Re: [atlas-devel] PowerPC performance, was Floating point operations in DGEMM and DSYRK
Brought to you by:
rwhaley,
tonyc040457
From: José L. G. P. <jgp...@gm...> - 2013-03-25 19:08:38
|
2013/3/25 Brooks Moses <br...@co...>: > Hello, José! > > José Luis García Pallero wrote, at 3/7/2013 3:20 PM: >> GEMM: M=N=K= 100 -> 2.568 GFLOPS/s >> GEMM: M=N=K= 200 -> 2.345 GFLOPS/s >> GEMM: M=N=K= 300 -> 2.447 GFLOPS/s >> GEMM: M=N=K= 400 -> 2.404 GFLOPS/s > [...] > > One last dumb question -- what program did you use to get these results? > I finally have a working ATLAS build on our Freescale e6500 machine and > am ready to see how the performance compares. Hi: Attached is my dummy program (is quite simple, but is suitable for a fast and simple test). Remember: 1. #define FLOAT or #define DOUBLE in order to work with float or double data 2. #define DLIM XXX to set the maximum matrix dimensions 3. #define START XXX to set the minumum matrix dimensions 4. #define INC XXX to set the increment in dimensions between matrices 5. #define R XX to set the number of times a computatios is repeated (the final result is the average) And remember too to link against -lrt library in order to use the function clock_gettime() for time computations gcc -Wall -Wextra benchGemmSyrk.c .o benchGemmSyrk -latlas -lcblas -lrt Cheers PD: And post your results, please ;) > > Thanks again, > - Brooks > -- ***************************************** José Luis García Pallero jgp...@gm... (o< / / \ V_/_ Use Debian GNU/Linux and enjoy! ***************************************** |