1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/common/x86/variance_impl_mmx.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,851 @@
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 +;unsigned int vp8_get_mb_ss_mmx( short *src_ptr )
1.18 +global sym(vp8_get_mb_ss_mmx) PRIVATE
1.19 +sym(vp8_get_mb_ss_mmx):
1.20 + push rbp
1.21 + mov rbp, rsp
1.22 + SHADOW_ARGS_TO_STACK 7
1.23 + GET_GOT rbx
1.24 + push rsi
1.25 + push rdi
1.26 + sub rsp, 8
1.27 + ; end prolog
1.28 +
1.29 + mov rax, arg(0) ;src_ptr
1.30 + mov rcx, 16
1.31 + pxor mm4, mm4
1.32 +
1.33 +.NEXTROW:
1.34 + movq mm0, [rax]
1.35 + movq mm1, [rax+8]
1.36 + movq mm2, [rax+16]
1.37 + movq mm3, [rax+24]
1.38 + pmaddwd mm0, mm0
1.39 + pmaddwd mm1, mm1
1.40 + pmaddwd mm2, mm2
1.41 + pmaddwd mm3, mm3
1.42 +
1.43 + paddd mm4, mm0
1.44 + paddd mm4, mm1
1.45 + paddd mm4, mm2
1.46 + paddd mm4, mm3
1.47 +
1.48 + add rax, 32
1.49 + dec rcx
1.50 + ja .NEXTROW
1.51 + movq QWORD PTR [rsp], mm4
1.52 +
1.53 + ;return sum[0]+sum[1];
1.54 + movsxd rax, dword ptr [rsp]
1.55 + movsxd rcx, dword ptr [rsp+4]
1.56 + add rax, rcx
1.57 +
1.58 +
1.59 + ; begin epilog
1.60 + add rsp, 8
1.61 + pop rdi
1.62 + pop rsi
1.63 + RESTORE_GOT
1.64 + UNSHADOW_ARGS
1.65 + pop rbp
1.66 + ret
1.67 +
1.68 +
1.69 +;unsigned int vp8_get8x8var_mmx
1.70 +;(
1.71 +; unsigned char *src_ptr,
1.72 +; int source_stride,
1.73 +; unsigned char *ref_ptr,
1.74 +; int recon_stride,
1.75 +; unsigned int *SSE,
1.76 +; int *Sum
1.77 +;)
1.78 +global sym(vp8_get8x8var_mmx) PRIVATE
1.79 +sym(vp8_get8x8var_mmx):
1.80 + push rbp
1.81 + mov rbp, rsp
1.82 + SHADOW_ARGS_TO_STACK 6
1.83 + push rsi
1.84 + push rdi
1.85 + push rbx
1.86 + sub rsp, 16
1.87 + ; end prolog
1.88 +
1.89 +
1.90 + pxor mm5, mm5 ; Blank mmx6
1.91 + pxor mm6, mm6 ; Blank mmx7
1.92 + pxor mm7, mm7 ; Blank mmx7
1.93 +
1.94 + mov rax, arg(0) ;[src_ptr] ; Load base addresses
1.95 + mov rbx, arg(2) ;[ref_ptr]
1.96 + movsxd rcx, dword ptr arg(1) ;[source_stride]
1.97 + movsxd rdx, dword ptr arg(3) ;[recon_stride]
1.98 +
1.99 + ; Row 1
1.100 + movq mm0, [rax] ; Copy eight bytes to mm0
1.101 + movq mm1, [rbx] ; Copy eight bytes to mm1
1.102 + movq mm2, mm0 ; Take copies
1.103 + movq mm3, mm1 ; Take copies
1.104 +
1.105 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.106 + punpcklbw mm1, mm6
1.107 + punpckhbw mm2, mm6 ; unpack to higher prrcision
1.108 + punpckhbw mm3, mm6
1.109 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.110 + psubsw mm2, mm3 ; A-B (high order) to MM2
1.111 +
1.112 + paddw mm5, mm0 ; accumulate differences in mm5
1.113 + paddw mm5, mm2 ; accumulate differences in mm5
1.114 +
1.115 + pmaddwd mm0, mm0 ; square and accumulate
1.116 + pmaddwd mm2, mm2 ; square and accumulate
1.117 + add rbx,rdx ; Inc pointer into ref data
1.118 + add rax,rcx ; Inc pointer into the new data
1.119 + movq mm1, [rbx] ; Copy eight bytes to mm1
1.120 + paddd mm7, mm0 ; accumulate in mm7
1.121 + paddd mm7, mm2 ; accumulate in mm7
1.122 +
1.123 +
1.124 + ; Row 2
1.125 + movq mm0, [rax] ; Copy eight bytes to mm0
1.126 + movq mm2, mm0 ; Take copies
1.127 + movq mm3, mm1 ; Take copies
1.128 +
1.129 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.130 + punpcklbw mm1, mm6
1.131 + punpckhbw mm2, mm6 ; unpack to higher prrcision
1.132 + punpckhbw mm3, mm6
1.133 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.134 + psubsw mm2, mm3 ; A-B (high order) to MM2
1.135 +
1.136 + paddw mm5, mm0 ; accumulate differences in mm5
1.137 + paddw mm5, mm2 ; accumulate differences in mm5
1.138 +
1.139 + pmaddwd mm0, mm0 ; square and accumulate
1.140 + pmaddwd mm2, mm2 ; square and accumulate
1.141 + add rbx,rdx ; Inc pointer into ref data
1.142 + add rax,rcx ; Inc pointer into the new data
1.143 + movq mm1, [rbx] ; Copy eight bytes to mm1
1.144 + paddd mm7, mm0 ; accumulate in mm7
1.145 + paddd mm7, mm2 ; accumulate in mm7
1.146 +
1.147 + ; Row 3
1.148 + movq mm0, [rax] ; Copy eight bytes to mm0
1.149 + movq mm2, mm0 ; Take copies
1.150 + movq mm3, mm1 ; Take copies
1.151 +
1.152 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.153 + punpcklbw mm1, mm6
1.154 + punpckhbw mm2, mm6 ; unpack to higher prrcision
1.155 + punpckhbw mm3, mm6
1.156 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.157 + psubsw mm2, mm3 ; A-B (high order) to MM2
1.158 +
1.159 + paddw mm5, mm0 ; accumulate differences in mm5
1.160 + paddw mm5, mm2 ; accumulate differences in mm5
1.161 +
1.162 + pmaddwd mm0, mm0 ; square and accumulate
1.163 + pmaddwd mm2, mm2 ; square and accumulate
1.164 + add rbx,rdx ; Inc pointer into ref data
1.165 + add rax,rcx ; Inc pointer into the new data
1.166 + movq mm1, [rbx] ; Copy eight bytes to mm1
1.167 + paddd mm7, mm0 ; accumulate in mm7
1.168 + paddd mm7, mm2 ; accumulate in mm7
1.169 +
1.170 + ; Row 4
1.171 + movq mm0, [rax] ; Copy eight bytes to mm0
1.172 + movq mm2, mm0 ; Take copies
1.173 + movq mm3, mm1 ; Take copies
1.174 +
1.175 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.176 + punpcklbw mm1, mm6
1.177 + punpckhbw mm2, mm6 ; unpack to higher prrcision
1.178 + punpckhbw mm3, mm6
1.179 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.180 + psubsw mm2, mm3 ; A-B (high order) to MM2
1.181 +
1.182 + paddw mm5, mm0 ; accumulate differences in mm5
1.183 + paddw mm5, mm2 ; accumulate differences in mm5
1.184 +
1.185 + pmaddwd mm0, mm0 ; square and accumulate
1.186 + pmaddwd mm2, mm2 ; square and accumulate
1.187 + add rbx,rdx ; Inc pointer into ref data
1.188 + add rax,rcx ; Inc pointer into the new data
1.189 + movq mm1, [rbx] ; Copy eight bytes to mm1
1.190 + paddd mm7, mm0 ; accumulate in mm7
1.191 + paddd mm7, mm2 ; accumulate in mm7
1.192 +
1.193 + ; Row 5
1.194 + movq mm0, [rax] ; Copy eight bytes to mm0
1.195 + movq mm2, mm0 ; Take copies
1.196 + movq mm3, mm1 ; Take copies
1.197 +
1.198 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.199 + punpcklbw mm1, mm6
1.200 + punpckhbw mm2, mm6 ; unpack to higher prrcision
1.201 + punpckhbw mm3, mm6
1.202 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.203 + psubsw mm2, mm3 ; A-B (high order) to MM2
1.204 +
1.205 + paddw mm5, mm0 ; accumulate differences in mm5
1.206 + paddw mm5, mm2 ; accumulate differences in mm5
1.207 +
1.208 + pmaddwd mm0, mm0 ; square and accumulate
1.209 + pmaddwd mm2, mm2 ; square and accumulate
1.210 + add rbx,rdx ; Inc pointer into ref data
1.211 + add rax,rcx ; Inc pointer into the new data
1.212 + movq mm1, [rbx] ; Copy eight bytes to mm1
1.213 + ; movq mm4, [rbx + rdx]
1.214 + paddd mm7, mm0 ; accumulate in mm7
1.215 + paddd mm7, mm2 ; accumulate in mm7
1.216 +
1.217 + ; Row 6
1.218 + movq mm0, [rax] ; Copy eight bytes to mm0
1.219 + movq mm2, mm0 ; Take copies
1.220 + movq mm3, mm1 ; Take copies
1.221 +
1.222 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.223 + punpcklbw mm1, mm6
1.224 + punpckhbw mm2, mm6 ; unpack to higher prrcision
1.225 + punpckhbw mm3, mm6
1.226 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.227 + psubsw mm2, mm3 ; A-B (high order) to MM2
1.228 +
1.229 + paddw mm5, mm0 ; accumulate differences in mm5
1.230 + paddw mm5, mm2 ; accumulate differences in mm5
1.231 +
1.232 + pmaddwd mm0, mm0 ; square and accumulate
1.233 + pmaddwd mm2, mm2 ; square and accumulate
1.234 + add rbx,rdx ; Inc pointer into ref data
1.235 + add rax,rcx ; Inc pointer into the new data
1.236 + movq mm1, [rbx] ; Copy eight bytes to mm1
1.237 + paddd mm7, mm0 ; accumulate in mm7
1.238 + paddd mm7, mm2 ; accumulate in mm7
1.239 +
1.240 + ; Row 7
1.241 + movq mm0, [rax] ; Copy eight bytes to mm0
1.242 + movq mm2, mm0 ; Take copies
1.243 + movq mm3, mm1 ; Take copies
1.244 +
1.245 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.246 + punpcklbw mm1, mm6
1.247 + punpckhbw mm2, mm6 ; unpack to higher prrcision
1.248 + punpckhbw mm3, mm6
1.249 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.250 + psubsw mm2, mm3 ; A-B (high order) to MM2
1.251 +
1.252 + paddw mm5, mm0 ; accumulate differences in mm5
1.253 + paddw mm5, mm2 ; accumulate differences in mm5
1.254 +
1.255 + pmaddwd mm0, mm0 ; square and accumulate
1.256 + pmaddwd mm2, mm2 ; square and accumulate
1.257 + add rbx,rdx ; Inc pointer into ref data
1.258 + add rax,rcx ; Inc pointer into the new data
1.259 + movq mm1, [rbx] ; Copy eight bytes to mm1
1.260 + paddd mm7, mm0 ; accumulate in mm7
1.261 + paddd mm7, mm2 ; accumulate in mm7
1.262 +
1.263 + ; Row 8
1.264 + movq mm0, [rax] ; Copy eight bytes to mm0
1.265 + movq mm2, mm0 ; Take copies
1.266 + movq mm3, mm1 ; Take copies
1.267 +
1.268 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.269 + punpcklbw mm1, mm6
1.270 + punpckhbw mm2, mm6 ; unpack to higher prrcision
1.271 + punpckhbw mm3, mm6
1.272 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.273 + psubsw mm2, mm3 ; A-B (high order) to MM2
1.274 +
1.275 + paddw mm5, mm0 ; accumulate differences in mm5
1.276 + paddw mm5, mm2 ; accumulate differences in mm5
1.277 +
1.278 + pmaddwd mm0, mm0 ; square and accumulate
1.279 + pmaddwd mm2, mm2 ; square and accumulate
1.280 + add rbx,rdx ; Inc pointer into ref data
1.281 + add rax,rcx ; Inc pointer into the new data
1.282 + paddd mm7, mm0 ; accumulate in mm7
1.283 + paddd mm7, mm2 ; accumulate in mm7
1.284 +
1.285 + ; Now accumulate the final results.
1.286 + movq QWORD PTR [rsp+8], mm5 ; copy back accumulated results into normal memory
1.287 + movq QWORD PTR [rsp], mm7 ; copy back accumulated results into normal memory
1.288 + movsx rdx, WORD PTR [rsp+8]
1.289 + movsx rcx, WORD PTR [rsp+10]
1.290 + movsx rbx, WORD PTR [rsp+12]
1.291 + movsx rax, WORD PTR [rsp+14]
1.292 + add rdx, rcx
1.293 + add rbx, rax
1.294 + add rdx, rbx ;XSum
1.295 + movsxd rax, DWORD PTR [rsp]
1.296 + movsxd rcx, DWORD PTR [rsp+4]
1.297 + add rax, rcx ;XXSum
1.298 + mov rsi, arg(4) ;SSE
1.299 + mov rdi, arg(5) ;Sum
1.300 + mov dword ptr [rsi], eax
1.301 + mov dword ptr [rdi], edx
1.302 + xor rax, rax ; return 0
1.303 +
1.304 +
1.305 + ; begin epilog
1.306 + add rsp, 16
1.307 + pop rbx
1.308 + pop rdi
1.309 + pop rsi
1.310 + UNSHADOW_ARGS
1.311 + pop rbp
1.312 + ret
1.313 +
1.314 +
1.315 +
1.316 +;unsigned int
1.317 +;vp8_get4x4var_mmx
1.318 +;(
1.319 +; unsigned char *src_ptr,
1.320 +; int source_stride,
1.321 +; unsigned char *ref_ptr,
1.322 +; int recon_stride,
1.323 +; unsigned int *SSE,
1.324 +; int *Sum
1.325 +;)
1.326 +global sym(vp8_get4x4var_mmx) PRIVATE
1.327 +sym(vp8_get4x4var_mmx):
1.328 + push rbp
1.329 + mov rbp, rsp
1.330 + SHADOW_ARGS_TO_STACK 6
1.331 + push rsi
1.332 + push rdi
1.333 + push rbx
1.334 + sub rsp, 16
1.335 + ; end prolog
1.336 +
1.337 +
1.338 + pxor mm5, mm5 ; Blank mmx6
1.339 + pxor mm6, mm6 ; Blank mmx7
1.340 + pxor mm7, mm7 ; Blank mmx7
1.341 +
1.342 + mov rax, arg(0) ;[src_ptr] ; Load base addresses
1.343 + mov rbx, arg(2) ;[ref_ptr]
1.344 + movsxd rcx, dword ptr arg(1) ;[source_stride]
1.345 + movsxd rdx, dword ptr arg(3) ;[recon_stride]
1.346 +
1.347 + ; Row 1
1.348 + movq mm0, [rax] ; Copy eight bytes to mm0
1.349 + movq mm1, [rbx] ; Copy eight bytes to mm1
1.350 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.351 + punpcklbw mm1, mm6
1.352 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.353 + paddw mm5, mm0 ; accumulate differences in mm5
1.354 + pmaddwd mm0, mm0 ; square and accumulate
1.355 + add rbx,rdx ; Inc pointer into ref data
1.356 + add rax,rcx ; Inc pointer into the new data
1.357 + movq mm1, [rbx] ; Copy eight bytes to mm1
1.358 + paddd mm7, mm0 ; accumulate in mm7
1.359 +
1.360 +
1.361 + ; Row 2
1.362 + movq mm0, [rax] ; Copy eight bytes to mm0
1.363 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.364 + punpcklbw mm1, mm6
1.365 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.366 + paddw mm5, mm0 ; accumulate differences in mm5
1.367 +
1.368 + pmaddwd mm0, mm0 ; square and accumulate
1.369 + add rbx,rdx ; Inc pointer into ref data
1.370 + add rax,rcx ; Inc pointer into the new data
1.371 + movq mm1, [rbx] ; Copy eight bytes to mm1
1.372 + paddd mm7, mm0 ; accumulate in mm7
1.373 +
1.374 + ; Row 3
1.375 + movq mm0, [rax] ; Copy eight bytes to mm0
1.376 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.377 + punpcklbw mm1, mm6
1.378 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.379 + paddw mm5, mm0 ; accumulate differences in mm5
1.380 +
1.381 + pmaddwd mm0, mm0 ; square and accumulate
1.382 + add rbx,rdx ; Inc pointer into ref data
1.383 + add rax,rcx ; Inc pointer into the new data
1.384 + movq mm1, [rbx] ; Copy eight bytes to mm1
1.385 + paddd mm7, mm0 ; accumulate in mm7
1.386 +
1.387 + ; Row 4
1.388 + movq mm0, [rax] ; Copy eight bytes to mm0
1.389 +
1.390 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.391 + punpcklbw mm1, mm6
1.392 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.393 +
1.394 + paddw mm5, mm0 ; accumulate differences in mm5
1.395 +
1.396 + pmaddwd mm0, mm0 ; square and accumulate
1.397 + paddd mm7, mm0 ; accumulate in mm7
1.398 +
1.399 +
1.400 + ; Now accumulate the final results.
1.401 + movq QWORD PTR [rsp+8], mm5 ; copy back accumulated results into normal memory
1.402 + movq QWORD PTR [rsp], mm7 ; copy back accumulated results into normal memory
1.403 + movsx rdx, WORD PTR [rsp+8]
1.404 + movsx rcx, WORD PTR [rsp+10]
1.405 + movsx rbx, WORD PTR [rsp+12]
1.406 + movsx rax, WORD PTR [rsp+14]
1.407 + add rdx, rcx
1.408 + add rbx, rax
1.409 + add rdx, rbx ;XSum
1.410 + movsxd rax, DWORD PTR [rsp]
1.411 + movsxd rcx, DWORD PTR [rsp+4]
1.412 + add rax, rcx ;XXSum
1.413 + mov rsi, arg(4) ;SSE
1.414 + mov rdi, arg(5) ;Sum
1.415 + mov dword ptr [rsi], eax
1.416 + mov dword ptr [rdi], edx
1.417 + xor rax, rax ; return 0
1.418 +
1.419 +
1.420 + ; begin epilog
1.421 + add rsp, 16
1.422 + pop rbx
1.423 + pop rdi
1.424 + pop rsi
1.425 + UNSHADOW_ARGS
1.426 + pop rbp
1.427 + ret
1.428 +
1.429 +
1.430 +
1.431 +;unsigned int
1.432 +;vp8_get4x4sse_cs_mmx
1.433 +;(
1.434 +; unsigned char *src_ptr,
1.435 +; int source_stride,
1.436 +; unsigned char *ref_ptr,
1.437 +; int recon_stride
1.438 +;)
1.439 +global sym(vp8_get4x4sse_cs_mmx) PRIVATE
1.440 +sym(vp8_get4x4sse_cs_mmx):
1.441 + push rbp
1.442 + mov rbp, rsp
1.443 + SHADOW_ARGS_TO_STACK 4
1.444 + push rsi
1.445 + push rdi
1.446 + push rbx
1.447 + ; end prolog
1.448 +
1.449 +
1.450 + pxor mm6, mm6 ; Blank mmx7
1.451 + pxor mm7, mm7 ; Blank mmx7
1.452 +
1.453 + mov rax, arg(0) ;[src_ptr] ; Load base addresses
1.454 + mov rbx, arg(2) ;[ref_ptr]
1.455 + movsxd rcx, dword ptr arg(1) ;[source_stride]
1.456 + movsxd rdx, dword ptr arg(3) ;[recon_stride]
1.457 + ; Row 1
1.458 + movd mm0, [rax] ; Copy eight bytes to mm0
1.459 + movd mm1, [rbx] ; Copy eight bytes to mm1
1.460 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.461 + punpcklbw mm1, mm6
1.462 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.463 + pmaddwd mm0, mm0 ; square and accumulate
1.464 + add rbx,rdx ; Inc pointer into ref data
1.465 + add rax,rcx ; Inc pointer into the new data
1.466 + movd mm1, [rbx] ; Copy eight bytes to mm1
1.467 + paddd mm7, mm0 ; accumulate in mm7
1.468 +
1.469 + ; Row 2
1.470 + movd mm0, [rax] ; Copy eight bytes to mm0
1.471 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.472 + punpcklbw mm1, mm6
1.473 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.474 + pmaddwd mm0, mm0 ; square and accumulate
1.475 + add rbx,rdx ; Inc pointer into ref data
1.476 + add rax,rcx ; Inc pointer into the new data
1.477 + movd mm1, [rbx] ; Copy eight bytes to mm1
1.478 + paddd mm7, mm0 ; accumulate in mm7
1.479 +
1.480 + ; Row 3
1.481 + movd mm0, [rax] ; Copy eight bytes to mm0
1.482 + punpcklbw mm1, mm6
1.483 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.484 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.485 +
1.486 + pmaddwd mm0, mm0 ; square and accumulate
1.487 + add rbx,rdx ; Inc pointer into ref data
1.488 + add rax,rcx ; Inc pointer into the new data
1.489 + movd mm1, [rbx] ; Copy eight bytes to mm1
1.490 + paddd mm7, mm0 ; accumulate in mm7
1.491 +
1.492 + ; Row 4
1.493 + movd mm0, [rax] ; Copy eight bytes to mm0
1.494 + punpcklbw mm0, mm6 ; unpack to higher prrcision
1.495 + punpcklbw mm1, mm6
1.496 + psubsw mm0, mm1 ; A-B (low order) to MM0
1.497 + pmaddwd mm0, mm0 ; square and accumulate
1.498 + paddd mm7, mm0 ; accumulate in mm7
1.499 +
1.500 + movq mm0, mm7 ;
1.501 + psrlq mm7, 32
1.502 +
1.503 + paddd mm0, mm7
1.504 + movq rax, mm0
1.505 +
1.506 +
1.507 + ; begin epilog
1.508 + pop rbx
1.509 + pop rdi
1.510 + pop rsi
1.511 + UNSHADOW_ARGS
1.512 + pop rbp
1.513 + ret
1.514 +
1.515 +%define mmx_filter_shift 7
1.516 +
1.517 +;void vp8_filter_block2d_bil4x4_var_mmx
1.518 +;(
1.519 +; unsigned char *ref_ptr,
1.520 +; int ref_pixels_per_line,
1.521 +; unsigned char *src_ptr,
1.522 +; int src_pixels_per_line,
1.523 +; unsigned short *HFilter,
1.524 +; unsigned short *VFilter,
1.525 +; int *sum,
1.526 +; unsigned int *sumsquared
1.527 +;)
1.528 +global sym(vp8_filter_block2d_bil4x4_var_mmx) PRIVATE
1.529 +sym(vp8_filter_block2d_bil4x4_var_mmx):
1.530 + push rbp
1.531 + mov rbp, rsp
1.532 + SHADOW_ARGS_TO_STACK 8
1.533 + GET_GOT rbx
1.534 + push rsi
1.535 + push rdi
1.536 + sub rsp, 16
1.537 + ; end prolog
1.538 +
1.539 +
1.540 + pxor mm6, mm6 ;
1.541 + pxor mm7, mm7 ;
1.542 +
1.543 + mov rax, arg(4) ;HFilter ;
1.544 + mov rdx, arg(5) ;VFilter ;
1.545 +
1.546 + mov rsi, arg(0) ;ref_ptr ;
1.547 + mov rdi, arg(2) ;src_ptr ;
1.548 +
1.549 + mov rcx, 4 ;
1.550 + pxor mm0, mm0 ;
1.551 +
1.552 + movd mm1, [rsi] ;
1.553 + movd mm3, [rsi+1] ;
1.554 +
1.555 + punpcklbw mm1, mm0 ;
1.556 + pmullw mm1, [rax] ;
1.557 +
1.558 + punpcklbw mm3, mm0 ;
1.559 + pmullw mm3, [rax+8] ;
1.560 +
1.561 + paddw mm1, mm3 ;
1.562 + paddw mm1, [GLOBAL(mmx_bi_rd)] ;
1.563 +
1.564 + psraw mm1, mmx_filter_shift ;
1.565 + movq mm5, mm1
1.566 +
1.567 +%if ABI_IS_32BIT
1.568 + add rsi, dword ptr arg(1) ;ref_pixels_per_line ;
1.569 +%else
1.570 + movsxd r8, dword ptr arg(1) ;ref_pixels_per_line ;
1.571 + add rsi, r8
1.572 +%endif
1.573 +
1.574 +.filter_block2d_bil4x4_var_mmx_loop:
1.575 +
1.576 + movd mm1, [rsi] ;
1.577 + movd mm3, [rsi+1] ;
1.578 +
1.579 + punpcklbw mm1, mm0 ;
1.580 + pmullw mm1, [rax] ;
1.581 +
1.582 + punpcklbw mm3, mm0 ;
1.583 + pmullw mm3, [rax+8] ;
1.584 +
1.585 + paddw mm1, mm3 ;
1.586 + paddw mm1, [GLOBAL(mmx_bi_rd)] ;
1.587 +
1.588 + psraw mm1, mmx_filter_shift ;
1.589 + movq mm3, mm5 ;
1.590 +
1.591 + movq mm5, mm1 ;
1.592 + pmullw mm3, [rdx] ;
1.593 +
1.594 + pmullw mm1, [rdx+8] ;
1.595 + paddw mm1, mm3 ;
1.596 +
1.597 +
1.598 + paddw mm1, [GLOBAL(mmx_bi_rd)] ;
1.599 + psraw mm1, mmx_filter_shift ;
1.600 +
1.601 + movd mm3, [rdi] ;
1.602 + punpcklbw mm3, mm0 ;
1.603 +
1.604 + psubw mm1, mm3 ;
1.605 + paddw mm6, mm1 ;
1.606 +
1.607 + pmaddwd mm1, mm1 ;
1.608 + paddd mm7, mm1 ;
1.609 +
1.610 +%if ABI_IS_32BIT
1.611 + add rsi, dword ptr arg(1) ;ref_pixels_per_line ;
1.612 + add rdi, dword ptr arg(3) ;src_pixels_per_line ;
1.613 +%else
1.614 + movsxd r8, dword ptr arg(1) ;ref_pixels_per_line
1.615 + movsxd r9, dword ptr arg(3) ;src_pixels_per_line
1.616 + add rsi, r8
1.617 + add rdi, r9
1.618 +%endif
1.619 + sub rcx, 1 ;
1.620 + jnz .filter_block2d_bil4x4_var_mmx_loop ;
1.621 +
1.622 +
1.623 + pxor mm3, mm3 ;
1.624 + pxor mm2, mm2 ;
1.625 +
1.626 + punpcklwd mm2, mm6 ;
1.627 + punpckhwd mm3, mm6 ;
1.628 +
1.629 + paddd mm2, mm3 ;
1.630 + movq mm6, mm2 ;
1.631 +
1.632 + psrlq mm6, 32 ;
1.633 + paddd mm2, mm6 ;
1.634 +
1.635 + psrad mm2, 16 ;
1.636 + movq mm4, mm7 ;
1.637 +
1.638 + psrlq mm4, 32 ;
1.639 + paddd mm4, mm7 ;
1.640 +
1.641 + mov rdi, arg(6) ;sum
1.642 + mov rsi, arg(7) ;sumsquared
1.643 +
1.644 + movd dword ptr [rdi], mm2 ;
1.645 + movd dword ptr [rsi], mm4 ;
1.646 +
1.647 +
1.648 +
1.649 + ; begin epilog
1.650 + add rsp, 16
1.651 + pop rdi
1.652 + pop rsi
1.653 + RESTORE_GOT
1.654 + UNSHADOW_ARGS
1.655 + pop rbp
1.656 + ret
1.657 +
1.658 +
1.659 +
1.660 +
1.661 +;void vp8_filter_block2d_bil_var_mmx
1.662 +;(
1.663 +; unsigned char *ref_ptr,
1.664 +; int ref_pixels_per_line,
1.665 +; unsigned char *src_ptr,
1.666 +; int src_pixels_per_line,
1.667 +; unsigned int Height,
1.668 +; unsigned short *HFilter,
1.669 +; unsigned short *VFilter,
1.670 +; int *sum,
1.671 +; unsigned int *sumsquared
1.672 +;)
1.673 +global sym(vp8_filter_block2d_bil_var_mmx) PRIVATE
1.674 +sym(vp8_filter_block2d_bil_var_mmx):
1.675 + push rbp
1.676 + mov rbp, rsp
1.677 + SHADOW_ARGS_TO_STACK 9
1.678 + GET_GOT rbx
1.679 + push rsi
1.680 + push rdi
1.681 + sub rsp, 16
1.682 + ; end prolog
1.683 +
1.684 + pxor mm6, mm6 ;
1.685 + pxor mm7, mm7 ;
1.686 + mov rax, arg(5) ;HFilter ;
1.687 +
1.688 + mov rdx, arg(6) ;VFilter ;
1.689 + mov rsi, arg(0) ;ref_ptr ;
1.690 +
1.691 + mov rdi, arg(2) ;src_ptr ;
1.692 + movsxd rcx, dword ptr arg(4) ;Height ;
1.693 +
1.694 + pxor mm0, mm0 ;
1.695 + movq mm1, [rsi] ;
1.696 +
1.697 + movq mm3, [rsi+1] ;
1.698 + movq mm2, mm1 ;
1.699 +
1.700 + movq mm4, mm3 ;
1.701 + punpcklbw mm1, mm0 ;
1.702 +
1.703 + punpckhbw mm2, mm0 ;
1.704 + pmullw mm1, [rax] ;
1.705 +
1.706 + pmullw mm2, [rax] ;
1.707 + punpcklbw mm3, mm0 ;
1.708 +
1.709 + punpckhbw mm4, mm0 ;
1.710 + pmullw mm3, [rax+8] ;
1.711 +
1.712 + pmullw mm4, [rax+8] ;
1.713 + paddw mm1, mm3 ;
1.714 +
1.715 + paddw mm2, mm4 ;
1.716 + paddw mm1, [GLOBAL(mmx_bi_rd)] ;
1.717 +
1.718 + psraw mm1, mmx_filter_shift ;
1.719 + paddw mm2, [GLOBAL(mmx_bi_rd)] ;
1.720 +
1.721 + psraw mm2, mmx_filter_shift ;
1.722 + movq mm5, mm1
1.723 +
1.724 + packuswb mm5, mm2 ;
1.725 +%if ABI_IS_32BIT
1.726 + add rsi, dword ptr arg(1) ;ref_pixels_per_line
1.727 +%else
1.728 + movsxd r8, dword ptr arg(1) ;ref_pixels_per_line
1.729 + add rsi, r8
1.730 +%endif
1.731 +
1.732 +.filter_block2d_bil_var_mmx_loop:
1.733 +
1.734 + movq mm1, [rsi] ;
1.735 + movq mm3, [rsi+1] ;
1.736 +
1.737 + movq mm2, mm1 ;
1.738 + movq mm4, mm3 ;
1.739 +
1.740 + punpcklbw mm1, mm0 ;
1.741 + punpckhbw mm2, mm0 ;
1.742 +
1.743 + pmullw mm1, [rax] ;
1.744 + pmullw mm2, [rax] ;
1.745 +
1.746 + punpcklbw mm3, mm0 ;
1.747 + punpckhbw mm4, mm0 ;
1.748 +
1.749 + pmullw mm3, [rax+8] ;
1.750 + pmullw mm4, [rax+8] ;
1.751 +
1.752 + paddw mm1, mm3 ;
1.753 + paddw mm2, mm4 ;
1.754 +
1.755 + paddw mm1, [GLOBAL(mmx_bi_rd)] ;
1.756 + psraw mm1, mmx_filter_shift ;
1.757 +
1.758 + paddw mm2, [GLOBAL(mmx_bi_rd)] ;
1.759 + psraw mm2, mmx_filter_shift ;
1.760 +
1.761 + movq mm3, mm5 ;
1.762 + movq mm4, mm5 ;
1.763 +
1.764 + punpcklbw mm3, mm0 ;
1.765 + punpckhbw mm4, mm0 ;
1.766 +
1.767 + movq mm5, mm1 ;
1.768 + packuswb mm5, mm2 ;
1.769 +
1.770 + pmullw mm3, [rdx] ;
1.771 + pmullw mm4, [rdx] ;
1.772 +
1.773 + pmullw mm1, [rdx+8] ;
1.774 + pmullw mm2, [rdx+8] ;
1.775 +
1.776 + paddw mm1, mm3 ;
1.777 + paddw mm2, mm4 ;
1.778 +
1.779 + paddw mm1, [GLOBAL(mmx_bi_rd)] ;
1.780 + paddw mm2, [GLOBAL(mmx_bi_rd)] ;
1.781 +
1.782 + psraw mm1, mmx_filter_shift ;
1.783 + psraw mm2, mmx_filter_shift ;
1.784 +
1.785 + movq mm3, [rdi] ;
1.786 + movq mm4, mm3 ;
1.787 +
1.788 + punpcklbw mm3, mm0 ;
1.789 + punpckhbw mm4, mm0 ;
1.790 +
1.791 + psubw mm1, mm3 ;
1.792 + psubw mm2, mm4 ;
1.793 +
1.794 + paddw mm6, mm1 ;
1.795 + pmaddwd mm1, mm1 ;
1.796 +
1.797 + paddw mm6, mm2 ;
1.798 + pmaddwd mm2, mm2 ;
1.799 +
1.800 + paddd mm7, mm1 ;
1.801 + paddd mm7, mm2 ;
1.802 +
1.803 +%if ABI_IS_32BIT
1.804 + add rsi, dword ptr arg(1) ;ref_pixels_per_line ;
1.805 + add rdi, dword ptr arg(3) ;src_pixels_per_line ;
1.806 +%else
1.807 + movsxd r8, dword ptr arg(1) ;ref_pixels_per_line ;
1.808 + movsxd r9, dword ptr arg(3) ;src_pixels_per_line ;
1.809 + add rsi, r8
1.810 + add rdi, r9
1.811 +%endif
1.812 + sub rcx, 1 ;
1.813 + jnz .filter_block2d_bil_var_mmx_loop ;
1.814 +
1.815 +
1.816 + pxor mm3, mm3 ;
1.817 + pxor mm2, mm2 ;
1.818 +
1.819 + punpcklwd mm2, mm6 ;
1.820 + punpckhwd mm3, mm6 ;
1.821 +
1.822 + paddd mm2, mm3 ;
1.823 + movq mm6, mm2 ;
1.824 +
1.825 + psrlq mm6, 32 ;
1.826 + paddd mm2, mm6 ;
1.827 +
1.828 + psrad mm2, 16 ;
1.829 + movq mm4, mm7 ;
1.830 +
1.831 + psrlq mm4, 32 ;
1.832 + paddd mm4, mm7 ;
1.833 +
1.834 + mov rdi, arg(7) ;sum
1.835 + mov rsi, arg(8) ;sumsquared
1.836 +
1.837 + movd dword ptr [rdi], mm2 ;
1.838 + movd dword ptr [rsi], mm4 ;
1.839 +
1.840 + ; begin epilog
1.841 + add rsp, 16
1.842 + pop rdi
1.843 + pop rsi
1.844 + RESTORE_GOT
1.845 + UNSHADOW_ARGS
1.846 + pop rbp
1.847 + ret
1.848 +
1.849 +
1.850 +SECTION_RODATA
1.851 +;short mmx_bi_rd[4] = { 64, 64, 64, 64};
1.852 +align 16
1.853 +mmx_bi_rd:
1.854 + times 4 dw 64
