[PATCH] D69493: Add -fconvergent-functions flag

[Matt Arsenault via Phabricator via cfe-commits]

arsenm created this revision.
arsenm added reviewers: yaxunl, tra, jlebar.
Herald added a subscriber: wdng.

The CUDA builtin library is apparently compiled in C++ mode, so the
assumption of convergent needs to be made in a typically non-SPMD
language. The functions in the library should still be assumed
convergent. Currently they are not, which is potentially incorrect and
this happens to work after the library is linked.

      

This is in preparation for a change to convergent handling to allow
downstream users to start using the flag where necessary to avoid
disruption.


https://reviews.llvm.org/D69493

Files:
  clang/include/clang/Basic/LangOptions.def
  clang/include/clang/Basic/LangOptions.h
  clang/include/clang/Driver/Options.td
  clang/lib/Frontend/CompilerInvocation.cpp
  clang/test/CodeGen/convergent-functions.cpp
  clang/test/CodeGenCUDA/propagate-metadata.cu


Index: clang/test/CodeGenCUDA/propagate-metadata.cu
===================================================================
--- clang/test/CodeGenCUDA/propagate-metadata.cu
+++ clang/test/CodeGenCUDA/propagate-metadata.cu
@@ -11,7 +11,7 @@
 
 // Build the bitcode library.  This is not built in CUDA mode, otherwise it
 // might have incompatible attributes.  This mirrors how libdevice is built.
-// RUN: %clang_cc1 -x c++ -emit-llvm-bc -ftrapping-math -DLIB \
+// RUN: %clang_cc1 -x c++ -fconvergent-functions -emit-llvm-bc -ftrapping-math -DLIB \
 // RUN:   %s -o %t.bc -triple nvptx-unknown-unknown
 
 // RUN: %clang_cc1 -x cuda %s -emit-llvm -mlink-builtin-bitcode %t.bc -o - \
@@ -53,7 +53,8 @@
 
 // Check the attribute list.
 // CHECK: attributes [[attr]] = {
-// CHECK: "no-trapping-math"="true"
+// CHECK-SAME: convergent
+// CHECK-SAME: "no-trapping-math"="true"
 
 // FTZ-SAME: "nvptx-f32ftz"="true"
 // NOFTZ-NOT: "nvptx-f32ftz"="true"
Index: clang/test/CodeGen/convergent-functions.cpp
===================================================================
--- /dev/null
+++ clang/test/CodeGen/convergent-functions.cpp
@@ -0,0 +1,8 @@
+// RUN: %clang_cc1 -triple i386-pc-win32 -emit-llvm -fconvergent-functions -o - < %s | FileCheck -check-prefix=CONVFUNC %s
+// RUN: %clang_cc1 -triple i386-pc-win32 -emit-llvm -o - < %s | FileCheck -check-prefix=NOCONVFUNC %s
+
+// Test that the -fconvergent-functions flag works
+
+// CONVFUNC: attributes #0 = { convergent {{.*}} }
+// NOCONVFUNC-NOT: convergent
+void func() { }
Index: clang/lib/Frontend/CompilerInvocation.cpp
===================================================================
--- clang/lib/Frontend/CompilerInvocation.cpp
+++ clang/lib/Frontend/CompilerInvocation.cpp
@@ -2756,6 +2756,9 @@
   Opts.BlocksRuntimeOptional = Args.hasArg(OPT_fblocks_runtime_optional);
   Opts.Coroutines = Opts.CPlusPlus2a || Args.hasArg(OPT_fcoroutines_ts);
 
+  Opts.ConvergentFunctions = Opts.OpenCL || (Opts.CUDA && Opts.CUDAIsDevice) ||
+    Args.hasArg(OPT_fconvergent_functions);
+
   Opts.DoubleSquareBracketAttributes =
       Args.hasFlag(OPT_fdouble_square_bracket_attributes,
                    OPT_fno_double_square_bracket_attributes,
Index: clang/include/clang/Driver/Options.td
===================================================================
--- clang/include/clang/Driver/Options.td
+++ clang/include/clang/Driver/Options.td
@@ -548,6 +548,9 @@
   MetaVarName<"<directory>">;
 def c : Flag<["-"], "c">, Flags<[DriverOption]>, Group<Action_Group>,
   HelpText<"Only run preprocess, compile, and assemble steps">;
+def fconvergent_functions : Joined<["-"], "fconvergent-functions">, Group<f_Group>, Flags<[CC1Option]>,
+  HelpText<"Assume functions may be convergent">;
+
 def cuda_device_only : Flag<["--"], "cuda-device-only">,
   HelpText<"Compile CUDA code for device only">;
 def cuda_host_only : Flag<["--"], "cuda-host-only">,
Index: clang/include/clang/Basic/LangOptions.h
===================================================================
--- clang/include/clang/Basic/LangOptions.h
+++ clang/include/clang/Basic/LangOptions.h
@@ -312,7 +312,7 @@
   }
 
   bool assumeFunctionsAreConvergent() const {
-    return (CUDA && CUDAIsDevice) || OpenCL;
+    return ConvergentFunctions;
   }
 
   /// Return the OpenCL C or C++ version as a VersionTuple.
Index: clang/include/clang/Basic/LangOptions.def
===================================================================
--- clang/include/clang/Basic/LangOptions.def
+++ clang/include/clang/Basic/LangOptions.def
@@ -122,6 +122,7 @@
 LANGOPT(ConstStrings      , 1, 0, "const-qualified string support")
 ENUM_LANGOPT(LaxVectorConversions, LaxVectorConversionKind, 2,
              LaxVectorConversionKind::All, "lax vector conversions")
+LANGOPT(ConvergentFunctions, 1, 1, "Assume convergent functions")
 LANGOPT(AltiVec           , 1, 0, "AltiVec-style vector initializers")
 LANGOPT(ZVector           , 1, 0, "System z vector extensions")
 LANGOPT(Exceptions        , 1, 0, "exception handling")


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D69493.226603.patch
Type: text/x-patch
Size: 4017 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20191028/33ab50cb/attachment-0001.bin>