1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/common/x86/postproc_sse2.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,721 @@
1.4 +;
1.5 +; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
1.6 +;
1.7 +; Use of this source code is governed by a BSD-style license
1.8 +; that can be found in the LICENSE file in the root of the source
1.9 +; tree. An additional intellectual property rights grant can be found
1.10 +; in the file PATENTS. All contributing project authors may
1.11 +; be found in the AUTHORS file in the root of the source tree.
1.12 +;
1.13 +
1.14 +
1.15 +%include "vpx_ports/x86_abi_support.asm"
1.16 +
1.17 +;macro in deblock functions
1.18 +%macro FIRST_2_ROWS 0
1.19 + movdqa xmm4, xmm0
1.20 + movdqa xmm6, xmm0
1.21 + movdqa xmm5, xmm1
1.22 + pavgb xmm5, xmm3
1.23 +
1.24 + ;calculate absolute value
1.25 + psubusb xmm4, xmm1
1.26 + psubusb xmm1, xmm0
1.27 + psubusb xmm6, xmm3
1.28 + psubusb xmm3, xmm0
1.29 + paddusb xmm4, xmm1
1.30 + paddusb xmm6, xmm3
1.31 +
1.32 + ;get threshold
1.33 + movdqa xmm2, flimit
1.34 + pxor xmm1, xmm1
1.35 + movdqa xmm7, xmm2
1.36 +
1.37 + ;get mask
1.38 + psubusb xmm2, xmm4
1.39 + psubusb xmm7, xmm6
1.40 + pcmpeqb xmm2, xmm1
1.41 + pcmpeqb xmm7, xmm1
1.42 + por xmm7, xmm2
1.43 +%endmacro
1.44 +
1.45 +%macro SECOND_2_ROWS 0
1.46 + movdqa xmm6, xmm0
1.47 + movdqa xmm4, xmm0
1.48 + movdqa xmm2, xmm1
1.49 + pavgb xmm1, xmm3
1.50 +
1.51 + ;calculate absolute value
1.52 + psubusb xmm6, xmm2
1.53 + psubusb xmm2, xmm0
1.54 + psubusb xmm4, xmm3
1.55 + psubusb xmm3, xmm0
1.56 + paddusb xmm6, xmm2
1.57 + paddusb xmm4, xmm3
1.58 +
1.59 + pavgb xmm5, xmm1
1.60 +
1.61 + ;get threshold
1.62 + movdqa xmm2, flimit
1.63 + pxor xmm1, xmm1
1.64 + movdqa xmm3, xmm2
1.65 +
1.66 + ;get mask
1.67 + psubusb xmm2, xmm6
1.68 + psubusb xmm3, xmm4
1.69 + pcmpeqb xmm2, xmm1
1.70 + pcmpeqb xmm3, xmm1
1.71 +
1.72 + por xmm7, xmm2
1.73 + por xmm7, xmm3
1.74 +
1.75 + pavgb xmm5, xmm0
1.76 +
1.77 + ;decide if or not to use filtered value
1.78 + pand xmm0, xmm7
1.79 + pandn xmm7, xmm5
1.80 + paddusb xmm0, xmm7
1.81 +%endmacro
1.82 +
1.83 +%macro UPDATE_FLIMIT 0
1.84 + movdqa xmm2, XMMWORD PTR [rbx]
1.85 + movdqa [rsp], xmm2
1.86 + add rbx, 16
1.87 +%endmacro
1.88 +
1.89 +;void vp8_post_proc_down_and_across_mb_row_sse2
1.90 +;(
1.91 +; unsigned char *src_ptr,
1.92 +; unsigned char *dst_ptr,
1.93 +; int src_pixels_per_line,
1.94 +; int dst_pixels_per_line,
1.95 +; int cols,
1.96 +; int *flimits,
1.97 +; int size
1.98 +;)
1.99 +global sym(vp8_post_proc_down_and_across_mb_row_sse2) PRIVATE
1.100 +sym(vp8_post_proc_down_and_across_mb_row_sse2):
1.101 + push rbp
1.102 + mov rbp, rsp
1.103 + SHADOW_ARGS_TO_STACK 7
1.104 + SAVE_XMM 7
1.105 + push rbx
1.106 + push rsi
1.107 + push rdi
1.108 + ; end prolog
1.109 + ALIGN_STACK 16, rax
1.110 + sub rsp, 16
1.111 +
1.112 + ; put flimit on stack
1.113 + mov rbx, arg(5) ;flimits ptr
1.114 + UPDATE_FLIMIT
1.115 +
1.116 +%define flimit [rsp]
1.117 +
1.118 + mov rsi, arg(0) ;src_ptr
1.119 + mov rdi, arg(1) ;dst_ptr
1.120 +
1.121 + movsxd rax, DWORD PTR arg(2) ;src_pixels_per_line
1.122 + movsxd rcx, DWORD PTR arg(6) ;rows in a macroblock
1.123 +.nextrow:
1.124 + xor rdx, rdx ;col
1.125 +.nextcol:
1.126 + ;load current and next 2 rows
1.127 + movdqu xmm0, XMMWORD PTR [rsi]
1.128 + movdqu xmm1, XMMWORD PTR [rsi + rax]
1.129 + movdqu xmm3, XMMWORD PTR [rsi + 2*rax]
1.130 +
1.131 + FIRST_2_ROWS
1.132 +
1.133 + ;load above 2 rows
1.134 + neg rax
1.135 + movdqu xmm1, XMMWORD PTR [rsi + 2*rax]
1.136 + movdqu xmm3, XMMWORD PTR [rsi + rax]
1.137 +
1.138 + SECOND_2_ROWS
1.139 +
1.140 + movdqu XMMWORD PTR [rdi], xmm0
1.141 +
1.142 + neg rax ; positive stride
1.143 + add rsi, 16
1.144 + add rdi, 16
1.145 +
1.146 + add rdx, 16
1.147 + cmp edx, dword arg(4) ;cols
1.148 + jge .downdone
1.149 + UPDATE_FLIMIT
1.150 + jmp .nextcol
1.151 +
1.152 +.downdone:
1.153 + ; done with the all cols, start the across filtering in place
1.154 + sub rsi, rdx
1.155 + sub rdi, rdx
1.156 +
1.157 + mov rbx, arg(5) ; flimits
1.158 + UPDATE_FLIMIT
1.159 +
1.160 + ; dup the first byte into the left border 8 times
1.161 + movq mm1, [rdi]
1.162 + punpcklbw mm1, mm1
1.163 + punpcklwd mm1, mm1
1.164 + punpckldq mm1, mm1
1.165 + mov rdx, -8
1.166 + movq [rdi+rdx], mm1
1.167 +
1.168 + ; dup the last byte into the right border
1.169 + movsxd rdx, dword arg(4)
1.170 + movq mm1, [rdi + rdx + -1]
1.171 + punpcklbw mm1, mm1
1.172 + punpcklwd mm1, mm1
1.173 + punpckldq mm1, mm1
1.174 + movq [rdi+rdx], mm1
1.175 +
1.176 + xor rdx, rdx
1.177 + movq mm0, QWORD PTR [rdi-16];
1.178 + movq mm1, QWORD PTR [rdi-8];
1.179 +
1.180 +.acrossnextcol:
1.181 + movdqu xmm0, XMMWORD PTR [rdi + rdx]
1.182 + movdqu xmm1, XMMWORD PTR [rdi + rdx -2]
1.183 + movdqu xmm3, XMMWORD PTR [rdi + rdx -1]
1.184 +
1.185 + FIRST_2_ROWS
1.186 +
1.187 + movdqu xmm1, XMMWORD PTR [rdi + rdx +1]
1.188 + movdqu xmm3, XMMWORD PTR [rdi + rdx +2]
1.189 +
1.190 + SECOND_2_ROWS
1.191 +
1.192 + movq QWORD PTR [rdi+rdx-16], mm0 ; store previous 8 bytes
1.193 + movq QWORD PTR [rdi+rdx-8], mm1 ; store previous 8 bytes
1.194 + movdq2q mm0, xmm0
1.195 + psrldq xmm0, 8
1.196 + movdq2q mm1, xmm0
1.197 +
1.198 + add rdx, 16
1.199 + cmp edx, dword arg(4) ;cols
1.200 + jge .acrossdone
1.201 + UPDATE_FLIMIT
1.202 + jmp .acrossnextcol
1.203 +
1.204 +.acrossdone
1.205 + ; last 16 pixels
1.206 + movq QWORD PTR [rdi+rdx-16], mm0
1.207 +
1.208 + cmp edx, dword arg(4)
1.209 + jne .throw_last_8
1.210 + movq QWORD PTR [rdi+rdx-8], mm1
1.211 +.throw_last_8:
1.212 + ; done with this rwo
1.213 + add rsi,rax ;next src line
1.214 + mov eax, dword arg(3) ;dst_pixels_per_line
1.215 + add rdi,rax ;next destination
1.216 + mov eax, dword arg(2) ;src_pixels_per_line
1.217 +
1.218 + mov rbx, arg(5) ;flimits
1.219 + UPDATE_FLIMIT
1.220 +
1.221 + dec rcx ;decrement count
1.222 + jnz .nextrow ;next row
1.223 +
1.224 + add rsp, 16
1.225 + pop rsp
1.226 + ; begin epilog
1.227 + pop rdi
1.228 + pop rsi
1.229 + pop rbx
1.230 + RESTORE_XMM
1.231 + UNSHADOW_ARGS
1.232 + pop rbp
1.233 + ret
1.234 +%undef flimit
1.235 +
1.236 +;void vp8_mbpost_proc_down_xmm(unsigned char *dst,
1.237 +; int pitch, int rows, int cols,int flimit)
1.238 +extern sym(vp8_rv)
1.239 +global sym(vp8_mbpost_proc_down_xmm) PRIVATE
1.240 +sym(vp8_mbpost_proc_down_xmm):
1.241 + push rbp
1.242 + mov rbp, rsp
1.243 + SHADOW_ARGS_TO_STACK 5
1.244 + SAVE_XMM 7
1.245 + GET_GOT rbx
1.246 + push rsi
1.247 + push rdi
1.248 + ; end prolog
1.249 +
1.250 + ALIGN_STACK 16, rax
1.251 + sub rsp, 128+16
1.252 +
1.253 + ; unsigned char d[16][8] at [rsp]
1.254 + ; create flimit2 at [rsp+128]
1.255 + mov eax, dword ptr arg(4) ;flimit
1.256 + mov [rsp+128], eax
1.257 + mov [rsp+128+4], eax
1.258 + mov [rsp+128+8], eax
1.259 + mov [rsp+128+12], eax
1.260 +%define flimit4 [rsp+128]
1.261 +
1.262 +%if ABI_IS_32BIT=0
1.263 + lea r8, [GLOBAL(sym(vp8_rv))]
1.264 +%endif
1.265 +
1.266 + ;rows +=8;
1.267 + add dword arg(2), 8
1.268 +
1.269 + ;for(c=0; c<cols; c+=8)
1.270 +.loop_col:
1.271 + mov rsi, arg(0) ; s
1.272 + pxor xmm0, xmm0 ;
1.273 +
1.274 + movsxd rax, dword ptr arg(1) ;pitch ;
1.275 +
1.276 + ; this copies the last row down into the border 8 rows
1.277 + mov rdi, rsi
1.278 + mov rdx, arg(2)
1.279 + sub rdx, 9
1.280 + imul rdx, rax
1.281 + lea rdi, [rdi+rdx]
1.282 + movq xmm1, QWORD ptr[rdi] ; first row
1.283 + mov rcx, 8
1.284 +.init_borderd ; initialize borders
1.285 + lea rdi, [rdi + rax]
1.286 + movq [rdi], xmm1
1.287 +
1.288 + dec rcx
1.289 + jne .init_borderd
1.290 +
1.291 + neg rax ; rax = -pitch
1.292 +
1.293 + ; this copies the first row up into the border 8 rows
1.294 + mov rdi, rsi
1.295 + movq xmm1, QWORD ptr[rdi] ; first row
1.296 + mov rcx, 8
1.297 +.init_border ; initialize borders
1.298 + lea rdi, [rdi + rax]
1.299 + movq [rdi], xmm1
1.300 +
1.301 + dec rcx
1.302 + jne .init_border
1.303 +
1.304 +
1.305 +
1.306 + lea rsi, [rsi + rax*8]; ; rdi = s[-pitch*8]
1.307 + neg rax
1.308 +
1.309 + pxor xmm5, xmm5
1.310 + pxor xmm6, xmm6 ;
1.311 +
1.312 + pxor xmm7, xmm7 ;
1.313 + mov rdi, rsi
1.314 +
1.315 + mov rcx, 15 ;
1.316 +
1.317 +.loop_initvar:
1.318 + movq xmm1, QWORD PTR [rdi];
1.319 + punpcklbw xmm1, xmm0 ;
1.320 +
1.321 + paddw xmm5, xmm1 ;
1.322 + pmullw xmm1, xmm1 ;
1.323 +
1.324 + movdqa xmm2, xmm1 ;
1.325 + punpcklwd xmm1, xmm0 ;
1.326 +
1.327 + punpckhwd xmm2, xmm0 ;
1.328 + paddd xmm6, xmm1 ;
1.329 +
1.330 + paddd xmm7, xmm2 ;
1.331 + lea rdi, [rdi+rax] ;
1.332 +
1.333 + dec rcx
1.334 + jne .loop_initvar
1.335 + ;save the var and sum
1.336 + xor rdx, rdx
1.337 +.loop_row:
1.338 + movq xmm1, QWORD PTR [rsi] ; [s-pitch*8]
1.339 + movq xmm2, QWORD PTR [rdi] ; [s+pitch*7]
1.340 +
1.341 + punpcklbw xmm1, xmm0
1.342 + punpcklbw xmm2, xmm0
1.343 +
1.344 + paddw xmm5, xmm2
1.345 + psubw xmm5, xmm1
1.346 +
1.347 + pmullw xmm2, xmm2
1.348 + movdqa xmm4, xmm2
1.349 +
1.350 + punpcklwd xmm2, xmm0
1.351 + punpckhwd xmm4, xmm0
1.352 +
1.353 + paddd xmm6, xmm2
1.354 + paddd xmm7, xmm4
1.355 +
1.356 + pmullw xmm1, xmm1
1.357 + movdqa xmm2, xmm1
1.358 +
1.359 + punpcklwd xmm1, xmm0
1.360 + psubd xmm6, xmm1
1.361 +
1.362 + punpckhwd xmm2, xmm0
1.363 + psubd xmm7, xmm2
1.364 +
1.365 +
1.366 + movdqa xmm3, xmm6
1.367 + pslld xmm3, 4
1.368 +
1.369 + psubd xmm3, xmm6
1.370 + movdqa xmm1, xmm5
1.371 +
1.372 + movdqa xmm4, xmm5
1.373 + pmullw xmm1, xmm1
1.374 +
1.375 + pmulhw xmm4, xmm4
1.376 + movdqa xmm2, xmm1
1.377 +
1.378 + punpcklwd xmm1, xmm4
1.379 + punpckhwd xmm2, xmm4
1.380 +
1.381 + movdqa xmm4, xmm7
1.382 + pslld xmm4, 4
1.383 +
1.384 + psubd xmm4, xmm7
1.385 +
1.386 + psubd xmm3, xmm1
1.387 + psubd xmm4, xmm2
1.388 +
1.389 + psubd xmm3, flimit4
1.390 + psubd xmm4, flimit4
1.391 +
1.392 + psrad xmm3, 31
1.393 + psrad xmm4, 31
1.394 +
1.395 + packssdw xmm3, xmm4
1.396 + packsswb xmm3, xmm0
1.397 +
1.398 + movq xmm1, QWORD PTR [rsi+rax*8]
1.399 +
1.400 + movq xmm2, xmm1
1.401 + punpcklbw xmm1, xmm0
1.402 +
1.403 + paddw xmm1, xmm5
1.404 + mov rcx, rdx
1.405 +
1.406 + and rcx, 127
1.407 +%if ABI_IS_32BIT=1 && CONFIG_PIC=1
1.408 + push rax
1.409 + lea rax, [GLOBAL(sym(vp8_rv))]
1.410 + movdqu xmm4, [rax + rcx*2] ;vp8_rv[rcx*2]
1.411 + pop rax
1.412 +%elif ABI_IS_32BIT=0
1.413 + movdqu xmm4, [r8 + rcx*2] ;vp8_rv[rcx*2]
1.414 +%else
1.415 + movdqu xmm4, [sym(vp8_rv) + rcx*2]
1.416 +%endif
1.417 +
1.418 + paddw xmm1, xmm4
1.419 + ;paddw xmm1, eight8s
1.420 + psraw xmm1, 4
1.421 +
1.422 + packuswb xmm1, xmm0
1.423 + pand xmm1, xmm3
1.424 +
1.425 + pandn xmm3, xmm2
1.426 + por xmm1, xmm3
1.427 +
1.428 + and rcx, 15
1.429 + movq QWORD PTR [rsp + rcx*8], xmm1 ;d[rcx*8]
1.430 +
1.431 + mov rcx, rdx
1.432 + sub rcx, 8
1.433 +
1.434 + and rcx, 15
1.435 + movq mm0, [rsp + rcx*8] ;d[rcx*8]
1.436 +
1.437 + movq [rsi], mm0
1.438 + lea rsi, [rsi+rax]
1.439 +
1.440 + lea rdi, [rdi+rax]
1.441 + add rdx, 1
1.442 +
1.443 + cmp edx, dword arg(2) ;rows
1.444 + jl .loop_row
1.445 +
1.446 + add dword arg(0), 8 ; s += 8
1.447 + sub dword arg(3), 8 ; cols -= 8
1.448 + cmp dword arg(3), 0
1.449 + jg .loop_col
1.450 +
1.451 + add rsp, 128+16
1.452 + pop rsp
1.453 +
1.454 + ; begin epilog
1.455 + pop rdi
1.456 + pop rsi
1.457 + RESTORE_GOT
1.458 + RESTORE_XMM
1.459 + UNSHADOW_ARGS
1.460 + pop rbp
1.461 + ret
1.462 +%undef flimit4
1.463 +
1.464 +
1.465 +;void vp8_mbpost_proc_across_ip_xmm(unsigned char *src,
1.466 +; int pitch, int rows, int cols,int flimit)
1.467 +global sym(vp8_mbpost_proc_across_ip_xmm) PRIVATE
1.468 +sym(vp8_mbpost_proc_across_ip_xmm):
1.469 + push rbp
1.470 + mov rbp, rsp
1.471 + SHADOW_ARGS_TO_STACK 5
1.472 + SAVE_XMM 7
1.473 + GET_GOT rbx
1.474 + push rsi
1.475 + push rdi
1.476 + ; end prolog
1.477 +
1.478 + ALIGN_STACK 16, rax
1.479 + sub rsp, 16
1.480 +
1.481 + ; create flimit4 at [rsp]
1.482 + mov eax, dword ptr arg(4) ;flimit
1.483 + mov [rsp], eax
1.484 + mov [rsp+4], eax
1.485 + mov [rsp+8], eax
1.486 + mov [rsp+12], eax
1.487 +%define flimit4 [rsp]
1.488 +
1.489 +
1.490 + ;for(r=0;r<rows;r++)
1.491 +.ip_row_loop:
1.492 +
1.493 + xor rdx, rdx ;sumsq=0;
1.494 + xor rcx, rcx ;sum=0;
1.495 + mov rsi, arg(0); s
1.496 +
1.497 +
1.498 + ; dup the first byte into the left border 8 times
1.499 + movq mm1, [rsi]
1.500 + punpcklbw mm1, mm1
1.501 + punpcklwd mm1, mm1
1.502 + punpckldq mm1, mm1
1.503 +
1.504 + mov rdi, -8
1.505 + movq [rsi+rdi], mm1
1.506 +
1.507 + ; dup the last byte into the right border
1.508 + movsxd rdx, dword arg(3)
1.509 + movq mm1, [rsi + rdx + -1]
1.510 + punpcklbw mm1, mm1
1.511 + punpcklwd mm1, mm1
1.512 + punpckldq mm1, mm1
1.513 + movq [rsi+rdx], mm1
1.514 +
1.515 +.ip_var_loop:
1.516 + ;for(i=-8;i<=6;i++)
1.517 + ;{
1.518 + ; sumsq += s[i]*s[i];
1.519 + ; sum += s[i];
1.520 + ;}
1.521 + movzx eax, byte [rsi+rdi]
1.522 + add ecx, eax
1.523 + mul al
1.524 + add edx, eax
1.525 + add rdi, 1
1.526 + cmp rdi, 6
1.527 + jle .ip_var_loop
1.528 +
1.529 +
1.530 + ;mov rax, sumsq
1.531 + ;movd xmm7, rax
1.532 + movd xmm7, edx
1.533 +
1.534 + ;mov rax, sum
1.535 + ;movd xmm6, rax
1.536 + movd xmm6, ecx
1.537 +
1.538 + mov rsi, arg(0) ;s
1.539 + xor rcx, rcx
1.540 +
1.541 + movsxd rdx, dword arg(3) ;cols
1.542 + add rdx, 8
1.543 + pxor mm0, mm0
1.544 + pxor mm1, mm1
1.545 +
1.546 + pxor xmm0, xmm0
1.547 +.nextcol4:
1.548 +
1.549 + movd xmm1, DWORD PTR [rsi+rcx-8] ; -8 -7 -6 -5
1.550 + movd xmm2, DWORD PTR [rsi+rcx+7] ; +7 +8 +9 +10
1.551 +
1.552 + punpcklbw xmm1, xmm0 ; expanding
1.553 + punpcklbw xmm2, xmm0 ; expanding
1.554 +
1.555 + punpcklwd xmm1, xmm0 ; expanding to dwords
1.556 + punpcklwd xmm2, xmm0 ; expanding to dwords
1.557 +
1.558 + psubd xmm2, xmm1 ; 7--8 8--7 9--6 10--5
1.559 + paddd xmm1, xmm1 ; -8*2 -7*2 -6*2 -5*2
1.560 +
1.561 + paddd xmm1, xmm2 ; 7+-8 8+-7 9+-6 10+-5
1.562 + pmaddwd xmm1, xmm2 ; squared of 7+-8 8+-7 9+-6 10+-5
1.563 +
1.564 + paddd xmm6, xmm2
1.565 + paddd xmm7, xmm1
1.566 +
1.567 + pshufd xmm6, xmm6, 0 ; duplicate the last ones
1.568 + pshufd xmm7, xmm7, 0 ; duplicate the last ones
1.569 +
1.570 + psrldq xmm1, 4 ; 8--7 9--6 10--5 0000
1.571 + psrldq xmm2, 4 ; 8--7 9--6 10--5 0000
1.572 +
1.573 + pshufd xmm3, xmm1, 3 ; 0000 8--7 8--7 8--7 squared
1.574 + pshufd xmm4, xmm2, 3 ; 0000 8--7 8--7 8--7 squared
1.575 +
1.576 + paddd xmm6, xmm4
1.577 + paddd xmm7, xmm3
1.578 +
1.579 + pshufd xmm3, xmm1, 01011111b ; 0000 0000 9--6 9--6 squared
1.580 + pshufd xmm4, xmm2, 01011111b ; 0000 0000 9--6 9--6 squared
1.581 +
1.582 + paddd xmm7, xmm3
1.583 + paddd xmm6, xmm4
1.584 +
1.585 + pshufd xmm3, xmm1, 10111111b ; 0000 0000 8--7 8--7 squared
1.586 + pshufd xmm4, xmm2, 10111111b ; 0000 0000 8--7 8--7 squared
1.587 +
1.588 + paddd xmm7, xmm3
1.589 + paddd xmm6, xmm4
1.590 +
1.591 + movdqa xmm3, xmm6
1.592 + pmaddwd xmm3, xmm3
1.593 +
1.594 + movdqa xmm5, xmm7
1.595 + pslld xmm5, 4
1.596 +
1.597 + psubd xmm5, xmm7
1.598 + psubd xmm5, xmm3
1.599 +
1.600 + psubd xmm5, flimit4
1.601 + psrad xmm5, 31
1.602 +
1.603 + packssdw xmm5, xmm0
1.604 + packsswb xmm5, xmm0
1.605 +
1.606 + movd xmm1, DWORD PTR [rsi+rcx]
1.607 + movq xmm2, xmm1
1.608 +
1.609 + punpcklbw xmm1, xmm0
1.610 + punpcklwd xmm1, xmm0
1.611 +
1.612 + paddd xmm1, xmm6
1.613 + paddd xmm1, [GLOBAL(four8s)]
1.614 +
1.615 + psrad xmm1, 4
1.616 + packssdw xmm1, xmm0
1.617 +
1.618 + packuswb xmm1, xmm0
1.619 + pand xmm1, xmm5
1.620 +
1.621 + pandn xmm5, xmm2
1.622 + por xmm5, xmm1
1.623 +
1.624 + movd [rsi+rcx-8], mm0
1.625 + movq mm0, mm1
1.626 +
1.627 + movdq2q mm1, xmm5
1.628 + psrldq xmm7, 12
1.629 +
1.630 + psrldq xmm6, 12
1.631 + add rcx, 4
1.632 +
1.633 + cmp rcx, rdx
1.634 + jl .nextcol4
1.635 +
1.636 + ;s+=pitch;
1.637 + movsxd rax, dword arg(1)
1.638 + add arg(0), rax
1.639 +
1.640 + sub dword arg(2), 1 ;rows-=1
1.641 + cmp dword arg(2), 0
1.642 + jg .ip_row_loop
1.643 +
1.644 + add rsp, 16
1.645 + pop rsp
1.646 +
1.647 + ; begin epilog
1.648 + pop rdi
1.649 + pop rsi
1.650 + RESTORE_GOT
1.651 + RESTORE_XMM
1.652 + UNSHADOW_ARGS
1.653 + pop rbp
1.654 + ret
1.655 +%undef flimit4
1.656 +
1.657 +
1.658 +;void vp8_plane_add_noise_wmt (unsigned char *Start, unsigned char *noise,
1.659 +; unsigned char blackclamp[16],
1.660 +; unsigned char whiteclamp[16],
1.661 +; unsigned char bothclamp[16],
1.662 +; unsigned int Width, unsigned int Height, int Pitch)
1.663 +extern sym(rand)
1.664 +global sym(vp8_plane_add_noise_wmt) PRIVATE
1.665 +sym(vp8_plane_add_noise_wmt):
1.666 + push rbp
1.667 + mov rbp, rsp
1.668 + SHADOW_ARGS_TO_STACK 8
1.669 + GET_GOT rbx
1.670 + push rsi
1.671 + push rdi
1.672 + ; end prolog
1.673 +
1.674 +.addnoise_loop:
1.675 + call sym(rand) WRT_PLT
1.676 + mov rcx, arg(1) ;noise
1.677 + and rax, 0xff
1.678 + add rcx, rax
1.679 +
1.680 + ; we rely on the fact that the clamping vectors are stored contiguously
1.681 + ; in black/white/both order. Note that we have to reload this here because
1.682 + ; rdx could be trashed by rand()
1.683 + mov rdx, arg(2) ; blackclamp
1.684 +
1.685 +
1.686 + mov rdi, rcx
1.687 + movsxd rcx, dword arg(5) ;[Width]
1.688 + mov rsi, arg(0) ;Pos
1.689 + xor rax,rax
1.690 +
1.691 +.addnoise_nextset:
1.692 + movdqu xmm1,[rsi+rax] ; get the source
1.693 +
1.694 + psubusb xmm1, [rdx] ;blackclamp ; clamp both sides so we don't outrange adding noise
1.695 + paddusb xmm1, [rdx+32] ;bothclamp
1.696 + psubusb xmm1, [rdx+16] ;whiteclamp
1.697 +
1.698 + movdqu xmm2,[rdi+rax] ; get the noise for this line
1.699 + paddb xmm1,xmm2 ; add it in
1.700 + movdqu [rsi+rax],xmm1 ; store the result
1.701 +
1.702 + add rax,16 ; move to the next line
1.703 +
1.704 + cmp rax, rcx
1.705 + jl .addnoise_nextset
1.706 +
1.707 + movsxd rax, dword arg(7) ; Pitch
1.708 + add arg(0), rax ; Start += Pitch
1.709 + sub dword arg(6), 1 ; Height -= 1
1.710 + jg .addnoise_loop
1.711 +
1.712 + ; begin epilog
1.713 + pop rdi
1.714 + pop rsi
1.715 + RESTORE_GOT
1.716 + UNSHADOW_ARGS
1.717 + pop rbp
1.718 + ret
1.719 +
1.720 +
1.721 +SECTION_RODATA
1.722 +align 16
1.723 +four8s:
1.724 + times 4 dd 8
