move forward kernels in elastic.c into separate functions so that the…

…y are vectorized even when using OpenMP
ar4 · Jul 21, 2023 · e5c54c7 · e5c54c7
1 parent 3e6de3c
commit e5c54c7
Show file tree

Hide file tree

Showing 6 changed files with 182 additions and 125 deletions.
diff --git a/src/deepwave/build_linux.sh b/src/deepwave/build_linux.sh
@@ -3,7 +3,7 @@
 set -e
 
 DW_OMP_NAME=libgomp.so.1
-CFLAGS="-Wall -Wextra -pedantic -DDW_USE_OPENMP -fPIC -fopenmp -Ofast -mavx2"
+CFLAGS="-Wall -Wextra -pedantic -fPIC -fopenmp -Ofast -mavx2"
 CUDAFLAGS="--restrict --use_fast_math -O3 -gencode=arch=compute_52,code=sm_52, -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=compute_80  --compiler-options -fPIC"
 gcc $CFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=float -c scalar.c -o scalar_cpu_iso_2_float.o
 gcc $CFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=float -c scalar.c -o scalar_cpu_iso_4_float.o

diff --git a/src/deepwave/build_macos.sh b/src/deepwave/build_macos.sh
@@ -3,7 +3,7 @@
 set -e
 
 DW_OMP_NAME=iomp5
-CFLAGS="-Wall -Wextra -pedantic -DDW_USE_OPENMP -fPIC -Ofast -Xpreprocessor -fopenmp -I`brew --prefix libomp`/include"
+CFLAGS="-Wall -Wextra -pedantic -fPIC -Ofast -Xpreprocessor -fopenmp -I`brew --prefix libomp`/include"
 clang $CFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=float -c scalar.c -o scalar_cpu_iso_2_float.o
 clang $CFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=float -c scalar.c -o scalar_cpu_iso_4_float.o
 clang $CFLAGS -DDW_ACCURACY=6 -DDW_DTYPE=float -c scalar.c -o scalar_cpu_iso_6_float.o

diff --git a/src/deepwave/common_cpu.h b/src/deepwave/common_cpu.h
@@ -1,12 +1,6 @@
 #ifndef DW_COMMON_CPU_H
 #define DW_COMMON_CPU_H
 
-//#ifdef DW_USE_OPENMP
-//int dw_use_openmp = 1;
-//#else
-//int dw_use_openmp = 0;
-//#endif /* DW_USE_OPENMP */
-
 static void add_sources(DW_DTYPE *__restrict const wf,
                         DW_DTYPE const *__restrict const f,
                         int64_t const *__restrict const sources_i,