Parameter 7 incorrect for cblas_dgemv

[Lavrijsen__Matthijs]

I have the following code:

	CBLAS_LAYOUT layout = CblasRowMajor;
	CBLAS_TRANSPOSE trans = CblasNoTrans;

	int m = 2;
	int n = 4;
	double a[8] = { 1,2,3,4,5,6,7,8 }; // m * n matrix
	double x[4] = { 1,2,3,4 }; // vector of size n
	double y[2] = { 0,0 }; // vector of size m
	
	double *a_ptr = &a[0];
	double *x_ptr = &x[0];
	double *y_ptr = &y[0];

	cblas_dgemv(layout,	// matrix has row-major layout
		trans,			// should not be transposed
		m,				// has m number of rows
		n,				// n number of columns
		1,				// alpha = 1
		a_ptr,			// pointer to matrix A
		n,				// dimensionality of leading dimension of A (#columns, in this case)
		x_ptr,			// pointer to vector x
		1,				// stride of x
		0,				// beta = 0
		y_ptr,			// pointer to vector y (output will also be placed in y)
		1);				// stride of y

This compiles fine. When running, I get 

Intel MKL ERROR: Parameter 7 was incorrect on entry to cblas_dgemv

Parameter 7 is the lda, which I set to be n, the nr of columns since I use row-major layout. According to the documentation and this post: https://software.intel.com/en-us/forums/intel-math-kernel-library/topic/501513 this should be correct.

So, why is it giving me this error?

Thanks in advance!

[mecej4]

Please provide details of how you compiled and linked the program -- OS, target 32 or 64 bit, compiler options, how MKL was linked into, etc.

[Dennis_F_2]

Hi Mecej,

Answering for Matthijs S, the platform is Windows 7 using MSVC 2015 (19.00.23918) in x64 release mode. Statically linked.

[TimP]

One of the questions implied by Sham is whether you might have linked the ilp64 MKL libraries (which require int64_t), seeing that your code uses 32-bit integer arguments.  The question would be answered by showing your build log.

[Lavrijsen__Matthijs]

Thanks. We're indeed linking to ilp64. Here are relevant CMake lines:

find_library(mkl_intel_ilp64 "mkl_intel_ilp64.lib" PATHS "C:/Program Files (x86)/IntelSWTools/compilers_and_libraries/windows/mkl/lib/intel64")
find_library(mkl_core "mkl_core.lib" PATHS "C:/Program Files (x86)/IntelSWTools/compilers_and_libraries/windows/mkl/lib/intel64")
find_library(mkl_sequential "mkl_sequential.lib" PATHS "C:/Program Files (x86)/IntelSWTools/compilers_and_libraries/windows/mkl/lib/intel64")

target_link_libraries(our_lib ${mkl_intel_ilp64})
target_link_libraries(our_lib ${mkl_core})
target_link_libraries(our_lib ${mkl_sequential})

Based on your comment, I changed line 4 and 5 in the code from the first post to int64_t. Still the same error.

EDIT: we're currently building in Debug mode. Not sure whether that's relevant.

[TimP]

Sorry, I meant to point out that for plain int arguments you would link the lp64 libraries, which are the ones you get with the -Qmkl options.

[Lavrijsen__Matthijs]

Sorry, not sure I follow:

1. "you would link the lp64 libraries." Do you agree that we're linking to these libraries with the above CMake lines, or is that not sufficient?
2. "For plain int arguments". I used plain int arguments in my original code, with the same CMake setup -- i.e. linking to ilp64, unless point 1 is incorrect. This didn't work. Then I switched to int64_t, which also did not work.

So -- if the CMake setup should indeed be sufficient for linking to the ilp64 libraries -- I'm not sure what to do differently.

[Gennady_F_Intel · Accepted solution]

if you want to link with ILP64 interfaces, then please don't forget to add /DMKL_ILP64 compiler option. Pls refer to the User's Guide for more details.

[Lavrijsen__Matthijs]

Thanks Gennady, that works!