diff options
Diffstat (limited to 'libavcodec/x86/dsputil_mmx.c')
-rw-r--r-- | libavcodec/x86/dsputil_mmx.c | 467 |
1 files changed, 199 insertions, 268 deletions
diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c index aa7b3984aa..fe59d22017 100644 --- a/libavcodec/x86/dsputil_mmx.c +++ b/libavcodec/x86/dsputil_mmx.c @@ -3,23 +3,23 @@ * Copyright (c) 2000, 2001 Fabrice Bellard * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> * - * MMX optimization by Nick Kurshev <nickols_k@mail.ru> - * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + * + * MMX optimization by Nick Kurshev <nickols_k@mail.ru> */ #include "libavutil/attributes.h" @@ -29,8 +29,10 @@ #include "libavcodec/h264dsp.h" #include "libavcodec/mpegvideo.h" #include "libavcodec/simple_idct.h" +#include "libavcodec/videodsp.h" #include "dsputil_mmx.h" #include "idct_xvid.h" +#include "diracdsp_mmx.h" //#undef NDEBUG //#include <assert.h> @@ -39,54 +41,25 @@ DECLARE_ALIGNED(8, const uint64_t, ff_bone) = 0x0101010101010101ULL; DECLARE_ALIGNED(8, const uint64_t, ff_wtwo) = 0x0002000200020002ULL; -DECLARE_ALIGNED(16, const xmm_reg, ff_pw_1) = { 0x0001000100010001ULL, 0x0001000100010001ULL }; -DECLARE_ALIGNED(16, const xmm_reg, ff_pw_2) = { 0x0002000200020002ULL, 0x0002000200020002ULL }; -DECLARE_ALIGNED(16, const xmm_reg, ff_pw_3) = { 0x0003000300030003ULL, 0x0003000300030003ULL }; -DECLARE_ALIGNED(16, const xmm_reg, ff_pw_4) = { 0x0004000400040004ULL, 0x0004000400040004ULL }; -DECLARE_ALIGNED(16, const xmm_reg, ff_pw_5) = { 0x0005000500050005ULL, 0x0005000500050005ULL }; -DECLARE_ALIGNED(16, const xmm_reg, ff_pw_8) = { 0x0008000800080008ULL, 0x0008000800080008ULL }; -DECLARE_ALIGNED(16, const xmm_reg, ff_pw_9) = { 0x0009000900090009ULL, 0x0009000900090009ULL }; DECLARE_ALIGNED(8, const uint64_t, ff_pw_15) = 0x000F000F000F000FULL; -DECLARE_ALIGNED(16, const xmm_reg, ff_pw_16) = { 0x0010001000100010ULL, 0x0010001000100010ULL }; DECLARE_ALIGNED(16, const xmm_reg, ff_pw_17) = { 0x0011001100110011ULL, 0x0011001100110011ULL }; -DECLARE_ALIGNED(16, const xmm_reg, ff_pw_18) = { 0x0012001200120012ULL, 0x0012001200120012ULL }; DECLARE_ALIGNED(8, const uint64_t, ff_pw_20) = 0x0014001400140014ULL; -DECLARE_ALIGNED(16, const xmm_reg, ff_pw_27) = { 0x001B001B001B001BULL, 0x001B001B001B001BULL }; -DECLARE_ALIGNED(16, const xmm_reg, ff_pw_28) = { 0x001C001C001C001CULL, 0x001C001C001C001CULL }; -DECLARE_ALIGNED(16, const xmm_reg, ff_pw_32) = { 0x0020002000200020ULL, 0x0020002000200020ULL }; DECLARE_ALIGNED(8, const uint64_t, ff_pw_42) = 0x002A002A002A002AULL; DECLARE_ALIGNED(8, const uint64_t, ff_pw_53) = 0x0035003500350035ULL; -DECLARE_ALIGNED(16, const xmm_reg, ff_pw_63) = { 0x003F003F003F003FULL, 0x003F003F003F003FULL }; -DECLARE_ALIGNED(16, const xmm_reg, ff_pw_64) = { 0x0040004000400040ULL, 0x0040004000400040ULL }; DECLARE_ALIGNED(8, const uint64_t, ff_pw_96) = 0x0060006000600060ULL; DECLARE_ALIGNED(8, const uint64_t, ff_pw_128) = 0x0080008000800080ULL; DECLARE_ALIGNED(8, const uint64_t, ff_pw_255) = 0x00ff00ff00ff00ffULL; DECLARE_ALIGNED(16, const xmm_reg, ff_pw_512) = { 0x0200020002000200ULL, 0x0200020002000200ULL }; DECLARE_ALIGNED(16, const xmm_reg, ff_pw_1019) = { 0x03FB03FB03FB03FBULL, 0x03FB03FB03FB03FBULL }; -DECLARE_ALIGNED(16, const xmm_reg, ff_pb_0) = { 0x0000000000000000ULL, 0x0000000000000000ULL }; -DECLARE_ALIGNED(16, const xmm_reg, ff_pb_1) = { 0x0101010101010101ULL, 0x0101010101010101ULL }; -DECLARE_ALIGNED(16, const xmm_reg, ff_pb_3) = { 0x0303030303030303ULL, 0x0303030303030303ULL }; -DECLARE_ALIGNED(16, const xmm_reg, ff_pb_4) = { 0x0404040404040404ULL, 0x0404040404040404ULL }; -DECLARE_ALIGNED(8, const uint64_t, ff_pb_7) = 0x0707070707070707ULL; -DECLARE_ALIGNED(8, const uint64_t, ff_pb_1F) = 0x1F1F1F1F1F1F1F1FULL; DECLARE_ALIGNED(8, const uint64_t, ff_pb_3F) = 0x3F3F3F3F3F3F3F3FULL; -DECLARE_ALIGNED(16, const xmm_reg, ff_pb_80) = { 0x8080808080808080ULL, 0x8080808080808080ULL }; -DECLARE_ALIGNED(8, const uint64_t, ff_pb_81) = 0x8181818181818181ULL; -DECLARE_ALIGNED(16, const xmm_reg, ff_pb_A1) = { 0xA1A1A1A1A1A1A1A1ULL, 0xA1A1A1A1A1A1A1A1ULL }; -DECLARE_ALIGNED(16, const xmm_reg, ff_pb_F8) = { 0xF8F8F8F8F8F8F8F8ULL, 0xF8F8F8F8F8F8F8F8ULL }; DECLARE_ALIGNED(8, const uint64_t, ff_pb_FC) = 0xFCFCFCFCFCFCFCFCULL; -DECLARE_ALIGNED(16, const xmm_reg, ff_pb_FE) = { 0xFEFEFEFEFEFEFEFEULL, 0xFEFEFEFEFEFEFEFEULL }; DECLARE_ALIGNED(16, const double, ff_pd_1)[2] = { 1.0, 1.0 }; DECLARE_ALIGNED(16, const double, ff_pd_2)[2] = { 2.0, 2.0 }; #if HAVE_YASM -void ff_put_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_put_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); void ff_put_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h); void ff_put_no_rnd_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, @@ -94,54 +67,14 @@ void ff_put_no_rnd_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, int src1Stride, int h); void ff_avg_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h); -void ff_put_pixels16_x2_mmxext(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_put_pixels16_x2_3dnow(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); void ff_put_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h); void ff_avg_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h); void ff_put_no_rnd_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h); -void ff_put_no_rnd_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_put_no_rnd_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_put_no_rnd_pixels8_x2_exact_mmxext(uint8_t *block, - const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_put_no_rnd_pixels8_x2_exact_3dnow(uint8_t *block, - const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_put_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_put_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_put_no_rnd_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_put_no_rnd_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_put_no_rnd_pixels8_y2_exact_mmxext(uint8_t *block, - const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_put_no_rnd_pixels8_y2_exact_3dnow(uint8_t *block, - const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_avg_pixels8_3dnow(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_avg_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_avg_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_avg_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_avg_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_avg_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); -void ff_avg_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels, - ptrdiff_t line_size, int h); +void ff_avg_pixels8_mmxext(uint8_t *block, const uint8_t *pixels, + ptrdiff_t line_size, int h); void ff_put_pixels8_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h); static void ff_put_pixels16_mmxext(uint8_t *block, const uint8_t *pixels, @@ -215,14 +148,6 @@ void ff_put_no_rnd_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, // using regr as temporary and for the output result // first argument is unmodifed and second is trashed // regfe is supposed to contain 0xfefefefefefefefe -#define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \ - "movq "#rega", "#regr" \n\t" \ - "pand "#regb", "#regr" \n\t" \ - "pxor "#rega", "#regb" \n\t" \ - "pand "#regfe", "#regb" \n\t" \ - "psrlq $1, "#regb" \n\t" \ - "paddb "#regb", "#regr" \n\t" - #define PAVGB_MMX(rega, regb, regr, regfe) \ "movq "#rega", "#regr" \n\t" \ "por "#regb", "#regr" \n\t" \ @@ -232,20 +157,6 @@ void ff_put_no_rnd_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, "psubb "#regb", "#regr" \n\t" // mm6 is supposed to contain 0xfefefefefefefefe -#define PAVGBP_MMX_NO_RND(rega, regb, regr, regc, regd, regp) \ - "movq "#rega", "#regr" \n\t" \ - "movq "#regc", "#regp" \n\t" \ - "pand "#regb", "#regr" \n\t" \ - "pand "#regd", "#regp" \n\t" \ - "pxor "#rega", "#regb" \n\t" \ - "pxor "#regc", "#regd" \n\t" \ - "pand %%mm6, "#regb" \n\t" \ - "pand %%mm6, "#regd" \n\t" \ - "psrlq $1, "#regb" \n\t" \ - "psrlq $1, "#regd" \n\t" \ - "paddb "#regb", "#regr" \n\t" \ - "paddb "#regd", "#regp" \n\t" - #define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \ "movq "#rega", "#regr" \n\t" \ "movq "#regc", "#regp" \n\t" \ @@ -261,28 +172,13 @@ void ff_put_no_rnd_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, "psubb "#regd", "#regp" \n\t" /***********************************/ -/* MMX no rounding */ -#define NO_RND 1 -#define DEF(x, y) x ## _no_rnd_ ## y ## _mmx -#define SET_RND MOVQ_WONE -#define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX_NO_RND(a, b, c, d, e, f) -#define PAVGB(a, b, c, e) PAVGB_MMX_NO_RND(a, b, c, e) -#define OP_AVG(a, b, c, e) PAVGB_MMX(a, b, c, e) - -#include "dsputil_rnd_template.c" - -#undef DEF -#undef SET_RND -#undef PAVGBP -#undef PAVGB -#undef NO_RND -/***********************************/ /* MMX rounding */ #define DEF(x, y) x ## _ ## y ## _mmx #define SET_RND MOVQ_WTWO #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX(a, b, c, d, e, f) #define PAVGB(a, b, c, e) PAVGB_MMX(a, b, c, e) +#define OP_AVG(a, b, c, e) PAVGB_MMX(a, b, c, e) #include "dsputil_rnd_template.c" @@ -298,30 +194,20 @@ void ff_put_no_rnd_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, #if HAVE_YASM /***********************************/ -/* 3Dnow specific */ - -#define DEF(x) x ## _3dnow - -#include "dsputil_avg_template.c" - -#undef DEF - -/***********************************/ /* MMXEXT specific */ -#define DEF(x) x ## _mmxext - -#include "dsputil_avg_template.c" - -#undef DEF +//FIXME the following could be optimized too ... +static void ff_avg_pixels16_mmxext(uint8_t *block, const uint8_t *pixels, + int line_size, int h) +{ + ff_avg_pixels8_mmxext(block, pixels, line_size, h); + ff_avg_pixels8_mmxext(block + 8, pixels + 8, line_size, h); +} #endif /* HAVE_YASM */ #if HAVE_INLINE_ASM -#define put_no_rnd_pixels16_mmx put_pixels16_mmx -#define put_no_rnd_pixels8_mmx put_pixels8_mmx - /***********************************/ /* standard MMX */ @@ -679,7 +565,7 @@ static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, : "+r"(ptr) : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height) ); - } else { + } else if(w==16){ __asm__ volatile ( "1: \n\t" "movd (%0), %%mm0 \n\t" @@ -700,6 +586,25 @@ static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, : "+r"(ptr) : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height) ); + } else { + av_assert1(w == 4); + __asm__ volatile ( + "1: \n\t" + "movd (%0), %%mm0 \n\t" + "punpcklbw %%mm0, %%mm0 \n\t" + "punpcklwd %%mm0, %%mm0 \n\t" + "movd %%mm0, -4(%0) \n\t" + "movd -4(%0, %2), %%mm1 \n\t" + "punpcklbw %%mm1, %%mm1 \n\t" + "punpckhwd %%mm1, %%mm1 \n\t" + "punpckhdq %%mm1, %%mm1 \n\t" + "movd %%mm1, (%0, %2) \n\t" + "add %1, %0 \n\t" + "cmp %3, %0 \n\t" + "jb 1b \n\t" + : "+r"(ptr) + : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height) + ); } /* top and bottom (and hopefully also the corners) */ @@ -1153,10 +1058,15 @@ void ff_avg_rv40_qpel16_mc33_mmx(uint8_t *dst, uint8_t *src, ptrdiff_t stride) avg_pixels16_xy2_mmx(dst, src, stride, 16); } -static void gmc_mmx(uint8_t *dst, uint8_t *src, - int stride, int h, int ox, int oy, - int dxx, int dxy, int dyx, int dyy, - int shift, int r, int width, int height) +typedef void emulated_edge_mc_func(uint8_t *dst, const uint8_t *src, + ptrdiff_t linesize, int block_w, int block_h, + int src_x, int src_y, int w, int h); + +static av_always_inline void gmc(uint8_t *dst, uint8_t *src, + int stride, int h, int ox, int oy, + int dxx, int dxy, int dyx, int dyy, + int shift, int r, int width, int height, + emulated_edge_mc_func *emu_edge_fn) { const int w = 8; const int ix = ox >> (16 + shift); @@ -1171,19 +1081,24 @@ static void gmc_mmx(uint8_t *dst, uint8_t *src, const uint16_t dxy4[4] = { dxys, dxys, dxys, dxys }; const uint16_t dyy4[4] = { dyys, dyys, dyys, dyys }; const uint64_t shift2 = 2 * shift; +#define MAX_STRIDE 4096U +#define MAX_H 8U + uint8_t edge_buf[(MAX_H + 1) * MAX_STRIDE]; int x, y; const int dxw = (dxx - (1 << (16 + shift))) * (w - 1); const int dyh = (dyy - (1 << (16 + shift))) * (h - 1); const int dxh = dxy * (h - 1); const int dyw = dyx * (w - 1); + int need_emu = (unsigned)ix >= width - w || + (unsigned)iy >= height - h; + if ( // non-constant fullpel offset (3% of blocks) ((ox ^ (ox + dxw)) | (ox ^ (ox + dxh)) | (ox ^ (ox + dxw + dxh)) | (oy ^ (oy + dyw)) | (oy ^ (oy + dyh)) | (oy ^ (oy + dyw + dyh))) >> (16 + shift) // uses more than 16 bits of subpel mv (only at huge resolution) - || (dxx | dxy | dyx | dyy) & 15 || - (unsigned)ix >= width - w || - (unsigned)iy >= height - h) { + || (dxx | dxy | dyx | dyy) & 15 + || (need_emu && (h > MAX_H || stride > MAX_STRIDE))) { // FIXME could still use mmx for some of the rows ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, width, height); @@ -1191,6 +1106,10 @@ static void gmc_mmx(uint8_t *dst, uint8_t *src, } src += ix + iy * stride; + if (need_emu) { + emu_edge_fn(edge_buf, src, stride, w + 1, h + 1, ix, iy, width, height); + src = edge_buf; + } __asm__ volatile ( "movd %0, %%mm6 \n\t" @@ -1268,6 +1187,39 @@ static void gmc_mmx(uint8_t *dst, uint8_t *src, src += 4 - h * stride; } } + +#if CONFIG_VIDEODSP +#if HAVE_YASM +#if ARCH_X86_32 +static void gmc_mmx(uint8_t *dst, uint8_t *src, + int stride, int h, int ox, int oy, + int dxx, int dxy, int dyx, int dyy, + int shift, int r, int width, int height) +{ + gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, + width, height, &ff_emulated_edge_mc_8); +} +#endif +static void gmc_sse(uint8_t *dst, uint8_t *src, + int stride, int h, int ox, int oy, + int dxx, int dxy, int dyx, int dyy, + int shift, int r, int width, int height) +{ + gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, + width, height, &ff_emulated_edge_mc_8); +} +#else +static void gmc_mmx(uint8_t *dst, uint8_t *src, + int stride, int h, int ox, int oy, + int dxx, int dxy, int dyx, int dyy, + int shift, int r, int width, int height) +{ + gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, + width, height, &ff_emulated_edge_mc_8); +} +#endif +#endif + #endif /* HAVE_INLINE_ASM */ void ff_put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, @@ -1305,6 +1257,75 @@ void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, put_pixels8_mmx(dst, src, stride, 8); } +#if CONFIG_DIRAC_DECODER +#define DIRAC_PIXOP(OPNAME2, OPNAME, EXT)\ +void ff_ ## OPNAME2 ## _dirac_pixels8_ ## EXT(uint8_t *dst, const uint8_t *src[5], int stride, int h)\ +{\ + if (h&3)\ + ff_ ## OPNAME2 ## _dirac_pixels8_c(dst, src, stride, h);\ + else\ + OPNAME ## _pixels8_ ## EXT(dst, src[0], stride, h);\ +}\ +void ff_ ## OPNAME2 ## _dirac_pixels16_ ## EXT(uint8_t *dst, const uint8_t *src[5], int stride, int h)\ +{\ + if (h&3)\ + ff_ ## OPNAME2 ## _dirac_pixels16_c(dst, src, stride, h);\ + else\ + OPNAME ## _pixels16_ ## EXT(dst, src[0], stride, h);\ +}\ +void ff_ ## OPNAME2 ## _dirac_pixels32_ ## EXT(uint8_t *dst, const uint8_t *src[5], int stride, int h)\ +{\ + if (h&3) {\ + ff_ ## OPNAME2 ## _dirac_pixels32_c(dst, src, stride, h);\ + } else {\ + OPNAME ## _pixels16_ ## EXT(dst , src[0] , stride, h);\ + OPNAME ## _pixels16_ ## EXT(dst+16, src[0]+16, stride, h);\ + }\ +} + +#if HAVE_MMX_INLINE +DIRAC_PIXOP(put, put, mmx) +DIRAC_PIXOP(avg, avg, mmx) +#endif + +#if HAVE_YASM +DIRAC_PIXOP(avg, ff_avg, mmxext) + +void ff_put_dirac_pixels16_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h) +{ + if (h&3) + ff_put_dirac_pixels16_c(dst, src, stride, h); + else + ff_put_pixels16_sse2(dst, src[0], stride, h); +} +void ff_avg_dirac_pixels16_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h) +{ + if (h&3) + ff_avg_dirac_pixels16_c(dst, src, stride, h); + else + ff_avg_pixels16_sse2(dst, src[0], stride, h); +} +void ff_put_dirac_pixels32_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h) +{ + if (h&3) { + ff_put_dirac_pixels32_c(dst, src, stride, h); + } else { + ff_put_pixels16_sse2(dst , src[0] , stride, h); + ff_put_pixels16_sse2(dst+16, src[0]+16, stride, h); + } +} +void ff_avg_dirac_pixels32_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h) +{ + if (h&3) { + ff_avg_dirac_pixels32_c(dst, src, stride, h); + } else { + ff_avg_pixels16_sse2(dst , src[0] , stride, h); + ff_avg_pixels16_sse2(dst+16, src[0]+16, stride, h); + } +} +#endif +#endif + static void vector_clipf_sse(float *dst, const float *src, float min, float max, int len) { @@ -1408,14 +1429,6 @@ void ff_vector_clip_int32_sse4 (int32_t *dst, const int32_t *src, c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \ } while (0) -#define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \ - do { \ - c->PFX ## _pixels_tab IDX [0] = PFX ## _pixels ## SIZE ## _ ## CPU; \ - c->PFX ## _pixels_tab IDX [1] = PFX ## _pixels ## SIZE ## _x2_ ## CPU; \ - c->PFX ## _pixels_tab IDX [2] = PFX ## _pixels ## SIZE ## _y2_ ## CPU; \ - c->PFX ## _pixels_tab IDX [3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU; \ - } while (0) - static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx, int mm_flags) { @@ -1430,32 +1443,11 @@ static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx, c->clear_block = clear_block_mmx; c->clear_blocks = clear_blocks_mmx; c->draw_edges = draw_edges_mmx; - - SET_HPEL_FUNCS(put, [0], 16, mmx); - SET_HPEL_FUNCS(put_no_rnd, [0], 16, mmx); - SET_HPEL_FUNCS(avg, [0], 16, mmx); - SET_HPEL_FUNCS(avg_no_rnd, , 16, mmx); - SET_HPEL_FUNCS(put, [1], 8, mmx); - SET_HPEL_FUNCS(put_no_rnd, [1], 8, mmx); - SET_HPEL_FUNCS(avg, [1], 8, mmx); - - switch (avctx->idct_algo) { - case FF_IDCT_AUTO: - case FF_IDCT_SIMPLEMMX: - c->idct_put = ff_simple_idct_put_mmx; - c->idct_add = ff_simple_idct_add_mmx; - c->idct = ff_simple_idct_mmx; - c->idct_permutation_type = FF_SIMPLE_IDCT_PERM; - break; - case FF_IDCT_XVIDMMX: - c->idct_put = ff_idct_xvid_mmx_put; - c->idct_add = ff_idct_xvid_mmx_add; - c->idct = ff_idct_xvid_mmx; - break; - } } +#if CONFIG_VIDEODSP && (ARCH_X86_32 || !HAVE_YASM) c->gmc = gmc_mmx; +#endif c->add_bytes = add_bytes_mmx; #endif /* HAVE_INLINE_ASM */ @@ -1474,8 +1466,6 @@ static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx, static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx, int mm_flags) { - const int bit_depth = avctx->bits_per_raw_sample; - const int high_bit_depth = bit_depth > 8; #if HAVE_YASM SET_QPEL_FUNCS(avg_qpel, 0, 16, mmxext, ); @@ -1485,51 +1475,9 @@ static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx, SET_QPEL_FUNCS(put_qpel, 1, 8, mmxext, ); SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmxext, ); SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, mmxext, ); - - if (!high_bit_depth) { - c->put_pixels_tab[0][1] = ff_put_pixels16_x2_mmxext; - c->put_pixels_tab[0][2] = ff_put_pixels16_y2_mmxext; - - c->avg_pixels_tab[0][0] = ff_avg_pixels16_mmxext; - c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_mmxext; - c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_mmxext; - - c->put_pixels_tab[1][1] = ff_put_pixels8_x2_mmxext; - c->put_pixels_tab[1][2] = ff_put_pixels8_y2_mmxext; - - c->avg_pixels_tab[1][0] = ff_avg_pixels8_mmxext; - c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_mmxext; - c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_mmxext; - } - - if (!(avctx->flags & CODEC_FLAG_BITEXACT)) { - if (!high_bit_depth) { - c->put_no_rnd_pixels_tab[0][1] = ff_put_no_rnd_pixels16_x2_mmxext; - c->put_no_rnd_pixels_tab[0][2] = ff_put_no_rnd_pixels16_y2_mmxext; - c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_mmxext; - c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_mmxext; - - c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_mmxext; - c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_mmxext; - } - } #endif /* HAVE_YASM */ -#if HAVE_INLINE_ASM - if (!high_bit_depth && avctx->idct_algo == FF_IDCT_XVIDMMX) { - c->idct_put = ff_idct_xvid_mmxext_put; - c->idct_add = ff_idct_xvid_mmxext_add; - c->idct = ff_idct_xvid_mmxext; - } -#endif /* HAVE_INLINE_ASM */ - #if HAVE_MMXEXT_EXTERNAL - if (CONFIG_VP3_DECODER && (avctx->codec_id == AV_CODEC_ID_VP3 || - avctx->codec_id == AV_CODEC_ID_THEORA)) { - c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_exact_mmxext; - c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_exact_mmxext; - } - /* slower than cmov version on AMD */ if (!(mm_flags & AV_CPU_FLAG_3DNOW)) c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmxext; @@ -1545,46 +1493,6 @@ static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx, #endif /* HAVE_MMXEXT_EXTERNAL */ } -static av_cold void dsputil_init_3dnow(DSPContext *c, AVCodecContext *avctx, - int mm_flags) -{ - const int high_bit_depth = avctx->bits_per_raw_sample > 8; - -#if HAVE_YASM - if (!high_bit_depth) { - c->put_pixels_tab[0][1] = ff_put_pixels16_x2_3dnow; - c->put_pixels_tab[0][2] = ff_put_pixels16_y2_3dnow; - - c->avg_pixels_tab[0][0] = ff_avg_pixels16_3dnow; - c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_3dnow; - c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_3dnow; - - c->put_pixels_tab[1][1] = ff_put_pixels8_x2_3dnow; - c->put_pixels_tab[1][2] = ff_put_pixels8_y2_3dnow; - - c->avg_pixels_tab[1][0] = ff_avg_pixels8_3dnow; - c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_3dnow; - c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_3dnow; - - if (!(avctx->flags & CODEC_FLAG_BITEXACT)){ - c->put_no_rnd_pixels_tab[0][1] = ff_put_no_rnd_pixels16_x2_3dnow; - c->put_no_rnd_pixels_tab[0][2] = ff_put_no_rnd_pixels16_y2_3dnow; - c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_3dnow; - c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_3dnow; - - c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_3dnow; - c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_3dnow; - } - } - - if (CONFIG_VP3_DECODER && (avctx->codec_id == AV_CODEC_ID_VP3 || - avctx->codec_id == AV_CODEC_ID_THEORA)) { - c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_exact_3dnow; - c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_exact_3dnow; - } -#endif /* HAVE_YASM */ -} - static av_cold void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx, int mm_flags) { @@ -1601,6 +1509,12 @@ static av_cold void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx, c->vector_clipf = vector_clipf_sse; #endif /* HAVE_INLINE_ASM */ + +#if HAVE_YASM +#if HAVE_INLINE_ASM && CONFIG_VIDEODSP + c->gmc = gmc_sse; +#endif +#endif /* HAVE_YASM */ } static av_cold void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx, @@ -1619,15 +1533,6 @@ static av_cold void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx, #endif /* HAVE_SSE2_INLINE */ #if HAVE_SSE2_EXTERNAL - if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) { - // these functions are slower than mmx on AMD, but faster on Intel - if (!high_bit_depth) { - c->put_pixels_tab[0][0] = ff_put_pixels16_sse2; - c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_sse2; - c->avg_pixels_tab[0][0] = ff_avg_pixels16_sse2; - } - } - c->scalarproduct_int16 = ff_scalarproduct_int16_sse2; c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_sse2; if (mm_flags & AV_CPU_FLAG_ATOM) { @@ -1679,15 +1584,41 @@ av_cold void ff_dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx) c->add_hfyu_median_prediction = add_hfyu_median_prediction_cmov; #endif - if (mm_flags & AV_CPU_FLAG_MMX) + if (mm_flags & AV_CPU_FLAG_MMX) { +#if HAVE_INLINE_ASM + const int idct_algo = avctx->idct_algo; + + if (avctx->lowres == 0 && avctx->bits_per_raw_sample <= 8) { + if (idct_algo == FF_IDCT_AUTO || idct_algo == FF_IDCT_SIMPLEMMX) { + c->idct_put = ff_simple_idct_put_mmx; + c->idct_add = ff_simple_idct_add_mmx; + c->idct = ff_simple_idct_mmx; + c->idct_permutation_type = FF_SIMPLE_IDCT_PERM; + } else if (idct_algo == FF_IDCT_XVIDMMX) { + if (mm_flags & AV_CPU_FLAG_SSE2) { + c->idct_put = ff_idct_xvid_sse2_put; + c->idct_add = ff_idct_xvid_sse2_add; + c->idct = ff_idct_xvid_sse2; + c->idct_permutation_type = FF_SSE2_IDCT_PERM; + } else if (mm_flags & AV_CPU_FLAG_MMXEXT) { + c->idct_put = ff_idct_xvid_mmxext_put; + c->idct_add = ff_idct_xvid_mmxext_add; + c->idct = ff_idct_xvid_mmxext; + } else { + c->idct_put = ff_idct_xvid_mmx_put; + c->idct_add = ff_idct_xvid_mmx_add; + c->idct = ff_idct_xvid_mmx; + } + } + } +#endif /* HAVE_INLINE_ASM */ + dsputil_init_mmx(c, avctx, mm_flags); + } if (mm_flags & AV_CPU_FLAG_MMXEXT) dsputil_init_mmxext(c, avctx, mm_flags); - if (mm_flags & AV_CPU_FLAG_3DNOW) - dsputil_init_3dnow(c, avctx, mm_flags); - if (mm_flags & AV_CPU_FLAG_SSE) dsputil_init_sse(c, avctx, mm_flags); |