/************************************************************************

		C-DAC Tech Workshop : hyPACK-2013
                           October 15-18, 2013

        Objective   :  A set of programs for Memory Optimisation 
                       README info...

   Created          : August-2013

   E-mail          : hpcfte@cdac.in     

***********************************************************************/
" README-Part-II " text file has information about the following

	coalescedFloat3Access.cu*
	deviceDetails.cu*
	globalMemoryAccessPatterns.cu*
	sharedMemoryReadingSameWord.cu*
	sharedMemoryRestructuringDataTypes.cu*
	sharedMemoryStridedAccessPatterns.cu*
	SOAvsAOS.cu*


  MEMORY OPTIMIZATION TECHNIQUES ON nVIDIA-GPUs
  ---------------------------------------------

	I.  List of Programs
	II. Compilation
	III.Execution
	IV. Expected results

I. List of Programs
   -------------------


  deviceDetails.cu -- 

        contains example code to find out the number of devices that are present on the 
        current system and also to query using the cuda API calls about the different 
        properties of the devices.

coalescedFloat3Access.cu -- 
        contains example code to demonstrate the different access patterns of float3 
        array in the global memory the corresponding advantages in terms of the 
        bandwidth that is achievable

globalMemoryAccessPatterns.cu -- 
        contains example code to demonstrate the different access patterns of the global 
        memory the corresponding advantages in terms of the bandwidth that is achievable

sharedMemoryReadingSameWord.cu -- 
        This code demonstrates achievable shared memory bandwidth while reading same word

sharedMemoryStridedAccessPatterns.cu -- 
        This code demonstrates bank conflicts that can occur while accessing the shared 
        memory with strides

sharedMemoryRestructuringDataTypes.cu -- 
        This code demonstrates achievable shared memory bandwidth for different in-built 
        data types

SOAvsAOS.cu -- 
        Example code to demonstrate the advantage of having Stucture of arrays rather 
        than array of structures in the application while representing data the corresponding 
        advantages in terms of the bandwidth of the global memory that is achievable

        For more details about the programs see the documentation in "doc" folder


II. Compilation
-----------------

to make: $$make

      the make file generates the following executables:

        deviceDetails, coalescedFloat3Access, 
        globalMemoryAccessPatterns, sharedMemoryReadingSameWord, 
        sharedMemoryStridedAccessPatterns, sharedMemoryRestructuringDataTypes, 
        SOAvsAOS
				

III.Execution
----------------

to run:  $$./{executable}

for example to run the deviceDetails executable
			$$./deviceDetails



IV. Expected results
---------------------

  "deviceDetails" --- 
     Prints the different properties of the devices that are present on the system

  "coalescedFloat3Access" --- 
     prints the bandwidths that are achieved by the two Float3 access patterns - naive and 
     share-memory assisted

  "globalMemoryAccessPatterns" --- 
     prints the bandwidths achieved by the six different global access patterns

  "sharedMemoryReadingSameWord" --- 
     prints the bandwidths got when, reading the same word and reading from different 
     banks without any bank conflicts

  "sharedMemoryStridedAccessPatterns" --- 
     prints the bandwidths achieved while accessing shared memory with different strides

  "sharedMemoryRestructuringDataTypes" --- 
     prints the different bandwidths that are got while accessing the arrays of different 
     inbuilt types

  "SOAvsAOS" --- prints the achieved bandwidths when the same data is represented 
     using Structure of arrays or Array of structures

/***********************************************************************************