sycl: Fix broken sqrt call

sycl: Better compiler flags
UoB-HPC · Nov 19, 2020 · f527c4c · f527c4c
1 parent 099e6ed
commit f527c4c
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 4 deletions.
diff --git a/sycl/CMakeLists.txt b/sycl/CMakeLists.txt
@@ -78,7 +78,7 @@ endif ()
 add_definitions(-DNUM_TD_PER_THREAD=${NUM_TD_PER_THREAD})
 
 set(DEBUG_OPTIONS -O2 -fno-omit-frame-pointer ${CXX_EXTRA_FLAGS})
-set(RELEASE_OPTIONS -Ofast -ffast-math -march=native ${CXX_EXTRA_FLAGS})
+set(RELEASE_OPTIONS -Ofast -ffast-math -march=native -mtune=native ${CXX_EXTRA_FLAGS})
 
 target_compile_options(bude PUBLIC "$<$<CONFIG:RelWithDebInfo>:${RELEASE_OPTIONS}>")
 target_compile_options(bude PUBLIC "$<$<CONFIG:Release>:${RELEASE_OPTIONS}>")

diff --git a/sycl/src/vec-pose-inner.cpp b/sycl/src/vec-pose-inner.cpp
@@ -150,10 +150,12 @@ void fasten_main(
 					const float x = lpos[i].x() - p_atom.x;
 					const float y = lpos[i].y() - p_atom.y;
 					const float z = lpos[i].z() - p_atom.z;
-					const float distij = clsycl::native::sqrt(x * x + y * y + z * z);
 
-					//TODO replace with:
-					// const float distij = clsycl::distance(lpos[id], clsycl::float3(p_atom.x, p_atom.y, p_atom.z));
+					// XXX as of oneapi-2021.1-beta10, the cl::sycl::native::sqrt variant is significantly slower for no apparent reason
+					const float distij = clsycl::sqrt(x * x + y * y + z * z);
+
+					// XXX as of oneapi-2021.1-beta10, the following variant is significantly slower for no apparent reason
+					//const float distij = clsycl::distance(lpos[i], clsycl::float3(p_atom.x, p_atom.y, p_atom.z));
 
 					// Calculate the sum of the sphere radii
 					const float distbb = distij - radij;