[CUDA] fix codegen for __nvvm_atom_min/max_gen_u*

Summary: Clang should emit "atomicrmw umin/umax" instead of "atomicrmw min/max".

Reviewers: eliben, tra

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D12487

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@246455 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/CGBuiltin.cpp b/lib/CodeGen/CGBuiltin.cpp
index 955dcfc96ca24c84e2a271d82dc3a13f69390eaa..6966b03ef3b66d0ae9cec1090b89bfc0559fa339 100644 (file)
--- a/lib/CodeGen/CGBuiltin.cpp
+++ b/lib/CodeGen/CGBuiltin.cpp
@@ -6985,18 +6985,22 @@ Value *CodeGenFunction::EmitNVPTXBuiltinExpr(unsigned BuiltinID,
   case NVPTX::BI__nvvm_atom_max_gen_i:
   case NVPTX::BI__nvvm_atom_max_gen_l:
   case NVPTX::BI__nvvm_atom_max_gen_ll:
+    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Max, E);
+
   case NVPTX::BI__nvvm_atom_max_gen_ui:
   case NVPTX::BI__nvvm_atom_max_gen_ul:
   case NVPTX::BI__nvvm_atom_max_gen_ull:
-    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Max, E);
+    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::UMax, E);
 
   case NVPTX::BI__nvvm_atom_min_gen_i:
   case NVPTX::BI__nvvm_atom_min_gen_l:
   case NVPTX::BI__nvvm_atom_min_gen_ll:
+    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Min, E);
+
   case NVPTX::BI__nvvm_atom_min_gen_ui:
   case NVPTX::BI__nvvm_atom_min_gen_ul:
   case NVPTX::BI__nvvm_atom_min_gen_ull:
-    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Min, E);
+    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::UMin, E);
 
   case NVPTX::BI__nvvm_atom_cas_gen_i:
   case NVPTX::BI__nvvm_atom_cas_gen_l:

diff --git a/test/CodeGen/builtins-nvptx.c b/test/CodeGen/builtins-nvptx.c
index ebf20673ddb48d74e3be0c09c4f95baa780acd25..498da476d102da572f20cfa24a75c1c26e824f22 100644 (file)
--- a/test/CodeGen/builtins-nvptx.c
+++ b/test/CodeGen/builtins-nvptx.c
@@ -234,30 +234,30 @@ __device__ void nvvm_atom(float *fp, float f, int *ip, int i, long *lp, long l,
   // CHECK: atomicrmw xchg
   __nvvm_atom_xchg_gen_ll(&sll, ll);
 
-  // CHECK: atomicrmw max
+  // CHECK: atomicrmw max i32*
   __nvvm_atom_max_gen_i(ip, i);
-  // CHECK: atomicrmw max
+  // CHECK: atomicrmw umax i32*
   __nvvm_atom_max_gen_ui((unsigned int *)ip, i);
   // CHECK: atomicrmw max
   __nvvm_atom_max_gen_l(&dl, l);
-  // CHECK: atomicrmw max
+  // CHECK: atomicrmw umax
   __nvvm_atom_max_gen_ul((unsigned long *)&dl, l);
-  // CHECK: atomicrmw max
+  // CHECK: atomicrmw max i64*
   __nvvm_atom_max_gen_ll(&sll, ll);
-  // CHECK: atomicrmw max
+  // CHECK: atomicrmw umax i64*
   __nvvm_atom_max_gen_ull((unsigned long long *)&sll, ll);
 
-  // CHECK: atomicrmw min
+  // CHECK: atomicrmw min i32*
   __nvvm_atom_min_gen_i(ip, i);
-  // CHECK: atomicrmw min
+  // CHECK: atomicrmw umin i32*
   __nvvm_atom_min_gen_ui((unsigned int *)ip, i);
   // CHECK: atomicrmw min
   __nvvm_atom_min_gen_l(&dl, l);
-  // CHECK: atomicrmw min
+  // CHECK: atomicrmw umin
   __nvvm_atom_min_gen_ul((unsigned long *)&dl, l);
-  // CHECK: atomicrmw min
+  // CHECK: atomicrmw min i64*
   __nvvm_atom_min_gen_ll(&sll, ll);
-  // CHECK: atomicrmw min
+  // CHECK: atomicrmw umin i64*
   __nvvm_atom_min_gen_ull((unsigned long long *)&sll, ll);
 
   // CHECK: cmpxchg