bool TestTranspose<T>::process()
{
NCVStatus ncvStat;
bool rcode = false;
NcvSize32u srcSize(this->width, this->height);
NCVMatrixAlloc<T> d_img(*this->allocatorGPU.get(), this->width, this->height);
ncvAssertReturn(d_img.isMemAllocated(), false);
NCVMatrixAlloc<T> h_img(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_img.isMemAllocated(), false);
NCVMatrixAlloc<T> d_dst(*this->allocatorGPU.get(), this->height, this->width);
ncvAssertReturn(d_dst.isMemAllocated(), false);
NCVMatrixAlloc<T> h_dst(*this->allocatorCPU.get(), this->height, this->width);
ncvAssertReturn(h_dst.isMemAllocated(), false);
NCVMatrixAlloc<T> h_dst_d(*this->allocatorCPU.get(), this->height, this->width);
ncvAssertReturn(h_dst_d.isMemAllocated(), false);
NCV_SET_SKIP_COND(this->allocatorGPU.get()->isCounting());
NCV_SKIP_COND_BEGIN
ncvAssertReturn(this->src.fill(h_img), false);
NCV_SKIP_COND_END
ncvStat = h_img.copySolid(d_img, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_BEGIN
if (sizeof(T) == sizeof(Ncv32u))
{
ncvStat = nppiStTranspose_32u_C1R((Ncv32u *)d_img.ptr(), d_img.pitch(),
(Ncv32u *)d_dst.ptr(), d_dst.pitch(),
NcvSize32u(this->width, this->height));
}
else if (sizeof(T) == sizeof(Ncv64u))
{
ncvStat = nppiStTranspose_64u_C1R((Ncv64u *)d_img.ptr(), d_img.pitch(),
(Ncv64u *)d_dst.ptr(), d_dst.pitch(),
NcvSize32u(this->width, this->height));
}
else
{
ncvAssertPrintReturn(false, "Incorrect transpose test instance", false);
}
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_END
ncvStat = d_dst.copySolid(h_dst_d, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_BEGIN
if (sizeof(T) == sizeof(Ncv32u))
{
ncvStat = nppiStTranspose_32u_C1R_host((Ncv32u *)h_img.ptr(), h_img.pitch(),
(Ncv32u *)h_dst.ptr(), h_dst.pitch(),
NcvSize32u(this->width, this->height));
}
else if (sizeof(T) == sizeof(Ncv64u))
{
ncvStat = nppiStTranspose_64u_C1R_host((Ncv64u *)h_img.ptr(), h_img.pitch(),
(Ncv64u *)h_dst.ptr(), h_dst.pitch(),
NcvSize32u(this->width, this->height));
}
else
{
ncvAssertPrintReturn(false, "Incorrect downsample test instance", false);
}
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_END
//bit-to-bit check
bool bLoopVirgin = true;
NCV_SKIP_COND_BEGIN
//const Ncv64f relEPS = 0.005;
for (Ncv32u i=0; bLoopVirgin && i < this->width; i++)
{
for (Ncv32u j=0; bLoopVirgin && j < this->height; j++)
{
if (h_dst.ptr()[h_dst.stride()*i+j] != h_dst_d.ptr()[h_dst_d.stride()*i+j])
{
bLoopVirgin = false;
}
}
}
NCV_SKIP_COND_END
if (bLoopVirgin)
{
rcode = true;
}
return rcode;
}
bool TestIntegralImage<T_in, T_out>::process()
{
NCVStatus ncvStat;
bool rcode = false;
Ncv32u widthII = this->width + 1;
Ncv32u heightII = this->height + 1;
NCVMatrixAlloc<T_in> d_img(*this->allocatorGPU.get(), this->width, this->height);
ncvAssertReturn(d_img.isMemAllocated(), false);
NCVMatrixAlloc<T_in> h_img(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_img.isMemAllocated(), false);
NCVMatrixAlloc<T_out> d_imgII(*this->allocatorGPU.get(), widthII, heightII);
ncvAssertReturn(d_imgII.isMemAllocated(), false);
NCVMatrixAlloc<T_out> h_imgII(*this->allocatorCPU.get(), widthII, heightII);
ncvAssertReturn(h_imgII.isMemAllocated(), false);
NCVMatrixAlloc<T_out> h_imgII_d(*this->allocatorCPU.get(), widthII, heightII);
ncvAssertReturn(h_imgII_d.isMemAllocated(), false);
Ncv32u bufSize;
if (sizeof(T_in) == sizeof(Ncv8u))
{
ncvStat = nppiStIntegralGetSize_8u32u(NcvSize32u(this->width, this->height), &bufSize, this->devProp);
ncvAssertReturn(NPPST_SUCCESS == ncvStat, false);
}
else if (sizeof(T_in) == sizeof(Ncv32f))
{
ncvStat = nppiStIntegralGetSize_32f32f(NcvSize32u(this->width, this->height), &bufSize, this->devProp);
ncvAssertReturn(NPPST_SUCCESS == ncvStat, false);
}
else
{
ncvAssertPrintReturn(false, "Incorrect integral image test instance", false);
}
NCVVectorAlloc<Ncv8u> d_tmpBuf(*this->allocatorGPU.get(), bufSize);
ncvAssertReturn(d_tmpBuf.isMemAllocated(), false);
NCV_SET_SKIP_COND(this->allocatorGPU.get()->isCounting());
NCV_SKIP_COND_BEGIN
ncvAssertReturn(this->src.fill(h_img), false);
ncvStat = h_img.copySolid(d_img, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
if (sizeof(T_in) == sizeof(Ncv8u))
{
ncvStat = nppiStIntegral_8u32u_C1R((Ncv8u *)d_img.ptr(), d_img.pitch(),
(Ncv32u *)d_imgII.ptr(), d_imgII.pitch(),
NcvSize32u(this->width, this->height),
d_tmpBuf.ptr(), bufSize, this->devProp);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
}
else if (sizeof(T_in) == sizeof(Ncv32f))
{
ncvStat = nppiStIntegral_32f32f_C1R((Ncv32f *)d_img.ptr(), d_img.pitch(),
(Ncv32f *)d_imgII.ptr(), d_imgII.pitch(),
NcvSize32u(this->width, this->height),
d_tmpBuf.ptr(), bufSize, this->devProp);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
}
else
{
ncvAssertPrintReturn(false, "Incorrect integral image test instance", false);
}
ncvStat = d_imgII.copySolid(h_imgII_d, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
if (sizeof(T_in) == sizeof(Ncv8u))
{
ncvStat = nppiStIntegral_8u32u_C1R_host((Ncv8u *)h_img.ptr(), h_img.pitch(),
(Ncv32u *)h_imgII.ptr(), h_imgII.pitch(),
NcvSize32u(this->width, this->height));
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
}
else if (sizeof(T_in) == sizeof(Ncv32f))
{
ncvStat = nppiStIntegral_32f32f_C1R_host((Ncv32f *)h_img.ptr(), h_img.pitch(),
(Ncv32f *)h_imgII.ptr(), h_imgII.pitch(),
NcvSize32u(this->width, this->height));
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
}
else
{
ncvAssertPrintReturn(false, "Incorrect integral image test instance", false);
}
NCV_SKIP_COND_END
//bit-to-bit check
bool bLoopVirgin = true;
NCV_SKIP_COND_BEGIN
for (Ncv32u i=0; bLoopVirgin && i < h_img.height() + 1; i++)
{
for (Ncv32u j=0; bLoopVirgin && j < h_img.width() + 1; j++)
{
if (sizeof(T_in) == sizeof(Ncv8u))
{
if (h_imgII.ptr()[h_imgII.stride()*i+j] != h_imgII_d.ptr()[h_imgII_d.stride()*i+j])
{
bLoopVirgin = false;
}
}
else if (sizeof(T_in) == sizeof(Ncv32f))
{
if (fabsf((float)h_imgII.ptr()[h_imgII.stride()*i+j] - (float)h_imgII_d.ptr()[h_imgII_d.stride()*i+j]) > 0.01f)
{
bLoopVirgin = false;
}
}
else
{
ncvAssertPrintReturn(false, "Incorrect integral image test instance", false);
}
}
}
NCV_SKIP_COND_END
if (bLoopVirgin)
{
rcode = true;
}
return rcode;
}
bool TestResize<T>::process()
{
NCVStatus ncvStat;
bool rcode = false;
Ncv32s smallWidth = this->width / this->scaleFactor;
Ncv32s smallHeight = this->height / this->scaleFactor;
if (smallWidth == 0 || smallHeight == 0)
{
return true;
}
NcvSize32u srcSize(this->width, this->height);
NCVMatrixAlloc<T> d_img(*this->allocatorGPU.get(), this->width, this->height);
ncvAssertReturn(d_img.isMemAllocated(), false);
NCVMatrixAlloc<T> h_img(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_img.isMemAllocated(), false);
NCVMatrixAlloc<T> d_small(*this->allocatorGPU.get(), smallWidth, smallHeight);
ncvAssertReturn(d_small.isMemAllocated(), false);
NCVMatrixAlloc<T> h_small(*this->allocatorCPU.get(), smallWidth, smallHeight);
ncvAssertReturn(h_small.isMemAllocated(), false);
NCVMatrixAlloc<T> h_small_d(*this->allocatorCPU.get(), smallWidth, smallHeight);
ncvAssertReturn(h_small_d.isMemAllocated(), false);
NCV_SET_SKIP_COND(this->allocatorGPU.get()->isCounting());
NCV_SKIP_COND_BEGIN
ncvAssertReturn(this->src.fill(h_img), false);
NCV_SKIP_COND_END
ncvStat = h_img.copySolid(d_img, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_BEGIN
if (sizeof(T) == sizeof(Ncv32u))
{
ncvStat = nppiStDecimate_32u_C1R((Ncv32u *)d_img.ptr(), d_img.pitch(),
(Ncv32u *)d_small.ptr(), d_small.pitch(),
srcSize, this->scaleFactor,
this->bTextureCache);
}
else if (sizeof(T) == sizeof(Ncv64u))
{
ncvStat = nppiStDecimate_64u_C1R((Ncv64u *)d_img.ptr(), d_img.pitch(),
(Ncv64u *)d_small.ptr(), d_small.pitch(),
srcSize, this->scaleFactor,
this->bTextureCache);
}
else
{
ncvAssertPrintReturn(false, "Incorrect downsample test instance", false);
}
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_END
ncvStat = d_small.copySolid(h_small_d, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_BEGIN
if (sizeof(T) == sizeof(Ncv32u))
{
ncvStat = nppiStDecimate_32u_C1R_host((Ncv32u *)h_img.ptr(), h_img.pitch(),
(Ncv32u *)h_small.ptr(), h_small.pitch(),
srcSize, this->scaleFactor);
}
else if (sizeof(T) == sizeof(Ncv64u))
{
ncvStat = nppiStDecimate_64u_C1R_host((Ncv64u *)h_img.ptr(), h_img.pitch(),
(Ncv64u *)h_small.ptr(), h_small.pitch(),
srcSize, this->scaleFactor);
}
else
{
ncvAssertPrintReturn(false, "Incorrect downsample test instance", false);
}
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_END
//bit-to-bit check
bool bLoopVirgin = true;
NCV_SKIP_COND_BEGIN
//const Ncv64f relEPS = 0.005;
for (Ncv32u i=0; bLoopVirgin && i < h_small.height(); i++)
{
for (Ncv32u j=0; bLoopVirgin && j < h_small.width(); j++)
{
if (h_small.ptr()[h_small.stride()*i+j] != h_small_d.ptr()[h_small_d.stride()*i+j])
{
bLoopVirgin = false;
}
}
}
NCV_SKIP_COND_END
if (bLoopVirgin)
{
rcode = true;
}
return rcode;
}
bool TestDrawRects<T>::process()
{
NCVStatus ncvStat;
bool rcode = false;
NCVMatrixAlloc<T> d_img(*this->allocatorGPU.get(), this->width, this->height);
ncvAssertReturn(d_img.isMemAllocated(), false);
NCVMatrixAlloc<T> h_img(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_img.isMemAllocated(), false);
NCVMatrixAlloc<T> h_img_d(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_img_d.isMemAllocated(), false);
NCVVectorAlloc<NcvRect32u> d_rects(*this->allocatorGPU.get(), this->numRects);
ncvAssertReturn(d_rects.isMemAllocated(), false);
NCVVectorAlloc<NcvRect32u> h_rects(*this->allocatorCPU.get(), this->numRects);
ncvAssertReturn(h_rects.isMemAllocated(), false);
NCV_SET_SKIP_COND(this->allocatorGPU.get()->isCounting());
NCV_SKIP_COND_BEGIN
ncvAssertReturn(this->src.fill(h_img), false);
ncvStat = h_img.copySolid(d_img, 0);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvAssertCUDAReturn(cudaStreamSynchronize(0), false);
//fill vector of rectangles with random rects covering the input
NCVVectorReuse<Ncv32u> h_rects_as32u(h_rects.getSegment());
ncvAssertReturn(h_rects_as32u.isMemReused(), false);
ncvAssertReturn(this->src32u.fill(h_rects_as32u), false);
for (Ncv32u i=0; i<this->numRects; i++)
{
h_rects.ptr()[i].x = (Ncv32u)(((1.0 * h_rects.ptr()[i].x) / RAND_MAX) * (this->width-2));
h_rects.ptr()[i].y = (Ncv32u)(((1.0 * h_rects.ptr()[i].y) / RAND_MAX) * (this->height-2));
h_rects.ptr()[i].width = (Ncv32u)(((1.0 * h_rects.ptr()[i].width) / RAND_MAX) * (this->width+10 - h_rects.ptr()[i].x));
h_rects.ptr()[i].height = (Ncv32u)(((1.0 * h_rects.ptr()[i].height) / RAND_MAX) * (this->height+10 - h_rects.ptr()[i].y));
}
ncvStat = h_rects.copySolid(d_rects, 0);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvAssertCUDAReturn(cudaStreamSynchronize(0), false);
if (sizeof(T) == sizeof(Ncv32u))
{
ncvStat = ncvDrawRects_32u_device((Ncv32u *)d_img.ptr(), d_img.stride(), this->width, this->height,
(NcvRect32u *)d_rects.ptr(), this->numRects, this->color, 0);
}
else if (sizeof(T) == sizeof(Ncv8u))
{
ncvStat = ncvDrawRects_8u_device((Ncv8u *)d_img.ptr(), d_img.stride(), this->width, this->height,
(NcvRect32u *)d_rects.ptr(), this->numRects, (Ncv8u)this->color, 0);
}
else
{
ncvAssertPrintReturn(false, "Incorrect drawrects test instance", false);
}
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
NCV_SKIP_COND_END
ncvStat = d_img.copySolid(h_img_d, 0);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvAssertCUDAReturn(cudaStreamSynchronize(0), false);
NCV_SKIP_COND_BEGIN
if (sizeof(T) == sizeof(Ncv32u))
{
ncvStat = ncvDrawRects_32u_host((Ncv32u *)h_img.ptr(), h_img.stride(), this->width, this->height,
(NcvRect32u *)h_rects.ptr(), this->numRects, this->color);
}
else if (sizeof(T) == sizeof(Ncv8u))
{
ncvStat = ncvDrawRects_8u_host((Ncv8u *)h_img.ptr(), h_img.stride(), this->width, this->height,
(NcvRect32u *)h_rects.ptr(), this->numRects, (Ncv8u)this->color);
}
else
{
ncvAssertPrintReturn(false, "Incorrect drawrects test instance", false);
}
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
NCV_SKIP_COND_END
//bit-to-bit check
bool bLoopVirgin = true;
NCV_SKIP_COND_BEGIN
//const Ncv64f relEPS = 0.005;
for (Ncv32u i=0; bLoopVirgin && i < h_img.height(); i++)
{
for (Ncv32u j=0; bLoopVirgin && j < h_img.width(); j++)
{
if (h_img.ptr()[h_img.stride()*i+j] != h_img_d.ptr()[h_img_d.stride()*i+j])
{
bLoopVirgin = false;
}
}
}
NCV_SKIP_COND_END
if (bLoopVirgin)
{
rcode = true;
}
return rcode;
}
