1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/common/x86/subpixel_mmx.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,702 @@
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 +extern sym(vp8_bilinear_filters_x86_8)
1.17 +
1.18 +
1.19 +%define BLOCK_HEIGHT_WIDTH 4
1.20 +%define vp8_filter_weight 128
1.21 +%define VP8_FILTER_SHIFT 7
1.22 +
1.23 +
1.24 +;void vp8_filter_block1d_h6_mmx
1.25 +;(
1.26 +; unsigned char *src_ptr,
1.27 +; unsigned short *output_ptr,
1.28 +; unsigned int src_pixels_per_line,
1.29 +; unsigned int pixel_step,
1.30 +; unsigned int output_height,
1.31 +; unsigned int output_width,
1.32 +; short * vp8_filter
1.33 +;)
1.34 +global sym(vp8_filter_block1d_h6_mmx) PRIVATE
1.35 +sym(vp8_filter_block1d_h6_mmx):
1.36 + push rbp
1.37 + mov rbp, rsp
1.38 + SHADOW_ARGS_TO_STACK 7
1.39 + GET_GOT rbx
1.40 + push rsi
1.41 + push rdi
1.42 + ; end prolog
1.43 +
1.44 + mov rdx, arg(6) ;vp8_filter
1.45 +
1.46 + movq mm1, [rdx + 16] ; do both the negative taps first!!!
1.47 + movq mm2, [rdx + 32] ;
1.48 + movq mm6, [rdx + 48] ;
1.49 + movq mm7, [rdx + 64] ;
1.50 +
1.51 + mov rdi, arg(1) ;output_ptr
1.52 + mov rsi, arg(0) ;src_ptr
1.53 + movsxd rcx, dword ptr arg(4) ;output_height
1.54 + movsxd rax, dword ptr arg(5) ;output_width ; destination pitch?
1.55 + pxor mm0, mm0 ; mm0 = 00000000
1.56 +
1.57 +.nextrow:
1.58 + movq mm3, [rsi-2] ; mm3 = p-2..p5
1.59 + movq mm4, mm3 ; mm4 = p-2..p5
1.60 + psrlq mm3, 8 ; mm3 = p-1..p5
1.61 + punpcklbw mm3, mm0 ; mm3 = p-1..p2
1.62 + pmullw mm3, mm1 ; mm3 *= kernel 1 modifiers.
1.63 +
1.64 + movq mm5, mm4 ; mm5 = p-2..p5
1.65 + punpckhbw mm4, mm0 ; mm5 = p2..p5
1.66 + pmullw mm4, mm7 ; mm5 *= kernel 4 modifiers
1.67 + paddsw mm3, mm4 ; mm3 += mm5
1.68 +
1.69 + movq mm4, mm5 ; mm4 = p-2..p5;
1.70 + psrlq mm5, 16 ; mm5 = p0..p5;
1.71 + punpcklbw mm5, mm0 ; mm5 = p0..p3
1.72 + pmullw mm5, mm2 ; mm5 *= kernel 2 modifiers
1.73 + paddsw mm3, mm5 ; mm3 += mm5
1.74 +
1.75 + movq mm5, mm4 ; mm5 = p-2..p5
1.76 + psrlq mm4, 24 ; mm4 = p1..p5
1.77 + punpcklbw mm4, mm0 ; mm4 = p1..p4
1.78 + pmullw mm4, mm6 ; mm5 *= kernel 3 modifiers
1.79 + paddsw mm3, mm4 ; mm3 += mm5
1.80 +
1.81 + ; do outer positive taps
1.82 + movd mm4, [rsi+3]
1.83 + punpcklbw mm4, mm0 ; mm5 = p3..p6
1.84 + pmullw mm4, [rdx+80] ; mm5 *= kernel 0 modifiers
1.85 + paddsw mm3, mm4 ; mm3 += mm5
1.86 +
1.87 + punpcklbw mm5, mm0 ; mm5 = p-2..p1
1.88 + pmullw mm5, [rdx] ; mm5 *= kernel 5 modifiers
1.89 + paddsw mm3, mm5 ; mm3 += mm5
1.90 +
1.91 + paddsw mm3, [GLOBAL(rd)] ; mm3 += round value
1.92 + psraw mm3, VP8_FILTER_SHIFT ; mm3 /= 128
1.93 + packuswb mm3, mm0 ; pack and unpack to saturate
1.94 + punpcklbw mm3, mm0 ;
1.95 +
1.96 + movq [rdi], mm3 ; store the results in the destination
1.97 +
1.98 +%if ABI_IS_32BIT
1.99 + add rsi, dword ptr arg(2) ;src_pixels_per_line ; next line
1.100 + add rdi, rax;
1.101 +%else
1.102 + movsxd r8, dword ptr arg(2) ;src_pixels_per_line
1.103 + add rdi, rax;
1.104 +
1.105 + add rsi, r8 ; next line
1.106 +%endif
1.107 +
1.108 + dec rcx ; decrement count
1.109 + jnz .nextrow ; next row
1.110 +
1.111 + ; begin epilog
1.112 + pop rdi
1.113 + pop rsi
1.114 + RESTORE_GOT
1.115 + UNSHADOW_ARGS
1.116 + pop rbp
1.117 + ret
1.118 +
1.119 +
1.120 +;void vp8_filter_block1dc_v6_mmx
1.121 +;(
1.122 +; short *src_ptr,
1.123 +; unsigned char *output_ptr,
1.124 +; int output_pitch,
1.125 +; unsigned int pixels_per_line,
1.126 +; unsigned int pixel_step,
1.127 +; unsigned int output_height,
1.128 +; unsigned int output_width,
1.129 +; short * vp8_filter
1.130 +;)
1.131 +global sym(vp8_filter_block1dc_v6_mmx) PRIVATE
1.132 +sym(vp8_filter_block1dc_v6_mmx):
1.133 + push rbp
1.134 + mov rbp, rsp
1.135 + SHADOW_ARGS_TO_STACK 8
1.136 + GET_GOT rbx
1.137 + push rsi
1.138 + push rdi
1.139 + ; end prolog
1.140 +
1.141 + movq mm5, [GLOBAL(rd)]
1.142 + push rbx
1.143 + mov rbx, arg(7) ;vp8_filter
1.144 + movq mm1, [rbx + 16] ; do both the negative taps first!!!
1.145 + movq mm2, [rbx + 32] ;
1.146 + movq mm6, [rbx + 48] ;
1.147 + movq mm7, [rbx + 64] ;
1.148 +
1.149 + movsxd rdx, dword ptr arg(3) ;pixels_per_line
1.150 + mov rdi, arg(1) ;output_ptr
1.151 + mov rsi, arg(0) ;src_ptr
1.152 + sub rsi, rdx
1.153 + sub rsi, rdx
1.154 + movsxd rcx, DWORD PTR arg(5) ;output_height
1.155 + movsxd rax, DWORD PTR arg(2) ;output_pitch ; destination pitch?
1.156 + pxor mm0, mm0 ; mm0 = 00000000
1.157 +
1.158 +
1.159 +.nextrow_cv:
1.160 + movq mm3, [rsi+rdx] ; mm3 = p0..p8 = row -1
1.161 + pmullw mm3, mm1 ; mm3 *= kernel 1 modifiers.
1.162 +
1.163 +
1.164 + movq mm4, [rsi + 4*rdx] ; mm4 = p0..p3 = row 2
1.165 + pmullw mm4, mm7 ; mm4 *= kernel 4 modifiers.
1.166 + paddsw mm3, mm4 ; mm3 += mm4
1.167 +
1.168 + movq mm4, [rsi + 2*rdx] ; mm4 = p0..p3 = row 0
1.169 + pmullw mm4, mm2 ; mm4 *= kernel 2 modifiers.
1.170 + paddsw mm3, mm4 ; mm3 += mm4
1.171 +
1.172 + movq mm4, [rsi] ; mm4 = p0..p3 = row -2
1.173 + pmullw mm4, [rbx] ; mm4 *= kernel 0 modifiers.
1.174 + paddsw mm3, mm4 ; mm3 += mm4
1.175 +
1.176 +
1.177 + add rsi, rdx ; move source forward 1 line to avoid 3 * pitch
1.178 + movq mm4, [rsi + 2*rdx] ; mm4 = p0..p3 = row 1
1.179 + pmullw mm4, mm6 ; mm4 *= kernel 3 modifiers.
1.180 + paddsw mm3, mm4 ; mm3 += mm4
1.181 +
1.182 + movq mm4, [rsi + 4*rdx] ; mm4 = p0..p3 = row 3
1.183 + pmullw mm4, [rbx +80] ; mm4 *= kernel 3 modifiers.
1.184 + paddsw mm3, mm4 ; mm3 += mm4
1.185 +
1.186 +
1.187 + paddsw mm3, mm5 ; mm3 += round value
1.188 + psraw mm3, VP8_FILTER_SHIFT ; mm3 /= 128
1.189 + packuswb mm3, mm0 ; pack and saturate
1.190 +
1.191 + movd [rdi],mm3 ; store the results in the destination
1.192 + ; the subsequent iterations repeat 3 out of 4 of these reads. Since the
1.193 + ; recon block should be in cache this shouldn't cost much. Its obviously
1.194 + ; avoidable!!!.
1.195 + lea rdi, [rdi+rax] ;
1.196 + dec rcx ; decrement count
1.197 + jnz .nextrow_cv ; next row
1.198 +
1.199 + pop rbx
1.200 +
1.201 + ; begin epilog
1.202 + pop rdi
1.203 + pop rsi
1.204 + RESTORE_GOT
1.205 + UNSHADOW_ARGS
1.206 + pop rbp
1.207 + ret
1.208 +
1.209 +
1.210 +;void bilinear_predict8x8_mmx
1.211 +;(
1.212 +; unsigned char *src_ptr,
1.213 +; int src_pixels_per_line,
1.214 +; int xoffset,
1.215 +; int yoffset,
1.216 +; unsigned char *dst_ptr,
1.217 +; int dst_pitch
1.218 +;)
1.219 +global sym(vp8_bilinear_predict8x8_mmx) PRIVATE
1.220 +sym(vp8_bilinear_predict8x8_mmx):
1.221 + push rbp
1.222 + mov rbp, rsp
1.223 + SHADOW_ARGS_TO_STACK 6
1.224 + GET_GOT rbx
1.225 + push rsi
1.226 + push rdi
1.227 + ; end prolog
1.228 +
1.229 + ;const short *HFilter = vp8_bilinear_filters_x86_8[xoffset];
1.230 + ;const short *VFilter = vp8_bilinear_filters_x86_8[yoffset];
1.231 +
1.232 + movsxd rax, dword ptr arg(2) ;xoffset
1.233 + mov rdi, arg(4) ;dst_ptr ;
1.234 +
1.235 + shl rax, 5 ; offset * 32
1.236 + lea rcx, [GLOBAL(sym(vp8_bilinear_filters_x86_8))]
1.237 +
1.238 + add rax, rcx ; HFilter
1.239 + mov rsi, arg(0) ;src_ptr ;
1.240 +
1.241 + movsxd rdx, dword ptr arg(5) ;dst_pitch
1.242 + movq mm1, [rax] ;
1.243 +
1.244 + movq mm2, [rax+16] ;
1.245 + movsxd rax, dword ptr arg(3) ;yoffset
1.246 +
1.247 + pxor mm0, mm0 ;
1.248 +
1.249 + shl rax, 5 ; offset*32
1.250 + add rax, rcx ; VFilter
1.251 +
1.252 + lea rcx, [rdi+rdx*8] ;
1.253 + movsxd rdx, dword ptr arg(1) ;src_pixels_per_line ;
1.254 +
1.255 +
1.256 +
1.257 + ; get the first horizontal line done ;
1.258 + movq mm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14
1.259 + movq mm4, mm3 ; make a copy of current line
1.260 +
1.261 + punpcklbw mm3, mm0 ; xx 00 01 02 03 04 05 06
1.262 + punpckhbw mm4, mm0 ;
1.263 +
1.264 + pmullw mm3, mm1 ;
1.265 + pmullw mm4, mm1 ;
1.266 +
1.267 + movq mm5, [rsi+1] ;
1.268 + movq mm6, mm5 ;
1.269 +
1.270 + punpcklbw mm5, mm0 ;
1.271 + punpckhbw mm6, mm0 ;
1.272 +
1.273 + pmullw mm5, mm2 ;
1.274 + pmullw mm6, mm2 ;
1.275 +
1.276 + paddw mm3, mm5 ;
1.277 + paddw mm4, mm6 ;
1.278 +
1.279 + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value
1.280 + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128
1.281 +
1.282 + paddw mm4, [GLOBAL(rd)] ;
1.283 + psraw mm4, VP8_FILTER_SHIFT ;
1.284 +
1.285 + movq mm7, mm3 ;
1.286 + packuswb mm7, mm4 ;
1.287 +
1.288 + add rsi, rdx ; next line
1.289 +.next_row_8x8:
1.290 + movq mm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14
1.291 + movq mm4, mm3 ; make a copy of current line
1.292 +
1.293 + punpcklbw mm3, mm0 ; xx 00 01 02 03 04 05 06
1.294 + punpckhbw mm4, mm0 ;
1.295 +
1.296 + pmullw mm3, mm1 ;
1.297 + pmullw mm4, mm1 ;
1.298 +
1.299 + movq mm5, [rsi+1] ;
1.300 + movq mm6, mm5 ;
1.301 +
1.302 + punpcklbw mm5, mm0 ;
1.303 + punpckhbw mm6, mm0 ;
1.304 +
1.305 + pmullw mm5, mm2 ;
1.306 + pmullw mm6, mm2 ;
1.307 +
1.308 + paddw mm3, mm5 ;
1.309 + paddw mm4, mm6 ;
1.310 +
1.311 + movq mm5, mm7 ;
1.312 + movq mm6, mm7 ;
1.313 +
1.314 + punpcklbw mm5, mm0 ;
1.315 + punpckhbw mm6, mm0
1.316 +
1.317 + pmullw mm5, [rax] ;
1.318 + pmullw mm6, [rax] ;
1.319 +
1.320 + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value
1.321 + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128
1.322 +
1.323 + paddw mm4, [GLOBAL(rd)] ;
1.324 + psraw mm4, VP8_FILTER_SHIFT ;
1.325 +
1.326 + movq mm7, mm3 ;
1.327 + packuswb mm7, mm4 ;
1.328 +
1.329 +
1.330 + pmullw mm3, [rax+16] ;
1.331 + pmullw mm4, [rax+16] ;
1.332 +
1.333 + paddw mm3, mm5 ;
1.334 + paddw mm4, mm6 ;
1.335 +
1.336 +
1.337 + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value
1.338 + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128
1.339 +
1.340 + paddw mm4, [GLOBAL(rd)] ;
1.341 + psraw mm4, VP8_FILTER_SHIFT ;
1.342 +
1.343 + packuswb mm3, mm4
1.344 +
1.345 + movq [rdi], mm3 ; store the results in the destination
1.346 +
1.347 +%if ABI_IS_32BIT
1.348 + add rsi, rdx ; next line
1.349 + add rdi, dword ptr arg(5) ;dst_pitch ;
1.350 +%else
1.351 + movsxd r8, dword ptr arg(5) ;dst_pitch
1.352 + add rsi, rdx ; next line
1.353 + add rdi, r8 ;dst_pitch
1.354 +%endif
1.355 + cmp rdi, rcx ;
1.356 + jne .next_row_8x8
1.357 +
1.358 + ; begin epilog
1.359 + pop rdi
1.360 + pop rsi
1.361 + RESTORE_GOT
1.362 + UNSHADOW_ARGS
1.363 + pop rbp
1.364 + ret
1.365 +
1.366 +
1.367 +;void bilinear_predict8x4_mmx
1.368 +;(
1.369 +; unsigned char *src_ptr,
1.370 +; int src_pixels_per_line,
1.371 +; int xoffset,
1.372 +; int yoffset,
1.373 +; unsigned char *dst_ptr,
1.374 +; int dst_pitch
1.375 +;)
1.376 +global sym(vp8_bilinear_predict8x4_mmx) PRIVATE
1.377 +sym(vp8_bilinear_predict8x4_mmx):
1.378 + push rbp
1.379 + mov rbp, rsp
1.380 + SHADOW_ARGS_TO_STACK 6
1.381 + GET_GOT rbx
1.382 + push rsi
1.383 + push rdi
1.384 + ; end prolog
1.385 +
1.386 + ;const short *HFilter = vp8_bilinear_filters_x86_8[xoffset];
1.387 + ;const short *VFilter = vp8_bilinear_filters_x86_8[yoffset];
1.388 +
1.389 + movsxd rax, dword ptr arg(2) ;xoffset
1.390 + mov rdi, arg(4) ;dst_ptr ;
1.391 +
1.392 + lea rcx, [GLOBAL(sym(vp8_bilinear_filters_x86_8))]
1.393 + shl rax, 5
1.394 +
1.395 + mov rsi, arg(0) ;src_ptr ;
1.396 + add rax, rcx
1.397 +
1.398 + movsxd rdx, dword ptr arg(5) ;dst_pitch
1.399 + movq mm1, [rax] ;
1.400 +
1.401 + movq mm2, [rax+16] ;
1.402 + movsxd rax, dword ptr arg(3) ;yoffset
1.403 +
1.404 + pxor mm0, mm0 ;
1.405 + shl rax, 5
1.406 +
1.407 + add rax, rcx
1.408 + lea rcx, [rdi+rdx*4] ;
1.409 +
1.410 + movsxd rdx, dword ptr arg(1) ;src_pixels_per_line ;
1.411 +
1.412 + ; get the first horizontal line done ;
1.413 + movq mm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14
1.414 + movq mm4, mm3 ; make a copy of current line
1.415 +
1.416 + punpcklbw mm3, mm0 ; xx 00 01 02 03 04 05 06
1.417 + punpckhbw mm4, mm0 ;
1.418 +
1.419 + pmullw mm3, mm1 ;
1.420 + pmullw mm4, mm1 ;
1.421 +
1.422 + movq mm5, [rsi+1] ;
1.423 + movq mm6, mm5 ;
1.424 +
1.425 + punpcklbw mm5, mm0 ;
1.426 + punpckhbw mm6, mm0 ;
1.427 +
1.428 + pmullw mm5, mm2 ;
1.429 + pmullw mm6, mm2 ;
1.430 +
1.431 + paddw mm3, mm5 ;
1.432 + paddw mm4, mm6 ;
1.433 +
1.434 + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value
1.435 + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128
1.436 +
1.437 + paddw mm4, [GLOBAL(rd)] ;
1.438 + psraw mm4, VP8_FILTER_SHIFT ;
1.439 +
1.440 + movq mm7, mm3 ;
1.441 + packuswb mm7, mm4 ;
1.442 +
1.443 + add rsi, rdx ; next line
1.444 +.next_row_8x4:
1.445 + movq mm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14
1.446 + movq mm4, mm3 ; make a copy of current line
1.447 +
1.448 + punpcklbw mm3, mm0 ; xx 00 01 02 03 04 05 06
1.449 + punpckhbw mm4, mm0 ;
1.450 +
1.451 + pmullw mm3, mm1 ;
1.452 + pmullw mm4, mm1 ;
1.453 +
1.454 + movq mm5, [rsi+1] ;
1.455 + movq mm6, mm5 ;
1.456 +
1.457 + punpcklbw mm5, mm0 ;
1.458 + punpckhbw mm6, mm0 ;
1.459 +
1.460 + pmullw mm5, mm2 ;
1.461 + pmullw mm6, mm2 ;
1.462 +
1.463 + paddw mm3, mm5 ;
1.464 + paddw mm4, mm6 ;
1.465 +
1.466 + movq mm5, mm7 ;
1.467 + movq mm6, mm7 ;
1.468 +
1.469 + punpcklbw mm5, mm0 ;
1.470 + punpckhbw mm6, mm0
1.471 +
1.472 + pmullw mm5, [rax] ;
1.473 + pmullw mm6, [rax] ;
1.474 +
1.475 + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value
1.476 + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128
1.477 +
1.478 + paddw mm4, [GLOBAL(rd)] ;
1.479 + psraw mm4, VP8_FILTER_SHIFT ;
1.480 +
1.481 + movq mm7, mm3 ;
1.482 + packuswb mm7, mm4 ;
1.483 +
1.484 +
1.485 + pmullw mm3, [rax+16] ;
1.486 + pmullw mm4, [rax+16] ;
1.487 +
1.488 + paddw mm3, mm5 ;
1.489 + paddw mm4, mm6 ;
1.490 +
1.491 +
1.492 + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value
1.493 + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128
1.494 +
1.495 + paddw mm4, [GLOBAL(rd)] ;
1.496 + psraw mm4, VP8_FILTER_SHIFT ;
1.497 +
1.498 + packuswb mm3, mm4
1.499 +
1.500 + movq [rdi], mm3 ; store the results in the destination
1.501 +
1.502 +%if ABI_IS_32BIT
1.503 + add rsi, rdx ; next line
1.504 + add rdi, dword ptr arg(5) ;dst_pitch ;
1.505 +%else
1.506 + movsxd r8, dword ptr arg(5) ;dst_pitch
1.507 + add rsi, rdx ; next line
1.508 + add rdi, r8
1.509 +%endif
1.510 + cmp rdi, rcx ;
1.511 + jne .next_row_8x4
1.512 +
1.513 + ; begin epilog
1.514 + pop rdi
1.515 + pop rsi
1.516 + RESTORE_GOT
1.517 + UNSHADOW_ARGS
1.518 + pop rbp
1.519 + ret
1.520 +
1.521 +
1.522 +;void bilinear_predict4x4_mmx
1.523 +;(
1.524 +; unsigned char *src_ptr,
1.525 +; int src_pixels_per_line,
1.526 +; int xoffset,
1.527 +; int yoffset,
1.528 +; unsigned char *dst_ptr,
1.529 +; int dst_pitch
1.530 +;)
1.531 +global sym(vp8_bilinear_predict4x4_mmx) PRIVATE
1.532 +sym(vp8_bilinear_predict4x4_mmx):
1.533 + push rbp
1.534 + mov rbp, rsp
1.535 + SHADOW_ARGS_TO_STACK 6
1.536 + GET_GOT rbx
1.537 + push rsi
1.538 + push rdi
1.539 + ; end prolog
1.540 +
1.541 + ;const short *HFilter = vp8_bilinear_filters_x86_8[xoffset];
1.542 + ;const short *VFilter = vp8_bilinear_filters_x86_8[yoffset];
1.543 +
1.544 + movsxd rax, dword ptr arg(2) ;xoffset
1.545 + mov rdi, arg(4) ;dst_ptr ;
1.546 +
1.547 + lea rcx, [GLOBAL(sym(vp8_bilinear_filters_x86_8))]
1.548 + shl rax, 5
1.549 +
1.550 + add rax, rcx ; HFilter
1.551 + mov rsi, arg(0) ;src_ptr ;
1.552 +
1.553 + movsxd rdx, dword ptr arg(5) ;ldst_pitch
1.554 + movq mm1, [rax] ;
1.555 +
1.556 + movq mm2, [rax+16] ;
1.557 + movsxd rax, dword ptr arg(3) ;yoffset
1.558 +
1.559 + pxor mm0, mm0 ;
1.560 + shl rax, 5
1.561 +
1.562 + add rax, rcx
1.563 + lea rcx, [rdi+rdx*4] ;
1.564 +
1.565 + movsxd rdx, dword ptr arg(1) ;src_pixels_per_line ;
1.566 +
1.567 + ; get the first horizontal line done ;
1.568 + movd mm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14
1.569 + punpcklbw mm3, mm0 ; xx 00 01 02 03 04 05 06
1.570 +
1.571 + pmullw mm3, mm1 ;
1.572 + movd mm5, [rsi+1] ;
1.573 +
1.574 + punpcklbw mm5, mm0 ;
1.575 + pmullw mm5, mm2 ;
1.576 +
1.577 + paddw mm3, mm5 ;
1.578 + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value
1.579 +
1.580 + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128
1.581 +
1.582 + movq mm7, mm3 ;
1.583 + packuswb mm7, mm0 ;
1.584 +
1.585 + add rsi, rdx ; next line
1.586 +.next_row_4x4:
1.587 + movd mm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14
1.588 + punpcklbw mm3, mm0 ; xx 00 01 02 03 04 05 06
1.589 +
1.590 + pmullw mm3, mm1 ;
1.591 + movd mm5, [rsi+1] ;
1.592 +
1.593 + punpcklbw mm5, mm0 ;
1.594 + pmullw mm5, mm2 ;
1.595 +
1.596 + paddw mm3, mm5 ;
1.597 +
1.598 + movq mm5, mm7 ;
1.599 + punpcklbw mm5, mm0 ;
1.600 +
1.601 + pmullw mm5, [rax] ;
1.602 + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value
1.603 +
1.604 + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128
1.605 + movq mm7, mm3 ;
1.606 +
1.607 + packuswb mm7, mm0 ;
1.608 +
1.609 + pmullw mm3, [rax+16] ;
1.610 + paddw mm3, mm5 ;
1.611 +
1.612 +
1.613 + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value
1.614 + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128
1.615 +
1.616 + packuswb mm3, mm0
1.617 + movd [rdi], mm3 ; store the results in the destination
1.618 +
1.619 +%if ABI_IS_32BIT
1.620 + add rsi, rdx ; next line
1.621 + add rdi, dword ptr arg(5) ;dst_pitch ;
1.622 +%else
1.623 + movsxd r8, dword ptr arg(5) ;dst_pitch ;
1.624 + add rsi, rdx ; next line
1.625 + add rdi, r8
1.626 +%endif
1.627 +
1.628 + cmp rdi, rcx ;
1.629 + jne .next_row_4x4
1.630 +
1.631 + ; begin epilog
1.632 + pop rdi
1.633 + pop rsi
1.634 + RESTORE_GOT
1.635 + UNSHADOW_ARGS
1.636 + pop rbp
1.637 + ret
1.638 +
1.639 +
1.640 +
1.641 +SECTION_RODATA
1.642 +align 16
1.643 +rd:
1.644 + times 4 dw 0x40
1.645 +
1.646 +align 16
1.647 +global HIDDEN_DATA(sym(vp8_six_tap_mmx))
1.648 +sym(vp8_six_tap_mmx):
1.649 + times 8 dw 0
1.650 + times 8 dw 0
1.651 + times 8 dw 128
1.652 + times 8 dw 0
1.653 + times 8 dw 0
1.654 + times 8 dw 0
1.655 +
1.656 + times 8 dw 0
1.657 + times 8 dw -6
1.658 + times 8 dw 123
1.659 + times 8 dw 12
1.660 + times 8 dw -1
1.661 + times 8 dw 0
1.662 +
1.663 + times 8 dw 2
1.664 + times 8 dw -11
1.665 + times 8 dw 108
1.666 + times 8 dw 36
1.667 + times 8 dw -8
1.668 + times 8 dw 1
1.669 +
1.670 + times 8 dw 0
1.671 + times 8 dw -9
1.672 + times 8 dw 93
1.673 + times 8 dw 50
1.674 + times 8 dw -6
1.675 + times 8 dw 0
1.676 +
1.677 + times 8 dw 3
1.678 + times 8 dw -16
1.679 + times 8 dw 77
1.680 + times 8 dw 77
1.681 + times 8 dw -16
1.682 + times 8 dw 3
1.683 +
1.684 + times 8 dw 0
1.685 + times 8 dw -6
1.686 + times 8 dw 50
1.687 + times 8 dw 93
1.688 + times 8 dw -9
1.689 + times 8 dw 0
1.690 +
1.691 + times 8 dw 1
1.692 + times 8 dw -8
1.693 + times 8 dw 36
1.694 + times 8 dw 108
1.695 + times 8 dw -11
1.696 + times 8 dw 2
1.697 +
1.698 + times 8 dw 0
1.699 + times 8 dw -1
1.700 + times 8 dw 12
1.701 + times 8 dw 123
1.702 + times 8 dw -6
1.703 + times 8 dw 0
1.704 +
1.705 +
