[Openmp-commits] [PATCH] D87084: [OpenMP][AMDGPU] Use DS_Max_Warp_Number instead of WARPSIZE

[Pushpinder Singh via Phabricator via Openmp-commits]

pdhaliwal created this revision.
pdhaliwal added reviewers: JonChesterfield, ABataev, jdoerfert, grokos.
Herald added subscribers: openmp-commits, guansong, t-tye, tpr, dstuttard, yaxunl, kzhuravl.
Herald added a project: OpenMP.
pdhaliwal requested review of this revision.
Herald added subscribers: sstefan1, wdng.

The size of worker_rootS should have been DS_Max_Warp_Number.
This reduces memory usage by deviceRTL on AMDGPU from around 2.3GB
to around 770MB.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D87084

Files:
  openmp/libomptarget/deviceRTLs/common/omptarget.h


Index: openmp/libomptarget/deviceRTLs/common/omptarget.h
===================================================================
--- openmp/libomptarget/deviceRTLs/common/omptarget.h
+++ openmp/libomptarget/deviceRTLs/common/omptarget.h
@@ -252,7 +252,7 @@
       workDescrForActiveParallel; // one, ONLY for the active par
 
   ALIGN(16)
-  __kmpc_data_sharing_worker_slot_static worker_rootS[WARPSIZE];
+  __kmpc_data_sharing_worker_slot_static worker_rootS[DS_Max_Warp_Number];
   ALIGN(16) __kmpc_data_sharing_master_slot_static master_rootS[1];
 };
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D87084.289691.patch
Type: text/x-patch
Size: 552 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/openmp-commits/attachments/20200903/72b7e0f3/attachment.bin>