1 ;*****************************************************************************
2 ;* x86-optimized AC-3 DSP utils
3 ;* Copyright (c) 2011 Justin Ruggles
5 ;* This file is part of Libav.
7 ;* Libav 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 ;* Libav 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 Libav; 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
35 ; used in ff_ac3_extract_exponents()
37 pd_151: times 4 dd 151
41 ;-----------------------------------------------------------------------------
42 ; void ff_ac3_exponent_min(uint8_t *exp, int num_reuse_blocks, int nb_coefs)
43 ;-----------------------------------------------------------------------------
45 %macro AC3_EXPONENT_MIN 1
46 cglobal ac3_exponent_min_%1, 3,4,2, exp, reuse_blks, expn, offset
51 mov offsetq, reuse_blksq
52 mova m0, [expq+offsetq]
56 PMINUB m0, [expq+offsetq], m1
67 %define PMINUB PMINUB_MMX
72 %define PMINUB PMINUB_MMXEXT
73 %define LOOP_ALIGN ALIGN 16
74 AC3_EXPONENT_MIN mmxext
83 ;-----------------------------------------------------------------------------
84 ; int ff_ac3_max_msb_abs_int16(const int16_t *src, int len)
86 ; This function uses 2 different methods to calculate a valid result.
87 ; 1) logical 'or' of abs of each element
88 ; This is used for ssse3 because of the pabsw instruction.
89 ; It is also used for mmx because of the lack of min/max instructions.
90 ; 2) calculate min/max for the array, then or(abs(min),abs(max))
91 ; This is used for mmxext and sse2 because they have pminsw/pmaxsw.
92 ;-----------------------------------------------------------------------------
94 %macro AC3_MAX_MSB_ABS_INT16 2
95 cglobal ac3_max_msb_abs_int16_%1, 2,2,5, src, len
101 mova m1, [srcq+mmsize]
109 mova m1, [srcq+mmsize]
112 ; using memory args is faster for ssse3
114 pabsw m1, [srcq+mmsize]
140 %define ABS2 ABS2_MMX
141 %define PSHUFLW pshufw
142 AC3_MAX_MSB_ABS_INT16 mmx, or_abs
143 %define ABS2 ABS2_MMX2
144 AC3_MAX_MSB_ABS_INT16 mmxext, min_max
146 %define PSHUFLW pshuflw
147 AC3_MAX_MSB_ABS_INT16 sse2, min_max
148 %define ABS2 ABS2_SSSE3
149 AC3_MAX_MSB_ABS_INT16 ssse3, or_abs
151 ;-----------------------------------------------------------------------------
152 ; macro used for ff_ac3_lshift_int16() and ff_ac3_rshift_int32()
153 ;-----------------------------------------------------------------------------
155 %macro AC3_SHIFT 4 ; l/r, 16/32, shift instruction, instruction set
156 cglobal ac3_%1shift_int%2_%4, 3,3,5, src, len, shift
160 mova m2, [srcq+mmsize ]
161 mova m3, [srcq+mmsize*2]
162 mova m4, [srcq+mmsize*3]
168 mova [srcq+mmsize ], m2
169 mova [srcq+mmsize*2], m3
170 mova [srcq+mmsize*3], m4
172 sub lend, mmsize*32/%2
178 ;-----------------------------------------------------------------------------
179 ; void ff_ac3_lshift_int16(int16_t *src, unsigned int len, unsigned int shift)
180 ;-----------------------------------------------------------------------------
183 AC3_SHIFT l, 16, psllw, mmx
185 AC3_SHIFT l, 16, psllw, sse2
187 ;-----------------------------------------------------------------------------
188 ; void ff_ac3_rshift_int32(int32_t *src, unsigned int len, unsigned int shift)
189 ;-----------------------------------------------------------------------------
192 AC3_SHIFT r, 32, psrad, mmx
194 AC3_SHIFT r, 32, psrad, sse2
196 ;-----------------------------------------------------------------------------
197 ; void ff_float_to_fixed24(int32_t *dst, const float *src, unsigned int len)
198 ;-----------------------------------------------------------------------------
200 ; The 3DNow! version is not bit-identical because pf2id uses truncation rather
201 ; than round-to-nearest.
203 cglobal float_to_fixed24_3dnow, 3,3,0, dst, src, len
229 cglobal float_to_fixed24_sse, 3,3,3, dst, src, len
253 cglobal float_to_fixed24_sse2, 3,3,9, dst, src, len
257 movaps m2, [srcq+16 ]
258 movaps m3, [srcq+32 ]
259 movaps m4, [srcq+48 ]
261 movaps m5, [srcq+64 ]
262 movaps m6, [srcq+80 ]
263 movaps m7, [srcq+96 ]
264 movaps m8, [srcq+112]
287 movdqa [dstq+16 ], m2
288 movdqa [dstq+32 ], m3
289 movdqa [dstq+48 ], m4
291 movdqa [dstq+64 ], m5
292 movdqa [dstq+80 ], m6
293 movdqa [dstq+96 ], m7
294 movdqa [dstq+112], m8
306 ;------------------------------------------------------------------------------
307 ; int ff_ac3_compute_mantissa_size(uint16_t mant_cnt[6][16])
308 ;------------------------------------------------------------------------------
310 %macro PHADDD4 2 ; xmm src, xmm tmp
318 cglobal ac3_compute_mantissa_size_sse2, 1,2,4, mant_cnt, sum
319 movdqa m0, [mant_cntq ]
320 movdqa m1, [mant_cntq+ 1*16]
321 paddw m0, [mant_cntq+ 2*16]
322 paddw m1, [mant_cntq+ 3*16]
323 paddw m0, [mant_cntq+ 4*16]
324 paddw m1, [mant_cntq+ 5*16]
325 paddw m0, [mant_cntq+ 6*16]
326 paddw m1, [mant_cntq+ 7*16]
327 paddw m0, [mant_cntq+ 8*16]
328 paddw m1, [mant_cntq+ 9*16]
329 paddw m0, [mant_cntq+10*16]
330 paddw m1, [mant_cntq+11*16]
331 pmaddwd m0, [ac3_bap_bits ]
332 pmaddwd m1, [ac3_bap_bits+16]
336 movdqa m3, [pw_bap_mul1]
337 movhpd m0, [mant_cntq +2]
338 movlpd m0, [mant_cntq+1*32+2]
339 movhpd m1, [mant_cntq+2*32+2]
340 movlpd m1, [mant_cntq+3*32+2]
341 movhpd m2, [mant_cntq+4*32+2]
342 movlpd m2, [mant_cntq+5*32+2]
348 pmaddwd m0, [pw_bap_mul2]
354 ;------------------------------------------------------------------------------
355 ; void ff_ac3_extract_exponents(uint8_t *exp, int32_t *coef, int nb_coefs)
356 ;------------------------------------------------------------------------------
358 %macro PABSD_MMX 2 ; src/dst, tmp
365 %macro PABSD_SSSE3 1-2 ; src/dst, unused
371 cglobal ac3_extract_exponents_3dnow, 3,3,0, exp, coef, len
373 lea coefq, [coefq+4*lenq]
378 movq m0, [coefq+4*lenq ]
379 movq m1, [coefq+4*lenq+8]
405 %macro AC3_EXTRACT_EXPONENTS 1
406 cglobal ac3_extract_exponents_%1, 3,3,4, exp, coef, len
408 lea coefq, [coefq+4*lenq]
413 ; move 4 32-bit coefs to xmm0
414 mova m0, [coefq+4*lenq]
417 ; convert to float and extract exponents
424 ; move the lowest byte in each of 4 dwords to the low dword
425 ; NOTE: We cannot just extract the low bytes with pshufb because the dword
426 ; result for 16777215 is -1 due to float inaccuracy. Using packuswb
427 ; clips this to 0, which is the correct exponent.
439 %define PABSD PABSD_MMX
440 AC3_EXTRACT_EXPONENTS sse2
442 %define PABSD PABSD_SSSE3
443 AC3_EXTRACT_EXPONENTS ssse3