diff options
Diffstat (limited to 'libavcodec/ppc')
35 files changed, 2000 insertions, 860 deletions
diff --git a/libavcodec/ppc/Makefile b/libavcodec/ppc/Makefile index c6ff0f1132..6f7eeb7286 100644 --- a/libavcodec/ppc/Makefile +++ b/libavcodec/ppc/Makefile @@ -4,9 +4,10 @@ OBJS += ppc/fmtconvert_altivec.o \ OBJS-$(CONFIG_AUDIODSP) += ppc/audiodsp.o OBJS-$(CONFIG_BLOCKDSP) += ppc/blockdsp.o OBJS-$(CONFIG_FFT) += ppc/fft_init.o \ - ppc/fft_altivec.o + ppc/fft_altivec.o \ + ppc/fft_vsx.o OBJS-$(CONFIG_H264CHROMA) += ppc/h264chroma_init.o -OBJS-$(CONFIG_H264DSP) += ppc/h264dsp.o +OBJS-$(CONFIG_H264DSP) += ppc/h264dsp.o ppc/hpeldsp_altivec.o OBJS-$(CONFIG_H264QPEL) += ppc/h264qpel.o OBJS-$(CONFIG_HPELDSP) += ppc/hpeldsp_altivec.o OBJS-$(CONFIG_HUFFYUVDSP) += ppc/huffyuvdsp_altivec.o @@ -22,7 +23,7 @@ OBJS-$(CONFIG_VIDEODSP) += ppc/videodsp_ppc.o OBJS-$(CONFIG_VP3DSP) += ppc/vp3dsp_altivec.o # decoders/encoders -OBJS-$(CONFIG_APE_DECODER) += ppc/apedsp_altivec.o +OBJS-$(CONFIG_LLAUDDSP) += ppc/lossless_audiodsp_altivec.o OBJS-$(CONFIG_SVQ1_ENCODER) += ppc/svq1enc_altivec.o OBJS-$(CONFIG_VC1_DECODER) += ppc/vc1dsp_altivec.o OBJS-$(CONFIG_VORBIS_DECODER) += ppc/vorbisdsp_altivec.o diff --git a/libavcodec/ppc/asm.S b/libavcodec/ppc/asm.S index 141dee9b78..a3edeed202 100644 --- a/libavcodec/ppc/asm.S +++ b/libavcodec/ppc/asm.S @@ -1,20 +1,20 @@ /* * Copyright (c) 2009 Loren Merritt * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/ppc/audiodsp.c b/libavcodec/ppc/audiodsp.c index 36506ce902..c88c3d9167 100644 --- a/libavcodec/ppc/audiodsp.c +++ b/libavcodec/ppc/audiodsp.c @@ -1,20 +1,20 @@ /* * Copyright (c) 2007 Luca Barbato <lu_zero@gentoo.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/ppc/blockdsp.c b/libavcodec/ppc/blockdsp.c index 679bc0454f..0059b3b448 100644 --- a/libavcodec/ppc/blockdsp.c +++ b/libavcodec/ppc/blockdsp.c @@ -3,20 +3,20 @@ * Copyright (c) 2002 Dieter Shirley * Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/ppc/dct-test.c b/libavcodec/ppc/dct-test.c index 37fd8bbd23..2328516ca4 100644 --- a/libavcodec/ppc/dct-test.c +++ b/libavcodec/ppc/dct-test.c @@ -1,18 +1,18 @@ /* - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/ppc/fdct.h b/libavcodec/ppc/fdct.h index 74710354ef..437f815258 100644 --- a/libavcodec/ppc/fdct.h +++ b/libavcodec/ppc/fdct.h @@ -1,18 +1,18 @@ /* - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/ppc/fdctdsp.c b/libavcodec/ppc/fdctdsp.c index 51417a5828..f2efc5ddd0 100644 --- a/libavcodec/ppc/fdctdsp.c +++ b/libavcodec/ppc/fdctdsp.c @@ -1,20 +1,20 @@ /* * Copyright (C) 2003 James Klicman <james@klicman.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -59,7 +59,7 @@ #define WA (SQRT_2 * (-C3 - C5)) #define WB (SQRT_2 * (C5 - C3)) -static vector float fdctconsts[3] = { +static const vector float fdctconsts[3] = { { W0, W1, W2, W3 }, { W4, W5, W6, W7 }, { W8, W9, WA, WB } diff --git a/libavcodec/ppc/fft_altivec.S b/libavcodec/ppc/fft_altivec.S index c92b30b897..aab669ea45 100644 --- a/libavcodec/ppc/fft_altivec.S +++ b/libavcodec/ppc/fft_altivec.S @@ -5,20 +5,20 @@ * This algorithm (though not any of the implementation details) is * based on libdjbfft by D. J. Bernstein. * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/ppc/fft_init.c b/libavcodec/ppc/fft_init.c index 8fcc033b53..675fa33a95 100644 --- a/libavcodec/ppc/fft_init.c +++ b/libavcodec/ppc/fft_init.c @@ -3,20 +3,20 @@ * AltiVec-enabled * Copyright (c) 2009 Loren Merritt * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -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 */ diff --git a/libavcodec/ppc/fmtconvert_altivec.c b/libavcodec/ppc/fmtconvert_altivec.c index 14aaf6bc9a..10a7169d1b 100644 --- a/libavcodec/ppc/fmtconvert_altivec.c +++ b/libavcodec/ppc/fmtconvert_altivec.c @@ -1,20 +1,20 @@ /* * Copyright (c) 2006 Luca Barbato <lu_zero@gentoo.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/ppc/h264chroma_init.c b/libavcodec/ppc/h264chroma_init.c index 6d656d46ba..876efeca09 100644 --- a/libavcodec/ppc/h264chroma_init.c +++ b/libavcodec/ppc/h264chroma_init.c @@ -1,20 +1,20 @@ /* * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/ppc/h264chroma_template.c b/libavcodec/ppc/h264chroma_template.c index 293fef5c90..cb1e0957e1 100644 --- a/libavcodec/ppc/h264chroma_template.c +++ b/libavcodec/ppc/h264chroma_template.c @@ -1,30 +1,32 @@ /* * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ #include "libavutil/mem.h" +#include "libavutil/ppc/types_altivec.h" +#include "libavutil/ppc/util_altivec.h" /* this code assume that stride % 16 == 0 */ #define CHROMA_MC8_ALTIVEC_CORE(BIAS1, BIAS2) \ - vsrc2ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc2uc);\ - vsrc3ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc3uc);\ + vsrc2ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc2uc);\ + vsrc3ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc3uc);\ \ psum = vec_mladd(vA, vsrc0ssH, BIAS1);\ psum = vec_mladd(vB, vsrc1ssH, psum);\ @@ -49,8 +51,8 @@ #define CHROMA_MC8_ALTIVEC_CORE_SIMPLE \ \ - vsrc0ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc0uc);\ - vsrc1ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc1uc);\ + vsrc0ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc0uc);\ + vsrc1ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc1uc);\ \ psum = vec_mladd(vA, vsrc0ssH, v32ss);\ psum = vec_mladd(vE, vsrc1ssH, psum);\ @@ -70,6 +72,43 @@ #define noop(a) a #define add28(a) vec_add(v28ss, a) +#if HAVE_BIGENDIAN +#define GET_VSRC1(vs0, off, b, perm0, s){ \ + vec_u8 vsrcCuc, vsrcDuc; \ + vsrcCuc = vec_ld(off, s); \ + if (loadSecond){ \ + vsrcDuc = vec_ld(off + b, s); \ + } else \ + vsrcDuc = vsrcCuc; \ + \ + vs0 = vec_perm(vsrcCuc, vsrcDuc, perm0); \ +} +#define GET_VSRC(vs0, vs1, off, b, perm0, perm1, s){ \ + vec_u8 vsrcCuc, vsrcDuc; \ + vsrcCuc = vec_ld(off, s); \ + if (loadSecond){ \ + vsrcDuc = vec_ld(off + b, s); \ + } else \ + vsrcDuc = vsrcCuc; \ + \ + vs0 = vec_perm(vsrcCuc, vsrcDuc, perm0); \ + if (reallyBadAlign){ \ + vs1 = vsrcDuc; \ + } else \ + vs1 = vec_perm(vsrcCuc, vsrcDuc, perm1); \ + } + +#else + +#define GET_VSRC1(vs0, off, b, perm0, s){ \ + vs0 = vec_vsx_ld(off, s); \ + } +#define GET_VSRC(vs0, vs1, off, b, perm0, perm1, s){ \ + vs0 = vec_vsx_ld(off, s); \ + vs1 = vec_vsx_ld(off + 1, s); \ + } +#endif /* HAVE_BIGENDIAN */ + #ifdef PREFIX_h264_chroma_mc8_altivec static void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) { @@ -80,23 +119,27 @@ static void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, (( x) * ( y))}; register int i; vec_u8 fperm; - const vec_s32 vABCD = vec_ld(0, ABCD); - const vec_s16 vA = vec_splat((vec_s16)vABCD, 1); - const vec_s16 vB = vec_splat((vec_s16)vABCD, 3); - const vec_s16 vC = vec_splat((vec_s16)vABCD, 5); - const vec_s16 vD = vec_splat((vec_s16)vABCD, 7); LOAD_ZERO; + const vec_s32 vABCD = vec_ld(0, ABCD); + const vec_s16 vA = VEC_SPLAT16(vABCD, 1); + const vec_s16 vB = VEC_SPLAT16(vABCD, 3); + const vec_s16 vC = VEC_SPLAT16(vABCD, 5); + const vec_s16 vD = VEC_SPLAT16(vABCD, 7); const vec_s16 v32ss = vec_sl(vec_splat_s16(1),vec_splat_u16(5)); const vec_u16 v6us = vec_splat_u16(6); - register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1; - register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0; - vec_u8 vsrcAuc, av_uninit(vsrcBuc), vsrcperm0, vsrcperm1; + vec_u8 vsrcperm0, vsrcperm1; vec_u8 vsrc0uc, vsrc1uc; vec_s16 vsrc0ssH, vsrc1ssH; - vec_u8 vsrcCuc, vsrc2uc, vsrc3uc; + vec_u8 vsrc2uc, vsrc3uc; vec_s16 vsrc2ssH, vsrc3ssH, psum; vec_u8 vdst, ppsum, vfdst, fsum; +#if HAVE_BIGENDIAN + register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1; + register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0; + vsrcperm0 = vec_lvsl(0, src); + vsrcperm1 = vec_lvsl(1, src); +#endif if (((unsigned long)dst) % 16 == 0) { fperm = (vec_u8){0x10, 0x11, 0x12, 0x13, @@ -110,89 +153,28 @@ static void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, 0x1C, 0x1D, 0x1E, 0x1F}; } - vsrcAuc = vec_ld(0, src); - - if (loadSecond) - vsrcBuc = vec_ld(16, src); - vsrcperm0 = vec_lvsl(0, src); - vsrcperm1 = vec_lvsl(1, src); - - vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0); - if (reallyBadAlign) - vsrc1uc = vsrcBuc; - else - vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1); + GET_VSRC(vsrc0uc, vsrc1uc, 0, 16, vsrcperm0, vsrcperm1, src); - vsrc0ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc0uc); - vsrc1ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc1uc); + vsrc0ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc0uc); + vsrc1ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc1uc); if (ABCD[3]) { - if (!loadSecond) {// -> !reallyBadAlign - for (i = 0 ; i < h ; i++) { - vsrcCuc = vec_ld(stride + 0, src); - vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0); - vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1); - - CHROMA_MC8_ALTIVEC_CORE(v32ss, noop) - } - } else { - vec_u8 vsrcDuc; - for (i = 0 ; i < h ; i++) { - vsrcCuc = vec_ld(stride + 0, src); - vsrcDuc = vec_ld(stride + 16, src); - vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0); - if (reallyBadAlign) - vsrc3uc = vsrcDuc; - else - vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1); - - CHROMA_MC8_ALTIVEC_CORE(v32ss, noop) - } + for (i = 0 ; i < h ; i++) { + GET_VSRC(vsrc2uc, vsrc3uc, stride, 16, vsrcperm0, vsrcperm1, src); + CHROMA_MC8_ALTIVEC_CORE(v32ss, noop); } } else { const vec_s16 vE = vec_add(vB, vC); if (ABCD[2]) { // x == 0 B == 0 - if (!loadSecond) {// -> !reallyBadAlign - for (i = 0 ; i < h ; i++) { - vsrcCuc = vec_ld(stride + 0, src); - vsrc1uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0); - CHROMA_MC8_ALTIVEC_CORE_SIMPLE - - vsrc0uc = vsrc1uc; - } - } else { - vec_u8 vsrcDuc; - for (i = 0 ; i < h ; i++) { - vsrcCuc = vec_ld(stride + 0, src); - vsrcDuc = vec_ld(stride + 15, src); - vsrc1uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0); - CHROMA_MC8_ALTIVEC_CORE_SIMPLE - - vsrc0uc = vsrc1uc; - } + for (i = 0 ; i < h ; i++) { + GET_VSRC1(vsrc1uc, stride, 15, vsrcperm0, src); + CHROMA_MC8_ALTIVEC_CORE_SIMPLE; + vsrc0uc = vsrc1uc; } } else { // y == 0 C == 0 - if (!loadSecond) {// -> !reallyBadAlign - for (i = 0 ; i < h ; i++) { - vsrcCuc = vec_ld(0, src); - vsrc0uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0); - vsrc1uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1); - - CHROMA_MC8_ALTIVEC_CORE_SIMPLE - } - } else { - vec_u8 vsrcDuc; - for (i = 0 ; i < h ; i++) { - vsrcCuc = vec_ld(0, src); - vsrcDuc = vec_ld(15, src); - vsrc0uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0); - if (reallyBadAlign) - vsrc1uc = vsrcDuc; - else - vsrc1uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1); - - CHROMA_MC8_ALTIVEC_CORE_SIMPLE - } + for (i = 0 ; i < h ; i++) { + GET_VSRC(vsrc0uc, vsrc1uc, 0, 15, vsrcperm0, vsrcperm1, src); + CHROMA_MC8_ALTIVEC_CORE_SIMPLE; } } } @@ -209,23 +191,27 @@ static void PREFIX_no_rnd_vc1_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, i (( x) * ( y))}; register int i; vec_u8 fperm; - const vec_s32 vABCD = vec_ld(0, ABCD); - const vec_s16 vA = vec_splat((vec_s16)vABCD, 1); - const vec_s16 vB = vec_splat((vec_s16)vABCD, 3); - const vec_s16 vC = vec_splat((vec_s16)vABCD, 5); - const vec_s16 vD = vec_splat((vec_s16)vABCD, 7); LOAD_ZERO; + const vec_s32 vABCD = vec_ld(0, ABCD); + const vec_s16 vA = VEC_SPLAT16(vABCD, 1); + const vec_s16 vB = VEC_SPLAT16(vABCD, 3); + const vec_s16 vC = VEC_SPLAT16(vABCD, 5); + const vec_s16 vD = VEC_SPLAT16(vABCD, 7); const vec_s16 v28ss = vec_sub(vec_sl(vec_splat_s16(1),vec_splat_u16(5)),vec_splat_s16(4)); const vec_u16 v6us = vec_splat_u16(6); - register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1; - register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0; - vec_u8 vsrcAuc, av_uninit(vsrcBuc), vsrcperm0, vsrcperm1; + vec_u8 vsrcperm0, vsrcperm1; vec_u8 vsrc0uc, vsrc1uc; vec_s16 vsrc0ssH, vsrc1ssH; - vec_u8 vsrcCuc, vsrc2uc, vsrc3uc; + vec_u8 vsrc2uc, vsrc3uc; vec_s16 vsrc2ssH, vsrc3ssH, psum; vec_u8 vdst, ppsum, vfdst, fsum; +#if HAVE_BIGENDIAN + register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1; + register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0; + vsrcperm0 = vec_lvsl(0, src); + vsrcperm1 = vec_lvsl(1, src); +#endif if (((unsigned long)dst) % 16 == 0) { fperm = (vec_u8){0x10, 0x11, 0x12, 0x13, @@ -239,47 +225,14 @@ static void PREFIX_no_rnd_vc1_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, i 0x1C, 0x1D, 0x1E, 0x1F}; } - vsrcAuc = vec_ld(0, src); - - if (loadSecond) - vsrcBuc = vec_ld(16, src); - vsrcperm0 = vec_lvsl(0, src); - vsrcperm1 = vec_lvsl(1, src); - - vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0); - if (reallyBadAlign) - vsrc1uc = vsrcBuc; - else - vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1); - - vsrc0ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc0uc); - vsrc1ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc1uc); - - if (!loadSecond) {// -> !reallyBadAlign - for (i = 0 ; i < h ; i++) { - - - vsrcCuc = vec_ld(stride + 0, src); + GET_VSRC(vsrc0uc, vsrc1uc, 0, 16, vsrcperm0, vsrcperm1, src); - vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0); - vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1); + vsrc0ssH = (vec_s16)VEC_MERGEH(zero_u8v, (vec_u8)vsrc0uc); + vsrc1ssH = (vec_s16)VEC_MERGEH(zero_u8v, (vec_u8)vsrc1uc); - CHROMA_MC8_ALTIVEC_CORE(vec_splat_s16(0), add28) - } - } else { - vec_u8 vsrcDuc; - for (i = 0 ; i < h ; i++) { - vsrcCuc = vec_ld(stride + 0, src); - vsrcDuc = vec_ld(stride + 16, src); - - vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0); - if (reallyBadAlign) - vsrc3uc = vsrcDuc; - else - vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1); - - CHROMA_MC8_ALTIVEC_CORE(vec_splat_s16(0), add28) - } + for (i = 0 ; i < h ; i++) { + GET_VSRC(vsrc2uc, vsrc3uc, stride, 16, vsrcperm0, vsrcperm1, src); + CHROMA_MC8_ALTIVEC_CORE(vec_splat_s16(0), add28); } } #endif diff --git a/libavcodec/ppc/h264dsp.c b/libavcodec/ppc/h264dsp.c index 31dc141d29..da118a49b6 100644 --- a/libavcodec/ppc/h264dsp.c +++ b/libavcodec/ppc/h264dsp.c @@ -1,20 +1,20 @@ /* * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -62,10 +62,17 @@ b2 = vec_mergeh( a1, a3 ); \ b3 = vec_mergel( a1, a3 ) +#if HAVE_BIGENDIAN +#define vdst_load(d) \ + vdst_orig = vec_ld(0, dst); \ + vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask); +#else +#define vdst_load(d) vdst = vec_vsx_ld(0, dst) +#endif + #define VEC_LOAD_U8_ADD_S16_STORE_U8(va) \ - vdst_orig = vec_ld(0, dst); \ - vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask); \ - vdst_ss = (vec_s16) vec_mergeh(zero_u8v, vdst); \ + vdst_load(); \ + vdst_ss = (vec_s16) VEC_MERGEH(zero_u8v, vdst); \ va = vec_add(va, vdst_ss); \ va_u8 = vec_packsu(va, zero_s16v); \ va_u32 = vec_splat((vec_u32)va_u8, 0); \ @@ -165,26 +172,43 @@ static void h264_idct_add_altivec(uint8_t *dst, int16_t *block, int stride) d7 = vec_sub(b0v, b7v); \ } +#if HAVE_BIGENDIAN +#define GET_2PERM(ldv, stv, d) \ + ldv = vec_lvsl(0, d); \ + stv = vec_lvsr(8, d); +#define dstv_load(d) \ + vec_u8 hv = vec_ld( 0, d ); \ + vec_u8 lv = vec_ld( 7, d); \ + vec_u8 dstv = vec_perm( hv, lv, (vec_u8)perm_ldv ); +#define dest_unligned_store(d) \ + vec_u8 edgehv; \ + vec_u8 bodyv = vec_perm( idstsum8, idstsum8, perm_stv ); \ + vec_u8 edgelv = vec_perm( sel, zero_u8v, perm_stv ); \ + lv = vec_sel( lv, bodyv, edgelv ); \ + vec_st( lv, 7, d ); \ + hv = vec_ld( 0, d ); \ + edgehv = vec_perm( zero_u8v, sel, perm_stv ); \ + hv = vec_sel( hv, bodyv, edgehv ); \ + vec_st( hv, 0, d ); +#else + +#define GET_2PERM(ldv, stv, d) {} +#define dstv_load(d) vec_u8 dstv = vec_vsx_ld(0, d) +#define dest_unligned_store(d)\ + vec_u8 dst8 = vec_perm((vec_u8)idstsum8, dstv, vcprm(2,3,s2,s3));\ + vec_vsx_st(dst8, 0, d) +#endif /* HAVE_BIGENDIAN */ + #define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \ /* unaligned load */ \ - vec_u8 hv = vec_ld( 0, dest ); \ - vec_u8 lv = vec_ld( 7, dest ); \ - vec_u8 dstv = vec_perm( hv, lv, (vec_u8)perm_ldv ); \ + dstv_load(dest); \ vec_s16 idct_sh6 = vec_sra(idctv, sixv); \ - vec_u16 dst16 = (vec_u16)vec_mergeh(zero_u8v, dstv); \ + vec_u16 dst16 = (vec_u16)VEC_MERGEH(zero_u8v, dstv); \ vec_s16 idstsum = vec_adds(idct_sh6, (vec_s16)dst16); \ vec_u8 idstsum8 = vec_packsu(zero_s16v, idstsum); \ - vec_u8 edgehv; \ /* unaligned store */ \ - vec_u8 bodyv = vec_perm( idstsum8, idstsum8, perm_stv );\ - vec_u8 edgelv = vec_perm( sel, zero_u8v, perm_stv ); \ - lv = vec_sel( lv, bodyv, edgelv ); \ - vec_st( lv, 7, dest ); \ - hv = vec_ld( 0, dest ); \ - edgehv = vec_perm( zero_u8v, sel, perm_stv ); \ - hv = vec_sel( hv, bodyv, edgehv ); \ - vec_st( hv, 0, dest ); \ - } + dest_unligned_store(dest);\ +} static void h264_idct8_add_altivec(uint8_t *dst, int16_t *dct, int stride) { @@ -192,8 +216,8 @@ static void h264_idct8_add_altivec(uint8_t *dst, int16_t *dct, int stride) vec_s16 d0, d1, d2, d3, d4, d5, d6, d7; vec_s16 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7; - vec_u8 perm_ldv = vec_lvsl(0, dst); - vec_u8 perm_stv = vec_lvsr(8, dst); + vec_u8 perm_ldv, perm_stv; + GET_2PERM(perm_ldv, perm_stv, dst); const vec_u16 onev = vec_splat_u16(1); const vec_u16 twov = vec_splat_u16(2); @@ -236,20 +260,25 @@ static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, int16_t *bl { vec_s16 dc16; vec_u8 dcplus, dcminus, v0, v1, v2, v3, aligner; + vec_s32 v_dc32; LOAD_ZERO; DECLARE_ALIGNED(16, int, dc); int i; dc = (block[0] + 32) >> 6; block[0] = 0; - dc16 = vec_splat((vec_s16) vec_lde(0, &dc), 1); + v_dc32 = vec_lde(0, &dc); + dc16 = VEC_SPLAT16((vec_s16)v_dc32, 1); if (size == 4) - dc16 = vec_sld(dc16, zero_s16v, 8); + dc16 = VEC_SLD16(dc16, zero_s16v, 8); dcplus = vec_packsu(dc16, zero_s16v); dcminus = vec_packsu(vec_sub(zero_s16v, dc16), zero_s16v); aligner = vec_lvsr(0, dst); +#if !HAVE_BIGENDIAN + aligner = vec_perm(aligner, zero_u8v, vcswapc()); +#endif dcplus = vec_perm(dcplus, dcplus, aligner); dcminus = vec_perm(dcminus, dcminus, aligner); @@ -633,6 +662,9 @@ void weight_h264_W_altivec(uint8_t *block, int stride, int height, temp[2] = offset; vtemp = (vec_s16)vec_ld(0, temp); +#if !HAVE_BIGENDIAN + vtemp =(vec_s16)vec_perm(vtemp, vtemp, vcswapi2s(0,1,2,3)); +#endif vlog2_denom = (vec_u16)vec_splat(vtemp, 1); vweight = vec_splat(vtemp, 3); voffset = vec_splat(vtemp, 5); @@ -641,8 +673,8 @@ void weight_h264_W_altivec(uint8_t *block, int stride, int height, for (y = 0; y < height; y++) { vblock = vec_ld(0, block); - v0 = (vec_s16)vec_mergeh(zero_u8v, vblock); - v1 = (vec_s16)vec_mergel(zero_u8v, vblock); + v0 = (vec_s16)VEC_MERGEH(zero_u8v, vblock); + v1 = (vec_s16)VEC_MERGEL(zero_u8v, vblock); if (w == 16 || aligned) { v0 = vec_mladd(v0, vweight, zero_s16v); @@ -679,6 +711,9 @@ void biweight_h264_W_altivec(uint8_t *dst, uint8_t *src, int stride, int height, temp[3] = offset; vtemp = (vec_s16)vec_ld(0, temp); +#if !HAVE_BIGENDIAN + vtemp =(vec_s16)vec_perm(vtemp, vtemp, vcswapi2s(0,1,2,3)); +#endif vlog2_denom = (vec_u16)vec_splat(vtemp, 1); vweights = vec_splat(vtemp, 3); vweightd = vec_splat(vtemp, 5); @@ -690,10 +725,10 @@ void biweight_h264_W_altivec(uint8_t *dst, uint8_t *src, int stride, int height, vdst = vec_ld(0, dst); vsrc = vec_ld(0, src); - v0 = (vec_s16)vec_mergeh(zero_u8v, vdst); - v1 = (vec_s16)vec_mergel(zero_u8v, vdst); - v2 = (vec_s16)vec_mergeh(zero_u8v, vsrc); - v3 = (vec_s16)vec_mergel(zero_u8v, vsrc); + v0 = (vec_s16)VEC_MERGEH(zero_u8v, vdst); + v1 = (vec_s16)VEC_MERGEL(zero_u8v, vdst); + v2 = (vec_s16)VEC_MERGEH(zero_u8v, vsrc); + v3 = (vec_s16)VEC_MERGEL(zero_u8v, vsrc); if (w == 8) { if (src_aligned) diff --git a/libavcodec/ppc/h264qpel.c b/libavcodec/ppc/h264qpel.c index f840277c88..575f504d32 100644 --- a/libavcodec/ppc/h264qpel.c +++ b/libavcodec/ppc/h264qpel.c @@ -1,20 +1,20 @@ /* * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -191,86 +191,79 @@ static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, cons OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\ }\ +#if HAVE_BIGENDIAN +#define put_unligned_store(s, dest) { \ + tmp1 = vec_ld(0, dest); \ + mask = vec_lvsl(0, dest); \ + tmp2 = vec_ld(15, dest); \ + edges = vec_perm(tmp2, tmp1, mask); \ + align = vec_lvsr(0, dest); \ + tmp2 = vec_perm(s, edges, align); \ + tmp1 = vec_perm(edges, s, align); \ + vec_st(tmp2, 15, dest); \ + vec_st(tmp1, 0 , dest); \ + } +#else +#define put_unligned_store(s, dest) vec_vsx_st(s, 0, dest); +#endif /* HAVE_BIGENDIAN */ + static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1, const uint8_t * src2, int dst_stride, int src_stride1, int h) { int i; - vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align; - + vec_u8 a, b, d, mask_; +#if HAVE_BIGENDIAN + vec_u8 tmp1, tmp2, mask, edges, align; mask_ = vec_lvsl(0, src2); +#endif for (i = 0; i < h; i++) { - - tmp1 = vec_ld(i * src_stride1, src1); - mask = vec_lvsl(i * src_stride1, src1); - tmp2 = vec_ld(i * src_stride1 + 15, src1); - - a = vec_perm(tmp1, tmp2, mask); - - tmp1 = vec_ld(i * 16, src2); - tmp2 = vec_ld(i * 16 + 15, src2); - - b = vec_perm(tmp1, tmp2, mask_); - - tmp1 = vec_ld(0, dst); - mask = vec_lvsl(0, dst); - tmp2 = vec_ld(15, dst); - + a = unaligned_load(i * src_stride1, src1); + b = load_with_perm_vec(i * 16, src2, mask_); d = vec_avg(a, b); - - edges = vec_perm(tmp2, tmp1, mask); - - align = vec_lvsr(0, dst); - - tmp2 = vec_perm(d, edges, align); - tmp1 = vec_perm(edges, d, align); - - vec_st(tmp2, 15, dst); - vec_st(tmp1, 0 , dst); - + put_unligned_store(d, dst); dst += dst_stride; } } +#if HAVE_BIGENDIAN +#define avg_unligned_store(s, dest){ \ + tmp1 = vec_ld(0, dest); \ + mask = vec_lvsl(0, dest); \ + tmp2 = vec_ld(15, dest); \ + a = vec_avg(vec_perm(tmp1, tmp2, mask), s); \ + edges = vec_perm(tmp2, tmp1, mask); \ + align = vec_lvsr(0, dest); \ + tmp2 = vec_perm(a, edges, align); \ + tmp1 = vec_perm(edges, a, align); \ + vec_st(tmp2, 15, dest); \ + vec_st(tmp1, 0 , dest); \ + } +#else +#define avg_unligned_store(s, dest){ \ + a = vec_avg(vec_vsx_ld(0, dst), s); \ + vec_vsx_st(a, 0, dst); \ + } +#endif /* HAVE_BIGENDIAN */ + static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1, const uint8_t * src2, int dst_stride, int src_stride1, int h) { int i; - vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align; + vec_u8 a, b, d, mask_; +#if HAVE_BIGENDIAN + vec_u8 tmp1, tmp2, mask, edges, align; mask_ = vec_lvsl(0, src2); +#endif for (i = 0; i < h; i++) { - - tmp1 = vec_ld(i * src_stride1, src1); - mask = vec_lvsl(i * src_stride1, src1); - tmp2 = vec_ld(i * src_stride1 + 15, src1); - - a = vec_perm(tmp1, tmp2, mask); - - tmp1 = vec_ld(i * 16, src2); - tmp2 = vec_ld(i * 16 + 15, src2); - - b = vec_perm(tmp1, tmp2, mask_); - - tmp1 = vec_ld(0, dst); - mask = vec_lvsl(0, dst); - tmp2 = vec_ld(15, dst); - - d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b)); - - edges = vec_perm(tmp2, tmp1, mask); - - align = vec_lvsr(0, dst); - - tmp2 = vec_perm(d, edges, align); - tmp1 = vec_perm(edges, d, align); - - vec_st(tmp2, 15, dst); - vec_st(tmp1, 0 , dst); - + a = unaligned_load(i * src_stride1, src1); + b = load_with_perm_vec(i * 16, src2, mask_); + d = vec_avg(a, b); + avg_unligned_store(d, dst); dst += dst_stride; } } diff --git a/libavcodec/ppc/h264qpel_template.c b/libavcodec/ppc/h264qpel_template.c index fe83146e63..9a345485ba 100644 --- a/libavcodec/ppc/h264qpel_template.c +++ b/libavcodec/ppc/h264qpel_template.c @@ -1,24 +1,31 @@ /* * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * 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" +#if HAVE_UNISTD_H +#include <unistd.h> +#endif + #include "libavutil/mem.h" +#include "libavutil/ppc/types_altivec.h" +#include "libavutil/ppc/util_altivec.h" #ifdef DEBUG #define ASSERT_ALIGNED(ptr) assert(((unsigned long)ptr&0x0000000F)); @@ -26,6 +33,76 @@ #define ASSERT_ALIGNED(ptr) ; #endif +#if HAVE_BIGENDIAN +#define load_alignment(s, ali, pm2, pm1, pp0, pp1, pp2, pp3){\ + vec_u8 srcR1 = vec_ld(-2, s);\ + vec_u8 srcR2 = vec_ld(14, s);\ + switch (ali) {\ + default: {\ + srcM2 = vec_perm(srcR1, srcR2, pm2);\ + srcM1 = vec_perm(srcR1, srcR2, pm1);\ + srcP0 = vec_perm(srcR1, srcR2, pp0);\ + srcP1 = vec_perm(srcR1, srcR2, pp1);\ + srcP2 = vec_perm(srcR1, srcR2, pp2);\ + srcP3 = vec_perm(srcR1, srcR2, pp3);\ + } break;\ + case 11: {\ + srcM2 = vec_perm(srcR1, srcR2, pm2);\ + srcM1 = vec_perm(srcR1, srcR2, pm1);\ + srcP0 = vec_perm(srcR1, srcR2, pp0);\ + srcP1 = vec_perm(srcR1, srcR2, pp1);\ + srcP2 = vec_perm(srcR1, srcR2, pp2);\ + srcP3 = srcR2;\ + } break;\ + case 12: {\ + vec_u8 srcR3 = vec_ld(30, s);\ + srcM2 = vec_perm(srcR1, srcR2, pm2);\ + srcM1 = vec_perm(srcR1, srcR2, pm1);\ + srcP0 = vec_perm(srcR1, srcR2, pp0);\ + srcP1 = vec_perm(srcR1, srcR2, pp1);\ + srcP2 = srcR2;\ + srcP3 = vec_perm(srcR2, srcR3, pp3);\ + } break;\ + case 13: {\ + vec_u8 srcR3 = vec_ld(30, s);\ + srcM2 = vec_perm(srcR1, srcR2, pm2);\ + srcM1 = vec_perm(srcR1, srcR2, pm1);\ + srcP0 = vec_perm(srcR1, srcR2, pp0);\ + srcP1 = srcR2;\ + srcP2 = vec_perm(srcR2, srcR3, pp2);\ + srcP3 = vec_perm(srcR2, srcR3, pp3);\ + } break;\ + case 14: {\ + vec_u8 srcR3 = vec_ld(30, s);\ + srcM2 = vec_perm(srcR1, srcR2, pm2);\ + srcM1 = vec_perm(srcR1, srcR2, pm1);\ + srcP0 = srcR2;\ + srcP1 = vec_perm(srcR2, srcR3, pp1);\ + srcP2 = vec_perm(srcR2, srcR3, pp2);\ + srcP3 = vec_perm(srcR2, srcR3, pp3);\ + } break;\ + case 15: {\ + vec_u8 srcR3 = vec_ld(30, s);\ + srcM2 = vec_perm(srcR1, srcR2, pm2);\ + srcM1 = srcR2;\ + srcP0 = vec_perm(srcR2, srcR3, pp0);\ + srcP1 = vec_perm(srcR2, srcR3, pp1);\ + srcP2 = vec_perm(srcR2, srcR3, pp2);\ + srcP3 = vec_perm(srcR2, srcR3, pp3);\ + } break;\ + }\ + } +#else +#define load_alignment(s, ali, pm2, pm1, pp0, pp1, pp2, pp3){\ + srcM2 = vec_vsx_ld(-2, s);\ + srcM1 = vec_vsx_ld(-1, s);\ + srcP0 = vec_vsx_ld(0, s);\ + srcP1 = vec_vsx_ld(1, s);\ + srcP2 = vec_vsx_ld(2, s);\ + srcP3 = vec_vsx_ld(3, s);\ + } +#endif /* HAVE_BIGENDIAN */ + /* this code assume stride % 16 == 0 */ #ifdef PREFIX_h264_qpel16_h_lowpass_altivec static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t *dst, @@ -35,12 +112,7 @@ static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t *dst, register int i; LOAD_ZERO; - const vec_u8 permM2 = vec_lvsl(-2, src); - const vec_u8 permM1 = vec_lvsl(-1, src); - const vec_u8 permP0 = vec_lvsl(+0, src); - const vec_u8 permP1 = vec_lvsl(+1, src); - const vec_u8 permP2 = vec_lvsl(+2, src); - const vec_u8 permP3 = vec_lvsl(+3, src); + vec_u8 permM2, permM1, permP0, permP1, permP2, permP3; const vec_s16 v5ss = vec_splat_s16(5); const vec_u16 v5us = vec_splat_u16(5); const vec_s16 v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); @@ -59,79 +131,32 @@ static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t *dst, vec_u8 sum, fsum; +#if HAVE_BIGENDIAN + permM2 = vec_lvsl(-2, src); + permM1 = vec_lvsl(-1, src); + permP0 = vec_lvsl(+0, src); + permP1 = vec_lvsl(+1, src); + permP2 = vec_lvsl(+2, src); + permP3 = vec_lvsl(+3, src); +#endif /* HAVE_BIGENDIAN */ + for (i = 0 ; i < 16 ; i ++) { - vec_u8 srcR1 = vec_ld(-2, src); - vec_u8 srcR2 = vec_ld(14, src); - - switch (align) { - default: { - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = vec_perm(srcR1, srcR2, permP2); - srcP3 = vec_perm(srcR1, srcR2, permP3); - } break; - case 11: { - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = vec_perm(srcR1, srcR2, permP2); - srcP3 = srcR2; - } break; - case 12: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = srcR2; - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 13: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = srcR2; - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 14: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = srcR2; - srcP1 = vec_perm(srcR2, srcR3, permP1); - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 15: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = srcR2; - srcP0 = vec_perm(srcR2, srcR3, permP0); - srcP1 = vec_perm(srcR2, srcR3, permP1); - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - } - - srcP0A = (vec_s16) vec_mergeh(zero_u8v, srcP0); - srcP0B = (vec_s16) vec_mergel(zero_u8v, srcP0); - srcP1A = (vec_s16) vec_mergeh(zero_u8v, srcP1); - srcP1B = (vec_s16) vec_mergel(zero_u8v, srcP1); - - srcP2A = (vec_s16) vec_mergeh(zero_u8v, srcP2); - srcP2B = (vec_s16) vec_mergel(zero_u8v, srcP2); - srcP3A = (vec_s16) vec_mergeh(zero_u8v, srcP3); - srcP3B = (vec_s16) vec_mergel(zero_u8v, srcP3); - - srcM1A = (vec_s16) vec_mergeh(zero_u8v, srcM1); - srcM1B = (vec_s16) vec_mergel(zero_u8v, srcM1); - srcM2A = (vec_s16) vec_mergeh(zero_u8v, srcM2); - srcM2B = (vec_s16) vec_mergel(zero_u8v, srcM2); + load_alignment(src, align, permM2, permM1, permP0, permP1, permP2, permP3); + + srcP0A = (vec_s16) VEC_MERGEH(zero_u8v, srcP0); + srcP0B = (vec_s16) VEC_MERGEL(zero_u8v, srcP0); + srcP1A = (vec_s16) VEC_MERGEH(zero_u8v, srcP1); + srcP1B = (vec_s16) VEC_MERGEL(zero_u8v, srcP1); + + srcP2A = (vec_s16) VEC_MERGEH(zero_u8v, srcP2); + srcP2B = (vec_s16) VEC_MERGEL(zero_u8v, srcP2); + srcP3A = (vec_s16) VEC_MERGEH(zero_u8v, srcP3); + srcP3B = (vec_s16) VEC_MERGEL(zero_u8v, srcP3); + + srcM1A = (vec_s16) VEC_MERGEH(zero_u8v, srcM1); + srcM1B = (vec_s16) VEC_MERGEL(zero_u8v, srcM1); + srcM2A = (vec_s16) VEC_MERGEH(zero_u8v, srcM2); + srcM2B = (vec_s16) VEC_MERGEL(zero_u8v, srcM2); sum1A = vec_adds(srcP0A, srcP1A); sum1B = vec_adds(srcP0B, srcP1B); @@ -178,7 +203,10 @@ static void PREFIX_h264_qpel16_v_lowpass_altivec(uint8_t *dst, register int i; LOAD_ZERO; - const vec_u8 perm = vec_lvsl(0, src); + vec_u8 perm; +#if HAVE_BIGENDIAN + perm = vec_lvsl(0, src); +#endif const vec_s16 v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); const vec_u16 v5us = vec_splat_u16(5); const vec_s16 v5ss = vec_splat_s16(5); @@ -186,52 +214,41 @@ static void PREFIX_h264_qpel16_v_lowpass_altivec(uint8_t *dst, const uint8_t *srcbis = src - (srcStride * 2); - const vec_u8 srcM2a = vec_ld(0, srcbis); - const vec_u8 srcM2b = vec_ld(16, srcbis); - const vec_u8 srcM2 = vec_perm(srcM2a, srcM2b, perm); - //srcbis += srcStride; - const vec_u8 srcM1a = vec_ld(0, srcbis += srcStride); - const vec_u8 srcM1b = vec_ld(16, srcbis); - const vec_u8 srcM1 = vec_perm(srcM1a, srcM1b, perm); - //srcbis += srcStride; - const vec_u8 srcP0a = vec_ld(0, srcbis += srcStride); - const vec_u8 srcP0b = vec_ld(16, srcbis); - const vec_u8 srcP0 = vec_perm(srcP0a, srcP0b, perm); - //srcbis += srcStride; - const vec_u8 srcP1a = vec_ld(0, srcbis += srcStride); - const vec_u8 srcP1b = vec_ld(16, srcbis); - const vec_u8 srcP1 = vec_perm(srcP1a, srcP1b, perm); - //srcbis += srcStride; - const vec_u8 srcP2a = vec_ld(0, srcbis += srcStride); - const vec_u8 srcP2b = vec_ld(16, srcbis); - const vec_u8 srcP2 = vec_perm(srcP2a, srcP2b, perm); - //srcbis += srcStride; - - vec_s16 srcM2ssA = (vec_s16) vec_mergeh(zero_u8v, srcM2); - vec_s16 srcM2ssB = (vec_s16) vec_mergel(zero_u8v, srcM2); - vec_s16 srcM1ssA = (vec_s16) vec_mergeh(zero_u8v, srcM1); - vec_s16 srcM1ssB = (vec_s16) vec_mergel(zero_u8v, srcM1); - vec_s16 srcP0ssA = (vec_s16) vec_mergeh(zero_u8v, srcP0); - vec_s16 srcP0ssB = (vec_s16) vec_mergel(zero_u8v, srcP0); - vec_s16 srcP1ssA = (vec_s16) vec_mergeh(zero_u8v, srcP1); - vec_s16 srcP1ssB = (vec_s16) vec_mergel(zero_u8v, srcP1); - vec_s16 srcP2ssA = (vec_s16) vec_mergeh(zero_u8v, srcP2); - vec_s16 srcP2ssB = (vec_s16) vec_mergel(zero_u8v, srcP2); + const vec_u8 srcM2 = load_with_perm_vec(0, srcbis, perm); + srcbis += srcStride; + const vec_u8 srcM1 = load_with_perm_vec(0, srcbis, perm); + srcbis += srcStride; + const vec_u8 srcP0 = load_with_perm_vec(0, srcbis, perm); + srcbis += srcStride; + const vec_u8 srcP1 = load_with_perm_vec(0, srcbis, perm); + srcbis += srcStride; + const vec_u8 srcP2 = load_with_perm_vec(0, srcbis, perm); + srcbis += srcStride; + + vec_s16 srcM2ssA = (vec_s16) VEC_MERGEH(zero_u8v, srcM2); + vec_s16 srcM2ssB = (vec_s16) VEC_MERGEL(zero_u8v, srcM2); + vec_s16 srcM1ssA = (vec_s16) VEC_MERGEH(zero_u8v, srcM1); + vec_s16 srcM1ssB = (vec_s16) VEC_MERGEL(zero_u8v, srcM1); + vec_s16 srcP0ssA = (vec_s16) VEC_MERGEH(zero_u8v, srcP0); + vec_s16 srcP0ssB = (vec_s16) VEC_MERGEL(zero_u8v, srcP0); + vec_s16 srcP1ssA = (vec_s16) VEC_MERGEH(zero_u8v, srcP1); + vec_s16 srcP1ssB = (vec_s16) VEC_MERGEL(zero_u8v, srcP1); + vec_s16 srcP2ssA = (vec_s16) VEC_MERGEH(zero_u8v, srcP2); + vec_s16 srcP2ssB = (vec_s16) VEC_MERGEL(zero_u8v, srcP2); vec_s16 pp1A, pp1B, pp2A, pp2B, pp3A, pp3B, psumA, psumB, sumA, sumB, srcP3ssA, srcP3ssB, sum1A, sum1B, sum2A, sum2B, sum3A, sum3B; - vec_u8 sum, fsum, srcP3a, srcP3b, srcP3; + vec_u8 sum, fsum, srcP3; for (i = 0 ; i < 16 ; i++) { - srcP3a = vec_ld(0, srcbis += srcStride); - srcP3b = vec_ld(16, srcbis); - srcP3 = vec_perm(srcP3a, srcP3b, perm); - srcP3ssA = (vec_s16) vec_mergeh(zero_u8v, srcP3); - srcP3ssB = (vec_s16) vec_mergel(zero_u8v, srcP3); - //srcbis += srcStride; + srcP3 = load_with_perm_vec(0, srcbis, perm); + srcbis += srcStride; + + srcP3ssA = (vec_s16) VEC_MERGEH(zero_u8v, srcP3); + srcP3ssB = (vec_s16) VEC_MERGEL(zero_u8v, srcP3); sum1A = vec_adds(srcP0ssA, srcP1ssA); sum1B = vec_adds(srcP0ssB, srcP1ssB); @@ -288,12 +305,7 @@ static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t *dst, int16_t *tmp, { register int i; LOAD_ZERO; - const vec_u8 permM2 = vec_lvsl(-2, src); - const vec_u8 permM1 = vec_lvsl(-1, src); - const vec_u8 permP0 = vec_lvsl(+0, src); - const vec_u8 permP1 = vec_lvsl(+1, src); - const vec_u8 permP2 = vec_lvsl(+2, src); - const vec_u8 permP3 = vec_lvsl(+3, src); + vec_u8 permM2, permM1, permP0, permP1, permP2, permP3; const vec_s16 v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); const vec_u32 v10ui = vec_splat_u32(10); const vec_s16 v5ss = vec_splat_s16(5); @@ -325,81 +337,35 @@ static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t *dst, int16_t *tmp, vec_u8 fsum, sumv, sum; vec_s16 ssume, ssumo; +#if HAVE_BIGENDIAN + permM2 = vec_lvsl(-2, src); + permM1 = vec_lvsl(-1, src); + permP0 = vec_lvsl(+0, src); + permP1 = vec_lvsl(+1, src); + permP2 = vec_lvsl(+2, src); + permP3 = vec_lvsl(+3, src); +#endif /* HAVE_BIGENDIAN */ + src -= (2 * srcStride); for (i = 0 ; i < 21 ; i ++) { vec_u8 srcM2, srcM1, srcP0, srcP1, srcP2, srcP3; - vec_u8 srcR1 = vec_ld(-2, src); - vec_u8 srcR2 = vec_ld(14, src); - - switch (align) { - default: { - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = vec_perm(srcR1, srcR2, permP2); - srcP3 = vec_perm(srcR1, srcR2, permP3); - } break; - case 11: { - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = vec_perm(srcR1, srcR2, permP2); - srcP3 = srcR2; - } break; - case 12: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = srcR2; - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 13: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = srcR2; - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 14: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = srcR2; - srcP1 = vec_perm(srcR2, srcR3, permP1); - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 15: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = srcR2; - srcP0 = vec_perm(srcR2, srcR3, permP0); - srcP1 = vec_perm(srcR2, srcR3, permP1); - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - } - - srcP0A = (vec_s16) vec_mergeh(zero_u8v, srcP0); - srcP0B = (vec_s16) vec_mergel(zero_u8v, srcP0); - srcP1A = (vec_s16) vec_mergeh(zero_u8v, srcP1); - srcP1B = (vec_s16) vec_mergel(zero_u8v, srcP1); - - srcP2A = (vec_s16) vec_mergeh(zero_u8v, srcP2); - srcP2B = (vec_s16) vec_mergel(zero_u8v, srcP2); - srcP3A = (vec_s16) vec_mergeh(zero_u8v, srcP3); - srcP3B = (vec_s16) vec_mergel(zero_u8v, srcP3); - - srcM1A = (vec_s16) vec_mergeh(zero_u8v, srcM1); - srcM1B = (vec_s16) vec_mergel(zero_u8v, srcM1); - srcM2A = (vec_s16) vec_mergeh(zero_u8v, srcM2); - srcM2B = (vec_s16) vec_mergel(zero_u8v, srcM2); + + load_alignment(src, align, permM2, permM1, permP0, permP1, permP2, permP3); + + srcP0A = (vec_s16) VEC_MERGEH(zero_u8v, srcP0); + srcP0B = (vec_s16) VEC_MERGEL(zero_u8v, srcP0); + srcP1A = (vec_s16) VEC_MERGEH(zero_u8v, srcP1); + srcP1B = (vec_s16) VEC_MERGEL(zero_u8v, srcP1); + + srcP2A = (vec_s16) VEC_MERGEH(zero_u8v, srcP2); + srcP2B = (vec_s16) VEC_MERGEL(zero_u8v, srcP2); + srcP3A = (vec_s16) VEC_MERGEH(zero_u8v, srcP3); + srcP3B = (vec_s16) VEC_MERGEL(zero_u8v, srcP3); + + srcM1A = (vec_s16) VEC_MERGEH(zero_u8v, srcM1); + srcM1B = (vec_s16) VEC_MERGEL(zero_u8v, srcM1); + srcM2A = (vec_s16) VEC_MERGEH(zero_u8v, srcM2); + srcM2B = (vec_s16) VEC_MERGEL(zero_u8v, srcM2); sum1A = vec_adds(srcP0A, srcP1A); sum1B = vec_adds(srcP0B, srcP1B); @@ -448,8 +414,8 @@ static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t *dst, int16_t *tmp, const vec_s16 sum1B = vec_adds(tmpP0ssB, tmpP1ssB); const vec_s16 sum2A = vec_adds(tmpM1ssA, tmpP2ssA); const vec_s16 sum2B = vec_adds(tmpM1ssB, tmpP2ssB); - const vec_s16 sum3A = vec_adds(tmpM2ssA, tmpP3ssA); - const vec_s16 sum3B = vec_adds(tmpM2ssB, tmpP3ssB); + vec_s16 sum3A = vec_adds(tmpM2ssA, tmpP3ssA); + vec_s16 sum3B = vec_adds(tmpM2ssB, tmpP3ssB); tmpbis += tmpStride; @@ -474,10 +440,14 @@ static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t *dst, int16_t *tmp, pp2Be = vec_mule(sum2B, v5ss); pp2Bo = vec_mulo(sum2B, v5ss); - pp3Ae = vec_sra((vec_s32)sum3A, v16ui); pp3Ao = vec_mulo(sum3A, v1ss); - pp3Be = vec_sra((vec_s32)sum3B, v16ui); pp3Bo = vec_mulo(sum3B, v1ss); +#if !HAVE_BIGENDIAN + sum3A = (vec_s16)vec_perm(sum3A, sum3A,vcswapi2s(0,1,2,3)); + sum3B = (vec_s16)vec_perm(sum3B, sum3B,vcswapi2s(0,1,2,3)); +#endif + pp3Ae = vec_sra((vec_s32)sum3A, v16ui); + pp3Be = vec_sra((vec_s32)sum3B, v16ui); pp1cAe = vec_add(pp1Ae, v512si); pp1cAo = vec_add(pp1Ao, v512si); diff --git a/libavcodec/ppc/hpeldsp_altivec.c b/libavcodec/ppc/hpeldsp_altivec.c index fd6ae73915..87a1f05b6a 100644 --- a/libavcodec/ppc/hpeldsp_altivec.c +++ b/libavcodec/ppc/hpeldsp_altivec.c @@ -3,20 +3,20 @@ * Copyright (c) 2002 Dieter Shirley * Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -38,12 +38,11 @@ /* next one assumes that ((line_size % 16) == 0) */ void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) { - register vector unsigned char pixelsv1, pixelsv2; - register vector unsigned char pixelsv1B, pixelsv2B; - register vector unsigned char pixelsv1C, pixelsv2C; - register vector unsigned char pixelsv1D, pixelsv2D; + register vector unsigned char pixelsv1; + register vector unsigned char pixelsv1B; + register vector unsigned char pixelsv1C; + register vector unsigned char pixelsv1D; - register vector unsigned char perm = vec_lvsl(0, pixels); int i; register ptrdiff_t line_size_2 = line_size << 1; register ptrdiff_t line_size_3 = line_size + line_size_2; @@ -55,22 +54,14 @@ void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t li // -funroll-loops w/ this is bad - 74 cycles again. // all this is on a 7450, tuning for the 7450 for (i = 0; i < h; i += 4) { - pixelsv1 = vec_ld( 0, pixels); - pixelsv2 = vec_ld(15, pixels); - pixelsv1B = vec_ld(line_size, pixels); - pixelsv2B = vec_ld(15 + line_size, pixels); - pixelsv1C = vec_ld(line_size_2, pixels); - pixelsv2C = vec_ld(15 + line_size_2, pixels); - pixelsv1D = vec_ld(line_size_3, pixels); - pixelsv2D = vec_ld(15 + line_size_3, pixels); - vec_st(vec_perm(pixelsv1, pixelsv2, perm), - 0, (unsigned char*)block); - vec_st(vec_perm(pixelsv1B, pixelsv2B, perm), - line_size, (unsigned char*)block); - vec_st(vec_perm(pixelsv1C, pixelsv2C, perm), - line_size_2, (unsigned char*)block); - vec_st(vec_perm(pixelsv1D, pixelsv2D, perm), - line_size_3, (unsigned char*)block); + pixelsv1 = unaligned_load( 0, pixels); + pixelsv1B = unaligned_load(line_size, pixels); + pixelsv1C = unaligned_load(line_size_2, pixels); + pixelsv1D = unaligned_load(line_size_3, pixels); + VEC_ST(pixelsv1, 0, (unsigned char*)block); + VEC_ST(pixelsv1B, line_size, (unsigned char*)block); + VEC_ST(pixelsv1C, line_size_2, (unsigned char*)block); + VEC_ST(pixelsv1D, line_size_3, (unsigned char*)block); pixels+=line_size_4; block +=line_size_4; } @@ -80,15 +71,12 @@ void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t li #define op_avg(a,b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEUL)>>1) ) void ff_avg_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) { - register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv; - register vector unsigned char perm = vec_lvsl(0, pixels); - int i; + register vector unsigned char pixelsv, blockv; + int i; for (i = 0; i < h; i++) { - pixelsv1 = vec_ld( 0, pixels); - pixelsv2 = vec_ld(16,pixels); blockv = vec_ld(0, block); - pixelsv = vec_perm(pixelsv1, pixelsv2, perm); + pixelsv = VEC_LD( 0, pixels); blockv = vec_avg(blockv,pixelsv); vec_st(blockv, 0, (unsigned char*)block); pixels+=line_size; @@ -108,9 +96,7 @@ static void avg_pixels8_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff int rightside = ((unsigned long)block & 0x0000000F); blockv = vec_ld(0, block); - pixelsv1 = vec_ld( 0, pixels); - pixelsv2 = vec_ld(16, pixels); - pixelsv = vec_perm(pixelsv1, pixelsv2, vec_lvsl(0, pixels)); + pixelsv = VEC_LD( 0, pixels); if (rightside) { pixelsv = vec_perm(blockv, pixelsv, vcprm(0,1,s0,s1)); @@ -132,21 +118,16 @@ static void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi { register int i; register vector unsigned char pixelsv1, pixelsv2, pixelsavg; - register vector unsigned char blockv, temp1, temp2; + register vector unsigned char blockv; register vector unsigned short pixelssum1, pixelssum2, temp3; register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0); register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); - temp1 = vec_ld(0, pixels); - temp2 = vec_ld(16, pixels); - pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels)); - if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) { - pixelsv2 = temp2; - } else { - pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels)); - } - pixelsv1 = vec_mergeh(vczero, pixelsv1); - pixelsv2 = vec_mergeh(vczero, pixelsv2); + pixelsv1 = VEC_LD(0, pixels); + pixelsv2 = VEC_LD(1, pixels); + pixelsv1 = VEC_MERGEH(vczero, pixelsv1); + pixelsv2 = VEC_MERGEH(vczero, pixelsv2); + pixelssum1 = vec_add((vector unsigned short)pixelsv1, (vector unsigned short)pixelsv2); pixelssum1 = vec_add(pixelssum1, vctwo); @@ -155,17 +136,10 @@ static void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi int rightside = ((unsigned long)block & 0x0000000F); blockv = vec_ld(0, block); - temp1 = vec_ld(line_size, pixels); - temp2 = vec_ld(line_size + 16, pixels); - pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels)); - if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) { - pixelsv2 = temp2; - } else { - pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels)); - } - - pixelsv1 = vec_mergeh(vczero, pixelsv1); - pixelsv2 = vec_mergeh(vczero, pixelsv2); + pixelsv1 = unaligned_load(line_size, pixels); + pixelsv2 = unaligned_load(line_size+1, pixels); + pixelsv1 = VEC_MERGEH(vczero, pixelsv1); + pixelsv2 = VEC_MERGEH(vczero, pixelsv2); pixelssum2 = vec_add((vector unsigned short)pixelsv1, (vector unsigned short)pixelsv2); temp3 = vec_add(pixelssum1, pixelssum2); @@ -191,22 +165,16 @@ static void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels { register int i; register vector unsigned char pixelsv1, pixelsv2, pixelsavg; - register vector unsigned char blockv, temp1, temp2; + register vector unsigned char blockv; register vector unsigned short pixelssum1, pixelssum2, temp3; register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0); register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1); register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); - temp1 = vec_ld(0, pixels); - temp2 = vec_ld(16, pixels); - pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels)); - if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) { - pixelsv2 = temp2; - } else { - pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels)); - } - pixelsv1 = vec_mergeh(vczero, pixelsv1); - pixelsv2 = vec_mergeh(vczero, pixelsv2); + pixelsv1 = VEC_LD(0, pixels); + pixelsv2 = VEC_LD(1, pixels); + pixelsv1 = VEC_MERGEH(vczero, pixelsv1); + pixelsv2 = VEC_MERGEH(vczero, pixelsv2); pixelssum1 = vec_add((vector unsigned short)pixelsv1, (vector unsigned short)pixelsv2); pixelssum1 = vec_add(pixelssum1, vcone); @@ -215,17 +183,10 @@ static void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels int rightside = ((unsigned long)block & 0x0000000F); blockv = vec_ld(0, block); - temp1 = vec_ld(line_size, pixels); - temp2 = vec_ld(line_size + 16, pixels); - pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels)); - if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) { - pixelsv2 = temp2; - } else { - pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels)); - } - - pixelsv1 = vec_mergeh(vczero, pixelsv1); - pixelsv2 = vec_mergeh(vczero, pixelsv2); + pixelsv1 = unaligned_load(line_size, pixels); + pixelsv2 = unaligned_load(line_size+1, pixels); + pixelsv1 = VEC_MERGEH(vczero, pixelsv1); + pixelsv2 = VEC_MERGEH(vczero, pixelsv2); pixelssum2 = vec_add((vector unsigned short)pixelsv1, (vector unsigned short)pixelsv2); temp3 = vec_add(pixelssum1, pixelssum2); @@ -251,24 +212,18 @@ static void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, pt { register int i; register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4; - register vector unsigned char blockv, temp1, temp2; + register vector unsigned char blockv; register vector unsigned short temp3, temp4, pixelssum1, pixelssum2, pixelssum3, pixelssum4; register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0); register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); - temp1 = vec_ld(0, pixels); - temp2 = vec_ld(16, pixels); - pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels)); - if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) { - pixelsv2 = temp2; - } else { - pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels)); - } - pixelsv3 = vec_mergel(vczero, pixelsv1); - pixelsv4 = vec_mergel(vczero, pixelsv2); - pixelsv1 = vec_mergeh(vczero, pixelsv1); - pixelsv2 = vec_mergeh(vczero, pixelsv2); + pixelsv1 = VEC_LD(0, pixels); + pixelsv2 = VEC_LD(1, pixels); + pixelsv3 = VEC_MERGEL(vczero, pixelsv1); + pixelsv4 = VEC_MERGEL(vczero, pixelsv2); + pixelsv1 = VEC_MERGEH(vczero, pixelsv1); + pixelsv2 = VEC_MERGEH(vczero, pixelsv2); pixelssum3 = vec_add((vector unsigned short)pixelsv3, (vector unsigned short)pixelsv4); pixelssum3 = vec_add(pixelssum3, vctwo); @@ -279,20 +234,13 @@ static void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, pt for (i = 0; i < h ; i++) { blockv = vec_ld(0, block); - temp1 = vec_ld(line_size, pixels); - temp2 = vec_ld(line_size + 16, pixels); - pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels)); - if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) { - pixelsv2 = temp2; - } else { - pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels)); - } - - pixelsv3 = vec_mergel(vczero, pixelsv1); - pixelsv4 = vec_mergel(vczero, pixelsv2); - pixelsv1 = vec_mergeh(vczero, pixelsv1); - pixelsv2 = vec_mergeh(vczero, pixelsv2); + pixelsv1 = unaligned_load(line_size, pixels); + pixelsv2 = unaligned_load(line_size+1, pixels); + pixelsv3 = VEC_MERGEL(vczero, pixelsv1); + pixelsv4 = VEC_MERGEL(vczero, pixelsv2); + pixelsv1 = VEC_MERGEH(vczero, pixelsv1); + pixelsv2 = VEC_MERGEH(vczero, pixelsv2); pixelssum4 = vec_add((vector unsigned short)pixelsv3, (vector unsigned short)pixelsv4); pixelssum2 = vec_add((vector unsigned short)pixelsv1, @@ -319,25 +267,19 @@ static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pix { register int i; register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4; - register vector unsigned char blockv, temp1, temp2; + register vector unsigned char blockv; register vector unsigned short temp3, temp4, pixelssum1, pixelssum2, pixelssum3, pixelssum4; register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0); register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1); register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); - temp1 = vec_ld(0, pixels); - temp2 = vec_ld(16, pixels); - pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels)); - if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) { - pixelsv2 = temp2; - } else { - pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels)); - } - pixelsv3 = vec_mergel(vczero, pixelsv1); - pixelsv4 = vec_mergel(vczero, pixelsv2); - pixelsv1 = vec_mergeh(vczero, pixelsv1); - pixelsv2 = vec_mergeh(vczero, pixelsv2); + pixelsv1 = VEC_LD(0, pixels); + pixelsv2 = VEC_LD(1, pixels); + pixelsv3 = VEC_MERGEL(vczero, pixelsv1); + pixelsv4 = VEC_MERGEL(vczero, pixelsv2); + pixelsv1 = VEC_MERGEH(vczero, pixelsv1); + pixelsv2 = VEC_MERGEH(vczero, pixelsv2); pixelssum3 = vec_add((vector unsigned short)pixelsv3, (vector unsigned short)pixelsv4); pixelssum3 = vec_add(pixelssum3, vcone); @@ -346,22 +288,13 @@ static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pix pixelssum1 = vec_add(pixelssum1, vcone); for (i = 0; i < h ; i++) { - blockv = vec_ld(0, block); - - temp1 = vec_ld(line_size, pixels); - temp2 = vec_ld(line_size + 16, pixels); - pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels)); - if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) { - pixelsv2 = temp2; - } else { - pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels)); - } - - pixelsv3 = vec_mergel(vczero, pixelsv1); - pixelsv4 = vec_mergel(vczero, pixelsv2); - pixelsv1 = vec_mergeh(vczero, pixelsv1); - pixelsv2 = vec_mergeh(vczero, pixelsv2); + pixelsv1 = unaligned_load(line_size, pixels); + pixelsv2 = unaligned_load(line_size+1, pixels); + pixelsv3 = VEC_MERGEL(vczero, pixelsv1); + pixelsv4 = VEC_MERGEL(vczero, pixelsv2); + pixelsv1 = VEC_MERGEH(vczero, pixelsv1); + pixelsv2 = VEC_MERGEH(vczero, pixelsv2); pixelssum4 = vec_add((vector unsigned short)pixelsv3, (vector unsigned short)pixelsv4); pixelssum2 = vec_add((vector unsigned short)pixelsv1, @@ -376,7 +309,7 @@ static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pix blockv = vec_packsu(temp3, temp4); - vec_st(blockv, 0, block); + VEC_ST(blockv, 0, block); block += line_size; pixels += line_size; @@ -388,7 +321,7 @@ static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi { register int i; register vector unsigned char pixelsv1, pixelsv2, pixelsavg; - register vector unsigned char blockv, temp1, temp2, blocktemp; + register vector unsigned char blockv, blocktemp; register vector unsigned short pixelssum1, pixelssum2, temp3; register const vector unsigned char vczero = (const vector unsigned char) @@ -396,16 +329,10 @@ static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi register const vector unsigned short vctwo = (const vector unsigned short) vec_splat_u16(2); - temp1 = vec_ld(0, pixels); - temp2 = vec_ld(16, pixels); - pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels)); - if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) { - pixelsv2 = temp2; - } else { - pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels)); - } - pixelsv1 = vec_mergeh(vczero, pixelsv1); - pixelsv2 = vec_mergeh(vczero, pixelsv2); + pixelsv1 = VEC_LD(0, pixels); + pixelsv2 = VEC_LD(1, pixels); + pixelsv1 = VEC_MERGEH(vczero, pixelsv1); + pixelsv2 = VEC_MERGEH(vczero, pixelsv2); pixelssum1 = vec_add((vector unsigned short)pixelsv1, (vector unsigned short)pixelsv2); pixelssum1 = vec_add(pixelssum1, vctwo); @@ -414,17 +341,11 @@ static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi int rightside = ((unsigned long)block & 0x0000000F); blockv = vec_ld(0, block); - temp1 = vec_ld(line_size, pixels); - temp2 = vec_ld(line_size + 16, pixels); - pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels)); - if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) { - pixelsv2 = temp2; - } else { - pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels)); - } + pixelsv1 = unaligned_load(line_size, pixels); + pixelsv2 = unaligned_load(line_size+1, pixels); - pixelsv1 = vec_mergeh(vczero, pixelsv1); - pixelsv2 = vec_mergeh(vczero, pixelsv2); + pixelsv1 = VEC_MERGEH(vczero, pixelsv1); + pixelsv2 = VEC_MERGEH(vczero, pixelsv2); pixelssum2 = vec_add((vector unsigned short)pixelsv1, (vector unsigned short)pixelsv2); temp3 = vec_add(pixelssum1, pixelssum2); diff --git a/libavcodec/ppc/hpeldsp_altivec.h b/libavcodec/ppc/hpeldsp_altivec.h index 98dd80ea1c..590809f539 100644 --- a/libavcodec/ppc/hpeldsp_altivec.h +++ b/libavcodec/ppc/hpeldsp_altivec.h @@ -3,20 +3,20 @@ * Copyright (c) 2002 Dieter Shirley * Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/ppc/huffyuvdsp_altivec.c b/libavcodec/ppc/huffyuvdsp_altivec.c index 7c34a67ea4..6701524e4a 100644 --- a/libavcodec/ppc/huffyuvdsp_altivec.c +++ b/libavcodec/ppc/huffyuvdsp_altivec.c @@ -3,20 +3,20 @@ * Copyright (c) 2002 Dieter Shirley * Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -33,7 +33,7 @@ #include "libavcodec/huffyuvdsp.h" #if HAVE_ALTIVEC -static void add_bytes_altivec(uint8_t *dst, uint8_t *src, int w) +static void add_bytes_altivec(uint8_t *dst, uint8_t *src, intptr_t w) { register int i; register vector unsigned char vdst, vsrc; diff --git a/libavcodec/ppc/idctdsp.c b/libavcodec/ppc/idctdsp.c index 17f7dbbc7f..5ef514b51b 100644 --- a/libavcodec/ppc/idctdsp.c +++ b/libavcodec/ppc/idctdsp.c @@ -1,28 +1,28 @@ /* * Copyright (c) 2001 Michel Lespinasse * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ /* NOTE: This code is based on GPL code from the libmpeg2 project. The * author, Michel Lespinasses, has given explicit permission to release - * under LGPL as part of Libav. + * under LGPL as part of FFmpeg. * - * Libav integration by Dieter Shirley + * FFmpeg integration by Dieter Shirley * * This file is a direct copy of the AltiVec IDCT module from the libmpeg2 * project. I've deleted all of the libmpeg2-specific code, renamed the @@ -153,6 +153,22 @@ static const vec_s16 constants[5] = { { 19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722 } }; +static void idct_altivec(int16_t *blk) +{ + vec_s16 *block = (vec_s16 *) blk; + + IDCT; + + block[0] = vx0; + block[1] = vx1; + block[2] = vx2; + block[3] = vx3; + block[4] = vx4; + block[5] = vx5; + block[6] = vx6; + block[7] = vx7; +} + static void idct_put_altivec(uint8_t *dest, int stride, int16_t *blk) { vec_s16 *block = (vec_s16 *) blk; @@ -234,9 +250,10 @@ av_cold void ff_idctdsp_init_ppc(IDCTDSPContext *c, AVCodecContext *avctx, if (!PPC_ALTIVEC(av_get_cpu_flags())) return; - if (!high_bit_depth) { - if ((avctx->idct_algo == FF_IDCT_AUTO) || + if (!high_bit_depth && avctx->lowres == 0) { + if ((avctx->idct_algo == FF_IDCT_AUTO && !(avctx->flags & CODEC_FLAG_BITEXACT)) || (avctx->idct_algo == FF_IDCT_ALTIVEC)) { + c->idct = idct_altivec; c->idct_add = idct_add_altivec; c->idct_put = idct_put_altivec; c->perm_type = FF_IDCT_PERM_TRANSPOSE; diff --git a/libavcodec/ppc/apedsp_altivec.c b/libavcodec/ppc/lossless_audiodsp_altivec.c index d8bf4bd167..bdec25223d 100644 --- a/libavcodec/ppc/apedsp_altivec.c +++ b/libavcodec/ppc/lossless_audiodsp_altivec.c @@ -1,20 +1,20 @@ /* * Copyright (c) 2007 Luca Barbato <lu_zero@gentoo.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -27,7 +27,21 @@ #include "libavutil/cpu.h" #include "libavutil/ppc/cpu.h" #include "libavutil/ppc/types_altivec.h" -#include "libavcodec/apedsp.h" +#include "libavcodec/lossless_audiodsp.h" + +#if HAVE_BIGENDIAN +#define GET_T(tt0,tt1,src,a,b){ \ + a = vec_ld(16, src); \ + tt0 = vec_perm(b, a, align); \ + b = vec_ld(32, src); \ + tt1 = vec_perm(a, b, align); \ + } +#else +#define GET_T(tt0,tt1,src,a,b){ \ + tt0 = vec_vsx_ld(0, src); \ + tt1 = vec_vsx_ld(16, src); \ + } +#endif #if HAVE_ALTIVEC static int32_t scalarproduct_and_madd_int16_altivec(int16_t *v1, @@ -38,26 +52,23 @@ static int32_t scalarproduct_and_madd_int16_altivec(int16_t *v1, LOAD_ZERO; vec_s16 *pv1 = (vec_s16 *) v1; register vec_s16 muls = { mul, mul, mul, mul, mul, mul, mul, mul }; - register vec_s16 t0, t1, i0, i1, i4; - register vec_s16 i2 = vec_ld(0, v2), i3 = vec_ld(0, v3); + register vec_s16 t0, t1, i0, i1, i4, i2, i3; register vec_s32 res = zero_s32v; +#if HAVE_BIGENDIAN register vec_u8 align = vec_lvsl(0, v2); + i2 = vec_ld(0, v2); + i3 = vec_ld(0, v3); +#endif int32_t ires; order >>= 4; do { - i1 = vec_ld(16, v2); - t0 = vec_perm(i2, i1, align); - i2 = vec_ld(32, v2); - t1 = vec_perm(i1, i2, align); + GET_T(t0,t1,v2,i1,i2); i0 = pv1[0]; i1 = pv1[1]; res = vec_msum(t0, i0, res); res = vec_msum(t1, i1, res); - i4 = vec_ld(16, v3); - t0 = vec_perm(i3, i4, align); - i3 = vec_ld(32, v3); - t1 = vec_perm(i4, i3, align); + GET_T(t0,t1,v3,i4,i3); pv1[0] = vec_mladd(t0, muls, i0); pv1[1] = vec_mladd(t1, muls, i1); pv1 += 2; @@ -71,7 +82,7 @@ static int32_t scalarproduct_and_madd_int16_altivec(int16_t *v1, } #endif /* HAVE_ALTIVEC */ -av_cold void ff_apedsp_init_ppc(APEDSPContext *c) +av_cold void ff_llauddsp_init_ppc(LLAudDSPContext *c) { #if HAVE_ALTIVEC if (!PPC_ALTIVEC(av_get_cpu_flags())) diff --git a/libavcodec/ppc/mathops.h b/libavcodec/ppc/mathops.h index 34ddb11800..dbd714fcd4 100644 --- a/libavcodec/ppc/mathops.h +++ b/libavcodec/ppc/mathops.h @@ -3,20 +3,20 @@ * Copyright (c) 2001, 2002 Fabrice Bellard * Copyright (c) 2006 Michael Niedermayer <michaelni@gmx.at> et al * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/ppc/me_cmp.c b/libavcodec/ppc/me_cmp.c index fd67cf34ee..38a7ba1476 100644 --- a/libavcodec/ppc/me_cmp.c +++ b/libavcodec/ppc/me_cmp.c @@ -3,20 +3,20 @@ * Copyright (c) 2002 Dieter Shirley * Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -35,26 +35,43 @@ #include "libavcodec/me_cmp.h" #if HAVE_ALTIVEC + +#if HAVE_BIGENDIAN +#define GET_PERM(per1, per2, pix) {\ + per1 = vec_lvsl(0, pix);\ + per2 = vec_add(per1, vec_splat_u8(1));\ +} +#define LOAD_PIX(v, iv, pix, per1, per2) {\ + vector unsigned char pix2l = vec_ld(0, pix);\ + vector unsigned char pix2r = vec_ld(16, pix);\ + v = vec_perm(pix2l, pix2r, per1);\ + iv = vec_perm(pix2l, pix2r, per2);\ +} +#else +#define GET_PERM(per1, per2, pix) {} +#define LOAD_PIX(v, iv, pix, per1, per2) {\ + v = vec_vsx_ld(0, pix);\ + iv = vec_vsx_ld(1, pix);\ +} +#endif static int sad16_x2_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, ptrdiff_t stride, int h) { - int i, s = 0; + int i; + int __attribute__((aligned(16))) s = 0; const vector unsigned char zero = (const vector unsigned char) vec_splat_u8(0); - vector unsigned char perm1 = vec_lvsl(0, pix2); - vector unsigned char perm2 = vec_add(perm1, vec_splat_u8(1)); vector unsigned int sad = (vector unsigned int) vec_splat_u32(0); vector signed int sumdiffs; + vector unsigned char perm1, perm2, pix2v, pix2iv; + GET_PERM(perm1, perm2, pix2); for (i = 0; i < h; i++) { /* Read unaligned pixels into our vectors. The vectors are as follows: * pix1v: pix1[0] - pix1[15] * pix2v: pix2[0] - pix2[15] pix2iv: pix2[1] - pix2[16] */ vector unsigned char pix1v = vec_ld(0, pix1); - vector unsigned char pix2l = vec_ld(0, pix2); - vector unsigned char pix2r = vec_ld(16, pix2); - vector unsigned char pix2v = vec_perm(pix2l, pix2r, perm1); - vector unsigned char pix2iv = vec_perm(pix2l, pix2r, perm2); + LOAD_PIX(pix2v, pix2iv, pix2, perm1, perm2); /* Calculate the average vector. */ vector unsigned char avgv = vec_avg(pix2v, pix2iv); @@ -80,13 +97,14 @@ static int sad16_x2_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, static int sad16_y2_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, ptrdiff_t stride, int h) { - int i, s = 0; + int i; + int __attribute__((aligned(16))) s = 0; const vector unsigned char zero = (const vector unsigned char) vec_splat_u8(0); - vector unsigned char perm = vec_lvsl(0, pix2); vector unsigned char pix1v, pix3v, avgv, t5; vector unsigned int sad = (vector unsigned int) vec_splat_u32(0); vector signed int sumdiffs; + uint8_t *pix3 = pix2 + stride; /* Due to the fact that pix3 = pix2 + stride, the pix3 of one @@ -96,19 +114,14 @@ static int sad16_y2_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, * Read unaligned pixels into our vectors. The vectors are as follows: * pix2v: pix2[0] - pix2[15] * Split the pixel vectors into shorts. */ - vector unsigned char pix2l = vec_ld(0, pix2); - vector unsigned char pix2r = vec_ld(15, pix2); - vector unsigned char pix2v = vec_perm(pix2l, pix2r, perm); + vector unsigned char pix2v = VEC_LD(0, pix2); for (i = 0; i < h; i++) { /* Read unaligned pixels into our vectors. The vectors are as follows: * pix1v: pix1[0] - pix1[15] * pix3v: pix3[0] - pix3[15] */ pix1v = vec_ld(0, pix1); - - pix2l = vec_ld(0, pix3); - pix2r = vec_ld(15, pix3); - pix3v = vec_perm(pix2l, pix2r, perm); + pix3v = VEC_LD(0, pix3); /* Calculate the average vector. */ avgv = vec_avg(pix2v, pix3v); @@ -134,20 +147,21 @@ static int sad16_y2_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, static int sad16_xy2_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, ptrdiff_t stride, int h) { - int i, s = 0; + int i; + int __attribute__((aligned(16))) s = 0; uint8_t *pix3 = pix2 + stride; const vector unsigned char zero = (const vector unsigned char) vec_splat_u8(0); const vector unsigned short two = (const vector unsigned short) vec_splat_u16(2); vector unsigned char avgv, t5; - vector unsigned char perm1 = vec_lvsl(0, pix2); - vector unsigned char perm2 = vec_add(perm1, vec_splat_u8(1)); vector unsigned char pix1v, pix3v, pix3iv; vector unsigned short pix3lv, pix3hv, pix3ilv, pix3ihv; vector unsigned short avghv, avglv; vector unsigned int sad = (vector unsigned int) vec_splat_u32(0); vector signed int sumdiffs; + vector unsigned char perm1, perm2, pix2v, pix2iv; + GET_PERM(perm1, perm2, pix2); /* Due to the fact that pix3 = pix2 + stride, the pix3 of one * iteration becomes pix2 in the next iteration. We can use this @@ -156,19 +170,16 @@ static int sad16_xy2_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, * Read unaligned pixels into our vectors. The vectors are as follows: * pix2v: pix2[0] - pix2[15] pix2iv: pix2[1] - pix2[16] * Split the pixel vectors into shorts. */ - vector unsigned char pix2l = vec_ld(0, pix2); - vector unsigned char pix2r = vec_ld(16, pix2); - vector unsigned char pix2v = vec_perm(pix2l, pix2r, perm1); - vector unsigned char pix2iv = vec_perm(pix2l, pix2r, perm2); - + LOAD_PIX(pix2v, pix2iv, pix2, perm1, perm2); vector unsigned short pix2hv = - (vector unsigned short) vec_mergeh(zero, pix2v); + (vector unsigned short) VEC_MERGEH(zero, pix2v); vector unsigned short pix2lv = - (vector unsigned short) vec_mergel(zero, pix2v); + (vector unsigned short) VEC_MERGEL(zero, pix2v); vector unsigned short pix2ihv = - (vector unsigned short) vec_mergeh(zero, pix2iv); + (vector unsigned short) VEC_MERGEH(zero, pix2iv); vector unsigned short pix2ilv = - (vector unsigned short) vec_mergel(zero, pix2iv); + (vector unsigned short) VEC_MERGEL(zero, pix2iv); + vector unsigned short t1 = vec_add(pix2hv, pix2ihv); vector unsigned short t2 = vec_add(pix2lv, pix2ilv); vector unsigned short t3, t4; @@ -178,11 +189,7 @@ static int sad16_xy2_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, * pix1v: pix1[0] - pix1[15] * pix3v: pix3[0] - pix3[15] pix3iv: pix3[1] - pix3[16] */ pix1v = vec_ld(0, pix1); - - pix2l = vec_ld(0, pix3); - pix2r = vec_ld(16, pix3); - pix3v = vec_perm(pix2l, pix2r, perm1); - pix3iv = vec_perm(pix2l, pix2r, perm2); + LOAD_PIX(pix3v, pix3iv, pix3, perm1, perm2); /* Note that AltiVec does have vec_avg, but this works on vector pairs * and rounds up. We could do avg(avg(a, b), avg(c, d)), but the @@ -191,10 +198,10 @@ static int sad16_xy2_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, * vectors of shorts and do the averaging by hand. */ /* Split the pixel vectors into shorts. */ - pix3hv = (vector unsigned short) vec_mergeh(zero, pix3v); - pix3lv = (vector unsigned short) vec_mergel(zero, pix3v); - pix3ihv = (vector unsigned short) vec_mergeh(zero, pix3iv); - pix3ilv = (vector unsigned short) vec_mergel(zero, pix3iv); + pix3hv = (vector unsigned short) VEC_MERGEH(zero, pix3v); + pix3lv = (vector unsigned short) VEC_MERGEL(zero, pix3v); + pix3ihv = (vector unsigned short) VEC_MERGEH(zero, pix3iv); + pix3ilv = (vector unsigned short) VEC_MERGEL(zero, pix3iv); /* Do the averaging on them. */ t3 = vec_add(pix3hv, pix3ihv); @@ -229,19 +236,17 @@ static int sad16_xy2_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, static int sad16_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, ptrdiff_t stride, int h) { - int i, s; + int i; + int __attribute__((aligned(16))) s; const vector unsigned int zero = (const vector unsigned int) vec_splat_u32(0); - vector unsigned char perm = vec_lvsl(0, pix2); vector unsigned int sad = (vector unsigned int) vec_splat_u32(0); vector signed int sumdiffs; for (i = 0; i < h; i++) { /* Read potentially unaligned pixels into t1 and t2. */ - vector unsigned char pix2l = vec_ld(0, pix2); - vector unsigned char pix2r = vec_ld(15, pix2); - vector unsigned char t1 = vec_ld(0, pix1); - vector unsigned char t2 = vec_perm(pix2l, pix2r, perm); + vector unsigned char t1 =vec_ld(0, pix1); + vector unsigned char t2 = VEC_LD(0, pix2); /* Calculate a sum of abs differences vector. */ vector unsigned char t3 = vec_max(t1, t2); @@ -266,14 +271,13 @@ static int sad16_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, static int sad8_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, ptrdiff_t stride, int h) { - int i, s; + int i; + int __attribute__((aligned(16))) s; const vector unsigned int zero = (const vector unsigned int) vec_splat_u32(0); const vector unsigned char permclear = (vector unsigned char) { 255, 255, 255, 255, 255, 255, 255, 255, 0, 0, 0, 0, 0, 0, 0, 0 }; - vector unsigned char perm1 = vec_lvsl(0, pix1); - vector unsigned char perm2 = vec_lvsl(0, pix2); vector unsigned int sad = (vector unsigned int) vec_splat_u32(0); vector signed int sumdiffs; @@ -281,14 +285,10 @@ static int sad8_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, /* Read potentially unaligned pixels into t1 and t2. * Since we're reading 16 pixels, and actually only want 8, * mask out the last 8 pixels. The 0s don't change the sum. */ - vector unsigned char pix1l = vec_ld(0, pix1); - vector unsigned char pix1r = vec_ld(7, pix1); - vector unsigned char pix2l = vec_ld(0, pix2); - vector unsigned char pix2r = vec_ld(7, pix2); - vector unsigned char t1 = vec_and(vec_perm(pix1l, pix1r, perm1), - permclear); - vector unsigned char t2 = vec_and(vec_perm(pix2l, pix2r, perm2), - permclear); + vector unsigned char pix1l = VEC_LD(0, pix1); + vector unsigned char pix2l = VEC_LD(0, pix2); + vector unsigned char t1 = vec_and(pix1l, permclear); + vector unsigned char t2 = vec_and(pix2l, permclear); /* Calculate a sum of abs differences vector. */ vector unsigned char t3 = vec_max(t1, t2); @@ -315,14 +315,13 @@ static int sad8_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, static int sse8_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, ptrdiff_t stride, int h) { - int i, s; + int i; + int __attribute__((aligned(16))) s; const vector unsigned int zero = (const vector unsigned int) vec_splat_u32(0); const vector unsigned char permclear = (vector unsigned char) { 255, 255, 255, 255, 255, 255, 255, 255, 0, 0, 0, 0, 0, 0, 0, 0 }; - vector unsigned char perm1 = vec_lvsl(0, pix1); - vector unsigned char perm2 = vec_lvsl(0, pix2); vector unsigned int sum = (vector unsigned int) vec_splat_u32(0); vector signed int sumsqr; @@ -330,14 +329,8 @@ static int sse8_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, /* Read potentially unaligned pixels into t1 and t2. * Since we're reading 16 pixels, and actually only want 8, * mask out the last 8 pixels. The 0s don't change the sum. */ - vector unsigned char pix1l = vec_ld(0, pix1); - vector unsigned char pix1r = vec_ld(7, pix1); - vector unsigned char pix2l = vec_ld(0, pix2); - vector unsigned char pix2r = vec_ld(7, pix2); - vector unsigned char t1 = vec_and(vec_perm(pix1l, pix1r, perm1), - permclear); - vector unsigned char t2 = vec_and(vec_perm(pix2l, pix2r, perm2), - permclear); + vector unsigned char t1 = vec_and(VEC_LD(0, pix1), permclear); + vector unsigned char t2 = vec_and(VEC_LD(0, pix2), permclear); /* Since we want to use unsigned chars, we can take advantage * of the fact that abs(a - b) ^ 2 = (a - b) ^ 2. */ @@ -367,19 +360,17 @@ static int sse8_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, static int sse16_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, ptrdiff_t stride, int h) { - int i, s; + int i; + int __attribute__((aligned(16))) s; const vector unsigned int zero = (const vector unsigned int) vec_splat_u32(0); - vector unsigned char perm = vec_lvsl(0, pix2); vector unsigned int sum = (vector unsigned int) vec_splat_u32(0); vector signed int sumsqr; for (i = 0; i < h; i++) { /* Read potentially unaligned pixels into t1 and t2. */ - vector unsigned char pix2l = vec_ld(0, pix2); - vector unsigned char pix2r = vec_ld(15, pix2); vector unsigned char t1 = vec_ld(0, pix1); - vector unsigned char t2 = vec_perm(pix2l, pix2r, perm); + vector unsigned char t2 = VEC_LD(0, pix2); /* Since we want to use unsigned chars, we can take advantage * of the fact that abs(a - b) ^ 2 = (a - b) ^ 2. */ @@ -399,15 +390,15 @@ static int sse16_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, /* Sum up the four partial sums, and put the result into s. */ sumsqr = vec_sums((vector signed int) sum, (vector signed int) zero); sumsqr = vec_splat(sumsqr, 3); - vec_ste(sumsqr, 0, &s); + vec_ste(sumsqr, 0, &s); return s; } static int hadamard8_diff8x8_altivec(MpegEncContext *s, uint8_t *dst, uint8_t *src, ptrdiff_t stride, int h) { - int sum; + int __attribute__((aligned(16))) sum; register const vector unsigned char vzero = (const vector unsigned char) vec_splat_u8(0); register vector signed short temp0, temp1, temp2, temp3, temp4, @@ -432,24 +423,19 @@ static int hadamard8_diff8x8_altivec(MpegEncContext *s, uint8_t *dst, { 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07 }; + #define ONEITERBUTTERFLY(i, res) \ { \ - register vector unsigned char src1 = vec_ld(stride * i, src); \ - register vector unsigned char src2 = vec_ld(stride * i + 15, src); \ - register vector unsigned char srcO = \ - vec_perm(src1, src2, vec_lvsl(stride * i, src)); \ - register vector unsigned char dst1 = vec_ld(stride * i, dst); \ - register vector unsigned char dst2 = vec_ld(stride * i + 15, dst); \ - register vector unsigned char dstO = \ - vec_perm(dst1, dst2, vec_lvsl(stride * i, dst)); \ + register vector unsigned char srcO = unaligned_load(stride * i, src); \ + register vector unsigned char dstO = unaligned_load(stride * i, dst);\ \ /* Promote the unsigned chars to signed shorts. */ \ /* We're in the 8x8 function, we only care for the first 8. */ \ register vector signed short srcV = \ - (vector signed short) vec_mergeh((vector signed char) vzero, \ + (vector signed short) VEC_MERGEH((vector signed char) vzero, \ (vector signed char) srcO); \ register vector signed short dstV = \ - (vector signed short) vec_mergeh((vector signed char) vzero, \ + (vector signed short) VEC_MERGEH((vector signed char) vzero, \ (vector signed char) dstO); \ \ /* subtractions inside the first butterfly */ \ @@ -461,6 +447,7 @@ static int hadamard8_diff8x8_altivec(MpegEncContext *s, uint8_t *dst, register vector signed short op3 = vec_perm(but2, but2, perm3); \ res = vec_mladd(but2, vprod3, op3); \ } + ONEITERBUTTERFLY(0, temp0); ONEITERBUTTERFLY(1, temp1); ONEITERBUTTERFLY(2, temp2); @@ -510,6 +497,7 @@ static int hadamard8_diff8x8_altivec(MpegEncContext *s, uint8_t *dst, vsum = vec_sum4s(vec_abs(line7C), vsum); vsum = vec_sums(vsum, (vector signed int) vzero); vsum = vec_splat(vsum, 3); + vec_ste(vsum, 0, &sum); } return sum; @@ -536,7 +524,7 @@ static int hadamard8_diff8x8_altivec(MpegEncContext *s, uint8_t *dst, static int hadamard8_diff16x8_altivec(MpegEncContext *s, uint8_t *dst, uint8_t *src, ptrdiff_t stride, int h) { - int sum; + int __attribute__((aligned(16))) sum; register vector signed short temp0 __asm__ ("v0"), temp1 __asm__ ("v1"), @@ -584,31 +572,23 @@ static int hadamard8_diff16x8_altivec(MpegEncContext *s, uint8_t *dst, #define ONEITERBUTTERFLY(i, res1, res2) \ { \ - register vector unsigned char src1 __asm__ ("v22") = \ - vec_ld(stride * i, src); \ - register vector unsigned char src2 __asm__ ("v23") = \ - vec_ld(stride * i + 16, src); \ register vector unsigned char srcO __asm__ ("v22") = \ - vec_perm(src1, src2, vec_lvsl(stride * i, src)); \ - register vector unsigned char dst1 __asm__ ("v24") = \ - vec_ld(stride * i, dst); \ - register vector unsigned char dst2 __asm__ ("v25") = \ - vec_ld(stride * i + 16, dst); \ + unaligned_load(stride * i, src); \ register vector unsigned char dstO __asm__ ("v23") = \ - vec_perm(dst1, dst2, vec_lvsl(stride * i, dst)); \ + unaligned_load(stride * i, dst);\ \ /* Promote the unsigned chars to signed shorts. */ \ register vector signed short srcV __asm__ ("v24") = \ - (vector signed short) vec_mergeh((vector signed char) vzero, \ + (vector signed short) VEC_MERGEH((vector signed char) vzero, \ (vector signed char) srcO); \ register vector signed short dstV __asm__ ("v25") = \ - (vector signed short) vec_mergeh((vector signed char) vzero, \ + (vector signed short) VEC_MERGEH((vector signed char) vzero, \ (vector signed char) dstO); \ register vector signed short srcW __asm__ ("v26") = \ - (vector signed short) vec_mergel((vector signed char) vzero, \ + (vector signed short) VEC_MERGEL((vector signed char) vzero, \ (vector signed char) srcO); \ register vector signed short dstW __asm__ ("v27") = \ - (vector signed short) vec_mergel((vector signed char) vzero, \ + (vector signed short) VEC_MERGEL((vector signed char) vzero, \ (vector signed char) dstO); \ \ /* subtractions inside the first butterfly */ \ @@ -639,6 +619,7 @@ static int hadamard8_diff16x8_altivec(MpegEncContext *s, uint8_t *dst, res1 = vec_mladd(but2, vprod3, op3); \ res2 = vec_mladd(but2S, vprod3, op3S); \ } + ONEITERBUTTERFLY(0, temp0, temp0S); ONEITERBUTTERFLY(1, temp1, temp1S); ONEITERBUTTERFLY(2, temp2, temp2S); @@ -725,6 +706,7 @@ static int hadamard8_diff16x8_altivec(MpegEncContext *s, uint8_t *dst, vsum = vec_sum4s(vec_abs(line7CS), vsum); vsum = vec_sums(vsum, (vector signed int) vzero); vsum = vec_splat(vsum, 3); + vec_ste(vsum, 0, &sum); } return sum; diff --git a/libavcodec/ppc/mpegaudiodsp_altivec.c b/libavcodec/ppc/mpegaudiodsp_altivec.c index c37f8ec975..ddfe5dcbb7 100644 --- a/libavcodec/ppc/mpegaudiodsp_altivec.c +++ b/libavcodec/ppc/mpegaudiodsp_altivec.c @@ -2,20 +2,20 @@ * Altivec optimized MP3 decoding functions * Copyright (c) 2010 Vitor Sessak * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/ppc/mpegvideo_altivec.c b/libavcodec/ppc/mpegvideo_altivec.c index 9ae849f173..ce53ae4b61 100644 --- a/libavcodec/ppc/mpegvideo_altivec.c +++ b/libavcodec/ppc/mpegvideo_altivec.c @@ -4,20 +4,20 @@ * dct_unquantize_h263_altivec: * Copyright (c) 2003 Romain Dolbeau <romain@dolbeau.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/ppc/mpegvideodsp.c b/libavcodec/ppc/mpegvideodsp.c index 2bdf909e4a..7696954335 100644 --- a/libavcodec/ppc/mpegvideodsp.c +++ b/libavcodec/ppc/mpegvideodsp.c @@ -3,20 +3,20 @@ * * Copyright (c) 2003 Romain Dolbeau <romain@dolbeau.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -66,7 +66,7 @@ static void gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */, vec_lvsl(0, src)); if (src_really_odd != 0x0000000F) - /* If src & 0xF == 0xF, then (src + 1) is properly aligned + /* If (src & 0xF) == 0xF, then (src + 1) is properly aligned * on the second vector. */ srcvB = vec_perm(src_0, src_1, vec_lvsl(1, src)); else @@ -88,7 +88,7 @@ static void gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */, srcvC = vec_perm(src_0, src_1, vec_lvsl(stride + 0, src)); if (src_really_odd != 0x0000000F) - /* If src & 0xF == 0xF, then (src + 1) is properly aligned + /* If (src & 0xF) == 0xF, then (src + 1) is properly aligned * on the second vector. */ srcvD = vec_perm(src_0, src_1, vec_lvsl(stride + 1, src)); else diff --git a/libavcodec/ppc/mpegvideoencdsp.c b/libavcodec/ppc/mpegvideoencdsp.c index b5348e6abd..e91ba5d25f 100644 --- a/libavcodec/ppc/mpegvideoencdsp.c +++ b/libavcodec/ppc/mpegvideoencdsp.c @@ -1,18 +1,18 @@ /* - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -31,6 +31,34 @@ #if HAVE_ALTIVEC +#if HAVE_VSX +static int pix_norm1_altivec(uint8_t *pix, int line_size) +{ + int i, s = 0; + const vector unsigned int zero = + (const vector unsigned int) vec_splat_u32(0); + vector unsigned int sv = (vector unsigned int) vec_splat_u32(0); + vector signed int sum; + + for (i = 0; i < 16; i++) { + /* Read the potentially unaligned pixels. */ + //vector unsigned char pixl = vec_ld(0, pix); + //vector unsigned char pixr = vec_ld(15, pix); + //vector unsigned char pixv = vec_perm(pixl, pixr, perm); + vector unsigned char pixv = vec_vsx_ld(0, pix); + + /* Square the values, and add them to our sum. */ + sv = vec_msum(pixv, pixv, sv); + + pix += line_size; + } + /* Sum up the four partial sums, and put the result into s. */ + sum = vec_sums((vector signed int) sv, (vector signed int) zero); + sum = vec_splat(sum, 3); + vec_vsx_st(sum, 0, &s); + return s; +} +#else static int pix_norm1_altivec(uint8_t *pix, int line_size) { int i, s = 0; @@ -58,7 +86,37 @@ static int pix_norm1_altivec(uint8_t *pix, int line_size) return s; } +#endif /* HAVE_VSX */ + +#if HAVE_VSX +static int pix_sum_altivec(uint8_t *pix, int line_size) +{ + int i, s; + const vector unsigned int zero = + (const vector unsigned int) vec_splat_u32(0); + vector unsigned int sad = (vector unsigned int) vec_splat_u32(0); + vector signed int sumdiffs; + + for (i = 0; i < 16; i++) { + /* Read the potentially unaligned 16 pixels into t1. */ + //vector unsigned char pixl = vec_ld(0, pix); + //vector unsigned char pixr = vec_ld(15, pix); + //vector unsigned char t1 = vec_perm(pixl, pixr, perm); + vector unsigned char t1 = vec_vsx_ld(0, pix); + + /* Add each 4 pixel group together and put 4 results into sad. */ + sad = vec_sum4s(t1, sad); + + pix += line_size; + } + /* Sum up the four partial sums, and put the result into s. */ + sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero); + sumdiffs = vec_splat(sumdiffs, 3); + vec_vsx_st(sumdiffs, 0, &s); + return s; +} +#else static int pix_sum_altivec(uint8_t *pix, int line_size) { int i, s; @@ -88,6 +146,8 @@ static int pix_sum_altivec(uint8_t *pix, int line_size) return s; } +#endif /* HAVE_VSX */ + #endif /* HAVE_ALTIVEC */ av_cold void ff_mpegvideoencdsp_init_ppc(MpegvideoEncDSPContext *c, diff --git a/libavcodec/ppc/pixblockdsp.c b/libavcodec/ppc/pixblockdsp.c index 698d655fc6..9bbdf96d12 100644 --- a/libavcodec/ppc/pixblockdsp.c +++ b/libavcodec/ppc/pixblockdsp.c @@ -3,20 +3,20 @@ * Copyright (c) 2002 Dieter Shirley * Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -35,8 +35,36 @@ #if HAVE_ALTIVEC +#if HAVE_VSX static void get_pixels_altivec(int16_t *restrict block, const uint8_t *pixels, - int line_size) + ptrdiff_t line_size) +{ + int i; + vector unsigned char perm = + (vector unsigned char) {0x00,0x10, 0x01,0x11,0x02,0x12,0x03,0x13,\ + 0x04,0x14,0x05,0x15,0x06,0x16,0x07,0x17}; + const vector unsigned char zero = + (const vector unsigned char) vec_splat_u8(0); + + for (i = 0; i < 8; i++) { + /* Read potentially unaligned pixels. + * We're reading 16 pixels, and actually only want 8, + * but we simply ignore the extras. */ + vector unsigned char bytes = vec_vsx_ld(0, pixels); + + // Convert the bytes into shorts. + //vector signed short shorts = (vector signed short) vec_perm(zero, bytes, perm); + vector signed short shorts = (vector signed short) vec_perm(bytes, zero, perm); + + // Save the data to the block, we assume the block is 16-byte aligned. + vec_vsx_st(shorts, i * 16, (vector signed short *) block); + + pixels += line_size; + } +} +#else +static void get_pixels_altivec(int16_t *restrict block, const uint8_t *pixels, + ptrdiff_t line_size) { int i; vector unsigned char perm = vec_lvsl(0, pixels); @@ -62,6 +90,71 @@ static void get_pixels_altivec(int16_t *restrict block, const uint8_t *pixels, } } +#endif /* HAVE_VSX */ + +#if HAVE_VSX +static void diff_pixels_altivec(int16_t *restrict block, const uint8_t *s1, + const uint8_t *s2, int stride) +{ + int i; + const vector unsigned char zero = + (const vector unsigned char) vec_splat_u8(0); + vector signed short shorts1, shorts2; + + for (i = 0; i < 4; i++) { + /* Read potentially unaligned pixels. + * We're reading 16 pixels, and actually only want 8, + * but we simply ignore the extras. */ + vector unsigned char bytes = vec_vsx_ld(0, s1); + + // Convert the bytes into shorts. + shorts1 = (vector signed short) vec_mergeh(bytes, zero); + + // Do the same for the second block of pixels. + bytes =vec_vsx_ld(0, s2); + + // Convert the bytes into shorts. + shorts2 = (vector signed short) vec_mergeh(bytes, zero); + + // Do the subtraction. + shorts1 = vec_sub(shorts1, shorts2); + + // Save the data to the block, we assume the block is 16-byte aligned. + vec_vsx_st(shorts1, 0, (vector signed short *) block); + + s1 += stride; + s2 += stride; + block += 8; + + /* The code below is a copy of the code above... + * This is a manual unroll. */ + + /* Read potentially unaligned pixels. + * We're reading 16 pixels, and actually only want 8, + * but we simply ignore the extras. */ + bytes = vec_vsx_ld(0, s1); + + // Convert the bytes into shorts. + shorts1 = (vector signed short) vec_mergeh(bytes, zero); + + // Do the same for the second block of pixels. + bytes = vec_vsx_ld(0, s2); + + // Convert the bytes into shorts. + shorts2 = (vector signed short) vec_mergeh(bytes, zero); + + // Do the subtraction. + shorts1 = vec_sub(shorts1, shorts2); + + // Save the data to the block, we assume the block is 16-byte aligned. + vec_vsx_st(shorts1, 0, (vector signed short *) block); + + s1 += stride; + s2 += stride; + block += 8; + } +} +#else static void diff_pixels_altivec(int16_t *restrict block, const uint8_t *s1, const uint8_t *s2, int stride) { @@ -134,6 +227,8 @@ static void diff_pixels_altivec(int16_t *restrict block, const uint8_t *s1, } } +#endif /* HAVE_VSX */ + #endif /* HAVE_ALTIVEC */ av_cold void ff_pixblockdsp_init_ppc(PixblockDSPContext *c, diff --git a/libavcodec/ppc/svq1enc_altivec.c b/libavcodec/ppc/svq1enc_altivec.c index 564f12986b..4e25e253f6 100644 --- a/libavcodec/ppc/svq1enc_altivec.c +++ b/libavcodec/ppc/svq1enc_altivec.c @@ -1,20 +1,20 @@ /* * Copyright (c) 2007 Luca Barbato <lu_zero@gentoo.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/ppc/vc1dsp_altivec.c b/libavcodec/ppc/vc1dsp_altivec.c index 90c3d27e82..35bb280842 100644 --- a/libavcodec/ppc/vc1dsp_altivec.c +++ b/libavcodec/ppc/vc1dsp_altivec.c @@ -2,20 +2,20 @@ * VC-1 and WMV3 decoder - DSP functions AltiVec-optimized * Copyright (c) 2006 Konstantin Shishkov * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -304,16 +304,23 @@ static void vc1_inv_trans_8x4_altivec(uint8_t *dest, int stride, int16_t *block) src2 = vec_pack(s2, sA); src3 = vec_pack(s3, sB); +#if HAVE_BIGENDIAN p0 = vec_lvsl (0, dest); p1 = vec_lvsl (stride, dest); p = vec_splat_u8 (-1); perm0 = vec_mergeh (p, p0); perm1 = vec_mergeh (p, p1); +#define GET_TMP2(dst, p) \ + tmp = vec_ld (0, dest); \ + tmp2 = (vector signed short)vec_perm (tmp, vec_splat_u8(0), p); +#else +#define GET_TMP2(dst,p) \ + tmp = vec_vsx_ld (0, dst); \ + tmp2 = (vector signed short)vec_mergeh (tmp, vec_splat_u8(0)); +#endif #define ADD(dest,src,perm) \ - /* *(uint64_t *)&tmp = *(uint64_t *)dest; */ \ - tmp = vec_ld (0, dest); \ - tmp2 = (vector signed short)vec_perm (tmp, vec_splat_u8(0), perm); \ + GET_TMP2(dest, perm); \ tmp3 = vec_adds (tmp2, src); \ tmp = vec_packsu (tmp3, tmp3); \ vec_ste ((vector unsigned int)tmp, 0, (unsigned int *)dest); \ diff --git a/libavcodec/ppc/videodsp_ppc.c b/libavcodec/ppc/videodsp_ppc.c index b9e003b487..915702252e 100644 --- a/libavcodec/ppc/videodsp_ppc.c +++ b/libavcodec/ppc/videodsp_ppc.c @@ -1,20 +1,20 @@ /* * Copyright (c) 2003-2004 Romain Dolbeau * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/ppc/vorbisdsp_altivec.c b/libavcodec/ppc/vorbisdsp_altivec.c index 43f4d0325b..d7557c815b 100644 --- a/libavcodec/ppc/vorbisdsp_altivec.c +++ b/libavcodec/ppc/vorbisdsp_altivec.c @@ -1,20 +1,20 @@ /* * Copyright (c) 2006 Luca Barbato <lu_zero@gentoo.org> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/ppc/vp3dsp_altivec.c b/libavcodec/ppc/vp3dsp_altivec.c index bce49e3170..4a367b655e 100644 --- a/libavcodec/ppc/vp3dsp_altivec.c +++ b/libavcodec/ppc/vp3dsp_altivec.c @@ -1,20 +1,20 @@ /* * Copyright (C) 2009 David Conrad * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -32,8 +32,13 @@ static const vec_s16 constants = {0, 64277, 60547, 54491, 46341, 36410, 25080, 12785}; +#if HAVE_BIGENDIAN static const vec_u8 interleave_high = {0, 1, 16, 17, 4, 5, 20, 21, 8, 9, 24, 25, 12, 13, 28, 29}; +#else +static const vec_u8 interleave_high = + {2, 3, 18, 19, 6, 7, 22, 23, 10, 11, 26, 27, 14, 15, 30, 31}; +#endif #define IDCT_START \ vec_s16 A, B, C, D, Ad, Bd, Cd, Dd, E, F, G, H;\ @@ -156,9 +161,18 @@ static void vp3_idct_add_altivec(uint8_t *dst, int stride, int16_t block[64]) TRANSPOSE8(b0, b1, b2, b3, b4, b5, b6, b7); IDCT_1D(ADD8, SHIFT4) -#define ADD(a)\ +#if HAVE_BIGENDIAN +#define GET_VDST16\ vdst = vec_ld(0, dst);\ - vdst_16 = (vec_s16)vec_perm(vdst, zero_u8v, vdst_mask);\ + vdst_16 = (vec_s16)vec_perm(vdst, zero_u8v, vdst_mask); +#else +#define GET_VDST16\ + vdst = vec_vsx_ld(0,dst);\ + vdst_16 = (vec_s16)vec_mergeh(vdst, zero_u8v); +#endif + +#define ADD(a)\ + GET_VDST16;\ vdst_16 = vec_adds(a, vdst_16);\ t = vec_packsu(vdst_16, vdst_16);\ vec_ste((vec_u32)t, 0, (unsigned int *)dst);\ diff --git a/libavcodec/ppc/vp8dsp_altivec.c b/libavcodec/ppc/vp8dsp_altivec.c index e010dee4d2..23e4ace7da 100644 --- a/libavcodec/ppc/vp8dsp_altivec.c +++ b/libavcodec/ppc/vp8dsp_altivec.c @@ -3,20 +3,20 @@ * * Copyright (C) 2010 David Conrad * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -59,17 +59,30 @@ static const vec_s8 h_subpel_filters_outer[3] = vec_s8 filter_outerh = h_subpel_filters_outer[(i)>>1]; \ vec_s8 filter_outerl = vec_sld(filter_outerh, filter_outerh, 2) +#if HAVE_BIGENDIAN +#define GET_PIXHL(offset) \ + a = vec_ld((offset)-is6tap-1, src); \ + b = vec_ld((offset)-is6tap-1+15, src); \ + pixh = vec_perm(a, b, permh##offset); \ + pixl = vec_perm(a, b, perml##offset) + +#define GET_OUTER(offset) outer = vec_perm(a, b, perm_6tap##offset) +#else +#define GET_PIXHL(offset) \ + a = vec_vsx_ld((offset)-is6tap-1, src); \ + pixh = vec_perm(a, a, perm_inner); \ + pixl = vec_perm(a, a, vec_add(perm_inner, vec_splat_u8(4))) + +#define GET_OUTER(offset) outer = vec_perm(a, a, perm_outer) +#endif + #define FILTER_H(dstv, off) \ - a = vec_ld((off)-is6tap-1, src); \ - b = vec_ld((off)-is6tap-1+15, src); \ -\ - pixh = vec_perm(a, b, permh##off); \ - pixl = vec_perm(a, b, perml##off); \ + GET_PIXHL(off); \ filth = vec_msum(filter_inner, pixh, c64); \ filtl = vec_msum(filter_inner, pixl, c64); \ \ if (is6tap) { \ - outer = vec_perm(a, b, perm_6tap##off); \ + GET_OUTER(off); \ filth = vec_msum(filter_outerh, outer, filth); \ filtl = vec_msum(filter_outerl, outer, filtl); \ } \ @@ -84,9 +97,12 @@ void put_vp8_epel_h_altivec_core(uint8_t *dst, ptrdiff_t dst_stride, int h, int mx, int w, int is6tap) { LOAD_H_SUBPEL_FILTER(mx-1); - vec_u8 align_vec0, align_vec8, permh0, permh8, filt; +#if HAVE_BIGENDIAN + vec_u8 align_vec0, align_vec8, permh0, permh8; vec_u8 perm_6tap0, perm_6tap8, perml0, perml8; - vec_u8 a, b, pixh, pixl, outer; + vec_u8 b; +#endif + vec_u8 filt, a, pixh, pixl, outer; vec_s16 f16h, f16l; vec_s32 filth, filtl; @@ -97,6 +113,7 @@ void put_vp8_epel_h_altivec_core(uint8_t *dst, ptrdiff_t dst_stride, vec_s32 c64 = vec_sl(vec_splat_s32(1), vec_splat_u32(6)); vec_u16 c7 = vec_splat_u16(7); +#if HAVE_BIGENDIAN align_vec0 = vec_lvsl( -is6tap-1, src); align_vec8 = vec_lvsl(8-is6tap-1, src); @@ -107,6 +124,7 @@ void put_vp8_epel_h_altivec_core(uint8_t *dst, ptrdiff_t dst_stride, perml8 = vec_perm(align_vec8, align_vec8, perm_inner); perm_6tap0 = vec_perm(align_vec0, align_vec0, perm_outer); perm_6tap8 = vec_perm(align_vec8, align_vec8, perm_outer); +#endif while (h --> 0) { FILTER_H(f16h, 0); @@ -164,6 +182,12 @@ static const vec_u8 v_subpel_filters[7] = dstv = vec_adds(dstv, c64); \ dstv = vec_sra(dstv, c7) +#if HAVE_BIGENDIAN +#define LOAD_HL(off, s, perm) load_with_perm_vec(off, s, perm) +#else +#define LOAD_HL(off, s, perm) vec_mergeh(vec_vsx_ld(off,s), vec_vsx_ld(off+8,s)) +#endif + static av_always_inline void put_vp8_epel_v_altivec_core(uint8_t *dst, ptrdiff_t dst_stride, uint8_t *src, ptrdiff_t src_stride, @@ -175,6 +199,7 @@ void put_vp8_epel_v_altivec_core(uint8_t *dst, ptrdiff_t dst_stride, vec_s16 c64 = vec_sl(vec_splat_s16(1), vec_splat_u16(6)); vec_u16 c7 = vec_splat_u16(7); +#if HAVE_BIGENDIAN // we want pixels 0-7 to be in the even positions and 8-15 in the odd, // so combine this permute with the alignment permute vector align_vech = vec_lvsl(0, src); @@ -183,22 +208,23 @@ void put_vp8_epel_v_altivec_core(uint8_t *dst, ptrdiff_t dst_stride, perm_vec = vec_mergeh(align_vech, align_vecl); else perm_vec = vec_mergeh(align_vech, align_vech); +#endif if (is6tap) - s0 = load_with_perm_vec(-2*src_stride, src, perm_vec); - s1 = load_with_perm_vec(-1*src_stride, src, perm_vec); - s2 = load_with_perm_vec( 0*src_stride, src, perm_vec); - s3 = load_with_perm_vec( 1*src_stride, src, perm_vec); + s0 = LOAD_HL(-2*src_stride, src, perm_vec); + s1 = LOAD_HL(-1*src_stride, src, perm_vec); + s2 = LOAD_HL( 0*src_stride, src, perm_vec); + s3 = LOAD_HL( 1*src_stride, src, perm_vec); if (is6tap) - s4 = load_with_perm_vec( 2*src_stride, src, perm_vec); + s4 = LOAD_HL( 2*src_stride, src, perm_vec); src += (2+is6tap)*src_stride; while (h --> 0) { if (is6tap) - s5 = load_with_perm_vec(0, src, perm_vec); + s5 = LOAD_HL(0, src, perm_vec); else - s4 = load_with_perm_vec(0, src, perm_vec); + s4 = LOAD_HL(0, src, perm_vec); FILTER_V(f16h, vec_mule); @@ -272,39 +298,25 @@ EPEL_HV(4, 4,4) static void put_vp8_pixels16_altivec(uint8_t *dst, ptrdiff_t dstride, uint8_t *src, ptrdiff_t sstride, int h, int mx, int my) { - register vector unsigned char pixelsv1, pixelsv2; - register vector unsigned char pixelsv1B, pixelsv2B; - register vector unsigned char pixelsv1C, pixelsv2C; - register vector unsigned char pixelsv1D, pixelsv2D; - - register vector unsigned char perm = vec_lvsl(0, src); + register vector unsigned char perm; int i; register ptrdiff_t dstride2 = dstride << 1, sstride2 = sstride << 1; register ptrdiff_t dstride3 = dstride2 + dstride, sstride3 = sstride + sstride2; register ptrdiff_t dstride4 = dstride << 2, sstride4 = sstride << 2; +#if HAVE_BIGENDIAN + perm = vec_lvsl(0, src); +#endif // hand-unrolling the loop by 4 gains about 15% // mininum execution time goes from 74 to 60 cycles // it's faster than -funroll-loops, but using // -funroll-loops w/ this is bad - 74 cycles again. // all this is on a 7450, tuning for the 7450 for (i = 0; i < h; i += 4) { - pixelsv1 = vec_ld( 0, src); - pixelsv2 = vec_ld(15, src); - pixelsv1B = vec_ld(sstride, src); - pixelsv2B = vec_ld(15 + sstride, src); - pixelsv1C = vec_ld(sstride2, src); - pixelsv2C = vec_ld(15 + sstride2, src); - pixelsv1D = vec_ld(sstride3, src); - pixelsv2D = vec_ld(15 + sstride3, src); - vec_st(vec_perm(pixelsv1, pixelsv2, perm), - 0, (unsigned char*)dst); - vec_st(vec_perm(pixelsv1B, pixelsv2B, perm), - dstride, (unsigned char*)dst); - vec_st(vec_perm(pixelsv1C, pixelsv2C, perm), - dstride2, (unsigned char*)dst); - vec_st(vec_perm(pixelsv1D, pixelsv2D, perm), - dstride3, (unsigned char*)dst); + vec_st(load_with_perm_vec(0, src, perm), 0, dst); + vec_st(load_with_perm_vec(sstride, src, perm), dstride, dst); + vec_st(load_with_perm_vec(sstride2, src, perm), dstride2, dst); + vec_st(load_with_perm_vec(sstride3, src, perm), dstride3, dst); src += sstride4; dst += dstride4; } |