yading@10: /* yading@10: * (I)DCT Transforms yading@10: * Copyright (c) 2009 Peter Ross yading@10: * Copyright (c) 2010 Alex Converse yading@10: * Copyright (c) 2010 Vitor Sessak yading@10: * yading@10: * This file is part of FFmpeg. yading@10: * yading@10: * FFmpeg is free software; you can redistribute it and/or yading@10: * modify it under the terms of the GNU Lesser General Public yading@10: * License as published by the Free Software Foundation; either yading@10: * version 2.1 of the License, or (at your option) any later version. yading@10: * yading@10: * FFmpeg is distributed in the hope that it will be useful, yading@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of yading@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@10: * Lesser General Public License for more details. yading@10: * yading@10: * You should have received a copy of the GNU Lesser General Public yading@10: * License along with FFmpeg; if not, write to the Free Software yading@10: * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA yading@10: */ yading@10: yading@10: /** yading@10: * @file yading@10: * (Inverse) Discrete Cosine Transforms. These are also known as the yading@10: * type II and type III DCTs respectively. yading@10: */ yading@10: yading@10: #include yading@10: #include yading@10: yading@10: #include "libavutil/mathematics.h" yading@10: #include "dct.h" yading@10: #include "dct32.h" yading@10: yading@10: /* sin((M_PI * x / (2 * n)) */ yading@10: #define SIN(s, n, x) (s->costab[(n) - (x)]) yading@10: yading@10: /* cos((M_PI * x / (2 * n)) */ yading@10: #define COS(s, n, x) (s->costab[x]) yading@10: yading@10: static void ff_dst_calc_I_c(DCTContext *ctx, FFTSample *data) yading@10: { yading@10: int n = 1 << ctx->nbits; yading@10: int i; yading@10: yading@10: data[0] = 0; yading@10: for (i = 1; i < n / 2; i++) { yading@10: float tmp1 = data[i ]; yading@10: float tmp2 = data[n - i]; yading@10: float s = SIN(ctx, n, 2 * i); yading@10: yading@10: s *= tmp1 + tmp2; yading@10: tmp1 = (tmp1 - tmp2) * 0.5f; yading@10: data[i] = s + tmp1; yading@10: data[n - i] = s - tmp1; yading@10: } yading@10: yading@10: data[n / 2] *= 2; yading@10: ctx->rdft.rdft_calc(&ctx->rdft, data); yading@10: yading@10: data[0] *= 0.5f; yading@10: yading@10: for (i = 1; i < n - 2; i += 2) { yading@10: data[i + 1] += data[i - 1]; yading@10: data[i] = -data[i + 2]; yading@10: } yading@10: yading@10: data[n - 1] = 0; yading@10: } yading@10: yading@10: static void ff_dct_calc_I_c(DCTContext *ctx, FFTSample *data) yading@10: { yading@10: int n = 1 << ctx->nbits; yading@10: int i; yading@10: float next = -0.5f * (data[0] - data[n]); yading@10: yading@10: for (i = 0; i < n / 2; i++) { yading@10: float tmp1 = data[i]; yading@10: float tmp2 = data[n - i]; yading@10: float s = SIN(ctx, n, 2 * i); yading@10: float c = COS(ctx, n, 2 * i); yading@10: yading@10: c *= tmp1 - tmp2; yading@10: s *= tmp1 - tmp2; yading@10: yading@10: next += c; yading@10: yading@10: tmp1 = (tmp1 + tmp2) * 0.5f; yading@10: data[i] = tmp1 - s; yading@10: data[n - i] = tmp1 + s; yading@10: } yading@10: yading@10: ctx->rdft.rdft_calc(&ctx->rdft, data); yading@10: data[n] = data[1]; yading@10: data[1] = next; yading@10: yading@10: for (i = 3; i <= n; i += 2) yading@10: data[i] = data[i - 2] - data[i]; yading@10: } yading@10: yading@10: static void ff_dct_calc_III_c(DCTContext *ctx, FFTSample *data) yading@10: { yading@10: int n = 1 << ctx->nbits; yading@10: int i; yading@10: yading@10: float next = data[n - 1]; yading@10: float inv_n = 1.0f / n; yading@10: yading@10: for (i = n - 2; i >= 2; i -= 2) { yading@10: float val1 = data[i]; yading@10: float val2 = data[i - 1] - data[i + 1]; yading@10: float c = COS(ctx, n, i); yading@10: float s = SIN(ctx, n, i); yading@10: yading@10: data[i] = c * val1 + s * val2; yading@10: data[i + 1] = s * val1 - c * val2; yading@10: } yading@10: yading@10: data[1] = 2 * next; yading@10: yading@10: ctx->rdft.rdft_calc(&ctx->rdft, data); yading@10: yading@10: for (i = 0; i < n / 2; i++) { yading@10: float tmp1 = data[i] * inv_n; yading@10: float tmp2 = data[n - i - 1] * inv_n; yading@10: float csc = ctx->csc2[i] * (tmp1 - tmp2); yading@10: yading@10: tmp1 += tmp2; yading@10: data[i] = tmp1 + csc; yading@10: data[n - i - 1] = tmp1 - csc; yading@10: } yading@10: } yading@10: yading@10: static void ff_dct_calc_II_c(DCTContext *ctx, FFTSample *data) yading@10: { yading@10: int n = 1 << ctx->nbits; yading@10: int i; yading@10: float next; yading@10: yading@10: for (i = 0; i < n / 2; i++) { yading@10: float tmp1 = data[i]; yading@10: float tmp2 = data[n - i - 1]; yading@10: float s = SIN(ctx, n, 2 * i + 1); yading@10: yading@10: s *= tmp1 - tmp2; yading@10: tmp1 = (tmp1 + tmp2) * 0.5f; yading@10: yading@10: data[i] = tmp1 + s; yading@10: data[n-i-1] = tmp1 - s; yading@10: } yading@10: yading@10: ctx->rdft.rdft_calc(&ctx->rdft, data); yading@10: yading@10: next = data[1] * 0.5; yading@10: data[1] *= -1; yading@10: yading@10: for (i = n - 2; i >= 0; i -= 2) { yading@10: float inr = data[i ]; yading@10: float ini = data[i + 1]; yading@10: float c = COS(ctx, n, i); yading@10: float s = SIN(ctx, n, i); yading@10: yading@10: data[i] = c * inr + s * ini; yading@10: data[i + 1] = next; yading@10: yading@10: next += s * inr - c * ini; yading@10: } yading@10: } yading@10: yading@10: static void dct32_func(DCTContext *ctx, FFTSample *data) yading@10: { yading@10: ctx->dct32(data, data); yading@10: } yading@10: yading@10: av_cold int ff_dct_init(DCTContext *s, int nbits, enum DCTTransformType inverse) yading@10: { yading@10: int n = 1 << nbits; yading@10: int i; yading@10: yading@10: memset(s, 0, sizeof(*s)); yading@10: yading@10: s->nbits = nbits; yading@10: s->inverse = inverse; yading@10: yading@10: if (inverse == DCT_II && nbits == 5) { yading@10: s->dct_calc = dct32_func; yading@10: } else { yading@10: ff_init_ff_cos_tabs(nbits + 2); yading@10: yading@10: s->costab = ff_cos_tabs[nbits + 2]; yading@10: s->csc2 = av_malloc(n / 2 * sizeof(FFTSample)); yading@10: yading@10: if (ff_rdft_init(&s->rdft, nbits, inverse == DCT_III) < 0) { yading@10: av_free(s->csc2); yading@10: return -1; yading@10: } yading@10: yading@10: for (i = 0; i < n / 2; i++) yading@10: s->csc2[i] = 0.5 / sin((M_PI / (2 * n) * (2 * i + 1))); yading@10: yading@10: switch (inverse) { yading@10: case DCT_I : s->dct_calc = ff_dct_calc_I_c; break; yading@10: case DCT_II : s->dct_calc = ff_dct_calc_II_c; break; yading@10: case DCT_III: s->dct_calc = ff_dct_calc_III_c; break; yading@10: case DST_I : s->dct_calc = ff_dst_calc_I_c; break; yading@10: } yading@10: } yading@10: yading@10: s->dct32 = ff_dct32_float; yading@10: if (ARCH_X86) yading@10: ff_dct_init_x86(s); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: av_cold void ff_dct_end(DCTContext *s) yading@10: { yading@10: ff_rdft_end(&s->rdft); yading@10: av_free(s->csc2); yading@10: }