From ab12373956e92b865bced9b05b12971f062cfd3e Mon Sep 17 00:00:00 2001 From: Rong Yan Date: Wed, 18 Jun 2014 04:51:35 -0400 Subject: [PATCH] ppc: Fix the bug of fft for little endian environment on POWER7 and later Signed-off-by: Michael Niedermayer --- configure | 15 +- libavcodec/ppc/Makefile | 1 + libavcodec/ppc/fft_altivec.c | 12 + libavcodec/ppc/fft_vsx.c | 227 ++++++++++ libavcodec/ppc/fft_vsx.h | 830 +++++++++++++++++++++++++++++++++++ 5 files changed, 1084 insertions(+), 1 deletion(-) create mode 100644 libavcodec/ppc/fft_vsx.c create mode 100644 libavcodec/ppc/fft_vsx.h diff --git a/configure b/configure index d1eb8da92e..75740f976a 100755 --- a/configure +++ b/configure @@ -1548,6 +1548,7 @@ ARCH_EXT_LIST_PPC=" dcbzl ldbrx ppc4xx + vsx " ARCH_EXT_LIST_X86=" @@ -1933,6 +1934,7 @@ mipsdspr2_deps="mips" altivec_deps="ppc" ppc4xx_deps="ppc" +vsx_deps="ppc" cpunop_deps="i686" x86_64_select="i686" @@ -3691,6 +3693,7 @@ elif enabled mips; then elif enabled ppc; then disable ldbrx + disable vsx case $(tolower $cpu) in 601|ppc601|powerpc601) @@ -3718,7 +3721,7 @@ elif enabled ppc; then g5|970|ppc970|powerpc970) cpuflags="-mcpu=970" ;; - power[3-7]*) + power[3-8]*) cpuflags="-mcpu=$cpu" ;; cell) @@ -4286,6 +4289,12 @@ unsigned int endian = 'B' << 24 | 'I' << 16 | 'G' << 8 | 'E'; EOF od -t x1 $TMPO | grep -q '42 *49 *47 *45' && enable bigendian +if [ "$cpu" = "power7" ] || [ "$cpu" = "power8" ] ;then + if ! enabled bigendian ;then + enable vsx + fi +fi + if enabled asm; then enabled arm && nogas=die @@ -4415,6 +4424,9 @@ EOF enabled altivec || warn "Altivec disabled, possibly missing --cpu flag" fi + if enabled vsx; then + check_cflags -mvsx + fi elif enabled x86; then check_builtin rdtsc intrin.h "__rdtsc()" @@ -5256,6 +5268,7 @@ fi if enabled ppc; then echo "AltiVec enabled ${altivec-no}" echo "PPC 4xx optimizations ${ppc4xx-no}" + echo "PPC VSX optimizations ${vsx-no}" echo "dcbzl available ${dcbzl-no}" fi echo "debug symbols ${debug-no}" diff --git a/libavcodec/ppc/Makefile b/libavcodec/ppc/Makefile index b04aa0ebc9..bec7845198 100644 --- a/libavcodec/ppc/Makefile +++ b/libavcodec/ppc/Makefile @@ -26,4 +26,5 @@ ALTIVEC-OBJS-$(CONFIG_DSPUTIL) += ppc/dsputil_altivec.o \ ppc/int_altivec.o \ FFT-OBJS-$(HAVE_GNU_AS) += ppc/fft_altivec_s.o +FFT-OBJS-$(HAVE_VSX) += ppc/fft_vsx.o ALTIVEC-OBJS-$(CONFIG_FFT) += $(FFT-OBJS-yes) diff --git a/libavcodec/ppc/fft_altivec.c b/libavcodec/ppc/fft_altivec.c index 8d7fd465ed..675fa33a95 100644 --- a/libavcodec/ppc/fft_altivec.c +++ b/libavcodec/ppc/fft_altivec.c @@ -36,8 +36,12 @@ * It also assumes all FFTComplex are 8 bytes-aligned pairs of floats. */ +#if HAVE_VSX +#include "fft_vsx.h" +#else void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z); void ff_fft_calc_interleave_altivec(FFTContext *s, FFTComplex *z); +#endif #if HAVE_GNU_AS && HAVE_ALTIVEC static void imdct_half_altivec(FFTContext *s, FFTSample *output, const FFTSample *input) @@ -94,7 +98,11 @@ static void imdct_half_altivec(FFTContext *s, FFTSample *output, const FFTSample k--; } while(k >= 0); +#if HAVE_VSX + ff_fft_calc_vsx(s, (FFTComplex*)output); +#else ff_fft_calc_altivec(s, (FFTComplex*)output); +#endif /* post rotation + reordering */ j = -n32; @@ -147,7 +155,11 @@ av_cold void ff_fft_init_ppc(FFTContext *s) if (!PPC_ALTIVEC(av_get_cpu_flags())) return; +#if HAVE_VSX + s->fft_calc = ff_fft_calc_interleave_vsx; +#else s->fft_calc = ff_fft_calc_interleave_altivec; +#endif if (s->mdct_bits >= 5) { s->imdct_calc = imdct_calc_altivec; s->imdct_half = imdct_half_altivec; diff --git a/libavcodec/ppc/fft_vsx.c b/libavcodec/ppc/fft_vsx.c new file mode 100644 index 0000000000..e92975f74e --- /dev/null +++ b/libavcodec/ppc/fft_vsx.c @@ -0,0 +1,227 @@ +/* + * FFT transform, optimized with VSX built-in functions + * Copyright (c) 2014 Rong Yan + * + * This algorithm (though not any of the implementation details) is + * based on libdjbfft by D. J. Bernstein. + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + + +#include "config.h" +#include "libavutil/cpu.h" +#include "libavutil/ppc/types_altivec.h" +#include "libavutil/ppc/util_altivec.h" +#include "libavcodec/fft.h" +#include "libavcodec/fft-internal.h" +#include "fft_vsx.h" + +#if HAVE_VSX + +static void fft32_vsx_interleave(FFTComplex *z) +{ + fft16_vsx_interleave(z); + fft8_vsx_interleave(z+16); + fft8_vsx_interleave(z+24); + pass_vsx_interleave(z,ff_cos_32,4); +} + +static void fft64_vsx_interleave(FFTComplex *z) +{ + fft32_vsx_interleave(z); + fft16_vsx_interleave(z+32); + fft16_vsx_interleave(z+48); + pass_vsx_interleave(z,ff_cos_64, 8); +} +static void fft128_vsx_interleave(FFTComplex *z) +{ + fft64_vsx_interleave(z); + fft32_vsx_interleave(z+64); + fft32_vsx_interleave(z+96); + pass_vsx_interleave(z,ff_cos_128,16); +} +static void fft256_vsx_interleave(FFTComplex *z) +{ + fft128_vsx_interleave(z); + fft64_vsx_interleave(z+128); + fft64_vsx_interleave(z+192); + pass_vsx_interleave(z,ff_cos_256,32); +} +static void fft512_vsx_interleave(FFTComplex *z) +{ + fft256_vsx_interleave(z); + fft128_vsx_interleave(z+256); + fft128_vsx_interleave(z+384); + pass_vsx_interleave(z,ff_cos_512,64); +} +static void fft1024_vsx_interleave(FFTComplex *z) +{ + fft512_vsx_interleave(z); + fft256_vsx_interleave(z+512); + fft256_vsx_interleave(z+768); + pass_vsx_interleave(z,ff_cos_1024,128); + +} +static void fft2048_vsx_interleave(FFTComplex *z) +{ + fft1024_vsx_interleave(z); + fft512_vsx_interleave(z+1024); + fft512_vsx_interleave(z+1536); + pass_vsx_interleave(z,ff_cos_2048,256); +} +static void fft4096_vsx_interleave(FFTComplex *z) +{ + fft2048_vsx_interleave(z); + fft1024_vsx_interleave(z+2048); + fft1024_vsx_interleave(z+3072); + pass_vsx_interleave(z,ff_cos_4096, 512); +} +static void fft8192_vsx_interleave(FFTComplex *z) +{ + fft4096_vsx_interleave(z); + fft2048_vsx_interleave(z+4096); + fft2048_vsx_interleave(z+6144); + pass_vsx_interleave(z,ff_cos_8192,1024); +} +static void fft16384_vsx_interleave(FFTComplex *z) +{ + fft8192_vsx_interleave(z); + fft4096_vsx_interleave(z+8192); + fft4096_vsx_interleave(z+12288); + pass_vsx_interleave(z,ff_cos_16384,2048); +} +static void fft32768_vsx_interleave(FFTComplex *z) +{ + fft16384_vsx_interleave(z); + fft8192_vsx_interleave(z+16384); + fft8192_vsx_interleave(z+24576); + pass_vsx_interleave(z,ff_cos_32768,4096); +} +static void fft65536_vsx_interleave(FFTComplex *z) +{ + fft32768_vsx_interleave(z); + fft16384_vsx_interleave(z+32768); + fft16384_vsx_interleave(z+49152); + pass_vsx_interleave(z,ff_cos_65536,8192); +} + +static void fft32_vsx(FFTComplex *z) +{ + fft16_vsx(z); + fft8_vsx(z+16); + fft8_vsx(z+24); + pass_vsx(z,ff_cos_32,4); +} + +static void fft64_vsx(FFTComplex *z) +{ + fft32_vsx(z); + fft16_vsx(z+32); + fft16_vsx(z+48); + pass_vsx(z,ff_cos_64, 8); +} +static void fft128_vsx(FFTComplex *z) +{ + fft64_vsx(z); + fft32_vsx(z+64); + fft32_vsx(z+96); + pass_vsx(z,ff_cos_128,16); +} +static void fft256_vsx(FFTComplex *z) +{ + fft128_vsx(z); + fft64_vsx(z+128); + fft64_vsx(z+192); + pass_vsx(z,ff_cos_256,32); +} +static void fft512_vsx(FFTComplex *z) +{ + fft256_vsx(z); + fft128_vsx(z+256); + fft128_vsx(z+384); + pass_vsx(z,ff_cos_512,64); +} +static void fft1024_vsx(FFTComplex *z) +{ + fft512_vsx(z); + fft256_vsx(z+512); + fft256_vsx(z+768); + pass_vsx(z,ff_cos_1024,128); + +} +static void fft2048_vsx(FFTComplex *z) +{ + fft1024_vsx(z); + fft512_vsx(z+1024); + fft512_vsx(z+1536); + pass_vsx(z,ff_cos_2048,256); +} +static void fft4096_vsx(FFTComplex *z) +{ + fft2048_vsx(z); + fft1024_vsx(z+2048); + fft1024_vsx(z+3072); + pass_vsx(z,ff_cos_4096, 512); +} +static void fft8192_vsx(FFTComplex *z) +{ + fft4096_vsx(z); + fft2048_vsx(z+4096); + fft2048_vsx(z+6144); + pass_vsx(z,ff_cos_8192,1024); +} +static void fft16384_vsx(FFTComplex *z) +{ + fft8192_vsx(z); + fft4096_vsx(z+8192); + fft4096_vsx(z+12288); + pass_vsx(z,ff_cos_16384,2048); +} +static void fft32768_vsx(FFTComplex *z) +{ + fft16384_vsx(z); + fft8192_vsx(z+16384); + fft8192_vsx(z+24576); + pass_vsx(z,ff_cos_32768,4096); +} +static void fft65536_vsx(FFTComplex *z) +{ + fft32768_vsx(z); + fft16384_vsx(z+32768); + fft16384_vsx(z+49152); + pass_vsx(z,ff_cos_65536,8192); +} + +static void (* const fft_dispatch_vsx[])(FFTComplex*) = { + fft4_vsx, fft8_vsx, fft16_vsx, fft32_vsx, fft64_vsx, fft128_vsx, fft256_vsx, fft512_vsx, fft1024_vsx, + fft2048_vsx, fft4096_vsx, fft8192_vsx, fft16384_vsx, fft32768_vsx, fft65536_vsx, +}; +static void (* const fft_dispatch_vsx_interleave[])(FFTComplex*) = { + fft4_vsx_interleave, fft8_vsx_interleave, fft16_vsx_interleave, fft32_vsx_interleave, fft64_vsx_interleave, + fft128_vsx_interleave, fft256_vsx_interleave, fft512_vsx_interleave, fft1024_vsx_interleave, + fft2048_vsx_interleave, fft4096_vsx_interleave, fft8192_vsx_interleave, fft16384_vsx_interleave, fft32768_vsx_interleave, fft65536_vsx_interleave, +}; +void ff_fft_calc_interleave_vsx(FFTContext *s, FFTComplex *z) +{ + fft_dispatch_vsx_interleave[s->nbits-2](z); +} +void ff_fft_calc_vsx(FFTContext *s, FFTComplex *z) +{ + fft_dispatch_vsx[s->nbits-2](z); +} +#endif /* HAVE_VSX */ diff --git a/libavcodec/ppc/fft_vsx.h b/libavcodec/ppc/fft_vsx.h new file mode 100644 index 0000000000..a85475d160 --- /dev/null +++ b/libavcodec/ppc/fft_vsx.h @@ -0,0 +1,830 @@ +#ifndef AVCODEC_PPC_FFT_VSX_H +#define AVCODEC_PPC_FFT_VSX_H +/* + * FFT transform, optimized with VSX built-in functions + * Copyright (c) 2014 Rong Yan Copyright (c) 2009 Loren Merritt + * + * This algorithm (though not any of the implementation details) is + * based on libdjbfft by D. J. Bernstein, and fft_altivec_s.S. + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + + +#include "config.h" +#include "libavutil/cpu.h" +#include "libavutil/ppc/types_altivec.h" +#include "libavutil/ppc/util_altivec.h" +#include "libavcodec/fft.h" +#include "libavcodec/fft-internal.h" + +#if HAVE_VSX + +void ff_fft_calc_interleave_vsx(FFTContext *s, FFTComplex *z); +void ff_fft_calc_vsx(FFTContext *s, FFTComplex *z); + + +#define byte_2complex (2*sizeof(FFTComplex)) +#define byte_4complex (4*sizeof(FFTComplex)) +#define byte_6complex (6*sizeof(FFTComplex)) +#define byte_8complex (8*sizeof(FFTComplex)) +#define byte_10complex (10*sizeof(FFTComplex)) +#define byte_12complex (12*sizeof(FFTComplex)) +#define byte_14complex (14*sizeof(FFTComplex)) + +inline static void pass_vsx_interleave(FFTComplex *z, const FFTSample *wre, unsigned int n) +{ + int o1 = n<<1; + int o2 = n<<2; + int o3 = o1+o2; + int i1, i2, i3; + FFTSample* out = (FFTSample*)z; + const FFTSample *wim = wre+o1; + vec_f vz0, vzo1, vzo2, vzo3; + vec_f x0, x1, x2, x3; + vec_f x4, x5, x6, x7; + vec_f x8, x9, x10, x11; + vec_f x12, x13, x14, x15; + vec_f x16, x17, x18, x19; + vec_f x20, x21, x22, x23; + vec_f vz0plus1, vzo1plus1, vzo2plus1, vzo3plus1; + vec_f y0, y1, y2, y3; + vec_f y4, y5, y8, y9; + vec_f y10, y13, y14, y15; + vec_f y16, y17, y18, y19; + vec_f y20, y21, y22, y23; + vec_f wr1, wi1, wr0, wi0; + vec_f wr2, wi2, wr3, wi3; + vec_f xmulwi0, xmulwi1, ymulwi2, ymulwi3; + + n = n-2; + i1 = o1*sizeof(FFTComplex); + i2 = o2*sizeof(FFTComplex); + i3 = o3*sizeof(FFTComplex); + vzo2 = vec_ld(i2, &(out[0])); // zo2.r zo2.i z(o2+1).r z(o2+1).i + vzo2plus1 = vec_ld(i2+16, &(out[0])); + vzo3 = vec_ld(i3, &(out[0])); // zo3.r zo3.i z(o3+1).r z(o3+1).i + vzo3plus1 = vec_ld(i3+16, &(out[0])); + vz0 = vec_ld(0, &(out[0])); // z0.r z0.i z1.r z1.i + vz0plus1 = vec_ld(16, &(out[0])); + vzo1 = vec_ld(i1, &(out[0])); // zo1.r zo1.i z(o1+1).r z(o1+1).i + vzo1plus1 = vec_ld(i1+16, &(out[0])); + + x0 = vec_add(vzo2, vzo3); + x1 = vec_sub(vzo2, vzo3); + y0 = vec_add(vzo2plus1, vzo3plus1); + y1 = vec_sub(vzo2plus1, vzo3plus1); + + wr1 = vec_splats(wre[1]); + wi1 = vec_splats(wim[-1]); + wi2 = vec_splats(wim[-2]); + wi3 = vec_splats(wim[-3]); + wr2 = vec_splats(wre[2]); + wr3 = vec_splats(wre[3]); + + x2 = vec_perm(x0, x1, vcprm(2,s2,3,s3)); + x3 = vec_perm(x0, x1, vcprm(s3,3,s2,2)); + + y4 = vec_perm(y0, y1, vcprm(s1,1,s0,0)); + y5 = vec_perm(y0, y1, vcprm(s3,3,s2,2)); + y2 = vec_perm(y0, y1, vcprm(0,s0,1,s1)); + y3 = vec_perm(y0, y1, vcprm(2,s2,3,s3)); + + ymulwi2 = vec_mul(y4, wi2); + ymulwi3 = vec_mul(y5, wi3); + x4 = vec_mul(x2, wr1); + x5 = vec_mul(x3, wi1); + y8 = vec_madd(y2, wr2, ymulwi2); + y9 = vec_msub(y2, wr2, ymulwi2); + x6 = vec_add(x4, x5); + x7 = vec_sub(x4, x5); + y13 = vec_madd(y3, wr3, ymulwi3); + y14 = vec_msub(y3, wr3, ymulwi3); + + x8 = vec_perm(x6, x7, vcprm(0,1,s2,s3)); + y10 = vec_perm(y8, y9, vcprm(0,1,s2,s3)); + y15 = vec_perm(y13, y14, vcprm(0,1,s2,s3)); + + x9 = vec_perm(x0, x8, vcprm(0,1,s0,s2)); + x10 = vec_perm(x1, x8, vcprm(1,0,s3,s1)); + + y16 = vec_perm(y10, y15, vcprm(0,2,s0,s2)); + y17 = vec_perm(y10, y15, vcprm(3,1,s3,s1)); + + x11 = vec_add(vz0, x9); + x12 = vec_sub(vz0, x9); + x13 = vec_add(vzo1, x10); + x14 = vec_sub(vzo1, x10); + + y18 = vec_add(vz0plus1, y16); + y19 = vec_sub(vz0plus1, y16); + y20 = vec_add(vzo1plus1, y17); + y21 = vec_sub(vzo1plus1, y17); + + x15 = vec_perm(x13, x14, vcprm(0,s1,2,s3)); + x16 = vec_perm(x13, x14, vcprm(s0,1,s2,3)); + y22 = vec_perm(y20, y21, vcprm(0,s1,2,s3)); + y23 = vec_perm(y20, y21, vcprm(s0,1,s2,3)); + + + vec_st(x11, 0, &(out[0])); + vec_st(y18, 16, &(out[0])); + vec_st(x15, i1, &(out[0])); + vec_st(y22, i1+16, &(out[0])); + vec_st(x12, i2, &(out[0])); + vec_st(y19, i2+16, &(out[0])); + vec_st(x16, i3, &(out[0])); + vec_st(y23, i3+16, &(out[0])); + + do { + out += 8; + wre += 4; + wim -= 4; + wr0 = vec_splats(wre[0]); + wr1 = vec_splats(wre[1]); + wi0 = vec_splats(wim[0]); + wi1 = vec_splats(wim[-1]); + + wr2 = vec_splats(wre[2]); + wr3 = vec_splats(wre[3]); + wi2 = vec_splats(wim[-2]); + wi3 = vec_splats(wim[-3]); + + vzo2 = vec_ld(i2, &(out[0])); // zo2.r zo2.i z(o2+1).r z(o2+1).i + vzo2plus1 = vec_ld(i2+16, &(out[0])); + vzo3 = vec_ld(i3, &(out[0])); // zo3.r zo3.i z(o3+1).r z(o3+1).i + vzo3plus1 = vec_ld(i3+16, &(out[0])); + vz0 = vec_ld(0, &(out[0])); // z0.r z0.i z1.r z1.i + vz0plus1 = vec_ld(16, &(out[0])); + vzo1 = vec_ld(i1, &(out[0])); // zo1.r zo1.i z(o1+1).r z(o1+1).i + vzo1plus1 = vec_ld(i1+16, &(out[0])); + + x0 = vec_add(vzo2, vzo3); + x1 = vec_sub(vzo2, vzo3); + + y0 = vec_add(vzo2plus1, vzo3plus1); + y1 = vec_sub(vzo2plus1, vzo3plus1); + + x4 = vec_perm(x0, x1, vcprm(s1,1,s0,0)); + x5 = vec_perm(x0, x1, vcprm(s3,3,s2,2)); + x2 = vec_perm(x0, x1, vcprm(0,s0,1,s1)); + x3 = vec_perm(x0, x1, vcprm(2,s2,3,s3)); + + y2 = vec_perm(y0, y1, vcprm(0,s0,1,s1)); + y3 = vec_perm(y0, y1, vcprm(2,s2,3,s3)); + xmulwi0 = vec_mul(x4, wi0); + xmulwi1 = vec_mul(x5, wi1); + + y4 = vec_perm(y0, y1, vcprm(s1,1,s0,0)); + y5 = vec_perm(y0, y1, vcprm(s3,3,s2,2)); + + x8 = vec_madd(x2, wr0, xmulwi0); + x9 = vec_msub(x2, wr0, xmulwi0); + ymulwi2 = vec_mul(y4, wi2); + ymulwi3 = vec_mul(y5, wi3); + + x13 = vec_madd(x3, wr1, xmulwi1); + x14 = vec_msub(x3, wr1, xmulwi1); + + y8 = vec_madd(y2, wr2, ymulwi2); + y9 = vec_msub(y2, wr2, ymulwi2); + y13 = vec_madd(y3, wr3, ymulwi3); + y14 = vec_msub(y3, wr3, ymulwi3); + + x10 = vec_perm(x8, x9, vcprm(0,1,s2,s3)); + x15 = vec_perm(x13, x14, vcprm(0,1,s2,s3)); + + y10 = vec_perm(y8, y9, vcprm(0,1,s2,s3)); + y15 = vec_perm(y13, y14, vcprm(0,1,s2,s3)); + + x16 = vec_perm(x10, x15, vcprm(0,2,s0,s2)); + x17 = vec_perm(x10, x15, vcprm(3,1,s3,s1)); + + y16 = vec_perm(y10, y15, vcprm(0,2,s0,s2)); + y17 = vec_perm(y10, y15, vcprm(3,1,s3,s1)); + + x18 = vec_add(vz0, x16); + x19 = vec_sub(vz0, x16); + x20 = vec_add(vzo1, x17); + x21 = vec_sub(vzo1, x17); + + y18 = vec_add(vz0plus1, y16); + y19 = vec_sub(vz0plus1, y16); + y20 = vec_add(vzo1plus1, y17); + y21 = vec_sub(vzo1plus1, y17); + + x22 = vec_perm(x20, x21, vcprm(0,s1,2,s3)); + x23 = vec_perm(x20, x21, vcprm(s0,1,s2,3)); + + y22 = vec_perm(y20, y21, vcprm(0,s1,2,s3)); + y23 = vec_perm(y20, y21, vcprm(s0,1,s2,3)); + + vec_st(x18, 0, &(out[0])); + vec_st(y18, 16, &(out[0])); + vec_st(x22, i1, &(out[0])); + vec_st(y22, i1+16, &(out[0])); + vec_st(x19, i2, &(out[0])); + vec_st(y19, i2+16, &(out[0])); + vec_st(x23, i3, &(out[0])); + vec_st(y23, i3+16, &(out[0])); + } while (n-=2); +} + +inline static void fft2_vsx_interleave(FFTComplex *z) +{ + FFTSample r1, i1; + + r1 = z[0].re - z[1].re; + z[0].re += z[1].re; + z[1].re = r1; + + i1 = z[0].im - z[1].im; + z[0].im += z[1].im; + z[1].im = i1; + } + +inline static void fft4_vsx_interleave(FFTComplex *z) +{ + vec_f a, b, c, d; + float* out= (float*)z; + a = vec_ld(0, &(out[0])); + b = vec_ld(byte_2complex, &(out[0])); + + c = vec_perm(a, b, vcprm(0,1,s2,s1)); + d = vec_perm(a, b, vcprm(2,3,s0,s3)); + a = vec_add(c, d); + b = vec_sub(c, d); + + c = vec_perm(a, b, vcprm(0,1,s0,s1)); + d = vec_perm(a, b, vcprm(2,3,s3,s2)); + + a = vec_add(c, d); + b = vec_sub(c, d); + vec_st(a, 0, &(out[0])); + vec_st(b, byte_2complex, &(out[0])); +} + +inline static void fft8_vsx_interleave(FFTComplex *z) +{ + vec_f vz0, vz1, vz2, vz3; + vec_f x0, x1, x2, x3; + vec_f x4, x5, x6, x7; + vec_f x8, x9, x10, x11; + vec_f x12, x13, x14, x15; + vec_f x16, x17, x18, x19; + vec_f x20, x21, x22, x23; + vec_f x24, x25, x26, x27; + vec_f x28, x29, x30, x31; + vec_f x32, x33, x34; + + float* out= (float*)z; + vec_f vc1 = {sqrthalf, sqrthalf, sqrthalf, sqrthalf}; + + vz0 = vec_ld(0, &(out[0])); + vz1 = vec_ld(byte_2complex, &(out[0])); + vz2 = vec_ld(byte_4complex, &(out[0])); + vz3 = vec_ld(byte_6complex, &(out[0])); + + x0 = vec_perm(vz0, vz1, vcprm(0,1,s2,s1)); + x1 = vec_perm(vz0, vz1, vcprm(2,3,s0,s3)); + x2 = vec_perm(vz2, vz3, vcprm(2,1,s0,s1)); + x3 = vec_perm(vz2, vz3, vcprm(0,3,s2,s3)); + + x4 = vec_add(x0, x1); + x5 = vec_sub(x0, x1); + x6 = vec_add(x2, x3); + x7 = vec_sub(x2, x3); + + x8 = vec_perm(x4, x5, vcprm(0,1,s0,s1)); + x9 = vec_perm(x4, x5, vcprm(2,3,s3,s2)); + x10 = vec_perm(x6, x7, vcprm(2,1,s2,s1)); + x11 = vec_perm(x6, x7, vcprm(0,3,s0,s3)); + + x12 = vec_add(x8, x9); + x13 = vec_sub(x8, x9); + x14 = vec_add(x10, x11); + x15 = vec_sub(x10, x11); + x16 = vec_perm(x12, x13, vcprm(0,s0,1,s1)); + x17 = vec_perm(x14, x15, vcprm(0,s0,1,s1)); + x18 = vec_perm(x16, x17, vcprm(s0,s3,s2,s1)); + x19 = vec_add(x16, x18); // z0.r z2.r z0.i z2.i + x20 = vec_sub(x16, x18); // z4.r z6.r z4.i z6.i + + x21 = vec_perm(x12, x13, vcprm(2,s2,3,s3)); + x22 = vec_perm(x14, x15, vcprm(2,3,s2,s3)); + x23 = vec_perm(x14, x15, vcprm(3,2,s3,s2)); + x24 = vec_add(x22, x23); + x25 = vec_sub(x22, x23); + x26 = vec_mul( vec_perm(x24, x25, vcprm(2,s2,0,s0)), vc1); + + x27 = vec_add(x21, x26); // z1.r z7.r z1.i z3.i + x28 = vec_sub(x21, x26); //z5.r z3.r z5.i z7.i + + x29 = vec_perm(x19, x27, vcprm(0,2,s0,s2)); // z0.r z0.i z1.r z1.i + x30 = vec_perm(x19, x27, vcprm(1,3,s1,s3)); // z2.r z2.i z7.r z3.i + x31 = vec_perm(x20, x28, vcprm(0,2,s0,s2)); // z4.r z4.i z5.r z5.i + x32 = vec_perm(x20, x28, vcprm(1,3,s1,s3)); // z6.r z6.i z3.r z7.i + x33 = vec_perm(x30, x32, vcprm(0,1,s2,3)); // z2.r z2.i z3.r z3.i + x34 = vec_perm(x30, x32, vcprm(s0,s1,2,s3)); // z6.r z6.i z7.r z7.i + + vec_st(x29, 0, &(out[0])); + vec_st(x33, byte_2complex, &(out[0])); + vec_st(x31, byte_4complex, &(out[0])); + vec_st(x34, byte_6complex, &(out[0])); +} + +inline static void fft16_vsx_interleave(FFTComplex *z) +{ + float* out= (float*)z; + vec_f vc0 = {sqrthalf, sqrthalf, sqrthalf, sqrthalf}; + vec_f vc1 = {ff_cos_16[1], ff_cos_16[1], ff_cos_16[1], ff_cos_16[1]}; + vec_f vc2 = {ff_cos_16[3], ff_cos_16[3], ff_cos_16[3], ff_cos_16[3]}; + vec_f vz0, vz1, vz2, vz3; + vec_f vz4, vz5, vz6, vz7; + vec_f x0, x1, x2, x3; + vec_f x4, x5, x6, x7; + vec_f x8, x9, x10, x11; + vec_f x12, x13, x14, x15; + vec_f x16, x17, x18, x19; + vec_f x20, x21, x22, x23; + vec_f x24, x25, x26, x27; + vec_f x28, x29, x30, x31; + vec_f x32, x33, x34, x35; + vec_f x36, x37, x38, x39; + vec_f x40, x41, x42, x43; + vec_f x44, x45, x46, x47; + vec_f x48, x49, x50, x51; + vec_f x52, x53, x54, x55; + vec_f x56, x57, x58, x59; + vec_f x60, x61, x62, x63; + vec_f x64, x65, x66, x67; + vec_f x68, x69, x70, x71; + vec_f x72, x73, x74, x75; + vec_f x76, x77, x78, x79; + vec_f x80, x81, x82, x83; + vec_f x84, x85, x86; + + vz0 = vec_ld(0, &(out[0])); + vz1 = vec_ld(byte_2complex, &(out[0])); + vz2 = vec_ld(byte_4complex, &(out[0])); + vz3 = vec_ld(byte_6complex, &(out[0])); + vz4 = vec_ld(byte_8complex, &(out[0])); + vz5 = vec_ld(byte_10complex, &(out[0])); + vz6 = vec_ld(byte_12complex, &(out[0])); + vz7 = vec_ld(byte_14complex, &(out[0])); + + x0 = vec_perm(vz0, vz1, vcprm(0,1,s2,s1)); + x1 = vec_perm(vz0, vz1, vcprm(2,3,s0,s3)); + x2 = vec_perm(vz2, vz3, vcprm(0,1,s0,s1)); + x3 = vec_perm(vz2, vz3, vcprm(2,3,s2,s3)); + + x4 = vec_perm(vz4, vz5, vcprm(0,1,s2,s1)); + x5 = vec_perm(vz4, vz5, vcprm(2,3,s0,s3)); + x6 = vec_perm(vz6, vz7, vcprm(0,1,s2,s1)); + x7 = vec_perm(vz6, vz7, vcprm(2,3,s0,s3)); + + x8 = vec_add(x0, x1); + x9 = vec_sub(x0, x1); + x10 = vec_add(x2, x3); + x11 = vec_sub(x2, x3); + + x12 = vec_add(x4, x5); + x13 = vec_sub(x4, x5); + x14 = vec_add(x6, x7); + x15 = vec_sub(x6, x7); + + x16 = vec_perm(x8, x9, vcprm(0,1,s0,s1)); + x17 = vec_perm(x8, x9, vcprm(2,3,s3,s2)); + x18 = vec_perm(x10, x11, vcprm(2,1,s1,s2)); + x19 = vec_perm(x10, x11, vcprm(0,3,s0,s3)); + x20 = vec_perm(x12, x14, vcprm(0,1,s0, s1)); + x21 = vec_perm(x12, x14, vcprm(2,3,s2,s3)); + x22 = vec_perm(x13, x15, vcprm(0,1,s0,s1)); + x23 = vec_perm(x13, x15, vcprm(3,2,s3,s2)); + + x24 = vec_add(x16, x17); + x25 = vec_sub(x16, x17); + x26 = vec_add(x18, x19); + x27 = vec_sub(x18, x19); + x28 = vec_add(x20, x21); + x29 = vec_sub(x20, x21); + x30 = vec_add(x22, x23); + x31 = vec_sub(x22, x23); + + x32 = vec_add(x24, x26); + x33 = vec_sub(x24, x26); + x34 = vec_perm(x32, x33, vcprm(0,1,s0,s1)); + + x35 = vec_perm(x28, x29, vcprm(2,1,s1,s2)); + x36 = vec_perm(x28, x29, vcprm(0,3,s0,s3)); + x37 = vec_add(x35, x36); + x38 = vec_sub(x35, x36); + x39 = vec_perm(x37, x38, vcprm(0,1,s1,s0)); + + x40 = vec_perm(x27, x38, vcprm(3,2,s2,s3)); + x41 = vec_perm(x26, x37, vcprm(2,3,s3,s2)); + x42 = vec_add(x40, x41); + x43 = vec_sub(x40, x41); + x44 = vec_mul(x42, vc0); + x45 = vec_mul(x43, vc0); + + x46 = vec_add(x34, x39); // z0.r z0.i z4.r z4.i + x47 = vec_sub(x34, x39); // z8.r z8.i z12.r z12.i + + x48 = vec_perm(x30, x31, vcprm(2,1,s1,s2)); + x49 = vec_perm(x30, x31, vcprm(0,3,s3,s0)); + x50 = vec_add(x48, x49); + x51 = vec_sub(x48, x49); + x52 = vec_mul(x50, vc1); + x53 = vec_mul(x50, vc2); + x54 = vec_mul(x51, vc1); + x55 = vec_mul(x51, vc2); + + x56 = vec_perm(x24, x25, vcprm(2,3,s2,s3)); + x57 = vec_perm(x44, x45, vcprm(0,1,s1,s0)); + x58 = vec_add(x56, x57); + x59 = vec_sub(x56, x57); + + x60 = vec_perm(x54, x55, vcprm(1,0,3,2)); + x61 = vec_perm(x54, x55, vcprm(s1,s0,s3,s2)); + x62 = vec_add(x52, x61); + x63 = vec_sub(x52, x61); + x64 = vec_add(x60, x53); + x65 = vec_sub(x60, x53); + x66 = vec_perm(x62, x64, vcprm(0,1,s3,s2)); + x67 = vec_perm(x63, x65, vcprm(s0,s1,3,2)); + + x68 = vec_add(x58, x66); // z1.r z1.i z3.r z3.i + x69 = vec_sub(x58, x66); // z9.r z9.i z11.r z11.i + x70 = vec_add(x59, x67); // z5.r z5.i z15.r z15.i + x71 = vec_sub(x59, x67); // z13.r z13.i z7.r z7.i + + x72 = vec_perm(x25, x27, vcprm(s1,s0,s2,s3)); + x73 = vec_add(x25, x72); + x74 = vec_sub(x25, x72); + x75 = vec_perm(x73, x74, vcprm(0,1,s0,s1)); + x76 = vec_perm(x44, x45, vcprm(3,2,s2,s3)); + x77 = vec_add(x75, x76); // z2.r z2.i z6.r z6.i + x78 = vec_sub(x75, x76); // z10.r z10.i z14.r z14.i + + x79 = vec_perm(x46, x68, vcprm(0,1,s0,s1)); // z0.r z0.i z1.r z1.i + x80 = vec_perm(x77, x68, vcprm(0,1,s2,s3)); // z2.r z2.i z3.r z3.i + x81 = vec_perm(x46, x70, vcprm(2,3,s0,s1)); // z4.r z4.i z5.r z5.i + x82 = vec_perm(x71, x77, vcprm(s2,s3,2,3)); // z6.r z6.i z7.r z7.i + vec_st(x79, 0, &(out[0])); + vec_st(x80, byte_2complex, &(out[0])); + vec_st(x81, byte_4complex, &(out[0])); + vec_st(x82, byte_6complex, &(out[0])); + x83 = vec_perm(x47, x69, vcprm(0,1,s0,s1)); // z8.r z8.i z9.r z9.i + x84 = vec_perm(x78, x69, vcprm(0,1,s2,s3)); // z10.r z10.i z11.r z11.i + x85 = vec_perm(x47, x71, vcprm(2,3,s0,s1)); // z12.r z12.i z13.r z13.i + x86 = vec_perm(x70, x78, vcprm(s2,s3,2,3)); // z14.r z14.i z15.r z15.i + vec_st(x83, byte_8complex, &(out[0])); + vec_st(x84, byte_10complex, &(out[0])); + vec_st(x85, byte_12complex, &(out[0])); + vec_st(x86, byte_14complex, &(out[0])); +} + +inline static void fft4_vsx(FFTComplex *z) +{ + vec_f a, b, c, d; + float* out= (float*)z; + a = vec_ld(0, &(out[0])); + b = vec_ld(byte_2complex, &(out[0])); + + c = vec_perm(a, b, vcprm(0,1,s2,s1)); + d = vec_perm(a, b, vcprm(2,3,s0,s3)); + a = vec_add(c, d); + b = vec_sub(c, d); + + c = vec_perm(a,b, vcprm(0,s0,1,s1)); + d = vec_perm(a, b, vcprm(2,s3,3,s2)); + + a = vec_add(c, d); + b = vec_sub(c, d); + + c = vec_perm(a, b, vcprm(0,1,s0,s1)); + d = vec_perm(a, b, vcprm(2,3,s2,s3)); + + vec_st(c, 0, &(out[0])); + vec_st(d, byte_2complex, &(out[0])); + return; +} + +inline static void fft8_vsx(FFTComplex *z) +{ + vec_f vz0, vz1, vz2, vz3; + vec_f vz4, vz5, vz6, vz7, vz8; + + float* out= (float*)z; + vec_f vc0 = {0.0, 0.0, 0.0, 0.0}; + vec_f vc1 = {-sqrthalf, sqrthalf, sqrthalf, -sqrthalf}; + vec_f vc2 = {sqrthalf, sqrthalf, sqrthalf, sqrthalf}; + + vz0 = vec_ld(0, &(out[0])); + vz1 = vec_ld(byte_2complex, &(out[0])); + vz2 = vec_ld(byte_4complex, &(out[0])); + vz3 = vec_ld(byte_6complex, &(out[0])); + + vz6 = vec_perm(vz2, vz3, vcprm(0,s0,1,s1)); + vz7 = vec_perm(vz2, vz3, vcprm(2,s2,3,s3)); + vz4 = vec_perm(vz0, vz1, vcprm(0,1,s2,s1)); + vz5 = vec_perm(vz0, vz1, vcprm(2,3,s0,s3)); + + vz2 = vec_add(vz6, vz7); + vz3 = vec_sub(vz6, vz7); + vz8 = vec_perm(vz3, vz3, vcprm(2,3,0,1)); + + vz0 = vec_add(vz4, vz5); + vz1 = vec_sub(vz4, vz5); + + vz3 = vec_madd(vz3, vc1, vc0); + vz3 = vec_madd(vz8, vc2, vz3); + + vz4 = vec_perm(vz0, vz1, vcprm(0,s0,1,s1)); + vz5 = vec_perm(vz0, vz1, vcprm(2,s3,3,s2)); + vz6 = vec_perm(vz2, vz3, vcprm(1,2,s3,s0)); + vz7 = vec_perm(vz2, vz3, vcprm(0,3,s2,s1)); + + vz0 = vec_add(vz4, vz5); + vz1 = vec_sub(vz4, vz5); + vz2 = vec_add(vz6, vz7); + vz3 = vec_sub(vz6, vz7); + + vz4 = vec_perm(vz0, vz1, vcprm(0,1,s0,s1)); + vz5 = vec_perm(vz0, vz1, vcprm(2,3,s2,s3)); + vz6 = vec_perm(vz2, vz3, vcprm(0,2,s1,s3)); + vz7 = vec_perm(vz2, vz3, vcprm(1,3,s0,s2)); + + + vz2 = vec_sub(vz4, vz6); + vz3 = vec_sub(vz5, vz7); + + vz0 = vec_add(vz4, vz6); + vz1 = vec_add(vz5, vz7); + + vec_st(vz0, 0, &(out[0])); + vec_st(vz1, byte_2complex, &(out[0])); + vec_st(vz2, byte_4complex, &(out[0])); + vec_st(vz3, byte_6complex, &(out[0])); + return; +} + +inline static void fft16_vsx(FFTComplex *z) +{ + float* out= (float*)z; + vec_f vc0 = {0.0, 0.0, 0.0, 0.0}; + vec_f vc1 = {-sqrthalf, sqrthalf, sqrthalf, -sqrthalf}; + vec_f vc2 = {sqrthalf, sqrthalf, sqrthalf, sqrthalf}; + vec_f vc3 = {1.0, 0.92387953, sqrthalf, 0.38268343}; + vec_f vc4 = {0.0, 0.38268343, sqrthalf, 0.92387953}; + vec_f vc5 = {-0.0, -0.38268343, -sqrthalf, -0.92387953}; + + vec_f vz0, vz1, vz2, vz3; + vec_f vz4, vz5, vz6, vz7; + vec_f vz8, vz9, vz10, vz11; + vec_f vz12, vz13; + + vz0 = vec_ld(byte_8complex, &(out[0])); + vz1 = vec_ld(byte_10complex, &(out[0])); + vz2 = vec_ld(byte_12complex, &(out[0])); + vz3 = vec_ld(byte_14complex, &(out[0])); + + vz4 = vec_perm(vz0, vz1, vcprm(0,1,s2,s1)); + vz5 = vec_perm(vz0, vz1, vcprm(2,3,s0,s3)); + vz6 = vec_perm(vz2, vz3, vcprm(0,1,s2,s1)); + vz7 = vec_perm(vz2, vz3, vcprm(2,3,s0,s3)); + + vz0 = vec_add(vz4, vz5); + vz1= vec_sub(vz4, vz5); + vz2 = vec_add(vz6, vz7); + vz3 = vec_sub(vz6, vz7); + + vz4 = vec_perm(vz0, vz1, vcprm(0,s0,1,s1)); + vz5 = vec_perm(vz0, vz1, vcprm(2,s3,3,s2)); + vz6 = vec_perm(vz2, vz3, vcprm(0,s0,1,s1)); + vz7 = vec_perm(vz2, vz3, vcprm(2,s3,3,s2)); + + vz0 = vec_add(vz4, vz5); + vz1 = vec_sub(vz4, vz5); + vz2 = vec_add(vz6, vz7); + vz3 = vec_sub(vz6, vz7); + + vz4 = vec_perm(vz0, vz1, vcprm(0,1,s0,s1)); + vz5 = vec_perm(vz0, vz1, vcprm(2,3,s2,s3)); + + vz6 = vec_perm(vz2, vz3, vcprm(0,1,s0,s1)); + vz7 = vec_perm(vz2, vz3, vcprm(2,3,s2,s3)); + + vz0 = vec_ld(0, &(out[0])); + vz1 = vec_ld(byte_2complex, &(out[0])); + vz2 = vec_ld(byte_4complex, &(out[0])); + vz3 = vec_ld(byte_6complex, &(out[0])); + vz10 = vec_perm(vz2, vz3, vcprm(0,s0,1,s1)); + vz11 = vec_perm(vz2, vz3, vcprm(2,s2,3,s3)); + vz8 = vec_perm(vz0, vz1, vcprm(0,1,s2,s1)); + vz9 = vec_perm(vz0, vz1, vcprm(2,3,s0,s3)); + + vz2 = vec_add(vz10, vz11); + vz3 = vec_sub(vz10, vz11); + vz12 = vec_perm(vz3, vz3, vcprm(2,3,0,1)); + vz0 = vec_add(vz8, vz9); + vz1 = vec_sub(vz8, vz9); + + vz3 = vec_madd(vz3, vc1, vc0); + vz3 = vec_madd(vz12, vc2, vz3); + vz8 = vec_perm(vz0, vz1, vcprm(0,s0,1,s1)); + vz9 = vec_perm(vz0, vz1, vcprm(2,s3,3,s2)); + vz10 = vec_perm(vz2, vz3, vcprm(1,2,s3,s0)); + vz11 = vec_perm(vz2, vz3, vcprm(0,3,s2,s1)); + + vz0 = vec_add(vz8, vz9); + vz1 = vec_sub(vz8, vz9); + vz2 = vec_add(vz10, vz11); + vz3 = vec_sub(vz10, vz11); + + vz8 = vec_perm(vz0, vz1, vcprm(0,1,s0,s1)); + vz9 = vec_perm(vz0, vz1, vcprm(2,3,s2,s3)); + vz10 = vec_perm(vz2, vz3, vcprm(0,2,s1,s3)); + vz11 = vec_perm(vz2, vz3, vcprm(1,3,s0,s2)); + + vz2 = vec_sub(vz8, vz10); + vz3 = vec_sub(vz9, vz11); + vz0 = vec_add(vz8, vz10); + vz1 = vec_add(vz9, vz11); + + vz8 = vec_madd(vz4, vc3, vc0); + vz9 = vec_madd(vz5, vc3, vc0); + vz10 = vec_madd(vz6, vc3, vc0); + vz11 = vec_madd(vz7, vc3, vc0); + + vz8 = vec_madd(vz5, vc4, vz8); + vz9 = vec_madd(vz4, vc5, vz9); + vz10 = vec_madd(vz7, vc5, vz10); + vz11 = vec_madd(vz6, vc4, vz11); + + vz12 = vec_sub(vz10, vz8); + vz10 = vec_add(vz10, vz8); + + vz13 = vec_sub(vz9, vz11); + vz11 = vec_add(vz9, vz11); + + vz4 = vec_sub(vz0, vz10); + vz0 = vec_add(vz0, vz10); + + vz7= vec_sub(vz3, vz12); + vz3= vec_add(vz3, vz12); + + vz5 = vec_sub(vz1, vz11); + vz1 = vec_add(vz1, vz11); + + vz6 = vec_sub(vz2, vz13); + vz2 = vec_add(vz2, vz13); + + vec_st(vz0, 0, &(out[0])); + vec_st(vz1, byte_2complex, &(out[0])); + vec_st(vz2, byte_4complex, &(out[0])); + vec_st(vz3, byte_6complex, &(out[0])); + vec_st(vz4, byte_8complex, &(out[0])); + vec_st(vz5, byte_10complex, &(out[0])); + vec_st(vz6, byte_12complex, &(out[0])); + vec_st(vz7, byte_14complex, &(out[0])); + return; + +} +inline static void pass_vsx(FFTComplex * z, const FFTSample * wre, unsigned int n) +{ + int o1 = n<<1; + int o2 = n<<2; + int o3 = o1+o2; + int i1, i2, i3; + FFTSample* out = (FFTSample*)z; + const FFTSample *wim = wre+o1; + vec_f v0, v1, v2, v3; + vec_f v4, v5, v6, v7; + vec_f v8, v9, v10, v11; + vec_f v12, v13; + + n = n-2; + i1 = o1*sizeof(FFTComplex); + i2 = o2*sizeof(FFTComplex); + i3 = o3*sizeof(FFTComplex); + + v8 = vec_ld(0, &(wre[0])); + v10 = vec_ld(0, &(wim[0])); + v9 = vec_ld(0, &(wim[-4])); + v9 = vec_perm(v9, v10, vcprm(s0,3,2,1)); + + v4 = vec_ld(i2, &(out[0])); + v5 = vec_ld(i2+16, &(out[0])); + v6 = vec_ld(i3, &(out[0])); + v7 = vec_ld(i3+16, &(out[0])); + v10 = vec_mul(v4, v8); // r2*wre + v11 = vec_mul(v5, v8); // i2*wre + v12 = vec_mul(v6, v8); // r3*wre + v13 = vec_mul(v7, v8); // i3*wre + + v0 = vec_ld(0, &(out[0])); // r0 + v3 = vec_ld(i1+16, &(out[0])); // i1 + v10 = vec_madd(v5, v9, v10); // r2*wim + v11 = vec_nmsub(v4, v9, v11); // i2*wim + v12 = vec_nmsub(v7, v9, v12); // r3*wim + v13 = vec_madd(v6, v9, v13); // i3*wim + + v1 = vec_ld(16, &(out[0])); // i0 + v2 = vec_ld(i1, &(out[0])); // r1 + v8 = vec_sub(v12, v10); + v12 = vec_add(v12, v10); + v9 = vec_sub(v11, v13); + v13 = vec_add(v11, v13); + v4 = vec_sub(v0, v12); + v0 = vec_add(v0, v12); + v7 = vec_sub(v3, v8); + v3 = vec_add(v3, v8); + + vec_st(v0, 0, &(out[0])); // r0 + vec_st(v3, i1+16, &(out[0])); // i1 + vec_st(v4, i2, &(out[0])); // r2 + vec_st(v7, i3+16, &(out[0]));// i3 + + v5 = vec_sub(v1, v13); + v1 = vec_add(v1, v13); + v6 = vec_sub(v2, v9); + v2 = vec_add(v2, v9); + + vec_st(v1, 16, &(out[0])); // i0 + vec_st(v2, i1, &(out[0])); // r1 + vec_st(v5, i2+16, &(out[0])); // i2 + vec_st(v6, i3, &(out[0])); // r3 + + do { + out += 8; + wre += 4; + wim -= 4; + + v8 = vec_ld(0, &(wre[0])); + v10 = vec_ld(0, &(wim[0])); + v9 = vec_ld(0, &(wim[-4])); + v9 = vec_perm(v9, v10, vcprm(s0,3,2,1)); + + v4 = vec_ld(i2, &(out[0])); // r2 + v5 = vec_ld(i2+16, &(out[0])); // i2 + v6 = vec_ld(i3, &(out[0])); // r3 + v7 = vec_ld(i3+16, &(out[0]));// i3 + v10 = vec_mul(v4, v8); // r2*wre + v11 = vec_mul(v5, v8); // i2*wre + v12 = vec_mul(v6, v8); // r3*wre + v13 = vec_mul(v7, v8); // i3*wre + + v0 = vec_ld(0, &(out[0])); // r0 + v3 = vec_ld(i1+16, &(out[0])); // i1 + v10 = vec_madd(v5, v9, v10); // r2*wim + v11 = vec_nmsub(v4, v9, v11); // i2*wim + v12 = vec_nmsub(v7, v9, v12); // r3*wim + v13 = vec_madd(v6, v9, v13); // i3*wim + + v1 = vec_ld(16, &(out[0])); // i0 + v2 = vec_ld(i1, &(out[0])); // r1 + v8 = vec_sub(v12, v10); + v12 = vec_add(v12, v10); + v9 = vec_sub(v11, v13); + v13 = vec_add(v11, v13); + v4 = vec_sub(v0, v12); + v0 = vec_add(v0, v12); + v7 = vec_sub(v3, v8); + v3 = vec_add(v3, v8); + + vec_st(v0, 0, &(out[0])); // r0 + vec_st(v3, i1+16, &(out[0])); // i1 + vec_st(v4, i2, &(out[0])); // r2 + vec_st(v7, i3+16, &(out[0])); // i3 + + v5 = vec_sub(v1, v13); + v1 = vec_add(v1, v13); + v6 = vec_sub(v2, v9); + v2 = vec_add(v2, v9); + + vec_st(v1, 16, &(out[0])); // i0 + vec_st(v2, i1, &(out[0])); // r1 + vec_st(v5, i2+16, &(out[0])); // i2 + vec_st(v6, i3, &(out[0])); // r3 + } while (n-=2); +} + +#endif + +#endif /* AVCODEC_PPC_FFT_VSX_H */