diff options
Diffstat (limited to 'libavcodec/x86')
-rw-r--r-- | libavcodec/x86/Makefile | 2 | ||||
-rw-r--r-- | libavcodec/x86/mdct15.asm | 141 | ||||
-rw-r--r-- | libavcodec/x86/mdct15_init.c | 95 |
3 files changed, 238 insertions, 0 deletions
diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile index e65118d134..b86700b675 100644 --- a/libavcodec/x86/Makefile +++ b/libavcodec/x86/Makefile @@ -25,6 +25,7 @@ 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.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 \ @@ -117,6 +118,7 @@ X86ASM-OBJS-$(CONFIG_IDCTDSP) += x86/idctdsp.o X86ASM-OBJS-$(CONFIG_LLAUDDSP) += x86/lossless_audiodsp.o X86ASM-OBJS-$(CONFIG_LLVIDDSP) += x86/lossless_videodsp.o X86ASM-OBJS-$(CONFIG_LLVIDENCDSP) += x86/lossless_videoencdsp.o +X86ASM-OBJS-$(CONFIG_MDCT15) += x86/mdct15.o X86ASM-OBJS-$(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 new file mode 100644 index 0000000000..8d01675956 --- /dev/null +++ b/libavcodec/x86/mdct15.asm @@ -0,0 +1,141 @@ +;****************************************************************************** +;* SIMD optimized non-power-of-two MDCT functions +;* +;* Copyright (C) 2017 Rostislav Pehlivanov <atomnuker@gmail.com> +;* +;* 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" + +%if ARCH_X86_64 + +SECTION_RODATA + +sign_adjust_5: dd 0x00000000, 0x80000000, 0x80000000, 0x00000000 + +SECTION .text + +%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, m0 ; Finally offset with DCs +%endmacro + +%macro BUTTERFLIES_DC 2 ; %1 - exptab_offset, %2 - out + 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 [%2q], xm0 +%endmacro + +%macro BUTTERFLIES_AC 2 ; exptab, exptab_offset, src1, src2, src3, out (uses m0-m3) + 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, m2 + addps m1, m3 + addps m0, m11 + + shufps m1, m1, q2301 + addps m0, m1 + + vextractf128 xm1, m0, 1 + + movlps [%2q + strideq*1], xm0 + movhps [%2q + strideq*2], xm0 + movlps [%2q + stride3q], xm1 + movhps [%2q + strideq*4], xm1 +%endmacro + +;***************************************************************************************** +;void ff_fft15_avx(FFTComplex *out, FFTComplex *in, FFTComplex *exptab, ptrdiff_t stride); +;***************************************************************************************** +INIT_YMM avx +cglobal fft15, 4, 6, 14, out, in, exptab, stride, stride3, stride5 +%define out0q inq + 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 + + lea stride3q, [strideq + strideq*2] + lea stride5q, [strideq + strideq*4] + + mov out0q, outq + + BUTTERFLIES_DC (8*6 + 4*0)*2*4, out0 + lea outq, [out0q + stride5q*1] + BUTTERFLIES_DC (8*6 + 4*1)*2*4, out + lea outq, [out0q + stride5q*2] + BUTTERFLIES_DC (8*6 + 4*2)*2*4, out + + BUTTERFLIES_AC (8*0)*2*4, out0 + lea outq, [out0q + stride5q*1] + BUTTERFLIES_AC (8*2)*2*4, out + lea outq, [out0q + stride5q*2] + BUTTERFLIES_AC (8*4)*2*4, out + + RET + +%endif diff --git a/libavcodec/x86/mdct15_init.c b/libavcodec/x86/mdct15_init.c new file mode 100644 index 0000000000..ba3d94c2ec --- /dev/null +++ b/libavcodec/x86/mdct15_init.c @@ -0,0 +1,95 @@ +/* + * SIMD optimized non-power-of-two MDCT functions + * + * Copyright (C) 2017 Rostislav Pehlivanov <atomnuker@gmail.com> + * + * 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/x86/cpu.h" +#include "libavcodec/mdct15.h" + +void ff_fft15_avx(FFTComplex *out, FFTComplex *in, FFTComplex *exptab, ptrdiff_t stride); + +static void perm_twiddles(MDCT15Context *s) +{ + int k; + FFTComplex exp_5point[4]; + + FFTComplex tmp[21], tmp2[30]; + memcpy(tmp, s->exptab, sizeof(FFTComplex)*21); + + /* 15-point FFT twiddles */ + for (k = 0; k < 5; k++) { + tmp2[6*k + 0] = tmp[k + 0]; + tmp2[6*k + 2] = tmp[k + 5]; + tmp2[6*k + 4] = tmp[k + 10]; + + tmp2[6*k + 1] = tmp[2 * (k + 0)]; + tmp2[6*k + 3] = tmp[2 * (k + 5)]; + tmp2[6*k + 5] = tmp[2 * k + 5 ]; + } + + for (k = 0; k < 6; k++) { + FFTComplex ac_exp[] = { + { tmp2[6*1 + k].re, tmp2[6*1 + k].re }, + { tmp2[6*2 + k].re, tmp2[6*2 + k].re }, + { tmp2[6*3 + k].re, tmp2[6*3 + k].re }, + { tmp2[6*4 + k].re, tmp2[6*4 + k].re }, + { tmp2[6*1 + k].im, -tmp2[6*1 + k].im }, + { tmp2[6*2 + k].im, -tmp2[6*2 + k].im }, + { tmp2[6*3 + k].im, -tmp2[6*3 + k].im }, + { tmp2[6*4 + k].im, -tmp2[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[] = { + { tmp2[2*k + 0].re, -tmp2[2*k + 0].im }, + { tmp2[2*k + 0].im, tmp2[2*k + 0].re }, + { tmp2[2*k + 1].re, -tmp2[2*k + 1].im }, + { tmp2[2*k + 1].im, tmp2[2*k + 1].re }, + }; + memcpy(s->exptab + 8*6 + 4*k, dc_exp, 4*sizeof(FFTComplex)); + } + + /* 5-point FFT twiddles */ + exp_5point[0].re = exp_5point[0].im = tmp[19].re; + exp_5point[1].re = exp_5point[1].im = tmp[19].im; + exp_5point[2].re = exp_5point[2].im = tmp[20].re; + exp_5point[3].re = exp_5point[3].im = tmp[20].im; + + memcpy(s->exptab + 8*6 + 4*3, exp_5point, 4*sizeof(FFTComplex)); +} + +av_cold void ff_mdct15_init_x86(MDCT15Context *s) +{ + int adjust_twiddles = 0; + int cpu_flags = av_get_cpu_flags(); + + if (ARCH_X86_64 && EXTERNAL_AVX(cpu_flags)) { + s->fft15 = ff_fft15_avx; + adjust_twiddles = 1; + } + + if (adjust_twiddles) + perm_twiddles(s); +} |