1.Whetstone-A synthetic benchmark for evaluating the overall performance of computing float-point data rating in Millions of Whetstone Instructions Per Second (MWIPS) plus derived measurements of Millions of Floating Point Instructions Per Second (MFLOPS) . The overall MWIPS calculation proportional to CPU MHz on a given type of processor is used as the measuring benchmark via MWIPS/MHZ.
2.Use the Whetstone demo to evaluate the performance of D10 with SPU in AndeSight IDE.First,import the Whetstone demo (attached at the end of this page) into AndeSight,and then start to set Build Settings as follows.
(1).Select ADP-AG101P-4GB-N1068A-SPU-32I as Target Configuration and nds32le-elf-mculib-v3s
as Tool Chain.
Attachment:
Build Settings-Target Configuration.png [ 25.84 KiB | Viewed 9160 times ]
Attachment:
Build Setting-tool chain.png [ 24.07 KiB | Viewed 9158 times ]
(2)For compiler's Build Optimization Seting,add"-funroll-loops --param max-unroll-times=4 -fsched-pressure --param sched-pressure-algorithm=2 -fno-strict-aliasing"in other optimization flags option.At the same time,select"Link Time Optimization(-flto)"option.
Attachment:
Build optimization setting.png [ 48.81 KiB | Viewed 9156 times ]
(3).For compiler's Build Micesllaneous Seting,select "Single Precision Constant"option.
Attachment:
Build Settings compiler-Micesllaneous.png [ 30.44 KiB | Viewed 9155 times ]
(4).For Linker setting,add math library in Libraries.
Attachment:
Build settings-math library.png [ 39.54 KiB | Viewed 9154 times ]
(5).For Micesllaneous Linker flags,add "-funroll-loops --param max-unroll-times=4 -fsched-pressure --param sched-pressure-algorithm=2 -fno-strict-aliasing".
Attachment:
Build settings-linker-Micesllaneous.png [ 39.77 KiB | Viewed 9153 times ]
3.Then compile the whole project and run on FPGA after having done the Build Settings.For D10 with SPU,the Whetstone Score on FPGA is 1.094 MWIPS/MHZ.
Attachment:
Whetstone run result for D10+SPU.png [ 70.94 KiB | Viewed 9152 times ]