[PATCH] D44541: [OpenMP][Clang] Move device global stack init before master-workers split

[Gheorghe-Teodor Bercea via Phabricator via cfe-commits]

gtbercea created this revision.
gtbercea added reviewers: ABataev, carlo.bertolli, grokos, caomhin.
Herald added subscribers: cfe-commits, guansong, jholewinski.

This patch moves the call to the stack init data sharing function before the splitting of threads into master and workers. This ensures that the initialization happens for all active warp master threads. Test is modified appropriately.


Repository:
  rC Clang

https://reviews.llvm.org/D44541

Files:
  lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
  test/OpenMP/nvptx_data_sharing.cpp


Index: test/OpenMP/nvptx_data_sharing.cpp
===================================================================
--- test/OpenMP/nvptx_data_sharing.cpp
+++ test/OpenMP/nvptx_data_sharing.cpp
@@ -30,8 +30,9 @@
 // CK1: {{.*}}define void @__omp_offloading{{.*}}test_ds{{.*}}()
 // CK1: [[SHAREDARGS1:%.+]] = alloca i8**
 // CK1: [[SHAREDARGS2:%.+]] = alloca i8**
-// CK1: call void @__kmpc_kernel_init
 // CK1: call void @__kmpc_data_sharing_init_stack
+// CK1: call void @__omp_offloading{{.*}}_worker()
+// CK1: call void @__kmpc_kernel_init
 // CK1: [[GLOBALSTACK:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i64 8, i16 0)
 // CK1: [[GLOBALSTACK2:%.+]] = bitcast i8* [[GLOBALSTACK]] to %struct._globalized_locals_ty*
 // CK1: [[A:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[GLOBALSTACK2]], i32 0, i32 0
Index: lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
===================================================================
--- lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
+++ lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
@@ -598,6 +598,11 @@
                                                   WorkerFunctionState &WST) {
   CGBuilderTy &Bld = CGF.Builder;
 
+  // For data sharing, initialize the stack for master and workers.
+  CGF.EmitRuntimeCall(
+      createNVPTXRuntimeFunction(
+          OMPRTL_NVPTX__kmpc_data_sharing_init_stack));
+
   llvm::BasicBlock *WorkerBB = CGF.createBasicBlock(".worker");
   llvm::BasicBlock *MasterCheckBB = CGF.createBasicBlock(".mastercheck");
   llvm::BasicBlock *MasterBB = CGF.createBasicBlock(".master");
@@ -626,11 +631,6 @@
   CGF.EmitRuntimeCall(
       createNVPTXRuntimeFunction(OMPRTL_NVPTX__kmpc_kernel_init), Args);
 
-  // For data sharing, we need to initialize the stack.
-  CGF.EmitRuntimeCall(
-      createNVPTXRuntimeFunction(
-          OMPRTL_NVPTX__kmpc_data_sharing_init_stack));
-
   emitGenericVarsProlog(CGF, WST.Loc);
 }
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D44541.138622.patch
Type: text/x-patch
Size: 1920 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20180315/2cd813e9/attachment.bin>