1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/common/x86/sad_sse2.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,410 @@
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_sad16x16_wmt(
1.18 +; unsigned char *src_ptr,
1.19 +; int src_stride,
1.20 +; unsigned char *ref_ptr,
1.21 +; int ref_stride)
1.22 +global sym(vp8_sad16x16_wmt) PRIVATE
1.23 +sym(vp8_sad16x16_wmt):
1.24 + push rbp
1.25 + mov rbp, rsp
1.26 + SHADOW_ARGS_TO_STACK 4
1.27 + SAVE_XMM 6
1.28 + push rsi
1.29 + push rdi
1.30 + ; end prolog
1.31 +
1.32 + mov rsi, arg(0) ;src_ptr
1.33 + mov rdi, arg(2) ;ref_ptr
1.34 +
1.35 + movsxd rax, dword ptr arg(1) ;src_stride
1.36 + movsxd rdx, dword ptr arg(3) ;ref_stride
1.37 +
1.38 + lea rcx, [rsi+rax*8]
1.39 +
1.40 + lea rcx, [rcx+rax*8]
1.41 + pxor xmm6, xmm6
1.42 +
1.43 +.x16x16sad_wmt_loop:
1.44 +
1.45 + movq xmm0, QWORD PTR [rsi]
1.46 + movq xmm2, QWORD PTR [rsi+8]
1.47 +
1.48 + movq xmm1, QWORD PTR [rdi]
1.49 + movq xmm3, QWORD PTR [rdi+8]
1.50 +
1.51 + movq xmm4, QWORD PTR [rsi+rax]
1.52 + movq xmm5, QWORD PTR [rdi+rdx]
1.53 +
1.54 +
1.55 + punpcklbw xmm0, xmm2
1.56 + punpcklbw xmm1, xmm3
1.57 +
1.58 + psadbw xmm0, xmm1
1.59 + movq xmm2, QWORD PTR [rsi+rax+8]
1.60 +
1.61 + movq xmm3, QWORD PTR [rdi+rdx+8]
1.62 + lea rsi, [rsi+rax*2]
1.63 +
1.64 + lea rdi, [rdi+rdx*2]
1.65 + punpcklbw xmm4, xmm2
1.66 +
1.67 + punpcklbw xmm5, xmm3
1.68 + psadbw xmm4, xmm5
1.69 +
1.70 + paddw xmm6, xmm0
1.71 + paddw xmm6, xmm4
1.72 +
1.73 + cmp rsi, rcx
1.74 + jne .x16x16sad_wmt_loop
1.75 +
1.76 + movq xmm0, xmm6
1.77 + psrldq xmm6, 8
1.78 +
1.79 + paddw xmm0, xmm6
1.80 + movq rax, xmm0
1.81 +
1.82 + ; begin epilog
1.83 + pop rdi
1.84 + pop rsi
1.85 + RESTORE_XMM
1.86 + UNSHADOW_ARGS
1.87 + pop rbp
1.88 + ret
1.89 +
1.90 +;unsigned int vp8_sad8x16_wmt(
1.91 +; unsigned char *src_ptr,
1.92 +; int src_stride,
1.93 +; unsigned char *ref_ptr,
1.94 +; int ref_stride,
1.95 +; int max_sad)
1.96 +global sym(vp8_sad8x16_wmt) PRIVATE
1.97 +sym(vp8_sad8x16_wmt):
1.98 + push rbp
1.99 + mov rbp, rsp
1.100 + SHADOW_ARGS_TO_STACK 5
1.101 + push rbx
1.102 + push rsi
1.103 + push rdi
1.104 + ; end prolog
1.105 +
1.106 + mov rsi, arg(0) ;src_ptr
1.107 + mov rdi, arg(2) ;ref_ptr
1.108 +
1.109 + movsxd rbx, dword ptr arg(1) ;src_stride
1.110 + movsxd rdx, dword ptr arg(3) ;ref_stride
1.111 +
1.112 + lea rcx, [rsi+rbx*8]
1.113 +
1.114 + lea rcx, [rcx+rbx*8]
1.115 + pxor mm7, mm7
1.116 +
1.117 +.x8x16sad_wmt_loop:
1.118 +
1.119 + movq rax, mm7
1.120 + cmp eax, arg(4)
1.121 + ja .x8x16sad_wmt_early_exit
1.122 +
1.123 + movq mm0, QWORD PTR [rsi]
1.124 + movq mm1, QWORD PTR [rdi]
1.125 +
1.126 + movq mm2, QWORD PTR [rsi+rbx]
1.127 + movq mm3, QWORD PTR [rdi+rdx]
1.128 +
1.129 + psadbw mm0, mm1
1.130 + psadbw mm2, mm3
1.131 +
1.132 + lea rsi, [rsi+rbx*2]
1.133 + lea rdi, [rdi+rdx*2]
1.134 +
1.135 + paddw mm7, mm0
1.136 + paddw mm7, mm2
1.137 +
1.138 + cmp rsi, rcx
1.139 + jne .x8x16sad_wmt_loop
1.140 +
1.141 + movq rax, mm7
1.142 +
1.143 +.x8x16sad_wmt_early_exit:
1.144 +
1.145 + ; begin epilog
1.146 + pop rdi
1.147 + pop rsi
1.148 + pop rbx
1.149 + UNSHADOW_ARGS
1.150 + pop rbp
1.151 + ret
1.152 +
1.153 +
1.154 +;unsigned int vp8_sad8x8_wmt(
1.155 +; unsigned char *src_ptr,
1.156 +; int src_stride,
1.157 +; unsigned char *ref_ptr,
1.158 +; int ref_stride)
1.159 +global sym(vp8_sad8x8_wmt) PRIVATE
1.160 +sym(vp8_sad8x8_wmt):
1.161 + push rbp
1.162 + mov rbp, rsp
1.163 + SHADOW_ARGS_TO_STACK 5
1.164 + push rbx
1.165 + push rsi
1.166 + push rdi
1.167 + ; end prolog
1.168 +
1.169 + mov rsi, arg(0) ;src_ptr
1.170 + mov rdi, arg(2) ;ref_ptr
1.171 +
1.172 + movsxd rbx, dword ptr arg(1) ;src_stride
1.173 + movsxd rdx, dword ptr arg(3) ;ref_stride
1.174 +
1.175 + lea rcx, [rsi+rbx*8]
1.176 + pxor mm7, mm7
1.177 +
1.178 +.x8x8sad_wmt_loop:
1.179 +
1.180 + movq rax, mm7
1.181 + cmp eax, arg(4)
1.182 + ja .x8x8sad_wmt_early_exit
1.183 +
1.184 + movq mm0, QWORD PTR [rsi]
1.185 + movq mm1, QWORD PTR [rdi]
1.186 +
1.187 + psadbw mm0, mm1
1.188 + lea rsi, [rsi+rbx]
1.189 +
1.190 + add rdi, rdx
1.191 + paddw mm7, mm0
1.192 +
1.193 + cmp rsi, rcx
1.194 + jne .x8x8sad_wmt_loop
1.195 +
1.196 + movq rax, mm7
1.197 +.x8x8sad_wmt_early_exit:
1.198 +
1.199 + ; begin epilog
1.200 + pop rdi
1.201 + pop rsi
1.202 + pop rbx
1.203 + UNSHADOW_ARGS
1.204 + pop rbp
1.205 + ret
1.206 +
1.207 +;unsigned int vp8_sad4x4_wmt(
1.208 +; unsigned char *src_ptr,
1.209 +; int src_stride,
1.210 +; unsigned char *ref_ptr,
1.211 +; int ref_stride)
1.212 +global sym(vp8_sad4x4_wmt) PRIVATE
1.213 +sym(vp8_sad4x4_wmt):
1.214 + push rbp
1.215 + mov rbp, rsp
1.216 + SHADOW_ARGS_TO_STACK 4
1.217 + push rsi
1.218 + push rdi
1.219 + ; end prolog
1.220 +
1.221 + mov rsi, arg(0) ;src_ptr
1.222 + mov rdi, arg(2) ;ref_ptr
1.223 +
1.224 + movsxd rax, dword ptr arg(1) ;src_stride
1.225 + movsxd rdx, dword ptr arg(3) ;ref_stride
1.226 +
1.227 + movd mm0, DWORD PTR [rsi]
1.228 + movd mm1, DWORD PTR [rdi]
1.229 +
1.230 + movd mm2, DWORD PTR [rsi+rax]
1.231 + movd mm3, DWORD PTR [rdi+rdx]
1.232 +
1.233 + punpcklbw mm0, mm2
1.234 + punpcklbw mm1, mm3
1.235 +
1.236 + psadbw mm0, mm1
1.237 + lea rsi, [rsi+rax*2]
1.238 +
1.239 + lea rdi, [rdi+rdx*2]
1.240 + movd mm4, DWORD PTR [rsi]
1.241 +
1.242 + movd mm5, DWORD PTR [rdi]
1.243 + movd mm6, DWORD PTR [rsi+rax]
1.244 +
1.245 + movd mm7, DWORD PTR [rdi+rdx]
1.246 + punpcklbw mm4, mm6
1.247 +
1.248 + punpcklbw mm5, mm7
1.249 + psadbw mm4, mm5
1.250 +
1.251 + paddw mm0, mm4
1.252 + movq rax, mm0
1.253 +
1.254 + ; begin epilog
1.255 + pop rdi
1.256 + pop rsi
1.257 + UNSHADOW_ARGS
1.258 + pop rbp
1.259 + ret
1.260 +
1.261 +
1.262 +;unsigned int vp8_sad16x8_wmt(
1.263 +; unsigned char *src_ptr,
1.264 +; int src_stride,
1.265 +; unsigned char *ref_ptr,
1.266 +; int ref_stride)
1.267 +global sym(vp8_sad16x8_wmt) PRIVATE
1.268 +sym(vp8_sad16x8_wmt):
1.269 + push rbp
1.270 + mov rbp, rsp
1.271 + SHADOW_ARGS_TO_STACK 5
1.272 + push rbx
1.273 + push rsi
1.274 + push rdi
1.275 + ; end prolog
1.276 +
1.277 +
1.278 + mov rsi, arg(0) ;src_ptr
1.279 + mov rdi, arg(2) ;ref_ptr
1.280 +
1.281 + movsxd rbx, dword ptr arg(1) ;src_stride
1.282 + movsxd rdx, dword ptr arg(3) ;ref_stride
1.283 +
1.284 + lea rcx, [rsi+rbx*8]
1.285 + pxor mm7, mm7
1.286 +
1.287 +.x16x8sad_wmt_loop:
1.288 +
1.289 + movq rax, mm7
1.290 + cmp eax, arg(4)
1.291 + ja .x16x8sad_wmt_early_exit
1.292 +
1.293 + movq mm0, QWORD PTR [rsi]
1.294 + movq mm2, QWORD PTR [rsi+8]
1.295 +
1.296 + movq mm1, QWORD PTR [rdi]
1.297 + movq mm3, QWORD PTR [rdi+8]
1.298 +
1.299 + movq mm4, QWORD PTR [rsi+rbx]
1.300 + movq mm5, QWORD PTR [rdi+rdx]
1.301 +
1.302 + psadbw mm0, mm1
1.303 + psadbw mm2, mm3
1.304 +
1.305 + movq mm1, QWORD PTR [rsi+rbx+8]
1.306 + movq mm3, QWORD PTR [rdi+rdx+8]
1.307 +
1.308 + psadbw mm4, mm5
1.309 + psadbw mm1, mm3
1.310 +
1.311 + lea rsi, [rsi+rbx*2]
1.312 + lea rdi, [rdi+rdx*2]
1.313 +
1.314 + paddw mm0, mm2
1.315 + paddw mm4, mm1
1.316 +
1.317 + paddw mm7, mm0
1.318 + paddw mm7, mm4
1.319 +
1.320 + cmp rsi, rcx
1.321 + jne .x16x8sad_wmt_loop
1.322 +
1.323 + movq rax, mm7
1.324 +
1.325 +.x16x8sad_wmt_early_exit:
1.326 +
1.327 + ; begin epilog
1.328 + pop rdi
1.329 + pop rsi
1.330 + pop rbx
1.331 + UNSHADOW_ARGS
1.332 + pop rbp
1.333 + ret
1.334 +
1.335 +;void vp8_copy32xn_sse2(
1.336 +; unsigned char *src_ptr,
1.337 +; int src_stride,
1.338 +; unsigned char *dst_ptr,
1.339 +; int dst_stride,
1.340 +; int height);
1.341 +global sym(vp8_copy32xn_sse2) PRIVATE
1.342 +sym(vp8_copy32xn_sse2):
1.343 + push rbp
1.344 + mov rbp, rsp
1.345 + SHADOW_ARGS_TO_STACK 5
1.346 + SAVE_XMM 7
1.347 + push rsi
1.348 + push rdi
1.349 + ; end prolog
1.350 +
1.351 + mov rsi, arg(0) ;src_ptr
1.352 + mov rdi, arg(2) ;dst_ptr
1.353 +
1.354 + movsxd rax, dword ptr arg(1) ;src_stride
1.355 + movsxd rdx, dword ptr arg(3) ;dst_stride
1.356 + movsxd rcx, dword ptr arg(4) ;height
1.357 +
1.358 +.block_copy_sse2_loopx4:
1.359 + movdqu xmm0, XMMWORD PTR [rsi]
1.360 + movdqu xmm1, XMMWORD PTR [rsi + 16]
1.361 + movdqu xmm2, XMMWORD PTR [rsi + rax]
1.362 + movdqu xmm3, XMMWORD PTR [rsi + rax + 16]
1.363 +
1.364 + lea rsi, [rsi+rax*2]
1.365 +
1.366 + movdqu xmm4, XMMWORD PTR [rsi]
1.367 + movdqu xmm5, XMMWORD PTR [rsi + 16]
1.368 + movdqu xmm6, XMMWORD PTR [rsi + rax]
1.369 + movdqu xmm7, XMMWORD PTR [rsi + rax + 16]
1.370 +
1.371 + lea rsi, [rsi+rax*2]
1.372 +
1.373 + movdqa XMMWORD PTR [rdi], xmm0
1.374 + movdqa XMMWORD PTR [rdi + 16], xmm1
1.375 + movdqa XMMWORD PTR [rdi + rdx], xmm2
1.376 + movdqa XMMWORD PTR [rdi + rdx + 16], xmm3
1.377 +
1.378 + lea rdi, [rdi+rdx*2]
1.379 +
1.380 + movdqa XMMWORD PTR [rdi], xmm4
1.381 + movdqa XMMWORD PTR [rdi + 16], xmm5
1.382 + movdqa XMMWORD PTR [rdi + rdx], xmm6
1.383 + movdqa XMMWORD PTR [rdi + rdx + 16], xmm7
1.384 +
1.385 + lea rdi, [rdi+rdx*2]
1.386 +
1.387 + sub rcx, 4
1.388 + cmp rcx, 4
1.389 + jge .block_copy_sse2_loopx4
1.390 +
1.391 + cmp rcx, 0
1.392 + je .copy_is_done
1.393 +
1.394 +.block_copy_sse2_loop:
1.395 + movdqu xmm0, XMMWORD PTR [rsi]
1.396 + movdqu xmm1, XMMWORD PTR [rsi + 16]
1.397 + lea rsi, [rsi+rax]
1.398 +
1.399 + movdqa XMMWORD PTR [rdi], xmm0
1.400 + movdqa XMMWORD PTR [rdi + 16], xmm1
1.401 + lea rdi, [rdi+rdx]
1.402 +
1.403 + sub rcx, 1
1.404 + jne .block_copy_sse2_loop
1.405 +
1.406 +.copy_is_done:
1.407 + ; begin epilog
1.408 + pop rdi
1.409 + pop rsi
1.410 + RESTORE_XMM
1.411 + UNSHADOW_ARGS
1.412 + pop rbp
1.413 + ret
