AM5728: GLSDK CMEM issue

Part Number: AM5728

Hello,

I'm using GLSDK-7.04 setup. I doing read and write from ARM core to shared memory CMEM.

To read and write 320 KB of data to CMEM it is taking more time(around 9.7 ms). 

Code:

#include <ti/ipc/Std.h>
#include <ti/cmem.h>
#include <pthread.h>
#include <semaphore.h>
#include <ti/ipc/Ipc.h>
#include <ti/ipc/MessageQ.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/time.h>
#include <time.h>
#include <string.h>

#define SUCCESS  (0)
#define FAILURE  (-1)
#define SHARED_MEM_SIZE (320*1024u)
#define HEAPID   0

char *s8SharedMem;

static long get_nanos(void) {
	struct timespec ts;
	clock_gettime(CLOCK_REALTIME, &ts);
	return (long)ts.tv_sec * 1000000000L + ts.tv_nsec;
}


int main(void)
{

	int s32ReturnValue;
	char num;
	CMEM_AllocParams    cmemAttrs;
		
	int status;
	char *ptr;
	long time1, time2;
	int i,j,k;
	int s32CacheInvChk;
	int s32Status;
	s32Status = 0;

	cmemAttrs.type = CMEM_HEAP;
	cmemAttrs.flags = CMEM_NONCACHED;
	cmemAttrs.alignment = 4;

	s32ReturnValue = SUCCESS;
	s32CacheInvChk		= 0;


	if(s32ReturnValue == SUCCESS)
	{
		if(CMEM_init() != SUCCESS)
		{
			printf("\nCmem initialization failed");
			s32ReturnValue = FAILURE;
		}
		else
		{
			printf("\ncmem init successfull");
		}
	}

	if(s32ReturnValue == SUCCESS)
	{
		if((s8SharedMem = CMEM_alloc(SHARED_MEM_SIZE, &cmemAttrs)) == NULL)
		{
			printf("\nCmem allocation failed");
			s32ReturnValue = FAILURE;
		}
		else
		{
			printf("\ncmem alloc successfull");
		}
		
	}


	if(s32ReturnValue == SUCCESS)
	{
		if((ptr = CMEM_alloc(SHARED_MEM_SIZE, &cmemAttrs)) == NULL)
		{
			printf("\nCmem allocation failed");
			s32ReturnValue = FAILURE;
		}
		else
		{
			printf("\ncmem alloc successfull");
		}
	}


	for(i=0; i < 1024; i++)
	{
		ptr[i] = 'c';
	}


	s32CacheInvChk = CMEM_cacheInv ((void *)ptr, 1024);

	if (0 == s32CacheInvChk )
	{
		time1 = get_nanos();
		for(i=0,j=0; i< 320; i++)
		{
			memcpy(s8SharedMem+j, (const char *)ptr, 1024);
			/*printf("s8SharedMem: = %x\n", s8SharedMem+j);*/
			j += 1024;
		}

		time2 = get_nanos();
		printf("\ntime taken to write:%ld",time2-time1);
		time1 = 0;
		time2 = 0;

		time1 = get_nanos();
		for(i=0,j=0; i< 320; i++)
		{
			memcpy(ptr+j, s8SharedMem+j, 1024);
			j += 1024;
		}
		time2 = get_nanos();

		printf("\ntime taken to read:%ld",time2-time1);
	}
	else
	{
		s32ReturnValue = FAILURE;
	}




	CMEM_free(s8SharedMem, &cmemAttrs);

	return s32ReturnValue;
}

Logs:

root@dra7xx-evm:~# ./pjtArm_profile 
CMEM Debug: init: entered - ref_count 0, cmem_fd -2
CMEM Debug: init: successfully opened /dev/cmem, matching driver version...
CMEM Debug: getVersion: entered
CMEM Debug: getVersion: exiting, ioctl CMEM_IOCGETVERSION returned 0x4000000
CMEM Debug: init: ... match good (0x4000000)
CMEM Debug: init: exiting, returning success

CMEM Debug: alloc: entered w/ size 0x50000, params - type HEAP, flags NONCACHED, align 0x4
CMEM Debug: allocHeap: allocated phys buffer 0xb5200000
CMEM Debug: allocHeap: mmap succeeded, returning virt buffer 0xb6e2b000
cmem init successfull
CMEM Debug: alloc: entered w/ size 0x50000, params - type HEAP, flags NONCACHED, align 0x4
CMEM Debug: allocHeap: allocated phys buffer 0xb5250000
CMEM Debug: allocHeap: mmap succeeded, returning virt buffer 0xb6ddb000
cmem alloc successfull
cmem alloc successfull
time taken to write:4571452
CMEM Debug: free: entered w/ ptr 0xb6e2b000, params - type HEAP
CMEM Debug: free: ioctl CMEM_IOCFREEHEAP succeeded, size 0x50000
CMEM Debug: free: munmap succeeded, returning 0
time taken to read:5150228

If we calculate the total time to perform write and read to CMEM is 9.7 ms(milli-seconds).

If we do memcpy of 320 KB data from heap to heap or global buffer to another global buffer it is taking 220 micro-seconds. so the problem is not with memcpy.

If we do same operation between heap/global buffer and CMEM it is taking 4.5 ms. So the problem is with CMEM only. 

Can anyone suggest me how to increase the speed ?