IISWC

Reproducing The Results

Figure 6

Figure 7

The figure shows the percentage of time spent in data movement latency, PIM kernel execution and host execution. Following is an example to show how the bars for vector addition can be reproduced:

1. cd <PIMeval-PIMbench-dir>
2. Run make clean & make -j
3. cd <PIMbench/vec-add/PIM/>
4. run ./vec-add.out -l 2035544320 -c -c ../../../configs/iiswc/PIMeval_BitSerial_Rank32.cfg
5. The terminal should produce following stats:

PIM-Info: Current Device = PIM_FUNCTIONAL, Simulation Target = PIM_DEVICE_BITSIMD_V_AP
PIM-Info: Config: #ranks = 32, #bankPerRank = 128, #subarrayPerBank = 32, #rowsPerSubarray = 1024, #colsPerRow = 8192
PIM-Info: Aggregate every two subarrays as a single core
PIM-Info: Created PIM device with 65536 cores of 2048 rows and 8192 columns.
PIM-Info: Created thread pool with 63 threads.
----------------------------------------
PIM Params:
       	PIM Device Type Enum : PIM_FUNCTIONAL
      	PIM Simulation Target : PIM_DEVICE_BITSIMD_V_AP
 Rank, Bank, Subarray, Row, Col : 32, 128, 32, 1024, 8192
        	Number of PIM Cores : 65536
    	Number of Rows per Core : 2048
    	Number of Cols per Core : 8192
            	Typical Rank BW : 25.600000 GB/s
              	Row Read (ns) : 28.500000
             	Row Write (ns) : 43.500000
                  	tCCD (ns) : 3.000000
Data Copy Stats:
                           	Host to Device : 16284354560 bytes
                           	Device to Host : 8142177280 bytes
                         	Device to Device : 0 bytes
                          	TOTAL --------- : 24426531840 bytes  	27.769752 ms Estimated Runtime	1592.428655 mj Estimated Energy
PIM Command Stats:
                                  	PIM-CMD :    	CNT EstimatedRuntime(ms) EstimatedEnergyConsumption(mJ)
                              	add.int32.v :      	1   	0.014028  	13.261247
                          	TOTAL --------- :      	1   	0.014028  	13.261247
----------------------------------------