注:问号以及未注释部分 会在x265-1.8版本内更新
/*****************************************************************************
* Copyright (C) 2015 x265 project
*
* Authors: Steve Borho <steve@borho.org>
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program 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 General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
*
* This program is also available under a commercial proprietary license.
* For more information, contact us at license @ x265.com.
*****************************************************************************/
#include "common.h"
#include "primitives.h"
#include "scalinglist.h"
namespace {
// file-anonymous namespace
/* Strings for scaling list file parsing */
/* 不同的量化表类型
* 这个表对量化非常重要,量化表是根据这个表中 不同的TU尺寸(4x4、8x8、16x16、32x32)、不同的图像分量(Y、U、V)分别进行建立的
*/
const char MatrixType[4][6][20] =
{
{ // Intra 4x4 + Inter 4x4
"INTRA4X4_LUMA",
"INTRA4X4_CHROMAU",
"INTRA4X4_CHROMAV",
"INTER4X4_LUMA",
"INTER4X4_CHROMAU",
"INTER4X4_CHROMAV"
},
{ // Intra 8x8 + Inter 8x8
"INTRA8X8_LUMA",
"INTRA8X8_CHROMAU",
"INTRA8X8_CHROMAV",
"INTER8X8_LUMA",
"INTER8X8_CHROMAU",
"INTER8X8_CHROMAV"
},
{ // Intra 16x16 + Inter 16x16
"INTRA16X16_LUMA",
"INTRA16X16_CHROMAU",
"INTRA16X16_CHROMAV",
"INTER16X16_LUMA",
"INTER16X16_CHROMAU",
"INTER16X16_CHROMAV"
},
{ // Intra 32x32 Luma + Inter 32x32 Luma
"INTRA32X32_LUMA",
"INTER32X32_LUMA",
},
};
const char MatrixType_DC[4][12][22] =
{
{
},
{
},
{
"INTRA16X16_LUMA_DC",
"INTRA16X16_CHROMAU_DC",
"INTRA16X16_CHROMAV_DC",
"INTER16X16_LUMA_DC",
"INTER16X16_CHROMAU_DC",
"INTER16X16_CHROMAV_DC"
},
{
"INTRA32X32_LUMA_DC",
"INTER32X32_LUMA_DC",
},
};
int quantTSDefault4x4[16] =
{
16, 16, 16, 16,
16, 16, 16, 16,
16, 16, 16, 16,
16, 16, 16, 16
};
int quantIntraDefault8x8[64] =
{
16, 16, 16, 16, 17, 18, 21, 24,
16, 16, 16, 16, 17, 19, 22, 25,
16, 16, 17, 18, 20, 22, 25, 29,
16, 16, 18, 21, 24, 27, 31, 36,
17, 17, 20, 24, 30, 35, 41, 47,
18, 19, 22, 27, 35, 44, 54, 65,
21, 22, 25, 31, 41, 54, 70, 88,
24, 25, 29, 36, 47, 65, 88, 115
};
int quantInterDefault8x8[64] =
{
16, 16, 16, 16, 17, 18, 20, 24,
16, 16, 16, 17, 18, 20, 24, 25,
16, 16, 17, 18, 20, 24, 25, 28,
16, 17, 18, 20, 24, 25, 28, 33,
17, 18, 20, 24, 25, 28, 33, 41,
18, 20, 24, 25, 28, 33, 41, 54,
20, 24, 25, 28, 33, 41, 54, 71,
24, 25, 28, 33, 41, 54, 71, 91
};
}
namespace x265 {
// private namespace
const int ScalingList::s_numCoefPerSize[NUM_SIZES] = { 16, 64, 256, 1024 }; // 不同变换块大小,对应的变换系数个数,(4x4:16,8x8:64,16x16:265,32x32:1024)
const int32_t ScalingList::s_quantScales[NUM_REM] = { 26214, 23302, 20560, 18396, 16384, 14564 }; // 前向量化系数表的值,分别对应不同的Qp余数(Qp%6)
const int32_t ScalingList::s_invQuantScales[NUM_REM] = { 40, 45, 51, 57, 64, 72 }; // 反量化系数表的值,分别对应不同的Qp余数(Qp%6)
ScalingList::ScalingList()
{
memset(m_quantCoef, 0, sizeof(m_quantCoef));
memset(m_dequantCoef, 0, sizeof(m_dequantCoef));
memset(m_scalingListCoef, 0, sizeof(m_scalingListCoef));
}
/** 函数功能 : 初始化量化中所需要的几个表格
* 前向量化表"m_quantCoef"/反量化表"m_dequantCoef"/量化矩阵表"m_scalingListCoef"
* \返回值 numSig : 初始化成功则返回true,否则返回false
*/
bool ScalingList::init()
{
bool ok = true;
for (int sizeId = 0; sizeId < NUM_SIZES; sizeId++)
{
for (int listId = 0; listId < NUM_LISTS; listId++)
{
m_scalingListCoef[sizeId][listId] = X265_MALLOC(int32_t, X265_MIN(MAX_MATRIX_COEF_NUM, s_numCoefPerSize[sizeId])); // 为m_scalingListCoef分配空间,最大为8x8=64
ok &= !!m_scalingListCoef[sizeId][listId];
for (int rem = 0; rem < NUM_REM; rem++)
{
m_quantCoef[sizeId][listId][rem] = X265_MALLOC(int32_t, s_numCoefPerSize[sizeId]); // 为m_quantCoef分配空间,空间大小是s_numCoefPerSize[sizeId],即根据TU尺寸分配
m_dequantCoef[sizeId][listId][rem] = X265_MALLOC(int32_t, s_numCoefPerSize[sizeId]);// 为m_dequantCoef分配空间,空间大小是s_numCoefPerSize[sizeId],即根据TU尺寸分配
ok &= m_quantCoef[sizeId][listId][rem] && m_dequantCoef[sizeId][listId][rem];
}
}
}
return ok;
}
ScalingList::~ScalingList()
{
for (int sizeId = 0; sizeId < NUM_SIZES; sizeId++)
{
for (int listId = 0; listId < NUM_LISTS; listId++)
{
X265_FREE(m_scalingListCoef[sizeId][listId]);
for (int rem = 0; rem < NUM_REM; rem++)
{
X265_FREE(m_quantCoef[sizeId][listId][rem]);
X265_FREE(m_dequantCoef[sizeId][listId][rem]);
}
}
}
}
/* returns predicted list index if a match is found, else -1 */
int ScalingList::checkPredMode(int size, int list) const
{
for (int predList = list; predList >= 0; predList--)
{
// check DC value
if (size < BLOCK_16x16 && m_scalingListDC[size][list] != m_scalingListDC[size][predList])
continue;
// check value of matrix
if (!memcmp(m_scalingListCoef[size][list],
list == predList ? getScalingListDefaultAddress(size, predList) : m_scalingListCoef[size][predList],
sizeof(int32_t) * X265_MIN(MAX_MATRIX_COEF_NUM, s_numCoefPerSize[size])))
return predList;
}
return -1;
}
/* check if use default quantization matrix
* returns true if default quantization matrix is used in all sizes */
bool ScalingList::checkDefaultScalingList() const
{
int defaultCounter = 0;
for (int s = 0; s < NUM_SIZES; s++)
for (int l = 0; l < NUM_LISTS; l++)
if (!memcmp(m_scalingListCoef[s][l], getScalingListDefaultAddress(s, l),
sizeof(int32_t) * X265_MIN(MAX_MATRIX_COEF_NUM, s_numCoefPerSize[s])) &&
((s < BLOCK_16x16) || (m_scalingListDC[s][l] == 16)))
defaultCounter++;
return defaultCounter != (NUM_LISTS * NUM_SIZES - 4); // -4 for 32x32
}
/* get address of default quantization matrix */
/** 函数功能 : 根据不同的TU尺寸,不同的list类型得到不同的默认量化矩阵
* \参数 sizeId : TU尺寸的ID
* \参数 listId : list类型的ID
*/
const int32_t* ScalingList::getScalingListDefaultAddress(int sizeId, int listId) const
{
switch (sizeId)
{
case BLOCK_4x4: // 4x4 量化矩阵
return quantTSDefault4x4;
case BLOCK_8x8: // 8x8 量化矩阵
return (listId < 3) ? quantIntraDefault8x8 : quantInterDefault8x8; // listId:0~2为Intra的量化矩阵(0~2分别对应Y、U、V),3~5为Inter的量化矩阵(3~5分别对应Y、U、V)
case BLOCK_16x16: // 16x16 量化矩阵
return (listId < 3) ? quantIntraDefault8x8 : quantInterDefault8x8; // listId:0~2为Intra的量化矩阵(0~2分别对应Y、U、V),3~5为Inter的量化矩阵(3~5分别对应Y、U、V)
case BLOCK_32x32:
return (listId < 1) ? quantIntraDefault8x8 : quantInterDefault8x8; // 对于32x32,0为Intra Luma的量化矩阵,1为Inter Luma的量化矩阵
default:
break;
}
X265_CHECK(0, "invalid scaling list size\n");
return NULL;
}
/** 函数功能 : 将默认量化矩阵中拷贝到m_scalingListCoef,用于之后的量化
* \参数 sizeId : TU尺寸的ID
* \参数 listId : list类型的ID
*/
void ScalingList::processDefaultMarix(int sizeId, int listId)
{ // 根据不同的TU尺寸,不同的list类型来拷贝对应的默认量化矩阵
memcpy(m_scalingListCoef[sizeId][listId], getScalingListDefaultAddress(sizeId, listId), sizeof(int) * X265_MIN(MAX_MATRIX_COEF_NUM, s_numCoefPerSize[sizeId]));
m_scalingListDC[sizeId][listId] = SCALING_LIST_DC; // 设置DC系数的量化矩阵值
}
/** 函数功能 : 设置默认的量化矩阵
** 调用范围 : 仅在Encoder::create()中被调用
*/
void ScalingList::setDefaultScalingList()
{
for (int sizeId = 0; sizeId < NUM_SIZES; sizeId++) // 设置不同TU尺寸的量化表
for (int listId = 0; listId < NUM_LISTS; listId++) // 设置不同list类型的量化表
processDefaultMarix(sizeId, listId);
m_bEnabled = true; // 使能量化矩阵
m_bDataPresent = false; // 使用默认的量化矩阵
}
bool ScalingList::parseScalingList(const char* filename)
{
FILE *fp = fopen(filename, "r");
if (!fp)
{
x265_log(NULL, X265_LOG_ERROR, "can't open scaling list file %s\n", filename);
return true;
}
char line[1024];
int32_t *src = NULL;
for (int sizeIdc = 0; sizeIdc < NUM_SIZES; sizeIdc++)
{
int size = X265_MIN(MAX_MATRIX_COEF_NUM, s_numCoefPerSize[sizeIdc]);
for (int listIdc = 0; listIdc < NUM_LISTS; listIdc++)
{
src = m_scalingListCoef[sizeIdc][listIdc];
fseek(fp, 0, 0);
do
{
char *ret = fgets(line, 1024, fp);
if (!ret || (!strstr(line, MatrixType[sizeIdc][listIdc]) && feof(fp)))
{
x265_log(NULL, X265_LOG_ERROR, "can't read matrix from %s\n", filename);
return true;
}
}
while (!strstr(line, MatrixType[sizeIdc][listIdc]));
for (int i = 0; i < size; i++)
{
int data;
if (fscanf(fp, "%d,", &data) != 1)
{
x265_log(NULL, X265_LOG_ERROR, "can't read matrix from %s\n", filename);
return true;
}
src[i] = data;
}
// set DC value for default matrix check
m_scalingListDC[sizeIdc][listIdc] = src[0];
if (sizeIdc > BLOCK_8x8)
{
fseek(fp, 0, 0);
do
{
char *ret = fgets(line, 1024, fp);
if (!ret || (!strstr(line, MatrixType_DC[sizeIdc][listIdc]) && feof(fp)))
{
x265_log(NULL, X265_LOG_ERROR, "can't read DC from %s\n", filename);
return true;
}
}
while (!strstr(line, MatrixType_DC[sizeIdc][listIdc]));
int data;
if (fscanf(fp, "%d,", &data) != 1)
{
x265_log(NULL, X265_LOG_ERROR, "can't read matrix from %s\n", filename);
return true;
}
// overwrite DC value when size of matrix is larger than 16x16
m_scalingListDC[sizeIdc][listIdc] = data;
}
}
}
fclose(fp);
m_bEnabled = true;
m_bDataPresent = !checkDefaultScalingList();
return false;
}
/** set quantized matrix coefficient for encode */
/** 函数功能 : 根据不同的TU尺寸,不同的list类型,不同的量化矩阵,生成不同的量化矩阵。
** 如果m_bEnabled为true,则使用默认的量化矩阵或者从文件中读取的量化矩阵,生成新的非均匀量化矩阵。
** 如果m_bEnabled为false,则直接使用默认的量化系数得到均匀量化矩阵。
** 调用范围 : 仅在Encoder::create()中被调用
*/
void ScalingList::setupQuantMatrices()
{
for (int size = 0; size < NUM_SIZES; size++) // 对不同的TU尺寸设置量化矩阵表
{
int width = 1 << (size + 2); // 得到TU的宽度
int ratio = width / X265_MIN(MAX_MATRIX_SIZE_NUM, width); // ratio = width / min(8, width); 假如width <= 8,ratio = 1; 假如width > 8,ratio = width/8
int stride = X265_MIN(MAX_MATRIX_SIZE_NUM, width); // stride = min(8, width)
int count = s_numCoefPerSize[size]; // 得到TU中变换系数的个数
for (int list = 0; list < NUM_LISTS; list++) // 对每个list type设置量化矩阵表
{
int32_t *coeff = m_scalingListCoef[size][list]; // 得到的量化矩阵(HEVC中规定的默认量化矩阵或者是从文件中读取的量化矩阵)
int32_t dc = m_scalingListDC[size][list]; // 得到的DC量化系数(HEVC中规定的默认DC量化系数或者是从文件中读取的DC量化系数)
for (int rem = 0; rem < NUM_REM; rem++) // 对于不同的Qp余数,设置不同的量化矩阵
{
int32_t *quantCoeff = m_quantCoef[size][list][rem]; // 得到均匀量化的量化系数
int32_t *dequantCoeff = m_dequantCoef[size][list][rem]; // 得到均匀量化的量化系数
if (m_bEnabled) // 如果使能非均匀量化
{
processScalingListEnc(coeff, quantCoeff, s_quantScales[rem] << 4, width, width, ratio, stride, dc); // 生成新的非均匀量化矩阵
processScalingListDec(coeff, dequantCoeff, s_invQuantScales[rem], width, width, ratio, stride, dc); // 生成新的非均匀反量化矩阵
}
else // 如果不支持非均匀量化,则只能使用均匀量化
{
/* flat quant and dequant coefficients */
for (int i = 0; i < count; i++) // 均匀量化表中的每一个值都设置为默认的量化系数
{ // 使用的量化系数只与Qp余数相关
quantCoeff[i] = s_quantScales[rem];
dequantCoeff[i] = s_invQuantScales[rem];
}
}
}
}
}
}
/** 函数功能 : 生成非均匀量化矩阵
** 调用范围 : 仅在ScalingList::setupQuantMatrices()中被调用
* \参数 coeff : 输入的非均匀量化矩阵
* \参数 quantcoeff : 输出的新建立的非均匀量化矩阵
* \参数 quantScales: 默认的均匀量化系数
* \参数 height : TU的高度
* \参数 width : TU的宽度
* \参数 ratio : TU尺寸与量化矩阵尺寸的比例 = TU width / min(8, TU width). 假如TU width<=8, 则ratio = 1; 假如TU height>8, ratio = TU width/8
* \参数 stride : TU的步长, = width
* \参数 dc : 输入的非均匀DC量化系数
*/
void ScalingList::processScalingListEnc(int32_t *coeff, int32_t *quantcoeff, int32_t quantScales, int height, int width,
int ratio, int stride, int32_t dc)
{
// 如果width<= 8, ratio = 1, 量化矩阵 quantcoeff = quantscale*16/coeff[stride*j+i]
// 如果width==16, ratio = 2, 量化矩阵 quantcoeff = quantscale*16/coeff[stride*j/2+i/2],由于TU过大,而量化矩阵只有8x8,所以需要对TU长和宽进行2倍的下采样,来匹配8x8的量化矩阵
// 如果width==32, ratio = 4, 量化矩阵 quantcoeff = quantscale*16/coeff[stride*j/4+i/4],对TU长和宽进行4倍的下采样,来匹配8x8的量化矩阵
for (int j = 0; j < height; j++)
for (int i = 0; i < width; i++)
quantcoeff[j * width + i] = quantScales / coeff[stride * (j / ratio) + i / ratio];
if (ratio > 1) // 对于ratio>1,也就是TU尺寸大于量化矩阵的情况,可以对DC量化系数进行单独的设置
quantcoeff[0] = quantScales / dc; // = quantscale*16/ dc
}
/** 函数功能 : 生成非均匀反量化矩阵
** 调用范围 : 仅在ScalingList::setupQuantMatrices()中被调用
* \参数 coeff : 输入的非均匀量化矩阵
* \参数 dequantcoeff : 输出的新建立的非均匀反量化矩阵
* \参数 invQuantScales : 默认的均匀反量化系数
* \参数 height : TU的高度
* \参数 width : TU的宽度
* \参数 ratio : TU尺寸与量化矩阵尺寸的比例 = TU width / min(8, TU width). 假如TU width<=8, 则ratio = 1; 假如TU height>8, ratio = TU width/8
* \参数 stride : TU的步长, = width
* \参数 dc : 输入的非均匀DC量化系数
*/
void ScalingList::processScalingListDec(int32_t *coeff, int32_t *dequantcoeff, int32_t invQuantScales, int height, int width,
int ratio, int stride, int32_t dc)
{
// 如果width<= 8, ratio = 1, 反量化矩阵 dequantcoeff = invQuantScales * coeff[stride*j+i]
// 如果width==16, ratio = 2, 反量化矩阵 dequantcoeff = invQuantScales * coeff[stride*j/2+i/2],由于TU过大,而量化矩阵只有8x8,所以需要对TU长和宽进行2倍的下采样,来匹配8x8的量化矩阵
// 如果width==32, ratio = 4, 反量化矩阵 dequantcoeff = invQuantScales * coeff[stride*j/4+i/4],对TU长和宽进行4倍的下采样,来匹配8x8的量化矩阵
for (int j = 0; j < height; j++)
for (int i = 0; i < width; i++)
dequantcoeff[j * width + i] = invQuantScales * coeff[stride * (j / ratio) + i / ratio];
if (ratio > 1) // 对于ratio>1,也就是TU尺寸大于量化矩阵的情况,可以对DC反量化系数进行单独的设置
dequantcoeff[0] = invQuantScales * dc;
}
}