1 ;*****************************************************************************
2 ;* x86-optimized AC-3 DSP utils
3 ;* Copyright (c) 2011 Justin Ruggles
5 ;* This file is part of FFmpeg.
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.
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.
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 ;******************************************************************************
23 %include "x86util.asm"
27 ; 16777216.0f - used in ff_float_to_fixed24()
28 pf_1_24: times 4 dd 0x4B800000
30 ; used in ff_ac3_compute_mantissa_size()
32 pw_bap_mul1: dw 21846, 21846, 0, 32768, 21846, 21846, 0, 32768
33 pw_bap_mul2: dw 5, 7, 0, 7, 5, 7, 0, 7
37 ;-----------------------------------------------------------------------------
38 ; void ff_ac3_exponent_min(uint8_t *exp, int num_reuse_blocks, int nb_coefs)
39 ;-----------------------------------------------------------------------------
41 %macro AC3_EXPONENT_MIN 1
42 cglobal ac3_exponent_min_%1, 3,4,2, exp, reuse_blks, expn, offset
47 mov offsetq, reuse_blksq
48 mova m0, [expq+offsetq]
52 PMINUB m0, [expq+offsetq], m1
63 %define PMINUB PMINUB_MMX
68 %define PMINUB PMINUB_MMXEXT
69 %define LOOP_ALIGN ALIGN 16
70 AC3_EXPONENT_MIN mmxext
79 ;-----------------------------------------------------------------------------
80 ; int ff_ac3_max_msb_abs_int16(const int16_t *src, int len)
82 ; This function uses 2 different methods to calculate a valid result.
83 ; 1) logical 'or' of abs of each element
84 ; This is used for ssse3 because of the pabsw instruction.
85 ; It is also used for mmx because of the lack of min/max instructions.
86 ; 2) calculate min/max for the array, then or(abs(min),abs(max))
87 ; This is used for mmxext and sse2 because they have pminsw/pmaxsw.
88 ;-----------------------------------------------------------------------------
90 %macro AC3_MAX_MSB_ABS_INT16 2
91 cglobal ac3_max_msb_abs_int16_%1, 2,2,5, src, len
97 mova m1, [srcq+mmsize]
105 mova m1, [srcq+mmsize]
108 ; using memory args is faster for ssse3
110 pabsw m1, [srcq+mmsize]
136 %define ABS2 ABS2_MMX
137 %define PSHUFLW pshufw
138 AC3_MAX_MSB_ABS_INT16 mmx, or_abs
139 %define ABS2 ABS2_MMX2
140 AC3_MAX_MSB_ABS_INT16 mmxext, min_max
142 %define PSHUFLW pshuflw
143 AC3_MAX_MSB_ABS_INT16 sse2, min_max
144 %define ABS2 ABS2_SSSE3
145 AC3_MAX_MSB_ABS_INT16 ssse3, or_abs
147 ;-----------------------------------------------------------------------------
148 ; macro used for ff_ac3_lshift_int16() and ff_ac3_rshift_int32()
149 ;-----------------------------------------------------------------------------
151 %macro AC3_SHIFT 4 ; l/r, 16/32, shift instruction, instruction set
152 cglobal ac3_%1shift_int%2_%4, 3,3,5, src, len, shift
156 mova m2, [srcq+mmsize ]
157 mova m3, [srcq+mmsize*2]
158 mova m4, [srcq+mmsize*3]
164 mova [srcq+mmsize ], m2
165 mova [srcq+mmsize*2], m3
166 mova [srcq+mmsize*3], m4
168 sub lend, mmsize*32/%2
174 ;-----------------------------------------------------------------------------
175 ; void ff_ac3_lshift_int16(int16_t *src, unsigned int len, unsigned int shift)
176 ;-----------------------------------------------------------------------------
179 AC3_SHIFT l, 16, psllw, mmx
181 AC3_SHIFT l, 16, psllw, sse2
183 ;-----------------------------------------------------------------------------
184 ; void ff_ac3_rshift_int32(int32_t *src, unsigned int len, unsigned int shift)
185 ;-----------------------------------------------------------------------------
188 AC3_SHIFT r, 32, psrad, mmx
190 AC3_SHIFT r, 32, psrad, sse2
192 ;-----------------------------------------------------------------------------
193 ; void ff_float_to_fixed24(int32_t *dst, const float *src, unsigned int len)
194 ;-----------------------------------------------------------------------------
196 ; The 3DNow! version is not bit-identical because pf2id uses truncation rather
197 ; than round-to-nearest.
199 cglobal float_to_fixed24_3dnow, 3,3,0, dst, src, len
225 cglobal float_to_fixed24_sse, 3,3,3, dst, src, len
249 cglobal float_to_fixed24_sse2, 3,3,9, dst, src, len
253 movaps m2, [srcq+16 ]
254 movaps m3, [srcq+32 ]
255 movaps m4, [srcq+48 ]
257 movaps m5, [srcq+64 ]
258 movaps m6, [srcq+80 ]
259 movaps m7, [srcq+96 ]
260 movaps m8, [srcq+112]
283 movdqa [dstq+16 ], m2
284 movdqa [dstq+32 ], m3
285 movdqa [dstq+48 ], m4
287 movdqa [dstq+64 ], m5
288 movdqa [dstq+80 ], m6
289 movdqa [dstq+96 ], m7
290 movdqa [dstq+112], m8
302 ;------------------------------------------------------------------------------
303 ; int ff_ac3_compute_mantissa_size(uint16_t mant_cnt[6][16])
304 ;------------------------------------------------------------------------------
306 %macro PHADDD4 2 ; xmm src, xmm tmp
314 cglobal ac3_compute_mantissa_size_sse2, 1,2,4, mant_cnt, sum
315 movdqa m0, [mant_cntq ]
316 movdqa m1, [mant_cntq+ 1*16]
317 paddw m0, [mant_cntq+ 2*16]
318 paddw m1, [mant_cntq+ 3*16]
319 paddw m0, [mant_cntq+ 4*16]
320 paddw m1, [mant_cntq+ 5*16]
321 paddw m0, [mant_cntq+ 6*16]
322 paddw m1, [mant_cntq+ 7*16]
323 paddw m0, [mant_cntq+ 8*16]
324 paddw m1, [mant_cntq+ 9*16]
325 paddw m0, [mant_cntq+10*16]
326 paddw m1, [mant_cntq+11*16]
327 pmaddwd m0, [ac3_bap_bits ]
328 pmaddwd m1, [ac3_bap_bits+16]
332 movdqa m3, [pw_bap_mul1]
333 movhpd m0, [mant_cntq +2]
334 movlpd m0, [mant_cntq+1*32+2]
335 movhpd m1, [mant_cntq+2*32+2]
336 movlpd m1, [mant_cntq+3*32+2]
337 movhpd m2, [mant_cntq+4*32+2]
338 movlpd m2, [mant_cntq+5*32+2]
344 pmaddwd m0, [pw_bap_mul2]