cuda malloc and mecpy error or is it?

sachin · May 22, 2009, 6:15am

is it just me or am i going mad. i send out a vector to the GPU so that it can take the absolute value. the cpu version works fine. but everytime i send it to GPU version it sends me back the exact same vector.

im guessing that somewhere its just copying the input pointer to the output pointer.

heres my test function. just look at bold section

[codebox]#include <cuda.h>

#include <cuda_runtime_api.h>

#include “text_write.h”

#define ELEMENTS 100

int main (void){

PIX_TYPE *pixarray=NULL;

INT1_TYPE *int1array=NULL, *int1_GPU=NULL, *int1_CPU=NULL;

INT4_TYPE *int4array=NULL;

UINT1_TYPE *uint1array=NULL;

UINT4_TYPE *uint4array=NULL;

DBL_TYPE *dblarray=NULL;

int i=0;



/*create a matrix of ELEMENTS for each of above types randomly*/

pixarray=(PIX_TYPE *)malloc(sizeof(PIX_TYPE)*ELEMENTS);

for (i=0; i<ELEMENTS;i++){

	pixarray[i]=(PIX_TYPE)rand();

}

[b]int1array=(INT1_TYPE *)malloc(sizeof(INT1_TYPE)*ELEMENTS);

for (i=0; i<ELEMENTS;i++){

	int1array[i]=(INT1_TYPE)rand();/*create random array*/

	if (i%2==0)/*for every second item*/

		int1array[i]=-int1array[i];/*make number negative*/

}

[/b]

int4array=(INT4_TYPE *)malloc(sizeof(INT4_TYPE)*ELEMENTS);

for (i=0; i<ELEMENTS;i++){

	int4array[i]=(INT4_TYPE)rand();

}

uint1array=(UINT1_TYPE *)malloc(sizeof(UINT1_TYPE)*ELEMENTS);

for (i=0; i<ELEMENTS;i++){

	uint1array[i]=(UINT1_TYPE)rand();

}

uint4array=(UINT4_TYPE *)malloc(sizeof(UINT4_TYPE)*ELEMENTS);

for (i=0; i<ELEMENTS;i++){

	uint4array[i]=(UINT4_TYPE)rand();

}

[b]int1_CPU=(INT1_TYPE *)calloc(ELEMENTS,sizeof(INT1_TYPE));

int1_GPU=(INT1_TYPE *)calloc(ELEMENTS,sizeof(INT1_TYPE));[/b]

//for (i=0; i<ELEMENTS;i++){

//	int1array[i]=(INT1_TYPE)rand()/rand())-5;

//}

[b]if (l_abs_gpu((void *)int1array,INT1_CODE, int1_GPU, ELEMENTS))/*l_abs_gpu returns 1 if int1array and int1GPU are the same pointer*/

	printf("Error\n");

l_abs((void *)int1array,INT1_CODE, int1_CPU, ELEMENTS);[/b]

text_write("int1 array input 2nd try", INT1_CODE, int1array);

text_write("int1 abs for GPU", INT1_CODE, int1_GPU);

text_write("int1 abs for CPU", INT1_CODE, int1_CPU);

free(int1_CPU);

free(int1_GPU);

free(pixarray);

free(int1array);

free(int4array);

free(uint1array);

free(uint4array);

free(dblarray);

return 1;

}[/codebox]

Heres my kernel:

[codebox]#include <cuda.h>

#include <cuda_runtime_api.h>

#include <liarp.h>

#include “arith_const_type.h”

//prototypes

global void abs_ (void *in1, int code_in, long elements);

global void double_ (INT1_TYPE *data, int elements);

int l_abs_gpu(void in, / ptr to first input image */

  int code_in,  /* code of the input pixel type */

  void *out,    /* ptr to output image */

  long nbpix)   /* size of input & ouput in pixels */

{

INT1_TYPE    *ci;

INT1_TYPE *dev_ci;

INT4_TYPE    *ii, *dev_ii=NULL;

DBL_TYPE     *di, *dev_di=NULL;

dim3 dimBlock(512,1 );

dim3 dimGrid(round_up(nbpix,dimBlock.x),1);

[b] if (code_in == INT1_CODE) {

	ci = (INT1_TYPE *)in;

	cudaMalloc((void **)&dev_ci, sizeof(INT1_TYPE)*nbpix);/*allocate memory in GPU*/

	cudaMemcpy(dev_ci,ci,sizeof(INT1_TYPE)*nbpix,cudaMemcpyHostT

oDevice);/copy to GPU/

	double_<<<dimGrid, dimBlock>>>(dev_ci, nbpix);

	//abs_<<<dimGrid, dimBlock>>>(dev_ci, code_in, nbpix);/*get absolute value of elements*/

	cudaMemcpy(out,dev_ci,sizeof(INT1_TYPE)*nbpix,cudaMemcpyDevi

ceToHost);/copy back to normal memory/

	cudaFree(dev_ci);/*free cuda pointer*/

}[/b]

else if (code_in == INT4_CODE) {

	ii = (INT4_TYPE *)in;

	cudaMalloc((void **)&dev_ii, sizeof(INT4_TYPE)*nbpix);/*allocate memory in GPU*/

	cudaMemcpy(dev_ii,ii,sizeof(INT4_TYPE)*nbpix,cudaMemcpyHostT

oDevice);/copy to GPU/

	abs_<<<dimGrid, dimBlock>>>(dev_ii, code_in, nbpix);/*get absolute value of elements*/

	cudaMemcpy(out,dev_ii,sizeof(INT4_TYPE)*nbpix,cudaMemcpyDevi

ceToHost);/copy back to normal memory/

	cudaFree(dev_ii);/*free cuda pointer*/

}

else if(code_in == DBL_CODE) {

	di = (DBL_TYPE *)in;

	CUDA_SAFE_CALL(cudaMalloc((void **)&dev_di, sizeof(DBL_TYPE)*nbpix));/*allocate memory in GPU*/

	CUT_CHECK_ERROR("cudaMalloc failed");

	CUDA_SAFE_CALL(cudaMemcpy(dev_di,di,sizeof(DBL_TYPE)*nbpix,c

udaMemcpyHostToDevice));/copy to GPU/

	CUT_CHECK_ERROR("cudaMemcpy failed");

	abs_<<<dimGrid, dimBlock>>>(dev_di, code_in, nbpix);/*get absolute value of elements*/

	CUDA_SAFE_CALL(cudaMemcpy(out,dev_di,sizeof(DBL_TYPE)*nbpix,

cudaMemcpyDeviceToHost));/copy back to normal memory/

	CUT_CHECK_ERROR("cudaMemcpy failed");

	cudaFree(dev_di);/*free cuda pointer*/

}	

else {

  LIARerror("Input type not supported\n");

  return(1);

}

if (out==in)

	return 1;

return 0;

}

global void abs_ (void *in1, int code_in, long elements){

INT1_TYPE 	*ci;/*same as char*/

INT4_TYPE 	*ii;/*same as int*/

DBL_TYPE	*di;/*same as double*/

int i = blockIdx.x * blockDim.x + threadIdx.x;/*get thread number*/

[b] if (i<elements){/if less than elements/

	if (code_in == INT1_CODE ) {

		ci=(INT1_TYPE *)in1;

		if (ci[i]<0)/*if negative number*/

			ci[i]=-ci[i];/*times by minus one*/

		//ci[i] = ci[i] >= 0 ? ci[i] : -ci[i];

	} [/b]

	else if (code_in == INT4_CODE) {

		ii = (INT4_TYPE *)in1;

		ii[i] = ii[i] >= 0 ? ii[i] : -ii[i];

	} 

	else if (code_in == DBL_CODE) {

		di = (DBL_TYPE *)in1;

		di[i] = di[i] >= 0 ? di[i] : -di[i];

	}

}

}

[b]/this is an alternate function to abs that i created that should double the values/

global void double_ (INT1_TYPE *data, int elements){

int i = blockIdx.x * blockDim.x + threadIdx.x;/*get thread no*/

if (i<elements){

	data[i]=2*data[i];

}

}[/b][/codebox]

i doubt you will need it but the text_write function is as follows:

[codebox]void text_write(const char *msg, int type, void *data){

FILE *file=NULL;

PIX_TYPE *pixarray=NULL;

INT1_TYPE *int1array=NULL;

INT4_TYPE *int4array=NULL;

UINT1_TYPE *uint1array=NULL;

UINT4_TYPE *uint4array=NULL;

DBL_TYPE *dblarray=NULL;

int i=0;

file=fopen("test.txt","a");	

fprintf(file, msg);/*type in the message as required*/

switch (type){

	case PIX_CODE:

		for (i=0; i<100;i++){

			if (i%10==0)

				fprintf(file,"\n"); /*insert new line every 10 lines*/

			fprintf(file,"%d\t",((PIX_TYPE *)data)[i]);

		}

		break;

	case INT1_CODE:

		for (i=0; i<100;i++){

			if (i%10==0)

				fprintf(file,"\n"); /*insert new line every 10 lines*/

			fprintf(file,"%d\t",((INT1_TYPE *)data)[i]);

		}

		break;

	case INT4_CODE:

		for (i=0; i<100;i++){

			if (i%10==0)

				fprintf(file,"\n"); /*insert new line every 10 lines*/

			fprintf(file,"%d\t",((INT4_TYPE *)data)[i]);

		}

		break;

	//case UINT1_CODE:

	//	for (i=0; i<100;i++){

	//		if (i%10==0)

	//			fprintf(file,"\n"); /*insert new line every 10 lines*/

	//		fprintf(file,"%d\t",((UINT1_TYPE *)data)[i]);

	//	}

	//	break;

	case UINT4_CODE:

		for (i=0; i<100;i++){

			if (i%10==0)

				fprintf(file,"\n"); /*insert new line every 10 lines*/

			fprintf(file,"%d\t",((UINT4_TYPE *)data)[i]);

		}

		break;

	case DBL_CODE:

		for (i=0; i<100;i++){

			if (i%10==0)

				fprintf(file,"\n"); /*insert new line every 10 lines*/

			fprintf(file,"%.4g\t",((DBL_TYPE *)data)[i]);

		}

		break;

	default:

		printf("Error:wrong code\n");

		break;

}

fprintf(file,"\n"); /*insert new line*/

fclose(file);

}[/codebox]

and my errors are

[codebox]‘LIAR.exe’: Loaded ‘C:\Documents and Settings\abe044\My Documents\Visual Studio 2005\liar_arith\LIAR\win32\LIAR\Debug\LIAR.exe’, Symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\ntdll.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\kernel32.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\CUDA\bin\cudart.dll’, Binary was not built with debug information.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\user32.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\gdi32.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\shimeng.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\imm32.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\advapi32.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\rpcrt4.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\secur32.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\lpk.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\usp10.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\msvcrt.dll’, No symbols loaded.

‘LIAR.exe’: Unloaded ‘C:\WINDOWS\system32\shimeng.dll’

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\nvcuda.dll’, Binary was not built with debug information.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\nvapi.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\ole32.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\oleaut32.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\shlwapi.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\shell32.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\setupapi.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\version.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\WinSxS\x86_Microsoft.Windows.Common-Controls_6595b64144ccf1df_6.0.2600.2982_x-ww_ac3f9c03\comctl32.dll’, No symbols loaded.

‘LIAR.exe’: Loaded ‘C:\WINDOWS\system32\comctl32.dll’, No symbols loaded.

First-chance exception at 0x7c812a6b in LIAR.exe: Microsoft C++ exception: cudaError_enum at memory location 0x0013fcfc…

First-chance exception at 0x7c812a6b in LIAR.exe: Microsoft C++ exception: cudaError at memory location 0x0013fd4c…

[/codebox]

sry about the length of this post but im really desperate for some answers

sachin · May 23, 2009, 7:10am

ok well aftert some research iver narrowed the error down to one POSSIBLE ERROR. tell me if im right… i cant test it out myself since i dont have a nvidia card at home. and all work done above was at work.

neways… seems something to do with the fact that im not using cudaSymbolCpy?? from what i understnd it seems that its telling me that im sending a CPU pointer to the GPU (which i get is nto what your meant to do). what i thought would happen is if you declare say GPU_pointer=NULL then you can just give it a GPU address at will. any thoughts?

Topic		Replies	Views
CUT_DEVICE_INIT exited with code 1 CUDA Programming and Performance	10	8637	June 24, 2008
cuda Error at memory location CUDA Programming and Performance	1	2583	April 5, 2012
Strange cudaErrors popping up during dll loading. CUDA Programming and Performance	5	12494	October 20, 2011
Emulation works, Debug doesn't CUDA Programming and Performance	12	2655	January 29, 2010
Exception in execution CUDA Programming and Performance	8	11119	February 14, 2009
CudaAPI calls in functions, compiler/linking bug? CUDA Programming and Performance	6	367	August 17, 2023
CUDA Error: Invalid Device Function Debugging CUDA errors CUDA Programming and Performance	3	5762	July 29, 2009
unresolved external symbol _main referenced in function ___tmainCRTStartup CUDA Programming and Performance	7	9312	February 22, 2011
First Chance Exception After Program Exits MS Visual Studio 2005 Exception in Provided Example Code CUDA Programming and Performance	7	6328	August 4, 2009
Can we do malloc inside a __global__ function CUDA Programming and Performance	26	9686	February 21, 2010

cuda malloc and mecpy error or is it?

Related topics