• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 -nvvm-reflect -O2 -mtriple=nvptx64 \
5; RUN:   | FileCheck %s --check-prefixes=COMMON,SM20
6; RUN: opt %s -S -nvvm-reflect -O2 -mtriple=nvptx64 -mcpu=sm_35 \
7; RUN:   | FileCheck %s --check-prefixes=COMMON,SM35
8
9@"$str" = private addrspace(1) constant [12 x i8] c"__CUDA_ARCH\00"
10
11declare i32 @__nvvm_reflect(i8*)
12
13; COMMON-LABEL: @foo
14define i32 @foo(float %a, float %b) {
15; COMMON-NOT: call i32 @__nvvm_reflect
16  %reflect = call i32 @__nvvm_reflect(i8* addrspacecast (i8 addrspace(1)* getelementptr inbounds ([12 x i8], [12 x i8] addrspace(1)* @"$str", i32 0, i32 0) to i8*))
17; SM20: ret i32 200
18; SM35: ret i32 350
19  ret i32 %reflect
20}
21
22