topic ipprWarpAffine bug in Intel® Integrated Performance Primitives

ipprWarpAffine bug

Ken_Thornton — Fri, 07 Sep 2018 22:48:21 GMT

I'm running tests on ipprWarpAffine and get failures for certain permutation matrices.

My IPP build is 58644 (April 6, 2018). I'm running on Windows 7 (Intel Core i7) and using the ia32_win IPP libraries.

Below is a simple example to illustrate the problem.

void TestWarpAffine()
{
	double coefs[3][4] ={{0.0, 0.0, 1.0, 0.0}, {1.0, 0.0, 0.0, 0.0}, {0.0, 1.0, 0.0, 0.0}};
	unsigned char in[27] ={0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26};
	unsigned char out[27] ={0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};

	IpprVolume srcSize ={3,3,3}, dstSize ={3,3,3};
	IpprCuboid srcVOI ={0,0,0,3,3,3}, dstVOI ={0,0,0,3,3,3};

	//Step 1: Allocate buffer
	int nBufferSize = 0;
	IppStatus sts = ipprWarpAffineGetBufSize(srcSize, srcVOI, dstVOI, coefs, 1, ippLinear, &nBufferSize); //1 means 1 channel
	ASSERT(sts == ippStsOk && nBufferSize >= 0);
	nBufferSize = __max(nBufferSize, 10); //Protect against allocating 0 bytes
	Ipp8u* pBuffer = ippsMalloc_8u(nBufferSize);
	ASSERT(pBuffer != NULL);

	//Step 2: Warp
	sts = ipprWarpAffine_8u_C1V(in, srcSize, 3, 9, srcVOI, out, 3, 9, dstVOI, coefs, ippLinear, pBuffer); //3 bytes between rows, 9 bytes between planes
	ippsFree(pBuffer);
	ASSERT(sts == ippStsOk); //Will get sts == -56

	const IppLibraryVersion* pVer = ippGetLibVersion();
}

is that 64 or 32 bit code or

Gennady_F_Intel — Sat, 08 Sep 2018 02:41:47 GMT

is that 64 or 32 bit code or both?

-56 error code means that

Gennady_F_Intel — Sat, 08 Sep 2018 03:17:01 GMT

-56 error code means that ippStsCoeffErr determinant of the transform matrix Aij is equal to zero. if you will fix this - your case will work.

I know what the error code

Ken_Thornton — Mon, 10 Sep 2018 13:25:57 GMT

Gennady,

No the determinant is 1, not 0. So it is invertible.

https://software.intel.com/en-us/ipp-dev-reference-warpaffine

If p is a 3D point in the source volume and q is a 3D point in the destination volume:

| 0 0 1 | | 0 |

q = | 1 0 0 | p + | 0 |

| 0 1 0 | | 0 |

The inverse transformation is:

| 0 1 0 |

p = | 0 0 1 | q

| 1 0 0 |

-1

| 0 1 0 | | 0 0 1 |

Since | 0 0 1 | = | 1 0 0 |

| 1 0 0 | | 0 1 0 |

Ken, we confirmed the problem

Gennady_F_Intel — Wed, 12 Sep 2018 08:56:18 GMT

Ken, we confirmed the problem and will add the fix into one of the nearest updates. We will keep you updated with the status of this issue. Thanks for report.

the fix of the problem

Gennady_F_Intel — Thu, 07 May 2020 09:31:32 GMT

the fix of the problem available since IPP 2019 u1 and the latest 2020 versions as well.