From 6e0593e80b30d9225e1adba32f7dfccddf2fffb0 Mon Sep 17 00:00:00 2001 From: Michael Niedermayer Date: Sat, 25 Oct 2003 21:59:28 +0000 Subject: [PATCH] reorder table instead of wasting instructions to reorder the input to match the table Originally committed as revision 2436 to svn://svn.ffmpeg.org/ffmpeg/trunk --- libavcodec/i386/fdct_mmx.c | 232 +++++++++++++++++-------------------- 1 file changed, 108 insertions(+), 124 deletions(-) diff --git a/libavcodec/i386/fdct_mmx.c b/libavcodec/i386/fdct_mmx.c index 2f9a5aef96..5f2074804a 100644 --- a/libavcodec/i386/fdct_mmx.c +++ b/libavcodec/i386/fdct_mmx.c @@ -49,85 +49,77 @@ static const long long fdct_one_corr ATTR_ALIGN(8) = 0x0001000100010001LL; static const long fdct_r_row[2] ATTR_ALIGN(8) = {RND_FRW_ROW, RND_FRW_ROW }; static const int16_t tab_frw_01234567[] ATTR_ALIGN(8) = { // forward_dct coeff table - //row0 - 16384, 16384, 21407, -8867, // w09 w01 w08 w00 - 16384, 16384, 8867, -21407, // w13 w05 w12 w04 - 16384, -16384, 8867, 21407, // w11 w03 w10 w02 - -16384, 16384, -21407, -8867, // w15 w07 w14 w06 - 22725, 12873, 19266, -22725, // w22 w20 w18 w16 - 19266, 4520, -4520, -12873, // w23 w21 w19 w17 - 12873, 4520, 4520, 19266, // w30 w28 w26 w24 - -22725, 19266, -12873, -22725, // w31 w29 w27 w25 + 16384, 16384, -8867, -21407, + 16384, 16384, 21407, 8867, + 16384, -16384, 21407, -8867, + -16384, 16384, 8867, -21407, + 22725, 19266, -22725, -12873, + 12873, 4520, 19266, -4520, + 12873, -22725, 19266, -22725, + 4520, 19266, 4520, -12873, - //row1 - 22725, 22725, 29692, -12299, // w09 w01 w08 w00 - 22725, 22725, 12299, -29692, // w13 w05 w12 w04 - 22725, -22725, 12299, 29692, // w11 w03 w10 w02 - -22725, 22725, -29692, -12299, // w15 w07 w14 w06 - 31521, 17855, 26722, -31521, // w22 w20 w18 w16 - 26722, 6270, -6270, -17855, // w23 w21 w19 w17 - 17855, 6270, 6270, 26722, // w30 w28 w26 w24 - -31521, 26722, -17855, -31521, // w31 w29 w27 w25 + 22725, 22725, -12299, -29692, + 22725, 22725, 29692, 12299, + 22725, -22725, 29692, -12299, + -22725, 22725, 12299, -29692, + 31521, 26722, -31521, -17855, + 17855, 6270, 26722, -6270, + 17855, -31521, 26722, -31521, + 6270, 26722, 6270, -17855, - //row2 - 21407, 21407, 27969, -11585, // w09 w01 w08 w00 - 21407, 21407, 11585, -27969, // w13 w05 w12 w04 - 21407, -21407, 11585, 27969, // w11 w03 w10 w02 - -21407, 21407, -27969, -11585, // w15 w07 w14 w06 - 29692, 16819, 25172, -29692, // w22 w20 w18 w16 - 25172, 5906, -5906, -16819, // w23 w21 w19 w17 - 16819, 5906, 5906, 25172, // w30 w28 w26 w24 - -29692, 25172, -16819, -29692, // w31 w29 w27 w25 + 21407, 21407, -11585, -27969, + 21407, 21407, 27969, 11585, + 21407, -21407, 27969, -11585, + -21407, 21407, 11585, -27969, + 29692, 25172, -29692, -16819, + 16819, 5906, 25172, -5906, + 16819, -29692, 25172, -29692, + 5906, 25172, 5906, -16819, - //row3 - 19266, 19266, 25172, -10426, // w09 w01 w08 w00 - 19266, 19266, 10426, -25172, // w13 w05 w12 w04 - 19266, -19266, 10426, 25172, // w11 w03 w10 w02 - -19266, 19266, -25172, -10426, // w15 w07 w14 w06, - 26722, 15137, 22654, -26722, // w22 w20 w18 w16 - 22654, 5315, -5315, -15137, // w23 w21 w19 w17 - 15137, 5315, 5315, 22654, // w30 w28 w26 w24 - -26722, 22654, -15137, -26722, // w31 w29 w27 w25, + 19266, 19266, -10426, -25172, + 19266, 19266, 25172, 10426, + 19266, -19266, 25172, -10426, + -19266, 19266, 10426, -25172, + 26722, 22654, -26722, -15137, + 15137, 5315, 22654, -5315, + 15137, -26722, 22654, -26722, + 5315, 22654, 5315, -15137, - //row4 - 16384, 16384, 21407, -8867, // w09 w01 w08 w00 - 16384, 16384, 8867, -21407, // w13 w05 w12 w04 - 16384, -16384, 8867, 21407, // w11 w03 w10 w02 - -16384, 16384, -21407, -8867, // w15 w07 w14 w06 - 22725, 12873, 19266, -22725, // w22 w20 w18 w16 - 19266, 4520, -4520, -12873, // w23 w21 w19 w17 - 12873, 4520, 4520, 19266, // w30 w28 w26 w24 - -22725, 19266, -12873, -22725, // w31 w29 w27 w25 + 16384, 16384, -8867, -21407, + 16384, 16384, 21407, 8867, + 16384, -16384, 21407, -8867, + -16384, 16384, 8867, -21407, + 22725, 19266, -22725, -12873, + 12873, 4520, 19266, -4520, + 12873, -22725, 19266, -22725, + 4520, 19266, 4520, -12873, - //row5 - 19266, 19266, 25172, -10426, // w09 w01 w08 w00 - 19266, 19266, 10426, -25172, // w13 w05 w12 w04 - 19266, -19266, 10426, 25172, // w11 w03 w10 w02 - -19266, 19266, -25172, -10426, // w15 w07 w14 w06 - 26722, 15137, 22654, -26722, // w22 w20 w18 w16 - 22654, 5315, -5315, -15137, // w23 w21 w19 w17 - 15137, 5315, 5315, 22654, // w30 w28 w26 w24 - -26722, 22654, -15137, -26722, // w31 w29 w27 w25 + 19266, 19266, -10426, -25172, + 19266, 19266, 25172, 10426, + 19266, -19266, 25172, -10426, + -19266, 19266, 10426, -25172, + 26722, 22654, -26722, -15137, + 15137, 5315, 22654, -5315, + 15137, -26722, 22654, -26722, + 5315, 22654, 5315, -15137, - //row6 - 21407, 21407, 27969, -11585, // w09 w01 w08 w00 - 21407, 21407, 11585, -27969, // w13 w05 w12 w04 - 21407, -21407, 11585, 27969, // w11 w03 w10 w02 - -21407, 21407, -27969, -11585, // w15 w07 w14 w06, - 29692, 16819, 25172, -29692, // w22 w20 w18 w16 - 25172, 5906, -5906, -16819, // w23 w21 w19 w17 - 16819, 5906, 5906, 25172, // w30 w28 w26 w24 - -29692, 25172, -16819, -29692, // w31 w29 w27 w25, + 21407, 21407, -11585, -27969, + 21407, 21407, 27969, 11585, + 21407, -21407, 27969, -11585, + -21407, 21407, 11585, -27969, + 29692, 25172, -29692, -16819, + 16819, 5906, 25172, -5906, + 16819, -29692, 25172, -29692, + 5906, 25172, 5906, -16819, - //row7 - 22725, 22725, 29692, -12299, // w09 w01 w08 w00 - 22725, 22725, 12299, -29692, // w13 w05 w12 w04 - 22725, -22725, 12299, 29692, // w11 w03 w10 w02 - -22725, 22725, -29692, -12299, // w15 w07 w14 w06, - 31521, 17855, 26722, -31521, // w22 w20 w18 w16 - 26722, 6270, -6270, -17855, // w23 w21 w19 w17 - 17855, 6270, 6270, 26722, // w30 w28 w26 w24 - -31521, 26722, -17855, -31521 // w31 w29 w27 w25 + 22725, 22725, -12299, -29692, + 22725, 22725, 29692, 12299, + 22725, -22725, 29692, -12299, + -22725, 22725, 12299, -29692, + 31521, 26722, -31521, -17855, + 17855, 6270, 26722, -6270, + 17855, -31521, 26722, -31521, + 6270, 26722, 6270, -17855, }; @@ -217,73 +209,65 @@ static always_inline void fdct_row_mmx2(const int16_t *in, int16_t *out, const i movq_r2r(mm0, mm1); paddsw_r2r(mm5, mm0); psubsw_r2r(mm5, mm1); - pshufw_r2r(mm1, mm5, 0x88); - pshufw_r2r(mm1, mm6, 0xDD); - pshufw_r2r(mm0, mm1, 0xDD); - pshufw_r2r(mm0, mm0, 0x88); - movq_m2r(*(table + 0), mm3); - movq_m2r(*(table + 4), mm4); - movq_m2r(*(table + 16), mm2); + pshufw_r2r(mm0, mm2, 0x4E); + pshufw_r2r(mm1, mm3, 0x4E); + movq_m2r(*(table + 0), mm4); + movq_m2r(*(table + 4), mm6); + movq_m2r(*(table + 16), mm5); movq_m2r(*(table + 20), mm7); - pmaddwd_r2r(mm0, mm3); - pmaddwd_r2r(mm1, mm4); - pmaddwd_r2r(mm5, mm2); - pmaddwd_r2r(mm6, mm7); + pmaddwd_r2r(mm0, mm4); + pmaddwd_r2r(mm1, mm5); + pmaddwd_r2r(mm2, mm6); + pmaddwd_r2r(mm3, mm7); pmaddwd_m2r(*(table + 8), mm0); - pmaddwd_m2r(*(table + 12), mm1); - pmaddwd_m2r(*(table + 24), mm5); - pmaddwd_m2r(*(table + 28), mm6); - paddd_r2r(mm4, mm3); - paddd_r2r(mm7, mm2); - paddd_r2r(mm1, mm0); - paddd_r2r(mm6, mm5); - movq_m2r(*fdct_r_row, mm7); - paddd_r2r(mm7, mm3); - paddd_r2r(mm7, mm0); - paddd_r2r(mm7, mm2); + pmaddwd_m2r(*(table + 12), mm2); + pmaddwd_m2r(*(table + 24), mm1); + pmaddwd_m2r(*(table + 28), mm3); + paddd_r2r(mm6, mm4); paddd_r2r(mm7, mm5); - psrad_i2r(SHIFT_FRW_ROW, mm3); - psrad_i2r(SHIFT_FRW_ROW, mm2); - psrad_i2r(SHIFT_FRW_ROW, mm0); + paddd_r2r(mm2, mm0); + paddd_r2r(mm3, mm1); + movq_m2r(*fdct_r_row, mm7); + paddd_r2r(mm7, mm4); + paddd_r2r(mm7, mm5); + paddd_r2r(mm7, mm0); + paddd_r2r(mm7, mm1); + psrad_i2r(SHIFT_FRW_ROW, mm4); psrad_i2r(SHIFT_FRW_ROW, mm5); - packssdw_r2r(mm0, mm3); - packssdw_r2r(mm5, mm2); - movq_r2r(mm3, mm6); - punpcklwd_r2r(mm2, mm3); - punpckhwd_r2r(mm2, mm6); - movq_r2m(mm3, *(out + 0)); - movq_r2m(mm6, *(out + 4)); + psrad_i2r(SHIFT_FRW_ROW, mm0); + psrad_i2r(SHIFT_FRW_ROW, mm1); + packssdw_r2r(mm0, mm4); + packssdw_r2r(mm1, mm5); + movq_r2r(mm4, mm2); + punpcklwd_r2r(mm5, mm4); + punpckhwd_r2r(mm5, mm2); + movq_r2m(mm4, *(out + 0)); + movq_r2m(mm2, *(out + 4)); } static always_inline void fdct_row_mmx(const int16_t *in, int16_t *out, const int16_t *table) { - movd_m2r(*(in + 6), mm5); - punpcklwd_m2r(*(in + 4), mm5); - movq_r2r(mm5, mm2); - psrlq_i2r(0x20, mm5); + movd_m2r(*(in + 6), mm1); + punpcklwd_m2r(*(in + 4), mm1); + movq_r2r(mm1, mm2); + psrlq_i2r(0x20, mm1); movq_m2r(*(in + 0), mm0); - punpcklwd_r2r(mm2, mm5); - movq_r2r(mm0, mm1); - paddsw_r2r(mm5, mm0); - psubsw_r2r(mm5, mm1); - movq_r2r(mm0, mm2); - punpcklwd_r2r(mm1, mm0); - punpckhwd_r2r(mm1, mm2); - movq_r2r(mm0, mm1); + punpcklwd_r2r(mm2, mm1); + movq_r2r(mm0, mm5); + paddsw_r2r(mm1, mm0); + psubsw_r2r(mm1, mm5); + movq_r2r(mm0, mm1); + movq_r2r(mm5, mm6); + punpckldq_r2r(mm5, mm3); + punpckhdq_r2r(mm3, mm6); movq_m2r(*(table + 0), mm3); - punpcklwd_r2r(mm2, mm0); - movq_r2r(mm0, mm5); - punpckldq_r2r(mm0, mm0); movq_m2r(*(table + 4), mm4); - punpckhwd_r2r(mm2, mm1); + punpckldq_r2r(mm0, mm2); pmaddwd_r2r(mm0, mm3); - movq_r2r(mm1, mm6); + punpckhdq_r2r(mm2, mm1); movq_m2r(*(table + 16), mm2); - punpckldq_r2r(mm1, mm1); pmaddwd_r2r(mm1, mm4); - punpckhdq_r2r(mm5, mm5); pmaddwd_m2r(*(table + 8), mm0); - punpckhdq_r2r(mm6, mm6); movq_m2r(*(table + 20), mm7); pmaddwd_r2r(mm5, mm2); paddd_m2r(*fdct_r_row, mm3);