xref: /llvm-project/llvm/test/CodeGen/NVPTX/nvvm-reflect-arch.ll (revision ffabcbcf8f9fc7ef5fd29e2a711f95aba0ef7808)
1; Libdevice in recent CUDA versions relies on __CUDA_ARCH reflecting GPU type.
2; Verify that __nvvm_reflect() is replaced with an appropriate value.
3;
4; RUN: opt %s -S -passes='default<O2>' -mtriple=nvptx64 -mcpu=sm_20 \
5; RUN:   | FileCheck %s --check-prefixes=COMMON,SM20
6; RUN: opt %s -S -passes='default<O2>' -mtriple=nvptx64 -mcpu=sm_35 \
7; RUN:   | FileCheck %s --check-prefixes=COMMON,SM35
8
9@"$str" = private addrspace(1) constant [12 x i8] c"__CUDA_ARCH\00"
10
11declare i32 @__nvvm_reflect(ptr)
12
13; COMMON-LABEL: @foo
14define i32 @foo(float %a, float %b) {
15; COMMON-NOT: call i32 @__nvvm_reflect
16  %reflect = call i32 @__nvvm_reflect(ptr addrspacecast (ptr addrspace(1) @"$str" to ptr))
17; SM20: ret i32 200
18; SM35: ret i32 350
19  ret i32 %reflect
20}
21