diff --git a/libavcodec/avcodec.h b/libavcodec/avcodec.h index 08dd8c99f0117376dca70a1953670c3b36b6b394..ae96558c95020e306d020fcbc47c8c5e8e3ee503 100644 --- a/libavcodec/avcodec.h +++ b/libavcodec/avcodec.h @@ -5,8 +5,8 @@ #define LIBAVCODEC_VERSION_INT 0x000406 #define LIBAVCODEC_VERSION "0.4.6" -#define LIBAVCODEC_BUILD 4651 -#define LIBAVCODEC_BUILD_STR "4651" +#define LIBAVCODEC_BUILD 4652 +#define LIBAVCODEC_BUILD_STR "4652" enum CodecID { CODEC_ID_NONE, @@ -909,7 +909,7 @@ typedef struct AVCodecContext { * decoding: unused */ int me_pre_cmp; - + /** * ME pre pass diamond size & shape * encoding: set by user. @@ -917,6 +917,13 @@ typedef struct AVCodecContext { */ int pre_dia_size; + /** + * subpel ME quality + * encoding: set by user. + * decoding: unused + */ + int me_subpel_quality; + } AVCodecContext; typedef struct AVCodec { diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c index 80ff8f4022a1961a43c0dce4546ccd7b3d4af913..0d7556f6579a42085eace53e099ddf63e39e0151 100644 --- a/libavcodec/dsputil.c +++ b/libavcodec/dsputil.c @@ -781,6 +781,7 @@ static inline void copy_block9(UINT8 *dst, UINT8 *src, int dstStride, int srcStr } } + #define QPEL_MC(r, OPNAME, RND, OP) \ static void OPNAME ## mpeg4_qpel8_h_lowpass(UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int h){\ UINT8 *cm = cropTbl + MAX_NEG_CROP;\ @@ -830,6 +831,7 @@ static void OPNAME ## mpeg4_qpel8_v_lowpass(UINT8 *dst, UINT8 *src, int dstStrid static void OPNAME ## mpeg4_qpel16_h_lowpass(UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int h){\ UINT8 *cm = cropTbl + MAX_NEG_CROP;\ int i;\ + \ for(i=0; i>5]+1)>>1) diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h index b2cac91c9c286b9b9f84b6f99b34653fe4144c0e..a672f89863d2f21dfd8d9689e223630b6ea7a1b6 100644 --- a/libavcodec/dsputil.h +++ b/libavcodec/dsputil.h @@ -102,6 +102,7 @@ typedef struct DSPContext { me_cmp_func quant_psnr[2]; int (*hadamard8_abs )(uint8_t *src, int stride, int mean); + me_cmp_func me_pre_cmp[11]; me_cmp_func me_cmp[11]; me_cmp_func me_sub_cmp[11]; me_cmp_func mb_cmp[11]; diff --git a/libavcodec/i386/dsputil_mmx.c b/libavcodec/i386/dsputil_mmx.c index b9ebc311364d3875e83c153f452fc6412a7cafd9..b6795c035c5137d87fa5ac4351eb3400cb1510a2 100644 --- a/libavcodec/i386/dsputil_mmx.c +++ b/libavcodec/i386/dsputil_mmx.c @@ -53,6 +53,11 @@ static const uint64_t mm_bone __attribute__ ((aligned(8))) = 0x0101010101010101U static const uint64_t mm_wone __attribute__ ((aligned(8))) = 0x0001000100010001ULL; static const uint64_t mm_wtwo __attribute__ ((aligned(8))) = 0x0002000200020002ULL; +static const uint64_t ff_pw_20 __attribute__ ((aligned(8))) = 0x0014001400140014ULL; +static const uint64_t ff_pw_3 __attribute__ ((aligned(8))) = 0x0003000300030003ULL; +static const uint64_t ff_pw_16 __attribute__ ((aligned(8))) = 0x0010001000100010ULL; +static const uint64_t ff_pw_15 __attribute__ ((aligned(8))) = 0x000F000F000F000FULL; + #define JUMPALIGN() __asm __volatile (".balign 8"::) #define MOVQ_ZERO(regd) __asm __volatile ("pxor %%" #regd ", %%" #regd ::) @@ -646,10 +651,698 @@ static int hadamard8_diff_mmx(void *s, uint8_t *src1, uint8_t *src2, int stride) WARPER88_1616(hadamard8_diff_mmx, hadamard8_diff16_mmx) +#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\ + "paddw " #m4 ", " #m3 " \n\t" /* x1 */\ + "movq " #pw_20 ", %%mm4 \n\t" /* 20 */\ + "pmullw " #m3 ", %%mm4 \n\t" /* 20x1 */\ + "movq "#in7", " #m3 " \n\t" /* d */\ + "movq "#in0", %%mm5 \n\t" /* D */\ + "paddw " #m3 ", %%mm5 \n\t" /* x4 */\ + "psubw %%mm5, %%mm4 \n\t" /* 20x1 - x4 */\ + "movq "#in1", %%mm5 \n\t" /* C */\ + "movq "#in2", %%mm6 \n\t" /* B */\ + "paddw " #m6 ", %%mm5 \n\t" /* x3 */\ + "paddw " #m5 ", %%mm6 \n\t" /* x2 */\ + "paddw %%mm6, %%mm6 \n\t" /* 2x2 */\ + "psubw %%mm6, %%mm5 \n\t" /* -2x2 + x3 */\ + "pmullw " #pw_3 ", %%mm5 \n\t" /* -6x2 + 3x3 */\ + "paddw " #rnd ", %%mm4 \n\t" /* x2 */\ + "paddw %%mm4, %%mm5 \n\t" /* 20x1 - 6x2 + 3x3 - x4 */\ + "psraw $5, %%mm5 \n\t"\ + "packuswb %%mm5, %%mm5 \n\t"\ + OP(%%mm5, out, %%mm7, d) + +#define QPEL_BASE(OPNAME, ROUNDER, RND, OP)\ +void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ + uint64_t temp;\ +\ + asm volatile(\ + "pxor %%mm7, %%mm7 \n\t"\ + "1: \n\t"\ + "movq (%0), %%mm0 \n\t" /* ABCDEFGH */\ + "movq %%mm0, %%mm1 \n\t" /* ABCDEFGH */\ + "movq %%mm0, %%mm2 \n\t" /* ABCDEFGH */\ + "punpcklbw %%mm7, %%mm0 \n\t" /* 0A0B0C0D */\ + "punpckhbw %%mm7, %%mm1 \n\t" /* 0E0F0G0H */\ + "pshufw $0x90, %%mm0, %%mm5 \n\t" /* 0A0A0B0C */\ + "pshufw $0x41, %%mm0, %%mm6 \n\t" /* 0B0A0A0B */\ + "movq %%mm2, %%mm3 \n\t" /* ABCDEFGH */\ + "movq %%mm2, %%mm4 \n\t" /* ABCDEFGH */\ + "psllq $8, %%mm2 \n\t" /* 0ABCDEFG */\ + "psllq $16, %%mm3 \n\t" /* 00ABCDEF */\ + "psllq $24, %%mm4 \n\t" /* 000ABCDE */\ + "punpckhbw %%mm7, %%mm2 \n\t" /* 0D0E0F0G */\ + "punpckhbw %%mm7, %%mm3 \n\t" /* 0C0D0E0F */\ + "punpckhbw %%mm7, %%mm4 \n\t" /* 0B0C0D0E */\ + "paddw %%mm3, %%mm5 \n\t" /* b */\ + "paddw %%mm2, %%mm6 \n\t" /* c */\ + "paddw %%mm5, %%mm5 \n\t" /* 2b */\ + "psubw %%mm5, %%mm6 \n\t" /* c - 2b */\ + "pshufw $0x06, %%mm0, %%mm5 \n\t" /* 0C0B0A0A */\ + "pmullw %6, %%mm6 \n\t" /* 3c - 6b */\ + "paddw %%mm4, %%mm0 \n\t" /* a */\ + "paddw %%mm1, %%mm5 \n\t" /* d */\ + "pmullw %5, %%mm0 \n\t" /* 20a */\ + "psubw %%mm5, %%mm0 \n\t" /* 20a - d */\ + "paddw %8, %%mm6 \n\t"\ + "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\ + "psraw $5, %%mm0 \n\t"\ + "movq %%mm0, %7 \n\t"\ + /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\ + \ + "movq 5(%0), %%mm0 \n\t" /* FGHIJKLM */\ + "movq %%mm0, %%mm5 \n\t" /* FGHIJKLM */\ + "movq %%mm0, %%mm6 \n\t" /* FGHIJKLM */\ + "psrlq $8, %%mm0 \n\t" /* GHIJKLM0 */\ + "psrlq $16, %%mm5 \n\t" /* HIJKLM00 */\ + "punpcklbw %%mm7, %%mm0 \n\t" /* 0G0H0I0J */\ + "punpcklbw %%mm7, %%mm5 \n\t" /* 0H0I0J0K */\ + "paddw %%mm0, %%mm2 \n\t" /* b */\ + "paddw %%mm5, %%mm3 \n\t" /* c */\ + "paddw %%mm2, %%mm2 \n\t" /* 2b */\ + "psubw %%mm2, %%mm3 \n\t" /* c - 2b */\ + "movq %%mm6, %%mm2 \n\t" /* FGHIJKLM */\ + "psrlq $24, %%mm6 \n\t" /* IJKLM000 */\ + "punpcklbw %%mm7, %%mm2 \n\t" /* 0F0G0H0I */\ + "punpcklbw %%mm7, %%mm6 \n\t" /* 0I0J0K0L */\ + "pmullw %6, %%mm3 \n\t" /* 3c - 6b */\ + "paddw %%mm2, %%mm1 \n\t" /* a */\ + "paddw %%mm6, %%mm4 \n\t" /* d */\ + "pmullw %5, %%mm1 \n\t" /* 20a */\ + "psubw %%mm4, %%mm3 \n\t" /* - 6b +3c - d */\ + "paddw %8, %%mm1 \n\t"\ + "paddw %%mm1, %%mm3 \n\t" /* 20a - 6b +3c - d */\ + "psraw $5, %%mm3 \n\t"\ + "movq %7, %%mm1 \n\t"\ + "packuswb %%mm3, %%mm1 \n\t"\ + OP(%%mm1, (%1),%%mm4, q)\ + /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\ + \ + "movq 9(%0), %%mm1 \n\t" /* JKLMNOPQ */\ + "movq %%mm1, %%mm4 \n\t" /* JKLMNOPQ */\ + "movq %%mm1, %%mm3 \n\t" /* JKLMNOPQ */\ + "psrlq $8, %%mm1 \n\t" /* KLMNOPQ0 */\ + "psrlq $16, %%mm4 \n\t" /* LMNOPQ00 */\ + "punpcklbw %%mm7, %%mm1 \n\t" /* 0K0L0M0N */\ + "punpcklbw %%mm7, %%mm4 \n\t" /* 0L0M0N0O */\ + "paddw %%mm1, %%mm5 \n\t" /* b */\ + "paddw %%mm4, %%mm0 \n\t" /* c */\ + "paddw %%mm5, %%mm5 \n\t" /* 2b */\ + "psubw %%mm5, %%mm0 \n\t" /* c - 2b */\ + "movq %%mm3, %%mm5 \n\t" /* JKLMNOPQ */\ + "psrlq $24, %%mm3 \n\t" /* MNOPQ000 */\ + "pmullw %6, %%mm0 \n\t" /* 3c - 6b */\ + "punpcklbw %%mm7, %%mm3 \n\t" /* 0M0N0O0P */\ + "paddw %%mm3, %%mm2 \n\t" /* d */\ + "psubw %%mm2, %%mm0 \n\t" /* -6b + 3c - d */\ + "movq %%mm5, %%mm2 \n\t" /* JKLMNOPQ */\ + "punpcklbw %%mm7, %%mm2 \n\t" /* 0J0K0L0M */\ + "punpckhbw %%mm7, %%mm5 \n\t" /* 0N0O0P0Q */\ + "paddw %%mm2, %%mm6 \n\t" /* a */\ + "pmullw %5, %%mm6 \n\t" /* 20a */\ + "paddw %8, %%mm0 \n\t"\ + "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\ + "psraw $5, %%mm0 \n\t"\ + /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\ + \ + "paddw %%mm5, %%mm3 \n\t" /* a */\ + "pshufw $0xF9, %%mm5, %%mm6 \n\t" /* 0O0P0Q0Q */\ + "paddw %%mm4, %%mm6 \n\t" /* b */\ + "pshufw $0xBE, %%mm5, %%mm4 \n\t" /* 0P0Q0Q0P */\ + "pshufw $0x6F, %%mm5, %%mm5 \n\t" /* 0Q0Q0P0O */\ + "paddw %%mm1, %%mm4 \n\t" /* c */\ + "paddw %%mm2, %%mm5 \n\t" /* d */\ + "paddw %%mm6, %%mm6 \n\t" /* 2b */\ + "psubw %%mm6, %%mm4 \n\t" /* c - 2b */\ + "pmullw %5, %%mm3 \n\t" /* 20a */\ + "pmullw %6, %%mm4 \n\t" /* 3c - 6b */\ + "psubw %%mm5, %%mm3 \n\t" /* -6b + 3c - d */\ + "paddw %8, %%mm4 \n\t"\ + "paddw %%mm3, %%mm4 \n\t" /* 20a - 6b + 3c - d */\ + "psraw $5, %%mm4 \n\t"\ + "packuswb %%mm4, %%mm0 \n\t"\ + OP(%%mm0, 8(%1), %%mm4, q)\ + \ + "addl %3, %0 \n\t"\ + "addl %4, %1 \n\t"\ + "decl %2 \n\t"\ + " jnz 1b \n\t"\ + : "+r"(src), "+r"(dst), "+g"(h)\ + : "r"(srcStride), "r"(dstStride), "m"(ff_pw_20), "m"(ff_pw_3), "m"(temp), "m"(ROUNDER)\ + );\ +}\ +\ +static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ + int i;\ + int16_t temp[16];\ + /* quick HACK, XXX FIXME MUST be optimized */\ + for(i=0; iput_ ## postfix1 = put_ ## postfix2;\ + c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\ + c->avg_ ## postfix1 = avg_ ## postfix2; + void dsputil_init_mmx(DSPContext* c, unsigned mask) { mm_flags = mm_support(); @@ -724,7 +1417,7 @@ void dsputil_init_mmx(DSPContext* c, unsigned mask) c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx; c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx; c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx; - + c->add_bytes= add_bytes_mmx; c->diff_bytes= diff_bytes_mmx; @@ -767,6 +1460,38 @@ void dsputil_init_mmx(DSPContext* c, unsigned mask) c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2; c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2; c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2; + SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_mmx2) + SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_mmx2) } else if (mm_flags & MM_3DNOW) { c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow; c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow; @@ -787,6 +1512,39 @@ void dsputil_init_mmx(DSPContext* c, unsigned mask) c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow; c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow; c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow; + + SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow) + SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow) } } diff --git a/libavcodec/i386/dsputil_mmx_avg.h b/libavcodec/i386/dsputil_mmx_avg.h index 6873432ce8bd248025596adc55e0450c2075794d..4a8841156856ed1dc2c518f26475ce2e41dadb25 100644 --- a/libavcodec/i386/dsputil_mmx_avg.h +++ b/libavcodec/i386/dsputil_mmx_avg.h @@ -53,6 +53,38 @@ static void DEF(put_pixels8_x2)(UINT8 *block, const UINT8 *pixels, int line_size :"%eax", "memory"); } +static void DEF(put_pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h) +{ + __asm __volatile( + "1: \n\t" + "movq (%1), %%mm0 \n\t" + "addl %4, %1 \n\t" + "movq (%1), %%mm1 \n\t" + "addl %4, %1 \n\t" + PAVGB" (%2), %%mm0 \n\t" + PAVGB" 8(%2), %%mm1 \n\t" + "movq %%mm0, (%3) \n\t" + "addl %5, %3 \n\t" + "movq %%mm1, (%3) \n\t" + "addl %5, %3 \n\t" + "movq (%1), %%mm0 \n\t" + "addl %4, %1 \n\t" + "movq (%1), %%mm1 \n\t" + "addl %4, %1 \n\t" + PAVGB" 16(%2), %%mm0 \n\t" + PAVGB" 24(%2), %%mm1 \n\t" + "movq %%mm0, (%3) \n\t" + "addl %5, %3 \n\t" + "movq %%mm1, (%3) \n\t" + "addl %5, %3 \n\t" + "addl $32, %2 \n\t" + "subl $4, %0 \n\t" + "jnz 1b \n\t" + :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst) + :"r"(src1Stride), "r"(dstStride) + :"memory"); +} + static void DEF(put_pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h) { __asm __volatile( @@ -92,6 +124,34 @@ static void DEF(put_pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_siz :"r" (line_size) :"%eax", "memory"); } + +static void DEF(put_pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h) +{ + __asm __volatile( + "1: \n\t" + "movq (%1), %%mm0 \n\t" + "movq 8(%1), %%mm1 \n\t" + "addl %4, %1 \n\t" + PAVGB" (%2), %%mm0 \n\t" + PAVGB" 8(%2), %%mm1 \n\t" + "movq %%mm0, (%3) \n\t" + "movq %%mm1, 8(%3) \n\t" + "addl %5, %3 \n\t" + "movq (%1), %%mm0 \n\t" + "movq 8(%1), %%mm1 \n\t" + "addl %4, %1 \n\t" + PAVGB" 16(%2), %%mm0 \n\t" + PAVGB" 24(%2), %%mm1 \n\t" + "movq %%mm0, (%3) \n\t" + "movq %%mm1, 8(%3) \n\t" + "addl %5, %3 \n\t" + "addl $32, %2 \n\t" + "subl $2, %0 \n\t" + "jnz 1b \n\t" + :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst) + :"r"(src1Stride), "r"(dstStride) + :"memory"); +} /* GL: this function does incorrect rounding if overflow */ static void DEF(put_no_rnd_pixels8_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h) diff --git a/libavcodec/i386/dsputil_mmx_rnd.h b/libavcodec/i386/dsputil_mmx_rnd.h index 3605e03f9ccca2eccf1c3b89d1ee0ddf3208557f..f72ded1bcf4e37027232d527a39ab22b1336962d 100644 --- a/libavcodec/i386/dsputil_mmx_rnd.h +++ b/libavcodec/i386/dsputil_mmx_rnd.h @@ -54,6 +54,42 @@ static void DEF(put, pixels8_x2)(UINT8 *block, const UINT8 *pixels, int line_siz :"eax", "memory"); } +static void DEF(put, pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h) +{ + MOVQ_BFE(mm6); + __asm __volatile( + ".balign 8 \n\t" + "1: \n\t" + "movq (%1), %%mm0 \n\t" + "movq (%2), %%mm1 \n\t" + "addl %4, %1 \n\t" + "movq (%1), %%mm2 \n\t" + "movq 8(%2), %%mm3 \n\t" + "addl %4, %1 \n\t" + PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) + "movq %%mm4, (%3) \n\t" + "addl %5, %3 \n\t" + "movq %%mm5, (%3) \n\t" + "addl %5, %3 \n\t" + "movq (%1), %%mm0 \n\t" + "movq 16(%2), %%mm1 \n\t" + "addl %4, %1 \n\t" + "movq (%1), %%mm2 \n\t" + "movq 24(%2), %%mm3 \n\t" + "addl %4, %1 \n\t" + "addl $32, %2 \n\t" + PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) + "movq %%mm4, (%3) \n\t" + "addl %5, %3 \n\t" + "movq %%mm5, (%3) \n\t" + "addl %5, %3 \n\t" + "subl $4, %0 \n\t" + "jnz 1b \n\t" + :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst) + :"r"(src1Stride), "r"(dstStride) + :"memory"); +} + static void DEF(put, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h) { MOVQ_BFE(mm6); @@ -90,7 +126,7 @@ static void DEF(put, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_si "movq 9(%1, %3), %%mm3 \n\t" PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) "movq %%mm4, 8(%2) \n\t" - "movq %%mm5, 8(%2, %3) \n\t" + "movq %%mm5, 8(%2, %3) \n\t" "addl %%eax, %1 \n\t" "addl %%eax, %2 \n\t" "subl $4, %0 \n\t" @@ -100,6 +136,38 @@ static void DEF(put, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_si :"eax", "memory"); } +static void DEF(put, pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h) +{ + MOVQ_BFE(mm6); + __asm __volatile( + ".balign 8 \n\t" + "1: \n\t" + "movq (%1), %%mm0 \n\t" + "movq (%2), %%mm1 \n\t" + "movq 8(%1), %%mm2 \n\t" + "movq 8(%2), %%mm3 \n\t" + "addl %4, %1 \n\t" + PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) + "movq %%mm4, (%3) \n\t" + "movq %%mm5, 8(%3) \n\t" + "addl %5, %3 \n\t" + "movq (%1), %%mm0 \n\t" + "movq 16(%2), %%mm1 \n\t" + "movq 8(%1), %%mm2 \n\t" + "movq 24(%2), %%mm3 \n\t" + "addl %4, %1 \n\t" + PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) + "movq %%mm4, (%3) \n\t" + "movq %%mm5, 8(%3) \n\t" + "addl %5, %3 \n\t" + "addl $32, %2 \n\t" + "subl $2, %0 \n\t" + "jnz 1b \n\t" + :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst) + :"r"(src1Stride), "r"(dstStride) + :"memory"); +} + static void DEF(put, pixels8_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h) { MOVQ_BFE(mm6); @@ -195,6 +263,124 @@ static void DEF(put, pixels8_xy2)(UINT8 *block, const UINT8 *pixels, int line_si :"eax", "memory"); } +static void DEF(put, pixels8_l4)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int stride, int h) +{ + MOVQ_ZERO(mm7); + SET_RND(mm6); // =2 for rnd and =1 for no_rnd version + __asm __volatile( + ".balign 8 \n\t" + "1: \n\t" + "movq (%1), %%mm0 \n\t" + "movq (%2), %%mm1 \n\t" + "movq 64(%2), %%mm2 \n\t" + "movq 136(%2), %%mm3 \n\t" + "punpcklbw %%mm7, %%mm0 \n\t" + "punpcklbw %%mm7, %%mm1 \n\t" + "punpcklbw %%mm7, %%mm2 \n\t" + "punpcklbw %%mm7, %%mm3 \n\t" + "paddusw %%mm6, %%mm0 \n\t" + "paddusw %%mm0, %%mm1 \n\t" + "paddusw %%mm2, %%mm3 \n\t" + "paddusw %%mm1, %%mm3 \n\t" + "psrlw $2, %%mm3 \n\t" + "movq (%1), %%mm0 \n\t" + "movq (%2), %%mm1 \n\t" + "movq 64(%2), %%mm2 \n\t" + "movq 136(%2), %%mm4 \n\t" + "punpckhbw %%mm7, %%mm0 \n\t" + "punpckhbw %%mm7, %%mm1 \n\t" + "punpckhbw %%mm7, %%mm2 \n\t" + "punpckhbw %%mm7, %%mm4 \n\t" + "paddusw %%mm6, %%mm0 \n\t" + "paddusw %%mm0, %%mm1 \n\t" + "paddusw %%mm2, %%mm4 \n\t" + "paddusw %%mm1, %%mm4 \n\t" + "psrlw $2, %%mm4 \n\t" + "packuswb %%mm4, %%mm3 \n\t" + "movq %%mm3, (%0) \n\t" + "addl %4, %0 \n\t" + "addl %4, %1 \n\t" + "addl $8, %2 \n\t" + "decl %3 \n\t" + "jnz 1b \n\t" + :"+r"(dst), "+r"(src1), "+r"(src2), "+r"(h) + :"r"(stride) + :"memory"); +} + +static void DEF(put, pixels16_l4)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int stride, int h) +{ + MOVQ_ZERO(mm7); + SET_RND(mm6); // =2 for rnd and =1 for no_rnd version + __asm __volatile( + ".balign 8 \n\t" + "1: \n\t" + "movq (%1), %%mm0 \n\t" + "movq (%2), %%mm1 \n\t" + "movq 256(%2), %%mm2 \n\t" + "movq 528(%2), %%mm3 \n\t" + "punpcklbw %%mm7, %%mm0 \n\t" + "punpcklbw %%mm7, %%mm1 \n\t" + "punpcklbw %%mm7, %%mm2 \n\t" + "punpcklbw %%mm7, %%mm3 \n\t" + "paddusw %%mm6, %%mm0 \n\t" + "paddusw %%mm0, %%mm1 \n\t" + "paddusw %%mm2, %%mm3 \n\t" + "paddusw %%mm1, %%mm3 \n\t" + "psrlw $2, %%mm3 \n\t" + "movq (%1), %%mm0 \n\t" + "movq (%2), %%mm1 \n\t" + "movq 256(%2), %%mm2 \n\t" + "movq 528(%2), %%mm4 \n\t" + "punpckhbw %%mm7, %%mm0 \n\t" + "punpckhbw %%mm7, %%mm1 \n\t" + "punpckhbw %%mm7, %%mm2 \n\t" + "punpckhbw %%mm7, %%mm4 \n\t" + "paddusw %%mm6, %%mm0 \n\t" + "paddusw %%mm0, %%mm1 \n\t" + "paddusw %%mm2, %%mm4 \n\t" + "paddusw %%mm1, %%mm4 \n\t" + "psrlw $2, %%mm4 \n\t" + "packuswb %%mm4, %%mm3 \n\t" + "movq %%mm3, (%0) \n\t" + "movq 8(%1), %%mm0 \n\t" + "movq 8(%2), %%mm1 \n\t" + "movq 264(%2), %%mm2 \n\t" + "movq 536(%2), %%mm3 \n\t" + "punpcklbw %%mm7, %%mm0 \n\t" + "punpcklbw %%mm7, %%mm1 \n\t" + "punpcklbw %%mm7, %%mm2 \n\t" + "punpcklbw %%mm7, %%mm3 \n\t" + "paddusw %%mm6, %%mm0 \n\t" + "paddusw %%mm0, %%mm1 \n\t" + "paddusw %%mm2, %%mm3 \n\t" + "paddusw %%mm1, %%mm3 \n\t" + "psrlw $2, %%mm3 \n\t" + "movq 8(%1), %%mm0 \n\t" + "movq 8(%2), %%mm1 \n\t" + "movq 264(%2), %%mm2 \n\t" + "movq 536(%2), %%mm4 \n\t" + "punpckhbw %%mm7, %%mm0 \n\t" + "punpckhbw %%mm7, %%mm1 \n\t" + "punpckhbw %%mm7, %%mm2 \n\t" + "punpckhbw %%mm7, %%mm4 \n\t" + "paddusw %%mm6, %%mm0 \n\t" + "paddusw %%mm0, %%mm1 \n\t" + "paddusw %%mm2, %%mm4 \n\t" + "paddusw %%mm1, %%mm4 \n\t" + "psrlw $2, %%mm4 \n\t" + "packuswb %%mm4, %%mm3 \n\t" + "movq %%mm3, 8(%0) \n\t" + "addl %4, %0 \n\t" + "addl %4, %1 \n\t" + "addl $16, %2 \n\t" + "decl %3 \n\t" + "jnz 1b \n\t" + :"+r"(dst), "+r"(src1), "+r"(src2), "+r"(h) + :"r"(stride) + :"memory"); +} + // avg_pixels // in case more speed is needed - unroling would certainly help static void DEF(avg, pixels8)(UINT8 *block, const UINT8 *pixels, int line_size, int h) @@ -259,6 +445,27 @@ static void DEF(avg, pixels8_x2)(UINT8 *block, const UINT8 *pixels, int line_siz } while (--h); } +static void DEF(avg, pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h) +{ + MOVQ_BFE(mm6); + JUMPALIGN(); + do { + __asm __volatile( + "movq %1, %%mm0 \n\t" + "movq %2, %%mm1 \n\t" + "movq %0, %%mm3 \n\t" + PAVGB(%%mm0, %%mm1, %%mm2, %%mm6) + PAVGB(%%mm3, %%mm2, %%mm0, %%mm6) + "movq %%mm0, %0 \n\t" + :"+m"(*dst) + :"m"(*src1), "m"(*src2) + :"memory"); + dst += dstStride; + src1 += src1Stride; + src2 += 8; + } while (--h); +} + static void DEF(avg, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h) { MOVQ_BFE(mm6); @@ -285,6 +492,33 @@ static void DEF(avg, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_si } while (--h); } +static void DEF(avg, pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h) +{ + MOVQ_BFE(mm6); + JUMPALIGN(); + do { + __asm __volatile( + "movq %1, %%mm0 \n\t" + "movq %2, %%mm1 \n\t" + "movq %0, %%mm3 \n\t" + PAVGB(%%mm0, %%mm1, %%mm2, %%mm6) + PAVGB(%%mm3, %%mm2, %%mm0, %%mm6) + "movq %%mm0, %0 \n\t" + "movq 8%1, %%mm0 \n\t" + "movq 8%2, %%mm1 \n\t" + "movq 8%0, %%mm3 \n\t" + PAVGB(%%mm0, %%mm1, %%mm2, %%mm6) + PAVGB(%%mm3, %%mm2, %%mm0, %%mm6) + "movq %%mm0, 8%0 \n\t" + :"+m"(*dst) + :"m"(*src1), "m"(*src2) + :"memory"); + dst += dstStride; + src1 += src1Stride; + src2 += 16; + } while (--h); +} + static void DEF(avg, pixels8_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h) { MOVQ_BFE(mm6); @@ -399,6 +633,133 @@ static void DEF(avg, pixels8_xy2)(UINT8 *block, const UINT8 *pixels, int line_si :"eax", "memory"); } +static void DEF(avg, pixels8_l4)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int stride, int h) +{ + MOVQ_ZERO(mm7); + SET_RND(mm6); // =2 for rnd and =1 for no_rnd version + MOVQ_BFE(mm5); + __asm __volatile( + ".balign 8 \n\t" + "1: \n\t" + "movq (%1), %%mm0 \n\t" + "movq (%2), %%mm1 \n\t" + "movq 64(%2), %%mm2 \n\t" + "movq 136(%2), %%mm3 \n\t" + "punpcklbw %%mm7, %%mm0 \n\t" + "punpcklbw %%mm7, %%mm1 \n\t" + "punpcklbw %%mm7, %%mm2 \n\t" + "punpcklbw %%mm7, %%mm3 \n\t" + "paddusw %%mm6, %%mm0 \n\t" + "paddusw %%mm0, %%mm1 \n\t" + "paddusw %%mm2, %%mm3 \n\t" + "paddusw %%mm1, %%mm3 \n\t" + "psrlw $2, %%mm3 \n\t" + "movq (%1), %%mm0 \n\t" + "movq (%2), %%mm1 \n\t" + "movq 64(%2), %%mm2 \n\t" + "movq 136(%4), %%mm4 \n\t" + "punpckhbw %%mm7, %%mm0 \n\t" + "punpckhbw %%mm7, %%mm1 \n\t" + "punpckhbw %%mm7, %%mm2 \n\t" + "punpckhbw %%mm7, %%mm4 \n\t" + "paddusw %%mm6, %%mm0 \n\t" + "paddusw %%mm0, %%mm1 \n\t" + "paddusw %%mm2, %%mm4 \n\t" + "paddusw %%mm1, %%mm4 \n\t" + "psrlw $2, %%mm4 \n\t" + "packuswb %%mm4, %%mm3 \n\t" + "movq (%0), %%mm4 \n\t" + PAVGB(%%mm3, %%mm4, %%mm0, %%mm5) + "movq %%mm3, (%0) \n\t" + "addl %4, %0 \n\t" + "addl %4, %1 \n\t" + "addl $8, %2 \n\t" + "decl %3 \n\t" + "jnz 1b \n\t" + :"+r"(dst), "+r"(src1), "+r"(src2), "+r"(h) + :"r"(stride) + :"memory"); +} + +static void DEF(avg, pixels16_l4)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int stride, int h) +{ + MOVQ_ZERO(mm7); + SET_RND(mm6); // =2 for rnd and =1 for no_rnd version + MOVQ_BFE(mm5); + __asm __volatile( + ".balign 8 \n\t" + "1: \n\t" + "movq (%1), %%mm0 \n\t" + "movq (%2), %%mm1 \n\t" + "movq 256(%2), %%mm2 \n\t" + "movq 528(%2), %%mm3 \n\t" + "punpcklbw %%mm7, %%mm0 \n\t" + "punpcklbw %%mm7, %%mm1 \n\t" + "punpcklbw %%mm7, %%mm2 \n\t" + "punpcklbw %%mm7, %%mm3 \n\t" + "paddusw %%mm6, %%mm0 \n\t" + "paddusw %%mm0, %%mm1 \n\t" + "paddusw %%mm2, %%mm3 \n\t" + "paddusw %%mm1, %%mm3 \n\t" + "psrlw $2, %%mm3 \n\t" + "movq (%1), %%mm0 \n\t" + "movq (%2), %%mm1 \n\t" + "movq 256(%2), %%mm2 \n\t" + "movq 528(%4), %%mm4 \n\t" + "punpckhbw %%mm7, %%mm0 \n\t" + "punpckhbw %%mm7, %%mm1 \n\t" + "punpckhbw %%mm7, %%mm2 \n\t" + "punpckhbw %%mm7, %%mm4 \n\t" + "paddusw %%mm6, %%mm0 \n\t" + "paddusw %%mm0, %%mm1 \n\t" + "paddusw %%mm2, %%mm4 \n\t" + "paddusw %%mm1, %%mm4 \n\t" + "psrlw $2, %%mm4 \n\t" + "packuswb %%mm4, %%mm3 \n\t" + "movq (%0), %%mm4 \n\t" + PAVGB(%%mm3, %%mm4, %%mm0, %%mm5) + "movq %%mm3, (%0) \n\t" + "movq 8(%1), %%mm0 \n\t" + "movq 8(%2), %%mm1 \n\t" + "movq 264(%2), %%mm2 \n\t" + "movq 536(%2), %%mm3 \n\t" + "punpcklbw %%mm7, %%mm0 \n\t" + "punpcklbw %%mm7, %%mm1 \n\t" + "punpcklbw %%mm7, %%mm2 \n\t" + "punpcklbw %%mm7, %%mm3 \n\t" + "paddusw %%mm6, %%mm0 \n\t" + "paddusw %%mm0, %%mm1 \n\t" + "paddusw %%mm2, %%mm3 \n\t" + "paddusw %%mm1, %%mm3 \n\t" + "psrlw $2, %%mm3 \n\t" + "movq 8(%1), %%mm0 \n\t" + "movq 8(%2), %%mm1 \n\t" + "movq 264(%2), %%mm2 \n\t" + "movq 536(%4), %%mm4 \n\t" + "punpckhbw %%mm7, %%mm0 \n\t" + "punpckhbw %%mm7, %%mm1 \n\t" + "punpckhbw %%mm7, %%mm2 \n\t" + "punpckhbw %%mm7, %%mm4 \n\t" + "paddusw %%mm6, %%mm0 \n\t" + "paddusw %%mm0, %%mm1 \n\t" + "paddusw %%mm2, %%mm4 \n\t" + "paddusw %%mm1, %%mm4 \n\t" + "psrlw $2, %%mm4 \n\t" + "packuswb %%mm4, %%mm3 \n\t" + "movq 8(%0), %%mm4 \n\t" + PAVGB(%%mm3, %%mm4, %%mm0, %%mm5) + "movq %%mm3, 8(%0) \n\t" + "addl %4, %0 \n\t" + "addl %4, %1 \n\t" + "addl $16, %2 \n\t" + "decl %3 \n\t" + "jnz 1b \n\t" + :"+r"(dst), "+r"(src1), "+r"(src2), "+r"(h) + :"r"(stride) + :"memory"); +} + + //FIXME optimize static void DEF(put, pixels16_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h){ DEF(put, pixels8_y2)(block , pixels , line_size, h); diff --git a/libavcodec/motion_est.c b/libavcodec/motion_est.c index dcc31a4a1a6f714416596a6de7582ffacb7769ea..da565eb95d916f3848bea0864d4608e8641f6478 100644 --- a/libavcodec/motion_est.c +++ b/libavcodec/motion_est.c @@ -311,6 +311,7 @@ static inline int get_penalty_factor(MpegEncContext *s, int type){ } void ff_init_me(MpegEncContext *s){ + set_cmp(s, s->dsp.me_pre_cmp, s->avctx->me_pre_cmp); set_cmp(s, s->dsp.me_cmp, s->avctx->me_cmp); set_cmp(s, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp); set_cmp(s, s->dsp.mb_cmp, s->avctx->mb_cmp); @@ -336,6 +337,12 @@ void ff_init_me(MpegEncContext *s){ s->me.motion_search[0]= simple_epzs_motion_search; s->me.motion_search[1]= simple_epzs_motion_search4; } + + if(s->avctx->me_pre_cmp&FF_CMP_CHROMA){ + s->me.pre_motion_search= simple_chroma_epzs_motion_search; + }else{ + s->me.pre_motion_search= simple_epzs_motion_search; + } } static int pix_dev(UINT8 * pix, int line_size, int mean) @@ -1037,7 +1044,7 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s, assert(s->quarter_sample==0 || s->quarter_sample==1); - s->me.penalty_factor = get_penalty_factor(s, s->avctx->me_cmp); + s->me.pre_penalty_factor = get_penalty_factor(s, s->avctx->me_pre_cmp); get_limits(s, &range, &xmin, &ymin, &xmax, &ymax, s->f_code); rel_xmin= xmin - mb_x*16; @@ -1072,8 +1079,8 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s, pred_x = P_MEDIAN[0]; pred_y = P_MEDIAN[1]; } - dmin = s->me.motion_search[0](s, 0, &mx, &my, P, pred_x, pred_y, rel_xmin, rel_ymin, rel_xmax, rel_ymax, - &s->last_picture, s->p_mv_table, (1<<16)>>shift, mv_penalty); + dmin = s->me.pre_motion_search(s, 0, &mx, &my, P, pred_x, pred_y, rel_xmin, rel_ymin, rel_xmax, rel_ymax, + &s->last_picture, s->p_mv_table, (1<<16)>>shift, mv_penalty); s->p_mv_table[xy][0] = mx<p_mv_table[xy][1] = my<me.sub_penalty_factor; const int map_generation= s->me.map_generation; + const int subpel_quality= s->avctx->me_subpel_quality; uint32_t *map= s->me.map; me_cmp_func cmp, chroma_cmp; me_cmp_func cmp_sub, chroma_cmp_sub; @@ -309,7 +310,7 @@ static int RENAME(qpel_motion_search)(MpegEncContext * s, memset(best, 64, sizeof(int)*8); #if 1 - if(s->avctx->dia_size>=2){ + if(s->me.dia_size>=2){ const int tl= score_map[(index-(1<>2, ny>>2) } + #if 0 - nx= FFMAX(4*mx - bx, bx - 4*mx); - ny= FFMAX(4*my - by, by - 4*my); + const int tl= score_map[(index-(1<avctx->dia_size); + const int minima_count= ABS(s->me.dia_size); int i, j; LOAD_COMMON(s->mb_x*16, s->mb_y*16); @@ -744,7 +755,7 @@ static inline int RENAME(var_diamond_search)(MpegEncContext * s, int *best, int cmp= s->dsp.me_cmp[size]; chroma_cmp= s->dsp.me_cmp[size+1]; - for(dia_size=1; dia_size<=s->avctx->dia_size; dia_size++){ + for(dia_size=1; dia_size<=s->me.dia_size; dia_size++){ int dir, start, end; const int x= best[0]; const int y= best[1]; @@ -893,15 +904,15 @@ static int RENAME(epzs_motion_search)(MpegEncContext * s, int block, } //check(best[0],best[1],0, b0) - if(s->avctx->dia_size==-1) + if(s->me.dia_size==-1) dmin= RENAME(funny_diamond_search)(s, best, dmin, ref_picture, pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax, shift, map, map_generation, size, mv_penalty); - else if(s->avctx->dia_size<-1) + else if(s->me.dia_size<-1) dmin= RENAME(sab_diamond_search)(s, best, dmin, ref_picture, pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax, shift, map, map_generation, size, mv_penalty); - else if(s->avctx->dia_size<2) + else if(s->me.dia_size<2) dmin= RENAME(small_diamond_search)(s, best, dmin, ref_picture, pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax, shift, map, map_generation, size, mv_penalty); @@ -969,15 +980,15 @@ static int RENAME(epzs_motion_search4)(MpegEncContext * s, int block, (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16) } - if(s->avctx->dia_size==-1) + if(s->me.dia_size==-1) dmin= RENAME(funny_diamond_search)(s, best, dmin, ref_picture, pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax, shift, map, map_generation, size, mv_penalty); - else if(s->avctx->dia_size<-1) + else if(s->me.dia_size<-1) dmin= RENAME(sab_diamond_search)(s, best, dmin, ref_picture, pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax, shift, map, map_generation, size, mv_penalty); - else if(s->avctx->dia_size<2) + else if(s->me.dia_size<2) dmin= RENAME(small_diamond_search)(s, best, dmin, ref_picture, pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax, shift, map, map_generation, size, mv_penalty); diff --git a/libavcodec/mpegvideo.c b/libavcodec/mpegvideo.c index fec95f4ca1024d434906b8070f6a7c6aa4b842db..219a86767f2cc5dbb3f39d1b6454b97d1bb9407e 100644 --- a/libavcodec/mpegvideo.c +++ b/libavcodec/mpegvideo.c @@ -2786,12 +2786,12 @@ static void encode_picture(MpegEncContext *s, int picture_number) else if(s->pict_type!=B_TYPE) s->no_rounding ^= 1; } - /* Estimate motion for every MB */ if(s->pict_type != I_TYPE){ if(s->pict_type != B_TYPE){ if((s->avctx->pre_me && s->last_non_b_pict_type==I_TYPE) || s->avctx->pre_me==2){ s->me.pre_pass=1; + s->me.dia_size= s->avctx->pre_dia_size; for(mb_y=s->mb_height-1; mb_y >=0 ; mb_y--) { for(mb_x=s->mb_width-1; mb_x >=0 ; mb_x--) { @@ -2804,6 +2804,7 @@ static void encode_picture(MpegEncContext *s, int picture_number) } } + s->me.dia_size= s->avctx->dia_size; for(mb_y=0; mb_y < s->mb_height; mb_y++) { s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1; s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1); @@ -2816,7 +2817,7 @@ static void encode_picture(MpegEncContext *s, int picture_number) s->block_index[1]+=2; s->block_index[2]+=2; s->block_index[3]+=2; - + /* compute motion vector & mb_type and store in context */ if(s->pict_type==B_TYPE) ff_estimate_b_frame_motion(s, mb_x, mb_y); diff --git a/libavcodec/mpegvideo.h b/libavcodec/mpegvideo.h index 1bd9bf6628f2855d47e6ec70c3d999b0cf8f1d8e..6d794d8626ddb844ec85e14732af45135c135cb4 100644 --- a/libavcodec/mpegvideo.h +++ b/libavcodec/mpegvideo.h @@ -139,9 +139,11 @@ typedef struct MotionEstContext{ uint32_t *map; /* map to avoid duplicate evaluations */ uint32_t *score_map; /* map to store the scores */ int map_generation; + int pre_penalty_factor; int penalty_factor; int sub_penalty_factor; int pre_pass; /* = 1 for the pre pass */ + int dia_size; UINT16 (*mv_penalty)[MAX_MV*2+1]; /* amount of bits needed to encode a MV */ int (*sub_motion_search)(struct MpegEncContext * s, int *mx_ptr, int *my_ptr, int dmin, @@ -153,6 +155,11 @@ typedef struct MotionEstContext{ int P[10][2], int pred_x, int pred_y, int xmin, int ymin, int xmax, int ymax, Picture *ref_picture, int16_t (*last_mv)[2], int ref_mv_scale, uint16_t * const mv_penalty); + int (*pre_motion_search)(struct MpegEncContext * s, int block, + int *mx_ptr, int *my_ptr, + int P[10][2], int pred_x, int pred_y, + int xmin, int ymin, int xmax, int ymax, Picture *ref_picture, int16_t (*last_mv)[2], + int ref_mv_scale, uint16_t * const mv_penalty); }MotionEstContext; typedef struct MpegEncContext { diff --git a/libavcodec/utils.c b/libavcodec/utils.c index 7ebc7d29e5fc3f5bc783c55055e04a4f73643eea..d593a0de78e3224e0b13dab5ea67ec231e655ff6 100644 --- a/libavcodec/utils.c +++ b/libavcodec/utils.c @@ -234,6 +234,7 @@ void avcodec_get_context_defaults(AVCodecContext *s){ s->me_method= ME_EPZS; s->get_buffer= avcodec_default_get_buffer; s->release_buffer= avcodec_default_release_buffer; + s->me_subpel_quality=8; } /**