Merge pull request #45 from roussian23/addcuda

add cuda benchmark

Author: jaimesouza

benchmark/marmousi_2D.py

@@ -49,6 +49,12 @@ def get_args(args=sys.argv[1:]):
         cflags = '-O3 -fPIC -ffast-math -fopenmp \
                -fopenmp-targets=nvptx64-nvidia-cuda \
                -Xopenmp-target -march=sm_75'
+    elif language == 'cuda':
+        cc = 'nvcc'
+        cflags = '-O3 -gencode arch=compute_75,code=sm_75 \
+                  --compiler-options -fPIC,-Wall \
+                  --use_fast_math -std=c++17 -shared \
+                  -DDEBUG -DTX=32 -DTY=4 -DTZ=2'
     else:
         raise ValueError('Language not available')
 

benchmark/overthrust_3D.py

@@ -56,6 +56,12 @@ def read_model(filename):
         cflags = '-O3 -fPIC -ffast-math -fopenmp \
                -fopenmp-targets=nvptx64-nvidia-cuda \
                -Xopenmp-target -march=sm_75'
+    elif language == 'cuda':
+        cc = 'nvcc'
+        cflags = '-O3 -gencode arch=compute_75,code=sm_75 \
+                  --compiler-options -fPIC,-Wall \
+                  --use_fast_math -std=c++17 -shared \
+                  -DDEBUG -DTX=32 -DTY=4 -DTZ=2'
     else:
         raise ValueError('Language not available')