1;*****************************************************************************
2;* x86-optimized AC-3 DSP functions
3;* Copyright (c) 2011 Justin Ruggles
4;*
5;* This file is part of FFmpeg.
6;*
7;* FFmpeg is free software; you can redistribute it and/or
8;* modify it under the terms of the GNU Lesser General Public
9;* License as published by the Free Software Foundation; either
10;* version 2.1 of the License, or (at your option) any later version.
11;*
12;* FFmpeg is distributed in the hope that it will be useful,
13;* but WITHOUT ANY WARRANTY; without even the implied warranty of
14;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15;* Lesser General Public License for more details.
16;*
17;* You should have received a copy of the GNU Lesser General Public
18;* License along with FFmpeg; if not, write to the Free Software
19;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20;******************************************************************************
21
22%include "libavutil/x86/x86util.asm"
23
24SECTION_RODATA
25
26; 16777216.0f - used in ff_float_to_fixed24()
27pf_1_24: times 4 dd 0x4B800000
28
29; used in ff_ac3_compute_mantissa_size()
30cextern ac3_bap_bits
31pw_bap_mul1: dw 21846, 21846, 0, 32768, 21846, 21846, 0, 32768
32pw_bap_mul2: dw 5, 7, 0, 7, 5, 7, 0, 7
33
34; used in ff_ac3_extract_exponents()
35cextern pd_1
36pd_151: times 4 dd 151
37
38SECTION .text
39
40;-----------------------------------------------------------------------------
41; void ff_ac3_exponent_min(uint8_t *exp, int num_reuse_blocks, int nb_coefs)
42;-----------------------------------------------------------------------------
43
44%macro AC3_EXPONENT_MIN 0
45cglobal ac3_exponent_min, 3, 4, 2, exp, reuse_blks, expn, offset
46    shl  reuse_blksq, 8
47    jz .end
48    LOOP_ALIGN
49.nextexp:
50    mov      offsetq, reuse_blksq
51    mova          m0, [expq+offsetq]
52    sub      offsetq, 256
53    LOOP_ALIGN
54.nextblk:
55    PMINUB        m0, [expq+offsetq], m1
56    sub      offsetq, 256
57    jae .nextblk
58    mova      [expq], m0
59    add         expq, mmsize
60    sub        expnq, mmsize
61    jg .nextexp
62.end:
63    REP_RET
64%endmacro
65
66%define LOOP_ALIGN ALIGN 16
67%if HAVE_SSE2_EXTERNAL
68INIT_XMM sse2
69AC3_EXPONENT_MIN
70%endif
71%undef LOOP_ALIGN
72
73;-----------------------------------------------------------------------------
74; void ff_float_to_fixed24(int32_t *dst, const float *src, unsigned int len)
75;-----------------------------------------------------------------------------
76
77INIT_XMM sse2
78cglobal float_to_fixed24, 3, 3, 9, dst, src, len
79    movaps     m0, [pf_1_24]
80.loop:
81    movaps     m1, [srcq    ]
82    movaps     m2, [srcq+16 ]
83    movaps     m3, [srcq+32 ]
84    movaps     m4, [srcq+48 ]
85%ifdef m8
86    movaps     m5, [srcq+64 ]
87    movaps     m6, [srcq+80 ]
88    movaps     m7, [srcq+96 ]
89    movaps     m8, [srcq+112]
90%endif
91    mulps      m1, m0
92    mulps      m2, m0
93    mulps      m3, m0
94    mulps      m4, m0
95%ifdef m8
96    mulps      m5, m0
97    mulps      m6, m0
98    mulps      m7, m0
99    mulps      m8, m0
100%endif
101    cvtps2dq   m1, m1
102    cvtps2dq   m2, m2
103    cvtps2dq   m3, m3
104    cvtps2dq   m4, m4
105%ifdef m8
106    cvtps2dq   m5, m5
107    cvtps2dq   m6, m6
108    cvtps2dq   m7, m7
109    cvtps2dq   m8, m8
110%endif
111    movdqa  [dstq    ], m1
112    movdqa  [dstq+16 ], m2
113    movdqa  [dstq+32 ], m3
114    movdqa  [dstq+48 ], m4
115%ifdef m8
116    movdqa  [dstq+64 ], m5
117    movdqa  [dstq+80 ], m6
118    movdqa  [dstq+96 ], m7
119    movdqa  [dstq+112], m8
120    add      srcq, 128
121    add      dstq, 128
122    sub      lenq, 32
123%else
124    add      srcq, 64
125    add      dstq, 64
126    sub      lenq, 16
127%endif
128    ja .loop
129    REP_RET
130
131;------------------------------------------------------------------------------
132; int ff_ac3_compute_mantissa_size(uint16_t mant_cnt[6][16])
133;------------------------------------------------------------------------------
134
135%macro PHADDD4 2 ; xmm src, xmm tmp
136    movhlps  %2, %1
137    paddd    %1, %2
138    pshufd   %2, %1, 0x1
139    paddd    %1, %2
140%endmacro
141
142INIT_XMM sse2
143cglobal ac3_compute_mantissa_size, 1, 2, 4, mant_cnt, sum
144    movdqa      m0, [mant_cntq      ]
145    movdqa      m1, [mant_cntq+ 1*16]
146    paddw       m0, [mant_cntq+ 2*16]
147    paddw       m1, [mant_cntq+ 3*16]
148    paddw       m0, [mant_cntq+ 4*16]
149    paddw       m1, [mant_cntq+ 5*16]
150    paddw       m0, [mant_cntq+ 6*16]
151    paddw       m1, [mant_cntq+ 7*16]
152    paddw       m0, [mant_cntq+ 8*16]
153    paddw       m1, [mant_cntq+ 9*16]
154    paddw       m0, [mant_cntq+10*16]
155    paddw       m1, [mant_cntq+11*16]
156    pmaddwd     m0, [ac3_bap_bits   ]
157    pmaddwd     m1, [ac3_bap_bits+16]
158    paddd       m0, m1
159    PHADDD4     m0, m1
160    movd      sumd, m0
161    movdqa      m3, [pw_bap_mul1]
162    movhpd      m0, [mant_cntq     +2]
163    movlpd      m0, [mant_cntq+1*32+2]
164    movhpd      m1, [mant_cntq+2*32+2]
165    movlpd      m1, [mant_cntq+3*32+2]
166    movhpd      m2, [mant_cntq+4*32+2]
167    movlpd      m2, [mant_cntq+5*32+2]
168    pmulhuw     m0, m3
169    pmulhuw     m1, m3
170    pmulhuw     m2, m3
171    paddusw     m0, m1
172    paddusw     m0, m2
173    pmaddwd     m0, [pw_bap_mul2]
174    PHADDD4     m0, m1
175    movd       eax, m0
176    add        eax, sumd
177    RET
178
179;------------------------------------------------------------------------------
180; void ff_ac3_extract_exponents(uint8_t *exp, int32_t *coef, int nb_coefs)
181;------------------------------------------------------------------------------
182
183%macro PABSD 1-2 ; src/dst, unused
184%if cpuflag(ssse3)
185    pabsd    %1, %1
186%else ; src/dst, tmp
187    pxor     %2, %2
188    pcmpgtd  %2, %1
189    pxor     %1, %2
190    psubd    %1, %2
191%endif
192%endmacro
193
194%macro AC3_EXTRACT_EXPONENTS 0
195cglobal ac3_extract_exponents, 3, 3, 4, exp, coef, len
196    add     expq, lenq
197    lea    coefq, [coefq+4*lenq]
198    neg     lenq
199    mova      m2, [pd_1]
200    mova      m3, [pd_151]
201.loop:
202    ; move 4 32-bit coefs to xmm0
203    mova      m0, [coefq+4*lenq]
204    ; absolute value
205    PABSD     m0, m1
206    ; convert to float and extract exponents
207    pslld     m0, 1
208    por       m0, m2
209    cvtdq2ps  m1, m0
210    psrld     m1, 23
211    mova      m0, m3
212    psubd     m0, m1
213    ; move the lowest byte in each of 4 dwords to the low dword
214    ; NOTE: We cannot just extract the low bytes with pshufb because the dword
215    ;       result for 16777215 is -1 due to float inaccuracy. Using packuswb
216    ;       clips this to 0, which is the correct exponent.
217    packssdw  m0, m0
218    packuswb  m0, m0
219    movd  [expq+lenq], m0
220
221    add     lenq, 4
222    jl .loop
223    REP_RET
224%endmacro
225
226%if HAVE_SSE2_EXTERNAL
227INIT_XMM sse2
228AC3_EXTRACT_EXPONENTS
229%endif
230%if HAVE_SSSE3_EXTERNAL
231INIT_XMM ssse3
232AC3_EXTRACT_EXPONENTS
233%endif
234