changed kernel.set to support unsuccssful set

This commit is contained in:
mlyashko 2014-03-07 11:15:54 +04:00
parent 66ed6a8a50
commit eb2e8a11fa
4 changed files with 6 additions and 130 deletions

View File

@ -599,17 +599,6 @@ protected:
CV_EXPORTS MatAllocator* getOpenCLAllocator();
#define SAFE_KERNEL_SET_ARG(idx, arg) \
{\
int idxNew = kernel.set(idx, arg); \
if (-1 == idxNew)\
{\
printf("algorithm can't setup argument index = %d to kernel\n", idx); \
return false; \
}\
idx = idxNew; \
}
}}
#endif

View File

@ -2796,7 +2796,8 @@ int Kernel::set(int i, const void* value, size_t sz)
{
if (!p || !p->handle)
return -1;
CV_Assert(i >= 0);
if (i < 0)
return i;
if( i == 0 )
p->cleanupUMats();
@ -2822,7 +2823,8 @@ int Kernel::set(int i, const KernelArg& arg)
{
if( !p || !p->handle )
return -1;
CV_Assert( i >= 0 );
if (i < 0)
return i;
if( i == 0 )
p->cleanupUMats();
if( arg.m )

View File

@ -726,7 +726,6 @@ namespace cv
ocl::Image2D imageI(I);
ocl::Image2D imageJ(J);
int idxArg = 0;
#if 0
idxArg = kernel.set(idxArg, imageI); //image2d_t I
idxArg = kernel.set(idxArg, imageJ); //image2d_t J
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(prevPts)); // __global const float2* prevPts
@ -744,26 +743,6 @@ namespace cv
idxArg = kernel.set(idxArg, (int)winSize.height); // int c_winSize_y
idxArg = kernel.set(idxArg, (int)iters); // int c_iters
idxArg = kernel.set(idxArg, (char)calcErr); //char calcErr
#else
SAFE_KERNEL_SET_ARG(idxArg, imageI); //image2d_t I
SAFE_KERNEL_SET_ARG(idxArg, imageJ); //image2d_t J
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadOnly(prevPts)); // __global const float2* prevPts
SAFE_KERNEL_SET_ARG(idxArg, (int)prevPts.step); // int prevPtsStep
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadWrite(nextPts)); // __global const float2* nextPts
SAFE_KERNEL_SET_ARG(idxArg, (int)nextPts.step); // int nextPtsStep
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadWrite(status)); // __global uchar* status
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadWrite(err)); // __global float* err
SAFE_KERNEL_SET_ARG(idxArg, (int)level); // const int level
SAFE_KERNEL_SET_ARG(idxArg, (int)I.rows); // const int rows
SAFE_KERNEL_SET_ARG(idxArg, (int)I.cols); // const int cols
SAFE_KERNEL_SET_ARG(idxArg, (int)patch.x); // int PATCH_X
SAFE_KERNEL_SET_ARG(idxArg, (int)patch.y); // int PATCH_Y
SAFE_KERNEL_SET_ARG(idxArg, (int)winSize.width); // int c_winSize_x
SAFE_KERNEL_SET_ARG(idxArg, (int)winSize.height); // int c_winSize_y
SAFE_KERNEL_SET_ARG(idxArg, (int)iters); // int c_iters
SAFE_KERNEL_SET_ARG(idxArg, (char)calcErr); //char calcErr
#endif
return kernel.run(2, globalThreads, localThreads, true);
}
private:

View File

@ -204,7 +204,6 @@ bool cv_ocl_tvl1flow::centeredGradient(const UMat &src, UMat &dx, UMat &dy)
return false;
int idxArg = 0;
#if 0
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(src));//src mat
idxArg = kernel.set(idxArg, (int)(src.cols));//src mat col
idxArg = kernel.set(idxArg, (int)(src.rows));//src mat rows
@ -212,15 +211,6 @@ bool cv_ocl_tvl1flow::centeredGradient(const UMat &src, UMat &dx, UMat &dy)
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrWriteOnly(dx));//res mat dx
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrWriteOnly(dy));//res mat dy
idxArg = kernel.set(idxArg, (int)(dx.step/dx.elemSize()));//res mat step
#else
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadOnly(src));
SAFE_KERNEL_SET_ARG(idxArg, (int)(src.cols));
SAFE_KERNEL_SET_ARG(idxArg, (int)(src.rows));
SAFE_KERNEL_SET_ARG(idxArg, (int)(src.step / src.elemSize()));
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrWriteOnly(dx));
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrWriteOnly(dy));
SAFE_KERNEL_SET_ARG(idxArg, (int)(dx.step / dx.elemSize()));
#endif
return kernel.run(2, globalsize, NULL, false);
}
@ -235,7 +225,6 @@ bool cv_ocl_tvl1flow::warpBackward(const UMat &I0, const UMat &I1, UMat &I1x, UM
return false;
int idxArg = 0;
#if 0
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(I0));//I0 mat
int I0_step = (int)(I0.step / I0.elemSize());
idxArg = kernel.set(idxArg, I0_step);//I0_step
@ -265,37 +254,6 @@ bool cv_ocl_tvl1flow::warpBackward(const UMat &I0, const UMat &I1, UMat &I1x, UM
u2_offset_x = (int) (u2_offset_x / u2.elemSize());
idxArg = kernel.set(idxArg, (int)u2_offset_x);//u2_offset_x
idxArg = kernel.set(idxArg, (int)(u2.offset / u2.step));//u2_offset_y
#else
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadOnly(I0));//I0 mat
int I0_step = (int)(I0.step / I0.elemSize());
SAFE_KERNEL_SET_ARG(idxArg, I0_step);//I0_step
SAFE_KERNEL_SET_ARG(idxArg, (int)(I0.cols));//I0_col
SAFE_KERNEL_SET_ARG(idxArg, (int)(I0.rows));//I0_row
ocl::Image2D imageI1(I1);
ocl::Image2D imageI1x(I1x);
ocl::Image2D imageI1y(I1y);
SAFE_KERNEL_SET_ARG(idxArg, imageI1);//image2d_t tex_I1
SAFE_KERNEL_SET_ARG(idxArg, imageI1x);//image2d_t tex_I1x
SAFE_KERNEL_SET_ARG(idxArg, imageI1y);//image2d_t tex_I1y
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadOnly(u1));//const float* u1
SAFE_KERNEL_SET_ARG(idxArg, (int)(u1.step / u1.elemSize()));//int u1_step
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadOnly(u2));//const float* u2
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrWriteOnly(I1w));///float* I1w
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrWriteOnly(I1wx));//float* I1wx
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrWriteOnly(I1wy));//float* I1wy
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrWriteOnly(grad));//float* grad
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrWriteOnly(rho));//float* rho
SAFE_KERNEL_SET_ARG(idxArg, (int)(I1w.step / I1w.elemSize()));//I1w_step
SAFE_KERNEL_SET_ARG(idxArg, (int)(u2.step / u2.elemSize()));//u2_step
int u1_offset_x = (int)((u1.offset) % (u1.step));
u1_offset_x = (int)(u1_offset_x / u1.elemSize());
SAFE_KERNEL_SET_ARG(idxArg, (int)u1_offset_x);//u1_offset_x
SAFE_KERNEL_SET_ARG(idxArg, (int)(u1.offset / u1.step));//u1_offset_y
int u2_offset_x = (int)((u2.offset) % (u2.step));
u2_offset_x = (int)(u2_offset_x / u2.elemSize());
SAFE_KERNEL_SET_ARG(idxArg, (int)u2_offset_x);//u2_offset_x
SAFE_KERNEL_SET_ARG(idxArg, (int)(u2.offset / u2.step));//u2_offset_y
#endif
return kernel.run(2, globalsize, NULL, false);
}
@ -311,7 +269,6 @@ bool cv_ocl_tvl1flow::estimateU(UMat &I1wx, UMat &I1wy, UMat &grad,
return false;
int idxArg = 0;
#if 0
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(I1wx)); //const float* I1wx
idxArg = kernel.set(idxArg, (int)(I1wx.cols)); //int I1wx_col
idxArg = kernel.set(idxArg, (int)(I1wx.rows)); //int I1wx_row
@ -339,35 +296,7 @@ bool cv_ocl_tvl1flow::estimateU(UMat &I1wx, UMat &I1wy, UMat &grad,
idxArg = kernel.set(idxArg, (int)u2_offset_x ); //int u2_offset_x
idxArg = kernel.set(idxArg, (int)(u2.offset / u2.step)); //int u2_offset_y
idxArg = kernel.set(idxArg, (char)calc_error); //char calc_error
#else
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadOnly(I1wx)); //const float* I1wx
SAFE_KERNEL_SET_ARG(idxArg, (int)(I1wx.cols)); //int I1wx_col
SAFE_KERNEL_SET_ARG(idxArg, (int)(I1wx.rows)); //int I1wx_row
SAFE_KERNEL_SET_ARG(idxArg, (int)(I1wx.step / I1wx.elemSize())); //int I1wx_step
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadOnly(I1wy)); //const float* I1wy
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadOnly(grad)); //const float* grad
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadOnly(rho_c)); //const float* rho_c
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadOnly(p11)); //const float* p11
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadOnly(p12)); //const float* p12
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadOnly(p21)); //const float* p21
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadOnly(p22)); //const float* p22
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadWrite(u1)); //float* u1
SAFE_KERNEL_SET_ARG(idxArg, (int)(u1.step / u1.elemSize())); //int u1_step
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadWrite(u2)); //float* u2
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrWriteOnly(error)); //float* error
SAFE_KERNEL_SET_ARG(idxArg, (float)l_t); //float l_t
SAFE_KERNEL_SET_ARG(idxArg, (float)theta); //float theta
SAFE_KERNEL_SET_ARG(idxArg, (int)(u2.step / u2.elemSize()));//int u2_step
int u1_offset_x = (int)(u1.offset % u1.step);
u1_offset_x = (int)(u1_offset_x / u1.elemSize());
SAFE_KERNEL_SET_ARG(idxArg, (int)u1_offset_x); //int u1_offset_x
SAFE_KERNEL_SET_ARG(idxArg, (int)(u1.offset / u1.step)); //int u1_offset_y
int u2_offset_x = (int)(u2.offset % u2.step);
u2_offset_x = (int)(u2_offset_x / u2.elemSize());
SAFE_KERNEL_SET_ARG(idxArg, (int)u2_offset_x); //int u2_offset_x
SAFE_KERNEL_SET_ARG(idxArg, (int)(u2.offset / u2.step)); //int u2_offset_y
SAFE_KERNEL_SET_ARG(idxArg, (char)calc_error); //char calc_error
#endif
return kernel.run(2, globalsize, NULL, false);
}
@ -381,7 +310,6 @@ bool cv_ocl_tvl1flow::estimateDualVariables(UMat &u1, UMat &u2,
return false;
int idxArg = 0;
#if 0
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(u1));// const float* u1
idxArg = kernel.set(idxArg, (int)(u1.cols)); //int u1_col
idxArg = kernel.set(idxArg, (int)(u1.rows)); //int u1_row
@ -402,33 +330,11 @@ bool cv_ocl_tvl1flow::estimateDualVariables(UMat &u1, UMat &u2,
u2_offset_x = (int)(u2_offset_x / u2.elemSize());
idxArg = kernel.set(idxArg, u2_offset_x); //int u2_offset_x
idxArg = kernel.set(idxArg, (int)(u2.offset / u2.step)); //int u2_offset_y
#else
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadOnly(u1));// const float* u1
SAFE_KERNEL_SET_ARG(idxArg, (int)(u1.cols)); //int u1_col
SAFE_KERNEL_SET_ARG(idxArg, (int)(u1.rows)); //int u1_row
SAFE_KERNEL_SET_ARG(idxArg, (int)(u1.step / u1.elemSize())); //int u1_step
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadOnly(u2)); // const float* u2
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadWrite(p11)); // float* p11
SAFE_KERNEL_SET_ARG(idxArg, (int)(p11.step / p11.elemSize())); //int p11_step
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadWrite(p12)); // float* p12
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadWrite(p21)); // float* p21
SAFE_KERNEL_SET_ARG(idxArg, ocl::KernelArg::PtrReadWrite(p22)); // float* p22
SAFE_KERNEL_SET_ARG(idxArg, (float)(taut)); //float taut
SAFE_KERNEL_SET_ARG(idxArg, (int)(u2.step / u2.elemSize())); //int u2_step
int u1_offset_x = (int)(u1.offset % u1.step);
u1_offset_x = (int)(u1_offset_x / u1.elemSize());
SAFE_KERNEL_SET_ARG(idxArg, u1_offset_x); //int u1_offset_x
SAFE_KERNEL_SET_ARG(idxArg, (int)(u1.offset / u1.step)); //int u1_offset_y
int u2_offset_x = (int)(u2.offset % u2.step);
u2_offset_x = (int)(u2_offset_x / u2.elemSize());
SAFE_KERNEL_SET_ARG(idxArg, u2_offset_x); //int u2_offset_x
idxArg = kernel.set(idxArg, (int)(u2.offset / u2.step)); //int u2_offset_y
#endif
return kernel.run(2, globalsize, NULL, false);
}
OpticalFlowDual_TVL1::OpticalFlowDual_TVL1()
{
tau = 0.25;