LLVM，polly，最新测试

1）gemm 测试

GCC编译：

gcc -O3 -I/home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities    /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities/polybench.c  /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/gemm/gemm.c  -o a1.gemm.gcc.o3

clang编译：

clang -O3 -I/home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities    /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities/polybench.c  /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/gemm/gemm.c  -o a2.gemm.clang.o3

clang+polly编译：

clang -O3 -mllvm -polly -I/home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities    /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities/polybench.c  /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/gemm/gemm.c  -o a3.gemm.polly.o3

运行时间

a@a:~/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/gemm$ time ./a1.gemm.gcc.o3 

real    0m10.964s
user    0m10.895s
sys    0m0.043s
 

a@a:~/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/gemm$ time ./a2.gemm.clang.o3 

real    0m1.060s
user    0m0.983s
sys    0m0.070s

@a:~/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/gemm$ time ./a3.gemm.polly.o3 

real    0m0.775s
user    0m0.684s
sys    0m0.089s
 

2）

gcc -DLARGE_DATASET -O3 -I/home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities    /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities/polybench.c  /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/gemm/gemm.c  -o a1.gemm.gcc.o3

clang -DLARGE_DATASET -O3 -I/home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities    /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities/polybench.c  /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/gemm/gemm.c  -o a2.gemm.clang.o3

clang -DLARGE_DATASET -O3 -mllvm -polly -I/home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities    /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities/polybench.c  /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/gemm/gemm.c  -o a3.gemm.polly.o3

time ./a1.gemm.gcc.o3
time ./a2.gemm.clang.o3
time ./a3.gemm.polly.o3

real    0m1.181s
user    0m1.104s
sys    0m0.065s

real    0m1.305s
user    0m1.261s
sys    0m0.040s

real    0m0.775s
user    0m0.635s
sys    0m0.140s
 

这个结果就有点正确 了

3）

a@a:~/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/gemm$ 
gcc -DEXTRALARGE_DATASET -O3 -I/home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities    /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities/polybench.c  /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/gemm/gemm.c  -o a1.gemm.gcc.o3

clang -DEXTRALARGE_DATASET -O3 -I/home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities    /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities/polybench.c  /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/gemm/gemm.c  -o a2.gemm.clang.o3

clang -DEXTRALARGE_DATASET -O3 -mllvm -polly -I/home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities    /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities/polybench.c  /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/gemm/gemm.c  -o a3.gemm.polly.o3

time ./a1.gemm.gcc.o3
time ./a2.gemm.clang.o3
time ./a3.gemm.polly.o3

real    0m14.763s
user    0m14.061s
sys    0m0.680s

real    0m12.724s
user    0m12.534s
sys    0m0.175s

real    0m5.839s
user    0m5.677s
sys    0m0.153s
 

polly的效果非常好了。

4）

对于symm，clang和polly的效果都不好。

gcc -DLARGE_DATASET -O3 -I/home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities    /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities/polybench.c  /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/symm/symm.c  -o a1.gcc.o3

clang -DLARGE_DATASET -O3 -I/home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities    /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities/polybench.c  /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/symm/symm.c  -o a2.clang.o3

clang -DLARGE_DATASET -O3 -mllvm -polly -I/home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities    /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/utilities/polybench.c  /home/a/src/a3.llvm.polly/a1.polybench-c-4.2/linear-algebra/blas/symm/symm.c  -o a3.polly.o3

time ./a1.gcc.o3
time ./a2.clang.o3
time ./a3.polly.o3

real    0m5.529s
user    0m5.467s
sys    0m0.040s

real    0m6.629s
user    0m6.551s
sys    0m0.050s

real    0m6.223s
user    0m6.171s
sys    0m0.040s
 

5）

a@a:~/src/a3.llvm.polly/a1.polybench-c-4.2$ ./run.sh -DEXTRALARGE_DATASET linear-algebra/blas/gemver/gemver.c

real    0m0.793s
user    0m0.462s
sys    0m0.330s

real    0m0.666s
user    0m0.521s
sys    0m0.141s

real    0m0.315s
user    0m0.148s
sys    0m0.166s
gemver效果很好

6）

a@a:~/src/a3.llvm.polly/a1.polybench-c-4.2$ ./run.sh -DEXTRALARGE_DATASET linear-algebra/blas/gesummv/gesummv.c

real    0m0.465s
user    0m0.133s
sys    0m0.330s

real    0m0.295s
user    0m0.131s
sys    0m0.163s

real    0m0.277s
user    0m0.128s
sys    0m0.149s
 

7）syr2k效果也不行

a@a:~/src/a3.llvm.polly/a1.polybench-c-4.2$ ./r2.2p.sh -DLARGE_DATASET linear-algebra/blas/syr2k/syr2k.c

real    0m23.820s
user    0m23.369s
sys    0m0.346s

real    0m24.247s
user    0m24.160s
sys    0m0.063s

real    0m25.786s
user    0m25.684s
sys    0m0.063s
 

8）syrk这个也不行

a@a:~/src/a3.llvm.polly/a1.polybench-c-4.2$ ./r2.2p.sh -DLARGE_DATASET linear-algebra/blas/syrk/syrk.c

real    0m1.881s
user    0m1.831s
sys    0m0.046s

real    0m1.833s
user    0m1.792s
sys    0m0.037s

real    0m1.858s
user    0m1.809s
sys    0m0.045s
9）trmm效果很好

a@a:~/src/a3.llvm.polly/a1.polybench-c-4.2$ ./r2.2p.sh -DLARGE_DATASET linear-algebra/blas/trmm/trmm.c

real    0m7.962s
user    0m7.877s
sys    0m0.074s

real    0m8.054s
user    0m7.928s
sys    0m0.099s

real    0m3.783s
user    0m3.749s
sys    0m0.029s