1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/encoder/x86/dct_mmx.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,241 @@
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 +;void vp8_short_fdct4x4_mmx(short *input, short *output, int pitch)
1.18 +global sym(vp8_short_fdct4x4_mmx) PRIVATE
1.19 +sym(vp8_short_fdct4x4_mmx):
1.20 + push rbp
1.21 + mov rbp, rsp
1.22 + SHADOW_ARGS_TO_STACK 3
1.23 + GET_GOT rbx
1.24 + push rsi
1.25 + push rdi
1.26 + ; end prolog
1.27 +
1.28 + mov rsi, arg(0) ; input
1.29 + mov rdi, arg(1) ; output
1.30 +
1.31 + movsxd rax, dword ptr arg(2) ;pitch
1.32 +
1.33 + lea rcx, [rsi + rax*2]
1.34 + ; read the input data
1.35 + movq mm0, [rsi]
1.36 + movq mm1, [rsi + rax]
1.37 +
1.38 + movq mm2, [rcx]
1.39 + movq mm4, [rcx + rax]
1.40 +
1.41 + ; transpose for the first stage
1.42 + movq mm3, mm0 ; 00 01 02 03
1.43 + movq mm5, mm2 ; 20 21 22 23
1.44 +
1.45 + punpcklwd mm0, mm1 ; 00 10 01 11
1.46 + punpckhwd mm3, mm1 ; 02 12 03 13
1.47 +
1.48 + punpcklwd mm2, mm4 ; 20 30 21 31
1.49 + punpckhwd mm5, mm4 ; 22 32 23 33
1.50 +
1.51 + movq mm1, mm0 ; 00 10 01 11
1.52 + punpckldq mm0, mm2 ; 00 10 20 30
1.53 +
1.54 + punpckhdq mm1, mm2 ; 01 11 21 31
1.55 +
1.56 + movq mm2, mm3 ; 02 12 03 13
1.57 + punpckldq mm2, mm5 ; 02 12 22 32
1.58 +
1.59 + punpckhdq mm3, mm5 ; 03 13 23 33
1.60 +
1.61 + ; mm0 0
1.62 + ; mm1 1
1.63 + ; mm2 2
1.64 + ; mm3 3
1.65 +
1.66 + ; first stage
1.67 + movq mm5, mm0
1.68 + movq mm4, mm1
1.69 +
1.70 + paddw mm0, mm3 ; a1 = 0 + 3
1.71 + paddw mm1, mm2 ; b1 = 1 + 2
1.72 +
1.73 + psubw mm4, mm2 ; c1 = 1 - 2
1.74 + psubw mm5, mm3 ; d1 = 0 - 3
1.75 +
1.76 + psllw mm5, 3
1.77 + psllw mm4, 3
1.78 +
1.79 + psllw mm0, 3
1.80 + psllw mm1, 3
1.81 +
1.82 + ; output 0 and 2
1.83 + movq mm2, mm0 ; a1
1.84 +
1.85 + paddw mm0, mm1 ; op[0] = a1 + b1
1.86 + psubw mm2, mm1 ; op[2] = a1 - b1
1.87 +
1.88 + ; output 1 and 3
1.89 + ; interleave c1, d1
1.90 + movq mm1, mm5 ; d1
1.91 + punpcklwd mm1, mm4 ; c1 d1
1.92 + punpckhwd mm5, mm4 ; c1 d1
1.93 +
1.94 + movq mm3, mm1
1.95 + movq mm4, mm5
1.96 +
1.97 + pmaddwd mm1, MMWORD PTR[GLOBAL (_5352_2217)] ; c1*2217 + d1*5352
1.98 + pmaddwd mm4, MMWORD PTR[GLOBAL (_5352_2217)] ; c1*2217 + d1*5352
1.99 +
1.100 + pmaddwd mm3, MMWORD PTR[GLOBAL(_2217_neg5352)] ; d1*2217 - c1*5352
1.101 + pmaddwd mm5, MMWORD PTR[GLOBAL(_2217_neg5352)] ; d1*2217 - c1*5352
1.102 +
1.103 + paddd mm1, MMWORD PTR[GLOBAL(_14500)]
1.104 + paddd mm4, MMWORD PTR[GLOBAL(_14500)]
1.105 + paddd mm3, MMWORD PTR[GLOBAL(_7500)]
1.106 + paddd mm5, MMWORD PTR[GLOBAL(_7500)]
1.107 +
1.108 + psrad mm1, 12 ; (c1 * 2217 + d1 * 5352 + 14500)>>12
1.109 + psrad mm4, 12 ; (c1 * 2217 + d1 * 5352 + 14500)>>12
1.110 + psrad mm3, 12 ; (d1 * 2217 - c1 * 5352 + 7500)>>12
1.111 + psrad mm5, 12 ; (d1 * 2217 - c1 * 5352 + 7500)>>12
1.112 +
1.113 + packssdw mm1, mm4 ; op[1]
1.114 + packssdw mm3, mm5 ; op[3]
1.115 +
1.116 + ; done with vertical
1.117 + ; transpose for the second stage
1.118 + movq mm4, mm0 ; 00 10 20 30
1.119 + movq mm5, mm2 ; 02 12 22 32
1.120 +
1.121 + punpcklwd mm0, mm1 ; 00 01 10 11
1.122 + punpckhwd mm4, mm1 ; 20 21 30 31
1.123 +
1.124 + punpcklwd mm2, mm3 ; 02 03 12 13
1.125 + punpckhwd mm5, mm3 ; 22 23 32 33
1.126 +
1.127 + movq mm1, mm0 ; 00 01 10 11
1.128 + punpckldq mm0, mm2 ; 00 01 02 03
1.129 +
1.130 + punpckhdq mm1, mm2 ; 01 22 12 13
1.131 +
1.132 + movq mm2, mm4 ; 20 31 30 31
1.133 + punpckldq mm2, mm5 ; 20 21 22 23
1.134 +
1.135 + punpckhdq mm4, mm5 ; 30 31 32 33
1.136 +
1.137 + ; mm0 0
1.138 + ; mm1 1
1.139 + ; mm2 2
1.140 + ; mm3 4
1.141 +
1.142 + movq mm5, mm0
1.143 + movq mm3, mm1
1.144 +
1.145 + paddw mm0, mm4 ; a1 = 0 + 3
1.146 + paddw mm1, mm2 ; b1 = 1 + 2
1.147 +
1.148 + psubw mm3, mm2 ; c1 = 1 - 2
1.149 + psubw mm5, mm4 ; d1 = 0 - 3
1.150 +
1.151 + pxor mm6, mm6 ; zero out for compare
1.152 +
1.153 + pcmpeqw mm6, mm5 ; d1 != 0
1.154 +
1.155 + pandn mm6, MMWORD PTR[GLOBAL(_cmp_mask)] ; clear upper,
1.156 + ; and keep bit 0 of lower
1.157 +
1.158 + ; output 0 and 2
1.159 + movq mm2, mm0 ; a1
1.160 +
1.161 + paddw mm0, mm1 ; a1 + b1
1.162 + psubw mm2, mm1 ; a1 - b1
1.163 +
1.164 + paddw mm0, MMWORD PTR[GLOBAL(_7w)]
1.165 + paddw mm2, MMWORD PTR[GLOBAL(_7w)]
1.166 +
1.167 + psraw mm0, 4 ; op[0] = (a1 + b1 + 7)>>4
1.168 + psraw mm2, 4 ; op[8] = (a1 - b1 + 7)>>4
1.169 +
1.170 + movq MMWORD PTR[rdi + 0 ], mm0
1.171 + movq MMWORD PTR[rdi + 16], mm2
1.172 +
1.173 + ; output 1 and 3
1.174 + ; interleave c1, d1
1.175 + movq mm1, mm5 ; d1
1.176 + punpcklwd mm1, mm3 ; c1 d1
1.177 + punpckhwd mm5, mm3 ; c1 d1
1.178 +
1.179 + movq mm3, mm1
1.180 + movq mm4, mm5
1.181 +
1.182 + pmaddwd mm1, MMWORD PTR[GLOBAL (_5352_2217)] ; c1*2217 + d1*5352
1.183 + pmaddwd mm4, MMWORD PTR[GLOBAL (_5352_2217)] ; c1*2217 + d1*5352
1.184 +
1.185 + pmaddwd mm3, MMWORD PTR[GLOBAL(_2217_neg5352)] ; d1*2217 - c1*5352
1.186 + pmaddwd mm5, MMWORD PTR[GLOBAL(_2217_neg5352)] ; d1*2217 - c1*5352
1.187 +
1.188 + paddd mm1, MMWORD PTR[GLOBAL(_12000)]
1.189 + paddd mm4, MMWORD PTR[GLOBAL(_12000)]
1.190 + paddd mm3, MMWORD PTR[GLOBAL(_51000)]
1.191 + paddd mm5, MMWORD PTR[GLOBAL(_51000)]
1.192 +
1.193 + psrad mm1, 16 ; (c1 * 2217 + d1 * 5352 + 14500)>>16
1.194 + psrad mm4, 16 ; (c1 * 2217 + d1 * 5352 + 14500)>>16
1.195 + psrad mm3, 16 ; (d1 * 2217 - c1 * 5352 + 7500)>>16
1.196 + psrad mm5, 16 ; (d1 * 2217 - c1 * 5352 + 7500)>>16
1.197 +
1.198 + packssdw mm1, mm4 ; op[4]
1.199 + packssdw mm3, mm5 ; op[12]
1.200 +
1.201 + paddw mm1, mm6 ; op[4] += (d1!=0)
1.202 +
1.203 + movq MMWORD PTR[rdi + 8 ], mm1
1.204 + movq MMWORD PTR[rdi + 24], mm3
1.205 +
1.206 + ; begin epilog
1.207 + pop rdi
1.208 + pop rsi
1.209 + RESTORE_GOT
1.210 + UNSHADOW_ARGS
1.211 + pop rbp
1.212 + ret
1.213 +
1.214 +SECTION_RODATA
1.215 +align 8
1.216 +_5352_2217:
1.217 + dw 5352
1.218 + dw 2217
1.219 + dw 5352
1.220 + dw 2217
1.221 +align 8
1.222 +_2217_neg5352:
1.223 + dw 2217
1.224 + dw -5352
1.225 + dw 2217
1.226 + dw -5352
1.227 +align 8
1.228 +_cmp_mask:
1.229 + times 4 dw 1
1.230 +align 8
1.231 +_7w:
1.232 + times 4 dw 7
1.233 +align 8
1.234 +_14500:
1.235 + times 2 dd 14500
1.236 +align 8
1.237 +_7500:
1.238 + times 2 dd 7500
1.239 +align 8
1.240 +_12000:
1.241 + times 2 dd 12000
1.242 +align 8
1.243 +_51000:
1.244 + times 2 dd 51000
