r318098 - OpenCL: Assume inline asm is convergent

[Matt Arsenault via cfe-commits]

Author: arsenm
Date: Mon Nov 13 14:40:55 2017
New Revision: 318098

URL: http://llvm.org/viewvc/llvm-project?rev=318098&view=rev
Log:
OpenCL: Assume inline asm is convergent

Already done for CUDA.

Modified:
    cfe/trunk/lib/CodeGen/CGStmt.cpp
    cfe/trunk/test/CodeGenOpenCL/convergent.cl

Modified: cfe/trunk/lib/CodeGen/CGStmt.cpp
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CGStmt.cpp?rev=318098&r1=318097&r2=318098&view=diff
==============================================================================
--- cfe/trunk/lib/CodeGen/CGStmt.cpp (original)
+++ cfe/trunk/lib/CodeGen/CGStmt.cpp Mon Nov 13 14:40:55 2017
@@ -2149,10 +2149,11 @@ void CodeGenFunction::EmitAsmStmt(const
                                           llvm::ConstantAsMetadata::get(Loc)));
   }
 
-  if (getLangOpts().CUDA && getLangOpts().CUDAIsDevice) {
-    // Conservatively, mark all inline asm blocks in CUDA as convergent
-    // (meaning, they may call an intrinsically convergent op, such as bar.sync,
-    // and so can't have certain optimizations applied around them).
+  if (getLangOpts().assumeFunctionsAreConvergent()) {
+    // Conservatively, mark all inline asm blocks in CUDA or OpenCL as
+    // convergent (meaning, they may call an intrinsically convergent op, such
+    // as bar.sync, and so can't have certain optimizations applied around
+    // them).
     Result->addAttribute(llvm::AttributeList::FunctionIndex,
                          llvm::Attribute::Convergent);
   }

Modified: cfe/trunk/test/CodeGenOpenCL/convergent.cl
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGenOpenCL/convergent.cl?rev=318098&r1=318097&r2=318098&view=diff
==============================================================================
--- cfe/trunk/test/CodeGenOpenCL/convergent.cl (original)
+++ cfe/trunk/test/CodeGenOpenCL/convergent.cl Mon Nov 13 14:40:55 2017
@@ -126,6 +126,13 @@ void test_not_unroll() {
 
 // CHECK: declare spir_func void @nodupfun(){{[^#]*}} #[[attr3:[0-9]+]]
 
+// CHECK-LABEL: @assume_convergent_asm
+// CHECK: tail call void asm sideeffect "s_barrier", ""() #4
+kernel void assume_convergent_asm()
+{
+  __asm__ volatile("s_barrier");
+}
+
 // CHECK: attributes #0 = { noinline norecurse nounwind "
 // CHECK: attributes #1 = { {{[^}]*}}convergent{{[^}]*}} }
 // CHECK: attributes #2 = { {{[^}]*}}convergent{{[^}]*}} }