[PATCH] D78655: [HIP] Let lambda be host device by default

[Yaxun Liu via Phabricator via cfe-commits]

yaxunl created this revision.
yaxunl added reviewers: tra, rjmccall.

Lambda functions do not have names, therefore they do not need host/device
attribute for overloading resolution. They are also have internal linkage and
is only emitted if used, therefore no need to use host/device attribute to
indicate that they should only be emitted for host or device, since clang
can detect whether they are used and emitted accordingly.

Therefore it seems letting lambda functions have host device attributes
by default should not cause ambiguity or unexpected emission.

On the other hand, inferring host/device attribute of lambda function
by context is inaccurate, since a lambda function can be defined in
a host function and passed to a template kernel as template argument
and called in that kernel, i.e., many cases a lambda function defined in
a host function is intended to be a device function.

This patch let lambda function be host device by default for HIP.
This should make lambda easier to use without unwanted side effect.


https://reviews.llvm.org/D78655

Files:
  clang/lib/Sema/SemaCUDA.cpp
  clang/test/CodeGenCUDA/lambda.cu


Index: clang/test/CodeGenCUDA/lambda.cu
===================================================================
--- /dev/null
+++ clang/test/CodeGenCUDA/lambda.cu
@@ -0,0 +1,26 @@
+// RUN: %clang_cc1 -x hip -emit-llvm -std=c++11 %s -o - \
+// RUN:   -triple x86_64-linux-gnu | FileCheck -check-prefix=HOST %s
+// RUN: %clang_cc1 -x hip -emit-llvm -std=c++11 %s -o - \
+// RUN:   -triple amdgcn-amd-amdhsa -fcuda-is-device \
+// RUN:   | FileCheck -check-prefix=DEV %s
+
+#include "Inputs/cuda.h"
+
+// HOST: @[[KERN:[0-9]+]] = private unnamed_addr constant [22 x i8] c"_Z1gIZ4mainEUlvE_EvT_\00"
+// HOST: define internal void @_Z1hIZ4mainEUlvE_EvT_
+// HOST: define internal void @_Z16__device_stub__gIZ4mainEUlvE_EvT_
+// HOST: @__hipRegisterFunction(i8** %0, i8* bitcast ({{.*}}@[[KERN]]
+// HOST-NOT: define{{.*}}@_ZZ4mainENKUlvE_clEv
+// DEV: define amdgpu_kernel void @_Z1gIZ4mainEUlvE_EvT_
+// DEV: define internal void @_ZZ4mainENKUlvE_clEv
+template<class F>
+__global__ void g(F f) { f(); }
+
+template<class F>
+void h(F f) { g<<<1,1>>>(f); }
+
+__device__ int a;
+
+int main(void) {
+  h([&](){ a=1;});
+}
Index: clang/lib/Sema/SemaCUDA.cpp
===================================================================
--- clang/lib/Sema/SemaCUDA.cpp
+++ clang/lib/Sema/SemaCUDA.cpp
@@ -718,6 +718,11 @@
   FunctionDecl *CurFn = dyn_cast<FunctionDecl>(CurContext);
   if (!CurFn)
     return;
+  if (getLangOpts().HIP) {
+    Method->addAttr(CUDADeviceAttr::CreateImplicit(Context));
+    Method->addAttr(CUDAHostAttr::CreateImplicit(Context));
+    return;
+  }
   CUDAFunctionTarget Target = IdentifyCUDATarget(CurFn);
   if (Target == CFT_Global || Target == CFT_Device) {
     Method->addAttr(CUDADeviceAttr::CreateImplicit(Context));


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D78655.259330.patch
Type: text/x-patch
Size: 1736 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20200422/6e6f58c5/attachment.bin>