[llvm-branch-commits] [llvm-branch] r333475 - Merging r328885:

[Tom Stellard via llvm-branch-commits]

Author: tstellar
Date: Tue May 29 16:03:38 2018
New Revision: 333475

URL: http://llvm.org/viewvc/llvm-project?rev=333475&view=rev
Log:
Merging r328885:

------------------------------------------------------------------------
r328885 | timshen | 2018-03-30 10:51:03 -0700 (Fri, 30 Mar 2018) | 14 lines

[NVPTX] Enable StructuredCFG for NVPTX

Summary:
Make NVPTX require structured CFG. Added a temporary flag to
"roll back" the behavior for easy deployment.

Combined with D45008, this fixes several internal Nvidia GPU test
failures that we suspect to be ptxas miscompiles (PR27738).

Reviewers: jlebar

Subscribers: jholewinski, sanjoy, llvm-commits, hiraditya

Differential Revision: https://reviews.llvm.org/D45070
------------------------------------------------------------------------

Modified:
    llvm/branches/release_60/lib/Target/NVPTX/NVPTXTargetMachine.cpp

Modified: llvm/branches/release_60/lib/Target/NVPTX/NVPTXTargetMachine.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/release_60/lib/Target/NVPTX/NVPTXTargetMachine.cpp?rev=333475&r1=333474&r2=333475&view=diff
==============================================================================
--- llvm/branches/release_60/lib/Target/NVPTX/NVPTXTargetMachine.cpp (original)
+++ llvm/branches/release_60/lib/Target/NVPTX/NVPTXTargetMachine.cpp Tue May 29 16:03:38 2018
@@ -44,6 +44,14 @@ static cl::opt<bool>
                                cl::desc("Disable load/store vectorizer"),
                                cl::init(false), cl::Hidden);
 
+// TODO: Remove this flag when we are confident with no regressions.
+static cl::opt<bool> DisableRequireStructuredCFG(
+    "disable-nvptx-require-structured-cfg",
+    cl::desc("Transitional flag to turn off NVPTX's requirement on preserving "
+             "structured CFG. The requirement should be disabled only when "
+             "unexpected regressions happen."),
+    cl::init(false), cl::Hidden);
+
 namespace llvm {
 
 void initializeNVVMIntrRangePass(PassRegistry&);
@@ -108,6 +116,8 @@ NVPTXTargetMachine::NVPTXTargetMachine(c
     drvInterface = NVPTX::NVCL;
   else
     drvInterface = NVPTX::CUDA;
+  if (!DisableRequireStructuredCFG)
+    setRequiresStructuredCFG(true);
   initAsmInfo();
 }