[PATCH] D28323: [CUDA] Let NVPTX inherit the host's calling conventions.

Justin Lebar via Phabricator via cfe-commits cfe-commits at lists.llvm.org
Wed Jan 4 14:00:55 PST 2017


jlebar created this revision.
jlebar added a reviewer: tra.
jlebar added a subscriber: cfe-commits.

When compiling device code, we may still see host code with explicit
calling conventions.  NVPTX needs to claim that it supports these CCs,
so that (a) we don't raise noisy warnings, and (b) we don't break
existing code which relies on the existence of these CCs when
specializing templates.  (If a CC doesn't exist, clang ignores it, so
two template specializations which are different only insofar as one
specifies a CC are considered identical and therefore are an error if
that CC is not supported.)


https://reviews.llvm.org/D28323

Files:
  clang/lib/Basic/Targets.cpp
  clang/test/SemaCUDA/cuda-inherits-calling-conv.cu


Index: clang/test/SemaCUDA/cuda-inherits-calling-conv.cu
===================================================================
--- /dev/null
+++ clang/test/SemaCUDA/cuda-inherits-calling-conv.cu
@@ -0,0 +1,30 @@
+// RUN: %clang_cc1 -std=c++11 -triple i386-windows-msvc \
+// RUN:   -aux-triple nvptx-nvidia-cuda -fsyntax-only -verify %s
+
+// RUN: %clang_cc1 -std=c++11 -triple nvptx-nvidia-cuda \
+// RUN:   -aux-triple i386-windows-msvc -fsyntax-only \
+// RUN:   -fcuda-is-device -verify %s
+
+// RUN: %clang_cc1 -std=c++11 -triple nvptx-nvidia-cuda \
+// RUN:   -aux-triple x86_64-linux-gnu -fsyntax-only \
+// RUN:   -fcuda-is-device -verify -verify-ignore-unexpected=note \
+// RUN:   -DEXPECT_ERR %s
+
+// CUDA device code should inherit the host's calling conventions.
+
+template <class T>
+struct Foo;
+
+template <class T>
+struct Foo<T()> {};
+
+// On x86_64-linux-gnu, this is a redefinition of the template, because the
+// __fastcall calling convention doesn't exist (and is therefore ignored).
+#ifndef EXPECT_ERR
+// expected-no-diagnostics
+#else
+// expected-error at +4 {{redefinition of 'Foo}}
+// expected-warning at +3 {{calling convention '__fastcall' ignored}}
+#endif
+template <class T>
+struct Foo<T __fastcall()> {};
Index: clang/lib/Basic/Targets.cpp
===================================================================
--- clang/lib/Basic/Targets.cpp
+++ clang/lib/Basic/Targets.cpp
@@ -1962,6 +1962,16 @@
     Opts.support("cl_khr_local_int32_base_atomics");
     Opts.support("cl_khr_local_int32_extended_atomics");
   }
+
+  CallingConvCheckResult checkCallingConvention(CallingConv CC) const override {
+    // CUDA compilations support all of the host's calling conventions.
+    //
+    // TODO: We should warn if you apply a non-default CC to anything other than
+    // a host function.
+    if (HostTarget)
+      return HostTarget->checkCallingConvention(CC);
+    return CCCR_Warning;
+  }
 };
 
 const Builtin::Info NVPTXTargetInfo::BuiltinInfo[] = {


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D28323.83130.patch
Type: text/x-patch
Size: 1985 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20170104/b077d2d3/attachment-0001.bin>


More information about the cfe-commits mailing list