History

avcodec: Remove DCT, FFT, MDCT and RDFT

They were replaced by TX from libavutil; the tremendous work to get to this point (both creating TX as well as porting the users of the components removed in this commit) was completely performed by Lynne alone. Removing the subsystems from configure may break some command lines, because the --disable-fft etc. options are no longer recognized.

Andreas Rheinhardt committed 3 years ago

Commit 6f7bf641 parent d9464f352 files · +2 −8799Browse files

Unified · Split

52 files
  1. configure +0 −11
  2. doc/mips.txt +0 −8
  3. libavcodec/Makefile +1 −15
  4. libavcodec/aarch64/Makefile +0 −3
  5. libavcodec/aarch64/asm-offsets.h +0 −25
  6. libavcodec/aarch64/fft_init_aarch64.c +0 −52
  7. libavcodec/aarch64/fft_neon.S +0 −447
  8. libavcodec/aarch64/mdct_neon.S +0 −326
  9. libavcodec/aarch64/synth_filter_init.c +0 −7
  10. libavcodec/aarch64/synth_filter_neon.S +0 −2
  11. libavcodec/arm/Makefile +0 −7
  12. libavcodec/arm/fft_init_arm.c +0 −63
  13. libavcodec/arm/fft_neon.S +0 −375
  14. libavcodec/arm/fft_vfp.S +0 −530
  15. libavcodec/arm/mdct_neon.S +0 −301
  16. libavcodec/arm/mdct_vfp.S +0 −347
  17. libavcodec/arm/rdft_init_arm.c +0 −33
  18. libavcodec/arm/rdft_neon.S +0 −155
  19. libavcodec/arm/synth_filter_init_arm.c +0 −1
  20. libavcodec/cos_tablegen.c +0 −80
  21. libavcodec/dct.c +0 −228
  22. libavcodec/dct.h +1 −26
  23. libavcodec/fft-internal.h +0 −62
  24. libavcodec/fft.h +0 −160
  25. libavcodec/fft_fixed_32.c +0 −51
  26. libavcodec/fft_float.c +0 −20
  27. libavcodec/fft_init_table.c +0 −344
  28. libavcodec/fft_table.h +0 −66
  29. libavcodec/fft_template.c +0 −628
  30. libavcodec/mdct_fixed_32.c +0 −51
  31. libavcodec/mdct_float.c +0 −20
  32. libavcodec/mdct_template.c +0 −209
  33. libavcodec/mips/Makefile +0 −1
  34. libavcodec/mips/fft_mips.c +0 −516
  35. libavcodec/ppc/Makefile +0 −3
  36. libavcodec/ppc/fft_altivec.S +0 −458
  37. libavcodec/ppc/fft_init.c +0 −168
  38. libavcodec/ppc/fft_vsx.c +0 −226
  39. libavcodec/ppc/fft_vsx.h +0 −829
  40. libavcodec/rdft.c +0 −120
  41. libavcodec/rdft.h +0 −52
  42. libavcodec/tests/.gitignore +0 −3
  43. libavcodec/tests/avfft.c +0 −25
  44. libavcodec/tests/fft-fixed32.c +0 −21
  45. libavcodec/tests/fft.c +0 −683
  46. libavcodec/x86/Makefile +0 −4
  47. libavcodec/x86/dct_init.c +0 −36
  48. libavcodec/x86/fft.asm +0 −838
  49. libavcodec/x86/fft.h +0 −32
  50. libavcodec/x86/fft_init.c +0 −47
  51. tests/Makefile +0 −1
  52. tests/fate/fft.mak +0 −83
configure+0−11
@@ -136,13 +136,9 @@ Component options: · 135 unmodified lines
136136 --disable-w32threads disable Win32 threads [autodetect]
137137 --disable-os2threads disable OS/2 threads [autodetect]
138138 --disable-network disable network support [no]
139− --disable-dct disable DCT code
140139 --disable-dwt disable DWT code
141140 --disable-error-resilience disable error resilience code
142141 --disable-lsp disable LSP code
143− --disable-mdct disable MDCT code
144− --disable-rdft disable RDFT code
145− --disable-fft disable FFT code
146142 --disable-faan disable floating point AAN (I)DCT code
147143 --disable-pixelutils disable pixel utils in libavutil
148144
@@ -2004,17 +2000,13 @@ PROGRAM_LIST=" · 1855 unmodified lines
20042000 "
20052001
20062002 SUBSYSTEM_LIST="
2007− dct
20082003 dwt
20092004 error_resilience
20102005 faan
20112006 fast_unaligned
2012− fft
20132007 lsp
2014− mdct
20152008 pixelutils
20162009 network
2017− rdft
20182010 "
20192011
20202012 # COMPONENT_LIST needs to come last to ensure correct dependency checking
@@ -2766,7 +2758,6 @@ cbs_h266_select="cbs" · 745 unmodified lines
27662758 cbs_jpeg_select="cbs"
27672759 cbs_mpeg2_select="cbs"
27682760 cbs_vp9_select="cbs"
2769−dct_select="rdft"
27702761 deflate_wrapper_deps="zlib"
27712762 dirac_parse_select="golomb"
27722763 dovi_rpu_select="golomb"
@@ -2786,7 +2777,6 @@ frame_thread_encoder_deps="encoders threads" · 13 unmodified lines
27862777 inflate_wrapper_deps="zlib"
27872778 intrax8_select="blockdsp wmv2dsp"
27882779 iso_media_select="mpeg4audio"
2789−mdct_select="fft"
27902780 me_cmp_select="idctdsp"
27912781 mpeg_er_select="error_resilience"
27922782 mpegaudio_select="mpegaudiodsp mpegaudioheader"
@@ -2796,7 +2786,6 @@ mpegvideoenc_select="aandcttables fdctdsp me_cmp mpegvideo pixblockdsp" · 3 unmodified lines
27962786 msmpeg4dec_select="h263_decoder"
27972787 msmpeg4enc_select="h263_encoder"
27982788 vc1dsp_select="h264chroma qpeldsp startcode"
2799−rdft_select="fft"
28002789
28012790 # decoders / encoders
28022791 aac_decoder_select="adts_header mpeg4audio sinewin"
doc/mips.txt+0−8
@@ -48,11 +48,6 @@ Files that have MIPS copyright notice in them: · 47 unmodified lines
4848 float_dsp_mips.c
4949 libm_mips.h
5050 softfloat_tables.h
51−* libavcodec/
52− fft_fixed_32.c
53− fft_init_table.c
54− fft_table.h
55− mdct_fixed_32.c
5651 * libavcodec/mips/
5752 aacdec_fixed.c
5853 aacsbr_fixed.c
@@ -70,9 +65,6 @@ Files that have MIPS copyright notice in them: · 11 unmodified lines
7065 compute_antialias_float.h
7166 lsp_mips.h
7267 dsputil_mips.c
73− fft_mips.c
74− fft_table.h
75− fft_init_table.c
7668 fmtconvert_mips.c
7769 iirfilter_mips.c
7870 mpegaudiodsp_mips_fixed.c
libavcodec/Makefile+1−15
@@ -32,6 +32,7 @@ OBJS = ac3_parser.o \ · 31 unmodified lines
3232 allcodecs.o \
3333 avcodec.o \
3434 avdct.o \
35+ avfft.o \
3536 avpacket.o \
3637 bitstream.o \
3738 bitstream_filters.o \
@@ -81,7 +82,6 @@ OBJS-$(CONFIG_CBS_JPEG) += cbs_jpeg.o · 43 unmodified lines
8182 OBJS-$(CONFIG_CBS_MPEG2) += cbs_mpeg2.o
8283 OBJS-$(CONFIG_CBS_VP9) += cbs_vp9.o
8384 OBJS-$(CONFIG_CRYSTALHD) += crystalhd.o
84−OBJS-$(CONFIG_DCT) += dct.o dct32_fixed.o dct32_float.o
8585 OBJS-$(CONFIG_DEFLATE_WRAPPER) += zlib_wrapper.o
8686 OBJS-$(CONFIG_DOVI_RPU) += dovi_rpu.o
8787 OBJS-$(CONFIG_ERROR_RESILIENCE) += error_resilience.o
@@ -90,9 +90,6 @@ OBJS-$(CONFIG_EXIF) += exif.o tiff_common.o · 2 unmodified lines
9090 OBJS-$(CONFIG_FAANDCT) += faandct.o
9191 OBJS-$(CONFIG_FAANIDCT) += faanidct.o
9292 OBJS-$(CONFIG_FDCTDSP) += fdctdsp.o jfdctfst.o jfdctint.o
93−FFT-OBJS-$(CONFIG_HARDCODED_TABLES) += cos_tables.o
94−OBJS-$(CONFIG_FFT) += avfft.o fft_float.o fft_fixed_32.o \
95− fft_init_table.o $(FFT-OBJS-yes)
9693 OBJS-$(CONFIG_FMTCONVERT) += fmtconvert.o
9794 OBJS-$(CONFIG_GOLOMB) += golomb.o
9895 OBJS-$(CONFIG_H263DSP) += h263dsp.o
@@ -125,7 +122,6 @@ OBJS-$(CONFIG_LLVIDENCDSP) += lossless_videoencdsp.o · 26 unmodified lines
125122 OBJS-$(CONFIG_LPC) += lpc.o
126123 OBJS-$(CONFIG_LSP) += lsp.o
127124 OBJS-$(CONFIG_LZF) += lzf.o
128−OBJS-$(CONFIG_MDCT) += mdct_float.o mdct_fixed_32.o
129125 OBJS-$(CONFIG_ME_CMP) += me_cmp.o
130126 OBJS-$(CONFIG_MEDIACODEC) += mediacodecdec_common.o mediacodec_surface.o mediacodec_wrapper.o mediacodec_sw_buffer.o
131127 OBJS-$(CONFIG_MPEG_ER) += mpeg_er.o
@@ -157,7 +153,6 @@ OBJS-$(CONFIG_QSV) += qsv.o · 25 unmodified lines
157153 OBJS-$(CONFIG_QSVDEC) += qsvdec.o
158154 OBJS-$(CONFIG_QSVENC) += qsvenc.o
159155 OBJS-$(CONFIG_RANGECODER) += rangecoder.o
160−OBJS-$(CONFIG_RDFT) += rdft.o
161156 OBJS-$(CONFIG_RV34DSP) += rv34dsp.o
162157 OBJS-$(CONFIG_SINEWIN) += sinewin.o
163158 OBJS-$(CONFIG_SNAPPY) += snappy.o
@@ -1326,8 +1321,6 @@ TESTPROGS = avcodec \ · 1162 unmodified lines
13261321
13271322 TESTPROGS-$(CONFIG_AV1_VAAPI_ENCODER) += av1_levels
13281323 TESTPROGS-$(CONFIG_CABAC) += cabac
1329−TESTPROGS-$(CONFIG_DCT) += avfft
1330−TESTPROGS-$(CONFIG_FFT) += fft fft-fixed32
13311324 TESTPROGS-$(CONFIG_GOLOMB) += golomb
13321325 TESTPROGS-$(CONFIG_IDCTDSP) += dct
13331326 TESTPROGS-$(CONFIG_IIRFILTER) += iirfilter
@@ -1347,7 +1340,6 @@ HOSTPROGS = aacps_tablegen \ · 13 unmodified lines
13471340 aacps_fixed_tablegen \
13481341 cbrt_tablegen \
13491342 cbrt_fixed_tablegen \
1350− cos_tablegen \
13511343 dv_tablegen \
13521344 motionpixels_tablegen \
13531345 mpegaudio_tablegen \
@@ -1362,12 +1354,6 @@ CLEANFILES = *_tables.c *_tables.h *_tablegen$(HOSTEXESUF) · 8 unmodified lines
13621354 $(SUBDIR)tests/dct$(EXESUF): $(SUBDIR)dctref.o $(SUBDIR)aandcttab.o
13631355 $(SUBDIR)dv_tablegen$(HOSTEXESUF): $(SUBDIR)dvdata_host.o
13641356
1365−TRIG_TABLES = cos cos_fixed sin
1366−TRIG_TABLES := $(TRIG_TABLES:%=$(SUBDIR)%_tables.c)
1367−
1368−$(TRIG_TABLES): $(SUBDIR)%_tables.c: $(SUBDIR)cos_tablegen$(HOSTEXESUF)
1369− $(M)./$< $* > $@
1370−
13711357 ifdef CONFIG_SMALL
13721358 $(SUBDIR)%_tablegen$(HOSTEXESUF): HOSTCFLAGS += -DCONFIG_SMALL=1
13731359 else
libavcodec/aarch64/Makefile+0−3
@@ -1,5 +1,4 @@
11 # subsystems
2−OBJS-$(CONFIG_FFT) += aarch64/fft_init_aarch64.o
32 OBJS-$(CONFIG_FMTCONVERT) += aarch64/fmtconvert_init.o
43 OBJS-$(CONFIG_H264CHROMA) += aarch64/h264chroma_init_aarch64.o
54 OBJS-$(CONFIG_H264DSP) += aarch64/h264dsp_init_aarch64.o
@@ -36,7 +35,6 @@ ARMV8-OBJS-$(CONFIG_VIDEODSP) += aarch64/videodsp.o · 30 unmodified lines
3635
3736 # subsystems
3837 NEON-OBJS-$(CONFIG_AAC_DECODER) += aarch64/sbrdsp_neon.o
39−NEON-OBJS-$(CONFIG_FFT) += aarch64/fft_neon.o
4038 NEON-OBJS-$(CONFIG_FMTCONVERT) += aarch64/fmtconvert_neon.o
4139 NEON-OBJS-$(CONFIG_H264CHROMA) += aarch64/h264cmc_neon.o
4240 NEON-OBJS-$(CONFIG_H264DSP) += aarch64/h264dsp_neon.o \
@@ -47,7 +45,6 @@ NEON-OBJS-$(CONFIG_H264QPEL) += aarch64/h264qpel_neon.o \ · 4 unmodified lines
4745 NEON-OBJS-$(CONFIG_HPELDSP) += aarch64/hpeldsp_neon.o
4846 NEON-OBJS-$(CONFIG_IDCTDSP) += aarch64/idctdsp_neon.o \
4947 aarch64/simple_idct_neon.o
50−NEON-OBJS-$(CONFIG_MDCT) += aarch64/mdct_neon.o
5148 NEON-OBJS-$(CONFIG_ME_CMP) += aarch64/me_cmp_neon.o
5249 NEON-OBJS-$(CONFIG_MPEGAUDIODSP) += aarch64/mpegaudiodsp_neon.o
5350 NEON-OBJS-$(CONFIG_PIXBLOCKDSP) += aarch64/pixblockdsp_neon.o
libavcodec/aarch64/asm-offsets.h+0−25
@@ -1,25 +0,0 @@
1−/*
2− * This file is part of FFmpeg.
3− *
4− * FFmpeg is free software; you can redistribute it and/or
5− * modify it under the terms of the GNU Lesser General Public
6− * License as published by the Free Software Foundation; either
7− * version 2.1 of the License, or (at your option) any later version.
8− *
9− * FFmpeg is distributed in the hope that it will be useful,
10− * but WITHOUT ANY WARRANTY; without even the implied warranty of
11− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12− * Lesser General Public License for more details.
13− *
14− * You should have received a copy of the GNU Lesser General Public
15− * License along with FFmpeg; if not, write to the Free Software
16− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17− */
18−
19−#ifndef AVCODEC_AARCH64_ASM_OFFSETS_H
20−#define AVCODEC_AARCH64_ASM_OFFSETS_H
21−
22−/* FFTContext */
23−#define IMDCT_HALF 0x48
24−
25−#endif /* AVCODEC_AARCH64_ASM_OFFSETS_H */
libavcodec/aarch64/fft_init_aarch64.c+0−52
@@ -1,52 +0,0 @@
1−/*
2− * Copyright (c) 2009 Mans Rullgard <mans@mansr.com>
3− *
4− * This file is part of FFmpeg.
5− *
6− * FFmpeg is free software; you can redistribute it and/or
7− * modify it under the terms of the GNU Lesser General Public
8− * License as published by the Free Software Foundation; either
9− * version 2.1 of the License, or (at your option) any later version.
10− *
11− * FFmpeg is distributed in the hope that it will be useful,
12− * but WITHOUT ANY WARRANTY; without even the implied warranty of
13− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14− * Lesser General Public License for more details.
15− *
16− * You should have received a copy of the GNU Lesser General Public
17− * License along with FFmpeg; if not, write to the Free Software
18− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19− */
20−
21−#include "config.h"
22−
23−#include "libavutil/attributes.h"
24−#include "libavutil/cpu.h"
25−#include "libavutil/aarch64/cpu.h"
26−
27−#include "libavcodec/fft.h"
28−
29−void ff_fft_permute_neon(FFTContext *s, FFTComplex *z);
30−void ff_fft_calc_neon(FFTContext *s, FFTComplex *z);
31−
32−void ff_imdct_calc_neon(FFTContext *s, FFTSample *output, const FFTSample *input);
33−void ff_imdct_half_neon(FFTContext *s, FFTSample *output, const FFTSample *input);
34−void ff_mdct_calc_neon(FFTContext *s, FFTSample *output, const FFTSample *input);
35−
36−av_cold void ff_fft_init_aarch64(FFTContext *s)
37−{
38− int cpu_flags = av_get_cpu_flags();
39−
40− if (have_neon(cpu_flags)) {
41− if (s->nbits < 17) {
42− s->fft_permute = ff_fft_permute_neon;
43− s->fft_calc = ff_fft_calc_neon;
44− }
45−#if CONFIG_MDCT
46− s->imdct_calc = ff_imdct_calc_neon;
47− s->imdct_half = ff_imdct_half_neon;
48− s->mdct_calc = ff_mdct_calc_neon;
49− s->mdct_permutation = FF_MDCT_PERM_INTERLEAVE;
50−#endif
51− }
52−}
libavcodec/aarch64/fft_neon.S+0−447
@@ -1,447 +0,0 @@
1−/*
2− * ARM NEON optimised FFT
3− *
4− * Copyright (c) 2009 Mans Rullgard <mans@mansr.com>
5− * Copyright (c) 2009 Naotoshi Nojiri
6− * Copyright (c) 2014 Janne Grunau <janne-libav@jannau.net>
7− *
8− * This algorithm (though not any of the implementation details) is
9− * based on libdjbfft by D. J. Bernstein.
10− *
11− * This file is part of FFmpeg.
12− *
13− * FFmpeg is free software; you can redistribute it and/or
14− * modify it under the terms of the GNU Lesser General Public
15− * License as published by the Free Software Foundation; either
16− * version 2.1 of the License, or (at your option) any later version.
17− *
18− * FFmpeg is distributed in the hope that it will be useful,
19− * but WITHOUT ANY WARRANTY; without even the implied warranty of
20− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
21− * Lesser General Public License for more details.
22− *
23− * You should have received a copy of the GNU Lesser General Public
24− * License along with FFmpeg; if not, write to the Free Software
25− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26− */
27−
28−#include "libavutil/aarch64/asm.S"
29−
30−#define M_SQRT1_2 0.70710678118654752440
31−
32−.macro transpose d0, d1, s0, s1
33− trn1 \d0, \s0, \s1
34− trn2 \d1, \s0, \s1
35−.endm
36−
37−
38−function fft4_neon
39− AARCH64_VALID_JUMP_TARGET
40− ld1 {v0.2s,v1.2s,v2.2s,v3.2s}, [x0]
41−
42− fadd v4.2s, v0.2s, v1.2s // r0+r1,i0+i1
43− fsub v6.2s, v0.2s, v1.2s // r0-r1,i0-i1
44−
45− ext v16.8b, v2.8b, v3.8b, #4
46− ext v17.8b, v3.8b, v2.8b, #4
47−
48− fadd v5.2s, v2.2s, v3.2s // i2+i3,r2+r3
49− fsub v7.2s, v16.2s, v17.2s // r3-r2,i2-i3
50−
51− fadd v0.2s, v4.2s, v5.2s
52− fsub v2.2s, v4.2s, v5.2s
53− fadd v1.2s, v6.2s, v7.2s
54− fsub v3.2s, v6.2s, v7.2s
55−
56− st1 {v0.2s,v1.2s,v2.2s,v3.2s}, [x0]
57−
58− ret
59−endfunc
60−
61−function fft8_neon
62− AARCH64_VALID_JUMP_TARGET
63− mov x1, x0
64− ld1 {v0.2s, v1.2s, v2.2s, v3.2s}, [x0], #32
65− ld1 {v16.2s,v17.2s,v18.2s,v19.2s}, [x0]
66− ext v22.8b, v2.8b, v3.8b, #4
67− ext v23.8b, v3.8b, v2.8b, #4
68− fadd v4.2s, v16.2s, v17.2s // r4+r5,i4+i5
69− fadd v5.2s, v18.2s, v19.2s // r6+r7,i6+i7
70− fsub v17.2s, v16.2s, v17.2s // r4-r5,i4-i5
71− fsub v19.2s, v18.2s, v19.2s // r6-r7,i6-i7
72− rev64 v27.2s, v28.2s // ???
73− fadd v20.2s, v0.2s, v1.2s // r0+r1,i0+i1
74− fadd v21.2s, v2.2s, v3.2s // r2+r3,i2+i3
75− fmul v26.2s, v17.2s, v28.2s // -a2r*w,a2i*w
76− ext v6.8b, v4.8b, v5.8b, #4
77− ext v7.8b, v5.8b, v4.8b, #4
78− fmul v27.2s, v19.2s, v27.2s // a3r*w,-a3i*w
79− fsub v23.2s, v22.2s, v23.2s // i2-i3,r3-r2
80− fsub v22.2s, v0.2s, v1.2s // r0-r1,i0-i1
81− fmul v24.2s, v17.2s, v28.s[1] // a2r*w,a2i*w
82− fmul v25.2s, v19.2s, v28.s[1] // a3r*w,a3i*w
83− fadd v0.2s, v20.2s, v21.2s
84− fsub v2.2s, v20.2s, v21.2s
85− fadd v1.2s, v22.2s, v23.2s
86− rev64 v26.2s, v26.2s
87− rev64 v27.2s, v27.2s
88− fsub v3.2s, v22.2s, v23.2s
89− fsub v6.2s, v6.2s, v7.2s
90− fadd v24.2s, v24.2s, v26.2s // a2r+a2i,a2i-a2r t1,t2
91− fadd v25.2s, v25.2s, v27.2s // a3r-a3i,a3i+a3r t5,t6
92− fadd v7.2s, v4.2s, v5.2s
93− fsub v18.2s, v2.2s, v6.2s
94− ext v26.8b, v24.8b, v25.8b, #4
95− ext v27.8b, v25.8b, v24.8b, #4
96− fadd v2.2s, v2.2s, v6.2s
97− fsub v16.2s, v0.2s, v7.2s
98− fadd v5.2s, v25.2s, v24.2s
99− fsub v4.2s, v26.2s, v27.2s
100− fadd v0.2s, v0.2s, v7.2s
101− fsub v17.2s, v1.2s, v5.2s
102− fsub v19.2s, v3.2s, v4.2s
103− fadd v3.2s, v3.2s, v4.2s
104− fadd v1.2s, v1.2s, v5.2s
105−
106− st1 {v16.2s,v17.2s,v18.2s,v19.2s}, [x0]
107− st1 {v0.2s, v1.2s, v2.2s, v3.2s}, [x1]
108−
109− ret
110−endfunc
111−
112−function fft16_neon
113− AARCH64_VALID_JUMP_TARGET
114− mov x1, x0
115− ld1 {v0.2s, v1.2s, v2.2s, v3.2s}, [x0], #32
116− ld1 {v16.2s,v17.2s,v18.2s,v19.2s}, [x0], #32
117− ext v22.8b, v2.8b, v3.8b, #4
118− ext v23.8b, v3.8b, v2.8b, #4
119− fadd v4.2s, v16.2s, v17.2s // r4+r5,i4+i5
120− fadd v5.2s, v18.2s, v19.2s // r6+r7,i6+i7
121− fsub v17.2s, v16.2s, v17.2s // r4-r5,i4-i5
122− fsub v19.2s, v18.2s, v19.2s // r6-r7,i6-i7
123− rev64 v27.2s, v28.2s // ???
124− fadd v20.2s, v0.2s, v1.2s // r0+r1,i0+i1
125− fadd v21.2s, v2.2s, v3.2s // r2+r3,i2+i3
126− fmul v26.2s, v17.2s, v28.2s // -a2r*w,a2i*w
127− ext v6.8b, v4.8b, v5.8b, #4
128− ext v7.8b, v5.8b, v4.8b, #4
129− fmul v27.2s, v19.2s, v27.2s // a3r*w,-a3i*w
130− fsub v23.2s, v22.2s, v23.2s // i2-i3,r3-r2
131− fsub v22.2s, v0.2s, v1.2s // r0-r1,i0-i1
132− fmul v24.2s, v17.2s, v28.s[1] // a2r*w,a2i*w
133− fmul v25.2s, v19.2s, v28.s[1] // a3r*w,a3i*w
134− fadd v0.2s, v20.2s, v21.2s
135− fsub v2.2s, v20.2s, v21.2s
136− fadd v1.2s, v22.2s, v23.2s
137− rev64 v26.2s, v26.2s
138− rev64 v27.2s, v27.2s
139− fsub v3.2s, v22.2s, v23.2s
140− fsub v6.2s, v6.2s, v7.2s
141− fadd v24.2s, v24.2s, v26.2s // a2r+a2i,a2i-a2r t1,t2
142− fadd v25.2s, v25.2s, v27.2s // a3r-a3i,a3i+a3r t5,t6
143− fadd v7.2s, v4.2s, v5.2s
144− fsub v18.2s, v2.2s, v6.2s
145− ld1 {v20.4s,v21.4s}, [x0], #32
146− ld1 {v22.4s,v23.4s}, [x0], #32
147− ext v26.8b, v24.8b, v25.8b, #4
148− ext v27.8b, v25.8b, v24.8b, #4
149− fadd v2.2s, v2.2s, v6.2s
150− fsub v16.2s, v0.2s, v7.2s
151− fadd v5.2s, v25.2s, v24.2s
152− fsub v4.2s, v26.2s, v27.2s
153− transpose v24.2d, v25.2d, v20.2d, v22.2d
154− transpose v26.2d, v27.2d, v21.2d, v23.2d
155− fadd v0.2s, v0.2s, v7.2s
156− fsub v17.2s, v1.2s, v5.2s
157− fsub v19.2s, v3.2s, v4.2s
158− fadd v3.2s, v3.2s, v4.2s
159− fadd v1.2s, v1.2s, v5.2s
160− ext v20.16b, v21.16b, v21.16b, #4
161− ext v21.16b, v23.16b, v23.16b, #4
162−
163− zip1 v0.2d, v0.2d, v1.2d // {z[0], z[1]}
164− zip1 v1.2d, v2.2d, v3.2d // {z[2], z[3]}
165− zip1 v2.2d, v16.2d, v17.2d // {z[o1], z[o1+1]}
166− zip1 v3.2d, v18.2d, v19.2d // {z[o1+2],z[o1+3]}
167−
168− // 2 x fft4
169− transpose v22.2d, v23.2d, v20.2d, v21.2d
170−
171− fadd v4.4s, v24.4s, v25.4s
172− fadd v5.4s, v26.4s, v27.4s
173− fsub v6.4s, v24.4s, v25.4s
174− fsub v7.4s, v22.4s, v23.4s
175−
176− ld1 {v23.4s}, [x14]
177−
178− fadd v24.4s, v4.4s, v5.4s // {z[o2+0],z[o2+1]}
179− fsub v26.4s, v4.4s, v5.4s // {z[o2+2],z[o2+3]}
180− fadd v25.4s, v6.4s, v7.4s // {z[o3+0],z[o3+1]}
181− fsub v27.4s, v6.4s, v7.4s // {z[o3+2],z[o3+3]}
182−
183− //fft_pass_neon_16
184− rev64 v7.4s, v25.4s
185− fmul v25.4s, v25.4s, v23.s[1]
186− fmul v7.4s, v7.4s, v29.4s
187− fmla v25.4s, v7.4s, v23.s[3] // {t1a,t2a,t5a,t6a}
188−
189− zip1 v20.4s, v24.4s, v25.4s
190− zip2 v21.4s, v24.4s, v25.4s
191− fneg v22.4s, v20.4s
192− fadd v4.4s, v21.4s, v20.4s
193− fsub v6.4s, v20.4s, v21.4s // just the second half
194− fadd v5.4s, v21.4s, v22.4s // just the first half
195−
196− tbl v4.16b, {v4.16b}, v30.16b // trans4_float
197− tbl v5.16b, {v5.16b,v6.16b}, v31.16b // trans8_float
198−
199− fsub v20.4s, v0.4s, v4.4s // {z[o2],z[o2+1]}
200− fadd v16.4s, v0.4s, v4.4s // {z[0], z[1]}
201− fsub v22.4s, v2.4s, v5.4s // {z[o3],z[o3+1]}
202− fadd v18.4s, v2.4s, v5.4s // {z[o1],z[o1+1]}
203−
204−//second half
205− rev64 v6.4s, v26.4s
206− fmul v26.4s, v26.4s, v23.s[2]
207− rev64 v7.4s, v27.4s
208− fmul v27.4s, v27.4s, v23.s[3]
209− fmul v6.4s, v6.4s, v29.4s
210− fmul v7.4s, v7.4s, v29.4s
211− fmla v26.4s, v6.4s, v23.s[2] // {t1,t2,t5,t6}
212− fmla v27.4s, v7.4s, v23.s[1] // {t1a,t2a,t5a,t6a}
213−
214− zip1 v24.4s, v26.4s, v27.4s
215− zip2 v25.4s, v26.4s, v27.4s
216− fneg v26.4s, v24.4s
217− fadd v4.4s, v25.4s, v24.4s
218− fsub v6.4s, v24.4s, v25.4s // just the second half
219− fadd v5.4s, v25.4s, v26.4s // just the first half
220−
221− tbl v4.16b, {v4.16b}, v30.16b // trans4_float
222− tbl v5.16b, {v5.16b,v6.16b}, v31.16b // trans8_float
223−
224− fadd v17.4s, v1.4s, v4.4s // {z[2], z[3]}
225− fsub v21.4s, v1.4s, v4.4s // {z[o2+2],z[o2+3]}
226− fadd v19.4s, v3.4s, v5.4s // {z[o1+2],z[o1+3]}
227− fsub v23.4s, v3.4s, v5.4s // {z[o3+2],z[o3+3]}
228−
229− st1 {v16.4s,v17.4s}, [x1], #32
230− st1 {v18.4s,v19.4s}, [x1], #32
231− st1 {v20.4s,v21.4s}, [x1], #32
232− st1 {v22.4s,v23.4s}, [x1], #32
233−
234− ret
235−endfunc
236−
237−
238−const trans4_float, align=4
239− .byte 0, 1, 2, 3
240− .byte 8, 9, 10, 11
241− .byte 4, 5, 6, 7
242− .byte 12, 13, 14, 15
243−endconst
244−
245−const trans8_float, align=4
246− .byte 24, 25, 26, 27
247− .byte 0, 1, 2, 3
248− .byte 28, 29, 30, 31
249− .byte 4, 5, 6, 7
250−endconst
251−
252−function fft_pass_neon
253− sub x6, x2, #1 // n - 1, loop counter
254− lsl x5, x2, #3 // 2 * n * sizeof FFTSample
255− lsl x1, x2, #4 // 2 * n * sizeof FFTComplex
256− add x5, x4, x5 // wim
257− add x3, x1, x2, lsl #5 // 4 * n * sizeof FFTComplex
258− add x2, x0, x2, lsl #5 // &z[o2]
259− add x3, x0, x3 // &z[o3]
260− add x1, x0, x1 // &z[o1]
261− ld1 {v20.4s},[x2] // {z[o2],z[o2+1]}
262− ld1 {v22.4s},[x3] // {z[o3],z[o3+1]}
263− ld1 {v4.2s}, [x4], #8 // {wre[0],wre[1]}
264− trn2 v25.2d, v20.2d, v22.2d
265− sub x5, x5, #4 // wim--
266− trn1 v24.2d, v20.2d, v22.2d
267− ld1 {v5.s}[0], [x5], x7 // d5[0] = wim[-1]
268− rev64 v7.4s, v25.4s
269− fmul v25.4s, v25.4s, v4.s[1]
270− ld1 {v16.4s}, [x0] // {z[0],z[1]}
271− fmul v7.4s, v7.4s, v29.4s
272− ld1 {v17.4s}, [x1] // {z[o1],z[o1+1]}
273− prfm pldl1keep, [x2, #16]
274− prfm pldl1keep, [x3, #16]
275− fmla v25.4s, v7.4s, v5.s[0] // {t1a,t2a,t5a,t6a}
276− prfm pldl1keep, [x0, #16]
277− prfm pldl1keep, [x1, #16]
278−
279− zip1 v20.4s, v24.4s, v25.4s
280− zip2 v21.4s, v24.4s, v25.4s
281− fneg v22.4s, v20.4s
282− fadd v4.4s, v21.4s, v20.4s
283− fsub v6.4s, v20.4s, v21.4s // just the second half
284− fadd v5.4s, v21.4s, v22.4s // just the first half
285−
286− tbl v4.16b, {v4.16b}, v30.16b // trans4_float
287− tbl v5.16b, {v5.16b,v6.16b}, v31.16b // trans8_float
288−
289− fadd v20.4s, v16.4s, v4.4s
290− fsub v22.4s, v16.4s, v4.4s
291− fadd v21.4s, v17.4s, v5.4s
292− st1 {v20.4s}, [x0], #16 // {z[0], z[1]}
293− fsub v23.4s, v17.4s, v5.4s
294−
295− st1 {v21.4s}, [x1], #16 // {z[o1],z[o1+1]}
296− st1 {v22.4s}, [x2], #16 // {z[o2],z[o2+1]}
297− st1 {v23.4s}, [x3], #16 // {z[o3],z[o3+1]}
298−1:
299− ld1 {v20.4s},[x2] // {z[o2],z[o2+1]}
300− ld1 {v22.4s},[x3] // {z[o3],z[o3+1]}
301− ld1 {v4.2s}, [x4], #8 // {wre[0],wre[1]}
302− transpose v26.2d, v27.2d, v20.2d, v22.2d
303− ld1 {v5.2s}, [x5], x7 // {wim[-1],wim[0]}
304− rev64 v6.4s, v26.4s
305− fmul v26.4s, v26.4s, v4.s[0]
306− rev64 v7.4s, v27.4s
307− fmul v27.4s, v27.4s, v4.s[1]
308− fmul v6.4s, v6.4s, v29.4s
309− fmul v7.4s, v7.4s, v29.4s
310− ld1 {v16.4s},[x0] // {z[0],z[1]}
311− fmla v26.4s, v6.4s, v5.s[1] // {t1,t2,t5,t6}
312− fmla v27.4s, v7.4s, v5.s[0] // {t1a,t2a,t5a,t6a}
313− ld1 {v17.4s},[x1] // {z[o1],z[o1+1]}
314−
315− subs x6, x6, #1 // n--
316−
317− zip1 v20.4s, v26.4s, v27.4s
318− zip2 v21.4s, v26.4s, v27.4s
319− fneg v22.4s, v20.4s
320− fadd v4.4s, v21.4s, v20.4s
321− fsub v6.4s, v20.4s, v21.4s // just the second half
322− fadd v5.4s, v21.4s, v22.4s // just the first half
323−
324− tbl v4.16b, {v4.16b}, v30.16b // trans4_float
325− tbl v5.16b, {v5.16b,v6.16b}, v31.16b // trans8_float
326−
327− fadd v20.4s, v16.4s, v4.4s
328− fsub v22.4s, v16.4s, v4.4s
329− fadd v21.4s, v17.4s, v5.4s
330− st1 {v20.4s}, [x0], #16 // {z[0], z[1]}
331− fsub v23.4s, v17.4s, v5.4s
332−
333− st1 {v21.4s}, [x1], #16 // {z[o1],z[o1+1]}
334− st1 {v22.4s}, [x2], #16 // {z[o2],z[o2+1]}
335− st1 {v23.4s}, [x3], #16 // {z[o3],z[o3+1]}
336− b.ne 1b
337−
338− ret
339−endfunc
340−
341−.macro def_fft n, n2, n4
342−function fft\n\()_neon, align=6
343− AARCH64_VALID_JUMP_TARGET
344− AARCH64_SIGN_LINK_REGISTER
345− stp x28, x30, [sp, #-16]!
346− add x28, x0, #\n4*2*8
347− bl fft\n2\()_neon
348− mov x0, x28
349− bl fft\n4\()_neon
350− add x0, x28, #\n4*1*8
351− bl fft\n4\()_neon
352− sub x0, x28, #\n4*2*8
353− ldp x28, x30, [sp], #16
354− AARCH64_VALIDATE_LINK_REGISTER
355− movrel x4, X(ff_cos_\n)
356− mov x2, #\n4>>1
357− b fft_pass_neon
358−endfunc
359−.endm
360−
361− def_fft 32, 16, 8
362− def_fft 64, 32, 16
363− def_fft 128, 64, 32
364− def_fft 256, 128, 64
365− def_fft 512, 256, 128
366− def_fft 1024, 512, 256
367− def_fft 2048, 1024, 512
368− def_fft 4096, 2048, 1024
369− def_fft 8192, 4096, 2048
370− def_fft 16384, 8192, 4096
371− def_fft 32768, 16384, 8192
372− def_fft 65536, 32768, 16384
373−
374−function ff_fft_calc_neon, export=1
375− prfm pldl1keep, [x1]
376− movrel x10, trans4_float
377− ldr w2, [x0]
378− movrel x11, trans8_float
379− sub w2, w2, #2
380− movrel x3, fft_tab_neon
381− ld1 {v30.16b}, [x10]
382− mov x7, #-8
383− movrel x12, pmmp
384− ldr x3, [x3, x2, lsl #3]
385− movrel x13, mppm
386− movrel x14, X(ff_cos_16)
387− ld1 {v31.16b}, [x11]
388− mov x0, x1
389− ld1 {v29.4s}, [x12] // pmmp
390− ld1 {v28.4s}, [x13]
391− br x3
392−endfunc
393−
394−function ff_fft_permute_neon, export=1
395− mov x6, #1
396− ldr w2, [x0] // nbits
397− ldr x3, [x0, #16] // tmp_buf
398− ldr x0, [x0, #8] // revtab
399− lsl x6, x6, x2
400− mov x2, x6
401−1:
402− ld1 {v0.2s,v1.2s}, [x1], #16
403− ldr w4, [x0], #4
404− uxth w5, w4
405− lsr w4, w4, #16
406− add x5, x3, x5, lsl #3
407− add x4, x3, x4, lsl #3
408− st1 {v0.2s}, [x5]
409− st1 {v1.2s}, [x4]
410− subs x6, x6, #2
411− b.gt 1b
412−
413− sub x1, x1, x2, lsl #3
414−1:
415− ld1 {v0.4s,v1.4s}, [x3], #32
416− st1 {v0.4s,v1.4s}, [x1], #32
417− subs x2, x2, #4
418− b.gt 1b
419−
420− ret
421−endfunc
422−
423−const fft_tab_neon, relocate=1
424− .quad fft4_neon
425− .quad fft8_neon
426− .quad fft16_neon
427− .quad fft32_neon
428− .quad fft64_neon
429− .quad fft128_neon
430− .quad fft256_neon
431− .quad fft512_neon
432− .quad fft1024_neon
433− .quad fft2048_neon
434− .quad fft4096_neon
435− .quad fft8192_neon
436− .quad fft16384_neon
437− .quad fft32768_neon
438− .quad fft65536_neon
439−endconst
440−
441−const pmmp, align=4
442− .float +1.0, -1.0, -1.0, +1.0
443−endconst
444−
445−const mppm, align=4
446− .float -M_SQRT1_2, M_SQRT1_2, M_SQRT1_2, -M_SQRT1_2
447−endconst
libavcodec/aarch64/mdct_neon.S+0−326
@@ -1,326 +0,0 @@
1−/*
2− * AArch64 NEON optimised MDCT
3− * Copyright (c) 2009 Mans Rullgard <mans@mansr.com>
4− * Copyright (c) 2014 Janne Grunau <janne-libav@jannau.net>
5− *
6− * This file is part of FFmpeg.
7− *
8− * FFmpeg is free software; you can redistribute it and/or
9− * modify it under the terms of the GNU Lesser General Public
10− * License as published by the Free Software Foundation; either
11− * version 2.1 of the License, or (at your option) any later version.
12− *
13− * FFmpeg is distributed in the hope that it will be useful,
14− * but WITHOUT ANY WARRANTY; without even the implied warranty of
15− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16− * Lesser General Public License for more details.
17− *
18− * You should have received a copy of the GNU Lesser General Public
19− * License along with FFmpeg; if not, write to the Free Software
20− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21− */
22−
23−#include "libavutil/aarch64/asm.S"
24−
25−function ff_imdct_half_neon, export=1
26− stp x19, x20, [sp, #-32]!
27− AARCH64_SIGN_LINK_REGISTER
28− str x30, [sp, #16]
29− mov x12, #1
30− ldr w14, [x0, #28] // mdct_bits
31− ldr x4, [x0, #32] // tcos
32− ldr x3, [x0, #8] // revtab
33− lsl x12, x12, x14 // n = 1 << nbits
34− lsr x14, x12, #2 // n4 = n >> 2
35− add x7, x2, x12, lsl #1
36− mov x12, #-16
37− sub x7, x7, #16
38−
39− ld2 {v16.2s,v17.2s}, [x7], x12 // d16=x,n1 d17=x,n0
40− ld2 {v0.2s,v1.2s}, [x2], #16 // d0 =m0,x d1 =m1,x
41− rev64 v17.2s, v17.2s
42− ld2 {v2.2s,v3.2s}, [x4], #16 // d2=c0,c1 d3=s0,s2
43− fmul v6.2s, v17.2s, v2.2s
44− fmul v7.2s, v0.2s, v2.2s
45−1:
46− subs x14, x14, #2
47− ldr w6, [x3], #4
48− fmul v4.2s, v0.2s, v3.2s
49− fmul v5.2s, v17.2s, v3.2s
50− fsub v4.2s, v6.2s, v4.2s
51− fadd v5.2s, v5.2s, v7.2s
52− ubfm x8, x6, #16, #31
53− ubfm x6, x6, #0, #15
54− add x8, x1, x8, lsl #3
55− add x6, x1, x6, lsl #3
56− b.eq 2f
57− ld2 {v16.2s,v17.2s}, [x7], x12
58− ld2 {v0.2s,v1.2s}, [x2], #16
59− rev64 v17.2s, v17.2s
60− ld2 {v2.2s,v3.2s}, [x4], #16 // d2=c0,c1 d3=s0,s2
61− fmul v6.2s, v17.2s, v2.2s
62− fmul v7.2s, v0.2s, v2.2s
63− st2 {v4.s,v5.s}[0], [x6]
64− st2 {v4.s,v5.s}[1], [x8]
65− b 1b
66−2:
67− st2 {v4.s,v5.s}[0], [x6]
68− st2 {v4.s,v5.s}[1], [x8]
69−
70− mov x19, x0
71− mov x20, x1
72− bl X(ff_fft_calc_neon)
73−
74− mov x12, #1
75− ldr w14, [x19, #28] // mdct_bits
76− ldr x4, [x19, #32] // tcos
77− lsl x12, x12, x14 // n = 1 << nbits
78− lsr x14, x12, #3 // n8 = n >> 3
79−
80− add x4, x4, x14, lsl #3
81− add x6, x20, x14, lsl #3
82− sub x1, x4, #16
83− sub x3, x6, #16
84−
85− mov x7, #-16
86− mov x8, x6
87− mov x0, x3
88−
89− ld2 {v0.2s,v1.2s}, [x3], x7 // d0 =i1,r1 d1 =i0,r0
90− ld2 {v20.2s,v21.2s},[x6], #16 // d20=i2,r2 d21=i3,r3
91− ld2 {v16.2s,v17.2s},[x1], x7 // d16=c1,c0 d18=s1,s0
92−3:
93− subs x14, x14, #2
94− fmul v7.2s, v0.2s, v17.2s
95− ld2 {v18.2s,v19.2s},[x4], #16 // d17=c2,c3 d19=s2,s3
96− fmul v4.2s, v1.2s, v17.2s
97− fmul v6.2s, v21.2s, v19.2s
98− fmul v5.2s, v20.2s, v19.2s
99− fmul v22.2s, v1.2s, v16.2s
100− fmul v23.2s, v21.2s, v18.2s
101− fmul v24.2s, v0.2s, v16.2s
102− fmul v25.2s, v20.2s, v18.2s
103− fadd v7.2s, v7.2s, v22.2s
104− fadd v5.2s, v5.2s, v23.2s
105− fsub v4.2s, v4.2s, v24.2s
106− fsub v6.2s, v6.2s, v25.2s
107− b.eq 4f
108− ld2 {v0.2s,v1.2s}, [x3], x7
109− ld2 {v20.2s,v21.2s},[x6], #16
110− ld2 {v16.2s,v17.2s},[x1], x7 // d16=c1,c0 d18=s1,s0
111− rev64 v5.2s, v5.2s
112− rev64 v7.2s, v7.2s
113− st2 {v4.2s,v5.2s}, [x0], x7
114− st2 {v6.2s,v7.2s}, [x8], #16
115− b 3b
116−4:
117− rev64 v5.2s, v5.2s
118− rev64 v7.2s, v7.2s
119− st2 {v4.2s,v5.2s}, [x0]
120− st2 {v6.2s,v7.2s}, [x8]
121−
122− ldr x30, [sp, #16]
123− AARCH64_VALIDATE_LINK_REGISTER
124− ldp x19, x20, [sp], #32
125−
126− ret
127−endfunc
128−
129−function ff_imdct_calc_neon, export=1
130− stp x19, x20, [sp, #-32]!
131− AARCH64_SIGN_LINK_REGISTER
132− str x30, [sp, #16]
133− ldr w3, [x0, #28] // mdct_bits
134− mov x19, #1
135− mov x20, x1
136− lsl x19, x19, x3
137− add x1, x1, x19
138−
139− bl X(ff_imdct_half_neon)
140−
141− add x0, x20, x19, lsl #2
142− add x1, x20, x19, lsl #1
143− sub x0, x0, #8
144− sub x2, x1, #16
145− mov x3, #-16
146− mov x6, #-8
147−1:
148− ld1 {v0.4s}, [x2], x3
149− prfum pldl1keep, [x0, #-16]
150− rev64 v0.4s, v0.4s
151− ld1 {v2.2s,v3.2s}, [x1], #16
152− fneg v4.4s, v0.4s
153− prfum pldl1keep, [x2, #-16]
154− rev64 v2.2s, v2.2s
155− rev64 v3.2s, v3.2s
156− ext v4.16b, v4.16b, v4.16b, #8
157− st1 {v2.2s}, [x0], x6
158− st1 {v3.2s}, [x0], x6
159− st1 {v4.4s}, [x20], #16
160− subs x19, x19, #16
161− b.gt 1b
162−
163− ldr x30, [sp, #16]
164− AARCH64_VALIDATE_LINK_REGISTER
165− ldp x19, x20, [sp], #32
166−
167− ret
168−endfunc
169−
170−
171−function ff_mdct_calc_neon, export=1
172− stp x19, x20, [sp, #-32]!
173− AARCH64_SIGN_LINK_REGISTER
174− str x30, [sp, #16]
175−
176− mov x12, #1
177− ldr w14, [x0, #28] // mdct_bits
178− ldr x4, [x0, #32] // tcos
179− ldr x3, [x0, #8] // revtab
180− lsl x14, x12, x14 // n = 1 << nbits
181− add x7, x2, x14 // in4u
182− sub x9, x7, #16 // in4d
183− add x2, x7, x14, lsl #1 // in3u
184− add x8, x9, x14, lsl #1 // in3d
185− add x5, x4, x14, lsl #1
186− sub x5, x5, #16
187− sub x3, x3, #4
188− mov x12, #-16
189− lsr x13, x14, #1
190−
191− ld2 {v16.2s,v17.2s}, [x9], x12 // in0u0,in0u1 in4d1,in4d0
192− ld2 {v18.2s,v19.2s}, [x8], x12 // in2u0,in2u1 in3d1,in3d0
193− ld2 {v0.2s, v1.2s}, [x7], #16 // in4u0,in4u1 in2d1,in2d0
194− rev64 v17.2s, v17.2s // in4d0,in4d1 in3d0,in3d1
195− rev64 v19.2s, v19.2s // in4d0,in4d1 in3d0,in3d1
196− ld2 {v2.2s, v3.2s}, [x2], #16 // in3u0,in3u1 in1d1,in1d0
197− fsub v0.2s, v17.2s, v0.2s // in4d-in4u I
198− ld2 {v20.2s,v21.2s}, [x4], #16 // c0,c1 s0,s1
199− rev64 v1.2s, v1.2s // in2d0,in2d1 in1d0,in1d1
200− rev64 v3.2s, v3.2s // in2d0,in2d1 in1d0,in1d1
201− ld2 {v30.2s,v31.2s}, [x5], x12 // c2,c3 s2,s3
202− fadd v2.2s, v2.2s, v19.2s // in3u+in3d -R
203− fsub v16.2s, v16.2s, v1.2s // in0u-in2d R
204− fadd v18.2s, v18.2s, v3.2s // in2u+in1d -I
205−1:
206− fmul v7.2s, v0.2s, v21.2s // I*s
207− ldr w10, [x3, x13]
208− fmul v6.2s, v2.2s, v20.2s // -R*c
209− ldr w6, [x3, #4]!
210− fmul v4.2s, v2.2s, v21.2s // -R*s
211− fmul v5.2s, v0.2s, v20.2s // I*c
212− fmul v24.2s, v16.2s, v30.2s // R*c
213− fmul v25.2s, v18.2s, v31.2s // -I*s
214− fmul v22.2s, v16.2s, v31.2s // R*s
215− fmul v23.2s, v18.2s, v30.2s // I*c
216− subs x14, x14, #16
217− subs x13, x13, #8
218− fsub v6.2s, v6.2s, v7.2s // -R*c-I*s
219− fadd v7.2s, v4.2s, v5.2s // -R*s+I*c
220− fsub v24.2s, v25.2s, v24.2s // I*s-R*c
221− fadd v25.2s, v22.2s, v23.2s // R*s-I*c
222− b.eq 1f
223− mov x12, #-16
224− ld2 {v16.2s,v17.2s}, [x9], x12 // in0u0,in0u1 in4d1,in4d0
225− ld2 {v18.2s,v19.2s}, [x8], x12 // in2u0,in2u1 in3d1,in3d0
226− fneg v7.2s, v7.2s // R*s-I*c
227− ld2 {v0.2s, v1.2s}, [x7], #16 // in4u0,in4u1 in2d1,in2d0
228− rev64 v17.2s, v17.2s // in4d0,in4d1 in3d0,in3d1
229− rev64 v19.2s, v19.2s // in4d0,in4d1 in3d0,in3d1
230− ld2 {v2.2s, v3.2s}, [x2], #16 // in3u0,in3u1 in1d1,in1d0
231− fsub v0.2s, v17.2s, v0.2s // in4d-in4u I
232− ld2 {v20.2s,v21.2s}, [x4], #16 // c0,c1 s0,s1
233− rev64 v1.2s, v1.2s // in2d0,in2d1 in1d0,in1d1
234− rev64 v3.2s, v3.2s // in2d0,in2d1 in1d0,in1d1
235− ld2 {v30.2s,v31.2s}, [x5], x12 // c2,c3 s2,s3
236− fadd v2.2s, v2.2s, v19.2s // in3u+in3d -R
237− fsub v16.2s, v16.2s, v1.2s // in0u-in2d R
238− fadd v18.2s, v18.2s, v3.2s // in2u+in1d -I
239− ubfm x12, x6, #16, #31
240− ubfm x6, x6, #0, #15
241− add x12, x1, x12, lsl #3
242− add x6, x1, x6, lsl #3
243− st2 {v6.s,v7.s}[0], [x6]
244− st2 {v6.s,v7.s}[1], [x12]
245− ubfm x6, x10, #16, #31
246− ubfm x10, x10, #0, #15
247− add x6 , x1, x6, lsl #3
248− add x10, x1, x10, lsl #3
249− st2 {v24.s,v25.s}[0], [x10]
250− st2 {v24.s,v25.s}[1], [x6]
251− b 1b
252−1:
253− fneg v7.2s, v7.2s // R*s-I*c
254− ubfm x12, x6, #16, #31
255− ubfm x6, x6, #0, #15
256− add x12, x1, x12, lsl #3
257− add x6, x1, x6, lsl #3
258− st2 {v6.s,v7.s}[0], [x6]
259− st2 {v6.s,v7.s}[1], [x12]
260− ubfm x6, x10, #16, #31
261− ubfm x10, x10, #0, #15
262− add x6 , x1, x6, lsl #3
263− add x10, x1, x10, lsl #3
264− st2 {v24.s,v25.s}[0], [x10]
265− st2 {v24.s,v25.s}[1], [x6]
266−
267− mov x19, x0
268− mov x20, x1
269− bl X(ff_fft_calc_neon)
270−
271− mov x12, #1
272− ldr w14, [x19, #28] // mdct_bits
273− ldr x4, [x19, #32] // tcos
274− lsl x12, x12, x14 // n = 1 << nbits
275− lsr x14, x12, #3 // n8 = n >> 3
276−
277− add x4, x4, x14, lsl #3
278− add x6, x20, x14, lsl #3
279− sub x1, x4, #16
280− sub x3, x6, #16
281−
282− mov x7, #-16
283− mov x8, x6
284− mov x0, x3
285−
286− ld2 {v0.2s,v1.2s}, [x3], x7 // d0 =r1,i1 d1 =r0,i0
287− ld2 {v20.2s,v21.2s}, [x6], #16 // d20=r2,i2 d21=r3,i3
288− ld2 {v16.2s,v17.2s}, [x1], x7 // c1,c0 s1,s0
289−1:
290− subs x14, x14, #2
291− fmul v7.2s, v0.2s, v17.2s // r1*s1,r0*s0
292− ld2 {v18.2s,v19.2s}, [x4], #16 // c2,c3 s2,s3
293− fmul v4.2s, v1.2s, v17.2s // i1*s1,i0*s0
294− fmul v6.2s, v21.2s, v19.2s // i2*s2,i3*s3
295− fmul v5.2s, v20.2s, v19.2s // r2*s2,r3*s3
296− fmul v24.2s, v0.2s, v16.2s // r1*c1,r0*c0
297− fmul v25.2s, v20.2s, v18.2s // r2*c2,r3*c3
298− fmul v22.2s, v21.2s, v18.2s // i2*c2,i3*c3
299− fmul v23.2s, v1.2s, v16.2s // i1*c1,i0*c0
300− fadd v4.2s, v4.2s, v24.2s // i1*s1+r1*c1,i0*s0+r0*c0
301− fadd v6.2s, v6.2s, v25.2s // i2*s2+r2*c2,i3*s3+r3*c3
302− fsub v5.2s, v22.2s, v5.2s // i2*c2-r2*s2,i3*c3-r3*s3
303− fsub v7.2s, v23.2s, v7.2s // i1*c1-r1*s1,i0*c0-r0*s0
304− fneg v4.2s, v4.2s
305− fneg v6.2s, v6.2s
306− b.eq 1f
307− ld2 {v0.2s, v1.2s}, [x3], x7
308− ld2 {v20.2s,v21.2s}, [x6], #16
309− ld2 {v16.2s,v17.2s}, [x1], x7 // c1,c0 s1,s0
310− rev64 v5.2s, v5.2s
311− rev64 v7.2s, v7.2s
312− st2 {v4.2s,v5.2s}, [x0], x7
313− st2 {v6.2s,v7.2s}, [x8], #16
314− b 1b
315−1:
316− rev64 v5.2s, v5.2s
317− rev64 v7.2s, v7.2s
318− st2 {v4.2s,v5.2s}, [x0]
319− st2 {v6.2s,v7.2s}, [x8]
320−
321− ldr x30, [sp, #16]
322− AARCH64_VALIDATE_LINK_REGISTER
323− ldp x19, x20, [sp], #32
324−
325− ret
326−endfunc
libavcodec/aarch64/synth_filter_init.c+0−7
@@ -23,15 +23,8 @@ · 22 unmodified lines
2323 #include "libavutil/aarch64/cpu.h"
2424 #include "libavutil/attributes.h"
2525 #include "libavutil/internal.h"
26−#include "libavcodec/fft.h"
2726 #include "libavcodec/synth_filter.h"
2827
29−#include "asm-offsets.h"
30−
31−#if HAVE_NEON
32−AV_CHECK_OFFSET(FFTContext, imdct_half, IMDCT_HALF);
33−#endif
34−
3528 void ff_synth_filter_float_neon(AVTXContext *imdct,
3629 float *synth_buf_ptr, int *synth_buf_offset,
3730 float synth_buf2[32], const float window[512],
libavcodec/aarch64/synth_filter_neon.S+0−2
@@ -19,8 +19,6 @@ · 18 unmodified lines
1919 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
2020 */
2121
22−#include "asm-offsets.h"
23−
2422 #include "libavutil/aarch64/asm.S"
2523
2624 .macro inner_loop
libavcodec/arm/Makefile+0−7
@@ -5,7 +5,6 @@ OBJS-$(CONFIG_AC3DSP) += arm/ac3dsp_init_arm.o \ · 4 unmodified lines
55 arm/ac3dsp_arm.o
66 OBJS-$(CONFIG_AUDIODSP) += arm/audiodsp_init_arm.o
77 OBJS-$(CONFIG_BLOCKDSP) += arm/blockdsp_init_arm.o
8−OBJS-$(CONFIG_FFT) += arm/fft_init_arm.o
98 OBJS-$(CONFIG_FMTCONVERT) += arm/fmtconvert_init_arm.o
109 OBJS-$(CONFIG_G722DSP) += arm/g722dsp_init_arm.o
1110 OBJS-$(CONFIG_H264CHROMA) += arm/h264chroma_init_arm.o
@@ -25,7 +24,6 @@ OBJS-$(CONFIG_MPEGVIDEO) += arm/mpegvideo_arm.o · 13 unmodified lines
2524 OBJS-$(CONFIG_MPEGVIDEOENC) += arm/mpegvideoencdsp_init_arm.o
2625 OBJS-$(CONFIG_NEON_CLOBBER_TEST) += arm/neontest.o
2726 OBJS-$(CONFIG_PIXBLOCKDSP) += arm/pixblockdsp_init_arm.o
28−OBJS-$(CONFIG_RDFT) += arm/rdft_init_arm.o
2927 OBJS-$(CONFIG_RV34DSP) += arm/rv34dsp_init_arm.o
3028 OBJS-$(CONFIG_VC1DSP) += arm/vc1dsp_init_arm.o
3129 OBJS-$(CONFIG_VIDEODSP) += arm/videodsp_init_arm.o
@@ -90,9 +88,7 @@ ARMV6-OBJS-$(CONFIG_TRUEHD_DECODER) += arm/mlpdsp_armv6.o · 58 unmodified lines
9088 # VFP optimizations
9189
9290 # subsystems
93−VFP-OBJS-$(CONFIG_FFT) += arm/fft_vfp.o
9491 VFP-OBJS-$(CONFIG_FMTCONVERT) += arm/fmtconvert_vfp.o
95−VFP-OBJS-$(CONFIG_MDCT) += arm/mdct_vfp.o
9692
9793 # decoders/encoders
9894 VFP-OBJS-$(CONFIG_DCA_DECODER) += arm/synth_filter_vfp.o
@@ -107,7 +103,6 @@ NEON-OBJS-$(CONFIG_AUDIODSP) += arm/audiodsp_init_neon.o \ · 8 unmodified lines
107103 arm/int_neon.o
108104 NEON-OBJS-$(CONFIG_BLOCKDSP) += arm/blockdsp_init_neon.o \
109105 arm/blockdsp_neon.o
110−NEON-OBJS-$(CONFIG_FFT) += arm/fft_neon.o
111106 NEON-OBJS-$(CONFIG_FMTCONVERT) += arm/fmtconvert_neon.o
112107 NEON-OBJS-$(CONFIG_G722DSP) += arm/g722dsp_neon.o
113108 NEON-OBJS-$(CONFIG_H264CHROMA) += arm/h264cmc_neon.o
@@ -121,10 +116,8 @@ NEON-OBJS-$(CONFIG_HPELDSP) += arm/hpeldsp_init_neon.o \ · 7 unmodified lines
121116 NEON-OBJS-$(CONFIG_IDCTDSP) += arm/idctdsp_init_neon.o \
122117 arm/idctdsp_neon.o \
123118 arm/simple_idct_neon.o
124−NEON-OBJS-$(CONFIG_MDCT) += arm/mdct_neon.o
125119 NEON-OBJS-$(CONFIG_MPEGVIDEO) += arm/mpegvideo_neon.o
126120 NEON-OBJS-$(CONFIG_PIXBLOCKDSP) += arm/pixblockdsp_neon.o
127−NEON-OBJS-$(CONFIG_RDFT) += arm/rdft_neon.o
128121 NEON-OBJS-$(CONFIG_VC1DSP) += arm/vc1dsp_init_neon.o \
129122 arm/vc1dsp_neon.o
130123 NEON-OBJS-$(CONFIG_VP3DSP) += arm/vp3dsp_neon.o
libavcodec/arm/fft_init_arm.c+0−63
@@ -1,63 +0,0 @@
1−/*
2− * Copyright (c) 2009 Mans Rullgard <mans@mansr.com>
3− *
4− * This file is part of FFmpeg.
5− *
6− * FFmpeg is free software; you can redistribute it and/or
7− * modify it under the terms of the GNU Lesser General Public
8− * License as published by the Free Software Foundation; either
9− * version 2.1 of the License, or (at your option) any later version.
10− *
11− * FFmpeg is distributed in the hope that it will be useful,
12− * but WITHOUT ANY WARRANTY; without even the implied warranty of
13− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14− * Lesser General Public License for more details.
15− *
16− * You should have received a copy of the GNU Lesser General Public
17− * License along with FFmpeg; if not, write to the Free Software
18− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19− */
20−
21−#include "libavutil/attributes.h"
22−#include "libavutil/cpu.h"
23−#include "libavutil/arm/cpu.h"
24−
25−#include "libavcodec/fft.h"
26−
27−void ff_fft_calc_vfp(FFTContext *s, FFTComplex *z);
28−
29−void ff_fft_permute_neon(FFTContext *s, FFTComplex *z);
30−void ff_fft_calc_neon(FFTContext *s, FFTComplex *z);
31−
32−void ff_imdct_half_vfp(FFTContext *s, FFTSample *output, const FFTSample *input);
33−
34−void ff_imdct_calc_neon(FFTContext *s, FFTSample *output, const FFTSample *input);
35−void ff_imdct_half_neon(FFTContext *s, FFTSample *output, const FFTSample *input);
36−void ff_mdct_calc_neon(FFTContext *s, FFTSample *output, const FFTSample *input);
37−
38−av_cold void ff_fft_init_arm(FFTContext *s)
39−{
40− int cpu_flags = av_get_cpu_flags();
41−
42− if (have_vfp_vm(cpu_flags)) {
43− s->fft_calc = ff_fft_calc_vfp;
44−#if CONFIG_MDCT
45− s->imdct_half = ff_imdct_half_vfp;
46−#endif
47− }
48−
49− if (have_neon(cpu_flags)) {
50−#if CONFIG_FFT
51− if (s->nbits < 17) {
52− s->fft_permute = ff_fft_permute_neon;
53− s->fft_calc = ff_fft_calc_neon;
54− }
55−#endif
56−#if CONFIG_MDCT
57− s->imdct_calc = ff_imdct_calc_neon;
58− s->imdct_half = ff_imdct_half_neon;
59− s->mdct_calc = ff_mdct_calc_neon;
60− s->mdct_permutation = FF_MDCT_PERM_INTERLEAVE;
61−#endif
62− }
63−}
libavcodec/arm/fft_neon.S+0−375
@@ -1,375 +0,0 @@
1−/*
2− * ARM NEON optimised FFT
3− *
4− * Copyright (c) 2009 Mans Rullgard <mans@mansr.com>
5− * Copyright (c) 2009 Naotoshi Nojiri
6− *
7− * This algorithm (though not any of the implementation details) is
8− * based on libdjbfft by D. J. Bernstein.
9− *
10− * This file is part of FFmpeg.
11− *
12− * FFmpeg is free software; you can redistribute it and/or
13− * modify it under the terms of the GNU Lesser General Public
14− * License as published by the Free Software Foundation; either
15− * version 2.1 of the License, or (at your option) any later version.
16− *
17− * FFmpeg is distributed in the hope that it will be useful,
18− * but WITHOUT ANY WARRANTY; without even the implied warranty of
19− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20− * Lesser General Public License for more details.
21− *
22− * You should have received a copy of the GNU Lesser General Public
23− * License along with FFmpeg; if not, write to the Free Software
24− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25− */
26−
27−#include "libavutil/arm/asm.S"
28−
29−#define M_SQRT1_2 0.70710678118654752440
30−
31−
32−function fft4_neon
33− vld1.32 {d0-d3}, [r0,:128]
34−
35− vext.32 q8, q1, q1, #1 @ i2,r3 d3=i3,r2
36− vsub.f32 d6, d0, d1 @ r0-r1,i0-i1
37− vsub.f32 d7, d16, d17 @ r3-r2,i2-i3
38− vadd.f32 d4, d0, d1 @ r0+r1,i0+i1
39− vadd.f32 d5, d2, d3 @ i2+i3,r2+r3
40− vadd.f32 d1, d6, d7
41− vsub.f32 d3, d6, d7
42− vadd.f32 d0, d4, d5
43− vsub.f32 d2, d4, d5
44−
45− vst1.32 {d0-d3}, [r0,:128]
46−
47− bx lr
48−endfunc
49−
50−function fft8_neon
51− mov r1, r0
52− vld1.32 {d0-d3}, [r1,:128]!
53− vld1.32 {d16-d19}, [r1,:128]
54−
55− movw r2, #0x04f3 @ sqrt(1/2)
56− movt r2, #0x3f35
57− eor r3, r2, #1<<31
58− vdup.32 d31, r2
59−
60− vext.32 q11, q1, q1, #1 @ i2,r3,i3,r2
61− vadd.f32 d4, d16, d17 @ r4+r5,i4+i5
62− vmov d28, r3, r2
63− vadd.f32 d5, d18, d19 @ r6+r7,i6+i7
64− vsub.f32 d17, d16, d17 @ r4-r5,i4-i5
65− vsub.f32 d19, d18, d19 @ r6-r7,i6-i7
66− vrev64.32 d29, d28
67− vadd.f32 d20, d0, d1 @ r0+r1,i0+i1
68− vadd.f32 d21, d2, d3 @ r2+r3,i2+i3
69− vmul.f32 d26, d17, d28 @ -a2r*w,a2i*w
70− vext.32 q3, q2, q2, #1
71− vmul.f32 d27, d19, d29 @ a3r*w,-a3i*w
72− vsub.f32 d23, d22, d23 @ i2-i3,r3-r2
73− vsub.f32 d22, d0, d1 @ r0-r1,i0-i1
74− vmul.f32 d24, d17, d31 @ a2r*w,a2i*w
75− vmul.f32 d25, d19, d31 @ a3r*w,a3i*w
76− vadd.f32 d0, d20, d21
77− vsub.f32 d2, d20, d21
78− vadd.f32 d1, d22, d23
79− vrev64.32 q13, q13
80− vsub.f32 d3, d22, d23
81− vsub.f32 d6, d6, d7
82− vadd.f32 d24, d24, d26 @ a2r+a2i,a2i-a2r t1,t2
83− vadd.f32 d25, d25, d27 @ a3r-a3i,a3i+a3r t5,t6
84− vadd.f32 d7, d4, d5
85− vsub.f32 d18, d2, d6
86− vext.32 q13, q12, q12, #1
87− vadd.f32 d2, d2, d6
88− vsub.f32 d16, d0, d7
89− vadd.f32 d5, d25, d24
90− vsub.f32 d4, d26, d27
91− vadd.f32 d0, d0, d7
92− vsub.f32 d17, d1, d5
93− vsub.f32 d19, d3, d4
94− vadd.f32 d3, d3, d4
95− vadd.f32 d1, d1, d5
96−
97− vst1.32 {d16-d19}, [r1,:128]
98− vst1.32 {d0-d3}, [r0,:128]
99−
100− bx lr
101−endfunc
102−
103−function fft16_neon
104− movrel r1, mppm
105− vld1.32 {d16-d19}, [r0,:128]! @ q8{r0,i0,r1,i1} q9{r2,i2,r3,i3}
106− pld [r0, #32]
107− vld1.32 {d2-d3}, [r1,:128]
108− vext.32 q13, q9, q9, #1
109− vld1.32 {d22-d25}, [r0,:128]! @ q11{r4,i4,r5,i5} q12{r6,i5,r7,i7}
110− vadd.f32 d4, d16, d17
111− vsub.f32 d5, d16, d17
112− vadd.f32 d18, d18, d19
113− vsub.f32 d19, d26, d27
114−
115− vadd.f32 d20, d22, d23
116− vsub.f32 d22, d22, d23
117− vsub.f32 d23, d24, d25
118− vadd.f32 q8, q2, q9 @ {r0,i0,r1,i1}
119− vadd.f32 d21, d24, d25
120− vmul.f32 d24, d22, d2
121− vsub.f32 q9, q2, q9 @ {r2,i2,r3,i3}
122− vmul.f32 d25, d23, d3
123− vuzp.32 d16, d17 @ {r0,r1,i0,i1}
124− vmul.f32 q1, q11, d2[1]
125− vuzp.32 d18, d19 @ {r2,r3,i2,i3}
126− vrev64.32 q12, q12
127− vadd.f32 q11, q12, q1 @ {t1a,t2a,t5,t6}
128− vld1.32 {d24-d27}, [r0,:128]! @ q12{r8,i8,r9,i9} q13{r10,i10,r11,i11}
129− vzip.32 q10, q11
130− vld1.32 {d28-d31}, [r0,:128] @ q14{r12,i12,r13,i13} q15{r14,i14,r15,i15}
131− vadd.f32 d0, d22, d20
132− vadd.f32 d1, d21, d23
133− vsub.f32 d2, d21, d23
134− vsub.f32 d3, d22, d20
135− sub r0, r0, #96
136− vext.32 q13, q13, q13, #1
137− vsub.f32 q10, q8, q0 @ {r4,r5,i4,i5}
138− vadd.f32 q8, q8, q0 @ {r0,r1,i0,i1}
139− vext.32 q15, q15, q15, #1
140− vsub.f32 q11, q9, q1 @ {r6,r7,i6,i7}
141− vswp d25, d26 @ q12{r8,i8,i10,r11} q13{r9,i9,i11,r10}
142− vadd.f32 q9, q9, q1 @ {r2,r3,i2,i3}
143− vswp d29, d30 @ q14{r12,i12,i14,r15} q15{r13,i13,i15,r14}
144− vadd.f32 q0, q12, q13 @ {t1,t2,t5,t6}
145− vadd.f32 q1, q14, q15 @ {t1a,t2a,t5a,t6a}
146− movrelx r2, X(ff_cos_16)
147− vsub.f32 q13, q12, q13 @ {t3,t4,t7,t8}
148− vrev64.32 d1, d1
149− vsub.f32 q15, q14, q15 @ {t3a,t4a,t7a,t8a}
150− vrev64.32 d3, d3
151− movrel r3, pmmp
152− vswp d1, d26 @ q0{t1,t2,t3,t4} q13{t6,t5,t7,t8}
153− vswp d3, d30 @ q1{t1a,t2a,t3a,t4a} q15{t6a,t5a,t7a,t8a}
154− vadd.f32 q12, q0, q13 @ {r8,i8,r9,i9}
155− vadd.f32 q14, q1, q15 @ {r12,i12,r13,i13}
156− vld1.32 {d4-d5}, [r2,:64]
157− vsub.f32 q13, q0, q13 @ {r10,i10,r11,i11}
158− vsub.f32 q15, q1, q15 @ {r14,i14,r15,i15}
159− vswp d25, d28 @ q12{r8,i8,r12,i12} q14{r9,i9,r13,i13}
160− vld1.32 {d6-d7}, [r3,:128]
161− vrev64.32 q1, q14
162− vmul.f32 q14, q14, d4[1]
163− vmul.f32 q1, q1, q3
164− vmla.f32 q14, q1, d5[1] @ {t1a,t2a,t5a,t6a}
165− vswp d27, d30 @ q13{r10,i10,r14,i14} q15{r11,i11,r15,i15}
166− vzip.32 q12, q14
167− vadd.f32 d0, d28, d24
168− vadd.f32 d1, d25, d29
169− vsub.f32 d2, d25, d29
170− vsub.f32 d3, d28, d24
171− vsub.f32 q12, q8, q0 @ {r8,r9,i8,i9}
172− vadd.f32 q8, q8, q0 @ {r0,r1,i0,i1}
173− vsub.f32 q14, q10, q1 @ {r12,r13,i12,i13}
174− mov r1, #32
175− vadd.f32 q10, q10, q1 @ {r4,r5,i4,i5}
176− vrev64.32 q0, q13
177− vmul.f32 q13, q13, d5[0]
178− vrev64.32 q1, q15
179− vmul.f32 q15, q15, d5[1]
180− vst2.32 {d16-d17},[r0,:128], r1
181− vmul.f32 q0, q0, q3
182− vst2.32 {d20-d21},[r0,:128], r1
183− vmul.f32 q1, q1, q3
184− vmla.f32 q13, q0, d5[0] @ {t1,t2,t5,t6}
185− vmla.f32 q15, q1, d4[1] @ {t1a,t2a,t5a,t6a}
186− vst2.32 {d24-d25},[r0,:128], r1
187− vst2.32 {d28-d29},[r0,:128]
188− vzip.32 q13, q15
189− sub r0, r0, #80
190− vadd.f32 d0, d30, d26
191− vadd.f32 d1, d27, d31
192− vsub.f32 d2, d27, d31
193− vsub.f32 d3, d30, d26
194− vsub.f32 q13, q9, q0 @ {r10,r11,i10,i11}
195− vadd.f32 q9, q9, q0 @ {r2,r3,i2,i3}
196− vsub.f32 q15, q11, q1 @ {r14,r15,i14,i15}
197− vadd.f32 q11, q11, q1 @ {r6,r7,i6,i7}
198− vst2.32 {d18-d19},[r0,:128], r1
199− vst2.32 {d22-d23},[r0,:128], r1
200− vst2.32 {d26-d27},[r0,:128], r1
201− vst2.32 {d30-d31},[r0,:128]
202− bx lr
203−endfunc
204−
205−function fft_pass_neon
206− push {r4-r6,lr}
207− mov r6, r2 @ n
208− lsl r5, r2, #3 @ 2 * n * sizeof FFTSample
209− lsl r4, r2, #4 @ 2 * n * sizeof FFTComplex
210− lsl r2, r2, #5 @ 4 * n * sizeof FFTComplex
211− add r3, r2, r4
212− add r4, r4, r0 @ &z[o1]
213− add r2, r2, r0 @ &z[o2]
214− add r3, r3, r0 @ &z[o3]
215− vld1.32 {d20-d21},[r2,:128] @ {z[o2],z[o2+1]}
216− movrel r12, pmmp
217− vld1.32 {d22-d23},[r3,:128] @ {z[o3],z[o3+1]}
218− add r5, r5, r1 @ wim
219− vld1.32 {d6-d7}, [r12,:128] @ pmmp
220− vswp d21, d22
221− vld1.32 {d4}, [r1,:64]! @ {wre[0],wre[1]}
222− sub r5, r5, #4 @ wim--
223− vrev64.32 q1, q11
224− vmul.f32 q11, q11, d4[1]
225− vmul.f32 q1, q1, q3
226− vld1.32 {d5[0]}, [r5,:32] @ d5[0] = wim[-1]
227− vmla.f32 q11, q1, d5[0] @ {t1a,t2a,t5a,t6a}
228− vld2.32 {d16-d17},[r0,:128] @ {z[0],z[1]}
229− sub r6, r6, #1 @ n--
230− vld2.32 {d18-d19},[r4,:128] @ {z[o1],z[o1+1]}
231− vzip.32 q10, q11
232− vadd.f32 d0, d22, d20
233− vadd.f32 d1, d21, d23
234− vsub.f32 d2, d21, d23
235− vsub.f32 d3, d22, d20
236− vsub.f32 q10, q8, q0
237− vadd.f32 q8, q8, q0
238− vsub.f32 q11, q9, q1
239− vadd.f32 q9, q9, q1
240− vst2.32 {d20-d21},[r2,:128]! @ {z[o2],z[o2+1]}
241− vst2.32 {d16-d17},[r0,:128]! @ {z[0],z[1]}
242− vst2.32 {d22-d23},[r3,:128]! @ {z[o3],z[o3+1]}
243− vst2.32 {d18-d19},[r4,:128]! @ {z[o1],z[o1+1]}
244− sub r5, r5, #8 @ wim -= 2
245−1:
246− vld1.32 {d20-d21},[r2,:128] @ {z[o2],z[o2+1]}
247− vld1.32 {d22-d23},[r3,:128] @ {z[o3],z[o3+1]}
248− vswp d21, d22
249− vld1.32 {d4}, [r1]! @ {wre[0],wre[1]}
250− vrev64.32 q0, q10
251− vmul.f32 q10, q10, d4[0]
252− vrev64.32 q1, q11
253− vmul.f32 q11, q11, d4[1]
254− vld1.32 {d5}, [r5] @ {wim[-1],wim[0]}
255− vmul.f32 q0, q0, q3
256− sub r5, r5, #8 @ wim -= 2
257− vmul.f32 q1, q1, q3
258− vmla.f32 q10, q0, d5[1] @ {t1,t2,t5,t6}
259− vmla.f32 q11, q1, d5[0] @ {t1a,t2a,t5a,t6a}
260− vld2.32 {d16-d17},[r0,:128] @ {z[0],z[1]}
261− subs r6, r6, #1 @ n--
262− vld2.32 {d18-d19},[r4,:128] @ {z[o1],z[o1+1]}
263− vzip.32 q10, q11
264− vadd.f32 d0, d22, d20
265− vadd.f32 d1, d21, d23
266− vsub.f32 d2, d21, d23
267− vsub.f32 d3, d22, d20
268− vsub.f32 q10, q8, q0
269− vadd.f32 q8, q8, q0
270− vsub.f32 q11, q9, q1
271− vadd.f32 q9, q9, q1
272− vst2.32 {d20-d21}, [r2,:128]! @ {z[o2],z[o2+1]}
273− vst2.32 {d16-d17}, [r0,:128]! @ {z[0],z[1]}
274− vst2.32 {d22-d23}, [r3,:128]! @ {z[o3],z[o3+1]}
275− vst2.32 {d18-d19}, [r4,:128]! @ {z[o1],z[o1+1]}
276− bne 1b
277−
278− pop {r4-r6,pc}
279−endfunc
280−
281−.macro def_fft n, n2, n4
282− .align 6
283−function fft\n\()_neon
284− push {r4, lr}
285− mov r4, r0
286− bl fft\n2\()_neon
287− add r0, r4, #\n4*2*8
288− bl fft\n4\()_neon
289− add r0, r4, #\n4*3*8
290− bl fft\n4\()_neon
291− mov r0, r4
292− pop {r4, lr}
293− movrelx r1, X(ff_cos_\n)
294− mov r2, #\n4/2
295− b fft_pass_neon
296−endfunc
297−.endm
298−
299− def_fft 32, 16, 8
300− def_fft 64, 32, 16
301− def_fft 128, 64, 32
302− def_fft 256, 128, 64
303− def_fft 512, 256, 128
304− def_fft 1024, 512, 256
305− def_fft 2048, 1024, 512
306− def_fft 4096, 2048, 1024
307− def_fft 8192, 4096, 2048
308− def_fft 16384, 8192, 4096
309− def_fft 32768, 16384, 8192
310− def_fft 65536, 32768, 16384
311−
312−function ff_fft_calc_neon, export=1
313− ldr r2, [r0]
314− sub r2, r2, #2
315− movrel r3, fft_tab_neon
316− ldr r3, [r3, r2, lsl #2]
317− mov r0, r1
318− bx r3
319−endfunc
320−
321−function ff_fft_permute_neon, export=1
322− push {r4,lr}
323− mov r12, #1
324− ldr r2, [r0] @ nbits
325− ldr r3, [r0, #12] @ tmp_buf
326− ldr r0, [r0, #8] @ revtab
327− lsl r12, r12, r2
328− mov r2, r12
329−1:
330− vld1.32 {d0-d1}, [r1,:128]!
331− ldr r4, [r0], #4
332− uxth lr, r4
333− uxth r4, r4, ror #16
334− add lr, r3, lr, lsl #3
335− add r4, r3, r4, lsl #3
336− vst1.32 {d0}, [lr,:64]
337− vst1.32 {d1}, [r4,:64]
338− subs r12, r12, #2
339− bgt 1b
340−
341− sub r1, r1, r2, lsl #3
342−1:
343− vld1.32 {d0-d3}, [r3,:128]!
344− vst1.32 {d0-d3}, [r1,:128]!
345− subs r2, r2, #4
346− bgt 1b
347−
348− pop {r4,pc}
349−endfunc
350−
351−const fft_tab_neon, relocate=1
352− .word fft4_neon
353− .word fft8_neon
354− .word fft16_neon
355− .word fft32_neon
356− .word fft64_neon
357− .word fft128_neon
358− .word fft256_neon
359− .word fft512_neon
360− .word fft1024_neon
361− .word fft2048_neon
362− .word fft4096_neon
363− .word fft8192_neon
364− .word fft16384_neon
365− .word fft32768_neon
366− .word fft65536_neon
367−endconst
368−
369−const pmmp, align=4
370− .float +1.0, -1.0, -1.0, +1.0
371−endconst
372−
373−const mppm, align=4
374− .float -M_SQRT1_2, M_SQRT1_2, M_SQRT1_2, -M_SQRT1_2
375−endconst
libavcodec/arm/fft_vfp.S+0−530
@@ -1,530 +0,0 @@
1−/*
2− * Copyright (c) 2013 RISC OS Open Ltd
3− * Author: Ben Avison <bavison@riscosopen.org>
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/arm/asm.S"
23−
24−@ The fftx_internal_vfp versions of the functions obey a modified AAPCS:
25−@ VFP is in RunFast mode, vector length 4, stride 1 thoroughout, and
26−@ all single-precision VFP registers may be corrupted on exit. The a2
27−@ register may not be clobbered in these functions, as it holds the
28−@ stored original FPSCR.
29−
30−function ff_fft_calc_vfp, export=1
31− ldr ip, [a1, #0] @ nbits
32− mov a1, a2
33− movrel a2, (fft_tab_vfp - 8)
34− ldr pc, [a2, ip, lsl #2]
35−endfunc
36−const fft_tab_vfp, relocate=1
37− .word fft4_vfp
38− .word fft8_vfp
39− .word X(ff_fft16_vfp) @ this one alone is exported
40− .word fft32_vfp
41− .word fft64_vfp
42− .word fft128_vfp
43− .word fft256_vfp
44− .word fft512_vfp
45− .word fft1024_vfp
46− .word fft2048_vfp
47− .word fft4096_vfp
48− .word fft8192_vfp
49− .word fft16384_vfp
50− .word fft32768_vfp
51− .word fft65536_vfp
52−endconst
53−
54−function fft4_vfp
55− vldr d0, [a1, #0*2*4] @ s0,s1 = z[0]
56− vldr d4, [a1, #1*2*4] @ s8,s9 = z[1]
57− vldr d1, [a1, #2*2*4] @ s2,s3 = z[2]
58− vldr d5, [a1, #3*2*4] @ s10,s11 = z[3]
59− @ stall
60− vadd.f s12, s0, s8 @ i0
61− vadd.f s13, s1, s9 @ i1
62− vadd.f s14, s2, s10 @ i2
63− vadd.f s15, s3, s11 @ i3
64− vsub.f s8, s0, s8 @ i4
65− vsub.f s9, s1, s9 @ i5
66− vsub.f s10, s2, s10 @ i6
67− vsub.f s11, s3, s11 @ i7
68− @ stall
69− @ stall
70− vadd.f s0, s12, s14 @ z[0].re
71− vsub.f s4, s12, s14 @ z[2].re
72− vadd.f s1, s13, s15 @ z[0].im
73− vsub.f s5, s13, s15 @ z[2].im
74− vadd.f s7, s9, s10 @ z[3].im
75− vsub.f s3, s9, s10 @ z[1].im
76− vadd.f s2, s8, s11 @ z[1].re
77− vsub.f s6, s8, s11 @ z[3].re
78− @ stall
79− @ stall
80− vstr d0, [a1, #0*2*4]
81− vstr d2, [a1, #2*2*4]
82− @ stall
83− @ stall
84− vstr d1, [a1, #1*2*4]
85− vstr d3, [a1, #3*2*4]
86−
87− bx lr
88−endfunc
89−
90−.macro macro_fft8_head
91− @ FFT4
92− vldr d4, [a1, #0 * 2*4]
93− vldr d6, [a1, #1 * 2*4]
94− vldr d5, [a1, #2 * 2*4]
95− vldr d7, [a1, #3 * 2*4]
96− @ BF
97− vldr d12, [a1, #4 * 2*4]
98− vadd.f s16, s8, s12 @ vector op
99− vldr d14, [a1, #5 * 2*4]
100− vldr d13, [a1, #6 * 2*4]
101− vldr d15, [a1, #7 * 2*4]
102− vsub.f s20, s8, s12 @ vector op
103− vadd.f s0, s16, s18
104− vsub.f s2, s16, s18
105− vadd.f s1, s17, s19
106− vsub.f s3, s17, s19
107− vadd.f s7, s21, s22
108− vsub.f s5, s21, s22
109− vadd.f s4, s20, s23
110− vsub.f s6, s20, s23
111− vsub.f s20, s24, s28 @ vector op
112− vstr d0, [a1, #0 * 2*4] @ transfer s0-s7 to s24-s31 via memory
113− vstr d1, [a1, #1 * 2*4]
114− vldr s0, cos1pi4
115− vadd.f s16, s24, s28 @ vector op
116− vstr d2, [a1, #2 * 2*4]
117− vstr d3, [a1, #3 * 2*4]
118− vldr d12, [a1, #0 * 2*4]
119− @ TRANSFORM
120− vmul.f s20, s20, s0 @ vector x scalar op
121− vldr d13, [a1, #1 * 2*4]
122− vldr d14, [a1, #2 * 2*4]
123− vldr d15, [a1, #3 * 2*4]
124− @ BUTTERFLIES
125− vadd.f s0, s18, s16
126− vadd.f s1, s17, s19
127− vsub.f s2, s17, s19
128− vsub.f s3, s18, s16
129− vadd.f s4, s21, s20
130− vsub.f s5, s21, s20
131− vadd.f s6, s22, s23
132− vsub.f s7, s22, s23
133− vadd.f s8, s0, s24 @ vector op
134− vstr d0, [a1, #0 * 2*4] @ transfer s0-s3 to s12-s15 via memory
135− vstr d1, [a1, #1 * 2*4]
136− vldr d6, [a1, #0 * 2*4]
137− vldr d7, [a1, #1 * 2*4]
138− vadd.f s1, s5, s6
139− vadd.f s0, s7, s4
140− vsub.f s2, s5, s6
141− vsub.f s3, s7, s4
142− vsub.f s12, s24, s12 @ vector op
143− vsub.f s5, s29, s1
144− vsub.f s4, s28, s0
145− vsub.f s6, s30, s2
146− vsub.f s7, s31, s3
147− vadd.f s16, s0, s28 @ vector op
148− vstr d6, [a1, #4 * 2*4]
149− vstr d7, [a1, #6 * 2*4]
150− vstr d4, [a1, #0 * 2*4]
151− vstr d5, [a1, #2 * 2*4]
152− vstr d2, [a1, #5 * 2*4]
153− vstr d3, [a1, #7 * 2*4]
154−.endm
155−
156−.macro macro_fft8_tail
157− vstr d8, [a1, #1 * 2*4]
158− vstr d9, [a1, #3 * 2*4]
159−.endm
160−
161−function .Lfft8_internal_vfp
162− macro_fft8_head
163− macro_fft8_tail
164− bx lr
165−endfunc
166−
167−function fft8_vfp
168− ldr a3, =0x03030000 @ RunFast mode, vector length 4, stride 1
169− fmrx a2, FPSCR
170− fmxr FPSCR, a3
171− vpush {s16-s31}
172− mov ip, lr
173− bl .Lfft8_internal_vfp
174− vpop {s16-s31}
175− fmxr FPSCR, a2
176− bx ip
177−endfunc
178−
179−.align 3
180−cos1pi4: @ cos(1*pi/4) = sqrt(2)
181− .float 0.707106769084930419921875
182−cos1pi8: @ cos(1*pi/8) = sqrt(2+sqrt(2))/2
183− .float 0.92387950420379638671875
184−cos3pi8: @ cos(2*pi/8) = sqrt(2-sqrt(2))/2
185− .float 0.3826834261417388916015625
186−
187−function .Lfft16_internal_vfp
188− macro_fft8_head
189− @ FFT4(z+8)
190− vldr d10, [a1, #8 * 2*4]
191− vldr d12, [a1, #9 * 2*4]
192− vldr d11, [a1, #10 * 2*4]
193− vldr d13, [a1, #11 * 2*4]
194− macro_fft8_tail
195− vadd.f s16, s20, s24 @ vector op
196− @ FFT4(z+12)
197− vldr d4, [a1, #12 * 2*4]
198− vldr d6, [a1, #13 * 2*4]
199− vldr d5, [a1, #14 * 2*4]
200− vsub.f s20, s20, s24 @ vector op
201− vldr d7, [a1, #15 * 2*4]
202− vadd.f s0, s16, s18
203− vsub.f s4, s16, s18
204− vadd.f s1, s17, s19
205− vsub.f s5, s17, s19
206− vadd.f s7, s21, s22
207− vsub.f s3, s21, s22
208− vadd.f s2, s20, s23
209− vsub.f s6, s20, s23
210− vadd.f s16, s8, s12 @ vector op
211− vstr d0, [a1, #8 * 2*4]
212− vstr d2, [a1, #10 * 2*4]
213− vstr d1, [a1, #9 * 2*4]
214− vsub.f s20, s8, s12
215− vstr d3, [a1, #11 * 2*4]
216− @ TRANSFORM(z[2],z[6],z[10],z[14],cos1pi4,cos1pi4)
217− vldr d12, [a1, #10 * 2*4]
218− vadd.f s0, s16, s18
219− vadd.f s1, s17, s19
220− vsub.f s6, s16, s18
221− vsub.f s7, s17, s19
222− vsub.f s3, s21, s22
223− vadd.f s2, s20, s23
224− vadd.f s5, s21, s22
225− vsub.f s4, s20, s23
226− vstr d0, [a1, #12 * 2*4]
227− vmov s0, s6
228− @ TRANSFORM(z[1],z[5],z[9],z[13],cos1pi8,cos3pi8)
229− vldr d6, [a1, #9 * 2*4]
230− vstr d1, [a1, #13 * 2*4]
231− vldr d1, cos1pi4 @ s2 = cos1pi4, s3 = cos1pi8
232− vstr d2, [a1, #15 * 2*4]
233− vldr d7, [a1, #13 * 2*4]
234− vadd.f s4, s25, s24
235− vsub.f s5, s25, s24
236− vsub.f s6, s0, s7
237− vadd.f s7, s0, s7
238− vmul.f s20, s12, s3 @ vector op
239− @ TRANSFORM(z[3],z[7],z[11],z[15],cos3pi8,cos1pi8)
240− vldr d4, [a1, #11 * 2*4]
241− vldr d5, [a1, #15 * 2*4]
242− vldr s1, cos3pi8
243− vmul.f s24, s4, s2 @ vector * scalar op
244− vmul.f s28, s12, s1 @ vector * scalar op
245− vmul.f s12, s8, s1 @ vector * scalar op
246− vadd.f s4, s20, s29
247− vsub.f s5, s21, s28
248− vsub.f s6, s22, s31
249− vadd.f s7, s23, s30
250− vmul.f s8, s8, s3 @ vector * scalar op
251− vldr d8, [a1, #1 * 2*4]
252− vldr d9, [a1, #5 * 2*4]
253− vldr d10, [a1, #3 * 2*4]
254− vldr d11, [a1, #7 * 2*4]
255− vldr d14, [a1, #2 * 2*4]
256− vadd.f s0, s6, s4
257− vadd.f s1, s5, s7
258− vsub.f s2, s5, s7
259− vsub.f s3, s6, s4
260− vadd.f s4, s12, s9
261− vsub.f s5, s13, s8
262− vsub.f s6, s14, s11
263− vadd.f s7, s15, s10
264− vadd.f s12, s0, s16 @ vector op
265− vstr d0, [a1, #1 * 2*4]
266− vstr d1, [a1, #5 * 2*4]
267− vldr d4, [a1, #1 * 2*4]
268− vldr d5, [a1, #5 * 2*4]
269− vadd.f s0, s6, s4
270− vadd.f s1, s5, s7
271− vsub.f s2, s5, s7
272− vsub.f s3, s6, s4
273− vsub.f s8, s16, s8 @ vector op
274− vstr d6, [a1, #1 * 2*4]
275− vstr d7, [a1, #5 * 2*4]
276− vldr d15, [a1, #6 * 2*4]
277− vsub.f s4, s20, s0
278− vsub.f s5, s21, s1
279− vsub.f s6, s22, s2
280− vsub.f s7, s23, s3
281− vadd.f s20, s0, s20 @ vector op
282− vstr d4, [a1, #9 * 2*4]
283− @ TRANSFORM_ZERO(z[0],z[4],z[8],z[12])
284− vldr d6, [a1, #8 * 2*4]
285− vstr d5, [a1, #13 * 2*4]
286− vldr d7, [a1, #12 * 2*4]
287− vstr d2, [a1, #11 * 2*4]
288− vldr d8, [a1, #0 * 2*4]
289− vstr d3, [a1, #15 * 2*4]
290− vldr d9, [a1, #4 * 2*4]
291− vadd.f s0, s26, s24
292− vadd.f s1, s25, s27
293− vsub.f s2, s25, s27
294− vsub.f s3, s26, s24
295− vadd.f s4, s14, s12
296− vadd.f s5, s13, s15
297− vsub.f s6, s13, s15
298− vsub.f s7, s14, s12
299− vadd.f s8, s0, s28 @ vector op
300− vstr d0, [a1, #3 * 2*4]
301− vstr d1, [a1, #7 * 2*4]
302− vldr d6, [a1, #3 * 2*4]
303− vldr d7, [a1, #7 * 2*4]
304− vsub.f s0, s16, s4
305− vsub.f s1, s17, s5
306− vsub.f s2, s18, s6
307− vsub.f s3, s19, s7
308− vsub.f s12, s28, s12 @ vector op
309− vadd.f s16, s4, s16 @ vector op
310− vstr d10, [a1, #3 * 2*4]
311− vstr d11, [a1, #7 * 2*4]
312− vstr d4, [a1, #2 * 2*4]
313− vstr d5, [a1, #6 * 2*4]
314− vstr d0, [a1, #8 * 2*4]
315− vstr d1, [a1, #12 * 2*4]
316− vstr d6, [a1, #10 * 2*4]
317− vstr d7, [a1, #14 * 2*4]
318− vstr d8, [a1, #0 * 2*4]
319− vstr d9, [a1, #4 * 2*4]
320−
321− bx lr
322−endfunc
323−
324−function ff_fft16_vfp, export=1
325− ldr a3, =0x03030000 @ RunFast mode, vector length 4, stride 1
326− fmrx a2, FPSCR
327− fmxr FPSCR, a3
328− vpush {s16-s31}
329− mov ip, lr
330− bl .Lfft16_internal_vfp
331− vpop {s16-s31}
332− fmxr FPSCR, a2
333− bx ip
334−endfunc
335−
336−.macro pass n, z0, z1, z2, z3
337− add v6, v5, #4*2*\n
338− @ TRANSFORM_ZERO(z[0],z[o1],z[o2],z[o3])
339− @ TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1])
340− @ TRANSFORM(z[0],z[o1],z[o2],z[o3],wre[0],wim[0])
341− @ TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1])
342− vldr d8, [\z2, #8*(o2+1)] @ s16,s17
343− vldmdb v6!, {s2}
344− vldr d9, [\z3, #8*(o3+1)] @ s18,s19
345− vldmia v5!, {s0,s1} @ s0 is unused
346− vldr s7, [\z2, #8*o2] @ t1
347− vmul.f s20, s16, s2 @ vector * scalar
348− vldr s0, [\z3, #8*o3] @ t5
349− vldr s6, [\z2, #8*o2+4] @ t2
350− vldr s3, [\z3, #8*o3+4] @ t6
351− vmul.f s16, s16, s1 @ vector * scalar
352− ldr a4, =\n-1
353−1: add \z0, \z0, #8*2
354− .if \n*4*2 >= 512
355− add \z1, \z1, #8*2
356− .endif
357− .if \n*4*2 >= 256
358− add \z2, \z2, #8*2
359− .endif
360− .if \n*4*2 >= 512
361− add \z3, \z3, #8*2
362− .endif
363− @ up to 2 stalls (VFP vector issuing / waiting for s0)
364− @ depending upon whether this is the first iteration and
365− @ how many add instructions are inserted above
366− vadd.f s4, s0, s7 @ t5
367− vadd.f s5, s6, s3 @ t6
368− vsub.f s6, s6, s3 @ t4
369− vsub.f s7, s0, s7 @ t3
370− vldr d6, [\z0, #8*0-8*2] @ s12,s13
371− vadd.f s0, s16, s21 @ t1
372− vldr d7, [\z1, #8*o1-8*2] @ s14,s15
373− vsub.f s1, s18, s23 @ t5
374− vadd.f s8, s4, s12 @ vector + vector
375− @ stall (VFP vector issuing)
376− @ stall (VFP vector issuing)
377− @ stall (VFP vector issuing)
378− vsub.f s4, s12, s4
379− vsub.f s5, s13, s5
380− vsub.f s6, s14, s6
381− vsub.f s7, s15, s7
382− vsub.f s2, s17, s20 @ t2
383− vadd.f s3, s19, s22 @ t6
384− vstr d4, [\z0, #8*0-8*2] @ s8,s9
385− vstr d5, [\z1, #8*o1-8*2] @ s10,s11
386− @ stall (waiting for s5)
387− vstr d2, [\z2, #8*o2-8*2] @ s4,s5
388− vadd.f s4, s1, s0 @ t5
389− vstr d3, [\z3, #8*o3-8*2] @ s6,s7
390− vsub.f s7, s1, s0 @ t3
391− vadd.f s5, s2, s3 @ t6
392− vsub.f s6, s2, s3 @ t4
393− vldr d6, [\z0, #8*1-8*2] @ s12,s13
394− vldr d7, [\z1, #8*(o1+1)-8*2] @ s14,s15
395− vldr d4, [\z2, #8*o2] @ s8,s9
396− vldmdb v6!, {s2,s3}
397− vldr d5, [\z3, #8*o3] @ s10,s11
398− vadd.f s20, s4, s12 @ vector + vector
399− vldmia v5!, {s0,s1}
400− vldr d8, [\z2, #8*(o2+1)] @ s16,s17
401− @ stall (VFP vector issuing)
402− vsub.f s4, s12, s4
403− vsub.f s5, s13, s5
404− vsub.f s6, s14, s6
405− vsub.f s7, s15, s7
406− vmul.f s12, s8, s3 @ vector * scalar
407− vstr d10, [\z0, #8*1-8*2] @ s20,s21
408− vldr d9, [\z3, #8*(o3+1)] @ s18,s19
409− vstr d11, [\z1, #8*(o1+1)-8*2] @ s22,s23
410− vmul.f s8, s8, s0 @ vector * scalar
411− vstr d2, [\z2, #8*(o2+1)-8*2] @ s4,s5
412− @ stall (waiting for s7)
413− vstr d3, [\z3, #8*(o3+1)-8*2] @ s6,s7
414− vmul.f s20, s16, s2 @ vector * scalar
415− @ stall (VFP vector issuing)
416− @ stall (VFP vector issuing)
417− @ stall (VFP vector issuing)
418− vadd.f s7, s8, s13 @ t1
419− vsub.f s6, s9, s12 @ t2
420− vsub.f s0, s10, s15 @ t5
421− vadd.f s3, s11, s14 @ t6
422− vmul.f s16, s16, s1 @ vector * scalar
423− subs a4, a4, #1
424− bne 1b
425− @ What remains is identical to the first two indentations of
426− @ the above, but without the increment of z
427− vadd.f s4, s0, s7 @ t5
428− vadd.f s5, s6, s3 @ t6
429− vsub.f s6, s6, s3 @ t4
430− vsub.f s7, s0, s7 @ t3
431− vldr d6, [\z0, #8*0] @ s12,s13
432− vadd.f s0, s16, s21 @ t1
433− vldr d7, [\z1, #8*o1] @ s14,s15
434− vsub.f s1, s18, s23 @ t5
435− vadd.f s8, s4, s12 @ vector + vector
436− vsub.f s4, s12, s4
437− vsub.f s5, s13, s5
438− vsub.f s6, s14, s6
439− vsub.f s7, s15, s7
440− vsub.f s2, s17, s20 @ t2
441− vadd.f s3, s19, s22 @ t6
442− vstr d4, [\z0, #8*0] @ s8,s9
443− vstr d5, [\z1, #8*o1] @ s10,s11
444− vstr d2, [\z2, #8*o2] @ s4,s5
445− vadd.f s4, s1, s0 @ t5
446− vstr d3, [\z3, #8*o3] @ s6,s7
447− vsub.f s7, s1, s0 @ t3
448− vadd.f s5, s2, s3 @ t6
449− vsub.f s6, s2, s3 @ t4
450− vldr d6, [\z0, #8*1] @ s12,s13
451− vldr d7, [\z1, #8*(o1+1)] @ s14,s15
452− vadd.f s20, s4, s12 @ vector + vector
453− vsub.f s4, s12, s4
454− vsub.f s5, s13, s5
455− vsub.f s6, s14, s6
456− vsub.f s7, s15, s7
457− vstr d10, [\z0, #8*1] @ s20,s21
458− vstr d11, [\z1, #8*(o1+1)] @ s22,s23
459− vstr d2, [\z2, #8*(o2+1)] @ s4,s5
460− vstr d3, [\z3, #8*(o3+1)] @ s6,s7
461−.endm
462−
463−.macro def_fft n, n2, n4
464−function .Lfft\n\()_internal_vfp
465− .if \n >= 512
466− push {v1-v6,lr}
467− .elseif \n >= 256
468− push {v1-v2,v5-v6,lr}
469− .else
470− push {v1,v5-v6,lr}
471− .endif
472− mov v1, a1
473− bl .Lfft\n2\()_internal_vfp
474− add a1, v1, #8*(\n/4)*2
475− bl .Lfft\n4\()_internal_vfp
476− movrelx v5, X(ff_cos_\n), a1
477− add a1, v1, #8*(\n/4)*3
478− bl .Lfft\n4\()_internal_vfp
479− .if \n >= 512
480− .set o1, 0*(\n/4/2)
481− .set o2, 0*(\n/4/2)
482− .set o3, 0*(\n/4/2)
483− add v2, v1, #8*2*(\n/4/2)
484− add v3, v1, #8*4*(\n/4/2)
485− add v4, v1, #8*6*(\n/4/2)
486− pass (\n/4/2), v1, v2, v3, v4
487− pop {v1-v6,pc}
488− .elseif \n >= 256
489− .set o1, 2*(\n/4/2)
490− .set o2, 0*(\n/4/2)
491− .set o3, 2*(\n/4/2)
492− add v2, v1, #8*4*(\n/4/2)
493− pass (\n/4/2), v1, v1, v2, v2
494− pop {v1-v2,v5-v6,pc}
495− .else
496− .set o1, 2*(\n/4/2)
497− .set o2, 4*(\n/4/2)
498− .set o3, 6*(\n/4/2)
499− pass (\n/4/2), v1, v1, v1, v1
500− pop {v1,v5-v6,pc}
501− .endif
502−endfunc
503−
504−function fft\n\()_vfp
505− ldr a3, =0x03030000 /* RunFast mode, vector length 4, stride 1 */
506− fmrx a2, FPSCR
507− fmxr FPSCR, a3
508− vpush {s16-s31}
509− mov ip, lr
510− bl .Lfft\n\()_internal_vfp
511− vpop {s16-s31}
512− fmxr FPSCR, a2
513− bx ip
514−endfunc
515−
516−.ltorg
517−.endm
518−
519− def_fft 32, 16, 8
520− def_fft 64, 32, 16
521− def_fft 128, 64, 32
522− def_fft 256, 128, 64
523− def_fft 512, 256, 128
524− def_fft 1024, 512, 256
525− def_fft 2048, 1024, 512
526− def_fft 4096, 2048, 1024
527− def_fft 8192, 4096, 2048
528− def_fft 16384, 8192, 4096
529− def_fft 32768, 16384, 8192
530− def_fft 65536, 32768, 16384
libavcodec/arm/mdct_neon.S+0−301
@@ -1,301 +0,0 @@
1−/*
2− * ARM NEON optimised MDCT
3− * Copyright (c) 2009 Mans Rullgard <mans@mansr.com>
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/arm/asm.S"
23−
24−#define ff_fft_calc_neon X(ff_fft_calc_neon)
25−
26−function ff_imdct_half_neon, export=1
27− push {r4-r8,lr}
28−
29− mov r12, #1
30− ldr lr, [r0, #20] @ mdct_bits
31− ldr r4, [r0, #24] @ tcos
32− ldr r3, [r0, #8] @ revtab
33− lsl r12, r12, lr @ n = 1 << nbits
34− lsr lr, r12, #2 @ n4 = n >> 2
35− add r7, r2, r12, lsl #1
36− mov r12, #-16
37− sub r7, r7, #16
38−
39− vld2.32 {d16-d17},[r7,:128],r12 @ d16=x,n1 d17=x,n0
40− vld2.32 {d0-d1}, [r2,:128]! @ d0 =m0,x d1 =m1,x
41− vrev64.32 d17, d17
42− vld2.32 {d2,d3}, [r4,:128]! @ d2=c0,c1 d3=s0,s2
43− vmul.f32 d6, d17, d2
44− vmul.f32 d7, d0, d2
45−1:
46− subs lr, lr, #2
47− ldr r6, [r3], #4
48− vmul.f32 d4, d0, d3
49− vmul.f32 d5, d17, d3
50− vsub.f32 d4, d6, d4
51− vadd.f32 d5, d5, d7
52− uxth r8, r6, ror #16
53− uxth r6, r6
54− add r8, r1, r8, lsl #3
55− add r6, r1, r6, lsl #3
56− beq 1f
57− vld2.32 {d16-d17},[r7,:128],r12
58− vld2.32 {d0-d1}, [r2,:128]!
59− vrev64.32 d17, d17
60− vld2.32 {d2,d3}, [r4,:128]! @ d2=c0,c1 d3=s0,s2
61− vmul.f32 d6, d17, d2
62− vmul.f32 d7, d0, d2
63− vst2.32 {d4[0],d5[0]}, [r6,:64]
64− vst2.32 {d4[1],d5[1]}, [r8,:64]
65− b 1b
66−1:
67− vst2.32 {d4[0],d5[0]}, [r6,:64]
68− vst2.32 {d4[1],d5[1]}, [r8,:64]
69−
70− mov r4, r0
71− mov r6, r1
72− bl ff_fft_calc_neon
73−
74− mov r12, #1
75− ldr lr, [r4, #20] @ mdct_bits
76− ldr r4, [r4, #24] @ tcos
77− lsl r12, r12, lr @ n = 1 << nbits
78− lsr lr, r12, #3 @ n8 = n >> 3
79−
80− add r4, r4, lr, lsl #3
81− add r6, r6, lr, lsl #3
82− sub r1, r4, #16
83− sub r3, r6, #16
84−
85− mov r7, #-16
86− mov r8, r6
87− mov r0, r3
88−
89− vld2.32 {d0-d1}, [r3,:128], r7 @ d0 =i1,r1 d1 =i0,r0
90− vld2.32 {d20-d21},[r6,:128]! @ d20=i2,r2 d21=i3,r3
91− vld2.32 {d16,d18},[r1,:128], r7 @ d16=c1,c0 d18=s1,s0
92−1:
93− subs lr, lr, #2
94− vmul.f32 d7, d0, d18
95− vld2.32 {d17,d19},[r4,:128]! @ d17=c2,c3 d19=s2,s3
96− vmul.f32 d4, d1, d18
97− vmul.f32 d5, d21, d19
98− vmul.f32 d6, d20, d19
99− vmul.f32 d22, d1, d16
100− vmul.f32 d23, d21, d17
101− vmul.f32 d24, d0, d16
102− vmul.f32 d25, d20, d17
103− vadd.f32 d7, d7, d22
104− vadd.f32 d6, d6, d23
105− vsub.f32 d4, d4, d24
106− vsub.f32 d5, d5, d25
107− beq 1f
108− vld2.32 {d0-d1}, [r3,:128], r7
109− vld2.32 {d20-d21},[r6,:128]!
110− vld2.32 {d16,d18},[r1,:128], r7 @ d16=c1,c0 d18=s1,s0
111− vrev64.32 q3, q3
112− vst2.32 {d4,d6}, [r0,:128], r7
113− vst2.32 {d5,d7}, [r8,:128]!
114− b 1b
115−1:
116− vrev64.32 q3, q3
117− vst2.32 {d4,d6}, [r0,:128]
118− vst2.32 {d5,d7}, [r8,:128]
119−
120− pop {r4-r8,pc}
121−endfunc
122−
123−function ff_imdct_calc_neon, export=1
124− push {r4-r6,lr}
125−
126− ldr r3, [r0, #20]
127− mov r4, #1
128− mov r5, r1
129− lsl r4, r4, r3
130− add r1, r1, r4
131−
132− bl X(ff_imdct_half_neon)
133−
134− add r0, r5, r4, lsl #2
135− add r1, r5, r4, lsl #1
136− sub r0, r0, #8
137− sub r2, r1, #16
138− mov r3, #-16
139− mov r6, #-8
140− vmov.i32 d30, #1<<31
141−1:
142− vld1.32 {d0-d1}, [r2,:128], r3
143− pld [r0, #-16]
144− vrev64.32 q0, q0
145− vld1.32 {d2-d3}, [r1,:128]!
146− veor d4, d1, d30
147− pld [r2, #-16]
148− vrev64.32 q1, q1
149− veor d5, d0, d30
150− vst1.32 {d2}, [r0,:64], r6
151− vst1.32 {d3}, [r0,:64], r6
152− vst1.32 {d4-d5}, [r5,:128]!
153− subs r4, r4, #16
154− bgt 1b
155−
156− pop {r4-r6,pc}
157−endfunc
158−
159−function ff_mdct_calc_neon, export=1
160− push {r4-r10,lr}
161−
162− mov r12, #1
163− ldr lr, [r0, #20] @ mdct_bits
164− ldr r4, [r0, #24] @ tcos
165− ldr r3, [r0, #8] @ revtab
166− lsl lr, r12, lr @ n = 1 << nbits
167− add r7, r2, lr @ in4u
168− sub r9, r7, #16 @ in4d
169− add r2, r7, lr, lsl #1 @ in3u
170− add r8, r9, lr, lsl #1 @ in3d
171− add r5, r4, lr, lsl #1
172− sub r5, r5, #16
173− sub r3, r3, #4
174− mov r12, #-16
175−
176− vld2.32 {d16,d18},[r9,:128],r12 @ in0u0,in0u1 in4d1,in4d0
177− vld2.32 {d17,d19},[r8,:128],r12 @ in2u0,in2u1 in3d1,in3d0
178− vld2.32 {d0, d2}, [r7,:128]! @ in4u0,in4u1 in2d1,in2d0
179− vrev64.32 q9, q9 @ in4d0,in4d1 in3d0,in3d1
180− vld2.32 {d1, d3}, [r2,:128]! @ in3u0,in3u1 in1d1,in1d0
181− vsub.f32 d0, d18, d0 @ in4d-in4u I
182− vld2.32 {d20,d21},[r4,:128]! @ c0,c1 s0,s1
183− vrev64.32 q1, q1 @ in2d0,in2d1 in1d0,in1d1
184− vld2.32 {d30,d31},[r5,:128],r12 @ c2,c3 s2,s3
185− vadd.f32 d1, d1, d19 @ in3u+in3d -R
186− vsub.f32 d16, d16, d2 @ in0u-in2d R
187− vadd.f32 d17, d17, d3 @ in2u+in1d -I
188−1:
189− vmul.f32 d7, d0, d21 @ I*s
190−A ldr r10, [r3, lr, lsr #1]
191−T lsr r10, lr, #1
192−T ldr r10, [r3, r10]
193− vmul.f32 d6, d1, d20 @ -R*c
194− ldr r6, [r3, #4]!
195− vmul.f32 d4, d1, d21 @ -R*s
196− vmul.f32 d5, d0, d20 @ I*c
197− vmul.f32 d24, d16, d30 @ R*c
198− vmul.f32 d25, d17, d31 @ -I*s
199− vmul.f32 d22, d16, d31 @ R*s
200− vmul.f32 d23, d17, d30 @ I*c
201− subs lr, lr, #16
202− vsub.f32 d6, d6, d7 @ -R*c-I*s
203− vadd.f32 d7, d4, d5 @ -R*s+I*c
204− vsub.f32 d24, d25, d24 @ I*s-R*c
205− vadd.f32 d25, d22, d23 @ R*s-I*c
206− beq 1f
207− mov r12, #-16
208− vld2.32 {d16,d18},[r9,:128],r12 @ in0u0,in0u1 in4d1,in4d0
209− vld2.32 {d17,d19},[r8,:128],r12 @ in2u0,in2u1 in3d1,in3d0
210− vneg.f32 d7, d7 @ R*s-I*c
211− vld2.32 {d0, d2}, [r7,:128]! @ in4u0,in4u1 in2d1,in2d0
212− vrev64.32 q9, q9 @ in4d0,in4d1 in3d0,in3d1
213− vld2.32 {d1, d3}, [r2,:128]! @ in3u0,in3u1 in1d1,in1d0
214− vsub.f32 d0, d18, d0 @ in4d-in4u I
215− vld2.32 {d20,d21},[r4,:128]! @ c0,c1 s0,s1
216− vrev64.32 q1, q1 @ in2d0,in2d1 in1d0,in1d1
217− vld2.32 {d30,d31},[r5,:128],r12 @ c2,c3 s2,s3
218− vadd.f32 d1, d1, d19 @ in3u+in3d -R
219− vsub.f32 d16, d16, d2 @ in0u-in2d R
220− vadd.f32 d17, d17, d3 @ in2u+in1d -I
221− uxth r12, r6, ror #16
222− uxth r6, r6
223− add r12, r1, r12, lsl #3
224− add r6, r1, r6, lsl #3
225− vst2.32 {d6[0],d7[0]}, [r6,:64]
226− vst2.32 {d6[1],d7[1]}, [r12,:64]
227− uxth r6, r10, ror #16
228− uxth r10, r10
229− add r6 , r1, r6, lsl #3
230− add r10, r1, r10, lsl #3
231− vst2.32 {d24[0],d25[0]},[r10,:64]
232− vst2.32 {d24[1],d25[1]},[r6,:64]
233− b 1b
234−1:
235− vneg.f32 d7, d7 @ R*s-I*c
236− uxth r12, r6, ror #16
237− uxth r6, r6
238− add r12, r1, r12, lsl #3
239− add r6, r1, r6, lsl #3
240− vst2.32 {d6[0],d7[0]}, [r6,:64]
241− vst2.32 {d6[1],d7[1]}, [r12,:64]
242− uxth r6, r10, ror #16
243− uxth r10, r10
244− add r6 , r1, r6, lsl #3
245− add r10, r1, r10, lsl #3
246− vst2.32 {d24[0],d25[0]},[r10,:64]
247− vst2.32 {d24[1],d25[1]},[r6,:64]
248−
249− mov r4, r0
250− mov r6, r1
251− bl ff_fft_calc_neon
252−
253− mov r12, #1
254− ldr lr, [r4, #20] @ mdct_bits
255− ldr r4, [r4, #24] @ tcos
256− lsl r12, r12, lr @ n = 1 << nbits
257− lsr lr, r12, #3 @ n8 = n >> 3
258−
259− add r4, r4, lr, lsl #3
260− add r6, r6, lr, lsl #3
261− sub r1, r4, #16
262− sub r3, r6, #16
263−
264− mov r7, #-16
265− mov r8, r6
266− mov r0, r3
267−
268− vld2.32 {d0-d1}, [r3,:128], r7 @ d0 =r1,i1 d1 =r0,i0
269− vld2.32 {d20-d21},[r6,:128]! @ d20=r2,i2 d21=r3,i3
270− vld2.32 {d16,d18},[r1,:128], r7 @ c1,c0 s1,s0
271−1:
272− subs lr, lr, #2
273− vmul.f32 d7, d0, d18 @ r1*s1,r0*s0
274− vld2.32 {d17,d19},[r4,:128]! @ c2,c3 s2,s3
275− vmul.f32 d4, d1, d18 @ i1*s1,i0*s0
276− vmul.f32 d5, d21, d19 @ i2*s2,i3*s3
277− vmul.f32 d6, d20, d19 @ r2*s2,r3*s3
278− vmul.f32 d24, d0, d16 @ r1*c1,r0*c0
279− vmul.f32 d25, d20, d17 @ r2*c2,r3*c3
280− vmul.f32 d22, d21, d17 @ i2*c2,i3*c3
281− vmul.f32 d23, d1, d16 @ i1*c1,i0*c0
282− vadd.f32 d4, d4, d24 @ i1*s1+r1*c1,i0*s0+r0*c0
283− vadd.f32 d5, d5, d25 @ i2*s2+r2*c2,i3*s3+r3*c3
284− vsub.f32 d6, d22, d6 @ i2*c2-r2*s2,i3*c3-r3*s3
285− vsub.f32 d7, d23, d7 @ i1*c1-r1*s1,i0*c0-r0*s0
286− vneg.f32 q2, q2
287− beq 1f
288− vld2.32 {d0-d1}, [r3,:128], r7
289− vld2.32 {d20-d21},[r6,:128]!
290− vld2.32 {d16,d18},[r1,:128], r7 @ c1,c0 s1,s0
291− vrev64.32 q3, q3
292− vst2.32 {d4,d6}, [r0,:128], r7
293− vst2.32 {d5,d7}, [r8,:128]!
294− b 1b
295−1:
296− vrev64.32 q3, q3
297− vst2.32 {d4,d6}, [r0,:128]
298− vst2.32 {d5,d7}, [r8,:128]
299−
300− pop {r4-r10,pc}
301−endfunc
libavcodec/arm/mdct_vfp.S+0−347
@@ -1,347 +0,0 @@
1−/*
2− * Copyright (c) 2013 RISC OS Open Ltd
3− * Author: Ben Avison <bavison@riscosopen.org>
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/arm/asm.S"
23−
24−CONTEXT .req a1
25−ORIGOUT .req a2
26−IN .req a3
27−OUT .req v1
28−REVTAB .req v2
29−TCOS .req v3
30−TSIN .req v4
31−OLDFPSCR .req v5
32−J0 .req a2
33−J1 .req a4
34−J2 .req ip
35−J3 .req lr
36−REVTAB_HI .req v5
37−IN_HI .req v6
38−OUT_HI .req v6
39−TCOS_HI .req sl
40−TSIN_HI .req fp
41−
42−.macro prerotation_innerloop
43− .set trig_lo, k
44− .set trig_hi, n4 - k - 2
45− .set in_lo, trig_lo * 2
46− .set in_hi, trig_hi * 2
47− vldr d8, [TCOS, #trig_lo*4] @ s16,s17
48− vldr d9, [TCOS, #trig_hi*4] @ s18,s19
49− vldr s0, [IN, #in_hi*4 + 12]
50− vldr s1, [IN, #in_hi*4 + 4]
51− vldr s2, [IN, #in_lo*4 + 12]
52− vldr s3, [IN, #in_lo*4 + 4]
53− vmul.f s8, s0, s16 @ vector operation
54− vldr d10, [TSIN, #trig_lo*4] @ s20,s21
55− vldr d11, [TSIN, #trig_hi*4] @ s22,s23
56− vldr s4, [IN, #in_lo*4]
57− vldr s5, [IN, #in_lo*4 + 8]
58− vldr s6, [IN, #in_hi*4]
59− vldr s7, [IN, #in_hi*4 + 8]
60− ldr J0, [REVTAB, #trig_lo*2]
61− vmul.f s12, s0, s20 @ vector operation
62− ldr J2, [REVTAB, #trig_hi*2]
63− mov J1, J0, lsr #16
64− and J0, J0, #255 @ halfword value will be < n4
65− vmls.f s8, s4, s20 @ vector operation
66− mov J3, J2, lsr #16
67− and J2, J2, #255 @ halfword value will be < n4
68− add J0, OUT, J0, lsl #3
69− vmla.f s12, s4, s16 @ vector operation
70− add J1, OUT, J1, lsl #3
71− add J2, OUT, J2, lsl #3
72− add J3, OUT, J3, lsl #3
73− vstr s8, [J0]
74− vstr s9, [J1]
75− vstr s10, [J2]
76− vstr s11, [J3]
77− vstr s12, [J0, #4]
78− vstr s13, [J1, #4]
79− vstr s14, [J2, #4]
80− vstr s15, [J3, #4]
81− .set k, k + 2
82−.endm
83−
84−.macro prerotation_innerloop_rolled
85− vldmia TCOS!, {s16,s17}
86− vldmdb TCOS_HI!, {s18,s19}
87− vldr s0, [IN_HI, #-4]
88− vldr s1, [IN_HI, #-12]
89− vldr s2, [IN, #12]
90− vldr s3, [IN, #4]
91− vmul.f s8, s0, s16 @ vector operation
92− vldmia TSIN!, {s20,s21}
93− vldmdb TSIN_HI!, {s22,s23}
94− vldr s4, [IN]
95− vldr s5, [IN, #8]
96− vldr s6, [IN_HI, #-16]
97− vldr s7, [IN_HI, #-8]
98− vmul.f s12, s0, s20 @ vector operation
99− add IN, IN, #16
100− sub IN_HI, IN_HI, #16
101− ldrh J0, [REVTAB], #2
102− ldrh J1, [REVTAB], #2
103− vmls.f s8, s4, s20 @ vector operation
104− ldrh J3, [REVTAB_HI, #-2]!
105− ldrh J2, [REVTAB_HI, #-2]!
106− add J0, OUT, J0, lsl #3
107− vmla.f s12, s4, s16 @ vector operation
108− add J1, OUT, J1, lsl #3
109− add J2, OUT, J2, lsl #3
110− add J3, OUT, J3, lsl #3
111− vstr s8, [J0]
112− vstr s9, [J1]
113− vstr s10, [J2]
114− vstr s11, [J3]
115− vstr s12, [J0, #4]
116− vstr s13, [J1, #4]
117− vstr s14, [J2, #4]
118− vstr s15, [J3, #4]
119−.endm
120−
121−.macro postrotation_innerloop tail, head
122− .set trig_lo_head, n8 - k - 2
123− .set trig_hi_head, n8 + k
124− .set out_lo_head, trig_lo_head * 2
125− .set out_hi_head, trig_hi_head * 2
126− .set trig_lo_tail, n8 - (k - 2) - 2
127− .set trig_hi_tail, n8 + (k - 2)
128− .set out_lo_tail, trig_lo_tail * 2
129− .set out_hi_tail, trig_hi_tail * 2
130− .if (k & 2) == 0
131− TCOS_D0_HEAD .req d10 @ s20,s21
132− TCOS_D1_HEAD .req d11 @ s22,s23
133− TCOS_S0_TAIL .req s24
134− .else
135− TCOS_D0_HEAD .req d12 @ s24,s25
136− TCOS_D1_HEAD .req d13 @ s26,s27
137− TCOS_S0_TAIL .req s20
138− .endif
139− .ifnc "\tail",""
140− vmls.f s8, s0, TCOS_S0_TAIL @ vector operation
141− .endif
142− .ifnc "\head",""
143− vldr d8, [TSIN, #trig_lo_head*4] @ s16,s17
144− vldr d9, [TSIN, #trig_hi_head*4] @ s18,s19
145− vldr TCOS_D0_HEAD, [TCOS, #trig_lo_head*4]
146− .endif
147− .ifnc "\tail",""
148− vmla.f s12, s4, TCOS_S0_TAIL @ vector operation
149− .endif
150− .ifnc "\head",""
151− vldr s0, [OUT, #out_lo_head*4]
152− vldr s1, [OUT, #out_lo_head*4 + 8]
153− vldr s2, [OUT, #out_hi_head*4]
154− vldr s3, [OUT, #out_hi_head*4 + 8]
155− vldr s4, [OUT, #out_lo_head*4 + 4]
156− vldr s5, [OUT, #out_lo_head*4 + 12]
157− vldr s6, [OUT, #out_hi_head*4 + 4]
158− vldr s7, [OUT, #out_hi_head*4 + 12]
159− .endif
160− .ifnc "\tail",""
161− vstr s8, [OUT, #out_lo_tail*4]
162− vstr s9, [OUT, #out_lo_tail*4 + 8]
163− vstr s10, [OUT, #out_hi_tail*4]
164− vstr s11, [OUT, #out_hi_tail*4 + 8]
165− .endif
166− .ifnc "\head",""
167− vmul.f s8, s4, s16 @ vector operation
168− .endif
169− .ifnc "\tail",""
170− vstr s12, [OUT, #out_hi_tail*4 + 12]
171− vstr s13, [OUT, #out_hi_tail*4 + 4]
172− vstr s14, [OUT, #out_lo_tail*4 + 12]
173− vstr s15, [OUT, #out_lo_tail*4 + 4]
174− .endif
175− .ifnc "\head",""
176− vmul.f s12, s0, s16 @ vector operation
177− vldr TCOS_D1_HEAD, [TCOS, #trig_hi_head*4]
178− .endif
179− .unreq TCOS_D0_HEAD
180− .unreq TCOS_D1_HEAD
181− .unreq TCOS_S0_TAIL
182− .ifnc "\head",""
183− .set k, k + 2
184− .endif
185−.endm
186−
187−.macro postrotation_innerloop_rolled tail, head, tcos_s0_head, tcos_s1_head, tcos_s2_head, tcos_s3_head, tcos_s0_tail, out_offset_head, out_offset_tail
188− .ifnc "\tail",""
189− vmls.f s8, s0, \tcos_s0_tail @ vector operation
190− .endif
191− .ifnc "\head",""
192− vldmia TSIN!, {s16,s17}
193− vldmdb TSIN_HI!, {s18,s19}
194− vldmia TCOS!, {\tcos_s0_head,\tcos_s1_head}
195− .endif
196− .ifnc "\tail",""
197− vmla.f s12, s4, \tcos_s0_tail @ vector operation
198− .endif
199− .ifnc "\head",""
200− vldr s0, [OUT, #+\out_offset_head+0]
201− vldr s1, [OUT, #+\out_offset_head+8]
202− vldr s2, [OUT_HI, #-\out_offset_head-16]
203− vldr s3, [OUT_HI, #-\out_offset_head-8]
204− vldr s4, [OUT, #+\out_offset_head+4]
205− vldr s5, [OUT, #+\out_offset_head+12]
206− vldr s6, [OUT_HI, #-\out_offset_head-12]
207− vldr s7, [OUT_HI, #-\out_offset_head-4]
208− .endif
209− .ifnc "\tail",""
210− vstr s8, [OUT, #+\out_offset_tail+0]
211− vstr s9, [OUT, #+\out_offset_tail+8]
212− vstr s10, [OUT_HI, #-\out_offset_tail-16]
213− vstr s11, [OUT_HI, #-\out_offset_tail-8]
214− .endif
215− .ifnc "\head",""
216− vmul.f s8, s4, s16 @ vector operation
217− .endif
218− .ifnc "\tail",""
219− vstr s12, [OUT_HI, #-\out_offset_tail-4]
220− vstr s13, [OUT_HI, #-\out_offset_tail-12]
221− vstr s14, [OUT, #+\out_offset_tail+12]
222− vstr s15, [OUT, #+\out_offset_tail+4]
223− .endif
224− .ifnc "\head",""
225− vmul.f s12, s0, s16 @ vector operation
226− vldmdb TCOS_HI!, {\tcos_s2_head,\tcos_s3_head}
227− .endif
228−.endm
229−
230−
231−/* void ff_imdct_half_vfp(FFTContext *s,
232− * FFTSample *output,
233− * const FFTSample *input)
234− */
235−function ff_imdct_half_vfp, export=1
236− ldr ip, [CONTEXT, #5*4] @ mdct_bits
237− teq ip, #6
238− bne 10f
239−
240− .set n, 1<<6
241− .set n2, n/2
242− .set n4, n/4
243− .set n8, n/8
244−
245− push {v1-v5,lr}
246− vpush {s16-s27}
247− fmrx OLDFPSCR, FPSCR
248− ldr lr, =0x03030000 @ RunFast mode, short vectors of length 4, stride 1
249− fmxr FPSCR, lr
250− mov OUT, ORIGOUT
251− ldr REVTAB, [CONTEXT, #2*4]
252− ldr TCOS, [CONTEXT, #6*4]
253− ldr TSIN, [CONTEXT, #7*4]
254−
255− .set k, 0
256− .rept n8/2
257− prerotation_innerloop
258− .endr
259−
260− fmxr FPSCR, OLDFPSCR
261− mov a1, OUT
262− bl X(ff_fft16_vfp)
263− ldr lr, =0x03030000 @ RunFast mode, short vectors of length 4, stride 1
264− fmxr FPSCR, lr
265−
266− .set k, 0
267− postrotation_innerloop , head
268− .rept n8/2 - 1
269− postrotation_innerloop tail, head
270− .endr
271− postrotation_innerloop tail
272−
273− fmxr FPSCR, OLDFPSCR
274− vpop {s16-s27}
275− pop {v1-v5,pc}
276−
277−10:
278− push {v1-v6,sl,fp,lr}
279− vpush {s16-s27}
280− fmrx OLDFPSCR, FPSCR
281− ldr lr, =0x03030000 @ RunFast mode, short vectors of length 4, stride 1
282− fmxr FPSCR, lr
283− mov lr, #1
284− mov OUT, ORIGOUT
285− ldr REVTAB, [CONTEXT, #2*4]
286− ldr TCOS, [CONTEXT, #6*4]
287− ldr TSIN, [CONTEXT, #7*4]
288− mov lr, lr, lsl ip
289−
290− push {CONTEXT,OLDFPSCR}
291− add IN_HI, IN, lr, lsl #1
292− add REVTAB_HI, REVTAB, lr, lsr #1
293− add TCOS_HI, TCOS, lr
294− add TSIN_HI, TSIN, lr
295−0: prerotation_innerloop_rolled
296− teq IN, IN_HI
297− bne 0b
298− ldmia sp, {CONTEXT,OLDFPSCR}
299−
300− mov ORIGOUT, OUT
301− fmxr FPSCR, OLDFPSCR
302− ldr ip, [CONTEXT, #9*4]
303− blx ip @ s->fft_calc(s, output)
304−
305− pop {CONTEXT,OLDFPSCR}
306− ldr lr, =0x03030000 @ RunFast mode, short vectors of length 4, stride 1
307− ldr ip, [CONTEXT, #5*4] @ mdct_bits
308− fmxr FPSCR, lr
309− mov lr, #1
310− mov lr, lr, lsl ip
311− sub TCOS, TCOS, lr, lsr #1
312− sub TSIN, TSIN, lr, lsr #1
313− add OUT_HI, OUT, lr, lsl #1
314− add TCOS_HI, TCOS, lr
315− add TSIN_HI, TSIN, lr
316− postrotation_innerloop_rolled , head, s20, s21, s22, s23,, 0
317− b 1f
318−0: add OUT, OUT, #32
319− sub OUT_HI, OUT_HI, #32
320− postrotation_innerloop_rolled tail, head, s20, s21, s22, s23, s24, 0, -16
321−1: postrotation_innerloop_rolled tail, head, s24, s25, s26, s27, s20, 16, 0
322− teq TSIN, TSIN_HI
323− bne 0b
324− postrotation_innerloop_rolled tail,,,,,, s24,, 16
325−
326− fmxr FPSCR, OLDFPSCR
327− vpop {s16-s27}
328− pop {v1-v6,sl,fp,pc}
329−endfunc
330−
331− .unreq CONTEXT
332− .unreq ORIGOUT
333− .unreq IN
334− .unreq OUT
335− .unreq REVTAB
336− .unreq TCOS
337− .unreq TSIN
338− .unreq OLDFPSCR
339− .unreq J0
340− .unreq J1
341− .unreq J2
342− .unreq J3
343− .unreq REVTAB_HI
344− .unreq IN_HI
345− .unreq OUT_HI
346− .unreq TCOS_HI
347− .unreq TSIN_HI
libavcodec/arm/rdft_init_arm.c+0−33
@@ -1,33 +0,0 @@
1−/*
2− * This file is part of FFmpeg.
3− *
4− * FFmpeg is free software; you can redistribute it and/or
5− * modify it under the terms of the GNU Lesser General Public
6− * License as published by the Free Software Foundation; either
7− * version 2.1 of the License, or (at your option) any later version.
8− *
9− * FFmpeg is distributed in the hope that it will be useful,
10− * but WITHOUT ANY WARRANTY; without even the implied warranty of
11− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12− * Lesser General Public License for more details.
13− *
14− * You should have received a copy of the GNU Lesser General Public
15− * License along with FFmpeg; if not, write to the Free Software
16− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17− */
18−
19−#include "libavutil/attributes.h"
20−#include "libavutil/cpu.h"
21−#include "libavutil/arm/cpu.h"
22−
23−#include "libavcodec/rdft.h"
24−
25−void ff_rdft_calc_neon(struct RDFTContext *s, FFTSample *z);
26−
27−av_cold void ff_rdft_init_arm(RDFTContext *s)
28−{
29− int cpu_flags = av_get_cpu_flags();
30−
31− if (have_neon(cpu_flags))
32− s->rdft_calc = ff_rdft_calc_neon;
33−}
libavcodec/arm/rdft_neon.S+0−155
@@ -1,155 +0,0 @@
1−/*
2− * ARM NEON optimised RDFT
3− * Copyright (c) 2009 Mans Rullgard <mans@mansr.com>
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/arm/asm.S"
23−
24−function ff_rdft_calc_neon, export=1
25− push {r4-r8,lr}
26−
27− ldr r6, [r0, #4] @ inverse
28− mov r4, r0
29− mov r5, r1
30−
31− lsls r6, r6, #31
32− bne 1f
33− add r0, r4, #24
34− bl X(ff_fft_permute_neon)
35− add r0, r4, #24
36− mov r1, r5
37− bl X(ff_fft_calc_neon)
38−1:
39− ldr r12, [r4, #0] @ nbits
40− mov r2, #1
41− ldr r8, [r4, #20] @ negative_sin
42− lsl r12, r2, r12
43− add r0, r5, #8
44− lsl r8, r8, #31
45− add r1, r5, r12, lsl #2
46− lsr r12, r12, #2
47− vdup.32 d26, r8
48− ldr r2, [r4, #12] @ tcos
49− sub r12, r12, #2
50− ldr r3, [r4, #16] @ tsin
51− mov r7, r0
52− sub r1, r1, #8
53− mov lr, r1
54− mov r8, #-8
55− vld1.32 {d0}, [r0,:64]! @ d1[0,1]
56− vld1.32 {d1}, [r1,:64], r8 @ d2[0,1]
57− vld1.32 {d4}, [r2,:64]! @ tcos[i]
58− vld1.32 {d5}, [r3,:64]! @ tsin[i]
59− vmov.f32 d18, #0.5 @ k1
60− vdup.32 d19, r6
61− veor d5, d26, d5
62− pld [r0, #32]
63− veor d19, d18, d19 @ k2
64− vmov.i32 d16, #0
65− vmov.i32 d17, #1<<31
66− pld [r1, #-32]
67− vtrn.32 d16, d17
68− pld [r2, #32]
69− vrev64.32 d16, d16 @ d16=1,0 d17=0,1
70− pld [r3, #32]
71−2:
72− veor q1, q0, q8 @ -d1[0],d1[1], d2[0],-d2[1]
73− vld1.32 {d24}, [r0,:64]! @ d1[0,1]
74− vadd.f32 d0, d0, d3 @ d1[0]+d2[0], d1[1]-d2[1]
75− vld1.32 {d25}, [r1,:64], r8 @ d2[0,1]
76− vadd.f32 d1, d2, d1 @ -d1[0]+d2[0], d1[1]+d2[1]
77− veor q3, q12, q8 @ -d1[0],d1[1], d2[0],-d2[1]
78− pld [r0, #32]
79− vmul.f32 q10, q0, q9 @ ev.re, ev.im, od.im, od.re
80− pld [r1, #-32]
81− vadd.f32 d0, d24, d7 @ d1[0]+d2[0], d1[1]-d2[1]
82− vadd.f32 d1, d6, d25 @ -d1[0]+d2[0], d1[1]+d2[1]
83− vmul.f32 q11, q0, q9 @ ev.re, ev.im, od.im, od.re
84− veor d7, d21, d16 @ -od.im, od.re
85− vrev64.32 d3, d21 @ od.re, od.im
86− veor d6, d20, d17 @ ev.re,-ev.im
87− veor d2, d3, d16 @ -od.re, od.im
88− vmla.f32 d20, d3, d4[1]
89− vmla.f32 d20, d7, d5[1]
90− vmla.f32 d6, d2, d4[1]
91− vmla.f32 d6, d21, d5[1]
92− vld1.32 {d4}, [r2,:64]! @ tcos[i]
93− veor d7, d23, d16 @ -od.im, od.re
94− vld1.32 {d5}, [r3,:64]! @ tsin[i]
95− veor d24, d22, d17 @ ev.re,-ev.im
96− vrev64.32 d3, d23 @ od.re, od.im
97− veor d5, d26, d5
98− pld [r2, #32]
99− veor d2, d3, d16 @ -od.re, od.im
100− pld [r3, #32]
101− vmla.f32 d22, d3, d4[0]
102− vmla.f32 d22, d7, d5[0]
103− vmla.f32 d24, d2, d4[0]
104− vmla.f32 d24, d23, d5[0]
105− vld1.32 {d0}, [r0,:64]! @ d1[0,1]
106− vld1.32 {d1}, [r1,:64], r8 @ d2[0,1]
107− vst1.32 {d20}, [r7,:64]!
108− vst1.32 {d6}, [lr,:64], r8
109− vst1.32 {d22}, [r7,:64]!
110− vst1.32 {d24}, [lr,:64], r8
111− subs r12, r12, #2
112− bgt 2b
113−
114− veor q1, q0, q8 @ -d1[0],d1[1], d2[0],-d2[1]
115− vadd.f32 d0, d0, d3 @ d1[0]+d2[0], d1[1]-d2[1]
116− vadd.f32 d1, d2, d1 @ -d1[0]+d2[0], d1[1]+d2[1]
117− ldr r2, [r4, #8] @ sign_convention
118− vmul.f32 q10, q0, q9 @ ev.re, ev.im, od.im, od.re
119− add r0, r0, #4
120− bfc r2, #0, #31
121− vld1.32 {d0[0]}, [r0,:32]
122− veor d7, d21, d16 @ -od.im, od.re
123− vrev64.32 d3, d21 @ od.re, od.im
124− veor d6, d20, d17 @ ev.re,-ev.im
125− vld1.32 {d22}, [r5,:64]
126− vdup.32 d1, r2
127− vmov d23, d22
128− veor d2, d3, d16 @ -od.re, od.im
129− vtrn.32 d22, d23
130− veor d0, d0, d1
131− veor d23, d23, d17
132− vmla.f32 d20, d3, d4[1]
133− vmla.f32 d20, d7, d5[1]
134− vmla.f32 d6, d2, d4[1]
135− vmla.f32 d6, d21, d5[1]
136− vadd.f32 d22, d22, d23
137− vst1.32 {d20}, [r7,:64]
138− vst1.32 {d6}, [lr,:64]
139− vst1.32 {d0[0]}, [r0,:32]
140− vst1.32 {d22}, [r5,:64]
141−
142− cmp r6, #0
143− it eq
144− popeq {r4-r8,pc}
145−
146− vmul.f32 d22, d22, d18
147− vst1.32 {d22}, [r5,:64]
148− add r0, r4, #24
149− mov r1, r5
150− bl X(ff_fft_permute_neon)
151− add r0, r4, #24
152− mov r1, r5
153− pop {r4-r8,lr}
154− b X(ff_fft_calc_neon)
155−endfunc
libavcodec/arm/synth_filter_init_arm.c+0−1
@@ -23,7 +23,6 @@ · 22 unmodified lines
2323 #include "libavutil/arm/cpu.h"
2424 #include "libavutil/attributes.h"
2525 #include "libavutil/internal.h"
26−#include "libavcodec/fft.h"
2726 #include "libavcodec/synth_filter.h"
2827
2928 void ff_synth_filter_float_vfp(AVTXContext *imdct,
libavcodec/cos_tablegen.c+0−80
@@ -1,80 +0,0 @@
1−/*
2− * Generate a header file for hardcoded ff_cos_* tables
3− *
4− * Copyright (c) 2009 Reimar Döffinger <Reimar.Doeffinger@gmx.de>
5− *
6− * This file is part of FFmpeg.
7− *
8− * FFmpeg is free software; you can redistribute it and/or
9− * modify it under the terms of the GNU Lesser General Public
10− * License as published by the Free Software Foundation; either
11− * version 2.1 of the License, or (at your option) any later version.
12− *
13− * FFmpeg is distributed in the hope that it will be useful,
14− * but WITHOUT ANY WARRANTY; without even the implied warranty of
15− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16− * Lesser General Public License for more details.
17− *
18− * You should have received a copy of the GNU Lesser General Public
19− * License along with FFmpeg; if not, write to the Free Software
20− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21− */
22−
23−#include <stdio.h>
24−#include <string.h>
25−#include <math.h>
26−
27−#include "libavutil/mathematics.h"
28−
29−#define BITS 17
30−#define FLOATFMT "%.18e"
31−#define FIXEDFMT "%6d"
32−
33−static int clip_f15(int v)
34−{
35− return v < -32767 ? -32767 :
36− v > 32767 ? 32767 :
37− v;
38−}
39−
40−static void printval(double val, int fixed)
41−{
42− if (fixed) {
43− /* lrint() isn't always available, so round and cast manually. */
44− double new_val = val * (double) (1 << 15);
45−
46− new_val = new_val >= 0 ? floor(new_val + 0.5) : ceil(new_val - 0.5);
47−
48− printf(" "FIXEDFMT",", clip_f15((long int) new_val));
49− } else {
50− printf(" "FLOATFMT",", val);
51− }
52−}
53−
54−int main(int argc, char *argv[])
55−{
56− int i, j;
57− int do_sin = argc > 1 && !strcmp(argv[1], "sin");
58− int fixed = argc > 1 && strstr(argv[1], "fixed");
59− double (*func)(double) = do_sin ? sin : cos;
60−
61− printf("/* This file was automatically generated. */\n");
62− printf("#define FFT_FLOAT %d\n", !fixed);
63− printf("#include \"libavcodec/%s\"\n", do_sin ? "rdft.h" : "fft.h");
64− for (i = 4; i <= BITS; i++) {
65− int m = 1 << i;
66− double freq = 2*M_PI/m;
67− printf("%s(%i) = {\n ", do_sin ? "SINTABLE" : "COSTABLE", m);
68− for (j = 0; j < m/2 - 1; j++) {
69− int idx = j > m/4 ? m/2 - j : j;
70− if (do_sin && j >= m/4)
71− idx = m/4 - j;
72− printval(func(idx*freq), fixed);
73− if ((j & 3) == 3)
74− printf("\n ");
75− }
76− printval(func(do_sin ? -(m/4 - 1)*freq : freq), fixed);
77− printf("\n};\n");
78− }
79− return 0;
80−}
libavcodec/dct.c+0−228
@@ -1,228 +0,0 @@
1−/*
2− * (I)DCT Transforms
3− * Copyright (c) 2009 Peter Ross <pross@xvid.org>
4− * Copyright (c) 2010 Alex Converse <alex.converse@gmail.com>
5− * Copyright (c) 2010 Vitor Sessak
6− *
7− * This file is part of FFmpeg.
8− *
9− * FFmpeg is free software; you can redistribute it and/or
10− * modify it under the terms of the GNU Lesser General Public
11− * License as published by the Free Software Foundation; either
12− * version 2.1 of the License, or (at your option) any later version.
13− *
14− * FFmpeg is distributed in the hope that it will be useful,
15− * but WITHOUT ANY WARRANTY; without even the implied warranty of
16− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17− * Lesser General Public License for more details.
18− *
19− * You should have received a copy of the GNU Lesser General Public
20− * License along with FFmpeg; if not, write to the Free Software
21− * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
22− */
23−
24−/**
25− * @file
26− * (Inverse) Discrete Cosine Transforms. These are also known as the
27− * type II and type III DCTs respectively.
28− */
29−
30−#include <math.h>
31−#include <string.h>
32−
33−#include "libavutil/error.h"
34−#include "libavutil/mathematics.h"
35−#include "libavutil/mem.h"
36−#include "dct.h"
37−#include "dct32.h"
38−
39−/* sin((M_PI * x / (2 * n)) */
40−#define SIN(s, n, x) (s->costab[(n) - (x)])
41−
42−/* cos((M_PI * x / (2 * n)) */
43−#define COS(s, n, x) (s->costab[x])
44−
45−static void dst_calc_I_c(DCTContext *ctx, FFTSample *data)
46−{
47− int n = 1 << ctx->nbits;
48− int i;
49−
50− data[0] = 0;
51− for (i = 1; i < n / 2; i++) {
52− float tmp1 = data[i ];
53− float tmp2 = data[n - i];
54− float s = SIN(ctx, n, 2 * i);
55−
56− s *= tmp1 + tmp2;
57− tmp1 = (tmp1 - tmp2) * 0.5f;
58− data[i] = s + tmp1;
59− data[n - i] = s - tmp1;
60− }
61−
62− data[n / 2] *= 2;
63− ctx->rdft.rdft_calc(&ctx->rdft, data);
64−
65− data[0] *= 0.5f;
66−
67− for (i = 1; i < n - 2; i += 2) {
68− data[i + 1] += data[i - 1];
69− data[i] = -data[i + 2];
70− }
71−
72− data[n - 1] = 0;
73−}
74−
75−static void dct_calc_I_c(DCTContext *ctx, FFTSample *data)
76−{
77− int n = 1 << ctx->nbits;
78− int i;
79− float next = -0.5f * (data[0] - data[n]);
80−
81− for (i = 0; i < n / 2; i++) {
82− float tmp1 = data[i];
83− float tmp2 = data[n - i];
84− float s = SIN(ctx, n, 2 * i);
85− float c = COS(ctx, n, 2 * i);
86−
87− c *= tmp1 - tmp2;
88− s *= tmp1 - tmp2;
89−
90− next += c;
91−
92− tmp1 = (tmp1 + tmp2) * 0.5f;
93− data[i] = tmp1 - s;
94− data[n - i] = tmp1 + s;
95− }
96−
97− ctx->rdft.rdft_calc(&ctx->rdft, data);
98− data[n] = data[1];
99− data[1] = next;
100−
101− for (i = 3; i <= n; i += 2)
102− data[i] = data[i - 2] - data[i];
103−}
104−
105−static void dct_calc_III_c(DCTContext *ctx, FFTSample *data)
106−{
107− int n = 1 << ctx->nbits;
108− int i;
109−
110− float next = data[n - 1];
111− float inv_n = 1.0f / n;
112−
113− for (i = n - 2; i >= 2; i -= 2) {
114− float val1 = data[i];
115− float val2 = data[i - 1] - data[i + 1];
116− float c = COS(ctx, n, i);
117− float s = SIN(ctx, n, i);
118−
119− data[i] = c * val1 + s * val2;
120− data[i + 1] = s * val1 - c * val2;
121− }
122−
123− data[1] = 2 * next;
124−
125− ctx->rdft.rdft_calc(&ctx->rdft, data);
126−
127− for (i = 0; i < n / 2; i++) {
128− float tmp1 = data[i] * inv_n;
129− float tmp2 = data[n - i - 1] * inv_n;
130− float csc = ctx->csc2[i] * (tmp1 - tmp2);
131−
132− tmp1 += tmp2;
133− data[i] = tmp1 + csc;
134− data[n - i - 1] = tmp1 - csc;
135− }
136−}
137−
138−static void dct_calc_II_c(DCTContext *ctx, FFTSample *data)
139−{
140− int n = 1 << ctx->nbits;
141− int i;
142− float next;
143−
144− for (i = 0; i < n / 2; i++) {
145− float tmp1 = data[i];
146− float tmp2 = data[n - i - 1];
147− float s = SIN(ctx, n, 2 * i + 1);
148−
149− s *= tmp1 - tmp2;
150− tmp1 = (tmp1 + tmp2) * 0.5f;
151−
152− data[i] = tmp1 + s;
153− data[n-i-1] = tmp1 - s;
154− }
155−
156− ctx->rdft.rdft_calc(&ctx->rdft, data);
157−
158− next = data[1] * 0.5;
159− data[1] *= -1;
160−
161− for (i = n - 2; i >= 0; i -= 2) {
162− float inr = data[i ];
163− float ini = data[i + 1];
164− float c = COS(ctx, n, i);
165− float s = SIN(ctx, n, i);
166−
167− data[i] = c * inr + s * ini;
168− data[i + 1] = next;
169−
170− next += s * inr - c * ini;
171− }
172−}
173−
174−static void dct32_func(DCTContext *ctx, FFTSample *data)
175−{
176− ctx->dct32(data, data);
177−}
178−
179−av_cold int ff_dct_init(DCTContext *s, int nbits, enum DCTTransformType inverse)
180−{
181− int n = 1 << nbits;
182− int i;
183− int ret;
184−
185− memset(s, 0, sizeof(*s));
186−
187− s->nbits = nbits;
188− s->inverse = inverse;
189−
190− if (inverse == DCT_II && nbits == 5) {
191− s->dct_calc = dct32_func;
192− } else {
193− ff_init_ff_cos_tabs(nbits + 2);
194−
195− s->costab = ff_cos_tabs[nbits + 2];
196− s->csc2 = av_malloc_array(n / 2, sizeof(FFTSample));
197− if (!s->csc2)
198− return AVERROR(ENOMEM);
199−
200− if ((ret = ff_rdft_init(&s->rdft, nbits, inverse == DCT_III)) < 0) {
201− av_freep(&s->csc2);
202− return ret;
203− }
204−
205− for (i = 0; i < n / 2; i++)
206− s->csc2[i] = 0.5 / sin((M_PI / (2 * n) * (2 * i + 1)));
207−
208− switch (inverse) {
209− case DCT_I : s->dct_calc = dct_calc_I_c; break;
210− case DCT_II : s->dct_calc = dct_calc_II_c; break;
211− case DCT_III: s->dct_calc = dct_calc_III_c; break;
212− case DST_I : s->dct_calc = dst_calc_I_c; break;
213− }
214− }
215−
216− s->dct32 = ff_dct32_float;
217−#if ARCH_X86
218− ff_dct_init_x86(s);
219−#endif
220−
221− return 0;
222−}
223−
224−av_cold void ff_dct_end(DCTContext *s)
225−{
226− ff_rdft_end(&s->rdft);
227− av_freep(&s->csc2);
228−}
libavcodec/dct.h+1−26
@@ -21,37 +21,12 @@ · 20 unmodified lines
2121 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
2222 */
2323
24−#if !defined(AVCODEC_DCT_H) && (!defined(FFT_FLOAT) || FFT_FLOAT)
24+#ifndef AVCODEC_DCT_H
2525 #define AVCODEC_DCT_H
2626
2727 #include <stddef.h>
2828 #include <stdint.h>
2929
30−#include "rdft.h"
31−
32−struct DCTContext {
33− int nbits;
34− int inverse;
35− RDFTContext rdft;
36− const float *costab;
37− FFTSample *csc2;
38− void (*dct_calc)(struct DCTContext *s, FFTSample *data);
39− void (*dct32)(FFTSample *out, const FFTSample *in);
40−};
41−
42−/**
43− * Set up DCT.
44− * @param nbits size of the input array:
45− * (1 << nbits) for DCT-II, DCT-III and DST-I
46− * (1 << nbits) + 1 for DCT-I
47− *
48− * @note the first element of the input of DST-I is ignored
49− */
50−int ff_dct_init(DCTContext *s, int nbits, enum DCTTransformType type);
51−void ff_dct_end (DCTContext *s);
52−
53−void ff_dct_init_x86(DCTContext *s);
54−
5530 void ff_j_rev_dct(int16_t *data);
5631 void ff_j_rev_dct4(int16_t *data);
5732 void ff_j_rev_dct2(int16_t *data);
libavcodec/fft-internal.h+0−62
@@ -1,62 +0,0 @@
1−/*
2− * This file is part of FFmpeg.
3− *
4− * FFmpeg is free software; you can redistribute it and/or
5− * modify it under the terms of the GNU Lesser General Public
6− * License as published by the Free Software Foundation; either
7− * version 2.1 of the License, or (at your option) any later version.
8− *
9− * FFmpeg is distributed in the hope that it will be useful,
10− * but WITHOUT ANY WARRANTY; without even the implied warranty of
11− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12− * Lesser General Public License for more details.
13− *
14− * You should have received a copy of the GNU Lesser General Public
15− * License along with FFmpeg; if not, write to the Free Software
16− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17− */
18−
19−#ifndef AVCODEC_FFT_INTERNAL_H
20−#define AVCODEC_FFT_INTERNAL_H
21−
22−#include "libavutil/mathematics.h"
23−#include "fft.h"
24−
25−#if FFT_FLOAT
26−
27−#define FIX15(v) (v)
28−#define sqrthalf (float)M_SQRT1_2
29−
30−#define BF(x, y, a, b) do { \
31− x = a - b; \
32− y = a + b; \
33− } while (0)
34−
35−#define CMUL(dre, dim, are, aim, bre, bim) do { \
36− (dre) = (are) * (bre) - (aim) * (bim); \
37− (dim) = (are) * (bim) + (aim) * (bre); \
38− } while (0)
39−
40−#else /* FFT_FLOAT */
41−
42−#define CMUL(dre, dim, are, aim, bre, bim) do { \
43− int64_t accu; \
44− (accu) = (int64_t)(bre) * (are); \
45− (accu) -= (int64_t)(bim) * (aim); \
46− (dre) = (int)(((accu) + 0x40000000) >> 31); \
47− (accu) = (int64_t)(bre) * (aim); \
48− (accu) += (int64_t)(bim) * (are); \
49− (dim) = (int)(((accu) + 0x40000000) >> 31); \
50− } while (0)
51−
52−#endif /* FFT_FLOAT */
53−
54−#define ff_imdct_calc_c FFT_NAME(ff_imdct_calc_c)
55−#define ff_imdct_half_c FFT_NAME(ff_imdct_half_c)
56−#define ff_mdct_calc_c FFT_NAME(ff_mdct_calc_c)
57−
58−void ff_imdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input);
59−void ff_imdct_half_c(FFTContext *s, FFTSample *output, const FFTSample *input);
60−void ff_mdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input);
61−
62−#endif /* AVCODEC_FFT_INTERNAL_H */
libavcodec/fft.h+0−160
@@ -1,160 +0,0 @@
1−/*
2− * Copyright (c) 2000, 2001, 2002 Fabrice Bellard
3− * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
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−#ifndef AVCODEC_FFT_H
23−#define AVCODEC_FFT_H
24−
25−#ifndef FFT_FLOAT
26−#define FFT_FLOAT 1
27−#endif
28−
29−#include <stdint.h>
30−#include "config.h"
31−
32−#include "libavutil/attributes_internal.h"
33−#include "libavutil/mem_internal.h"
34−
35−#if FFT_FLOAT
36−
37−#include "avfft.h"
38−
39−#define FFT_NAME(x) x
40−
41−typedef float FFTDouble;
42−
43−#else
44−
45−#define Q31(x) (int)((x)*2147483648.0 + 0.5)
46−#define FFT_NAME(x) x ## _fixed_32
47−
48−typedef int32_t FFTSample;
49−
50−typedef struct FFTComplex {
51− FFTSample re, im;
52−} FFTComplex;
53−
54−typedef int FFTDouble;
55−typedef struct FFTContext FFTContext;
56−
57−#endif /* FFT_FLOAT */
58−
59−typedef struct FFTDComplex {
60− FFTDouble re, im;
61−} FFTDComplex;
62−
63−/* FFT computation */
64−
65−enum fft_permutation_type {
66− FF_FFT_PERM_DEFAULT,
67− FF_FFT_PERM_SWAP_LSBS,
68− FF_FFT_PERM_AVX,
69−};
70−
71−enum mdct_permutation_type {
72− FF_MDCT_PERM_NONE,
73− FF_MDCT_PERM_INTERLEAVE,
74−};
75−
76−struct FFTContext {
77− int nbits;
78− int inverse;
79− uint16_t *revtab;
80− FFTComplex *tmp_buf;
81− int mdct_size; /* size of MDCT (i.e. number of input data * 2) */
82− int mdct_bits; /* n = 2^nbits */
83− /* pre/post rotation tables */
84− FFTSample *tcos;
85− FFTSample *tsin;
86− /**
87− * Do the permutation needed BEFORE calling fft_calc().
88− */
89− void (*fft_permute)(struct FFTContext *s, FFTComplex *z);
90− /**
91− * Do a complex FFT with the parameters defined in ff_fft_init(). The
92− * input data must be permuted before. No 1.0/sqrt(n) normalization is done.
93− */
94− void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
95− void (*imdct_calc)(struct FFTContext *s, FFTSample *output, const FFTSample *input);
96− void (*imdct_half)(struct FFTContext *s, FFTSample *output, const FFTSample *input);
97− void (*mdct_calc)(struct FFTContext *s, FFTSample *output, const FFTSample *input);
98− enum fft_permutation_type fft_permutation;
99− enum mdct_permutation_type mdct_permutation;
100− uint32_t *revtab32;
101−};
102−
103−#if CONFIG_HARDCODED_TABLES
104−#define COSTABLE_CONST const
105−#define ff_init_ff_cos_tabs(index)
106−#else
107−#define COSTABLE_CONST
108−#define ff_init_ff_cos_tabs FFT_NAME(ff_init_ff_cos_tabs)
109−
110−/**
111− * Initialize the cosine table in ff_cos_tabs[index]
112− * @param index index in ff_cos_tabs array of the table to initialize
113− */
114−void ff_init_ff_cos_tabs(int index);
115−#endif
116−
117−#define COSTABLE(size) \
118− COSTABLE_CONST attribute_visibility_hidden DECLARE_ALIGNED(32, FFTSample, FFT_NAME(ff_cos_##size))[size/2]
119−
120−extern COSTABLE(16);
121−extern COSTABLE(32);
122−extern COSTABLE(64);
123−extern COSTABLE(128);
124−extern COSTABLE(256);
125−extern COSTABLE(512);
126−extern COSTABLE(1024);
127−extern COSTABLE(2048);
128−extern COSTABLE(4096);
129−extern COSTABLE(8192);
130−extern COSTABLE(16384);
131−extern COSTABLE(32768);
132−extern COSTABLE(65536);
133−extern COSTABLE(131072);
134−extern COSTABLE_CONST FFTSample* const FFT_NAME(ff_cos_tabs)[18];
135−
136−#define ff_fft_init FFT_NAME(ff_fft_init)
137−#define ff_fft_end FFT_NAME(ff_fft_end)
138−
139−/**
140− * Set up a complex FFT.
141− * @param nbits log2 of the length of the input array
142− * @param inverse if 0 perform the forward transform, if 1 perform the inverse
143− */
144−int ff_fft_init(FFTContext *s, int nbits, int inverse);
145−
146−void ff_fft_init_aarch64(FFTContext *s);
147−void ff_fft_init_x86(FFTContext *s);
148−void ff_fft_init_arm(FFTContext *s);
149−void ff_fft_init_mips(FFTContext *s);
150−void ff_fft_init_ppc(FFTContext *s);
151−
152−void ff_fft_end(FFTContext *s);
153−
154−#define ff_mdct_init FFT_NAME(ff_mdct_init)
155−#define ff_mdct_end FFT_NAME(ff_mdct_end)
156−
157−int ff_mdct_init(FFTContext *s, int nbits, int inverse, double scale);
158−void ff_mdct_end(FFTContext *s);
159−
160−#endif /* AVCODEC_FFT_H */
libavcodec/fft_fixed_32.c+0−51
@@ -1,51 +0,0 @@
1−/*
2− * Copyright (c) 2012
3− * MIPS Technologies, Inc., California.
4− *
5− * Redistribution and use in source and binary forms, with or without
6− * modification, are permitted provided that the following conditions
7− * are met:
8− * 1. Redistributions of source code must retain the above copyright
9− * notice, this list of conditions and the following disclaimer.
10− * 2. Redistributions in binary form must reproduce the above copyright
11− * notice, this list of conditions and the following disclaimer in the
12− * documentation and/or other materials provided with the distribution.
13− * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
14− * contributors may be used to endorse or promote products derived from
15− * this software without specific prior written permission.
16− *
17− * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
18− * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19− * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20− * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
21− * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22− * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23− * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24− * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25− * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26− * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27− * SUCH DAMAGE.
28− *
29− * Authors: Stanislav Ocovaj (socovaj@mips.com)
30− * Goran Cordasic (goran@mips.com)
31− * Djordje Pesut (djordje@mips.com)
32− *
33− * This file is part of FFmpeg.
34− *
35− * FFmpeg is free software; you can redistribute it and/or
36− * modify it under the terms of the GNU Lesser General Public
37− * License as published by the Free Software Foundation; either
38− * version 2.1 of the License, or (at your option) any later version.
39− *
40− * FFmpeg is distributed in the hope that it will be useful,
41− * but WITHOUT ANY WARRANTY; without even the implied warranty of
42− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
43− * Lesser General Public License for more details.
44− *
45− * You should have received a copy of the GNU Lesser General Public
46− * License along with FFmpeg; if not, write to the Free Software
47− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
48− */
49−
50−#define FFT_FLOAT 0
51−#include "fft_template.c"
libavcodec/fft_float.c+0−20
@@ -1,20 +0,0 @@
1−/*
2− * This file is part of FFmpeg.
3− *
4− * FFmpeg is free software; you can redistribute it and/or
5− * modify it under the terms of the GNU Lesser General Public
6− * License as published by the Free Software Foundation; either
7− * version 2.1 of the License, or (at your option) any later version.
8− *
9− * FFmpeg is distributed in the hope that it will be useful,
10− * but WITHOUT ANY WARRANTY; without even the implied warranty of
11− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12− * Lesser General Public License for more details.
13− *
14− * You should have received a copy of the GNU Lesser General Public
15− * License along with FFmpeg; if not, write to the Free Software
16− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17− */
18−
19−#define FFT_FLOAT 1
20−#include "fft_template.c"
libavcodec/fft_init_table.c+0−344
@@ -1,344 +0,0 @@
1−/*
2− * Copyright (c) 2012
3− * MIPS Technologies, Inc., California.
4− *
5− * Redistribution and use in source and binary forms, with or without
6− * modification, are permitted provided that the following conditions
7− * are met:
8− * 1. Redistributions of source code must retain the above copyright
9− * notice, this list of conditions and the following disclaimer.
10− * 2. Redistributions in binary form must reproduce the above copyright
11− * notice, this list of conditions and the following disclaimer in the
12− * documentation and/or other materials provided with the distribution.
13− * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
14− * contributors may be used to endorse or promote products derived from
15− * this software without specific prior written permission.
16− *
17− * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
18− * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19− * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20− * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
21− * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22− * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23− * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24− * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25− * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26− * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27− * SUCH DAMAGE.
28− *
29− * Authors: Stanislav Ocovaj (socovaj@mips.com)
30− * Goran Cordasic (goran@mips.com)
31− * Djordje Pesut (djordje@mips.com)
32− *
33− * This file is part of FFmpeg.
34− *
35− * FFmpeg is free software; you can redistribute it and/or
36− * modify it under the terms of the GNU Lesser General Public
37− * License as published by the Free Software Foundation; either
38− * version 2.1 of the License, or (at your option) any later version.
39− *
40− * FFmpeg is distributed in the hope that it will be useful,
41− * but WITHOUT ANY WARRANTY; without even the implied warranty of
42− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
43− * Lesser General Public License for more details.
44− *
45− * You should have received a copy of the GNU Lesser General Public
46− * License along with FFmpeg; if not, write to the Free Software
47− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
48− */
49−
50−/**
51− * @file
52− * definitions and initialization of LUT table for FFT
53− */
54−#include "libavutil/thread.h"
55−
56−#include "libavcodec/fft_table.h"
57−
58−const int32_t ff_w_tab_sr[MAX_FFT_SIZE/(4*16)] = {
59−2147483647, 2147483016, 2147481121, 2147477963, 2147473542, 2147467857, 2147460908, 2147452697,
60−2147443222, 2147432484, 2147420483, 2147407218, 2147392690, 2147376899, 2147359845, 2147341527,
61−2147321946, 2147301102, 2147278995, 2147255625, 2147230991, 2147205094, 2147177934, 2147149511,
62−2147119825, 2147088876, 2147056664, 2147023188, 2146988450, 2146952448, 2146915184, 2146876656,
63−2146836866, 2146795813, 2146753497, 2146709917, 2146665076, 2146618971, 2146571603, 2146522973,
64−2146473080, 2146421924, 2146369505, 2146315824, 2146260881, 2146204674, 2146147205, 2146088474,
65−2146028480, 2145967224, 2145904705, 2145840924, 2145775880, 2145709574, 2145642006, 2145573176,
66−2145503083, 2145431729, 2145359112, 2145285233, 2145210092, 2145133690, 2145056025, 2144977098,
67−2144896910, 2144815460, 2144732748, 2144648774, 2144563539, 2144477042, 2144389283, 2144300264,
68−2144209982, 2144118439, 2144025635, 2143931570, 2143836244, 2143739656, 2143641807, 2143542697,
69−2143442326, 2143340694, 2143237802, 2143133648, 2143028234, 2142921559, 2142813624, 2142704427,
70−2142593971, 2142482254, 2142369276, 2142255039, 2142139541, 2142022783, 2141904764, 2141785486,
71−2141664948, 2141543150, 2141420092, 2141295774, 2141170197, 2141043360, 2140915264, 2140785908,
72−2140655293, 2140523418, 2140390284, 2140255892, 2140120240, 2139983329, 2139845159, 2139705730,
73−2139565043, 2139423097, 2139279892, 2139135429, 2138989708, 2138842728, 2138694490, 2138544994,
74−2138394240, 2138242228, 2138088958, 2137934430, 2137778644, 2137621601, 2137463301, 2137303743,
75−2137142927, 2136980855, 2136817525, 2136652938, 2136487095, 2136319994, 2136151637, 2135982023,
76−2135811153, 2135639026, 2135465642, 2135291003, 2135115107, 2134937956, 2134759548, 2134579885,
77−2134398966, 2134216791, 2134033361, 2133848675, 2133662734, 2133475538, 2133287087, 2133097381,
78−2132906420, 2132714204, 2132520734, 2132326009, 2132130030, 2131932796, 2131734309, 2131534567,
79−2131333572, 2131131322, 2130927819, 2130723062, 2130517052, 2130309789, 2130101272, 2129891502,
80−2129680480, 2129468204, 2129254676, 2129039895, 2128823862, 2128606576, 2128388038, 2128168248,
81−2127947206, 2127724913, 2127501367, 2127276570, 2127050522, 2126823222, 2126594672, 2126364870,
82−2126133817, 2125901514, 2125667960, 2125433155, 2125197100, 2124959795, 2124721240, 2124481435,
83−2124240380, 2123998076, 2123754522, 2123509718, 2123263666, 2123016364, 2122767814, 2122518015,
84−2122266967, 2122014670, 2121761126, 2121506333, 2121250292, 2120993003, 2120734467, 2120474683,
85−2120213651, 2119951372, 2119687847, 2119423074, 2119157054, 2118889788, 2118621275, 2118351516,
86−2118080511, 2117808259, 2117534762, 2117260020, 2116984031, 2116706797, 2116428319, 2116148595,
87−2115867626, 2115585412, 2115301954, 2115017252, 2114731305, 2114444114, 2114155680, 2113866001,
88−2113575080, 2113282914, 2112989506, 2112694855, 2112398960, 2112101824, 2111803444, 2111503822,
89−2111202959, 2110900853, 2110597505, 2110292916, 2109987085, 2109680013, 2109371700, 2109062146,
90−2108751352, 2108439317, 2108126041, 2107811526, 2107495770, 2107178775, 2106860540, 2106541065,
91−2106220352, 2105898399, 2105575208, 2105250778, 2104925109, 2104598202, 2104270057, 2103940674,
92−2103610054, 2103278196, 2102945101, 2102610768, 2102275199, 2101938393, 2101600350, 2101261071,
93−2100920556, 2100578805, 2100235819, 2099891596, 2099546139, 2099199446, 2098851519, 2098502357,
94−2098151960, 2097800329, 2097447464, 2097093365, 2096738032, 2096381466, 2096023667, 2095664635,
95−2095304370, 2094942872, 2094580142, 2094216179, 2093850985, 2093484559, 2093116901, 2092748012,
96−2092377892, 2092006541, 2091633960, 2091260147, 2090885105, 2090508833, 2090131331, 2089752599,
97−2089372638, 2088991448, 2088609029, 2088225381, 2087840505, 2087454400, 2087067068, 2086678508,
98−2086288720, 2085897705, 2085505463, 2085111994, 2084717298, 2084321376, 2083924228, 2083525854,
99−2083126254, 2082725429, 2082323379, 2081920103, 2081515603, 2081109879, 2080702930, 2080294757,
100−2079885360, 2079474740, 2079062896, 2078649830, 2078235540, 2077820028, 2077403294, 2076985338,
101−2076566160, 2076145760, 2075724139, 2075301296, 2074877233, 2074451950, 2074025446, 2073597721,
102−2073168777, 2072738614, 2072307231, 2071874629, 2071440808, 2071005769, 2070569511, 2070132035,
103−2069693342, 2069253430, 2068812302, 2068369957, 2067926394, 2067481616, 2067035621, 2066588410,
104−2066139983, 2065690341, 2065239484, 2064787411, 2064334124, 2063879623, 2063423908, 2062966978,
105−2062508835, 2062049479, 2061588910, 2061127128, 2060664133, 2060199927, 2059734508, 2059267877,
106−2058800036, 2058330983, 2057860719, 2057389244, 2056916560, 2056442665, 2055967560, 2055491246,
107−2055013723, 2054534991, 2054055050, 2053573901, 2053091544, 2052607979, 2052123207, 2051637227,
108−2051150040, 2050661647, 2050172048, 2049681242, 2049189231, 2048696014, 2048201592, 2047705965,
109−2047209133, 2046711097, 2046211857, 2045711414, 2045209767, 2044706916, 2044202863, 2043697608,
110−2043191150, 2042683490, 2042174628, 2041664565, 2041153301, 2040640837, 2040127172, 2039612306,
111−2039096241, 2038578976, 2038060512, 2037540850, 2037019988, 2036497928, 2035974670, 2035450215,
112−2034924562, 2034397712, 2033869665, 2033340422, 2032809982, 2032278347, 2031745516, 2031211490,
113−2030676269, 2030139853, 2029602243, 2029063439, 2028523442, 2027982251, 2027439867, 2026896291,
114−2026351522, 2025805561, 2025258408, 2024710064, 2024160529, 2023609803, 2023057887, 2022504780,
115−2021950484, 2021394998, 2020838323, 2020280460, 2019721407, 2019161167, 2018599739, 2018037123,
116−2017473321, 2016908331, 2016342155, 2015774793, 2015206245, 2014636511, 2014065592, 2013493489,
117−2012920201, 2012345729, 2011770073, 2011193233, 2010615210, 2010036005, 2009455617, 2008874047,
118−2008291295, 2007707362, 2007122248, 2006535953, 2005948478, 2005359822, 2004769987, 2004178973,
119−2003586779, 2002993407, 2002398857, 2001803128, 2001206222, 2000608139, 2000008879, 1999408442,
120−1998806829, 1998204040, 1997600076, 1996994937, 1996388622, 1995781134, 1995172471, 1994562635,
121−1993951625, 1993339442, 1992726087, 1992111559, 1991495860, 1990878989, 1990260946, 1989641733,
122−1989021350, 1988399796, 1987777073, 1987153180, 1986528118, 1985901888, 1985274489, 1984645923,
123−1984016189, 1983385288, 1982753220, 1982119985, 1981485585, 1980850019, 1980213288, 1979575392,
124−1978936331, 1978296106, 1977654717, 1977012165, 1976368450, 1975723572, 1975077532, 1974430331,
125−1973781967, 1973132443, 1972481757, 1971829912, 1971176906, 1970522741, 1969867417, 1969210933,
126−1968553292, 1967894492, 1967234535, 1966573420, 1965911148, 1965247720, 1964583136, 1963917396,
127−1963250501, 1962582451, 1961913246, 1961242888, 1960571375, 1959898709, 1959224890, 1958549919,
128−1957873796, 1957196520, 1956518093, 1955838516, 1955157788, 1954475909, 1953792881, 1953108703,
129−1952423377, 1951736902, 1951049279, 1950360508, 1949670589, 1948979524, 1948287312, 1947593954,
130−1946899451, 1946203802, 1945507008, 1944809070, 1944109987, 1943409761, 1942708392, 1942005880,
131−1941302225, 1940597428, 1939891490, 1939184411, 1938476190, 1937766830, 1937056329, 1936344689,
132−1935631910, 1934917992, 1934202936, 1933486742, 1932769411, 1932050943, 1931331338, 1930610597,
133−1929888720, 1929165708, 1928441561, 1927716279, 1926989864, 1926262315, 1925533633, 1924803818,
134−1924072871, 1923340791, 1922607581, 1921873239, 1921137767, 1920401165, 1919663432, 1918924571,
135−1918184581, 1917443462, 1916701216, 1915957841, 1915213340, 1914467712, 1913720958, 1912973078,
136−1912224073, 1911473942, 1910722688, 1909970309, 1909216806, 1908462181, 1907706433, 1906949562,
137−1906191570, 1905432457, 1904672222, 1903910867, 1903148392, 1902384797, 1901620084, 1900854251,
138−1900087301, 1899319232, 1898550047, 1897779744, 1897008325, 1896235790, 1895462140, 1894687374,
139−1893911494, 1893134500, 1892356392, 1891577171, 1890796837, 1890015391, 1889232832, 1888449163,
140−1887664383, 1886878492, 1886091491, 1885303381, 1884514161, 1883723833, 1882932397, 1882139853,
141−1881346202, 1880551444, 1879755580, 1878958610, 1878160535, 1877361354, 1876561070, 1875759681,
142−1874957189, 1874153594, 1873348897, 1872543097, 1871736196, 1870928194, 1870119091, 1869308888,
143−1868497586, 1867685184, 1866871683, 1866057085, 1865241388, 1864424594, 1863606704, 1862787717,
144−1861967634, 1861146456, 1860324183, 1859500816, 1858676355, 1857850800, 1857024153, 1856196413,
145−1855367581, 1854537657, 1853706643, 1852874538, 1852041343, 1851207059, 1850371686, 1849535224,
146−1848697674, 1847859036, 1847019312, 1846178501, 1845336604, 1844493621, 1843649553, 1842804401,
147−1841958164, 1841110844, 1840262441, 1839412956, 1838562388, 1837710739, 1836858008, 1836004197,
148−1835149306, 1834293336, 1833436286, 1832578158, 1831718951, 1830858668, 1829997307, 1829134869,
149−1828271356, 1827406767, 1826541103, 1825674364, 1824806552, 1823937666, 1823067707, 1822196675,
150−1821324572, 1820451397, 1819577151, 1818701835, 1817825449, 1816947994, 1816069469, 1815189877,
151−1814309216, 1813427489, 1812544694, 1811660833, 1810775906, 1809889915, 1809002858, 1808114737,
152−1807225553, 1806335305, 1805443995, 1804551623, 1803658189, 1802763694, 1801868139, 1800971523,
153−1800073849, 1799175115, 1798275323, 1797374472, 1796472565, 1795569601, 1794665580, 1793760504,
154−1792854372, 1791947186, 1791038946, 1790129652, 1789219305, 1788307905, 1787395453, 1786481950,
155−1785567396, 1784651792, 1783735137, 1782817434, 1781898681, 1780978881, 1780058032, 1779136137,
156−1778213194, 1777289206, 1776364172, 1775438094, 1774510970, 1773582803, 1772653593, 1771723340,
157−1770792044, 1769859707, 1768926328, 1767991909, 1767056450, 1766119952, 1765182414, 1764243838,
158−1763304224, 1762363573, 1761421885, 1760479161, 1759535401, 1758590607, 1757644777, 1756697914,
159−1755750017, 1754801087, 1753851126, 1752900132, 1751948107, 1750995052, 1750040966, 1749085851,
160−1748129707, 1747172535, 1746214334, 1745255107, 1744294853, 1743333573, 1742371267, 1741407936,
161−1740443581, 1739478202, 1738511799, 1737544374, 1736575927, 1735606458, 1734635968, 1733664458,
162−1732691928, 1731718378, 1730743810, 1729768224, 1728791620, 1727813999, 1726835361, 1725855708,
163−1724875040, 1723893357, 1722910659, 1721926948, 1720942225, 1719956488, 1718969740, 1717981981,
164−1716993211, 1716003431, 1715012642, 1714020844, 1713028037, 1712034223, 1711039401, 1710043573,
165−1709046739, 1708048900, 1707050055, 1706050207, 1705049355, 1704047500, 1703044642, 1702040783,
166−1701035922, 1700030061, 1699023199, 1698015339, 1697006479, 1695996621, 1694985765, 1693973912,
167−1692961062, 1691947217, 1690932376, 1689916541, 1688899711, 1687881888, 1686863072, 1685843263,
168−1684822463, 1683800672, 1682777890, 1681754118, 1680729357, 1679703608, 1678676870, 1677649144,
169−1676620432, 1675590733, 1674560049, 1673528379, 1672495725, 1671462087, 1670427466, 1669391862,
170−1668355276, 1667317709, 1666279161, 1665239632, 1664199124, 1663157637, 1662115172, 1661071729,
171−1660027308, 1658981911, 1657935539, 1656888190, 1655839867, 1654790570, 1653740300, 1652689057,
172−1651636841, 1650583654, 1649529496, 1648474367, 1647418269, 1646361202, 1645303166, 1644244162,
173−1643184191, 1642123253, 1641061349, 1639998480, 1638934646, 1637869848, 1636804087, 1635737362,
174−1634669676, 1633601027, 1632531418, 1631460848, 1630389319, 1629316830, 1628243383, 1627168978,
175−1626093616, 1625017297, 1623940023, 1622861793, 1621782608, 1620702469, 1619621377, 1618539332,
176−1617456335, 1616372386, 1615287487, 1614201637, 1613114838, 1612027089, 1610938393, 1609848749,
177−1608758157, 1607666620, 1606574136, 1605480708, 1604386335, 1603291018, 1602194758, 1601097555,
178−1599999411, 1598900325, 1597800299, 1596699333, 1595597428, 1594494583, 1593390801, 1592286082,
179−1591180426, 1590073833, 1588966306, 1587857843, 1586748447, 1585638117, 1584526854, 1583414660,
180−1582301533, 1581187476, 1580072489, 1578956572, 1577839726, 1576721952, 1575603251, 1574483623,
181−1573363068, 1572241588, 1571119183, 1569995854, 1568871601, 1567746425, 1566620327, 1565493307,
182−1564365367, 1563236506, 1562106725, 1560976026, 1559844408, 1558711873, 1557578421, 1556444052,
183−1555308768, 1554172569, 1553035455, 1551897428, 1550758488, 1549618636, 1548477872, 1547336197,
184−1546193612, 1545050118, 1543905714, 1542760402, 1541614183, 1540467057, 1539319024, 1538170087,
185−1537020244, 1535869497, 1534717846, 1533565293, 1532411837, 1531257480, 1530102222, 1528946064,
186−1527789007, 1526631051, 1525472197, 1524312445, 1523151797, 1521990252, 1520827813, 1519664478,
187−1518500250, 1517335128, 1516169114, 1515002208, 1513834411, 1512665723, 1511496145, 1510325678,
188−1509154322, 1507982079, 1506808949, 1505634932, 1504460029, 1503284242, 1502107570, 1500930014,
189−1499751576, 1498572255, 1497392053, 1496210969, 1495029006, 1493846163, 1492662441, 1491477842,
190−1490292364, 1489106011, 1487918781, 1486730675, 1485541696, 1484351842, 1483161115, 1481969516,
191−1480777044, 1479583702, 1478389489, 1477194407, 1475998456, 1474801636, 1473603949, 1472405394,
192−1471205974, 1470005688, 1468804538, 1467602523, 1466399645, 1465195904, 1463991302, 1462785838,
193−1461579514, 1460372329, 1459164286, 1457955385, 1456745625, 1455535009, 1454323536, 1453111208,
194−1451898025, 1450683988, 1449469098, 1448253355, 1447036760, 1445819314, 1444601017, 1443381870,
195−1442161874, 1440941030, 1439719338, 1438496799, 1437273414, 1436049184, 1434824109, 1433598189,
196−1432371426, 1431143821, 1429915374, 1428686085, 1427455956, 1426224988, 1424993180, 1423760534,
197−1422527051, 1421292730, 1420057574, 1418821582, 1417584755, 1416347095, 1415108601, 1413869275,
198−1412629117, 1411388129, 1410146309, 1408903661, 1407660183, 1406415878, 1405170745, 1403924785,
199−1402678000, 1401430389, 1400181954, 1398932695, 1397682613, 1396431709, 1395179984, 1393927438,
200−1392674072, 1391419886, 1390164882, 1388909060, 1387652422, 1386394966, 1385136696, 1383877610,
201−1382617710, 1381356997, 1380095472, 1378833134, 1377569986, 1376306026, 1375041258, 1373775680,
202−1372509294, 1371242101, 1369974101, 1368705296, 1367435685, 1366165269, 1364894050, 1363622028,
203−1362349204, 1361075579, 1359801152, 1358525926, 1357249901, 1355973077, 1354695455, 1353417037,
204−1352137822, 1350857812, 1349577007, 1348295409, 1347013017, 1345729833, 1344445857, 1343161090,
205−1341875533, 1340589187, 1339302052, 1338014129, 1336725419, 1335435923, 1334145641, 1332854574,
206−1331562723, 1330270089, 1328976672, 1327682474, 1326387494, 1325091734, 1323795195, 1322497877,
207−1321199781, 1319900907, 1318601257, 1317300832, 1315999631, 1314697657, 1313394909, 1312091388,
208−1310787095, 1309482032, 1308176198, 1306869594, 1305562222, 1304254082, 1302945174, 1301635500,
209−1300325060, 1299013855, 1297701886, 1296389154, 1295075659, 1293761402, 1292446384, 1291130606,
210−1289814068, 1288496772, 1287178717, 1285859905, 1284540337, 1283220013, 1281898935, 1280577102,
211−1279254516, 1277931177, 1276607086, 1275282245, 1273956653, 1272630312, 1271303222, 1269975384,
212−1268646800, 1267317469, 1265987392, 1264656571, 1263325005, 1261992697, 1260659646, 1259325853,
213−1257991320, 1256656047, 1255320034, 1253983283, 1252645794, 1251307568, 1249968606, 1248628909,
214−1247288478, 1245947312, 1244605414, 1243262783, 1241919421, 1240575329, 1239230506, 1237884955,
215−1236538675, 1235191668, 1233843935, 1232495475, 1231146291, 1229796382, 1228445750, 1227094395,
216−1225742318, 1224389521, 1223036002, 1221681765, 1220326809, 1218971135, 1217614743, 1216257636,
217−1214899813, 1213541275, 1212182024, 1210822059, 1209461382, 1208099993, 1206737894, 1205375085,
218−1204011567, 1202647340, 1201282407, 1199916766, 1198550419, 1197183368, 1195815612, 1194447153,
219−1193077991, 1191708127, 1190337562, 1188966297, 1187594332, 1186221669, 1184848308, 1183474250,
220−1182099496, 1180724046, 1179347902, 1177971064, 1176593533, 1175215310, 1173836395, 1172456790,
221−1171076495, 1169695512, 1168313840, 1166931481, 1165548435, 1164164704, 1162780288, 1161395188,
222−1160009405, 1158622939, 1157235792, 1155847964, 1154459456, 1153070269, 1151680403, 1150289860,
223−1148898640, 1147506745, 1146114174, 1144720929, 1143327011, 1141932420, 1140537158, 1139141224,
224−1137744621, 1136347348, 1134949406, 1133550797, 1132151521, 1130751579, 1129350972, 1127949701,
225−1126547765, 1125145168, 1123741908, 1122337987, 1120933406, 1119528166, 1118122267, 1116715710,
226−1115308496, 1113900627, 1112492101, 1111082922, 1109673089, 1108262603, 1106851465, 1105439676,
227−1104027237, 1102614148, 1101200410, 1099786025, 1098370993, 1096955314, 1095538991, 1094122023,
228−1092704411, 1091286156, 1089867259, 1088447722, 1087027544, 1085606726, 1084185270, 1082763176,
229−1081340445, 1079917078, 1078493076, 1077068439, 1075643169, 1074217266, 1072790730, 1071363564,
230−1069935768, 1068507342, 1067078288, 1065648605, 1064218296, 1062787361, 1061355801, 1059923616,
231−1058490808, 1057057377, 1055623324, 1054188651, 1052753357, 1051317443, 1049880912, 1048443763,
232−1047005996, 1045567615, 1044128617, 1042689006, 1041248781, 1039807944, 1038366495, 1036924436,
233−1035481766, 1034038487, 1032594600, 1031150105, 1029705004, 1028259297, 1026812985, 1025366069,
234−1023918550, 1022470428, 1021021705, 1019572382, 1018122458, 1016671936, 1015220816, 1013769098,
235−1012316784, 1010863875, 1009410370, 1007956272, 1006501581, 1005046298, 1003590424, 1002133959,
236−1000676905, 999219262, 997761031, 996302214, 994842810, 993382821, 991922248, 990461091,
237−988999351, 987537030, 986074127, 984610645, 983146583, 981681943, 980216726, 978750932,
238−977284562, 975817617, 974350098, 972882006, 971413342, 969944106, 968474300, 967003923,
239−965532978, 964061465, 962589385, 961116739, 959643527, 958169751, 956695411, 955220508,
240−953745043, 952269017, 950792431, 949315286, 947837582, 946359321, 944880503, 943401129,
241−941921200, 940440717, 938959681, 937478092, 935995952, 934513261, 933030021, 931546231,
242−930061894, 928577010, 927091579, 925605603, 924119082, 922632018, 921144411, 919656262,
243−918167572, 916678342, 915188572, 913698265, 912207419, 910716038, 909224120, 907731667,
244−906238681, 904745161, 903251110, 901756526, 900261413, 898765769, 897269597, 895772898,
245−894275671, 892777918, 891279640, 889780838, 888281512, 886781663, 885281293, 883780402,
246−882278992, 880777062, 879274614, 877771649, 876268167, 874764170, 873259659, 871754633,
247−870249095, 868743045, 867236484, 865729413, 864221832, 862713743, 861205147, 859696043,
248−858186435, 856676321, 855165703, 853654582, 852142959, 850630835, 849118210, 847605086,
249−846091463, 844577343, 843062726, 841547612, 840032004, 838515901, 836999305, 835482217,
250−833964638, 832446567, 830928007, 829408958, 827889422, 826369398, 824848888, 823327893,
251−821806413, 820284450, 818762005, 817239078, 815715670, 814191782, 812667415, 811142571,
252−809617249, 808091450, 806565177, 805038429, 803511207, 801983513, 800455346, 798926709,
253−797397602, 795868026, 794337982, 792807470, 791276492, 789745049, 788213141, 786680769,
254−785147934, 783614638, 782080880, 780546663, 779011986, 777476851, 775941259, 774405210,
255−772868706, 771331747, 769794334, 768256469, 766718151, 765179382, 763640164, 762100496,
256−760560380, 759019816, 757478806, 755937350, 754395449, 752853105, 751310318, 749767089,
257−748223418, 746679308, 745134758, 743589770, 742044345, 740498483, 738952186, 737405453,
258−735858287, 734310688, 732762657, 731214195, 729665303, 728115982, 726566232, 725016055,
259−723465451, 721914422, 720362968, 718811090, 717258790, 715706067, 714152924, 712599360,
260−711045377, 709490976, 707936158, 706380923, 704825272, 703269207, 701712728, 700155836,
261−698598533, 697040818, 695482694, 693924160, 692365218, 690805869, 689246113, 687685952,
262−686125387, 684564417, 683003045, 681441272, 679879097, 678316522, 676753549, 675190177,
263−673626408, 672062243, 670497682, 668932727, 667367379, 665801638, 664235505, 662668981,
264−661102068, 659534766, 657967075, 656398998, 654830535, 653261686, 651692453, 650122837,
265−648552838, 646982457, 645411696, 643840556, 642269036, 640697139, 639124865, 637552215,
266−635979190, 634405791, 632832018, 631257873, 629683357, 628108471, 626533215, 624957590,
267−623381598, 621805239, 620228514, 618651424, 617073971, 615496154, 613917975, 612339436,
268−610760536, 609181276, 607601658, 606021683, 604441352, 602860664, 601279623, 599698227,
269−598116479, 596534378, 594951927, 593369126, 591785976, 590202477, 588618632, 587034440,
270−585449903, 583865021, 582279796, 580694229, 579108320, 577522070, 575935480, 574348552,
271−572761285, 571173682, 569585743, 567997469, 566408860, 564819919, 563230645, 561641039,
272−560051104, 558460839, 556870245, 555279324, 553688076, 552096502, 550504604, 548912382,
273−547319836, 545726969, 544133781, 542540273, 540946445, 539352300, 537757837, 536163058,
274−534567963, 532972554, 531376831, 529780796, 528184449, 526587791, 524990824, 523393547,
275−521795963, 520198072, 518599875, 517001373, 515402566, 513803457, 512204045, 510604332,
276−509004318, 507404005, 505803394, 504202485, 502601279, 500999778, 499397982, 497795892,
277−496193509, 494590835, 492987869, 491384614, 489781069, 488177236, 486573117, 484968710,
278−483364019, 481759043, 480153784, 478548243, 476942419, 475336316, 473729932, 472123270,
279−470516330, 468909114, 467301622, 465693854, 464085813, 462477499, 460868912, 459260055,
280−457650927, 456041530, 454431865, 452821933, 451211734, 449601270, 447990541, 446379549,
281−444768294, 443156777, 441545000, 439932963, 438320667, 436708113, 435095303, 433482236,
282−431868915, 430255339, 428641511, 427027430, 425413098, 423798515, 422183684, 420568604,
283−418953276, 417337703, 415721883, 414105819, 412489512, 410872962, 409256170, 407639137,
284−406021865, 404404353, 402786604, 401168618, 399550396, 397931939, 396313247, 394694323,
285−393075166, 391455778, 389836160, 388216313, 386596237, 384975934, 383355404, 381734649,
286−380113669, 378492466, 376871039, 375249392, 373627523, 372005435, 370383128, 368760603,
287−367137861, 365514903, 363891730, 362268343, 360644742, 359020930, 357396906, 355772673,
288−354148230, 352523578, 350898719, 349273654, 347648383, 346022908, 344397230, 342771348,
289−341145265, 339518981, 337892498, 336265816, 334638936, 333011859, 331384586, 329757119,
290−328129457, 326501602, 324873555, 323245317, 321616889, 319988272, 318359466, 316730474,
291−315101295, 313471930, 311842381, 310212649, 308582734, 306952638, 305322361, 303691904,
292−302061269, 300430456, 298799466, 297168301, 295536961, 293905447, 292273760, 290641901,
293−289009871, 287377671, 285745302, 284112765, 282480061, 280847190, 279214155, 277580955,
294−275947592, 274314066, 272680379, 271046532, 269412525, 267778360, 266144038, 264509558,
295−262874923, 261240134, 259605191, 257970095, 256334847, 254699448, 253063900, 251428203,
296−249792358, 248156366, 246520228, 244883945, 243247518, 241610947, 239974235, 238337382,
297−236700388, 235063255, 233425984, 231788575, 230151030, 228513350, 226875535, 225237587,
298−223599506, 221961294, 220322951, 218684479, 217045878, 215407149, 213768293, 212129312,
299−210490206, 208850976, 207211624, 205572149, 203932553, 202292838, 200653003, 199013051,
300−197372981, 195732795, 194092495, 192452080, 190811551, 189170911, 187530159, 185889297,
301−184248325, 182607245, 180966058, 179324764, 177683365, 176041861, 174400254, 172758544,
302−171116733, 169474820, 167832808, 166190698, 164548489, 162906184, 161263783, 159621287,
303−157978697, 156336015, 154693240, 153050374, 151407418, 149764374, 148121241, 146478021,
304−144834714, 143191323, 141547847, 139904288, 138260647, 136616925, 134973122, 133329239,
305−131685278, 130041240, 128397125, 126752935, 125108670, 123464332, 121819921, 120175438,
306−118530885, 116886262, 115241570, 113596810, 111951983, 110307091, 108662134, 107017112,
307−105372028, 103726882, 102081675, 100436408, 98791081, 97145697, 95500255, 93854758,
308− 92209205, 90563597, 88917937, 87272224, 85626460, 83980645, 82334782, 80688869,
309− 79042909, 77396903, 75750851, 74104755, 72458615, 70812432, 69166208, 67519943,
310− 65873638, 64227295, 62580914, 60934496, 59288042, 57641553, 55995030, 54348475,
311− 52701887, 51055268, 49408620, 47761942, 46115236, 44468503, 42821744, 41174960,
312− 39528151, 37881320, 36234466, 34587590, 32940695, 31293780, 29646846, 27999895,
313− 26352928, 24705945, 23058947, 21411936, 19764913, 18117878, 16470832, 14823776,
314− 13176712, 11529640, 9882561, 8235476, 6588387, 4941294, 3294197, 1647099
315−};
316−
317−uint16_t ff_fft_offsets_lut[21845];
318−
319−static void fft_lut_init(uint16_t *table, int off, int size, int *index)
320−{
321− if (size < 16) {
322− table[*index] = off >> 2;
323− (*index)++;
324− }
325− else {
326− fft_lut_init(table, off, size >> 1, index);
327− fft_lut_init(table, off + (size >> 1), size >> 2, index);
328− fft_lut_init(table, off + 3 * (size >> 2), size >> 2, index);
329− }
330−}
331−
332−static void fft_lut_init_start(void)
333−{
334− int n = 0;
335−
336− fft_lut_init(ff_fft_offsets_lut, 0, 1 << 17, &n);
337−}
338−
339−void ff_fft_lut_init(void)
340−{
341− static AVOnce init_once = AV_ONCE_INIT;
342−
343− ff_thread_once(&init_once, fft_lut_init_start);
344−}
libavcodec/fft_table.h+0−66
@@ -1,66 +0,0 @@
1−/*
2− * Copyright (c) 2012
3− * MIPS Technologies, Inc., California.
4− *
5− * Redistribution and use in source and binary forms, with or without
6− * modification, are permitted provided that the following conditions
7− * are met:
8− * 1. Redistributions of source code must retain the above copyright
9− * notice, this list of conditions and the following disclaimer.
10− * 2. Redistributions in binary form must reproduce the above copyright
11− * notice, this list of conditions and the following disclaimer in the
12− * documentation and/or other materials provided with the distribution.
13− * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
14− * contributors may be used to endorse or promote products derived from
15− * this software without specific prior written permission.
16− *
17− * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
18− * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19− * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20− * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
21− * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22− * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23− * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24− * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25− * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26− * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27− * SUCH DAMAGE.
28− *
29− * Authors: Stanislav Ocovaj (socovaj@mips.com)
30− * Goran Cordasic (goran@mips.com)
31− * Djordje Pesut (djordje@mips.com)
32− *
33− * This file is part of FFmpeg.
34− *
35− * FFmpeg is free software; you can redistribute it and/or
36− * modify it under the terms of the GNU Lesser General Public
37− * License as published by the Free Software Foundation; either
38− * version 2.1 of the License, or (at your option) any later version.
39− *
40− * FFmpeg is distributed in the hope that it will be useful,
41− * but WITHOUT ANY WARRANTY; without even the implied warranty of
42− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
43− * Lesser General Public License for more details.
44− *
45− * You should have received a copy of the GNU Lesser General Public
46− * License along with FFmpeg; if not, write to the Free Software
47− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
48− */
49−
50−/**
51− * @file
52− * definitions and tables for FFT
53− */
54−#ifndef AVCODEC_FFT_TABLE_H
55−#define AVCODEC_FFT_TABLE_H
56−
57−#include "libavcodec/fft.h"
58−
59−#define MAX_LOG2_NFFT 17 //!< Specifies maximum allowed fft size
60−#define MAX_FFT_SIZE (1 << MAX_LOG2_NFFT)
61−
62−extern const int32_t ff_w_tab_sr[];
63−extern uint16_t ff_fft_offsets_lut[];
64−void ff_fft_lut_init(void);
65−
66−#endif /* AVCODEC_FFT_TABLE_H */
libavcodec/fft_template.c+0−628
@@ -1,628 +0,0 @@
1−/*
2− * FFT/IFFT transforms
3− * Copyright (c) 2008 Loren Merritt
4− * Copyright (c) 2002 Fabrice Bellard
5− * Partly based on libdjbfft by D. J. Bernstein
6− *
7− * This file is part of FFmpeg.
8− *
9− * FFmpeg is free software; you can redistribute it and/or
10− * modify it under the terms of the GNU Lesser General Public
11− * License as published by the Free Software Foundation; either
12− * version 2.1 of the License, or (at your option) any later version.
13− *
14− * FFmpeg is distributed in the hope that it will be useful,
15− * but WITHOUT ANY WARRANTY; without even the implied warranty of
16− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17− * Lesser General Public License for more details.
18− *
19− * You should have received a copy of the GNU Lesser General Public
20− * License along with FFmpeg; if not, write to the Free Software
21− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22− */
23−
24−/**
25− * @file
26− * FFT/IFFT transforms.
27− */
28−
29−#include <stdlib.h>
30−#include <string.h>
31−#include "libavutil/mathematics.h"
32−#include "libavutil/thread.h"
33−#include "fft.h"
34−#include "fft-internal.h"
35−
36−#if !FFT_FLOAT
37−#include "fft_table.h"
38−#else /* !FFT_FLOAT */
39−
40−/* cos(2*pi*x/n) for 0<=x<=n/4, followed by its reverse */
41−#if !CONFIG_HARDCODED_TABLES
42−COSTABLE(16);
43−COSTABLE(32);
44−COSTABLE(64);
45−COSTABLE(128);
46−COSTABLE(256);
47−COSTABLE(512);
48−COSTABLE(1024);
49−COSTABLE(2048);
50−COSTABLE(4096);
51−COSTABLE(8192);
52−COSTABLE(16384);
53−COSTABLE(32768);
54−COSTABLE(65536);
55−COSTABLE(131072);
56−
57−static av_cold void init_ff_cos_tabs(int index)
58−{
59− int i;
60− int m = 1<<index;
61− double freq = 2*M_PI/m;
62− FFTSample *tab = FFT_NAME(ff_cos_tabs)[index];
63− for(i=0; i<=m/4; i++)
64− tab[i] = FIX15(cos(i*freq));
65− for(i=1; i<m/4; i++)
66− tab[m/2-i] = tab[i];
67−}
68−
69−typedef struct CosTabsInitOnce {
70− void (*func)(void);
71− AVOnce control;
72−} CosTabsInitOnce;
73−
74−#define INIT_FF_COS_TABS_FUNC(index, size) \
75−static av_cold void init_ff_cos_tabs_ ## size (void)\
76−{ \
77− init_ff_cos_tabs(index); \
78−}
79−
80−INIT_FF_COS_TABS_FUNC(4, 16)
81−INIT_FF_COS_TABS_FUNC(5, 32)
82−INIT_FF_COS_TABS_FUNC(6, 64)
83−INIT_FF_COS_TABS_FUNC(7, 128)
84−INIT_FF_COS_TABS_FUNC(8, 256)
85−INIT_FF_COS_TABS_FUNC(9, 512)
86−INIT_FF_COS_TABS_FUNC(10, 1024)
87−INIT_FF_COS_TABS_FUNC(11, 2048)
88−INIT_FF_COS_TABS_FUNC(12, 4096)
89−INIT_FF_COS_TABS_FUNC(13, 8192)
90−INIT_FF_COS_TABS_FUNC(14, 16384)
91−INIT_FF_COS_TABS_FUNC(15, 32768)
92−INIT_FF_COS_TABS_FUNC(16, 65536)
93−INIT_FF_COS_TABS_FUNC(17, 131072)
94−
95−static CosTabsInitOnce cos_tabs_init_once[] = {
96− { NULL },
97− { NULL },
98− { NULL },
99− { NULL },
100− { init_ff_cos_tabs_16, AV_ONCE_INIT },
101− { init_ff_cos_tabs_32, AV_ONCE_INIT },
102− { init_ff_cos_tabs_64, AV_ONCE_INIT },
103− { init_ff_cos_tabs_128, AV_ONCE_INIT },
104− { init_ff_cos_tabs_256, AV_ONCE_INIT },
105− { init_ff_cos_tabs_512, AV_ONCE_INIT },
106− { init_ff_cos_tabs_1024, AV_ONCE_INIT },
107− { init_ff_cos_tabs_2048, AV_ONCE_INIT },
108− { init_ff_cos_tabs_4096, AV_ONCE_INIT },
109− { init_ff_cos_tabs_8192, AV_ONCE_INIT },
110− { init_ff_cos_tabs_16384, AV_ONCE_INIT },
111− { init_ff_cos_tabs_32768, AV_ONCE_INIT },
112− { init_ff_cos_tabs_65536, AV_ONCE_INIT },
113− { init_ff_cos_tabs_131072, AV_ONCE_INIT },
114−};
115−
116−av_cold void ff_init_ff_cos_tabs(int index)
117−{
118− ff_thread_once(&cos_tabs_init_once[index].control, cos_tabs_init_once[index].func);
119−}
120−#endif
121−COSTABLE_CONST FFTSample * const FFT_NAME(ff_cos_tabs)[] = {
122− NULL, NULL, NULL, NULL,
123− FFT_NAME(ff_cos_16),
124− FFT_NAME(ff_cos_32),
125− FFT_NAME(ff_cos_64),
126− FFT_NAME(ff_cos_128),
127− FFT_NAME(ff_cos_256),
128− FFT_NAME(ff_cos_512),
129− FFT_NAME(ff_cos_1024),
130− FFT_NAME(ff_cos_2048),
131− FFT_NAME(ff_cos_4096),
132− FFT_NAME(ff_cos_8192),
133− FFT_NAME(ff_cos_16384),
134− FFT_NAME(ff_cos_32768),
135− FFT_NAME(ff_cos_65536),
136− FFT_NAME(ff_cos_131072),
137−};
138−
139−#endif /* FFT_FLOAT */
140−
141−static void fft_permute_c(FFTContext *s, FFTComplex *z);
142−static void fft_calc_c(FFTContext *s, FFTComplex *z);
143−
144−static int split_radix_permutation(int i, int n, int inverse)
145−{
146− int m;
147− if(n <= 2) return i&1;
148− m = n >> 1;
149− if(!(i&m)) return split_radix_permutation(i, m, inverse)*2;
150− m >>= 1;
151− if(inverse == !(i&m)) return split_radix_permutation(i, m, inverse)*4 + 1;
152− else return split_radix_permutation(i, m, inverse)*4 - 1;
153−}
154−
155−
156−static const int avx_tab[] = {
157− 0, 4, 1, 5, 8, 12, 9, 13, 2, 6, 3, 7, 10, 14, 11, 15
158−};
159−
160−static int is_second_half_of_fft32(int i, int n)
161−{
162− if (n <= 32)
163− return i >= 16;
164− else if (i < n/2)
165− return is_second_half_of_fft32(i, n/2);
166− else if (i < 3*n/4)
167− return is_second_half_of_fft32(i - n/2, n/4);
168− else
169− return is_second_half_of_fft32(i - 3*n/4, n/4);
170−}
171−
172−static av_cold void fft_perm_avx(FFTContext *s)
173−{
174− int i;
175− int n = 1 << s->nbits;
176−
177− for (i = 0; i < n; i += 16) {
178− int k;
179− if (is_second_half_of_fft32(i, n)) {
180− for (k = 0; k < 16; k++)
181− s->revtab[-split_radix_permutation(i + k, n, s->inverse) & (n - 1)] =
182− i + avx_tab[k];
183−
184− } else {
185− for (k = 0; k < 16; k++) {
186− int j = i + k;
187− j = (j & ~7) | ((j >> 1) & 3) | ((j << 2) & 4);
188− s->revtab[-split_radix_permutation(i + k, n, s->inverse) & (n - 1)] = j;
189− }
190− }
191− }
192−}
193−
194−av_cold int ff_fft_init(FFTContext *s, int nbits, int inverse)
195−{
196− int i, j, n;
197−
198− s->revtab = NULL;
199− s->revtab32 = NULL;
200−
201− if (nbits < 2 || nbits > 17)
202− goto fail;
203− s->nbits = nbits;
204− n = 1 << nbits;
205−
206− if (nbits <= 16) {
207− s->revtab = av_malloc(n * sizeof(uint16_t));
208− if (!s->revtab)
209− goto fail;
210− } else {
211− s->revtab32 = av_malloc(n * sizeof(uint32_t));
212− if (!s->revtab32)
213− goto fail;
214− }
215− s->tmp_buf = av_malloc(n * sizeof(FFTComplex));
216− if (!s->tmp_buf)
217− goto fail;
218− s->inverse = inverse;
219− s->fft_permutation = FF_FFT_PERM_DEFAULT;
220−
221− s->fft_permute = fft_permute_c;
222− s->fft_calc = fft_calc_c;
223−#if CONFIG_MDCT
224− s->imdct_calc = ff_imdct_calc_c;
225− s->imdct_half = ff_imdct_half_c;
226− s->mdct_calc = ff_mdct_calc_c;
227−#endif
228−
229−#if FFT_FLOAT
230−#if ARCH_AARCH64
231− ff_fft_init_aarch64(s);
232−#elif ARCH_ARM
233− ff_fft_init_arm(s);
234−#elif ARCH_PPC
235− ff_fft_init_ppc(s);
236−#elif ARCH_X86
237− ff_fft_init_x86(s);
238−#endif
239−#if HAVE_MIPSFPU
240− ff_fft_init_mips(s);
241−#endif
242− for(j=4; j<=nbits; j++) {
243− ff_init_ff_cos_tabs(j);
244− }
245−#else /* FFT_FLOAT */
246− ff_fft_lut_init();
247−#endif
248−
249−
250− if (ARCH_X86 && FFT_FLOAT && s->fft_permutation == FF_FFT_PERM_AVX) {
251− fft_perm_avx(s);
252− } else {
253−#define PROCESS_FFT_PERM_SWAP_LSBS(num) do {\
254− for(i = 0; i < n; i++) {\
255− int k;\
256− j = i;\
257− j = (j & ~3) | ((j >> 1) & 1) | ((j << 1) & 2);\
258− k = -split_radix_permutation(i, n, s->inverse) & (n - 1);\
259− s->revtab##num[k] = j;\
260− } \
261−} while(0);
262−
263−#define PROCESS_FFT_PERM_DEFAULT(num) do {\
264− for(i = 0; i < n; i++) {\
265− int k;\
266− j = i;\
267− k = -split_radix_permutation(i, n, s->inverse) & (n - 1);\
268− s->revtab##num[k] = j;\
269− } \
270−} while(0);
271−
272−#define SPLIT_RADIX_PERMUTATION(num) do { \
273− if (s->fft_permutation == FF_FFT_PERM_SWAP_LSBS) {\
274− PROCESS_FFT_PERM_SWAP_LSBS(num) \
275− } else {\
276− PROCESS_FFT_PERM_DEFAULT(num) \
277− }\
278−} while(0);
279−
280− if (s->revtab)
281− SPLIT_RADIX_PERMUTATION()
282− if (s->revtab32)
283− SPLIT_RADIX_PERMUTATION(32)
284−
285−#undef PROCESS_FFT_PERM_DEFAULT
286−#undef PROCESS_FFT_PERM_SWAP_LSBS
287−#undef SPLIT_RADIX_PERMUTATION
288− }
289−
290− return 0;
291− fail:
292− av_freep(&s->revtab);
293− av_freep(&s->revtab32);
294− av_freep(&s->tmp_buf);
295− return -1;
296−}
297−
298−static void fft_permute_c(FFTContext *s, FFTComplex *z)
299−{
300− int j, np;
301− const uint16_t *revtab = s->revtab;
302− const uint32_t *revtab32 = s->revtab32;
303− np = 1 << s->nbits;
304− /* TODO: handle split-radix permute in a more optimal way, probably in-place */
305− if (revtab) {
306− for(j=0;j<np;j++) s->tmp_buf[revtab[j]] = z[j];
307− } else
308− for(j=0;j<np;j++) s->tmp_buf[revtab32[j]] = z[j];
309−
310− memcpy(z, s->tmp_buf, np * sizeof(FFTComplex));
311−}
312−
313−av_cold void ff_fft_end(FFTContext *s)
314−{
315− av_freep(&s->revtab);
316− av_freep(&s->revtab32);
317− av_freep(&s->tmp_buf);
318−}
319−
320−#if !FFT_FLOAT
321−
322−static void fft_calc_c(FFTContext *s, FFTComplex *z) {
323−
324− int nbits, i, n, num_transforms, offset, step;
325− int n4, n2, n34;
326− unsigned tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
327− FFTComplex *tmpz;
328− const int fft_size = (1 << s->nbits);
329− int64_t accu;
330−
331− num_transforms = (0x2aab >> (16 - s->nbits)) | 1;
332−
333− for (n=0; n<num_transforms; n++){
334− offset = ff_fft_offsets_lut[n] << 2;
335− tmpz = z + offset;
336−
337− tmp1 = tmpz[0].re + (unsigned)tmpz[1].re;
338− tmp5 = tmpz[2].re + (unsigned)tmpz[3].re;
339− tmp2 = tmpz[0].im + (unsigned)tmpz[1].im;
340− tmp6 = tmpz[2].im + (unsigned)tmpz[3].im;
341− tmp3 = tmpz[0].re - (unsigned)tmpz[1].re;
342− tmp8 = tmpz[2].im - (unsigned)tmpz[3].im;
343− tmp4 = tmpz[0].im - (unsigned)tmpz[1].im;
344− tmp7 = tmpz[2].re - (unsigned)tmpz[3].re;
345−
346− tmpz[0].re = tmp1 + tmp5;
347− tmpz[2].re = tmp1 - tmp5;
348− tmpz[0].im = tmp2 + tmp6;
349− tmpz[2].im = tmp2 - tmp6;
350− tmpz[1].re = tmp3 + tmp8;
351− tmpz[3].re = tmp3 - tmp8;
352− tmpz[1].im = tmp4 - tmp7;
353− tmpz[3].im = tmp4 + tmp7;
354− }
355−
356− if (fft_size < 8)
357− return;
358−
359− num_transforms = (num_transforms >> 1) | 1;
360−
361− for (n=0; n<num_transforms; n++){
362− offset = ff_fft_offsets_lut[n] << 3;
363− tmpz = z + offset;
364−
365− tmp1 = tmpz[4].re + (unsigned)tmpz[5].re;
366− tmp3 = tmpz[6].re + (unsigned)tmpz[7].re;
367− tmp2 = tmpz[4].im + (unsigned)tmpz[5].im;
368− tmp4 = tmpz[6].im + (unsigned)tmpz[7].im;
369− tmp5 = tmp1 + tmp3;
370− tmp7 = tmp1 - tmp3;
371− tmp6 = tmp2 + tmp4;
372− tmp8 = tmp2 - tmp4;
373−
374− tmp1 = tmpz[4].re - (unsigned)tmpz[5].re;
375− tmp2 = tmpz[4].im - (unsigned)tmpz[5].im;
376− tmp3 = tmpz[6].re - (unsigned)tmpz[7].re;
377− tmp4 = tmpz[6].im - (unsigned)tmpz[7].im;
378−
379− tmpz[4].re = tmpz[0].re - tmp5;
380− tmpz[0].re = tmpz[0].re + tmp5;
381− tmpz[4].im = tmpz[0].im - tmp6;
382− tmpz[0].im = tmpz[0].im + tmp6;
383− tmpz[6].re = tmpz[2].re - tmp8;
384− tmpz[2].re = tmpz[2].re + tmp8;
385− tmpz[6].im = tmpz[2].im + tmp7;
386− tmpz[2].im = tmpz[2].im - tmp7;
387−
388− accu = (int64_t)Q31(M_SQRT1_2)*(int)(tmp1 + tmp2);
389− tmp5 = (int32_t)((accu + 0x40000000) >> 31);
390− accu = (int64_t)Q31(M_SQRT1_2)*(int)(tmp3 - tmp4);
391− tmp7 = (int32_t)((accu + 0x40000000) >> 31);
392− accu = (int64_t)Q31(M_SQRT1_2)*(int)(tmp2 - tmp1);
393− tmp6 = (int32_t)((accu + 0x40000000) >> 31);
394− accu = (int64_t)Q31(M_SQRT1_2)*(int)(tmp3 + tmp4);
395− tmp8 = (int32_t)((accu + 0x40000000) >> 31);
396− tmp1 = tmp5 + tmp7;
397− tmp3 = tmp5 - tmp7;
398− tmp2 = tmp6 + tmp8;
399− tmp4 = tmp6 - tmp8;
400−
401− tmpz[5].re = tmpz[1].re - tmp1;
402− tmpz[1].re = tmpz[1].re + tmp1;
403− tmpz[5].im = tmpz[1].im - tmp2;
404− tmpz[1].im = tmpz[1].im + tmp2;
405− tmpz[7].re = tmpz[3].re - tmp4;
406− tmpz[3].re = tmpz[3].re + tmp4;
407− tmpz[7].im = tmpz[3].im + tmp3;
408− tmpz[3].im = tmpz[3].im - tmp3;
409− }
410−
411− step = 1 << ((MAX_LOG2_NFFT-4) - 4);
412− n4 = 4;
413−
414− for (nbits=4; nbits<=s->nbits; nbits++){
415− n2 = 2*n4;
416− n34 = 3*n4;
417− num_transforms = (num_transforms >> 1) | 1;
418−
419− for (n=0; n<num_transforms; n++){
420− const FFTSample *w_re_ptr = ff_w_tab_sr + step;
421− const FFTSample *w_im_ptr = ff_w_tab_sr + MAX_FFT_SIZE/(4*16) - step;
422− offset = ff_fft_offsets_lut[n] << nbits;
423− tmpz = z + offset;
424−
425− tmp5 = tmpz[ n2].re + (unsigned)tmpz[n34].re;
426− tmp1 = tmpz[ n2].re - (unsigned)tmpz[n34].re;
427− tmp6 = tmpz[ n2].im + (unsigned)tmpz[n34].im;
428− tmp2 = tmpz[ n2].im - (unsigned)tmpz[n34].im;
429−
430− tmpz[ n2].re = tmpz[ 0].re - tmp5;
431− tmpz[ 0].re = tmpz[ 0].re + tmp5;
432− tmpz[ n2].im = tmpz[ 0].im - tmp6;
433− tmpz[ 0].im = tmpz[ 0].im + tmp6;
434− tmpz[n34].re = tmpz[n4].re - tmp2;
435− tmpz[ n4].re = tmpz[n4].re + tmp2;
436− tmpz[n34].im = tmpz[n4].im + tmp1;
437− tmpz[ n4].im = tmpz[n4].im - tmp1;
438−
439− for (i=1; i<n4; i++){
440− FFTSample w_re = w_re_ptr[0];
441− FFTSample w_im = w_im_ptr[0];
442− accu = (int64_t)w_re*tmpz[ n2+i].re;
443− accu += (int64_t)w_im*tmpz[ n2+i].im;
444− tmp1 = (int32_t)((accu + 0x40000000) >> 31);
445− accu = (int64_t)w_re*tmpz[ n2+i].im;
446− accu -= (int64_t)w_im*tmpz[ n2+i].re;
447− tmp2 = (int32_t)((accu + 0x40000000) >> 31);
448− accu = (int64_t)w_re*tmpz[n34+i].re;
449− accu -= (int64_t)w_im*tmpz[n34+i].im;
450− tmp3 = (int32_t)((accu + 0x40000000) >> 31);
451− accu = (int64_t)w_re*tmpz[n34+i].im;
452− accu += (int64_t)w_im*tmpz[n34+i].re;
453− tmp4 = (int32_t)((accu + 0x40000000) >> 31);
454−
455− tmp5 = tmp1 + tmp3;
456− tmp1 = tmp1 - tmp3;
457− tmp6 = tmp2 + tmp4;
458− tmp2 = tmp2 - tmp4;
459−
460− tmpz[ n2+i].re = tmpz[ i].re - tmp5;
461− tmpz[ i].re = tmpz[ i].re + tmp5;
462− tmpz[ n2+i].im = tmpz[ i].im - tmp6;
463− tmpz[ i].im = tmpz[ i].im + tmp6;
464− tmpz[n34+i].re = tmpz[n4+i].re - tmp2;
465− tmpz[ n4+i].re = tmpz[n4+i].re + tmp2;
466− tmpz[n34+i].im = tmpz[n4+i].im + tmp1;
467− tmpz[ n4+i].im = tmpz[n4+i].im - tmp1;
468−
469− w_re_ptr += step;
470− w_im_ptr -= step;
471− }
472− }
473− step >>= 1;
474− n4 <<= 1;
475− }
476−}
477−
478−#else /* !FFT_FLOAT */
479−
480−#define BUTTERFLIES(a0,a1,a2,a3) {\
481− BF(t3, t5, t5, t1);\
482− BF(a2.re, a0.re, a0.re, t5);\
483− BF(a3.im, a1.im, a1.im, t3);\
484− BF(t4, t6, t2, t6);\
485− BF(a3.re, a1.re, a1.re, t4);\
486− BF(a2.im, a0.im, a0.im, t6);\
487−}
488−
489−// force loading all the inputs before storing any.
490−// this is slightly slower for small data, but avoids store->load aliasing
491−// for addresses separated by large powers of 2.
492−#define BUTTERFLIES_BIG(a0,a1,a2,a3) {\
493− FFTSample r0=a0.re, i0=a0.im, r1=a1.re, i1=a1.im;\
494− BF(t3, t5, t5, t1);\
495− BF(a2.re, a0.re, r0, t5);\
496− BF(a3.im, a1.im, i1, t3);\
497− BF(t4, t6, t2, t6);\
498− BF(a3.re, a1.re, r1, t4);\
499− BF(a2.im, a0.im, i0, t6);\
500−}
501−
502−#define TRANSFORM(a0,a1,a2,a3,wre,wim) {\
503− CMUL(t1, t2, a2.re, a2.im, wre, -wim);\
504− CMUL(t5, t6, a3.re, a3.im, wre, wim);\
505− BUTTERFLIES(a0,a1,a2,a3)\
506−}
507−
508−#define TRANSFORM_ZERO(a0,a1,a2,a3) {\
509− t1 = a2.re;\
510− t2 = a2.im;\
511− t5 = a3.re;\
512− t6 = a3.im;\
513− BUTTERFLIES(a0,a1,a2,a3)\
514−}
515−
516−/* z[0...8n-1], w[1...2n-1] */
517−#define PASS(name)\
518−static void name(FFTComplex *z, const FFTSample *wre, unsigned int n)\
519−{\
520− FFTDouble t1, t2, t3, t4, t5, t6;\
521− int o1 = 2*n;\
522− int o2 = 4*n;\
523− int o3 = 6*n;\
524− const FFTSample *wim = wre+o1;\
525− n--;\
526−\
527− TRANSFORM_ZERO(z[0],z[o1],z[o2],z[o3]);\
528− TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\
529− do {\
530− z += 2;\
531− wre += 2;\
532− wim -= 2;\
533− TRANSFORM(z[0],z[o1],z[o2],z[o3],wre[0],wim[0]);\
534− TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\
535− } while(--n);\
536−}
537−
538−PASS(pass)
539−#if !CONFIG_SMALL
540−#undef BUTTERFLIES
541−#define BUTTERFLIES BUTTERFLIES_BIG
542−PASS(pass_big)
543−#endif
544−
545−#define DECL_FFT(n,n2,n4)\
546−static void fft##n(FFTComplex *z)\
547−{\
548− fft##n2(z);\
549− fft##n4(z+n4*2);\
550− fft##n4(z+n4*3);\
551− pass(z,FFT_NAME(ff_cos_##n),n4/2);\
552−}
553−
554−static void fft4(FFTComplex *z)
555−{
556− FFTDouble t1, t2, t3, t4, t5, t6, t7, t8;
557−
558− BF(t3, t1, z[0].re, z[1].re);
559− BF(t8, t6, z[3].re, z[2].re);
560− BF(z[2].re, z[0].re, t1, t6);
561− BF(t4, t2, z[0].im, z[1].im);
562− BF(t7, t5, z[2].im, z[3].im);
563− BF(z[3].im, z[1].im, t4, t8);
564− BF(z[3].re, z[1].re, t3, t7);
565− BF(z[2].im, z[0].im, t2, t5);
566−}
567−
568−static void fft8(FFTComplex *z)
569−{
570− FFTDouble t1, t2, t3, t4, t5, t6;
571−
572− fft4(z);
573−
574− BF(t1, z[5].re, z[4].re, -z[5].re);
575− BF(t2, z[5].im, z[4].im, -z[5].im);
576− BF(t5, z[7].re, z[6].re, -z[7].re);
577− BF(t6, z[7].im, z[6].im, -z[7].im);
578−
579− BUTTERFLIES(z[0],z[2],z[4],z[6]);
580− TRANSFORM(z[1],z[3],z[5],z[7],sqrthalf,sqrthalf);
581−}
582−
583−#if !CONFIG_SMALL
584−static void fft16(FFTComplex *z)
585−{
586− FFTDouble t1, t2, t3, t4, t5, t6;
587− FFTSample cos_16_1 = FFT_NAME(ff_cos_16)[1];
588− FFTSample cos_16_3 = FFT_NAME(ff_cos_16)[3];
589−
590− fft8(z);
591− fft4(z+8);
592− fft4(z+12);
593−
594− TRANSFORM_ZERO(z[0],z[4],z[8],z[12]);
595− TRANSFORM(z[2],z[6],z[10],z[14],sqrthalf,sqrthalf);
596− TRANSFORM(z[1],z[5],z[9],z[13],cos_16_1,cos_16_3);
597− TRANSFORM(z[3],z[7],z[11],z[15],cos_16_3,cos_16_1);
598−}
599−#else
600−DECL_FFT(16,8,4)
601−#endif
602−DECL_FFT(32,16,8)
603−DECL_FFT(64,32,16)
604−DECL_FFT(128,64,32)
605−DECL_FFT(256,128,64)
606−DECL_FFT(512,256,128)
607−#if !CONFIG_SMALL
608−#define pass pass_big
609−#endif
610−DECL_FFT(1024,512,256)
611−DECL_FFT(2048,1024,512)
612−DECL_FFT(4096,2048,1024)
613−DECL_FFT(8192,4096,2048)
614−DECL_FFT(16384,8192,4096)
615−DECL_FFT(32768,16384,8192)
616−DECL_FFT(65536,32768,16384)
617−DECL_FFT(131072,65536,32768)
618−
619−static void (* const fft_dispatch[])(FFTComplex*) = {
620− fft4, fft8, fft16, fft32, fft64, fft128, fft256, fft512, fft1024,
621− fft2048, fft4096, fft8192, fft16384, fft32768, fft65536, fft131072
622−};
623−
624−static void fft_calc_c(FFTContext *s, FFTComplex *z)
625−{
626− fft_dispatch[s->nbits-2](z);
627−}
628−#endif /* !FFT_FLOAT */
libavcodec/mdct_fixed_32.c+0−51
@@ -1,51 +0,0 @@
1−/*
2− * Copyright (c) 2012
3− * MIPS Technologies, Inc., California.
4− *
5− * Redistribution and use in source and binary forms, with or without
6− * modification, are permitted provided that the following conditions
7− * are met:
8− * 1. Redistributions of source code must retain the above copyright
9− * notice, this list of conditions and the following disclaimer.
10− * 2. Redistributions in binary form must reproduce the above copyright
11− * notice, this list of conditions and the following disclaimer in the
12− * documentation and/or other materials provided with the distribution.
13− * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
14− * contributors may be used to endorse or promote products derived from
15− * this software without specific prior written permission.
16− *
17− * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
18− * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19− * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20− * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
21− * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22− * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23− * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24− * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25− * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26− * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27− * SUCH DAMAGE.
28− *
29− * Authors: Stanislav Ocovaj (socovaj@mips.com)
30− * Goran Cordasic (goran@mips.com)
31− * Djordje Pesut (djordje@mips.com)
32− *
33− * This file is part of FFmpeg.
34− *
35− * FFmpeg is free software; you can redistribute it and/or
36− * modify it under the terms of the GNU Lesser General Public
37− * License as published by the Free Software Foundation; either
38− * version 2.1 of the License, or (at your option) any later version.
39− *
40− * FFmpeg is distributed in the hope that it will be useful,
41− * but WITHOUT ANY WARRANTY; without even the implied warranty of
42− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
43− * Lesser General Public License for more details.
44− *
45− * You should have received a copy of the GNU Lesser General Public
46− * License along with FFmpeg; if not, write to the Free Software
47− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
48− */
49−
50−#define FFT_FLOAT 0
51−#include "mdct_template.c"
libavcodec/mdct_float.c+0−20
@@ -1,20 +0,0 @@
1−/*
2− * This file is part of FFmpeg.
3− *
4− * FFmpeg is free software; you can redistribute it and/or
5− * modify it under the terms of the GNU Lesser General Public
6− * License as published by the Free Software Foundation; either
7− * version 2.1 of the License, or (at your option) any later version.
8− *
9− * FFmpeg is distributed in the hope that it will be useful,
10− * but WITHOUT ANY WARRANTY; without even the implied warranty of
11− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12− * Lesser General Public License for more details.
13− *
14− * You should have received a copy of the GNU Lesser General Public
15− * License along with FFmpeg; if not, write to the Free Software
16− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17− */
18−
19−#define FFT_FLOAT 1
20−#include "mdct_template.c"
libavcodec/mdct_template.c+0−209
@@ -1,209 +0,0 @@
1−/*
2− * MDCT/IMDCT transforms
3− * Copyright (c) 2002 Fabrice Bellard
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 <stdlib.h>
23−#include <string.h>
24−#include "libavutil/common.h"
25−#include "libavutil/libm.h"
26−#include "libavutil/mathematics.h"
27−#include "fft.h"
28−#include "fft-internal.h"
29−
30−/**
31− * @file
32− * MDCT/IMDCT transforms.
33− */
34−
35−#if FFT_FLOAT
36−# define RSCALE(x, y) ((x) + (y))
37−#else
38−# define RSCALE(x, y) ((int)((x) + (unsigned)(y) + 32) >> 6)
39−#endif
40−
41−/**
42− * init MDCT or IMDCT computation.
43− */
44−av_cold int ff_mdct_init(FFTContext *s, int nbits, int inverse, double scale)
45−{
46− int n, n4, i;
47− double alpha, theta;
48− int tstep;
49−
50− memset(s, 0, sizeof(*s));
51− n = 1 << nbits;
52− s->mdct_bits = nbits;
53− s->mdct_size = n;
54− n4 = n >> 2;
55− s->mdct_permutation = FF_MDCT_PERM_NONE;
56−
57− if (ff_fft_init(s, s->mdct_bits - 2, inverse) < 0)
58− goto fail;
59−
60− s->tcos = av_malloc_array(n/2, sizeof(FFTSample));
61− if (!s->tcos)
62− goto fail;
63−
64− switch (s->mdct_permutation) {
65− case FF_MDCT_PERM_NONE:
66− s->tsin = s->tcos + n4;
67− tstep = 1;
68− break;
69− case FF_MDCT_PERM_INTERLEAVE:
70− s->tsin = s->tcos + 1;
71− tstep = 2;
72− break;
73− default:
74− goto fail;
75− }
76−
77− theta = 1.0 / 8.0 + (scale < 0 ? n4 : 0);
78− scale = sqrt(fabs(scale));
79− for(i=0;i<n4;i++) {
80− alpha = 2 * M_PI * (i + theta) / n;
81−#if !FFT_FLOAT
82− s->tcos[i*tstep] = lrint(-cos(alpha) * 2147483648.0);
83− s->tsin[i*tstep] = lrint(-sin(alpha) * 2147483648.0);
84−#else
85− s->tcos[i*tstep] = FIX15(-cos(alpha) * scale);
86− s->tsin[i*tstep] = FIX15(-sin(alpha) * scale);
87−#endif
88− }
89− return 0;
90− fail:
91− ff_mdct_end(s);
92− return -1;
93−}
94−
95−/**
96− * Compute the middle half of the inverse MDCT of size N = 2^nbits,
97− * thus excluding the parts that can be derived by symmetry
98− * @param output N/2 samples
99− * @param input N/2 samples
100− */
101−void ff_imdct_half_c(FFTContext *s, FFTSample *output, const FFTSample *input)
102−{
103− int k, n8, n4, n2, n, j;
104− const uint16_t *revtab = s->revtab;
105− const FFTSample *tcos = s->tcos;
106− const FFTSample *tsin = s->tsin;
107− const FFTSample *in1, *in2;
108− FFTComplex *z = (FFTComplex *)output;
109−
110− n = 1 << s->mdct_bits;
111− n2 = n >> 1;
112− n4 = n >> 2;
113− n8 = n >> 3;
114−
115− /* pre rotation */
116− in1 = input;
117− in2 = input + n2 - 1;
118− for(k = 0; k < n4; k++) {
119− j=revtab[k];
120− CMUL(z[j].re, z[j].im, *in2, *in1, tcos[k], tsin[k]);
121− in1 += 2;
122− in2 -= 2;
123− }
124− s->fft_calc(s, z);
125−
126− /* post rotation + reordering */
127− for(k = 0; k < n8; k++) {
128− FFTSample r0, i0, r1, i1;
129− CMUL(r0, i1, z[n8-k-1].im, z[n8-k-1].re, tsin[n8-k-1], tcos[n8-k-1]);
130− CMUL(r1, i0, z[n8+k ].im, z[n8+k ].re, tsin[n8+k ], tcos[n8+k ]);
131− z[n8-k-1].re = r0;
132− z[n8-k-1].im = i0;
133− z[n8+k ].re = r1;
134− z[n8+k ].im = i1;
135− }
136−}
137−
138−/**
139− * Compute inverse MDCT of size N = 2^nbits
140− * @param output N samples
141− * @param input N/2 samples
142− */
143−void ff_imdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input)
144−{
145− int k;
146− int n = 1 << s->mdct_bits;
147− int n2 = n >> 1;
148− int n4 = n >> 2;
149−
150− ff_imdct_half_c(s, output+n4, input);
151−
152− for(k = 0; k < n4; k++) {
153− output[k] = -output[n2-k-1];
154− output[n-k-1] = output[n2+k];
155− }
156−}
157−
158−/**
159− * Compute MDCT of size N = 2^nbits
160− * @param input N samples
161− * @param out N/2 samples
162− */
163−void ff_mdct_calc_c(FFTContext *s, FFTSample *out, const FFTSample *input)
164−{
165− int i, j, n, n8, n4, n2, n3;
166− FFTDouble re, im;
167− const uint16_t *revtab = s->revtab;
168− const FFTSample *tcos = s->tcos;
169− const FFTSample *tsin = s->tsin;
170− FFTComplex *x = (FFTComplex *)out;
171−
172− n = 1 << s->mdct_bits;
173− n2 = n >> 1;
174− n4 = n >> 2;
175− n8 = n >> 3;
176− n3 = 3 * n4;
177−
178− /* pre rotation */
179− for(i=0;i<n8;i++) {
180− re = RSCALE(-input[2*i+n3], - input[n3-1-2*i]);
181− im = RSCALE(-input[n4+2*i], + input[n4-1-2*i]);
182− j = revtab[i];
183− CMUL(x[j].re, x[j].im, re, im, -tcos[i], tsin[i]);
184−
185− re = RSCALE( input[2*i] , - input[n2-1-2*i]);
186− im = RSCALE(-input[n2+2*i], - input[ n-1-2*i]);
187− j = revtab[n8 + i];
188− CMUL(x[j].re, x[j].im, re, im, -tcos[n8 + i], tsin[n8 + i]);
189− }
190−
191− s->fft_calc(s, x);
192−
193− /* post rotation */
194− for(i=0;i<n8;i++) {
195− FFTSample r0, i0, r1, i1;
196− CMUL(i1, r0, x[n8-i-1].re, x[n8-i-1].im, -tsin[n8-i-1], -tcos[n8-i-1]);
197− CMUL(i0, r1, x[n8+i ].re, x[n8+i ].im, -tsin[n8+i ], -tcos[n8+i ]);
198− x[n8-i-1].re = r0;
199− x[n8-i-1].im = i0;
200− x[n8+i ].re = r1;
201− x[n8+i ].im = i1;
202− }
203−}
204−
205−av_cold void ff_mdct_end(FFTContext *s)
206−{
207− av_freep(&s->tcos);
208− ff_fft_end(s);
209−}
libavcodec/mips/Makefile+0−1
@@ -13,7 +13,6 @@ MIPSFPU-OBJS-$(CONFIG_AMRWB_DECODER) += mips/acelp_filters_mips.o \ · 12 unmodified lines
1313 mips/acelp_vectors_mips.o
1414 MIPSFPU-OBJS-$(CONFIG_MPEGAUDIODSP) += mips/mpegaudiodsp_mips_float.o
1515 MIPSDSP-OBJS-$(CONFIG_MPEGAUDIODSP) += mips/mpegaudiodsp_mips_fixed.o
16−MIPSFPU-OBJS-$(CONFIG_FFT) += mips/fft_mips.o
1716 MIPSFPU-OBJS-$(CONFIG_FMTCONVERT) += mips/fmtconvert_mips.o
1817 OBJS-$(CONFIG_AC3DSP) += mips/ac3dsp_mips.o
1918 OBJS-$(CONFIG_AAC_DECODER) += mips/aacdec_mips.o \
libavcodec/mips/fft_mips.c+0−516
@@ -1,516 +0,0 @@
1−/*
2− * Copyright (c) 2012
3− * MIPS Technologies, Inc., California.
4− *
5− * Redistribution and use in source and binary forms, with or without
6− * modification, are permitted provided that the following conditions
7− * are met:
8− * 1. Redistributions of source code must retain the above copyright
9− * notice, this list of conditions and the following disclaimer.
10− * 2. Redistributions in binary form must reproduce the above copyright
11− * notice, this list of conditions and the following disclaimer in the
12− * documentation and/or other materials provided with the distribution.
13− * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
14− * contributors may be used to endorse or promote products derived from
15− * this software without specific prior written permission.
16− *
17− * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
18− * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19− * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20− * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
21− * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22− * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23− * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24− * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25− * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26− * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27− * SUCH DAMAGE.
28− *
29− * Author: Stanislav Ocovaj (socovaj@mips.com)
30− * Author: Zoran Lukic (zoranl@mips.com)
31− *
32− * Optimized MDCT/IMDCT and FFT transforms
33− *
34− * This file is part of FFmpeg.
35− *
36− * FFmpeg is free software; you can redistribute it and/or
37− * modify it under the terms of the GNU Lesser General Public
38− * License as published by the Free Software Foundation; either
39− * version 2.1 of the License, or (at your option) any later version.
40− *
41− * FFmpeg is distributed in the hope that it will be useful,
42− * but WITHOUT ANY WARRANTY; without even the implied warranty of
43− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
44− * Lesser General Public License for more details.
45− *
46− * You should have received a copy of the GNU Lesser General Public
47− * License along with FFmpeg; if not, write to the Free Software
48− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
49− */
50−#include "config.h"
51−#include "libavutil/attributes.h"
52−#include "libavcodec/fft.h"
53−#include "libavcodec/fft_table.h"
54−#include "libavutil/mips/asmdefs.h"
55−
56−/**
57− * FFT transform
58− */
59−
60−#if HAVE_INLINE_ASM
61−#if !HAVE_MIPS32R6 && !HAVE_MIPS64R6
62−static void ff_fft_calc_mips(FFTContext *s, FFTComplex *z)
63−{
64− int nbits, i, n, num_transforms, offset, step;
65− int n4, n2, n34;
66− FFTSample tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
67− FFTComplex *tmpz;
68− float w_re, w_im;
69− float *w_re_ptr, *w_im_ptr;
70− const int fft_size = (1 << s->nbits);
71− float pom, pom1, pom2, pom3;
72− float temp, temp1, temp3, temp4;
73− FFTComplex * tmpz_n2, * tmpz_n34, * tmpz_n4;
74− FFTComplex * tmpz_n2_i, * tmpz_n34_i, * tmpz_n4_i, * tmpz_i;
75− float f1 = 0.7071067812;
76−
77− num_transforms = (21845 >> (17 - s->nbits)) | 1;
78−
79− for (n=0; n<num_transforms; n++) {
80− offset = ff_fft_offsets_lut[n] << 2;
81− tmpz = z + offset;
82−
83− tmp1 = tmpz[0].re + tmpz[1].re;
84− tmp5 = tmpz[2].re + tmpz[3].re;
85− tmp2 = tmpz[0].im + tmpz[1].im;
86− tmp6 = tmpz[2].im + tmpz[3].im;
87− tmp3 = tmpz[0].re - tmpz[1].re;
88− tmp8 = tmpz[2].im - tmpz[3].im;
89− tmp4 = tmpz[0].im - tmpz[1].im;
90− tmp7 = tmpz[2].re - tmpz[3].re;
91−
92− tmpz[0].re = tmp1 + tmp5;
93− tmpz[2].re = tmp1 - tmp5;
94− tmpz[0].im = tmp2 + tmp6;
95− tmpz[2].im = tmp2 - tmp6;
96− tmpz[1].re = tmp3 + tmp8;
97− tmpz[3].re = tmp3 - tmp8;
98− tmpz[1].im = tmp4 - tmp7;
99− tmpz[3].im = tmp4 + tmp7;
100−
101− }
102−
103− if (fft_size < 8)
104− return;
105−
106− num_transforms = (num_transforms >> 1) | 1;
107−
108− for (n=0; n<num_transforms; n++) {
109− offset = ff_fft_offsets_lut[n] << 3;
110− tmpz = z + offset;
111−
112− __asm__ volatile (
113− "lwc1 %[tmp1], 32(%[tmpz]) \n\t"
114− "lwc1 %[pom], 40(%[tmpz]) \n\t"
115− "lwc1 %[tmp3], 48(%[tmpz]) \n\t"
116− "lwc1 %[pom1], 56(%[tmpz]) \n\t"
117− "lwc1 %[tmp2], 36(%[tmpz]) \n\t"
118− "lwc1 %[pom2], 44(%[tmpz]) \n\t"
119− "lwc1 %[pom3], 60(%[tmpz]) \n\t"
120− "lwc1 %[tmp4], 52(%[tmpz]) \n\t"
121− "add.s %[tmp1], %[tmp1], %[pom] \n\t" // tmp1 = tmpz[4].re + tmpz[5].re;
122− "add.s %[tmp3], %[tmp3], %[pom1] \n\t" // tmp3 = tmpz[6].re + tmpz[7].re;
123− "add.s %[tmp2], %[tmp2], %[pom2] \n\t" // tmp2 = tmpz[4].im + tmpz[5].im;
124− "lwc1 %[pom], 40(%[tmpz]) \n\t"
125− "add.s %[tmp4], %[tmp4], %[pom3] \n\t" // tmp4 = tmpz[6].im + tmpz[7].im;
126− "add.s %[tmp5], %[tmp1], %[tmp3] \n\t" // tmp5 = tmp1 + tmp3;
127− "sub.s %[tmp7], %[tmp1], %[tmp3] \n\t" // tmp7 = tmp1 - tmp3;
128− "lwc1 %[tmp1], 32(%[tmpz]) \n\t"
129− "lwc1 %[pom1], 44(%[tmpz]) \n\t"
130− "add.s %[tmp6], %[tmp2], %[tmp4] \n\t" // tmp6 = tmp2 + tmp4;
131− "sub.s %[tmp8], %[tmp2], %[tmp4] \n\t" // tmp8 = tmp2 - tmp4;
132− "lwc1 %[tmp2], 36(%[tmpz]) \n\t"
133− "lwc1 %[pom2], 56(%[tmpz]) \n\t"
134− "lwc1 %[pom3], 60(%[tmpz]) \n\t"
135− "lwc1 %[tmp3], 48(%[tmpz]) \n\t"
136− "lwc1 %[tmp4], 52(%[tmpz]) \n\t"
137− "sub.s %[tmp1], %[tmp1], %[pom] \n\t" // tmp1 = tmpz[4].re - tmpz[5].re;
138− "lwc1 %[pom], 0(%[tmpz]) \n\t"
139− "sub.s %[tmp2], %[tmp2], %[pom1] \n\t" // tmp2 = tmpz[4].im - tmpz[5].im;
140− "sub.s %[tmp3], %[tmp3], %[pom2] \n\t" // tmp3 = tmpz[6].re - tmpz[7].re;
141− "lwc1 %[pom2], 4(%[tmpz]) \n\t"
142− "sub.s %[pom1], %[pom], %[tmp5] \n\t"
143− "sub.s %[tmp4], %[tmp4], %[pom3] \n\t" // tmp4 = tmpz[6].im - tmpz[7].im;
144− "add.s %[pom3], %[pom], %[tmp5] \n\t"
145− "sub.s %[pom], %[pom2], %[tmp6] \n\t"
146− "add.s %[pom2], %[pom2], %[tmp6] \n\t"
147− "swc1 %[pom1], 32(%[tmpz]) \n\t" // tmpz[4].re = tmpz[0].re - tmp5;
148− "swc1 %[pom3], 0(%[tmpz]) \n\t" // tmpz[0].re = tmpz[0].re + tmp5;
149− "swc1 %[pom], 36(%[tmpz]) \n\t" // tmpz[4].im = tmpz[0].im - tmp6;
150− "swc1 %[pom2], 4(%[tmpz]) \n\t" // tmpz[0].im = tmpz[0].im + tmp6;
151− "lwc1 %[pom1], 16(%[tmpz]) \n\t"
152− "lwc1 %[pom3], 20(%[tmpz]) \n\t"
153− "add.s %[temp1],%[tmp1], %[tmp2] \n\t"
154− "sub.s %[temp], %[pom1], %[tmp8] \n\t"
155− "add.s %[pom2], %[pom3], %[tmp7] \n\t"
156− "sub.s %[temp3],%[tmp3], %[tmp4] \n\t"
157− "sub.s %[temp4],%[tmp2], %[tmp1] \n\t"
158− "swc1 %[temp], 48(%[tmpz]) \n\t" // tmpz[6].re = tmpz[2].re - tmp8;
159− "swc1 %[pom2], 52(%[tmpz]) \n\t" // tmpz[6].im = tmpz[2].im + tmp7;
160− "add.s %[pom1], %[pom1], %[tmp8] \n\t"
161− "sub.s %[pom3], %[pom3], %[tmp7] \n\t"
162− "add.s %[tmp3], %[tmp3], %[tmp4] \n\t"
163− "mul.s %[tmp5], %[f1], %[temp1] \n\t" // tmp5 = pom * (tmp1 + tmp2);
164− "mul.s %[tmp7], %[f1], %[temp3] \n\t" // tmp7 = pom * (tmp3 - tmp4);
165− "mul.s %[tmp6], %[f1], %[temp4] \n\t" // tmp6 = pom * (tmp2 - tmp1);
166− "mul.s %[tmp8], %[f1], %[tmp3] \n\t" // tmp8 = pom * (tmp3 + tmp4);
167− "swc1 %[pom1], 16(%[tmpz]) \n\t" // tmpz[2].re = tmpz[2].re + tmp8;
168− "swc1 %[pom3], 20(%[tmpz]) \n\t" // tmpz[2].im = tmpz[2].im - tmp7;
169− "add.s %[tmp1], %[tmp5], %[tmp7] \n\t" // tmp1 = tmp5 + tmp7;
170− "sub.s %[tmp3], %[tmp5], %[tmp7] \n\t" // tmp3 = tmp5 - tmp7;
171− "add.s %[tmp2], %[tmp6], %[tmp8] \n\t" // tmp2 = tmp6 + tmp8;
172− "sub.s %[tmp4], %[tmp6], %[tmp8] \n\t" // tmp4 = tmp6 - tmp8;
173− "lwc1 %[temp], 8(%[tmpz]) \n\t"
174− "lwc1 %[temp1],12(%[tmpz]) \n\t"
175− "lwc1 %[pom], 24(%[tmpz]) \n\t"
176− "lwc1 %[pom2], 28(%[tmpz]) \n\t"
177− "sub.s %[temp4],%[temp], %[tmp1] \n\t"
178− "sub.s %[temp3],%[temp1], %[tmp2] \n\t"
179− "add.s %[temp], %[temp], %[tmp1] \n\t"
180− "add.s %[temp1],%[temp1], %[tmp2] \n\t"
181− "sub.s %[pom1], %[pom], %[tmp4] \n\t"
182− "add.s %[pom3], %[pom2], %[tmp3] \n\t"
183− "add.s %[pom], %[pom], %[tmp4] \n\t"
184− "sub.s %[pom2], %[pom2], %[tmp3] \n\t"
185− "swc1 %[temp4],40(%[tmpz]) \n\t" // tmpz[5].re = tmpz[1].re - tmp1;
186− "swc1 %[temp3],44(%[tmpz]) \n\t" // tmpz[5].im = tmpz[1].im - tmp2;
187− "swc1 %[temp], 8(%[tmpz]) \n\t" // tmpz[1].re = tmpz[1].re + tmp1;
188− "swc1 %[temp1],12(%[tmpz]) \n\t" // tmpz[1].im = tmpz[1].im + tmp2;
189− "swc1 %[pom1], 56(%[tmpz]) \n\t" // tmpz[7].re = tmpz[3].re - tmp4;
190− "swc1 %[pom3], 60(%[tmpz]) \n\t" // tmpz[7].im = tmpz[3].im + tmp3;
191− "swc1 %[pom], 24(%[tmpz]) \n\t" // tmpz[3].re = tmpz[3].re + tmp4;
192− "swc1 %[pom2], 28(%[tmpz]) \n\t" // tmpz[3].im = tmpz[3].im - tmp3;
193− : [tmp1]"=&f"(tmp1), [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2),
194− [tmp3]"=&f"(tmp3), [tmp2]"=&f"(tmp2), [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5), [tmp7]"=&f"(tmp7),
195− [tmp6]"=&f"(tmp6), [tmp8]"=&f"(tmp8), [pom3]"=&f"(pom3),[temp]"=&f"(temp), [temp1]"=&f"(temp1),
196− [temp3]"=&f"(temp3), [temp4]"=&f"(temp4)
197− : [tmpz]"r"(tmpz), [f1]"f"(f1)
198− : "memory"
199− );
200− }
201−
202− step = 1 << (MAX_LOG2_NFFT - 4);
203− n4 = 4;
204−
205− for (nbits=4; nbits<=s->nbits; nbits++) {
206− num_transforms = (num_transforms >> 1) | 1;
207− n2 = 2 * n4;
208− n34 = 3 * n4;
209−
210− for (n=0; n<num_transforms; n++) {
211− offset = ff_fft_offsets_lut[n] << nbits;
212− tmpz = z + offset;
213−
214− tmpz_n2 = tmpz + n2;
215− tmpz_n4 = tmpz + n4;
216− tmpz_n34 = tmpz + n34;
217−
218− __asm__ volatile (
219− "lwc1 %[pom1], 0(%[tmpz_n2]) \n\t"
220− "lwc1 %[pom], 0(%[tmpz_n34]) \n\t"
221− "lwc1 %[pom2], 4(%[tmpz_n2]) \n\t"
222− "lwc1 %[pom3], 4(%[tmpz_n34]) \n\t"
223− "lwc1 %[temp1],0(%[tmpz]) \n\t"
224− "lwc1 %[temp3],4(%[tmpz]) \n\t"
225− "add.s %[tmp5], %[pom1], %[pom] \n\t" // tmp5 = tmpz[ n2].re + tmpz[n34].re;
226− "sub.s %[tmp1], %[pom1], %[pom] \n\t" // tmp1 = tmpz[ n2].re - tmpz[n34].re;
227− "add.s %[tmp6], %[pom2], %[pom3] \n\t" // tmp6 = tmpz[ n2].im + tmpz[n34].im;
228− "sub.s %[tmp2], %[pom2], %[pom3] \n\t" // tmp2 = tmpz[ n2].im - tmpz[n34].im;
229− "sub.s %[temp], %[temp1], %[tmp5] \n\t"
230− "add.s %[temp1],%[temp1], %[tmp5] \n\t"
231− "sub.s %[temp4],%[temp3], %[tmp6] \n\t"
232− "add.s %[temp3],%[temp3], %[tmp6] \n\t"
233− "swc1 %[temp], 0(%[tmpz_n2]) \n\t" // tmpz[ n2].re = tmpz[ 0].re - tmp5;
234− "swc1 %[temp1],0(%[tmpz]) \n\t" // tmpz[ 0].re = tmpz[ 0].re + tmp5;
235− "lwc1 %[pom1], 0(%[tmpz_n4]) \n\t"
236− "swc1 %[temp4],4(%[tmpz_n2]) \n\t" // tmpz[ n2].im = tmpz[ 0].im - tmp6;
237− "lwc1 %[temp], 4(%[tmpz_n4]) \n\t"
238− "swc1 %[temp3],4(%[tmpz]) \n\t" // tmpz[ 0].im = tmpz[ 0].im + tmp6;
239− "sub.s %[pom], %[pom1], %[tmp2] \n\t"
240− "add.s %[pom1], %[pom1], %[tmp2] \n\t"
241− "add.s %[temp1],%[temp], %[tmp1] \n\t"
242− "sub.s %[temp], %[temp], %[tmp1] \n\t"
243− "swc1 %[pom], 0(%[tmpz_n34]) \n\t" // tmpz[n34].re = tmpz[n4].re - tmp2;
244− "swc1 %[pom1], 0(%[tmpz_n4]) \n\t" // tmpz[ n4].re = tmpz[n4].re + tmp2;
245− "swc1 %[temp1],4(%[tmpz_n34]) \n\t" // tmpz[n34].im = tmpz[n4].im + tmp1;
246− "swc1 %[temp], 4(%[tmpz_n4]) \n\t" // tmpz[ n4].im = tmpz[n4].im - tmp1;
247− : [tmp5]"=&f"(tmp5),
248− [tmp1]"=&f"(tmp1), [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2),
249− [tmp2]"=&f"(tmp2), [tmp6]"=&f"(tmp6), [pom3]"=&f"(pom3),
250− [temp]"=&f"(temp), [temp1]"=&f"(temp1), [temp3]"=&f"(temp3), [temp4]"=&f"(temp4)
251− : [tmpz]"r"(tmpz), [tmpz_n2]"r"(tmpz_n2), [tmpz_n34]"r"(tmpz_n34), [tmpz_n4]"r"(tmpz_n4)
252− : "memory"
253− );
254−
255− w_re_ptr = (float*)(ff_cos_131072 + step);
256− w_im_ptr = (float*)(ff_cos_131072 + MAX_FFT_SIZE/4 - step);
257−
258− for (i=1; i<n4; i++) {
259− w_re = w_re_ptr[0];
260− w_im = w_im_ptr[0];
261− tmpz_n2_i = tmpz_n2 + i;
262− tmpz_n4_i = tmpz_n4 + i;
263− tmpz_n34_i= tmpz_n34 + i;
264− tmpz_i = tmpz + i;
265−
266− __asm__ volatile (
267− "lwc1 %[temp], 0(%[tmpz_n2_i]) \n\t"
268− "lwc1 %[temp1], 4(%[tmpz_n2_i]) \n\t"
269− "lwc1 %[pom], 0(%[tmpz_n34_i]) \n\t"
270− "lwc1 %[pom1], 4(%[tmpz_n34_i]) \n\t"
271− "mul.s %[temp3], %[w_im], %[temp] \n\t"
272− "mul.s %[temp4], %[w_im], %[temp1] \n\t"
273− "mul.s %[pom2], %[w_im], %[pom1] \n\t"
274− "mul.s %[pom3], %[w_im], %[pom] \n\t"
275− "msub.s %[tmp2], %[temp3], %[w_re], %[temp1] \n\t" // tmp2 = w_re * tmpz[ n2+i].im - w_im * tmpz[ n2+i].re;
276− "madd.s %[tmp1], %[temp4], %[w_re], %[temp] \n\t" // tmp1 = w_re * tmpz[ n2+i].re + w_im * tmpz[ n2+i].im;
277− "msub.s %[tmp3], %[pom2], %[w_re], %[pom] \n\t" // tmp3 = w_re * tmpz[n34+i].re - w_im * tmpz[n34+i].im;
278− "madd.s %[tmp4], %[pom3], %[w_re], %[pom1] \n\t" // tmp4 = w_re * tmpz[n34+i].im + w_im * tmpz[n34+i].re;
279− "lwc1 %[temp], 0(%[tmpz_i]) \n\t"
280− "lwc1 %[pom], 4(%[tmpz_i]) \n\t"
281− "add.s %[tmp5], %[tmp1], %[tmp3] \n\t" // tmp5 = tmp1 + tmp3;
282− "sub.s %[tmp1], %[tmp1], %[tmp3] \n\t" // tmp1 = tmp1 - tmp3;
283− "add.s %[tmp6], %[tmp2], %[tmp4] \n\t" // tmp6 = tmp2 + tmp4;
284− "sub.s %[tmp2], %[tmp2], %[tmp4] \n\t" // tmp2 = tmp2 - tmp4;
285− "sub.s %[temp1], %[temp], %[tmp5] \n\t"
286− "add.s %[temp], %[temp], %[tmp5] \n\t"
287− "sub.s %[pom1], %[pom], %[tmp6] \n\t"
288− "add.s %[pom], %[pom], %[tmp6] \n\t"
289− "lwc1 %[temp3], 0(%[tmpz_n4_i]) \n\t"
290− "lwc1 %[pom2], 4(%[tmpz_n4_i]) \n\t"
291− "swc1 %[temp1], 0(%[tmpz_n2_i]) \n\t" // tmpz[ n2+i].re = tmpz[ i].re - tmp5;
292− "swc1 %[temp], 0(%[tmpz_i]) \n\t" // tmpz[ i].re = tmpz[ i].re + tmp5;
293− "swc1 %[pom1], 4(%[tmpz_n2_i]) \n\t" // tmpz[ n2+i].im = tmpz[ i].im - tmp6;
294− "swc1 %[pom] , 4(%[tmpz_i]) \n\t" // tmpz[ i].im = tmpz[ i].im + tmp6;
295− "sub.s %[temp4], %[temp3], %[tmp2] \n\t"
296− "add.s %[pom3], %[pom2], %[tmp1] \n\t"
297− "add.s %[temp3], %[temp3], %[tmp2] \n\t"
298− "sub.s %[pom2], %[pom2], %[tmp1] \n\t"
299− "swc1 %[temp4], 0(%[tmpz_n34_i]) \n\t" // tmpz[n34+i].re = tmpz[n4+i].re - tmp2;
300− "swc1 %[pom3], 4(%[tmpz_n34_i]) \n\t" // tmpz[n34+i].im = tmpz[n4+i].im + tmp1;
301− "swc1 %[temp3], 0(%[tmpz_n4_i]) \n\t" // tmpz[ n4+i].re = tmpz[n4+i].re + tmp2;
302− "swc1 %[pom2], 4(%[tmpz_n4_i]) \n\t" // tmpz[ n4+i].im = tmpz[n4+i].im - tmp1;
303− : [tmp1]"=&f"(tmp1), [tmp2]"=&f" (tmp2), [temp]"=&f"(temp), [tmp3]"=&f"(tmp3),
304− [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5), [tmp6]"=&f"(tmp6),
305− [temp1]"=&f"(temp1), [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
306− [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2), [pom3]"=&f"(pom3)
307− : [w_re]"f"(w_re), [w_im]"f"(w_im),
308− [tmpz_i]"r"(tmpz_i),[tmpz_n2_i]"r"(tmpz_n2_i),
309− [tmpz_n34_i]"r"(tmpz_n34_i), [tmpz_n4_i]"r"(tmpz_n4_i)
310− : "memory"
311− );
312− w_re_ptr += step;
313− w_im_ptr -= step;
314− }
315− }
316− step >>= 1;
317− n4 <<= 1;
318− }
319−}
320−
321−/**
322− * MDCT/IMDCT transforms.
323− */
324−
325−static void ff_imdct_half_mips(FFTContext *s, FFTSample *output, const FFTSample *input)
326−{
327− int k, n8, n4, n2, n, j;
328− const uint16_t *revtab = s->revtab;
329− const FFTSample *tcos = s->tcos;
330− const FFTSample *tsin = s->tsin;
331− const FFTSample *in1, *in2, *in3, *in4;
332− FFTComplex *z = (FFTComplex *)output;
333−
334− int j1;
335− const float *tcos1, *tsin1, *tcos2, *tsin2;
336− float temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8,
337− temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16;
338− FFTComplex *z1, *z2;
339−
340− n = 1 << s->mdct_bits;
341− n2 = n >> 1;
342− n4 = n >> 2;
343− n8 = n >> 3;
344−
345− /* pre rotation */
346− in1 = input;
347− in2 = input + n2 - 1;
348− in3 = input + 2;
349− in4 = input + n2 - 3;
350−
351− tcos1 = tcos;
352− tsin1 = tsin;
353−
354− /* n4 = 64 or 128 */
355− for(k = 0; k < n4; k += 2) {
356− j = revtab[k ];
357− j1 = revtab[k + 1];
358−
359− __asm__ volatile (
360− "lwc1 %[temp1], 0(%[in2]) \t\n"
361− "lwc1 %[temp2], 0(%[tcos1]) \t\n"
362− "lwc1 %[temp3], 0(%[tsin1]) \t\n"
363− "lwc1 %[temp4], 0(%[in1]) \t\n"
364− "lwc1 %[temp5], 0(%[in4]) \t\n"
365− "mul.s %[temp9], %[temp1], %[temp2] \t\n"
366− "mul.s %[temp10], %[temp1], %[temp3] \t\n"
367− "lwc1 %[temp6], 4(%[tcos1]) \t\n"
368− "lwc1 %[temp7], 4(%[tsin1]) \t\n"
369− "nmsub.s %[temp9], %[temp9], %[temp4], %[temp3] \t\n"
370− "madd.s %[temp10], %[temp10], %[temp4], %[temp2] \t\n"
371− "mul.s %[temp11], %[temp5], %[temp6] \t\n"
372− "mul.s %[temp12], %[temp5], %[temp7] \t\n"
373− "lwc1 %[temp8], 0(%[in3]) \t\n"
374− PTR_ADDIU " %[tcos1], %[tcos1], 8 \t\n"
375− PTR_ADDIU " %[tsin1], %[tsin1], 8 \t\n"
376− PTR_ADDIU " %[in1], %[in1], 16 \t\n"
377− "nmsub.s %[temp11], %[temp11], %[temp8], %[temp7] \t\n"
378− "madd.s %[temp12], %[temp12], %[temp8], %[temp6] \t\n"
379− PTR_ADDIU " %[in2], %[in2], -16 \t\n"
380− PTR_ADDIU " %[in3], %[in3], 16 \t\n"
381− PTR_ADDIU " %[in4], %[in4], -16 \t\n"
382−
383− : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2),
384− [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
385− [temp5]"=&f"(temp5), [temp6]"=&f"(temp6),
386− [temp7]"=&f"(temp7), [temp8]"=&f"(temp8),
387− [temp9]"=&f"(temp9), [temp10]"=&f"(temp10),
388− [temp11]"=&f"(temp11), [temp12]"=&f"(temp12),
389− [tsin1]"+r"(tsin1), [tcos1]"+r"(tcos1),
390− [in1]"+r"(in1), [in2]"+r"(in2),
391− [in3]"+r"(in3), [in4]"+r"(in4)
392− :
393− : "memory"
394− );
395−
396− z[j ].re = temp9;
397− z[j ].im = temp10;
398− z[j1].re = temp11;
399− z[j1].im = temp12;
400− }
401−
402− s->fft_calc(s, z);
403−
404− /* post rotation + reordering */
405− /* n8 = 32 or 64 */
406− for(k = 0; k < n8; k += 2) {
407− tcos1 = &tcos[n8 - k - 2];
408− tsin1 = &tsin[n8 - k - 2];
409− tcos2 = &tcos[n8 + k];
410− tsin2 = &tsin[n8 + k];
411− z1 = &z[n8 - k - 2];
412− z2 = &z[n8 + k ];
413−
414− __asm__ volatile (
415− "lwc1 %[temp1], 12(%[z1]) \t\n"
416− "lwc1 %[temp2], 4(%[tsin1]) \t\n"
417− "lwc1 %[temp3], 4(%[tcos1]) \t\n"
418− "lwc1 %[temp4], 8(%[z1]) \t\n"
419− "lwc1 %[temp5], 4(%[z1]) \t\n"
420− "mul.s %[temp9], %[temp1], %[temp2] \t\n"
421− "mul.s %[temp10], %[temp1], %[temp3] \t\n"
422− "lwc1 %[temp6], 0(%[tsin1]) \t\n"
423− "lwc1 %[temp7], 0(%[tcos1]) \t\n"
424− "nmsub.s %[temp9], %[temp9], %[temp4], %[temp3] \t\n"
425− "madd.s %[temp10], %[temp10], %[temp4], %[temp2] \t\n"
426− "mul.s %[temp11], %[temp5], %[temp6] \t\n"
427− "mul.s %[temp12], %[temp5], %[temp7] \t\n"
428− "lwc1 %[temp8], 0(%[z1]) \t\n"
429− "lwc1 %[temp1], 4(%[z2]) \t\n"
430− "lwc1 %[temp2], 0(%[tsin2]) \t\n"
431− "lwc1 %[temp3], 0(%[tcos2]) \t\n"
432− "nmsub.s %[temp11], %[temp11], %[temp8], %[temp7] \t\n"
433− "madd.s %[temp12], %[temp12], %[temp8], %[temp6] \t\n"
434− "mul.s %[temp13], %[temp1], %[temp2] \t\n"
435− "mul.s %[temp14], %[temp1], %[temp3] \t\n"
436− "lwc1 %[temp4], 0(%[z2]) \t\n"
437− "lwc1 %[temp5], 12(%[z2]) \t\n"
438− "lwc1 %[temp6], 4(%[tsin2]) \t\n"
439− "lwc1 %[temp7], 4(%[tcos2]) \t\n"
440− "nmsub.s %[temp13], %[temp13], %[temp4], %[temp3] \t\n"
441− "madd.s %[temp14], %[temp14], %[temp4], %[temp2] \t\n"
442− "mul.s %[temp15], %[temp5], %[temp6] \t\n"
443− "mul.s %[temp16], %[temp5], %[temp7] \t\n"
444− "lwc1 %[temp8], 8(%[z2]) \t\n"
445− "nmsub.s %[temp15], %[temp15], %[temp8], %[temp7] \t\n"
446− "madd.s %[temp16], %[temp16], %[temp8], %[temp6] \t\n"
447− : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2),
448− [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
449− [temp5]"=&f"(temp5), [temp6]"=&f"(temp6),
450− [temp7]"=&f"(temp7), [temp8]"=&f"(temp8),
451− [temp9]"=&f"(temp9), [temp10]"=&f"(temp10),
452− [temp11]"=&f"(temp11), [temp12]"=&f"(temp12),
453− [temp13]"=&f"(temp13), [temp14]"=&f"(temp14),
454− [temp15]"=&f"(temp15), [temp16]"=&f"(temp16)
455− : [z1]"r"(z1), [z2]"r"(z2),
456− [tsin1]"r"(tsin1), [tcos1]"r"(tcos1),
457− [tsin2]"r"(tsin2), [tcos2]"r"(tcos2)
458− : "memory"
459− );
460−
461− z1[1].re = temp9;
462− z1[1].im = temp14;
463− z2[0].re = temp13;
464− z2[0].im = temp10;
465−
466− z1[0].re = temp11;
467− z1[0].im = temp16;
468− z2[1].re = temp15;
469− z2[1].im = temp12;
470− }
471−}
472−
473−/**
474− * Compute inverse MDCT of size N = 2^nbits
475− * @param output N samples
476− * @param input N/2 samples
477− */
478−static void ff_imdct_calc_mips(FFTContext *s, FFTSample *output, const FFTSample *input)
479−{
480− int k;
481− int n = 1 << s->mdct_bits;
482− int n2 = n >> 1;
483− int n4 = n >> 2;
484−
485− ff_imdct_half_mips(s, output+n4, input);
486−
487− for(k = 0; k < n4; k+=4) {
488− output[k] = -output[n2-k-1];
489− output[k+1] = -output[n2-k-2];
490− output[k+2] = -output[n2-k-3];
491− output[k+3] = -output[n2-k-4];
492−
493− output[n-k-1] = output[n2+k];
494− output[n-k-2] = output[n2+k+1];
495− output[n-k-3] = output[n2+k+2];
496− output[n-k-4] = output[n2+k+3];
497− }
498−}
499−#endif /* !HAVE_MIPS32R6 && !HAVE_MIPS64R6 */
500−#endif /* HAVE_INLINE_ASM */
501−
502−av_cold void ff_fft_init_mips(FFTContext *s)
503−{
504− ff_fft_lut_init();
505− ff_init_ff_cos_tabs(17);
506−
507−#if HAVE_INLINE_ASM
508−#if !HAVE_MIPS32R6 && !HAVE_MIPS64R6
509− s->fft_calc = ff_fft_calc_mips;
510−#if CONFIG_MDCT
511− s->imdct_calc = ff_imdct_calc_mips;
512− s->imdct_half = ff_imdct_half_mips;
513−#endif
514−#endif
515−#endif
516−}
libavcodec/ppc/Makefile+0−3
@@ -1,9 +1,6 @@
11 # subsystems
22 OBJS-$(CONFIG_AUDIODSP) += ppc/audiodsp.o
33 OBJS-$(CONFIG_BLOCKDSP) += ppc/blockdsp.o
4−OBJS-$(CONFIG_FFT) += ppc/fft_init.o \
5− ppc/fft_altivec.o \
6− ppc/fft_vsx.o
74 OBJS-$(CONFIG_FDCTDSP) += ppc/fdctdsp.o
85 OBJS-$(CONFIG_FMTCONVERT) += ppc/fmtconvert_altivec.o
96 OBJS-$(CONFIG_H264CHROMA) += ppc/h264chroma_init.o
libavcodec/ppc/fft_altivec.S+0−458
@@ -1,458 +0,0 @@
1−/*
2− * FFT transform with Altivec optimizations
3− * Copyright (c) 2009 Loren Merritt
4− *
5− * This algorithm (though not any of the implementation details) is
6− * based on libdjbfft by D. J. Bernstein.
7− *
8− * This file is part of FFmpeg.
9− *
10− * FFmpeg is free software; you can redistribute it and/or
11− * modify it under the terms of the GNU Lesser General Public
12− * License as published by the Free Software Foundation; either
13− * version 2.1 of the License, or (at your option) any later version.
14− *
15− * FFmpeg is distributed in the hope that it will be useful,
16− * but WITHOUT ANY WARRANTY; without even the implied warranty of
17− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18− * Lesser General Public License for more details.
19− *
20− * You should have received a copy of the GNU Lesser General Public
21− * License along with FFmpeg; if not, write to the Free Software
22− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23− */
24−
25−/*
26− * These functions are not individually interchangeable with the C versions.
27− * While C takes arrays of FFTComplex, Altivec leaves intermediate results
28− * in blocks as convenient to the vector size.
29− * i.e. {4x real, 4x imaginary, 4x real, ...}
30− *
31− * I ignore standard calling convention.
32− * Instead, the following registers are treated as global constants:
33− * v14: zero
34− * v15..v18: cosines
35− * v19..v29: permutations
36− * r9: 16
37− * r12: ff_cos_tabs
38− * and the rest are free for local use.
39− */
40−
41−#include "config.h"
42−
43−#if HAVE_GNU_AS && HAVE_ALTIVEC && HAVE_BIGENDIAN
44−
45−#include "asm.S"
46−
47−.text
48−
49−.macro addi2 ra, imm // add 32-bit immediate
50−.if \imm & 0xffff
51− addi \ra, \ra, \imm@l
52−.endif
53−.if (\imm+0x8000)>>16
54− addis \ra, \ra, \imm@ha
55−.endif
56−.endm
57−
58−.macro FFT4 a0, a1, a2, a3 // in:0-1 out:2-3
59− vperm \a2,\a0,\a1,v20 // vcprm(0,1,s2,s1) // {r0,i0,r3,i2}
60− vperm \a3,\a0,\a1,v21 // vcprm(2,3,s0,s3) // {r1,i1,r2,i3}
61− vaddfp \a0,\a2,\a3 // {t1,t2,t6,t5}
62− vsubfp \a1,\a2,\a3 // {t3,t4,t8,t7}
63− vmrghw \a2,\a0,\a1 // vcprm(0,s0,1,s1) // {t1,t3,t2,t4}
64− vperm \a3,\a0,\a1,v22 // vcprm(2,s3,3,s2) // {t6,t7,t5,t8}
65− vaddfp \a0,\a2,\a3 // {r0,r1,i0,i1}
66− vsubfp \a1,\a2,\a3 // {r2,r3,i2,i3}
67− vperm \a2,\a0,\a1,v23 // vcprm(0,1,s0,s1) // {r0,r1,r2,r3}
68− vperm \a3,\a0,\a1,v24 // vcprm(2,3,s2,s3) // {i0,i1,i2,i3}
69−.endm
70−
71−.macro FFT4x2 a0, a1, b0, b1, a2, a3, b2, b3
72− vperm \a2,\a0,\a1,v20 // vcprm(0,1,s2,s1) // {r0,i0,r3,i2}
73− vperm \a3,\a0,\a1,v21 // vcprm(2,3,s0,s3) // {r1,i1,r2,i3}
74− vperm \b2,\b0,\b1,v20
75− vperm \b3,\b0,\b1,v21
76− vaddfp \a0,\a2,\a3 // {t1,t2,t6,t5}
77− vsubfp \a1,\a2,\a3 // {t3,t4,t8,t7}
78− vaddfp \b0,\b2,\b3
79− vsubfp \b1,\b2,\b3
80− vmrghw \a2,\a0,\a1 // vcprm(0,s0,1,s1) // {t1,t3,t2,t4}
81− vperm \a3,\a0,\a1,v22 // vcprm(2,s3,3,s2) // {t6,t7,t5,t8}
82− vmrghw \b2,\b0,\b1
83− vperm \b3,\b0,\b1,v22
84− vaddfp \a0,\a2,\a3 // {r0,r1,i0,i1}
85− vsubfp \a1,\a2,\a3 // {r2,r3,i2,i3}
86− vaddfp \b0,\b2,\b3
87− vsubfp \b1,\b2,\b3
88− vperm \a2,\a0,\a1,v23 // vcprm(0,1,s0,s1) // {r0,r1,r2,r3}
89− vperm \a3,\a0,\a1,v24 // vcprm(2,3,s2,s3) // {i0,i1,i2,i3}
90− vperm \b2,\b0,\b1,v23
91− vperm \b3,\b0,\b1,v24
92−.endm
93−
94−.macro FFT8 a0, a1, b0, b1, a2, a3, b2, b3, b4 // in,out:a0-b1
95− vmrghw \b2,\b0,\b1 // vcprm(0,s0,1,s1) // {r4,r6,i4,i6}
96− vmrglw \b3,\b0,\b1 // vcprm(2,s2,3,s3) // {r5,r7,i5,i7}
97− vperm \a2,\a0,\a1,v20 // FFT4 ...
98− vperm \a3,\a0,\a1,v21
99− vaddfp \b0,\b2,\b3 // {t1,t3,t2,t4}
100− vsubfp \b1,\b2,\b3 // {r5,r7,i5,i7}
101− vperm \b4,\b1,\b1,v25 // vcprm(2,3,0,1) // {i5,i7,r5,r7}
102− vaddfp \a0,\a2,\a3
103− vsubfp \a1,\a2,\a3
104− vmaddfp \b1,\b1,v17,v14 // * {-1,1,1,-1}/sqrt(2)
105− vmaddfp \b1,\b4,v18,\b1 // * { 1,1,1,1 }/sqrt(2) // {t8,ta,t7,t9}
106− vmrghw \a2,\a0,\a1
107− vperm \a3,\a0,\a1,v22
108− vperm \b2,\b0,\b1,v26 // vcprm(1,2,s3,s0) // {t3,t2,t9,t8}
109− vperm \b3,\b0,\b1,v27 // vcprm(0,3,s2,s1) // {t1,t4,t7,ta}
110− vaddfp \a0,\a2,\a3
111− vsubfp \a1,\a2,\a3
112− vaddfp \b0,\b2,\b3 // {t1,t2,t9,ta}
113− vsubfp \b1,\b2,\b3 // {t6,t5,tc,tb}
114− vperm \a2,\a0,\a1,v23
115− vperm \a3,\a0,\a1,v24
116− vperm \b2,\b0,\b1,v28 // vcprm(0,2,s1,s3) // {t1,t9,t5,tb}
117− vperm \b3,\b0,\b1,v29 // vcprm(1,3,s0,s2) // {t2,ta,t6,tc}
118− vsubfp \b0,\a2,\b2 // {r4,r5,r6,r7}
119− vsubfp \b1,\a3,\b3 // {i4,i5,i6,i7}
120− vaddfp \a0,\a2,\b2 // {r0,r1,r2,r3}
121− vaddfp \a1,\a3,\b3 // {i0,i1,i2,i3}
122−.endm
123−
124−.macro BF d0,d1,s0,s1
125− vsubfp \d1,\s0,\s1
126− vaddfp \d0,\s0,\s1
127−.endm
128−
129−.macro zip d0,d1,s0,s1
130− vmrghw \d0,\s0,\s1
131− vmrglw \d1,\s0,\s1
132−.endm
133−
134−.macro def_fft4 interleave
135−fft4\interleave\()_altivec:
136− lvx v0, 0,r3
137− lvx v1,r9,r3
138− FFT4 v0,v1,v2,v3
139−.ifnb \interleave
140− zip v0,v1,v2,v3
141− stvx v0, 0,r3
142− stvx v1,r9,r3
143−.else
144− stvx v2, 0,r3
145− stvx v3,r9,r3
146−.endif
147− blr
148−.endm
149−
150−.macro def_fft8 interleave
151−fft8\interleave\()_altivec:
152− addi r4,r3,32
153− lvx v0, 0,r3
154− lvx v1,r9,r3
155− lvx v2, 0,r4
156− lvx v3,r9,r4
157− FFT8 v0,v1,v2,v3,v4,v5,v6,v7,v8
158−.ifnb \interleave
159− zip v4,v5,v0,v1
160− zip v6,v7,v2,v3
161− stvx v4, 0,r3
162− stvx v5,r9,r3
163− stvx v6, 0,r4
164− stvx v7,r9,r4
165−.else
166− stvx v0, 0,r3
167− stvx v1,r9,r3
168− stvx v2, 0,r4
169− stvx v3,r9,r4
170−.endif
171− blr
172−.endm
173−
174−.macro def_fft16 interleave
175−fft16\interleave\()_altivec:
176− addi r5,r3,64
177− addi r6,r3,96
178− addi r4,r3,32
179− lvx v0, 0,r5
180− lvx v1,r9,r5
181− lvx v2, 0,r6
182− lvx v3,r9,r6
183− FFT4x2 v0,v1,v2,v3,v4,v5,v6,v7
184− lvx v0, 0,r3
185− lvx v1,r9,r3
186− lvx v2, 0,r4
187− lvx v3,r9,r4
188− FFT8 v0,v1,v2,v3,v8,v9,v10,v11,v12
189− vmaddfp v8,v4,v15,v14 // r2*wre
190− vmaddfp v9,v5,v15,v14 // i2*wre
191− vmaddfp v10,v6,v15,v14 // r3*wre
192− vmaddfp v11,v7,v15,v14 // i3*wre
193− vmaddfp v8,v5,v16,v8 // i2*wim
194− vnmsubfp v9,v4,v16,v9 // r2*wim
195− vnmsubfp v10,v7,v16,v10 // i3*wim
196− vmaddfp v11,v6,v16,v11 // r3*wim
197− BF v10,v12,v10,v8
198− BF v11,v13,v9,v11
199− BF v0,v4,v0,v10
200− BF v3,v7,v3,v12
201− BF v1,v5,v1,v11
202− BF v2,v6,v2,v13
203−.ifnb \interleave
204− zip v8, v9,v0,v1
205− zip v10,v11,v2,v3
206− zip v12,v13,v4,v5
207− zip v14,v15,v6,v7
208− stvx v8, 0,r3
209− stvx v9,r9,r3
210− stvx v10, 0,r4
211− stvx v11,r9,r4
212− stvx v12, 0,r5
213− stvx v13,r9,r5
214− stvx v14, 0,r6
215− stvx v15,r9,r6
216−.else
217− stvx v0, 0,r3
218− stvx v4, 0,r5
219− stvx v3,r9,r4
220− stvx v7,r9,r6
221− stvx v1,r9,r3
222− stvx v5,r9,r5
223− stvx v2, 0,r4
224− stvx v6, 0,r6
225−.endif
226− blr
227−.endm
228−
229−// void pass(float *z, float *wre, int n)
230−.macro PASS interleave, suffix
231−fft_pass\suffix\()_altivec:
232− mtctr r5
233− slwi r0,r5,4
234− slwi r7,r5,6 // o2
235− slwi r5,r5,5 // o1
236− add r10,r5,r7 // o3
237− add r0,r4,r0 // wim
238− addi r6,r5,16 // o1+16
239− addi r8,r7,16 // o2+16
240− addi r11,r10,16 // o3+16
241−1:
242− lvx v8, 0,r4 // wre
243− lvx v10, 0,r0 // wim
244− sub r0,r0,r9
245− lvx v9, 0,r0
246− vperm v9,v9,v10,v19 // vcprm(s0,3,2,1) => wim[0 .. -3]
247− lvx v4,r3,r7 // r2 = z[o2]
248− lvx v5,r3,r8 // i2 = z[o2+16]
249− lvx v6,r3,r10 // r3 = z[o3]
250− lvx v7,r3,r11 // i3 = z[o3+16]
251− vmaddfp v10,v4,v8,v14 // r2*wre
252− vmaddfp v11,v5,v8,v14 // i2*wre
253− vmaddfp v12,v6,v8,v14 // r3*wre
254− vmaddfp v13,v7,v8,v14 // i3*wre
255− lvx v0, 0,r3 // r0 = z[0]
256− lvx v3,r3,r6 // i1 = z[o1+16]
257− vmaddfp v10,v5,v9,v10 // i2*wim
258− vnmsubfp v11,v4,v9,v11 // r2*wim
259− vnmsubfp v12,v7,v9,v12 // i3*wim
260− vmaddfp v13,v6,v9,v13 // r3*wim
261− lvx v1,r3,r9 // i0 = z[16]
262− lvx v2,r3,r5 // r1 = z[o1]
263− BF v12,v8,v12,v10
264− BF v13,v9,v11,v13
265− BF v0,v4,v0,v12
266− BF v3,v7,v3,v8
267−.if !\interleave
268− stvx v0, 0,r3
269− stvx v4,r3,r7
270− stvx v3,r3,r6
271− stvx v7,r3,r11
272−.endif
273− BF v1,v5,v1,v13
274− BF v2,v6,v2,v9
275−.if !\interleave
276− stvx v1,r3,r9
277− stvx v2,r3,r5
278− stvx v5,r3,r8
279− stvx v6,r3,r10
280−.else
281− vmrghw v8,v0,v1
282− vmrglw v9,v0,v1
283− stvx v8, 0,r3
284− stvx v9,r3,r9
285− vmrghw v8,v2,v3
286− vmrglw v9,v2,v3
287− stvx v8,r3,r5
288− stvx v9,r3,r6
289− vmrghw v8,v4,v5
290− vmrglw v9,v4,v5
291− stvx v8,r3,r7
292− stvx v9,r3,r8
293− vmrghw v8,v6,v7
294− vmrglw v9,v6,v7
295− stvx v8,r3,r10
296− stvx v9,r3,r11
297−.endif
298− addi r3,r3,32
299− addi r4,r4,16
300− bdnz 1b
301− sub r3,r3,r5
302− blr
303−.endm
304−
305−#define M_SQRT1_2 0.70710678118654752440 /* 1/sqrt(2) */
306−
307−#define WORD_0 0x00,0x01,0x02,0x03
308−#define WORD_1 0x04,0x05,0x06,0x07
309−#define WORD_2 0x08,0x09,0x0a,0x0b
310−#define WORD_3 0x0c,0x0d,0x0e,0x0f
311−#define WORD_s0 0x10,0x11,0x12,0x13
312−#define WORD_s1 0x14,0x15,0x16,0x17
313−#define WORD_s2 0x18,0x19,0x1a,0x1b
314−#define WORD_s3 0x1c,0x1d,0x1e,0x1f
315−
316−#define vcprm(a, b, c, d) .byte WORD_##a, WORD_##b, WORD_##c, WORD_##d
317−
318− .rodata
319− .align 4
320−fft_data:
321− .float 0, 0, 0, 0
322− .float 1, 0.92387953, M_SQRT1_2, 0.38268343
323− .float 0, 0.38268343, M_SQRT1_2, 0.92387953
324− .float -M_SQRT1_2, M_SQRT1_2, M_SQRT1_2,-M_SQRT1_2
325− .float M_SQRT1_2, M_SQRT1_2, M_SQRT1_2, M_SQRT1_2
326− vcprm(s0,3,2,1)
327− vcprm(0,1,s2,s1)
328− vcprm(2,3,s0,s3)
329− vcprm(2,s3,3,s2)
330− vcprm(0,1,s0,s1)
331− vcprm(2,3,s2,s3)
332− vcprm(2,3,0,1)
333− vcprm(1,2,s3,s0)
334− vcprm(0,3,s2,s1)
335− vcprm(0,2,s1,s3)
336− vcprm(1,3,s0,s2)
337−
338−.macro lvm b, r, regs:vararg
339− lvx \r, 0, \b
340− addi \b, \b, 16
341− .ifnb \regs
342− lvm \b, \regs
343− .endif
344−.endm
345−
346−.macro stvm b, r, regs:vararg
347− stvx \r, 0, \b
348− addi \b, \b, 16
349− .ifnb \regs
350− stvm \b, \regs
351− .endif
352−.endm
353−
354−.macro fft_calc interleave
355−extfunc ff_fft_calc\interleave\()_altivec
356− mflr r0
357− stp r0, 2*PS(R(1))
358− stpu r1, -(160+16*PS)(R(1))
359− get_got r11
360− addi r6, r1, 16*PS
361− stvm r6, v20, v21, v22, v23, v24, v25, v26, v27, v28, v29
362− mfvrsave r0
363− stw r0, 15*PS(R(1))
364−#if __APPLE__
365− li r6, 0xfffffffc
366−#else
367− li r6, -4
368−#endif
369− mtvrsave r6
370−
371− movrel r6, fft_data, r11
372− lvm r6, v14, v15, v16, v17, v18, v19, v20, v21
373− lvm r6, v22, v23, v24, v25, v26, v27, v28, v29
374−
375− li r9, 16
376− movrel r12, X(ff_cos_tabs), r11
377−
378− movrel r6, fft_dispatch_tab\interleave\()_altivec, r11
379− lwz r3, 0(R(3))
380− subi r3, r3, 2
381− slwi r3, r3, 2+ARCH_PPC64
382− lpx r3, r3, r6
383− mtctr r3
384− mr r3, r4
385− bctrl
386−
387− addi r6, r1, 16*PS
388− lvm r6, v20, v21, v22, v23, v24, v25, v26, v27, v28, v29
389− lwz r6, 15*PS(R(1))
390− mtvrsave r6
391− lp r1, 0(R(1))
392− lp r0, 2*PS(R(1))
393− mtlr r0
394− blr
395−.endm
396−
397−.macro DECL_FFT suffix, bits, n, n2, n4
398−fft\n\suffix\()_altivec:
399− mflr r0
400− stp r0,PS*(\bits-3)(R(1))
401− bl fft\n2\()_altivec
402− addi2 r3,\n*4
403− bl fft\n4\()_altivec
404− addi2 r3,\n*2
405− bl fft\n4\()_altivec
406− addi2 r3,\n*-6
407− lp r0,PS*(\bits-3)(R(1))
408− lp r4,\bits*PS(R(12))
409− mtlr r0
410− li r5,\n/16
411− b fft_pass\suffix\()_altivec
412−.endm
413−
414−.macro DECL_FFTS interleave, suffix
415− .text
416− def_fft4 \suffix
417− def_fft8 \suffix
418− def_fft16 \suffix
419− PASS \interleave, \suffix
420− DECL_FFT \suffix, 5, 32, 16, 8
421− DECL_FFT \suffix, 6, 64, 32, 16
422− DECL_FFT \suffix, 7, 128, 64, 32
423− DECL_FFT \suffix, 8, 256, 128, 64
424− DECL_FFT \suffix, 9, 512, 256, 128
425− DECL_FFT \suffix,10, 1024, 512, 256
426− DECL_FFT \suffix,11, 2048, 1024, 512
427− DECL_FFT \suffix,12, 4096, 2048, 1024
428− DECL_FFT \suffix,13, 8192, 4096, 2048
429− DECL_FFT \suffix,14,16384, 8192, 4096
430− DECL_FFT \suffix,15,32768,16384, 8192
431− DECL_FFT \suffix,16,65536,32768,16384
432−
433− fft_calc \suffix
434−
435− .rodata
436− .align 3
437−fft_dispatch_tab\suffix\()_altivec:
438− PTR fft4\suffix\()_altivec
439− PTR fft8\suffix\()_altivec
440− PTR fft16\suffix\()_altivec
441− PTR fft32\suffix\()_altivec
442− PTR fft64\suffix\()_altivec
443− PTR fft128\suffix\()_altivec
444− PTR fft256\suffix\()_altivec
445− PTR fft512\suffix\()_altivec
446− PTR fft1024\suffix\()_altivec
447− PTR fft2048\suffix\()_altivec
448− PTR fft4096\suffix\()_altivec
449− PTR fft8192\suffix\()_altivec
450− PTR fft16384\suffix\()_altivec
451− PTR fft32768\suffix\()_altivec
452− PTR fft65536\suffix\()_altivec
453−.endm
454−
455−DECL_FFTS 0
456−DECL_FFTS 1, _interleave
457−
458−#endif /* HAVE_GNU_AS && HAVE_ALTIVEC && HAVE_BIGENDIAN */
libavcodec/ppc/fft_init.c+0−168
@@ -1,168 +0,0 @@
1−/*
2− * FFT/IFFT transforms
3− * AltiVec-enabled
4− * Copyright (c) 2009 Loren Merritt
5− *
6− * This file is part of FFmpeg.
7− *
8− * FFmpeg is free software; you can redistribute it and/or
9− * modify it under the terms of the GNU Lesser General Public
10− * License as published by the Free Software Foundation; either
11− * version 2.1 of the License, or (at your option) any later version.
12− *
13− * FFmpeg is distributed in the hope that it will be useful,
14− * but WITHOUT ANY WARRANTY; without even the implied warranty of
15− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16− * Lesser General Public License for more details.
17− *
18− * You should have received a copy of the GNU Lesser General Public
19− * License along with FFmpeg; if not, write to the Free Software
20− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21− */
22−
23−#include "config.h"
24−#include "libavutil/attributes.h"
25−#include "libavutil/cpu.h"
26−#include "libavutil/ppc/cpu.h"
27−#include "libavutil/ppc/util_altivec.h"
28−#include "libavcodec/fft.h"
29−
30−/**
31− * Do a complex FFT with the parameters defined in ff_fft_init().
32− * The input data must be permuted before with s->revtab table.
33− * No 1.0 / sqrt(n) normalization is done.
34− * AltiVec-enabled:
35− * This code assumes that the 'z' pointer is 16 bytes-aligned.
36− * It also assumes all FFTComplex are 8 bytes-aligned pairs of floats.
37− */
38−
39−#if HAVE_VSX
40−#include "fft_vsx.h"
41−#else
42−void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
43−void ff_fft_calc_interleave_altivec(FFTContext *s, FFTComplex *z);
44−#endif
45−
46−#if HAVE_GNU_AS && HAVE_ALTIVEC && (HAVE_BIGENDIAN || HAVE_VSX)
47−static void imdct_half_altivec(FFTContext *s, FFTSample *output, const FFTSample *input)
48−{
49− int j, k;
50− int n = 1 << s->mdct_bits;
51− int n4 = n >> 2;
52− int n8 = n >> 3;
53− int n32 = n >> 5;
54− const uint16_t *revtabj = s->revtab;
55− const uint16_t *revtabk = s->revtab+n4;
56− const vec_f *tcos = (const vec_f*)(s->tcos+n8);
57− const vec_f *tsin = (const vec_f*)(s->tsin+n8);
58− const vec_f *pin = (const vec_f*)(input+n4);
59− vec_f *pout = (vec_f*)(output+n4);
60−
61− /* pre rotation */
62− k = n32-1;
63− do {
64− vec_f cos,sin,cos0,sin0,cos1,sin1,re,im,r0,i0,r1,i1,a,b,c,d;
65−#define CMULA(p,o0,o1,o2,o3)\
66− a = pin[ k*2+p]; /* { z[k].re, z[k].im, z[k+1].re, z[k+1].im } */\
67− b = pin[-k*2-p-1]; /* { z[-k-2].re, z[-k-2].im, z[-k-1].re, z[-k-1].im } */\
68− re = vec_perm(a, b, vcprm(0,2,s0,s2)); /* { z[k].re, z[k+1].re, z[-k-2].re, z[-k-1].re } */\
69− im = vec_perm(a, b, vcprm(s3,s1,3,1)); /* { z[-k-1].im, z[-k-2].im, z[k+1].im, z[k].im } */\
70− cos = vec_perm(cos0, cos1, vcprm(o0,o1,s##o2,s##o3)); /* { cos[k], cos[k+1], cos[-k-2], cos[-k-1] } */\
71− sin = vec_perm(sin0, sin1, vcprm(o0,o1,s##o2,s##o3));\
72− r##p = im*cos - re*sin;\
73− i##p = re*cos + im*sin;
74−#define STORE2(v,dst)\
75− j = dst;\
76− vec_ste(v, 0, output+j*2);\
77− vec_ste(v, 4, output+j*2);
78−#define STORE8(p)\
79− a = vec_perm(r##p, i##p, vcprm(0,s0,0,s0));\
80− b = vec_perm(r##p, i##p, vcprm(1,s1,1,s1));\
81− c = vec_perm(r##p, i##p, vcprm(2,s2,2,s2));\
82− d = vec_perm(r##p, i##p, vcprm(3,s3,3,s3));\
83− STORE2(a, revtabk[ p*2-4]);\
84− STORE2(b, revtabk[ p*2-3]);\
85− STORE2(c, revtabj[-p*2+2]);\
86− STORE2(d, revtabj[-p*2+3]);
87−
88− cos0 = tcos[k];
89− sin0 = tsin[k];
90− cos1 = tcos[-k-1];
91− sin1 = tsin[-k-1];
92− CMULA(0, 0,1,2,3);
93− CMULA(1, 2,3,0,1);
94− STORE8(0);
95− STORE8(1);
96− revtabj += 4;
97− revtabk -= 4;
98− k--;
99− } while(k >= 0);
100−
101−#if HAVE_VSX
102− ff_fft_calc_vsx(s, (FFTComplex*)output);
103−#else
104− ff_fft_calc_altivec(s, (FFTComplex*)output);
105−#endif
106−
107− /* post rotation + reordering */
108− j = -n32;
109− k = n32-1;
110− do {
111− vec_f cos,sin,re,im,a,b,c,d;
112−#define CMULB(d0,d1,o)\
113− re = pout[o*2];\
114− im = pout[o*2+1];\
115− cos = tcos[o];\
116− sin = tsin[o];\
117− d0 = im*sin - re*cos;\
118− d1 = re*sin + im*cos;
119−
120− CMULB(a,b,j);
121− CMULB(c,d,k);
122− pout[2*j] = vec_perm(a, d, vcprm(0,s3,1,s2));
123− pout[2*j+1] = vec_perm(a, d, vcprm(2,s1,3,s0));
124− pout[2*k] = vec_perm(c, b, vcprm(0,s3,1,s2));
125− pout[2*k+1] = vec_perm(c, b, vcprm(2,s1,3,s0));
126− j++;
127− k--;
128− } while(k >= 0);
129−}
130−
131−static void imdct_calc_altivec(FFTContext *s, FFTSample *output, const FFTSample *input)
132−{
133− int k;
134− int n = 1 << s->mdct_bits;
135− int n4 = n >> 2;
136− int n16 = n >> 4;
137− vec_u32 sign = {1U<<31,1U<<31,1U<<31,1U<<31};
138− vec_u32 *p0 = (vec_u32*)(output+n4);
139− vec_u32 *p1 = (vec_u32*)(output+n4*3);
140−
141− imdct_half_altivec(s, output + n4, input);
142−
143− for (k = 0; k < n16; k++) {
144− vec_u32 a = p0[k] ^ sign;
145− vec_u32 b = p1[-k-1];
146− p0[-k-1] = vec_perm(a, a, vcprm(3,2,1,0));
147− p1[k] = vec_perm(b, b, vcprm(3,2,1,0));
148− }
149−}
150−#endif /* HAVE_GNU_AS && HAVE_ALTIVEC && (HAVE_BIGENDIAN || HAVE_VSX) */
151−
152−av_cold void ff_fft_init_ppc(FFTContext *s)
153−{
154−#if HAVE_GNU_AS && HAVE_ALTIVEC && (HAVE_BIGENDIAN || HAVE_VSX)
155− if (!PPC_ALTIVEC(av_get_cpu_flags()))
156− return;
157−
158−#if HAVE_VSX
159− s->fft_calc = ff_fft_calc_interleave_vsx;
160−#else
161− s->fft_calc = ff_fft_calc_interleave_altivec;
162−#endif
163− if (s->mdct_bits >= 5) {
164− s->imdct_calc = imdct_calc_altivec;
165− s->imdct_half = imdct_half_altivec;
166− }
167−#endif /* HAVE_GNU_AS && HAVE_ALTIVEC && HAVE_BIGENDIAN */
168−}
libavcodec/ppc/fft_vsx.c+0−226
@@ -1,226 +0,0 @@
1−/*
2− * FFT transform, optimized with VSX built-in functions
3− * Copyright (c) 2014 Rong Yan
4− *
5− * This algorithm (though not any of the implementation details) is
6− * based on libdjbfft by D. J. Bernstein.
7− *
8− * This file is part of FFmpeg.
9− *
10− * FFmpeg is free software; you can redistribute it and/or
11− * modify it under the terms of the GNU Lesser General Public
12− * License as published by the Free Software Foundation; either
13− * version 2.1 of the License, or (at your option) any later version.
14− *
15− * FFmpeg is distributed in the hope that it will be useful,
16− * but WITHOUT ANY WARRANTY; without even the implied warranty of
17− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18− * Lesser General Public License for more details.
19− *
20− * You should have received a copy of the GNU Lesser General Public
21− * License along with FFmpeg; if not, write to the Free Software
22− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23− */
24−
25−
26−#include "config.h"
27−#include "libavutil/cpu.h"
28−#include "libavutil/ppc/util_altivec.h"
29−#include "libavcodec/fft.h"
30−#include "libavcodec/fft-internal.h"
31−#include "fft_vsx.h"
32−
33−#if HAVE_VSX
34−
35−static void fft32_vsx_interleave(FFTComplex *z)
36−{
37− fft16_vsx_interleave(z);
38− fft8_vsx_interleave(z+16);
39− fft8_vsx_interleave(z+24);
40− pass_vsx_interleave(z,ff_cos_32,4);
41−}
42−
43−static void fft64_vsx_interleave(FFTComplex *z)
44−{
45− fft32_vsx_interleave(z);
46− fft16_vsx_interleave(z+32);
47− fft16_vsx_interleave(z+48);
48− pass_vsx_interleave(z,ff_cos_64, 8);
49−}
50−static void fft128_vsx_interleave(FFTComplex *z)
51−{
52− fft64_vsx_interleave(z);
53− fft32_vsx_interleave(z+64);
54− fft32_vsx_interleave(z+96);
55− pass_vsx_interleave(z,ff_cos_128,16);
56−}
57−static void fft256_vsx_interleave(FFTComplex *z)
58−{
59− fft128_vsx_interleave(z);
60− fft64_vsx_interleave(z+128);
61− fft64_vsx_interleave(z+192);
62− pass_vsx_interleave(z,ff_cos_256,32);
63−}
64−static void fft512_vsx_interleave(FFTComplex *z)
65−{
66− fft256_vsx_interleave(z);
67− fft128_vsx_interleave(z+256);
68− fft128_vsx_interleave(z+384);
69− pass_vsx_interleave(z,ff_cos_512,64);
70−}
71−static void fft1024_vsx_interleave(FFTComplex *z)
72−{
73− fft512_vsx_interleave(z);
74− fft256_vsx_interleave(z+512);
75− fft256_vsx_interleave(z+768);
76− pass_vsx_interleave(z,ff_cos_1024,128);
77−
78−}
79−static void fft2048_vsx_interleave(FFTComplex *z)
80−{
81− fft1024_vsx_interleave(z);
82− fft512_vsx_interleave(z+1024);
83− fft512_vsx_interleave(z+1536);
84− pass_vsx_interleave(z,ff_cos_2048,256);
85−}
86−static void fft4096_vsx_interleave(FFTComplex *z)
87−{
88− fft2048_vsx_interleave(z);
89− fft1024_vsx_interleave(z+2048);
90− fft1024_vsx_interleave(z+3072);
91− pass_vsx_interleave(z,ff_cos_4096, 512);
92−}
93−static void fft8192_vsx_interleave(FFTComplex *z)
94−{
95− fft4096_vsx_interleave(z);
96− fft2048_vsx_interleave(z+4096);
97− fft2048_vsx_interleave(z+6144);
98− pass_vsx_interleave(z,ff_cos_8192,1024);
99−}
100−static void fft16384_vsx_interleave(FFTComplex *z)
101−{
102− fft8192_vsx_interleave(z);
103− fft4096_vsx_interleave(z+8192);
104− fft4096_vsx_interleave(z+12288);
105− pass_vsx_interleave(z,ff_cos_16384,2048);
106−}
107−static void fft32768_vsx_interleave(FFTComplex *z)
108−{
109− fft16384_vsx_interleave(z);
110− fft8192_vsx_interleave(z+16384);
111− fft8192_vsx_interleave(z+24576);
112− pass_vsx_interleave(z,ff_cos_32768,4096);
113−}
114−static void fft65536_vsx_interleave(FFTComplex *z)
115−{
116− fft32768_vsx_interleave(z);
117− fft16384_vsx_interleave(z+32768);
118− fft16384_vsx_interleave(z+49152);
119− pass_vsx_interleave(z,ff_cos_65536,8192);
120−}
121−
122−static void fft32_vsx(FFTComplex *z)
123−{
124− fft16_vsx(z);
125− fft8_vsx(z+16);
126− fft8_vsx(z+24);
127− pass_vsx(z,ff_cos_32,4);
128−}
129−
130−static void fft64_vsx(FFTComplex *z)
131−{
132− fft32_vsx(z);
133− fft16_vsx(z+32);
134− fft16_vsx(z+48);
135− pass_vsx(z,ff_cos_64, 8);
136−}
137−static void fft128_vsx(FFTComplex *z)
138−{
139− fft64_vsx(z);
140− fft32_vsx(z+64);
141− fft32_vsx(z+96);
142− pass_vsx(z,ff_cos_128,16);
143−}
144−static void fft256_vsx(FFTComplex *z)
145−{
146− fft128_vsx(z);
147− fft64_vsx(z+128);
148− fft64_vsx(z+192);
149− pass_vsx(z,ff_cos_256,32);
150−}
151−static void fft512_vsx(FFTComplex *z)
152−{
153− fft256_vsx(z);
154− fft128_vsx(z+256);
155− fft128_vsx(z+384);
156− pass_vsx(z,ff_cos_512,64);
157−}
158−static void fft1024_vsx(FFTComplex *z)
159−{
160− fft512_vsx(z);
161− fft256_vsx(z+512);
162− fft256_vsx(z+768);
163− pass_vsx(z,ff_cos_1024,128);
164−
165−}
166−static void fft2048_vsx(FFTComplex *z)
167−{
168− fft1024_vsx(z);
169− fft512_vsx(z+1024);
170− fft512_vsx(z+1536);
171− pass_vsx(z,ff_cos_2048,256);
172−}
173−static void fft4096_vsx(FFTComplex *z)
174−{
175− fft2048_vsx(z);
176− fft1024_vsx(z+2048);
177− fft1024_vsx(z+3072);
178− pass_vsx(z,ff_cos_4096, 512);
179−}
180−static void fft8192_vsx(FFTComplex *z)
181−{
182− fft4096_vsx(z);
183− fft2048_vsx(z+4096);
184− fft2048_vsx(z+6144);
185− pass_vsx(z,ff_cos_8192,1024);
186−}
187−static void fft16384_vsx(FFTComplex *z)
188−{
189− fft8192_vsx(z);
190− fft4096_vsx(z+8192);
191− fft4096_vsx(z+12288);
192− pass_vsx(z,ff_cos_16384,2048);
193−}
194−static void fft32768_vsx(FFTComplex *z)
195−{
196− fft16384_vsx(z);
197− fft8192_vsx(z+16384);
198− fft8192_vsx(z+24576);
199− pass_vsx(z,ff_cos_32768,4096);
200−}
201−static void fft65536_vsx(FFTComplex *z)
202−{
203− fft32768_vsx(z);
204− fft16384_vsx(z+32768);
205− fft16384_vsx(z+49152);
206− pass_vsx(z,ff_cos_65536,8192);
207−}
208−
209−static void (* const fft_dispatch_vsx[])(FFTComplex*) = {
210− fft4_vsx, fft8_vsx, fft16_vsx, fft32_vsx, fft64_vsx, fft128_vsx, fft256_vsx, fft512_vsx, fft1024_vsx,
211− fft2048_vsx, fft4096_vsx, fft8192_vsx, fft16384_vsx, fft32768_vsx, fft65536_vsx,
212−};
213−static void (* const fft_dispatch_vsx_interleave[])(FFTComplex*) = {
214− fft4_vsx_interleave, fft8_vsx_interleave, fft16_vsx_interleave, fft32_vsx_interleave, fft64_vsx_interleave,
215− fft128_vsx_interleave, fft256_vsx_interleave, fft512_vsx_interleave, fft1024_vsx_interleave,
216− fft2048_vsx_interleave, fft4096_vsx_interleave, fft8192_vsx_interleave, fft16384_vsx_interleave, fft32768_vsx_interleave, fft65536_vsx_interleave,
217−};
218−void ff_fft_calc_interleave_vsx(FFTContext *s, FFTComplex *z)
219−{
220− fft_dispatch_vsx_interleave[s->nbits-2](z);
221−}
222−void ff_fft_calc_vsx(FFTContext *s, FFTComplex *z)
223−{
224− fft_dispatch_vsx[s->nbits-2](z);
225−}
226−#endif /* HAVE_VSX */
libavcodec/ppc/fft_vsx.h+0−829
@@ -1,829 +0,0 @@
1−#ifndef AVCODEC_PPC_FFT_VSX_H
2−#define AVCODEC_PPC_FFT_VSX_H
3−/*
4− * FFT transform, optimized with VSX built-in functions
5− * Copyright (c) 2014 Rong Yan Copyright (c) 2009 Loren Merritt
6− *
7− * This algorithm (though not any of the implementation details) is
8− * based on libdjbfft by D. J. Bernstein, and fft_altivec_s.S.
9− *
10− * This file is part of FFmpeg.
11− *
12− * FFmpeg is free software; you can redistribute it and/or
13− * modify it under the terms of the GNU Lesser General Public
14− * License as published by the Free Software Foundation; either
15− * version 2.1 of the License, or (at your option) any later version.
16− *
17− * FFmpeg is distributed in the hope that it will be useful,
18− * but WITHOUT ANY WARRANTY; without even the implied warranty of
19− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20− * Lesser General Public License for more details.
21− *
22− * You should have received a copy of the GNU Lesser General Public
23− * License along with FFmpeg; if not, write to the Free Software
24− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25− */
26−
27−
28−#include "config.h"
29−#include "libavutil/cpu.h"
30−#include "libavutil/ppc/util_altivec.h"
31−#include "libavcodec/fft.h"
32−#include "libavcodec/fft-internal.h"
33−
34−#if HAVE_VSX
35−
36−void ff_fft_calc_interleave_vsx(FFTContext *s, FFTComplex *z);
37−void ff_fft_calc_vsx(FFTContext *s, FFTComplex *z);
38−
39−
40−#define byte_2complex (2*sizeof(FFTComplex))
41−#define byte_4complex (4*sizeof(FFTComplex))
42−#define byte_6complex (6*sizeof(FFTComplex))
43−#define byte_8complex (8*sizeof(FFTComplex))
44−#define byte_10complex (10*sizeof(FFTComplex))
45−#define byte_12complex (12*sizeof(FFTComplex))
46−#define byte_14complex (14*sizeof(FFTComplex))
47−
48−inline static void pass_vsx_interleave(FFTComplex *z, const FFTSample *wre, unsigned int n)
49−{
50− int o1 = n<<1;
51− int o2 = n<<2;
52− int o3 = o1+o2;
53− int i1, i2, i3;
54− FFTSample* out = (FFTSample*)z;
55− const FFTSample *wim = wre+o1;
56− vec_f vz0, vzo1, vzo2, vzo3;
57− vec_f x0, x1, x2, x3;
58− vec_f x4, x5, x6, x7;
59− vec_f x8, x9, x10, x11;
60− vec_f x12, x13, x14, x15;
61− vec_f x16, x17, x18, x19;
62− vec_f x20, x21, x22, x23;
63− vec_f vz0plus1, vzo1plus1, vzo2plus1, vzo3plus1;
64− vec_f y0, y1, y2, y3;
65− vec_f y4, y5, y8, y9;
66− vec_f y10, y13, y14, y15;
67− vec_f y16, y17, y18, y19;
68− vec_f y20, y21, y22, y23;
69− vec_f wr1, wi1, wr0, wi0;
70− vec_f wr2, wi2, wr3, wi3;
71− vec_f xmulwi0, xmulwi1, ymulwi2, ymulwi3;
72−
73− n = n-2;
74− i1 = o1*sizeof(FFTComplex);
75− i2 = o2*sizeof(FFTComplex);
76− i3 = o3*sizeof(FFTComplex);
77− vzo2 = vec_ld(i2, &(out[0])); // zo2.r zo2.i z(o2+1).r z(o2+1).i
78− vzo2plus1 = vec_ld(i2+16, &(out[0]));
79− vzo3 = vec_ld(i3, &(out[0])); // zo3.r zo3.i z(o3+1).r z(o3+1).i
80− vzo3plus1 = vec_ld(i3+16, &(out[0]));
81− vz0 = vec_ld(0, &(out[0])); // z0.r z0.i z1.r z1.i
82− vz0plus1 = vec_ld(16, &(out[0]));
83− vzo1 = vec_ld(i1, &(out[0])); // zo1.r zo1.i z(o1+1).r z(o1+1).i
84− vzo1plus1 = vec_ld(i1+16, &(out[0]));
85−
86− x0 = vec_add(vzo2, vzo3);
87− x1 = vec_sub(vzo2, vzo3);
88− y0 = vec_add(vzo2plus1, vzo3plus1);
89− y1 = vec_sub(vzo2plus1, vzo3plus1);
90−
91− wr1 = vec_splats(wre[1]);
92− wi1 = vec_splats(wim[-1]);
93− wi2 = vec_splats(wim[-2]);
94− wi3 = vec_splats(wim[-3]);
95− wr2 = vec_splats(wre[2]);
96− wr3 = vec_splats(wre[3]);
97−
98− x2 = vec_perm(x0, x1, vcprm(2,s2,3,s3));
99− x3 = vec_perm(x0, x1, vcprm(s3,3,s2,2));
100−
101− y4 = vec_perm(y0, y1, vcprm(s1,1,s0,0));
102− y5 = vec_perm(y0, y1, vcprm(s3,3,s2,2));
103− y2 = vec_perm(y0, y1, vcprm(0,s0,1,s1));
104− y3 = vec_perm(y0, y1, vcprm(2,s2,3,s3));
105−
106− ymulwi2 = vec_mul(y4, wi2);
107− ymulwi3 = vec_mul(y5, wi3);
108− x4 = vec_mul(x2, wr1);
109− x5 = vec_mul(x3, wi1);
110− y8 = vec_madd(y2, wr2, ymulwi2);
111− y9 = vec_msub(y2, wr2, ymulwi2);
112− x6 = vec_add(x4, x5);
113− x7 = vec_sub(x4, x5);
114− y13 = vec_madd(y3, wr3, ymulwi3);
115− y14 = vec_msub(y3, wr3, ymulwi3);
116−
117− x8 = vec_perm(x6, x7, vcprm(0,1,s2,s3));
118− y10 = vec_perm(y8, y9, vcprm(0,1,s2,s3));
119− y15 = vec_perm(y13, y14, vcprm(0,1,s2,s3));
120−
121− x9 = vec_perm(x0, x8, vcprm(0,1,s0,s2));
122− x10 = vec_perm(x1, x8, vcprm(1,0,s3,s1));
123−
124− y16 = vec_perm(y10, y15, vcprm(0,2,s0,s2));
125− y17 = vec_perm(y10, y15, vcprm(3,1,s3,s1));
126−
127− x11 = vec_add(vz0, x9);
128− x12 = vec_sub(vz0, x9);
129− x13 = vec_add(vzo1, x10);
130− x14 = vec_sub(vzo1, x10);
131−
132− y18 = vec_add(vz0plus1, y16);
133− y19 = vec_sub(vz0plus1, y16);
134− y20 = vec_add(vzo1plus1, y17);
135− y21 = vec_sub(vzo1plus1, y17);
136−
137− x15 = vec_perm(x13, x14, vcprm(0,s1,2,s3));
138− x16 = vec_perm(x13, x14, vcprm(s0,1,s2,3));
139− y22 = vec_perm(y20, y21, vcprm(0,s1,2,s3));
140− y23 = vec_perm(y20, y21, vcprm(s0,1,s2,3));
141−
142−
143− vec_st(x11, 0, &(out[0]));
144− vec_st(y18, 16, &(out[0]));
145− vec_st(x15, i1, &(out[0]));
146− vec_st(y22, i1+16, &(out[0]));
147− vec_st(x12, i2, &(out[0]));
148− vec_st(y19, i2+16, &(out[0]));
149− vec_st(x16, i3, &(out[0]));
150− vec_st(y23, i3+16, &(out[0]));
151−
152− do {
153− out += 8;
154− wre += 4;
155− wim -= 4;
156− wr0 = vec_splats(wre[0]);
157− wr1 = vec_splats(wre[1]);
158− wi0 = vec_splats(wim[0]);
159− wi1 = vec_splats(wim[-1]);
160−
161− wr2 = vec_splats(wre[2]);
162− wr3 = vec_splats(wre[3]);
163− wi2 = vec_splats(wim[-2]);
164− wi3 = vec_splats(wim[-3]);
165−
166− vzo2 = vec_ld(i2, &(out[0])); // zo2.r zo2.i z(o2+1).r z(o2+1).i
167− vzo2plus1 = vec_ld(i2+16, &(out[0]));
168− vzo3 = vec_ld(i3, &(out[0])); // zo3.r zo3.i z(o3+1).r z(o3+1).i
169− vzo3plus1 = vec_ld(i3+16, &(out[0]));
170− vz0 = vec_ld(0, &(out[0])); // z0.r z0.i z1.r z1.i
171− vz0plus1 = vec_ld(16, &(out[0]));
172− vzo1 = vec_ld(i1, &(out[0])); // zo1.r zo1.i z(o1+1).r z(o1+1).i
173− vzo1plus1 = vec_ld(i1+16, &(out[0]));
174−
175− x0 = vec_add(vzo2, vzo3);
176− x1 = vec_sub(vzo2, vzo3);
177−
178− y0 = vec_add(vzo2plus1, vzo3plus1);
179− y1 = vec_sub(vzo2plus1, vzo3plus1);
180−
181− x4 = vec_perm(x0, x1, vcprm(s1,1,s0,0));
182− x5 = vec_perm(x0, x1, vcprm(s3,3,s2,2));
183− x2 = vec_perm(x0, x1, vcprm(0,s0,1,s1));
184− x3 = vec_perm(x0, x1, vcprm(2,s2,3,s3));
185−
186− y2 = vec_perm(y0, y1, vcprm(0,s0,1,s1));
187− y3 = vec_perm(y0, y1, vcprm(2,s2,3,s3));
188− xmulwi0 = vec_mul(x4, wi0);
189− xmulwi1 = vec_mul(x5, wi1);
190−
191− y4 = vec_perm(y0, y1, vcprm(s1,1,s0,0));
192− y5 = vec_perm(y0, y1, vcprm(s3,3,s2,2));
193−
194− x8 = vec_madd(x2, wr0, xmulwi0);
195− x9 = vec_msub(x2, wr0, xmulwi0);
196− ymulwi2 = vec_mul(y4, wi2);
197− ymulwi3 = vec_mul(y5, wi3);
198−
199− x13 = vec_madd(x3, wr1, xmulwi1);
200− x14 = vec_msub(x3, wr1, xmulwi1);
201−
202− y8 = vec_madd(y2, wr2, ymulwi2);
203− y9 = vec_msub(y2, wr2, ymulwi2);
204− y13 = vec_madd(y3, wr3, ymulwi3);
205− y14 = vec_msub(y3, wr3, ymulwi3);
206−
207− x10 = vec_perm(x8, x9, vcprm(0,1,s2,s3));
208− x15 = vec_perm(x13, x14, vcprm(0,1,s2,s3));
209−
210− y10 = vec_perm(y8, y9, vcprm(0,1,s2,s3));
211− y15 = vec_perm(y13, y14, vcprm(0,1,s2,s3));
212−
213− x16 = vec_perm(x10, x15, vcprm(0,2,s0,s2));
214− x17 = vec_perm(x10, x15, vcprm(3,1,s3,s1));
215−
216− y16 = vec_perm(y10, y15, vcprm(0,2,s0,s2));
217− y17 = vec_perm(y10, y15, vcprm(3,1,s3,s1));
218−
219− x18 = vec_add(vz0, x16);
220− x19 = vec_sub(vz0, x16);
221− x20 = vec_add(vzo1, x17);
222− x21 = vec_sub(vzo1, x17);
223−
224− y18 = vec_add(vz0plus1, y16);
225− y19 = vec_sub(vz0plus1, y16);
226− y20 = vec_add(vzo1plus1, y17);
227− y21 = vec_sub(vzo1plus1, y17);
228−
229− x22 = vec_perm(x20, x21, vcprm(0,s1,2,s3));
230− x23 = vec_perm(x20, x21, vcprm(s0,1,s2,3));
231−
232− y22 = vec_perm(y20, y21, vcprm(0,s1,2,s3));
233− y23 = vec_perm(y20, y21, vcprm(s0,1,s2,3));
234−
235− vec_st(x18, 0, &(out[0]));
236− vec_st(y18, 16, &(out[0]));
237− vec_st(x22, i1, &(out[0]));
238− vec_st(y22, i1+16, &(out[0]));
239− vec_st(x19, i2, &(out[0]));
240− vec_st(y19, i2+16, &(out[0]));
241− vec_st(x23, i3, &(out[0]));
242− vec_st(y23, i3+16, &(out[0]));
243− } while (n-=2);
244−}
245−
246−inline static void fft2_vsx_interleave(FFTComplex *z)
247−{
248− FFTSample r1, i1;
249−
250− r1 = z[0].re - z[1].re;
251− z[0].re += z[1].re;
252− z[1].re = r1;
253−
254− i1 = z[0].im - z[1].im;
255− z[0].im += z[1].im;
256− z[1].im = i1;
257− }
258−
259−inline static void fft4_vsx_interleave(FFTComplex *z)
260−{
261− vec_f a, b, c, d;
262− float* out= (float*)z;
263− a = vec_ld(0, &(out[0]));
264− b = vec_ld(byte_2complex, &(out[0]));
265−
266− c = vec_perm(a, b, vcprm(0,1,s2,s1));
267− d = vec_perm(a, b, vcprm(2,3,s0,s3));
268− a = vec_add(c, d);
269− b = vec_sub(c, d);
270−
271− c = vec_perm(a, b, vcprm(0,1,s0,s1));
272− d = vec_perm(a, b, vcprm(2,3,s3,s2));
273−
274− a = vec_add(c, d);
275− b = vec_sub(c, d);
276− vec_st(a, 0, &(out[0]));
277− vec_st(b, byte_2complex, &(out[0]));
278−}
279−
280−inline static void fft8_vsx_interleave(FFTComplex *z)
281−{
282− vec_f vz0, vz1, vz2, vz3;
283− vec_f x0, x1, x2, x3;
284− vec_f x4, x5, x6, x7;
285− vec_f x8, x9, x10, x11;
286− vec_f x12, x13, x14, x15;
287− vec_f x16, x17, x18, x19;
288− vec_f x20, x21, x22, x23;
289− vec_f x24, x25, x26, x27;
290− vec_f x28, x29, x30, x31;
291− vec_f x32, x33, x34;
292−
293− float* out= (float*)z;
294− vec_f vc1 = {sqrthalf, sqrthalf, sqrthalf, sqrthalf};
295−
296− vz0 = vec_ld(0, &(out[0]));
297− vz1 = vec_ld(byte_2complex, &(out[0]));
298− vz2 = vec_ld(byte_4complex, &(out[0]));
299− vz3 = vec_ld(byte_6complex, &(out[0]));
300−
301− x0 = vec_perm(vz0, vz1, vcprm(0,1,s2,s1));
302− x1 = vec_perm(vz0, vz1, vcprm(2,3,s0,s3));
303− x2 = vec_perm(vz2, vz3, vcprm(2,1,s0,s1));
304− x3 = vec_perm(vz2, vz3, vcprm(0,3,s2,s3));
305−
306− x4 = vec_add(x0, x1);
307− x5 = vec_sub(x0, x1);
308− x6 = vec_add(x2, x3);
309− x7 = vec_sub(x2, x3);
310−
311− x8 = vec_perm(x4, x5, vcprm(0,1,s0,s1));
312− x9 = vec_perm(x4, x5, vcprm(2,3,s3,s2));
313− x10 = vec_perm(x6, x7, vcprm(2,1,s2,s1));
314− x11 = vec_perm(x6, x7, vcprm(0,3,s0,s3));
315−
316− x12 = vec_add(x8, x9);
317− x13 = vec_sub(x8, x9);
318− x14 = vec_add(x10, x11);
319− x15 = vec_sub(x10, x11);
320− x16 = vec_perm(x12, x13, vcprm(0,s0,1,s1));
321− x17 = vec_perm(x14, x15, vcprm(0,s0,1,s1));
322− x18 = vec_perm(x16, x17, vcprm(s0,s3,s2,s1));
323− x19 = vec_add(x16, x18); // z0.r z2.r z0.i z2.i
324− x20 = vec_sub(x16, x18); // z4.r z6.r z4.i z6.i
325−
326− x21 = vec_perm(x12, x13, vcprm(2,s2,3,s3));
327− x22 = vec_perm(x14, x15, vcprm(2,3,s2,s3));
328− x23 = vec_perm(x14, x15, vcprm(3,2,s3,s2));
329− x24 = vec_add(x22, x23);
330− x25 = vec_sub(x22, x23);
331− x26 = vec_mul( vec_perm(x24, x25, vcprm(2,s2,0,s0)), vc1);
332−
333− x27 = vec_add(x21, x26); // z1.r z7.r z1.i z3.i
334− x28 = vec_sub(x21, x26); //z5.r z3.r z5.i z7.i
335−
336− x29 = vec_perm(x19, x27, vcprm(0,2,s0,s2)); // z0.r z0.i z1.r z1.i
337− x30 = vec_perm(x19, x27, vcprm(1,3,s1,s3)); // z2.r z2.i z7.r z3.i
338− x31 = vec_perm(x20, x28, vcprm(0,2,s0,s2)); // z4.r z4.i z5.r z5.i
339− x32 = vec_perm(x20, x28, vcprm(1,3,s1,s3)); // z6.r z6.i z3.r z7.i
340− x33 = vec_perm(x30, x32, vcprm(0,1,s2,3)); // z2.r z2.i z3.r z3.i
341− x34 = vec_perm(x30, x32, vcprm(s0,s1,2,s3)); // z6.r z6.i z7.r z7.i
342−
343− vec_st(x29, 0, &(out[0]));
344− vec_st(x33, byte_2complex, &(out[0]));
345− vec_st(x31, byte_4complex, &(out[0]));
346− vec_st(x34, byte_6complex, &(out[0]));
347−}
348−
349−inline static void fft16_vsx_interleave(FFTComplex *z)
350−{
351− float* out= (float*)z;
352− vec_f vc0 = {sqrthalf, sqrthalf, sqrthalf, sqrthalf};
353− vec_f vc1 = {ff_cos_16[1], ff_cos_16[1], ff_cos_16[1], ff_cos_16[1]};
354− vec_f vc2 = {ff_cos_16[3], ff_cos_16[3], ff_cos_16[3], ff_cos_16[3]};
355− vec_f vz0, vz1, vz2, vz3;
356− vec_f vz4, vz5, vz6, vz7;
357− vec_f x0, x1, x2, x3;
358− vec_f x4, x5, x6, x7;
359− vec_f x8, x9, x10, x11;
360− vec_f x12, x13, x14, x15;
361− vec_f x16, x17, x18, x19;
362− vec_f x20, x21, x22, x23;
363− vec_f x24, x25, x26, x27;
364− vec_f x28, x29, x30, x31;
365− vec_f x32, x33, x34, x35;
366− vec_f x36, x37, x38, x39;
367− vec_f x40, x41, x42, x43;
368− vec_f x44, x45, x46, x47;
369− vec_f x48, x49, x50, x51;
370− vec_f x52, x53, x54, x55;
371− vec_f x56, x57, x58, x59;
372− vec_f x60, x61, x62, x63;
373− vec_f x64, x65, x66, x67;
374− vec_f x68, x69, x70, x71;
375− vec_f x72, x73, x74, x75;
376− vec_f x76, x77, x78, x79;
377− vec_f x80, x81, x82, x83;
378− vec_f x84, x85, x86;
379−
380− vz0 = vec_ld(0, &(out[0]));
381− vz1 = vec_ld(byte_2complex, &(out[0]));
382− vz2 = vec_ld(byte_4complex, &(out[0]));
383− vz3 = vec_ld(byte_6complex, &(out[0]));
384− vz4 = vec_ld(byte_8complex, &(out[0]));
385− vz5 = vec_ld(byte_10complex, &(out[0]));
386− vz6 = vec_ld(byte_12complex, &(out[0]));
387− vz7 = vec_ld(byte_14complex, &(out[0]));
388−
389− x0 = vec_perm(vz0, vz1, vcprm(0,1,s2,s1));
390− x1 = vec_perm(vz0, vz1, vcprm(2,3,s0,s3));
391− x2 = vec_perm(vz2, vz3, vcprm(0,1,s0,s1));
392− x3 = vec_perm(vz2, vz3, vcprm(2,3,s2,s3));
393−
394− x4 = vec_perm(vz4, vz5, vcprm(0,1,s2,s1));
395− x5 = vec_perm(vz4, vz5, vcprm(2,3,s0,s3));
396− x6 = vec_perm(vz6, vz7, vcprm(0,1,s2,s1));
397− x7 = vec_perm(vz6, vz7, vcprm(2,3,s0,s3));
398−
399− x8 = vec_add(x0, x1);
400− x9 = vec_sub(x0, x1);
401− x10 = vec_add(x2, x3);
402− x11 = vec_sub(x2, x3);
403−
404− x12 = vec_add(x4, x5);
405− x13 = vec_sub(x4, x5);
406− x14 = vec_add(x6, x7);
407− x15 = vec_sub(x6, x7);
408−
409− x16 = vec_perm(x8, x9, vcprm(0,1,s0,s1));
410− x17 = vec_perm(x8, x9, vcprm(2,3,s3,s2));
411− x18 = vec_perm(x10, x11, vcprm(2,1,s1,s2));
412− x19 = vec_perm(x10, x11, vcprm(0,3,s0,s3));
413− x20 = vec_perm(x12, x14, vcprm(0,1,s0, s1));
414− x21 = vec_perm(x12, x14, vcprm(2,3,s2,s3));
415− x22 = vec_perm(x13, x15, vcprm(0,1,s0,s1));
416− x23 = vec_perm(x13, x15, vcprm(3,2,s3,s2));
417−
418− x24 = vec_add(x16, x17);
419− x25 = vec_sub(x16, x17);
420− x26 = vec_add(x18, x19);
421− x27 = vec_sub(x18, x19);
422− x28 = vec_add(x20, x21);
423− x29 = vec_sub(x20, x21);
424− x30 = vec_add(x22, x23);
425− x31 = vec_sub(x22, x23);
426−
427− x32 = vec_add(x24, x26);
428− x33 = vec_sub(x24, x26);
429− x34 = vec_perm(x32, x33, vcprm(0,1,s0,s1));
430−
431− x35 = vec_perm(x28, x29, vcprm(2,1,s1,s2));
432− x36 = vec_perm(x28, x29, vcprm(0,3,s0,s3));
433− x37 = vec_add(x35, x36);
434− x38 = vec_sub(x35, x36);
435− x39 = vec_perm(x37, x38, vcprm(0,1,s1,s0));
436−
437− x40 = vec_perm(x27, x38, vcprm(3,2,s2,s3));
438− x41 = vec_perm(x26, x37, vcprm(2,3,s3,s2));
439− x42 = vec_add(x40, x41);
440− x43 = vec_sub(x40, x41);
441− x44 = vec_mul(x42, vc0);
442− x45 = vec_mul(x43, vc0);
443−
444− x46 = vec_add(x34, x39); // z0.r z0.i z4.r z4.i
445− x47 = vec_sub(x34, x39); // z8.r z8.i z12.r z12.i
446−
447− x48 = vec_perm(x30, x31, vcprm(2,1,s1,s2));
448− x49 = vec_perm(x30, x31, vcprm(0,3,s3,s0));
449− x50 = vec_add(x48, x49);
450− x51 = vec_sub(x48, x49);
451− x52 = vec_mul(x50, vc1);
452− x53 = vec_mul(x50, vc2);
453− x54 = vec_mul(x51, vc1);
454− x55 = vec_mul(x51, vc2);
455−
456− x56 = vec_perm(x24, x25, vcprm(2,3,s2,s3));
457− x57 = vec_perm(x44, x45, vcprm(0,1,s1,s0));
458− x58 = vec_add(x56, x57);
459− x59 = vec_sub(x56, x57);
460−
461− x60 = vec_perm(x54, x55, vcprm(1,0,3,2));
462− x61 = vec_perm(x54, x55, vcprm(s1,s0,s3,s2));
463− x62 = vec_add(x52, x61);
464− x63 = vec_sub(x52, x61);
465− x64 = vec_add(x60, x53);
466− x65 = vec_sub(x60, x53);
467− x66 = vec_perm(x62, x64, vcprm(0,1,s3,s2));
468− x67 = vec_perm(x63, x65, vcprm(s0,s1,3,2));
469−
470− x68 = vec_add(x58, x66); // z1.r z1.i z3.r z3.i
471− x69 = vec_sub(x58, x66); // z9.r z9.i z11.r z11.i
472− x70 = vec_add(x59, x67); // z5.r z5.i z15.r z15.i
473− x71 = vec_sub(x59, x67); // z13.r z13.i z7.r z7.i
474−
475− x72 = vec_perm(x25, x27, vcprm(s1,s0,s2,s3));
476− x73 = vec_add(x25, x72);
477− x74 = vec_sub(x25, x72);
478− x75 = vec_perm(x73, x74, vcprm(0,1,s0,s1));
479− x76 = vec_perm(x44, x45, vcprm(3,2,s2,s3));
480− x77 = vec_add(x75, x76); // z2.r z2.i z6.r z6.i
481− x78 = vec_sub(x75, x76); // z10.r z10.i z14.r z14.i
482−
483− x79 = vec_perm(x46, x68, vcprm(0,1,s0,s1)); // z0.r z0.i z1.r z1.i
484− x80 = vec_perm(x77, x68, vcprm(0,1,s2,s3)); // z2.r z2.i z3.r z3.i
485− x81 = vec_perm(x46, x70, vcprm(2,3,s0,s1)); // z4.r z4.i z5.r z5.i
486− x82 = vec_perm(x71, x77, vcprm(s2,s3,2,3)); // z6.r z6.i z7.r z7.i
487− vec_st(x79, 0, &(out[0]));
488− vec_st(x80, byte_2complex, &(out[0]));
489− vec_st(x81, byte_4complex, &(out[0]));
490− vec_st(x82, byte_6complex, &(out[0]));
491− x83 = vec_perm(x47, x69, vcprm(0,1,s0,s1)); // z8.r z8.i z9.r z9.i
492− x84 = vec_perm(x78, x69, vcprm(0,1,s2,s3)); // z10.r z10.i z11.r z11.i
493− x85 = vec_perm(x47, x71, vcprm(2,3,s0,s1)); // z12.r z12.i z13.r z13.i
494− x86 = vec_perm(x70, x78, vcprm(s2,s3,2,3)); // z14.r z14.i z15.r z15.i
495− vec_st(x83, byte_8complex, &(out[0]));
496− vec_st(x84, byte_10complex, &(out[0]));
497− vec_st(x85, byte_12complex, &(out[0]));
498− vec_st(x86, byte_14complex, &(out[0]));
499−}
500−
501−inline static void fft4_vsx(FFTComplex *z)
502−{
503− vec_f a, b, c, d;
504− float* out= (float*)z;
505− a = vec_ld(0, &(out[0]));
506− b = vec_ld(byte_2complex, &(out[0]));
507−
508− c = vec_perm(a, b, vcprm(0,1,s2,s1));
509− d = vec_perm(a, b, vcprm(2,3,s0,s3));
510− a = vec_add(c, d);
511− b = vec_sub(c, d);
512−
513− c = vec_perm(a,b, vcprm(0,s0,1,s1));
514− d = vec_perm(a, b, vcprm(2,s3,3,s2));
515−
516− a = vec_add(c, d);
517− b = vec_sub(c, d);
518−
519− c = vec_perm(a, b, vcprm(0,1,s0,s1));
520− d = vec_perm(a, b, vcprm(2,3,s2,s3));
521−
522− vec_st(c, 0, &(out[0]));
523− vec_st(d, byte_2complex, &(out[0]));
524− return;
525−}
526−
527−inline static void fft8_vsx(FFTComplex *z)
528−{
529− vec_f vz0, vz1, vz2, vz3;
530− vec_f vz4, vz5, vz6, vz7, vz8;
531−
532− float* out= (float*)z;
533− vec_f vc0 = {0.0, 0.0, 0.0, 0.0};
534− vec_f vc1 = {-sqrthalf, sqrthalf, sqrthalf, -sqrthalf};
535− vec_f vc2 = {sqrthalf, sqrthalf, sqrthalf, sqrthalf};
536−
537− vz0 = vec_ld(0, &(out[0]));
538− vz1 = vec_ld(byte_2complex, &(out[0]));
539− vz2 = vec_ld(byte_4complex, &(out[0]));
540− vz3 = vec_ld(byte_6complex, &(out[0]));
541−
542− vz6 = vec_perm(vz2, vz3, vcprm(0,s0,1,s1));
543− vz7 = vec_perm(vz2, vz3, vcprm(2,s2,3,s3));
544− vz4 = vec_perm(vz0, vz1, vcprm(0,1,s2,s1));
545− vz5 = vec_perm(vz0, vz1, vcprm(2,3,s0,s3));
546−
547− vz2 = vec_add(vz6, vz7);
548− vz3 = vec_sub(vz6, vz7);
549− vz8 = vec_perm(vz3, vz3, vcprm(2,3,0,1));
550−
551− vz0 = vec_add(vz4, vz5);
552− vz1 = vec_sub(vz4, vz5);
553−
554− vz3 = vec_madd(vz3, vc1, vc0);
555− vz3 = vec_madd(vz8, vc2, vz3);
556−
557− vz4 = vec_perm(vz0, vz1, vcprm(0,s0,1,s1));
558− vz5 = vec_perm(vz0, vz1, vcprm(2,s3,3,s2));
559− vz6 = vec_perm(vz2, vz3, vcprm(1,2,s3,s0));
560− vz7 = vec_perm(vz2, vz3, vcprm(0,3,s2,s1));
561−
562− vz0 = vec_add(vz4, vz5);
563− vz1 = vec_sub(vz4, vz5);
564− vz2 = vec_add(vz6, vz7);
565− vz3 = vec_sub(vz6, vz7);
566−
567− vz4 = vec_perm(vz0, vz1, vcprm(0,1,s0,s1));
568− vz5 = vec_perm(vz0, vz1, vcprm(2,3,s2,s3));
569− vz6 = vec_perm(vz2, vz3, vcprm(0,2,s1,s3));
570− vz7 = vec_perm(vz2, vz3, vcprm(1,3,s0,s2));
571−
572−
573− vz2 = vec_sub(vz4, vz6);
574− vz3 = vec_sub(vz5, vz7);
575−
576− vz0 = vec_add(vz4, vz6);
577− vz1 = vec_add(vz5, vz7);
578−
579− vec_st(vz0, 0, &(out[0]));
580− vec_st(vz1, byte_2complex, &(out[0]));
581− vec_st(vz2, byte_4complex, &(out[0]));
582− vec_st(vz3, byte_6complex, &(out[0]));
583− return;
584−}
585−
586−inline static void fft16_vsx(FFTComplex *z)
587−{
588− float* out= (float*)z;
589− vec_f vc0 = {0.0, 0.0, 0.0, 0.0};
590− vec_f vc1 = {-sqrthalf, sqrthalf, sqrthalf, -sqrthalf};
591− vec_f vc2 = {sqrthalf, sqrthalf, sqrthalf, sqrthalf};
592− vec_f vc3 = {1.0, 0.92387953, sqrthalf, 0.38268343};
593− vec_f vc4 = {0.0, 0.38268343, sqrthalf, 0.92387953};
594− vec_f vc5 = {-0.0, -0.38268343, -sqrthalf, -0.92387953};
595−
596− vec_f vz0, vz1, vz2, vz3;
597− vec_f vz4, vz5, vz6, vz7;
598− vec_f vz8, vz9, vz10, vz11;
599− vec_f vz12, vz13;
600−
601− vz0 = vec_ld(byte_8complex, &(out[0]));
602− vz1 = vec_ld(byte_10complex, &(out[0]));
603− vz2 = vec_ld(byte_12complex, &(out[0]));
604− vz3 = vec_ld(byte_14complex, &(out[0]));
605−
606− vz4 = vec_perm(vz0, vz1, vcprm(0,1,s2,s1));
607− vz5 = vec_perm(vz0, vz1, vcprm(2,3,s0,s3));
608− vz6 = vec_perm(vz2, vz3, vcprm(0,1,s2,s1));
609− vz7 = vec_perm(vz2, vz3, vcprm(2,3,s0,s3));
610−
611− vz0 = vec_add(vz4, vz5);
612− vz1= vec_sub(vz4, vz5);
613− vz2 = vec_add(vz6, vz7);
614− vz3 = vec_sub(vz6, vz7);
615−
616− vz4 = vec_perm(vz0, vz1, vcprm(0,s0,1,s1));
617− vz5 = vec_perm(vz0, vz1, vcprm(2,s3,3,s2));
618− vz6 = vec_perm(vz2, vz3, vcprm(0,s0,1,s1));
619− vz7 = vec_perm(vz2, vz3, vcprm(2,s3,3,s2));
620−
621− vz0 = vec_add(vz4, vz5);
622− vz1 = vec_sub(vz4, vz5);
623− vz2 = vec_add(vz6, vz7);
624− vz3 = vec_sub(vz6, vz7);
625−
626− vz4 = vec_perm(vz0, vz1, vcprm(0,1,s0,s1));
627− vz5 = vec_perm(vz0, vz1, vcprm(2,3,s2,s3));
628−
629− vz6 = vec_perm(vz2, vz3, vcprm(0,1,s0,s1));
630− vz7 = vec_perm(vz2, vz3, vcprm(2,3,s2,s3));
631−
632− vz0 = vec_ld(0, &(out[0]));
633− vz1 = vec_ld(byte_2complex, &(out[0]));
634− vz2 = vec_ld(byte_4complex, &(out[0]));
635− vz3 = vec_ld(byte_6complex, &(out[0]));
636− vz10 = vec_perm(vz2, vz3, vcprm(0,s0,1,s1));
637− vz11 = vec_perm(vz2, vz3, vcprm(2,s2,3,s3));
638− vz8 = vec_perm(vz0, vz1, vcprm(0,1,s2,s1));
639− vz9 = vec_perm(vz0, vz1, vcprm(2,3,s0,s3));
640−
641− vz2 = vec_add(vz10, vz11);
642− vz3 = vec_sub(vz10, vz11);
643− vz12 = vec_perm(vz3, vz3, vcprm(2,3,0,1));
644− vz0 = vec_add(vz8, vz9);
645− vz1 = vec_sub(vz8, vz9);
646−
647− vz3 = vec_madd(vz3, vc1, vc0);
648− vz3 = vec_madd(vz12, vc2, vz3);
649− vz8 = vec_perm(vz0, vz1, vcprm(0,s0,1,s1));
650− vz9 = vec_perm(vz0, vz1, vcprm(2,s3,3,s2));
651− vz10 = vec_perm(vz2, vz3, vcprm(1,2,s3,s0));
652− vz11 = vec_perm(vz2, vz3, vcprm(0,3,s2,s1));
653−
654− vz0 = vec_add(vz8, vz9);
655− vz1 = vec_sub(vz8, vz9);
656− vz2 = vec_add(vz10, vz11);
657− vz3 = vec_sub(vz10, vz11);
658−
659− vz8 = vec_perm(vz0, vz1, vcprm(0,1,s0,s1));
660− vz9 = vec_perm(vz0, vz1, vcprm(2,3,s2,s3));
661− vz10 = vec_perm(vz2, vz3, vcprm(0,2,s1,s3));
662− vz11 = vec_perm(vz2, vz3, vcprm(1,3,s0,s2));
663−
664− vz2 = vec_sub(vz8, vz10);
665− vz3 = vec_sub(vz9, vz11);
666− vz0 = vec_add(vz8, vz10);
667− vz1 = vec_add(vz9, vz11);
668−
669− vz8 = vec_madd(vz4, vc3, vc0);
670− vz9 = vec_madd(vz5, vc3, vc0);
671− vz10 = vec_madd(vz6, vc3, vc0);
672− vz11 = vec_madd(vz7, vc3, vc0);
673−
674− vz8 = vec_madd(vz5, vc4, vz8);
675− vz9 = vec_madd(vz4, vc5, vz9);
676− vz10 = vec_madd(vz7, vc5, vz10);
677− vz11 = vec_madd(vz6, vc4, vz11);
678−
679− vz12 = vec_sub(vz10, vz8);
680− vz10 = vec_add(vz10, vz8);
681−
682− vz13 = vec_sub(vz9, vz11);
683− vz11 = vec_add(vz9, vz11);
684−
685− vz4 = vec_sub(vz0, vz10);
686− vz0 = vec_add(vz0, vz10);
687−
688− vz7= vec_sub(vz3, vz12);
689− vz3= vec_add(vz3, vz12);
690−
691− vz5 = vec_sub(vz1, vz11);
692− vz1 = vec_add(vz1, vz11);
693−
694− vz6 = vec_sub(vz2, vz13);
695− vz2 = vec_add(vz2, vz13);
696−
697− vec_st(vz0, 0, &(out[0]));
698− vec_st(vz1, byte_2complex, &(out[0]));
699− vec_st(vz2, byte_4complex, &(out[0]));
700− vec_st(vz3, byte_6complex, &(out[0]));
701− vec_st(vz4, byte_8complex, &(out[0]));
702− vec_st(vz5, byte_10complex, &(out[0]));
703− vec_st(vz6, byte_12complex, &(out[0]));
704− vec_st(vz7, byte_14complex, &(out[0]));
705− return;
706−
707−}
708−inline static void pass_vsx(FFTComplex * z, const FFTSample * wre, unsigned int n)
709−{
710− int o1 = n<<1;
711− int o2 = n<<2;
712− int o3 = o1+o2;
713− int i1, i2, i3;
714− FFTSample* out = (FFTSample*)z;
715− const FFTSample *wim = wre+o1;
716− vec_f v0, v1, v2, v3;
717− vec_f v4, v5, v6, v7;
718− vec_f v8, v9, v10, v11;
719− vec_f v12, v13;
720−
721− n = n-2;
722− i1 = o1*sizeof(FFTComplex);
723− i2 = o2*sizeof(FFTComplex);
724− i3 = o3*sizeof(FFTComplex);
725−
726− v8 = vec_ld(0, &(wre[0]));
727− v10 = vec_ld(0, &(wim[0]));
728− v9 = vec_ld(0, &(wim[-4]));
729− v9 = vec_perm(v9, v10, vcprm(s0,3,2,1));
730−
731− v4 = vec_ld(i2, &(out[0]));
732− v5 = vec_ld(i2+16, &(out[0]));
733− v6 = vec_ld(i3, &(out[0]));
734− v7 = vec_ld(i3+16, &(out[0]));
735− v10 = vec_mul(v4, v8); // r2*wre
736− v11 = vec_mul(v5, v8); // i2*wre
737− v12 = vec_mul(v6, v8); // r3*wre
738− v13 = vec_mul(v7, v8); // i3*wre
739−
740− v0 = vec_ld(0, &(out[0])); // r0
741− v3 = vec_ld(i1+16, &(out[0])); // i1
742− v10 = vec_madd(v5, v9, v10); // r2*wim
743− v11 = vec_nmsub(v4, v9, v11); // i2*wim
744− v12 = vec_nmsub(v7, v9, v12); // r3*wim
745− v13 = vec_madd(v6, v9, v13); // i3*wim
746−
747− v1 = vec_ld(16, &(out[0])); // i0
748− v2 = vec_ld(i1, &(out[0])); // r1
749− v8 = vec_sub(v12, v10);
750− v12 = vec_add(v12, v10);
751− v9 = vec_sub(v11, v13);
752− v13 = vec_add(v11, v13);
753− v4 = vec_sub(v0, v12);
754− v0 = vec_add(v0, v12);
755− v7 = vec_sub(v3, v8);
756− v3 = vec_add(v3, v8);
757−
758− vec_st(v0, 0, &(out[0])); // r0
759− vec_st(v3, i1+16, &(out[0])); // i1
760− vec_st(v4, i2, &(out[0])); // r2
761− vec_st(v7, i3+16, &(out[0]));// i3
762−
763− v5 = vec_sub(v1, v13);
764− v1 = vec_add(v1, v13);
765− v6 = vec_sub(v2, v9);
766− v2 = vec_add(v2, v9);
767−
768− vec_st(v1, 16, &(out[0])); // i0
769− vec_st(v2, i1, &(out[0])); // r1
770− vec_st(v5, i2+16, &(out[0])); // i2
771− vec_st(v6, i3, &(out[0])); // r3
772−
773− do {
774− out += 8;
775− wre += 4;
776− wim -= 4;
777−
778− v8 = vec_ld(0, &(wre[0]));
779− v10 = vec_ld(0, &(wim[0]));
780− v9 = vec_ld(0, &(wim[-4]));
781− v9 = vec_perm(v9, v10, vcprm(s0,3,2,1));
782−
783− v4 = vec_ld(i2, &(out[0])); // r2
784− v5 = vec_ld(i2+16, &(out[0])); // i2
785− v6 = vec_ld(i3, &(out[0])); // r3
786− v7 = vec_ld(i3+16, &(out[0]));// i3
787− v10 = vec_mul(v4, v8); // r2*wre
788− v11 = vec_mul(v5, v8); // i2*wre
789− v12 = vec_mul(v6, v8); // r3*wre
790− v13 = vec_mul(v7, v8); // i3*wre
791−
792− v0 = vec_ld(0, &(out[0])); // r0
793− v3 = vec_ld(i1+16, &(out[0])); // i1
794− v10 = vec_madd(v5, v9, v10); // r2*wim
795− v11 = vec_nmsub(v4, v9, v11); // i2*wim
796− v12 = vec_nmsub(v7, v9, v12); // r3*wim
797− v13 = vec_madd(v6, v9, v13); // i3*wim
798−
799− v1 = vec_ld(16, &(out[0])); // i0
800− v2 = vec_ld(i1, &(out[0])); // r1
801− v8 = vec_sub(v12, v10);
802− v12 = vec_add(v12, v10);
803− v9 = vec_sub(v11, v13);
804− v13 = vec_add(v11, v13);
805− v4 = vec_sub(v0, v12);
806− v0 = vec_add(v0, v12);
807− v7 = vec_sub(v3, v8);
808− v3 = vec_add(v3, v8);
809−
810− vec_st(v0, 0, &(out[0])); // r0
811− vec_st(v3, i1+16, &(out[0])); // i1
812− vec_st(v4, i2, &(out[0])); // r2
813− vec_st(v7, i3+16, &(out[0])); // i3
814−
815− v5 = vec_sub(v1, v13);
816− v1 = vec_add(v1, v13);
817− v6 = vec_sub(v2, v9);
818− v2 = vec_add(v2, v9);
819−
820− vec_st(v1, 16, &(out[0])); // i0
821− vec_st(v2, i1, &(out[0])); // r1
822− vec_st(v5, i2+16, &(out[0])); // i2
823− vec_st(v6, i3, &(out[0])); // r3
824− } while (n-=2);
825−}
826−
827−#endif
828−
829−#endif /* AVCODEC_PPC_FFT_VSX_H */
libavcodec/rdft.c+0−120
@@ -1,120 +0,0 @@
1−/*
2− * (I)RDFT transforms
3− * Copyright (c) 2009 Alex Converse <alex dot converse at gmail dot com>
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−#include <stdlib.h>
22−#include <math.h>
23−#include "libavutil/error.h"
24−#include "libavutil/mathematics.h"
25−#include "rdft.h"
26−
27−/**
28− * @file
29− * (Inverse) Real Discrete Fourier Transforms.
30− */
31−
32−/** Map one real FFT into two parallel real even and odd FFTs. Then interleave
33− * the two real FFTs into one complex FFT. Unmangle the results.
34− * ref: http://www.engineeringproductivitytools.com/stuff/T0001/PT10.HTM
35− */
36−static void rdft_calc_c(RDFTContext *s, FFTSample *data)
37−{
38− int i, i1, i2;
39− FFTComplex ev, od, odsum;
40− const int n = 1 << s->nbits;
41− const float k1 = 0.5;
42− const float k2 = 0.5 - s->inverse;
43− const FFTSample *tcos = s->tcos;
44− const FFTSample *tsin = s->tsin;
45−
46− if (!s->inverse) {
47− s->fft.fft_permute(&s->fft, (FFTComplex*)data);
48− s->fft.fft_calc(&s->fft, (FFTComplex*)data);
49− }
50− /* i=0 is a special case because of packing, the DC term is real, so we
51− are going to throw the N/2 term (also real) in with it. */
52− ev.re = data[0];
53− data[0] = ev.re+data[1];
54− data[1] = ev.re-data[1];
55−
56−#define RDFT_UNMANGLE(sign0, sign1) \
57− for (i = 1; i < (n>>2); i++) { \
58− i1 = 2*i; \
59− i2 = n-i1; \
60− /* Separate even and odd FFTs */ \
61− ev.re = k1*(data[i1 ]+data[i2 ]); \
62− od.im = k2*(data[i2 ]-data[i1 ]); \
63− ev.im = k1*(data[i1+1]-data[i2+1]); \
64− od.re = k2*(data[i1+1]+data[i2+1]); \
65− /* Apply twiddle factors to the odd FFT and add to the even FFT */ \
66− odsum.re = od.re*tcos[i] sign0 od.im*tsin[i]; \
67− odsum.im = od.im*tcos[i] sign1 od.re*tsin[i]; \
68− data[i1 ] = ev.re + odsum.re; \
69− data[i1+1] = ev.im + odsum.im; \
70− data[i2 ] = ev.re - odsum.re; \
71− data[i2+1] = odsum.im - ev.im; \
72− }
73−
74− if (s->negative_sin) {
75− RDFT_UNMANGLE(+,-)
76− } else {
77− RDFT_UNMANGLE(-,+)
78− }
79−
80− data[2*i+1]=s->sign_convention*data[2*i+1];
81− if (s->inverse) {
82− data[0] *= k1;
83− data[1] *= k1;
84− s->fft.fft_permute(&s->fft, (FFTComplex*)data);
85− s->fft.fft_calc(&s->fft, (FFTComplex*)data);
86− }
87−}
88−
89−av_cold int ff_rdft_init(RDFTContext *s, int nbits, enum RDFTransformType trans)
90−{
91− int n = 1 << nbits;
92− int ret;
93−
94− s->nbits = nbits;
95− s->inverse = trans == IDFT_C2R || trans == DFT_C2R;
96− s->sign_convention = trans == IDFT_R2C || trans == DFT_C2R ? 1 : -1;
97− s->negative_sin = trans == DFT_C2R || trans == DFT_R2C;
98−
99− if (nbits < 4 || nbits > 16)
100− return AVERROR(EINVAL);
101−
102− if ((ret = ff_fft_init(&s->fft, nbits-1, trans == IDFT_C2R || trans == IDFT_R2C)) < 0)
103− return ret;
104−
105− ff_init_ff_cos_tabs(nbits);
106− s->tcos = ff_cos_tabs[nbits];
107− s->tsin = ff_cos_tabs[nbits] + (n >> 2);
108− s->rdft_calc = rdft_calc_c;
109−
110−#if ARCH_ARM
111− ff_rdft_init_arm(s);
112−#endif
113−
114− return 0;
115−}
116−
117−av_cold void ff_rdft_end(RDFTContext *s)
118−{
119− ff_fft_end(&s->fft);
120−}
libavcodec/rdft.h+0−52
@@ -1,52 +0,0 @@
1−/*
2− * (I)RDFT transforms
3− * Copyright (c) 2009 Alex Converse <alex dot converse at gmail dot com>
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−#if !defined(AVCODEC_RDFT_H) && (!defined(FFT_FLOAT) || FFT_FLOAT)
23−#define AVCODEC_RDFT_H
24−
25−#include "config.h"
26−#include "fft.h"
27−
28−struct RDFTContext {
29− int nbits;
30− int inverse;
31− int sign_convention;
32−
33− /* pre/post rotation tables */
34− const FFTSample *tcos;
35− const FFTSample *tsin;
36− int negative_sin;
37− FFTContext fft;
38− void (*rdft_calc)(struct RDFTContext *s, FFTSample *z);
39−};
40−
41−/**
42− * Set up a real FFT.
43− * @param nbits log2 of the length of the input array
44− * @param trans the type of transform
45− */
46−int ff_rdft_init(RDFTContext *s, int nbits, enum RDFTransformType trans);
47−void ff_rdft_end(RDFTContext *s);
48−
49−void ff_rdft_init_arm(RDFTContext *s);
50−
51−
52−#endif /* AVCODEC_RDFT_H */
libavcodec/tests/.gitignore+0−3
@@ -1,6 +1,5 @@
11 /av1_levels
22 /avcodec
3−/avfft
43 /avpacket
54 /bitstream_be
65 /bitstream_le
@@ -8,8 +7,6 @@ · 1 unmodified line
87 /celp_math
98 /codec_desc
109 /dct
11−/fft
12−/fft-fixed32
1310 /golomb
1411 /h264_levels
1512 /h265_levels
libavcodec/tests/avfft.c+0−25
@@ -1,25 +0,0 @@
1−/*
2− * This file is part of FFmpeg.
3− *
4− * FFmpeg is free software; you can redistribute it and/or
5− * modify it under the terms of the GNU Lesser General Public
6− * License as published by the Free Software Foundation; either
7− * version 2.1 of the License, or (at your option) any later version.
8− *
9− * FFmpeg is distributed in the hope that it will be useful,
10− * but WITHOUT ANY WARRANTY; without even the implied warranty of
11− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12− * Lesser General Public License for more details.
13− *
14− * You should have received a copy of the GNU Lesser General Public
15− * License along with FFmpeg; if not, write to the Free Software
16− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17− */
18−
19−/*
20− * This test is similar to fft-fixed.c or fft-fixed32.c
21− */
22−
23−#define AVFFT 1
24−#define FFT_FLOAT 1
25−#include "fft.c"
libavcodec/tests/fft-fixed32.c+0−21
@@ -1,21 +0,0 @@
1−/*
2− * This file is part of FFmpeg.
3− *
4− * FFmpeg is free software; you can redistribute it and/or
5− * modify it under the terms of the GNU Lesser General Public
6− * License as published by the Free Software Foundation; either
7− * version 2.1 of the License, or (at your option) any later version.
8− *
9− * FFmpeg is distributed in the hope that it will be useful,
10− * but WITHOUT ANY WARRANTY; without even the implied warranty of
11− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12− * Lesser General Public License for more details.
13− *
14− * You should have received a copy of the GNU Lesser General Public
15− * License along with FFmpeg; if not, write to the Free Software
16− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17− */
18−
19−#define FFT_FLOAT 0
20−#define AVFFT 0
21−#include "fft.c"
libavcodec/tests/fft.c+0−683
@@ -1,683 +0,0 @@
1−/*
2− * (c) 2002 Fabrice Bellard
3− *
4− * This file is part of FFmpeg.
5− *
6− * FFmpeg is free software; you can redistribute it and/or
7− * modify it under the terms of the GNU Lesser General Public
8− * License as published by the Free Software Foundation; either
9− * version 2.1 of the License, or (at your option) any later version.
10− *
11− * FFmpeg is distributed in the hope that it will be useful,
12− * but WITHOUT ANY WARRANTY; without even the implied warranty of
13− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14− * Lesser General Public License for more details.
15− *
16− * You should have received a copy of the GNU Lesser General Public
17− * License along with FFmpeg; if not, write to the Free Software
18− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19− */
20−
21−#include "libavutil/internal.h"
22−
23−FF_DISABLE_DEPRECATION_WARNINGS
24−
25−/**
26− * @file
27− * FFT and MDCT tests.
28− */
29−
30−#include "config.h"
31−
32−#ifndef AVFFT
33−#define AVFFT 0
34−#endif
35−
36−#include <math.h>
37−#if HAVE_UNISTD_H
38−#include <unistd.h>
39−#endif
40−#include <stdio.h>
41−#include <stdlib.h>
42−#include <string.h>
43−
44−#include "libavutil/cpu.h"
45−#include "libavutil/error.h"
46−#include "libavutil/lfg.h"
47−#include "libavutil/log.h"
48−#include "libavutil/mathematics.h"
49−#include "libavutil/time.h"
50−
51−#if AVFFT
52−#include "libavcodec/avfft.h"
53−#else
54−#include "libavcodec/fft.h"
55−#endif
56−
57−#if FFT_FLOAT
58−#include "libavcodec/dct.h"
59−#include "libavcodec/rdft.h"
60−#endif
61−
62−/* reference fft */
63−
64−#define MUL16(a, b) ((a) * (b))
65−
66−#define CMAC(pre, pim, are, aim, bre, bim) \
67− { \
68− pre += (MUL16(are, bre) - MUL16(aim, bim)); \
69− pim += (MUL16(are, bim) + MUL16(bre, aim)); \
70− }
71−
72−#if FFT_FLOAT || AVFFT
73−#define RANGE 1.0
74−#define REF_SCALE(x, bits) (x)
75−#define FMT "%10.6f"
76−#else
77−#define RANGE 8388608
78−#define REF_SCALE(x, bits) (x)
79−#define FMT "%6d"
80−#endif
81−
82−static struct {
83− float re, im;
84−} *exptab;
85−
86−static int fft_ref_init(int nbits, int inverse)
87−{
88− int i, n = 1 << nbits;
89−
90− exptab = av_malloc_array((n / 2), sizeof(*exptab));
91− if (!exptab)
92− return AVERROR(ENOMEM);
93−
94− for (i = 0; i < (n / 2); i++) {
95− double alpha = 2 * M_PI * (float) i / (float) n;
96− double c1 = cos(alpha), s1 = sin(alpha);
97− if (!inverse)
98− s1 = -s1;
99− exptab[i].re = c1;
100− exptab[i].im = s1;
101− }
102− return 0;
103−}
104−
105−static void fft_ref(FFTComplex *tabr, FFTComplex *tab, int nbits)
106−{
107− int i, j;
108− int n = 1 << nbits;
109− int n2 = n >> 1;
110−
111− for (i = 0; i < n; i++) {
112− double tmp_re = 0, tmp_im = 0;
113− FFTComplex *q = tab;
114− for (j = 0; j < n; j++) {
115− double s, c;
116− int k = (i * j) & (n - 1);
117− if (k >= n2) {
118− c = -exptab[k - n2].re;
119− s = -exptab[k - n2].im;
120− } else {
121− c = exptab[k].re;
122− s = exptab[k].im;
123− }
124− CMAC(tmp_re, tmp_im, c, s, q->re, q->im);
125− q++;
126− }
127− tabr[i].re = REF_SCALE(tmp_re, nbits);
128− tabr[i].im = REF_SCALE(tmp_im, nbits);
129− }
130−}
131−
132−#if CONFIG_MDCT
133−static void imdct_ref(FFTSample *out, FFTSample *in, int nbits)
134−{
135− int i, k, n = 1 << nbits;
136−
137− for (i = 0; i < n; i++) {
138− double sum = 0;
139− for (k = 0; k < n / 2; k++) {
140− int a = (2 * i + 1 + (n / 2)) * (2 * k + 1);
141− double f = cos(M_PI * a / (double) (2 * n));
142− sum += f * in[k];
143− }
144− out[i] = REF_SCALE(-sum, nbits - 2);
145− }
146−}
147−
148−/* NOTE: no normalisation by 1 / N is done */
149−static void mdct_ref(FFTSample *output, FFTSample *input, int nbits)
150−{
151− int i, k, n = 1 << nbits;
152−
153− /* do it by hand */
154− for (k = 0; k < n / 2; k++) {
155− double s = 0;
156− for (i = 0; i < n; i++) {
157− double a = (2 * M_PI * (2 * i + 1 + n / 2) * (2 * k + 1) / (4 * n));
158− s += input[i] * cos(a);
159− }
160− output[k] = REF_SCALE(s, nbits - 1);
161− }
162−}
163−#endif /* CONFIG_MDCT */
164−
165−#if FFT_FLOAT
166−#if CONFIG_DCT
167−static void idct_ref(FFTSample *output, FFTSample *input, int nbits)
168−{
169− int i, k, n = 1 << nbits;
170−
171− /* do it by hand */
172− for (i = 0; i < n; i++) {
173− double s = 0.5 * input[0];
174− for (k = 1; k < n; k++) {
175− double a = M_PI * k * (i + 0.5) / n;
176− s += input[k] * cos(a);
177− }
178− output[i] = 2 * s / n;
179− }
180−}
181−
182−static void dct_ref(FFTSample *output, FFTSample *input, int nbits)
183−{
184− int i, k, n = 1 << nbits;
185−
186− /* do it by hand */
187− for (k = 0; k < n; k++) {
188− double s = 0;
189− for (i = 0; i < n; i++) {
190− double a = M_PI * k * (i + 0.5) / n;
191− s += input[i] * cos(a);
192− }
193− output[k] = s;
194− }
195−}
196−#endif /* CONFIG_DCT */
197−#endif /* FFT_FLOAT */
198−
199−static FFTSample frandom(AVLFG *prng)
200−{
201− return (int16_t) av_lfg_get(prng) / 32768.0 * RANGE;
202−}
203−
204−static int check_diff(FFTSample *tab1, FFTSample *tab2, int n, double scale)
205−{
206− int i, err = 0;
207− double error = 0, max = 0;
208−
209− for (i = 0; i < n; i++) {
210− double e = fabs(tab1[i] - (tab2[i] / scale)) / RANGE;
211− if (e >= 1e-3) {
212− av_log(NULL, AV_LOG_ERROR, "ERROR %5d: "FMT" "FMT"\n",
213− i, tab1[i], tab2[i]);
214− err = 1;
215− }
216− error += e * e;
217− if (e > max)
218− max = e;
219− }
220− av_log(NULL, AV_LOG_INFO, "max:%f e:%g\n", max, sqrt(error / n));
221− return err;
222−}
223−
224−static inline void fft_init(FFTContext **s, int nbits, int inverse)
225−{
226−#if AVFFT
227− *s = av_fft_init(nbits, inverse);
228−#else
229− ff_fft_init(*s, nbits, inverse);
230−#endif
231−}
232−
233−#if CONFIG_MDCT
234−static inline void mdct_init(FFTContext **s, int nbits, int inverse, double scale)
235−{
236−#if AVFFT
237− *s = av_mdct_init(nbits, inverse, scale);
238−#else
239− ff_mdct_init(*s, nbits, inverse, scale);
240−#endif
241−}
242−
243−static inline void mdct_calc(FFTContext *s, FFTSample *output, const FFTSample *input)
244−{
245−#if AVFFT
246− av_mdct_calc(s, output, input);
247−#else
248− s->mdct_calc(s, output, input);
249−#endif
250−}
251−
252−static inline void imdct_calc(struct FFTContext *s, FFTSample *output, const FFTSample *input)
253−{
254−#if AVFFT
255− av_imdct_calc(s, output, input);
256−#else
257− s->imdct_calc(s, output, input);
258−#endif
259−}
260−#endif
261−
262−static inline void fft_permute(FFTContext *s, FFTComplex *z)
263−{
264−#if AVFFT
265− av_fft_permute(s, z);
266−#else
267− s->fft_permute(s, z);
268−#endif
269−}
270−
271−static inline void fft_calc(FFTContext *s, FFTComplex *z)
272−{
273−#if AVFFT
274− av_fft_calc(s, z);
275−#else
276− s->fft_calc(s, z);
277−#endif
278−}
279−
280−static inline void mdct_end(FFTContext *s)
281−{
282−#if AVFFT
283− av_mdct_end(s);
284−#else
285− ff_mdct_end(s);
286−#endif
287−}
288−
289−static inline void fft_end(FFTContext *s)
290−{
291−#if AVFFT
292− av_fft_end(s);
293−#else
294− ff_fft_end(s);
295−#endif
296−}
297−
298−#if FFT_FLOAT
299−static inline void rdft_init(RDFTContext **r, int nbits, enum RDFTransformType trans)
300−{
301−#if AVFFT
302− *r = av_rdft_init(nbits, trans);
303−#else
304− ff_rdft_init(*r, nbits, trans);
305−#endif
306−}
307−
308−static inline void dct_init(DCTContext **d, int nbits, enum DCTTransformType trans)
309−{
310−#if AVFFT
311− *d = av_dct_init(nbits, trans);
312−#else
313− ff_dct_init(*d, nbits, trans);
314−#endif
315−}
316−
317−static inline void rdft_calc(RDFTContext *r, FFTSample *tab)
318−{
319−#if AVFFT
320− av_rdft_calc(r, tab);
321−#else
322− r->rdft_calc(r, tab);
323−#endif
324−}
325−
326−static inline void dct_calc(DCTContext *d, FFTSample *data)
327−{
328−#if AVFFT
329− av_dct_calc(d, data);
330−#else
331− d->dct_calc(d, data);
332−#endif
333−}
334−
335−static inline void rdft_end(RDFTContext *r)
336−{
337−#if AVFFT
338− av_rdft_end(r);
339−#else
340− ff_rdft_end(r);
341−#endif
342−}
343−
344−static inline void dct_end(DCTContext *d)
345−{
346−#if AVFFT
347− av_dct_end(d);
348−#else
349− ff_dct_end(d);
350−#endif
351−}
352−#endif /* FFT_FLOAT */
353−
354−static void help(void)
355−{
356− av_log(NULL, AV_LOG_INFO,
357− "usage: fft-test [-h] [-s] [-i] [-n b]\n"
358− "-h print this help\n"
359− "-s speed test\n"
360− "-m (I)MDCT test\n"
361− "-d (I)DCT test\n"
362− "-r (I)RDFT test\n"
363− "-i inverse transform test\n"
364− "-n b set the transform size to 2^b\n"
365− "-f x set scale factor for output data of (I)MDCT to x\n");
366−}
367−
368−enum tf_transform {
369− TRANSFORM_FFT,
370− TRANSFORM_MDCT,
371− TRANSFORM_RDFT,
372− TRANSFORM_DCT,
373−};
374−
375−#if !HAVE_GETOPT
376−#include "compat/getopt.c"
377−#endif
378−
379−int main(int argc, char **argv)
380−{
381− FFTComplex *tab, *tab1, *tab_ref;
382− FFTSample *tab2;
383− enum tf_transform transform = TRANSFORM_FFT;
384− FFTContext *m, *s;
385−#if FFT_FLOAT
386− RDFTContext *r;
387− DCTContext *d;
388−#endif /* FFT_FLOAT */
389− int it, i, err = 1;
390− int do_speed = 0, do_inverse = 0;
391− int fft_nbits = 9, fft_size;
392− double scale = 1.0;
393− AVLFG prng;
394−
395−#if !AVFFT
396− s = av_mallocz(sizeof(*s));
397− m = av_mallocz(sizeof(*m));
398−#endif
399−
400−#if !AVFFT && FFT_FLOAT
401− r = av_mallocz(sizeof(*r));
402− d = av_mallocz(sizeof(*d));
403−#endif
404−
405− av_lfg_init(&prng, 1);
406−
407− for (;;) {
408− int c = getopt(argc, argv, "hsimrdn:f:c:");
409− if (c == -1)
410− break;
411− switch (c) {
412− case 'h':
413− help();
414− return 1;
415− case 's':
416− do_speed = 1;
417− break;
418− case 'i':
419− do_inverse = 1;
420− break;
421− case 'm':
422− transform = TRANSFORM_MDCT;
423− break;
424− case 'r':
425− transform = TRANSFORM_RDFT;
426− break;
427− case 'd':
428− transform = TRANSFORM_DCT;
429− break;
430− case 'n':
431− fft_nbits = atoi(optarg);
432− break;
433− case 'f':
434− scale = atof(optarg);
435− break;
436− case 'c':
437− {
438− unsigned cpuflags = av_get_cpu_flags();
439−
440− if (av_parse_cpu_caps(&cpuflags, optarg) < 0)
441− return 1;
442−
443− av_force_cpu_flags(cpuflags);
444− break;
445− }
446− }
447− }
448−
449− fft_size = 1 << fft_nbits;
450− tab = av_malloc_array(fft_size, sizeof(FFTComplex));
451− tab1 = av_malloc_array(fft_size, sizeof(FFTComplex));
452− tab_ref = av_malloc_array(fft_size, sizeof(FFTComplex));
453− tab2 = av_malloc_array(fft_size, sizeof(FFTSample));
454−
455− if (!(tab && tab1 && tab_ref && tab2))
456− goto cleanup;
457−
458− switch (transform) {
459−#if CONFIG_MDCT
460− case TRANSFORM_MDCT:
461− av_log(NULL, AV_LOG_INFO, "Scale factor is set to %f\n", scale);
462− if (do_inverse)
463− av_log(NULL, AV_LOG_INFO, "IMDCT");
464− else
465− av_log(NULL, AV_LOG_INFO, "MDCT");
466− mdct_init(&m, fft_nbits, do_inverse, scale);
467− break;
468−#endif /* CONFIG_MDCT */
469− case TRANSFORM_FFT:
470− if (do_inverse)
471− av_log(NULL, AV_LOG_INFO, "IFFT");
472− else
473− av_log(NULL, AV_LOG_INFO, "FFT");
474− fft_init(&s, fft_nbits, do_inverse);
475− if ((err = fft_ref_init(fft_nbits, do_inverse)) < 0)
476− goto cleanup;
477− break;
478−#if FFT_FLOAT
479−# if CONFIG_RDFT
480− case TRANSFORM_RDFT:
481− if (do_inverse)
482− av_log(NULL, AV_LOG_INFO, "IDFT_C2R");
483− else
484− av_log(NULL, AV_LOG_INFO, "DFT_R2C");
485− rdft_init(&r, fft_nbits, do_inverse ? IDFT_C2R : DFT_R2C);
486− if ((err = fft_ref_init(fft_nbits, do_inverse)) < 0)
487− goto cleanup;
488− break;
489−# endif /* CONFIG_RDFT */
490−# if CONFIG_DCT
491− case TRANSFORM_DCT:
492− if (do_inverse)
493− av_log(NULL, AV_LOG_INFO, "DCT_III");
494− else
495− av_log(NULL, AV_LOG_INFO, "DCT_II");
496− dct_init(&d, fft_nbits, do_inverse ? DCT_III : DCT_II);
497− break;
498−# endif /* CONFIG_DCT */
499−#endif /* FFT_FLOAT */
500− default:
501− av_log(NULL, AV_LOG_ERROR, "Requested transform not supported\n");
502− goto cleanup;
503− }
504− av_log(NULL, AV_LOG_INFO, " %d test\n", fft_size);
505−
506− /* generate random data */
507−
508− for (i = 0; i < fft_size; i++) {
509− tab1[i].re = frandom(&prng);
510− tab1[i].im = frandom(&prng);
511− }
512−
513− /* checking result */
514− av_log(NULL, AV_LOG_INFO, "Checking...\n");
515−
516− switch (transform) {
517−#if CONFIG_MDCT
518− case TRANSFORM_MDCT:
519− if (do_inverse) {
520− imdct_ref(&tab_ref->re, &tab1->re, fft_nbits);
521− imdct_calc(m, tab2, &tab1->re);
522− err = check_diff(&tab_ref->re, tab2, fft_size, scale);
523− } else {
524− mdct_ref(&tab_ref->re, &tab1->re, fft_nbits);
525− mdct_calc(m, tab2, &tab1->re);
526− err = check_diff(&tab_ref->re, tab2, fft_size / 2, scale);
527− }
528− break;
529−#endif /* CONFIG_MDCT */
530− case TRANSFORM_FFT:
531− memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
532− fft_permute(s, tab);
533− fft_calc(s, tab);
534−
535− fft_ref(tab_ref, tab1, fft_nbits);
536− err = check_diff(&tab_ref->re, &tab->re, fft_size * 2, 1.0);
537− break;
538−#if FFT_FLOAT
539−#if CONFIG_RDFT
540− case TRANSFORM_RDFT:
541− {
542− int fft_size_2 = fft_size >> 1;
543− if (do_inverse) {
544− tab1[0].im = 0;
545− tab1[fft_size_2].im = 0;
546− for (i = 1; i < fft_size_2; i++) {
547− tab1[fft_size_2 + i].re = tab1[fft_size_2 - i].re;
548− tab1[fft_size_2 + i].im = -tab1[fft_size_2 - i].im;
549− }
550−
551− memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
552− tab2[1] = tab1[fft_size_2].re;
553−
554− rdft_calc(r, tab2);
555− fft_ref(tab_ref, tab1, fft_nbits);
556− for (i = 0; i < fft_size; i++) {
557− tab[i].re = tab2[i];
558− tab[i].im = 0;
559− }
560− err = check_diff(&tab_ref->re, &tab->re, fft_size * 2, 0.5);
561− } else {
562− for (i = 0; i < fft_size; i++) {
563− tab2[i] = tab1[i].re;
564− tab1[i].im = 0;
565− }
566− rdft_calc(r, tab2);
567− fft_ref(tab_ref, tab1, fft_nbits);
568− tab_ref[0].im = tab_ref[fft_size_2].re;
569− err = check_diff(&tab_ref->re, tab2, fft_size, 1.0);
570− }
571− break;
572− }
573−#endif /* CONFIG_RDFT */
574−#if CONFIG_DCT
575− case TRANSFORM_DCT:
576− memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
577− dct_calc(d, &tab->re);
578− if (do_inverse)
579− idct_ref(&tab_ref->re, &tab1->re, fft_nbits);
580− else
581− dct_ref(&tab_ref->re, &tab1->re, fft_nbits);
582− err = check_diff(&tab_ref->re, &tab->re, fft_size, 1.0);
583− break;
584−#endif /* CONFIG_DCT */
585−#endif /* FFT_FLOAT */
586− }
587−
588− /* do a speed test */
589−
590− if (do_speed) {
591− int64_t time_start, duration;
592− int nb_its;
593−
594− av_log(NULL, AV_LOG_INFO, "Speed test...\n");
595− /* we measure during about 1 seconds */
596− nb_its = 1;
597− for (;;) {
598− time_start = av_gettime_relative();
599− for (it = 0; it < nb_its; it++) {
600− switch (transform) {
601−#if CONFIG_MDCT
602− case TRANSFORM_MDCT:
603− if (do_inverse)
604− imdct_calc(m, &tab->re, &tab1->re);
605− else
606− mdct_calc(m, &tab->re, &tab1->re);
607− break;
608−#endif
609− case TRANSFORM_FFT:
610− memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
611− fft_calc(s, tab);
612− break;
613−#if FFT_FLOAT
614− case TRANSFORM_RDFT:
615− memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
616− rdft_calc(r, tab2);
617− break;
618− case TRANSFORM_DCT:
619− memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
620− dct_calc(d, tab2);
621− break;
622−#endif /* FFT_FLOAT */
623− }
624− }
625− duration = av_gettime_relative() - time_start;
626− if (duration >= 1000000)
627− break;
628− nb_its *= 2;
629− }
630− av_log(NULL, AV_LOG_INFO,
631− "time: %0.1f us/transform [total time=%0.2f s its=%d]\n",
632− (double) duration / nb_its,
633− (double) duration / 1000000.0,
634− nb_its);
635− }
636−
637− switch (transform) {
638−#if CONFIG_MDCT
639− case TRANSFORM_MDCT:
640− mdct_end(m);
641− break;
642−#endif /* CONFIG_MDCT */
643− case TRANSFORM_FFT:
644− fft_end(s);
645− break;
646−#if FFT_FLOAT
647−# if CONFIG_RDFT
648− case TRANSFORM_RDFT:
649− rdft_end(r);
650− break;
651−# endif /* CONFIG_RDFT */
652−# if CONFIG_DCT
653− case TRANSFORM_DCT:
654− dct_end(d);
655− break;
656−# endif /* CONFIG_DCT */
657−#endif /* FFT_FLOAT */
658− }
659−
660−cleanup:
661− av_free(tab);
662− av_free(tab1);
663− av_free(tab2);
664− av_free(tab_ref);
665− av_free(exptab);
666−
667−#if !AVFFT
668− av_free(s);
669− av_free(m);
670−#endif
671−
672−#if !AVFFT && FFT_FLOAT
673− av_free(r);
674− av_free(d);
675−#endif
676−
677− if (err)
678− printf("Error: %d.\n", err);
679−
680− return !!err;
681−}
682−
683−FF_ENABLE_DEPRECATION_WARNINGS
libavcodec/x86/Makefile+0−4
@@ -5,11 +5,9 @@ OBJS-$(CONFIG_AC3DSP) += x86/ac3dsp_init.o · 4 unmodified lines
55 OBJS-$(CONFIG_AUDIODSP) += x86/audiodsp_init.o
66 OBJS-$(CONFIG_BLOCKDSP) += x86/blockdsp_init.o
77 OBJS-$(CONFIG_BSWAPDSP) += x86/bswapdsp_init.o
8−OBJS-$(CONFIG_DCT) += x86/dct_init.o
98 OBJS-$(CONFIG_DIRAC_DECODER) += x86/diracdsp_init.o \
109 x86/dirac_dwt_init.o
1110 OBJS-$(CONFIG_FDCTDSP) += x86/fdctdsp_init.o
12−OBJS-$(CONFIG_FFT) += x86/fft_init.o
1311 OBJS-$(CONFIG_FMTCONVERT) += x86/fmtconvert_init.o
1412 OBJS-$(CONFIG_H263DSP) += x86/h263dsp_init.o
1513 OBJS-$(CONFIG_H264CHROMA) += x86/h264chroma_init.o
@@ -98,8 +96,6 @@ X86ASM-OBJS-$(CONFIG_AC3DSP) += x86/ac3dsp.o \ · 82 unmodified lines
9896 X86ASM-OBJS-$(CONFIG_AUDIODSP) += x86/audiodsp.o
9997 X86ASM-OBJS-$(CONFIG_BLOCKDSP) += x86/blockdsp.o
10098 X86ASM-OBJS-$(CONFIG_BSWAPDSP) += x86/bswapdsp.o
101−X86ASM-OBJS-$(CONFIG_DCT) += x86/dct32.o
102−X86ASM-OBJS-$(CONFIG_FFT) += x86/fft.o
10399 X86ASM-OBJS-$(CONFIG_FMTCONVERT) += x86/fmtconvert.o
104100 X86ASM-OBJS-$(CONFIG_H263DSP) += x86/h263_loopfilter.o
105101 X86ASM-OBJS-$(CONFIG_H264CHROMA) += x86/h264_chromamc.o \
libavcodec/x86/dct_init.c+0−36
@@ -1,36 +0,0 @@
1−/*
2− * This file is part of FFmpeg.
3− *
4− * FFmpeg is free software; you can redistribute it and/or
5− * modify it under the terms of the GNU Lesser General Public
6− * License as published by the Free Software Foundation; either
7− * version 2.1 of the License, or (at your option) any later version.
8− *
9− * FFmpeg is distributed in the hope that it will be useful,
10− * but WITHOUT ANY WARRANTY; without even the implied warranty of
11− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12− * Lesser General Public License for more details.
13− *
14− * You should have received a copy of the GNU Lesser General Public
15− * License along with FFmpeg; if not, write to the Free Software
16− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17− */
18−
19−#include "config.h"
20−#include "libavutil/attributes.h"
21−#include "libavutil/cpu.h"
22−#include "libavutil/x86/cpu.h"
23−#include "libavcodec/dct.h"
24−
25−void ff_dct32_float_sse2(FFTSample *out, const FFTSample *in);
26−void ff_dct32_float_avx(FFTSample *out, const FFTSample *in);
27−
28−av_cold void ff_dct_init_x86(DCTContext *s)
29−{
30− int cpu_flags = av_get_cpu_flags();
31−
32− if (EXTERNAL_SSE2(cpu_flags))
33− s->dct32 = ff_dct32_float_sse2;
34− if (EXTERNAL_AVX_FAST(cpu_flags))
35− s->dct32 = ff_dct32_float_avx;
36−}
libavcodec/x86/fft.asm+0−838
@@ -1,838 +0,0 @@
1−;******************************************************************************
2−;* FFT transform with SSE/AVX optimizations
3−;* Copyright (c) 2008 Loren Merritt
4−;* Copyright (c) 2011 Vitor Sessak
5−;*
6−;* This algorithm (though not any of the implementation details) is
7−;* based on libdjbfft by D. J. Bernstein.
8−;*
9−;* This file is part of FFmpeg.
10−;*
11−;* FFmpeg is free software; you can redistribute it and/or
12−;* modify it under the terms of the GNU Lesser General Public
13−;* License as published by the Free Software Foundation; either
14−;* version 2.1 of the License, or (at your option) any later version.
15−;*
16−;* FFmpeg is distributed in the hope that it will be useful,
17−;* but WITHOUT ANY WARRANTY; without even the implied warranty of
18−;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19−;* Lesser General Public License for more details.
20−;*
21−;* You should have received a copy of the GNU Lesser General Public
22−;* License along with FFmpeg; if not, write to the Free Software
23−;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24−;******************************************************************************
25−
26−; These functions are not individually interchangeable with the C versions.
27−; While C takes arrays of FFTComplex, SSE/3DNow leave intermediate results
28−; in blocks as conventient to the vector size.
29−; i.e. {4x real, 4x imaginary, 4x real, ...} (or 2x respectively)
30−
31−%include "libavutil/x86/x86util.asm"
32−
33−%if ARCH_X86_64
34−%define pointer resq
35−%else
36−%define pointer resd
37−%endif
38−
39−struc FFTContext
40− .nbits: resd 1
41− .reverse: resd 1
42− .revtab: pointer 1
43− .tmpbuf: pointer 1
44− .mdctsize: resd 1
45− .mdctbits: resd 1
46− .tcos: pointer 1
47− .tsin: pointer 1
48− .fftperm: pointer 1
49− .fftcalc: pointer 1
50− .imdctcalc:pointer 1
51− .imdcthalf:pointer 1
52−endstruc
53−
54−SECTION_RODATA 32
55−
56−%define M_SQRT1_2 0.70710678118654752440
57−%define M_COS_PI_1_8 0.923879532511287
58−%define M_COS_PI_3_8 0.38268343236509
59−
60−ps_cos16_1: dd 1.0, M_COS_PI_1_8, M_SQRT1_2, M_COS_PI_3_8, 1.0, M_COS_PI_1_8, M_SQRT1_2, M_COS_PI_3_8
61−ps_cos16_2: dd 0, M_COS_PI_3_8, M_SQRT1_2, M_COS_PI_1_8, 0, -M_COS_PI_3_8, -M_SQRT1_2, -M_COS_PI_1_8
62−
63−ps_root2: times 8 dd M_SQRT1_2
64−ps_root2mppm: dd -M_SQRT1_2, M_SQRT1_2, M_SQRT1_2, -M_SQRT1_2, -M_SQRT1_2, M_SQRT1_2, M_SQRT1_2, -M_SQRT1_2
65−ps_p1p1m1p1: dd 0, 0, 1<<31, 0, 0, 0, 1<<31, 0
66−
67−perm1: dd 0x00, 0x02, 0x03, 0x01, 0x03, 0x00, 0x02, 0x01
68−perm2: dd 0x00, 0x01, 0x02, 0x03, 0x01, 0x00, 0x02, 0x03
69−ps_p1p1m1p1root2: dd 1.0, 1.0, -1.0, 1.0, M_SQRT1_2, M_SQRT1_2, M_SQRT1_2, M_SQRT1_2
70−ps_m1m1p1m1p1m1m1m1: dd 1<<31, 1<<31, 0, 1<<31, 0, 1<<31, 1<<31, 1<<31
71−ps_m1p1: dd 1<<31, 0
72−
73−cextern ps_neg
74−
75−%assign i 16
76−%rep 14
77−cextern cos_ %+ i
78−%assign i i<<1
79−%endrep
80−
81−%if ARCH_X86_64
82− %define pointer dq
83−%else
84− %define pointer dd
85−%endif
86−
87−%macro IF0 1+
88−%endmacro
89−%macro IF1 1+
90− %1
91−%endmacro
92−
93−SECTION .text
94−
95−; in: %1 = {r0,i0,r2,i2,r4,i4,r6,i6}
96−; %2 = {r1,i1,r3,i3,r5,i5,r7,i7}
97−; %3, %4, %5 tmp
98−; out: %1 = {r0,r1,r2,r3,i0,i1,i2,i3}
99−; %2 = {r4,r5,r6,r7,i4,i5,i6,i7}
100−%macro T8_AVX 5
101− vsubps %5, %1, %2 ; v = %1 - %2
102− vaddps %3, %1, %2 ; w = %1 + %2
103− vmulps %2, %5, [ps_p1p1m1p1root2] ; v *= vals1
104− vpermilps %2, %2, [perm1]
105− vblendps %1, %2, %3, 0x33 ; q = {w1,w2,v4,v2,w5,w6,v7,v6}
106− vshufps %5, %3, %2, 0x4e ; r = {w3,w4,v1,v3,w7,w8,v8,v5}
107− vsubps %4, %5, %1 ; s = r - q
108− vaddps %1, %5, %1 ; u = r + q
109− vpermilps %1, %1, [perm2] ; k = {u1,u2,u3,u4,u6,u5,u7,u8}
110− vshufps %5, %4, %1, 0xbb
111− vshufps %3, %4, %1, 0xee
112− vperm2f128 %3, %3, %5, 0x13
113− vxorps %4, %4, [ps_m1m1p1m1p1m1m1m1] ; s *= {1,1,-1,-1,1,-1,-1,-1}
114− vshufps %2, %1, %4, 0xdd
115− vshufps %1, %1, %4, 0x88
116− vperm2f128 %4, %2, %1, 0x02 ; v = {k1,k3,s1,s3,k2,k4,s2,s4}
117− vperm2f128 %1, %1, %2, 0x13 ; w = {k6,k8,s6,s8,k5,k7,s5,s7}
118− vsubps %5, %1, %3
119− vblendps %1, %5, %1, 0x55 ; w -= {0,s7,0,k7,0,s8,0,k8}
120− vsubps %2, %4, %1 ; %2 = v - w
121− vaddps %1, %4, %1 ; %1 = v + w
122−%endmacro
123−
124−; In SSE mode do one fft4 transforms
125−; in: %1={r0,i0,r2,i2} %2={r1,i1,r3,i3}
126−; out: %1={r0,r1,r2,r3} %2={i0,i1,i2,i3}
127−;
128−; In AVX mode do two fft4 transforms
129−; in: %1={r0,i0,r2,i2,r4,i4,r6,i6} %2={r1,i1,r3,i3,r5,i5,r7,i7}
130−; out: %1={r0,r1,r2,r3,r4,r5,r6,r7} %2={i0,i1,i2,i3,i4,i5,i6,i7}
131−%macro T4_SSE 3
132− subps %3, %1, %2 ; {t3,t4,-t8,t7}
133− addps %1, %1, %2 ; {t1,t2,t6,t5}
134− xorps %3, %3, [ps_p1p1m1p1]
135− shufps %2, %1, %3, 0xbe ; {t6,t5,t7,t8}
136− shufps %1, %1, %3, 0x44 ; {t1,t2,t3,t4}
137− subps %3, %1, %2 ; {r2,i2,r3,i3}
138− addps %1, %1, %2 ; {r0,i0,r1,i1}
139− shufps %2, %1, %3, 0xdd ; {i0,i1,i2,i3}
140− shufps %1, %1, %3, 0x88 ; {r0,r1,r2,r3}
141−%endmacro
142−
143−; In SSE mode do one FFT8
144−; in: %1={r0,r1,r2,r3} %2={i0,i1,i2,i3} %3={r4,i4,r6,i6} %4={r5,i5,r7,i7}
145−; out: %1={r0,r1,r2,r3} %2={i0,i1,i2,i3} %1={r4,r5,r6,r7} %2={i4,i5,i6,i7}
146−;
147−; In AVX mode do two FFT8
148−; in: %1={r0,i0,r2,i2,r8, i8, r10,i10} %2={r1,i1,r3,i3,r9, i9, r11,i11}
149−; %3={r4,i4,r6,i6,r12,i12,r14,i14} %4={r5,i5,r7,i7,r13,i13,r15,i15}
150−; out: %1={r0,r1,r2,r3,r8, r9, r10,r11} %2={i0,i1,i2,i3,i8, i9, i10,i11}
151−; %3={r4,r5,r6,r7,r12,r13,r14,r15} %4={i4,i5,i6,i7,i12,i13,i14,i15}
152−%macro T8_SSE 6
153− addps %6, %3, %4 ; {t1,t2,t3,t4}
154− subps %3, %3, %4 ; {r5,i5,r7,i7}
155− shufps %4, %3, %3, 0xb1 ; {i5,r5,i7,r7}
156− mulps %3, %3, [ps_root2mppm] ; {-r5,i5,r7,-i7}
157− mulps %4, %4, [ps_root2]
158− addps %3, %3, %4 ; {t8,t7,ta,t9}
159− shufps %4, %6, %3, 0x9c ; {t1,t4,t7,ta}
160− shufps %6, %6, %3, 0x36 ; {t3,t2,t9,t8}
161− subps %3, %6, %4 ; {t6,t5,tc,tb}
162− addps %6, %6, %4 ; {t1,t2,t9,ta}
163− shufps %5, %6, %3, 0x8d ; {t2,ta,t6,tc}
164− shufps %6, %6, %3, 0xd8 ; {t1,t9,t5,tb}
165− subps %3, %1, %6 ; {r4,r5,r6,r7}
166− addps %1, %1, %6 ; {r0,r1,r2,r3}
167− subps %4, %2, %5 ; {i4,i5,i6,i7}
168− addps %2, %2, %5 ; {i0,i1,i2,i3}
169−%endmacro
170−
171−%macro INTERL 5
172−%if cpuflag(avx)
173− vunpckhps %3, %2, %1
174− vunpcklps %2, %2, %1
175− vextractf128 %4(%5), %2, 0
176− vextractf128 %4 %+ H(%5), %3, 0
177− vextractf128 %4(%5 + 1), %2, 1
178− vextractf128 %4 %+ H(%5 + 1), %3, 1
179−%elif cpuflag(sse)
180− mova %3, %2
181− unpcklps %2, %1
182− unpckhps %3, %1
183− mova %4(%5), %2
184− mova %4(%5+1), %3
185−%endif
186−%endmacro
187−
188−; scheduled for cpu-bound sizes
189−%macro PASS_SMALL 3 ; (to load m4-m7), wre, wim
190−IF%1 mova m4, Z(4)
191−IF%1 mova m5, Z(5)
192− mova m0, %2 ; wre
193− mova m1, %3 ; wim
194− mulps m2, m4, m0 ; r2*wre
195−IF%1 mova m6, Z2(6)
196− mulps m3, m5, m1 ; i2*wim
197−IF%1 mova m7, Z2(7)
198− mulps m4, m4, m1 ; r2*wim
199− mulps m5, m5, m0 ; i2*wre
200− addps m2, m2, m3 ; r2*wre + i2*wim
201− mulps m3, m1, m7 ; i3*wim
202− subps m5, m5, m4 ; i2*wre - r2*wim
203− mulps m1, m1, m6 ; r3*wim
204− mulps m4, m0, m6 ; r3*wre
205− mulps m0, m0, m7 ; i3*wre
206− subps m4, m4, m3 ; r3*wre - i3*wim
207− mova m3, Z(0)
208− addps m0, m0, m1 ; i3*wre + r3*wim
209− subps m1, m4, m2 ; t3
210− addps m4, m4, m2 ; t5
211− subps m3, m3, m4 ; r2
212− addps m4, m4, Z(0) ; r0
213− mova m6, Z(2)
214− mova Z(4), m3
215− mova Z(0), m4
216− subps m3, m5, m0 ; t4
217− subps m4, m6, m3 ; r3
218− addps m3, m3, m6 ; r1
219− mova Z2(6), m4
220− mova Z(2), m3
221− mova m2, Z(3)
222− addps m3, m5, m0 ; t6
223− subps m2, m2, m1 ; i3
224− mova m7, Z(1)
225− addps m1, m1, Z(3) ; i1
226− mova Z2(7), m2
227− mova Z(3), m1
228− subps m4, m7, m3 ; i2
229− addps m3, m3, m7 ; i0
230− mova Z(5), m4
231− mova Z(1), m3
232−%endmacro
233−
234−; scheduled to avoid store->load aliasing
235−%macro PASS_BIG 1 ; (!interleave)
236− mova m4, Z(4) ; r2
237− mova m5, Z(5) ; i2
238− mova m0, [wq] ; wre
239− mova m1, [wq+o1q] ; wim
240− mulps m2, m4, m0 ; r2*wre
241− mova m6, Z2(6) ; r3
242− mulps m3, m5, m1 ; i2*wim
243− mova m7, Z2(7) ; i3
244− mulps m4, m4, m1 ; r2*wim
245− mulps m5, m5, m0 ; i2*wre
246− addps m2, m2, m3 ; r2*wre + i2*wim
247− mulps m3, m1, m7 ; i3*wim
248− mulps m1, m1, m6 ; r3*wim
249− subps m5, m5, m4 ; i2*wre - r2*wim
250− mulps m4, m0, m6 ; r3*wre
251− mulps m0, m0, m7 ; i3*wre
252− subps m4, m4, m3 ; r3*wre - i3*wim
253− mova m3, Z(0)
254− addps m0, m0, m1 ; i3*wre + r3*wim
255− subps m1, m4, m2 ; t3
256− addps m4, m4, m2 ; t5
257− subps m3, m3, m4 ; r2
258− addps m4, m4, Z(0) ; r0
259− mova m6, Z(2)
260− mova Z(4), m3
261− mova Z(0), m4
262− subps m3, m5, m0 ; t4
263− subps m4, m6, m3 ; r3
264− addps m3, m3, m6 ; r1
265−IF%1 mova Z2(6), m4
266−IF%1 mova Z(2), m3
267− mova m2, Z(3)
268− addps m5, m5, m0 ; t6
269− subps m2, m2, m1 ; i3
270− mova m7, Z(1)
271− addps m1, m1, Z(3) ; i1
272−IF%1 mova Z2(7), m2
273−IF%1 mova Z(3), m1
274− subps m6, m7, m5 ; i2
275− addps m5, m5, m7 ; i0
276−IF%1 mova Z(5), m6
277−IF%1 mova Z(1), m5
278−%if %1==0
279− INTERL m1, m3, m7, Z, 2
280− INTERL m2, m4, m0, Z2, 6
281−
282− mova m1, Z(0)
283− mova m2, Z(4)
284−
285− INTERL m5, m1, m3, Z, 0
286− INTERL m6, m2, m7, Z, 4
287−%endif
288−%endmacro
289−
290−%define Z(x) [r0+mmsize*x]
291−%define Z2(x) [r0+mmsize*x]
292−%define ZH(x) [r0+mmsize*x+mmsize/2]
293−
294−INIT_YMM avx
295−
296−%if HAVE_AVX_EXTERNAL
297−align 16
298−fft8_avx:
299− mova m0, Z(0)
300− mova m1, Z(1)
301− T8_AVX m0, m1, m2, m3, m4
302− mova Z(0), m0
303− mova Z(1), m1
304− ret
305−
306−
307−align 16
308−fft16_avx:
309− mova m2, Z(2)
310− mova m3, Z(3)
311− T4_SSE m2, m3, m7
312−
313− mova m0, Z(0)
314− mova m1, Z(1)
315− T8_AVX m0, m1, m4, m5, m7
316−
317− mova m4, [ps_cos16_1]
318− mova m5, [ps_cos16_2]
319− vmulps m6, m2, m4
320− vmulps m7, m3, m5
321− vaddps m7, m7, m6
322− vmulps m2, m2, m5
323− vmulps m3, m3, m4
324− vsubps m3, m3, m2
325− vblendps m2, m7, m3, 0xf0
326− vperm2f128 m3, m7, m3, 0x21
327− vaddps m4, m2, m3
328− vsubps m2, m3, m2
329− vperm2f128 m2, m2, m2, 0x01
330− vsubps m3, m1, m2
331− vaddps m1, m1, m2
332− vsubps m5, m0, m4
333− vaddps m0, m0, m4
334− vextractf128 Z(0), m0, 0
335− vextractf128 ZH(0), m1, 0
336− vextractf128 Z(1), m0, 1
337− vextractf128 ZH(1), m1, 1
338− vextractf128 Z(2), m5, 0
339− vextractf128 ZH(2), m3, 0
340− vextractf128 Z(3), m5, 1
341− vextractf128 ZH(3), m3, 1
342− ret
343−
344−align 16
345−fft32_avx:
346− call fft16_avx
347−
348− mova m0, Z(4)
349− mova m1, Z(5)
350−
351− T4_SSE m0, m1, m4
352−
353− mova m2, Z(6)
354− mova m3, Z(7)
355−
356− T8_SSE m0, m1, m2, m3, m4, m6
357− ; m0={r0,r1,r2,r3,r8, r9, r10,r11} m1={i0,i1,i2,i3,i8, i9, i10,i11}
358− ; m2={r4,r5,r6,r7,r12,r13,r14,r15} m3={i4,i5,i6,i7,i12,i13,i14,i15}
359−
360− vperm2f128 m4, m0, m2, 0x20
361− vperm2f128 m5, m1, m3, 0x20
362− vperm2f128 m6, m0, m2, 0x31
363− vperm2f128 m7, m1, m3, 0x31
364−
365− PASS_SMALL 0, [cos_32], [cos_32+32]
366−
367− ret
368−
369−fft32_interleave_avx:
370− call fft32_avx
371− mov r2d, 32
372−.deint_loop:
373− mova m2, Z(0)
374− mova m3, Z(1)
375− vunpcklps m0, m2, m3
376− vunpckhps m1, m2, m3
377− vextractf128 Z(0), m0, 0
378− vextractf128 ZH(0), m1, 0
379− vextractf128 Z(1), m0, 1
380− vextractf128 ZH(1), m1, 1
381− add r0, mmsize*2
382− sub r2d, mmsize/4
383− jg .deint_loop
384− ret
385−
386−%endif
387−
388−INIT_XMM sse
389−
390−align 16
391−fft4_avx:
392−fft4_sse:
393− mova m0, Z(0)
394− mova m1, Z(1)
395− T4_SSE m0, m1, m2
396− mova Z(0), m0
397− mova Z(1), m1
398− ret
399−
400−align 16
401−fft8_sse:
402− mova m0, Z(0)
403− mova m1, Z(1)
404− T4_SSE m0, m1, m2
405− mova m2, Z(2)
406− mova m3, Z(3)
407− T8_SSE m0, m1, m2, m3, m4, m5
408− mova Z(0), m0
409− mova Z(1), m1
410− mova Z(2), m2
411− mova Z(3), m3
412− ret
413−
414−align 16
415−fft16_sse:
416− mova m0, Z(0)
417− mova m1, Z(1)
418− T4_SSE m0, m1, m2
419− mova m2, Z(2)
420− mova m3, Z(3)
421− T8_SSE m0, m1, m2, m3, m4, m5
422− mova m4, Z(4)
423− mova m5, Z(5)
424− mova Z(0), m0
425− mova Z(1), m1
426− mova Z(2), m2
427− mova Z(3), m3
428− T4_SSE m4, m5, m6
429− mova m6, Z2(6)
430− mova m7, Z2(7)
431− T4_SSE m6, m7, m0
432− PASS_SMALL 0, [cos_16], [cos_16+16]
433− ret
434−
435−
436−%define Z(x) [zcq + o1q*(x&6) + mmsize*(x&1)]
437−%define Z2(x) [zcq + o3q + mmsize*(x&1)]
438−%define ZH(x) [zcq + o1q*(x&6) + mmsize*(x&1) + mmsize/2]
439−%define Z2H(x) [zcq + o3q + mmsize*(x&1) + mmsize/2]
440−
441−%macro DECL_PASS 2+ ; name, payload
442−align 16
443−%1:
444−DEFINE_ARGS zc, w, n, o1, o3
445− lea o3q, [nq*3]
446− lea o1q, [nq*8]
447− shl o3q, 4
448−.loop:
449− %2
450− add zcq, mmsize*2
451− add wq, mmsize
452− sub nd, mmsize/8
453− jg .loop
454− rep ret
455−%endmacro
456−
457−%macro FFT_DISPATCH 2; clobbers 5 GPRs, 8 XMMs
458− lea r2, [dispatch_tab%1]
459− mov r2, [r2 + (%2q-2)*gprsize]
460−%ifdef PIC
461− lea r3, [$$]
462− add r2, r3
463−%endif
464− call r2
465−%endmacro ; FFT_DISPATCH
466−
467−INIT_YMM avx
468−
469−%if HAVE_AVX_EXTERNAL
470−DECL_PASS pass_avx, PASS_BIG 1
471−DECL_PASS pass_interleave_avx, PASS_BIG 0
472−
473−cglobal fft_calc, 2,5,8
474− mov r3d, [r0 + FFTContext.nbits]
475− mov r0, r1
476− mov r1, r3
477− FFT_DISPATCH _interleave %+ SUFFIX, r1
478− RET
479−
480−%endif
481−
482−INIT_XMM sse
483−
484−DECL_PASS pass_sse, PASS_BIG 1
485−DECL_PASS pass_interleave_sse, PASS_BIG 0
486−
487−INIT_XMM sse
488−cglobal fft_calc, 2,5,8
489− mov r3d, [r0 + FFTContext.nbits]
490− PUSH r1
491− PUSH r3
492− mov r0, r1
493− mov r1, r3
494− FFT_DISPATCH _interleave %+ SUFFIX, r1
495− POP rcx
496− POP r4
497− cmp rcx, 3+(mmsize/16)
498− jg .end
499− mov r2, -1
500− add rcx, 3
501− shl r2, cl
502− sub r4, r2
503−.loop:
504− movaps xmm0, [r4 + r2]
505− movaps xmm1, xmm0
506− unpcklps xmm0, [r4 + r2 + 16]
507− unpckhps xmm1, [r4 + r2 + 16]
508− movaps [r4 + r2], xmm0
509− movaps [r4 + r2 + 16], xmm1
510− add r2, mmsize*2
511− jl .loop
512−.end:
513− RET
514−
515−cglobal fft_permute, 2,7,1
516− mov r4, [r0 + FFTContext.revtab]
517− mov r5, [r0 + FFTContext.tmpbuf]
518− mov ecx, [r0 + FFTContext.nbits]
519− mov r2, 1
520− shl r2, cl
521− xor r0, r0
522−%if ARCH_X86_32
523− mov r1, r1m
524−%endif
525−.loop:
526− movaps xmm0, [r1 + 8*r0]
527− movzx r6, word [r4 + 2*r0]
528− movzx r3, word [r4 + 2*r0 + 2]
529− movlps [r5 + 8*r6], xmm0
530− movhps [r5 + 8*r3], xmm0
531− add r0, 2
532− cmp r0, r2
533− jl .loop
534− shl r2, 3
535− add r1, r2
536− add r5, r2
537− neg r2
538−; nbits >= 2 (FFT4) and sizeof(FFTComplex)=8 => at least 32B
539−.loopcopy:
540− movaps xmm0, [r5 + r2]
541− movaps xmm1, [r5 + r2 + 16]
542− movaps [r1 + r2], xmm0
543− movaps [r1 + r2 + 16], xmm1
544− add r2, 32
545− jl .loopcopy
546− RET
547−
548−INIT_XMM sse
549−cglobal imdct_calc, 3,5,3
550− mov r3d, [r0 + FFTContext.mdctsize]
551− mov r4, [r0 + FFTContext.imdcthalf]
552− add r1, r3
553− PUSH r3
554− PUSH r1
555−%if ARCH_X86_32
556− push r2
557− push r1
558− push r0
559−%else
560− sub rsp, 8+32*WIN64 ; allocate win64 shadow space
561−%endif
562− call r4
563−%if ARCH_X86_32
564− add esp, 12
565−%else
566− add rsp, 8+32*WIN64
567−%endif
568− POP r1
569− POP r3
570− lea r0, [r1 + 2*r3]
571− mov r2, r3
572− sub r3, mmsize
573− neg r2
574− mova m2, [ps_neg]
575−.loop:
576− mova m0, [r1 + r3]
577− mova m1, [r0 + r2]
578− shufps m0, m0, 0x1b
579− shufps m1, m1, 0x1b
580− xorps m0, m2
581− mova [r0 + r3], m1
582− mova [r1 + r2], m0
583− sub r3, mmsize
584− add r2, mmsize
585− jl .loop
586− RET
587−
588−%ifdef PIC
589−%define SECTION_REL - $$
590−%else
591−%define SECTION_REL
592−%endif
593−
594−%macro DECL_FFT 1-2 ; nbits, suffix
595−%ifidn %0, 1
596−%xdefine fullsuffix SUFFIX
597−%else
598−%xdefine fullsuffix %2 %+ SUFFIX
599−%endif
600−%xdefine list_of_fft fft4 %+ SUFFIX SECTION_REL, fft8 %+ SUFFIX SECTION_REL
601−%if %1>=5
602−%xdefine list_of_fft list_of_fft, fft16 %+ SUFFIX SECTION_REL
603−%endif
604−%if %1>=6
605−%xdefine list_of_fft list_of_fft, fft32 %+ fullsuffix SECTION_REL
606−%endif
607−
608−%assign n 1<<%1
609−%rep 18-%1
610−%assign n2 n/2
611−%assign n4 n/4
612−%xdefine list_of_fft list_of_fft, fft %+ n %+ fullsuffix SECTION_REL
613−
614−align 16
615−fft %+ n %+ fullsuffix:
616− call fft %+ n2 %+ SUFFIX
617− add r0, n*4 - (n&(-2<<%1))
618− call fft %+ n4 %+ SUFFIX
619− add r0, n*2 - (n2&(-2<<%1))
620− call fft %+ n4 %+ SUFFIX
621− sub r0, n*6 + (n2&(-2<<%1))
622− lea r1, [cos_ %+ n]
623− mov r2d, n4/2
624− jmp pass %+ fullsuffix
625−
626−%assign n n*2
627−%endrep
628−%undef n
629−
630−align 8
631−dispatch_tab %+ fullsuffix: pointer list_of_fft
632−%endmacro ; DECL_FFT
633−
634−%if HAVE_AVX_EXTERNAL
635−INIT_YMM avx
636−DECL_FFT 6
637−DECL_FFT 6, _interleave
638−%endif
639−INIT_XMM sse
640−DECL_FFT 5
641−DECL_FFT 5, _interleave
642−
643−INIT_XMM sse
644−%undef mulps
645−%undef addps
646−%undef subps
647−%undef unpcklps
648−%undef unpckhps
649−
650−%macro PREROTATER 5 ;-2*k, 2*k, input+n4, tcos+n8, tsin+n8
651− movaps xmm0, [%3+%2*4]
652− movaps xmm1, [%3+%1*4-0x10]
653− movaps xmm2, xmm0
654− shufps xmm0, xmm1, 0x88
655− shufps xmm1, xmm2, 0x77
656− movlps xmm4, [%4+%2*2]
657− movlps xmm5, [%5+%2*2+0x0]
658− movhps xmm4, [%4+%1*2-0x8]
659− movhps xmm5, [%5+%1*2-0x8]
660− movaps xmm2, xmm0
661− movaps xmm3, xmm1
662− mulps xmm0, xmm5
663− mulps xmm1, xmm4
664− mulps xmm2, xmm4
665− mulps xmm3, xmm5
666− subps xmm1, xmm0
667− addps xmm2, xmm3
668− movaps xmm0, xmm1
669− unpcklps xmm1, xmm2
670− unpckhps xmm0, xmm2
671−%endmacro
672−
673−%macro CMUL 6 ;j, xmm0, xmm1, 3, 4, 5
674− mulps m6, %3, [%5+%1]
675− mulps m7, %2, [%5+%1]
676− mulps %2, %2, [%6+%1]
677− mulps %3, %3, [%6+%1]
678− subps %2, %2, m6
679− addps %3, %3, m7
680−%endmacro
681−
682−%macro POSROTATESHUF 5 ;j, k, z+n8, tcos+n8, tsin+n8
683−.post:
684−%if cpuflag(avx)
685− vmovaps ymm1, [%3+%1*2]
686− vmovaps ymm0, [%3+%1*2+0x20]
687− vmovaps ymm3, [%3+%2*2]
688− vmovaps ymm2, [%3+%2*2+0x20]
689−
690− CMUL %1, ymm0, ymm1, %3, %4, %5
691− CMUL %2, ymm2, ymm3, %3, %4, %5
692− vshufps ymm1, ymm1, ymm1, 0x1b
693− vshufps ymm3, ymm3, ymm3, 0x1b
694− vperm2f128 ymm1, ymm1, ymm1, 0x01
695− vperm2f128 ymm3, ymm3, ymm3, 0x01
696− vunpcklps ymm6, ymm2, ymm1
697− vunpckhps ymm4, ymm2, ymm1
698− vunpcklps ymm7, ymm0, ymm3
699− vunpckhps ymm5, ymm0, ymm3
700−
701− vextractf128 [%3+%1*2], ymm7, 0
702− vextractf128 [%3+%1*2+0x10], ymm5, 0
703− vextractf128 [%3+%1*2+0x20], ymm7, 1
704− vextractf128 [%3+%1*2+0x30], ymm5, 1
705−
706− vextractf128 [%3+%2*2], ymm6, 0
707− vextractf128 [%3+%2*2+0x10], ymm4, 0
708− vextractf128 [%3+%2*2+0x20], ymm6, 1
709− vextractf128 [%3+%2*2+0x30], ymm4, 1
710− sub %2, 0x20
711− add %1, 0x20
712− jl .post
713−%else
714− movaps xmm1, [%3+%1*2]
715− movaps xmm0, [%3+%1*2+0x10]
716− CMUL %1, xmm0, xmm1, %3, %4, %5
717− movaps xmm5, [%3+%2*2]
718− movaps xmm4, [%3+%2*2+0x10]
719− CMUL %2, xmm4, xmm5, %3, %4, %5
720− shufps xmm1, xmm1, 0x1b
721− shufps xmm5, xmm5, 0x1b
722− movaps xmm6, xmm4
723− unpckhps xmm4, xmm1
724− unpcklps xmm6, xmm1
725− movaps xmm2, xmm0
726− unpcklps xmm0, xmm5
727− unpckhps xmm2, xmm5
728− movaps [%3+%2*2], xmm6
729− movaps [%3+%2*2+0x10], xmm4
730− movaps [%3+%1*2], xmm0
731− movaps [%3+%1*2+0x10], xmm2
732− sub %2, 0x10
733− add %1, 0x10
734− jl .post
735−%endif
736−%endmacro
737−
738−%macro DECL_IMDCT 0
739−cglobal imdct_half, 3,12,8; FFTContext *s, FFTSample *output, const FFTSample *input
740−%if ARCH_X86_64
741−%define rrevtab r7
742−%define rtcos r8
743−%define rtsin r9
744−%else
745−%define rrevtab r6
746−%define rtsin r6
747−%define rtcos r5
748−%endif
749− mov r3d, [r0+FFTContext.mdctsize]
750− add r2, r3
751− shr r3, 1
752− mov rtcos, [r0+FFTContext.tcos]
753− mov rtsin, [r0+FFTContext.tsin]
754− add rtcos, r3
755− add rtsin, r3
756−%if ARCH_X86_64 == 0
757− push rtcos
758− push rtsin
759−%endif
760− shr r3, 1
761− mov rrevtab, [r0+FFTContext.revtab]
762− add rrevtab, r3
763−%if ARCH_X86_64 == 0
764− push rrevtab
765−%endif
766−
767− sub r3, 4
768−%if ARCH_X86_64
769− xor r4, r4
770− sub r4, r3
771−%endif
772−.pre:
773−%if ARCH_X86_64 == 0
774−;unspill
775− xor r4, r4
776− sub r4, r3
777− mov rtcos, [esp+8]
778− mov rtsin, [esp+4]
779−%endif
780−
781− PREROTATER r4, r3, r2, rtcos, rtsin
782−%if ARCH_X86_64
783− movzx r5, word [rrevtab+r4-4]
784− movzx r6, word [rrevtab+r4-2]
785− movzx r10, word [rrevtab+r3]
786− movzx r11, word [rrevtab+r3+2]
787− movlps [r1+r5 *8], xmm0
788− movhps [r1+r6 *8], xmm0
789− movlps [r1+r10*8], xmm1
790− movhps [r1+r11*8], xmm1
791− add r4, 4
792−%else
793− mov r6, [esp]
794− movzx r5, word [r6+r4-4]
795− movzx r4, word [r6+r4-2]
796− movlps [r1+r5*8], xmm0
797− movhps [r1+r4*8], xmm0
798− movzx r5, word [r6+r3]
799− movzx r4, word [r6+r3+2]
800− movlps [r1+r5*8], xmm1
801− movhps [r1+r4*8], xmm1
802−%endif
803− sub r3, 4
804− jns .pre
805−
806− mov r5, r0
807− mov r6, r1
808− mov r0, r1
809− mov r1d, [r5+FFTContext.nbits]
810−
811− FFT_DISPATCH SUFFIX, r1
812−
813− mov r0d, [r5+FFTContext.mdctsize]
814− add r6, r0
815− shr r0, 1
816−%if ARCH_X86_64 == 0
817−%define rtcos r2
818−%define rtsin r3
819− mov rtcos, [esp+8]
820− mov rtsin, [esp+4]
821−%endif
822− neg r0
823− mov r1, -mmsize
824− sub r1, r0
825− POSROTATESHUF r0, r1, r6, rtcos, rtsin
826−%if ARCH_X86_64 == 0
827− add esp, 12
828−%endif
829− RET
830−%endmacro
831−
832−DECL_IMDCT
833−
834−INIT_YMM avx
835−
836−%if HAVE_AVX_EXTERNAL
837−DECL_IMDCT
838−%endif
libavcodec/x86/fft.h+0−32
@@ -1,32 +0,0 @@
1−/*
2− * This file is part of FFmpeg.
3− *
4− * FFmpeg is free software; you can redistribute it and/or
5− * modify it under the terms of the GNU Lesser General Public
6− * License as published by the Free Software Foundation; either
7− * version 2.1 of the License, or (at your option) any later version.
8− *
9− * FFmpeg is distributed in the hope that it will be useful,
10− * but WITHOUT ANY WARRANTY; without even the implied warranty of
11− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12− * Lesser General Public License for more details.
13− *
14− * You should have received a copy of the GNU Lesser General Public
15− * License along with FFmpeg; if not, write to the Free Software
16− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17− */
18−
19−#ifndef AVCODEC_X86_FFT_H
20−#define AVCODEC_X86_FFT_H
21−
22−#include "libavcodec/fft.h"
23−
24−void ff_fft_permute_sse(FFTContext *s, FFTComplex *z);
25−void ff_fft_calc_avx(FFTContext *s, FFTComplex *z);
26−void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
27−
28−void ff_imdct_calc_sse(FFTContext *s, FFTSample *output, const FFTSample *input);
29−void ff_imdct_half_sse(FFTContext *s, FFTSample *output, const FFTSample *input);
30−void ff_imdct_half_avx(FFTContext *s, FFTSample *output, const FFTSample *input);
31−
32−#endif /* AVCODEC_X86_FFT_H */
libavcodec/x86/fft_init.c+0−47
@@ -1,47 +0,0 @@
1−/*
2− * This file is part of FFmpeg.
3− *
4− * FFmpeg is free software; you can redistribute it and/or
5− * modify it under the terms of the GNU Lesser General Public
6− * License as published by the Free Software Foundation; either
7− * version 2.1 of the License, or (at your option) any later version.
8− *
9− * FFmpeg is distributed in the hope that it will be useful,
10− * but WITHOUT ANY WARRANTY; without even the implied warranty of
11− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12− * Lesser General Public License for more details.
13− *
14− * You should have received a copy of the GNU Lesser General Public
15− * License along with FFmpeg; if not, write to the Free Software
16− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17− */
18−
19−#include "config.h"
20−
21−#include "libavutil/attributes.h"
22−#include "libavutil/cpu.h"
23−#include "libavutil/x86/cpu.h"
24−
25−#include "fft.h"
26−
27−av_cold void ff_fft_init_x86(FFTContext *s)
28−{
29− int cpu_flags = av_get_cpu_flags();
30−
31− if (s->nbits > 16)
32− return;
33−
34− if (EXTERNAL_SSE(cpu_flags)) {
35− s->imdct_calc = ff_imdct_calc_sse;
36− s->imdct_half = ff_imdct_half_sse;
37− s->fft_permute = ff_fft_permute_sse;
38− s->fft_calc = ff_fft_calc_sse;
39− s->fft_permutation = FF_FFT_PERM_SWAP_LSBS;
40− }
41−
42− if (EXTERNAL_AVX_FAST(cpu_flags) && s->nbits >= 5) {
43− s->imdct_half = ff_imdct_half_avx;
44− s->fft_calc = ff_fft_calc_avx;
45− s->fft_permutation = FF_FFT_PERM_AVX;
46− }
47−}
tests/Makefile+0−1
@@ -181,7 +181,6 @@ include $(SRC_PATH)/tests/fate/enc_external.mak · 180 unmodified lines
181181 # Must be included after lavf-video.mak
182182 include $(SRC_PATH)/tests/fate/ffmpeg.mak
183183 include $(SRC_PATH)/tests/fate/ffprobe.mak
184−include $(SRC_PATH)/tests/fate/fft.mak
185184 include $(SRC_PATH)/tests/fate/fifo-muxer.mak
186185 include $(SRC_PATH)/tests/fate/filter-audio.mak
187186 # Must be included after vcodec.mak
tests/fate/fft.mak+0−83
@@ -1,83 +0,0 @@
1−define DEF_FFT
2−FATE_DCT-$(CONFIG_DCT) += fate-dct1d-$(1) fate-idct1d-$(1)
3−FATE_FFT-$(CONFIG_FFT) += fate-fft-$(1) fate-ifft-$(1)
4−FATE_MDCT-$(CONFIG_MDCT) += fate-mdct-$(1) fate-imdct-$(1)
5−FATE_RDFT-$(CONFIG_RDFT) += fate-rdft-$(1) fate-irdft-$(1)
6−
7−fate-fft-$(N): ARGS = -n$(1)
8−fate-ifft-$(N): ARGS = -n$(1) -i
9−fate-mdct-$(N): ARGS = -n$(1) -m
10−fate-imdct-$(N): ARGS = -n$(1) -m -i
11−fate-rdft-$(N): ARGS = -n$(1) -r
12−fate-irdft-$(N): ARGS = -n$(1) -r -i
13−fate-dct1d-$(N): ARGS = -n$(1) -d
14−fate-idct1d-$(N): ARGS = -n$(1) -d -i
15−endef
16−
17−$(foreach N, 4 5 6 7 8 9 10 11 12, $(eval $(call DEF_FFT,$(N))))
18−
19−fate-dct-float: $(FATE_DCT-yes)
20−fate-fft-float: $(FATE_FFT-yes)
21−fate-mdct-float: $(FATE_MDCT-yes)
22−fate-rdft-float: $(FATE_RDFT-yes)
23−
24−FATE_FFT_ALL = $(FATE_DCT-yes) $(FATE_FFT-yes) $(FATE_MDCT-yes) $(FATE_RDFT-yes)
25−
26−$(FATE_FFT_ALL): libavcodec/tests/fft$(EXESUF)
27−$(FATE_FFT_ALL): CMD = run libavcodec/tests/fft$(EXESUF) $(CPUFLAGS:%=-c%) $(ARGS)
28−
29−$(FATE_FFT_ALL): CMP = null
30−
31−define DEF_FFT_FIXED32
32−FATE_FFT_FIXED32 += fate-fft-fixed32-$(1) fate-ifft-fixed32-$(1) \
33− fate-mdct-fixed32-$(1) fate-imdct-fixed32-$(1)
34−
35−fate-fft-fixed32-$(1): ARGS = -n$(1)
36−fate-ifft-fixed32-$(1): ARGS = -n$(1) -i
37−#fate-mdct-fixed32-$(1): ARGS = -n$(1) -m
38−fate-imdct-fixed32-$(1): ARGS = -n$(1) -m -i
39−endef
40−
41−$(foreach N, 4 5 6 7 8 9 10 11 12, $(eval $(call DEF_FFT_FIXED32,$(N))))
42−
43−fate-fft-fixed32: $(FATE_FFT_FIXED32)
44−$(FATE_FFT_FIXED32): libavcodec/tests/fft-fixed32$(EXESUF)
45−$(FATE_FFT_FIXED32): CMD = run libavcodec/tests/fft-fixed32$(EXESUF) $(CPUFLAGS:%=-c%) $(ARGS)
46−$(FATE_FFT_FIXED32): CMP = null
47−
48−define DEF_AV_FFT
49−FATE_AV_DCT-$(CONFIG_DCT) += fate-av-dct1d-$(1) fate-av-idct1d-$(1)
50−FATE_AV_FFT-$(CONFIG_FFT) += fate-av-fft-$(1) fate-av-ifft-$(1)
51−FATE_AV_MDCT-$(CONFIG_MDCT) += fate-av-mdct-$(1) fate-av-imdct-$(1)
52−FATE_AV_RDFT-$(CONFIG_RDFT) += fate-av-rdft-$(1) fate-av-irdft-$(1)
53−
54−fate-av-fft-$(N): ARGS = -n$(1)
55−fate-av-ifft-$(N): ARGS = -n$(1) -i
56−fate-av-mdct-$(N): ARGS = -n$(1) -m
57−fate-av-imdct-$(N): ARGS = -n$(1) -m -i
58−fate-av-rdft-$(N): ARGS = -n$(1) -r
59−fate-av-irdft-$(N): ARGS = -n$(1) -r -i
60−fate-av-dct1d-$(N): ARGS = -n$(1) -d
61−fate-av-idct1d-$(N): ARGS = -n$(1) -d -i
62−endef
63−
64−$(foreach N, 4 5 6 7 8 9 10 11 12, $(eval $(call DEF_AV_FFT,$(N))))
65−
66−fate-av-dct-float: $(FATE_AV_DCT-yes)
67−fate-av-fft-float: $(FATE_AV_FFT-yes)
68−fate-av-mdct-float: $(FATE_AV_MDCT-yes)
69−fate-av-rdft-float: $(FATE_AV_RDFT-yes)
70−
71−FATE_AV_FFT_ALL = $(FATE_AV_DCT-yes) $(FATE_AV_FFT-yes) $(FATE_AV_MDCT-yes) $(FATE_AV_RDFT-yes)
72−
73−$(FATE_AV_FFT_ALL): libavcodec/tests/avfft$(EXESUF)
74−$(FATE_AV_FFT_ALL): CMD = run libavcodec/tests/avfft$(EXESUF) $(CPUFLAGS:%=-c%) $(ARGS)
75−$(FATE_AV_FFT_ALL): CMP = null
76−
77−fate-dct: fate-dct-float
78−fate-fft: fate-fft-float fate-fft-fixed32
79−fate-mdct: fate-mdct-float
80−fate-rdft: fate-rdft-float
81−
82−FATE-$(call ALLYES, AVCODEC FFT MDCT) += $(FATE_FFT_ALL) $(FATE_FFT_FIXED32) $(FATE_AV_FFT_ALL)
83−fate-fft-all: $(FATE_FFT_ALL) $(FATE_FFT_FIXED32) $(FATE_AV_FFT_ALL)

Keyboard shortcuts

General

Show keyboard shortcuts
?
Focus the current search
/

Issues and pull requests

Open filters
f

Issues, pull requests and Inbox

Select next item
j↓
Select previous item
k↑
Open selected item
Enter
Dismiss filters or clear selection
Esc

Repository code

Go to file
t