media/libvpx/vp8/encoder/x86/dct_mmx.asm@b8a032363ba2
media/libvpx/vp8/encoder/x86/dct_mmx.asm
Thu, 22 Jan 2015 13:21:57 +0100
- author
- Michael Schloh von Bennewitz <michael@schloh.com>
- date
- Thu, 22 Jan 2015 13:21:57 +0100
- branch
- TOR_BUG_9701
- changeset 15
- b8a032363ba2
- permissions
- -rw-r--r--
Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6
1 ;
2 ; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
3 ;
4 ; Use of this source code is governed by a BSD-style license
5 ; that can be found in the LICENSE file in the root of the source
6 ; tree. An additional intellectual property rights grant can be found
7 ; in the file PATENTS. All contributing project authors may
8 ; be found in the AUTHORS file in the root of the source tree.
9 ;
12 %include "vpx_ports/x86_abi_support.asm"
14 ;void vp8_short_fdct4x4_mmx(short *input, short *output, int pitch)
15 global sym(vp8_short_fdct4x4_mmx) PRIVATE
16 sym(vp8_short_fdct4x4_mmx):
17 push rbp
18 mov rbp, rsp
19 SHADOW_ARGS_TO_STACK 3
20 GET_GOT rbx
21 push rsi
22 push rdi
23 ; end prolog
25 mov rsi, arg(0) ; input
26 mov rdi, arg(1) ; output
28 movsxd rax, dword ptr arg(2) ;pitch
30 lea rcx, [rsi + rax*2]
31 ; read the input data
32 movq mm0, [rsi]
33 movq mm1, [rsi + rax]
35 movq mm2, [rcx]
36 movq mm4, [rcx + rax]
38 ; transpose for the first stage
39 movq mm3, mm0 ; 00 01 02 03
40 movq mm5, mm2 ; 20 21 22 23
42 punpcklwd mm0, mm1 ; 00 10 01 11
43 punpckhwd mm3, mm1 ; 02 12 03 13
45 punpcklwd mm2, mm4 ; 20 30 21 31
46 punpckhwd mm5, mm4 ; 22 32 23 33
48 movq mm1, mm0 ; 00 10 01 11
49 punpckldq mm0, mm2 ; 00 10 20 30
51 punpckhdq mm1, mm2 ; 01 11 21 31
53 movq mm2, mm3 ; 02 12 03 13
54 punpckldq mm2, mm5 ; 02 12 22 32
56 punpckhdq mm3, mm5 ; 03 13 23 33
58 ; mm0 0
59 ; mm1 1
60 ; mm2 2
61 ; mm3 3
63 ; first stage
64 movq mm5, mm0
65 movq mm4, mm1
67 paddw mm0, mm3 ; a1 = 0 + 3
68 paddw mm1, mm2 ; b1 = 1 + 2
70 psubw mm4, mm2 ; c1 = 1 - 2
71 psubw mm5, mm3 ; d1 = 0 - 3
73 psllw mm5, 3
74 psllw mm4, 3
76 psllw mm0, 3
77 psllw mm1, 3
79 ; output 0 and 2
80 movq mm2, mm0 ; a1
82 paddw mm0, mm1 ; op[0] = a1 + b1
83 psubw mm2, mm1 ; op[2] = a1 - b1
85 ; output 1 and 3
86 ; interleave c1, d1
87 movq mm1, mm5 ; d1
88 punpcklwd mm1, mm4 ; c1 d1
89 punpckhwd mm5, mm4 ; c1 d1
91 movq mm3, mm1
92 movq mm4, mm5
94 pmaddwd mm1, MMWORD PTR[GLOBAL (_5352_2217)] ; c1*2217 + d1*5352
95 pmaddwd mm4, MMWORD PTR[GLOBAL (_5352_2217)] ; c1*2217 + d1*5352
97 pmaddwd mm3, MMWORD PTR[GLOBAL(_2217_neg5352)] ; d1*2217 - c1*5352
98 pmaddwd mm5, MMWORD PTR[GLOBAL(_2217_neg5352)] ; d1*2217 - c1*5352
100 paddd mm1, MMWORD PTR[GLOBAL(_14500)]
101 paddd mm4, MMWORD PTR[GLOBAL(_14500)]
102 paddd mm3, MMWORD PTR[GLOBAL(_7500)]
103 paddd mm5, MMWORD PTR[GLOBAL(_7500)]
105 psrad mm1, 12 ; (c1 * 2217 + d1 * 5352 + 14500)>>12
106 psrad mm4, 12 ; (c1 * 2217 + d1 * 5352 + 14500)>>12
107 psrad mm3, 12 ; (d1 * 2217 - c1 * 5352 + 7500)>>12
108 psrad mm5, 12 ; (d1 * 2217 - c1 * 5352 + 7500)>>12
110 packssdw mm1, mm4 ; op[1]
111 packssdw mm3, mm5 ; op[3]
113 ; done with vertical
114 ; transpose for the second stage
115 movq mm4, mm0 ; 00 10 20 30
116 movq mm5, mm2 ; 02 12 22 32
118 punpcklwd mm0, mm1 ; 00 01 10 11
119 punpckhwd mm4, mm1 ; 20 21 30 31
121 punpcklwd mm2, mm3 ; 02 03 12 13
122 punpckhwd mm5, mm3 ; 22 23 32 33
124 movq mm1, mm0 ; 00 01 10 11
125 punpckldq mm0, mm2 ; 00 01 02 03
127 punpckhdq mm1, mm2 ; 01 22 12 13
129 movq mm2, mm4 ; 20 31 30 31
130 punpckldq mm2, mm5 ; 20 21 22 23
132 punpckhdq mm4, mm5 ; 30 31 32 33
134 ; mm0 0
135 ; mm1 1
136 ; mm2 2
137 ; mm3 4
139 movq mm5, mm0
140 movq mm3, mm1
142 paddw mm0, mm4 ; a1 = 0 + 3
143 paddw mm1, mm2 ; b1 = 1 + 2
145 psubw mm3, mm2 ; c1 = 1 - 2
146 psubw mm5, mm4 ; d1 = 0 - 3
148 pxor mm6, mm6 ; zero out for compare
150 pcmpeqw mm6, mm5 ; d1 != 0
152 pandn mm6, MMWORD PTR[GLOBAL(_cmp_mask)] ; clear upper,
153 ; and keep bit 0 of lower
155 ; output 0 and 2
156 movq mm2, mm0 ; a1
158 paddw mm0, mm1 ; a1 + b1
159 psubw mm2, mm1 ; a1 - b1
161 paddw mm0, MMWORD PTR[GLOBAL(_7w)]
162 paddw mm2, MMWORD PTR[GLOBAL(_7w)]
164 psraw mm0, 4 ; op[0] = (a1 + b1 + 7)>>4
165 psraw mm2, 4 ; op[8] = (a1 - b1 + 7)>>4
167 movq MMWORD PTR[rdi + 0 ], mm0
168 movq MMWORD PTR[rdi + 16], mm2
170 ; output 1 and 3
171 ; interleave c1, d1
172 movq mm1, mm5 ; d1
173 punpcklwd mm1, mm3 ; c1 d1
174 punpckhwd mm5, mm3 ; c1 d1
176 movq mm3, mm1
177 movq mm4, mm5
179 pmaddwd mm1, MMWORD PTR[GLOBAL (_5352_2217)] ; c1*2217 + d1*5352
180 pmaddwd mm4, MMWORD PTR[GLOBAL (_5352_2217)] ; c1*2217 + d1*5352
182 pmaddwd mm3, MMWORD PTR[GLOBAL(_2217_neg5352)] ; d1*2217 - c1*5352
183 pmaddwd mm5, MMWORD PTR[GLOBAL(_2217_neg5352)] ; d1*2217 - c1*5352
185 paddd mm1, MMWORD PTR[GLOBAL(_12000)]
186 paddd mm4, MMWORD PTR[GLOBAL(_12000)]
187 paddd mm3, MMWORD PTR[GLOBAL(_51000)]
188 paddd mm5, MMWORD PTR[GLOBAL(_51000)]
190 psrad mm1, 16 ; (c1 * 2217 + d1 * 5352 + 14500)>>16
191 psrad mm4, 16 ; (c1 * 2217 + d1 * 5352 + 14500)>>16
192 psrad mm3, 16 ; (d1 * 2217 - c1 * 5352 + 7500)>>16
193 psrad mm5, 16 ; (d1 * 2217 - c1 * 5352 + 7500)>>16
195 packssdw mm1, mm4 ; op[4]
196 packssdw mm3, mm5 ; op[12]
198 paddw mm1, mm6 ; op[4] += (d1!=0)
200 movq MMWORD PTR[rdi + 8 ], mm1
201 movq MMWORD PTR[rdi + 24], mm3
203 ; begin epilog
204 pop rdi
205 pop rsi
206 RESTORE_GOT
207 UNSHADOW_ARGS
208 pop rbp
209 ret
211 SECTION_RODATA
212 align 8
213 _5352_2217:
214 dw 5352
215 dw 2217
216 dw 5352
217 dw 2217
218 align 8
219 _2217_neg5352:
220 dw 2217
221 dw -5352
222 dw 2217
223 dw -5352
224 align 8
225 _cmp_mask:
226 times 4 dw 1
227 align 8
228 _7w:
229 times 4 dw 7
230 align 8
231 _14500:
232 times 2 dd 14500
233 align 8
234 _7500:
235 times 2 dd 7500
236 align 8
237 _12000:
238 times 2 dd 12000
239 align 8
240 _51000:
241 times 2 dd 51000
