use cms::alpakatools::once_per_block

cms-sw · Jul 23, 2024 · 69debc2 · 69debc2
1 parent fac2958
commit 69debc2
Show file tree

Hide file tree

Showing 4 changed files with 22 additions and 6 deletions.
diff --git a/RecoTracker/LSTCore/src/alpaka/MiniDoublet.h b/RecoTracker/LSTCore/src/alpaka/MiniDoublet.h
@@ -1,6 +1,8 @@
 #ifndef RecoTracker_LSTCore_src_alpaka_MiniDoublet_h
 #define RecoTracker_LSTCore_src_alpaka_MiniDoublet_h
 
+#include "HeterogeneousCore/AlpakaInterface/interface/workdivision.h"
+
 #include "RecoTracker/LSTCore/interface/alpaka/Constants.h"
 #include "RecoTracker/LSTCore/interface/Module.h"
 #include "RecoTracker/LSTCore/interface/EndcapGeometry.h"
@@ -994,7 +996,9 @@ namespace lst {
 
       // Declare variables in shared memory and set to 0
       int& nTotalMDs = alpaka::declareSharedVar<int, __COUNTER__>(acc);
-      nTotalMDs = 0;
+      if (cms::alpakatools::once_per_block(acc)) {
+        nTotalMDs = 0;
+      }
       alpaka::syncBlockThreads(acc);
 
       // Initialize variables outside of the for loop.
@@ -1067,7 +1071,7 @@ namespace lst {
 
       // Wait for all threads to finish before reporting final values
       alpaka::syncBlockThreads(acc);
-      if (globalThreadIdx[2] == 0) {
+      if (cms::alpakatools::once_per_block(acc)) {
         rangesInGPU.miniDoubletModuleIndices[*modulesInGPU.nLowerModules] = nTotalMDs;
         *rangesInGPU.device_nTotalMDs = nTotalMDs;
       }

diff --git a/RecoTracker/LSTCore/src/alpaka/Quintuplet.h b/RecoTracker/LSTCore/src/alpaka/Quintuplet.h
@@ -1,6 +1,8 @@
 #ifndef RecoTracker_LSTCore_src_alpaka_Quintuplet_h
 #define RecoTracker_LSTCore_src_alpaka_Quintuplet_h
 
+#include "HeterogeneousCore/AlpakaInterface/interface/workdivision.h"
+
 #include "RecoTracker/LSTCore/interface/alpaka/Constants.h"
 #include "RecoTracker/LSTCore/interface/Module.h"
 #include "RecoTracker/LSTCore/interface/EndcapGeometry.h"
@@ -3109,8 +3111,10 @@ namespace lst {
       // Initialize variables in shared memory and set to 0
       int& nEligibleT5Modulesx = alpaka::declareSharedVar<int, __COUNTER__>(acc);
       int& nTotalQuintupletsx = alpaka::declareSharedVar<int, __COUNTER__>(acc);
-      nTotalQuintupletsx = 0;
-      nEligibleT5Modulesx = 0;
+      if (cms::alpakatools::once_per_block(acc)) {
+        nTotalQuintupletsx = 0;
+        nEligibleT5Modulesx = 0;
+      }
       alpaka::syncBlockThreads(acc);
 
       // Initialize variables outside of the for loop.

diff --git a/RecoTracker/LSTCore/src/alpaka/Segment.h b/RecoTracker/LSTCore/src/alpaka/Segment.h
@@ -1,6 +1,8 @@
 #ifndef RecoTracker_LSTCore_src_alpaka_Segment_h
 #define RecoTracker_LSTCore_src_alpaka_Segment_h
 
+#include "HeterogeneousCore/AlpakaInterface/interface/workdivision.h"
+
 #include "RecoTracker/LSTCore/interface/alpaka/Constants.h"
 #include "RecoTracker/LSTCore/interface/Module.h"
 #include "RecoTracker/LSTCore/interface/EndcapGeometry.h"
@@ -900,7 +902,9 @@ namespace lst {
 
       // Initialize variables in shared memory and set to 0
       int& nTotalSegments = alpaka::declareSharedVar<int, __COUNTER__>(acc);
-      nTotalSegments = 0;
+      if (cms::alpakatools::once_per_block(acc)) {
+        nTotalSegments = 0;
+      }
       alpaka::syncBlockThreads(acc);
 
       // Initialize variables outside of the for loop.

diff --git a/RecoTracker/LSTCore/src/alpaka/Triplet.h b/RecoTracker/LSTCore/src/alpaka/Triplet.h
@@ -1,6 +1,8 @@
 #ifndef RecoTracker_LSTCore_src_alpaka_Triplet_h
 #define RecoTracker_LSTCore_src_alpaka_Triplet_h
 
+#include "HeterogeneousCore/AlpakaInterface/interface/workdivision.h"
+
 #include "RecoTracker/LSTCore/interface/alpaka/Constants.h"
 #include "RecoTracker/LSTCore/interface/Module.h"
 
@@ -997,7 +999,9 @@ namespace lst {
 
       // Initialize variables in shared memory and set to 0
       int& nTotalTriplets = alpaka::declareSharedVar<int, __COUNTER__>(acc);
-      nTotalTriplets = 0;
+      if (cms::alpakatools::once_per_block(acc)) {
+        nTotalTriplets = 0;
+      }
       alpaka::syncBlockThreads(acc);
 
       // Initialize variables outside of the for loop.