Skip to content

Commit c76792f

Browse files
asmorkalovasmorkalov
authored
Merge pull request #3803 from cudawarped:cuda_update_to_npp_stream_ctx
cuda - update npp calls to use the new NppStreamContext API if available
2 parents 5741f22 + 06f764a commit c76792f

File tree

11 files changed

+571
-82
lines changed

11 files changed

+571
-82
lines changed

‎modules/cudaarithm/src/core.cpp‎

Lines changed: 43 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -79,7 +79,11 @@ namespace
7979
{
8080
typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
8181

82-
typedef NppStatus (*func_t)(const npp_t* pSrc, int nSrcStep, npp_t* pDst, int nDstStep, NppiSize oROI, NppiAxis flip);
82+
#if USE_NPP_STREAM_CTX
83+
typedef NppStatus (*func_t)(const npp_t* pSrc, int nSrcStep, npp_t* pDst, int nDstStep, NppiSize oROI, NppiAxis flip, NppStreamContext ctx);
84+
#else
85+
typedef NppStatus(*func_t)(const npp_t* pSrc, int nSrcStep, npp_t* pDst, int nDstStep, NppiSize oROI, NppiAxis flip);
86+
#endif
8387
};
8488

8589
template <int DEPTH, typename NppMirrorFunc<DEPTH>::func_t func> struct NppMirror
@@ -94,9 +98,15 @@ namespace
9498
sz.width = src.cols;
9599
sz.height = src.rows;
96100

101+
#if USE_NPP_STREAM_CTX
97102
nppSafeCall( func(src.ptr<npp_t>(), static_cast<int>(src.step),
98103
dst.ptr<npp_t>(), static_cast<int>(dst.step), sz,
99-
(flipCode == 0 ? NPP_HORIZONTAL_AXIS : (flipCode > 0 ? NPP_VERTICAL_AXIS : NPP_BOTH_AXIS))) );
104+
(flipCode == 0 ? NPP_HORIZONTAL_AXIS : (flipCode > 0 ? NPP_VERTICAL_AXIS : NPP_BOTH_AXIS)), h) );
105+
#else
106+
nppSafeCall( func(src.ptr<npp_t>(), static_cast<int>(src.step),
107+
dst.ptr<npp_t>(), static_cast<int>(dst.step), sz,
108+
(flipCode == 0 ? NPP_HORIZONTAL_AXIS : (flipCode > 0 ? NPP_VERTICAL_AXIS : NPP_BOTH_AXIS))) );
109+
#endif
100110

101111
if (stream == 0)
102112
cudaSafeCall( cudaDeviceSynchronize() );
@@ -107,7 +117,11 @@ namespace
107117
{
108118
typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
109119

110-
typedef NppStatus (*func_t)(npp_t* pSrcDst, int nSrcDstStep, NppiSize oROI, NppiAxis flip);
120+
#if USE_NPP_STREAM_CTX
121+
typedef NppStatus (*func_t)(npp_t* pSrcDst, int nSrcDstStep, NppiSize oROI, NppiAxis flip, NppStreamContext ctx);
122+
#else
123+
typedef NppStatus(*func_t)(npp_t* pSrcDst, int nSrcDstStep, NppiSize oROI, NppiAxis flip);
124+
#endif
111125
};
112126

113127
template <int DEPTH, typename NppMirrorIFunc<DEPTH>::func_t func> struct NppMirrorI
@@ -121,10 +135,15 @@ namespace
121135
NppiSize sz;
122136
sz.width = srcDst.cols;
123137
sz.height = srcDst.rows;
124-
138+
#if USE_NPP_STREAM_CTX
139+
nppSafeCall(func(srcDst.ptr<npp_t>(), static_cast<int>(srcDst.step),
140+
sz,
141+
(flipCode == 0 ? NPP_HORIZONTAL_AXIS : (flipCode > 0 ? NPP_VERTICAL_AXIS : NPP_BOTH_AXIS)), h) );
142+
#else
125143
nppSafeCall( func(srcDst.ptr<npp_t>(), static_cast<int>(srcDst.step),
126144
sz,
127145
(flipCode == 0 ? NPP_HORIZONTAL_AXIS : (flipCode > 0 ? NPP_VERTICAL_AXIS : NPP_BOTH_AXIS))) );
146+
#endif
128147

129148
if (stream == 0)
130149
cudaSafeCall( cudaDeviceSynchronize() );
@@ -137,23 +156,41 @@ void cv::cuda::flip(InputArray _src, OutputArray _dst, int flipCode, Stream& str
137156
typedef void (*func_t)(const GpuMat& src, GpuMat& dst, int flipCode, cudaStream_t stream);
138157
static const func_t funcs[6][4] =
139158
{
140-
{NppMirror<CV_8U, nppiMirror_8u_C1R>::call, 0, NppMirror<CV_8U, nppiMirror_8u_C3R>::call, NppMirror<CV_8U, nppiMirror_8u_C4R>::call},
159+
#if USE_NPP_STREAM_CTX
160+
{NppMirror<CV_8U, nppiMirror_8u_C1R_Ctx>::call, 0, NppMirror<CV_8U, nppiMirror_8u_C3R_Ctx>::call, NppMirror<CV_8U, nppiMirror_8u_C4R_Ctx>::call},
161+
{0,0,0,0},
162+
{NppMirror<CV_16U, nppiMirror_16u_C1R_Ctx>::call, 0, NppMirror<CV_16U, nppiMirror_16u_C3R_Ctx>::call, NppMirror<CV_16U, nppiMirror_16u_C4R_Ctx>::call},
163+
{0,0,0,0},
164+
{NppMirror<CV_32S, nppiMirror_32s_C1R_Ctx>::call, 0, NppMirror<CV_32S, nppiMirror_32s_C3R_Ctx>::call, NppMirror<CV_32S, nppiMirror_32s_C4R_Ctx>::call},
165+
{NppMirror<CV_32F, nppiMirror_32f_C1R_Ctx>::call, 0, NppMirror<CV_32F, nppiMirror_32f_C3R_Ctx>::call, NppMirror<CV_32F, nppiMirror_32f_C4R_Ctx>::call}
166+
#else
167+
{ NppMirror<CV_8U, nppiMirror_8u_C1R>::call, 0, NppMirror<CV_8U, nppiMirror_8u_C3R>::call, NppMirror<CV_8U, nppiMirror_8u_C4R>::call },
141168
{0,0,0,0},
142169
{NppMirror<CV_16U, nppiMirror_16u_C1R>::call, 0, NppMirror<CV_16U, nppiMirror_16u_C3R>::call, NppMirror<CV_16U, nppiMirror_16u_C4R>::call},
143170
{0,0,0,0},
144171
{NppMirror<CV_32S, nppiMirror_32s_C1R>::call, 0, NppMirror<CV_32S, nppiMirror_32s_C3R>::call, NppMirror<CV_32S, nppiMirror_32s_C4R>::call},
145172
{NppMirror<CV_32F, nppiMirror_32f_C1R>::call, 0, NppMirror<CV_32F, nppiMirror_32f_C3R>::call, NppMirror<CV_32F, nppiMirror_32f_C4R>::call}
173+
#endif
146174
};
147175

148176
typedef void (*ifunc_t)(GpuMat& srcDst, int flipCode, cudaStream_t stream);
149177
static const ifunc_t ifuncs[6][4] =
150178
{
151-
{NppMirrorI<CV_8U, nppiMirror_8u_C1IR>::call, 0, NppMirrorI<CV_8U, nppiMirror_8u_C3IR>::call, NppMirrorI<CV_8U, nppiMirror_8u_C4IR>::call},
179+
#if USE_NPP_STREAM_CTX
180+
{NppMirrorI<CV_8U, nppiMirror_8u_C1IR_Ctx>::call, 0, NppMirrorI<CV_8U, nppiMirror_8u_C3IR_Ctx>::call, NppMirrorI<CV_8U, nppiMirror_8u_C4IR_Ctx>::call},
181+
{0,0,0,0},
182+
{NppMirrorI<CV_16U, nppiMirror_16u_C1IR_Ctx>::call, 0, NppMirrorI<CV_16U, nppiMirror_16u_C3IR_Ctx>::call, NppMirrorI<CV_16U, nppiMirror_16u_C4IR_Ctx>::call},
183+
{0,0,0,0},
184+
{NppMirrorI<CV_32S, nppiMirror_32s_C1IR_Ctx>::call, 0, NppMirrorI<CV_32S, nppiMirror_32s_C3IR_Ctx>::call, NppMirrorI<CV_32S, nppiMirror_32s_C4IR_Ctx>::call},
185+
{NppMirrorI<CV_32F, nppiMirror_32f_C1IR_Ctx>::call, 0, NppMirrorI<CV_32F, nppiMirror_32f_C3IR_Ctx>::call, NppMirrorI<CV_32F, nppiMirror_32f_C4IR_Ctx>::call}
186+
#else
187+
{ NppMirrorI<CV_8U, nppiMirror_8u_C1IR>::call, 0, NppMirrorI<CV_8U, nppiMirror_8u_C3IR>::call, NppMirrorI<CV_8U, nppiMirror_8u_C4IR>::call },
152188
{0,0,0,0},
153189
{NppMirrorI<CV_16U, nppiMirror_16u_C1IR>::call, 0, NppMirrorI<CV_16U, nppiMirror_16u_C3IR>::call, NppMirrorI<CV_16U, nppiMirror_16u_C4IR>::call},
154190
{0,0,0,0},
155191
{NppMirrorI<CV_32S, nppiMirror_32s_C1IR>::call, 0, NppMirrorI<CV_32S, nppiMirror_32s_C3IR>::call, NppMirrorI<CV_32S, nppiMirror_32s_C4IR>::call},
156192
{NppMirrorI<CV_32F, nppiMirror_32f_C1IR>::call, 0, NppMirrorI<CV_32F, nppiMirror_32f_C3IR>::call, NppMirrorI<CV_32F, nppiMirror_32f_C4IR>::call}
193+
#endif
157194
};
158195

159196
GpuMat src = getInputMat(_src, stream);

‎modules/cudaarithm/src/cuda/bitwise_scalar.cu‎

Lines changed: 41 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -92,7 +92,11 @@ namespace
9292
{
9393
typedef typename NPPTypeTraits<DEPTH>::npp_type npp_type;
9494

95+
#if USE_NPP_STREAM_CTX
96+
typedef NppStatus(*func_t)(const npp_type* pSrc1, int nSrc1Step, const npp_type* pConstants, npp_type* pDst, int nDstStep, NppiSize oSizeROI, NppStreamContext ctx);
97+
#else
9598
typedef NppStatus (*func_t)(const npp_type* pSrc1, int nSrc1Step, const npp_type* pConstants, npp_type* pDst, int nDstStep, NppiSize oSizeROI);
99+
#endif
96100
};
97101

98102
template <int DEPTH, int cn, typename NppBitwiseCFunc<DEPTH, cn>::func_t func> struct NppBitwiseC
@@ -116,7 +120,11 @@ namespace
116120
cv::saturate_cast<npp_type>(value[3])
117121
};
118122

123+
#if USE_NPP_STREAM_CTX
124+
nppSafeCall(func(src.ptr<npp_type>(), static_cast<int>(src.step), pConstants, dst.ptr<npp_type>(), static_cast<int>(dst.step), oSizeROI, h));
125+
#else
119126
nppSafeCall( func(src.ptr<npp_type>(), static_cast<int>(src.step), pConstants, dst.ptr<npp_type>(), static_cast<int>(dst.step), oSizeROI) );
127+
#endif
120128

121129
if (stream == 0)
122130
CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
@@ -131,13 +139,39 @@ void bitScalar(const GpuMat& src, cv::Scalar value, bool, GpuMat& dst, const Gpu
131139
typedef void (*func_t)(const GpuMat& src, cv::Scalar value, GpuMat& dst, Stream& stream);
132140
static const func_t funcs[3][6][4] =
133141
{
142+
#if USE_NPP_STREAM_CTX
143+
{
144+
{BitScalar<uchar, bitScalarOp<bit_and, uchar> >::call , 0, NppBitwiseC<CV_8U , 3, nppiAndC_8u_C3R_Ctx >::call, BitScalar4< bitScalarOp<bit_and, uint> >::call},
145+
{BitScalar<uchar, bitScalarOp<bit_and, uchar> >::call , 0, NppBitwiseC<CV_8U , 3, nppiAndC_8u_C3R_Ctx >::call, BitScalar4< bitScalarOp<bit_and, uint> >::call},
146+
{BitScalar<ushort, bitScalarOp<bit_and, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiAndC_16u_C3R_Ctx>::call, NppBitwiseC<CV_16U, 4, nppiAndC_16u_C4R_Ctx>::call},
147+
{BitScalar<ushort, bitScalarOp<bit_and, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiAndC_16u_C3R_Ctx>::call, NppBitwiseC<CV_16U, 4, nppiAndC_16u_C4R_Ctx>::call},
148+
{BitScalar<uint, bitScalarOp<bit_and, uint> >::call , 0, NppBitwiseC<CV_32S, 3, nppiAndC_32s_C3R_Ctx>::call, NppBitwiseC<CV_32S, 4, nppiAndC_32s_C4R_Ctx>::call},
149+
{BitScalar<uint, bitScalarOp<bit_and, uint> >::call , 0, NppBitwiseC<CV_32S, 3, nppiAndC_32s_C3R_Ctx>::call, NppBitwiseC<CV_32S, 4, nppiAndC_32s_C4R_Ctx>::call}
150+
},
134151
{
135-
{BitScalar<uchar, bitScalarOp<bit_and, uchar> >::call , 0, NppBitwiseC<CV_8U , 3, nppiAndC_8u_C3R >::call, BitScalar4< bitScalarOp<bit_and, uint> >::call},
136-
{BitScalar<uchar, bitScalarOp<bit_and, uchar> >::call , 0, NppBitwiseC<CV_8U , 3, nppiAndC_8u_C3R >::call, BitScalar4< bitScalarOp<bit_and, uint> >::call},
137-
{BitScalar<ushort, bitScalarOp<bit_and, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiAndC_16u_C3R>::call, NppBitwiseC<CV_16U, 4, nppiAndC_16u_C4R>::call},
138-
{BitScalar<ushort, bitScalarOp<bit_and, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiAndC_16u_C3R>::call, NppBitwiseC<CV_16U, 4, nppiAndC_16u_C4R>::call},
139-
{BitScalar<uint, bitScalarOp<bit_and, uint> >::call , 0, NppBitwiseC<CV_32S, 3, nppiAndC_32s_C3R>::call, NppBitwiseC<CV_32S, 4, nppiAndC_32s_C4R>::call},
140-
{BitScalar<uint, bitScalarOp<bit_and, uint> >::call , 0, NppBitwiseC<CV_32S, 3, nppiAndC_32s_C3R>::call, NppBitwiseC<CV_32S, 4, nppiAndC_32s_C4R>::call}
152+
{BitScalar<uchar, bitScalarOp<bit_or, uchar> >::call , 0, NppBitwiseC<CV_8U , 3, nppiOrC_8u_C3R_Ctx >::call, BitScalar4< bitScalarOp<bit_or, uint> >::call},
153+
{BitScalar<uchar, bitScalarOp<bit_or, uchar> >::call , 0, NppBitwiseC<CV_8U , 3, nppiOrC_8u_C3R_Ctx >::call, BitScalar4< bitScalarOp<bit_or, uint> >::call},
154+
{BitScalar<ushort, bitScalarOp<bit_or, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiOrC_16u_C3R_Ctx>::call, NppBitwiseC<CV_16U, 4, nppiOrC_16u_C4R_Ctx>::call},
155+
{BitScalar<ushort, bitScalarOp<bit_or, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiOrC_16u_C3R_Ctx>::call, NppBitwiseC<CV_16U, 4, nppiOrC_16u_C4R_Ctx>::call},
156+
{BitScalar<uint, bitScalarOp<bit_or, uint> >::call , 0, NppBitwiseC<CV_32S, 3, nppiOrC_32s_C3R_Ctx>::call, NppBitwiseC<CV_32S, 4, nppiOrC_32s_C4R_Ctx>::call},
157+
{BitScalar<uint, bitScalarOp<bit_or, uint> >::call , 0, NppBitwiseC<CV_32S, 3, nppiOrC_32s_C3R_Ctx>::call, NppBitwiseC<CV_32S, 4, nppiOrC_32s_C4R_Ctx>::call}
158+
},
159+
{
160+
{BitScalar<uchar, bitScalarOp<bit_xor, uchar> >::call , 0, NppBitwiseC<CV_8U , 3, nppiXorC_8u_C3R_Ctx >::call, BitScalar4< bitScalarOp<bit_xor, uint> >::call},
161+
{BitScalar<uchar, bitScalarOp<bit_xor, uchar> >::call , 0, NppBitwiseC<CV_8U , 3, nppiXorC_8u_C3R_Ctx >::call, BitScalar4< bitScalarOp<bit_xor, uint> >::call},
162+
{BitScalar<ushort, bitScalarOp<bit_xor, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiXorC_16u_C3R_Ctx>::call, NppBitwiseC<CV_16U, 4, nppiXorC_16u_C4R_Ctx>::call},
163+
{BitScalar<ushort, bitScalarOp<bit_xor, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiXorC_16u_C3R_Ctx>::call, NppBitwiseC<CV_16U, 4, nppiXorC_16u_C4R_Ctx>::call},
164+
{BitScalar<uint, bitScalarOp<bit_xor, uint> >::call , 0, NppBitwiseC<CV_32S, 3, nppiXorC_32s_C3R_Ctx>::call, NppBitwiseC<CV_32S, 4, nppiXorC_32s_C4R_Ctx>::call},
165+
{BitScalar<uint, bitScalarOp<bit_xor, uint> >::call , 0, NppBitwiseC<CV_32S, 3, nppiXorC_32s_C3R_Ctx>::call, NppBitwiseC<CV_32S, 4, nppiXorC_32s_C4R_Ctx>::call}
166+
}
167+
#else
168+
{
169+
{ BitScalar<uchar, bitScalarOp<bit_and, uchar> >::call, 0, NppBitwiseC<CV_8U, 3, nppiAndC_8u_C3R >::call, BitScalar4< bitScalarOp<bit_and, uint> >::call },
170+
{ BitScalar<uchar, bitScalarOp<bit_and, uchar> >::call , 0, NppBitwiseC<CV_8U , 3, nppiAndC_8u_C3R >::call, BitScalar4< bitScalarOp<bit_and, uint> >::call },
171+
{ BitScalar<ushort, bitScalarOp<bit_and, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiAndC_16u_C3R>::call, NppBitwiseC<CV_16U, 4, nppiAndC_16u_C4R>::call },
172+
{ BitScalar<ushort, bitScalarOp<bit_and, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiAndC_16u_C3R>::call, NppBitwiseC<CV_16U, 4, nppiAndC_16u_C4R>::call },
173+
{ BitScalar<uint, bitScalarOp<bit_and, uint> >::call , 0, NppBitwiseC<CV_32S, 3, nppiAndC_32s_C3R>::call, NppBitwiseC<CV_32S, 4, nppiAndC_32s_C4R>::call },
174+
{ BitScalar<uint, bitScalarOp<bit_and, uint> >::call , 0, NppBitwiseC<CV_32S, 3, nppiAndC_32s_C3R>::call, NppBitwiseC<CV_32S, 4, nppiAndC_32s_C4R>::call }
141175
},
142176
{
143177
{BitScalar<uchar, bitScalarOp<bit_or, uchar> >::call , 0, NppBitwiseC<CV_8U , 3, nppiOrC_8u_C3R >::call, BitScalar4< bitScalarOp<bit_or, uint> >::call},
@@ -155,6 +189,7 @@ void bitScalar(const GpuMat& src, cv::Scalar value, bool, GpuMat& dst, const Gpu
155189
{BitScalar<uint, bitScalarOp<bit_xor, uint> >::call , 0, NppBitwiseC<CV_32S, 3, nppiXorC_32s_C3R>::call, NppBitwiseC<CV_32S, 4, nppiXorC_32s_C4R>::call},
156190
{BitScalar<uint, bitScalarOp<bit_xor, uint> >::call , 0, NppBitwiseC<CV_32S, 3, nppiXorC_32s_C3R>::call, NppBitwiseC<CV_32S, 4, nppiXorC_32s_C4R>::call}
157191
}
192+
#endif
158193
};
159194

160195
const int depth = src.depth();

‎modules/cudaarithm/src/cuda/threshold.cu‎

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -116,8 +116,13 @@ double cv::cuda::threshold(InputArray _src, OutputArray _dst, double thresh, dou
116116
sz.width = src.cols;
117117
sz.height = src.rows;
118118

119+
#if USE_NPP_STREAM_CTX
120+
nppSafeCall(nppiThreshold_32f_C1R_Ctx(src.ptr<Npp32f>(), static_cast<int>(src.step),
121+
dst.ptr<Npp32f>(), static_cast<int>(dst.step), sz, static_cast<Npp32f>(thresh), NPP_CMP_GREATER, h));
122+
#else
119123
nppSafeCall( nppiThreshold_32f_C1R(src.ptr<Npp32f>(), static_cast<int>(src.step),
120124
dst.ptr<Npp32f>(), static_cast<int>(dst.step), sz, static_cast<Npp32f>(thresh), NPP_CMP_GREATER) );
125+
#endif
121126

122127
if (!stream)
123128
CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );

‎modules/cudaarithm/src/cuda/transpose.cu‎

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -74,8 +74,13 @@ void cv::cuda::transpose(InputArray _src, OutputArray _dst, Stream& stream)
7474
sz.width = src.cols;
7575
sz.height = src.rows;
7676

77+
#if USE_NPP_STREAM_CTX
78+
nppSafeCall(nppiTranspose_8u_C1R_Ctx(src.ptr<Npp8u>(), static_cast<int>(src.step),
79+
dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz, h));
80+
#else
7781
nppSafeCall( nppiTranspose_8u_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step),
7882
dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz) );
83+
#endif
7984

8085
if (!stream)
8186
CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );

0 commit comments

Comments
 (0)