Integrated Performance Primitives (Intel® IPP) 是Intel提供的一套软件库， 支持在使用intel处理器的windows、linux、mac平台上对常用的一些运算库、多媒体处理等高处理器需求的库进行硬件加速。 IPP库提供了丰富的一维信号、数据处理函数和图像处理函数。
兴趣区域 Regions of Interest in Intel IPP
IPP对图像处理的支持可以不是整张图的buffer，也可以对图像中的一个区域进行操作。我们把这个区域叫做兴趣区域ROI。 ROI由区域大小roisize、起点地址psrc、区域行宽srcStep来确定。
ippiCopyWrapBorder 用于把一个图像buffer拷贝到一个新的buffer，新buffer的尺寸会比原图像大。输入位置参数就可以决定原图像会被放在新buffer的什么位置。
zigzagFwd 用于把像素顺序按zigzag重排，在JPEG压缩中会用到，将高低频系数分开。
ippi图像处理函数接口的类型选择
同一个ipp api会随着输入数据类型和用法有几个不同的接口，形如ippiBasename[_modifiers]_types[_descriptors].
https://software.intel.com/en-us/articles/descriptor-codes-and-parameters-for-ippi-functions
小波变换DWT运算的IPP实现
调用dwt函数时要注意wavelet卷积运算进行到图像边缘的处理，ipp接口需要调用者自己根据filter的宽度和偏移，事先在图像外围做好buffer补充。 比如先用ippiCopyWrapBorder_32s_C1R补充原始图像src_buffer，之后调用ippiWTFwd_32f_C1R时传入src_buffer中原图的所在位置地址。逆向小波变换的时候也一样，先把approx，x，y，xy四个频率成分的buffer扩充，之后调用ippiWTInv_32f_C1R时传入新buffer中原数据的所在位置地址。 值得注意的是，这样做其实增加了两次内存拷贝的损耗。
IPP wavelet decomposition and reconstruction use fast polyphase algorithm, which is equivalent to traditional application of separable convolution and dyadic resampling in different order.
As source ROIs do not include border pixels required to computations, the application program have to apply a border extension model (symmetrical, wraparound or another) to ROIs of all source images filling the neighboring memory locations. Note the border sizes may be different for different source images.
/* /////////////////////////////////////////////////////////////////////////////
// Wavelet Transform Fucntions for User Filter Banks
///////////////////////////////////////////////////////////////////////////// */
/* //////////////////////////////////////////////////////////////////////
// Name: ippsWTFwdGetSize 创建DWT结构
//
// Purpose: Get sizes, in bytes, of the ippsWTFwd state structure.
//
// Parameters:
// srcType - Data type of the source vector.
// lenLow - Length of lowpass filter.
// offsLow - Input delay of lowpass filter.
// lenHigh - Length of highpass filter.
// offsHigh - Input delay of highpass filter.
// pStateSize- Pointer to the size of the ippsWTFwd state structure (in bytes).
//
// Returns:
// ippStsNoErr - Ok.
// ippStsNullPtrErr - Error when any of the specified pointers is NULL.
// ippStsSizeErr - Error when filters length is negative, or equal to zero.
// ippStsWtOffsetErr - Error when filter delay is less than (-1).
*/
IPPAPI(IppStatus, ippsWTFwdGetSize, (IppDataType srcType, int lenLow, int offsLow, int lenHigh, int offsHigh, int* pStateSize))
/* //////////////////////////////////////////////////////////////////////
// Name: ippsWTFwdInit 初始化
//
// Purpose: Initialize forward wavelet transform state structure.
//
// Parameters:
// pState - Pointer to allocated ippsWTFwd state structure.
// pTapsLow - Pointer to lowpass filter taps.
// lenLow - Length of lowpass filter.
// offsLow - Input delay of lowpass filter.
// pTapsHigh - Pointer to highpass filter taps.
// lenHigh - Length of highpass filter.
// offsHigh - Input delay of highpass filter.
//
// Returns:
// ippStsNoErr - Ok.
// ippStsNullPtrErr - Error when any of the specified pointers is NULL.
// ippStsSizeErr - Error when filters length is negative, or equal to zero.
// ippStsWtOffsetErr - Error when filter delay is less than (-1).
*/
IPPAPI(IppStatus, ippsWTFwdInit_8u32f, (IppsWTFwdState_8u32f* pState, const Ipp32f* pTapsLow, int lenLow, int offsLow, const Ipp32f* pTapsHigh, int lenHigh, int offsHigh))
IPPAPI(IppStatus, ippsWTFwdInit_16s32f, (IppsWTFwdState_16s32f* pState, const Ipp32f* pTapsLow, int lenLow, int offsLow, const Ipp32f* pTapsHigh, int lenHigh, int offsHigh))
IPPAPI(IppStatus, ippsWTFwdInit_16u32f, (IppsWTFwdState_16u32f* pState, const Ipp32f* pTapsLow, int lenLow, int offsLow, const Ipp32f* pTapsHigh, int lenHigh, int offsHigh))
IPPAPI(IppStatus, ippsWTFwdInit_32f, (IppsWTFwdState_32f* pState, const Ipp32f* pTapsLow, int lenLow, int offsLow, const Ipp32f* pTapsHigh, int lenHigh, int offsHigh))
/* //////////////////////////////////////////////////////////////////////
// Name: ippsWTFwdSetDlyLine_32f, ippsWTFwdSetDlyLine_8s32f,
// ippsWTFwdSetDlyLine_8u32f, ippsWTFwdSetDlyLine_16s32f,
// ippsWTFwdSetDlyLine_16u32f
//
// Purpose: The function copies the pointed vectors to internal delay lines.
//
// Parameters:
// pState - pointer to pState structure;
// pDlyLow - pointer to delay line for lowpass filtering;
// pDlyHigh - pointer to delay line for highpass filtering.
//
// Returns:
// ippStsNoErr - Ok;
// ippStsNullPtrErr - some of pointers pDlyLow
// or pDlyHigh vectors are NULL;
// ippStspStateMatchErr - mismatch pState structure.
//
// Notes: lengths of delay lines:
// len(pDlyLow) = lenLow + offsLow - 1;
// len(pDlyHigh) = lenHigh + offsHigh - 1;
// lenLow, offsLow, lenHigh, offsHigh - parameters
// for ippsWTFwdInitAlloc function.
*/
IPPAPI(IppStatus, ippsWTFwdSetDlyLine_8u32f, (IppsWTFwdState_8u32f* pState, const Ipp32f* pDlyLow, const Ipp32f* pDlyHigh))
IPPAPI(IppStatus, ippsWTFwdSetDlyLine_16s32f, (IppsWTFwdState_16s32f* pState, const Ipp32f* pDlyLow, const Ipp32f* pDlyHigh))
IPPAPI(IppStatus, ippsWTFwdSetDlyLine_16u32f, (IppsWTFwdState_16u32f* pState, const Ipp32f* pDlyLow, const Ipp32f* pDlyHigh))
IPPAPI(IppStatus, ippsWTFwdSetDlyLine_32f, (IppsWTFwdState_32f* pState, const Ipp32f* pDlyLow, const Ipp32f* pDlyHigh))
/* //////////////////////////////////////////////////////////////////////
// Name: ippsWTFwdGetDlyLine_32f, ippsWTFwdGetDlyLine_8s32f,
// ippsWTFwdGetDlyLine_8u32f, ippsWTFwdGetDlyLine_16s32f,
// ippsWTFwdGetDlyLine_16u32f
//
// Purpose: The function copies data from interanl delay lines
// to the pointed vectors.
// Parameters:
// pState - pointer to pState structure;
// pDlyLow - pointer to delay line for lowpass filtering;
// pDlyHigh - pointer to delay line for highpass filtering.
//
// Returns:
// ippStsNoErr - Ok;
// ippStsNullPtrErr - some of pointers pDlyLow
// or pDlyHigh vectors are NULL;
// ippStspStateMatchErr - mismatch pState structure.
//
// Notes: lengths of delay lines:
// len(pDlyLow) = lenLow + offsLow - 1;
// len(pDlyHigh) = lenHigh + offsHigh - 1;
// lenLow, offsLow, lenHigh, offsHigh - parameters
// for ippsWTFwdInitAlloc function.
*/
IPPAPI(IppStatus, ippsWTFwdGetDlyLine_8u32f, (IppsWTFwdState_8u32f* pState, Ipp32f* pDlyLow, Ipp32f* pDlyHigh))
IPPAPI(IppStatus, ippsWTFwdGetDlyLine_16s32f, (IppsWTFwdState_16s32f* pState, Ipp32f* pDlyLow, Ipp32f* pDlyHigh))
IPPAPI(IppStatus, ippsWTFwdGetDlyLine_16u32f, (IppsWTFwdState_16u32f* pState, Ipp32f* pDlyLow, Ipp32f* pDlyHigh))
IPPAPI(IppStatus, ippsWTFwdGetDlyLine_32f, (IppsWTFwdState_32f* pState, Ipp32f* pDlyLow, Ipp32f* pDlyHigh))
/* //////////////////////////////////////////////////////////////////////
// Name: ippsWTFwd_32f, ippsWTFwd_16s32f, ippsWTFwd_16u32f,
// ippsWTFwd_8s32f, ippsWTFwd_8u32f
//
// Purpose: Forward wavelet transform. 小波变换
//
// Parameters:
// pSrc - pointer to source block of data;
// pDstLow - pointer to destination block of
// "low-frequency" component;
// pDstHigh - pointer to destination block of
// "high-frequency" component;
// dstLen - length of destination;
// pState - pointer to pState structure.
//
// Returns:
// ippStsNoErr - Ok;
// ippStsNullPtrErr - some of pointers to pSrc, pDstLow
// or pDstHigh vectors are NULL;
// ippStsSizeErr - the length is less or equal zero;
// ippStspStateMatchErr - mismatch pState structure.
//
// Notes: source block length must be 2 * dstLen.
*/
IPPAPI(IppStatus, ippsWTFwd_8u32f, (const Ipp8u* pSrc, Ipp32f* pDstLow, Ipp32f* pDstHigh, int dstLen, IppsWTFwdState_8u32f* pState))
IPPAPI(IppStatus, ippsWTFwd_16s32f, (const Ipp16s* pSrc, Ipp32f* pDstLow, Ipp32f* pDstHigh, int dstLen, IppsWTFwdState_16s32f* pState))
IPPAPI(IppStatus, ippsWTFwd_16u32f, (const Ipp16u* pSrc, Ipp32f* pDstLow, Ipp32f* pDstHigh, int dstLen, IppsWTFwdState_16u32f* pState))
IPPAPI(IppStatus, ippsWTFwd_32f, (const Ipp32f* pSrc, Ipp32f* pDstLow, Ipp32f* pDstHigh, int dstLen, IppsWTFwdState_32f* pState))
/* //////////////////////////////////////////////////////////////////////
// Name: ippsWTInvGetSize
//
// Purpose: Get sizes, in bytes, of the ippsWTInv state structure.
//
// Parameters:
// dstType - Data type of the destination vector.
// lenLow - Length of lowpass filter.
// offsLow - Input delay of lowpass filter.
// lenHigh - Length of highpass filter.
// offsHigh - Input delay of highpass filter.
// pStateSize- Pointer to the size of the ippsWTInv state structure (in bytes).
//
// Returns:
// ippStsNoErr - Ok.
// ippStsNullPtrErr - Error when any of the specified pointers is NULL.
// ippStsSizeErr - Error when filters length is negative, or equal to zero.
// ippStsWtOffsetErr - Error when filter delay is less than (-1).
*/
IPPAPI(IppStatus, ippsWTInvGetSize, (IppDataType dstType, int lenLow, int offsLow, int lenHigh, int offsHigh, int* pStateSize))
/* //////////////////////////////////////////////////////////////////////
// Name: ippsWTInvInit
//
// Purpose: Initialize inverse wavelet transform state structure.
//
// Parameters:
// pState - Pointer to allocated ippsWTInv state structure.
// pTapsLow - Pointer to lowpass filter taps.
// lenLow - Length of lowpass filter.
// offsLow - Input delay of lowpass filter.
// pTapsHigh - Pointer to highpass filter taps.
// lenHigh - Length of highpass filter.
// offsHigh - Input delay of highpass filter.
//
// Returns:
// ippStsNoErr - Ok.
// ippStsNullPtrErr - Error when any of the specified pointers is NULL.
// ippStsSizeErr - Error when filters length is negative, or equal to zero.
// ippStsWtOffsetErr - Error when filter delay is less than (-1).
*/
IPPAPI(IppStatus, ippsWTInvInit_32f8u, (IppsWTInvState_32f8u* pState, const Ipp32f* pTapsLow, int lenLow, int offsLow, const Ipp32f* pTapsHigh, int lenHigh, int offsHigh))
IPPAPI(IppStatus, ippsWTInvInit_32f16u, (IppsWTInvState_32f16u* pState, const Ipp32f* pTapsLow, int lenLow, int offsLow, const Ipp32f* pTapsHigh, int lenHigh, int offsHigh))
IPPAPI(IppStatus, ippsWTInvInit_32f16s, (IppsWTInvState_32f16s* pState, const Ipp32f* pTapsLow, int lenLow, int offsLow, const Ipp32f* pTapsHigh, int lenHigh, int offsHigh))
IPPAPI(IppStatus, ippsWTInvInit_32f, (IppsWTInvState_32f* pState, const Ipp32f* pTapsLow, int lenLow, int offsLow, const Ipp32f* pTapsHigh, int lenHigh, int offsHigh))
/* //////////////////////////////////////////////////////////////////////
// Name: ippsWTInvSetDlyLine_32f, ippsWTInvSetDlyLine_32f8s,
// ippsWTInvSetDlyLine_32f8u, ippsWTInvSetDlyLine_32f16s,
// ippsWTInvSetDlyLine_32f16u
//
// Purpose: The function copies the pointed vectors to internal delay lines.
//
// Parameters:
// pState - pointer to pState structure;
// pDlyLow - pointer to delay line for lowpass filtering;
// pDlyHigh - pointer to delay line for highpass filtering.
//
// Returns:
// ippStsNoErr - Ok;
// ippStsNullPtrErr - some of pointers pDlyLow
// or pDlyHigh vectors are NULL;
// ippStspStateMatchErr - mismatch pState structure.
//
// Notes: lengths of delay lines (as "C" expression):
// len(pDlyLow) = (lenLow + offsLow - 1) / 2;
// len(pDlyHigh) = (lenHigh + offsHigh - 1) / 2;
// lenLow, offsLow, lenHigh, offsHigh - parameters
// for ippsWTInvInitAlloc function.
*/
IPPAPI(IppStatus, ippsWTInvSetDlyLine_32f8u, (IppsWTInvState_32f8u* pState, const Ipp32f* pDlyLow, const Ipp32f* pDlyHigh))
IPPAPI(IppStatus, ippsWTInvSetDlyLine_32f16s, (IppsWTInvState_32f16s* pState, const Ipp32f* pDlyLow, const Ipp32f* pDlyHigh))
IPPAPI(IppStatus, ippsWTInvSetDlyLine_32f16u, (IppsWTInvState_32f16u* pState, const Ipp32f* pDlyLow, const Ipp32f* pDlyHigh))
IPPAPI(IppStatus, ippsWTInvSetDlyLine_32f, (IppsWTInvState_32f* pState, const Ipp32f* pDlyLow, const Ipp32f* pDlyHigh))
/* //////////////////////////////////////////////////////////////////////
// Name: ippsWTInvGetDlyLine_32f, ippsWTInvGetDlyLine_32f8s,
// ippsWTInvGetDlyLine_32f8u, ippsWTInvGetDlyLine_32f16s,
// ippsWTInvGetDlyLine_32f16u
//
// Purpose: The function copies data from interanl delay lines
// to the pointed vectors.
// Parameters:
// pState - pointer to pState structure;
// pDlyLow - pointer to delay line for lowpass filtering;
// pDlyHigh - pointer to delay line for highpass filtering.
//
// Returns:
// ippStsNoErr - Ok;
// ippStsNullPtrErr - some of pointers pDlyLow
// or pDlyHigh vectors are NULL;
// ippStspStateMatchErr - mismatch pState structure.
//
// Notes: lengths of delay lines (as "C" expression):
// len(pDlyLow) = (lenLow + offsLow - 1) / 2;
// len(pDlyHigh) = (lenHigh + offsHigh - 1) / 2;
// lenLow, offsLow, lenHigh, offsHigh - parameters
// for ippsWTInvInitAlloc function.
*/
IPPAPI(IppStatus, ippsWTInvGetDlyLine_32f8u, (IppsWTInvState_32f8u* pState, Ipp32f* pDlyLow, Ipp32f* pDlyHigh))
IPPAPI(IppStatus, ippsWTInvGetDlyLine_32f16s, (IppsWTInvState_32f16s* pState, Ipp32f* pDlyLow, Ipp32f* pDlyHigh))
IPPAPI(IppStatus, ippsWTInvGetDlyLine_32f16u, (IppsWTInvState_32f16u* pState, Ipp32f* pDlyLow, Ipp32f* pDlyHigh))
IPPAPI(IppStatus, ippsWTInvGetDlyLine_32f, (IppsWTInvState_32f* pState, Ipp32f* pDlyLow, Ipp32f* pDlyHigh))
/* //////////////////////////////////////////////////////////////////////
// Name: ippsWTInv_32f, ippsWTInv_32f16s, ippsWTInv_32f16u,
// ippsWTInv_32f8u
//
// Purpose: Inverse wavelet transform.
//
// Parameters:
// srcLow - pointer to source block of "low-frequency" component;
// srcHigh - pointer to source block of "high-frequency" component;
// dstLen - length of components.
// dst - pointer to destination block of reconstructed data;
// pState - pointer to pState structure;
//
// Returns:
// ippStsNoErr - Ok;
// ippStsNullPtrErr - some of pointers to pDst pSrcLow or pSrcHigh vectors are NULL;
// ippStsSizeErr - the length is less or equal zero;
// ippStspStateMatchErr - mismatch pState structure.
//
// Notes: destination block length must be 2 * srcLen.
*/
IPPAPI(IppStatus, ippsWTInv_32f8u, (const Ipp32f* pSrcLow, const Ipp32f* pSrcHigh, int srcLen, Ipp8u* pDst, IppsWTInvState_32f8u* pState))
IPPAPI(IppStatus, ippsWTInv_32f16s, (const Ipp32f* pSrcLow, const Ipp32f* pSrcHigh, int srcLen, Ipp16s* pDst, IppsWTInvState_32f16s* pState))
IPPAPI(IppStatus, ippsWTInv_32f16u, (const Ipp32f* pSrcLow, const Ipp32f* pSrcHigh, int srcLen, Ipp16u* pDst, IppsWTInvState_32f16u* pState))
IPPAPI(IppStatus, ippsWTInv_32f, (const Ipp32f* pSrcLow, const Ipp32f* pSrcHigh, int srcLen, Ipp32f* pDst, IppsWTInvState_32f* pState))