|
@@ -1,1104 +0,0 @@
|
|
|
-/*
|
|
|
-* Copyright (C) 2016 Intel Corporation. All rights reserved.
|
|
|
-*
|
|
|
-* Redistribution and use in source and binary forms, with or without
|
|
|
-* modification, are permitted provided that the following conditions
|
|
|
-* are met:
|
|
|
-*
|
|
|
-* * Redistributions of source code must retain the above copyright
|
|
|
-* notice, this list of conditions and the following disclaimer.
|
|
|
-* * Redistributions in binary form must reproduce the above copyright
|
|
|
-* notice, this list of conditions and the following disclaimer in
|
|
|
-* the documentation and/or other materials provided with the
|
|
|
-* distribution.
|
|
|
-* * Neither the name of Intel Corporation nor the names of its
|
|
|
-* contributors may be used to endorse or promote products derived
|
|
|
-* from this software without specific prior written permission.
|
|
|
-*
|
|
|
-* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
|
|
-* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
|
|
-* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
|
|
-* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
|
|
|
-* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
|
|
-* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
|
|
-* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
|
|
-* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
|
|
-* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
|
-* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
|
|
-* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
|
-*
|
|
|
-*/
|
|
|
-
|
|
|
-#ifndef __IPPTYPES_H__
|
|
|
-#define __IPPTYPES_H__
|
|
|
-
|
|
|
-#ifdef __cplusplus
|
|
|
-extern "C" {
|
|
|
-#endif
|
|
|
-
|
|
|
-/*****************************************************************************/
|
|
|
-/* Below are ippCore domain specific definitions */
|
|
|
-/*****************************************************************************/
|
|
|
-#if !defined( _OWN_BLDPCS )
|
|
|
-
|
|
|
-#define ippCPUID_MMX 0x00000001 /* Intel Architecture MMX technology supported */
|
|
|
-#define ippCPUID_SSE 0x00000002 /* Streaming SIMD Extensions */
|
|
|
-#define ippCPUID_SSE2 0x00000004 /* Streaming SIMD Extensions 2 */
|
|
|
-#define ippCPUID_SSE3 0x00000008 /* Streaming SIMD Extensions 3 */
|
|
|
-#define ippCPUID_SSSE3 0x00000010 /* Supplemental Streaming SIMD Extensions 3 */
|
|
|
-#define ippCPUID_MOVBE 0x00000020 /* The processor supports MOVBE instruction */
|
|
|
-#define ippCPUID_SSE41 0x00000040 /* Streaming SIMD Extensions 4.1 */
|
|
|
-#define ippCPUID_SSE42 0x00000080 /* Streaming SIMD Extensions 4.2 */
|
|
|
-#define ippCPUID_AVX 0x00000100 /* Advanced Vector Extensions instruction set */
|
|
|
-#define ippAVX_ENABLEDBYOS 0x00000200 /* The operating system supports AVX */
|
|
|
-#define ippCPUID_AES 0x00000400 /* AES instruction */
|
|
|
-#define ippCPUID_CLMUL 0x00000800 /* PCLMULQDQ instruction */
|
|
|
-#define ippCPUID_ABR 0x00001000 /* Reserved */
|
|
|
-#define ippCPUID_RDRAND 0x00002000 /* Read Random Number instructions */
|
|
|
-#define ippCPUID_F16C 0x00004000 /* Float16 instructions */
|
|
|
-#define ippCPUID_AVX2 0x00008000 /* Advanced Vector Extensions 2 instruction set */
|
|
|
-#define ippCPUID_ADCOX 0x00010000 /* ADCX and ADOX instructions */
|
|
|
-#define ippCPUID_RDSEED 0x00020000 /* The RDSEED instruction */
|
|
|
-#define ippCPUID_PREFETCHW 0x00040000 /* The PREFETCHW instruction */
|
|
|
-#define ippCPUID_SHA 0x00080000 /* Intel (R) SHA Extensions */
|
|
|
-#define ippCPUID_AVX512F 0x00100000 /* AVX-512 Foundation instructions */
|
|
|
-#define ippCPUID_AVX512CD 0x00200000 /* AVX-512 Conflict Detection instructions */
|
|
|
-#define ippCPUID_AVX512ER 0x00400000 /* AVX-512 Exponential & Reciprocal instructions*/
|
|
|
-#define ippCPUID_AVX512PF 0x00800000 /* AVX-512 Prefetch instructions */
|
|
|
-#define ippCPUID_AVX512BW 0x01000000 /* AVX-512 Byte & Word instructions */
|
|
|
-#define ippCPUID_AVX512DQ 0x02000000 /* AVX-512 DWord & QWord instructions */
|
|
|
-#define ippCPUID_AVX512VL 0x04000000 /* AVX-512 Vector Length extensions */
|
|
|
-#define ippCPUID_KNC 0x80000000 /* Intel(R) Xeon Phi(TM) Coprocessor */
|
|
|
-#if defined( _WIN32 ) || defined ( _WIN64 )
|
|
|
- #define ippCPUID_NOCHECK 0x8000000000000000L /* Force ippSetCpuFeatures to set CPU features without check */
|
|
|
-#else
|
|
|
- #define ippCPUID_NOCHECK 0x8000000000000000LL
|
|
|
-#endif
|
|
|
-
|
|
|
-#if defined( _WIN32 ) || defined ( _WIN64 )
|
|
|
- #define ippCPUID_GETINFO_A 0x616f666e69746567L /* Force ipp_GetCpuFeatures to work as cpuid instruction */
|
|
|
-#else
|
|
|
- #define ippCPUID_GETINFO_A 0x616f666e69746567LL
|
|
|
-#endif
|
|
|
-
|
|
|
-#define IPP_COUNT_OF( obj ) (sizeof(obj)/sizeof(obj[0]))
|
|
|
-
|
|
|
-/*****************************************************************************/
|
|
|
-/* Below are ippSP domain specific definitions */
|
|
|
-/*****************************************************************************/
|
|
|
-typedef enum {
|
|
|
- ippRndZero,
|
|
|
- ippRndNear,
|
|
|
- ippRndFinancial,
|
|
|
- ippRndHintAccurate=0x10
|
|
|
-} IppRoundMode;
|
|
|
-
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippAlgHintNone,
|
|
|
- ippAlgHintFast,
|
|
|
- ippAlgHintAccurate
|
|
|
-} IppHintAlgorithm;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippCmpLess,
|
|
|
- ippCmpLessEq,
|
|
|
- ippCmpEq,
|
|
|
- ippCmpGreaterEq,
|
|
|
- ippCmpGreater
|
|
|
-} IppCmpOp;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippAlgAuto = 0x00000000,
|
|
|
- ippAlgDirect = 0x00000001,
|
|
|
- ippAlgFFT = 0x00000002,
|
|
|
- ippAlgMask = 0x000000FF
|
|
|
-} IppAlgType;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippsNormNone = 0x00000000, /* default */
|
|
|
- ippsNormA = 0x00000100, /* biased normalization */
|
|
|
- ippsNormB = 0x00000200, /* unbiased normalization */
|
|
|
- ippsNormMask = 0x0000FF00
|
|
|
-} IppsNormOp;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippNormInf = 0x00000001,
|
|
|
- ippNormL1 = 0x00000002,
|
|
|
- ippNormL2 = 0x00000004
|
|
|
-} IppNormType;
|
|
|
-
|
|
|
-enum {
|
|
|
- IPP_FFT_DIV_FWD_BY_N = 1,
|
|
|
- IPP_FFT_DIV_INV_BY_N = 2,
|
|
|
- IPP_FFT_DIV_BY_SQRTN = 4,
|
|
|
- IPP_FFT_NODIV_BY_ANY = 8
|
|
|
-};
|
|
|
-
|
|
|
-enum {
|
|
|
- IPP_DIV_FWD_BY_N = 1,
|
|
|
- IPP_DIV_INV_BY_N = 2,
|
|
|
- IPP_DIV_BY_SQRTN = 4,
|
|
|
- IPP_NODIV_BY_ANY = 8
|
|
|
-};
|
|
|
-
|
|
|
-typedef struct {
|
|
|
- Ipp32f rho;
|
|
|
- Ipp32f theta;
|
|
|
-} IppPointPolar;
|
|
|
-
|
|
|
-typedef enum {ippWinBartlett,ippWinBlackman,ippWinHamming,ippWinHann,ippWinRect} IppWinType;
|
|
|
-
|
|
|
-typedef enum { ippButterworth, ippChebyshev1 } IppsIIRFilterType;
|
|
|
-
|
|
|
-typedef enum { ippZCR=0, ippZCXor, ippZCC } IppsZCType;
|
|
|
-
|
|
|
-#if !defined( _OWN_BLDPCS )
|
|
|
-
|
|
|
-typedef struct {
|
|
|
- int left;
|
|
|
- int right;
|
|
|
-} IppsROI;
|
|
|
-
|
|
|
-typedef struct RandUniState_8u IppsRandUniState_8u;
|
|
|
-typedef struct RandUniState_16s IppsRandUniState_16s;
|
|
|
-typedef struct RandUniState_32f IppsRandUniState_32f;
|
|
|
-typedef struct RandUniState_64f IppsRandUniState_64f;
|
|
|
-
|
|
|
-typedef struct RandGaussState_8u IppsRandGaussState_8u;
|
|
|
-typedef struct RandGaussState_16s IppsRandGaussState_16s;
|
|
|
-typedef struct RandGaussState_32f IppsRandGaussState_32f;
|
|
|
-typedef struct RandGaussState_64f IppsRandGaussState_64f;
|
|
|
-
|
|
|
-typedef struct FFTSpec_C_32fc IppsFFTSpec_C_32fc;
|
|
|
-typedef struct FFTSpec_C_32f IppsFFTSpec_C_32f;
|
|
|
-typedef struct FFTSpec_R_32f IppsFFTSpec_R_32f;
|
|
|
-
|
|
|
-typedef struct FFTSpec_C_64fc IppsFFTSpec_C_64fc;
|
|
|
-typedef struct FFTSpec_C_64f IppsFFTSpec_C_64f;
|
|
|
-typedef struct FFTSpec_R_64f IppsFFTSpec_R_64f;
|
|
|
-
|
|
|
-typedef struct DFTSpec_C_32fc IppsDFTSpec_C_32fc;
|
|
|
-typedef struct DFTSpec_C_32f IppsDFTSpec_C_32f;
|
|
|
-typedef struct DFTSpec_R_32f IppsDFTSpec_R_32f;
|
|
|
-
|
|
|
-typedef struct DFTSpec_C_64fc IppsDFTSpec_C_64fc;
|
|
|
-typedef struct DFTSpec_C_64f IppsDFTSpec_C_64f;
|
|
|
-typedef struct DFTSpec_R_64f IppsDFTSpec_R_64f;
|
|
|
-
|
|
|
-typedef struct DCTFwdSpec_32f IppsDCTFwdSpec_32f;
|
|
|
-typedef struct DCTInvSpec_32f IppsDCTInvSpec_32f;
|
|
|
-
|
|
|
-typedef struct DCTFwdSpec_64f IppsDCTFwdSpec_64f;
|
|
|
-typedef struct DCTInvSpec_64f IppsDCTInvSpec_64f;
|
|
|
-
|
|
|
-typedef struct sWTFwdState_32f IppsWTFwdState_32f;
|
|
|
-typedef struct sWTFwdState_8u32f IppsWTFwdState_8u32f;
|
|
|
-typedef struct sWTFwdState_16s32f IppsWTFwdState_16s32f;
|
|
|
-typedef struct sWTFwdState_16u32f IppsWTFwdState_16u32f;
|
|
|
-typedef struct sWTInvState_32f IppsWTInvState_32f;
|
|
|
-typedef struct sWTInvState_32f8u IppsWTInvState_32f8u;
|
|
|
-typedef struct sWTInvState_32f16s IppsWTInvState_32f16s;
|
|
|
-typedef struct sWTInvState_32f16u IppsWTInvState_32f16u;
|
|
|
-
|
|
|
-typedef struct IIRState_32f IppsIIRState_32f;
|
|
|
-typedef struct IIRState_32fc IppsIIRState_32fc;
|
|
|
-typedef struct IIRState32f_16s IppsIIRState32f_16s;
|
|
|
-typedef struct IIRState32fc_16sc IppsIIRState32fc_16sc;
|
|
|
-typedef struct IIRState_64f IppsIIRState_64f;
|
|
|
-typedef struct IIRState_64fc IppsIIRState_64fc;
|
|
|
-typedef struct IIRState64f_32f IppsIIRState64f_32f;
|
|
|
-typedef struct IIRState64fc_32fc IppsIIRState64fc_32fc;
|
|
|
-typedef struct IIRState64f_32s IppsIIRState64f_32s;
|
|
|
-typedef struct IIRState64fc_32sc IppsIIRState64fc_32sc;
|
|
|
-typedef struct IIRState64f_16s IppsIIRState64f_16s;
|
|
|
-typedef struct IIRState64fc_16sc IppsIIRState64fc_16sc;
|
|
|
-
|
|
|
-typedef struct FIRSpec_32f IppsFIRSpec_32f;
|
|
|
-typedef struct FIRSpec_64f IppsFIRSpec_64f;
|
|
|
-typedef struct FIRSpec_32fc IppsFIRSpec_32fc;
|
|
|
-typedef struct FIRSpec_64fc IppsFIRSpec_64fc;
|
|
|
-
|
|
|
-typedef struct FIRLMSState_32f IppsFIRLMSState_32f;
|
|
|
-typedef struct FIRLMSState32f_16s IppsFIRLMSState32f_16s;
|
|
|
-
|
|
|
-typedef struct HilbertSpec IppsHilbertSpec;
|
|
|
-
|
|
|
-typedef struct FIRSparseState_32f IppsFIRSparseState_32f;
|
|
|
-typedef struct IIRSparseState_32f IppsIIRSparseState_32f;
|
|
|
-
|
|
|
-typedef struct ResamplingPolyphase_16s IppsResamplingPolyphase_16s;
|
|
|
-typedef struct ResamplingPolyphaseFixed_16s IppsResamplingPolyphaseFixed_16s;
|
|
|
-typedef struct ResamplingPolyphase_32f IppsResamplingPolyphase_32f;
|
|
|
-typedef struct ResamplingPolyphaseFixed_32f IppsResamplingPolyphaseFixed_32f;
|
|
|
-
|
|
|
-#endif /* _OWN_BLDPCS */
|
|
|
-
|
|
|
-/*****************************************************************************/
|
|
|
-/* Below are ippIP domain specific definitions */
|
|
|
-/*****************************************************************************/
|
|
|
-#define IPP_TEMPORAL_COPY 0x0
|
|
|
-#define IPP_NONTEMPORAL_STORE 0x01
|
|
|
-#define IPP_NONTEMPORAL_LOAD 0x02
|
|
|
-
|
|
|
-typedef int IppEnum;
|
|
|
-
|
|
|
-#define IPP_DEG_TO_RAD( deg ) ( (deg)/180.0 * IPP_PI )
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippiNormNone = 0x00000000, /* default */
|
|
|
- ippiNorm = 0x00000100, /* normalized form */
|
|
|
- ippiNormCoefficient = 0x00000200, /* correlation coefficient in the range [-1.0 ... 1.0] */
|
|
|
- ippiNormMask = 0x0000FF00
|
|
|
-} IppiNormOp;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippiROIFull = 0x00000000,
|
|
|
- ippiROIValid = 0x00010000,
|
|
|
- ippiROISame = 0x00020000,
|
|
|
- ippiROIMask = 0x00FF0000
|
|
|
-} IppiROIShape;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippC0 = 0,
|
|
|
- ippC1 = 1,
|
|
|
- ippC2 = 2,
|
|
|
- ippC3 = 3,
|
|
|
- ippC4 = 4,
|
|
|
- ippP2 = 5,
|
|
|
- ippP3 = 6,
|
|
|
- ippP4 = 7,
|
|
|
- ippAC1 = 8,
|
|
|
- ippAC4 = 9,
|
|
|
- ippA0C4 = 10,
|
|
|
- ippAP4 = 11
|
|
|
-} IppChannels;
|
|
|
-
|
|
|
-typedef enum _IppiBorderType {
|
|
|
- ippBorderConst = 0,
|
|
|
- ippBorderRepl = 1,
|
|
|
- ippBorderWrap = 2,
|
|
|
- ippBorderMirror = 3, /* left border: 012... -> 21012... */
|
|
|
- ippBorderMirrorR = 4, /* left border: 012... -> 210012... */
|
|
|
- ippBorderInMem = 6,
|
|
|
- ippBorderTransp = 7,
|
|
|
- ippBorderInMemTop = 0x0010,
|
|
|
- ippBorderInMemBottom = 0x0020,
|
|
|
- ippBorderInMemLeft = 0x0040,
|
|
|
- ippBorderInMemRight = 0x0080
|
|
|
-} IppiBorderType;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippAxsHorizontal,
|
|
|
- ippAxsVertical,
|
|
|
- ippAxsBoth,
|
|
|
- ippAxs45,
|
|
|
- ippAxs135
|
|
|
-} IppiAxis;
|
|
|
-
|
|
|
-typedef struct {
|
|
|
- int x;
|
|
|
- int y;
|
|
|
- int width;
|
|
|
- int height;
|
|
|
-} IppiRect;
|
|
|
-
|
|
|
-typedef struct {
|
|
|
- int x;
|
|
|
- int y;
|
|
|
-} IppiPoint;
|
|
|
-
|
|
|
-typedef struct {
|
|
|
- int width;
|
|
|
- int height;
|
|
|
-} IppiSize;
|
|
|
-
|
|
|
-typedef struct {
|
|
|
- Ipp32f x;
|
|
|
- Ipp32f y;
|
|
|
-} IppiPoint_32f;
|
|
|
-
|
|
|
-typedef enum _IppiMaskSize {
|
|
|
- ippMskSize1x3 = 13,
|
|
|
- ippMskSize1x5 = 15,
|
|
|
- ippMskSize3x1 = 31,
|
|
|
- ippMskSize3x3 = 33,
|
|
|
- ippMskSize5x1 = 51,
|
|
|
- ippMskSize5x5 = 55
|
|
|
-} IppiMaskSize;
|
|
|
-
|
|
|
-enum {
|
|
|
- IPPI_INTER_NN = 1,
|
|
|
- IPPI_INTER_LINEAR = 2,
|
|
|
- IPPI_INTER_CUBIC = 4,
|
|
|
- IPPI_INTER_CUBIC2P_BSPLINE, /* two-parameter cubic filter (B=1, C=0) */
|
|
|
- IPPI_INTER_CUBIC2P_CATMULLROM, /* two-parameter cubic filter (B=0, C=1/2) */
|
|
|
- IPPI_INTER_CUBIC2P_B05C03, /* two-parameter cubic filter (B=1/2, C=3/10) */
|
|
|
- IPPI_INTER_SUPER = 8,
|
|
|
- IPPI_INTER_LANCZOS = 16,
|
|
|
- IPPI_ANTIALIASING = (1 << 29),
|
|
|
- IPPI_SUBPIXEL_EDGE = (1 << 30),
|
|
|
- IPPI_SMOOTH_EDGE = (1 << 31)
|
|
|
-};
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippNearest = IPPI_INTER_NN,
|
|
|
- ippLinear = IPPI_INTER_LINEAR,
|
|
|
- ippCubic = IPPI_INTER_CUBIC2P_CATMULLROM,
|
|
|
- ippLanczos = IPPI_INTER_LANCZOS,
|
|
|
- ippHahn = 0,
|
|
|
- ippSuper = IPPI_INTER_SUPER
|
|
|
-} IppiInterpolationType;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippPolyphase_1_2,
|
|
|
- ippPolyphase_3_5,
|
|
|
- ippPolyphase_2_3,
|
|
|
- ippPolyphase_7_10,
|
|
|
- ippPolyphase_3_4
|
|
|
-} IppiFraction;
|
|
|
-
|
|
|
-enum {
|
|
|
- IPP_FASTN_ORIENTATION = 0x0001,
|
|
|
- IPP_FASTN_NMS = 0x0002,
|
|
|
- IPP_FASTN_CIRCLE = 0X0004,
|
|
|
- IPP_FASTN_SCORE_MODE0 = 0X0020
|
|
|
-};
|
|
|
-
|
|
|
-#if !defined( _OWN_BLDPCS )
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippAlphaOver,
|
|
|
- ippAlphaIn,
|
|
|
- ippAlphaOut,
|
|
|
- ippAlphaATop,
|
|
|
- ippAlphaXor,
|
|
|
- ippAlphaPlus,
|
|
|
- ippAlphaOverPremul,
|
|
|
- ippAlphaInPremul,
|
|
|
- ippAlphaOutPremul,
|
|
|
- ippAlphaATopPremul,
|
|
|
- ippAlphaXorPremul,
|
|
|
- ippAlphaPlusPremul
|
|
|
-} IppiAlphaType;
|
|
|
-
|
|
|
-typedef struct DeconvFFTState_32f_C1R IppiDeconvFFTState_32f_C1R;
|
|
|
-typedef struct DeconvFFTState_32f_C3R IppiDeconvFFTState_32f_C3R;
|
|
|
-typedef struct DeconvLR_32f_C1R IppiDeconvLR_32f_C1R;
|
|
|
-typedef struct DeconvLR_32f_C3R IppiDeconvLR_32f_C3R;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippiFilterBilateralGauss = 100,
|
|
|
- ippiFilterBilateralGaussFast = 101
|
|
|
-} IppiFilterBilateralType;
|
|
|
-
|
|
|
-typedef struct FilterBilateralSpec IppiFilterBilateralSpec;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippDistNormL1 = 0x00000002
|
|
|
-} IppiDistanceMethodType;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippResizeFilterHann,
|
|
|
- ippResizeFilterLanczos
|
|
|
-} IppiResizeFilterType;
|
|
|
-
|
|
|
-typedef struct ResizeFilterState IppiResizeFilterState;
|
|
|
-
|
|
|
-typedef struct {
|
|
|
- Ipp32u borderLeft;
|
|
|
- Ipp32u borderTop;
|
|
|
- Ipp32u borderRight;
|
|
|
- Ipp32u borderBottom;
|
|
|
-} IppiBorderSize;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippWarpForward,
|
|
|
- ippWarpBackward,
|
|
|
-} IppiWarpDirection;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippWarpAffine,
|
|
|
- ippWarpPerspective,
|
|
|
- ippWarpBilinear,
|
|
|
-} IppiWarpTransformType;
|
|
|
-
|
|
|
-
|
|
|
-typedef struct ResizeSpec_32f IppiResizeSpec_32f;
|
|
|
-typedef struct ResizeYUV422Spec IppiResizeYUV422Spec;
|
|
|
-typedef struct ResizeYUV420Spec IppiResizeYUV420Spec;
|
|
|
-
|
|
|
-typedef struct ResizeSpec_64f IppiResizeSpec_64f;
|
|
|
-
|
|
|
-typedef struct IppiWarpSpec IppiWarpSpec;
|
|
|
-
|
|
|
-typedef struct FilterBorderSpec IppiFilterBorderSpec;
|
|
|
-
|
|
|
-typedef struct ThresholdAdaptiveSpec IppiThresholdAdaptiveSpec;
|
|
|
-
|
|
|
-typedef struct HistogramSpec IppiHistogramSpec;
|
|
|
-
|
|
|
-typedef struct {
|
|
|
- int cvCompatible; /* openCV compatible output format */
|
|
|
- int cellSize; /* squre cell size (pixels) */
|
|
|
- int blockSize; /* square block size (pixels) */
|
|
|
- int blockStride; /* block displacement (the same for x- and y- directions) */
|
|
|
- int nbins; /* required number of bins */
|
|
|
- Ipp32f sigma; /* gaussian factor of HOG block weights */
|
|
|
- Ipp32f l2thresh; /* normalization factor */
|
|
|
- IppiSize winSize; /* detection window size (pixels) */
|
|
|
-} IppiHOGConfig;
|
|
|
-
|
|
|
-typedef struct FFT2DSpec_C_32fc IppiFFTSpec_C_32fc;
|
|
|
-typedef struct FFT2DSpec_R_32f IppiFFTSpec_R_32f;
|
|
|
-
|
|
|
-typedef struct DFT2DSpec_C_32fc IppiDFTSpec_C_32fc;
|
|
|
-typedef struct DFT2DSpec_R_32f IppiDFTSpec_R_32f;
|
|
|
-
|
|
|
-typedef struct DCT2DFwdSpec_32f IppiDCTFwdSpec_32f;
|
|
|
-typedef struct DCT2DInvSpec_32f IppiDCTInvSpec_32f;
|
|
|
-
|
|
|
-typedef struct iWTFwdSpec_32f_C1R IppiWTFwdSpec_32f_C1R;
|
|
|
-typedef struct iWTInvSpec_32f_C1R IppiWTInvSpec_32f_C1R;
|
|
|
-typedef struct iWTFwdSpec_32f_C3R IppiWTFwdSpec_32f_C3R;
|
|
|
-typedef struct iWTInvSpec_32f_C3R IppiWTInvSpec_32f_C3R;
|
|
|
-
|
|
|
-typedef struct MomentState64f IppiMomentState_64f;
|
|
|
-typedef Ipp64f IppiHuMoment_64f[7];
|
|
|
-
|
|
|
-typedef struct LUT_Spec IppiLUT_Spec;
|
|
|
-
|
|
|
-#define IPP_HOG_MAX_CELL (16) /* max size of CELL */
|
|
|
-#define IPP_HOG_MAX_BLOCK (64) /* max size of BLOCK */
|
|
|
-#define IPP_HOG_MAX_BINS (16) /* max number of BINS */
|
|
|
-
|
|
|
-typedef struct _ipHOG IppiHOGSpec;
|
|
|
-
|
|
|
-#endif /* _OWN_BLDPCS */
|
|
|
-
|
|
|
-
|
|
|
- /**** Below are 3D Image (Volume) Processing specific definitions ****/
|
|
|
-
|
|
|
-typedef struct {
|
|
|
- int width;
|
|
|
- int height;
|
|
|
- int depth;
|
|
|
-} IpprVolume;
|
|
|
-
|
|
|
-typedef struct {
|
|
|
- int x;
|
|
|
- int y;
|
|
|
- int z;
|
|
|
- int width;
|
|
|
- int height;
|
|
|
- int depth;
|
|
|
-} IpprCuboid;
|
|
|
-
|
|
|
-typedef struct {
|
|
|
- int x;
|
|
|
- int y;
|
|
|
- int z;
|
|
|
-} IpprPoint;
|
|
|
-
|
|
|
-/*****************************************************************************/
|
|
|
-/* Below are ippCV domain specific definitions */
|
|
|
-/*****************************************************************************/
|
|
|
-
|
|
|
-typedef enum _IppiDifferentialKernel
|
|
|
-{
|
|
|
- ippFilterSobelVert,
|
|
|
- ippFilterSobelHoriz,
|
|
|
- ippFilterSobel,
|
|
|
- ippFilterScharrVert,
|
|
|
- ippFilterScharrHoriz,
|
|
|
- ippFilterScharr,
|
|
|
- ippFilterCentralDiffVert,
|
|
|
- ippFilterCentralDiffHoriz,
|
|
|
- ippFilterCentralDiff,
|
|
|
-}IppiDifferentialKernel;
|
|
|
-
|
|
|
-#if !defined( _OWN_BLDPCS )
|
|
|
-
|
|
|
-typedef enum _IppiKernelType {
|
|
|
- ippKernelSobel = 0,
|
|
|
- ippKernelScharr = 1,
|
|
|
- ippKernelSobelNeg = 2
|
|
|
-} IppiKernelType;
|
|
|
-
|
|
|
-typedef enum _IppiNorm {
|
|
|
- ippiNormInf = 0,
|
|
|
- ippiNormL1 = 1,
|
|
|
- ippiNormL2 = 2,
|
|
|
- ippiNormFM = 3
|
|
|
-} IppiNorm;
|
|
|
-
|
|
|
-typedef struct ipcvMorphState IppiMorphState;
|
|
|
-typedef struct ipcvMorphAdvState IppiMorphAdvState;
|
|
|
-typedef struct ipcvMorphGrayState_8u IppiMorphGrayState_8u;
|
|
|
-typedef struct ipcvMorphGrayState_32f IppiMorphGrayState_32f;
|
|
|
-
|
|
|
-typedef struct ipcvConvState IppiConvState;
|
|
|
-
|
|
|
-typedef struct _IppiConnectedComp {
|
|
|
- Ipp64f area; /* area of the segmented component */
|
|
|
- Ipp64f value[3];/* gray scale value of the segmented component */
|
|
|
- IppiRect rect; /* bounding rectangle of the segmented component */
|
|
|
-} IppiConnectedComp;
|
|
|
-
|
|
|
-typedef struct PyramidState IppiPyramidState;
|
|
|
-
|
|
|
-typedef IppiPyramidState IppiPyramidDownState_8u_C1R;
|
|
|
-typedef IppiPyramidState IppiPyramidDownState_16u_C1R;
|
|
|
-typedef IppiPyramidState IppiPyramidDownState_32f_C1R;
|
|
|
-typedef IppiPyramidState IppiPyramidDownState_8u_C3R;
|
|
|
-typedef IppiPyramidState IppiPyramidDownState_16u_C3R;
|
|
|
-typedef IppiPyramidState IppiPyramidDownState_32f_C3R;
|
|
|
-typedef IppiPyramidState IppiPyramidUpState_8u_C1R;
|
|
|
-typedef IppiPyramidState IppiPyramidUpState_16u_C1R;
|
|
|
-typedef IppiPyramidState IppiPyramidUpState_32f_C1R;
|
|
|
-typedef IppiPyramidState IppiPyramidUpState_8u_C3R;
|
|
|
-typedef IppiPyramidState IppiPyramidUpState_16u_C3R;
|
|
|
-typedef IppiPyramidState IppiPyramidUpState_32f_C3R;
|
|
|
-
|
|
|
-
|
|
|
-typedef struct _IppiPyramid {
|
|
|
- Ipp8u **pImage;
|
|
|
- IppiSize *pRoi;
|
|
|
- Ipp64f *pRate;
|
|
|
- int *pStep;
|
|
|
- Ipp8u *pState;
|
|
|
- int level;
|
|
|
-} IppiPyramid;
|
|
|
-
|
|
|
-typedef struct OptFlowPyrLK IppiOptFlowPyrLK;
|
|
|
-
|
|
|
-typedef IppiOptFlowPyrLK IppiOptFlowPyrLK_8u_C1R;
|
|
|
-typedef IppiOptFlowPyrLK IppiOptFlowPyrLK_16u_C1R;
|
|
|
-typedef IppiOptFlowPyrLK IppiOptFlowPyrLK_32f_C1R;
|
|
|
-
|
|
|
-typedef struct ipcvHaarClassifier_32f IppiHaarClassifier_32f;
|
|
|
-typedef struct ipcvHaarClassifier_32s IppiHaarClassifier_32s;
|
|
|
-
|
|
|
-typedef struct ipcvFGHistogramState_8u_C1R IppFGHistogramState_8u_C1R;
|
|
|
-typedef struct ipcvFGHistogramState_8u_C3R IppFGHistogramState_8u_C3R;
|
|
|
-
|
|
|
-typedef struct ipcvFGGaussianState_8u_C1R IppFGGaussianState_8u_C1R;
|
|
|
-typedef struct ipcvFGGaussianState_8u_C3R IppFGGaussianState_8u_C3R;
|
|
|
-
|
|
|
-typedef enum _IppiInpaintFlag {
|
|
|
- IPP_INPAINT_TELEA = 0,
|
|
|
- IPP_INPAINT_NS = 1
|
|
|
-} IppiInpaintFlag;
|
|
|
-
|
|
|
-typedef struct ippcvFilterGaussianSpec IppFilterGaussianSpec;
|
|
|
-
|
|
|
-typedef struct ipcvInpaintState_8u IppiInpaintState_8u_C1R;
|
|
|
-typedef struct ipcvInpaintState_8u IppiInpaintState_8u_C3R;
|
|
|
-
|
|
|
-typedef struct HoughProbSpec IppiHoughProbSpec;
|
|
|
-
|
|
|
-typedef struct FastNSpec IppiFastNSpec;
|
|
|
-
|
|
|
-typedef struct _IppiCornerFastN {
|
|
|
- int x;
|
|
|
- int y;
|
|
|
- int cornerType;
|
|
|
- int orientation;
|
|
|
- float angle;
|
|
|
- float score;
|
|
|
-} IppiCornerFastN;
|
|
|
-
|
|
|
-typedef struct FGMMState_8u_C3R IppFGMMState_8u_C3R;
|
|
|
-
|
|
|
-typedef struct
|
|
|
-{
|
|
|
- unsigned int numFrames;
|
|
|
- unsigned int maxNGauss;
|
|
|
- Ipp32f varInit;
|
|
|
- Ipp32f varMin;
|
|
|
- Ipp32f varMax;
|
|
|
- Ipp32f varWBRatio;
|
|
|
- Ipp32f bckgThr;
|
|
|
- Ipp32f varNGRatio;
|
|
|
- Ipp32f reduction;
|
|
|
- Ipp8u shadowValue;
|
|
|
- char shadowFlag;
|
|
|
- Ipp32f shadowRatio;
|
|
|
-} IppFGMModel;
|
|
|
-
|
|
|
-#endif /* _OWN_BLDPCS */
|
|
|
-
|
|
|
-#define IPP_SEGMENT_QUEUE 0x01
|
|
|
-#define IPP_SEGMENT_DISTANCE 0x02
|
|
|
-#define IPP_SEGMENT_BORDER_4 0x40
|
|
|
-#define IPP_SEGMENT_BORDER_8 0x80
|
|
|
-
|
|
|
-#define IPP_TRUNC(a,b) ((a)&~((b)-1))
|
|
|
-#define IPP_APPEND(a,b) (((a)+(b)-1)&~((b)-1))
|
|
|
-
|
|
|
-/*****************************************************************************/
|
|
|
-/* Below are ippCC domain specific definitions */
|
|
|
-/*****************************************************************************/
|
|
|
-enum {
|
|
|
- IPP_UPPER = 1,
|
|
|
- IPP_LEFT = 2,
|
|
|
- IPP_CENTER = 4,
|
|
|
- IPP_RIGHT = 8,
|
|
|
- IPP_LOWER = 16,
|
|
|
- IPP_UPPER_LEFT = 32,
|
|
|
- IPP_UPPER_RIGHT = 64,
|
|
|
- IPP_LOWER_LEFT = 128,
|
|
|
- IPP_LOWER_RIGHT = 256
|
|
|
-};
|
|
|
-
|
|
|
-#if !defined( _OWN_BLDPCS )
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippDitherNone,
|
|
|
- ippDitherFS,
|
|
|
- ippDitherJJN,
|
|
|
- ippDitherStucki,
|
|
|
- ippDitherBayer
|
|
|
-} IppiDitherType;
|
|
|
-
|
|
|
-#endif /* _OWN_BLDPCS */
|
|
|
-
|
|
|
-/*****************************************************************************/
|
|
|
-/* Below are ippCH domain specific definitions */
|
|
|
-/*****************************************************************************/
|
|
|
-
|
|
|
-#if !defined( _OWN_BLDPCS )
|
|
|
-
|
|
|
-typedef struct {
|
|
|
- void *pFind;
|
|
|
- int lenFind;
|
|
|
-} IppRegExpFind;
|
|
|
-
|
|
|
-typedef struct RegExpState IppRegExpState;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippFmtASCII = 0,
|
|
|
- ippFmtUTF8
|
|
|
-} IppRegExpFormat;
|
|
|
-
|
|
|
-typedef struct RegExpReplaceState IppRegExpReplaceState;
|
|
|
-
|
|
|
-#endif /* _OWN_BLDPCS */
|
|
|
-
|
|
|
-/*****************************************************************************/
|
|
|
-/* Below are ippDC domain specific definitions */
|
|
|
-/*****************************************************************************/
|
|
|
-
|
|
|
-#if !defined ( _OWN_BLDPCS )
|
|
|
-
|
|
|
-typedef struct MTFState_8u IppMTFState_8u;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- ippBWTItohTanakaLimSort,
|
|
|
- ippBWTItohTanakaUnlimSort,
|
|
|
- ippBWTSuffixSort,
|
|
|
- ippBWTAutoSort
|
|
|
-} IppBWTSortAlgorithmHint;
|
|
|
-
|
|
|
-typedef struct LZSSState_8u IppLZSSState_8u;
|
|
|
-
|
|
|
-typedef struct LZ77State_8u IppLZ77State_8u;
|
|
|
-typedef enum{
|
|
|
- IppLZ77FastCompr,
|
|
|
- IppLZ77AverageCompr,
|
|
|
- IppLZ77BestCompr
|
|
|
-} IppLZ77ComprLevel;
|
|
|
-typedef enum{
|
|
|
- IppLZ77NoChcksm,
|
|
|
- IppLZ77Adler32,
|
|
|
- IppLZ77CRC32
|
|
|
-} IppLZ77Chcksm;
|
|
|
-typedef enum {
|
|
|
- IppLZ77NoFlush,
|
|
|
- IppLZ77SyncFlush,
|
|
|
- IppLZ77FullFlush,
|
|
|
- IppLZ77FinishFlush
|
|
|
-} IppLZ77Flush;
|
|
|
-typedef struct IppLZ77Pairs_16u {
|
|
|
- Ipp16u length;
|
|
|
- Ipp16u offset;
|
|
|
-} IppLZ77Pair;
|
|
|
-typedef enum {
|
|
|
- IppLZ77StatusInit,
|
|
|
- IppLZ77StatusLZ77Process,
|
|
|
- IppLZ77StatusHuffProcess,
|
|
|
- IppLZ77StatusFinal
|
|
|
-} IppLZ77DeflateStatus;
|
|
|
-typedef enum {
|
|
|
- IppLZ77UseFixed,
|
|
|
- IppLZ77UseDynamic,
|
|
|
- IppLZ77UseStored
|
|
|
-} IppLZ77HuffMode;
|
|
|
-typedef enum {
|
|
|
- IppLZ77InflateStatusInit,
|
|
|
- IppLZ77InflateStatusHuffProcess,
|
|
|
- IppLZ77InflateStatusLZ77Process,
|
|
|
- IppLZ77InflateStatusFinal
|
|
|
-} IppLZ77InflateStatus;
|
|
|
-
|
|
|
-typedef struct IppInflateState {
|
|
|
- const Ipp8u* pWindow; /* pointer to the sliding window
|
|
|
- (the dictionary for the LZ77 algorithm) */
|
|
|
- unsigned int winSize; /* size of the sliding window */
|
|
|
- unsigned int tableType; /* type of Huffman code tables
|
|
|
- (for example, 0 - tables for Fixed Huffman codes) */
|
|
|
- unsigned int tableBufferSize; /* (ENOUGH = 2048) * (sizeof(code) = 4) -
|
|
|
- sizeof(IppInflateState) */
|
|
|
-} IppInflateState;
|
|
|
-
|
|
|
-typedef enum { /* this type is used as a translator of the inflate_mode type from zlib */
|
|
|
- ippTYPE,
|
|
|
- ippLEN,
|
|
|
- ippLENEXT
|
|
|
-} IppInflateMode;
|
|
|
-
|
|
|
-typedef struct {
|
|
|
- Ipp16u freq;
|
|
|
- Ipp16u code;
|
|
|
-} IppDeflateFreqTable;
|
|
|
-
|
|
|
-typedef struct {
|
|
|
- Ipp16u code;
|
|
|
- Ipp16u len;
|
|
|
-} IppDeflateHuffCode;
|
|
|
-
|
|
|
-typedef struct RLEState_BZ2 IppRLEState_BZ2;
|
|
|
-
|
|
|
-typedef struct EncodeHuffState_BZ2 IppEncodeHuffState_BZ2;
|
|
|
-
|
|
|
-typedef struct DecodeHuffState_BZ2 IppDecodeHuffState_BZ2;
|
|
|
-
|
|
|
-typedef enum {
|
|
|
- IppLZO1XST, /* Single-threaded, generic LZO-compatible*/
|
|
|
- IppLZO1XMT /* Multi-threaded */
|
|
|
-} IppLZOMethod ;
|
|
|
-
|
|
|
-typedef struct LZOState_8u IppLZOState_8u;
|
|
|
-
|
|
|
-#endif /* _OWN_BLDPCS */
|
|
|
-
|
|
|
-/* /////////////////////////////////////////////////////////////////////////////
|
|
|
-// The following enumerator defines a status of IPP operations
|
|
|
-// negative value means error
|
|
|
-*/
|
|
|
-typedef enum {
|
|
|
- /* errors */
|
|
|
- ippStsNotSupportedModeErr = -9999,/* The requested mode is currently not supported. */
|
|
|
- ippStsCpuNotSupportedErr = -9998,/* The target CPU is not supported. */
|
|
|
- ippStsInplaceModeNotSupportedErr = -9997,/* The inplace operation is currently not supported. */
|
|
|
-
|
|
|
- ippStsIIRIIRLengthErr = -234, /* Vector length for IIRIIR function is less than 3*(IIR order) */
|
|
|
- ippStsWarpTransformTypeErr = -233, /* The warp transform type is illegal */
|
|
|
- ippStsExceededSizeErr = -232, /* Requested size exceeded the maximum supported ROI size */
|
|
|
- ippStsWarpDirectionErr = -231, /* The warp transform direction is illegal */
|
|
|
-
|
|
|
- ippStsFilterTypeErr = -230, /* The filter type is incorrect or not supported */
|
|
|
-
|
|
|
- ippStsNormErr = -229, /* The norm is incorrect or not supported */
|
|
|
-
|
|
|
- ippStsAlgTypeErr = -228, /* Algorithm type is not supported. */
|
|
|
- ippStsMisalignedOffsetErr = -227, /* The offset is not aligned with an element. */
|
|
|
-
|
|
|
- ippStsQuadraticNonResidueErr = -226, /* SQRT operation on quadratic non-residue value. */
|
|
|
-
|
|
|
- ippStsBorderErr = -225, /* Illegal value for border type.*/
|
|
|
-
|
|
|
- ippStsDitherTypeErr = -224, /* Dithering type is not supported. */
|
|
|
- ippStsH264BufferFullErr = -223, /* Buffer for the output bitstream is full. */
|
|
|
- ippStsWrongAffinitySettingErr= -222, /* An affinity setting does not correspond to the affinity setting that was set by f.ippSetAffinity(). */
|
|
|
- ippStsLoadDynErr = -221, /* Error when loading the dynamic library. */
|
|
|
-
|
|
|
- ippStsPointAtInfinity = -220, /* Point at infinity is detected. */
|
|
|
-
|
|
|
- ippStsUnknownStatusCodeErr = -216, /* Unknown status code. */
|
|
|
-
|
|
|
- ippStsOFBSizeErr = -215, /* Incorrect value for crypto OFB block size. */
|
|
|
- ippStsLzoBrokenStreamErr = -214, /* LZO safe decompression function cannot decode LZO stream. */
|
|
|
-
|
|
|
- ippStsRoundModeNotSupportedErr = -213, /* Rounding mode is not supported. */
|
|
|
- ippStsDecimateFractionErr = -212, /* Fraction in Decimate is not supported. */
|
|
|
- ippStsWeightErr = -211, /* Incorrect value for weight. */
|
|
|
-
|
|
|
- ippStsQualityIndexErr = -210, /* Cannot calculate the quality index for an image filled with a constant. */
|
|
|
- ippStsIIRPassbandRippleErr = -209, /* Ripple in passband for Chebyshev1 design is less than zero, equal to zero, or greater than 29. */
|
|
|
- ippStsFilterFrequencyErr = -208, /* Cutoff frequency of filter is less than zero, equal to zero, or greater than 0.5. */
|
|
|
- ippStsFIRGenOrderErr = -207, /* Order of the FIR filter for design is less than 1. */
|
|
|
- ippStsIIRGenOrderErr = -206, /* Order of the IIR filter for design is less than 1, or greater than 12. */
|
|
|
-
|
|
|
- ippStsConvergeErr = -205, /* The algorithm does not converge. */
|
|
|
- ippStsSizeMatchMatrixErr = -204, /* The sizes of the source matrices are unsuitable. */
|
|
|
- ippStsCountMatrixErr = -203, /* Count value is less than, or equal to zero. */
|
|
|
- ippStsRoiShiftMatrixErr = -202, /* RoiShift value is negative or not divisible by the size of the data type. */
|
|
|
-
|
|
|
- ippStsResizeNoOperationErr = -201, /* One of the output image dimensions is less than 1 pixel. */
|
|
|
- ippStsSrcDataErr = -200, /* The source buffer contains unsupported data. */
|
|
|
- ippStsMaxLenHuffCodeErr = -199, /* Huff: Max length of Huffman code is more than the expected one. */
|
|
|
- ippStsCodeLenTableErr = -198, /* Huff: Invalid codeLenTable. */
|
|
|
- ippStsFreqTableErr = -197, /* Huff: Invalid freqTable. */
|
|
|
-
|
|
|
- ippStsIncompleteContextErr = -196, /* Crypto: set up of context is not complete. */
|
|
|
-
|
|
|
- ippStsSingularErr = -195, /* Matrix is singular. */
|
|
|
- ippStsSparseErr = -194, /* Positions of taps are not in ascending order, or are negative, or repetitive. */
|
|
|
- ippStsBitOffsetErr = -193, /* Incorrect bit offset value. */
|
|
|
- ippStsQPErr = -192, /* Incorrect quantization parameter value. */
|
|
|
- ippStsVLCErr = -191, /* Illegal VLC or FLC is detected during stream decoding. */
|
|
|
- ippStsRegExpOptionsErr = -190, /* RegExp: Options for the pattern are incorrect. */
|
|
|
- ippStsRegExpErr = -189, /* RegExp: The structure pRegExpState contains incorrect data. */
|
|
|
- ippStsRegExpMatchLimitErr = -188, /* RegExp: The match limit is exhausted. */
|
|
|
- ippStsRegExpQuantifierErr = -187, /* RegExp: Incorrect quantifier. */
|
|
|
- ippStsRegExpGroupingErr = -186, /* RegExp: Incorrect grouping. */
|
|
|
- ippStsRegExpBackRefErr = -185, /* RegExp: Incorrect back reference. */
|
|
|
- ippStsRegExpChClassErr = -184, /* RegExp: Incorrect character class. */
|
|
|
- ippStsRegExpMetaChErr = -183, /* RegExp: Incorrect metacharacter. */
|
|
|
- ippStsStrideMatrixErr = -182, /* Stride value is not positive or not divisible by the size of the data type. */
|
|
|
- ippStsCTRSizeErr = -181, /* Incorrect value for crypto CTR block size. */
|
|
|
- ippStsJPEG2KCodeBlockIsNotAttached =-180, /* Codeblock parameters are not attached to the state structure. */
|
|
|
- ippStsNotPosDefErr = -179, /* Matrix is not positive definite. */
|
|
|
-
|
|
|
- ippStsEphemeralKeyErr = -178, /* ECC: Invalid ephemeral key. */
|
|
|
- ippStsMessageErr = -177, /* ECC: Invalid message digest. */
|
|
|
- ippStsShareKeyErr = -176, /* ECC: Invalid share key. */
|
|
|
- ippStsIvalidPublicKey = -175, /* ECC: Invalid public key. */
|
|
|
- ippStsIvalidPrivateKey = -174, /* ECC: Invalid private key. */
|
|
|
- ippStsOutOfECErr = -173, /* ECC: Point out of EC. */
|
|
|
- ippStsECCInvalidFlagErr = -172, /* ECC: Invalid Flag. */
|
|
|
-
|
|
|
- ippStsMP3FrameHeaderErr = -171, /* Error in fields of the IppMP3FrameHeader structure. */
|
|
|
- ippStsMP3SideInfoErr = -170, /* Error in fields of the IppMP3SideInfo structure. */
|
|
|
-
|
|
|
- ippStsBlockStepErr = -169, /* Step for Block is less than 8. */
|
|
|
- ippStsMBStepErr = -168, /* Step for MB is less than 16. */
|
|
|
-
|
|
|
- ippStsAacPrgNumErr = -167, /* AAC: Invalid number of elements for one program. */
|
|
|
- ippStsAacSectCbErr = -166, /* AAC: Invalid section codebook. */
|
|
|
- ippStsAacSfValErr = -164, /* AAC: Invalid scalefactor value. */
|
|
|
- ippStsAacCoefValErr = -163, /* AAC: Invalid quantized coefficient value. */
|
|
|
- ippStsAacMaxSfbErr = -162, /* AAC: Invalid coefficient index. */
|
|
|
- ippStsAacPredSfbErr = -161, /* AAC: Invalid predicted coefficient index. */
|
|
|
- ippStsAacPlsDataErr = -160, /* AAC: Invalid pulse data attributes. */
|
|
|
- ippStsAacGainCtrErr = -159, /* AAC: Gain control is not supported. */
|
|
|
- ippStsAacSectErr = -158, /* AAC: Invalid number of sections. */
|
|
|
- ippStsAacTnsNumFiltErr = -157, /* AAC: Invalid number of TNS filters. */
|
|
|
- ippStsAacTnsLenErr = -156, /* AAC: Invalid length of TNS region. */
|
|
|
- ippStsAacTnsOrderErr = -155, /* AAC: Invalid order of TNS filter. */
|
|
|
- ippStsAacTnsCoefResErr = -154, /* AAC: Invalid bit-resolution for TNS filter coefficients. */
|
|
|
- ippStsAacTnsCoefErr = -153, /* AAC: Invalid coefficients of TNS filter. */
|
|
|
- ippStsAacTnsDirectErr = -152, /* AAC: Invalid direction TNS filter. */
|
|
|
- ippStsAacTnsProfileErr = -151, /* AAC: Invalid TNS profile. */
|
|
|
- ippStsAacErr = -150, /* AAC: Internal error. */
|
|
|
- ippStsAacBitOffsetErr = -149, /* AAC: Invalid current bit offset in bitstream. */
|
|
|
- ippStsAacAdtsSyncWordErr = -148, /* AAC: Invalid ADTS syncword. */
|
|
|
- ippStsAacSmplRateIdxErr = -147, /* AAC: Invalid sample rate index. */
|
|
|
- ippStsAacWinLenErr = -146, /* AAC: Invalid window length (not short or long). */
|
|
|
- ippStsAacWinGrpErr = -145, /* AAC: Invalid number of groups for current window length. */
|
|
|
- ippStsAacWinSeqErr = -144, /* AAC: Invalid window sequence range. */
|
|
|
- ippStsAacComWinErr = -143, /* AAC: Invalid common window flag. */
|
|
|
- ippStsAacStereoMaskErr = -142, /* AAC: Invalid stereo mask. */
|
|
|
- ippStsAacChanErr = -141, /* AAC: Invalid channel number. */
|
|
|
- ippStsAacMonoStereoErr = -140, /* AAC: Invalid mono-stereo flag. */
|
|
|
- ippStsAacStereoLayerErr = -139, /* AAC: Invalid this Stereo Layer flag. */
|
|
|
- ippStsAacMonoLayerErr = -138, /* AAC: Invalid this Mono Layer flag. */
|
|
|
- ippStsAacScalableErr = -137, /* AAC: Invalid scalable object flag. */
|
|
|
- ippStsAacObjTypeErr = -136, /* AAC: Invalid audio object type. */
|
|
|
- ippStsAacWinShapeErr = -135, /* AAC: Invalid window shape. */
|
|
|
- ippStsAacPcmModeErr = -134, /* AAC: Invalid PCM output interleaving indicator. */
|
|
|
- ippStsVLCUsrTblHeaderErr = -133, /* VLC: Invalid header inside table. */
|
|
|
- ippStsVLCUsrTblUnsupportedFmtErr = -132, /* VLC: Table format is not supported. */
|
|
|
- ippStsVLCUsrTblEscAlgTypeErr = -131, /* VLC: Ecs-algorithm is not supported. */
|
|
|
- ippStsVLCUsrTblEscCodeLengthErr = -130, /* VLC: Esc-code length inside table header is incorrect. */
|
|
|
- ippStsVLCUsrTblCodeLengthErr = -129, /* VLC: Code length inside table is incorrect. */
|
|
|
- ippStsVLCInternalTblErr = -128, /* VLC: Invalid internal table. */
|
|
|
- ippStsVLCInputDataErr = -127, /* VLC: Invalid input data. */
|
|
|
- ippStsVLCAACEscCodeLengthErr = -126, /* VLC: Invalid AAC-Esc code length. */
|
|
|
- ippStsNoiseRangeErr = -125, /* Noise value for Wiener Filter is out of range. */
|
|
|
- ippStsUnderRunErr = -124, /* Error in data under run. */
|
|
|
- ippStsPaddingErr = -123, /* Detected padding error indicates the possible data corruption. */
|
|
|
- ippStsCFBSizeErr = -122, /* Incorrect value for crypto CFB block size. */
|
|
|
- ippStsPaddingSchemeErr = -121, /* Invalid padding scheme. */
|
|
|
- ippStsInvalidCryptoKeyErr = -120, /* A compromised key causes suspansion of the requested cryptographic operation. */
|
|
|
- ippStsLengthErr = -119, /* Incorrect value for string length. */
|
|
|
- ippStsBadModulusErr = -118, /* Bad modulus caused a failure in module inversion. */
|
|
|
- ippStsLPCCalcErr = -117, /* Cannot evaluate linear prediction. */
|
|
|
- ippStsRCCalcErr = -116, /* Cannot compute reflection coefficients. */
|
|
|
- ippStsIncorrectLSPErr = -115, /* Incorrect values for Linear Spectral Pair. */
|
|
|
- ippStsNoRootFoundErr = -114, /* No roots are found for equation. */
|
|
|
- ippStsJPEG2KBadPassNumber = -113, /* Pass number exceeds allowed boundaries [0,nOfPasses-1]. */
|
|
|
- ippStsJPEG2KDamagedCodeBlock= -112, /* Codeblock for decoding contains damaged data. */
|
|
|
- ippStsH263CBPYCodeErr = -111, /* Illegal Huffman code is detected through CBPY stream processing. */
|
|
|
- ippStsH263MCBPCInterCodeErr = -110, /* Illegal Huffman code is detected through MCBPC Inter stream processing. */
|
|
|
- ippStsH263MCBPCIntraCodeErr = -109, /* Illegal Huffman code is detected through MCBPC Intra stream processing. */
|
|
|
- ippStsNotEvenStepErr = -108, /* Step value is not pixel multiple. */
|
|
|
- ippStsHistoNofLevelsErr = -107, /* Number of levels for histogram is less than 2. */
|
|
|
- ippStsLUTNofLevelsErr = -106, /* Number of levels for LUT is less than 2. */
|
|
|
- ippStsMP4BitOffsetErr = -105, /* Incorrect bit offset value. */
|
|
|
- ippStsMP4QPErr = -104, /* Incorrect quantization parameter. */
|
|
|
- ippStsMP4BlockIdxErr = -103, /* Incorrect block index. */
|
|
|
- ippStsMP4BlockTypeErr = -102, /* Incorrect block type. */
|
|
|
- ippStsMP4MVCodeErr = -101, /* Illegal Huffman code is detected during MV stream processing. */
|
|
|
- ippStsMP4VLCCodeErr = -100, /* Illegal Huffman code is detected during VLC stream processing. */
|
|
|
- ippStsMP4DCCodeErr = -99, /* Illegal code is detected during DC stream processing. */
|
|
|
- ippStsMP4FcodeErr = -98, /* Incorrect fcode value. */
|
|
|
- ippStsMP4AlignErr = -97, /* Incorrect buffer alignment . */
|
|
|
- ippStsMP4TempDiffErr = -96, /* Incorrect temporal difference. */
|
|
|
- ippStsMP4BlockSizeErr = -95, /* Incorrect size of a block or macroblock. */
|
|
|
- ippStsMP4ZeroBABErr = -94, /* All BAB values are equal to zero. */
|
|
|
- ippStsMP4PredDirErr = -93, /* Incorrect prediction direction. */
|
|
|
- ippStsMP4BitsPerPixelErr = -92, /* Incorrect number of bits per pixel. */
|
|
|
- ippStsMP4VideoCompModeErr = -91, /* Incorrect video component mode. */
|
|
|
- ippStsMP4LinearModeErr = -90, /* Incorrect DC linear mode. */
|
|
|
- ippStsH263PredModeErr = -83, /* Incorrect Prediction Mode value. */
|
|
|
- ippStsH263BlockStepErr = -82, /* The step value is less than 8. */
|
|
|
- ippStsH263MBStepErr = -81, /* The step value is less than 16. */
|
|
|
- ippStsH263FrameWidthErr = -80, /* The frame width is less than 8. */
|
|
|
- ippStsH263FrameHeightErr = -79, /* The frame height is less than, or equal to zero. */
|
|
|
- ippStsH263ExpandPelsErr = -78, /* Expand pixels number is less than 8. */
|
|
|
- ippStsH263PlaneStepErr = -77, /* Step value is less than the plane width. */
|
|
|
- ippStsH263QuantErr = -76, /* Quantizer value is less than, or equal to zero, or greater than 31. */
|
|
|
- ippStsH263MVCodeErr = -75, /* Illegal Huffman code is detected during MV stream processing. */
|
|
|
- ippStsH263VLCCodeErr = -74, /* Illegal Huffman code is detected during VLC stream processing. */
|
|
|
- ippStsH263DCCodeErr = -73, /* Illegal code is detected during DC stream processing. */
|
|
|
- ippStsH263ZigzagLenErr = -72, /* Zigzag compact length is more than 64. */
|
|
|
- ippStsFBankFreqErr = -71, /* Incorrect value for the filter bank frequency parameter. */
|
|
|
- ippStsFBankFlagErr = -70, /* Incorrect value for the filter bank parameter. */
|
|
|
- ippStsFBankErr = -69, /* Filter bank is not correctly initialized. */
|
|
|
- ippStsNegOccErr = -67, /* Occupation count is negative. */
|
|
|
- ippStsCdbkFlagErr = -66, /* Incorrect value for the codebook flag parameter. */
|
|
|
- ippStsSVDCnvgErr = -65, /* SVD algorithm does not converge. */
|
|
|
- ippStsJPEGHuffTableErr = -64, /* JPEG Huffman table is destroyed. */
|
|
|
- ippStsJPEGDCTRangeErr = -63, /* JPEG DCT coefficient is out of range. */
|
|
|
- ippStsJPEGOutOfBufErr = -62, /* Attempt to access out of the buffer limits. */
|
|
|
- ippStsDrawTextErr = -61, /* System error in the draw text operation. */
|
|
|
- ippStsChannelOrderErr = -60, /* Incorrect order of the destination channels. */
|
|
|
- ippStsZeroMaskValuesErr = -59, /* All values of the mask are equal to zero. */
|
|
|
- ippStsQuadErr = -58, /* The quadrangle is nonconvex or degenerates into triangle, line, or point */
|
|
|
- ippStsRectErr = -57, /* Size of the rectangle region is less than, or equal to 1. */
|
|
|
- ippStsCoeffErr = -56, /* Incorrect values for transformation coefficients. */
|
|
|
- ippStsNoiseValErr = -55, /* Incorrect value for noise amplitude for dithering. */
|
|
|
- ippStsDitherLevelsErr = -54, /* Number of dithering levels is out of range. */
|
|
|
- ippStsNumChannelsErr = -53, /* Number of channels is incorrect, or not supported. */
|
|
|
- ippStsCOIErr = -52, /* COI is out of range. */
|
|
|
- ippStsDivisorErr = -51, /* Divisor is equal to zero, function is aborted. */
|
|
|
- ippStsAlphaTypeErr = -50, /* Illegal type of image compositing operation. */
|
|
|
- ippStsGammaRangeErr = -49, /* Gamma range bounds is less than, or equal to zero. */
|
|
|
- ippStsGrayCoefSumErr = -48, /* Sum of the conversion coefficients must be less than, or equal to 1. */
|
|
|
- ippStsChannelErr = -47, /* Illegal channel number. */
|
|
|
- ippStsToneMagnErr = -46, /* Tone magnitude is less than, or equal to zero. */
|
|
|
- ippStsToneFreqErr = -45, /* Tone frequency is negative, or greater than, or equal to 0.5. */
|
|
|
- ippStsTonePhaseErr = -44, /* Tone phase is negative, or greater than, or equal to 2*PI. */
|
|
|
- ippStsTrnglMagnErr = -43, /* Triangle magnitude is less than, or equal to zero. */
|
|
|
- ippStsTrnglFreqErr = -42, /* Triangle frequency is negative, or greater than, or equal to 0.5. */
|
|
|
- ippStsTrnglPhaseErr = -41, /* Triangle phase is negative, or greater than, or equal to 2*PI. */
|
|
|
- ippStsTrnglAsymErr = -40, /* Triangle asymmetry is less than -PI, or greater than, or equal to PI. */
|
|
|
- ippStsHugeWinErr = -39, /* Kaiser window is too big. */
|
|
|
- ippStsJaehneErr = -38, /* Magnitude value is negative. */
|
|
|
- ippStsStrideErr = -37, /* Stride value is less than the length of the row. */
|
|
|
- ippStsEpsValErr = -36, /* Negative epsilon value. */
|
|
|
- ippStsWtOffsetErr = -35, /* Invalid offset value for wavelet filter. */
|
|
|
- ippStsAnchorErr = -34, /* Anchor point is outside the mask. */
|
|
|
- ippStsMaskSizeErr = -33, /* Invalid mask size. */
|
|
|
- ippStsShiftErr = -32, /* Shift value is less than zero. */
|
|
|
- ippStsSampleFactorErr = -31, /* Sampling factor is less than, or equal to zero. */
|
|
|
- ippStsSamplePhaseErr = -30, /* Phase value is out of range: 0 <= phase < factor. */
|
|
|
- ippStsFIRMRFactorErr = -29, /* MR FIR sampling factor is less than, or equal to zero. */
|
|
|
- ippStsFIRMRPhaseErr = -28, /* MR FIR sampling phase is negative, or greater than, or equal to the sampling factor. */
|
|
|
- ippStsRelFreqErr = -27, /* Relative frequency value is out of range. */
|
|
|
- ippStsFIRLenErr = -26, /* Length of a FIR filter is less than, or equal to zero. */
|
|
|
- ippStsIIROrderErr = -25, /* Order of an IIR filter is not valid. */
|
|
|
- ippStsDlyLineIndexErr = -24, /* Invalid value for the delay line sample index. */
|
|
|
- ippStsResizeFactorErr = -23, /* Resize factor(s) is less than, or equal to zero. */
|
|
|
- ippStsInterpolationErr = -22, /* Invalid interpolation mode. */
|
|
|
- ippStsMirrorFlipErr = -21, /* Invalid flip mode. */
|
|
|
- ippStsMoment00ZeroErr = -20, /* Moment value M(0,0) is too small to continue calculations. */
|
|
|
- ippStsThreshNegLevelErr = -19, /* Negative value of the level in the threshold operation. */
|
|
|
- ippStsThresholdErr = -18, /* Invalid threshold bounds. */
|
|
|
- ippStsContextMatchErr = -17, /* Context parameter does not match the operation. */
|
|
|
- ippStsFftFlagErr = -16, /* Invalid value for the FFT flag parameter. */
|
|
|
- ippStsFftOrderErr = -15, /* Invalid value for the FFT order parameter. */
|
|
|
- ippStsStepErr = -14, /* Step value is not valid. */
|
|
|
- ippStsScaleRangeErr = -13, /* Scale bounds are out of range. */
|
|
|
- ippStsDataTypeErr = -12, /* Data type is incorrect or not supported. */
|
|
|
- ippStsOutOfRangeErr = -11, /* Argument is out of range, or point is outside the image. */
|
|
|
- ippStsDivByZeroErr = -10, /* An attempt to divide by zero. */
|
|
|
- ippStsMemAllocErr = -9, /* Memory allocated for the operation is not enough.*/
|
|
|
- ippStsNullPtrErr = -8, /* Null pointer error. */
|
|
|
- ippStsRangeErr = -7, /* Incorrect values for bounds: the lower bound is greater than the upper bound. */
|
|
|
- ippStsSizeErr = -6, /* Incorrect value for data size. */
|
|
|
- ippStsBadArgErr = -5, /* Incorrect arg/param of the function. */
|
|
|
- ippStsNoMemErr = -4, /* Not enough memory for the operation. */
|
|
|
- ippStsSAReservedErr3 = -3, /* Unknown/unspecified error, -3. */
|
|
|
- ippStsErr = -2, /* Unknown/unspecified error, -2. */
|
|
|
- ippStsSAReservedErr1 = -1, /* Unknown/unspecified error, -1. */
|
|
|
-
|
|
|
- /* no errors */
|
|
|
- ippStsNoErr = 0, /* No errors. */
|
|
|
-
|
|
|
- /* warnings */
|
|
|
- ippStsNoOperation = 1, /* No operation has been executed. */
|
|
|
- ippStsMisalignedBuf = 2, /* Misaligned pointer in operation in which it must be aligned. */
|
|
|
- ippStsSqrtNegArg = 3, /* Negative value(s) for the argument in the Sqrt function. */
|
|
|
- ippStsInvZero = 4, /* INF result. Zero value was met by InvThresh with zero level. */
|
|
|
- ippStsEvenMedianMaskSize= 5, /* Even size of the Median Filter mask was replaced with the odd one. */
|
|
|
- ippStsDivByZero = 6, /* Zero value(s) for the divisor in the Div function. */
|
|
|
- ippStsLnZeroArg = 7, /* Zero value(s) for the argument in the Ln function. */
|
|
|
- ippStsLnNegArg = 8, /* Negative value(s) for the argument in the Ln function. */
|
|
|
- ippStsNanArg = 9, /* Argument value is not a number. */
|
|
|
- ippStsJPEGMarker = 10, /* JPEG marker in the bitstream. */
|
|
|
- ippStsResFloor = 11, /* All result values are floored. */
|
|
|
- ippStsOverflow = 12, /* Overflow in the operation. */
|
|
|
- ippStsLSFLow = 13, /* Quantized LP synthesis filter stability check is applied at the low boundary of [0,pi]. */
|
|
|
- ippStsLSFHigh = 14, /* Quantized LP synthesis filter stability check is applied at the high boundary of [0,pi]. */
|
|
|
- ippStsLSFLowAndHigh = 15, /* Quantized LP synthesis filter stability check is applied at both boundaries of [0,pi]. */
|
|
|
- ippStsZeroOcc = 16, /* Zero occupation count. */
|
|
|
- ippStsUnderflow = 17, /* Underflow in the operation. */
|
|
|
- ippStsSingularity = 18, /* Singularity in the operation. */
|
|
|
- ippStsDomain = 19, /* Argument is out of the function domain. */
|
|
|
- ippStsNonIntelCpu = 20, /* The target CPU is not Genuine Intel. */
|
|
|
- ippStsCpuMismatch = 21, /* Cannot set the library for the given CPU. */
|
|
|
- ippStsNoIppFunctionFound = 22, /* Application does not contain Intel IPP function calls. */
|
|
|
- ippStsDllNotFoundBestUsed = 23, /* Dispatcher cannot find the newest version of the Intel IPP dll. */
|
|
|
- ippStsNoOperationInDll = 24, /* The function does nothing in the dynamic version of the library. */
|
|
|
- ippStsInsufficientEntropy= 25, /* Generation of the prime/key failed due to insufficient entropy in the random seed and stimulus bit string. */
|
|
|
- ippStsOvermuchStrings = 26, /* Number of destination strings is more than expected. */
|
|
|
- ippStsOverlongString = 27, /* Length of one of the destination strings is more than expected. */
|
|
|
- ippStsAffineQuadChanged = 28, /* 4th vertex of destination quad is not equal to customer's one. */
|
|
|
- ippStsWrongIntersectROI = 29, /* ROI has no intersection with the source or destination ROI. No operation. */
|
|
|
- ippStsWrongIntersectQuad = 30, /* Quadrangle has no intersection with the source or destination ROI. No operation. */
|
|
|
- ippStsSmallerCodebook = 31, /* Size of created codebook is less than the cdbkSize argument. */
|
|
|
- ippStsSrcSizeLessExpected = 32, /* DC: Size of the source buffer is less than the expected one. */
|
|
|
- ippStsDstSizeLessExpected = 33, /* DC: Size of the destination buffer is less than the expected one. */
|
|
|
- ippStsStreamEnd = 34, /* DC: The end of stream processed. */
|
|
|
- ippStsDoubleSize = 35, /* Width or height of image is odd. */
|
|
|
- ippStsNotSupportedCpu = 36, /* The CPU is not supported. */
|
|
|
- ippStsUnknownCacheSize = 37, /* The CPU is supported, but the size of the cache is unknown. */
|
|
|
- ippStsSymKernelExpected = 38, /* The Kernel is not symmetric. */
|
|
|
- ippStsEvenMedianWeight = 39, /* Even weight of the Weighted Median Filter is replaced with the odd one. */
|
|
|
- ippStsWrongIntersectVOI = 40, /* VOI has no intersection with the source or destination volume. No operation. */
|
|
|
- ippStsI18nMsgCatalogInvalid=41, /* Message Catalog is invalid, English message returned. */
|
|
|
- ippStsI18nGetMessageFail = 42, /* Failed to fetch a localized message, English message returned. For more information use errno on Linux* OS and GetLastError on Windows* OS. */
|
|
|
- ippStsWaterfall = 43, /* Cannot load required library, waterfall is used. */
|
|
|
- ippStsPrevLibraryUsed = 44, /* Cannot load required library, previous dynamic library is used. */
|
|
|
- ippStsLLADisabled = 45, /* OpenMP* Low Level Affinity is disabled. */
|
|
|
- ippStsNoAntialiasing = 46, /* The mode does not support antialiasing. */
|
|
|
- ippStsRepetitiveSrcData = 47, /* DC: The source data is too repetitive. */
|
|
|
- ippStsSizeWrn = 48, /* The size does not allow to perform full operation. */
|
|
|
- ippStsFeatureNotSupported = 49, /* Current CPU doesn't support at least 1 of the desired features. */
|
|
|
- ippStsUnknownFeature = 50, /* At least one of the desired features is unknown. */
|
|
|
- ippStsFeaturesCombination = 51, /* Wrong combination of features. */
|
|
|
- ippStsAccurateModeNotSupported = 52 /* Accurate mode is not supported. */
|
|
|
-} IppStatus;
|
|
|
-
|
|
|
-#define ippStsOk ippStsNoErr
|
|
|
-
|
|
|
-#endif
|
|
|
-
|
|
|
-#ifdef __cplusplus
|
|
|
-}
|
|
|
-#endif
|
|
|
-
|
|
|
-#endif /* __IPPTYPES_H__ */
|