Author: jlebar Date: Tue Nov 7 14:10:54 2017 New Revision: 317623 URL: http://llvm.org/viewvc/llvm-project?rev=317623&view=rev Log: [NVPTX] Implement __nvvm_atom_add_gen_d builtin.
Summary: This just seems to have been an oversight. We already supported the f64 atomic add with an explicit scope (e.g. "cta"), but not the scopeless version. Reviewers: tra Subscribers: jholewinski, sanjoy, cfe-commits, llvm-commits, hiraditya Differential Revision: https://reviews.llvm.org/D39638 Added: cfe/trunk/test/CodeGen/builtins-nvptx-ptx50.cu Modified: cfe/trunk/include/clang/Basic/BuiltinsNVPTX.def cfe/trunk/lib/CodeGen/CGBuiltin.cpp Modified: cfe/trunk/include/clang/Basic/BuiltinsNVPTX.def URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Basic/BuiltinsNVPTX.def?rev=317623&r1=317622&r2=317623&view=diff ============================================================================== --- cfe/trunk/include/clang/Basic/BuiltinsNVPTX.def (original) +++ cfe/trunk/include/clang/Basic/BuiltinsNVPTX.def Tue Nov 7 14:10:54 2017 @@ -481,7 +481,7 @@ TARGET_BUILTIN(__nvvm_atom_cta_add_gen_f TARGET_BUILTIN(__nvvm_atom_sys_add_gen_f, "ffD*f", "n", "satom") BUILTIN(__nvvm_atom_add_g_d, "ddD*1d", "n") BUILTIN(__nvvm_atom_add_s_d, "ddD*3d", "n") -BUILTIN(__nvvm_atom_add_gen_d, "ddD*d", "n") +TARGET_BUILTIN(__nvvm_atom_add_gen_d, "ddD*d", "n", "satom") TARGET_BUILTIN(__nvvm_atom_cta_add_gen_d, "ddD*d", "n", "satom") TARGET_BUILTIN(__nvvm_atom_sys_add_gen_d, "ddD*d", "n", "satom") Modified: cfe/trunk/lib/CodeGen/CGBuiltin.cpp URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CGBuiltin.cpp?rev=317623&r1=317622&r2=317623&view=diff ============================================================================== --- cfe/trunk/lib/CodeGen/CGBuiltin.cpp (original) +++ cfe/trunk/lib/CodeGen/CGBuiltin.cpp Tue Nov 7 14:10:54 2017 @@ -9554,6 +9554,16 @@ Value *CodeGenFunction::EmitNVPTXBuiltin return Builder.CreateCall(FnALAF32, {Ptr, Val}); } + case NVPTX::BI__nvvm_atom_add_gen_d: { + Value *Ptr = EmitScalarExpr(E->getArg(0)); + Value *Val = EmitScalarExpr(E->getArg(1)); + // atomicrmw only deals with integer arguments, so we need to use + // LLVM's nvvm_atomic_load_add_f64 intrinsic. + Value *FnALAF64 = + CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_add_f64, Ptr->getType()); + return Builder.CreateCall(FnALAF64, {Ptr, Val}); + } + case NVPTX::BI__nvvm_atom_inc_gen_ui: { Value *Ptr = EmitScalarExpr(E->getArg(0)); Value *Val = EmitScalarExpr(E->getArg(1)); Added: cfe/trunk/test/CodeGen/builtins-nvptx-ptx50.cu URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGen/builtins-nvptx-ptx50.cu?rev=317623&view=auto ============================================================================== --- cfe/trunk/test/CodeGen/builtins-nvptx-ptx50.cu (added) +++ cfe/trunk/test/CodeGen/builtins-nvptx-ptx50.cu Tue Nov 7 14:10:54 2017 @@ -0,0 +1,23 @@ +// RUN: %clang_cc1 -triple nvptx64-unknown-unknown -target-cpu sm_60 \ +// RUN: -fcuda-is-device -S -emit-llvm -o - -x cuda %s \ +// RUN: | FileCheck -check-prefix=CHECK %s +// +// RUN: %clang_cc1 -triple nvptx-unknown-unknown -target-cpu sm_50 \ +// RUN: -fcuda-is-device -S -o /dev/null -x cuda -verify %s + +#define __device__ __attribute__((device)) +#define __global__ __attribute__((global)) +#define __shared__ __attribute__((shared)) +#define __constant__ __attribute__((constant)) + +// We have to keep all builtins that depend on particular target feature in the +// same function, because the codegen will stop after the very first function +// that encounters an error, so -verify will not be able to find errors in +// subsequent functions. + +// CHECK-LABEL: test_fn +__device__ void test_fn(double d, double* double_ptr) { + // CHECK: call double @llvm.nvvm.atomic.load.add.f64.p0f64 + // expected-error@+1 {{'__nvvm_atom_add_gen_d' needs target feature satom}} + __nvvm_atom_add_gen_d(double_ptr, d); +} _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits