yading@10: /* yading@10: * Simple IDCT (Alpha optimized) yading@10: * yading@10: * Copyright (c) 2001 Michael Niedermayer yading@10: * yading@10: * based upon some outcommented C code from mpeg2dec (idct_mmx.c yading@10: * written by Aaron Holtzman ) yading@10: * yading@10: * Alpha optimizations by Måns Rullgård yading@10: * and Falk Hueffner 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 Street, Fifth Floor, Boston, MA 02110-1301 USA yading@10: */ yading@10: yading@10: #include "dsputil_alpha.h" yading@10: #include "asm.h" yading@10: yading@10: // cos(i * M_PI / 16) * sqrt(2) * (1 << 14) yading@10: // W4 is actually exactly 16384, but using 16383 works around yading@10: // accumulating rounding errors for some encoders yading@10: #define W1 22725 yading@10: #define W2 21407 yading@10: #define W3 19266 yading@10: #define W4 16383 yading@10: #define W5 12873 yading@10: #define W6 8867 yading@10: #define W7 4520 yading@10: #define ROW_SHIFT 11 yading@10: #define COL_SHIFT 20 yading@10: yading@10: /* 0: all entries 0, 1: only first entry nonzero, 2: otherwise */ yading@10: static inline int idct_row(int16_t *row) yading@10: { yading@10: int a0, a1, a2, a3, b0, b1, b2, b3, t; yading@10: uint64_t l, r, t2; yading@10: l = ldq(row); yading@10: r = ldq(row + 4); yading@10: yading@10: if (l == 0 && r == 0) yading@10: return 0; yading@10: yading@10: a0 = W4 * sextw(l) + (1 << (ROW_SHIFT - 1)); yading@10: yading@10: if (((l & ~0xffffUL) | r) == 0) { yading@10: a0 >>= ROW_SHIFT; yading@10: t2 = (uint16_t) a0; yading@10: t2 |= t2 << 16; yading@10: t2 |= t2 << 32; yading@10: yading@10: stq(t2, row); yading@10: stq(t2, row + 4); yading@10: return 1; yading@10: } yading@10: yading@10: a1 = a0; yading@10: a2 = a0; yading@10: a3 = a0; yading@10: yading@10: t = extwl(l, 4); /* row[2] */ yading@10: if (t != 0) { yading@10: t = sextw(t); yading@10: a0 += W2 * t; yading@10: a1 += W6 * t; yading@10: a2 -= W6 * t; yading@10: a3 -= W2 * t; yading@10: } yading@10: yading@10: t = extwl(r, 0); /* row[4] */ yading@10: if (t != 0) { yading@10: t = sextw(t); yading@10: a0 += W4 * t; yading@10: a1 -= W4 * t; yading@10: a2 -= W4 * t; yading@10: a3 += W4 * t; yading@10: } yading@10: yading@10: t = extwl(r, 4); /* row[6] */ yading@10: if (t != 0) { yading@10: t = sextw(t); yading@10: a0 += W6 * t; yading@10: a1 -= W2 * t; yading@10: a2 += W2 * t; yading@10: a3 -= W6 * t; yading@10: } yading@10: yading@10: t = extwl(l, 2); /* row[1] */ yading@10: if (t != 0) { yading@10: t = sextw(t); yading@10: b0 = W1 * t; yading@10: b1 = W3 * t; yading@10: b2 = W5 * t; yading@10: b3 = W7 * t; yading@10: } else { yading@10: b0 = 0; yading@10: b1 = 0; yading@10: b2 = 0; yading@10: b3 = 0; yading@10: } yading@10: yading@10: t = extwl(l, 6); /* row[3] */ yading@10: if (t) { yading@10: t = sextw(t); yading@10: b0 += W3 * t; yading@10: b1 -= W7 * t; yading@10: b2 -= W1 * t; yading@10: b3 -= W5 * t; yading@10: } yading@10: yading@10: yading@10: t = extwl(r, 2); /* row[5] */ yading@10: if (t) { yading@10: t = sextw(t); yading@10: b0 += W5 * t; yading@10: b1 -= W1 * t; yading@10: b2 += W7 * t; yading@10: b3 += W3 * t; yading@10: } yading@10: yading@10: t = extwl(r, 6); /* row[7] */ yading@10: if (t) { yading@10: t = sextw(t); yading@10: b0 += W7 * t; yading@10: b1 -= W5 * t; yading@10: b2 += W3 * t; yading@10: b3 -= W1 * t; yading@10: } yading@10: yading@10: row[0] = (a0 + b0) >> ROW_SHIFT; yading@10: row[1] = (a1 + b1) >> ROW_SHIFT; yading@10: row[2] = (a2 + b2) >> ROW_SHIFT; yading@10: row[3] = (a3 + b3) >> ROW_SHIFT; yading@10: row[4] = (a3 - b3) >> ROW_SHIFT; yading@10: row[5] = (a2 - b2) >> ROW_SHIFT; yading@10: row[6] = (a1 - b1) >> ROW_SHIFT; yading@10: row[7] = (a0 - b0) >> ROW_SHIFT; yading@10: yading@10: return 2; yading@10: } yading@10: yading@10: static inline void idct_col(int16_t *col) yading@10: { yading@10: int a0, a1, a2, a3, b0, b1, b2, b3; yading@10: yading@10: col[0] += (1 << (COL_SHIFT - 1)) / W4; yading@10: yading@10: a0 = W4 * col[8 * 0]; yading@10: a1 = W4 * col[8 * 0]; yading@10: a2 = W4 * col[8 * 0]; yading@10: a3 = W4 * col[8 * 0]; yading@10: yading@10: if (col[8 * 2]) { yading@10: a0 += W2 * col[8 * 2]; yading@10: a1 += W6 * col[8 * 2]; yading@10: a2 -= W6 * col[8 * 2]; yading@10: a3 -= W2 * col[8 * 2]; yading@10: } yading@10: yading@10: if (col[8 * 4]) { yading@10: a0 += W4 * col[8 * 4]; yading@10: a1 -= W4 * col[8 * 4]; yading@10: a2 -= W4 * col[8 * 4]; yading@10: a3 += W4 * col[8 * 4]; yading@10: } yading@10: yading@10: if (col[8 * 6]) { yading@10: a0 += W6 * col[8 * 6]; yading@10: a1 -= W2 * col[8 * 6]; yading@10: a2 += W2 * col[8 * 6]; yading@10: a3 -= W6 * col[8 * 6]; yading@10: } yading@10: yading@10: if (col[8 * 1]) { yading@10: b0 = W1 * col[8 * 1]; yading@10: b1 = W3 * col[8 * 1]; yading@10: b2 = W5 * col[8 * 1]; yading@10: b3 = W7 * col[8 * 1]; yading@10: } else { yading@10: b0 = 0; yading@10: b1 = 0; yading@10: b2 = 0; yading@10: b3 = 0; yading@10: } yading@10: yading@10: if (col[8 * 3]) { yading@10: b0 += W3 * col[8 * 3]; yading@10: b1 -= W7 * col[8 * 3]; yading@10: b2 -= W1 * col[8 * 3]; yading@10: b3 -= W5 * col[8 * 3]; yading@10: } yading@10: yading@10: if (col[8 * 5]) { yading@10: b0 += W5 * col[8 * 5]; yading@10: b1 -= W1 * col[8 * 5]; yading@10: b2 += W7 * col[8 * 5]; yading@10: b3 += W3 * col[8 * 5]; yading@10: } yading@10: yading@10: if (col[8 * 7]) { yading@10: b0 += W7 * col[8 * 7]; yading@10: b1 -= W5 * col[8 * 7]; yading@10: b2 += W3 * col[8 * 7]; yading@10: b3 -= W1 * col[8 * 7]; yading@10: } yading@10: yading@10: col[8 * 0] = (a0 + b0) >> COL_SHIFT; yading@10: col[8 * 7] = (a0 - b0) >> COL_SHIFT; yading@10: col[8 * 1] = (a1 + b1) >> COL_SHIFT; yading@10: col[8 * 6] = (a1 - b1) >> COL_SHIFT; yading@10: col[8 * 2] = (a2 + b2) >> COL_SHIFT; yading@10: col[8 * 5] = (a2 - b2) >> COL_SHIFT; yading@10: col[8 * 3] = (a3 + b3) >> COL_SHIFT; yading@10: col[8 * 4] = (a3 - b3) >> COL_SHIFT; yading@10: } yading@10: yading@10: /* If all rows but the first one are zero after row transformation, yading@10: all rows will be identical after column transformation. */ yading@10: static inline void idct_col2(int16_t *col) yading@10: { yading@10: int i; yading@10: uint64_t l, r; yading@10: yading@10: for (i = 0; i < 8; ++i) { yading@10: int a0 = col[i] + (1 << (COL_SHIFT - 1)) / W4; yading@10: yading@10: a0 *= W4; yading@10: col[i] = a0 >> COL_SHIFT; yading@10: } yading@10: yading@10: l = ldq(col + 0 * 4); r = ldq(col + 1 * 4); yading@10: stq(l, col + 2 * 4); stq(r, col + 3 * 4); yading@10: stq(l, col + 4 * 4); stq(r, col + 5 * 4); yading@10: stq(l, col + 6 * 4); stq(r, col + 7 * 4); yading@10: stq(l, col + 8 * 4); stq(r, col + 9 * 4); yading@10: stq(l, col + 10 * 4); stq(r, col + 11 * 4); yading@10: stq(l, col + 12 * 4); stq(r, col + 13 * 4); yading@10: stq(l, col + 14 * 4); stq(r, col + 15 * 4); yading@10: } yading@10: yading@10: void ff_simple_idct_axp(int16_t *block) yading@10: { yading@10: yading@10: int i; yading@10: int rowsZero = 1; /* all rows except row 0 zero */ yading@10: int rowsConstant = 1; /* all rows consist of a constant value */ yading@10: yading@10: for (i = 0; i < 8; i++) { yading@10: int sparseness = idct_row(block + 8 * i); yading@10: yading@10: if (i > 0 && sparseness > 0) yading@10: rowsZero = 0; yading@10: if (sparseness == 2) yading@10: rowsConstant = 0; yading@10: } yading@10: yading@10: if (rowsZero) { yading@10: idct_col2(block); yading@10: } else if (rowsConstant) { yading@10: idct_col(block); yading@10: for (i = 0; i < 8; i += 2) { yading@10: uint64_t v = (uint16_t) block[0]; yading@10: uint64_t w = (uint16_t) block[8]; yading@10: yading@10: v |= v << 16; yading@10: w |= w << 16; yading@10: v |= v << 32; yading@10: w |= w << 32; yading@10: stq(v, block + 0 * 4); yading@10: stq(v, block + 1 * 4); yading@10: stq(w, block + 2 * 4); yading@10: stq(w, block + 3 * 4); yading@10: block += 4 * 4; yading@10: } yading@10: } else { yading@10: for (i = 0; i < 8; i++) yading@10: idct_col(block + i); yading@10: } yading@10: } yading@10: yading@10: void ff_simple_idct_put_axp(uint8_t *dest, int line_size, int16_t *block) yading@10: { yading@10: ff_simple_idct_axp(block); yading@10: put_pixels_clamped_axp_p(block, dest, line_size); yading@10: } yading@10: yading@10: void ff_simple_idct_add_axp(uint8_t *dest, int line_size, int16_t *block) yading@10: { yading@10: ff_simple_idct_axp(block); yading@10: add_pixels_clamped_axp_p(block, dest, line_size); yading@10: }