[OpenMP] Fix incorrect use of align instead of round (#196191)

openmp/device/src/Mapping.cpp

diff --git a/openmp/device/src/Mapping.cpp b/openmp/device/src/Mapping.cpp
index 5c51da6..1096e5f 100644
--- a/openmp/device/src/Mapping.cpp
+++ b/openmp/device/src/Mapping.cpp
@@ -131,7 +131,7 @@
 }
 
 uint32_t mapping::getNumberOfWarpsInBlock() {
-  return utils::alignUp(mapping::getNumberOfThreadsInBlock(),
+  return utils::roundUp(mapping::getNumberOfThreadsInBlock(),
                         mapping::getWarpSize());
 }
 

openmp/device/src/Reduction.cpp

diff --git a/openmp/device/src/Reduction.cpp b/openmp/device/src/Reduction.cpp
index 271d316..b1b1e0d 100644
--- a/openmp/device/src/Reduction.cpp
+++ b/openmp/device/src/Reduction.cpp
@@ -135,7 +135,7 @@
   //
   // Only L1 parallel region can enter this if condition.
   if (NumThreads > mapping::getWarpSize()) {
-    uint32_t WarpsNeeded = utils::alignUp(NumThreads, mapping::getWarpSize());
+    uint32_t WarpsNeeded = utils::roundUp(NumThreads, mapping::getWarpSize());
     // Gather all the reduced values from each warp
     // to the first warp.
     cpyFct(reduce_data, WarpsNeeded);
@@ -280,7 +280,7 @@
       uint32_t ActiveThreads = kmpcMin(NumRecs, NumThreads);
       if (ActiveThreads > mapping::getWarpSize()) {
         uint32_t WarpsNeeded =
-            utils::alignUp(ActiveThreads, mapping::getWarpSize());
+            utils::roundUp(ActiveThreads, mapping::getWarpSize());
         // Gather all the reduced values from each warp
         // to the first warp.
         cpyFct(reduce_data, WarpsNeeded);