bool TestRectStdDev::process()
{
NCVStatus ncvStat;
bool rcode = false;
Ncv32s _normWidth = (Ncv32s)this->width - this->rect.x - this->rect.width + 1;
Ncv32s _normHeight = (Ncv32s)this->height - this->rect.y - this->rect.height + 1;
if (_normWidth <= 0 || _normHeight <= 0)
{
return true;
}
Ncv32u normWidth = (Ncv32u)_normWidth;
Ncv32u normHeight = (Ncv32u)_normHeight;
NcvSize32u szNormRoi(normWidth, normHeight);
Ncv32u widthII = this->width + 1;
Ncv32u heightII = this->height + 1;
Ncv32u widthSII = this->width + 1;
Ncv32u heightSII = this->height + 1;
NCVMatrixAlloc<Ncv8u> d_img(*this->allocatorGPU.get(), this->width, this->height);
ncvAssertReturn(d_img.isMemAllocated(), false);
NCVMatrixAlloc<Ncv8u> h_img(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_img.isMemAllocated(), false);
NCVMatrixAlloc<Ncv32u> d_imgII(*this->allocatorGPU.get(), widthII, heightII);
ncvAssertReturn(d_imgII.isMemAllocated(), false);
NCVMatrixAlloc<Ncv32u> h_imgII(*this->allocatorCPU.get(), widthII, heightII);
ncvAssertReturn(h_imgII.isMemAllocated(), false);
NCVMatrixAlloc<Ncv64u> d_imgSII(*this->allocatorGPU.get(), widthSII, heightSII);
ncvAssertReturn(d_imgSII.isMemAllocated(), false);
NCVMatrixAlloc<Ncv64u> h_imgSII(*this->allocatorCPU.get(), widthSII, heightSII);
ncvAssertReturn(h_imgSII.isMemAllocated(), false);
NCVMatrixAlloc<Ncv32f> d_norm(*this->allocatorGPU.get(), normWidth, normHeight);
ncvAssertReturn(d_norm.isMemAllocated(), false);
NCVMatrixAlloc<Ncv32f> h_norm(*this->allocatorCPU.get(), normWidth, normHeight);
ncvAssertReturn(h_norm.isMemAllocated(), false);
NCVMatrixAlloc<Ncv32f> h_norm_d(*this->allocatorCPU.get(), normWidth, normHeight);
ncvAssertReturn(h_norm_d.isMemAllocated(), false);
Ncv32u bufSizeII, bufSizeSII;
ncvStat = nppiStIntegralGetSize_8u32u(NcvSize32u(this->width, this->height), &bufSizeII, this->devProp);
ncvAssertReturn(NPPST_SUCCESS == ncvStat, false);
ncvStat = nppiStSqrIntegralGetSize_8u64u(NcvSize32u(this->width, this->height), &bufSizeSII, this->devProp);
ncvAssertReturn(NPPST_SUCCESS == ncvStat, false);
Ncv32u bufSize = bufSizeII > bufSizeSII ? bufSizeII : bufSizeSII;
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);
ncvStat = nppiStIntegral_8u32u_C1R(d_img.ptr(), d_img.pitch(),
d_imgII.ptr(), d_imgII.pitch(),
NcvSize32u(this->width, this->height),
d_tmpBuf.ptr(), bufSize, this->devProp);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = nppiStSqrIntegral_8u64u_C1R(d_img.ptr(), d_img.pitch(),
d_imgSII.ptr(), d_imgSII.pitch(),
NcvSize32u(this->width, this->height),
d_tmpBuf.ptr(), bufSize, this->devProp);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = nppiStRectStdDev_32f_C1R(d_imgII.ptr(), d_imgII.pitch(),
d_imgSII.ptr(), d_imgSII.pitch(),
d_norm.ptr(), d_norm.pitch(),
szNormRoi, this->rect,
this->scaleFactor,
this->bTextureCache);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = d_norm.copySolid(h_norm_d, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = nppiStIntegral_8u32u_C1R_host(h_img.ptr(), h_img.pitch(),
h_imgII.ptr(), h_imgII.pitch(),
NcvSize32u(this->width, this->height));
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = nppiStSqrIntegral_8u64u_C1R_host(h_img.ptr(), h_img.pitch(),
h_imgSII.ptr(), h_imgSII.pitch(),
NcvSize32u(this->width, this->height));
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = nppiStRectStdDev_32f_C1R_host(h_imgII.ptr(), h_imgII.pitch(),
h_imgSII.ptr(), h_imgSII.pitch(),
h_norm.ptr(), h_norm.pitch(),
szNormRoi, this->rect,
this->scaleFactor);
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_norm.height(); i++)
{
for (Ncv32u j=0; bLoopVirgin && j < h_norm.width(); j++)
{
Ncv64f absErr = fabs(h_norm.ptr()[h_norm.stride()*i+j] - h_norm_d.ptr()[h_norm_d.stride()*i+j]);
Ncv64f relErr = absErr / h_norm.ptr()[h_norm.stride()*i+j];
if (relErr > relEPS)
{
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;
}