inv_txfm_ssse3_x86_64.asm revision 7ce0a1d1337c01056ba24006efab21f00e179e04 
1;
2;  Copyright (c) 2014 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;
10
11%include "third_party/x86inc/x86inc.asm"
12
13; This file provides SSSE3 version of the inverse transformation. Part
14; of the functions are originally derived from the ffmpeg project.
15; Note that the current version applies to x86 64-bit only.
16
17SECTION_RODATA
18
19pw_11585x2: times 8 dw 23170
20pd_8192:    times 4 dd 8192
21pw_16:      times 8 dw 16
22
23%macro TRANSFORM_COEFFS 2
24pw_%1_%2:   dw  %1,  %2,  %1,  %2,  %1,  %2,  %1,  %2
25pw_m%2_%1:  dw -%2,  %1, -%2,  %1, -%2,  %1, -%2,  %1
26%endmacro
27
28TRANSFORM_COEFFS    6270, 15137
29TRANSFORM_COEFFS    3196, 16069
30TRANSFORM_COEFFS   13623,  9102
31
32%macro PAIR_PP_COEFFS 2
33dpw_%1_%2:   dw  %1,  %1,  %1,  %1,  %2,  %2,  %2,  %2
34%endmacro
35
36%macro PAIR_MP_COEFFS 2
37dpw_m%1_%2:  dw -%1, -%1, -%1, -%1,  %2,  %2,  %2,  %2
38%endmacro
39
40%macro PAIR_MM_COEFFS 2
41dpw_m%1_m%2: dw -%1, -%1, -%1, -%1, -%2, -%2, -%2, -%2
42%endmacro
43
44PAIR_PP_COEFFS     30274, 12540
45PAIR_PP_COEFFS      6392, 32138
46PAIR_MP_COEFFS     18204, 27246
47
48PAIR_PP_COEFFS     12540, 12540
49PAIR_PP_COEFFS     30274, 30274
50PAIR_PP_COEFFS      6392,  6392
51PAIR_PP_COEFFS     32138, 32138
52PAIR_MM_COEFFS     18204, 18204
53PAIR_PP_COEFFS     27246, 27246
54
55SECTION .text
56
57%if ARCH_X86_64
58%macro SUM_SUB 3
59  psubw  m%3, m%1, m%2
60  paddw  m%1, m%2
61  SWAP    %2, %3
62%endmacro
63
64; butterfly operation
65%macro MUL_ADD_2X 6 ; dst1, dst2, src, round, coefs1, coefs2
66  pmaddwd            m%1, m%3, %5
67  pmaddwd            m%2, m%3, %6
68  paddd              m%1,  %4
69  paddd              m%2,  %4
70  psrad              m%1,  14
71  psrad              m%2,  14
72%endmacro
73
74%macro BUTTERFLY_4X 7 ; dst1, dst2, coef1, coef2, round, tmp1, tmp2
75  punpckhwd          m%6, m%2, m%1
76  MUL_ADD_2X         %7,  %6,  %6,  %5, [pw_m%4_%3], [pw_%3_%4]
77  punpcklwd          m%2, m%1
78  MUL_ADD_2X         %1,  %2,  %2,  %5, [pw_m%4_%3], [pw_%3_%4]
79  packssdw           m%1, m%7
80  packssdw           m%2, m%6
81%endmacro
82
83; matrix transpose
84%macro INTERLEAVE_2X 4
85  punpckh%1          m%4, m%2, m%3
86  punpckl%1          m%2, m%3
87  SWAP               %3,  %4
88%endmacro
89
90%macro TRANSPOSE8X8 9
91  INTERLEAVE_2X  wd, %1, %2, %9
92  INTERLEAVE_2X  wd, %3, %4, %9
93  INTERLEAVE_2X  wd, %5, %6, %9
94  INTERLEAVE_2X  wd, %7, %8, %9
95
96  INTERLEAVE_2X  dq, %1, %3, %9
97  INTERLEAVE_2X  dq, %2, %4, %9
98  INTERLEAVE_2X  dq, %5, %7, %9
99  INTERLEAVE_2X  dq, %6, %8, %9
100
101  INTERLEAVE_2X  qdq, %1, %5, %9
102  INTERLEAVE_2X  qdq, %3, %7, %9
103  INTERLEAVE_2X  qdq, %2, %6, %9
104  INTERLEAVE_2X  qdq, %4, %8, %9
105
106  SWAP  %2, %5
107  SWAP  %4, %7
108%endmacro
109
110%macro IDCT8_1D 0
111  SUM_SUB          0,    4,    9
112  BUTTERFLY_4X     2,    6,    6270, 15137,  m8,  9,  10
113  pmulhrsw        m0,  m12
114  pmulhrsw        m4,  m12
115  BUTTERFLY_4X     1,    7,    3196, 16069,  m8,  9,  10
116  BUTTERFLY_4X     5,    3,   13623,  9102,  m8,  9,  10
117
118  SUM_SUB          1,    5,    9
119  SUM_SUB          7,    3,    9
120  SUM_SUB          0,    6,    9
121  SUM_SUB          4,    2,    9
122  SUM_SUB          3,    5,    9
123  pmulhrsw        m3,  m12
124  pmulhrsw        m5,  m12
125
126  SUM_SUB          0,    7,    9
127  SUM_SUB          4,    3,    9
128  SUM_SUB          2,    5,    9
129  SUM_SUB          6,    1,    9
130
131  SWAP             3,    6
132  SWAP             1,    4
133%endmacro
134
135; This macro handles 8 pixels per line
136%macro ADD_STORE_8P_2X 5;  src1, src2, tmp1, tmp2, zero
137  paddw           m%1, m11
138  paddw           m%2, m11
139  psraw           m%1, 5
140  psraw           m%2, 5
141
142  movh            m%3, [outputq]
143  movh            m%4, [outputq + strideq]
144  punpcklbw       m%3, m%5
145  punpcklbw       m%4, m%5
146  paddw           m%3, m%1
147  paddw           m%4, m%2
148  packuswb        m%3, m%5
149  packuswb        m%4, m%5
150  movh               [outputq], m%3
151  movh     [outputq + strideq], m%4
152%endmacro
153
154INIT_XMM ssse3
155; full inverse 8x8 2D-DCT transform
156cglobal idct8x8_64_add, 3, 5, 13, input, output, stride
157  mova     m8, [pd_8192]
158  mova    m11, [pw_16]
159  mova    m12, [pw_11585x2]
160
161  lea      r3, [2 * strideq]
162
163  mova     m0, [inputq +   0]
164  mova     m1, [inputq +  16]
165  mova     m2, [inputq +  32]
166  mova     m3, [inputq +  48]
167  mova     m4, [inputq +  64]
168  mova     m5, [inputq +  80]
169  mova     m6, [inputq +  96]
170  mova     m7, [inputq + 112]
171
172  TRANSPOSE8X8  0, 1, 2, 3, 4, 5, 6, 7, 9
173  IDCT8_1D
174  TRANSPOSE8X8  0, 1, 2, 3, 4, 5, 6, 7, 9
175  IDCT8_1D
176
177  pxor    m12, m12
178  ADD_STORE_8P_2X  0, 1, 9, 10, 12
179  lea              outputq, [outputq + r3]
180  ADD_STORE_8P_2X  2, 3, 9, 10, 12
181  lea              outputq, [outputq + r3]
182  ADD_STORE_8P_2X  4, 5, 9, 10, 12
183  lea              outputq, [outputq + r3]
184  ADD_STORE_8P_2X  6, 7, 9, 10, 12
185
186  RET
187
188; inverse 8x8 2D-DCT transform with only first 10 coeffs non-zero
189cglobal idct8x8_12_add, 3, 5, 13, input, output, stride
190  mova       m8, [pd_8192]
191  mova      m11, [pw_16]
192  mova      m12, [pw_11585x2]
193
194  lea        r3, [2 * strideq]
195
196  mova       m0, [inputq +  0]
197  mova       m1, [inputq + 16]
198  mova       m2, [inputq + 32]
199  mova       m3, [inputq + 48]
200
201  punpcklwd  m0, m1
202  punpcklwd  m2, m3
203  punpckhdq  m9, m0, m2
204  punpckldq  m0, m2
205  SWAP       2, 9
206
207  ; m0 -> [0], [0]
208  ; m1 -> [1], [1]
209  ; m2 -> [2], [2]
210  ; m3 -> [3], [3]
211  punpckhqdq m10, m0, m0
212  punpcklqdq m0,  m0
213  punpckhqdq m9,  m2, m2
214  punpcklqdq m2,  m2
215  SWAP       1, 10
216  SWAP       3,  9
217
218  pmulhrsw   m0, m12
219  pmulhrsw   m2, [dpw_30274_12540]
220  pmulhrsw   m1, [dpw_6392_32138]
221  pmulhrsw   m3, [dpw_m18204_27246]
222
223  SUM_SUB    0, 2, 9
224  SUM_SUB    1, 3, 9
225
226  punpcklqdq m9, m3, m3
227  punpckhqdq m5, m3, m9
228
229  SUM_SUB    3, 5, 9
230  punpckhqdq m5, m3
231  pmulhrsw   m5, m12
232
233  punpckhqdq m9, m1, m5
234  punpcklqdq m1, m5
235  SWAP       5, 9
236
237  SUM_SUB    0, 5, 9
238  SUM_SUB    2, 1, 9
239
240  punpckhqdq m3, m0, m0
241  punpckhqdq m4, m1, m1
242  punpckhqdq m6, m5, m5
243  punpckhqdq m7, m2, m2
244
245  punpcklwd  m0, m3
246  punpcklwd  m7, m2
247  punpcklwd  m1, m4
248  punpcklwd  m6, m5
249
250  punpckhdq  m4, m0, m7
251  punpckldq  m0, m7
252  punpckhdq  m10, m1, m6
253  punpckldq  m5, m1, m6
254
255  punpckhqdq m1, m0, m5
256  punpcklqdq m0, m5
257  punpckhqdq m3, m4, m10
258  punpcklqdq m2, m4, m10
259
260
261  pmulhrsw   m0, m12
262  pmulhrsw   m6, m2, [dpw_30274_30274]
263  pmulhrsw   m4, m2, [dpw_12540_12540]
264
265  pmulhrsw   m7, m1, [dpw_32138_32138]
266  pmulhrsw   m1, [dpw_6392_6392]
267  pmulhrsw   m5, m3, [dpw_m18204_m18204]
268  pmulhrsw   m3, [dpw_27246_27246]
269
270  mova       m2, m0
271  SUM_SUB    0, 6, 9
272  SUM_SUB    2, 4, 9
273  SUM_SUB    1, 5, 9
274  SUM_SUB    7, 3, 9
275
276  SUM_SUB    3, 5, 9
277  pmulhrsw   m3, m12
278  pmulhrsw   m5, m12
279
280  SUM_SUB    0, 7, 9
281  SUM_SUB    2, 3, 9
282  SUM_SUB    4, 5, 9
283  SUM_SUB    6, 1, 9
284
285  SWAP       3, 6
286  SWAP       1, 2
287  SWAP       2, 4
288
289
290  pxor    m12, m12
291  ADD_STORE_8P_2X  0, 1, 9, 10, 12
292  lea              outputq, [outputq + r3]
293  ADD_STORE_8P_2X  2, 3, 9, 10, 12
294  lea              outputq, [outputq + r3]
295  ADD_STORE_8P_2X  4, 5, 9, 10, 12
296  lea              outputq, [outputq + r3]
297  ADD_STORE_8P_2X  6, 7, 9, 10, 12
298
299  RET
300
301%endif
302