[OpenMP] Make default distribute schedule for NVPTX target regions in SPMD mode achieve coalescing

Summary: For the OpenMP NVPTX toolchain choose a default distribute schedule that ensures coalescing on the GPU when in SPMD mode. This significantly increases the performance of offloaded target code and reduces the number of registers used on the GPU side.

Reviewers: ABataev, caomhin, Hahnfeld

Reviewed By: ABataev, Hahnfeld

Subscribers: Hahnfeld, jholewinski, guansong, cfe-commits

Differential Revision: https://reviews.llvm.org/D52434

llvm-svn: 343253

1 files changed, 12 insertions, 0 deletions

diff --git a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
index 45aafaa5c3c..56b244d0ae8 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
@@ -4081,3 +4081,15 @@ void CGOpenMPRuntimeNVPTX::functionFinished(CodeGenFunction &CGF) {
   FunctionGlobalizedDecls.erase(CGF.CurFn);
   CGOpenMPRuntime::functionFinished(CGF);
 }
+
+void CGOpenMPRuntimeNVPTX::getDefaultDistScheduleAndChunk(
+    CodeGenFunction &CGF, const OMPLoopDirective &S,
+    OpenMPDistScheduleClauseKind &ScheduleKind,
+    llvm::Value *&Chunk) const {
+  if (getExecutionMode() == CGOpenMPRuntimeNVPTX::EM_SPMD) {
+    ScheduleKind = OMPC_DIST_SCHEDULE_static;
+    Chunk = CGF.EmitScalarConversion(getNVPTXNumThreads(CGF),
+        CGF.getContext().getIntTypeForBitwidth(32, /*Signed=*/0),
+        S.getIterationVariable()->getType(), S.getBeginLoc());
+  }
+}