1; RUN: llc < %s -march=nvptx -mcpu=sm_20 | FileCheck %s
2; RUN: %if ptxas %{ llc < %s -march=nvptx -mcpu=sm_20 | %ptxas-verify %}
3
4;;; Need 4-byte alignment on float* passed byval
5define ptx_device void @t1(float* byval(float) %x) {
6; CHECK: .func t1
7; CHECK: .param .align 4 .b8 t1_param_0[4]
8  ret void
9}
10
11
12;;; Need 8-byte alignment on double* passed byval
13define ptx_device void @t2(double* byval(double) %x) {
14; CHECK: .func t2
15; CHECK: .param .align 8 .b8 t2_param_0[8]
16  ret void
17}
18
19
20;;; Need 4-byte alignment on float2* passed byval
21%struct.float2 = type { float, float }
22define ptx_device void @t3(%struct.float2* byval(%struct.float2) %x) {
23; CHECK: .func t3
24; CHECK: .param .align 4 .b8 t3_param_0[8]
25  ret void
26}
27
28;;; Need at least 4-byte alignment in order to avoid miscompilation by
29;;; ptxas for sm_50+
30define ptx_device void @t4(i8* byval(i8) %x) {
31; CHECK: .func t4
32; CHECK: .param .align 4 .b8 t4_param_0[1]
33  ret void
34}
35
36;;; Make sure we adjust alignment at the call site as well.
37define ptx_device void @t5(i8* align 2 byval(i8) %x) {
38; CHECK: .func t5
39; CHECK: .param .align 4 .b8 t5_param_0[1]
40; CHECK: {
41; CHECK: .param .align 4 .b8 param0[1];
42; CHECK: call.uni
43  call void @t4(i8* byval(i8) %x)
44  ret void
45}
46