diff --git a/MAINTAINERS b/MAINTAINERS index eebfa5cfb7..f9adca2197 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -110,8 +110,6 @@ Generic Parts: lzw.* Michael Niedermayer floating point AAN DCT: faandct.c, faandct.h Michael Niedermayer - Non-power-of-two MDCT: - mdct15.c, mdct15.h Rostislav Pehlivanov Golomb coding: golomb.c, golomb.h Michael Niedermayer motion estimation: diff --git a/configure b/configure index 1ff9b0de62..590a260e52 100755 --- a/configure +++ b/configure @@ -2477,7 +2477,6 @@ CONFIG_EXTRA=" huffyuvencdsp idctdsp iirfilter - mdct15 inflate_wrapper intrax8 iso_media @@ -2749,7 +2748,6 @@ inflate_wrapper_deps="zlib" intrax8_select="blockdsp wmv2dsp" iso_media_select="mpeg4audio" mdct_select="fft" -mdct15_select="fft" me_cmp_select="idctdsp" mpeg_er_select="error_resilience" mpegaudio_select="mpegaudiodsp mpegaudioheader" diff --git a/libavcodec/Makefile b/libavcodec/Makefile index 32318fd7ed..23e368137a 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -107,7 +107,6 @@ OBJS-$(CONFIG_HUFFYUVDSP) += huffyuvdsp.o OBJS-$(CONFIG_HUFFYUVENCDSP) += huffyuvencdsp.o OBJS-$(CONFIG_IDCTDSP) += idctdsp.o simple_idct.o jrevdct.o OBJS-$(CONFIG_IIRFILTER) += iirfilter.o -OBJS-$(CONFIG_MDCT15) += mdct15.o OBJS-$(CONFIG_INFLATE_WRAPPER) += zlib_wrapper.o OBJS-$(CONFIG_INTRAX8) += intrax8.o intrax8dsp.o msmpeg4data.o OBJS-$(CONFIG_IVIDSP) += ivi_dsp.o diff --git a/libavcodec/mdct15.c b/libavcodec/mdct15.c deleted file mode 100644 index f8ed74d15e..0000000000 --- a/libavcodec/mdct15.c +++ /dev/null @@ -1,331 +0,0 @@ -/* - * Copyright (c) 2013-2014 Mozilla Corporation - * Copyright (c) 2017 Rostislav Pehlivanov - * - * 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 - */ - -/** - * @file - * Celt non-power of 2 iMDCT - */ - -#include -#include -#include -#include - -#include "config.h" - -#include "libavutil/attributes.h" -#include "libavutil/error.h" - -#include "mdct15.h" - -#define FFT_FLOAT 1 -#include "fft-internal.h" - -#define CMUL3(c, a, b) CMUL((c).re, (c).im, (a).re, (a).im, (b).re, (b).im) - -av_cold void ff_mdct15_uninit(MDCT15Context **ps) -{ - MDCT15Context *s = *ps; - - if (!s) - return; - - ff_fft_end(&s->ptwo_fft); - - av_freep(&s->pfa_prereindex); - av_freep(&s->pfa_postreindex); - av_freep(&s->twiddle_exptab); - av_freep(&s->tmp); - - av_freep(ps); -} - -static inline int init_pfa_reindex_tabs(MDCT15Context *s) -{ - int i, j; - const int b_ptwo = s->ptwo_fft.nbits; /* Bits for the power of two FFTs */ - const int l_ptwo = 1 << b_ptwo; /* Total length for the power of two FFTs */ - const int inv_1 = l_ptwo << ((4 - b_ptwo) & 3); /* (2^b_ptwo)^-1 mod 15 */ - const int inv_2 = 0xeeeeeeef & ((1U << b_ptwo) - 1); /* 15^-1 mod 2^b_ptwo */ - - s->pfa_prereindex = av_malloc_array(15 * l_ptwo, sizeof(*s->pfa_prereindex)); - if (!s->pfa_prereindex) - return 1; - - s->pfa_postreindex = av_malloc_array(15 * l_ptwo, sizeof(*s->pfa_postreindex)); - if (!s->pfa_postreindex) - return 1; - - /* Pre/Post-reindex */ - for (i = 0; i < l_ptwo; i++) { - for (j = 0; j < 15; j++) { - const int q_pre = ((l_ptwo * j)/15 + i) >> b_ptwo; - const int q_post = (((j*inv_1)/15) + (i*inv_2)) >> b_ptwo; - const int k_pre = 15*i + (j - q_pre*15)*(1 << b_ptwo); - const int k_post = i*inv_2*15 + j*inv_1 - 15*q_post*l_ptwo; - s->pfa_prereindex[i*15 + j] = k_pre << 1; - s->pfa_postreindex[k_post] = l_ptwo*j + i; - } - } - - return 0; -} - -/* Stride is hardcoded to 3 */ -static inline void fft5(FFTComplex *out, FFTComplex *in, FFTComplex exptab[2]) -{ - FFTComplex z0[4], t[6]; - - t[0].re = in[3].re + in[12].re; - t[0].im = in[3].im + in[12].im; - t[1].im = in[3].re - in[12].re; - t[1].re = in[3].im - in[12].im; - t[2].re = in[6].re + in[ 9].re; - t[2].im = in[6].im + in[ 9].im; - t[3].im = in[6].re - in[ 9].re; - t[3].re = in[6].im - in[ 9].im; - - out[0].re = in[0].re + in[3].re + in[6].re + in[9].re + in[12].re; - out[0].im = in[0].im + in[3].im + in[6].im + in[9].im + in[12].im; - - t[4].re = exptab[0].re * t[2].re - exptab[1].re * t[0].re; - t[4].im = exptab[0].re * t[2].im - exptab[1].re * t[0].im; - t[0].re = exptab[0].re * t[0].re - exptab[1].re * t[2].re; - t[0].im = exptab[0].re * t[0].im - exptab[1].re * t[2].im; - t[5].re = exptab[0].im * t[3].re - exptab[1].im * t[1].re; - t[5].im = exptab[0].im * t[3].im - exptab[1].im * t[1].im; - t[1].re = exptab[0].im * t[1].re + exptab[1].im * t[3].re; - t[1].im = exptab[0].im * t[1].im + exptab[1].im * t[3].im; - - z0[0].re = t[0].re - t[1].re; - z0[0].im = t[0].im - t[1].im; - z0[1].re = t[4].re + t[5].re; - z0[1].im = t[4].im + t[5].im; - - z0[2].re = t[4].re - t[5].re; - z0[2].im = t[4].im - t[5].im; - z0[3].re = t[0].re + t[1].re; - z0[3].im = t[0].im + t[1].im; - - out[1].re = in[0].re + z0[3].re; - out[1].im = in[0].im + z0[0].im; - out[2].re = in[0].re + z0[2].re; - out[2].im = in[0].im + z0[1].im; - out[3].re = in[0].re + z0[1].re; - out[3].im = in[0].im + z0[2].im; - out[4].re = in[0].re + z0[0].re; - out[4].im = in[0].im + z0[3].im; -} - -static void fft15_c(FFTComplex *out, FFTComplex *in, FFTComplex *exptab, ptrdiff_t stride) -{ - int k; - FFTComplex tmp1[5], tmp2[5], tmp3[5]; - - fft5(tmp1, in + 0, exptab + 19); - fft5(tmp2, in + 1, exptab + 19); - fft5(tmp3, in + 2, exptab + 19); - - for (k = 0; k < 5; k++) { - FFTComplex t[2]; - - CMUL3(t[0], tmp2[k], exptab[k]); - CMUL3(t[1], tmp3[k], exptab[2 * k]); - out[stride*k].re = tmp1[k].re + t[0].re + t[1].re; - out[stride*k].im = tmp1[k].im + t[0].im + t[1].im; - - CMUL3(t[0], tmp2[k], exptab[k + 5]); - CMUL3(t[1], tmp3[k], exptab[2 * (k + 5)]); - out[stride*(k + 5)].re = tmp1[k].re + t[0].re + t[1].re; - out[stride*(k + 5)].im = tmp1[k].im + t[0].im + t[1].im; - - CMUL3(t[0], tmp2[k], exptab[k + 10]); - CMUL3(t[1], tmp3[k], exptab[2 * k + 5]); - out[stride*(k + 10)].re = tmp1[k].re + t[0].re + t[1].re; - out[stride*(k + 10)].im = tmp1[k].im + t[0].im + t[1].im; - } -} - -static void mdct15(MDCT15Context *s, float *dst, const float *src, ptrdiff_t stride) -{ - int i, j; - const int len4 = s->len4, len3 = len4 * 3, len8 = len4 >> 1; - const int l_ptwo = 1 << s->ptwo_fft.nbits; - FFTComplex fft15in[15]; - - /* Folding and pre-reindexing */ - for (i = 0; i < l_ptwo; i++) { - for (j = 0; j < 15; j++) { - const int k = s->pfa_prereindex[i*15 + j]; - FFTComplex tmp, exp = s->twiddle_exptab[k >> 1]; - if (k < len4) { - tmp.re = -src[ len4 + k] + src[1*len4 - 1 - k]; - tmp.im = -src[ len3 + k] - src[1*len3 - 1 - k]; - } else { - tmp.re = -src[ len4 + k] - src[5*len4 - 1 - k]; - tmp.im = src[-len4 + k] - src[1*len3 - 1 - k]; - } - CMUL(fft15in[j].im, fft15in[j].re, tmp.re, tmp.im, exp.re, exp.im); - } - s->fft15(s->tmp + s->ptwo_fft.revtab[i], fft15in, s->exptab, l_ptwo); - } - - /* Then a 15xN FFT (where N is a power of two) */ - for (i = 0; i < 15; i++) - s->ptwo_fft.fft_calc(&s->ptwo_fft, s->tmp + l_ptwo*i); - - /* Reindex again, apply twiddles and output */ - for (i = 0; i < len8; i++) { - const int i0 = len8 + i, i1 = len8 - i - 1; - const int s0 = s->pfa_postreindex[i0], s1 = s->pfa_postreindex[i1]; - - CMUL(dst[2*i1*stride + stride], dst[2*i0*stride], s->tmp[s0].re, s->tmp[s0].im, - s->twiddle_exptab[i0].im, s->twiddle_exptab[i0].re); - CMUL(dst[2*i0*stride + stride], dst[2*i1*stride], s->tmp[s1].re, s->tmp[s1].im, - s->twiddle_exptab[i1].im, s->twiddle_exptab[i1].re); - } -} - -static void imdct15_half(MDCT15Context *s, float *dst, const float *src, - ptrdiff_t stride) -{ - FFTComplex fft15in[15]; - FFTComplex *z = (FFTComplex *)dst; - int i, j, len8 = s->len4 >> 1, l_ptwo = 1 << s->ptwo_fft.nbits; - const float *in1 = src, *in2 = src + (s->len2 - 1) * stride; - - /* Reindex input, putting it into a buffer and doing an Nx15 FFT */ - for (i = 0; i < l_ptwo; i++) { - for (j = 0; j < 15; j++) { - const int k = s->pfa_prereindex[i*15 + j]; - FFTComplex tmp = { in2[-k*stride], in1[k*stride] }; - CMUL3(fft15in[j], tmp, s->twiddle_exptab[k >> 1]); - } - s->fft15(s->tmp + s->ptwo_fft.revtab[i], fft15in, s->exptab, l_ptwo); - } - - /* Then a 15xN FFT (where N is a power of two) */ - for (i = 0; i < 15; i++) - s->ptwo_fft.fft_calc(&s->ptwo_fft, s->tmp + l_ptwo*i); - - /* Reindex again, apply twiddles and output */ - s->postreindex(z, s->tmp, s->twiddle_exptab, s->pfa_postreindex, len8); -} - -static void postrotate_c(FFTComplex *out, FFTComplex *in, FFTComplex *exp, - int *lut, ptrdiff_t len8) -{ - int i; - - /* Reindex again, apply twiddles and output */ - for (i = 0; i < len8; i++) { - const int i0 = len8 + i, i1 = len8 - i - 1; - const int s0 = lut[i0], s1 = lut[i1]; - - CMUL(out[i1].re, out[i0].im, in[s1].im, in[s1].re, exp[i1].im, exp[i1].re); - CMUL(out[i0].re, out[i1].im, in[s0].im, in[s0].re, exp[i0].im, exp[i0].re); - } -} - -av_cold int ff_mdct15_init(MDCT15Context **ps, int inverse, int N, double scale) -{ - MDCT15Context *s; - double alpha, theta; - int len2 = 15 * (1 << N); - int len = 2 * len2; - int i; - - /* Tested and verified to work on everything in between */ - if ((N < 2) || (N > 13)) - return AVERROR(EINVAL); - - s = av_mallocz(sizeof(*s)); - if (!s) - return AVERROR(ENOMEM); - - s->fft_n = N - 1; - s->len4 = len2 / 2; - s->len2 = len2; - s->inverse = inverse; - s->fft15 = fft15_c; - s->mdct = mdct15; - s->imdct_half = imdct15_half; - s->postreindex = postrotate_c; - - if (ff_fft_init(&s->ptwo_fft, N - 1, s->inverse) < 0) - goto fail; - - if (init_pfa_reindex_tabs(s)) - goto fail; - - s->tmp = av_malloc_array(len, 2 * sizeof(*s->tmp)); - if (!s->tmp) - goto fail; - - s->twiddle_exptab = av_malloc_array(s->len4, sizeof(*s->twiddle_exptab)); - if (!s->twiddle_exptab) - goto fail; - - theta = 0.125f + (scale < 0 ? s->len4 : 0); - scale = sqrt(fabs(scale)); - for (i = 0; i < s->len4; i++) { - alpha = 2 * M_PI * (i + theta) / len; - s->twiddle_exptab[i].re = cosf(alpha) * scale; - s->twiddle_exptab[i].im = sinf(alpha) * scale; - } - - /* 15-point FFT exptab */ - for (i = 0; i < 19; i++) { - if (i < 15) { - double theta = (2.0f * M_PI * i) / 15.0f; - if (!s->inverse) - theta *= -1; - s->exptab[i].re = cosf(theta); - s->exptab[i].im = sinf(theta); - } else { /* Wrap around to simplify fft15 */ - s->exptab[i] = s->exptab[i - 15]; - } - } - - /* 5-point FFT exptab */ - s->exptab[19].re = cosf(2.0f * M_PI / 5.0f); - s->exptab[19].im = sinf(2.0f * M_PI / 5.0f); - s->exptab[20].re = cosf(1.0f * M_PI / 5.0f); - s->exptab[20].im = sinf(1.0f * M_PI / 5.0f); - - /* Invert the phase for an inverse transform, do nothing for a forward transform */ - if (s->inverse) { - s->exptab[19].im *= -1; - s->exptab[20].im *= -1; - } - -#if ARCH_X86 - ff_mdct15_init_x86(s); -#endif - - *ps = s; - - return 0; - -fail: - ff_mdct15_uninit(&s); - return AVERROR(ENOMEM); -} diff --git a/libavcodec/mdct15.h b/libavcodec/mdct15.h deleted file mode 100644 index 035246ff97..0000000000 --- a/libavcodec/mdct15.h +++ /dev/null @@ -1,64 +0,0 @@ -/* - * Copyright (c) 2017 Rostislav Pehlivanov - * - * 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 - */ - -#ifndef AVCODEC_MDCT15_H -#define AVCODEC_MDCT15_H - -#include - -#include "libavutil/mem_internal.h" - -#include "fft.h" - -typedef struct MDCT15Context { - int fft_n; - int len2; - int len4; - int inverse; - int *pfa_prereindex; - int *pfa_postreindex; - - FFTContext ptwo_fft; - FFTComplex *tmp; - FFTComplex *twiddle_exptab; - - DECLARE_ALIGNED(32, FFTComplex, exptab)[64]; - - /* 15-point FFT */ - void (*fft15)(FFTComplex *out, FFTComplex *in, FFTComplex *exptab, ptrdiff_t stride); - - /* PFA postrotate and exptab */ - void (*postreindex)(FFTComplex *out, FFTComplex *in, FFTComplex *exp, int *lut, ptrdiff_t len8); - - /* Calculate a full 2N -> N MDCT */ - void (*mdct)(struct MDCT15Context *s, float *dst, const float *src, ptrdiff_t stride); - - /* Calculate the middle half of the iMDCT */ - void (*imdct_half)(struct MDCT15Context *s, float *dst, const float *src, - ptrdiff_t stride); -} MDCT15Context; - -/* Init an (i)MDCT of the length 2 * 15 * (2^N) */ -int ff_mdct15_init(MDCT15Context **ps, int inverse, int N, double scale); -void ff_mdct15_uninit(MDCT15Context **ps); - -void ff_mdct15_init_x86(MDCT15Context *s); - -#endif /* AVCODEC_MDCT15_H */ diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile index ec6adcd8b0..6f62878081 100644 --- a/libavcodec/x86/Makefile +++ b/libavcodec/x86/Makefile @@ -24,7 +24,6 @@ OBJS-$(CONFIG_HUFFYUVDSP) += x86/huffyuvdsp_init.o OBJS-$(CONFIG_HUFFYUVENCDSP) += x86/huffyuvencdsp_init.o OBJS-$(CONFIG_IDCTDSP) += x86/idctdsp_init.o OBJS-$(CONFIG_LPC) += x86/lpc_init.o -OBJS-$(CONFIG_MDCT15) += x86/mdct15_init.o OBJS-$(CONFIG_ME_CMP) += x86/me_cmp_init.o OBJS-$(CONFIG_MPEGAUDIODSP) += x86/mpegaudiodsp.o OBJS-$(CONFIG_MPEGVIDEO) += x86/mpegvideo.o @@ -127,7 +126,6 @@ X86ASM-OBJS-$(CONFIG_LLAUDDSP) += x86/lossless_audiodsp.o X86ASM-OBJS-$(CONFIG_LLVIDDSP) += x86/lossless_videodsp.o X86ASM-OBJS-$(CONFIG_LLVIDENCDSP) += x86/lossless_videoencdsp.o X86ASM-OBJS-$(CONFIG_LPC) += x86/lpc.o -X86ASM-OBJS-$(CONFIG_MDCT15) += x86/mdct15.o X86ASM-OBJS-$(CONFIG_ME_CMP) += x86/me_cmp.o X86ASM-OBJS-$(CONFIG_MPEGAUDIODSP) += x86/imdct36.o X86ASM-OBJS-$(CONFIG_MPEGVIDEOENC) += x86/mpegvideoencdsp.o diff --git a/libavcodec/x86/mdct15.asm b/libavcodec/x86/mdct15.asm deleted file mode 100644 index 2a2cdbd21b..0000000000 --- a/libavcodec/x86/mdct15.asm +++ /dev/null @@ -1,221 +0,0 @@ -;****************************************************************************** -;* SIMD optimized non-power-of-two MDCT functions -;* -;* Copyright (C) 2017 Rostislav Pehlivanov -;* -;* 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 "libavutil/x86/x86util.asm" - -SECTION_RODATA 32 - -perm_neg: dd 2, 5, 3, 4, 6, 1, 7, 0 -perm_pos: dd 0, 7, 1, 6, 4, 3, 5, 2 -sign_adjust_r: times 4 dd 0x80000000, 0x00000000 - -sign_adjust_5: dd 0x00000000, 0x80000000, 0x80000000, 0x00000000 - -SECTION .text - -%if ARCH_X86_64 - -;***************************************************************************************** -;void ff_fft15_avx(FFTComplex *out, FFTComplex *in, FFTComplex *exptab, ptrdiff_t stride); -;***************************************************************************************** -%macro FFT5 3 ; %1 - in_offset, %2 - dst1 (64bit used), %3 - dst2 - VBROADCASTSD m0, [inq + %1] ; in[ 0].re, in[ 0].im, in[ 0].re, in[ 0].im - movsd xm1, [inq + 1*16 + 8 + %1] ; in[ 3].re, in[ 3].im, 0, 0 - movsd xm4, [inq + 6*16 + 0 + %1] ; in[12].re, in[12].im, 0, 0 - movhps xm1, [inq + 3*16 + 0 + %1] ; in[ 3].re, in[ 3].im, in[ 6].re, in[ 6].im - movhps xm4, [inq + 4*16 + 8 + %1] ; in[12].re, in[12].im, in[ 9].re, in[ 9].im - - subps xm2, xm1, xm4 ; t[2].im, t[2].re, t[3].im, t[3].re - addps xm1, xm4 ; t[0].re, t[0].im, t[1].re, t[1].im - - movhlps %2, xm1 ; t[0].re, t[1].re, t[0].im, t[1].im - addps %2, xm1 - addps %2, xm0 ; DC[0].re, DC[0].im, junk... - movlhps %2, %2 ; DC[0].re, DC[0].im, DC[0].re, DC[0].im - - shufps xm3, xm1, xm2, q0110 ; t[0].re, t[0].im, t[2].re, t[2].im - shufps xm1, xm2, q2332 ; t[1].re, t[1].im, t[3].re, t[3].im - - mulps xm%3, xm1, xm5 - mulps xm4, xm3, xm6 - mulps xm1, xm6 - - xorps xm1, xm7 - mulps xm3, xm5 - addsubps xm3, xm1 ; t[0].re, t[0].im, t[2].re, t[2].im - subps xm%3, xm4 ; t[4].re, t[4].im, t[5].re, t[5].im - - movhlps xm2, xm%3, xm3 ; t[2].re, t[2].im, t[5].re, t[5].im - movlhps xm3, xm%3 ; t[0].re, t[0].im, t[4].re, t[4].im - - xorps xm2, xm7 - addps xm%3, xm2, xm3 - subps xm3, xm2 - - shufps xm3, xm3, q1032 - vinsertf128 m%3, m%3, xm3, 1 ; All ACs (tmp[1] through to tmp[4]) - addps m%3, m%3, m0 ; Finally offset with DCs -%endmacro - -%macro BUTTERFLIES_DC 1 ; %1 - exptab_offset - mulps xm0, xm9, [exptabq + %1 + 16*0] - mulps xm1, xm10, [exptabq + %1 + 16*1] - - haddps xm0, xm1 - movhlps xm1, xm0 ; t[0].re, t[1].re, t[0].im, t[1].im - - addps xm0, xm1 - addps xm0, xm8 - - movsd [outq], xm0 -%endmacro - -%macro BUTTERFLIES_AC 1 ; %1 - exptab_offset - mulps m0, m12, [exptabq + 64*0 + 0*mmsize + %1] - mulps m1, m12, [exptabq + 64*0 + 1*mmsize + %1] - mulps m2, m13, [exptabq + 64*1 + 0*mmsize + %1] - mulps m3, m13, [exptabq + 64*1 + 1*mmsize + %1] - - addps m0, m0, m2 - addps m1, m1, m3 - addps m0, m0, m11 - - shufps m1, m1, m1, q2301 - addps m0, m0, m1 - - vextractf128 xm1, m0, 1 - - movlps [outq + strideq*1], xm0 - movhps [outq + strideq*2], xm0 - movlps [outq + stride3q], xm1 - movhps [outq + strideq*4], xm1 -%endmacro - -INIT_YMM avx -cglobal fft15, 4, 5, 14, out, in, exptab, stride, stride5 - shl strideq, 3 - - movaps xm5, [exptabq + 480 + 16*0] - movaps xm6, [exptabq + 480 + 16*1] - movaps xm7, [sign_adjust_5] - - FFT5 0, xm8, 11 - FFT5 8, xm9, 12 - FFT5 16, xm10, 13 - -%define stride3q inq - lea stride3q, [strideq + strideq*2] - lea stride5q, [strideq + strideq*4] - - BUTTERFLIES_DC (8*6 + 4*0)*2*4 - BUTTERFLIES_AC (8*0 + 0*0)*2*4 - - add outq, stride5q - BUTTERFLIES_DC (8*6 + 4*1)*2*4 - BUTTERFLIES_AC (8*2 + 0*0)*2*4 - - add outq, stride5q - BUTTERFLIES_DC (8*6 + 4*2)*2*4 - BUTTERFLIES_AC (8*4 + 0*0)*2*4 - - RET - -%endif ; ARCH_X86_64 - -;******************************************************************************************************* -;void ff_mdct15_postreindex(FFTComplex *out, FFTComplex *in, FFTComplex *exp, int *lut, ptrdiff_t len8); -;******************************************************************************************************* -%macro LUT_LOAD_4D 3 - mov r4d, [lutq + %3q*4 + 0] - movsd xmm%1, [inq + r4q*8] - mov r4d, [lutq + %3q*4 + 4] - movhps xmm%1, [inq + r4q*8] -%if cpuflag(avx2) - mov r4d, [lutq + %3q*4 + 8] - movsd %2, [inq + r4q*8] - mov r4d, [lutq + %3q*4 + 12] - movhps %2, [inq + r4q*8] - vinsertf128 %1, %1, %2, 1 -%endif -%endmacro - -%macro POSTROTATE_FN 1 -cglobal mdct15_postreindex, 5, 7, 8 + cpuflag(avx2)*2, out, in, exp, lut, len8, offset_p, offset_n - - xor offset_nq, offset_nq - lea offset_pq, [len8q*2 - %1] - - movaps m7, [sign_adjust_r] - -%if cpuflag(avx2) - movaps m8, [perm_pos] - movaps m9, [perm_neg] -%endif - -.loop: - movups m0, [expq + offset_pq*8] ; exp[p0].re, exp[p0].im, exp[p1].re, exp[p1].im, exp[p2].re, exp[p2].im, exp[p3].re, exp[p3].im - movups m1, [expq + offset_nq*8] ; exp[n3].re, exp[n3].im, exp[n2].re, exp[n2].im, exp[n1].re, exp[n1].im, exp[n0].re, exp[n0].im - - LUT_LOAD_4D m3, xm4, offset_p ; in[p0].re, in[p0].im, in[p1].re, in[p1].im, in[p2].re, in[p2].im, in[p3].re, in[p3].im - LUT_LOAD_4D m4, xm5, offset_n ; in[n3].re, in[n3].im, in[n2].re, in[n2].im, in[n1].re, in[n1].im, in[n0].re, in[n0].im - - mulps m5, m3, m0 ; in[p].reim * exp[p].reim - mulps m6, m4, m1 ; in[n].reim * exp[n].reim - - xorps m5, m7 ; in[p].re *= -1, in[p].im *= 1 - xorps m6, m7 ; in[n].re *= -1, in[n].im *= 1 - - shufps m3, m3, m3, q2301 ; in[p].imre - shufps m4, m4, m4, q2301 ; in[n].imre - - mulps m3, m0 ; in[p].imre * exp[p].reim - mulps m4, m1 ; in[n].imre * exp[n].reim - - haddps m3, m6 ; out[n0].im, out[n1].im, out[n3].re, out[n2].re, out[n2].im, out[n3].im, out[n1].re, out[n0].re - haddps m5, m4 ; out[p0].re, out[p1].re, out[p3].im, out[p2].im, out[p2].re, out[p3].re, out[p1].im, out[p0].im - -%if cpuflag(avx2) - vpermps m3, m9, m3 ; out[n3].im, out[n3].re, out[n2].im, out[n2].re, out[n1].im, out[n1].re, out[n0].im, out[n0].re - vpermps m5, m8, m5 ; out[p0].re, out[p0].im, out[p1].re, out[p1].im, out[p2].re, out[p2].im, out[p3].re, out[p3].im -%else - shufps m3, m3, m3, q0312 - shufps m5, m5, m5, q2130 -%endif - - movups [outq + offset_nq*8], m3 - movups [outq + offset_pq*8], m5 - - sub offset_pq, %1 - add offset_nq, %1 - cmp offset_nq, offset_pq - jle .loop - - REP_RET -%endmacro - -INIT_XMM sse3 -POSTROTATE_FN 2 - -%if ARCH_X86_64 && HAVE_AVX2_EXTERNAL -INIT_YMM avx2 -POSTROTATE_FN 4 -%endif diff --git a/libavcodec/x86/mdct15_init.c b/libavcodec/x86/mdct15_init.c deleted file mode 100644 index ea7301e1e7..0000000000 --- a/libavcodec/x86/mdct15_init.c +++ /dev/null @@ -1,104 +0,0 @@ -/* - * SIMD optimized non-power-of-two MDCT functions - * - * Copyright (C) 2017 Rostislav Pehlivanov - * - * 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 - -#include "config.h" - -#include "libavutil/attributes.h" -#include "libavutil/x86/cpu.h" -#include "libavcodec/mdct15.h" - -void ff_mdct15_postreindex_sse3(FFTComplex *out, FFTComplex *in, FFTComplex *exp, int *lut, ptrdiff_t len8); -void ff_mdct15_postreindex_avx2(FFTComplex *out, FFTComplex *in, FFTComplex *exp, int *lut, ptrdiff_t len8); - -void ff_fft15_avx(FFTComplex *out, FFTComplex *in, FFTComplex *exptab, ptrdiff_t stride); - -static void perm_twiddles(MDCT15Context *s) -{ - int k; - FFTComplex tmp[30]; - - /* 5-point FFT twiddles */ - s->exptab[60].re = s->exptab[60].im = s->exptab[19].re; - s->exptab[61].re = s->exptab[61].im = s->exptab[19].im; - s->exptab[62].re = s->exptab[62].im = s->exptab[20].re; - s->exptab[63].re = s->exptab[63].im = s->exptab[20].im; - - /* 15-point FFT twiddles */ - for (k = 0; k < 5; k++) { - tmp[6*k + 0] = s->exptab[k + 0]; - tmp[6*k + 2] = s->exptab[k + 5]; - tmp[6*k + 4] = s->exptab[k + 10]; - - tmp[6*k + 1] = s->exptab[2 * (k + 0)]; - tmp[6*k + 3] = s->exptab[2 * (k + 5)]; - tmp[6*k + 5] = s->exptab[2 * k + 5 ]; - } - - for (k = 0; k < 6; k++) { - FFTComplex ac_exp[] = { - { tmp[6*1 + k].re, tmp[6*1 + k].re }, - { tmp[6*2 + k].re, tmp[6*2 + k].re }, - { tmp[6*3 + k].re, tmp[6*3 + k].re }, - { tmp[6*4 + k].re, tmp[6*4 + k].re }, - { tmp[6*1 + k].im, -tmp[6*1 + k].im }, - { tmp[6*2 + k].im, -tmp[6*2 + k].im }, - { tmp[6*3 + k].im, -tmp[6*3 + k].im }, - { tmp[6*4 + k].im, -tmp[6*4 + k].im }, - }; - memcpy(s->exptab + 8*k, ac_exp, 8*sizeof(FFTComplex)); - } - - /* Specialcase when k = 0 */ - for (k = 0; k < 3; k++) { - FFTComplex dc_exp[] = { - { tmp[2*k + 0].re, -tmp[2*k + 0].im }, - { tmp[2*k + 0].im, tmp[2*k + 0].re }, - { tmp[2*k + 1].re, -tmp[2*k + 1].im }, - { tmp[2*k + 1].im, tmp[2*k + 1].re }, - }; - memcpy(s->exptab + 8*6 + 4*k, dc_exp, 4*sizeof(FFTComplex)); - } -} - -av_cold void ff_mdct15_init_x86(MDCT15Context *s) -{ - int adjust_twiddles = 0; - int cpu_flags = av_get_cpu_flags(); - - if (EXTERNAL_SSE3(cpu_flags)) - s->postreindex = ff_mdct15_postreindex_sse3; - -#if ARCH_X86_64 - if (EXTERNAL_AVX(cpu_flags)) { - s->fft15 = ff_fft15_avx; - adjust_twiddles = 1; - } - - if (EXTERNAL_AVX2_FAST(cpu_flags)) - s->postreindex = ff_mdct15_postreindex_avx2; -#endif - - if (adjust_twiddles) - perm_twiddles(s); -}