Merge pull request #70 from slava77/CMSSW_14_1_0_pre5/LSTb4-buf-lite

alpaka modernise/improve for batch4
cms-sw · Jul 30, 2024 · 3fc9904 · 3fc9904
2 parents fac2958 + d5e0b69
commit 3fc9904
Show file tree

Hide file tree

Showing 8 changed files with 90 additions and 89 deletions.
diff --git a/RecoTracker/LSTCore/interface/alpaka/Constants.h b/RecoTracker/LSTCore/interface/alpaka/Constants.h
@@ -35,11 +35,6 @@ namespace lst {
   };
 #endif
 
-  // Wrapper function to reduce code boilerplate for defining grid/block sizes.
-  ALPAKA_FN_HOST ALPAKA_FN_INLINE Vec3D createVec(int x, int y, int z) {
-    return Vec3D(static_cast<Idx>(x), static_cast<Idx>(y), static_cast<Idx>(z));
-  }
-
   // Adjust grid and block sizes based on backend configuration
   template <typename Vec>
   ALPAKA_FN_HOST ALPAKA_FN_INLINE WorkDiv3D createWorkDiv(const Vec& blocksPerGrid,

diff --git a/RecoTracker/LSTCore/src/EndcapGeometry.cc b/RecoTracker/LSTCore/src/EndcapGeometry.cc
@@ -49,9 +49,6 @@ void lst::EndcapGeometry::fillGeoMapArraysExplicit() {
 }
 
 float lst::EndcapGeometry::getdxdy_slope(unsigned int detid) const {
-  if (dxdy_slope_.find(detid) != dxdy_slope_.end()) {
-    return dxdy_slope_.at(detid);
-  } else {
-    return 0;
-  }
+  auto res = dxdy_slope_.find(detid);
+  return res == dxdy_slope_.end() ? 0.f : res->second;
 }
diff --git a/RecoTracker/LSTCore/src/TiltedGeometry.cc b/RecoTracker/LSTCore/src/TiltedGeometry.cc
@@ -33,17 +33,11 @@ void lst::TiltedGeometry::load(std::string filename) {
 }
 
 float lst::TiltedGeometry::getDrDz(unsigned int detid) const {
-  if (drdzs_.find(detid) != drdzs_.end()) {
-    return drdzs_.at(detid);
-  } else {
-    return 0;
-  }
+  auto res = drdzs_.find(detid);
+  return res == drdzs_.end() ? 0.f : res->second;
 }
 
 float lst::TiltedGeometry::getDxDy(unsigned int detid) const {
-  if (dxdys_.find(detid) != dxdys_.end()) {
-    return dxdys_.at(detid);
-  } else {
-    return 0;
-  }
+  auto res = dxdys_.find(detid);
+  return res == dxdys_.end() ? 0.f : res->second;
 }
diff --git a/RecoTracker/LSTCore/src/alpaka/Event.dev.cc b/RecoTracker/LSTCore/src/alpaka/Event.dev.cc
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.