xref: /llvm-project/llvm/test/CodeGen/NVPTX/generic-to-nvvm.ll (revision 4583f6d3443c8dc6605c868724e3743161954210)
1; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_20 | FileCheck %s
2; RUN: %if ptxas %{ llc < %s -mtriple=nvptx64 -mcpu=sm_20 | %ptxas-verify %}
3
4target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64"
5target triple = "nvptx-nvidia-cuda"
6
7; Ensure global variables in address space 0 are promoted to address space 1
8
9; CHECK: .global .align 4 .u32 myglobal = 42;
10@myglobal = internal global i32 42, align 4
11; CHECK: .global .align 4 .u32 myconst = 420;
12@myconst = internal constant i32 420, align 4
13
14
15define ptx_kernel void @foo(ptr %a, ptr %b) {
16; Expect one load -- @myconst isn't loaded from, because we know its value
17; statically.
18; CHECK: ld.global.u32
19; CHECK: st.global.u32
20; CHECK: st.global.u32
21  %ld1 = load i32, ptr @myglobal
22  %ld2 = load i32, ptr @myconst
23  store i32 %ld1, ptr %a
24  store i32 %ld2, ptr %b
25  ret void
26}
27