi
 257
 258         ; dest2
 259         pshufb  xgft2_hi, x0            ;Lookup mul table of high nibble
 260         pshufb  xgft2_lo, xtmpa         ;Lookup mul table of low nibble
 261         pxor    xgft2_hi, xgft2_lo      ;GF add high and low partials
 262         pand    xgft2_hi, xtmpl3
 263         pxor    xd2, xgft2_hi
 264
 265         ; dest3
 266         pshufb  xgft3_hi, x0            ;Lookup mul table of high nibble
 267         pshufb  xgft3_lo, xtmpa         ;Lookup mul table of low nibble
 268         pxor    xgft3_hi, xgft3_lo      ;GF add high and low partials
 269         pand    xgft3_hi, xtmpl3
 270         pxor    xd3, xgft3_hi
 271
 272         XSTR    [dest1+tmp], xd1        ;Store result
 273         XSTR    [dest2+tmp], xd2        ;Store result
 274         XSTR    [dest3+tmp], xd3        ;Store result
 275
 276 .return_pass:
 277         FUNC_RESTORE
 278         mov     return, 0
 279         ret
 280
 281 .return_fail:
 282         FUNC_RESTORE
 283         mov     return, 1
 284         ret
 285
 286 endproc_frame
 287
 288 section .data
 289
 290 align 16
 291
 292 mask0f:
 293         dq 0x0f0f0f0f0f0f0f0f, 0x0f0f0f0f0f0f0f0f
 294 constip16:
 295         dq 0xf8f9fafbfcfdfeff, 0xf0f1f2f3f4f5f6f7
 296
 297 ;;;       func             core, ver, snum
 298 slversion gf_3vect_mad_sse, 00,  01,  0206