📄 quantize.c
字号:
/* quantize.c, 量化和逆量化 */
#include <stdio.h>#include "global.h"
// 下面的这些常量将用在与MMX相关的代码中
static short PACKED_0xFF00[4] = { 0xff00U, 0xff00U, 0xff00U, 0xff00U };
static unsigned short PACKED_32768[4] = { 32768, 32768, 32768, 32768 };
static unsigned short PACKED_30720[4] = { 30720, 30720, 30720, 30720 };
static unsigned short PACKED_61440[4] = { 61440, 61440, 61440, 61440 };
static unsigned short PACKED_2048[4] = { 2048, 2048, 2048, 2048 };
/* 该量化器在趋近于0时采用1/8的步长（DC系数除外） * */int quant_intra(src,dst,dc_prec,quant_mat,recip_quant_mat,mquant)short *src, *dst;int dc_prec;unsigned char *quant_mat;
unsigned short *recip_quant_mat;int mquant;{ int i; int x, y, d;
static short mquantm3p2div4[4];
static unsigned short recip_mquant[4];
/*下面的代码是针对MMX处理器或快速量化的*/
if(cpu_MMX && fastQuantization)
{
mquantm3p2div4[0] = ((mquant*3)+2)>>2;
mquantm3p2div4[1] = mquantm3p2div4[0];
mquantm3p2div4[2] = mquantm3p2div4[0];
mquantm3p2div4[3] = mquantm3p2div4[0];
recip_mquant[0] = (unsigned short)((16384.0 + (double) mquant/2.0)/mquant);
recip_mquant[1] = recip_mquant[0];
recip_mquant[2] = recip_mquant[0];
recip_mquant[3] = recip_mquant[0];
x = src[0];
_asm
{
mov esi, src ;// esi = src
mov edx, recip_quant_mat ;// edx = recip_quant_mat
mov edi, dst ;// edi = dst
mov ecx, 8 ;// ecx = 8
quant_intra__l1:
movq mm0, [esi] ;// mm0 = 4 words into esi[0..7]
movq mm1, [esi+8] ;// mm1 = 4 words into esi[8..15]
movq mm2, mm0
movq mm3, mm1
psraw mm2, 15 ;// copy sign of mm0 into mm2
psraw mm3, 15 ;// copy sign of mm3 into mm3
pxor mm0, mm2
pxor mm1, mm3
psubw mm0, mm2 ;// mm0 = abs(mm0)
psubw mm1, mm3 ;// mm1 = abs(mm1)
psllw mm0, 5 ;// mm0 *= 32
psllw mm1, 5 ;// mm1 *= 32
movq mm4, mm0
movq mm6, [edx] ;// mm6 = 4 words into edx[0..7]
movq mm5, mm1
pmullw mm0, mm6 ;// mm0 = LSW(mm0 * mm6)
pmulhw mm4, mm6 ;// mm4 = MSW(mm0 * mm6)
psrlw mm0, 14 ;// shift LSW
psllw mm4, 2 ;// shift MSW
movq mm6, [edx+8] ;// mm6 = 4 words into edx[8..11]
por mm0, mm4 ;// combine mm0 and mm4 to get result
pmullw mm1, mm6 ;// mm1 = LSW(mm1 * mm6)
pmulhw mm5, mm6 ;// mm5 = MSW(mm5 * mm6)
psrlw mm1, 14 ;// shift LSW
psllw mm5, 2 ;// shift MSW
por mm1, mm5 ;// combine mm1 and mm5 to get result
paddw mm0, mquantm3p2div4 ;// mm0 += (2 * mquant + 2) >> 2
paddw mm1, mquantm3p2div4 ;// mm1 += (2 * mquant + 2) >> 2
movq mm4, mm0
movq mm6, recip_mquant ;// mm6 = recip_mquant
movq mm5, mm1
pmullw mm0, mm6 ;// mm0 = LSW(mm0 * mm6)
pmulhw mm4, mm6 ;// mm4 = MSW(mm0 * mm6)
psrlw mm0, 15 ;// shift LSW
psllw mm4, 1 ;// shift MSW
por mm0, mm4 ;// combine mm0 and mm4 to get result
pmullw mm1, mm6 ;// mm1 = LSW(mm1 * mm6)
pmulhw mm5, mm6 ;// mm5 = MSW(mm5 * mm6)
psrlw mm1, 15 ;// shift LSW
psllw mm5, 1 ;// shift MSW
movq mm6, PACKED_0xFF00 ;// mm6 = (0xff00, 0xff00, 0xff00, 0xff00)
por mm1, mm5 ;// combine mm1 and mm5 to get result
paddusw mm0, mm6
paddusw mm1, mm6
psubusw mm0, mm6 ;// clip each word in mm0 to [0..255]
psubusw mm1, mm6 ;// clip each word in mm1 to [0..255]
paddw mm0, mm2
paddw mm1, mm3
pxor mm0, mm2 ;// restore original sign of mm0
pxor mm1, mm3 ;// restore original sign of mm1
movq [edi], mm0 ;// store mm0 into edi[0..7]
movq [edi+8], mm1 ;// store mm1 into edi[8..15]
add esi, 16 ;// esi += 16
add edx, 16 ;// edx += 16
add edi, 16 ;// edi += 16
dec ecx ;// decrement ecx
jnz quant_intra__l1 ;// loop while not zero
emms ;// empty MMX state
}
d = 8>>dc_prec;
dst[0] = (x>=0) ? (x+(d>>1))/d : -((-x+(d>>1))/d); // 将 (x/d) 四舍五入
}
else
{
x = src[0]; d = 8>>dc_prec; /* intra_dc_mult */ dst[0] = (x>=0) ? (x+(d>>1))/d : -((-x+(d>>1))/d); /* 将 (x/d) 四舍五入 */ for (i=1; i<64; i++) {
x = src[i];
d = quant_mat[i]; y = (32*(x>=0 ? x : -x) + (d>>1))/d; /* 将(32*x/quant_mat) 四舍五入 */
d = (3*mquant+2)>>2;
y = (y+d)/(2*mquant); if (y > 255) { y = 255; }
dst[i] = (x>=0) ? y : -y; #if 0 if (x<0) x = -x; d = mquant*quant_mat[i]; y = (16*x + ((3*d)>>3)) / d; dst[i] = (src[i]<0) ? -y : y; #endif }
} return 1;}
int quant_non_intra(src,dst,quant_mat, recip_quant_mat,mquant)short *src, *dst;
unsigned char *quant_mat;unsigned short *recip_quant_mat;int mquant;{ int i; int x, y, d; int nzflag;
short recip_mquant[4];
nzflag = 0;
if(cpu_MMX && fastQuantization)
{
recip_mquant[0] = (unsigned short)((16384.0 + (double) mquant/2.0)/mquant);
recip_mquant[1] = recip_mquant[0];
recip_mquant[2] = recip_mquant[0];
recip_mquant[3] = recip_mquant[0];
_asm
{
mov esi, src ;// esi = src
mov edx, recip_quant_mat ;// edx = recip_quant_mat
mov edi, dst ;// edi = dst
mov ecx, 8 ;// ecx = 8
pxor mm7, mm7 ;// mm7 = nzflag
quant_non_intra__l1:
movq mm0, [esi] ;// mm0 = 4 words into src[0..7]
movq mm1, [esi+8] ;// mm1 = 4 words into src[8..15]
movq mm2, mm0
movq mm3, mm1
psraw mm2, 15 ;// copy sign of mm0 into mm2
psraw mm3, 15 ;// copy sign of mm3 into mm3
pxor mm0, mm2
pxor mm1, mm3
psubw mm0, mm2 ;// mm0 = abs(mm0)
psubw mm1, mm3 ;// mm1 = abs(mm1)
psllw mm0, 5 ;// mm0 *= 32
psllw mm1, 5 ;// mm1 *= 32
movq mm4, mm0
movq mm6, [edx] ;// mm6 = 4 words into edx[0..7]
movq mm5, mm1
pmullw mm0, mm6 ;// mm0 = LSW(mm0 * mm6)
pmulhw mm4, mm6 ;// mm4 = MSW(mm0 * mm6)
psrlw mm0, 14 ;// shift LSW
psllw mm4, 2 ;// shift MSW
movq mm6, [edx+8] ;// mm6 = 4 words into edx[8..15]
por mm0, mm4 ;// combine mm0 and mm4 to get result
pmullw mm1, mm6 ;// mm1 = LSW(mm1 * mm6)
pmulhw mm5, mm6 ;// mm5 = MSW(mm1 * mm6)
psrlw mm1, 14 ;// shift LSW
psllw mm5, 2 ;// shift MSW
por mm1, mm5 ;// combine mm1 and mm5 to get result
movq mm4, mm0
movq mm6, recip_mquant ;// mm6 = recip_mquant
movq mm5, mm1
pmullw mm0, mm6 ;// mm0 = LSW(mm0 * mm6)
pmulhw mm4, mm6 ;// mm4 = MSW(mm0 * mm6)
psrlw mm0, 15 ;// shift LSW
psllw mm4, 1 ;// shift MSW
por mm0, mm4 ;// combine mm0 and mm4 to get result
pmullw mm1, mm6 ;// mm1 = LSW(mm1 * mm6)
pmulhw mm5, mm6 ;// mm5 = MSW(mm1 * mm6)
psrlw mm1, 15 ;// shift LSW
psllw mm5, 1 ;// shift MSW
movq mm6, PACKED_0xFF00 ;// mm6 = (0xff00, 0xff00, 0xff00, 0xff00)
por mm1, mm5 ;// combine mm1 and mm5 to get result
paddusw mm0, mm6
paddusw mm1, mm6
psubusw mm0, mm6 ;// clip each word in mm0 to [0..255]
psubusw mm1, mm6 ;// clip each word in mm1 to [0..255]
paddw mm0, mm2
paddw mm1, mm3
pxor mm0, mm2 ;// restore original sign of mm0
pxor mm1, mm3 ;// restore original sign of mm1
por mm7, mm0 ;// update nzflag
movq [edi], mm0 ;// store mm0 into dst[0..7]
por mm7, mm1 ;// update nzflag
movq [edi+8], mm1 ;// store mm1 into dst[8..15]
add esi, 16 ;// esi += 16
add edx, 16 ;// edx += 16
add edi, 16 ;// edi += 16
dec ecx ;// decrement ecx
jnz quant_non_intra__l1 ;// loop while not zero
movd eax, mm7 ;// load the lower dword of mm7 into eax
psrlq mm7, 32 ;// shift mm7 to get upper dword
movd ebx, mm7 ;// load the lower dword of mm7 into ebx
or eax, ebx ;// update nzflag
setnz byte ptr nzflag ;// generate nzflag
emms ;// empty MMX state
}
}
else
{
for (i=0; i<64; i++) {
x = src[i]; d = quant_mat[i]; y = (32*(x>=0 ? x : -x) + (d>>1))/d; y /= (2*mquant);
if (y > 255) { y = 255; }
if ((dst[i] = (x>=0 ? y : -y)) != 0) nzflag=1; }
}
return nzflag;
}/* MPEG-1 逆量化，针对I帧*/extern void iquant_intra(src,dst,dc_prec,quant_mat,mquant)short *src, *dst;int dc_prec;unsigned char *quant_mat;int mquant;{
int x, i, val;
unsigned short PACKED_MQUANT[4];
if(cpu_MMX && fastQuantization)
{
PACKED_MQUANT[0] = mquant;
PACKED_MQUANT[1] = mquant;
PACKED_MQUANT[2] = mquant;
PACKED_MQUANT[3] = mquant;
x = src[0];
_asm
{
mov esi, src ;// esi = src
mov edx, quant_mat ;// edx = quant_mat
mov edi, dst ;// edi = dst
mov ecx, 8 ;// ecx = 8
pxor mm7, mm7 ;// mm7 = 0
iquant_intra__l1:
movq mm0, [esi] ;// mm0 = 4 words into esi[0..7]
movq mm2, [esi+8] ;// mm2 = 4 words into esi[8..15]
movq mm4, [edx] ;// mm4 = 4 words into edx[0..7]
movq mm1, mm0
movq mm3, mm2
movq mm5, mm4
psraw mm1, 15 ;// copy sign of mm0 into mm1
psraw mm3, 15 ;// copy sign of mm2 into mm3
pxor mm0, mm1
pxor mm2, mm3
psubw mm0, mm1 ;// mm0 = abs(mm0)
psubw mm2, mm3 ;// mm2 = abs(mm2)
punpcklbw mm4, mm7 ;// unpack the lower 4 bytes into mm4
punpckhbw mm5, mm7 ;// unpack the lower 4 bytes into mm5
pmullw mm0, mm4 ;// mm0 *= mm4
pmullw mm2, mm5 ;// mm2 *= mm5
movq mm4, mm0
movq mm6, PACKED_MQUANT ;// mm6 = (mquant, mquant, mquant, mquant)
movq mm5, mm2
pmullw mm0, mm6 ;// mm0 = LSW(mm0 * mm6)
pmulhw mm4, mm6 ;// mm4 = MSW(mm4 * mm6)
pmullw mm2, mm6 ;// mm2 = LSW(mm2 * mm6)
pmulhw mm5, mm6 ;// mm5 = MSW(mm5 * mm6)
psrlw mm0, 4 ;// shift LSW
psllw mm4, 12 ;// shift MSW
psrlw mm2, 4 ;// shift LSW
psllw mm5, 12 ;// shift MSW
por mm0, mm4 ;// combine mm0 and mm4 to get result
por mm2, mm5 ;// combine mm2 and mm5 to get result
paddw mm0, mm1
paddw mm2, mm3
pxor mm0, mm1 ;// restore original sign of mm0
pxor mm2, mm3 ;// restore original sign of mm2
paddw mm0, PACKED_32768
paddw mm2, PACKED_32768
paddusw mm0, PACKED_30720
paddusw mm2, PACKED_30720
psubusw mm0, PACKED_61440
psubusw mm2, PACKED_61440
psubw mm0, PACKED_2048 ;// clip each words into mm0 to [-2048..2047]
psubw mm2, PACKED_2048 ;// clip each words into mm2 to [-2048..2047]
movq [edi], mm0 ;// store mm0 into edi[0..7]
movq [edi+8], mm2 ;// store mm2 into edi[8..15]
add esi, 16 ;// esi += 16
add edx, 8 ;// edx += 8
add edi, 16 ;// edi += 16
dec ecx ;// decrement ecx
jnz iquant_intra__l1 ;// loop while not zero
emms ;// empty MMX state
}
dst[0] = x << (3-dc_prec);
}
else
{ dst[0] = src[0] << (3-dc_prec); for (i=1; i<64; i++) { val = (int)(src[i]*quant_mat[i]*mquant)/16;
/* 如果不匹配则进行如下操作 */ if ((val&1)==0 && val!=0) val+= (val>0) ? -1 : 1; /* 色饱和 */ dst[i] = (val>2047) ? 2047 : ((val<-2048) ? -2048 : val); }
}}
/*逆量化针对非I帧图像*/extern void iquant_non_intra(src,dst,quant_mat,mquant)short *src, *dst;unsigned char *quant_mat;int mquant;{
int i, val;
unsigned short PACKED_MQUANT[4];
if(cpu_MMX && fastQuantization)
{
PACKED_MQUANT[0] = mquant;
PACKED_MQUANT[1] = mquant;
PACKED_MQUANT[2] = mquant;
PACKED_MQUANT[3] = mquant;
_asm
{
mov esi, src ;// esi = src
mov edx, quant_mat ;// edx = quant_mat
mov edi, dst ;// edi = dst
mov ecx, 8 ;// ecx = 8
pxor mm7, mm7 ;// mm7 = 0
iquant_non_intra__l1:
movq mm0, [esi] ;// mm0 = 4 words into esi[0..7]
movq mm2, [esi+8] ;// mm2 = 4 words into esi[8..15]
movq mm4, [edx] ;// mm4 = 4 words into edx[0..7]
movq mm1, mm0
movq mm3, mm2
movq mm5, mm4
psraw mm1, 15 ;// copy sign of mm0 into mm1
psraw mm3, 15 ;// copy sign of mm2 into mm3
pxor mm0, mm1
pxor mm2, mm3
psubw mm0, mm1 ;// mm0 = abs(mm0)
psubw mm2, mm3 ;// mm2 = abs(mm2)
psllw mm0, 1 ;// mm0 *= 2
psllw mm2, 1 ;// mm2 *= 2
punpcklbw mm4, mm7 ;// unpack the lower 4 bytes into mm4
punpckhbw mm5, mm7 ;// unpack the lower 4 bytes into mm5
pmullw mm0, mm4 ;// mm0 *= mm4
pmullw mm2, mm5 ;// mm2 *= mm5
movq mm4, mm0
movq mm6, PACKED_MQUANT ;// mm6 = (mquant, mquant, mquant, mquant)
movq mm5, mm2
pmullw mm0, mm6 ;// mm0 = LSW(mm0 * mm6)
pmulhw mm4, mm6 ;// mm4 = MSW(mm4 * mm6)
pmullw mm2, mm6 ;// mm2 = LSW(mm2 * mm6)
pmulhw mm5, mm6 ;// mm5 = MSW(mm5 * mm6)
psrlw mm0, 5 ;// shift LSW
psllw mm4, 11 ;// shift MSW
psrlw mm2, 5 ;// shift LSW
psllw mm5, 11 ;// shift MSW
por mm0, mm4 ;// combine mm0 and mm4 to get result
por mm2, mm5 ;// combine mm2 and mm5 to get result
paddw mm0, mm1
paddw mm2, mm3
pxor mm0, mm1 ;// restore original sign of mm0
pxor mm2, mm3 ;// restore original sign of mm2
paddw mm0, PACKED_32768
paddw mm2, PACKED_32768
paddusw mm0, PACKED_30720
paddusw mm2, PACKED_30720
psubusw mm0, PACKED_61440
psubusw mm2, PACKED_61440
psubw mm0, PACKED_2048 ;// clip each words into mm0 to [-2048..2047]
psubw mm2, PACKED_2048 ;// clip each words into mm2 to [-2048..2047]
movq [edi], mm0 ;// store mm0 into edi[0..7]
movq [edi+8], mm2 ;// store mm2 into edi[8..15]
add esi, 16 ;// esi += 16
add edx, 8 ;// edx += 8
add edi, 16 ;// edi += 16
dec ecx ;// decrement ecx
jnz iquant_non_intra__l1;// loop while not zero
emms ;// empty MMX state
}
}
else
{
for (i=0; i<64; i++) { val = src[i]; if (val!=0) { val = (int)((2*val+(val>0 ? 1 : -1))*quant_mat[i]*mquant)/32; /* 如果不匹配则进行如下操作 */ if ((val&1)==0 && val!=0) val+= (val>0) ? -1 : 1; } /* 色饱和 */ dst[i] = (val>2047) ? 2047 : ((val<-2048) ? -2048 : val); }
}
}⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -