1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/encoder/x86/encodeopt.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,386 @@
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 +;int vp8_block_error_xmm(short *coeff_ptr, short *dcoef_ptr)
1.18 +global sym(vp8_block_error_xmm) PRIVATE
1.19 +sym(vp8_block_error_xmm):
1.20 + push rbp
1.21 + mov rbp, rsp
1.22 + SHADOW_ARGS_TO_STACK 2
1.23 + push rsi
1.24 + push rdi
1.25 + ; end prologue
1.26 +
1.27 + mov rsi, arg(0) ;coeff_ptr
1.28 + mov rdi, arg(1) ;dcoef_ptr
1.29 +
1.30 + movdqa xmm0, [rsi]
1.31 + movdqa xmm1, [rdi]
1.32 +
1.33 + movdqa xmm2, [rsi+16]
1.34 + movdqa xmm3, [rdi+16]
1.35 +
1.36 + psubw xmm0, xmm1
1.37 + psubw xmm2, xmm3
1.38 +
1.39 + pmaddwd xmm0, xmm0
1.40 + pmaddwd xmm2, xmm2
1.41 +
1.42 + paddd xmm0, xmm2
1.43 +
1.44 + pxor xmm5, xmm5
1.45 + movdqa xmm1, xmm0
1.46 +
1.47 + punpckldq xmm0, xmm5
1.48 + punpckhdq xmm1, xmm5
1.49 +
1.50 + paddd xmm0, xmm1
1.51 + movdqa xmm1, xmm0
1.52 +
1.53 + psrldq xmm0, 8
1.54 + paddd xmm0, xmm1
1.55 +
1.56 + movq rax, xmm0
1.57 +
1.58 + pop rdi
1.59 + pop rsi
1.60 + ; begin epilog
1.61 + UNSHADOW_ARGS
1.62 + pop rbp
1.63 + ret
1.64 +
1.65 +;int vp8_block_error_mmx(short *coeff_ptr, short *dcoef_ptr)
1.66 +global sym(vp8_block_error_mmx) PRIVATE
1.67 +sym(vp8_block_error_mmx):
1.68 + push rbp
1.69 + mov rbp, rsp
1.70 + SHADOW_ARGS_TO_STACK 2
1.71 + push rsi
1.72 + push rdi
1.73 + ; end prolog
1.74 +
1.75 +
1.76 + mov rsi, arg(0) ;coeff_ptr
1.77 + pxor mm7, mm7
1.78 +
1.79 + mov rdi, arg(1) ;dcoef_ptr
1.80 + movq mm3, [rsi]
1.81 +
1.82 + movq mm4, [rdi]
1.83 + movq mm5, [rsi+8]
1.84 +
1.85 + movq mm6, [rdi+8]
1.86 + pxor mm1, mm1 ; from movd mm1, dc ; dc =0
1.87 +
1.88 + movq mm2, mm7
1.89 + psubw mm5, mm6
1.90 +
1.91 + por mm1, mm2
1.92 + pmaddwd mm5, mm5
1.93 +
1.94 + pcmpeqw mm1, mm7
1.95 + psubw mm3, mm4
1.96 +
1.97 + pand mm1, mm3
1.98 + pmaddwd mm1, mm1
1.99 +
1.100 + paddd mm1, mm5
1.101 + movq mm3, [rsi+16]
1.102 +
1.103 + movq mm4, [rdi+16]
1.104 + movq mm5, [rsi+24]
1.105 +
1.106 + movq mm6, [rdi+24]
1.107 + psubw mm5, mm6
1.108 +
1.109 + pmaddwd mm5, mm5
1.110 + psubw mm3, mm4
1.111 +
1.112 + pmaddwd mm3, mm3
1.113 + paddd mm3, mm5
1.114 +
1.115 + paddd mm1, mm3
1.116 + movq mm0, mm1
1.117 +
1.118 + psrlq mm1, 32
1.119 + paddd mm0, mm1
1.120 +
1.121 + movq rax, mm0
1.122 +
1.123 + pop rdi
1.124 + pop rsi
1.125 + ; begin epilog
1.126 + UNSHADOW_ARGS
1.127 + pop rbp
1.128 + ret
1.129 +
1.130 +
1.131 +;int vp8_mbblock_error_mmx_impl(short *coeff_ptr, short *dcoef_ptr, int dc);
1.132 +global sym(vp8_mbblock_error_mmx_impl) PRIVATE
1.133 +sym(vp8_mbblock_error_mmx_impl):
1.134 + push rbp
1.135 + mov rbp, rsp
1.136 + SHADOW_ARGS_TO_STACK 3
1.137 + push rsi
1.138 + push rdi
1.139 + ; end prolog
1.140 +
1.141 +
1.142 + mov rsi, arg(0) ;coeff_ptr
1.143 + pxor mm7, mm7
1.144 +
1.145 + mov rdi, arg(1) ;dcoef_ptr
1.146 + pxor mm2, mm2
1.147 +
1.148 + movd mm1, dword ptr arg(2) ;dc
1.149 + por mm1, mm2
1.150 +
1.151 + pcmpeqw mm1, mm7
1.152 + mov rcx, 16
1.153 +
1.154 +.mberror_loop_mmx:
1.155 + movq mm3, [rsi]
1.156 + movq mm4, [rdi]
1.157 +
1.158 + movq mm5, [rsi+8]
1.159 + movq mm6, [rdi+8]
1.160 +
1.161 +
1.162 + psubw mm5, mm6
1.163 + pmaddwd mm5, mm5
1.164 +
1.165 + psubw mm3, mm4
1.166 + pand mm3, mm1
1.167 +
1.168 + pmaddwd mm3, mm3
1.169 + paddd mm2, mm5
1.170 +
1.171 + paddd mm2, mm3
1.172 + movq mm3, [rsi+16]
1.173 +
1.174 + movq mm4, [rdi+16]
1.175 + movq mm5, [rsi+24]
1.176 +
1.177 + movq mm6, [rdi+24]
1.178 + psubw mm5, mm6
1.179 +
1.180 + pmaddwd mm5, mm5
1.181 + psubw mm3, mm4
1.182 +
1.183 + pmaddwd mm3, mm3
1.184 + paddd mm2, mm5
1.185 +
1.186 + paddd mm2, mm3
1.187 + add rsi, 32
1.188 +
1.189 + add rdi, 32
1.190 + sub rcx, 1
1.191 +
1.192 + jnz .mberror_loop_mmx
1.193 +
1.194 + movq mm0, mm2
1.195 + psrlq mm2, 32
1.196 +
1.197 + paddd mm0, mm2
1.198 + movq rax, mm0
1.199 +
1.200 + pop rdi
1.201 + pop rsi
1.202 + ; begin epilog
1.203 + UNSHADOW_ARGS
1.204 + pop rbp
1.205 + ret
1.206 +
1.207 +
1.208 +;int vp8_mbblock_error_xmm_impl(short *coeff_ptr, short *dcoef_ptr, int dc);
1.209 +global sym(vp8_mbblock_error_xmm_impl) PRIVATE
1.210 +sym(vp8_mbblock_error_xmm_impl):
1.211 + push rbp
1.212 + mov rbp, rsp
1.213 + SHADOW_ARGS_TO_STACK 3
1.214 + SAVE_XMM 6
1.215 + push rsi
1.216 + push rdi
1.217 + ; end prolog
1.218 +
1.219 +
1.220 + mov rsi, arg(0) ;coeff_ptr
1.221 + pxor xmm6, xmm6
1.222 +
1.223 + mov rdi, arg(1) ;dcoef_ptr
1.224 + pxor xmm4, xmm4
1.225 +
1.226 + movd xmm5, dword ptr arg(2) ;dc
1.227 + por xmm5, xmm4
1.228 +
1.229 + pcmpeqw xmm5, xmm6
1.230 + mov rcx, 16
1.231 +
1.232 +.mberror_loop:
1.233 + movdqa xmm0, [rsi]
1.234 + movdqa xmm1, [rdi]
1.235 +
1.236 + movdqa xmm2, [rsi+16]
1.237 + movdqa xmm3, [rdi+16]
1.238 +
1.239 +
1.240 + psubw xmm2, xmm3
1.241 + pmaddwd xmm2, xmm2
1.242 +
1.243 + psubw xmm0, xmm1
1.244 + pand xmm0, xmm5
1.245 +
1.246 + pmaddwd xmm0, xmm0
1.247 + add rsi, 32
1.248 +
1.249 + add rdi, 32
1.250 +
1.251 + sub rcx, 1
1.252 + paddd xmm4, xmm2
1.253 +
1.254 + paddd xmm4, xmm0
1.255 + jnz .mberror_loop
1.256 +
1.257 + movdqa xmm0, xmm4
1.258 + punpckldq xmm0, xmm6
1.259 +
1.260 + punpckhdq xmm4, xmm6
1.261 + paddd xmm0, xmm4
1.262 +
1.263 + movdqa xmm1, xmm0
1.264 + psrldq xmm0, 8
1.265 +
1.266 + paddd xmm0, xmm1
1.267 + movq rax, xmm0
1.268 +
1.269 + pop rdi
1.270 + pop rsi
1.271 + ; begin epilog
1.272 + RESTORE_XMM
1.273 + UNSHADOW_ARGS
1.274 + pop rbp
1.275 + ret
1.276 +
1.277 +
1.278 +;int vp8_mbuverror_mmx_impl(short *s_ptr, short *d_ptr);
1.279 +global sym(vp8_mbuverror_mmx_impl) PRIVATE
1.280 +sym(vp8_mbuverror_mmx_impl):
1.281 + push rbp
1.282 + mov rbp, rsp
1.283 + SHADOW_ARGS_TO_STACK 2
1.284 + push rsi
1.285 + push rdi
1.286 + ; end prolog
1.287 +
1.288 +
1.289 + mov rsi, arg(0) ;s_ptr
1.290 + mov rdi, arg(1) ;d_ptr
1.291 +
1.292 + mov rcx, 16
1.293 + pxor mm7, mm7
1.294 +
1.295 +.mbuverror_loop_mmx:
1.296 +
1.297 + movq mm1, [rsi]
1.298 + movq mm2, [rdi]
1.299 +
1.300 + psubw mm1, mm2
1.301 + pmaddwd mm1, mm1
1.302 +
1.303 +
1.304 + movq mm3, [rsi+8]
1.305 + movq mm4, [rdi+8]
1.306 +
1.307 + psubw mm3, mm4
1.308 + pmaddwd mm3, mm3
1.309 +
1.310 +
1.311 + paddd mm7, mm1
1.312 + paddd mm7, mm3
1.313 +
1.314 +
1.315 + add rsi, 16
1.316 + add rdi, 16
1.317 +
1.318 + dec rcx
1.319 + jnz .mbuverror_loop_mmx
1.320 +
1.321 + movq mm0, mm7
1.322 + psrlq mm7, 32
1.323 +
1.324 + paddd mm0, mm7
1.325 + movq rax, mm0
1.326 +
1.327 + pop rdi
1.328 + pop rsi
1.329 + ; begin epilog
1.330 + UNSHADOW_ARGS
1.331 + pop rbp
1.332 + ret
1.333 +
1.334 +
1.335 +;int vp8_mbuverror_xmm_impl(short *s_ptr, short *d_ptr);
1.336 +global sym(vp8_mbuverror_xmm_impl) PRIVATE
1.337 +sym(vp8_mbuverror_xmm_impl):
1.338 + push rbp
1.339 + mov rbp, rsp
1.340 + SHADOW_ARGS_TO_STACK 2
1.341 + push rsi
1.342 + push rdi
1.343 + ; end prolog
1.344 +
1.345 +
1.346 + mov rsi, arg(0) ;s_ptr
1.347 + mov rdi, arg(1) ;d_ptr
1.348 +
1.349 + mov rcx, 16
1.350 + pxor xmm3, xmm3
1.351 +
1.352 +.mbuverror_loop:
1.353 +
1.354 + movdqa xmm1, [rsi]
1.355 + movdqa xmm2, [rdi]
1.356 +
1.357 + psubw xmm1, xmm2
1.358 + pmaddwd xmm1, xmm1
1.359 +
1.360 + paddd xmm3, xmm1
1.361 +
1.362 + add rsi, 16
1.363 + add rdi, 16
1.364 +
1.365 + dec rcx
1.366 + jnz .mbuverror_loop
1.367 +
1.368 + pxor xmm0, xmm0
1.369 + movdqa xmm1, xmm3
1.370 +
1.371 + movdqa xmm2, xmm1
1.372 + punpckldq xmm1, xmm0
1.373 +
1.374 + punpckhdq xmm2, xmm0
1.375 + paddd xmm1, xmm2
1.376 +
1.377 + movdqa xmm2, xmm1
1.378 +
1.379 + psrldq xmm1, 8
1.380 + paddd xmm1, xmm2
1.381 +
1.382 + movq rax, xmm1
1.383 +
1.384 + pop rdi
1.385 + pop rsi
1.386 + ; begin epilog
1.387 + UNSHADOW_ARGS
1.388 + pop rbp
1.389 + ret
