From patchwork Thu Jun 22 22:44:35 2017 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Rostislav Pehlivanov X-Patchwork-Id: 4084 Delivered-To: ffmpegpatchwork@gmail.com Received: by 10.103.22.4 with SMTP id 4csp279036vsw; Thu, 22 Jun 2017 15:52:50 -0700 (PDT) X-Received: by 10.223.176.164 with SMTP id i33mr3371148wra.165.1498171970694; Thu, 22 Jun 2017 15:52:50 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1498171970; cv=none; d=google.com; s=arc-20160816; b=zY2EdzE7+wOYICeXSuf3+XrZOPMhtFlT5Rfm2G/qIHAMTBCitXyHcQnLHIe2CLiGVd z3jI4sJvsW3aj9NUE6YmKF6NblcYhYmnqzxX+YXUlChZA/3slyMpwYZQe6osXZAGVXn8 J6YnbdhJr/A3oxZLnjN8QAecSF51Uxp+xD5CVnYWaZmvST+ONlmEU9R4R5q192iP+M5R lbjKuVFaY7SI9IiZj0lhvEeLabDR1BvJVLuu1Mu/GxZlimQbqROxTy6Z6k/MeXbkOXo/ 8bl3g5MLievkjLB8F1EJJE2zzG1wfe9v4siCFUB8e6B0A3l33yhGDyGJZQLbPCSj3l84 HAhQ== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:mime-version:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:references:in-reply-to:message-id:date :to:from:dkim-signature:delivered-to:arc-authentication-results; bh=QWVUwSYIFkmD8a9NAq7BOy4h67xj5GJOK0r5Vk6NOPA=; b=RaLdWh4FSu7s3zy0/ccdDg5BRceUXM2Tfwj6XCvUtdTVWncFpESS4V5J2fI6e8ZMYi aombL/pGZ6AEHwgrThf/z95hLDV0ycnwv3cdNiKOw9gPa5dZuZk+nOsVbIPx0RtqcNNs qONcpGKhGzWaEpogfywelJYkGzEDfoFrkLiV6aUmnOl5RzCiLLMise5QWGdCKeOQfgBM h6ZU9qiU6P7nCNme2sXJh4B4qBYuLELnvGyTCOlnCJS9VKbKPoqnsuCzUEXQIxYYd7wY JW8JHFrofHFZ8f0U8ux/BNWc5IQk/dF1S/QPZgJYEThxIHlxjtH1AIMkJbg95wHD1x9c v2dg== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@gmail.com header.b=QAO06Avd; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org; dmarc=fail (p=NONE sp=NONE dis=NONE) header.from=gmail.com Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id f126si2349160wmf.177.2017.06.22.15.52.50; Thu, 22 Jun 2017 15:52:50 -0700 (PDT) Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; Authentication-Results: mx.google.com; dkim=neutral (body hash did not verify) header.i=@gmail.com header.b=QAO06Avd; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org; dmarc=fail (p=NONE sp=NONE dis=NONE) header.from=gmail.com Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 5A86D68A422; Fri, 23 Jun 2017 01:52:47 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mail-wr0-f193.google.com (mail-wr0-f193.google.com [209.85.128.193]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 0F1F168A0CA for ; Fri, 23 Jun 2017 01:52:41 +0300 (EEST) Received: by mail-wr0-f193.google.com with SMTP id x23so8201978wrb.0 for ; Thu, 22 Jun 2017 15:52:41 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20161025; h=from:to:cc:subject:date:message-id:in-reply-to:references; bh=9kBqoIbzYBkwpJmLuHoOiTX3z57dNe29ZnXpo3AbnKw=; b=QAO06AvdsshWUYP0R2wfBVy7bBGdK4YYH5wbE2pBdOo8OgWaR4SSfctHv39q6vu932 HPk4T2pmLrigjWFcidyFzJq/q9qHQMR1YrjTcOVbw2BiRsDGguZbS4HFb9JkMjawDpk/ 91TUe5lvKcBE8Tmci6pXObO5FjcbPVDirur7pQH2aZ4QJiyswhord7si8x1uva8B3n8A 1O8ucHvPT8U8ZkAlcIsfjaiF1R9YltqYKpd8rewUB7GSldnEkDMwFDbQOJzGwLUch1Q1 rwDBn2ughLbT0Cjy/3hJxqzI3+hBH3J7qStG8y67rJFxc/99z24UrpaElAyIORXRleCO kPNg== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references; bh=9kBqoIbzYBkwpJmLuHoOiTX3z57dNe29ZnXpo3AbnKw=; b=s4mKKO1Ls1UoB8rtxCZGTis8cvcvXlwSivxosa8/JcMEbBGuxXB2PjbutGRlpjw1nC S5btCHGcYtYZkpTz0S1mXgGmnnwuB/YqKfiRw4Pd1w2pMfF5ObFL6fB/gQhHp3D+7FLv LOhMxRbGieEZMycolxxeO9n73In5gPG0Fej+tHYy4K9kVWhOMlWH4CrfznnbkCXz/mGi eA+70oo0Ax5UEnNBRX8qbtSp7k15OCtcRT1vd+ke+0tExb1MhLLLmEBLeX3KSf9lONrg 4tkfBscP6eRrfnMS+a1LNMrEFp2mAOfY3GHg32ZezXr2HnDhQC5U6i8gPDCUlLO2hUvU Qidw== X-Gm-Message-State: AKS2vOxI+KR99nml4lBNeUGW/YwktAn5RrXjOKfpnZU0G5ek9NEL5eea Ustt1PYGU++hALfYnG8= X-Received: by 10.28.113.85 with SMTP id m82mr3297462wmc.21.1498171478884; Thu, 22 Jun 2017 15:44:38 -0700 (PDT) Received: from moonbase.lan ([2a00:23c4:7c80:f00:f121:7da9:4a59:e523]) by smtp.gmail.com with ESMTPSA id k12sm3846075wrc.10.2017.06.22.15.44.37 (version=TLS1_2 cipher=ECDHE-RSA-AES128-GCM-SHA256 bits=128/128); Thu, 22 Jun 2017 15:44:38 -0700 (PDT) From: Rostislav Pehlivanov To: ffmpeg-devel@ffmpeg.org Date: Thu, 22 Jun 2017 23:44:35 +0100 Message-Id: <20170622224435.9684-1-atomnuker@gmail.com> X-Mailer: git-send-email 2.13.1.704.gde00cce3c In-Reply-To: <016add2a-330b-dd43-4d2c-cd2a42cf6d6f@gmail.com> References: <016add2a-330b-dd43-4d2c-cd2a42cf6d6f@gmail.com> Subject: [FFmpeg-devel] [PATCH v3] mdct15: add assembly optimizations for the 15-point FFT X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.20 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: Rostislav Pehlivanov MIME-Version: 1.0 Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" c: 1802 decicycles in fft15,16774635 runs, 2581 skips fma3: 935 decicycles in fft15,16775893 runs, 1323 skips Signed-off-by: Rostislav Pehlivanov --- libavcodec/mdct15.c | 182 +++++++++++++++++++++---------------------- libavcodec/mdct15.h | 26 +++---- libavcodec/x86/Makefile | 2 + libavcodec/x86/mdct15.asm | 146 ++++++++++++++++++++++++++++++++++ libavcodec/x86/mdct15_init.c | 95 ++++++++++++++++++++++ 5 files changed, 342 insertions(+), 109 deletions(-) create mode 100644 libavcodec/x86/mdct15.asm create mode 100644 libavcodec/x86/mdct15_init.c diff --git a/libavcodec/mdct15.c b/libavcodec/mdct15.c index 8c42ece483..f93881fbed 100644 --- a/libavcodec/mdct15.c +++ b/libavcodec/mdct15.c @@ -57,11 +57,6 @@ av_cold void ff_mdct15_uninit(MDCT15Context **ps) av_freep(ps); } -static void mdct15(MDCT15Context *s, float *dst, const float *src, ptrdiff_t stride); - -static void imdct15_half(MDCT15Context *s, float *dst, const float *src, - ptrdiff_t stride, float scale); - static inline int init_pfa_reindex_tabs(MDCT15Context *s) { int i, j; @@ -93,88 +88,8 @@ static inline int init_pfa_reindex_tabs(MDCT15Context *s) return 0; } -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->mdct = mdct15; - s->imdct_half = imdct15_half; - - 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 = cos(alpha) * scale; - s->twiddle_exptab[i].im = sin(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 = cos(theta); - s->exptab[i].im = sin(theta); - } else { /* Wrap around to simplify fft15 */ - s->exptab[i] = s->exptab[i - 15]; - } - } - - /* 5-point FFT exptab */ - s->exptab[19].re = cos(2.0f * M_PI / 5.0f); - s->exptab[19].im = sin(2.0f * M_PI / 5.0f); - s->exptab[20].re = cos(1.0f * M_PI / 5.0f); - s->exptab[20].im = sin(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; - } - - *ps = s; - - return 0; - -fail: - ff_mdct15_uninit(&s); - return AVERROR(ENOMEM); -} - /* Stride is hardcoded to 3 */ -static inline void fft5(const FFTComplex exptab[2], FFTComplex *out, - const FFTComplex *in) +static inline void fft5(FFTComplex *out, FFTComplex *in, FFTComplex exptab[2]) { FFTComplex z0[4], t[6]; @@ -219,14 +134,14 @@ static inline void fft5(const FFTComplex exptab[2], FFTComplex *out, out[4].im = in[0].im + z0[3].im; } -static void fft15(const FFTComplex exptab[22], FFTComplex *out, const FFTComplex *in, size_t stride) +static void fft15_c(FFTComplex *out, FFTComplex *in, FFTComplex *exptab, ptrdiff_t stride) { int k; FFTComplex tmp1[5], tmp2[5], tmp3[5]; - fft5(exptab + 19, tmp1, in + 0); - fft5(exptab + 19, tmp2, in + 1); - fft5(exptab + 19, tmp3, in + 2); + 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]; @@ -269,7 +184,7 @@ static void mdct15(MDCT15Context *s, float *dst, const float *src, ptrdiff_t str } CMUL(fft15in[j].re, fft15in[j].im, re, im, s->twiddle_exptab[k].re, -s->twiddle_exptab[k].im); } - fft15(s->exptab, s->tmp + s->ptwo_fft.revtab[i], fft15in, l_ptwo); + 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) */ @@ -306,7 +221,7 @@ static void imdct15_half(MDCT15Context *s, float *dst, const float *src, FFTComplex tmp = { *(in2 - 2*k*stride), *(in1 + 2*k*stride) }; CMUL3(fft15in[j], tmp, s->twiddle_exptab[k]); } - fft15(s->exptab, s->tmp + s->ptwo_fft.revtab[i], fft15in, l_ptwo); + 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) */ @@ -327,3 +242,86 @@ static void imdct15_half(MDCT15Context *s, float *dst, const float *src, z[i0].im = scale * im1; } } + +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; + + 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); + + *ps = s; + + return 0; + +fail: + ff_mdct15_uninit(&s); + return AVERROR(ENOMEM); +} diff --git a/libavcodec/mdct15.h b/libavcodec/mdct15.h index ef94edff6c..7d83f3ebdf 100644 --- a/libavcodec/mdct15.h +++ b/libavcodec/mdct15.h @@ -34,34 +34,26 @@ typedef struct MDCT15Context { int *pfa_postreindex; FFTContext ptwo_fft; - FFTComplex *tmp; - FFTComplex *twiddle_exptab; - /* 0 - 18: fft15 twiddles, 19 - 20: fft5 twiddles */ - FFTComplex exptab[21]; + DECLARE_ALIGNED(32, FFTComplex, exptab)[64]; - /** - * Calculate a full 2N -> N MDCT - */ + /* 15-point FFT */ + void (*fft15)(FFTComplex *out, FFTComplex *in, FFTComplex *exptab, ptrdiff_t stride); + + /* 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 - */ + /* Calculate the middle half of the iMDCT */ void (*imdct_half)(struct MDCT15Context *s, float *dst, const float *src, ptrdiff_t src_stride, float scale); } MDCT15Context; -/** - * Init an (i)MDCT of the length 2 * 15 * (2^N) - */ +/* Init an (i)MDCT of the length 2 * 15 * (2^N) */ int ff_mdct15_init(MDCT15Context **ps, int inverse, int N, double scale); - -/** - * Frees a context - */ 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 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..ec167fbaec --- /dev/null +++ b/libavcodec/x86/mdct15.asm @@ -0,0 +1,146 @@ +;****************************************************************************** +;* 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 + +sign_adjust_5: dd 0x00000000, 0x80000000, 0x80000000, 0x00000000 + +SECTION .text + +%macro FFT5 3 ; %1 - in_offset, %2 - dst1 (64bit used), %3 - dst2 + movddup xm0, [inq + 0*16 + 0 + %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 xm2, [inq + 2*16 + 16 + %1] ; in[ 5].re, in[ 5].im, in[ 6].re, in[ 6].im + movsd xm3, [inq + 4*16 + 8 + %1] ; in[ 8].re, in[ 8].im, in[ 9].re, in[ 9].im + movsd xm4, [inq + 6*16 + 0 + %1] ; in[12].re, in[12].im, 0, 0 + + vinsertf128 m0, xm0, 1 + + shufps xm1, xm2, q1010 ; in[ 3].re, in[ 3].im, in[ 6].re, in[ 6].im + shufps xm4, xm3, q1010 ; 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, 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 + movaps m0, [exptabq + %1] + vextractf128 xm1, m0, 1 + + mulps xm1, xm10 + mulps xm0, xm9 + + 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] + + shufps m1, m1, q2301 + shufps m3, m3, q2301 + + addps m0, m1 + addps m2, m3 + addps m0, m2 + addps m0, m11 + + 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 m5, [exptabq + 480] + vextractf128 xm6, m5, 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, [stride3q + strideq*2] + + 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 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 + * + * 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); +}