1 ; RUN: opt -mtriple=amdgcn--amdhsa -S -inline -inline-threshold=0 < %s | FileCheck %s
2 ; RUN: opt -mtriple=amdgcn--amdhsa -S -passes=inline -inline-threshold=0 < %s | FileCheck %s
4 target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5"
6 define void @use_flat_ptr_arg(float* nocapture %p) {
8 %tmp1 = load float, float* %p, align 4
9 %div = fdiv float 1.000000e+00, %tmp1
10 %add0 = fadd float %div, 1.0
11 %add1 = fadd float %add0, 1.0
12 %add2 = fadd float %add1, 1.0
13 %add3 = fadd float %add2, 1.0
14 %add4 = fadd float %add3, 1.0
15 %add5 = fadd float %add4, 1.0
16 %add6 = fadd float %add5, 1.0
17 %add7 = fadd float %add6, 1.0
18 %add8 = fadd float %add7, 1.0
19 %add9 = fadd float %add8, 1.0
20 %add10 = fadd float %add9, 1.0
21 store float %add10, float* %p, align 4
25 define void @use_private_ptr_arg(float addrspace(5)* nocapture %p) {
27 %tmp1 = load float, float addrspace(5)* %p, align 4
28 %div = fdiv float 1.000000e+00, %tmp1
29 %add0 = fadd float %div, 1.0
30 %add1 = fadd float %add0, 1.0
31 %add2 = fadd float %add1, 1.0
32 %add3 = fadd float %add2, 1.0
33 %add4 = fadd float %add3, 1.0
34 %add5 = fadd float %add4, 1.0
35 %add6 = fadd float %add5, 1.0
36 %add7 = fadd float %add6, 1.0
37 %add8 = fadd float %add7, 1.0
38 %add9 = fadd float %add8, 1.0
39 %add10 = fadd float %add9, 1.0
40 store float %add10, float addrspace(5)* %p, align 4
44 ; Test that the inline threshold is boosted if called with an
45 ; addrspacecasted' alloca.
46 ; CHECK-LABEL: @test_inliner_flat_ptr(
47 ; CHECK: call i32 @llvm.amdgcn.workitem.id.x()
50 define amdgpu_kernel void @test_inliner_flat_ptr(float addrspace(1)* nocapture %a, i32 %n) {
52 %pvt_arr = alloca [64 x float], align 4, addrspace(5)
53 %tid = tail call i32 @llvm.amdgcn.workitem.id.x()
54 %arrayidx = getelementptr inbounds float, float addrspace(1)* %a, i32 %tid
55 %tmp2 = load float, float addrspace(1)* %arrayidx, align 4
56 %add = add i32 %tid, 1
57 %arrayidx2 = getelementptr inbounds float, float addrspace(1)* %a, i32 %add
58 %tmp5 = load float, float addrspace(1)* %arrayidx2, align 4
60 %arrayidx5 = getelementptr inbounds [64 x float], [64 x float] addrspace(5)* %pvt_arr, i32 0, i32 %or
61 %arrayidx7 = getelementptr inbounds [64 x float], [64 x float] addrspace(5)* %pvt_arr, i32 0, i32 %or
62 %to.flat = addrspacecast float addrspace(5)* %arrayidx7 to float*
63 call void @use_private_ptr_arg(float addrspace(5)* %arrayidx7)
64 call void @use_flat_ptr_arg(float* %to.flat)
68 declare i32 @llvm.amdgcn.workitem.id.x() #1
70 attributes #0 = { noinline }
71 attributes #1 = { nounwind readnone }