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