Error was detected while a MEX-file was running and MATLAB is exiting because of fatal error

I am trying to run the batched version of QR (dgeqrfbatched) in matlab using CUBLAS by calling it from a mex file. I am struck with this error which i am not able to find the answer , Is there any work around for this problem? I am attaching the code that i am running and also the crash report

#include "mex.h"
#include "cublas_v2.h"

// The MEX gateway function.
    void mexFunction(int nlhs, mxArray *plhs[], int nrhs,const mxArray *prhs[]) 
    {
    
    // Get input variables from Matlab (host variables).
    
    double **A;
    // Get dimensions of input variables from Matlab.
    size_t m, n, k;
    const mwSize *Adims;
    Adims = mxGetDimensions(prhs[0]);
    //Bdims = mxGetDimensions(prhs[1]);
    m = Adims[0];
    n = Adims[1];
    k = Adims[2];
    A = (double**)mxGetPr(prhs[0]);
    int lda = m;
    const int batchSize=k;

//step -1 Allocate storage for batch count 

    double **tau;
    tau = (double**)malloc(batchSize * sizeof(double*));        
    for (int i = 0; i < batchSize; i++) 
    {
    		tau[i] = (double*)malloc(n * sizeof(double));
    }

      int *info;
	info = (int*)malloc(batchSize * sizeof(int));


//step -2 create host pointer array to the gpu array


    double **d_A, **d_TAU, **h_d_A, **h_d_TAU;
	h_d_A = (double**)malloc(batchSize * sizeof(double*));
	h_d_TAU = (double**)malloc(batchSize * sizeof(double*));
	for (int i = 0; i < batchSize; i++) {
		cudaMalloc((double**)&h_d_A[i], m*n * sizeof(double));
		cudaMalloc((double**)&h_d_TAU[i], n * sizeof(double));
	}
	
	
//step -3 copy host array of pointers to device
	cudaMalloc((double**)&d_A, batchSize * sizeof(double*));
	cudaMalloc((double**)&d_TAU, batchSize * sizeof(double));
	
	cudaMemcpy(d_A, h_d_A, batchSize * sizeof(double*), cudaMemcpyHostToDevice);
	cudaMemcpy(d_TAU, h_d_TAU, batchSize * sizeof(double*), cudaMemcpyHostToDevice);
	
	
	
	for (int i = 0; i < batchSize; i++)
	{
		cudaMemcpy(h_d_A[i], A[i], m *n * sizeof(double), cudaMemcpyHostToDevice);
		cudaMemcpy(h_d_TAU[i], tau[i], n * sizeof(double), cudaMemcpyHostToDevice);
	}
	
// --- CUBLAS initialization
	cublasHandle_t cublas_handle;
	cublasCreate(&cublas_handle);
	
	cublasDgeqrfBatched(cublas_handle, m, n, d_A, lda, d_TAU, info, batchSize);
	
	for (int i = 0; i < batchSize; i++)
		cudaMemcpy(A[i], h_d_A[i], m*n * sizeof(double), cudaMemcpyDeviceToHost);
	
//print the A matrix
 
	for (int k = 0; k < batchSize; k++) {
		for (int j = 0; j < m; j++) {
			for (int i = 0; i < n; i++) {
				int index = j * m + i;//not tested
				//count = count + 1;
				printf("\n %d The values are %lf",k+index, A[k][index]);

			} // i
		} // j
	} // k

}

When i execute the above program this is the crash report i am getting.

--------------------------------------------------------------------------------

Segmentation violation detected at Thu Mar 14 08:52:21 2019 -0700

--------------------------------------------------------------------------------

Configuration:

Crash Decoding : Disabled - No sandbox or build area path

Crash Mode : continue (default)

Default Encoding : UTF-8

Deployed : false

GNU C Library : 2.24 stable

Graphics Driver : Unknown software

Java Version : Java 1.8.0_144-b01 with Oracle Corporation Java HotSpot(TM) 64-Bit Server VM mixed mode

MATLAB Architecture : glnxa64

MATLAB Entitlement ID : 1378095

MATLAB Root : /cvmfs/restricted.computecanada.ca/easybuild/software/2017/Core/matlab/2018a

MATLAB Version : 9.4.0.813654 (R2018a)

OpenGL : software

Operating System : "CentOS Linux release 7.6.1810 (Core) "

Process ID : 171494

Processor ID : x86 Family 6 Model 79 Stepping 1, GenuineIntel

Session Key : 32be7088-53bb-46ac-a878-a0e4028bfd50

Static TLS mitigation : Disabled: Unnecessary 1

Window System : No active display

Fault Count: 1

Abnormal termination

Register State (from fault):

RAX = 0000000000000000 RBX = 0000000000000000

RCX = 00002b105df7b080 RDX = 0000000000000000

RSP = 00002b1073ffcc00 RBP = 00002b1073ffcc10

RSI = 00002b1073ffcf10 RDI = 0000000000000000

R8 = 00002b1042d133e8 R9 = 0000000000000030

R10 = 000000000000042b R11 = 00002b1047aff750

R12 = 0000000000000000 R13 = 00002b1073ffcf10

R14 = 0000000000000000 R15 = 00002b105df7b080

RIP = 00002b1047aa710c EFL = 0000000000010206

CS = 0033 FS = 0000 GS = 0000

Stack Trace (from fault):

[ 0] 0x00002b1047aa710c bin/glnxa64/libmx.so+00499980 _ZN6matrix6detail10noninlined12mx_array_api15mxGetDimensionsEPK11mxArray_tag+00000012

[ 1] 0x00002b10e29b6c00 /home/naveen/Matlab/Mexcuda/example.mexa64+00003072 mexFunction+00000106

[ 2] 0x00002b105dd49080 bin/glnxa64/libmex.so+00413824

[ 3] 0x00002b105dd49447 bin/glnxa64/libmex.so+00414791

[ 4] 0x00002b105dd49f2b bin/glnxa64/libmex.so+00417579

[ 5] 0x00002b105dd3430c bin/glnxa64/libmex.so+00328460

[ 6] 0x00002b105bdca2ad bin/glnxa64/libmwm_dispatcher.so+00979629 _ZN8Mfh_file16dispatch_fh_implEMS_FviPP11mxArray_tagiS2_EiS2_iS2_+00000829

[ 7] 0x00002b105bdcabae bin/glnxa64/libmwm_dispatcher.so+00981934 _ZN8Mfh_file11dispatch_fhEiPP11mxArray_tagiS2_+00000030

[ 8] 0x00002b105ee70da1 bin/glnxa64/libmwm_lxe.so+12619169

[ 9] 0x00002b105ee71982 bin/glnxa64/libmwm_lxe.so+12622210

[ 10] 0x00002b105ef59fc9 bin/glnxa64/libmwm_lxe.so+13574089

[ 11] 0x00002b105eefb431 bin/glnxa64/libmwm_lxe.so+13186097

[ 12] 0x00002b105e7015a8 bin/glnxa64/libmwm_lxe.so+04822440

[ 13] 0x00002b105e703cbc bin/glnxa64/libmwm_lxe.so+04832444

[ 14] 0x00002b105e70001d bin/glnxa64/libmwm_lxe.so+04816925

[ 15] 0x00002b105e6f9ba1 bin/glnxa64/libmwm_lxe.so+04791201

[ 16] 0x00002b105e6f9dd9 bin/glnxa64/libmwm_lxe.so+04791769

[ 17] 0x00002b105e6ff846 bin/glnxa64/libmwm_lxe.so+04814918

[ 18] 0x00002b105e6ff92f bin/glnxa64/libmwm_lxe.so+04815151

[ 19] 0x00002b105e82e503 bin/glnxa64/libmwm_lxe.so+06055171

[ 20] 0x00002b105e831cf3 bin/glnxa64/libmwm_lxe.so+06069491

[ 21] 0x00002b105ed41f6d bin/glnxa64/libmwm_lxe.so+11378541

[ 22] 0x00002b105ecef60c bin/glnxa64/libmwm_lxe.so+11040268

[ 23] 0x00002b105ecf6448 bin/glnxa64/libmwm_lxe.so+11068488

[ 24] 0x00002b105ecf7e22 bin/glnxa64/libmwm_lxe.so+11075106

[ 25] 0x00002b105ed85807 bin/glnxa64/libmwm_lxe.so+11655175

[ 26] 0x00002b105ed85aea bin/glnxa64/libmwm_lxe.so+11655914

[ 27] 0x00002b105dab591a bin/glnxa64/libmwbridge.so+00207130 _Z8mnParserv+00000874

[ 28] 0x00002b105b7bebb8 bin/glnxa64/libmwmcr.so+00641976

[ 29] 0x00002b1048475e9f bin/glnxa64/libmwmlutil.so+06524575 _ZNSt13__future_base13_State_baseV29_M_do_setEPSt8functionIFSt10unique_ptrINS_12_Result_baseENS3_8_DeleterEEvEEPb+00000031

[ 30] 0x00002b1046e464f9 /cvmfs/soft.computecanada.ca/nix/var/nix/profiles/16.09/lib/libpthread.so.0+00058617

[ 31] 0x00002b1048476126 bin/glnxa64/libmwmlutil.so+06525222 _ZSt9call_onceIMNSt13__future_base13_State_baseV2EFvPSt8functionIFSt10unique_ptrINS0_12_Result_baseENS4_8_DeleterEEvEEPbEJPS1_S9_SA_EEvRSt9once_flagOT_DpOT0_+00000102

[ 32] 0x00002b105b7be9d3 bin/glnxa64/libmwmcr.so+00641491

[ 33] 0x00002b10435a61a2 bin/glnxa64/libmwmvm.so+03367330 _ZN14cmddistributor15PackagedTaskIIP10invokeFuncIN7mwboost8functionIFvvEEEEENS2_10shared_ptrINS2_13unique_futureIDTclfp_EEEEEERKT_+00000082

[ 34] 0x00002b10435a64e8 bin/glnxa64/libmwmvm.so+03368168 _ZNSt17_Function_handlerIFN7mwboost3anyEvEZN14cmddistributor15PackagedTaskIIP10createFuncINS0_8functionIFvvEEEEESt8functionIS2_ET_EUlvE_E9_M_invokeERKSt9_Any_data+00000024

[ 35] 0x00002b105b206e6c bin/glnxa64/libmwiqm.so+00867948 _ZN7mwboost6detail8function21function_obj_invoker0ISt8functionIFNS_3anyEvEES4_E6invokeERNS1_15function_bufferE+00000028

[ 36] 0x00002b105b20697f bin/glnxa64/libmwiqm.so+00866687 _ZN3iqm18PackagedTaskPlugin7executeEP15inWorkSpace_tagRN7mwboost10shared_ptrIN14cmddistributor17IIPCompletedEventEEE+00000447

[ 37] 0x00002b105b1e4ab1 bin/glnxa64/libmwiqm.so+00727729

[ 38] 0x00002b105b1c7ac8 bin/glnxa64/libmwiqm.so+00608968

[ 39] 0x00002b105b1c28bf bin/glnxa64/libmwiqm.so+00587967

[ 40] 0x00002b1044694a05 bin/glnxa64/libmwservices.so+03262981

[ 41] 0x00002b1044695ff2 bin/glnxa64/libmwservices.so+03268594

[ 42] 0x00002b10446968fb bin/glnxa64/libmwservices.so+03270907 _Z25svWS_ProcessPendingEventsiib+00000187

[ 43] 0x00002b105b7bffc3 bin/glnxa64/libmwmcr.so+00647107

[ 44] 0x00002b105b7c06a4 bin/glnxa64/libmwmcr.so+00648868

[ 45] 0x00002b105b7b93f1 bin/glnxa64/libmwmcr.so+00619505

[ 46] 0x00002b1046e3f1f4 /cvmfs/soft.computecanada.ca/nix/var/nix/profiles/16.09/lib/libpthread.so.0+00029172

[ 47] 0x00002b104554c16f /cvmfs/soft.computecanada.ca/nix/var/nix/profiles/16.09/lib/libc.so.6+00950639 clone+00000095

[ 48] 0x0000000000000000 <unknown-module>+00000000

This error was detected while a MEX-file was running. If the MEX-file

is not an official MathWorks function, please examine its source code

for errors. Please consult the External Interfaces Guide for information

on debugging MEX-files.

[Edric Ellis]

Not really an answer to your question as such - but note that if you have Parallel Computing Toolbox, you might be able to use pagefun - it doesn't support QR directly, but it does support batched mldivide...

[Naveen kumar Elumalai]

Thank you Edric Ellis for your answer, but i want my batched qr alone being done in my gpu.

Answer 1

[James Tursa]

Can you explain what you intended with these lines for A:

    double **A;
        :
    A = (double**)mxGetPr(prhs[0]);

If you pass in a regular double array, there are doubles in the data area of prhs[0], not pointers to doubles. You've got one too many levels of indirection here. What were your intentions with this?

Downstream in your code you appear to use A[i] as a pointer in a memory copy. Since there are doubles behind A, and not pointers to doubles behind A, you would be using a floating point double bit pattern as a pointer and this will crash MATLAB.

[Naveen kumar Elumalai]

• Hey james, thank you for your answer , A is a 3 dimensional matrix , I was using double **A in the C code as one pointer for the batchsize and the other for the size of the individual matrix . I thought i could do the same in the mex code. Sorry i am new to mex. in this code i wanted to copy each batch of the matrix one by one to the device memory. I wasnt sure how to do it
• i incorporated your changes to the code that is changing the double **A to double *A and in the cudamemcopy section i changed A[i] to A and now my code is not crashing , but i am getting the QR for only single batch of the A matrix as i reckon i am sending only one batch of the A matrix in the cudamemcopy as i changed from A[i] to A, so how do i send each and evey batch to the device memory?

[James Tursa]

Using A will point to the first batch (we typically use the term "plane" or "page" here to refer to the first 2D slice of a multi-dimensional array). To point to the next plane, simply increment the pointer by the appropriate amount. E.g.,

A points to the first plane
A+m*n points to the second plane
A+m*n*2 points to the third plane
A+m*n*3 points to the fourth plane
etc.

So, programatically you would simply use A+m*n*i as your pointer to the plane you want to process, where i is a 0-based index (like you currently have in your for-loop).

[Srinidhi Ganeshan]

Thank you james your answer worked