Skip to content

F
Ffmpeg

小白菜888 / Ffmpeg

通知 3
Star 0
Fork 0
F
Ffmpeg

体验新版 GitCode，发现更多精彩内容 >>

提交 826f429a 编写于 作者: M Michael Niedermayer
浏览文件
操作

qpel in mmx2/3dnow 

qpel refinement quality parameter

Originally committed as revision 1393 to svn://svn.ffmpeg.org/ffmpeg/trunk
上级 70ac76c0
展开全部 隐藏空白更改
内联 并排
Showing with 1263 addition and 46 deletion
libavcodec/avcodec.h
浏览文件 @ 826f429a
......@@ -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 {
......
libavcodec/dsputil.c
浏览文件 @ 826f429a
......@@ -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<h; i++)\
{\
OP(dst[ 0], (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]));\
......@@ -853,9 +855,10 @@ static void OPNAME ## mpeg4_qpel16_h_lowpass(UINT8 *dst, UINT8 *src, int dstStri
}\
}\
\
static void OPNAME ## mpeg4_qpel16_v_lowpass(UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int w){\
static void OPNAME ## mpeg4_qpel16_v_lowpass(UINT8 *dst, UINT8 *src, int dstStride, int srcStride){\
UINT8 *cm = cropTbl + MAX_NEG_CROP;\
int i;\
const int w=16;\
for(i=0; i<w; i++)\
{\
const int src0= src[0*srcStride];\
......@@ -1046,21 +1049,21 @@ static void OPNAME ## qpel16_mc01_c(UINT8 *dst, UINT8 *src, int stride){\
UINT8 full[24*17];\
UINT8 half[256];\
copy_block17(full, src, 24, stride, 17);\
put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\
OPNAME ## pixels16_l2(dst, full, half, stride, 24, 16, 16);\
}\
\
static void OPNAME ## qpel16_mc02_c(UINT8 *dst, UINT8 *src, int stride){\
UINT8 full[24*17];\
copy_block17(full, src, 24, stride, 17);\
OPNAME ## mpeg4_qpel16_v_lowpass(dst, full, stride, 24, 16);\
OPNAME ## mpeg4_qpel16_v_lowpass(dst, full, stride, 24);\
}\
\
static void OPNAME ## qpel16_mc03_c(UINT8 *dst, UINT8 *src, int stride){\
UINT8 full[24*17];\
UINT8 half[256];\
copy_block17(full, src, 24, stride, 17);\
put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\
OPNAME ## pixels16_l2(dst, full+24, half, stride, 24, 16, 16);\
}\
static void OPNAME ## qpel16_mc11_c(UINT8 *dst, UINT8 *src, int stride){\
......@@ -1070,8 +1073,8 @@ static void OPNAME ## qpel16_mc11_c(UINT8 *dst, UINT8 *src, int stride){\
UINT8 halfHV[256];\
copy_block17(full, src, 24, stride, 17);\
put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
OPNAME ## pixels16_l4(dst, full, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc31_c(UINT8 *dst, UINT8 *src, int stride){\
......@@ -1081,8 +1084,8 @@ static void OPNAME ## qpel16_mc31_c(UINT8 *dst, UINT8 *src, int stride){\
UINT8 halfHV[256];\
copy_block17(full, src, 24, stride, 17);\
put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
OPNAME ## pixels16_l4(dst, full+1, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc13_c(UINT8 *dst, UINT8 *src, int stride){\
......@@ -1092,8 +1095,8 @@ static void OPNAME ## qpel16_mc13_c(UINT8 *dst, UINT8 *src, int stride){\
UINT8 halfHV[256];\
copy_block17(full, src, 24, stride, 17);\
put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
OPNAME ## pixels16_l4(dst, full+24, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc33_c(UINT8 *dst, UINT8 *src, int stride){\
......@@ -1103,22 +1106,22 @@ static void OPNAME ## qpel16_mc33_c(UINT8 *dst, UINT8 *src, int stride){\
UINT8 halfHV[256];\
copy_block17(full, src, 24, stride, 17);\
put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full , 16, 24, 17);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
OPNAME ## pixels16_l4(dst, full+25, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc21_c(UINT8 *dst, UINT8 *src, int stride){\
UINT8 halfH[272];\
UINT8 halfHV[256];\
put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc23_c(UINT8 *dst, UINT8 *src, int stride){\
UINT8 halfH[272];\
UINT8 halfHV[256];\
put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc12_c(UINT8 *dst, UINT8 *src, int stride){\
......@@ -1128,8 +1131,8 @@ static void OPNAME ## qpel16_mc12_c(UINT8 *dst, UINT8 *src, int stride){\
UINT8 halfHV[256];\
copy_block17(full, src, 24, stride, 17);\
put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc32_c(UINT8 *dst, UINT8 *src, int stride){\
......@@ -1139,14 +1142,14 @@ static void OPNAME ## qpel16_mc32_c(UINT8 *dst, UINT8 *src, int stride){\
UINT8 halfHV[256];\
copy_block17(full, src, 24, stride, 17);\
put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\
put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc22_c(UINT8 *dst, UINT8 *src, int stride){\
UINT8 halfH[272];\
put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\
OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16, 16);\
OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\
}
#define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1)
......
libavcodec/dsputil.h
浏览文件 @ 826f429a
......@@ -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];
......
libavcodec/i386/dsputil_mmx.c
浏览文件 @ 826f429a
此差异已折叠。
libavcodec/i386/dsputil_mmx_avg.h
浏览文件 @ 826f429a
......@@ -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)
......
libavcodec/i386/dsputil_mmx_rnd.h
浏览文件 @ 826f429a
......@@ -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);
......
libavcodec/motion_est.c
浏览文件 @ 826f429a
......@@ -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<<shift;
s->p_mv_table[xy][1] = my<<shift;
......
libavcodec/motion_est_template.c
浏览文件 @ 826f429a
......@@ -268,6 +268,7 @@ static int RENAME(qpel_motion_search)(MpegEncContext * s,
const int my = *my_ptr;
const int penalty_factor= s->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<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
const int bl= score_map[(index+(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
const int tr= score_map[(index-(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
......@@ -388,24 +389,34 @@ static int RENAME(qpel_motion_search)(MpegEncContext * s,
}
}
}
for(i=0; i<8; i++){
for(i=0; i<subpel_quality; i++){
nx= best_pos[i][0];
ny= best_pos[i][1];
CHECK_QUARTER_MV(nx&3, ny&3, nx>>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<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
const int bl= score_map[(index+(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
const int tr= score_map[(index-(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
const int br= score_map[(index+(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
// if(l < r && l < t && l < b && l < tl && l < bl && l < tr && l < br && bl < tl){
if(tl<br){
// nx= FFMAX(4*mx - bx, bx - 4*mx);
// ny= FFMAX(4*my - by, by - 4*my);
static int stats[4][4];
stats[nx][ny]++;
if(256*256*256*64 % (stats[0][0]+1) ==0){
for(i=0; i<16; i++){
if((i&3)==0) printf("\n");
static int stats[7][7], count;
count++;
stats[4*mx - bx + 3][4*my - by + 3]++;
if(256*256*256*64 % count ==0){
for(i=0; i<49; i++){
if((i%7)==0) printf("\n");
printf("%6d ", stats[0][i]);
}
printf("\n");
}
}
#endif
#else
......@@ -659,7 +670,7 @@ static inline int RENAME(sab_diamond_search)(MpegEncContext * s, int *best, int
{
me_cmp_func cmp, chroma_cmp;
Minima minima[MAX_SAB_SIZE];
const int minima_count= ABS(s->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);
......
libavcodec/mpegvideo.c
浏览文件 @ 826f429a
......@@ -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);
......
libavcodec/mpegvideo.h
浏览文件 @ 826f429a
......@@ -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 {
......
libavcodec/utils.c
浏览文件 @ 826f429a
......@@ -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;
}
/**
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或