1/* 2 * Copyright (C) 2004 the ffmpeg project 3 * 4 * This file is part of FFmpeg. 5 * 6 * FFmpeg is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU Lesser General Public 8 * License as published by the Free Software Foundation; either 9 * version 2.1 of the License, or (at your option) any later version. 10 * 11 * FFmpeg is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 14 * Lesser General Public License for more details. 15 * 16 * You should have received a copy of the GNU Lesser General Public 17 * License along with FFmpeg; if not, write to the Free Software 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 19 */ 20 21/** 22 * @file 23 * SSE2-optimized functions cribbed from the original VP3 source code. 24 */ 25 26#include "libavcodec/dsputil.h" 27#include "dsputil_mmx.h" 28#include "vp3dsp_sse2.h" 29 30DECLARE_ALIGNED(16, const uint16_t, ff_vp3_idct_data)[7 * 8] = 31{ 32 64277,64277,64277,64277,64277,64277,64277,64277, 33 60547,60547,60547,60547,60547,60547,60547,60547, 34 54491,54491,54491,54491,54491,54491,54491,54491, 35 46341,46341,46341,46341,46341,46341,46341,46341, 36 36410,36410,36410,36410,36410,36410,36410,36410, 37 25080,25080,25080,25080,25080,25080,25080,25080, 38 12785,12785,12785,12785,12785,12785,12785,12785 39}; 40 41 42#define VP3_1D_IDCT_SSE2(ADD, SHIFT) \ 43 "movdqa "I(3)", %%xmm2 \n\t" /* xmm2 = i3 */ \ 44 "movdqa "C(3)", %%xmm6 \n\t" /* xmm6 = c3 */ \ 45 "movdqa %%xmm2, %%xmm4 \n\t" /* xmm4 = i3 */ \ 46 "movdqa "I(5)", %%xmm7 \n\t" /* xmm7 = i5 */ \ 47 "pmulhw %%xmm6, %%xmm4 \n\t" /* xmm4 = c3 * i3 - i3 */ \ 48 "movdqa "C(5)", %%xmm1 \n\t" /* xmm1 = c5 */ \ 49 "pmulhw %%xmm7, %%xmm6 \n\t" /* xmm6 = c3 * i5 - i5 */ \ 50 "movdqa %%xmm1, %%xmm5 \n\t" /* xmm5 = c5 */ \ 51 "pmulhw %%xmm2, %%xmm1 \n\t" /* xmm1 = c5 * i3 - i3 */ \ 52 "movdqa "I(1)", %%xmm3 \n\t" /* xmm3 = i1 */ \ 53 "pmulhw %%xmm7, %%xmm5 \n\t" /* xmm5 = c5 * i5 - i5 */ \ 54 "movdqa "C(1)", %%xmm0 \n\t" /* xmm0 = c1 */ \ 55 "paddw %%xmm2, %%xmm4 \n\t" /* xmm4 = c3 * i3 */ \ 56 "paddw %%xmm7, %%xmm6 \n\t" /* xmm6 = c3 * i5 */ \ 57 "paddw %%xmm1, %%xmm2 \n\t" /* xmm2 = c5 * i3 */ \ 58 "movdqa "I(7)", %%xmm1 \n\t" /* xmm1 = i7 */ \ 59 "paddw %%xmm5, %%xmm7 \n\t" /* xmm7 = c5 * i5 */ \ 60 "movdqa %%xmm0, %%xmm5 \n\t" /* xmm5 = c1 */ \ 61 "pmulhw %%xmm3, %%xmm0 \n\t" /* xmm0 = c1 * i1 - i1 */ \ 62 "paddsw %%xmm7, %%xmm4 \n\t" /* xmm4 = c3 * i3 + c5 * i5 = C */ \ 63 "pmulhw %%xmm1, %%xmm5 \n\t" /* xmm5 = c1 * i7 - i7 */ \ 64 "movdqa "C(7)", %%xmm7 \n\t" /* xmm7 = c7 */ \ 65 "psubsw %%xmm2, %%xmm6 \n\t" /* xmm6 = c3 * i5 - c5 * i3 = D */ \ 66 "paddw %%xmm3, %%xmm0 \n\t" /* xmm0 = c1 * i1 */ \ 67 "pmulhw %%xmm7, %%xmm3 \n\t" /* xmm3 = c7 * i1 */ \ 68 "movdqa "I(2)", %%xmm2 \n\t" /* xmm2 = i2 */ \ 69 "pmulhw %%xmm1, %%xmm7 \n\t" /* xmm7 = c7 * i7 */ \ 70 "paddw %%xmm1, %%xmm5 \n\t" /* xmm5 = c1 * i7 */ \ 71 "movdqa %%xmm2, %%xmm1 \n\t" /* xmm1 = i2 */ \ 72 "pmulhw "C(2)", %%xmm2 \n\t" /* xmm2 = i2 * c2 -i2 */ \ 73 "psubsw %%xmm5, %%xmm3 \n\t" /* xmm3 = c7 * i1 - c1 * i7 = B */ \ 74 "movdqa "I(6)", %%xmm5 \n\t" /* xmm5 = i6 */ \ 75 "paddsw %%xmm7, %%xmm0 \n\t" /* xmm0 = c1 * i1 + c7 * i7 = A */ \ 76 "movdqa %%xmm5, %%xmm7 \n\t" /* xmm7 = i6 */ \ 77 "psubsw %%xmm4, %%xmm0 \n\t" /* xmm0 = A - C */ \ 78 "pmulhw "C(2)", %%xmm5 \n\t" /* xmm5 = c2 * i6 - i6 */ \ 79 "paddw %%xmm1, %%xmm2 \n\t" /* xmm2 = i2 * c2 */ \ 80 "pmulhw "C(6)", %%xmm1 \n\t" /* xmm1 = c6 * i2 */ \ 81 "paddsw %%xmm4, %%xmm4 \n\t" /* xmm4 = C + C */ \ 82 "paddsw %%xmm0, %%xmm4 \n\t" /* xmm4 = A + C = C. */ \ 83 "psubsw %%xmm6, %%xmm3 \n\t" /* xmm3 = B - D */ \ 84 "paddw %%xmm7, %%xmm5 \n\t" /* xmm5 = c2 * i6 */ \ 85 "paddsw %%xmm6, %%xmm6 \n\t" /* xmm6 = D + D */ \ 86 "pmulhw "C(6)", %%xmm7 \n\t" /* xmm7 = c6 * i6 */ \ 87 "paddsw %%xmm3, %%xmm6 \n\t" /* xmm6 = B + D = D. */ \ 88 "movdqa %%xmm4, "I(1)" \n\t" /* Save C. at I(1) */ \ 89 "psubsw %%xmm5, %%xmm1 \n\t" /* xmm1 = c6 * i2 - c2 * i6 = H */ \ 90 "movdqa "C(4)", %%xmm4 \n\t" /* xmm4 = c4 */ \ 91 "movdqa %%xmm3, %%xmm5 \n\t" /* xmm5 = B - D */ \ 92 "pmulhw %%xmm4, %%xmm3 \n\t" /* xmm3 = ( c4 -1 ) * ( B - D ) */ \ 93 "paddsw %%xmm2, %%xmm7 \n\t" /* xmm7 = c2 * i2 + c6 * i6 = G */ \ 94 "movdqa %%xmm6, "I(2)" \n\t" /* Save D. at I(2) */ \ 95 "movdqa %%xmm0, %%xmm2 \n\t" /* xmm2 = A - C */ \ 96 "movdqa "I(0)", %%xmm6 \n\t" /* xmm6 = i0 */ \ 97 "pmulhw %%xmm4, %%xmm0 \n\t" /* xmm0 = ( c4 - 1 ) * ( A - C ) = A. */ \ 98 "paddw %%xmm3, %%xmm5 \n\t" /* xmm5 = c4 * ( B - D ) = B. */ \ 99 "movdqa "I(4)", %%xmm3 \n\t" /* xmm3 = i4 */ \ 100 "psubsw %%xmm1, %%xmm5 \n\t" /* xmm5 = B. - H = B.. */ \ 101 "paddw %%xmm0, %%xmm2 \n\t" /* xmm2 = c4 * ( A - C) = A. */ \ 102 "psubsw %%xmm3, %%xmm6 \n\t" /* xmm6 = i0 - i4 */ \ 103 "movdqa %%xmm6, %%xmm0 \n\t" /* xmm0 = i0 - i4 */ \ 104 "pmulhw %%xmm4, %%xmm6 \n\t" /* xmm6 = (c4 - 1) * (i0 - i4) = F */ \ 105 "paddsw %%xmm3, %%xmm3 \n\t" /* xmm3 = i4 + i4 */ \ 106 "paddsw %%xmm1, %%xmm1 \n\t" /* xmm1 = H + H */ \ 107 "paddsw %%xmm0, %%xmm3 \n\t" /* xmm3 = i0 + i4 */ \ 108 "paddsw %%xmm5, %%xmm1 \n\t" /* xmm1 = B. + H = H. */ \ 109 "pmulhw %%xmm3, %%xmm4 \n\t" /* xmm4 = ( c4 - 1 ) * ( i0 + i4 ) */ \ 110 "paddw %%xmm0, %%xmm6 \n\t" /* xmm6 = c4 * ( i0 - i4 ) */ \ 111 "psubsw %%xmm2, %%xmm6 \n\t" /* xmm6 = F - A. = F. */ \ 112 "paddsw %%xmm2, %%xmm2 \n\t" /* xmm2 = A. + A. */ \ 113 "movdqa "I(1)", %%xmm0 \n\t" /* Load C. from I(1) */ \ 114 "paddsw %%xmm6, %%xmm2 \n\t" /* xmm2 = F + A. = A.. */ \ 115 "paddw %%xmm3, %%xmm4 \n\t" /* xmm4 = c4 * ( i0 + i4 ) = 3 */ \ 116 "psubsw %%xmm1, %%xmm2 \n\t" /* xmm2 = A.. - H. = R2 */ \ 117 ADD(%%xmm2) /* Adjust R2 and R1 before shifting */ \ 118 "paddsw %%xmm1, %%xmm1 \n\t" /* xmm1 = H. + H. */ \ 119 "paddsw %%xmm2, %%xmm1 \n\t" /* xmm1 = A.. + H. = R1 */ \ 120 SHIFT(%%xmm2) /* xmm2 = op2 */ \ 121 "psubsw %%xmm7, %%xmm4 \n\t" /* xmm4 = E - G = E. */ \ 122 SHIFT(%%xmm1) /* xmm1 = op1 */ \ 123 "movdqa "I(2)", %%xmm3 \n\t" /* Load D. from I(2) */ \ 124 "paddsw %%xmm7, %%xmm7 \n\t" /* xmm7 = G + G */ \ 125 "paddsw %%xmm4, %%xmm7 \n\t" /* xmm7 = E + G = G. */ \ 126 "psubsw %%xmm3, %%xmm4 \n\t" /* xmm4 = E. - D. = R4 */ \ 127 ADD(%%xmm4) /* Adjust R4 and R3 before shifting */ \ 128 "paddsw %%xmm3, %%xmm3 \n\t" /* xmm3 = D. + D. */ \ 129 "paddsw %%xmm4, %%xmm3 \n\t" /* xmm3 = E. + D. = R3 */ \ 130 SHIFT(%%xmm4) /* xmm4 = op4 */ \ 131 "psubsw %%xmm5, %%xmm6 \n\t" /* xmm6 = F. - B..= R6 */ \ 132 SHIFT(%%xmm3) /* xmm3 = op3 */ \ 133 ADD(%%xmm6) /* Adjust R6 and R5 before shifting */ \ 134 "paddsw %%xmm5, %%xmm5 \n\t" /* xmm5 = B.. + B.. */ \ 135 "paddsw %%xmm6, %%xmm5 \n\t" /* xmm5 = F. + B.. = R5 */ \ 136 SHIFT(%%xmm6) /* xmm6 = op6 */ \ 137 SHIFT(%%xmm5) /* xmm5 = op5 */ \ 138 "psubsw %%xmm0, %%xmm7 \n\t" /* xmm7 = G. - C. = R7 */ \ 139 ADD(%%xmm7) /* Adjust R7 and R0 before shifting */ \ 140 "paddsw %%xmm0, %%xmm0 \n\t" /* xmm0 = C. + C. */ \ 141 "paddsw %%xmm7, %%xmm0 \n\t" /* xmm0 = G. + C. */ \ 142 SHIFT(%%xmm7) /* xmm7 = op7 */ \ 143 SHIFT(%%xmm0) /* xmm0 = op0 */ 144 145#define PUT_BLOCK(r0, r1, r2, r3, r4, r5, r6, r7) \ 146 "movdqa " #r0 ", " O(0) "\n\t" \ 147 "movdqa " #r1 ", " O(1) "\n\t" \ 148 "movdqa " #r2 ", " O(2) "\n\t" \ 149 "movdqa " #r3 ", " O(3) "\n\t" \ 150 "movdqa " #r4 ", " O(4) "\n\t" \ 151 "movdqa " #r5 ", " O(5) "\n\t" \ 152 "movdqa " #r6 ", " O(6) "\n\t" \ 153 "movdqa " #r7 ", " O(7) "\n\t" 154 155#define NOP(xmm) 156#define SHIFT4(xmm) "psraw $4, "#xmm"\n\t" 157#define ADD8(xmm) "paddsw %2, "#xmm"\n\t" 158 159void ff_vp3_idct_sse2(int16_t *input_data) 160{ 161#define I(x) AV_STRINGIFY(16*x)"(%0)" 162#define O(x) I(x) 163#define C(x) AV_STRINGIFY(16*(x-1))"(%1)" 164 165 __asm__ volatile ( 166 VP3_1D_IDCT_SSE2(NOP, NOP) 167 168 TRANSPOSE8(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm6, %%xmm7, (%0)) 169 PUT_BLOCK(%%xmm0, %%xmm5, %%xmm7, %%xmm3, %%xmm6, %%xmm4, %%xmm2, %%xmm1) 170 171 VP3_1D_IDCT_SSE2(ADD8, SHIFT4) 172 PUT_BLOCK(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm6, %%xmm7) 173 :: "r"(input_data), "r"(ff_vp3_idct_data), "m"(ff_pw_8) 174 ); 175} 176 177void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, DCTELEM *block) 178{ 179 ff_vp3_idct_sse2(block); 180 put_signed_pixels_clamped_mmx(block, dest, line_size); 181} 182 183void ff_vp3_idct_add_sse2(uint8_t *dest, int line_size, DCTELEM *block) 184{ 185 ff_vp3_idct_sse2(block); 186 add_pixels_clamped_mmx(block, dest, line_size); 187} 188