OLD | NEW |
1 ; | 1 ; |
2 ; Copyright (c) 2010 The WebM project authors. All Rights Reserved. | 2 ; Copyright (c) 2010 The WebM project authors. All Rights Reserved. |
3 ; | 3 ; |
4 ; Use of this source code is governed by a BSD-style license | 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 | 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 | 6 ; tree. An additional intellectual property rights grant can be found |
7 ; in the file PATENTS. All contributing project authors may | 7 ; in the file PATENTS. All contributing project authors may |
8 ; be found in the AUTHORS file in the root of the source tree. | 8 ; be found in the AUTHORS file in the root of the source tree. |
9 ; | 9 ; |
10 | 10 |
(...skipping 41 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
52 RESTORE_GOT | 52 RESTORE_GOT |
53 pop rbp | 53 pop rbp |
54 %else | 54 %else |
55 %ifidn __OUTPUT_FORMAT__,x64 | 55 %ifidn __OUTPUT_FORMAT__,x64 |
56 %endif | 56 %endif |
57 %endif | 57 %endif |
58 ret | 58 ret |
59 %endmacro | 59 %endmacro |
60 | 60 |
61 ;void vp8_short_fdct4x4_sse2(short *input, short *output, int pitch) | 61 ;void vp8_short_fdct4x4_sse2(short *input, short *output, int pitch) |
62 global sym(vp8_short_fdct4x4_sse2) | 62 global sym(vp8_short_fdct4x4_sse2) PRIVATE |
63 sym(vp8_short_fdct4x4_sse2): | 63 sym(vp8_short_fdct4x4_sse2): |
64 | 64 |
65 STACK_FRAME_CREATE | 65 STACK_FRAME_CREATE |
66 | 66 |
67 movq xmm0, MMWORD PTR[input ] ;03 02 01 00 | 67 movq xmm0, MMWORD PTR[input ] ;03 02 01 00 |
68 movq xmm2, MMWORD PTR[input+ pitch] ;13 12 11 10 | 68 movq xmm2, MMWORD PTR[input+ pitch] ;13 12 11 10 |
69 lea input, [input+2*pitch] | 69 lea input, [input+2*pitch] |
70 movq xmm1, MMWORD PTR[input ] ;23 22 21 20 | 70 movq xmm1, MMWORD PTR[input ] ;23 22 21 20 |
71 movq xmm3, MMWORD PTR[input+ pitch] ;33 32 31 30 | 71 movq xmm3, MMWORD PTR[input+ pitch] ;33 32 31 30 |
72 | 72 |
(...skipping 84 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
157 paddw xmm3, xmm2 ;op[4] += (d1!=0) | 157 paddw xmm3, xmm2 ;op[4] += (d1!=0) |
158 punpcklqdq xmm0, xmm3 ;op[4] op[0] | 158 punpcklqdq xmm0, xmm3 ;op[4] op[0] |
159 punpckhqdq xmm1, xmm3 ;op[12] op[8] | 159 punpckhqdq xmm1, xmm3 ;op[12] op[8] |
160 | 160 |
161 movdqa XMMWORD PTR[output + 0], xmm0 | 161 movdqa XMMWORD PTR[output + 0], xmm0 |
162 movdqa XMMWORD PTR[output + 16], xmm1 | 162 movdqa XMMWORD PTR[output + 16], xmm1 |
163 | 163 |
164 STACK_FRAME_DESTROY | 164 STACK_FRAME_DESTROY |
165 | 165 |
166 ;void vp8_short_fdct8x4_sse2(short *input, short *output, int pitch) | 166 ;void vp8_short_fdct8x4_sse2(short *input, short *output, int pitch) |
167 global sym(vp8_short_fdct8x4_sse2) | 167 global sym(vp8_short_fdct8x4_sse2) PRIVATE |
168 sym(vp8_short_fdct8x4_sse2): | 168 sym(vp8_short_fdct8x4_sse2): |
169 | 169 |
170 STACK_FRAME_CREATE | 170 STACK_FRAME_CREATE |
171 | 171 |
172 ; read the input data | 172 ; read the input data |
173 movdqa xmm0, [input ] | 173 movdqa xmm0, [input ] |
174 movdqa xmm2, [input+ pitch] | 174 movdqa xmm2, [input+ pitch] |
175 lea input, [input+2*pitch] | 175 lea input, [input+2*pitch] |
176 movdqa xmm4, [input ] | 176 movdqa xmm4, [input ] |
177 movdqa xmm3, [input+ pitch] | 177 movdqa xmm3, [input+ pitch] |
(...skipping 243 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
421 times 4 dd 14500 | 421 times 4 dd 14500 |
422 align 16 | 422 align 16 |
423 _7500: | 423 _7500: |
424 times 4 dd 7500 | 424 times 4 dd 7500 |
425 align 16 | 425 align 16 |
426 _12000: | 426 _12000: |
427 times 4 dd 12000 | 427 times 4 dd 12000 |
428 align 16 | 428 align 16 |
429 _51000: | 429 _51000: |
430 times 4 dd 51000 | 430 times 4 dd 51000 |
OLD | NEW |